diff options
Diffstat (limited to 'contrib/llvm/lib/Target/AArch64/AArch64InstrNEON.td')
| -rw-r--r-- | contrib/llvm/lib/Target/AArch64/AArch64InstrNEON.td | 8671 |
1 files changed, 8671 insertions, 0 deletions
diff --git a/contrib/llvm/lib/Target/AArch64/AArch64InstrNEON.td b/contrib/llvm/lib/Target/AArch64/AArch64InstrNEON.td new file mode 100644 index 0000000..d71749d --- /dev/null +++ b/contrib/llvm/lib/Target/AArch64/AArch64InstrNEON.td @@ -0,0 +1,8671 @@ +//===-- AArch64InstrNEON.td - NEON support for AArch64 -----*- 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 AArch64 NEON instruction set. +// +//===----------------------------------------------------------------------===// + +//===----------------------------------------------------------------------===// +// NEON-specific DAG Nodes. +//===----------------------------------------------------------------------===// +def Neon_bsl : SDNode<"AArch64ISD::NEON_BSL", SDTypeProfile<1, 3, + [SDTCisVec<0>, SDTCisSameAs<0, 1>, SDTCisSameAs<0, 2>, + SDTCisSameAs<0, 3>]>>; + +// (outs Result), (ins Imm, OpCmode) +def SDT_Neon_movi : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisVT<1, i32>]>; + +def Neon_movi : SDNode<"AArch64ISD::NEON_MOVIMM", SDT_Neon_movi>; + +def Neon_mvni : SDNode<"AArch64ISD::NEON_MVNIMM", SDT_Neon_movi>; + +// (outs Result), (ins Imm) +def Neon_fmovi : SDNode<"AArch64ISD::NEON_FMOVIMM", SDTypeProfile<1, 1, + [SDTCisVec<0>, SDTCisVT<1, i32>]>>; + +// (outs Result), (ins LHS, RHS, CondCode) +def Neon_cmp : SDNode<"AArch64ISD::NEON_CMP", SDTypeProfile<1, 3, + [SDTCisVec<0>, SDTCisSameAs<1, 2>]>>; + +// (outs Result), (ins LHS, 0/0.0 constant, CondCode) +def Neon_cmpz : SDNode<"AArch64ISD::NEON_CMPZ", SDTypeProfile<1, 3, + [SDTCisVec<0>, SDTCisVec<1>]>>; + +// (outs Result), (ins LHS, RHS) +def Neon_tst : SDNode<"AArch64ISD::NEON_TST", SDTypeProfile<1, 2, + [SDTCisVec<0>, SDTCisSameAs<1, 2>]>>; + +def SDTARMVSH : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>, + SDTCisVT<2, i32>]>; +def Neon_sqrshlImm : SDNode<"AArch64ISD::NEON_QSHLs", SDTARMVSH>; +def Neon_uqrshlImm : SDNode<"AArch64ISD::NEON_QSHLu", SDTARMVSH>; + +def SDTPERMUTE : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>, + SDTCisSameAs<0, 2>]>; +def Neon_uzp1 : SDNode<"AArch64ISD::NEON_UZP1", SDTPERMUTE>; +def Neon_uzp2 : SDNode<"AArch64ISD::NEON_UZP2", SDTPERMUTE>; +def Neon_zip1 : SDNode<"AArch64ISD::NEON_ZIP1", SDTPERMUTE>; +def Neon_zip2 : SDNode<"AArch64ISD::NEON_ZIP2", SDTPERMUTE>; +def Neon_trn1 : SDNode<"AArch64ISD::NEON_TRN1", SDTPERMUTE>; +def Neon_trn2 : SDNode<"AArch64ISD::NEON_TRN2", SDTPERMUTE>; + +def SDTVSHUF : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0, 1>]>; +def Neon_rev64 : SDNode<"AArch64ISD::NEON_REV64", SDTVSHUF>; +def Neon_rev32 : SDNode<"AArch64ISD::NEON_REV32", SDTVSHUF>; +def Neon_rev16 : SDNode<"AArch64ISD::NEON_REV16", SDTVSHUF>; +def Neon_vdup : SDNode<"AArch64ISD::NEON_VDUP", SDTypeProfile<1, 1, + [SDTCisVec<0>]>>; +def Neon_vduplane : SDNode<"AArch64ISD::NEON_VDUPLANE", SDTypeProfile<1, 2, + [SDTCisVec<0>, SDTCisVec<1>, SDTCisVT<2, i64>]>>; +def Neon_vextract : SDNode<"AArch64ISD::NEON_VEXTRACT", SDTypeProfile<1, 3, + [SDTCisVec<0>, SDTCisSameAs<0, 1>, + SDTCisSameAs<0, 2>, SDTCisVT<3, i64>]>>; + +def SDT_assertext : SDTypeProfile<1, 1, + [SDTCisInt<0>, SDTCisInt<1>, SDTCisSameAs<1, 0>]>; +def assertsext : SDNode<"ISD::AssertSext", SDT_assertext>; +def assertzext : SDNode<"ISD::AssertZext", SDT_assertext>; + +//===----------------------------------------------------------------------===// +// Multiclasses +//===----------------------------------------------------------------------===// + +multiclass NeonI_3VSame_B_sizes<bit u, bits<2> size, bits<5> opcode, + string asmop, SDPatternOperator opnode8B, + SDPatternOperator opnode16B, + bit Commutable = 0> { + let isCommutable = Commutable in { + def _8B : NeonI_3VSame<0b0, u, size, opcode, + (outs VPR64:$Rd), (ins VPR64:$Rn, VPR64:$Rm), + asmop # "\t$Rd.8b, $Rn.8b, $Rm.8b", + [(set (v8i8 VPR64:$Rd), + (v8i8 (opnode8B (v8i8 VPR64:$Rn), (v8i8 VPR64:$Rm))))], + NoItinerary>; + + def _16B : NeonI_3VSame<0b1, u, size, opcode, + (outs VPR128:$Rd), (ins VPR128:$Rn, VPR128:$Rm), + asmop # "\t$Rd.16b, $Rn.16b, $Rm.16b", + [(set (v16i8 VPR128:$Rd), + (v16i8 (opnode16B (v16i8 VPR128:$Rn), (v16i8 VPR128:$Rm))))], + NoItinerary>; + } + +} + +multiclass NeonI_3VSame_HS_sizes<bit u, bits<5> opcode, + string asmop, SDPatternOperator opnode, + bit Commutable = 0> { + let isCommutable = Commutable in { + def _4H : NeonI_3VSame<0b0, u, 0b01, opcode, + (outs VPR64:$Rd), (ins VPR64:$Rn, VPR64:$Rm), + asmop # "\t$Rd.4h, $Rn.4h, $Rm.4h", + [(set (v4i16 VPR64:$Rd), + (v4i16 (opnode (v4i16 VPR64:$Rn), (v4i16 VPR64:$Rm))))], + NoItinerary>; + + def _8H : NeonI_3VSame<0b1, u, 0b01, opcode, + (outs VPR128:$Rd), (ins VPR128:$Rn, VPR128:$Rm), + asmop # "\t$Rd.8h, $Rn.8h, $Rm.8h", + [(set (v8i16 VPR128:$Rd), + (v8i16 (opnode (v8i16 VPR128:$Rn), (v8i16 VPR128:$Rm))))], + NoItinerary>; + + def _2S : NeonI_3VSame<0b0, u, 0b10, opcode, + (outs VPR64:$Rd), (ins VPR64:$Rn, VPR64:$Rm), + asmop # "\t$Rd.2s, $Rn.2s, $Rm.2s", + [(set (v2i32 VPR64:$Rd), + (v2i32 (opnode (v2i32 VPR64:$Rn), (v2i32 VPR64:$Rm))))], + NoItinerary>; + + def _4S : NeonI_3VSame<0b1, u, 0b10, opcode, + (outs VPR128:$Rd), (ins VPR128:$Rn, VPR128:$Rm), + asmop # "\t$Rd.4s, $Rn.4s, $Rm.4s", + [(set (v4i32 VPR128:$Rd), + (v4i32 (opnode (v4i32 VPR128:$Rn), (v4i32 VPR128:$Rm))))], + NoItinerary>; + } +} +multiclass NeonI_3VSame_BHS_sizes<bit u, bits<5> opcode, + string asmop, SDPatternOperator opnode, + bit Commutable = 0> + : NeonI_3VSame_HS_sizes<u, opcode, asmop, opnode, Commutable> { + let isCommutable = Commutable in { + def _8B : NeonI_3VSame<0b0, u, 0b00, opcode, + (outs VPR64:$Rd), (ins VPR64:$Rn, VPR64:$Rm), + asmop # "\t$Rd.8b, $Rn.8b, $Rm.8b", + [(set (v8i8 VPR64:$Rd), + (v8i8 (opnode (v8i8 VPR64:$Rn), (v8i8 VPR64:$Rm))))], + NoItinerary>; + + def _16B : NeonI_3VSame<0b1, u, 0b00, opcode, + (outs VPR128:$Rd), (ins VPR128:$Rn, VPR128:$Rm), + asmop # "\t$Rd.16b, $Rn.16b, $Rm.16b", + [(set (v16i8 VPR128:$Rd), + (v16i8 (opnode (v16i8 VPR128:$Rn), (v16i8 VPR128:$Rm))))], + NoItinerary>; + } +} + +multiclass NeonI_3VSame_BHSD_sizes<bit u, bits<5> opcode, + string asmop, SDPatternOperator opnode, + bit Commutable = 0> + : NeonI_3VSame_BHS_sizes<u, opcode, asmop, opnode, Commutable> { + let isCommutable = Commutable in { + def _2D : NeonI_3VSame<0b1, u, 0b11, opcode, + (outs VPR128:$Rd), (ins VPR128:$Rn, VPR128:$Rm), + asmop # "\t$Rd.2d, $Rn.2d, $Rm.2d", + [(set (v2i64 VPR128:$Rd), + (v2i64 (opnode (v2i64 VPR128:$Rn), (v2i64 VPR128:$Rm))))], + NoItinerary>; + } +} + +// Multiclass NeonI_3VSame_SD_sizes: Operand types are floating point types, +// but Result types can be integer or floating point types. +multiclass NeonI_3VSame_SD_sizes<bit u, bit size, bits<5> opcode, + string asmop, SDPatternOperator opnode2S, + SDPatternOperator opnode4S, + SDPatternOperator opnode2D, + ValueType ResTy2S, ValueType ResTy4S, + ValueType ResTy2D, bit Commutable = 0> { + let isCommutable = Commutable in { + def _2S : NeonI_3VSame<0b0, u, {size, 0b0}, opcode, + (outs VPR64:$Rd), (ins VPR64:$Rn, VPR64:$Rm), + asmop # "\t$Rd.2s, $Rn.2s, $Rm.2s", + [(set (ResTy2S VPR64:$Rd), + (ResTy2S (opnode2S (v2f32 VPR64:$Rn), (v2f32 VPR64:$Rm))))], + NoItinerary>; + + def _4S : NeonI_3VSame<0b1, u, {size, 0b0}, opcode, + (outs VPR128:$Rd), (ins VPR128:$Rn, VPR128:$Rm), + asmop # "\t$Rd.4s, $Rn.4s, $Rm.4s", + [(set (ResTy4S VPR128:$Rd), + (ResTy4S (opnode4S (v4f32 VPR128:$Rn), (v4f32 VPR128:$Rm))))], + NoItinerary>; + + def _2D : NeonI_3VSame<0b1, u, {size, 0b1}, opcode, + (outs VPR128:$Rd), (ins VPR128:$Rn, VPR128:$Rm), + asmop # "\t$Rd.2d, $Rn.2d, $Rm.2d", + [(set (ResTy2D VPR128:$Rd), + (ResTy2D (opnode2D (v2f64 VPR128:$Rn), (v2f64 VPR128:$Rm))))], + NoItinerary>; + } +} + +//===----------------------------------------------------------------------===// +// Instruction Definitions +//===----------------------------------------------------------------------===// + +// Vector Arithmetic Instructions + +// Vector Add (Integer and Floating-Point) + +defm ADDvvv : NeonI_3VSame_BHSD_sizes<0b0, 0b10000, "add", add, 1>; +defm FADDvvv : NeonI_3VSame_SD_sizes<0b0, 0b0, 0b11010, "fadd", fadd, fadd, fadd, + v2f32, v4f32, v2f64, 1>; + +// Vector Sub (Integer and Floating-Point) + +defm SUBvvv : NeonI_3VSame_BHSD_sizes<0b1, 0b10000, "sub", sub, 0>; +defm FSUBvvv : NeonI_3VSame_SD_sizes<0b0, 0b1, 0b11010, "fsub", fsub, fsub, fsub, + v2f32, v4f32, v2f64, 0>; + +// Vector Multiply (Integer and Floating-Point) + +defm MULvvv : NeonI_3VSame_BHS_sizes<0b0, 0b10011, "mul", mul, 1>; +defm FMULvvv : NeonI_3VSame_SD_sizes<0b1, 0b0, 0b11011, "fmul", fmul, fmul, fmul, + v2f32, v4f32, v2f64, 1>; + +// Vector Multiply (Polynomial) + +defm PMULvvv : NeonI_3VSame_B_sizes<0b1, 0b00, 0b10011, "pmul", + int_arm_neon_vmulp, int_arm_neon_vmulp, 1>; + +// Vector Multiply-accumulate and Multiply-subtract (Integer) + +// class NeonI_3VSame_Constraint_impl: NeonI_3VSame with no data type and +// two operands constraints. +class NeonI_3VSame_Constraint_impl<string asmop, string asmlane, + RegisterOperand VPRC, ValueType OpTy, bit q, bit u, bits<2> size, + bits<5> opcode, SDPatternOperator opnode> + : NeonI_3VSame<q, u, size, opcode, + (outs VPRC:$Rd), (ins VPRC:$src, VPRC:$Rn, VPRC:$Rm), + asmop # "\t$Rd" # asmlane # ", $Rn" # asmlane # ", $Rm" # asmlane, + [(set (OpTy VPRC:$Rd), + (OpTy (opnode (OpTy VPRC:$src), (OpTy VPRC:$Rn), (OpTy VPRC:$Rm))))], + NoItinerary> { + let Constraints = "$src = $Rd"; +} + +def Neon_mla : PatFrag<(ops node:$Ra, node:$Rn, node:$Rm), + (add node:$Ra, (mul node:$Rn, node:$Rm))>; + +def Neon_mls : PatFrag<(ops node:$Ra, node:$Rn, node:$Rm), + (sub node:$Ra, (mul node:$Rn, node:$Rm))>; + + +def MLAvvv_8B: NeonI_3VSame_Constraint_impl<"mla", ".8b", VPR64, v8i8, + 0b0, 0b0, 0b00, 0b10010, Neon_mla>; +def MLAvvv_16B: NeonI_3VSame_Constraint_impl<"mla", ".16b", VPR128, v16i8, + 0b1, 0b0, 0b00, 0b10010, Neon_mla>; +def MLAvvv_4H: NeonI_3VSame_Constraint_impl<"mla", ".4h", VPR64, v4i16, + 0b0, 0b0, 0b01, 0b10010, Neon_mla>; +def MLAvvv_8H: NeonI_3VSame_Constraint_impl<"mla", ".8h", VPR128, v8i16, + 0b1, 0b0, 0b01, 0b10010, Neon_mla>; +def MLAvvv_2S: NeonI_3VSame_Constraint_impl<"mla", ".2s", VPR64, v2i32, + 0b0, 0b0, 0b10, 0b10010, Neon_mla>; +def MLAvvv_4S: NeonI_3VSame_Constraint_impl<"mla", ".4s", VPR128, v4i32, + 0b1, 0b0, 0b10, 0b10010, Neon_mla>; + +def MLSvvv_8B: NeonI_3VSame_Constraint_impl<"mls", ".8b", VPR64, v8i8, + 0b0, 0b1, 0b00, 0b10010, Neon_mls>; +def MLSvvv_16B: NeonI_3VSame_Constraint_impl<"mls", ".16b", VPR128, v16i8, + 0b1, 0b1, 0b00, 0b10010, Neon_mls>; +def MLSvvv_4H: NeonI_3VSame_Constraint_impl<"mls", ".4h", VPR64, v4i16, + 0b0, 0b1, 0b01, 0b10010, Neon_mls>; +def MLSvvv_8H: NeonI_3VSame_Constraint_impl<"mls", ".8h", VPR128, v8i16, + 0b1, 0b1, 0b01, 0b10010, Neon_mls>; +def MLSvvv_2S: NeonI_3VSame_Constraint_impl<"mls", ".2s", VPR64, v2i32, + 0b0, 0b1, 0b10, 0b10010, Neon_mls>; +def MLSvvv_4S: NeonI_3VSame_Constraint_impl<"mls", ".4s", VPR128, v4i32, + 0b1, 0b1, 0b10, 0b10010, Neon_mls>; + +// Vector Multiply-accumulate and Multiply-subtract (Floating Point) + +def Neon_fmla : PatFrag<(ops node:$Ra, node:$Rn, node:$Rm), + (fadd node:$Ra, (fmul node:$Rn, node:$Rm))>; + +def Neon_fmls : PatFrag<(ops node:$Ra, node:$Rn, node:$Rm), + (fsub node:$Ra, (fmul node:$Rn, node:$Rm))>; + +let Predicates = [HasNEON, UseFusedMAC] in { +def FMLAvvv_2S: NeonI_3VSame_Constraint_impl<"fmla", ".2s", VPR64, v2f32, + 0b0, 0b0, 0b00, 0b11001, Neon_fmla>; +def FMLAvvv_4S: NeonI_3VSame_Constraint_impl<"fmla", ".4s", VPR128, v4f32, + 0b1, 0b0, 0b00, 0b11001, Neon_fmla>; +def FMLAvvv_2D: NeonI_3VSame_Constraint_impl<"fmla", ".2d", VPR128, v2f64, + 0b1, 0b0, 0b01, 0b11001, Neon_fmla>; + +def FMLSvvv_2S: NeonI_3VSame_Constraint_impl<"fmls", ".2s", VPR64, v2f32, + 0b0, 0b0, 0b10, 0b11001, Neon_fmls>; +def FMLSvvv_4S: NeonI_3VSame_Constraint_impl<"fmls", ".4s", VPR128, v4f32, + 0b1, 0b0, 0b10, 0b11001, Neon_fmls>; +def FMLSvvv_2D: NeonI_3VSame_Constraint_impl<"fmls", ".2d", VPR128, v2f64, + 0b1, 0b0, 0b11, 0b11001, Neon_fmls>; +} + +// We're also allowed to match the fma instruction regardless of compile +// options. +def : Pat<(v2f32 (fma VPR64:$Rn, VPR64:$Rm, VPR64:$Ra)), + (FMLAvvv_2S VPR64:$Ra, VPR64:$Rn, VPR64:$Rm)>; +def : Pat<(v4f32 (fma VPR128:$Rn, VPR128:$Rm, VPR128:$Ra)), + (FMLAvvv_4S VPR128:$Ra, VPR128:$Rn, VPR128:$Rm)>; +def : Pat<(v2f64 (fma VPR128:$Rn, VPR128:$Rm, VPR128:$Ra)), + (FMLAvvv_2D VPR128:$Ra, VPR128:$Rn, VPR128:$Rm)>; + +def : Pat<(v2f32 (fma (fneg VPR64:$Rn), VPR64:$Rm, VPR64:$Ra)), + (FMLSvvv_2S VPR64:$Ra, VPR64:$Rn, VPR64:$Rm)>; +def : Pat<(v4f32 (fma (fneg VPR128:$Rn), VPR128:$Rm, VPR128:$Ra)), + (FMLSvvv_4S VPR128:$Ra, VPR128:$Rn, VPR128:$Rm)>; +def : Pat<(v2f64 (fma (fneg VPR128:$Rn), VPR128:$Rm, VPR128:$Ra)), + (FMLSvvv_2D VPR128:$Ra, VPR128:$Rn, VPR128:$Rm)>; + +// Vector Divide (Floating-Point) + +defm FDIVvvv : NeonI_3VSame_SD_sizes<0b1, 0b0, 0b11111, "fdiv", fdiv, fdiv, fdiv, + v2f32, v4f32, v2f64, 0>; + +// Vector Bitwise Operations + +// Vector Bitwise AND + +defm ANDvvv : NeonI_3VSame_B_sizes<0b0, 0b00, 0b00011, "and", and, and, 1>; + +// Vector Bitwise Exclusive OR + +defm EORvvv : NeonI_3VSame_B_sizes<0b1, 0b00, 0b00011, "eor", xor, xor, 1>; + +// Vector Bitwise OR + +defm ORRvvv : NeonI_3VSame_B_sizes<0b0, 0b10, 0b00011, "orr", or, or, 1>; + +// ORR disassembled as MOV if Vn==Vm + +// Vector Move - register +// Alias for ORR if Vn=Vm. +// FIXME: This is actually the preferred syntax but TableGen can't deal with +// custom printing of aliases. +def : NeonInstAlias<"mov $Rd.8b, $Rn.8b", + (ORRvvv_8B VPR64:$Rd, VPR64:$Rn, VPR64:$Rn), 0>; +def : NeonInstAlias<"mov $Rd.16b, $Rn.16b", + (ORRvvv_16B VPR128:$Rd, VPR128:$Rn, VPR128:$Rn), 0>; + +// The MOVI instruction takes two immediate operands. The first is the +// immediate encoding, while the second is the cmode. A cmode of 14, or +// 0b1110, produces a MOVI operation, rather than a MVNI, ORR, or BIC. +def Neon_AllZero : PatFrag<(ops), (Neon_movi (i32 0), (i32 14))>; +def Neon_AllOne : PatFrag<(ops), (Neon_movi (i32 255), (i32 14))>; + +def Neon_not8B : PatFrag<(ops node:$in), + (xor node:$in, (bitconvert (v8i8 Neon_AllOne)))>; +def Neon_not16B : PatFrag<(ops node:$in), + (xor node:$in, (bitconvert (v16i8 Neon_AllOne)))>; + +def Neon_orn8B : PatFrag<(ops node:$Rn, node:$Rm), + (or node:$Rn, (Neon_not8B node:$Rm))>; + +def Neon_orn16B : PatFrag<(ops node:$Rn, node:$Rm), + (or node:$Rn, (Neon_not16B node:$Rm))>; + +def Neon_bic8B : PatFrag<(ops node:$Rn, node:$Rm), + (and node:$Rn, (Neon_not8B node:$Rm))>; + +def Neon_bic16B : PatFrag<(ops node:$Rn, node:$Rm), + (and node:$Rn, (Neon_not16B node:$Rm))>; + + +// Vector Bitwise OR NOT - register + +defm ORNvvv : NeonI_3VSame_B_sizes<0b0, 0b11, 0b00011, "orn", + Neon_orn8B, Neon_orn16B, 0>; + +// Vector Bitwise Bit Clear (AND NOT) - register + +defm BICvvv : NeonI_3VSame_B_sizes<0b0, 0b01, 0b00011, "bic", + Neon_bic8B, Neon_bic16B, 0>; + +multiclass Neon_bitwise2V_patterns<SDPatternOperator opnode8B, + SDPatternOperator opnode16B, + Instruction INST8B, + Instruction INST16B> { + def : Pat<(v2i32 (opnode8B VPR64:$Rn, VPR64:$Rm)), + (INST8B VPR64:$Rn, VPR64:$Rm)>; + def : Pat<(v4i16 (opnode8B VPR64:$Rn, VPR64:$Rm)), + (INST8B VPR64:$Rn, VPR64:$Rm)>; + def : Pat<(v1i64 (opnode8B VPR64:$Rn, VPR64:$Rm)), + (INST8B VPR64:$Rn, VPR64:$Rm)>; + def : Pat<(v4i32 (opnode16B VPR128:$Rn, VPR128:$Rm)), + (INST16B VPR128:$Rn, VPR128:$Rm)>; + def : Pat<(v8i16 (opnode16B VPR128:$Rn, VPR128:$Rm)), + (INST16B VPR128:$Rn, VPR128:$Rm)>; + def : Pat<(v2i64 (opnode16B VPR128:$Rn, VPR128:$Rm)), + (INST16B VPR128:$Rn, VPR128:$Rm)>; +} + +// Additional patterns for bitwise instructions AND, EOR, ORR, BIC, ORN +defm : Neon_bitwise2V_patterns<and, and, ANDvvv_8B, ANDvvv_16B>; +defm : Neon_bitwise2V_patterns<or, or, ORRvvv_8B, ORRvvv_16B>; +defm : Neon_bitwise2V_patterns<xor, xor, EORvvv_8B, EORvvv_16B>; +defm : Neon_bitwise2V_patterns<Neon_bic8B, Neon_bic16B, BICvvv_8B, BICvvv_16B>; +defm : Neon_bitwise2V_patterns<Neon_orn8B, Neon_orn16B, ORNvvv_8B, ORNvvv_16B>; + +// Vector Bitwise Select +def BSLvvv_8B : NeonI_3VSame_Constraint_impl<"bsl", ".8b", VPR64, v8i8, + 0b0, 0b1, 0b01, 0b00011, Neon_bsl>; + +def BSLvvv_16B : NeonI_3VSame_Constraint_impl<"bsl", ".16b", VPR128, v16i8, + 0b1, 0b1, 0b01, 0b00011, Neon_bsl>; + +multiclass Neon_bitwise3V_patterns<SDPatternOperator opnode, + Instruction INST8B, + Instruction INST16B> { + // Disassociate type from instruction definition + def : Pat<(v2i32 (opnode VPR64:$src,VPR64:$Rn, VPR64:$Rm)), + (INST8B VPR64:$src, VPR64:$Rn, VPR64:$Rm)>; + def : Pat<(v4i16 (opnode VPR64:$src, VPR64:$Rn, VPR64:$Rm)), + (INST8B VPR64:$src, VPR64:$Rn, VPR64:$Rm)>; + def : Pat<(v1i64 (opnode VPR64:$src, VPR64:$Rn, VPR64:$Rm)), + (INST8B VPR64:$src, VPR64:$Rn, VPR64:$Rm)>; + def : Pat<(v4i32 (opnode VPR128:$src, VPR128:$Rn, VPR128:$Rm)), + (INST16B VPR128:$src, VPR128:$Rn, VPR128:$Rm)>; + def : Pat<(v8i16 (opnode VPR128:$src, VPR128:$Rn, VPR128:$Rm)), + (INST16B VPR128:$src, VPR128:$Rn, VPR128:$Rm)>; + def : Pat<(v2i64 (opnode VPR128:$src, VPR128:$Rn, VPR128:$Rm)), + (INST16B VPR128:$src, VPR128:$Rn, VPR128:$Rm)>; + + // Allow to match BSL instruction pattern with non-constant operand + def : Pat<(v8i8 (or (and VPR64:$Rn, VPR64:$Rd), + (and VPR64:$Rm, (Neon_not8B VPR64:$Rd)))), + (INST8B VPR64:$Rd, VPR64:$Rn, VPR64:$Rm)>; + def : Pat<(v4i16 (or (and VPR64:$Rn, VPR64:$Rd), + (and VPR64:$Rm, (Neon_not8B VPR64:$Rd)))), + (INST8B VPR64:$Rd, VPR64:$Rn, VPR64:$Rm)>; + def : Pat<(v2i32 (or (and VPR64:$Rn, VPR64:$Rd), + (and VPR64:$Rm, (Neon_not8B VPR64:$Rd)))), + (INST8B VPR64:$Rd, VPR64:$Rn, VPR64:$Rm)>; + def : Pat<(v1i64 (or (and VPR64:$Rn, VPR64:$Rd), + (and VPR64:$Rm, (Neon_not8B VPR64:$Rd)))), + (INST8B VPR64:$Rd, VPR64:$Rn, VPR64:$Rm)>; + def : Pat<(v16i8 (or (and VPR128:$Rn, VPR128:$Rd), + (and VPR128:$Rm, (Neon_not16B VPR128:$Rd)))), + (INST16B VPR128:$Rd, VPR128:$Rn, VPR128:$Rm)>; + def : Pat<(v8i16 (or (and VPR128:$Rn, VPR128:$Rd), + (and VPR128:$Rm, (Neon_not16B VPR128:$Rd)))), + (INST16B VPR128:$Rd, VPR128:$Rn, VPR128:$Rm)>; + def : Pat<(v4i32 (or (and VPR128:$Rn, VPR128:$Rd), + (and VPR128:$Rm, (Neon_not16B VPR128:$Rd)))), + (INST16B VPR128:$Rd, VPR128:$Rn, VPR128:$Rm)>; + def : Pat<(v2i64 (or (and VPR128:$Rn, VPR128:$Rd), + (and VPR128:$Rm, (Neon_not16B VPR128:$Rd)))), + (INST16B VPR128:$Rd, VPR128:$Rn, VPR128:$Rm)>; + + // Allow to match llvm.arm.* intrinsics. + def : Pat<(v8i8 (int_arm_neon_vbsl (v8i8 VPR64:$src), + (v8i8 VPR64:$Rn), (v8i8 VPR64:$Rm))), + (INST8B VPR64:$src, VPR64:$Rn, VPR64:$Rm)>; + def : Pat<(v4i16 (int_arm_neon_vbsl (v4i16 VPR64:$src), + (v4i16 VPR64:$Rn), (v4i16 VPR64:$Rm))), + (INST8B VPR64:$src, VPR64:$Rn, VPR64:$Rm)>; + def : Pat<(v2i32 (int_arm_neon_vbsl (v2i32 VPR64:$src), + (v2i32 VPR64:$Rn), (v2i32 VPR64:$Rm))), + (INST8B VPR64:$src, VPR64:$Rn, VPR64:$Rm)>; + def : Pat<(v1i64 (int_arm_neon_vbsl (v1i64 VPR64:$src), + (v1i64 VPR64:$Rn), (v1i64 VPR64:$Rm))), + (INST8B VPR64:$src, VPR64:$Rn, VPR64:$Rm)>; + def : Pat<(v2f32 (int_arm_neon_vbsl (v2f32 VPR64:$src), + (v2f32 VPR64:$Rn), (v2f32 VPR64:$Rm))), + (INST8B VPR64:$src, VPR64:$Rn, VPR64:$Rm)>; + def : Pat<(v1f64 (int_arm_neon_vbsl (v1f64 VPR64:$src), + (v1f64 VPR64:$Rn), (v1f64 VPR64:$Rm))), + (INST8B VPR64:$src, VPR64:$Rn, VPR64:$Rm)>; + def : Pat<(v16i8 (int_arm_neon_vbsl (v16i8 VPR128:$src), + (v16i8 VPR128:$Rn), (v16i8 VPR128:$Rm))), + (INST16B VPR128:$src, VPR128:$Rn, VPR128:$Rm)>; + def : Pat<(v8i16 (int_arm_neon_vbsl (v8i16 VPR128:$src), + (v8i16 VPR128:$Rn), (v8i16 VPR128:$Rm))), + (INST16B VPR128:$src, VPR128:$Rn, VPR128:$Rm)>; + def : Pat<(v4i32 (int_arm_neon_vbsl (v4i32 VPR128:$src), + (v4i32 VPR128:$Rn), (v4i32 VPR128:$Rm))), + (INST16B VPR128:$src, VPR128:$Rn, VPR128:$Rm)>; + def : Pat<(v2i64 (int_arm_neon_vbsl (v2i64 VPR128:$src), + (v2i64 VPR128:$Rn), (v2i64 VPR128:$Rm))), + (INST16B VPR128:$src, VPR128:$Rn, VPR128:$Rm)>; + def : Pat<(v4f32 (int_arm_neon_vbsl (v4f32 VPR128:$src), + (v4f32 VPR128:$Rn), (v4f32 VPR128:$Rm))), + (INST16B VPR128:$src, VPR128:$Rn, VPR128:$Rm)>; + def : Pat<(v2f64 (int_arm_neon_vbsl (v2f64 VPR128:$src), + (v2f64 VPR128:$Rn), (v2f64 VPR128:$Rm))), + (INST16B VPR128:$src, VPR128:$Rn, VPR128:$Rm)>; +} + +// Additional patterns for bitwise instruction BSL +defm: Neon_bitwise3V_patterns<Neon_bsl, BSLvvv_8B, BSLvvv_16B>; + +def Neon_NoBSLop : PatFrag<(ops node:$src, node:$Rn, node:$Rm), + (Neon_bsl node:$src, node:$Rn, node:$Rm), + [{ (void)N; return false; }]>; + +// Vector Bitwise Insert if True + +def BITvvv_8B : NeonI_3VSame_Constraint_impl<"bit", ".8b", VPR64, v8i8, + 0b0, 0b1, 0b10, 0b00011, Neon_NoBSLop>; +def BITvvv_16B : NeonI_3VSame_Constraint_impl<"bit", ".16b", VPR128, v16i8, + 0b1, 0b1, 0b10, 0b00011, Neon_NoBSLop>; + +// Vector Bitwise Insert if False + +def BIFvvv_8B : NeonI_3VSame_Constraint_impl<"bif", ".8b", VPR64, v8i8, + 0b0, 0b1, 0b11, 0b00011, Neon_NoBSLop>; +def BIFvvv_16B : NeonI_3VSame_Constraint_impl<"bif", ".16b", VPR128, v16i8, + 0b1, 0b1, 0b11, 0b00011, Neon_NoBSLop>; + +// Vector Absolute Difference and Accumulate (Signed, Unsigned) + +def Neon_uaba : PatFrag<(ops node:$Ra, node:$Rn, node:$Rm), + (add node:$Ra, (int_arm_neon_vabdu node:$Rn, node:$Rm))>; +def Neon_saba : PatFrag<(ops node:$Ra, node:$Rn, node:$Rm), + (add node:$Ra, (int_arm_neon_vabds node:$Rn, node:$Rm))>; + +// Vector Absolute Difference and Accumulate (Unsigned) +def UABAvvv_8B : NeonI_3VSame_Constraint_impl<"uaba", ".8b", VPR64, v8i8, + 0b0, 0b1, 0b00, 0b01111, Neon_uaba>; +def UABAvvv_16B : NeonI_3VSame_Constraint_impl<"uaba", ".16b", VPR128, v16i8, + 0b1, 0b1, 0b00, 0b01111, Neon_uaba>; +def UABAvvv_4H : NeonI_3VSame_Constraint_impl<"uaba", ".4h", VPR64, v4i16, + 0b0, 0b1, 0b01, 0b01111, Neon_uaba>; +def UABAvvv_8H : NeonI_3VSame_Constraint_impl<"uaba", ".8h", VPR128, v8i16, + 0b1, 0b1, 0b01, 0b01111, Neon_uaba>; +def UABAvvv_2S : NeonI_3VSame_Constraint_impl<"uaba", ".2s", VPR64, v2i32, + 0b0, 0b1, 0b10, 0b01111, Neon_uaba>; +def UABAvvv_4S : NeonI_3VSame_Constraint_impl<"uaba", ".4s", VPR128, v4i32, + 0b1, 0b1, 0b10, 0b01111, Neon_uaba>; + +// Vector Absolute Difference and Accumulate (Signed) +def SABAvvv_8B : NeonI_3VSame_Constraint_impl<"saba", ".8b", VPR64, v8i8, + 0b0, 0b0, 0b00, 0b01111, Neon_saba>; +def SABAvvv_16B : NeonI_3VSame_Constraint_impl<"saba", ".16b", VPR128, v16i8, + 0b1, 0b0, 0b00, 0b01111, Neon_saba>; +def SABAvvv_4H : NeonI_3VSame_Constraint_impl<"saba", ".4h", VPR64, v4i16, + 0b0, 0b0, 0b01, 0b01111, Neon_saba>; +def SABAvvv_8H : NeonI_3VSame_Constraint_impl<"saba", ".8h", VPR128, v8i16, + 0b1, 0b0, 0b01, 0b01111, Neon_saba>; +def SABAvvv_2S : NeonI_3VSame_Constraint_impl<"saba", ".2s", VPR64, v2i32, + 0b0, 0b0, 0b10, 0b01111, Neon_saba>; +def SABAvvv_4S : NeonI_3VSame_Constraint_impl<"saba", ".4s", VPR128, v4i32, + 0b1, 0b0, 0b10, 0b01111, Neon_saba>; + + +// Vector Absolute Difference (Signed, Unsigned) +defm UABDvvv : NeonI_3VSame_BHS_sizes<0b1, 0b01110, "uabd", int_arm_neon_vabdu, 0>; +defm SABDvvv : NeonI_3VSame_BHS_sizes<0b0, 0b01110, "sabd", int_arm_neon_vabds, 0>; + +// Vector Absolute Difference (Floating Point) +defm FABDvvv: NeonI_3VSame_SD_sizes<0b1, 0b1, 0b11010, "fabd", + int_arm_neon_vabds, int_arm_neon_vabds, + int_arm_neon_vabds, v2f32, v4f32, v2f64, 0>; + +// Vector Reciprocal Step (Floating Point) +defm FRECPSvvv : NeonI_3VSame_SD_sizes<0b0, 0b0, 0b11111, "frecps", + int_arm_neon_vrecps, int_arm_neon_vrecps, + int_arm_neon_vrecps, + v2f32, v4f32, v2f64, 0>; + +// Vector Reciprocal Square Root Step (Floating Point) +defm FRSQRTSvvv : NeonI_3VSame_SD_sizes<0b0, 0b1, 0b11111, "frsqrts", + int_arm_neon_vrsqrts, + int_arm_neon_vrsqrts, + int_arm_neon_vrsqrts, + v2f32, v4f32, v2f64, 0>; + +// Vector Comparisons + +def Neon_cmeq : PatFrag<(ops node:$lhs, node:$rhs), + (Neon_cmp node:$lhs, node:$rhs, SETEQ)>; +def Neon_cmphs : PatFrag<(ops node:$lhs, node:$rhs), + (Neon_cmp node:$lhs, node:$rhs, SETUGE)>; +def Neon_cmge : PatFrag<(ops node:$lhs, node:$rhs), + (Neon_cmp node:$lhs, node:$rhs, SETGE)>; +def Neon_cmhi : PatFrag<(ops node:$lhs, node:$rhs), + (Neon_cmp node:$lhs, node:$rhs, SETUGT)>; +def Neon_cmgt : PatFrag<(ops node:$lhs, node:$rhs), + (Neon_cmp node:$lhs, node:$rhs, SETGT)>; + +// NeonI_compare_aliases class: swaps register operands to implement +// comparison aliases, e.g., CMLE is alias for CMGE with operands reversed. +class NeonI_compare_aliases<string asmop, string asmlane, + Instruction inst, RegisterOperand VPRC> + : NeonInstAlias<asmop # "\t$Rd" # asmlane #", $Rn" # asmlane # + ", $Rm" # asmlane, + (inst VPRC:$Rd, VPRC:$Rm, VPRC:$Rn), 0b0>; + +// Vector Comparisons (Integer) + +// Vector Compare Mask Equal (Integer) +let isCommutable =1 in { +defm CMEQvvv : NeonI_3VSame_BHSD_sizes<0b1, 0b10001, "cmeq", Neon_cmeq, 0>; +} + +// Vector Compare Mask Higher or Same (Unsigned Integer) +defm CMHSvvv : NeonI_3VSame_BHSD_sizes<0b1, 0b00111, "cmhs", Neon_cmphs, 0>; + +// Vector Compare Mask Greater Than or Equal (Integer) +defm CMGEvvv : NeonI_3VSame_BHSD_sizes<0b0, 0b00111, "cmge", Neon_cmge, 0>; + +// Vector Compare Mask Higher (Unsigned Integer) +defm CMHIvvv : NeonI_3VSame_BHSD_sizes<0b1, 0b00110, "cmhi", Neon_cmhi, 0>; + +// Vector Compare Mask Greater Than (Integer) +defm CMGTvvv : NeonI_3VSame_BHSD_sizes<0b0, 0b00110, "cmgt", Neon_cmgt, 0>; + +// Vector Compare Mask Bitwise Test (Integer) +defm CMTSTvvv: NeonI_3VSame_BHSD_sizes<0b0, 0b10001, "cmtst", Neon_tst, 0>; + +// Vector Compare Mask Less or Same (Unsigned Integer) +// CMLS is alias for CMHS with operands reversed. +def CMLSvvv_8B : NeonI_compare_aliases<"cmls", ".8b", CMHSvvv_8B, VPR64>; +def CMLSvvv_16B : NeonI_compare_aliases<"cmls", ".16b", CMHSvvv_16B, VPR128>; +def CMLSvvv_4H : NeonI_compare_aliases<"cmls", ".4h", CMHSvvv_4H, VPR64>; +def CMLSvvv_8H : NeonI_compare_aliases<"cmls", ".8h", CMHSvvv_8H, VPR128>; +def CMLSvvv_2S : NeonI_compare_aliases<"cmls", ".2s", CMHSvvv_2S, VPR64>; +def CMLSvvv_4S : NeonI_compare_aliases<"cmls", ".4s", CMHSvvv_4S, VPR128>; +def CMLSvvv_2D : NeonI_compare_aliases<"cmls", ".2d", CMHSvvv_2D, VPR128>; + +// Vector Compare Mask Less Than or Equal (Integer) +// CMLE is alias for CMGE with operands reversed. +def CMLEvvv_8B : NeonI_compare_aliases<"cmle", ".8b", CMGEvvv_8B, VPR64>; +def CMLEvvv_16B : NeonI_compare_aliases<"cmle", ".16b", CMGEvvv_16B, VPR128>; +def CMLEvvv_4H : NeonI_compare_aliases<"cmle", ".4h", CMGEvvv_4H, VPR64>; +def CMLEvvv_8H : NeonI_compare_aliases<"cmle", ".8h", CMGEvvv_8H, VPR128>; +def CMLEvvv_2S : NeonI_compare_aliases<"cmle", ".2s", CMGEvvv_2S, VPR64>; +def CMLEvvv_4S : NeonI_compare_aliases<"cmle", ".4s", CMGEvvv_4S, VPR128>; +def CMLEvvv_2D : NeonI_compare_aliases<"cmle", ".2d", CMGEvvv_2D, VPR128>; + +// Vector Compare Mask Lower (Unsigned Integer) +// CMLO is alias for CMHI with operands reversed. +def CMLOvvv_8B : NeonI_compare_aliases<"cmlo", ".8b", CMHIvvv_8B, VPR64>; +def CMLOvvv_16B : NeonI_compare_aliases<"cmlo", ".16b", CMHIvvv_16B, VPR128>; +def CMLOvvv_4H : NeonI_compare_aliases<"cmlo", ".4h", CMHIvvv_4H, VPR64>; +def CMLOvvv_8H : NeonI_compare_aliases<"cmlo", ".8h", CMHIvvv_8H, VPR128>; +def CMLOvvv_2S : NeonI_compare_aliases<"cmlo", ".2s", CMHIvvv_2S, VPR64>; +def CMLOvvv_4S : NeonI_compare_aliases<"cmlo", ".4s", CMHIvvv_4S, VPR128>; +def CMLOvvv_2D : NeonI_compare_aliases<"cmlo", ".2d", CMHIvvv_2D, VPR128>; + +// Vector Compare Mask Less Than (Integer) +// CMLT is alias for CMGT with operands reversed. +def CMLTvvv_8B : NeonI_compare_aliases<"cmlt", ".8b", CMGTvvv_8B, VPR64>; +def CMLTvvv_16B : NeonI_compare_aliases<"cmlt", ".16b", CMGTvvv_16B, VPR128>; +def CMLTvvv_4H : NeonI_compare_aliases<"cmlt", ".4h", CMGTvvv_4H, VPR64>; +def CMLTvvv_8H : NeonI_compare_aliases<"cmlt", ".8h", CMGTvvv_8H, VPR128>; +def CMLTvvv_2S : NeonI_compare_aliases<"cmlt", ".2s", CMGTvvv_2S, VPR64>; +def CMLTvvv_4S : NeonI_compare_aliases<"cmlt", ".4s", CMGTvvv_4S, VPR128>; +def CMLTvvv_2D : NeonI_compare_aliases<"cmlt", ".2d", CMGTvvv_2D, VPR128>; + + +def neon_uimm0_asmoperand : AsmOperandClass +{ + let Name = "UImm0"; + let PredicateMethod = "isUImm<0>"; + let RenderMethod = "addImmOperands"; +} + +def neon_uimm0 : Operand<i32>, ImmLeaf<i32, [{return Imm == 0;}]> { + let ParserMatchClass = neon_uimm0_asmoperand; + let PrintMethod = "printNeonUImm0Operand"; + +} + +multiclass NeonI_cmpz_sizes<bit u, bits<5> opcode, string asmop, CondCode CC> +{ + def _8B : NeonI_2VMisc<0b0, u, 0b00, opcode, + (outs VPR64:$Rd), (ins VPR64:$Rn, neon_uimm0:$Imm), + asmop # "\t$Rd.8b, $Rn.8b, $Imm", + [(set (v8i8 VPR64:$Rd), + (v8i8 (Neon_cmpz (v8i8 VPR64:$Rn), (i32 imm:$Imm), CC)))], + NoItinerary>; + + def _16B : NeonI_2VMisc<0b1, u, 0b00, opcode, + (outs VPR128:$Rd), (ins VPR128:$Rn, neon_uimm0:$Imm), + asmop # "\t$Rd.16b, $Rn.16b, $Imm", + [(set (v16i8 VPR128:$Rd), + (v16i8 (Neon_cmpz (v16i8 VPR128:$Rn), (i32 imm:$Imm), CC)))], + NoItinerary>; + + def _4H : NeonI_2VMisc<0b0, u, 0b01, opcode, + (outs VPR64:$Rd), (ins VPR64:$Rn, neon_uimm0:$Imm), + asmop # "\t$Rd.4h, $Rn.4h, $Imm", + [(set (v4i16 VPR64:$Rd), + (v4i16 (Neon_cmpz (v4i16 VPR64:$Rn), (i32 imm:$Imm), CC)))], + NoItinerary>; + + def _8H : NeonI_2VMisc<0b1, u, 0b01, opcode, + (outs VPR128:$Rd), (ins VPR128:$Rn, neon_uimm0:$Imm), + asmop # "\t$Rd.8h, $Rn.8h, $Imm", + [(set (v8i16 VPR128:$Rd), + (v8i16 (Neon_cmpz (v8i16 VPR128:$Rn), (i32 imm:$Imm), CC)))], + NoItinerary>; + + def _2S : NeonI_2VMisc<0b0, u, 0b10, opcode, + (outs VPR64:$Rd), (ins VPR64:$Rn, neon_uimm0:$Imm), + asmop # "\t$Rd.2s, $Rn.2s, $Imm", + [(set (v2i32 VPR64:$Rd), + (v2i32 (Neon_cmpz (v2i32 VPR64:$Rn), (i32 imm:$Imm), CC)))], + NoItinerary>; + + def _4S : NeonI_2VMisc<0b1, u, 0b10, opcode, + (outs VPR128:$Rd), (ins VPR128:$Rn, neon_uimm0:$Imm), + asmop # "\t$Rd.4s, $Rn.4s, $Imm", + [(set (v4i32 VPR128:$Rd), + (v4i32 (Neon_cmpz (v4i32 VPR128:$Rn), (i32 imm:$Imm), CC)))], + NoItinerary>; + + def _2D : NeonI_2VMisc<0b1, u, 0b11, opcode, + (outs VPR128:$Rd), (ins VPR128:$Rn, neon_uimm0:$Imm), + asmop # "\t$Rd.2d, $Rn.2d, $Imm", + [(set (v2i64 VPR128:$Rd), + (v2i64 (Neon_cmpz (v2i64 VPR128:$Rn), (i32 imm:$Imm), CC)))], + NoItinerary>; +} + +// Vector Compare Mask Equal to Zero (Integer) +defm CMEQvvi : NeonI_cmpz_sizes<0b0, 0b01001, "cmeq", SETEQ>; + +// Vector Compare Mask Greater Than or Equal to Zero (Signed Integer) +defm CMGEvvi : NeonI_cmpz_sizes<0b1, 0b01000, "cmge", SETGE>; + +// Vector Compare Mask Greater Than Zero (Signed Integer) +defm CMGTvvi : NeonI_cmpz_sizes<0b0, 0b01000, "cmgt", SETGT>; + +// Vector Compare Mask Less Than or Equal To Zero (Signed Integer) +defm CMLEvvi : NeonI_cmpz_sizes<0b1, 0b01001, "cmle", SETLE>; + +// Vector Compare Mask Less Than Zero (Signed Integer) +defm CMLTvvi : NeonI_cmpz_sizes<0b0, 0b01010, "cmlt", SETLT>; + +// Vector Comparisons (Floating Point) + +// Vector Compare Mask Equal (Floating Point) +let isCommutable =1 in { +defm FCMEQvvv : NeonI_3VSame_SD_sizes<0b0, 0b0, 0b11100, "fcmeq", Neon_cmeq, + Neon_cmeq, Neon_cmeq, + v2i32, v4i32, v2i64, 0>; +} + +// Vector Compare Mask Greater Than Or Equal (Floating Point) +defm FCMGEvvv : NeonI_3VSame_SD_sizes<0b1, 0b0, 0b11100, "fcmge", Neon_cmge, + Neon_cmge, Neon_cmge, + v2i32, v4i32, v2i64, 0>; + +// Vector Compare Mask Greater Than (Floating Point) +defm FCMGTvvv : NeonI_3VSame_SD_sizes<0b1, 0b1, 0b11100, "fcmgt", Neon_cmgt, + Neon_cmgt, Neon_cmgt, + v2i32, v4i32, v2i64, 0>; + +// Vector Compare Mask Less Than Or Equal (Floating Point) +// FCMLE is alias for FCMGE with operands reversed. +def FCMLEvvv_2S : NeonI_compare_aliases<"fcmle", ".2s", FCMGEvvv_2S, VPR64>; +def FCMLEvvv_4S : NeonI_compare_aliases<"fcmle", ".4s", FCMGEvvv_4S, VPR128>; +def FCMLEvvv_2D : NeonI_compare_aliases<"fcmle", ".2d", FCMGEvvv_2D, VPR128>; + +// Vector Compare Mask Less Than (Floating Point) +// FCMLT is alias for FCMGT with operands reversed. +def FCMLTvvv_2S : NeonI_compare_aliases<"fcmlt", ".2s", FCMGTvvv_2S, VPR64>; +def FCMLTvvv_4S : NeonI_compare_aliases<"fcmlt", ".4s", FCMGTvvv_4S, VPR128>; +def FCMLTvvv_2D : NeonI_compare_aliases<"fcmlt", ".2d", FCMGTvvv_2D, VPR128>; + + +multiclass NeonI_fpcmpz_sizes<bit u, bit size, bits<5> opcode, + string asmop, CondCode CC> +{ + def _2S : NeonI_2VMisc<0b0, u, {size, 0b0}, opcode, + (outs VPR64:$Rd), (ins VPR64:$Rn, fpz32:$FPImm), + asmop # "\t$Rd.2s, $Rn.2s, $FPImm", + [(set (v2i32 VPR64:$Rd), + (v2i32 (Neon_cmpz (v2f32 VPR64:$Rn), (f32 fpimm:$FPImm), CC)))], + NoItinerary>; + + def _4S : NeonI_2VMisc<0b1, u, {size, 0b0}, opcode, + (outs VPR128:$Rd), (ins VPR128:$Rn, fpz32:$FPImm), + asmop # "\t$Rd.4s, $Rn.4s, $FPImm", + [(set (v4i32 VPR128:$Rd), + (v4i32 (Neon_cmpz (v4f32 VPR128:$Rn), (f32 fpimm:$FPImm), CC)))], + NoItinerary>; + + def _2D : NeonI_2VMisc<0b1, u, {size, 0b1}, opcode, + (outs VPR128:$Rd), (ins VPR128:$Rn, fpz32:$FPImm), + asmop # "\t$Rd.2d, $Rn.2d, $FPImm", + [(set (v2i64 VPR128:$Rd), + (v2i64 (Neon_cmpz (v2f64 VPR128:$Rn), (f32 fpimm:$FPImm), CC)))], + NoItinerary>; +} + +// Vector Compare Mask Equal to Zero (Floating Point) +defm FCMEQvvi : NeonI_fpcmpz_sizes<0b0, 0b1, 0b01101, "fcmeq", SETEQ>; + +// Vector Compare Mask Greater Than or Equal to Zero (Floating Point) +defm FCMGEvvi : NeonI_fpcmpz_sizes<0b1, 0b1, 0b01100, "fcmge", SETGE>; + +// Vector Compare Mask Greater Than Zero (Floating Point) +defm FCMGTvvi : NeonI_fpcmpz_sizes<0b0, 0b1, 0b01100, "fcmgt", SETGT>; + +// Vector Compare Mask Less Than or Equal To Zero (Floating Point) +defm FCMLEvvi : NeonI_fpcmpz_sizes<0b1, 0b1, 0b01101, "fcmle", SETLE>; + +// Vector Compare Mask Less Than Zero (Floating Point) +defm FCMLTvvi : NeonI_fpcmpz_sizes<0b0, 0b1, 0b01110, "fcmlt", SETLT>; + +// Vector Absolute Comparisons (Floating Point) + +// Vector Absolute Compare Mask Greater Than Or Equal (Floating Point) +defm FACGEvvv : NeonI_3VSame_SD_sizes<0b1, 0b0, 0b11101, "facge", + int_arm_neon_vacged, int_arm_neon_vacgeq, + int_aarch64_neon_vacgeq, + v2i32, v4i32, v2i64, 0>; + +// Vector Absolute Compare Mask Greater Than (Floating Point) +defm FACGTvvv : NeonI_3VSame_SD_sizes<0b1, 0b1, 0b11101, "facgt", + int_arm_neon_vacgtd, int_arm_neon_vacgtq, + int_aarch64_neon_vacgtq, + v2i32, v4i32, v2i64, 0>; + +// Vector Absolute Compare Mask Less Than Or Equal (Floating Point) +// FACLE is alias for FACGE with operands reversed. +def FACLEvvv_2S : NeonI_compare_aliases<"facle", ".2s", FACGEvvv_2S, VPR64>; +def FACLEvvv_4S : NeonI_compare_aliases<"facle", ".4s", FACGEvvv_4S, VPR128>; +def FACLEvvv_2D : NeonI_compare_aliases<"facle", ".2d", FACGEvvv_2D, VPR128>; + +// Vector Absolute Compare Mask Less Than (Floating Point) +// FACLT is alias for FACGT with operands reversed. +def FACLTvvv_2S : NeonI_compare_aliases<"faclt", ".2s", FACGTvvv_2S, VPR64>; +def FACLTvvv_4S : NeonI_compare_aliases<"faclt", ".4s", FACGTvvv_4S, VPR128>; +def FACLTvvv_2D : NeonI_compare_aliases<"faclt", ".2d", FACGTvvv_2D, VPR128>; + +// Vector halving add (Integer Signed, Unsigned) +defm SHADDvvv : NeonI_3VSame_BHS_sizes<0b0, 0b00000, "shadd", + int_arm_neon_vhadds, 1>; +defm UHADDvvv : NeonI_3VSame_BHS_sizes<0b1, 0b00000, "uhadd", + int_arm_neon_vhaddu, 1>; + +// Vector halving sub (Integer Signed, Unsigned) +defm SHSUBvvv : NeonI_3VSame_BHS_sizes<0b0, 0b00100, "shsub", + int_arm_neon_vhsubs, 0>; +defm UHSUBvvv : NeonI_3VSame_BHS_sizes<0b1, 0b00100, "uhsub", + int_arm_neon_vhsubu, 0>; + +// Vector rouding halving add (Integer Signed, Unsigned) +defm SRHADDvvv : NeonI_3VSame_BHS_sizes<0b0, 0b00010, "srhadd", + int_arm_neon_vrhadds, 1>; +defm URHADDvvv : NeonI_3VSame_BHS_sizes<0b1, 0b00010, "urhadd", + int_arm_neon_vrhaddu, 1>; + +// Vector Saturating add (Integer Signed, Unsigned) +defm SQADDvvv : NeonI_3VSame_BHSD_sizes<0b0, 0b00001, "sqadd", + int_arm_neon_vqadds, 1>; +defm UQADDvvv : NeonI_3VSame_BHSD_sizes<0b1, 0b00001, "uqadd", + int_arm_neon_vqaddu, 1>; + +// Vector Saturating sub (Integer Signed, Unsigned) +defm SQSUBvvv : NeonI_3VSame_BHSD_sizes<0b0, 0b00101, "sqsub", + int_arm_neon_vqsubs, 1>; +defm UQSUBvvv : NeonI_3VSame_BHSD_sizes<0b1, 0b00101, "uqsub", + int_arm_neon_vqsubu, 1>; + +// Vector Shift Left (Signed and Unsigned Integer) +defm SSHLvvv : NeonI_3VSame_BHSD_sizes<0b0, 0b01000, "sshl", + int_arm_neon_vshifts, 1>; +defm USHLvvv : NeonI_3VSame_BHSD_sizes<0b1, 0b01000, "ushl", + int_arm_neon_vshiftu, 1>; + +// Vector Saturating Shift Left (Signed and Unsigned Integer) +defm SQSHLvvv : NeonI_3VSame_BHSD_sizes<0b0, 0b01001, "sqshl", + int_arm_neon_vqshifts, 1>; +defm UQSHLvvv : NeonI_3VSame_BHSD_sizes<0b1, 0b01001, "uqshl", + int_arm_neon_vqshiftu, 1>; + +// Vector Rouding Shift Left (Signed and Unsigned Integer) +defm SRSHLvvv : NeonI_3VSame_BHSD_sizes<0b0, 0b01010, "srshl", + int_arm_neon_vrshifts, 1>; +defm URSHLvvv : NeonI_3VSame_BHSD_sizes<0b1, 0b01010, "urshl", + int_arm_neon_vrshiftu, 1>; + +// Vector Saturating Rouding Shift Left (Signed and Unsigned Integer) +defm SQRSHLvvv : NeonI_3VSame_BHSD_sizes<0b0, 0b01011, "sqrshl", + int_arm_neon_vqrshifts, 1>; +defm UQRSHLvvv : NeonI_3VSame_BHSD_sizes<0b1, 0b01011, "uqrshl", + int_arm_neon_vqrshiftu, 1>; + +// Vector Maximum (Signed and Unsigned Integer) +defm SMAXvvv : NeonI_3VSame_BHS_sizes<0b0, 0b01100, "smax", int_arm_neon_vmaxs, 1>; +defm UMAXvvv : NeonI_3VSame_BHS_sizes<0b1, 0b01100, "umax", int_arm_neon_vmaxu, 1>; + +// Vector Minimum (Signed and Unsigned Integer) +defm SMINvvv : NeonI_3VSame_BHS_sizes<0b0, 0b01101, "smin", int_arm_neon_vmins, 1>; +defm UMINvvv : NeonI_3VSame_BHS_sizes<0b1, 0b01101, "umin", int_arm_neon_vminu, 1>; + +// Vector Maximum (Floating Point) +defm FMAXvvv : NeonI_3VSame_SD_sizes<0b0, 0b0, 0b11110, "fmax", + int_arm_neon_vmaxs, int_arm_neon_vmaxs, + int_arm_neon_vmaxs, v2f32, v4f32, v2f64, 1>; + +// Vector Minimum (Floating Point) +defm FMINvvv : NeonI_3VSame_SD_sizes<0b0, 0b1, 0b11110, "fmin", + int_arm_neon_vmins, int_arm_neon_vmins, + int_arm_neon_vmins, v2f32, v4f32, v2f64, 1>; + +// Vector maxNum (Floating Point) - prefer a number over a quiet NaN) +defm FMAXNMvvv : NeonI_3VSame_SD_sizes<0b0, 0b0, 0b11000, "fmaxnm", + int_aarch64_neon_vmaxnm, + int_aarch64_neon_vmaxnm, + int_aarch64_neon_vmaxnm, + v2f32, v4f32, v2f64, 1>; + +// Vector minNum (Floating Point) - prefer a number over a quiet NaN) +defm FMINNMvvv : NeonI_3VSame_SD_sizes<0b0, 0b1, 0b11000, "fminnm", + int_aarch64_neon_vminnm, + int_aarch64_neon_vminnm, + int_aarch64_neon_vminnm, + v2f32, v4f32, v2f64, 1>; + +// Vector Maximum Pairwise (Signed and Unsigned Integer) +defm SMAXPvvv : NeonI_3VSame_BHS_sizes<0b0, 0b10100, "smaxp", int_arm_neon_vpmaxs, 1>; +defm UMAXPvvv : NeonI_3VSame_BHS_sizes<0b1, 0b10100, "umaxp", int_arm_neon_vpmaxu, 1>; + +// Vector Minimum Pairwise (Signed and Unsigned Integer) +defm SMINPvvv : NeonI_3VSame_BHS_sizes<0b0, 0b10101, "sminp", int_arm_neon_vpmins, 1>; +defm UMINPvvv : NeonI_3VSame_BHS_sizes<0b1, 0b10101, "uminp", int_arm_neon_vpminu, 1>; + +// Vector Maximum Pairwise (Floating Point) +defm FMAXPvvv : NeonI_3VSame_SD_sizes<0b1, 0b0, 0b11110, "fmaxp", + int_arm_neon_vpmaxs, int_arm_neon_vpmaxs, + int_arm_neon_vpmaxs, v2f32, v4f32, v2f64, 1>; + +// Vector Minimum Pairwise (Floating Point) +defm FMINPvvv : NeonI_3VSame_SD_sizes<0b1, 0b1, 0b11110, "fminp", + int_arm_neon_vpmins, int_arm_neon_vpmins, + int_arm_neon_vpmins, v2f32, v4f32, v2f64, 1>; + +// Vector maxNum Pairwise (Floating Point) - prefer a number over a quiet NaN) +defm FMAXNMPvvv : NeonI_3VSame_SD_sizes<0b1, 0b0, 0b11000, "fmaxnmp", + int_aarch64_neon_vpmaxnm, + int_aarch64_neon_vpmaxnm, + int_aarch64_neon_vpmaxnm, + v2f32, v4f32, v2f64, 1>; + +// Vector minNum Pairwise (Floating Point) - prefer a number over a quiet NaN) +defm FMINNMPvvv : NeonI_3VSame_SD_sizes<0b1, 0b1, 0b11000, "fminnmp", + int_aarch64_neon_vpminnm, + int_aarch64_neon_vpminnm, + int_aarch64_neon_vpminnm, + v2f32, v4f32, v2f64, 1>; + +// Vector Addition Pairwise (Integer) +defm ADDP : NeonI_3VSame_BHSD_sizes<0b0, 0b10111, "addp", int_arm_neon_vpadd, 1>; + +// Vector Addition Pairwise (Floating Point) +defm FADDP : NeonI_3VSame_SD_sizes<0b1, 0b0, 0b11010, "faddp", + int_arm_neon_vpadd, + int_arm_neon_vpadd, + int_arm_neon_vpadd, + v2f32, v4f32, v2f64, 1>; + +// Vector Saturating Doubling Multiply High +defm SQDMULHvvv : NeonI_3VSame_HS_sizes<0b0, 0b10110, "sqdmulh", + int_arm_neon_vqdmulh, 1>; + +// Vector Saturating Rouding Doubling Multiply High +defm SQRDMULHvvv : NeonI_3VSame_HS_sizes<0b1, 0b10110, "sqrdmulh", + int_arm_neon_vqrdmulh, 1>; + +// Vector Multiply Extended (Floating Point) +defm FMULXvvv : NeonI_3VSame_SD_sizes<0b0, 0b0, 0b11011, "fmulx", + int_aarch64_neon_vmulx, + int_aarch64_neon_vmulx, + int_aarch64_neon_vmulx, + v2f32, v4f32, v2f64, 1>; + +// Vector Immediate Instructions + +multiclass neon_mov_imm_shift_asmoperands<string PREFIX> +{ + def _asmoperand : AsmOperandClass + { + let Name = "NeonMovImmShift" # PREFIX; + let RenderMethod = "addNeonMovImmShift" # PREFIX # "Operands"; + let PredicateMethod = "isNeonMovImmShift" # PREFIX; + } +} + +// Definition of vector immediates shift operands + +// The selectable use-cases extract the shift operation +// information from the OpCmode fields encoded in the immediate. +def neon_mod_shift_imm_XFORM : SDNodeXForm<imm, [{ + uint64_t OpCmode = N->getZExtValue(); + unsigned ShiftImm; + unsigned ShiftOnesIn; + unsigned HasShift = + A64Imms::decodeNeonModShiftImm(OpCmode, ShiftImm, ShiftOnesIn); + if (!HasShift) return SDValue(); + return CurDAG->getTargetConstant(ShiftImm, MVT::i32); +}]>; + +// Vector immediates shift operands which accept LSL and MSL +// shift operators with shift value in the range of 0, 8, 16, 24 (LSL), +// or 0, 8 (LSLH) or 8, 16 (MSL). +defm neon_mov_imm_LSL : neon_mov_imm_shift_asmoperands<"LSL">; +defm neon_mov_imm_MSL : neon_mov_imm_shift_asmoperands<"MSL">; +// LSLH restricts shift amount to 0, 8 out of 0, 8, 16, 24 +defm neon_mov_imm_LSLH : neon_mov_imm_shift_asmoperands<"LSLH">; + +multiclass neon_mov_imm_shift_operands<string PREFIX, + string HALF, string ISHALF, code pred> +{ + def _operand : Operand<i32>, ImmLeaf<i32, pred, neon_mod_shift_imm_XFORM> + { + let PrintMethod = + "printNeonMovImmShiftOperand<A64SE::" # PREFIX # ", " # ISHALF # ">"; + let DecoderMethod = + "DecodeNeonMovImmShiftOperand<A64SE::" # PREFIX # ", " # ISHALF # ">"; + let ParserMatchClass = + !cast<AsmOperandClass>("neon_mov_imm_" # PREFIX # HALF # "_asmoperand"); + } +} + +defm neon_mov_imm_LSL : neon_mov_imm_shift_operands<"LSL", "", "false", [{ + unsigned ShiftImm; + unsigned ShiftOnesIn; + unsigned HasShift = + A64Imms::decodeNeonModShiftImm(Imm, ShiftImm, ShiftOnesIn); + return (HasShift && !ShiftOnesIn); +}]>; + +defm neon_mov_imm_MSL : neon_mov_imm_shift_operands<"MSL", "", "false", [{ + unsigned ShiftImm; + unsigned ShiftOnesIn; + unsigned HasShift = + A64Imms::decodeNeonModShiftImm(Imm, ShiftImm, ShiftOnesIn); + return (HasShift && ShiftOnesIn); +}]>; + +defm neon_mov_imm_LSLH : neon_mov_imm_shift_operands<"LSL", "H", "true", [{ + unsigned ShiftImm; + unsigned ShiftOnesIn; + unsigned HasShift = + A64Imms::decodeNeonModShiftImm(Imm, ShiftImm, ShiftOnesIn); + return (HasShift && !ShiftOnesIn); +}]>; + +def neon_uimm1_asmoperand : AsmOperandClass +{ + let Name = "UImm1"; + let PredicateMethod = "isUImm<1>"; + let RenderMethod = "addImmOperands"; +} + +def neon_uimm2_asmoperand : AsmOperandClass +{ + let Name = "UImm2"; + let PredicateMethod = "isUImm<2>"; + let RenderMethod = "addImmOperands"; +} + +def neon_uimm8_asmoperand : AsmOperandClass +{ + let Name = "UImm8"; + let PredicateMethod = "isUImm<8>"; + let RenderMethod = "addImmOperands"; +} + +def neon_uimm8 : Operand<i32>, ImmLeaf<i32, [{(void)Imm; return true;}]> { + let ParserMatchClass = neon_uimm8_asmoperand; + let PrintMethod = "printUImmHexOperand"; +} + +def neon_uimm64_mask_asmoperand : AsmOperandClass +{ + let Name = "NeonUImm64Mask"; + let PredicateMethod = "isNeonUImm64Mask"; + let RenderMethod = "addNeonUImm64MaskOperands"; +} + +// MCOperand for 64-bit bytemask with each byte having only the +// value 0x00 and 0xff is encoded as an unsigned 8-bit value +def neon_uimm64_mask : Operand<i32>, ImmLeaf<i32, [{(void)Imm; return true;}]> { + let ParserMatchClass = neon_uimm64_mask_asmoperand; + let PrintMethod = "printNeonUImm64MaskOperand"; +} + +multiclass NeonI_mov_imm_lsl_sizes<string asmop, bit op, + SDPatternOperator opnode> +{ + // shift zeros, per word + def _2S : NeonI_1VModImm<0b0, op, + (outs VPR64:$Rd), + (ins neon_uimm8:$Imm, + neon_mov_imm_LSL_operand:$Simm), + !strconcat(asmop, "\t$Rd.2s, $Imm$Simm"), + [(set (v2i32 VPR64:$Rd), + (v2i32 (opnode (timm:$Imm), + (neon_mov_imm_LSL_operand:$Simm))))], + NoItinerary> { + bits<2> Simm; + let cmode = {0b0, Simm{1}, Simm{0}, 0b0}; + } + + def _4S : NeonI_1VModImm<0b1, op, + (outs VPR128:$Rd), + (ins neon_uimm8:$Imm, + neon_mov_imm_LSL_operand:$Simm), + !strconcat(asmop, "\t$Rd.4s, $Imm$Simm"), + [(set (v4i32 VPR128:$Rd), + (v4i32 (opnode (timm:$Imm), + (neon_mov_imm_LSL_operand:$Simm))))], + NoItinerary> { + bits<2> Simm; + let cmode = {0b0, Simm{1}, Simm{0}, 0b0}; + } + + // shift zeros, per halfword + def _4H : NeonI_1VModImm<0b0, op, + (outs VPR64:$Rd), + (ins neon_uimm8:$Imm, + neon_mov_imm_LSLH_operand:$Simm), + !strconcat(asmop, "\t$Rd.4h, $Imm$Simm"), + [(set (v4i16 VPR64:$Rd), + (v4i16 (opnode (timm:$Imm), + (neon_mov_imm_LSLH_operand:$Simm))))], + NoItinerary> { + bit Simm; + let cmode = {0b1, 0b0, Simm, 0b0}; + } + + def _8H : NeonI_1VModImm<0b1, op, + (outs VPR128:$Rd), + (ins neon_uimm8:$Imm, + neon_mov_imm_LSLH_operand:$Simm), + !strconcat(asmop, "\t$Rd.8h, $Imm$Simm"), + [(set (v8i16 VPR128:$Rd), + (v8i16 (opnode (timm:$Imm), + (neon_mov_imm_LSLH_operand:$Simm))))], + NoItinerary> { + bit Simm; + let cmode = {0b1, 0b0, Simm, 0b0}; + } +} + +multiclass NeonI_mov_imm_with_constraint_lsl_sizes<string asmop, bit op, + SDPatternOperator opnode, + SDPatternOperator neonopnode> +{ + let Constraints = "$src = $Rd" in { + // shift zeros, per word + def _2S : NeonI_1VModImm<0b0, op, + (outs VPR64:$Rd), + (ins VPR64:$src, neon_uimm8:$Imm, + neon_mov_imm_LSL_operand:$Simm), + !strconcat(asmop, "\t$Rd.2s, $Imm$Simm"), + [(set (v2i32 VPR64:$Rd), + (v2i32 (opnode (v2i32 VPR64:$src), + (v2i32 (bitconvert (v2i32 (neonopnode timm:$Imm, + neon_mov_imm_LSL_operand:$Simm)))))))], + NoItinerary> { + bits<2> Simm; + let cmode = {0b0, Simm{1}, Simm{0}, 0b1}; + } + + def _4S : NeonI_1VModImm<0b1, op, + (outs VPR128:$Rd), + (ins VPR128:$src, neon_uimm8:$Imm, + neon_mov_imm_LSL_operand:$Simm), + !strconcat(asmop, "\t$Rd.4s, $Imm$Simm"), + [(set (v4i32 VPR128:$Rd), + (v4i32 (opnode (v4i32 VPR128:$src), + (v4i32 (bitconvert (v4i32 (neonopnode timm:$Imm, + neon_mov_imm_LSL_operand:$Simm)))))))], + NoItinerary> { + bits<2> Simm; + let cmode = {0b0, Simm{1}, Simm{0}, 0b1}; + } + + // shift zeros, per halfword + def _4H : NeonI_1VModImm<0b0, op, + (outs VPR64:$Rd), + (ins VPR64:$src, neon_uimm8:$Imm, + neon_mov_imm_LSLH_operand:$Simm), + !strconcat(asmop, "\t$Rd.4h, $Imm$Simm"), + [(set (v4i16 VPR64:$Rd), + (v4i16 (opnode (v4i16 VPR64:$src), + (v4i16 (bitconvert (v4i16 (neonopnode timm:$Imm, + neon_mov_imm_LSL_operand:$Simm)))))))], + NoItinerary> { + bit Simm; + let cmode = {0b1, 0b0, Simm, 0b1}; + } + + def _8H : NeonI_1VModImm<0b1, op, + (outs VPR128:$Rd), + (ins VPR128:$src, neon_uimm8:$Imm, + neon_mov_imm_LSLH_operand:$Simm), + !strconcat(asmop, "\t$Rd.8h, $Imm$Simm"), + [(set (v8i16 VPR128:$Rd), + (v8i16 (opnode (v8i16 VPR128:$src), + (v8i16 (bitconvert (v8i16 (neonopnode timm:$Imm, + neon_mov_imm_LSL_operand:$Simm)))))))], + NoItinerary> { + bit Simm; + let cmode = {0b1, 0b0, Simm, 0b1}; + } + } +} + +multiclass NeonI_mov_imm_msl_sizes<string asmop, bit op, + SDPatternOperator opnode> +{ + // shift ones, per word + def _2S : NeonI_1VModImm<0b0, op, + (outs VPR64:$Rd), + (ins neon_uimm8:$Imm, + neon_mov_imm_MSL_operand:$Simm), + !strconcat(asmop, "\t$Rd.2s, $Imm$Simm"), + [(set (v2i32 VPR64:$Rd), + (v2i32 (opnode (timm:$Imm), + (neon_mov_imm_MSL_operand:$Simm))))], + NoItinerary> { + bit Simm; + let cmode = {0b1, 0b1, 0b0, Simm}; + } + + def _4S : NeonI_1VModImm<0b1, op, + (outs VPR128:$Rd), + (ins neon_uimm8:$Imm, + neon_mov_imm_MSL_operand:$Simm), + !strconcat(asmop, "\t$Rd.4s, $Imm$Simm"), + [(set (v4i32 VPR128:$Rd), + (v4i32 (opnode (timm:$Imm), + (neon_mov_imm_MSL_operand:$Simm))))], + NoItinerary> { + bit Simm; + let cmode = {0b1, 0b1, 0b0, Simm}; + } +} + +// Vector Move Immediate Shifted +let isReMaterializable = 1 in { +defm MOVIvi_lsl : NeonI_mov_imm_lsl_sizes<"movi", 0b0, Neon_movi>; +} + +// Vector Move Inverted Immediate Shifted +let isReMaterializable = 1 in { +defm MVNIvi_lsl : NeonI_mov_imm_lsl_sizes<"mvni", 0b1, Neon_mvni>; +} + +// Vector Bitwise Bit Clear (AND NOT) - immediate +let isReMaterializable = 1 in { +defm BICvi_lsl : NeonI_mov_imm_with_constraint_lsl_sizes<"bic", 0b1, + and, Neon_mvni>; +} + +// Vector Bitwise OR - immedidate + +let isReMaterializable = 1 in { +defm ORRvi_lsl : NeonI_mov_imm_with_constraint_lsl_sizes<"orr", 0b0, + or, Neon_movi>; +} + +// Additional patterns for Vector Bitwise Bit Clear (AND NOT) - immedidate +// LowerBUILD_VECTOR favors lowering MOVI over MVNI. +// BIC immediate instructions selection requires additional patterns to +// transform Neon_movi operands into BIC immediate operands + +def neon_mov_imm_LSLH_transform_XFORM : SDNodeXForm<imm, [{ + uint64_t OpCmode = N->getZExtValue(); + unsigned ShiftImm; + unsigned ShiftOnesIn; + (void)A64Imms::decodeNeonModShiftImm(OpCmode, ShiftImm, ShiftOnesIn); + // LSLH restricts shift amount to 0, 8 which are encoded as 0 and 1 + // Transform encoded shift amount 0 to 1 and 1 to 0. + return CurDAG->getTargetConstant(!ShiftImm, MVT::i32); +}]>; + +def neon_mov_imm_LSLH_transform_operand + : ImmLeaf<i32, [{ + unsigned ShiftImm; + unsigned ShiftOnesIn; + unsigned HasShift = + A64Imms::decodeNeonModShiftImm(Imm, ShiftImm, ShiftOnesIn); + return (HasShift && !ShiftOnesIn); }], + neon_mov_imm_LSLH_transform_XFORM>; + +// Transform (and A, (4h Neon_movi 0xff)) -> BIC 4h (A, 0x00, LSL 8) +// Transform (and A, (4h Neon_movi 0xff LSL #8)) -> BIC 4h (A, 0x00) +def : Pat<(v4i16 (and VPR64:$src, + (v4i16 (Neon_movi 255, neon_mov_imm_LSLH_transform_operand:$Simm)))), + (BICvi_lsl_4H VPR64:$src, 0, + neon_mov_imm_LSLH_transform_operand:$Simm)>; + +// Transform (and A, (8h Neon_movi 8h 0xff)) -> BIC 8h (A, 0x00, LSL 8) +// Transform (and A, (8h Neon_movi 0xff LSL #8)) -> BIC 8h (A, 0x00) +def : Pat<(v8i16 (and VPR128:$src, + (v8i16 (Neon_movi 255, neon_mov_imm_LSLH_transform_operand:$Simm)))), + (BICvi_lsl_8H VPR128:$src, 0, + neon_mov_imm_LSLH_transform_operand:$Simm)>; + + +multiclass Neon_bitwiseVi_patterns<SDPatternOperator opnode, + SDPatternOperator neonopnode, + Instruction INST4H, + Instruction INST8H> { + def : Pat<(v8i8 (opnode VPR64:$src, + (bitconvert(v4i16 (neonopnode timm:$Imm, + neon_mov_imm_LSLH_operand:$Simm))))), + (INST4H VPR64:$src, neon_uimm8:$Imm, + neon_mov_imm_LSLH_operand:$Simm)>; + def : Pat<(v1i64 (opnode VPR64:$src, + (bitconvert(v4i16 (neonopnode timm:$Imm, + neon_mov_imm_LSLH_operand:$Simm))))), + (INST4H VPR64:$src, neon_uimm8:$Imm, + neon_mov_imm_LSLH_operand:$Simm)>; + + def : Pat<(v16i8 (opnode VPR128:$src, + (bitconvert(v8i16 (neonopnode timm:$Imm, + neon_mov_imm_LSLH_operand:$Simm))))), + (INST8H VPR128:$src, neon_uimm8:$Imm, + neon_mov_imm_LSLH_operand:$Simm)>; + def : Pat<(v4i32 (opnode VPR128:$src, + (bitconvert(v8i16 (neonopnode timm:$Imm, + neon_mov_imm_LSLH_operand:$Simm))))), + (INST8H VPR128:$src, neon_uimm8:$Imm, + neon_mov_imm_LSLH_operand:$Simm)>; + def : Pat<(v2i64 (opnode VPR128:$src, + (bitconvert(v8i16 (neonopnode timm:$Imm, + neon_mov_imm_LSLH_operand:$Simm))))), + (INST8H VPR128:$src, neon_uimm8:$Imm, + neon_mov_imm_LSLH_operand:$Simm)>; +} + +// Additional patterns for Vector Vector Bitwise Bit Clear (AND NOT) - immediate +defm : Neon_bitwiseVi_patterns<or, Neon_mvni, BICvi_lsl_4H, BICvi_lsl_8H>; + +// Additional patterns for Vector Bitwise OR - immedidate +defm : Neon_bitwiseVi_patterns<or, Neon_movi, ORRvi_lsl_4H, ORRvi_lsl_8H>; + + +// Vector Move Immediate Masked +let isReMaterializable = 1 in { +defm MOVIvi_msl : NeonI_mov_imm_msl_sizes<"movi", 0b0, Neon_movi>; +} + +// Vector Move Inverted Immediate Masked +let isReMaterializable = 1 in { +defm MVNIvi_msl : NeonI_mov_imm_msl_sizes<"mvni", 0b1, Neon_mvni>; +} + +class NeonI_mov_imm_lsl_aliases<string asmop, string asmlane, + Instruction inst, RegisterOperand VPRC> + : NeonInstAlias<!strconcat(asmop, "\t$Rd," # asmlane # ", $Imm"), + (inst VPRC:$Rd, neon_uimm8:$Imm, 0), 0b0>; + +// Aliases for Vector Move Immediate Shifted +def : NeonI_mov_imm_lsl_aliases<"movi", ".2s", MOVIvi_lsl_2S, VPR64>; +def : NeonI_mov_imm_lsl_aliases<"movi", ".4s", MOVIvi_lsl_4S, VPR128>; +def : NeonI_mov_imm_lsl_aliases<"movi", ".4h", MOVIvi_lsl_4H, VPR64>; +def : NeonI_mov_imm_lsl_aliases<"movi", ".8h", MOVIvi_lsl_8H, VPR128>; + +// Aliases for Vector Move Inverted Immediate Shifted +def : NeonI_mov_imm_lsl_aliases<"mvni", ".2s", MVNIvi_lsl_2S, VPR64>; +def : NeonI_mov_imm_lsl_aliases<"mvni", ".4s", MVNIvi_lsl_4S, VPR128>; +def : NeonI_mov_imm_lsl_aliases<"mvni", ".4h", MVNIvi_lsl_4H, VPR64>; +def : NeonI_mov_imm_lsl_aliases<"mvni", ".8h", MVNIvi_lsl_8H, VPR128>; + +// Aliases for Vector Bitwise Bit Clear (AND NOT) - immediate +def : NeonI_mov_imm_lsl_aliases<"bic", ".2s", BICvi_lsl_2S, VPR64>; +def : NeonI_mov_imm_lsl_aliases<"bic", ".4s", BICvi_lsl_4S, VPR128>; +def : NeonI_mov_imm_lsl_aliases<"bic", ".4h", BICvi_lsl_4H, VPR64>; +def : NeonI_mov_imm_lsl_aliases<"bic", ".8h", BICvi_lsl_8H, VPR128>; + +// Aliases for Vector Bitwise OR - immedidate +def : NeonI_mov_imm_lsl_aliases<"orr", ".2s", ORRvi_lsl_2S, VPR64>; +def : NeonI_mov_imm_lsl_aliases<"orr", ".4s", ORRvi_lsl_4S, VPR128>; +def : NeonI_mov_imm_lsl_aliases<"orr", ".4h", ORRvi_lsl_4H, VPR64>; +def : NeonI_mov_imm_lsl_aliases<"orr", ".8h", ORRvi_lsl_8H, VPR128>; + +// Vector Move Immediate - per byte +let isReMaterializable = 1 in { +def MOVIvi_8B : NeonI_1VModImm<0b0, 0b0, + (outs VPR64:$Rd), (ins neon_uimm8:$Imm), + "movi\t$Rd.8b, $Imm", + [(set (v8i8 VPR64:$Rd), + (v8i8 (Neon_movi (timm:$Imm), (i32 imm))))], + NoItinerary> { + let cmode = 0b1110; +} + +def MOVIvi_16B : NeonI_1VModImm<0b1, 0b0, + (outs VPR128:$Rd), (ins neon_uimm8:$Imm), + "movi\t$Rd.16b, $Imm", + [(set (v16i8 VPR128:$Rd), + (v16i8 (Neon_movi (timm:$Imm), (i32 imm))))], + NoItinerary> { + let cmode = 0b1110; +} +} + +// Vector Move Immediate - bytemask, per double word +let isReMaterializable = 1 in { +def MOVIvi_2D : NeonI_1VModImm<0b1, 0b1, + (outs VPR128:$Rd), (ins neon_uimm64_mask:$Imm), + "movi\t $Rd.2d, $Imm", + [(set (v2i64 VPR128:$Rd), + (v2i64 (Neon_movi (timm:$Imm), (i32 imm))))], + NoItinerary> { + let cmode = 0b1110; +} +} + +// Vector Move Immediate - bytemask, one doubleword + +let isReMaterializable = 1 in { +def MOVIdi : NeonI_1VModImm<0b0, 0b1, + (outs FPR64:$Rd), (ins neon_uimm64_mask:$Imm), + "movi\t $Rd, $Imm", + [(set (v1i64 FPR64:$Rd), + (v1i64 (Neon_movi (timm:$Imm), (i32 imm))))], + NoItinerary> { + let cmode = 0b1110; +} +} + +// Vector Floating Point Move Immediate + +class NeonI_FMOV_impl<string asmlane, RegisterOperand VPRC, ValueType OpTy, + Operand immOpType, bit q, bit op> + : NeonI_1VModImm<q, op, + (outs VPRC:$Rd), (ins immOpType:$Imm), + "fmov\t$Rd" # asmlane # ", $Imm", + [(set (OpTy VPRC:$Rd), + (OpTy (Neon_fmovi (timm:$Imm))))], + NoItinerary> { + let cmode = 0b1111; + } + +let isReMaterializable = 1 in { +def FMOVvi_2S : NeonI_FMOV_impl<".2s", VPR64, v2f32, fmov32_operand, 0b0, 0b0>; +def FMOVvi_4S : NeonI_FMOV_impl<".4s", VPR128, v4f32, fmov32_operand, 0b1, 0b0>; +def FMOVvi_2D : NeonI_FMOV_impl<".2d", VPR128, v2f64, fmov64_operand, 0b1, 0b1>; +} + +// Vector Shift (Immediate) +// Immediate in [0, 63] +def imm0_63 : Operand<i32> { + let ParserMatchClass = uimm6_asmoperand; +} + +// Shift Right/Left Immediate - The immh:immb field of these shifts are encoded +// as follows: +// +// Offset Encoding +// 8 immh:immb<6:3> = '0001xxx', <imm> is encoded in immh:immb<2:0> +// 16 immh:immb<6:4> = '001xxxx', <imm> is encoded in immh:immb<3:0> +// 32 immh:immb<6:5> = '01xxxxx', <imm> is encoded in immh:immb<4:0> +// 64 immh:immb<6> = '1xxxxxx', <imm> is encoded in immh:immb<5:0> +// +// The shift right immediate amount, in the range 1 to element bits, is computed +// as Offset - UInt(immh:immb). The shift left immediate amount, in the range 0 +// to element bits - 1, is computed as UInt(immh:immb) - Offset. + +class shr_imm_asmoperands<string OFFSET> : AsmOperandClass { + let Name = "ShrImm" # OFFSET; + let RenderMethod = "addImmOperands"; + let DiagnosticType = "ShrImm" # OFFSET; +} + +class shr_imm<string OFFSET> : Operand<i32> { + let EncoderMethod = "getShiftRightImm" # OFFSET; + let DecoderMethod = "DecodeShiftRightImm" # OFFSET; + let ParserMatchClass = + !cast<AsmOperandClass>("shr_imm" # OFFSET # "_asmoperand"); +} + +def shr_imm8_asmoperand : shr_imm_asmoperands<"8">; +def shr_imm16_asmoperand : shr_imm_asmoperands<"16">; +def shr_imm32_asmoperand : shr_imm_asmoperands<"32">; +def shr_imm64_asmoperand : shr_imm_asmoperands<"64">; + +def shr_imm8 : shr_imm<"8">, ImmLeaf<i32, [{return Imm > 0 && Imm <= 8;}]>; +def shr_imm16 : shr_imm<"16">, ImmLeaf<i32, [{return Imm > 0 && Imm <= 16;}]>; +def shr_imm32 : shr_imm<"32">, ImmLeaf<i32, [{return Imm > 0 && Imm <= 32;}]>; +def shr_imm64 : shr_imm<"64">, ImmLeaf<i32, [{return Imm > 0 && Imm <= 64;}]>; + +class shl_imm_asmoperands<string OFFSET> : AsmOperandClass { + let Name = "ShlImm" # OFFSET; + let RenderMethod = "addImmOperands"; + let DiagnosticType = "ShlImm" # OFFSET; +} + +class shl_imm<string OFFSET> : Operand<i32> { + let EncoderMethod = "getShiftLeftImm" # OFFSET; + let DecoderMethod = "DecodeShiftLeftImm" # OFFSET; + let ParserMatchClass = + !cast<AsmOperandClass>("shl_imm" # OFFSET # "_asmoperand"); +} + +def shl_imm8_asmoperand : shl_imm_asmoperands<"8">; +def shl_imm16_asmoperand : shl_imm_asmoperands<"16">; +def shl_imm32_asmoperand : shl_imm_asmoperands<"32">; +def shl_imm64_asmoperand : shl_imm_asmoperands<"64">; + +def shl_imm8 : shl_imm<"8">, ImmLeaf<i32, [{return Imm >= 0 && Imm < 8;}]>; +def shl_imm16 : shl_imm<"16">, ImmLeaf<i32, [{return Imm >= 0 && Imm < 16;}]>; +def shl_imm32 : shl_imm<"32">, ImmLeaf<i32, [{return Imm >= 0 && Imm < 32;}]>; +def shl_imm64 : shl_imm<"64">, ImmLeaf<i32, [{return Imm >= 0 && Imm < 64;}]>; + +class N2VShift<bit q, bit u, bits<5> opcode, string asmop, string T, + RegisterOperand VPRC, ValueType Ty, Operand ImmTy, SDNode OpNode> + : NeonI_2VShiftImm<q, u, opcode, + (outs VPRC:$Rd), (ins VPRC:$Rn, ImmTy:$Imm), + asmop # "\t$Rd." # T # ", $Rn." # T # ", $Imm", + [(set (Ty VPRC:$Rd), + (Ty (OpNode (Ty VPRC:$Rn), + (Ty (Neon_vdup (i32 ImmTy:$Imm))))))], + NoItinerary>; + +multiclass NeonI_N2VShL<bit u, bits<5> opcode, string asmop> { + // 64-bit vector types. + def _8B : N2VShift<0b0, u, opcode, asmop, "8b", VPR64, v8i8, shl_imm8, shl> { + let Inst{22-19} = 0b0001; // immh:immb = 0001xxx + } + + def _4H : N2VShift<0b0, u, opcode, asmop, "4h", VPR64, v4i16, shl_imm16, shl> { + let Inst{22-20} = 0b001; // immh:immb = 001xxxx + } + + def _2S : N2VShift<0b0, u, opcode, asmop, "2s", VPR64, v2i32, shl_imm32, shl> { + let Inst{22-21} = 0b01; // immh:immb = 01xxxxx + } + + // 128-bit vector types. + def _16B : N2VShift<0b1, u, opcode, asmop, "16b", VPR128, v16i8, shl_imm8, shl> { + let Inst{22-19} = 0b0001; // immh:immb = 0001xxx + } + + def _8H : N2VShift<0b1, u, opcode, asmop, "8h", VPR128, v8i16, shl_imm16, shl> { + let Inst{22-20} = 0b001; // immh:immb = 001xxxx + } + + def _4S : N2VShift<0b1, u, opcode, asmop, "4s", VPR128, v4i32, shl_imm32, shl> { + let Inst{22-21} = 0b01; // immh:immb = 01xxxxx + } + + def _2D : N2VShift<0b1, u, opcode, asmop, "2d", VPR128, v2i64, shl_imm64, shl> { + let Inst{22} = 0b1; // immh:immb = 1xxxxxx + } +} + +multiclass NeonI_N2VShR<bit u, bits<5> opcode, string asmop, SDNode OpNode> { + def _8B : N2VShift<0b0, u, opcode, asmop, "8b", VPR64, v8i8, shr_imm8, + OpNode> { + let Inst{22-19} = 0b0001; + } + + def _4H : N2VShift<0b0, u, opcode, asmop, "4h", VPR64, v4i16, shr_imm16, + OpNode> { + let Inst{22-20} = 0b001; + } + + def _2S : N2VShift<0b0, u, opcode, asmop, "2s", VPR64, v2i32, shr_imm32, + OpNode> { + let Inst{22-21} = 0b01; + } + + def _16B : N2VShift<0b1, u, opcode, asmop, "16b", VPR128, v16i8, shr_imm8, + OpNode> { + let Inst{22-19} = 0b0001; + } + + def _8H : N2VShift<0b1, u, opcode, asmop, "8h", VPR128, v8i16, shr_imm16, + OpNode> { + let Inst{22-20} = 0b001; + } + + def _4S : N2VShift<0b1, u, opcode, asmop, "4s", VPR128, v4i32, shr_imm32, + OpNode> { + let Inst{22-21} = 0b01; + } + + def _2D : N2VShift<0b1, u, opcode, asmop, "2d", VPR128, v2i64, shr_imm64, + OpNode> { + let Inst{22} = 0b1; + } +} + +// Shift left +defm SHLvvi : NeonI_N2VShL<0b0, 0b01010, "shl">; + +// Shift right +defm SSHRvvi : NeonI_N2VShR<0b0, 0b00000, "sshr", sra>; +defm USHRvvi : NeonI_N2VShR<0b1, 0b00000, "ushr", srl>; + +def Neon_High16B : PatFrag<(ops node:$in), + (extract_subvector (v16i8 node:$in), (iPTR 8))>; +def Neon_High8H : PatFrag<(ops node:$in), + (extract_subvector (v8i16 node:$in), (iPTR 4))>; +def Neon_High4S : PatFrag<(ops node:$in), + (extract_subvector (v4i32 node:$in), (iPTR 2))>; +def Neon_High2D : PatFrag<(ops node:$in), + (extract_subvector (v2i64 node:$in), (iPTR 1))>; +def Neon_High4float : PatFrag<(ops node:$in), + (extract_subvector (v4f32 node:$in), (iPTR 2))>; +def Neon_High2double : PatFrag<(ops node:$in), + (extract_subvector (v2f64 node:$in), (iPTR 1))>; + +def Neon_Low16B : PatFrag<(ops node:$in), + (v8i8 (extract_subvector (v16i8 node:$in), + (iPTR 0)))>; +def Neon_Low8H : PatFrag<(ops node:$in), + (v4i16 (extract_subvector (v8i16 node:$in), + (iPTR 0)))>; +def Neon_Low4S : PatFrag<(ops node:$in), + (v2i32 (extract_subvector (v4i32 node:$in), + (iPTR 0)))>; +def Neon_Low2D : PatFrag<(ops node:$in), + (v1i64 (extract_subvector (v2i64 node:$in), + (iPTR 0)))>; +def Neon_Low4float : PatFrag<(ops node:$in), + (v2f32 (extract_subvector (v4f32 node:$in), + (iPTR 0)))>; +def Neon_Low2double : PatFrag<(ops node:$in), + (v1f64 (extract_subvector (v2f64 node:$in), + (iPTR 0)))>; + +class N2VShiftLong<bit q, bit u, bits<5> opcode, string asmop, string DestT, + string SrcT, ValueType DestTy, ValueType SrcTy, + Operand ImmTy, SDPatternOperator ExtOp> + : NeonI_2VShiftImm<q, u, opcode, (outs VPR128:$Rd), + (ins VPR64:$Rn, ImmTy:$Imm), + asmop # "\t$Rd." # DestT # ", $Rn." # SrcT # ", $Imm", + [(set (DestTy VPR128:$Rd), + (DestTy (shl + (DestTy (ExtOp (SrcTy VPR64:$Rn))), + (DestTy (Neon_vdup (i32 ImmTy:$Imm))))))], + NoItinerary>; + +class N2VShiftLongHigh<bit q, bit u, bits<5> opcode, string asmop, string DestT, + string SrcT, ValueType DestTy, ValueType SrcTy, + int StartIndex, Operand ImmTy, + SDPatternOperator ExtOp, PatFrag getTop> + : NeonI_2VShiftImm<q, u, opcode, (outs VPR128:$Rd), + (ins VPR128:$Rn, ImmTy:$Imm), + asmop # "2\t$Rd." # DestT # ", $Rn." # SrcT # ", $Imm", + [(set (DestTy VPR128:$Rd), + (DestTy (shl + (DestTy (ExtOp + (SrcTy (getTop VPR128:$Rn)))), + (DestTy (Neon_vdup (i32 ImmTy:$Imm))))))], + NoItinerary>; + +multiclass NeonI_N2VShLL<string prefix, bit u, bits<5> opcode, string asmop, + SDNode ExtOp> { + // 64-bit vector types. + def _8B : N2VShiftLong<0b0, u, opcode, asmop, "8h", "8b", v8i16, v8i8, + shl_imm8, ExtOp> { + let Inst{22-19} = 0b0001; // immh:immb = 0001xxx + } + + def _4H : N2VShiftLong<0b0, u, opcode, asmop, "4s", "4h", v4i32, v4i16, + shl_imm16, ExtOp> { + let Inst{22-20} = 0b001; // immh:immb = 001xxxx + } + + def _2S : N2VShiftLong<0b0, u, opcode, asmop, "2d", "2s", v2i64, v2i32, + shl_imm32, ExtOp> { + let Inst{22-21} = 0b01; // immh:immb = 01xxxxx + } + + // 128-bit vector types + def _16B : N2VShiftLongHigh<0b1, u, opcode, asmop, "8h", "16b", v8i16, v8i8, + 8, shl_imm8, ExtOp, Neon_High16B> { + let Inst{22-19} = 0b0001; // immh:immb = 0001xxx + } + + def _8H : N2VShiftLongHigh<0b1, u, opcode, asmop, "4s", "8h", v4i32, v4i16, + 4, shl_imm16, ExtOp, Neon_High8H> { + let Inst{22-20} = 0b001; // immh:immb = 001xxxx + } + + def _4S : N2VShiftLongHigh<0b1, u, opcode, asmop, "2d", "4s", v2i64, v2i32, + 2, shl_imm32, ExtOp, Neon_High4S> { + let Inst{22-21} = 0b01; // immh:immb = 01xxxxx + } + + // Use other patterns to match when the immediate is 0. + def : Pat<(v8i16 (ExtOp (v8i8 VPR64:$Rn))), + (!cast<Instruction>(prefix # "_8B") VPR64:$Rn, 0)>; + + def : Pat<(v4i32 (ExtOp (v4i16 VPR64:$Rn))), + (!cast<Instruction>(prefix # "_4H") VPR64:$Rn, 0)>; + + def : Pat<(v2i64 (ExtOp (v2i32 VPR64:$Rn))), + (!cast<Instruction>(prefix # "_2S") VPR64:$Rn, 0)>; + + def : Pat<(v8i16 (ExtOp (v8i8 (Neon_High16B VPR128:$Rn)))), + (!cast<Instruction>(prefix # "_16B") VPR128:$Rn, 0)>; + + def : Pat<(v4i32 (ExtOp (v4i16 (Neon_High8H VPR128:$Rn)))), + (!cast<Instruction>(prefix # "_8H") VPR128:$Rn, 0)>; + + def : Pat<(v2i64 (ExtOp (v2i32 (Neon_High4S VPR128:$Rn)))), + (!cast<Instruction>(prefix # "_4S") VPR128:$Rn, 0)>; +} + +// Shift left long +defm SSHLLvvi : NeonI_N2VShLL<"SSHLLvvi", 0b0, 0b10100, "sshll", sext>; +defm USHLLvvi : NeonI_N2VShLL<"USHLLvvi", 0b1, 0b10100, "ushll", zext>; + +// Rounding/Saturating shift +class N2VShift_RQ<bit q, bit u, bits<5> opcode, string asmop, string T, + RegisterOperand VPRC, ValueType Ty, Operand ImmTy, + SDPatternOperator OpNode> + : NeonI_2VShiftImm<q, u, opcode, + (outs VPRC:$Rd), (ins VPRC:$Rn, ImmTy:$Imm), + asmop # "\t$Rd." # T # ", $Rn." # T # ", $Imm", + [(set (Ty VPRC:$Rd), (Ty (OpNode (Ty VPRC:$Rn), + (i32 ImmTy:$Imm))))], + NoItinerary>; + +// shift right (vector by immediate) +multiclass NeonI_N2VShR_RQ<bit u, bits<5> opcode, string asmop, + SDPatternOperator OpNode> { + def _8B : N2VShift_RQ<0b0, u, opcode, asmop, "8b", VPR64, v8i8, shr_imm8, + OpNode> { + let Inst{22-19} = 0b0001; + } + + def _4H : N2VShift_RQ<0b0, u, opcode, asmop, "4h", VPR64, v4i16, shr_imm16, + OpNode> { + let Inst{22-20} = 0b001; + } + + def _2S : N2VShift_RQ<0b0, u, opcode, asmop, "2s", VPR64, v2i32, shr_imm32, + OpNode> { + let Inst{22-21} = 0b01; + } + + def _16B : N2VShift_RQ<0b1, u, opcode, asmop, "16b", VPR128, v16i8, shr_imm8, + OpNode> { + let Inst{22-19} = 0b0001; + } + + def _8H : N2VShift_RQ<0b1, u, opcode, asmop, "8h", VPR128, v8i16, shr_imm16, + OpNode> { + let Inst{22-20} = 0b001; + } + + def _4S : N2VShift_RQ<0b1, u, opcode, asmop, "4s", VPR128, v4i32, shr_imm32, + OpNode> { + let Inst{22-21} = 0b01; + } + + def _2D : N2VShift_RQ<0b1, u, opcode, asmop, "2d", VPR128, v2i64, shr_imm64, + OpNode> { + let Inst{22} = 0b1; + } +} + +multiclass NeonI_N2VShL_Q<bit u, bits<5> opcode, string asmop, + SDPatternOperator OpNode> { + // 64-bit vector types. + def _8B : N2VShift_RQ<0b0, u, opcode, asmop, "8b", VPR64, v8i8, shl_imm8, + OpNode> { + let Inst{22-19} = 0b0001; + } + + def _4H : N2VShift_RQ<0b0, u, opcode, asmop, "4h", VPR64, v4i16, shl_imm16, + OpNode> { + let Inst{22-20} = 0b001; + } + + def _2S : N2VShift_RQ<0b0, u, opcode, asmop, "2s", VPR64, v2i32, shl_imm32, + OpNode> { + let Inst{22-21} = 0b01; + } + + // 128-bit vector types. + def _16B : N2VShift_RQ<0b1, u, opcode, asmop, "16b", VPR128, v16i8, shl_imm8, + OpNode> { + let Inst{22-19} = 0b0001; + } + + def _8H : N2VShift_RQ<0b1, u, opcode, asmop, "8h", VPR128, v8i16, shl_imm16, + OpNode> { + let Inst{22-20} = 0b001; + } + + def _4S : N2VShift_RQ<0b1, u, opcode, asmop, "4s", VPR128, v4i32, shl_imm32, + OpNode> { + let Inst{22-21} = 0b01; + } + + def _2D : N2VShift_RQ<0b1, u, opcode, asmop, "2d", VPR128, v2i64, shl_imm64, + OpNode> { + let Inst{22} = 0b1; + } +} + +// Rounding shift right +defm SRSHRvvi : NeonI_N2VShR_RQ<0b0, 0b00100, "srshr", + int_aarch64_neon_vsrshr>; +defm URSHRvvi : NeonI_N2VShR_RQ<0b1, 0b00100, "urshr", + int_aarch64_neon_vurshr>; + +// Saturating shift left unsigned +defm SQSHLUvvi : NeonI_N2VShL_Q<0b1, 0b01100, "sqshlu", int_aarch64_neon_vsqshlu>; + +// Saturating shift left +defm SQSHLvvi : NeonI_N2VShL_Q<0b0, 0b01110, "sqshl", Neon_sqrshlImm>; +defm UQSHLvvi : NeonI_N2VShL_Q<0b1, 0b01110, "uqshl", Neon_uqrshlImm>; + +class N2VShiftAdd<bit q, bit u, bits<5> opcode, string asmop, string T, + RegisterOperand VPRC, ValueType Ty, Operand ImmTy, + SDNode OpNode> + : NeonI_2VShiftImm<q, u, opcode, + (outs VPRC:$Rd), (ins VPRC:$src, VPRC:$Rn, ImmTy:$Imm), + asmop # "\t$Rd." # T # ", $Rn." # T # ", $Imm", + [(set (Ty VPRC:$Rd), (Ty (add (Ty VPRC:$src), + (Ty (OpNode (Ty VPRC:$Rn), + (Ty (Neon_vdup (i32 ImmTy:$Imm))))))))], + NoItinerary> { + let Constraints = "$src = $Rd"; +} + +// Shift Right accumulate +multiclass NeonI_N2VShRAdd<bit u, bits<5> opcode, string asmop, SDNode OpNode> { + def _8B : N2VShiftAdd<0b0, u, opcode, asmop, "8b", VPR64, v8i8, shr_imm8, + OpNode> { + let Inst{22-19} = 0b0001; + } + + def _4H : N2VShiftAdd<0b0, u, opcode, asmop, "4h", VPR64, v4i16, shr_imm16, + OpNode> { + let Inst{22-20} = 0b001; + } + + def _2S : N2VShiftAdd<0b0, u, opcode, asmop, "2s", VPR64, v2i32, shr_imm32, + OpNode> { + let Inst{22-21} = 0b01; + } + + def _16B : N2VShiftAdd<0b1, u, opcode, asmop, "16b", VPR128, v16i8, shr_imm8, + OpNode> { + let Inst{22-19} = 0b0001; + } + + def _8H : N2VShiftAdd<0b1, u, opcode, asmop, "8h", VPR128, v8i16, shr_imm16, + OpNode> { + let Inst{22-20} = 0b001; + } + + def _4S : N2VShiftAdd<0b1, u, opcode, asmop, "4s", VPR128, v4i32, shr_imm32, + OpNode> { + let Inst{22-21} = 0b01; + } + + def _2D : N2VShiftAdd<0b1, u, opcode, asmop, "2d", VPR128, v2i64, shr_imm64, + OpNode> { + let Inst{22} = 0b1; + } +} + +// Shift right and accumulate +defm SSRAvvi : NeonI_N2VShRAdd<0, 0b00010, "ssra", sra>; +defm USRAvvi : NeonI_N2VShRAdd<1, 0b00010, "usra", srl>; + +// Rounding shift accumulate +class N2VShiftAdd_R<bit q, bit u, bits<5> opcode, string asmop, string T, + RegisterOperand VPRC, ValueType Ty, Operand ImmTy, + SDPatternOperator OpNode> + : NeonI_2VShiftImm<q, u, opcode, + (outs VPRC:$Rd), (ins VPRC:$src, VPRC:$Rn, ImmTy:$Imm), + asmop # "\t$Rd." # T # ", $Rn." # T # ", $Imm", + [(set (Ty VPRC:$Rd), (Ty (add (Ty VPRC:$src), + (Ty (OpNode (Ty VPRC:$Rn), (i32 ImmTy:$Imm))))))], + NoItinerary> { + let Constraints = "$src = $Rd"; +} + +multiclass NeonI_N2VShRAdd_R<bit u, bits<5> opcode, string asmop, + SDPatternOperator OpNode> { + def _8B : N2VShiftAdd_R<0b0, u, opcode, asmop, "8b", VPR64, v8i8, shr_imm8, + OpNode> { + let Inst{22-19} = 0b0001; + } + + def _4H : N2VShiftAdd_R<0b0, u, opcode, asmop, "4h", VPR64, v4i16, shr_imm16, + OpNode> { + let Inst{22-20} = 0b001; + } + + def _2S : N2VShiftAdd_R<0b0, u, opcode, asmop, "2s", VPR64, v2i32, shr_imm32, + OpNode> { + let Inst{22-21} = 0b01; + } + + def _16B : N2VShiftAdd_R<0b1, u, opcode, asmop, "16b", VPR128, v16i8, shr_imm8, + OpNode> { + let Inst{22-19} = 0b0001; + } + + def _8H : N2VShiftAdd_R<0b1, u, opcode, asmop, "8h", VPR128, v8i16, shr_imm16, + OpNode> { + let Inst{22-20} = 0b001; + } + + def _4S : N2VShiftAdd_R<0b1, u, opcode, asmop, "4s", VPR128, v4i32, shr_imm32, + OpNode> { + let Inst{22-21} = 0b01; + } + + def _2D : N2VShiftAdd_R<0b1, u, opcode, asmop, "2d", VPR128, v2i64, shr_imm64, + OpNode> { + let Inst{22} = 0b1; + } +} + +// Rounding shift right and accumulate +defm SRSRAvvi : NeonI_N2VShRAdd_R<0, 0b00110, "srsra", int_aarch64_neon_vsrshr>; +defm URSRAvvi : NeonI_N2VShRAdd_R<1, 0b00110, "ursra", int_aarch64_neon_vurshr>; + +// Shift insert by immediate +class N2VShiftIns<bit q, bit u, bits<5> opcode, string asmop, string T, + RegisterOperand VPRC, ValueType Ty, Operand ImmTy, + SDPatternOperator OpNode> + : NeonI_2VShiftImm<q, u, opcode, + (outs VPRC:$Rd), (ins VPRC:$src, VPRC:$Rn, ImmTy:$Imm), + asmop # "\t$Rd." # T # ", $Rn." # T # ", $Imm", + [(set (Ty VPRC:$Rd), (Ty (OpNode (Ty VPRC:$src), (Ty VPRC:$Rn), + (i32 ImmTy:$Imm))))], + NoItinerary> { + let Constraints = "$src = $Rd"; +} + +// shift left insert (vector by immediate) +multiclass NeonI_N2VShLIns<bit u, bits<5> opcode, string asmop> { + def _8B : N2VShiftIns<0b0, u, opcode, asmop, "8b", VPR64, v8i8, shl_imm8, + int_aarch64_neon_vsli> { + let Inst{22-19} = 0b0001; + } + + def _4H : N2VShiftIns<0b0, u, opcode, asmop, "4h", VPR64, v4i16, shl_imm16, + int_aarch64_neon_vsli> { + let Inst{22-20} = 0b001; + } + + def _2S : N2VShiftIns<0b0, u, opcode, asmop, "2s", VPR64, v2i32, shl_imm32, + int_aarch64_neon_vsli> { + let Inst{22-21} = 0b01; + } + + // 128-bit vector types + def _16B : N2VShiftIns<0b1, u, opcode, asmop, "16b", VPR128, v16i8, shl_imm8, + int_aarch64_neon_vsli> { + let Inst{22-19} = 0b0001; + } + + def _8H : N2VShiftIns<0b1, u, opcode, asmop, "8h", VPR128, v8i16, shl_imm16, + int_aarch64_neon_vsli> { + let Inst{22-20} = 0b001; + } + + def _4S : N2VShiftIns<0b1, u, opcode, asmop, "4s", VPR128, v4i32, shl_imm32, + int_aarch64_neon_vsli> { + let Inst{22-21} = 0b01; + } + + def _2D : N2VShiftIns<0b1, u, opcode, asmop, "2d", VPR128, v2i64, shl_imm64, + int_aarch64_neon_vsli> { + let Inst{22} = 0b1; + } +} + +// shift right insert (vector by immediate) +multiclass NeonI_N2VShRIns<bit u, bits<5> opcode, string asmop> { + // 64-bit vector types. + def _8B : N2VShiftIns<0b0, u, opcode, asmop, "8b", VPR64, v8i8, shr_imm8, + int_aarch64_neon_vsri> { + let Inst{22-19} = 0b0001; + } + + def _4H : N2VShiftIns<0b0, u, opcode, asmop, "4h", VPR64, v4i16, shr_imm16, + int_aarch64_neon_vsri> { + let Inst{22-20} = 0b001; + } + + def _2S : N2VShiftIns<0b0, u, opcode, asmop, "2s", VPR64, v2i32, shr_imm32, + int_aarch64_neon_vsri> { + let Inst{22-21} = 0b01; + } + + // 128-bit vector types + def _16B : N2VShiftIns<0b1, u, opcode, asmop, "16b", VPR128, v16i8, shr_imm8, + int_aarch64_neon_vsri> { + let Inst{22-19} = 0b0001; + } + + def _8H : N2VShiftIns<0b1, u, opcode, asmop, "8h", VPR128, v8i16, shr_imm16, + int_aarch64_neon_vsri> { + let Inst{22-20} = 0b001; + } + + def _4S : N2VShiftIns<0b1, u, opcode, asmop, "4s", VPR128, v4i32, shr_imm32, + int_aarch64_neon_vsri> { + let Inst{22-21} = 0b01; + } + + def _2D : N2VShiftIns<0b1, u, opcode, asmop, "2d", VPR128, v2i64, shr_imm64, + int_aarch64_neon_vsri> { + let Inst{22} = 0b1; + } +} + +// Shift left and insert +defm SLIvvi : NeonI_N2VShLIns<0b1, 0b01010, "sli">; + +// Shift right and insert +defm SRIvvi : NeonI_N2VShRIns<0b1, 0b01000, "sri">; + +class N2VShR_Narrow<bit q, bit u, bits<5> opcode, string asmop, string DestT, + string SrcT, Operand ImmTy> + : NeonI_2VShiftImm<q, u, opcode, + (outs VPR64:$Rd), (ins VPR128:$Rn, ImmTy:$Imm), + asmop # "\t$Rd." # DestT # ", $Rn." # SrcT # ", $Imm", + [], NoItinerary>; + +class N2VShR_Narrow_Hi<bit q, bit u, bits<5> opcode, string asmop, string DestT, + string SrcT, Operand ImmTy> + : NeonI_2VShiftImm<q, u, opcode, (outs VPR128:$Rd), + (ins VPR128:$src, VPR128:$Rn, ImmTy:$Imm), + asmop # "\t$Rd." # DestT # ", $Rn." # SrcT # ", $Imm", + [], NoItinerary> { + let Constraints = "$src = $Rd"; +} + +// left long shift by immediate +multiclass NeonI_N2VShR_Narrow<bit u, bits<5> opcode, string asmop> { + def _8B : N2VShR_Narrow<0b0, u, opcode, asmop, "8b", "8h", shr_imm8> { + let Inst{22-19} = 0b0001; + } + + def _4H : N2VShR_Narrow<0b0, u, opcode, asmop, "4h", "4s", shr_imm16> { + let Inst{22-20} = 0b001; + } + + def _2S : N2VShR_Narrow<0b0, u, opcode, asmop, "2s", "2d", shr_imm32> { + let Inst{22-21} = 0b01; + } + + // Shift Narrow High + def _16B : N2VShR_Narrow_Hi<0b1, u, opcode, asmop # "2", "16b", "8h", + shr_imm8> { + let Inst{22-19} = 0b0001; + } + + def _8H : N2VShR_Narrow_Hi<0b1, u, opcode, asmop # "2", "8h", "4s", + shr_imm16> { + let Inst{22-20} = 0b001; + } + + def _4S : N2VShR_Narrow_Hi<0b1, u, opcode, asmop # "2", "4s", "2d", + shr_imm32> { + let Inst{22-21} = 0b01; + } +} + +// Shift right narrow +defm SHRNvvi : NeonI_N2VShR_Narrow<0b0, 0b10000, "shrn">; + +// Shift right narrow (prefix Q is saturating, prefix R is rounding) +defm QSHRUNvvi :NeonI_N2VShR_Narrow<0b1, 0b10000, "sqshrun">; +defm RSHRNvvi : NeonI_N2VShR_Narrow<0b0, 0b10001, "rshrn">; +defm QRSHRUNvvi : NeonI_N2VShR_Narrow<0b1, 0b10001, "sqrshrun">; +defm SQSHRNvvi : NeonI_N2VShR_Narrow<0b0, 0b10010, "sqshrn">; +defm UQSHRNvvi : NeonI_N2VShR_Narrow<0b1, 0b10010, "uqshrn">; +defm SQRSHRNvvi : NeonI_N2VShR_Narrow<0b0, 0b10011, "sqrshrn">; +defm UQRSHRNvvi : NeonI_N2VShR_Narrow<0b1, 0b10011, "uqrshrn">; + +def Neon_combine_2D : PatFrag<(ops node:$Rm, node:$Rn), + (v2i64 (concat_vectors (v1i64 node:$Rm), + (v1i64 node:$Rn)))>; +def Neon_combine_8H : PatFrag<(ops node:$Rm, node:$Rn), + (v8i16 (concat_vectors (v4i16 node:$Rm), + (v4i16 node:$Rn)))>; +def Neon_combine_4S : PatFrag<(ops node:$Rm, node:$Rn), + (v4i32 (concat_vectors (v2i32 node:$Rm), + (v2i32 node:$Rn)))>; +def Neon_combine_4f : PatFrag<(ops node:$Rm, node:$Rn), + (v4f32 (concat_vectors (v2f32 node:$Rm), + (v2f32 node:$Rn)))>; +def Neon_combine_2d : PatFrag<(ops node:$Rm, node:$Rn), + (v2f64 (concat_vectors (v1f64 node:$Rm), + (v1f64 node:$Rn)))>; + +def Neon_lshrImm8H : PatFrag<(ops node:$lhs, node:$rhs), + (v8i16 (srl (v8i16 node:$lhs), + (v8i16 (Neon_vdup (i32 node:$rhs)))))>; +def Neon_lshrImm4S : PatFrag<(ops node:$lhs, node:$rhs), + (v4i32 (srl (v4i32 node:$lhs), + (v4i32 (Neon_vdup (i32 node:$rhs)))))>; +def Neon_lshrImm2D : PatFrag<(ops node:$lhs, node:$rhs), + (v2i64 (srl (v2i64 node:$lhs), + (v2i64 (Neon_vdup (i32 node:$rhs)))))>; +def Neon_ashrImm8H : PatFrag<(ops node:$lhs, node:$rhs), + (v8i16 (sra (v8i16 node:$lhs), + (v8i16 (Neon_vdup (i32 node:$rhs)))))>; +def Neon_ashrImm4S : PatFrag<(ops node:$lhs, node:$rhs), + (v4i32 (sra (v4i32 node:$lhs), + (v4i32 (Neon_vdup (i32 node:$rhs)))))>; +def Neon_ashrImm2D : PatFrag<(ops node:$lhs, node:$rhs), + (v2i64 (sra (v2i64 node:$lhs), + (v2i64 (Neon_vdup (i32 node:$rhs)))))>; + +// Normal shift right narrow is matched by IR (srl/sra, trunc, concat_vectors) +multiclass Neon_shiftNarrow_patterns<string shr> { + def : Pat<(v8i8 (trunc (!cast<PatFrag>("Neon_" # shr # "Imm8H") VPR128:$Rn, + (i32 shr_imm8:$Imm)))), + (SHRNvvi_8B VPR128:$Rn, imm:$Imm)>; + def : Pat<(v4i16 (trunc (!cast<PatFrag>("Neon_" # shr # "Imm4S") VPR128:$Rn, + (i32 shr_imm16:$Imm)))), + (SHRNvvi_4H VPR128:$Rn, imm:$Imm)>; + def : Pat<(v2i32 (trunc (!cast<PatFrag>("Neon_" # shr # "Imm2D") VPR128:$Rn, + (i32 shr_imm32:$Imm)))), + (SHRNvvi_2S VPR128:$Rn, imm:$Imm)>; + + def : Pat<(Neon_combine_2D (v1i64 VPR64:$src), (v1i64 (bitconvert + (v8i8 (trunc (!cast<PatFrag>("Neon_" # shr # "Imm8H") + VPR128:$Rn, (i32 shr_imm8:$Imm))))))), + (SHRNvvi_16B (v2i64 (SUBREG_TO_REG (i64 0), VPR64:$src, sub_64)), + VPR128:$Rn, imm:$Imm)>; + def : Pat<(Neon_combine_2D (v1i64 VPR64:$src), (v1i64 (bitconvert + (v4i16 (trunc (!cast<PatFrag>("Neon_" # shr # "Imm4S") + VPR128:$Rn, (i32 shr_imm16:$Imm))))))), + (SHRNvvi_8H (SUBREG_TO_REG (i64 0), VPR64:$src, sub_64), + VPR128:$Rn, imm:$Imm)>; + def : Pat<(Neon_combine_2D (v1i64 VPR64:$src), (v1i64 (bitconvert + (v2i32 (trunc (!cast<PatFrag>("Neon_" # shr # "Imm2D") + VPR128:$Rn, (i32 shr_imm32:$Imm))))))), + (SHRNvvi_4S (SUBREG_TO_REG (i64 0), VPR64:$src, sub_64), + VPR128:$Rn, imm:$Imm)>; +} + +multiclass Neon_shiftNarrow_QR_patterns<SDPatternOperator op, string prefix> { + def : Pat<(v8i8 (op (v8i16 VPR128:$Rn), shr_imm8:$Imm)), + (!cast<Instruction>(prefix # "_8B") VPR128:$Rn, imm:$Imm)>; + def : Pat<(v4i16 (op (v4i32 VPR128:$Rn), shr_imm16:$Imm)), + (!cast<Instruction>(prefix # "_4H") VPR128:$Rn, imm:$Imm)>; + def : Pat<(v2i32 (op (v2i64 VPR128:$Rn), shr_imm32:$Imm)), + (!cast<Instruction>(prefix # "_2S") VPR128:$Rn, imm:$Imm)>; + + def : Pat<(Neon_combine_2D (v1i64 VPR64:$src), + (v1i64 (bitconvert (v8i8 + (op (v8i16 VPR128:$Rn), shr_imm8:$Imm))))), + (!cast<Instruction>(prefix # "_16B") + (SUBREG_TO_REG (i64 0), VPR64:$src, sub_64), + VPR128:$Rn, imm:$Imm)>; + def : Pat<(Neon_combine_2D (v1i64 VPR64:$src), + (v1i64 (bitconvert (v4i16 + (op (v4i32 VPR128:$Rn), shr_imm16:$Imm))))), + (!cast<Instruction>(prefix # "_8H") + (SUBREG_TO_REG (i64 0), VPR64:$src, sub_64), + VPR128:$Rn, imm:$Imm)>; + def : Pat<(Neon_combine_2D (v1i64 VPR64:$src), + (v1i64 (bitconvert (v2i32 + (op (v2i64 VPR128:$Rn), shr_imm32:$Imm))))), + (!cast<Instruction>(prefix # "_4S") + (SUBREG_TO_REG (i64 0), VPR64:$src, sub_64), + VPR128:$Rn, imm:$Imm)>; +} + +defm : Neon_shiftNarrow_patterns<"lshr">; +defm : Neon_shiftNarrow_patterns<"ashr">; + +defm : Neon_shiftNarrow_QR_patterns<int_aarch64_neon_vsqshrun, "QSHRUNvvi">; +defm : Neon_shiftNarrow_QR_patterns<int_aarch64_neon_vrshrn, "RSHRNvvi">; +defm : Neon_shiftNarrow_QR_patterns<int_aarch64_neon_vsqrshrun, "QRSHRUNvvi">; +defm : Neon_shiftNarrow_QR_patterns<int_aarch64_neon_vsqshrn, "SQSHRNvvi">; +defm : Neon_shiftNarrow_QR_patterns<int_aarch64_neon_vuqshrn, "UQSHRNvvi">; +defm : Neon_shiftNarrow_QR_patterns<int_aarch64_neon_vsqrshrn, "SQRSHRNvvi">; +defm : Neon_shiftNarrow_QR_patterns<int_aarch64_neon_vuqrshrn, "UQRSHRNvvi">; + +// Convert fix-point and float-pointing +class N2VCvt_Fx<bit q, bit u, bits<5> opcode, string asmop, string T, + RegisterOperand VPRC, ValueType DestTy, ValueType SrcTy, + Operand ImmTy, SDPatternOperator IntOp> + : NeonI_2VShiftImm<q, u, opcode, + (outs VPRC:$Rd), (ins VPRC:$Rn, ImmTy:$Imm), + asmop # "\t$Rd." # T # ", $Rn." # T # ", $Imm", + [(set (DestTy VPRC:$Rd), (DestTy (IntOp (SrcTy VPRC:$Rn), + (i32 ImmTy:$Imm))))], + NoItinerary>; + +multiclass NeonI_N2VCvt_Fx2fp<bit u, bits<5> opcode, string asmop, + SDPatternOperator IntOp> { + def _2S : N2VCvt_Fx<0, u, opcode, asmop, "2s", VPR64, v2f32, v2i32, + shr_imm32, IntOp> { + let Inst{22-21} = 0b01; + } + + def _4S : N2VCvt_Fx<1, u, opcode, asmop, "4s", VPR128, v4f32, v4i32, + shr_imm32, IntOp> { + let Inst{22-21} = 0b01; + } + + def _2D : N2VCvt_Fx<1, u, opcode, asmop, "2d", VPR128, v2f64, v2i64, + shr_imm64, IntOp> { + let Inst{22} = 0b1; + } +} + +multiclass NeonI_N2VCvt_Fp2fx<bit u, bits<5> opcode, string asmop, + SDPatternOperator IntOp> { + def _2S : N2VCvt_Fx<0, u, opcode, asmop, "2s", VPR64, v2i32, v2f32, + shr_imm32, IntOp> { + let Inst{22-21} = 0b01; + } + + def _4S : N2VCvt_Fx<1, u, opcode, asmop, "4s", VPR128, v4i32, v4f32, + shr_imm32, IntOp> { + let Inst{22-21} = 0b01; + } + + def _2D : N2VCvt_Fx<1, u, opcode, asmop, "2d", VPR128, v2i64, v2f64, + shr_imm64, IntOp> { + let Inst{22} = 0b1; + } +} + +// Convert fixed-point to floating-point +defm VCVTxs2f : NeonI_N2VCvt_Fx2fp<0, 0b11100, "scvtf", + int_arm_neon_vcvtfxs2fp>; +defm VCVTxu2f : NeonI_N2VCvt_Fx2fp<1, 0b11100, "ucvtf", + int_arm_neon_vcvtfxu2fp>; + +// Convert floating-point to fixed-point +defm VCVTf2xs : NeonI_N2VCvt_Fp2fx<0, 0b11111, "fcvtzs", + int_arm_neon_vcvtfp2fxs>; +defm VCVTf2xu : NeonI_N2VCvt_Fp2fx<1, 0b11111, "fcvtzu", + int_arm_neon_vcvtfp2fxu>; + +multiclass Neon_sshll2_0<SDNode ext> +{ + def _v8i8 : PatFrag<(ops node:$Rn), + (v8i16 (ext (v8i8 (Neon_High16B node:$Rn))))>; + def _v4i16 : PatFrag<(ops node:$Rn), + (v4i32 (ext (v4i16 (Neon_High8H node:$Rn))))>; + def _v2i32 : PatFrag<(ops node:$Rn), + (v2i64 (ext (v2i32 (Neon_High4S node:$Rn))))>; +} + +defm NI_sext_high : Neon_sshll2_0<sext>; +defm NI_zext_high : Neon_sshll2_0<zext>; + + +//===----------------------------------------------------------------------===// +// Multiclasses for NeonI_Across +//===----------------------------------------------------------------------===// + +// Variant 1 + +multiclass NeonI_2VAcross_1<bit u, bits<5> opcode, + string asmop, SDPatternOperator opnode> +{ + def _1h8b: NeonI_2VAcross<0b0, u, 0b00, opcode, + (outs FPR16:$Rd), (ins VPR64:$Rn), + asmop # "\t$Rd, $Rn.8b", + [(set (v1i16 FPR16:$Rd), + (v1i16 (opnode (v8i8 VPR64:$Rn))))], + NoItinerary>; + + def _1h16b: NeonI_2VAcross<0b1, u, 0b00, opcode, + (outs FPR16:$Rd), (ins VPR128:$Rn), + asmop # "\t$Rd, $Rn.16b", + [(set (v1i16 FPR16:$Rd), + (v1i16 (opnode (v16i8 VPR128:$Rn))))], + NoItinerary>; + + def _1s4h: NeonI_2VAcross<0b0, u, 0b01, opcode, + (outs FPR32:$Rd), (ins VPR64:$Rn), + asmop # "\t$Rd, $Rn.4h", + [(set (v1i32 FPR32:$Rd), + (v1i32 (opnode (v4i16 VPR64:$Rn))))], + NoItinerary>; + + def _1s8h: NeonI_2VAcross<0b1, u, 0b01, opcode, + (outs FPR32:$Rd), (ins VPR128:$Rn), + asmop # "\t$Rd, $Rn.8h", + [(set (v1i32 FPR32:$Rd), + (v1i32 (opnode (v8i16 VPR128:$Rn))))], + NoItinerary>; + + // _1d2s doesn't exist! + + def _1d4s: NeonI_2VAcross<0b1, u, 0b10, opcode, + (outs FPR64:$Rd), (ins VPR128:$Rn), + asmop # "\t$Rd, $Rn.4s", + [(set (v1i64 FPR64:$Rd), + (v1i64 (opnode (v4i32 VPR128:$Rn))))], + NoItinerary>; +} + +defm SADDLV : NeonI_2VAcross_1<0b0, 0b00011, "saddlv", int_aarch64_neon_saddlv>; +defm UADDLV : NeonI_2VAcross_1<0b1, 0b00011, "uaddlv", int_aarch64_neon_uaddlv>; + +// Variant 2 + +multiclass NeonI_2VAcross_2<bit u, bits<5> opcode, + string asmop, SDPatternOperator opnode> +{ + def _1b8b: NeonI_2VAcross<0b0, u, 0b00, opcode, + (outs FPR8:$Rd), (ins VPR64:$Rn), + asmop # "\t$Rd, $Rn.8b", + [(set (v1i8 FPR8:$Rd), + (v1i8 (opnode (v8i8 VPR64:$Rn))))], + NoItinerary>; + + def _1b16b: NeonI_2VAcross<0b1, u, 0b00, opcode, + (outs FPR8:$Rd), (ins VPR128:$Rn), + asmop # "\t$Rd, $Rn.16b", + [(set (v1i8 FPR8:$Rd), + (v1i8 (opnode (v16i8 VPR128:$Rn))))], + NoItinerary>; + + def _1h4h: NeonI_2VAcross<0b0, u, 0b01, opcode, + (outs FPR16:$Rd), (ins VPR64:$Rn), + asmop # "\t$Rd, $Rn.4h", + [(set (v1i16 FPR16:$Rd), + (v1i16 (opnode (v4i16 VPR64:$Rn))))], + NoItinerary>; + + def _1h8h: NeonI_2VAcross<0b1, u, 0b01, opcode, + (outs FPR16:$Rd), (ins VPR128:$Rn), + asmop # "\t$Rd, $Rn.8h", + [(set (v1i16 FPR16:$Rd), + (v1i16 (opnode (v8i16 VPR128:$Rn))))], + NoItinerary>; + + // _1s2s doesn't exist! + + def _1s4s: NeonI_2VAcross<0b1, u, 0b10, opcode, + (outs FPR32:$Rd), (ins VPR128:$Rn), + asmop # "\t$Rd, $Rn.4s", + [(set (v1i32 FPR32:$Rd), + (v1i32 (opnode (v4i32 VPR128:$Rn))))], + NoItinerary>; +} + +defm SMAXV : NeonI_2VAcross_2<0b0, 0b01010, "smaxv", int_aarch64_neon_smaxv>; +defm UMAXV : NeonI_2VAcross_2<0b1, 0b01010, "umaxv", int_aarch64_neon_umaxv>; + +defm SMINV : NeonI_2VAcross_2<0b0, 0b11010, "sminv", int_aarch64_neon_sminv>; +defm UMINV : NeonI_2VAcross_2<0b1, 0b11010, "uminv", int_aarch64_neon_uminv>; + +defm ADDV : NeonI_2VAcross_2<0b0, 0b11011, "addv", int_aarch64_neon_vaddv>; + +// Variant 3 + +multiclass NeonI_2VAcross_3<bit u, bits<5> opcode, bits<2> size, + string asmop, SDPatternOperator opnode> { + def _1s4s: NeonI_2VAcross<0b1, u, size, opcode, + (outs FPR32:$Rd), (ins VPR128:$Rn), + asmop # "\t$Rd, $Rn.4s", + [(set (v1f32 FPR32:$Rd), + (v1f32 (opnode (v4f32 VPR128:$Rn))))], + NoItinerary>; +} + +defm FMAXNMV : NeonI_2VAcross_3<0b1, 0b01100, 0b00, "fmaxnmv", + int_aarch64_neon_vmaxnmv>; +defm FMINNMV : NeonI_2VAcross_3<0b1, 0b01100, 0b10, "fminnmv", + int_aarch64_neon_vminnmv>; + +defm FMAXV : NeonI_2VAcross_3<0b1, 0b01111, 0b00, "fmaxv", + int_aarch64_neon_vmaxv>; +defm FMINV : NeonI_2VAcross_3<0b1, 0b01111, 0b10, "fminv", + int_aarch64_neon_vminv>; + +// The followings are for instruction class (Perm) + +class NeonI_Permute<bit q, bits<2> size, bits<3> opcode, + string asmop, RegisterOperand OpVPR, string OpS, + SDPatternOperator opnode, ValueType Ty> + : NeonI_Perm<q, size, opcode, + (outs OpVPR:$Rd), (ins OpVPR:$Rn, OpVPR:$Rm), + asmop # "\t$Rd." # OpS # ", $Rn." # OpS # ", $Rm." # OpS, + [(set (Ty OpVPR:$Rd), + (Ty (opnode (Ty OpVPR:$Rn), (Ty OpVPR:$Rm))))], + NoItinerary>; + +multiclass NeonI_Perm_pat<bits<3> opcode, string asmop, + SDPatternOperator opnode> { + def _8b : NeonI_Permute<0b0, 0b00, opcode, asmop, + VPR64, "8b", opnode, v8i8>; + def _16b : NeonI_Permute<0b1, 0b00, opcode, asmop, + VPR128, "16b",opnode, v16i8>; + def _4h : NeonI_Permute<0b0, 0b01, opcode, asmop, + VPR64, "4h", opnode, v4i16>; + def _8h : NeonI_Permute<0b1, 0b01, opcode, asmop, + VPR128, "8h", opnode, v8i16>; + def _2s : NeonI_Permute<0b0, 0b10, opcode, asmop, + VPR64, "2s", opnode, v2i32>; + def _4s : NeonI_Permute<0b1, 0b10, opcode, asmop, + VPR128, "4s", opnode, v4i32>; + def _2d : NeonI_Permute<0b1, 0b11, opcode, asmop, + VPR128, "2d", opnode, v2i64>; +} + +defm UZP1vvv : NeonI_Perm_pat<0b001, "uzp1", Neon_uzp1>; +defm TRN1vvv : NeonI_Perm_pat<0b010, "trn1", Neon_trn1>; +defm ZIP1vvv : NeonI_Perm_pat<0b011, "zip1", Neon_zip1>; +defm UZP2vvv : NeonI_Perm_pat<0b101, "uzp2", Neon_uzp2>; +defm TRN2vvv : NeonI_Perm_pat<0b110, "trn2", Neon_trn2>; +defm ZIP2vvv : NeonI_Perm_pat<0b111, "zip2", Neon_zip2>; + +multiclass NeonI_Perm_float_pat<string INS, SDPatternOperator opnode> { + def : Pat<(v2f32 (opnode (v2f32 VPR64:$Rn), (v2f32 VPR64:$Rm))), + (!cast<Instruction>(INS # "_2s") VPR64:$Rn, VPR64:$Rm)>; + + def : Pat<(v4f32 (opnode (v4f32 VPR128:$Rn), (v4f32 VPR128:$Rm))), + (!cast<Instruction>(INS # "_4s") VPR128:$Rn, VPR128:$Rm)>; + + def : Pat<(v2f64 (opnode (v2f64 VPR128:$Rn), (v2f64 VPR128:$Rm))), + (!cast<Instruction>(INS # "_2d") VPR128:$Rn, VPR128:$Rm)>; +} + +defm : NeonI_Perm_float_pat<"UZP1vvv", Neon_uzp1>; +defm : NeonI_Perm_float_pat<"UZP2vvv", Neon_uzp2>; +defm : NeonI_Perm_float_pat<"ZIP1vvv", Neon_zip1>; +defm : NeonI_Perm_float_pat<"ZIP2vvv", Neon_zip2>; +defm : NeonI_Perm_float_pat<"TRN1vvv", Neon_trn1>; +defm : NeonI_Perm_float_pat<"TRN2vvv", Neon_trn2>; + +// The followings are for instruction class (3V Diff) + +// normal long/long2 pattern +class NeonI_3VDL<bit q, bit u, bits<2> size, bits<4> opcode, + string asmop, string ResS, string OpS, + SDPatternOperator opnode, SDPatternOperator ext, + RegisterOperand OpVPR, + ValueType ResTy, ValueType OpTy> + : NeonI_3VDiff<q, u, size, opcode, + (outs VPR128:$Rd), (ins OpVPR:$Rn, OpVPR:$Rm), + asmop # "\t$Rd." # ResS # ", $Rn." # OpS # ", $Rm." # OpS, + [(set (ResTy VPR128:$Rd), + (ResTy (opnode (ResTy (ext (OpTy OpVPR:$Rn))), + (ResTy (ext (OpTy OpVPR:$Rm))))))], + NoItinerary>; + +multiclass NeonI_3VDL_s<bit u, bits<4> opcode, + string asmop, SDPatternOperator opnode, + bit Commutable = 0> { + let isCommutable = Commutable in { + def _8h8b : NeonI_3VDL<0b0, u, 0b00, opcode, asmop, "8h", "8b", + opnode, sext, VPR64, v8i16, v8i8>; + def _4s4h : NeonI_3VDL<0b0, u, 0b01, opcode, asmop, "4s", "4h", + opnode, sext, VPR64, v4i32, v4i16>; + def _2d2s : NeonI_3VDL<0b0, u, 0b10, opcode, asmop, "2d", "2s", + opnode, sext, VPR64, v2i64, v2i32>; + } +} + +multiclass NeonI_3VDL2_s<bit u, bits<4> opcode, string asmop, + SDPatternOperator opnode, bit Commutable = 0> { + let isCommutable = Commutable in { + def _8h16b : NeonI_3VDL<0b1, u, 0b00, opcode, asmop, "8h", "16b", + opnode, NI_sext_high_v8i8, VPR128, v8i16, v16i8>; + def _4s8h : NeonI_3VDL<0b1, u, 0b01, opcode, asmop, "4s", "8h", + opnode, NI_sext_high_v4i16, VPR128, v4i32, v8i16>; + def _2d4s : NeonI_3VDL<0b1, u, 0b10, opcode, asmop, "2d", "4s", + opnode, NI_sext_high_v2i32, VPR128, v2i64, v4i32>; + } +} + +multiclass NeonI_3VDL_u<bit u, bits<4> opcode, string asmop, + SDPatternOperator opnode, bit Commutable = 0> { + let isCommutable = Commutable in { + def _8h8b : NeonI_3VDL<0b0, u, 0b00, opcode, asmop, "8h", "8b", + opnode, zext, VPR64, v8i16, v8i8>; + def _4s4h : NeonI_3VDL<0b0, u, 0b01, opcode, asmop, "4s", "4h", + opnode, zext, VPR64, v4i32, v4i16>; + def _2d2s : NeonI_3VDL<0b0, u, 0b10, opcode, asmop, "2d", "2s", + opnode, zext, VPR64, v2i64, v2i32>; + } +} + +multiclass NeonI_3VDL2_u<bit u, bits<4> opcode, string asmop, + SDPatternOperator opnode, bit Commutable = 0> { + let isCommutable = Commutable in { + def _8h16b : NeonI_3VDL<0b1, u, 0b00, opcode, asmop, "8h", "16b", + opnode, NI_zext_high_v8i8, VPR128, v8i16, v16i8>; + def _4s8h : NeonI_3VDL<0b1, u, 0b01, opcode, asmop, "4s", "8h", + opnode, NI_zext_high_v4i16, VPR128, v4i32, v8i16>; + def _2d4s : NeonI_3VDL<0b1, u, 0b10, opcode, asmop, "2d", "4s", + opnode, NI_zext_high_v2i32, VPR128, v2i64, v4i32>; + } +} + +defm SADDLvvv : NeonI_3VDL_s<0b0, 0b0000, "saddl", add, 1>; +defm UADDLvvv : NeonI_3VDL_u<0b1, 0b0000, "uaddl", add, 1>; + +defm SADDL2vvv : NeonI_3VDL2_s<0b0, 0b0000, "saddl2", add, 1>; +defm UADDL2vvv : NeonI_3VDL2_u<0b1, 0b0000, "uaddl2", add, 1>; + +defm SSUBLvvv : NeonI_3VDL_s<0b0, 0b0010, "ssubl", sub, 0>; +defm USUBLvvv : NeonI_3VDL_u<0b1, 0b0010, "usubl", sub, 0>; + +defm SSUBL2vvv : NeonI_3VDL2_s<0b0, 0b0010, "ssubl2", sub, 0>; +defm USUBL2vvv : NeonI_3VDL2_u<0b1, 0b0010, "usubl2", sub, 0>; + +// normal wide/wide2 pattern +class NeonI_3VDW<bit q, bit u, bits<2> size, bits<4> opcode, + string asmop, string ResS, string OpS, + SDPatternOperator opnode, SDPatternOperator ext, + RegisterOperand OpVPR, + ValueType ResTy, ValueType OpTy> + : NeonI_3VDiff<q, u, size, opcode, + (outs VPR128:$Rd), (ins VPR128:$Rn, OpVPR:$Rm), + asmop # "\t$Rd." # ResS # ", $Rn." # ResS # ", $Rm." # OpS, + [(set (ResTy VPR128:$Rd), + (ResTy (opnode (ResTy VPR128:$Rn), + (ResTy (ext (OpTy OpVPR:$Rm))))))], + NoItinerary>; + +multiclass NeonI_3VDW_s<bit u, bits<4> opcode, string asmop, + SDPatternOperator opnode> { + def _8h8b : NeonI_3VDW<0b0, u, 0b00, opcode, asmop, "8h", "8b", + opnode, sext, VPR64, v8i16, v8i8>; + def _4s4h : NeonI_3VDW<0b0, u, 0b01, opcode, asmop, "4s", "4h", + opnode, sext, VPR64, v4i32, v4i16>; + def _2d2s : NeonI_3VDW<0b0, u, 0b10, opcode, asmop, "2d", "2s", + opnode, sext, VPR64, v2i64, v2i32>; +} + +defm SADDWvvv : NeonI_3VDW_s<0b0, 0b0001, "saddw", add>; +defm SSUBWvvv : NeonI_3VDW_s<0b0, 0b0011, "ssubw", sub>; + +multiclass NeonI_3VDW2_s<bit u, bits<4> opcode, string asmop, + SDPatternOperator opnode> { + def _8h16b : NeonI_3VDW<0b1, u, 0b00, opcode, asmop, "8h", "16b", + opnode, NI_sext_high_v8i8, VPR128, v8i16, v16i8>; + def _4s8h : NeonI_3VDW<0b1, u, 0b01, opcode, asmop, "4s", "8h", + opnode, NI_sext_high_v4i16, VPR128, v4i32, v8i16>; + def _2d4s : NeonI_3VDW<0b1, u, 0b10, opcode, asmop, "2d", "4s", + opnode, NI_sext_high_v2i32, VPR128, v2i64, v4i32>; +} + +defm SADDW2vvv : NeonI_3VDW2_s<0b0, 0b0001, "saddw2", add>; +defm SSUBW2vvv : NeonI_3VDW2_s<0b0, 0b0011, "ssubw2", sub>; + +multiclass NeonI_3VDW_u<bit u, bits<4> opcode, string asmop, + SDPatternOperator opnode> { + def _8h8b : NeonI_3VDW<0b0, u, 0b00, opcode, asmop, "8h", "8b", + opnode, zext, VPR64, v8i16, v8i8>; + def _4s4h : NeonI_3VDW<0b0, u, 0b01, opcode, asmop, "4s", "4h", + opnode, zext, VPR64, v4i32, v4i16>; + def _2d2s : NeonI_3VDW<0b0, u, 0b10, opcode, asmop, "2d", "2s", + opnode, zext, VPR64, v2i64, v2i32>; +} + +defm UADDWvvv : NeonI_3VDW_u<0b1, 0b0001, "uaddw", add>; +defm USUBWvvv : NeonI_3VDW_u<0b1, 0b0011, "usubw", sub>; + +multiclass NeonI_3VDW2_u<bit u, bits<4> opcode, string asmop, + SDPatternOperator opnode> { + def _8h16b : NeonI_3VDW<0b1, u, 0b00, opcode, asmop, "8h", "16b", + opnode, NI_zext_high_v8i8, VPR128, v8i16, v16i8>; + def _4s8h : NeonI_3VDW<0b1, u, 0b01, opcode, asmop, "4s", "8h", + opnode, NI_zext_high_v4i16, VPR128, v4i32, v8i16>; + def _2d4s : NeonI_3VDW<0b1, u, 0b10, opcode, asmop, "2d", "4s", + opnode, NI_zext_high_v2i32, VPR128, v2i64, v4i32>; +} + +defm UADDW2vvv : NeonI_3VDW2_u<0b1, 0b0001, "uaddw2", add>; +defm USUBW2vvv : NeonI_3VDW2_u<0b1, 0b0011, "usubw2", sub>; + +// Get the high half part of the vector element. +multiclass NeonI_get_high { + def _8h : PatFrag<(ops node:$Rn), + (v8i8 (trunc (v8i16 (srl (v8i16 node:$Rn), + (v8i16 (Neon_vdup (i32 8)))))))>; + def _4s : PatFrag<(ops node:$Rn), + (v4i16 (trunc (v4i32 (srl (v4i32 node:$Rn), + (v4i32 (Neon_vdup (i32 16)))))))>; + def _2d : PatFrag<(ops node:$Rn), + (v2i32 (trunc (v2i64 (srl (v2i64 node:$Rn), + (v2i64 (Neon_vdup (i32 32)))))))>; +} + +defm NI_get_hi : NeonI_get_high; + +// pattern for addhn/subhn with 2 operands +class NeonI_3VDN_addhn_2Op<bit q, bit u, bits<2> size, bits<4> opcode, + string asmop, string ResS, string OpS, + SDPatternOperator opnode, SDPatternOperator get_hi, + ValueType ResTy, ValueType OpTy> + : NeonI_3VDiff<q, u, size, opcode, + (outs VPR64:$Rd), (ins VPR128:$Rn, VPR128:$Rm), + asmop # "\t$Rd." # ResS # ", $Rn." # OpS # ", $Rm." # OpS, + [(set (ResTy VPR64:$Rd), + (ResTy (get_hi + (OpTy (opnode (OpTy VPR128:$Rn), + (OpTy VPR128:$Rm))))))], + NoItinerary>; + +multiclass NeonI_3VDN_addhn_2Op<bit u, bits<4> opcode, string asmop, + SDPatternOperator opnode, bit Commutable = 0> { + let isCommutable = Commutable in { + def _8b8h : NeonI_3VDN_addhn_2Op<0b0, u, 0b00, opcode, asmop, "8b", "8h", + opnode, NI_get_hi_8h, v8i8, v8i16>; + def _4h4s : NeonI_3VDN_addhn_2Op<0b0, u, 0b01, opcode, asmop, "4h", "4s", + opnode, NI_get_hi_4s, v4i16, v4i32>; + def _2s2d : NeonI_3VDN_addhn_2Op<0b0, u, 0b10, opcode, asmop, "2s", "2d", + opnode, NI_get_hi_2d, v2i32, v2i64>; + } +} + +defm ADDHNvvv : NeonI_3VDN_addhn_2Op<0b0, 0b0100, "addhn", add, 1>; +defm SUBHNvvv : NeonI_3VDN_addhn_2Op<0b0, 0b0110, "subhn", sub, 0>; + +// pattern for operation with 2 operands +class NeonI_3VD_2Op<bit q, bit u, bits<2> size, bits<4> opcode, + string asmop, string ResS, string OpS, + SDPatternOperator opnode, + RegisterOperand ResVPR, RegisterOperand OpVPR, + ValueType ResTy, ValueType OpTy> + : NeonI_3VDiff<q, u, size, opcode, + (outs ResVPR:$Rd), (ins OpVPR:$Rn, OpVPR:$Rm), + asmop # "\t$Rd." # ResS # ", $Rn." # OpS # ", $Rm." # OpS, + [(set (ResTy ResVPR:$Rd), + (ResTy (opnode (OpTy OpVPR:$Rn), (OpTy OpVPR:$Rm))))], + NoItinerary>; + +// normal narrow pattern +multiclass NeonI_3VDN_2Op<bit u, bits<4> opcode, string asmop, + SDPatternOperator opnode, bit Commutable = 0> { + let isCommutable = Commutable in { + def _8b8h : NeonI_3VD_2Op<0b0, u, 0b00, opcode, asmop, "8b", "8h", + opnode, VPR64, VPR128, v8i8, v8i16>; + def _4h4s : NeonI_3VD_2Op<0b0, u, 0b01, opcode, asmop, "4h", "4s", + opnode, VPR64, VPR128, v4i16, v4i32>; + def _2s2d : NeonI_3VD_2Op<0b0, u, 0b10, opcode, asmop, "2s", "2d", + opnode, VPR64, VPR128, v2i32, v2i64>; + } +} + +defm RADDHNvvv : NeonI_3VDN_2Op<0b1, 0b0100, "raddhn", int_arm_neon_vraddhn, 1>; +defm RSUBHNvvv : NeonI_3VDN_2Op<0b1, 0b0110, "rsubhn", int_arm_neon_vrsubhn, 0>; + +// pattern for acle intrinsic with 3 operands +class NeonI_3VDN_3Op<bit q, bit u, bits<2> size, bits<4> opcode, + string asmop, string ResS, string OpS> + : NeonI_3VDiff<q, u, size, opcode, + (outs VPR128:$Rd), (ins VPR128:$src, VPR128:$Rn, VPR128:$Rm), + asmop # "\t$Rd." # ResS # ", $Rn." # OpS # ", $Rm." # OpS, + [], NoItinerary> { + let Constraints = "$src = $Rd"; + let neverHasSideEffects = 1; +} + +multiclass NeonI_3VDN_3Op_v1<bit u, bits<4> opcode, string asmop> { + def _16b8h : NeonI_3VDN_3Op<0b1, u, 0b00, opcode, asmop, "16b", "8h">; + def _8h4s : NeonI_3VDN_3Op<0b1, u, 0b01, opcode, asmop, "8h", "4s">; + def _4s2d : NeonI_3VDN_3Op<0b1, u, 0b10, opcode, asmop, "4s", "2d">; +} + +defm ADDHN2vvv : NeonI_3VDN_3Op_v1<0b0, 0b0100, "addhn2">; +defm SUBHN2vvv : NeonI_3VDN_3Op_v1<0b0, 0b0110, "subhn2">; + +defm RADDHN2vvv : NeonI_3VDN_3Op_v1<0b1, 0b0100, "raddhn2">; +defm RSUBHN2vvv : NeonI_3VDN_3Op_v1<0b1, 0b0110, "rsubhn2">; + +// Patterns have to be separate because there's a SUBREG_TO_REG in the output +// part. +class NarrowHighHalfPat<Instruction INST, ValueType DstTy, ValueType SrcTy, + SDPatternOperator coreop> + : Pat<(Neon_combine_2D (v1i64 VPR64:$src), + (v1i64 (bitconvert (DstTy (coreop (SrcTy VPR128:$Rn), + (SrcTy VPR128:$Rm)))))), + (INST (SUBREG_TO_REG (i64 0), VPR64:$src, sub_64), + VPR128:$Rn, VPR128:$Rm)>; + +// addhn2 patterns +def : NarrowHighHalfPat<ADDHN2vvv_16b8h, v8i8, v8i16, + BinOpFrag<(NI_get_hi_8h (add node:$LHS, node:$RHS))>>; +def : NarrowHighHalfPat<ADDHN2vvv_8h4s, v4i16, v4i32, + BinOpFrag<(NI_get_hi_4s (add node:$LHS, node:$RHS))>>; +def : NarrowHighHalfPat<ADDHN2vvv_4s2d, v2i32, v2i64, + BinOpFrag<(NI_get_hi_2d (add node:$LHS, node:$RHS))>>; + +// subhn2 patterns +def : NarrowHighHalfPat<SUBHN2vvv_16b8h, v8i8, v8i16, + BinOpFrag<(NI_get_hi_8h (sub node:$LHS, node:$RHS))>>; +def : NarrowHighHalfPat<SUBHN2vvv_8h4s, v4i16, v4i32, + BinOpFrag<(NI_get_hi_4s (sub node:$LHS, node:$RHS))>>; +def : NarrowHighHalfPat<SUBHN2vvv_4s2d, v2i32, v2i64, + BinOpFrag<(NI_get_hi_2d (sub node:$LHS, node:$RHS))>>; + +// raddhn2 patterns +def : NarrowHighHalfPat<RADDHN2vvv_16b8h, v8i8, v8i16, int_arm_neon_vraddhn>; +def : NarrowHighHalfPat<RADDHN2vvv_8h4s, v4i16, v4i32, int_arm_neon_vraddhn>; +def : NarrowHighHalfPat<RADDHN2vvv_4s2d, v2i32, v2i64, int_arm_neon_vraddhn>; + +// rsubhn2 patterns +def : NarrowHighHalfPat<RSUBHN2vvv_16b8h, v8i8, v8i16, int_arm_neon_vrsubhn>; +def : NarrowHighHalfPat<RSUBHN2vvv_8h4s, v4i16, v4i32, int_arm_neon_vrsubhn>; +def : NarrowHighHalfPat<RSUBHN2vvv_4s2d, v2i32, v2i64, int_arm_neon_vrsubhn>; + +// pattern that need to extend result +class NeonI_3VDL_Ext<bit q, bit u, bits<2> size, bits<4> opcode, + string asmop, string ResS, string OpS, + SDPatternOperator opnode, + RegisterOperand OpVPR, + ValueType ResTy, ValueType OpTy, ValueType OpSTy> + : NeonI_3VDiff<q, u, size, opcode, + (outs VPR128:$Rd), (ins OpVPR:$Rn, OpVPR:$Rm), + asmop # "\t$Rd." # ResS # ", $Rn." # OpS # ", $Rm." # OpS, + [(set (ResTy VPR128:$Rd), + (ResTy (zext (OpSTy (opnode (OpTy OpVPR:$Rn), + (OpTy OpVPR:$Rm))))))], + NoItinerary>; + +multiclass NeonI_3VDL_zext<bit u, bits<4> opcode, string asmop, + SDPatternOperator opnode, bit Commutable = 0> { + let isCommutable = Commutable in { + def _8h8b : NeonI_3VDL_Ext<0b0, u, 0b00, opcode, asmop, "8h", "8b", + opnode, VPR64, v8i16, v8i8, v8i8>; + def _4s4h : NeonI_3VDL_Ext<0b0, u, 0b01, opcode, asmop, "4s", "4h", + opnode, VPR64, v4i32, v4i16, v4i16>; + def _2d2s : NeonI_3VDL_Ext<0b0, u, 0b10, opcode, asmop, "2d", "2s", + opnode, VPR64, v2i64, v2i32, v2i32>; + } +} + +defm SABDLvvv : NeonI_3VDL_zext<0b0, 0b0111, "sabdl", int_arm_neon_vabds, 1>; +defm UABDLvvv : NeonI_3VDL_zext<0b1, 0b0111, "uabdl", int_arm_neon_vabdu, 1>; + +multiclass NeonI_Op_High<SDPatternOperator op> { + def _16B : PatFrag<(ops node:$Rn, node:$Rm), + (op (v8i8 (Neon_High16B node:$Rn)), + (v8i8 (Neon_High16B node:$Rm)))>; + def _8H : PatFrag<(ops node:$Rn, node:$Rm), + (op (v4i16 (Neon_High8H node:$Rn)), + (v4i16 (Neon_High8H node:$Rm)))>; + def _4S : PatFrag<(ops node:$Rn, node:$Rm), + (op (v2i32 (Neon_High4S node:$Rn)), + (v2i32 (Neon_High4S node:$Rm)))>; +} + +defm NI_sabdl_hi : NeonI_Op_High<int_arm_neon_vabds>; +defm NI_uabdl_hi : NeonI_Op_High<int_arm_neon_vabdu>; +defm NI_smull_hi : NeonI_Op_High<int_arm_neon_vmulls>; +defm NI_umull_hi : NeonI_Op_High<int_arm_neon_vmullu>; +defm NI_qdmull_hi : NeonI_Op_High<int_arm_neon_vqdmull>; +defm NI_pmull_hi : NeonI_Op_High<int_arm_neon_vmullp>; + +multiclass NeonI_3VDL_Abd_u<bit u, bits<4> opcode, string asmop, string opnode, + bit Commutable = 0> { + let isCommutable = Commutable in { + def _8h8b : NeonI_3VDL_Ext<0b1, u, 0b00, opcode, asmop, "8h", "16b", + !cast<PatFrag>(opnode # "_16B"), + VPR128, v8i16, v16i8, v8i8>; + def _4s4h : NeonI_3VDL_Ext<0b1, u, 0b01, opcode, asmop, "4s", "8h", + !cast<PatFrag>(opnode # "_8H"), + VPR128, v4i32, v8i16, v4i16>; + def _2d2s : NeonI_3VDL_Ext<0b1, u, 0b10, opcode, asmop, "2d", "4s", + !cast<PatFrag>(opnode # "_4S"), + VPR128, v2i64, v4i32, v2i32>; + } +} + +defm SABDL2vvv : NeonI_3VDL_Abd_u<0b0, 0b0111, "sabdl2", "NI_sabdl_hi", 1>; +defm UABDL2vvv : NeonI_3VDL_Abd_u<0b1, 0b0111, "uabdl2", "NI_uabdl_hi", 1>; + +// For pattern that need two operators being chained. +class NeonI_3VDL_Aba<bit q, bit u, bits<2> size, bits<4> opcode, + string asmop, string ResS, string OpS, + SDPatternOperator opnode, SDPatternOperator subop, + RegisterOperand OpVPR, + ValueType ResTy, ValueType OpTy, ValueType OpSTy> + : NeonI_3VDiff<q, u, size, opcode, + (outs VPR128:$Rd), (ins VPR128:$src, OpVPR:$Rn, OpVPR:$Rm), + asmop # "\t$Rd." # ResS # ", $Rn." # OpS # ", $Rm." # OpS, + [(set (ResTy VPR128:$Rd), + (ResTy (opnode + (ResTy VPR128:$src), + (ResTy (zext (OpSTy (subop (OpTy OpVPR:$Rn), + (OpTy OpVPR:$Rm))))))))], + NoItinerary> { + let Constraints = "$src = $Rd"; +} + +multiclass NeonI_3VDL_Aba_v1<bit u, bits<4> opcode, string asmop, + SDPatternOperator opnode, SDPatternOperator subop>{ + def _8h8b : NeonI_3VDL_Aba<0b0, u, 0b00, opcode, asmop, "8h", "8b", + opnode, subop, VPR64, v8i16, v8i8, v8i8>; + def _4s4h : NeonI_3VDL_Aba<0b0, u, 0b01, opcode, asmop, "4s", "4h", + opnode, subop, VPR64, v4i32, v4i16, v4i16>; + def _2d2s : NeonI_3VDL_Aba<0b0, u, 0b10, opcode, asmop, "2d", "2s", + opnode, subop, VPR64, v2i64, v2i32, v2i32>; +} + +defm SABALvvv : NeonI_3VDL_Aba_v1<0b0, 0b0101, "sabal", + add, int_arm_neon_vabds>; +defm UABALvvv : NeonI_3VDL_Aba_v1<0b1, 0b0101, "uabal", + add, int_arm_neon_vabdu>; + +multiclass NeonI_3VDL2_Aba_v1<bit u, bits<4> opcode, string asmop, + SDPatternOperator opnode, string subop> { + def _8h8b : NeonI_3VDL_Aba<0b1, u, 0b00, opcode, asmop, "8h", "16b", + opnode, !cast<PatFrag>(subop # "_16B"), + VPR128, v8i16, v16i8, v8i8>; + def _4s4h : NeonI_3VDL_Aba<0b1, u, 0b01, opcode, asmop, "4s", "8h", + opnode, !cast<PatFrag>(subop # "_8H"), + VPR128, v4i32, v8i16, v4i16>; + def _2d2s : NeonI_3VDL_Aba<0b1, u, 0b10, opcode, asmop, "2d", "4s", + opnode, !cast<PatFrag>(subop # "_4S"), + VPR128, v2i64, v4i32, v2i32>; +} + +defm SABAL2vvv : NeonI_3VDL2_Aba_v1<0b0, 0b0101, "sabal2", add, + "NI_sabdl_hi">; +defm UABAL2vvv : NeonI_3VDL2_Aba_v1<0b1, 0b0101, "uabal2", add, + "NI_uabdl_hi">; + +// Long pattern with 2 operands +multiclass NeonI_3VDL_2Op<bit u, bits<4> opcode, string asmop, + SDPatternOperator opnode, bit Commutable = 0> { + let isCommutable = Commutable in { + def _8h8b : NeonI_3VD_2Op<0b0, u, 0b00, opcode, asmop, "8h", "8b", + opnode, VPR128, VPR64, v8i16, v8i8>; + def _4s4h : NeonI_3VD_2Op<0b0, u, 0b01, opcode, asmop, "4s", "4h", + opnode, VPR128, VPR64, v4i32, v4i16>; + def _2d2s : NeonI_3VD_2Op<0b0, u, 0b10, opcode, asmop, "2d", "2s", + opnode, VPR128, VPR64, v2i64, v2i32>; + } +} + +defm SMULLvvv : NeonI_3VDL_2Op<0b0, 0b1100, "smull", int_arm_neon_vmulls, 1>; +defm UMULLvvv : NeonI_3VDL_2Op<0b1, 0b1100, "umull", int_arm_neon_vmullu, 1>; + +class NeonI_3VDL2_2Op_mull<bit q, bit u, bits<2> size, bits<4> opcode, + string asmop, string ResS, string OpS, + SDPatternOperator opnode, + ValueType ResTy, ValueType OpTy> + : NeonI_3VDiff<q, u, size, opcode, + (outs VPR128:$Rd), (ins VPR128:$Rn, VPR128:$Rm), + asmop # "\t$Rd." # ResS # ", $Rn." # OpS # ", $Rm." # OpS, + [(set (ResTy VPR128:$Rd), + (ResTy (opnode (OpTy VPR128:$Rn), (OpTy VPR128:$Rm))))], + NoItinerary>; + +multiclass NeonI_3VDL2_2Op_mull_v1<bit u, bits<4> opcode, string asmop, + string opnode, bit Commutable = 0> { + let isCommutable = Commutable in { + def _8h16b : NeonI_3VDL2_2Op_mull<0b1, u, 0b00, opcode, asmop, "8h", "16b", + !cast<PatFrag>(opnode # "_16B"), + v8i16, v16i8>; + def _4s8h : NeonI_3VDL2_2Op_mull<0b1, u, 0b01, opcode, asmop, "4s", "8h", + !cast<PatFrag>(opnode # "_8H"), + v4i32, v8i16>; + def _2d4s : NeonI_3VDL2_2Op_mull<0b1, u, 0b10, opcode, asmop, "2d", "4s", + !cast<PatFrag>(opnode # "_4S"), + v2i64, v4i32>; + } +} + +defm SMULL2vvv : NeonI_3VDL2_2Op_mull_v1<0b0, 0b1100, "smull2", + "NI_smull_hi", 1>; +defm UMULL2vvv : NeonI_3VDL2_2Op_mull_v1<0b1, 0b1100, "umull2", + "NI_umull_hi", 1>; + +// Long pattern with 3 operands +class NeonI_3VDL_3Op<bit q, bit u, bits<2> size, bits<4> opcode, + string asmop, string ResS, string OpS, + SDPatternOperator opnode, + ValueType ResTy, ValueType OpTy> + : NeonI_3VDiff<q, u, size, opcode, + (outs VPR128:$Rd), (ins VPR128:$src, VPR64:$Rn, VPR64:$Rm), + asmop # "\t$Rd." # ResS # ", $Rn." # OpS # ", $Rm." # OpS, + [(set (ResTy VPR128:$Rd), + (ResTy (opnode + (ResTy VPR128:$src), + (OpTy VPR64:$Rn), (OpTy VPR64:$Rm))))], + NoItinerary> { + let Constraints = "$src = $Rd"; +} + +multiclass NeonI_3VDL_3Op_v1<bit u, bits<4> opcode, string asmop, + SDPatternOperator opnode> { + def _8h8b : NeonI_3VDL_3Op<0b0, u, 0b00, opcode, asmop, "8h", "8b", + opnode, v8i16, v8i8>; + def _4s4h : NeonI_3VDL_3Op<0b0, u, 0b01, opcode, asmop, "4s", "4h", + opnode, v4i32, v4i16>; + def _2d2s : NeonI_3VDL_3Op<0b0, u, 0b10, opcode, asmop, "2d", "2s", + opnode, v2i64, v2i32>; +} + +def Neon_smlal : PatFrag<(ops node:$Rd, node:$Rn, node:$Rm), + (add node:$Rd, + (int_arm_neon_vmulls node:$Rn, node:$Rm))>; + +def Neon_umlal : PatFrag<(ops node:$Rd, node:$Rn, node:$Rm), + (add node:$Rd, + (int_arm_neon_vmullu node:$Rn, node:$Rm))>; + +def Neon_smlsl : PatFrag<(ops node:$Rd, node:$Rn, node:$Rm), + (sub node:$Rd, + (int_arm_neon_vmulls node:$Rn, node:$Rm))>; + +def Neon_umlsl : PatFrag<(ops node:$Rd, node:$Rn, node:$Rm), + (sub node:$Rd, + (int_arm_neon_vmullu node:$Rn, node:$Rm))>; + +defm SMLALvvv : NeonI_3VDL_3Op_v1<0b0, 0b1000, "smlal", Neon_smlal>; +defm UMLALvvv : NeonI_3VDL_3Op_v1<0b1, 0b1000, "umlal", Neon_umlal>; + +defm SMLSLvvv : NeonI_3VDL_3Op_v1<0b0, 0b1010, "smlsl", Neon_smlsl>; +defm UMLSLvvv : NeonI_3VDL_3Op_v1<0b1, 0b1010, "umlsl", Neon_umlsl>; + +class NeonI_3VDL2_3Op_mlas<bit q, bit u, bits<2> size, bits<4> opcode, + string asmop, string ResS, string OpS, + SDPatternOperator subop, SDPatternOperator opnode, + RegisterOperand OpVPR, + ValueType ResTy, ValueType OpTy> + : NeonI_3VDiff<q, u, size, opcode, + (outs VPR128:$Rd), (ins VPR128:$src, OpVPR:$Rn, OpVPR:$Rm), + asmop # "\t$Rd." # ResS # ", $Rn." # OpS # ", $Rm." # OpS, + [(set (ResTy VPR128:$Rd), + (ResTy (subop + (ResTy VPR128:$src), + (ResTy (opnode (OpTy OpVPR:$Rn), (OpTy OpVPR:$Rm))))))], + NoItinerary> { + let Constraints = "$src = $Rd"; +} + +multiclass NeonI_3VDL2_3Op_mlas_v1<bit u, bits<4> opcode, string asmop, + SDPatternOperator subop, string opnode> { + def _8h16b : NeonI_3VDL2_3Op_mlas<0b1, u, 0b00, opcode, asmop, "8h", "16b", + subop, !cast<PatFrag>(opnode # "_16B"), + VPR128, v8i16, v16i8>; + def _4s8h : NeonI_3VDL2_3Op_mlas<0b1, u, 0b01, opcode, asmop, "4s", "8h", + subop, !cast<PatFrag>(opnode # "_8H"), + VPR128, v4i32, v8i16>; + def _2d4s : NeonI_3VDL2_3Op_mlas<0b1, u, 0b10, opcode, asmop, "2d", "4s", + subop, !cast<PatFrag>(opnode # "_4S"), + VPR128, v2i64, v4i32>; +} + +defm SMLAL2vvv : NeonI_3VDL2_3Op_mlas_v1<0b0, 0b1000, "smlal2", + add, "NI_smull_hi">; +defm UMLAL2vvv : NeonI_3VDL2_3Op_mlas_v1<0b1, 0b1000, "umlal2", + add, "NI_umull_hi">; + +defm SMLSL2vvv : NeonI_3VDL2_3Op_mlas_v1<0b0, 0b1010, "smlsl2", + sub, "NI_smull_hi">; +defm UMLSL2vvv : NeonI_3VDL2_3Op_mlas_v1<0b1, 0b1010, "umlsl2", + sub, "NI_umull_hi">; + +multiclass NeonI_3VDL_qdmlal_3Op_v2<bit u, bits<4> opcode, string asmop, + SDPatternOperator opnode> { + def _4s4h : NeonI_3VDL2_3Op_mlas<0b0, u, 0b01, opcode, asmop, "4s", "4h", + opnode, int_arm_neon_vqdmull, + VPR64, v4i32, v4i16>; + def _2d2s : NeonI_3VDL2_3Op_mlas<0b0, u, 0b10, opcode, asmop, "2d", "2s", + opnode, int_arm_neon_vqdmull, + VPR64, v2i64, v2i32>; +} + +defm SQDMLALvvv : NeonI_3VDL_qdmlal_3Op_v2<0b0, 0b1001, "sqdmlal", + int_arm_neon_vqadds>; +defm SQDMLSLvvv : NeonI_3VDL_qdmlal_3Op_v2<0b0, 0b1011, "sqdmlsl", + int_arm_neon_vqsubs>; + +multiclass NeonI_3VDL_v2<bit u, bits<4> opcode, string asmop, + SDPatternOperator opnode, bit Commutable = 0> { + let isCommutable = Commutable in { + def _4s4h : NeonI_3VD_2Op<0b0, u, 0b01, opcode, asmop, "4s", "4h", + opnode, VPR128, VPR64, v4i32, v4i16>; + def _2d2s : NeonI_3VD_2Op<0b0, u, 0b10, opcode, asmop, "2d", "2s", + opnode, VPR128, VPR64, v2i64, v2i32>; + } +} + +defm SQDMULLvvv : NeonI_3VDL_v2<0b0, 0b1101, "sqdmull", + int_arm_neon_vqdmull, 1>; + +multiclass NeonI_3VDL2_2Op_mull_v2<bit u, bits<4> opcode, string asmop, + string opnode, bit Commutable = 0> { + let isCommutable = Commutable in { + def _4s8h : NeonI_3VDL2_2Op_mull<0b1, u, 0b01, opcode, asmop, "4s", "8h", + !cast<PatFrag>(opnode # "_8H"), + v4i32, v8i16>; + def _2d4s : NeonI_3VDL2_2Op_mull<0b1, u, 0b10, opcode, asmop, "2d", "4s", + !cast<PatFrag>(opnode # "_4S"), + v2i64, v4i32>; + } +} + +defm SQDMULL2vvv : NeonI_3VDL2_2Op_mull_v2<0b0, 0b1101, "sqdmull2", + "NI_qdmull_hi", 1>; + +multiclass NeonI_3VDL2_3Op_qdmlal_v2<bit u, bits<4> opcode, string asmop, + SDPatternOperator opnode> { + def _4s8h : NeonI_3VDL2_3Op_mlas<0b1, u, 0b01, opcode, asmop, "4s", "8h", + opnode, NI_qdmull_hi_8H, + VPR128, v4i32, v8i16>; + def _2d4s : NeonI_3VDL2_3Op_mlas<0b1, u, 0b10, opcode, asmop, "2d", "4s", + opnode, NI_qdmull_hi_4S, + VPR128, v2i64, v4i32>; +} + +defm SQDMLAL2vvv : NeonI_3VDL2_3Op_qdmlal_v2<0b0, 0b1001, "sqdmlal2", + int_arm_neon_vqadds>; +defm SQDMLSL2vvv : NeonI_3VDL2_3Op_qdmlal_v2<0b0, 0b1011, "sqdmlsl2", + int_arm_neon_vqsubs>; + +multiclass NeonI_3VDL_v3<bit u, bits<4> opcode, string asmop, + SDPatternOperator opnode, bit Commutable = 0> { + let isCommutable = Commutable in { + def _8h8b : NeonI_3VD_2Op<0b0, u, 0b00, opcode, asmop, "8h", "8b", + opnode, VPR128, VPR64, v8i16, v8i8>; + + def _1q1d : NeonI_3VDiff<0b0, u, 0b11, opcode, + (outs VPR128:$Rd), (ins VPR64:$Rn, VPR64:$Rm), + asmop # "\t$Rd.1q, $Rn.1d, $Rm.1d", + [], NoItinerary>; + } +} + +defm PMULLvvv : NeonI_3VDL_v3<0b0, 0b1110, "pmull", int_arm_neon_vmullp, 1>; + +multiclass NeonI_3VDL2_2Op_mull_v3<bit u, bits<4> opcode, string asmop, + string opnode, bit Commutable = 0> { + let isCommutable = Commutable in { + def _8h16b : NeonI_3VDL2_2Op_mull<0b1, u, 0b00, opcode, asmop, "8h", "16b", + !cast<PatFrag>(opnode # "_16B"), + v8i16, v16i8>; + + def _1q2d : NeonI_3VDiff<0b1, u, 0b11, opcode, + (outs VPR128:$Rd), (ins VPR128:$Rn, VPR128:$Rm), + asmop # "\t$Rd.1q, $Rn.2d, $Rm.2d", + [], NoItinerary>; + } +} + +defm PMULL2vvv : NeonI_3VDL2_2Op_mull_v3<0b0, 0b1110, "pmull2", "NI_pmull_hi", + 1>; + +// End of implementation for instruction class (3V Diff) + +// The followings are vector load/store multiple N-element structure +// (class SIMD lselem). + +// ld1: load multiple 1-element structure to 1/2/3/4 registers. +// ld2/ld3/ld4: load multiple N-element structure to N registers (N = 2, 3, 4). +// The structure consists of a sequence of sets of N values. +// The first element of the structure is placed in the first lane +// of the first first vector, the second element in the first lane +// of the second vector, and so on. +// E.g. LD1_3V_2S will load 32-bit elements {A, B, C, D, E, F} sequentially into +// the three 64-bit vectors list {BA, DC, FE}. +// E.g. LD3_2S will load 32-bit elements {A, B, C, D, E, F} into the three +// 64-bit vectors list {DA, EB, FC}. +// Store instructions store multiple structure to N registers like load. + + +class NeonI_LDVList<bit q, bits<4> opcode, bits<2> size, + RegisterOperand VecList, string asmop> + : NeonI_LdStMult<q, 1, opcode, size, + (outs VecList:$Rt), (ins GPR64xsp:$Rn), + asmop # "\t$Rt, [$Rn]", + [], + NoItinerary> { + let mayLoad = 1; + let neverHasSideEffects = 1; +} + +multiclass LDVList_BHSD<bits<4> opcode, string List, string asmop> { + def _8B : NeonI_LDVList<0, opcode, 0b00, + !cast<RegisterOperand>(List # "8B_operand"), asmop>; + + def _4H : NeonI_LDVList<0, opcode, 0b01, + !cast<RegisterOperand>(List # "4H_operand"), asmop>; + + def _2S : NeonI_LDVList<0, opcode, 0b10, + !cast<RegisterOperand>(List # "2S_operand"), asmop>; + + def _16B : NeonI_LDVList<1, opcode, 0b00, + !cast<RegisterOperand>(List # "16B_operand"), asmop>; + + def _8H : NeonI_LDVList<1, opcode, 0b01, + !cast<RegisterOperand>(List # "8H_operand"), asmop>; + + def _4S : NeonI_LDVList<1, opcode, 0b10, + !cast<RegisterOperand>(List # "4S_operand"), asmop>; + + def _2D : NeonI_LDVList<1, opcode, 0b11, + !cast<RegisterOperand>(List # "2D_operand"), asmop>; +} + +// Load multiple N-element structure to N consecutive registers (N = 1,2,3,4) +defm LD1 : LDVList_BHSD<0b0111, "VOne", "ld1">; +def LD1_1D : NeonI_LDVList<0, 0b0111, 0b11, VOne1D_operand, "ld1">; + +defm LD2 : LDVList_BHSD<0b1000, "VPair", "ld2">; + +defm LD3 : LDVList_BHSD<0b0100, "VTriple", "ld3">; + +defm LD4 : LDVList_BHSD<0b0000, "VQuad", "ld4">; + +// Load multiple 1-element structure to N consecutive registers (N = 2,3,4) +defm LD1x2 : LDVList_BHSD<0b1010, "VPair", "ld1">; +def LD1x2_1D : NeonI_LDVList<0, 0b1010, 0b11, VPair1D_operand, "ld1">; + +defm LD1x3 : LDVList_BHSD<0b0110, "VTriple", "ld1">; +def LD1x3_1D : NeonI_LDVList<0, 0b0110, 0b11, VTriple1D_operand, "ld1">; + +defm LD1x4 : LDVList_BHSD<0b0010, "VQuad", "ld1">; +def LD1x4_1D : NeonI_LDVList<0, 0b0010, 0b11, VQuad1D_operand, "ld1">; + +class NeonI_STVList<bit q, bits<4> opcode, bits<2> size, + RegisterOperand VecList, string asmop> + : NeonI_LdStMult<q, 0, opcode, size, + (outs), (ins GPR64xsp:$Rn, VecList:$Rt), + asmop # "\t$Rt, [$Rn]", + [], + NoItinerary> { + let mayStore = 1; + let neverHasSideEffects = 1; +} + +multiclass STVList_BHSD<bits<4> opcode, string List, string asmop> { + def _8B : NeonI_STVList<0, opcode, 0b00, + !cast<RegisterOperand>(List # "8B_operand"), asmop>; + + def _4H : NeonI_STVList<0, opcode, 0b01, + !cast<RegisterOperand>(List # "4H_operand"), asmop>; + + def _2S : NeonI_STVList<0, opcode, 0b10, + !cast<RegisterOperand>(List # "2S_operand"), asmop>; + + def _16B : NeonI_STVList<1, opcode, 0b00, + !cast<RegisterOperand>(List # "16B_operand"), asmop>; + + def _8H : NeonI_STVList<1, opcode, 0b01, + !cast<RegisterOperand>(List # "8H_operand"), asmop>; + + def _4S : NeonI_STVList<1, opcode, 0b10, + !cast<RegisterOperand>(List # "4S_operand"), asmop>; + + def _2D : NeonI_STVList<1, opcode, 0b11, + !cast<RegisterOperand>(List # "2D_operand"), asmop>; +} + +// Store multiple N-element structures from N registers (N = 1,2,3,4) +defm ST1 : STVList_BHSD<0b0111, "VOne", "st1">; +def ST1_1D : NeonI_STVList<0, 0b0111, 0b11, VOne1D_operand, "st1">; + +defm ST2 : STVList_BHSD<0b1000, "VPair", "st2">; + +defm ST3 : STVList_BHSD<0b0100, "VTriple", "st3">; + +defm ST4 : STVList_BHSD<0b0000, "VQuad", "st4">; + +// Store multiple 1-element structures from N consecutive registers (N = 2,3,4) +defm ST1x2 : STVList_BHSD<0b1010, "VPair", "st1">; +def ST1x2_1D : NeonI_STVList<0, 0b1010, 0b11, VPair1D_operand, "st1">; + +defm ST1x3 : STVList_BHSD<0b0110, "VTriple", "st1">; +def ST1x3_1D : NeonI_STVList<0, 0b0110, 0b11, VTriple1D_operand, "st1">; + +defm ST1x4 : STVList_BHSD<0b0010, "VQuad", "st1">; +def ST1x4_1D : NeonI_STVList<0, 0b0010, 0b11, VQuad1D_operand, "st1">; + +def : Pat<(v2f64 (load GPR64xsp:$addr)), (LD1_2D GPR64xsp:$addr)>; +def : Pat<(v2i64 (load GPR64xsp:$addr)), (LD1_2D GPR64xsp:$addr)>; + +def : Pat<(v4f32 (load GPR64xsp:$addr)), (LD1_4S GPR64xsp:$addr)>; +def : Pat<(v4i32 (load GPR64xsp:$addr)), (LD1_4S GPR64xsp:$addr)>; + +def : Pat<(v8i16 (load GPR64xsp:$addr)), (LD1_8H GPR64xsp:$addr)>; +def : Pat<(v16i8 (load GPR64xsp:$addr)), (LD1_16B GPR64xsp:$addr)>; + +def : Pat<(v1f64 (load GPR64xsp:$addr)), (LD1_1D GPR64xsp:$addr)>; +def : Pat<(v1i64 (load GPR64xsp:$addr)), (LD1_1D GPR64xsp:$addr)>; + +def : Pat<(v2f32 (load GPR64xsp:$addr)), (LD1_2S GPR64xsp:$addr)>; +def : Pat<(v2i32 (load GPR64xsp:$addr)), (LD1_2S GPR64xsp:$addr)>; + +def : Pat<(v4i16 (load GPR64xsp:$addr)), (LD1_4H GPR64xsp:$addr)>; +def : Pat<(v8i8 (load GPR64xsp:$addr)), (LD1_8B GPR64xsp:$addr)>; + +def : Pat<(store (v2i64 VPR128:$value), GPR64xsp:$addr), + (ST1_2D GPR64xsp:$addr, VPR128:$value)>; +def : Pat<(store (v2f64 VPR128:$value), GPR64xsp:$addr), + (ST1_2D GPR64xsp:$addr, VPR128:$value)>; + +def : Pat<(store (v4i32 VPR128:$value), GPR64xsp:$addr), + (ST1_4S GPR64xsp:$addr, VPR128:$value)>; +def : Pat<(store (v4f32 VPR128:$value), GPR64xsp:$addr), + (ST1_4S GPR64xsp:$addr, VPR128:$value)>; + +def : Pat<(store (v8i16 VPR128:$value), GPR64xsp:$addr), + (ST1_8H GPR64xsp:$addr, VPR128:$value)>; +def : Pat<(store (v16i8 VPR128:$value), GPR64xsp:$addr), + (ST1_16B GPR64xsp:$addr, VPR128:$value)>; + +def : Pat<(store (v1i64 VPR64:$value), GPR64xsp:$addr), + (ST1_1D GPR64xsp:$addr, VPR64:$value)>; +def : Pat<(store (v1f64 VPR64:$value), GPR64xsp:$addr), + (ST1_1D GPR64xsp:$addr, VPR64:$value)>; + +def : Pat<(store (v2i32 VPR64:$value), GPR64xsp:$addr), + (ST1_2S GPR64xsp:$addr, VPR64:$value)>; +def : Pat<(store (v2f32 VPR64:$value), GPR64xsp:$addr), + (ST1_2S GPR64xsp:$addr, VPR64:$value)>; + +def : Pat<(store (v4i16 VPR64:$value), GPR64xsp:$addr), + (ST1_4H GPR64xsp:$addr, VPR64:$value)>; +def : Pat<(store (v8i8 VPR64:$value), GPR64xsp:$addr), + (ST1_8B GPR64xsp:$addr, VPR64:$value)>; + +// End of vector load/store multiple N-element structure(class SIMD lselem) + +// The followings are post-index vector load/store multiple N-element +// structure(class SIMD lselem-post) +def exact1_asmoperand : AsmOperandClass { + let Name = "Exact1"; + let PredicateMethod = "isExactImm<1>"; + let RenderMethod = "addImmOperands"; +} +def uimm_exact1 : Operand<i32>, ImmLeaf<i32, [{return Imm == 1;}]> { + let ParserMatchClass = exact1_asmoperand; +} + +def exact2_asmoperand : AsmOperandClass { + let Name = "Exact2"; + let PredicateMethod = "isExactImm<2>"; + let RenderMethod = "addImmOperands"; +} +def uimm_exact2 : Operand<i32>, ImmLeaf<i32, [{return Imm == 2;}]> { + let ParserMatchClass = exact2_asmoperand; +} + +def exact3_asmoperand : AsmOperandClass { + let Name = "Exact3"; + let PredicateMethod = "isExactImm<3>"; + let RenderMethod = "addImmOperands"; +} +def uimm_exact3 : Operand<i32>, ImmLeaf<i32, [{return Imm == 3;}]> { + let ParserMatchClass = exact3_asmoperand; +} + +def exact4_asmoperand : AsmOperandClass { + let Name = "Exact4"; + let PredicateMethod = "isExactImm<4>"; + let RenderMethod = "addImmOperands"; +} +def uimm_exact4 : Operand<i32>, ImmLeaf<i32, [{return Imm == 4;}]> { + let ParserMatchClass = exact4_asmoperand; +} + +def exact6_asmoperand : AsmOperandClass { + let Name = "Exact6"; + let PredicateMethod = "isExactImm<6>"; + let RenderMethod = "addImmOperands"; +} +def uimm_exact6 : Operand<i32>, ImmLeaf<i32, [{return Imm == 6;}]> { + let ParserMatchClass = exact6_asmoperand; +} + +def exact8_asmoperand : AsmOperandClass { + let Name = "Exact8"; + let PredicateMethod = "isExactImm<8>"; + let RenderMethod = "addImmOperands"; +} +def uimm_exact8 : Operand<i32>, ImmLeaf<i32, [{return Imm == 8;}]> { + let ParserMatchClass = exact8_asmoperand; +} + +def exact12_asmoperand : AsmOperandClass { + let Name = "Exact12"; + let PredicateMethod = "isExactImm<12>"; + let RenderMethod = "addImmOperands"; +} +def uimm_exact12 : Operand<i32>, ImmLeaf<i32, [{return Imm == 12;}]> { + let ParserMatchClass = exact12_asmoperand; +} + +def exact16_asmoperand : AsmOperandClass { + let Name = "Exact16"; + let PredicateMethod = "isExactImm<16>"; + let RenderMethod = "addImmOperands"; +} +def uimm_exact16 : Operand<i32>, ImmLeaf<i32, [{return Imm == 16;}]> { + let ParserMatchClass = exact16_asmoperand; +} + +def exact24_asmoperand : AsmOperandClass { + let Name = "Exact24"; + let PredicateMethod = "isExactImm<24>"; + let RenderMethod = "addImmOperands"; +} +def uimm_exact24 : Operand<i32>, ImmLeaf<i32, [{return Imm == 24;}]> { + let ParserMatchClass = exact24_asmoperand; +} + +def exact32_asmoperand : AsmOperandClass { + let Name = "Exact32"; + let PredicateMethod = "isExactImm<32>"; + let RenderMethod = "addImmOperands"; +} +def uimm_exact32 : Operand<i32>, ImmLeaf<i32, [{return Imm == 32;}]> { + let ParserMatchClass = exact32_asmoperand; +} + +def exact48_asmoperand : AsmOperandClass { + let Name = "Exact48"; + let PredicateMethod = "isExactImm<48>"; + let RenderMethod = "addImmOperands"; +} +def uimm_exact48 : Operand<i32>, ImmLeaf<i32, [{return Imm == 48;}]> { + let ParserMatchClass = exact48_asmoperand; +} + +def exact64_asmoperand : AsmOperandClass { + let Name = "Exact64"; + let PredicateMethod = "isExactImm<64>"; + let RenderMethod = "addImmOperands"; +} +def uimm_exact64 : Operand<i32>, ImmLeaf<i32, [{return Imm == 64;}]> { + let ParserMatchClass = exact64_asmoperand; +} + +multiclass NeonI_LDWB_VList<bit q, bits<4> opcode, bits<2> size, + RegisterOperand VecList, Operand ImmTy, + string asmop> { + let Constraints = "$Rn = $wb", mayLoad = 1, neverHasSideEffects = 1, + DecoderMethod = "DecodeVLDSTPostInstruction" in { + def _fixed : NeonI_LdStMult_Post<q, 1, opcode, size, + (outs VecList:$Rt, GPR64xsp:$wb), + (ins GPR64xsp:$Rn, ImmTy:$amt), + asmop # "\t$Rt, [$Rn], $amt", + [], + NoItinerary> { + let Rm = 0b11111; + } + + def _register : NeonI_LdStMult_Post<q, 1, opcode, size, + (outs VecList:$Rt, GPR64xsp:$wb), + (ins GPR64xsp:$Rn, GPR64noxzr:$Rm), + asmop # "\t$Rt, [$Rn], $Rm", + [], + NoItinerary>; + } +} + +multiclass LDWB_VList_BHSD<bits<4> opcode, string List, Operand ImmTy, + Operand ImmTy2, string asmop> { + defm _8B : NeonI_LDWB_VList<0, opcode, 0b00, + !cast<RegisterOperand>(List # "8B_operand"), + ImmTy, asmop>; + + defm _4H : NeonI_LDWB_VList<0, opcode, 0b01, + !cast<RegisterOperand>(List # "4H_operand"), + ImmTy, asmop>; + + defm _2S : NeonI_LDWB_VList<0, opcode, 0b10, + !cast<RegisterOperand>(List # "2S_operand"), + ImmTy, asmop>; + + defm _16B : NeonI_LDWB_VList<1, opcode, 0b00, + !cast<RegisterOperand>(List # "16B_operand"), + ImmTy2, asmop>; + + defm _8H : NeonI_LDWB_VList<1, opcode, 0b01, + !cast<RegisterOperand>(List # "8H_operand"), + ImmTy2, asmop>; + + defm _4S : NeonI_LDWB_VList<1, opcode, 0b10, + !cast<RegisterOperand>(List # "4S_operand"), + ImmTy2, asmop>; + + defm _2D : NeonI_LDWB_VList<1, opcode, 0b11, + !cast<RegisterOperand>(List # "2D_operand"), + ImmTy2, asmop>; +} + +// Post-index load multiple N-element structures from N registers (N = 1,2,3,4) +defm LD1WB : LDWB_VList_BHSD<0b0111, "VOne", uimm_exact8, uimm_exact16, "ld1">; +defm LD1WB_1D : NeonI_LDWB_VList<0, 0b0111, 0b11, VOne1D_operand, uimm_exact8, + "ld1">; + +defm LD2WB : LDWB_VList_BHSD<0b1000, "VPair", uimm_exact16, uimm_exact32, "ld2">; + +defm LD3WB : LDWB_VList_BHSD<0b0100, "VTriple", uimm_exact24, uimm_exact48, + "ld3">; + +defm LD4WB : LDWB_VList_BHSD<0b0000, "VQuad", uimm_exact32, uimm_exact64, "ld4">; + +// Post-index load multiple 1-element structures from N consecutive registers +// (N = 2,3,4) +defm LD1x2WB : LDWB_VList_BHSD<0b1010, "VPair", uimm_exact16, uimm_exact32, + "ld1">; +defm LD1x2WB_1D : NeonI_LDWB_VList<0, 0b1010, 0b11, VPair1D_operand, + uimm_exact16, "ld1">; + +defm LD1x3WB : LDWB_VList_BHSD<0b0110, "VTriple", uimm_exact24, uimm_exact48, + "ld1">; +defm LD1x3WB_1D : NeonI_LDWB_VList<0, 0b0110, 0b11, VTriple1D_operand, + uimm_exact24, "ld1">; + +defm LD1x4WB : LDWB_VList_BHSD<0b0010, "VQuad", uimm_exact32, uimm_exact64, + "ld1">; +defm LD1x4WB_1D : NeonI_LDWB_VList<0, 0b0010, 0b11, VQuad1D_operand, + uimm_exact32, "ld1">; + +multiclass NeonI_STWB_VList<bit q, bits<4> opcode, bits<2> size, + RegisterOperand VecList, Operand ImmTy, + string asmop> { + let Constraints = "$Rn = $wb", mayStore = 1, neverHasSideEffects = 1, + DecoderMethod = "DecodeVLDSTPostInstruction" in { + def _fixed : NeonI_LdStMult_Post<q, 0, opcode, size, + (outs GPR64xsp:$wb), + (ins GPR64xsp:$Rn, ImmTy:$amt, VecList:$Rt), + asmop # "\t$Rt, [$Rn], $amt", + [], + NoItinerary> { + let Rm = 0b11111; + } + + def _register : NeonI_LdStMult_Post<q, 0, opcode, size, + (outs GPR64xsp:$wb), + (ins GPR64xsp:$Rn, GPR64noxzr:$Rm, VecList:$Rt), + asmop # "\t$Rt, [$Rn], $Rm", + [], + NoItinerary>; + } +} + +multiclass STWB_VList_BHSD<bits<4> opcode, string List, Operand ImmTy, + Operand ImmTy2, string asmop> { + defm _8B : NeonI_STWB_VList<0, opcode, 0b00, + !cast<RegisterOperand>(List # "8B_operand"), ImmTy, asmop>; + + defm _4H : NeonI_STWB_VList<0, opcode, 0b01, + !cast<RegisterOperand>(List # "4H_operand"), + ImmTy, asmop>; + + defm _2S : NeonI_STWB_VList<0, opcode, 0b10, + !cast<RegisterOperand>(List # "2S_operand"), + ImmTy, asmop>; + + defm _16B : NeonI_STWB_VList<1, opcode, 0b00, + !cast<RegisterOperand>(List # "16B_operand"), + ImmTy2, asmop>; + + defm _8H : NeonI_STWB_VList<1, opcode, 0b01, + !cast<RegisterOperand>(List # "8H_operand"), + ImmTy2, asmop>; + + defm _4S : NeonI_STWB_VList<1, opcode, 0b10, + !cast<RegisterOperand>(List # "4S_operand"), + ImmTy2, asmop>; + + defm _2D : NeonI_STWB_VList<1, opcode, 0b11, + !cast<RegisterOperand>(List # "2D_operand"), + ImmTy2, asmop>; +} + +// Post-index load multiple N-element structures from N registers (N = 1,2,3,4) +defm ST1WB : STWB_VList_BHSD<0b0111, "VOne", uimm_exact8, uimm_exact16, "st1">; +defm ST1WB_1D : NeonI_STWB_VList<0, 0b0111, 0b11, VOne1D_operand, uimm_exact8, + "st1">; + +defm ST2WB : STWB_VList_BHSD<0b1000, "VPair", uimm_exact16, uimm_exact32, "st2">; + +defm ST3WB : STWB_VList_BHSD<0b0100, "VTriple", uimm_exact24, uimm_exact48, + "st3">; + +defm ST4WB : STWB_VList_BHSD<0b0000, "VQuad", uimm_exact32, uimm_exact64, "st4">; + +// Post-index load multiple 1-element structures from N consecutive registers +// (N = 2,3,4) +defm ST1x2WB : STWB_VList_BHSD<0b1010, "VPair", uimm_exact16, uimm_exact32, + "st1">; +defm ST1x2WB_1D : NeonI_STWB_VList<0, 0b1010, 0b11, VPair1D_operand, + uimm_exact16, "st1">; + +defm ST1x3WB : STWB_VList_BHSD<0b0110, "VTriple", uimm_exact24, uimm_exact48, + "st1">; +defm ST1x3WB_1D : NeonI_STWB_VList<0, 0b0110, 0b11, VTriple1D_operand, + uimm_exact24, "st1">; + +defm ST1x4WB : STWB_VList_BHSD<0b0010, "VQuad", uimm_exact32, uimm_exact64, + "st1">; +defm ST1x4WB_1D : NeonI_STWB_VList<0, 0b0010, 0b11, VQuad1D_operand, + uimm_exact32, "st1">; + +// End of post-index vector load/store multiple N-element structure +// (class SIMD lselem-post) + +// The followings are vector load/store single N-element structure +// (class SIMD lsone). +def neon_uimm0_bare : Operand<i64>, + ImmLeaf<i64, [{return Imm == 0;}]> { + let ParserMatchClass = neon_uimm0_asmoperand; + let PrintMethod = "printUImmBareOperand"; +} + +def neon_uimm1_bare : Operand<i64>, + ImmLeaf<i64, [{return Imm < 2;}]> { + let ParserMatchClass = neon_uimm1_asmoperand; + let PrintMethod = "printUImmBareOperand"; +} + +def neon_uimm2_bare : Operand<i64>, + ImmLeaf<i64, [{return Imm < 4;}]> { + let ParserMatchClass = neon_uimm2_asmoperand; + let PrintMethod = "printUImmBareOperand"; +} + +def neon_uimm3_bare : Operand<i64>, + ImmLeaf<i64, [{return Imm < 8;}]> { + let ParserMatchClass = uimm3_asmoperand; + let PrintMethod = "printUImmBareOperand"; +} + +def neon_uimm4_bare : Operand<i64>, + ImmLeaf<i64, [{return Imm < 16;}]> { + let ParserMatchClass = uimm4_asmoperand; + let PrintMethod = "printUImmBareOperand"; +} + +class NeonI_LDN_Dup<bit q, bit r, bits<3> opcode, bits<2> size, + RegisterOperand VecList, string asmop> + : NeonI_LdOne_Dup<q, r, opcode, size, + (outs VecList:$Rt), (ins GPR64xsp:$Rn), + asmop # "\t$Rt, [$Rn]", + [], + NoItinerary> { + let mayLoad = 1; + let neverHasSideEffects = 1; +} + +multiclass LDN_Dup_BHSD<bit r, bits<3> opcode, string List, string asmop> { + def _8B : NeonI_LDN_Dup<0, r, opcode, 0b00, + !cast<RegisterOperand>(List # "8B_operand"), asmop>; + + def _4H : NeonI_LDN_Dup<0, r, opcode, 0b01, + !cast<RegisterOperand>(List # "4H_operand"), asmop>; + + def _2S : NeonI_LDN_Dup<0, r, opcode, 0b10, + !cast<RegisterOperand>(List # "2S_operand"), asmop>; + + def _1D : NeonI_LDN_Dup<0, r, opcode, 0b11, + !cast<RegisterOperand>(List # "1D_operand"), asmop>; + + def _16B : NeonI_LDN_Dup<1, r, opcode, 0b00, + !cast<RegisterOperand>(List # "16B_operand"), asmop>; + + def _8H : NeonI_LDN_Dup<1, r, opcode, 0b01, + !cast<RegisterOperand>(List # "8H_operand"), asmop>; + + def _4S : NeonI_LDN_Dup<1, r, opcode, 0b10, + !cast<RegisterOperand>(List # "4S_operand"), asmop>; + + def _2D : NeonI_LDN_Dup<1, r, opcode, 0b11, + !cast<RegisterOperand>(List # "2D_operand"), asmop>; +} + +// Load single 1-element structure to all lanes of 1 register +defm LD1R : LDN_Dup_BHSD<0b0, 0b110, "VOne", "ld1r">; + +// Load single N-element structure to all lanes of N consecutive +// registers (N = 2,3,4) +defm LD2R : LDN_Dup_BHSD<0b1, 0b110, "VPair", "ld2r">; +defm LD3R : LDN_Dup_BHSD<0b0, 0b111, "VTriple", "ld3r">; +defm LD4R : LDN_Dup_BHSD<0b1, 0b111, "VQuad", "ld4r">; + + +class LD1R_pattern <ValueType VTy, ValueType DTy, PatFrag LoadOp, + Instruction INST> + : Pat<(VTy (Neon_vdup (DTy (LoadOp GPR64xsp:$Rn)))), + (VTy (INST GPR64xsp:$Rn))>; + +// Match all LD1R instructions +def : LD1R_pattern<v8i8, i32, extloadi8, LD1R_8B>; + +def : LD1R_pattern<v16i8, i32, extloadi8, LD1R_16B>; + +def : LD1R_pattern<v4i16, i32, extloadi16, LD1R_4H>; + +def : LD1R_pattern<v8i16, i32, extloadi16, LD1R_8H>; + +def : LD1R_pattern<v2i32, i32, load, LD1R_2S>; +def : LD1R_pattern<v2f32, f32, load, LD1R_2S>; + +def : LD1R_pattern<v4i32, i32, load, LD1R_4S>; +def : LD1R_pattern<v4f32, f32, load, LD1R_4S>; + +def : LD1R_pattern<v1i64, i64, load, LD1R_1D>; +def : LD1R_pattern<v1f64, f64, load, LD1R_1D>; + +def : LD1R_pattern<v2i64, i64, load, LD1R_2D>; +def : LD1R_pattern<v2f64, f64, load, LD1R_2D>; + + +multiclass VectorList_Bare_BHSD<string PREFIX, int Count, + RegisterClass RegList> { + defm B : VectorList_operands<PREFIX, "B", Count, RegList>; + defm H : VectorList_operands<PREFIX, "H", Count, RegList>; + defm S : VectorList_operands<PREFIX, "S", Count, RegList>; + defm D : VectorList_operands<PREFIX, "D", Count, RegList>; +} + +// Special vector list operand of 128-bit vectors with bare layout. +// i.e. only show ".b", ".h", ".s", ".d" +defm VOne : VectorList_Bare_BHSD<"VOne", 1, FPR128>; +defm VPair : VectorList_Bare_BHSD<"VPair", 2, QPair>; +defm VTriple : VectorList_Bare_BHSD<"VTriple", 3, QTriple>; +defm VQuad : VectorList_Bare_BHSD<"VQuad", 4, QQuad>; + +class NeonI_LDN_Lane<bit r, bits<2> op2_1, bit op0, RegisterOperand VList, + Operand ImmOp, string asmop> + : NeonI_LdStOne_Lane<1, r, op2_1, op0, + (outs VList:$Rt), + (ins GPR64xsp:$Rn, VList:$src, ImmOp:$lane), + asmop # "\t$Rt[$lane], [$Rn]", + [], + NoItinerary> { + let mayLoad = 1; + let neverHasSideEffects = 1; + let hasExtraDefRegAllocReq = 1; + let Constraints = "$src = $Rt"; +} + +multiclass LDN_Lane_BHSD<bit r, bit op0, string List, string asmop> { + def _B : NeonI_LDN_Lane<r, 0b00, op0, + !cast<RegisterOperand>(List # "B_operand"), + neon_uimm4_bare, asmop> { + let Inst{12-10} = lane{2-0}; + let Inst{30} = lane{3}; + } + + def _H : NeonI_LDN_Lane<r, 0b01, op0, + !cast<RegisterOperand>(List # "H_operand"), + neon_uimm3_bare, asmop> { + let Inst{12-10} = {lane{1}, lane{0}, 0b0}; + let Inst{30} = lane{2}; + } + + def _S : NeonI_LDN_Lane<r, 0b10, op0, + !cast<RegisterOperand>(List # "S_operand"), + neon_uimm2_bare, asmop> { + let Inst{12-10} = {lane{0}, 0b0, 0b0}; + let Inst{30} = lane{1}; + } + + def _D : NeonI_LDN_Lane<r, 0b10, op0, + !cast<RegisterOperand>(List # "D_operand"), + neon_uimm1_bare, asmop> { + let Inst{12-10} = 0b001; + let Inst{30} = lane{0}; + } +} + +// Load single 1-element structure to one lane of 1 register. +defm LD1LN : LDN_Lane_BHSD<0b0, 0b0, "VOne", "ld1">; + +// Load single N-element structure to one lane of N consecutive registers +// (N = 2,3,4) +defm LD2LN : LDN_Lane_BHSD<0b1, 0b0, "VPair", "ld2">; +defm LD3LN : LDN_Lane_BHSD<0b0, 0b1, "VTriple", "ld3">; +defm LD4LN : LDN_Lane_BHSD<0b1, 0b1, "VQuad", "ld4">; + +multiclass LD1LN_patterns<ValueType VTy, ValueType VTy2, ValueType DTy, + Operand ImmOp, Operand ImmOp2, PatFrag LoadOp, + Instruction INST> { + def : Pat<(VTy (vector_insert (VTy VPR64:$src), + (DTy (LoadOp GPR64xsp:$Rn)), (ImmOp:$lane))), + (VTy (EXTRACT_SUBREG + (INST GPR64xsp:$Rn, + (SUBREG_TO_REG (i64 0), VPR64:$src, sub_64), + ImmOp:$lane), + sub_64))>; + + def : Pat<(VTy2 (vector_insert (VTy2 VPR128:$src), + (DTy (LoadOp GPR64xsp:$Rn)), (ImmOp2:$lane))), + (VTy2 (INST GPR64xsp:$Rn, VPR128:$src, ImmOp2:$lane))>; +} + +// Match all LD1LN instructions +defm : LD1LN_patterns<v8i8, v16i8, i32, neon_uimm3_bare, neon_uimm4_bare, + extloadi8, LD1LN_B>; + +defm : LD1LN_patterns<v4i16, v8i16, i32, neon_uimm2_bare, neon_uimm3_bare, + extloadi16, LD1LN_H>; + +defm : LD1LN_patterns<v2i32, v4i32, i32, neon_uimm1_bare, neon_uimm2_bare, + load, LD1LN_S>; +defm : LD1LN_patterns<v2f32, v4f32, f32, neon_uimm1_bare, neon_uimm2_bare, + load, LD1LN_S>; + +defm : LD1LN_patterns<v1i64, v2i64, i64, neon_uimm0_bare, neon_uimm1_bare, + load, LD1LN_D>; +defm : LD1LN_patterns<v1f64, v2f64, f64, neon_uimm0_bare, neon_uimm1_bare, + load, LD1LN_D>; + +class NeonI_STN_Lane<bit r, bits<2> op2_1, bit op0, RegisterOperand VList, + Operand ImmOp, string asmop> + : NeonI_LdStOne_Lane<0, r, op2_1, op0, + (outs), (ins GPR64xsp:$Rn, VList:$Rt, ImmOp:$lane), + asmop # "\t$Rt[$lane], [$Rn]", + [], + NoItinerary> { + let mayStore = 1; + let neverHasSideEffects = 1; + let hasExtraDefRegAllocReq = 1; +} + +multiclass STN_Lane_BHSD<bit r, bit op0, string List, string asmop> { + def _B : NeonI_STN_Lane<r, 0b00, op0, + !cast<RegisterOperand>(List # "B_operand"), + neon_uimm4_bare, asmop> { + let Inst{12-10} = lane{2-0}; + let Inst{30} = lane{3}; + } + + def _H : NeonI_STN_Lane<r, 0b01, op0, + !cast<RegisterOperand>(List # "H_operand"), + neon_uimm3_bare, asmop> { + let Inst{12-10} = {lane{1}, lane{0}, 0b0}; + let Inst{30} = lane{2}; + } + + def _S : NeonI_STN_Lane<r, 0b10, op0, + !cast<RegisterOperand>(List # "S_operand"), + neon_uimm2_bare, asmop> { + let Inst{12-10} = {lane{0}, 0b0, 0b0}; + let Inst{30} = lane{1}; + } + + def _D : NeonI_STN_Lane<r, 0b10, op0, + !cast<RegisterOperand>(List # "D_operand"), + neon_uimm1_bare, asmop>{ + let Inst{12-10} = 0b001; + let Inst{30} = lane{0}; + } +} + +// Store single 1-element structure from one lane of 1 register. +defm ST1LN : STN_Lane_BHSD<0b0, 0b0, "VOne", "st1">; + +// Store single N-element structure from one lane of N consecutive registers +// (N = 2,3,4) +defm ST2LN : STN_Lane_BHSD<0b1, 0b0, "VPair", "st2">; +defm ST3LN : STN_Lane_BHSD<0b0, 0b1, "VTriple", "st3">; +defm ST4LN : STN_Lane_BHSD<0b1, 0b1, "VQuad", "st4">; + +multiclass ST1LN_patterns<ValueType VTy, ValueType VTy2, ValueType DTy, + Operand ImmOp, Operand ImmOp2, PatFrag StoreOp, + Instruction INST> { + def : Pat<(StoreOp (DTy (vector_extract (VTy VPR64:$Rt), ImmOp:$lane)), + GPR64xsp:$Rn), + (INST GPR64xsp:$Rn, + (SUBREG_TO_REG (i64 0), VPR64:$Rt, sub_64), + ImmOp:$lane)>; + + def : Pat<(StoreOp (DTy (vector_extract (VTy2 VPR128:$Rt), ImmOp2:$lane)), + GPR64xsp:$Rn), + (INST GPR64xsp:$Rn, VPR128:$Rt, ImmOp2:$lane)>; +} + +// Match all ST1LN instructions +defm : ST1LN_patterns<v8i8, v16i8, i32, neon_uimm3_bare, neon_uimm4_bare, + truncstorei8, ST1LN_B>; + +defm : ST1LN_patterns<v4i16, v8i16, i32, neon_uimm2_bare, neon_uimm3_bare, + truncstorei16, ST1LN_H>; + +defm : ST1LN_patterns<v2i32, v4i32, i32, neon_uimm1_bare, neon_uimm2_bare, + store, ST1LN_S>; +defm : ST1LN_patterns<v2f32, v4f32, f32, neon_uimm1_bare, neon_uimm2_bare, + store, ST1LN_S>; + +defm : ST1LN_patterns<v1i64, v2i64, i64, neon_uimm0_bare, neon_uimm1_bare, + store, ST1LN_D>; +defm : ST1LN_patterns<v1f64, v2f64, f64, neon_uimm0_bare, neon_uimm1_bare, + store, ST1LN_D>; + +// End of vector load/store single N-element structure (class SIMD lsone). + + +// The following are post-index load/store single N-element instructions +// (class SIMD lsone-post) + +multiclass NeonI_LDN_WB_Dup<bit q, bit r, bits<3> opcode, bits<2> size, + RegisterOperand VecList, Operand ImmTy, + string asmop> { + let mayLoad = 1, neverHasSideEffects = 1, Constraints = "$wb = $Rn", + DecoderMethod = "DecodeVLDSTLanePostInstruction" in { + def _fixed : NeonI_LdOne_Dup_Post<q, r, opcode, size, + (outs VecList:$Rt, GPR64xsp:$wb), + (ins GPR64xsp:$Rn, ImmTy:$amt), + asmop # "\t$Rt, [$Rn], $amt", + [], + NoItinerary> { + let Rm = 0b11111; + } + + def _register : NeonI_LdOne_Dup_Post<q, r, opcode, size, + (outs VecList:$Rt, GPR64xsp:$wb), + (ins GPR64xsp:$Rn, GPR64noxzr:$Rm), + asmop # "\t$Rt, [$Rn], $Rm", + [], + NoItinerary>; + } +} + +multiclass LDWB_Dup_BHSD<bit r, bits<3> opcode, string List, string asmop, + Operand uimm_b, Operand uimm_h, + Operand uimm_s, Operand uimm_d> { + defm _8B : NeonI_LDN_WB_Dup<0, r, opcode, 0b00, + !cast<RegisterOperand>(List # "8B_operand"), + uimm_b, asmop>; + + defm _4H : NeonI_LDN_WB_Dup<0, r, opcode, 0b01, + !cast<RegisterOperand>(List # "4H_operand"), + uimm_h, asmop>; + + defm _2S : NeonI_LDN_WB_Dup<0, r, opcode, 0b10, + !cast<RegisterOperand>(List # "2S_operand"), + uimm_s, asmop>; + + defm _1D : NeonI_LDN_WB_Dup<0, r, opcode, 0b11, + !cast<RegisterOperand>(List # "1D_operand"), + uimm_d, asmop>; + + defm _16B : NeonI_LDN_WB_Dup<1, r, opcode, 0b00, + !cast<RegisterOperand>(List # "16B_operand"), + uimm_b, asmop>; + + defm _8H : NeonI_LDN_WB_Dup<1, r, opcode, 0b01, + !cast<RegisterOperand>(List # "8H_operand"), + uimm_h, asmop>; + + defm _4S : NeonI_LDN_WB_Dup<1, r, opcode, 0b10, + !cast<RegisterOperand>(List # "4S_operand"), + uimm_s, asmop>; + + defm _2D : NeonI_LDN_WB_Dup<1, r, opcode, 0b11, + !cast<RegisterOperand>(List # "2D_operand"), + uimm_d, asmop>; +} + +// Post-index load single 1-element structure to all lanes of 1 register +defm LD1R_WB : LDWB_Dup_BHSD<0b0, 0b110, "VOne", "ld1r", uimm_exact1, + uimm_exact2, uimm_exact4, uimm_exact8>; + +// Post-index load single N-element structure to all lanes of N consecutive +// registers (N = 2,3,4) +defm LD2R_WB : LDWB_Dup_BHSD<0b1, 0b110, "VPair", "ld2r", uimm_exact2, + uimm_exact4, uimm_exact8, uimm_exact16>; +defm LD3R_WB : LDWB_Dup_BHSD<0b0, 0b111, "VTriple", "ld3r", uimm_exact3, + uimm_exact6, uimm_exact12, uimm_exact24>; +defm LD4R_WB : LDWB_Dup_BHSD<0b1, 0b111, "VQuad", "ld4r", uimm_exact4, + uimm_exact8, uimm_exact16, uimm_exact32>; + +let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1, + Constraints = "$Rn = $wb, $Rt = $src", + DecoderMethod = "DecodeVLDSTLanePostInstruction" in { + class LDN_WBFx_Lane<bit r, bits<2> op2_1, bit op0, RegisterOperand VList, + Operand ImmTy, Operand ImmOp, string asmop> + : NeonI_LdStOne_Lane_Post<1, r, op2_1, op0, + (outs VList:$Rt, GPR64xsp:$wb), + (ins GPR64xsp:$Rn, ImmTy:$amt, + VList:$src, ImmOp:$lane), + asmop # "\t$Rt[$lane], [$Rn], $amt", + [], + NoItinerary> { + let Rm = 0b11111; + } + + class LDN_WBReg_Lane<bit r, bits<2> op2_1, bit op0, RegisterOperand VList, + Operand ImmTy, Operand ImmOp, string asmop> + : NeonI_LdStOne_Lane_Post<1, r, op2_1, op0, + (outs VList:$Rt, GPR64xsp:$wb), + (ins GPR64xsp:$Rn, GPR64noxzr:$Rm, + VList:$src, ImmOp:$lane), + asmop # "\t$Rt[$lane], [$Rn], $Rm", + [], + NoItinerary>; +} + +multiclass LD_Lane_WB_BHSD<bit r, bit op0, string List, string asmop, + Operand uimm_b, Operand uimm_h, + Operand uimm_s, Operand uimm_d> { + def _B_fixed : LDN_WBFx_Lane<r, 0b00, op0, + !cast<RegisterOperand>(List # "B_operand"), + uimm_b, neon_uimm4_bare, asmop> { + let Inst{12-10} = lane{2-0}; + let Inst{30} = lane{3}; + } + + def _B_register : LDN_WBReg_Lane<r, 0b00, op0, + !cast<RegisterOperand>(List # "B_operand"), + uimm_b, neon_uimm4_bare, asmop> { + let Inst{12-10} = lane{2-0}; + let Inst{30} = lane{3}; + } + + def _H_fixed : LDN_WBFx_Lane<r, 0b01, op0, + !cast<RegisterOperand>(List # "H_operand"), + uimm_h, neon_uimm3_bare, asmop> { + let Inst{12-10} = {lane{1}, lane{0}, 0b0}; + let Inst{30} = lane{2}; + } + + def _H_register : LDN_WBReg_Lane<r, 0b01, op0, + !cast<RegisterOperand>(List # "H_operand"), + uimm_h, neon_uimm3_bare, asmop> { + let Inst{12-10} = {lane{1}, lane{0}, 0b0}; + let Inst{30} = lane{2}; + } + + def _S_fixed : LDN_WBFx_Lane<r, 0b10, op0, + !cast<RegisterOperand>(List # "S_operand"), + uimm_s, neon_uimm2_bare, asmop> { + let Inst{12-10} = {lane{0}, 0b0, 0b0}; + let Inst{30} = lane{1}; + } + + def _S_register : LDN_WBReg_Lane<r, 0b10, op0, + !cast<RegisterOperand>(List # "S_operand"), + uimm_s, neon_uimm2_bare, asmop> { + let Inst{12-10} = {lane{0}, 0b0, 0b0}; + let Inst{30} = lane{1}; + } + + def _D_fixed : LDN_WBFx_Lane<r, 0b10, op0, + !cast<RegisterOperand>(List # "D_operand"), + uimm_d, neon_uimm1_bare, asmop> { + let Inst{12-10} = 0b001; + let Inst{30} = lane{0}; + } + + def _D_register : LDN_WBReg_Lane<r, 0b10, op0, + !cast<RegisterOperand>(List # "D_operand"), + uimm_d, neon_uimm1_bare, asmop> { + let Inst{12-10} = 0b001; + let Inst{30} = lane{0}; + } +} + +// Post-index load single 1-element structure to one lane of 1 register. +defm LD1LN_WB : LD_Lane_WB_BHSD<0b0, 0b0, "VOne", "ld1", uimm_exact1, + uimm_exact2, uimm_exact4, uimm_exact8>; + +// Post-index load single N-element structure to one lane of N consecutive +// registers +// (N = 2,3,4) +defm LD2LN_WB : LD_Lane_WB_BHSD<0b1, 0b0, "VPair", "ld2", uimm_exact2, + uimm_exact4, uimm_exact8, uimm_exact16>; +defm LD3LN_WB : LD_Lane_WB_BHSD<0b0, 0b1, "VTriple", "ld3", uimm_exact3, + uimm_exact6, uimm_exact12, uimm_exact24>; +defm LD4LN_WB : LD_Lane_WB_BHSD<0b1, 0b1, "VQuad", "ld4", uimm_exact4, + uimm_exact8, uimm_exact16, uimm_exact32>; + +let mayStore = 1, neverHasSideEffects = 1, + hasExtraDefRegAllocReq = 1, Constraints = "$Rn = $wb", + DecoderMethod = "DecodeVLDSTLanePostInstruction" in { + class STN_WBFx_Lane<bit r, bits<2> op2_1, bit op0, RegisterOperand VList, + Operand ImmTy, Operand ImmOp, string asmop> + : NeonI_LdStOne_Lane_Post<0, r, op2_1, op0, + (outs GPR64xsp:$wb), + (ins GPR64xsp:$Rn, ImmTy:$amt, + VList:$Rt, ImmOp:$lane), + asmop # "\t$Rt[$lane], [$Rn], $amt", + [], + NoItinerary> { + let Rm = 0b11111; + } + + class STN_WBReg_Lane<bit r, bits<2> op2_1, bit op0, RegisterOperand VList, + Operand ImmTy, Operand ImmOp, string asmop> + : NeonI_LdStOne_Lane_Post<0, r, op2_1, op0, + (outs GPR64xsp:$wb), + (ins GPR64xsp:$Rn, GPR64noxzr:$Rm, VList:$Rt, + ImmOp:$lane), + asmop # "\t$Rt[$lane], [$Rn], $Rm", + [], + NoItinerary>; +} + +multiclass ST_Lane_WB_BHSD<bit r, bit op0, string List, string asmop, + Operand uimm_b, Operand uimm_h, + Operand uimm_s, Operand uimm_d> { + def _B_fixed : STN_WBFx_Lane<r, 0b00, op0, + !cast<RegisterOperand>(List # "B_operand"), + uimm_b, neon_uimm4_bare, asmop> { + let Inst{12-10} = lane{2-0}; + let Inst{30} = lane{3}; + } + + def _B_register : STN_WBReg_Lane<r, 0b00, op0, + !cast<RegisterOperand>(List # "B_operand"), + uimm_b, neon_uimm4_bare, asmop> { + let Inst{12-10} = lane{2-0}; + let Inst{30} = lane{3}; + } + + def _H_fixed : STN_WBFx_Lane<r, 0b01, op0, + !cast<RegisterOperand>(List # "H_operand"), + uimm_h, neon_uimm3_bare, asmop> { + let Inst{12-10} = {lane{1}, lane{0}, 0b0}; + let Inst{30} = lane{2}; + } + + def _H_register : STN_WBReg_Lane<r, 0b01, op0, + !cast<RegisterOperand>(List # "H_operand"), + uimm_h, neon_uimm3_bare, asmop> { + let Inst{12-10} = {lane{1}, lane{0}, 0b0}; + let Inst{30} = lane{2}; + } + + def _S_fixed : STN_WBFx_Lane<r, 0b10, op0, + !cast<RegisterOperand>(List # "S_operand"), + uimm_s, neon_uimm2_bare, asmop> { + let Inst{12-10} = {lane{0}, 0b0, 0b0}; + let Inst{30} = lane{1}; + } + + def _S_register : STN_WBReg_Lane<r, 0b10, op0, + !cast<RegisterOperand>(List # "S_operand"), + uimm_s, neon_uimm2_bare, asmop> { + let Inst{12-10} = {lane{0}, 0b0, 0b0}; + let Inst{30} = lane{1}; + } + + def _D_fixed : STN_WBFx_Lane<r, 0b10, op0, + !cast<RegisterOperand>(List # "D_operand"), + uimm_d, neon_uimm1_bare, asmop> { + let Inst{12-10} = 0b001; + let Inst{30} = lane{0}; + } + + def _D_register : STN_WBReg_Lane<r, 0b10, op0, + !cast<RegisterOperand>(List # "D_operand"), + uimm_d, neon_uimm1_bare, asmop> { + let Inst{12-10} = 0b001; + let Inst{30} = lane{0}; + } +} + +// Post-index store single 1-element structure from one lane of 1 register. +defm ST1LN_WB : ST_Lane_WB_BHSD<0b0, 0b0, "VOne", "st1", uimm_exact1, + uimm_exact2, uimm_exact4, uimm_exact8>; + +// Post-index store single N-element structure from one lane of N consecutive +// registers (N = 2,3,4) +defm ST2LN_WB : ST_Lane_WB_BHSD<0b1, 0b0, "VPair", "st2", uimm_exact2, + uimm_exact4, uimm_exact8, uimm_exact16>; +defm ST3LN_WB : ST_Lane_WB_BHSD<0b0, 0b1, "VTriple", "st3", uimm_exact3, + uimm_exact6, uimm_exact12, uimm_exact24>; +defm ST4LN_WB : ST_Lane_WB_BHSD<0b1, 0b1, "VQuad", "st4", uimm_exact4, + uimm_exact8, uimm_exact16, uimm_exact32>; + +// End of post-index load/store single N-element instructions +// (class SIMD lsone-post) + +// Neon Scalar instructions implementation +// Scalar Three Same + +class NeonI_Scalar3Same_size<bit u, bits<2> size, bits<5> opcode, string asmop, + RegisterClass FPRC> + : NeonI_Scalar3Same<u, size, opcode, + (outs FPRC:$Rd), (ins FPRC:$Rn, FPRC:$Rm), + !strconcat(asmop, "\t$Rd, $Rn, $Rm"), + [], + NoItinerary>; + +class NeonI_Scalar3Same_D_size<bit u, bits<5> opcode, string asmop> + : NeonI_Scalar3Same_size<u, 0b11, opcode, asmop, FPR64>; + +multiclass NeonI_Scalar3Same_HS_sizes<bit u, bits<5> opcode, string asmop, + bit Commutable = 0> { + let isCommutable = Commutable in { + def hhh : NeonI_Scalar3Same_size<u, 0b01, opcode, asmop, FPR16>; + def sss : NeonI_Scalar3Same_size<u, 0b10, opcode, asmop, FPR32>; + } +} + +multiclass NeonI_Scalar3Same_SD_sizes<bit u, bit size_high, bits<5> opcode, + string asmop, bit Commutable = 0> { + let isCommutable = Commutable in { + def sss : NeonI_Scalar3Same_size<u, {size_high, 0b0}, opcode, asmop, FPR32>; + def ddd : NeonI_Scalar3Same_size<u, {size_high, 0b1}, opcode, asmop, FPR64>; + } +} + +multiclass NeonI_Scalar3Same_BHSD_sizes<bit u, bits<5> opcode, + string asmop, bit Commutable = 0> { + let isCommutable = Commutable in { + def bbb : NeonI_Scalar3Same_size<u, 0b00, opcode, asmop, FPR8>; + def hhh : NeonI_Scalar3Same_size<u, 0b01, opcode, asmop, FPR16>; + def sss : NeonI_Scalar3Same_size<u, 0b10, opcode, asmop, FPR32>; + def ddd : NeonI_Scalar3Same_size<u, 0b11, opcode, asmop, FPR64>; + } +} + +multiclass Neon_Scalar3Same_D_size_patterns<SDPatternOperator opnode, + Instruction INSTD> { + def : Pat<(v1i64 (opnode (v1i64 FPR64:$Rn), (v1i64 FPR64:$Rm))), + (INSTD FPR64:$Rn, FPR64:$Rm)>; +} + +multiclass Neon_Scalar3Same_BHSD_size_patterns<SDPatternOperator opnode, + Instruction INSTB, + Instruction INSTH, + Instruction INSTS, + Instruction INSTD> + : Neon_Scalar3Same_D_size_patterns<opnode, INSTD> { + def: Pat<(v1i8 (opnode (v1i8 FPR8:$Rn), (v1i8 FPR8:$Rm))), + (INSTB FPR8:$Rn, FPR8:$Rm)>; + + def: Pat<(v1i16 (opnode (v1i16 FPR16:$Rn), (v1i16 FPR16:$Rm))), + (INSTH FPR16:$Rn, FPR16:$Rm)>; + + def: Pat<(v1i32 (opnode (v1i32 FPR32:$Rn), (v1i32 FPR32:$Rm))), + (INSTS FPR32:$Rn, FPR32:$Rm)>; +} + +class Neon_Scalar3Same_cmp_D_size_patterns<SDPatternOperator opnode, + Instruction INSTD> + : Pat<(v1i64 (opnode (v1i64 FPR64:$Rn), (v1i64 FPR64:$Rm))), + (INSTD FPR64:$Rn, FPR64:$Rm)>; + +multiclass Neon_Scalar3Same_HS_size_patterns<SDPatternOperator opnode, + Instruction INSTH, + Instruction INSTS> { + def : Pat<(v1i16 (opnode (v1i16 FPR16:$Rn), (v1i16 FPR16:$Rm))), + (INSTH FPR16:$Rn, FPR16:$Rm)>; + def : Pat<(v1i32 (opnode (v1i32 FPR32:$Rn), (v1i32 FPR32:$Rm))), + (INSTS FPR32:$Rn, FPR32:$Rm)>; +} + +multiclass Neon_Scalar3Same_SD_size_patterns<SDPatternOperator opnode, + Instruction INSTS, + Instruction INSTD> { + def : Pat<(v1f32 (opnode (v1f32 FPR32:$Rn), (v1f32 FPR32:$Rm))), + (INSTS FPR32:$Rn, FPR32:$Rm)>; + def : Pat<(v1f64 (opnode (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))), + (INSTD FPR64:$Rn, FPR64:$Rm)>; +} + +multiclass Neon_Scalar3Same_cmp_SD_size_patterns<SDPatternOperator opnode, + Instruction INSTS, + Instruction INSTD> { + def : Pat<(v1i32 (opnode (v1f32 FPR32:$Rn), (v1f32 FPR32:$Rm))), + (INSTS FPR32:$Rn, FPR32:$Rm)>; + def : Pat<(v1i64 (opnode (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))), + (INSTD FPR64:$Rn, FPR64:$Rm)>; +} + +class Neon_Scalar3Same_cmp_V1_D_size_patterns<CondCode CC, + Instruction INSTD> + : Pat<(v1i64 (Neon_cmp (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm), CC)), + (INSTD FPR64:$Rn, FPR64:$Rm)>; + +// Scalar Three Different + +class NeonI_Scalar3Diff_size<bit u, bits<2> size, bits<4> opcode, string asmop, + RegisterClass FPRCD, RegisterClass FPRCS> + : NeonI_Scalar3Diff<u, size, opcode, + (outs FPRCD:$Rd), (ins FPRCS:$Rn, FPRCS:$Rm), + !strconcat(asmop, "\t$Rd, $Rn, $Rm"), + [], + NoItinerary>; + +multiclass NeonI_Scalar3Diff_HS_size<bit u, bits<4> opcode, string asmop> { + def shh : NeonI_Scalar3Diff_size<u, 0b01, opcode, asmop, FPR32, FPR16>; + def dss : NeonI_Scalar3Diff_size<u, 0b10, opcode, asmop, FPR64, FPR32>; +} + +multiclass NeonI_Scalar3Diff_ml_HS_size<bit u, bits<4> opcode, string asmop> { + let Constraints = "$Src = $Rd" in { + def shh : NeonI_Scalar3Diff<u, 0b01, opcode, + (outs FPR32:$Rd), (ins FPR32:$Src, FPR16:$Rn, FPR16:$Rm), + !strconcat(asmop, "\t$Rd, $Rn, $Rm"), + [], + NoItinerary>; + def dss : NeonI_Scalar3Diff<u, 0b10, opcode, + (outs FPR64:$Rd), (ins FPR64:$Src, FPR32:$Rn, FPR32:$Rm), + !strconcat(asmop, "\t$Rd, $Rn, $Rm"), + [], + NoItinerary>; + } +} + +multiclass Neon_Scalar3Diff_HS_size_patterns<SDPatternOperator opnode, + Instruction INSTH, + Instruction INSTS> { + def : Pat<(v1i32 (opnode (v1i16 FPR16:$Rn), (v1i16 FPR16:$Rm))), + (INSTH FPR16:$Rn, FPR16:$Rm)>; + def : Pat<(v1i64 (opnode (v1i32 FPR32:$Rn), (v1i32 FPR32:$Rm))), + (INSTS FPR32:$Rn, FPR32:$Rm)>; +} + +multiclass Neon_Scalar3Diff_ml_HS_size_patterns<SDPatternOperator opnode, + Instruction INSTH, + Instruction INSTS> { + def : Pat<(v1i32 (opnode (v1i32 FPR32:$Src), (v1i16 FPR16:$Rn), (v1i16 FPR16:$Rm))), + (INSTH FPR32:$Src, FPR16:$Rn, FPR16:$Rm)>; + def : Pat<(v1i64 (opnode (v1i64 FPR64:$Src), (v1i32 FPR32:$Rn), (v1i32 FPR32:$Rm))), + (INSTS FPR64:$Src, FPR32:$Rn, FPR32:$Rm)>; +} + +// Scalar Two Registers Miscellaneous + +class NeonI_Scalar2SameMisc_size<bit u, bits<2> size, bits<5> opcode, string asmop, + RegisterClass FPRCD, RegisterClass FPRCS> + : NeonI_Scalar2SameMisc<u, size, opcode, + (outs FPRCD:$Rd), (ins FPRCS:$Rn), + !strconcat(asmop, "\t$Rd, $Rn"), + [], + NoItinerary>; + +multiclass NeonI_Scalar2SameMisc_SD_size<bit u, bit size_high, bits<5> opcode, + string asmop> { + def ss : NeonI_Scalar2SameMisc_size<u, {size_high, 0b0}, opcode, asmop, FPR32, + FPR32>; + def dd : NeonI_Scalar2SameMisc_size<u, {size_high, 0b1}, opcode, asmop, FPR64, + FPR64>; +} + +multiclass NeonI_Scalar2SameMisc_D_size<bit u, bits<5> opcode, string asmop> { + def dd : NeonI_Scalar2SameMisc_size<u, 0b11, opcode, asmop, FPR64, FPR64>; +} + +multiclass NeonI_Scalar2SameMisc_BHSD_size<bit u, bits<5> opcode, string asmop> + : NeonI_Scalar2SameMisc_D_size<u, opcode, asmop> { + def bb : NeonI_Scalar2SameMisc_size<u, 0b00, opcode, asmop, FPR8, FPR8>; + def hh : NeonI_Scalar2SameMisc_size<u, 0b01, opcode, asmop, FPR16, FPR16>; + def ss : NeonI_Scalar2SameMisc_size<u, 0b10, opcode, asmop, FPR32, FPR32>; +} + +class NeonI_Scalar2SameMisc_fcvtxn_D_size<bit u, bits<5> opcode, string asmop> + : NeonI_Scalar2SameMisc_size<u, 0b01, opcode, asmop, FPR32, FPR64>; + +multiclass NeonI_Scalar2SameMisc_narrow_HSD_size<bit u, bits<5> opcode, + string asmop> { + def bh : NeonI_Scalar2SameMisc_size<u, 0b00, opcode, asmop, FPR8, FPR16>; + def hs : NeonI_Scalar2SameMisc_size<u, 0b01, opcode, asmop, FPR16, FPR32>; + def sd : NeonI_Scalar2SameMisc_size<u, 0b10, opcode, asmop, FPR32, FPR64>; +} + +class NeonI_Scalar2SameMisc_accum_size<bit u, bits<2> size, bits<5> opcode, + string asmop, RegisterClass FPRC> + : NeonI_Scalar2SameMisc<u, size, opcode, + (outs FPRC:$Rd), (ins FPRC:$Src, FPRC:$Rn), + !strconcat(asmop, "\t$Rd, $Rn"), + [], + NoItinerary>; + +multiclass NeonI_Scalar2SameMisc_accum_BHSD_size<bit u, bits<5> opcode, + string asmop> { + + let Constraints = "$Src = $Rd" in { + def bb : NeonI_Scalar2SameMisc_accum_size<u, 0b00, opcode, asmop, FPR8>; + def hh : NeonI_Scalar2SameMisc_accum_size<u, 0b01, opcode, asmop, FPR16>; + def ss : NeonI_Scalar2SameMisc_accum_size<u, 0b10, opcode, asmop, FPR32>; + def dd : NeonI_Scalar2SameMisc_accum_size<u, 0b11, opcode, asmop, FPR64>; + } +} + +class Neon_Scalar2SameMisc_fcvtxn_D_size_patterns<SDPatternOperator opnode, + Instruction INSTD> + : Pat<(v1f32 (opnode (v1f64 FPR64:$Rn))), + (INSTD FPR64:$Rn)>; + +multiclass Neon_Scalar2SameMisc_fcvt_SD_size_patterns<SDPatternOperator opnode, + Instruction INSTS, + Instruction INSTD> { + def : Pat<(v1i32 (opnode (v1f32 FPR32:$Rn))), + (INSTS FPR32:$Rn)>; + def : Pat<(v1i64 (opnode (v1f64 FPR64:$Rn))), + (INSTD FPR64:$Rn)>; +} + +multiclass Neon_Scalar2SameMisc_cvt_SD_size_patterns<SDPatternOperator Sopnode, + SDPatternOperator Dopnode, + Instruction INSTS, + Instruction INSTD> { + def : Pat<(f32 (Sopnode (v1i32 FPR32:$Rn))), + (INSTS FPR32:$Rn)>; + def : Pat<(f64 (Dopnode (v1i64 FPR64:$Rn))), + (INSTD FPR64:$Rn)>; +} + +multiclass Neon_Scalar2SameMisc_SD_size_patterns<SDPatternOperator opnode, + Instruction INSTS, + Instruction INSTD> { + def : Pat<(v1f32 (opnode (v1f32 FPR32:$Rn))), + (INSTS FPR32:$Rn)>; + def : Pat<(v1f64 (opnode (v1f64 FPR64:$Rn))), + (INSTD FPR64:$Rn)>; +} + +class NeonI_Scalar2SameMisc_cmpz_D_size<bit u, bits<5> opcode, string asmop> + : NeonI_Scalar2SameMisc<u, 0b11, opcode, + (outs FPR64:$Rd), (ins FPR64:$Rn, neon_uimm0:$Imm), + !strconcat(asmop, "\t$Rd, $Rn, $Imm"), + [], + NoItinerary>; + +multiclass NeonI_Scalar2SameMisc_cmpz_SD_size<bit u, bits<5> opcode, + string asmop> { + def ssi : NeonI_Scalar2SameMisc<u, 0b10, opcode, + (outs FPR32:$Rd), (ins FPR32:$Rn, fpz32:$FPImm), + !strconcat(asmop, "\t$Rd, $Rn, $FPImm"), + [], + NoItinerary>; + def ddi : NeonI_Scalar2SameMisc<u, 0b11, opcode, + (outs FPR64:$Rd), (ins FPR64:$Rn, fpz32:$FPImm), + !strconcat(asmop, "\t$Rd, $Rn, $FPImm"), + [], + NoItinerary>; +} + +class Neon_Scalar2SameMisc_cmpz_D_size_patterns<SDPatternOperator opnode, + Instruction INSTD> + : Pat<(v1i64 (opnode (v1i64 FPR64:$Rn), + (v1i64 (bitconvert (v8i8 Neon_AllZero))))), + (INSTD FPR64:$Rn, 0)>; + +class Neon_Scalar2SameMisc_cmpz_D_V1_size_patterns<CondCode CC, + Instruction INSTD> + : Pat<(v1i64 (Neon_cmpz (v1i64 FPR64:$Rn), + (i32 neon_uimm0:$Imm), CC)), + (INSTD FPR64:$Rn, neon_uimm0:$Imm)>; + +multiclass Neon_Scalar2SameMisc_cmpz_SD_size_patterns<SDPatternOperator opnode, + Instruction INSTS, + Instruction INSTD> { + def : Pat<(v1i32 (opnode (v1f32 FPR32:$Rn), + (v1f32 (scalar_to_vector (f32 fpz32:$FPImm))))), + (INSTS FPR32:$Rn, fpz32:$FPImm)>; + def : Pat<(v1i64 (opnode (v1f64 FPR64:$Rn), + (v1f32 (scalar_to_vector (f32 fpz32:$FPImm))))), + (INSTD FPR64:$Rn, fpz32:$FPImm)>; +} + +multiclass Neon_Scalar2SameMisc_D_size_patterns<SDPatternOperator opnode, + Instruction INSTD> { + def : Pat<(v1i64 (opnode (v1i64 FPR64:$Rn))), + (INSTD FPR64:$Rn)>; +} + +multiclass Neon_Scalar2SameMisc_BHSD_size_patterns<SDPatternOperator opnode, + Instruction INSTB, + Instruction INSTH, + Instruction INSTS, + Instruction INSTD> + : Neon_Scalar2SameMisc_D_size_patterns<opnode, INSTD> { + def : Pat<(v1i8 (opnode (v1i8 FPR8:$Rn))), + (INSTB FPR8:$Rn)>; + def : Pat<(v1i16 (opnode (v1i16 FPR16:$Rn))), + (INSTH FPR16:$Rn)>; + def : Pat<(v1i32 (opnode (v1i32 FPR32:$Rn))), + (INSTS FPR32:$Rn)>; +} + +multiclass Neon_Scalar2SameMisc_narrow_HSD_size_patterns< + SDPatternOperator opnode, + Instruction INSTH, + Instruction INSTS, + Instruction INSTD> { + def : Pat<(v1i8 (opnode (v1i16 FPR16:$Rn))), + (INSTH FPR16:$Rn)>; + def : Pat<(v1i16 (opnode (v1i32 FPR32:$Rn))), + (INSTS FPR32:$Rn)>; + def : Pat<(v1i32 (opnode (v1i64 FPR64:$Rn))), + (INSTD FPR64:$Rn)>; + +} + +multiclass Neon_Scalar2SameMisc_accum_BHSD_size_patterns< + SDPatternOperator opnode, + Instruction INSTB, + Instruction INSTH, + Instruction INSTS, + Instruction INSTD> { + def : Pat<(v1i8 (opnode (v1i8 FPR8:$Src), (v1i8 FPR8:$Rn))), + (INSTB FPR8:$Src, FPR8:$Rn)>; + def : Pat<(v1i16 (opnode (v1i16 FPR16:$Src), (v1i16 FPR16:$Rn))), + (INSTH FPR16:$Src, FPR16:$Rn)>; + def : Pat<(v1i32 (opnode (v1i32 FPR32:$Src), (v1i32 FPR32:$Rn))), + (INSTS FPR32:$Src, FPR32:$Rn)>; + def : Pat<(v1i64 (opnode (v1i64 FPR64:$Src), (v1i64 FPR64:$Rn))), + (INSTD FPR64:$Src, FPR64:$Rn)>; +} + +// Scalar Shift By Immediate + +class NeonI_ScalarShiftImm_size<bit u, bits<5> opcode, string asmop, + RegisterClass FPRC, Operand ImmTy> + : NeonI_ScalarShiftImm<u, opcode, + (outs FPRC:$Rd), (ins FPRC:$Rn, ImmTy:$Imm), + !strconcat(asmop, "\t$Rd, $Rn, $Imm"), + [], NoItinerary>; + +multiclass NeonI_ScalarShiftRightImm_D_size<bit u, bits<5> opcode, + string asmop> { + def ddi : NeonI_ScalarShiftImm_size<u, opcode, asmop, FPR64, shr_imm64> { + bits<6> Imm; + let Inst{22} = 0b1; // immh:immb = 1xxxxxx + let Inst{21-16} = Imm; + } +} + +multiclass NeonI_ScalarShiftRightImm_BHSD_size<bit u, bits<5> opcode, + string asmop> + : NeonI_ScalarShiftRightImm_D_size<u, opcode, asmop> { + def bbi : NeonI_ScalarShiftImm_size<u, opcode, asmop, FPR8, shr_imm8> { + bits<3> Imm; + let Inst{22-19} = 0b0001; // immh:immb = 0001xxx + let Inst{18-16} = Imm; + } + def hhi : NeonI_ScalarShiftImm_size<u, opcode, asmop, FPR16, shr_imm16> { + bits<4> Imm; + let Inst{22-20} = 0b001; // immh:immb = 001xxxx + let Inst{19-16} = Imm; + } + def ssi : NeonI_ScalarShiftImm_size<u, opcode, asmop, FPR32, shr_imm32> { + bits<5> Imm; + let Inst{22-21} = 0b01; // immh:immb = 01xxxxx + let Inst{20-16} = Imm; + } +} + +multiclass NeonI_ScalarShiftLeftImm_D_size<bit u, bits<5> opcode, + string asmop> { + def ddi : NeonI_ScalarShiftImm_size<u, opcode, asmop, FPR64, shl_imm64> { + bits<6> Imm; + let Inst{22} = 0b1; // immh:immb = 1xxxxxx + let Inst{21-16} = Imm; + } +} + +multiclass NeonI_ScalarShiftLeftImm_BHSD_size<bit u, bits<5> opcode, + string asmop> + : NeonI_ScalarShiftLeftImm_D_size<u, opcode, asmop> { + def bbi : NeonI_ScalarShiftImm_size<u, opcode, asmop, FPR8, shl_imm8> { + bits<3> Imm; + let Inst{22-19} = 0b0001; // immh:immb = 0001xxx + let Inst{18-16} = Imm; + } + def hhi : NeonI_ScalarShiftImm_size<u, opcode, asmop, FPR16, shl_imm16> { + bits<4> Imm; + let Inst{22-20} = 0b001; // immh:immb = 001xxxx + let Inst{19-16} = Imm; + } + def ssi : NeonI_ScalarShiftImm_size<u, opcode, asmop, FPR32, shl_imm32> { + bits<5> Imm; + let Inst{22-21} = 0b01; // immh:immb = 01xxxxx + let Inst{20-16} = Imm; + } +} + +class NeonI_ScalarShiftRightImm_accum_D_size<bit u, bits<5> opcode, string asmop> + : NeonI_ScalarShiftImm<u, opcode, + (outs FPR64:$Rd), + (ins FPR64:$Src, FPR64:$Rn, shr_imm64:$Imm), + !strconcat(asmop, "\t$Rd, $Rn, $Imm"), + [], NoItinerary> { + bits<6> Imm; + let Inst{22} = 0b1; // immh:immb = 1xxxxxx + let Inst{21-16} = Imm; + let Constraints = "$Src = $Rd"; +} + +class NeonI_ScalarShiftLeftImm_accum_D_size<bit u, bits<5> opcode, string asmop> + : NeonI_ScalarShiftImm<u, opcode, + (outs FPR64:$Rd), + (ins FPR64:$Src, FPR64:$Rn, shl_imm64:$Imm), + !strconcat(asmop, "\t$Rd, $Rn, $Imm"), + [], NoItinerary> { + bits<6> Imm; + let Inst{22} = 0b1; // immh:immb = 1xxxxxx + let Inst{21-16} = Imm; + let Constraints = "$Src = $Rd"; +} + +class NeonI_ScalarShiftImm_narrow_size<bit u, bits<5> opcode, string asmop, + RegisterClass FPRCD, RegisterClass FPRCS, + Operand ImmTy> + : NeonI_ScalarShiftImm<u, opcode, + (outs FPRCD:$Rd), (ins FPRCS:$Rn, ImmTy:$Imm), + !strconcat(asmop, "\t$Rd, $Rn, $Imm"), + [], NoItinerary>; + +multiclass NeonI_ScalarShiftImm_narrow_HSD_size<bit u, bits<5> opcode, + string asmop> { + def bhi : NeonI_ScalarShiftImm_narrow_size<u, opcode, asmop, FPR8, FPR16, + shr_imm8> { + bits<3> Imm; + let Inst{22-19} = 0b0001; // immh:immb = 0001xxx + let Inst{18-16} = Imm; + } + def hsi : NeonI_ScalarShiftImm_narrow_size<u, opcode, asmop, FPR16, FPR32, + shr_imm16> { + bits<4> Imm; + let Inst{22-20} = 0b001; // immh:immb = 001xxxx + let Inst{19-16} = Imm; + } + def sdi : NeonI_ScalarShiftImm_narrow_size<u, opcode, asmop, FPR32, FPR64, + shr_imm32> { + bits<5> Imm; + let Inst{22-21} = 0b01; // immh:immb = 01xxxxx + let Inst{20-16} = Imm; + } +} + +multiclass NeonI_ScalarShiftImm_cvt_SD_size<bit u, bits<5> opcode, string asmop> { + def ssi : NeonI_ScalarShiftImm_size<u, opcode, asmop, FPR32, shr_imm32> { + bits<5> Imm; + let Inst{22-21} = 0b01; // immh:immb = 01xxxxx + let Inst{20-16} = Imm; + } + def ddi : NeonI_ScalarShiftImm_size<u, opcode, asmop, FPR64, shr_imm64> { + bits<6> Imm; + let Inst{22} = 0b1; // immh:immb = 1xxxxxx + let Inst{21-16} = Imm; + } +} + +multiclass Neon_ScalarShiftRImm_D_size_patterns<SDPatternOperator opnode, + Instruction INSTD> { + def ddi : Pat<(v1i64 (opnode (v1i64 FPR64:$Rn), (i32 shr_imm64:$Imm))), + (INSTD FPR64:$Rn, imm:$Imm)>; +} + +multiclass Neon_ScalarShiftLImm_D_size_patterns<SDPatternOperator opnode, + Instruction INSTD> { + def ddi : Pat<(v1i64 (opnode (v1i64 FPR64:$Rn), (i32 shl_imm64:$Imm))), + (INSTD FPR64:$Rn, imm:$Imm)>; +} + +class Neon_ScalarShiftImm_arm_D_size_patterns<SDPatternOperator opnode, + Instruction INSTD> + : Pat<(v1i64 (opnode (v1i64 FPR64:$Rn), + (v1i64 (Neon_vdup (i32 shr_imm64:$Imm))))), + (INSTD FPR64:$Rn, imm:$Imm)>; + +multiclass Neon_ScalarShiftLImm_BHSD_size_patterns<SDPatternOperator opnode, + Instruction INSTB, + Instruction INSTH, + Instruction INSTS, + Instruction INSTD> + : Neon_ScalarShiftLImm_D_size_patterns<opnode, INSTD> { + def bbi : Pat<(v1i8 (opnode (v1i8 FPR8:$Rn), (i32 shl_imm8:$Imm))), + (INSTB FPR8:$Rn, imm:$Imm)>; + def hhi : Pat<(v1i16 (opnode (v1i16 FPR16:$Rn), (i32 shl_imm16:$Imm))), + (INSTH FPR16:$Rn, imm:$Imm)>; + def ssi : Pat<(v1i32 (opnode (v1i32 FPR32:$Rn), (i32 shl_imm32:$Imm))), + (INSTS FPR32:$Rn, imm:$Imm)>; +} + +class Neon_ScalarShiftLImm_accum_D_size_patterns<SDPatternOperator opnode, + Instruction INSTD> + : Pat<(v1i64 (opnode (v1i64 FPR64:$Src), (v1i64 FPR64:$Rn), + (i32 shl_imm64:$Imm))), + (INSTD FPR64:$Src, FPR64:$Rn, imm:$Imm)>; + +class Neon_ScalarShiftRImm_accum_D_size_patterns<SDPatternOperator opnode, + Instruction INSTD> + : Pat<(v1i64 (opnode (v1i64 FPR64:$Src), (v1i64 FPR64:$Rn), + (i32 shr_imm64:$Imm))), + (INSTD FPR64:$Src, FPR64:$Rn, imm:$Imm)>; + +multiclass Neon_ScalarShiftImm_narrow_HSD_size_patterns< + SDPatternOperator opnode, + Instruction INSTH, + Instruction INSTS, + Instruction INSTD> { + def bhi : Pat<(v1i8 (opnode (v1i16 FPR16:$Rn), (i32 shr_imm16:$Imm))), + (INSTH FPR16:$Rn, imm:$Imm)>; + def hsi : Pat<(v1i16 (opnode (v1i32 FPR32:$Rn), (i32 shr_imm32:$Imm))), + (INSTS FPR32:$Rn, imm:$Imm)>; + def sdi : Pat<(v1i32 (opnode (v1i64 FPR64:$Rn), (i32 shr_imm64:$Imm))), + (INSTD FPR64:$Rn, imm:$Imm)>; +} + +multiclass Neon_ScalarShiftImm_scvtf_SD_size_patterns<SDPatternOperator Sopnode, + SDPatternOperator Dopnode, + Instruction INSTS, + Instruction INSTD> { + def ssi : Pat<(f32 (Sopnode (v1i32 FPR32:$Rn), (i32 shr_imm32:$Imm))), + (INSTS FPR32:$Rn, imm:$Imm)>; + def ddi : Pat<(f64 (Dopnode (v1i64 FPR64:$Rn), (i32 shr_imm64:$Imm))), + (INSTD FPR64:$Rn, imm:$Imm)>; +} + +multiclass Neon_ScalarShiftImm_fcvts_SD_size_patterns<SDPatternOperator Sopnode, + SDPatternOperator Dopnode, + Instruction INSTS, + Instruction INSTD> { + def ssi : Pat<(v1i32 (Sopnode (v1f32 FPR32:$Rn), (i32 shr_imm32:$Imm))), + (INSTS FPR32:$Rn, imm:$Imm)>; + def ddi : Pat<(v1i64 (Dopnode (v1f64 FPR64:$Rn), (i32 shr_imm64:$Imm))), + (INSTD FPR64:$Rn, imm:$Imm)>; +} + +// Scalar Signed Shift Right (Immediate) +defm SSHR : NeonI_ScalarShiftRightImm_D_size<0b0, 0b00000, "sshr">; +defm : Neon_ScalarShiftRImm_D_size_patterns<int_aarch64_neon_vshrds_n, SSHRddi>; +// Pattern to match llvm.arm.* intrinsic. +def : Neon_ScalarShiftImm_arm_D_size_patterns<sra, SSHRddi>; + +// Scalar Unsigned Shift Right (Immediate) +defm USHR : NeonI_ScalarShiftRightImm_D_size<0b1, 0b00000, "ushr">; +defm : Neon_ScalarShiftRImm_D_size_patterns<int_aarch64_neon_vshrdu_n, USHRddi>; +// Pattern to match llvm.arm.* intrinsic. +def : Neon_ScalarShiftImm_arm_D_size_patterns<srl, USHRddi>; + +// Scalar Signed Rounding Shift Right (Immediate) +defm SRSHR : NeonI_ScalarShiftRightImm_D_size<0b0, 0b00100, "srshr">; +defm : Neon_ScalarShiftRImm_D_size_patterns<int_aarch64_neon_vsrshr, SRSHRddi>; + +// Scalar Unigned Rounding Shift Right (Immediate) +defm URSHR : NeonI_ScalarShiftRightImm_D_size<0b1, 0b00100, "urshr">; +defm : Neon_ScalarShiftRImm_D_size_patterns<int_aarch64_neon_vurshr, URSHRddi>; + +// Scalar Signed Shift Right and Accumulate (Immediate) +def SSRA : NeonI_ScalarShiftRightImm_accum_D_size<0b0, 0b00010, "ssra">; +def : Neon_ScalarShiftRImm_accum_D_size_patterns + <int_aarch64_neon_vsrads_n, SSRA>; + +// Scalar Unsigned Shift Right and Accumulate (Immediate) +def USRA : NeonI_ScalarShiftRightImm_accum_D_size<0b1, 0b00010, "usra">; +def : Neon_ScalarShiftRImm_accum_D_size_patterns + <int_aarch64_neon_vsradu_n, USRA>; + +// Scalar Signed Rounding Shift Right and Accumulate (Immediate) +def SRSRA : NeonI_ScalarShiftRightImm_accum_D_size<0b0, 0b00110, "srsra">; +def : Neon_ScalarShiftRImm_accum_D_size_patterns + <int_aarch64_neon_vrsrads_n, SRSRA>; + +// Scalar Unsigned Rounding Shift Right and Accumulate (Immediate) +def URSRA : NeonI_ScalarShiftRightImm_accum_D_size<0b1, 0b00110, "ursra">; +def : Neon_ScalarShiftRImm_accum_D_size_patterns + <int_aarch64_neon_vrsradu_n, URSRA>; + +// Scalar Shift Left (Immediate) +defm SHL : NeonI_ScalarShiftLeftImm_D_size<0b0, 0b01010, "shl">; +defm : Neon_ScalarShiftLImm_D_size_patterns<int_aarch64_neon_vshld_n, SHLddi>; +// Pattern to match llvm.arm.* intrinsic. +def : Neon_ScalarShiftImm_arm_D_size_patterns<shl, SHLddi>; + +// Signed Saturating Shift Left (Immediate) +defm SQSHL : NeonI_ScalarShiftLeftImm_BHSD_size<0b0, 0b01110, "sqshl">; +defm : Neon_ScalarShiftLImm_BHSD_size_patterns<int_aarch64_neon_vqshls_n, + SQSHLbbi, SQSHLhhi, + SQSHLssi, SQSHLddi>; +// Pattern to match llvm.arm.* intrinsic. +defm : Neon_ScalarShiftLImm_D_size_patterns<Neon_sqrshlImm, SQSHLddi>; + +// Unsigned Saturating Shift Left (Immediate) +defm UQSHL : NeonI_ScalarShiftLeftImm_BHSD_size<0b1, 0b01110, "uqshl">; +defm : Neon_ScalarShiftLImm_BHSD_size_patterns<int_aarch64_neon_vqshlu_n, + UQSHLbbi, UQSHLhhi, + UQSHLssi, UQSHLddi>; +// Pattern to match llvm.arm.* intrinsic. +defm : Neon_ScalarShiftLImm_D_size_patterns<Neon_uqrshlImm, UQSHLddi>; + +// Signed Saturating Shift Left Unsigned (Immediate) +defm SQSHLU : NeonI_ScalarShiftLeftImm_BHSD_size<0b1, 0b01100, "sqshlu">; +defm : Neon_ScalarShiftLImm_BHSD_size_patterns<int_aarch64_neon_vsqshlu, + SQSHLUbbi, SQSHLUhhi, + SQSHLUssi, SQSHLUddi>; + +// Shift Right And Insert (Immediate) +def SRI : NeonI_ScalarShiftRightImm_accum_D_size<0b1, 0b01000, "sri">; +def : Neon_ScalarShiftRImm_accum_D_size_patterns + <int_aarch64_neon_vsri, SRI>; + +// Shift Left And Insert (Immediate) +def SLI : NeonI_ScalarShiftLeftImm_accum_D_size<0b1, 0b01010, "sli">; +def : Neon_ScalarShiftLImm_accum_D_size_patterns + <int_aarch64_neon_vsli, SLI>; + +// Signed Saturating Shift Right Narrow (Immediate) +defm SQSHRN : NeonI_ScalarShiftImm_narrow_HSD_size<0b0, 0b10010, "sqshrn">; +defm : Neon_ScalarShiftImm_narrow_HSD_size_patterns<int_aarch64_neon_vsqshrn, + SQSHRNbhi, SQSHRNhsi, + SQSHRNsdi>; + +// Unsigned Saturating Shift Right Narrow (Immediate) +defm UQSHRN : NeonI_ScalarShiftImm_narrow_HSD_size<0b1, 0b10010, "uqshrn">; +defm : Neon_ScalarShiftImm_narrow_HSD_size_patterns<int_aarch64_neon_vuqshrn, + UQSHRNbhi, UQSHRNhsi, + UQSHRNsdi>; + +// Signed Saturating Rounded Shift Right Narrow (Immediate) +defm SQRSHRN : NeonI_ScalarShiftImm_narrow_HSD_size<0b0, 0b10011, "sqrshrn">; +defm : Neon_ScalarShiftImm_narrow_HSD_size_patterns<int_aarch64_neon_vsqrshrn, + SQRSHRNbhi, SQRSHRNhsi, + SQRSHRNsdi>; + +// Unsigned Saturating Rounded Shift Right Narrow (Immediate) +defm UQRSHRN : NeonI_ScalarShiftImm_narrow_HSD_size<0b1, 0b10011, "uqrshrn">; +defm : Neon_ScalarShiftImm_narrow_HSD_size_patterns<int_aarch64_neon_vuqrshrn, + UQRSHRNbhi, UQRSHRNhsi, + UQRSHRNsdi>; + +// Signed Saturating Shift Right Unsigned Narrow (Immediate) +defm SQSHRUN : NeonI_ScalarShiftImm_narrow_HSD_size<0b1, 0b10000, "sqshrun">; +defm : Neon_ScalarShiftImm_narrow_HSD_size_patterns<int_aarch64_neon_vsqshrun, + SQSHRUNbhi, SQSHRUNhsi, + SQSHRUNsdi>; + +// Signed Saturating Rounded Shift Right Unsigned Narrow (Immediate) +defm SQRSHRUN : NeonI_ScalarShiftImm_narrow_HSD_size<0b1, 0b10001, "sqrshrun">; +defm : Neon_ScalarShiftImm_narrow_HSD_size_patterns<int_aarch64_neon_vsqrshrun, + SQRSHRUNbhi, SQRSHRUNhsi, + SQRSHRUNsdi>; + +// Scalar Signed Fixed-point Convert To Floating-Point (Immediate) +defm SCVTF_N : NeonI_ScalarShiftImm_cvt_SD_size<0b0, 0b11100, "scvtf">; +defm : Neon_ScalarShiftImm_scvtf_SD_size_patterns<int_aarch64_neon_vcvtf32_n_s32, + int_aarch64_neon_vcvtf64_n_s64, + SCVTF_Nssi, SCVTF_Nddi>; + +// Scalar Unsigned Fixed-point Convert To Floating-Point (Immediate) +defm UCVTF_N : NeonI_ScalarShiftImm_cvt_SD_size<0b1, 0b11100, "ucvtf">; +defm : Neon_ScalarShiftImm_scvtf_SD_size_patterns<int_aarch64_neon_vcvtf32_n_u32, + int_aarch64_neon_vcvtf64_n_u64, + UCVTF_Nssi, UCVTF_Nddi>; + +// Scalar Floating-point Convert To Signed Fixed-point (Immediate) +defm FCVTZS_N : NeonI_ScalarShiftImm_cvt_SD_size<0b0, 0b11111, "fcvtzs">; +defm : Neon_ScalarShiftImm_fcvts_SD_size_patterns<int_aarch64_neon_vcvts_n_s32_f32, + int_aarch64_neon_vcvtd_n_s64_f64, + FCVTZS_Nssi, FCVTZS_Nddi>; + +// Scalar Floating-point Convert To Unsigned Fixed-point (Immediate) +defm FCVTZU_N : NeonI_ScalarShiftImm_cvt_SD_size<0b1, 0b11111, "fcvtzu">; +defm : Neon_ScalarShiftImm_fcvts_SD_size_patterns<int_aarch64_neon_vcvts_n_u32_f32, + int_aarch64_neon_vcvtd_n_u64_f64, + FCVTZU_Nssi, FCVTZU_Nddi>; + +// Patterns For Convert Instructions Between v1f64 and v1i64 +class Neon_ScalarShiftImm_cvtf_v1f64_pattern<SDPatternOperator opnode, + Instruction INST> + : Pat<(v1f64 (opnode (v1i64 FPR64:$Rn), (i32 shr_imm64:$Imm))), + (INST FPR64:$Rn, imm:$Imm)>; + +class Neon_ScalarShiftImm_fcvt_v1f64_pattern<SDPatternOperator opnode, + Instruction INST> + : Pat<(v1i64 (opnode (v1f64 FPR64:$Rn), (i32 shr_imm64:$Imm))), + (INST FPR64:$Rn, imm:$Imm)>; + +def : Neon_ScalarShiftImm_cvtf_v1f64_pattern<int_arm_neon_vcvtfxs2fp, + SCVTF_Nddi>; + +def : Neon_ScalarShiftImm_cvtf_v1f64_pattern<int_arm_neon_vcvtfxu2fp, + UCVTF_Nddi>; + +def : Neon_ScalarShiftImm_fcvt_v1f64_pattern<int_arm_neon_vcvtfp2fxs, + FCVTZS_Nddi>; + +def : Neon_ScalarShiftImm_fcvt_v1f64_pattern<int_arm_neon_vcvtfp2fxu, + FCVTZU_Nddi>; + +// Scalar Integer Add +let isCommutable = 1 in { +def ADDddd : NeonI_Scalar3Same_D_size<0b0, 0b10000, "add">; +} + +// Scalar Integer Sub +def SUBddd : NeonI_Scalar3Same_D_size<0b1, 0b10000, "sub">; + +// Pattern for Scalar Integer Add and Sub with D register only +defm : Neon_Scalar3Same_D_size_patterns<add, ADDddd>; +defm : Neon_Scalar3Same_D_size_patterns<sub, SUBddd>; + +// Patterns to match llvm.aarch64.* intrinsic for Scalar Add, Sub +defm : Neon_Scalar3Same_D_size_patterns<int_aarch64_neon_vaddds, ADDddd>; +defm : Neon_Scalar3Same_D_size_patterns<int_aarch64_neon_vadddu, ADDddd>; +defm : Neon_Scalar3Same_D_size_patterns<int_aarch64_neon_vsubds, SUBddd>; +defm : Neon_Scalar3Same_D_size_patterns<int_aarch64_neon_vsubdu, SUBddd>; + +// Scalar Integer Saturating Add (Signed, Unsigned) +defm SQADD : NeonI_Scalar3Same_BHSD_sizes<0b0, 0b00001, "sqadd", 1>; +defm UQADD : NeonI_Scalar3Same_BHSD_sizes<0b1, 0b00001, "uqadd", 1>; + +// Scalar Integer Saturating Sub (Signed, Unsigned) +defm SQSUB : NeonI_Scalar3Same_BHSD_sizes<0b0, 0b00101, "sqsub", 0>; +defm UQSUB : NeonI_Scalar3Same_BHSD_sizes<0b1, 0b00101, "uqsub", 0>; + + +// Patterns to match llvm.aarch64.* intrinsic for +// Scalar Integer Saturating Add, Sub (Signed, Unsigned) +defm : Neon_Scalar3Same_BHSD_size_patterns<int_arm_neon_vqadds, SQADDbbb, + SQADDhhh, SQADDsss, SQADDddd>; +defm : Neon_Scalar3Same_BHSD_size_patterns<int_arm_neon_vqaddu, UQADDbbb, + UQADDhhh, UQADDsss, UQADDddd>; +defm : Neon_Scalar3Same_BHSD_size_patterns<int_arm_neon_vqsubs, SQSUBbbb, + SQSUBhhh, SQSUBsss, SQSUBddd>; +defm : Neon_Scalar3Same_BHSD_size_patterns<int_arm_neon_vqsubu, UQSUBbbb, + UQSUBhhh, UQSUBsss, UQSUBddd>; + +// Scalar Integer Saturating Doubling Multiply Half High +defm SQDMULH : NeonI_Scalar3Same_HS_sizes<0b0, 0b10110, "sqdmulh", 1>; + +// Scalar Integer Saturating Rounding Doubling Multiply Half High +defm SQRDMULH : NeonI_Scalar3Same_HS_sizes<0b1, 0b10110, "sqrdmulh", 1>; + +// Patterns to match llvm.arm.* intrinsic for +// Scalar Integer Saturating Doubling Multiply Half High and +// Scalar Integer Saturating Rounding Doubling Multiply Half High +defm : Neon_Scalar3Same_HS_size_patterns<int_arm_neon_vqdmulh, SQDMULHhhh, + SQDMULHsss>; +defm : Neon_Scalar3Same_HS_size_patterns<int_arm_neon_vqrdmulh, SQRDMULHhhh, + SQRDMULHsss>; + +// Scalar Floating-point Multiply Extended +defm FMULX : NeonI_Scalar3Same_SD_sizes<0b0, 0b0, 0b11011, "fmulx", 1>; + +// Scalar Floating-point Reciprocal Step +defm FRECPS : NeonI_Scalar3Same_SD_sizes<0b0, 0b0, 0b11111, "frecps", 0>; + +// Scalar Floating-point Reciprocal Square Root Step +defm FRSQRTS : NeonI_Scalar3Same_SD_sizes<0b0, 0b1, 0b11111, "frsqrts", 0>; + +// Patterns to match llvm.arm.* intrinsic for +// Scalar Floating-point Reciprocal Step and +// Scalar Floating-point Reciprocal Square Root Step +defm : Neon_Scalar3Same_SD_size_patterns<int_arm_neon_vrecps, FRECPSsss, + FRECPSddd>; +defm : Neon_Scalar3Same_SD_size_patterns<int_arm_neon_vrsqrts, FRSQRTSsss, + FRSQRTSddd>; + +def : Pat<(v1f64 (fsqrt (v1f64 FPR64:$Rn))), (FSQRTdd FPR64:$Rn)>; + +// Patterns to match llvm.aarch64.* intrinsic for +// Scalar Floating-point Multiply Extended, +multiclass Neon_Scalar3Same_MULX_SD_size_patterns<SDPatternOperator opnode, + Instruction INSTS, + Instruction INSTD> { + def : Pat<(f32 (opnode (f32 FPR32:$Rn), (f32 FPR32:$Rm))), + (INSTS FPR32:$Rn, FPR32:$Rm)>; + def : Pat<(f64 (opnode (f64 FPR64:$Rn), (f64 FPR64:$Rm))), + (INSTD FPR64:$Rn, FPR64:$Rm)>; +} + +defm : Neon_Scalar3Same_MULX_SD_size_patterns<int_aarch64_neon_vmulx, + FMULXsss,FMULXddd>; + +// Scalar Integer Shift Left (Signed, Unsigned) +def SSHLddd : NeonI_Scalar3Same_D_size<0b0, 0b01000, "sshl">; +def USHLddd : NeonI_Scalar3Same_D_size<0b1, 0b01000, "ushl">; + +// Patterns to match llvm.arm.* intrinsic for +// Scalar Integer Shift Left (Signed, Unsigned) +defm : Neon_Scalar3Same_D_size_patterns<int_arm_neon_vshifts, SSHLddd>; +defm : Neon_Scalar3Same_D_size_patterns<int_arm_neon_vshiftu, USHLddd>; + +// Patterns to match llvm.aarch64.* intrinsic for +// Scalar Integer Shift Left (Signed, Unsigned) +defm : Neon_Scalar3Same_D_size_patterns<int_aarch64_neon_vshlds, SSHLddd>; +defm : Neon_Scalar3Same_D_size_patterns<int_aarch64_neon_vshldu, USHLddd>; + +// Scalar Integer Saturating Shift Left (Signed, Unsigned) +defm SQSHL: NeonI_Scalar3Same_BHSD_sizes<0b0, 0b01001, "sqshl", 0>; +defm UQSHL: NeonI_Scalar3Same_BHSD_sizes<0b1, 0b01001, "uqshl", 0>; + +// Patterns to match llvm.aarch64.* intrinsic for +// Scalar Integer Saturating Shift Letf (Signed, Unsigned) +defm : Neon_Scalar3Same_BHSD_size_patterns<int_aarch64_neon_vqshls, SQSHLbbb, + SQSHLhhh, SQSHLsss, SQSHLddd>; +defm : Neon_Scalar3Same_BHSD_size_patterns<int_aarch64_neon_vqshlu, UQSHLbbb, + UQSHLhhh, UQSHLsss, UQSHLddd>; + +// Patterns to match llvm.arm.* intrinsic for +// Scalar Integer Saturating Shift Letf (Signed, Unsigned) +defm : Neon_Scalar3Same_D_size_patterns<int_arm_neon_vqshifts, SQSHLddd>; +defm : Neon_Scalar3Same_D_size_patterns<int_arm_neon_vqshiftu, UQSHLddd>; + +// Scalar Integer Rounding Shift Left (Signed, Unsigned) +def SRSHLddd: NeonI_Scalar3Same_D_size<0b0, 0b01010, "srshl">; +def URSHLddd: NeonI_Scalar3Same_D_size<0b1, 0b01010, "urshl">; + +// Patterns to match llvm.aarch64.* intrinsic for +// Scalar Integer Rounding Shift Left (Signed, Unsigned) +defm : Neon_Scalar3Same_D_size_patterns<int_aarch64_neon_vrshlds, SRSHLddd>; +defm : Neon_Scalar3Same_D_size_patterns<int_aarch64_neon_vrshldu, URSHLddd>; + +// Patterns to match llvm.arm.* intrinsic for +// Scalar Integer Rounding Shift Left (Signed, Unsigned) +defm : Neon_Scalar3Same_D_size_patterns<int_arm_neon_vrshifts, SRSHLddd>; +defm : Neon_Scalar3Same_D_size_patterns<int_arm_neon_vrshiftu, URSHLddd>; + +// Scalar Integer Saturating Rounding Shift Left (Signed, Unsigned) +defm SQRSHL: NeonI_Scalar3Same_BHSD_sizes<0b0, 0b01011, "sqrshl", 0>; +defm UQRSHL: NeonI_Scalar3Same_BHSD_sizes<0b1, 0b01011, "uqrshl", 0>; + +// Patterns to match llvm.aarch64.* intrinsic for +// Scalar Integer Saturating Rounding Shift Left (Signed, Unsigned) +defm : Neon_Scalar3Same_BHSD_size_patterns<int_aarch64_neon_vqrshls, SQRSHLbbb, + SQRSHLhhh, SQRSHLsss, SQRSHLddd>; +defm : Neon_Scalar3Same_BHSD_size_patterns<int_aarch64_neon_vqrshlu, UQRSHLbbb, + UQRSHLhhh, UQRSHLsss, UQRSHLddd>; + +// Patterns to match llvm.arm.* intrinsic for +// Scalar Integer Saturating Rounding Shift Left (Signed, Unsigned) +defm : Neon_Scalar3Same_D_size_patterns<int_arm_neon_vqrshifts, SQRSHLddd>; +defm : Neon_Scalar3Same_D_size_patterns<int_arm_neon_vqrshiftu, UQRSHLddd>; + +// Signed Saturating Doubling Multiply-Add Long +defm SQDMLAL : NeonI_Scalar3Diff_ml_HS_size<0b0, 0b1001, "sqdmlal">; +defm : Neon_Scalar3Diff_ml_HS_size_patterns<int_aarch64_neon_vqdmlal, + SQDMLALshh, SQDMLALdss>; + +// Signed Saturating Doubling Multiply-Subtract Long +defm SQDMLSL : NeonI_Scalar3Diff_ml_HS_size<0b0, 0b1011, "sqdmlsl">; +defm : Neon_Scalar3Diff_ml_HS_size_patterns<int_aarch64_neon_vqdmlsl, + SQDMLSLshh, SQDMLSLdss>; + +// Signed Saturating Doubling Multiply Long +defm SQDMULL : NeonI_Scalar3Diff_HS_size<0b0, 0b1101, "sqdmull">; +defm : Neon_Scalar3Diff_HS_size_patterns<int_arm_neon_vqdmull, + SQDMULLshh, SQDMULLdss>; + +// Scalar Signed Integer Convert To Floating-point +defm SCVTF : NeonI_Scalar2SameMisc_SD_size<0b0, 0b0, 0b11101, "scvtf">; +defm : Neon_Scalar2SameMisc_cvt_SD_size_patterns<int_aarch64_neon_vcvtf32_s32, + int_aarch64_neon_vcvtf64_s64, + SCVTFss, SCVTFdd>; + +// Scalar Unsigned Integer Convert To Floating-point +defm UCVTF : NeonI_Scalar2SameMisc_SD_size<0b1, 0b0, 0b11101, "ucvtf">; +defm : Neon_Scalar2SameMisc_cvt_SD_size_patterns<int_aarch64_neon_vcvtf32_u32, + int_aarch64_neon_vcvtf64_u64, + UCVTFss, UCVTFdd>; + +// Scalar Floating-point Converts +def FCVTXN : NeonI_Scalar2SameMisc_fcvtxn_D_size<0b1, 0b10110, "fcvtxn">; +def : Neon_Scalar2SameMisc_fcvtxn_D_size_patterns<int_aarch64_neon_fcvtxn, + FCVTXN>; + +defm FCVTNS : NeonI_Scalar2SameMisc_SD_size<0b0, 0b0, 0b11010, "fcvtns">; +defm : Neon_Scalar2SameMisc_fcvt_SD_size_patterns<int_aarch64_neon_fcvtns, + FCVTNSss, FCVTNSdd>; + +defm FCVTNU : NeonI_Scalar2SameMisc_SD_size<0b1, 0b0, 0b11010, "fcvtnu">; +defm : Neon_Scalar2SameMisc_fcvt_SD_size_patterns<int_aarch64_neon_fcvtnu, + FCVTNUss, FCVTNUdd>; + +defm FCVTMS : NeonI_Scalar2SameMisc_SD_size<0b0, 0b0, 0b11011, "fcvtms">; +defm : Neon_Scalar2SameMisc_fcvt_SD_size_patterns<int_aarch64_neon_fcvtms, + FCVTMSss, FCVTMSdd>; + +defm FCVTMU : NeonI_Scalar2SameMisc_SD_size<0b1, 0b0, 0b11011, "fcvtmu">; +defm : Neon_Scalar2SameMisc_fcvt_SD_size_patterns<int_aarch64_neon_fcvtmu, + FCVTMUss, FCVTMUdd>; + +defm FCVTAS : NeonI_Scalar2SameMisc_SD_size<0b0, 0b0, 0b11100, "fcvtas">; +defm : Neon_Scalar2SameMisc_fcvt_SD_size_patterns<int_aarch64_neon_fcvtas, + FCVTASss, FCVTASdd>; + +defm FCVTAU : NeonI_Scalar2SameMisc_SD_size<0b1, 0b0, 0b11100, "fcvtau">; +defm : Neon_Scalar2SameMisc_fcvt_SD_size_patterns<int_aarch64_neon_fcvtau, + FCVTAUss, FCVTAUdd>; + +defm FCVTPS : NeonI_Scalar2SameMisc_SD_size<0b0, 0b1, 0b11010, "fcvtps">; +defm : Neon_Scalar2SameMisc_fcvt_SD_size_patterns<int_aarch64_neon_fcvtps, + FCVTPSss, FCVTPSdd>; + +defm FCVTPU : NeonI_Scalar2SameMisc_SD_size<0b1, 0b1, 0b11010, "fcvtpu">; +defm : Neon_Scalar2SameMisc_fcvt_SD_size_patterns<int_aarch64_neon_fcvtpu, + FCVTPUss, FCVTPUdd>; + +defm FCVTZS : NeonI_Scalar2SameMisc_SD_size<0b0, 0b1, 0b11011, "fcvtzs">; +defm : Neon_Scalar2SameMisc_fcvt_SD_size_patterns<int_aarch64_neon_fcvtzs, + FCVTZSss, FCVTZSdd>; + +defm FCVTZU : NeonI_Scalar2SameMisc_SD_size<0b1, 0b1, 0b11011, "fcvtzu">; +defm : Neon_Scalar2SameMisc_fcvt_SD_size_patterns<int_aarch64_neon_fcvtzu, + FCVTZUss, FCVTZUdd>; + +// Patterns For Convert Instructions Between v1f64 and v1i64 +class Neon_Scalar2SameMisc_cvtf_v1f64_pattern<SDPatternOperator opnode, + Instruction INST> + : Pat<(v1f64 (opnode (v1i64 FPR64:$Rn))), (INST FPR64:$Rn)>; + +class Neon_Scalar2SameMisc_fcvt_v1f64_pattern<SDPatternOperator opnode, + Instruction INST> + : Pat<(v1i64 (opnode (v1f64 FPR64:$Rn))), (INST FPR64:$Rn)>; + +def : Neon_Scalar2SameMisc_cvtf_v1f64_pattern<sint_to_fp, SCVTFdd>; +def : Neon_Scalar2SameMisc_cvtf_v1f64_pattern<uint_to_fp, UCVTFdd>; + +def : Neon_Scalar2SameMisc_fcvt_v1f64_pattern<fp_to_sint, FCVTZSdd>; +def : Neon_Scalar2SameMisc_fcvt_v1f64_pattern<fp_to_uint, FCVTZUdd>; + +// Scalar Floating-point Reciprocal Estimate +defm FRECPE : NeonI_Scalar2SameMisc_SD_size<0b0, 0b1, 0b11101, "frecpe">; +defm : Neon_Scalar2SameMisc_SD_size_patterns<int_arm_neon_vrecpe, + FRECPEss, FRECPEdd>; + +// Scalar Floating-point Reciprocal Exponent +defm FRECPX : NeonI_Scalar2SameMisc_SD_size<0b0, 0b1, 0b11111, "frecpx">; +defm : Neon_Scalar2SameMisc_SD_size_patterns<int_aarch64_neon_vrecpx, + FRECPXss, FRECPXdd>; + +// Scalar Floating-point Reciprocal Square Root Estimate +defm FRSQRTE: NeonI_Scalar2SameMisc_SD_size<0b1, 0b1, 0b11101, "frsqrte">; +defm : Neon_Scalar2SameMisc_SD_size_patterns<int_arm_neon_vrsqrte, + FRSQRTEss, FRSQRTEdd>; + +// Scalar Floating-point Round +class Neon_ScalarFloatRound_pattern<SDPatternOperator opnode, Instruction INST> + : Pat<(v1f64 (opnode (v1f64 FPR64:$Rn))), (INST FPR64:$Rn)>; + +def : Neon_ScalarFloatRound_pattern<fceil, FRINTPdd>; +def : Neon_ScalarFloatRound_pattern<ffloor, FRINTMdd>; +def : Neon_ScalarFloatRound_pattern<ftrunc, FRINTZdd>; +def : Neon_ScalarFloatRound_pattern<frint, FRINTXdd>; +def : Neon_ScalarFloatRound_pattern<fnearbyint, FRINTIdd>; +def : Neon_ScalarFloatRound_pattern<frnd, FRINTAdd>; +def : Neon_ScalarFloatRound_pattern<int_aarch64_neon_frintn, FRINTNdd>; + +// Scalar Integer Compare + +// Scalar Compare Bitwise Equal +def CMEQddd: NeonI_Scalar3Same_D_size<0b1, 0b10001, "cmeq">; +def : Neon_Scalar3Same_cmp_D_size_patterns<int_aarch64_neon_vceq, CMEQddd>; + +class Neon_Scalar3Same_cmp_D_size_v1_patterns<SDPatternOperator opnode, + Instruction INSTD, + CondCode CC> + : Pat<(v1i64 (opnode (v1i64 FPR64:$Rn), (v1i64 FPR64:$Rm), CC)), + (INSTD FPR64:$Rn, FPR64:$Rm)>; + +def : Neon_Scalar3Same_cmp_D_size_v1_patterns<Neon_cmp, CMEQddd, SETEQ>; + +// Scalar Compare Signed Greather Than Or Equal +def CMGEddd: NeonI_Scalar3Same_D_size<0b0, 0b00111, "cmge">; +def : Neon_Scalar3Same_cmp_D_size_patterns<int_aarch64_neon_vcge, CMGEddd>; +def : Neon_Scalar3Same_cmp_D_size_v1_patterns<Neon_cmp, CMGEddd, SETGE>; + +// Scalar Compare Unsigned Higher Or Same +def CMHSddd: NeonI_Scalar3Same_D_size<0b1, 0b00111, "cmhs">; +def : Neon_Scalar3Same_cmp_D_size_patterns<int_aarch64_neon_vchs, CMHSddd>; +def : Neon_Scalar3Same_cmp_D_size_v1_patterns<Neon_cmp, CMHSddd, SETUGE>; + +// Scalar Compare Unsigned Higher +def CMHIddd: NeonI_Scalar3Same_D_size<0b1, 0b00110, "cmhi">; +def : Neon_Scalar3Same_cmp_D_size_patterns<int_aarch64_neon_vchi, CMHIddd>; +def : Neon_Scalar3Same_cmp_D_size_v1_patterns<Neon_cmp, CMHIddd, SETUGT>; + +// Scalar Compare Signed Greater Than +def CMGTddd: NeonI_Scalar3Same_D_size<0b0, 0b00110, "cmgt">; +def : Neon_Scalar3Same_cmp_D_size_patterns<int_aarch64_neon_vcgt, CMGTddd>; +def : Neon_Scalar3Same_cmp_D_size_v1_patterns<Neon_cmp, CMGTddd, SETGT>; + +// Scalar Compare Bitwise Test Bits +def CMTSTddd: NeonI_Scalar3Same_D_size<0b0, 0b10001, "cmtst">; +def : Neon_Scalar3Same_cmp_D_size_patterns<int_aarch64_neon_vtstd, CMTSTddd>; +def : Neon_Scalar3Same_cmp_D_size_patterns<Neon_tst, CMTSTddd>; + +// Scalar Compare Bitwise Equal To Zero +def CMEQddi: NeonI_Scalar2SameMisc_cmpz_D_size<0b0, 0b01001, "cmeq">; +def : Neon_Scalar2SameMisc_cmpz_D_size_patterns<int_aarch64_neon_vceq, + CMEQddi>; +def : Neon_Scalar2SameMisc_cmpz_D_V1_size_patterns<SETEQ, CMEQddi>; + +// Scalar Compare Signed Greather Than Or Equal To Zero +def CMGEddi: NeonI_Scalar2SameMisc_cmpz_D_size<0b1, 0b01000, "cmge">; +def : Neon_Scalar2SameMisc_cmpz_D_size_patterns<int_aarch64_neon_vcge, + CMGEddi>; +def : Neon_Scalar2SameMisc_cmpz_D_V1_size_patterns<SETGE, CMGEddi>; + +// Scalar Compare Signed Greater Than Zero +def CMGTddi: NeonI_Scalar2SameMisc_cmpz_D_size<0b0, 0b01000, "cmgt">; +def : Neon_Scalar2SameMisc_cmpz_D_size_patterns<int_aarch64_neon_vcgt, + CMGTddi>; +def : Neon_Scalar2SameMisc_cmpz_D_V1_size_patterns<SETGT, CMGTddi>; + +// Scalar Compare Signed Less Than Or Equal To Zero +def CMLEddi: NeonI_Scalar2SameMisc_cmpz_D_size<0b1, 0b01001, "cmle">; +def : Neon_Scalar2SameMisc_cmpz_D_size_patterns<int_aarch64_neon_vclez, + CMLEddi>; +def : Neon_Scalar2SameMisc_cmpz_D_V1_size_patterns<SETLE, CMLEddi>; + +// Scalar Compare Less Than Zero +def CMLTddi: NeonI_Scalar2SameMisc_cmpz_D_size<0b0, 0b01010, "cmlt">; +def : Neon_Scalar2SameMisc_cmpz_D_size_patterns<int_aarch64_neon_vcltz, + CMLTddi>; +def : Neon_Scalar2SameMisc_cmpz_D_V1_size_patterns<SETLT, CMLTddi>; + +// Scalar Floating-point Compare + +// Scalar Floating-point Compare Mask Equal +defm FCMEQ: NeonI_Scalar3Same_SD_sizes<0b0, 0b0, 0b11100, "fcmeq">; +defm : Neon_Scalar3Same_cmp_SD_size_patterns<int_aarch64_neon_vceq, + FCMEQsss, FCMEQddd>; +def : Neon_Scalar3Same_cmp_V1_D_size_patterns<SETEQ, FCMEQddd>; + +// Scalar Floating-point Compare Mask Equal To Zero +defm FCMEQZ: NeonI_Scalar2SameMisc_cmpz_SD_size<0b0, 0b01101, "fcmeq">; +defm : Neon_Scalar2SameMisc_cmpz_SD_size_patterns<int_aarch64_neon_vceq, + FCMEQZssi, FCMEQZddi>; +def : Pat<(v1i64 (Neon_cmpz (v1f64 FPR64:$Rn), (f32 fpz32:$FPImm), SETEQ)), + (FCMEQZddi FPR64:$Rn, fpz32:$FPImm)>; + +// Scalar Floating-point Compare Mask Greater Than Or Equal +defm FCMGE: NeonI_Scalar3Same_SD_sizes<0b1, 0b0, 0b11100, "fcmge">; +defm : Neon_Scalar3Same_cmp_SD_size_patterns<int_aarch64_neon_vcge, + FCMGEsss, FCMGEddd>; +def : Neon_Scalar3Same_cmp_V1_D_size_patterns<SETGE, FCMGEddd>; + +// Scalar Floating-point Compare Mask Greater Than Or Equal To Zero +defm FCMGEZ: NeonI_Scalar2SameMisc_cmpz_SD_size<0b1, 0b01100, "fcmge">; +defm : Neon_Scalar2SameMisc_cmpz_SD_size_patterns<int_aarch64_neon_vcge, + FCMGEZssi, FCMGEZddi>; + +// Scalar Floating-point Compare Mask Greather Than +defm FCMGT: NeonI_Scalar3Same_SD_sizes<0b1, 0b1, 0b11100, "fcmgt">; +defm : Neon_Scalar3Same_cmp_SD_size_patterns<int_aarch64_neon_vcgt, + FCMGTsss, FCMGTddd>; +def : Neon_Scalar3Same_cmp_V1_D_size_patterns<SETGT, FCMGTddd>; + +// Scalar Floating-point Compare Mask Greather Than Zero +defm FCMGTZ: NeonI_Scalar2SameMisc_cmpz_SD_size<0b0, 0b01100, "fcmgt">; +defm : Neon_Scalar2SameMisc_cmpz_SD_size_patterns<int_aarch64_neon_vcgt, + FCMGTZssi, FCMGTZddi>; + +// Scalar Floating-point Compare Mask Less Than Or Equal To Zero +defm FCMLEZ: NeonI_Scalar2SameMisc_cmpz_SD_size<0b1, 0b01101, "fcmle">; +defm : Neon_Scalar2SameMisc_cmpz_SD_size_patterns<int_aarch64_neon_vclez, + FCMLEZssi, FCMLEZddi>; + +// Scalar Floating-point Compare Mask Less Than Zero +defm FCMLTZ: NeonI_Scalar2SameMisc_cmpz_SD_size<0b0, 0b01110, "fcmlt">; +defm : Neon_Scalar2SameMisc_cmpz_SD_size_patterns<int_aarch64_neon_vcltz, + FCMLTZssi, FCMLTZddi>; + +// Scalar Floating-point Absolute Compare Mask Greater Than Or Equal +defm FACGE: NeonI_Scalar3Same_SD_sizes<0b1, 0b0, 0b11101, "facge">; +defm : Neon_Scalar3Same_cmp_SD_size_patterns<int_aarch64_neon_vcage, + FACGEsss, FACGEddd>; + +// Scalar Floating-point Absolute Compare Mask Greater Than +defm FACGT: NeonI_Scalar3Same_SD_sizes<0b1, 0b1, 0b11101, "facgt">; +defm : Neon_Scalar3Same_cmp_SD_size_patterns<int_aarch64_neon_vcagt, + FACGTsss, FACGTddd>; + +// Scakar Floating-point Absolute Difference +defm FABD: NeonI_Scalar3Same_SD_sizes<0b1, 0b1, 0b11010, "fabd">; +defm : Neon_Scalar3Same_SD_size_patterns<int_aarch64_neon_vabd, + FABDsss, FABDddd>; + +// Scalar Absolute Value +defm ABS : NeonI_Scalar2SameMisc_D_size<0b0, 0b01011, "abs">; +defm : Neon_Scalar2SameMisc_D_size_patterns<int_aarch64_neon_vabs, ABSdd>; + +// Scalar Signed Saturating Absolute Value +defm SQABS : NeonI_Scalar2SameMisc_BHSD_size<0b0, 0b00111, "sqabs">; +defm : Neon_Scalar2SameMisc_BHSD_size_patterns<int_arm_neon_vqabs, + SQABSbb, SQABShh, SQABSss, SQABSdd>; + +// Scalar Negate +defm NEG : NeonI_Scalar2SameMisc_D_size<0b1, 0b01011, "neg">; +defm : Neon_Scalar2SameMisc_D_size_patterns<int_aarch64_neon_vneg, NEGdd>; + +// Scalar Signed Saturating Negate +defm SQNEG : NeonI_Scalar2SameMisc_BHSD_size<0b1, 0b00111, "sqneg">; +defm : Neon_Scalar2SameMisc_BHSD_size_patterns<int_arm_neon_vqneg, + SQNEGbb, SQNEGhh, SQNEGss, SQNEGdd>; + +// Scalar Signed Saturating Accumulated of Unsigned Value +defm SUQADD : NeonI_Scalar2SameMisc_accum_BHSD_size<0b0, 0b00011, "suqadd">; +defm : Neon_Scalar2SameMisc_accum_BHSD_size_patterns<int_aarch64_neon_vuqadd, + SUQADDbb, SUQADDhh, + SUQADDss, SUQADDdd>; + +// Scalar Unsigned Saturating Accumulated of Signed Value +defm USQADD : NeonI_Scalar2SameMisc_accum_BHSD_size<0b1, 0b00011, "usqadd">; +defm : Neon_Scalar2SameMisc_accum_BHSD_size_patterns<int_aarch64_neon_vsqadd, + USQADDbb, USQADDhh, + USQADDss, USQADDdd>; + +def : Pat<(v1i64 (int_aarch64_neon_suqadd (v1i64 FPR64:$Src), + (v1i64 FPR64:$Rn))), + (SUQADDdd FPR64:$Src, FPR64:$Rn)>; + +def : Pat<(v1i64 (int_aarch64_neon_usqadd (v1i64 FPR64:$Src), + (v1i64 FPR64:$Rn))), + (USQADDdd FPR64:$Src, FPR64:$Rn)>; + +def : Pat<(v1i64 (int_arm_neon_vabs (v1i64 FPR64:$Rn))), + (ABSdd FPR64:$Rn)>; + +def : Pat<(v1i64 (int_arm_neon_vqabs (v1i64 FPR64:$Rn))), + (SQABSdd FPR64:$Rn)>; + +def : Pat<(v1i64 (int_arm_neon_vqneg (v1i64 FPR64:$Rn))), + (SQNEGdd FPR64:$Rn)>; + +def : Pat<(v1i64 (sub (v1i64 (bitconvert (v8i8 Neon_AllZero))), + (v1i64 FPR64:$Rn))), + (NEGdd FPR64:$Rn)>; + +// Scalar Signed Saturating Extract Unsigned Narrow +defm SQXTUN : NeonI_Scalar2SameMisc_narrow_HSD_size<0b1, 0b10010, "sqxtun">; +defm : Neon_Scalar2SameMisc_narrow_HSD_size_patterns<int_arm_neon_vqmovnsu, + SQXTUNbh, SQXTUNhs, + SQXTUNsd>; + +// Scalar Signed Saturating Extract Narrow +defm SQXTN : NeonI_Scalar2SameMisc_narrow_HSD_size<0b0, 0b10100, "sqxtn">; +defm : Neon_Scalar2SameMisc_narrow_HSD_size_patterns<int_arm_neon_vqmovns, + SQXTNbh, SQXTNhs, + SQXTNsd>; + +// Scalar Unsigned Saturating Extract Narrow +defm UQXTN : NeonI_Scalar2SameMisc_narrow_HSD_size<0b1, 0b10100, "uqxtn">; +defm : Neon_Scalar2SameMisc_narrow_HSD_size_patterns<int_arm_neon_vqmovnu, + UQXTNbh, UQXTNhs, + UQXTNsd>; + +// Scalar Reduce Pairwise + +multiclass NeonI_ScalarPair_D_sizes<bit u, bit size, bits<5> opcode, + string asmop, bit Commutable = 0> { + let isCommutable = Commutable in { + def _D_2D : NeonI_ScalarPair<u, {size, 0b1}, opcode, + (outs FPR64:$Rd), (ins VPR128:$Rn), + !strconcat(asmop, "\t$Rd, $Rn.2d"), + [], + NoItinerary>; + } +} + +multiclass NeonI_ScalarPair_SD_sizes<bit u, bit size, bits<5> opcode, + string asmop, bit Commutable = 0> + : NeonI_ScalarPair_D_sizes<u, size, opcode, asmop, Commutable> { + let isCommutable = Commutable in { + def _S_2S : NeonI_ScalarPair<u, {size, 0b0}, opcode, + (outs FPR32:$Rd), (ins VPR64:$Rn), + !strconcat(asmop, "\t$Rd, $Rn.2s"), + [], + NoItinerary>; + } +} + +// Scalar Reduce Addition Pairwise (Integer) with +// Pattern to match llvm.arm.* intrinsic +defm ADDPvv : NeonI_ScalarPair_D_sizes<0b0, 0b1, 0b11011, "addp", 0>; + +// Pattern to match llvm.aarch64.* intrinsic for +// Scalar Reduce Addition Pairwise (Integer) +def : Pat<(v1i64 (int_aarch64_neon_vpadd (v2i64 VPR128:$Rn))), + (ADDPvv_D_2D VPR128:$Rn)>; +def : Pat<(v1i64 (int_aarch64_neon_vaddv (v2i64 VPR128:$Rn))), + (ADDPvv_D_2D VPR128:$Rn)>; + +// Scalar Reduce Addition Pairwise (Floating Point) +defm FADDPvv : NeonI_ScalarPair_SD_sizes<0b1, 0b0, 0b01101, "faddp", 0>; + +// Scalar Reduce Maximum Pairwise (Floating Point) +defm FMAXPvv : NeonI_ScalarPair_SD_sizes<0b1, 0b0, 0b01111, "fmaxp", 0>; + +// Scalar Reduce Minimum Pairwise (Floating Point) +defm FMINPvv : NeonI_ScalarPair_SD_sizes<0b1, 0b1, 0b01111, "fminp", 0>; + +// Scalar Reduce maxNum Pairwise (Floating Point) +defm FMAXNMPvv : NeonI_ScalarPair_SD_sizes<0b1, 0b0, 0b01100, "fmaxnmp", 0>; + +// Scalar Reduce minNum Pairwise (Floating Point) +defm FMINNMPvv : NeonI_ScalarPair_SD_sizes<0b1, 0b1, 0b01100, "fminnmp", 0>; + +multiclass Neon_ScalarPair_SD_size_patterns<SDPatternOperator opnodeS, + SDPatternOperator opnodeD, + Instruction INSTS, + Instruction INSTD> { + def : Pat<(v1f32 (opnodeS (v2f32 VPR64:$Rn))), + (INSTS VPR64:$Rn)>; + def : Pat<(v1f64 (opnodeD (v2f64 VPR128:$Rn))), + (INSTD VPR128:$Rn)>; +} + +// Patterns to match llvm.aarch64.* intrinsic for +// Scalar Reduce Add, Max, Min, MaxiNum, MinNum Pairwise (Floating Point) +defm : Neon_ScalarPair_SD_size_patterns<int_aarch64_neon_vpfadd, + int_aarch64_neon_vpfaddq, FADDPvv_S_2S, FADDPvv_D_2D>; + +defm : Neon_ScalarPair_SD_size_patterns<int_aarch64_neon_vpmax, + int_aarch64_neon_vpmaxq, FMAXPvv_S_2S, FMAXPvv_D_2D>; + +defm : Neon_ScalarPair_SD_size_patterns<int_aarch64_neon_vpmin, + int_aarch64_neon_vpminq, FMINPvv_S_2S, FMINPvv_D_2D>; + +defm : Neon_ScalarPair_SD_size_patterns<int_aarch64_neon_vpfmaxnm, + int_aarch64_neon_vpfmaxnmq, FMAXNMPvv_S_2S, FMAXNMPvv_D_2D>; + +defm : Neon_ScalarPair_SD_size_patterns<int_aarch64_neon_vpfminnm, + int_aarch64_neon_vpfminnmq, FMINNMPvv_S_2S, FMINNMPvv_D_2D>; + +defm : Neon_ScalarPair_SD_size_patterns<int_aarch64_neon_vaddv, + int_aarch64_neon_vaddv, FADDPvv_S_2S, FADDPvv_D_2D>; + +def : Pat<(v1f32 (int_aarch64_neon_vaddv (v4f32 VPR128:$Rn))), + (FADDPvv_S_2S (v2f32 + (EXTRACT_SUBREG + (v4f32 (FADDP_4S (v4f32 VPR128:$Rn), (v4f32 VPR128:$Rn))), + sub_64)))>; + +defm : Neon_ScalarPair_SD_size_patterns<int_aarch64_neon_vmaxv, + int_aarch64_neon_vmaxv, FMAXPvv_S_2S, FMAXPvv_D_2D>; + +defm : Neon_ScalarPair_SD_size_patterns<int_aarch64_neon_vminv, + int_aarch64_neon_vminv, FMINPvv_S_2S, FMINPvv_D_2D>; + +defm : Neon_ScalarPair_SD_size_patterns<int_aarch64_neon_vmaxnmv, + int_aarch64_neon_vmaxnmv, FMAXNMPvv_S_2S, FMAXNMPvv_D_2D>; + +defm : Neon_ScalarPair_SD_size_patterns<int_aarch64_neon_vminnmv, + int_aarch64_neon_vminnmv, FMINNMPvv_S_2S, FMINNMPvv_D_2D>; + +// Scalar by element Arithmetic + +class NeonI_ScalarXIndexedElemArith<string asmop, bits<4> opcode, + string rmlane, bit u, bit szhi, bit szlo, + RegisterClass ResFPR, RegisterClass OpFPR, + RegisterOperand OpVPR, Operand OpImm> + : NeonI_ScalarXIndexedElem<u, szhi, szlo, opcode, + (outs ResFPR:$Rd), + (ins OpFPR:$Rn, OpVPR:$MRm, OpImm:$Imm), + asmop # "\t$Rd, $Rn, $MRm" # rmlane # "[$Imm]", + [], + NoItinerary> { + bits<3> Imm; + bits<5> MRm; +} + +class NeonI_ScalarXIndexedElemArith_Constraint_Impl<string asmop, bits<4> opcode, + string rmlane, + bit u, bit szhi, bit szlo, + RegisterClass ResFPR, + RegisterClass OpFPR, + RegisterOperand OpVPR, + Operand OpImm> + : NeonI_ScalarXIndexedElem<u, szhi, szlo, opcode, + (outs ResFPR:$Rd), + (ins ResFPR:$src, OpFPR:$Rn, OpVPR:$MRm, OpImm:$Imm), + asmop # "\t$Rd, $Rn, $MRm" # rmlane # "[$Imm]", + [], + NoItinerary> { + let Constraints = "$src = $Rd"; + bits<3> Imm; + bits<5> MRm; +} + +// Scalar Floating Point multiply (scalar, by element) +def FMULssv_4S : NeonI_ScalarXIndexedElemArith<"fmul", + 0b1001, ".s", 0b0, 0b1, 0b0, FPR32, FPR32, VPR128, neon_uimm2_bare> { + let Inst{11} = Imm{1}; // h + let Inst{21} = Imm{0}; // l + let Inst{20-16} = MRm; +} +def FMULddv_2D : NeonI_ScalarXIndexedElemArith<"fmul", + 0b1001, ".d", 0b0, 0b1, 0b1, FPR64, FPR64, VPR128, neon_uimm1_bare> { + let Inst{11} = Imm{0}; // h + let Inst{21} = 0b0; // l + let Inst{20-16} = MRm; +} + +// Scalar Floating Point multiply extended (scalar, by element) +def FMULXssv_4S : NeonI_ScalarXIndexedElemArith<"fmulx", + 0b1001, ".s", 0b1, 0b1, 0b0, FPR32, FPR32, VPR128, neon_uimm2_bare> { + let Inst{11} = Imm{1}; // h + let Inst{21} = Imm{0}; // l + let Inst{20-16} = MRm; +} +def FMULXddv_2D : NeonI_ScalarXIndexedElemArith<"fmulx", + 0b1001, ".d", 0b1, 0b1, 0b1, FPR64, FPR64, VPR128, neon_uimm1_bare> { + let Inst{11} = Imm{0}; // h + let Inst{21} = 0b0; // l + let Inst{20-16} = MRm; +} + +multiclass Neon_ScalarXIndexedElem_MUL_MULX_Patterns< + SDPatternOperator opnode, + Instruction INST, + ValueType ResTy, RegisterClass FPRC, ValueType OpTy, Operand OpImm, + ValueType OpNTy, ValueType ExTy, Operand OpNImm> { + + def : Pat<(ResTy (opnode (ResTy FPRC:$Rn), + (ResTy (vector_extract (OpTy VPR128:$MRm), OpImm:$Imm)))), + (ResTy (INST (ResTy FPRC:$Rn), (OpTy VPR128:$MRm), OpImm:$Imm))>; + + def : Pat<(ResTy (opnode (ResTy FPRC:$Rn), + (ResTy (vector_extract (OpNTy VPR64:$MRm), OpNImm:$Imm)))), + (ResTy (INST (ResTy FPRC:$Rn), + (ExTy (SUBREG_TO_REG (i64 0), VPR64:$MRm, sub_64)), + OpNImm:$Imm))>; + + // swapped operands + def : Pat<(ResTy (opnode + (ResTy (vector_extract (OpTy VPR128:$MRm), OpImm:$Imm)), + (ResTy FPRC:$Rn))), + (ResTy (INST (ResTy FPRC:$Rn), (OpTy VPR128:$MRm), OpImm:$Imm))>; + + def : Pat<(ResTy (opnode + (ResTy (vector_extract (OpNTy VPR64:$MRm), OpNImm:$Imm)), + (ResTy FPRC:$Rn))), + (ResTy (INST (ResTy FPRC:$Rn), + (ExTy (SUBREG_TO_REG (i64 0), VPR64:$MRm, sub_64)), + OpNImm:$Imm))>; +} + +// Patterns for Scalar Floating Point multiply (scalar, by element) +defm : Neon_ScalarXIndexedElem_MUL_MULX_Patterns<fmul, FMULssv_4S, + f32, FPR32, v4f32, neon_uimm2_bare, v2f32, v4f32, neon_uimm1_bare>; +defm : Neon_ScalarXIndexedElem_MUL_MULX_Patterns<fmul, FMULddv_2D, + f64, FPR64, v2f64, neon_uimm1_bare, v1f64, v2f64, neon_uimm0_bare>; + +// Patterns for Scalar Floating Point multiply extended (scalar, by element) +defm : Neon_ScalarXIndexedElem_MUL_MULX_Patterns<int_aarch64_neon_vmulx, + FMULXssv_4S, f32, FPR32, v4f32, neon_uimm2_bare, + v2f32, v4f32, neon_uimm1_bare>; +defm : Neon_ScalarXIndexedElem_MUL_MULX_Patterns<int_aarch64_neon_vmulx, + FMULXddv_2D, f64, FPR64, v2f64, neon_uimm1_bare, + v1f64, v2f64, neon_uimm0_bare>; + + +// Scalar Floating Point fused multiply-add (scalar, by element) +def FMLAssv_4S : NeonI_ScalarXIndexedElemArith_Constraint_Impl<"fmla", + 0b0001, ".s", 0b0, 0b1, 0b0, FPR32, FPR32, VPR128, neon_uimm2_bare> { + let Inst{11} = Imm{1}; // h + let Inst{21} = Imm{0}; // l + let Inst{20-16} = MRm; +} +def FMLAddv_2D : NeonI_ScalarXIndexedElemArith_Constraint_Impl<"fmla", + 0b0001, ".d", 0b0, 0b1, 0b1, FPR64, FPR64, VPR128, neon_uimm1_bare> { + let Inst{11} = Imm{0}; // h + let Inst{21} = 0b0; // l + let Inst{20-16} = MRm; +} + +// Scalar Floating Point fused multiply-subtract (scalar, by element) +def FMLSssv_4S : NeonI_ScalarXIndexedElemArith_Constraint_Impl<"fmls", + 0b0101, ".s", 0b0, 0b1, 0b0, FPR32, FPR32, VPR128, neon_uimm2_bare> { + let Inst{11} = Imm{1}; // h + let Inst{21} = Imm{0}; // l + let Inst{20-16} = MRm; +} +def FMLSddv_2D : NeonI_ScalarXIndexedElemArith_Constraint_Impl<"fmls", + 0b0101, ".d", 0b0, 0b1, 0b1, FPR64, FPR64, VPR128, neon_uimm1_bare> { + let Inst{11} = Imm{0}; // h + let Inst{21} = 0b0; // l + let Inst{20-16} = MRm; +} +// We are allowed to match the fma instruction regardless of compile options. +multiclass Neon_ScalarXIndexedElem_FMA_Patterns< + Instruction FMLAI, Instruction FMLSI, + ValueType ResTy, RegisterClass FPRC, ValueType OpTy, Operand OpImm, + ValueType OpNTy, ValueType ExTy, Operand OpNImm> { + // fmla + def : Pat<(ResTy (fma (ResTy FPRC:$Rn), + (ResTy (vector_extract (OpTy VPR128:$MRm), OpImm:$Imm)), + (ResTy FPRC:$Ra))), + (ResTy (FMLAI (ResTy FPRC:$Ra), + (ResTy FPRC:$Rn), (OpTy VPR128:$MRm), OpImm:$Imm))>; + + def : Pat<(ResTy (fma (ResTy FPRC:$Rn), + (ResTy (vector_extract (OpNTy VPR64:$MRm), OpNImm:$Imm)), + (ResTy FPRC:$Ra))), + (ResTy (FMLAI (ResTy FPRC:$Ra), + (ResTy FPRC:$Rn), + (ExTy (SUBREG_TO_REG (i64 0), VPR64:$MRm, sub_64)), + OpNImm:$Imm))>; + + // swapped fmla operands + def : Pat<(ResTy (fma + (ResTy (vector_extract (OpTy VPR128:$MRm), OpImm:$Imm)), + (ResTy FPRC:$Rn), + (ResTy FPRC:$Ra))), + (ResTy (FMLAI (ResTy FPRC:$Ra), + (ResTy FPRC:$Rn), (OpTy VPR128:$MRm), OpImm:$Imm))>; + + def : Pat<(ResTy (fma + (ResTy (vector_extract (OpNTy VPR64:$MRm), OpNImm:$Imm)), + (ResTy FPRC:$Rn), + (ResTy FPRC:$Ra))), + (ResTy (FMLAI (ResTy FPRC:$Ra), + (ResTy FPRC:$Rn), + (ExTy (SUBREG_TO_REG (i64 0), VPR64:$MRm, sub_64)), + OpNImm:$Imm))>; + + // fmls + def : Pat<(ResTy (fma (ResTy FPRC:$Rn), + (fneg (ResTy (vector_extract (OpTy VPR128:$MRm), OpImm:$Imm))), + (ResTy FPRC:$Ra))), + (ResTy (FMLSI (ResTy FPRC:$Ra), + (ResTy FPRC:$Rn), (OpTy VPR128:$MRm), OpImm:$Imm))>; + + def : Pat<(ResTy (fma (ResTy FPRC:$Rn), + (fneg (ResTy (vector_extract (OpNTy VPR64:$MRm), OpNImm:$Imm))), + (ResTy FPRC:$Ra))), + (ResTy (FMLSI (ResTy FPRC:$Ra), + (ResTy FPRC:$Rn), + (ExTy (SUBREG_TO_REG (i64 0), VPR64:$MRm, sub_64)), + OpNImm:$Imm))>; + + // swapped fmls operands + def : Pat<(ResTy (fma + (fneg (ResTy (vector_extract (OpTy VPR128:$MRm), OpImm:$Imm))), + (ResTy FPRC:$Rn), + (ResTy FPRC:$Ra))), + (ResTy (FMLSI (ResTy FPRC:$Ra), + (ResTy FPRC:$Rn), (OpTy VPR128:$MRm), OpImm:$Imm))>; + + def : Pat<(ResTy (fma + (fneg (ResTy (vector_extract (OpNTy VPR64:$MRm), OpNImm:$Imm))), + (ResTy FPRC:$Rn), + (ResTy FPRC:$Ra))), + (ResTy (FMLSI (ResTy FPRC:$Ra), + (ResTy FPRC:$Rn), + (ExTy (SUBREG_TO_REG (i64 0), VPR64:$MRm, sub_64)), + OpNImm:$Imm))>; +} + +// Scalar Floating Point fused multiply-add and +// multiply-subtract (scalar, by element) +defm : Neon_ScalarXIndexedElem_FMA_Patterns<FMLAssv_4S, FMLSssv_4S, + f32, FPR32, v4f32, neon_uimm2_bare, v2f32, v4f32, neon_uimm1_bare>; +defm : Neon_ScalarXIndexedElem_FMA_Patterns<FMLAddv_2D, FMLSddv_2D, + f64, FPR64, v2f64, neon_uimm1_bare, v1f64, v2f64, neon_uimm0_bare>; +defm : Neon_ScalarXIndexedElem_FMA_Patterns<FMLAddv_2D, FMLSddv_2D, + f64, FPR64, v2f64, neon_uimm1_bare, v1f64, v2f64, neon_uimm0_bare>; + +// Scalar Signed saturating doubling multiply long (scalar, by element) +def SQDMULLshv_4H : NeonI_ScalarXIndexedElemArith<"sqdmull", + 0b1011, ".h", 0b0, 0b0, 0b1, FPR32, FPR16, VPR64Lo, neon_uimm2_bare> { + let Inst{11} = 0b0; // h + let Inst{21} = Imm{1}; // l + let Inst{20} = Imm{0}; // m + let Inst{19-16} = MRm{3-0}; +} +def SQDMULLshv_8H : NeonI_ScalarXIndexedElemArith<"sqdmull", + 0b1011, ".h", 0b0, 0b0, 0b1, FPR32, FPR16, VPR128Lo, neon_uimm3_bare> { + let Inst{11} = Imm{2}; // h + let Inst{21} = Imm{1}; // l + let Inst{20} = Imm{0}; // m + let Inst{19-16} = MRm{3-0}; +} +def SQDMULLdsv_2S : NeonI_ScalarXIndexedElemArith<"sqdmull", + 0b1011, ".s", 0b0, 0b1, 0b0, FPR64, FPR32, VPR64, neon_uimm1_bare> { + let Inst{11} = 0b0; // h + let Inst{21} = Imm{0}; // l + let Inst{20-16} = MRm; +} +def SQDMULLdsv_4S : NeonI_ScalarXIndexedElemArith<"sqdmull", + 0b1011, ".s", 0b0, 0b1, 0b0, FPR64, FPR32, VPR128, neon_uimm2_bare> { + let Inst{11} = Imm{1}; // h + let Inst{21} = Imm{0}; // l + let Inst{20-16} = MRm; +} + +multiclass Neon_ScalarXIndexedElem_MUL_Patterns< + SDPatternOperator opnode, + Instruction INST, + ValueType ResTy, RegisterClass FPRC, + ValueType OpVTy, ValueType OpTy, + ValueType VecOpTy, ValueType ExTy, RegisterOperand VPRC, Operand OpImm> { + + def : Pat<(ResTy (opnode (OpVTy FPRC:$Rn), + (OpVTy (scalar_to_vector + (ExTy (vector_extract (VecOpTy VPRC:$MRm), OpImm:$Imm)))))), + (ResTy (INST (OpVTy FPRC:$Rn), (VecOpTy VPRC:$MRm), OpImm:$Imm))>; + + //swapped operands + def : Pat<(ResTy (opnode + (OpVTy (scalar_to_vector + (ExTy (vector_extract (VecOpTy VPRC:$MRm), OpImm:$Imm)))), + (OpVTy FPRC:$Rn))), + (ResTy (INST (OpVTy FPRC:$Rn), (VecOpTy VPRC:$MRm), OpImm:$Imm))>; +} + + +// Patterns for Scalar Signed saturating doubling +// multiply long (scalar, by element) +defm : Neon_ScalarXIndexedElem_MUL_Patterns<int_arm_neon_vqdmull, + SQDMULLshv_4H, v1i32, FPR16, v1i16, i16, v4i16, + i32, VPR64Lo, neon_uimm2_bare>; +defm : Neon_ScalarXIndexedElem_MUL_Patterns<int_arm_neon_vqdmull, + SQDMULLshv_8H, v1i32, FPR16, v1i16, i16, v8i16, + i32, VPR128Lo, neon_uimm3_bare>; +defm : Neon_ScalarXIndexedElem_MUL_Patterns<int_arm_neon_vqdmull, + SQDMULLdsv_2S, v1i64, FPR32, v1i32, i32, v2i32, + i32, VPR64Lo, neon_uimm1_bare>; +defm : Neon_ScalarXIndexedElem_MUL_Patterns<int_arm_neon_vqdmull, + SQDMULLdsv_4S, v1i64, FPR32, v1i32, i32, v4i32, + i32, VPR128Lo, neon_uimm2_bare>; + +// Scalar Signed saturating doubling multiply-add long (scalar, by element) +def SQDMLALshv_4H : NeonI_ScalarXIndexedElemArith_Constraint_Impl<"sqdmlal", + 0b0011, ".h", 0b0, 0b0, 0b1, FPR32, FPR16, VPR64Lo, neon_uimm2_bare> { + let Inst{11} = 0b0; // h + let Inst{21} = Imm{1}; // l + let Inst{20} = Imm{0}; // m + let Inst{19-16} = MRm{3-0}; +} +def SQDMLALshv_8H : NeonI_ScalarXIndexedElemArith_Constraint_Impl<"sqdmlal", + 0b0011, ".h", 0b0, 0b0, 0b1, FPR32, FPR16, VPR128Lo, neon_uimm3_bare> { + let Inst{11} = Imm{2}; // h + let Inst{21} = Imm{1}; // l + let Inst{20} = Imm{0}; // m + let Inst{19-16} = MRm{3-0}; +} +def SQDMLALdsv_2S : NeonI_ScalarXIndexedElemArith_Constraint_Impl<"sqdmlal", + 0b0011, ".s", 0b0, 0b1, 0b0, FPR64, FPR32, VPR64, neon_uimm1_bare> { + let Inst{11} = 0b0; // h + let Inst{21} = Imm{0}; // l + let Inst{20-16} = MRm; +} +def SQDMLALdsv_4S : NeonI_ScalarXIndexedElemArith_Constraint_Impl<"sqdmlal", + 0b0011, ".s", 0b0, 0b1, 0b0, FPR64, FPR32, VPR128, neon_uimm2_bare> { + let Inst{11} = Imm{1}; // h + let Inst{21} = Imm{0}; // l + let Inst{20-16} = MRm; +} + +// Scalar Signed saturating doubling +// multiply-subtract long (scalar, by element) +def SQDMLSLshv_4H : NeonI_ScalarXIndexedElemArith_Constraint_Impl<"sqdmlsl", + 0b0111, ".h", 0b0, 0b0, 0b1, FPR32, FPR16, VPR64Lo, neon_uimm2_bare> { + let Inst{11} = 0b0; // h + let Inst{21} = Imm{1}; // l + let Inst{20} = Imm{0}; // m + let Inst{19-16} = MRm{3-0}; +} +def SQDMLSLshv_8H : NeonI_ScalarXIndexedElemArith_Constraint_Impl<"sqdmlsl", + 0b0111, ".h", 0b0, 0b0, 0b1, FPR32, FPR16, VPR128Lo, neon_uimm3_bare> { + let Inst{11} = Imm{2}; // h + let Inst{21} = Imm{1}; // l + let Inst{20} = Imm{0}; // m + let Inst{19-16} = MRm{3-0}; +} +def SQDMLSLdsv_2S : NeonI_ScalarXIndexedElemArith_Constraint_Impl<"sqdmlsl", + 0b0111, ".s", 0b0, 0b1, 0b0, FPR64, FPR32, VPR64, neon_uimm1_bare> { + let Inst{11} = 0b0; // h + let Inst{21} = Imm{0}; // l + let Inst{20-16} = MRm; +} +def SQDMLSLdsv_4S : NeonI_ScalarXIndexedElemArith_Constraint_Impl<"sqdmlsl", + 0b0111, ".s", 0b0, 0b1, 0b0, FPR64, FPR32, VPR128, neon_uimm2_bare> { + let Inst{11} = Imm{1}; // h + let Inst{21} = Imm{0}; // l + let Inst{20-16} = MRm; +} + +multiclass Neon_ScalarXIndexedElem_MLAL_Patterns< + SDPatternOperator opnode, + SDPatternOperator coreopnode, + Instruction INST, + ValueType ResTy, RegisterClass ResFPRC, RegisterClass FPRC, + ValueType OpTy, + ValueType OpVTy, ValueType ExTy, RegisterOperand VPRC, Operand OpImm> { + + def : Pat<(ResTy (opnode + (ResTy ResFPRC:$Ra), + (ResTy (coreopnode (OpTy FPRC:$Rn), + (OpTy (scalar_to_vector + (ExTy (vector_extract (OpVTy VPRC:$MRm), OpImm:$Imm)))))))), + (ResTy (INST (ResTy ResFPRC:$Ra), + (OpTy FPRC:$Rn), (OpVTy VPRC:$MRm), OpImm:$Imm))>; + + // swapped operands + def : Pat<(ResTy (opnode + (ResTy ResFPRC:$Ra), + (ResTy (coreopnode + (OpTy (scalar_to_vector + (ExTy (vector_extract (OpVTy VPRC:$MRm), OpImm:$Imm)))), + (OpTy FPRC:$Rn))))), + (ResTy (INST (ResTy ResFPRC:$Ra), + (OpTy FPRC:$Rn), (OpVTy VPRC:$MRm), OpImm:$Imm))>; +} + +// Patterns for Scalar Signed saturating +// doubling multiply-add long (scalar, by element) +defm : Neon_ScalarXIndexedElem_MLAL_Patterns<int_arm_neon_vqadds, + int_arm_neon_vqdmull, SQDMLALshv_4H, v1i32, FPR32, FPR16, v1i16, v4i16, + i32, VPR64Lo, neon_uimm2_bare>; +defm : Neon_ScalarXIndexedElem_MLAL_Patterns<int_arm_neon_vqadds, + int_arm_neon_vqdmull, SQDMLALshv_8H, v1i32, FPR32, FPR16, v1i16, v8i16, + i32, VPR128Lo, neon_uimm3_bare>; +defm : Neon_ScalarXIndexedElem_MLAL_Patterns<int_arm_neon_vqadds, + int_arm_neon_vqdmull, SQDMLALdsv_2S, v1i64, FPR64, FPR32, v1i32, v2i32, + i32, VPR64Lo, neon_uimm1_bare>; +defm : Neon_ScalarXIndexedElem_MLAL_Patterns<int_arm_neon_vqadds, + int_arm_neon_vqdmull, SQDMLALdsv_4S, v1i64, FPR64, FPR32, v1i32, v4i32, + i32, VPR128Lo, neon_uimm2_bare>; + +// Patterns for Scalar Signed saturating +// doubling multiply-sub long (scalar, by element) +defm : Neon_ScalarXIndexedElem_MLAL_Patterns<int_arm_neon_vqsubs, + int_arm_neon_vqdmull, SQDMLSLshv_4H, v1i32, FPR32, FPR16, v1i16, v4i16, + i32, VPR64Lo, neon_uimm2_bare>; +defm : Neon_ScalarXIndexedElem_MLAL_Patterns<int_arm_neon_vqsubs, + int_arm_neon_vqdmull, SQDMLSLshv_8H, v1i32, FPR32, FPR16, v1i16, v8i16, + i32, VPR128Lo, neon_uimm3_bare>; +defm : Neon_ScalarXIndexedElem_MLAL_Patterns<int_arm_neon_vqsubs, + int_arm_neon_vqdmull, SQDMLSLdsv_2S, v1i64, FPR64, FPR32, v1i32, v2i32, + i32, VPR64Lo, neon_uimm1_bare>; +defm : Neon_ScalarXIndexedElem_MLAL_Patterns<int_arm_neon_vqsubs, + int_arm_neon_vqdmull, SQDMLSLdsv_4S, v1i64, FPR64, FPR32, v1i32, v4i32, + i32, VPR128Lo, neon_uimm2_bare>; + +// Scalar general arithmetic operation +class Neon_Scalar_GeneralMath2D_pattern<SDPatternOperator opnode, + Instruction INST> + : Pat<(v1f64 (opnode (v1f64 FPR64:$Rn))), (INST FPR64:$Rn)>; + +class Neon_Scalar_GeneralMath3D_pattern<SDPatternOperator opnode, + Instruction INST> + : Pat<(v1f64 (opnode (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))), + (INST FPR64:$Rn, FPR64:$Rm)>; + +class Neon_Scalar_GeneralMath4D_pattern<SDPatternOperator opnode, + Instruction INST> + : Pat<(v1f64 (opnode (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm), + (v1f64 FPR64:$Ra))), + (INST FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>; + +def : Neon_Scalar_GeneralMath3D_pattern<fadd, FADDddd>; +def : Neon_Scalar_GeneralMath3D_pattern<fmul, FMULddd>; +def : Neon_Scalar_GeneralMath3D_pattern<fsub, FSUBddd>; +def : Neon_Scalar_GeneralMath3D_pattern<fdiv, FDIVddd>; +def : Neon_Scalar_GeneralMath3D_pattern<int_arm_neon_vabds, FABDddd>; +def : Neon_Scalar_GeneralMath3D_pattern<int_arm_neon_vmaxs, FMAXddd>; +def : Neon_Scalar_GeneralMath3D_pattern<int_arm_neon_vmins, FMINddd>; +def : Neon_Scalar_GeneralMath3D_pattern<int_aarch64_neon_vmaxnm, FMAXNMddd>; +def : Neon_Scalar_GeneralMath3D_pattern<int_aarch64_neon_vminnm, FMINNMddd>; + +def : Neon_Scalar_GeneralMath2D_pattern<fabs, FABSdd>; +def : Neon_Scalar_GeneralMath2D_pattern<fneg, FNEGdd>; + +def : Neon_Scalar_GeneralMath4D_pattern<fma, FMADDdddd>; +def : Neon_Scalar_GeneralMath4D_pattern<fmsub, FMSUBdddd>; + +// Scalar Signed saturating doubling multiply returning +// high half (scalar, by element) +def SQDMULHhhv_4H : NeonI_ScalarXIndexedElemArith<"sqdmulh", + 0b1100, ".h", 0b0, 0b0, 0b1, FPR16, FPR16, VPR64Lo, neon_uimm2_bare> { + let Inst{11} = 0b0; // h + let Inst{21} = Imm{1}; // l + let Inst{20} = Imm{0}; // m + let Inst{19-16} = MRm{3-0}; +} +def SQDMULHhhv_8H : NeonI_ScalarXIndexedElemArith<"sqdmulh", + 0b1100, ".h", 0b0, 0b0, 0b1, FPR16, FPR16, VPR128Lo, neon_uimm3_bare> { + let Inst{11} = Imm{2}; // h + let Inst{21} = Imm{1}; // l + let Inst{20} = Imm{0}; // m + let Inst{19-16} = MRm{3-0}; +} +def SQDMULHssv_2S : NeonI_ScalarXIndexedElemArith<"sqdmulh", + 0b1100, ".s", 0b0, 0b1, 0b0, FPR32, FPR32, VPR64, neon_uimm1_bare> { + let Inst{11} = 0b0; // h + let Inst{21} = Imm{0}; // l + let Inst{20-16} = MRm; +} +def SQDMULHssv_4S : NeonI_ScalarXIndexedElemArith<"sqdmulh", + 0b1100, ".s", 0b0, 0b1, 0b0, FPR32, FPR32, VPR128, neon_uimm2_bare> { + let Inst{11} = Imm{1}; // h + let Inst{21} = Imm{0}; // l + let Inst{20-16} = MRm; +} + +// Patterns for Scalar Signed saturating doubling multiply returning +// high half (scalar, by element) +defm : Neon_ScalarXIndexedElem_MUL_Patterns<int_arm_neon_vqdmulh, + SQDMULHhhv_4H, v1i16, FPR16, v1i16, i16, v4i16, + i32, VPR64Lo, neon_uimm2_bare>; +defm : Neon_ScalarXIndexedElem_MUL_Patterns<int_arm_neon_vqdmulh, + SQDMULHhhv_8H, v1i16, FPR16, v1i16, i16, v8i16, + i32, VPR128Lo, neon_uimm3_bare>; +defm : Neon_ScalarXIndexedElem_MUL_Patterns<int_arm_neon_vqdmulh, + SQDMULHssv_2S, v1i32, FPR32, v1i32, i32, v2i32, + i32, VPR64Lo, neon_uimm1_bare>; +defm : Neon_ScalarXIndexedElem_MUL_Patterns<int_arm_neon_vqdmulh, + SQDMULHssv_4S, v1i32, FPR32, v1i32, i32, v4i32, + i32, VPR128Lo, neon_uimm2_bare>; + +// Scalar Signed saturating rounding doubling multiply +// returning high half (scalar, by element) +def SQRDMULHhhv_4H : NeonI_ScalarXIndexedElemArith<"sqrdmulh", + 0b1101, ".h", 0b0, 0b0, 0b1, FPR16, FPR16, VPR64Lo, neon_uimm2_bare> { + let Inst{11} = 0b0; // h + let Inst{21} = Imm{1}; // l + let Inst{20} = Imm{0}; // m + let Inst{19-16} = MRm{3-0}; +} +def SQRDMULHhhv_8H : NeonI_ScalarXIndexedElemArith<"sqrdmulh", + 0b1101, ".h", 0b0, 0b0, 0b1, FPR16, FPR16, VPR128Lo, neon_uimm3_bare> { + let Inst{11} = Imm{2}; // h + let Inst{21} = Imm{1}; // l + let Inst{20} = Imm{0}; // m + let Inst{19-16} = MRm{3-0}; +} +def SQRDMULHssv_2S : NeonI_ScalarXIndexedElemArith<"sqrdmulh", + 0b1101, ".s", 0b0, 0b1, 0b0, FPR32, FPR32, VPR64, neon_uimm1_bare> { + let Inst{11} = 0b0; // h + let Inst{21} = Imm{0}; // l + let Inst{20-16} = MRm; +} +def SQRDMULHssv_4S : NeonI_ScalarXIndexedElemArith<"sqrdmulh", + 0b1101, ".s", 0b0, 0b1, 0b0, FPR32, FPR32, VPR128, neon_uimm2_bare> { + let Inst{11} = Imm{1}; // h + let Inst{21} = Imm{0}; // l + let Inst{20-16} = MRm; +} + +defm : Neon_ScalarXIndexedElem_MUL_Patterns<int_arm_neon_vqrdmulh, + SQRDMULHhhv_4H, v1i16, FPR16, v1i16, i16, v4i16, i32, + VPR64Lo, neon_uimm2_bare>; +defm : Neon_ScalarXIndexedElem_MUL_Patterns<int_arm_neon_vqrdmulh, + SQRDMULHhhv_8H, v1i16, FPR16, v1i16, i16, v8i16, i32, + VPR128Lo, neon_uimm3_bare>; +defm : Neon_ScalarXIndexedElem_MUL_Patterns<int_arm_neon_vqrdmulh, + SQRDMULHssv_2S, v1i32, FPR32, v1i32, i32, v2i32, i32, + VPR64Lo, neon_uimm1_bare>; +defm : Neon_ScalarXIndexedElem_MUL_Patterns<int_arm_neon_vqrdmulh, + SQRDMULHssv_4S, v1i32, FPR32, v1i32, i32, v4i32, i32, + VPR128Lo, neon_uimm2_bare>; + +// Scalar Copy - DUP element to scalar +class NeonI_Scalar_DUP<string asmop, string asmlane, + RegisterClass ResRC, RegisterOperand VPRC, + Operand OpImm> + : NeonI_ScalarCopy<(outs ResRC:$Rd), (ins VPRC:$Rn, OpImm:$Imm), + asmop # "\t$Rd, $Rn." # asmlane # "[$Imm]", + [], + NoItinerary> { + bits<4> Imm; +} + +def DUPbv_B : NeonI_Scalar_DUP<"dup", "b", FPR8, VPR128, neon_uimm4_bare> { + let Inst{20-16} = {Imm{3}, Imm{2}, Imm{1}, Imm{0}, 0b1}; +} +def DUPhv_H : NeonI_Scalar_DUP<"dup", "h", FPR16, VPR128, neon_uimm3_bare> { + let Inst{20-16} = {Imm{2}, Imm{1}, Imm{0}, 0b1, 0b0}; +} +def DUPsv_S : NeonI_Scalar_DUP<"dup", "s", FPR32, VPR128, neon_uimm2_bare> { + let Inst{20-16} = {Imm{1}, Imm{0}, 0b1, 0b0, 0b0}; +} +def DUPdv_D : NeonI_Scalar_DUP<"dup", "d", FPR64, VPR128, neon_uimm1_bare> { + let Inst{20-16} = {Imm, 0b1, 0b0, 0b0, 0b0}; +} + +multiclass NeonI_Scalar_DUP_Elt_pattern<Instruction DUPI, ValueType ResTy, + ValueType OpTy, Operand OpImm, + ValueType OpNTy, ValueType ExTy, Operand OpNImm> { + def : Pat<(ResTy (vector_extract (OpTy VPR128:$Rn), OpImm:$Imm)), + (ResTy (DUPI (OpTy VPR128:$Rn), OpImm:$Imm))>; + + def : Pat<(ResTy (vector_extract (OpNTy VPR64:$Rn), OpNImm:$Imm)), + (ResTy (DUPI + (ExTy (SUBREG_TO_REG (i64 0), VPR64:$Rn, sub_64)), + OpNImm:$Imm))>; +} + +// Patterns for vector extract of FP data using scalar DUP instructions +defm : NeonI_Scalar_DUP_Elt_pattern<DUPsv_S, f32, + v4f32, neon_uimm2_bare, v2f32, v4f32, neon_uimm1_bare>; +defm : NeonI_Scalar_DUP_Elt_pattern<DUPdv_D, f64, + v2f64, neon_uimm1_bare, v1f64, v2f64, neon_uimm0_bare>; + +multiclass NeonI_Scalar_DUP_Ext_Vec_pattern<Instruction DUPI, + ValueType ResTy, ValueType OpTy,Operand OpLImm, + ValueType NOpTy, ValueType ExTy, Operand OpNImm> { + + def : Pat<(ResTy (extract_subvector (OpTy VPR128:$Rn), OpLImm:$Imm)), + (ResTy (DUPI VPR128:$Rn, OpLImm:$Imm))>; + + def : Pat<(ResTy (extract_subvector (NOpTy VPR64:$Rn), OpNImm:$Imm)), + (ResTy (DUPI + (ExTy (SUBREG_TO_REG (i64 0), VPR64:$Rn, sub_64)), + OpNImm:$Imm))>; +} + +// Patterns for extract subvectors of v1ix data using scalar DUP instructions. +defm : NeonI_Scalar_DUP_Ext_Vec_pattern<DUPbv_B, v1i8, v16i8, neon_uimm4_bare, + v8i8, v16i8, neon_uimm3_bare>; +defm : NeonI_Scalar_DUP_Ext_Vec_pattern<DUPhv_H, v1i16, v8i16, neon_uimm3_bare, + v4i16, v8i16, neon_uimm2_bare>; +defm : NeonI_Scalar_DUP_Ext_Vec_pattern<DUPsv_S, v1i32, v4i32, neon_uimm2_bare, + v2i32, v4i32, neon_uimm1_bare>; + +multiclass NeonI_Scalar_DUP_Copy_pattern1<Instruction DUPI, ValueType ResTy, + ValueType OpTy, ValueType ElemTy, + Operand OpImm, ValueType OpNTy, + ValueType ExTy, Operand OpNImm> { + + def : Pat<(ResTy (vector_insert (ResTy undef), + (ElemTy (vector_extract (OpTy VPR128:$Rn), OpImm:$Imm)), + (neon_uimm0_bare:$Imm))), + (ResTy (DUPI (OpTy VPR128:$Rn), OpImm:$Imm))>; + + def : Pat<(ResTy (vector_insert (ResTy undef), + (ElemTy (vector_extract (OpNTy VPR64:$Rn), OpNImm:$Imm)), + (OpNImm:$Imm))), + (ResTy (DUPI + (ExTy (SUBREG_TO_REG (i64 0), VPR64:$Rn, sub_64)), + OpNImm:$Imm))>; +} + +multiclass NeonI_Scalar_DUP_Copy_pattern2<Instruction DUPI, ValueType ResTy, + ValueType OpTy, ValueType ElemTy, + Operand OpImm, ValueType OpNTy, + ValueType ExTy, Operand OpNImm> { + + def : Pat<(ResTy (scalar_to_vector + (ElemTy (vector_extract (OpTy VPR128:$Rn), OpImm:$Imm)))), + (ResTy (DUPI (OpTy VPR128:$Rn), OpImm:$Imm))>; + + def : Pat<(ResTy (scalar_to_vector + (ElemTy (vector_extract (OpNTy VPR64:$Rn), OpNImm:$Imm)))), + (ResTy (DUPI + (ExTy (SUBREG_TO_REG (i64 0), VPR64:$Rn, sub_64)), + OpNImm:$Imm))>; +} + +// Patterns for vector copy to v1ix and v1fx vectors using scalar DUP +// instructions. +defm : NeonI_Scalar_DUP_Copy_pattern1<DUPdv_D, + v1i64, v2i64, i64, neon_uimm1_bare, + v1i64, v2i64, neon_uimm0_bare>; +defm : NeonI_Scalar_DUP_Copy_pattern1<DUPsv_S, + v1i32, v4i32, i32, neon_uimm2_bare, + v2i32, v4i32, neon_uimm1_bare>; +defm : NeonI_Scalar_DUP_Copy_pattern1<DUPhv_H, + v1i16, v8i16, i32, neon_uimm3_bare, + v4i16, v8i16, neon_uimm2_bare>; +defm : NeonI_Scalar_DUP_Copy_pattern1<DUPbv_B, + v1i8, v16i8, i32, neon_uimm4_bare, + v8i8, v16i8, neon_uimm3_bare>; +defm : NeonI_Scalar_DUP_Copy_pattern1<DUPdv_D, + v1f64, v2f64, f64, neon_uimm1_bare, + v1f64, v2f64, neon_uimm0_bare>; +defm : NeonI_Scalar_DUP_Copy_pattern1<DUPsv_S, + v1f32, v4f32, f32, neon_uimm2_bare, + v2f32, v4f32, neon_uimm1_bare>; +defm : NeonI_Scalar_DUP_Copy_pattern2<DUPdv_D, + v1i64, v2i64, i64, neon_uimm1_bare, + v1i64, v2i64, neon_uimm0_bare>; +defm : NeonI_Scalar_DUP_Copy_pattern2<DUPsv_S, + v1i32, v4i32, i32, neon_uimm2_bare, + v2i32, v4i32, neon_uimm1_bare>; +defm : NeonI_Scalar_DUP_Copy_pattern2<DUPhv_H, + v1i16, v8i16, i32, neon_uimm3_bare, + v4i16, v8i16, neon_uimm2_bare>; +defm : NeonI_Scalar_DUP_Copy_pattern2<DUPbv_B, + v1i8, v16i8, i32, neon_uimm4_bare, + v8i8, v16i8, neon_uimm3_bare>; +defm : NeonI_Scalar_DUP_Copy_pattern2<DUPdv_D, + v1f64, v2f64, f64, neon_uimm1_bare, + v1f64, v2f64, neon_uimm0_bare>; +defm : NeonI_Scalar_DUP_Copy_pattern2<DUPsv_S, + v1f32, v4f32, f32, neon_uimm2_bare, + v2f32, v4f32, neon_uimm1_bare>; + +multiclass NeonI_Scalar_DUP_alias<string asmop, string asmlane, + Instruction DUPI, Operand OpImm, + RegisterClass ResRC> { + def : NeonInstAlias<!strconcat(asmop, "$Rd, $Rn" # asmlane # "[$Imm]"), + (DUPI ResRC:$Rd, VPR128:$Rn, OpImm:$Imm), 0b0>; +} + +// Aliases for Scalar copy - DUP element (scalar) +// FIXME: This is actually the preferred syntax but TableGen can't deal with +// custom printing of aliases. +defm : NeonI_Scalar_DUP_alias<"mov", ".b", DUPbv_B, neon_uimm4_bare, FPR8>; +defm : NeonI_Scalar_DUP_alias<"mov", ".h", DUPhv_H, neon_uimm3_bare, FPR16>; +defm : NeonI_Scalar_DUP_alias<"mov", ".s", DUPsv_S, neon_uimm2_bare, FPR32>; +defm : NeonI_Scalar_DUP_alias<"mov", ".d", DUPdv_D, neon_uimm1_bare, FPR64>; + +multiclass NeonI_SDUP<PatFrag GetLow, PatFrag GetHigh, ValueType ResTy, + ValueType OpTy> { + def : Pat<(ResTy (GetLow VPR128:$Rn)), + (ResTy (EXTRACT_SUBREG (OpTy VPR128:$Rn), sub_64))>; + def : Pat<(ResTy (GetHigh VPR128:$Rn)), + (ResTy (DUPdv_D (OpTy VPR128:$Rn), 1))>; +} + +defm : NeonI_SDUP<Neon_Low16B, Neon_High16B, v8i8, v16i8>; +defm : NeonI_SDUP<Neon_Low8H, Neon_High8H, v4i16, v8i16>; +defm : NeonI_SDUP<Neon_Low4S, Neon_High4S, v2i32, v4i32>; +defm : NeonI_SDUP<Neon_Low2D, Neon_High2D, v1i64, v2i64>; +defm : NeonI_SDUP<Neon_Low4float, Neon_High4float, v2f32, v4f32>; +defm : NeonI_SDUP<Neon_Low2double, Neon_High2double, v1f64, v2f64>; + +//===----------------------------------------------------------------------===// +// Non-Instruction Patterns +//===----------------------------------------------------------------------===// + +// 64-bit vector bitcasts... + +def : Pat<(v1i64 (bitconvert (v8i8 VPR64:$src))), (v1i64 VPR64:$src)>; +def : Pat<(v2f32 (bitconvert (v8i8 VPR64:$src))), (v2f32 VPR64:$src)>; +def : Pat<(v2i32 (bitconvert (v8i8 VPR64:$src))), (v2i32 VPR64:$src)>; +def : Pat<(v4i16 (bitconvert (v8i8 VPR64:$src))), (v4i16 VPR64:$src)>; + +def : Pat<(v1i64 (bitconvert (v4i16 VPR64:$src))), (v1i64 VPR64:$src)>; +def : Pat<(v2i32 (bitconvert (v4i16 VPR64:$src))), (v2i32 VPR64:$src)>; +def : Pat<(v2f32 (bitconvert (v4i16 VPR64:$src))), (v2f32 VPR64:$src)>; +def : Pat<(v8i8 (bitconvert (v4i16 VPR64:$src))), (v8i8 VPR64:$src)>; + +def : Pat<(v1i64 (bitconvert (v2i32 VPR64:$src))), (v1i64 VPR64:$src)>; +def : Pat<(v2f32 (bitconvert (v2i32 VPR64:$src))), (v2f32 VPR64:$src)>; +def : Pat<(v4i16 (bitconvert (v2i32 VPR64:$src))), (v4i16 VPR64:$src)>; +def : Pat<(v8i8 (bitconvert (v2i32 VPR64:$src))), (v8i8 VPR64:$src)>; + +def : Pat<(v1i64 (bitconvert (v2f32 VPR64:$src))), (v1i64 VPR64:$src)>; +def : Pat<(v2i32 (bitconvert (v2f32 VPR64:$src))), (v2i32 VPR64:$src)>; +def : Pat<(v4i16 (bitconvert (v2f32 VPR64:$src))), (v4i16 VPR64:$src)>; +def : Pat<(v8i8 (bitconvert (v2f32 VPR64:$src))), (v8i8 VPR64:$src)>; + +def : Pat<(v2f32 (bitconvert (v1i64 VPR64:$src))), (v2f32 VPR64:$src)>; +def : Pat<(v2i32 (bitconvert (v1i64 VPR64:$src))), (v2i32 VPR64:$src)>; +def : Pat<(v4i16 (bitconvert (v1i64 VPR64:$src))), (v4i16 VPR64:$src)>; +def : Pat<(v8i8 (bitconvert (v1i64 VPR64:$src))), (v8i8 VPR64:$src)>; + +// ..and 128-bit vector bitcasts... + +def : Pat<(v2f64 (bitconvert (v16i8 VPR128:$src))), (v2f64 VPR128:$src)>; +def : Pat<(v2i64 (bitconvert (v16i8 VPR128:$src))), (v2i64 VPR128:$src)>; +def : Pat<(v4f32 (bitconvert (v16i8 VPR128:$src))), (v4f32 VPR128:$src)>; +def : Pat<(v4i32 (bitconvert (v16i8 VPR128:$src))), (v4i32 VPR128:$src)>; +def : Pat<(v8i16 (bitconvert (v16i8 VPR128:$src))), (v8i16 VPR128:$src)>; + +def : Pat<(v2f64 (bitconvert (v8i16 VPR128:$src))), (v2f64 VPR128:$src)>; +def : Pat<(v2i64 (bitconvert (v8i16 VPR128:$src))), (v2i64 VPR128:$src)>; +def : Pat<(v4i32 (bitconvert (v8i16 VPR128:$src))), (v4i32 VPR128:$src)>; +def : Pat<(v4f32 (bitconvert (v8i16 VPR128:$src))), (v4f32 VPR128:$src)>; +def : Pat<(v16i8 (bitconvert (v8i16 VPR128:$src))), (v16i8 VPR128:$src)>; + +def : Pat<(v2f64 (bitconvert (v4i32 VPR128:$src))), (v2f64 VPR128:$src)>; +def : Pat<(v2i64 (bitconvert (v4i32 VPR128:$src))), (v2i64 VPR128:$src)>; +def : Pat<(v4f32 (bitconvert (v4i32 VPR128:$src))), (v4f32 VPR128:$src)>; +def : Pat<(v8i16 (bitconvert (v4i32 VPR128:$src))), (v8i16 VPR128:$src)>; +def : Pat<(v16i8 (bitconvert (v4i32 VPR128:$src))), (v16i8 VPR128:$src)>; + +def : Pat<(v2f64 (bitconvert (v4f32 VPR128:$src))), (v2f64 VPR128:$src)>; +def : Pat<(v2i64 (bitconvert (v4f32 VPR128:$src))), (v2i64 VPR128:$src)>; +def : Pat<(v4i32 (bitconvert (v4f32 VPR128:$src))), (v4i32 VPR128:$src)>; +def : Pat<(v8i16 (bitconvert (v4f32 VPR128:$src))), (v8i16 VPR128:$src)>; +def : Pat<(v16i8 (bitconvert (v4f32 VPR128:$src))), (v16i8 VPR128:$src)>; + +def : Pat<(v2f64 (bitconvert (v2i64 VPR128:$src))), (v2f64 VPR128:$src)>; +def : Pat<(v4f32 (bitconvert (v2i64 VPR128:$src))), (v4f32 VPR128:$src)>; +def : Pat<(v4i32 (bitconvert (v2i64 VPR128:$src))), (v4i32 VPR128:$src)>; +def : Pat<(v8i16 (bitconvert (v2i64 VPR128:$src))), (v8i16 VPR128:$src)>; +def : Pat<(v16i8 (bitconvert (v2i64 VPR128:$src))), (v16i8 VPR128:$src)>; + +def : Pat<(v2i64 (bitconvert (v2f64 VPR128:$src))), (v2i64 VPR128:$src)>; +def : Pat<(v4f32 (bitconvert (v2f64 VPR128:$src))), (v4f32 VPR128:$src)>; +def : Pat<(v4i32 (bitconvert (v2f64 VPR128:$src))), (v4i32 VPR128:$src)>; +def : Pat<(v8i16 (bitconvert (v2f64 VPR128:$src))), (v8i16 VPR128:$src)>; +def : Pat<(v16i8 (bitconvert (v2f64 VPR128:$src))), (v16i8 VPR128:$src)>; + +// ...and scalar bitcasts... +def : Pat<(f16 (bitconvert (v1i16 FPR16:$src))), (f16 FPR16:$src)>; +def : Pat<(f32 (bitconvert (v1i32 FPR32:$src))), (f32 FPR32:$src)>; +def : Pat<(f64 (bitconvert (v1i64 FPR64:$src))), (f64 FPR64:$src)>; +def : Pat<(f32 (bitconvert (v1f32 FPR32:$src))), (f32 FPR32:$src)>; +def : Pat<(f64 (bitconvert (v1f64 FPR64:$src))), (f64 FPR64:$src)>; + +def : Pat<(i64 (bitconvert (v1i64 FPR64:$src))), (FMOVxd $src)>; +def : Pat<(i64 (bitconvert (v1f64 FPR64:$src))), (FMOVxd $src)>; +def : Pat<(i64 (bitconvert (v2i32 FPR64:$src))), (FMOVxd $src)>; +def : Pat<(i64 (bitconvert (v2f32 FPR64:$src))), (FMOVxd $src)>; +def : Pat<(i64 (bitconvert (v4i16 FPR64:$src))), (FMOVxd $src)>; +def : Pat<(i64 (bitconvert (v8i8 FPR64:$src))), (FMOVxd $src)>; + +def : Pat<(i32 (bitconvert (v1i32 FPR32:$src))), (FMOVws $src)>; + +def : Pat<(v8i8 (bitconvert (v1i64 VPR64:$src))), (v8i8 VPR64:$src)>; +def : Pat<(v4i16 (bitconvert (v1i64 VPR64:$src))), (v4i16 VPR64:$src)>; +def : Pat<(v2i32 (bitconvert (v1i64 VPR64:$src))), (v2i32 VPR64:$src)>; + +def : Pat<(f64 (bitconvert (v8i8 VPR64:$src))), (f64 VPR64:$src)>; +def : Pat<(f64 (bitconvert (v4i16 VPR64:$src))), (f64 VPR64:$src)>; +def : Pat<(f64 (bitconvert (v2i32 VPR64:$src))), (f64 VPR64:$src)>; +def : Pat<(f64 (bitconvert (v2f32 VPR64:$src))), (f64 VPR64:$src)>; +def : Pat<(f64 (bitconvert (v1i64 VPR64:$src))), (f64 VPR64:$src)>; + +def : Pat<(f128 (bitconvert (v16i8 VPR128:$src))), (f128 VPR128:$src)>; +def : Pat<(f128 (bitconvert (v8i16 VPR128:$src))), (f128 VPR128:$src)>; +def : Pat<(f128 (bitconvert (v4i32 VPR128:$src))), (f128 VPR128:$src)>; +def : Pat<(f128 (bitconvert (v2i64 VPR128:$src))), (f128 VPR128:$src)>; +def : Pat<(f128 (bitconvert (v4f32 VPR128:$src))), (f128 VPR128:$src)>; +def : Pat<(f128 (bitconvert (v2f64 VPR128:$src))), (f128 VPR128:$src)>; + +def : Pat<(v1i16 (bitconvert (f16 FPR16:$src))), (v1i16 FPR16:$src)>; +def : Pat<(v1i32 (bitconvert (f32 FPR32:$src))), (v1i32 FPR32:$src)>; +def : Pat<(v1i64 (bitconvert (f64 FPR64:$src))), (v1i64 FPR64:$src)>; +def : Pat<(v1f32 (bitconvert (f32 FPR32:$src))), (v1f32 FPR32:$src)>; +def : Pat<(v1f64 (bitconvert (f64 FPR64:$src))), (v1f64 FPR64:$src)>; + +def : Pat<(v1i64 (bitconvert (i64 GPR64:$src))), (FMOVdx $src)>; +def : Pat<(v1f64 (bitconvert (i64 GPR64:$src))), (FMOVdx $src)>; +def : Pat<(v2i32 (bitconvert (i64 GPR64:$src))), (FMOVdx $src)>; +def : Pat<(v2f32 (bitconvert (i64 GPR64:$src))), (FMOVdx $src)>; +def : Pat<(v4i16 (bitconvert (i64 GPR64:$src))), (FMOVdx $src)>; +def : Pat<(v8i8 (bitconvert (i64 GPR64:$src))), (FMOVdx $src)>; + +def : Pat<(v1i32 (bitconvert (i32 GPR32:$src))), (FMOVsw $src)>; + +def : Pat<(v8i8 (bitconvert (f64 FPR64:$src))), (v8i8 FPR64:$src)>; +def : Pat<(v4i16 (bitconvert (f64 FPR64:$src))), (v4i16 FPR64:$src)>; +def : Pat<(v2i32 (bitconvert (f64 FPR64:$src))), (v2i32 FPR64:$src)>; +def : Pat<(v2f32 (bitconvert (f64 FPR64:$src))), (v2f32 FPR64:$src)>; +def : Pat<(v1i64 (bitconvert (f64 FPR64:$src))), (v1i64 FPR64:$src)>; + +def : Pat<(v16i8 (bitconvert (f128 FPR128:$src))), (v16i8 FPR128:$src)>; +def : Pat<(v8i16 (bitconvert (f128 FPR128:$src))), (v8i16 FPR128:$src)>; +def : Pat<(v4i32 (bitconvert (f128 FPR128:$src))), (v4i32 FPR128:$src)>; +def : Pat<(v2i64 (bitconvert (f128 FPR128:$src))), (v2i64 FPR128:$src)>; +def : Pat<(v4f32 (bitconvert (f128 FPR128:$src))), (v4f32 FPR128:$src)>; +def : Pat<(v2f64 (bitconvert (f128 FPR128:$src))), (v2f64 FPR128:$src)>; + +// Scalar Three Same + +def neon_uimm3 : Operand<i64>, + ImmLeaf<i64, [{return Imm < 8;}]> { + let ParserMatchClass = uimm3_asmoperand; + let PrintMethod = "printUImmHexOperand"; +} + +def neon_uimm4 : Operand<i64>, + ImmLeaf<i64, [{return Imm < 16;}]> { + let ParserMatchClass = uimm4_asmoperand; + let PrintMethod = "printUImmHexOperand"; +} + +// Bitwise Extract +class NeonI_Extract<bit q, bits<2> op2, string asmop, + string OpS, RegisterOperand OpVPR, Operand OpImm> + : NeonI_BitExtract<q, op2, (outs OpVPR:$Rd), + (ins OpVPR:$Rn, OpVPR:$Rm, OpImm:$Index), + asmop # "\t$Rd." # OpS # ", $Rn." # OpS # + ", $Rm." # OpS # ", $Index", + [], + NoItinerary>{ + bits<4> Index; +} + +def EXTvvvi_8b : NeonI_Extract<0b0, 0b00, "ext", "8b", + VPR64, neon_uimm3> { + let Inst{14-11} = {0b0, Index{2}, Index{1}, Index{0}}; +} + +def EXTvvvi_16b: NeonI_Extract<0b1, 0b00, "ext", "16b", + VPR128, neon_uimm4> { + let Inst{14-11} = Index; +} + +class NI_Extract<ValueType OpTy, RegisterOperand OpVPR, Instruction INST, + Operand OpImm> + : Pat<(OpTy (Neon_vextract (OpTy OpVPR:$Rn), (OpTy OpVPR:$Rm), + (i64 OpImm:$Imm))), + (INST OpVPR:$Rn, OpVPR:$Rm, OpImm:$Imm)>; + +def : NI_Extract<v8i8, VPR64, EXTvvvi_8b, neon_uimm3>; +def : NI_Extract<v4i16, VPR64, EXTvvvi_8b, neon_uimm3>; +def : NI_Extract<v2i32, VPR64, EXTvvvi_8b, neon_uimm3>; +def : NI_Extract<v1i64, VPR64, EXTvvvi_8b, neon_uimm3>; +def : NI_Extract<v2f32, VPR64, EXTvvvi_8b, neon_uimm3>; +def : NI_Extract<v1f64, VPR64, EXTvvvi_8b, neon_uimm3>; +def : NI_Extract<v16i8, VPR128, EXTvvvi_16b, neon_uimm4>; +def : NI_Extract<v8i16, VPR128, EXTvvvi_16b, neon_uimm4>; +def : NI_Extract<v4i32, VPR128, EXTvvvi_16b, neon_uimm4>; +def : NI_Extract<v2i64, VPR128, EXTvvvi_16b, neon_uimm4>; +def : NI_Extract<v4f32, VPR128, EXTvvvi_16b, neon_uimm4>; +def : NI_Extract<v2f64, VPR128, EXTvvvi_16b, neon_uimm4>; + +// Table lookup +class NI_TBL<bit q, bits<2> op2, bits<2> len, bit op, + string asmop, string OpS, RegisterOperand OpVPR, + RegisterOperand VecList> + : NeonI_TBL<q, op2, len, op, + (outs OpVPR:$Rd), (ins VecList:$Rn, OpVPR:$Rm), + asmop # "\t$Rd." # OpS # ", $Rn, $Rm." # OpS, + [], + NoItinerary>; + +// The vectors in look up table are always 16b +multiclass NI_TBL_pat<bits<2> len, bit op, string asmop, string List> { + def _8b : NI_TBL<0, 0b00, len, op, asmop, "8b", VPR64, + !cast<RegisterOperand>(List # "16B_operand")>; + + def _16b : NI_TBL<1, 0b00, len, op, asmop, "16b", VPR128, + !cast<RegisterOperand>(List # "16B_operand")>; +} + +defm TBL1 : NI_TBL_pat<0b00, 0b0, "tbl", "VOne">; +defm TBL2 : NI_TBL_pat<0b01, 0b0, "tbl", "VPair">; +defm TBL3 : NI_TBL_pat<0b10, 0b0, "tbl", "VTriple">; +defm TBL4 : NI_TBL_pat<0b11, 0b0, "tbl", "VQuad">; + +// Table lookup extention +class NI_TBX<bit q, bits<2> op2, bits<2> len, bit op, + string asmop, string OpS, RegisterOperand OpVPR, + RegisterOperand VecList> + : NeonI_TBL<q, op2, len, op, + (outs OpVPR:$Rd), (ins OpVPR:$src, VecList:$Rn, OpVPR:$Rm), + asmop # "\t$Rd." # OpS # ", $Rn, $Rm." # OpS, + [], + NoItinerary> { + let Constraints = "$src = $Rd"; +} + +// The vectors in look up table are always 16b +multiclass NI_TBX_pat<bits<2> len, bit op, string asmop, string List> { + def _8b : NI_TBX<0, 0b00, len, op, asmop, "8b", VPR64, + !cast<RegisterOperand>(List # "16B_operand")>; + + def _16b : NI_TBX<1, 0b00, len, op, asmop, "16b", VPR128, + !cast<RegisterOperand>(List # "16B_operand")>; +} + +defm TBX1 : NI_TBX_pat<0b00, 0b1, "tbx", "VOne">; +defm TBX2 : NI_TBX_pat<0b01, 0b1, "tbx", "VPair">; +defm TBX3 : NI_TBX_pat<0b10, 0b1, "tbx", "VTriple">; +defm TBX4 : NI_TBX_pat<0b11, 0b1, "tbx", "VQuad">; + +class NeonI_INS_main<string asmop, string Res, ValueType ResTy, + RegisterClass OpGPR, ValueType OpTy, Operand OpImm> + : NeonI_copy<0b1, 0b0, 0b0011, + (outs VPR128:$Rd), (ins VPR128:$src, OpGPR:$Rn, OpImm:$Imm), + asmop # "\t$Rd." # Res # "[$Imm], $Rn", + [(set (ResTy VPR128:$Rd), + (ResTy (vector_insert + (ResTy VPR128:$src), + (OpTy OpGPR:$Rn), + (OpImm:$Imm))))], + NoItinerary> { + bits<4> Imm; + let Constraints = "$src = $Rd"; +} + +//Insert element (vector, from main) +def INSbw : NeonI_INS_main<"ins", "b", v16i8, GPR32, i32, + neon_uimm4_bare> { + let Inst{20-16} = {Imm{3}, Imm{2}, Imm{1}, Imm{0}, 0b1}; +} +def INShw : NeonI_INS_main<"ins", "h", v8i16, GPR32, i32, + neon_uimm3_bare> { + let Inst{20-16} = {Imm{2}, Imm{1}, Imm{0}, 0b1, 0b0}; +} +def INSsw : NeonI_INS_main<"ins", "s", v4i32, GPR32, i32, + neon_uimm2_bare> { + let Inst{20-16} = {Imm{1}, Imm{0}, 0b1, 0b0, 0b0}; +} +def INSdx : NeonI_INS_main<"ins", "d", v2i64, GPR64, i64, + neon_uimm1_bare> { + let Inst{20-16} = {Imm, 0b1, 0b0, 0b0, 0b0}; +} + +def : NeonInstAlias<"mov $Rd.b[$Imm], $Rn", + (INSbw VPR128:$Rd, GPR32:$Rn, neon_uimm4_bare:$Imm), 0>; +def : NeonInstAlias<"mov $Rd.h[$Imm], $Rn", + (INShw VPR128:$Rd, GPR32:$Rn, neon_uimm3_bare:$Imm), 0>; +def : NeonInstAlias<"mov $Rd.s[$Imm], $Rn", + (INSsw VPR128:$Rd, GPR32:$Rn, neon_uimm2_bare:$Imm), 0>; +def : NeonInstAlias<"mov $Rd.d[$Imm], $Rn", + (INSdx VPR128:$Rd, GPR64:$Rn, neon_uimm1_bare:$Imm), 0>; + +class Neon_INS_main_pattern <ValueType ResTy,ValueType ExtResTy, + RegisterClass OpGPR, ValueType OpTy, + Operand OpImm, Instruction INS> + : Pat<(ResTy (vector_insert + (ResTy VPR64:$src), + (OpTy OpGPR:$Rn), + (OpImm:$Imm))), + (ResTy (EXTRACT_SUBREG + (ExtResTy (INS (ExtResTy (SUBREG_TO_REG (i64 0), VPR64:$src, sub_64)), + OpGPR:$Rn, OpImm:$Imm)), sub_64))>; + +def INSbw_pattern : Neon_INS_main_pattern<v8i8, v16i8, GPR32, i32, + neon_uimm3_bare, INSbw>; +def INShw_pattern : Neon_INS_main_pattern<v4i16, v8i16, GPR32, i32, + neon_uimm2_bare, INShw>; +def INSsw_pattern : Neon_INS_main_pattern<v2i32, v4i32, GPR32, i32, + neon_uimm1_bare, INSsw>; +def INSdx_pattern : Neon_INS_main_pattern<v1i64, v2i64, GPR64, i64, + neon_uimm0_bare, INSdx>; + +class NeonI_INS_element<string asmop, string Res, Operand ResImm> + : NeonI_insert<0b1, 0b1, + (outs VPR128:$Rd), (ins VPR128:$src, VPR128:$Rn, + ResImm:$Immd, ResImm:$Immn), + asmop # "\t$Rd." # Res # "[$Immd], $Rn." # Res # "[$Immn]", + [], + NoItinerary> { + let Constraints = "$src = $Rd"; + bits<4> Immd; + bits<4> Immn; +} + +//Insert element (vector, from element) +def INSELb : NeonI_INS_element<"ins", "b", neon_uimm4_bare> { + let Inst{20-16} = {Immd{3}, Immd{2}, Immd{1}, Immd{0}, 0b1}; + let Inst{14-11} = {Immn{3}, Immn{2}, Immn{1}, Immn{0}}; +} +def INSELh : NeonI_INS_element<"ins", "h", neon_uimm3_bare> { + let Inst{20-16} = {Immd{2}, Immd{1}, Immd{0}, 0b1, 0b0}; + let Inst{14-11} = {Immn{2}, Immn{1}, Immn{0}, 0b0}; + // bit 11 is unspecified, but should be set to zero. +} +def INSELs : NeonI_INS_element<"ins", "s", neon_uimm2_bare> { + let Inst{20-16} = {Immd{1}, Immd{0}, 0b1, 0b0, 0b0}; + let Inst{14-11} = {Immn{1}, Immn{0}, 0b0, 0b0}; + // bits 11-12 are unspecified, but should be set to zero. +} +def INSELd : NeonI_INS_element<"ins", "d", neon_uimm1_bare> { + let Inst{20-16} = {Immd, 0b1, 0b0, 0b0, 0b0}; + let Inst{14-11} = {Immn{0}, 0b0, 0b0, 0b0}; + // bits 11-13 are unspecified, but should be set to zero. +} + +def : NeonInstAlias<"mov $Rd.b[$Immd], $Rn.b[$Immn]", + (INSELb VPR128:$Rd, VPR128:$Rn, + neon_uimm4_bare:$Immd, neon_uimm4_bare:$Immn), 0>; +def : NeonInstAlias<"mov $Rd.h[$Immd], $Rn.h[$Immn]", + (INSELh VPR128:$Rd, VPR128:$Rn, + neon_uimm3_bare:$Immd, neon_uimm3_bare:$Immn), 0>; +def : NeonInstAlias<"mov $Rd.s[$Immd], $Rn.s[$Immn]", + (INSELs VPR128:$Rd, VPR128:$Rn, + neon_uimm2_bare:$Immd, neon_uimm2_bare:$Immn), 0>; +def : NeonInstAlias<"mov $Rd.d[$Immd], $Rn.d[$Immn]", + (INSELd VPR128:$Rd, VPR128:$Rn, + neon_uimm1_bare:$Immd, neon_uimm1_bare:$Immn), 0>; + +multiclass Neon_INS_elt_pattern<ValueType ResTy, ValueType NaTy, + ValueType MidTy, Operand StImm, Operand NaImm, + Instruction INS> { +def : Pat<(ResTy (vector_insert + (ResTy VPR128:$src), + (MidTy (vector_extract + (ResTy VPR128:$Rn), + (StImm:$Immn))), + (StImm:$Immd))), + (INS (ResTy VPR128:$src), (ResTy VPR128:$Rn), + StImm:$Immd, StImm:$Immn)>; + +def : Pat <(ResTy (vector_insert + (ResTy VPR128:$src), + (MidTy (vector_extract + (NaTy VPR64:$Rn), + (NaImm:$Immn))), + (StImm:$Immd))), + (INS (ResTy VPR128:$src), + (ResTy (SUBREG_TO_REG (i64 0), (NaTy VPR64:$Rn), sub_64)), + StImm:$Immd, NaImm:$Immn)>; + +def : Pat <(NaTy (vector_insert + (NaTy VPR64:$src), + (MidTy (vector_extract + (ResTy VPR128:$Rn), + (StImm:$Immn))), + (NaImm:$Immd))), + (NaTy (EXTRACT_SUBREG + (ResTy (INS + (ResTy (SUBREG_TO_REG (i64 0), (NaTy VPR64:$src), sub_64)), + (ResTy VPR128:$Rn), + NaImm:$Immd, StImm:$Immn)), + sub_64))>; + +def : Pat <(NaTy (vector_insert + (NaTy VPR64:$src), + (MidTy (vector_extract + (NaTy VPR64:$Rn), + (NaImm:$Immn))), + (NaImm:$Immd))), + (NaTy (EXTRACT_SUBREG + (ResTy (INS + (ResTy (SUBREG_TO_REG (i64 0), (NaTy VPR64:$src), sub_64)), + (ResTy (SUBREG_TO_REG (i64 0), (NaTy VPR64:$Rn), sub_64)), + NaImm:$Immd, NaImm:$Immn)), + sub_64))>; +} + +defm : Neon_INS_elt_pattern<v4f32, v2f32, f32, neon_uimm2_bare, + neon_uimm1_bare, INSELs>; +defm : Neon_INS_elt_pattern<v2f64, v1f64, f64, neon_uimm1_bare, + neon_uimm0_bare, INSELd>; +defm : Neon_INS_elt_pattern<v16i8, v8i8, i32, neon_uimm4_bare, + neon_uimm3_bare, INSELb>; +defm : Neon_INS_elt_pattern<v8i16, v4i16, i32, neon_uimm3_bare, + neon_uimm2_bare, INSELh>; +defm : Neon_INS_elt_pattern<v4i32, v2i32, i32, neon_uimm2_bare, + neon_uimm1_bare, INSELs>; +defm : Neon_INS_elt_pattern<v2i64, v1i64, i64, neon_uimm1_bare, + neon_uimm0_bare, INSELd>; + +multiclass Neon_INS_elt_float_pattern<ValueType ResTy, ValueType NaTy, + ValueType MidTy, + RegisterClass OpFPR, Operand ResImm, + SubRegIndex SubIndex, Instruction INS> { +def : Pat <(ResTy (vector_insert + (ResTy VPR128:$src), + (MidTy OpFPR:$Rn), + (ResImm:$Imm))), + (INS (ResTy VPR128:$src), + (ResTy (SUBREG_TO_REG (i64 0), OpFPR:$Rn, SubIndex)), + ResImm:$Imm, + (i64 0))>; + +def : Pat <(NaTy (vector_insert + (NaTy VPR64:$src), + (MidTy OpFPR:$Rn), + (ResImm:$Imm))), + (NaTy (EXTRACT_SUBREG + (ResTy (INS + (ResTy (SUBREG_TO_REG (i64 0), (NaTy VPR64:$src), sub_64)), + (ResTy (SUBREG_TO_REG (i64 0), (MidTy OpFPR:$Rn), SubIndex)), + ResImm:$Imm, + (i64 0))), + sub_64))>; +} + +defm : Neon_INS_elt_float_pattern<v4f32, v2f32, f32, FPR32, neon_uimm2_bare, + sub_32, INSELs>; +defm : Neon_INS_elt_float_pattern<v2f64, v1f64, f64, FPR64, neon_uimm1_bare, + sub_64, INSELd>; + +class NeonI_SMOV<string asmop, string Res, bit Q, + ValueType OpTy, ValueType eleTy, + Operand OpImm, RegisterClass ResGPR, ValueType ResTy> + : NeonI_copy<Q, 0b0, 0b0101, + (outs ResGPR:$Rd), (ins VPR128:$Rn, OpImm:$Imm), + asmop # "\t$Rd, $Rn." # Res # "[$Imm]", + [(set (ResTy ResGPR:$Rd), + (ResTy (sext_inreg + (ResTy (vector_extract + (OpTy VPR128:$Rn), (OpImm:$Imm))), + eleTy)))], + NoItinerary> { + bits<4> Imm; +} + +//Signed integer move (main, from element) +def SMOVwb : NeonI_SMOV<"smov", "b", 0b0, v16i8, i8, neon_uimm4_bare, + GPR32, i32> { + let Inst{20-16} = {Imm{3}, Imm{2}, Imm{1}, Imm{0}, 0b1}; +} +def SMOVwh : NeonI_SMOV<"smov", "h", 0b0, v8i16, i16, neon_uimm3_bare, + GPR32, i32> { + let Inst{20-16} = {Imm{2}, Imm{1}, Imm{0}, 0b1, 0b0}; +} +def SMOVxb : NeonI_SMOV<"smov", "b", 0b1, v16i8, i8, neon_uimm4_bare, + GPR64, i64> { + let Inst{20-16} = {Imm{3}, Imm{2}, Imm{1}, Imm{0}, 0b1}; +} +def SMOVxh : NeonI_SMOV<"smov", "h", 0b1, v8i16, i16, neon_uimm3_bare, + GPR64, i64> { + let Inst{20-16} = {Imm{2}, Imm{1}, Imm{0}, 0b1, 0b0}; +} +def SMOVxs : NeonI_SMOV<"smov", "s", 0b1, v4i32, i32, neon_uimm2_bare, + GPR64, i64> { + let Inst{20-16} = {Imm{1}, Imm{0}, 0b1, 0b0, 0b0}; +} + +multiclass Neon_SMOVx_pattern <ValueType StTy, ValueType NaTy, + ValueType eleTy, Operand StImm, Operand NaImm, + Instruction SMOVI> { + def : Pat<(i64 (sext_inreg + (i64 (anyext + (i32 (vector_extract + (StTy VPR128:$Rn), (StImm:$Imm))))), + eleTy)), + (SMOVI VPR128:$Rn, StImm:$Imm)>; + + def : Pat<(i64 (sext + (i32 (vector_extract + (StTy VPR128:$Rn), (StImm:$Imm))))), + (SMOVI VPR128:$Rn, StImm:$Imm)>; + + def : Pat<(i64 (sext_inreg + (i64 (vector_extract + (NaTy VPR64:$Rn), (NaImm:$Imm))), + eleTy)), + (SMOVI (StTy (SUBREG_TO_REG (i64 0), VPR64:$Rn, sub_64)), + NaImm:$Imm)>; + + def : Pat<(i64 (sext_inreg + (i64 (anyext + (i32 (vector_extract + (NaTy VPR64:$Rn), (NaImm:$Imm))))), + eleTy)), + (SMOVI (StTy (SUBREG_TO_REG (i64 0), VPR64:$Rn, sub_64)), + NaImm:$Imm)>; + + def : Pat<(i64 (sext + (i32 (vector_extract + (NaTy VPR64:$Rn), (NaImm:$Imm))))), + (SMOVI (StTy (SUBREG_TO_REG (i64 0), VPR64:$Rn, sub_64)), + NaImm:$Imm)>; +} + +defm : Neon_SMOVx_pattern<v16i8, v8i8, i8, neon_uimm4_bare, + neon_uimm3_bare, SMOVxb>; +defm : Neon_SMOVx_pattern<v8i16, v4i16, i16, neon_uimm3_bare, + neon_uimm2_bare, SMOVxh>; +defm : Neon_SMOVx_pattern<v4i32, v2i32, i32, neon_uimm2_bare, + neon_uimm1_bare, SMOVxs>; + +class Neon_SMOVw_pattern <ValueType StTy, ValueType NaTy, + ValueType eleTy, Operand StImm, Operand NaImm, + Instruction SMOVI> + : Pat<(i32 (sext_inreg + (i32 (vector_extract + (NaTy VPR64:$Rn), (NaImm:$Imm))), + eleTy)), + (SMOVI (StTy (SUBREG_TO_REG (i64 0), VPR64:$Rn, sub_64)), + NaImm:$Imm)>; + +def : Neon_SMOVw_pattern<v16i8, v8i8, i8, neon_uimm4_bare, + neon_uimm3_bare, SMOVwb>; +def : Neon_SMOVw_pattern<v8i16, v4i16, i16, neon_uimm3_bare, + neon_uimm2_bare, SMOVwh>; + +class NeonI_UMOV<string asmop, string Res, bit Q, + ValueType OpTy, Operand OpImm, + RegisterClass ResGPR, ValueType ResTy> + : NeonI_copy<Q, 0b0, 0b0111, + (outs ResGPR:$Rd), (ins VPR128:$Rn, OpImm:$Imm), + asmop # "\t$Rd, $Rn." # Res # "[$Imm]", + [(set (ResTy ResGPR:$Rd), + (ResTy (vector_extract + (OpTy VPR128:$Rn), (OpImm:$Imm))))], + NoItinerary> { + bits<4> Imm; +} + +//Unsigned integer move (main, from element) +def UMOVwb : NeonI_UMOV<"umov", "b", 0b0, v16i8, neon_uimm4_bare, + GPR32, i32> { + let Inst{20-16} = {Imm{3}, Imm{2}, Imm{1}, Imm{0}, 0b1}; +} +def UMOVwh : NeonI_UMOV<"umov", "h", 0b0, v8i16, neon_uimm3_bare, + GPR32, i32> { + let Inst{20-16} = {Imm{2}, Imm{1}, Imm{0}, 0b1, 0b0}; +} +def UMOVws : NeonI_UMOV<"umov", "s", 0b0, v4i32, neon_uimm2_bare, + GPR32, i32> { + let Inst{20-16} = {Imm{1}, Imm{0}, 0b1, 0b0, 0b0}; +} +def UMOVxd : NeonI_UMOV<"umov", "d", 0b1, v2i64, neon_uimm1_bare, + GPR64, i64> { + let Inst{20-16} = {Imm, 0b1, 0b0, 0b0, 0b0}; +} + +def : NeonInstAlias<"mov $Rd, $Rn.s[$Imm]", + (UMOVws GPR32:$Rd, VPR128:$Rn, neon_uimm2_bare:$Imm), 0>; +def : NeonInstAlias<"mov $Rd, $Rn.d[$Imm]", + (UMOVxd GPR64:$Rd, VPR128:$Rn, neon_uimm1_bare:$Imm), 0>; + +class Neon_UMOV_pattern <ValueType StTy, ValueType NaTy, ValueType ResTy, + Operand StImm, Operand NaImm, + Instruction SMOVI> + : Pat<(ResTy (vector_extract + (NaTy VPR64:$Rn), NaImm:$Imm)), + (SMOVI (StTy (SUBREG_TO_REG (i64 0), VPR64:$Rn, sub_64)), + NaImm:$Imm)>; + +def : Neon_UMOV_pattern<v16i8, v8i8, i32, neon_uimm4_bare, + neon_uimm3_bare, UMOVwb>; +def : Neon_UMOV_pattern<v8i16, v4i16, i32, neon_uimm3_bare, + neon_uimm2_bare, UMOVwh>; +def : Neon_UMOV_pattern<v4i32, v2i32, i32, neon_uimm2_bare, + neon_uimm1_bare, UMOVws>; + +def : Pat<(i32 (and + (i32 (vector_extract + (v16i8 VPR128:$Rn), (neon_uimm4_bare:$Imm))), + 255)), + (UMOVwb VPR128:$Rn, neon_uimm4_bare:$Imm)>; + +def : Pat<(i32 (and + (i32 (vector_extract + (v8i16 VPR128:$Rn), (neon_uimm3_bare:$Imm))), + 65535)), + (UMOVwh VPR128:$Rn, neon_uimm3_bare:$Imm)>; + +def : Pat<(i64 (zext + (i32 (vector_extract + (v2i64 VPR128:$Rn), (neon_uimm1_bare:$Imm))))), + (UMOVxd VPR128:$Rn, neon_uimm1_bare:$Imm)>; + +def : Pat<(i32 (and + (i32 (vector_extract + (v8i8 VPR64:$Rn), (neon_uimm3_bare:$Imm))), + 255)), + (UMOVwb (SUBREG_TO_REG (i64 0), VPR64:$Rn, sub_64), + neon_uimm3_bare:$Imm)>; + +def : Pat<(i32 (and + (i32 (vector_extract + (v4i16 VPR64:$Rn), (neon_uimm2_bare:$Imm))), + 65535)), + (UMOVwh (SUBREG_TO_REG (i64 0), VPR64:$Rn, sub_64), + neon_uimm2_bare:$Imm)>; + +def : Pat<(i64 (zext + (i32 (vector_extract + (v1i64 VPR64:$Rn), (neon_uimm0_bare:$Imm))))), + (UMOVxd (SUBREG_TO_REG (i64 0), VPR64:$Rn, sub_64), + neon_uimm0_bare:$Imm)>; + +// Additional copy patterns for scalar types +def : Pat<(i32 (vector_extract (v1i8 FPR8:$Rn), (i64 0))), + (UMOVwb (v16i8 + (SUBREG_TO_REG (i64 0), FPR8:$Rn, sub_8)), (i64 0))>; + +def : Pat<(i32 (vector_extract (v1i16 FPR16:$Rn), (i64 0))), + (UMOVwh (v8i16 + (SUBREG_TO_REG (i64 0), FPR16:$Rn, sub_16)), (i64 0))>; + +def : Pat<(i32 (vector_extract (v1i32 FPR32:$Rn), (i64 0))), + (FMOVws FPR32:$Rn)>; + +def : Pat<(i64 (vector_extract (v1i64 FPR64:$Rn), (i64 0))), + (FMOVxd FPR64:$Rn)>; + +def : Pat<(f64 (vector_extract (v1f64 FPR64:$Rn), (i64 0))), + (f64 FPR64:$Rn)>; + +def : Pat<(f32 (vector_extract (v1f32 FPR32:$Rn), (i64 0))), + (f32 FPR32:$Rn)>; + +def : Pat<(v1i8 (scalar_to_vector GPR32:$Rn)), + (v1i8 (EXTRACT_SUBREG (v16i8 + (INSbw (v16i8 (IMPLICIT_DEF)), $Rn, (i64 0))), + sub_8))>; + +def : Pat<(v1i16 (scalar_to_vector GPR32:$Rn)), + (v1i16 (EXTRACT_SUBREG (v8i16 + (INShw (v8i16 (IMPLICIT_DEF)), $Rn, (i64 0))), + sub_16))>; + +def : Pat<(v1i32 (scalar_to_vector GPR32:$src)), + (FMOVsw $src)>; + +def : Pat<(v1i64 (scalar_to_vector GPR64:$src)), + (FMOVdx $src)>; + +def : Pat<(v1f32 (scalar_to_vector (f32 FPR32:$Rn))), + (v1f32 FPR32:$Rn)>; +def : Pat<(v1f64 (scalar_to_vector (f64 FPR64:$Rn))), + (v1f64 FPR64:$Rn)>; + +def : Pat<(v1f64 (scalar_to_vector (f64 FPR64:$src))), + (FMOVdd $src)>; + +def : Pat<(v2f64 (scalar_to_vector (f64 FPR64:$src))), + (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), + (f64 FPR64:$src), sub_64)>; + +class NeonI_DUP_Elt<bit Q, string asmop, string rdlane, string rnlane, + RegisterOperand ResVPR, Operand OpImm> + : NeonI_copy<Q, 0b0, 0b0000, (outs ResVPR:$Rd), + (ins VPR128:$Rn, OpImm:$Imm), + asmop # "\t$Rd" # rdlane # ", $Rn" # rnlane # "[$Imm]", + [], + NoItinerary> { + bits<4> Imm; +} + +def DUPELT16b : NeonI_DUP_Elt<0b1, "dup", ".16b", ".b", VPR128, + neon_uimm4_bare> { + let Inst{20-16} = {Imm{3}, Imm{2}, Imm{1}, Imm{0}, 0b1}; +} + +def DUPELT8h : NeonI_DUP_Elt<0b1, "dup", ".8h", ".h", VPR128, + neon_uimm3_bare> { + let Inst{20-16} = {Imm{2}, Imm{1}, Imm{0}, 0b1, 0b0}; +} + +def DUPELT4s : NeonI_DUP_Elt<0b1, "dup", ".4s", ".s", VPR128, + neon_uimm2_bare> { + let Inst{20-16} = {Imm{1}, Imm{0}, 0b1, 0b0, 0b0}; +} + +def DUPELT2d : NeonI_DUP_Elt<0b1, "dup", ".2d", ".d", VPR128, + neon_uimm1_bare> { + let Inst{20-16} = {Imm, 0b1, 0b0, 0b0, 0b0}; +} + +def DUPELT8b : NeonI_DUP_Elt<0b0, "dup", ".8b", ".b", VPR64, + neon_uimm4_bare> { + let Inst{20-16} = {Imm{3}, Imm{2}, Imm{1}, Imm{0}, 0b1}; +} + +def DUPELT4h : NeonI_DUP_Elt<0b0, "dup", ".4h", ".h", VPR64, + neon_uimm3_bare> { + let Inst{20-16} = {Imm{2}, Imm{1}, Imm{0}, 0b1, 0b0}; +} + +def DUPELT2s : NeonI_DUP_Elt<0b0, "dup", ".2s", ".s", VPR64, + neon_uimm2_bare> { + let Inst{20-16} = {Imm{1}, Imm{0}, 0b1, 0b0, 0b0}; +} + +multiclass NeonI_DUP_Elt_pattern<Instruction DUPELT, ValueType ResTy, + ValueType OpTy,ValueType NaTy, + ValueType ExTy, Operand OpLImm, + Operand OpNImm> { +def : Pat<(ResTy (Neon_vduplane (OpTy VPR128:$Rn), OpLImm:$Imm)), + (ResTy (DUPELT (OpTy VPR128:$Rn), OpLImm:$Imm))>; + +def : Pat<(ResTy (Neon_vduplane + (NaTy VPR64:$Rn), OpNImm:$Imm)), + (ResTy (DUPELT + (ExTy (SUBREG_TO_REG (i64 0), VPR64:$Rn, sub_64)), OpNImm:$Imm))>; +} +defm : NeonI_DUP_Elt_pattern<DUPELT16b, v16i8, v16i8, v8i8, v16i8, + neon_uimm4_bare, neon_uimm3_bare>; +defm : NeonI_DUP_Elt_pattern<DUPELT8b, v8i8, v16i8, v8i8, v16i8, + neon_uimm4_bare, neon_uimm3_bare>; +defm : NeonI_DUP_Elt_pattern<DUPELT8h, v8i16, v8i16, v4i16, v8i16, + neon_uimm3_bare, neon_uimm2_bare>; +defm : NeonI_DUP_Elt_pattern<DUPELT4h, v4i16, v8i16, v4i16, v8i16, + neon_uimm3_bare, neon_uimm2_bare>; +defm : NeonI_DUP_Elt_pattern<DUPELT4s, v4i32, v4i32, v2i32, v4i32, + neon_uimm2_bare, neon_uimm1_bare>; +defm : NeonI_DUP_Elt_pattern<DUPELT2s, v2i32, v4i32, v2i32, v4i32, + neon_uimm2_bare, neon_uimm1_bare>; +defm : NeonI_DUP_Elt_pattern<DUPELT2d, v2i64, v2i64, v1i64, v2i64, + neon_uimm1_bare, neon_uimm0_bare>; +defm : NeonI_DUP_Elt_pattern<DUPELT4s, v4f32, v4f32, v2f32, v4f32, + neon_uimm2_bare, neon_uimm1_bare>; +defm : NeonI_DUP_Elt_pattern<DUPELT2s, v2f32, v4f32, v2f32, v4f32, + neon_uimm2_bare, neon_uimm1_bare>; +defm : NeonI_DUP_Elt_pattern<DUPELT2d, v2f64, v2f64, v1f64, v2f64, + neon_uimm1_bare, neon_uimm0_bare>; + +def : Pat<(v2f32 (Neon_vdup (f32 FPR32:$Rn))), + (v2f32 (DUPELT2s + (SUBREG_TO_REG (i64 0), FPR32:$Rn, sub_32), + (i64 0)))>; +def : Pat<(v4f32 (Neon_vdup (f32 FPR32:$Rn))), + (v4f32 (DUPELT4s + (SUBREG_TO_REG (i64 0), FPR32:$Rn, sub_32), + (i64 0)))>; +def : Pat<(v2f64 (Neon_vdup (f64 FPR64:$Rn))), + (v2f64 (DUPELT2d + (SUBREG_TO_REG (i64 0), FPR64:$Rn, sub_64), + (i64 0)))>; + +class NeonI_DUP<bit Q, string asmop, string rdlane, + RegisterOperand ResVPR, ValueType ResTy, + RegisterClass OpGPR, ValueType OpTy> + : NeonI_copy<Q, 0b0, 0b0001, (outs ResVPR:$Rd), (ins OpGPR:$Rn), + asmop # "\t$Rd" # rdlane # ", $Rn", + [(set (ResTy ResVPR:$Rd), + (ResTy (Neon_vdup (OpTy OpGPR:$Rn))))], + NoItinerary>; + +def DUP16b : NeonI_DUP<0b1, "dup", ".16b", VPR128, v16i8, GPR32, i32> { + let Inst{20-16} = 0b00001; + // bits 17-20 are unspecified, but should be set to zero. +} + +def DUP8h : NeonI_DUP<0b1, "dup", ".8h", VPR128, v8i16, GPR32, i32> { + let Inst{20-16} = 0b00010; + // bits 18-20 are unspecified, but should be set to zero. +} + +def DUP4s : NeonI_DUP<0b1, "dup", ".4s", VPR128, v4i32, GPR32, i32> { + let Inst{20-16} = 0b00100; + // bits 19-20 are unspecified, but should be set to zero. +} + +def DUP2d : NeonI_DUP<0b1, "dup", ".2d", VPR128, v2i64, GPR64, i64> { + let Inst{20-16} = 0b01000; + // bit 20 is unspecified, but should be set to zero. +} + +def DUP8b : NeonI_DUP<0b0, "dup", ".8b", VPR64, v8i8, GPR32, i32> { + let Inst{20-16} = 0b00001; + // bits 17-20 are unspecified, but should be set to zero. +} + +def DUP4h : NeonI_DUP<0b0, "dup", ".4h", VPR64, v4i16, GPR32, i32> { + let Inst{20-16} = 0b00010; + // bits 18-20 are unspecified, but should be set to zero. +} + +def DUP2s : NeonI_DUP<0b0, "dup", ".2s", VPR64, v2i32, GPR32, i32> { + let Inst{20-16} = 0b00100; + // bits 19-20 are unspecified, but should be set to zero. +} + +// patterns for CONCAT_VECTORS +multiclass Concat_Vector_Pattern<ValueType ResTy, ValueType OpTy> { +def : Pat<(ResTy (concat_vectors (OpTy VPR64:$Rn), undef)), + (SUBREG_TO_REG (i64 0), VPR64:$Rn, sub_64)>; +def : Pat<(ResTy (concat_vectors (OpTy VPR64:$Rn), (OpTy VPR64:$Rm))), + (INSELd + (v2i64 (SUBREG_TO_REG (i64 0), VPR64:$Rn, sub_64)), + (v2i64 (SUBREG_TO_REG (i64 0), VPR64:$Rm, sub_64)), + (i64 1), + (i64 0))>; +def : Pat<(ResTy (concat_vectors (OpTy VPR64:$Rn), (OpTy VPR64:$Rn))), + (DUPELT2d + (v2i64 (SUBREG_TO_REG (i64 0), VPR64:$Rn, sub_64)), + (i64 0))> ; +} + +defm : Concat_Vector_Pattern<v16i8, v8i8>; +defm : Concat_Vector_Pattern<v8i16, v4i16>; +defm : Concat_Vector_Pattern<v4i32, v2i32>; +defm : Concat_Vector_Pattern<v2i64, v1i64>; +defm : Concat_Vector_Pattern<v4f32, v2f32>; +defm : Concat_Vector_Pattern<v2f64, v1f64>; + +//patterns for EXTRACT_SUBVECTOR +def : Pat<(v8i8 (extract_subvector (v16i8 VPR128:$Rn), (i64 0))), + (v8i8 (EXTRACT_SUBREG VPR128:$Rn, sub_64))>; +def : Pat<(v4i16 (extract_subvector (v8i16 VPR128:$Rn), (i64 0))), + (v4i16 (EXTRACT_SUBREG VPR128:$Rn, sub_64))>; +def : Pat<(v2i32 (extract_subvector (v4i32 VPR128:$Rn), (i64 0))), + (v2i32 (EXTRACT_SUBREG VPR128:$Rn, sub_64))>; +def : Pat<(v1i64 (extract_subvector (v2i64 VPR128:$Rn), (i64 0))), + (v1i64 (EXTRACT_SUBREG VPR128:$Rn, sub_64))>; +def : Pat<(v2f32 (extract_subvector (v4f32 VPR128:$Rn), (i64 0))), + (v2f32 (EXTRACT_SUBREG VPR128:$Rn, sub_64))>; +def : Pat<(v1f64 (extract_subvector (v2f64 VPR128:$Rn), (i64 0))), + (v1f64 (EXTRACT_SUBREG VPR128:$Rn, sub_64))>; + +// The followings are for instruction class (3V Elem) + +// Variant 1 + +class NI_2VE<bit q, bit u, bits<2> size, bits<4> opcode, + string asmop, string ResS, string OpS, string EleOpS, + Operand OpImm, RegisterOperand ResVPR, + RegisterOperand OpVPR, RegisterOperand EleOpVPR> + : NeonI_2VElem<q, u, size, opcode, + (outs ResVPR:$Rd), (ins ResVPR:$src, OpVPR:$Rn, + EleOpVPR:$Re, OpImm:$Index), + asmop # "\t$Rd." # ResS # ", $Rn." # OpS # + ", $Re." # EleOpS # "[$Index]", + [], + NoItinerary> { + bits<3> Index; + bits<5> Re; + + let Constraints = "$src = $Rd"; +} + +multiclass NI_2VE_v1<bit u, bits<4> opcode, string asmop> { + // vector register class for element is always 128-bit to cover the max index + def _2s4s : NI_2VE<0b0, u, 0b10, opcode, asmop, "2s", "2s", "s", + neon_uimm2_bare, VPR64, VPR64, VPR128> { + let Inst{11} = {Index{1}}; + let Inst{21} = {Index{0}}; + let Inst{20-16} = Re; + } + + def _4s4s : NI_2VE<0b1, u, 0b10, opcode, asmop, "4s", "4s", "s", + neon_uimm2_bare, VPR128, VPR128, VPR128> { + let Inst{11} = {Index{1}}; + let Inst{21} = {Index{0}}; + let Inst{20-16} = Re; + } + + // Index operations on 16-bit(H) elements are restricted to using v0-v15. + def _4h8h : NI_2VE<0b0, u, 0b01, opcode, asmop, "4h", "4h", "h", + neon_uimm3_bare, VPR64, VPR64, VPR128Lo> { + let Inst{11} = {Index{2}}; + let Inst{21} = {Index{1}}; + let Inst{20} = {Index{0}}; + let Inst{19-16} = Re{3-0}; + } + + def _8h8h : NI_2VE<0b1, u, 0b01, opcode, asmop, "8h", "8h", "h", + neon_uimm3_bare, VPR128, VPR128, VPR128Lo> { + let Inst{11} = {Index{2}}; + let Inst{21} = {Index{1}}; + let Inst{20} = {Index{0}}; + let Inst{19-16} = Re{3-0}; + } +} + +defm MLAvve : NI_2VE_v1<0b1, 0b0000, "mla">; +defm MLSvve : NI_2VE_v1<0b1, 0b0100, "mls">; + +// Pattern for lane in 128-bit vector +class NI_2VE_laneq<Instruction INST, Operand OpImm, SDPatternOperator op, + RegisterOperand ResVPR, RegisterOperand OpVPR, + RegisterOperand EleOpVPR, ValueType ResTy, ValueType OpTy, + ValueType EleOpTy> + : Pat<(ResTy (op (ResTy ResVPR:$src), (OpTy OpVPR:$Rn), + (OpTy (Neon_vduplane (EleOpTy EleOpVPR:$Re), (i64 OpImm:$Index))))), + (INST ResVPR:$src, OpVPR:$Rn, EleOpVPR:$Re, OpImm:$Index)>; + +// Pattern for lane in 64-bit vector +class NI_2VE_lane<Instruction INST, Operand OpImm, SDPatternOperator op, + RegisterOperand ResVPR, RegisterOperand OpVPR, + RegisterOperand EleOpVPR, ValueType ResTy, ValueType OpTy, + ValueType EleOpTy> + : Pat<(ResTy (op (ResTy ResVPR:$src), (OpTy OpVPR:$Rn), + (OpTy (Neon_vduplane (EleOpTy EleOpVPR:$Re), (i64 OpImm:$Index))))), + (INST ResVPR:$src, OpVPR:$Rn, + (SUBREG_TO_REG (i64 0), EleOpVPR:$Re, sub_64), OpImm:$Index)>; + +multiclass NI_2VE_v1_pat<string subop, SDPatternOperator op> +{ + def : NI_2VE_laneq<!cast<Instruction>(subop # "_2s4s"), neon_uimm2_bare, + op, VPR64, VPR64, VPR128, v2i32, v2i32, v4i32>; + + def : NI_2VE_laneq<!cast<Instruction>(subop # "_4s4s"), neon_uimm2_bare, + op, VPR128, VPR128, VPR128, v4i32, v4i32, v4i32>; + + def : NI_2VE_laneq<!cast<Instruction>(subop # "_4h8h"), neon_uimm3_bare, + op, VPR64, VPR64, VPR128Lo, v4i16, v4i16, v8i16>; + + def : NI_2VE_laneq<!cast<Instruction>(subop # "_8h8h"), neon_uimm3_bare, + op, VPR128, VPR128, VPR128Lo, v8i16, v8i16, v8i16>; + + // Index can only be half of the max value for lane in 64-bit vector + + def : NI_2VE_lane<!cast<Instruction>(subop # "_2s4s"), neon_uimm1_bare, + op, VPR64, VPR64, VPR64, v2i32, v2i32, v2i32>; + + def : NI_2VE_lane<!cast<Instruction>(subop # "_4h8h"), neon_uimm2_bare, + op, VPR64, VPR64, VPR64Lo, v4i16, v4i16, v4i16>; +} + +defm MLA_lane_v1 : NI_2VE_v1_pat<"MLAvve", Neon_mla>; +defm MLS_lane_v1 : NI_2VE_v1_pat<"MLSvve", Neon_mls>; + +class NI_2VE_2op<bit q, bit u, bits<2> size, bits<4> opcode, + string asmop, string ResS, string OpS, string EleOpS, + Operand OpImm, RegisterOperand ResVPR, + RegisterOperand OpVPR, RegisterOperand EleOpVPR> + : NeonI_2VElem<q, u, size, opcode, + (outs ResVPR:$Rd), (ins OpVPR:$Rn, + EleOpVPR:$Re, OpImm:$Index), + asmop # "\t$Rd." # ResS # ", $Rn." # OpS # + ", $Re." # EleOpS # "[$Index]", + [], + NoItinerary> { + bits<3> Index; + bits<5> Re; +} + +multiclass NI_2VE_v1_2op<bit u, bits<4> opcode, string asmop> { + // vector register class for element is always 128-bit to cover the max index + def _2s4s : NI_2VE_2op<0b0, u, 0b10, opcode, asmop, "2s", "2s", "s", + neon_uimm2_bare, VPR64, VPR64, VPR128> { + let Inst{11} = {Index{1}}; + let Inst{21} = {Index{0}}; + let Inst{20-16} = Re; + } + + def _4s4s : NI_2VE_2op<0b1, u, 0b10, opcode, asmop, "4s", "4s", "s", + neon_uimm2_bare, VPR128, VPR128, VPR128> { + let Inst{11} = {Index{1}}; + let Inst{21} = {Index{0}}; + let Inst{20-16} = Re; + } + + // Index operations on 16-bit(H) elements are restricted to using v0-v15. + def _4h8h : NI_2VE_2op<0b0, u, 0b01, opcode, asmop, "4h", "4h", "h", + neon_uimm3_bare, VPR64, VPR64, VPR128Lo> { + let Inst{11} = {Index{2}}; + let Inst{21} = {Index{1}}; + let Inst{20} = {Index{0}}; + let Inst{19-16} = Re{3-0}; + } + + def _8h8h : NI_2VE_2op<0b1, u, 0b01, opcode, asmop, "8h", "8h", "h", + neon_uimm3_bare, VPR128, VPR128, VPR128Lo> { + let Inst{11} = {Index{2}}; + let Inst{21} = {Index{1}}; + let Inst{20} = {Index{0}}; + let Inst{19-16} = Re{3-0}; + } +} + +defm MULve : NI_2VE_v1_2op<0b0, 0b1000, "mul">; +defm SQDMULHve : NI_2VE_v1_2op<0b0, 0b1100, "sqdmulh">; +defm SQRDMULHve : NI_2VE_v1_2op<0b0, 0b1101, "sqrdmulh">; + +// Pattern for lane in 128-bit vector +class NI_2VE_mul_laneq<Instruction INST, Operand OpImm, SDPatternOperator op, + RegisterOperand OpVPR, RegisterOperand EleOpVPR, + ValueType ResTy, ValueType OpTy, ValueType EleOpTy> + : Pat<(ResTy (op (OpTy OpVPR:$Rn), + (OpTy (Neon_vduplane (EleOpTy EleOpVPR:$Re), (i64 OpImm:$Index))))), + (INST OpVPR:$Rn, EleOpVPR:$Re, OpImm:$Index)>; + +// Pattern for lane in 64-bit vector +class NI_2VE_mul_lane<Instruction INST, Operand OpImm, SDPatternOperator op, + RegisterOperand OpVPR, RegisterOperand EleOpVPR, + ValueType ResTy, ValueType OpTy, ValueType EleOpTy> + : Pat<(ResTy (op (OpTy OpVPR:$Rn), + (OpTy (Neon_vduplane (EleOpTy EleOpVPR:$Re), (i64 OpImm:$Index))))), + (INST OpVPR:$Rn, + (SUBREG_TO_REG (i64 0), EleOpVPR:$Re, sub_64), OpImm:$Index)>; + +multiclass NI_2VE_mul_v1_pat<string subop, SDPatternOperator op> { + def : NI_2VE_mul_laneq<!cast<Instruction>(subop # "_2s4s"), neon_uimm2_bare, + op, VPR64, VPR128, v2i32, v2i32, v4i32>; + + def : NI_2VE_mul_laneq<!cast<Instruction>(subop # "_4s4s"), neon_uimm2_bare, + op, VPR128, VPR128, v4i32, v4i32, v4i32>; + + def : NI_2VE_mul_laneq<!cast<Instruction>(subop # "_4h8h"), neon_uimm3_bare, + op, VPR64, VPR128Lo, v4i16, v4i16, v8i16>; + + def : NI_2VE_mul_laneq<!cast<Instruction>(subop # "_8h8h"), neon_uimm3_bare, + op, VPR128, VPR128Lo, v8i16, v8i16, v8i16>; + + // Index can only be half of the max value for lane in 64-bit vector + + def : NI_2VE_mul_lane<!cast<Instruction>(subop # "_2s4s"), neon_uimm1_bare, + op, VPR64, VPR64, v2i32, v2i32, v2i32>; + + def : NI_2VE_mul_lane<!cast<Instruction>(subop # "_4h8h"), neon_uimm2_bare, + op, VPR64, VPR64Lo, v4i16, v4i16, v4i16>; +} + +defm MUL_lane_v1 : NI_2VE_mul_v1_pat<"MULve", mul>; +defm SQDMULH_lane_v1 : NI_2VE_mul_v1_pat<"SQDMULHve", int_arm_neon_vqdmulh>; +defm SQRDMULH_lane_v1 : NI_2VE_mul_v1_pat<"SQRDMULHve", int_arm_neon_vqrdmulh>; + +// Variant 2 + +multiclass NI_2VE_v2_2op<bit u, bits<4> opcode, string asmop> { + // vector register class for element is always 128-bit to cover the max index + def _2s4s : NI_2VE_2op<0b0, u, 0b10, opcode, asmop, "2s", "2s", "s", + neon_uimm2_bare, VPR64, VPR64, VPR128> { + let Inst{11} = {Index{1}}; + let Inst{21} = {Index{0}}; + let Inst{20-16} = Re; + } + + def _4s4s : NI_2VE_2op<0b1, u, 0b10, opcode, asmop, "4s", "4s", "s", + neon_uimm2_bare, VPR128, VPR128, VPR128> { + let Inst{11} = {Index{1}}; + let Inst{21} = {Index{0}}; + let Inst{20-16} = Re; + } + + // _1d2d doesn't exist! + + def _2d2d : NI_2VE_2op<0b1, u, 0b11, opcode, asmop, "2d", "2d", "d", + neon_uimm1_bare, VPR128, VPR128, VPR128> { + let Inst{11} = {Index{0}}; + let Inst{21} = 0b0; + let Inst{20-16} = Re; + } +} + +defm FMULve : NI_2VE_v2_2op<0b0, 0b1001, "fmul">; +defm FMULXve : NI_2VE_v2_2op<0b1, 0b1001, "fmulx">; + +class NI_2VE_mul_lane_2d<Instruction INST, Operand OpImm, SDPatternOperator op, + RegisterOperand OpVPR, RegisterOperand EleOpVPR, + ValueType ResTy, ValueType OpTy, ValueType EleOpTy, + SDPatternOperator coreop> + : Pat<(ResTy (op (OpTy OpVPR:$Rn), + (OpTy (coreop (EleOpTy EleOpVPR:$Re), (EleOpTy EleOpVPR:$Re))))), + (INST OpVPR:$Rn, + (SUBREG_TO_REG (i64 0), EleOpVPR:$Re, sub_64), 0)>; + +multiclass NI_2VE_mul_v2_pat<string subop, SDPatternOperator op> { + def : NI_2VE_mul_laneq<!cast<Instruction>(subop # "_2s4s"), neon_uimm2_bare, + op, VPR64, VPR128, v2f32, v2f32, v4f32>; + + def : NI_2VE_mul_laneq<!cast<Instruction>(subop # "_4s4s"), neon_uimm2_bare, + op, VPR128, VPR128, v4f32, v4f32, v4f32>; + + def : NI_2VE_mul_laneq<!cast<Instruction>(subop # "_2d2d"), neon_uimm1_bare, + op, VPR128, VPR128, v2f64, v2f64, v2f64>; + + // Index can only be half of the max value for lane in 64-bit vector + + def : NI_2VE_mul_lane<!cast<Instruction>(subop # "_2s4s"), neon_uimm1_bare, + op, VPR64, VPR64, v2f32, v2f32, v2f32>; + + def : NI_2VE_mul_lane_2d<!cast<Instruction>(subop # "_2d2d"), neon_uimm1_bare, + op, VPR128, VPR64, v2f64, v2f64, v1f64, + BinOpFrag<(Neon_combine_2d node:$LHS, node:$RHS)>>; +} + +defm FMUL_lane_v2 : NI_2VE_mul_v2_pat<"FMULve", fmul>; +defm FMULX_lane_v2 : NI_2VE_mul_v2_pat<"FMULXve", int_aarch64_neon_vmulx>; + +def : Pat<(v2f32 (fmul (v2f32 (Neon_vdup (f32 FPR32:$Re))), + (v2f32 VPR64:$Rn))), + (FMULve_2s4s VPR64:$Rn, (SUBREG_TO_REG (i32 0), $Re, sub_32), 0)>; + +def : Pat<(v4f32 (fmul (v4f32 (Neon_vdup (f32 FPR32:$Re))), + (v4f32 VPR128:$Rn))), + (FMULve_4s4s VPR128:$Rn, (SUBREG_TO_REG (i32 0), $Re, sub_32), 0)>; + +def : Pat<(v2f64 (fmul (v2f64 (Neon_vdup (f64 FPR64:$Re))), + (v2f64 VPR128:$Rn))), + (FMULve_2d2d VPR128:$Rn, (SUBREG_TO_REG (i64 0), $Re, sub_64), 0)>; + +// The followings are patterns using fma +// -ffp-contract=fast generates fma + +multiclass NI_2VE_v2<bit u, bits<4> opcode, string asmop> { + // vector register class for element is always 128-bit to cover the max index + def _2s4s : NI_2VE<0b0, u, 0b10, opcode, asmop, "2s", "2s", "s", + neon_uimm2_bare, VPR64, VPR64, VPR128> { + let Inst{11} = {Index{1}}; + let Inst{21} = {Index{0}}; + let Inst{20-16} = Re; + } + + def _4s4s : NI_2VE<0b1, u, 0b10, opcode, asmop, "4s", "4s", "s", + neon_uimm2_bare, VPR128, VPR128, VPR128> { + let Inst{11} = {Index{1}}; + let Inst{21} = {Index{0}}; + let Inst{20-16} = Re; + } + + // _1d2d doesn't exist! + + def _2d2d : NI_2VE<0b1, u, 0b11, opcode, asmop, "2d", "2d", "d", + neon_uimm1_bare, VPR128, VPR128, VPR128> { + let Inst{11} = {Index{0}}; + let Inst{21} = 0b0; + let Inst{20-16} = Re; + } +} + +defm FMLAvve : NI_2VE_v2<0b0, 0b0001, "fmla">; +defm FMLSvve : NI_2VE_v2<0b0, 0b0101, "fmls">; + +// Pattern for lane in 128-bit vector +class NI_2VEswap_laneq<Instruction INST, Operand OpImm, SDPatternOperator op, + RegisterOperand ResVPR, RegisterOperand OpVPR, + ValueType ResTy, ValueType OpTy, + SDPatternOperator coreop> + : Pat<(ResTy (op (ResTy (coreop (OpTy OpVPR:$Re), (i64 OpImm:$Index))), + (ResTy ResVPR:$src), (ResTy ResVPR:$Rn))), + (INST ResVPR:$src, ResVPR:$Rn, OpVPR:$Re, OpImm:$Index)>; + +// Pattern for lane 0 +class NI_2VEfma_lane0<Instruction INST, SDPatternOperator op, + RegisterOperand ResVPR, ValueType ResTy> + : Pat<(ResTy (op (ResTy ResVPR:$Rn), + (ResTy (Neon_vdup (f32 FPR32:$Re))), + (ResTy ResVPR:$src))), + (INST ResVPR:$src, ResVPR:$Rn, + (SUBREG_TO_REG (i32 0), $Re, sub_32), 0)>; + +// Pattern for lane in 64-bit vector +class NI_2VEswap_lane<Instruction INST, Operand OpImm, SDPatternOperator op, + RegisterOperand ResVPR, RegisterOperand OpVPR, + ValueType ResTy, ValueType OpTy, + SDPatternOperator coreop> + : Pat<(ResTy (op (ResTy (coreop (OpTy OpVPR:$Re), (i64 OpImm:$Index))), + (ResTy ResVPR:$Rn), (ResTy ResVPR:$src))), + (INST ResVPR:$src, ResVPR:$Rn, + (SUBREG_TO_REG (i64 0), OpVPR:$Re, sub_64), OpImm:$Index)>; + +// Pattern for lane in 64-bit vector +class NI_2VEswap_lane_2d2d<Instruction INST, Operand OpImm, + SDPatternOperator op, + RegisterOperand ResVPR, RegisterOperand OpVPR, + ValueType ResTy, ValueType OpTy, + SDPatternOperator coreop> + : Pat<(ResTy (op (ResTy (coreop (OpTy OpVPR:$Re), (OpTy OpVPR:$Re))), + (ResTy ResVPR:$Rn), (ResTy ResVPR:$src))), + (INST ResVPR:$src, ResVPR:$Rn, + (SUBREG_TO_REG (i64 0), OpVPR:$Re, sub_64), 0)>; + + +multiclass NI_2VE_fma_v2_pat<string subop, SDPatternOperator op> { + def : NI_2VEswap_laneq<!cast<Instruction>(subop # "_2s4s"), + neon_uimm2_bare, op, VPR64, VPR128, v2f32, v4f32, + BinOpFrag<(Neon_vduplane node:$LHS, node:$RHS)>>; + + def : NI_2VEfma_lane0<!cast<Instruction>(subop # "_2s4s"), + op, VPR64, v2f32>; + + def : NI_2VEswap_laneq<!cast<Instruction>(subop # "_4s4s"), + neon_uimm2_bare, op, VPR128, VPR128, v4f32, v4f32, + BinOpFrag<(Neon_vduplane node:$LHS, node:$RHS)>>; + + def : NI_2VEfma_lane0<!cast<Instruction>(subop # "_4s4s"), + op, VPR128, v4f32>; + + def : NI_2VEswap_laneq<!cast<Instruction>(subop # "_2d2d"), + neon_uimm1_bare, op, VPR128, VPR128, v2f64, v2f64, + BinOpFrag<(Neon_vduplane node:$LHS, node:$RHS)>>; + + // Index can only be half of the max value for lane in 64-bit vector + + def : NI_2VEswap_lane<!cast<Instruction>(subop # "_2s4s"), + neon_uimm1_bare, op, VPR64, VPR64, v2f32, v2f32, + BinOpFrag<(Neon_vduplane node:$LHS, node:$RHS)>>; + + def : NI_2VEswap_lane_2d2d<!cast<Instruction>(subop # "_2d2d"), + neon_uimm1_bare, op, VPR128, VPR64, v2f64, v1f64, + BinOpFrag<(Neon_combine_2d node:$LHS, node:$RHS)>>; +} + +defm FMLA_lane_v2_s : NI_2VE_fma_v2_pat<"FMLAvve", fma>; + +// Pattern for lane 0 +class NI_2VEfms_lane0<Instruction INST, SDPatternOperator op, + RegisterOperand ResVPR, ValueType ResTy> + : Pat<(ResTy (op (ResTy (fneg ResVPR:$Rn)), + (ResTy (Neon_vdup (f32 FPR32:$Re))), + (ResTy ResVPR:$src))), + (INST ResVPR:$src, ResVPR:$Rn, + (SUBREG_TO_REG (i32 0), $Re, sub_32), 0)>; + +multiclass NI_2VE_fms_v2_pat<string subop, SDPatternOperator op> +{ + def : NI_2VEswap_laneq<!cast<Instruction>(subop # "_2s4s"), + neon_uimm2_bare, op, VPR64, VPR128, v2f32, v4f32, + BinOpFrag<(fneg (Neon_vduplane node:$LHS, node:$RHS))>>; + + def : NI_2VEswap_laneq<!cast<Instruction>(subop # "_2s4s"), + neon_uimm2_bare, op, VPR64, VPR128, v2f32, v4f32, + BinOpFrag<(Neon_vduplane + (fneg node:$LHS), node:$RHS)>>; + + def : NI_2VEfms_lane0<!cast<Instruction>(subop # "_2s4s"), + op, VPR64, v2f32>; + + def : NI_2VEswap_laneq<!cast<Instruction>(subop # "_4s4s"), + neon_uimm2_bare, op, VPR128, VPR128, v4f32, v4f32, + BinOpFrag<(fneg (Neon_vduplane + node:$LHS, node:$RHS))>>; + + def : NI_2VEswap_laneq<!cast<Instruction>(subop # "_4s4s"), + neon_uimm2_bare, op, VPR128, VPR128, v4f32, v4f32, + BinOpFrag<(Neon_vduplane + (fneg node:$LHS), node:$RHS)>>; + + def : NI_2VEfms_lane0<!cast<Instruction>(subop # "_4s4s"), + op, VPR128, v4f32>; + + def : NI_2VEswap_laneq<!cast<Instruction>(subop # "_2d2d"), + neon_uimm1_bare, op, VPR128, VPR128, v2f64, v2f64, + BinOpFrag<(fneg (Neon_vduplane + node:$LHS, node:$RHS))>>; + + def : NI_2VEswap_laneq<!cast<Instruction>(subop # "_2d2d"), + neon_uimm1_bare, op, VPR128, VPR128, v2f64, v2f64, + BinOpFrag<(Neon_vduplane + (fneg node:$LHS), node:$RHS)>>; + + // Index can only be half of the max value for lane in 64-bit vector + + def : NI_2VEswap_lane<!cast<Instruction>(subop # "_2s4s"), + neon_uimm1_bare, op, VPR64, VPR64, v2f32, v2f32, + BinOpFrag<(fneg (Neon_vduplane + node:$LHS, node:$RHS))>>; + + def : NI_2VEswap_lane<!cast<Instruction>(subop # "_2s4s"), + neon_uimm1_bare, op, VPR64, VPR64, v2f32, v2f32, + BinOpFrag<(Neon_vduplane + (fneg node:$LHS), node:$RHS)>>; + + def : NI_2VEswap_lane<!cast<Instruction>(subop # "_4s4s"), + neon_uimm1_bare, op, VPR128, VPR64, v4f32, v2f32, + BinOpFrag<(fneg (Neon_vduplane node:$LHS, node:$RHS))>>; + + def : NI_2VEswap_lane<!cast<Instruction>(subop # "_4s4s"), + neon_uimm1_bare, op, VPR128, VPR64, v4f32, v2f32, + BinOpFrag<(Neon_vduplane (fneg node:$LHS), node:$RHS)>>; + + def : NI_2VEswap_lane_2d2d<!cast<Instruction>(subop # "_2d2d"), + neon_uimm1_bare, op, VPR128, VPR64, v2f64, v1f64, + BinOpFrag<(fneg (Neon_combine_2d + node:$LHS, node:$RHS))>>; + + def : NI_2VEswap_lane_2d2d<!cast<Instruction>(subop # "_2d2d"), + neon_uimm1_bare, op, VPR128, VPR64, v2f64, v1f64, + BinOpFrag<(Neon_combine_2d + (fneg node:$LHS), (fneg node:$RHS))>>; +} + +defm FMLS_lane_v2_s : NI_2VE_fms_v2_pat<"FMLSvve", fma>; + +// Variant 3: Long type +// E.g. SMLAL : 4S/4H/H (v0-v15), 2D/2S/S +// SMLAL2: 4S/8H/H (v0-v15), 2D/4S/S + +multiclass NI_2VE_v3<bit u, bits<4> opcode, string asmop> { + // vector register class for element is always 128-bit to cover the max index + def _2d2s : NI_2VE<0b0, u, 0b10, opcode, asmop, "2d", "2s", "s", + neon_uimm2_bare, VPR128, VPR64, VPR128> { + let Inst{11} = {Index{1}}; + let Inst{21} = {Index{0}}; + let Inst{20-16} = Re; + } + + def _2d4s : NI_2VE<0b1, u, 0b10, opcode, asmop # "2", "2d", "4s", "s", + neon_uimm2_bare, VPR128, VPR128, VPR128> { + let Inst{11} = {Index{1}}; + let Inst{21} = {Index{0}}; + let Inst{20-16} = Re; + } + + // Index operations on 16-bit(H) elements are restricted to using v0-v15. + def _4s8h : NI_2VE<0b1, u, 0b01, opcode, asmop # "2", "4s", "8h", "h", + neon_uimm3_bare, VPR128, VPR128, VPR128Lo> { + let Inst{11} = {Index{2}}; + let Inst{21} = {Index{1}}; + let Inst{20} = {Index{0}}; + let Inst{19-16} = Re{3-0}; + } + + def _4s4h : NI_2VE<0b0, u, 0b01, opcode, asmop, "4s", "4h", "h", + neon_uimm3_bare, VPR128, VPR64, VPR128Lo> { + let Inst{11} = {Index{2}}; + let Inst{21} = {Index{1}}; + let Inst{20} = {Index{0}}; + let Inst{19-16} = Re{3-0}; + } +} + +defm SMLALvve : NI_2VE_v3<0b0, 0b0010, "smlal">; +defm UMLALvve : NI_2VE_v3<0b1, 0b0010, "umlal">; +defm SMLSLvve : NI_2VE_v3<0b0, 0b0110, "smlsl">; +defm UMLSLvve : NI_2VE_v3<0b1, 0b0110, "umlsl">; +defm SQDMLALvve : NI_2VE_v3<0b0, 0b0011, "sqdmlal">; +defm SQDMLSLvve : NI_2VE_v3<0b0, 0b0111, "sqdmlsl">; + +multiclass NI_2VE_v3_2op<bit u, bits<4> opcode, string asmop> { + // vector register class for element is always 128-bit to cover the max index + def _2d2s : NI_2VE_2op<0b0, u, 0b10, opcode, asmop, "2d", "2s", "s", + neon_uimm2_bare, VPR128, VPR64, VPR128> { + let Inst{11} = {Index{1}}; + let Inst{21} = {Index{0}}; + let Inst{20-16} = Re; + } + + def _2d4s : NI_2VE_2op<0b1, u, 0b10, opcode, asmop # "2", "2d", "4s", "s", + neon_uimm2_bare, VPR128, VPR128, VPR128> { + let Inst{11} = {Index{1}}; + let Inst{21} = {Index{0}}; + let Inst{20-16} = Re; + } + + // Index operations on 16-bit(H) elements are restricted to using v0-v15. + def _4s8h : NI_2VE_2op<0b1, u, 0b01, opcode, asmop # "2", "4s", "8h", "h", + neon_uimm3_bare, VPR128, VPR128, VPR128Lo> { + let Inst{11} = {Index{2}}; + let Inst{21} = {Index{1}}; + let Inst{20} = {Index{0}}; + let Inst{19-16} = Re{3-0}; + } + + def _4s4h : NI_2VE_2op<0b0, u, 0b01, opcode, asmop, "4s", "4h", "h", + neon_uimm3_bare, VPR128, VPR64, VPR128Lo> { + let Inst{11} = {Index{2}}; + let Inst{21} = {Index{1}}; + let Inst{20} = {Index{0}}; + let Inst{19-16} = Re{3-0}; + } +} + +defm SMULLve : NI_2VE_v3_2op<0b0, 0b1010, "smull">; +defm UMULLve : NI_2VE_v3_2op<0b1, 0b1010, "umull">; +defm SQDMULLve : NI_2VE_v3_2op<0b0, 0b1011, "sqdmull">; + +def : Pat<(v1f64 (scalar_to_vector (f64 FPR64:$src))), + (FMOVdd $src)>; +def : Pat<(v1f32 (scalar_to_vector (f32 FPR32:$src))), + (FMOVss $src)>; + +// Pattern for lane in 128-bit vector +class NI_2VEL2_laneq<Instruction INST, Operand OpImm, SDPatternOperator op, + RegisterOperand EleOpVPR, ValueType ResTy, + ValueType OpTy, ValueType EleOpTy, ValueType HalfOpTy, + SDPatternOperator hiop> + : Pat<(ResTy (op (ResTy VPR128:$src), + (HalfOpTy (hiop (OpTy VPR128:$Rn))), + (HalfOpTy (Neon_vduplane + (EleOpTy EleOpVPR:$Re), (i64 OpImm:$Index))))), + (INST VPR128:$src, VPR128:$Rn, EleOpVPR:$Re, OpImm:$Index)>; + +// Pattern for lane in 64-bit vector +class NI_2VEL2_lane<Instruction INST, Operand OpImm, SDPatternOperator op, + RegisterOperand EleOpVPR, ValueType ResTy, + ValueType OpTy, ValueType EleOpTy, ValueType HalfOpTy, + SDPatternOperator hiop> + : Pat<(ResTy (op (ResTy VPR128:$src), + (HalfOpTy (hiop (OpTy VPR128:$Rn))), + (HalfOpTy (Neon_vduplane + (EleOpTy EleOpVPR:$Re), (i64 OpImm:$Index))))), + (INST VPR128:$src, VPR128:$Rn, + (SUBREG_TO_REG (i64 0), EleOpVPR:$Re, sub_64), OpImm:$Index)>; + +class NI_2VEL2_lane0<Instruction INST, SDPatternOperator op, + ValueType ResTy, ValueType OpTy, ValueType HalfOpTy, + SDPatternOperator hiop, Instruction DupInst> + : Pat<(ResTy (op (ResTy VPR128:$src), + (HalfOpTy (hiop (OpTy VPR128:$Rn))), + (HalfOpTy (Neon_vdup (i32 GPR32:$Re))))), + (INST VPR128:$src, VPR128:$Rn, (DupInst $Re), 0)>; + +multiclass NI_2VEL_v3_pat<string subop, SDPatternOperator op> { + def : NI_2VE_laneq<!cast<Instruction>(subop # "_4s4h"), neon_uimm3_bare, + op, VPR128, VPR64, VPR128Lo, v4i32, v4i16, v8i16>; + + def : NI_2VE_laneq<!cast<Instruction>(subop # "_2d2s"), neon_uimm2_bare, + op, VPR128, VPR64, VPR128, v2i64, v2i32, v4i32>; + + def : NI_2VEL2_laneq<!cast<Instruction>(subop # "_4s8h"), neon_uimm3_bare, + op, VPR128Lo, v4i32, v8i16, v8i16, v4i16, Neon_High8H>; + + def : NI_2VEL2_laneq<!cast<Instruction>(subop # "_2d4s"), neon_uimm2_bare, + op, VPR128, v2i64, v4i32, v4i32, v2i32, Neon_High4S>; + + def : NI_2VEL2_lane0<!cast<Instruction>(subop # "_4s8h"), + op, v4i32, v8i16, v4i16, Neon_High8H, DUP8h>; + + def : NI_2VEL2_lane0<!cast<Instruction>(subop # "_2d4s"), + op, v2i64, v4i32, v2i32, Neon_High4S, DUP4s>; + + // Index can only be half of the max value for lane in 64-bit vector + + def : NI_2VE_lane<!cast<Instruction>(subop # "_4s4h"), neon_uimm2_bare, + op, VPR128, VPR64, VPR64Lo, v4i32, v4i16, v4i16>; + + def : NI_2VE_lane<!cast<Instruction>(subop # "_2d2s"), neon_uimm1_bare, + op, VPR128, VPR64, VPR64, v2i64, v2i32, v2i32>; + + def : NI_2VEL2_lane<!cast<Instruction>(subop # "_4s8h"), neon_uimm2_bare, + op, VPR64Lo, v4i32, v8i16, v4i16, v4i16, Neon_High8H>; + + def : NI_2VEL2_lane<!cast<Instruction>(subop # "_2d4s"), neon_uimm1_bare, + op, VPR64, v2i64, v4i32, v2i32, v2i32, Neon_High4S>; +} + +defm SMLAL_lane_v3 : NI_2VEL_v3_pat<"SMLALvve", Neon_smlal>; +defm UMLAL_lane_v3 : NI_2VEL_v3_pat<"UMLALvve", Neon_umlal>; +defm SMLSL_lane_v3 : NI_2VEL_v3_pat<"SMLSLvve", Neon_smlsl>; +defm UMLSL_lane_v3 : NI_2VEL_v3_pat<"UMLSLvve", Neon_umlsl>; + +// Pattern for lane in 128-bit vector +class NI_2VEL2_mul_laneq<Instruction INST, Operand OpImm, SDPatternOperator op, + RegisterOperand EleOpVPR, ValueType ResTy, + ValueType OpTy, ValueType EleOpTy, ValueType HalfOpTy, + SDPatternOperator hiop> + : Pat<(ResTy (op + (HalfOpTy (hiop (OpTy VPR128:$Rn))), + (HalfOpTy (Neon_vduplane + (EleOpTy EleOpVPR:$Re), (i64 OpImm:$Index))))), + (INST VPR128:$Rn, EleOpVPR:$Re, OpImm:$Index)>; + +// Pattern for lane in 64-bit vector +class NI_2VEL2_mul_lane<Instruction INST, Operand OpImm, SDPatternOperator op, + RegisterOperand EleOpVPR, ValueType ResTy, + ValueType OpTy, ValueType EleOpTy, ValueType HalfOpTy, + SDPatternOperator hiop> + : Pat<(ResTy (op + (HalfOpTy (hiop (OpTy VPR128:$Rn))), + (HalfOpTy (Neon_vduplane + (EleOpTy EleOpVPR:$Re), (i64 OpImm:$Index))))), + (INST VPR128:$Rn, + (SUBREG_TO_REG (i64 0), EleOpVPR:$Re, sub_64), OpImm:$Index)>; + +// Pattern for fixed lane 0 +class NI_2VEL2_mul_lane0<Instruction INST, SDPatternOperator op, + ValueType ResTy, ValueType OpTy, ValueType HalfOpTy, + SDPatternOperator hiop, Instruction DupInst> + : Pat<(ResTy (op + (HalfOpTy (hiop (OpTy VPR128:$Rn))), + (HalfOpTy (Neon_vdup (i32 GPR32:$Re))))), + (INST VPR128:$Rn, (DupInst $Re), 0)>; + +multiclass NI_2VEL_mul_v3_pat<string subop, SDPatternOperator op> { + def : NI_2VE_mul_laneq<!cast<Instruction>(subop # "_4s4h"), neon_uimm3_bare, + op, VPR64, VPR128Lo, v4i32, v4i16, v8i16>; + + def : NI_2VE_mul_laneq<!cast<Instruction>(subop # "_2d2s"), neon_uimm2_bare, + op, VPR64, VPR128, v2i64, v2i32, v4i32>; + + def : NI_2VEL2_mul_laneq<!cast<Instruction>(subop # "_4s8h"), neon_uimm3_bare, + op, VPR128Lo, v4i32, v8i16, v8i16, v4i16, Neon_High8H>; + + def : NI_2VEL2_mul_laneq<!cast<Instruction>(subop # "_2d4s"), neon_uimm2_bare, + op, VPR128, v2i64, v4i32, v4i32, v2i32, Neon_High4S>; + + def : NI_2VEL2_mul_lane0<!cast<Instruction>(subop # "_4s8h"), + op, v4i32, v8i16, v4i16, Neon_High8H, DUP8h>; + + def : NI_2VEL2_mul_lane0<!cast<Instruction>(subop # "_2d4s"), + op, v2i64, v4i32, v2i32, Neon_High4S, DUP4s>; + + // Index can only be half of the max value for lane in 64-bit vector + + def : NI_2VE_mul_lane<!cast<Instruction>(subop # "_4s4h"), neon_uimm2_bare, + op, VPR64, VPR64Lo, v4i32, v4i16, v4i16>; + + def : NI_2VE_mul_lane<!cast<Instruction>(subop # "_2d2s"), neon_uimm1_bare, + op, VPR64, VPR64, v2i64, v2i32, v2i32>; + + def : NI_2VEL2_mul_lane<!cast<Instruction>(subop # "_4s8h"), neon_uimm2_bare, + op, VPR64Lo, v4i32, v8i16, v4i16, v4i16, Neon_High8H>; + + def : NI_2VEL2_mul_lane<!cast<Instruction>(subop # "_2d4s"), neon_uimm1_bare, + op, VPR64, v2i64, v4i32, v2i32, v2i32, Neon_High4S>; +} + +defm SMULL_lane_v3 : NI_2VEL_mul_v3_pat<"SMULLve", int_arm_neon_vmulls>; +defm UMULL_lane_v3 : NI_2VEL_mul_v3_pat<"UMULLve", int_arm_neon_vmullu>; +defm SQDMULL_lane_v3 : NI_2VEL_mul_v3_pat<"SQDMULLve", int_arm_neon_vqdmull>; + +multiclass NI_qdma<SDPatternOperator op> { + def _4s : PatFrag<(ops node:$Ra, node:$Rn, node:$Rm), + (op node:$Ra, + (v4i32 (int_arm_neon_vqdmull node:$Rn, node:$Rm)))>; + + def _2d : PatFrag<(ops node:$Ra, node:$Rn, node:$Rm), + (op node:$Ra, + (v2i64 (int_arm_neon_vqdmull node:$Rn, node:$Rm)))>; +} + +defm Neon_qdmlal : NI_qdma<int_arm_neon_vqadds>; +defm Neon_qdmlsl : NI_qdma<int_arm_neon_vqsubs>; + +multiclass NI_2VEL_v3_qdma_pat<string subop, string op> { + def : NI_2VE_laneq<!cast<Instruction>(subop # "_4s4h"), neon_uimm3_bare, + !cast<PatFrag>(op # "_4s"), VPR128, VPR64, VPR128Lo, + v4i32, v4i16, v8i16>; + + def : NI_2VE_laneq<!cast<Instruction>(subop # "_2d2s"), neon_uimm2_bare, + !cast<PatFrag>(op # "_2d"), VPR128, VPR64, VPR128, + v2i64, v2i32, v4i32>; + + def : NI_2VEL2_laneq<!cast<Instruction>(subop # "_4s8h"), neon_uimm3_bare, + !cast<PatFrag>(op # "_4s"), VPR128Lo, + v4i32, v8i16, v8i16, v4i16, Neon_High8H>; + + def : NI_2VEL2_laneq<!cast<Instruction>(subop # "_2d4s"), neon_uimm2_bare, + !cast<PatFrag>(op # "_2d"), VPR128, + v2i64, v4i32, v4i32, v2i32, Neon_High4S>; + + def : NI_2VEL2_lane0<!cast<Instruction>(subop # "_4s8h"), + !cast<PatFrag>(op # "_4s"), + v4i32, v8i16, v4i16, Neon_High8H, DUP8h>; + + def : NI_2VEL2_lane0<!cast<Instruction>(subop # "_2d4s"), + !cast<PatFrag>(op # "_2d"), + v2i64, v4i32, v2i32, Neon_High4S, DUP4s>; + + // Index can only be half of the max value for lane in 64-bit vector + + def : NI_2VE_lane<!cast<Instruction>(subop # "_4s4h"), neon_uimm2_bare, + !cast<PatFrag>(op # "_4s"), VPR128, VPR64, VPR64Lo, + v4i32, v4i16, v4i16>; + + def : NI_2VE_lane<!cast<Instruction>(subop # "_2d2s"), neon_uimm1_bare, + !cast<PatFrag>(op # "_2d"), VPR128, VPR64, VPR64, + v2i64, v2i32, v2i32>; + + def : NI_2VEL2_lane<!cast<Instruction>(subop # "_4s8h"), neon_uimm2_bare, + !cast<PatFrag>(op # "_4s"), VPR64Lo, + v4i32, v8i16, v4i16, v4i16, Neon_High8H>; + + def : NI_2VEL2_lane<!cast<Instruction>(subop # "_2d4s"), neon_uimm1_bare, + !cast<PatFrag>(op # "_2d"), VPR64, + v2i64, v4i32, v2i32, v2i32, Neon_High4S>; +} + +defm SQDMLAL_lane_v3 : NI_2VEL_v3_qdma_pat<"SQDMLALvve", "Neon_qdmlal">; +defm SQDMLSL_lane_v3 : NI_2VEL_v3_qdma_pat<"SQDMLSLvve", "Neon_qdmlsl">; + +// End of implementation for instruction class (3V Elem) + +class NeonI_REV<string asmop, string Res, bits<2> size, bit Q, bit U, + bits<5> opcode, RegisterOperand ResVPR, ValueType ResTy, + SDPatternOperator Neon_Rev> + : NeonI_2VMisc<Q, U, size, opcode, + (outs ResVPR:$Rd), (ins ResVPR:$Rn), + asmop # "\t$Rd." # Res # ", $Rn." # Res, + [(set (ResTy ResVPR:$Rd), + (ResTy (Neon_Rev (ResTy ResVPR:$Rn))))], + NoItinerary> ; + +def REV64_16b : NeonI_REV<"rev64", "16b", 0b00, 0b1, 0b0, 0b00000, VPR128, + v16i8, Neon_rev64>; +def REV64_8h : NeonI_REV<"rev64", "8h", 0b01, 0b1, 0b0, 0b00000, VPR128, + v8i16, Neon_rev64>; +def REV64_4s : NeonI_REV<"rev64", "4s", 0b10, 0b1, 0b0, 0b00000, VPR128, + v4i32, Neon_rev64>; +def REV64_8b : NeonI_REV<"rev64", "8b", 0b00, 0b0, 0b0, 0b00000, VPR64, + v8i8, Neon_rev64>; +def REV64_4h : NeonI_REV<"rev64", "4h", 0b01, 0b0, 0b0, 0b00000, VPR64, + v4i16, Neon_rev64>; +def REV64_2s : NeonI_REV<"rev64", "2s", 0b10, 0b0, 0b0, 0b00000, VPR64, + v2i32, Neon_rev64>; + +def : Pat<(v4f32 (Neon_rev64 (v4f32 VPR128:$Rn))), (REV64_4s VPR128:$Rn)>; +def : Pat<(v2f32 (Neon_rev64 (v2f32 VPR64:$Rn))), (REV64_2s VPR64:$Rn)>; + +def REV32_16b : NeonI_REV<"rev32", "16b", 0b00, 0b1, 0b1, 0b00000, VPR128, + v16i8, Neon_rev32>; +def REV32_8h : NeonI_REV<"rev32", "8h", 0b01, 0b1, 0b1, 0b00000, VPR128, + v8i16, Neon_rev32>; +def REV32_8b : NeonI_REV<"rev32", "8b", 0b00, 0b0, 0b1, 0b00000, VPR64, + v8i8, Neon_rev32>; +def REV32_4h : NeonI_REV<"rev32", "4h", 0b01, 0b0, 0b1, 0b00000, VPR64, + v4i16, Neon_rev32>; + +def REV16_16b : NeonI_REV<"rev16", "16b", 0b00, 0b1, 0b0, 0b00001, VPR128, + v16i8, Neon_rev16>; +def REV16_8b : NeonI_REV<"rev16", "8b", 0b00, 0b0, 0b0, 0b00001, VPR64, + v8i8, Neon_rev16>; + +multiclass NeonI_PairwiseAdd<string asmop, bit U, bits<5> opcode, + SDPatternOperator Neon_Padd> { + def 16b8h : NeonI_2VMisc<0b1, U, 0b00, opcode, + (outs VPR128:$Rd), (ins VPR128:$Rn), + asmop # "\t$Rd.8h, $Rn.16b", + [(set (v8i16 VPR128:$Rd), + (v8i16 (Neon_Padd (v16i8 VPR128:$Rn))))], + NoItinerary>; + + def 8b4h : NeonI_2VMisc<0b0, U, 0b00, opcode, + (outs VPR64:$Rd), (ins VPR64:$Rn), + asmop # "\t$Rd.4h, $Rn.8b", + [(set (v4i16 VPR64:$Rd), + (v4i16 (Neon_Padd (v8i8 VPR64:$Rn))))], + NoItinerary>; + + def 8h4s : NeonI_2VMisc<0b1, U, 0b01, opcode, + (outs VPR128:$Rd), (ins VPR128:$Rn), + asmop # "\t$Rd.4s, $Rn.8h", + [(set (v4i32 VPR128:$Rd), + (v4i32 (Neon_Padd (v8i16 VPR128:$Rn))))], + NoItinerary>; + + def 4h2s : NeonI_2VMisc<0b0, U, 0b01, opcode, + (outs VPR64:$Rd), (ins VPR64:$Rn), + asmop # "\t$Rd.2s, $Rn.4h", + [(set (v2i32 VPR64:$Rd), + (v2i32 (Neon_Padd (v4i16 VPR64:$Rn))))], + NoItinerary>; + + def 4s2d : NeonI_2VMisc<0b1, U, 0b10, opcode, + (outs VPR128:$Rd), (ins VPR128:$Rn), + asmop # "\t$Rd.2d, $Rn.4s", + [(set (v2i64 VPR128:$Rd), + (v2i64 (Neon_Padd (v4i32 VPR128:$Rn))))], + NoItinerary>; + + def 2s1d : NeonI_2VMisc<0b0, U, 0b10, opcode, + (outs VPR64:$Rd), (ins VPR64:$Rn), + asmop # "\t$Rd.1d, $Rn.2s", + [(set (v1i64 VPR64:$Rd), + (v1i64 (Neon_Padd (v2i32 VPR64:$Rn))))], + NoItinerary>; +} + +defm SADDLP : NeonI_PairwiseAdd<"saddlp", 0b0, 0b00010, + int_arm_neon_vpaddls>; +defm UADDLP : NeonI_PairwiseAdd<"uaddlp", 0b1, 0b00010, + int_arm_neon_vpaddlu>; + +multiclass NeonI_PairwiseAddAcc<string asmop, bit U, bits<5> opcode, + SDPatternOperator Neon_Padd> { + let Constraints = "$src = $Rd" in { + def 16b8h : NeonI_2VMisc<0b1, U, 0b00, opcode, + (outs VPR128:$Rd), (ins VPR128:$src, VPR128:$Rn), + asmop # "\t$Rd.8h, $Rn.16b", + [(set (v8i16 VPR128:$Rd), + (v8i16 (Neon_Padd + (v8i16 VPR128:$src), (v16i8 VPR128:$Rn))))], + NoItinerary>; + + def 8b4h : NeonI_2VMisc<0b0, U, 0b00, opcode, + (outs VPR64:$Rd), (ins VPR64:$src, VPR64:$Rn), + asmop # "\t$Rd.4h, $Rn.8b", + [(set (v4i16 VPR64:$Rd), + (v4i16 (Neon_Padd + (v4i16 VPR64:$src), (v8i8 VPR64:$Rn))))], + NoItinerary>; + + def 8h4s : NeonI_2VMisc<0b1, U, 0b01, opcode, + (outs VPR128:$Rd), (ins VPR128:$src, VPR128:$Rn), + asmop # "\t$Rd.4s, $Rn.8h", + [(set (v4i32 VPR128:$Rd), + (v4i32 (Neon_Padd + (v4i32 VPR128:$src), (v8i16 VPR128:$Rn))))], + NoItinerary>; + + def 4h2s : NeonI_2VMisc<0b0, U, 0b01, opcode, + (outs VPR64:$Rd), (ins VPR64:$src, VPR64:$Rn), + asmop # "\t$Rd.2s, $Rn.4h", + [(set (v2i32 VPR64:$Rd), + (v2i32 (Neon_Padd + (v2i32 VPR64:$src), (v4i16 VPR64:$Rn))))], + NoItinerary>; + + def 4s2d : NeonI_2VMisc<0b1, U, 0b10, opcode, + (outs VPR128:$Rd), (ins VPR128:$src, VPR128:$Rn), + asmop # "\t$Rd.2d, $Rn.4s", + [(set (v2i64 VPR128:$Rd), + (v2i64 (Neon_Padd + (v2i64 VPR128:$src), (v4i32 VPR128:$Rn))))], + NoItinerary>; + + def 2s1d : NeonI_2VMisc<0b0, U, 0b10, opcode, + (outs VPR64:$Rd), (ins VPR64:$src, VPR64:$Rn), + asmop # "\t$Rd.1d, $Rn.2s", + [(set (v1i64 VPR64:$Rd), + (v1i64 (Neon_Padd + (v1i64 VPR64:$src), (v2i32 VPR64:$Rn))))], + NoItinerary>; + } +} + +defm SADALP : NeonI_PairwiseAddAcc<"sadalp", 0b0, 0b00110, + int_arm_neon_vpadals>; +defm UADALP : NeonI_PairwiseAddAcc<"uadalp", 0b1, 0b00110, + int_arm_neon_vpadalu>; + +multiclass NeonI_2VMisc_BHSDsize_1Arg<string asmop, bit U, bits<5> opcode> { + def 16b : NeonI_2VMisc<0b1, U, 0b00, opcode, + (outs VPR128:$Rd), (ins VPR128:$Rn), + asmop # "\t$Rd.16b, $Rn.16b", + [], NoItinerary>; + + def 8h : NeonI_2VMisc<0b1, U, 0b01, opcode, + (outs VPR128:$Rd), (ins VPR128:$Rn), + asmop # "\t$Rd.8h, $Rn.8h", + [], NoItinerary>; + + def 4s : NeonI_2VMisc<0b1, U, 0b10, opcode, + (outs VPR128:$Rd), (ins VPR128:$Rn), + asmop # "\t$Rd.4s, $Rn.4s", + [], NoItinerary>; + + def 2d : NeonI_2VMisc<0b1, U, 0b11, opcode, + (outs VPR128:$Rd), (ins VPR128:$Rn), + asmop # "\t$Rd.2d, $Rn.2d", + [], NoItinerary>; + + def 8b : NeonI_2VMisc<0b0, U, 0b00, opcode, + (outs VPR64:$Rd), (ins VPR64:$Rn), + asmop # "\t$Rd.8b, $Rn.8b", + [], NoItinerary>; + + def 4h : NeonI_2VMisc<0b0, U, 0b01, opcode, + (outs VPR64:$Rd), (ins VPR64:$Rn), + asmop # "\t$Rd.4h, $Rn.4h", + [], NoItinerary>; + + def 2s : NeonI_2VMisc<0b0, U, 0b10, opcode, + (outs VPR64:$Rd), (ins VPR64:$Rn), + asmop # "\t$Rd.2s, $Rn.2s", + [], NoItinerary>; +} + +defm SQABS : NeonI_2VMisc_BHSDsize_1Arg<"sqabs", 0b0, 0b00111>; +defm SQNEG : NeonI_2VMisc_BHSDsize_1Arg<"sqneg", 0b1, 0b00111>; +defm ABS : NeonI_2VMisc_BHSDsize_1Arg<"abs", 0b0, 0b01011>; +defm NEG : NeonI_2VMisc_BHSDsize_1Arg<"neg", 0b1, 0b01011>; + +multiclass NeonI_2VMisc_BHSD_1Arg_Pattern<string Prefix, + SDPatternOperator Neon_Op> { + def : Pat<(v16i8 (Neon_Op (v16i8 VPR128:$Rn))), + (v16i8 (!cast<Instruction>(Prefix # 16b) (v16i8 VPR128:$Rn)))>; + + def : Pat<(v8i16 (Neon_Op (v8i16 VPR128:$Rn))), + (v8i16 (!cast<Instruction>(Prefix # 8h) (v8i16 VPR128:$Rn)))>; + + def : Pat<(v4i32 (Neon_Op (v4i32 VPR128:$Rn))), + (v4i32 (!cast<Instruction>(Prefix # 4s) (v4i32 VPR128:$Rn)))>; + + def : Pat<(v2i64 (Neon_Op (v2i64 VPR128:$Rn))), + (v2i64 (!cast<Instruction>(Prefix # 2d) (v2i64 VPR128:$Rn)))>; + + def : Pat<(v8i8 (Neon_Op (v8i8 VPR64:$Rn))), + (v8i8 (!cast<Instruction>(Prefix # 8b) (v8i8 VPR64:$Rn)))>; + + def : Pat<(v4i16 (Neon_Op (v4i16 VPR64:$Rn))), + (v4i16 (!cast<Instruction>(Prefix # 4h) (v4i16 VPR64:$Rn)))>; + + def : Pat<(v2i32 (Neon_Op (v2i32 VPR64:$Rn))), + (v2i32 (!cast<Instruction>(Prefix # 2s) (v2i32 VPR64:$Rn)))>; +} + +defm : NeonI_2VMisc_BHSD_1Arg_Pattern<"SQABS", int_arm_neon_vqabs>; +defm : NeonI_2VMisc_BHSD_1Arg_Pattern<"SQNEG", int_arm_neon_vqneg>; +defm : NeonI_2VMisc_BHSD_1Arg_Pattern<"ABS", int_arm_neon_vabs>; + +def : Pat<(v16i8 (sub + (v16i8 Neon_AllZero), + (v16i8 VPR128:$Rn))), + (v16i8 (NEG16b (v16i8 VPR128:$Rn)))>; +def : Pat<(v8i8 (sub + (v8i8 Neon_AllZero), + (v8i8 VPR64:$Rn))), + (v8i8 (NEG8b (v8i8 VPR64:$Rn)))>; +def : Pat<(v8i16 (sub + (v8i16 (bitconvert (v16i8 Neon_AllZero))), + (v8i16 VPR128:$Rn))), + (v8i16 (NEG8h (v8i16 VPR128:$Rn)))>; +def : Pat<(v4i16 (sub + (v4i16 (bitconvert (v8i8 Neon_AllZero))), + (v4i16 VPR64:$Rn))), + (v4i16 (NEG4h (v4i16 VPR64:$Rn)))>; +def : Pat<(v4i32 (sub + (v4i32 (bitconvert (v16i8 Neon_AllZero))), + (v4i32 VPR128:$Rn))), + (v4i32 (NEG4s (v4i32 VPR128:$Rn)))>; +def : Pat<(v2i32 (sub + (v2i32 (bitconvert (v8i8 Neon_AllZero))), + (v2i32 VPR64:$Rn))), + (v2i32 (NEG2s (v2i32 VPR64:$Rn)))>; +def : Pat<(v2i64 (sub + (v2i64 (bitconvert (v16i8 Neon_AllZero))), + (v2i64 VPR128:$Rn))), + (v2i64 (NEG2d (v2i64 VPR128:$Rn)))>; + +multiclass NeonI_2VMisc_BHSDsize_2Args<string asmop, bit U, bits<5> opcode> { + let Constraints = "$src = $Rd" in { + def 16b : NeonI_2VMisc<0b1, U, 0b00, opcode, + (outs VPR128:$Rd), (ins VPR128:$src, VPR128:$Rn), + asmop # "\t$Rd.16b, $Rn.16b", + [], NoItinerary>; + + def 8h : NeonI_2VMisc<0b1, U, 0b01, opcode, + (outs VPR128:$Rd), (ins VPR128:$src, VPR128:$Rn), + asmop # "\t$Rd.8h, $Rn.8h", + [], NoItinerary>; + + def 4s : NeonI_2VMisc<0b1, U, 0b10, opcode, + (outs VPR128:$Rd), (ins VPR128:$src, VPR128:$Rn), + asmop # "\t$Rd.4s, $Rn.4s", + [], NoItinerary>; + + def 2d : NeonI_2VMisc<0b1, U, 0b11, opcode, + (outs VPR128:$Rd), (ins VPR128:$src, VPR128:$Rn), + asmop # "\t$Rd.2d, $Rn.2d", + [], NoItinerary>; + + def 8b : NeonI_2VMisc<0b0, U, 0b00, opcode, + (outs VPR64:$Rd), (ins VPR64:$src, VPR64:$Rn), + asmop # "\t$Rd.8b, $Rn.8b", + [], NoItinerary>; + + def 4h : NeonI_2VMisc<0b0, U, 0b01, opcode, + (outs VPR64:$Rd), (ins VPR64:$src, VPR64:$Rn), + asmop # "\t$Rd.4h, $Rn.4h", + [], NoItinerary>; + + def 2s : NeonI_2VMisc<0b0, U, 0b10, opcode, + (outs VPR64:$Rd), (ins VPR64:$src, VPR64:$Rn), + asmop # "\t$Rd.2s, $Rn.2s", + [], NoItinerary>; + } +} + +defm SUQADD : NeonI_2VMisc_BHSDsize_2Args<"suqadd", 0b0, 0b00011>; +defm USQADD : NeonI_2VMisc_BHSDsize_2Args<"usqadd", 0b1, 0b00011>; + +multiclass NeonI_2VMisc_BHSD_2Args_Pattern<string Prefix, + SDPatternOperator Neon_Op> { + def : Pat<(v16i8 (Neon_Op (v16i8 VPR128:$src), (v16i8 VPR128:$Rn))), + (v16i8 (!cast<Instruction>(Prefix # 16b) + (v16i8 VPR128:$src), (v16i8 VPR128:$Rn)))>; + + def : Pat<(v8i16 (Neon_Op (v8i16 VPR128:$src), (v8i16 VPR128:$Rn))), + (v8i16 (!cast<Instruction>(Prefix # 8h) + (v8i16 VPR128:$src), (v8i16 VPR128:$Rn)))>; + + def : Pat<(v4i32 (Neon_Op (v4i32 VPR128:$src), (v4i32 VPR128:$Rn))), + (v4i32 (!cast<Instruction>(Prefix # 4s) + (v4i32 VPR128:$src), (v4i32 VPR128:$Rn)))>; + + def : Pat<(v2i64 (Neon_Op (v2i64 VPR128:$src), (v2i64 VPR128:$Rn))), + (v2i64 (!cast<Instruction>(Prefix # 2d) + (v2i64 VPR128:$src), (v2i64 VPR128:$Rn)))>; + + def : Pat<(v8i8 (Neon_Op (v8i8 VPR64:$src), (v8i8 VPR64:$Rn))), + (v8i8 (!cast<Instruction>(Prefix # 8b) + (v8i8 VPR64:$src), (v8i8 VPR64:$Rn)))>; + + def : Pat<(v4i16 (Neon_Op (v4i16 VPR64:$src), (v4i16 VPR64:$Rn))), + (v4i16 (!cast<Instruction>(Prefix # 4h) + (v4i16 VPR64:$src), (v4i16 VPR64:$Rn)))>; + + def : Pat<(v2i32 (Neon_Op (v2i32 VPR64:$src), (v2i32 VPR64:$Rn))), + (v2i32 (!cast<Instruction>(Prefix # 2s) + (v2i32 VPR64:$src), (v2i32 VPR64:$Rn)))>; +} + +defm : NeonI_2VMisc_BHSD_2Args_Pattern<"SUQADD", int_aarch64_neon_suqadd>; +defm : NeonI_2VMisc_BHSD_2Args_Pattern<"USQADD", int_aarch64_neon_usqadd>; + +multiclass NeonI_2VMisc_BHSsizes<string asmop, bit U, + SDPatternOperator Neon_Op> { + def 16b : NeonI_2VMisc<0b1, U, 0b00, 0b00100, + (outs VPR128:$Rd), (ins VPR128:$Rn), + asmop # "\t$Rd.16b, $Rn.16b", + [(set (v16i8 VPR128:$Rd), + (v16i8 (Neon_Op (v16i8 VPR128:$Rn))))], + NoItinerary>; + + def 8h : NeonI_2VMisc<0b1, U, 0b01, 0b00100, + (outs VPR128:$Rd), (ins VPR128:$Rn), + asmop # "\t$Rd.8h, $Rn.8h", + [(set (v8i16 VPR128:$Rd), + (v8i16 (Neon_Op (v8i16 VPR128:$Rn))))], + NoItinerary>; + + def 4s : NeonI_2VMisc<0b1, U, 0b10, 0b00100, + (outs VPR128:$Rd), (ins VPR128:$Rn), + asmop # "\t$Rd.4s, $Rn.4s", + [(set (v4i32 VPR128:$Rd), + (v4i32 (Neon_Op (v4i32 VPR128:$Rn))))], + NoItinerary>; + + def 8b : NeonI_2VMisc<0b0, U, 0b00, 0b00100, + (outs VPR64:$Rd), (ins VPR64:$Rn), + asmop # "\t$Rd.8b, $Rn.8b", + [(set (v8i8 VPR64:$Rd), + (v8i8 (Neon_Op (v8i8 VPR64:$Rn))))], + NoItinerary>; + + def 4h : NeonI_2VMisc<0b0, U, 0b01, 0b00100, + (outs VPR64:$Rd), (ins VPR64:$Rn), + asmop # "\t$Rd.4h, $Rn.4h", + [(set (v4i16 VPR64:$Rd), + (v4i16 (Neon_Op (v4i16 VPR64:$Rn))))], + NoItinerary>; + + def 2s : NeonI_2VMisc<0b0, U, 0b10, 0b00100, + (outs VPR64:$Rd), (ins VPR64:$Rn), + asmop # "\t$Rd.2s, $Rn.2s", + [(set (v2i32 VPR64:$Rd), + (v2i32 (Neon_Op (v2i32 VPR64:$Rn))))], + NoItinerary>; +} + +defm CLS : NeonI_2VMisc_BHSsizes<"cls", 0b0, int_arm_neon_vcls>; +defm CLZ : NeonI_2VMisc_BHSsizes<"clz", 0b1, ctlz>; + +multiclass NeonI_2VMisc_Bsize<string asmop, bit U, bits<2> size, + bits<5> Opcode> { + def 16b : NeonI_2VMisc<0b1, U, size, Opcode, + (outs VPR128:$Rd), (ins VPR128:$Rn), + asmop # "\t$Rd.16b, $Rn.16b", + [], NoItinerary>; + + def 8b : NeonI_2VMisc<0b0, U, size, Opcode, + (outs VPR64:$Rd), (ins VPR64:$Rn), + asmop # "\t$Rd.8b, $Rn.8b", + [], NoItinerary>; +} + +defm CNT : NeonI_2VMisc_Bsize<"cnt", 0b0, 0b00, 0b00101>; +defm NOT : NeonI_2VMisc_Bsize<"not", 0b1, 0b00, 0b00101>; +defm RBIT : NeonI_2VMisc_Bsize<"rbit", 0b1, 0b01, 0b00101>; + +def : NeonInstAlias<"mvn $Rd.16b, $Rn.16b", + (NOT16b VPR128:$Rd, VPR128:$Rn), 0>; +def : NeonInstAlias<"mvn $Rd.8b, $Rn.8b", + (NOT8b VPR64:$Rd, VPR64:$Rn), 0>; + +def : Pat<(v16i8 (ctpop (v16i8 VPR128:$Rn))), + (v16i8 (CNT16b (v16i8 VPR128:$Rn)))>; +def : Pat<(v8i8 (ctpop (v8i8 VPR64:$Rn))), + (v8i8 (CNT8b (v8i8 VPR64:$Rn)))>; + +def : Pat<(v16i8 (xor + (v16i8 VPR128:$Rn), + (v16i8 Neon_AllOne))), + (v16i8 (NOT16b (v16i8 VPR128:$Rn)))>; +def : Pat<(v8i8 (xor + (v8i8 VPR64:$Rn), + (v8i8 Neon_AllOne))), + (v8i8 (NOT8b (v8i8 VPR64:$Rn)))>; +def : Pat<(v8i16 (xor + (v8i16 VPR128:$Rn), + (v8i16 (bitconvert (v16i8 Neon_AllOne))))), + (NOT16b VPR128:$Rn)>; +def : Pat<(v4i16 (xor + (v4i16 VPR64:$Rn), + (v4i16 (bitconvert (v8i8 Neon_AllOne))))), + (NOT8b VPR64:$Rn)>; +def : Pat<(v4i32 (xor + (v4i32 VPR128:$Rn), + (v4i32 (bitconvert (v16i8 Neon_AllOne))))), + (NOT16b VPR128:$Rn)>; +def : Pat<(v2i32 (xor + (v2i32 VPR64:$Rn), + (v2i32 (bitconvert (v8i8 Neon_AllOne))))), + (NOT8b VPR64:$Rn)>; +def : Pat<(v2i64 (xor + (v2i64 VPR128:$Rn), + (v2i64 (bitconvert (v16i8 Neon_AllOne))))), + (NOT16b VPR128:$Rn)>; + +def : Pat<(v16i8 (int_aarch64_neon_rbit (v16i8 VPR128:$Rn))), + (v16i8 (RBIT16b (v16i8 VPR128:$Rn)))>; +def : Pat<(v8i8 (int_aarch64_neon_rbit (v8i8 VPR64:$Rn))), + (v8i8 (RBIT8b (v8i8 VPR64:$Rn)))>; + +multiclass NeonI_2VMisc_SDsizes<string asmop, bit U, bits<5> opcode, + SDPatternOperator Neon_Op> { + def 4s : NeonI_2VMisc<0b1, U, 0b10, opcode, + (outs VPR128:$Rd), (ins VPR128:$Rn), + asmop # "\t$Rd.4s, $Rn.4s", + [(set (v4f32 VPR128:$Rd), + (v4f32 (Neon_Op (v4f32 VPR128:$Rn))))], + NoItinerary>; + + def 2d : NeonI_2VMisc<0b1, U, 0b11, opcode, + (outs VPR128:$Rd), (ins VPR128:$Rn), + asmop # "\t$Rd.2d, $Rn.2d", + [(set (v2f64 VPR128:$Rd), + (v2f64 (Neon_Op (v2f64 VPR128:$Rn))))], + NoItinerary>; + + def 2s : NeonI_2VMisc<0b0, U, 0b10, opcode, + (outs VPR64:$Rd), (ins VPR64:$Rn), + asmop # "\t$Rd.2s, $Rn.2s", + [(set (v2f32 VPR64:$Rd), + (v2f32 (Neon_Op (v2f32 VPR64:$Rn))))], + NoItinerary>; +} + +defm FABS : NeonI_2VMisc_SDsizes<"fabs", 0b0, 0b01111, fabs>; +defm FNEG : NeonI_2VMisc_SDsizes<"fneg", 0b1, 0b01111, fneg>; + +multiclass NeonI_2VMisc_HSD_Narrow<string asmop, bit U, bits<5> opcode> { + def 8h8b : NeonI_2VMisc<0b0, U, 0b00, opcode, + (outs VPR64:$Rd), (ins VPR128:$Rn), + asmop # "\t$Rd.8b, $Rn.8h", + [], NoItinerary>; + + def 4s4h : NeonI_2VMisc<0b0, U, 0b01, opcode, + (outs VPR64:$Rd), (ins VPR128:$Rn), + asmop # "\t$Rd.4h, $Rn.4s", + [], NoItinerary>; + + def 2d2s : NeonI_2VMisc<0b0, U, 0b10, opcode, + (outs VPR64:$Rd), (ins VPR128:$Rn), + asmop # "\t$Rd.2s, $Rn.2d", + [], NoItinerary>; + + let Constraints = "$Rd = $src" in { + def 8h16b : NeonI_2VMisc<0b1, U, 0b00, opcode, + (outs VPR128:$Rd), (ins VPR128:$src, VPR128:$Rn), + asmop # "2\t$Rd.16b, $Rn.8h", + [], NoItinerary>; + + def 4s8h : NeonI_2VMisc<0b1, U, 0b01, opcode, + (outs VPR128:$Rd), (ins VPR128:$src, VPR128:$Rn), + asmop # "2\t$Rd.8h, $Rn.4s", + [], NoItinerary>; + + def 2d4s : NeonI_2VMisc<0b1, U, 0b10, opcode, + (outs VPR128:$Rd), (ins VPR128:$src, VPR128:$Rn), + asmop # "2\t$Rd.4s, $Rn.2d", + [], NoItinerary>; + } +} + +defm XTN : NeonI_2VMisc_HSD_Narrow<"xtn", 0b0, 0b10010>; +defm SQXTUN : NeonI_2VMisc_HSD_Narrow<"sqxtun", 0b1, 0b10010>; +defm SQXTN : NeonI_2VMisc_HSD_Narrow<"sqxtn", 0b0, 0b10100>; +defm UQXTN : NeonI_2VMisc_HSD_Narrow<"uqxtn", 0b1, 0b10100>; + +multiclass NeonI_2VMisc_Narrow_Patterns<string Prefix, + SDPatternOperator Neon_Op> { + def : Pat<(v8i8 (Neon_Op (v8i16 VPR128:$Rn))), + (v8i8 (!cast<Instruction>(Prefix # 8h8b) (v8i16 VPR128:$Rn)))>; + + def : Pat<(v4i16 (Neon_Op (v4i32 VPR128:$Rn))), + (v4i16 (!cast<Instruction>(Prefix # 4s4h) (v4i32 VPR128:$Rn)))>; + + def : Pat<(v2i32 (Neon_Op (v2i64 VPR128:$Rn))), + (v2i32 (!cast<Instruction>(Prefix # 2d2s) (v2i64 VPR128:$Rn)))>; + + def : Pat<(v16i8 (concat_vectors + (v8i8 VPR64:$src), + (v8i8 (Neon_Op (v8i16 VPR128:$Rn))))), + (!cast<Instruction>(Prefix # 8h16b) + (SUBREG_TO_REG (i32 0), VPR64:$src, sub_64), + VPR128:$Rn)>; + + def : Pat<(v8i16 (concat_vectors + (v4i16 VPR64:$src), + (v4i16 (Neon_Op (v4i32 VPR128:$Rn))))), + (!cast<Instruction>(Prefix # 4s8h) + (SUBREG_TO_REG (i32 0), VPR64:$src, sub_64), + VPR128:$Rn)>; + + def : Pat<(v4i32 (concat_vectors + (v2i32 VPR64:$src), + (v2i32 (Neon_Op (v2i64 VPR128:$Rn))))), + (!cast<Instruction>(Prefix # 2d4s) + (SUBREG_TO_REG (i32 0), VPR64:$src, sub_64), + VPR128:$Rn)>; +} + +defm : NeonI_2VMisc_Narrow_Patterns<"XTN", trunc>; +defm : NeonI_2VMisc_Narrow_Patterns<"SQXTUN", int_arm_neon_vqmovnsu>; +defm : NeonI_2VMisc_Narrow_Patterns<"SQXTN", int_arm_neon_vqmovns>; +defm : NeonI_2VMisc_Narrow_Patterns<"UQXTN", int_arm_neon_vqmovnu>; + +multiclass NeonI_2VMisc_SHIFT<string asmop, bit U, bits<5> opcode> { + let DecoderMethod = "DecodeSHLLInstruction" in { + def 8b8h : NeonI_2VMisc<0b0, U, 0b00, opcode, + (outs VPR128:$Rd), + (ins VPR64:$Rn, uimm_exact8:$Imm), + asmop # "\t$Rd.8h, $Rn.8b, $Imm", + [], NoItinerary>; + + def 4h4s : NeonI_2VMisc<0b0, U, 0b01, opcode, + (outs VPR128:$Rd), + (ins VPR64:$Rn, uimm_exact16:$Imm), + asmop # "\t$Rd.4s, $Rn.4h, $Imm", + [], NoItinerary>; + + def 2s2d : NeonI_2VMisc<0b0, U, 0b10, opcode, + (outs VPR128:$Rd), + (ins VPR64:$Rn, uimm_exact32:$Imm), + asmop # "\t$Rd.2d, $Rn.2s, $Imm", + [], NoItinerary>; + + def 16b8h : NeonI_2VMisc<0b1, U, 0b00, opcode, + (outs VPR128:$Rd), + (ins VPR128:$Rn, uimm_exact8:$Imm), + asmop # "2\t$Rd.8h, $Rn.16b, $Imm", + [], NoItinerary>; + + def 8h4s : NeonI_2VMisc<0b1, U, 0b01, opcode, + (outs VPR128:$Rd), + (ins VPR128:$Rn, uimm_exact16:$Imm), + asmop # "2\t$Rd.4s, $Rn.8h, $Imm", + [], NoItinerary>; + + def 4s2d : NeonI_2VMisc<0b1, U, 0b10, opcode, + (outs VPR128:$Rd), + (ins VPR128:$Rn, uimm_exact32:$Imm), + asmop # "2\t$Rd.2d, $Rn.4s, $Imm", + [], NoItinerary>; + } +} + +defm SHLL : NeonI_2VMisc_SHIFT<"shll", 0b1, 0b10011>; + +class NeonI_SHLL_Patterns<ValueType OpTy, ValueType DesTy, + SDPatternOperator ExtOp, Operand Neon_Imm, + string suffix> + : Pat<(DesTy (shl + (DesTy (ExtOp (OpTy VPR64:$Rn))), + (DesTy (Neon_vdup + (i32 Neon_Imm:$Imm))))), + (!cast<Instruction>("SHLL" # suffix) VPR64:$Rn, Neon_Imm:$Imm)>; + +class NeonI_SHLL_High_Patterns<ValueType OpTy, ValueType DesTy, + SDPatternOperator ExtOp, Operand Neon_Imm, + string suffix, PatFrag GetHigh> + : Pat<(DesTy (shl + (DesTy (ExtOp + (OpTy (GetHigh VPR128:$Rn)))), + (DesTy (Neon_vdup + (i32 Neon_Imm:$Imm))))), + (!cast<Instruction>("SHLL" # suffix) VPR128:$Rn, Neon_Imm:$Imm)>; + +def : NeonI_SHLL_Patterns<v8i8, v8i16, zext, uimm_exact8, "8b8h">; +def : NeonI_SHLL_Patterns<v8i8, v8i16, sext, uimm_exact8, "8b8h">; +def : NeonI_SHLL_Patterns<v4i16, v4i32, zext, uimm_exact16, "4h4s">; +def : NeonI_SHLL_Patterns<v4i16, v4i32, sext, uimm_exact16, "4h4s">; +def : NeonI_SHLL_Patterns<v2i32, v2i64, zext, uimm_exact32, "2s2d">; +def : NeonI_SHLL_Patterns<v2i32, v2i64, sext, uimm_exact32, "2s2d">; +def : NeonI_SHLL_High_Patterns<v8i8, v8i16, zext, uimm_exact8, "16b8h", + Neon_High16B>; +def : NeonI_SHLL_High_Patterns<v8i8, v8i16, sext, uimm_exact8, "16b8h", + Neon_High16B>; +def : NeonI_SHLL_High_Patterns<v4i16, v4i32, zext, uimm_exact16, "8h4s", + Neon_High8H>; +def : NeonI_SHLL_High_Patterns<v4i16, v4i32, sext, uimm_exact16, "8h4s", + Neon_High8H>; +def : NeonI_SHLL_High_Patterns<v2i32, v2i64, zext, uimm_exact32, "4s2d", + Neon_High4S>; +def : NeonI_SHLL_High_Patterns<v2i32, v2i64, sext, uimm_exact32, "4s2d", + Neon_High4S>; + +multiclass NeonI_2VMisc_SD_Narrow<string asmop, bit U, bits<5> opcode> { + def 4s4h : NeonI_2VMisc<0b0, U, 0b00, opcode, + (outs VPR64:$Rd), (ins VPR128:$Rn), + asmop # "\t$Rd.4h, $Rn.4s", + [], NoItinerary>; + + def 2d2s : NeonI_2VMisc<0b0, U, 0b01, opcode, + (outs VPR64:$Rd), (ins VPR128:$Rn), + asmop # "\t$Rd.2s, $Rn.2d", + [], NoItinerary>; + + let Constraints = "$src = $Rd" in { + def 4s8h : NeonI_2VMisc<0b1, U, 0b00, opcode, + (outs VPR128:$Rd), (ins VPR128:$src, VPR128:$Rn), + asmop # "2\t$Rd.8h, $Rn.4s", + [], NoItinerary>; + + def 2d4s : NeonI_2VMisc<0b1, U, 0b01, opcode, + (outs VPR128:$Rd), (ins VPR128:$src, VPR128:$Rn), + asmop # "2\t$Rd.4s, $Rn.2d", + [], NoItinerary>; + } +} + +defm FCVTN : NeonI_2VMisc_SD_Narrow<"fcvtn", 0b0, 0b10110>; + +multiclass NeonI_2VMisc_Narrow_Pattern<string prefix, + SDPatternOperator f32_to_f16_Op, + SDPatternOperator f64_to_f32_Op> { + + def : Pat<(v4i16 (f32_to_f16_Op (v4f32 VPR128:$Rn))), + (!cast<Instruction>(prefix # "4s4h") (v4f32 VPR128:$Rn))>; + + def : Pat<(v8i16 (concat_vectors + (v4i16 VPR64:$src), + (v4i16 (f32_to_f16_Op (v4f32 VPR128:$Rn))))), + (!cast<Instruction>(prefix # "4s8h") + (v4f32 (SUBREG_TO_REG (i32 0), VPR64:$src, sub_64)), + (v4f32 VPR128:$Rn))>; + + def : Pat<(v2f32 (f64_to_f32_Op (v2f64 VPR128:$Rn))), + (!cast<Instruction>(prefix # "2d2s") (v2f64 VPR128:$Rn))>; + + def : Pat<(v4f32 (concat_vectors + (v2f32 VPR64:$src), + (v2f32 (f64_to_f32_Op (v2f64 VPR128:$Rn))))), + (!cast<Instruction>(prefix # "2d4s") + (v4f32 (SUBREG_TO_REG (i32 0), VPR64:$src, sub_64)), + (v2f64 VPR128:$Rn))>; +} + +defm : NeonI_2VMisc_Narrow_Pattern<"FCVTN", int_arm_neon_vcvtfp2hf, fround>; + +multiclass NeonI_2VMisc_D_Narrow<string asmop, string prefix, bit U, + bits<5> opcode> { + def 2d2s : NeonI_2VMisc<0b0, U, 0b01, opcode, + (outs VPR64:$Rd), (ins VPR128:$Rn), + asmop # "\t$Rd.2s, $Rn.2d", + [], NoItinerary>; + + def 2d4s : NeonI_2VMisc<0b1, U, 0b01, opcode, + (outs VPR128:$Rd), (ins VPR128:$src, VPR128:$Rn), + asmop # "2\t$Rd.4s, $Rn.2d", + [], NoItinerary> { + let Constraints = "$src = $Rd"; + } + + def : Pat<(v2f32 (int_aarch64_neon_fcvtxn (v2f64 VPR128:$Rn))), + (!cast<Instruction>(prefix # "2d2s") VPR128:$Rn)>; + + def : Pat<(v4f32 (concat_vectors + (v2f32 VPR64:$src), + (v2f32 (int_aarch64_neon_fcvtxn (v2f64 VPR128:$Rn))))), + (!cast<Instruction>(prefix # "2d4s") + (v4f32 (SUBREG_TO_REG (i32 0), VPR64:$src, sub_64)), + VPR128:$Rn)>; +} + +defm FCVTXN : NeonI_2VMisc_D_Narrow<"fcvtxn","FCVTXN", 0b1, 0b10110>; + +def Neon_High4Float : PatFrag<(ops node:$in), + (extract_subvector (v4f32 node:$in), (iPTR 2))>; + +multiclass NeonI_2VMisc_HS_Extend<string asmop, bit U, bits<5> opcode> { + def 4h4s : NeonI_2VMisc<0b0, U, 0b00, opcode, + (outs VPR128:$Rd), (ins VPR64:$Rn), + asmop # "\t$Rd.4s, $Rn.4h", + [], NoItinerary>; + + def 2s2d : NeonI_2VMisc<0b0, U, 0b01, opcode, + (outs VPR128:$Rd), (ins VPR64:$Rn), + asmop # "\t$Rd.2d, $Rn.2s", + [], NoItinerary>; + + def 8h4s : NeonI_2VMisc<0b1, U, 0b00, opcode, + (outs VPR128:$Rd), (ins VPR128:$Rn), + asmop # "2\t$Rd.4s, $Rn.8h", + [], NoItinerary>; + + def 4s2d : NeonI_2VMisc<0b1, U, 0b01, opcode, + (outs VPR128:$Rd), (ins VPR128:$Rn), + asmop # "2\t$Rd.2d, $Rn.4s", + [], NoItinerary>; +} + +defm FCVTL : NeonI_2VMisc_HS_Extend<"fcvtl", 0b0, 0b10111>; + +multiclass NeonI_2VMisc_Extend_Pattern<string prefix> { + def : Pat<(v4f32 (int_arm_neon_vcvthf2fp (v4i16 VPR64:$Rn))), + (!cast<Instruction>(prefix # "4h4s") VPR64:$Rn)>; + + def : Pat<(v4f32 (int_arm_neon_vcvthf2fp + (v4i16 (Neon_High8H + (v8i16 VPR128:$Rn))))), + (!cast<Instruction>(prefix # "8h4s") VPR128:$Rn)>; + + def : Pat<(v2f64 (fextend (v2f32 VPR64:$Rn))), + (!cast<Instruction>(prefix # "2s2d") VPR64:$Rn)>; + + def : Pat<(v2f64 (fextend + (v2f32 (Neon_High4Float + (v4f32 VPR128:$Rn))))), + (!cast<Instruction>(prefix # "4s2d") VPR128:$Rn)>; +} + +defm : NeonI_2VMisc_Extend_Pattern<"FCVTL">; + +multiclass NeonI_2VMisc_SD_Conv<string asmop, bit Size, bit U, bits<5> opcode, + ValueType ResTy4s, ValueType OpTy4s, + ValueType ResTy2d, ValueType OpTy2d, + ValueType ResTy2s, ValueType OpTy2s, + SDPatternOperator Neon_Op> { + + def 4s : NeonI_2VMisc<0b1, U, {Size, 0b0}, opcode, + (outs VPR128:$Rd), (ins VPR128:$Rn), + asmop # "\t$Rd.4s, $Rn.4s", + [(set (ResTy4s VPR128:$Rd), + (ResTy4s (Neon_Op (OpTy4s VPR128:$Rn))))], + NoItinerary>; + + def 2d : NeonI_2VMisc<0b1, U, {Size, 0b1}, opcode, + (outs VPR128:$Rd), (ins VPR128:$Rn), + asmop # "\t$Rd.2d, $Rn.2d", + [(set (ResTy2d VPR128:$Rd), + (ResTy2d (Neon_Op (OpTy2d VPR128:$Rn))))], + NoItinerary>; + + def 2s : NeonI_2VMisc<0b0, U, {Size, 0b0}, opcode, + (outs VPR64:$Rd), (ins VPR64:$Rn), + asmop # "\t$Rd.2s, $Rn.2s", + [(set (ResTy2s VPR64:$Rd), + (ResTy2s (Neon_Op (OpTy2s VPR64:$Rn))))], + NoItinerary>; +} + +multiclass NeonI_2VMisc_fp_to_int<string asmop, bit Size, bit U, + bits<5> opcode, SDPatternOperator Neon_Op> { + defm _ : NeonI_2VMisc_SD_Conv<asmop, Size, U, opcode, v4i32, v4f32, v2i64, + v2f64, v2i32, v2f32, Neon_Op>; +} + +defm FCVTNS : NeonI_2VMisc_fp_to_int<"fcvtns", 0b0, 0b0, 0b11010, + int_aarch64_neon_fcvtns>; +defm FCVTNU : NeonI_2VMisc_fp_to_int<"fcvtnu", 0b0, 0b1, 0b11010, + int_aarch64_neon_fcvtnu>; +defm FCVTPS : NeonI_2VMisc_fp_to_int<"fcvtps", 0b1, 0b0, 0b11010, + int_aarch64_neon_fcvtps>; +defm FCVTPU : NeonI_2VMisc_fp_to_int<"fcvtpu", 0b1, 0b1, 0b11010, + int_aarch64_neon_fcvtpu>; +defm FCVTMS : NeonI_2VMisc_fp_to_int<"fcvtms", 0b0, 0b0, 0b11011, + int_aarch64_neon_fcvtms>; +defm FCVTMU : NeonI_2VMisc_fp_to_int<"fcvtmu", 0b0, 0b1, 0b11011, + int_aarch64_neon_fcvtmu>; +defm FCVTZS : NeonI_2VMisc_fp_to_int<"fcvtzs", 0b1, 0b0, 0b11011, fp_to_sint>; +defm FCVTZU : NeonI_2VMisc_fp_to_int<"fcvtzu", 0b1, 0b1, 0b11011, fp_to_uint>; +defm FCVTAS : NeonI_2VMisc_fp_to_int<"fcvtas", 0b0, 0b0, 0b11100, + int_aarch64_neon_fcvtas>; +defm FCVTAU : NeonI_2VMisc_fp_to_int<"fcvtau", 0b0, 0b1, 0b11100, + int_aarch64_neon_fcvtau>; + +multiclass NeonI_2VMisc_int_to_fp<string asmop, bit Size, bit U, + bits<5> opcode, SDPatternOperator Neon_Op> { + defm _ : NeonI_2VMisc_SD_Conv<asmop, Size, U, opcode, v4f32, v4i32, v2f64, + v2i64, v2f32, v2i32, Neon_Op>; +} + +defm SCVTF : NeonI_2VMisc_int_to_fp<"scvtf", 0b0, 0b0, 0b11101, sint_to_fp>; +defm UCVTF : NeonI_2VMisc_int_to_fp<"ucvtf", 0b0, 0b1, 0b11101, uint_to_fp>; + +multiclass NeonI_2VMisc_fp_to_fp<string asmop, bit Size, bit U, + bits<5> opcode, SDPatternOperator Neon_Op> { + defm _ : NeonI_2VMisc_SD_Conv<asmop, Size, U, opcode, v4f32, v4f32, v2f64, + v2f64, v2f32, v2f32, Neon_Op>; +} + +defm FRINTN : NeonI_2VMisc_fp_to_fp<"frintn", 0b0, 0b0, 0b11000, + int_aarch64_neon_frintn>; +defm FRINTA : NeonI_2VMisc_fp_to_fp<"frinta", 0b0, 0b1, 0b11000, frnd>; +defm FRINTP : NeonI_2VMisc_fp_to_fp<"frintp", 0b1, 0b0, 0b11000, fceil>; +defm FRINTM : NeonI_2VMisc_fp_to_fp<"frintm", 0b0, 0b0, 0b11001, ffloor>; +defm FRINTX : NeonI_2VMisc_fp_to_fp<"frintx", 0b0, 0b1, 0b11001, frint>; +defm FRINTZ : NeonI_2VMisc_fp_to_fp<"frintz", 0b1, 0b0, 0b11001, ftrunc>; +defm FRINTI : NeonI_2VMisc_fp_to_fp<"frinti", 0b1, 0b1, 0b11001, fnearbyint>; +defm FRECPE : NeonI_2VMisc_fp_to_fp<"frecpe", 0b1, 0b0, 0b11101, + int_arm_neon_vrecpe>; +defm FRSQRTE : NeonI_2VMisc_fp_to_fp<"frsqrte", 0b1, 0b1, 0b11101, + int_arm_neon_vrsqrte>; +defm FSQRT : NeonI_2VMisc_fp_to_fp<"fsqrt", 0b1, 0b1, 0b11111, fsqrt>; + +multiclass NeonI_2VMisc_S_Conv<string asmop, bit Size, bit U, + bits<5> opcode, SDPatternOperator Neon_Op> { + def 4s : NeonI_2VMisc<0b1, U, {Size, 0b0}, opcode, + (outs VPR128:$Rd), (ins VPR128:$Rn), + asmop # "\t$Rd.4s, $Rn.4s", + [(set (v4i32 VPR128:$Rd), + (v4i32 (Neon_Op (v4i32 VPR128:$Rn))))], + NoItinerary>; + + def 2s : NeonI_2VMisc<0b0, U, {Size, 0b0}, opcode, + (outs VPR64:$Rd), (ins VPR64:$Rn), + asmop # "\t$Rd.2s, $Rn.2s", + [(set (v2i32 VPR64:$Rd), + (v2i32 (Neon_Op (v2i32 VPR64:$Rn))))], + NoItinerary>; +} + +defm URECPE : NeonI_2VMisc_S_Conv<"urecpe", 0b1, 0b0, 0b11100, + int_arm_neon_vrecpe>; +defm URSQRTE : NeonI_2VMisc_S_Conv<"ursqrte", 0b1, 0b1, 0b11100, + int_arm_neon_vrsqrte>; + +// Crypto Class +class NeonI_Cryptoaes_2v<bits<2> size, bits<5> opcode, + string asmop, SDPatternOperator opnode> + : NeonI_Crypto_AES<size, opcode, + (outs VPR128:$Rd), (ins VPR128:$src, VPR128:$Rn), + asmop # "\t$Rd.16b, $Rn.16b", + [(set (v16i8 VPR128:$Rd), + (v16i8 (opnode (v16i8 VPR128:$src), + (v16i8 VPR128:$Rn))))], + NoItinerary>{ + let Constraints = "$src = $Rd"; + let Predicates = [HasNEON, HasCrypto]; +} + +def AESE : NeonI_Cryptoaes_2v<0b00, 0b00100, "aese", int_arm_neon_aese>; +def AESD : NeonI_Cryptoaes_2v<0b00, 0b00101, "aesd", int_arm_neon_aesd>; + +class NeonI_Cryptoaes<bits<2> size, bits<5> opcode, + string asmop, SDPatternOperator opnode> + : NeonI_Crypto_AES<size, opcode, + (outs VPR128:$Rd), (ins VPR128:$Rn), + asmop # "\t$Rd.16b, $Rn.16b", + [(set (v16i8 VPR128:$Rd), + (v16i8 (opnode (v16i8 VPR128:$Rn))))], + NoItinerary>; + +def AESMC : NeonI_Cryptoaes<0b00, 0b00110, "aesmc", int_arm_neon_aesmc>; +def AESIMC : NeonI_Cryptoaes<0b00, 0b00111, "aesimc", int_arm_neon_aesimc>; + +class NeonI_Cryptosha_vv<bits<2> size, bits<5> opcode, + string asmop, SDPatternOperator opnode> + : NeonI_Crypto_SHA<size, opcode, + (outs VPR128:$Rd), (ins VPR128:$src, VPR128:$Rn), + asmop # "\t$Rd.4s, $Rn.4s", + [(set (v4i32 VPR128:$Rd), + (v4i32 (opnode (v4i32 VPR128:$src), + (v4i32 VPR128:$Rn))))], + NoItinerary> { + let Constraints = "$src = $Rd"; + let Predicates = [HasNEON, HasCrypto]; +} + +def SHA1SU1 : NeonI_Cryptosha_vv<0b00, 0b00001, "sha1su1", + int_arm_neon_sha1su1>; +def SHA256SU0 : NeonI_Cryptosha_vv<0b00, 0b00010, "sha256su0", + int_arm_neon_sha256su0>; + +class NeonI_Cryptosha_ss<bits<2> size, bits<5> opcode, + string asmop, SDPatternOperator opnode> + : NeonI_Crypto_SHA<size, opcode, + (outs FPR32:$Rd), (ins FPR32:$Rn), + asmop # "\t$Rd, $Rn", + [(set (v1i32 FPR32:$Rd), + (v1i32 (opnode (v1i32 FPR32:$Rn))))], + NoItinerary> { + let Predicates = [HasNEON, HasCrypto]; +} + +def SHA1H : NeonI_Cryptosha_ss<0b00, 0b00000, "sha1h", int_arm_neon_sha1h>; + +class NeonI_Cryptosha3_vvv<bits<2> size, bits<3> opcode, string asmop, + SDPatternOperator opnode> + : NeonI_Crypto_3VSHA<size, opcode, + (outs VPR128:$Rd), + (ins VPR128:$src, VPR128:$Rn, VPR128:$Rm), + asmop # "\t$Rd.4s, $Rn.4s, $Rm.4s", + [(set (v4i32 VPR128:$Rd), + (v4i32 (opnode (v4i32 VPR128:$src), + (v4i32 VPR128:$Rn), + (v4i32 VPR128:$Rm))))], + NoItinerary> { + let Constraints = "$src = $Rd"; + let Predicates = [HasNEON, HasCrypto]; +} + +def SHA1SU0 : NeonI_Cryptosha3_vvv<0b00, 0b011, "sha1su0", + int_arm_neon_sha1su0>; +def SHA256SU1 : NeonI_Cryptosha3_vvv<0b00, 0b110, "sha256su1", + int_arm_neon_sha256su1>; + +class NeonI_Cryptosha3_qqv<bits<2> size, bits<3> opcode, string asmop, + SDPatternOperator opnode> + : NeonI_Crypto_3VSHA<size, opcode, + (outs FPR128:$Rd), + (ins FPR128:$src, FPR128:$Rn, VPR128:$Rm), + asmop # "\t$Rd, $Rn, $Rm.4s", + [(set (v4i32 FPR128:$Rd), + (v4i32 (opnode (v4i32 FPR128:$src), + (v4i32 FPR128:$Rn), + (v4i32 VPR128:$Rm))))], + NoItinerary> { + let Constraints = "$src = $Rd"; + let Predicates = [HasNEON, HasCrypto]; +} + +def SHA256H : NeonI_Cryptosha3_qqv<0b00, 0b100, "sha256h", + int_arm_neon_sha256h>; +def SHA256H2 : NeonI_Cryptosha3_qqv<0b00, 0b101, "sha256h2", + int_arm_neon_sha256h2>; + +class NeonI_Cryptosha3_qsv<bits<2> size, bits<3> opcode, string asmop, + SDPatternOperator opnode> + : NeonI_Crypto_3VSHA<size, opcode, + (outs FPR128:$Rd), + (ins FPR128:$src, FPR32:$Rn, VPR128:$Rm), + asmop # "\t$Rd, $Rn, $Rm.4s", + [(set (v4i32 FPR128:$Rd), + (v4i32 (opnode (v4i32 FPR128:$src), + (v1i32 FPR32:$Rn), + (v4i32 VPR128:$Rm))))], + NoItinerary> { + let Constraints = "$src = $Rd"; + let Predicates = [HasNEON, HasCrypto]; +} + +def SHA1C : NeonI_Cryptosha3_qsv<0b00, 0b000, "sha1c", int_aarch64_neon_sha1c>; +def SHA1P : NeonI_Cryptosha3_qsv<0b00, 0b001, "sha1p", int_aarch64_neon_sha1p>; +def SHA1M : NeonI_Cryptosha3_qsv<0b00, 0b010, "sha1m", int_aarch64_neon_sha1m>; + +// +// Patterns for handling half-precision values +// + +// Convert f16 value coming in as i16 value to f32 +def : Pat<(f32 (f16_to_f32 (i32 (and (i32 GPR32:$Rn), 65535)))), + (FCVTsh (EXTRACT_SUBREG (FMOVsw GPR32:$Rn), sub_16))>; +def : Pat<(f32 (f16_to_f32 (i32 (assertzext GPR32:$Rn)))), + (FCVTsh (EXTRACT_SUBREG (FMOVsw GPR32:$Rn), sub_16))>; + +def : Pat<(f32 (f16_to_f32 (i32 (assertzext (i32 ( + f32_to_f16 (f32 FPR32:$Rn))))))), + (f32 FPR32:$Rn)>; + +// Patterns for vector extract of half-precision FP value in i16 storage type +def : Pat<(f32 (f16_to_f32 ( i32 (and (i32 (vector_extract + (v4i16 VPR64:$Rn), neon_uimm2_bare:$Imm)), 65535)))), + (FCVTsh (f16 (DUPhv_H + (v8i16 (SUBREG_TO_REG (i64 0), VPR64:$Rn, sub_64)), + neon_uimm2_bare:$Imm)))>; + +def : Pat<(f32 (f16_to_f32 ( i32 (and (i32 (vector_extract + (v8i16 VPR128:$Rn), neon_uimm3_bare:$Imm)), 65535)))), + (FCVTsh (f16 (DUPhv_H (v8i16 VPR128:$Rn), neon_uimm3_bare:$Imm)))>; + +// Patterns for vector insert of half-precision FP value 0 in i16 storage type +def : Pat<(v8i16 (vector_insert (v8i16 VPR128:$Rn), + (i32 (assertsext (i32 (fp_to_sint(f32 (f16_to_f32 (i32 0))))))), + (neon_uimm3_bare:$Imm))), + (v8i16 (INSELh (v8i16 VPR128:$Rn), + (v8i16 (SUBREG_TO_REG (i64 0), + (f16 (EXTRACT_SUBREG (f32 (FMOVsw (i32 WZR))), sub_16)), + sub_16)), + neon_uimm3_bare:$Imm, 0))>; + +def : Pat<(v4i16 (vector_insert (v4i16 VPR64:$Rn), + (i32 (assertsext (i32 (fp_to_sint(f32 (f16_to_f32 (i32 0))))))), + (neon_uimm2_bare:$Imm))), + (v4i16 (EXTRACT_SUBREG + (v8i16 (INSELh + (v8i16 (SUBREG_TO_REG (i64 0), VPR64:$Rn, sub_64)), + (v8i16 (SUBREG_TO_REG (i64 0), + (f16 (EXTRACT_SUBREG (f32 (FMOVsw (i32 WZR))), sub_16)), + sub_16)), + neon_uimm2_bare:$Imm, 0)), + sub_64))>; + +// Patterns for vector insert of half-precision FP value in i16 storage type +def : Pat<(v8i16 (vector_insert (v8i16 VPR128:$Rn), + (i32 (assertsext (i32 (fp_to_sint + (f32 (f16_to_f32 (i32 (and (i32 GPR32:$src), 65535)))))))), + (neon_uimm3_bare:$Imm))), + (v8i16 (INSELh (v8i16 VPR128:$Rn), + (v8i16 (SUBREG_TO_REG (i64 0), + (f16 (EXTRACT_SUBREG (f32 (FMOVsw (i32 GPR32:$src))), sub_16)), + sub_16)), + neon_uimm3_bare:$Imm, 0))>; + +def : Pat<(v4i16 (vector_insert (v4i16 VPR64:$Rn), + (i32 (assertsext (i32 (fp_to_sint + (f32 (f16_to_f32 (i32 (and (i32 GPR32:$src), 65535)))))))), + (neon_uimm2_bare:$Imm))), + (v4i16 (EXTRACT_SUBREG + (v8i16 (INSELh + (v8i16 (SUBREG_TO_REG (i64 0), VPR64:$Rn, sub_64)), + (v8i16 (SUBREG_TO_REG (i64 0), + (f16 (EXTRACT_SUBREG (f32 (FMOVsw (i32 GPR32:$src))), sub_16)), + sub_16)), + neon_uimm2_bare:$Imm, 0)), + sub_64))>; + +def : Pat<(v8i16 (vector_insert (v8i16 VPR128:$Rn), + (i32 (vector_extract (v8i16 VPR128:$src), neon_uimm3_bare:$Imm2)), + (neon_uimm3_bare:$Imm1))), + (v8i16 (INSELh (v8i16 VPR128:$Rn), (v8i16 VPR128:$src), + neon_uimm3_bare:$Imm1, neon_uimm3_bare:$Imm2))>; + +// Patterns for vector copy of half-precision FP value in i16 storage type +def : Pat<(v8i16 (vector_insert (v8i16 VPR128:$Rn), + (i32 (assertsext (i32 (fp_to_sint(f32 (f16_to_f32 (i32 (and (i32 + (vector_extract (v8i16 VPR128:$src), neon_uimm3_bare:$Imm2)), + 65535)))))))), + (neon_uimm3_bare:$Imm1))), + (v8i16 (INSELh (v8i16 VPR128:$Rn), (v8i16 VPR128:$src), + neon_uimm3_bare:$Imm1, neon_uimm3_bare:$Imm2))>; + +def : Pat<(v4i16 (vector_insert (v4i16 VPR64:$Rn), + (i32 (assertsext (i32 (fp_to_sint(f32 (f16_to_f32 (i32 (and (i32 + (vector_extract (v4i16 VPR64:$src), neon_uimm3_bare:$Imm2)), + 65535)))))))), + (neon_uimm3_bare:$Imm1))), + (v4i16 (EXTRACT_SUBREG + (v8i16 (INSELh + (v8i16 (SUBREG_TO_REG (i64 0), VPR64:$Rn, sub_64)), + (v8i16 (SUBREG_TO_REG (i64 0), VPR64:$src, sub_64)), + neon_uimm3_bare:$Imm1, neon_uimm3_bare:$Imm2)), + sub_64))>; + + |
