xref: /freebsd/contrib/llvm-project/llvm/lib/Target/AArch64/AArch64InstrInfo.td (revision a7dea1671b87c07d2d266f836bfa8b58efc7c134)
1//=- AArch64InstrInfo.td - Describe the AArch64 Instructions -*- tablegen -*-=//
2//
3// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4// See https://llvm.org/LICENSE.txt for license information.
5// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6//
7//===----------------------------------------------------------------------===//
8//
9// AArch64 Instruction definitions.
10//
11//===----------------------------------------------------------------------===//
12
13//===----------------------------------------------------------------------===//
14// ARM Instruction Predicate Definitions.
15//
16def HasV8_1a         : Predicate<"Subtarget->hasV8_1aOps()">,
17                                 AssemblerPredicate<"HasV8_1aOps", "armv8.1a">;
18def HasV8_2a         : Predicate<"Subtarget->hasV8_2aOps()">,
19                                 AssemblerPredicate<"HasV8_2aOps", "armv8.2a">;
20def HasV8_3a         : Predicate<"Subtarget->hasV8_3aOps()">,
21                                 AssemblerPredicate<"HasV8_3aOps", "armv8.3a">;
22def HasV8_4a         : Predicate<"Subtarget->hasV8_4aOps()">,
23                                 AssemblerPredicate<"HasV8_4aOps", "armv8.4a">;
24def HasV8_5a         : Predicate<"Subtarget->hasV8_5aOps()">,
25                                 AssemblerPredicate<"HasV8_5aOps", "armv8.5a">;
26def HasVH            : Predicate<"Subtarget->hasVH()">,
27                       AssemblerPredicate<"FeatureVH", "vh">;
28
29def HasLOR           : Predicate<"Subtarget->hasLOR()">,
30                       AssemblerPredicate<"FeatureLOR", "lor">;
31
32def HasPA            : Predicate<"Subtarget->hasPA()">,
33                       AssemblerPredicate<"FeaturePA", "pa">;
34
35def HasJS            : Predicate<"Subtarget->hasJS()">,
36                       AssemblerPredicate<"FeatureJS", "jsconv">;
37
38def HasCCIDX         : Predicate<"Subtarget->hasCCIDX()">,
39                       AssemblerPredicate<"FeatureCCIDX", "ccidx">;
40
41def HasComplxNum      : Predicate<"Subtarget->hasComplxNum()">,
42                       AssemblerPredicate<"FeatureComplxNum", "complxnum">;
43
44def HasNV            : Predicate<"Subtarget->hasNV()">,
45                       AssemblerPredicate<"FeatureNV", "nv">;
46
47def HasRASv8_4       : Predicate<"Subtarget->hasRASv8_4()">,
48                       AssemblerPredicate<"FeatureRASv8_4", "rasv8_4">;
49
50def HasMPAM          : Predicate<"Subtarget->hasMPAM()">,
51                       AssemblerPredicate<"FeatureMPAM", "mpam">;
52
53def HasDIT           : Predicate<"Subtarget->hasDIT()">,
54                       AssemblerPredicate<"FeatureDIT", "dit">;
55
56def HasTRACEV8_4         : Predicate<"Subtarget->hasTRACEV8_4()">,
57                       AssemblerPredicate<"FeatureTRACEV8_4", "tracev8.4">;
58
59def HasAM            : Predicate<"Subtarget->hasAM()">,
60                       AssemblerPredicate<"FeatureAM", "am">;
61
62def HasSEL2          : Predicate<"Subtarget->hasSEL2()">,
63                       AssemblerPredicate<"FeatureSEL2", "sel2">;
64
65def HasPMU           : Predicate<"Subtarget->hasPMU()">,
66                       AssemblerPredicate<"FeaturePMU", "pmu">;
67
68def HasTLB_RMI          : Predicate<"Subtarget->hasTLB_RMI()">,
69                       AssemblerPredicate<"FeatureTLB_RMI", "tlb-rmi">;
70
71def HasFMI           : Predicate<"Subtarget->hasFMI()">,
72                       AssemblerPredicate<"FeatureFMI", "fmi">;
73
74def HasRCPC_IMMO      : Predicate<"Subtarget->hasRCPCImm()">,
75                       AssemblerPredicate<"FeatureRCPC_IMMO", "rcpc-immo">;
76
77def HasFPARMv8       : Predicate<"Subtarget->hasFPARMv8()">,
78                               AssemblerPredicate<"FeatureFPARMv8", "fp-armv8">;
79def HasNEON          : Predicate<"Subtarget->hasNEON()">,
80                                 AssemblerPredicate<"FeatureNEON", "neon">;
81def HasCrypto        : Predicate<"Subtarget->hasCrypto()">,
82                                 AssemblerPredicate<"FeatureCrypto", "crypto">;
83def HasSM4           : Predicate<"Subtarget->hasSM4()">,
84                                 AssemblerPredicate<"FeatureSM4", "sm4">;
85def HasSHA3          : Predicate<"Subtarget->hasSHA3()">,
86                                 AssemblerPredicate<"FeatureSHA3", "sha3">;
87def HasSHA2          : Predicate<"Subtarget->hasSHA2()">,
88                                 AssemblerPredicate<"FeatureSHA2", "sha2">;
89def HasAES           : Predicate<"Subtarget->hasAES()">,
90                                 AssemblerPredicate<"FeatureAES", "aes">;
91def HasDotProd       : Predicate<"Subtarget->hasDotProd()">,
92                                 AssemblerPredicate<"FeatureDotProd", "dotprod">;
93def HasCRC           : Predicate<"Subtarget->hasCRC()">,
94                                 AssemblerPredicate<"FeatureCRC", "crc">;
95def HasLSE           : Predicate<"Subtarget->hasLSE()">,
96                                 AssemblerPredicate<"FeatureLSE", "lse">;
97def HasRAS           : Predicate<"Subtarget->hasRAS()">,
98                                 AssemblerPredicate<"FeatureRAS", "ras">;
99def HasRDM           : Predicate<"Subtarget->hasRDM()">,
100                                 AssemblerPredicate<"FeatureRDM", "rdm">;
101def HasPerfMon       : Predicate<"Subtarget->hasPerfMon()">;
102def HasFullFP16      : Predicate<"Subtarget->hasFullFP16()">,
103                                 AssemblerPredicate<"FeatureFullFP16", "fullfp16">;
104def HasFP16FML       : Predicate<"Subtarget->hasFP16FML()">,
105                                 AssemblerPredicate<"FeatureFP16FML", "fp16fml">;
106def HasSPE           : Predicate<"Subtarget->hasSPE()">,
107                                 AssemblerPredicate<"FeatureSPE", "spe">;
108def HasFuseAES       : Predicate<"Subtarget->hasFuseAES()">,
109                                 AssemblerPredicate<"FeatureFuseAES",
110                                 "fuse-aes">;
111def HasSVE           : Predicate<"Subtarget->hasSVE()">,
112                                 AssemblerPredicate<"FeatureSVE", "sve">;
113def HasSVE2          : Predicate<"Subtarget->hasSVE2()">,
114                                 AssemblerPredicate<"FeatureSVE2", "sve2">;
115def HasSVE2AES       : Predicate<"Subtarget->hasSVE2AES()">,
116                                 AssemblerPredicate<"FeatureSVE2AES", "sve2-aes">;
117def HasSVE2SM4       : Predicate<"Subtarget->hasSVE2SM4()">,
118                                 AssemblerPredicate<"FeatureSVE2SM4", "sve2-sm4">;
119def HasSVE2SHA3      : Predicate<"Subtarget->hasSVE2SHA3()">,
120                                 AssemblerPredicate<"FeatureSVE2SHA3", "sve2-sha3">;
121def HasSVE2BitPerm   : Predicate<"Subtarget->hasSVE2BitPerm()">,
122                                 AssemblerPredicate<"FeatureSVE2BitPerm", "sve2-bitperm">;
123def HasRCPC          : Predicate<"Subtarget->hasRCPC()">,
124                                 AssemblerPredicate<"FeatureRCPC", "rcpc">;
125def HasAltNZCV       : Predicate<"Subtarget->hasAlternativeNZCV()">,
126                       AssemblerPredicate<"FeatureAltFPCmp", "altnzcv">;
127def HasFRInt3264     : Predicate<"Subtarget->hasFRInt3264()">,
128                       AssemblerPredicate<"FeatureFRInt3264", "frint3264">;
129def HasSB            : Predicate<"Subtarget->hasSB()">,
130                       AssemblerPredicate<"FeatureSB", "sb">;
131def HasPredRes      : Predicate<"Subtarget->hasPredRes()">,
132                       AssemblerPredicate<"FeaturePredRes", "predres">;
133def HasCCDP          : Predicate<"Subtarget->hasCCDP()">,
134                       AssemblerPredicate<"FeatureCacheDeepPersist", "ccdp">;
135def HasBTI           : Predicate<"Subtarget->hasBTI()">,
136                       AssemblerPredicate<"FeatureBranchTargetId", "bti">;
137def HasMTE           : Predicate<"Subtarget->hasMTE()">,
138                       AssemblerPredicate<"FeatureMTE", "mte">;
139def HasTME           : Predicate<"Subtarget->hasTME()">,
140                       AssemblerPredicate<"FeatureTME", "tme">;
141def HasETE           : Predicate<"Subtarget->hasETE()">,
142                       AssemblerPredicate<"FeatureETE", "ete">;
143def HasTRBE          : Predicate<"Subtarget->hasTRBE()">,
144                       AssemblerPredicate<"FeatureTRBE", "trbe">;
145def IsLE             : Predicate<"Subtarget->isLittleEndian()">;
146def IsBE             : Predicate<"!Subtarget->isLittleEndian()">;
147def IsWindows        : Predicate<"Subtarget->isTargetWindows()">;
148def UseAlternateSExtLoadCVTF32
149    : Predicate<"Subtarget->useAlternateSExtLoadCVTF32Pattern()">;
150
151def UseNegativeImmediates
152    : Predicate<"false">, AssemblerPredicate<"!FeatureNoNegativeImmediates",
153                                             "NegativeImmediates">;
154
155def AArch64LocalRecover : SDNode<"ISD::LOCAL_RECOVER",
156                                  SDTypeProfile<1, 1, [SDTCisSameAs<0, 1>,
157                                                       SDTCisInt<1>]>>;
158
159
160//===----------------------------------------------------------------------===//
161// AArch64-specific DAG Nodes.
162//
163
164// SDTBinaryArithWithFlagsOut - RES1, FLAGS = op LHS, RHS
165def SDTBinaryArithWithFlagsOut : SDTypeProfile<2, 2,
166                                              [SDTCisSameAs<0, 2>,
167                                               SDTCisSameAs<0, 3>,
168                                               SDTCisInt<0>, SDTCisVT<1, i32>]>;
169
170// SDTBinaryArithWithFlagsIn - RES1, FLAGS = op LHS, RHS, FLAGS
171def SDTBinaryArithWithFlagsIn : SDTypeProfile<1, 3,
172                                            [SDTCisSameAs<0, 1>,
173                                             SDTCisSameAs<0, 2>,
174                                             SDTCisInt<0>,
175                                             SDTCisVT<3, i32>]>;
176
177// SDTBinaryArithWithFlagsInOut - RES1, FLAGS = op LHS, RHS, FLAGS
178def SDTBinaryArithWithFlagsInOut : SDTypeProfile<2, 3,
179                                            [SDTCisSameAs<0, 2>,
180                                             SDTCisSameAs<0, 3>,
181                                             SDTCisInt<0>,
182                                             SDTCisVT<1, i32>,
183                                             SDTCisVT<4, i32>]>;
184
185def SDT_AArch64Brcond  : SDTypeProfile<0, 3,
186                                     [SDTCisVT<0, OtherVT>, SDTCisVT<1, i32>,
187                                      SDTCisVT<2, i32>]>;
188def SDT_AArch64cbz : SDTypeProfile<0, 2, [SDTCisInt<0>, SDTCisVT<1, OtherVT>]>;
189def SDT_AArch64tbz : SDTypeProfile<0, 3, [SDTCisInt<0>, SDTCisInt<1>,
190                                        SDTCisVT<2, OtherVT>]>;
191
192
193def SDT_AArch64CSel  : SDTypeProfile<1, 4,
194                                   [SDTCisSameAs<0, 1>,
195                                    SDTCisSameAs<0, 2>,
196                                    SDTCisInt<3>,
197                                    SDTCisVT<4, i32>]>;
198def SDT_AArch64CCMP : SDTypeProfile<1, 5,
199                                    [SDTCisVT<0, i32>,
200                                     SDTCisInt<1>,
201                                     SDTCisSameAs<1, 2>,
202                                     SDTCisInt<3>,
203                                     SDTCisInt<4>,
204                                     SDTCisVT<5, i32>]>;
205def SDT_AArch64FCCMP : SDTypeProfile<1, 5,
206                                     [SDTCisVT<0, i32>,
207                                      SDTCisFP<1>,
208                                      SDTCisSameAs<1, 2>,
209                                      SDTCisInt<3>,
210                                      SDTCisInt<4>,
211                                      SDTCisVT<5, i32>]>;
212def SDT_AArch64FCmp   : SDTypeProfile<0, 2,
213                                   [SDTCisFP<0>,
214                                    SDTCisSameAs<0, 1>]>;
215def SDT_AArch64Dup   : SDTypeProfile<1, 1, [SDTCisVec<0>]>;
216def SDT_AArch64DupLane   : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisInt<2>]>;
217def SDT_AArch64Zip   : SDTypeProfile<1, 2, [SDTCisVec<0>,
218                                          SDTCisSameAs<0, 1>,
219                                          SDTCisSameAs<0, 2>]>;
220def SDT_AArch64MOVIedit : SDTypeProfile<1, 1, [SDTCisInt<1>]>;
221def SDT_AArch64MOVIshift : SDTypeProfile<1, 2, [SDTCisInt<1>, SDTCisInt<2>]>;
222def SDT_AArch64vecimm : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0,1>,
223                                           SDTCisInt<2>, SDTCisInt<3>]>;
224def SDT_AArch64UnaryVec: SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0,1>]>;
225def SDT_AArch64ExtVec: SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0,1>,
226                                          SDTCisSameAs<0,2>, SDTCisInt<3>]>;
227def SDT_AArch64vshift : SDTypeProfile<1, 2, [SDTCisSameAs<0,1>, SDTCisInt<2>]>;
228
229def SDT_AArch64unvec : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0,1>]>;
230def SDT_AArch64fcmpz : SDTypeProfile<1, 1, []>;
231def SDT_AArch64fcmp  : SDTypeProfile<1, 2, [SDTCisSameAs<1,2>]>;
232def SDT_AArch64binvec : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisSameAs<0,1>,
233                                           SDTCisSameAs<0,2>]>;
234def SDT_AArch64trivec : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0,1>,
235                                           SDTCisSameAs<0,2>,
236                                           SDTCisSameAs<0,3>]>;
237def SDT_AArch64TCRET : SDTypeProfile<0, 2, [SDTCisPtrTy<0>]>;
238def SDT_AArch64PREFETCH : SDTypeProfile<0, 2, [SDTCisVT<0, i32>, SDTCisPtrTy<1>]>;
239
240def SDT_AArch64ITOF  : SDTypeProfile<1, 1, [SDTCisFP<0>, SDTCisSameAs<0,1>]>;
241
242def SDT_AArch64TLSDescCall : SDTypeProfile<0, -2, [SDTCisPtrTy<0>,
243                                                 SDTCisPtrTy<1>]>;
244
245// Generates the general dynamic sequences, i.e.
246//  adrp  x0, :tlsdesc:var
247//  ldr   x1, [x0, #:tlsdesc_lo12:var]
248//  add   x0, x0, #:tlsdesc_lo12:var
249//  .tlsdesccall var
250//  blr   x1
251
252// (the TPIDR_EL0 offset is put directly in X0, hence no "result" here)
253// number of operands (the variable)
254def SDT_AArch64TLSDescCallSeq : SDTypeProfile<0,1,
255                                          [SDTCisPtrTy<0>]>;
256
257def SDT_AArch64WrapperLarge : SDTypeProfile<1, 4,
258                                        [SDTCisVT<0, i64>, SDTCisVT<1, i32>,
259                                         SDTCisSameAs<1, 2>, SDTCisSameAs<1, 3>,
260                                         SDTCisSameAs<1, 4>]>;
261
262
263// Node definitions.
264def AArch64adrp          : SDNode<"AArch64ISD::ADRP", SDTIntUnaryOp, []>;
265def AArch64adr           : SDNode<"AArch64ISD::ADR", SDTIntUnaryOp, []>;
266def AArch64addlow        : SDNode<"AArch64ISD::ADDlow", SDTIntBinOp, []>;
267def AArch64LOADgot       : SDNode<"AArch64ISD::LOADgot", SDTIntUnaryOp>;
268def AArch64callseq_start : SDNode<"ISD::CALLSEQ_START",
269                                SDCallSeqStart<[ SDTCisVT<0, i32>,
270                                                 SDTCisVT<1, i32> ]>,
271                                [SDNPHasChain, SDNPOutGlue]>;
272def AArch64callseq_end   : SDNode<"ISD::CALLSEQ_END",
273                                SDCallSeqEnd<[ SDTCisVT<0, i32>,
274                                               SDTCisVT<1, i32> ]>,
275                                [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue]>;
276def AArch64call          : SDNode<"AArch64ISD::CALL",
277                                SDTypeProfile<0, -1, [SDTCisPtrTy<0>]>,
278                                [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue,
279                                 SDNPVariadic]>;
280def AArch64brcond        : SDNode<"AArch64ISD::BRCOND", SDT_AArch64Brcond,
281                                [SDNPHasChain]>;
282def AArch64cbz           : SDNode<"AArch64ISD::CBZ", SDT_AArch64cbz,
283                                [SDNPHasChain]>;
284def AArch64cbnz           : SDNode<"AArch64ISD::CBNZ", SDT_AArch64cbz,
285                                [SDNPHasChain]>;
286def AArch64tbz           : SDNode<"AArch64ISD::TBZ", SDT_AArch64tbz,
287                                [SDNPHasChain]>;
288def AArch64tbnz           : SDNode<"AArch64ISD::TBNZ", SDT_AArch64tbz,
289                                [SDNPHasChain]>;
290
291
292def AArch64csel          : SDNode<"AArch64ISD::CSEL", SDT_AArch64CSel>;
293def AArch64csinv         : SDNode<"AArch64ISD::CSINV", SDT_AArch64CSel>;
294def AArch64csneg         : SDNode<"AArch64ISD::CSNEG", SDT_AArch64CSel>;
295def AArch64csinc         : SDNode<"AArch64ISD::CSINC", SDT_AArch64CSel>;
296def AArch64retflag       : SDNode<"AArch64ISD::RET_FLAG", SDTNone,
297                                [SDNPHasChain, SDNPOptInGlue, SDNPVariadic]>;
298def AArch64adc       : SDNode<"AArch64ISD::ADC",  SDTBinaryArithWithFlagsIn >;
299def AArch64sbc       : SDNode<"AArch64ISD::SBC",  SDTBinaryArithWithFlagsIn>;
300def AArch64add_flag  : SDNode<"AArch64ISD::ADDS",  SDTBinaryArithWithFlagsOut,
301                            [SDNPCommutative]>;
302def AArch64sub_flag  : SDNode<"AArch64ISD::SUBS",  SDTBinaryArithWithFlagsOut>;
303def AArch64and_flag  : SDNode<"AArch64ISD::ANDS",  SDTBinaryArithWithFlagsOut,
304                            [SDNPCommutative]>;
305def AArch64adc_flag  : SDNode<"AArch64ISD::ADCS",  SDTBinaryArithWithFlagsInOut>;
306def AArch64sbc_flag  : SDNode<"AArch64ISD::SBCS",  SDTBinaryArithWithFlagsInOut>;
307
308def AArch64ccmp      : SDNode<"AArch64ISD::CCMP",  SDT_AArch64CCMP>;
309def AArch64ccmn      : SDNode<"AArch64ISD::CCMN",  SDT_AArch64CCMP>;
310def AArch64fccmp     : SDNode<"AArch64ISD::FCCMP", SDT_AArch64FCCMP>;
311
312def AArch64threadpointer : SDNode<"AArch64ISD::THREAD_POINTER", SDTPtrLeaf>;
313
314def AArch64fcmp      : SDNode<"AArch64ISD::FCMP", SDT_AArch64FCmp>;
315
316def AArch64dup       : SDNode<"AArch64ISD::DUP", SDT_AArch64Dup>;
317def AArch64duplane8  : SDNode<"AArch64ISD::DUPLANE8", SDT_AArch64DupLane>;
318def AArch64duplane16 : SDNode<"AArch64ISD::DUPLANE16", SDT_AArch64DupLane>;
319def AArch64duplane32 : SDNode<"AArch64ISD::DUPLANE32", SDT_AArch64DupLane>;
320def AArch64duplane64 : SDNode<"AArch64ISD::DUPLANE64", SDT_AArch64DupLane>;
321
322def AArch64zip1      : SDNode<"AArch64ISD::ZIP1", SDT_AArch64Zip>;
323def AArch64zip2      : SDNode<"AArch64ISD::ZIP2", SDT_AArch64Zip>;
324def AArch64uzp1      : SDNode<"AArch64ISD::UZP1", SDT_AArch64Zip>;
325def AArch64uzp2      : SDNode<"AArch64ISD::UZP2", SDT_AArch64Zip>;
326def AArch64trn1      : SDNode<"AArch64ISD::TRN1", SDT_AArch64Zip>;
327def AArch64trn2      : SDNode<"AArch64ISD::TRN2", SDT_AArch64Zip>;
328
329def AArch64movi_edit : SDNode<"AArch64ISD::MOVIedit", SDT_AArch64MOVIedit>;
330def AArch64movi_shift : SDNode<"AArch64ISD::MOVIshift", SDT_AArch64MOVIshift>;
331def AArch64movi_msl : SDNode<"AArch64ISD::MOVImsl", SDT_AArch64MOVIshift>;
332def AArch64mvni_shift : SDNode<"AArch64ISD::MVNIshift", SDT_AArch64MOVIshift>;
333def AArch64mvni_msl : SDNode<"AArch64ISD::MVNImsl", SDT_AArch64MOVIshift>;
334def AArch64movi : SDNode<"AArch64ISD::MOVI", SDT_AArch64MOVIedit>;
335def AArch64fmov : SDNode<"AArch64ISD::FMOV", SDT_AArch64MOVIedit>;
336
337def AArch64rev16 : SDNode<"AArch64ISD::REV16", SDT_AArch64UnaryVec>;
338def AArch64rev32 : SDNode<"AArch64ISD::REV32", SDT_AArch64UnaryVec>;
339def AArch64rev64 : SDNode<"AArch64ISD::REV64", SDT_AArch64UnaryVec>;
340def AArch64ext : SDNode<"AArch64ISD::EXT", SDT_AArch64ExtVec>;
341
342def AArch64vashr : SDNode<"AArch64ISD::VASHR", SDT_AArch64vshift>;
343def AArch64vlshr : SDNode<"AArch64ISD::VLSHR", SDT_AArch64vshift>;
344def AArch64vshl : SDNode<"AArch64ISD::VSHL", SDT_AArch64vshift>;
345def AArch64sqshli : SDNode<"AArch64ISD::SQSHL_I", SDT_AArch64vshift>;
346def AArch64uqshli : SDNode<"AArch64ISD::UQSHL_I", SDT_AArch64vshift>;
347def AArch64sqshlui : SDNode<"AArch64ISD::SQSHLU_I", SDT_AArch64vshift>;
348def AArch64srshri : SDNode<"AArch64ISD::SRSHR_I", SDT_AArch64vshift>;
349def AArch64urshri : SDNode<"AArch64ISD::URSHR_I", SDT_AArch64vshift>;
350
351def AArch64not: SDNode<"AArch64ISD::NOT", SDT_AArch64unvec>;
352def AArch64bit: SDNode<"AArch64ISD::BIT", SDT_AArch64trivec>;
353def AArch64bsl: SDNode<"AArch64ISD::BSL", SDT_AArch64trivec>;
354
355def AArch64cmeq: SDNode<"AArch64ISD::CMEQ", SDT_AArch64binvec>;
356def AArch64cmge: SDNode<"AArch64ISD::CMGE", SDT_AArch64binvec>;
357def AArch64cmgt: SDNode<"AArch64ISD::CMGT", SDT_AArch64binvec>;
358def AArch64cmhi: SDNode<"AArch64ISD::CMHI", SDT_AArch64binvec>;
359def AArch64cmhs: SDNode<"AArch64ISD::CMHS", SDT_AArch64binvec>;
360
361def AArch64fcmeq: SDNode<"AArch64ISD::FCMEQ", SDT_AArch64fcmp>;
362def AArch64fcmge: SDNode<"AArch64ISD::FCMGE", SDT_AArch64fcmp>;
363def AArch64fcmgt: SDNode<"AArch64ISD::FCMGT", SDT_AArch64fcmp>;
364
365def AArch64cmeqz: SDNode<"AArch64ISD::CMEQz", SDT_AArch64unvec>;
366def AArch64cmgez: SDNode<"AArch64ISD::CMGEz", SDT_AArch64unvec>;
367def AArch64cmgtz: SDNode<"AArch64ISD::CMGTz", SDT_AArch64unvec>;
368def AArch64cmlez: SDNode<"AArch64ISD::CMLEz", SDT_AArch64unvec>;
369def AArch64cmltz: SDNode<"AArch64ISD::CMLTz", SDT_AArch64unvec>;
370def AArch64cmtst : PatFrag<(ops node:$LHS, node:$RHS),
371                        (AArch64not (AArch64cmeqz (and node:$LHS, node:$RHS)))>;
372
373def AArch64fcmeqz: SDNode<"AArch64ISD::FCMEQz", SDT_AArch64fcmpz>;
374def AArch64fcmgez: SDNode<"AArch64ISD::FCMGEz", SDT_AArch64fcmpz>;
375def AArch64fcmgtz: SDNode<"AArch64ISD::FCMGTz", SDT_AArch64fcmpz>;
376def AArch64fcmlez: SDNode<"AArch64ISD::FCMLEz", SDT_AArch64fcmpz>;
377def AArch64fcmltz: SDNode<"AArch64ISD::FCMLTz", SDT_AArch64fcmpz>;
378
379def AArch64bici: SDNode<"AArch64ISD::BICi", SDT_AArch64vecimm>;
380def AArch64orri: SDNode<"AArch64ISD::ORRi", SDT_AArch64vecimm>;
381
382def AArch64neg : SDNode<"AArch64ISD::NEG", SDT_AArch64unvec>;
383
384def AArch64tcret: SDNode<"AArch64ISD::TC_RETURN", SDT_AArch64TCRET,
385                  [SDNPHasChain,  SDNPOptInGlue, SDNPVariadic]>;
386
387def AArch64Prefetch        : SDNode<"AArch64ISD::PREFETCH", SDT_AArch64PREFETCH,
388                               [SDNPHasChain, SDNPSideEffect]>;
389
390def AArch64sitof: SDNode<"AArch64ISD::SITOF", SDT_AArch64ITOF>;
391def AArch64uitof: SDNode<"AArch64ISD::UITOF", SDT_AArch64ITOF>;
392
393def AArch64tlsdesc_callseq : SDNode<"AArch64ISD::TLSDESC_CALLSEQ",
394                                    SDT_AArch64TLSDescCallSeq,
395                                    [SDNPInGlue, SDNPOutGlue, SDNPHasChain,
396                                     SDNPVariadic]>;
397
398
399def AArch64WrapperLarge : SDNode<"AArch64ISD::WrapperLarge",
400                                 SDT_AArch64WrapperLarge>;
401
402def AArch64NvCast : SDNode<"AArch64ISD::NVCAST", SDTUnaryOp>;
403
404def SDT_AArch64mull : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
405                                    SDTCisSameAs<1, 2>]>;
406def AArch64smull    : SDNode<"AArch64ISD::SMULL", SDT_AArch64mull>;
407def AArch64umull    : SDNode<"AArch64ISD::UMULL", SDT_AArch64mull>;
408
409def AArch64frecpe   : SDNode<"AArch64ISD::FRECPE", SDTFPUnaryOp>;
410def AArch64frecps   : SDNode<"AArch64ISD::FRECPS", SDTFPBinOp>;
411def AArch64frsqrte  : SDNode<"AArch64ISD::FRSQRTE", SDTFPUnaryOp>;
412def AArch64frsqrts  : SDNode<"AArch64ISD::FRSQRTS", SDTFPBinOp>;
413
414def AArch64saddv    : SDNode<"AArch64ISD::SADDV", SDT_AArch64UnaryVec>;
415def AArch64uaddv    : SDNode<"AArch64ISD::UADDV", SDT_AArch64UnaryVec>;
416def AArch64sminv    : SDNode<"AArch64ISD::SMINV", SDT_AArch64UnaryVec>;
417def AArch64uminv    : SDNode<"AArch64ISD::UMINV", SDT_AArch64UnaryVec>;
418def AArch64smaxv    : SDNode<"AArch64ISD::SMAXV", SDT_AArch64UnaryVec>;
419def AArch64umaxv    : SDNode<"AArch64ISD::UMAXV", SDT_AArch64UnaryVec>;
420
421def SDT_AArch64SETTAG : SDTypeProfile<0, 2, [SDTCisPtrTy<0>, SDTCisPtrTy<1>]>;
422def AArch64stg : SDNode<"AArch64ISD::STG", SDT_AArch64SETTAG, [SDNPHasChain, SDNPMayStore, SDNPMemOperand]>;
423def AArch64stzg : SDNode<"AArch64ISD::STZG", SDT_AArch64SETTAG, [SDNPHasChain, SDNPMayStore, SDNPMemOperand]>;
424def AArch64st2g : SDNode<"AArch64ISD::ST2G", SDT_AArch64SETTAG, [SDNPHasChain, SDNPMayStore, SDNPMemOperand]>;
425def AArch64stz2g : SDNode<"AArch64ISD::STZ2G", SDT_AArch64SETTAG, [SDNPHasChain, SDNPMayStore, SDNPMemOperand]>;
426
427def SDT_AArch64unpk : SDTypeProfile<1, 1, [
428    SDTCisInt<0>, SDTCisInt<1>, SDTCisOpSmallerThanOp<1, 0>
429]>;
430def AArch64sunpkhi : SDNode<"AArch64ISD::SUNPKHI", SDT_AArch64unpk>;
431def AArch64sunpklo : SDNode<"AArch64ISD::SUNPKLO", SDT_AArch64unpk>;
432def AArch64uunpkhi : SDNode<"AArch64ISD::UUNPKHI", SDT_AArch64unpk>;
433def AArch64uunpklo : SDNode<"AArch64ISD::UUNPKLO", SDT_AArch64unpk>;
434
435//===----------------------------------------------------------------------===//
436
437//===----------------------------------------------------------------------===//
438
439// AArch64 Instruction Predicate Definitions.
440// We could compute these on a per-module basis but doing so requires accessing
441// the Function object through the <Target>Subtarget and objections were raised
442// to that (see post-commit review comments for r301750).
443let RecomputePerFunction = 1 in {
444  def ForCodeSize   : Predicate<"MF->getFunction().hasOptSize()">;
445  def NotForCodeSize   : Predicate<"!MF->getFunction().hasOptSize()">;
446  // Avoid generating STRQro if it is slow, unless we're optimizing for code size.
447  def UseSTRQro : Predicate<"!Subtarget->isSTRQroSlow() || MF->getFunction().hasOptSize()">;
448
449  def UseBTI : Predicate<[{ MF->getFunction().hasFnAttribute("branch-target-enforcement") }]>;
450  def NotUseBTI : Predicate<[{ !MF->getFunction().hasFnAttribute("branch-target-enforcement") }]>;
451
452  // Toggles patterns which aren't beneficial in GlobalISel when we aren't
453  // optimizing. This allows us to selectively use patterns without impacting
454  // SelectionDAG's behaviour.
455  // FIXME: One day there will probably be a nicer way to check for this, but
456  // today is not that day.
457  def OptimizedGISelOrOtherSelector : Predicate<"!MF->getFunction().hasOptNone() || MF->getProperties().hasProperty(MachineFunctionProperties::Property::FailedISel) || !MF->getProperties().hasProperty(MachineFunctionProperties::Property::Legalized)">;
458}
459
460include "AArch64InstrFormats.td"
461include "SVEInstrFormats.td"
462
463//===----------------------------------------------------------------------===//
464
465//===----------------------------------------------------------------------===//
466// Miscellaneous instructions.
467//===----------------------------------------------------------------------===//
468
469let Defs = [SP], Uses = [SP], hasSideEffects = 1, isCodeGenOnly = 1 in {
470// We set Sched to empty list because we expect these instructions to simply get
471// removed in most cases.
472def ADJCALLSTACKDOWN : Pseudo<(outs), (ins i32imm:$amt1, i32imm:$amt2),
473                              [(AArch64callseq_start timm:$amt1, timm:$amt2)]>,
474                              Sched<[]>;
475def ADJCALLSTACKUP : Pseudo<(outs), (ins i32imm:$amt1, i32imm:$amt2),
476                            [(AArch64callseq_end timm:$amt1, timm:$amt2)]>,
477                            Sched<[]>;
478} // Defs = [SP], Uses = [SP], hasSideEffects = 1, isCodeGenOnly = 1
479
480let isReMaterializable = 1, isCodeGenOnly = 1 in {
481// FIXME: The following pseudo instructions are only needed because remat
482// cannot handle multiple instructions.  When that changes, they can be
483// removed, along with the AArch64Wrapper node.
484
485let AddedComplexity = 10 in
486def LOADgot : Pseudo<(outs GPR64:$dst), (ins i64imm:$addr),
487                     [(set GPR64:$dst, (AArch64LOADgot tglobaladdr:$addr))]>,
488              Sched<[WriteLDAdr]>;
489
490// The MOVaddr instruction should match only when the add is not folded
491// into a load or store address.
492def MOVaddr
493    : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
494             [(set GPR64:$dst, (AArch64addlow (AArch64adrp tglobaladdr:$hi),
495                                            tglobaladdr:$low))]>,
496      Sched<[WriteAdrAdr]>;
497def MOVaddrJT
498    : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
499             [(set GPR64:$dst, (AArch64addlow (AArch64adrp tjumptable:$hi),
500                                             tjumptable:$low))]>,
501      Sched<[WriteAdrAdr]>;
502def MOVaddrCP
503    : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
504             [(set GPR64:$dst, (AArch64addlow (AArch64adrp tconstpool:$hi),
505                                             tconstpool:$low))]>,
506      Sched<[WriteAdrAdr]>;
507def MOVaddrBA
508    : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
509             [(set GPR64:$dst, (AArch64addlow (AArch64adrp tblockaddress:$hi),
510                                             tblockaddress:$low))]>,
511      Sched<[WriteAdrAdr]>;
512def MOVaddrTLS
513    : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
514             [(set GPR64:$dst, (AArch64addlow (AArch64adrp tglobaltlsaddr:$hi),
515                                            tglobaltlsaddr:$low))]>,
516      Sched<[WriteAdrAdr]>;
517def MOVaddrEXT
518    : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
519             [(set GPR64:$dst, (AArch64addlow (AArch64adrp texternalsym:$hi),
520                                            texternalsym:$low))]>,
521      Sched<[WriteAdrAdr]>;
522// Normally AArch64addlow either gets folded into a following ldr/str,
523// or together with an adrp into MOVaddr above. For cases with TLS, it
524// might appear without either of them, so allow lowering it into a plain
525// add.
526def ADDlowTLS
527    : Pseudo<(outs GPR64:$dst), (ins GPR64:$src, i64imm:$low),
528             [(set GPR64:$dst, (AArch64addlow GPR64:$src,
529                                            tglobaltlsaddr:$low))]>,
530      Sched<[WriteAdr]>;
531
532} // isReMaterializable, isCodeGenOnly
533
534def : Pat<(AArch64LOADgot tglobaltlsaddr:$addr),
535          (LOADgot tglobaltlsaddr:$addr)>;
536
537def : Pat<(AArch64LOADgot texternalsym:$addr),
538          (LOADgot texternalsym:$addr)>;
539
540def : Pat<(AArch64LOADgot tconstpool:$addr),
541          (LOADgot tconstpool:$addr)>;
542
543// 32-bit jump table destination is actually only 2 instructions since we can
544// use the table itself as a PC-relative base. But optimization occurs after
545// branch relaxation so be pessimistic.
546let Size = 12, Constraints = "@earlyclobber $dst,@earlyclobber $scratch" in {
547def JumpTableDest32 : Pseudo<(outs GPR64:$dst, GPR64sp:$scratch),
548                             (ins GPR64:$table, GPR64:$entry, i32imm:$jti), []>,
549                      Sched<[]>;
550def JumpTableDest16 : Pseudo<(outs GPR64:$dst, GPR64sp:$scratch),
551                             (ins GPR64:$table, GPR64:$entry, i32imm:$jti), []>,
552                      Sched<[]>;
553def JumpTableDest8 : Pseudo<(outs GPR64:$dst, GPR64sp:$scratch),
554                            (ins GPR64:$table, GPR64:$entry, i32imm:$jti), []>,
555                     Sched<[]>;
556}
557
558// Space-consuming pseudo to aid testing of placement and reachability
559// algorithms. Immediate operand is the number of bytes this "instruction"
560// occupies; register operands can be used to enforce dependency and constrain
561// the scheduler.
562let hasSideEffects = 1, mayLoad = 1, mayStore = 1 in
563def SPACE : Pseudo<(outs GPR64:$Rd), (ins i32imm:$size, GPR64:$Rn),
564                   [(set GPR64:$Rd, (int_aarch64_space imm:$size, GPR64:$Rn))]>,
565            Sched<[]>;
566
567let hasSideEffects = 1, isCodeGenOnly = 1 in {
568  def SpeculationSafeValueX
569      : Pseudo<(outs GPR64:$dst), (ins GPR64:$src), []>, Sched<[]>;
570  def SpeculationSafeValueW
571      : Pseudo<(outs GPR32:$dst), (ins GPR32:$src), []>, Sched<[]>;
572}
573
574
575//===----------------------------------------------------------------------===//
576// System instructions.
577//===----------------------------------------------------------------------===//
578
579def HINT : HintI<"hint">;
580def : InstAlias<"nop",  (HINT 0b000)>;
581def : InstAlias<"yield",(HINT 0b001)>;
582def : InstAlias<"wfe",  (HINT 0b010)>;
583def : InstAlias<"wfi",  (HINT 0b011)>;
584def : InstAlias<"sev",  (HINT 0b100)>;
585def : InstAlias<"sevl", (HINT 0b101)>;
586def : InstAlias<"esb",  (HINT 0b10000)>, Requires<[HasRAS]>;
587def : InstAlias<"csdb", (HINT 20)>;
588def : InstAlias<"bti",  (HINT 32)>, Requires<[HasBTI]>;
589def : InstAlias<"bti $op", (HINT btihint_op:$op)>, Requires<[HasBTI]>;
590
591// v8.2a Statistical Profiling extension
592def : InstAlias<"psb $op",  (HINT psbhint_op:$op)>, Requires<[HasSPE]>;
593
594// As far as LLVM is concerned this writes to the system's exclusive monitors.
595let mayLoad = 1, mayStore = 1 in
596def CLREX : CRmSystemI<imm0_15, 0b010, "clrex">;
597
598// NOTE: ideally, this would have mayStore = 0, mayLoad = 0, but we cannot
599// model patterns with sufficiently fine granularity.
600let mayLoad = ?, mayStore = ? in {
601def DMB   : CRmSystemI<barrier_op, 0b101, "dmb",
602                       [(int_aarch64_dmb (i32 imm32_0_15:$CRm))]>;
603
604def DSB   : CRmSystemI<barrier_op, 0b100, "dsb",
605                       [(int_aarch64_dsb (i32 imm32_0_15:$CRm))]>;
606
607def ISB   : CRmSystemI<barrier_op, 0b110, "isb",
608                       [(int_aarch64_isb (i32 imm32_0_15:$CRm))]>;
609
610def TSB   : CRmSystemI<barrier_op, 0b010, "tsb", []> {
611  let CRm        = 0b0010;
612  let Inst{12}   = 0;
613  let Predicates = [HasTRACEV8_4];
614}
615}
616
617// ARMv8.2-A Dot Product
618let Predicates = [HasDotProd] in {
619defm SDOT : SIMDThreeSameVectorDot<0, "sdot", int_aarch64_neon_sdot>;
620defm UDOT : SIMDThreeSameVectorDot<1, "udot", int_aarch64_neon_udot>;
621defm SDOTlane : SIMDThreeSameVectorDotIndex<0, "sdot", int_aarch64_neon_sdot>;
622defm UDOTlane : SIMDThreeSameVectorDotIndex<1, "udot", int_aarch64_neon_udot>;
623}
624
625// ARMv8.2-A FP16 Fused Multiply-Add Long
626let Predicates = [HasNEON, HasFP16FML] in {
627defm FMLAL      : SIMDThreeSameVectorFML<0, 1, 0b001, "fmlal", int_aarch64_neon_fmlal>;
628defm FMLSL      : SIMDThreeSameVectorFML<0, 1, 0b101, "fmlsl", int_aarch64_neon_fmlsl>;
629defm FMLAL2     : SIMDThreeSameVectorFML<1, 0, 0b001, "fmlal2", int_aarch64_neon_fmlal2>;
630defm FMLSL2     : SIMDThreeSameVectorFML<1, 0, 0b101, "fmlsl2", int_aarch64_neon_fmlsl2>;
631defm FMLALlane  : SIMDThreeSameVectorFMLIndex<0, 0b0000, "fmlal", int_aarch64_neon_fmlal>;
632defm FMLSLlane  : SIMDThreeSameVectorFMLIndex<0, 0b0100, "fmlsl", int_aarch64_neon_fmlsl>;
633defm FMLAL2lane : SIMDThreeSameVectorFMLIndex<1, 0b1000, "fmlal2", int_aarch64_neon_fmlal2>;
634defm FMLSL2lane : SIMDThreeSameVectorFMLIndex<1, 0b1100, "fmlsl2", int_aarch64_neon_fmlsl2>;
635}
636
637// Armv8.2-A Crypto extensions
638let Predicates = [HasSHA3] in {
639def SHA512H   : CryptoRRRTied<0b0, 0b00, "sha512h">;
640def SHA512H2  : CryptoRRRTied<0b0, 0b01, "sha512h2">;
641def SHA512SU0 : CryptoRRTied_2D<0b0, 0b00, "sha512su0">;
642def SHA512SU1 : CryptoRRRTied_2D<0b0, 0b10, "sha512su1">;
643def RAX1      : CryptoRRR_2D<0b0,0b11, "rax1">;
644def EOR3      : CryptoRRRR_16B<0b00, "eor3">;
645def BCAX      : CryptoRRRR_16B<0b01, "bcax">;
646def XAR       : CryptoRRRi6<"xar">;
647} // HasSHA3
648
649let Predicates = [HasSM4] in {
650def SM3TT1A   : CryptoRRRi2Tied<0b0, 0b00, "sm3tt1a">;
651def SM3TT1B   : CryptoRRRi2Tied<0b0, 0b01, "sm3tt1b">;
652def SM3TT2A   : CryptoRRRi2Tied<0b0, 0b10, "sm3tt2a">;
653def SM3TT2B   : CryptoRRRi2Tied<0b0, 0b11, "sm3tt2b">;
654def SM3SS1    : CryptoRRRR_4S<0b10, "sm3ss1">;
655def SM3PARTW1 : CryptoRRRTied_4S<0b1, 0b00, "sm3partw1">;
656def SM3PARTW2 : CryptoRRRTied_4S<0b1, 0b01, "sm3partw2">;
657def SM4ENCKEY : CryptoRRR_4S<0b1, 0b10, "sm4ekey">;
658def SM4E      : CryptoRRTied_4S<0b0, 0b01, "sm4e">;
659} // HasSM4
660
661let Predicates = [HasRCPC] in {
662  // v8.3 Release Consistent Processor Consistent support, optional in v8.2.
663  def LDAPRB  : RCPCLoad<0b00, "ldaprb", GPR32>;
664  def LDAPRH  : RCPCLoad<0b01, "ldaprh", GPR32>;
665  def LDAPRW  : RCPCLoad<0b10, "ldapr", GPR32>;
666  def LDAPRX  : RCPCLoad<0b11, "ldapr", GPR64>;
667}
668
669// v8.3a complex add and multiply-accumulate. No predicate here, that is done
670// inside the multiclass as the FP16 versions need different predicates.
671defm FCMLA : SIMDThreeSameVectorTiedComplexHSD<1, 0b110, complexrotateop,
672                                               "fcmla", null_frag>;
673defm FCADD : SIMDThreeSameVectorComplexHSD<1, 0b111, complexrotateopodd,
674                                           "fcadd", null_frag>;
675defm FCMLA : SIMDIndexedTiedComplexHSD<1, 0, 1, complexrotateop, "fcmla",
676                                       null_frag>;
677
678// v8.3a Pointer Authentication
679// These instructions inhabit part of the hint space and so can be used for
680// armv8 targets
681let Uses = [LR], Defs = [LR] in {
682  def PACIAZ   : SystemNoOperands<0b000, "paciaz">;
683  def PACIBZ   : SystemNoOperands<0b010, "pacibz">;
684  def AUTIAZ   : SystemNoOperands<0b100, "autiaz">;
685  def AUTIBZ   : SystemNoOperands<0b110, "autibz">;
686}
687let Uses = [LR, SP], Defs = [LR] in {
688  def PACIASP  : SystemNoOperands<0b001, "paciasp">;
689  def PACIBSP  : SystemNoOperands<0b011, "pacibsp">;
690  def AUTIASP  : SystemNoOperands<0b101, "autiasp">;
691  def AUTIBSP  : SystemNoOperands<0b111, "autibsp">;
692}
693let Uses = [X16, X17], Defs = [X17], CRm = 0b0001 in {
694  def PACIA1716  : SystemNoOperands<0b000, "pacia1716">;
695  def PACIB1716  : SystemNoOperands<0b010, "pacib1716">;
696  def AUTIA1716  : SystemNoOperands<0b100, "autia1716">;
697  def AUTIB1716  : SystemNoOperands<0b110, "autib1716">;
698}
699
700let Uses = [LR], Defs = [LR], CRm = 0b0000 in {
701  def XPACLRI   : SystemNoOperands<0b111, "xpaclri">;
702}
703
704// These pointer authentication isntructions require armv8.3a
705let Predicates = [HasPA] in {
706  multiclass SignAuth<bits<3> prefix, bits<3> prefix_z, string asm> {
707    def IA   : SignAuthOneData<prefix, 0b00, !strconcat(asm, "ia")>;
708    def IB   : SignAuthOneData<prefix, 0b01, !strconcat(asm, "ib")>;
709    def DA   : SignAuthOneData<prefix, 0b10, !strconcat(asm, "da")>;
710    def DB   : SignAuthOneData<prefix, 0b11, !strconcat(asm, "db")>;
711    def IZA  : SignAuthZero<prefix_z, 0b00, !strconcat(asm, "iza")>;
712    def DZA  : SignAuthZero<prefix_z, 0b10, !strconcat(asm, "dza")>;
713    def IZB  : SignAuthZero<prefix_z, 0b01, !strconcat(asm, "izb")>;
714    def DZB  : SignAuthZero<prefix_z, 0b11, !strconcat(asm, "dzb")>;
715  }
716
717  defm PAC : SignAuth<0b000, 0b010, "pac">;
718  defm AUT : SignAuth<0b001, 0b011, "aut">;
719
720  def XPACI : SignAuthZero<0b100, 0b00, "xpaci">;
721  def XPACD : SignAuthZero<0b100, 0b01, "xpacd">;
722  def PACGA : SignAuthTwoOperand<0b1100, "pacga", null_frag>;
723
724  // Combined Instructions
725  def BRAA    : AuthBranchTwoOperands<0, 0, "braa">;
726  def BRAB    : AuthBranchTwoOperands<0, 1, "brab">;
727  def BLRAA   : AuthBranchTwoOperands<1, 0, "blraa">;
728  def BLRAB   : AuthBranchTwoOperands<1, 1, "blrab">;
729
730  def BRAAZ   : AuthOneOperand<0b000, 0, "braaz">;
731  def BRABZ   : AuthOneOperand<0b000, 1, "brabz">;
732  def BLRAAZ  : AuthOneOperand<0b001, 0, "blraaz">;
733  def BLRABZ  : AuthOneOperand<0b001, 1, "blrabz">;
734
735  let isReturn = 1, isTerminator = 1, isBarrier = 1 in {
736    def RETAA   : AuthReturn<0b010, 0, "retaa">;
737    def RETAB   : AuthReturn<0b010, 1, "retab">;
738    def ERETAA  : AuthReturn<0b100, 0, "eretaa">;
739    def ERETAB  : AuthReturn<0b100, 1, "eretab">;
740  }
741
742  defm LDRAA  : AuthLoad<0, "ldraa", simm10Scaled>;
743  defm LDRAB  : AuthLoad<1, "ldrab", simm10Scaled>;
744
745}
746
747// v8.3a floating point conversion for javascript
748let Predicates = [HasJS, HasFPARMv8] in
749def FJCVTZS  : BaseFPToIntegerUnscaled<0b01, 0b11, 0b110, FPR64, GPR32,
750                                      "fjcvtzs",
751                                      [(set GPR32:$Rd,
752                                         (int_aarch64_fjcvtzs FPR64:$Rn))]> {
753  let Inst{31} = 0;
754} // HasJS, HasFPARMv8
755
756// v8.4 Flag manipulation instructions
757let Predicates = [HasFMI] in {
758def CFINV : SimpleSystemI<0, (ins), "cfinv", "">, Sched<[WriteSys]> {
759  let Inst{20-5} = 0b0000001000000000;
760}
761def SETF8  : BaseFlagManipulation<0, 0, (ins GPR32:$Rn), "setf8", "{\t$Rn}">;
762def SETF16 : BaseFlagManipulation<0, 1, (ins GPR32:$Rn), "setf16", "{\t$Rn}">;
763def RMIF   : FlagRotate<(ins GPR64:$Rn, uimm6:$imm, imm0_15:$mask), "rmif",
764                        "{\t$Rn, $imm, $mask}">;
765} // HasFMI
766
767// v8.5 flag manipulation instructions
768let Predicates = [HasAltNZCV], Uses = [NZCV], Defs = [NZCV] in {
769
770def XAFLAG : PstateWriteSimple<(ins), "xaflag", "">, Sched<[WriteSys]> {
771  let Inst{18-16} = 0b000;
772  let Inst{11-8} = 0b0000;
773  let Unpredictable{11-8} = 0b1111;
774  let Inst{7-5} = 0b001;
775}
776
777def AXFLAG : PstateWriteSimple<(ins), "axflag", "">, Sched<[WriteSys]> {
778  let Inst{18-16} = 0b000;
779  let Inst{11-8} = 0b0000;
780  let Unpredictable{11-8} = 0b1111;
781  let Inst{7-5} = 0b010;
782}
783} // HasAltNZCV
784
785
786// Armv8.5-A speculation barrier
787def SB : SimpleSystemI<0, (ins), "sb", "">, Sched<[]> {
788  let Inst{20-5} = 0b0001100110000111;
789  let Unpredictable{11-8} = 0b1111;
790  let Predicates = [HasSB];
791  let hasSideEffects = 1;
792}
793
794def : InstAlias<"clrex", (CLREX 0xf)>;
795def : InstAlias<"isb", (ISB 0xf)>;
796def : InstAlias<"ssbb", (DSB 0)>;
797def : InstAlias<"pssbb", (DSB 4)>;
798
799def MRS    : MRSI;
800def MSR    : MSRI;
801def MSRpstateImm1 : MSRpstateImm0_1;
802def MSRpstateImm4 : MSRpstateImm0_15;
803
804// The thread pointer (on Linux, at least, where this has been implemented) is
805// TPIDR_EL0.
806def MOVbaseTLS : Pseudo<(outs GPR64:$dst), (ins),
807                       [(set GPR64:$dst, AArch64threadpointer)]>, Sched<[WriteSys]>;
808
809let Uses = [ X9 ], Defs = [ X16, X17, LR, NZCV ] in {
810def HWASAN_CHECK_MEMACCESS : Pseudo<
811  (outs), (ins GPR64noip:$ptr, i32imm:$accessinfo),
812  [(int_hwasan_check_memaccess X9, GPR64noip:$ptr, (i32 timm:$accessinfo))]>,
813  Sched<[]>;
814def HWASAN_CHECK_MEMACCESS_SHORTGRANULES : Pseudo<
815  (outs), (ins GPR64noip:$ptr, i32imm:$accessinfo),
816  [(int_hwasan_check_memaccess_shortgranules X9, GPR64noip:$ptr, (i32 timm:$accessinfo))]>,
817  Sched<[]>;
818}
819
820// The cycle counter PMC register is PMCCNTR_EL0.
821let Predicates = [HasPerfMon] in
822def : Pat<(readcyclecounter), (MRS 0xdce8)>;
823
824// FPCR register
825def : Pat<(i64 (int_aarch64_get_fpcr)), (MRS 0xda20)>;
826
827// Generic system instructions
828def SYSxt  : SystemXtI<0, "sys">;
829def SYSLxt : SystemLXtI<1, "sysl">;
830
831def : InstAlias<"sys $op1, $Cn, $Cm, $op2",
832                (SYSxt imm0_7:$op1, sys_cr_op:$Cn,
833                 sys_cr_op:$Cm, imm0_7:$op2, XZR)>;
834
835
836let Predicates = [HasTME] in {
837
838def TSTART : TMSystemI<0b0000, "tstart",
839                      [(set GPR64:$Rt, (int_aarch64_tstart))]>;
840
841def TCOMMIT : TMSystemINoOperand<0b0000, "tcommit", [(int_aarch64_tcommit)]>;
842
843def TCANCEL : TMSystemException<0b011, "tcancel",
844                                [(int_aarch64_tcancel i64_imm0_65535:$imm)]>;
845
846def TTEST : TMSystemI<0b0001, "ttest", [(set GPR64:$Rt, (int_aarch64_ttest))]> {
847  let mayLoad = 0;
848  let mayStore = 0;
849}
850} // HasTME
851
852//===----------------------------------------------------------------------===//
853// Move immediate instructions.
854//===----------------------------------------------------------------------===//
855
856defm MOVK : InsertImmediate<0b11, "movk">;
857defm MOVN : MoveImmediate<0b00, "movn">;
858
859let PostEncoderMethod = "fixMOVZ" in
860defm MOVZ : MoveImmediate<0b10, "movz">;
861
862// First group of aliases covers an implicit "lsl #0".
863def : InstAlias<"movk $dst, $imm", (MOVKWi GPR32:$dst, i32_imm0_65535:$imm, 0), 0>;
864def : InstAlias<"movk $dst, $imm", (MOVKXi GPR64:$dst, i32_imm0_65535:$imm, 0), 0>;
865def : InstAlias<"movn $dst, $imm", (MOVNWi GPR32:$dst, i32_imm0_65535:$imm, 0)>;
866def : InstAlias<"movn $dst, $imm", (MOVNXi GPR64:$dst, i32_imm0_65535:$imm, 0)>;
867def : InstAlias<"movz $dst, $imm", (MOVZWi GPR32:$dst, i32_imm0_65535:$imm, 0)>;
868def : InstAlias<"movz $dst, $imm", (MOVZXi GPR64:$dst, i32_imm0_65535:$imm, 0)>;
869
870// Next, we have various ELF relocations with the ":XYZ_g0:sym" syntax.
871def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movw_symbol_g3:$sym, 48)>;
872def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movw_symbol_g2:$sym, 32)>;
873def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movw_symbol_g1:$sym, 16)>;
874def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movw_symbol_g0:$sym, 0)>;
875
876def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movw_symbol_g3:$sym, 48)>;
877def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movw_symbol_g2:$sym, 32)>;
878def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movw_symbol_g1:$sym, 16)>;
879def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movw_symbol_g0:$sym, 0)>;
880
881def : InstAlias<"movk $Rd, $sym", (MOVKXi GPR64:$Rd, movw_symbol_g3:$sym, 48), 0>;
882def : InstAlias<"movk $Rd, $sym", (MOVKXi GPR64:$Rd, movw_symbol_g2:$sym, 32), 0>;
883def : InstAlias<"movk $Rd, $sym", (MOVKXi GPR64:$Rd, movw_symbol_g1:$sym, 16), 0>;
884def : InstAlias<"movk $Rd, $sym", (MOVKXi GPR64:$Rd, movw_symbol_g0:$sym, 0), 0>;
885
886def : InstAlias<"movz $Rd, $sym", (MOVZWi GPR32:$Rd, movw_symbol_g1:$sym, 16)>;
887def : InstAlias<"movz $Rd, $sym", (MOVZWi GPR32:$Rd, movw_symbol_g0:$sym, 0)>;
888
889def : InstAlias<"movn $Rd, $sym", (MOVNWi GPR32:$Rd, movw_symbol_g1:$sym, 16)>;
890def : InstAlias<"movn $Rd, $sym", (MOVNWi GPR32:$Rd, movw_symbol_g0:$sym, 0)>;
891
892def : InstAlias<"movk $Rd, $sym", (MOVKWi GPR32:$Rd, movw_symbol_g1:$sym, 16), 0>;
893def : InstAlias<"movk $Rd, $sym", (MOVKWi GPR32:$Rd, movw_symbol_g0:$sym, 0), 0>;
894
895// Final group of aliases covers true "mov $Rd, $imm" cases.
896multiclass movw_mov_alias<string basename,Instruction INST, RegisterClass GPR,
897                          int width, int shift> {
898  def _asmoperand : AsmOperandClass {
899    let Name = basename # width # "_lsl" # shift # "MovAlias";
900    let PredicateMethod = "is" # basename # "MovAlias<" # width # ", "
901                               # shift # ">";
902    let RenderMethod = "add" # basename # "MovAliasOperands<" # shift # ">";
903  }
904
905  def _movimm : Operand<i32> {
906    let ParserMatchClass = !cast<AsmOperandClass>(NAME # "_asmoperand");
907  }
908
909  def : InstAlias<"mov $Rd, $imm",
910                  (INST GPR:$Rd, !cast<Operand>(NAME # "_movimm"):$imm, shift)>;
911}
912
913defm : movw_mov_alias<"MOVZ", MOVZWi, GPR32, 32, 0>;
914defm : movw_mov_alias<"MOVZ", MOVZWi, GPR32, 32, 16>;
915
916defm : movw_mov_alias<"MOVZ", MOVZXi, GPR64, 64, 0>;
917defm : movw_mov_alias<"MOVZ", MOVZXi, GPR64, 64, 16>;
918defm : movw_mov_alias<"MOVZ", MOVZXi, GPR64, 64, 32>;
919defm : movw_mov_alias<"MOVZ", MOVZXi, GPR64, 64, 48>;
920
921defm : movw_mov_alias<"MOVN", MOVNWi, GPR32, 32, 0>;
922defm : movw_mov_alias<"MOVN", MOVNWi, GPR32, 32, 16>;
923
924defm : movw_mov_alias<"MOVN", MOVNXi, GPR64, 64, 0>;
925defm : movw_mov_alias<"MOVN", MOVNXi, GPR64, 64, 16>;
926defm : movw_mov_alias<"MOVN", MOVNXi, GPR64, 64, 32>;
927defm : movw_mov_alias<"MOVN", MOVNXi, GPR64, 64, 48>;
928
929let isReMaterializable = 1, isCodeGenOnly = 1, isMoveImm = 1,
930    isAsCheapAsAMove = 1 in {
931// FIXME: The following pseudo instructions are only needed because remat
932// cannot handle multiple instructions.  When that changes, we can select
933// directly to the real instructions and get rid of these pseudos.
934
935def MOVi32imm
936    : Pseudo<(outs GPR32:$dst), (ins i32imm:$src),
937             [(set GPR32:$dst, imm:$src)]>,
938      Sched<[WriteImm]>;
939def MOVi64imm
940    : Pseudo<(outs GPR64:$dst), (ins i64imm:$src),
941             [(set GPR64:$dst, imm:$src)]>,
942      Sched<[WriteImm]>;
943} // isReMaterializable, isCodeGenOnly
944
945// If possible, we want to use MOVi32imm even for 64-bit moves. This gives the
946// eventual expansion code fewer bits to worry about getting right. Marshalling
947// the types is a little tricky though:
948def i64imm_32bit : ImmLeaf<i64, [{
949  return (Imm & 0xffffffffULL) == static_cast<uint64_t>(Imm);
950}]>;
951
952def s64imm_32bit : ImmLeaf<i64, [{
953  int64_t Imm64 = static_cast<int64_t>(Imm);
954  return Imm64 >= std::numeric_limits<int32_t>::min() &&
955         Imm64 <= std::numeric_limits<int32_t>::max();
956}]>;
957
958def trunc_imm : SDNodeXForm<imm, [{
959  return CurDAG->getTargetConstant(N->getZExtValue(), SDLoc(N), MVT::i32);
960}]>;
961
962def gi_trunc_imm : GICustomOperandRenderer<"renderTruncImm">,
963  GISDNodeXFormEquiv<trunc_imm>;
964
965let Predicates = [OptimizedGISelOrOtherSelector] in {
966// The SUBREG_TO_REG isn't eliminated at -O0, which can result in pointless
967// copies.
968def : Pat<(i64 i64imm_32bit:$src),
969          (SUBREG_TO_REG (i64 0), (MOVi32imm (trunc_imm imm:$src)), sub_32)>;
970}
971
972// Materialize FP constants via MOVi32imm/MOVi64imm (MachO large code model).
973def bitcast_fpimm_to_i32 : SDNodeXForm<fpimm, [{
974return CurDAG->getTargetConstant(
975  N->getValueAPF().bitcastToAPInt().getZExtValue(), SDLoc(N), MVT::i32);
976}]>;
977
978def bitcast_fpimm_to_i64 : SDNodeXForm<fpimm, [{
979return CurDAG->getTargetConstant(
980  N->getValueAPF().bitcastToAPInt().getZExtValue(), SDLoc(N), MVT::i64);
981}]>;
982
983
984def : Pat<(f32 fpimm:$in),
985  (COPY_TO_REGCLASS (MOVi32imm (bitcast_fpimm_to_i32 f32:$in)), FPR32)>;
986def : Pat<(f64 fpimm:$in),
987  (COPY_TO_REGCLASS (MOVi64imm (bitcast_fpimm_to_i64 f64:$in)), FPR64)>;
988
989
990// Deal with the various forms of (ELF) large addressing with MOVZ/MOVK
991// sequences.
992def : Pat<(AArch64WrapperLarge tglobaladdr:$g3, tglobaladdr:$g2,
993                             tglobaladdr:$g1, tglobaladdr:$g0),
994          (MOVKXi (MOVKXi (MOVKXi (MOVZXi tglobaladdr:$g0, 0),
995                                  tglobaladdr:$g1, 16),
996                          tglobaladdr:$g2, 32),
997                  tglobaladdr:$g3, 48)>;
998
999def : Pat<(AArch64WrapperLarge tblockaddress:$g3, tblockaddress:$g2,
1000                             tblockaddress:$g1, tblockaddress:$g0),
1001          (MOVKXi (MOVKXi (MOVKXi (MOVZXi tblockaddress:$g0, 0),
1002                                  tblockaddress:$g1, 16),
1003                          tblockaddress:$g2, 32),
1004                  tblockaddress:$g3, 48)>;
1005
1006def : Pat<(AArch64WrapperLarge tconstpool:$g3, tconstpool:$g2,
1007                             tconstpool:$g1, tconstpool:$g0),
1008          (MOVKXi (MOVKXi (MOVKXi (MOVZXi tconstpool:$g0, 0),
1009                                  tconstpool:$g1, 16),
1010                          tconstpool:$g2, 32),
1011                  tconstpool:$g3, 48)>;
1012
1013def : Pat<(AArch64WrapperLarge tjumptable:$g3, tjumptable:$g2,
1014                             tjumptable:$g1, tjumptable:$g0),
1015          (MOVKXi (MOVKXi (MOVKXi (MOVZXi tjumptable:$g0, 0),
1016                                  tjumptable:$g1, 16),
1017                          tjumptable:$g2, 32),
1018                  tjumptable:$g3, 48)>;
1019
1020
1021//===----------------------------------------------------------------------===//
1022// Arithmetic instructions.
1023//===----------------------------------------------------------------------===//
1024
1025// Add/subtract with carry.
1026defm ADC : AddSubCarry<0, "adc", "adcs", AArch64adc, AArch64adc_flag>;
1027defm SBC : AddSubCarry<1, "sbc", "sbcs", AArch64sbc, AArch64sbc_flag>;
1028
1029def : InstAlias<"ngc $dst, $src",  (SBCWr  GPR32:$dst, WZR, GPR32:$src)>;
1030def : InstAlias<"ngc $dst, $src",  (SBCXr  GPR64:$dst, XZR, GPR64:$src)>;
1031def : InstAlias<"ngcs $dst, $src", (SBCSWr GPR32:$dst, WZR, GPR32:$src)>;
1032def : InstAlias<"ngcs $dst, $src", (SBCSXr GPR64:$dst, XZR, GPR64:$src)>;
1033
1034// Add/subtract
1035defm ADD : AddSub<0, "add", "sub", add>;
1036defm SUB : AddSub<1, "sub", "add">;
1037
1038def : InstAlias<"mov $dst, $src",
1039                (ADDWri GPR32sponly:$dst, GPR32sp:$src, 0, 0)>;
1040def : InstAlias<"mov $dst, $src",
1041                (ADDWri GPR32sp:$dst, GPR32sponly:$src, 0, 0)>;
1042def : InstAlias<"mov $dst, $src",
1043                (ADDXri GPR64sponly:$dst, GPR64sp:$src, 0, 0)>;
1044def : InstAlias<"mov $dst, $src",
1045                (ADDXri GPR64sp:$dst, GPR64sponly:$src, 0, 0)>;
1046
1047defm ADDS : AddSubS<0, "adds", AArch64add_flag, "cmn", "subs", "cmp">;
1048defm SUBS : AddSubS<1, "subs", AArch64sub_flag, "cmp", "adds", "cmn">;
1049
1050// Use SUBS instead of SUB to enable CSE between SUBS and SUB.
1051def : Pat<(sub GPR32sp:$Rn, addsub_shifted_imm32:$imm),
1052          (SUBSWri GPR32sp:$Rn, addsub_shifted_imm32:$imm)>;
1053def : Pat<(sub GPR64sp:$Rn, addsub_shifted_imm64:$imm),
1054          (SUBSXri GPR64sp:$Rn, addsub_shifted_imm64:$imm)>;
1055def : Pat<(sub GPR32:$Rn, GPR32:$Rm),
1056          (SUBSWrr GPR32:$Rn, GPR32:$Rm)>;
1057def : Pat<(sub GPR64:$Rn, GPR64:$Rm),
1058          (SUBSXrr GPR64:$Rn, GPR64:$Rm)>;
1059def : Pat<(sub GPR32:$Rn, arith_shifted_reg32:$Rm),
1060          (SUBSWrs GPR32:$Rn, arith_shifted_reg32:$Rm)>;
1061def : Pat<(sub GPR64:$Rn, arith_shifted_reg64:$Rm),
1062          (SUBSXrs GPR64:$Rn, arith_shifted_reg64:$Rm)>;
1063let AddedComplexity = 1 in {
1064def : Pat<(sub GPR32sp:$R2, arith_extended_reg32_i32:$R3),
1065          (SUBSWrx GPR32sp:$R2, arith_extended_reg32_i32:$R3)>;
1066def : Pat<(sub GPR64sp:$R2, arith_extended_reg32to64_i64:$R3),
1067          (SUBSXrx GPR64sp:$R2, arith_extended_reg32to64_i64:$R3)>;
1068}
1069
1070// Because of the immediate format for add/sub-imm instructions, the
1071// expression (add x, -1) must be transformed to (SUB{W,X}ri x, 1).
1072//  These patterns capture that transformation.
1073let AddedComplexity = 1 in {
1074def : Pat<(add GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
1075          (SUBSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
1076def : Pat<(add GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
1077          (SUBSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
1078def : Pat<(sub GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
1079          (ADDWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
1080def : Pat<(sub GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
1081          (ADDXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
1082}
1083
1084// Because of the immediate format for add/sub-imm instructions, the
1085// expression (add x, -1) must be transformed to (SUB{W,X}ri x, 1).
1086//  These patterns capture that transformation.
1087let AddedComplexity = 1 in {
1088def : Pat<(AArch64add_flag GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
1089          (SUBSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
1090def : Pat<(AArch64add_flag GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
1091          (SUBSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
1092def : Pat<(AArch64sub_flag GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
1093          (ADDSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
1094def : Pat<(AArch64sub_flag GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
1095          (ADDSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
1096}
1097
1098def : InstAlias<"neg $dst, $src", (SUBWrs GPR32:$dst, WZR, GPR32:$src, 0), 3>;
1099def : InstAlias<"neg $dst, $src", (SUBXrs GPR64:$dst, XZR, GPR64:$src, 0), 3>;
1100def : InstAlias<"neg $dst, $src$shift",
1101                (SUBWrs GPR32:$dst, WZR, GPR32:$src, arith_shift32:$shift), 2>;
1102def : InstAlias<"neg $dst, $src$shift",
1103                (SUBXrs GPR64:$dst, XZR, GPR64:$src, arith_shift64:$shift), 2>;
1104
1105def : InstAlias<"negs $dst, $src", (SUBSWrs GPR32:$dst, WZR, GPR32:$src, 0), 3>;
1106def : InstAlias<"negs $dst, $src", (SUBSXrs GPR64:$dst, XZR, GPR64:$src, 0), 3>;
1107def : InstAlias<"negs $dst, $src$shift",
1108                (SUBSWrs GPR32:$dst, WZR, GPR32:$src, arith_shift32:$shift), 2>;
1109def : InstAlias<"negs $dst, $src$shift",
1110                (SUBSXrs GPR64:$dst, XZR, GPR64:$src, arith_shift64:$shift), 2>;
1111
1112
1113// Unsigned/Signed divide
1114defm UDIV : Div<0, "udiv", udiv>;
1115defm SDIV : Div<1, "sdiv", sdiv>;
1116
1117def : Pat<(int_aarch64_udiv GPR32:$Rn, GPR32:$Rm), (UDIVWr GPR32:$Rn, GPR32:$Rm)>;
1118def : Pat<(int_aarch64_udiv GPR64:$Rn, GPR64:$Rm), (UDIVXr GPR64:$Rn, GPR64:$Rm)>;
1119def : Pat<(int_aarch64_sdiv GPR32:$Rn, GPR32:$Rm), (SDIVWr GPR32:$Rn, GPR32:$Rm)>;
1120def : Pat<(int_aarch64_sdiv GPR64:$Rn, GPR64:$Rm), (SDIVXr GPR64:$Rn, GPR64:$Rm)>;
1121
1122// Variable shift
1123defm ASRV : Shift<0b10, "asr", sra>;
1124defm LSLV : Shift<0b00, "lsl", shl>;
1125defm LSRV : Shift<0b01, "lsr", srl>;
1126defm RORV : Shift<0b11, "ror", rotr>;
1127
1128def : ShiftAlias<"asrv", ASRVWr, GPR32>;
1129def : ShiftAlias<"asrv", ASRVXr, GPR64>;
1130def : ShiftAlias<"lslv", LSLVWr, GPR32>;
1131def : ShiftAlias<"lslv", LSLVXr, GPR64>;
1132def : ShiftAlias<"lsrv", LSRVWr, GPR32>;
1133def : ShiftAlias<"lsrv", LSRVXr, GPR64>;
1134def : ShiftAlias<"rorv", RORVWr, GPR32>;
1135def : ShiftAlias<"rorv", RORVXr, GPR64>;
1136
1137// Multiply-add
1138let AddedComplexity = 5 in {
1139defm MADD : MulAccum<0, "madd", add>;
1140defm MSUB : MulAccum<1, "msub", sub>;
1141
1142def : Pat<(i32 (mul GPR32:$Rn, GPR32:$Rm)),
1143          (MADDWrrr GPR32:$Rn, GPR32:$Rm, WZR)>;
1144def : Pat<(i64 (mul GPR64:$Rn, GPR64:$Rm)),
1145          (MADDXrrr GPR64:$Rn, GPR64:$Rm, XZR)>;
1146
1147def : Pat<(i32 (ineg (mul GPR32:$Rn, GPR32:$Rm))),
1148          (MSUBWrrr GPR32:$Rn, GPR32:$Rm, WZR)>;
1149def : Pat<(i64 (ineg (mul GPR64:$Rn, GPR64:$Rm))),
1150          (MSUBXrrr GPR64:$Rn, GPR64:$Rm, XZR)>;
1151def : Pat<(i32 (mul (ineg GPR32:$Rn), GPR32:$Rm)),
1152          (MSUBWrrr GPR32:$Rn, GPR32:$Rm, WZR)>;
1153def : Pat<(i64 (mul (ineg GPR64:$Rn), GPR64:$Rm)),
1154          (MSUBXrrr GPR64:$Rn, GPR64:$Rm, XZR)>;
1155} // AddedComplexity = 5
1156
1157let AddedComplexity = 5 in {
1158def SMADDLrrr : WideMulAccum<0, 0b001, "smaddl", add, sext>;
1159def SMSUBLrrr : WideMulAccum<1, 0b001, "smsubl", sub, sext>;
1160def UMADDLrrr : WideMulAccum<0, 0b101, "umaddl", add, zext>;
1161def UMSUBLrrr : WideMulAccum<1, 0b101, "umsubl", sub, zext>;
1162
1163def : Pat<(i64 (mul (sext GPR32:$Rn), (sext GPR32:$Rm))),
1164          (SMADDLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
1165def : Pat<(i64 (mul (zext GPR32:$Rn), (zext GPR32:$Rm))),
1166          (UMADDLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
1167
1168def : Pat<(i64 (ineg (mul (sext GPR32:$Rn), (sext GPR32:$Rm)))),
1169          (SMSUBLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
1170def : Pat<(i64 (ineg (mul (zext GPR32:$Rn), (zext GPR32:$Rm)))),
1171          (UMSUBLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
1172
1173def : Pat<(i64 (mul (sext GPR32:$Rn), (s64imm_32bit:$C))),
1174          (SMADDLrrr GPR32:$Rn, (MOVi32imm (trunc_imm imm:$C)), XZR)>;
1175def : Pat<(i64 (mul (zext GPR32:$Rn), (i64imm_32bit:$C))),
1176          (UMADDLrrr GPR32:$Rn, (MOVi32imm (trunc_imm imm:$C)), XZR)>;
1177def : Pat<(i64 (mul (sext_inreg GPR64:$Rn, i32), (s64imm_32bit:$C))),
1178          (SMADDLrrr (i32 (EXTRACT_SUBREG GPR64:$Rn, sub_32)),
1179                     (MOVi32imm (trunc_imm imm:$C)), XZR)>;
1180
1181def : Pat<(i64 (ineg (mul (sext GPR32:$Rn), (s64imm_32bit:$C)))),
1182          (SMSUBLrrr GPR32:$Rn, (MOVi32imm (trunc_imm imm:$C)), XZR)>;
1183def : Pat<(i64 (ineg (mul (zext GPR32:$Rn), (i64imm_32bit:$C)))),
1184          (UMSUBLrrr GPR32:$Rn, (MOVi32imm (trunc_imm imm:$C)), XZR)>;
1185def : Pat<(i64 (ineg (mul (sext_inreg GPR64:$Rn, i32), (s64imm_32bit:$C)))),
1186          (SMSUBLrrr (i32 (EXTRACT_SUBREG GPR64:$Rn, sub_32)),
1187                     (MOVi32imm (trunc_imm imm:$C)), XZR)>;
1188
1189def : Pat<(i64 (add (mul (sext GPR32:$Rn), (s64imm_32bit:$C)), GPR64:$Ra)),
1190          (SMADDLrrr GPR32:$Rn, (MOVi32imm (trunc_imm imm:$C)), GPR64:$Ra)>;
1191def : Pat<(i64 (add (mul (zext GPR32:$Rn), (i64imm_32bit:$C)), GPR64:$Ra)),
1192          (UMADDLrrr GPR32:$Rn, (MOVi32imm (trunc_imm imm:$C)), GPR64:$Ra)>;
1193def : Pat<(i64 (add (mul (sext_inreg GPR64:$Rn, i32), (s64imm_32bit:$C)),
1194                    GPR64:$Ra)),
1195          (SMADDLrrr (i32 (EXTRACT_SUBREG GPR64:$Rn, sub_32)),
1196                     (MOVi32imm (trunc_imm imm:$C)), GPR64:$Ra)>;
1197
1198def : Pat<(i64 (sub GPR64:$Ra, (mul (sext GPR32:$Rn), (s64imm_32bit:$C)))),
1199          (SMSUBLrrr GPR32:$Rn, (MOVi32imm (trunc_imm imm:$C)), GPR64:$Ra)>;
1200def : Pat<(i64 (sub GPR64:$Ra, (mul (zext GPR32:$Rn), (i64imm_32bit:$C)))),
1201          (UMSUBLrrr GPR32:$Rn, (MOVi32imm (trunc_imm imm:$C)), GPR64:$Ra)>;
1202def : Pat<(i64 (sub GPR64:$Ra, (mul (sext_inreg GPR64:$Rn, i32),
1203                                    (s64imm_32bit:$C)))),
1204          (SMSUBLrrr (i32 (EXTRACT_SUBREG GPR64:$Rn, sub_32)),
1205                     (MOVi32imm (trunc_imm imm:$C)), GPR64:$Ra)>;
1206} // AddedComplexity = 5
1207
1208def : MulAccumWAlias<"mul", MADDWrrr>;
1209def : MulAccumXAlias<"mul", MADDXrrr>;
1210def : MulAccumWAlias<"mneg", MSUBWrrr>;
1211def : MulAccumXAlias<"mneg", MSUBXrrr>;
1212def : WideMulAccumAlias<"smull", SMADDLrrr>;
1213def : WideMulAccumAlias<"smnegl", SMSUBLrrr>;
1214def : WideMulAccumAlias<"umull", UMADDLrrr>;
1215def : WideMulAccumAlias<"umnegl", UMSUBLrrr>;
1216
1217// Multiply-high
1218def SMULHrr : MulHi<0b010, "smulh", mulhs>;
1219def UMULHrr : MulHi<0b110, "umulh", mulhu>;
1220
1221// CRC32
1222def CRC32Brr : BaseCRC32<0, 0b00, 0, GPR32, int_aarch64_crc32b, "crc32b">;
1223def CRC32Hrr : BaseCRC32<0, 0b01, 0, GPR32, int_aarch64_crc32h, "crc32h">;
1224def CRC32Wrr : BaseCRC32<0, 0b10, 0, GPR32, int_aarch64_crc32w, "crc32w">;
1225def CRC32Xrr : BaseCRC32<1, 0b11, 0, GPR64, int_aarch64_crc32x, "crc32x">;
1226
1227def CRC32CBrr : BaseCRC32<0, 0b00, 1, GPR32, int_aarch64_crc32cb, "crc32cb">;
1228def CRC32CHrr : BaseCRC32<0, 0b01, 1, GPR32, int_aarch64_crc32ch, "crc32ch">;
1229def CRC32CWrr : BaseCRC32<0, 0b10, 1, GPR32, int_aarch64_crc32cw, "crc32cw">;
1230def CRC32CXrr : BaseCRC32<1, 0b11, 1, GPR64, int_aarch64_crc32cx, "crc32cx">;
1231
1232// v8.1 atomic CAS
1233defm CAS   : CompareAndSwap<0, 0, "">;
1234defm CASA  : CompareAndSwap<1, 0, "a">;
1235defm CASL  : CompareAndSwap<0, 1, "l">;
1236defm CASAL : CompareAndSwap<1, 1, "al">;
1237
1238// v8.1 atomic CASP
1239defm CASP   : CompareAndSwapPair<0, 0, "">;
1240defm CASPA  : CompareAndSwapPair<1, 0, "a">;
1241defm CASPL  : CompareAndSwapPair<0, 1, "l">;
1242defm CASPAL : CompareAndSwapPair<1, 1, "al">;
1243
1244// v8.1 atomic SWP
1245defm SWP   : Swap<0, 0, "">;
1246defm SWPA  : Swap<1, 0, "a">;
1247defm SWPL  : Swap<0, 1, "l">;
1248defm SWPAL : Swap<1, 1, "al">;
1249
1250// v8.1 atomic LD<OP>(register). Performs load and then ST<OP>(register)
1251defm LDADD   : LDOPregister<0b000, "add", 0, 0, "">;
1252defm LDADDA  : LDOPregister<0b000, "add", 1, 0, "a">;
1253defm LDADDL  : LDOPregister<0b000, "add", 0, 1, "l">;
1254defm LDADDAL : LDOPregister<0b000, "add", 1, 1, "al">;
1255
1256defm LDCLR   : LDOPregister<0b001, "clr", 0, 0, "">;
1257defm LDCLRA  : LDOPregister<0b001, "clr", 1, 0, "a">;
1258defm LDCLRL  : LDOPregister<0b001, "clr", 0, 1, "l">;
1259defm LDCLRAL : LDOPregister<0b001, "clr", 1, 1, "al">;
1260
1261defm LDEOR   : LDOPregister<0b010, "eor", 0, 0, "">;
1262defm LDEORA  : LDOPregister<0b010, "eor", 1, 0, "a">;
1263defm LDEORL  : LDOPregister<0b010, "eor", 0, 1, "l">;
1264defm LDEORAL : LDOPregister<0b010, "eor", 1, 1, "al">;
1265
1266defm LDSET   : LDOPregister<0b011, "set", 0, 0, "">;
1267defm LDSETA  : LDOPregister<0b011, "set", 1, 0, "a">;
1268defm LDSETL  : LDOPregister<0b011, "set", 0, 1, "l">;
1269defm LDSETAL : LDOPregister<0b011, "set", 1, 1, "al">;
1270
1271defm LDSMAX   : LDOPregister<0b100, "smax", 0, 0, "">;
1272defm LDSMAXA  : LDOPregister<0b100, "smax", 1, 0, "a">;
1273defm LDSMAXL  : LDOPregister<0b100, "smax", 0, 1, "l">;
1274defm LDSMAXAL : LDOPregister<0b100, "smax", 1, 1, "al">;
1275
1276defm LDSMIN   : LDOPregister<0b101, "smin", 0, 0, "">;
1277defm LDSMINA  : LDOPregister<0b101, "smin", 1, 0, "a">;
1278defm LDSMINL  : LDOPregister<0b101, "smin", 0, 1, "l">;
1279defm LDSMINAL : LDOPregister<0b101, "smin", 1, 1, "al">;
1280
1281defm LDUMAX   : LDOPregister<0b110, "umax", 0, 0, "">;
1282defm LDUMAXA  : LDOPregister<0b110, "umax", 1, 0, "a">;
1283defm LDUMAXL  : LDOPregister<0b110, "umax", 0, 1, "l">;
1284defm LDUMAXAL : LDOPregister<0b110, "umax", 1, 1, "al">;
1285
1286defm LDUMIN   : LDOPregister<0b111, "umin", 0, 0, "">;
1287defm LDUMINA  : LDOPregister<0b111, "umin", 1, 0, "a">;
1288defm LDUMINL  : LDOPregister<0b111, "umin", 0, 1, "l">;
1289defm LDUMINAL : LDOPregister<0b111, "umin", 1, 1, "al">;
1290
1291// v8.1 atomic ST<OP>(register) as aliases to "LD<OP>(register) when Rt=xZR"
1292defm : STOPregister<"stadd","LDADD">; // STADDx
1293defm : STOPregister<"stclr","LDCLR">; // STCLRx
1294defm : STOPregister<"steor","LDEOR">; // STEORx
1295defm : STOPregister<"stset","LDSET">; // STSETx
1296defm : STOPregister<"stsmax","LDSMAX">;// STSMAXx
1297defm : STOPregister<"stsmin","LDSMIN">;// STSMINx
1298defm : STOPregister<"stumax","LDUMAX">;// STUMAXx
1299defm : STOPregister<"stumin","LDUMIN">;// STUMINx
1300
1301// v8.5 Memory Tagging Extension
1302let Predicates = [HasMTE] in {
1303
1304def IRG   : BaseTwoOperand<0b0100, GPR64sp, "irg", int_aarch64_irg, GPR64sp, GPR64>,
1305            Sched<[]>{
1306  let Inst{31} = 1;
1307}
1308def GMI   : BaseTwoOperand<0b0101, GPR64, "gmi", int_aarch64_gmi, GPR64sp>, Sched<[]>{
1309  let Inst{31} = 1;
1310  let isNotDuplicable = 1;
1311}
1312def ADDG  : AddSubG<0, "addg", null_frag>;
1313def SUBG  : AddSubG<1, "subg", null_frag>;
1314
1315def : InstAlias<"irg $dst, $src", (IRG GPR64sp:$dst, GPR64sp:$src, XZR), 1>;
1316
1317def SUBP : SUBP<0, "subp", int_aarch64_subp>, Sched<[]>;
1318def SUBPS : SUBP<1, "subps", null_frag>, Sched<[]>{
1319  let Defs = [NZCV];
1320}
1321
1322def : InstAlias<"cmpp $lhs, $rhs", (SUBPS XZR, GPR64sp:$lhs, GPR64sp:$rhs), 0>;
1323
1324def LDG : MemTagLoad<"ldg", "\t$Rt, [$Rn, $offset]">;
1325
1326def : Pat<(int_aarch64_addg (am_indexedu6s128 GPR64sp:$Rn, uimm6s16:$imm6), imm0_15:$imm4),
1327          (ADDG GPR64sp:$Rn, imm0_63:$imm6, imm0_15:$imm4)>;
1328def : Pat<(int_aarch64_ldg GPR64:$Rt, (am_indexeds9s128 GPR64sp:$Rn,  simm9s16:$offset)),
1329          (LDG GPR64:$Rt, GPR64sp:$Rn,  simm9s16:$offset)>;
1330
1331def : InstAlias<"ldg $Rt, [$Rn]", (LDG GPR64:$Rt, GPR64sp:$Rn, 0), 1>;
1332
1333def LDGM : MemTagVector<1, "ldgm", "\t$Rt, [$Rn]",
1334                   (outs GPR64:$Rt), (ins GPR64sp:$Rn)>;
1335def STGM : MemTagVector<0, "stgm", "\t$Rt, [$Rn]",
1336                   (outs), (ins GPR64:$Rt, GPR64sp:$Rn)>;
1337def STZGM : MemTagVector<0, "stzgm", "\t$Rt, [$Rn]",
1338                   (outs), (ins GPR64:$Rt, GPR64sp:$Rn)> {
1339  let Inst{23} = 0;
1340}
1341
1342defm STG   : MemTagStore<0b00, "stg">;
1343defm STZG  : MemTagStore<0b01, "stzg">;
1344defm ST2G  : MemTagStore<0b10, "st2g">;
1345defm STZ2G : MemTagStore<0b11, "stz2g">;
1346
1347def : Pat<(AArch64stg GPR64sp:$Rn, (am_indexeds9s128 GPR64sp:$Rm, simm9s16:$imm)),
1348          (STGOffset $Rn, $Rm, $imm)>;
1349def : Pat<(AArch64stzg GPR64sp:$Rn, (am_indexeds9s128 GPR64sp:$Rm, simm9s16:$imm)),
1350          (STZGOffset $Rn, $Rm, $imm)>;
1351def : Pat<(AArch64st2g GPR64sp:$Rn, (am_indexeds9s128 GPR64sp:$Rm, simm9s16:$imm)),
1352          (ST2GOffset $Rn, $Rm, $imm)>;
1353def : Pat<(AArch64stz2g GPR64sp:$Rn, (am_indexeds9s128 GPR64sp:$Rm, simm9s16:$imm)),
1354          (STZ2GOffset $Rn, $Rm, $imm)>;
1355
1356defm STGP     : StorePairOffset <0b01, 0, GPR64z, simm7s16, "stgp">;
1357def  STGPpre  : StorePairPreIdx <0b01, 0, GPR64z, simm7s16, "stgp">;
1358def  STGPpost : StorePairPostIdx<0b01, 0, GPR64z, simm7s16, "stgp">;
1359
1360def : Pat<(int_aarch64_stg GPR64:$Rt, (am_indexeds9s128 GPR64sp:$Rn, simm9s16:$offset)),
1361          (STGOffset GPR64:$Rt, GPR64sp:$Rn,  simm9s16:$offset)>;
1362
1363def : Pat<(int_aarch64_stgp (am_indexed7s128 GPR64sp:$Rn, simm7s16:$imm), GPR64:$Rt, GPR64:$Rt2),
1364          (STGPi $Rt, $Rt2, $Rn, $imm)>;
1365
1366def IRGstack
1367    : Pseudo<(outs GPR64sp:$Rd), (ins GPR64sp:$Rsp, GPR64:$Rm), []>,
1368      Sched<[]>;
1369def TAGPstack
1370    : Pseudo<(outs GPR64sp:$Rd), (ins GPR64sp:$Rn, uimm6s16:$imm6, GPR64sp:$Rm, imm0_15:$imm4), []>,
1371      Sched<[]>;
1372
1373// Explicit SP in the first operand prevents ShrinkWrap optimization
1374// from leaving this instruction out of the stack frame. When IRGstack
1375// is transformed into IRG, this operand is replaced with the actual
1376// register / expression for the tagged base pointer of the current function.
1377def : Pat<(int_aarch64_irg_sp i64:$Rm), (IRGstack SP, i64:$Rm)>;
1378
1379// Large STG to be expanded into a loop. $Rm is the size, $Rn is start address.
1380// $Rn_wback is one past the end of the range.
1381let isCodeGenOnly=1, mayStore=1 in {
1382def STGloop
1383    : Pseudo<(outs GPR64common:$Rm_wback, GPR64sp:$Rn_wback), (ins GPR64common:$Rm, GPR64sp:$Rn),
1384             [], "$Rn = $Rn_wback,@earlyclobber $Rn_wback,$Rm = $Rm_wback,@earlyclobber $Rm_wback" >,
1385      Sched<[WriteAdr, WriteST]>;
1386
1387def STZGloop
1388    : Pseudo<(outs GPR64common:$Rm_wback, GPR64sp:$Rn_wback), (ins GPR64common:$Rm, GPR64sp:$Rn),
1389             [], "$Rn = $Rn_wback,@earlyclobber $Rn_wback,$Rm = $Rm_wback,@earlyclobber $Rm_wback" >,
1390      Sched<[WriteAdr, WriteST]>;
1391}
1392
1393} // Predicates = [HasMTE]
1394
1395//===----------------------------------------------------------------------===//
1396// Logical instructions.
1397//===----------------------------------------------------------------------===//
1398
1399// (immediate)
1400defm ANDS : LogicalImmS<0b11, "ands", AArch64and_flag, "bics">;
1401defm AND  : LogicalImm<0b00, "and", and, "bic">;
1402defm EOR  : LogicalImm<0b10, "eor", xor, "eon">;
1403defm ORR  : LogicalImm<0b01, "orr", or, "orn">;
1404
1405// FIXME: these aliases *are* canonical sometimes (when movz can't be
1406// used). Actually, it seems to be working right now, but putting logical_immXX
1407// here is a bit dodgy on the AsmParser side too.
1408def : InstAlias<"mov $dst, $imm", (ORRWri GPR32sp:$dst, WZR,
1409                                          logical_imm32:$imm), 0>;
1410def : InstAlias<"mov $dst, $imm", (ORRXri GPR64sp:$dst, XZR,
1411                                          logical_imm64:$imm), 0>;
1412
1413
1414// (register)
1415defm ANDS : LogicalRegS<0b11, 0, "ands", AArch64and_flag>;
1416defm BICS : LogicalRegS<0b11, 1, "bics",
1417                        BinOpFrag<(AArch64and_flag node:$LHS, (not node:$RHS))>>;
1418defm AND  : LogicalReg<0b00, 0, "and", and>;
1419defm BIC  : LogicalReg<0b00, 1, "bic",
1420                       BinOpFrag<(and node:$LHS, (not node:$RHS))>>;
1421defm EON  : LogicalReg<0b10, 1, "eon",
1422                       BinOpFrag<(not (xor node:$LHS, node:$RHS))>>;
1423defm EOR  : LogicalReg<0b10, 0, "eor", xor>;
1424defm ORN  : LogicalReg<0b01, 1, "orn",
1425                       BinOpFrag<(or node:$LHS, (not node:$RHS))>>;
1426defm ORR  : LogicalReg<0b01, 0, "orr", or>;
1427
1428def : InstAlias<"mov $dst, $src", (ORRWrs GPR32:$dst, WZR, GPR32:$src, 0), 2>;
1429def : InstAlias<"mov $dst, $src", (ORRXrs GPR64:$dst, XZR, GPR64:$src, 0), 2>;
1430
1431def : InstAlias<"mvn $Wd, $Wm", (ORNWrs GPR32:$Wd, WZR, GPR32:$Wm, 0), 3>;
1432def : InstAlias<"mvn $Xd, $Xm", (ORNXrs GPR64:$Xd, XZR, GPR64:$Xm, 0), 3>;
1433
1434def : InstAlias<"mvn $Wd, $Wm$sh",
1435                (ORNWrs GPR32:$Wd, WZR, GPR32:$Wm, logical_shift32:$sh), 2>;
1436def : InstAlias<"mvn $Xd, $Xm$sh",
1437                (ORNXrs GPR64:$Xd, XZR, GPR64:$Xm, logical_shift64:$sh), 2>;
1438
1439def : InstAlias<"tst $src1, $src2",
1440                (ANDSWri WZR, GPR32:$src1, logical_imm32:$src2), 2>;
1441def : InstAlias<"tst $src1, $src2",
1442                (ANDSXri XZR, GPR64:$src1, logical_imm64:$src2), 2>;
1443
1444def : InstAlias<"tst $src1, $src2",
1445                        (ANDSWrs WZR, GPR32:$src1, GPR32:$src2, 0), 3>;
1446def : InstAlias<"tst $src1, $src2",
1447                        (ANDSXrs XZR, GPR64:$src1, GPR64:$src2, 0), 3>;
1448
1449def : InstAlias<"tst $src1, $src2$sh",
1450               (ANDSWrs WZR, GPR32:$src1, GPR32:$src2, logical_shift32:$sh), 2>;
1451def : InstAlias<"tst $src1, $src2$sh",
1452               (ANDSXrs XZR, GPR64:$src1, GPR64:$src2, logical_shift64:$sh), 2>;
1453
1454
1455def : Pat<(not GPR32:$Wm), (ORNWrr WZR, GPR32:$Wm)>;
1456def : Pat<(not GPR64:$Xm), (ORNXrr XZR, GPR64:$Xm)>;
1457
1458
1459//===----------------------------------------------------------------------===//
1460// One operand data processing instructions.
1461//===----------------------------------------------------------------------===//
1462
1463defm CLS    : OneOperandData<0b101, "cls">;
1464defm CLZ    : OneOperandData<0b100, "clz", ctlz>;
1465defm RBIT   : OneOperandData<0b000, "rbit", bitreverse>;
1466
1467def  REV16Wr : OneWRegData<0b001, "rev16",
1468                                  UnOpFrag<(rotr (bswap node:$LHS), (i64 16))>>;
1469def  REV16Xr : OneXRegData<0b001, "rev16", null_frag>;
1470
1471def : Pat<(cttz GPR32:$Rn),
1472          (CLZWr (RBITWr GPR32:$Rn))>;
1473def : Pat<(cttz GPR64:$Rn),
1474          (CLZXr (RBITXr GPR64:$Rn))>;
1475def : Pat<(ctlz (or (shl (xor (sra GPR32:$Rn, (i64 31)), GPR32:$Rn), (i64 1)),
1476                (i32 1))),
1477          (CLSWr GPR32:$Rn)>;
1478def : Pat<(ctlz (or (shl (xor (sra GPR64:$Rn, (i64 63)), GPR64:$Rn), (i64 1)),
1479                (i64 1))),
1480          (CLSXr GPR64:$Rn)>;
1481
1482// Unlike the other one operand instructions, the instructions with the "rev"
1483// mnemonic do *not* just different in the size bit, but actually use different
1484// opcode bits for the different sizes.
1485def REVWr   : OneWRegData<0b010, "rev", bswap>;
1486def REVXr   : OneXRegData<0b011, "rev", bswap>;
1487def REV32Xr : OneXRegData<0b010, "rev32",
1488                                 UnOpFrag<(rotr (bswap node:$LHS), (i64 32))>>;
1489
1490def : InstAlias<"rev64 $Rd, $Rn", (REVXr GPR64:$Rd, GPR64:$Rn), 0>;
1491
1492// The bswap commutes with the rotr so we want a pattern for both possible
1493// orders.
1494def : Pat<(bswap (rotr GPR32:$Rn, (i64 16))), (REV16Wr GPR32:$Rn)>;
1495def : Pat<(bswap (rotr GPR64:$Rn, (i64 32))), (REV32Xr GPR64:$Rn)>;
1496
1497//===----------------------------------------------------------------------===//
1498// Bitfield immediate extraction instruction.
1499//===----------------------------------------------------------------------===//
1500let hasSideEffects = 0 in
1501defm EXTR : ExtractImm<"extr">;
1502def : InstAlias<"ror $dst, $src, $shift",
1503            (EXTRWrri GPR32:$dst, GPR32:$src, GPR32:$src, imm0_31:$shift)>;
1504def : InstAlias<"ror $dst, $src, $shift",
1505            (EXTRXrri GPR64:$dst, GPR64:$src, GPR64:$src, imm0_63:$shift)>;
1506
1507def : Pat<(rotr GPR32:$Rn, (i64 imm0_31:$imm)),
1508          (EXTRWrri GPR32:$Rn, GPR32:$Rn, imm0_31:$imm)>;
1509def : Pat<(rotr GPR64:$Rn, (i64 imm0_63:$imm)),
1510          (EXTRXrri GPR64:$Rn, GPR64:$Rn, imm0_63:$imm)>;
1511
1512//===----------------------------------------------------------------------===//
1513// Other bitfield immediate instructions.
1514//===----------------------------------------------------------------------===//
1515let hasSideEffects = 0 in {
1516defm BFM  : BitfieldImmWith2RegArgs<0b01, "bfm">;
1517defm SBFM : BitfieldImm<0b00, "sbfm">;
1518defm UBFM : BitfieldImm<0b10, "ubfm">;
1519}
1520
1521def i32shift_a : Operand<i64>, SDNodeXForm<imm, [{
1522  uint64_t enc = (32 - N->getZExtValue()) & 0x1f;
1523  return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
1524}]>;
1525
1526def i32shift_b : Operand<i64>, SDNodeXForm<imm, [{
1527  uint64_t enc = 31 - N->getZExtValue();
1528  return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
1529}]>;
1530
1531// min(7, 31 - shift_amt)
1532def i32shift_sext_i8 : Operand<i64>, SDNodeXForm<imm, [{
1533  uint64_t enc = 31 - N->getZExtValue();
1534  enc = enc > 7 ? 7 : enc;
1535  return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
1536}]>;
1537
1538// min(15, 31 - shift_amt)
1539def i32shift_sext_i16 : Operand<i64>, SDNodeXForm<imm, [{
1540  uint64_t enc = 31 - N->getZExtValue();
1541  enc = enc > 15 ? 15 : enc;
1542  return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
1543}]>;
1544
1545def i64shift_a : Operand<i64>, SDNodeXForm<imm, [{
1546  uint64_t enc = (64 - N->getZExtValue()) & 0x3f;
1547  return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
1548}]>;
1549
1550def i64shift_b : Operand<i64>, SDNodeXForm<imm, [{
1551  uint64_t enc = 63 - N->getZExtValue();
1552  return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
1553}]>;
1554
1555// min(7, 63 - shift_amt)
1556def i64shift_sext_i8 : Operand<i64>, SDNodeXForm<imm, [{
1557  uint64_t enc = 63 - N->getZExtValue();
1558  enc = enc > 7 ? 7 : enc;
1559  return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
1560}]>;
1561
1562// min(15, 63 - shift_amt)
1563def i64shift_sext_i16 : Operand<i64>, SDNodeXForm<imm, [{
1564  uint64_t enc = 63 - N->getZExtValue();
1565  enc = enc > 15 ? 15 : enc;
1566  return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
1567}]>;
1568
1569// min(31, 63 - shift_amt)
1570def i64shift_sext_i32 : Operand<i64>, SDNodeXForm<imm, [{
1571  uint64_t enc = 63 - N->getZExtValue();
1572  enc = enc > 31 ? 31 : enc;
1573  return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
1574}]>;
1575
1576def : Pat<(shl GPR32:$Rn, (i64 imm0_31:$imm)),
1577          (UBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
1578                              (i64 (i32shift_b imm0_31:$imm)))>;
1579def : Pat<(shl GPR64:$Rn, (i64 imm0_63:$imm)),
1580          (UBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
1581                              (i64 (i64shift_b imm0_63:$imm)))>;
1582
1583let AddedComplexity = 10 in {
1584def : Pat<(sra GPR32:$Rn, (i64 imm0_31:$imm)),
1585          (SBFMWri GPR32:$Rn, imm0_31:$imm, 31)>;
1586def : Pat<(sra GPR64:$Rn, (i64 imm0_63:$imm)),
1587          (SBFMXri GPR64:$Rn, imm0_63:$imm, 63)>;
1588}
1589
1590def : InstAlias<"asr $dst, $src, $shift",
1591                (SBFMWri GPR32:$dst, GPR32:$src, imm0_31:$shift, 31)>;
1592def : InstAlias<"asr $dst, $src, $shift",
1593                (SBFMXri GPR64:$dst, GPR64:$src, imm0_63:$shift, 63)>;
1594def : InstAlias<"sxtb $dst, $src", (SBFMWri GPR32:$dst, GPR32:$src, 0, 7)>;
1595def : InstAlias<"sxtb $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 7)>;
1596def : InstAlias<"sxth $dst, $src", (SBFMWri GPR32:$dst, GPR32:$src, 0, 15)>;
1597def : InstAlias<"sxth $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 15)>;
1598def : InstAlias<"sxtw $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 31)>;
1599
1600def : Pat<(srl GPR32:$Rn, (i64 imm0_31:$imm)),
1601          (UBFMWri GPR32:$Rn, imm0_31:$imm, 31)>;
1602def : Pat<(srl GPR64:$Rn, (i64 imm0_63:$imm)),
1603          (UBFMXri GPR64:$Rn, imm0_63:$imm, 63)>;
1604
1605def : InstAlias<"lsr $dst, $src, $shift",
1606                (UBFMWri GPR32:$dst, GPR32:$src, imm0_31:$shift, 31)>;
1607def : InstAlias<"lsr $dst, $src, $shift",
1608                (UBFMXri GPR64:$dst, GPR64:$src, imm0_63:$shift, 63)>;
1609def : InstAlias<"uxtb $dst, $src", (UBFMWri GPR32:$dst, GPR32:$src, 0, 7)>;
1610def : InstAlias<"uxtb $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 7)>;
1611def : InstAlias<"uxth $dst, $src", (UBFMWri GPR32:$dst, GPR32:$src, 0, 15)>;
1612def : InstAlias<"uxth $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 15)>;
1613def : InstAlias<"uxtw $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 31)>;
1614
1615//===----------------------------------------------------------------------===//
1616// Conditional comparison instructions.
1617//===----------------------------------------------------------------------===//
1618defm CCMN : CondComparison<0, "ccmn", AArch64ccmn>;
1619defm CCMP : CondComparison<1, "ccmp", AArch64ccmp>;
1620
1621//===----------------------------------------------------------------------===//
1622// Conditional select instructions.
1623//===----------------------------------------------------------------------===//
1624defm CSEL  : CondSelect<0, 0b00, "csel">;
1625
1626def inc : PatFrag<(ops node:$in), (add node:$in, 1)>;
1627defm CSINC : CondSelectOp<0, 0b01, "csinc", inc>;
1628defm CSINV : CondSelectOp<1, 0b00, "csinv", not>;
1629defm CSNEG : CondSelectOp<1, 0b01, "csneg", ineg>;
1630
1631def : Pat<(AArch64csinv GPR32:$tval, GPR32:$fval, (i32 imm:$cc), NZCV),
1632          (CSINVWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
1633def : Pat<(AArch64csinv GPR64:$tval, GPR64:$fval, (i32 imm:$cc), NZCV),
1634          (CSINVXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
1635def : Pat<(AArch64csneg GPR32:$tval, GPR32:$fval, (i32 imm:$cc), NZCV),
1636          (CSNEGWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
1637def : Pat<(AArch64csneg GPR64:$tval, GPR64:$fval, (i32 imm:$cc), NZCV),
1638          (CSNEGXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
1639def : Pat<(AArch64csinc GPR32:$tval, GPR32:$fval, (i32 imm:$cc), NZCV),
1640          (CSINCWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
1641def : Pat<(AArch64csinc GPR64:$tval, GPR64:$fval, (i32 imm:$cc), NZCV),
1642          (CSINCXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
1643
1644def : Pat<(AArch64csel (i32 0), (i32 1), (i32 imm:$cc), NZCV),
1645          (CSINCWr WZR, WZR, (i32 imm:$cc))>;
1646def : Pat<(AArch64csel (i64 0), (i64 1), (i32 imm:$cc), NZCV),
1647          (CSINCXr XZR, XZR, (i32 imm:$cc))>;
1648def : Pat<(AArch64csel GPR32:$tval, (i32 1), (i32 imm:$cc), NZCV),
1649          (CSINCWr GPR32:$tval, WZR, (i32 imm:$cc))>;
1650def : Pat<(AArch64csel GPR64:$tval, (i64 1), (i32 imm:$cc), NZCV),
1651          (CSINCXr GPR64:$tval, XZR, (i32 imm:$cc))>;
1652def : Pat<(AArch64csel (i32 1), GPR32:$fval, (i32 imm:$cc), NZCV),
1653          (CSINCWr GPR32:$fval, WZR, (i32 (inv_cond_XFORM imm:$cc)))>;
1654def : Pat<(AArch64csel (i64 1), GPR64:$fval, (i32 imm:$cc), NZCV),
1655          (CSINCXr GPR64:$fval, XZR, (i32 (inv_cond_XFORM imm:$cc)))>;
1656def : Pat<(AArch64csel (i32 0), (i32 -1), (i32 imm:$cc), NZCV),
1657          (CSINVWr WZR, WZR, (i32 imm:$cc))>;
1658def : Pat<(AArch64csel (i64 0), (i64 -1), (i32 imm:$cc), NZCV),
1659          (CSINVXr XZR, XZR, (i32 imm:$cc))>;
1660def : Pat<(AArch64csel GPR32:$tval, (i32 -1), (i32 imm:$cc), NZCV),
1661          (CSINVWr GPR32:$tval, WZR, (i32 imm:$cc))>;
1662def : Pat<(AArch64csel GPR64:$tval, (i64 -1), (i32 imm:$cc), NZCV),
1663          (CSINVXr GPR64:$tval, XZR, (i32 imm:$cc))>;
1664def : Pat<(AArch64csel (i32 -1), GPR32:$fval, (i32 imm:$cc), NZCV),
1665          (CSINVWr GPR32:$fval, WZR, (i32 (inv_cond_XFORM imm:$cc)))>;
1666def : Pat<(AArch64csel (i64 -1), GPR64:$fval, (i32 imm:$cc), NZCV),
1667          (CSINVXr GPR64:$fval, XZR, (i32 (inv_cond_XFORM imm:$cc)))>;
1668
1669// The inverse of the condition code from the alias instruction is what is used
1670// in the aliased instruction. The parser all ready inverts the condition code
1671// for these aliases.
1672def : InstAlias<"cset $dst, $cc",
1673                (CSINCWr GPR32:$dst, WZR, WZR, inv_ccode:$cc)>;
1674def : InstAlias<"cset $dst, $cc",
1675                (CSINCXr GPR64:$dst, XZR, XZR, inv_ccode:$cc)>;
1676
1677def : InstAlias<"csetm $dst, $cc",
1678                (CSINVWr GPR32:$dst, WZR, WZR, inv_ccode:$cc)>;
1679def : InstAlias<"csetm $dst, $cc",
1680                (CSINVXr GPR64:$dst, XZR, XZR, inv_ccode:$cc)>;
1681
1682def : InstAlias<"cinc $dst, $src, $cc",
1683                (CSINCWr GPR32:$dst, GPR32:$src, GPR32:$src, inv_ccode:$cc)>;
1684def : InstAlias<"cinc $dst, $src, $cc",
1685                (CSINCXr GPR64:$dst, GPR64:$src, GPR64:$src, inv_ccode:$cc)>;
1686
1687def : InstAlias<"cinv $dst, $src, $cc",
1688                (CSINVWr GPR32:$dst, GPR32:$src, GPR32:$src, inv_ccode:$cc)>;
1689def : InstAlias<"cinv $dst, $src, $cc",
1690                (CSINVXr GPR64:$dst, GPR64:$src, GPR64:$src, inv_ccode:$cc)>;
1691
1692def : InstAlias<"cneg $dst, $src, $cc",
1693                (CSNEGWr GPR32:$dst, GPR32:$src, GPR32:$src, inv_ccode:$cc)>;
1694def : InstAlias<"cneg $dst, $src, $cc",
1695                (CSNEGXr GPR64:$dst, GPR64:$src, GPR64:$src, inv_ccode:$cc)>;
1696
1697//===----------------------------------------------------------------------===//
1698// PC-relative instructions.
1699//===----------------------------------------------------------------------===//
1700let isReMaterializable = 1 in {
1701let hasSideEffects = 0, mayStore = 0, mayLoad = 0 in {
1702def ADR  : ADRI<0, "adr", adrlabel,
1703                [(set GPR64:$Xd, (AArch64adr tglobaladdr:$label))]>;
1704} // hasSideEffects = 0
1705
1706def ADRP : ADRI<1, "adrp", adrplabel,
1707                [(set GPR64:$Xd, (AArch64adrp tglobaladdr:$label))]>;
1708} // isReMaterializable = 1
1709
1710// page address of a constant pool entry, block address
1711def : Pat<(AArch64adr tconstpool:$cp), (ADR tconstpool:$cp)>;
1712def : Pat<(AArch64adr tblockaddress:$cp), (ADR tblockaddress:$cp)>;
1713def : Pat<(AArch64adr texternalsym:$sym), (ADR texternalsym:$sym)>;
1714def : Pat<(AArch64adr tjumptable:$sym), (ADR tjumptable:$sym)>;
1715def : Pat<(AArch64adrp tconstpool:$cp), (ADRP tconstpool:$cp)>;
1716def : Pat<(AArch64adrp tblockaddress:$cp), (ADRP tblockaddress:$cp)>;
1717def : Pat<(AArch64adrp texternalsym:$sym), (ADRP texternalsym:$sym)>;
1718
1719//===----------------------------------------------------------------------===//
1720// Unconditional branch (register) instructions.
1721//===----------------------------------------------------------------------===//
1722
1723let isReturn = 1, isTerminator = 1, isBarrier = 1 in {
1724def RET  : BranchReg<0b0010, "ret", []>;
1725def DRPS : SpecialReturn<0b0101, "drps">;
1726def ERET : SpecialReturn<0b0100, "eret">;
1727} // isReturn = 1, isTerminator = 1, isBarrier = 1
1728
1729// Default to the LR register.
1730def : InstAlias<"ret", (RET LR)>;
1731
1732let isCall = 1, Defs = [LR], Uses = [SP] in {
1733def BLR : BranchReg<0b0001, "blr", [(AArch64call GPR64:$Rn)]>;
1734} // isCall
1735
1736let isBranch = 1, isTerminator = 1, isBarrier = 1, isIndirectBranch = 1 in {
1737def BR  : BranchReg<0b0000, "br", [(brind GPR64:$Rn)]>;
1738} // isBranch, isTerminator, isBarrier, isIndirectBranch
1739
1740// Create a separate pseudo-instruction for codegen to use so that we don't
1741// flag lr as used in every function. It'll be restored before the RET by the
1742// epilogue if it's legitimately used.
1743def RET_ReallyLR : Pseudo<(outs), (ins), [(AArch64retflag)]>,
1744                   Sched<[WriteBrReg]> {
1745  let isTerminator = 1;
1746  let isBarrier = 1;
1747  let isReturn = 1;
1748}
1749
1750// This is a directive-like pseudo-instruction. The purpose is to insert an
1751// R_AARCH64_TLSDESC_CALL relocation at the offset of the following instruction
1752// (which in the usual case is a BLR).
1753let hasSideEffects = 1 in
1754def TLSDESCCALL : Pseudo<(outs), (ins i64imm:$sym), []>, Sched<[]> {
1755  let AsmString = ".tlsdesccall $sym";
1756}
1757
1758// Pseudo instruction to tell the streamer to emit a 'B' character into the
1759// augmentation string.
1760def EMITBKEY : Pseudo<(outs), (ins), []>, Sched<[]> {}
1761
1762// FIXME: maybe the scratch register used shouldn't be fixed to X1?
1763// FIXME: can "hasSideEffects be dropped?
1764let isCall = 1, Defs = [LR, X0, X1], hasSideEffects = 1,
1765    isCodeGenOnly = 1 in
1766def TLSDESC_CALLSEQ
1767    : Pseudo<(outs), (ins i64imm:$sym),
1768             [(AArch64tlsdesc_callseq tglobaltlsaddr:$sym)]>,
1769      Sched<[WriteI, WriteLD, WriteI, WriteBrReg]>;
1770def : Pat<(AArch64tlsdesc_callseq texternalsym:$sym),
1771          (TLSDESC_CALLSEQ texternalsym:$sym)>;
1772
1773//===----------------------------------------------------------------------===//
1774// Conditional branch (immediate) instruction.
1775//===----------------------------------------------------------------------===//
1776def Bcc : BranchCond;
1777
1778//===----------------------------------------------------------------------===//
1779// Compare-and-branch instructions.
1780//===----------------------------------------------------------------------===//
1781defm CBZ  : CmpBranch<0, "cbz", AArch64cbz>;
1782defm CBNZ : CmpBranch<1, "cbnz", AArch64cbnz>;
1783
1784//===----------------------------------------------------------------------===//
1785// Test-bit-and-branch instructions.
1786//===----------------------------------------------------------------------===//
1787defm TBZ  : TestBranch<0, "tbz", AArch64tbz>;
1788defm TBNZ : TestBranch<1, "tbnz", AArch64tbnz>;
1789
1790//===----------------------------------------------------------------------===//
1791// Unconditional branch (immediate) instructions.
1792//===----------------------------------------------------------------------===//
1793let isBranch = 1, isTerminator = 1, isBarrier = 1 in {
1794def B  : BranchImm<0, "b", [(br bb:$addr)]>;
1795} // isBranch, isTerminator, isBarrier
1796
1797let isCall = 1, Defs = [LR], Uses = [SP] in {
1798def BL : CallImm<1, "bl", [(AArch64call tglobaladdr:$addr)]>;
1799} // isCall
1800def : Pat<(AArch64call texternalsym:$func), (BL texternalsym:$func)>;
1801
1802//===----------------------------------------------------------------------===//
1803// Exception generation instructions.
1804//===----------------------------------------------------------------------===//
1805let isTrap = 1 in {
1806def BRK   : ExceptionGeneration<0b001, 0b00, "brk">;
1807}
1808def DCPS1 : ExceptionGeneration<0b101, 0b01, "dcps1">;
1809def DCPS2 : ExceptionGeneration<0b101, 0b10, "dcps2">;
1810def DCPS3 : ExceptionGeneration<0b101, 0b11, "dcps3">;
1811def HLT   : ExceptionGeneration<0b010, 0b00, "hlt">;
1812def HVC   : ExceptionGeneration<0b000, 0b10, "hvc">;
1813def SMC   : ExceptionGeneration<0b000, 0b11, "smc">;
1814def SVC   : ExceptionGeneration<0b000, 0b01, "svc">;
1815
1816// DCPSn defaults to an immediate operand of zero if unspecified.
1817def : InstAlias<"dcps1", (DCPS1 0)>;
1818def : InstAlias<"dcps2", (DCPS2 0)>;
1819def : InstAlias<"dcps3", (DCPS3 0)>;
1820
1821def UDF : UDFType<0, "udf">;
1822
1823//===----------------------------------------------------------------------===//
1824// Load instructions.
1825//===----------------------------------------------------------------------===//
1826
1827// Pair (indexed, offset)
1828defm LDPW : LoadPairOffset<0b00, 0, GPR32z, simm7s4, "ldp">;
1829defm LDPX : LoadPairOffset<0b10, 0, GPR64z, simm7s8, "ldp">;
1830defm LDPS : LoadPairOffset<0b00, 1, FPR32Op, simm7s4, "ldp">;
1831defm LDPD : LoadPairOffset<0b01, 1, FPR64Op, simm7s8, "ldp">;
1832defm LDPQ : LoadPairOffset<0b10, 1, FPR128Op, simm7s16, "ldp">;
1833
1834defm LDPSW : LoadPairOffset<0b01, 0, GPR64z, simm7s4, "ldpsw">;
1835
1836// Pair (pre-indexed)
1837def LDPWpre : LoadPairPreIdx<0b00, 0, GPR32z, simm7s4, "ldp">;
1838def LDPXpre : LoadPairPreIdx<0b10, 0, GPR64z, simm7s8, "ldp">;
1839def LDPSpre : LoadPairPreIdx<0b00, 1, FPR32Op, simm7s4, "ldp">;
1840def LDPDpre : LoadPairPreIdx<0b01, 1, FPR64Op, simm7s8, "ldp">;
1841def LDPQpre : LoadPairPreIdx<0b10, 1, FPR128Op, simm7s16, "ldp">;
1842
1843def LDPSWpre : LoadPairPreIdx<0b01, 0, GPR64z, simm7s4, "ldpsw">;
1844
1845// Pair (post-indexed)
1846def LDPWpost : LoadPairPostIdx<0b00, 0, GPR32z, simm7s4, "ldp">;
1847def LDPXpost : LoadPairPostIdx<0b10, 0, GPR64z, simm7s8, "ldp">;
1848def LDPSpost : LoadPairPostIdx<0b00, 1, FPR32Op, simm7s4, "ldp">;
1849def LDPDpost : LoadPairPostIdx<0b01, 1, FPR64Op, simm7s8, "ldp">;
1850def LDPQpost : LoadPairPostIdx<0b10, 1, FPR128Op, simm7s16, "ldp">;
1851
1852def LDPSWpost : LoadPairPostIdx<0b01, 0, GPR64z, simm7s4, "ldpsw">;
1853
1854
1855// Pair (no allocate)
1856defm LDNPW : LoadPairNoAlloc<0b00, 0, GPR32z, simm7s4, "ldnp">;
1857defm LDNPX : LoadPairNoAlloc<0b10, 0, GPR64z, simm7s8, "ldnp">;
1858defm LDNPS : LoadPairNoAlloc<0b00, 1, FPR32Op, simm7s4, "ldnp">;
1859defm LDNPD : LoadPairNoAlloc<0b01, 1, FPR64Op, simm7s8, "ldnp">;
1860defm LDNPQ : LoadPairNoAlloc<0b10, 1, FPR128Op, simm7s16, "ldnp">;
1861
1862//---
1863// (register offset)
1864//---
1865
1866// Integer
1867defm LDRBB : Load8RO<0b00,  0, 0b01, GPR32, "ldrb", i32, zextloadi8>;
1868defm LDRHH : Load16RO<0b01, 0, 0b01, GPR32, "ldrh", i32, zextloadi16>;
1869defm LDRW  : Load32RO<0b10, 0, 0b01, GPR32, "ldr", i32, load>;
1870defm LDRX  : Load64RO<0b11, 0, 0b01, GPR64, "ldr", i64, load>;
1871
1872// Floating-point
1873defm LDRB : Load8RO<0b00,   1, 0b01, FPR8Op,   "ldr", untyped, load>;
1874defm LDRH : Load16RO<0b01,  1, 0b01, FPR16Op,  "ldr", f16, load>;
1875defm LDRS : Load32RO<0b10,  1, 0b01, FPR32Op,  "ldr", f32, load>;
1876defm LDRD : Load64RO<0b11,  1, 0b01, FPR64Op,  "ldr", f64, load>;
1877defm LDRQ : Load128RO<0b00, 1, 0b11, FPR128Op, "ldr", f128, load>;
1878
1879// Load sign-extended half-word
1880defm LDRSHW : Load16RO<0b01, 0, 0b11, GPR32, "ldrsh", i32, sextloadi16>;
1881defm LDRSHX : Load16RO<0b01, 0, 0b10, GPR64, "ldrsh", i64, sextloadi16>;
1882
1883// Load sign-extended byte
1884defm LDRSBW : Load8RO<0b00, 0, 0b11, GPR32, "ldrsb", i32, sextloadi8>;
1885defm LDRSBX : Load8RO<0b00, 0, 0b10, GPR64, "ldrsb", i64, sextloadi8>;
1886
1887// Load sign-extended word
1888defm LDRSW  : Load32RO<0b10, 0, 0b10, GPR64, "ldrsw", i64, sextloadi32>;
1889
1890// Pre-fetch.
1891defm PRFM : PrefetchRO<0b11, 0, 0b10, "prfm">;
1892
1893// For regular load, we do not have any alignment requirement.
1894// Thus, it is safe to directly map the vector loads with interesting
1895// addressing modes.
1896// FIXME: We could do the same for bitconvert to floating point vectors.
1897multiclass ScalToVecROLoadPat<ROAddrMode ro, SDPatternOperator loadop,
1898                              ValueType ScalTy, ValueType VecTy,
1899                              Instruction LOADW, Instruction LOADX,
1900                              SubRegIndex sub> {
1901  def : Pat<(VecTy (scalar_to_vector (ScalTy
1902              (loadop (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$offset))))),
1903            (INSERT_SUBREG (VecTy (IMPLICIT_DEF)),
1904                           (LOADW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$offset),
1905                           sub)>;
1906
1907  def : Pat<(VecTy (scalar_to_vector (ScalTy
1908              (loadop (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$offset))))),
1909            (INSERT_SUBREG (VecTy (IMPLICIT_DEF)),
1910                           (LOADX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$offset),
1911                           sub)>;
1912}
1913
1914let AddedComplexity = 10 in {
1915defm : ScalToVecROLoadPat<ro8,  extloadi8,  i32, v8i8,  LDRBroW, LDRBroX, bsub>;
1916defm : ScalToVecROLoadPat<ro8,  extloadi8,  i32, v16i8, LDRBroW, LDRBroX, bsub>;
1917
1918defm : ScalToVecROLoadPat<ro16, extloadi16, i32, v4i16, LDRHroW, LDRHroX, hsub>;
1919defm : ScalToVecROLoadPat<ro16, extloadi16, i32, v8i16, LDRHroW, LDRHroX, hsub>;
1920
1921defm : ScalToVecROLoadPat<ro16, load,       i32, v4f16, LDRHroW, LDRHroX, hsub>;
1922defm : ScalToVecROLoadPat<ro16, load,       i32, v8f16, LDRHroW, LDRHroX, hsub>;
1923
1924defm : ScalToVecROLoadPat<ro32, load,       i32, v2i32, LDRSroW, LDRSroX, ssub>;
1925defm : ScalToVecROLoadPat<ro32, load,       i32, v4i32, LDRSroW, LDRSroX, ssub>;
1926
1927defm : ScalToVecROLoadPat<ro32, load,       f32, v2f32, LDRSroW, LDRSroX, ssub>;
1928defm : ScalToVecROLoadPat<ro32, load,       f32, v4f32, LDRSroW, LDRSroX, ssub>;
1929
1930defm : ScalToVecROLoadPat<ro64, load,       i64, v2i64, LDRDroW, LDRDroX, dsub>;
1931
1932defm : ScalToVecROLoadPat<ro64, load,       f64, v2f64, LDRDroW, LDRDroX, dsub>;
1933
1934
1935def : Pat <(v1i64 (scalar_to_vector (i64
1936                      (load (ro_Windexed64 GPR64sp:$Rn, GPR32:$Rm,
1937                                           ro_Wextend64:$extend))))),
1938           (LDRDroW GPR64sp:$Rn, GPR32:$Rm, ro_Wextend64:$extend)>;
1939
1940def : Pat <(v1i64 (scalar_to_vector (i64
1941                      (load (ro_Xindexed64 GPR64sp:$Rn, GPR64:$Rm,
1942                                           ro_Xextend64:$extend))))),
1943           (LDRDroX GPR64sp:$Rn, GPR64:$Rm, ro_Xextend64:$extend)>;
1944}
1945
1946// Match all load 64 bits width whose type is compatible with FPR64
1947multiclass VecROLoadPat<ROAddrMode ro, ValueType VecTy,
1948                        Instruction LOADW, Instruction LOADX> {
1949
1950  def : Pat<(VecTy (load (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend))),
1951            (LOADW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
1952
1953  def : Pat<(VecTy (load (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend))),
1954            (LOADX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
1955}
1956
1957let AddedComplexity = 10 in {
1958let Predicates = [IsLE] in {
1959  // We must do vector loads with LD1 in big-endian.
1960  defm : VecROLoadPat<ro64, v2i32, LDRDroW, LDRDroX>;
1961  defm : VecROLoadPat<ro64, v2f32, LDRDroW, LDRDroX>;
1962  defm : VecROLoadPat<ro64, v8i8,  LDRDroW, LDRDroX>;
1963  defm : VecROLoadPat<ro64, v4i16, LDRDroW, LDRDroX>;
1964  defm : VecROLoadPat<ro64, v4f16, LDRDroW, LDRDroX>;
1965}
1966
1967defm : VecROLoadPat<ro64, v1i64,  LDRDroW, LDRDroX>;
1968defm : VecROLoadPat<ro64, v1f64,  LDRDroW, LDRDroX>;
1969
1970// Match all load 128 bits width whose type is compatible with FPR128
1971let Predicates = [IsLE] in {
1972  // We must do vector loads with LD1 in big-endian.
1973  defm : VecROLoadPat<ro128, v2i64,  LDRQroW, LDRQroX>;
1974  defm : VecROLoadPat<ro128, v2f64,  LDRQroW, LDRQroX>;
1975  defm : VecROLoadPat<ro128, v4i32,  LDRQroW, LDRQroX>;
1976  defm : VecROLoadPat<ro128, v4f32,  LDRQroW, LDRQroX>;
1977  defm : VecROLoadPat<ro128, v8i16,  LDRQroW, LDRQroX>;
1978  defm : VecROLoadPat<ro128, v8f16,  LDRQroW, LDRQroX>;
1979  defm : VecROLoadPat<ro128, v16i8,  LDRQroW, LDRQroX>;
1980}
1981} // AddedComplexity = 10
1982
1983// zextload -> i64
1984multiclass ExtLoadTo64ROPat<ROAddrMode ro, SDPatternOperator loadop,
1985                            Instruction INSTW, Instruction INSTX> {
1986  def : Pat<(i64 (loadop (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend))),
1987            (SUBREG_TO_REG (i64 0),
1988                           (INSTW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend),
1989                           sub_32)>;
1990
1991  def : Pat<(i64 (loadop (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend))),
1992            (SUBREG_TO_REG (i64 0),
1993                           (INSTX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend),
1994                           sub_32)>;
1995}
1996
1997let AddedComplexity = 10 in {
1998  defm : ExtLoadTo64ROPat<ro8,  zextloadi8,  LDRBBroW, LDRBBroX>;
1999  defm : ExtLoadTo64ROPat<ro16, zextloadi16, LDRHHroW, LDRHHroX>;
2000  defm : ExtLoadTo64ROPat<ro32, zextloadi32, LDRWroW,  LDRWroX>;
2001
2002  // zextloadi1 -> zextloadi8
2003  defm : ExtLoadTo64ROPat<ro8,  zextloadi1,  LDRBBroW, LDRBBroX>;
2004
2005  // extload -> zextload
2006  defm : ExtLoadTo64ROPat<ro8,  extloadi8,   LDRBBroW, LDRBBroX>;
2007  defm : ExtLoadTo64ROPat<ro16, extloadi16,  LDRHHroW, LDRHHroX>;
2008  defm : ExtLoadTo64ROPat<ro32, extloadi32,  LDRWroW,  LDRWroX>;
2009
2010  // extloadi1 -> zextloadi8
2011  defm : ExtLoadTo64ROPat<ro8,  extloadi1,   LDRBBroW, LDRBBroX>;
2012}
2013
2014
2015// zextload -> i64
2016multiclass ExtLoadTo32ROPat<ROAddrMode ro, SDPatternOperator loadop,
2017                            Instruction INSTW, Instruction INSTX> {
2018  def : Pat<(i32 (loadop (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend))),
2019            (INSTW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
2020
2021  def : Pat<(i32 (loadop (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend))),
2022            (INSTX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
2023
2024}
2025
2026let AddedComplexity = 10 in {
2027  // extload -> zextload
2028  defm : ExtLoadTo32ROPat<ro8,  extloadi8,   LDRBBroW, LDRBBroX>;
2029  defm : ExtLoadTo32ROPat<ro16, extloadi16,  LDRHHroW, LDRHHroX>;
2030  defm : ExtLoadTo32ROPat<ro32, extloadi32,  LDRWroW,  LDRWroX>;
2031
2032  // zextloadi1 -> zextloadi8
2033  defm : ExtLoadTo32ROPat<ro8, zextloadi1, LDRBBroW, LDRBBroX>;
2034}
2035
2036//---
2037// (unsigned immediate)
2038//---
2039defm LDRX : LoadUI<0b11, 0, 0b01, GPR64z, uimm12s8, "ldr",
2040                   [(set GPR64z:$Rt,
2041                         (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)))]>;
2042defm LDRW : LoadUI<0b10, 0, 0b01, GPR32z, uimm12s4, "ldr",
2043                   [(set GPR32z:$Rt,
2044                         (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset)))]>;
2045defm LDRB : LoadUI<0b00, 1, 0b01, FPR8Op, uimm12s1, "ldr",
2046                   [(set FPR8Op:$Rt,
2047                         (load (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset)))]>;
2048defm LDRH : LoadUI<0b01, 1, 0b01, FPR16Op, uimm12s2, "ldr",
2049                   [(set (f16 FPR16Op:$Rt),
2050                         (load (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset)))]>;
2051defm LDRS : LoadUI<0b10, 1, 0b01, FPR32Op, uimm12s4, "ldr",
2052                   [(set (f32 FPR32Op:$Rt),
2053                         (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset)))]>;
2054defm LDRD : LoadUI<0b11, 1, 0b01, FPR64Op, uimm12s8, "ldr",
2055                   [(set (f64 FPR64Op:$Rt),
2056                         (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)))]>;
2057defm LDRQ : LoadUI<0b00, 1, 0b11, FPR128Op, uimm12s16, "ldr",
2058                 [(set (f128 FPR128Op:$Rt),
2059                       (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)))]>;
2060
2061// For regular load, we do not have any alignment requirement.
2062// Thus, it is safe to directly map the vector loads with interesting
2063// addressing modes.
2064// FIXME: We could do the same for bitconvert to floating point vectors.
2065def : Pat <(v8i8 (scalar_to_vector (i32
2066               (extloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))))),
2067           (INSERT_SUBREG (v8i8 (IMPLICIT_DEF)),
2068                          (LDRBui GPR64sp:$Rn, uimm12s1:$offset), bsub)>;
2069def : Pat <(v16i8 (scalar_to_vector (i32
2070               (extloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))))),
2071           (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
2072                          (LDRBui GPR64sp:$Rn, uimm12s1:$offset), bsub)>;
2073def : Pat <(v4i16 (scalar_to_vector (i32
2074               (extloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))))),
2075           (INSERT_SUBREG (v4i16 (IMPLICIT_DEF)),
2076                          (LDRHui GPR64sp:$Rn, uimm12s2:$offset), hsub)>;
2077def : Pat <(v8i16 (scalar_to_vector (i32
2078               (extloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))))),
2079           (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
2080                          (LDRHui GPR64sp:$Rn, uimm12s2:$offset), hsub)>;
2081def : Pat <(v2i32 (scalar_to_vector (i32
2082               (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))))),
2083           (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)),
2084                          (LDRSui GPR64sp:$Rn, uimm12s4:$offset), ssub)>;
2085def : Pat <(v4i32 (scalar_to_vector (i32
2086               (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))))),
2087           (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
2088                          (LDRSui GPR64sp:$Rn, uimm12s4:$offset), ssub)>;
2089def : Pat <(v1i64 (scalar_to_vector (i64
2090               (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))))),
2091           (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
2092def : Pat <(v2i64 (scalar_to_vector (i64
2093               (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))))),
2094           (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
2095                          (LDRDui GPR64sp:$Rn, uimm12s8:$offset), dsub)>;
2096
2097// Match all load 64 bits width whose type is compatible with FPR64
2098let Predicates = [IsLE] in {
2099  // We must use LD1 to perform vector loads in big-endian.
2100  def : Pat<(v2f32 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
2101            (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
2102  def : Pat<(v8i8 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
2103            (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
2104  def : Pat<(v4i16 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
2105            (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
2106  def : Pat<(v2i32 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
2107            (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
2108  def : Pat<(v4f16 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
2109            (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
2110}
2111def : Pat<(v1f64 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
2112          (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
2113def : Pat<(v1i64 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
2114          (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
2115
2116// Match all load 128 bits width whose type is compatible with FPR128
2117let Predicates = [IsLE] in {
2118  // We must use LD1 to perform vector loads in big-endian.
2119  def : Pat<(v4f32 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
2120            (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
2121  def : Pat<(v2f64 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
2122            (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
2123  def : Pat<(v16i8 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
2124            (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
2125  def : Pat<(v8i16 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
2126            (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
2127  def : Pat<(v4i32 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
2128            (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
2129  def : Pat<(v2i64 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
2130            (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
2131  def : Pat<(v8f16 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
2132            (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
2133}
2134def : Pat<(f128  (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
2135          (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
2136
2137defm LDRHH : LoadUI<0b01, 0, 0b01, GPR32, uimm12s2, "ldrh",
2138                    [(set GPR32:$Rt,
2139                          (zextloadi16 (am_indexed16 GPR64sp:$Rn,
2140                                                     uimm12s2:$offset)))]>;
2141defm LDRBB : LoadUI<0b00, 0, 0b01, GPR32, uimm12s1, "ldrb",
2142                    [(set GPR32:$Rt,
2143                          (zextloadi8 (am_indexed8 GPR64sp:$Rn,
2144                                                   uimm12s1:$offset)))]>;
2145// zextload -> i64
2146def : Pat<(i64 (zextloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
2147    (SUBREG_TO_REG (i64 0), (LDRBBui GPR64sp:$Rn, uimm12s1:$offset), sub_32)>;
2148def : Pat<(i64 (zextloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))),
2149    (SUBREG_TO_REG (i64 0), (LDRHHui GPR64sp:$Rn, uimm12s2:$offset), sub_32)>;
2150
2151// zextloadi1 -> zextloadi8
2152def : Pat<(i32 (zextloadi1 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
2153          (LDRBBui GPR64sp:$Rn, uimm12s1:$offset)>;
2154def : Pat<(i64 (zextloadi1 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
2155    (SUBREG_TO_REG (i64 0), (LDRBBui GPR64sp:$Rn, uimm12s1:$offset), sub_32)>;
2156
2157// extload -> zextload
2158def : Pat<(i32 (extloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))),
2159          (LDRHHui GPR64sp:$Rn, uimm12s2:$offset)>;
2160def : Pat<(i32 (extloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
2161          (LDRBBui GPR64sp:$Rn, uimm12s1:$offset)>;
2162def : Pat<(i32 (extloadi1 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
2163          (LDRBBui GPR64sp:$Rn, uimm12s1:$offset)>;
2164def : Pat<(i64 (extloadi32 (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))),
2165    (SUBREG_TO_REG (i64 0), (LDRWui GPR64sp:$Rn, uimm12s4:$offset), sub_32)>;
2166def : Pat<(i64 (extloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))),
2167    (SUBREG_TO_REG (i64 0), (LDRHHui GPR64sp:$Rn, uimm12s2:$offset), sub_32)>;
2168def : Pat<(i64 (extloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
2169    (SUBREG_TO_REG (i64 0), (LDRBBui GPR64sp:$Rn, uimm12s1:$offset), sub_32)>;
2170def : Pat<(i64 (extloadi1 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
2171    (SUBREG_TO_REG (i64 0), (LDRBBui GPR64sp:$Rn, uimm12s1:$offset), sub_32)>;
2172
2173// load sign-extended half-word
2174defm LDRSHW : LoadUI<0b01, 0, 0b11, GPR32, uimm12s2, "ldrsh",
2175                     [(set GPR32:$Rt,
2176                           (sextloadi16 (am_indexed16 GPR64sp:$Rn,
2177                                                      uimm12s2:$offset)))]>;
2178defm LDRSHX : LoadUI<0b01, 0, 0b10, GPR64, uimm12s2, "ldrsh",
2179                     [(set GPR64:$Rt,
2180                           (sextloadi16 (am_indexed16 GPR64sp:$Rn,
2181                                                      uimm12s2:$offset)))]>;
2182
2183// load sign-extended byte
2184defm LDRSBW : LoadUI<0b00, 0, 0b11, GPR32, uimm12s1, "ldrsb",
2185                     [(set GPR32:$Rt,
2186                           (sextloadi8 (am_indexed8 GPR64sp:$Rn,
2187                                                    uimm12s1:$offset)))]>;
2188defm LDRSBX : LoadUI<0b00, 0, 0b10, GPR64, uimm12s1, "ldrsb",
2189                     [(set GPR64:$Rt,
2190                           (sextloadi8 (am_indexed8 GPR64sp:$Rn,
2191                                                    uimm12s1:$offset)))]>;
2192
2193// load sign-extended word
2194defm LDRSW  : LoadUI<0b10, 0, 0b10, GPR64, uimm12s4, "ldrsw",
2195                     [(set GPR64:$Rt,
2196                           (sextloadi32 (am_indexed32 GPR64sp:$Rn,
2197                                                      uimm12s4:$offset)))]>;
2198
2199// load zero-extended word
2200def : Pat<(i64 (zextloadi32 (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))),
2201      (SUBREG_TO_REG (i64 0), (LDRWui GPR64sp:$Rn, uimm12s4:$offset), sub_32)>;
2202
2203// Pre-fetch.
2204def PRFMui : PrefetchUI<0b11, 0, 0b10, "prfm",
2205                        [(AArch64Prefetch imm:$Rt,
2206                                        (am_indexed64 GPR64sp:$Rn,
2207                                                      uimm12s8:$offset))]>;
2208
2209def : InstAlias<"prfm $Rt, [$Rn]", (PRFMui prfop:$Rt, GPR64sp:$Rn, 0)>;
2210
2211//---
2212// (literal)
2213
2214def alignedglobal : PatLeaf<(iPTR iPTR:$label), [{
2215  if (auto *G = dyn_cast<GlobalAddressSDNode>(N)) {
2216    const DataLayout &DL = MF->getDataLayout();
2217    MaybeAlign Align = G->getGlobal()->getPointerAlignment(DL);
2218    return Align && *Align >= 4 && G->getOffset() % 4 == 0;
2219  }
2220  if (auto *C = dyn_cast<ConstantPoolSDNode>(N))
2221    return C->getAlignment() >= 4 && C->getOffset() % 4 == 0;
2222  return false;
2223}]>;
2224
2225def LDRWl : LoadLiteral<0b00, 0, GPR32z, "ldr",
2226  [(set GPR32z:$Rt, (load (AArch64adr alignedglobal:$label)))]>;
2227def LDRXl : LoadLiteral<0b01, 0, GPR64z, "ldr",
2228  [(set GPR64z:$Rt, (load (AArch64adr alignedglobal:$label)))]>;
2229def LDRSl : LoadLiteral<0b00, 1, FPR32Op, "ldr",
2230  [(set (f32 FPR32Op:$Rt), (load (AArch64adr alignedglobal:$label)))]>;
2231def LDRDl : LoadLiteral<0b01, 1, FPR64Op, "ldr",
2232  [(set (f64 FPR64Op:$Rt), (load (AArch64adr alignedglobal:$label)))]>;
2233def LDRQl : LoadLiteral<0b10, 1, FPR128Op, "ldr",
2234  [(set (f128 FPR128Op:$Rt), (load (AArch64adr alignedglobal:$label)))]>;
2235
2236// load sign-extended word
2237def LDRSWl : LoadLiteral<0b10, 0, GPR64z, "ldrsw",
2238  [(set GPR64z:$Rt, (sextloadi32 (AArch64adr alignedglobal:$label)))]>;
2239
2240let AddedComplexity = 20 in {
2241def : Pat<(i64 (zextloadi32 (AArch64adr alignedglobal:$label))),
2242        (SUBREG_TO_REG (i64 0), (LDRWl $label), sub_32)>;
2243}
2244
2245// prefetch
2246def PRFMl : PrefetchLiteral<0b11, 0, "prfm", []>;
2247//                   [(AArch64Prefetch imm:$Rt, tglobaladdr:$label)]>;
2248
2249//---
2250// (unscaled immediate)
2251defm LDURX : LoadUnscaled<0b11, 0, 0b01, GPR64z, "ldur",
2252                    [(set GPR64z:$Rt,
2253                          (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset)))]>;
2254defm LDURW : LoadUnscaled<0b10, 0, 0b01, GPR32z, "ldur",
2255                    [(set GPR32z:$Rt,
2256                          (load (am_unscaled32 GPR64sp:$Rn, simm9:$offset)))]>;
2257defm LDURB : LoadUnscaled<0b00, 1, 0b01, FPR8Op, "ldur",
2258                    [(set FPR8Op:$Rt,
2259                          (load (am_unscaled8 GPR64sp:$Rn, simm9:$offset)))]>;
2260defm LDURH : LoadUnscaled<0b01, 1, 0b01, FPR16Op, "ldur",
2261                    [(set FPR16Op:$Rt,
2262                          (load (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
2263defm LDURS : LoadUnscaled<0b10, 1, 0b01, FPR32Op, "ldur",
2264                    [(set (f32 FPR32Op:$Rt),
2265                          (load (am_unscaled32 GPR64sp:$Rn, simm9:$offset)))]>;
2266defm LDURD : LoadUnscaled<0b11, 1, 0b01, FPR64Op, "ldur",
2267                    [(set (f64 FPR64Op:$Rt),
2268                          (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset)))]>;
2269defm LDURQ : LoadUnscaled<0b00, 1, 0b11, FPR128Op, "ldur",
2270                    [(set (f128 FPR128Op:$Rt),
2271                          (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset)))]>;
2272
2273defm LDURHH
2274    : LoadUnscaled<0b01, 0, 0b01, GPR32, "ldurh",
2275             [(set GPR32:$Rt,
2276                    (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
2277defm LDURBB
2278    : LoadUnscaled<0b00, 0, 0b01, GPR32, "ldurb",
2279             [(set GPR32:$Rt,
2280                    (zextloadi8 (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
2281
2282// Match all load 64 bits width whose type is compatible with FPR64
2283let Predicates = [IsLE] in {
2284  def : Pat<(v2f32 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
2285            (LDURDi GPR64sp:$Rn, simm9:$offset)>;
2286  def : Pat<(v2i32 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
2287            (LDURDi GPR64sp:$Rn, simm9:$offset)>;
2288  def : Pat<(v4i16 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
2289            (LDURDi GPR64sp:$Rn, simm9:$offset)>;
2290  def : Pat<(v8i8 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
2291            (LDURDi GPR64sp:$Rn, simm9:$offset)>;
2292  def : Pat<(v4f16 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
2293            (LDURDi GPR64sp:$Rn, simm9:$offset)>;
2294}
2295def : Pat<(v1f64 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
2296          (LDURDi GPR64sp:$Rn, simm9:$offset)>;
2297def : Pat<(v1i64 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
2298          (LDURDi GPR64sp:$Rn, simm9:$offset)>;
2299
2300// Match all load 128 bits width whose type is compatible with FPR128
2301let Predicates = [IsLE] in {
2302  def : Pat<(v2f64 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
2303            (LDURQi GPR64sp:$Rn, simm9:$offset)>;
2304  def : Pat<(v2i64 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
2305            (LDURQi GPR64sp:$Rn, simm9:$offset)>;
2306  def : Pat<(v4f32 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
2307            (LDURQi GPR64sp:$Rn, simm9:$offset)>;
2308  def : Pat<(v4i32 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
2309            (LDURQi GPR64sp:$Rn, simm9:$offset)>;
2310  def : Pat<(v8i16 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
2311            (LDURQi GPR64sp:$Rn, simm9:$offset)>;
2312  def : Pat<(v16i8 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
2313            (LDURQi GPR64sp:$Rn, simm9:$offset)>;
2314  def : Pat<(v8f16 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
2315            (LDURQi GPR64sp:$Rn, simm9:$offset)>;
2316}
2317
2318//  anyext -> zext
2319def : Pat<(i32 (extloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
2320          (LDURHHi GPR64sp:$Rn, simm9:$offset)>;
2321def : Pat<(i32 (extloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
2322          (LDURBBi GPR64sp:$Rn, simm9:$offset)>;
2323def : Pat<(i32 (extloadi1 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
2324          (LDURBBi GPR64sp:$Rn, simm9:$offset)>;
2325def : Pat<(i64 (extloadi32 (am_unscaled32 GPR64sp:$Rn, simm9:$offset))),
2326    (SUBREG_TO_REG (i64 0), (LDURWi GPR64sp:$Rn, simm9:$offset), sub_32)>;
2327def : Pat<(i64 (extloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
2328    (SUBREG_TO_REG (i64 0), (LDURHHi GPR64sp:$Rn, simm9:$offset), sub_32)>;
2329def : Pat<(i64 (extloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
2330    (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
2331def : Pat<(i64 (extloadi1 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
2332    (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
2333// unscaled zext
2334def : Pat<(i32 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
2335          (LDURHHi GPR64sp:$Rn, simm9:$offset)>;
2336def : Pat<(i32 (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
2337          (LDURBBi GPR64sp:$Rn, simm9:$offset)>;
2338def : Pat<(i32 (zextloadi1 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
2339          (LDURBBi GPR64sp:$Rn, simm9:$offset)>;
2340def : Pat<(i64 (zextloadi32 (am_unscaled32 GPR64sp:$Rn, simm9:$offset))),
2341    (SUBREG_TO_REG (i64 0), (LDURWi GPR64sp:$Rn, simm9:$offset), sub_32)>;
2342def : Pat<(i64 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
2343    (SUBREG_TO_REG (i64 0), (LDURHHi GPR64sp:$Rn, simm9:$offset), sub_32)>;
2344def : Pat<(i64 (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
2345    (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
2346def : Pat<(i64 (zextloadi1 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
2347    (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
2348
2349
2350//---
2351// LDR mnemonics fall back to LDUR for negative or unaligned offsets.
2352
2353// Define new assembler match classes as we want to only match these when
2354// the don't otherwise match the scaled addressing mode for LDR/STR. Don't
2355// associate a DiagnosticType either, as we want the diagnostic for the
2356// canonical form (the scaled operand) to take precedence.
2357class SImm9OffsetOperand<int Width> : AsmOperandClass {
2358  let Name = "SImm9OffsetFB" # Width;
2359  let PredicateMethod = "isSImm9OffsetFB<" # Width # ">";
2360  let RenderMethod = "addImmOperands";
2361}
2362
2363def SImm9OffsetFB8Operand : SImm9OffsetOperand<8>;
2364def SImm9OffsetFB16Operand : SImm9OffsetOperand<16>;
2365def SImm9OffsetFB32Operand : SImm9OffsetOperand<32>;
2366def SImm9OffsetFB64Operand : SImm9OffsetOperand<64>;
2367def SImm9OffsetFB128Operand : SImm9OffsetOperand<128>;
2368
2369def simm9_offset_fb8 : Operand<i64> {
2370  let ParserMatchClass = SImm9OffsetFB8Operand;
2371}
2372def simm9_offset_fb16 : Operand<i64> {
2373  let ParserMatchClass = SImm9OffsetFB16Operand;
2374}
2375def simm9_offset_fb32 : Operand<i64> {
2376  let ParserMatchClass = SImm9OffsetFB32Operand;
2377}
2378def simm9_offset_fb64 : Operand<i64> {
2379  let ParserMatchClass = SImm9OffsetFB64Operand;
2380}
2381def simm9_offset_fb128 : Operand<i64> {
2382  let ParserMatchClass = SImm9OffsetFB128Operand;
2383}
2384
2385def : InstAlias<"ldr $Rt, [$Rn, $offset]",
2386                (LDURXi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb64:$offset), 0>;
2387def : InstAlias<"ldr $Rt, [$Rn, $offset]",
2388                (LDURWi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
2389def : InstAlias<"ldr $Rt, [$Rn, $offset]",
2390                (LDURBi FPR8Op:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
2391def : InstAlias<"ldr $Rt, [$Rn, $offset]",
2392                (LDURHi FPR16Op:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
2393def : InstAlias<"ldr $Rt, [$Rn, $offset]",
2394                (LDURSi FPR32Op:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
2395def : InstAlias<"ldr $Rt, [$Rn, $offset]",
2396                (LDURDi FPR64Op:$Rt, GPR64sp:$Rn, simm9_offset_fb64:$offset), 0>;
2397def : InstAlias<"ldr $Rt, [$Rn, $offset]",
2398               (LDURQi FPR128Op:$Rt, GPR64sp:$Rn, simm9_offset_fb128:$offset), 0>;
2399
2400// zextload -> i64
2401def : Pat<(i64 (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
2402  (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
2403def : Pat<(i64 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
2404  (SUBREG_TO_REG (i64 0), (LDURHHi GPR64sp:$Rn, simm9:$offset), sub_32)>;
2405
2406// load sign-extended half-word
2407defm LDURSHW
2408    : LoadUnscaled<0b01, 0, 0b11, GPR32, "ldursh",
2409               [(set GPR32:$Rt,
2410                    (sextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
2411defm LDURSHX
2412    : LoadUnscaled<0b01, 0, 0b10, GPR64, "ldursh",
2413              [(set GPR64:$Rt,
2414                    (sextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
2415
2416// load sign-extended byte
2417defm LDURSBW
2418    : LoadUnscaled<0b00, 0, 0b11, GPR32, "ldursb",
2419                [(set GPR32:$Rt,
2420                      (sextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset)))]>;
2421defm LDURSBX
2422    : LoadUnscaled<0b00, 0, 0b10, GPR64, "ldursb",
2423                [(set GPR64:$Rt,
2424                      (sextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset)))]>;
2425
2426// load sign-extended word
2427defm LDURSW
2428    : LoadUnscaled<0b10, 0, 0b10, GPR64, "ldursw",
2429              [(set GPR64:$Rt,
2430                    (sextloadi32 (am_unscaled32 GPR64sp:$Rn, simm9:$offset)))]>;
2431
2432// zero and sign extending aliases from generic LDR* mnemonics to LDUR*.
2433def : InstAlias<"ldrb $Rt, [$Rn, $offset]",
2434                (LDURBBi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
2435def : InstAlias<"ldrh $Rt, [$Rn, $offset]",
2436                (LDURHHi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
2437def : InstAlias<"ldrsb $Rt, [$Rn, $offset]",
2438                (LDURSBWi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
2439def : InstAlias<"ldrsb $Rt, [$Rn, $offset]",
2440                (LDURSBXi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
2441def : InstAlias<"ldrsh $Rt, [$Rn, $offset]",
2442                (LDURSHWi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
2443def : InstAlias<"ldrsh $Rt, [$Rn, $offset]",
2444                (LDURSHXi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
2445def : InstAlias<"ldrsw $Rt, [$Rn, $offset]",
2446                (LDURSWi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
2447
2448// Pre-fetch.
2449defm PRFUM : PrefetchUnscaled<0b11, 0, 0b10, "prfum",
2450                  [(AArch64Prefetch imm:$Rt,
2451                                  (am_unscaled64 GPR64sp:$Rn, simm9:$offset))]>;
2452
2453//---
2454// (unscaled immediate, unprivileged)
2455defm LDTRX : LoadUnprivileged<0b11, 0, 0b01, GPR64, "ldtr">;
2456defm LDTRW : LoadUnprivileged<0b10, 0, 0b01, GPR32, "ldtr">;
2457
2458defm LDTRH : LoadUnprivileged<0b01, 0, 0b01, GPR32, "ldtrh">;
2459defm LDTRB : LoadUnprivileged<0b00, 0, 0b01, GPR32, "ldtrb">;
2460
2461// load sign-extended half-word
2462defm LDTRSHW : LoadUnprivileged<0b01, 0, 0b11, GPR32, "ldtrsh">;
2463defm LDTRSHX : LoadUnprivileged<0b01, 0, 0b10, GPR64, "ldtrsh">;
2464
2465// load sign-extended byte
2466defm LDTRSBW : LoadUnprivileged<0b00, 0, 0b11, GPR32, "ldtrsb">;
2467defm LDTRSBX : LoadUnprivileged<0b00, 0, 0b10, GPR64, "ldtrsb">;
2468
2469// load sign-extended word
2470defm LDTRSW  : LoadUnprivileged<0b10, 0, 0b10, GPR64, "ldtrsw">;
2471
2472//---
2473// (immediate pre-indexed)
2474def LDRWpre : LoadPreIdx<0b10, 0, 0b01, GPR32z, "ldr">;
2475def LDRXpre : LoadPreIdx<0b11, 0, 0b01, GPR64z, "ldr">;
2476def LDRBpre : LoadPreIdx<0b00, 1, 0b01, FPR8Op,  "ldr">;
2477def LDRHpre : LoadPreIdx<0b01, 1, 0b01, FPR16Op, "ldr">;
2478def LDRSpre : LoadPreIdx<0b10, 1, 0b01, FPR32Op, "ldr">;
2479def LDRDpre : LoadPreIdx<0b11, 1, 0b01, FPR64Op, "ldr">;
2480def LDRQpre : LoadPreIdx<0b00, 1, 0b11, FPR128Op, "ldr">;
2481
2482// load sign-extended half-word
2483def LDRSHWpre : LoadPreIdx<0b01, 0, 0b11, GPR32z, "ldrsh">;
2484def LDRSHXpre : LoadPreIdx<0b01, 0, 0b10, GPR64z, "ldrsh">;
2485
2486// load sign-extended byte
2487def LDRSBWpre : LoadPreIdx<0b00, 0, 0b11, GPR32z, "ldrsb">;
2488def LDRSBXpre : LoadPreIdx<0b00, 0, 0b10, GPR64z, "ldrsb">;
2489
2490// load zero-extended byte
2491def LDRBBpre : LoadPreIdx<0b00, 0, 0b01, GPR32z, "ldrb">;
2492def LDRHHpre : LoadPreIdx<0b01, 0, 0b01, GPR32z, "ldrh">;
2493
2494// load sign-extended word
2495def LDRSWpre : LoadPreIdx<0b10, 0, 0b10, GPR64z, "ldrsw">;
2496
2497//---
2498// (immediate post-indexed)
2499def LDRWpost : LoadPostIdx<0b10, 0, 0b01, GPR32z, "ldr">;
2500def LDRXpost : LoadPostIdx<0b11, 0, 0b01, GPR64z, "ldr">;
2501def LDRBpost : LoadPostIdx<0b00, 1, 0b01, FPR8Op,  "ldr">;
2502def LDRHpost : LoadPostIdx<0b01, 1, 0b01, FPR16Op, "ldr">;
2503def LDRSpost : LoadPostIdx<0b10, 1, 0b01, FPR32Op, "ldr">;
2504def LDRDpost : LoadPostIdx<0b11, 1, 0b01, FPR64Op, "ldr">;
2505def LDRQpost : LoadPostIdx<0b00, 1, 0b11, FPR128Op, "ldr">;
2506
2507// load sign-extended half-word
2508def LDRSHWpost : LoadPostIdx<0b01, 0, 0b11, GPR32z, "ldrsh">;
2509def LDRSHXpost : LoadPostIdx<0b01, 0, 0b10, GPR64z, "ldrsh">;
2510
2511// load sign-extended byte
2512def LDRSBWpost : LoadPostIdx<0b00, 0, 0b11, GPR32z, "ldrsb">;
2513def LDRSBXpost : LoadPostIdx<0b00, 0, 0b10, GPR64z, "ldrsb">;
2514
2515// load zero-extended byte
2516def LDRBBpost : LoadPostIdx<0b00, 0, 0b01, GPR32z, "ldrb">;
2517def LDRHHpost : LoadPostIdx<0b01, 0, 0b01, GPR32z, "ldrh">;
2518
2519// load sign-extended word
2520def LDRSWpost : LoadPostIdx<0b10, 0, 0b10, GPR64z, "ldrsw">;
2521
2522//===----------------------------------------------------------------------===//
2523// Store instructions.
2524//===----------------------------------------------------------------------===//
2525
2526// Pair (indexed, offset)
2527// FIXME: Use dedicated range-checked addressing mode operand here.
2528defm STPW : StorePairOffset<0b00, 0, GPR32z, simm7s4, "stp">;
2529defm STPX : StorePairOffset<0b10, 0, GPR64z, simm7s8, "stp">;
2530defm STPS : StorePairOffset<0b00, 1, FPR32Op, simm7s4, "stp">;
2531defm STPD : StorePairOffset<0b01, 1, FPR64Op, simm7s8, "stp">;
2532defm STPQ : StorePairOffset<0b10, 1, FPR128Op, simm7s16, "stp">;
2533
2534// Pair (pre-indexed)
2535def STPWpre : StorePairPreIdx<0b00, 0, GPR32z, simm7s4, "stp">;
2536def STPXpre : StorePairPreIdx<0b10, 0, GPR64z, simm7s8, "stp">;
2537def STPSpre : StorePairPreIdx<0b00, 1, FPR32Op, simm7s4, "stp">;
2538def STPDpre : StorePairPreIdx<0b01, 1, FPR64Op, simm7s8, "stp">;
2539def STPQpre : StorePairPreIdx<0b10, 1, FPR128Op, simm7s16, "stp">;
2540
2541// Pair (pre-indexed)
2542def STPWpost : StorePairPostIdx<0b00, 0, GPR32z, simm7s4, "stp">;
2543def STPXpost : StorePairPostIdx<0b10, 0, GPR64z, simm7s8, "stp">;
2544def STPSpost : StorePairPostIdx<0b00, 1, FPR32Op, simm7s4, "stp">;
2545def STPDpost : StorePairPostIdx<0b01, 1, FPR64Op, simm7s8, "stp">;
2546def STPQpost : StorePairPostIdx<0b10, 1, FPR128Op, simm7s16, "stp">;
2547
2548// Pair (no allocate)
2549defm STNPW : StorePairNoAlloc<0b00, 0, GPR32z, simm7s4, "stnp">;
2550defm STNPX : StorePairNoAlloc<0b10, 0, GPR64z, simm7s8, "stnp">;
2551defm STNPS : StorePairNoAlloc<0b00, 1, FPR32Op, simm7s4, "stnp">;
2552defm STNPD : StorePairNoAlloc<0b01, 1, FPR64Op, simm7s8, "stnp">;
2553defm STNPQ : StorePairNoAlloc<0b10, 1, FPR128Op, simm7s16, "stnp">;
2554
2555//---
2556// (Register offset)
2557
2558// Integer
2559defm STRBB : Store8RO< 0b00, 0, 0b00, GPR32, "strb", i32, truncstorei8>;
2560defm STRHH : Store16RO<0b01, 0, 0b00, GPR32, "strh", i32, truncstorei16>;
2561defm STRW  : Store32RO<0b10, 0, 0b00, GPR32, "str",  i32, store>;
2562defm STRX  : Store64RO<0b11, 0, 0b00, GPR64, "str",  i64, store>;
2563
2564
2565// Floating-point
2566defm STRB : Store8RO< 0b00,  1, 0b00, FPR8Op,   "str", untyped, store>;
2567defm STRH : Store16RO<0b01,  1, 0b00, FPR16Op,  "str", f16,     store>;
2568defm STRS : Store32RO<0b10,  1, 0b00, FPR32Op,  "str", f32,     store>;
2569defm STRD : Store64RO<0b11,  1, 0b00, FPR64Op,  "str", f64,     store>;
2570defm STRQ : Store128RO<0b00, 1, 0b10, FPR128Op, "str", f128,    store>;
2571
2572let Predicates = [UseSTRQro], AddedComplexity = 10 in {
2573  def : Pat<(store (f128 FPR128:$Rt),
2574                        (ro_Windexed128 GPR64sp:$Rn, GPR32:$Rm,
2575                                        ro_Wextend128:$extend)),
2576            (STRQroW FPR128:$Rt, GPR64sp:$Rn, GPR32:$Rm, ro_Wextend128:$extend)>;
2577  def : Pat<(store (f128 FPR128:$Rt),
2578                        (ro_Xindexed128 GPR64sp:$Rn, GPR64:$Rm,
2579                                        ro_Xextend128:$extend)),
2580            (STRQroX FPR128:$Rt, GPR64sp:$Rn, GPR64:$Rm, ro_Wextend128:$extend)>;
2581}
2582
2583multiclass TruncStoreFrom64ROPat<ROAddrMode ro, SDPatternOperator storeop,
2584                                 Instruction STRW, Instruction STRX> {
2585
2586  def : Pat<(storeop GPR64:$Rt,
2587                     (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)),
2588            (STRW (EXTRACT_SUBREG GPR64:$Rt, sub_32),
2589                  GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
2590
2591  def : Pat<(storeop GPR64:$Rt,
2592                     (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)),
2593            (STRX (EXTRACT_SUBREG GPR64:$Rt, sub_32),
2594                  GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
2595}
2596
2597let AddedComplexity = 10 in {
2598  // truncstore i64
2599  defm : TruncStoreFrom64ROPat<ro8,  truncstorei8,  STRBBroW, STRBBroX>;
2600  defm : TruncStoreFrom64ROPat<ro16, truncstorei16, STRHHroW, STRHHroX>;
2601  defm : TruncStoreFrom64ROPat<ro32, truncstorei32, STRWroW,  STRWroX>;
2602}
2603
2604multiclass VecROStorePat<ROAddrMode ro, ValueType VecTy, RegisterClass FPR,
2605                         Instruction STRW, Instruction STRX> {
2606  def : Pat<(store (VecTy FPR:$Rt),
2607                   (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)),
2608            (STRW FPR:$Rt, GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
2609
2610  def : Pat<(store (VecTy FPR:$Rt),
2611                   (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)),
2612            (STRX FPR:$Rt, GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
2613}
2614
2615let AddedComplexity = 10 in {
2616// Match all store 64 bits width whose type is compatible with FPR64
2617let Predicates = [IsLE] in {
2618  // We must use ST1 to store vectors in big-endian.
2619  defm : VecROStorePat<ro64, v2i32, FPR64, STRDroW, STRDroX>;
2620  defm : VecROStorePat<ro64, v2f32, FPR64, STRDroW, STRDroX>;
2621  defm : VecROStorePat<ro64, v4i16, FPR64, STRDroW, STRDroX>;
2622  defm : VecROStorePat<ro64, v8i8, FPR64, STRDroW, STRDroX>;
2623  defm : VecROStorePat<ro64, v4f16, FPR64, STRDroW, STRDroX>;
2624}
2625
2626defm : VecROStorePat<ro64, v1i64, FPR64, STRDroW, STRDroX>;
2627defm : VecROStorePat<ro64, v1f64, FPR64, STRDroW, STRDroX>;
2628
2629// Match all store 128 bits width whose type is compatible with FPR128
2630let Predicates = [IsLE, UseSTRQro] in {
2631  // We must use ST1 to store vectors in big-endian.
2632  defm : VecROStorePat<ro128, v2i64, FPR128, STRQroW, STRQroX>;
2633  defm : VecROStorePat<ro128, v2f64, FPR128, STRQroW, STRQroX>;
2634  defm : VecROStorePat<ro128, v4i32, FPR128, STRQroW, STRQroX>;
2635  defm : VecROStorePat<ro128, v4f32, FPR128, STRQroW, STRQroX>;
2636  defm : VecROStorePat<ro128, v8i16, FPR128, STRQroW, STRQroX>;
2637  defm : VecROStorePat<ro128, v16i8, FPR128, STRQroW, STRQroX>;
2638  defm : VecROStorePat<ro128, v8f16, FPR128, STRQroW, STRQroX>;
2639}
2640} // AddedComplexity = 10
2641
2642// Match stores from lane 0 to the appropriate subreg's store.
2643multiclass VecROStoreLane0Pat<ROAddrMode ro, SDPatternOperator storeop,
2644                              ValueType VecTy, ValueType STy,
2645                              SubRegIndex SubRegIdx,
2646                              Instruction STRW, Instruction STRX> {
2647
2648  def : Pat<(storeop (STy (vector_extract (VecTy VecListOne128:$Vt), 0)),
2649                     (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)),
2650            (STRW (EXTRACT_SUBREG VecListOne128:$Vt, SubRegIdx),
2651                  GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
2652
2653  def : Pat<(storeop (STy (vector_extract (VecTy VecListOne128:$Vt), 0)),
2654                     (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)),
2655            (STRX (EXTRACT_SUBREG VecListOne128:$Vt, SubRegIdx),
2656                  GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
2657}
2658
2659let AddedComplexity = 19 in {
2660  defm : VecROStoreLane0Pat<ro16, truncstorei16, v8i16, i32, hsub, STRHroW, STRHroX>;
2661  defm : VecROStoreLane0Pat<ro16,         store, v8f16, f16, hsub, STRHroW, STRHroX>;
2662  defm : VecROStoreLane0Pat<ro32,         store, v4i32, i32, ssub, STRSroW, STRSroX>;
2663  defm : VecROStoreLane0Pat<ro32,         store, v4f32, f32, ssub, STRSroW, STRSroX>;
2664  defm : VecROStoreLane0Pat<ro64,         store, v2i64, i64, dsub, STRDroW, STRDroX>;
2665  defm : VecROStoreLane0Pat<ro64,         store, v2f64, f64, dsub, STRDroW, STRDroX>;
2666}
2667
2668//---
2669// (unsigned immediate)
2670defm STRX : StoreUIz<0b11, 0, 0b00, GPR64z, uimm12s8, "str",
2671                   [(store GPR64z:$Rt,
2672                            (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))]>;
2673defm STRW : StoreUIz<0b10, 0, 0b00, GPR32z, uimm12s4, "str",
2674                    [(store GPR32z:$Rt,
2675                            (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))]>;
2676defm STRB : StoreUI<0b00, 1, 0b00, FPR8Op, uimm12s1, "str",
2677                    [(store FPR8Op:$Rt,
2678                            (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))]>;
2679defm STRH : StoreUI<0b01, 1, 0b00, FPR16Op, uimm12s2, "str",
2680                    [(store (f16 FPR16Op:$Rt),
2681                            (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))]>;
2682defm STRS : StoreUI<0b10, 1, 0b00, FPR32Op, uimm12s4, "str",
2683                    [(store (f32 FPR32Op:$Rt),
2684                            (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))]>;
2685defm STRD : StoreUI<0b11, 1, 0b00, FPR64Op, uimm12s8, "str",
2686                    [(store (f64 FPR64Op:$Rt),
2687                            (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))]>;
2688defm STRQ : StoreUI<0b00, 1, 0b10, FPR128Op, uimm12s16, "str", []>;
2689
2690defm STRHH : StoreUIz<0b01, 0, 0b00, GPR32z, uimm12s2, "strh",
2691                     [(truncstorei16 GPR32z:$Rt,
2692                                     (am_indexed16 GPR64sp:$Rn,
2693                                                   uimm12s2:$offset))]>;
2694defm STRBB : StoreUIz<0b00, 0, 0b00, GPR32z, uimm12s1,  "strb",
2695                     [(truncstorei8 GPR32z:$Rt,
2696                                    (am_indexed8 GPR64sp:$Rn,
2697                                                 uimm12s1:$offset))]>;
2698
2699let AddedComplexity = 10 in {
2700
2701// Match all store 64 bits width whose type is compatible with FPR64
2702def : Pat<(store (v1i64 FPR64:$Rt),
2703                 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
2704          (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
2705def : Pat<(store (v1f64 FPR64:$Rt),
2706                 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
2707          (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
2708
2709let Predicates = [IsLE] in {
2710  // We must use ST1 to store vectors in big-endian.
2711  def : Pat<(store (v2f32 FPR64:$Rt),
2712                   (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
2713            (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
2714  def : Pat<(store (v8i8 FPR64:$Rt),
2715                   (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
2716            (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
2717  def : Pat<(store (v4i16 FPR64:$Rt),
2718                   (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
2719            (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
2720  def : Pat<(store (v2i32 FPR64:$Rt),
2721                   (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
2722            (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
2723  def : Pat<(store (v4f16 FPR64:$Rt),
2724                   (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
2725            (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
2726}
2727
2728// Match all store 128 bits width whose type is compatible with FPR128
2729def : Pat<(store (f128  FPR128:$Rt),
2730                 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
2731          (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
2732
2733let Predicates = [IsLE] in {
2734  // We must use ST1 to store vectors in big-endian.
2735  def : Pat<(store (v4f32 FPR128:$Rt),
2736                   (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
2737            (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
2738  def : Pat<(store (v2f64 FPR128:$Rt),
2739                   (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
2740            (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
2741  def : Pat<(store (v16i8 FPR128:$Rt),
2742                   (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
2743            (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
2744  def : Pat<(store (v8i16 FPR128:$Rt),
2745                   (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
2746            (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
2747  def : Pat<(store (v4i32 FPR128:$Rt),
2748                   (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
2749            (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
2750  def : Pat<(store (v2i64 FPR128:$Rt),
2751                   (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
2752            (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
2753  def : Pat<(store (v8f16 FPR128:$Rt),
2754                   (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
2755            (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
2756}
2757
2758// truncstore i64
2759def : Pat<(truncstorei32 GPR64:$Rt,
2760                         (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset)),
2761  (STRWui (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, uimm12s4:$offset)>;
2762def : Pat<(truncstorei16 GPR64:$Rt,
2763                         (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset)),
2764  (STRHHui (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, uimm12s2:$offset)>;
2765def : Pat<(truncstorei8 GPR64:$Rt, (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset)),
2766  (STRBBui (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, uimm12s1:$offset)>;
2767
2768} // AddedComplexity = 10
2769
2770// Match stores from lane 0 to the appropriate subreg's store.
2771multiclass VecStoreLane0Pat<Operand UIAddrMode, SDPatternOperator storeop,
2772                            ValueType VTy, ValueType STy,
2773                            SubRegIndex SubRegIdx, Operand IndexType,
2774                            Instruction STR> {
2775  def : Pat<(storeop (STy (vector_extract (VTy VecListOne128:$Vt), 0)),
2776                     (UIAddrMode GPR64sp:$Rn, IndexType:$offset)),
2777            (STR (EXTRACT_SUBREG VecListOne128:$Vt, SubRegIdx),
2778                 GPR64sp:$Rn, IndexType:$offset)>;
2779}
2780
2781let AddedComplexity = 19 in {
2782  defm : VecStoreLane0Pat<am_indexed16, truncstorei16, v8i16, i32, hsub, uimm12s2, STRHui>;
2783  defm : VecStoreLane0Pat<am_indexed16,         store, v8f16, f16, hsub, uimm12s2, STRHui>;
2784  defm : VecStoreLane0Pat<am_indexed32,         store, v4i32, i32, ssub, uimm12s4, STRSui>;
2785  defm : VecStoreLane0Pat<am_indexed32,         store, v4f32, f32, ssub, uimm12s4, STRSui>;
2786  defm : VecStoreLane0Pat<am_indexed64,         store, v2i64, i64, dsub, uimm12s8, STRDui>;
2787  defm : VecStoreLane0Pat<am_indexed64,         store, v2f64, f64, dsub, uimm12s8, STRDui>;
2788}
2789
2790//---
2791// (unscaled immediate)
2792defm STURX : StoreUnscaled<0b11, 0, 0b00, GPR64z, "stur",
2793                         [(store GPR64z:$Rt,
2794                                 (am_unscaled64 GPR64sp:$Rn, simm9:$offset))]>;
2795defm STURW : StoreUnscaled<0b10, 0, 0b00, GPR32z, "stur",
2796                         [(store GPR32z:$Rt,
2797                                 (am_unscaled32 GPR64sp:$Rn, simm9:$offset))]>;
2798defm STURB : StoreUnscaled<0b00, 1, 0b00, FPR8Op, "stur",
2799                         [(store FPR8Op:$Rt,
2800                                 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))]>;
2801defm STURH : StoreUnscaled<0b01, 1, 0b00, FPR16Op, "stur",
2802                         [(store (f16 FPR16Op:$Rt),
2803                                 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))]>;
2804defm STURS : StoreUnscaled<0b10, 1, 0b00, FPR32Op, "stur",
2805                         [(store (f32 FPR32Op:$Rt),
2806                                 (am_unscaled32 GPR64sp:$Rn, simm9:$offset))]>;
2807defm STURD : StoreUnscaled<0b11, 1, 0b00, FPR64Op, "stur",
2808                         [(store (f64 FPR64Op:$Rt),
2809                                 (am_unscaled64 GPR64sp:$Rn, simm9:$offset))]>;
2810defm STURQ : StoreUnscaled<0b00, 1, 0b10, FPR128Op, "stur",
2811                         [(store (f128 FPR128Op:$Rt),
2812                                 (am_unscaled128 GPR64sp:$Rn, simm9:$offset))]>;
2813defm STURHH : StoreUnscaled<0b01, 0, 0b00, GPR32z, "sturh",
2814                         [(truncstorei16 GPR32z:$Rt,
2815                                 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))]>;
2816defm STURBB : StoreUnscaled<0b00, 0, 0b00, GPR32z, "sturb",
2817                         [(truncstorei8 GPR32z:$Rt,
2818                                  (am_unscaled8 GPR64sp:$Rn, simm9:$offset))]>;
2819
2820// Armv8.4 Weaker Release Consistency enhancements
2821//         LDAPR & STLR with Immediate Offset instructions
2822let Predicates = [HasRCPC_IMMO] in {
2823defm STLURB     : BaseStoreUnscaleV84<"stlurb",  0b00, 0b00, GPR32>;
2824defm STLURH     : BaseStoreUnscaleV84<"stlurh",  0b01, 0b00, GPR32>;
2825defm STLURW     : BaseStoreUnscaleV84<"stlur",   0b10, 0b00, GPR32>;
2826defm STLURX     : BaseStoreUnscaleV84<"stlur",   0b11, 0b00, GPR64>;
2827defm LDAPURB    : BaseLoadUnscaleV84<"ldapurb",  0b00, 0b01, GPR32>;
2828defm LDAPURSBW  : BaseLoadUnscaleV84<"ldapursb", 0b00, 0b11, GPR32>;
2829defm LDAPURSBX  : BaseLoadUnscaleV84<"ldapursb", 0b00, 0b10, GPR64>;
2830defm LDAPURH    : BaseLoadUnscaleV84<"ldapurh",  0b01, 0b01, GPR32>;
2831defm LDAPURSHW  : BaseLoadUnscaleV84<"ldapursh", 0b01, 0b11, GPR32>;
2832defm LDAPURSHX  : BaseLoadUnscaleV84<"ldapursh", 0b01, 0b10, GPR64>;
2833defm LDAPUR     : BaseLoadUnscaleV84<"ldapur",   0b10, 0b01, GPR32>;
2834defm LDAPURSW   : BaseLoadUnscaleV84<"ldapursw", 0b10, 0b10, GPR64>;
2835defm LDAPURX    : BaseLoadUnscaleV84<"ldapur",   0b11, 0b01, GPR64>;
2836}
2837
2838// Match all store 64 bits width whose type is compatible with FPR64
2839def : Pat<(store (v1f64 FPR64:$Rt), (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2840          (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2841def : Pat<(store (v1i64 FPR64:$Rt), (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2842          (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2843
2844let AddedComplexity = 10 in {
2845
2846let Predicates = [IsLE] in {
2847  // We must use ST1 to store vectors in big-endian.
2848  def : Pat<(store (v2f32 FPR64:$Rt),
2849                   (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2850            (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2851  def : Pat<(store (v8i8 FPR64:$Rt),
2852                   (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2853            (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2854  def : Pat<(store (v4i16 FPR64:$Rt),
2855                   (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2856            (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2857  def : Pat<(store (v2i32 FPR64:$Rt),
2858                   (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2859            (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2860  def : Pat<(store (v4f16 FPR64:$Rt),
2861                   (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2862            (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2863}
2864
2865// Match all store 128 bits width whose type is compatible with FPR128
2866def : Pat<(store (f128 FPR128:$Rt), (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2867          (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2868
2869let Predicates = [IsLE] in {
2870  // We must use ST1 to store vectors in big-endian.
2871  def : Pat<(store (v4f32 FPR128:$Rt),
2872                   (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2873            (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2874  def : Pat<(store (v2f64 FPR128:$Rt),
2875                   (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2876            (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2877  def : Pat<(store (v16i8 FPR128:$Rt),
2878                   (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2879            (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2880  def : Pat<(store (v8i16 FPR128:$Rt),
2881                   (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2882            (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2883  def : Pat<(store (v4i32 FPR128:$Rt),
2884                   (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2885            (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2886  def : Pat<(store (v2i64 FPR128:$Rt),
2887                   (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2888            (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2889  def : Pat<(store (v2f64 FPR128:$Rt),
2890                   (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2891            (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2892  def : Pat<(store (v8f16 FPR128:$Rt),
2893                   (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2894            (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2895}
2896
2897} // AddedComplexity = 10
2898
2899// unscaled i64 truncating stores
2900def : Pat<(truncstorei32 GPR64:$Rt, (am_unscaled32 GPR64sp:$Rn, simm9:$offset)),
2901  (STURWi (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, simm9:$offset)>;
2902def : Pat<(truncstorei16 GPR64:$Rt, (am_unscaled16 GPR64sp:$Rn, simm9:$offset)),
2903  (STURHHi (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, simm9:$offset)>;
2904def : Pat<(truncstorei8 GPR64:$Rt, (am_unscaled8 GPR64sp:$Rn, simm9:$offset)),
2905  (STURBBi (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, simm9:$offset)>;
2906
2907// Match stores from lane 0 to the appropriate subreg's store.
2908multiclass VecStoreULane0Pat<SDPatternOperator StoreOp,
2909                             ValueType VTy, ValueType STy,
2910                             SubRegIndex SubRegIdx, Instruction STR> {
2911  defm : VecStoreLane0Pat<am_unscaled128, StoreOp, VTy, STy, SubRegIdx, simm9, STR>;
2912}
2913
2914let AddedComplexity = 19 in {
2915  defm : VecStoreULane0Pat<truncstorei16, v8i16, i32, hsub, STURHi>;
2916  defm : VecStoreULane0Pat<store,         v8f16, f16, hsub, STURHi>;
2917  defm : VecStoreULane0Pat<store,         v4i32, i32, ssub, STURSi>;
2918  defm : VecStoreULane0Pat<store,         v4f32, f32, ssub, STURSi>;
2919  defm : VecStoreULane0Pat<store,         v2i64, i64, dsub, STURDi>;
2920  defm : VecStoreULane0Pat<store,         v2f64, f64, dsub, STURDi>;
2921}
2922
2923//---
2924// STR mnemonics fall back to STUR for negative or unaligned offsets.
2925def : InstAlias<"str $Rt, [$Rn, $offset]",
2926                (STURXi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb64:$offset), 0>;
2927def : InstAlias<"str $Rt, [$Rn, $offset]",
2928                (STURWi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
2929def : InstAlias<"str $Rt, [$Rn, $offset]",
2930                (STURBi FPR8Op:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
2931def : InstAlias<"str $Rt, [$Rn, $offset]",
2932                (STURHi FPR16Op:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
2933def : InstAlias<"str $Rt, [$Rn, $offset]",
2934                (STURSi FPR32Op:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
2935def : InstAlias<"str $Rt, [$Rn, $offset]",
2936                (STURDi FPR64Op:$Rt, GPR64sp:$Rn, simm9_offset_fb64:$offset), 0>;
2937def : InstAlias<"str $Rt, [$Rn, $offset]",
2938                (STURQi FPR128Op:$Rt, GPR64sp:$Rn, simm9_offset_fb128:$offset), 0>;
2939
2940def : InstAlias<"strb $Rt, [$Rn, $offset]",
2941                (STURBBi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
2942def : InstAlias<"strh $Rt, [$Rn, $offset]",
2943                (STURHHi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
2944
2945//---
2946// (unscaled immediate, unprivileged)
2947defm STTRW : StoreUnprivileged<0b10, 0, 0b00, GPR32, "sttr">;
2948defm STTRX : StoreUnprivileged<0b11, 0, 0b00, GPR64, "sttr">;
2949
2950defm STTRH : StoreUnprivileged<0b01, 0, 0b00, GPR32, "sttrh">;
2951defm STTRB : StoreUnprivileged<0b00, 0, 0b00, GPR32, "sttrb">;
2952
2953//---
2954// (immediate pre-indexed)
2955def STRWpre : StorePreIdx<0b10, 0, 0b00, GPR32z, "str",  pre_store, i32>;
2956def STRXpre : StorePreIdx<0b11, 0, 0b00, GPR64z, "str",  pre_store, i64>;
2957def STRBpre : StorePreIdx<0b00, 1, 0b00, FPR8Op,  "str",  pre_store, untyped>;
2958def STRHpre : StorePreIdx<0b01, 1, 0b00, FPR16Op, "str",  pre_store, f16>;
2959def STRSpre : StorePreIdx<0b10, 1, 0b00, FPR32Op, "str",  pre_store, f32>;
2960def STRDpre : StorePreIdx<0b11, 1, 0b00, FPR64Op, "str",  pre_store, f64>;
2961def STRQpre : StorePreIdx<0b00, 1, 0b10, FPR128Op, "str", pre_store, f128>;
2962
2963def STRBBpre : StorePreIdx<0b00, 0, 0b00, GPR32z, "strb", pre_truncsti8,  i32>;
2964def STRHHpre : StorePreIdx<0b01, 0, 0b00, GPR32z, "strh", pre_truncsti16, i32>;
2965
2966// truncstore i64
2967def : Pat<(pre_truncsti32 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
2968  (STRWpre (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
2969           simm9:$off)>;
2970def : Pat<(pre_truncsti16 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
2971  (STRHHpre (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
2972            simm9:$off)>;
2973def : Pat<(pre_truncsti8 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
2974  (STRBBpre (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
2975            simm9:$off)>;
2976
2977def : Pat<(pre_store (v8i8 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2978          (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2979def : Pat<(pre_store (v4i16 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2980          (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2981def : Pat<(pre_store (v2i32 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2982          (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2983def : Pat<(pre_store (v2f32 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2984          (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2985def : Pat<(pre_store (v1i64 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2986          (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2987def : Pat<(pre_store (v1f64 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2988          (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2989def : Pat<(pre_store (v4f16 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2990          (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2991
2992def : Pat<(pre_store (v16i8 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2993          (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2994def : Pat<(pre_store (v8i16 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2995          (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2996def : Pat<(pre_store (v4i32 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2997          (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2998def : Pat<(pre_store (v4f32 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2999          (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
3000def : Pat<(pre_store (v2i64 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
3001          (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
3002def : Pat<(pre_store (v2f64 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
3003          (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
3004def : Pat<(pre_store (v8f16 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
3005          (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
3006
3007//---
3008// (immediate post-indexed)
3009def STRWpost : StorePostIdx<0b10, 0, 0b00, GPR32z,  "str", post_store, i32>;
3010def STRXpost : StorePostIdx<0b11, 0, 0b00, GPR64z,  "str", post_store, i64>;
3011def STRBpost : StorePostIdx<0b00, 1, 0b00, FPR8Op,   "str", post_store, untyped>;
3012def STRHpost : StorePostIdx<0b01, 1, 0b00, FPR16Op,  "str", post_store, f16>;
3013def STRSpost : StorePostIdx<0b10, 1, 0b00, FPR32Op,  "str", post_store, f32>;
3014def STRDpost : StorePostIdx<0b11, 1, 0b00, FPR64Op,  "str", post_store, f64>;
3015def STRQpost : StorePostIdx<0b00, 1, 0b10, FPR128Op, "str", post_store, f128>;
3016
3017def STRBBpost : StorePostIdx<0b00, 0, 0b00, GPR32z, "strb", post_truncsti8, i32>;
3018def STRHHpost : StorePostIdx<0b01, 0, 0b00, GPR32z, "strh", post_truncsti16, i32>;
3019
3020// truncstore i64
3021def : Pat<(post_truncsti32 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
3022  (STRWpost (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
3023            simm9:$off)>;
3024def : Pat<(post_truncsti16 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
3025  (STRHHpost (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
3026             simm9:$off)>;
3027def : Pat<(post_truncsti8 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
3028  (STRBBpost (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
3029             simm9:$off)>;
3030
3031def : Pat<(post_store (v8i8 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
3032          (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
3033def : Pat<(post_store (v4i16 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
3034          (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
3035def : Pat<(post_store (v2i32 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
3036          (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
3037def : Pat<(post_store (v2f32 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
3038          (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
3039def : Pat<(post_store (v1i64 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
3040          (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
3041def : Pat<(post_store (v1f64 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
3042          (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
3043def : Pat<(post_store (v4f16 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
3044          (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
3045
3046def : Pat<(post_store (v16i8 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
3047          (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
3048def : Pat<(post_store (v8i16 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
3049          (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
3050def : Pat<(post_store (v4i32 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
3051          (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
3052def : Pat<(post_store (v4f32 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
3053          (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
3054def : Pat<(post_store (v2i64 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
3055          (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
3056def : Pat<(post_store (v2f64 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
3057          (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
3058def : Pat<(post_store (v8f16 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
3059          (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
3060
3061//===----------------------------------------------------------------------===//
3062// Load/store exclusive instructions.
3063//===----------------------------------------------------------------------===//
3064
3065def LDARW  : LoadAcquire   <0b10, 1, 1, 0, 1, GPR32, "ldar">;
3066def LDARX  : LoadAcquire   <0b11, 1, 1, 0, 1, GPR64, "ldar">;
3067def LDARB  : LoadAcquire   <0b00, 1, 1, 0, 1, GPR32, "ldarb">;
3068def LDARH  : LoadAcquire   <0b01, 1, 1, 0, 1, GPR32, "ldarh">;
3069
3070def LDAXRW : LoadExclusive <0b10, 0, 1, 0, 1, GPR32, "ldaxr">;
3071def LDAXRX : LoadExclusive <0b11, 0, 1, 0, 1, GPR64, "ldaxr">;
3072def LDAXRB : LoadExclusive <0b00, 0, 1, 0, 1, GPR32, "ldaxrb">;
3073def LDAXRH : LoadExclusive <0b01, 0, 1, 0, 1, GPR32, "ldaxrh">;
3074
3075def LDXRW  : LoadExclusive <0b10, 0, 1, 0, 0, GPR32, "ldxr">;
3076def LDXRX  : LoadExclusive <0b11, 0, 1, 0, 0, GPR64, "ldxr">;
3077def LDXRB  : LoadExclusive <0b00, 0, 1, 0, 0, GPR32, "ldxrb">;
3078def LDXRH  : LoadExclusive <0b01, 0, 1, 0, 0, GPR32, "ldxrh">;
3079
3080def STLRW  : StoreRelease  <0b10, 1, 0, 0, 1, GPR32, "stlr">;
3081def STLRX  : StoreRelease  <0b11, 1, 0, 0, 1, GPR64, "stlr">;
3082def STLRB  : StoreRelease  <0b00, 1, 0, 0, 1, GPR32, "stlrb">;
3083def STLRH  : StoreRelease  <0b01, 1, 0, 0, 1, GPR32, "stlrh">;
3084
3085def STLXRW : StoreExclusive<0b10, 0, 0, 0, 1, GPR32, "stlxr">;
3086def STLXRX : StoreExclusive<0b11, 0, 0, 0, 1, GPR64, "stlxr">;
3087def STLXRB : StoreExclusive<0b00, 0, 0, 0, 1, GPR32, "stlxrb">;
3088def STLXRH : StoreExclusive<0b01, 0, 0, 0, 1, GPR32, "stlxrh">;
3089
3090def STXRW  : StoreExclusive<0b10, 0, 0, 0, 0, GPR32, "stxr">;
3091def STXRX  : StoreExclusive<0b11, 0, 0, 0, 0, GPR64, "stxr">;
3092def STXRB  : StoreExclusive<0b00, 0, 0, 0, 0, GPR32, "stxrb">;
3093def STXRH  : StoreExclusive<0b01, 0, 0, 0, 0, GPR32, "stxrh">;
3094
3095def LDAXPW : LoadExclusivePair<0b10, 0, 1, 1, 1, GPR32, "ldaxp">;
3096def LDAXPX : LoadExclusivePair<0b11, 0, 1, 1, 1, GPR64, "ldaxp">;
3097
3098def LDXPW  : LoadExclusivePair<0b10, 0, 1, 1, 0, GPR32, "ldxp">;
3099def LDXPX  : LoadExclusivePair<0b11, 0, 1, 1, 0, GPR64, "ldxp">;
3100
3101def STLXPW : StoreExclusivePair<0b10, 0, 0, 1, 1, GPR32, "stlxp">;
3102def STLXPX : StoreExclusivePair<0b11, 0, 0, 1, 1, GPR64, "stlxp">;
3103
3104def STXPW  : StoreExclusivePair<0b10, 0, 0, 1, 0, GPR32, "stxp">;
3105def STXPX  : StoreExclusivePair<0b11, 0, 0, 1, 0, GPR64, "stxp">;
3106
3107let Predicates = [HasLOR] in {
3108  // v8.1a "Limited Order Region" extension load-acquire instructions
3109  def LDLARW  : LoadAcquire   <0b10, 1, 1, 0, 0, GPR32, "ldlar">;
3110  def LDLARX  : LoadAcquire   <0b11, 1, 1, 0, 0, GPR64, "ldlar">;
3111  def LDLARB  : LoadAcquire   <0b00, 1, 1, 0, 0, GPR32, "ldlarb">;
3112  def LDLARH  : LoadAcquire   <0b01, 1, 1, 0, 0, GPR32, "ldlarh">;
3113
3114  // v8.1a "Limited Order Region" extension store-release instructions
3115  def STLLRW  : StoreRelease   <0b10, 1, 0, 0, 0, GPR32, "stllr">;
3116  def STLLRX  : StoreRelease   <0b11, 1, 0, 0, 0, GPR64, "stllr">;
3117  def STLLRB  : StoreRelease   <0b00, 1, 0, 0, 0, GPR32, "stllrb">;
3118  def STLLRH  : StoreRelease   <0b01, 1, 0, 0, 0, GPR32, "stllrh">;
3119}
3120
3121//===----------------------------------------------------------------------===//
3122// Scaled floating point to integer conversion instructions.
3123//===----------------------------------------------------------------------===//
3124
3125defm FCVTAS : FPToIntegerUnscaled<0b00, 0b100, "fcvtas", int_aarch64_neon_fcvtas>;
3126defm FCVTAU : FPToIntegerUnscaled<0b00, 0b101, "fcvtau", int_aarch64_neon_fcvtau>;
3127defm FCVTMS : FPToIntegerUnscaled<0b10, 0b000, "fcvtms", int_aarch64_neon_fcvtms>;
3128defm FCVTMU : FPToIntegerUnscaled<0b10, 0b001, "fcvtmu", int_aarch64_neon_fcvtmu>;
3129defm FCVTNS : FPToIntegerUnscaled<0b00, 0b000, "fcvtns", int_aarch64_neon_fcvtns>;
3130defm FCVTNU : FPToIntegerUnscaled<0b00, 0b001, "fcvtnu", int_aarch64_neon_fcvtnu>;
3131defm FCVTPS : FPToIntegerUnscaled<0b01, 0b000, "fcvtps", int_aarch64_neon_fcvtps>;
3132defm FCVTPU : FPToIntegerUnscaled<0b01, 0b001, "fcvtpu", int_aarch64_neon_fcvtpu>;
3133defm FCVTZS : FPToIntegerUnscaled<0b11, 0b000, "fcvtzs", fp_to_sint>;
3134defm FCVTZU : FPToIntegerUnscaled<0b11, 0b001, "fcvtzu", fp_to_uint>;
3135defm FCVTZS : FPToIntegerScaled<0b11, 0b000, "fcvtzs", fp_to_sint>;
3136defm FCVTZU : FPToIntegerScaled<0b11, 0b001, "fcvtzu", fp_to_uint>;
3137
3138multiclass FPToIntegerIntPats<Intrinsic round, string INST> {
3139  def : Pat<(i32 (round f16:$Rn)), (!cast<Instruction>(INST # UWHr) $Rn)>;
3140  def : Pat<(i64 (round f16:$Rn)), (!cast<Instruction>(INST # UXHr) $Rn)>;
3141  def : Pat<(i32 (round f32:$Rn)), (!cast<Instruction>(INST # UWSr) $Rn)>;
3142  def : Pat<(i64 (round f32:$Rn)), (!cast<Instruction>(INST # UXSr) $Rn)>;
3143  def : Pat<(i32 (round f64:$Rn)), (!cast<Instruction>(INST # UWDr) $Rn)>;
3144  def : Pat<(i64 (round f64:$Rn)), (!cast<Instruction>(INST # UXDr) $Rn)>;
3145
3146  def : Pat<(i32 (round (fmul f16:$Rn, fixedpoint_f16_i32:$scale))),
3147            (!cast<Instruction>(INST # SWHri) $Rn, $scale)>;
3148  def : Pat<(i64 (round (fmul f16:$Rn, fixedpoint_f16_i64:$scale))),
3149            (!cast<Instruction>(INST # SXHri) $Rn, $scale)>;
3150  def : Pat<(i32 (round (fmul f32:$Rn, fixedpoint_f32_i32:$scale))),
3151            (!cast<Instruction>(INST # SWSri) $Rn, $scale)>;
3152  def : Pat<(i64 (round (fmul f32:$Rn, fixedpoint_f32_i64:$scale))),
3153            (!cast<Instruction>(INST # SXSri) $Rn, $scale)>;
3154  def : Pat<(i32 (round (fmul f64:$Rn, fixedpoint_f64_i32:$scale))),
3155            (!cast<Instruction>(INST # SWDri) $Rn, $scale)>;
3156  def : Pat<(i64 (round (fmul f64:$Rn, fixedpoint_f64_i64:$scale))),
3157            (!cast<Instruction>(INST # SXDri) $Rn, $scale)>;
3158}
3159
3160defm : FPToIntegerIntPats<int_aarch64_neon_fcvtzs, "FCVTZS">;
3161defm : FPToIntegerIntPats<int_aarch64_neon_fcvtzu, "FCVTZU">;
3162
3163multiclass FPToIntegerPats<SDNode to_int, SDNode round, string INST> {
3164  def : Pat<(i32 (to_int (round f32:$Rn))),
3165            (!cast<Instruction>(INST # UWSr) f32:$Rn)>;
3166  def : Pat<(i64 (to_int (round f32:$Rn))),
3167            (!cast<Instruction>(INST # UXSr) f32:$Rn)>;
3168  def : Pat<(i32 (to_int (round f64:$Rn))),
3169            (!cast<Instruction>(INST # UWDr) f64:$Rn)>;
3170  def : Pat<(i64 (to_int (round f64:$Rn))),
3171            (!cast<Instruction>(INST # UXDr) f64:$Rn)>;
3172}
3173
3174defm : FPToIntegerPats<fp_to_sint, fceil,  "FCVTPS">;
3175defm : FPToIntegerPats<fp_to_uint, fceil,  "FCVTPU">;
3176defm : FPToIntegerPats<fp_to_sint, ffloor, "FCVTMS">;
3177defm : FPToIntegerPats<fp_to_uint, ffloor, "FCVTMU">;
3178defm : FPToIntegerPats<fp_to_sint, ftrunc, "FCVTZS">;
3179defm : FPToIntegerPats<fp_to_uint, ftrunc, "FCVTZU">;
3180defm : FPToIntegerPats<fp_to_sint, fround, "FCVTAS">;
3181defm : FPToIntegerPats<fp_to_uint, fround, "FCVTAU">;
3182
3183let Predicates = [HasFullFP16] in {
3184  def : Pat<(i32 (lround f16:$Rn)),
3185            (!cast<Instruction>(FCVTASUWHr) f16:$Rn)>;
3186  def : Pat<(i64 (lround f16:$Rn)),
3187            (!cast<Instruction>(FCVTASUXHr) f16:$Rn)>;
3188  def : Pat<(i64 (llround f16:$Rn)),
3189            (!cast<Instruction>(FCVTASUXHr) f16:$Rn)>;
3190}
3191def : Pat<(i32 (lround f32:$Rn)),
3192          (!cast<Instruction>(FCVTASUWSr) f32:$Rn)>;
3193def : Pat<(i32 (lround f64:$Rn)),
3194          (!cast<Instruction>(FCVTASUWDr) f64:$Rn)>;
3195def : Pat<(i64 (lround f32:$Rn)),
3196          (!cast<Instruction>(FCVTASUXSr) f32:$Rn)>;
3197def : Pat<(i64 (lround f64:$Rn)),
3198          (!cast<Instruction>(FCVTASUXDr) f64:$Rn)>;
3199def : Pat<(i64 (llround f32:$Rn)),
3200          (!cast<Instruction>(FCVTASUXSr) f32:$Rn)>;
3201def : Pat<(i64 (llround f64:$Rn)),
3202          (!cast<Instruction>(FCVTASUXDr) f64:$Rn)>;
3203
3204//===----------------------------------------------------------------------===//
3205// Scaled integer to floating point conversion instructions.
3206//===----------------------------------------------------------------------===//
3207
3208defm SCVTF : IntegerToFP<0, "scvtf", sint_to_fp>;
3209defm UCVTF : IntegerToFP<1, "ucvtf", uint_to_fp>;
3210
3211//===----------------------------------------------------------------------===//
3212// Unscaled integer to floating point conversion instruction.
3213//===----------------------------------------------------------------------===//
3214
3215defm FMOV : UnscaledConversion<"fmov">;
3216
3217// Add pseudo ops for FMOV 0 so we can mark them as isReMaterializable
3218let isReMaterializable = 1, isCodeGenOnly = 1, isAsCheapAsAMove = 1 in {
3219def FMOVH0 : Pseudo<(outs FPR16:$Rd), (ins), [(set f16:$Rd, (fpimm0))]>,
3220    Sched<[WriteF]>, Requires<[HasFullFP16]>;
3221def FMOVS0 : Pseudo<(outs FPR32:$Rd), (ins), [(set f32:$Rd, (fpimm0))]>,
3222    Sched<[WriteF]>;
3223def FMOVD0 : Pseudo<(outs FPR64:$Rd), (ins), [(set f64:$Rd, (fpimm0))]>,
3224    Sched<[WriteF]>;
3225}
3226// Similarly add aliases
3227def : InstAlias<"fmov $Rd, #0.0", (FMOVWHr FPR16:$Rd, WZR), 0>,
3228    Requires<[HasFullFP16]>;
3229def : InstAlias<"fmov $Rd, #0.0", (FMOVWSr FPR32:$Rd, WZR), 0>;
3230def : InstAlias<"fmov $Rd, #0.0", (FMOVXDr FPR64:$Rd, XZR), 0>;
3231
3232//===----------------------------------------------------------------------===//
3233// Floating point conversion instruction.
3234//===----------------------------------------------------------------------===//
3235
3236defm FCVT : FPConversion<"fcvt">;
3237
3238//===----------------------------------------------------------------------===//
3239// Floating point single operand instructions.
3240//===----------------------------------------------------------------------===//
3241
3242defm FABS   : SingleOperandFPData<0b0001, "fabs", fabs>;
3243defm FMOV   : SingleOperandFPData<0b0000, "fmov">;
3244defm FNEG   : SingleOperandFPData<0b0010, "fneg", fneg>;
3245defm FRINTA : SingleOperandFPData<0b1100, "frinta", fround>;
3246defm FRINTI : SingleOperandFPData<0b1111, "frinti", fnearbyint>;
3247defm FRINTM : SingleOperandFPData<0b1010, "frintm", ffloor>;
3248defm FRINTN : SingleOperandFPData<0b1000, "frintn", int_aarch64_neon_frintn>;
3249defm FRINTP : SingleOperandFPData<0b1001, "frintp", fceil>;
3250
3251def : Pat<(v1f64 (int_aarch64_neon_frintn (v1f64 FPR64:$Rn))),
3252          (FRINTNDr FPR64:$Rn)>;
3253
3254defm FRINTX : SingleOperandFPData<0b1110, "frintx", frint>;
3255defm FRINTZ : SingleOperandFPData<0b1011, "frintz", ftrunc>;
3256
3257let SchedRW = [WriteFDiv] in {
3258defm FSQRT  : SingleOperandFPData<0b0011, "fsqrt", fsqrt>;
3259}
3260
3261let Predicates = [HasFRInt3264] in {
3262  defm FRINT32Z : FRIntNNT<0b00, "frint32z">;
3263  defm FRINT64Z : FRIntNNT<0b10, "frint64z">;
3264  defm FRINT32X : FRIntNNT<0b01, "frint32x">;
3265  defm FRINT64X : FRIntNNT<0b11, "frint64x">;
3266} // HasFRInt3264
3267
3268let Predicates = [HasFullFP16] in {
3269  def : Pat<(i32 (lrint f16:$Rn)),
3270            (FCVTZSUWHr (!cast<Instruction>(FRINTXHr) f16:$Rn))>;
3271  def : Pat<(i64 (lrint f16:$Rn)),
3272            (FCVTZSUXHr (!cast<Instruction>(FRINTXHr) f16:$Rn))>;
3273  def : Pat<(i64 (llrint f16:$Rn)),
3274            (FCVTZSUXHr (!cast<Instruction>(FRINTXHr) f16:$Rn))>;
3275}
3276def : Pat<(i32 (lrint f32:$Rn)),
3277          (FCVTZSUWSr (!cast<Instruction>(FRINTXSr) f32:$Rn))>;
3278def : Pat<(i32 (lrint f64:$Rn)),
3279          (FCVTZSUWDr (!cast<Instruction>(FRINTXDr) f64:$Rn))>;
3280def : Pat<(i64 (lrint f32:$Rn)),
3281          (FCVTZSUXSr (!cast<Instruction>(FRINTXSr) f32:$Rn))>;
3282def : Pat<(i64 (lrint f64:$Rn)),
3283          (FCVTZSUXDr (!cast<Instruction>(FRINTXDr) f64:$Rn))>;
3284def : Pat<(i64 (llrint f32:$Rn)),
3285          (FCVTZSUXSr (!cast<Instruction>(FRINTXSr) f32:$Rn))>;
3286def : Pat<(i64 (llrint f64:$Rn)),
3287          (FCVTZSUXDr (!cast<Instruction>(FRINTXDr) f64:$Rn))>;
3288
3289//===----------------------------------------------------------------------===//
3290// Floating point two operand instructions.
3291//===----------------------------------------------------------------------===//
3292
3293defm FADD   : TwoOperandFPData<0b0010, "fadd", fadd>;
3294let SchedRW = [WriteFDiv] in {
3295defm FDIV   : TwoOperandFPData<0b0001, "fdiv", fdiv>;
3296}
3297defm FMAXNM : TwoOperandFPData<0b0110, "fmaxnm", fmaxnum>;
3298defm FMAX   : TwoOperandFPData<0b0100, "fmax", fmaximum>;
3299defm FMINNM : TwoOperandFPData<0b0111, "fminnm", fminnum>;
3300defm FMIN   : TwoOperandFPData<0b0101, "fmin", fminimum>;
3301let SchedRW = [WriteFMul] in {
3302defm FMUL   : TwoOperandFPData<0b0000, "fmul", fmul>;
3303defm FNMUL  : TwoOperandFPDataNeg<0b1000, "fnmul", fmul>;
3304}
3305defm FSUB   : TwoOperandFPData<0b0011, "fsub", fsub>;
3306
3307def : Pat<(v1f64 (fmaximum (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
3308          (FMAXDrr FPR64:$Rn, FPR64:$Rm)>;
3309def : Pat<(v1f64 (fminimum (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
3310          (FMINDrr FPR64:$Rn, FPR64:$Rm)>;
3311def : Pat<(v1f64 (fmaxnum (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
3312          (FMAXNMDrr FPR64:$Rn, FPR64:$Rm)>;
3313def : Pat<(v1f64 (fminnum (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
3314          (FMINNMDrr FPR64:$Rn, FPR64:$Rm)>;
3315
3316//===----------------------------------------------------------------------===//
3317// Floating point three operand instructions.
3318//===----------------------------------------------------------------------===//
3319
3320defm FMADD  : ThreeOperandFPData<0, 0, "fmadd", fma>;
3321defm FMSUB  : ThreeOperandFPData<0, 1, "fmsub",
3322     TriOpFrag<(fma node:$LHS, (fneg node:$MHS), node:$RHS)> >;
3323defm FNMADD : ThreeOperandFPData<1, 0, "fnmadd",
3324     TriOpFrag<(fneg (fma node:$LHS, node:$MHS, node:$RHS))> >;
3325defm FNMSUB : ThreeOperandFPData<1, 1, "fnmsub",
3326     TriOpFrag<(fma node:$LHS, node:$MHS, (fneg node:$RHS))> >;
3327
3328// The following def pats catch the case where the LHS of an FMA is negated.
3329// The TriOpFrag above catches the case where the middle operand is negated.
3330
3331// N.b. FMSUB etc have the accumulator at the *end* of (outs), unlike
3332// the NEON variant.
3333
3334// Here we handle first -(a + b*c) for FNMADD:
3335
3336let Predicates = [HasNEON, HasFullFP16] in
3337def : Pat<(f16 (fma (fneg FPR16:$Rn), FPR16:$Rm, FPR16:$Ra)),
3338          (FMSUBHrrr FPR16:$Rn, FPR16:$Rm, FPR16:$Ra)>;
3339
3340def : Pat<(f32 (fma (fneg FPR32:$Rn), FPR32:$Rm, FPR32:$Ra)),
3341          (FMSUBSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
3342
3343def : Pat<(f64 (fma (fneg FPR64:$Rn), FPR64:$Rm, FPR64:$Ra)),
3344          (FMSUBDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
3345
3346// Now it's time for "(-a) + (-b)*c"
3347
3348let Predicates = [HasNEON, HasFullFP16] in
3349def : Pat<(f16 (fma (fneg FPR16:$Rn), FPR16:$Rm, (fneg FPR16:$Ra))),
3350          (FNMADDHrrr FPR16:$Rn, FPR16:$Rm, FPR16:$Ra)>;
3351
3352def : Pat<(f32 (fma (fneg FPR32:$Rn), FPR32:$Rm, (fneg FPR32:$Ra))),
3353          (FNMADDSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
3354
3355def : Pat<(f64 (fma (fneg FPR64:$Rn), FPR64:$Rm, (fneg FPR64:$Ra))),
3356          (FNMADDDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
3357
3358// And here "(-a) + b*(-c)"
3359
3360let Predicates = [HasNEON, HasFullFP16] in
3361def : Pat<(f16 (fma FPR16:$Rn, (fneg FPR16:$Rm), (fneg FPR16:$Ra))),
3362          (FNMADDHrrr FPR16:$Rn, FPR16:$Rm, FPR16:$Ra)>;
3363
3364def : Pat<(f32 (fma FPR32:$Rn, (fneg FPR32:$Rm), (fneg FPR32:$Ra))),
3365          (FNMADDSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
3366
3367def : Pat<(f64 (fma FPR64:$Rn, (fneg FPR64:$Rm), (fneg FPR64:$Ra))),
3368          (FNMADDDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
3369
3370//===----------------------------------------------------------------------===//
3371// Floating point comparison instructions.
3372//===----------------------------------------------------------------------===//
3373
3374defm FCMPE : FPComparison<1, "fcmpe">;
3375defm FCMP  : FPComparison<0, "fcmp", AArch64fcmp>;
3376
3377//===----------------------------------------------------------------------===//
3378// Floating point conditional comparison instructions.
3379//===----------------------------------------------------------------------===//
3380
3381defm FCCMPE : FPCondComparison<1, "fccmpe">;
3382defm FCCMP  : FPCondComparison<0, "fccmp", AArch64fccmp>;
3383
3384//===----------------------------------------------------------------------===//
3385// Floating point conditional select instruction.
3386//===----------------------------------------------------------------------===//
3387
3388defm FCSEL : FPCondSelect<"fcsel">;
3389
3390// CSEL instructions providing f128 types need to be handled by a
3391// pseudo-instruction since the eventual code will need to introduce basic
3392// blocks and control flow.
3393def F128CSEL : Pseudo<(outs FPR128:$Rd),
3394                      (ins FPR128:$Rn, FPR128:$Rm, ccode:$cond),
3395                      [(set (f128 FPR128:$Rd),
3396                            (AArch64csel FPR128:$Rn, FPR128:$Rm,
3397                                       (i32 imm:$cond), NZCV))]> {
3398  let Uses = [NZCV];
3399  let usesCustomInserter = 1;
3400  let hasNoSchedulingInfo = 1;
3401}
3402
3403//===----------------------------------------------------------------------===//
3404// Instructions used for emitting unwind opcodes on ARM64 Windows.
3405//===----------------------------------------------------------------------===//
3406let isPseudo = 1 in {
3407  def SEH_StackAlloc : Pseudo<(outs), (ins i32imm:$size), []>, Sched<[]>;
3408  def SEH_SaveFPLR : Pseudo<(outs), (ins i32imm:$offs), []>, Sched<[]>;
3409  def SEH_SaveFPLR_X : Pseudo<(outs), (ins i32imm:$offs), []>, Sched<[]>;
3410  def SEH_SaveReg : Pseudo<(outs), (ins i32imm:$reg, i32imm:$offs), []>, Sched<[]>;
3411  def SEH_SaveReg_X : Pseudo<(outs), (ins i32imm:$reg, i32imm:$offs), []>, Sched<[]>;
3412  def SEH_SaveRegP : Pseudo<(outs), (ins i32imm:$reg0, i32imm:$reg1, i32imm:$offs), []>, Sched<[]>;
3413  def SEH_SaveRegP_X : Pseudo<(outs), (ins i32imm:$reg0, i32imm:$reg1, i32imm:$offs), []>, Sched<[]>;
3414  def SEH_SaveFReg : Pseudo<(outs), (ins i32imm:$reg, i32imm:$offs), []>, Sched<[]>;
3415  def SEH_SaveFReg_X :  Pseudo<(outs), (ins i32imm:$reg, i32imm:$offs), []>, Sched<[]>;
3416  def SEH_SaveFRegP : Pseudo<(outs), (ins i32imm:$reg0, i32imm:$reg1, i32imm:$offs), []>, Sched<[]>;
3417  def SEH_SaveFRegP_X : Pseudo<(outs), (ins i32imm:$reg0, i32imm:$reg1, i32imm:$offs), []>, Sched<[]>;
3418  def SEH_SetFP : Pseudo<(outs), (ins), []>, Sched<[]>;
3419  def SEH_AddFP : Pseudo<(outs), (ins i32imm:$offs), []>, Sched<[]>;
3420  def SEH_Nop : Pseudo<(outs), (ins), []>, Sched<[]>;
3421  def SEH_PrologEnd : Pseudo<(outs), (ins), []>, Sched<[]>;
3422  def SEH_EpilogStart : Pseudo<(outs), (ins), []>, Sched<[]>;
3423  def SEH_EpilogEnd : Pseudo<(outs), (ins), []>, Sched<[]>;
3424}
3425
3426// Pseudo instructions for Windows EH
3427//===----------------------------------------------------------------------===//
3428let isTerminator = 1, hasSideEffects = 1, isBarrier = 1, hasCtrlDep = 1,
3429    isCodeGenOnly = 1, isReturn = 1, isEHScopeReturn = 1, isPseudo = 1 in {
3430   def CLEANUPRET : Pseudo<(outs), (ins), [(cleanupret)]>, Sched<[]>;
3431   let usesCustomInserter = 1 in
3432     def CATCHRET : Pseudo<(outs), (ins am_brcond:$dst, am_brcond:$src), [(catchret bb:$dst, bb:$src)]>,
3433                    Sched<[]>;
3434}
3435
3436let hasSideEffects = 1, hasCtrlDep = 1, isCodeGenOnly = 1,
3437    usesCustomInserter = 1 in
3438def CATCHPAD : Pseudo<(outs), (ins), [(catchpad)]>, Sched<[]>;
3439
3440//===----------------------------------------------------------------------===//
3441// Floating point immediate move.
3442//===----------------------------------------------------------------------===//
3443
3444let isReMaterializable = 1 in {
3445defm FMOV : FPMoveImmediate<"fmov">;
3446}
3447
3448//===----------------------------------------------------------------------===//
3449// Advanced SIMD two vector instructions.
3450//===----------------------------------------------------------------------===//
3451
3452defm UABDL   : SIMDLongThreeVectorBHSabdl<1, 0b0111, "uabdl",
3453                                          int_aarch64_neon_uabd>;
3454// Match UABDL in log2-shuffle patterns.
3455def : Pat<(abs (v8i16 (sub (zext (v8i8 V64:$opA)),
3456                           (zext (v8i8 V64:$opB))))),
3457          (UABDLv8i8_v8i16 V64:$opA, V64:$opB)>;
3458def : Pat<(xor (v8i16 (AArch64vashr v8i16:$src, (i32 15))),
3459               (v8i16 (add (sub (zext (v8i8 V64:$opA)),
3460                                (zext (v8i8 V64:$opB))),
3461                           (AArch64vashr v8i16:$src, (i32 15))))),
3462          (UABDLv8i8_v8i16 V64:$opA, V64:$opB)>;
3463def : Pat<(abs (v8i16 (sub (zext (extract_high_v16i8 V128:$opA)),
3464                           (zext (extract_high_v16i8 V128:$opB))))),
3465          (UABDLv16i8_v8i16 V128:$opA, V128:$opB)>;
3466def : Pat<(xor (v8i16 (AArch64vashr v8i16:$src, (i32 15))),
3467               (v8i16 (add (sub (zext (extract_high_v16i8 V128:$opA)),
3468                                (zext (extract_high_v16i8 V128:$opB))),
3469                           (AArch64vashr v8i16:$src, (i32 15))))),
3470          (UABDLv16i8_v8i16 V128:$opA, V128:$opB)>;
3471def : Pat<(abs (v4i32 (sub (zext (v4i16 V64:$opA)),
3472                           (zext (v4i16 V64:$opB))))),
3473          (UABDLv4i16_v4i32 V64:$opA, V64:$opB)>;
3474def : Pat<(abs (v4i32 (sub (zext (extract_high_v8i16 V128:$opA)),
3475                           (zext (extract_high_v8i16 V128:$opB))))),
3476          (UABDLv8i16_v4i32 V128:$opA, V128:$opB)>;
3477def : Pat<(abs (v2i64 (sub (zext (v2i32 V64:$opA)),
3478                           (zext (v2i32 V64:$opB))))),
3479          (UABDLv2i32_v2i64 V64:$opA, V64:$opB)>;
3480def : Pat<(abs (v2i64 (sub (zext (extract_high_v4i32 V128:$opA)),
3481                           (zext (extract_high_v4i32 V128:$opB))))),
3482          (UABDLv4i32_v2i64 V128:$opA, V128:$opB)>;
3483
3484defm ABS    : SIMDTwoVectorBHSD<0, 0b01011, "abs", abs>;
3485defm CLS    : SIMDTwoVectorBHS<0, 0b00100, "cls", int_aarch64_neon_cls>;
3486defm CLZ    : SIMDTwoVectorBHS<1, 0b00100, "clz", ctlz>;
3487defm CMEQ   : SIMDCmpTwoVector<0, 0b01001, "cmeq", AArch64cmeqz>;
3488defm CMGE   : SIMDCmpTwoVector<1, 0b01000, "cmge", AArch64cmgez>;
3489defm CMGT   : SIMDCmpTwoVector<0, 0b01000, "cmgt", AArch64cmgtz>;
3490defm CMLE   : SIMDCmpTwoVector<1, 0b01001, "cmle", AArch64cmlez>;
3491defm CMLT   : SIMDCmpTwoVector<0, 0b01010, "cmlt", AArch64cmltz>;
3492defm CNT    : SIMDTwoVectorB<0, 0b00, 0b00101, "cnt", ctpop>;
3493defm FABS   : SIMDTwoVectorFP<0, 1, 0b01111, "fabs", fabs>;
3494
3495defm FCMEQ  : SIMDFPCmpTwoVector<0, 1, 0b01101, "fcmeq", AArch64fcmeqz>;
3496defm FCMGE  : SIMDFPCmpTwoVector<1, 1, 0b01100, "fcmge", AArch64fcmgez>;
3497defm FCMGT  : SIMDFPCmpTwoVector<0, 1, 0b01100, "fcmgt", AArch64fcmgtz>;
3498defm FCMLE  : SIMDFPCmpTwoVector<1, 1, 0b01101, "fcmle", AArch64fcmlez>;
3499defm FCMLT  : SIMDFPCmpTwoVector<0, 1, 0b01110, "fcmlt", AArch64fcmltz>;
3500defm FCVTAS : SIMDTwoVectorFPToInt<0,0,0b11100, "fcvtas",int_aarch64_neon_fcvtas>;
3501defm FCVTAU : SIMDTwoVectorFPToInt<1,0,0b11100, "fcvtau",int_aarch64_neon_fcvtau>;
3502defm FCVTL  : SIMDFPWidenTwoVector<0, 0, 0b10111, "fcvtl">;
3503def : Pat<(v4f32 (int_aarch64_neon_vcvthf2fp (v4i16 V64:$Rn))),
3504          (FCVTLv4i16 V64:$Rn)>;
3505def : Pat<(v4f32 (int_aarch64_neon_vcvthf2fp (extract_subvector (v8i16 V128:$Rn),
3506                                                              (i64 4)))),
3507          (FCVTLv8i16 V128:$Rn)>;
3508def : Pat<(v2f64 (fpextend (v2f32 V64:$Rn))), (FCVTLv2i32 V64:$Rn)>;
3509def : Pat<(v2f64 (fpextend (v2f32 (extract_subvector (v4f32 V128:$Rn),
3510                                                    (i64 2))))),
3511          (FCVTLv4i32 V128:$Rn)>;
3512
3513def : Pat<(v4f32 (fpextend (v4f16 V64:$Rn))), (FCVTLv4i16 V64:$Rn)>;
3514def : Pat<(v4f32 (fpextend (v4f16 (extract_subvector (v8f16 V128:$Rn),
3515                                                    (i64 4))))),
3516          (FCVTLv8i16 V128:$Rn)>;
3517
3518defm FCVTMS : SIMDTwoVectorFPToInt<0,0,0b11011, "fcvtms",int_aarch64_neon_fcvtms>;
3519defm FCVTMU : SIMDTwoVectorFPToInt<1,0,0b11011, "fcvtmu",int_aarch64_neon_fcvtmu>;
3520defm FCVTNS : SIMDTwoVectorFPToInt<0,0,0b11010, "fcvtns",int_aarch64_neon_fcvtns>;
3521defm FCVTNU : SIMDTwoVectorFPToInt<1,0,0b11010, "fcvtnu",int_aarch64_neon_fcvtnu>;
3522defm FCVTN  : SIMDFPNarrowTwoVector<0, 0, 0b10110, "fcvtn">;
3523def : Pat<(v4i16 (int_aarch64_neon_vcvtfp2hf (v4f32 V128:$Rn))),
3524          (FCVTNv4i16 V128:$Rn)>;
3525def : Pat<(concat_vectors V64:$Rd,
3526                          (v4i16 (int_aarch64_neon_vcvtfp2hf (v4f32 V128:$Rn)))),
3527          (FCVTNv8i16 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
3528def : Pat<(v2f32 (fpround (v2f64 V128:$Rn))), (FCVTNv2i32 V128:$Rn)>;
3529def : Pat<(v4f16 (fpround (v4f32 V128:$Rn))), (FCVTNv4i16 V128:$Rn)>;
3530def : Pat<(concat_vectors V64:$Rd, (v2f32 (fpround (v2f64 V128:$Rn)))),
3531          (FCVTNv4i32 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
3532defm FCVTPS : SIMDTwoVectorFPToInt<0,1,0b11010, "fcvtps",int_aarch64_neon_fcvtps>;
3533defm FCVTPU : SIMDTwoVectorFPToInt<1,1,0b11010, "fcvtpu",int_aarch64_neon_fcvtpu>;
3534defm FCVTXN : SIMDFPInexactCvtTwoVector<1, 0, 0b10110, "fcvtxn",
3535                                        int_aarch64_neon_fcvtxn>;
3536defm FCVTZS : SIMDTwoVectorFPToInt<0, 1, 0b11011, "fcvtzs", fp_to_sint>;
3537defm FCVTZU : SIMDTwoVectorFPToInt<1, 1, 0b11011, "fcvtzu", fp_to_uint>;
3538
3539def : Pat<(v4i16 (int_aarch64_neon_fcvtzs v4f16:$Rn)), (FCVTZSv4f16 $Rn)>;
3540def : Pat<(v8i16 (int_aarch64_neon_fcvtzs v8f16:$Rn)), (FCVTZSv8f16 $Rn)>;
3541def : Pat<(v2i32 (int_aarch64_neon_fcvtzs v2f32:$Rn)), (FCVTZSv2f32 $Rn)>;
3542def : Pat<(v4i32 (int_aarch64_neon_fcvtzs v4f32:$Rn)), (FCVTZSv4f32 $Rn)>;
3543def : Pat<(v2i64 (int_aarch64_neon_fcvtzs v2f64:$Rn)), (FCVTZSv2f64 $Rn)>;
3544
3545def : Pat<(v4i16 (int_aarch64_neon_fcvtzu v4f16:$Rn)), (FCVTZUv4f16 $Rn)>;
3546def : Pat<(v8i16 (int_aarch64_neon_fcvtzu v8f16:$Rn)), (FCVTZUv8f16 $Rn)>;
3547def : Pat<(v2i32 (int_aarch64_neon_fcvtzu v2f32:$Rn)), (FCVTZUv2f32 $Rn)>;
3548def : Pat<(v4i32 (int_aarch64_neon_fcvtzu v4f32:$Rn)), (FCVTZUv4f32 $Rn)>;
3549def : Pat<(v2i64 (int_aarch64_neon_fcvtzu v2f64:$Rn)), (FCVTZUv2f64 $Rn)>;
3550
3551defm FNEG   : SIMDTwoVectorFP<1, 1, 0b01111, "fneg", fneg>;
3552defm FRECPE : SIMDTwoVectorFP<0, 1, 0b11101, "frecpe", int_aarch64_neon_frecpe>;
3553defm FRINTA : SIMDTwoVectorFP<1, 0, 0b11000, "frinta", fround>;
3554defm FRINTI : SIMDTwoVectorFP<1, 1, 0b11001, "frinti", fnearbyint>;
3555defm FRINTM : SIMDTwoVectorFP<0, 0, 0b11001, "frintm", ffloor>;
3556defm FRINTN : SIMDTwoVectorFP<0, 0, 0b11000, "frintn", int_aarch64_neon_frintn>;
3557defm FRINTP : SIMDTwoVectorFP<0, 1, 0b11000, "frintp", fceil>;
3558defm FRINTX : SIMDTwoVectorFP<1, 0, 0b11001, "frintx", frint>;
3559defm FRINTZ : SIMDTwoVectorFP<0, 1, 0b11001, "frintz", ftrunc>;
3560
3561let Predicates = [HasFRInt3264] in {
3562  defm FRINT32Z : FRIntNNTVector<0, 0, "frint32z">;
3563  defm FRINT64Z : FRIntNNTVector<0, 1, "frint64z">;
3564  defm FRINT32X : FRIntNNTVector<1, 0, "frint32x">;
3565  defm FRINT64X : FRIntNNTVector<1, 1, "frint64x">;
3566} // HasFRInt3264
3567
3568defm FRSQRTE: SIMDTwoVectorFP<1, 1, 0b11101, "frsqrte", int_aarch64_neon_frsqrte>;
3569defm FSQRT  : SIMDTwoVectorFP<1, 1, 0b11111, "fsqrt", fsqrt>;
3570defm NEG    : SIMDTwoVectorBHSD<1, 0b01011, "neg",
3571                               UnOpFrag<(sub immAllZerosV, node:$LHS)> >;
3572defm NOT    : SIMDTwoVectorB<1, 0b00, 0b00101, "not", vnot>;
3573// Aliases for MVN -> NOT.
3574def : InstAlias<"mvn{ $Vd.8b, $Vn.8b|.8b $Vd, $Vn}",
3575                (NOTv8i8 V64:$Vd, V64:$Vn)>;
3576def : InstAlias<"mvn{ $Vd.16b, $Vn.16b|.16b $Vd, $Vn}",
3577                (NOTv16i8 V128:$Vd, V128:$Vn)>;
3578
3579def : Pat<(AArch64neg (v8i8  V64:$Rn)),  (NEGv8i8  V64:$Rn)>;
3580def : Pat<(AArch64neg (v16i8 V128:$Rn)), (NEGv16i8 V128:$Rn)>;
3581def : Pat<(AArch64neg (v4i16 V64:$Rn)),  (NEGv4i16 V64:$Rn)>;
3582def : Pat<(AArch64neg (v8i16 V128:$Rn)), (NEGv8i16 V128:$Rn)>;
3583def : Pat<(AArch64neg (v2i32 V64:$Rn)),  (NEGv2i32 V64:$Rn)>;
3584def : Pat<(AArch64neg (v4i32 V128:$Rn)), (NEGv4i32 V128:$Rn)>;
3585def : Pat<(AArch64neg (v2i64 V128:$Rn)), (NEGv2i64 V128:$Rn)>;
3586
3587def : Pat<(AArch64not (v8i8 V64:$Rn)),   (NOTv8i8  V64:$Rn)>;
3588def : Pat<(AArch64not (v16i8 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
3589def : Pat<(AArch64not (v4i16 V64:$Rn)),  (NOTv8i8  V64:$Rn)>;
3590def : Pat<(AArch64not (v8i16 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
3591def : Pat<(AArch64not (v2i32 V64:$Rn)),  (NOTv8i8  V64:$Rn)>;
3592def : Pat<(AArch64not (v1i64 V64:$Rn)),  (NOTv8i8  V64:$Rn)>;
3593def : Pat<(AArch64not (v4i32 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
3594def : Pat<(AArch64not (v2i64 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
3595
3596def : Pat<(vnot (v4i16 V64:$Rn)),  (NOTv8i8  V64:$Rn)>;
3597def : Pat<(vnot (v8i16 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
3598def : Pat<(vnot (v2i32 V64:$Rn)),  (NOTv8i8  V64:$Rn)>;
3599def : Pat<(vnot (v4i32 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
3600def : Pat<(vnot (v2i64 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
3601
3602defm RBIT   : SIMDTwoVectorB<1, 0b01, 0b00101, "rbit", int_aarch64_neon_rbit>;
3603defm REV16  : SIMDTwoVectorB<0, 0b00, 0b00001, "rev16", AArch64rev16>;
3604defm REV32  : SIMDTwoVectorBH<1, 0b00000, "rev32", AArch64rev32>;
3605defm REV64  : SIMDTwoVectorBHS<0, 0b00000, "rev64", AArch64rev64>;
3606defm SADALP : SIMDLongTwoVectorTied<0, 0b00110, "sadalp",
3607       BinOpFrag<(add node:$LHS, (int_aarch64_neon_saddlp node:$RHS))> >;
3608defm SADDLP : SIMDLongTwoVector<0, 0b00010, "saddlp", int_aarch64_neon_saddlp>;
3609defm SCVTF  : SIMDTwoVectorIntToFP<0, 0, 0b11101, "scvtf", sint_to_fp>;
3610defm SHLL   : SIMDVectorLShiftLongBySizeBHS;
3611defm SQABS  : SIMDTwoVectorBHSD<0, 0b00111, "sqabs", int_aarch64_neon_sqabs>;
3612defm SQNEG  : SIMDTwoVectorBHSD<1, 0b00111, "sqneg", int_aarch64_neon_sqneg>;
3613defm SQXTN  : SIMDMixedTwoVector<0, 0b10100, "sqxtn", int_aarch64_neon_sqxtn>;
3614defm SQXTUN : SIMDMixedTwoVector<1, 0b10010, "sqxtun", int_aarch64_neon_sqxtun>;
3615defm SUQADD : SIMDTwoVectorBHSDTied<0, 0b00011, "suqadd",int_aarch64_neon_suqadd>;
3616defm UADALP : SIMDLongTwoVectorTied<1, 0b00110, "uadalp",
3617       BinOpFrag<(add node:$LHS, (int_aarch64_neon_uaddlp node:$RHS))> >;
3618defm UADDLP : SIMDLongTwoVector<1, 0b00010, "uaddlp",
3619                    int_aarch64_neon_uaddlp>;
3620defm UCVTF  : SIMDTwoVectorIntToFP<1, 0, 0b11101, "ucvtf", uint_to_fp>;
3621defm UQXTN  : SIMDMixedTwoVector<1, 0b10100, "uqxtn", int_aarch64_neon_uqxtn>;
3622defm URECPE : SIMDTwoVectorS<0, 1, 0b11100, "urecpe", int_aarch64_neon_urecpe>;
3623defm URSQRTE: SIMDTwoVectorS<1, 1, 0b11100, "ursqrte", int_aarch64_neon_ursqrte>;
3624defm USQADD : SIMDTwoVectorBHSDTied<1, 0b00011, "usqadd",int_aarch64_neon_usqadd>;
3625defm XTN    : SIMDMixedTwoVector<0, 0b10010, "xtn", trunc>;
3626
3627def : Pat<(v4f16 (AArch64rev32 V64:$Rn)), (REV32v4i16 V64:$Rn)>;
3628def : Pat<(v4f16 (AArch64rev64 V64:$Rn)), (REV64v4i16 V64:$Rn)>;
3629def : Pat<(v8f16 (AArch64rev32 V128:$Rn)), (REV32v8i16 V128:$Rn)>;
3630def : Pat<(v8f16 (AArch64rev64 V128:$Rn)), (REV64v8i16 V128:$Rn)>;
3631def : Pat<(v2f32 (AArch64rev64 V64:$Rn)), (REV64v2i32 V64:$Rn)>;
3632def : Pat<(v4f32 (AArch64rev64 V128:$Rn)), (REV64v4i32 V128:$Rn)>;
3633
3634// Patterns for vector long shift (by element width). These need to match all
3635// three of zext, sext and anyext so it's easier to pull the patterns out of the
3636// definition.
3637multiclass SIMDVectorLShiftLongBySizeBHSPats<SDPatternOperator ext> {
3638  def : Pat<(AArch64vshl (v8i16 (ext (v8i8 V64:$Rn))), (i32 8)),
3639            (SHLLv8i8 V64:$Rn)>;
3640  def : Pat<(AArch64vshl (v8i16 (ext (extract_high_v16i8 V128:$Rn))), (i32 8)),
3641            (SHLLv16i8 V128:$Rn)>;
3642  def : Pat<(AArch64vshl (v4i32 (ext (v4i16 V64:$Rn))), (i32 16)),
3643            (SHLLv4i16 V64:$Rn)>;
3644  def : Pat<(AArch64vshl (v4i32 (ext (extract_high_v8i16 V128:$Rn))), (i32 16)),
3645            (SHLLv8i16 V128:$Rn)>;
3646  def : Pat<(AArch64vshl (v2i64 (ext (v2i32 V64:$Rn))), (i32 32)),
3647            (SHLLv2i32 V64:$Rn)>;
3648  def : Pat<(AArch64vshl (v2i64 (ext (extract_high_v4i32 V128:$Rn))), (i32 32)),
3649            (SHLLv4i32 V128:$Rn)>;
3650}
3651
3652defm : SIMDVectorLShiftLongBySizeBHSPats<anyext>;
3653defm : SIMDVectorLShiftLongBySizeBHSPats<zext>;
3654defm : SIMDVectorLShiftLongBySizeBHSPats<sext>;
3655
3656//===----------------------------------------------------------------------===//
3657// Advanced SIMD three vector instructions.
3658//===----------------------------------------------------------------------===//
3659
3660defm ADD     : SIMDThreeSameVector<0, 0b10000, "add", add>;
3661defm ADDP    : SIMDThreeSameVector<0, 0b10111, "addp", int_aarch64_neon_addp>;
3662defm CMEQ    : SIMDThreeSameVector<1, 0b10001, "cmeq", AArch64cmeq>;
3663defm CMGE    : SIMDThreeSameVector<0, 0b00111, "cmge", AArch64cmge>;
3664defm CMGT    : SIMDThreeSameVector<0, 0b00110, "cmgt", AArch64cmgt>;
3665defm CMHI    : SIMDThreeSameVector<1, 0b00110, "cmhi", AArch64cmhi>;
3666defm CMHS    : SIMDThreeSameVector<1, 0b00111, "cmhs", AArch64cmhs>;
3667defm CMTST   : SIMDThreeSameVector<0, 0b10001, "cmtst", AArch64cmtst>;
3668defm FABD    : SIMDThreeSameVectorFP<1,1,0b010,"fabd", int_aarch64_neon_fabd>;
3669let Predicates = [HasNEON] in {
3670foreach VT = [ v2f32, v4f32, v2f64 ] in
3671def : Pat<(fabs (fsub VT:$Rn, VT:$Rm)), (!cast<Instruction>("FABD"#VT) VT:$Rn, VT:$Rm)>;
3672}
3673let Predicates = [HasNEON, HasFullFP16] in {
3674foreach VT = [ v4f16, v8f16 ] in
3675def : Pat<(fabs (fsub VT:$Rn, VT:$Rm)), (!cast<Instruction>("FABD"#VT) VT:$Rn, VT:$Rm)>;
3676}
3677defm FACGE   : SIMDThreeSameVectorFPCmp<1,0,0b101,"facge",int_aarch64_neon_facge>;
3678defm FACGT   : SIMDThreeSameVectorFPCmp<1,1,0b101,"facgt",int_aarch64_neon_facgt>;
3679defm FADDP   : SIMDThreeSameVectorFP<1,0,0b010,"faddp",int_aarch64_neon_faddp>;
3680defm FADD    : SIMDThreeSameVectorFP<0,0,0b010,"fadd", fadd>;
3681defm FCMEQ   : SIMDThreeSameVectorFPCmp<0, 0, 0b100, "fcmeq", AArch64fcmeq>;
3682defm FCMGE   : SIMDThreeSameVectorFPCmp<1, 0, 0b100, "fcmge", AArch64fcmge>;
3683defm FCMGT   : SIMDThreeSameVectorFPCmp<1, 1, 0b100, "fcmgt", AArch64fcmgt>;
3684defm FDIV    : SIMDThreeSameVectorFP<1,0,0b111,"fdiv", fdiv>;
3685defm FMAXNMP : SIMDThreeSameVectorFP<1,0,0b000,"fmaxnmp", int_aarch64_neon_fmaxnmp>;
3686defm FMAXNM  : SIMDThreeSameVectorFP<0,0,0b000,"fmaxnm", fmaxnum>;
3687defm FMAXP   : SIMDThreeSameVectorFP<1,0,0b110,"fmaxp", int_aarch64_neon_fmaxp>;
3688defm FMAX    : SIMDThreeSameVectorFP<0,0,0b110,"fmax", fmaximum>;
3689defm FMINNMP : SIMDThreeSameVectorFP<1,1,0b000,"fminnmp", int_aarch64_neon_fminnmp>;
3690defm FMINNM  : SIMDThreeSameVectorFP<0,1,0b000,"fminnm", fminnum>;
3691defm FMINP   : SIMDThreeSameVectorFP<1,1,0b110,"fminp", int_aarch64_neon_fminp>;
3692defm FMIN    : SIMDThreeSameVectorFP<0,1,0b110,"fmin", fminimum>;
3693
3694// NOTE: The operands of the PatFrag are reordered on FMLA/FMLS because the
3695// instruction expects the addend first, while the fma intrinsic puts it last.
3696defm FMLA     : SIMDThreeSameVectorFPTied<0, 0, 0b001, "fmla",
3697            TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)> >;
3698defm FMLS     : SIMDThreeSameVectorFPTied<0, 1, 0b001, "fmls",
3699            TriOpFrag<(fma node:$MHS, (fneg node:$RHS), node:$LHS)> >;
3700
3701// The following def pats catch the case where the LHS of an FMA is negated.
3702// The TriOpFrag above catches the case where the middle operand is negated.
3703def : Pat<(v2f32 (fma (fneg V64:$Rn), V64:$Rm, V64:$Rd)),
3704          (FMLSv2f32 V64:$Rd, V64:$Rn, V64:$Rm)>;
3705
3706def : Pat<(v4f32 (fma (fneg V128:$Rn), V128:$Rm, V128:$Rd)),
3707          (FMLSv4f32 V128:$Rd, V128:$Rn, V128:$Rm)>;
3708
3709def : Pat<(v2f64 (fma (fneg V128:$Rn), V128:$Rm, V128:$Rd)),
3710          (FMLSv2f64 V128:$Rd, V128:$Rn, V128:$Rm)>;
3711
3712defm FMULX    : SIMDThreeSameVectorFP<0,0,0b011,"fmulx", int_aarch64_neon_fmulx>;
3713defm FMUL     : SIMDThreeSameVectorFP<1,0,0b011,"fmul", fmul>;
3714defm FRECPS   : SIMDThreeSameVectorFP<0,0,0b111,"frecps", int_aarch64_neon_frecps>;
3715defm FRSQRTS  : SIMDThreeSameVectorFP<0,1,0b111,"frsqrts", int_aarch64_neon_frsqrts>;
3716defm FSUB     : SIMDThreeSameVectorFP<0,1,0b010,"fsub", fsub>;
3717defm MLA      : SIMDThreeSameVectorBHSTied<0, 0b10010, "mla",
3718                      TriOpFrag<(add node:$LHS, (mul node:$MHS, node:$RHS))> >;
3719defm MLS      : SIMDThreeSameVectorBHSTied<1, 0b10010, "mls",
3720                      TriOpFrag<(sub node:$LHS, (mul node:$MHS, node:$RHS))> >;
3721defm MUL      : SIMDThreeSameVectorBHS<0, 0b10011, "mul", mul>;
3722defm PMUL     : SIMDThreeSameVectorB<1, 0b10011, "pmul", int_aarch64_neon_pmul>;
3723defm SABA     : SIMDThreeSameVectorBHSTied<0, 0b01111, "saba",
3724      TriOpFrag<(add node:$LHS, (int_aarch64_neon_sabd node:$MHS, node:$RHS))> >;
3725defm SABD     : SIMDThreeSameVectorBHS<0,0b01110,"sabd", int_aarch64_neon_sabd>;
3726defm SHADD    : SIMDThreeSameVectorBHS<0,0b00000,"shadd", int_aarch64_neon_shadd>;
3727defm SHSUB    : SIMDThreeSameVectorBHS<0,0b00100,"shsub", int_aarch64_neon_shsub>;
3728defm SMAXP    : SIMDThreeSameVectorBHS<0,0b10100,"smaxp", int_aarch64_neon_smaxp>;
3729defm SMAX     : SIMDThreeSameVectorBHS<0,0b01100,"smax", smax>;
3730defm SMINP    : SIMDThreeSameVectorBHS<0,0b10101,"sminp", int_aarch64_neon_sminp>;
3731defm SMIN     : SIMDThreeSameVectorBHS<0,0b01101,"smin", smin>;
3732defm SQADD    : SIMDThreeSameVector<0,0b00001,"sqadd", int_aarch64_neon_sqadd>;
3733defm SQDMULH  : SIMDThreeSameVectorHS<0,0b10110,"sqdmulh",int_aarch64_neon_sqdmulh>;
3734defm SQRDMULH : SIMDThreeSameVectorHS<1,0b10110,"sqrdmulh",int_aarch64_neon_sqrdmulh>;
3735defm SQRSHL   : SIMDThreeSameVector<0,0b01011,"sqrshl", int_aarch64_neon_sqrshl>;
3736defm SQSHL    : SIMDThreeSameVector<0,0b01001,"sqshl", int_aarch64_neon_sqshl>;
3737defm SQSUB    : SIMDThreeSameVector<0,0b00101,"sqsub", int_aarch64_neon_sqsub>;
3738defm SRHADD   : SIMDThreeSameVectorBHS<0,0b00010,"srhadd",int_aarch64_neon_srhadd>;
3739defm SRSHL    : SIMDThreeSameVector<0,0b01010,"srshl", int_aarch64_neon_srshl>;
3740defm SSHL     : SIMDThreeSameVector<0,0b01000,"sshl", int_aarch64_neon_sshl>;
3741defm SUB      : SIMDThreeSameVector<1,0b10000,"sub", sub>;
3742defm UABA     : SIMDThreeSameVectorBHSTied<1, 0b01111, "uaba",
3743      TriOpFrag<(add node:$LHS, (int_aarch64_neon_uabd node:$MHS, node:$RHS))> >;
3744defm UABD     : SIMDThreeSameVectorBHS<1,0b01110,"uabd", int_aarch64_neon_uabd>;
3745defm UHADD    : SIMDThreeSameVectorBHS<1,0b00000,"uhadd", int_aarch64_neon_uhadd>;
3746defm UHSUB    : SIMDThreeSameVectorBHS<1,0b00100,"uhsub", int_aarch64_neon_uhsub>;
3747defm UMAXP    : SIMDThreeSameVectorBHS<1,0b10100,"umaxp", int_aarch64_neon_umaxp>;
3748defm UMAX     : SIMDThreeSameVectorBHS<1,0b01100,"umax", umax>;
3749defm UMINP    : SIMDThreeSameVectorBHS<1,0b10101,"uminp", int_aarch64_neon_uminp>;
3750defm UMIN     : SIMDThreeSameVectorBHS<1,0b01101,"umin", umin>;
3751defm UQADD    : SIMDThreeSameVector<1,0b00001,"uqadd", int_aarch64_neon_uqadd>;
3752defm UQRSHL   : SIMDThreeSameVector<1,0b01011,"uqrshl", int_aarch64_neon_uqrshl>;
3753defm UQSHL    : SIMDThreeSameVector<1,0b01001,"uqshl", int_aarch64_neon_uqshl>;
3754defm UQSUB    : SIMDThreeSameVector<1,0b00101,"uqsub", int_aarch64_neon_uqsub>;
3755defm URHADD   : SIMDThreeSameVectorBHS<1,0b00010,"urhadd", int_aarch64_neon_urhadd>;
3756defm URSHL    : SIMDThreeSameVector<1,0b01010,"urshl", int_aarch64_neon_urshl>;
3757defm USHL     : SIMDThreeSameVector<1,0b01000,"ushl", int_aarch64_neon_ushl>;
3758defm SQRDMLAH : SIMDThreeSameVectorSQRDMLxHTiedHS<1,0b10000,"sqrdmlah",
3759                                                  int_aarch64_neon_sqadd>;
3760defm SQRDMLSH : SIMDThreeSameVectorSQRDMLxHTiedHS<1,0b10001,"sqrdmlsh",
3761                                                    int_aarch64_neon_sqsub>;
3762
3763defm AND : SIMDLogicalThreeVector<0, 0b00, "and", and>;
3764defm BIC : SIMDLogicalThreeVector<0, 0b01, "bic",
3765                                  BinOpFrag<(and node:$LHS, (vnot node:$RHS))> >;
3766defm BIF : SIMDLogicalThreeVector<1, 0b11, "bif">;
3767defm BIT : SIMDLogicalThreeVectorTied<1, 0b10, "bit", AArch64bit>;
3768defm BSL : SIMDLogicalThreeVectorTied<1, 0b01, "bsl",
3769    TriOpFrag<(or (and node:$LHS, node:$MHS), (and (vnot node:$LHS), node:$RHS))>>;
3770defm EOR : SIMDLogicalThreeVector<1, 0b00, "eor", xor>;
3771defm ORN : SIMDLogicalThreeVector<0, 0b11, "orn",
3772                                  BinOpFrag<(or node:$LHS, (vnot node:$RHS))> >;
3773defm ORR : SIMDLogicalThreeVector<0, 0b10, "orr", or>;
3774
3775
3776def : Pat<(AArch64bsl (v8i8 V64:$Rd), V64:$Rn, V64:$Rm),
3777          (BSLv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
3778def : Pat<(AArch64bsl (v4i16 V64:$Rd), V64:$Rn, V64:$Rm),
3779          (BSLv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
3780def : Pat<(AArch64bsl (v2i32 V64:$Rd), V64:$Rn, V64:$Rm),
3781          (BSLv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
3782def : Pat<(AArch64bsl (v1i64 V64:$Rd), V64:$Rn, V64:$Rm),
3783          (BSLv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
3784
3785def : Pat<(AArch64bsl (v16i8 V128:$Rd), V128:$Rn, V128:$Rm),
3786          (BSLv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
3787def : Pat<(AArch64bsl (v8i16 V128:$Rd), V128:$Rn, V128:$Rm),
3788          (BSLv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
3789def : Pat<(AArch64bsl (v4i32 V128:$Rd), V128:$Rn, V128:$Rm),
3790          (BSLv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
3791def : Pat<(AArch64bsl (v2i64 V128:$Rd), V128:$Rn, V128:$Rm),
3792          (BSLv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
3793
3794def : InstAlias<"mov{\t$dst.16b, $src.16b|.16b\t$dst, $src}",
3795                (ORRv16i8 V128:$dst, V128:$src, V128:$src), 1>;
3796def : InstAlias<"mov{\t$dst.8h, $src.8h|.8h\t$dst, $src}",
3797                (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
3798def : InstAlias<"mov{\t$dst.4s, $src.4s|.4s\t$dst, $src}",
3799                (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
3800def : InstAlias<"mov{\t$dst.2d, $src.2d|.2d\t$dst, $src}",
3801                (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
3802
3803def : InstAlias<"mov{\t$dst.8b, $src.8b|.8b\t$dst, $src}",
3804                (ORRv8i8 V64:$dst, V64:$src, V64:$src), 1>;
3805def : InstAlias<"mov{\t$dst.4h, $src.4h|.4h\t$dst, $src}",
3806                (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
3807def : InstAlias<"mov{\t$dst.2s, $src.2s|.2s\t$dst, $src}",
3808                (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
3809def : InstAlias<"mov{\t$dst.1d, $src.1d|.1d\t$dst, $src}",
3810                (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
3811
3812def : InstAlias<"{cmls\t$dst.8b, $src1.8b, $src2.8b" #
3813                "|cmls.8b\t$dst, $src1, $src2}",
3814                (CMHSv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
3815def : InstAlias<"{cmls\t$dst.16b, $src1.16b, $src2.16b" #
3816                "|cmls.16b\t$dst, $src1, $src2}",
3817                (CMHSv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
3818def : InstAlias<"{cmls\t$dst.4h, $src1.4h, $src2.4h" #
3819                "|cmls.4h\t$dst, $src1, $src2}",
3820                (CMHSv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
3821def : InstAlias<"{cmls\t$dst.8h, $src1.8h, $src2.8h" #
3822                "|cmls.8h\t$dst, $src1, $src2}",
3823                (CMHSv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
3824def : InstAlias<"{cmls\t$dst.2s, $src1.2s, $src2.2s" #
3825                "|cmls.2s\t$dst, $src1, $src2}",
3826                (CMHSv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
3827def : InstAlias<"{cmls\t$dst.4s, $src1.4s, $src2.4s" #
3828                "|cmls.4s\t$dst, $src1, $src2}",
3829                (CMHSv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
3830def : InstAlias<"{cmls\t$dst.2d, $src1.2d, $src2.2d" #
3831                "|cmls.2d\t$dst, $src1, $src2}",
3832                (CMHSv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
3833
3834def : InstAlias<"{cmlo\t$dst.8b, $src1.8b, $src2.8b" #
3835                "|cmlo.8b\t$dst, $src1, $src2}",
3836                (CMHIv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
3837def : InstAlias<"{cmlo\t$dst.16b, $src1.16b, $src2.16b" #
3838                "|cmlo.16b\t$dst, $src1, $src2}",
3839                (CMHIv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
3840def : InstAlias<"{cmlo\t$dst.4h, $src1.4h, $src2.4h" #
3841                "|cmlo.4h\t$dst, $src1, $src2}",
3842                (CMHIv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
3843def : InstAlias<"{cmlo\t$dst.8h, $src1.8h, $src2.8h" #
3844                "|cmlo.8h\t$dst, $src1, $src2}",
3845                (CMHIv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
3846def : InstAlias<"{cmlo\t$dst.2s, $src1.2s, $src2.2s" #
3847                "|cmlo.2s\t$dst, $src1, $src2}",
3848                (CMHIv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
3849def : InstAlias<"{cmlo\t$dst.4s, $src1.4s, $src2.4s" #
3850                "|cmlo.4s\t$dst, $src1, $src2}",
3851                (CMHIv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
3852def : InstAlias<"{cmlo\t$dst.2d, $src1.2d, $src2.2d" #
3853                "|cmlo.2d\t$dst, $src1, $src2}",
3854                (CMHIv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
3855
3856def : InstAlias<"{cmle\t$dst.8b, $src1.8b, $src2.8b" #
3857                "|cmle.8b\t$dst, $src1, $src2}",
3858                (CMGEv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
3859def : InstAlias<"{cmle\t$dst.16b, $src1.16b, $src2.16b" #
3860                "|cmle.16b\t$dst, $src1, $src2}",
3861                (CMGEv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
3862def : InstAlias<"{cmle\t$dst.4h, $src1.4h, $src2.4h" #
3863                "|cmle.4h\t$dst, $src1, $src2}",
3864                (CMGEv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
3865def : InstAlias<"{cmle\t$dst.8h, $src1.8h, $src2.8h" #
3866                "|cmle.8h\t$dst, $src1, $src2}",
3867                (CMGEv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
3868def : InstAlias<"{cmle\t$dst.2s, $src1.2s, $src2.2s" #
3869                "|cmle.2s\t$dst, $src1, $src2}",
3870                (CMGEv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
3871def : InstAlias<"{cmle\t$dst.4s, $src1.4s, $src2.4s" #
3872                "|cmle.4s\t$dst, $src1, $src2}",
3873                (CMGEv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
3874def : InstAlias<"{cmle\t$dst.2d, $src1.2d, $src2.2d" #
3875                "|cmle.2d\t$dst, $src1, $src2}",
3876                (CMGEv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
3877
3878def : InstAlias<"{cmlt\t$dst.8b, $src1.8b, $src2.8b" #
3879                "|cmlt.8b\t$dst, $src1, $src2}",
3880                (CMGTv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
3881def : InstAlias<"{cmlt\t$dst.16b, $src1.16b, $src2.16b" #
3882                "|cmlt.16b\t$dst, $src1, $src2}",
3883                (CMGTv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
3884def : InstAlias<"{cmlt\t$dst.4h, $src1.4h, $src2.4h" #
3885                "|cmlt.4h\t$dst, $src1, $src2}",
3886                (CMGTv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
3887def : InstAlias<"{cmlt\t$dst.8h, $src1.8h, $src2.8h" #
3888                "|cmlt.8h\t$dst, $src1, $src2}",
3889                (CMGTv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
3890def : InstAlias<"{cmlt\t$dst.2s, $src1.2s, $src2.2s" #
3891                "|cmlt.2s\t$dst, $src1, $src2}",
3892                (CMGTv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
3893def : InstAlias<"{cmlt\t$dst.4s, $src1.4s, $src2.4s" #
3894                "|cmlt.4s\t$dst, $src1, $src2}",
3895                (CMGTv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
3896def : InstAlias<"{cmlt\t$dst.2d, $src1.2d, $src2.2d" #
3897                "|cmlt.2d\t$dst, $src1, $src2}",
3898                (CMGTv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
3899
3900let Predicates = [HasNEON, HasFullFP16] in {
3901def : InstAlias<"{fcmle\t$dst.4h, $src1.4h, $src2.4h" #
3902                "|fcmle.4h\t$dst, $src1, $src2}",
3903                (FCMGEv4f16 V64:$dst, V64:$src2, V64:$src1), 0>;
3904def : InstAlias<"{fcmle\t$dst.8h, $src1.8h, $src2.8h" #
3905                "|fcmle.8h\t$dst, $src1, $src2}",
3906                (FCMGEv8f16 V128:$dst, V128:$src2, V128:$src1), 0>;
3907}
3908def : InstAlias<"{fcmle\t$dst.2s, $src1.2s, $src2.2s" #
3909                "|fcmle.2s\t$dst, $src1, $src2}",
3910                (FCMGEv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
3911def : InstAlias<"{fcmle\t$dst.4s, $src1.4s, $src2.4s" #
3912                "|fcmle.4s\t$dst, $src1, $src2}",
3913                (FCMGEv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
3914def : InstAlias<"{fcmle\t$dst.2d, $src1.2d, $src2.2d" #
3915                "|fcmle.2d\t$dst, $src1, $src2}",
3916                (FCMGEv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
3917
3918let Predicates = [HasNEON, HasFullFP16] in {
3919def : InstAlias<"{fcmlt\t$dst.4h, $src1.4h, $src2.4h" #
3920                "|fcmlt.4h\t$dst, $src1, $src2}",
3921                (FCMGTv4f16 V64:$dst, V64:$src2, V64:$src1), 0>;
3922def : InstAlias<"{fcmlt\t$dst.8h, $src1.8h, $src2.8h" #
3923                "|fcmlt.8h\t$dst, $src1, $src2}",
3924                (FCMGTv8f16 V128:$dst, V128:$src2, V128:$src1), 0>;
3925}
3926def : InstAlias<"{fcmlt\t$dst.2s, $src1.2s, $src2.2s" #
3927                "|fcmlt.2s\t$dst, $src1, $src2}",
3928                (FCMGTv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
3929def : InstAlias<"{fcmlt\t$dst.4s, $src1.4s, $src2.4s" #
3930                "|fcmlt.4s\t$dst, $src1, $src2}",
3931                (FCMGTv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
3932def : InstAlias<"{fcmlt\t$dst.2d, $src1.2d, $src2.2d" #
3933                "|fcmlt.2d\t$dst, $src1, $src2}",
3934                (FCMGTv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
3935
3936let Predicates = [HasNEON, HasFullFP16] in {
3937def : InstAlias<"{facle\t$dst.4h, $src1.4h, $src2.4h" #
3938                "|facle.4h\t$dst, $src1, $src2}",
3939                (FACGEv4f16 V64:$dst, V64:$src2, V64:$src1), 0>;
3940def : InstAlias<"{facle\t$dst.8h, $src1.8h, $src2.8h" #
3941                "|facle.8h\t$dst, $src1, $src2}",
3942                (FACGEv8f16 V128:$dst, V128:$src2, V128:$src1), 0>;
3943}
3944def : InstAlias<"{facle\t$dst.2s, $src1.2s, $src2.2s" #
3945                "|facle.2s\t$dst, $src1, $src2}",
3946                (FACGEv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
3947def : InstAlias<"{facle\t$dst.4s, $src1.4s, $src2.4s" #
3948                "|facle.4s\t$dst, $src1, $src2}",
3949                (FACGEv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
3950def : InstAlias<"{facle\t$dst.2d, $src1.2d, $src2.2d" #
3951                "|facle.2d\t$dst, $src1, $src2}",
3952                (FACGEv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
3953
3954let Predicates = [HasNEON, HasFullFP16] in {
3955def : InstAlias<"{faclt\t$dst.4h, $src1.4h, $src2.4h" #
3956                "|faclt.4h\t$dst, $src1, $src2}",
3957                (FACGTv4f16 V64:$dst, V64:$src2, V64:$src1), 0>;
3958def : InstAlias<"{faclt\t$dst.8h, $src1.8h, $src2.8h" #
3959                "|faclt.8h\t$dst, $src1, $src2}",
3960                (FACGTv8f16 V128:$dst, V128:$src2, V128:$src1), 0>;
3961}
3962def : InstAlias<"{faclt\t$dst.2s, $src1.2s, $src2.2s" #
3963                "|faclt.2s\t$dst, $src1, $src2}",
3964                (FACGTv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
3965def : InstAlias<"{faclt\t$dst.4s, $src1.4s, $src2.4s" #
3966                "|faclt.4s\t$dst, $src1, $src2}",
3967                (FACGTv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
3968def : InstAlias<"{faclt\t$dst.2d, $src1.2d, $src2.2d" #
3969                "|faclt.2d\t$dst, $src1, $src2}",
3970                (FACGTv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
3971
3972//===----------------------------------------------------------------------===//
3973// Advanced SIMD three scalar instructions.
3974//===----------------------------------------------------------------------===//
3975
3976defm ADD      : SIMDThreeScalarD<0, 0b10000, "add", add>;
3977defm CMEQ     : SIMDThreeScalarD<1, 0b10001, "cmeq", AArch64cmeq>;
3978defm CMGE     : SIMDThreeScalarD<0, 0b00111, "cmge", AArch64cmge>;
3979defm CMGT     : SIMDThreeScalarD<0, 0b00110, "cmgt", AArch64cmgt>;
3980defm CMHI     : SIMDThreeScalarD<1, 0b00110, "cmhi", AArch64cmhi>;
3981defm CMHS     : SIMDThreeScalarD<1, 0b00111, "cmhs", AArch64cmhs>;
3982defm CMTST    : SIMDThreeScalarD<0, 0b10001, "cmtst", AArch64cmtst>;
3983defm FABD     : SIMDFPThreeScalar<1, 1, 0b010, "fabd", int_aarch64_sisd_fabd>;
3984def : Pat<(v1f64 (int_aarch64_neon_fabd (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
3985          (FABD64 FPR64:$Rn, FPR64:$Rm)>;
3986let Predicates = [HasFullFP16] in {
3987def : Pat<(fabs (fsub f16:$Rn, f16:$Rm)), (FABD16 f16:$Rn, f16:$Rm)>;
3988}
3989def : Pat<(fabs (fsub f32:$Rn, f32:$Rm)), (FABD32 f32:$Rn, f32:$Rm)>;
3990def : Pat<(fabs (fsub f64:$Rn, f64:$Rm)), (FABD64 f64:$Rn, f64:$Rm)>;
3991defm FACGE    : SIMDThreeScalarFPCmp<1, 0, 0b101, "facge",
3992                                     int_aarch64_neon_facge>;
3993defm FACGT    : SIMDThreeScalarFPCmp<1, 1, 0b101, "facgt",
3994                                     int_aarch64_neon_facgt>;
3995defm FCMEQ    : SIMDThreeScalarFPCmp<0, 0, 0b100, "fcmeq", AArch64fcmeq>;
3996defm FCMGE    : SIMDThreeScalarFPCmp<1, 0, 0b100, "fcmge", AArch64fcmge>;
3997defm FCMGT    : SIMDThreeScalarFPCmp<1, 1, 0b100, "fcmgt", AArch64fcmgt>;
3998defm FMULX    : SIMDFPThreeScalar<0, 0, 0b011, "fmulx", int_aarch64_neon_fmulx>;
3999defm FRECPS   : SIMDFPThreeScalar<0, 0, 0b111, "frecps", int_aarch64_neon_frecps>;
4000defm FRSQRTS  : SIMDFPThreeScalar<0, 1, 0b111, "frsqrts", int_aarch64_neon_frsqrts>;
4001defm SQADD    : SIMDThreeScalarBHSD<0, 0b00001, "sqadd", int_aarch64_neon_sqadd>;
4002defm SQDMULH  : SIMDThreeScalarHS<  0, 0b10110, "sqdmulh", int_aarch64_neon_sqdmulh>;
4003defm SQRDMULH : SIMDThreeScalarHS<  1, 0b10110, "sqrdmulh", int_aarch64_neon_sqrdmulh>;
4004defm SQRSHL   : SIMDThreeScalarBHSD<0, 0b01011, "sqrshl",int_aarch64_neon_sqrshl>;
4005defm SQSHL    : SIMDThreeScalarBHSD<0, 0b01001, "sqshl", int_aarch64_neon_sqshl>;
4006defm SQSUB    : SIMDThreeScalarBHSD<0, 0b00101, "sqsub", int_aarch64_neon_sqsub>;
4007defm SRSHL    : SIMDThreeScalarD<   0, 0b01010, "srshl", int_aarch64_neon_srshl>;
4008defm SSHL     : SIMDThreeScalarD<   0, 0b01000, "sshl", int_aarch64_neon_sshl>;
4009defm SUB      : SIMDThreeScalarD<   1, 0b10000, "sub", sub>;
4010defm UQADD    : SIMDThreeScalarBHSD<1, 0b00001, "uqadd", int_aarch64_neon_uqadd>;
4011defm UQRSHL   : SIMDThreeScalarBHSD<1, 0b01011, "uqrshl",int_aarch64_neon_uqrshl>;
4012defm UQSHL    : SIMDThreeScalarBHSD<1, 0b01001, "uqshl", int_aarch64_neon_uqshl>;
4013defm UQSUB    : SIMDThreeScalarBHSD<1, 0b00101, "uqsub", int_aarch64_neon_uqsub>;
4014defm URSHL    : SIMDThreeScalarD<   1, 0b01010, "urshl", int_aarch64_neon_urshl>;
4015defm USHL     : SIMDThreeScalarD<   1, 0b01000, "ushl", int_aarch64_neon_ushl>;
4016let Predicates = [HasRDM] in {
4017  defm SQRDMLAH : SIMDThreeScalarHSTied<1, 0, 0b10000, "sqrdmlah">;
4018  defm SQRDMLSH : SIMDThreeScalarHSTied<1, 0, 0b10001, "sqrdmlsh">;
4019  def : Pat<(i32 (int_aarch64_neon_sqadd
4020                   (i32 FPR32:$Rd),
4021                   (i32 (int_aarch64_neon_sqrdmulh (i32 FPR32:$Rn),
4022                                                   (i32 FPR32:$Rm))))),
4023            (SQRDMLAHv1i32 FPR32:$Rd, FPR32:$Rn, FPR32:$Rm)>;
4024  def : Pat<(i32 (int_aarch64_neon_sqsub
4025                   (i32 FPR32:$Rd),
4026                   (i32 (int_aarch64_neon_sqrdmulh (i32 FPR32:$Rn),
4027                                                   (i32 FPR32:$Rm))))),
4028            (SQRDMLSHv1i32 FPR32:$Rd, FPR32:$Rn, FPR32:$Rm)>;
4029}
4030
4031def : InstAlias<"cmls $dst, $src1, $src2",
4032                (CMHSv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
4033def : InstAlias<"cmle $dst, $src1, $src2",
4034                (CMGEv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
4035def : InstAlias<"cmlo $dst, $src1, $src2",
4036                (CMHIv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
4037def : InstAlias<"cmlt $dst, $src1, $src2",
4038                (CMGTv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
4039def : InstAlias<"fcmle $dst, $src1, $src2",
4040                (FCMGE32 FPR32:$dst, FPR32:$src2, FPR32:$src1), 0>;
4041def : InstAlias<"fcmle $dst, $src1, $src2",
4042                (FCMGE64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
4043def : InstAlias<"fcmlt $dst, $src1, $src2",
4044                (FCMGT32 FPR32:$dst, FPR32:$src2, FPR32:$src1), 0>;
4045def : InstAlias<"fcmlt $dst, $src1, $src2",
4046                (FCMGT64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
4047def : InstAlias<"facle $dst, $src1, $src2",
4048                (FACGE32 FPR32:$dst, FPR32:$src2, FPR32:$src1), 0>;
4049def : InstAlias<"facle $dst, $src1, $src2",
4050                (FACGE64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
4051def : InstAlias<"faclt $dst, $src1, $src2",
4052                (FACGT32 FPR32:$dst, FPR32:$src2, FPR32:$src1), 0>;
4053def : InstAlias<"faclt $dst, $src1, $src2",
4054                (FACGT64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
4055
4056//===----------------------------------------------------------------------===//
4057// Advanced SIMD three scalar instructions (mixed operands).
4058//===----------------------------------------------------------------------===//
4059defm SQDMULL  : SIMDThreeScalarMixedHS<0, 0b11010, "sqdmull",
4060                                       int_aarch64_neon_sqdmulls_scalar>;
4061defm SQDMLAL  : SIMDThreeScalarMixedTiedHS<0, 0b10010, "sqdmlal">;
4062defm SQDMLSL  : SIMDThreeScalarMixedTiedHS<0, 0b10110, "sqdmlsl">;
4063
4064def : Pat<(i64 (int_aarch64_neon_sqadd (i64 FPR64:$Rd),
4065                   (i64 (int_aarch64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
4066                                                        (i32 FPR32:$Rm))))),
4067          (SQDMLALi32 FPR64:$Rd, FPR32:$Rn, FPR32:$Rm)>;
4068def : Pat<(i64 (int_aarch64_neon_sqsub (i64 FPR64:$Rd),
4069                   (i64 (int_aarch64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
4070                                                        (i32 FPR32:$Rm))))),
4071          (SQDMLSLi32 FPR64:$Rd, FPR32:$Rn, FPR32:$Rm)>;
4072
4073//===----------------------------------------------------------------------===//
4074// Advanced SIMD two scalar instructions.
4075//===----------------------------------------------------------------------===//
4076
4077defm ABS    : SIMDTwoScalarD<    0, 0b01011, "abs", abs>;
4078defm CMEQ   : SIMDCmpTwoScalarD< 0, 0b01001, "cmeq", AArch64cmeqz>;
4079defm CMGE   : SIMDCmpTwoScalarD< 1, 0b01000, "cmge", AArch64cmgez>;
4080defm CMGT   : SIMDCmpTwoScalarD< 0, 0b01000, "cmgt", AArch64cmgtz>;
4081defm CMLE   : SIMDCmpTwoScalarD< 1, 0b01001, "cmle", AArch64cmlez>;
4082defm CMLT   : SIMDCmpTwoScalarD< 0, 0b01010, "cmlt", AArch64cmltz>;
4083defm FCMEQ  : SIMDFPCmpTwoScalar<0, 1, 0b01101, "fcmeq", AArch64fcmeqz>;
4084defm FCMGE  : SIMDFPCmpTwoScalar<1, 1, 0b01100, "fcmge", AArch64fcmgez>;
4085defm FCMGT  : SIMDFPCmpTwoScalar<0, 1, 0b01100, "fcmgt", AArch64fcmgtz>;
4086defm FCMLE  : SIMDFPCmpTwoScalar<1, 1, 0b01101, "fcmle", AArch64fcmlez>;
4087defm FCMLT  : SIMDFPCmpTwoScalar<0, 1, 0b01110, "fcmlt", AArch64fcmltz>;
4088defm FCVTAS : SIMDFPTwoScalar<   0, 0, 0b11100, "fcvtas">;
4089defm FCVTAU : SIMDFPTwoScalar<   1, 0, 0b11100, "fcvtau">;
4090defm FCVTMS : SIMDFPTwoScalar<   0, 0, 0b11011, "fcvtms">;
4091defm FCVTMU : SIMDFPTwoScalar<   1, 0, 0b11011, "fcvtmu">;
4092defm FCVTNS : SIMDFPTwoScalar<   0, 0, 0b11010, "fcvtns">;
4093defm FCVTNU : SIMDFPTwoScalar<   1, 0, 0b11010, "fcvtnu">;
4094defm FCVTPS : SIMDFPTwoScalar<   0, 1, 0b11010, "fcvtps">;
4095defm FCVTPU : SIMDFPTwoScalar<   1, 1, 0b11010, "fcvtpu">;
4096def  FCVTXNv1i64 : SIMDInexactCvtTwoScalar<0b10110, "fcvtxn">;
4097defm FCVTZS : SIMDFPTwoScalar<   0, 1, 0b11011, "fcvtzs">;
4098defm FCVTZU : SIMDFPTwoScalar<   1, 1, 0b11011, "fcvtzu">;
4099defm FRECPE : SIMDFPTwoScalar<   0, 1, 0b11101, "frecpe">;
4100defm FRECPX : SIMDFPTwoScalar<   0, 1, 0b11111, "frecpx">;
4101defm FRSQRTE : SIMDFPTwoScalar<  1, 1, 0b11101, "frsqrte">;
4102defm NEG    : SIMDTwoScalarD<    1, 0b01011, "neg",
4103                                 UnOpFrag<(sub immAllZerosV, node:$LHS)> >;
4104defm SCVTF  : SIMDFPTwoScalarCVT<   0, 0, 0b11101, "scvtf", AArch64sitof>;
4105defm SQABS  : SIMDTwoScalarBHSD< 0, 0b00111, "sqabs", int_aarch64_neon_sqabs>;
4106defm SQNEG  : SIMDTwoScalarBHSD< 1, 0b00111, "sqneg", int_aarch64_neon_sqneg>;
4107defm SQXTN  : SIMDTwoScalarMixedBHS< 0, 0b10100, "sqxtn", int_aarch64_neon_scalar_sqxtn>;
4108defm SQXTUN : SIMDTwoScalarMixedBHS< 1, 0b10010, "sqxtun", int_aarch64_neon_scalar_sqxtun>;
4109defm SUQADD : SIMDTwoScalarBHSDTied< 0, 0b00011, "suqadd",
4110                                     int_aarch64_neon_suqadd>;
4111defm UCVTF  : SIMDFPTwoScalarCVT<   1, 0, 0b11101, "ucvtf", AArch64uitof>;
4112defm UQXTN  : SIMDTwoScalarMixedBHS<1, 0b10100, "uqxtn", int_aarch64_neon_scalar_uqxtn>;
4113defm USQADD : SIMDTwoScalarBHSDTied< 1, 0b00011, "usqadd",
4114                                    int_aarch64_neon_usqadd>;
4115
4116def : Pat<(AArch64neg (v1i64 V64:$Rn)), (NEGv1i64 V64:$Rn)>;
4117
4118def : Pat<(v1i64 (int_aarch64_neon_fcvtas (v1f64 FPR64:$Rn))),
4119          (FCVTASv1i64 FPR64:$Rn)>;
4120def : Pat<(v1i64 (int_aarch64_neon_fcvtau (v1f64 FPR64:$Rn))),
4121          (FCVTAUv1i64 FPR64:$Rn)>;
4122def : Pat<(v1i64 (int_aarch64_neon_fcvtms (v1f64 FPR64:$Rn))),
4123          (FCVTMSv1i64 FPR64:$Rn)>;
4124def : Pat<(v1i64 (int_aarch64_neon_fcvtmu (v1f64 FPR64:$Rn))),
4125          (FCVTMUv1i64 FPR64:$Rn)>;
4126def : Pat<(v1i64 (int_aarch64_neon_fcvtns (v1f64 FPR64:$Rn))),
4127          (FCVTNSv1i64 FPR64:$Rn)>;
4128def : Pat<(v1i64 (int_aarch64_neon_fcvtnu (v1f64 FPR64:$Rn))),
4129          (FCVTNUv1i64 FPR64:$Rn)>;
4130def : Pat<(v1i64 (int_aarch64_neon_fcvtps (v1f64 FPR64:$Rn))),
4131          (FCVTPSv1i64 FPR64:$Rn)>;
4132def : Pat<(v1i64 (int_aarch64_neon_fcvtpu (v1f64 FPR64:$Rn))),
4133          (FCVTPUv1i64 FPR64:$Rn)>;
4134
4135def : Pat<(f16 (int_aarch64_neon_frecpe (f16 FPR16:$Rn))),
4136          (FRECPEv1f16 FPR16:$Rn)>;
4137def : Pat<(f32 (int_aarch64_neon_frecpe (f32 FPR32:$Rn))),
4138          (FRECPEv1i32 FPR32:$Rn)>;
4139def : Pat<(f64 (int_aarch64_neon_frecpe (f64 FPR64:$Rn))),
4140          (FRECPEv1i64 FPR64:$Rn)>;
4141def : Pat<(v1f64 (int_aarch64_neon_frecpe (v1f64 FPR64:$Rn))),
4142          (FRECPEv1i64 FPR64:$Rn)>;
4143
4144def : Pat<(f32 (AArch64frecpe (f32 FPR32:$Rn))),
4145          (FRECPEv1i32 FPR32:$Rn)>;
4146def : Pat<(v2f32 (AArch64frecpe (v2f32 V64:$Rn))),
4147          (FRECPEv2f32 V64:$Rn)>;
4148def : Pat<(v4f32 (AArch64frecpe (v4f32 FPR128:$Rn))),
4149          (FRECPEv4f32 FPR128:$Rn)>;
4150def : Pat<(f64 (AArch64frecpe (f64 FPR64:$Rn))),
4151          (FRECPEv1i64 FPR64:$Rn)>;
4152def : Pat<(v1f64 (AArch64frecpe (v1f64 FPR64:$Rn))),
4153          (FRECPEv1i64 FPR64:$Rn)>;
4154def : Pat<(v2f64 (AArch64frecpe (v2f64 FPR128:$Rn))),
4155          (FRECPEv2f64 FPR128:$Rn)>;
4156
4157def : Pat<(f32 (AArch64frecps (f32 FPR32:$Rn), (f32 FPR32:$Rm))),
4158          (FRECPS32 FPR32:$Rn, FPR32:$Rm)>;
4159def : Pat<(v2f32 (AArch64frecps (v2f32 V64:$Rn), (v2f32 V64:$Rm))),
4160          (FRECPSv2f32 V64:$Rn, V64:$Rm)>;
4161def : Pat<(v4f32 (AArch64frecps (v4f32 FPR128:$Rn), (v4f32 FPR128:$Rm))),
4162          (FRECPSv4f32 FPR128:$Rn, FPR128:$Rm)>;
4163def : Pat<(f64 (AArch64frecps (f64 FPR64:$Rn), (f64 FPR64:$Rm))),
4164          (FRECPS64 FPR64:$Rn, FPR64:$Rm)>;
4165def : Pat<(v2f64 (AArch64frecps (v2f64 FPR128:$Rn), (v2f64 FPR128:$Rm))),
4166          (FRECPSv2f64 FPR128:$Rn, FPR128:$Rm)>;
4167
4168def : Pat<(f16 (int_aarch64_neon_frecpx (f16 FPR16:$Rn))),
4169          (FRECPXv1f16 FPR16:$Rn)>;
4170def : Pat<(f32 (int_aarch64_neon_frecpx (f32 FPR32:$Rn))),
4171          (FRECPXv1i32 FPR32:$Rn)>;
4172def : Pat<(f64 (int_aarch64_neon_frecpx (f64 FPR64:$Rn))),
4173          (FRECPXv1i64 FPR64:$Rn)>;
4174
4175def : Pat<(f16 (int_aarch64_neon_frsqrte (f16 FPR16:$Rn))),
4176          (FRSQRTEv1f16 FPR16:$Rn)>;
4177def : Pat<(f32 (int_aarch64_neon_frsqrte (f32 FPR32:$Rn))),
4178          (FRSQRTEv1i32 FPR32:$Rn)>;
4179def : Pat<(f64 (int_aarch64_neon_frsqrte (f64 FPR64:$Rn))),
4180          (FRSQRTEv1i64 FPR64:$Rn)>;
4181def : Pat<(v1f64 (int_aarch64_neon_frsqrte (v1f64 FPR64:$Rn))),
4182          (FRSQRTEv1i64 FPR64:$Rn)>;
4183
4184def : Pat<(f32 (AArch64frsqrte (f32 FPR32:$Rn))),
4185          (FRSQRTEv1i32 FPR32:$Rn)>;
4186def : Pat<(v2f32 (AArch64frsqrte (v2f32 V64:$Rn))),
4187          (FRSQRTEv2f32 V64:$Rn)>;
4188def : Pat<(v4f32 (AArch64frsqrte (v4f32 FPR128:$Rn))),
4189          (FRSQRTEv4f32 FPR128:$Rn)>;
4190def : Pat<(f64 (AArch64frsqrte (f64 FPR64:$Rn))),
4191          (FRSQRTEv1i64 FPR64:$Rn)>;
4192def : Pat<(v1f64 (AArch64frsqrte (v1f64 FPR64:$Rn))),
4193          (FRSQRTEv1i64 FPR64:$Rn)>;
4194def : Pat<(v2f64 (AArch64frsqrte (v2f64 FPR128:$Rn))),
4195          (FRSQRTEv2f64 FPR128:$Rn)>;
4196
4197def : Pat<(f32 (AArch64frsqrts (f32 FPR32:$Rn), (f32 FPR32:$Rm))),
4198          (FRSQRTS32 FPR32:$Rn, FPR32:$Rm)>;
4199def : Pat<(v2f32 (AArch64frsqrts (v2f32 V64:$Rn), (v2f32 V64:$Rm))),
4200          (FRSQRTSv2f32 V64:$Rn, V64:$Rm)>;
4201def : Pat<(v4f32 (AArch64frsqrts (v4f32 FPR128:$Rn), (v4f32 FPR128:$Rm))),
4202          (FRSQRTSv4f32 FPR128:$Rn, FPR128:$Rm)>;
4203def : Pat<(f64 (AArch64frsqrts (f64 FPR64:$Rn), (f64 FPR64:$Rm))),
4204          (FRSQRTS64 FPR64:$Rn, FPR64:$Rm)>;
4205def : Pat<(v2f64 (AArch64frsqrts (v2f64 FPR128:$Rn), (v2f64 FPR128:$Rm))),
4206          (FRSQRTSv2f64 FPR128:$Rn, FPR128:$Rm)>;
4207
4208// If an integer is about to be converted to a floating point value,
4209// just load it on the floating point unit.
4210// Here are the patterns for 8 and 16-bits to float.
4211// 8-bits -> float.
4212multiclass UIntToFPROLoadPat<ValueType DstTy, ValueType SrcTy,
4213                             SDPatternOperator loadop, Instruction UCVTF,
4214                             ROAddrMode ro, Instruction LDRW, Instruction LDRX,
4215                             SubRegIndex sub> {
4216  def : Pat<(DstTy (uint_to_fp (SrcTy
4217                     (loadop (ro.Wpat GPR64sp:$Rn, GPR32:$Rm,
4218                                      ro.Wext:$extend))))),
4219           (UCVTF (INSERT_SUBREG (DstTy (IMPLICIT_DEF)),
4220                                 (LDRW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend),
4221                                 sub))>;
4222
4223  def : Pat<(DstTy (uint_to_fp (SrcTy
4224                     (loadop (ro.Xpat GPR64sp:$Rn, GPR64:$Rm,
4225                                      ro.Wext:$extend))))),
4226           (UCVTF (INSERT_SUBREG (DstTy (IMPLICIT_DEF)),
4227                                 (LDRX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend),
4228                                 sub))>;
4229}
4230
4231defm : UIntToFPROLoadPat<f32, i32, zextloadi8,
4232                         UCVTFv1i32, ro8, LDRBroW, LDRBroX, bsub>;
4233def : Pat <(f32 (uint_to_fp (i32
4234               (zextloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))))),
4235           (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
4236                          (LDRBui GPR64sp:$Rn, uimm12s1:$offset), bsub))>;
4237def : Pat <(f32 (uint_to_fp (i32
4238                     (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))))),
4239           (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
4240                          (LDURBi GPR64sp:$Rn, simm9:$offset), bsub))>;
4241// 16-bits -> float.
4242defm : UIntToFPROLoadPat<f32, i32, zextloadi16,
4243                         UCVTFv1i32, ro16, LDRHroW, LDRHroX, hsub>;
4244def : Pat <(f32 (uint_to_fp (i32
4245                  (zextloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))))),
4246           (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
4247                          (LDRHui GPR64sp:$Rn, uimm12s2:$offset), hsub))>;
4248def : Pat <(f32 (uint_to_fp (i32
4249                  (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))))),
4250           (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
4251                          (LDURHi GPR64sp:$Rn, simm9:$offset), hsub))>;
4252// 32-bits are handled in target specific dag combine:
4253// performIntToFpCombine.
4254// 64-bits integer to 32-bits floating point, not possible with
4255// UCVTF on floating point registers (both source and destination
4256// must have the same size).
4257
4258// Here are the patterns for 8, 16, 32, and 64-bits to double.
4259// 8-bits -> double.
4260defm : UIntToFPROLoadPat<f64, i32, zextloadi8,
4261                         UCVTFv1i64, ro8, LDRBroW, LDRBroX, bsub>;
4262def : Pat <(f64 (uint_to_fp (i32
4263                    (zextloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))))),
4264           (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4265                          (LDRBui GPR64sp:$Rn, uimm12s1:$offset), bsub))>;
4266def : Pat <(f64 (uint_to_fp (i32
4267                  (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))))),
4268           (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4269                          (LDURBi GPR64sp:$Rn, simm9:$offset), bsub))>;
4270// 16-bits -> double.
4271defm : UIntToFPROLoadPat<f64, i32, zextloadi16,
4272                         UCVTFv1i64, ro16, LDRHroW, LDRHroX, hsub>;
4273def : Pat <(f64 (uint_to_fp (i32
4274                  (zextloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))))),
4275           (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4276                          (LDRHui GPR64sp:$Rn, uimm12s2:$offset), hsub))>;
4277def : Pat <(f64 (uint_to_fp (i32
4278                  (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))))),
4279           (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4280                          (LDURHi GPR64sp:$Rn, simm9:$offset), hsub))>;
4281// 32-bits -> double.
4282defm : UIntToFPROLoadPat<f64, i32, load,
4283                         UCVTFv1i64, ro32, LDRSroW, LDRSroX, ssub>;
4284def : Pat <(f64 (uint_to_fp (i32
4285                  (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))))),
4286           (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4287                          (LDRSui GPR64sp:$Rn, uimm12s4:$offset), ssub))>;
4288def : Pat <(f64 (uint_to_fp (i32
4289                  (load (am_unscaled32 GPR64sp:$Rn, simm9:$offset))))),
4290           (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4291                          (LDURSi GPR64sp:$Rn, simm9:$offset), ssub))>;
4292// 64-bits -> double are handled in target specific dag combine:
4293// performIntToFpCombine.
4294
4295//===----------------------------------------------------------------------===//
4296// Advanced SIMD three different-sized vector instructions.
4297//===----------------------------------------------------------------------===//
4298
4299defm ADDHN  : SIMDNarrowThreeVectorBHS<0,0b0100,"addhn", int_aarch64_neon_addhn>;
4300defm SUBHN  : SIMDNarrowThreeVectorBHS<0,0b0110,"subhn", int_aarch64_neon_subhn>;
4301defm RADDHN : SIMDNarrowThreeVectorBHS<1,0b0100,"raddhn",int_aarch64_neon_raddhn>;
4302defm RSUBHN : SIMDNarrowThreeVectorBHS<1,0b0110,"rsubhn",int_aarch64_neon_rsubhn>;
4303defm PMULL  : SIMDDifferentThreeVectorBD<0,0b1110,"pmull",int_aarch64_neon_pmull>;
4304defm SABAL  : SIMDLongThreeVectorTiedBHSabal<0,0b0101,"sabal",
4305                                             int_aarch64_neon_sabd>;
4306defm SABDL   : SIMDLongThreeVectorBHSabdl<0, 0b0111, "sabdl",
4307                                          int_aarch64_neon_sabd>;
4308defm SADDL   : SIMDLongThreeVectorBHS<   0, 0b0000, "saddl",
4309            BinOpFrag<(add (sext node:$LHS), (sext node:$RHS))>>;
4310defm SADDW   : SIMDWideThreeVectorBHS<   0, 0b0001, "saddw",
4311                 BinOpFrag<(add node:$LHS, (sext node:$RHS))>>;
4312defm SMLAL   : SIMDLongThreeVectorTiedBHS<0, 0b1000, "smlal",
4313    TriOpFrag<(add node:$LHS, (int_aarch64_neon_smull node:$MHS, node:$RHS))>>;
4314defm SMLSL   : SIMDLongThreeVectorTiedBHS<0, 0b1010, "smlsl",
4315    TriOpFrag<(sub node:$LHS, (int_aarch64_neon_smull node:$MHS, node:$RHS))>>;
4316defm SMULL   : SIMDLongThreeVectorBHS<0, 0b1100, "smull", int_aarch64_neon_smull>;
4317defm SQDMLAL : SIMDLongThreeVectorSQDMLXTiedHS<0, 0b1001, "sqdmlal",
4318                                               int_aarch64_neon_sqadd>;
4319defm SQDMLSL : SIMDLongThreeVectorSQDMLXTiedHS<0, 0b1011, "sqdmlsl",
4320                                               int_aarch64_neon_sqsub>;
4321defm SQDMULL : SIMDLongThreeVectorHS<0, 0b1101, "sqdmull",
4322                                     int_aarch64_neon_sqdmull>;
4323defm SSUBL   : SIMDLongThreeVectorBHS<0, 0b0010, "ssubl",
4324                 BinOpFrag<(sub (sext node:$LHS), (sext node:$RHS))>>;
4325defm SSUBW   : SIMDWideThreeVectorBHS<0, 0b0011, "ssubw",
4326                 BinOpFrag<(sub node:$LHS, (sext node:$RHS))>>;
4327defm UABAL   : SIMDLongThreeVectorTiedBHSabal<1, 0b0101, "uabal",
4328                                              int_aarch64_neon_uabd>;
4329defm UADDL   : SIMDLongThreeVectorBHS<1, 0b0000, "uaddl",
4330                 BinOpFrag<(add (zext node:$LHS), (zext node:$RHS))>>;
4331defm UADDW   : SIMDWideThreeVectorBHS<1, 0b0001, "uaddw",
4332                 BinOpFrag<(add node:$LHS, (zext node:$RHS))>>;
4333defm UMLAL   : SIMDLongThreeVectorTiedBHS<1, 0b1000, "umlal",
4334    TriOpFrag<(add node:$LHS, (int_aarch64_neon_umull node:$MHS, node:$RHS))>>;
4335defm UMLSL   : SIMDLongThreeVectorTiedBHS<1, 0b1010, "umlsl",
4336    TriOpFrag<(sub node:$LHS, (int_aarch64_neon_umull node:$MHS, node:$RHS))>>;
4337defm UMULL   : SIMDLongThreeVectorBHS<1, 0b1100, "umull", int_aarch64_neon_umull>;
4338defm USUBL   : SIMDLongThreeVectorBHS<1, 0b0010, "usubl",
4339                 BinOpFrag<(sub (zext node:$LHS), (zext node:$RHS))>>;
4340defm USUBW   : SIMDWideThreeVectorBHS<   1, 0b0011, "usubw",
4341                 BinOpFrag<(sub node:$LHS, (zext node:$RHS))>>;
4342
4343// Additional patterns for SMULL and UMULL
4344multiclass Neon_mul_widen_patterns<SDPatternOperator opnode,
4345  Instruction INST8B, Instruction INST4H, Instruction INST2S> {
4346  def : Pat<(v8i16 (opnode (v8i8 V64:$Rn), (v8i8 V64:$Rm))),
4347            (INST8B V64:$Rn, V64:$Rm)>;
4348  def : Pat<(v4i32 (opnode (v4i16 V64:$Rn), (v4i16 V64:$Rm))),
4349            (INST4H V64:$Rn, V64:$Rm)>;
4350  def : Pat<(v2i64 (opnode (v2i32 V64:$Rn), (v2i32 V64:$Rm))),
4351            (INST2S V64:$Rn, V64:$Rm)>;
4352}
4353
4354defm : Neon_mul_widen_patterns<AArch64smull, SMULLv8i8_v8i16,
4355  SMULLv4i16_v4i32, SMULLv2i32_v2i64>;
4356defm : Neon_mul_widen_patterns<AArch64umull, UMULLv8i8_v8i16,
4357  UMULLv4i16_v4i32, UMULLv2i32_v2i64>;
4358
4359// Additional patterns for SMLAL/SMLSL and UMLAL/UMLSL
4360multiclass Neon_mulacc_widen_patterns<SDPatternOperator opnode,
4361  Instruction INST8B, Instruction INST4H, Instruction INST2S> {
4362  def : Pat<(v8i16 (opnode (v8i16 V128:$Rd), (v8i8 V64:$Rn), (v8i8 V64:$Rm))),
4363            (INST8B V128:$Rd, V64:$Rn, V64:$Rm)>;
4364  def : Pat<(v4i32 (opnode (v4i32 V128:$Rd), (v4i16 V64:$Rn), (v4i16 V64:$Rm))),
4365            (INST4H V128:$Rd, V64:$Rn, V64:$Rm)>;
4366  def : Pat<(v2i64 (opnode (v2i64 V128:$Rd), (v2i32 V64:$Rn), (v2i32 V64:$Rm))),
4367            (INST2S  V128:$Rd, V64:$Rn, V64:$Rm)>;
4368}
4369
4370defm : Neon_mulacc_widen_patterns<
4371  TriOpFrag<(add node:$LHS, (AArch64smull node:$MHS, node:$RHS))>,
4372  SMLALv8i8_v8i16, SMLALv4i16_v4i32, SMLALv2i32_v2i64>;
4373defm : Neon_mulacc_widen_patterns<
4374  TriOpFrag<(add node:$LHS, (AArch64umull node:$MHS, node:$RHS))>,
4375  UMLALv8i8_v8i16, UMLALv4i16_v4i32, UMLALv2i32_v2i64>;
4376defm : Neon_mulacc_widen_patterns<
4377  TriOpFrag<(sub node:$LHS, (AArch64smull node:$MHS, node:$RHS))>,
4378  SMLSLv8i8_v8i16, SMLSLv4i16_v4i32, SMLSLv2i32_v2i64>;
4379defm : Neon_mulacc_widen_patterns<
4380  TriOpFrag<(sub node:$LHS, (AArch64umull node:$MHS, node:$RHS))>,
4381  UMLSLv8i8_v8i16, UMLSLv4i16_v4i32, UMLSLv2i32_v2i64>;
4382
4383// Patterns for 64-bit pmull
4384def : Pat<(int_aarch64_neon_pmull64 V64:$Rn, V64:$Rm),
4385          (PMULLv1i64 V64:$Rn, V64:$Rm)>;
4386def : Pat<(int_aarch64_neon_pmull64 (extractelt (v2i64 V128:$Rn), (i64 1)),
4387                                    (extractelt (v2i64 V128:$Rm), (i64 1))),
4388          (PMULLv2i64 V128:$Rn, V128:$Rm)>;
4389
4390// CodeGen patterns for addhn and subhn instructions, which can actually be
4391// written in LLVM IR without too much difficulty.
4392
4393// ADDHN
4394def : Pat<(v8i8 (trunc (v8i16 (AArch64vlshr (add V128:$Rn, V128:$Rm), (i32 8))))),
4395          (ADDHNv8i16_v8i8 V128:$Rn, V128:$Rm)>;
4396def : Pat<(v4i16 (trunc (v4i32 (AArch64vlshr (add V128:$Rn, V128:$Rm),
4397                                           (i32 16))))),
4398          (ADDHNv4i32_v4i16 V128:$Rn, V128:$Rm)>;
4399def : Pat<(v2i32 (trunc (v2i64 (AArch64vlshr (add V128:$Rn, V128:$Rm),
4400                                           (i32 32))))),
4401          (ADDHNv2i64_v2i32 V128:$Rn, V128:$Rm)>;
4402def : Pat<(concat_vectors (v8i8 V64:$Rd),
4403                          (trunc (v8i16 (AArch64vlshr (add V128:$Rn, V128:$Rm),
4404                                                    (i32 8))))),
4405          (ADDHNv8i16_v16i8 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
4406                            V128:$Rn, V128:$Rm)>;
4407def : Pat<(concat_vectors (v4i16 V64:$Rd),
4408                          (trunc (v4i32 (AArch64vlshr (add V128:$Rn, V128:$Rm),
4409                                                    (i32 16))))),
4410          (ADDHNv4i32_v8i16 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
4411                            V128:$Rn, V128:$Rm)>;
4412def : Pat<(concat_vectors (v2i32 V64:$Rd),
4413                          (trunc (v2i64 (AArch64vlshr (add V128:$Rn, V128:$Rm),
4414                                                    (i32 32))))),
4415          (ADDHNv2i64_v4i32 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
4416                            V128:$Rn, V128:$Rm)>;
4417
4418// SUBHN
4419def : Pat<(v8i8 (trunc (v8i16 (AArch64vlshr (sub V128:$Rn, V128:$Rm), (i32 8))))),
4420          (SUBHNv8i16_v8i8 V128:$Rn, V128:$Rm)>;
4421def : Pat<(v4i16 (trunc (v4i32 (AArch64vlshr (sub V128:$Rn, V128:$Rm),
4422                                           (i32 16))))),
4423          (SUBHNv4i32_v4i16 V128:$Rn, V128:$Rm)>;
4424def : Pat<(v2i32 (trunc (v2i64 (AArch64vlshr (sub V128:$Rn, V128:$Rm),
4425                                           (i32 32))))),
4426          (SUBHNv2i64_v2i32 V128:$Rn, V128:$Rm)>;
4427def : Pat<(concat_vectors (v8i8 V64:$Rd),
4428                          (trunc (v8i16 (AArch64vlshr (sub V128:$Rn, V128:$Rm),
4429                                                    (i32 8))))),
4430          (SUBHNv8i16_v16i8 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
4431                            V128:$Rn, V128:$Rm)>;
4432def : Pat<(concat_vectors (v4i16 V64:$Rd),
4433                          (trunc (v4i32 (AArch64vlshr (sub V128:$Rn, V128:$Rm),
4434                                                    (i32 16))))),
4435          (SUBHNv4i32_v8i16 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
4436                            V128:$Rn, V128:$Rm)>;
4437def : Pat<(concat_vectors (v2i32 V64:$Rd),
4438                          (trunc (v2i64 (AArch64vlshr (sub V128:$Rn, V128:$Rm),
4439                                                    (i32 32))))),
4440          (SUBHNv2i64_v4i32 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
4441                            V128:$Rn, V128:$Rm)>;
4442
4443//----------------------------------------------------------------------------
4444// AdvSIMD bitwise extract from vector instruction.
4445//----------------------------------------------------------------------------
4446
4447defm EXT : SIMDBitwiseExtract<"ext">;
4448
4449def AdjustExtImm : SDNodeXForm<imm, [{
4450  return CurDAG->getTargetConstant(8 + N->getZExtValue(), SDLoc(N), MVT::i32);
4451}]>;
4452multiclass ExtPat<ValueType VT64, ValueType VT128, int N> {
4453  def : Pat<(VT64 (AArch64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
4454            (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
4455  def : Pat<(VT128 (AArch64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
4456            (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
4457  // We use EXT to handle extract_subvector to copy the upper 64-bits of a
4458  // 128-bit vector.
4459  def : Pat<(VT64 (extract_subvector V128:$Rn, (i64 N))),
4460            (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
4461  // A 64-bit EXT of two halves of the same 128-bit register can be done as a
4462  // single 128-bit EXT.
4463  def : Pat<(VT64 (AArch64ext (extract_subvector V128:$Rn, (i64 0)),
4464                              (extract_subvector V128:$Rn, (i64 N)),
4465                              (i32 imm:$imm))),
4466            (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, imm:$imm), dsub)>;
4467  // A 64-bit EXT of the high half of a 128-bit register can be done using a
4468  // 128-bit EXT of the whole register with an adjustment to the immediate. The
4469  // top half of the other operand will be unset, but that doesn't matter as it
4470  // will not be used.
4471  def : Pat<(VT64 (AArch64ext (extract_subvector V128:$Rn, (i64 N)),
4472                              V64:$Rm,
4473                              (i32 imm:$imm))),
4474            (EXTRACT_SUBREG (EXTv16i8 V128:$Rn,
4475                                      (SUBREG_TO_REG (i32 0), V64:$Rm, dsub),
4476                                      (AdjustExtImm imm:$imm)), dsub)>;
4477}
4478
4479defm : ExtPat<v8i8, v16i8, 8>;
4480defm : ExtPat<v4i16, v8i16, 4>;
4481defm : ExtPat<v4f16, v8f16, 4>;
4482defm : ExtPat<v2i32, v4i32, 2>;
4483defm : ExtPat<v2f32, v4f32, 2>;
4484defm : ExtPat<v1i64, v2i64, 1>;
4485defm : ExtPat<v1f64, v2f64, 1>;
4486
4487//----------------------------------------------------------------------------
4488// AdvSIMD zip vector
4489//----------------------------------------------------------------------------
4490
4491defm TRN1 : SIMDZipVector<0b010, "trn1", AArch64trn1>;
4492defm TRN2 : SIMDZipVector<0b110, "trn2", AArch64trn2>;
4493defm UZP1 : SIMDZipVector<0b001, "uzp1", AArch64uzp1>;
4494defm UZP2 : SIMDZipVector<0b101, "uzp2", AArch64uzp2>;
4495defm ZIP1 : SIMDZipVector<0b011, "zip1", AArch64zip1>;
4496defm ZIP2 : SIMDZipVector<0b111, "zip2", AArch64zip2>;
4497
4498//----------------------------------------------------------------------------
4499// AdvSIMD TBL/TBX instructions
4500//----------------------------------------------------------------------------
4501
4502defm TBL : SIMDTableLookup<    0, "tbl">;
4503defm TBX : SIMDTableLookupTied<1, "tbx">;
4504
4505def : Pat<(v8i8 (int_aarch64_neon_tbl1 (v16i8 VecListOne128:$Rn), (v8i8 V64:$Ri))),
4506          (TBLv8i8One VecListOne128:$Rn, V64:$Ri)>;
4507def : Pat<(v16i8 (int_aarch64_neon_tbl1 (v16i8 V128:$Ri), (v16i8 V128:$Rn))),
4508          (TBLv16i8One V128:$Ri, V128:$Rn)>;
4509
4510def : Pat<(v8i8 (int_aarch64_neon_tbx1 (v8i8 V64:$Rd),
4511                  (v16i8 VecListOne128:$Rn), (v8i8 V64:$Ri))),
4512          (TBXv8i8One V64:$Rd, VecListOne128:$Rn, V64:$Ri)>;
4513def : Pat<(v16i8 (int_aarch64_neon_tbx1 (v16i8 V128:$Rd),
4514                   (v16i8 V128:$Ri), (v16i8 V128:$Rn))),
4515          (TBXv16i8One V128:$Rd, V128:$Ri, V128:$Rn)>;
4516
4517
4518//----------------------------------------------------------------------------
4519// AdvSIMD scalar CPY instruction
4520//----------------------------------------------------------------------------
4521
4522defm CPY : SIMDScalarCPY<"cpy">;
4523
4524//----------------------------------------------------------------------------
4525// AdvSIMD scalar pairwise instructions
4526//----------------------------------------------------------------------------
4527
4528defm ADDP    : SIMDPairwiseScalarD<0, 0b11011, "addp">;
4529defm FADDP   : SIMDFPPairwiseScalar<0, 0b01101, "faddp">;
4530defm FMAXNMP : SIMDFPPairwiseScalar<0, 0b01100, "fmaxnmp">;
4531defm FMAXP   : SIMDFPPairwiseScalar<0, 0b01111, "fmaxp">;
4532defm FMINNMP : SIMDFPPairwiseScalar<1, 0b01100, "fminnmp">;
4533defm FMINP   : SIMDFPPairwiseScalar<1, 0b01111, "fminp">;
4534def : Pat<(v2i64 (AArch64saddv V128:$Rn)),
4535          (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), (ADDPv2i64p V128:$Rn), dsub)>;
4536def : Pat<(v2i64 (AArch64uaddv V128:$Rn)),
4537          (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), (ADDPv2i64p V128:$Rn), dsub)>;
4538def : Pat<(f32 (int_aarch64_neon_faddv (v2f32 V64:$Rn))),
4539          (FADDPv2i32p V64:$Rn)>;
4540def : Pat<(f32 (int_aarch64_neon_faddv (v4f32 V128:$Rn))),
4541          (FADDPv2i32p (EXTRACT_SUBREG (FADDPv4f32 V128:$Rn, V128:$Rn), dsub))>;
4542def : Pat<(f64 (int_aarch64_neon_faddv (v2f64 V128:$Rn))),
4543          (FADDPv2i64p V128:$Rn)>;
4544def : Pat<(f32 (int_aarch64_neon_fmaxnmv (v2f32 V64:$Rn))),
4545          (FMAXNMPv2i32p V64:$Rn)>;
4546def : Pat<(f64 (int_aarch64_neon_fmaxnmv (v2f64 V128:$Rn))),
4547          (FMAXNMPv2i64p V128:$Rn)>;
4548def : Pat<(f32 (int_aarch64_neon_fmaxv (v2f32 V64:$Rn))),
4549          (FMAXPv2i32p V64:$Rn)>;
4550def : Pat<(f64 (int_aarch64_neon_fmaxv (v2f64 V128:$Rn))),
4551          (FMAXPv2i64p V128:$Rn)>;
4552def : Pat<(f32 (int_aarch64_neon_fminnmv (v2f32 V64:$Rn))),
4553          (FMINNMPv2i32p V64:$Rn)>;
4554def : Pat<(f64 (int_aarch64_neon_fminnmv (v2f64 V128:$Rn))),
4555          (FMINNMPv2i64p V128:$Rn)>;
4556def : Pat<(f32 (int_aarch64_neon_fminv (v2f32 V64:$Rn))),
4557          (FMINPv2i32p V64:$Rn)>;
4558def : Pat<(f64 (int_aarch64_neon_fminv (v2f64 V128:$Rn))),
4559          (FMINPv2i64p V128:$Rn)>;
4560
4561//----------------------------------------------------------------------------
4562// AdvSIMD INS/DUP instructions
4563//----------------------------------------------------------------------------
4564
4565def DUPv8i8gpr  : SIMDDupFromMain<0, {?,?,?,?,1}, ".8b", v8i8, V64, GPR32>;
4566def DUPv16i8gpr : SIMDDupFromMain<1, {?,?,?,?,1}, ".16b", v16i8, V128, GPR32>;
4567def DUPv4i16gpr : SIMDDupFromMain<0, {?,?,?,1,0}, ".4h", v4i16, V64, GPR32>;
4568def DUPv8i16gpr : SIMDDupFromMain<1, {?,?,?,1,0}, ".8h", v8i16, V128, GPR32>;
4569def DUPv2i32gpr : SIMDDupFromMain<0, {?,?,1,0,0}, ".2s", v2i32, V64, GPR32>;
4570def DUPv4i32gpr : SIMDDupFromMain<1, {?,?,1,0,0}, ".4s", v4i32, V128, GPR32>;
4571def DUPv2i64gpr : SIMDDupFromMain<1, {?,1,0,0,0}, ".2d", v2i64, V128, GPR64>;
4572
4573def DUPv2i64lane : SIMDDup64FromElement;
4574def DUPv2i32lane : SIMDDup32FromElement<0, ".2s", v2i32, V64>;
4575def DUPv4i32lane : SIMDDup32FromElement<1, ".4s", v4i32, V128>;
4576def DUPv4i16lane : SIMDDup16FromElement<0, ".4h", v4i16, V64>;
4577def DUPv8i16lane : SIMDDup16FromElement<1, ".8h", v8i16, V128>;
4578def DUPv8i8lane  : SIMDDup8FromElement <0, ".8b", v8i8, V64>;
4579def DUPv16i8lane : SIMDDup8FromElement <1, ".16b", v16i8, V128>;
4580
4581// DUP from a 64-bit register to a 64-bit register is just a copy
4582def : Pat<(v1i64 (AArch64dup (i64 GPR64:$Rn))),
4583          (COPY_TO_REGCLASS GPR64:$Rn, FPR64)>;
4584def : Pat<(v1f64 (AArch64dup (f64 FPR64:$Rn))),
4585          (COPY_TO_REGCLASS FPR64:$Rn, FPR64)>;
4586
4587def : Pat<(v2f32 (AArch64dup (f32 FPR32:$Rn))),
4588          (v2f32 (DUPv2i32lane
4589            (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rn, ssub),
4590            (i64 0)))>;
4591def : Pat<(v4f32 (AArch64dup (f32 FPR32:$Rn))),
4592          (v4f32 (DUPv4i32lane
4593            (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rn, ssub),
4594            (i64 0)))>;
4595def : Pat<(v2f64 (AArch64dup (f64 FPR64:$Rn))),
4596          (v2f64 (DUPv2i64lane
4597            (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$Rn, dsub),
4598            (i64 0)))>;
4599def : Pat<(v4f16 (AArch64dup (f16 FPR16:$Rn))),
4600          (v4f16 (DUPv4i16lane
4601            (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), FPR16:$Rn, hsub),
4602            (i64 0)))>;
4603def : Pat<(v8f16 (AArch64dup (f16 FPR16:$Rn))),
4604          (v8f16 (DUPv8i16lane
4605            (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), FPR16:$Rn, hsub),
4606            (i64 0)))>;
4607
4608def : Pat<(v4f16 (AArch64duplane16 (v8f16 V128:$Rn), VectorIndexH:$imm)),
4609          (DUPv4i16lane V128:$Rn, VectorIndexH:$imm)>;
4610def : Pat<(v8f16 (AArch64duplane16 (v8f16 V128:$Rn), VectorIndexH:$imm)),
4611          (DUPv8i16lane V128:$Rn, VectorIndexH:$imm)>;
4612
4613def : Pat<(v2f32 (AArch64duplane32 (v4f32 V128:$Rn), VectorIndexS:$imm)),
4614          (DUPv2i32lane V128:$Rn, VectorIndexS:$imm)>;
4615def : Pat<(v4f32 (AArch64duplane32 (v4f32 V128:$Rn), VectorIndexS:$imm)),
4616         (DUPv4i32lane V128:$Rn, VectorIndexS:$imm)>;
4617def : Pat<(v2f64 (AArch64duplane64 (v2f64 V128:$Rn), VectorIndexD:$imm)),
4618          (DUPv2i64lane V128:$Rn, VectorIndexD:$imm)>;
4619
4620// If there's an (AArch64dup (vector_extract ...) ...), we can use a duplane
4621// instruction even if the types don't match: we just have to remap the lane
4622// carefully. N.b. this trick only applies to truncations.
4623def VecIndex_x2 : SDNodeXForm<imm, [{
4624  return CurDAG->getTargetConstant(2 * N->getZExtValue(), SDLoc(N), MVT::i64);
4625}]>;
4626def VecIndex_x4 : SDNodeXForm<imm, [{
4627  return CurDAG->getTargetConstant(4 * N->getZExtValue(), SDLoc(N), MVT::i64);
4628}]>;
4629def VecIndex_x8 : SDNodeXForm<imm, [{
4630  return CurDAG->getTargetConstant(8 * N->getZExtValue(), SDLoc(N), MVT::i64);
4631}]>;
4632
4633multiclass DUPWithTruncPats<ValueType ResVT, ValueType Src64VT,
4634                            ValueType Src128VT, ValueType ScalVT,
4635                            Instruction DUP, SDNodeXForm IdxXFORM> {
4636  def : Pat<(ResVT (AArch64dup (ScalVT (vector_extract (Src128VT V128:$Rn),
4637                                                     imm:$idx)))),
4638            (DUP V128:$Rn, (IdxXFORM imm:$idx))>;
4639
4640  def : Pat<(ResVT (AArch64dup (ScalVT (vector_extract (Src64VT V64:$Rn),
4641                                                     imm:$idx)))),
4642            (DUP (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), (IdxXFORM imm:$idx))>;
4643}
4644
4645defm : DUPWithTruncPats<v8i8,   v4i16, v8i16, i32, DUPv8i8lane,  VecIndex_x2>;
4646defm : DUPWithTruncPats<v8i8,   v2i32, v4i32, i32, DUPv8i8lane,  VecIndex_x4>;
4647defm : DUPWithTruncPats<v4i16,  v2i32, v4i32, i32, DUPv4i16lane, VecIndex_x2>;
4648
4649defm : DUPWithTruncPats<v16i8,  v4i16, v8i16, i32, DUPv16i8lane, VecIndex_x2>;
4650defm : DUPWithTruncPats<v16i8,  v2i32, v4i32, i32, DUPv16i8lane, VecIndex_x4>;
4651defm : DUPWithTruncPats<v8i16,  v2i32, v4i32, i32, DUPv8i16lane, VecIndex_x2>;
4652
4653multiclass DUPWithTrunci64Pats<ValueType ResVT, Instruction DUP,
4654                               SDNodeXForm IdxXFORM> {
4655  def : Pat<(ResVT (AArch64dup (i32 (trunc (extractelt (v2i64 V128:$Rn),
4656                                                         imm:$idx))))),
4657            (DUP V128:$Rn, (IdxXFORM imm:$idx))>;
4658
4659  def : Pat<(ResVT (AArch64dup (i32 (trunc (extractelt (v1i64 V64:$Rn),
4660                                                       imm:$idx))))),
4661            (DUP (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), (IdxXFORM imm:$idx))>;
4662}
4663
4664defm : DUPWithTrunci64Pats<v8i8,  DUPv8i8lane,   VecIndex_x8>;
4665defm : DUPWithTrunci64Pats<v4i16, DUPv4i16lane,  VecIndex_x4>;
4666defm : DUPWithTrunci64Pats<v2i32, DUPv2i32lane,  VecIndex_x2>;
4667
4668defm : DUPWithTrunci64Pats<v16i8, DUPv16i8lane, VecIndex_x8>;
4669defm : DUPWithTrunci64Pats<v8i16, DUPv8i16lane, VecIndex_x4>;
4670defm : DUPWithTrunci64Pats<v4i32, DUPv4i32lane, VecIndex_x2>;
4671
4672// SMOV and UMOV definitions, with some extra patterns for convenience
4673defm SMOV : SMov;
4674defm UMOV : UMov;
4675
4676def : Pat<(sext_inreg (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx), i8),
4677          (i32 (SMOVvi8to32 V128:$Rn, VectorIndexB:$idx))>;
4678def : Pat<(sext_inreg (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx), i8),
4679          (i64 (SMOVvi8to64 V128:$Rn, VectorIndexB:$idx))>;
4680def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
4681          (i32 (SMOVvi16to32 V128:$Rn, VectorIndexH:$idx))>;
4682def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
4683          (i64 (SMOVvi16to64 V128:$Rn, VectorIndexH:$idx))>;
4684def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
4685          (i32 (SMOVvi16to32 V128:$Rn, VectorIndexH:$idx))>;
4686def : Pat<(sext (i32 (vector_extract (v4i32 V128:$Rn), VectorIndexS:$idx))),
4687          (i64 (SMOVvi32to64 V128:$Rn, VectorIndexS:$idx))>;
4688
4689def : Pat<(sext_inreg (i64 (anyext (i32 (vector_extract (v16i8 V128:$Rn),
4690            VectorIndexB:$idx)))), i8),
4691          (i64 (SMOVvi8to64 V128:$Rn, VectorIndexB:$idx))>;
4692def : Pat<(sext_inreg (i64 (anyext (i32 (vector_extract (v8i16 V128:$Rn),
4693            VectorIndexH:$idx)))), i16),
4694          (i64 (SMOVvi16to64 V128:$Rn, VectorIndexH:$idx))>;
4695
4696// Extracting i8 or i16 elements will have the zero-extend transformed to
4697// an 'and' mask by type legalization since neither i8 nor i16 are legal types
4698// for AArch64. Match these patterns here since UMOV already zeroes out the high
4699// bits of the destination register.
4700def : Pat<(and (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx),
4701               (i32 0xff)),
4702          (i32 (UMOVvi8 V128:$Rn, VectorIndexB:$idx))>;
4703def : Pat<(and (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),
4704               (i32 0xffff)),
4705          (i32 (UMOVvi16 V128:$Rn, VectorIndexH:$idx))>;
4706
4707defm INS : SIMDIns;
4708
4709def : Pat<(v16i8 (scalar_to_vector GPR32:$Rn)),
4710          (SUBREG_TO_REG (i32 0),
4711                         (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
4712def : Pat<(v8i8 (scalar_to_vector GPR32:$Rn)),
4713          (SUBREG_TO_REG (i32 0),
4714                         (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
4715
4716def : Pat<(v8i16 (scalar_to_vector GPR32:$Rn)),
4717          (SUBREG_TO_REG (i32 0),
4718                         (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
4719def : Pat<(v4i16 (scalar_to_vector GPR32:$Rn)),
4720          (SUBREG_TO_REG (i32 0),
4721                         (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
4722
4723def : Pat<(v4f16 (scalar_to_vector (f16 FPR16:$Rn))),
4724          (INSERT_SUBREG (v4f16 (IMPLICIT_DEF)), FPR16:$Rn, hsub)>;
4725def : Pat<(v8f16 (scalar_to_vector (f16 FPR16:$Rn))),
4726          (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), FPR16:$Rn, hsub)>;
4727
4728def : Pat<(v2i32 (scalar_to_vector (i32 FPR32:$Rn))),
4729            (v2i32 (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)),
4730                                  (i32 FPR32:$Rn), ssub))>;
4731def : Pat<(v4i32 (scalar_to_vector (i32 FPR32:$Rn))),
4732            (v4i32 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
4733                                  (i32 FPR32:$Rn), ssub))>;
4734
4735def : Pat<(v2i64 (scalar_to_vector (i64 FPR64:$Rn))),
4736            (v2i64 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
4737                                  (i64 FPR64:$Rn), dsub))>;
4738
4739def : Pat<(v4f16 (scalar_to_vector (f16 FPR16:$Rn))),
4740          (INSERT_SUBREG (v4f16 (IMPLICIT_DEF)), FPR16:$Rn, hsub)>;
4741def : Pat<(v8f16 (scalar_to_vector (f16 FPR16:$Rn))),
4742          (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), FPR16:$Rn, hsub)>;
4743
4744def : Pat<(v4f32 (scalar_to_vector (f32 FPR32:$Rn))),
4745          (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rn, ssub)>;
4746def : Pat<(v2f32 (scalar_to_vector (f32 FPR32:$Rn))),
4747          (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), FPR32:$Rn, ssub)>;
4748
4749def : Pat<(v2f64 (scalar_to_vector (f64 FPR64:$Rn))),
4750          (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$Rn, dsub)>;
4751
4752def : Pat<(v4f16 (vector_insert (v4f16 V64:$Rn),
4753            (f16 FPR16:$Rm), (i64 VectorIndexS:$imm))),
4754          (EXTRACT_SUBREG
4755            (INSvi16lane
4756              (v8f16 (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), V64:$Rn, dsub)),
4757              VectorIndexS:$imm,
4758              (v8f16 (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), FPR16:$Rm, hsub)),
4759              (i64 0)),
4760            dsub)>;
4761
4762def : Pat<(v8f16 (vector_insert (v8f16 V128:$Rn),
4763            (f16 FPR16:$Rm), (i64 VectorIndexH:$imm))),
4764          (INSvi16lane
4765            V128:$Rn, VectorIndexH:$imm,
4766            (v8f16 (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), FPR16:$Rm, hsub)),
4767            (i64 0))>;
4768
4769def : Pat<(v2f32 (vector_insert (v2f32 V64:$Rn),
4770            (f32 FPR32:$Rm), (i64 VectorIndexS:$imm))),
4771          (EXTRACT_SUBREG
4772            (INSvi32lane
4773              (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), V64:$Rn, dsub)),
4774              VectorIndexS:$imm,
4775              (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rm, ssub)),
4776              (i64 0)),
4777            dsub)>;
4778def : Pat<(v4f32 (vector_insert (v4f32 V128:$Rn),
4779            (f32 FPR32:$Rm), (i64 VectorIndexS:$imm))),
4780          (INSvi32lane
4781            V128:$Rn, VectorIndexS:$imm,
4782            (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rm, ssub)),
4783            (i64 0))>;
4784def : Pat<(v2f64 (vector_insert (v2f64 V128:$Rn),
4785            (f64 FPR64:$Rm), (i64 VectorIndexD:$imm))),
4786          (INSvi64lane
4787            V128:$Rn, VectorIndexD:$imm,
4788            (v2f64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$Rm, dsub)),
4789            (i64 0))>;
4790
4791// Copy an element at a constant index in one vector into a constant indexed
4792// element of another.
4793// FIXME refactor to a shared class/dev parameterized on vector type, vector
4794// index type and INS extension
4795def : Pat<(v16i8 (int_aarch64_neon_vcopy_lane
4796                   (v16i8 V128:$Vd), VectorIndexB:$idx, (v16i8 V128:$Vs),
4797                   VectorIndexB:$idx2)),
4798          (v16i8 (INSvi8lane
4799                   V128:$Vd, VectorIndexB:$idx, V128:$Vs, VectorIndexB:$idx2)
4800          )>;
4801def : Pat<(v8i16 (int_aarch64_neon_vcopy_lane
4802                   (v8i16 V128:$Vd), VectorIndexH:$idx, (v8i16 V128:$Vs),
4803                   VectorIndexH:$idx2)),
4804          (v8i16 (INSvi16lane
4805                   V128:$Vd, VectorIndexH:$idx, V128:$Vs, VectorIndexH:$idx2)
4806          )>;
4807def : Pat<(v4i32 (int_aarch64_neon_vcopy_lane
4808                   (v4i32 V128:$Vd), VectorIndexS:$idx, (v4i32 V128:$Vs),
4809                   VectorIndexS:$idx2)),
4810          (v4i32 (INSvi32lane
4811                   V128:$Vd, VectorIndexS:$idx, V128:$Vs, VectorIndexS:$idx2)
4812          )>;
4813def : Pat<(v2i64 (int_aarch64_neon_vcopy_lane
4814                   (v2i64 V128:$Vd), VectorIndexD:$idx, (v2i64 V128:$Vs),
4815                   VectorIndexD:$idx2)),
4816          (v2i64 (INSvi64lane
4817                   V128:$Vd, VectorIndexD:$idx, V128:$Vs, VectorIndexD:$idx2)
4818          )>;
4819
4820multiclass Neon_INS_elt_pattern<ValueType VT128, ValueType VT64,
4821                                ValueType VTScal, Instruction INS> {
4822  def : Pat<(VT128 (vector_insert V128:$src,
4823                        (VTScal (vector_extract (VT128 V128:$Rn), imm:$Immn)),
4824                        imm:$Immd)),
4825            (INS V128:$src, imm:$Immd, V128:$Rn, imm:$Immn)>;
4826
4827  def : Pat<(VT128 (vector_insert V128:$src,
4828                        (VTScal (vector_extract (VT64 V64:$Rn), imm:$Immn)),
4829                        imm:$Immd)),
4830            (INS V128:$src, imm:$Immd,
4831                 (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), imm:$Immn)>;
4832
4833  def : Pat<(VT64 (vector_insert V64:$src,
4834                        (VTScal (vector_extract (VT128 V128:$Rn), imm:$Immn)),
4835                        imm:$Immd)),
4836            (EXTRACT_SUBREG (INS (SUBREG_TO_REG (i64 0), V64:$src, dsub),
4837                                 imm:$Immd, V128:$Rn, imm:$Immn),
4838                            dsub)>;
4839
4840  def : Pat<(VT64 (vector_insert V64:$src,
4841                        (VTScal (vector_extract (VT64 V64:$Rn), imm:$Immn)),
4842                        imm:$Immd)),
4843            (EXTRACT_SUBREG
4844                (INS (SUBREG_TO_REG (i64 0), V64:$src, dsub), imm:$Immd,
4845                     (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), imm:$Immn),
4846                dsub)>;
4847}
4848
4849defm : Neon_INS_elt_pattern<v8f16, v4f16, f16, INSvi16lane>;
4850defm : Neon_INS_elt_pattern<v4f32, v2f32, f32, INSvi32lane>;
4851defm : Neon_INS_elt_pattern<v2f64, v1f64, f64, INSvi64lane>;
4852
4853
4854// Floating point vector extractions are codegen'd as either a sequence of
4855// subregister extractions, or a MOV (aka CPY here, alias for DUP) if
4856// the lane number is anything other than zero.
4857def : Pat<(vector_extract (v2f64 V128:$Rn), 0),
4858          (f64 (EXTRACT_SUBREG V128:$Rn, dsub))>;
4859def : Pat<(vector_extract (v4f32 V128:$Rn), 0),
4860          (f32 (EXTRACT_SUBREG V128:$Rn, ssub))>;
4861def : Pat<(vector_extract (v8f16 V128:$Rn), 0),
4862          (f16 (EXTRACT_SUBREG V128:$Rn, hsub))>;
4863
4864def : Pat<(vector_extract (v2f64 V128:$Rn), VectorIndexD:$idx),
4865          (f64 (CPYi64 V128:$Rn, VectorIndexD:$idx))>;
4866def : Pat<(vector_extract (v4f32 V128:$Rn), VectorIndexS:$idx),
4867          (f32 (CPYi32 V128:$Rn, VectorIndexS:$idx))>;
4868def : Pat<(vector_extract (v8f16 V128:$Rn), VectorIndexH:$idx),
4869          (f16 (CPYi16 V128:$Rn, VectorIndexH:$idx))>;
4870
4871// All concat_vectors operations are canonicalised to act on i64 vectors for
4872// AArch64. In the general case we need an instruction, which had just as well be
4873// INS.
4874class ConcatPat<ValueType DstTy, ValueType SrcTy>
4875  : Pat<(DstTy (concat_vectors (SrcTy V64:$Rd), V64:$Rn)),
4876        (INSvi64lane (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), 1,
4877                     (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rn, dsub), 0)>;
4878
4879def : ConcatPat<v2i64, v1i64>;
4880def : ConcatPat<v2f64, v1f64>;
4881def : ConcatPat<v4i32, v2i32>;
4882def : ConcatPat<v4f32, v2f32>;
4883def : ConcatPat<v8i16, v4i16>;
4884def : ConcatPat<v8f16, v4f16>;
4885def : ConcatPat<v16i8, v8i8>;
4886
4887// If the high lanes are undef, though, we can just ignore them:
4888class ConcatUndefPat<ValueType DstTy, ValueType SrcTy>
4889  : Pat<(DstTy (concat_vectors (SrcTy V64:$Rn), undef)),
4890        (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rn, dsub)>;
4891
4892def : ConcatUndefPat<v2i64, v1i64>;
4893def : ConcatUndefPat<v2f64, v1f64>;
4894def : ConcatUndefPat<v4i32, v2i32>;
4895def : ConcatUndefPat<v4f32, v2f32>;
4896def : ConcatUndefPat<v8i16, v4i16>;
4897def : ConcatUndefPat<v16i8, v8i8>;
4898
4899//----------------------------------------------------------------------------
4900// AdvSIMD across lanes instructions
4901//----------------------------------------------------------------------------
4902
4903defm ADDV    : SIMDAcrossLanesBHS<0, 0b11011, "addv">;
4904defm SMAXV   : SIMDAcrossLanesBHS<0, 0b01010, "smaxv">;
4905defm SMINV   : SIMDAcrossLanesBHS<0, 0b11010, "sminv">;
4906defm UMAXV   : SIMDAcrossLanesBHS<1, 0b01010, "umaxv">;
4907defm UMINV   : SIMDAcrossLanesBHS<1, 0b11010, "uminv">;
4908defm SADDLV  : SIMDAcrossLanesHSD<0, 0b00011, "saddlv">;
4909defm UADDLV  : SIMDAcrossLanesHSD<1, 0b00011, "uaddlv">;
4910defm FMAXNMV : SIMDFPAcrossLanes<0b01100, 0, "fmaxnmv", int_aarch64_neon_fmaxnmv>;
4911defm FMAXV   : SIMDFPAcrossLanes<0b01111, 0, "fmaxv", int_aarch64_neon_fmaxv>;
4912defm FMINNMV : SIMDFPAcrossLanes<0b01100, 1, "fminnmv", int_aarch64_neon_fminnmv>;
4913defm FMINV   : SIMDFPAcrossLanes<0b01111, 1, "fminv", int_aarch64_neon_fminv>;
4914
4915// Patterns for across-vector intrinsics, that have a node equivalent, that
4916// returns a vector (with only the low lane defined) instead of a scalar.
4917// In effect, opNode is the same as (scalar_to_vector (IntNode)).
4918multiclass SIMDAcrossLanesIntrinsic<string baseOpc,
4919                                    SDPatternOperator opNode> {
4920// If a lane instruction caught the vector_extract around opNode, we can
4921// directly match the latter to the instruction.
4922def : Pat<(v8i8 (opNode V64:$Rn)),
4923          (INSERT_SUBREG (v8i8 (IMPLICIT_DEF)),
4924           (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub)>;
4925def : Pat<(v16i8 (opNode V128:$Rn)),
4926          (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4927           (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub)>;
4928def : Pat<(v4i16 (opNode V64:$Rn)),
4929          (INSERT_SUBREG (v4i16 (IMPLICIT_DEF)),
4930           (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub)>;
4931def : Pat<(v8i16 (opNode V128:$Rn)),
4932          (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
4933           (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub)>;
4934def : Pat<(v4i32 (opNode V128:$Rn)),
4935          (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
4936           (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), ssub)>;
4937
4938
4939// If none did, fallback to the explicit patterns, consuming the vector_extract.
4940def : Pat<(i32 (vector_extract (insert_subvector undef, (v8i8 (opNode V64:$Rn)),
4941            (i32 0)), (i64 0))),
4942          (EXTRACT_SUBREG (INSERT_SUBREG (v8i8 (IMPLICIT_DEF)),
4943            (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn),
4944            bsub), ssub)>;
4945def : Pat<(i32 (vector_extract (v16i8 (opNode V128:$Rn)), (i64 0))),
4946          (EXTRACT_SUBREG (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4947            (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn),
4948            bsub), ssub)>;
4949def : Pat<(i32 (vector_extract (insert_subvector undef,
4950            (v4i16 (opNode V64:$Rn)), (i32 0)), (i64 0))),
4951          (EXTRACT_SUBREG (INSERT_SUBREG (v4i16 (IMPLICIT_DEF)),
4952            (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn),
4953            hsub), ssub)>;
4954def : Pat<(i32 (vector_extract (v8i16 (opNode V128:$Rn)), (i64 0))),
4955          (EXTRACT_SUBREG (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
4956            (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn),
4957            hsub), ssub)>;
4958def : Pat<(i32 (vector_extract (v4i32 (opNode V128:$Rn)), (i64 0))),
4959          (EXTRACT_SUBREG (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
4960            (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn),
4961            ssub), ssub)>;
4962
4963}
4964
4965multiclass SIMDAcrossLanesSignedIntrinsic<string baseOpc,
4966                                          SDPatternOperator opNode>
4967    : SIMDAcrossLanesIntrinsic<baseOpc, opNode> {
4968// If there is a sign extension after this intrinsic, consume it as smov already
4969// performed it
4970def : Pat<(i32 (sext_inreg (i32 (vector_extract (insert_subvector undef,
4971            (opNode (v8i8 V64:$Rn)), (i32 0)), (i64 0))), i8)),
4972          (i32 (SMOVvi8to32
4973            (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4974              (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
4975            (i64 0)))>;
4976def : Pat<(i32 (sext_inreg (i32 (vector_extract
4977            (opNode (v16i8 V128:$Rn)), (i64 0))), i8)),
4978          (i32 (SMOVvi8to32
4979            (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4980             (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
4981            (i64 0)))>;
4982def : Pat<(i32 (sext_inreg (i32 (vector_extract (insert_subvector undef,
4983            (opNode (v4i16 V64:$Rn)), (i32 0)), (i64 0))), i16)),
4984          (i32 (SMOVvi16to32
4985           (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4986            (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
4987           (i64 0)))>;
4988def : Pat<(i32 (sext_inreg (i32 (vector_extract
4989            (opNode (v8i16 V128:$Rn)), (i64 0))), i16)),
4990          (i32 (SMOVvi16to32
4991            (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4992             (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
4993            (i64 0)))>;
4994}
4995
4996multiclass SIMDAcrossLanesUnsignedIntrinsic<string baseOpc,
4997                                            SDPatternOperator opNode>
4998    : SIMDAcrossLanesIntrinsic<baseOpc, opNode> {
4999// If there is a masking operation keeping only what has been actually
5000// generated, consume it.
5001def : Pat<(i32 (and (i32 (vector_extract (insert_subvector undef,
5002            (opNode (v8i8 V64:$Rn)), (i32 0)), (i64 0))), maski8_or_more)),
5003      (i32 (EXTRACT_SUBREG
5004        (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
5005          (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
5006        ssub))>;
5007def : Pat<(i32 (and (i32 (vector_extract (opNode (v16i8 V128:$Rn)), (i64 0))),
5008            maski8_or_more)),
5009        (i32 (EXTRACT_SUBREG
5010          (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
5011            (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
5012          ssub))>;
5013def : Pat<(i32 (and (i32 (vector_extract (insert_subvector undef,
5014            (opNode (v4i16 V64:$Rn)), (i32 0)), (i64 0))), maski16_or_more)),
5015          (i32 (EXTRACT_SUBREG
5016            (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
5017              (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
5018            ssub))>;
5019def : Pat<(i32 (and (i32 (vector_extract (opNode (v8i16 V128:$Rn)), (i64 0))),
5020            maski16_or_more)),
5021        (i32 (EXTRACT_SUBREG
5022          (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
5023            (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
5024          ssub))>;
5025}
5026
5027defm : SIMDAcrossLanesSignedIntrinsic<"ADDV",  AArch64saddv>;
5028// vaddv_[su]32 is special; -> ADDP Vd.2S,Vn.2S,Vm.2S; return Vd.s[0];Vn==Vm
5029def : Pat<(v2i32 (AArch64saddv (v2i32 V64:$Rn))),
5030          (ADDPv2i32 V64:$Rn, V64:$Rn)>;
5031
5032defm : SIMDAcrossLanesUnsignedIntrinsic<"ADDV", AArch64uaddv>;
5033// vaddv_[su]32 is special; -> ADDP Vd.2S,Vn.2S,Vm.2S; return Vd.s[0];Vn==Vm
5034def : Pat<(v2i32 (AArch64uaddv (v2i32 V64:$Rn))),
5035          (ADDPv2i32 V64:$Rn, V64:$Rn)>;
5036
5037defm : SIMDAcrossLanesSignedIntrinsic<"SMAXV", AArch64smaxv>;
5038def : Pat<(v2i32 (AArch64smaxv (v2i32 V64:$Rn))),
5039          (SMAXPv2i32 V64:$Rn, V64:$Rn)>;
5040
5041defm : SIMDAcrossLanesSignedIntrinsic<"SMINV", AArch64sminv>;
5042def : Pat<(v2i32 (AArch64sminv (v2i32 V64:$Rn))),
5043          (SMINPv2i32 V64:$Rn, V64:$Rn)>;
5044
5045defm : SIMDAcrossLanesUnsignedIntrinsic<"UMAXV", AArch64umaxv>;
5046def : Pat<(v2i32 (AArch64umaxv (v2i32 V64:$Rn))),
5047          (UMAXPv2i32 V64:$Rn, V64:$Rn)>;
5048
5049defm : SIMDAcrossLanesUnsignedIntrinsic<"UMINV", AArch64uminv>;
5050def : Pat<(v2i32 (AArch64uminv (v2i32 V64:$Rn))),
5051          (UMINPv2i32 V64:$Rn, V64:$Rn)>;
5052
5053multiclass SIMDAcrossLanesSignedLongIntrinsic<string baseOpc, Intrinsic intOp> {
5054  def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
5055        (i32 (SMOVvi16to32
5056          (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
5057            (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), hsub),
5058          (i64 0)))>;
5059def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
5060        (i32 (SMOVvi16to32
5061          (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
5062           (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), hsub),
5063          (i64 0)))>;
5064
5065def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
5066          (i32 (EXTRACT_SUBREG
5067           (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
5068            (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), ssub),
5069           ssub))>;
5070def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
5071        (i32 (EXTRACT_SUBREG
5072          (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
5073           (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), ssub),
5074          ssub))>;
5075
5076def : Pat<(i64 (intOp (v4i32 V128:$Rn))),
5077        (i64 (EXTRACT_SUBREG
5078          (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
5079           (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), dsub),
5080          dsub))>;
5081}
5082
5083multiclass SIMDAcrossLanesUnsignedLongIntrinsic<string baseOpc,
5084                                                Intrinsic intOp> {
5085  def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
5086        (i32 (EXTRACT_SUBREG
5087          (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
5088            (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), hsub),
5089          ssub))>;
5090def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
5091        (i32 (EXTRACT_SUBREG
5092          (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
5093            (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), hsub),
5094          ssub))>;
5095
5096def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
5097          (i32 (EXTRACT_SUBREG
5098            (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
5099              (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), ssub),
5100            ssub))>;
5101def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
5102        (i32 (EXTRACT_SUBREG
5103          (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
5104            (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), ssub),
5105          ssub))>;
5106
5107def : Pat<(i64 (intOp (v4i32 V128:$Rn))),
5108        (i64 (EXTRACT_SUBREG
5109          (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
5110            (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), dsub),
5111          dsub))>;
5112}
5113
5114defm : SIMDAcrossLanesSignedLongIntrinsic<"SADDLV", int_aarch64_neon_saddlv>;
5115defm : SIMDAcrossLanesUnsignedLongIntrinsic<"UADDLV", int_aarch64_neon_uaddlv>;
5116
5117// The vaddlv_s32 intrinsic gets mapped to SADDLP.
5118def : Pat<(i64 (int_aarch64_neon_saddlv (v2i32 V64:$Rn))),
5119          (i64 (EXTRACT_SUBREG
5120            (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
5121              (SADDLPv2i32_v1i64 V64:$Rn), dsub),
5122            dsub))>;
5123// The vaddlv_u32 intrinsic gets mapped to UADDLP.
5124def : Pat<(i64 (int_aarch64_neon_uaddlv (v2i32 V64:$Rn))),
5125          (i64 (EXTRACT_SUBREG
5126            (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
5127              (UADDLPv2i32_v1i64 V64:$Rn), dsub),
5128            dsub))>;
5129
5130//------------------------------------------------------------------------------
5131// AdvSIMD modified immediate instructions
5132//------------------------------------------------------------------------------
5133
5134// AdvSIMD BIC
5135defm BIC : SIMDModifiedImmVectorShiftTied<1, 0b11, 0b01, "bic", AArch64bici>;
5136// AdvSIMD ORR
5137defm ORR : SIMDModifiedImmVectorShiftTied<0, 0b11, 0b01, "orr", AArch64orri>;
5138
5139def : InstAlias<"bic $Vd.4h, $imm", (BICv4i16 V64:$Vd,  imm0_255:$imm, 0)>;
5140def : InstAlias<"bic $Vd.8h, $imm", (BICv8i16 V128:$Vd, imm0_255:$imm, 0)>;
5141def : InstAlias<"bic $Vd.2s, $imm", (BICv2i32 V64:$Vd,  imm0_255:$imm, 0)>;
5142def : InstAlias<"bic $Vd.4s, $imm", (BICv4i32 V128:$Vd, imm0_255:$imm, 0)>;
5143
5144def : InstAlias<"bic.4h $Vd, $imm", (BICv4i16 V64:$Vd,  imm0_255:$imm, 0)>;
5145def : InstAlias<"bic.8h $Vd, $imm", (BICv8i16 V128:$Vd, imm0_255:$imm, 0)>;
5146def : InstAlias<"bic.2s $Vd, $imm", (BICv2i32 V64:$Vd,  imm0_255:$imm, 0)>;
5147def : InstAlias<"bic.4s $Vd, $imm", (BICv4i32 V128:$Vd, imm0_255:$imm, 0)>;
5148
5149def : InstAlias<"orr $Vd.4h, $imm", (ORRv4i16 V64:$Vd,  imm0_255:$imm, 0)>;
5150def : InstAlias<"orr $Vd.8h, $imm", (ORRv8i16 V128:$Vd, imm0_255:$imm, 0)>;
5151def : InstAlias<"orr $Vd.2s, $imm", (ORRv2i32 V64:$Vd,  imm0_255:$imm, 0)>;
5152def : InstAlias<"orr $Vd.4s, $imm", (ORRv4i32 V128:$Vd, imm0_255:$imm, 0)>;
5153
5154def : InstAlias<"orr.4h $Vd, $imm", (ORRv4i16 V64:$Vd,  imm0_255:$imm, 0)>;
5155def : InstAlias<"orr.8h $Vd, $imm", (ORRv8i16 V128:$Vd, imm0_255:$imm, 0)>;
5156def : InstAlias<"orr.2s $Vd, $imm", (ORRv2i32 V64:$Vd,  imm0_255:$imm, 0)>;
5157def : InstAlias<"orr.4s $Vd, $imm", (ORRv4i32 V128:$Vd, imm0_255:$imm, 0)>;
5158
5159// AdvSIMD FMOV
5160def FMOVv2f64_ns : SIMDModifiedImmVectorNoShift<1, 1, 0, 0b1111, V128, fpimm8,
5161                                              "fmov", ".2d",
5162                       [(set (v2f64 V128:$Rd), (AArch64fmov imm0_255:$imm8))]>;
5163def FMOVv2f32_ns : SIMDModifiedImmVectorNoShift<0, 0, 0, 0b1111, V64,  fpimm8,
5164                                              "fmov", ".2s",
5165                       [(set (v2f32 V64:$Rd), (AArch64fmov imm0_255:$imm8))]>;
5166def FMOVv4f32_ns : SIMDModifiedImmVectorNoShift<1, 0, 0, 0b1111, V128, fpimm8,
5167                                              "fmov", ".4s",
5168                       [(set (v4f32 V128:$Rd), (AArch64fmov imm0_255:$imm8))]>;
5169let Predicates = [HasNEON, HasFullFP16] in {
5170def FMOVv4f16_ns : SIMDModifiedImmVectorNoShift<0, 0, 1, 0b1111, V64,  fpimm8,
5171                                              "fmov", ".4h",
5172                       [(set (v4f16 V64:$Rd), (AArch64fmov imm0_255:$imm8))]>;
5173def FMOVv8f16_ns : SIMDModifiedImmVectorNoShift<1, 0, 1, 0b1111, V128, fpimm8,
5174                                              "fmov", ".8h",
5175                       [(set (v8f16 V128:$Rd), (AArch64fmov imm0_255:$imm8))]>;
5176} // Predicates = [HasNEON, HasFullFP16]
5177
5178// AdvSIMD MOVI
5179
5180// EDIT byte mask: scalar
5181let isReMaterializable = 1, isAsCheapAsAMove = 1 in
5182def MOVID      : SIMDModifiedImmScalarNoShift<0, 1, 0b1110, "movi",
5183                    [(set FPR64:$Rd, simdimmtype10:$imm8)]>;
5184// The movi_edit node has the immediate value already encoded, so we use
5185// a plain imm0_255 here.
5186def : Pat<(f64 (AArch64movi_edit imm0_255:$shift)),
5187          (MOVID imm0_255:$shift)>;
5188
5189// EDIT byte mask: 2d
5190
5191// The movi_edit node has the immediate value already encoded, so we use
5192// a plain imm0_255 in the pattern
5193let isReMaterializable = 1, isAsCheapAsAMove = 1 in
5194def MOVIv2d_ns   : SIMDModifiedImmVectorNoShift<1, 1, 0, 0b1110, V128,
5195                                                simdimmtype10,
5196                                                "movi", ".2d",
5197                   [(set (v2i64 V128:$Rd), (AArch64movi_edit imm0_255:$imm8))]>;
5198
5199def : Pat<(v2i64 immAllZerosV), (MOVIv2d_ns (i32 0))>;
5200def : Pat<(v4i32 immAllZerosV), (MOVIv2d_ns (i32 0))>;
5201def : Pat<(v8i16 immAllZerosV), (MOVIv2d_ns (i32 0))>;
5202def : Pat<(v16i8 immAllZerosV), (MOVIv2d_ns (i32 0))>;
5203
5204def : Pat<(v2i64 immAllOnesV), (MOVIv2d_ns (i32 255))>;
5205def : Pat<(v4i32 immAllOnesV), (MOVIv2d_ns (i32 255))>;
5206def : Pat<(v8i16 immAllOnesV), (MOVIv2d_ns (i32 255))>;
5207def : Pat<(v16i8 immAllOnesV), (MOVIv2d_ns (i32 255))>;
5208
5209// Set 64-bit vectors to all 0/1 by extracting from a 128-bit register as the
5210// extract is free and this gives better MachineCSE results.
5211def : Pat<(v1i64 immAllZerosV), (EXTRACT_SUBREG (MOVIv2d_ns (i32 0)), dsub)>;
5212def : Pat<(v2i32 immAllZerosV), (EXTRACT_SUBREG (MOVIv2d_ns (i32 0)), dsub)>;
5213def : Pat<(v4i16 immAllZerosV), (EXTRACT_SUBREG (MOVIv2d_ns (i32 0)), dsub)>;
5214def : Pat<(v8i8  immAllZerosV), (EXTRACT_SUBREG (MOVIv2d_ns (i32 0)), dsub)>;
5215
5216def : Pat<(v1i64 immAllOnesV), (EXTRACT_SUBREG (MOVIv2d_ns (i32 255)), dsub)>;
5217def : Pat<(v2i32 immAllOnesV), (EXTRACT_SUBREG (MOVIv2d_ns (i32 255)), dsub)>;
5218def : Pat<(v4i16 immAllOnesV), (EXTRACT_SUBREG (MOVIv2d_ns (i32 255)), dsub)>;
5219def : Pat<(v8i8  immAllOnesV), (EXTRACT_SUBREG (MOVIv2d_ns (i32 255)), dsub)>;
5220
5221// EDIT per word & halfword: 2s, 4h, 4s, & 8h
5222let isReMaterializable = 1, isAsCheapAsAMove = 1 in
5223defm MOVI      : SIMDModifiedImmVectorShift<0, 0b10, 0b00, "movi">;
5224
5225def : InstAlias<"movi $Vd.4h, $imm", (MOVIv4i16 V64:$Vd,  imm0_255:$imm, 0), 0>;
5226def : InstAlias<"movi $Vd.8h, $imm", (MOVIv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
5227def : InstAlias<"movi $Vd.2s, $imm", (MOVIv2i32 V64:$Vd,  imm0_255:$imm, 0), 0>;
5228def : InstAlias<"movi $Vd.4s, $imm", (MOVIv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
5229
5230def : InstAlias<"movi.4h $Vd, $imm", (MOVIv4i16 V64:$Vd,  imm0_255:$imm, 0), 0>;
5231def : InstAlias<"movi.8h $Vd, $imm", (MOVIv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
5232def : InstAlias<"movi.2s $Vd, $imm", (MOVIv2i32 V64:$Vd,  imm0_255:$imm, 0), 0>;
5233def : InstAlias<"movi.4s $Vd, $imm", (MOVIv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
5234
5235def : Pat<(v2i32 (AArch64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
5236          (MOVIv2i32 imm0_255:$imm8, imm:$shift)>;
5237def : Pat<(v4i32 (AArch64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
5238          (MOVIv4i32 imm0_255:$imm8, imm:$shift)>;
5239def : Pat<(v4i16 (AArch64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
5240          (MOVIv4i16 imm0_255:$imm8, imm:$shift)>;
5241def : Pat<(v8i16 (AArch64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
5242          (MOVIv8i16 imm0_255:$imm8, imm:$shift)>;
5243
5244let isReMaterializable = 1, isAsCheapAsAMove = 1 in {
5245// EDIT per word: 2s & 4s with MSL shifter
5246def MOVIv2s_msl  : SIMDModifiedImmMoveMSL<0, 0, {1,1,0,?}, V64, "movi", ".2s",
5247                      [(set (v2i32 V64:$Rd),
5248                            (AArch64movi_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
5249def MOVIv4s_msl  : SIMDModifiedImmMoveMSL<1, 0, {1,1,0,?}, V128, "movi", ".4s",
5250                      [(set (v4i32 V128:$Rd),
5251                            (AArch64movi_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
5252
5253// Per byte: 8b & 16b
5254def MOVIv8b_ns   : SIMDModifiedImmVectorNoShift<0, 0, 0, 0b1110, V64,  imm0_255,
5255                                                 "movi", ".8b",
5256                       [(set (v8i8 V64:$Rd), (AArch64movi imm0_255:$imm8))]>;
5257
5258def MOVIv16b_ns  : SIMDModifiedImmVectorNoShift<1, 0, 0, 0b1110, V128, imm0_255,
5259                                                 "movi", ".16b",
5260                       [(set (v16i8 V128:$Rd), (AArch64movi imm0_255:$imm8))]>;
5261}
5262
5263// AdvSIMD MVNI
5264
5265// EDIT per word & halfword: 2s, 4h, 4s, & 8h
5266let isReMaterializable = 1, isAsCheapAsAMove = 1 in
5267defm MVNI      : SIMDModifiedImmVectorShift<1, 0b10, 0b00, "mvni">;
5268
5269def : InstAlias<"mvni $Vd.4h, $imm", (MVNIv4i16 V64:$Vd,  imm0_255:$imm, 0), 0>;
5270def : InstAlias<"mvni $Vd.8h, $imm", (MVNIv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
5271def : InstAlias<"mvni $Vd.2s, $imm", (MVNIv2i32 V64:$Vd,  imm0_255:$imm, 0), 0>;
5272def : InstAlias<"mvni $Vd.4s, $imm", (MVNIv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
5273
5274def : InstAlias<"mvni.4h $Vd, $imm", (MVNIv4i16 V64:$Vd,  imm0_255:$imm, 0), 0>;
5275def : InstAlias<"mvni.8h $Vd, $imm", (MVNIv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
5276def : InstAlias<"mvni.2s $Vd, $imm", (MVNIv2i32 V64:$Vd,  imm0_255:$imm, 0), 0>;
5277def : InstAlias<"mvni.4s $Vd, $imm", (MVNIv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
5278
5279def : Pat<(v2i32 (AArch64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
5280          (MVNIv2i32 imm0_255:$imm8, imm:$shift)>;
5281def : Pat<(v4i32 (AArch64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
5282          (MVNIv4i32 imm0_255:$imm8, imm:$shift)>;
5283def : Pat<(v4i16 (AArch64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
5284          (MVNIv4i16 imm0_255:$imm8, imm:$shift)>;
5285def : Pat<(v8i16 (AArch64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
5286          (MVNIv8i16 imm0_255:$imm8, imm:$shift)>;
5287
5288// EDIT per word: 2s & 4s with MSL shifter
5289let isReMaterializable = 1, isAsCheapAsAMove = 1 in {
5290def MVNIv2s_msl   : SIMDModifiedImmMoveMSL<0, 1, {1,1,0,?}, V64, "mvni", ".2s",
5291                      [(set (v2i32 V64:$Rd),
5292                            (AArch64mvni_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
5293def MVNIv4s_msl   : SIMDModifiedImmMoveMSL<1, 1, {1,1,0,?}, V128, "mvni", ".4s",
5294                      [(set (v4i32 V128:$Rd),
5295                            (AArch64mvni_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
5296}
5297
5298//----------------------------------------------------------------------------
5299// AdvSIMD indexed element
5300//----------------------------------------------------------------------------
5301
5302let hasSideEffects = 0 in {
5303  defm FMLA  : SIMDFPIndexedTied<0, 0b0001, "fmla">;
5304  defm FMLS  : SIMDFPIndexedTied<0, 0b0101, "fmls">;
5305}
5306
5307// NOTE: Operands are reordered in the FMLA/FMLS PatFrags because the
5308// instruction expects the addend first, while the intrinsic expects it last.
5309
5310// On the other hand, there are quite a few valid combinatorial options due to
5311// the commutativity of multiplication and the fact that (-x) * y = x * (-y).
5312defm : SIMDFPIndexedTiedPatterns<"FMLA",
5313           TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)>>;
5314defm : SIMDFPIndexedTiedPatterns<"FMLA",
5315           TriOpFrag<(fma node:$MHS, node:$RHS, node:$LHS)>>;
5316
5317defm : SIMDFPIndexedTiedPatterns<"FMLS",
5318           TriOpFrag<(fma node:$MHS, (fneg node:$RHS), node:$LHS)> >;
5319defm : SIMDFPIndexedTiedPatterns<"FMLS",
5320           TriOpFrag<(fma node:$RHS, (fneg node:$MHS), node:$LHS)> >;
5321defm : SIMDFPIndexedTiedPatterns<"FMLS",
5322           TriOpFrag<(fma (fneg node:$RHS), node:$MHS, node:$LHS)> >;
5323defm : SIMDFPIndexedTiedPatterns<"FMLS",
5324           TriOpFrag<(fma (fneg node:$MHS), node:$RHS, node:$LHS)> >;
5325
5326multiclass FMLSIndexedAfterNegPatterns<SDPatternOperator OpNode> {
5327  // 3 variants for the .2s version: DUPLANE from 128-bit, DUPLANE from 64-bit
5328  // and DUP scalar.
5329  def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
5330                           (AArch64duplane32 (v4f32 (fneg V128:$Rm)),
5331                                           VectorIndexS:$idx))),
5332            (FMLSv2i32_indexed V64:$Rd, V64:$Rn, V128:$Rm, VectorIndexS:$idx)>;
5333  def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
5334                           (v2f32 (AArch64duplane32
5335                                      (v4f32 (insert_subvector undef,
5336                                                 (v2f32 (fneg V64:$Rm)),
5337                                                 (i32 0))),
5338                                      VectorIndexS:$idx)))),
5339            (FMLSv2i32_indexed V64:$Rd, V64:$Rn,
5340                               (SUBREG_TO_REG (i32 0), V64:$Rm, dsub),
5341                               VectorIndexS:$idx)>;
5342  def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
5343                           (AArch64dup (f32 (fneg FPR32Op:$Rm))))),
5344            (FMLSv2i32_indexed V64:$Rd, V64:$Rn,
5345                (SUBREG_TO_REG (i32 0), FPR32Op:$Rm, ssub), (i64 0))>;
5346
5347  // 3 variants for the .4s version: DUPLANE from 128-bit, DUPLANE from 64-bit
5348  // and DUP scalar.
5349  def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
5350                           (AArch64duplane32 (v4f32 (fneg V128:$Rm)),
5351                                           VectorIndexS:$idx))),
5352            (FMLSv4i32_indexed V128:$Rd, V128:$Rn, V128:$Rm,
5353                               VectorIndexS:$idx)>;
5354  def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
5355                           (v4f32 (AArch64duplane32
5356                                      (v4f32 (insert_subvector undef,
5357                                                 (v2f32 (fneg V64:$Rm)),
5358                                                 (i32 0))),
5359                                      VectorIndexS:$idx)))),
5360            (FMLSv4i32_indexed V128:$Rd, V128:$Rn,
5361                               (SUBREG_TO_REG (i32 0), V64:$Rm, dsub),
5362                               VectorIndexS:$idx)>;
5363  def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
5364                           (AArch64dup (f32 (fneg FPR32Op:$Rm))))),
5365            (FMLSv4i32_indexed V128:$Rd, V128:$Rn,
5366                (SUBREG_TO_REG (i32 0), FPR32Op:$Rm, ssub), (i64 0))>;
5367
5368  // 2 variants for the .2d version: DUPLANE from 128-bit, and DUP scalar
5369  // (DUPLANE from 64-bit would be trivial).
5370  def : Pat<(v2f64 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn),
5371                           (AArch64duplane64 (v2f64 (fneg V128:$Rm)),
5372                                           VectorIndexD:$idx))),
5373            (FMLSv2i64_indexed
5374                V128:$Rd, V128:$Rn, V128:$Rm, VectorIndexS:$idx)>;
5375  def : Pat<(v2f64 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn),
5376                           (AArch64dup (f64 (fneg FPR64Op:$Rm))))),
5377            (FMLSv2i64_indexed V128:$Rd, V128:$Rn,
5378                (SUBREG_TO_REG (i32 0), FPR64Op:$Rm, dsub), (i64 0))>;
5379
5380  // 2 variants for 32-bit scalar version: extract from .2s or from .4s
5381  def : Pat<(f32 (OpNode (f32 FPR32:$Rd), (f32 FPR32:$Rn),
5382                         (vector_extract (v4f32 (fneg V128:$Rm)),
5383                                         VectorIndexS:$idx))),
5384            (FMLSv1i32_indexed FPR32:$Rd, FPR32:$Rn,
5385                V128:$Rm, VectorIndexS:$idx)>;
5386  def : Pat<(f32 (OpNode (f32 FPR32:$Rd), (f32 FPR32:$Rn),
5387                         (vector_extract (v4f32 (insert_subvector undef,
5388                                                    (v2f32 (fneg V64:$Rm)),
5389                                                    (i32 0))),
5390                                         VectorIndexS:$idx))),
5391            (FMLSv1i32_indexed FPR32:$Rd, FPR32:$Rn,
5392                (SUBREG_TO_REG (i32 0), V64:$Rm, dsub), VectorIndexS:$idx)>;
5393
5394  // 1 variant for 64-bit scalar version: extract from .1d or from .2d
5395  def : Pat<(f64 (OpNode (f64 FPR64:$Rd), (f64 FPR64:$Rn),
5396                         (vector_extract (v2f64 (fneg V128:$Rm)),
5397                                         VectorIndexS:$idx))),
5398            (FMLSv1i64_indexed FPR64:$Rd, FPR64:$Rn,
5399                V128:$Rm, VectorIndexS:$idx)>;
5400}
5401
5402defm : FMLSIndexedAfterNegPatterns<
5403           TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)> >;
5404defm : FMLSIndexedAfterNegPatterns<
5405           TriOpFrag<(fma node:$MHS, node:$RHS, node:$LHS)> >;
5406
5407defm FMULX : SIMDFPIndexed<1, 0b1001, "fmulx", int_aarch64_neon_fmulx>;
5408defm FMUL  : SIMDFPIndexed<0, 0b1001, "fmul", fmul>;
5409
5410def : Pat<(v2f32 (fmul V64:$Rn, (AArch64dup (f32 FPR32:$Rm)))),
5411          (FMULv2i32_indexed V64:$Rn,
5412            (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rm, ssub),
5413            (i64 0))>;
5414def : Pat<(v4f32 (fmul V128:$Rn, (AArch64dup (f32 FPR32:$Rm)))),
5415          (FMULv4i32_indexed V128:$Rn,
5416            (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rm, ssub),
5417            (i64 0))>;
5418def : Pat<(v2f64 (fmul V128:$Rn, (AArch64dup (f64 FPR64:$Rm)))),
5419          (FMULv2i64_indexed V128:$Rn,
5420            (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$Rm, dsub),
5421            (i64 0))>;
5422
5423defm SQDMULH : SIMDIndexedHS<0, 0b1100, "sqdmulh", int_aarch64_neon_sqdmulh>;
5424defm SQRDMULH : SIMDIndexedHS<0, 0b1101, "sqrdmulh", int_aarch64_neon_sqrdmulh>;
5425defm MLA   : SIMDVectorIndexedHSTied<1, 0b0000, "mla",
5426              TriOpFrag<(add node:$LHS, (mul node:$MHS, node:$RHS))>>;
5427defm MLS   : SIMDVectorIndexedHSTied<1, 0b0100, "mls",
5428              TriOpFrag<(sub node:$LHS, (mul node:$MHS, node:$RHS))>>;
5429defm MUL   : SIMDVectorIndexedHS<0, 0b1000, "mul", mul>;
5430defm SMLAL : SIMDVectorIndexedLongSDTied<0, 0b0010, "smlal",
5431    TriOpFrag<(add node:$LHS, (int_aarch64_neon_smull node:$MHS, node:$RHS))>>;
5432defm SMLSL : SIMDVectorIndexedLongSDTied<0, 0b0110, "smlsl",
5433    TriOpFrag<(sub node:$LHS, (int_aarch64_neon_smull node:$MHS, node:$RHS))>>;
5434defm SMULL : SIMDVectorIndexedLongSD<0, 0b1010, "smull",
5435                int_aarch64_neon_smull>;
5436defm SQDMLAL : SIMDIndexedLongSQDMLXSDTied<0, 0b0011, "sqdmlal",
5437                                           int_aarch64_neon_sqadd>;
5438defm SQDMLSL : SIMDIndexedLongSQDMLXSDTied<0, 0b0111, "sqdmlsl",
5439                                           int_aarch64_neon_sqsub>;
5440defm SQRDMLAH : SIMDIndexedSQRDMLxHSDTied<1, 0b1101, "sqrdmlah",
5441                                          int_aarch64_neon_sqadd>;
5442defm SQRDMLSH : SIMDIndexedSQRDMLxHSDTied<1, 0b1111, "sqrdmlsh",
5443                                          int_aarch64_neon_sqsub>;
5444defm SQDMULL : SIMDIndexedLongSD<0, 0b1011, "sqdmull", int_aarch64_neon_sqdmull>;
5445defm UMLAL   : SIMDVectorIndexedLongSDTied<1, 0b0010, "umlal",
5446    TriOpFrag<(add node:$LHS, (int_aarch64_neon_umull node:$MHS, node:$RHS))>>;
5447defm UMLSL   : SIMDVectorIndexedLongSDTied<1, 0b0110, "umlsl",
5448    TriOpFrag<(sub node:$LHS, (int_aarch64_neon_umull node:$MHS, node:$RHS))>>;
5449defm UMULL   : SIMDVectorIndexedLongSD<1, 0b1010, "umull",
5450                int_aarch64_neon_umull>;
5451
5452// A scalar sqdmull with the second operand being a vector lane can be
5453// handled directly with the indexed instruction encoding.
5454def : Pat<(int_aarch64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
5455                                          (vector_extract (v4i32 V128:$Vm),
5456                                                           VectorIndexS:$idx)),
5457          (SQDMULLv1i64_indexed FPR32:$Rn, V128:$Vm, VectorIndexS:$idx)>;
5458
5459//----------------------------------------------------------------------------
5460// AdvSIMD scalar shift instructions
5461//----------------------------------------------------------------------------
5462defm FCVTZS : SIMDFPScalarRShift<0, 0b11111, "fcvtzs">;
5463defm FCVTZU : SIMDFPScalarRShift<1, 0b11111, "fcvtzu">;
5464defm SCVTF  : SIMDFPScalarRShift<0, 0b11100, "scvtf">;
5465defm UCVTF  : SIMDFPScalarRShift<1, 0b11100, "ucvtf">;
5466// Codegen patterns for the above. We don't put these directly on the
5467// instructions because TableGen's type inference can't handle the truth.
5468// Having the same base pattern for fp <--> int totally freaks it out.
5469def : Pat<(int_aarch64_neon_vcvtfp2fxs FPR32:$Rn, vecshiftR32:$imm),
5470          (FCVTZSs FPR32:$Rn, vecshiftR32:$imm)>;
5471def : Pat<(int_aarch64_neon_vcvtfp2fxu FPR32:$Rn, vecshiftR32:$imm),
5472          (FCVTZUs FPR32:$Rn, vecshiftR32:$imm)>;
5473def : Pat<(i64 (int_aarch64_neon_vcvtfp2fxs (f64 FPR64:$Rn), vecshiftR64:$imm)),
5474          (FCVTZSd FPR64:$Rn, vecshiftR64:$imm)>;
5475def : Pat<(i64 (int_aarch64_neon_vcvtfp2fxu (f64 FPR64:$Rn), vecshiftR64:$imm)),
5476          (FCVTZUd FPR64:$Rn, vecshiftR64:$imm)>;
5477def : Pat<(v1i64 (int_aarch64_neon_vcvtfp2fxs (v1f64 FPR64:$Rn),
5478                                            vecshiftR64:$imm)),
5479          (FCVTZSd FPR64:$Rn, vecshiftR64:$imm)>;
5480def : Pat<(v1i64 (int_aarch64_neon_vcvtfp2fxu (v1f64 FPR64:$Rn),
5481                                            vecshiftR64:$imm)),
5482          (FCVTZUd FPR64:$Rn, vecshiftR64:$imm)>;
5483def : Pat<(int_aarch64_neon_vcvtfxu2fp FPR32:$Rn, vecshiftR32:$imm),
5484          (UCVTFs FPR32:$Rn, vecshiftR32:$imm)>;
5485def : Pat<(f64 (int_aarch64_neon_vcvtfxu2fp (i64 FPR64:$Rn), vecshiftR64:$imm)),
5486          (UCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
5487def : Pat<(v1f64 (int_aarch64_neon_vcvtfxs2fp (v1i64 FPR64:$Rn),
5488                                            vecshiftR64:$imm)),
5489          (SCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
5490def : Pat<(f64 (int_aarch64_neon_vcvtfxs2fp (i64 FPR64:$Rn), vecshiftR64:$imm)),
5491          (SCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
5492def : Pat<(v1f64 (int_aarch64_neon_vcvtfxu2fp (v1i64 FPR64:$Rn),
5493                                            vecshiftR64:$imm)),
5494          (UCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
5495def : Pat<(int_aarch64_neon_vcvtfxs2fp FPR32:$Rn, vecshiftR32:$imm),
5496          (SCVTFs FPR32:$Rn, vecshiftR32:$imm)>;
5497
5498// Patterns for FP16 Instrinsics - requires reg copy to/from as i16s not supported.
5499
5500def : Pat<(f16 (int_aarch64_neon_vcvtfxs2fp (i32 (sext_inreg FPR32:$Rn, i16)), vecshiftR16:$imm)),
5501          (SCVTFh (EXTRACT_SUBREG FPR32:$Rn, hsub), vecshiftR16:$imm)>;
5502def : Pat<(f16 (int_aarch64_neon_vcvtfxs2fp (i32 FPR32:$Rn), vecshiftR16:$imm)),
5503          (SCVTFh (EXTRACT_SUBREG FPR32:$Rn, hsub), vecshiftR16:$imm)>;
5504def : Pat<(f16 (int_aarch64_neon_vcvtfxs2fp (i64 FPR64:$Rn), vecshiftR16:$imm)),
5505          (SCVTFh (EXTRACT_SUBREG FPR64:$Rn, hsub), vecshiftR16:$imm)>;
5506def : Pat<(f16 (int_aarch64_neon_vcvtfxu2fp
5507            (and FPR32:$Rn, (i32 65535)),
5508            vecshiftR16:$imm)),
5509          (UCVTFh (EXTRACT_SUBREG FPR32:$Rn, hsub), vecshiftR16:$imm)>;
5510def : Pat<(f16 (int_aarch64_neon_vcvtfxu2fp FPR32:$Rn, vecshiftR16:$imm)),
5511          (UCVTFh (EXTRACT_SUBREG FPR32:$Rn, hsub), vecshiftR16:$imm)>;
5512def : Pat<(f16 (int_aarch64_neon_vcvtfxu2fp (i64 FPR64:$Rn), vecshiftR16:$imm)),
5513          (UCVTFh (EXTRACT_SUBREG FPR64:$Rn, hsub), vecshiftR16:$imm)>;
5514def : Pat<(i32 (int_aarch64_neon_vcvtfp2fxs (f16 FPR16:$Rn), vecshiftR32:$imm)),
5515          (i32 (INSERT_SUBREG
5516            (i32 (IMPLICIT_DEF)),
5517            (FCVTZSh FPR16:$Rn, vecshiftR32:$imm),
5518            hsub))>;
5519def : Pat<(i64 (int_aarch64_neon_vcvtfp2fxs (f16 FPR16:$Rn), vecshiftR64:$imm)),
5520          (i64 (INSERT_SUBREG
5521            (i64 (IMPLICIT_DEF)),
5522            (FCVTZSh FPR16:$Rn, vecshiftR64:$imm),
5523            hsub))>;
5524def : Pat<(i32 (int_aarch64_neon_vcvtfp2fxu (f16 FPR16:$Rn), vecshiftR32:$imm)),
5525          (i32 (INSERT_SUBREG
5526            (i32 (IMPLICIT_DEF)),
5527            (FCVTZUh FPR16:$Rn, vecshiftR32:$imm),
5528            hsub))>;
5529def : Pat<(i64 (int_aarch64_neon_vcvtfp2fxu (f16 FPR16:$Rn), vecshiftR64:$imm)),
5530          (i64 (INSERT_SUBREG
5531            (i64 (IMPLICIT_DEF)),
5532            (FCVTZUh FPR16:$Rn, vecshiftR64:$imm),
5533            hsub))>;
5534def : Pat<(i32 (int_aarch64_neon_facge (f16 FPR16:$Rn), (f16 FPR16:$Rm))),
5535          (i32 (INSERT_SUBREG
5536            (i32 (IMPLICIT_DEF)),
5537            (FACGE16 FPR16:$Rn, FPR16:$Rm),
5538            hsub))>;
5539def : Pat<(i32 (int_aarch64_neon_facgt (f16 FPR16:$Rn), (f16 FPR16:$Rm))),
5540          (i32 (INSERT_SUBREG
5541            (i32 (IMPLICIT_DEF)),
5542            (FACGT16 FPR16:$Rn, FPR16:$Rm),
5543            hsub))>;
5544
5545defm SHL      : SIMDScalarLShiftD<   0, 0b01010, "shl", AArch64vshl>;
5546defm SLI      : SIMDScalarLShiftDTied<1, 0b01010, "sli">;
5547defm SQRSHRN  : SIMDScalarRShiftBHS< 0, 0b10011, "sqrshrn",
5548                                     int_aarch64_neon_sqrshrn>;
5549defm SQRSHRUN : SIMDScalarRShiftBHS< 1, 0b10001, "sqrshrun",
5550                                     int_aarch64_neon_sqrshrun>;
5551defm SQSHLU   : SIMDScalarLShiftBHSD<1, 0b01100, "sqshlu", AArch64sqshlui>;
5552defm SQSHL    : SIMDScalarLShiftBHSD<0, 0b01110, "sqshl", AArch64sqshli>;
5553defm SQSHRN   : SIMDScalarRShiftBHS< 0, 0b10010, "sqshrn",
5554                                     int_aarch64_neon_sqshrn>;
5555defm SQSHRUN  : SIMDScalarRShiftBHS< 1, 0b10000, "sqshrun",
5556                                     int_aarch64_neon_sqshrun>;
5557defm SRI      : SIMDScalarRShiftDTied<   1, 0b01000, "sri">;
5558defm SRSHR    : SIMDScalarRShiftD<   0, 0b00100, "srshr", AArch64srshri>;
5559defm SRSRA    : SIMDScalarRShiftDTied<   0, 0b00110, "srsra",
5560    TriOpFrag<(add node:$LHS,
5561                   (AArch64srshri node:$MHS, node:$RHS))>>;
5562defm SSHR     : SIMDScalarRShiftD<   0, 0b00000, "sshr", AArch64vashr>;
5563defm SSRA     : SIMDScalarRShiftDTied<   0, 0b00010, "ssra",
5564    TriOpFrag<(add node:$LHS,
5565                   (AArch64vashr node:$MHS, node:$RHS))>>;
5566defm UQRSHRN  : SIMDScalarRShiftBHS< 1, 0b10011, "uqrshrn",
5567                                     int_aarch64_neon_uqrshrn>;
5568defm UQSHL    : SIMDScalarLShiftBHSD<1, 0b01110, "uqshl", AArch64uqshli>;
5569defm UQSHRN   : SIMDScalarRShiftBHS< 1, 0b10010, "uqshrn",
5570                                     int_aarch64_neon_uqshrn>;
5571defm URSHR    : SIMDScalarRShiftD<   1, 0b00100, "urshr", AArch64urshri>;
5572defm URSRA    : SIMDScalarRShiftDTied<   1, 0b00110, "ursra",
5573    TriOpFrag<(add node:$LHS,
5574                   (AArch64urshri node:$MHS, node:$RHS))>>;
5575defm USHR     : SIMDScalarRShiftD<   1, 0b00000, "ushr", AArch64vlshr>;
5576defm USRA     : SIMDScalarRShiftDTied<   1, 0b00010, "usra",
5577    TriOpFrag<(add node:$LHS,
5578                   (AArch64vlshr node:$MHS, node:$RHS))>>;
5579
5580//----------------------------------------------------------------------------
5581// AdvSIMD vector shift instructions
5582//----------------------------------------------------------------------------
5583defm FCVTZS:SIMDVectorRShiftSD<0, 0b11111, "fcvtzs", int_aarch64_neon_vcvtfp2fxs>;
5584defm FCVTZU:SIMDVectorRShiftSD<1, 0b11111, "fcvtzu", int_aarch64_neon_vcvtfp2fxu>;
5585defm SCVTF: SIMDVectorRShiftToFP<0, 0b11100, "scvtf",
5586                                   int_aarch64_neon_vcvtfxs2fp>;
5587defm RSHRN   : SIMDVectorRShiftNarrowBHS<0, 0b10001, "rshrn",
5588                                         int_aarch64_neon_rshrn>;
5589defm SHL     : SIMDVectorLShiftBHSD<0, 0b01010, "shl", AArch64vshl>;
5590defm SHRN    : SIMDVectorRShiftNarrowBHS<0, 0b10000, "shrn",
5591                          BinOpFrag<(trunc (AArch64vashr node:$LHS, node:$RHS))>>;
5592defm SLI     : SIMDVectorLShiftBHSDTied<1, 0b01010, "sli", int_aarch64_neon_vsli>;
5593def : Pat<(v1i64 (int_aarch64_neon_vsli (v1i64 FPR64:$Rd), (v1i64 FPR64:$Rn),
5594                                      (i32 vecshiftL64:$imm))),
5595          (SLId FPR64:$Rd, FPR64:$Rn, vecshiftL64:$imm)>;
5596defm SQRSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10011, "sqrshrn",
5597                                         int_aarch64_neon_sqrshrn>;
5598defm SQRSHRUN: SIMDVectorRShiftNarrowBHS<1, 0b10001, "sqrshrun",
5599                                         int_aarch64_neon_sqrshrun>;
5600defm SQSHLU : SIMDVectorLShiftBHSD<1, 0b01100, "sqshlu", AArch64sqshlui>;
5601defm SQSHL  : SIMDVectorLShiftBHSD<0, 0b01110, "sqshl", AArch64sqshli>;
5602defm SQSHRN  : SIMDVectorRShiftNarrowBHS<0, 0b10010, "sqshrn",
5603                                         int_aarch64_neon_sqshrn>;
5604defm SQSHRUN : SIMDVectorRShiftNarrowBHS<1, 0b10000, "sqshrun",
5605                                         int_aarch64_neon_sqshrun>;
5606defm SRI     : SIMDVectorRShiftBHSDTied<1, 0b01000, "sri", int_aarch64_neon_vsri>;
5607def : Pat<(v1i64 (int_aarch64_neon_vsri (v1i64 FPR64:$Rd), (v1i64 FPR64:$Rn),
5608                                      (i32 vecshiftR64:$imm))),
5609          (SRId FPR64:$Rd, FPR64:$Rn, vecshiftR64:$imm)>;
5610defm SRSHR   : SIMDVectorRShiftBHSD<0, 0b00100, "srshr", AArch64srshri>;
5611defm SRSRA   : SIMDVectorRShiftBHSDTied<0, 0b00110, "srsra",
5612                 TriOpFrag<(add node:$LHS,
5613                                (AArch64srshri node:$MHS, node:$RHS))> >;
5614defm SSHLL   : SIMDVectorLShiftLongBHSD<0, 0b10100, "sshll",
5615                BinOpFrag<(AArch64vshl (sext node:$LHS), node:$RHS)>>;
5616
5617defm SSHR    : SIMDVectorRShiftBHSD<0, 0b00000, "sshr", AArch64vashr>;
5618defm SSRA    : SIMDVectorRShiftBHSDTied<0, 0b00010, "ssra",
5619                TriOpFrag<(add node:$LHS, (AArch64vashr node:$MHS, node:$RHS))>>;
5620defm UCVTF   : SIMDVectorRShiftToFP<1, 0b11100, "ucvtf",
5621                        int_aarch64_neon_vcvtfxu2fp>;
5622defm UQRSHRN : SIMDVectorRShiftNarrowBHS<1, 0b10011, "uqrshrn",
5623                                         int_aarch64_neon_uqrshrn>;
5624defm UQSHL   : SIMDVectorLShiftBHSD<1, 0b01110, "uqshl", AArch64uqshli>;
5625defm UQSHRN  : SIMDVectorRShiftNarrowBHS<1, 0b10010, "uqshrn",
5626                                         int_aarch64_neon_uqshrn>;
5627defm URSHR   : SIMDVectorRShiftBHSD<1, 0b00100, "urshr", AArch64urshri>;
5628defm URSRA   : SIMDVectorRShiftBHSDTied<1, 0b00110, "ursra",
5629                TriOpFrag<(add node:$LHS,
5630                               (AArch64urshri node:$MHS, node:$RHS))> >;
5631defm USHLL   : SIMDVectorLShiftLongBHSD<1, 0b10100, "ushll",
5632                BinOpFrag<(AArch64vshl (zext node:$LHS), node:$RHS)>>;
5633defm USHR    : SIMDVectorRShiftBHSD<1, 0b00000, "ushr", AArch64vlshr>;
5634defm USRA    : SIMDVectorRShiftBHSDTied<1, 0b00010, "usra",
5635                TriOpFrag<(add node:$LHS, (AArch64vlshr node:$MHS, node:$RHS))> >;
5636
5637// SHRN patterns for when a logical right shift was used instead of arithmetic
5638// (the immediate guarantees no sign bits actually end up in the result so it
5639// doesn't matter).
5640def : Pat<(v8i8 (trunc (AArch64vlshr (v8i16 V128:$Rn), vecshiftR16Narrow:$imm))),
5641          (SHRNv8i8_shift V128:$Rn, vecshiftR16Narrow:$imm)>;
5642def : Pat<(v4i16 (trunc (AArch64vlshr (v4i32 V128:$Rn), vecshiftR32Narrow:$imm))),
5643          (SHRNv4i16_shift V128:$Rn, vecshiftR32Narrow:$imm)>;
5644def : Pat<(v2i32 (trunc (AArch64vlshr (v2i64 V128:$Rn), vecshiftR64Narrow:$imm))),
5645          (SHRNv2i32_shift V128:$Rn, vecshiftR64Narrow:$imm)>;
5646
5647def : Pat<(v16i8 (concat_vectors (v8i8 V64:$Rd),
5648                                 (trunc (AArch64vlshr (v8i16 V128:$Rn),
5649                                                    vecshiftR16Narrow:$imm)))),
5650          (SHRNv16i8_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
5651                           V128:$Rn, vecshiftR16Narrow:$imm)>;
5652def : Pat<(v8i16 (concat_vectors (v4i16 V64:$Rd),
5653                                 (trunc (AArch64vlshr (v4i32 V128:$Rn),
5654                                                    vecshiftR32Narrow:$imm)))),
5655          (SHRNv8i16_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
5656                           V128:$Rn, vecshiftR32Narrow:$imm)>;
5657def : Pat<(v4i32 (concat_vectors (v2i32 V64:$Rd),
5658                                 (trunc (AArch64vlshr (v2i64 V128:$Rn),
5659                                                    vecshiftR64Narrow:$imm)))),
5660          (SHRNv4i32_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
5661                           V128:$Rn, vecshiftR32Narrow:$imm)>;
5662
5663// Vector sign and zero extensions are implemented with SSHLL and USSHLL.
5664// Anyexts are implemented as zexts.
5665def : Pat<(v8i16 (sext   (v8i8 V64:$Rn))),  (SSHLLv8i8_shift  V64:$Rn, (i32 0))>;
5666def : Pat<(v8i16 (zext   (v8i8 V64:$Rn))),  (USHLLv8i8_shift  V64:$Rn, (i32 0))>;
5667def : Pat<(v8i16 (anyext (v8i8 V64:$Rn))),  (USHLLv8i8_shift  V64:$Rn, (i32 0))>;
5668def : Pat<(v4i32 (sext   (v4i16 V64:$Rn))), (SSHLLv4i16_shift V64:$Rn, (i32 0))>;
5669def : Pat<(v4i32 (zext   (v4i16 V64:$Rn))), (USHLLv4i16_shift V64:$Rn, (i32 0))>;
5670def : Pat<(v4i32 (anyext (v4i16 V64:$Rn))), (USHLLv4i16_shift V64:$Rn, (i32 0))>;
5671def : Pat<(v2i64 (sext   (v2i32 V64:$Rn))), (SSHLLv2i32_shift V64:$Rn, (i32 0))>;
5672def : Pat<(v2i64 (zext   (v2i32 V64:$Rn))), (USHLLv2i32_shift V64:$Rn, (i32 0))>;
5673def : Pat<(v2i64 (anyext (v2i32 V64:$Rn))), (USHLLv2i32_shift V64:$Rn, (i32 0))>;
5674// Also match an extend from the upper half of a 128 bit source register.
5675def : Pat<(v8i16 (anyext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
5676          (USHLLv16i8_shift V128:$Rn, (i32 0))>;
5677def : Pat<(v8i16 (zext   (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
5678          (USHLLv16i8_shift V128:$Rn, (i32 0))>;
5679def : Pat<(v8i16 (sext   (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
5680          (SSHLLv16i8_shift V128:$Rn, (i32 0))>;
5681def : Pat<(v4i32 (anyext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
5682          (USHLLv8i16_shift V128:$Rn, (i32 0))>;
5683def : Pat<(v4i32 (zext   (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
5684          (USHLLv8i16_shift V128:$Rn, (i32 0))>;
5685def : Pat<(v4i32 (sext   (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
5686          (SSHLLv8i16_shift V128:$Rn, (i32 0))>;
5687def : Pat<(v2i64 (anyext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
5688          (USHLLv4i32_shift V128:$Rn, (i32 0))>;
5689def : Pat<(v2i64 (zext   (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
5690          (USHLLv4i32_shift V128:$Rn, (i32 0))>;
5691def : Pat<(v2i64 (sext   (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
5692          (SSHLLv4i32_shift V128:$Rn, (i32 0))>;
5693
5694// Vector shift sxtl aliases
5695def : InstAlias<"sxtl.8h $dst, $src1",
5696                (SSHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
5697def : InstAlias<"sxtl $dst.8h, $src1.8b",
5698                (SSHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
5699def : InstAlias<"sxtl.4s $dst, $src1",
5700                (SSHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
5701def : InstAlias<"sxtl $dst.4s, $src1.4h",
5702                (SSHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
5703def : InstAlias<"sxtl.2d $dst, $src1",
5704                (SSHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
5705def : InstAlias<"sxtl $dst.2d, $src1.2s",
5706                (SSHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
5707
5708// Vector shift sxtl2 aliases
5709def : InstAlias<"sxtl2.8h $dst, $src1",
5710                (SSHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
5711def : InstAlias<"sxtl2 $dst.8h, $src1.16b",
5712                (SSHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
5713def : InstAlias<"sxtl2.4s $dst, $src1",
5714                (SSHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
5715def : InstAlias<"sxtl2 $dst.4s, $src1.8h",
5716                (SSHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
5717def : InstAlias<"sxtl2.2d $dst, $src1",
5718                (SSHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
5719def : InstAlias<"sxtl2 $dst.2d, $src1.4s",
5720                (SSHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
5721
5722// Vector shift uxtl aliases
5723def : InstAlias<"uxtl.8h $dst, $src1",
5724                (USHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
5725def : InstAlias<"uxtl $dst.8h, $src1.8b",
5726                (USHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
5727def : InstAlias<"uxtl.4s $dst, $src1",
5728                (USHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
5729def : InstAlias<"uxtl $dst.4s, $src1.4h",
5730                (USHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
5731def : InstAlias<"uxtl.2d $dst, $src1",
5732                (USHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
5733def : InstAlias<"uxtl $dst.2d, $src1.2s",
5734                (USHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
5735
5736// Vector shift uxtl2 aliases
5737def : InstAlias<"uxtl2.8h $dst, $src1",
5738                (USHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
5739def : InstAlias<"uxtl2 $dst.8h, $src1.16b",
5740                (USHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
5741def : InstAlias<"uxtl2.4s $dst, $src1",
5742                (USHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
5743def : InstAlias<"uxtl2 $dst.4s, $src1.8h",
5744                (USHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
5745def : InstAlias<"uxtl2.2d $dst, $src1",
5746                (USHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
5747def : InstAlias<"uxtl2 $dst.2d, $src1.4s",
5748                (USHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
5749
5750// If an integer is about to be converted to a floating point value,
5751// just load it on the floating point unit.
5752// These patterns are more complex because floating point loads do not
5753// support sign extension.
5754// The sign extension has to be explicitly added and is only supported for
5755// one step: byte-to-half, half-to-word, word-to-doubleword.
5756// SCVTF GPR -> FPR is 9 cycles.
5757// SCVTF FPR -> FPR is 4 cyclces.
5758// (sign extension with lengthen) SXTL FPR -> FPR is 2 cycles.
5759// Therefore, we can do 2 sign extensions and one SCVTF FPR -> FPR
5760// and still being faster.
5761// However, this is not good for code size.
5762// 8-bits -> float. 2 sizes step-up.
5763class SExtLoadi8CVTf32Pat<dag addrmode, dag INST>
5764  : Pat<(f32 (sint_to_fp (i32 (sextloadi8 addrmode)))),
5765        (SCVTFv1i32 (f32 (EXTRACT_SUBREG
5766                            (SSHLLv4i16_shift
5767                              (f64
5768                                (EXTRACT_SUBREG
5769                                  (SSHLLv8i8_shift
5770                                    (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
5771                                        INST,
5772                                        bsub),
5773                                    0),
5774                                  dsub)),
5775                               0),
5776                             ssub)))>,
5777    Requires<[NotForCodeSize, UseAlternateSExtLoadCVTF32]>;
5778
5779def : SExtLoadi8CVTf32Pat<(ro8.Wpat GPR64sp:$Rn, GPR32:$Rm, ro8.Wext:$ext),
5780                          (LDRBroW  GPR64sp:$Rn, GPR32:$Rm, ro8.Wext:$ext)>;
5781def : SExtLoadi8CVTf32Pat<(ro8.Xpat GPR64sp:$Rn, GPR64:$Rm, ro8.Xext:$ext),
5782                          (LDRBroX  GPR64sp:$Rn, GPR64:$Rm, ro8.Xext:$ext)>;
5783def : SExtLoadi8CVTf32Pat<(am_indexed8 GPR64sp:$Rn, uimm12s1:$offset),
5784                          (LDRBui GPR64sp:$Rn, uimm12s1:$offset)>;
5785def : SExtLoadi8CVTf32Pat<(am_unscaled8 GPR64sp:$Rn, simm9:$offset),
5786                          (LDURBi GPR64sp:$Rn, simm9:$offset)>;
5787
5788// 16-bits -> float. 1 size step-up.
5789class SExtLoadi16CVTf32Pat<dag addrmode, dag INST>
5790  : Pat<(f32 (sint_to_fp (i32 (sextloadi16 addrmode)))),
5791        (SCVTFv1i32 (f32 (EXTRACT_SUBREG
5792                            (SSHLLv4i16_shift
5793                                (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
5794                                  INST,
5795                                  hsub),
5796                                0),
5797                            ssub)))>, Requires<[NotForCodeSize]>;
5798
5799def : SExtLoadi16CVTf32Pat<(ro16.Wpat GPR64sp:$Rn, GPR32:$Rm, ro16.Wext:$ext),
5800                           (LDRHroW   GPR64sp:$Rn, GPR32:$Rm, ro16.Wext:$ext)>;
5801def : SExtLoadi16CVTf32Pat<(ro16.Xpat GPR64sp:$Rn, GPR64:$Rm, ro16.Xext:$ext),
5802                           (LDRHroX   GPR64sp:$Rn, GPR64:$Rm, ro16.Xext:$ext)>;
5803def : SExtLoadi16CVTf32Pat<(am_indexed16 GPR64sp:$Rn, uimm12s2:$offset),
5804                           (LDRHui GPR64sp:$Rn, uimm12s2:$offset)>;
5805def : SExtLoadi16CVTf32Pat<(am_unscaled16 GPR64sp:$Rn, simm9:$offset),
5806                           (LDURHi GPR64sp:$Rn, simm9:$offset)>;
5807
5808// 32-bits to 32-bits are handled in target specific dag combine:
5809// performIntToFpCombine.
5810// 64-bits integer to 32-bits floating point, not possible with
5811// SCVTF on floating point registers (both source and destination
5812// must have the same size).
5813
5814// Here are the patterns for 8, 16, 32, and 64-bits to double.
5815// 8-bits -> double. 3 size step-up: give up.
5816// 16-bits -> double. 2 size step.
5817class SExtLoadi16CVTf64Pat<dag addrmode, dag INST>
5818  : Pat <(f64 (sint_to_fp (i32 (sextloadi16 addrmode)))),
5819           (SCVTFv1i64 (f64 (EXTRACT_SUBREG
5820                              (SSHLLv2i32_shift
5821                                 (f64
5822                                  (EXTRACT_SUBREG
5823                                    (SSHLLv4i16_shift
5824                                      (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
5825                                        INST,
5826                                        hsub),
5827                                     0),
5828                                   dsub)),
5829                               0),
5830                             dsub)))>,
5831    Requires<[NotForCodeSize, UseAlternateSExtLoadCVTF32]>;
5832
5833def : SExtLoadi16CVTf64Pat<(ro16.Wpat GPR64sp:$Rn, GPR32:$Rm, ro16.Wext:$ext),
5834                           (LDRHroW GPR64sp:$Rn, GPR32:$Rm, ro16.Wext:$ext)>;
5835def : SExtLoadi16CVTf64Pat<(ro16.Xpat GPR64sp:$Rn, GPR64:$Rm, ro16.Xext:$ext),
5836                           (LDRHroX GPR64sp:$Rn, GPR64:$Rm, ro16.Xext:$ext)>;
5837def : SExtLoadi16CVTf64Pat<(am_indexed16 GPR64sp:$Rn, uimm12s2:$offset),
5838                           (LDRHui GPR64sp:$Rn, uimm12s2:$offset)>;
5839def : SExtLoadi16CVTf64Pat<(am_unscaled16 GPR64sp:$Rn, simm9:$offset),
5840                           (LDURHi GPR64sp:$Rn, simm9:$offset)>;
5841// 32-bits -> double. 1 size step-up.
5842class SExtLoadi32CVTf64Pat<dag addrmode, dag INST>
5843  : Pat <(f64 (sint_to_fp (i32 (load addrmode)))),
5844           (SCVTFv1i64 (f64 (EXTRACT_SUBREG
5845                              (SSHLLv2i32_shift
5846                                (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
5847                                  INST,
5848                                  ssub),
5849                               0),
5850                             dsub)))>, Requires<[NotForCodeSize]>;
5851
5852def : SExtLoadi32CVTf64Pat<(ro32.Wpat GPR64sp:$Rn, GPR32:$Rm, ro32.Wext:$ext),
5853                           (LDRSroW GPR64sp:$Rn, GPR32:$Rm, ro32.Wext:$ext)>;
5854def : SExtLoadi32CVTf64Pat<(ro32.Xpat GPR64sp:$Rn, GPR64:$Rm, ro32.Xext:$ext),
5855                           (LDRSroX GPR64sp:$Rn, GPR64:$Rm, ro32.Xext:$ext)>;
5856def : SExtLoadi32CVTf64Pat<(am_indexed32 GPR64sp:$Rn, uimm12s4:$offset),
5857                           (LDRSui GPR64sp:$Rn, uimm12s4:$offset)>;
5858def : SExtLoadi32CVTf64Pat<(am_unscaled32 GPR64sp:$Rn, simm9:$offset),
5859                           (LDURSi GPR64sp:$Rn, simm9:$offset)>;
5860
5861// 64-bits -> double are handled in target specific dag combine:
5862// performIntToFpCombine.
5863
5864
5865//----------------------------------------------------------------------------
5866// AdvSIMD Load-Store Structure
5867//----------------------------------------------------------------------------
5868defm LD1 : SIMDLd1Multiple<"ld1">;
5869defm LD2 : SIMDLd2Multiple<"ld2">;
5870defm LD3 : SIMDLd3Multiple<"ld3">;
5871defm LD4 : SIMDLd4Multiple<"ld4">;
5872
5873defm ST1 : SIMDSt1Multiple<"st1">;
5874defm ST2 : SIMDSt2Multiple<"st2">;
5875defm ST3 : SIMDSt3Multiple<"st3">;
5876defm ST4 : SIMDSt4Multiple<"st4">;
5877
5878class Ld1Pat<ValueType ty, Instruction INST>
5879  : Pat<(ty (load GPR64sp:$Rn)), (INST GPR64sp:$Rn)>;
5880
5881def : Ld1Pat<v16i8, LD1Onev16b>;
5882def : Ld1Pat<v8i16, LD1Onev8h>;
5883def : Ld1Pat<v4i32, LD1Onev4s>;
5884def : Ld1Pat<v2i64, LD1Onev2d>;
5885def : Ld1Pat<v8i8,  LD1Onev8b>;
5886def : Ld1Pat<v4i16, LD1Onev4h>;
5887def : Ld1Pat<v2i32, LD1Onev2s>;
5888def : Ld1Pat<v1i64, LD1Onev1d>;
5889
5890class St1Pat<ValueType ty, Instruction INST>
5891  : Pat<(store ty:$Vt, GPR64sp:$Rn),
5892        (INST ty:$Vt, GPR64sp:$Rn)>;
5893
5894def : St1Pat<v16i8, ST1Onev16b>;
5895def : St1Pat<v8i16, ST1Onev8h>;
5896def : St1Pat<v4i32, ST1Onev4s>;
5897def : St1Pat<v2i64, ST1Onev2d>;
5898def : St1Pat<v8i8,  ST1Onev8b>;
5899def : St1Pat<v4i16, ST1Onev4h>;
5900def : St1Pat<v2i32, ST1Onev2s>;
5901def : St1Pat<v1i64, ST1Onev1d>;
5902
5903//---
5904// Single-element
5905//---
5906
5907defm LD1R          : SIMDLdR<0, 0b110, 0, "ld1r", "One", 1, 2, 4, 8>;
5908defm LD2R          : SIMDLdR<1, 0b110, 0, "ld2r", "Two", 2, 4, 8, 16>;
5909defm LD3R          : SIMDLdR<0, 0b111, 0, "ld3r", "Three", 3, 6, 12, 24>;
5910defm LD4R          : SIMDLdR<1, 0b111, 0, "ld4r", "Four", 4, 8, 16, 32>;
5911let mayLoad = 1, hasSideEffects = 0 in {
5912defm LD1 : SIMDLdSingleBTied<0, 0b000,       "ld1", VecListOneb,   GPR64pi1>;
5913defm LD1 : SIMDLdSingleHTied<0, 0b010, 0,    "ld1", VecListOneh,   GPR64pi2>;
5914defm LD1 : SIMDLdSingleSTied<0, 0b100, 0b00, "ld1", VecListOnes,   GPR64pi4>;
5915defm LD1 : SIMDLdSingleDTied<0, 0b100, 0b01, "ld1", VecListOned,   GPR64pi8>;
5916defm LD2 : SIMDLdSingleBTied<1, 0b000,       "ld2", VecListTwob,   GPR64pi2>;
5917defm LD2 : SIMDLdSingleHTied<1, 0b010, 0,    "ld2", VecListTwoh,   GPR64pi4>;
5918defm LD2 : SIMDLdSingleSTied<1, 0b100, 0b00, "ld2", VecListTwos,   GPR64pi8>;
5919defm LD2 : SIMDLdSingleDTied<1, 0b100, 0b01, "ld2", VecListTwod,   GPR64pi16>;
5920defm LD3 : SIMDLdSingleBTied<0, 0b001,       "ld3", VecListThreeb, GPR64pi3>;
5921defm LD3 : SIMDLdSingleHTied<0, 0b011, 0,    "ld3", VecListThreeh, GPR64pi6>;
5922defm LD3 : SIMDLdSingleSTied<0, 0b101, 0b00, "ld3", VecListThrees, GPR64pi12>;
5923defm LD3 : SIMDLdSingleDTied<0, 0b101, 0b01, "ld3", VecListThreed, GPR64pi24>;
5924defm LD4 : SIMDLdSingleBTied<1, 0b001,       "ld4", VecListFourb,  GPR64pi4>;
5925defm LD4 : SIMDLdSingleHTied<1, 0b011, 0,    "ld4", VecListFourh,  GPR64pi8>;
5926defm LD4 : SIMDLdSingleSTied<1, 0b101, 0b00, "ld4", VecListFours,  GPR64pi16>;
5927defm LD4 : SIMDLdSingleDTied<1, 0b101, 0b01, "ld4", VecListFourd,  GPR64pi32>;
5928}
5929
5930def : Pat<(v8i8 (AArch64dup (i32 (extloadi8 GPR64sp:$Rn)))),
5931          (LD1Rv8b GPR64sp:$Rn)>;
5932def : Pat<(v16i8 (AArch64dup (i32 (extloadi8 GPR64sp:$Rn)))),
5933          (LD1Rv16b GPR64sp:$Rn)>;
5934def : Pat<(v4i16 (AArch64dup (i32 (extloadi16 GPR64sp:$Rn)))),
5935          (LD1Rv4h GPR64sp:$Rn)>;
5936def : Pat<(v8i16 (AArch64dup (i32 (extloadi16 GPR64sp:$Rn)))),
5937          (LD1Rv8h GPR64sp:$Rn)>;
5938def : Pat<(v2i32 (AArch64dup (i32 (load GPR64sp:$Rn)))),
5939          (LD1Rv2s GPR64sp:$Rn)>;
5940def : Pat<(v4i32 (AArch64dup (i32 (load GPR64sp:$Rn)))),
5941          (LD1Rv4s GPR64sp:$Rn)>;
5942def : Pat<(v2i64 (AArch64dup (i64 (load GPR64sp:$Rn)))),
5943          (LD1Rv2d GPR64sp:$Rn)>;
5944def : Pat<(v1i64 (AArch64dup (i64 (load GPR64sp:$Rn)))),
5945          (LD1Rv1d GPR64sp:$Rn)>;
5946// Grab the floating point version too
5947def : Pat<(v2f32 (AArch64dup (f32 (load GPR64sp:$Rn)))),
5948          (LD1Rv2s GPR64sp:$Rn)>;
5949def : Pat<(v4f32 (AArch64dup (f32 (load GPR64sp:$Rn)))),
5950          (LD1Rv4s GPR64sp:$Rn)>;
5951def : Pat<(v2f64 (AArch64dup (f64 (load GPR64sp:$Rn)))),
5952          (LD1Rv2d GPR64sp:$Rn)>;
5953def : Pat<(v1f64 (AArch64dup (f64 (load GPR64sp:$Rn)))),
5954          (LD1Rv1d GPR64sp:$Rn)>;
5955def : Pat<(v4f16 (AArch64dup (f16 (load GPR64sp:$Rn)))),
5956          (LD1Rv4h GPR64sp:$Rn)>;
5957def : Pat<(v8f16 (AArch64dup (f16 (load GPR64sp:$Rn)))),
5958          (LD1Rv8h GPR64sp:$Rn)>;
5959
5960class Ld1Lane128Pat<SDPatternOperator scalar_load, Operand VecIndex,
5961                    ValueType VTy, ValueType STy, Instruction LD1>
5962  : Pat<(vector_insert (VTy VecListOne128:$Rd),
5963           (STy (scalar_load GPR64sp:$Rn)), VecIndex:$idx),
5964        (LD1 VecListOne128:$Rd, VecIndex:$idx, GPR64sp:$Rn)>;
5965
5966def : Ld1Lane128Pat<extloadi8,  VectorIndexB, v16i8, i32, LD1i8>;
5967def : Ld1Lane128Pat<extloadi16, VectorIndexH, v8i16, i32, LD1i16>;
5968def : Ld1Lane128Pat<load,       VectorIndexS, v4i32, i32, LD1i32>;
5969def : Ld1Lane128Pat<load,       VectorIndexS, v4f32, f32, LD1i32>;
5970def : Ld1Lane128Pat<load,       VectorIndexD, v2i64, i64, LD1i64>;
5971def : Ld1Lane128Pat<load,       VectorIndexD, v2f64, f64, LD1i64>;
5972def : Ld1Lane128Pat<load,       VectorIndexH, v8f16, f16, LD1i16>;
5973
5974class Ld1Lane64Pat<SDPatternOperator scalar_load, Operand VecIndex,
5975                   ValueType VTy, ValueType STy, Instruction LD1>
5976  : Pat<(vector_insert (VTy VecListOne64:$Rd),
5977           (STy (scalar_load GPR64sp:$Rn)), VecIndex:$idx),
5978        (EXTRACT_SUBREG
5979            (LD1 (SUBREG_TO_REG (i32 0), VecListOne64:$Rd, dsub),
5980                          VecIndex:$idx, GPR64sp:$Rn),
5981            dsub)>;
5982
5983def : Ld1Lane64Pat<extloadi8,  VectorIndexB, v8i8,  i32, LD1i8>;
5984def : Ld1Lane64Pat<extloadi16, VectorIndexH, v4i16, i32, LD1i16>;
5985def : Ld1Lane64Pat<load,       VectorIndexS, v2i32, i32, LD1i32>;
5986def : Ld1Lane64Pat<load,       VectorIndexS, v2f32, f32, LD1i32>;
5987def : Ld1Lane64Pat<load,       VectorIndexH, v4f16, f16, LD1i16>;
5988
5989
5990defm LD1 : SIMDLdSt1SingleAliases<"ld1">;
5991defm LD2 : SIMDLdSt2SingleAliases<"ld2">;
5992defm LD3 : SIMDLdSt3SingleAliases<"ld3">;
5993defm LD4 : SIMDLdSt4SingleAliases<"ld4">;
5994
5995// Stores
5996defm ST1 : SIMDStSingleB<0, 0b000,       "st1", VecListOneb, GPR64pi1>;
5997defm ST1 : SIMDStSingleH<0, 0b010, 0,    "st1", VecListOneh, GPR64pi2>;
5998defm ST1 : SIMDStSingleS<0, 0b100, 0b00, "st1", VecListOnes, GPR64pi4>;
5999defm ST1 : SIMDStSingleD<0, 0b100, 0b01, "st1", VecListOned, GPR64pi8>;
6000
6001let AddedComplexity = 19 in
6002class St1Lane128Pat<SDPatternOperator scalar_store, Operand VecIndex,
6003                    ValueType VTy, ValueType STy, Instruction ST1>
6004  : Pat<(scalar_store
6005             (STy (vector_extract (VTy VecListOne128:$Vt), VecIndex:$idx)),
6006             GPR64sp:$Rn),
6007        (ST1 VecListOne128:$Vt, VecIndex:$idx, GPR64sp:$Rn)>;
6008
6009def : St1Lane128Pat<truncstorei8,  VectorIndexB, v16i8, i32, ST1i8>;
6010def : St1Lane128Pat<truncstorei16, VectorIndexH, v8i16, i32, ST1i16>;
6011def : St1Lane128Pat<store,         VectorIndexS, v4i32, i32, ST1i32>;
6012def : St1Lane128Pat<store,         VectorIndexS, v4f32, f32, ST1i32>;
6013def : St1Lane128Pat<store,         VectorIndexD, v2i64, i64, ST1i64>;
6014def : St1Lane128Pat<store,         VectorIndexD, v2f64, f64, ST1i64>;
6015def : St1Lane128Pat<store,         VectorIndexH, v8f16, f16, ST1i16>;
6016
6017let AddedComplexity = 19 in
6018class St1Lane64Pat<SDPatternOperator scalar_store, Operand VecIndex,
6019                   ValueType VTy, ValueType STy, Instruction ST1>
6020  : Pat<(scalar_store
6021             (STy (vector_extract (VTy VecListOne64:$Vt), VecIndex:$idx)),
6022             GPR64sp:$Rn),
6023        (ST1 (SUBREG_TO_REG (i32 0), VecListOne64:$Vt, dsub),
6024             VecIndex:$idx, GPR64sp:$Rn)>;
6025
6026def : St1Lane64Pat<truncstorei8,  VectorIndexB, v8i8, i32, ST1i8>;
6027def : St1Lane64Pat<truncstorei16, VectorIndexH, v4i16, i32, ST1i16>;
6028def : St1Lane64Pat<store,         VectorIndexS, v2i32, i32, ST1i32>;
6029def : St1Lane64Pat<store,         VectorIndexS, v2f32, f32, ST1i32>;
6030def : St1Lane64Pat<store,         VectorIndexH, v4f16, f16, ST1i16>;
6031
6032multiclass St1LanePost64Pat<SDPatternOperator scalar_store, Operand VecIndex,
6033                             ValueType VTy, ValueType STy, Instruction ST1,
6034                             int offset> {
6035  def : Pat<(scalar_store
6036              (STy (vector_extract (VTy VecListOne64:$Vt), VecIndex:$idx)),
6037              GPR64sp:$Rn, offset),
6038        (ST1 (SUBREG_TO_REG (i32 0), VecListOne64:$Vt, dsub),
6039             VecIndex:$idx, GPR64sp:$Rn, XZR)>;
6040
6041  def : Pat<(scalar_store
6042              (STy (vector_extract (VTy VecListOne64:$Vt), VecIndex:$idx)),
6043              GPR64sp:$Rn, GPR64:$Rm),
6044        (ST1 (SUBREG_TO_REG (i32 0), VecListOne64:$Vt, dsub),
6045             VecIndex:$idx, GPR64sp:$Rn, $Rm)>;
6046}
6047
6048defm : St1LanePost64Pat<post_truncsti8, VectorIndexB, v8i8, i32, ST1i8_POST, 1>;
6049defm : St1LanePost64Pat<post_truncsti16, VectorIndexH, v4i16, i32, ST1i16_POST,
6050                        2>;
6051defm : St1LanePost64Pat<post_store, VectorIndexS, v2i32, i32, ST1i32_POST, 4>;
6052defm : St1LanePost64Pat<post_store, VectorIndexS, v2f32, f32, ST1i32_POST, 4>;
6053defm : St1LanePost64Pat<post_store, VectorIndexD, v1i64, i64, ST1i64_POST, 8>;
6054defm : St1LanePost64Pat<post_store, VectorIndexD, v1f64, f64, ST1i64_POST, 8>;
6055defm : St1LanePost64Pat<post_store, VectorIndexH, v4f16, f16, ST1i16_POST, 2>;
6056
6057multiclass St1LanePost128Pat<SDPatternOperator scalar_store, Operand VecIndex,
6058                             ValueType VTy, ValueType STy, Instruction ST1,
6059                             int offset> {
6060  def : Pat<(scalar_store
6061              (STy (vector_extract (VTy VecListOne128:$Vt), VecIndex:$idx)),
6062              GPR64sp:$Rn, offset),
6063        (ST1 VecListOne128:$Vt, VecIndex:$idx, GPR64sp:$Rn, XZR)>;
6064
6065  def : Pat<(scalar_store
6066              (STy (vector_extract (VTy VecListOne128:$Vt), VecIndex:$idx)),
6067              GPR64sp:$Rn, GPR64:$Rm),
6068        (ST1 VecListOne128:$Vt, VecIndex:$idx, GPR64sp:$Rn, $Rm)>;
6069}
6070
6071defm : St1LanePost128Pat<post_truncsti8, VectorIndexB, v16i8, i32, ST1i8_POST,
6072                         1>;
6073defm : St1LanePost128Pat<post_truncsti16, VectorIndexH, v8i16, i32, ST1i16_POST,
6074                         2>;
6075defm : St1LanePost128Pat<post_store, VectorIndexS, v4i32, i32, ST1i32_POST, 4>;
6076defm : St1LanePost128Pat<post_store, VectorIndexS, v4f32, f32, ST1i32_POST, 4>;
6077defm : St1LanePost128Pat<post_store, VectorIndexD, v2i64, i64, ST1i64_POST, 8>;
6078defm : St1LanePost128Pat<post_store, VectorIndexD, v2f64, f64, ST1i64_POST, 8>;
6079defm : St1LanePost128Pat<post_store, VectorIndexH, v8f16, f16, ST1i16_POST, 2>;
6080
6081let mayStore = 1, hasSideEffects = 0 in {
6082defm ST2 : SIMDStSingleB<1, 0b000,       "st2", VecListTwob,   GPR64pi2>;
6083defm ST2 : SIMDStSingleH<1, 0b010, 0,    "st2", VecListTwoh,   GPR64pi4>;
6084defm ST2 : SIMDStSingleS<1, 0b100, 0b00, "st2", VecListTwos,   GPR64pi8>;
6085defm ST2 : SIMDStSingleD<1, 0b100, 0b01, "st2", VecListTwod,   GPR64pi16>;
6086defm ST3 : SIMDStSingleB<0, 0b001,       "st3", VecListThreeb, GPR64pi3>;
6087defm ST3 : SIMDStSingleH<0, 0b011, 0,    "st3", VecListThreeh, GPR64pi6>;
6088defm ST3 : SIMDStSingleS<0, 0b101, 0b00, "st3", VecListThrees, GPR64pi12>;
6089defm ST3 : SIMDStSingleD<0, 0b101, 0b01, "st3", VecListThreed, GPR64pi24>;
6090defm ST4 : SIMDStSingleB<1, 0b001,       "st4", VecListFourb,  GPR64pi4>;
6091defm ST4 : SIMDStSingleH<1, 0b011, 0,    "st4", VecListFourh,  GPR64pi8>;
6092defm ST4 : SIMDStSingleS<1, 0b101, 0b00, "st4", VecListFours,  GPR64pi16>;
6093defm ST4 : SIMDStSingleD<1, 0b101, 0b01, "st4", VecListFourd,  GPR64pi32>;
6094}
6095
6096defm ST1 : SIMDLdSt1SingleAliases<"st1">;
6097defm ST2 : SIMDLdSt2SingleAliases<"st2">;
6098defm ST3 : SIMDLdSt3SingleAliases<"st3">;
6099defm ST4 : SIMDLdSt4SingleAliases<"st4">;
6100
6101//----------------------------------------------------------------------------
6102// Crypto extensions
6103//----------------------------------------------------------------------------
6104
6105let Predicates = [HasAES] in {
6106def AESErr   : AESTiedInst<0b0100, "aese",   int_aarch64_crypto_aese>;
6107def AESDrr   : AESTiedInst<0b0101, "aesd",   int_aarch64_crypto_aesd>;
6108def AESMCrr  : AESInst<    0b0110, "aesmc",  int_aarch64_crypto_aesmc>;
6109def AESIMCrr : AESInst<    0b0111, "aesimc", int_aarch64_crypto_aesimc>;
6110}
6111
6112// Pseudo instructions for AESMCrr/AESIMCrr with a register constraint required
6113// for AES fusion on some CPUs.
6114let hasSideEffects = 0, mayStore = 0, mayLoad = 0 in {
6115def AESMCrrTied: Pseudo<(outs V128:$Rd), (ins V128:$Rn), [], "$Rn = $Rd">,
6116                        Sched<[WriteV]>;
6117def AESIMCrrTied: Pseudo<(outs V128:$Rd), (ins V128:$Rn), [], "$Rn = $Rd">,
6118                         Sched<[WriteV]>;
6119}
6120
6121// Only use constrained versions of AES(I)MC instructions if they are paired with
6122// AESE/AESD.
6123def : Pat<(v16i8 (int_aarch64_crypto_aesmc
6124            (v16i8 (int_aarch64_crypto_aese (v16i8 V128:$src1),
6125                                            (v16i8 V128:$src2))))),
6126          (v16i8 (AESMCrrTied (v16i8 (AESErr (v16i8 V128:$src1),
6127                                             (v16i8 V128:$src2)))))>,
6128          Requires<[HasFuseAES]>;
6129
6130def : Pat<(v16i8 (int_aarch64_crypto_aesimc
6131            (v16i8 (int_aarch64_crypto_aesd (v16i8 V128:$src1),
6132                                            (v16i8 V128:$src2))))),
6133          (v16i8 (AESIMCrrTied (v16i8 (AESDrr (v16i8 V128:$src1),
6134                                              (v16i8 V128:$src2)))))>,
6135          Requires<[HasFuseAES]>;
6136
6137let Predicates = [HasSHA2] in {
6138def SHA1Crrr     : SHATiedInstQSV<0b000, "sha1c",   int_aarch64_crypto_sha1c>;
6139def SHA1Prrr     : SHATiedInstQSV<0b001, "sha1p",   int_aarch64_crypto_sha1p>;
6140def SHA1Mrrr     : SHATiedInstQSV<0b010, "sha1m",   int_aarch64_crypto_sha1m>;
6141def SHA1SU0rrr   : SHATiedInstVVV<0b011, "sha1su0", int_aarch64_crypto_sha1su0>;
6142def SHA256Hrrr   : SHATiedInstQQV<0b100, "sha256h", int_aarch64_crypto_sha256h>;
6143def SHA256H2rrr  : SHATiedInstQQV<0b101, "sha256h2",int_aarch64_crypto_sha256h2>;
6144def SHA256SU1rrr :SHATiedInstVVV<0b110, "sha256su1",int_aarch64_crypto_sha256su1>;
6145
6146def SHA1Hrr     : SHAInstSS<    0b0000, "sha1h",    int_aarch64_crypto_sha1h>;
6147def SHA1SU1rr   : SHATiedInstVV<0b0001, "sha1su1",  int_aarch64_crypto_sha1su1>;
6148def SHA256SU0rr : SHATiedInstVV<0b0010, "sha256su0",int_aarch64_crypto_sha256su0>;
6149}
6150
6151//----------------------------------------------------------------------------
6152// Compiler-pseudos
6153//----------------------------------------------------------------------------
6154// FIXME: Like for X86, these should go in their own separate .td file.
6155
6156def def32 : PatLeaf<(i32 GPR32:$src), [{
6157  return isDef32(*N);
6158}]>;
6159
6160// In the case of a 32-bit def that is known to implicitly zero-extend,
6161// we can use a SUBREG_TO_REG.
6162def : Pat<(i64 (zext def32:$src)), (SUBREG_TO_REG (i64 0), GPR32:$src, sub_32)>;
6163
6164// For an anyext, we don't care what the high bits are, so we can perform an
6165// INSERT_SUBREF into an IMPLICIT_DEF.
6166def : Pat<(i64 (anyext GPR32:$src)),
6167          (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32)>;
6168
6169// When we need to explicitly zero-extend, we use a 32-bit MOV instruction and
6170// then assert the extension has happened.
6171def : Pat<(i64 (zext GPR32:$src)),
6172          (SUBREG_TO_REG (i32 0), (ORRWrs WZR, GPR32:$src, 0), sub_32)>;
6173
6174// To sign extend, we use a signed bitfield move instruction (SBFM) on the
6175// containing super-reg.
6176def : Pat<(i64 (sext GPR32:$src)),
6177   (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32), 0, 31)>;
6178def : Pat<(i64 (sext_inreg GPR64:$src, i32)), (SBFMXri GPR64:$src, 0, 31)>;
6179def : Pat<(i64 (sext_inreg GPR64:$src, i16)), (SBFMXri GPR64:$src, 0, 15)>;
6180def : Pat<(i64 (sext_inreg GPR64:$src, i8)),  (SBFMXri GPR64:$src, 0, 7)>;
6181def : Pat<(i64 (sext_inreg GPR64:$src, i1)),  (SBFMXri GPR64:$src, 0, 0)>;
6182def : Pat<(i32 (sext_inreg GPR32:$src, i16)), (SBFMWri GPR32:$src, 0, 15)>;
6183def : Pat<(i32 (sext_inreg GPR32:$src, i8)),  (SBFMWri GPR32:$src, 0, 7)>;
6184def : Pat<(i32 (sext_inreg GPR32:$src, i1)),  (SBFMWri GPR32:$src, 0, 0)>;
6185
6186def : Pat<(shl (sext_inreg GPR32:$Rn, i8), (i64 imm0_31:$imm)),
6187          (SBFMWri GPR32:$Rn, (i64 (i32shift_a       imm0_31:$imm)),
6188                              (i64 (i32shift_sext_i8 imm0_31:$imm)))>;
6189def : Pat<(shl (sext_inreg GPR64:$Rn, i8), (i64 imm0_63:$imm)),
6190          (SBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
6191                              (i64 (i64shift_sext_i8 imm0_63:$imm)))>;
6192
6193def : Pat<(shl (sext_inreg GPR32:$Rn, i16), (i64 imm0_31:$imm)),
6194          (SBFMWri GPR32:$Rn, (i64 (i32shift_a        imm0_31:$imm)),
6195                              (i64 (i32shift_sext_i16 imm0_31:$imm)))>;
6196def : Pat<(shl (sext_inreg GPR64:$Rn, i16), (i64 imm0_63:$imm)),
6197          (SBFMXri GPR64:$Rn, (i64 (i64shift_a        imm0_63:$imm)),
6198                              (i64 (i64shift_sext_i16 imm0_63:$imm)))>;
6199
6200def : Pat<(shl (i64 (sext GPR32:$Rn)), (i64 imm0_63:$imm)),
6201          (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$Rn, sub_32),
6202                   (i64 (i64shift_a        imm0_63:$imm)),
6203                   (i64 (i64shift_sext_i32 imm0_63:$imm)))>;
6204
6205// sra patterns have an AddedComplexity of 10, so make sure we have a higher
6206// AddedComplexity for the following patterns since we want to match sext + sra
6207// patterns before we attempt to match a single sra node.
6208let AddedComplexity = 20 in {
6209// We support all sext + sra combinations which preserve at least one bit of the
6210// original value which is to be sign extended. E.g. we support shifts up to
6211// bitwidth-1 bits.
6212def : Pat<(sra (sext_inreg GPR32:$Rn, i8), (i64 imm0_7:$imm)),
6213          (SBFMWri GPR32:$Rn, (i64 imm0_7:$imm), 7)>;
6214def : Pat<(sra (sext_inreg GPR64:$Rn, i8), (i64 imm0_7:$imm)),
6215          (SBFMXri GPR64:$Rn, (i64 imm0_7:$imm), 7)>;
6216
6217def : Pat<(sra (sext_inreg GPR32:$Rn, i16), (i64 imm0_15:$imm)),
6218          (SBFMWri GPR32:$Rn, (i64 imm0_15:$imm), 15)>;
6219def : Pat<(sra (sext_inreg GPR64:$Rn, i16), (i64 imm0_15:$imm)),
6220          (SBFMXri GPR64:$Rn, (i64 imm0_15:$imm), 15)>;
6221
6222def : Pat<(sra (i64 (sext GPR32:$Rn)), (i64 imm0_31:$imm)),
6223          (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$Rn, sub_32),
6224                   (i64 imm0_31:$imm), 31)>;
6225} // AddedComplexity = 20
6226
6227// To truncate, we can simply extract from a subregister.
6228def : Pat<(i32 (trunc GPR64sp:$src)),
6229          (i32 (EXTRACT_SUBREG GPR64sp:$src, sub_32))>;
6230
6231// __builtin_trap() uses the BRK instruction on AArch64.
6232def : Pat<(trap), (BRK 1)>;
6233def : Pat<(debugtrap), (BRK 0xF000)>, Requires<[IsWindows]>;
6234
6235// Multiply high patterns which multiply the lower subvector using smull/umull
6236// and the upper subvector with smull2/umull2. Then shuffle the high the high
6237// part of both results together.
6238def : Pat<(v16i8 (mulhs V128:$Rn, V128:$Rm)),
6239          (UZP2v16i8
6240           (SMULLv8i8_v8i16 (EXTRACT_SUBREG V128:$Rn, dsub),
6241                            (EXTRACT_SUBREG V128:$Rm, dsub)),
6242           (SMULLv16i8_v8i16 V128:$Rn, V128:$Rm))>;
6243def : Pat<(v8i16 (mulhs V128:$Rn, V128:$Rm)),
6244          (UZP2v8i16
6245           (SMULLv4i16_v4i32 (EXTRACT_SUBREG V128:$Rn, dsub),
6246                             (EXTRACT_SUBREG V128:$Rm, dsub)),
6247           (SMULLv8i16_v4i32 V128:$Rn, V128:$Rm))>;
6248def : Pat<(v4i32 (mulhs V128:$Rn, V128:$Rm)),
6249          (UZP2v4i32
6250           (SMULLv2i32_v2i64 (EXTRACT_SUBREG V128:$Rn, dsub),
6251                             (EXTRACT_SUBREG V128:$Rm, dsub)),
6252           (SMULLv4i32_v2i64 V128:$Rn, V128:$Rm))>;
6253
6254def : Pat<(v16i8 (mulhu V128:$Rn, V128:$Rm)),
6255          (UZP2v16i8
6256           (UMULLv8i8_v8i16 (EXTRACT_SUBREG V128:$Rn, dsub),
6257                            (EXTRACT_SUBREG V128:$Rm, dsub)),
6258           (UMULLv16i8_v8i16 V128:$Rn, V128:$Rm))>;
6259def : Pat<(v8i16 (mulhu V128:$Rn, V128:$Rm)),
6260          (UZP2v8i16
6261           (UMULLv4i16_v4i32 (EXTRACT_SUBREG V128:$Rn, dsub),
6262                             (EXTRACT_SUBREG V128:$Rm, dsub)),
6263           (UMULLv8i16_v4i32 V128:$Rn, V128:$Rm))>;
6264def : Pat<(v4i32 (mulhu V128:$Rn, V128:$Rm)),
6265          (UZP2v4i32
6266           (UMULLv2i32_v2i64 (EXTRACT_SUBREG V128:$Rn, dsub),
6267                             (EXTRACT_SUBREG V128:$Rm, dsub)),
6268           (UMULLv4i32_v2i64 V128:$Rn, V128:$Rm))>;
6269
6270// Conversions within AdvSIMD types in the same register size are free.
6271// But because we need a consistent lane ordering, in big endian many
6272// conversions require one or more REV instructions.
6273//
6274// Consider a simple memory load followed by a bitconvert then a store.
6275//   v0 = load v2i32
6276//   v1 = BITCAST v2i32 v0 to v4i16
6277//        store v4i16 v2
6278//
6279// In big endian mode every memory access has an implicit byte swap. LDR and
6280// STR do a 64-bit byte swap, whereas LD1/ST1 do a byte swap per lane - that
6281// is, they treat the vector as a sequence of elements to be byte-swapped.
6282// The two pairs of instructions are fundamentally incompatible. We've decided
6283// to use LD1/ST1 only to simplify compiler implementation.
6284//
6285// LD1/ST1 perform the equivalent of a sequence of LDR/STR + REV. This makes
6286// the original code sequence:
6287//   v0 = load v2i32
6288//   v1 = REV v2i32                  (implicit)
6289//   v2 = BITCAST v2i32 v1 to v4i16
6290//   v3 = REV v4i16 v2               (implicit)
6291//        store v4i16 v3
6292//
6293// But this is now broken - the value stored is different to the value loaded
6294// due to lane reordering. To fix this, on every BITCAST we must perform two
6295// other REVs:
6296//   v0 = load v2i32
6297//   v1 = REV v2i32                  (implicit)
6298//   v2 = REV v2i32
6299//   v3 = BITCAST v2i32 v2 to v4i16
6300//   v4 = REV v4i16
6301//   v5 = REV v4i16 v4               (implicit)
6302//        store v4i16 v5
6303//
6304// This means an extra two instructions, but actually in most cases the two REV
6305// instructions can be combined into one. For example:
6306//   (REV64_2s (REV64_4h X)) === (REV32_4h X)
6307//
6308// There is also no 128-bit REV instruction. This must be synthesized with an
6309// EXT instruction.
6310//
6311// Most bitconverts require some sort of conversion. The only exceptions are:
6312//   a) Identity conversions -  vNfX <-> vNiX
6313//   b) Single-lane-to-scalar - v1fX <-> fX or v1iX <-> iX
6314//
6315
6316// Natural vector casts (64 bit)
6317def : Pat<(v8i8 (AArch64NvCast (v2i32 FPR64:$src))), (v8i8 FPR64:$src)>;
6318def : Pat<(v4i16 (AArch64NvCast (v2i32 FPR64:$src))), (v4i16 FPR64:$src)>;
6319def : Pat<(v4f16 (AArch64NvCast (v2i32 FPR64:$src))), (v4f16 FPR64:$src)>;
6320def : Pat<(v2i32 (AArch64NvCast (v2i32 FPR64:$src))), (v2i32 FPR64:$src)>;
6321def : Pat<(v2f32 (AArch64NvCast (v2i32 FPR64:$src))), (v2f32 FPR64:$src)>;
6322def : Pat<(v1i64 (AArch64NvCast (v2i32 FPR64:$src))), (v1i64 FPR64:$src)>;
6323
6324def : Pat<(v8i8 (AArch64NvCast (v4i16 FPR64:$src))), (v8i8 FPR64:$src)>;
6325def : Pat<(v4i16 (AArch64NvCast (v4i16 FPR64:$src))), (v4i16 FPR64:$src)>;
6326def : Pat<(v4f16 (AArch64NvCast (v4i16 FPR64:$src))), (v4f16 FPR64:$src)>;
6327def : Pat<(v2i32 (AArch64NvCast (v4i16 FPR64:$src))), (v2i32 FPR64:$src)>;
6328def : Pat<(v1i64 (AArch64NvCast (v4i16 FPR64:$src))), (v1i64 FPR64:$src)>;
6329
6330def : Pat<(v8i8 (AArch64NvCast (v8i8 FPR64:$src))), (v8i8 FPR64:$src)>;
6331def : Pat<(v4i16 (AArch64NvCast (v8i8 FPR64:$src))), (v4i16 FPR64:$src)>;
6332def : Pat<(v4f16 (AArch64NvCast (v8i8 FPR64:$src))), (v4f16 FPR64:$src)>;
6333def : Pat<(v2i32 (AArch64NvCast (v8i8 FPR64:$src))), (v2i32 FPR64:$src)>;
6334def : Pat<(v2f32 (AArch64NvCast (v8i8 FPR64:$src))), (v2f32 FPR64:$src)>;
6335def : Pat<(v1i64 (AArch64NvCast (v8i8 FPR64:$src))), (v1i64 FPR64:$src)>;
6336
6337def : Pat<(v8i8 (AArch64NvCast (f64 FPR64:$src))), (v8i8 FPR64:$src)>;
6338def : Pat<(v4i16 (AArch64NvCast (f64 FPR64:$src))), (v4i16 FPR64:$src)>;
6339def : Pat<(v4f16 (AArch64NvCast (f64 FPR64:$src))), (v4f16 FPR64:$src)>;
6340def : Pat<(v2i32 (AArch64NvCast (f64 FPR64:$src))), (v2i32 FPR64:$src)>;
6341def : Pat<(v2f32 (AArch64NvCast (f64 FPR64:$src))), (v2f32 FPR64:$src)>;
6342def : Pat<(v1i64 (AArch64NvCast (f64 FPR64:$src))), (v1i64 FPR64:$src)>;
6343def : Pat<(v1f64 (AArch64NvCast (f64 FPR64:$src))), (v1f64 FPR64:$src)>;
6344
6345def : Pat<(v8i8 (AArch64NvCast (v2f32 FPR64:$src))), (v8i8 FPR64:$src)>;
6346def : Pat<(v4i16 (AArch64NvCast (v2f32 FPR64:$src))), (v4i16 FPR64:$src)>;
6347def : Pat<(v2i32 (AArch64NvCast (v2f32 FPR64:$src))), (v2i32 FPR64:$src)>;
6348def : Pat<(v2f32 (AArch64NvCast (v2f32 FPR64:$src))), (v2f32 FPR64:$src)>;
6349def : Pat<(v1i64 (AArch64NvCast (v2f32 FPR64:$src))), (v1i64 FPR64:$src)>;
6350def : Pat<(v1f64 (AArch64NvCast (v2f32 FPR64:$src))), (v1f64 FPR64:$src)>;
6351
6352// Natural vector casts (128 bit)
6353def : Pat<(v16i8 (AArch64NvCast (v4i32 FPR128:$src))), (v16i8 FPR128:$src)>;
6354def : Pat<(v8i16 (AArch64NvCast (v4i32 FPR128:$src))), (v8i16 FPR128:$src)>;
6355def : Pat<(v8f16 (AArch64NvCast (v4i32 FPR128:$src))), (v8f16 FPR128:$src)>;
6356def : Pat<(v4i32 (AArch64NvCast (v4i32 FPR128:$src))), (v4i32 FPR128:$src)>;
6357def : Pat<(v4f32 (AArch64NvCast (v4i32 FPR128:$src))), (v4f32 FPR128:$src)>;
6358def : Pat<(v2i64 (AArch64NvCast (v4i32 FPR128:$src))), (v2i64 FPR128:$src)>;
6359def : Pat<(v2f64 (AArch64NvCast (v4i32 FPR128:$src))), (v2f64 FPR128:$src)>;
6360
6361def : Pat<(v16i8 (AArch64NvCast (v8i16 FPR128:$src))), (v16i8 FPR128:$src)>;
6362def : Pat<(v8i16 (AArch64NvCast (v8i16 FPR128:$src))), (v8i16 FPR128:$src)>;
6363def : Pat<(v8f16 (AArch64NvCast (v8i16 FPR128:$src))), (v8f16 FPR128:$src)>;
6364def : Pat<(v4i32 (AArch64NvCast (v8i16 FPR128:$src))), (v4i32 FPR128:$src)>;
6365def : Pat<(v2i64 (AArch64NvCast (v8i16 FPR128:$src))), (v2i64 FPR128:$src)>;
6366def : Pat<(v4f32 (AArch64NvCast (v8i16 FPR128:$src))), (v4f32 FPR128:$src)>;
6367def : Pat<(v2f64 (AArch64NvCast (v8i16 FPR128:$src))), (v2f64 FPR128:$src)>;
6368
6369def : Pat<(v16i8 (AArch64NvCast (v16i8 FPR128:$src))), (v16i8 FPR128:$src)>;
6370def : Pat<(v8i16 (AArch64NvCast (v16i8 FPR128:$src))), (v8i16 FPR128:$src)>;
6371def : Pat<(v8f16 (AArch64NvCast (v16i8 FPR128:$src))), (v8f16 FPR128:$src)>;
6372def : Pat<(v4i32 (AArch64NvCast (v16i8 FPR128:$src))), (v4i32 FPR128:$src)>;
6373def : Pat<(v2i64 (AArch64NvCast (v16i8 FPR128:$src))), (v2i64 FPR128:$src)>;
6374def : Pat<(v4f32 (AArch64NvCast (v16i8 FPR128:$src))), (v4f32 FPR128:$src)>;
6375def : Pat<(v2f64 (AArch64NvCast (v16i8 FPR128:$src))), (v2f64 FPR128:$src)>;
6376
6377def : Pat<(v16i8 (AArch64NvCast (v2i64 FPR128:$src))), (v16i8 FPR128:$src)>;
6378def : Pat<(v8i16 (AArch64NvCast (v2i64 FPR128:$src))), (v8i16 FPR128:$src)>;
6379def : Pat<(v8f16 (AArch64NvCast (v2i64 FPR128:$src))), (v8f16 FPR128:$src)>;
6380def : Pat<(v4i32 (AArch64NvCast (v2i64 FPR128:$src))), (v4i32 FPR128:$src)>;
6381def : Pat<(v2i64 (AArch64NvCast (v2i64 FPR128:$src))), (v2i64 FPR128:$src)>;
6382def : Pat<(v4f32 (AArch64NvCast (v2i64 FPR128:$src))), (v4f32 FPR128:$src)>;
6383def : Pat<(v2f64 (AArch64NvCast (v2i64 FPR128:$src))), (v2f64 FPR128:$src)>;
6384
6385def : Pat<(v16i8 (AArch64NvCast (v4f32 FPR128:$src))), (v16i8 FPR128:$src)>;
6386def : Pat<(v8i16 (AArch64NvCast (v4f32 FPR128:$src))), (v8i16 FPR128:$src)>;
6387def : Pat<(v4i32 (AArch64NvCast (v4f32 FPR128:$src))), (v4i32 FPR128:$src)>;
6388def : Pat<(v4f32 (AArch64NvCast (v4f32 FPR128:$src))), (v4f32 FPR128:$src)>;
6389def : Pat<(v2i64 (AArch64NvCast (v4f32 FPR128:$src))), (v2i64 FPR128:$src)>;
6390def : Pat<(v8f16 (AArch64NvCast (v4f32 FPR128:$src))), (v8f16 FPR128:$src)>;
6391def : Pat<(v2f64 (AArch64NvCast (v4f32 FPR128:$src))), (v2f64 FPR128:$src)>;
6392
6393def : Pat<(v16i8 (AArch64NvCast (v2f64 FPR128:$src))), (v16i8 FPR128:$src)>;
6394def : Pat<(v8i16 (AArch64NvCast (v2f64 FPR128:$src))), (v8i16 FPR128:$src)>;
6395def : Pat<(v4i32 (AArch64NvCast (v2f64 FPR128:$src))), (v4i32 FPR128:$src)>;
6396def : Pat<(v2i64 (AArch64NvCast (v2f64 FPR128:$src))), (v2i64 FPR128:$src)>;
6397def : Pat<(v2f64 (AArch64NvCast (v2f64 FPR128:$src))), (v2f64 FPR128:$src)>;
6398def : Pat<(v8f16 (AArch64NvCast (v2f64 FPR128:$src))), (v8f16 FPR128:$src)>;
6399def : Pat<(v4f32 (AArch64NvCast (v2f64 FPR128:$src))), (v4f32 FPR128:$src)>;
6400
6401let Predicates = [IsLE] in {
6402def : Pat<(v8i8  (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
6403def : Pat<(v4i16 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
6404def : Pat<(v2i32 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
6405def : Pat<(v4f16 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
6406def : Pat<(v2f32 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
6407
6408def : Pat<(i64 (bitconvert (v8i8  V64:$Vn))),
6409          (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
6410def : Pat<(i64 (bitconvert (v4i16 V64:$Vn))),
6411          (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
6412def : Pat<(i64 (bitconvert (v2i32 V64:$Vn))),
6413          (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
6414def : Pat<(i64 (bitconvert (v4f16 V64:$Vn))),
6415          (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
6416def : Pat<(i64 (bitconvert (v2f32 V64:$Vn))),
6417          (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
6418def : Pat<(i64 (bitconvert (v1f64 V64:$Vn))),
6419          (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
6420}
6421let Predicates = [IsBE] in {
6422def : Pat<(v8i8  (bitconvert GPR64:$Xn)),
6423                 (REV64v8i8 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
6424def : Pat<(v4i16 (bitconvert GPR64:$Xn)),
6425                 (REV64v4i16 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
6426def : Pat<(v2i32 (bitconvert GPR64:$Xn)),
6427                 (REV64v2i32 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
6428def : Pat<(v4f16 (bitconvert GPR64:$Xn)),
6429                 (REV64v4i16 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
6430def : Pat<(v2f32 (bitconvert GPR64:$Xn)),
6431                 (REV64v2i32 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
6432
6433def : Pat<(i64 (bitconvert (v8i8  V64:$Vn))),
6434          (REV64v8i8 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
6435def : Pat<(i64 (bitconvert (v4i16 V64:$Vn))),
6436          (REV64v4i16 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
6437def : Pat<(i64 (bitconvert (v2i32 V64:$Vn))),
6438          (REV64v2i32 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
6439def : Pat<(i64 (bitconvert (v4f16 V64:$Vn))),
6440          (REV64v4i16 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
6441def : Pat<(i64 (bitconvert (v2f32 V64:$Vn))),
6442          (REV64v2i32 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
6443}
6444def : Pat<(v1i64 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
6445def : Pat<(v1f64 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
6446def : Pat<(i64 (bitconvert (v1i64 V64:$Vn))),
6447          (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
6448def : Pat<(v1i64 (scalar_to_vector GPR64:$Xn)),
6449          (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
6450def : Pat<(v1f64 (scalar_to_vector GPR64:$Xn)),
6451          (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
6452def : Pat<(v1f64 (scalar_to_vector (f64 FPR64:$Xn))), (v1f64 FPR64:$Xn)>;
6453
6454def : Pat<(f32 (bitconvert (i32 GPR32:$Xn))),
6455          (COPY_TO_REGCLASS GPR32:$Xn, FPR32)>;
6456def : Pat<(i32 (bitconvert (f32 FPR32:$Xn))),
6457          (COPY_TO_REGCLASS FPR32:$Xn, GPR32)>;
6458def : Pat<(f64 (bitconvert (i64 GPR64:$Xn))),
6459          (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
6460def : Pat<(i64 (bitconvert (f64 FPR64:$Xn))),
6461          (COPY_TO_REGCLASS FPR64:$Xn, GPR64)>;
6462def : Pat<(i64 (bitconvert (v1f64 V64:$Vn))),
6463          (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
6464
6465let Predicates = [IsLE] in {
6466def : Pat<(v1i64 (bitconvert (v2i32 FPR64:$src))), (v1i64 FPR64:$src)>;
6467def : Pat<(v1i64 (bitconvert (v4i16 FPR64:$src))), (v1i64 FPR64:$src)>;
6468def : Pat<(v1i64 (bitconvert (v8i8  FPR64:$src))), (v1i64 FPR64:$src)>;
6469def : Pat<(v1i64 (bitconvert (v4f16 FPR64:$src))), (v1i64 FPR64:$src)>;
6470def : Pat<(v1i64 (bitconvert (v2f32 FPR64:$src))), (v1i64 FPR64:$src)>;
6471}
6472let Predicates = [IsBE] in {
6473def : Pat<(v1i64 (bitconvert (v2i32 FPR64:$src))),
6474                             (v1i64 (REV64v2i32 FPR64:$src))>;
6475def : Pat<(v1i64 (bitconvert (v4i16 FPR64:$src))),
6476                             (v1i64 (REV64v4i16 FPR64:$src))>;
6477def : Pat<(v1i64 (bitconvert (v8i8  FPR64:$src))),
6478                             (v1i64 (REV64v8i8 FPR64:$src))>;
6479def : Pat<(v1i64 (bitconvert (v4f16 FPR64:$src))),
6480                             (v1i64 (REV64v4i16 FPR64:$src))>;
6481def : Pat<(v1i64 (bitconvert (v2f32 FPR64:$src))),
6482                             (v1i64 (REV64v2i32 FPR64:$src))>;
6483}
6484def : Pat<(v1i64 (bitconvert (v1f64 FPR64:$src))), (v1i64 FPR64:$src)>;
6485def : Pat<(v1i64 (bitconvert (f64   FPR64:$src))), (v1i64 FPR64:$src)>;
6486
6487let Predicates = [IsLE] in {
6488def : Pat<(v2i32 (bitconvert (v1i64 FPR64:$src))), (v2i32 FPR64:$src)>;
6489def : Pat<(v2i32 (bitconvert (v4i16 FPR64:$src))), (v2i32 FPR64:$src)>;
6490def : Pat<(v2i32 (bitconvert (v8i8  FPR64:$src))), (v2i32 FPR64:$src)>;
6491def : Pat<(v2i32 (bitconvert (f64   FPR64:$src))), (v2i32 FPR64:$src)>;
6492def : Pat<(v2i32 (bitconvert (v1f64 FPR64:$src))), (v2i32 FPR64:$src)>;
6493def : Pat<(v2i32 (bitconvert (v4f16 FPR64:$src))), (v2i32 FPR64:$src)>;
6494}
6495let Predicates = [IsBE] in {
6496def : Pat<(v2i32 (bitconvert (v1i64 FPR64:$src))),
6497                             (v2i32 (REV64v2i32 FPR64:$src))>;
6498def : Pat<(v2i32 (bitconvert (v4i16 FPR64:$src))),
6499                             (v2i32 (REV32v4i16 FPR64:$src))>;
6500def : Pat<(v2i32 (bitconvert (v8i8  FPR64:$src))),
6501                             (v2i32 (REV32v8i8 FPR64:$src))>;
6502def : Pat<(v2i32 (bitconvert (f64   FPR64:$src))),
6503                             (v2i32 (REV64v2i32 FPR64:$src))>;
6504def : Pat<(v2i32 (bitconvert (v1f64 FPR64:$src))),
6505                             (v2i32 (REV64v2i32 FPR64:$src))>;
6506def : Pat<(v2i32 (bitconvert (v4f16 FPR64:$src))),
6507                             (v2i32 (REV32v4i16 FPR64:$src))>;
6508}
6509def : Pat<(v2i32 (bitconvert (v2f32 FPR64:$src))), (v2i32 FPR64:$src)>;
6510
6511let Predicates = [IsLE] in {
6512def : Pat<(v4i16 (bitconvert (v1i64 FPR64:$src))), (v4i16 FPR64:$src)>;
6513def : Pat<(v4i16 (bitconvert (v2i32 FPR64:$src))), (v4i16 FPR64:$src)>;
6514def : Pat<(v4i16 (bitconvert (v8i8  FPR64:$src))), (v4i16 FPR64:$src)>;
6515def : Pat<(v4i16 (bitconvert (f64   FPR64:$src))), (v4i16 FPR64:$src)>;
6516def : Pat<(v4i16 (bitconvert (v2f32 FPR64:$src))), (v4i16 FPR64:$src)>;
6517def : Pat<(v4i16 (bitconvert (v1f64 FPR64:$src))), (v4i16 FPR64:$src)>;
6518}
6519let Predicates = [IsBE] in {
6520def : Pat<(v4i16 (bitconvert (v1i64 FPR64:$src))),
6521                             (v4i16 (REV64v4i16 FPR64:$src))>;
6522def : Pat<(v4i16 (bitconvert (v2i32 FPR64:$src))),
6523                             (v4i16 (REV32v4i16 FPR64:$src))>;
6524def : Pat<(v4i16 (bitconvert (v8i8  FPR64:$src))),
6525                             (v4i16 (REV16v8i8 FPR64:$src))>;
6526def : Pat<(v4i16 (bitconvert (f64   FPR64:$src))),
6527                             (v4i16 (REV64v4i16 FPR64:$src))>;
6528def : Pat<(v4i16 (bitconvert (v2f32 FPR64:$src))),
6529                             (v4i16 (REV32v4i16 FPR64:$src))>;
6530def : Pat<(v4i16 (bitconvert (v1f64 FPR64:$src))),
6531                             (v4i16 (REV64v4i16 FPR64:$src))>;
6532}
6533def : Pat<(v4i16 (bitconvert (v4f16 FPR64:$src))), (v4i16 FPR64:$src)>;
6534
6535let Predicates = [IsLE] in {
6536def : Pat<(v4f16 (bitconvert (v1i64 FPR64:$src))), (v4f16 FPR64:$src)>;
6537def : Pat<(v4f16 (bitconvert (v2i32 FPR64:$src))), (v4f16 FPR64:$src)>;
6538def : Pat<(v4f16 (bitconvert (v8i8  FPR64:$src))), (v4f16 FPR64:$src)>;
6539def : Pat<(v4f16 (bitconvert (f64   FPR64:$src))), (v4f16 FPR64:$src)>;
6540def : Pat<(v4f16 (bitconvert (v2f32 FPR64:$src))), (v4f16 FPR64:$src)>;
6541def : Pat<(v4f16 (bitconvert (v1f64 FPR64:$src))), (v4f16 FPR64:$src)>;
6542}
6543let Predicates = [IsBE] in {
6544def : Pat<(v4f16 (bitconvert (v1i64 FPR64:$src))),
6545                             (v4f16 (REV64v4i16 FPR64:$src))>;
6546def : Pat<(v4f16 (bitconvert (v2i32 FPR64:$src))),
6547                             (v4f16 (REV32v4i16 FPR64:$src))>;
6548def : Pat<(v4f16 (bitconvert (v8i8  FPR64:$src))),
6549                             (v4f16 (REV16v8i8 FPR64:$src))>;
6550def : Pat<(v4f16 (bitconvert (f64   FPR64:$src))),
6551                             (v4f16 (REV64v4i16 FPR64:$src))>;
6552def : Pat<(v4f16 (bitconvert (v2f32 FPR64:$src))),
6553                             (v4f16 (REV32v4i16 FPR64:$src))>;
6554def : Pat<(v4f16 (bitconvert (v1f64 FPR64:$src))),
6555                             (v4f16 (REV64v4i16 FPR64:$src))>;
6556}
6557def : Pat<(v4f16 (bitconvert (v4i16 FPR64:$src))), (v4f16 FPR64:$src)>;
6558
6559let Predicates = [IsLE] in {
6560def : Pat<(v8i8  (bitconvert (v1i64 FPR64:$src))), (v8i8  FPR64:$src)>;
6561def : Pat<(v8i8  (bitconvert (v2i32 FPR64:$src))), (v8i8  FPR64:$src)>;
6562def : Pat<(v8i8  (bitconvert (v4i16 FPR64:$src))), (v8i8  FPR64:$src)>;
6563def : Pat<(v8i8  (bitconvert (f64   FPR64:$src))), (v8i8  FPR64:$src)>;
6564def : Pat<(v8i8  (bitconvert (v2f32 FPR64:$src))), (v8i8  FPR64:$src)>;
6565def : Pat<(v8i8  (bitconvert (v1f64 FPR64:$src))), (v8i8  FPR64:$src)>;
6566def : Pat<(v8i8  (bitconvert (v4f16 FPR64:$src))), (v8i8  FPR64:$src)>;
6567}
6568let Predicates = [IsBE] in {
6569def : Pat<(v8i8  (bitconvert (v1i64 FPR64:$src))),
6570                             (v8i8 (REV64v8i8 FPR64:$src))>;
6571def : Pat<(v8i8  (bitconvert (v2i32 FPR64:$src))),
6572                             (v8i8 (REV32v8i8 FPR64:$src))>;
6573def : Pat<(v8i8  (bitconvert (v4i16 FPR64:$src))),
6574                             (v8i8 (REV16v8i8 FPR64:$src))>;
6575def : Pat<(v8i8  (bitconvert (f64   FPR64:$src))),
6576                             (v8i8 (REV64v8i8 FPR64:$src))>;
6577def : Pat<(v8i8  (bitconvert (v2f32 FPR64:$src))),
6578                             (v8i8 (REV32v8i8 FPR64:$src))>;
6579def : Pat<(v8i8  (bitconvert (v1f64 FPR64:$src))),
6580                             (v8i8 (REV64v8i8 FPR64:$src))>;
6581def : Pat<(v8i8  (bitconvert (v4f16 FPR64:$src))),
6582                             (v8i8 (REV16v8i8 FPR64:$src))>;
6583}
6584
6585let Predicates = [IsLE] in {
6586def : Pat<(f64   (bitconvert (v2i32 FPR64:$src))), (f64   FPR64:$src)>;
6587def : Pat<(f64   (bitconvert (v4i16 FPR64:$src))), (f64   FPR64:$src)>;
6588def : Pat<(f64   (bitconvert (v2f32 FPR64:$src))), (f64   FPR64:$src)>;
6589def : Pat<(f64   (bitconvert (v8i8  FPR64:$src))), (f64   FPR64:$src)>;
6590def : Pat<(f64   (bitconvert (v4f16 FPR64:$src))), (f64   FPR64:$src)>;
6591}
6592let Predicates = [IsBE] in {
6593def : Pat<(f64   (bitconvert (v2i32 FPR64:$src))),
6594                             (f64 (REV64v2i32 FPR64:$src))>;
6595def : Pat<(f64   (bitconvert (v4i16 FPR64:$src))),
6596                             (f64 (REV64v4i16 FPR64:$src))>;
6597def : Pat<(f64   (bitconvert (v2f32 FPR64:$src))),
6598                             (f64 (REV64v2i32 FPR64:$src))>;
6599def : Pat<(f64   (bitconvert (v8i8  FPR64:$src))),
6600                             (f64 (REV64v8i8 FPR64:$src))>;
6601def : Pat<(f64   (bitconvert (v4f16 FPR64:$src))),
6602                             (f64 (REV64v4i16 FPR64:$src))>;
6603}
6604def : Pat<(f64   (bitconvert (v1i64 FPR64:$src))), (f64   FPR64:$src)>;
6605def : Pat<(f64   (bitconvert (v1f64 FPR64:$src))), (f64   FPR64:$src)>;
6606
6607let Predicates = [IsLE] in {
6608def : Pat<(v1f64 (bitconvert (v2i32 FPR64:$src))), (v1f64 FPR64:$src)>;
6609def : Pat<(v1f64 (bitconvert (v4i16 FPR64:$src))), (v1f64 FPR64:$src)>;
6610def : Pat<(v1f64 (bitconvert (v8i8  FPR64:$src))), (v1f64 FPR64:$src)>;
6611def : Pat<(v1f64 (bitconvert (v2f32 FPR64:$src))), (v1f64 FPR64:$src)>;
6612def : Pat<(v1f64 (bitconvert (v4f16 FPR64:$src))), (v1f64 FPR64:$src)>;
6613}
6614let Predicates = [IsBE] in {
6615def : Pat<(v1f64 (bitconvert (v2i32 FPR64:$src))),
6616                             (v1f64 (REV64v2i32 FPR64:$src))>;
6617def : Pat<(v1f64 (bitconvert (v4i16 FPR64:$src))),
6618                             (v1f64 (REV64v4i16 FPR64:$src))>;
6619def : Pat<(v1f64 (bitconvert (v8i8  FPR64:$src))),
6620                             (v1f64 (REV64v8i8 FPR64:$src))>;
6621def : Pat<(v1f64 (bitconvert (v2f32 FPR64:$src))),
6622                             (v1f64 (REV64v2i32 FPR64:$src))>;
6623def : Pat<(v1f64 (bitconvert (v4f16 FPR64:$src))),
6624                             (v1f64 (REV64v4i16 FPR64:$src))>;
6625}
6626def : Pat<(v1f64 (bitconvert (v1i64 FPR64:$src))), (v1f64 FPR64:$src)>;
6627def : Pat<(v1f64 (bitconvert (f64   FPR64:$src))), (v1f64 FPR64:$src)>;
6628
6629let Predicates = [IsLE] in {
6630def : Pat<(v2f32 (bitconvert (v1i64 FPR64:$src))), (v2f32 FPR64:$src)>;
6631def : Pat<(v2f32 (bitconvert (v4i16 FPR64:$src))), (v2f32 FPR64:$src)>;
6632def : Pat<(v2f32 (bitconvert (v8i8  FPR64:$src))), (v2f32 FPR64:$src)>;
6633def : Pat<(v2f32 (bitconvert (v1f64 FPR64:$src))), (v2f32 FPR64:$src)>;
6634def : Pat<(v2f32 (bitconvert (f64   FPR64:$src))), (v2f32 FPR64:$src)>;
6635def : Pat<(v2f32 (bitconvert (v4f16 FPR64:$src))), (v2f32 FPR64:$src)>;
6636}
6637let Predicates = [IsBE] in {
6638def : Pat<(v2f32 (bitconvert (v1i64 FPR64:$src))),
6639                             (v2f32 (REV64v2i32 FPR64:$src))>;
6640def : Pat<(v2f32 (bitconvert (v4i16 FPR64:$src))),
6641                             (v2f32 (REV32v4i16 FPR64:$src))>;
6642def : Pat<(v2f32 (bitconvert (v8i8  FPR64:$src))),
6643                             (v2f32 (REV32v8i8 FPR64:$src))>;
6644def : Pat<(v2f32 (bitconvert (v1f64 FPR64:$src))),
6645                             (v2f32 (REV64v2i32 FPR64:$src))>;
6646def : Pat<(v2f32 (bitconvert (f64   FPR64:$src))),
6647                             (v2f32 (REV64v2i32 FPR64:$src))>;
6648def : Pat<(v2f32 (bitconvert (v4f16 FPR64:$src))),
6649                             (v2f32 (REV32v4i16 FPR64:$src))>;
6650}
6651def : Pat<(v2f32 (bitconvert (v2i32 FPR64:$src))), (v2f32 FPR64:$src)>;
6652
6653let Predicates = [IsLE] in {
6654def : Pat<(f128 (bitconvert (v2i64 FPR128:$src))), (f128 FPR128:$src)>;
6655def : Pat<(f128 (bitconvert (v4i32 FPR128:$src))), (f128 FPR128:$src)>;
6656def : Pat<(f128 (bitconvert (v8i16 FPR128:$src))), (f128 FPR128:$src)>;
6657def : Pat<(f128 (bitconvert (v2f64 FPR128:$src))), (f128 FPR128:$src)>;
6658def : Pat<(f128 (bitconvert (v4f32 FPR128:$src))), (f128 FPR128:$src)>;
6659def : Pat<(f128 (bitconvert (v8f16 FPR128:$src))), (f128 FPR128:$src)>;
6660def : Pat<(f128 (bitconvert (v16i8 FPR128:$src))), (f128 FPR128:$src)>;
6661}
6662let Predicates = [IsBE] in {
6663def : Pat<(f128 (bitconvert (v2i64 FPR128:$src))),
6664                            (f128 (EXTv16i8 FPR128:$src, FPR128:$src, (i32 8)))>;
6665def : Pat<(f128 (bitconvert (v4i32 FPR128:$src))),
6666                            (f128 (EXTv16i8 (REV64v4i32 FPR128:$src),
6667                                            (REV64v4i32 FPR128:$src), (i32 8)))>;
6668def : Pat<(f128 (bitconvert (v8i16 FPR128:$src))),
6669                            (f128 (EXTv16i8 (REV64v8i16 FPR128:$src),
6670                                            (REV64v8i16 FPR128:$src), (i32 8)))>;
6671def : Pat<(f128 (bitconvert (v8f16 FPR128:$src))),
6672                            (f128 (EXTv16i8 (REV64v8i16 FPR128:$src),
6673                                            (REV64v8i16 FPR128:$src), (i32 8)))>;
6674def : Pat<(f128 (bitconvert (v2f64 FPR128:$src))),
6675                            (f128 (EXTv16i8 FPR128:$src, FPR128:$src, (i32 8)))>;
6676def : Pat<(f128 (bitconvert (v4f32 FPR128:$src))),
6677                            (f128 (EXTv16i8 (REV64v4i32 FPR128:$src),
6678                                            (REV64v4i32 FPR128:$src), (i32 8)))>;
6679def : Pat<(f128 (bitconvert (v16i8 FPR128:$src))),
6680                            (f128 (EXTv16i8 (REV64v16i8 FPR128:$src),
6681                                            (REV64v16i8 FPR128:$src), (i32 8)))>;
6682}
6683
6684let Predicates = [IsLE] in {
6685def : Pat<(v2f64 (bitconvert (f128  FPR128:$src))), (v2f64 FPR128:$src)>;
6686def : Pat<(v2f64 (bitconvert (v4i32 FPR128:$src))), (v2f64 FPR128:$src)>;
6687def : Pat<(v2f64 (bitconvert (v8i16 FPR128:$src))), (v2f64 FPR128:$src)>;
6688def : Pat<(v2f64 (bitconvert (v8f16 FPR128:$src))), (v2f64 FPR128:$src)>;
6689def : Pat<(v2f64 (bitconvert (v16i8 FPR128:$src))), (v2f64 FPR128:$src)>;
6690def : Pat<(v2f64 (bitconvert (v4f32 FPR128:$src))), (v2f64 FPR128:$src)>;
6691}
6692let Predicates = [IsBE] in {
6693def : Pat<(v2f64 (bitconvert (f128  FPR128:$src))),
6694                             (v2f64 (EXTv16i8 FPR128:$src,
6695                                              FPR128:$src, (i32 8)))>;
6696def : Pat<(v2f64 (bitconvert (v4i32 FPR128:$src))),
6697                             (v2f64 (REV64v4i32 FPR128:$src))>;
6698def : Pat<(v2f64 (bitconvert (v8i16 FPR128:$src))),
6699                             (v2f64 (REV64v8i16 FPR128:$src))>;
6700def : Pat<(v2f64 (bitconvert (v8f16 FPR128:$src))),
6701                             (v2f64 (REV64v8i16 FPR128:$src))>;
6702def : Pat<(v2f64 (bitconvert (v16i8 FPR128:$src))),
6703                             (v2f64 (REV64v16i8 FPR128:$src))>;
6704def : Pat<(v2f64 (bitconvert (v4f32 FPR128:$src))),
6705                             (v2f64 (REV64v4i32 FPR128:$src))>;
6706}
6707def : Pat<(v2f64 (bitconvert (v2i64 FPR128:$src))), (v2f64 FPR128:$src)>;
6708
6709let Predicates = [IsLE] in {
6710def : Pat<(v4f32 (bitconvert (f128  FPR128:$src))), (v4f32 FPR128:$src)>;
6711def : Pat<(v4f32 (bitconvert (v8i16 FPR128:$src))), (v4f32 FPR128:$src)>;
6712def : Pat<(v4f32 (bitconvert (v8f16 FPR128:$src))), (v4f32 FPR128:$src)>;
6713def : Pat<(v4f32 (bitconvert (v16i8 FPR128:$src))), (v4f32 FPR128:$src)>;
6714def : Pat<(v4f32 (bitconvert (v2i64 FPR128:$src))), (v4f32 FPR128:$src)>;
6715def : Pat<(v4f32 (bitconvert (v2f64 FPR128:$src))), (v4f32 FPR128:$src)>;
6716}
6717let Predicates = [IsBE] in {
6718def : Pat<(v4f32 (bitconvert (f128  FPR128:$src))),
6719                             (v4f32 (EXTv16i8 (REV64v4i32 FPR128:$src),
6720                                    (REV64v4i32 FPR128:$src), (i32 8)))>;
6721def : Pat<(v4f32 (bitconvert (v8i16 FPR128:$src))),
6722                             (v4f32 (REV32v8i16 FPR128:$src))>;
6723def : Pat<(v4f32 (bitconvert (v8f16 FPR128:$src))),
6724                             (v4f32 (REV32v8i16 FPR128:$src))>;
6725def : Pat<(v4f32 (bitconvert (v16i8 FPR128:$src))),
6726                             (v4f32 (REV32v16i8 FPR128:$src))>;
6727def : Pat<(v4f32 (bitconvert (v2i64 FPR128:$src))),
6728                             (v4f32 (REV64v4i32 FPR128:$src))>;
6729def : Pat<(v4f32 (bitconvert (v2f64 FPR128:$src))),
6730                             (v4f32 (REV64v4i32 FPR128:$src))>;
6731}
6732def : Pat<(v4f32 (bitconvert (v4i32 FPR128:$src))), (v4f32 FPR128:$src)>;
6733
6734let Predicates = [IsLE] in {
6735def : Pat<(v2i64 (bitconvert (f128  FPR128:$src))), (v2i64 FPR128:$src)>;
6736def : Pat<(v2i64 (bitconvert (v4i32 FPR128:$src))), (v2i64 FPR128:$src)>;
6737def : Pat<(v2i64 (bitconvert (v8i16 FPR128:$src))), (v2i64 FPR128:$src)>;
6738def : Pat<(v2i64 (bitconvert (v16i8 FPR128:$src))), (v2i64 FPR128:$src)>;
6739def : Pat<(v2i64 (bitconvert (v4f32 FPR128:$src))), (v2i64 FPR128:$src)>;
6740def : Pat<(v2i64 (bitconvert (v8f16 FPR128:$src))), (v2i64 FPR128:$src)>;
6741}
6742let Predicates = [IsBE] in {
6743def : Pat<(v2i64 (bitconvert (f128  FPR128:$src))),
6744                             (v2i64 (EXTv16i8 FPR128:$src,
6745                                              FPR128:$src, (i32 8)))>;
6746def : Pat<(v2i64 (bitconvert (v4i32 FPR128:$src))),
6747                             (v2i64 (REV64v4i32 FPR128:$src))>;
6748def : Pat<(v2i64 (bitconvert (v8i16 FPR128:$src))),
6749                             (v2i64 (REV64v8i16 FPR128:$src))>;
6750def : Pat<(v2i64 (bitconvert (v16i8 FPR128:$src))),
6751                             (v2i64 (REV64v16i8 FPR128:$src))>;
6752def : Pat<(v2i64 (bitconvert (v4f32 FPR128:$src))),
6753                             (v2i64 (REV64v4i32 FPR128:$src))>;
6754def : Pat<(v2i64 (bitconvert (v8f16 FPR128:$src))),
6755                             (v2i64 (REV64v8i16 FPR128:$src))>;
6756}
6757def : Pat<(v2i64 (bitconvert (v2f64 FPR128:$src))), (v2i64 FPR128:$src)>;
6758
6759let Predicates = [IsLE] in {
6760def : Pat<(v4i32 (bitconvert (f128  FPR128:$src))), (v4i32 FPR128:$src)>;
6761def : Pat<(v4i32 (bitconvert (v2i64 FPR128:$src))), (v4i32 FPR128:$src)>;
6762def : Pat<(v4i32 (bitconvert (v8i16 FPR128:$src))), (v4i32 FPR128:$src)>;
6763def : Pat<(v4i32 (bitconvert (v16i8 FPR128:$src))), (v4i32 FPR128:$src)>;
6764def : Pat<(v4i32 (bitconvert (v2f64 FPR128:$src))), (v4i32 FPR128:$src)>;
6765def : Pat<(v4i32 (bitconvert (v8f16 FPR128:$src))), (v4i32 FPR128:$src)>;
6766}
6767let Predicates = [IsBE] in {
6768def : Pat<(v4i32 (bitconvert (f128  FPR128:$src))),
6769                             (v4i32 (EXTv16i8 (REV64v4i32 FPR128:$src),
6770                                              (REV64v4i32 FPR128:$src),
6771                                              (i32 8)))>;
6772def : Pat<(v4i32 (bitconvert (v2i64 FPR128:$src))),
6773                             (v4i32 (REV64v4i32 FPR128:$src))>;
6774def : Pat<(v4i32 (bitconvert (v8i16 FPR128:$src))),
6775                             (v4i32 (REV32v8i16 FPR128:$src))>;
6776def : Pat<(v4i32 (bitconvert (v16i8 FPR128:$src))),
6777                             (v4i32 (REV32v16i8 FPR128:$src))>;
6778def : Pat<(v4i32 (bitconvert (v2f64 FPR128:$src))),
6779                             (v4i32 (REV64v4i32 FPR128:$src))>;
6780def : Pat<(v4i32 (bitconvert (v8f16 FPR128:$src))),
6781                             (v4i32 (REV32v8i16 FPR128:$src))>;
6782}
6783def : Pat<(v4i32 (bitconvert (v4f32 FPR128:$src))), (v4i32 FPR128:$src)>;
6784
6785let Predicates = [IsLE] in {
6786def : Pat<(v8i16 (bitconvert (f128  FPR128:$src))), (v8i16 FPR128:$src)>;
6787def : Pat<(v8i16 (bitconvert (v2i64 FPR128:$src))), (v8i16 FPR128:$src)>;
6788def : Pat<(v8i16 (bitconvert (v4i32 FPR128:$src))), (v8i16 FPR128:$src)>;
6789def : Pat<(v8i16 (bitconvert (v16i8 FPR128:$src))), (v8i16 FPR128:$src)>;
6790def : Pat<(v8i16 (bitconvert (v2f64 FPR128:$src))), (v8i16 FPR128:$src)>;
6791def : Pat<(v8i16 (bitconvert (v4f32 FPR128:$src))), (v8i16 FPR128:$src)>;
6792}
6793let Predicates = [IsBE] in {
6794def : Pat<(v8i16 (bitconvert (f128  FPR128:$src))),
6795                             (v8i16 (EXTv16i8 (REV64v8i16 FPR128:$src),
6796                                              (REV64v8i16 FPR128:$src),
6797                                              (i32 8)))>;
6798def : Pat<(v8i16 (bitconvert (v2i64 FPR128:$src))),
6799                             (v8i16 (REV64v8i16 FPR128:$src))>;
6800def : Pat<(v8i16 (bitconvert (v4i32 FPR128:$src))),
6801                             (v8i16 (REV32v8i16 FPR128:$src))>;
6802def : Pat<(v8i16 (bitconvert (v16i8 FPR128:$src))),
6803                             (v8i16 (REV16v16i8 FPR128:$src))>;
6804def : Pat<(v8i16 (bitconvert (v2f64 FPR128:$src))),
6805                             (v8i16 (REV64v8i16 FPR128:$src))>;
6806def : Pat<(v8i16 (bitconvert (v4f32 FPR128:$src))),
6807                             (v8i16 (REV32v8i16 FPR128:$src))>;
6808}
6809def : Pat<(v8i16 (bitconvert (v8f16 FPR128:$src))), (v8i16 FPR128:$src)>;
6810
6811let Predicates = [IsLE] in {
6812def : Pat<(v8f16 (bitconvert (f128  FPR128:$src))), (v8f16 FPR128:$src)>;
6813def : Pat<(v8f16 (bitconvert (v2i64 FPR128:$src))), (v8f16 FPR128:$src)>;
6814def : Pat<(v8f16 (bitconvert (v4i32 FPR128:$src))), (v8f16 FPR128:$src)>;
6815def : Pat<(v8f16 (bitconvert (v16i8 FPR128:$src))), (v8f16 FPR128:$src)>;
6816def : Pat<(v8f16 (bitconvert (v2f64 FPR128:$src))), (v8f16 FPR128:$src)>;
6817def : Pat<(v8f16 (bitconvert (v4f32 FPR128:$src))), (v8f16 FPR128:$src)>;
6818}
6819let Predicates = [IsBE] in {
6820def : Pat<(v8f16 (bitconvert (f128  FPR128:$src))),
6821                             (v8f16 (EXTv16i8 (REV64v8i16 FPR128:$src),
6822                                              (REV64v8i16 FPR128:$src),
6823                                              (i32 8)))>;
6824def : Pat<(v8f16 (bitconvert (v2i64 FPR128:$src))),
6825                             (v8f16 (REV64v8i16 FPR128:$src))>;
6826def : Pat<(v8f16 (bitconvert (v4i32 FPR128:$src))),
6827                             (v8f16 (REV32v8i16 FPR128:$src))>;
6828def : Pat<(v8f16 (bitconvert (v16i8 FPR128:$src))),
6829                             (v8f16 (REV16v16i8 FPR128:$src))>;
6830def : Pat<(v8f16 (bitconvert (v2f64 FPR128:$src))),
6831                             (v8f16 (REV64v8i16 FPR128:$src))>;
6832def : Pat<(v8f16 (bitconvert (v4f32 FPR128:$src))),
6833                             (v8f16 (REV32v8i16 FPR128:$src))>;
6834}
6835def : Pat<(v8f16 (bitconvert (v8i16 FPR128:$src))), (v8f16 FPR128:$src)>;
6836
6837let Predicates = [IsLE] in {
6838def : Pat<(v16i8 (bitconvert (f128  FPR128:$src))), (v16i8 FPR128:$src)>;
6839def : Pat<(v16i8 (bitconvert (v2i64 FPR128:$src))), (v16i8 FPR128:$src)>;
6840def : Pat<(v16i8 (bitconvert (v4i32 FPR128:$src))), (v16i8 FPR128:$src)>;
6841def : Pat<(v16i8 (bitconvert (v8i16 FPR128:$src))), (v16i8 FPR128:$src)>;
6842def : Pat<(v16i8 (bitconvert (v2f64 FPR128:$src))), (v16i8 FPR128:$src)>;
6843def : Pat<(v16i8 (bitconvert (v4f32 FPR128:$src))), (v16i8 FPR128:$src)>;
6844def : Pat<(v16i8 (bitconvert (v8f16 FPR128:$src))), (v16i8 FPR128:$src)>;
6845}
6846let Predicates = [IsBE] in {
6847def : Pat<(v16i8 (bitconvert (f128  FPR128:$src))),
6848                             (v16i8 (EXTv16i8 (REV64v16i8 FPR128:$src),
6849                                              (REV64v16i8 FPR128:$src),
6850                                              (i32 8)))>;
6851def : Pat<(v16i8 (bitconvert (v2i64 FPR128:$src))),
6852                             (v16i8 (REV64v16i8 FPR128:$src))>;
6853def : Pat<(v16i8 (bitconvert (v4i32 FPR128:$src))),
6854                             (v16i8 (REV32v16i8 FPR128:$src))>;
6855def : Pat<(v16i8 (bitconvert (v8i16 FPR128:$src))),
6856                             (v16i8 (REV16v16i8 FPR128:$src))>;
6857def : Pat<(v16i8 (bitconvert (v2f64 FPR128:$src))),
6858                             (v16i8 (REV64v16i8 FPR128:$src))>;
6859def : Pat<(v16i8 (bitconvert (v4f32 FPR128:$src))),
6860                             (v16i8 (REV32v16i8 FPR128:$src))>;
6861def : Pat<(v16i8 (bitconvert (v8f16 FPR128:$src))),
6862                             (v16i8 (REV16v16i8 FPR128:$src))>;
6863}
6864
6865def : Pat<(v4i16 (extract_subvector V128:$Rn, (i64 0))),
6866           (EXTRACT_SUBREG V128:$Rn, dsub)>;
6867def : Pat<(v8i8 (extract_subvector V128:$Rn, (i64 0))),
6868           (EXTRACT_SUBREG V128:$Rn, dsub)>;
6869def : Pat<(v2f32 (extract_subvector V128:$Rn, (i64 0))),
6870           (EXTRACT_SUBREG V128:$Rn, dsub)>;
6871def : Pat<(v4f16 (extract_subvector V128:$Rn, (i64 0))),
6872           (EXTRACT_SUBREG V128:$Rn, dsub)>;
6873def : Pat<(v2i32 (extract_subvector V128:$Rn, (i64 0))),
6874           (EXTRACT_SUBREG V128:$Rn, dsub)>;
6875def : Pat<(v1i64 (extract_subvector V128:$Rn, (i64 0))),
6876           (EXTRACT_SUBREG V128:$Rn, dsub)>;
6877def : Pat<(v1f64 (extract_subvector V128:$Rn, (i64 0))),
6878           (EXTRACT_SUBREG V128:$Rn, dsub)>;
6879
6880def : Pat<(v8i8 (extract_subvector (v16i8 FPR128:$Rn), (i64 1))),
6881          (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
6882def : Pat<(v4i16 (extract_subvector (v8i16 FPR128:$Rn), (i64 1))),
6883          (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
6884def : Pat<(v2i32 (extract_subvector (v4i32 FPR128:$Rn), (i64 1))),
6885          (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
6886def : Pat<(v1i64 (extract_subvector (v2i64 FPR128:$Rn), (i64 1))),
6887          (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
6888
6889// A 64-bit subvector insert to the first 128-bit vector position
6890// is a subregister copy that needs no instruction.
6891multiclass InsertSubvectorUndef<ValueType Ty> {
6892  def : Pat<(insert_subvector undef, (v1i64 FPR64:$src), (Ty 0)),
6893            (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
6894  def : Pat<(insert_subvector undef, (v1f64 FPR64:$src), (Ty 0)),
6895            (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
6896  def : Pat<(insert_subvector undef, (v2i32 FPR64:$src), (Ty 0)),
6897            (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
6898  def : Pat<(insert_subvector undef, (v2f32 FPR64:$src), (Ty 0)),
6899            (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
6900  def : Pat<(insert_subvector undef, (v4i16 FPR64:$src), (Ty 0)),
6901            (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
6902  def : Pat<(insert_subvector undef, (v4f16 FPR64:$src), (Ty 0)),
6903            (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
6904  def : Pat<(insert_subvector undef, (v8i8 FPR64:$src), (Ty 0)),
6905            (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
6906}
6907
6908defm : InsertSubvectorUndef<i32>;
6909defm : InsertSubvectorUndef<i64>;
6910
6911// Use pair-wise add instructions when summing up the lanes for v2f64, v2i64
6912// or v2f32.
6913def : Pat<(i64 (add (vector_extract (v2i64 FPR128:$Rn), (i64 0)),
6914                    (vector_extract (v2i64 FPR128:$Rn), (i64 1)))),
6915           (i64 (ADDPv2i64p (v2i64 FPR128:$Rn)))>;
6916def : Pat<(f64 (fadd (vector_extract (v2f64 FPR128:$Rn), (i64 0)),
6917                     (vector_extract (v2f64 FPR128:$Rn), (i64 1)))),
6918           (f64 (FADDPv2i64p (v2f64 FPR128:$Rn)))>;
6919    // vector_extract on 64-bit vectors gets promoted to a 128 bit vector,
6920    // so we match on v4f32 here, not v2f32. This will also catch adding
6921    // the low two lanes of a true v4f32 vector.
6922def : Pat<(fadd (vector_extract (v4f32 FPR128:$Rn), (i64 0)),
6923                (vector_extract (v4f32 FPR128:$Rn), (i64 1))),
6924          (f32 (FADDPv2i32p (EXTRACT_SUBREG FPR128:$Rn, dsub)))>;
6925
6926// Scalar 64-bit shifts in FPR64 registers.
6927def : Pat<(i64 (int_aarch64_neon_sshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
6928          (SSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
6929def : Pat<(i64 (int_aarch64_neon_ushl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
6930          (USHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
6931def : Pat<(i64 (int_aarch64_neon_srshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
6932          (SRSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
6933def : Pat<(i64 (int_aarch64_neon_urshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
6934          (URSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
6935
6936// Patterns for nontemporal/no-allocate stores.
6937// We have to resort to tricks to turn a single-input store into a store pair,
6938// because there is no single-input nontemporal store, only STNP.
6939let Predicates = [IsLE] in {
6940let AddedComplexity = 15 in {
6941class NTStore128Pat<ValueType VT> :
6942  Pat<(nontemporalstore (VT FPR128:$Rt),
6943        (am_indexed7s64 GPR64sp:$Rn, simm7s8:$offset)),
6944      (STNPDi (EXTRACT_SUBREG FPR128:$Rt, dsub),
6945              (CPYi64 FPR128:$Rt, (i64 1)),
6946              GPR64sp:$Rn, simm7s8:$offset)>;
6947
6948def : NTStore128Pat<v2i64>;
6949def : NTStore128Pat<v4i32>;
6950def : NTStore128Pat<v8i16>;
6951def : NTStore128Pat<v16i8>;
6952
6953class NTStore64Pat<ValueType VT> :
6954  Pat<(nontemporalstore (VT FPR64:$Rt),
6955        (am_indexed7s32 GPR64sp:$Rn, simm7s4:$offset)),
6956      (STNPSi (EXTRACT_SUBREG FPR64:$Rt, ssub),
6957              (CPYi32 (SUBREG_TO_REG (i64 0), FPR64:$Rt, dsub), (i64 1)),
6958              GPR64sp:$Rn, simm7s4:$offset)>;
6959
6960// FIXME: Shouldn't v1f64 loads/stores be promoted to v1i64?
6961def : NTStore64Pat<v1f64>;
6962def : NTStore64Pat<v1i64>;
6963def : NTStore64Pat<v2i32>;
6964def : NTStore64Pat<v4i16>;
6965def : NTStore64Pat<v8i8>;
6966
6967def : Pat<(nontemporalstore GPR64:$Rt,
6968            (am_indexed7s32 GPR64sp:$Rn, simm7s4:$offset)),
6969          (STNPWi (EXTRACT_SUBREG GPR64:$Rt, sub_32),
6970                  (EXTRACT_SUBREG (UBFMXri GPR64:$Rt, 32, 63), sub_32),
6971                  GPR64sp:$Rn, simm7s4:$offset)>;
6972} // AddedComplexity=10
6973} // Predicates = [IsLE]
6974
6975// Tail call return handling. These are all compiler pseudo-instructions,
6976// so no encoding information or anything like that.
6977let isCall = 1, isTerminator = 1, isReturn = 1, isBarrier = 1, Uses = [SP] in {
6978  def TCRETURNdi : Pseudo<(outs), (ins i64imm:$dst, i32imm:$FPDiff), []>,
6979                   Sched<[WriteBrReg]>;
6980  def TCRETURNri : Pseudo<(outs), (ins tcGPR64:$dst, i32imm:$FPDiff), []>,
6981                   Sched<[WriteBrReg]>;
6982  // Indirect tail-call with any register allowed, used by MachineOutliner when
6983  // this is proven safe.
6984  // FIXME: If we have to add any more hacks like this, we should instead relax
6985  // some verifier checks for outlined functions.
6986  def TCRETURNriALL : Pseudo<(outs), (ins GPR64:$dst, i32imm:$FPDiff), []>,
6987                      Sched<[WriteBrReg]>;
6988  // Indirect tail-call limited to only use registers (x16 and x17) which are
6989  // allowed to tail-call a "BTI c" instruction.
6990  def TCRETURNriBTI : Pseudo<(outs), (ins rtcGPR64:$dst, i32imm:$FPDiff), []>,
6991                      Sched<[WriteBrReg]>;
6992}
6993
6994def : Pat<(AArch64tcret tcGPR64:$dst, (i32 timm:$FPDiff)),
6995          (TCRETURNri tcGPR64:$dst, imm:$FPDiff)>,
6996      Requires<[NotUseBTI]>;
6997def : Pat<(AArch64tcret rtcGPR64:$dst, (i32 timm:$FPDiff)),
6998          (TCRETURNriBTI rtcGPR64:$dst, imm:$FPDiff)>,
6999      Requires<[UseBTI]>;
7000def : Pat<(AArch64tcret tglobaladdr:$dst, (i32 timm:$FPDiff)),
7001          (TCRETURNdi texternalsym:$dst, imm:$FPDiff)>;
7002def : Pat<(AArch64tcret texternalsym:$dst, (i32 timm:$FPDiff)),
7003          (TCRETURNdi texternalsym:$dst, imm:$FPDiff)>;
7004
7005def MOVMCSym : Pseudo<(outs GPR64:$dst), (ins i64imm:$sym), []>, Sched<[]>;
7006def : Pat<(i64 (AArch64LocalRecover mcsym:$sym)), (MOVMCSym mcsym:$sym)>;
7007
7008// Extracting lane zero is a special case where we can just use a plain
7009// EXTRACT_SUBREG instruction, which will become FMOV. This is easier for the
7010// rest of the compiler, especially the register allocator and copy propagation,
7011// to reason about, so is preferred when it's possible to use it.
7012let AddedComplexity = 10 in {
7013  def : Pat<(i64 (extractelt (v2i64 V128:$V), (i64 0))), (EXTRACT_SUBREG V128:$V, dsub)>;
7014  def : Pat<(i32 (extractelt (v4i32 V128:$V), (i64 0))), (EXTRACT_SUBREG V128:$V, ssub)>;
7015  def : Pat<(i32 (extractelt (v2i32 V64:$V), (i64 0))), (EXTRACT_SUBREG V64:$V, ssub)>;
7016}
7017
7018// dot_v4i8
7019class mul_v4i8<SDPatternOperator ldop> :
7020  PatFrag<(ops node:$Rn, node:$Rm, node:$offset),
7021          (mul (ldop (add node:$Rn, node:$offset)),
7022               (ldop (add node:$Rm, node:$offset)))>;
7023class mulz_v4i8<SDPatternOperator ldop> :
7024  PatFrag<(ops node:$Rn, node:$Rm),
7025          (mul (ldop node:$Rn), (ldop node:$Rm))>;
7026
7027def load_v4i8 :
7028  OutPatFrag<(ops node:$R),
7029             (INSERT_SUBREG
7030              (v2i32 (IMPLICIT_DEF)),
7031               (i32 (COPY_TO_REGCLASS (LDRWui node:$R, (i64 0)), FPR32)),
7032              ssub)>;
7033
7034class dot_v4i8<Instruction DOT, SDPatternOperator ldop> :
7035  Pat<(i32 (add (mul_v4i8<ldop> GPR64sp:$Rn, GPR64sp:$Rm, (i64 3)),
7036           (add (mul_v4i8<ldop> GPR64sp:$Rn, GPR64sp:$Rm, (i64 2)),
7037           (add (mul_v4i8<ldop> GPR64sp:$Rn, GPR64sp:$Rm, (i64 1)),
7038                (mulz_v4i8<ldop> GPR64sp:$Rn, GPR64sp:$Rm))))),
7039      (EXTRACT_SUBREG (i64 (DOT (DUPv2i32gpr WZR),
7040                                (load_v4i8 GPR64sp:$Rn),
7041                                (load_v4i8 GPR64sp:$Rm))),
7042                      sub_32)>, Requires<[HasDotProd]>;
7043
7044// dot_v8i8
7045class ee_v8i8<SDPatternOperator extend> :
7046  PatFrag<(ops node:$V, node:$K),
7047          (v4i16 (extract_subvector (v8i16 (extend node:$V)), node:$K))>;
7048
7049class mul_v8i8<SDPatternOperator mulop, SDPatternOperator extend> :
7050  PatFrag<(ops node:$M, node:$N, node:$K),
7051          (mulop (v4i16 (ee_v8i8<extend> node:$M, node:$K)),
7052                 (v4i16 (ee_v8i8<extend> node:$N, node:$K)))>;
7053
7054class idot_v8i8<SDPatternOperator mulop, SDPatternOperator extend> :
7055  PatFrag<(ops node:$M, node:$N),
7056          (i32 (extractelt
7057           (v4i32 (AArch64uaddv
7058            (add (mul_v8i8<mulop, extend> node:$M, node:$N, (i64 0)),
7059                 (mul_v8i8<mulop, extend> node:$M, node:$N, (i64 4))))),
7060           (i64 0)))>;
7061
7062// vaddv_[su]32 is special; -> ADDP Vd.2S,Vn.2S,Vm.2S; return Vd.s[0];Vn==Vm
7063def VADDV_32 : OutPatFrag<(ops node:$R), (ADDPv2i32 node:$R, node:$R)>;
7064
7065class odot_v8i8<Instruction DOT> :
7066  OutPatFrag<(ops node:$Vm, node:$Vn),
7067             (EXTRACT_SUBREG
7068              (VADDV_32
7069               (i64 (DOT (DUPv2i32gpr WZR),
7070                         (v8i8 node:$Vm),
7071                         (v8i8 node:$Vn)))),
7072              sub_32)>;
7073
7074class dot_v8i8<Instruction DOT, SDPatternOperator mulop,
7075                    SDPatternOperator extend> :
7076  Pat<(idot_v8i8<mulop, extend> V64:$Vm, V64:$Vn),
7077      (odot_v8i8<DOT> V64:$Vm, V64:$Vn)>,
7078  Requires<[HasDotProd]>;
7079
7080// dot_v16i8
7081class ee_v16i8<SDPatternOperator extend> :
7082  PatFrag<(ops node:$V, node:$K1, node:$K2),
7083          (v4i16 (extract_subvector
7084           (v8i16 (extend
7085            (v8i8 (extract_subvector node:$V, node:$K1)))), node:$K2))>;
7086
7087class mul_v16i8<SDPatternOperator mulop, SDPatternOperator extend> :
7088  PatFrag<(ops node:$M, node:$N, node:$K1, node:$K2),
7089          (v4i32
7090           (mulop (v4i16 (ee_v16i8<extend> node:$M, node:$K1, node:$K2)),
7091                  (v4i16 (ee_v16i8<extend> node:$N, node:$K1, node:$K2))))>;
7092
7093class idot_v16i8<SDPatternOperator m, SDPatternOperator x> :
7094  PatFrag<(ops node:$M, node:$N),
7095          (i32 (extractelt
7096           (v4i32 (AArch64uaddv
7097            (add
7098             (add (mul_v16i8<m, x> node:$M, node:$N, (i64 0), (i64 0)),
7099                  (mul_v16i8<m, x> node:$M, node:$N, (i64 8), (i64 0))),
7100             (add (mul_v16i8<m, x> node:$M, node:$N, (i64 0), (i64 4)),
7101                  (mul_v16i8<m, x> node:$M, node:$N, (i64 8), (i64 4)))))),
7102           (i64 0)))>;
7103
7104class odot_v16i8<Instruction DOT> :
7105  OutPatFrag<(ops node:$Vm, node:$Vn),
7106             (i32 (ADDVv4i32v
7107              (DOT (DUPv4i32gpr WZR), node:$Vm, node:$Vn)))>;
7108
7109class dot_v16i8<Instruction DOT, SDPatternOperator mulop,
7110                SDPatternOperator extend> :
7111  Pat<(idot_v16i8<mulop, extend> V128:$Vm, V128:$Vn),
7112      (odot_v16i8<DOT> V128:$Vm, V128:$Vn)>,
7113  Requires<[HasDotProd]>;
7114
7115let AddedComplexity = 10 in {
7116  def : dot_v4i8<SDOTv8i8, sextloadi8>;
7117  def : dot_v4i8<UDOTv8i8, zextloadi8>;
7118  def : dot_v8i8<SDOTv8i8, AArch64smull, sext>;
7119  def : dot_v8i8<UDOTv8i8, AArch64umull, zext>;
7120  def : dot_v16i8<SDOTv16i8, AArch64smull, sext>;
7121  def : dot_v16i8<UDOTv16i8, AArch64umull, zext>;
7122
7123  // FIXME: add patterns to generate vector by element dot product.
7124  // FIXME: add SVE dot-product patterns.
7125}
7126
7127include "AArch64InstrAtomics.td"
7128include "AArch64SVEInstrInfo.td"
7129