Google Git
Sign in
llvm / llvm-project / 4b30076f16fc55891da8952bda3f2d81dbeec5a5 / . / llvm / lib / Target / ARM / ARMInstrNEON.td
blob: aaf3280ea15085bbd5a0dc2bee5441c365621f14 [file] [log] [blame]
//===-- ARMInstrNEON.td - NEON support for ARM -------------*- tablegen -*-===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
//
// This file describes the ARM NEON instruction set.
//
//===----------------------------------------------------------------------===//
//===----------------------------------------------------------------------===//
// NEON-specific Operands.
//===----------------------------------------------------------------------===//
def nModImm : Operand<i32> {
let PrintMethod = "printVMOVModImmOperand";
}
def nImmSplatI8AsmOperand : AsmOperandClass { let Name = "NEONi8splat"; }
def nImmSplatI8 : Operand<i32> {
let PrintMethod = "printVMOVModImmOperand";
let ParserMatchClass = nImmSplatI8AsmOperand;
}
def nImmSplatI16AsmOperand : AsmOperandClass { let Name = "NEONi16splat"; }
def nImmSplatI16 : Operand<i32> {
let PrintMethod = "printVMOVModImmOperand";
let ParserMatchClass = nImmSplatI16AsmOperand;
}
def nImmSplatI32AsmOperand : AsmOperandClass { let Name = "NEONi32splat"; }
def nImmSplatI32 : Operand<i32> {
let PrintMethod = "printVMOVModImmOperand";
let ParserMatchClass = nImmSplatI32AsmOperand;
}
def nImmSplatNotI16AsmOperand : AsmOperandClass { let Name = "NEONi16splatNot"; }
def nImmSplatNotI16 : Operand<i32> {
let ParserMatchClass = nImmSplatNotI16AsmOperand;
}
def nImmSplatNotI32AsmOperand : AsmOperandClass { let Name = "NEONi32splatNot"; }
def nImmSplatNotI32 : Operand<i32> {
let ParserMatchClass = nImmSplatNotI32AsmOperand;
}
def nImmVMOVI32AsmOperand : AsmOperandClass { let Name = "NEONi32vmov"; }
def nImmVMOVI32 : Operand<i32> {
let PrintMethod = "printVMOVModImmOperand";
let ParserMatchClass = nImmVMOVI32AsmOperand;
}
class nImmVMOVIAsmOperandReplicate<ValueType From, ValueType To>
: AsmOperandClass {
let Name = "NEONi" # To.Size # "vmovi" # From.Size # "Replicate";
let PredicateMethod = "isNEONmovReplicate<" # From.Size # ", " # To.Size # ">";
let RenderMethod = "addNEONvmovi" # From.Size # "ReplicateOperands";
}
class nImmVINVIAsmOperandReplicate<ValueType From, ValueType To>
: AsmOperandClass {
let Name = "NEONi" # To.Size # "invi" # From.Size # "Replicate";
let PredicateMethod = "isNEONinvReplicate<" # From.Size # ", " # To.Size # ">";
let RenderMethod = "addNEONinvi" # From.Size # "ReplicateOperands";
}
class nImmVMOVIReplicate<ValueType From, ValueType To> : Operand<i32> {
let PrintMethod = "printVMOVModImmOperand";
let ParserMatchClass = nImmVMOVIAsmOperandReplicate<From, To>;
}
class nImmVINVIReplicate<ValueType From, ValueType To> : Operand<i32> {
let PrintMethod = "printVMOVModImmOperand";
let ParserMatchClass = nImmVINVIAsmOperandReplicate<From, To>;
}
def nImmVMOVI32NegAsmOperand : AsmOperandClass { let Name = "NEONi32vmovNeg"; }
def nImmVMOVI32Neg : Operand<i32> {
let PrintMethod = "printVMOVModImmOperand";
let ParserMatchClass = nImmVMOVI32NegAsmOperand;
}
def nImmVMOVF32 : Operand<i32> {
let PrintMethod = "printFPImmOperand";
let ParserMatchClass = FPImmOperand;
}
def nImmSplatI64AsmOperand : AsmOperandClass { let Name = "NEONi64splat"; }
def nImmSplatI64 : Operand<i32> {
let PrintMethod = "printVMOVModImmOperand";
let ParserMatchClass = nImmSplatI64AsmOperand;
}
def VectorIndex8Operand : AsmOperandClass { let Name = "VectorIndex8"; }
def VectorIndex16Operand : AsmOperandClass { let Name = "VectorIndex16"; }
def VectorIndex32Operand : AsmOperandClass { let Name = "VectorIndex32"; }
def VectorIndex64Operand : AsmOperandClass { let Name = "VectorIndex64"; }
def VectorIndex8 : Operand<i32>, ImmLeaf<i32, [{
return ((uint64_t)Imm) < 8;
}]> {
let ParserMatchClass = VectorIndex8Operand;
let PrintMethod = "printVectorIndex";
let MIOperandInfo = (ops i32imm);
}
def VectorIndex16 : Operand<i32>, ImmLeaf<i32, [{
return ((uint64_t)Imm) < 4;
}]> {
let ParserMatchClass = VectorIndex16Operand;
let PrintMethod = "printVectorIndex";
let MIOperandInfo = (ops i32imm);
}
def VectorIndex32 : Operand<i32>, ImmLeaf<i32, [{
return ((uint64_t)Imm) < 2;
}]> {
let ParserMatchClass = VectorIndex32Operand;
let PrintMethod = "printVectorIndex";
let MIOperandInfo = (ops i32imm);
}
def VectorIndex64 : Operand<i32>, ImmLeaf<i32, [{
return ((uint64_t)Imm) < 1;
}]> {
let ParserMatchClass = VectorIndex64Operand;
let PrintMethod = "printVectorIndex";
let MIOperandInfo = (ops i32imm);
}
// Register list of one D register.
def VecListOneDAsmOperand : AsmOperandClass {
let Name = "VecListOneD";
let ParserMethod = "parseVectorList";
let RenderMethod = "addVecListOperands";
}
def VecListOneD : RegisterOperand<DPR, "printVectorListOne"> {
let ParserMatchClass = VecListOneDAsmOperand;
}
// Register list of two sequential D registers.
def VecListDPairAsmOperand : AsmOperandClass {
let Name = "VecListDPair";
let ParserMethod = "parseVectorList";
let RenderMethod = "addVecListOperands";
}
def VecListDPair : RegisterOperand<DPair, "printVectorListTwo"> {
let ParserMatchClass = VecListDPairAsmOperand;
}
// Register list of three sequential D registers.
def VecListThreeDAsmOperand : AsmOperandClass {
let Name = "VecListThreeD";
let ParserMethod = "parseVectorList";
let RenderMethod = "addVecListOperands";
}
def VecListThreeD : RegisterOperand<DPR, "printVectorListThree"> {
let ParserMatchClass = VecListThreeDAsmOperand;
}
// Register list of four sequential D registers.
def VecListFourDAsmOperand : AsmOperandClass {
let Name = "VecListFourD";
let ParserMethod = "parseVectorList";
let RenderMethod = "addVecListOperands";
}
def VecListFourD : RegisterOperand<DPR, "printVectorListFour"> {
let ParserMatchClass = VecListFourDAsmOperand;
}
// Register list of two D registers spaced by 2 (two sequential Q registers).
def VecListDPairSpacedAsmOperand : AsmOperandClass {
let Name = "VecListDPairSpaced";
let ParserMethod = "parseVectorList";
let RenderMethod = "addVecListOperands";
}
def VecListDPairSpaced : RegisterOperand<DPair, "printVectorListTwoSpaced"> {
let ParserMatchClass = VecListDPairSpacedAsmOperand;
}
// Register list of three D registers spaced by 2 (three Q registers).
def VecListThreeQAsmOperand : AsmOperandClass {
let Name = "VecListThreeQ";
let ParserMethod = "parseVectorList";
let RenderMethod = "addVecListOperands";
}
def VecListThreeQ : RegisterOperand<DPR, "printVectorListThreeSpaced"> {
let ParserMatchClass = VecListThreeQAsmOperand;
}
// Register list of three D registers spaced by 2 (three Q registers).
def VecListFourQAsmOperand : AsmOperandClass {
let Name = "VecListFourQ";
let ParserMethod = "parseVectorList";
let RenderMethod = "addVecListOperands";
}
def VecListFourQ : RegisterOperand<DPR, "printVectorListFourSpaced"> {
let ParserMatchClass = VecListFourQAsmOperand;
}
// Register list of one D register, with "all lanes" subscripting.
def VecListOneDAllLanesAsmOperand : AsmOperandClass {
let Name = "VecListOneDAllLanes";
let ParserMethod = "parseVectorList";
let RenderMethod = "addVecListOperands";
}
def VecListOneDAllLanes : RegisterOperand<DPR, "printVectorListOneAllLanes"> {
let ParserMatchClass = VecListOneDAllLanesAsmOperand;
}
// Register list of two D registers, with "all lanes" subscripting.
def VecListDPairAllLanesAsmOperand : AsmOperandClass {
let Name = "VecListDPairAllLanes";
let ParserMethod = "parseVectorList";
let RenderMethod = "addVecListOperands";
}
def VecListDPairAllLanes : RegisterOperand<DPair,
"printVectorListTwoAllLanes"> {
let ParserMatchClass = VecListDPairAllLanesAsmOperand;
}
// Register list of two D registers spaced by 2 (two sequential Q registers).
def VecListDPairSpacedAllLanesAsmOperand : AsmOperandClass {
let Name = "VecListDPairSpacedAllLanes";
let ParserMethod = "parseVectorList";
let RenderMethod = "addVecListOperands";
}
def VecListDPairSpacedAllLanes : RegisterOperand<DPairSpc,
"printVectorListTwoSpacedAllLanes"> {
let ParserMatchClass = VecListDPairSpacedAllLanesAsmOperand;
}
// Register list of three D registers, with "all lanes" subscripting.
def VecListThreeDAllLanesAsmOperand : AsmOperandClass {
let Name = "VecListThreeDAllLanes";
let ParserMethod = "parseVectorList";
let RenderMethod = "addVecListOperands";
}
def VecListThreeDAllLanes : RegisterOperand<DPR,
"printVectorListThreeAllLanes"> {
let ParserMatchClass = VecListThreeDAllLanesAsmOperand;
}
// Register list of three D registers spaced by 2 (three sequential Q regs).
def VecListThreeQAllLanesAsmOperand : AsmOperandClass {
let Name = "VecListThreeQAllLanes";
let ParserMethod = "parseVectorList";
let RenderMethod = "addVecListOperands";
}
def VecListThreeQAllLanes : RegisterOperand<DPR,
"printVectorListThreeSpacedAllLanes"> {
let ParserMatchClass = VecListThreeQAllLanesAsmOperand;
}
// Register list of four D registers, with "all lanes" subscripting.
def VecListFourDAllLanesAsmOperand : AsmOperandClass {
let Name = "VecListFourDAllLanes";
let ParserMethod = "parseVectorList";
let RenderMethod = "addVecListOperands";
}
def VecListFourDAllLanes : RegisterOperand<DPR, "printVectorListFourAllLanes"> {
let ParserMatchClass = VecListFourDAllLanesAsmOperand;
}
// Register list of four D registers spaced by 2 (four sequential Q regs).
def VecListFourQAllLanesAsmOperand : AsmOperandClass {
let Name = "VecListFourQAllLanes";
let ParserMethod = "parseVectorList";
let RenderMethod = "addVecListOperands";
}
def VecListFourQAllLanes : RegisterOperand<DPR,
"printVectorListFourSpacedAllLanes"> {
let ParserMatchClass = VecListFourQAllLanesAsmOperand;
}
// Register list of one D register, with byte lane subscripting.
def VecListOneDByteIndexAsmOperand : AsmOperandClass {
let Name = "VecListOneDByteIndexed";
let ParserMethod = "parseVectorList";
let RenderMethod = "addVecListIndexedOperands";
}
def VecListOneDByteIndexed : Operand<i32> {
let ParserMatchClass = VecListOneDByteIndexAsmOperand;
let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
}
// ...with half-word lane subscripting.
def VecListOneDHWordIndexAsmOperand : AsmOperandClass {
let Name = "VecListOneDHWordIndexed";
let ParserMethod = "parseVectorList";
let RenderMethod = "addVecListIndexedOperands";
}
def VecListOneDHWordIndexed : Operand<i32> {
let ParserMatchClass = VecListOneDHWordIndexAsmOperand;
let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
}
// ...with word lane subscripting.
def VecListOneDWordIndexAsmOperand : AsmOperandClass {
let Name = "VecListOneDWordIndexed";
let ParserMethod = "parseVectorList";
let RenderMethod = "addVecListIndexedOperands";
}
def VecListOneDWordIndexed : Operand<i32> {
let ParserMatchClass = VecListOneDWordIndexAsmOperand;
let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
}
// Register list of two D registers with byte lane subscripting.
def VecListTwoDByteIndexAsmOperand : AsmOperandClass {
let Name = "VecListTwoDByteIndexed";
let ParserMethod = "parseVectorList";
let RenderMethod = "addVecListIndexedOperands";
}
def VecListTwoDByteIndexed : Operand<i32> {
let ParserMatchClass = VecListTwoDByteIndexAsmOperand;
let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
}
// ...with half-word lane subscripting.
def VecListTwoDHWordIndexAsmOperand : AsmOperandClass {
let Name = "VecListTwoDHWordIndexed";
let ParserMethod = "parseVectorList";
let RenderMethod = "addVecListIndexedOperands";
}
def VecListTwoDHWordIndexed : Operand<i32> {
let ParserMatchClass = VecListTwoDHWordIndexAsmOperand;
let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
}
// ...with word lane subscripting.
def VecListTwoDWordIndexAsmOperand : AsmOperandClass {
let Name = "VecListTwoDWordIndexed";
let ParserMethod = "parseVectorList";
let RenderMethod = "addVecListIndexedOperands";
}
def VecListTwoDWordIndexed : Operand<i32> {
let ParserMatchClass = VecListTwoDWordIndexAsmOperand;
let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
}
// Register list of two Q registers with half-word lane subscripting.
def VecListTwoQHWordIndexAsmOperand : AsmOperandClass {
let Name = "VecListTwoQHWordIndexed";
let ParserMethod = "parseVectorList";
let RenderMethod = "addVecListIndexedOperands";
}
def VecListTwoQHWordIndexed : Operand<i32> {
let ParserMatchClass = VecListTwoQHWordIndexAsmOperand;
let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
}
// ...with word lane subscripting.
def VecListTwoQWordIndexAsmOperand : AsmOperandClass {
let Name = "VecListTwoQWordIndexed";
let ParserMethod = "parseVectorList";
let RenderMethod = "addVecListIndexedOperands";
}
def VecListTwoQWordIndexed : Operand<i32> {
let ParserMatchClass = VecListTwoQWordIndexAsmOperand;
let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
}
// Register list of three D registers with byte lane subscripting.
def VecListThreeDByteIndexAsmOperand : AsmOperandClass {
let Name = "VecListThreeDByteIndexed";
let ParserMethod = "parseVectorList";
let RenderMethod = "addVecListIndexedOperands";
}
def VecListThreeDByteIndexed : Operand<i32> {
let ParserMatchClass = VecListThreeDByteIndexAsmOperand;
let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
}
// ...with half-word lane subscripting.
def VecListThreeDHWordIndexAsmOperand : AsmOperandClass {
let Name = "VecListThreeDHWordIndexed";
let ParserMethod = "parseVectorList";
let RenderMethod = "addVecListIndexedOperands";
}
def VecListThreeDHWordIndexed : Operand<i32> {
let ParserMatchClass = VecListThreeDHWordIndexAsmOperand;
let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
}
// ...with word lane subscripting.
def VecListThreeDWordIndexAsmOperand : AsmOperandClass {
let Name = "VecListThreeDWordIndexed";
let ParserMethod = "parseVectorList";
let RenderMethod = "addVecListIndexedOperands";
}
def VecListThreeDWordIndexed : Operand<i32> {
let ParserMatchClass = VecListThreeDWordIndexAsmOperand;
let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
}
// Register list of three Q registers with half-word lane subscripting.
def VecListThreeQHWordIndexAsmOperand : AsmOperandClass {
let Name = "VecListThreeQHWordIndexed";
let ParserMethod = "parseVectorList";
let RenderMethod = "addVecListIndexedOperands";
}
def VecListThreeQHWordIndexed : Operand<i32> {
let ParserMatchClass = VecListThreeQHWordIndexAsmOperand;
let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
}
// ...with word lane subscripting.
def VecListThreeQWordIndexAsmOperand : AsmOperandClass {
let Name = "VecListThreeQWordIndexed";
let ParserMethod = "parseVectorList";
let RenderMethod = "addVecListIndexedOperands";
}
def VecListThreeQWordIndexed : Operand<i32> {
let ParserMatchClass = VecListThreeQWordIndexAsmOperand;
let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
}
// Register list of four D registers with byte lane subscripting.
def VecListFourDByteIndexAsmOperand : AsmOperandClass {
let Name = "VecListFourDByteIndexed";
let ParserMethod = "parseVectorList";
let RenderMethod = "addVecListIndexedOperands";
}
def VecListFourDByteIndexed : Operand<i32> {
let ParserMatchClass = VecListFourDByteIndexAsmOperand;
let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
}
// ...with half-word lane subscripting.
def VecListFourDHWordIndexAsmOperand : AsmOperandClass {
let Name = "VecListFourDHWordIndexed";
let ParserMethod = "parseVectorList";
let RenderMethod = "addVecListIndexedOperands";
}
def VecListFourDHWordIndexed : Operand<i32> {
let ParserMatchClass = VecListFourDHWordIndexAsmOperand;
let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
}
// ...with word lane subscripting.
def VecListFourDWordIndexAsmOperand : AsmOperandClass {
let Name = "VecListFourDWordIndexed";
let ParserMethod = "parseVectorList";
let RenderMethod = "addVecListIndexedOperands";
}
def VecListFourDWordIndexed : Operand<i32> {
let ParserMatchClass = VecListFourDWordIndexAsmOperand;
let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
}
// Register list of four Q registers with half-word lane subscripting.
def VecListFourQHWordIndexAsmOperand : AsmOperandClass {
let Name = "VecListFourQHWordIndexed";
let ParserMethod = "parseVectorList";
let RenderMethod = "addVecListIndexedOperands";
}
def VecListFourQHWordIndexed : Operand<i32> {
let ParserMatchClass = VecListFourQHWordIndexAsmOperand;
let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
}
// ...with word lane subscripting.
def VecListFourQWordIndexAsmOperand : AsmOperandClass {
let Name = "VecListFourQWordIndexed";
let ParserMethod = "parseVectorList";
let RenderMethod = "addVecListIndexedOperands";
}
def VecListFourQWordIndexed : Operand<i32> {
let ParserMatchClass = VecListFourQWordIndexAsmOperand;
let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
}
def dword_alignedload : PatFrag<(ops node:$ptr), (load node:$ptr), [{
return cast<LoadSDNode>(N)->getAlignment() >= 8;
}]>;
def dword_alignedstore : PatFrag<(ops node:$val, node:$ptr),
(store node:$val, node:$ptr), [{
return cast<StoreSDNode>(N)->getAlignment() >= 8;
}]>;
def word_alignedload : PatFrag<(ops node:$ptr), (load node:$ptr), [{
return cast<LoadSDNode>(N)->getAlignment() == 4;
}]>;
def word_alignedstore : PatFrag<(ops node:$val, node:$ptr),
(store node:$val, node:$ptr), [{
return cast<StoreSDNode>(N)->getAlignment() == 4;
}]>;
def hword_alignedload : PatFrag<(ops node:$ptr), (load node:$ptr), [{
return cast<LoadSDNode>(N)->getAlignment() == 2;
}]>;
def hword_alignedstore : PatFrag<(ops node:$val, node:$ptr),
(store node:$val, node:$ptr), [{
return cast<StoreSDNode>(N)->getAlignment() == 2;
}]>;
def byte_alignedload : PatFrag<(ops node:$ptr), (load node:$ptr), [{
return cast<LoadSDNode>(N)->getAlignment() == 1;
}]>;
def byte_alignedstore : PatFrag<(ops node:$val, node:$ptr),
(store node:$val, node:$ptr), [{
return cast<StoreSDNode>(N)->getAlignment() == 1;
}]>;
def non_word_alignedload : PatFrag<(ops node:$ptr), (load node:$ptr), [{
return cast<LoadSDNode>(N)->getAlignment() < 4;
}]>;
def non_word_alignedstore : PatFrag<(ops node:$val, node:$ptr),
(store node:$val, node:$ptr), [{
return cast<StoreSDNode>(N)->getAlignment() < 4;
}]>;
//===----------------------------------------------------------------------===//
// NEON-specific DAG Nodes.
//===----------------------------------------------------------------------===//
def SDTARMVTST : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<1, 2>]>;
def NEONvtst : SDNode<"ARMISD::VTST", SDTARMVTST>;
// Types for vector shift by immediates. The "SHX" version is for long and
// narrow operations where the source and destination vectors have different
// types. The "SHINS" version is for shift and insert operations.
def SDTARMVSHXIMM : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
SDTCisVT<2, i32>]>;
def SDTARMVSHINSIMM : SDTypeProfile<1, 3, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
def NEONvshrnImm : SDNode<"ARMISD::VSHRNIMM", SDTARMVSHXIMM>;
def NEONvrshrsImm : SDNode<"ARMISD::VRSHRsIMM", SDTARMVSHIMM>;
def NEONvrshruImm : SDNode<"ARMISD::VRSHRuIMM", SDTARMVSHIMM>;
def NEONvrshrnImm : SDNode<"ARMISD::VRSHRNIMM", SDTARMVSHXIMM>;
def NEONvqshlsImm : SDNode<"ARMISD::VQSHLsIMM", SDTARMVSHIMM>;
def NEONvqshluImm : SDNode<"ARMISD::VQSHLuIMM", SDTARMVSHIMM>;
def NEONvqshlsuImm : SDNode<"ARMISD::VQSHLsuIMM", SDTARMVSHIMM>;
def NEONvqshrnsImm : SDNode<"ARMISD::VQSHRNsIMM", SDTARMVSHXIMM>;
def NEONvqshrnuImm : SDNode<"ARMISD::VQSHRNuIMM", SDTARMVSHXIMM>;
def NEONvqshrnsuImm : SDNode<"ARMISD::VQSHRNsuIMM", SDTARMVSHXIMM>;
def NEONvqrshrnsImm : SDNode<"ARMISD::VQRSHRNsIMM", SDTARMVSHXIMM>;
def NEONvqrshrnuImm : SDNode<"ARMISD::VQRSHRNuIMM", SDTARMVSHXIMM>;
def NEONvqrshrnsuImm : SDNode<"ARMISD::VQRSHRNsuIMM", SDTARMVSHXIMM>;
def NEONvsliImm : SDNode<"ARMISD::VSLIIMM", SDTARMVSHINSIMM>;
def NEONvsriImm : SDNode<"ARMISD::VSRIIMM", SDTARMVSHINSIMM>;
def NEONvbsp : SDNode<"ARMISD::VBSP",
SDTypeProfile<1, 3, [SDTCisVec<0>,
SDTCisSameAs<0, 1>,
SDTCisSameAs<0, 2>,
SDTCisSameAs<0, 3>]>>;
def SDTARMVEXT : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
def NEONvext : SDNode<"ARMISD::VEXT", SDTARMVEXT>;
def SDTARMVSHUF2 : SDTypeProfile<2, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
SDTCisSameAs<0, 2>,
SDTCisSameAs<0, 3>]>;
def NEONzip : SDNode<"ARMISD::VZIP", SDTARMVSHUF2>;
def NEONuzp : SDNode<"ARMISD::VUZP", SDTARMVSHUF2>;
def NEONtrn : SDNode<"ARMISD::VTRN", SDTARMVSHUF2>;
def SDTARMVTBL1 : SDTypeProfile<1, 2, [SDTCisVT<0, v8i8>, SDTCisVT<1, v8i8>,
SDTCisVT<2, v8i8>]>;
def SDTARMVTBL2 : SDTypeProfile<1, 3, [SDTCisVT<0, v8i8>, SDTCisVT<1, v8i8>,
SDTCisVT<2, v8i8>, SDTCisVT<3, v8i8>]>;
def NEONvtbl1 : SDNode<"ARMISD::VTBL1", SDTARMVTBL1>;
def NEONvtbl2 : SDNode<"ARMISD::VTBL2", SDTARMVTBL2>;
//===----------------------------------------------------------------------===//
// NEON load / store instructions
//===----------------------------------------------------------------------===//
// Use VLDM to load a Q register as a D register pair.
// This is a pseudo instruction that is expanded to VLDMD after reg alloc.
def VLDMQIA
: PseudoVFPLdStM<(outs DPair:$dst), (ins GPR:$Rn),
IIC_fpLoad_m, "",
[(set DPair:$dst, (v2f64 (word_alignedload GPR:$Rn)))]>;
// Use VSTM to store a Q register as a D register pair.
// This is a pseudo instruction that is expanded to VSTMD after reg alloc.
def VSTMQIA
: PseudoVFPLdStM<(outs), (ins DPair:$src, GPR:$Rn),
IIC_fpStore_m, "",
[(word_alignedstore (v2f64 DPair:$src), GPR:$Rn)]>;
// Classes for VLD* pseudo-instructions with multi-register operands.
// These are expanded to real instructions after register allocation.
class VLDQPseudo<InstrItinClass itin>
: PseudoNLdSt<(outs QPR:$dst), (ins addrmode6:$addr), itin, "">;
class VLDQWBPseudo<InstrItinClass itin>
: PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
(ins addrmode6:$addr, am6offset:$offset), itin,
"$addr.addr = $wb">;
class VLDQWBfixedPseudo<InstrItinClass itin>
: PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
(ins addrmode6:$addr), itin,
"$addr.addr = $wb">;
class VLDQWBregisterPseudo<InstrItinClass itin>
: PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
(ins addrmode6:$addr, rGPR:$offset), itin,
"$addr.addr = $wb">;
class VLDQQPseudo<InstrItinClass itin>
: PseudoNLdSt<(outs QQPR:$dst), (ins addrmode6:$addr), itin, "">;
class VLDQQWBPseudo<InstrItinClass itin>
: PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
(ins addrmode6:$addr, am6offset:$offset), itin,
"$addr.addr = $wb">;
class VLDQQWBfixedPseudo<InstrItinClass itin>
: PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
(ins addrmode6:$addr), itin,
"$addr.addr = $wb">;
class VLDQQWBregisterPseudo<InstrItinClass itin>
: PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
(ins addrmode6:$addr, rGPR:$offset), itin,
"$addr.addr = $wb">;
class VLDQQQQPseudo<InstrItinClass itin>
: PseudoNLdSt<(outs QQQQPR:$dst), (ins addrmode6:$addr, QQQQPR:$src),itin,
"$src = $dst">;
class VLDQQQQWBPseudo<InstrItinClass itin>
: PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
(ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
"$addr.addr = $wb, $src = $dst">;
let mayLoad = 1, hasSideEffects = 0, hasExtraDefRegAllocReq = 1 in {
// VLD1 : Vector Load (multiple single elements)
class VLD1D<bits<4> op7_4, string Dt, Operand AddrMode>
: NLdSt<0,0b10,0b0111,op7_4, (outs VecListOneD:$Vd),
(ins AddrMode:$Rn), IIC_VLD1,
"vld1", Dt, "$Vd, $Rn", "", []>, Sched<[WriteVLD1]> {
let Rm = 0b1111;
let Inst{4} = Rn{4};
let DecoderMethod = "DecodeVLDST1Instruction";
}
class VLD1Q<bits<4> op7_4, string Dt, Operand AddrMode>
: NLdSt<0,0b10,0b1010,op7_4, (outs VecListDPair:$Vd),
(ins AddrMode:$Rn), IIC_VLD1x2,
"vld1", Dt, "$Vd, $Rn", "", []>, Sched<[WriteVLD2]> {
let Rm = 0b1111;
let Inst{5-4} = Rn{5-4};
let DecoderMethod = "DecodeVLDST1Instruction";
}
def VLD1d8 : VLD1D<{0,0,0,?}, "8", addrmode6align64>;
def VLD1d16 : VLD1D<{0,1,0,?}, "16", addrmode6align64>;
def VLD1d32 : VLD1D<{1,0,0,?}, "32", addrmode6align64>;
def VLD1d64 : VLD1D<{1,1,0,?}, "64", addrmode6align64>;
def VLD1q8 : VLD1Q<{0,0,?,?}, "8", addrmode6align64or128>;
def VLD1q16 : VLD1Q<{0,1,?,?}, "16", addrmode6align64or128>;
def VLD1q32 : VLD1Q<{1,0,?,?}, "32", addrmode6align64or128>;
def VLD1q64 : VLD1Q<{1,1,?,?}, "64", addrmode6align64or128>;
// ...with address register writeback:
multiclass VLD1DWB<bits<4> op7_4, string Dt, Operand AddrMode> {
def _fixed : NLdSt<0,0b10, 0b0111,op7_4, (outs VecListOneD:$Vd, GPR:$wb),
(ins AddrMode:$Rn), IIC_VLD1u,
"vld1", Dt, "$Vd, $Rn!",
"$Rn.addr = $wb", []>, Sched<[WriteVLD1]> {
let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
let Inst{4} = Rn{4};
let DecoderMethod = "DecodeVLDST1Instruction";
}
def _register : NLdSt<0,0b10,0b0111,op7_4, (outs VecListOneD:$Vd, GPR:$wb),
(ins AddrMode:$Rn, rGPR:$Rm), IIC_VLD1u,
"vld1", Dt, "$Vd, $Rn, $Rm",
"$Rn.addr = $wb", []>, Sched<[WriteVLD1]> {
let Inst{4} = Rn{4};
let DecoderMethod = "DecodeVLDST1Instruction";
}
}
multiclass VLD1QWB<bits<4> op7_4, string Dt, Operand AddrMode> {
def _fixed : NLdSt<0,0b10,0b1010,op7_4, (outs VecListDPair:$Vd, GPR:$wb),
(ins AddrMode:$Rn), IIC_VLD1x2u,
"vld1", Dt, "$Vd, $Rn!",
"$Rn.addr = $wb", []>, Sched<[WriteVLD2]> {
let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
let Inst{5-4} = Rn{5-4};
let DecoderMethod = "DecodeVLDST1Instruction";
}
def _register : NLdSt<0,0b10,0b1010,op7_4, (outs VecListDPair:$Vd, GPR:$wb),
(ins AddrMode:$Rn, rGPR:$Rm), IIC_VLD1x2u,
"vld1", Dt, "$Vd, $Rn, $Rm",
"$Rn.addr = $wb", []>, Sched<[WriteVLD2]> {
let Inst{5-4} = Rn{5-4};
let DecoderMethod = "DecodeVLDST1Instruction";
}
}
defm VLD1d8wb : VLD1DWB<{0,0,0,?}, "8", addrmode6align64>;
defm VLD1d16wb : VLD1DWB<{0,1,0,?}, "16", addrmode6align64>;
defm VLD1d32wb : VLD1DWB<{1,0,0,?}, "32", addrmode6align64>;
defm VLD1d64wb : VLD1DWB<{1,1,0,?}, "64", addrmode6align64>;
defm VLD1q8wb : VLD1QWB<{0,0,?,?}, "8", addrmode6align64or128>;
defm VLD1q16wb : VLD1QWB<{0,1,?,?}, "16", addrmode6align64or128>;
defm VLD1q32wb : VLD1QWB<{1,0,?,?}, "32", addrmode6align64or128>;
defm VLD1q64wb : VLD1QWB<{1,1,?,?}, "64", addrmode6align64or128>;
// ...with 3 registers
class VLD1D3<bits<4> op7_4, string Dt, Operand AddrMode>
: NLdSt<0,0b10,0b0110,op7_4, (outs VecListThreeD:$Vd),
(ins AddrMode:$Rn), IIC_VLD1x3, "vld1", Dt,
"$Vd, $Rn", "", []>, Sched<[WriteVLD3]> {
let Rm = 0b1111;
let Inst{4} = Rn{4};
let DecoderMethod = "DecodeVLDST1Instruction";
}
multiclass VLD1D3WB<bits<4> op7_4, string Dt, Operand AddrMode> {
def _fixed : NLdSt<0,0b10,0b0110, op7_4, (outs VecListThreeD:$Vd, GPR:$wb),
(ins AddrMode:$Rn), IIC_VLD1x2u,
"vld1", Dt, "$Vd, $Rn!",
"$Rn.addr = $wb", []>, Sched<[WriteVLD3]> {
let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
let Inst{4} = Rn{4};
let DecoderMethod = "DecodeVLDST1Instruction";
}
def _register : NLdSt<0,0b10,0b0110,op7_4, (outs VecListThreeD:$Vd, GPR:$wb),
(ins AddrMode:$Rn, rGPR:$Rm), IIC_VLD1x2u,
"vld1", Dt, "$Vd, $Rn, $Rm",
"$Rn.addr = $wb", []>, Sched<[WriteVLD3]> {
let Inst{4} = Rn{4};
let DecoderMethod = "DecodeVLDST1Instruction";
}
}
def VLD1d8T : VLD1D3<{0,0,0,?}, "8", addrmode6align64>;
def VLD1d16T : VLD1D3<{0,1,0,?}, "16", addrmode6align64>;
def VLD1d32T : VLD1D3<{1,0,0,?}, "32", addrmode6align64>;
def VLD1d64T : VLD1D3<{1,1,0,?}, "64", addrmode6align64>;
defm VLD1d8Twb : VLD1D3WB<{0,0,0,?}, "8", addrmode6align64>;
defm VLD1d16Twb : VLD1D3WB<{0,1,0,?}, "16", addrmode6align64>;
defm VLD1d32Twb : VLD1D3WB<{1,0,0,?}, "32", addrmode6align64>;
defm VLD1d64Twb : VLD1D3WB<{1,1,0,?}, "64", addrmode6align64>;
def VLD1d8TPseudo : VLDQQPseudo<IIC_VLD1x3>, Sched<[WriteVLD3]>;
def VLD1d8TPseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD1x3>, Sched<[WriteVLD3]>;
def VLD1d8TPseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD1x3>, Sched<[WriteVLD3]>;
def VLD1d16TPseudo : VLDQQPseudo<IIC_VLD1x3>, Sched<[WriteVLD3]>;
def VLD1d16TPseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD1x3>, Sched<[WriteVLD3]>;
def VLD1d16TPseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD1x3>, Sched<[WriteVLD3]>;
def VLD1d32TPseudo : VLDQQPseudo<IIC_VLD1x3>, Sched<[WriteVLD3]>;
def VLD1d32TPseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD1x3>, Sched<[WriteVLD3]>;
def VLD1d32TPseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD1x3>, Sched<[WriteVLD3]>;
def VLD1d64TPseudo : VLDQQPseudo<IIC_VLD1x3>, Sched<[WriteVLD3]>;
def VLD1d64TPseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD1x3>, Sched<[WriteVLD3]>;
def VLD1d64TPseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD1x3>, Sched<[WriteVLD3]>;
def VLD1q8HighTPseudo : VLDQQQQPseudo<IIC_VLD1x3>, Sched<[WriteVLD3]>;
def VLD1q8HighTPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD1x3>, Sched<[WriteVLD3]>;
def VLD1q8LowTPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD1x3>, Sched<[WriteVLD3]>;
def VLD1q16HighTPseudo : VLDQQQQPseudo<IIC_VLD1x3>, Sched<[WriteVLD3]>;
def VLD1q16HighTPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD1x3>, Sched<[WriteVLD3]>;
def VLD1q16LowTPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD1x3>, Sched<[WriteVLD3]>;
def VLD1q32HighTPseudo : VLDQQQQPseudo<IIC_VLD1x3>, Sched<[WriteVLD3]>;
def VLD1q32HighTPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD1x3>, Sched<[WriteVLD3]>;
def VLD1q32LowTPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD1x3>, Sched<[WriteVLD3]>;
def VLD1q64HighTPseudo : VLDQQQQPseudo<IIC_VLD1x3>, Sched<[WriteVLD3]>;
def VLD1q64HighTPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD1x3>, Sched<[WriteVLD3]>;
def VLD1q64LowTPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD1x3>, Sched<[WriteVLD3]>;
// ...with 4 registers
class VLD1D4<bits<4> op7_4, string Dt, Operand AddrMode>
: NLdSt<0, 0b10, 0b0010, op7_4, (outs VecListFourD:$Vd),
(ins AddrMode:$Rn), IIC_VLD1x4, "vld1", Dt,
"$Vd, $Rn", "", []>, Sched<[WriteVLD4]> {
let Rm = 0b1111;
let Inst{5-4} = Rn{5-4};
let DecoderMethod = "DecodeVLDST1Instruction";
}
multiclass VLD1D4WB<bits<4> op7_4, string Dt, Operand AddrMode> {
def _fixed : NLdSt<0,0b10,0b0010, op7_4, (outs VecListFourD:$Vd, GPR:$wb),
(ins AddrMode:$Rn), IIC_VLD1x2u,
"vld1", Dt, "$Vd, $Rn!",
"$Rn.addr = $wb", []>, Sched<[WriteVLD4]> {
let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
let Inst{5-4} = Rn{5-4};
let DecoderMethod = "DecodeVLDST1Instruction";
}
def _register : NLdSt<0,0b10,0b0010,op7_4, (outs VecListFourD:$Vd, GPR:$wb),
(ins AddrMode:$Rn, rGPR:$Rm), IIC_VLD1x2u,
"vld1", Dt, "$Vd, $Rn, $Rm",
"$Rn.addr = $wb", []>, Sched<[WriteVLD4]> {
let Inst{5-4} = Rn{5-4};
let DecoderMethod = "DecodeVLDST1Instruction";
}
}
def VLD1d8Q : VLD1D4<{0,0,?,?}, "8", addrmode6align64or128or256>;
def VLD1d16Q : VLD1D4<{0,1,?,?}, "16", addrmode6align64or128or256>;
def VLD1d32Q : VLD1D4<{1,0,?,?}, "32", addrmode6align64or128or256>;
def VLD1d64Q : VLD1D4<{1,1,?,?}, "64", addrmode6align64or128or256>;
defm VLD1d8Qwb : VLD1D4WB<{0,0,?,?}, "8", addrmode6align64or128or256>;
defm VLD1d16Qwb : VLD1D4WB<{0,1,?,?}, "16", addrmode6align64or128or256>;
defm VLD1d32Qwb : VLD1D4WB<{1,0,?,?}, "32", addrmode6align64or128or256>;
defm VLD1d64Qwb : VLD1D4WB<{1,1,?,?}, "64", addrmode6align64or128or256>;
def VLD1d8QPseudo : VLDQQPseudo<IIC_VLD1x4>, Sched<[WriteVLD4]>;
def VLD1d8QPseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD1x4>, Sched<[WriteVLD4]>;
def VLD1d8QPseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD1x4>, Sched<[WriteVLD4]>;
def VLD1d16QPseudo : VLDQQPseudo<IIC_VLD1x4>, Sched<[WriteVLD4]>;
def VLD1d16QPseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD1x4>, Sched<[WriteVLD4]>;
def VLD1d16QPseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD1x4>, Sched<[WriteVLD4]>;
def VLD1d32QPseudo : VLDQQPseudo<IIC_VLD1x4>, Sched<[WriteVLD4]>;
def VLD1d32QPseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD1x4>, Sched<[WriteVLD4]>;
def VLD1d32QPseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD1x4>, Sched<[WriteVLD4]>;
def VLD1d64QPseudo : VLDQQPseudo<IIC_VLD1x4>, Sched<[WriteVLD4]>;
def VLD1d64QPseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD1x4>, Sched<[WriteVLD4]>;
def VLD1d64QPseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD1x4>, Sched<[WriteVLD4]>;
def VLD1q8LowQPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD1x4>, Sched<[WriteVLD4]>;
def VLD1q8HighQPseudo : VLDQQQQPseudo<IIC_VLD1x4>, Sched<[WriteVLD4]>;
def VLD1q8HighQPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD1x4>, Sched<[WriteVLD4]>;
def VLD1q16LowQPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD1x4>, Sched<[WriteVLD4]>;
def VLD1q16HighQPseudo : VLDQQQQPseudo<IIC_VLD1x4>, Sched<[WriteVLD4]>;
def VLD1q16HighQPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD1x4>, Sched<[WriteVLD4]>;
def VLD1q32LowQPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD1x4>, Sched<[WriteVLD4]>;
def VLD1q32HighQPseudo : VLDQQQQPseudo<IIC_VLD1x4>, Sched<[WriteVLD4]>;
def VLD1q32HighQPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD1x4>, Sched<[WriteVLD4]>;
def VLD1q64LowQPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD1x4>, Sched<[WriteVLD4]>;
def VLD1q64HighQPseudo : VLDQQQQPseudo<IIC_VLD1x4>, Sched<[WriteVLD4]>;
def VLD1q64HighQPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD1x4>, Sched<[WriteVLD4]>;
// VLD2 : Vector Load (multiple 2-element structures)
class VLD2<bits<4> op11_8, bits<4> op7_4, string Dt, RegisterOperand VdTy,
InstrItinClass itin, Operand AddrMode>
: NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd),
(ins AddrMode:$Rn), itin,
"vld2", Dt, "$Vd, $Rn", "", []> {
let Rm = 0b1111;
let Inst{5-4} = Rn{5-4};
let DecoderMethod = "DecodeVLDST2Instruction";
}
def VLD2d8 : VLD2<0b1000, {0,0,?,?}, "8", VecListDPair, IIC_VLD2,
addrmode6align64or128>, Sched<[WriteVLD2]>;
def VLD2d16 : VLD2<0b1000, {0,1,?,?}, "16", VecListDPair, IIC_VLD2,
addrmode6align64or128>, Sched<[WriteVLD2]>;
def VLD2d32 : VLD2<0b1000, {1,0,?,?}, "32", VecListDPair, IIC_VLD2,
addrmode6align64or128>, Sched<[WriteVLD2]>;
def VLD2q8 : VLD2<0b0011, {0,0,?,?}, "8", VecListFourD, IIC_VLD2x2,
addrmode6align64or128or256>, Sched<[WriteVLD4]>;
def VLD2q16 : VLD2<0b0011, {0,1,?,?}, "16", VecListFourD, IIC_VLD2x2,
addrmode6align64or128or256>, Sched<[WriteVLD4]>;
def VLD2q32 : VLD2<0b0011, {1,0,?,?}, "32", VecListFourD, IIC_VLD2x2,
addrmode6align64or128or256>, Sched<[WriteVLD4]>;
def VLD2q8Pseudo : VLDQQPseudo<IIC_VLD2x2>, Sched<[WriteVLD4]>;
def VLD2q16Pseudo : VLDQQPseudo<IIC_VLD2x2>, Sched<[WriteVLD4]>;
def VLD2q32Pseudo : VLDQQPseudo<IIC_VLD2x2>, Sched<[WriteVLD4]>;
// ...with address register writeback:
multiclass VLD2WB<bits<4> op11_8, bits<4> op7_4, string Dt,
RegisterOperand VdTy, InstrItinClass itin, Operand AddrMode> {
def _fixed : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd, GPR:$wb),
(ins AddrMode:$Rn), itin,
"vld2", Dt, "$Vd, $Rn!",
"$Rn.addr = $wb", []> {
let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
let Inst{5-4} = Rn{5-4};
let DecoderMethod = "DecodeVLDST2Instruction";
}
def _register : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd, GPR:$wb),
(ins AddrMode:$Rn, rGPR:$Rm), itin,
"vld2", Dt, "$Vd, $Rn, $Rm",
"$Rn.addr = $wb", []> {
let Inst{5-4} = Rn{5-4};
let DecoderMethod = "DecodeVLDST2Instruction";
}
}
defm VLD2d8wb : VLD2WB<0b1000, {0,0,?,?}, "8", VecListDPair, IIC_VLD2u,
addrmode6align64or128>, Sched<[WriteVLD2]>;
defm VLD2d16wb : VLD2WB<0b1000, {0,1,?,?}, "16", VecListDPair, IIC_VLD2u,
addrmode6align64or128>, Sched<[WriteVLD2]>;
defm VLD2d32wb : VLD2WB<0b1000, {1,0,?,?}, "32", VecListDPair, IIC_VLD2u,
addrmode6align64or128>, Sched<[WriteVLD2]>;
defm VLD2q8wb : VLD2WB<0b0011, {0,0,?,?}, "8", VecListFourD, IIC_VLD2x2u,
addrmode6align64or128or256>, Sched<[WriteVLD4]>;
defm VLD2q16wb : VLD2WB<0b0011, {0,1,?,?}, "16", VecListFourD, IIC_VLD2x2u,
addrmode6align64or128or256>, Sched<[WriteVLD4]>;
defm VLD2q32wb : VLD2WB<0b0011, {1,0,?,?}, "32", VecListFourD, IIC_VLD2x2u,
addrmode6align64or128or256>, Sched<[WriteVLD4]>;
def VLD2q8PseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD2x2u>, Sched<[WriteVLD4]>;
def VLD2q16PseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD2x2u>, Sched<[WriteVLD4]>;
def VLD2q32PseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD2x2u>, Sched<[WriteVLD4]>;
def VLD2q8PseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD2x2u>, Sched<[WriteVLD4]>;
def VLD2q16PseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD2x2u>, Sched<[WriteVLD4]>;
def VLD2q32PseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD2x2u>, Sched<[WriteVLD4]>;
// ...with double-spaced registers
def VLD2b8 : VLD2<0b1001, {0,0,?,?}, "8", VecListDPairSpaced, IIC_VLD2,
addrmode6align64or128>, Sched<[WriteVLD2]>;
def VLD2b16 : VLD2<0b1001, {0,1,?,?}, "16", VecListDPairSpaced, IIC_VLD2,
addrmode6align64or128>, Sched<[WriteVLD2]>;
def VLD2b32 : VLD2<0b1001, {1,0,?,?}, "32", VecListDPairSpaced, IIC_VLD2,
addrmode6align64or128>, Sched<[WriteVLD2]>;
defm VLD2b8wb : VLD2WB<0b1001, {0,0,?,?}, "8", VecListDPairSpaced, IIC_VLD2u,
addrmode6align64or128>, Sched<[WriteVLD2]>;
defm VLD2b16wb : VLD2WB<0b1001, {0,1,?,?}, "16", VecListDPairSpaced, IIC_VLD2u,
addrmode6align64or128>, Sched<[WriteVLD2]>;
defm VLD2b32wb : VLD2WB<0b1001, {1,0,?,?}, "32", VecListDPairSpaced, IIC_VLD2u,
addrmode6align64or128>, Sched<[WriteVLD2]>;
// VLD3 : Vector Load (multiple 3-element structures)
class VLD3D<bits<4> op11_8, bits<4> op7_4, string Dt>
: NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
(ins addrmode6:$Rn), IIC_VLD3,
"vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn", "", []>, Sched<[WriteVLD3]> {
let Rm = 0b1111;
let Inst{4} = Rn{4};
let DecoderMethod = "DecodeVLDST3Instruction";
}
def VLD3d8 : VLD3D<0b0100, {0,0,0,?}, "8">;
def VLD3d16 : VLD3D<0b0100, {0,1,0,?}, "16">;
def VLD3d32 : VLD3D<0b0100, {1,0,0,?}, "32">;
def VLD3d8Pseudo : VLDQQPseudo<IIC_VLD3>, Sched<[WriteVLD3]>;
def VLD3d16Pseudo : VLDQQPseudo<IIC_VLD3>, Sched<[WriteVLD3]>;
def VLD3d32Pseudo : VLDQQPseudo<IIC_VLD3>, Sched<[WriteVLD3]>;
// ...with address register writeback:
class VLD3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
: NLdSt<0, 0b10, op11_8, op7_4,
(outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
(ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD3u,
"vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn$Rm",
"$Rn.addr = $wb", []>, Sched<[WriteVLD3]> {
let Inst{4} = Rn{4};
let DecoderMethod = "DecodeVLDST3Instruction";
}
def VLD3d8_UPD : VLD3DWB<0b0100, {0,0,0,?}, "8">;
def VLD3d16_UPD : VLD3DWB<0b0100, {0,1,0,?}, "16">;
def VLD3d32_UPD : VLD3DWB<0b0100, {1,0,0,?}, "32">;
def VLD3d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>, Sched<[WriteVLD3]>;
def VLD3d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>, Sched<[WriteVLD3]>;
def VLD3d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>, Sched<[WriteVLD3]>;
// ...with double-spaced registers:
def VLD3q8 : VLD3D<0b0101, {0,0,0,?}, "8">;
def VLD3q16 : VLD3D<0b0101, {0,1,0,?}, "16">;
def VLD3q32 : VLD3D<0b0101, {1,0,0,?}, "32">;
def VLD3q8_UPD : VLD3DWB<0b0101, {0,0,0,?}, "8">;
def VLD3q16_UPD : VLD3DWB<0b0101, {0,1,0,?}, "16">;
def VLD3q32_UPD : VLD3DWB<0b0101, {1,0,0,?}, "32">;
def VLD3q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>, Sched<[WriteVLD3]>;
def VLD3q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>, Sched<[WriteVLD3]>;
def VLD3q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>, Sched<[WriteVLD3]>;
// ...alternate versions to be allocated odd register numbers:
def VLD3q8oddPseudo : VLDQQQQPseudo<IIC_VLD3>, Sched<[WriteVLD3]>;
def VLD3q16oddPseudo : VLDQQQQPseudo<IIC_VLD3>, Sched<[WriteVLD3]>;
def VLD3q32oddPseudo : VLDQQQQPseudo<IIC_VLD3>, Sched<[WriteVLD3]>;
def VLD3q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>, Sched<[WriteVLD3]>;
def VLD3q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>, Sched<[WriteVLD3]>;
def VLD3q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>, Sched<[WriteVLD3]>;
// VLD4 : Vector Load (multiple 4-element structures)
class VLD4D<bits<4> op11_8, bits<4> op7_4, string Dt>
: NLdSt<0, 0b10, op11_8, op7_4,
(outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
(ins addrmode6:$Rn), IIC_VLD4,
"vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []>,
Sched<[WriteVLD4]> {
let Rm = 0b1111;
let Inst{5-4} = Rn{5-4};
let DecoderMethod = "DecodeVLDST4Instruction";
}
def VLD4d8 : VLD4D<0b0000, {0,0,?,?}, "8">;
def VLD4d16 : VLD4D<0b0000, {0,1,?,?}, "16">;
def VLD4d32 : VLD4D<0b0000, {1,0,?,?}, "32">;
def VLD4d8Pseudo : VLDQQPseudo<IIC_VLD4>, Sched<[WriteVLD4]>;
def VLD4d16Pseudo : VLDQQPseudo<IIC_VLD4>, Sched<[WriteVLD4]>;
def VLD4d32Pseudo : VLDQQPseudo<IIC_VLD4>, Sched<[WriteVLD4]>;
// ...with address register writeback:
class VLD4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
: NLdSt<0, 0b10, op11_8, op7_4,
(outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
(ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD4u,
"vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm",
"$Rn.addr = $wb", []>, Sched<[WriteVLD4]> {
let Inst{5-4} = Rn{5-4};
let DecoderMethod = "DecodeVLDST4Instruction";
}
def VLD4d8_UPD : VLD4DWB<0b0000, {0,0,?,?}, "8">;
def VLD4d16_UPD : VLD4DWB<0b0000, {0,1,?,?}, "16">;
def VLD4d32_UPD : VLD4DWB<0b0000, {1,0,?,?}, "32">;
def VLD4d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>, Sched<[WriteVLD4]>;
def VLD4d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>, Sched<[WriteVLD4]>;
def VLD4d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>, Sched<[WriteVLD4]>;
// ...with double-spaced registers:
def VLD4q8 : VLD4D<0b0001, {0,0,?,?}, "8">;
def VLD4q16 : VLD4D<0b0001, {0,1,?,?}, "16">;
def VLD4q32 : VLD4D<0b0001, {1,0,?,?}, "32">;
def VLD4q8_UPD : VLD4DWB<0b0001, {0,0,?,?}, "8">;
def VLD4q16_UPD : VLD4DWB<0b0001, {0,1,?,?}, "16">;
def VLD4q32_UPD : VLD4DWB<0b0001, {1,0,?,?}, "32">;
def VLD4q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>, Sched<[WriteVLD4]>;
def VLD4q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>, Sched<[WriteVLD4]>;
def VLD4q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>, Sched<[WriteVLD4]>;
// ...alternate versions to be allocated odd register numbers:
def VLD4q8oddPseudo : VLDQQQQPseudo<IIC_VLD4>, Sched<[WriteVLD4]>;
def VLD4q16oddPseudo : VLDQQQQPseudo<IIC_VLD4>, Sched<[WriteVLD4]>;
def VLD4q32oddPseudo : VLDQQQQPseudo<IIC_VLD4>, Sched<[WriteVLD4]>;
def VLD4q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>, Sched<[WriteVLD4]>;
def VLD4q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>, Sched<[WriteVLD4]>;
def VLD4q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>, Sched<[WriteVLD4]>;
} // mayLoad = 1, hasSideEffects = 0, hasExtraDefRegAllocReq = 1
// Classes for VLD*LN pseudo-instructions with multi-register operands.
// These are expanded to real instructions after register allocation.
class VLDQLNPseudo<InstrItinClass itin>
: PseudoNLdSt<(outs QPR:$dst),
(ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
itin, "$src = $dst">;
class VLDQLNWBPseudo<InstrItinClass itin>
: PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
(ins addrmode6:$addr, am6offset:$offset, QPR:$src,
nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
class VLDQQLNPseudo<InstrItinClass itin>
: PseudoNLdSt<(outs QQPR:$dst),
(ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
itin, "$src = $dst">;
class VLDQQLNWBPseudo<InstrItinClass itin>
: PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
(ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
class VLDQQQQLNPseudo<InstrItinClass itin>
: PseudoNLdSt<(outs QQQQPR:$dst),
(ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
itin, "$src = $dst">;
class VLDQQQQLNWBPseudo<InstrItinClass itin>
: PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
(ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
// VLD1LN : Vector Load (single element to one lane)
class VLD1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
PatFrag LoadOp>
: NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
(ins addrmode6:$Rn, DPR:$src, nohash_imm:$lane),
IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
"$src = $Vd",
[(set DPR:$Vd, (vector_insert (Ty DPR:$src),
(i32 (LoadOp addrmode6:$Rn)),
imm:$lane))]> {
let Rm = 0b1111;
let DecoderMethod = "DecodeVLD1LN";
}
class VLD1LN32<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
PatFrag LoadOp>
: NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
(ins addrmode6oneL32:$Rn, DPR:$src, nohash_imm:$lane),
IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
"$src = $Vd",
[(set DPR:$Vd, (vector_insert (Ty DPR:$src),
(i32 (LoadOp addrmode6oneL32:$Rn)),
imm:$lane))]>, Sched<[WriteVLD1]> {
let Rm = 0b1111;
let DecoderMethod = "DecodeVLD1LN";
}
class VLD1QLNPseudo<ValueType Ty, PatFrag LoadOp> : VLDQLNPseudo<IIC_VLD1ln>,
Sched<[WriteVLD1]> {
let Pattern = [(set QPR:$dst, (vector_insert (Ty QPR:$src),
(i32 (LoadOp addrmode6:$addr)),
imm:$lane))];
}
def VLD1LNd8 : VLD1LN<0b0000, {?,?,?,0}, "8", v8i8, extloadi8> {
let Inst{7-5} = lane{2-0};
}
def VLD1LNd16 : VLD1LN<0b0100, {?,?,0,?}, "16", v4i16, extloadi16> {
let Inst{7-6} = lane{1-0};
let Inst{5-4} = Rn{5-4};
}
def VLD1LNd32 : VLD1LN32<0b1000, {?,0,?,?}, "32", v2i32, load> {
let Inst{7} = lane{0};
let Inst{5-4} = Rn{5-4};
}
def VLD1LNq8Pseudo : VLD1QLNPseudo<v16i8, extloadi8>;
def VLD1LNq16Pseudo : VLD1QLNPseudo<v8i16, extloadi16>;
def VLD1LNq32Pseudo : VLD1QLNPseudo<v4i32, load>;
let Predicates = [HasNEON] in {
def : Pat<(vector_insert (v4f16 DPR:$src),
(f16 (load addrmode6:$addr)), imm:$lane),
(VLD1LNd16 addrmode6:$addr, DPR:$src, imm:$lane)>;
def : Pat<(vector_insert (v8f16 QPR:$src),
(f16 (load addrmode6:$addr)), imm:$lane),
(VLD1LNq16Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
def : Pat<(vector_insert (v4bf16 DPR:$src),
(bf16 (load addrmode6:$addr)), imm:$lane),
(VLD1LNd16 addrmode6:$addr, DPR:$src, imm:$lane)>;
def : Pat<(vector_insert (v8bf16 QPR:$src),
(bf16 (load addrmode6:$addr)), imm:$lane),
(VLD1LNq16Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
def : Pat<(vector_insert (v2f32 DPR:$src),
(f32 (load addrmode6:$addr)), imm:$lane),
(VLD1LNd32 addrmode6:$addr, DPR:$src, imm:$lane)>;
def : Pat<(vector_insert (v4f32 QPR:$src),
(f32 (load addrmode6:$addr)), imm:$lane),
(VLD1LNq32Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
// A 64-bit subvector insert to the first 128-bit vector position
// is a subregister copy that needs no instruction.
def : Pat<(insert_subvector undef, (v1i64 DPR:$src), (i32 0)),
(INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
def : Pat<(insert_subvector undef, (v2i32 DPR:$src), (i32 0)),
(INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
def : Pat<(insert_subvector undef, (v2f32 DPR:$src), (i32 0)),
(INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
def : Pat<(insert_subvector undef, (v4i16 DPR:$src), (i32 0)),
(INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
def : Pat<(insert_subvector undef, (v4f16 DPR:$src), (i32 0)),
(INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
def : Pat<(insert_subvector (v16i8 undef), (v8i8 DPR:$src), (i32 0)),
(INSERT_SUBREG (v16i8 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
}
let mayLoad = 1, hasSideEffects = 0, hasExtraDefRegAllocReq = 1 in {
// ...with address register writeback:
class VLD1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
: NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, GPR:$wb),
(ins addrmode6:$Rn, am6offset:$Rm,
DPR:$src, nohash_imm:$lane), IIC_VLD1lnu, "vld1", Dt,
"\\{$Vd[$lane]\\}, $Rn$Rm",
"$src = $Vd, $Rn.addr = $wb", []>, Sched<[WriteVLD1]> {
let DecoderMethod = "DecodeVLD1LN";
}
def VLD1LNd8_UPD : VLD1LNWB<0b0000, {?,?,?,0}, "8"> {
let Inst{7-5} = lane{2-0};
}
def VLD1LNd16_UPD : VLD1LNWB<0b0100, {?,?,0,?}, "16"> {
let Inst{7-6} = lane{1-0};
let Inst{4} = Rn{4};
}
def VLD1LNd32_UPD : VLD1LNWB<0b1000, {?,0,?,?}, "32"> {
let Inst{7} = lane{0};
let Inst{5} = Rn{4};
let Inst{4} = Rn{4};
}
def VLD1LNq8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>, Sched<[WriteVLD1]>;
def VLD1LNq16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>, Sched<[WriteVLD1]>;
def VLD1LNq32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>, Sched<[WriteVLD1]>;
// VLD2LN : Vector Load (single 2-element structure to one lane)
class VLD2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
: NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2),
(ins addrmode6:$Rn, DPR:$src1, DPR:$src2, nohash_imm:$lane),
IIC_VLD2ln, "vld2", Dt, "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn",
"$src1 = $Vd, $src2 = $dst2", []>, Sched<[WriteVLD1]> {
let Rm = 0b1111;
let Inst{4} = Rn{4};
let DecoderMethod = "DecodeVLD2LN";
}
def VLD2LNd8 : VLD2LN<0b0001, {?,?,?,?}, "8"> {
let Inst{7-5} = lane{2-0};
}
def VLD2LNd16 : VLD2LN<0b0101, {?,?,0,?}, "16"> {
let Inst{7-6} = lane{1-0};
}
def VLD2LNd32 : VLD2LN<0b1001, {?,0,0,?}, "32"> {
let Inst{7} = lane{0};
}
def VLD2LNd8Pseudo : VLDQLNPseudo<IIC_VLD2ln>, Sched<[WriteVLD1]>;
def VLD2LNd16Pseudo : VLDQLNPseudo<IIC_VLD2ln>, Sched<[WriteVLD1]>;
def VLD2LNd32Pseudo : VLDQLNPseudo<IIC_VLD2ln>, Sched<[WriteVLD1]>;
// ...with double-spaced registers:
def VLD2LNq16 : VLD2LN<0b0101, {?,?,1,?}, "16"> {
let Inst{7-6} = lane{1-0};
}
def VLD2LNq32 : VLD2LN<0b1001, {?,1,0,?}, "32"> {
let Inst{7} = lane{0};
}
def VLD2LNq16Pseudo : VLDQQLNPseudo<IIC_VLD2ln>, Sched<[WriteVLD1]>;
def VLD2LNq32Pseudo : VLDQQLNPseudo<IIC_VLD2ln>, Sched<[WriteVLD1]>;
// ...with address register writeback:
class VLD2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
: NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
(ins addrmode6:$Rn, am6offset:$Rm,
DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VLD2lnu, "vld2", Dt,
"\\{$Vd[$lane], $dst2[$lane]\\}, $Rn$Rm",
"$src1 = $Vd, $src2 = $dst2, $Rn.addr = $wb", []> {
let Inst{4} = Rn{4};
let DecoderMethod = "DecodeVLD2LN";
}
def VLD2LNd8_UPD : VLD2LNWB<0b0001, {?,?,?,?}, "8"> {
let Inst{7-5} = lane{2-0};
}
def VLD2LNd16_UPD : VLD2LNWB<0b0101, {?,?,0,?}, "16"> {
let Inst{7-6} = lane{1-0};
}
def VLD2LNd32_UPD : VLD2LNWB<0b1001, {?,0,0,?}, "32"> {
let Inst{7} = lane{0};
}
def VLD2LNd8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>, Sched<[WriteVLD1]>;
def VLD2LNd16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>, Sched<[WriteVLD1]>;
def VLD2LNd32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>, Sched<[WriteVLD1]>;
def VLD2LNq16_UPD : VLD2LNWB<0b0101, {?,?,1,?}, "16"> {
let Inst{7-6} = lane{1-0};
}
def VLD2LNq32_UPD : VLD2LNWB<0b1001, {?,1,0,?}, "32"> {
let Inst{7} = lane{0};
}
def VLD2LNq16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>, Sched<[WriteVLD1]>;
def VLD2LNq32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>, Sched<[WriteVLD1]>;
// VLD3LN : Vector Load (single 3-element structure to one lane)
class VLD3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
: NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
(ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3,
nohash_imm:$lane), IIC_VLD3ln, "vld3", Dt,
"\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn",
"$src1 = $Vd, $src2 = $dst2, $src3 = $dst3", []>, Sched<[WriteVLD2]> {
let Rm = 0b1111;
let DecoderMethod = "DecodeVLD3LN";
}
def VLD3LNd8 : VLD3LN<0b0010, {?,?,?,0}, "8"> {
let Inst{7-5} = lane{2-0};
}
def VLD3LNd16 : VLD3LN<0b0110, {?,?,0,0}, "16"> {
let Inst{7-6} = lane{1-0};
}
def VLD3LNd32 : VLD3LN<0b1010, {?,0,0,0}, "32"> {
let Inst{7} = lane{0};
}
def VLD3LNd8Pseudo : VLDQQLNPseudo<IIC_VLD3ln>, Sched<[WriteVLD2]>;
def VLD3LNd16Pseudo : VLDQQLNPseudo<IIC_VLD3ln>, Sched<[WriteVLD2]>;
def VLD3LNd32Pseudo : VLDQQLNPseudo<IIC_VLD3ln>, Sched<[WriteVLD2]>;
// ...with double-spaced registers:
def VLD3LNq16 : VLD3LN<0b0110, {?,?,1,0}, "16"> {
let Inst{7-6} = lane{1-0};
}
def VLD3LNq32 : VLD3LN<0b1010, {?,1,0,0}, "32"> {
let Inst{7} = lane{0};
}
def VLD3LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>, Sched<[WriteVLD2]>;
def VLD3LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>, Sched<[WriteVLD2]>;
// ...with address register writeback:
class VLD3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
: NLdStLn<1, 0b10, op11_8, op7_4,
(outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
(ins addrmode6:$Rn, am6offset:$Rm,
DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
IIC_VLD3lnu, "vld3", Dt,
"\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn$Rm",
"$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $Rn.addr = $wb",
[]>, Sched<[WriteVLD2]> {
let DecoderMethod = "DecodeVLD3LN";
}
def VLD3LNd8_UPD : VLD3LNWB<0b0010, {?,?,?,0}, "8"> {
let Inst{7-5} = lane{2-0};
}
def VLD3LNd16_UPD : VLD3LNWB<0b0110, {?,?,0,0}, "16"> {
let Inst{7-6} = lane{1-0};
}
def VLD3LNd32_UPD : VLD3LNWB<0b1010, {?,0,0,0}, "32"> {
let Inst{7} = lane{0};
}
def VLD3LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>, Sched<[WriteVLD2]>;
def VLD3LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>, Sched<[WriteVLD2]>;
def VLD3LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>, Sched<[WriteVLD2]>;
def VLD3LNq16_UPD : VLD3LNWB<0b0110, {?,?,1,0}, "16"> {
let Inst{7-6} = lane{1-0};
}
def VLD3LNq32_UPD : VLD3LNWB<0b1010, {?,1,0,0}, "32"> {
let Inst{7} = lane{0};
}
def VLD3LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>, Sched<[WriteVLD2]>;
def VLD3LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>, Sched<[WriteVLD2]>;
// VLD4LN : Vector Load (single 4-element structure to one lane)
class VLD4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
: NLdStLn<1, 0b10, op11_8, op7_4,
(outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
(ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
nohash_imm:$lane), IIC_VLD4ln, "vld4", Dt,
"\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn",
"$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4", []>,
Sched<[WriteVLD2]> {
let Rm = 0b1111;
let Inst{4} = Rn{4};
let DecoderMethod = "DecodeVLD4LN";
}
def VLD4LNd8 : VLD4LN<0b0011, {?,?,?,?}, "8"> {
let Inst{7-5} = lane{2-0};
}
def VLD4LNd16 : VLD4LN<0b0111, {?,?,0,?}, "16"> {
let Inst{7-6} = lane{1-0};
}
def VLD4LNd32 : VLD4LN<0b1011, {?,0,?,?}, "32"> {
let Inst{7} = lane{0};
let Inst{5} = Rn{5};
}
def VLD4LNd8Pseudo : VLDQQLNPseudo<IIC_VLD4ln>, Sched<[WriteVLD2]>;
def VLD4LNd16Pseudo : VLDQQLNPseudo<IIC_VLD4ln>, Sched<[WriteVLD2]>;
def VLD4LNd32Pseudo : VLDQQLNPseudo<IIC_VLD4ln>, Sched<[WriteVLD2]>;
// ...with double-spaced registers:
def VLD4LNq16 : VLD4LN<0b0111, {?,?,1,?}, "16"> {
let Inst{7-6} = lane{1-0};
}
def VLD4LNq32 : VLD4LN<0b1011, {?,1,?,?}, "32"> {
let Inst{7} = lane{0};
let Inst{5} = Rn{5};
}
def VLD4LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>, Sched<[WriteVLD2]>;
def VLD4LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>, Sched<[WriteVLD2]>;
// ...with address register writeback:
class VLD4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
: NLdStLn<1, 0b10, op11_8, op7_4,
(outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
(ins addrmode6:$Rn, am6offset:$Rm,
DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
IIC_VLD4lnu, "vld4", Dt,
"\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn$Rm",
"$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4, $Rn.addr = $wb",
[]> {
let Inst{4} = Rn{4};
let DecoderMethod = "DecodeVLD4LN" ;
}
def VLD4LNd8_UPD : VLD4LNWB<0b0011, {?,?,?,?}, "8"> {
let Inst{7-5} = lane{2-0};
}
def VLD4LNd16_UPD : VLD4LNWB<0b0111, {?,?,0,?}, "16"> {
let Inst{7-6} = lane{1-0};
}
def VLD4LNd32_UPD : VLD4LNWB<0b1011, {?,0,?,?}, "32"> {
let Inst{7} = lane{0};
let Inst{5} = Rn{5};
}
def VLD4LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>, Sched<[WriteVLD2]>;
def VLD4LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>, Sched<[WriteVLD2]>;
def VLD4LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>, Sched<[WriteVLD2]>;
def VLD4LNq16_UPD : VLD4LNWB<0b0111, {?,?,1,?}, "16"> {
let Inst{7-6} = lane{1-0};
}
def VLD4LNq32_UPD : VLD4LNWB<0b1011, {?,1,?,?}, "32"> {
let Inst{7} = lane{0};
let Inst{5} = Rn{5};
}
def VLD4LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>, Sched<[WriteVLD2]>;
def VLD4LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>, Sched<[WriteVLD2]>;
} // mayLoad = 1, hasSideEffects = 0, hasExtraDefRegAllocReq = 1
// VLD1DUP : Vector Load (single element to all lanes)
class VLD1DUP<bits<4> op7_4, string Dt, ValueType Ty, PatFrag LoadOp,
Operand AddrMode>
: NLdSt<1, 0b10, 0b1100, op7_4, (outs VecListOneDAllLanes:$Vd),
(ins AddrMode:$Rn),
IIC_VLD1dup, "vld1", Dt, "$Vd, $Rn", "",
[(set VecListOneDAllLanes:$Vd,
(Ty (ARMvdup (i32 (LoadOp AddrMode:$Rn)))))]>,
Sched<[WriteVLD2]> {
let Rm = 0b1111;
let Inst{4} = Rn{4};
let DecoderMethod = "DecodeVLD1DupInstruction";
}
def VLD1DUPd8 : VLD1DUP<{0,0,0,?}, "8", v8i8, extloadi8,
addrmode6dupalignNone>;
def VLD1DUPd16 : VLD1DUP<{0,1,0,?}, "16", v4i16, extloadi16,
addrmode6dupalign16>;
def VLD1DUPd32 : VLD1DUP<{1,0,0,?}, "32", v2i32, load,
addrmode6dupalign32>;
let Predicates = [HasNEON] in {
def : Pat<(v2f32 (ARMvdup (f32 (load addrmode6dup:$addr)))),
(VLD1DUPd32 addrmode6:$addr)>;
}
class VLD1QDUP<bits<4> op7_4, string Dt, ValueType Ty, PatFrag LoadOp,
Operand AddrMode>
: NLdSt<1, 0b10, 0b1100, op7_4, (outs VecListDPairAllLanes:$Vd),
(ins AddrMode:$Rn), IIC_VLD1dup,
"vld1", Dt, "$Vd, $Rn", "",
[(set VecListDPairAllLanes:$Vd,
(Ty (ARMvdup (i32 (LoadOp AddrMode:$Rn)))))]> {
let Rm = 0b1111;
let Inst{4} = Rn{4};
let DecoderMethod = "DecodeVLD1DupInstruction";
}
def VLD1DUPq8 : VLD1QDUP<{0,0,1,0}, "8", v16i8, extloadi8,
addrmode6dupalignNone>;
def VLD1DUPq16 : VLD1QDUP<{0,1,1,?}, "16", v8i16, extloadi16,
addrmode6dupalign16>;
def VLD1DUPq32 : VLD1QDUP<{1,0,1,?}, "32", v4i32, load,
addrmode6dupalign32>;
let Predicates = [HasNEON] in {
def : Pat<(v4f32 (ARMvdup (f32 (load addrmode6dup:$addr)))),
(VLD1DUPq32 addrmode6:$addr)>;
}
let mayLoad = 1, hasSideEffects = 0, hasExtraDefRegAllocReq = 1 in {
// ...with address register writeback:
multiclass VLD1DUPWB<bits<4> op7_4, string Dt, Operand AddrMode> {
def _fixed : NLdSt<1, 0b10, 0b1100, op7_4,
(outs VecListOneDAllLanes:$Vd, GPR:$wb),
(ins AddrMode:$Rn), IIC_VLD1dupu,
"vld1", Dt, "$Vd, $Rn!",
"$Rn.addr = $wb", []> {
let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
let Inst{4} = Rn{4};
let DecoderMethod = "DecodeVLD1DupInstruction";
}
def _register : NLdSt<1, 0b10, 0b1100, op7_4,
(outs VecListOneDAllLanes:$Vd, GPR:$wb),
(ins AddrMode:$Rn, rGPR:$Rm), IIC_VLD1dupu,
"vld1", Dt, "$Vd, $Rn, $Rm",
"$Rn.addr = $wb", []> {
let Inst{4} = Rn{4};
let DecoderMethod = "DecodeVLD1DupInstruction";
}
}
multiclass VLD1QDUPWB<bits<4> op7_4, string Dt, Operand AddrMode> {
def _fixed : NLdSt<1, 0b10, 0b1100, op7_4,
(outs VecListDPairAllLanes:$Vd, GPR:$wb),
(ins AddrMode:$Rn), IIC_VLD1dupu,
"vld1", Dt, "$Vd, $Rn!",
"$Rn.addr = $wb", []>, Sched<[WriteVLD1]> {
let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
let Inst{4} = Rn{4};
let DecoderMethod = "DecodeVLD1DupInstruction";
}
def _register : NLdSt<1, 0b10, 0b1100, op7_4,
(outs VecListDPairAllLanes:$Vd, GPR:$wb),
(ins AddrMode:$Rn, rGPR:$Rm), IIC_VLD1dupu,
"vld1", Dt, "$Vd, $Rn, $Rm",
"$Rn.addr = $wb", []> {
let Inst{4} = Rn{4};
let DecoderMethod = "DecodeVLD1DupInstruction";
}
}
defm VLD1DUPd8wb : VLD1DUPWB<{0,0,0,0}, "8", addrmode6dupalignNone>;
defm VLD1DUPd16wb : VLD1DUPWB<{0,1,0,?}, "16", addrmode6dupalign16>;
defm VLD1DUPd32wb : VLD1DUPWB<{1,0,0,?}, "32", addrmode6dupalign32>;
defm VLD1DUPq8wb : VLD1QDUPWB<{0,0,1,0}, "8", addrmode6dupalignNone>;
defm VLD1DUPq16wb : VLD1QDUPWB<{0,1,1,?}, "16", addrmode6dupalign16>;
defm VLD1DUPq32wb : VLD1QDUPWB<{1,0,1,?}, "32", addrmode6dupalign32>;
// VLD2DUP : Vector Load (single 2-element structure to all lanes)
class VLD2DUP<bits<4> op7_4, string Dt, RegisterOperand VdTy, Operand AddrMode>
: NLdSt<1, 0b10, 0b1101, op7_4, (outs VdTy:$Vd),
(ins AddrMode:$Rn), IIC_VLD2dup,
"vld2", Dt, "$Vd, $Rn", "", []> {
let Rm = 0b1111;
let Inst{4} = Rn{4};
let DecoderMethod = "DecodeVLD2DupInstruction";
}
def VLD2DUPd8 : VLD2DUP<{0,0,0,?}, "8", VecListDPairAllLanes,
addrmode6dupalign16>;
def VLD2DUPd16 : VLD2DUP<{0,1,0,?}, "16", VecListDPairAllLanes,
addrmode6dupalign32>;
def VLD2DUPd32 : VLD2DUP<{1,0,0,?}, "32", VecListDPairAllLanes,
addrmode6dupalign64>;
// HACK this one, VLD2DUPd8x2 must be changed at the same time with VLD2b8 or
// "vld2.8 {d0[], d2[]}, [r4:32]" will become "vld2.8 {d0, d2}, [r4:32]".
// ...with double-spaced registers
def VLD2DUPd8x2 : VLD2DUP<{0,0,1,?}, "8", VecListDPairSpacedAllLanes,
addrmode6dupalign16>;
def VLD2DUPd16x2 : VLD2DUP<{0,1,1,?}, "16", VecListDPairSpacedAllLanes,
addrmode6dupalign32>;
def VLD2DUPd32x2 : VLD2DUP<{1,0,1,?}, "32", VecListDPairSpacedAllLanes,
addrmode6dupalign64>;
def VLD2DUPq8EvenPseudo : VLDQQPseudo<IIC_VLD2dup>, Sched<[WriteVLD2]>;
def VLD2DUPq8OddPseudo : VLDQQPseudo<IIC_VLD2dup>, Sched<[WriteVLD2]>;
def VLD2DUPq16EvenPseudo : VLDQQPseudo<IIC_VLD2dup>, Sched<[WriteVLD2]>;
def VLD2DUPq16OddPseudo : VLDQQPseudo<IIC_VLD2dup>, Sched<[WriteVLD2]>;
def VLD2DUPq32EvenPseudo : VLDQQPseudo<IIC_VLD2dup>, Sched<[WriteVLD2]>;
def VLD2DUPq32OddPseudo : VLDQQPseudo<IIC_VLD2dup>, Sched<[WriteVLD2]>;
// ...with address register writeback:
multiclass VLD2DUPWB<bits<4> op7_4, string Dt, RegisterOperand VdTy,
Operand AddrMode> {
def _fixed : NLdSt<1, 0b10, 0b1101, op7_4,
(outs VdTy:$Vd, GPR:$wb),
(ins AddrMode:$Rn), IIC_VLD2dupu,
"vld2", Dt, "$Vd, $Rn!",
"$Rn.addr = $wb", []>, Sched<[WriteVLD1]> {
let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
let Inst{4} = Rn{4};
let DecoderMethod = "DecodeVLD2DupInstruction";
}
def _register : NLdSt<1, 0b10, 0b1101, op7_4,
(outs VdTy:$Vd, GPR:$wb),
(ins AddrMode:$Rn, rGPR:$Rm), IIC_VLD2dupu,
"vld2", Dt, "$Vd, $Rn, $Rm",
"$Rn.addr = $wb", []>, Sched<[WriteVLD1]> {
let Inst{4} = Rn{4};
let DecoderMethod = "DecodeVLD2DupInstruction";
}
}
defm VLD2DUPd8wb : VLD2DUPWB<{0,0,0,0}, "8", VecListDPairAllLanes,
addrmode6dupalign16>;
defm VLD2DUPd16wb : VLD2DUPWB<{0,1,0,?}, "16", VecListDPairAllLanes,
addrmode6dupalign32>;
defm VLD2DUPd32wb : VLD2DUPWB<{1,0,0,?}, "32", VecListDPairAllLanes,
addrmode6dupalign64>;
defm VLD2DUPd8x2wb : VLD2DUPWB<{0,0,1,0}, "8", VecListDPairSpacedAllLanes,
addrmode6dupalign16>;
defm VLD2DUPd16x2wb : VLD2DUPWB<{0,1,1,?}, "16", VecListDPairSpacedAllLanes,
addrmode6dupalign32>;
defm VLD2DUPd32x2wb : VLD2DUPWB<{1,0,1,?}, "32", VecListDPairSpacedAllLanes,
addrmode6dupalign64>;
def VLD2DUPq8OddPseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD2dup>, Sched<[WriteVLD2]>;
def VLD2DUPq16OddPseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD2dup>, Sched<[WriteVLD2]>;
def VLD2DUPq32OddPseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD2dup>, Sched<[WriteVLD2]>;
def VLD2DUPq8OddPseudoWB_register : VLDQQWBPseudo<IIC_VLD2dup>, Sched<[WriteVLD2]>;
def VLD2DUPq16OddPseudoWB_register : VLDQQWBPseudo<IIC_VLD2dup>, Sched<[WriteVLD2]>;
def VLD2DUPq32OddPseudoWB_register : VLDQQWBPseudo<IIC_VLD2dup>, Sched<[WriteVLD2]>;
// VLD3DUP : Vector Load (single 3-element structure to all lanes)
class VLD3DUP<bits<4> op7_4, string Dt>
: NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
(ins addrmode6dup:$Rn), IIC_VLD3dup,
"vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn", "", []>,
Sched<[WriteVLD2]> {
let Rm = 0b1111;
let Inst{4} = 0;
let DecoderMethod = "DecodeVLD3DupInstruction";
}
def VLD3DUPd8 : VLD3DUP<{0,0,0,?}, "8">;
def VLD3DUPd16 : VLD3DUP<{0,1,0,?}, "16">;
def VLD3DUPd32 : VLD3DUP<{1,0,0,?}, "32">;
def VLD3DUPd8Pseudo : VLDQQPseudo<IIC_VLD3dup>, Sched<[WriteVLD2]>;
def VLD3DUPd16Pseudo : VLDQQPseudo<IIC_VLD3dup>, Sched<[WriteVLD2]>;
def VLD3DUPd32Pseudo : VLDQQPseudo<IIC_VLD3dup>, Sched<[WriteVLD2]>;
// ...with double-spaced registers (not used for codegen):
def VLD3DUPq8 : VLD3DUP<{0,0,1,?}, "8">;
def VLD3DUPq16 : VLD3DUP<{0,1,1,?}, "16">;
def VLD3DUPq32 : VLD3DUP<{1,0,1,?}, "32">;
def VLD3DUPq8EvenPseudo : VLDQQQQPseudo<IIC_VLD3dup>, Sched<[WriteVLD2]>;
def VLD3DUPq8OddPseudo : VLDQQQQPseudo<IIC_VLD3dup>, Sched<[WriteVLD2]>;
def VLD3DUPq16EvenPseudo : VLDQQQQPseudo<IIC_VLD3dup>, Sched<[WriteVLD2]>;
def VLD3DUPq16OddPseudo : VLDQQQQPseudo<IIC_VLD3dup>, Sched<[WriteVLD2]>;
def VLD3DUPq32EvenPseudo : VLDQQQQPseudo<IIC_VLD3dup>, Sched<[WriteVLD2]>;
def VLD3DUPq32OddPseudo : VLDQQQQPseudo<IIC_VLD3dup>, Sched<[WriteVLD2]>;
// ...with address register writeback:
class VLD3DUPWB<bits<4> op7_4, string Dt, Operand AddrMode>
: NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
(ins AddrMode:$Rn, am6offset:$Rm), IIC_VLD3dupu,
"vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn$Rm",
"$Rn.addr = $wb", []>, Sched<[WriteVLD2]> {
let Inst{4} = 0;
let DecoderMethod = "DecodeVLD3DupInstruction";
}
def VLD3DUPd8_UPD : VLD3DUPWB<{0,0,0,0}, "8", addrmode6dupalign64>;
def VLD3DUPd16_UPD : VLD3DUPWB<{0,1,0,?}, "16", addrmode6dupalign64>;
def VLD3DUPd32_UPD : VLD3DUPWB<{1,0,0,?}, "32", addrmode6dupalign64>;
def VLD3DUPq8_UPD : VLD3DUPWB<{0,0,1,0}, "8", addrmode6dupalign64>;
def VLD3DUPq16_UPD : VLD3DUPWB<{0,1,1,?}, "16", addrmode6dupalign64>;
def VLD3DUPq32_UPD : VLD3DUPWB<{1,0,1,?}, "32", addrmode6dupalign64>;
def VLD3DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>, Sched<[WriteVLD2]>;
def VLD3DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>, Sched<[WriteVLD2]>;
def VLD3DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>, Sched<[WriteVLD2]>;
def VLD3DUPq8OddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3dupu>, Sched<[WriteVLD2]>;
def VLD3DUPq16OddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3dupu>, Sched<[WriteVLD2]>;
def VLD3DUPq32OddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3dupu>, Sched<[WriteVLD2]>;
// VLD4DUP : Vector Load (single 4-element structure to all lanes)
class VLD4DUP<bits<4> op7_4, string Dt>
: NLdSt<1, 0b10, 0b1111, op7_4,
(outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
(ins addrmode6dup:$Rn), IIC_VLD4dup,
"vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn", "", []> {
let Rm = 0b1111;
let Inst{4} = Rn{4};
let DecoderMethod = "DecodeVLD4DupInstruction";
}
def VLD4DUPd8 : VLD4DUP<{0,0,0,?}, "8">;
def VLD4DUPd16 : VLD4DUP<{0,1,0,?}, "16">;
def VLD4DUPd32 : VLD4DUP<{1,?,0,?}, "32"> { let Inst{6} = Rn{5}; }
def VLD4DUPd8Pseudo : VLDQQPseudo<IIC_VLD4dup>, Sched<[WriteVLD2]>;
def VLD4DUPd16Pseudo : VLDQQPseudo<IIC_VLD4dup>, Sched<[WriteVLD2]>;
def VLD4DUPd32Pseudo : VLDQQPseudo<IIC_VLD4dup>, Sched<[WriteVLD2]>;
// ...with double-spaced registers (not used for codegen):
def VLD4DUPq8 : VLD4DUP<{0,0,1,?}, "8">;
def VLD4DUPq16 : VLD4DUP<{0,1,1,?}, "16">;
def VLD4DUPq32 : VLD4DUP<{1,?,1,?}, "32"> { let Inst{6} = Rn{5}; }
def VLD4DUPq8EvenPseudo : VLDQQQQPseudo<IIC_VLD4dup>, Sched<[WriteVLD2]>;
def VLD4DUPq8OddPseudo : VLDQQQQPseudo<IIC_VLD4dup>, Sched<[WriteVLD2]>;
def VLD4DUPq16EvenPseudo : VLDQQQQPseudo<IIC_VLD4dup>, Sched<[WriteVLD2]>;
def VLD4DUPq16OddPseudo : VLDQQQQPseudo<IIC_VLD4dup>, Sched<[WriteVLD2]>;
def VLD4DUPq32EvenPseudo : VLDQQQQPseudo<IIC_VLD4dup>, Sched<[WriteVLD2]>;
def VLD4DUPq32OddPseudo : VLDQQQQPseudo<IIC_VLD4dup>, Sched<[WriteVLD2]>;
// ...with address register writeback:
class VLD4DUPWB<bits<4> op7_4, string Dt>
: NLdSt<1, 0b10, 0b1111, op7_4,
(outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
(ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD4dupu,
"vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn$Rm",
"$Rn.addr = $wb", []>, Sched<[WriteVLD2]> {
let Inst{4} = Rn{4};
let DecoderMethod = "DecodeVLD4DupInstruction";
}
def VLD4DUPd8_UPD : VLD4DUPWB<{0,0,0,0}, "8">;
def VLD4DUPd16_UPD : VLD4DUPWB<{0,1,0,?}, "16">;
def VLD4DUPd32_UPD : VLD4DUPWB<{1,?,0,?}, "32"> { let Inst{6} = Rn{5}; }
def VLD4DUPq8_UPD : VLD4DUPWB<{0,0,1,0}, "8">;
def VLD4DUPq16_UPD : VLD4DUPWB<{0,1,1,?}, "16">;
def VLD4DUPq32_UPD : VLD4DUPWB<{1,?,1,?}, "32"> { let Inst{6} = Rn{5}; }
def VLD4DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>, Sched<[WriteVLD2]>;
def VLD4DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>, Sched<[WriteVLD2]>;
def VLD4DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>, Sched<[WriteVLD2]>;
def VLD4DUPq8OddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4dupu>, Sched<[WriteVLD2]>;
def VLD4DUPq16OddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4dupu>, Sched<[WriteVLD2]>;
def VLD4DUPq32OddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4dupu>, Sched<[WriteVLD2]>;
} // mayLoad = 1, hasSideEffects = 0, hasExtraDefRegAllocReq = 1
let mayStore = 1, hasSideEffects = 0, hasExtraSrcRegAllocReq = 1 in {
// Classes for VST* pseudo-instructions with multi-register operands.
// These are expanded to real instructions after register allocation.
class VSTQPseudo<InstrItinClass itin>
: PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src), itin, "">;
class VSTQWBPseudo<InstrItinClass itin>
: PseudoNLdSt<(outs GPR:$wb),
(ins addrmode6:$addr, am6offset:$offset, QPR:$src), itin,
"$addr.addr = $wb">;
class VSTQWBfixedPseudo<InstrItinClass itin>
: PseudoNLdSt<(outs GPR:$wb),
(ins addrmode6:$addr, QPR:$src), itin,
"$addr.addr = $wb">;
class VSTQWBregisterPseudo<InstrItinClass itin>
: PseudoNLdSt<(outs GPR:$wb),
(ins addrmode6:$addr, rGPR:$offset, QPR:$src), itin,
"$addr.addr = $wb">;
class VSTQQPseudo<InstrItinClass itin>
: PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src), itin, "">;
class VSTQQWBPseudo<InstrItinClass itin>
: PseudoNLdSt<(outs GPR:$wb),
(ins addrmode6:$addr, am6offset:$offset, QQPR:$src), itin,
"$addr.addr = $wb">;
class VSTQQWBfixedPseudo<InstrItinClass itin>
: PseudoNLdSt<(outs GPR:$wb),
(ins addrmode6:$addr, QQPR:$src), itin,
"$addr.addr = $wb">;
class VSTQQWBregisterPseudo<InstrItinClass itin>
: PseudoNLdSt<(outs GPR:$wb),
(ins addrmode6:$addr, rGPR:$offset, QQPR:$src), itin,
"$addr.addr = $wb">;
class VSTQQQQPseudo<InstrItinClass itin>
: PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src), itin, "">;
class VSTQQQQWBPseudo<InstrItinClass itin>
: PseudoNLdSt<(outs GPR:$wb),
(ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
"$addr.addr = $wb">;
// VST1 : Vector Store (multiple single elements)
class VST1D<bits<4> op7_4, string Dt, Operand AddrMode>
: NLdSt<0,0b00,0b0111,op7_4, (outs), (ins AddrMode:$Rn, VecListOneD:$Vd),
IIC_VST1, "vst1", Dt, "$Vd, $Rn", "", []>, Sched<[WriteVST1]> {
let Rm = 0b1111;
let Inst{4} = Rn{4};
let DecoderMethod = "DecodeVLDST1Instruction";
}
class VST1Q<bits<4> op7_4, string Dt, Operand AddrMode>
: NLdSt<0,0b00,0b1010,op7_4, (outs), (ins AddrMode:$Rn, VecListDPair:$Vd),
IIC_VST1x2, "vst1", Dt, "$Vd, $Rn", "", []>, Sched<[WriteVST2]> {
let Rm = 0b1111;
let Inst{5-4} = Rn{5-4};
let DecoderMethod = "DecodeVLDST1Instruction";
}
def VST1d8 : VST1D<{0,0,0,?}, "8", addrmode6align64>;
def VST1d16 : VST1D<{0,1,0,?}, "16", addrmode6align64>;
def VST1d32 : VST1D<{1,0,0,?}, "32", addrmode6align64>;
def VST1d64 : VST1D<{1,1,0,?}, "64", addrmode6align64>;
def VST1q8 : VST1Q<{0,0,?,?}, "8", addrmode6align64or128>;
def VST1q16 : VST1Q<{0,1,?,?}, "16", addrmode6align64or128>;
def VST1q32 : VST1Q<{1,0,?,?}, "32", addrmode6align64or128>;
def VST1q64 : VST1Q<{1,1,?,?}, "64", addrmode6align64or128>;
// ...with address register writeback:
multiclass VST1DWB<bits<4> op7_4, string Dt, Operand AddrMode> {
def _fixed : NLdSt<0,0b00, 0b0111,op7_4, (outs GPR:$wb),
(ins AddrMode:$Rn, VecListOneD:$Vd), IIC_VLD1u,
"vst1", Dt, "$Vd, $Rn!",
"$Rn.addr = $wb", []>, Sched<[WriteVST1]> {
let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
let Inst{4} = Rn{4};
let DecoderMethod = "DecodeVLDST1Instruction";
}
def _register : NLdSt<0,0b00,0b0111,op7_4, (outs GPR:$wb),
(ins AddrMode:$Rn, rGPR:$Rm, VecListOneD:$Vd),
IIC_VLD1u,
"vst1", Dt, "$Vd, $Rn, $Rm",
"$Rn.addr = $wb", []>, Sched<[WriteVST1]> {
let Inst{4} = Rn{4};
let DecoderMethod = "DecodeVLDST1Instruction";
}
}
multiclass VST1QWB<bits<4> op7_4, string Dt, Operand AddrMode> {
def _fixed : NLdSt<0,0b00,0b1010,op7_4, (outs GPR:$wb),
(ins AddrMode:$Rn, VecListDPair:$Vd), IIC_VLD1x2u,
"vst1", Dt, "$Vd, $Rn!",
"$Rn.addr = $wb", []>, Sched<[WriteVST2]> {
let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
let Inst{5-4} = Rn{5-4};
let DecoderMethod = "DecodeVLDST1Instruction";
}
def _register : NLdSt<0,0b00,0b1010,op7_4, (outs GPR:$wb),
(ins AddrMode:$Rn, rGPR:$Rm, VecListDPair:$Vd),
IIC_VLD1x2u,
"vst1", Dt, "$Vd, $Rn, $Rm",
"$Rn.addr = $wb", []>, Sched<[WriteVST2]> {
let Inst{5-4} = Rn{5-4};
let DecoderMethod = "DecodeVLDST1Instruction";
}
}
defm VST1d8wb : VST1DWB<{0,0,0,?}, "8", addrmode6align64>;
defm VST1d16wb : VST1DWB<{0,1,0,?}, "16", addrmode6align64>;
defm VST1d32wb : VST1DWB<{1,0,0,?}, "32", addrmode6align64>;
defm VST1d64wb : VST1DWB<{1,1,0,?}, "64", addrmode6align64>;
defm VST1q8wb : VST1QWB<{0,0,?,?}, "8", addrmode6align64or128>;
defm VST1q16wb : VST1QWB<{0,1,?,?}, "16", addrmode6align64or128>;
defm VST1q32wb : VST1QWB<{1,0,?,?}, "32", addrmode6align64or128>;
defm VST1q64wb : VST1QWB<{1,1,?,?}, "64", addrmode6align64or128>;
// ...with 3 registers
class VST1D3<bits<4> op7_4, string Dt, Operand AddrMode>
: NLdSt<0, 0b00, 0b0110, op7_4, (outs),
(ins AddrMode:$Rn, VecListThreeD:$Vd),
IIC_VST1x3, "vst1", Dt, "$Vd, $Rn", "", []>, Sched<[WriteVST3]> {
let Rm = 0b1111;
let Inst{4} = Rn{4};
let DecoderMethod = "DecodeVLDST1Instruction";
}
multiclass VST1D3WB<bits<4> op7_4, string Dt, Operand AddrMode> {
def _fixed : NLdSt<0,0b00,0b0110,op7_4, (outs GPR:$wb),
(ins AddrMode:$Rn, VecListThreeD:$Vd), IIC_VLD1x3u,
"vst1", Dt, "$Vd, $Rn!",
"$Rn.addr = $wb", []>, Sched<[WriteVST3]> {
let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
let Inst{5-4} = Rn{5-4};
let DecoderMethod = "DecodeVLDST1Instruction";
}
def _register : NLdSt<0,0b00,0b0110,op7_4, (outs GPR:$wb),
(ins AddrMode:$Rn, rGPR:$Rm, VecListThreeD:$Vd),
IIC_VLD1x3u,
"vst1", Dt, "$Vd, $Rn, $Rm",
"$Rn.addr = $wb", []>, Sched<[WriteVST3]> {
let Inst{5-4} = Rn{5-4};
let DecoderMethod = "DecodeVLDST1Instruction";
}
}
def VST1d8T : VST1D3<{0,0,0,?}, "8", addrmode6align64>;
def VST1d16T : VST1D3<{0,1,0,?}, "16", addrmode6align64>;
def VST1d32T : VST1D3<{1,0,0,?}, "32", addrmode6align64>;
def VST1d64T : VST1D3<{1,1,0,?}, "64", addrmode6align64>;
defm VST1d8Twb : VST1D3WB<{0,0,0,?}, "8", addrmode6align64>;
defm VST1d16Twb : VST1D3WB<{0,1,0,?}, "16", addrmode6align64>;
defm VST1d32Twb : VST1D3WB<{1,0,0,?}, "32", addrmode6align64>;
defm VST1d64Twb : VST1D3WB<{1,1,0,?}, "64", addrmode6align64>;
def VST1d8TPseudo : VSTQQPseudo<IIC_VST1x3>, Sched<[WriteVST3]>;
def VST1d8TPseudoWB_fixed : VSTQQWBfixedPseudo<IIC_VST1x3u>, Sched<[WriteVST3]>;
def VST1d8TPseudoWB_register : VSTQQWBPseudo<IIC_VST1x3u>, Sched<[WriteVST3]>;
def VST1d16TPseudo : VSTQQPseudo<IIC_VST1x3>, Sched<[WriteVST3]>;
def VST1d16TPseudoWB_fixed : VSTQQWBfixedPseudo<IIC_VST1x3u>, Sched<[WriteVST3]>;
def VST1d16TPseudoWB_register : VSTQQWBPseudo<IIC_VST1x3u>, Sched<[WriteVST3]>;
def VST1d32TPseudo : VSTQQPseudo<IIC_VST1x3>, Sched<[WriteVST3]>;
def VST1d32TPseudoWB_fixed : VSTQQWBfixedPseudo<IIC_VST1x3u>, Sched<[WriteVST3]>;
def VST1d32TPseudoWB_register : VSTQQWBPseudo<IIC_VST1x3u>, Sched<[WriteVST3]>;
def VST1d64TPseudo : VSTQQPseudo<IIC_VST1x3>, Sched<[WriteVST3]>;
def VST1d64TPseudoWB_fixed : VSTQQWBfixedPseudo<IIC_VST1x3u>, Sched<[WriteVST3]>;
def VST1d64TPseudoWB_register : VSTQQWBPseudo<IIC_VST1x3u>, Sched<[WriteVST3]>;
def VST1q8HighTPseudo : VSTQQQQPseudo<IIC_VST1x3>, Sched<[WriteVST3]>;
def VST1q16HighTPseudo : VSTQQQQPseudo<IIC_VST1x3>, Sched<[WriteVST3]>;
def VST1q32HighTPseudo : VSTQQQQPseudo<IIC_VST1x3>, Sched<[WriteVST3]>;
def VST1q64HighTPseudo : VSTQQQQPseudo<IIC_VST1x3>, Sched<[WriteVST3]>;
def VST1q8HighTPseudo_UPD : VSTQQQQWBPseudo<IIC_VST1x3>, Sched<[WriteVST3]>;
def VST1q16HighTPseudo_UPD : VSTQQQQWBPseudo<IIC_VST1x3>, Sched<[WriteVST3]>;
def VST1q32HighTPseudo_UPD : VSTQQQQWBPseudo<IIC_VST1x3>, Sched<[WriteVST3]>;
def VST1q64HighTPseudo_UPD : VSTQQQQWBPseudo<IIC_VST1x3>, Sched<[WriteVST3]>;
def VST1q8LowTPseudo_UPD : VSTQQQQWBPseudo<IIC_VST1x3>, Sched<[WriteVST3]>;
def VST1q16LowTPseudo_UPD : VSTQQQQWBPseudo<IIC_VST1x3>, Sched<[WriteVST3]>;
def VST1q32LowTPseudo_UPD : VSTQQQQWBPseudo<IIC_VST1x3>, Sched<[WriteVST3]>;
def VST1q64LowTPseudo_UPD : VSTQQQQWBPseudo<IIC_VST1x3>, Sched<[WriteVST3]>;
// ...with 4 registers
class VST1D4<bits<4> op7_4, string Dt, Operand AddrMode>
: NLdSt<0, 0b00, 0b0010, op7_4, (outs),
(ins AddrMode:$Rn, VecListFourD:$Vd),
IIC_VST1x4, "vst1", Dt, "$Vd, $Rn", "",
[]>, Sched<[WriteVST4]> {
let Rm = 0b1111;
let Inst{5-4} = Rn{5-4};
let DecoderMethod = "DecodeVLDST1Instruction";
}
multiclass VST1D4WB<bits<4> op7_4, string Dt, Operand AddrMode> {
def _fixed : NLdSt<0,0b00,0b0010,op7_4, (outs GPR:$wb),
(ins AddrMode:$Rn, VecListFourD:$Vd), IIC_VLD1x4u,
"vst1", Dt, "$Vd, $Rn!",
"$Rn.addr = $wb", []>, Sched<[WriteVST4]> {
let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
let Inst{5-4} = Rn{5-4};
let DecoderMethod = "DecodeVLDST1Instruction";
}
def _register : NLdSt<0,0b00,0b0010,op7_4, (outs GPR:$wb),
(ins AddrMode:$Rn, rGPR:$Rm, VecListFourD:$Vd),
IIC_VLD1x4u,
"vst1", Dt, "$Vd, $Rn, $Rm",
"$Rn.addr = $wb", []>, Sched<[WriteVST4]> {
let Inst{5-4} = Rn{5-4};
let DecoderMethod = "DecodeVLDST1Instruction";
}
}
def VST1d8Q : VST1D4<{0,0,?,?}, "8", addrmode6align64or128or256>;
def VST1d16Q : VST1D4<{0,1,?,?}, "16", addrmode6align64or128or256>;
def VST1d32Q : VST1D4<{1,0,?,?}, "32", addrmode6align64or128or256>;
def VST1d64Q : VST1D4<{1,1,?,?}, "64", addrmode6align64or128or256>;
defm VST1d8Qwb : VST1D4WB<{0,0,?,?}, "8", addrmode6align64or128or256>;
defm VST1d16Qwb : VST1D4WB<{0,1,?,?}, "16", addrmode6align64or128or256>;
defm VST1d32Qwb : VST1D4WB<{1,0,?,?}, "32", addrmode6align64or128or256>;
defm VST1d64Qwb : VST1D4WB<{1,1,?,?}, "64", addrmode6align64or128or256>;
def VST1d8QPseudo : VSTQQPseudo<IIC_VST1x4>, Sched<[WriteVST4]>;
def VST1d8QPseudoWB_fixed : VSTQQWBfixedPseudo<IIC_VST1x4u>, Sched<[WriteVST4]>;
def VST1d8QPseudoWB_register : VSTQQWBPseudo<IIC_VST1x4u>, Sched<[WriteVST4]>;
def VST1d16QPseudo : VSTQQPseudo<IIC_VST1x4>, Sched<[WriteVST4]>;
def VST1d16QPseudoWB_fixed : VSTQQWBfixedPseudo<IIC_VST1x4u>, Sched<[WriteVST4]>;
def VST1d16QPseudoWB_register : VSTQQWBPseudo<IIC_VST1x4u>, Sched<[WriteVST4]>;
def VST1d32QPseudo : VSTQQPseudo<IIC_VST1x4>, Sched<[WriteVST4]>;
def VST1d32QPseudoWB_fixed : VSTQQWBfixedPseudo<IIC_VST1x4u>, Sched<[WriteVST4]>;
def VST1d32QPseudoWB_register : VSTQQWBPseudo<IIC_VST1x4u>, Sched<[WriteVST4]>;
def VST1d64QPseudo : VSTQQPseudo<IIC_VST1x4>, Sched<[WriteVST4]>;
def VST1d64QPseudoWB_fixed : VSTQQWBfixedPseudo<IIC_VST1x4u>, Sched<[WriteVST4]>;
def VST1d64QPseudoWB_register : VSTQQWBPseudo<IIC_VST1x4u>, Sched<[WriteVST4]>;
def VST1q8HighQPseudo : VSTQQQQPseudo<IIC_VST1x4>, Sched<[WriteVST4]>;
def VST1q16HighQPseudo : VSTQQQQPseudo<IIC_VST1x4>, Sched<[WriteVST4]>;
def VST1q32HighQPseudo : VSTQQQQPseudo<IIC_VST1x4>, Sched<[WriteVST4]>;
def VST1q64HighQPseudo : VSTQQQQPseudo<IIC_VST1x4>, Sched<[WriteVST4]>;
def VST1q8HighQPseudo_UPD : VSTQQQQWBPseudo<IIC_VST1x4>, Sched<[WriteVST4]>;
def VST1q16HighQPseudo_UPD : VSTQQQQWBPseudo<IIC_VST1x4>, Sched<[WriteVST4]>;
def VST1q32HighQPseudo_UPD : VSTQQQQWBPseudo<IIC_VST1x4>, Sched<[WriteVST4]>;
def VST1q64HighQPseudo_UPD : VSTQQQQWBPseudo<IIC_VST1x4>, Sched<[WriteVST4]>;
def VST1q8LowQPseudo_UPD : VSTQQQQWBPseudo<IIC_VST1x4>, Sched<[WriteVST4]>;
def VST1q16LowQPseudo_UPD : VSTQQQQWBPseudo<IIC_VST1x4>, Sched<[WriteVST4]>;
def VST1q32LowQPseudo_UPD : VSTQQQQWBPseudo<IIC_VST1x4>, Sched<[WriteVST4]>;
def VST1q64LowQPseudo_UPD : VSTQQQQWBPseudo<IIC_VST1x4>, Sched<[WriteVST4]>;
// VST2 : Vector Store (multiple 2-element structures)
class VST2<bits<4> op11_8, bits<4> op7_4, string Dt, RegisterOperand VdTy,
InstrItinClass itin, Operand AddrMode>
: NLdSt<0, 0b00, op11_8, op7_4, (outs), (ins AddrMode:$Rn, VdTy:$Vd),
itin, "vst2", Dt, "$Vd, $Rn", "", []> {
let Rm = 0b1111;
let Inst{5-4} = Rn{5-4};
let DecoderMethod = "DecodeVLDST2Instruction";
}
def VST2d8 : VST2<0b1000, {0,0,?,?}, "8", VecListDPair, IIC_VST2,
addrmode6align64or128>, Sched<[WriteVST2]>;
def VST2d16 : VST2<0b1000, {0,1,?,?}, "16", VecListDPair, IIC_VST2,
addrmode6align64or128>, Sched<[WriteVST2]>;
def VST2d32 : VST2<0b1000, {1,0,?,?}, "32", VecListDPair, IIC_VST2,
addrmode6align64or128>, Sched<[WriteVST2]>;
def VST2q8 : VST2<0b0011, {0,0,?,?}, "8", VecListFourD, IIC_VST2x2,
addrmode6align64or128or256>, Sched<[WriteVST4]>;
def VST2q16 : VST2<0b0011, {0,1,?,?}, "16", VecListFourD, IIC_VST2x2,
addrmode6align64or128or256>, Sched<[WriteVST4]>;
def VST2q32 : VST2<0b0011, {1,0,?,?}, "32", VecListFourD, IIC_VST2x2,
addrmode6align64or128or256>, Sched<[WriteVST4]>;
def VST2q8Pseudo : VSTQQPseudo<IIC_VST2x2>, Sched<[WriteVST4]>;
def VST2q16Pseudo : VSTQQPseudo<IIC_VST2x2>, Sched<[WriteVST4]>;
def VST2q32Pseudo : VSTQQPseudo<IIC_VST2x2>, Sched<[WriteVST4]>;
// ...with address register writeback:
multiclass VST2DWB<bits<4> op11_8, bits<4> op7_4, string Dt,
RegisterOperand VdTy, Operand AddrMode> {
def _fixed : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
(ins AddrMode:$Rn, VdTy:$Vd), IIC_VLD1u,
"vst2", Dt, "$Vd, $Rn!",
"$Rn.addr = $wb", []>, Sched<[WriteVST2]> {
let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
let Inst{5-4} = Rn{5-4};
let DecoderMethod = "DecodeVLDST2Instruction";
}
def _register : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
(ins AddrMode:$Rn, rGPR:$Rm, VdTy:$Vd), IIC_VLD1u,
"vst2", Dt, "$Vd, $Rn, $Rm",
"$Rn.addr = $wb", []>, Sched<[WriteVST2]> {
let Inst{5-4} = Rn{5-4};
let DecoderMethod = "DecodeVLDST2Instruction";
}
}
multiclass VST2QWB<bits<4> op7_4, string Dt, Operand AddrMode> {
def _fixed : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
(ins AddrMode:$Rn, VecListFourD:$Vd), IIC_VLD1u,
"vst2", Dt, "$Vd, $Rn!",
"$Rn.addr = $wb", []>, Sched<[WriteVST4]> {
let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
let Inst{5-4} = Rn{5-4};
let DecoderMethod = "DecodeVLDST2Instruction";
}
def _register : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
(ins AddrMode:$Rn, rGPR:$Rm, VecListFourD:$Vd),
IIC_VLD1u,
"vst2", Dt, "$Vd, $Rn, $Rm",
"$Rn.addr = $wb", []>, Sched<[WriteVST4]> {
let Inst{5-4} = Rn{5-4};
let DecoderMethod = "DecodeVLDST2Instruction";
}
}
defm VST2d8wb : VST2DWB<0b1000, {0,0,?,?}, "8", VecListDPair,
addrmode6align64or128>;
defm VST2d16wb : VST2DWB<0b1000, {0,1,?,?}, "16", VecListDPair,
addrmode6align64or128>;
defm VST2d32wb : VST2DWB<0b1000, {1,0,?,?}, "32", VecListDPair,
addrmode6align64or128>;
defm VST2q8wb : VST2QWB<{0,0,?,?}, "8", addrmode6align64or128or256>;
defm VST2q16wb : VST2QWB<{0,1,?,?}, "16", addrmode6align64or128or256>;
defm VST2q32wb : VST2QWB<{1,0,?,?}, "32", addrmode6align64or128or256>;
def VST2q8PseudoWB_fixed : VSTQQWBfixedPseudo<IIC_VST2x2u>, Sched<[WriteVST4]>;
def VST2q16PseudoWB_fixed : VSTQQWBfixedPseudo<IIC_VST2x2u>, Sched<[WriteVST4]>;
def VST2q32PseudoWB_fixed : VSTQQWBfixedPseudo<IIC_VST2x2u>, Sched<[WriteVST4]>;
def VST2q8PseudoWB_register : VSTQQWBregisterPseudo<IIC_VST2x2u>, Sched<[WriteVST4]>;
def VST2q16PseudoWB_register : VSTQQWBregisterPseudo<IIC_VST2x2u>, Sched<[WriteVST4]>;
def VST2q32PseudoWB_register : VSTQQWBregisterPseudo<IIC_VST2x2u>, Sched<[WriteVST4]>;
// ...with double-spaced registers
def VST2b8 : VST2<0b1001, {0,0,?,?}, "8", VecListDPairSpaced, IIC_VST2,
addrmode6align64or128>;
def VST2b16 : VST2<0b1001, {0,1,?,?}, "16", VecListDPairSpaced, IIC_VST2,
addrmode6align64or128>;
def VST2b32 : VST2<0b1001, {1,0,?,?}, "32", VecListDPairSpaced, IIC_VST2,
addrmode6align64or128>;
defm VST2b8wb : VST2DWB<0b1001, {0,0,?,?}, "8", VecListDPairSpaced,
addrmode6align64or128>;
defm VST2b16wb : VST2DWB<0b1001, {0,1,?,?}, "16", VecListDPairSpaced,
addrmode6align64or128>;
defm VST2b32wb : VST2DWB<0b1001, {1,0,?,?}, "32", VecListDPairSpaced,
addrmode6align64or128>;
// VST3 : Vector Store (multiple 3-element structures)
class VST3D<bits<4> op11_8, bits<4> op7_4, string Dt>
: NLdSt<0, 0b00, op11_8, op7_4, (outs),
(ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3,
"vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn", "", []>, Sched<[WriteVST3]> {
let Rm = 0b1111;
let Inst{4} = Rn{4};
let DecoderMethod = "DecodeVLDST3Instruction";
}
def VST3d8 : VST3D<0b0100, {0,0,0,?}, "8">;
def VST3d16 : VST3D<0b0100, {0,1,0,?}, "16">;
def VST3d32 : VST3D<0b0100, {1,0,0,?}, "32">;
def VST3d8Pseudo : VSTQQPseudo<IIC_VST3>, Sched<[WriteVST3]>;
def VST3d16Pseudo : VSTQQPseudo<IIC_VST3>, Sched<[WriteVST3]>;
def VST3d32Pseudo : VSTQQPseudo<IIC_VST3>, Sched<[WriteVST3]>;
// ...with address register writeback:
class VST3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
: NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
(ins addrmode6:$Rn, am6offset:$Rm,
DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3u,
"vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn$Rm",
"$Rn.addr = $wb", []>, Sched<[WriteVST3]> {
let Inst{4} = Rn{4};
let DecoderMethod = "DecodeVLDST3Instruction";
}
def VST3d8_UPD : VST3DWB<0b0100, {0,0,0,?}, "8">;
def VST3d16_UPD : VST3DWB<0b0100, {0,1,0,?}, "16">;
def VST3d32_UPD : VST3DWB<0b0100, {1,0,0,?}, "32">;
def VST3d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>, Sched<[WriteVST3]>;
def VST3d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>, Sched<[WriteVST3]>;
def VST3d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>, Sched<[WriteVST3]>;
// ...with double-spaced registers:
def VST3q8 : VST3D<0b0101, {0,0,0,?}, "8">;
def VST3q16 : VST3D<0b0101, {0,1,0,?}, "16">;
def VST3q32 : VST3D<0b0101, {1,0,0,?}, "32">;
def VST3q8_UPD : VST3DWB<0b0101, {0,0,0,?}, "8">;
def VST3q16_UPD : VST3DWB<0b0101, {0,1,0,?}, "16">;
def VST3q32_UPD : VST3DWB<0b0101, {1,0,0,?}, "32">;
def VST3q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>, Sched<[WriteVST3]>;
def VST3q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>, Sched<[WriteVST3]>;
def VST3q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>, Sched<[WriteVST3]>;
// ...alternate versions to be allocated odd register numbers:
def VST3q8oddPseudo : VSTQQQQPseudo<IIC_VST3>, Sched<[WriteVST3]>;
def VST3q16oddPseudo : VSTQQQQPseudo<IIC_VST3>, Sched<[WriteVST3]>;
def VST3q32oddPseudo : VSTQQQQPseudo<IIC_VST3>, Sched<[WriteVST3]>;
def VST3q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>, Sched<[WriteVST3]>;
def VST3q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>, Sched<[WriteVST3]>;
def VST3q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>, Sched<[WriteVST3]>;
// VST4 : Vector Store (multiple 4-element structures)
class VST4D<bits<4> op11_8, bits<4> op7_4, string Dt>
: NLdSt<0, 0b00, op11_8, op7_4, (outs),
(ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
IIC_VST4, "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn",
"", []>, Sched<[WriteVST4]> {
let Rm = 0b1111;
let Inst{5-4} = Rn{5-4};
let DecoderMethod = "DecodeVLDST4Instruction";
}
def VST4d8 : VST4D<0b0000, {0,0,?,?}, "8">;
def VST4d16 : VST4D<0b0000, {0,1,?,?}, "16">;
def VST4d32 : VST4D<0b0000, {1,0,?,?}, "32">;
def VST4d8Pseudo : VSTQQPseudo<IIC_VST4>, Sched<[WriteVST4]>;
def VST4d16Pseudo : VSTQQPseudo<IIC_VST4>, Sched<[WriteVST4]>;
def VST4d32Pseudo : VSTQQPseudo<IIC_VST4>, Sched<[WriteVST4]>;
// ...with address register writeback:
class VST4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
: NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
(ins addrmode6:$Rn, am6offset:$Rm,
DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST4u,
"vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
"$Rn.addr = $wb", []>, Sched<[WriteVST4]> {
let Inst{5-4} = Rn{5-4};
let DecoderMethod = "DecodeVLDST4Instruction";
}
def VST4d8_UPD : VST4DWB<0b0000, {0,0,?,?}, "8">;
def VST4d16_UPD : VST4DWB<0b0000, {0,1,?,?}, "16">;
def VST4d32_UPD : VST4DWB<0b0000, {1,0,?,?}, "32">;
def VST4d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>, Sched<[WriteVST4]>;
def VST4d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>, Sched<[WriteVST4]>;
def VST4d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>, Sched<[WriteVST4]>;
// ...with double-spaced registers:
def VST4q8 : VST4D<0b0001, {0,0,?,?}, "8">;
def VST4q16 : VST4D<0b0001, {0,1,?,?}, "16">;
def VST4q32 : VST4D<0b0001, {1,0,?,?}, "32">;
def VST4q8_UPD : VST4DWB<0b0001, {0,0,?,?}, "8">;
def VST4q16_UPD : VST4DWB<0b0001, {0,1,?,?}, "16">;
def VST4q32_UPD : VST4DWB<0b0001, {1,0,?,?}, "32">;
def VST4q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>, Sched<[WriteVST4]>;
def VST4q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>, Sched<[WriteVST4]>;
def VST4q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>, Sched<[WriteVST4]>;
// ...alternate versions to be allocated odd register numbers:
def VST4q8oddPseudo : VSTQQQQPseudo<IIC_VST4>, Sched<[WriteVST4]>;
def VST4q16oddPseudo : VSTQQQQPseudo<IIC_VST4>, Sched<[WriteVST4]>;
def VST4q32oddPseudo : VSTQQQQPseudo<IIC_VST4>, Sched<[WriteVST4]>;
def VST4q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>, Sched<[WriteVST4]>;
def VST4q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>, Sched<[WriteVST4]>;
def VST4q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>, Sched<[WriteVST4]>;
} // mayStore = 1, hasSideEffects = 0, hasExtraSrcRegAllocReq = 1
// Classes for VST*LN pseudo-instructions with multi-register operands.
// These are expanded to real instructions after register allocation.
class VSTQLNPseudo<InstrItinClass itin>
: PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
itin, "">;
class VSTQLNWBPseudo<InstrItinClass itin>
: PseudoNLdSt<(outs GPR:$wb),
(ins addrmode6:$addr, am6offset:$offset, QPR:$src,
nohash_imm:$lane), itin, "$addr.addr = $wb">;
class VSTQQLNPseudo<InstrItinClass itin>
: PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
itin, "">;
class VSTQQLNWBPseudo<InstrItinClass itin>
: PseudoNLdSt<(outs GPR:$wb),
(ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
nohash_imm:$lane), itin, "$addr.addr = $wb">;
class VSTQQQQLNPseudo<InstrItinClass itin>
: PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
itin, "">;
class VSTQQQQLNWBPseudo<InstrItinClass itin>
: PseudoNLdSt<(outs GPR:$wb),
(ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
nohash_imm:$lane), itin, "$addr.addr = $wb">;
// VST1LN : Vector Store (single element from one lane)
class VST1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
PatFrag StoreOp, SDNode ExtractOp, Operand AddrMode>
: NLdStLn<1, 0b00, op11_8, op7_4, (outs),
(ins AddrMode:$Rn, DPR:$Vd, nohash_imm:$lane),
IIC_VST1ln, "vst1", Dt, "\\{$Vd[$lane]\\}, $Rn", "",
[(StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane), AddrMode:$Rn)]>,
Sched<[WriteVST1]> {
let Rm = 0b1111;
let DecoderMethod = "DecodeVST1LN";
}
class VST1QLNPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
: VSTQLNPseudo<IIC_VST1ln>, Sched<[WriteVST1]> {
let Pattern = [(StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
addrmode6:$addr)];
}
def VST1LNd8 : VST1LN<0b0000, {?,?,?,0}, "8", v8i8, truncstorei8,
ARMvgetlaneu, addrmode6> {
let Inst{7-5} = lane{2-0};
}
def VST1LNd16 : VST1LN<0b0100, {?,?,0,?}, "16", v4i16, truncstorei16,
ARMvgetlaneu, addrmode6> {
let Inst{7-6} = lane{1-0};
let Inst{4} = Rn{4};