xref: /freebsd/contrib/llvm-project/llvm/lib/Target/AArch64/AArch64InstrInfo.td (revision 7ef62cebc2f965b0f640263e179276928885e33d)
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_0a         : Predicate<"Subtarget->hasV8_0aOps()">,
17                                 AssemblerPredicate<(all_of HasV8_0aOps), "armv8.0a">;
18def HasV8_1a         : Predicate<"Subtarget->hasV8_1aOps()">,
19                                 AssemblerPredicateWithAll<(all_of HasV8_1aOps), "armv8.1a">;
20def HasV8_2a         : Predicate<"Subtarget->hasV8_2aOps()">,
21                                 AssemblerPredicateWithAll<(all_of HasV8_2aOps), "armv8.2a">;
22def HasV8_3a         : Predicate<"Subtarget->hasV8_3aOps()">,
23                                 AssemblerPredicateWithAll<(all_of HasV8_3aOps), "armv8.3a">;
24def HasV8_4a         : Predicate<"Subtarget->hasV8_4aOps()">,
25                                 AssemblerPredicateWithAll<(all_of HasV8_4aOps), "armv8.4a">;
26def HasV8_5a         : Predicate<"Subtarget->hasV8_5aOps()">,
27                                 AssemblerPredicateWithAll<(all_of HasV8_5aOps), "armv8.5a">;
28def HasV8_6a         : Predicate<"Subtarget->hasV8_6aOps()">,
29                                 AssemblerPredicateWithAll<(all_of HasV8_6aOps), "armv8.6a">;
30def HasV8_7a         : Predicate<"Subtarget->hasV8_7aOps()">,
31                                 AssemblerPredicateWithAll<(all_of HasV8_7aOps), "armv8.7a">;
32def HasV8_8a         : Predicate<"Subtarget->hasV8_8aOps()">,
33                                 AssemblerPredicateWithAll<(all_of HasV8_8aOps), "armv8.8a">;
34def HasV8_9a         : Predicate<"Subtarget->hasV8_9aOps()">,
35                                 AssemblerPredicateWithAll<(all_of HasV8_9aOps), "armv8.9a">;
36def HasV9_0a         : Predicate<"Subtarget->hasV9_0aOps()">,
37                                 AssemblerPredicateWithAll<(all_of HasV9_0aOps), "armv9-a">;
38def HasV9_1a         : Predicate<"Subtarget->hasV9_1aOps()">,
39                                 AssemblerPredicateWithAll<(all_of HasV9_1aOps), "armv9.1a">;
40def HasV9_2a         : Predicate<"Subtarget->hasV9_2aOps()">,
41                                 AssemblerPredicateWithAll<(all_of HasV9_2aOps), "armv9.2a">;
42def HasV9_3a         : Predicate<"Subtarget->hasV9_3aOps()">,
43                                 AssemblerPredicateWithAll<(all_of HasV9_3aOps), "armv9.3a">;
44def HasV9_4a         : Predicate<"Subtarget->hasV9_4aOps()">,
45                                 AssemblerPredicateWithAll<(all_of HasV9_4aOps), "armv9.4a">;
46def HasV8_0r         : Predicate<"Subtarget->hasV8_0rOps()">,
47                                 AssemblerPredicateWithAll<(all_of HasV8_0rOps), "armv8-r">;
48
49def HasEL2VMSA       : Predicate<"Subtarget->hasEL2VMSA()">,
50                       AssemblerPredicateWithAll<(all_of FeatureEL2VMSA), "el2vmsa">;
51
52def HasEL3           : Predicate<"Subtarget->hasEL3()">,
53                       AssemblerPredicateWithAll<(all_of FeatureEL3), "el3">;
54
55def HasVH            : Predicate<"Subtarget->hasVH()">,
56                       AssemblerPredicateWithAll<(all_of FeatureVH), "vh">;
57
58def HasLOR           : Predicate<"Subtarget->hasLOR()">,
59                       AssemblerPredicateWithAll<(all_of FeatureLOR), "lor">;
60
61def HasPAuth         : Predicate<"Subtarget->hasPAuth()">,
62                       AssemblerPredicateWithAll<(all_of FeaturePAuth), "pauth">;
63
64def HasJS            : Predicate<"Subtarget->hasJS()">,
65                       AssemblerPredicateWithAll<(all_of FeatureJS), "jsconv">;
66
67def HasCCIDX         : Predicate<"Subtarget->hasCCIDX()">,
68                       AssemblerPredicateWithAll<(all_of FeatureCCIDX), "ccidx">;
69
70def HasComplxNum      : Predicate<"Subtarget->hasComplxNum()">,
71                       AssemblerPredicateWithAll<(all_of FeatureComplxNum), "complxnum">;
72
73def HasNV            : Predicate<"Subtarget->hasNV()">,
74                       AssemblerPredicateWithAll<(all_of FeatureNV), "nv">;
75
76def HasMPAM          : Predicate<"Subtarget->hasMPAM()">,
77                       AssemblerPredicateWithAll<(all_of FeatureMPAM), "mpam">;
78
79def HasDIT           : Predicate<"Subtarget->hasDIT()">,
80                       AssemblerPredicateWithAll<(all_of FeatureDIT), "dit">;
81
82def HasTRACEV8_4         : Predicate<"Subtarget->hasTRACEV8_4()">,
83                       AssemblerPredicateWithAll<(all_of FeatureTRACEV8_4), "tracev8.4">;
84
85def HasAM            : Predicate<"Subtarget->hasAM()">,
86                       AssemblerPredicateWithAll<(all_of FeatureAM), "am">;
87
88def HasSEL2          : Predicate<"Subtarget->hasSEL2()">,
89                       AssemblerPredicateWithAll<(all_of FeatureSEL2), "sel2">;
90
91def HasTLB_RMI          : Predicate<"Subtarget->hasTLB_RMI()">,
92                       AssemblerPredicateWithAll<(all_of FeatureTLB_RMI), "tlb-rmi">;
93
94def HasFlagM         : Predicate<"Subtarget->hasFlagM()">,
95                       AssemblerPredicateWithAll<(all_of FeatureFlagM), "flagm">;
96
97def HasRCPC_IMMO      : Predicate<"Subtarget->hasRCPCImm()">,
98                       AssemblerPredicateWithAll<(all_of FeatureRCPC_IMMO), "rcpc-immo">;
99
100def HasFPARMv8       : Predicate<"Subtarget->hasFPARMv8()">,
101                               AssemblerPredicateWithAll<(all_of FeatureFPARMv8), "fp-armv8">;
102def HasNEON          : Predicate<"Subtarget->hasNEON()">,
103                                 AssemblerPredicateWithAll<(all_of FeatureNEON), "neon">;
104def HasCrypto        : Predicate<"Subtarget->hasCrypto()">,
105                                 AssemblerPredicateWithAll<(all_of FeatureCrypto), "crypto">;
106def HasSM4           : Predicate<"Subtarget->hasSM4()">,
107                                 AssemblerPredicateWithAll<(all_of FeatureSM4), "sm4">;
108def HasSHA3          : Predicate<"Subtarget->hasSHA3()">,
109                                 AssemblerPredicateWithAll<(all_of FeatureSHA3), "sha3">;
110def HasSHA2          : Predicate<"Subtarget->hasSHA2()">,
111                                 AssemblerPredicateWithAll<(all_of FeatureSHA2), "sha2">;
112def HasAES           : Predicate<"Subtarget->hasAES()">,
113                                 AssemblerPredicateWithAll<(all_of FeatureAES), "aes">;
114def HasDotProd       : Predicate<"Subtarget->hasDotProd()">,
115                                 AssemblerPredicateWithAll<(all_of FeatureDotProd), "dotprod">;
116def HasCRC           : Predicate<"Subtarget->hasCRC()">,
117                                 AssemblerPredicateWithAll<(all_of FeatureCRC), "crc">;
118def HasCSSC          : Predicate<"Subtarget->hasCSSC()">,
119                                 AssemblerPredicateWithAll<(all_of FeatureCSSC), "cssc">;
120def HasNoCSSC        : Predicate<"!Subtarget->hasCSSC()">;
121def HasLSE           : Predicate<"Subtarget->hasLSE()">,
122                                 AssemblerPredicateWithAll<(all_of FeatureLSE), "lse">;
123def HasNoLSE         : Predicate<"!Subtarget->hasLSE()">;
124def HasRAS           : Predicate<"Subtarget->hasRAS()">,
125                                 AssemblerPredicateWithAll<(all_of FeatureRAS), "ras">;
126def HasRDM           : Predicate<"Subtarget->hasRDM()">,
127                                 AssemblerPredicateWithAll<(all_of FeatureRDM), "rdm">;
128def HasFullFP16      : Predicate<"Subtarget->hasFullFP16()">,
129                                 AssemblerPredicateWithAll<(all_of FeatureFullFP16), "fullfp16">;
130def HasFP16FML       : Predicate<"Subtarget->hasFP16FML()">,
131                                 AssemblerPredicateWithAll<(all_of FeatureFP16FML), "fp16fml">;
132def HasSPE           : Predicate<"Subtarget->hasSPE()">,
133                                 AssemblerPredicateWithAll<(all_of FeatureSPE), "spe">;
134def HasFuseAES       : Predicate<"Subtarget->hasFuseAES()">,
135                                 AssemblerPredicateWithAll<(all_of FeatureFuseAES),
136                                 "fuse-aes">;
137def HasSVE           : Predicate<"Subtarget->hasSVE()">,
138                                 AssemblerPredicateWithAll<(all_of FeatureSVE), "sve">;
139def HasSVE2          : Predicate<"Subtarget->hasSVE2()">,
140                                 AssemblerPredicateWithAll<(all_of FeatureSVE2), "sve2">;
141def HasSVE2p1        : Predicate<"Subtarget->hasSVE2p1()">,
142                                 AssemblerPredicate<(any_of FeatureSVE2p1), "sve2p1">;
143def HasSVE2AES       : Predicate<"Subtarget->hasSVE2AES()">,
144                                 AssemblerPredicateWithAll<(all_of FeatureSVE2AES), "sve2-aes">;
145def HasSVE2SM4       : Predicate<"Subtarget->hasSVE2SM4()">,
146                                 AssemblerPredicateWithAll<(all_of FeatureSVE2SM4), "sve2-sm4">;
147def HasSVE2SHA3      : Predicate<"Subtarget->hasSVE2SHA3()">,
148                                 AssemblerPredicateWithAll<(all_of FeatureSVE2SHA3), "sve2-sha3">;
149def HasSVE2BitPerm   : Predicate<"Subtarget->hasSVE2BitPerm()">,
150                                 AssemblerPredicateWithAll<(all_of FeatureSVE2BitPerm), "sve2-bitperm">;
151def HasB16B16        : Predicate<"Subtarget->hasB16B16()">,
152                                 AssemblerPredicateWithAll<(all_of FeatureB16B16), "b16b16">;
153def HasSME           : Predicate<"Subtarget->hasSME()">,
154                                 AssemblerPredicateWithAll<(all_of FeatureSME), "sme">;
155def HasSMEF64F64     : Predicate<"Subtarget->hasSMEF64F64()">,
156                                 AssemblerPredicateWithAll<(all_of FeatureSMEF64F64), "sme-f64f64">;
157def HasSMEF16F16     : Predicate<"Subtarget->hasSMEF16F16()">,
158                                 AssemblerPredicateWithAll<(all_of FeatureSMEF16F16), "sme-f16f16">;
159def HasSMEI16I64     : Predicate<"Subtarget->hasSMEI16I64()">,
160                                 AssemblerPredicateWithAll<(all_of FeatureSMEI16I64), "sme-i16i64">;
161def HasSME2          : Predicate<"Subtarget->hasSME2()">,
162                                 AssemblerPredicateWithAll<(all_of FeatureSME2), "sme2">;
163def HasSME2p1        : Predicate<"Subtarget->hasSME2p1()">,
164                                 AssemblerPredicateWithAll<(all_of FeatureSME2p1), "sme2p1">;
165
166// A subset of SVE(2) instructions are legal in Streaming SVE execution mode,
167// they should be enabled if either has been specified.
168def HasSVEorSME
169    : Predicate<"Subtarget->hasSVEorSME()">,
170                AssemblerPredicateWithAll<(any_of FeatureSVE, FeatureSME),
171                "sve or sme">;
172def HasSVE2orSME
173    : Predicate<"Subtarget->hasSVE2() || Subtarget->hasSME()">,
174                AssemblerPredicateWithAll<(any_of FeatureSVE2, FeatureSME),
175                "sve2 or sme">;
176def HasSVE2p1_or_HasSME
177    : Predicate<"Subtarget->hasSVE2p1() || Subtarget->hasSME()">,
178                 AssemblerPredicateWithAll<(any_of FeatureSME, FeatureSVE2p1), "sme or sve2p1">;
179def HasSVE2p1_or_HasSME2
180    : Predicate<"Subtarget->hasSVE2p1() || Subtarget->hasSME2()">,
181                 AssemblerPredicateWithAll<(any_of FeatureSME2, FeatureSVE2p1), "sme2 or sve2p1">;
182def HasSVE2p1_or_HasSME2p1
183    : Predicate<"Subtarget->hasSVE2p1() || Subtarget->hasSME2p1()">,
184                 AssemblerPredicateWithAll<(any_of FeatureSME2p1, FeatureSVE2p1), "sme2p1 or sve2p1">;
185// A subset of NEON instructions are legal in Streaming SVE execution mode,
186// they should be enabled if either has been specified.
187def HasNEONorSME
188    : Predicate<"Subtarget->hasNEON() || Subtarget->hasSME()">,
189                AssemblerPredicateWithAll<(any_of FeatureNEON, FeatureSME),
190                "neon or sme">;
191def HasRCPC          : Predicate<"Subtarget->hasRCPC()">,
192                                 AssemblerPredicateWithAll<(all_of FeatureRCPC), "rcpc">;
193def HasAltNZCV       : Predicate<"Subtarget->hasAlternativeNZCV()">,
194                       AssemblerPredicateWithAll<(all_of FeatureAltFPCmp), "altnzcv">;
195def HasFRInt3264     : Predicate<"Subtarget->hasFRInt3264()">,
196                       AssemblerPredicateWithAll<(all_of FeatureFRInt3264), "frint3264">;
197def HasSB            : Predicate<"Subtarget->hasSB()">,
198                       AssemblerPredicateWithAll<(all_of FeatureSB), "sb">;
199def HasPredRes      : Predicate<"Subtarget->hasPredRes()">,
200                       AssemblerPredicateWithAll<(all_of FeaturePredRes), "predres">;
201def HasCCDP          : Predicate<"Subtarget->hasCCDP()">,
202                       AssemblerPredicateWithAll<(all_of FeatureCacheDeepPersist), "ccdp">;
203def HasBTI           : Predicate<"Subtarget->hasBTI()">,
204                       AssemblerPredicateWithAll<(all_of FeatureBranchTargetId), "bti">;
205def HasMTE           : Predicate<"Subtarget->hasMTE()">,
206                       AssemblerPredicateWithAll<(all_of FeatureMTE), "mte">;
207def HasTME           : Predicate<"Subtarget->hasTME()">,
208                       AssemblerPredicateWithAll<(all_of FeatureTME), "tme">;
209def HasETE           : Predicate<"Subtarget->hasETE()">,
210                       AssemblerPredicateWithAll<(all_of FeatureETE), "ete">;
211def HasTRBE          : Predicate<"Subtarget->hasTRBE()">,
212                       AssemblerPredicateWithAll<(all_of FeatureTRBE), "trbe">;
213def HasBF16          : Predicate<"Subtarget->hasBF16()">,
214                       AssemblerPredicateWithAll<(all_of FeatureBF16), "bf16">;
215def HasMatMulInt8    : Predicate<"Subtarget->hasMatMulInt8()">,
216                       AssemblerPredicateWithAll<(all_of FeatureMatMulInt8), "i8mm">;
217def HasMatMulFP32    : Predicate<"Subtarget->hasMatMulFP32()">,
218                       AssemblerPredicateWithAll<(all_of FeatureMatMulFP32), "f32mm">;
219def HasMatMulFP64    : Predicate<"Subtarget->hasMatMulFP64()">,
220                       AssemblerPredicateWithAll<(all_of FeatureMatMulFP64), "f64mm">;
221def HasXS            : Predicate<"Subtarget->hasXS()">,
222                       AssemblerPredicateWithAll<(all_of FeatureXS), "xs">;
223def HasWFxT          : Predicate<"Subtarget->hasWFxT()">,
224                       AssemblerPredicateWithAll<(all_of FeatureWFxT), "wfxt">;
225def HasLS64          : Predicate<"Subtarget->hasLS64()">,
226                       AssemblerPredicateWithAll<(all_of FeatureLS64), "ls64">;
227def HasBRBE          : Predicate<"Subtarget->hasBRBE()">,
228                       AssemblerPredicateWithAll<(all_of FeatureBRBE), "brbe">;
229def HasSPE_EEF       : Predicate<"Subtarget->hasSPE_EEF()">,
230                       AssemblerPredicateWithAll<(all_of FeatureSPE_EEF), "spe-eef">;
231def HasHBC           : Predicate<"Subtarget->hasHBC()">,
232                       AssemblerPredicateWithAll<(all_of FeatureHBC), "hbc">;
233def HasMOPS          : Predicate<"Subtarget->hasMOPS()">,
234                       AssemblerPredicateWithAll<(all_of FeatureMOPS), "mops">;
235def HasCLRBHB        : Predicate<"Subtarget->hasCLRBHB()">,
236                       AssemblerPredicateWithAll<(all_of FeatureCLRBHB), "clrbhb">;
237def HasSPECRES2      : Predicate<"Subtarget->hasSPECRES2()">,
238                       AssemblerPredicateWithAll<(all_of FeatureSPECRES2), "specres2">;
239def HasITE           : Predicate<"Subtarget->hasITE()">,
240                       AssemblerPredicateWithAll<(all_of FeatureITE), "ite">;
241def HasTHE           : Predicate<"Subtarget->hasTHE()">,
242                       AssemblerPredicateWithAll<(all_of FeatureTHE), "the">;
243def HasRCPC3         : Predicate<"Subtarget->hasRCPC3()">,
244                       AssemblerPredicateWithAll<(all_of FeatureRCPC3), "rcpc3">;
245def HasLSE128        : Predicate<"Subtarget->hasLSE128()">,
246                       AssemblerPredicateWithAll<(all_of FeatureLSE128), "lse128">;
247def HasD128          : Predicate<"Subtarget->hasD128()">,
248                       AssemblerPredicateWithAll<(all_of FeatureD128), "d128">;
249def IsLE             : Predicate<"Subtarget->isLittleEndian()">;
250def IsBE             : Predicate<"!Subtarget->isLittleEndian()">;
251def IsWindows        : Predicate<"Subtarget->isTargetWindows()">;
252def UseExperimentalZeroingPseudos
253    : Predicate<"Subtarget->useExperimentalZeroingPseudos()">;
254def UseAlternateSExtLoadCVTF32
255    : Predicate<"Subtarget->useAlternateSExtLoadCVTF32Pattern()">;
256
257def UseNegativeImmediates
258    : Predicate<"false">, AssemblerPredicate<(all_of (not FeatureNoNegativeImmediates)),
259                                             "NegativeImmediates">;
260
261def UseScalarIncVL : Predicate<"Subtarget->useScalarIncVL()">;
262
263def NotInStreamingSVEMode  : Predicate<"!Subtarget->forceStreamingCompatibleSVE()">;
264
265def AArch64LocalRecover : SDNode<"ISD::LOCAL_RECOVER",
266                                  SDTypeProfile<1, 1, [SDTCisSameAs<0, 1>,
267                                                       SDTCisInt<1>]>>;
268
269
270//===----------------------------------------------------------------------===//
271// AArch64-specific DAG Nodes.
272//
273
274// SDTBinaryArithWithFlagsOut - RES1, FLAGS = op LHS, RHS
275def SDTBinaryArithWithFlagsOut : SDTypeProfile<2, 2,
276                                              [SDTCisSameAs<0, 2>,
277                                               SDTCisSameAs<0, 3>,
278                                               SDTCisInt<0>, SDTCisVT<1, i32>]>;
279
280// SDTBinaryArithWithFlagsIn - RES1, FLAGS = op LHS, RHS, FLAGS
281def SDTBinaryArithWithFlagsIn : SDTypeProfile<1, 3,
282                                            [SDTCisSameAs<0, 1>,
283                                             SDTCisSameAs<0, 2>,
284                                             SDTCisInt<0>,
285                                             SDTCisVT<3, i32>]>;
286
287// SDTBinaryArithWithFlagsInOut - RES1, FLAGS = op LHS, RHS, FLAGS
288def SDTBinaryArithWithFlagsInOut : SDTypeProfile<2, 3,
289                                            [SDTCisSameAs<0, 2>,
290                                             SDTCisSameAs<0, 3>,
291                                             SDTCisInt<0>,
292                                             SDTCisVT<1, i32>,
293                                             SDTCisVT<4, i32>]>;
294
295def SDT_AArch64Brcond  : SDTypeProfile<0, 3,
296                                     [SDTCisVT<0, OtherVT>, SDTCisVT<1, i32>,
297                                      SDTCisVT<2, i32>]>;
298def SDT_AArch64cbz : SDTypeProfile<0, 2, [SDTCisInt<0>, SDTCisVT<1, OtherVT>]>;
299def SDT_AArch64tbz : SDTypeProfile<0, 3, [SDTCisInt<0>, SDTCisInt<1>,
300                                        SDTCisVT<2, OtherVT>]>;
301
302
303def SDT_AArch64CSel  : SDTypeProfile<1, 4,
304                                   [SDTCisSameAs<0, 1>,
305                                    SDTCisSameAs<0, 2>,
306                                    SDTCisInt<3>,
307                                    SDTCisVT<4, i32>]>;
308def SDT_AArch64CCMP : SDTypeProfile<1, 5,
309                                    [SDTCisVT<0, i32>,
310                                     SDTCisInt<1>,
311                                     SDTCisSameAs<1, 2>,
312                                     SDTCisInt<3>,
313                                     SDTCisInt<4>,
314                                     SDTCisVT<5, i32>]>;
315def SDT_AArch64FCCMP : SDTypeProfile<1, 5,
316                                     [SDTCisVT<0, i32>,
317                                      SDTCisFP<1>,
318                                      SDTCisSameAs<1, 2>,
319                                      SDTCisInt<3>,
320                                      SDTCisInt<4>,
321                                      SDTCisVT<5, i32>]>;
322def SDT_AArch64FCmp   : SDTypeProfile<0, 2,
323                                   [SDTCisFP<0>,
324                                    SDTCisSameAs<0, 1>]>;
325def SDT_AArch64Dup   : SDTypeProfile<1, 1, [SDTCisVec<0>]>;
326def SDT_AArch64DupLane   : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisInt<2>]>;
327def SDT_AArch64Insr  : SDTypeProfile<1, 2, [SDTCisVec<0>]>;
328def SDT_AArch64Zip   : SDTypeProfile<1, 2, [SDTCisVec<0>,
329                                          SDTCisSameAs<0, 1>,
330                                          SDTCisSameAs<0, 2>]>;
331def SDT_AArch64MOVIedit : SDTypeProfile<1, 1, [SDTCisInt<1>]>;
332def SDT_AArch64MOVIshift : SDTypeProfile<1, 2, [SDTCisInt<1>, SDTCisInt<2>]>;
333def SDT_AArch64vecimm : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0,1>,
334                                           SDTCisInt<2>, SDTCisInt<3>]>;
335def SDT_AArch64UnaryVec: SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0,1>]>;
336def SDT_AArch64ExtVec: SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0,1>,
337                                          SDTCisSameAs<0,2>, SDTCisInt<3>]>;
338def SDT_AArch64vshift : SDTypeProfile<1, 2, [SDTCisSameAs<0,1>, SDTCisInt<2>]>;
339def SDT_AArch64Dot: SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0,1>,
340                                         SDTCisVec<2>, SDTCisSameAs<2,3>]>;
341
342def SDT_AArch64vshiftinsert : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisInt<3>,
343                                                 SDTCisSameAs<0,1>,
344                                                 SDTCisSameAs<0,2>]>;
345
346def SDT_AArch64unvec : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0,1>]>;
347def SDT_AArch64fcmpz : SDTypeProfile<1, 1, []>;
348def SDT_AArch64fcmp  : SDTypeProfile<1, 2, [SDTCisSameAs<1,2>]>;
349def SDT_AArch64binvec : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisSameAs<0,1>,
350                                           SDTCisSameAs<0,2>]>;
351def SDT_AArch64trivec : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0,1>,
352                                           SDTCisSameAs<0,2>,
353                                           SDTCisSameAs<0,3>]>;
354def SDT_AArch64TCRET : SDTypeProfile<0, 2, [SDTCisPtrTy<0>]>;
355def SDT_AArch64PREFETCH : SDTypeProfile<0, 2, [SDTCisVT<0, i32>, SDTCisPtrTy<1>]>;
356
357def SDT_AArch64ITOF  : SDTypeProfile<1, 1, [SDTCisFP<0>, SDTCisSameAs<0,1>]>;
358
359def SDT_AArch64TLSDescCall : SDTypeProfile<0, -2, [SDTCisPtrTy<0>,
360                                                 SDTCisPtrTy<1>]>;
361
362def SDT_AArch64uaddlp : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisVec<1>]>;
363
364def SDT_AArch64ldp : SDTypeProfile<2, 1, [SDTCisVT<0, i64>, SDTCisSameAs<0, 1>, SDTCisPtrTy<2>]>;
365def SDT_AArch64ldnp : SDTypeProfile<2, 1, [SDTCisVT<0, v4i32>, SDTCisSameAs<0, 1>, SDTCisPtrTy<2>]>;
366def SDT_AArch64stp : SDTypeProfile<0, 3, [SDTCisVT<0, i64>, SDTCisSameAs<0, 1>, SDTCisPtrTy<2>]>;
367def SDT_AArch64stnp : SDTypeProfile<0, 3, [SDTCisVT<0, v4i32>, SDTCisSameAs<0, 1>, SDTCisPtrTy<2>]>;
368
369// Generates the general dynamic sequences, i.e.
370//  adrp  x0, :tlsdesc:var
371//  ldr   x1, [x0, #:tlsdesc_lo12:var]
372//  add   x0, x0, #:tlsdesc_lo12:var
373//  .tlsdesccall var
374//  blr   x1
375
376// (the TPIDR_EL0 offset is put directly in X0, hence no "result" here)
377// number of operands (the variable)
378def SDT_AArch64TLSDescCallSeq : SDTypeProfile<0,1,
379                                          [SDTCisPtrTy<0>]>;
380
381def SDT_AArch64WrapperLarge : SDTypeProfile<1, 4,
382                                        [SDTCisVT<0, i64>, SDTCisVT<1, i32>,
383                                         SDTCisSameAs<1, 2>, SDTCisSameAs<1, 3>,
384                                         SDTCisSameAs<1, 4>]>;
385
386def SDT_AArch64TBL : SDTypeProfile<1, 2, [
387  SDTCisVec<0>, SDTCisSameAs<0, 1>, SDTCisInt<2>
388]>;
389
390// non-extending masked load fragment.
391def nonext_masked_load :
392  PatFrag<(ops node:$ptr, node:$pred, node:$def),
393          (masked_ld node:$ptr, undef, node:$pred, node:$def), [{
394  return cast<MaskedLoadSDNode>(N)->getExtensionType() == ISD::NON_EXTLOAD &&
395         cast<MaskedLoadSDNode>(N)->isUnindexed() &&
396         !cast<MaskedLoadSDNode>(N)->isNonTemporal();
397}]>;
398// Any/Zero extending masked load fragments.
399def azext_masked_load :
400  PatFrag<(ops node:$ptr, node:$pred, node:$def),
401          (masked_ld node:$ptr, undef, node:$pred, node:$def),[{
402  return (cast<MaskedLoadSDNode>(N)->getExtensionType() == ISD::EXTLOAD ||
403          cast<MaskedLoadSDNode>(N)->getExtensionType() == ISD::ZEXTLOAD) &&
404         cast<MaskedLoadSDNode>(N)->isUnindexed();
405}]>;
406def azext_masked_load_i8 :
407  PatFrag<(ops node:$ptr, node:$pred, node:$def),
408          (azext_masked_load node:$ptr, node:$pred, node:$def), [{
409  return cast<MaskedLoadSDNode>(N)->getMemoryVT().getScalarType() == MVT::i8;
410}]>;
411def azext_masked_load_i16 :
412  PatFrag<(ops node:$ptr, node:$pred, node:$def),
413          (azext_masked_load node:$ptr, node:$pred, node:$def), [{
414  return cast<MaskedLoadSDNode>(N)->getMemoryVT().getScalarType() == MVT::i16;
415}]>;
416def azext_masked_load_i32 :
417  PatFrag<(ops node:$ptr, node:$pred, node:$def),
418          (azext_masked_load node:$ptr, node:$pred, node:$def), [{
419  return cast<MaskedLoadSDNode>(N)->getMemoryVT().getScalarType() == MVT::i32;
420}]>;
421// Sign extending masked load fragments.
422def sext_masked_load :
423  PatFrag<(ops node:$ptr, node:$pred, node:$def),
424          (masked_ld node:$ptr, undef, node:$pred, node:$def), [{
425  return cast<MaskedLoadSDNode>(N)->getExtensionType() == ISD::SEXTLOAD &&
426         cast<MaskedLoadSDNode>(N)->isUnindexed();
427}]>;
428def sext_masked_load_i8 :
429  PatFrag<(ops node:$ptr, node:$pred, node:$def),
430          (sext_masked_load node:$ptr, node:$pred, node:$def), [{
431  return cast<MaskedLoadSDNode>(N)->getMemoryVT().getScalarType() == MVT::i8;
432}]>;
433def sext_masked_load_i16 :
434  PatFrag<(ops node:$ptr, node:$pred, node:$def),
435          (sext_masked_load node:$ptr, node:$pred, node:$def), [{
436  return cast<MaskedLoadSDNode>(N)->getMemoryVT().getScalarType() == MVT::i16;
437}]>;
438def sext_masked_load_i32 :
439  PatFrag<(ops node:$ptr, node:$pred, node:$def),
440          (sext_masked_load node:$ptr, node:$pred, node:$def), [{
441  return cast<MaskedLoadSDNode>(N)->getMemoryVT().getScalarType() == MVT::i32;
442}]>;
443
444def non_temporal_load :
445   PatFrag<(ops node:$ptr, node:$pred, node:$def),
446           (masked_ld node:$ptr, undef, node:$pred, node:$def), [{
447   return cast<MaskedLoadSDNode>(N)->getExtensionType() == ISD::NON_EXTLOAD &&
448          cast<MaskedLoadSDNode>(N)->isUnindexed() &&
449          cast<MaskedLoadSDNode>(N)->isNonTemporal();
450}]>;
451
452// non-truncating masked store fragment.
453def nontrunc_masked_store :
454  PatFrag<(ops node:$val, node:$ptr, node:$pred),
455          (masked_st node:$val, node:$ptr, undef, node:$pred), [{
456  return !cast<MaskedStoreSDNode>(N)->isTruncatingStore() &&
457         cast<MaskedStoreSDNode>(N)->isUnindexed() &&
458         !cast<MaskedStoreSDNode>(N)->isNonTemporal();
459}]>;
460// truncating masked store fragments.
461def trunc_masked_store :
462  PatFrag<(ops node:$val, node:$ptr, node:$pred),
463          (masked_st node:$val, node:$ptr, undef, node:$pred), [{
464  return cast<MaskedStoreSDNode>(N)->isTruncatingStore() &&
465         cast<MaskedStoreSDNode>(N)->isUnindexed();
466}]>;
467def trunc_masked_store_i8 :
468  PatFrag<(ops node:$val, node:$ptr, node:$pred),
469          (trunc_masked_store node:$val, node:$ptr, node:$pred), [{
470  return cast<MaskedStoreSDNode>(N)->getMemoryVT().getScalarType() == MVT::i8;
471}]>;
472def trunc_masked_store_i16 :
473  PatFrag<(ops node:$val, node:$ptr, node:$pred),
474          (trunc_masked_store node:$val, node:$ptr, node:$pred), [{
475  return cast<MaskedStoreSDNode>(N)->getMemoryVT().getScalarType() == MVT::i16;
476}]>;
477def trunc_masked_store_i32 :
478  PatFrag<(ops node:$val, node:$ptr, node:$pred),
479          (trunc_masked_store node:$val, node:$ptr, node:$pred), [{
480  return cast<MaskedStoreSDNode>(N)->getMemoryVT().getScalarType() == MVT::i32;
481}]>;
482
483def non_temporal_store :
484  PatFrag<(ops node:$val, node:$ptr, node:$pred),
485          (masked_st node:$val, node:$ptr, undef, node:$pred), [{
486  return !cast<MaskedStoreSDNode>(N)->isTruncatingStore() &&
487         cast<MaskedStoreSDNode>(N)->isUnindexed() &&
488         cast<MaskedStoreSDNode>(N)->isNonTemporal();
489}]>;
490
491multiclass masked_gather_scatter<PatFrags GatherScatterOp> {
492  // offsets = (signed)Index << sizeof(elt)
493  def NAME#_signed_scaled :
494    PatFrag<(ops node:$val, node:$pred, node:$ptr, node:$idx),
495            (GatherScatterOp node:$val, node:$pred, node:$ptr, node:$idx),[{
496    auto MGS = cast<MaskedGatherScatterSDNode>(N);
497    bool Signed = MGS->isIndexSigned() ||
498        MGS->getIndex().getValueType().getVectorElementType() == MVT::i64;
499    return Signed && MGS->isIndexScaled();
500  }]>;
501  // offsets = (signed)Index
502  def NAME#_signed_unscaled :
503    PatFrag<(ops node:$val, node:$pred, node:$ptr, node:$idx),
504            (GatherScatterOp node:$val, node:$pred, node:$ptr, node:$idx),[{
505    auto MGS = cast<MaskedGatherScatterSDNode>(N);
506    bool Signed = MGS->isIndexSigned() ||
507        MGS->getIndex().getValueType().getVectorElementType() == MVT::i64;
508    return Signed && !MGS->isIndexScaled();
509  }]>;
510  // offsets = (unsigned)Index << sizeof(elt)
511  def NAME#_unsigned_scaled :
512    PatFrag<(ops node:$val, node:$pred, node:$ptr, node:$idx),
513            (GatherScatterOp node:$val, node:$pred, node:$ptr, node:$idx),[{
514    auto MGS = cast<MaskedGatherScatterSDNode>(N);
515    bool Signed = MGS->isIndexSigned() ||
516        MGS->getIndex().getValueType().getVectorElementType() == MVT::i64;
517    return !Signed && MGS->isIndexScaled();
518  }]>;
519  // offsets = (unsigned)Index
520  def NAME#_unsigned_unscaled :
521    PatFrag<(ops node:$val, node:$pred, node:$ptr, node:$idx),
522            (GatherScatterOp node:$val, node:$pred, node:$ptr, node:$idx),[{
523    auto MGS = cast<MaskedGatherScatterSDNode>(N);
524    bool Signed = MGS->isIndexSigned() ||
525        MGS->getIndex().getValueType().getVectorElementType() == MVT::i64;
526    return !Signed && !MGS->isIndexScaled();
527  }]>;
528}
529
530defm nonext_masked_gather    : masked_gather_scatter<nonext_masked_gather>;
531defm azext_masked_gather_i8  : masked_gather_scatter<azext_masked_gather_i8>;
532defm azext_masked_gather_i16 : masked_gather_scatter<azext_masked_gather_i16>;
533defm azext_masked_gather_i32 : masked_gather_scatter<azext_masked_gather_i32>;
534defm sext_masked_gather_i8   : masked_gather_scatter<sext_masked_gather_i8>;
535defm sext_masked_gather_i16  : masked_gather_scatter<sext_masked_gather_i16>;
536defm sext_masked_gather_i32  : masked_gather_scatter<sext_masked_gather_i32>;
537
538defm nontrunc_masked_scatter  : masked_gather_scatter<nontrunc_masked_scatter>;
539defm trunc_masked_scatter_i8  : masked_gather_scatter<trunc_masked_scatter_i8>;
540defm trunc_masked_scatter_i16 : masked_gather_scatter<trunc_masked_scatter_i16>;
541defm trunc_masked_scatter_i32 : masked_gather_scatter<trunc_masked_scatter_i32>;
542
543// top16Zero - answer true if the upper 16 bits of $src are 0, false otherwise
544def top16Zero: PatLeaf<(i32 GPR32:$src), [{
545  return SDValue(N,0)->getValueType(0) == MVT::i32 &&
546         CurDAG->MaskedValueIsZero(SDValue(N,0), APInt::getHighBitsSet(32, 16));
547  }]>;
548
549// top32Zero - answer true if the upper 32 bits of $src are 0, false otherwise
550def top32Zero: PatLeaf<(i64 GPR64:$src), [{
551  return SDValue(N,0)->getValueType(0) == MVT::i64 &&
552         CurDAG->MaskedValueIsZero(SDValue(N,0), APInt::getHighBitsSet(64, 32));
553  }]>;
554
555// topbitsallzero - Return true if all bits except the lowest bit are known zero
556def topbitsallzero32: PatLeaf<(i32 GPR32:$src), [{
557  return SDValue(N,0)->getValueType(0) == MVT::i32 &&
558         CurDAG->MaskedValueIsZero(SDValue(N,0), APInt::getHighBitsSet(32, 31));
559  }]>;
560def topbitsallzero64: PatLeaf<(i64 GPR64:$src), [{
561  return SDValue(N,0)->getValueType(0) == MVT::i64 &&
562         CurDAG->MaskedValueIsZero(SDValue(N,0), APInt::getHighBitsSet(64, 63));
563  }]>;
564
565// Node definitions.
566def AArch64adrp          : SDNode<"AArch64ISD::ADRP", SDTIntUnaryOp, []>;
567def AArch64adr           : SDNode<"AArch64ISD::ADR", SDTIntUnaryOp, []>;
568def AArch64addlow        : SDNode<"AArch64ISD::ADDlow", SDTIntBinOp, []>;
569def AArch64LOADgot       : SDNode<"AArch64ISD::LOADgot", SDTIntUnaryOp>;
570def AArch64callseq_start : SDNode<"ISD::CALLSEQ_START",
571                                SDCallSeqStart<[ SDTCisVT<0, i32>,
572                                                 SDTCisVT<1, i32> ]>,
573                                [SDNPHasChain, SDNPOutGlue]>;
574def AArch64callseq_end   : SDNode<"ISD::CALLSEQ_END",
575                                SDCallSeqEnd<[ SDTCisVT<0, i32>,
576                                               SDTCisVT<1, i32> ]>,
577                                [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue]>;
578def AArch64call          : SDNode<"AArch64ISD::CALL",
579                                SDTypeProfile<0, -1, [SDTCisPtrTy<0>]>,
580                                [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue,
581                                 SDNPVariadic]>;
582
583def AArch64call_bti      : SDNode<"AArch64ISD::CALL_BTI",
584                                SDTypeProfile<0, -1, [SDTCisPtrTy<0>]>,
585                                [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue,
586                                 SDNPVariadic]>;
587
588def AArch64call_rvmarker: SDNode<"AArch64ISD::CALL_RVMARKER",
589                             SDTypeProfile<0, -1, [SDTCisPtrTy<0>]>,
590                             [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue,
591                              SDNPVariadic]>;
592
593def AArch64brcond        : SDNode<"AArch64ISD::BRCOND", SDT_AArch64Brcond,
594                                [SDNPHasChain]>;
595def AArch64cbz           : SDNode<"AArch64ISD::CBZ", SDT_AArch64cbz,
596                                [SDNPHasChain]>;
597def AArch64cbnz           : SDNode<"AArch64ISD::CBNZ", SDT_AArch64cbz,
598                                [SDNPHasChain]>;
599def AArch64tbz           : SDNode<"AArch64ISD::TBZ", SDT_AArch64tbz,
600                                [SDNPHasChain]>;
601def AArch64tbnz           : SDNode<"AArch64ISD::TBNZ", SDT_AArch64tbz,
602                                [SDNPHasChain]>;
603
604
605def AArch64csel          : SDNode<"AArch64ISD::CSEL", SDT_AArch64CSel>;
606def AArch64csinv         : SDNode<"AArch64ISD::CSINV", SDT_AArch64CSel>;
607def AArch64csneg         : SDNode<"AArch64ISD::CSNEG", SDT_AArch64CSel>;
608def AArch64csinc         : SDNode<"AArch64ISD::CSINC", SDT_AArch64CSel>;
609def AArch64retflag       : SDNode<"AArch64ISD::RET_FLAG", SDTNone,
610                                [SDNPHasChain, SDNPOptInGlue, SDNPVariadic]>;
611def AArch64adc       : SDNode<"AArch64ISD::ADC",  SDTBinaryArithWithFlagsIn >;
612def AArch64sbc       : SDNode<"AArch64ISD::SBC",  SDTBinaryArithWithFlagsIn>;
613def AArch64add_flag  : SDNode<"AArch64ISD::ADDS",  SDTBinaryArithWithFlagsOut,
614                            [SDNPCommutative]>;
615def AArch64sub_flag  : SDNode<"AArch64ISD::SUBS",  SDTBinaryArithWithFlagsOut>;
616def AArch64and_flag  : SDNode<"AArch64ISD::ANDS",  SDTBinaryArithWithFlagsOut,
617                            [SDNPCommutative]>;
618def AArch64adc_flag  : SDNode<"AArch64ISD::ADCS",  SDTBinaryArithWithFlagsInOut>;
619def AArch64sbc_flag  : SDNode<"AArch64ISD::SBCS",  SDTBinaryArithWithFlagsInOut>;
620
621def AArch64ccmp      : SDNode<"AArch64ISD::CCMP",  SDT_AArch64CCMP>;
622def AArch64ccmn      : SDNode<"AArch64ISD::CCMN",  SDT_AArch64CCMP>;
623def AArch64fccmp     : SDNode<"AArch64ISD::FCCMP", SDT_AArch64FCCMP>;
624
625def AArch64threadpointer : SDNode<"AArch64ISD::THREAD_POINTER", SDTPtrLeaf>;
626
627def AArch64fcmp         : SDNode<"AArch64ISD::FCMP", SDT_AArch64FCmp>;
628def AArch64strict_fcmp  : SDNode<"AArch64ISD::STRICT_FCMP", SDT_AArch64FCmp,
629                                 [SDNPHasChain]>;
630def AArch64strict_fcmpe : SDNode<"AArch64ISD::STRICT_FCMPE", SDT_AArch64FCmp,
631                                 [SDNPHasChain]>;
632def AArch64any_fcmp     : PatFrags<(ops node:$lhs, node:$rhs),
633                                   [(AArch64strict_fcmp node:$lhs, node:$rhs),
634                                    (AArch64fcmp node:$lhs, node:$rhs)]>;
635
636def AArch64dup       : SDNode<"AArch64ISD::DUP", SDT_AArch64Dup>;
637def AArch64duplane8  : SDNode<"AArch64ISD::DUPLANE8", SDT_AArch64DupLane>;
638def AArch64duplane16 : SDNode<"AArch64ISD::DUPLANE16", SDT_AArch64DupLane>;
639def AArch64duplane32 : SDNode<"AArch64ISD::DUPLANE32", SDT_AArch64DupLane>;
640def AArch64duplane64 : SDNode<"AArch64ISD::DUPLANE64", SDT_AArch64DupLane>;
641def AArch64duplane128 : SDNode<"AArch64ISD::DUPLANE128", SDT_AArch64DupLane>;
642
643def AArch64insr      : SDNode<"AArch64ISD::INSR", SDT_AArch64Insr>;
644
645def AArch64zip1      : SDNode<"AArch64ISD::ZIP1", SDT_AArch64Zip>;
646def AArch64zip2      : SDNode<"AArch64ISD::ZIP2", SDT_AArch64Zip>;
647def AArch64uzp1      : SDNode<"AArch64ISD::UZP1", SDT_AArch64Zip>;
648def AArch64uzp2      : SDNode<"AArch64ISD::UZP2", SDT_AArch64Zip>;
649def AArch64trn1      : SDNode<"AArch64ISD::TRN1", SDT_AArch64Zip>;
650def AArch64trn2      : SDNode<"AArch64ISD::TRN2", SDT_AArch64Zip>;
651
652def AArch64movi_edit : SDNode<"AArch64ISD::MOVIedit", SDT_AArch64MOVIedit>;
653def AArch64movi_shift : SDNode<"AArch64ISD::MOVIshift", SDT_AArch64MOVIshift>;
654def AArch64movi_msl : SDNode<"AArch64ISD::MOVImsl", SDT_AArch64MOVIshift>;
655def AArch64mvni_shift : SDNode<"AArch64ISD::MVNIshift", SDT_AArch64MOVIshift>;
656def AArch64mvni_msl : SDNode<"AArch64ISD::MVNImsl", SDT_AArch64MOVIshift>;
657def AArch64movi : SDNode<"AArch64ISD::MOVI", SDT_AArch64MOVIedit>;
658def AArch64fmov : SDNode<"AArch64ISD::FMOV", SDT_AArch64MOVIedit>;
659
660def AArch64rev16 : SDNode<"AArch64ISD::REV16", SDT_AArch64UnaryVec>;
661def AArch64rev32 : SDNode<"AArch64ISD::REV32", SDT_AArch64UnaryVec>;
662def AArch64rev64 : SDNode<"AArch64ISD::REV64", SDT_AArch64UnaryVec>;
663def AArch64ext : SDNode<"AArch64ISD::EXT", SDT_AArch64ExtVec>;
664
665def AArch64vashr : SDNode<"AArch64ISD::VASHR", SDT_AArch64vshift>;
666def AArch64vlshr : SDNode<"AArch64ISD::VLSHR", SDT_AArch64vshift>;
667def AArch64vshl : SDNode<"AArch64ISD::VSHL", SDT_AArch64vshift>;
668def AArch64sqshli : SDNode<"AArch64ISD::SQSHL_I", SDT_AArch64vshift>;
669def AArch64uqshli : SDNode<"AArch64ISD::UQSHL_I", SDT_AArch64vshift>;
670def AArch64sqshlui : SDNode<"AArch64ISD::SQSHLU_I", SDT_AArch64vshift>;
671def AArch64srshri : SDNode<"AArch64ISD::SRSHR_I", SDT_AArch64vshift>;
672def AArch64urshri : SDNode<"AArch64ISD::URSHR_I", SDT_AArch64vshift>;
673def AArch64vsli : SDNode<"AArch64ISD::VSLI", SDT_AArch64vshiftinsert>;
674def AArch64vsri : SDNode<"AArch64ISD::VSRI", SDT_AArch64vshiftinsert>;
675
676def AArch64bit: SDNode<"AArch64ISD::BIT", SDT_AArch64trivec>;
677def AArch64bsp: SDNode<"AArch64ISD::BSP", SDT_AArch64trivec>;
678
679def AArch64cmeq: SDNode<"AArch64ISD::CMEQ", SDT_AArch64binvec>;
680def AArch64cmge: SDNode<"AArch64ISD::CMGE", SDT_AArch64binvec>;
681def AArch64cmgt: SDNode<"AArch64ISD::CMGT", SDT_AArch64binvec>;
682def AArch64cmhi: SDNode<"AArch64ISD::CMHI", SDT_AArch64binvec>;
683def AArch64cmhs: SDNode<"AArch64ISD::CMHS", SDT_AArch64binvec>;
684
685def AArch64fcmeq: SDNode<"AArch64ISD::FCMEQ", SDT_AArch64fcmp>;
686def AArch64fcmge: SDNode<"AArch64ISD::FCMGE", SDT_AArch64fcmp>;
687def AArch64fcmgt: SDNode<"AArch64ISD::FCMGT", SDT_AArch64fcmp>;
688
689def AArch64cmeqz: SDNode<"AArch64ISD::CMEQz", SDT_AArch64unvec>;
690def AArch64cmgez: SDNode<"AArch64ISD::CMGEz", SDT_AArch64unvec>;
691def AArch64cmgtz: SDNode<"AArch64ISD::CMGTz", SDT_AArch64unvec>;
692def AArch64cmlez: SDNode<"AArch64ISD::CMLEz", SDT_AArch64unvec>;
693def AArch64cmltz: SDNode<"AArch64ISD::CMLTz", SDT_AArch64unvec>;
694def AArch64cmtst : PatFrag<(ops node:$LHS, node:$RHS),
695                        (vnot (AArch64cmeqz (and node:$LHS, node:$RHS)))>;
696
697def AArch64fcmeqz: SDNode<"AArch64ISD::FCMEQz", SDT_AArch64fcmpz>;
698def AArch64fcmgez: SDNode<"AArch64ISD::FCMGEz", SDT_AArch64fcmpz>;
699def AArch64fcmgtz: SDNode<"AArch64ISD::FCMGTz", SDT_AArch64fcmpz>;
700def AArch64fcmlez: SDNode<"AArch64ISD::FCMLEz", SDT_AArch64fcmpz>;
701def AArch64fcmltz: SDNode<"AArch64ISD::FCMLTz", SDT_AArch64fcmpz>;
702
703def AArch64bici: SDNode<"AArch64ISD::BICi", SDT_AArch64vecimm>;
704def AArch64orri: SDNode<"AArch64ISD::ORRi", SDT_AArch64vecimm>;
705
706def AArch64tcret: SDNode<"AArch64ISD::TC_RETURN", SDT_AArch64TCRET,
707                  [SDNPHasChain,  SDNPOptInGlue, SDNPVariadic]>;
708
709def AArch64Prefetch        : SDNode<"AArch64ISD::PREFETCH", SDT_AArch64PREFETCH,
710                               [SDNPHasChain, SDNPSideEffect]>;
711
712def AArch64sitof: SDNode<"AArch64ISD::SITOF", SDT_AArch64ITOF>;
713def AArch64uitof: SDNode<"AArch64ISD::UITOF", SDT_AArch64ITOF>;
714
715def AArch64tlsdesc_callseq : SDNode<"AArch64ISD::TLSDESC_CALLSEQ",
716                                    SDT_AArch64TLSDescCallSeq,
717                                    [SDNPInGlue, SDNPOutGlue, SDNPHasChain,
718                                     SDNPVariadic]>;
719
720
721def AArch64WrapperLarge : SDNode<"AArch64ISD::WrapperLarge",
722                                 SDT_AArch64WrapperLarge>;
723
724def AArch64NvCast : SDNode<"AArch64ISD::NVCAST", SDTUnaryOp>;
725
726def SDT_AArch64mull : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
727                                    SDTCisSameAs<1, 2>]>;
728def AArch64pmull    : SDNode<"AArch64ISD::PMULL", SDT_AArch64mull,
729                             [SDNPCommutative]>;
730def AArch64smull    : SDNode<"AArch64ISD::SMULL", SDT_AArch64mull,
731                             [SDNPCommutative]>;
732def AArch64umull    : SDNode<"AArch64ISD::UMULL", SDT_AArch64mull,
733                             [SDNPCommutative]>;
734
735def AArch64frecpe   : SDNode<"AArch64ISD::FRECPE", SDTFPUnaryOp>;
736def AArch64frecps   : SDNode<"AArch64ISD::FRECPS", SDTFPBinOp>;
737def AArch64frsqrte  : SDNode<"AArch64ISD::FRSQRTE", SDTFPUnaryOp>;
738def AArch64frsqrts  : SDNode<"AArch64ISD::FRSQRTS", SDTFPBinOp>;
739
740def AArch64sdot     : SDNode<"AArch64ISD::SDOT", SDT_AArch64Dot>;
741def AArch64udot     : SDNode<"AArch64ISD::UDOT", SDT_AArch64Dot>;
742
743def AArch64saddv    : SDNode<"AArch64ISD::SADDV", SDT_AArch64UnaryVec>;
744def AArch64uaddv    : SDNode<"AArch64ISD::UADDV", SDT_AArch64UnaryVec>;
745def AArch64sminv    : SDNode<"AArch64ISD::SMINV", SDT_AArch64UnaryVec>;
746def AArch64uminv    : SDNode<"AArch64ISD::UMINV", SDT_AArch64UnaryVec>;
747def AArch64smaxv    : SDNode<"AArch64ISD::SMAXV", SDT_AArch64UnaryVec>;
748def AArch64umaxv    : SDNode<"AArch64ISD::UMAXV", SDT_AArch64UnaryVec>;
749
750def AArch64uabd     : PatFrags<(ops node:$lhs, node:$rhs),
751                               [(abdu node:$lhs, node:$rhs),
752                                (int_aarch64_neon_uabd node:$lhs, node:$rhs)]>;
753def AArch64sabd     : PatFrags<(ops node:$lhs, node:$rhs),
754                               [(abds node:$lhs, node:$rhs),
755                                (int_aarch64_neon_sabd node:$lhs, node:$rhs)]>;
756
757def AArch64addp_n   : SDNode<"AArch64ISD::ADDP", SDT_AArch64Zip>;
758def AArch64uaddlp_n : SDNode<"AArch64ISD::UADDLP", SDT_AArch64uaddlp>;
759def AArch64saddlp_n : SDNode<"AArch64ISD::SADDLP", SDT_AArch64uaddlp>;
760def AArch64addp     : PatFrags<(ops node:$Rn, node:$Rm),
761                               [(AArch64addp_n node:$Rn, node:$Rm),
762                                (int_aarch64_neon_addp node:$Rn, node:$Rm)]>;
763def AArch64uaddlp   : PatFrags<(ops node:$src),
764                               [(AArch64uaddlp_n node:$src),
765                                (int_aarch64_neon_uaddlp node:$src)]>;
766def AArch64saddlp   : PatFrags<(ops node:$src),
767                               [(AArch64saddlp_n node:$src),
768                                (int_aarch64_neon_saddlp node:$src)]>;
769def AArch64faddp     : PatFrags<(ops node:$Rn, node:$Rm),
770                                [(AArch64addp_n node:$Rn, node:$Rm),
771                                 (int_aarch64_neon_faddp node:$Rn, node:$Rm)]>;
772def AArch64roundingvlshr : ComplexPattern<vAny, 2, "SelectRoundingVLShr", [AArch64vlshr]>;
773
774def SDT_AArch64SETTAG : SDTypeProfile<0, 2, [SDTCisPtrTy<0>, SDTCisPtrTy<1>]>;
775def AArch64stg : SDNode<"AArch64ISD::STG", SDT_AArch64SETTAG, [SDNPHasChain, SDNPMayStore, SDNPMemOperand]>;
776def AArch64stzg : SDNode<"AArch64ISD::STZG", SDT_AArch64SETTAG, [SDNPHasChain, SDNPMayStore, SDNPMemOperand]>;
777def AArch64st2g : SDNode<"AArch64ISD::ST2G", SDT_AArch64SETTAG, [SDNPHasChain, SDNPMayStore, SDNPMemOperand]>;
778def AArch64stz2g : SDNode<"AArch64ISD::STZ2G", SDT_AArch64SETTAG, [SDNPHasChain, SDNPMayStore, SDNPMemOperand]>;
779
780def SDT_AArch64unpk : SDTypeProfile<1, 1, [
781    SDTCisInt<0>, SDTCisInt<1>, SDTCisOpSmallerThanOp<1, 0>
782]>;
783def AArch64sunpkhi : SDNode<"AArch64ISD::SUNPKHI", SDT_AArch64unpk>;
784def AArch64sunpklo : SDNode<"AArch64ISD::SUNPKLO", SDT_AArch64unpk>;
785def AArch64uunpkhi : SDNode<"AArch64ISD::UUNPKHI", SDT_AArch64unpk>;
786def AArch64uunpklo : SDNode<"AArch64ISD::UUNPKLO", SDT_AArch64unpk>;
787
788def AArch64ldp : SDNode<"AArch64ISD::LDP", SDT_AArch64ldp, [SDNPHasChain, SDNPMayLoad, SDNPMemOperand]>;
789def AArch64ldnp : SDNode<"AArch64ISD::LDNP", SDT_AArch64ldnp, [SDNPHasChain, SDNPMayLoad, SDNPMemOperand]>;
790def AArch64stp : SDNode<"AArch64ISD::STP", SDT_AArch64stp, [SDNPHasChain, SDNPMayStore, SDNPMemOperand]>;
791def AArch64stnp : SDNode<"AArch64ISD::STNP", SDT_AArch64stnp, [SDNPHasChain, SDNPMayStore, SDNPMemOperand]>;
792
793def AArch64tbl : SDNode<"AArch64ISD::TBL", SDT_AArch64TBL>;
794def AArch64mrs : SDNode<"AArch64ISD::MRS",
795                        SDTypeProfile<1, 1, [SDTCisVT<0, i64>, SDTCisVT<1, i32>]>,
796                        [SDNPHasChain, SDNPOutGlue]>;
797
798// Match add node and also treat an 'or' node is as an 'add' if the or'ed operands
799// have no common bits.
800def add_and_or_is_add : PatFrags<(ops node:$lhs, node:$rhs),
801                         [(add node:$lhs, node:$rhs), (or node:$lhs, node:$rhs)],[{
802   if (N->getOpcode() == ISD::ADD)
803     return true;
804   return CurDAG->haveNoCommonBitsSet(N->getOperand(0), N->getOperand(1));
805}]> {
806  let GISelPredicateCode = [{
807     // Only handle G_ADD for now. FIXME. build capability to compute whether
808     // operands of G_OR have common bits set or not.
809     return MI.getOpcode() == TargetOpcode::G_ADD;
810  }];
811}
812
813// Match mul with enough sign-bits. Can be reduced to a smaller mul operand.
814def smullwithsignbits : PatFrag<(ops node:$l, node:$r), (mul node:$l, node:$r), [{
815  return CurDAG->ComputeNumSignBits(N->getOperand(0)) > 32 &&
816         CurDAG->ComputeNumSignBits(N->getOperand(1)) > 32;
817}]>;
818
819//===----------------------------------------------------------------------===//
820
821//===----------------------------------------------------------------------===//
822
823// AArch64 Instruction Predicate Definitions.
824// We could compute these on a per-module basis but doing so requires accessing
825// the Function object through the <Target>Subtarget and objections were raised
826// to that (see post-commit review comments for r301750).
827let RecomputePerFunction = 1 in {
828  def ForCodeSize   : Predicate<"shouldOptForSize(MF)">;
829  def NotForCodeSize   : Predicate<"!shouldOptForSize(MF)">;
830  // Avoid generating STRQro if it is slow, unless we're optimizing for code size.
831  def UseSTRQro : Predicate<"!Subtarget->isSTRQroSlow() || shouldOptForSize(MF)">;
832
833  def UseBTI : Predicate<[{ MF->getInfo<AArch64FunctionInfo>()->branchTargetEnforcement() }]>;
834  def NotUseBTI : Predicate<[{ !MF->getInfo<AArch64FunctionInfo>()->branchTargetEnforcement() }]>;
835
836  def SLSBLRMitigation : Predicate<[{ MF->getSubtarget<AArch64Subtarget>().hardenSlsBlr() }]>;
837  def NoSLSBLRMitigation : Predicate<[{ !MF->getSubtarget<AArch64Subtarget>().hardenSlsBlr() }]>;
838  // Toggles patterns which aren't beneficial in GlobalISel when we aren't
839  // optimizing. This allows us to selectively use patterns without impacting
840  // SelectionDAG's behaviour.
841  // FIXME: One day there will probably be a nicer way to check for this, but
842  // today is not that day.
843  def OptimizedGISelOrOtherSelector : Predicate<"!MF->getFunction().hasOptNone() || MF->getProperties().hasProperty(MachineFunctionProperties::Property::FailedISel) || !MF->getProperties().hasProperty(MachineFunctionProperties::Property::Legalized)">;
844}
845
846include "AArch64InstrFormats.td"
847include "SVEInstrFormats.td"
848include "SMEInstrFormats.td"
849
850//===----------------------------------------------------------------------===//
851
852//===----------------------------------------------------------------------===//
853// Miscellaneous instructions.
854//===----------------------------------------------------------------------===//
855
856let Defs = [SP], Uses = [SP], hasSideEffects = 1, isCodeGenOnly = 1 in {
857// We set Sched to empty list because we expect these instructions to simply get
858// removed in most cases.
859def ADJCALLSTACKDOWN : Pseudo<(outs), (ins i32imm:$amt1, i32imm:$amt2),
860                              [(AArch64callseq_start timm:$amt1, timm:$amt2)]>,
861                              Sched<[]>;
862def ADJCALLSTACKUP : Pseudo<(outs), (ins i32imm:$amt1, i32imm:$amt2),
863                            [(AArch64callseq_end timm:$amt1, timm:$amt2)]>,
864                            Sched<[]>;
865} // Defs = [SP], Uses = [SP], hasSideEffects = 1, isCodeGenOnly = 1
866
867let isReMaterializable = 1, isCodeGenOnly = 1 in {
868// FIXME: The following pseudo instructions are only needed because remat
869// cannot handle multiple instructions.  When that changes, they can be
870// removed, along with the AArch64Wrapper node.
871
872let AddedComplexity = 10 in
873def LOADgot : Pseudo<(outs GPR64common:$dst), (ins i64imm:$addr),
874                     [(set GPR64common:$dst, (AArch64LOADgot tglobaladdr:$addr))]>,
875              Sched<[WriteLDAdr]>;
876
877// The MOVaddr instruction should match only when the add is not folded
878// into a load or store address.
879def MOVaddr
880    : Pseudo<(outs GPR64common:$dst), (ins i64imm:$hi, i64imm:$low),
881             [(set GPR64common:$dst, (AArch64addlow (AArch64adrp tglobaladdr:$hi),
882                                            tglobaladdr:$low))]>,
883      Sched<[WriteAdrAdr]>;
884def MOVaddrJT
885    : Pseudo<(outs GPR64common:$dst), (ins i64imm:$hi, i64imm:$low),
886             [(set GPR64common:$dst, (AArch64addlow (AArch64adrp tjumptable:$hi),
887                                             tjumptable:$low))]>,
888      Sched<[WriteAdrAdr]>;
889def MOVaddrCP
890    : Pseudo<(outs GPR64common:$dst), (ins i64imm:$hi, i64imm:$low),
891             [(set GPR64common:$dst, (AArch64addlow (AArch64adrp tconstpool:$hi),
892                                             tconstpool:$low))]>,
893      Sched<[WriteAdrAdr]>;
894def MOVaddrBA
895    : Pseudo<(outs GPR64common:$dst), (ins i64imm:$hi, i64imm:$low),
896             [(set GPR64common:$dst, (AArch64addlow (AArch64adrp tblockaddress:$hi),
897                                             tblockaddress:$low))]>,
898      Sched<[WriteAdrAdr]>;
899def MOVaddrTLS
900    : Pseudo<(outs GPR64common:$dst), (ins i64imm:$hi, i64imm:$low),
901             [(set GPR64common:$dst, (AArch64addlow (AArch64adrp tglobaltlsaddr:$hi),
902                                            tglobaltlsaddr:$low))]>,
903      Sched<[WriteAdrAdr]>;
904def MOVaddrEXT
905    : Pseudo<(outs GPR64common:$dst), (ins i64imm:$hi, i64imm:$low),
906             [(set GPR64common:$dst, (AArch64addlow (AArch64adrp texternalsym:$hi),
907                                            texternalsym:$low))]>,
908      Sched<[WriteAdrAdr]>;
909// Normally AArch64addlow either gets folded into a following ldr/str,
910// or together with an adrp into MOVaddr above. For cases with TLS, it
911// might appear without either of them, so allow lowering it into a plain
912// add.
913def ADDlowTLS
914    : Pseudo<(outs GPR64sp:$dst), (ins GPR64sp:$src, i64imm:$low),
915             [(set GPR64sp:$dst, (AArch64addlow GPR64sp:$src,
916                                            tglobaltlsaddr:$low))]>,
917      Sched<[WriteAdr]>;
918
919} // isReMaterializable, isCodeGenOnly
920
921def : Pat<(AArch64LOADgot tglobaltlsaddr:$addr),
922          (LOADgot tglobaltlsaddr:$addr)>;
923
924def : Pat<(AArch64LOADgot texternalsym:$addr),
925          (LOADgot texternalsym:$addr)>;
926
927def : Pat<(AArch64LOADgot tconstpool:$addr),
928          (LOADgot tconstpool:$addr)>;
929
930// In general these get lowered into a sequence of three 4-byte instructions.
931// 32-bit jump table destination is actually only 2 instructions since we can
932// use the table itself as a PC-relative base. But optimization occurs after
933// branch relaxation so be pessimistic.
934let Size = 12, Constraints = "@earlyclobber $dst,@earlyclobber $scratch",
935    isNotDuplicable = 1 in {
936def JumpTableDest32 : Pseudo<(outs GPR64:$dst, GPR64sp:$scratch),
937                             (ins GPR64:$table, GPR64:$entry, i32imm:$jti), []>,
938                      Sched<[]>;
939def JumpTableDest16 : Pseudo<(outs GPR64:$dst, GPR64sp:$scratch),
940                             (ins GPR64:$table, GPR64:$entry, i32imm:$jti), []>,
941                      Sched<[]>;
942def JumpTableDest8 : Pseudo<(outs GPR64:$dst, GPR64sp:$scratch),
943                            (ins GPR64:$table, GPR64:$entry, i32imm:$jti), []>,
944                     Sched<[]>;
945}
946
947// Space-consuming pseudo to aid testing of placement and reachability
948// algorithms. Immediate operand is the number of bytes this "instruction"
949// occupies; register operands can be used to enforce dependency and constrain
950// the scheduler.
951let hasSideEffects = 1, mayLoad = 1, mayStore = 1 in
952def SPACE : Pseudo<(outs GPR64:$Rd), (ins i32imm:$size, GPR64:$Rn),
953                   [(set GPR64:$Rd, (int_aarch64_space imm:$size, GPR64:$Rn))]>,
954            Sched<[]>;
955
956let hasSideEffects = 1, isCodeGenOnly = 1 in {
957  def SpeculationSafeValueX
958      : Pseudo<(outs GPR64:$dst), (ins GPR64:$src), []>, Sched<[]>;
959  def SpeculationSafeValueW
960      : Pseudo<(outs GPR32:$dst), (ins GPR32:$src), []>, Sched<[]>;
961}
962
963// SpeculationBarrierEndBB must only be used after an unconditional control
964// flow, i.e. after a terminator for which isBarrier is True.
965let hasSideEffects = 1, isCodeGenOnly = 1, isTerminator = 1, isBarrier = 1 in {
966  // This gets lowered to a pair of 4-byte instructions.
967  let Size = 8 in
968  def SpeculationBarrierISBDSBEndBB
969      : Pseudo<(outs), (ins), []>, Sched<[]>;
970  // This gets lowered to a 4-byte instruction.
971  let Size = 4 in
972  def SpeculationBarrierSBEndBB
973      : Pseudo<(outs), (ins), []>, Sched<[]>;
974}
975
976//===----------------------------------------------------------------------===//
977// System instructions.
978//===----------------------------------------------------------------------===//
979
980def HINT : HintI<"hint">;
981def : InstAlias<"nop",  (HINT 0b000)>;
982def : InstAlias<"yield",(HINT 0b001)>;
983def : InstAlias<"wfe",  (HINT 0b010)>;
984def : InstAlias<"wfi",  (HINT 0b011)>;
985def : InstAlias<"sev",  (HINT 0b100)>;
986def : InstAlias<"sevl", (HINT 0b101)>;
987def : InstAlias<"dgh",  (HINT 0b110)>;
988def : InstAlias<"esb",  (HINT 0b10000)>, Requires<[HasRAS]>;
989def : InstAlias<"csdb", (HINT 20)>;
990// In order to be able to write readable assembly, LLVM should accept assembly
991// inputs that use Branch Target Indentification mnemonics, even with BTI disabled.
992// However, in order to be compatible with other assemblers (e.g. GAS), LLVM
993// should not emit these mnemonics unless BTI is enabled.
994def : InstAlias<"bti",  (HINT 32), 0>;
995def : InstAlias<"bti $op", (HINT btihint_op:$op), 0>;
996def : InstAlias<"bti",  (HINT 32)>, Requires<[HasBTI]>;
997def : InstAlias<"bti $op", (HINT btihint_op:$op)>, Requires<[HasBTI]>;
998
999// v8.2a Statistical Profiling extension
1000def : InstAlias<"psb $op",  (HINT psbhint_op:$op)>, Requires<[HasSPE]>;
1001
1002// As far as LLVM is concerned this writes to the system's exclusive monitors.
1003let mayLoad = 1, mayStore = 1 in
1004def CLREX : CRmSystemI<imm0_15, 0b010, "clrex">;
1005
1006// NOTE: ideally, this would have mayStore = 0, mayLoad = 0, but we cannot
1007// model patterns with sufficiently fine granularity.
1008let mayLoad = ?, mayStore = ? in {
1009def DMB   : CRmSystemI<barrier_op, 0b101, "dmb",
1010                       [(int_aarch64_dmb (i32 imm32_0_15:$CRm))]>;
1011
1012def DSB   : CRmSystemI<barrier_op, 0b100, "dsb",
1013                       [(int_aarch64_dsb (i32 imm32_0_15:$CRm))]>;
1014
1015def ISB   : CRmSystemI<barrier_op, 0b110, "isb",
1016                       [(int_aarch64_isb (i32 imm32_0_15:$CRm))]>;
1017
1018def TSB   : CRmSystemI<barrier_op, 0b010, "tsb", []> {
1019  let CRm        = 0b0010;
1020  let Inst{12}   = 0;
1021  let Predicates = [HasTRACEV8_4];
1022}
1023
1024def DSBnXS  : CRmSystemI<barrier_nxs_op, 0b001, "dsb"> {
1025  let CRm{1-0}   = 0b11;
1026  let Inst{9-8}  = 0b10;
1027  let Predicates = [HasXS];
1028}
1029
1030let Predicates = [HasWFxT] in {
1031def WFET : RegInputSystemI<0b0000, 0b000, "wfet">;
1032def WFIT : RegInputSystemI<0b0000, 0b001, "wfit">;
1033}
1034
1035// Branch Record Buffer two-word mnemonic instructions
1036class BRBEI<bits<3> op2, string keyword>
1037    : SimpleSystemI<0, (ins), "brb", keyword>, Sched<[WriteSys]> {
1038  let Inst{31-8} = 0b110101010000100101110010;
1039  let Inst{7-5} = op2;
1040  let Predicates = [HasBRBE];
1041}
1042def BRB_IALL: BRBEI<0b100, "\tiall">;
1043def BRB_INJ:  BRBEI<0b101, "\tinj">;
1044
1045}
1046
1047// Allow uppercase and lowercase keyword arguments for BRB IALL and BRB INJ
1048def : TokenAlias<"INJ", "inj">;
1049def : TokenAlias<"IALL", "iall">;
1050
1051// ARMv8.2-A Dot Product
1052let Predicates = [HasDotProd] in {
1053defm SDOT : SIMDThreeSameVectorDot<0, 0, "sdot", AArch64sdot>;
1054defm UDOT : SIMDThreeSameVectorDot<1, 0, "udot", AArch64udot>;
1055defm SDOTlane : SIMDThreeSameVectorDotIndex<0, 0, 0b10, "sdot", AArch64sdot>;
1056defm UDOTlane : SIMDThreeSameVectorDotIndex<1, 0, 0b10, "udot", AArch64udot>;
1057}
1058
1059// ARMv8.6-A BFloat
1060let Predicates = [HasNEON, HasBF16] in {
1061defm BFDOT       : SIMDThreeSameVectorBFDot<1, "bfdot">;
1062defm BF16DOTlane : SIMDThreeSameVectorBF16DotI<0, "bfdot">;
1063def BFMMLA       : SIMDThreeSameVectorBF16MatrixMul<"bfmmla">;
1064def BFMLALB      : SIMDBF16MLAL<0, "bfmlalb", int_aarch64_neon_bfmlalb>;
1065def BFMLALT      : SIMDBF16MLAL<1, "bfmlalt", int_aarch64_neon_bfmlalt>;
1066def BFMLALBIdx   : SIMDBF16MLALIndex<0, "bfmlalb", int_aarch64_neon_bfmlalb>;
1067def BFMLALTIdx   : SIMDBF16MLALIndex<1, "bfmlalt", int_aarch64_neon_bfmlalt>;
1068def BFCVTN       : SIMD_BFCVTN;
1069def BFCVTN2      : SIMD_BFCVTN2;
1070
1071// Vector-scalar BFDOT:
1072// The second source operand of the 64-bit variant of BF16DOTlane is a 128-bit
1073// register (the instruction uses a single 32-bit lane from it), so the pattern
1074// is a bit tricky.
1075def : Pat<(v2f32 (int_aarch64_neon_bfdot
1076                    (v2f32 V64:$Rd), (v4bf16 V64:$Rn),
1077                    (v4bf16 (bitconvert
1078                      (v2i32 (AArch64duplane32
1079                        (v4i32 (bitconvert
1080                          (v8bf16 (insert_subvector undef,
1081                            (v4bf16 V64:$Rm),
1082                            (i64 0))))),
1083                        VectorIndexS:$idx)))))),
1084          (BF16DOTlanev4bf16 (v2f32 V64:$Rd), (v4bf16 V64:$Rn),
1085                             (SUBREG_TO_REG (i32 0), V64:$Rm, dsub),
1086                             VectorIndexS:$idx)>;
1087}
1088
1089let Predicates = [HasNEONorSME, HasBF16] in {
1090def BFCVT : BF16ToSinglePrecision<"bfcvt">;
1091}
1092
1093// ARMv8.6A AArch64 matrix multiplication
1094let Predicates = [HasMatMulInt8] in {
1095def  SMMLA : SIMDThreeSameVectorMatMul<0, 0, "smmla", int_aarch64_neon_smmla>;
1096def  UMMLA : SIMDThreeSameVectorMatMul<0, 1, "ummla", int_aarch64_neon_ummla>;
1097def USMMLA : SIMDThreeSameVectorMatMul<1, 0, "usmmla", int_aarch64_neon_usmmla>;
1098defm USDOT : SIMDThreeSameVectorDot<0, 1, "usdot", int_aarch64_neon_usdot>;
1099defm USDOTlane : SIMDThreeSameVectorDotIndex<0, 1, 0b10, "usdot", int_aarch64_neon_usdot>;
1100
1101// sudot lane has a pattern where usdot is expected (there is no sudot).
1102// The second operand is used in the dup operation to repeat the indexed
1103// element.
1104class BaseSIMDSUDOTIndex<bit Q, string dst_kind, string lhs_kind,
1105                         string rhs_kind, RegisterOperand RegType,
1106                         ValueType AccumType, ValueType InputType>
1107      : BaseSIMDThreeSameVectorDotIndex<Q, 0, 1, 0b00, "sudot", dst_kind,
1108                                        lhs_kind, rhs_kind, RegType, AccumType,
1109                                        InputType, null_frag> {
1110  let Pattern = [(set (AccumType RegType:$dst),
1111                      (AccumType (int_aarch64_neon_usdot (AccumType RegType:$Rd),
1112                                 (InputType (bitconvert (AccumType
1113                                    (AArch64duplane32 (v4i32 V128:$Rm),
1114                                        VectorIndexS:$idx)))),
1115                                 (InputType RegType:$Rn))))];
1116}
1117
1118multiclass SIMDSUDOTIndex {
1119  def v8i8  : BaseSIMDSUDOTIndex<0, ".2s", ".8b", ".4b", V64, v2i32, v8i8>;
1120  def v16i8 : BaseSIMDSUDOTIndex<1, ".4s", ".16b", ".4b", V128, v4i32, v16i8>;
1121}
1122
1123defm SUDOTlane : SIMDSUDOTIndex;
1124
1125}
1126
1127// ARMv8.2-A FP16 Fused Multiply-Add Long
1128let Predicates = [HasNEON, HasFP16FML] in {
1129defm FMLAL      : SIMDThreeSameVectorFML<0, 1, 0b001, "fmlal", int_aarch64_neon_fmlal>;
1130defm FMLSL      : SIMDThreeSameVectorFML<0, 1, 0b101, "fmlsl", int_aarch64_neon_fmlsl>;
1131defm FMLAL2     : SIMDThreeSameVectorFML<1, 0, 0b001, "fmlal2", int_aarch64_neon_fmlal2>;
1132defm FMLSL2     : SIMDThreeSameVectorFML<1, 0, 0b101, "fmlsl2", int_aarch64_neon_fmlsl2>;
1133defm FMLALlane  : SIMDThreeSameVectorFMLIndex<0, 0b0000, "fmlal", int_aarch64_neon_fmlal>;
1134defm FMLSLlane  : SIMDThreeSameVectorFMLIndex<0, 0b0100, "fmlsl", int_aarch64_neon_fmlsl>;
1135defm FMLAL2lane : SIMDThreeSameVectorFMLIndex<1, 0b1000, "fmlal2", int_aarch64_neon_fmlal2>;
1136defm FMLSL2lane : SIMDThreeSameVectorFMLIndex<1, 0b1100, "fmlsl2", int_aarch64_neon_fmlsl2>;
1137}
1138
1139// Armv8.2-A Crypto extensions
1140let Predicates = [HasSHA3] in {
1141def SHA512H   : CryptoRRRTied<0b0, 0b00, "sha512h">;
1142def SHA512H2  : CryptoRRRTied<0b0, 0b01, "sha512h2">;
1143def SHA512SU0 : CryptoRRTied_2D<0b0, 0b00, "sha512su0">;
1144def SHA512SU1 : CryptoRRRTied_2D<0b0, 0b10, "sha512su1">;
1145def RAX1      : CryptoRRR_2D<0b0,0b11, "rax1">;
1146def EOR3      : CryptoRRRR_16B<0b00, "eor3">;
1147def BCAX      : CryptoRRRR_16B<0b01, "bcax">;
1148def XAR       : CryptoRRRi6<"xar">;
1149
1150class SHA3_pattern<Instruction INST, Intrinsic OpNode, ValueType VecTy>
1151  : Pat<(VecTy (OpNode (VecTy V128:$Vd), (VecTy V128:$Vn), (VecTy V128:$Vm))),
1152        (INST (VecTy V128:$Vd), (VecTy V128:$Vn), (VecTy V128:$Vm))>;
1153
1154def : Pat<(v2i64 (int_aarch64_crypto_sha512su0 (v2i64 V128:$Vn), (v2i64 V128:$Vm))),
1155          (SHA512SU0 (v2i64 V128:$Vn), (v2i64 V128:$Vm))>;
1156
1157def : SHA3_pattern<SHA512H, int_aarch64_crypto_sha512h, v2i64>;
1158def : SHA3_pattern<SHA512H2, int_aarch64_crypto_sha512h2, v2i64>;
1159def : SHA3_pattern<SHA512SU1, int_aarch64_crypto_sha512su1, v2i64>;
1160
1161def : SHA3_pattern<EOR3, int_aarch64_crypto_eor3u, v16i8>;
1162def : SHA3_pattern<EOR3, int_aarch64_crypto_eor3u, v8i16>;
1163def : SHA3_pattern<EOR3, int_aarch64_crypto_eor3u, v4i32>;
1164def : SHA3_pattern<EOR3, int_aarch64_crypto_eor3u, v2i64>;
1165
1166class EOR3_pattern<ValueType VecTy>
1167  : Pat<(xor (xor (VecTy V128:$Vn), (VecTy V128:$Vm)), (VecTy V128:$Va)),
1168        (EOR3 (VecTy V128:$Vn), (VecTy V128:$Vm), (VecTy V128:$Va))>;
1169
1170def : EOR3_pattern<v16i8>;
1171def : EOR3_pattern<v8i16>;
1172def : EOR3_pattern<v4i32>;
1173def : EOR3_pattern<v2i64>;
1174
1175class BCAX_pattern<ValueType VecTy>
1176  : Pat<(xor (VecTy V128:$Vn), (and (VecTy V128:$Vm), (vnot (VecTy V128:$Va)))),
1177        (BCAX (VecTy V128:$Vn), (VecTy V128:$Vm), (VecTy V128:$Va))>;
1178
1179def : BCAX_pattern<v16i8>;
1180def : BCAX_pattern<v8i16>;
1181def : BCAX_pattern<v4i32>;
1182def : BCAX_pattern<v2i64>;
1183
1184def : SHA3_pattern<BCAX, int_aarch64_crypto_bcaxu, v16i8>;
1185def : SHA3_pattern<BCAX, int_aarch64_crypto_bcaxu, v8i16>;
1186def : SHA3_pattern<BCAX, int_aarch64_crypto_bcaxu, v4i32>;
1187def : SHA3_pattern<BCAX, int_aarch64_crypto_bcaxu, v2i64>;
1188
1189def : SHA3_pattern<EOR3, int_aarch64_crypto_eor3s, v16i8>;
1190def : SHA3_pattern<EOR3, int_aarch64_crypto_eor3s, v8i16>;
1191def : SHA3_pattern<EOR3, int_aarch64_crypto_eor3s, v4i32>;
1192def : SHA3_pattern<EOR3, int_aarch64_crypto_eor3s, v2i64>;
1193
1194def : SHA3_pattern<BCAX, int_aarch64_crypto_bcaxs, v16i8>;
1195def : SHA3_pattern<BCAX, int_aarch64_crypto_bcaxs, v8i16>;
1196def : SHA3_pattern<BCAX, int_aarch64_crypto_bcaxs, v4i32>;
1197def : SHA3_pattern<BCAX, int_aarch64_crypto_bcaxs, v2i64>;
1198
1199def : Pat<(v2i64 (int_aarch64_crypto_rax1 (v2i64 V128:$Vn), (v2i64 V128:$Vm))),
1200          (RAX1 (v2i64 V128:$Vn), (v2i64 V128:$Vm))>;
1201
1202def : Pat<(v2i64 (int_aarch64_crypto_xar (v2i64 V128:$Vn), (v2i64 V128:$Vm), (i64 timm0_63:$imm))),
1203          (XAR (v2i64 V128:$Vn), (v2i64 V128:$Vm), (timm0_63:$imm))>;
1204
1205
1206} // HasSHA3
1207
1208let Predicates = [HasSM4] in {
1209def SM3TT1A   : CryptoRRRi2Tied<0b0, 0b00, "sm3tt1a">;
1210def SM3TT1B   : CryptoRRRi2Tied<0b0, 0b01, "sm3tt1b">;
1211def SM3TT2A   : CryptoRRRi2Tied<0b0, 0b10, "sm3tt2a">;
1212def SM3TT2B   : CryptoRRRi2Tied<0b0, 0b11, "sm3tt2b">;
1213def SM3SS1    : CryptoRRRR_4S<0b10, "sm3ss1">;
1214def SM3PARTW1 : CryptoRRRTied_4S<0b1, 0b00, "sm3partw1">;
1215def SM3PARTW2 : CryptoRRRTied_4S<0b1, 0b01, "sm3partw2">;
1216def SM4ENCKEY : CryptoRRR_4S<0b1, 0b10, "sm4ekey">;
1217def SM4E      : CryptoRRTied_4S<0b0, 0b01, "sm4e">;
1218
1219def : Pat<(v4i32 (int_aarch64_crypto_sm3ss1 (v4i32 V128:$Vn), (v4i32 V128:$Vm), (v4i32 V128:$Va))),
1220          (SM3SS1 (v4i32 V128:$Vn), (v4i32 V128:$Vm), (v4i32 V128:$Va))>;
1221
1222class SM3PARTW_pattern<Instruction INST, Intrinsic OpNode>
1223  : Pat<(v4i32 (OpNode (v4i32 V128:$Vd), (v4i32 V128:$Vn), (v4i32 V128:$Vm))),
1224        (INST (v4i32 V128:$Vd), (v4i32 V128:$Vn), (v4i32 V128:$Vm))>;
1225
1226class SM3TT_pattern<Instruction INST, Intrinsic OpNode>
1227  : Pat<(v4i32 (OpNode (v4i32 V128:$Vd), (v4i32 V128:$Vn), (v4i32 V128:$Vm), (i64 VectorIndexS_timm:$imm) )),
1228        (INST (v4i32 V128:$Vd), (v4i32 V128:$Vn), (v4i32 V128:$Vm), (VectorIndexS_timm:$imm))>;
1229
1230class SM4_pattern<Instruction INST, Intrinsic OpNode>
1231  : Pat<(v4i32 (OpNode (v4i32 V128:$Vn), (v4i32 V128:$Vm))),
1232        (INST (v4i32 V128:$Vn), (v4i32 V128:$Vm))>;
1233
1234def : SM3PARTW_pattern<SM3PARTW1, int_aarch64_crypto_sm3partw1>;
1235def : SM3PARTW_pattern<SM3PARTW2, int_aarch64_crypto_sm3partw2>;
1236
1237def : SM3TT_pattern<SM3TT1A, int_aarch64_crypto_sm3tt1a>;
1238def : SM3TT_pattern<SM3TT1B, int_aarch64_crypto_sm3tt1b>;
1239def : SM3TT_pattern<SM3TT2A, int_aarch64_crypto_sm3tt2a>;
1240def : SM3TT_pattern<SM3TT2B, int_aarch64_crypto_sm3tt2b>;
1241
1242def : SM4_pattern<SM4ENCKEY, int_aarch64_crypto_sm4ekey>;
1243def : SM4_pattern<SM4E, int_aarch64_crypto_sm4e>;
1244} // HasSM4
1245
1246let Predicates = [HasRCPC] in {
1247  // v8.3 Release Consistent Processor Consistent support, optional in v8.2.
1248  def LDAPRB  : RCPCLoad<0b00, "ldaprb", GPR32>;
1249  def LDAPRH  : RCPCLoad<0b01, "ldaprh", GPR32>;
1250  def LDAPRW  : RCPCLoad<0b10, "ldapr", GPR32>;
1251  def LDAPRX  : RCPCLoad<0b11, "ldapr", GPR64>;
1252}
1253
1254// v8.3a complex add and multiply-accumulate. No predicate here, that is done
1255// inside the multiclass as the FP16 versions need different predicates.
1256defm FCMLA : SIMDThreeSameVectorTiedComplexHSD<1, 0b110, complexrotateop,
1257                                               "fcmla", null_frag>;
1258defm FCADD : SIMDThreeSameVectorComplexHSD<1, 0b111, complexrotateopodd,
1259                                           "fcadd", null_frag>;
1260defm FCMLA : SIMDIndexedTiedComplexHSD<0, 1, complexrotateop, "fcmla">;
1261
1262let Predicates = [HasComplxNum, HasNEON, HasFullFP16] in {
1263  def : Pat<(v4f16 (int_aarch64_neon_vcadd_rot90 (v4f16 V64:$Rn), (v4f16 V64:$Rm))),
1264            (FCADDv4f16 (v4f16 V64:$Rn), (v4f16 V64:$Rm), (i32 0))>;
1265  def : Pat<(v4f16 (int_aarch64_neon_vcadd_rot270 (v4f16 V64:$Rn), (v4f16 V64:$Rm))),
1266            (FCADDv4f16 (v4f16 V64:$Rn), (v4f16 V64:$Rm), (i32 1))>;
1267  def : Pat<(v8f16 (int_aarch64_neon_vcadd_rot90 (v8f16 V128:$Rn), (v8f16 V128:$Rm))),
1268            (FCADDv8f16 (v8f16 V128:$Rn), (v8f16 V128:$Rm), (i32 0))>;
1269  def : Pat<(v8f16 (int_aarch64_neon_vcadd_rot270 (v8f16 V128:$Rn), (v8f16 V128:$Rm))),
1270            (FCADDv8f16 (v8f16 V128:$Rn), (v8f16 V128:$Rm), (i32 1))>;
1271}
1272
1273let Predicates = [HasComplxNum, HasNEON] in {
1274  def : Pat<(v2f32 (int_aarch64_neon_vcadd_rot90 (v2f32 V64:$Rn), (v2f32 V64:$Rm))),
1275            (FCADDv2f32 (v2f32 V64:$Rn), (v2f32 V64:$Rm), (i32 0))>;
1276  def : Pat<(v2f32 (int_aarch64_neon_vcadd_rot270 (v2f32 V64:$Rn), (v2f32 V64:$Rm))),
1277            (FCADDv2f32 (v2f32 V64:$Rn), (v2f32 V64:$Rm), (i32 1))>;
1278  foreach Ty = [v4f32, v2f64] in {
1279    def : Pat<(Ty (int_aarch64_neon_vcadd_rot90 (Ty V128:$Rn), (Ty V128:$Rm))),
1280              (!cast<Instruction>("FCADD"#Ty) (Ty V128:$Rn), (Ty V128:$Rm), (i32 0))>;
1281    def : Pat<(Ty (int_aarch64_neon_vcadd_rot270 (Ty V128:$Rn), (Ty V128:$Rm))),
1282              (!cast<Instruction>("FCADD"#Ty) (Ty V128:$Rn), (Ty V128:$Rm), (i32 1))>;
1283  }
1284}
1285
1286multiclass FCMLA_PATS<ValueType ty, DAGOperand Reg> {
1287  def : Pat<(ty (int_aarch64_neon_vcmla_rot0 (ty Reg:$Rd), (ty Reg:$Rn), (ty Reg:$Rm))),
1288            (!cast<Instruction>("FCMLA" # ty) $Rd, $Rn, $Rm, 0)>;
1289  def : Pat<(ty (int_aarch64_neon_vcmla_rot90 (ty Reg:$Rd), (ty Reg:$Rn), (ty Reg:$Rm))),
1290            (!cast<Instruction>("FCMLA" # ty) $Rd, $Rn, $Rm, 1)>;
1291  def : Pat<(ty (int_aarch64_neon_vcmla_rot180 (ty Reg:$Rd), (ty Reg:$Rn), (ty Reg:$Rm))),
1292            (!cast<Instruction>("FCMLA" # ty) $Rd, $Rn, $Rm, 2)>;
1293  def : Pat<(ty (int_aarch64_neon_vcmla_rot270 (ty Reg:$Rd), (ty Reg:$Rn), (ty Reg:$Rm))),
1294            (!cast<Instruction>("FCMLA" # ty) $Rd, $Rn, $Rm, 3)>;
1295}
1296
1297multiclass FCMLA_LANE_PATS<ValueType ty, DAGOperand Reg, dag RHSDup> {
1298  def : Pat<(ty (int_aarch64_neon_vcmla_rot0 (ty Reg:$Rd), (ty Reg:$Rn), RHSDup)),
1299            (!cast<Instruction>("FCMLA" # ty # "_indexed") $Rd, $Rn, $Rm, VectorIndexS:$idx, 0)>;
1300  def : Pat<(ty (int_aarch64_neon_vcmla_rot90 (ty Reg:$Rd), (ty Reg:$Rn), RHSDup)),
1301            (!cast<Instruction>("FCMLA" # ty # "_indexed") $Rd, $Rn, $Rm, VectorIndexS:$idx, 1)>;
1302  def : Pat<(ty (int_aarch64_neon_vcmla_rot180 (ty Reg:$Rd), (ty Reg:$Rn), RHSDup)),
1303            (!cast<Instruction>("FCMLA" # ty # "_indexed") $Rd, $Rn, $Rm, VectorIndexS:$idx, 2)>;
1304  def : Pat<(ty (int_aarch64_neon_vcmla_rot270 (ty Reg:$Rd), (ty Reg:$Rn), RHSDup)),
1305            (!cast<Instruction>("FCMLA" # ty # "_indexed") $Rd, $Rn, $Rm, VectorIndexS:$idx, 3)>;
1306}
1307
1308
1309let Predicates = [HasComplxNum, HasNEON, HasFullFP16] in {
1310  defm : FCMLA_PATS<v4f16, V64>;
1311  defm : FCMLA_PATS<v8f16, V128>;
1312
1313  defm : FCMLA_LANE_PATS<v4f16, V64,
1314                         (v4f16 (bitconvert (v2i32 (AArch64duplane32 (v4i32 V128:$Rm), VectorIndexD:$idx))))>;
1315  defm : FCMLA_LANE_PATS<v8f16, V128,
1316                         (v8f16 (bitconvert (v4i32 (AArch64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))>;
1317}
1318let Predicates = [HasComplxNum, HasNEON] in {
1319  defm : FCMLA_PATS<v2f32, V64>;
1320  defm : FCMLA_PATS<v4f32, V128>;
1321  defm : FCMLA_PATS<v2f64, V128>;
1322
1323  defm : FCMLA_LANE_PATS<v4f32, V128,
1324                         (v4f32 (bitconvert (v2i64 (AArch64duplane64 (v2i64 V128:$Rm), VectorIndexD:$idx))))>;
1325}
1326
1327// v8.3a Pointer Authentication
1328// These instructions inhabit part of the hint space and so can be used for
1329// armv8 targets. Keeping the old HINT mnemonic when compiling without PA is
1330// important for compatibility with other assemblers (e.g. GAS) when building
1331// software compatible with both CPUs that do or don't implement PA.
1332let Uses = [LR], Defs = [LR] in {
1333  def PACIAZ   : SystemNoOperands<0b000, "hint\t#24">;
1334  def PACIBZ   : SystemNoOperands<0b010, "hint\t#26">;
1335  let isAuthenticated = 1 in {
1336    def AUTIAZ   : SystemNoOperands<0b100, "hint\t#28">;
1337    def AUTIBZ   : SystemNoOperands<0b110, "hint\t#30">;
1338  }
1339}
1340let Uses = [LR, SP], Defs = [LR] in {
1341  def PACIASP  : SystemNoOperands<0b001, "hint\t#25">;
1342  def PACIBSP  : SystemNoOperands<0b011, "hint\t#27">;
1343  let isAuthenticated = 1 in {
1344    def AUTIASP  : SystemNoOperands<0b101, "hint\t#29">;
1345    def AUTIBSP  : SystemNoOperands<0b111, "hint\t#31">;
1346  }
1347}
1348let Uses = [X16, X17], Defs = [X17], CRm = 0b0001 in {
1349  def PACIA1716  : SystemNoOperands<0b000, "hint\t#8">;
1350  def PACIB1716  : SystemNoOperands<0b010, "hint\t#10">;
1351  let isAuthenticated = 1 in {
1352    def AUTIA1716  : SystemNoOperands<0b100, "hint\t#12">;
1353    def AUTIB1716  : SystemNoOperands<0b110, "hint\t#14">;
1354  }
1355}
1356
1357let Uses = [LR], Defs = [LR], CRm = 0b0000 in {
1358  def XPACLRI   : SystemNoOperands<0b111, "hint\t#7">;
1359}
1360
1361// In order to be able to write readable assembly, LLVM should accept assembly
1362// inputs that use pointer authentication mnemonics, even with PA disabled.
1363// However, in order to be compatible with other assemblers (e.g. GAS), LLVM
1364// should not emit these mnemonics unless PA is enabled.
1365def : InstAlias<"paciaz", (PACIAZ), 0>;
1366def : InstAlias<"pacibz", (PACIBZ), 0>;
1367def : InstAlias<"autiaz", (AUTIAZ), 0>;
1368def : InstAlias<"autibz", (AUTIBZ), 0>;
1369def : InstAlias<"paciasp", (PACIASP), 0>;
1370def : InstAlias<"pacibsp", (PACIBSP), 0>;
1371def : InstAlias<"autiasp", (AUTIASP), 0>;
1372def : InstAlias<"autibsp", (AUTIBSP), 0>;
1373def : InstAlias<"pacia1716", (PACIA1716), 0>;
1374def : InstAlias<"pacib1716", (PACIB1716), 0>;
1375def : InstAlias<"autia1716", (AUTIA1716), 0>;
1376def : InstAlias<"autib1716", (AUTIB1716), 0>;
1377def : InstAlias<"xpaclri", (XPACLRI), 0>;
1378
1379// These pointer authentication instructions require armv8.3a
1380let Predicates = [HasPAuth] in {
1381
1382  // When PA is enabled, a better mnemonic should be emitted.
1383  def : InstAlias<"paciaz", (PACIAZ), 1>;
1384  def : InstAlias<"pacibz", (PACIBZ), 1>;
1385  def : InstAlias<"autiaz", (AUTIAZ), 1>;
1386  def : InstAlias<"autibz", (AUTIBZ), 1>;
1387  def : InstAlias<"paciasp", (PACIASP), 1>;
1388  def : InstAlias<"pacibsp", (PACIBSP), 1>;
1389  def : InstAlias<"autiasp", (AUTIASP), 1>;
1390  def : InstAlias<"autibsp", (AUTIBSP), 1>;
1391  def : InstAlias<"pacia1716", (PACIA1716), 1>;
1392  def : InstAlias<"pacib1716", (PACIB1716), 1>;
1393  def : InstAlias<"autia1716", (AUTIA1716), 1>;
1394  def : InstAlias<"autib1716", (AUTIB1716), 1>;
1395  def : InstAlias<"xpaclri", (XPACLRI), 1>;
1396
1397  multiclass SignAuth<bits<3> prefix, bits<3> prefix_z, string asm,
1398                      SDPatternOperator op> {
1399    def IA   : SignAuthOneData<prefix, 0b00, !strconcat(asm,  "ia"), op>;
1400    def IB   : SignAuthOneData<prefix, 0b01, !strconcat(asm,  "ib"), op>;
1401    def DA   : SignAuthOneData<prefix, 0b10, !strconcat(asm,  "da"), op>;
1402    def DB   : SignAuthOneData<prefix, 0b11, !strconcat(asm,  "db"), op>;
1403    def IZA  : SignAuthZero<prefix_z,  0b00, !strconcat(asm, "iza"), op>;
1404    def DZA  : SignAuthZero<prefix_z,  0b10, !strconcat(asm, "dza"), op>;
1405    def IZB  : SignAuthZero<prefix_z,  0b01, !strconcat(asm, "izb"), op>;
1406    def DZB  : SignAuthZero<prefix_z,  0b11, !strconcat(asm, "dzb"), op>;
1407  }
1408
1409  defm PAC : SignAuth<0b000, 0b010, "pac", int_ptrauth_sign>;
1410  defm AUT : SignAuth<0b001, 0b011, "aut", null_frag>;
1411
1412  def XPACI : ClearAuth<0, "xpaci">;
1413  def : Pat<(int_ptrauth_strip GPR64:$Rd, 0), (XPACI GPR64:$Rd)>;
1414  def : Pat<(int_ptrauth_strip GPR64:$Rd, 1), (XPACI GPR64:$Rd)>;
1415
1416  def XPACD : ClearAuth<1, "xpacd">;
1417  def : Pat<(int_ptrauth_strip GPR64:$Rd, 2), (XPACD GPR64:$Rd)>;
1418  def : Pat<(int_ptrauth_strip GPR64:$Rd, 3), (XPACD GPR64:$Rd)>;
1419
1420  def PACGA : SignAuthTwoOperand<0b1100, "pacga", int_ptrauth_sign_generic>;
1421
1422  // Combined Instructions
1423  let isBranch = 1, isTerminator = 1, isBarrier = 1, isIndirectBranch = 1  in {
1424    def BRAA    : AuthBranchTwoOperands<0, 0, "braa">;
1425    def BRAB    : AuthBranchTwoOperands<0, 1, "brab">;
1426  }
1427  let isCall = 1, Defs = [LR], Uses = [SP] in {
1428    def BLRAA   : AuthBranchTwoOperands<1, 0, "blraa">;
1429    def BLRAB   : AuthBranchTwoOperands<1, 1, "blrab">;
1430  }
1431
1432  let isBranch = 1, isTerminator = 1, isBarrier = 1, isIndirectBranch = 1  in {
1433    def BRAAZ   : AuthOneOperand<0b000, 0, "braaz">;
1434    def BRABZ   : AuthOneOperand<0b000, 1, "brabz">;
1435  }
1436  let isCall = 1, Defs = [LR], Uses = [SP] in {
1437    def BLRAAZ  : AuthOneOperand<0b001, 0, "blraaz">;
1438    def BLRABZ  : AuthOneOperand<0b001, 1, "blrabz">;
1439  }
1440
1441  let isReturn = 1, isTerminator = 1, isBarrier = 1 in {
1442    def RETAA   : AuthReturn<0b010, 0, "retaa">;
1443    def RETAB   : AuthReturn<0b010, 1, "retab">;
1444    def ERETAA  : AuthReturn<0b100, 0, "eretaa">;
1445    def ERETAB  : AuthReturn<0b100, 1, "eretab">;
1446  }
1447
1448  defm LDRAA  : AuthLoad<0, "ldraa", simm10Scaled>;
1449  defm LDRAB  : AuthLoad<1, "ldrab", simm10Scaled>;
1450
1451}
1452
1453// v8.3a floating point conversion for javascript
1454let Predicates = [HasJS, HasFPARMv8], Defs = [NZCV] in
1455def FJCVTZS  : BaseFPToIntegerUnscaled<0b01, 0b11, 0b110, FPR64, GPR32,
1456                                      "fjcvtzs",
1457                                      [(set GPR32:$Rd,
1458                                         (int_aarch64_fjcvtzs FPR64:$Rn))]> {
1459  let Inst{31} = 0;
1460} // HasJS, HasFPARMv8
1461
1462// v8.4 Flag manipulation instructions
1463let Predicates = [HasFlagM], Defs = [NZCV], Uses = [NZCV] in {
1464def CFINV : SimpleSystemI<0, (ins), "cfinv", "">, Sched<[WriteSys]> {
1465  let Inst{20-5} = 0b0000001000000000;
1466}
1467def SETF8  : BaseFlagManipulation<0, 0, (ins GPR32:$Rn), "setf8", "{\t$Rn}">;
1468def SETF16 : BaseFlagManipulation<0, 1, (ins GPR32:$Rn), "setf16", "{\t$Rn}">;
1469def RMIF   : FlagRotate<(ins GPR64:$Rn, uimm6:$imm, imm0_15:$mask), "rmif",
1470                        "{\t$Rn, $imm, $mask}">;
1471} // HasFlagM
1472
1473// v8.5 flag manipulation instructions
1474let Predicates = [HasAltNZCV], Uses = [NZCV], Defs = [NZCV] in {
1475
1476def XAFLAG : PstateWriteSimple<(ins), "xaflag", "">, Sched<[WriteSys]> {
1477  let Inst{18-16} = 0b000;
1478  let Inst{11-8} = 0b0000;
1479  let Unpredictable{11-8} = 0b1111;
1480  let Inst{7-5} = 0b001;
1481}
1482
1483def AXFLAG : PstateWriteSimple<(ins), "axflag", "">, Sched<[WriteSys]> {
1484  let Inst{18-16} = 0b000;
1485  let Inst{11-8} = 0b0000;
1486  let Unpredictable{11-8} = 0b1111;
1487  let Inst{7-5} = 0b010;
1488}
1489} // HasAltNZCV
1490
1491
1492// Armv8.5-A speculation barrier
1493def SB : SimpleSystemI<0, (ins), "sb", "">, Sched<[]> {
1494  let Inst{20-5} = 0b0001100110000111;
1495  let Unpredictable{11-8} = 0b1111;
1496  let Predicates = [HasSB];
1497  let hasSideEffects = 1;
1498}
1499
1500def : InstAlias<"clrex", (CLREX 0xf)>;
1501def : InstAlias<"isb", (ISB 0xf)>;
1502def : InstAlias<"ssbb", (DSB 0)>;
1503def : InstAlias<"pssbb", (DSB 4)>;
1504def : InstAlias<"dfb", (DSB 0b1100)>, Requires<[HasV8_0r]>;
1505
1506def MRS    : MRSI;
1507def MSR    : MSRI;
1508def MSRpstateImm1 : MSRpstateImm0_1;
1509def MSRpstateImm4 : MSRpstateImm0_15;
1510
1511def : Pat<(AArch64mrs imm:$id),
1512          (MRS imm:$id)>;
1513
1514// The thread pointer (on Linux, at least, where this has been implemented) is
1515// TPIDR_EL0.
1516def MOVbaseTLS : Pseudo<(outs GPR64:$dst), (ins),
1517                       [(set GPR64:$dst, AArch64threadpointer)]>, Sched<[WriteSys]>;
1518
1519// This gets lowered into a 24-byte instruction sequence
1520let Defs = [ X9, X16, X17, NZCV ], Size = 24 in {
1521def KCFI_CHECK : Pseudo<
1522  (outs), (ins GPR64:$ptr, i32imm:$type), []>, Sched<[]>;
1523}
1524
1525let Uses = [ X9 ], Defs = [ X16, X17, LR, NZCV ] in {
1526def HWASAN_CHECK_MEMACCESS : Pseudo<
1527  (outs), (ins GPR64noip:$ptr, i32imm:$accessinfo),
1528  [(int_hwasan_check_memaccess X9, GPR64noip:$ptr, (i32 timm:$accessinfo))]>,
1529  Sched<[]>;
1530}
1531
1532let Uses = [ X20 ], Defs = [ X16, X17, LR, NZCV ] in {
1533def HWASAN_CHECK_MEMACCESS_SHORTGRANULES : Pseudo<
1534  (outs), (ins GPR64noip:$ptr, i32imm:$accessinfo),
1535  [(int_hwasan_check_memaccess_shortgranules X20, GPR64noip:$ptr, (i32 timm:$accessinfo))]>,
1536  Sched<[]>;
1537}
1538
1539// The virtual cycle counter register is CNTVCT_EL0.
1540def : Pat<(readcyclecounter), (MRS 0xdf02)>;
1541
1542// FPCR register
1543let Uses = [FPCR] in
1544def MRS_FPCR : Pseudo<(outs GPR64:$dst), (ins),
1545                      [(set GPR64:$dst, (int_aarch64_get_fpcr))]>,
1546               PseudoInstExpansion<(MRS GPR64:$dst, 0xda20)>,
1547               Sched<[WriteSys]>;
1548let Defs = [FPCR] in
1549def MSR_FPCR : Pseudo<(outs), (ins GPR64:$val),
1550                      [(int_aarch64_set_fpcr i64:$val)]>,
1551               PseudoInstExpansion<(MSR 0xda20, GPR64:$val)>,
1552               Sched<[WriteSys]>;
1553
1554// Generic system instructions
1555def SYSxt  : SystemXtI<0, "sys">;
1556def SYSLxt : SystemLXtI<1, "sysl">;
1557
1558def : InstAlias<"sys $op1, $Cn, $Cm, $op2",
1559                (SYSxt imm0_7:$op1, sys_cr_op:$Cn,
1560                 sys_cr_op:$Cm, imm0_7:$op2, XZR)>;
1561
1562
1563let Predicates = [HasTME] in {
1564
1565def TSTART : TMSystemI<0b0000, "tstart",
1566                      [(set GPR64:$Rt, (int_aarch64_tstart))]>;
1567
1568def TCOMMIT : TMSystemINoOperand<0b0000, "tcommit", [(int_aarch64_tcommit)]>;
1569
1570def TCANCEL : TMSystemException<0b011, "tcancel",
1571                                [(int_aarch64_tcancel timm64_0_65535:$imm)]>;
1572
1573def TTEST : TMSystemI<0b0001, "ttest", [(set GPR64:$Rt, (int_aarch64_ttest))]> {
1574  let mayLoad = 0;
1575  let mayStore = 0;
1576}
1577} // HasTME
1578
1579//===----------------------------------------------------------------------===//
1580// Move immediate instructions.
1581//===----------------------------------------------------------------------===//
1582
1583defm MOVK : InsertImmediate<0b11, "movk">;
1584defm MOVN : MoveImmediate<0b00, "movn">;
1585
1586let PostEncoderMethod = "fixMOVZ" in
1587defm MOVZ : MoveImmediate<0b10, "movz">;
1588
1589// First group of aliases covers an implicit "lsl #0".
1590def : InstAlias<"movk $dst, $imm", (MOVKWi GPR32:$dst, timm32_0_65535:$imm, 0), 0>;
1591def : InstAlias<"movk $dst, $imm", (MOVKXi GPR64:$dst, timm32_0_65535:$imm, 0), 0>;
1592def : InstAlias<"movn $dst, $imm", (MOVNWi GPR32:$dst, timm32_0_65535:$imm, 0)>;
1593def : InstAlias<"movn $dst, $imm", (MOVNXi GPR64:$dst, timm32_0_65535:$imm, 0)>;
1594def : InstAlias<"movz $dst, $imm", (MOVZWi GPR32:$dst, timm32_0_65535:$imm, 0)>;
1595def : InstAlias<"movz $dst, $imm", (MOVZXi GPR64:$dst, timm32_0_65535:$imm, 0)>;
1596
1597// Next, we have various ELF relocations with the ":XYZ_g0:sym" syntax.
1598def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movw_symbol_g3:$sym, 48)>;
1599def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movw_symbol_g2:$sym, 32)>;
1600def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movw_symbol_g1:$sym, 16)>;
1601def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movw_symbol_g0:$sym, 0)>;
1602
1603def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movw_symbol_g3:$sym, 48)>;
1604def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movw_symbol_g2:$sym, 32)>;
1605def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movw_symbol_g1:$sym, 16)>;
1606def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movw_symbol_g0:$sym, 0)>;
1607
1608def : InstAlias<"movk $Rd, $sym", (MOVKXi GPR64:$Rd, movw_symbol_g3:$sym, 48), 0>;
1609def : InstAlias<"movk $Rd, $sym", (MOVKXi GPR64:$Rd, movw_symbol_g2:$sym, 32), 0>;
1610def : InstAlias<"movk $Rd, $sym", (MOVKXi GPR64:$Rd, movw_symbol_g1:$sym, 16), 0>;
1611def : InstAlias<"movk $Rd, $sym", (MOVKXi GPR64:$Rd, movw_symbol_g0:$sym, 0), 0>;
1612
1613def : InstAlias<"movz $Rd, $sym", (MOVZWi GPR32:$Rd, movw_symbol_g1:$sym, 16)>;
1614def : InstAlias<"movz $Rd, $sym", (MOVZWi GPR32:$Rd, movw_symbol_g0:$sym, 0)>;
1615
1616def : InstAlias<"movn $Rd, $sym", (MOVNWi GPR32:$Rd, movw_symbol_g1:$sym, 16)>;
1617def : InstAlias<"movn $Rd, $sym", (MOVNWi GPR32:$Rd, movw_symbol_g0:$sym, 0)>;
1618
1619def : InstAlias<"movk $Rd, $sym", (MOVKWi GPR32:$Rd, movw_symbol_g1:$sym, 16), 0>;
1620def : InstAlias<"movk $Rd, $sym", (MOVKWi GPR32:$Rd, movw_symbol_g0:$sym, 0), 0>;
1621
1622// Final group of aliases covers true "mov $Rd, $imm" cases.
1623multiclass movw_mov_alias<string basename,Instruction INST, RegisterClass GPR,
1624                          int width, int shift> {
1625  def _asmoperand : AsmOperandClass {
1626    let Name = basename # width # "_lsl" # shift # "MovAlias";
1627    let PredicateMethod = "is" # basename # "MovAlias<" # width # ", "
1628                               # shift # ">";
1629    let RenderMethod = "add" # basename # "MovAliasOperands<" # shift # ">";
1630  }
1631
1632  def _movimm : Operand<i32> {
1633    let ParserMatchClass = !cast<AsmOperandClass>(NAME # "_asmoperand");
1634  }
1635
1636  def : InstAlias<"mov $Rd, $imm",
1637                  (INST GPR:$Rd, !cast<Operand>(NAME # "_movimm"):$imm, shift)>;
1638}
1639
1640defm : movw_mov_alias<"MOVZ", MOVZWi, GPR32, 32, 0>;
1641defm : movw_mov_alias<"MOVZ", MOVZWi, GPR32, 32, 16>;
1642
1643defm : movw_mov_alias<"MOVZ", MOVZXi, GPR64, 64, 0>;
1644defm : movw_mov_alias<"MOVZ", MOVZXi, GPR64, 64, 16>;
1645defm : movw_mov_alias<"MOVZ", MOVZXi, GPR64, 64, 32>;
1646defm : movw_mov_alias<"MOVZ", MOVZXi, GPR64, 64, 48>;
1647
1648defm : movw_mov_alias<"MOVN", MOVNWi, GPR32, 32, 0>;
1649defm : movw_mov_alias<"MOVN", MOVNWi, GPR32, 32, 16>;
1650
1651defm : movw_mov_alias<"MOVN", MOVNXi, GPR64, 64, 0>;
1652defm : movw_mov_alias<"MOVN", MOVNXi, GPR64, 64, 16>;
1653defm : movw_mov_alias<"MOVN", MOVNXi, GPR64, 64, 32>;
1654defm : movw_mov_alias<"MOVN", MOVNXi, GPR64, 64, 48>;
1655
1656let isReMaterializable = 1, isCodeGenOnly = 1, isMoveImm = 1,
1657    isAsCheapAsAMove = 1 in {
1658// FIXME: The following pseudo instructions are only needed because remat
1659// cannot handle multiple instructions.  When that changes, we can select
1660// directly to the real instructions and get rid of these pseudos.
1661
1662def MOVi32imm
1663    : Pseudo<(outs GPR32:$dst), (ins i32imm:$src),
1664             [(set GPR32:$dst, imm:$src)]>,
1665      Sched<[WriteImm]>;
1666def MOVi64imm
1667    : Pseudo<(outs GPR64:$dst), (ins i64imm:$src),
1668             [(set GPR64:$dst, imm:$src)]>,
1669      Sched<[WriteImm]>;
1670} // isReMaterializable, isCodeGenOnly
1671
1672// If possible, we want to use MOVi32imm even for 64-bit moves. This gives the
1673// eventual expansion code fewer bits to worry about getting right. Marshalling
1674// the types is a little tricky though:
1675def i64imm_32bit : ImmLeaf<i64, [{
1676  return (Imm & 0xffffffffULL) == static_cast<uint64_t>(Imm);
1677}]>;
1678
1679def s64imm_32bit : ImmLeaf<i64, [{
1680  int64_t Imm64 = static_cast<int64_t>(Imm);
1681  return Imm64 >= std::numeric_limits<int32_t>::min() &&
1682         Imm64 <= std::numeric_limits<int32_t>::max();
1683}]>;
1684
1685def trunc_imm : SDNodeXForm<imm, [{
1686  return CurDAG->getTargetConstant(N->getZExtValue(), SDLoc(N), MVT::i32);
1687}]>;
1688
1689def gi_trunc_imm : GICustomOperandRenderer<"renderTruncImm">,
1690  GISDNodeXFormEquiv<trunc_imm>;
1691
1692let Predicates = [OptimizedGISelOrOtherSelector] in {
1693// The SUBREG_TO_REG isn't eliminated at -O0, which can result in pointless
1694// copies.
1695def : Pat<(i64 i64imm_32bit:$src),
1696          (SUBREG_TO_REG (i64 0), (MOVi32imm (trunc_imm imm:$src)), sub_32)>;
1697}
1698
1699// Materialize FP constants via MOVi32imm/MOVi64imm (MachO large code model).
1700def bitcast_fpimm_to_i32 : SDNodeXForm<fpimm, [{
1701return CurDAG->getTargetConstant(
1702  N->getValueAPF().bitcastToAPInt().getZExtValue(), SDLoc(N), MVT::i32);
1703}]>;
1704
1705def bitcast_fpimm_to_i64 : SDNodeXForm<fpimm, [{
1706return CurDAG->getTargetConstant(
1707  N->getValueAPF().bitcastToAPInt().getZExtValue(), SDLoc(N), MVT::i64);
1708}]>;
1709
1710
1711def : Pat<(f32 fpimm:$in),
1712  (COPY_TO_REGCLASS (MOVi32imm (bitcast_fpimm_to_i32 f32:$in)), FPR32)>;
1713def : Pat<(f64 fpimm:$in),
1714  (COPY_TO_REGCLASS (MOVi64imm (bitcast_fpimm_to_i64 f64:$in)), FPR64)>;
1715
1716
1717// Deal with the various forms of (ELF) large addressing with MOVZ/MOVK
1718// sequences.
1719def : Pat<(AArch64WrapperLarge tglobaladdr:$g3, tglobaladdr:$g2,
1720                             tglobaladdr:$g1, tglobaladdr:$g0),
1721          (MOVKXi (MOVKXi (MOVKXi (MOVZXi tglobaladdr:$g0, 0),
1722                                  tglobaladdr:$g1, 16),
1723                          tglobaladdr:$g2, 32),
1724                  tglobaladdr:$g3, 48)>;
1725
1726def : Pat<(AArch64WrapperLarge tblockaddress:$g3, tblockaddress:$g2,
1727                             tblockaddress:$g1, tblockaddress:$g0),
1728          (MOVKXi (MOVKXi (MOVKXi (MOVZXi tblockaddress:$g0, 0),
1729                                  tblockaddress:$g1, 16),
1730                          tblockaddress:$g2, 32),
1731                  tblockaddress:$g3, 48)>;
1732
1733def : Pat<(AArch64WrapperLarge tconstpool:$g3, tconstpool:$g2,
1734                             tconstpool:$g1, tconstpool:$g0),
1735          (MOVKXi (MOVKXi (MOVKXi (MOVZXi tconstpool:$g0, 0),
1736                                  tconstpool:$g1, 16),
1737                          tconstpool:$g2, 32),
1738                  tconstpool:$g3, 48)>;
1739
1740def : Pat<(AArch64WrapperLarge tjumptable:$g3, tjumptable:$g2,
1741                             tjumptable:$g1, tjumptable:$g0),
1742          (MOVKXi (MOVKXi (MOVKXi (MOVZXi tjumptable:$g0, 0),
1743                                  tjumptable:$g1, 16),
1744                          tjumptable:$g2, 32),
1745                  tjumptable:$g3, 48)>;
1746
1747
1748//===----------------------------------------------------------------------===//
1749// Arithmetic instructions.
1750//===----------------------------------------------------------------------===//
1751
1752// Add/subtract with carry.
1753defm ADC : AddSubCarry<0, "adc", "adcs", AArch64adc, AArch64adc_flag>;
1754defm SBC : AddSubCarry<1, "sbc", "sbcs", AArch64sbc, AArch64sbc_flag>;
1755
1756def : InstAlias<"ngc $dst, $src",  (SBCWr  GPR32:$dst, WZR, GPR32:$src)>;
1757def : InstAlias<"ngc $dst, $src",  (SBCXr  GPR64:$dst, XZR, GPR64:$src)>;
1758def : InstAlias<"ngcs $dst, $src", (SBCSWr GPR32:$dst, WZR, GPR32:$src)>;
1759def : InstAlias<"ngcs $dst, $src", (SBCSXr GPR64:$dst, XZR, GPR64:$src)>;
1760
1761// Add/subtract
1762defm ADD : AddSub<0, "add", "sub", add>;
1763defm SUB : AddSub<1, "sub", "add">;
1764
1765def : InstAlias<"mov $dst, $src",
1766                (ADDWri GPR32sponly:$dst, GPR32sp:$src, 0, 0)>;
1767def : InstAlias<"mov $dst, $src",
1768                (ADDWri GPR32sp:$dst, GPR32sponly:$src, 0, 0)>;
1769def : InstAlias<"mov $dst, $src",
1770                (ADDXri GPR64sponly:$dst, GPR64sp:$src, 0, 0)>;
1771def : InstAlias<"mov $dst, $src",
1772                (ADDXri GPR64sp:$dst, GPR64sponly:$src, 0, 0)>;
1773
1774defm ADDS : AddSubS<0, "adds", AArch64add_flag, "cmn", "subs", "cmp">;
1775defm SUBS : AddSubS<1, "subs", AArch64sub_flag, "cmp", "adds", "cmn">;
1776
1777def copyFromSP: PatLeaf<(i64 GPR64:$src), [{
1778  return N->getOpcode() == ISD::CopyFromReg &&
1779         cast<RegisterSDNode>(N->getOperand(1))->getReg() == AArch64::SP;
1780}]>;
1781
1782// Use SUBS instead of SUB to enable CSE between SUBS and SUB.
1783def : Pat<(sub GPR32sp:$Rn, addsub_shifted_imm32:$imm),
1784          (SUBSWri GPR32sp:$Rn, addsub_shifted_imm32:$imm)>;
1785def : Pat<(sub GPR64sp:$Rn, addsub_shifted_imm64:$imm),
1786          (SUBSXri GPR64sp:$Rn, addsub_shifted_imm64:$imm)>;
1787def : Pat<(sub GPR32:$Rn, GPR32:$Rm),
1788          (SUBSWrr GPR32:$Rn, GPR32:$Rm)>;
1789def : Pat<(sub GPR64:$Rn, GPR64:$Rm),
1790          (SUBSXrr GPR64:$Rn, GPR64:$Rm)>;
1791def : Pat<(sub GPR32:$Rn, arith_shifted_reg32:$Rm),
1792          (SUBSWrs GPR32:$Rn, arith_shifted_reg32:$Rm)>;
1793def : Pat<(sub GPR64:$Rn, arith_shifted_reg64:$Rm),
1794          (SUBSXrs GPR64:$Rn, arith_shifted_reg64:$Rm)>;
1795let AddedComplexity = 1 in {
1796def : Pat<(sub GPR32sp:$R2, arith_extended_reg32_i32:$R3),
1797          (SUBSWrx GPR32sp:$R2, arith_extended_reg32_i32:$R3)>;
1798def : Pat<(sub GPR64sp:$R2, arith_extended_reg32to64_i64:$R3),
1799          (SUBSXrx GPR64sp:$R2, arith_extended_reg32to64_i64:$R3)>;
1800def : Pat<(sub copyFromSP:$R2, (arith_uxtx GPR64:$R3, arith_extendlsl64:$imm)),
1801          (SUBXrx64 GPR64sp:$R2, GPR64:$R3, arith_extendlsl64:$imm)>;
1802}
1803
1804// Because of the immediate format for add/sub-imm instructions, the
1805// expression (add x, -1) must be transformed to (SUB{W,X}ri x, 1).
1806//  These patterns capture that transformation.
1807let AddedComplexity = 1 in {
1808def : Pat<(add GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
1809          (SUBSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
1810def : Pat<(add GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
1811          (SUBSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
1812def : Pat<(sub GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
1813          (ADDWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
1814def : Pat<(sub GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
1815          (ADDXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
1816}
1817
1818// Because of the immediate format for add/sub-imm instructions, the
1819// expression (add x, -1) must be transformed to (SUB{W,X}ri x, 1).
1820//  These patterns capture that transformation.
1821let AddedComplexity = 1 in {
1822def : Pat<(AArch64add_flag GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
1823          (SUBSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
1824def : Pat<(AArch64add_flag GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
1825          (SUBSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
1826def : Pat<(AArch64sub_flag GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
1827          (ADDSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
1828def : Pat<(AArch64sub_flag GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
1829          (ADDSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
1830}
1831
1832def : InstAlias<"neg $dst, $src", (SUBWrs GPR32:$dst, WZR, GPR32:$src, 0), 3>;
1833def : InstAlias<"neg $dst, $src", (SUBXrs GPR64:$dst, XZR, GPR64:$src, 0), 3>;
1834def : InstAlias<"neg $dst, $src$shift",
1835                (SUBWrs GPR32:$dst, WZR, GPR32:$src, arith_shift32:$shift), 2>;
1836def : InstAlias<"neg $dst, $src$shift",
1837                (SUBXrs GPR64:$dst, XZR, GPR64:$src, arith_shift64:$shift), 2>;
1838
1839def : InstAlias<"negs $dst, $src", (SUBSWrs GPR32:$dst, WZR, GPR32:$src, 0), 3>;
1840def : InstAlias<"negs $dst, $src", (SUBSXrs GPR64:$dst, XZR, GPR64:$src, 0), 3>;
1841def : InstAlias<"negs $dst, $src$shift",
1842                (SUBSWrs GPR32:$dst, WZR, GPR32:$src, arith_shift32:$shift), 2>;
1843def : InstAlias<"negs $dst, $src$shift",
1844                (SUBSXrs GPR64:$dst, XZR, GPR64:$src, arith_shift64:$shift), 2>;
1845
1846
1847// Unsigned/Signed divide
1848defm UDIV : Div<0, "udiv", udiv>;
1849defm SDIV : Div<1, "sdiv", sdiv>;
1850
1851def : Pat<(int_aarch64_udiv GPR32:$Rn, GPR32:$Rm), (UDIVWr GPR32:$Rn, GPR32:$Rm)>;
1852def : Pat<(int_aarch64_udiv GPR64:$Rn, GPR64:$Rm), (UDIVXr GPR64:$Rn, GPR64:$Rm)>;
1853def : Pat<(int_aarch64_sdiv GPR32:$Rn, GPR32:$Rm), (SDIVWr GPR32:$Rn, GPR32:$Rm)>;
1854def : Pat<(int_aarch64_sdiv GPR64:$Rn, GPR64:$Rm), (SDIVXr GPR64:$Rn, GPR64:$Rm)>;
1855
1856// Variable shift
1857defm ASRV : Shift<0b10, "asr", sra>;
1858defm LSLV : Shift<0b00, "lsl", shl>;
1859defm LSRV : Shift<0b01, "lsr", srl>;
1860defm RORV : Shift<0b11, "ror", rotr>;
1861
1862def : ShiftAlias<"asrv", ASRVWr, GPR32>;
1863def : ShiftAlias<"asrv", ASRVXr, GPR64>;
1864def : ShiftAlias<"lslv", LSLVWr, GPR32>;
1865def : ShiftAlias<"lslv", LSLVXr, GPR64>;
1866def : ShiftAlias<"lsrv", LSRVWr, GPR32>;
1867def : ShiftAlias<"lsrv", LSRVXr, GPR64>;
1868def : ShiftAlias<"rorv", RORVWr, GPR32>;
1869def : ShiftAlias<"rorv", RORVXr, GPR64>;
1870
1871// Multiply-add
1872let AddedComplexity = 5 in {
1873defm MADD : MulAccum<0, "madd">;
1874defm MSUB : MulAccum<1, "msub">;
1875
1876def : Pat<(i32 (mul GPR32:$Rn, GPR32:$Rm)),
1877          (MADDWrrr GPR32:$Rn, GPR32:$Rm, WZR)>;
1878def : Pat<(i64 (mul GPR64:$Rn, GPR64:$Rm)),
1879          (MADDXrrr GPR64:$Rn, GPR64:$Rm, XZR)>;
1880
1881def : Pat<(i32 (ineg (mul GPR32:$Rn, GPR32:$Rm))),
1882          (MSUBWrrr GPR32:$Rn, GPR32:$Rm, WZR)>;
1883def : Pat<(i64 (ineg (mul GPR64:$Rn, GPR64:$Rm))),
1884          (MSUBXrrr GPR64:$Rn, GPR64:$Rm, XZR)>;
1885def : Pat<(i32 (mul (ineg GPR32:$Rn), GPR32:$Rm)),
1886          (MSUBWrrr GPR32:$Rn, GPR32:$Rm, WZR)>;
1887def : Pat<(i64 (mul (ineg GPR64:$Rn), GPR64:$Rm)),
1888          (MSUBXrrr GPR64:$Rn, GPR64:$Rm, XZR)>;
1889} // AddedComplexity = 5
1890
1891let AddedComplexity = 5 in {
1892def SMADDLrrr : WideMulAccum<0, 0b001, "smaddl", add, sext>;
1893def SMSUBLrrr : WideMulAccum<1, 0b001, "smsubl", sub, sext>;
1894def UMADDLrrr : WideMulAccum<0, 0b101, "umaddl", add, zext>;
1895def UMSUBLrrr : WideMulAccum<1, 0b101, "umsubl", sub, zext>;
1896
1897def : Pat<(i64 (mul (sext_inreg GPR64:$Rn, i32), (sext_inreg GPR64:$Rm, i32))),
1898          (SMADDLrrr (EXTRACT_SUBREG $Rn, sub_32), (EXTRACT_SUBREG $Rm, sub_32), XZR)>;
1899def : Pat<(i64 (mul (sext_inreg GPR64:$Rn, i32), (sext GPR32:$Rm))),
1900          (SMADDLrrr (EXTRACT_SUBREG $Rn, sub_32), $Rm, XZR)>;
1901def : Pat<(i64 (mul (sext GPR32:$Rn), (sext GPR32:$Rm))),
1902          (SMADDLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
1903def : Pat<(i64 (mul (and GPR64:$Rn, 0xFFFFFFFF), (and GPR64:$Rm, 0xFFFFFFFF))),
1904          (UMADDLrrr (EXTRACT_SUBREG $Rn, sub_32), (EXTRACT_SUBREG $Rm, sub_32), XZR)>;
1905def : Pat<(i64 (mul (and GPR64:$Rn, 0xFFFFFFFF), (zext GPR32:$Rm))),
1906          (UMADDLrrr (EXTRACT_SUBREG $Rn, sub_32), $Rm, XZR)>;
1907def : Pat<(i64 (mul (zext GPR32:$Rn), (zext GPR32:$Rm))),
1908          (UMADDLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
1909
1910def : Pat<(i64 (ineg (mul (sext GPR32:$Rn), (sext GPR32:$Rm)))),
1911          (SMSUBLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
1912def : Pat<(i64 (ineg (mul (zext GPR32:$Rn), (zext GPR32:$Rm)))),
1913          (UMSUBLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
1914
1915def : Pat<(i64 (mul (sext GPR32:$Rn), (s64imm_32bit:$C))),
1916          (SMADDLrrr GPR32:$Rn, (MOVi32imm (trunc_imm imm:$C)), XZR)>;
1917def : Pat<(i64 (mul (zext GPR32:$Rn), (i64imm_32bit:$C))),
1918          (UMADDLrrr GPR32:$Rn, (MOVi32imm (trunc_imm imm:$C)), XZR)>;
1919def : Pat<(i64 (mul (sext_inreg GPR64:$Rn, i32), (s64imm_32bit:$C))),
1920          (SMADDLrrr (i32 (EXTRACT_SUBREG GPR64:$Rn, sub_32)),
1921                     (MOVi32imm (trunc_imm imm:$C)), XZR)>;
1922
1923def : Pat<(i64 (ineg (mul (sext GPR32:$Rn), (s64imm_32bit:$C)))),
1924          (SMSUBLrrr GPR32:$Rn, (MOVi32imm (trunc_imm imm:$C)), XZR)>;
1925def : Pat<(i64 (ineg (mul (zext GPR32:$Rn), (i64imm_32bit:$C)))),
1926          (UMSUBLrrr GPR32:$Rn, (MOVi32imm (trunc_imm imm:$C)), XZR)>;
1927def : Pat<(i64 (ineg (mul (sext_inreg GPR64:$Rn, i32), (s64imm_32bit:$C)))),
1928          (SMSUBLrrr (i32 (EXTRACT_SUBREG GPR64:$Rn, sub_32)),
1929                     (MOVi32imm (trunc_imm imm:$C)), XZR)>;
1930
1931def : Pat<(i64 (add (mul (sext GPR32:$Rn), (s64imm_32bit:$C)), GPR64:$Ra)),
1932          (SMADDLrrr GPR32:$Rn, (MOVi32imm (trunc_imm imm:$C)), GPR64:$Ra)>;
1933def : Pat<(i64 (add (mul (zext GPR32:$Rn), (i64imm_32bit:$C)), GPR64:$Ra)),
1934          (UMADDLrrr GPR32:$Rn, (MOVi32imm (trunc_imm imm:$C)), GPR64:$Ra)>;
1935def : Pat<(i64 (add (mul (sext_inreg GPR64:$Rn, i32), (s64imm_32bit:$C)),
1936                    GPR64:$Ra)),
1937          (SMADDLrrr (i32 (EXTRACT_SUBREG GPR64:$Rn, sub_32)),
1938                     (MOVi32imm (trunc_imm imm:$C)), GPR64:$Ra)>;
1939
1940def : Pat<(i64 (sub GPR64:$Ra, (mul (sext GPR32:$Rn), (s64imm_32bit:$C)))),
1941          (SMSUBLrrr GPR32:$Rn, (MOVi32imm (trunc_imm imm:$C)), GPR64:$Ra)>;
1942def : Pat<(i64 (sub GPR64:$Ra, (mul (zext GPR32:$Rn), (i64imm_32bit:$C)))),
1943          (UMSUBLrrr GPR32:$Rn, (MOVi32imm (trunc_imm imm:$C)), GPR64:$Ra)>;
1944def : Pat<(i64 (sub GPR64:$Ra, (mul (sext_inreg GPR64:$Rn, i32),
1945                                    (s64imm_32bit:$C)))),
1946          (SMSUBLrrr (i32 (EXTRACT_SUBREG GPR64:$Rn, sub_32)),
1947                     (MOVi32imm (trunc_imm imm:$C)), GPR64:$Ra)>;
1948
1949def : Pat<(i64 (smullwithsignbits GPR64:$Rn, GPR64:$Rm)),
1950          (SMADDLrrr (EXTRACT_SUBREG $Rn, sub_32), (EXTRACT_SUBREG $Rm, sub_32), XZR)>;
1951def : Pat<(i64 (smullwithsignbits GPR64:$Rn, (sext GPR32:$Rm))),
1952          (SMADDLrrr (EXTRACT_SUBREG $Rn, sub_32), $Rm, XZR)>;
1953
1954def : Pat<(i64 (add (smullwithsignbits GPR64:$Rn, GPR64:$Rm), GPR64:$Ra)),
1955          (SMADDLrrr (EXTRACT_SUBREG $Rn, sub_32), (EXTRACT_SUBREG $Rm, sub_32), GPR64:$Ra)>;
1956def : Pat<(i64 (add (smullwithsignbits GPR64:$Rn, (sext GPR32:$Rm)), GPR64:$Ra)),
1957          (SMADDLrrr (EXTRACT_SUBREG $Rn, sub_32), $Rm, GPR64:$Ra)>;
1958
1959def : Pat<(i64 (ineg (smullwithsignbits GPR64:$Rn, GPR64:$Rm))),
1960          (SMSUBLrrr (EXTRACT_SUBREG $Rn, sub_32), (EXTRACT_SUBREG $Rm, sub_32), XZR)>;
1961def : Pat<(i64 (ineg (smullwithsignbits GPR64:$Rn, (sext GPR32:$Rm)))),
1962          (SMSUBLrrr (EXTRACT_SUBREG $Rn, sub_32), $Rm, XZR)>;
1963
1964def : Pat<(i64 (sub GPR64:$Ra, (smullwithsignbits GPR64:$Rn, GPR64:$Rm))),
1965          (SMSUBLrrr (EXTRACT_SUBREG $Rn, sub_32), (EXTRACT_SUBREG $Rm, sub_32), GPR64:$Ra)>;
1966def : Pat<(i64 (sub GPR64:$Ra, (smullwithsignbits GPR64:$Rn, (sext GPR32:$Rm)))),
1967          (SMSUBLrrr (EXTRACT_SUBREG $Rn, sub_32), $Rm, GPR64:$Ra)>;
1968
1969def : Pat<(i64 (mul top32Zero:$Rn, top32Zero:$Rm)),
1970          (UMADDLrrr (EXTRACT_SUBREG $Rn, sub_32), (EXTRACT_SUBREG $Rm, sub_32), XZR)>;
1971def : Pat<(i64 (mul top32Zero:$Rn, (zext GPR32:$Rm))),
1972          (UMADDLrrr (EXTRACT_SUBREG $Rn, sub_32), $Rm, XZR)>;
1973
1974def : Pat<(i64 (add (mul top32Zero:$Rn, top32Zero:$Rm), GPR64:$Ra)),
1975          (UMADDLrrr (EXTRACT_SUBREG $Rn, sub_32), (EXTRACT_SUBREG $Rm, sub_32), GPR64:$Ra)>;
1976def : Pat<(i64 (add (mul top32Zero:$Rn, (zext GPR32:$Rm)), GPR64:$Ra)),
1977          (UMADDLrrr (EXTRACT_SUBREG $Rn, sub_32), $Rm, GPR64:$Ra)>;
1978
1979def : Pat<(i64 (ineg (mul top32Zero:$Rn, top32Zero:$Rm))),
1980          (UMSUBLrrr (EXTRACT_SUBREG $Rn, sub_32), (EXTRACT_SUBREG $Rm, sub_32), XZR)>;
1981def : Pat<(i64 (ineg (mul top32Zero:$Rn, (zext GPR32:$Rm)))),
1982          (UMSUBLrrr (EXTRACT_SUBREG $Rn, sub_32), $Rm, XZR)>;
1983
1984def : Pat<(i64 (sub GPR64:$Ra, (mul top32Zero:$Rn, top32Zero:$Rm))),
1985          (UMSUBLrrr (EXTRACT_SUBREG $Rn, sub_32), (EXTRACT_SUBREG $Rm, sub_32), GPR64:$Ra)>;
1986def : Pat<(i64 (sub GPR64:$Ra, (mul top32Zero:$Rn, (zext GPR32:$Rm)))),
1987          (UMSUBLrrr (EXTRACT_SUBREG $Rn, sub_32), $Rm, GPR64:$Ra)>;
1988} // AddedComplexity = 5
1989
1990def : MulAccumWAlias<"mul", MADDWrrr>;
1991def : MulAccumXAlias<"mul", MADDXrrr>;
1992def : MulAccumWAlias<"mneg", MSUBWrrr>;
1993def : MulAccumXAlias<"mneg", MSUBXrrr>;
1994def : WideMulAccumAlias<"smull", SMADDLrrr>;
1995def : WideMulAccumAlias<"smnegl", SMSUBLrrr>;
1996def : WideMulAccumAlias<"umull", UMADDLrrr>;
1997def : WideMulAccumAlias<"umnegl", UMSUBLrrr>;
1998
1999// Multiply-high
2000def SMULHrr : MulHi<0b010, "smulh", mulhs>;
2001def UMULHrr : MulHi<0b110, "umulh", mulhu>;
2002
2003// CRC32
2004def CRC32Brr : BaseCRC32<0, 0b00, 0, GPR32, int_aarch64_crc32b, "crc32b">;
2005def CRC32Hrr : BaseCRC32<0, 0b01, 0, GPR32, int_aarch64_crc32h, "crc32h">;
2006def CRC32Wrr : BaseCRC32<0, 0b10, 0, GPR32, int_aarch64_crc32w, "crc32w">;
2007def CRC32Xrr : BaseCRC32<1, 0b11, 0, GPR64, int_aarch64_crc32x, "crc32x">;
2008
2009def CRC32CBrr : BaseCRC32<0, 0b00, 1, GPR32, int_aarch64_crc32cb, "crc32cb">;
2010def CRC32CHrr : BaseCRC32<0, 0b01, 1, GPR32, int_aarch64_crc32ch, "crc32ch">;
2011def CRC32CWrr : BaseCRC32<0, 0b10, 1, GPR32, int_aarch64_crc32cw, "crc32cw">;
2012def CRC32CXrr : BaseCRC32<1, 0b11, 1, GPR64, int_aarch64_crc32cx, "crc32cx">;
2013
2014// v8.1 atomic CAS
2015defm CAS   : CompareAndSwap<0, 0, "">;
2016defm CASA  : CompareAndSwap<1, 0, "a">;
2017defm CASL  : CompareAndSwap<0, 1, "l">;
2018defm CASAL : CompareAndSwap<1, 1, "al">;
2019
2020// v8.1 atomic CASP
2021defm CASP   : CompareAndSwapPair<0, 0, "">;
2022defm CASPA  : CompareAndSwapPair<1, 0, "a">;
2023defm CASPL  : CompareAndSwapPair<0, 1, "l">;
2024defm CASPAL : CompareAndSwapPair<1, 1, "al">;
2025
2026// v8.1 atomic SWP
2027defm SWP   : Swap<0, 0, "">;
2028defm SWPA  : Swap<1, 0, "a">;
2029defm SWPL  : Swap<0, 1, "l">;
2030defm SWPAL : Swap<1, 1, "al">;
2031
2032// v8.1 atomic LD<OP>(register). Performs load and then ST<OP>(register)
2033defm LDADD   : LDOPregister<0b000, "add", 0, 0, "">;
2034defm LDADDA  : LDOPregister<0b000, "add", 1, 0, "a">;
2035defm LDADDL  : LDOPregister<0b000, "add", 0, 1, "l">;
2036defm LDADDAL : LDOPregister<0b000, "add", 1, 1, "al">;
2037
2038defm LDCLR   : LDOPregister<0b001, "clr", 0, 0, "">;
2039defm LDCLRA  : LDOPregister<0b001, "clr", 1, 0, "a">;
2040defm LDCLRL  : LDOPregister<0b001, "clr", 0, 1, "l">;
2041defm LDCLRAL : LDOPregister<0b001, "clr", 1, 1, "al">;
2042
2043defm LDEOR   : LDOPregister<0b010, "eor", 0, 0, "">;
2044defm LDEORA  : LDOPregister<0b010, "eor", 1, 0, "a">;
2045defm LDEORL  : LDOPregister<0b010, "eor", 0, 1, "l">;
2046defm LDEORAL : LDOPregister<0b010, "eor", 1, 1, "al">;
2047
2048defm LDSET   : LDOPregister<0b011, "set", 0, 0, "">;
2049defm LDSETA  : LDOPregister<0b011, "set", 1, 0, "a">;
2050defm LDSETL  : LDOPregister<0b011, "set", 0, 1, "l">;
2051defm LDSETAL : LDOPregister<0b011, "set", 1, 1, "al">;
2052
2053defm LDSMAX   : LDOPregister<0b100, "smax", 0, 0, "">;
2054defm LDSMAXA  : LDOPregister<0b100, "smax", 1, 0, "a">;
2055defm LDSMAXL  : LDOPregister<0b100, "smax", 0, 1, "l">;
2056defm LDSMAXAL : LDOPregister<0b100, "smax", 1, 1, "al">;
2057
2058defm LDSMIN   : LDOPregister<0b101, "smin", 0, 0, "">;
2059defm LDSMINA  : LDOPregister<0b101, "smin", 1, 0, "a">;
2060defm LDSMINL  : LDOPregister<0b101, "smin", 0, 1, "l">;
2061defm LDSMINAL : LDOPregister<0b101, "smin", 1, 1, "al">;
2062
2063defm LDUMAX   : LDOPregister<0b110, "umax", 0, 0, "">;
2064defm LDUMAXA  : LDOPregister<0b110, "umax", 1, 0, "a">;
2065defm LDUMAXL  : LDOPregister<0b110, "umax", 0, 1, "l">;
2066defm LDUMAXAL : LDOPregister<0b110, "umax", 1, 1, "al">;
2067
2068defm LDUMIN   : LDOPregister<0b111, "umin", 0, 0, "">;
2069defm LDUMINA  : LDOPregister<0b111, "umin", 1, 0, "a">;
2070defm LDUMINL  : LDOPregister<0b111, "umin", 0, 1, "l">;
2071defm LDUMINAL : LDOPregister<0b111, "umin", 1, 1, "al">;
2072
2073// v8.1 atomic ST<OP>(register) as aliases to "LD<OP>(register) when Rt=xZR"
2074defm : STOPregister<"stadd","LDADD">; // STADDx
2075defm : STOPregister<"stclr","LDCLR">; // STCLRx
2076defm : STOPregister<"steor","LDEOR">; // STEORx
2077defm : STOPregister<"stset","LDSET">; // STSETx
2078defm : STOPregister<"stsmax","LDSMAX">;// STSMAXx
2079defm : STOPregister<"stsmin","LDSMIN">;// STSMINx
2080defm : STOPregister<"stumax","LDUMAX">;// STUMAXx
2081defm : STOPregister<"stumin","LDUMIN">;// STUMINx
2082
2083// v8.5 Memory Tagging Extension
2084let Predicates = [HasMTE] in {
2085
2086def IRG   : BaseTwoOperandRegReg<0b1, 0b0, 0b000100, GPR64sp, "irg",
2087                                 int_aarch64_irg, GPR64sp, GPR64>, Sched<[]>;
2088
2089def GMI   : BaseTwoOperandRegReg<0b1, 0b0, 0b000101, GPR64, "gmi",
2090                                 int_aarch64_gmi, GPR64sp>, Sched<[]> {
2091  let isNotDuplicable = 1;
2092}
2093def ADDG  : AddSubG<0, "addg", null_frag>;
2094def SUBG  : AddSubG<1, "subg", null_frag>;
2095
2096def : InstAlias<"irg $dst, $src", (IRG GPR64sp:$dst, GPR64sp:$src, XZR), 1>;
2097
2098def SUBP : SUBP<0, "subp", int_aarch64_subp>, Sched<[]>;
2099def SUBPS : SUBP<1, "subps", null_frag>, Sched<[]>{
2100  let Defs = [NZCV];
2101}
2102
2103def : InstAlias<"cmpp $lhs, $rhs", (SUBPS XZR, GPR64sp:$lhs, GPR64sp:$rhs), 0>;
2104
2105def LDG : MemTagLoad<"ldg", "\t$Rt, [$Rn, $offset]">;
2106
2107def : Pat<(int_aarch64_addg (am_indexedu6s128 GPR64sp:$Rn, uimm6s16:$imm6), imm0_15:$imm4),
2108          (ADDG GPR64sp:$Rn, imm0_63:$imm6, imm0_15:$imm4)>;
2109def : Pat<(int_aarch64_ldg GPR64:$Rt, (am_indexeds9s128 GPR64sp:$Rn,  simm9s16:$offset)),
2110          (LDG GPR64:$Rt, GPR64sp:$Rn,  simm9s16:$offset)>;
2111
2112def : InstAlias<"ldg $Rt, [$Rn]", (LDG GPR64:$Rt, GPR64sp:$Rn, 0), 1>;
2113
2114def LDGM : MemTagVector<1, "ldgm", "\t$Rt, [$Rn]",
2115                   (outs GPR64:$Rt), (ins GPR64sp:$Rn)>;
2116def STGM : MemTagVector<0, "stgm", "\t$Rt, [$Rn]",
2117                   (outs), (ins GPR64:$Rt, GPR64sp:$Rn)>;
2118def STZGM : MemTagVector<0, "stzgm", "\t$Rt, [$Rn]",
2119                   (outs), (ins GPR64:$Rt, GPR64sp:$Rn)> {
2120  let Inst{23} = 0;
2121}
2122
2123defm STG   : MemTagStore<0b00, "stg">;
2124defm STZG  : MemTagStore<0b01, "stzg">;
2125defm ST2G  : MemTagStore<0b10, "st2g">;
2126defm STZ2G : MemTagStore<0b11, "stz2g">;
2127
2128def : Pat<(AArch64stg GPR64sp:$Rn, (am_indexeds9s128 GPR64sp:$Rm, simm9s16:$imm)),
2129          (STGOffset $Rn, $Rm, $imm)>;
2130def : Pat<(AArch64stzg GPR64sp:$Rn, (am_indexeds9s128 GPR64sp:$Rm, simm9s16:$imm)),
2131          (STZGOffset $Rn, $Rm, $imm)>;
2132def : Pat<(AArch64st2g GPR64sp:$Rn, (am_indexeds9s128 GPR64sp:$Rm, simm9s16:$imm)),
2133          (ST2GOffset $Rn, $Rm, $imm)>;
2134def : Pat<(AArch64stz2g GPR64sp:$Rn, (am_indexeds9s128 GPR64sp:$Rm, simm9s16:$imm)),
2135          (STZ2GOffset $Rn, $Rm, $imm)>;
2136
2137defm STGP     : StorePairOffset <0b01, 0, GPR64z, simm7s16, "stgp">;
2138def  STGPpre  : StorePairPreIdx <0b01, 0, GPR64z, simm7s16, "stgp">;
2139def  STGPpost : StorePairPostIdx<0b01, 0, GPR64z, simm7s16, "stgp">;
2140
2141def : Pat<(int_aarch64_stg GPR64:$Rt, (am_indexeds9s128 GPR64sp:$Rn, simm9s16:$offset)),
2142          (STGOffset GPR64:$Rt, GPR64sp:$Rn,  simm9s16:$offset)>;
2143
2144def : Pat<(int_aarch64_stgp (am_indexed7s128 GPR64sp:$Rn, simm7s16:$imm), GPR64:$Rt, GPR64:$Rt2),
2145          (STGPi $Rt, $Rt2, $Rn, $imm)>;
2146
2147def IRGstack
2148    : Pseudo<(outs GPR64sp:$Rd), (ins GPR64sp:$Rsp, GPR64:$Rm), []>,
2149      Sched<[]>;
2150def TAGPstack
2151    : Pseudo<(outs GPR64sp:$Rd), (ins GPR64sp:$Rn, uimm6s16:$imm6, GPR64sp:$Rm, imm0_15:$imm4), []>,
2152      Sched<[]>;
2153
2154// Explicit SP in the first operand prevents ShrinkWrap optimization
2155// from leaving this instruction out of the stack frame. When IRGstack
2156// is transformed into IRG, this operand is replaced with the actual
2157// register / expression for the tagged base pointer of the current function.
2158def : Pat<(int_aarch64_irg_sp i64:$Rm), (IRGstack SP, i64:$Rm)>;
2159
2160// Large STG to be expanded into a loop. $sz is the size, $Rn is start address.
2161// $Rn_wback is one past the end of the range. $Rm is the loop counter.
2162let isCodeGenOnly=1, mayStore=1 in {
2163def STGloop_wback
2164    : Pseudo<(outs GPR64common:$Rm, GPR64sp:$Rn_wback), (ins i64imm:$sz, GPR64sp:$Rn),
2165             [], "$Rn = $Rn_wback,@earlyclobber $Rn_wback,@earlyclobber $Rm" >,
2166      Sched<[WriteAdr, WriteST]>;
2167
2168def STZGloop_wback
2169    : Pseudo<(outs GPR64common:$Rm, GPR64sp:$Rn_wback), (ins i64imm:$sz, GPR64sp:$Rn),
2170             [], "$Rn = $Rn_wback,@earlyclobber $Rn_wback,@earlyclobber $Rm" >,
2171      Sched<[WriteAdr, WriteST]>;
2172
2173// A variant of the above where $Rn2 is an independent register not tied to the input register $Rn.
2174// Their purpose is to use a FrameIndex operand as $Rn (which of course can not be written back).
2175def STGloop
2176    : Pseudo<(outs GPR64common:$Rm, GPR64sp:$Rn2), (ins i64imm:$sz, GPR64sp:$Rn),
2177             [], "@earlyclobber $Rn2,@earlyclobber $Rm" >,
2178      Sched<[WriteAdr, WriteST]>;
2179
2180def STZGloop
2181    : Pseudo<(outs GPR64common:$Rm, GPR64sp:$Rn2), (ins i64imm:$sz, GPR64sp:$Rn),
2182             [], "@earlyclobber $Rn2,@earlyclobber $Rm" >,
2183      Sched<[WriteAdr, WriteST]>;
2184}
2185
2186} // Predicates = [HasMTE]
2187
2188//===----------------------------------------------------------------------===//
2189// Logical instructions.
2190//===----------------------------------------------------------------------===//
2191
2192// (immediate)
2193defm ANDS : LogicalImmS<0b11, "ands", AArch64and_flag, "bics">;
2194defm AND  : LogicalImm<0b00, "and", and, "bic">;
2195defm EOR  : LogicalImm<0b10, "eor", xor, "eon">;
2196defm ORR  : LogicalImm<0b01, "orr", or, "orn">;
2197
2198// FIXME: these aliases *are* canonical sometimes (when movz can't be
2199// used). Actually, it seems to be working right now, but putting logical_immXX
2200// here is a bit dodgy on the AsmParser side too.
2201def : InstAlias<"mov $dst, $imm", (ORRWri GPR32sp:$dst, WZR,
2202                                          logical_imm32:$imm), 0>;
2203def : InstAlias<"mov $dst, $imm", (ORRXri GPR64sp:$dst, XZR,
2204                                          logical_imm64:$imm), 0>;
2205
2206
2207// (register)
2208defm ANDS : LogicalRegS<0b11, 0, "ands", AArch64and_flag>;
2209defm BICS : LogicalRegS<0b11, 1, "bics",
2210                        BinOpFrag<(AArch64and_flag node:$LHS, (not node:$RHS))>>;
2211defm AND  : LogicalReg<0b00, 0, "and", and>;
2212defm BIC  : LogicalReg<0b00, 1, "bic",
2213                       BinOpFrag<(and node:$LHS, (not node:$RHS))>, 3>;
2214defm EON  : LogicalReg<0b10, 1, "eon",
2215                       BinOpFrag<(not (xor node:$LHS, node:$RHS))>>;
2216defm EOR  : LogicalReg<0b10, 0, "eor", xor>;
2217defm ORN  : LogicalReg<0b01, 1, "orn",
2218                       BinOpFrag<(or node:$LHS, (not node:$RHS))>>;
2219defm ORR  : LogicalReg<0b01, 0, "orr", or>;
2220
2221def : InstAlias<"mov $dst, $src", (ORRWrs GPR32:$dst, WZR, GPR32:$src, 0), 2>;
2222def : InstAlias<"mov $dst, $src", (ORRXrs GPR64:$dst, XZR, GPR64:$src, 0), 2>;
2223
2224def : InstAlias<"mvn $Wd, $Wm", (ORNWrs GPR32:$Wd, WZR, GPR32:$Wm, 0), 3>;
2225def : InstAlias<"mvn $Xd, $Xm", (ORNXrs GPR64:$Xd, XZR, GPR64:$Xm, 0), 3>;
2226
2227def : InstAlias<"mvn $Wd, $Wm$sh",
2228                (ORNWrs GPR32:$Wd, WZR, GPR32:$Wm, logical_shift32:$sh), 2>;
2229def : InstAlias<"mvn $Xd, $Xm$sh",
2230                (ORNXrs GPR64:$Xd, XZR, GPR64:$Xm, logical_shift64:$sh), 2>;
2231
2232def : InstAlias<"tst $src1, $src2",
2233                (ANDSWri WZR, GPR32:$src1, logical_imm32:$src2), 2>;
2234def : InstAlias<"tst $src1, $src2",
2235                (ANDSXri XZR, GPR64:$src1, logical_imm64:$src2), 2>;
2236
2237def : InstAlias<"tst $src1, $src2",
2238                        (ANDSWrs WZR, GPR32:$src1, GPR32:$src2, 0), 3>;
2239def : InstAlias<"tst $src1, $src2",
2240                        (ANDSXrs XZR, GPR64:$src1, GPR64:$src2, 0), 3>;
2241
2242def : InstAlias<"tst $src1, $src2$sh",
2243               (ANDSWrs WZR, GPR32:$src1, GPR32:$src2, logical_shift32:$sh), 2>;
2244def : InstAlias<"tst $src1, $src2$sh",
2245               (ANDSXrs XZR, GPR64:$src1, GPR64:$src2, logical_shift64:$sh), 2>;
2246
2247
2248def : Pat<(not GPR32:$Wm), (ORNWrr WZR, GPR32:$Wm)>;
2249def : Pat<(not GPR64:$Xm), (ORNXrr XZR, GPR64:$Xm)>;
2250
2251
2252//===----------------------------------------------------------------------===//
2253// One operand data processing instructions.
2254//===----------------------------------------------------------------------===//
2255
2256defm CLS    : OneOperandData<0b000101, "cls">;
2257defm CLZ    : OneOperandData<0b000100, "clz", ctlz>;
2258defm RBIT   : OneOperandData<0b000000, "rbit", bitreverse>;
2259
2260def  REV16Wr : OneWRegData<0b000001, "rev16",
2261                                     UnOpFrag<(rotr (bswap node:$LHS), (i64 16))>>;
2262def  REV16Xr : OneXRegData<0b000001, "rev16", null_frag>;
2263
2264def : Pat<(cttz GPR32:$Rn),
2265          (CLZWr (RBITWr GPR32:$Rn))>;
2266def : Pat<(cttz GPR64:$Rn),
2267          (CLZXr (RBITXr GPR64:$Rn))>;
2268def : Pat<(ctlz (or (shl (xor (sra GPR32:$Rn, (i64 31)), GPR32:$Rn), (i64 1)),
2269                (i32 1))),
2270          (CLSWr GPR32:$Rn)>;
2271def : Pat<(ctlz (or (shl (xor (sra GPR64:$Rn, (i64 63)), GPR64:$Rn), (i64 1)),
2272                (i64 1))),
2273          (CLSXr GPR64:$Rn)>;
2274def : Pat<(int_aarch64_cls GPR32:$Rn), (CLSWr GPR32:$Rn)>;
2275def : Pat<(int_aarch64_cls64 GPR64:$Rm), (EXTRACT_SUBREG (CLSXr GPR64:$Rm), sub_32)>;
2276
2277// Unlike the other one operand instructions, the instructions with the "rev"
2278// mnemonic do *not* just different in the size bit, but actually use different
2279// opcode bits for the different sizes.
2280def REVWr   : OneWRegData<0b000010, "rev", bswap>;
2281def REVXr   : OneXRegData<0b000011, "rev", bswap>;
2282def REV32Xr : OneXRegData<0b000010, "rev32",
2283                                    UnOpFrag<(rotr (bswap node:$LHS), (i64 32))>>;
2284
2285def : InstAlias<"rev64 $Rd, $Rn", (REVXr GPR64:$Rd, GPR64:$Rn), 0>;
2286
2287// The bswap commutes with the rotr so we want a pattern for both possible
2288// orders.
2289def : Pat<(bswap (rotr GPR32:$Rn, (i64 16))), (REV16Wr GPR32:$Rn)>;
2290def : Pat<(bswap (rotr GPR64:$Rn, (i64 32))), (REV32Xr GPR64:$Rn)>;
2291
2292// Match (srl (bswap x), C) -> revC if the upper bswap bits are known zero.
2293def : Pat<(srl (bswap top16Zero:$Rn), (i64 16)), (REV16Wr GPR32:$Rn)>;
2294def : Pat<(srl (bswap top32Zero:$Rn), (i64 32)), (REV32Xr GPR64:$Rn)>;
2295
2296def : Pat<(or (and (srl GPR64:$Rn, (i64 8)), (i64 0x00ff00ff00ff00ff)),
2297              (and (shl GPR64:$Rn, (i64 8)), (i64 0xff00ff00ff00ff00))),
2298          (REV16Xr GPR64:$Rn)>;
2299
2300//===----------------------------------------------------------------------===//
2301// Bitfield immediate extraction instruction.
2302//===----------------------------------------------------------------------===//
2303let hasSideEffects = 0 in
2304defm EXTR : ExtractImm<"extr">;
2305def : InstAlias<"ror $dst, $src, $shift",
2306            (EXTRWrri GPR32:$dst, GPR32:$src, GPR32:$src, imm0_31:$shift)>;
2307def : InstAlias<"ror $dst, $src, $shift",
2308            (EXTRXrri GPR64:$dst, GPR64:$src, GPR64:$src, imm0_63:$shift)>;
2309
2310def : Pat<(rotr GPR32:$Rn, (i64 imm0_31:$imm)),
2311          (EXTRWrri GPR32:$Rn, GPR32:$Rn, imm0_31:$imm)>;
2312def : Pat<(rotr GPR64:$Rn, (i64 imm0_63:$imm)),
2313          (EXTRXrri GPR64:$Rn, GPR64:$Rn, imm0_63:$imm)>;
2314
2315//===----------------------------------------------------------------------===//
2316// Other bitfield immediate instructions.
2317//===----------------------------------------------------------------------===//
2318let hasSideEffects = 0 in {
2319defm BFM  : BitfieldImmWith2RegArgs<0b01, "bfm">;
2320defm SBFM : BitfieldImm<0b00, "sbfm">;
2321defm UBFM : BitfieldImm<0b10, "ubfm">;
2322}
2323
2324def i32shift_a : Operand<i64>, SDNodeXForm<imm, [{
2325  uint64_t enc = (32 - N->getZExtValue()) & 0x1f;
2326  return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
2327}]>;
2328
2329def i32shift_b : Operand<i64>, SDNodeXForm<imm, [{
2330  uint64_t enc = 31 - N->getZExtValue();
2331  return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
2332}]>;
2333
2334// min(7, 31 - shift_amt)
2335def i32shift_sext_i8 : Operand<i64>, SDNodeXForm<imm, [{
2336  uint64_t enc = 31 - N->getZExtValue();
2337  enc = enc > 7 ? 7 : enc;
2338  return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
2339}]>;
2340
2341// min(15, 31 - shift_amt)
2342def i32shift_sext_i16 : Operand<i64>, SDNodeXForm<imm, [{
2343  uint64_t enc = 31 - N->getZExtValue();
2344  enc = enc > 15 ? 15 : enc;
2345  return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
2346}]>;
2347
2348def i64shift_a : Operand<i64>, SDNodeXForm<imm, [{
2349  uint64_t enc = (64 - N->getZExtValue()) & 0x3f;
2350  return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
2351}]>;
2352
2353def i64shift_b : Operand<i64>, SDNodeXForm<imm, [{
2354  uint64_t enc = 63 - N->getZExtValue();
2355  return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
2356}]>;
2357
2358// min(7, 63 - shift_amt)
2359def i64shift_sext_i8 : Operand<i64>, SDNodeXForm<imm, [{
2360  uint64_t enc = 63 - N->getZExtValue();
2361  enc = enc > 7 ? 7 : enc;
2362  return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
2363}]>;
2364
2365// min(15, 63 - shift_amt)
2366def i64shift_sext_i16 : Operand<i64>, SDNodeXForm<imm, [{
2367  uint64_t enc = 63 - N->getZExtValue();
2368  enc = enc > 15 ? 15 : enc;
2369  return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
2370}]>;
2371
2372// min(31, 63 - shift_amt)
2373def i64shift_sext_i32 : Operand<i64>, SDNodeXForm<imm, [{
2374  uint64_t enc = 63 - N->getZExtValue();
2375  enc = enc > 31 ? 31 : enc;
2376  return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
2377}]>;
2378
2379def : Pat<(shl GPR32:$Rn, (i64 imm0_31:$imm)),
2380          (UBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
2381                              (i64 (i32shift_b imm0_31:$imm)))>;
2382def : Pat<(shl GPR64:$Rn, (i64 imm0_63:$imm)),
2383          (UBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
2384                              (i64 (i64shift_b imm0_63:$imm)))>;
2385
2386let AddedComplexity = 10 in {
2387def : Pat<(sra GPR32:$Rn, (i64 imm0_31:$imm)),
2388          (SBFMWri GPR32:$Rn, imm0_31:$imm, 31)>;
2389def : Pat<(sra GPR64:$Rn, (i64 imm0_63:$imm)),
2390          (SBFMXri GPR64:$Rn, imm0_63:$imm, 63)>;
2391}
2392
2393def : InstAlias<"asr $dst, $src, $shift",
2394                (SBFMWri GPR32:$dst, GPR32:$src, imm0_31:$shift, 31)>;
2395def : InstAlias<"asr $dst, $src, $shift",
2396                (SBFMXri GPR64:$dst, GPR64:$src, imm0_63:$shift, 63)>;
2397def : InstAlias<"sxtb $dst, $src", (SBFMWri GPR32:$dst, GPR32:$src, 0, 7)>;
2398def : InstAlias<"sxtb $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 7)>;
2399def : InstAlias<"sxth $dst, $src", (SBFMWri GPR32:$dst, GPR32:$src, 0, 15)>;
2400def : InstAlias<"sxth $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 15)>;
2401def : InstAlias<"sxtw $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 31)>;
2402
2403def : Pat<(srl GPR32:$Rn, (i64 imm0_31:$imm)),
2404          (UBFMWri GPR32:$Rn, imm0_31:$imm, 31)>;
2405def : Pat<(srl GPR64:$Rn, (i64 imm0_63:$imm)),
2406          (UBFMXri GPR64:$Rn, imm0_63:$imm, 63)>;
2407
2408def : InstAlias<"lsr $dst, $src, $shift",
2409                (UBFMWri GPR32:$dst, GPR32:$src, imm0_31:$shift, 31)>;
2410def : InstAlias<"lsr $dst, $src, $shift",
2411                (UBFMXri GPR64:$dst, GPR64:$src, imm0_63:$shift, 63)>;
2412def : InstAlias<"uxtb $dst, $src", (UBFMWri GPR32:$dst, GPR32:$src, 0, 7)>;
2413def : InstAlias<"uxtb $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 7)>;
2414def : InstAlias<"uxth $dst, $src", (UBFMWri GPR32:$dst, GPR32:$src, 0, 15)>;
2415def : InstAlias<"uxth $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 15)>;
2416def : InstAlias<"uxtw $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 31)>;
2417
2418//===----------------------------------------------------------------------===//
2419// Conditional comparison instructions.
2420//===----------------------------------------------------------------------===//
2421defm CCMN : CondComparison<0, "ccmn", AArch64ccmn>;
2422defm CCMP : CondComparison<1, "ccmp", AArch64ccmp>;
2423
2424//===----------------------------------------------------------------------===//
2425// Conditional select instructions.
2426//===----------------------------------------------------------------------===//
2427defm CSEL  : CondSelect<0, 0b00, "csel">;
2428
2429def inc : PatFrag<(ops node:$in), (add node:$in, 1)>;
2430defm CSINC : CondSelectOp<0, 0b01, "csinc", inc>;
2431defm CSINV : CondSelectOp<1, 0b00, "csinv", not>;
2432defm CSNEG : CondSelectOp<1, 0b01, "csneg", ineg>;
2433
2434def : Pat<(AArch64csinv GPR32:$tval, GPR32:$fval, (i32 imm:$cc), NZCV),
2435          (CSINVWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
2436def : Pat<(AArch64csinv GPR64:$tval, GPR64:$fval, (i32 imm:$cc), NZCV),
2437          (CSINVXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
2438def : Pat<(AArch64csneg GPR32:$tval, GPR32:$fval, (i32 imm:$cc), NZCV),
2439          (CSNEGWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
2440def : Pat<(AArch64csneg GPR64:$tval, GPR64:$fval, (i32 imm:$cc), NZCV),
2441          (CSNEGXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
2442def : Pat<(AArch64csinc GPR32:$tval, GPR32:$fval, (i32 imm:$cc), NZCV),
2443          (CSINCWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
2444def : Pat<(AArch64csinc GPR64:$tval, GPR64:$fval, (i32 imm:$cc), NZCV),
2445          (CSINCXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
2446
2447def : Pat<(AArch64csel (i32 0), (i32 1), (i32 imm:$cc), NZCV),
2448          (CSINCWr WZR, WZR, (i32 imm:$cc))>;
2449def : Pat<(AArch64csel (i64 0), (i64 1), (i32 imm:$cc), NZCV),
2450          (CSINCXr XZR, XZR, (i32 imm:$cc))>;
2451def : Pat<(AArch64csel GPR32:$tval, (i32 1), (i32 imm:$cc), NZCV),
2452          (CSINCWr GPR32:$tval, WZR, (i32 imm:$cc))>;
2453def : Pat<(AArch64csel GPR64:$tval, (i64 1), (i32 imm:$cc), NZCV),
2454          (CSINCXr GPR64:$tval, XZR, (i32 imm:$cc))>;
2455def : Pat<(AArch64csel (i32 1), GPR32:$fval, (i32 imm:$cc), NZCV),
2456          (CSINCWr GPR32:$fval, WZR, (i32 (inv_cond_XFORM imm:$cc)))>;
2457def : Pat<(AArch64csel (i64 1), GPR64:$fval, (i32 imm:$cc), NZCV),
2458          (CSINCXr GPR64:$fval, XZR, (i32 (inv_cond_XFORM imm:$cc)))>;
2459def : Pat<(AArch64csel (i32 0), (i32 -1), (i32 imm:$cc), NZCV),
2460          (CSINVWr WZR, WZR, (i32 imm:$cc))>;
2461def : Pat<(AArch64csel (i64 0), (i64 -1), (i32 imm:$cc), NZCV),
2462          (CSINVXr XZR, XZR, (i32 imm:$cc))>;
2463def : Pat<(AArch64csel GPR32:$tval, (i32 -1), (i32 imm:$cc), NZCV),
2464          (CSINVWr GPR32:$tval, WZR, (i32 imm:$cc))>;
2465def : Pat<(AArch64csel GPR64:$tval, (i64 -1), (i32 imm:$cc), NZCV),
2466          (CSINVXr GPR64:$tval, XZR, (i32 imm:$cc))>;
2467def : Pat<(AArch64csel (i32 -1), GPR32:$fval, (i32 imm:$cc), NZCV),
2468          (CSINVWr GPR32:$fval, WZR, (i32 (inv_cond_XFORM imm:$cc)))>;
2469def : Pat<(AArch64csel (i64 -1), GPR64:$fval, (i32 imm:$cc), NZCV),
2470          (CSINVXr GPR64:$fval, XZR, (i32 (inv_cond_XFORM imm:$cc)))>;
2471
2472def : Pat<(add GPR32:$val, (AArch64csel (i32 0), (i32 1), (i32 imm:$cc), NZCV)),
2473          (CSINCWr GPR32:$val, GPR32:$val, (i32 imm:$cc))>;
2474def : Pat<(add GPR64:$val, (zext (AArch64csel (i32 0), (i32 1), (i32 imm:$cc), NZCV))),
2475          (CSINCXr GPR64:$val, GPR64:$val, (i32 imm:$cc))>;
2476
2477def : Pat<(or (topbitsallzero32:$val), (AArch64csel (i32 0), (i32 1), (i32 imm:$cc), NZCV)),
2478          (CSINCWr GPR32:$val, WZR, imm:$cc)>;
2479def : Pat<(or (topbitsallzero64:$val), (AArch64csel (i64 0), (i64 1), (i32 imm:$cc), NZCV)),
2480          (CSINCXr GPR64:$val, XZR, imm:$cc)>;
2481def : Pat<(or (topbitsallzero64:$val), (zext (AArch64csel (i32 0), (i32 1), (i32 imm:$cc), NZCV))),
2482          (CSINCXr GPR64:$val, XZR, imm:$cc)>;
2483
2484def : Pat<(and (topbitsallzero32:$val), (AArch64csel (i32 0), (i32 1), (i32 imm:$cc), NZCV)),
2485          (CSELWr WZR, GPR32:$val, imm:$cc)>;
2486def : Pat<(and (topbitsallzero64:$val), (AArch64csel (i64 0), (i64 1), (i32 imm:$cc), NZCV)),
2487          (CSELXr XZR, GPR64:$val, imm:$cc)>;
2488def : Pat<(and (topbitsallzero64:$val), (zext (AArch64csel (i32 0), (i32 1), (i32 imm:$cc), NZCV))),
2489          (CSELXr XZR, GPR64:$val, imm:$cc)>;
2490
2491// The inverse of the condition code from the alias instruction is what is used
2492// in the aliased instruction. The parser all ready inverts the condition code
2493// for these aliases.
2494def : InstAlias<"cset $dst, $cc",
2495                (CSINCWr GPR32:$dst, WZR, WZR, inv_ccode:$cc)>;
2496def : InstAlias<"cset $dst, $cc",
2497                (CSINCXr GPR64:$dst, XZR, XZR, inv_ccode:$cc)>;
2498
2499def : InstAlias<"csetm $dst, $cc",
2500                (CSINVWr GPR32:$dst, WZR, WZR, inv_ccode:$cc)>;
2501def : InstAlias<"csetm $dst, $cc",
2502                (CSINVXr GPR64:$dst, XZR, XZR, inv_ccode:$cc)>;
2503
2504def : InstAlias<"cinc $dst, $src, $cc",
2505                (CSINCWr GPR32:$dst, GPR32:$src, GPR32:$src, inv_ccode:$cc)>;
2506def : InstAlias<"cinc $dst, $src, $cc",
2507                (CSINCXr GPR64:$dst, GPR64:$src, GPR64:$src, inv_ccode:$cc)>;
2508
2509def : InstAlias<"cinv $dst, $src, $cc",
2510                (CSINVWr GPR32:$dst, GPR32:$src, GPR32:$src, inv_ccode:$cc)>;
2511def : InstAlias<"cinv $dst, $src, $cc",
2512                (CSINVXr GPR64:$dst, GPR64:$src, GPR64:$src, inv_ccode:$cc)>;
2513
2514def : InstAlias<"cneg $dst, $src, $cc",
2515                (CSNEGWr GPR32:$dst, GPR32:$src, GPR32:$src, inv_ccode:$cc)>;
2516def : InstAlias<"cneg $dst, $src, $cc",
2517                (CSNEGXr GPR64:$dst, GPR64:$src, GPR64:$src, inv_ccode:$cc)>;
2518
2519//===----------------------------------------------------------------------===//
2520// PC-relative instructions.
2521//===----------------------------------------------------------------------===//
2522let isReMaterializable = 1 in {
2523let hasSideEffects = 0, mayStore = 0, mayLoad = 0 in {
2524def ADR  : ADRI<0, "adr", adrlabel,
2525                [(set GPR64:$Xd, (AArch64adr tglobaladdr:$label))]>;
2526} // hasSideEffects = 0
2527
2528def ADRP : ADRI<1, "adrp", adrplabel,
2529                [(set GPR64:$Xd, (AArch64adrp tglobaladdr:$label))]>;
2530} // isReMaterializable = 1
2531
2532// page address of a constant pool entry, block address
2533def : Pat<(AArch64adr tconstpool:$cp), (ADR tconstpool:$cp)>;
2534def : Pat<(AArch64adr tblockaddress:$cp), (ADR tblockaddress:$cp)>;
2535def : Pat<(AArch64adr texternalsym:$sym), (ADR texternalsym:$sym)>;
2536def : Pat<(AArch64adr tjumptable:$sym), (ADR tjumptable:$sym)>;
2537def : Pat<(AArch64adrp tconstpool:$cp), (ADRP tconstpool:$cp)>;
2538def : Pat<(AArch64adrp tblockaddress:$cp), (ADRP tblockaddress:$cp)>;
2539def : Pat<(AArch64adrp texternalsym:$sym), (ADRP texternalsym:$sym)>;
2540
2541//===----------------------------------------------------------------------===//
2542// Unconditional branch (register) instructions.
2543//===----------------------------------------------------------------------===//
2544
2545let isReturn = 1, isTerminator = 1, isBarrier = 1 in {
2546def RET  : BranchReg<0b0010, "ret", []>;
2547def DRPS : SpecialReturn<0b0101, "drps">;
2548def ERET : SpecialReturn<0b0100, "eret">;
2549} // isReturn = 1, isTerminator = 1, isBarrier = 1
2550
2551// Default to the LR register.
2552def : InstAlias<"ret", (RET LR)>;
2553
2554let isCall = 1, Defs = [LR], Uses = [SP] in {
2555  def BLR : BranchReg<0b0001, "blr", []>;
2556  def BLRNoIP : Pseudo<(outs), (ins GPR64noip:$Rn), []>,
2557                Sched<[WriteBrReg]>,
2558                PseudoInstExpansion<(BLR GPR64:$Rn)>;
2559  def BLR_RVMARKER : Pseudo<(outs), (ins variable_ops), []>,
2560                     Sched<[WriteBrReg]>;
2561  def BLR_BTI : Pseudo<(outs), (ins variable_ops), []>,
2562                Sched<[WriteBrReg]>;
2563} // isCall
2564
2565def : Pat<(AArch64call GPR64:$Rn),
2566          (BLR GPR64:$Rn)>,
2567      Requires<[NoSLSBLRMitigation]>;
2568def : Pat<(AArch64call GPR64noip:$Rn),
2569          (BLRNoIP GPR64noip:$Rn)>,
2570      Requires<[SLSBLRMitigation]>;
2571
2572def : Pat<(AArch64call_rvmarker (i64 tglobaladdr:$rvfunc), GPR64:$Rn),
2573          (BLR_RVMARKER tglobaladdr:$rvfunc, GPR64:$Rn)>,
2574      Requires<[NoSLSBLRMitigation]>;
2575
2576def : Pat<(AArch64call_bti GPR64:$Rn),
2577          (BLR_BTI GPR64:$Rn)>,
2578      Requires<[NoSLSBLRMitigation]>;
2579
2580let isBranch = 1, isTerminator = 1, isBarrier = 1, isIndirectBranch = 1 in {
2581def BR  : BranchReg<0b0000, "br", [(brind GPR64:$Rn)]>;
2582} // isBranch, isTerminator, isBarrier, isIndirectBranch
2583
2584// Create a separate pseudo-instruction for codegen to use so that we don't
2585// flag lr as used in every function. It'll be restored before the RET by the
2586// epilogue if it's legitimately used.
2587def RET_ReallyLR : Pseudo<(outs), (ins), [(AArch64retflag)]>,
2588                   Sched<[WriteBrReg]> {
2589  let isTerminator = 1;
2590  let isBarrier = 1;
2591  let isReturn = 1;
2592}
2593
2594// This is a directive-like pseudo-instruction. The purpose is to insert an
2595// R_AARCH64_TLSDESC_CALL relocation at the offset of the following instruction
2596// (which in the usual case is a BLR).
2597let hasSideEffects = 1 in
2598def TLSDESCCALL : Pseudo<(outs), (ins i64imm:$sym), []>, Sched<[]> {
2599  let AsmString = ".tlsdesccall $sym";
2600}
2601
2602// Pseudo instruction to tell the streamer to emit a 'B' character into the
2603// augmentation string.
2604def EMITBKEY : Pseudo<(outs), (ins), []>, Sched<[]> {}
2605
2606// Pseudo instruction to tell the streamer to emit a 'G' character into the
2607// augmentation string.
2608def EMITMTETAGGED : Pseudo<(outs), (ins), []>, Sched<[]> {}
2609
2610// FIXME: maybe the scratch register used shouldn't be fixed to X1?
2611// FIXME: can "hasSideEffects be dropped?
2612// This gets lowered to an instruction sequence which takes 16 bytes
2613let isCall = 1, Defs = [LR, X0, X1], hasSideEffects = 1, Size = 16,
2614    isCodeGenOnly = 1 in
2615def TLSDESC_CALLSEQ
2616    : Pseudo<(outs), (ins i64imm:$sym),
2617             [(AArch64tlsdesc_callseq tglobaltlsaddr:$sym)]>,
2618      Sched<[WriteI, WriteLD, WriteI, WriteBrReg]>;
2619def : Pat<(AArch64tlsdesc_callseq texternalsym:$sym),
2620          (TLSDESC_CALLSEQ texternalsym:$sym)>;
2621
2622//===----------------------------------------------------------------------===//
2623// Conditional branch (immediate) instruction.
2624//===----------------------------------------------------------------------===//
2625def Bcc : BranchCond<0, "b">;
2626
2627// Armv8.8-A variant form which hints to the branch predictor that
2628// this branch is very likely to go the same way nearly all the time
2629// (even though it is not known at compile time _which_ way that is).
2630def BCcc : BranchCond<1, "bc">, Requires<[HasHBC]>;
2631
2632//===----------------------------------------------------------------------===//
2633// Compare-and-branch instructions.
2634//===----------------------------------------------------------------------===//
2635defm CBZ  : CmpBranch<0, "cbz", AArch64cbz>;
2636defm CBNZ : CmpBranch<1, "cbnz", AArch64cbnz>;
2637
2638//===----------------------------------------------------------------------===//
2639// Test-bit-and-branch instructions.
2640//===----------------------------------------------------------------------===//
2641defm TBZ  : TestBranch<0, "tbz", AArch64tbz>;
2642defm TBNZ : TestBranch<1, "tbnz", AArch64tbnz>;
2643
2644//===----------------------------------------------------------------------===//
2645// Unconditional branch (immediate) instructions.
2646//===----------------------------------------------------------------------===//
2647let isBranch = 1, isTerminator = 1, isBarrier = 1 in {
2648def B  : BranchImm<0, "b", [(br bb:$addr)]>;
2649} // isBranch, isTerminator, isBarrier
2650
2651let isCall = 1, Defs = [LR], Uses = [SP] in {
2652def BL : CallImm<1, "bl", [(AArch64call tglobaladdr:$addr)]>;
2653} // isCall
2654def : Pat<(AArch64call texternalsym:$func), (BL texternalsym:$func)>;
2655
2656//===----------------------------------------------------------------------===//
2657// Exception generation instructions.
2658//===----------------------------------------------------------------------===//
2659let isTrap = 1 in {
2660def BRK   : ExceptionGeneration<0b001, 0b00, "brk",
2661                                [(int_aarch64_break timm32_0_65535:$imm)]>;
2662}
2663def DCPS1 : ExceptionGeneration<0b101, 0b01, "dcps1">;
2664def DCPS2 : ExceptionGeneration<0b101, 0b10, "dcps2">;
2665def DCPS3 : ExceptionGeneration<0b101, 0b11, "dcps3">, Requires<[HasEL3]>;
2666def HLT   : ExceptionGeneration<0b010, 0b00, "hlt">;
2667def HVC   : ExceptionGeneration<0b000, 0b10, "hvc">;
2668def SMC   : ExceptionGeneration<0b000, 0b11, "smc">, Requires<[HasEL3]>;
2669def SVC   : ExceptionGeneration<0b000, 0b01, "svc">;
2670
2671// DCPSn defaults to an immediate operand of zero if unspecified.
2672def : InstAlias<"dcps1", (DCPS1 0)>;
2673def : InstAlias<"dcps2", (DCPS2 0)>;
2674def : InstAlias<"dcps3", (DCPS3 0)>, Requires<[HasEL3]>;
2675
2676def UDF : UDFType<0, "udf">;
2677
2678//===----------------------------------------------------------------------===//
2679// Load instructions.
2680//===----------------------------------------------------------------------===//
2681
2682// Pair (indexed, offset)
2683defm LDPW : LoadPairOffset<0b00, 0, GPR32z, simm7s4, "ldp">;
2684defm LDPX : LoadPairOffset<0b10, 0, GPR64z, simm7s8, "ldp">;
2685defm LDPS : LoadPairOffset<0b00, 1, FPR32Op, simm7s4, "ldp">;
2686defm LDPD : LoadPairOffset<0b01, 1, FPR64Op, simm7s8, "ldp">;
2687defm LDPQ : LoadPairOffset<0b10, 1, FPR128Op, simm7s16, "ldp">;
2688
2689defm LDPSW : LoadPairOffset<0b01, 0, GPR64z, simm7s4, "ldpsw">;
2690
2691// Pair (pre-indexed)
2692def LDPWpre : LoadPairPreIdx<0b00, 0, GPR32z, simm7s4, "ldp">;
2693def LDPXpre : LoadPairPreIdx<0b10, 0, GPR64z, simm7s8, "ldp">;
2694def LDPSpre : LoadPairPreIdx<0b00, 1, FPR32Op, simm7s4, "ldp">;
2695def LDPDpre : LoadPairPreIdx<0b01, 1, FPR64Op, simm7s8, "ldp">;
2696def LDPQpre : LoadPairPreIdx<0b10, 1, FPR128Op, simm7s16, "ldp">;
2697
2698def LDPSWpre : LoadPairPreIdx<0b01, 0, GPR64z, simm7s4, "ldpsw">;
2699
2700// Pair (post-indexed)
2701def LDPWpost : LoadPairPostIdx<0b00, 0, GPR32z, simm7s4, "ldp">;
2702def LDPXpost : LoadPairPostIdx<0b10, 0, GPR64z, simm7s8, "ldp">;
2703def LDPSpost : LoadPairPostIdx<0b00, 1, FPR32Op, simm7s4, "ldp">;
2704def LDPDpost : LoadPairPostIdx<0b01, 1, FPR64Op, simm7s8, "ldp">;
2705def LDPQpost : LoadPairPostIdx<0b10, 1, FPR128Op, simm7s16, "ldp">;
2706
2707def LDPSWpost : LoadPairPostIdx<0b01, 0, GPR64z, simm7s4, "ldpsw">;
2708
2709
2710// Pair (no allocate)
2711defm LDNPW : LoadPairNoAlloc<0b00, 0, GPR32z, simm7s4, "ldnp">;
2712defm LDNPX : LoadPairNoAlloc<0b10, 0, GPR64z, simm7s8, "ldnp">;
2713defm LDNPS : LoadPairNoAlloc<0b00, 1, FPR32Op, simm7s4, "ldnp">;
2714defm LDNPD : LoadPairNoAlloc<0b01, 1, FPR64Op, simm7s8, "ldnp">;
2715defm LDNPQ : LoadPairNoAlloc<0b10, 1, FPR128Op, simm7s16, "ldnp">;
2716
2717def : Pat<(AArch64ldp (am_indexed7s64 GPR64sp:$Rn, simm7s8:$offset)),
2718          (LDPXi GPR64sp:$Rn, simm7s8:$offset)>;
2719
2720def : Pat<(AArch64ldnp (am_indexed7s128 GPR64sp:$Rn, simm7s16:$offset)),
2721          (LDNPQi GPR64sp:$Rn, simm7s16:$offset)>;
2722//---
2723// (register offset)
2724//---
2725
2726// Integer
2727defm LDRBB : Load8RO<0b00,  0, 0b01, GPR32, "ldrb", i32, zextloadi8>;
2728defm LDRHH : Load16RO<0b01, 0, 0b01, GPR32, "ldrh", i32, zextloadi16>;
2729defm LDRW  : Load32RO<0b10, 0, 0b01, GPR32, "ldr", i32, load>;
2730defm LDRX  : Load64RO<0b11, 0, 0b01, GPR64, "ldr", i64, load>;
2731
2732// Floating-point
2733defm LDRB : Load8RO<0b00,   1, 0b01, FPR8Op,   "ldr", untyped, load>;
2734defm LDRH : Load16RO<0b01,  1, 0b01, FPR16Op,  "ldr", f16, load>;
2735defm LDRS : Load32RO<0b10,  1, 0b01, FPR32Op,  "ldr", f32, load>;
2736defm LDRD : Load64RO<0b11,  1, 0b01, FPR64Op,  "ldr", f64, load>;
2737defm LDRQ : Load128RO<0b00, 1, 0b11, FPR128Op, "ldr", f128, load>;
2738
2739// Load sign-extended half-word
2740defm LDRSHW : Load16RO<0b01, 0, 0b11, GPR32, "ldrsh", i32, sextloadi16>;
2741defm LDRSHX : Load16RO<0b01, 0, 0b10, GPR64, "ldrsh", i64, sextloadi16>;
2742
2743// Load sign-extended byte
2744defm LDRSBW : Load8RO<0b00, 0, 0b11, GPR32, "ldrsb", i32, sextloadi8>;
2745defm LDRSBX : Load8RO<0b00, 0, 0b10, GPR64, "ldrsb", i64, sextloadi8>;
2746
2747// Load sign-extended word
2748defm LDRSW  : Load32RO<0b10, 0, 0b10, GPR64, "ldrsw", i64, sextloadi32>;
2749
2750// Pre-fetch.
2751defm PRFM : PrefetchRO<0b11, 0, 0b10, "prfm">;
2752
2753// For regular load, we do not have any alignment requirement.
2754// Thus, it is safe to directly map the vector loads with interesting
2755// addressing modes.
2756// FIXME: We could do the same for bitconvert to floating point vectors.
2757multiclass ScalToVecROLoadPat<ROAddrMode ro, SDPatternOperator loadop,
2758                              ValueType ScalTy, ValueType VecTy,
2759                              Instruction LOADW, Instruction LOADX,
2760                              SubRegIndex sub> {
2761  def : Pat<(VecTy (scalar_to_vector (ScalTy
2762              (loadop (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$offset))))),
2763            (INSERT_SUBREG (VecTy (IMPLICIT_DEF)),
2764                           (LOADW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$offset),
2765                           sub)>;
2766
2767  def : Pat<(VecTy (scalar_to_vector (ScalTy
2768              (loadop (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$offset))))),
2769            (INSERT_SUBREG (VecTy (IMPLICIT_DEF)),
2770                           (LOADX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$offset),
2771                           sub)>;
2772}
2773
2774let AddedComplexity = 10 in {
2775defm : ScalToVecROLoadPat<ro8,  extloadi8,  i32, v8i8,  LDRBroW, LDRBroX, bsub>;
2776defm : ScalToVecROLoadPat<ro8,  extloadi8,  i32, v16i8, LDRBroW, LDRBroX, bsub>;
2777
2778defm : ScalToVecROLoadPat<ro16, extloadi16, i32, v4i16, LDRHroW, LDRHroX, hsub>;
2779defm : ScalToVecROLoadPat<ro16, extloadi16, i32, v8i16, LDRHroW, LDRHroX, hsub>;
2780
2781defm : ScalToVecROLoadPat<ro16, load,       i32, v4f16, LDRHroW, LDRHroX, hsub>;
2782defm : ScalToVecROLoadPat<ro16, load,       i32, v8f16, LDRHroW, LDRHroX, hsub>;
2783
2784defm : ScalToVecROLoadPat<ro32, load,       i32, v2i32, LDRSroW, LDRSroX, ssub>;
2785defm : ScalToVecROLoadPat<ro32, load,       i32, v4i32, LDRSroW, LDRSroX, ssub>;
2786
2787defm : ScalToVecROLoadPat<ro32, load,       f32, v2f32, LDRSroW, LDRSroX, ssub>;
2788defm : ScalToVecROLoadPat<ro32, load,       f32, v4f32, LDRSroW, LDRSroX, ssub>;
2789
2790defm : ScalToVecROLoadPat<ro64, load,       i64, v2i64, LDRDroW, LDRDroX, dsub>;
2791
2792defm : ScalToVecROLoadPat<ro64, load,       f64, v2f64, LDRDroW, LDRDroX, dsub>;
2793
2794
2795def : Pat <(v1i64 (scalar_to_vector (i64
2796                      (load (ro_Windexed64 GPR64sp:$Rn, GPR32:$Rm,
2797                                           ro_Wextend64:$extend))))),
2798           (LDRDroW GPR64sp:$Rn, GPR32:$Rm, ro_Wextend64:$extend)>;
2799
2800def : Pat <(v1i64 (scalar_to_vector (i64
2801                      (load (ro_Xindexed64 GPR64sp:$Rn, GPR64:$Rm,
2802                                           ro_Xextend64:$extend))))),
2803           (LDRDroX GPR64sp:$Rn, GPR64:$Rm, ro_Xextend64:$extend)>;
2804}
2805
2806// Match all load 64 bits width whose type is compatible with FPR64
2807multiclass VecROLoadPat<ROAddrMode ro, ValueType VecTy,
2808                        Instruction LOADW, Instruction LOADX> {
2809
2810  def : Pat<(VecTy (load (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend))),
2811            (LOADW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
2812
2813  def : Pat<(VecTy (load (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend))),
2814            (LOADX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
2815}
2816
2817let AddedComplexity = 10 in {
2818let Predicates = [IsLE] in {
2819  // We must do vector loads with LD1 in big-endian.
2820  defm : VecROLoadPat<ro64, v2i32, LDRDroW, LDRDroX>;
2821  defm : VecROLoadPat<ro64, v2f32, LDRDroW, LDRDroX>;
2822  defm : VecROLoadPat<ro64, v8i8,  LDRDroW, LDRDroX>;
2823  defm : VecROLoadPat<ro64, v4i16, LDRDroW, LDRDroX>;
2824  defm : VecROLoadPat<ro64, v4f16, LDRDroW, LDRDroX>;
2825  defm : VecROLoadPat<ro64, v4bf16, LDRDroW, LDRDroX>;
2826}
2827
2828defm : VecROLoadPat<ro64, v1i64,  LDRDroW, LDRDroX>;
2829defm : VecROLoadPat<ro64, v1f64,  LDRDroW, LDRDroX>;
2830
2831// Match all load 128 bits width whose type is compatible with FPR128
2832let Predicates = [IsLE] in {
2833  // We must do vector loads with LD1 in big-endian.
2834  defm : VecROLoadPat<ro128, v2i64,  LDRQroW, LDRQroX>;
2835  defm : VecROLoadPat<ro128, v2f64,  LDRQroW, LDRQroX>;
2836  defm : VecROLoadPat<ro128, v4i32,  LDRQroW, LDRQroX>;
2837  defm : VecROLoadPat<ro128, v4f32,  LDRQroW, LDRQroX>;
2838  defm : VecROLoadPat<ro128, v8i16,  LDRQroW, LDRQroX>;
2839  defm : VecROLoadPat<ro128, v8f16,  LDRQroW, LDRQroX>;
2840  defm : VecROLoadPat<ro128, v8bf16,  LDRQroW, LDRQroX>;
2841  defm : VecROLoadPat<ro128, v16i8,  LDRQroW, LDRQroX>;
2842}
2843} // AddedComplexity = 10
2844
2845// zextload -> i64
2846multiclass ExtLoadTo64ROPat<ROAddrMode ro, SDPatternOperator loadop,
2847                            Instruction INSTW, Instruction INSTX> {
2848  def : Pat<(i64 (loadop (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend))),
2849            (SUBREG_TO_REG (i64 0),
2850                           (INSTW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend),
2851                           sub_32)>;
2852
2853  def : Pat<(i64 (loadop (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend))),
2854            (SUBREG_TO_REG (i64 0),
2855                           (INSTX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend),
2856                           sub_32)>;
2857}
2858
2859let AddedComplexity = 10 in {
2860  defm : ExtLoadTo64ROPat<ro8,  zextloadi8,  LDRBBroW, LDRBBroX>;
2861  defm : ExtLoadTo64ROPat<ro16, zextloadi16, LDRHHroW, LDRHHroX>;
2862  defm : ExtLoadTo64ROPat<ro32, zextloadi32, LDRWroW,  LDRWroX>;
2863
2864  // zextloadi1 -> zextloadi8
2865  defm : ExtLoadTo64ROPat<ro8,  zextloadi1,  LDRBBroW, LDRBBroX>;
2866
2867  // extload -> zextload
2868  defm : ExtLoadTo64ROPat<ro8,  extloadi8,   LDRBBroW, LDRBBroX>;
2869  defm : ExtLoadTo64ROPat<ro16, extloadi16,  LDRHHroW, LDRHHroX>;
2870  defm : ExtLoadTo64ROPat<ro32, extloadi32,  LDRWroW,  LDRWroX>;
2871
2872  // extloadi1 -> zextloadi8
2873  defm : ExtLoadTo64ROPat<ro8,  extloadi1,   LDRBBroW, LDRBBroX>;
2874}
2875
2876
2877// zextload -> i64
2878multiclass ExtLoadTo32ROPat<ROAddrMode ro, SDPatternOperator loadop,
2879                            Instruction INSTW, Instruction INSTX> {
2880  def : Pat<(i32 (loadop (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend))),
2881            (INSTW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
2882
2883  def : Pat<(i32 (loadop (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend))),
2884            (INSTX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
2885
2886}
2887
2888let AddedComplexity = 10 in {
2889  // extload -> zextload
2890  defm : ExtLoadTo32ROPat<ro8,  extloadi8,   LDRBBroW, LDRBBroX>;
2891  defm : ExtLoadTo32ROPat<ro16, extloadi16,  LDRHHroW, LDRHHroX>;
2892  defm : ExtLoadTo32ROPat<ro32, extloadi32,  LDRWroW,  LDRWroX>;
2893
2894  // zextloadi1 -> zextloadi8
2895  defm : ExtLoadTo32ROPat<ro8, zextloadi1, LDRBBroW, LDRBBroX>;
2896}
2897
2898//---
2899// (unsigned immediate)
2900//---
2901defm LDRX : LoadUI<0b11, 0, 0b01, GPR64z, uimm12s8, "ldr",
2902                   [(set GPR64z:$Rt,
2903                         (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)))]>;
2904defm LDRW : LoadUI<0b10, 0, 0b01, GPR32z, uimm12s4, "ldr",
2905                   [(set GPR32z:$Rt,
2906                         (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset)))]>;
2907defm LDRB : LoadUI<0b00, 1, 0b01, FPR8Op, uimm12s1, "ldr",
2908                   [(set FPR8Op:$Rt,
2909                         (load (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset)))]>;
2910defm LDRH : LoadUI<0b01, 1, 0b01, FPR16Op, uimm12s2, "ldr",
2911                   [(set (f16 FPR16Op:$Rt),
2912                         (load (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset)))]>;
2913defm LDRS : LoadUI<0b10, 1, 0b01, FPR32Op, uimm12s4, "ldr",
2914                   [(set (f32 FPR32Op:$Rt),
2915                         (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset)))]>;
2916defm LDRD : LoadUI<0b11, 1, 0b01, FPR64Op, uimm12s8, "ldr",
2917                   [(set (f64 FPR64Op:$Rt),
2918                         (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)))]>;
2919defm LDRQ : LoadUI<0b00, 1, 0b11, FPR128Op, uimm12s16, "ldr",
2920                 [(set (f128 FPR128Op:$Rt),
2921                       (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)))]>;
2922
2923// bf16 load pattern
2924def : Pat <(bf16 (load (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))),
2925           (LDRHui GPR64sp:$Rn, uimm12s2:$offset)>;
2926
2927// For regular load, we do not have any alignment requirement.
2928// Thus, it is safe to directly map the vector loads with interesting
2929// addressing modes.
2930// FIXME: We could do the same for bitconvert to floating point vectors.
2931def : Pat <(v8i8 (scalar_to_vector (i32
2932               (extloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))))),
2933           (INSERT_SUBREG (v8i8 (IMPLICIT_DEF)),
2934                          (LDRBui GPR64sp:$Rn, uimm12s1:$offset), bsub)>;
2935def : Pat <(v16i8 (scalar_to_vector (i32
2936               (extloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))))),
2937           (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
2938                          (LDRBui GPR64sp:$Rn, uimm12s1:$offset), bsub)>;
2939def : Pat <(v4i16 (scalar_to_vector (i32
2940               (extloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))))),
2941           (INSERT_SUBREG (v4i16 (IMPLICIT_DEF)),
2942                          (LDRHui GPR64sp:$Rn, uimm12s2:$offset), hsub)>;
2943def : Pat <(v8i16 (scalar_to_vector (i32
2944               (extloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))))),
2945           (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
2946                          (LDRHui GPR64sp:$Rn, uimm12s2:$offset), hsub)>;
2947def : Pat <(v2i32 (scalar_to_vector (i32
2948               (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))))),
2949           (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)),
2950                          (LDRSui GPR64sp:$Rn, uimm12s4:$offset), ssub)>;
2951def : Pat <(v4i32 (scalar_to_vector (i32
2952               (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))))),
2953           (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
2954                          (LDRSui GPR64sp:$Rn, uimm12s4:$offset), ssub)>;
2955def : Pat <(v1i64 (scalar_to_vector (i64
2956               (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))))),
2957           (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
2958def : Pat <(v2i64 (scalar_to_vector (i64
2959               (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))))),
2960           (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
2961                          (LDRDui GPR64sp:$Rn, uimm12s8:$offset), dsub)>;
2962
2963// Match all load 64 bits width whose type is compatible with FPR64
2964let Predicates = [IsLE] in {
2965  // We must use LD1 to perform vector loads in big-endian.
2966  def : Pat<(v2f32 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
2967            (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
2968  def : Pat<(v8i8 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
2969            (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
2970  def : Pat<(v4i16 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
2971            (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
2972  def : Pat<(v2i32 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
2973            (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
2974  def : Pat<(v4f16 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
2975            (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
2976  def : Pat<(v4bf16 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
2977            (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
2978}
2979def : Pat<(v1f64 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
2980          (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
2981def : Pat<(v1i64 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
2982          (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
2983
2984// Match all load 128 bits width whose type is compatible with FPR128
2985let Predicates = [IsLE] in {
2986  // We must use LD1 to perform vector loads in big-endian.
2987  def : Pat<(v4f32 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
2988            (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
2989  def : Pat<(v2f64 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
2990            (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
2991  def : Pat<(v16i8 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
2992            (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
2993  def : Pat<(v8i16 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
2994            (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
2995  def : Pat<(v4i32 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
2996            (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
2997  def : Pat<(v2i64 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
2998            (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
2999  def : Pat<(v8f16 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
3000            (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
3001  def : Pat<(v8bf16 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
3002            (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
3003}
3004def : Pat<(f128  (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
3005          (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
3006
3007defm LDRHH : LoadUI<0b01, 0, 0b01, GPR32, uimm12s2, "ldrh",
3008                    [(set GPR32:$Rt,
3009                          (zextloadi16 (am_indexed16 GPR64sp:$Rn,
3010                                                     uimm12s2:$offset)))]>;
3011defm LDRBB : LoadUI<0b00, 0, 0b01, GPR32, uimm12s1, "ldrb",
3012                    [(set GPR32:$Rt,
3013                          (zextloadi8 (am_indexed8 GPR64sp:$Rn,
3014                                                   uimm12s1:$offset)))]>;
3015// zextload -> i64
3016def : Pat<(i64 (zextloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
3017    (SUBREG_TO_REG (i64 0), (LDRBBui GPR64sp:$Rn, uimm12s1:$offset), sub_32)>;
3018def : Pat<(i64 (zextloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))),
3019    (SUBREG_TO_REG (i64 0), (LDRHHui GPR64sp:$Rn, uimm12s2:$offset), sub_32)>;
3020
3021// zextloadi1 -> zextloadi8
3022def : Pat<(i32 (zextloadi1 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
3023          (LDRBBui GPR64sp:$Rn, uimm12s1:$offset)>;
3024def : Pat<(i64 (zextloadi1 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
3025    (SUBREG_TO_REG (i64 0), (LDRBBui GPR64sp:$Rn, uimm12s1:$offset), sub_32)>;
3026
3027// extload -> zextload
3028def : Pat<(i32 (extloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))),
3029          (LDRHHui GPR64sp:$Rn, uimm12s2:$offset)>;
3030def : Pat<(i32 (extloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
3031          (LDRBBui GPR64sp:$Rn, uimm12s1:$offset)>;
3032def : Pat<(i32 (extloadi1 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
3033          (LDRBBui GPR64sp:$Rn, uimm12s1:$offset)>;
3034def : Pat<(i64 (extloadi32 (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))),
3035    (SUBREG_TO_REG (i64 0), (LDRWui GPR64sp:$Rn, uimm12s4:$offset), sub_32)>;
3036def : Pat<(i64 (extloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))),
3037    (SUBREG_TO_REG (i64 0), (LDRHHui GPR64sp:$Rn, uimm12s2:$offset), sub_32)>;
3038def : Pat<(i64 (extloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
3039    (SUBREG_TO_REG (i64 0), (LDRBBui GPR64sp:$Rn, uimm12s1:$offset), sub_32)>;
3040def : Pat<(i64 (extloadi1 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
3041    (SUBREG_TO_REG (i64 0), (LDRBBui GPR64sp:$Rn, uimm12s1:$offset), sub_32)>;
3042
3043// load sign-extended half-word
3044defm LDRSHW : LoadUI<0b01, 0, 0b11, GPR32, uimm12s2, "ldrsh",
3045                     [(set GPR32:$Rt,
3046                           (sextloadi16 (am_indexed16 GPR64sp:$Rn,
3047                                                      uimm12s2:$offset)))]>;
3048defm LDRSHX : LoadUI<0b01, 0, 0b10, GPR64, uimm12s2, "ldrsh",
3049                     [(set GPR64:$Rt,
3050                           (sextloadi16 (am_indexed16 GPR64sp:$Rn,
3051                                                      uimm12s2:$offset)))]>;
3052
3053// load sign-extended byte
3054defm LDRSBW : LoadUI<0b00, 0, 0b11, GPR32, uimm12s1, "ldrsb",
3055                     [(set GPR32:$Rt,
3056                           (sextloadi8 (am_indexed8 GPR64sp:$Rn,
3057                                                    uimm12s1:$offset)))]>;
3058defm LDRSBX : LoadUI<0b00, 0, 0b10, GPR64, uimm12s1, "ldrsb",
3059                     [(set GPR64:$Rt,
3060                           (sextloadi8 (am_indexed8 GPR64sp:$Rn,
3061                                                    uimm12s1:$offset)))]>;
3062
3063// load sign-extended word
3064defm LDRSW  : LoadUI<0b10, 0, 0b10, GPR64, uimm12s4, "ldrsw",
3065                     [(set GPR64:$Rt,
3066                           (sextloadi32 (am_indexed32 GPR64sp:$Rn,
3067                                                      uimm12s4:$offset)))]>;
3068
3069// load zero-extended word
3070def : Pat<(i64 (zextloadi32 (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))),
3071      (SUBREG_TO_REG (i64 0), (LDRWui GPR64sp:$Rn, uimm12s4:$offset), sub_32)>;
3072
3073// Pre-fetch.
3074def PRFMui : PrefetchUI<0b11, 0, 0b10, "prfm",
3075                        [(AArch64Prefetch timm:$Rt,
3076                                        (am_indexed64 GPR64sp:$Rn,
3077                                                      uimm12s8:$offset))]>;
3078
3079def : InstAlias<"prfm $Rt, [$Rn]", (PRFMui prfop:$Rt, GPR64sp:$Rn, 0)>;
3080
3081//---
3082// (literal)
3083
3084def alignedglobal : PatLeaf<(iPTR iPTR:$label), [{
3085  if (auto *G = dyn_cast<GlobalAddressSDNode>(N)) {
3086    const DataLayout &DL = MF->getDataLayout();
3087    Align Align = G->getGlobal()->getPointerAlignment(DL);
3088    return Align >= 4 && G->getOffset() % 4 == 0;
3089  }
3090  if (auto *C = dyn_cast<ConstantPoolSDNode>(N))
3091    return C->getAlign() >= 4 && C->getOffset() % 4 == 0;
3092  return false;
3093}]>;
3094
3095def LDRWl : LoadLiteral<0b00, 0, GPR32z, "ldr",
3096  [(set GPR32z:$Rt, (load (AArch64adr alignedglobal:$label)))]>;
3097def LDRXl : LoadLiteral<0b01, 0, GPR64z, "ldr",
3098  [(set GPR64z:$Rt, (load (AArch64adr alignedglobal:$label)))]>;
3099def LDRSl : LoadLiteral<0b00, 1, FPR32Op, "ldr",
3100  [(set (f32 FPR32Op:$Rt), (load (AArch64adr alignedglobal:$label)))]>;
3101def LDRDl : LoadLiteral<0b01, 1, FPR64Op, "ldr",
3102  [(set (f64 FPR64Op:$Rt), (load (AArch64adr alignedglobal:$label)))]>;
3103def LDRQl : LoadLiteral<0b10, 1, FPR128Op, "ldr",
3104  [(set (f128 FPR128Op:$Rt), (load (AArch64adr alignedglobal:$label)))]>;
3105
3106// load sign-extended word
3107def LDRSWl : LoadLiteral<0b10, 0, GPR64z, "ldrsw",
3108  [(set GPR64z:$Rt, (sextloadi32 (AArch64adr alignedglobal:$label)))]>;
3109
3110let AddedComplexity = 20 in {
3111def : Pat<(i64 (zextloadi32 (AArch64adr alignedglobal:$label))),
3112        (SUBREG_TO_REG (i64 0), (LDRWl $label), sub_32)>;
3113}
3114
3115// prefetch
3116def PRFMl : PrefetchLiteral<0b11, 0, "prfm", []>;
3117//                   [(AArch64Prefetch imm:$Rt, tglobaladdr:$label)]>;
3118
3119//---
3120// (unscaled immediate)
3121defm LDURX : LoadUnscaled<0b11, 0, 0b01, GPR64z, "ldur",
3122                    [(set GPR64z:$Rt,
3123                          (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset)))]>;
3124defm LDURW : LoadUnscaled<0b10, 0, 0b01, GPR32z, "ldur",
3125                    [(set GPR32z:$Rt,
3126                          (load (am_unscaled32 GPR64sp:$Rn, simm9:$offset)))]>;
3127defm LDURB : LoadUnscaled<0b00, 1, 0b01, FPR8Op, "ldur",
3128                    [(set FPR8Op:$Rt,
3129                          (load (am_unscaled8 GPR64sp:$Rn, simm9:$offset)))]>;
3130defm LDURH : LoadUnscaled<0b01, 1, 0b01, FPR16Op, "ldur",
3131                    [(set (f16 FPR16Op:$Rt),
3132                          (load (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
3133defm LDURS : LoadUnscaled<0b10, 1, 0b01, FPR32Op, "ldur",
3134                    [(set (f32 FPR32Op:$Rt),
3135                          (load (am_unscaled32 GPR64sp:$Rn, simm9:$offset)))]>;
3136defm LDURD : LoadUnscaled<0b11, 1, 0b01, FPR64Op, "ldur",
3137                    [(set (f64 FPR64Op:$Rt),
3138                          (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset)))]>;
3139defm LDURQ : LoadUnscaled<0b00, 1, 0b11, FPR128Op, "ldur",
3140                    [(set (f128 FPR128Op:$Rt),
3141                          (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset)))]>;
3142
3143defm LDURHH
3144    : LoadUnscaled<0b01, 0, 0b01, GPR32, "ldurh",
3145             [(set GPR32:$Rt,
3146                    (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
3147defm LDURBB
3148    : LoadUnscaled<0b00, 0, 0b01, GPR32, "ldurb",
3149             [(set GPR32:$Rt,
3150                    (zextloadi8 (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
3151
3152// Match all load 64 bits width whose type is compatible with FPR64
3153let Predicates = [IsLE] in {
3154  def : Pat<(v2f32 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
3155            (LDURDi GPR64sp:$Rn, simm9:$offset)>;
3156  def : Pat<(v2i32 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
3157            (LDURDi GPR64sp:$Rn, simm9:$offset)>;
3158  def : Pat<(v4i16 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
3159            (LDURDi GPR64sp:$Rn, simm9:$offset)>;
3160  def : Pat<(v8i8 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
3161            (LDURDi GPR64sp:$Rn, simm9:$offset)>;
3162  def : Pat<(v4f16 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
3163            (LDURDi GPR64sp:$Rn, simm9:$offset)>;
3164}
3165def : Pat<(v1f64 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
3166          (LDURDi GPR64sp:$Rn, simm9:$offset)>;
3167def : Pat<(v1i64 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
3168          (LDURDi GPR64sp:$Rn, simm9:$offset)>;
3169
3170// Match all load 128 bits width whose type is compatible with FPR128
3171let Predicates = [IsLE] in {
3172  def : Pat<(v2f64 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
3173            (LDURQi GPR64sp:$Rn, simm9:$offset)>;
3174  def : Pat<(v2i64 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
3175            (LDURQi GPR64sp:$Rn, simm9:$offset)>;
3176  def : Pat<(v4f32 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
3177            (LDURQi GPR64sp:$Rn, simm9:$offset)>;
3178  def : Pat<(v4i32 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
3179            (LDURQi GPR64sp:$Rn, simm9:$offset)>;
3180  def : Pat<(v8i16 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
3181            (LDURQi GPR64sp:$Rn, simm9:$offset)>;
3182  def : Pat<(v16i8 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
3183            (LDURQi GPR64sp:$Rn, simm9:$offset)>;
3184  def : Pat<(v8f16 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
3185            (LDURQi GPR64sp:$Rn, simm9:$offset)>;
3186}
3187
3188//  anyext -> zext
3189def : Pat<(i32 (extloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
3190          (LDURHHi GPR64sp:$Rn, simm9:$offset)>;
3191def : Pat<(i32 (extloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
3192          (LDURBBi GPR64sp:$Rn, simm9:$offset)>;
3193def : Pat<(i32 (extloadi1 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
3194          (LDURBBi GPR64sp:$Rn, simm9:$offset)>;
3195def : Pat<(i64 (extloadi32 (am_unscaled32 GPR64sp:$Rn, simm9:$offset))),
3196    (SUBREG_TO_REG (i64 0), (LDURWi GPR64sp:$Rn, simm9:$offset), sub_32)>;
3197def : Pat<(i64 (extloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
3198    (SUBREG_TO_REG (i64 0), (LDURHHi GPR64sp:$Rn, simm9:$offset), sub_32)>;
3199def : Pat<(i64 (extloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
3200    (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
3201def : Pat<(i64 (extloadi1 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
3202    (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
3203// unscaled zext
3204def : Pat<(i32 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
3205          (LDURHHi GPR64sp:$Rn, simm9:$offset)>;
3206def : Pat<(i32 (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
3207          (LDURBBi GPR64sp:$Rn, simm9:$offset)>;
3208def : Pat<(i32 (zextloadi1 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
3209          (LDURBBi GPR64sp:$Rn, simm9:$offset)>;
3210def : Pat<(i64 (zextloadi32 (am_unscaled32 GPR64sp:$Rn, simm9:$offset))),
3211    (SUBREG_TO_REG (i64 0), (LDURWi GPR64sp:$Rn, simm9:$offset), sub_32)>;
3212def : Pat<(i64 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
3213    (SUBREG_TO_REG (i64 0), (LDURHHi GPR64sp:$Rn, simm9:$offset), sub_32)>;
3214def : Pat<(i64 (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
3215    (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
3216def : Pat<(i64 (zextloadi1 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
3217    (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
3218
3219
3220//---
3221// LDR mnemonics fall back to LDUR for negative or unaligned offsets.
3222
3223// Define new assembler match classes as we want to only match these when
3224// the don't otherwise match the scaled addressing mode for LDR/STR. Don't
3225// associate a DiagnosticType either, as we want the diagnostic for the
3226// canonical form (the scaled operand) to take precedence.
3227class SImm9OffsetOperand<int Width> : AsmOperandClass {
3228  let Name = "SImm9OffsetFB" # Width;
3229  let PredicateMethod = "isSImm9OffsetFB<" # Width # ">";
3230  let RenderMethod = "addImmOperands";
3231}
3232
3233def SImm9OffsetFB8Operand : SImm9OffsetOperand<8>;
3234def SImm9OffsetFB16Operand : SImm9OffsetOperand<16>;
3235def SImm9OffsetFB32Operand : SImm9OffsetOperand<32>;
3236def SImm9OffsetFB64Operand : SImm9OffsetOperand<64>;
3237def SImm9OffsetFB128Operand : SImm9OffsetOperand<128>;
3238
3239def simm9_offset_fb8 : Operand<i64> {
3240  let ParserMatchClass = SImm9OffsetFB8Operand;
3241}
3242def simm9_offset_fb16 : Operand<i64> {
3243  let ParserMatchClass = SImm9OffsetFB16Operand;
3244}
3245def simm9_offset_fb32 : Operand<i64> {
3246  let ParserMatchClass = SImm9OffsetFB32Operand;
3247}
3248def simm9_offset_fb64 : Operand<i64> {
3249  let ParserMatchClass = SImm9OffsetFB64Operand;
3250}
3251def simm9_offset_fb128 : Operand<i64> {
3252  let ParserMatchClass = SImm9OffsetFB128Operand;
3253}
3254
3255def : InstAlias<"ldr $Rt, [$Rn, $offset]",
3256                (LDURXi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb64:$offset), 0>;
3257def : InstAlias<"ldr $Rt, [$Rn, $offset]",
3258                (LDURWi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
3259def : InstAlias<"ldr $Rt, [$Rn, $offset]",
3260                (LDURBi FPR8Op:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
3261def : InstAlias<"ldr $Rt, [$Rn, $offset]",
3262                (LDURHi FPR16Op:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
3263def : InstAlias<"ldr $Rt, [$Rn, $offset]",
3264                (LDURSi FPR32Op:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
3265def : InstAlias<"ldr $Rt, [$Rn, $offset]",
3266                (LDURDi FPR64Op:$Rt, GPR64sp:$Rn, simm9_offset_fb64:$offset), 0>;
3267def : InstAlias<"ldr $Rt, [$Rn, $offset]",
3268               (LDURQi FPR128Op:$Rt, GPR64sp:$Rn, simm9_offset_fb128:$offset), 0>;
3269
3270// zextload -> i64
3271def : Pat<(i64 (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
3272  (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
3273def : Pat<(i64 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
3274  (SUBREG_TO_REG (i64 0), (LDURHHi GPR64sp:$Rn, simm9:$offset), sub_32)>;
3275
3276// load sign-extended half-word
3277defm LDURSHW
3278    : LoadUnscaled<0b01, 0, 0b11, GPR32, "ldursh",
3279               [(set GPR32:$Rt,
3280                    (sextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
3281defm LDURSHX
3282    : LoadUnscaled<0b01, 0, 0b10, GPR64, "ldursh",
3283              [(set GPR64:$Rt,
3284                    (sextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
3285
3286// load sign-extended byte
3287defm LDURSBW
3288    : LoadUnscaled<0b00, 0, 0b11, GPR32, "ldursb",
3289                [(set GPR32:$Rt,
3290                      (sextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset)))]>;
3291defm LDURSBX
3292    : LoadUnscaled<0b00, 0, 0b10, GPR64, "ldursb",
3293                [(set GPR64:$Rt,
3294                      (sextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset)))]>;
3295
3296// load sign-extended word
3297defm LDURSW
3298    : LoadUnscaled<0b10, 0, 0b10, GPR64, "ldursw",
3299              [(set GPR64:$Rt,
3300                    (sextloadi32 (am_unscaled32 GPR64sp:$Rn, simm9:$offset)))]>;
3301
3302// zero and sign extending aliases from generic LDR* mnemonics to LDUR*.
3303def : InstAlias<"ldrb $Rt, [$Rn, $offset]",
3304                (LDURBBi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
3305def : InstAlias<"ldrh $Rt, [$Rn, $offset]",
3306                (LDURHHi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
3307def : InstAlias<"ldrsb $Rt, [$Rn, $offset]",
3308                (LDURSBWi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
3309def : InstAlias<"ldrsb $Rt, [$Rn, $offset]",
3310                (LDURSBXi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
3311def : InstAlias<"ldrsh $Rt, [$Rn, $offset]",
3312                (LDURSHWi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
3313def : InstAlias<"ldrsh $Rt, [$Rn, $offset]",
3314                (LDURSHXi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
3315def : InstAlias<"ldrsw $Rt, [$Rn, $offset]",
3316                (LDURSWi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
3317
3318// Pre-fetch.
3319defm PRFUM : PrefetchUnscaled<0b11, 0, 0b10, "prfum",
3320                  [(AArch64Prefetch timm:$Rt,
3321                                  (am_unscaled64 GPR64sp:$Rn, simm9:$offset))]>;
3322
3323//---
3324// (unscaled immediate, unprivileged)
3325defm LDTRX : LoadUnprivileged<0b11, 0, 0b01, GPR64, "ldtr">;
3326defm LDTRW : LoadUnprivileged<0b10, 0, 0b01, GPR32, "ldtr">;
3327
3328defm LDTRH : LoadUnprivileged<0b01, 0, 0b01, GPR32, "ldtrh">;
3329defm LDTRB : LoadUnprivileged<0b00, 0, 0b01, GPR32, "ldtrb">;
3330
3331// load sign-extended half-word
3332defm LDTRSHW : LoadUnprivileged<0b01, 0, 0b11, GPR32, "ldtrsh">;
3333defm LDTRSHX : LoadUnprivileged<0b01, 0, 0b10, GPR64, "ldtrsh">;
3334
3335// load sign-extended byte
3336defm LDTRSBW : LoadUnprivileged<0b00, 0, 0b11, GPR32, "ldtrsb">;
3337defm LDTRSBX : LoadUnprivileged<0b00, 0, 0b10, GPR64, "ldtrsb">;
3338
3339// load sign-extended word
3340defm LDTRSW  : LoadUnprivileged<0b10, 0, 0b10, GPR64, "ldtrsw">;
3341
3342//---
3343// (immediate pre-indexed)
3344def LDRWpre : LoadPreIdx<0b10, 0, 0b01, GPR32z, "ldr">;
3345def LDRXpre : LoadPreIdx<0b11, 0, 0b01, GPR64z, "ldr">;
3346def LDRBpre : LoadPreIdx<0b00, 1, 0b01, FPR8Op,  "ldr">;
3347def LDRHpre : LoadPreIdx<0b01, 1, 0b01, FPR16Op, "ldr">;
3348def LDRSpre : LoadPreIdx<0b10, 1, 0b01, FPR32Op, "ldr">;
3349def LDRDpre : LoadPreIdx<0b11, 1, 0b01, FPR64Op, "ldr">;
3350def LDRQpre : LoadPreIdx<0b00, 1, 0b11, FPR128Op, "ldr">;
3351
3352// load sign-extended half-word
3353def LDRSHWpre : LoadPreIdx<0b01, 0, 0b11, GPR32z, "ldrsh">;
3354def LDRSHXpre : LoadPreIdx<0b01, 0, 0b10, GPR64z, "ldrsh">;
3355
3356// load sign-extended byte
3357def LDRSBWpre : LoadPreIdx<0b00, 0, 0b11, GPR32z, "ldrsb">;
3358def LDRSBXpre : LoadPreIdx<0b00, 0, 0b10, GPR64z, "ldrsb">;
3359
3360// load zero-extended byte
3361def LDRBBpre : LoadPreIdx<0b00, 0, 0b01, GPR32z, "ldrb">;
3362def LDRHHpre : LoadPreIdx<0b01, 0, 0b01, GPR32z, "ldrh">;
3363
3364// load sign-extended word
3365def LDRSWpre : LoadPreIdx<0b10, 0, 0b10, GPR64z, "ldrsw">;
3366
3367//---
3368// (immediate post-indexed)
3369def LDRWpost : LoadPostIdx<0b10, 0, 0b01, GPR32z, "ldr">;
3370def LDRXpost : LoadPostIdx<0b11, 0, 0b01, GPR64z, "ldr">;
3371def LDRBpost : LoadPostIdx<0b00, 1, 0b01, FPR8Op,  "ldr">;
3372def LDRHpost : LoadPostIdx<0b01, 1, 0b01, FPR16Op, "ldr">;
3373def LDRSpost : LoadPostIdx<0b10, 1, 0b01, FPR32Op, "ldr">;
3374def LDRDpost : LoadPostIdx<0b11, 1, 0b01, FPR64Op, "ldr">;
3375def LDRQpost : LoadPostIdx<0b00, 1, 0b11, FPR128Op, "ldr">;
3376
3377// load sign-extended half-word
3378def LDRSHWpost : LoadPostIdx<0b01, 0, 0b11, GPR32z, "ldrsh">;
3379def LDRSHXpost : LoadPostIdx<0b01, 0, 0b10, GPR64z, "ldrsh">;
3380
3381// load sign-extended byte
3382def LDRSBWpost : LoadPostIdx<0b00, 0, 0b11, GPR32z, "ldrsb">;
3383def LDRSBXpost : LoadPostIdx<0b00, 0, 0b10, GPR64z, "ldrsb">;
3384
3385// load zero-extended byte
3386def LDRBBpost : LoadPostIdx<0b00, 0, 0b01, GPR32z, "ldrb">;
3387def LDRHHpost : LoadPostIdx<0b01, 0, 0b01, GPR32z, "ldrh">;
3388
3389// load sign-extended word
3390def LDRSWpost : LoadPostIdx<0b10, 0, 0b10, GPR64z, "ldrsw">;
3391
3392//===----------------------------------------------------------------------===//
3393// Store instructions.
3394//===----------------------------------------------------------------------===//
3395
3396// Pair (indexed, offset)
3397// FIXME: Use dedicated range-checked addressing mode operand here.
3398defm STPW : StorePairOffset<0b00, 0, GPR32z, simm7s4, "stp">;
3399defm STPX : StorePairOffset<0b10, 0, GPR64z, simm7s8, "stp">;
3400defm STPS : StorePairOffset<0b00, 1, FPR32Op, simm7s4, "stp">;
3401defm STPD : StorePairOffset<0b01, 1, FPR64Op, simm7s8, "stp">;
3402defm STPQ : StorePairOffset<0b10, 1, FPR128Op, simm7s16, "stp">;
3403
3404// Pair (pre-indexed)
3405def STPWpre : StorePairPreIdx<0b00, 0, GPR32z, simm7s4, "stp">;
3406def STPXpre : StorePairPreIdx<0b10, 0, GPR64z, simm7s8, "stp">;
3407def STPSpre : StorePairPreIdx<0b00, 1, FPR32Op, simm7s4, "stp">;
3408def STPDpre : StorePairPreIdx<0b01, 1, FPR64Op, simm7s8, "stp">;
3409def STPQpre : StorePairPreIdx<0b10, 1, FPR128Op, simm7s16, "stp">;
3410
3411// Pair (pre-indexed)
3412def STPWpost : StorePairPostIdx<0b00, 0, GPR32z, simm7s4, "stp">;
3413def STPXpost : StorePairPostIdx<0b10, 0, GPR64z, simm7s8, "stp">;
3414def STPSpost : StorePairPostIdx<0b00, 1, FPR32Op, simm7s4, "stp">;
3415def STPDpost : StorePairPostIdx<0b01, 1, FPR64Op, simm7s8, "stp">;
3416def STPQpost : StorePairPostIdx<0b10, 1, FPR128Op, simm7s16, "stp">;
3417
3418// Pair (no allocate)
3419defm STNPW : StorePairNoAlloc<0b00, 0, GPR32z, simm7s4, "stnp">;
3420defm STNPX : StorePairNoAlloc<0b10, 0, GPR64z, simm7s8, "stnp">;
3421defm STNPS : StorePairNoAlloc<0b00, 1, FPR32Op, simm7s4, "stnp">;
3422defm STNPD : StorePairNoAlloc<0b01, 1, FPR64Op, simm7s8, "stnp">;
3423defm STNPQ : StorePairNoAlloc<0b10, 1, FPR128Op, simm7s16, "stnp">;
3424
3425def : Pat<(AArch64stp GPR64z:$Rt, GPR64z:$Rt2, (am_indexed7s64 GPR64sp:$Rn, simm7s8:$offset)),
3426          (STPXi GPR64z:$Rt, GPR64z:$Rt2, GPR64sp:$Rn, simm7s8:$offset)>;
3427
3428def : Pat<(AArch64stnp FPR128:$Rt, FPR128:$Rt2, (am_indexed7s128 GPR64sp:$Rn, simm7s16:$offset)),
3429          (STNPQi FPR128:$Rt, FPR128:$Rt2, GPR64sp:$Rn, simm7s16:$offset)>;
3430
3431
3432//---
3433// (Register offset)
3434
3435// Integer
3436defm STRBB : Store8RO< 0b00, 0, 0b00, GPR32, "strb", i32, truncstorei8>;
3437defm STRHH : Store16RO<0b01, 0, 0b00, GPR32, "strh", i32, truncstorei16>;
3438defm STRW  : Store32RO<0b10, 0, 0b00, GPR32, "str",  i32, store>;
3439defm STRX  : Store64RO<0b11, 0, 0b00, GPR64, "str",  i64, store>;
3440
3441
3442// Floating-point
3443defm STRB : Store8RO< 0b00,  1, 0b00, FPR8Op,   "str", untyped, store>;
3444defm STRH : Store16RO<0b01,  1, 0b00, FPR16Op,  "str", f16,     store>;
3445defm STRS : Store32RO<0b10,  1, 0b00, FPR32Op,  "str", f32,     store>;
3446defm STRD : Store64RO<0b11,  1, 0b00, FPR64Op,  "str", f64,     store>;
3447defm STRQ : Store128RO<0b00, 1, 0b10, FPR128Op, "str">;
3448
3449let Predicates = [UseSTRQro], AddedComplexity = 10 in {
3450  def : Pat<(store (f128 FPR128:$Rt),
3451                        (ro_Windexed128 GPR64sp:$Rn, GPR32:$Rm,
3452                                        ro_Wextend128:$extend)),
3453            (STRQroW FPR128:$Rt, GPR64sp:$Rn, GPR32:$Rm, ro_Wextend128:$extend)>;
3454  def : Pat<(store (f128 FPR128:$Rt),
3455                        (ro_Xindexed128 GPR64sp:$Rn, GPR64:$Rm,
3456                                        ro_Xextend128:$extend)),
3457            (STRQroX FPR128:$Rt, GPR64sp:$Rn, GPR64:$Rm, ro_Wextend128:$extend)>;
3458}
3459
3460multiclass TruncStoreFrom64ROPat<ROAddrMode ro, SDPatternOperator storeop,
3461                                 Instruction STRW, Instruction STRX> {
3462
3463  def : Pat<(storeop GPR64:$Rt,
3464                     (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)),
3465            (STRW (EXTRACT_SUBREG GPR64:$Rt, sub_32),
3466                  GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
3467
3468  def : Pat<(storeop GPR64:$Rt,
3469                     (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)),
3470            (STRX (EXTRACT_SUBREG GPR64:$Rt, sub_32),
3471                  GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
3472}
3473
3474let AddedComplexity = 10 in {
3475  // truncstore i64
3476  defm : TruncStoreFrom64ROPat<ro8,  truncstorei8,  STRBBroW, STRBBroX>;
3477  defm : TruncStoreFrom64ROPat<ro16, truncstorei16, STRHHroW, STRHHroX>;
3478  defm : TruncStoreFrom64ROPat<ro32, truncstorei32, STRWroW,  STRWroX>;
3479}
3480
3481multiclass VecROStorePat<ROAddrMode ro, ValueType VecTy, RegisterClass FPR,
3482                         Instruction STRW, Instruction STRX> {
3483  def : Pat<(store (VecTy FPR:$Rt),
3484                   (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)),
3485            (STRW FPR:$Rt, GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
3486
3487  def : Pat<(store (VecTy FPR:$Rt),
3488                   (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)),
3489            (STRX FPR:$Rt, GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
3490}
3491
3492let AddedComplexity = 10 in {
3493// Match all store 64 bits width whose type is compatible with FPR64
3494let Predicates = [IsLE] in {
3495  // We must use ST1 to store vectors in big-endian.
3496  defm : VecROStorePat<ro64, v2i32, FPR64, STRDroW, STRDroX>;
3497  defm : VecROStorePat<ro64, v2f32, FPR64, STRDroW, STRDroX>;
3498  defm : VecROStorePat<ro64, v4i16, FPR64, STRDroW, STRDroX>;
3499  defm : VecROStorePat<ro64, v8i8, FPR64, STRDroW, STRDroX>;
3500  defm : VecROStorePat<ro64, v4f16, FPR64, STRDroW, STRDroX>;
3501  defm : VecROStorePat<ro64, v4bf16, FPR64, STRDroW, STRDroX>;
3502}
3503
3504defm : VecROStorePat<ro64, v1i64, FPR64, STRDroW, STRDroX>;
3505defm : VecROStorePat<ro64, v1f64, FPR64, STRDroW, STRDroX>;
3506
3507// Match all store 128 bits width whose type is compatible with FPR128
3508let Predicates = [IsLE, UseSTRQro] in {
3509  // We must use ST1 to store vectors in big-endian.
3510  defm : VecROStorePat<ro128, v2i64, FPR128, STRQroW, STRQroX>;
3511  defm : VecROStorePat<ro128, v2f64, FPR128, STRQroW, STRQroX>;
3512  defm : VecROStorePat<ro128, v4i32, FPR128, STRQroW, STRQroX>;
3513  defm : VecROStorePat<ro128, v4f32, FPR128, STRQroW, STRQroX>;
3514  defm : VecROStorePat<ro128, v8i16, FPR128, STRQroW, STRQroX>;
3515  defm : VecROStorePat<ro128, v16i8, FPR128, STRQroW, STRQroX>;
3516  defm : VecROStorePat<ro128, v8f16, FPR128, STRQroW, STRQroX>;
3517  defm : VecROStorePat<ro128, v8bf16, FPR128, STRQroW, STRQroX>;
3518}
3519} // AddedComplexity = 10
3520
3521// Match stores from lane 0 to the appropriate subreg's store.
3522multiclass VecROStoreLane0Pat<ROAddrMode ro, SDPatternOperator storeop,
3523                              ValueType VecTy, ValueType STy,
3524                              SubRegIndex SubRegIdx,
3525                              Instruction STRW, Instruction STRX> {
3526
3527  def : Pat<(storeop (STy (vector_extract (VecTy VecListOne128:$Vt), 0)),
3528                     (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)),
3529            (STRW (EXTRACT_SUBREG VecListOne128:$Vt, SubRegIdx),
3530                  GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
3531
3532  def : Pat<(storeop (STy (vector_extract (VecTy VecListOne128:$Vt), 0)),
3533                     (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)),
3534            (STRX (EXTRACT_SUBREG VecListOne128:$Vt, SubRegIdx),
3535                  GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
3536}
3537
3538let AddedComplexity = 19 in {
3539  defm : VecROStoreLane0Pat<ro16, truncstorei16, v8i16, i32, hsub, STRHroW, STRHroX>;
3540  defm : VecROStoreLane0Pat<ro16,         store, v8f16, f16, hsub, STRHroW, STRHroX>;
3541  defm : VecROStoreLane0Pat<ro32,         store, v4i32, i32, ssub, STRSroW, STRSroX>;
3542  defm : VecROStoreLane0Pat<ro32,         store, v4f32, f32, ssub, STRSroW, STRSroX>;
3543  defm : VecROStoreLane0Pat<ro64,         store, v2i64, i64, dsub, STRDroW, STRDroX>;
3544  defm : VecROStoreLane0Pat<ro64,         store, v2f64, f64, dsub, STRDroW, STRDroX>;
3545}
3546
3547//---
3548// (unsigned immediate)
3549defm STRX : StoreUIz<0b11, 0, 0b00, GPR64z, uimm12s8, "str",
3550                   [(store GPR64z:$Rt,
3551                            (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))]>;
3552defm STRW : StoreUIz<0b10, 0, 0b00, GPR32z, uimm12s4, "str",
3553                    [(store GPR32z:$Rt,
3554                            (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))]>;
3555defm STRB : StoreUI<0b00, 1, 0b00, FPR8Op, uimm12s1, "str",
3556                    [(store FPR8Op:$Rt,
3557                            (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))]>;
3558defm STRH : StoreUI<0b01, 1, 0b00, FPR16Op, uimm12s2, "str",
3559                    [(store (f16 FPR16Op:$Rt),
3560                            (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))]>;
3561defm STRS : StoreUI<0b10, 1, 0b00, FPR32Op, uimm12s4, "str",
3562                    [(store (f32 FPR32Op:$Rt),
3563                            (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))]>;
3564defm STRD : StoreUI<0b11, 1, 0b00, FPR64Op, uimm12s8, "str",
3565                    [(store (f64 FPR64Op:$Rt),
3566                            (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))]>;
3567defm STRQ : StoreUI<0b00, 1, 0b10, FPR128Op, uimm12s16, "str", []>;
3568
3569defm STRHH : StoreUIz<0b01, 0, 0b00, GPR32z, uimm12s2, "strh",
3570                     [(truncstorei16 GPR32z:$Rt,
3571                                     (am_indexed16 GPR64sp:$Rn,
3572                                                   uimm12s2:$offset))]>;
3573defm STRBB : StoreUIz<0b00, 0, 0b00, GPR32z, uimm12s1,  "strb",
3574                     [(truncstorei8 GPR32z:$Rt,
3575                                    (am_indexed8 GPR64sp:$Rn,
3576                                                 uimm12s1:$offset))]>;
3577
3578// bf16 store pattern
3579def : Pat<(store (bf16 FPR16Op:$Rt),
3580                 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset)),
3581          (STRHui FPR16:$Rt, GPR64sp:$Rn, uimm12s2:$offset)>;
3582
3583let AddedComplexity = 10 in {
3584
3585// Match all store 64 bits width whose type is compatible with FPR64
3586def : Pat<(store (v1i64 FPR64:$Rt),
3587                 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
3588          (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
3589def : Pat<(store (v1f64 FPR64:$Rt),
3590                 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
3591          (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
3592
3593let Predicates = [IsLE] in {
3594  // We must use ST1 to store vectors in big-endian.
3595  def : Pat<(store (v2f32 FPR64:$Rt),
3596                   (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
3597            (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
3598  def : Pat<(store (v8i8 FPR64:$Rt),
3599                   (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
3600            (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
3601  def : Pat<(store (v4i16 FPR64:$Rt),
3602                   (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
3603            (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
3604  def : Pat<(store (v2i32 FPR64:$Rt),
3605                   (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
3606            (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
3607  def : Pat<(store (v4f16 FPR64:$Rt),
3608                   (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
3609            (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
3610  def : Pat<(store (v4bf16 FPR64:$Rt),
3611                   (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
3612            (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
3613}
3614
3615// Match all store 128 bits width whose type is compatible with FPR128
3616def : Pat<(store (f128  FPR128:$Rt),
3617                 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
3618          (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
3619
3620let Predicates = [IsLE] in {
3621  // We must use ST1 to store vectors in big-endian.
3622  def : Pat<(store (v4f32 FPR128:$Rt),
3623                   (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
3624            (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
3625  def : Pat<(store (v2f64 FPR128:$Rt),
3626                   (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
3627            (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
3628  def : Pat<(store (v16i8 FPR128:$Rt),
3629                   (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
3630            (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
3631  def : Pat<(store (v8i16 FPR128:$Rt),
3632                   (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
3633            (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
3634  def : Pat<(store (v4i32 FPR128:$Rt),
3635                   (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
3636            (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
3637  def : Pat<(store (v2i64 FPR128:$Rt),
3638                   (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
3639            (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
3640  def : Pat<(store (v8f16 FPR128:$Rt),
3641                   (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
3642            (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
3643  def : Pat<(store (v8bf16 FPR128:$Rt),
3644                   (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
3645            (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
3646}
3647
3648// truncstore i64
3649def : Pat<(truncstorei32 GPR64:$Rt,
3650                         (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset)),
3651  (STRWui (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, uimm12s4:$offset)>;
3652def : Pat<(truncstorei16 GPR64:$Rt,
3653                         (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset)),
3654  (STRHHui (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, uimm12s2:$offset)>;
3655def : Pat<(truncstorei8 GPR64:$Rt, (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset)),
3656  (STRBBui (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, uimm12s1:$offset)>;
3657
3658} // AddedComplexity = 10
3659
3660// Match stores from lane 0 to the appropriate subreg's store.
3661multiclass VecStoreLane0Pat<ComplexPattern UIAddrMode, SDPatternOperator storeop,
3662                            ValueType VTy, ValueType STy,
3663                            SubRegIndex SubRegIdx, Operand IndexType,
3664                            Instruction STR> {
3665  def : Pat<(storeop (STy (vector_extract (VTy VecListOne128:$Vt), 0)),
3666                     (UIAddrMode GPR64sp:$Rn, IndexType:$offset)),
3667            (STR (EXTRACT_SUBREG VecListOne128:$Vt, SubRegIdx),
3668                 GPR64sp:$Rn, IndexType:$offset)>;
3669}
3670
3671let AddedComplexity = 19 in {
3672  defm : VecStoreLane0Pat<am_indexed16, truncstorei16, v8i16, i32, hsub, uimm12s2, STRHui>;
3673  defm : VecStoreLane0Pat<am_indexed16,         store, v8f16, f16, hsub, uimm12s2, STRHui>;
3674  defm : VecStoreLane0Pat<am_indexed32,         store, v4i32, i32, ssub, uimm12s4, STRSui>;
3675  defm : VecStoreLane0Pat<am_indexed32,         store, v4f32, f32, ssub, uimm12s4, STRSui>;
3676  defm : VecStoreLane0Pat<am_indexed64,         store, v2i64, i64, dsub, uimm12s8, STRDui>;
3677  defm : VecStoreLane0Pat<am_indexed64,         store, v2f64, f64, dsub, uimm12s8, STRDui>;
3678}
3679
3680//---
3681// (unscaled immediate)
3682defm STURX : StoreUnscaled<0b11, 0, 0b00, GPR64z, "stur",
3683                         [(store GPR64z:$Rt,
3684                                 (am_unscaled64 GPR64sp:$Rn, simm9:$offset))]>;
3685defm STURW : StoreUnscaled<0b10, 0, 0b00, GPR32z, "stur",
3686                         [(store GPR32z:$Rt,
3687                                 (am_unscaled32 GPR64sp:$Rn, simm9:$offset))]>;
3688defm STURB : StoreUnscaled<0b00, 1, 0b00, FPR8Op, "stur",
3689                         [(store FPR8Op:$Rt,
3690                                 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))]>;
3691defm STURH : StoreUnscaled<0b01, 1, 0b00, FPR16Op, "stur",
3692                         [(store (f16 FPR16Op:$Rt),
3693                                 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))]>;
3694defm STURS : StoreUnscaled<0b10, 1, 0b00, FPR32Op, "stur",
3695                         [(store (f32 FPR32Op:$Rt),
3696                                 (am_unscaled32 GPR64sp:$Rn, simm9:$offset))]>;
3697defm STURD : StoreUnscaled<0b11, 1, 0b00, FPR64Op, "stur",
3698                         [(store (f64 FPR64Op:$Rt),
3699                                 (am_unscaled64 GPR64sp:$Rn, simm9:$offset))]>;
3700defm STURQ : StoreUnscaled<0b00, 1, 0b10, FPR128Op, "stur",
3701                         [(store (f128 FPR128Op:$Rt),
3702                                 (am_unscaled128 GPR64sp:$Rn, simm9:$offset))]>;
3703defm STURHH : StoreUnscaled<0b01, 0, 0b00, GPR32z, "sturh",
3704                         [(truncstorei16 GPR32z:$Rt,
3705                                 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))]>;
3706defm STURBB : StoreUnscaled<0b00, 0, 0b00, GPR32z, "sturb",
3707                         [(truncstorei8 GPR32z:$Rt,
3708                                  (am_unscaled8 GPR64sp:$Rn, simm9:$offset))]>;
3709
3710// Armv8.4 Weaker Release Consistency enhancements
3711//         LDAPR & STLR with Immediate Offset instructions
3712let Predicates = [HasRCPC_IMMO] in {
3713defm STLURB     : BaseStoreUnscaleV84<"stlurb",  0b00, 0b00, GPR32>;
3714defm STLURH     : BaseStoreUnscaleV84<"stlurh",  0b01, 0b00, GPR32>;
3715defm STLURW     : BaseStoreUnscaleV84<"stlur",   0b10, 0b00, GPR32>;
3716defm STLURX     : BaseStoreUnscaleV84<"stlur",   0b11, 0b00, GPR64>;
3717defm LDAPURB    : BaseLoadUnscaleV84<"ldapurb",  0b00, 0b01, GPR32>;
3718defm LDAPURSBW  : BaseLoadUnscaleV84<"ldapursb", 0b00, 0b11, GPR32>;
3719defm LDAPURSBX  : BaseLoadUnscaleV84<"ldapursb", 0b00, 0b10, GPR64>;
3720defm LDAPURH    : BaseLoadUnscaleV84<"ldapurh",  0b01, 0b01, GPR32>;
3721defm LDAPURSHW  : BaseLoadUnscaleV84<"ldapursh", 0b01, 0b11, GPR32>;
3722defm LDAPURSHX  : BaseLoadUnscaleV84<"ldapursh", 0b01, 0b10, GPR64>;
3723defm LDAPUR     : BaseLoadUnscaleV84<"ldapur",   0b10, 0b01, GPR32>;
3724defm LDAPURSW   : BaseLoadUnscaleV84<"ldapursw", 0b10, 0b10, GPR64>;
3725defm LDAPURX    : BaseLoadUnscaleV84<"ldapur",   0b11, 0b01, GPR64>;
3726}
3727
3728// Match all store 64 bits width whose type is compatible with FPR64
3729def : Pat<(store (v1f64 FPR64:$Rt), (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
3730          (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
3731def : Pat<(store (v1i64 FPR64:$Rt), (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
3732          (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
3733
3734let AddedComplexity = 10 in {
3735
3736let Predicates = [IsLE] in {
3737  // We must use ST1 to store vectors in big-endian.
3738  def : Pat<(store (v2f32 FPR64:$Rt),
3739                   (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
3740            (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
3741  def : Pat<(store (v8i8 FPR64:$Rt),
3742                   (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
3743            (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
3744  def : Pat<(store (v4i16 FPR64:$Rt),
3745                   (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
3746            (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
3747  def : Pat<(store (v2i32 FPR64:$Rt),
3748                   (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
3749            (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
3750  def : Pat<(store (v4f16 FPR64:$Rt),
3751                   (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
3752            (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
3753  def : Pat<(store (v4bf16 FPR64:$Rt),
3754                   (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
3755            (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
3756}
3757
3758// Match all store 128 bits width whose type is compatible with FPR128
3759def : Pat<(store (f128 FPR128:$Rt), (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
3760          (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
3761
3762let Predicates = [IsLE] in {
3763  // We must use ST1 to store vectors in big-endian.
3764  def : Pat<(store (v4f32 FPR128:$Rt),
3765                   (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
3766            (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
3767  def : Pat<(store (v2f64 FPR128:$Rt),
3768                   (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
3769            (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
3770  def : Pat<(store (v16i8 FPR128:$Rt),
3771                   (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
3772            (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
3773  def : Pat<(store (v8i16 FPR128:$Rt),
3774                   (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
3775            (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
3776  def : Pat<(store (v4i32 FPR128:$Rt),
3777                   (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
3778            (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
3779  def : Pat<(store (v2i64 FPR128:$Rt),
3780                   (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
3781            (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
3782  def : Pat<(store (v2f64 FPR128:$Rt),
3783                   (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
3784            (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
3785  def : Pat<(store (v8f16 FPR128:$Rt),
3786                   (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
3787            (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
3788  def : Pat<(store (v8bf16 FPR128:$Rt),
3789                   (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
3790            (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
3791}
3792
3793} // AddedComplexity = 10
3794
3795// unscaled i64 truncating stores
3796def : Pat<(truncstorei32 GPR64:$Rt, (am_unscaled32 GPR64sp:$Rn, simm9:$offset)),
3797  (STURWi (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, simm9:$offset)>;
3798def : Pat<(truncstorei16 GPR64:$Rt, (am_unscaled16 GPR64sp:$Rn, simm9:$offset)),
3799  (STURHHi (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, simm9:$offset)>;
3800def : Pat<(truncstorei8 GPR64:$Rt, (am_unscaled8 GPR64sp:$Rn, simm9:$offset)),
3801  (STURBBi (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, simm9:$offset)>;
3802
3803// Match stores from lane 0 to the appropriate subreg's store.
3804multiclass VecStoreULane0Pat<SDPatternOperator StoreOp,
3805                             ValueType VTy, ValueType STy,
3806                             SubRegIndex SubRegIdx, Instruction STR> {
3807  defm : VecStoreLane0Pat<am_unscaled128, StoreOp, VTy, STy, SubRegIdx, simm9, STR>;
3808}
3809
3810let AddedComplexity = 19 in {
3811  defm : VecStoreULane0Pat<truncstorei16, v8i16, i32, hsub, STURHi>;
3812  defm : VecStoreULane0Pat<store,         v8f16, f16, hsub, STURHi>;
3813  defm : VecStoreULane0Pat<store,         v4i32, i32, ssub, STURSi>;
3814  defm : VecStoreULane0Pat<store,         v4f32, f32, ssub, STURSi>;
3815  defm : VecStoreULane0Pat<store,         v2i64, i64, dsub, STURDi>;
3816  defm : VecStoreULane0Pat<store,         v2f64, f64, dsub, STURDi>;
3817}
3818
3819//---
3820// STR mnemonics fall back to STUR for negative or unaligned offsets.
3821def : InstAlias<"str $Rt, [$Rn, $offset]",
3822                (STURXi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb64:$offset), 0>;
3823def : InstAlias<"str $Rt, [$Rn, $offset]",
3824                (STURWi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
3825def : InstAlias<"str $Rt, [$Rn, $offset]",
3826                (STURBi FPR8Op:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
3827def : InstAlias<"str $Rt, [$Rn, $offset]",
3828                (STURHi FPR16Op:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
3829def : InstAlias<"str $Rt, [$Rn, $offset]",
3830                (STURSi FPR32Op:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
3831def : InstAlias<"str $Rt, [$Rn, $offset]",
3832                (STURDi FPR64Op:$Rt, GPR64sp:$Rn, simm9_offset_fb64:$offset), 0>;
3833def : InstAlias<"str $Rt, [$Rn, $offset]",
3834                (STURQi FPR128Op:$Rt, GPR64sp:$Rn, simm9_offset_fb128:$offset), 0>;
3835
3836def : InstAlias<"strb $Rt, [$Rn, $offset]",
3837                (STURBBi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
3838def : InstAlias<"strh $Rt, [$Rn, $offset]",
3839                (STURHHi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
3840
3841//---
3842// (unscaled immediate, unprivileged)
3843defm STTRW : StoreUnprivileged<0b10, 0, 0b00, GPR32, "sttr">;
3844defm STTRX : StoreUnprivileged<0b11, 0, 0b00, GPR64, "sttr">;
3845
3846defm STTRH : StoreUnprivileged<0b01, 0, 0b00, GPR32, "sttrh">;
3847defm STTRB : StoreUnprivileged<0b00, 0, 0b00, GPR32, "sttrb">;
3848
3849//---
3850// (immediate pre-indexed)
3851def STRWpre : StorePreIdx<0b10, 0, 0b00, GPR32z, "str",  pre_store, i32>;
3852def STRXpre : StorePreIdx<0b11, 0, 0b00, GPR64z, "str",  pre_store, i64>;
3853def STRBpre : StorePreIdx<0b00, 1, 0b00, FPR8Op,  "str",  pre_store, untyped>;
3854def STRHpre : StorePreIdx<0b01, 1, 0b00, FPR16Op, "str",  pre_store, f16>;
3855def STRSpre : StorePreIdx<0b10, 1, 0b00, FPR32Op, "str",  pre_store, f32>;
3856def STRDpre : StorePreIdx<0b11, 1, 0b00, FPR64Op, "str",  pre_store, f64>;
3857def STRQpre : StorePreIdx<0b00, 1, 0b10, FPR128Op, "str", pre_store, f128>;
3858
3859def STRBBpre : StorePreIdx<0b00, 0, 0b00, GPR32z, "strb", pre_truncsti8,  i32>;
3860def STRHHpre : StorePreIdx<0b01, 0, 0b00, GPR32z, "strh", pre_truncsti16, i32>;
3861
3862// truncstore i64
3863def : Pat<(pre_truncsti32 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
3864  (STRWpre (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
3865           simm9:$off)>;
3866def : Pat<(pre_truncsti16 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
3867  (STRHHpre (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
3868            simm9:$off)>;
3869def : Pat<(pre_truncsti8 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
3870  (STRBBpre (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
3871            simm9:$off)>;
3872
3873def : Pat<(pre_store (v8i8 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
3874          (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
3875def : Pat<(pre_store (v4i16 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
3876          (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
3877def : Pat<(pre_store (v2i32 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
3878          (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
3879def : Pat<(pre_store (v2f32 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
3880          (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
3881def : Pat<(pre_store (v1i64 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
3882          (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
3883def : Pat<(pre_store (v1f64 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
3884          (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
3885def : Pat<(pre_store (v4f16 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
3886          (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
3887
3888def : Pat<(pre_store (v16i8 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
3889          (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
3890def : Pat<(pre_store (v8i16 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
3891          (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
3892def : Pat<(pre_store (v4i32 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
3893          (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
3894def : Pat<(pre_store (v4f32 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
3895          (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
3896def : Pat<(pre_store (v2i64 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
3897          (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
3898def : Pat<(pre_store (v2f64 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
3899          (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
3900def : Pat<(pre_store (v8f16 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
3901          (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
3902
3903//---
3904// (immediate post-indexed)
3905def STRWpost : StorePostIdx<0b10, 0, 0b00, GPR32z,  "str", post_store, i32>;
3906def STRXpost : StorePostIdx<0b11, 0, 0b00, GPR64z,  "str", post_store, i64>;
3907def STRBpost : StorePostIdx<0b00, 1, 0b00, FPR8Op,   "str", post_store, untyped>;
3908def STRHpost : StorePostIdx<0b01, 1, 0b00, FPR16Op,  "str", post_store, f16>;
3909def STRSpost : StorePostIdx<0b10, 1, 0b00, FPR32Op,  "str", post_store, f32>;
3910def STRDpost : StorePostIdx<0b11, 1, 0b00, FPR64Op,  "str", post_store, f64>;
3911def STRQpost : StorePostIdx<0b00, 1, 0b10, FPR128Op, "str", post_store, f128>;
3912
3913def STRBBpost : StorePostIdx<0b00, 0, 0b00, GPR32z, "strb", post_truncsti8, i32>;
3914def STRHHpost : StorePostIdx<0b01, 0, 0b00, GPR32z, "strh", post_truncsti16, i32>;
3915
3916// truncstore i64
3917def : Pat<(post_truncsti32 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
3918  (STRWpost (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
3919            simm9:$off)>;
3920def : Pat<(post_truncsti16 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
3921  (STRHHpost (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
3922             simm9:$off)>;
3923def : Pat<(post_truncsti8 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
3924  (STRBBpost (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
3925             simm9:$off)>;
3926
3927def : Pat<(post_store (bf16 FPR16:$Rt), GPR64sp:$addr, simm9:$off),
3928          (STRHpost FPR16:$Rt, GPR64sp:$addr, simm9:$off)>;
3929
3930def : Pat<(post_store (v8i8 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
3931          (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
3932def : Pat<(post_store (v4i16 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
3933          (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
3934def : Pat<(post_store (v2i32 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
3935          (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
3936def : Pat<(post_store (v2f32 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
3937          (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
3938def : Pat<(post_store (v1i64 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
3939          (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
3940def : Pat<(post_store (v1f64 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
3941          (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
3942def : Pat<(post_store (v4f16 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
3943          (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
3944def : Pat<(post_store (v4bf16 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
3945          (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
3946
3947def : Pat<(post_store (v16i8 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
3948          (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
3949def : Pat<(post_store (v8i16 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
3950          (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
3951def : Pat<(post_store (v4i32 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
3952          (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
3953def : Pat<(post_store (v4f32 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
3954          (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
3955def : Pat<(post_store (v2i64 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
3956          (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
3957def : Pat<(post_store (v2f64 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
3958          (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
3959def : Pat<(post_store (v8f16 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
3960          (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
3961def : Pat<(post_store (v8bf16 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
3962          (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
3963
3964//===----------------------------------------------------------------------===//
3965// Load/store exclusive instructions.
3966//===----------------------------------------------------------------------===//
3967
3968def LDARW  : LoadAcquire   <0b10, 1, 1, 0, 1, GPR32, "ldar">;
3969def LDARX  : LoadAcquire   <0b11, 1, 1, 0, 1, GPR64, "ldar">;
3970def LDARB  : LoadAcquire   <0b00, 1, 1, 0, 1, GPR32, "ldarb">;
3971def LDARH  : LoadAcquire   <0b01, 1, 1, 0, 1, GPR32, "ldarh">;
3972
3973def LDAXRW : LoadExclusive <0b10, 0, 1, 0, 1, GPR32, "ldaxr">;
3974def LDAXRX : LoadExclusive <0b11, 0, 1, 0, 1, GPR64, "ldaxr">;
3975def LDAXRB : LoadExclusive <0b00, 0, 1, 0, 1, GPR32, "ldaxrb">;
3976def LDAXRH : LoadExclusive <0b01, 0, 1, 0, 1, GPR32, "ldaxrh">;
3977
3978def LDXRW  : LoadExclusive <0b10, 0, 1, 0, 0, GPR32, "ldxr">;
3979def LDXRX  : LoadExclusive <0b11, 0, 1, 0, 0, GPR64, "ldxr">;
3980def LDXRB  : LoadExclusive <0b00, 0, 1, 0, 0, GPR32, "ldxrb">;
3981def LDXRH  : LoadExclusive <0b01, 0, 1, 0, 0, GPR32, "ldxrh">;
3982
3983def STLRW  : StoreRelease  <0b10, 1, 0, 0, 1, GPR32, "stlr">;
3984def STLRX  : StoreRelease  <0b11, 1, 0, 0, 1, GPR64, "stlr">;
3985def STLRB  : StoreRelease  <0b00, 1, 0, 0, 1, GPR32, "stlrb">;
3986def STLRH  : StoreRelease  <0b01, 1, 0, 0, 1, GPR32, "stlrh">;
3987
3988/*
3989Aliases for when offset=0. Note that in contrast to LoadAcquire which has a $Rn
3990of type GPR64sp0, we deliberately choose to make $Rn of type GPR64sp and add an
3991alias for the case of immediate #0. This is because new STLR versions (from
3992LRCPC3 extension) do have a non-zero immediate value, so GPR64sp0 is not
3993appropriate anymore (it parses and discards the optional zero). This is not the
3994case for LoadAcquire because the new LRCPC3 LDAR instructions are post-indexed,
3995and the immediate values are not inside the [] brackets and thus not accepted
3996by GPR64sp0 parser.
3997*/
3998def STLRW0  : InstAlias<"stlr\t$Rt, [$Rn, #0]" , (STLRW   GPR32: $Rt, GPR64sp:$Rn)>;
3999def STLRX0  : InstAlias<"stlr\t$Rt, [$Rn, #0]" , (STLRX   GPR64: $Rt, GPR64sp:$Rn)>;
4000def STLRB0  : InstAlias<"stlrb\t$Rt, [$Rn, #0]", (STLRB   GPR32: $Rt, GPR64sp:$Rn)>;
4001def STLRH0  : InstAlias<"stlrh\t$Rt, [$Rn, #0]", (STLRH   GPR32: $Rt, GPR64sp:$Rn)>;
4002
4003def STLXRW : StoreExclusive<0b10, 0, 0, 0, 1, GPR32, "stlxr">;
4004def STLXRX : StoreExclusive<0b11, 0, 0, 0, 1, GPR64, "stlxr">;
4005def STLXRB : StoreExclusive<0b00, 0, 0, 0, 1, GPR32, "stlxrb">;
4006def STLXRH : StoreExclusive<0b01, 0, 0, 0, 1, GPR32, "stlxrh">;
4007
4008def STXRW  : StoreExclusive<0b10, 0, 0, 0, 0, GPR32, "stxr">;
4009def STXRX  : StoreExclusive<0b11, 0, 0, 0, 0, GPR64, "stxr">;
4010def STXRB  : StoreExclusive<0b00, 0, 0, 0, 0, GPR32, "stxrb">;
4011def STXRH  : StoreExclusive<0b01, 0, 0, 0, 0, GPR32, "stxrh">;
4012
4013def LDAXPW : LoadExclusivePair<0b10, 0, 1, 1, 1, GPR32, "ldaxp">;
4014def LDAXPX : LoadExclusivePair<0b11, 0, 1, 1, 1, GPR64, "ldaxp">;
4015
4016def LDXPW  : LoadExclusivePair<0b10, 0, 1, 1, 0, GPR32, "ldxp">;
4017def LDXPX  : LoadExclusivePair<0b11, 0, 1, 1, 0, GPR64, "ldxp">;
4018
4019def STLXPW : StoreExclusivePair<0b10, 0, 0, 1, 1, GPR32, "stlxp">;
4020def STLXPX : StoreExclusivePair<0b11, 0, 0, 1, 1, GPR64, "stlxp">;
4021
4022def STXPW  : StoreExclusivePair<0b10, 0, 0, 1, 0, GPR32, "stxp">;
4023def STXPX  : StoreExclusivePair<0b11, 0, 0, 1, 0, GPR64, "stxp">;
4024
4025let Predicates = [HasLOR] in {
4026  // v8.1a "Limited Order Region" extension load-acquire instructions
4027  def LDLARW  : LoadAcquire   <0b10, 1, 1, 0, 0, GPR32, "ldlar">;
4028  def LDLARX  : LoadAcquire   <0b11, 1, 1, 0, 0, GPR64, "ldlar">;
4029  def LDLARB  : LoadAcquire   <0b00, 1, 1, 0, 0, GPR32, "ldlarb">;
4030  def LDLARH  : LoadAcquire   <0b01, 1, 1, 0, 0, GPR32, "ldlarh">;
4031
4032  // v8.1a "Limited Order Region" extension store-release instructions
4033  def STLLRW  : StoreRelease   <0b10, 1, 0, 0, 0, GPR32, "stllr">;
4034  def STLLRX  : StoreRelease   <0b11, 1, 0, 0, 0, GPR64, "stllr">;
4035  def STLLRB  : StoreRelease   <0b00, 1, 0, 0, 0, GPR32, "stllrb">;
4036  def STLLRH  : StoreRelease   <0b01, 1, 0, 0, 0, GPR32, "stllrh">;
4037
4038  // Aliases for when offset=0
4039  def STLLRW0 : InstAlias<"stllr\t$Rt,  [$Rn, #0]",  (STLLRW   GPR32: $Rt, GPR64sp:$Rn)>;
4040  def STLLRX0 : InstAlias<"stllr\t$Rt,  [$Rn, #0]",  (STLLRX   GPR64: $Rt, GPR64sp:$Rn)>;
4041  def STLLRB0 : InstAlias<"stllrb\t$Rt, [$Rn, #0]",  (STLLRB   GPR32: $Rt, GPR64sp:$Rn)>;
4042  def STLLRH0 : InstAlias<"stllrh\t$Rt, [$Rn, #0]",  (STLLRH   GPR32: $Rt, GPR64sp:$Rn)>;
4043}
4044
4045//===----------------------------------------------------------------------===//
4046// Scaled floating point to integer conversion instructions.
4047//===----------------------------------------------------------------------===//
4048
4049defm FCVTAS : FPToIntegerUnscaled<0b00, 0b100, "fcvtas", int_aarch64_neon_fcvtas>;
4050defm FCVTAU : FPToIntegerUnscaled<0b00, 0b101, "fcvtau", int_aarch64_neon_fcvtau>;
4051defm FCVTMS : FPToIntegerUnscaled<0b10, 0b000, "fcvtms", int_aarch64_neon_fcvtms>;
4052defm FCVTMU : FPToIntegerUnscaled<0b10, 0b001, "fcvtmu", int_aarch64_neon_fcvtmu>;
4053defm FCVTNS : FPToIntegerUnscaled<0b00, 0b000, "fcvtns", int_aarch64_neon_fcvtns>;
4054defm FCVTNU : FPToIntegerUnscaled<0b00, 0b001, "fcvtnu", int_aarch64_neon_fcvtnu>;
4055defm FCVTPS : FPToIntegerUnscaled<0b01, 0b000, "fcvtps", int_aarch64_neon_fcvtps>;
4056defm FCVTPU : FPToIntegerUnscaled<0b01, 0b001, "fcvtpu", int_aarch64_neon_fcvtpu>;
4057defm FCVTZS : FPToIntegerUnscaled<0b11, 0b000, "fcvtzs", any_fp_to_sint>;
4058defm FCVTZU : FPToIntegerUnscaled<0b11, 0b001, "fcvtzu", any_fp_to_uint>;
4059defm FCVTZS : FPToIntegerScaled<0b11, 0b000, "fcvtzs", any_fp_to_sint>;
4060defm FCVTZU : FPToIntegerScaled<0b11, 0b001, "fcvtzu", any_fp_to_uint>;
4061
4062// AArch64's FCVT instructions saturate when out of range.
4063multiclass FPToIntegerSatPats<SDNode to_int_sat, string INST> {
4064  let Predicates = [HasFullFP16] in {
4065  def : Pat<(i32 (to_int_sat f16:$Rn, i32)),
4066            (!cast<Instruction>(INST # UWHr) f16:$Rn)>;
4067  def : Pat<(i64 (to_int_sat f16:$Rn, i64)),
4068            (!cast<Instruction>(INST # UXHr) f16:$Rn)>;
4069  }
4070  def : Pat<(i32 (to_int_sat f32:$Rn, i32)),
4071            (!cast<Instruction>(INST # UWSr) f32:$Rn)>;
4072  def : Pat<(i64 (to_int_sat f32:$Rn, i64)),
4073            (!cast<Instruction>(INST # UXSr) f32:$Rn)>;
4074  def : Pat<(i32 (to_int_sat f64:$Rn, i32)),
4075            (!cast<Instruction>(INST # UWDr) f64:$Rn)>;
4076  def : Pat<(i64 (to_int_sat f64:$Rn, i64)),
4077            (!cast<Instruction>(INST # UXDr) f64:$Rn)>;
4078
4079  let Predicates = [HasFullFP16] in {
4080  def : Pat<(i32 (to_int_sat (fmul f16:$Rn, fixedpoint_f16_i32:$scale), i32)),
4081            (!cast<Instruction>(INST # SWHri) $Rn, $scale)>;
4082  def : Pat<(i64 (to_int_sat (fmul f16:$Rn, fixedpoint_f16_i64:$scale), i64)),
4083            (!cast<Instruction>(INST # SXHri) $Rn, $scale)>;
4084  }
4085  def : Pat<(i32 (to_int_sat (fmul f32:$Rn, fixedpoint_f32_i32:$scale), i32)),
4086            (!cast<Instruction>(INST # SWSri) $Rn, $scale)>;
4087  def : Pat<(i64 (to_int_sat (fmul f32:$Rn, fixedpoint_f32_i64:$scale), i64)),
4088            (!cast<Instruction>(INST # SXSri) $Rn, $scale)>;
4089  def : Pat<(i32 (to_int_sat (fmul f64:$Rn, fixedpoint_f64_i32:$scale), i32)),
4090            (!cast<Instruction>(INST # SWDri) $Rn, $scale)>;
4091  def : Pat<(i64 (to_int_sat (fmul f64:$Rn, fixedpoint_f64_i64:$scale), i64)),
4092            (!cast<Instruction>(INST # SXDri) $Rn, $scale)>;
4093}
4094
4095defm : FPToIntegerSatPats<fp_to_sint_sat, "FCVTZS">;
4096defm : FPToIntegerSatPats<fp_to_uint_sat, "FCVTZU">;
4097
4098multiclass FPToIntegerIntPats<Intrinsic round, string INST> {
4099  let Predicates = [HasFullFP16] in {
4100  def : Pat<(i32 (round f16:$Rn)), (!cast<Instruction>(INST # UWHr) $Rn)>;
4101  def : Pat<(i64 (round f16:$Rn)), (!cast<Instruction>(INST # UXHr) $Rn)>;
4102  }
4103  def : Pat<(i32 (round f32:$Rn)), (!cast<Instruction>(INST # UWSr) $Rn)>;
4104  def : Pat<(i64 (round f32:$Rn)), (!cast<Instruction>(INST # UXSr) $Rn)>;
4105  def : Pat<(i32 (round f64:$Rn)), (!cast<Instruction>(INST # UWDr) $Rn)>;
4106  def : Pat<(i64 (round f64:$Rn)), (!cast<Instruction>(INST # UXDr) $Rn)>;
4107
4108  let Predicates = [HasFullFP16] in {
4109  def : Pat<(i32 (round (fmul f16:$Rn, fixedpoint_f16_i32:$scale))),
4110            (!cast<Instruction>(INST # SWHri) $Rn, $scale)>;
4111  def : Pat<(i64 (round (fmul f16:$Rn, fixedpoint_f16_i64:$scale))),
4112            (!cast<Instruction>(INST # SXHri) $Rn, $scale)>;
4113  }
4114  def : Pat<(i32 (round (fmul f32:$Rn, fixedpoint_f32_i32:$scale))),
4115            (!cast<Instruction>(INST # SWSri) $Rn, $scale)>;
4116  def : Pat<(i64 (round (fmul f32:$Rn, fixedpoint_f32_i64:$scale))),
4117            (!cast<Instruction>(INST # SXSri) $Rn, $scale)>;
4118  def : Pat<(i32 (round (fmul f64:$Rn, fixedpoint_f64_i32:$scale))),
4119            (!cast<Instruction>(INST # SWDri) $Rn, $scale)>;
4120  def : Pat<(i64 (round (fmul f64:$Rn, fixedpoint_f64_i64:$scale))),
4121            (!cast<Instruction>(INST # SXDri) $Rn, $scale)>;
4122}
4123
4124defm : FPToIntegerIntPats<int_aarch64_neon_fcvtzs, "FCVTZS">;
4125defm : FPToIntegerIntPats<int_aarch64_neon_fcvtzu, "FCVTZU">;
4126
4127multiclass FPToIntegerPats<SDNode to_int, SDNode to_int_sat, SDNode round, string INST> {
4128  def : Pat<(i32 (to_int (round f32:$Rn))),
4129            (!cast<Instruction>(INST # UWSr) f32:$Rn)>;
4130  def : Pat<(i64 (to_int (round f32:$Rn))),
4131            (!cast<Instruction>(INST # UXSr) f32:$Rn)>;
4132  def : Pat<(i32 (to_int (round f64:$Rn))),
4133            (!cast<Instruction>(INST # UWDr) f64:$Rn)>;
4134  def : Pat<(i64 (to_int (round f64:$Rn))),
4135            (!cast<Instruction>(INST # UXDr) f64:$Rn)>;
4136
4137  // These instructions saturate like fp_to_[su]int_sat.
4138  let Predicates = [HasFullFP16] in {
4139  def : Pat<(i32 (to_int_sat (round f16:$Rn), i32)),
4140            (!cast<Instruction>(INST # UWHr) f16:$Rn)>;
4141  def : Pat<(i64 (to_int_sat (round f16:$Rn), i64)),
4142            (!cast<Instruction>(INST # UXHr) f16:$Rn)>;
4143  }
4144  def : Pat<(i32 (to_int_sat (round f32:$Rn), i32)),
4145            (!cast<Instruction>(INST # UWSr) f32:$Rn)>;
4146  def : Pat<(i64 (to_int_sat (round f32:$Rn), i64)),
4147            (!cast<Instruction>(INST # UXSr) f32:$Rn)>;
4148  def : Pat<(i32 (to_int_sat (round f64:$Rn), i32)),
4149            (!cast<Instruction>(INST # UWDr) f64:$Rn)>;
4150  def : Pat<(i64 (to_int_sat (round f64:$Rn), i64)),
4151            (!cast<Instruction>(INST # UXDr) f64:$Rn)>;
4152}
4153
4154defm : FPToIntegerPats<fp_to_sint, fp_to_sint_sat, fceil,  "FCVTPS">;
4155defm : FPToIntegerPats<fp_to_uint, fp_to_uint_sat, fceil,  "FCVTPU">;
4156defm : FPToIntegerPats<fp_to_sint, fp_to_sint_sat, ffloor, "FCVTMS">;
4157defm : FPToIntegerPats<fp_to_uint, fp_to_uint_sat, ffloor, "FCVTMU">;
4158defm : FPToIntegerPats<fp_to_sint, fp_to_sint_sat, ftrunc, "FCVTZS">;
4159defm : FPToIntegerPats<fp_to_uint, fp_to_uint_sat, ftrunc, "FCVTZU">;
4160defm : FPToIntegerPats<fp_to_sint, fp_to_sint_sat, fround, "FCVTAS">;
4161defm : FPToIntegerPats<fp_to_uint, fp_to_uint_sat, fround, "FCVTAU">;
4162
4163
4164
4165let Predicates = [HasFullFP16] in {
4166  def : Pat<(i32 (any_lround f16:$Rn)),
4167            (!cast<Instruction>(FCVTASUWHr) f16:$Rn)>;
4168  def : Pat<(i64 (any_lround f16:$Rn)),
4169            (!cast<Instruction>(FCVTASUXHr) f16:$Rn)>;
4170  def : Pat<(i64 (any_llround f16:$Rn)),
4171            (!cast<Instruction>(FCVTASUXHr) f16:$Rn)>;
4172}
4173def : Pat<(i32 (any_lround f32:$Rn)),
4174          (!cast<Instruction>(FCVTASUWSr) f32:$Rn)>;
4175def : Pat<(i32 (any_lround f64:$Rn)),
4176          (!cast<Instruction>(FCVTASUWDr) f64:$Rn)>;
4177def : Pat<(i64 (any_lround f32:$Rn)),
4178          (!cast<Instruction>(FCVTASUXSr) f32:$Rn)>;
4179def : Pat<(i64 (any_lround f64:$Rn)),
4180          (!cast<Instruction>(FCVTASUXDr) f64:$Rn)>;
4181def : Pat<(i64 (any_llround f32:$Rn)),
4182          (!cast<Instruction>(FCVTASUXSr) f32:$Rn)>;
4183def : Pat<(i64 (any_llround f64:$Rn)),
4184          (!cast<Instruction>(FCVTASUXDr) f64:$Rn)>;
4185
4186//===----------------------------------------------------------------------===//
4187// Scaled integer to floating point conversion instructions.
4188//===----------------------------------------------------------------------===//
4189
4190defm SCVTF : IntegerToFP<0, "scvtf", any_sint_to_fp>;
4191defm UCVTF : IntegerToFP<1, "ucvtf", any_uint_to_fp>;
4192
4193//===----------------------------------------------------------------------===//
4194// Unscaled integer to floating point conversion instruction.
4195//===----------------------------------------------------------------------===//
4196
4197defm FMOV : UnscaledConversion<"fmov">;
4198
4199// Add pseudo ops for FMOV 0 so we can mark them as isReMaterializable
4200let isReMaterializable = 1, isCodeGenOnly = 1, isAsCheapAsAMove = 1 in {
4201def FMOVH0 : Pseudo<(outs FPR16:$Rd), (ins), [(set f16:$Rd, (fpimm0))]>,
4202    Sched<[WriteF]>, Requires<[HasFullFP16]>;
4203def FMOVS0 : Pseudo<(outs FPR32:$Rd), (ins), [(set f32:$Rd, (fpimm0))]>,
4204    Sched<[WriteF]>;
4205def FMOVD0 : Pseudo<(outs FPR64:$Rd), (ins), [(set f64:$Rd, (fpimm0))]>,
4206    Sched<[WriteF]>;
4207}
4208// Similarly add aliases
4209def : InstAlias<"fmov $Rd, #0.0", (FMOVWHr FPR16:$Rd, WZR), 0>,
4210    Requires<[HasFullFP16]>;
4211def : InstAlias<"fmov $Rd, #0.0", (FMOVWSr FPR32:$Rd, WZR), 0>;
4212def : InstAlias<"fmov $Rd, #0.0", (FMOVXDr FPR64:$Rd, XZR), 0>;
4213
4214// Pattern for FP16 immediates
4215let Predicates = [HasFullFP16] in {
4216  def : Pat<(f16 fpimm:$in),
4217    (FMOVWHr (MOVi32imm (bitcast_fpimm_to_i32 f16:$in)))>;
4218}
4219
4220//===----------------------------------------------------------------------===//
4221// Floating point conversion instruction.
4222//===----------------------------------------------------------------------===//
4223
4224defm FCVT : FPConversion<"fcvt">;
4225
4226//===----------------------------------------------------------------------===//
4227// Floating point single operand instructions.
4228//===----------------------------------------------------------------------===//
4229
4230defm FABS   : SingleOperandFPDataNoException<0b0001, "fabs", fabs>;
4231defm FMOV   : SingleOperandFPDataNoException<0b0000, "fmov">;
4232defm FNEG   : SingleOperandFPDataNoException<0b0010, "fneg", fneg>;
4233defm FRINTA : SingleOperandFPData<0b1100, "frinta", any_fround>;
4234defm FRINTI : SingleOperandFPData<0b1111, "frinti", any_fnearbyint>;
4235defm FRINTM : SingleOperandFPData<0b1010, "frintm", any_ffloor>;
4236defm FRINTN : SingleOperandFPData<0b1000, "frintn", any_froundeven>;
4237defm FRINTP : SingleOperandFPData<0b1001, "frintp", any_fceil>;
4238
4239defm FRINTX : SingleOperandFPData<0b1110, "frintx", any_frint>;
4240defm FRINTZ : SingleOperandFPData<0b1011, "frintz", any_ftrunc>;
4241
4242let SchedRW = [WriteFDiv] in {
4243defm FSQRT  : SingleOperandFPData<0b0011, "fsqrt", any_fsqrt>;
4244}
4245
4246let Predicates = [HasFRInt3264] in {
4247  defm FRINT32Z : FRIntNNT<0b00, "frint32z", int_aarch64_frint32z>;
4248  defm FRINT64Z : FRIntNNT<0b10, "frint64z", int_aarch64_frint64z>;
4249  defm FRINT32X : FRIntNNT<0b01, "frint32x", int_aarch64_frint32x>;
4250  defm FRINT64X : FRIntNNT<0b11, "frint64x", int_aarch64_frint64x>;
4251} // HasFRInt3264
4252
4253// Emitting strict_lrint as two instructions is valid as any exceptions that
4254// occur will happen in exactly one of the instructions (e.g. if the input is
4255// not an integer the inexact exception will happen in the FRINTX but not then
4256// in the FCVTZS as the output of FRINTX is an integer).
4257let Predicates = [HasFullFP16] in {
4258  def : Pat<(i32 (any_lrint f16:$Rn)),
4259            (FCVTZSUWHr (!cast<Instruction>(FRINTXHr) f16:$Rn))>;
4260  def : Pat<(i64 (any_lrint f16:$Rn)),
4261            (FCVTZSUXHr (!cast<Instruction>(FRINTXHr) f16:$Rn))>;
4262  def : Pat<(i64 (any_llrint f16:$Rn)),
4263            (FCVTZSUXHr (!cast<Instruction>(FRINTXHr) f16:$Rn))>;
4264}
4265def : Pat<(i32 (any_lrint f32:$Rn)),
4266          (FCVTZSUWSr (!cast<Instruction>(FRINTXSr) f32:$Rn))>;
4267def : Pat<(i32 (any_lrint f64:$Rn)),
4268          (FCVTZSUWDr (!cast<Instruction>(FRINTXDr) f64:$Rn))>;
4269def : Pat<(i64 (any_lrint f32:$Rn)),
4270          (FCVTZSUXSr (!cast<Instruction>(FRINTXSr) f32:$Rn))>;
4271def : Pat<(i64 (any_lrint f64:$Rn)),
4272          (FCVTZSUXDr (!cast<Instruction>(FRINTXDr) f64:$Rn))>;
4273def : Pat<(i64 (any_llrint f32:$Rn)),
4274          (FCVTZSUXSr (!cast<Instruction>(FRINTXSr) f32:$Rn))>;
4275def : Pat<(i64 (any_llrint f64:$Rn)),
4276          (FCVTZSUXDr (!cast<Instruction>(FRINTXDr) f64:$Rn))>;
4277
4278//===----------------------------------------------------------------------===//
4279// Floating point two operand instructions.
4280//===----------------------------------------------------------------------===//
4281
4282defm FADD   : TwoOperandFPData<0b0010, "fadd", any_fadd>;
4283let SchedRW = [WriteFDiv] in {
4284defm FDIV   : TwoOperandFPData<0b0001, "fdiv", any_fdiv>;
4285}
4286defm FMAXNM : TwoOperandFPData<0b0110, "fmaxnm", any_fmaxnum>;
4287defm FMAX   : TwoOperandFPData<0b0100, "fmax", any_fmaximum>;
4288defm FMINNM : TwoOperandFPData<0b0111, "fminnm", any_fminnum>;
4289defm FMIN   : TwoOperandFPData<0b0101, "fmin", any_fminimum>;
4290let SchedRW = [WriteFMul] in {
4291defm FMUL   : TwoOperandFPData<0b0000, "fmul", any_fmul>;
4292defm FNMUL  : TwoOperandFPDataNeg<0b1000, "fnmul", any_fmul>;
4293}
4294defm FSUB   : TwoOperandFPData<0b0011, "fsub", any_fsub>;
4295
4296// Match reassociated forms of FNMUL.
4297def : Pat<(fmul (fneg FPR16:$a), (f16 FPR16:$b)),
4298          (FNMULHrr FPR16:$a, FPR16:$b)>,
4299          Requires<[HasFullFP16]>;
4300def : Pat<(fmul (fneg FPR32:$a), (f32 FPR32:$b)),
4301          (FNMULSrr FPR32:$a, FPR32:$b)>;
4302def : Pat<(fmul (fneg FPR64:$a), (f64 FPR64:$b)),
4303          (FNMULDrr FPR64:$a, FPR64:$b)>;
4304
4305def : Pat<(v1f64 (fmaximum (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
4306          (FMAXDrr FPR64:$Rn, FPR64:$Rm)>;
4307def : Pat<(v1f64 (fminimum (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
4308          (FMINDrr FPR64:$Rn, FPR64:$Rm)>;
4309def : Pat<(v1f64 (fmaxnum (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
4310          (FMAXNMDrr FPR64:$Rn, FPR64:$Rm)>;
4311def : Pat<(v1f64 (fminnum (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
4312          (FMINNMDrr FPR64:$Rn, FPR64:$Rm)>;
4313
4314//===----------------------------------------------------------------------===//
4315// Floating point three operand instructions.
4316//===----------------------------------------------------------------------===//
4317
4318defm FMADD  : ThreeOperandFPData<0, 0, "fmadd", any_fma>;
4319defm FMSUB  : ThreeOperandFPData<0, 1, "fmsub",
4320     TriOpFrag<(any_fma node:$LHS, (fneg node:$MHS), node:$RHS)> >;
4321defm FNMADD : ThreeOperandFPData<1, 0, "fnmadd",
4322     TriOpFrag<(fneg (any_fma node:$LHS, node:$MHS, node:$RHS))> >;
4323defm FNMSUB : ThreeOperandFPData<1, 1, "fnmsub",
4324     TriOpFrag<(any_fma node:$LHS, node:$MHS, (fneg node:$RHS))> >;
4325
4326// The following def pats catch the case where the LHS of an FMA is negated.
4327// The TriOpFrag above catches the case where the middle operand is negated.
4328
4329// N.b. FMSUB etc have the accumulator at the *end* of (outs), unlike
4330// the NEON variant.
4331
4332// Here we handle first -(a + b*c) for FNMADD:
4333
4334let Predicates = [HasNEON, HasFullFP16] in
4335def : Pat<(f16 (fma (fneg FPR16:$Rn), FPR16:$Rm, FPR16:$Ra)),
4336          (FMSUBHrrr FPR16:$Rn, FPR16:$Rm, FPR16:$Ra)>;
4337
4338def : Pat<(f32 (fma (fneg FPR32:$Rn), FPR32:$Rm, FPR32:$Ra)),
4339          (FMSUBSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
4340
4341def : Pat<(f64 (fma (fneg FPR64:$Rn), FPR64:$Rm, FPR64:$Ra)),
4342          (FMSUBDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
4343
4344// Now it's time for "(-a) + (-b)*c"
4345
4346let Predicates = [HasNEON, HasFullFP16] in
4347def : Pat<(f16 (fma (fneg FPR16:$Rn), FPR16:$Rm, (fneg FPR16:$Ra))),
4348          (FNMADDHrrr FPR16:$Rn, FPR16:$Rm, FPR16:$Ra)>;
4349
4350def : Pat<(f32 (fma (fneg FPR32:$Rn), FPR32:$Rm, (fneg FPR32:$Ra))),
4351          (FNMADDSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
4352
4353def : Pat<(f64 (fma (fneg FPR64:$Rn), FPR64:$Rm, (fneg FPR64:$Ra))),
4354          (FNMADDDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
4355
4356//===----------------------------------------------------------------------===//
4357// Floating point comparison instructions.
4358//===----------------------------------------------------------------------===//
4359
4360defm FCMPE : FPComparison<1, "fcmpe", AArch64strict_fcmpe>;
4361defm FCMP  : FPComparison<0, "fcmp", AArch64any_fcmp>;
4362
4363//===----------------------------------------------------------------------===//
4364// Floating point conditional comparison instructions.
4365//===----------------------------------------------------------------------===//
4366
4367defm FCCMPE : FPCondComparison<1, "fccmpe">;
4368defm FCCMP  : FPCondComparison<0, "fccmp", AArch64fccmp>;
4369
4370//===----------------------------------------------------------------------===//
4371// Floating point conditional select instruction.
4372//===----------------------------------------------------------------------===//
4373
4374defm FCSEL : FPCondSelect<"fcsel">;
4375
4376let Predicates = [HasFullFP16] in
4377def : Pat<(bf16 (AArch64csel (bf16 FPR16:$Rn), (bf16 FPR16:$Rm), (i32 imm:$cond), NZCV)),
4378          (FCSELHrrr FPR16:$Rn, FPR16:$Rm, imm:$cond)>;
4379
4380// CSEL instructions providing f128 types need to be handled by a
4381// pseudo-instruction since the eventual code will need to introduce basic
4382// blocks and control flow.
4383def F128CSEL : Pseudo<(outs FPR128:$Rd),
4384                      (ins FPR128:$Rn, FPR128:$Rm, ccode:$cond),
4385                      [(set (f128 FPR128:$Rd),
4386                            (AArch64csel FPR128:$Rn, FPR128:$Rm,
4387                                       (i32 imm:$cond), NZCV))]> {
4388  let Uses = [NZCV];
4389  let usesCustomInserter = 1;
4390  let hasNoSchedulingInfo = 1;
4391}
4392
4393//===----------------------------------------------------------------------===//
4394// Instructions used for emitting unwind opcodes on ARM64 Windows.
4395//===----------------------------------------------------------------------===//
4396let isPseudo = 1 in {
4397  def SEH_StackAlloc : Pseudo<(outs), (ins i32imm:$size), []>, Sched<[]>;
4398  def SEH_SaveFPLR : Pseudo<(outs), (ins i32imm:$offs), []>, Sched<[]>;
4399  def SEH_SaveFPLR_X : Pseudo<(outs), (ins i32imm:$offs), []>, Sched<[]>;
4400  def SEH_SaveReg : Pseudo<(outs), (ins i32imm:$reg, i32imm:$offs), []>, Sched<[]>;
4401  def SEH_SaveReg_X : Pseudo<(outs), (ins i32imm:$reg, i32imm:$offs), []>, Sched<[]>;
4402  def SEH_SaveRegP : Pseudo<(outs), (ins i32imm:$reg0, i32imm:$reg1, i32imm:$offs), []>, Sched<[]>;
4403  def SEH_SaveRegP_X : Pseudo<(outs), (ins i32imm:$reg0, i32imm:$reg1, i32imm:$offs), []>, Sched<[]>;
4404  def SEH_SaveFReg : Pseudo<(outs), (ins i32imm:$reg, i32imm:$offs), []>, Sched<[]>;
4405  def SEH_SaveFReg_X :  Pseudo<(outs), (ins i32imm:$reg, i32imm:$offs), []>, Sched<[]>;
4406  def SEH_SaveFRegP : Pseudo<(outs), (ins i32imm:$reg0, i32imm:$reg1, i32imm:$offs), []>, Sched<[]>;
4407  def SEH_SaveFRegP_X : Pseudo<(outs), (ins i32imm:$reg0, i32imm:$reg1, i32imm:$offs), []>, Sched<[]>;
4408  def SEH_SetFP : Pseudo<(outs), (ins), []>, Sched<[]>;
4409  def SEH_AddFP : Pseudo<(outs), (ins i32imm:$offs), []>, Sched<[]>;
4410  def SEH_Nop : Pseudo<(outs), (ins), []>, Sched<[]>;
4411  def SEH_PrologEnd : Pseudo<(outs), (ins), []>, Sched<[]>;
4412  def SEH_EpilogStart : Pseudo<(outs), (ins), []>, Sched<[]>;
4413  def SEH_EpilogEnd : Pseudo<(outs), (ins), []>, Sched<[]>;
4414  def SEH_PACSignLR : Pseudo<(outs), (ins), []>, Sched<[]>;
4415}
4416
4417// Pseudo instructions for Windows EH
4418//===----------------------------------------------------------------------===//
4419let isTerminator = 1, hasSideEffects = 1, isBarrier = 1, hasCtrlDep = 1,
4420    isCodeGenOnly = 1, isReturn = 1, isEHScopeReturn = 1, isPseudo = 1 in {
4421   def CLEANUPRET : Pseudo<(outs), (ins), [(cleanupret)]>, Sched<[]>;
4422   let usesCustomInserter = 1 in
4423     def CATCHRET : Pseudo<(outs), (ins am_brcond:$dst, am_brcond:$src), [(catchret bb:$dst, bb:$src)]>,
4424                    Sched<[]>;
4425}
4426
4427// Pseudo instructions for homogeneous prolog/epilog
4428let isPseudo = 1 in {
4429  // Save CSRs in order, {FPOffset}
4430  def HOM_Prolog : Pseudo<(outs), (ins variable_ops), []>, Sched<[]>;
4431  // Restore CSRs in order
4432  def HOM_Epilog : Pseudo<(outs), (ins variable_ops), []>, Sched<[]>;
4433}
4434
4435//===----------------------------------------------------------------------===//
4436// Floating point immediate move.
4437//===----------------------------------------------------------------------===//
4438
4439let isReMaterializable = 1, isAsCheapAsAMove = 1 in {
4440defm FMOV : FPMoveImmediate<"fmov">;
4441}
4442
4443//===----------------------------------------------------------------------===//
4444// Advanced SIMD two vector instructions.
4445//===----------------------------------------------------------------------===//
4446
4447defm UABDL   : SIMDLongThreeVectorBHSabdl<1, 0b0111, "uabdl",
4448                                          AArch64uabd>;
4449// Match UABDL in log2-shuffle patterns.
4450def : Pat<(abs (v8i16 (sub (zext (v8i8 V64:$opA)),
4451                           (zext (v8i8 V64:$opB))))),
4452          (UABDLv8i8_v8i16 V64:$opA, V64:$opB)>;
4453def : Pat<(xor (v8i16 (AArch64vashr v8i16:$src, (i32 15))),
4454               (v8i16 (add (sub (zext (v8i8 V64:$opA)),
4455                                (zext (v8i8 V64:$opB))),
4456                           (AArch64vashr v8i16:$src, (i32 15))))),
4457          (UABDLv8i8_v8i16 V64:$opA, V64:$opB)>;
4458def : Pat<(abs (v8i16 (sub (zext (extract_high_v16i8 (v16i8 V128:$opA))),
4459                           (zext (extract_high_v16i8 (v16i8 V128:$opB)))))),
4460          (UABDLv16i8_v8i16 V128:$opA, V128:$opB)>;
4461def : Pat<(xor (v8i16 (AArch64vashr v8i16:$src, (i32 15))),
4462               (v8i16 (add (sub (zext (extract_high_v16i8 (v16i8 V128:$opA))),
4463                                (zext (extract_high_v16i8 (v16i8 V128:$opB)))),
4464                           (AArch64vashr v8i16:$src, (i32 15))))),
4465          (UABDLv16i8_v8i16 V128:$opA, V128:$opB)>;
4466def : Pat<(abs (v4i32 (sub (zext (v4i16 V64:$opA)),
4467                           (zext (v4i16 V64:$opB))))),
4468          (UABDLv4i16_v4i32 V64:$opA, V64:$opB)>;
4469def : Pat<(abs (v4i32 (sub (zext (extract_high_v8i16 (v8i16 V128:$opA))),
4470                           (zext (extract_high_v8i16 (v8i16 V128:$opB)))))),
4471          (UABDLv8i16_v4i32 V128:$opA, V128:$opB)>;
4472def : Pat<(abs (v2i64 (sub (zext (v2i32 V64:$opA)),
4473                           (zext (v2i32 V64:$opB))))),
4474          (UABDLv2i32_v2i64 V64:$opA, V64:$opB)>;
4475def : Pat<(abs (v2i64 (sub (zext (extract_high_v4i32 (v4i32 V128:$opA))),
4476                           (zext (extract_high_v4i32 (v4i32 V128:$opB)))))),
4477          (UABDLv4i32_v2i64 V128:$opA, V128:$opB)>;
4478
4479defm ABS    : SIMDTwoVectorBHSD<0, 0b01011, "abs", abs>;
4480defm CLS    : SIMDTwoVectorBHS<0, 0b00100, "cls", int_aarch64_neon_cls>;
4481defm CLZ    : SIMDTwoVectorBHS<1, 0b00100, "clz", ctlz>;
4482defm CMEQ   : SIMDCmpTwoVector<0, 0b01001, "cmeq", AArch64cmeqz>;
4483defm CMGE   : SIMDCmpTwoVector<1, 0b01000, "cmge", AArch64cmgez>;
4484defm CMGT   : SIMDCmpTwoVector<0, 0b01000, "cmgt", AArch64cmgtz>;
4485defm CMLE   : SIMDCmpTwoVector<1, 0b01001, "cmle", AArch64cmlez>;
4486defm CMLT   : SIMDCmpTwoVector<0, 0b01010, "cmlt", AArch64cmltz>;
4487defm CNT    : SIMDTwoVectorB<0, 0b00, 0b00101, "cnt", ctpop>;
4488defm FABS   : SIMDTwoVectorFPNoException<0, 1, 0b01111, "fabs", fabs>;
4489
4490def : Pat<(v8i8 (AArch64vashr (v8i8 V64:$Rn), (i32 7))),
4491          (CMLTv8i8rz V64:$Rn)>;
4492def : Pat<(v4i16 (AArch64vashr (v4i16 V64:$Rn), (i32 15))),
4493          (CMLTv4i16rz V64:$Rn)>;
4494def : Pat<(v2i32 (AArch64vashr (v2i32 V64:$Rn), (i32 31))),
4495          (CMLTv2i32rz V64:$Rn)>;
4496def : Pat<(v16i8 (AArch64vashr (v16i8 V128:$Rn), (i32 7))),
4497          (CMLTv16i8rz V128:$Rn)>;
4498def : Pat<(v8i16 (AArch64vashr (v8i16 V128:$Rn), (i32 15))),
4499          (CMLTv8i16rz V128:$Rn)>;
4500def : Pat<(v4i32 (AArch64vashr (v4i32 V128:$Rn), (i32 31))),
4501          (CMLTv4i32rz V128:$Rn)>;
4502def : Pat<(v2i64 (AArch64vashr (v2i64 V128:$Rn), (i32 63))),
4503          (CMLTv2i64rz V128:$Rn)>;
4504
4505defm FCMEQ  : SIMDFPCmpTwoVector<0, 1, 0b01101, "fcmeq", AArch64fcmeqz>;
4506defm FCMGE  : SIMDFPCmpTwoVector<1, 1, 0b01100, "fcmge", AArch64fcmgez>;
4507defm FCMGT  : SIMDFPCmpTwoVector<0, 1, 0b01100, "fcmgt", AArch64fcmgtz>;
4508defm FCMLE  : SIMDFPCmpTwoVector<1, 1, 0b01101, "fcmle", AArch64fcmlez>;
4509defm FCMLT  : SIMDFPCmpTwoVector<0, 1, 0b01110, "fcmlt", AArch64fcmltz>;
4510defm FCVTAS : SIMDTwoVectorFPToInt<0,0,0b11100, "fcvtas",int_aarch64_neon_fcvtas>;
4511defm FCVTAU : SIMDTwoVectorFPToInt<1,0,0b11100, "fcvtau",int_aarch64_neon_fcvtau>;
4512defm FCVTL  : SIMDFPWidenTwoVector<0, 0, 0b10111, "fcvtl">;
4513def : Pat<(v4f32 (int_aarch64_neon_vcvthf2fp (v4i16 V64:$Rn))),
4514          (FCVTLv4i16 V64:$Rn)>;
4515def : Pat<(v4f32 (int_aarch64_neon_vcvthf2fp (extract_subvector (v8i16 V128:$Rn),
4516                                                              (i64 4)))),
4517          (FCVTLv8i16 V128:$Rn)>;
4518def : Pat<(v2f64 (any_fpextend (v2f32 V64:$Rn))), (FCVTLv2i32 V64:$Rn)>;
4519
4520def : Pat<(v4f32 (any_fpextend (v4f16 V64:$Rn))), (FCVTLv4i16 V64:$Rn)>;
4521
4522defm FCVTMS : SIMDTwoVectorFPToInt<0,0,0b11011, "fcvtms",int_aarch64_neon_fcvtms>;
4523defm FCVTMU : SIMDTwoVectorFPToInt<1,0,0b11011, "fcvtmu",int_aarch64_neon_fcvtmu>;
4524defm FCVTNS : SIMDTwoVectorFPToInt<0,0,0b11010, "fcvtns",int_aarch64_neon_fcvtns>;
4525defm FCVTNU : SIMDTwoVectorFPToInt<1,0,0b11010, "fcvtnu",int_aarch64_neon_fcvtnu>;
4526defm FCVTN  : SIMDFPNarrowTwoVector<0, 0, 0b10110, "fcvtn">;
4527def : Pat<(v4i16 (int_aarch64_neon_vcvtfp2hf (v4f32 V128:$Rn))),
4528          (FCVTNv4i16 V128:$Rn)>;
4529def : Pat<(concat_vectors V64:$Rd,
4530                          (v4i16 (int_aarch64_neon_vcvtfp2hf (v4f32 V128:$Rn)))),
4531          (FCVTNv8i16 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
4532def : Pat<(v2f32 (any_fpround (v2f64 V128:$Rn))), (FCVTNv2i32 V128:$Rn)>;
4533def : Pat<(v4f16 (any_fpround (v4f32 V128:$Rn))), (FCVTNv4i16 V128:$Rn)>;
4534def : Pat<(concat_vectors V64:$Rd, (v2f32 (any_fpround (v2f64 V128:$Rn)))),
4535          (FCVTNv4i32 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
4536defm FCVTPS : SIMDTwoVectorFPToInt<0,1,0b11010, "fcvtps",int_aarch64_neon_fcvtps>;
4537defm FCVTPU : SIMDTwoVectorFPToInt<1,1,0b11010, "fcvtpu",int_aarch64_neon_fcvtpu>;
4538defm FCVTXN : SIMDFPInexactCvtTwoVector<1, 0, 0b10110, "fcvtxn",
4539                                        int_aarch64_neon_fcvtxn>;
4540defm FCVTZS : SIMDTwoVectorFPToInt<0, 1, 0b11011, "fcvtzs", any_fp_to_sint>;
4541defm FCVTZU : SIMDTwoVectorFPToInt<1, 1, 0b11011, "fcvtzu", any_fp_to_uint>;
4542
4543// AArch64's FCVT instructions saturate when out of range.
4544multiclass SIMDTwoVectorFPToIntSatPats<SDNode to_int_sat, string INST> {
4545  let Predicates = [HasFullFP16] in {
4546  def : Pat<(v4i16 (to_int_sat v4f16:$Rn, i16)),
4547            (!cast<Instruction>(INST # v4f16) v4f16:$Rn)>;
4548  def : Pat<(v8i16 (to_int_sat v8f16:$Rn, i16)),
4549            (!cast<Instruction>(INST # v8f16) v8f16:$Rn)>;
4550  }
4551  def : Pat<(v2i32 (to_int_sat v2f32:$Rn, i32)),
4552            (!cast<Instruction>(INST # v2f32) v2f32:$Rn)>;
4553  def : Pat<(v4i32 (to_int_sat v4f32:$Rn, i32)),
4554            (!cast<Instruction>(INST # v4f32) v4f32:$Rn)>;
4555  def : Pat<(v2i64 (to_int_sat v2f64:$Rn, i64)),
4556            (!cast<Instruction>(INST # v2f64) v2f64:$Rn)>;
4557}
4558defm : SIMDTwoVectorFPToIntSatPats<fp_to_sint_sat, "FCVTZS">;
4559defm : SIMDTwoVectorFPToIntSatPats<fp_to_uint_sat, "FCVTZU">;
4560
4561def : Pat<(v4i16 (int_aarch64_neon_fcvtzs v4f16:$Rn)), (FCVTZSv4f16 $Rn)>;
4562def : Pat<(v8i16 (int_aarch64_neon_fcvtzs v8f16:$Rn)), (FCVTZSv8f16 $Rn)>;
4563def : Pat<(v2i32 (int_aarch64_neon_fcvtzs v2f32:$Rn)), (FCVTZSv2f32 $Rn)>;
4564def : Pat<(v4i32 (int_aarch64_neon_fcvtzs v4f32:$Rn)), (FCVTZSv4f32 $Rn)>;
4565def : Pat<(v2i64 (int_aarch64_neon_fcvtzs v2f64:$Rn)), (FCVTZSv2f64 $Rn)>;
4566
4567def : Pat<(v4i16 (int_aarch64_neon_fcvtzu v4f16:$Rn)), (FCVTZUv4f16 $Rn)>;
4568def : Pat<(v8i16 (int_aarch64_neon_fcvtzu v8f16:$Rn)), (FCVTZUv8f16 $Rn)>;
4569def : Pat<(v2i32 (int_aarch64_neon_fcvtzu v2f32:$Rn)), (FCVTZUv2f32 $Rn)>;
4570def : Pat<(v4i32 (int_aarch64_neon_fcvtzu v4f32:$Rn)), (FCVTZUv4f32 $Rn)>;
4571def : Pat<(v2i64 (int_aarch64_neon_fcvtzu v2f64:$Rn)), (FCVTZUv2f64 $Rn)>;
4572
4573defm FNEG   : SIMDTwoVectorFPNoException<1, 1, 0b01111, "fneg", fneg>;
4574defm FRECPE : SIMDTwoVectorFP<0, 1, 0b11101, "frecpe", int_aarch64_neon_frecpe>;
4575defm FRINTA : SIMDTwoVectorFP<1, 0, 0b11000, "frinta", any_fround>;
4576defm FRINTI : SIMDTwoVectorFP<1, 1, 0b11001, "frinti", any_fnearbyint>;
4577defm FRINTM : SIMDTwoVectorFP<0, 0, 0b11001, "frintm", any_ffloor>;
4578defm FRINTN : SIMDTwoVectorFP<0, 0, 0b11000, "frintn", any_froundeven>;
4579defm FRINTP : SIMDTwoVectorFP<0, 1, 0b11000, "frintp", any_fceil>;
4580defm FRINTX : SIMDTwoVectorFP<1, 0, 0b11001, "frintx", any_frint>;
4581defm FRINTZ : SIMDTwoVectorFP<0, 1, 0b11001, "frintz", any_ftrunc>;
4582
4583let Predicates = [HasFRInt3264] in {
4584  defm FRINT32Z : FRIntNNTVector<0, 0, "frint32z", int_aarch64_neon_frint32z>;
4585  defm FRINT64Z : FRIntNNTVector<0, 1, "frint64z", int_aarch64_neon_frint64z>;
4586  defm FRINT32X : FRIntNNTVector<1, 0, "frint32x", int_aarch64_neon_frint32x>;
4587  defm FRINT64X : FRIntNNTVector<1, 1, "frint64x", int_aarch64_neon_frint64x>;
4588} // HasFRInt3264
4589
4590defm FRSQRTE: SIMDTwoVectorFP<1, 1, 0b11101, "frsqrte", int_aarch64_neon_frsqrte>;
4591defm FSQRT  : SIMDTwoVectorFP<1, 1, 0b11111, "fsqrt", any_fsqrt>;
4592defm NEG    : SIMDTwoVectorBHSD<1, 0b01011, "neg",
4593                               UnOpFrag<(sub immAllZerosV, node:$LHS)> >;
4594defm NOT    : SIMDTwoVectorB<1, 0b00, 0b00101, "not", vnot>;
4595// Aliases for MVN -> NOT.
4596def : InstAlias<"mvn{ $Vd.8b, $Vn.8b|.8b $Vd, $Vn}",
4597                (NOTv8i8 V64:$Vd, V64:$Vn)>;
4598def : InstAlias<"mvn{ $Vd.16b, $Vn.16b|.16b $Vd, $Vn}",
4599                (NOTv16i8 V128:$Vd, V128:$Vn)>;
4600
4601def : Pat<(vnot (v4i16 V64:$Rn)),  (NOTv8i8  V64:$Rn)>;
4602def : Pat<(vnot (v8i16 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
4603def : Pat<(vnot (v2i32 V64:$Rn)),  (NOTv8i8  V64:$Rn)>;
4604def : Pat<(vnot (v4i32 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
4605def : Pat<(vnot (v1i64 V64:$Rn)),  (NOTv8i8  V64:$Rn)>;
4606def : Pat<(vnot (v2i64 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
4607
4608defm RBIT   : SIMDTwoVectorB<1, 0b01, 0b00101, "rbit", bitreverse>;
4609defm REV16  : SIMDTwoVectorB<0, 0b00, 0b00001, "rev16", AArch64rev16>;
4610defm REV32  : SIMDTwoVectorBH<1, 0b00000, "rev32", AArch64rev32>;
4611defm REV64  : SIMDTwoVectorBHS<0, 0b00000, "rev64", AArch64rev64>;
4612defm SADALP : SIMDLongTwoVectorTied<0, 0b00110, "sadalp",
4613       BinOpFrag<(add node:$LHS, (AArch64saddlp node:$RHS))> >;
4614defm SADDLP : SIMDLongTwoVector<0, 0b00010, "saddlp", AArch64saddlp>;
4615defm SCVTF  : SIMDTwoVectorIntToFP<0, 0, 0b11101, "scvtf", any_sint_to_fp>;
4616defm SHLL   : SIMDVectorLShiftLongBySizeBHS;
4617defm SQABS  : SIMDTwoVectorBHSD<0, 0b00111, "sqabs", int_aarch64_neon_sqabs>;
4618defm SQNEG  : SIMDTwoVectorBHSD<1, 0b00111, "sqneg", int_aarch64_neon_sqneg>;
4619defm SQXTN  : SIMDMixedTwoVector<0, 0b10100, "sqxtn", int_aarch64_neon_sqxtn>;
4620defm SQXTUN : SIMDMixedTwoVector<1, 0b10010, "sqxtun", int_aarch64_neon_sqxtun>;
4621defm SUQADD : SIMDTwoVectorBHSDTied<0, 0b00011, "suqadd",int_aarch64_neon_suqadd>;
4622defm UADALP : SIMDLongTwoVectorTied<1, 0b00110, "uadalp",
4623       BinOpFrag<(add node:$LHS, (AArch64uaddlp node:$RHS))> >;
4624defm UADDLP : SIMDLongTwoVector<1, 0b00010, "uaddlp", AArch64uaddlp>;
4625defm UCVTF  : SIMDTwoVectorIntToFP<1, 0, 0b11101, "ucvtf", any_uint_to_fp>;
4626defm UQXTN  : SIMDMixedTwoVector<1, 0b10100, "uqxtn", int_aarch64_neon_uqxtn>;
4627defm URECPE : SIMDTwoVectorS<0, 1, 0b11100, "urecpe", int_aarch64_neon_urecpe>;
4628defm URSQRTE: SIMDTwoVectorS<1, 1, 0b11100, "ursqrte", int_aarch64_neon_ursqrte>;
4629defm USQADD : SIMDTwoVectorBHSDTied<1, 0b00011, "usqadd",int_aarch64_neon_usqadd>;
4630defm XTN    : SIMDMixedTwoVector<0, 0b10010, "xtn", trunc>;
4631
4632def : Pat<(v4f16  (AArch64rev32 V64:$Rn)),  (REV32v4i16 V64:$Rn)>;
4633def : Pat<(v4f16  (AArch64rev64 V64:$Rn)),  (REV64v4i16 V64:$Rn)>;
4634def : Pat<(v4bf16 (AArch64rev32 V64:$Rn)),  (REV32v4i16 V64:$Rn)>;
4635def : Pat<(v4bf16 (AArch64rev64 V64:$Rn)),  (REV64v4i16 V64:$Rn)>;
4636def : Pat<(v8f16  (AArch64rev32 V128:$Rn)), (REV32v8i16 V128:$Rn)>;
4637def : Pat<(v8f16  (AArch64rev64 V128:$Rn)), (REV64v8i16 V128:$Rn)>;
4638def : Pat<(v8bf16 (AArch64rev32 V128:$Rn)), (REV32v8i16 V128:$Rn)>;
4639def : Pat<(v8bf16 (AArch64rev64 V128:$Rn)), (REV64v8i16 V128:$Rn)>;
4640def : Pat<(v2f32  (AArch64rev64 V64:$Rn)),  (REV64v2i32 V64:$Rn)>;
4641def : Pat<(v4f32  (AArch64rev64 V128:$Rn)), (REV64v4i32 V128:$Rn)>;
4642
4643// Patterns for vector long shift (by element width). These need to match all
4644// three of zext, sext and anyext so it's easier to pull the patterns out of the
4645// definition.
4646multiclass SIMDVectorLShiftLongBySizeBHSPats<SDPatternOperator ext> {
4647  def : Pat<(AArch64vshl (v8i16 (ext (v8i8 V64:$Rn))), (i32 8)),
4648            (SHLLv8i8 V64:$Rn)>;
4649  def : Pat<(AArch64vshl (v8i16 (ext (extract_high_v16i8 (v16i8 V128:$Rn)))), (i32 8)),
4650            (SHLLv16i8 V128:$Rn)>;
4651  def : Pat<(AArch64vshl (v4i32 (ext (v4i16 V64:$Rn))), (i32 16)),
4652            (SHLLv4i16 V64:$Rn)>;
4653  def : Pat<(AArch64vshl (v4i32 (ext (extract_high_v8i16 (v8i16 V128:$Rn)))), (i32 16)),
4654            (SHLLv8i16 V128:$Rn)>;
4655  def : Pat<(AArch64vshl (v2i64 (ext (v2i32 V64:$Rn))), (i32 32)),
4656            (SHLLv2i32 V64:$Rn)>;
4657  def : Pat<(AArch64vshl (v2i64 (ext (extract_high_v4i32 (v4i32 V128:$Rn)))), (i32 32)),
4658            (SHLLv4i32 V128:$Rn)>;
4659}
4660
4661defm : SIMDVectorLShiftLongBySizeBHSPats<anyext>;
4662defm : SIMDVectorLShiftLongBySizeBHSPats<zext>;
4663defm : SIMDVectorLShiftLongBySizeBHSPats<sext>;
4664
4665// Constant vector values, used in the S/UQXTN patterns below.
4666def VImmFF:   PatLeaf<(AArch64NvCast (v2i64 (AArch64movi_edit (i32 85))))>;
4667def VImmFFFF: PatLeaf<(AArch64NvCast (v2i64 (AArch64movi_edit (i32 51))))>;
4668def VImm7F:   PatLeaf<(AArch64movi_shift (i32 127), (i32 0))>;
4669def VImm80:   PatLeaf<(AArch64mvni_shift (i32 127), (i32 0))>;
4670def VImm7FFF: PatLeaf<(AArch64movi_msl (i32 127), (i32 264))>;
4671def VImm8000: PatLeaf<(AArch64mvni_msl (i32 127), (i32 264))>;
4672
4673// trunc(umin(X, 255)) -> UQXTRN v8i8
4674def : Pat<(v8i8 (trunc (umin (v8i16 V128:$Vn), (v8i16 VImmFF)))),
4675          (UQXTNv8i8 V128:$Vn)>;
4676// trunc(umin(X, 65535)) -> UQXTRN v4i16
4677def : Pat<(v4i16 (trunc (umin (v4i32 V128:$Vn), (v4i32 VImmFFFF)))),
4678          (UQXTNv4i16 V128:$Vn)>;
4679// trunc(smin(smax(X, -128), 128)) -> SQXTRN
4680//  with reversed min/max
4681def : Pat<(v8i8 (trunc (smin (smax (v8i16 V128:$Vn), (v8i16 VImm80)),
4682                             (v8i16 VImm7F)))),
4683          (SQXTNv8i8 V128:$Vn)>;
4684def : Pat<(v8i8 (trunc (smax (smin (v8i16 V128:$Vn), (v8i16 VImm7F)),
4685                             (v8i16 VImm80)))),
4686          (SQXTNv8i8 V128:$Vn)>;
4687// trunc(smin(smax(X, -32768), 32767)) -> SQXTRN
4688//  with reversed min/max
4689def : Pat<(v4i16 (trunc (smin (smax (v4i32 V128:$Vn), (v4i32 VImm8000)),
4690                              (v4i32 VImm7FFF)))),
4691          (SQXTNv4i16 V128:$Vn)>;
4692def : Pat<(v4i16 (trunc (smax (smin (v4i32 V128:$Vn), (v4i32 VImm7FFF)),
4693                              (v4i32 VImm8000)))),
4694          (SQXTNv4i16 V128:$Vn)>;
4695
4696// concat_vectors(Vd, trunc(smin(smax Vm, -128), 127) ~> SQXTN2(Vd, Vn)
4697// with reversed min/max
4698def : Pat<(v16i8 (concat_vectors
4699                 (v8i8 V64:$Vd),
4700                 (v8i8 (trunc (smin (smax (v8i16 V128:$Vn), (v8i16 VImm80)),
4701                                          (v8i16 VImm7F)))))),
4702          (SQXTNv16i8 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Vd, dsub), V128:$Vn)>;
4703def : Pat<(v16i8 (concat_vectors
4704                 (v8i8 V64:$Vd),
4705                 (v8i8 (trunc (smax (smin (v8i16 V128:$Vn), (v8i16 VImm7F)),
4706                                          (v8i16 VImm80)))))),
4707          (SQXTNv16i8 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Vd, dsub), V128:$Vn)>;
4708
4709// concat_vectors(Vd, trunc(smin(smax Vm, -32768), 32767) ~> SQXTN2(Vd, Vn)
4710// with reversed min/max
4711def : Pat<(v8i16 (concat_vectors
4712                 (v4i16 V64:$Vd),
4713                 (v4i16 (trunc (smin (smax (v4i32 V128:$Vn), (v4i32 VImm8000)),
4714                                           (v4i32 VImm7FFF)))))),
4715          (SQXTNv8i16 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Vd, dsub), V128:$Vn)>;
4716def : Pat<(v8i16 (concat_vectors
4717                 (v4i16 V64:$Vd),
4718                 (v4i16 (trunc (smax (smin (v4i32 V128:$Vn), (v4i32 VImm7FFF)),
4719                                           (v4i32 VImm8000)))))),
4720          (SQXTNv8i16 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Vd, dsub), V128:$Vn)>;
4721
4722//===----------------------------------------------------------------------===//
4723// Advanced SIMD three vector instructions.
4724//===----------------------------------------------------------------------===//
4725
4726defm ADD     : SIMDThreeSameVector<0, 0b10000, "add", add>;
4727defm ADDP    : SIMDThreeSameVector<0, 0b10111, "addp", AArch64addp>;
4728defm CMEQ    : SIMDThreeSameVector<1, 0b10001, "cmeq", AArch64cmeq>;
4729defm CMGE    : SIMDThreeSameVector<0, 0b00111, "cmge", AArch64cmge>;
4730defm CMGT    : SIMDThreeSameVector<0, 0b00110, "cmgt", AArch64cmgt>;
4731defm CMHI    : SIMDThreeSameVector<1, 0b00110, "cmhi", AArch64cmhi>;
4732defm CMHS    : SIMDThreeSameVector<1, 0b00111, "cmhs", AArch64cmhs>;
4733defm CMTST   : SIMDThreeSameVector<0, 0b10001, "cmtst", AArch64cmtst>;
4734foreach VT = [ v8i8, v16i8, v4i16, v8i16, v2i32, v4i32, v2i64 ] in {
4735def : Pat<(vnot (AArch64cmeqz VT:$Rn)), (!cast<Instruction>("CMTST"#VT) VT:$Rn, VT:$Rn)>;
4736}
4737defm FABD    : SIMDThreeSameVectorFP<1,1,0b010,"fabd", int_aarch64_neon_fabd>;
4738let Predicates = [HasNEON] in {
4739foreach VT = [ v2f32, v4f32, v2f64 ] in
4740def : Pat<(fabs (fsub VT:$Rn, VT:$Rm)), (!cast<Instruction>("FABD"#VT) VT:$Rn, VT:$Rm)>;
4741}
4742let Predicates = [HasNEON, HasFullFP16] in {
4743foreach VT = [ v4f16, v8f16 ] in
4744def : Pat<(fabs (fsub VT:$Rn, VT:$Rm)), (!cast<Instruction>("FABD"#VT) VT:$Rn, VT:$Rm)>;
4745}
4746defm FACGE   : SIMDThreeSameVectorFPCmp<1,0,0b101,"facge",int_aarch64_neon_facge>;
4747defm FACGT   : SIMDThreeSameVectorFPCmp<1,1,0b101,"facgt",int_aarch64_neon_facgt>;
4748defm FADDP   : SIMDThreeSameVectorFP<1,0,0b010,"faddp", AArch64faddp>;
4749defm FADD    : SIMDThreeSameVectorFP<0,0,0b010,"fadd", any_fadd>;
4750defm FCMEQ   : SIMDThreeSameVectorFPCmp<0, 0, 0b100, "fcmeq", AArch64fcmeq>;
4751defm FCMGE   : SIMDThreeSameVectorFPCmp<1, 0, 0b100, "fcmge", AArch64fcmge>;
4752defm FCMGT   : SIMDThreeSameVectorFPCmp<1, 1, 0b100, "fcmgt", AArch64fcmgt>;
4753defm FDIV    : SIMDThreeSameVectorFP<1,0,0b111,"fdiv", any_fdiv>;
4754defm FMAXNMP : SIMDThreeSameVectorFP<1,0,0b000,"fmaxnmp", int_aarch64_neon_fmaxnmp>;
4755defm FMAXNM  : SIMDThreeSameVectorFP<0,0,0b000,"fmaxnm", any_fmaxnum>;
4756defm FMAXP   : SIMDThreeSameVectorFP<1,0,0b110,"fmaxp", int_aarch64_neon_fmaxp>;
4757defm FMAX    : SIMDThreeSameVectorFP<0,0,0b110,"fmax", any_fmaximum>;
4758defm FMINNMP : SIMDThreeSameVectorFP<1,1,0b000,"fminnmp", int_aarch64_neon_fminnmp>;
4759defm FMINNM  : SIMDThreeSameVectorFP<0,1,0b000,"fminnm", any_fminnum>;
4760defm FMINP   : SIMDThreeSameVectorFP<1,1,0b110,"fminp", int_aarch64_neon_fminp>;
4761defm FMIN    : SIMDThreeSameVectorFP<0,1,0b110,"fmin", any_fminimum>;
4762
4763// NOTE: The operands of the PatFrag are reordered on FMLA/FMLS because the
4764// instruction expects the addend first, while the fma intrinsic puts it last.
4765defm FMLA     : SIMDThreeSameVectorFPTied<0, 0, 0b001, "fmla",
4766            TriOpFrag<(any_fma node:$RHS, node:$MHS, node:$LHS)> >;
4767defm FMLS     : SIMDThreeSameVectorFPTied<0, 1, 0b001, "fmls",
4768            TriOpFrag<(any_fma node:$MHS, (fneg node:$RHS), node:$LHS)> >;
4769
4770defm FMULX    : SIMDThreeSameVectorFP<0,0,0b011,"fmulx", int_aarch64_neon_fmulx>;
4771defm FMUL     : SIMDThreeSameVectorFP<1,0,0b011,"fmul", any_fmul>;
4772defm FRECPS   : SIMDThreeSameVectorFP<0,0,0b111,"frecps", int_aarch64_neon_frecps>;
4773defm FRSQRTS  : SIMDThreeSameVectorFP<0,1,0b111,"frsqrts", int_aarch64_neon_frsqrts>;
4774defm FSUB     : SIMDThreeSameVectorFP<0,1,0b010,"fsub", any_fsub>;
4775
4776// MLA and MLS are generated in MachineCombine
4777defm MLA      : SIMDThreeSameVectorBHSTied<0, 0b10010, "mla", null_frag>;
4778defm MLS      : SIMDThreeSameVectorBHSTied<1, 0b10010, "mls", null_frag>;
4779
4780defm MUL      : SIMDThreeSameVectorBHS<0, 0b10011, "mul", mul>;
4781defm PMUL     : SIMDThreeSameVectorB<1, 0b10011, "pmul", int_aarch64_neon_pmul>;
4782defm SABA     : SIMDThreeSameVectorBHSTied<0, 0b01111, "saba",
4783      TriOpFrag<(add node:$LHS, (AArch64sabd node:$MHS, node:$RHS))> >;
4784defm SABD     : SIMDThreeSameVectorBHS<0,0b01110,"sabd", AArch64sabd>;
4785defm SHADD    : SIMDThreeSameVectorBHS<0,0b00000,"shadd", avgfloors>;
4786defm SHSUB    : SIMDThreeSameVectorBHS<0,0b00100,"shsub", int_aarch64_neon_shsub>;
4787defm SMAXP    : SIMDThreeSameVectorBHS<0,0b10100,"smaxp", int_aarch64_neon_smaxp>;
4788defm SMAX     : SIMDThreeSameVectorBHS<0,0b01100,"smax", smax>;
4789defm SMINP    : SIMDThreeSameVectorBHS<0,0b10101,"sminp", int_aarch64_neon_sminp>;
4790defm SMIN     : SIMDThreeSameVectorBHS<0,0b01101,"smin", smin>;
4791defm SQADD    : SIMDThreeSameVector<0,0b00001,"sqadd", int_aarch64_neon_sqadd>;
4792defm SQDMULH  : SIMDThreeSameVectorHS<0,0b10110,"sqdmulh",int_aarch64_neon_sqdmulh>;
4793defm SQRDMULH : SIMDThreeSameVectorHS<1,0b10110,"sqrdmulh",int_aarch64_neon_sqrdmulh>;
4794defm SQRSHL   : SIMDThreeSameVector<0,0b01011,"sqrshl", int_aarch64_neon_sqrshl>;
4795defm SQSHL    : SIMDThreeSameVector<0,0b01001,"sqshl", int_aarch64_neon_sqshl>;
4796defm SQSUB    : SIMDThreeSameVector<0,0b00101,"sqsub", int_aarch64_neon_sqsub>;
4797defm SRHADD   : SIMDThreeSameVectorBHS<0,0b00010,"srhadd", avgceils>;
4798defm SRSHL    : SIMDThreeSameVector<0,0b01010,"srshl", int_aarch64_neon_srshl>;
4799defm SSHL     : SIMDThreeSameVector<0,0b01000,"sshl", int_aarch64_neon_sshl>;
4800defm SUB      : SIMDThreeSameVector<1,0b10000,"sub", sub>;
4801defm UABA     : SIMDThreeSameVectorBHSTied<1, 0b01111, "uaba",
4802      TriOpFrag<(add node:$LHS, (AArch64uabd node:$MHS, node:$RHS))> >;
4803defm UABD     : SIMDThreeSameVectorBHS<1,0b01110,"uabd", AArch64uabd>;
4804defm UHADD    : SIMDThreeSameVectorBHS<1,0b00000,"uhadd", avgflooru>;
4805defm UHSUB    : SIMDThreeSameVectorBHS<1,0b00100,"uhsub", int_aarch64_neon_uhsub>;
4806defm UMAXP    : SIMDThreeSameVectorBHS<1,0b10100,"umaxp", int_aarch64_neon_umaxp>;
4807defm UMAX     : SIMDThreeSameVectorBHS<1,0b01100,"umax", umax>;
4808defm UMINP    : SIMDThreeSameVectorBHS<1,0b10101,"uminp", int_aarch64_neon_uminp>;
4809defm UMIN     : SIMDThreeSameVectorBHS<1,0b01101,"umin", umin>;
4810defm UQADD    : SIMDThreeSameVector<1,0b00001,"uqadd", int_aarch64_neon_uqadd>;
4811defm UQRSHL   : SIMDThreeSameVector<1,0b01011,"uqrshl", int_aarch64_neon_uqrshl>;
4812defm UQSHL    : SIMDThreeSameVector<1,0b01001,"uqshl", int_aarch64_neon_uqshl>;
4813defm UQSUB    : SIMDThreeSameVector<1,0b00101,"uqsub", int_aarch64_neon_uqsub>;
4814defm URHADD   : SIMDThreeSameVectorBHS<1,0b00010,"urhadd", avgceilu>;
4815defm URSHL    : SIMDThreeSameVector<1,0b01010,"urshl", int_aarch64_neon_urshl>;
4816defm USHL     : SIMDThreeSameVector<1,0b01000,"ushl", int_aarch64_neon_ushl>;
4817defm SQRDMLAH : SIMDThreeSameVectorSQRDMLxHTiedHS<1,0b10000,"sqrdmlah",
4818                                                  int_aarch64_neon_sqrdmlah>;
4819defm SQRDMLSH : SIMDThreeSameVectorSQRDMLxHTiedHS<1,0b10001,"sqrdmlsh",
4820                                                    int_aarch64_neon_sqrdmlsh>;
4821
4822// Extra saturate patterns, other than the intrinsics matches above
4823defm : SIMDThreeSameVectorExtraPatterns<"SQADD", saddsat>;
4824defm : SIMDThreeSameVectorExtraPatterns<"UQADD", uaddsat>;
4825defm : SIMDThreeSameVectorExtraPatterns<"SQSUB", ssubsat>;
4826defm : SIMDThreeSameVectorExtraPatterns<"UQSUB", usubsat>;
4827
4828defm AND : SIMDLogicalThreeVector<0, 0b00, "and", and>;
4829defm BIC : SIMDLogicalThreeVector<0, 0b01, "bic",
4830                                  BinOpFrag<(and node:$LHS, (vnot node:$RHS))> >;
4831defm EOR : SIMDLogicalThreeVector<1, 0b00, "eor", xor>;
4832defm ORN : SIMDLogicalThreeVector<0, 0b11, "orn",
4833                                  BinOpFrag<(or node:$LHS, (vnot node:$RHS))> >;
4834defm ORR : SIMDLogicalThreeVector<0, 0b10, "orr", or>;
4835
4836// Pseudo bitwise select pattern BSP.
4837// It is expanded into BSL/BIT/BIF after register allocation.
4838defm BSP : SIMDLogicalThreeVectorPseudo<TriOpFrag<(or (and node:$LHS, node:$MHS),
4839                                                      (and (vnot node:$LHS), node:$RHS))>>;
4840defm BSL : SIMDLogicalThreeVectorTied<1, 0b01, "bsl">;
4841defm BIT : SIMDLogicalThreeVectorTied<1, 0b10, "bit", AArch64bit>;
4842defm BIF : SIMDLogicalThreeVectorTied<1, 0b11, "bif">;
4843
4844def : Pat<(AArch64bsp (v8i8 V64:$Rd), V64:$Rn, V64:$Rm),
4845          (BSPv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
4846def : Pat<(AArch64bsp (v4i16 V64:$Rd), V64:$Rn, V64:$Rm),
4847          (BSPv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
4848def : Pat<(AArch64bsp (v2i32 V64:$Rd), V64:$Rn, V64:$Rm),
4849          (BSPv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
4850def : Pat<(AArch64bsp (v1i64 V64:$Rd), V64:$Rn, V64:$Rm),
4851          (BSPv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
4852
4853def : Pat<(AArch64bsp (v16i8 V128:$Rd), V128:$Rn, V128:$Rm),
4854          (BSPv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
4855def : Pat<(AArch64bsp (v8i16 V128:$Rd), V128:$Rn, V128:$Rm),
4856          (BSPv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
4857def : Pat<(AArch64bsp (v4i32 V128:$Rd), V128:$Rn, V128:$Rm),
4858          (BSPv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
4859def : Pat<(AArch64bsp (v2i64 V128:$Rd), V128:$Rn, V128:$Rm),
4860          (BSPv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
4861
4862def : InstAlias<"mov{\t$dst.16b, $src.16b|.16b\t$dst, $src}",
4863                (ORRv16i8 V128:$dst, V128:$src, V128:$src), 1>;
4864def : InstAlias<"mov{\t$dst.8h, $src.8h|.8h\t$dst, $src}",
4865                (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
4866def : InstAlias<"mov{\t$dst.4s, $src.4s|.4s\t$dst, $src}",
4867                (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
4868def : InstAlias<"mov{\t$dst.2d, $src.2d|.2d\t$dst, $src}",
4869                (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
4870
4871def : InstAlias<"mov{\t$dst.8b, $src.8b|.8b\t$dst, $src}",
4872                (ORRv8i8 V64:$dst, V64:$src, V64:$src), 1>;
4873def : InstAlias<"mov{\t$dst.4h, $src.4h|.4h\t$dst, $src}",
4874                (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
4875def : InstAlias<"mov{\t$dst.2s, $src.2s|.2s\t$dst, $src}",
4876                (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
4877def : InstAlias<"mov{\t$dst.1d, $src.1d|.1d\t$dst, $src}",
4878                (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
4879
4880def : InstAlias<"{cmls\t$dst.8b, $src1.8b, $src2.8b" #
4881                "|cmls.8b\t$dst, $src1, $src2}",
4882                (CMHSv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
4883def : InstAlias<"{cmls\t$dst.16b, $src1.16b, $src2.16b" #
4884                "|cmls.16b\t$dst, $src1, $src2}",
4885                (CMHSv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
4886def : InstAlias<"{cmls\t$dst.4h, $src1.4h, $src2.4h" #
4887                "|cmls.4h\t$dst, $src1, $src2}",
4888                (CMHSv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
4889def : InstAlias<"{cmls\t$dst.8h, $src1.8h, $src2.8h" #
4890                "|cmls.8h\t$dst, $src1, $src2}",
4891                (CMHSv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
4892def : InstAlias<"{cmls\t$dst.2s, $src1.2s, $src2.2s" #
4893                "|cmls.2s\t$dst, $src1, $src2}",
4894                (CMHSv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
4895def : InstAlias<"{cmls\t$dst.4s, $src1.4s, $src2.4s" #
4896                "|cmls.4s\t$dst, $src1, $src2}",
4897                (CMHSv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
4898def : InstAlias<"{cmls\t$dst.2d, $src1.2d, $src2.2d" #
4899                "|cmls.2d\t$dst, $src1, $src2}",
4900                (CMHSv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
4901
4902def : InstAlias<"{cmlo\t$dst.8b, $src1.8b, $src2.8b" #
4903                "|cmlo.8b\t$dst, $src1, $src2}",
4904                (CMHIv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
4905def : InstAlias<"{cmlo\t$dst.16b, $src1.16b, $src2.16b" #
4906                "|cmlo.16b\t$dst, $src1, $src2}",
4907                (CMHIv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
4908def : InstAlias<"{cmlo\t$dst.4h, $src1.4h, $src2.4h" #
4909                "|cmlo.4h\t$dst, $src1, $src2}",
4910                (CMHIv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
4911def : InstAlias<"{cmlo\t$dst.8h, $src1.8h, $src2.8h" #
4912                "|cmlo.8h\t$dst, $src1, $src2}",
4913                (CMHIv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
4914def : InstAlias<"{cmlo\t$dst.2s, $src1.2s, $src2.2s" #
4915                "|cmlo.2s\t$dst, $src1, $src2}",
4916                (CMHIv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
4917def : InstAlias<"{cmlo\t$dst.4s, $src1.4s, $src2.4s" #
4918                "|cmlo.4s\t$dst, $src1, $src2}",
4919                (CMHIv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
4920def : InstAlias<"{cmlo\t$dst.2d, $src1.2d, $src2.2d" #
4921                "|cmlo.2d\t$dst, $src1, $src2}",
4922                (CMHIv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
4923
4924def : InstAlias<"{cmle\t$dst.8b, $src1.8b, $src2.8b" #
4925                "|cmle.8b\t$dst, $src1, $src2}",
4926                (CMGEv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
4927def : InstAlias<"{cmle\t$dst.16b, $src1.16b, $src2.16b" #
4928                "|cmle.16b\t$dst, $src1, $src2}",
4929                (CMGEv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
4930def : InstAlias<"{cmle\t$dst.4h, $src1.4h, $src2.4h" #
4931                "|cmle.4h\t$dst, $src1, $src2}",
4932                (CMGEv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
4933def : InstAlias<"{cmle\t$dst.8h, $src1.8h, $src2.8h" #
4934                "|cmle.8h\t$dst, $src1, $src2}",
4935                (CMGEv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
4936def : InstAlias<"{cmle\t$dst.2s, $src1.2s, $src2.2s" #
4937                "|cmle.2s\t$dst, $src1, $src2}",
4938                (CMGEv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
4939def : InstAlias<"{cmle\t$dst.4s, $src1.4s, $src2.4s" #
4940                "|cmle.4s\t$dst, $src1, $src2}",
4941                (CMGEv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
4942def : InstAlias<"{cmle\t$dst.2d, $src1.2d, $src2.2d" #
4943                "|cmle.2d\t$dst, $src1, $src2}",
4944                (CMGEv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
4945
4946def : InstAlias<"{cmlt\t$dst.8b, $src1.8b, $src2.8b" #
4947                "|cmlt.8b\t$dst, $src1, $src2}",
4948                (CMGTv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
4949def : InstAlias<"{cmlt\t$dst.16b, $src1.16b, $src2.16b" #
4950                "|cmlt.16b\t$dst, $src1, $src2}",
4951                (CMGTv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
4952def : InstAlias<"{cmlt\t$dst.4h, $src1.4h, $src2.4h" #
4953                "|cmlt.4h\t$dst, $src1, $src2}",
4954                (CMGTv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
4955def : InstAlias<"{cmlt\t$dst.8h, $src1.8h, $src2.8h" #
4956                "|cmlt.8h\t$dst, $src1, $src2}",
4957                (CMGTv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
4958def : InstAlias<"{cmlt\t$dst.2s, $src1.2s, $src2.2s" #
4959                "|cmlt.2s\t$dst, $src1, $src2}",
4960                (CMGTv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
4961def : InstAlias<"{cmlt\t$dst.4s, $src1.4s, $src2.4s" #
4962                "|cmlt.4s\t$dst, $src1, $src2}",
4963                (CMGTv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
4964def : InstAlias<"{cmlt\t$dst.2d, $src1.2d, $src2.2d" #
4965                "|cmlt.2d\t$dst, $src1, $src2}",
4966                (CMGTv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
4967
4968let Predicates = [HasNEON, HasFullFP16] in {
4969def : InstAlias<"{fcmle\t$dst.4h, $src1.4h, $src2.4h" #
4970                "|fcmle.4h\t$dst, $src1, $src2}",
4971                (FCMGEv4f16 V64:$dst, V64:$src2, V64:$src1), 0>;
4972def : InstAlias<"{fcmle\t$dst.8h, $src1.8h, $src2.8h" #
4973                "|fcmle.8h\t$dst, $src1, $src2}",
4974                (FCMGEv8f16 V128:$dst, V128:$src2, V128:$src1), 0>;
4975}
4976def : InstAlias<"{fcmle\t$dst.2s, $src1.2s, $src2.2s" #
4977                "|fcmle.2s\t$dst, $src1, $src2}",
4978                (FCMGEv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
4979def : InstAlias<"{fcmle\t$dst.4s, $src1.4s, $src2.4s" #
4980                "|fcmle.4s\t$dst, $src1, $src2}",
4981                (FCMGEv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
4982def : InstAlias<"{fcmle\t$dst.2d, $src1.2d, $src2.2d" #
4983                "|fcmle.2d\t$dst, $src1, $src2}",
4984                (FCMGEv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
4985
4986let Predicates = [HasNEON, HasFullFP16] in {
4987def : InstAlias<"{fcmlt\t$dst.4h, $src1.4h, $src2.4h" #
4988                "|fcmlt.4h\t$dst, $src1, $src2}",
4989                (FCMGTv4f16 V64:$dst, V64:$src2, V64:$src1), 0>;
4990def : InstAlias<"{fcmlt\t$dst.8h, $src1.8h, $src2.8h" #
4991                "|fcmlt.8h\t$dst, $src1, $src2}",
4992                (FCMGTv8f16 V128:$dst, V128:$src2, V128:$src1), 0>;
4993}
4994def : InstAlias<"{fcmlt\t$dst.2s, $src1.2s, $src2.2s" #
4995                "|fcmlt.2s\t$dst, $src1, $src2}",
4996                (FCMGTv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
4997def : InstAlias<"{fcmlt\t$dst.4s, $src1.4s, $src2.4s" #
4998                "|fcmlt.4s\t$dst, $src1, $src2}",
4999                (FCMGTv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
5000def : InstAlias<"{fcmlt\t$dst.2d, $src1.2d, $src2.2d" #
5001                "|fcmlt.2d\t$dst, $src1, $src2}",
5002                (FCMGTv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
5003
5004let Predicates = [HasNEON, HasFullFP16] in {
5005def : InstAlias<"{facle\t$dst.4h, $src1.4h, $src2.4h" #
5006                "|facle.4h\t$dst, $src1, $src2}",
5007                (FACGEv4f16 V64:$dst, V64:$src2, V64:$src1), 0>;
5008def : InstAlias<"{facle\t$dst.8h, $src1.8h, $src2.8h" #
5009                "|facle.8h\t$dst, $src1, $src2}",
5010                (FACGEv8f16 V128:$dst, V128:$src2, V128:$src1), 0>;
5011}
5012def : InstAlias<"{facle\t$dst.2s, $src1.2s, $src2.2s" #
5013                "|facle.2s\t$dst, $src1, $src2}",
5014                (FACGEv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
5015def : InstAlias<"{facle\t$dst.4s, $src1.4s, $src2.4s" #
5016                "|facle.4s\t$dst, $src1, $src2}",
5017                (FACGEv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
5018def : InstAlias<"{facle\t$dst.2d, $src1.2d, $src2.2d" #
5019                "|facle.2d\t$dst, $src1, $src2}",
5020                (FACGEv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
5021
5022let Predicates = [HasNEON, HasFullFP16] in {
5023def : InstAlias<"{faclt\t$dst.4h, $src1.4h, $src2.4h" #
5024                "|faclt.4h\t$dst, $src1, $src2}",
5025                (FACGTv4f16 V64:$dst, V64:$src2, V64:$src1), 0>;
5026def : InstAlias<"{faclt\t$dst.8h, $src1.8h, $src2.8h" #
5027                "|faclt.8h\t$dst, $src1, $src2}",
5028                (FACGTv8f16 V128:$dst, V128:$src2, V128:$src1), 0>;
5029}
5030def : InstAlias<"{faclt\t$dst.2s, $src1.2s, $src2.2s" #
5031                "|faclt.2s\t$dst, $src1, $src2}",
5032                (FACGTv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
5033def : InstAlias<"{faclt\t$dst.4s, $src1.4s, $src2.4s" #
5034                "|faclt.4s\t$dst, $src1, $src2}",
5035                (FACGTv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
5036def : InstAlias<"{faclt\t$dst.2d, $src1.2d, $src2.2d" #
5037                "|faclt.2d\t$dst, $src1, $src2}",
5038                (FACGTv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
5039
5040//===----------------------------------------------------------------------===//
5041// Advanced SIMD three scalar instructions.
5042//===----------------------------------------------------------------------===//
5043
5044defm ADD      : SIMDThreeScalarD<0, 0b10000, "add", add>;
5045defm CMEQ     : SIMDThreeScalarD<1, 0b10001, "cmeq", AArch64cmeq>;
5046defm CMGE     : SIMDThreeScalarD<0, 0b00111, "cmge", AArch64cmge>;
5047defm CMGT     : SIMDThreeScalarD<0, 0b00110, "cmgt", AArch64cmgt>;
5048defm CMHI     : SIMDThreeScalarD<1, 0b00110, "cmhi", AArch64cmhi>;
5049defm CMHS     : SIMDThreeScalarD<1, 0b00111, "cmhs", AArch64cmhs>;
5050defm CMTST    : SIMDThreeScalarD<0, 0b10001, "cmtst", AArch64cmtst>;
5051defm FABD     : SIMDFPThreeScalar<1, 1, 0b010, "fabd", int_aarch64_sisd_fabd>;
5052def : Pat<(v1f64 (int_aarch64_neon_fabd (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
5053          (FABD64 FPR64:$Rn, FPR64:$Rm)>;
5054let Predicates = [HasNEON, HasFullFP16] in {
5055def : Pat<(fabs (fsub f16:$Rn, f16:$Rm)), (FABD16 f16:$Rn, f16:$Rm)>;
5056}
5057let Predicates = [HasNEON] in {
5058def : Pat<(fabs (fsub f32:$Rn, f32:$Rm)), (FABD32 f32:$Rn, f32:$Rm)>;
5059def : Pat<(fabs (fsub f64:$Rn, f64:$Rm)), (FABD64 f64:$Rn, f64:$Rm)>;
5060}
5061defm FACGE    : SIMDThreeScalarFPCmp<1, 0, 0b101, "facge",
5062                                     int_aarch64_neon_facge>;
5063defm FACGT    : SIMDThreeScalarFPCmp<1, 1, 0b101, "facgt",
5064                                     int_aarch64_neon_facgt>;
5065defm FCMEQ    : SIMDThreeScalarFPCmp<0, 0, 0b100, "fcmeq", AArch64fcmeq>;
5066defm FCMGE    : SIMDThreeScalarFPCmp<1, 0, 0b100, "fcmge", AArch64fcmge>;
5067defm FCMGT    : SIMDThreeScalarFPCmp<1, 1, 0b100, "fcmgt", AArch64fcmgt>;
5068defm FMULX    : SIMDFPThreeScalar<0, 0, 0b011, "fmulx", int_aarch64_neon_fmulx, HasNEONorSME>;
5069defm FRECPS   : SIMDFPThreeScalar<0, 0, 0b111, "frecps", int_aarch64_neon_frecps, HasNEONorSME>;
5070defm FRSQRTS  : SIMDFPThreeScalar<0, 1, 0b111, "frsqrts", int_aarch64_neon_frsqrts, HasNEONorSME>;
5071defm SQADD    : SIMDThreeScalarBHSD<0, 0b00001, "sqadd", int_aarch64_neon_sqadd>;
5072defm SQDMULH  : SIMDThreeScalarHS<  0, 0b10110, "sqdmulh", int_aarch64_neon_sqdmulh>;
5073defm SQRDMULH : SIMDThreeScalarHS<  1, 0b10110, "sqrdmulh", int_aarch64_neon_sqrdmulh>;
5074defm SQRSHL   : SIMDThreeScalarBHSD<0, 0b01011, "sqrshl",int_aarch64_neon_sqrshl>;
5075defm SQSHL    : SIMDThreeScalarBHSD<0, 0b01001, "sqshl", int_aarch64_neon_sqshl>;
5076defm SQSUB    : SIMDThreeScalarBHSD<0, 0b00101, "sqsub", int_aarch64_neon_sqsub>;
5077defm SRSHL    : SIMDThreeScalarD<   0, 0b01010, "srshl", int_aarch64_neon_srshl>;
5078defm SSHL     : SIMDThreeScalarD<   0, 0b01000, "sshl", int_aarch64_neon_sshl>;
5079defm SUB      : SIMDThreeScalarD<   1, 0b10000, "sub", sub>;
5080defm UQADD    : SIMDThreeScalarBHSD<1, 0b00001, "uqadd", int_aarch64_neon_uqadd>;
5081defm UQRSHL   : SIMDThreeScalarBHSD<1, 0b01011, "uqrshl",int_aarch64_neon_uqrshl>;
5082defm UQSHL    : SIMDThreeScalarBHSD<1, 0b01001, "uqshl", int_aarch64_neon_uqshl>;
5083defm UQSUB    : SIMDThreeScalarBHSD<1, 0b00101, "uqsub", int_aarch64_neon_uqsub>;
5084defm URSHL    : SIMDThreeScalarD<   1, 0b01010, "urshl", int_aarch64_neon_urshl>;
5085defm USHL     : SIMDThreeScalarD<   1, 0b01000, "ushl", int_aarch64_neon_ushl>;
5086let Predicates = [HasRDM] in {
5087  defm SQRDMLAH : SIMDThreeScalarHSTied<1, 0, 0b10000, "sqrdmlah">;
5088  defm SQRDMLSH : SIMDThreeScalarHSTied<1, 0, 0b10001, "sqrdmlsh">;
5089  def : Pat<(i32 (int_aarch64_neon_sqrdmlah (i32 FPR32:$Rd), (i32 FPR32:$Rn),
5090                                            (i32 FPR32:$Rm))),
5091            (SQRDMLAHv1i32 FPR32:$Rd, FPR32:$Rn, FPR32:$Rm)>;
5092  def : Pat<(i32 (int_aarch64_neon_sqrdmlsh (i32 FPR32:$Rd), (i32 FPR32:$Rn),
5093                                            (i32 FPR32:$Rm))),
5094            (SQRDMLSHv1i32 FPR32:$Rd, FPR32:$Rn, FPR32:$Rm)>;
5095}
5096
5097def : InstAlias<"cmls $dst, $src1, $src2",
5098                (CMHSv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
5099def : InstAlias<"cmle $dst, $src1, $src2",
5100                (CMGEv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
5101def : InstAlias<"cmlo $dst, $src1, $src2",
5102                (CMHIv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
5103def : InstAlias<"cmlt $dst, $src1, $src2",
5104                (CMGTv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
5105def : InstAlias<"fcmle $dst, $src1, $src2",
5106                (FCMGE32 FPR32:$dst, FPR32:$src2, FPR32:$src1), 0>;
5107def : InstAlias<"fcmle $dst, $src1, $src2",
5108                (FCMGE64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
5109def : InstAlias<"fcmlt $dst, $src1, $src2",
5110                (FCMGT32 FPR32:$dst, FPR32:$src2, FPR32:$src1), 0>;
5111def : InstAlias<"fcmlt $dst, $src1, $src2",
5112                (FCMGT64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
5113def : InstAlias<"facle $dst, $src1, $src2",
5114                (FACGE32 FPR32:$dst, FPR32:$src2, FPR32:$src1), 0>;
5115def : InstAlias<"facle $dst, $src1, $src2",
5116                (FACGE64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
5117def : InstAlias<"faclt $dst, $src1, $src2",
5118                (FACGT32 FPR32:$dst, FPR32:$src2, FPR32:$src1), 0>;
5119def : InstAlias<"faclt $dst, $src1, $src2",
5120                (FACGT64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
5121
5122//===----------------------------------------------------------------------===//
5123// Advanced SIMD three scalar instructions (mixed operands).
5124//===----------------------------------------------------------------------===//
5125defm SQDMULL  : SIMDThreeScalarMixedHS<0, 0b11010, "sqdmull",
5126                                       int_aarch64_neon_sqdmulls_scalar>;
5127defm SQDMLAL  : SIMDThreeScalarMixedTiedHS<0, 0b10010, "sqdmlal">;
5128defm SQDMLSL  : SIMDThreeScalarMixedTiedHS<0, 0b10110, "sqdmlsl">;
5129
5130def : Pat<(i64 (int_aarch64_neon_sqadd (i64 FPR64:$Rd),
5131                   (i64 (int_aarch64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
5132                                                        (i32 FPR32:$Rm))))),
5133          (SQDMLALi32 FPR64:$Rd, FPR32:$Rn, FPR32:$Rm)>;
5134def : Pat<(i64 (int_aarch64_neon_sqsub (i64 FPR64:$Rd),
5135                   (i64 (int_aarch64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
5136                                                        (i32 FPR32:$Rm))))),
5137          (SQDMLSLi32 FPR64:$Rd, FPR32:$Rn, FPR32:$Rm)>;
5138
5139//===----------------------------------------------------------------------===//
5140// Advanced SIMD two scalar instructions.
5141//===----------------------------------------------------------------------===//
5142
5143defm ABS    : SIMDTwoScalarD<    0, 0b01011, "abs", abs, [HasNoCSSC]>;
5144defm CMEQ   : SIMDCmpTwoScalarD< 0, 0b01001, "cmeq", AArch64cmeqz>;
5145defm CMGE   : SIMDCmpTwoScalarD< 1, 0b01000, "cmge", AArch64cmgez>;
5146defm CMGT   : SIMDCmpTwoScalarD< 0, 0b01000, "cmgt", AArch64cmgtz>;
5147defm CMLE   : SIMDCmpTwoScalarD< 1, 0b01001, "cmle", AArch64cmlez>;
5148defm CMLT   : SIMDCmpTwoScalarD< 0, 0b01010, "cmlt", AArch64cmltz>;
5149defm FCMEQ  : SIMDFPCmpTwoScalar<0, 1, 0b01101, "fcmeq", AArch64fcmeqz>;
5150defm FCMGE  : SIMDFPCmpTwoScalar<1, 1, 0b01100, "fcmge", AArch64fcmgez>;
5151defm FCMGT  : SIMDFPCmpTwoScalar<0, 1, 0b01100, "fcmgt", AArch64fcmgtz>;
5152defm FCMLE  : SIMDFPCmpTwoScalar<1, 1, 0b01101, "fcmle", AArch64fcmlez>;
5153defm FCMLT  : SIMDFPCmpTwoScalar<0, 1, 0b01110, "fcmlt", AArch64fcmltz>;
5154defm FCVTAS : SIMDFPTwoScalar<   0, 0, 0b11100, "fcvtas">;
5155defm FCVTAU : SIMDFPTwoScalar<   1, 0, 0b11100, "fcvtau">;
5156defm FCVTMS : SIMDFPTwoScalar<   0, 0, 0b11011, "fcvtms">;
5157defm FCVTMU : SIMDFPTwoScalar<   1, 0, 0b11011, "fcvtmu">;
5158defm FCVTNS : SIMDFPTwoScalar<   0, 0, 0b11010, "fcvtns">;
5159defm FCVTNU : SIMDFPTwoScalar<   1, 0, 0b11010, "fcvtnu">;
5160defm FCVTPS : SIMDFPTwoScalar<   0, 1, 0b11010, "fcvtps">;
5161defm FCVTPU : SIMDFPTwoScalar<   1, 1, 0b11010, "fcvtpu">;
5162def  FCVTXNv1i64 : SIMDInexactCvtTwoScalar<0b10110, "fcvtxn">;
5163defm FCVTZS : SIMDFPTwoScalar<   0, 1, 0b11011, "fcvtzs">;
5164defm FCVTZU : SIMDFPTwoScalar<   1, 1, 0b11011, "fcvtzu">;
5165defm FRECPE : SIMDFPTwoScalar<   0, 1, 0b11101, "frecpe", HasNEONorSME>;
5166defm FRECPX : SIMDFPTwoScalar<   0, 1, 0b11111, "frecpx", HasNEONorSME>;
5167defm FRSQRTE : SIMDFPTwoScalar<  1, 1, 0b11101, "frsqrte", HasNEONorSME>;
5168defm NEG    : SIMDTwoScalarD<    1, 0b01011, "neg",
5169                                 UnOpFrag<(sub immAllZerosV, node:$LHS)> >;
5170defm SCVTF  : SIMDFPTwoScalarCVT<   0, 0, 0b11101, "scvtf", AArch64sitof>;
5171defm SQABS  : SIMDTwoScalarBHSD< 0, 0b00111, "sqabs", int_aarch64_neon_sqabs>;
5172defm SQNEG  : SIMDTwoScalarBHSD< 1, 0b00111, "sqneg", int_aarch64_neon_sqneg>;
5173defm SQXTN  : SIMDTwoScalarMixedBHS< 0, 0b10100, "sqxtn", int_aarch64_neon_scalar_sqxtn>;
5174defm SQXTUN : SIMDTwoScalarMixedBHS< 1, 0b10010, "sqxtun", int_aarch64_neon_scalar_sqxtun>;
5175defm SUQADD : SIMDTwoScalarBHSDTied< 0, 0b00011, "suqadd",
5176                                     int_aarch64_neon_suqadd>;
5177defm UCVTF  : SIMDFPTwoScalarCVT<   1, 0, 0b11101, "ucvtf", AArch64uitof>;
5178defm UQXTN  : SIMDTwoScalarMixedBHS<1, 0b10100, "uqxtn", int_aarch64_neon_scalar_uqxtn>;
5179defm USQADD : SIMDTwoScalarBHSDTied< 1, 0b00011, "usqadd",
5180                                    int_aarch64_neon_usqadd>;
5181
5182def : Pat<(v1i64 (AArch64vashr (v1i64 V64:$Rn), (i32 63))),
5183          (CMLTv1i64rz V64:$Rn)>;
5184
5185def : Pat<(v1i64 (int_aarch64_neon_fcvtas (v1f64 FPR64:$Rn))),
5186          (FCVTASv1i64 FPR64:$Rn)>;
5187def : Pat<(v1i64 (int_aarch64_neon_fcvtau (v1f64 FPR64:$Rn))),
5188          (FCVTAUv1i64 FPR64:$Rn)>;
5189def : Pat<(v1i64 (int_aarch64_neon_fcvtms (v1f64 FPR64:$Rn))),
5190          (FCVTMSv1i64 FPR64:$Rn)>;
5191def : Pat<(v1i64 (int_aarch64_neon_fcvtmu (v1f64 FPR64:$Rn))),
5192          (FCVTMUv1i64 FPR64:$Rn)>;
5193def : Pat<(v1i64 (int_aarch64_neon_fcvtns (v1f64 FPR64:$Rn))),
5194          (FCVTNSv1i64 FPR64:$Rn)>;
5195def : Pat<(v1i64 (int_aarch64_neon_fcvtnu (v1f64 FPR64:$Rn))),
5196          (FCVTNUv1i64 FPR64:$Rn)>;
5197def : Pat<(v1i64 (int_aarch64_neon_fcvtps (v1f64 FPR64:$Rn))),
5198          (FCVTPSv1i64 FPR64:$Rn)>;
5199def : Pat<(v1i64 (int_aarch64_neon_fcvtpu (v1f64 FPR64:$Rn))),
5200          (FCVTPUv1i64 FPR64:$Rn)>;
5201def : Pat<(v1i64 (int_aarch64_neon_fcvtzs (v1f64 FPR64:$Rn))),
5202          (FCVTZSv1i64 FPR64:$Rn)>;
5203def : Pat<(v1i64 (int_aarch64_neon_fcvtzu (v1f64 FPR64:$Rn))),
5204          (FCVTZUv1i64 FPR64:$Rn)>;
5205
5206def : Pat<(f16 (int_aarch64_neon_frecpe (f16 FPR16:$Rn))),
5207          (FRECPEv1f16 FPR16:$Rn)>;
5208def : Pat<(f32 (int_aarch64_neon_frecpe (f32 FPR32:$Rn))),
5209          (FRECPEv1i32 FPR32:$Rn)>;
5210def : Pat<(f64 (int_aarch64_neon_frecpe (f64 FPR64:$Rn))),
5211          (FRECPEv1i64 FPR64:$Rn)>;
5212def : Pat<(v1f64 (int_aarch64_neon_frecpe (v1f64 FPR64:$Rn))),
5213          (FRECPEv1i64 FPR64:$Rn)>;
5214
5215def : Pat<(f32 (AArch64frecpe (f32 FPR32:$Rn))),
5216          (FRECPEv1i32 FPR32:$Rn)>;
5217def : Pat<(v2f32 (AArch64frecpe (v2f32 V64:$Rn))),
5218          (FRECPEv2f32 V64:$Rn)>;
5219def : Pat<(v4f32 (AArch64frecpe (v4f32 FPR128:$Rn))),
5220          (FRECPEv4f32 FPR128:$Rn)>;
5221def : Pat<(f64 (AArch64frecpe (f64 FPR64:$Rn))),
5222          (FRECPEv1i64 FPR64:$Rn)>;
5223def : Pat<(v1f64 (AArch64frecpe (v1f64 FPR64:$Rn))),
5224          (FRECPEv1i64 FPR64:$Rn)>;
5225def : Pat<(v2f64 (AArch64frecpe (v2f64 FPR128:$Rn))),
5226          (FRECPEv2f64 FPR128:$Rn)>;
5227
5228def : Pat<(f32 (AArch64frecps (f32 FPR32:$Rn), (f32 FPR32:$Rm))),
5229          (FRECPS32 FPR32:$Rn, FPR32:$Rm)>;
5230def : Pat<(v2f32 (AArch64frecps (v2f32 V64:$Rn), (v2f32 V64:$Rm))),
5231          (FRECPSv2f32 V64:$Rn, V64:$Rm)>;
5232def : Pat<(v4f32 (AArch64frecps (v4f32 FPR128:$Rn), (v4f32 FPR128:$Rm))),
5233          (FRECPSv4f32 FPR128:$Rn, FPR128:$Rm)>;
5234def : Pat<(f64 (AArch64frecps (f64 FPR64:$Rn), (f64 FPR64:$Rm))),
5235          (FRECPS64 FPR64:$Rn, FPR64:$Rm)>;
5236def : Pat<(v2f64 (AArch64frecps (v2f64 FPR128:$Rn), (v2f64 FPR128:$Rm))),
5237          (FRECPSv2f64 FPR128:$Rn, FPR128:$Rm)>;
5238
5239def : Pat<(f16 (int_aarch64_neon_frecpx (f16 FPR16:$Rn))),
5240          (FRECPXv1f16 FPR16:$Rn)>;
5241def : Pat<(f32 (int_aarch64_neon_frecpx (f32 FPR32:$Rn))),
5242          (FRECPXv1i32 FPR32:$Rn)>;
5243def : Pat<(f64 (int_aarch64_neon_frecpx (f64 FPR64:$Rn))),
5244          (FRECPXv1i64 FPR64:$Rn)>;
5245
5246def : Pat<(f16 (int_aarch64_neon_frsqrte (f16 FPR16:$Rn))),
5247          (FRSQRTEv1f16 FPR16:$Rn)>;
5248def : Pat<(f32 (int_aarch64_neon_frsqrte (f32 FPR32:$Rn))),
5249          (FRSQRTEv1i32 FPR32:$Rn)>;
5250def : Pat<(f64 (int_aarch64_neon_frsqrte (f64 FPR64:$Rn))),
5251          (FRSQRTEv1i64 FPR64:$Rn)>;
5252def : Pat<(v1f64 (int_aarch64_neon_frsqrte (v1f64 FPR64:$Rn))),
5253          (FRSQRTEv1i64 FPR64:$Rn)>;
5254
5255def : Pat<(f32 (AArch64frsqrte (f32 FPR32:$Rn))),
5256          (FRSQRTEv1i32 FPR32:$Rn)>;
5257def : Pat<(v2f32 (AArch64frsqrte (v2f32 V64:$Rn))),
5258          (FRSQRTEv2f32 V64:$Rn)>;
5259def : Pat<(v4f32 (AArch64frsqrte (v4f32 FPR128:$Rn))),
5260          (FRSQRTEv4f32 FPR128:$Rn)>;
5261def : Pat<(f64 (AArch64frsqrte (f64 FPR64:$Rn))),
5262          (FRSQRTEv1i64 FPR64:$Rn)>;
5263def : Pat<(v1f64 (AArch64frsqrte (v1f64 FPR64:$Rn))),
5264          (FRSQRTEv1i64 FPR64:$Rn)>;
5265def : Pat<(v2f64 (AArch64frsqrte (v2f64 FPR128:$Rn))),
5266          (FRSQRTEv2f64 FPR128:$Rn)>;
5267
5268def : Pat<(f32 (AArch64frsqrts (f32 FPR32:$Rn), (f32 FPR32:$Rm))),
5269          (FRSQRTS32 FPR32:$Rn, FPR32:$Rm)>;
5270def : Pat<(v2f32 (AArch64frsqrts (v2f32 V64:$Rn), (v2f32 V64:$Rm))),
5271          (FRSQRTSv2f32 V64:$Rn, V64:$Rm)>;
5272def : Pat<(v4f32 (AArch64frsqrts (v4f32 FPR128:$Rn), (v4f32 FPR128:$Rm))),
5273          (FRSQRTSv4f32 FPR128:$Rn, FPR128:$Rm)>;
5274def : Pat<(f64 (AArch64frsqrts (f64 FPR64:$Rn), (f64 FPR64:$Rm))),
5275          (FRSQRTS64 FPR64:$Rn, FPR64:$Rm)>;
5276def : Pat<(v2f64 (AArch64frsqrts (v2f64 FPR128:$Rn), (v2f64 FPR128:$Rm))),
5277          (FRSQRTSv2f64 FPR128:$Rn, FPR128:$Rm)>;
5278
5279// Some float -> int -> float conversion patterns for which we want to keep the
5280// int values in FP registers using the corresponding NEON instructions to
5281// avoid more costly int <-> fp register transfers.
5282let Predicates = [HasNEON] in {
5283def : Pat<(f64 (any_sint_to_fp (i64 (any_fp_to_sint f64:$Rn)))),
5284          (SCVTFv1i64 (i64 (FCVTZSv1i64 f64:$Rn)))>;
5285def : Pat<(f32 (any_sint_to_fp (i32 (any_fp_to_sint f32:$Rn)))),
5286          (SCVTFv1i32 (i32 (FCVTZSv1i32 f32:$Rn)))>;
5287def : Pat<(f64 (any_uint_to_fp (i64 (any_fp_to_uint f64:$Rn)))),
5288          (UCVTFv1i64 (i64 (FCVTZUv1i64 f64:$Rn)))>;
5289def : Pat<(f32 (any_uint_to_fp (i32 (any_fp_to_uint f32:$Rn)))),
5290          (UCVTFv1i32 (i32 (FCVTZUv1i32 f32:$Rn)))>;
5291
5292let Predicates = [HasFullFP16] in {
5293def : Pat<(f16 (any_sint_to_fp (i32 (any_fp_to_sint f16:$Rn)))),
5294          (SCVTFv1i16 (f16 (FCVTZSv1f16 f16:$Rn)))>;
5295def : Pat<(f16 (any_uint_to_fp (i32 (any_fp_to_uint f16:$Rn)))),
5296          (UCVTFv1i16 (f16 (FCVTZUv1f16 f16:$Rn)))>;
5297}
5298// If an integer is about to be converted to a floating point value,
5299// just load it on the floating point unit.
5300// Here are the patterns for 8 and 16-bits to float.
5301// 8-bits -> float.
5302multiclass UIntToFPROLoadPat<ValueType DstTy, ValueType SrcTy,
5303                             SDPatternOperator loadop, Instruction UCVTF,
5304                             ROAddrMode ro, Instruction LDRW, Instruction LDRX,
5305                             SubRegIndex sub> {
5306  def : Pat<(DstTy (uint_to_fp (SrcTy
5307                     (loadop (ro.Wpat GPR64sp:$Rn, GPR32:$Rm,
5308                                      ro.Wext:$extend))))),
5309           (UCVTF (INSERT_SUBREG (DstTy (IMPLICIT_DEF)),
5310                                 (LDRW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend),
5311                                 sub))>;
5312
5313  def : Pat<(DstTy (uint_to_fp (SrcTy
5314                     (loadop (ro.Xpat GPR64sp:$Rn, GPR64:$Rm,
5315                                      ro.Wext:$extend))))),
5316           (UCVTF (INSERT_SUBREG (DstTy (IMPLICIT_DEF)),
5317                                 (LDRX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend),
5318                                 sub))>;
5319}
5320
5321defm : UIntToFPROLoadPat<f32, i32, zextloadi8,
5322                         UCVTFv1i32, ro8, LDRBroW, LDRBroX, bsub>;
5323def : Pat <(f32 (uint_to_fp (i32
5324               (zextloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))))),
5325           (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
5326                          (LDRBui GPR64sp:$Rn, uimm12s1:$offset), bsub))>;
5327def : Pat <(f32 (uint_to_fp (i32
5328                     (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))))),
5329           (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
5330                          (LDURBi GPR64sp:$Rn, simm9:$offset), bsub))>;
5331// 16-bits -> float.
5332defm : UIntToFPROLoadPat<f32, i32, zextloadi16,
5333                         UCVTFv1i32, ro16, LDRHroW, LDRHroX, hsub>;
5334def : Pat <(f32 (uint_to_fp (i32
5335                  (zextloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))))),
5336           (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
5337                          (LDRHui GPR64sp:$Rn, uimm12s2:$offset), hsub))>;
5338def : Pat <(f32 (uint_to_fp (i32
5339                  (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))))),
5340           (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
5341                          (LDURHi GPR64sp:$Rn, simm9:$offset), hsub))>;
5342// 32-bits are handled in target specific dag combine:
5343// performIntToFpCombine.
5344// 64-bits integer to 32-bits floating point, not possible with
5345// UCVTF on floating point registers (both source and destination
5346// must have the same size).
5347
5348// Here are the patterns for 8, 16, 32, and 64-bits to double.
5349// 8-bits -> double.
5350defm : UIntToFPROLoadPat<f64, i32, zextloadi8,
5351                         UCVTFv1i64, ro8, LDRBroW, LDRBroX, bsub>;
5352def : Pat <(f64 (uint_to_fp (i32
5353                    (zextloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))))),
5354           (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
5355                          (LDRBui GPR64sp:$Rn, uimm12s1:$offset), bsub))>;
5356def : Pat <(f64 (uint_to_fp (i32
5357                  (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))))),
5358           (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
5359                          (LDURBi GPR64sp:$Rn, simm9:$offset), bsub))>;
5360// 16-bits -> double.
5361defm : UIntToFPROLoadPat<f64, i32, zextloadi16,
5362                         UCVTFv1i64, ro16, LDRHroW, LDRHroX, hsub>;
5363def : Pat <(f64 (uint_to_fp (i32
5364                  (zextloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))))),
5365           (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
5366                          (LDRHui GPR64sp:$Rn, uimm12s2:$offset), hsub))>;
5367def : Pat <(f64 (uint_to_fp (i32
5368                  (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))))),
5369           (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
5370                          (LDURHi GPR64sp:$Rn, simm9:$offset), hsub))>;
5371// 32-bits -> double.
5372defm : UIntToFPROLoadPat<f64, i32, load,
5373                         UCVTFv1i64, ro32, LDRSroW, LDRSroX, ssub>;
5374def : Pat <(f64 (uint_to_fp (i32
5375                  (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))))),
5376           (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
5377                          (LDRSui GPR64sp:$Rn, uimm12s4:$offset), ssub))>;
5378def : Pat <(f64 (uint_to_fp (i32
5379                  (load (am_unscaled32 GPR64sp:$Rn, simm9:$offset))))),
5380           (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
5381                          (LDURSi GPR64sp:$Rn, simm9:$offset), ssub))>;
5382// 64-bits -> double are handled in target specific dag combine:
5383// performIntToFpCombine.
5384} // let Predicates = [HasNEON]
5385
5386//===----------------------------------------------------------------------===//
5387// Advanced SIMD three different-sized vector instructions.
5388//===----------------------------------------------------------------------===//
5389
5390defm ADDHN  : SIMDNarrowThreeVectorBHS<0,0b0100,"addhn", int_aarch64_neon_addhn>;
5391defm SUBHN  : SIMDNarrowThreeVectorBHS<0,0b0110,"subhn", int_aarch64_neon_subhn>;
5392defm RADDHN : SIMDNarrowThreeVectorBHS<1,0b0100,"raddhn",int_aarch64_neon_raddhn>;
5393defm RSUBHN : SIMDNarrowThreeVectorBHS<1,0b0110,"rsubhn",int_aarch64_neon_rsubhn>;
5394defm PMULL  : SIMDDifferentThreeVectorBD<0,0b1110,"pmull", AArch64pmull>;
5395defm SABAL  : SIMDLongThreeVectorTiedBHSabal<0,0b0101,"sabal",
5396                                             AArch64sabd>;
5397defm SABDL   : SIMDLongThreeVectorBHSabdl<0, 0b0111, "sabdl",
5398                                          AArch64sabd>;
5399defm SADDL   : SIMDLongThreeVectorBHS<   0, 0b0000, "saddl",
5400            BinOpFrag<(add (sext node:$LHS), (sext node:$RHS))>>;
5401defm SADDW   : SIMDWideThreeVectorBHS<   0, 0b0001, "saddw",
5402                 BinOpFrag<(add node:$LHS, (sext node:$RHS))>>;
5403defm SMLAL   : SIMDLongThreeVectorTiedBHS<0, 0b1000, "smlal",
5404    TriOpFrag<(add node:$LHS, (AArch64smull node:$MHS, node:$RHS))>>;
5405defm SMLSL   : SIMDLongThreeVectorTiedBHS<0, 0b1010, "smlsl",
5406    TriOpFrag<(sub node:$LHS, (AArch64smull node:$MHS, node:$RHS))>>;
5407defm SMULL   : SIMDLongThreeVectorBHS<0, 0b1100, "smull", AArch64smull>;
5408defm SQDMLAL : SIMDLongThreeVectorSQDMLXTiedHS<0, 0b1001, "sqdmlal",
5409                                               int_aarch64_neon_sqadd>;
5410defm SQDMLSL : SIMDLongThreeVectorSQDMLXTiedHS<0, 0b1011, "sqdmlsl",
5411                                               int_aarch64_neon_sqsub>;
5412defm SQDMULL : SIMDLongThreeVectorHS<0, 0b1101, "sqdmull",
5413                                     int_aarch64_neon_sqdmull>;
5414defm SSUBL   : SIMDLongThreeVectorBHS<0, 0b0010, "ssubl",
5415                 BinOpFrag<(sub (sext node:$LHS), (sext node:$RHS))>>;
5416defm SSUBW   : SIMDWideThreeVectorBHS<0, 0b0011, "ssubw",
5417                 BinOpFrag<(sub node:$LHS, (sext node:$RHS))>>;
5418defm UABAL   : SIMDLongThreeVectorTiedBHSabal<1, 0b0101, "uabal",
5419                                              AArch64uabd>;
5420defm UADDL   : SIMDLongThreeVectorBHS<1, 0b0000, "uaddl",
5421                 BinOpFrag<(add (zanyext node:$LHS), (zanyext node:$RHS))>>;
5422defm UADDW   : SIMDWideThreeVectorBHS<1, 0b0001, "uaddw",
5423                 BinOpFrag<(add node:$LHS, (zanyext node:$RHS))>>;
5424defm UMLAL   : SIMDLongThreeVectorTiedBHS<1, 0b1000, "umlal",
5425    TriOpFrag<(add node:$LHS, (AArch64umull node:$MHS, node:$RHS))>>;
5426defm UMLSL   : SIMDLongThreeVectorTiedBHS<1, 0b1010, "umlsl",
5427    TriOpFrag<(sub node:$LHS, (AArch64umull node:$MHS, node:$RHS))>>;
5428defm UMULL   : SIMDLongThreeVectorBHS<1, 0b1100, "umull", AArch64umull>;
5429defm USUBL   : SIMDLongThreeVectorBHS<1, 0b0010, "usubl",
5430                 BinOpFrag<(sub (zanyext node:$LHS), (zanyext node:$RHS))>>;
5431defm USUBW   : SIMDWideThreeVectorBHS<   1, 0b0011, "usubw",
5432                 BinOpFrag<(sub node:$LHS, (zanyext node:$RHS))>>;
5433
5434// Additional patterns for [SU]ML[AS]L
5435multiclass Neon_mul_acc_widen_patterns<SDPatternOperator opnode, SDPatternOperator vecopnode,
5436  Instruction INST8B, Instruction INST4H, Instruction INST2S> {
5437  def : Pat<(v4i16 (opnode
5438                    V64:$Ra,
5439                    (v4i16 (extract_subvector
5440                            (vecopnode (v8i8 V64:$Rn),(v8i8 V64:$Rm)),
5441                            (i64 0))))),
5442             (EXTRACT_SUBREG (v8i16 (INST8B
5443                                     (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), V64:$Ra, dsub),
5444                                     V64:$Rn, V64:$Rm)), dsub)>;
5445  def : Pat<(v2i32 (opnode
5446                    V64:$Ra,
5447                    (v2i32 (extract_subvector
5448                            (vecopnode (v4i16 V64:$Rn),(v4i16 V64:$Rm)),
5449                            (i64 0))))),
5450             (EXTRACT_SUBREG (v4i32 (INST4H
5451                                     (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), V64:$Ra, dsub),
5452                                     V64:$Rn, V64:$Rm)), dsub)>;
5453  def : Pat<(v1i64 (opnode
5454                    V64:$Ra,
5455                    (v1i64 (extract_subvector
5456                            (vecopnode (v2i32 V64:$Rn),(v2i32 V64:$Rm)),
5457                            (i64 0))))),
5458             (EXTRACT_SUBREG (v2i64 (INST2S
5459                                     (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), V64:$Ra, dsub),
5460                                     V64:$Rn, V64:$Rm)), dsub)>;
5461}
5462
5463defm : Neon_mul_acc_widen_patterns<add, AArch64umull,
5464     UMLALv8i8_v8i16, UMLALv4i16_v4i32, UMLALv2i32_v2i64>;
5465defm : Neon_mul_acc_widen_patterns<add, AArch64smull,
5466     SMLALv8i8_v8i16, SMLALv4i16_v4i32, SMLALv2i32_v2i64>;
5467defm : Neon_mul_acc_widen_patterns<sub, AArch64umull,
5468     UMLSLv8i8_v8i16, UMLSLv4i16_v4i32, UMLSLv2i32_v2i64>;
5469defm : Neon_mul_acc_widen_patterns<sub, AArch64smull,
5470     SMLSLv8i8_v8i16, SMLSLv4i16_v4i32, SMLSLv2i32_v2i64>;
5471
5472// CodeGen patterns for addhn and subhn instructions, which can actually be
5473// written in LLVM IR without too much difficulty.
5474
5475// Prioritize ADDHN and SUBHN over UZP2.
5476let AddedComplexity = 10 in {
5477
5478// ADDHN
5479def : Pat<(v8i8 (trunc (v8i16 (AArch64vlshr (add V128:$Rn, V128:$Rm), (i32 8))))),
5480          (ADDHNv8i16_v8i8 V128:$Rn, V128:$Rm)>;
5481def : Pat<(v4i16 (trunc (v4i32 (AArch64vlshr (add V128:$Rn, V128:$Rm),
5482                                           (i32 16))))),
5483          (ADDHNv4i32_v4i16 V128:$Rn, V128:$Rm)>;
5484def : Pat<(v2i32 (trunc (v2i64 (AArch64vlshr (add V128:$Rn, V128:$Rm),
5485                                           (i32 32))))),
5486          (ADDHNv2i64_v2i32 V128:$Rn, V128:$Rm)>;
5487def : Pat<(concat_vectors (v8i8 V64:$Rd),
5488                          (trunc (v8i16 (AArch64vlshr (add V128:$Rn, V128:$Rm),
5489                                                    (i32 8))))),
5490          (ADDHNv8i16_v16i8 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
5491                            V128:$Rn, V128:$Rm)>;
5492def : Pat<(concat_vectors (v4i16 V64:$Rd),
5493                          (trunc (v4i32 (AArch64vlshr (add V128:$Rn, V128:$Rm),
5494                                                    (i32 16))))),
5495          (ADDHNv4i32_v8i16 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
5496                            V128:$Rn, V128:$Rm)>;
5497def : Pat<(concat_vectors (v2i32 V64:$Rd),
5498                          (trunc (v2i64 (AArch64vlshr (add V128:$Rn, V128:$Rm),
5499                                                    (i32 32))))),
5500          (ADDHNv2i64_v4i32 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
5501                            V128:$Rn, V128:$Rm)>;
5502
5503// SUBHN
5504def : Pat<(v8i8 (trunc (v8i16 (AArch64vlshr (sub V128:$Rn, V128:$Rm), (i32 8))))),
5505          (SUBHNv8i16_v8i8 V128:$Rn, V128:$Rm)>;
5506def : Pat<(v4i16 (trunc (v4i32 (AArch64vlshr (sub V128:$Rn, V128:$Rm),
5507                                           (i32 16))))),
5508          (SUBHNv4i32_v4i16 V128:$Rn, V128:$Rm)>;
5509def : Pat<(v2i32 (trunc (v2i64 (AArch64vlshr (sub V128:$Rn, V128:$Rm),
5510                                           (i32 32))))),
5511          (SUBHNv2i64_v2i32 V128:$Rn, V128:$Rm)>;
5512def : Pat<(concat_vectors (v8i8 V64:$Rd),
5513                          (trunc (v8i16 (AArch64vlshr (sub V128:$Rn, V128:$Rm),
5514                                                    (i32 8))))),
5515          (SUBHNv8i16_v16i8 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
5516                            V128:$Rn, V128:$Rm)>;
5517def : Pat<(concat_vectors (v4i16 V64:$Rd),
5518                          (trunc (v4i32 (AArch64vlshr (sub V128:$Rn, V128:$Rm),
5519                                                    (i32 16))))),
5520          (SUBHNv4i32_v8i16 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
5521                            V128:$Rn, V128:$Rm)>;
5522def : Pat<(concat_vectors (v2i32 V64:$Rd),
5523                          (trunc (v2i64 (AArch64vlshr (sub V128:$Rn, V128:$Rm),
5524                                                    (i32 32))))),
5525          (SUBHNv2i64_v4i32 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
5526                            V128:$Rn, V128:$Rm)>;
5527
5528} // AddedComplexity = 10
5529
5530//----------------------------------------------------------------------------
5531// AdvSIMD bitwise extract from vector instruction.
5532//----------------------------------------------------------------------------
5533
5534defm EXT : SIMDBitwiseExtract<"ext">;
5535
5536def AdjustExtImm : SDNodeXForm<imm, [{
5537  return CurDAG->getTargetConstant(8 + N->getZExtValue(), SDLoc(N), MVT::i32);
5538}]>;
5539multiclass ExtPat<ValueType VT64, ValueType VT128, int N> {
5540  def : Pat<(VT64 (AArch64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
5541            (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
5542  def : Pat<(VT128 (AArch64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
5543            (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
5544  // We use EXT to handle extract_subvector to copy the upper 64-bits of a
5545  // 128-bit vector.
5546  def : Pat<(VT64 (extract_subvector V128:$Rn, (i64 N))),
5547            (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
5548  // A 64-bit EXT of two halves of the same 128-bit register can be done as a
5549  // single 128-bit EXT.
5550  def : Pat<(VT64 (AArch64ext (extract_subvector V128:$Rn, (i64 0)),
5551                              (extract_subvector V128:$Rn, (i64 N)),
5552                              (i32 imm:$imm))),
5553            (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, imm:$imm), dsub)>;
5554  // A 64-bit EXT of the high half of a 128-bit register can be done using a
5555  // 128-bit EXT of the whole register with an adjustment to the immediate. The
5556  // top half of the other operand will be unset, but that doesn't matter as it
5557  // will not be used.
5558  def : Pat<(VT64 (AArch64ext (extract_subvector V128:$Rn, (i64 N)),
5559                              V64:$Rm,
5560                              (i32 imm:$imm))),
5561            (EXTRACT_SUBREG (EXTv16i8 V128:$Rn,
5562                                      (SUBREG_TO_REG (i32 0), V64:$Rm, dsub),
5563                                      (AdjustExtImm imm:$imm)), dsub)>;
5564}
5565
5566defm : ExtPat<v8i8, v16i8, 8>;
5567defm : ExtPat<v4i16, v8i16, 4>;
5568defm : ExtPat<v4f16, v8f16, 4>;
5569defm : ExtPat<v4bf16, v8bf16, 4>;
5570defm : ExtPat<v2i32, v4i32, 2>;
5571defm : ExtPat<v2f32, v4f32, 2>;
5572defm : ExtPat<v1i64, v2i64, 1>;
5573defm : ExtPat<v1f64, v2f64, 1>;
5574
5575//----------------------------------------------------------------------------
5576// AdvSIMD zip vector
5577//----------------------------------------------------------------------------
5578
5579defm TRN1 : SIMDZipVector<0b010, "trn1", AArch64trn1>;
5580defm TRN2 : SIMDZipVector<0b110, "trn2", AArch64trn2>;
5581defm UZP1 : SIMDZipVector<0b001, "uzp1", AArch64uzp1>;
5582defm UZP2 : SIMDZipVector<0b101, "uzp2", AArch64uzp2>;
5583defm ZIP1 : SIMDZipVector<0b011, "zip1", AArch64zip1>;
5584defm ZIP2 : SIMDZipVector<0b111, "zip2", AArch64zip2>;
5585
5586def : Pat<(v16i8 (concat_vectors (v8i8 (trunc (v8i16 V128:$Vn))),
5587                                 (v8i8 (trunc (v8i16 V128:$Vm))))),
5588          (UZP1v16i8 V128:$Vn, V128:$Vm)>;
5589def : Pat<(v8i16 (concat_vectors (v4i16 (trunc (v4i32 V128:$Vn))),
5590                                 (v4i16 (trunc (v4i32 V128:$Vm))))),
5591          (UZP1v8i16 V128:$Vn, V128:$Vm)>;
5592def : Pat<(v4i32 (concat_vectors (v2i32 (trunc (v2i64 V128:$Vn))),
5593                                 (v2i32 (trunc (v2i64 V128:$Vm))))),
5594          (UZP1v4i32 V128:$Vn, V128:$Vm)>;
5595
5596def : Pat<(v16i8 (concat_vectors
5597                 (v8i8 (trunc (AArch64vlshr (v8i16 V128:$Vn), (i32 8)))),
5598                 (v8i8 (trunc (AArch64vlshr (v8i16 V128:$Vm), (i32 8)))))),
5599          (UZP2v16i8 V128:$Vn, V128:$Vm)>;
5600def : Pat<(v8i16 (concat_vectors
5601                 (v4i16 (trunc (AArch64vlshr (v4i32 V128:$Vn), (i32 16)))),
5602                 (v4i16 (trunc (AArch64vlshr (v4i32 V128:$Vm), (i32 16)))))),
5603          (UZP2v8i16 V128:$Vn, V128:$Vm)>;
5604def : Pat<(v4i32 (concat_vectors
5605                 (v2i32 (trunc (AArch64vlshr (v2i64 V128:$Vn), (i32 32)))),
5606                 (v2i32 (trunc (AArch64vlshr (v2i64 V128:$Vm), (i32 32)))))),
5607          (UZP2v4i32 V128:$Vn, V128:$Vm)>;
5608
5609//----------------------------------------------------------------------------
5610// AdvSIMD TBL/TBX instructions
5611//----------------------------------------------------------------------------
5612
5613defm TBL : SIMDTableLookup<    0, "tbl">;
5614defm TBX : SIMDTableLookupTied<1, "tbx">;
5615
5616def : Pat<(v8i8 (int_aarch64_neon_tbl1 (v16i8 VecListOne128:$Rn), (v8i8 V64:$Ri))),
5617          (TBLv8i8One VecListOne128:$Rn, V64:$Ri)>;
5618def : Pat<(v16i8 (int_aarch64_neon_tbl1 (v16i8 V128:$Ri), (v16i8 V128:$Rn))),
5619          (TBLv16i8One V128:$Ri, V128:$Rn)>;
5620
5621def : Pat<(v8i8 (int_aarch64_neon_tbx1 (v8i8 V64:$Rd),
5622                  (v16i8 VecListOne128:$Rn), (v8i8 V64:$Ri))),
5623          (TBXv8i8One V64:$Rd, VecListOne128:$Rn, V64:$Ri)>;
5624def : Pat<(v16i8 (int_aarch64_neon_tbx1 (v16i8 V128:$Rd),
5625                   (v16i8 V128:$Ri), (v16i8 V128:$Rn))),
5626          (TBXv16i8One V128:$Rd, V128:$Ri, V128:$Rn)>;
5627
5628
5629//----------------------------------------------------------------------------
5630// AdvSIMD scalar DUP instruction
5631//----------------------------------------------------------------------------
5632
5633defm DUP : SIMDScalarDUP<"mov">;
5634
5635//----------------------------------------------------------------------------
5636// AdvSIMD scalar pairwise instructions
5637//----------------------------------------------------------------------------
5638
5639defm ADDP    : SIMDPairwiseScalarD<0, 0b11011, "addp">;
5640defm FADDP   : SIMDFPPairwiseScalar<0, 0b01101, "faddp">;
5641defm FMAXNMP : SIMDFPPairwiseScalar<0, 0b01100, "fmaxnmp">;
5642defm FMAXP   : SIMDFPPairwiseScalar<0, 0b01111, "fmaxp">;
5643defm FMINNMP : SIMDFPPairwiseScalar<1, 0b01100, "fminnmp">;
5644defm FMINP   : SIMDFPPairwiseScalar<1, 0b01111, "fminp">;
5645
5646// Only the lower half of the result of the inner FADDP is used in the patterns
5647// below, so the second operand does not matter. Re-use the first input
5648// operand, so no additional dependencies need to be introduced.
5649let Predicates = [HasFullFP16] in {
5650def : Pat<(f16 (vecreduce_fadd (v8f16 V128:$Rn))),
5651            (FADDPv2i16p
5652              (EXTRACT_SUBREG
5653                 (FADDPv8f16 (FADDPv8f16 V128:$Rn, V128:$Rn), V128:$Rn),
5654               dsub))>;
5655def : Pat<(f16 (vecreduce_fadd (v4f16 V64:$Rn))),
5656          (FADDPv2i16p (FADDPv4f16 V64:$Rn, V64:$Rn))>;
5657}
5658def : Pat<(f32 (vecreduce_fadd (v4f32 V128:$Rn))),
5659          (FADDPv2i32p
5660            (EXTRACT_SUBREG
5661              (FADDPv4f32 V128:$Rn, V128:$Rn),
5662             dsub))>;
5663def : Pat<(f32 (vecreduce_fadd (v2f32 V64:$Rn))),
5664          (FADDPv2i32p V64:$Rn)>;
5665def : Pat<(f64 (vecreduce_fadd (v2f64 V128:$Rn))),
5666          (FADDPv2i64p V128:$Rn)>;
5667
5668def : Pat<(v2i64 (AArch64saddv V128:$Rn)),
5669          (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), (ADDPv2i64p V128:$Rn), dsub)>;
5670def : Pat<(v2i64 (AArch64uaddv V128:$Rn)),
5671          (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), (ADDPv2i64p V128:$Rn), dsub)>;
5672def : Pat<(f32 (int_aarch64_neon_faddv (v2f32 V64:$Rn))),
5673          (FADDPv2i32p V64:$Rn)>;
5674def : Pat<(f32 (int_aarch64_neon_faddv (v4f32 V128:$Rn))),
5675          (FADDPv2i32p (EXTRACT_SUBREG (FADDPv4f32 V128:$Rn, V128:$Rn), dsub))>;
5676def : Pat<(f64 (int_aarch64_neon_faddv (v2f64 V128:$Rn))),
5677          (FADDPv2i64p V128:$Rn)>;
5678def : Pat<(f32 (int_aarch64_neon_fmaxnmv (v2f32 V64:$Rn))),
5679          (FMAXNMPv2i32p V64:$Rn)>;
5680def : Pat<(f64 (int_aarch64_neon_fmaxnmv (v2f64 V128:$Rn))),
5681          (FMAXNMPv2i64p V128:$Rn)>;
5682def : Pat<(f32 (int_aarch64_neon_fmaxv (v2f32 V64:$Rn))),
5683          (FMAXPv2i32p V64:$Rn)>;
5684def : Pat<(f64 (int_aarch64_neon_fmaxv (v2f64 V128:$Rn))),
5685          (FMAXPv2i64p V128:$Rn)>;
5686def : Pat<(f32 (int_aarch64_neon_fminnmv (v2f32 V64:$Rn))),
5687          (FMINNMPv2i32p V64:$Rn)>;
5688def : Pat<(f64 (int_aarch64_neon_fminnmv (v2f64 V128:$Rn))),
5689          (FMINNMPv2i64p V128:$Rn)>;
5690def : Pat<(f32 (int_aarch64_neon_fminv (v2f32 V64:$Rn))),
5691          (FMINPv2i32p V64:$Rn)>;
5692def : Pat<(f64 (int_aarch64_neon_fminv (v2f64 V128:$Rn))),
5693          (FMINPv2i64p V128:$Rn)>;
5694
5695//----------------------------------------------------------------------------
5696// AdvSIMD INS/DUP instructions
5697//----------------------------------------------------------------------------
5698
5699def DUPv8i8gpr  : SIMDDupFromMain<0, {?,?,?,?,1}, ".8b", v8i8, V64, GPR32>;
5700def DUPv16i8gpr : SIMDDupFromMain<1, {?,?,?,?,1}, ".16b", v16i8, V128, GPR32>;
5701def DUPv4i16gpr : SIMDDupFromMain<0, {?,?,?,1,0}, ".4h", v4i16, V64, GPR32>;
5702def DUPv8i16gpr : SIMDDupFromMain<1, {?,?,?,1,0}, ".8h", v8i16, V128, GPR32>;
5703def DUPv2i32gpr : SIMDDupFromMain<0, {?,?,1,0,0}, ".2s", v2i32, V64, GPR32>;
5704def DUPv4i32gpr : SIMDDupFromMain<1, {?,?,1,0,0}, ".4s", v4i32, V128, GPR32>;
5705def DUPv2i64gpr : SIMDDupFromMain<1, {?,1,0,0,0}, ".2d", v2i64, V128, GPR64>;
5706
5707def DUPv2i64lane : SIMDDup64FromElement;
5708def DUPv2i32lane : SIMDDup32FromElement<0, ".2s", v2i32, V64>;
5709def DUPv4i32lane : SIMDDup32FromElement<1, ".4s", v4i32, V128>;
5710def DUPv4i16lane : SIMDDup16FromElement<0, ".4h", v4i16, V64>;
5711def DUPv8i16lane : SIMDDup16FromElement<1, ".8h", v8i16, V128>;
5712def DUPv8i8lane  : SIMDDup8FromElement <0, ".8b", v8i8, V64>;
5713def DUPv16i8lane : SIMDDup8FromElement <1, ".16b", v16i8, V128>;
5714
5715// DUP from a 64-bit register to a 64-bit register is just a copy
5716def : Pat<(v1i64 (AArch64dup (i64 GPR64:$Rn))),
5717          (COPY_TO_REGCLASS GPR64:$Rn, FPR64)>;
5718def : Pat<(v1f64 (AArch64dup (f64 FPR64:$Rn))),
5719          (COPY_TO_REGCLASS FPR64:$Rn, FPR64)>;
5720
5721def : Pat<(v2f32 (AArch64dup (f32 FPR32:$Rn))),
5722          (v2f32 (DUPv2i32lane
5723            (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rn, ssub),
5724            (i64 0)))>;
5725def : Pat<(v4f32 (AArch64dup (f32 FPR32:$Rn))),
5726          (v4f32 (DUPv4i32lane
5727            (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rn, ssub),
5728            (i64 0)))>;
5729def : Pat<(v2f64 (AArch64dup (f64 FPR64:$Rn))),
5730          (v2f64 (DUPv2i64lane
5731            (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$Rn, dsub),
5732            (i64 0)))>;
5733def : Pat<(v4f16 (AArch64dup (f16 FPR16:$Rn))),
5734          (v4f16 (DUPv4i16lane
5735            (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), FPR16:$Rn, hsub),
5736            (i64 0)))>;
5737def : Pat<(v4bf16 (AArch64dup (bf16 FPR16:$Rn))),
5738          (v4bf16 (DUPv4i16lane
5739            (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), FPR16:$Rn, hsub),
5740            (i64 0)))>;
5741def : Pat<(v8f16 (AArch64dup (f16 FPR16:$Rn))),
5742          (v8f16 (DUPv8i16lane
5743            (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), FPR16:$Rn, hsub),
5744            (i64 0)))>;
5745def : Pat<(v8bf16 (AArch64dup (bf16 FPR16:$Rn))),
5746          (v8bf16 (DUPv8i16lane
5747            (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), FPR16:$Rn, hsub),
5748            (i64 0)))>;
5749
5750def : Pat<(v4f16 (AArch64duplane16 (v8f16 V128:$Rn), VectorIndexH:$imm)),
5751          (DUPv4i16lane V128:$Rn, VectorIndexH:$imm)>;
5752def : Pat<(v8f16 (AArch64duplane16 (v8f16 V128:$Rn), VectorIndexH:$imm)),
5753          (DUPv8i16lane V128:$Rn, VectorIndexH:$imm)>;
5754
5755def : Pat<(v4bf16 (AArch64duplane16 (v8bf16 V128:$Rn), VectorIndexH:$imm)),
5756          (DUPv4i16lane V128:$Rn, VectorIndexH:$imm)>;
5757def : Pat<(v8bf16 (AArch64duplane16 (v8bf16 V128:$Rn), VectorIndexH:$imm)),
5758          (DUPv8i16lane V128:$Rn, VectorIndexH:$imm)>;
5759
5760def : Pat<(v2f32 (AArch64duplane32 (v4f32 V128:$Rn), VectorIndexS:$imm)),
5761          (DUPv2i32lane V128:$Rn, VectorIndexS:$imm)>;
5762def : Pat<(v4f32 (AArch64duplane32 (v4f32 V128:$Rn), VectorIndexS:$imm)),
5763         (DUPv4i32lane V128:$Rn, VectorIndexS:$imm)>;
5764def : Pat<(v2f64 (AArch64duplane64 (v2f64 V128:$Rn), VectorIndexD:$imm)),
5765          (DUPv2i64lane V128:$Rn, VectorIndexD:$imm)>;
5766
5767// If there's an (AArch64dup (vector_extract ...) ...), we can use a duplane
5768// instruction even if the types don't match: we just have to remap the lane
5769// carefully. N.b. this trick only applies to truncations.
5770def VecIndex_x2 : SDNodeXForm<imm, [{
5771  return CurDAG->getTargetConstant(2 * N->getZExtValue(), SDLoc(N), MVT::i64);
5772}]>;
5773def VecIndex_x4 : SDNodeXForm<imm, [{
5774  return CurDAG->getTargetConstant(4 * N->getZExtValue(), SDLoc(N), MVT::i64);
5775}]>;
5776def VecIndex_x8 : SDNodeXForm<imm, [{
5777  return CurDAG->getTargetConstant(8 * N->getZExtValue(), SDLoc(N), MVT::i64);
5778}]>;
5779
5780multiclass DUPWithTruncPats<ValueType ResVT, ValueType Src64VT,
5781                            ValueType Src128VT, ValueType ScalVT,
5782                            Instruction DUP, SDNodeXForm IdxXFORM> {
5783  def : Pat<(ResVT (AArch64dup (ScalVT (vector_extract (Src128VT V128:$Rn),
5784                                                     imm:$idx)))),
5785            (DUP V128:$Rn, (IdxXFORM imm:$idx))>;
5786
5787  def : Pat<(ResVT (AArch64dup (ScalVT (vector_extract (Src64VT V64:$Rn),
5788                                                     imm:$idx)))),
5789            (DUP (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), (IdxXFORM imm:$idx))>;
5790}
5791
5792defm : DUPWithTruncPats<v8i8,   v4i16, v8i16, i32, DUPv8i8lane,  VecIndex_x2>;
5793defm : DUPWithTruncPats<v8i8,   v2i32, v4i32, i32, DUPv8i8lane,  VecIndex_x4>;
5794defm : DUPWithTruncPats<v4i16,  v2i32, v4i32, i32, DUPv4i16lane, VecIndex_x2>;
5795
5796defm : DUPWithTruncPats<v16i8,  v4i16, v8i16, i32, DUPv16i8lane, VecIndex_x2>;
5797defm : DUPWithTruncPats<v16i8,  v2i32, v4i32, i32, DUPv16i8lane, VecIndex_x4>;
5798defm : DUPWithTruncPats<v8i16,  v2i32, v4i32, i32, DUPv8i16lane, VecIndex_x2>;
5799
5800multiclass DUPWithTrunci64Pats<ValueType ResVT, Instruction DUP,
5801                               SDNodeXForm IdxXFORM> {
5802  def : Pat<(ResVT (AArch64dup (i32 (trunc (extractelt (v2i64 V128:$Rn),
5803                                                         imm:$idx))))),
5804            (DUP V128:$Rn, (IdxXFORM imm:$idx))>;
5805
5806  def : Pat<(ResVT (AArch64dup (i32 (trunc (extractelt (v1i64 V64:$Rn),
5807                                                       imm:$idx))))),
5808            (DUP (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), (IdxXFORM imm:$idx))>;
5809}
5810
5811defm : DUPWithTrunci64Pats<v8i8,  DUPv8i8lane,   VecIndex_x8>;
5812defm : DUPWithTrunci64Pats<v4i16, DUPv4i16lane,  VecIndex_x4>;
5813defm : DUPWithTrunci64Pats<v2i32, DUPv2i32lane,  VecIndex_x2>;
5814
5815defm : DUPWithTrunci64Pats<v16i8, DUPv16i8lane, VecIndex_x8>;
5816defm : DUPWithTrunci64Pats<v8i16, DUPv8i16lane, VecIndex_x4>;
5817defm : DUPWithTrunci64Pats<v4i32, DUPv4i32lane, VecIndex_x2>;
5818
5819// SMOV and UMOV definitions, with some extra patterns for convenience
5820defm SMOV : SMov;
5821defm UMOV : UMov;
5822
5823def : Pat<(sext_inreg (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx), i8),
5824          (i32 (SMOVvi8to32 V128:$Rn, VectorIndexB:$idx))>;
5825def : Pat<(sext_inreg (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx), i8),
5826          (i64 (SMOVvi8to64 V128:$Rn, VectorIndexB:$idx))>;
5827def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
5828          (i32 (SMOVvi16to32 V128:$Rn, VectorIndexH:$idx))>;
5829def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
5830          (i64 (SMOVvi16to64 V128:$Rn, VectorIndexH:$idx))>;
5831def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
5832          (i32 (SMOVvi16to32 V128:$Rn, VectorIndexH:$idx))>;
5833def : Pat<(sext (i32 (vector_extract (v4i32 V128:$Rn), VectorIndexS:$idx))),
5834          (i64 (SMOVvi32to64 V128:$Rn, VectorIndexS:$idx))>;
5835
5836def : Pat<(sext_inreg (i64 (anyext (i32 (vector_extract (v16i8 V128:$Rn),
5837            VectorIndexB:$idx)))), i8),
5838          (i64 (SMOVvi8to64 V128:$Rn, VectorIndexB:$idx))>;
5839def : Pat<(sext_inreg (i64 (anyext (i32 (vector_extract (v8i16 V128:$Rn),
5840            VectorIndexH:$idx)))), i16),
5841          (i64 (SMOVvi16to64 V128:$Rn, VectorIndexH:$idx))>;
5842
5843// Extracting i8 or i16 elements will have the zero-extend transformed to
5844// an 'and' mask by type legalization since neither i8 nor i16 are legal types
5845// for AArch64. Match these patterns here since UMOV already zeroes out the high
5846// bits of the destination register.
5847def : Pat<(and (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx),
5848               (i32 0xff)),
5849          (i32 (UMOVvi8 V128:$Rn, VectorIndexB:$idx))>;
5850def : Pat<(and (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),
5851               (i32 0xffff)),
5852          (i32 (UMOVvi16 V128:$Rn, VectorIndexH:$idx))>;
5853
5854def : Pat<(i64 (and (i64 (anyext (i32 (vector_extract (v16i8 V128:$Rn),
5855            VectorIndexB:$idx)))), (i64 0xff))),
5856          (SUBREG_TO_REG (i64 0), (i32 (UMOVvi8 V128:$Rn, VectorIndexB:$idx)), sub_32)>;
5857def : Pat<(i64 (and (i64 (anyext (i32 (vector_extract (v8i16 V128:$Rn),
5858            VectorIndexH:$idx)))), (i64 0xffff))),
5859          (SUBREG_TO_REG (i64 0), (i32 (UMOVvi16 V128:$Rn, VectorIndexH:$idx)), sub_32)>;
5860
5861defm INS : SIMDIns;
5862
5863def : Pat<(v16i8 (scalar_to_vector GPR32:$Rn)),
5864          (SUBREG_TO_REG (i32 0),
5865                         (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
5866def : Pat<(v8i8 (scalar_to_vector GPR32:$Rn)),
5867          (SUBREG_TO_REG (i32 0),
5868                         (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
5869
5870// The top bits will be zero from the FMOVWSr
5871def : Pat<(v8i8 (bitconvert (i64 (zext GPR32:$Rn)))),
5872          (SUBREG_TO_REG (i32 0), (f32 (FMOVWSr GPR32:$Rn)), ssub)>;
5873
5874def : Pat<(v8i16 (scalar_to_vector GPR32:$Rn)),
5875          (SUBREG_TO_REG (i32 0),
5876                         (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
5877def : Pat<(v4i16 (scalar_to_vector GPR32:$Rn)),
5878          (SUBREG_TO_REG (i32 0),
5879                         (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
5880
5881def : Pat<(v4f16 (scalar_to_vector (f16 FPR16:$Rn))),
5882          (INSERT_SUBREG (v4f16 (IMPLICIT_DEF)), FPR16:$Rn, hsub)>;
5883def : Pat<(v8f16 (scalar_to_vector (f16 FPR16:$Rn))),
5884          (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), FPR16:$Rn, hsub)>;
5885
5886def : Pat<(v4bf16 (scalar_to_vector (bf16 FPR16:$Rn))),
5887          (INSERT_SUBREG (v4bf16 (IMPLICIT_DEF)), FPR16:$Rn, hsub)>;
5888def : Pat<(v8bf16 (scalar_to_vector (bf16 FPR16:$Rn))),
5889          (INSERT_SUBREG (v8bf16 (IMPLICIT_DEF)), FPR16:$Rn, hsub)>;
5890
5891def : Pat<(v2i32 (scalar_to_vector (i32 FPR32:$Rn))),
5892            (v2i32 (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)),
5893                                  (i32 FPR32:$Rn), ssub))>;
5894def : Pat<(v4i32 (scalar_to_vector (i32 FPR32:$Rn))),
5895            (v4i32 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
5896                                  (i32 FPR32:$Rn), ssub))>;
5897
5898def : Pat<(v2i64 (scalar_to_vector (i64 FPR64:$Rn))),
5899            (v2i64 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
5900                                  (i64 FPR64:$Rn), dsub))>;
5901
5902def : Pat<(v4f16 (scalar_to_vector (f16 FPR16:$Rn))),
5903          (INSERT_SUBREG (v4f16 (IMPLICIT_DEF)), FPR16:$Rn, hsub)>;
5904def : Pat<(v8f16 (scalar_to_vector (f16 FPR16:$Rn))),
5905          (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), FPR16:$Rn, hsub)>;
5906
5907def : Pat<(v4bf16 (scalar_to_vector (bf16 FPR16:$Rn))),
5908          (INSERT_SUBREG (v4bf16 (IMPLICIT_DEF)), FPR16:$Rn, hsub)>;
5909def : Pat<(v8bf16 (scalar_to_vector (bf16 FPR16:$Rn))),
5910          (INSERT_SUBREG (v8bf16 (IMPLICIT_DEF)), FPR16:$Rn, hsub)>;
5911
5912def : Pat<(v4f32 (scalar_to_vector (f32 FPR32:$Rn))),
5913          (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rn, ssub)>;
5914def : Pat<(v2f32 (scalar_to_vector (f32 FPR32:$Rn))),
5915          (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), FPR32:$Rn, ssub)>;
5916
5917def : Pat<(v2f64 (scalar_to_vector (f64 FPR64:$Rn))),
5918          (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$Rn, dsub)>;
5919
5920def : Pat<(v4f16 (vector_insert (v4f16 V64:$Rn),
5921            (f16 FPR16:$Rm), (i64 VectorIndexS:$imm))),
5922          (EXTRACT_SUBREG
5923            (INSvi16lane
5924              (v8f16 (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), V64:$Rn, dsub)),
5925              VectorIndexS:$imm,
5926              (v8f16 (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), FPR16:$Rm, hsub)),
5927              (i64 0)),
5928            dsub)>;
5929
5930def : Pat<(vector_insert (v8f16 v8f16:$Rn), (f16 fpimm0),
5931            (i64 VectorIndexH:$imm)),
5932          (INSvi16gpr V128:$Rn, VectorIndexH:$imm, WZR)>;
5933def : Pat<(vector_insert v4f32:$Rn, (f32 fpimm0),
5934            (i64 VectorIndexS:$imm)),
5935          (INSvi32gpr V128:$Rn, VectorIndexS:$imm, WZR)>;
5936def : Pat<(vector_insert v2f64:$Rn, (f64 fpimm0),
5937            (i64 VectorIndexD:$imm)),
5938          (INSvi64gpr V128:$Rn, VectorIndexS:$imm, XZR)>;
5939
5940def : Pat<(v8f16 (vector_insert (v8f16 V128:$Rn),
5941            (f16 FPR16:$Rm), (i64 VectorIndexH:$imm))),
5942          (INSvi16lane
5943            V128:$Rn, VectorIndexH:$imm,
5944            (v8f16 (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), FPR16:$Rm, hsub)),
5945            (i64 0))>;
5946
5947def : Pat<(v4bf16 (vector_insert (v4bf16 V64:$Rn),
5948            (bf16 FPR16:$Rm), (i64 VectorIndexS:$imm))),
5949          (EXTRACT_SUBREG
5950            (INSvi16lane
5951              (v8bf16 (INSERT_SUBREG (v8bf16 (IMPLICIT_DEF)), V64:$Rn, dsub)),
5952              VectorIndexS:$imm,
5953              (v8bf16 (INSERT_SUBREG (v8bf16 (IMPLICIT_DEF)), FPR16:$Rm, hsub)),
5954              (i64 0)),
5955            dsub)>;
5956
5957def : Pat<(v8bf16 (vector_insert (v8bf16 V128:$Rn),
5958            (bf16 FPR16:$Rm), (i64 VectorIndexH:$imm))),
5959          (INSvi16lane
5960            V128:$Rn, VectorIndexH:$imm,
5961            (v8bf16 (INSERT_SUBREG (v8bf16 (IMPLICIT_DEF)), FPR16:$Rm, hsub)),
5962            (i64 0))>;
5963
5964def : Pat<(v2f32 (vector_insert (v2f32 V64:$Rn),
5965            (f32 FPR32:$Rm), (i64 VectorIndexS:$imm))),
5966          (EXTRACT_SUBREG
5967            (INSvi32lane
5968              (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), V64:$Rn, dsub)),
5969              VectorIndexS:$imm,
5970              (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rm, ssub)),
5971              (i64 0)),
5972            dsub)>;
5973def : Pat<(v4f32 (vector_insert (v4f32 V128:$Rn),
5974            (f32 FPR32:$Rm), (i64 VectorIndexS:$imm))),
5975          (INSvi32lane
5976            V128:$Rn, VectorIndexS:$imm,
5977            (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rm, ssub)),
5978            (i64 0))>;
5979def : Pat<(v2f64 (vector_insert (v2f64 V128:$Rn),
5980            (f64 FPR64:$Rm), (i64 VectorIndexD:$imm))),
5981          (INSvi64lane
5982            V128:$Rn, VectorIndexD:$imm,
5983            (v2f64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$Rm, dsub)),
5984            (i64 0))>;
5985
5986// Copy an element at a constant index in one vector into a constant indexed
5987// element of another.
5988// FIXME refactor to a shared class/dev parameterized on vector type, vector
5989// index type and INS extension
5990def : Pat<(v16i8 (int_aarch64_neon_vcopy_lane
5991                   (v16i8 V128:$Vd), VectorIndexB:$idx, (v16i8 V128:$Vs),
5992                   VectorIndexB:$idx2)),
5993          (v16i8 (INSvi8lane
5994                   V128:$Vd, VectorIndexB:$idx, V128:$Vs, VectorIndexB:$idx2)
5995          )>;
5996def : Pat<(v8i16 (int_aarch64_neon_vcopy_lane
5997                   (v8i16 V128:$Vd), VectorIndexH:$idx, (v8i16 V128:$Vs),
5998                   VectorIndexH:$idx2)),
5999          (v8i16 (INSvi16lane
6000                   V128:$Vd, VectorIndexH:$idx, V128:$Vs, VectorIndexH:$idx2)
6001          )>;
6002def : Pat<(v4i32 (int_aarch64_neon_vcopy_lane
6003                   (v4i32 V128:$Vd), VectorIndexS:$idx, (v4i32 V128:$Vs),
6004                   VectorIndexS:$idx2)),
6005          (v4i32 (INSvi32lane
6006                   V128:$Vd, VectorIndexS:$idx, V128:$Vs, VectorIndexS:$idx2)
6007          )>;
6008def : Pat<(v2i64 (int_aarch64_neon_vcopy_lane
6009                   (v2i64 V128:$Vd), VectorIndexD:$idx, (v2i64 V128:$Vs),
6010                   VectorIndexD:$idx2)),
6011          (v2i64 (INSvi64lane
6012                   V128:$Vd, VectorIndexD:$idx, V128:$Vs, VectorIndexD:$idx2)
6013          )>;
6014
6015multiclass Neon_INS_elt_pattern<ValueType VT128, ValueType VT64,
6016                                ValueType VTScal, Instruction INS> {
6017  def : Pat<(VT128 (vector_insert V128:$src,
6018                        (VTScal (vector_extract (VT128 V128:$Rn), imm:$Immn)),
6019                        imm:$Immd)),
6020            (INS V128:$src, imm:$Immd, V128:$Rn, imm:$Immn)>;
6021
6022  def : Pat<(VT128 (vector_insert V128:$src,
6023                        (VTScal (vector_extract (VT64 V64:$Rn), imm:$Immn)),
6024                        imm:$Immd)),
6025            (INS V128:$src, imm:$Immd,
6026                 (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), imm:$Immn)>;
6027
6028  def : Pat<(VT64 (vector_insert V64:$src,
6029                        (VTScal (vector_extract (VT128 V128:$Rn), imm:$Immn)),
6030                        imm:$Immd)),
6031            (EXTRACT_SUBREG (INS (SUBREG_TO_REG (i64 0), V64:$src, dsub),
6032                                 imm:$Immd, V128:$Rn, imm:$Immn),
6033                            dsub)>;
6034
6035  def : Pat<(VT64 (vector_insert V64:$src,
6036                        (VTScal (vector_extract (VT64 V64:$Rn), imm:$Immn)),
6037                        imm:$Immd)),
6038            (EXTRACT_SUBREG
6039                (INS (SUBREG_TO_REG (i64 0), V64:$src, dsub), imm:$Immd,
6040                     (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), imm:$Immn),
6041                dsub)>;
6042}
6043
6044defm : Neon_INS_elt_pattern<v8f16, v4f16, f16, INSvi16lane>;
6045defm : Neon_INS_elt_pattern<v8bf16, v4bf16, bf16, INSvi16lane>;
6046defm : Neon_INS_elt_pattern<v4f32, v2f32, f32, INSvi32lane>;
6047defm : Neon_INS_elt_pattern<v2f64, v1f64, f64, INSvi64lane>;
6048
6049// Insert from bitcast
6050// vector_insert(bitcast(f32 src), n, lane) -> INSvi32lane(src, lane, INSERT_SUBREG(-, n), 0)
6051def : Pat<(v4i32 (vector_insert v4i32:$src, (i32 (bitconvert (f32 FPR32:$Sn))), imm:$Immd)),
6052          (INSvi32lane V128:$src, imm:$Immd, (INSERT_SUBREG (IMPLICIT_DEF), FPR32:$Sn, ssub), 0)>;
6053def : Pat<(v2i64 (vector_insert v2i64:$src, (i64 (bitconvert (f64 FPR64:$Sn))), imm:$Immd)),
6054          (INSvi64lane V128:$src, imm:$Immd, (INSERT_SUBREG (IMPLICIT_DEF), FPR64:$Sn, dsub), 0)>;
6055
6056// bitcast of an extract
6057// f32 bitcast(vector_extract(v4i32 src, lane)) -> EXTRACT_SUBREG(INSvi32lane(-, 0, src, lane))
6058def : Pat<(f32 (bitconvert (i32 (vector_extract v4i32:$src, imm:$Immd)))),
6059          (EXTRACT_SUBREG (INSvi32lane (IMPLICIT_DEF), 0, V128:$src, imm:$Immd), ssub)>;
6060def : Pat<(f32 (bitconvert (i32 (vector_extract v4i32:$src, 0)))),
6061          (EXTRACT_SUBREG V128:$src, ssub)>;
6062def : Pat<(f64 (bitconvert (i64 (vector_extract v2i64:$src, imm:$Immd)))),
6063          (EXTRACT_SUBREG (INSvi64lane (IMPLICIT_DEF), 0, V128:$src, imm:$Immd), dsub)>;
6064def : Pat<(f64 (bitconvert (i64 (vector_extract v2i64:$src, 0)))),
6065          (EXTRACT_SUBREG V128:$src, dsub)>;
6066
6067// Floating point vector extractions are codegen'd as either a sequence of
6068// subregister extractions, or a MOV (aka DUP here) if
6069// the lane number is anything other than zero.
6070def : Pat<(vector_extract (v2f64 V128:$Rn), 0),
6071          (f64 (EXTRACT_SUBREG V128:$Rn, dsub))>;
6072def : Pat<(vector_extract (v4f32 V128:$Rn), 0),
6073          (f32 (EXTRACT_SUBREG V128:$Rn, ssub))>;
6074def : Pat<(vector_extract (v8f16 V128:$Rn), 0),
6075          (f16 (EXTRACT_SUBREG V128:$Rn, hsub))>;
6076def : Pat<(vector_extract (v8bf16 V128:$Rn), 0),
6077          (bf16 (EXTRACT_SUBREG V128:$Rn, hsub))>;
6078
6079
6080def : Pat<(vector_extract (v2f64 V128:$Rn), VectorIndexD:$idx),
6081          (f64 (DUPi64 V128:$Rn, VectorIndexD:$idx))>;
6082def : Pat<(vector_extract (v4f32 V128:$Rn), VectorIndexS:$idx),
6083          (f32 (DUPi32 V128:$Rn, VectorIndexS:$idx))>;
6084def : Pat<(vector_extract (v8f16 V128:$Rn), VectorIndexH:$idx),
6085          (f16 (DUPi16 V128:$Rn, VectorIndexH:$idx))>;
6086def : Pat<(vector_extract (v8bf16 V128:$Rn), VectorIndexH:$idx),
6087          (bf16 (DUPi16 V128:$Rn, VectorIndexH:$idx))>;
6088
6089// All concat_vectors operations are canonicalised to act on i64 vectors for
6090// AArch64. In the general case we need an instruction, which had just as well be
6091// INS.
6092class ConcatPat<ValueType DstTy, ValueType SrcTy>
6093  : Pat<(DstTy (concat_vectors (SrcTy V64:$Rd), V64:$Rn)),
6094        (INSvi64lane (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), 1,
6095                     (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rn, dsub), 0)>;
6096
6097def : ConcatPat<v2i64, v1i64>;
6098def : ConcatPat<v2f64, v1f64>;
6099def : ConcatPat<v4i32, v2i32>;
6100def : ConcatPat<v4f32, v2f32>;
6101def : ConcatPat<v8i16, v4i16>;
6102def : ConcatPat<v8f16, v4f16>;
6103def : ConcatPat<v8bf16, v4bf16>;
6104def : ConcatPat<v16i8, v8i8>;
6105
6106// If the high lanes are undef, though, we can just ignore them:
6107class ConcatUndefPat<ValueType DstTy, ValueType SrcTy>
6108  : Pat<(DstTy (concat_vectors (SrcTy V64:$Rn), undef)),
6109        (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rn, dsub)>;
6110
6111def : ConcatUndefPat<v2i64, v1i64>;
6112def : ConcatUndefPat<v2f64, v1f64>;
6113def : ConcatUndefPat<v4i32, v2i32>;
6114def : ConcatUndefPat<v4f32, v2f32>;
6115def : ConcatUndefPat<v8i16, v4i16>;
6116def : ConcatUndefPat<v16i8, v8i8>;
6117
6118//----------------------------------------------------------------------------
6119// AdvSIMD across lanes instructions
6120//----------------------------------------------------------------------------
6121
6122defm ADDV    : SIMDAcrossLanesBHS<0, 0b11011, "addv">;
6123defm SMAXV   : SIMDAcrossLanesBHS<0, 0b01010, "smaxv">;
6124defm SMINV   : SIMDAcrossLanesBHS<0, 0b11010, "sminv">;
6125defm UMAXV   : SIMDAcrossLanesBHS<1, 0b01010, "umaxv">;
6126defm UMINV   : SIMDAcrossLanesBHS<1, 0b11010, "uminv">;
6127defm SADDLV  : SIMDAcrossLanesHSD<0, 0b00011, "saddlv">;
6128defm UADDLV  : SIMDAcrossLanesHSD<1, 0b00011, "uaddlv">;
6129defm FMAXNMV : SIMDFPAcrossLanes<0b01100, 0, "fmaxnmv", int_aarch64_neon_fmaxnmv>;
6130defm FMAXV   : SIMDFPAcrossLanes<0b01111, 0, "fmaxv", int_aarch64_neon_fmaxv>;
6131defm FMINNMV : SIMDFPAcrossLanes<0b01100, 1, "fminnmv", int_aarch64_neon_fminnmv>;
6132defm FMINV   : SIMDFPAcrossLanes<0b01111, 1, "fminv", int_aarch64_neon_fminv>;
6133
6134multiclass SIMDAcrossLaneLongPairIntrinsic<string Opc, SDPatternOperator addlp> {
6135  // Patterns for addv(addlp(x)) ==> addlv
6136  def : Pat<(i32 (vector_extract (v8i16 (insert_subvector undef,
6137              (v4i16 (AArch64uaddv (v4i16 (addlp (v8i8 V64:$op))))),
6138              (i64 0))), (i64 0))),
6139            (EXTRACT_SUBREG (INSERT_SUBREG (v4i16 (IMPLICIT_DEF)),
6140              (!cast<Instruction>(Opc#"v8i8v") V64:$op), hsub), ssub)>;
6141  def : Pat<(i32 (vector_extract (v8i16 (AArch64uaddv (v8i16 (addlp (v16i8 V128:$op))))), (i64 0))),
6142            (EXTRACT_SUBREG (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
6143              (!cast<Instruction>(Opc#"v16i8v") V128:$op), hsub), ssub)>;
6144  def : Pat<(v4i32 (AArch64uaddv (v4i32 (addlp (v8i16 V128:$op))))),
6145            (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), (!cast<Instruction>(Opc#"v8i16v") V128:$op), ssub)>;
6146
6147  // Patterns for addp(addlp(x))) ==> addlv
6148  def : Pat<(v2i32 (AArch64uaddv (v2i32 (addlp (v4i16 V64:$op))))),
6149            (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)), (!cast<Instruction>(Opc#"v4i16v") V64:$op), ssub)>;
6150  def : Pat<(v2i64 (AArch64uaddv (v2i64 (addlp (v4i32 V128:$op))))),
6151            (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), (!cast<Instruction>(Opc#"v4i32v") V128:$op), dsub)>;
6152}
6153
6154defm : SIMDAcrossLaneLongPairIntrinsic<"UADDLV", AArch64uaddlp>;
6155defm : SIMDAcrossLaneLongPairIntrinsic<"SADDLV", AArch64saddlp>;
6156
6157// Patterns for across-vector intrinsics, that have a node equivalent, that
6158// returns a vector (with only the low lane defined) instead of a scalar.
6159// In effect, opNode is the same as (scalar_to_vector (IntNode)).
6160multiclass SIMDAcrossLanesIntrinsic<string baseOpc,
6161                                    SDPatternOperator opNode> {
6162// If a lane instruction caught the vector_extract around opNode, we can
6163// directly match the latter to the instruction.
6164def : Pat<(v8i8 (opNode V64:$Rn)),
6165          (INSERT_SUBREG (v8i8 (IMPLICIT_DEF)),
6166           (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub)>;
6167def : Pat<(v16i8 (opNode V128:$Rn)),
6168          (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
6169           (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub)>;
6170def : Pat<(v4i16 (opNode V64:$Rn)),
6171          (INSERT_SUBREG (v4i16 (IMPLICIT_DEF)),
6172           (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub)>;
6173def : Pat<(v8i16 (opNode V128:$Rn)),
6174          (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
6175           (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub)>;
6176def : Pat<(v4i32 (opNode V128:$Rn)),
6177          (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
6178           (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), ssub)>;
6179
6180
6181// If none did, fallback to the explicit patterns, consuming the vector_extract.
6182def : Pat<(i32 (vector_extract (insert_subvector undef, (v8i8 (opNode V64:$Rn)),
6183            (i64 0)), (i64 0))),
6184          (EXTRACT_SUBREG (INSERT_SUBREG (v8i8 (IMPLICIT_DEF)),
6185            (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn),
6186            bsub), ssub)>;
6187def : Pat<(i32 (vector_extract (v16i8 (opNode V128:$Rn)), (i64 0))),
6188          (EXTRACT_SUBREG (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
6189            (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn),
6190            bsub), ssub)>;
6191def : Pat<(i32 (vector_extract (insert_subvector undef,
6192            (v4i16 (opNode V64:$Rn)), (i64 0)), (i64 0))),
6193          (EXTRACT_SUBREG (INSERT_SUBREG (v4i16 (IMPLICIT_DEF)),
6194            (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn),
6195            hsub), ssub)>;
6196def : Pat<(i32 (vector_extract (v8i16 (opNode V128:$Rn)), (i64 0))),
6197          (EXTRACT_SUBREG (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
6198            (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn),
6199            hsub), ssub)>;
6200def : Pat<(i32 (vector_extract (v4i32 (opNode V128:$Rn)), (i64 0))),
6201          (EXTRACT_SUBREG (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
6202            (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn),
6203            ssub), ssub)>;
6204
6205}
6206
6207multiclass SIMDAcrossLanesSignedIntrinsic<string baseOpc,
6208                                          SDPatternOperator opNode>
6209    : SIMDAcrossLanesIntrinsic<baseOpc, opNode> {
6210// If there is a sign extension after this intrinsic, consume it as smov already
6211// performed it
6212def : Pat<(i32 (sext_inreg (i32 (vector_extract (insert_subvector undef,
6213            (opNode (v8i8 V64:$Rn)), (i64 0)), (i64 0))), i8)),
6214          (i32 (SMOVvi8to32
6215            (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
6216              (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
6217            (i64 0)))>;
6218def : Pat<(i32 (sext_inreg (i32 (vector_extract
6219            (opNode (v16i8 V128:$Rn)), (i64 0))), i8)),
6220          (i32 (SMOVvi8to32
6221            (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
6222             (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
6223            (i64 0)))>;
6224def : Pat<(i32 (sext_inreg (i32 (vector_extract (insert_subvector undef,
6225            (opNode (v4i16 V64:$Rn)), (i64 0)), (i64 0))), i16)),
6226          (i32 (SMOVvi16to32
6227           (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
6228            (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
6229           (i64 0)))>;
6230def : Pat<(i32 (sext_inreg (i32 (vector_extract
6231            (opNode (v8i16 V128:$Rn)), (i64 0))), i16)),
6232          (i32 (SMOVvi16to32
6233            (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
6234             (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
6235            (i64 0)))>;
6236}
6237
6238multiclass SIMDAcrossLanesUnsignedIntrinsic<string baseOpc,
6239                                            SDPatternOperator opNode>
6240    : SIMDAcrossLanesIntrinsic<baseOpc, opNode> {
6241// If there is a masking operation keeping only what has been actually
6242// generated, consume it.
6243def : Pat<(i32 (and (i32 (vector_extract (insert_subvector undef,
6244            (opNode (v8i8 V64:$Rn)), (i64 0)), (i64 0))), maski8_or_more)),
6245      (i32 (EXTRACT_SUBREG
6246        (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
6247          (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
6248        ssub))>;
6249def : Pat<(i32 (and (i32 (vector_extract (opNode (v16i8 V128:$Rn)), (i64 0))),
6250            maski8_or_more)),
6251        (i32 (EXTRACT_SUBREG
6252          (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
6253            (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
6254          ssub))>;
6255def : Pat<(i32 (and (i32 (vector_extract (insert_subvector undef,
6256            (opNode (v4i16 V64:$Rn)), (i64 0)), (i64 0))), maski16_or_more)),
6257          (i32 (EXTRACT_SUBREG
6258            (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
6259              (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
6260            ssub))>;
6261def : Pat<(i32 (and (i32 (vector_extract (opNode (v8i16 V128:$Rn)), (i64 0))),
6262            maski16_or_more)),
6263        (i32 (EXTRACT_SUBREG
6264          (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
6265            (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
6266          ssub))>;
6267}
6268
6269defm : SIMDAcrossLanesSignedIntrinsic<"ADDV",  AArch64saddv>;
6270// vaddv_[su]32 is special; -> ADDP Vd.2S,Vn.2S,Vm.2S; return Vd.s[0];Vn==Vm
6271def : Pat<(v2i32 (AArch64saddv (v2i32 V64:$Rn))),
6272          (ADDPv2i32 V64:$Rn, V64:$Rn)>;
6273
6274defm : SIMDAcrossLanesUnsignedIntrinsic<"ADDV", AArch64uaddv>;
6275// vaddv_[su]32 is special; -> ADDP Vd.2S,Vn.2S,Vm.2S; return Vd.s[0];Vn==Vm
6276def : Pat<(v2i32 (AArch64uaddv (v2i32 V64:$Rn))),
6277          (ADDPv2i32 V64:$Rn, V64:$Rn)>;
6278
6279defm : SIMDAcrossLanesSignedIntrinsic<"SMAXV", AArch64smaxv>;
6280def : Pat<(v2i32 (AArch64smaxv (v2i32 V64:$Rn))),
6281          (SMAXPv2i32 V64:$Rn, V64:$Rn)>;
6282
6283defm : SIMDAcrossLanesSignedIntrinsic<"SMINV", AArch64sminv>;
6284def : Pat<(v2i32 (AArch64sminv (v2i32 V64:$Rn))),
6285          (SMINPv2i32 V64:$Rn, V64:$Rn)>;
6286
6287defm : SIMDAcrossLanesUnsignedIntrinsic<"UMAXV", AArch64umaxv>;
6288def : Pat<(v2i32 (AArch64umaxv (v2i32 V64:$Rn))),
6289          (UMAXPv2i32 V64:$Rn, V64:$Rn)>;
6290
6291defm : SIMDAcrossLanesUnsignedIntrinsic<"UMINV", AArch64uminv>;
6292def : Pat<(v2i32 (AArch64uminv (v2i32 V64:$Rn))),
6293          (UMINPv2i32 V64:$Rn, V64:$Rn)>;
6294
6295multiclass SIMDAcrossLanesSignedLongIntrinsic<string baseOpc, Intrinsic intOp> {
6296  def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
6297        (i32 (SMOVvi16to32
6298          (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
6299            (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), hsub),
6300          (i64 0)))>;
6301def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
6302        (i32 (SMOVvi16to32
6303          (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
6304           (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), hsub),
6305          (i64 0)))>;
6306
6307def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
6308          (i32 (EXTRACT_SUBREG
6309           (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
6310            (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), ssub),
6311           ssub))>;
6312def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
6313        (i32 (EXTRACT_SUBREG
6314          (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
6315           (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), ssub),
6316          ssub))>;
6317
6318def : Pat<(i64 (intOp (v4i32 V128:$Rn))),
6319        (i64 (EXTRACT_SUBREG
6320          (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
6321           (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), dsub),
6322          dsub))>;
6323}
6324
6325multiclass SIMDAcrossLanesUnsignedLongIntrinsic<string baseOpc,
6326                                                Intrinsic intOp> {
6327  def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
6328        (i32 (EXTRACT_SUBREG
6329          (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
6330            (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), hsub),
6331          ssub))>;
6332def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
6333        (i32 (EXTRACT_SUBREG
6334          (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
6335            (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), hsub),
6336          ssub))>;
6337
6338def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
6339          (i32 (EXTRACT_SUBREG
6340            (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
6341              (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), ssub),
6342            ssub))>;
6343def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
6344        (i32 (EXTRACT_SUBREG
6345          (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
6346            (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), ssub),
6347          ssub))>;
6348
6349def : Pat<(i64 (intOp (v4i32 V128:$Rn))),
6350        (i64 (EXTRACT_SUBREG
6351          (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
6352            (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), dsub),
6353          dsub))>;
6354}
6355
6356defm : SIMDAcrossLanesSignedLongIntrinsic<"SADDLV", int_aarch64_neon_saddlv>;
6357defm : SIMDAcrossLanesUnsignedLongIntrinsic<"UADDLV", int_aarch64_neon_uaddlv>;
6358
6359// The vaddlv_s32 intrinsic gets mapped to SADDLP.
6360def : Pat<(i64 (int_aarch64_neon_saddlv (v2i32 V64:$Rn))),
6361          (i64 (EXTRACT_SUBREG
6362            (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
6363              (SADDLPv2i32_v1i64 V64:$Rn), dsub),
6364            dsub))>;
6365// The vaddlv_u32 intrinsic gets mapped to UADDLP.
6366def : Pat<(i64 (int_aarch64_neon_uaddlv (v2i32 V64:$Rn))),
6367          (i64 (EXTRACT_SUBREG
6368            (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
6369              (UADDLPv2i32_v1i64 V64:$Rn), dsub),
6370            dsub))>;
6371
6372//------------------------------------------------------------------------------
6373// AdvSIMD modified immediate instructions
6374//------------------------------------------------------------------------------
6375
6376// AdvSIMD BIC
6377defm BIC : SIMDModifiedImmVectorShiftTied<1, 0b11, 0b01, "bic", AArch64bici>;
6378// AdvSIMD ORR
6379defm ORR : SIMDModifiedImmVectorShiftTied<0, 0b11, 0b01, "orr", AArch64orri>;
6380
6381def : InstAlias<"bic $Vd.4h, $imm", (BICv4i16 V64:$Vd,  imm0_255:$imm, 0)>;
6382def : InstAlias<"bic $Vd.8h, $imm", (BICv8i16 V128:$Vd, imm0_255:$imm, 0)>;
6383def : InstAlias<"bic $Vd.2s, $imm", (BICv2i32 V64:$Vd,  imm0_255:$imm, 0)>;
6384def : InstAlias<"bic $Vd.4s, $imm", (BICv4i32 V128:$Vd, imm0_255:$imm, 0)>;
6385
6386def : InstAlias<"bic.4h $Vd, $imm", (BICv4i16 V64:$Vd,  imm0_255:$imm, 0)>;
6387def : InstAlias<"bic.8h $Vd, $imm", (BICv8i16 V128:$Vd, imm0_255:$imm, 0)>;
6388def : InstAlias<"bic.2s $Vd, $imm", (BICv2i32 V64:$Vd,  imm0_255:$imm, 0)>;
6389def : InstAlias<"bic.4s $Vd, $imm", (BICv4i32 V128:$Vd, imm0_255:$imm, 0)>;
6390
6391def : InstAlias<"orr $Vd.4h, $imm", (ORRv4i16 V64:$Vd,  imm0_255:$imm, 0)>;
6392def : InstAlias<"orr $Vd.8h, $imm", (ORRv8i16 V128:$Vd, imm0_255:$imm, 0)>;
6393def : InstAlias<"orr $Vd.2s, $imm", (ORRv2i32 V64:$Vd,  imm0_255:$imm, 0)>;
6394def : InstAlias<"orr $Vd.4s, $imm", (ORRv4i32 V128:$Vd, imm0_255:$imm, 0)>;
6395
6396def : InstAlias<"orr.4h $Vd, $imm", (ORRv4i16 V64:$Vd,  imm0_255:$imm, 0)>;
6397def : InstAlias<"orr.8h $Vd, $imm", (ORRv8i16 V128:$Vd, imm0_255:$imm, 0)>;
6398def : InstAlias<"orr.2s $Vd, $imm", (ORRv2i32 V64:$Vd,  imm0_255:$imm, 0)>;
6399def : InstAlias<"orr.4s $Vd, $imm", (ORRv4i32 V128:$Vd, imm0_255:$imm, 0)>;
6400
6401// AdvSIMD FMOV
6402def FMOVv2f64_ns : SIMDModifiedImmVectorNoShift<1, 1, 0, 0b1111, V128, fpimm8,
6403                                              "fmov", ".2d",
6404                       [(set (v2f64 V128:$Rd), (AArch64fmov imm0_255:$imm8))]>;
6405def FMOVv2f32_ns : SIMDModifiedImmVectorNoShift<0, 0, 0, 0b1111, V64,  fpimm8,
6406                                              "fmov", ".2s",
6407                       [(set (v2f32 V64:$Rd), (AArch64fmov imm0_255:$imm8))]>;
6408def FMOVv4f32_ns : SIMDModifiedImmVectorNoShift<1, 0, 0, 0b1111, V128, fpimm8,
6409                                              "fmov", ".4s",
6410                       [(set (v4f32 V128:$Rd), (AArch64fmov imm0_255:$imm8))]>;
6411let Predicates = [HasNEON, HasFullFP16] in {
6412def FMOVv4f16_ns : SIMDModifiedImmVectorNoShift<0, 0, 1, 0b1111, V64,  fpimm8,
6413                                              "fmov", ".4h",
6414                       [(set (v4f16 V64:$Rd), (AArch64fmov imm0_255:$imm8))]>;
6415def FMOVv8f16_ns : SIMDModifiedImmVectorNoShift<1, 0, 1, 0b1111, V128, fpimm8,
6416                                              "fmov", ".8h",
6417                       [(set (v8f16 V128:$Rd), (AArch64fmov imm0_255:$imm8))]>;
6418} // Predicates = [HasNEON, HasFullFP16]
6419
6420// AdvSIMD MOVI
6421
6422// EDIT byte mask: scalar
6423let isReMaterializable = 1, isAsCheapAsAMove = 1 in
6424def MOVID      : SIMDModifiedImmScalarNoShift<0, 1, 0b1110, "movi",
6425                    [(set FPR64:$Rd, simdimmtype10:$imm8)]>;
6426// The movi_edit node has the immediate value already encoded, so we use
6427// a plain imm0_255 here.
6428def : Pat<(f64 (AArch64movi_edit imm0_255:$shift)),
6429          (MOVID imm0_255:$shift)>;
6430
6431// EDIT byte mask: 2d
6432
6433// The movi_edit node has the immediate value already encoded, so we use
6434// a plain imm0_255 in the pattern
6435let isReMaterializable = 1, isAsCheapAsAMove = 1 in
6436def MOVIv2d_ns   : SIMDModifiedImmVectorNoShift<1, 1, 0, 0b1110, V128,
6437                                                simdimmtype10,
6438                                                "movi", ".2d",
6439                   [(set (v2i64 V128:$Rd), (AArch64movi_edit imm0_255:$imm8))]>;
6440
6441def : Pat<(v2i64 immAllZerosV), (MOVIv2d_ns (i32 0))>;
6442def : Pat<(v4i32 immAllZerosV), (MOVIv2d_ns (i32 0))>;
6443def : Pat<(v8i16 immAllZerosV), (MOVIv2d_ns (i32 0))>;
6444def : Pat<(v16i8 immAllZerosV), (MOVIv2d_ns (i32 0))>;
6445
6446def : Pat<(v2i64 immAllOnesV), (MOVIv2d_ns (i32 255))>;
6447def : Pat<(v4i32 immAllOnesV), (MOVIv2d_ns (i32 255))>;
6448def : Pat<(v8i16 immAllOnesV), (MOVIv2d_ns (i32 255))>;
6449def : Pat<(v16i8 immAllOnesV), (MOVIv2d_ns (i32 255))>;
6450
6451// Set 64-bit vectors to all 0/1 by extracting from a 128-bit register as the
6452// extract is free and this gives better MachineCSE results.
6453def : Pat<(v1i64 immAllZerosV), (EXTRACT_SUBREG (MOVIv2d_ns (i32 0)), dsub)>;
6454def : Pat<(v2i32 immAllZerosV), (EXTRACT_SUBREG (MOVIv2d_ns (i32 0)), dsub)>;
6455def : Pat<(v4i16 immAllZerosV), (EXTRACT_SUBREG (MOVIv2d_ns (i32 0)), dsub)>;
6456def : Pat<(v8i8  immAllZerosV), (EXTRACT_SUBREG (MOVIv2d_ns (i32 0)), dsub)>;
6457
6458def : Pat<(v1i64 immAllOnesV), (EXTRACT_SUBREG (MOVIv2d_ns (i32 255)), dsub)>;
6459def : Pat<(v2i32 immAllOnesV), (EXTRACT_SUBREG (MOVIv2d_ns (i32 255)), dsub)>;
6460def : Pat<(v4i16 immAllOnesV), (EXTRACT_SUBREG (MOVIv2d_ns (i32 255)), dsub)>;
6461def : Pat<(v8i8  immAllOnesV), (EXTRACT_SUBREG (MOVIv2d_ns (i32 255)), dsub)>;
6462
6463// EDIT per word & halfword: 2s, 4h, 4s, & 8h
6464let isReMaterializable = 1, isAsCheapAsAMove = 1 in
6465defm MOVI      : SIMDModifiedImmVectorShift<0, 0b10, 0b00, "movi">;
6466
6467let Predicates = [HasNEON] in {
6468  // Using the MOVI to materialize fp constants.
6469  def : Pat<(f32 fpimm32SIMDModImmType4:$in),
6470            (EXTRACT_SUBREG (MOVIv2i32 (fpimm32SIMDModImmType4XForm f32:$in),
6471                                       (i32 24)),
6472                            ssub)>;
6473}
6474
6475def : InstAlias<"movi $Vd.4h, $imm", (MOVIv4i16 V64:$Vd,  imm0_255:$imm, 0), 0>;
6476def : InstAlias<"movi $Vd.8h, $imm", (MOVIv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
6477def : InstAlias<"movi $Vd.2s, $imm", (MOVIv2i32 V64:$Vd,  imm0_255:$imm, 0), 0>;
6478def : InstAlias<"movi $Vd.4s, $imm", (MOVIv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
6479
6480def : InstAlias<"movi.4h $Vd, $imm", (MOVIv4i16 V64:$Vd,  imm0_255:$imm, 0), 0>;
6481def : InstAlias<"movi.8h $Vd, $imm", (MOVIv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
6482def : InstAlias<"movi.2s $Vd, $imm", (MOVIv2i32 V64:$Vd,  imm0_255:$imm, 0), 0>;
6483def : InstAlias<"movi.4s $Vd, $imm", (MOVIv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
6484
6485def : Pat<(v2i32 (AArch64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
6486          (MOVIv2i32 imm0_255:$imm8, imm:$shift)>;
6487def : Pat<(v4i32 (AArch64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
6488          (MOVIv4i32 imm0_255:$imm8, imm:$shift)>;
6489def : Pat<(v4i16 (AArch64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
6490          (MOVIv4i16 imm0_255:$imm8, imm:$shift)>;
6491def : Pat<(v8i16 (AArch64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
6492          (MOVIv8i16 imm0_255:$imm8, imm:$shift)>;
6493
6494let isReMaterializable = 1, isAsCheapAsAMove = 1 in {
6495// EDIT per word: 2s & 4s with MSL shifter
6496def MOVIv2s_msl  : SIMDModifiedImmMoveMSL<0, 0, {1,1,0,?}, V64, "movi", ".2s",
6497                      [(set (v2i32 V64:$Rd),
6498                            (AArch64movi_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
6499def MOVIv4s_msl  : SIMDModifiedImmMoveMSL<1, 0, {1,1,0,?}, V128, "movi", ".4s",
6500                      [(set (v4i32 V128:$Rd),
6501                            (AArch64movi_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
6502
6503// Per byte: 8b & 16b
6504def MOVIv8b_ns   : SIMDModifiedImmVectorNoShift<0, 0, 0, 0b1110, V64,  imm0_255,
6505                                                 "movi", ".8b",
6506                       [(set (v8i8 V64:$Rd), (AArch64movi imm0_255:$imm8))]>;
6507
6508def MOVIv16b_ns  : SIMDModifiedImmVectorNoShift<1, 0, 0, 0b1110, V128, imm0_255,
6509                                                 "movi", ".16b",
6510                       [(set (v16i8 V128:$Rd), (AArch64movi imm0_255:$imm8))]>;
6511}
6512
6513// AdvSIMD MVNI
6514
6515// EDIT per word & halfword: 2s, 4h, 4s, & 8h
6516let isReMaterializable = 1, isAsCheapAsAMove = 1 in
6517defm MVNI      : SIMDModifiedImmVectorShift<1, 0b10, 0b00, "mvni">;
6518
6519def : InstAlias<"mvni $Vd.4h, $imm", (MVNIv4i16 V64:$Vd,  imm0_255:$imm, 0), 0>;
6520def : InstAlias<"mvni $Vd.8h, $imm", (MVNIv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
6521def : InstAlias<"mvni $Vd.2s, $imm", (MVNIv2i32 V64:$Vd,  imm0_255:$imm, 0), 0>;
6522def : InstAlias<"mvni $Vd.4s, $imm", (MVNIv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
6523
6524def : InstAlias<"mvni.4h $Vd, $imm", (MVNIv4i16 V64:$Vd,  imm0_255:$imm, 0), 0>;
6525def : InstAlias<"mvni.8h $Vd, $imm", (MVNIv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
6526def : InstAlias<"mvni.2s $Vd, $imm", (MVNIv2i32 V64:$Vd,  imm0_255:$imm, 0), 0>;
6527def : InstAlias<"mvni.4s $Vd, $imm", (MVNIv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
6528
6529def : Pat<(v2i32 (AArch64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
6530          (MVNIv2i32 imm0_255:$imm8, imm:$shift)>;
6531def : Pat<(v4i32 (AArch64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
6532          (MVNIv4i32 imm0_255:$imm8, imm:$shift)>;
6533def : Pat<(v4i16 (AArch64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
6534          (MVNIv4i16 imm0_255:$imm8, imm:$shift)>;
6535def : Pat<(v8i16 (AArch64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
6536          (MVNIv8i16 imm0_255:$imm8, imm:$shift)>;
6537
6538// EDIT per word: 2s & 4s with MSL shifter
6539let isReMaterializable = 1, isAsCheapAsAMove = 1 in {
6540def MVNIv2s_msl   : SIMDModifiedImmMoveMSL<0, 1, {1,1,0,?}, V64, "mvni", ".2s",
6541                      [(set (v2i32 V64:$Rd),
6542                            (AArch64mvni_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
6543def MVNIv4s_msl   : SIMDModifiedImmMoveMSL<1, 1, {1,1,0,?}, V128, "mvni", ".4s",
6544                      [(set (v4i32 V128:$Rd),
6545                            (AArch64mvni_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
6546}
6547
6548//----------------------------------------------------------------------------
6549// AdvSIMD indexed element
6550//----------------------------------------------------------------------------
6551
6552let hasSideEffects = 0 in {
6553  defm FMLA  : SIMDFPIndexedTied<0, 0b0001, "fmla">;
6554  defm FMLS  : SIMDFPIndexedTied<0, 0b0101, "fmls">;
6555}
6556
6557// NOTE: Operands are reordered in the FMLA/FMLS PatFrags because the
6558// instruction expects the addend first, while the intrinsic expects it last.
6559
6560// On the other hand, there are quite a few valid combinatorial options due to
6561// the commutativity of multiplication and the fact that (-x) * y = x * (-y).
6562defm : SIMDFPIndexedTiedPatterns<"FMLA",
6563           TriOpFrag<(any_fma node:$RHS, node:$MHS, node:$LHS)>>;
6564defm : SIMDFPIndexedTiedPatterns<"FMLA",
6565           TriOpFrag<(any_fma node:$MHS, node:$RHS, node:$LHS)>>;
6566
6567defm : SIMDFPIndexedTiedPatterns<"FMLS",
6568           TriOpFrag<(any_fma node:$MHS, (fneg node:$RHS), node:$LHS)> >;
6569defm : SIMDFPIndexedTiedPatterns<"FMLS",
6570           TriOpFrag<(any_fma node:$RHS, (fneg node:$MHS), node:$LHS)> >;
6571defm : SIMDFPIndexedTiedPatterns<"FMLS",
6572           TriOpFrag<(any_fma (fneg node:$RHS), node:$MHS, node:$LHS)> >;
6573defm : SIMDFPIndexedTiedPatterns<"FMLS",
6574           TriOpFrag<(any_fma (fneg node:$MHS), node:$RHS, node:$LHS)> >;
6575
6576multiclass FMLSIndexedAfterNegPatterns<SDPatternOperator OpNode> {
6577  // 3 variants for the .2s version: DUPLANE from 128-bit, DUPLANE from 64-bit
6578  // and DUP scalar.
6579  def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
6580                           (AArch64duplane32 (v4f32 (fneg V128:$Rm)),
6581                                           VectorIndexS:$idx))),
6582            (FMLSv2i32_indexed V64:$Rd, V64:$Rn, V128:$Rm, VectorIndexS:$idx)>;
6583  def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
6584                           (v2f32 (AArch64duplane32
6585                                      (v4f32 (insert_subvector undef,
6586                                                 (v2f32 (fneg V64:$Rm)),
6587                                                 (i64 0))),
6588                                      VectorIndexS:$idx)))),
6589            (FMLSv2i32_indexed V64:$Rd, V64:$Rn,
6590                               (SUBREG_TO_REG (i32 0), V64:$Rm, dsub),
6591                               VectorIndexS:$idx)>;
6592  def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
6593                           (AArch64dup (f32 (fneg FPR32Op:$Rm))))),
6594            (FMLSv2i32_indexed V64:$Rd, V64:$Rn,
6595                (SUBREG_TO_REG (i32 0), FPR32Op:$Rm, ssub), (i64 0))>;
6596
6597  // 3 variants for the .4s version: DUPLANE from 128-bit, DUPLANE from 64-bit
6598  // and DUP scalar.
6599  def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
6600                           (AArch64duplane32 (v4f32 (fneg V128:$Rm)),
6601                                           VectorIndexS:$idx))),
6602            (FMLSv4i32_indexed V128:$Rd, V128:$Rn, V128:$Rm,
6603                               VectorIndexS:$idx)>;
6604  def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
6605                           (v4f32 (AArch64duplane32
6606                                      (v4f32 (insert_subvector undef,
6607                                                 (v2f32 (fneg V64:$Rm)),
6608                                                 (i64 0))),
6609                                      VectorIndexS:$idx)))),
6610            (FMLSv4i32_indexed V128:$Rd, V128:$Rn,
6611                               (SUBREG_TO_REG (i32 0), V64:$Rm, dsub),
6612                               VectorIndexS:$idx)>;
6613  def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
6614                           (AArch64dup (f32 (fneg FPR32Op:$Rm))))),
6615            (FMLSv4i32_indexed V128:$Rd, V128:$Rn,
6616                (SUBREG_TO_REG (i32 0), FPR32Op:$Rm, ssub), (i64 0))>;
6617
6618  // 2 variants for the .2d version: DUPLANE from 128-bit, and DUP scalar
6619  // (DUPLANE from 64-bit would be trivial).
6620  def : Pat<(v2f64 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn),
6621                           (AArch64duplane64 (v2f64 (fneg V128:$Rm)),
6622                                           VectorIndexD:$idx))),
6623            (FMLSv2i64_indexed
6624                V128:$Rd, V128:$Rn, V128:$Rm, VectorIndexS:$idx)>;
6625  def : Pat<(v2f64 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn),
6626                           (AArch64dup (f64 (fneg FPR64Op:$Rm))))),
6627            (FMLSv2i64_indexed V128:$Rd, V128:$Rn,
6628                (SUBREG_TO_REG (i32 0), FPR64Op:$Rm, dsub), (i64 0))>;
6629
6630  // 2 variants for 32-bit scalar version: extract from .2s or from .4s
6631  def : Pat<(f32 (OpNode (f32 FPR32:$Rd), (f32 FPR32:$Rn),
6632                         (vector_extract (v4f32 (fneg V128:$Rm)),
6633                                         VectorIndexS:$idx))),
6634            (FMLSv1i32_indexed FPR32:$Rd, FPR32:$Rn,
6635                V128:$Rm, VectorIndexS:$idx)>;
6636  def : Pat<(f32 (OpNode (f32 FPR32:$Rd), (f32 FPR32:$Rn),
6637                         (vector_extract (v4f32 (insert_subvector undef,
6638                                                    (v2f32 (fneg V64:$Rm)),
6639                                                    (i64 0))),
6640                                         VectorIndexS:$idx))),
6641            (FMLSv1i32_indexed FPR32:$Rd, FPR32:$Rn,
6642                (SUBREG_TO_REG (i32 0), V64:$Rm, dsub), VectorIndexS:$idx)>;
6643
6644  // 1 variant for 64-bit scalar version: extract from .1d or from .2d
6645  def : Pat<(f64 (OpNode (f64 FPR64:$Rd), (f64 FPR64:$Rn),
6646                         (vector_extract (v2f64 (fneg V128:$Rm)),
6647                                         VectorIndexS:$idx))),
6648            (FMLSv1i64_indexed FPR64:$Rd, FPR64:$Rn,
6649                V128:$Rm, VectorIndexS:$idx)>;
6650}
6651
6652defm : FMLSIndexedAfterNegPatterns<
6653           TriOpFrag<(any_fma node:$RHS, node:$MHS, node:$LHS)> >;
6654defm : FMLSIndexedAfterNegPatterns<
6655           TriOpFrag<(any_fma node:$MHS, node:$RHS, node:$LHS)> >;
6656
6657defm FMULX : SIMDFPIndexed<1, 0b1001, "fmulx", int_aarch64_neon_fmulx>;
6658defm FMUL  : SIMDFPIndexed<0, 0b1001, "fmul", any_fmul>;
6659
6660def : Pat<(v2f32 (any_fmul V64:$Rn, (AArch64dup (f32 FPR32:$Rm)))),
6661          (FMULv2i32_indexed V64:$Rn,
6662            (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rm, ssub),
6663            (i64 0))>;
6664def : Pat<(v4f32 (any_fmul V128:$Rn, (AArch64dup (f32 FPR32:$Rm)))),
6665          (FMULv4i32_indexed V128:$Rn,
6666            (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rm, ssub),
6667            (i64 0))>;
6668def : Pat<(v2f64 (any_fmul V128:$Rn, (AArch64dup (f64 FPR64:$Rm)))),
6669          (FMULv2i64_indexed V128:$Rn,
6670            (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$Rm, dsub),
6671            (i64 0))>;
6672
6673defm SQDMULH : SIMDIndexedHS<0, 0b1100, "sqdmulh", int_aarch64_neon_sqdmulh>;
6674defm SQRDMULH : SIMDIndexedHS<0, 0b1101, "sqrdmulh", int_aarch64_neon_sqrdmulh>;
6675
6676defm SQDMULH : SIMDIndexedHSPatterns<int_aarch64_neon_sqdmulh_lane,
6677                                     int_aarch64_neon_sqdmulh_laneq>;
6678defm SQRDMULH : SIMDIndexedHSPatterns<int_aarch64_neon_sqrdmulh_lane,
6679                                      int_aarch64_neon_sqrdmulh_laneq>;
6680
6681// Generated by MachineCombine
6682defm MLA   : SIMDVectorIndexedHSTied<1, 0b0000, "mla", null_frag>;
6683defm MLS   : SIMDVectorIndexedHSTied<1, 0b0100, "mls", null_frag>;
6684
6685defm MUL   : SIMDVectorIndexedHS<0, 0b1000, "mul", mul>;
6686defm SMLAL : SIMDVectorIndexedLongSDTied<0, 0b0010, "smlal",
6687    TriOpFrag<(add node:$LHS, (AArch64smull node:$MHS, node:$RHS))>>;
6688defm SMLSL : SIMDVectorIndexedLongSDTied<0, 0b0110, "smlsl",
6689    TriOpFrag<(sub node:$LHS, (AArch64smull node:$MHS, node:$RHS))>>;
6690defm SMULL : SIMDVectorIndexedLongSD<0, 0b1010, "smull", AArch64smull>;
6691defm SQDMLAL : SIMDIndexedLongSQDMLXSDTied<0, 0b0011, "sqdmlal",
6692                                           int_aarch64_neon_sqadd>;
6693defm SQDMLSL : SIMDIndexedLongSQDMLXSDTied<0, 0b0111, "sqdmlsl",
6694                                           int_aarch64_neon_sqsub>;
6695defm SQRDMLAH : SIMDIndexedSQRDMLxHSDTied<1, 0b1101, "sqrdmlah",
6696                                          int_aarch64_neon_sqrdmlah>;
6697defm SQRDMLSH : SIMDIndexedSQRDMLxHSDTied<1, 0b1111, "sqrdmlsh",
6698                                          int_aarch64_neon_sqrdmlsh>;
6699defm SQDMULL : SIMDIndexedLongSD<0, 0b1011, "sqdmull", int_aarch64_neon_sqdmull>;
6700defm UMLAL   : SIMDVectorIndexedLongSDTied<1, 0b0010, "umlal",
6701    TriOpFrag<(add node:$LHS, (AArch64umull node:$MHS, node:$RHS))>>;
6702defm UMLSL   : SIMDVectorIndexedLongSDTied<1, 0b0110, "umlsl",
6703    TriOpFrag<(sub node:$LHS, (AArch64umull node:$MHS, node:$RHS))>>;
6704defm UMULL   : SIMDVectorIndexedLongSD<1, 0b1010, "umull", AArch64umull>;
6705
6706// A scalar sqdmull with the second operand being a vector lane can be
6707// handled directly with the indexed instruction encoding.
6708def : Pat<(int_aarch64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
6709                                          (vector_extract (v4i32 V128:$Vm),
6710                                                           VectorIndexS:$idx)),
6711          (SQDMULLv1i64_indexed FPR32:$Rn, V128:$Vm, VectorIndexS:$idx)>;
6712
6713//----------------------------------------------------------------------------
6714// AdvSIMD scalar shift instructions
6715//----------------------------------------------------------------------------
6716defm FCVTZS : SIMDFPScalarRShift<0, 0b11111, "fcvtzs">;
6717defm FCVTZU : SIMDFPScalarRShift<1, 0b11111, "fcvtzu">;
6718defm SCVTF  : SIMDFPScalarRShift<0, 0b11100, "scvtf">;
6719defm UCVTF  : SIMDFPScalarRShift<1, 0b11100, "ucvtf">;
6720// Codegen patterns for the above. We don't put these directly on the
6721// instructions because TableGen's type inference can't handle the truth.
6722// Having the same base pattern for fp <--> int totally freaks it out.
6723def : Pat<(int_aarch64_neon_vcvtfp2fxs FPR32:$Rn, vecshiftR32:$imm),
6724          (FCVTZSs FPR32:$Rn, vecshiftR32:$imm)>;
6725def : Pat<(int_aarch64_neon_vcvtfp2fxu FPR32:$Rn, vecshiftR32:$imm),
6726          (FCVTZUs FPR32:$Rn, vecshiftR32:$imm)>;
6727def : Pat<(i64 (int_aarch64_neon_vcvtfp2fxs (f64 FPR64:$Rn), vecshiftR64:$imm)),
6728          (FCVTZSd FPR64:$Rn, vecshiftR64:$imm)>;
6729def : Pat<(i64 (int_aarch64_neon_vcvtfp2fxu (f64 FPR64:$Rn), vecshiftR64:$imm)),
6730          (FCVTZUd FPR64:$Rn, vecshiftR64:$imm)>;
6731def : Pat<(v1i64 (int_aarch64_neon_vcvtfp2fxs (v1f64 FPR64:$Rn),
6732                                            vecshiftR64:$imm)),
6733          (FCVTZSd FPR64:$Rn, vecshiftR64:$imm)>;
6734def : Pat<(v1i64 (int_aarch64_neon_vcvtfp2fxu (v1f64 FPR64:$Rn),
6735                                            vecshiftR64:$imm)),
6736          (FCVTZUd FPR64:$Rn, vecshiftR64:$imm)>;
6737def : Pat<(int_aarch64_neon_vcvtfxu2fp FPR32:$Rn, vecshiftR32:$imm),
6738          (UCVTFs FPR32:$Rn, vecshiftR32:$imm)>;
6739def : Pat<(f64 (int_aarch64_neon_vcvtfxu2fp (i64 FPR64:$Rn), vecshiftR64:$imm)),
6740          (UCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
6741def : Pat<(v1f64 (int_aarch64_neon_vcvtfxs2fp (v1i64 FPR64:$Rn),
6742                                            vecshiftR64:$imm)),
6743          (SCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
6744def : Pat<(f64 (int_aarch64_neon_vcvtfxs2fp (i64 FPR64:$Rn), vecshiftR64:$imm)),
6745          (SCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
6746def : Pat<(v1f64 (int_aarch64_neon_vcvtfxu2fp (v1i64 FPR64:$Rn),
6747                                            vecshiftR64:$imm)),
6748          (UCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
6749def : Pat<(int_aarch64_neon_vcvtfxs2fp FPR32:$Rn, vecshiftR32:$imm),
6750          (SCVTFs FPR32:$Rn, vecshiftR32:$imm)>;
6751
6752// Patterns for FP16 Intrinsics - requires reg copy to/from as i16s not supported.
6753
6754def : Pat<(f16 (int_aarch64_neon_vcvtfxs2fp (i32 (sext_inreg FPR32:$Rn, i16)), vecshiftR16:$imm)),
6755          (SCVTFh (EXTRACT_SUBREG FPR32:$Rn, hsub), vecshiftR16:$imm)>;
6756def : Pat<(f16 (int_aarch64_neon_vcvtfxs2fp (i32 FPR32:$Rn), vecshiftR16:$imm)),
6757          (SCVTFh (EXTRACT_SUBREG FPR32:$Rn, hsub), vecshiftR16:$imm)>;
6758def : Pat<(f16 (int_aarch64_neon_vcvtfxs2fp (i64 FPR64:$Rn), vecshiftR16:$imm)),
6759          (SCVTFh (EXTRACT_SUBREG FPR64:$Rn, hsub), vecshiftR16:$imm)>;
6760def : Pat<(f16 (int_aarch64_neon_vcvtfxu2fp
6761            (and FPR32:$Rn, (i32 65535)),
6762            vecshiftR16:$imm)),
6763          (UCVTFh (EXTRACT_SUBREG FPR32:$Rn, hsub), vecshiftR16:$imm)>;
6764def : Pat<(f16 (int_aarch64_neon_vcvtfxu2fp FPR32:$Rn, vecshiftR16:$imm)),
6765          (UCVTFh (EXTRACT_SUBREG FPR32:$Rn, hsub), vecshiftR16:$imm)>;
6766def : Pat<(f16 (int_aarch64_neon_vcvtfxu2fp (i64 FPR64:$Rn), vecshiftR16:$imm)),
6767          (UCVTFh (EXTRACT_SUBREG FPR64:$Rn, hsub), vecshiftR16:$imm)>;
6768def : Pat<(i32 (int_aarch64_neon_vcvtfp2fxs (f16 FPR16:$Rn), vecshiftR32:$imm)),
6769          (i32 (INSERT_SUBREG
6770            (i32 (IMPLICIT_DEF)),
6771            (FCVTZSh FPR16:$Rn, vecshiftR32:$imm),
6772            hsub))>;
6773def : Pat<(i64 (int_aarch64_neon_vcvtfp2fxs (f16 FPR16:$Rn), vecshiftR64:$imm)),
6774          (i64 (INSERT_SUBREG
6775            (i64 (IMPLICIT_DEF)),
6776            (FCVTZSh FPR16:$Rn, vecshiftR64:$imm),
6777            hsub))>;
6778def : Pat<(i32 (int_aarch64_neon_vcvtfp2fxu (f16 FPR16:$Rn), vecshiftR32:$imm)),
6779          (i32 (INSERT_SUBREG
6780            (i32 (IMPLICIT_DEF)),
6781            (FCVTZUh FPR16:$Rn, vecshiftR32:$imm),
6782            hsub))>;
6783def : Pat<(i64 (int_aarch64_neon_vcvtfp2fxu (f16 FPR16:$Rn), vecshiftR64:$imm)),
6784          (i64 (INSERT_SUBREG
6785            (i64 (IMPLICIT_DEF)),
6786            (FCVTZUh FPR16:$Rn, vecshiftR64:$imm),
6787            hsub))>;
6788def : Pat<(i32 (int_aarch64_neon_facge (f16 FPR16:$Rn), (f16 FPR16:$Rm))),
6789          (i32 (INSERT_SUBREG
6790            (i32 (IMPLICIT_DEF)),
6791            (FACGE16 FPR16:$Rn, FPR16:$Rm),
6792            hsub))>;
6793def : Pat<(i32 (int_aarch64_neon_facgt (f16 FPR16:$Rn), (f16 FPR16:$Rm))),
6794          (i32 (INSERT_SUBREG
6795            (i32 (IMPLICIT_DEF)),
6796            (FACGT16 FPR16:$Rn, FPR16:$Rm),
6797            hsub))>;
6798
6799defm SHL      : SIMDScalarLShiftD<   0, 0b01010, "shl", AArch64vshl>;
6800defm SLI      : SIMDScalarLShiftDTied<1, 0b01010, "sli">;
6801defm SQRSHRN  : SIMDScalarRShiftBHS< 0, 0b10011, "sqrshrn",
6802                                     int_aarch64_neon_sqrshrn>;
6803defm SQRSHRUN : SIMDScalarRShiftBHS< 1, 0b10001, "sqrshrun",
6804                                     int_aarch64_neon_sqrshrun>;
6805defm SQSHLU   : SIMDScalarLShiftBHSD<1, 0b01100, "sqshlu", AArch64sqshlui>;
6806defm SQSHL    : SIMDScalarLShiftBHSD<0, 0b01110, "sqshl", AArch64sqshli>;
6807defm SQSHRN   : SIMDScalarRShiftBHS< 0, 0b10010, "sqshrn",
6808                                     int_aarch64_neon_sqshrn>;
6809defm SQSHRUN  : SIMDScalarRShiftBHS< 1, 0b10000, "sqshrun",
6810                                     int_aarch64_neon_sqshrun>;
6811defm SRI      : SIMDScalarRShiftDTied<   1, 0b01000, "sri">;
6812defm SRSHR    : SIMDScalarRShiftD<   0, 0b00100, "srshr", AArch64srshri>;
6813defm SRSRA    : SIMDScalarRShiftDTied<   0, 0b00110, "srsra",
6814    TriOpFrag<(add node:$LHS,
6815                   (AArch64srshri node:$MHS, node:$RHS))>>;
6816defm SSHR     : SIMDScalarRShiftD<   0, 0b00000, "sshr", AArch64vashr>;
6817defm SSRA     : SIMDScalarRShiftDTied<   0, 0b00010, "ssra",
6818    TriOpFrag<(add_and_or_is_add node:$LHS,
6819                   (AArch64vashr node:$MHS, node:$RHS))>>;
6820defm UQRSHRN  : SIMDScalarRShiftBHS< 1, 0b10011, "uqrshrn",
6821                                     int_aarch64_neon_uqrshrn>;
6822defm UQSHL    : SIMDScalarLShiftBHSD<1, 0b01110, "uqshl", AArch64uqshli>;
6823defm UQSHRN   : SIMDScalarRShiftBHS< 1, 0b10010, "uqshrn",
6824                                     int_aarch64_neon_uqshrn>;
6825defm URSHR    : SIMDScalarRShiftD<   1, 0b00100, "urshr", AArch64urshri>;
6826defm URSRA    : SIMDScalarRShiftDTied<   1, 0b00110, "ursra",
6827    TriOpFrag<(add node:$LHS,
6828                   (AArch64urshri node:$MHS, node:$RHS))>>;
6829defm USHR     : SIMDScalarRShiftD<   1, 0b00000, "ushr", AArch64vlshr>;
6830defm USRA     : SIMDScalarRShiftDTied<   1, 0b00010, "usra",
6831    TriOpFrag<(add_and_or_is_add node:$LHS,
6832                   (AArch64vlshr node:$MHS, node:$RHS))>>;
6833
6834//----------------------------------------------------------------------------
6835// AdvSIMD vector shift instructions
6836//----------------------------------------------------------------------------
6837defm FCVTZS:SIMDVectorRShiftSD<0, 0b11111, "fcvtzs", int_aarch64_neon_vcvtfp2fxs>;
6838defm FCVTZU:SIMDVectorRShiftSD<1, 0b11111, "fcvtzu", int_aarch64_neon_vcvtfp2fxu>;
6839defm SCVTF: SIMDVectorRShiftToFP<0, 0b11100, "scvtf",
6840                                   int_aarch64_neon_vcvtfxs2fp>;
6841defm RSHRN   : SIMDVectorRShiftNarrowBHS<0, 0b10001, "rshrn",
6842                          BinOpFrag<(trunc (AArch64roundingvlshr node:$LHS, node:$RHS))>>;
6843defm SHL     : SIMDVectorLShiftBHSD<0, 0b01010, "shl", AArch64vshl>;
6844defm SHRN    : SIMDVectorRShiftNarrowBHS<0, 0b10000, "shrn",
6845                          BinOpFrag<(trunc (AArch64vashr node:$LHS, node:$RHS))>>;
6846defm SLI     : SIMDVectorLShiftBHSDTied<1, 0b01010, "sli", AArch64vsli>;
6847def : Pat<(v1i64 (AArch64vsli (v1i64 FPR64:$Rd), (v1i64 FPR64:$Rn),
6848                                      (i32 vecshiftL64:$imm))),
6849          (SLId FPR64:$Rd, FPR64:$Rn, vecshiftL64:$imm)>;
6850defm SQRSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10011, "sqrshrn",
6851                                         int_aarch64_neon_sqrshrn>;
6852defm SQRSHRUN: SIMDVectorRShiftNarrowBHS<1, 0b10001, "sqrshrun",
6853                                         int_aarch64_neon_sqrshrun>;
6854defm SQSHLU : SIMDVectorLShiftBHSD<1, 0b01100, "sqshlu", AArch64sqshlui>;
6855defm SQSHL  : SIMDVectorLShiftBHSD<0, 0b01110, "sqshl", AArch64sqshli>;
6856defm SQSHRN  : SIMDVectorRShiftNarrowBHS<0, 0b10010, "sqshrn",
6857                                         int_aarch64_neon_sqshrn>;
6858defm SQSHRUN : SIMDVectorRShiftNarrowBHS<1, 0b10000, "sqshrun",
6859                                         int_aarch64_neon_sqshrun>;
6860defm SRI     : SIMDVectorRShiftBHSDTied<1, 0b01000, "sri", AArch64vsri>;
6861def : Pat<(v1i64 (AArch64vsri (v1i64 FPR64:$Rd), (v1i64 FPR64:$Rn),
6862                                      (i32 vecshiftR64:$imm))),
6863          (SRId FPR64:$Rd, FPR64:$Rn, vecshiftR64:$imm)>;
6864defm SRSHR   : SIMDVectorRShiftBHSD<0, 0b00100, "srshr", AArch64srshri>;
6865defm SRSRA   : SIMDVectorRShiftBHSDTied<0, 0b00110, "srsra",
6866                 TriOpFrag<(add node:$LHS,
6867                                (AArch64srshri node:$MHS, node:$RHS))> >;
6868defm SSHLL   : SIMDVectorLShiftLongBHSD<0, 0b10100, "sshll",
6869                BinOpFrag<(AArch64vshl (sext node:$LHS), node:$RHS)>>;
6870
6871defm SSHR    : SIMDVectorRShiftBHSD<0, 0b00000, "sshr", AArch64vashr>;
6872defm SSRA    : SIMDVectorRShiftBHSDTied<0, 0b00010, "ssra",
6873                TriOpFrag<(add_and_or_is_add node:$LHS, (AArch64vashr node:$MHS, node:$RHS))>>;
6874defm UCVTF   : SIMDVectorRShiftToFP<1, 0b11100, "ucvtf",
6875                        int_aarch64_neon_vcvtfxu2fp>;
6876defm UQRSHRN : SIMDVectorRShiftNarrowBHS<1, 0b10011, "uqrshrn",
6877                                         int_aarch64_neon_uqrshrn>;
6878defm UQSHL   : SIMDVectorLShiftBHSD<1, 0b01110, "uqshl", AArch64uqshli>;
6879defm UQSHRN  : SIMDVectorRShiftNarrowBHS<1, 0b10010, "uqshrn",
6880                                         int_aarch64_neon_uqshrn>;
6881defm URSHR   : SIMDVectorRShiftBHSD<1, 0b00100, "urshr", AArch64urshri>;
6882defm URSRA   : SIMDVectorRShiftBHSDTied<1, 0b00110, "ursra",
6883                TriOpFrag<(add node:$LHS,
6884                               (AArch64urshri node:$MHS, node:$RHS))> >;
6885defm USHLL   : SIMDVectorLShiftLongBHSD<1, 0b10100, "ushll",
6886                BinOpFrag<(AArch64vshl (zext node:$LHS), node:$RHS)>>;
6887defm USHR    : SIMDVectorRShiftBHSD<1, 0b00000, "ushr", AArch64vlshr>;
6888defm USRA    : SIMDVectorRShiftBHSDTied<1, 0b00010, "usra",
6889                TriOpFrag<(add_and_or_is_add node:$LHS, (AArch64vlshr node:$MHS, node:$RHS))> >;
6890
6891// RADDHN patterns for when RSHRN shifts by half the size of the vector element
6892def : Pat<(v8i8 (trunc (AArch64vlshr (add (v8i16 V128:$Vn), (AArch64movi_shift (i32 128), (i32 0))), (i32 8)))),
6893          (RADDHNv8i16_v8i8 V128:$Vn, (v8i16 (MOVIv2d_ns (i32 0))))>;
6894def : Pat<(v4i16 (trunc (AArch64vlshr (add (v4i32 V128:$Vn), (AArch64movi_shift (i32 128), (i32 8))), (i32 16)))),
6895          (RADDHNv4i32_v4i16 V128:$Vn, (v4i32 (MOVIv2d_ns (i32 0))))>;
6896let AddedComplexity = 5 in
6897def : Pat<(v2i32 (trunc (AArch64vlshr (add (v2i64 V128:$Vn), (AArch64dup (i64 2147483648))), (i32 32)))),
6898          (RADDHNv2i64_v2i32 V128:$Vn, (v2i64 (MOVIv2d_ns (i32 0))))>;
6899
6900// RADDHN2 patterns for when RSHRN shifts by half the size of the vector element
6901def : Pat<(v16i8 (concat_vectors
6902                 (v8i8 V64:$Vd),
6903                 (v8i8 (trunc (AArch64vlshr (add (v8i16 V128:$Vn), (AArch64movi_shift (i32 128), (i32 0))), (i32 8)))))),
6904          (RADDHNv8i16_v16i8
6905                 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Vd, dsub), V128:$Vn,
6906                 (v8i16 (MOVIv2d_ns (i32 0))))>;
6907def : Pat<(v8i16 (concat_vectors
6908                 (v4i16 V64:$Vd),
6909                 (v4i16 (trunc (AArch64vlshr (add (v4i32 V128:$Vn), (AArch64movi_shift (i32 128), (i32 8))), (i32 16)))))),
6910          (RADDHNv4i32_v8i16
6911                 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Vd, dsub), V128:$Vn,
6912                 (v4i32 (MOVIv2d_ns (i32 0))))>;
6913let AddedComplexity = 5 in
6914def : Pat<(v4i32 (concat_vectors
6915                 (v2i32 V64:$Vd),
6916                 (v2i32 (trunc (AArch64vlshr (add (v2i64 V128:$Vn), (AArch64dup (i64 2147483648))), (i32 32)))))),
6917          (RADDHNv2i64_v4i32
6918                 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Vd, dsub), V128:$Vn,
6919                 (v2i64 (MOVIv2d_ns (i32 0))))>;
6920
6921// SHRN patterns for when a logical right shift was used instead of arithmetic
6922// (the immediate guarantees no sign bits actually end up in the result so it
6923// doesn't matter).
6924def : Pat<(v8i8 (trunc (AArch64vlshr (v8i16 V128:$Rn), vecshiftR16Narrow:$imm))),
6925          (SHRNv8i8_shift V128:$Rn, vecshiftR16Narrow:$imm)>;
6926def : Pat<(v4i16 (trunc (AArch64vlshr (v4i32 V128:$Rn), vecshiftR32Narrow:$imm))),
6927          (SHRNv4i16_shift V128:$Rn, vecshiftR32Narrow:$imm)>;
6928def : Pat<(v2i32 (trunc (AArch64vlshr (v2i64 V128:$Rn), vecshiftR64Narrow:$imm))),
6929          (SHRNv2i32_shift V128:$Rn, vecshiftR64Narrow:$imm)>;
6930
6931def : Pat<(v16i8 (concat_vectors (v8i8 V64:$Rd),
6932                                 (trunc (AArch64vlshr (v8i16 V128:$Rn),
6933                                                    vecshiftR16Narrow:$imm)))),
6934          (SHRNv16i8_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
6935                           V128:$Rn, vecshiftR16Narrow:$imm)>;
6936def : Pat<(v8i16 (concat_vectors (v4i16 V64:$Rd),
6937                                 (trunc (AArch64vlshr (v4i32 V128:$Rn),
6938                                                    vecshiftR32Narrow:$imm)))),
6939          (SHRNv8i16_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
6940                           V128:$Rn, vecshiftR32Narrow:$imm)>;
6941def : Pat<(v4i32 (concat_vectors (v2i32 V64:$Rd),
6942                                 (trunc (AArch64vlshr (v2i64 V128:$Rn),
6943                                                    vecshiftR64Narrow:$imm)))),
6944          (SHRNv4i32_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
6945                           V128:$Rn, vecshiftR32Narrow:$imm)>;
6946
6947// Vector sign and zero extensions are implemented with SSHLL and USSHLL.
6948// Anyexts are implemented as zexts.
6949def : Pat<(v8i16 (sext   (v8i8 V64:$Rn))),  (SSHLLv8i8_shift  V64:$Rn, (i32 0))>;
6950def : Pat<(v8i16 (zext   (v8i8 V64:$Rn))),  (USHLLv8i8_shift  V64:$Rn, (i32 0))>;
6951def : Pat<(v8i16 (anyext (v8i8 V64:$Rn))),  (USHLLv8i8_shift  V64:$Rn, (i32 0))>;
6952def : Pat<(v4i32 (sext   (v4i16 V64:$Rn))), (SSHLLv4i16_shift V64:$Rn, (i32 0))>;
6953def : Pat<(v4i32 (zext   (v4i16 V64:$Rn))), (USHLLv4i16_shift V64:$Rn, (i32 0))>;
6954def : Pat<(v4i32 (anyext (v4i16 V64:$Rn))), (USHLLv4i16_shift V64:$Rn, (i32 0))>;
6955def : Pat<(v2i64 (sext   (v2i32 V64:$Rn))), (SSHLLv2i32_shift V64:$Rn, (i32 0))>;
6956def : Pat<(v2i64 (zext   (v2i32 V64:$Rn))), (USHLLv2i32_shift V64:$Rn, (i32 0))>;
6957def : Pat<(v2i64 (anyext (v2i32 V64:$Rn))), (USHLLv2i32_shift V64:$Rn, (i32 0))>;
6958// Also match an extend from the upper half of a 128 bit source register.
6959def : Pat<(v8i16 (anyext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
6960          (USHLLv16i8_shift V128:$Rn, (i32 0))>;
6961def : Pat<(v8i16 (zext   (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
6962          (USHLLv16i8_shift V128:$Rn, (i32 0))>;
6963def : Pat<(v8i16 (sext   (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
6964          (SSHLLv16i8_shift V128:$Rn, (i32 0))>;
6965def : Pat<(v4i32 (anyext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
6966          (USHLLv8i16_shift V128:$Rn, (i32 0))>;
6967def : Pat<(v4i32 (zext   (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
6968          (USHLLv8i16_shift V128:$Rn, (i32 0))>;
6969def : Pat<(v4i32 (sext   (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
6970          (SSHLLv8i16_shift V128:$Rn, (i32 0))>;
6971def : Pat<(v2i64 (anyext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
6972          (USHLLv4i32_shift V128:$Rn, (i32 0))>;
6973def : Pat<(v2i64 (zext   (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
6974          (USHLLv4i32_shift V128:$Rn, (i32 0))>;
6975def : Pat<(v2i64 (sext   (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
6976          (SSHLLv4i32_shift V128:$Rn, (i32 0))>;
6977
6978// Vector shift sxtl aliases
6979def : InstAlias<"sxtl.8h $dst, $src1",
6980                (SSHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
6981def : InstAlias<"sxtl $dst.8h, $src1.8b",
6982                (SSHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
6983def : InstAlias<"sxtl.4s $dst, $src1",
6984                (SSHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
6985def : InstAlias<"sxtl $dst.4s, $src1.4h",
6986                (SSHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
6987def : InstAlias<"sxtl.2d $dst, $src1",
6988                (SSHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
6989def : InstAlias<"sxtl $dst.2d, $src1.2s",
6990                (SSHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
6991
6992// Vector shift sxtl2 aliases
6993def : InstAlias<"sxtl2.8h $dst, $src1",
6994                (SSHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
6995def : InstAlias<"sxtl2 $dst.8h, $src1.16b",
6996                (SSHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
6997def : InstAlias<"sxtl2.4s $dst, $src1",
6998                (SSHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
6999def : InstAlias<"sxtl2 $dst.4s, $src1.8h",
7000                (SSHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
7001def : InstAlias<"sxtl2.2d $dst, $src1",
7002                (SSHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
7003def : InstAlias<"sxtl2 $dst.2d, $src1.4s",
7004                (SSHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
7005
7006// Vector shift uxtl aliases
7007def : InstAlias<"uxtl.8h $dst, $src1",
7008                (USHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
7009def : InstAlias<"uxtl $dst.8h, $src1.8b",
7010                (USHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
7011def : InstAlias<"uxtl.4s $dst, $src1",
7012                (USHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
7013def : InstAlias<"uxtl $dst.4s, $src1.4h",
7014                (USHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
7015def : InstAlias<"uxtl.2d $dst, $src1",
7016                (USHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
7017def : InstAlias<"uxtl $dst.2d, $src1.2s",
7018                (USHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
7019
7020// Vector shift uxtl2 aliases
7021def : InstAlias<"uxtl2.8h $dst, $src1",
7022                (USHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
7023def : InstAlias<"uxtl2 $dst.8h, $src1.16b",
7024                (USHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
7025def : InstAlias<"uxtl2.4s $dst, $src1",
7026                (USHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
7027def : InstAlias<"uxtl2 $dst.4s, $src1.8h",
7028                (USHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
7029def : InstAlias<"uxtl2.2d $dst, $src1",
7030                (USHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
7031def : InstAlias<"uxtl2 $dst.2d, $src1.4s",
7032                (USHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
7033
7034// If an integer is about to be converted to a floating point value,
7035// just load it on the floating point unit.
7036// These patterns are more complex because floating point loads do not
7037// support sign extension.
7038// The sign extension has to be explicitly added and is only supported for
7039// one step: byte-to-half, half-to-word, word-to-doubleword.
7040// SCVTF GPR -> FPR is 9 cycles.
7041// SCVTF FPR -> FPR is 4 cyclces.
7042// (sign extension with lengthen) SXTL FPR -> FPR is 2 cycles.
7043// Therefore, we can do 2 sign extensions and one SCVTF FPR -> FPR
7044// and still being faster.
7045// However, this is not good for code size.
7046// 8-bits -> float. 2 sizes step-up.
7047class SExtLoadi8CVTf32Pat<dag addrmode, dag INST>
7048  : Pat<(f32 (sint_to_fp (i32 (sextloadi8 addrmode)))),
7049        (SCVTFv1i32 (f32 (EXTRACT_SUBREG
7050                            (SSHLLv4i16_shift
7051                              (f64
7052                                (EXTRACT_SUBREG
7053                                  (SSHLLv8i8_shift
7054                                    (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
7055                                        INST,
7056                                        bsub),
7057                                    0),
7058                                  dsub)),
7059                               0),
7060                             ssub)))>,
7061    Requires<[NotForCodeSize, UseAlternateSExtLoadCVTF32, HasNEON]>;
7062
7063def : SExtLoadi8CVTf32Pat<(ro8.Wpat GPR64sp:$Rn, GPR32:$Rm, ro8.Wext:$ext),
7064                          (LDRBroW  GPR64sp:$Rn, GPR32:$Rm, ro8.Wext:$ext)>;
7065def : SExtLoadi8CVTf32Pat<(ro8.Xpat GPR64sp:$Rn, GPR64:$Rm, ro8.Xext:$ext),
7066                          (LDRBroX  GPR64sp:$Rn, GPR64:$Rm, ro8.Xext:$ext)>;
7067def : SExtLoadi8CVTf32Pat<(am_indexed8 GPR64sp:$Rn, uimm12s1:$offset),
7068                          (LDRBui GPR64sp:$Rn, uimm12s1:$offset)>;
7069def : SExtLoadi8CVTf32Pat<(am_unscaled8 GPR64sp:$Rn, simm9:$offset),
7070                          (LDURBi GPR64sp:$Rn, simm9:$offset)>;
7071
7072// 16-bits -> float. 1 size step-up.
7073class SExtLoadi16CVTf32Pat<dag addrmode, dag INST>
7074  : Pat<(f32 (sint_to_fp (i32 (sextloadi16 addrmode)))),
7075        (SCVTFv1i32 (f32 (EXTRACT_SUBREG
7076                            (SSHLLv4i16_shift
7077                                (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
7078                                  INST,
7079                                  hsub),
7080                                0),
7081                            ssub)))>,
7082    Requires<[NotForCodeSize, UseAlternateSExtLoadCVTF32, HasNEON]>;
7083
7084def : SExtLoadi16CVTf32Pat<(ro16.Wpat GPR64sp:$Rn, GPR32:$Rm, ro16.Wext:$ext),
7085                           (LDRHroW   GPR64sp:$Rn, GPR32:$Rm, ro16.Wext:$ext)>;
7086def : SExtLoadi16CVTf32Pat<(ro16.Xpat GPR64sp:$Rn, GPR64:$Rm, ro16.Xext:$ext),
7087                           (LDRHroX   GPR64sp:$Rn, GPR64:$Rm, ro16.Xext:$ext)>;
7088def : SExtLoadi16CVTf32Pat<(am_indexed16 GPR64sp:$Rn, uimm12s2:$offset),
7089                           (LDRHui GPR64sp:$Rn, uimm12s2:$offset)>;
7090def : SExtLoadi16CVTf32Pat<(am_unscaled16 GPR64sp:$Rn, simm9:$offset),
7091                           (LDURHi GPR64sp:$Rn, simm9:$offset)>;
7092
7093// 32-bits to 32-bits are handled in target specific dag combine:
7094// performIntToFpCombine.
7095// 64-bits integer to 32-bits floating point, not possible with
7096// SCVTF on floating point registers (both source and destination
7097// must have the same size).
7098
7099// Here are the patterns for 8, 16, 32, and 64-bits to double.
7100// 8-bits -> double. 3 size step-up: give up.
7101// 16-bits -> double. 2 size step.
7102class SExtLoadi16CVTf64Pat<dag addrmode, dag INST>
7103  : Pat <(f64 (sint_to_fp (i32 (sextloadi16 addrmode)))),
7104           (SCVTFv1i64 (f64 (EXTRACT_SUBREG
7105                              (SSHLLv2i32_shift
7106                                 (f64
7107                                  (EXTRACT_SUBREG
7108                                    (SSHLLv4i16_shift
7109                                      (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
7110                                        INST,
7111                                        hsub),
7112                                     0),
7113                                   dsub)),
7114                               0),
7115                             dsub)))>,
7116    Requires<[NotForCodeSize, UseAlternateSExtLoadCVTF32, HasNEON]>;
7117
7118def : SExtLoadi16CVTf64Pat<(ro16.Wpat GPR64sp:$Rn, GPR32:$Rm, ro16.Wext:$ext),
7119                           (LDRHroW GPR64sp:$Rn, GPR32:$Rm, ro16.Wext:$ext)>;
7120def : SExtLoadi16CVTf64Pat<(ro16.Xpat GPR64sp:$Rn, GPR64:$Rm, ro16.Xext:$ext),
7121                           (LDRHroX GPR64sp:$Rn, GPR64:$Rm, ro16.Xext:$ext)>;
7122def : SExtLoadi16CVTf64Pat<(am_indexed16 GPR64sp:$Rn, uimm12s2:$offset),
7123                           (LDRHui GPR64sp:$Rn, uimm12s2:$offset)>;
7124def : SExtLoadi16CVTf64Pat<(am_unscaled16 GPR64sp:$Rn, simm9:$offset),
7125                           (LDURHi GPR64sp:$Rn, simm9:$offset)>;
7126// 32-bits -> double. 1 size step-up.
7127class SExtLoadi32CVTf64Pat<dag addrmode, dag INST>
7128  : Pat <(f64 (sint_to_fp (i32 (load addrmode)))),
7129           (SCVTFv1i64 (f64 (EXTRACT_SUBREG
7130                              (SSHLLv2i32_shift
7131                                (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
7132                                  INST,
7133                                  ssub),
7134                               0),
7135                             dsub)))>,
7136    Requires<[NotForCodeSize, UseAlternateSExtLoadCVTF32, HasNEON]>;
7137
7138def : SExtLoadi32CVTf64Pat<(ro32.Wpat GPR64sp:$Rn, GPR32:$Rm, ro32.Wext:$ext),
7139                           (LDRSroW GPR64sp:$Rn, GPR32:$Rm, ro32.Wext:$ext)>;
7140def : SExtLoadi32CVTf64Pat<(ro32.Xpat GPR64sp:$Rn, GPR64:$Rm, ro32.Xext:$ext),
7141                           (LDRSroX GPR64sp:$Rn, GPR64:$Rm, ro32.Xext:$ext)>;
7142def : SExtLoadi32CVTf64Pat<(am_indexed32 GPR64sp:$Rn, uimm12s4:$offset),
7143                           (LDRSui GPR64sp:$Rn, uimm12s4:$offset)>;
7144def : SExtLoadi32CVTf64Pat<(am_unscaled32 GPR64sp:$Rn, simm9:$offset),
7145                           (LDURSi GPR64sp:$Rn, simm9:$offset)>;
7146
7147// 64-bits -> double are handled in target specific dag combine:
7148// performIntToFpCombine.
7149
7150
7151//----------------------------------------------------------------------------
7152// AdvSIMD Load-Store Structure
7153//----------------------------------------------------------------------------
7154defm LD1 : SIMDLd1Multiple<"ld1">;
7155defm LD2 : SIMDLd2Multiple<"ld2">;
7156defm LD3 : SIMDLd3Multiple<"ld3">;
7157defm LD4 : SIMDLd4Multiple<"ld4">;
7158
7159defm ST1 : SIMDSt1Multiple<"st1">;
7160defm ST2 : SIMDSt2Multiple<"st2">;
7161defm ST3 : SIMDSt3Multiple<"st3">;
7162defm ST4 : SIMDSt4Multiple<"st4">;
7163
7164class Ld1Pat<ValueType ty, Instruction INST>
7165  : Pat<(ty (load GPR64sp:$Rn)), (INST GPR64sp:$Rn)>;
7166
7167def : Ld1Pat<v16i8, LD1Onev16b>;
7168def : Ld1Pat<v8i16, LD1Onev8h>;
7169def : Ld1Pat<v4i32, LD1Onev4s>;
7170def : Ld1Pat<v2i64, LD1Onev2d>;
7171def : Ld1Pat<v8i8,  LD1Onev8b>;
7172def : Ld1Pat<v4i16, LD1Onev4h>;
7173def : Ld1Pat<v2i32, LD1Onev2s>;
7174def : Ld1Pat<v1i64, LD1Onev1d>;
7175
7176class St1Pat<ValueType ty, Instruction INST>
7177  : Pat<(store ty:$Vt, GPR64sp:$Rn),
7178        (INST ty:$Vt, GPR64sp:$Rn)>;
7179
7180def : St1Pat<v16i8, ST1Onev16b>;
7181def : St1Pat<v8i16, ST1Onev8h>;
7182def : St1Pat<v4i32, ST1Onev4s>;
7183def : St1Pat<v2i64, ST1Onev2d>;
7184def : St1Pat<v8i8,  ST1Onev8b>;
7185def : St1Pat<v4i16, ST1Onev4h>;
7186def : St1Pat<v2i32, ST1Onev2s>;
7187def : St1Pat<v1i64, ST1Onev1d>;
7188
7189//---
7190// Single-element
7191//---
7192
7193defm LD1R          : SIMDLdR<0, 0b110, 0, "ld1r", "One", 1, 2, 4, 8>;
7194defm LD2R          : SIMDLdR<1, 0b110, 0, "ld2r", "Two", 2, 4, 8, 16>;
7195defm LD3R          : SIMDLdR<0, 0b111, 0, "ld3r", "Three", 3, 6, 12, 24>;
7196defm LD4R          : SIMDLdR<1, 0b111, 0, "ld4r", "Four", 4, 8, 16, 32>;
7197let mayLoad = 1, hasSideEffects = 0 in {
7198defm LD1 : SIMDLdSingleBTied<0, 0b000,       "ld1", VecListOneb,   GPR64pi1>;
7199defm LD1 : SIMDLdSingleHTied<0, 0b010, 0,    "ld1", VecListOneh,   GPR64pi2>;
7200defm LD1 : SIMDLdSingleSTied<0, 0b100, 0b00, "ld1", VecListOnes,   GPR64pi4>;
7201defm LD1 : SIMDLdSingleDTied<0, 0b100, 0b01, "ld1", VecListOned,   GPR64pi8>;
7202defm LD2 : SIMDLdSingleBTied<1, 0b000,       "ld2", VecListTwob,   GPR64pi2>;
7203defm LD2 : SIMDLdSingleHTied<1, 0b010, 0,    "ld2", VecListTwoh,   GPR64pi4>;
7204defm LD2 : SIMDLdSingleSTied<1, 0b100, 0b00, "ld2", VecListTwos,   GPR64pi8>;
7205defm LD2 : SIMDLdSingleDTied<1, 0b100, 0b01, "ld2", VecListTwod,   GPR64pi16>;
7206defm LD3 : SIMDLdSingleBTied<0, 0b001,       "ld3", VecListThreeb, GPR64pi3>;
7207defm LD3 : SIMDLdSingleHTied<0, 0b011, 0,    "ld3", VecListThreeh, GPR64pi6>;
7208defm LD3 : SIMDLdSingleSTied<0, 0b101, 0b00, "ld3", VecListThrees, GPR64pi12>;
7209defm LD3 : SIMDLdSingleDTied<0, 0b101, 0b01, "ld3", VecListThreed, GPR64pi24>;
7210defm LD4 : SIMDLdSingleBTied<1, 0b001,       "ld4", VecListFourb,  GPR64pi4>;
7211defm LD4 : SIMDLdSingleHTied<1, 0b011, 0,    "ld4", VecListFourh,  GPR64pi8>;
7212defm LD4 : SIMDLdSingleSTied<1, 0b101, 0b00, "ld4", VecListFours,  GPR64pi16>;
7213defm LD4 : SIMDLdSingleDTied<1, 0b101, 0b01, "ld4", VecListFourd,  GPR64pi32>;
7214}
7215
7216def : Pat<(v8i8 (AArch64dup (i32 (extloadi8 GPR64sp:$Rn)))),
7217          (LD1Rv8b GPR64sp:$Rn)>;
7218def : Pat<(v16i8 (AArch64dup (i32 (extloadi8 GPR64sp:$Rn)))),
7219          (LD1Rv16b GPR64sp:$Rn)>;
7220def : Pat<(v4i16 (AArch64dup (i32 (extloadi16 GPR64sp:$Rn)))),
7221          (LD1Rv4h GPR64sp:$Rn)>;
7222def : Pat<(v8i16 (AArch64dup (i32 (extloadi16 GPR64sp:$Rn)))),
7223          (LD1Rv8h GPR64sp:$Rn)>;
7224def : Pat<(v2i32 (AArch64dup (i32 (load GPR64sp:$Rn)))),
7225          (LD1Rv2s GPR64sp:$Rn)>;
7226def : Pat<(v4i32 (AArch64dup (i32 (load GPR64sp:$Rn)))),
7227          (LD1Rv4s GPR64sp:$Rn)>;
7228def : Pat<(v2i64 (AArch64dup (i64 (load GPR64sp:$Rn)))),
7229          (LD1Rv2d GPR64sp:$Rn)>;
7230def : Pat<(v1i64 (AArch64dup (i64 (load GPR64sp:$Rn)))),
7231          (LD1Rv1d GPR64sp:$Rn)>;
7232// Grab the floating point version too
7233def : Pat<(v2f32 (AArch64dup (f32 (load GPR64sp:$Rn)))),
7234          (LD1Rv2s GPR64sp:$Rn)>;
7235def : Pat<(v4f32 (AArch64dup (f32 (load GPR64sp:$Rn)))),
7236          (LD1Rv4s GPR64sp:$Rn)>;
7237def : Pat<(v2f64 (AArch64dup (f64 (load GPR64sp:$Rn)))),
7238          (LD1Rv2d GPR64sp:$Rn)>;
7239def : Pat<(v1f64 (AArch64dup (f64 (load GPR64sp:$Rn)))),
7240          (LD1Rv1d GPR64sp:$Rn)>;
7241def : Pat<(v4f16 (AArch64dup (f16 (load GPR64sp:$Rn)))),
7242          (LD1Rv4h GPR64sp:$Rn)>;
7243def : Pat<(v8f16 (AArch64dup (f16 (load GPR64sp:$Rn)))),
7244          (LD1Rv8h GPR64sp:$Rn)>;
7245def : Pat<(v4bf16 (AArch64dup (bf16 (load GPR64sp:$Rn)))),
7246          (LD1Rv4h GPR64sp:$Rn)>;
7247def : Pat<(v8bf16 (AArch64dup (bf16 (load GPR64sp:$Rn)))),
7248          (LD1Rv8h GPR64sp:$Rn)>;
7249
7250class Ld1Lane128Pat<SDPatternOperator scalar_load, Operand VecIndex,
7251                    ValueType VTy, ValueType STy, Instruction LD1>
7252  : Pat<(vector_insert (VTy VecListOne128:$Rd),
7253           (STy (scalar_load GPR64sp:$Rn)), VecIndex:$idx),
7254        (LD1 VecListOne128:$Rd, VecIndex:$idx, GPR64sp:$Rn)>;
7255
7256def : Ld1Lane128Pat<extloadi8,  VectorIndexB, v16i8, i32, LD1i8>;
7257def : Ld1Lane128Pat<extloadi16, VectorIndexH, v8i16, i32, LD1i16>;
7258def : Ld1Lane128Pat<load,       VectorIndexS, v4i32, i32, LD1i32>;
7259def : Ld1Lane128Pat<load,       VectorIndexS, v4f32, f32, LD1i32>;
7260def : Ld1Lane128Pat<load,       VectorIndexD, v2i64, i64, LD1i64>;
7261def : Ld1Lane128Pat<load,       VectorIndexD, v2f64, f64, LD1i64>;
7262def : Ld1Lane128Pat<load,       VectorIndexH, v8f16, f16, LD1i16>;
7263def : Ld1Lane128Pat<load,       VectorIndexH, v8bf16, bf16, LD1i16>;
7264
7265// Generate LD1 for extload if memory type does not match the
7266// destination type, for example:
7267//
7268//   (v4i32 (insert_vector_elt (load anyext from i8) idx))
7269//
7270// In this case, the index must be adjusted to match LD1 type.
7271//
7272class Ld1Lane128IdxOpPat<SDPatternOperator scalar_load, Operand
7273                    VecIndex, ValueType VTy, ValueType STy,
7274                    Instruction LD1, SDNodeXForm IdxOp>
7275  : Pat<(vector_insert (VTy VecListOne128:$Rd),
7276                       (STy (scalar_load GPR64sp:$Rn)), VecIndex:$idx),
7277        (LD1 VecListOne128:$Rd, (IdxOp VecIndex:$idx), GPR64sp:$Rn)>;
7278
7279def VectorIndexStoH : SDNodeXForm<imm, [{
7280  return CurDAG->getTargetConstant(N->getZExtValue() * 2, SDLoc(N), MVT::i64);
7281}]>;
7282def VectorIndexStoB : SDNodeXForm<imm, [{
7283  return CurDAG->getTargetConstant(N->getZExtValue() * 4, SDLoc(N), MVT::i64);
7284}]>;
7285def VectorIndexHtoB : SDNodeXForm<imm, [{
7286  return CurDAG->getTargetConstant(N->getZExtValue() * 2, SDLoc(N), MVT::i64);
7287}]>;
7288
7289def : Ld1Lane128IdxOpPat<extloadi16, VectorIndexS, v4i32, i32, LD1i16, VectorIndexStoH>;
7290def : Ld1Lane128IdxOpPat<extloadi8, VectorIndexS, v4i32, i32, LD1i8, VectorIndexStoB>;
7291def : Ld1Lane128IdxOpPat<extloadi8, VectorIndexH, v8i16, i32, LD1i8, VectorIndexHtoB>;
7292
7293// Same as above, but the first element is populated using
7294// scalar_to_vector + insert_subvector instead of insert_vector_elt.
7295let Predicates = [NotInStreamingSVEMode] in {
7296  class Ld1Lane128FirstElm<ValueType ResultTy, ValueType VecTy,
7297                          SDPatternOperator ExtLoad, Instruction LD1>
7298    : Pat<(ResultTy (scalar_to_vector (i32 (ExtLoad GPR64sp:$Rn)))),
7299            (ResultTy (EXTRACT_SUBREG
7300              (LD1 (VecTy (IMPLICIT_DEF)), 0, GPR64sp:$Rn), dsub))>;
7301
7302  def : Ld1Lane128FirstElm<v2i32, v8i16, extloadi16, LD1i16>;
7303  def : Ld1Lane128FirstElm<v2i32, v16i8, extloadi8, LD1i8>;
7304  def : Ld1Lane128FirstElm<v4i16, v16i8, extloadi8, LD1i8>;
7305}
7306class Ld1Lane64Pat<SDPatternOperator scalar_load, Operand VecIndex,
7307                   ValueType VTy, ValueType STy, Instruction LD1>
7308  : Pat<(vector_insert (VTy VecListOne64:$Rd),
7309           (STy (scalar_load GPR64sp:$Rn)), VecIndex:$idx),
7310        (EXTRACT_SUBREG
7311            (LD1 (SUBREG_TO_REG (i32 0), VecListOne64:$Rd, dsub),
7312                          VecIndex:$idx, GPR64sp:$Rn),
7313            dsub)>;
7314
7315def : Ld1Lane64Pat<extloadi8,  VectorIndexB, v8i8,  i32, LD1i8>;
7316def : Ld1Lane64Pat<extloadi16, VectorIndexH, v4i16, i32, LD1i16>;
7317def : Ld1Lane64Pat<load,       VectorIndexS, v2i32, i32, LD1i32>;
7318def : Ld1Lane64Pat<load,       VectorIndexS, v2f32, f32, LD1i32>;
7319def : Ld1Lane64Pat<load,       VectorIndexH, v4f16, f16, LD1i16>;
7320def : Ld1Lane64Pat<load,       VectorIndexH, v4bf16, bf16, LD1i16>;
7321
7322
7323defm LD1 : SIMDLdSt1SingleAliases<"ld1">;
7324defm LD2 : SIMDLdSt2SingleAliases<"ld2">;
7325defm LD3 : SIMDLdSt3SingleAliases<"ld3">;
7326defm LD4 : SIMDLdSt4SingleAliases<"ld4">;
7327
7328// Stores
7329defm ST1 : SIMDStSingleB<0, 0b000,       "st1", VecListOneb, GPR64pi1>;
7330defm ST1 : SIMDStSingleH<0, 0b010, 0,    "st1", VecListOneh, GPR64pi2>;
7331defm ST1 : SIMDStSingleS<0, 0b100, 0b00, "st1", VecListOnes, GPR64pi4>;
7332defm ST1 : SIMDStSingleD<0, 0b100, 0b01, "st1", VecListOned, GPR64pi8>;
7333
7334let AddedComplexity = 19 in
7335class St1Lane128Pat<SDPatternOperator scalar_store, Operand VecIndex,
7336                    ValueType VTy, ValueType STy, Instruction ST1>
7337  : Pat<(scalar_store
7338             (STy (vector_extract (VTy VecListOne128:$Vt), VecIndex:$idx)),
7339             GPR64sp:$Rn),
7340        (ST1 VecListOne128:$Vt, VecIndex:$idx, GPR64sp:$Rn)>;
7341
7342def : St1Lane128Pat<truncstorei8,  VectorIndexB, v16i8, i32, ST1i8>;
7343def : St1Lane128Pat<truncstorei16, VectorIndexH, v8i16, i32, ST1i16>;
7344def : St1Lane128Pat<store,         VectorIndexS, v4i32, i32, ST1i32>;
7345def : St1Lane128Pat<store,         VectorIndexS, v4f32, f32, ST1i32>;
7346def : St1Lane128Pat<store,         VectorIndexD, v2i64, i64, ST1i64>;
7347def : St1Lane128Pat<store,         VectorIndexD, v2f64, f64, ST1i64>;
7348def : St1Lane128Pat<store,         VectorIndexH, v8f16, f16, ST1i16>;
7349def : St1Lane128Pat<store,         VectorIndexH, v8bf16, bf16, ST1i16>;
7350
7351let AddedComplexity = 19 in
7352class St1Lane64Pat<SDPatternOperator scalar_store, Operand VecIndex,
7353                   ValueType VTy, ValueType STy, Instruction ST1>
7354  : Pat<(scalar_store
7355             (STy (vector_extract (VTy VecListOne64:$Vt), VecIndex:$idx)),
7356             GPR64sp:$Rn),
7357        (ST1 (SUBREG_TO_REG (i32 0), VecListOne64:$Vt, dsub),
7358             VecIndex:$idx, GPR64sp:$Rn)>;
7359
7360def : St1Lane64Pat<truncstorei8,  VectorIndexB, v8i8, i32, ST1i8>;
7361def : St1Lane64Pat<truncstorei16, VectorIndexH, v4i16, i32, ST1i16>;
7362def : St1Lane64Pat<store,         VectorIndexS, v2i32, i32, ST1i32>;
7363def : St1Lane64Pat<store,         VectorIndexS, v2f32, f32, ST1i32>;
7364def : St1Lane64Pat<store,         VectorIndexH, v4f16, f16, ST1i16>;
7365def : St1Lane64Pat<store,         VectorIndexH, v4bf16, bf16, ST1i16>;
7366
7367multiclass St1LanePost64Pat<SDPatternOperator scalar_store, Operand VecIndex,
7368                             ValueType VTy, ValueType STy, Instruction ST1,
7369                             int offset> {
7370  def : Pat<(scalar_store
7371              (STy (vector_extract (VTy VecListOne64:$Vt), VecIndex:$idx)),
7372              GPR64sp:$Rn, offset),
7373        (ST1 (SUBREG_TO_REG (i32 0), VecListOne64:$Vt, dsub),
7374             VecIndex:$idx, GPR64sp:$Rn, XZR)>;
7375
7376  def : Pat<(scalar_store
7377              (STy (vector_extract (VTy VecListOne64:$Vt), VecIndex:$idx)),
7378              GPR64sp:$Rn, GPR64:$Rm),
7379        (ST1 (SUBREG_TO_REG (i32 0), VecListOne64:$Vt, dsub),
7380             VecIndex:$idx, GPR64sp:$Rn, $Rm)>;
7381}
7382
7383defm : St1LanePost64Pat<post_truncsti8, VectorIndexB, v8i8, i32, ST1i8_POST, 1>;
7384defm : St1LanePost64Pat<post_truncsti16, VectorIndexH, v4i16, i32, ST1i16_POST,
7385                        2>;
7386defm : St1LanePost64Pat<post_store, VectorIndexS, v2i32, i32, ST1i32_POST, 4>;
7387defm : St1LanePost64Pat<post_store, VectorIndexS, v2f32, f32, ST1i32_POST, 4>;
7388defm : St1LanePost64Pat<post_store, VectorIndexD, v1i64, i64, ST1i64_POST, 8>;
7389defm : St1LanePost64Pat<post_store, VectorIndexD, v1f64, f64, ST1i64_POST, 8>;
7390defm : St1LanePost64Pat<post_store, VectorIndexH, v4f16, f16, ST1i16_POST, 2>;
7391defm : St1LanePost64Pat<post_store, VectorIndexH, v4bf16, bf16, ST1i16_POST, 2>;
7392
7393multiclass St1LanePost128Pat<SDPatternOperator scalar_store, Operand VecIndex,
7394                             ValueType VTy, ValueType STy, Instruction ST1,
7395                             int offset> {
7396  def : Pat<(scalar_store
7397              (STy (vector_extract (VTy VecListOne128:$Vt), VecIndex:$idx)),
7398              GPR64sp:$Rn, offset),
7399        (ST1 VecListOne128:$Vt, VecIndex:$idx, GPR64sp:$Rn, XZR)>;
7400
7401  def : Pat<(scalar_store
7402              (STy (vector_extract (VTy VecListOne128:$Vt), VecIndex:$idx)),
7403              GPR64sp:$Rn, GPR64:$Rm),
7404        (ST1 VecListOne128:$Vt, VecIndex:$idx, GPR64sp:$Rn, $Rm)>;
7405}
7406
7407defm : St1LanePost128Pat<post_truncsti8, VectorIndexB, v16i8, i32, ST1i8_POST,
7408                         1>;
7409defm : St1LanePost128Pat<post_truncsti16, VectorIndexH, v8i16, i32, ST1i16_POST,
7410                         2>;
7411defm : St1LanePost128Pat<post_store, VectorIndexS, v4i32, i32, ST1i32_POST, 4>;
7412defm : St1LanePost128Pat<post_store, VectorIndexS, v4f32, f32, ST1i32_POST, 4>;
7413defm : St1LanePost128Pat<post_store, VectorIndexD, v2i64, i64, ST1i64_POST, 8>;
7414defm : St1LanePost128Pat<post_store, VectorIndexD, v2f64, f64, ST1i64_POST, 8>;
7415defm : St1LanePost128Pat<post_store, VectorIndexH, v8f16, f16, ST1i16_POST, 2>;
7416defm : St1LanePost128Pat<post_store, VectorIndexH, v8bf16, bf16, ST1i16_POST, 2>;
7417
7418let mayStore = 1, hasSideEffects = 0 in {
7419defm ST2 : SIMDStSingleB<1, 0b000,       "st2", VecListTwob,   GPR64pi2>;
7420defm ST2 : SIMDStSingleH<1, 0b010, 0,    "st2", VecListTwoh,   GPR64pi4>;
7421defm ST2 : SIMDStSingleS<1, 0b100, 0b00, "st2", VecListTwos,   GPR64pi8>;
7422defm ST2 : SIMDStSingleD<1, 0b100, 0b01, "st2", VecListTwod,   GPR64pi16>;
7423defm ST3 : SIMDStSingleB<0, 0b001,       "st3", VecListThreeb, GPR64pi3>;
7424defm ST3 : SIMDStSingleH<0, 0b011, 0,    "st3", VecListThreeh, GPR64pi6>;
7425defm ST3 : SIMDStSingleS<0, 0b101, 0b00, "st3", VecListThrees, GPR64pi12>;
7426defm ST3 : SIMDStSingleD<0, 0b101, 0b01, "st3", VecListThreed, GPR64pi24>;
7427defm ST4 : SIMDStSingleB<1, 0b001,       "st4", VecListFourb,  GPR64pi4>;
7428defm ST4 : SIMDStSingleH<1, 0b011, 0,    "st4", VecListFourh,  GPR64pi8>;
7429defm ST4 : SIMDStSingleS<1, 0b101, 0b00, "st4", VecListFours,  GPR64pi16>;
7430defm ST4 : SIMDStSingleD<1, 0b101, 0b01, "st4", VecListFourd,  GPR64pi32>;
7431}
7432
7433defm ST1 : SIMDLdSt1SingleAliases<"st1">;
7434defm ST2 : SIMDLdSt2SingleAliases<"st2">;
7435defm ST3 : SIMDLdSt3SingleAliases<"st3">;
7436defm ST4 : SIMDLdSt4SingleAliases<"st4">;
7437
7438//----------------------------------------------------------------------------
7439// Crypto extensions
7440//----------------------------------------------------------------------------
7441
7442let Predicates = [HasAES] in {
7443def AESErr   : AESTiedInst<0b0100, "aese",   int_aarch64_crypto_aese>;
7444def AESDrr   : AESTiedInst<0b0101, "aesd",   int_aarch64_crypto_aesd>;
7445def AESMCrr  : AESInst<    0b0110, "aesmc",  int_aarch64_crypto_aesmc>;
7446def AESIMCrr : AESInst<    0b0111, "aesimc", int_aarch64_crypto_aesimc>;
7447}
7448
7449// Pseudo instructions for AESMCrr/AESIMCrr with a register constraint required
7450// for AES fusion on some CPUs.
7451let hasSideEffects = 0, mayStore = 0, mayLoad = 0 in {
7452def AESMCrrTied: Pseudo<(outs V128:$Rd), (ins V128:$Rn), [], "$Rn = $Rd">,
7453                        Sched<[WriteVq]>;
7454def AESIMCrrTied: Pseudo<(outs V128:$Rd), (ins V128:$Rn), [], "$Rn = $Rd">,
7455                         Sched<[WriteVq]>;
7456}
7457
7458// Only use constrained versions of AES(I)MC instructions if they are paired with
7459// AESE/AESD.
7460def : Pat<(v16i8 (int_aarch64_crypto_aesmc
7461            (v16i8 (int_aarch64_crypto_aese (v16i8 V128:$src1),
7462                                            (v16i8 V128:$src2))))),
7463          (v16i8 (AESMCrrTied (v16i8 (AESErr (v16i8 V128:$src1),
7464                                             (v16i8 V128:$src2)))))>,
7465          Requires<[HasFuseAES]>;
7466
7467def : Pat<(v16i8 (int_aarch64_crypto_aesimc
7468            (v16i8 (int_aarch64_crypto_aesd (v16i8 V128:$src1),
7469                                            (v16i8 V128:$src2))))),
7470          (v16i8 (AESIMCrrTied (v16i8 (AESDrr (v16i8 V128:$src1),
7471                                              (v16i8 V128:$src2)))))>,
7472          Requires<[HasFuseAES]>;
7473
7474let Predicates = [HasSHA2] in {
7475def SHA1Crrr     : SHATiedInstQSV<0b000, "sha1c",   int_aarch64_crypto_sha1c>;
7476def SHA1Prrr     : SHATiedInstQSV<0b001, "sha1p",   int_aarch64_crypto_sha1p>;
7477def SHA1Mrrr     : SHATiedInstQSV<0b010, "sha1m",   int_aarch64_crypto_sha1m>;
7478def SHA1SU0rrr   : SHATiedInstVVV<0b011, "sha1su0", int_aarch64_crypto_sha1su0>;
7479def SHA256Hrrr   : SHATiedInstQQV<0b100, "sha256h", int_aarch64_crypto_sha256h>;
7480def SHA256H2rrr  : SHATiedInstQQV<0b101, "sha256h2",int_aarch64_crypto_sha256h2>;
7481def SHA256SU1rrr :SHATiedInstVVV<0b110, "sha256su1",int_aarch64_crypto_sha256su1>;
7482
7483def SHA1Hrr     : SHAInstSS<    0b0000, "sha1h",    int_aarch64_crypto_sha1h>;
7484def SHA1SU1rr   : SHATiedInstVV<0b0001, "sha1su1",  int_aarch64_crypto_sha1su1>;
7485def SHA256SU0rr : SHATiedInstVV<0b0010, "sha256su0",int_aarch64_crypto_sha256su0>;
7486}
7487
7488//----------------------------------------------------------------------------
7489// Compiler-pseudos
7490//----------------------------------------------------------------------------
7491// FIXME: Like for X86, these should go in their own separate .td file.
7492
7493// For an anyext, we don't care what the high bits are, so we can perform an
7494// INSERT_SUBREF into an IMPLICIT_DEF.
7495def : Pat<(i64 (anyext GPR32:$src)),
7496          (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32)>;
7497
7498// When we need to explicitly zero-extend, we use a 32-bit MOV instruction and
7499// then assert the extension has happened.
7500def : Pat<(i64 (zext GPR32:$src)),
7501          (SUBREG_TO_REG (i32 0), (ORRWrs WZR, GPR32:$src, 0), sub_32)>;
7502
7503// To sign extend, we use a signed bitfield move instruction (SBFM) on the
7504// containing super-reg.
7505def : Pat<(i64 (sext GPR32:$src)),
7506   (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32), 0, 31)>;
7507def : Pat<(i64 (sext_inreg GPR64:$src, i32)), (SBFMXri GPR64:$src, 0, 31)>;
7508def : Pat<(i64 (sext_inreg GPR64:$src, i16)), (SBFMXri GPR64:$src, 0, 15)>;
7509def : Pat<(i64 (sext_inreg GPR64:$src, i8)),  (SBFMXri GPR64:$src, 0, 7)>;
7510def : Pat<(i64 (sext_inreg GPR64:$src, i1)),  (SBFMXri GPR64:$src, 0, 0)>;
7511def : Pat<(i32 (sext_inreg GPR32:$src, i16)), (SBFMWri GPR32:$src, 0, 15)>;
7512def : Pat<(i32 (sext_inreg GPR32:$src, i8)),  (SBFMWri GPR32:$src, 0, 7)>;
7513def : Pat<(i32 (sext_inreg GPR32:$src, i1)),  (SBFMWri GPR32:$src, 0, 0)>;
7514
7515def : Pat<(shl (sext_inreg GPR32:$Rn, i8), (i64 imm0_31:$imm)),
7516          (SBFMWri GPR32:$Rn, (i64 (i32shift_a       imm0_31:$imm)),
7517                              (i64 (i32shift_sext_i8 imm0_31:$imm)))>;
7518def : Pat<(shl (sext_inreg GPR64:$Rn, i8), (i64 imm0_63:$imm)),
7519          (SBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
7520                              (i64 (i64shift_sext_i8 imm0_63:$imm)))>;
7521
7522def : Pat<(shl (sext_inreg GPR32:$Rn, i16), (i64 imm0_31:$imm)),
7523          (SBFMWri GPR32:$Rn, (i64 (i32shift_a        imm0_31:$imm)),
7524                              (i64 (i32shift_sext_i16 imm0_31:$imm)))>;
7525def : Pat<(shl (sext_inreg GPR64:$Rn, i16), (i64 imm0_63:$imm)),
7526          (SBFMXri GPR64:$Rn, (i64 (i64shift_a        imm0_63:$imm)),
7527                              (i64 (i64shift_sext_i16 imm0_63:$imm)))>;
7528
7529def : Pat<(shl (i64 (sext GPR32:$Rn)), (i64 imm0_63:$imm)),
7530          (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$Rn, sub_32),
7531                   (i64 (i64shift_a        imm0_63:$imm)),
7532                   (i64 (i64shift_sext_i32 imm0_63:$imm)))>;
7533
7534def : Pat<(shl (i64 (zext GPR32:$Rn)), (i64 imm0_63:$imm)),
7535          (UBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$Rn, sub_32),
7536                   (i64 (i64shift_a        imm0_63:$imm)),
7537                   (i64 (i64shift_sext_i32 imm0_63:$imm)))>;
7538
7539// sra patterns have an AddedComplexity of 10, so make sure we have a higher
7540// AddedComplexity for the following patterns since we want to match sext + sra
7541// patterns before we attempt to match a single sra node.
7542let AddedComplexity = 20 in {
7543// We support all sext + sra combinations which preserve at least one bit of the
7544// original value which is to be sign extended. E.g. we support shifts up to
7545// bitwidth-1 bits.
7546def : Pat<(sra (sext_inreg GPR32:$Rn, i8), (i64 imm0_7:$imm)),
7547          (SBFMWri GPR32:$Rn, (i64 imm0_7:$imm), 7)>;
7548def : Pat<(sra (sext_inreg GPR64:$Rn, i8), (i64 imm0_7:$imm)),
7549          (SBFMXri GPR64:$Rn, (i64 imm0_7:$imm), 7)>;
7550
7551def : Pat<(sra (sext_inreg GPR32:$Rn, i16), (i64 imm0_15:$imm)),
7552          (SBFMWri GPR32:$Rn, (i64 imm0_15:$imm), 15)>;
7553def : Pat<(sra (sext_inreg GPR64:$Rn, i16), (i64 imm0_15:$imm)),
7554          (SBFMXri GPR64:$Rn, (i64 imm0_15:$imm), 15)>;
7555
7556def : Pat<(sra (i64 (sext GPR32:$Rn)), (i64 imm0_31:$imm)),
7557          (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$Rn, sub_32),
7558                   (i64 imm0_31:$imm), 31)>;
7559} // AddedComplexity = 20
7560
7561// To truncate, we can simply extract from a subregister.
7562def : Pat<(i32 (trunc GPR64sp:$src)),
7563          (i32 (EXTRACT_SUBREG GPR64sp:$src, sub_32))>;
7564
7565// __builtin_trap() uses the BRK instruction on AArch64.
7566def : Pat<(trap), (BRK 1)>;
7567def : Pat<(debugtrap), (BRK 0xF000)>;
7568
7569def ubsan_trap_xform : SDNodeXForm<timm, [{
7570  return CurDAG->getTargetConstant(N->getZExtValue() | ('U' << 8), SDLoc(N), MVT::i32);
7571}]>;
7572
7573def ubsan_trap_imm : TImmLeaf<i32, [{
7574  return isUInt<8>(Imm);
7575}], ubsan_trap_xform>;
7576
7577def : Pat<(ubsantrap ubsan_trap_imm:$kind), (BRK ubsan_trap_imm:$kind)>;
7578
7579// Multiply high patterns which multiply the lower subvector using smull/umull
7580// and the upper subvector with smull2/umull2. Then shuffle the high the high
7581// part of both results together.
7582def : Pat<(v16i8 (mulhs V128:$Rn, V128:$Rm)),
7583          (UZP2v16i8
7584           (SMULLv8i8_v8i16 (EXTRACT_SUBREG V128:$Rn, dsub),
7585                            (EXTRACT_SUBREG V128:$Rm, dsub)),
7586           (SMULLv16i8_v8i16 V128:$Rn, V128:$Rm))>;
7587def : Pat<(v8i16 (mulhs V128:$Rn, V128:$Rm)),
7588          (UZP2v8i16
7589           (SMULLv4i16_v4i32 (EXTRACT_SUBREG V128:$Rn, dsub),
7590                             (EXTRACT_SUBREG V128:$Rm, dsub)),
7591           (SMULLv8i16_v4i32 V128:$Rn, V128:$Rm))>;
7592def : Pat<(v4i32 (mulhs V128:$Rn, V128:$Rm)),
7593          (UZP2v4i32
7594           (SMULLv2i32_v2i64 (EXTRACT_SUBREG V128:$Rn, dsub),
7595                             (EXTRACT_SUBREG V128:$Rm, dsub)),
7596           (SMULLv4i32_v2i64 V128:$Rn, V128:$Rm))>;
7597
7598def : Pat<(v16i8 (mulhu V128:$Rn, V128:$Rm)),
7599          (UZP2v16i8
7600           (UMULLv8i8_v8i16 (EXTRACT_SUBREG V128:$Rn, dsub),
7601                            (EXTRACT_SUBREG V128:$Rm, dsub)),
7602           (UMULLv16i8_v8i16 V128:$Rn, V128:$Rm))>;
7603def : Pat<(v8i16 (mulhu V128:$Rn, V128:$Rm)),
7604          (UZP2v8i16
7605           (UMULLv4i16_v4i32 (EXTRACT_SUBREG V128:$Rn, dsub),
7606                             (EXTRACT_SUBREG V128:$Rm, dsub)),
7607           (UMULLv8i16_v4i32 V128:$Rn, V128:$Rm))>;
7608def : Pat<(v4i32 (mulhu V128:$Rn, V128:$Rm)),
7609          (UZP2v4i32
7610           (UMULLv2i32_v2i64 (EXTRACT_SUBREG V128:$Rn, dsub),
7611                             (EXTRACT_SUBREG V128:$Rm, dsub)),
7612           (UMULLv4i32_v2i64 V128:$Rn, V128:$Rm))>;
7613
7614// Conversions within AdvSIMD types in the same register size are free.
7615// But because we need a consistent lane ordering, in big endian many
7616// conversions require one or more REV instructions.
7617//
7618// Consider a simple memory load followed by a bitconvert then a store.
7619//   v0 = load v2i32
7620//   v1 = BITCAST v2i32 v0 to v4i16
7621//        store v4i16 v2
7622//
7623// In big endian mode every memory access has an implicit byte swap. LDR and
7624// STR do a 64-bit byte swap, whereas LD1/ST1 do a byte swap per lane - that
7625// is, they treat the vector as a sequence of elements to be byte-swapped.
7626// The two pairs of instructions are fundamentally incompatible. We've decided
7627// to use LD1/ST1 only to simplify compiler implementation.
7628//
7629// LD1/ST1 perform the equivalent of a sequence of LDR/STR + REV. This makes
7630// the original code sequence:
7631//   v0 = load v2i32
7632//   v1 = REV v2i32                  (implicit)
7633//   v2 = BITCAST v2i32 v1 to v4i16
7634//   v3 = REV v4i16 v2               (implicit)
7635//        store v4i16 v3
7636//
7637// But this is now broken - the value stored is different to the value loaded
7638// due to lane reordering. To fix this, on every BITCAST we must perform two
7639// other REVs:
7640//   v0 = load v2i32
7641//   v1 = REV v2i32                  (implicit)
7642//   v2 = REV v2i32
7643//   v3 = BITCAST v2i32 v2 to v4i16
7644//   v4 = REV v4i16
7645//   v5 = REV v4i16 v4               (implicit)
7646//        store v4i16 v5
7647//
7648// This means an extra two instructions, but actually in most cases the two REV
7649// instructions can be combined into one. For example:
7650//   (REV64_2s (REV64_4h X)) === (REV32_4h X)
7651//
7652// There is also no 128-bit REV instruction. This must be synthesized with an
7653// EXT instruction.
7654//
7655// Most bitconverts require some sort of conversion. The only exceptions are:
7656//   a) Identity conversions -  vNfX <-> vNiX
7657//   b) Single-lane-to-scalar - v1fX <-> fX or v1iX <-> iX
7658//
7659
7660// Natural vector casts (64 bit)
7661foreach VT = [ v8i8, v4i16, v4f16, v4bf16, v2i32, v2f32, v1i64, v1f64, f64 ] in
7662  foreach VT2 = [ v8i8, v4i16, v4f16, v4bf16, v2i32, v2f32, v1i64, v1f64, f64 ] in
7663    def : Pat<(VT (AArch64NvCast (VT2 FPR64:$src))),
7664              (VT FPR64:$src)>;
7665
7666// Natural vector casts (128 bit)
7667foreach VT = [ v16i8, v8i16, v8f16, v8bf16, v4i32, v4f32, v2i64, v2f64 ] in
7668  foreach VT2 = [ v16i8, v8i16, v8f16, v8bf16, v4i32, v4f32, v2i64, v2f64 ] in
7669    def : Pat<(VT (AArch64NvCast (VT2 FPR128:$src))),
7670              (VT FPR128:$src)>;
7671
7672let Predicates = [IsLE] in {
7673def : Pat<(v8i8  (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
7674def : Pat<(v4i16 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
7675def : Pat<(v2i32 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
7676def : Pat<(v4f16 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
7677def : Pat<(v4bf16 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
7678def : Pat<(v2f32 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
7679
7680def : Pat<(i64 (bitconvert (v8i8  V64:$Vn))),
7681          (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
7682def : Pat<(i64 (bitconvert (v4i16 V64:$Vn))),
7683          (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
7684def : Pat<(i64 (bitconvert (v2i32 V64:$Vn))),
7685          (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
7686def : Pat<(i64 (bitconvert (v4f16 V64:$Vn))),
7687          (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
7688def : Pat<(i64 (bitconvert (v4bf16 V64:$Vn))),
7689          (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
7690def : Pat<(i64 (bitconvert (v2f32 V64:$Vn))),
7691          (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
7692def : Pat<(i64 (bitconvert (v1f64 V64:$Vn))),
7693          (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
7694}
7695let Predicates = [IsBE] in {
7696def : Pat<(v8i8  (bitconvert GPR64:$Xn)),
7697                 (REV64v8i8 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
7698def : Pat<(v4i16 (bitconvert GPR64:$Xn)),
7699                 (REV64v4i16 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
7700def : Pat<(v2i32 (bitconvert GPR64:$Xn)),
7701                 (REV64v2i32 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
7702def : Pat<(v4f16 (bitconvert GPR64:$Xn)),
7703                 (REV64v4i16 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
7704def : Pat<(v4bf16 (bitconvert GPR64:$Xn)),
7705                  (REV64v4i16 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
7706def : Pat<(v2f32 (bitconvert GPR64:$Xn)),
7707                 (REV64v2i32 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
7708
7709def : Pat<(i64 (bitconvert (v8i8  V64:$Vn))),
7710          (REV64v8i8 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
7711def : Pat<(i64 (bitconvert (v4i16 V64:$Vn))),
7712          (REV64v4i16 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
7713def : Pat<(i64 (bitconvert (v2i32 V64:$Vn))),
7714          (REV64v2i32 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
7715def : Pat<(i64 (bitconvert (v4f16 V64:$Vn))),
7716          (REV64v4i16 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
7717def : Pat<(i64 (bitconvert (v4bf16 V64:$Vn))),
7718          (REV64v4i16 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
7719def : Pat<(i64 (bitconvert (v2f32 V64:$Vn))),
7720          (REV64v2i32 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
7721}
7722def : Pat<(v1i64 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
7723def : Pat<(v1f64 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
7724def : Pat<(i64 (bitconvert (v1i64 V64:$Vn))),
7725          (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
7726def : Pat<(v1i64 (scalar_to_vector GPR64:$Xn)),
7727          (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
7728def : Pat<(v1f64 (scalar_to_vector GPR64:$Xn)),
7729          (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
7730def : Pat<(v1f64 (scalar_to_vector (f64 FPR64:$Xn))), (v1f64 FPR64:$Xn)>;
7731
7732def : Pat<(f32 (bitconvert (i32 GPR32:$Xn))),
7733          (COPY_TO_REGCLASS GPR32:$Xn, FPR32)>;
7734def : Pat<(i32 (bitconvert (f32 FPR32:$Xn))),
7735          (COPY_TO_REGCLASS FPR32:$Xn, GPR32)>;
7736def : Pat<(f64 (bitconvert (i64 GPR64:$Xn))),
7737          (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
7738def : Pat<(i64 (bitconvert (f64 FPR64:$Xn))),
7739          (COPY_TO_REGCLASS FPR64:$Xn, GPR64)>;
7740def : Pat<(i64 (bitconvert (v1f64 V64:$Vn))),
7741          (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
7742
7743def : Pat<(f16 (bitconvert (bf16 FPR16:$src))), (f16 FPR16:$src)>;
7744def : Pat<(bf16 (bitconvert (f16 FPR16:$src))), (bf16 FPR16:$src)>;
7745
7746let Predicates = [IsLE] in {
7747def : Pat<(v1i64 (bitconvert (v2i32 FPR64:$src))), (v1i64 FPR64:$src)>;
7748def : Pat<(v1i64 (bitconvert (v4i16 FPR64:$src))), (v1i64 FPR64:$src)>;
7749def : Pat<(v1i64 (bitconvert (v8i8  FPR64:$src))), (v1i64 FPR64:$src)>;
7750def : Pat<(v1i64 (bitconvert (v4f16 FPR64:$src))), (v1i64 FPR64:$src)>;
7751def : Pat<(v1i64 (bitconvert (v4bf16 FPR64:$src))), (v1i64 FPR64:$src)>;
7752def : Pat<(v1i64 (bitconvert (v2f32 FPR64:$src))), (v1i64 FPR64:$src)>;
7753}
7754let Predicates = [IsBE] in {
7755def : Pat<(v1i64 (bitconvert (v2i32 FPR64:$src))),
7756                             (v1i64 (REV64v2i32 FPR64:$src))>;
7757def : Pat<(v1i64 (bitconvert (v4i16 FPR64:$src))),
7758                             (v1i64 (REV64v4i16 FPR64:$src))>;
7759def : Pat<(v1i64 (bitconvert (v8i8  FPR64:$src))),
7760                             (v1i64 (REV64v8i8 FPR64:$src))>;
7761def : Pat<(v1i64 (bitconvert (v4f16 FPR64:$src))),
7762                             (v1i64 (REV64v4i16 FPR64:$src))>;
7763def : Pat<(v1i64 (bitconvert (v4bf16 FPR64:$src))),
7764                             (v1i64 (REV64v4i16 FPR64:$src))>;
7765def : Pat<(v1i64 (bitconvert (v2f32 FPR64:$src))),
7766                             (v1i64 (REV64v2i32 FPR64:$src))>;
7767}
7768def : Pat<(v1i64 (bitconvert (v1f64 FPR64:$src))), (v1i64 FPR64:$src)>;
7769def : Pat<(v1i64 (bitconvert (f64   FPR64:$src))), (v1i64 FPR64:$src)>;
7770
7771let Predicates = [IsLE] in {
7772def : Pat<(v2i32 (bitconvert (v1i64 FPR64:$src))), (v2i32 FPR64:$src)>;
7773def : Pat<(v2i32 (bitconvert (v4i16 FPR64:$src))), (v2i32 FPR64:$src)>;
7774def : Pat<(v2i32 (bitconvert (v8i8  FPR64:$src))), (v2i32 FPR64:$src)>;
7775def : Pat<(v2i32 (bitconvert (f64   FPR64:$src))), (v2i32 FPR64:$src)>;
7776def : Pat<(v2i32 (bitconvert (v1f64 FPR64:$src))), (v2i32 FPR64:$src)>;
7777def : Pat<(v2i32 (bitconvert (v4f16 FPR64:$src))), (v2i32 FPR64:$src)>;
7778def : Pat<(v2i32 (bitconvert (v4bf16 FPR64:$src))), (v2i32 FPR64:$src)>;
7779}
7780let Predicates = [IsBE] in {
7781def : Pat<(v2i32 (bitconvert (v1i64 FPR64:$src))),
7782                             (v2i32 (REV64v2i32 FPR64:$src))>;
7783def : Pat<(v2i32 (bitconvert (v4i16 FPR64:$src))),
7784                             (v2i32 (REV32v4i16 FPR64:$src))>;
7785def : Pat<(v2i32 (bitconvert (v8i8  FPR64:$src))),
7786                             (v2i32 (REV32v8i8 FPR64:$src))>;
7787def : Pat<(v2i32 (bitconvert (f64   FPR64:$src))),
7788                             (v2i32 (REV64v2i32 FPR64:$src))>;
7789def : Pat<(v2i32 (bitconvert (v1f64 FPR64:$src))),
7790                             (v2i32 (REV64v2i32 FPR64:$src))>;
7791def : Pat<(v2i32 (bitconvert (v4f16 FPR64:$src))),
7792                             (v2i32 (REV32v4i16 FPR64:$src))>;
7793def : Pat<(v2i32 (bitconvert (v4bf16 FPR64:$src))),
7794                             (v2i32 (REV32v4i16 FPR64:$src))>;
7795}
7796def : Pat<(v2i32 (bitconvert (v2f32 FPR64:$src))), (v2i32 FPR64:$src)>;
7797
7798let Predicates = [IsLE] in {
7799def : Pat<(v4i16 (bitconvert (v1i64 FPR64:$src))), (v4i16 FPR64:$src)>;
7800def : Pat<(v4i16 (bitconvert (v2i32 FPR64:$src))), (v4i16 FPR64:$src)>;
7801def : Pat<(v4i16 (bitconvert (v8i8  FPR64:$src))), (v4i16 FPR64:$src)>;
7802def : Pat<(v4i16 (bitconvert (f64   FPR64:$src))), (v4i16 FPR64:$src)>;
7803def : Pat<(v4i16 (bitconvert (v2f32 FPR64:$src))), (v4i16 FPR64:$src)>;
7804def : Pat<(v4i16 (bitconvert (v1f64 FPR64:$src))), (v4i16 FPR64:$src)>;
7805}
7806let Predicates = [IsBE] in {
7807def : Pat<(v4i16 (bitconvert (v1i64 FPR64:$src))),
7808                             (v4i16 (REV64v4i16 FPR64:$src))>;
7809def : Pat<(v4i16 (bitconvert (v2i32 FPR64:$src))),
7810                             (v4i16 (REV32v4i16 FPR64:$src))>;
7811def : Pat<(v4i16 (bitconvert (v8i8  FPR64:$src))),
7812                             (v4i16 (REV16v8i8 FPR64:$src))>;
7813def : Pat<(v4i16 (bitconvert (f64   FPR64:$src))),
7814                             (v4i16 (REV64v4i16 FPR64:$src))>;
7815def : Pat<(v4i16 (bitconvert (v2f32 FPR64:$src))),
7816                             (v4i16 (REV32v4i16 FPR64:$src))>;
7817def : Pat<(v4i16 (bitconvert (v1f64 FPR64:$src))),
7818                             (v4i16 (REV64v4i16 FPR64:$src))>;
7819}
7820def : Pat<(v4i16 (bitconvert (v4f16 FPR64:$src))), (v4i16 FPR64:$src)>;
7821def : Pat<(v4i16 (bitconvert (v4bf16 FPR64:$src))), (v4i16 FPR64:$src)>;
7822
7823let Predicates = [IsLE] in {
7824def : Pat<(v4f16 (bitconvert (v1i64 FPR64:$src))), (v4f16 FPR64:$src)>;
7825def : Pat<(v4f16 (bitconvert (v2i32 FPR64:$src))), (v4f16 FPR64:$src)>;
7826def : Pat<(v4f16 (bitconvert (v8i8  FPR64:$src))), (v4f16 FPR64:$src)>;
7827def : Pat<(v4f16 (bitconvert (f64   FPR64:$src))), (v4f16 FPR64:$src)>;
7828def : Pat<(v4f16 (bitconvert (v2f32 FPR64:$src))), (v4f16 FPR64:$src)>;
7829def : Pat<(v4f16 (bitconvert (v1f64 FPR64:$src))), (v4f16 FPR64:$src)>;
7830
7831def : Pat<(v4bf16 (bitconvert (v1i64 FPR64:$src))), (v4bf16 FPR64:$src)>;
7832def : Pat<(v4bf16 (bitconvert (v2i32 FPR64:$src))), (v4bf16 FPR64:$src)>;
7833def : Pat<(v4bf16 (bitconvert (v8i8  FPR64:$src))), (v4bf16 FPR64:$src)>;
7834def : Pat<(v4bf16 (bitconvert (f64   FPR64:$src))), (v4bf16 FPR64:$src)>;
7835def : Pat<(v4bf16 (bitconvert (v2f32 FPR64:$src))), (v4bf16 FPR64:$src)>;
7836def : Pat<(v4bf16 (bitconvert (v1f64 FPR64:$src))), (v4bf16 FPR64:$src)>;
7837}
7838let Predicates = [IsBE] in {
7839def : Pat<(v4f16 (bitconvert (v1i64 FPR64:$src))),
7840                             (v4f16 (REV64v4i16 FPR64:$src))>;
7841def : Pat<(v4f16 (bitconvert (v2i32 FPR64:$src))),
7842                             (v4f16 (REV32v4i16 FPR64:$src))>;
7843def : Pat<(v4f16 (bitconvert (v8i8  FPR64:$src))),
7844                             (v4f16 (REV16v8i8 FPR64:$src))>;
7845def : Pat<(v4f16 (bitconvert (f64   FPR64:$src))),
7846                             (v4f16 (REV64v4i16 FPR64:$src))>;
7847def : Pat<(v4f16 (bitconvert (v2f32 FPR64:$src))),
7848                             (v4f16 (REV32v4i16 FPR64:$src))>;
7849def : Pat<(v4f16 (bitconvert (v1f64 FPR64:$src))),
7850                             (v4f16 (REV64v4i16 FPR64:$src))>;
7851
7852def : Pat<(v4bf16 (bitconvert (v1i64 FPR64:$src))),
7853                             (v4bf16 (REV64v4i16 FPR64:$src))>;
7854def : Pat<(v4bf16 (bitconvert (v2i32 FPR64:$src))),
7855                             (v4bf16 (REV32v4i16 FPR64:$src))>;
7856def : Pat<(v4bf16 (bitconvert (v8i8  FPR64:$src))),
7857                             (v4bf16 (REV16v8i8 FPR64:$src))>;
7858def : Pat<(v4bf16 (bitconvert (f64   FPR64:$src))),
7859                             (v4bf16 (REV64v4i16 FPR64:$src))>;
7860def : Pat<(v4bf16 (bitconvert (v2f32 FPR64:$src))),
7861                             (v4bf16 (REV32v4i16 FPR64:$src))>;
7862def : Pat<(v4bf16 (bitconvert (v1f64 FPR64:$src))),
7863                             (v4bf16 (REV64v4i16 FPR64:$src))>;
7864}
7865def : Pat<(v4f16 (bitconvert (v4i16 FPR64:$src))), (v4f16 FPR64:$src)>;
7866def : Pat<(v4bf16 (bitconvert (v4i16 FPR64:$src))), (v4bf16 FPR64:$src)>;
7867
7868let Predicates = [IsLE] in {
7869def : Pat<(v8i8  (bitconvert (v1i64 FPR64:$src))), (v8i8  FPR64:$src)>;
7870def : Pat<(v8i8  (bitconvert (v2i32 FPR64:$src))), (v8i8  FPR64:$src)>;
7871def : Pat<(v8i8  (bitconvert (v4i16 FPR64:$src))), (v8i8  FPR64:$src)>;
7872def : Pat<(v8i8  (bitconvert (f64   FPR64:$src))), (v8i8  FPR64:$src)>;
7873def : Pat<(v8i8  (bitconvert (v2f32 FPR64:$src))), (v8i8  FPR64:$src)>;
7874def : Pat<(v8i8  (bitconvert (v1f64 FPR64:$src))), (v8i8  FPR64:$src)>;
7875def : Pat<(v8i8  (bitconvert (v4f16 FPR64:$src))), (v8i8  FPR64:$src)>;
7876def : Pat<(v8i8  (bitconvert (v4bf16 FPR64:$src))), (v8i8  FPR64:$src)>;
7877}
7878let Predicates = [IsBE] in {
7879def : Pat<(v8i8  (bitconvert (v1i64 FPR64:$src))),
7880                             (v8i8 (REV64v8i8 FPR64:$src))>;
7881def : Pat<(v8i8  (bitconvert (v2i32 FPR64:$src))),
7882                             (v8i8 (REV32v8i8 FPR64:$src))>;
7883def : Pat<(v8i8  (bitconvert (v4i16 FPR64:$src))),
7884                             (v8i8 (REV16v8i8 FPR64:$src))>;
7885def : Pat<(v8i8  (bitconvert (f64   FPR64:$src))),
7886                             (v8i8 (REV64v8i8 FPR64:$src))>;
7887def : Pat<(v8i8  (bitconvert (v2f32 FPR64:$src))),
7888                             (v8i8 (REV32v8i8 FPR64:$src))>;
7889def : Pat<(v8i8  (bitconvert (v1f64 FPR64:$src))),
7890                             (v8i8 (REV64v8i8 FPR64:$src))>;
7891def : Pat<(v8i8  (bitconvert (v4f16 FPR64:$src))),
7892                             (v8i8 (REV16v8i8 FPR64:$src))>;
7893def : Pat<(v8i8  (bitconvert (v4bf16 FPR64:$src))),
7894                             (v8i8 (REV16v8i8 FPR64:$src))>;
7895}
7896
7897let Predicates = [IsLE] in {
7898def : Pat<(f64   (bitconvert (v2i32 FPR64:$src))), (f64   FPR64:$src)>;
7899def : Pat<(f64   (bitconvert (v4i16 FPR64:$src))), (f64   FPR64:$src)>;
7900def : Pat<(f64   (bitconvert (v2f32 FPR64:$src))), (f64   FPR64:$src)>;
7901def : Pat<(f64   (bitconvert (v8i8  FPR64:$src))), (f64   FPR64:$src)>;
7902def : Pat<(f64   (bitconvert (v4f16 FPR64:$src))), (f64   FPR64:$src)>;
7903def : Pat<(f64   (bitconvert (v4bf16 FPR64:$src))), (f64   FPR64:$src)>;
7904}
7905let Predicates = [IsBE] in {
7906def : Pat<(f64   (bitconvert (v2i32 FPR64:$src))),
7907                             (f64 (REV64v2i32 FPR64:$src))>;
7908def : Pat<(f64   (bitconvert (v4i16 FPR64:$src))),
7909                             (f64 (REV64v4i16 FPR64:$src))>;
7910def : Pat<(f64   (bitconvert (v2f32 FPR64:$src))),
7911                             (f64 (REV64v2i32 FPR64:$src))>;
7912def : Pat<(f64   (bitconvert (v8i8  FPR64:$src))),
7913                             (f64 (REV64v8i8 FPR64:$src))>;
7914def : Pat<(f64   (bitconvert (v4f16 FPR64:$src))),
7915                             (f64 (REV64v4i16 FPR64:$src))>;
7916def : Pat<(f64   (bitconvert (v4bf16 FPR64:$src))),
7917                             (f64 (REV64v4i16 FPR64:$src))>;
7918}
7919def : Pat<(f64   (bitconvert (v1i64 FPR64:$src))), (f64   FPR64:$src)>;
7920def : Pat<(f64   (bitconvert (v1f64 FPR64:$src))), (f64   FPR64:$src)>;
7921
7922let Predicates = [IsLE] in {
7923def : Pat<(v1f64 (bitconvert (v2i32 FPR64:$src))), (v1f64 FPR64:$src)>;
7924def : Pat<(v1f64 (bitconvert (v4i16 FPR64:$src))), (v1f64 FPR64:$src)>;
7925def : Pat<(v1f64 (bitconvert (v8i8  FPR64:$src))), (v1f64 FPR64:$src)>;
7926def : Pat<(v1f64 (bitconvert (v2f32 FPR64:$src))), (v1f64 FPR64:$src)>;
7927def : Pat<(v1f64 (bitconvert (v4f16 FPR64:$src))), (v1f64 FPR64:$src)>;
7928def : Pat<(v1f64 (bitconvert (v4bf16 FPR64:$src))), (v1f64 FPR64:$src)>;
7929}
7930let Predicates = [IsBE] in {
7931def : Pat<(v1f64 (bitconvert (v2i32 FPR64:$src))),
7932                             (v1f64 (REV64v2i32 FPR64:$src))>;
7933def : Pat<(v1f64 (bitconvert (v4i16 FPR64:$src))),
7934                             (v1f64 (REV64v4i16 FPR64:$src))>;
7935def : Pat<(v1f64 (bitconvert (v8i8  FPR64:$src))),
7936                             (v1f64 (REV64v8i8 FPR64:$src))>;
7937def : Pat<(v1f64 (bitconvert (v2f32 FPR64:$src))),
7938                             (v1f64 (REV64v2i32 FPR64:$src))>;
7939def : Pat<(v1f64 (bitconvert (v4f16 FPR64:$src))),
7940                             (v1f64 (REV64v4i16 FPR64:$src))>;
7941def : Pat<(v1f64 (bitconvert (v4bf16 FPR64:$src))),
7942                             (v1f64 (REV64v4i16 FPR64:$src))>;
7943}
7944def : Pat<(v1f64 (bitconvert (v1i64 FPR64:$src))), (v1f64 FPR64:$src)>;
7945def : Pat<(v1f64 (bitconvert (f64   FPR64:$src))), (v1f64 FPR64:$src)>;
7946
7947let Predicates = [IsLE] in {
7948def : Pat<(v2f32 (bitconvert (v1i64 FPR64:$src))), (v2f32 FPR64:$src)>;
7949def : Pat<(v2f32 (bitconvert (v4i16 FPR64:$src))), (v2f32 FPR64:$src)>;
7950def : Pat<(v2f32 (bitconvert (v8i8  FPR64:$src))), (v2f32 FPR64:$src)>;
7951def : Pat<(v2f32 (bitconvert (v1f64 FPR64:$src))), (v2f32 FPR64:$src)>;
7952def : Pat<(v2f32 (bitconvert (f64   FPR64:$src))), (v2f32 FPR64:$src)>;
7953def : Pat<(v2f32 (bitconvert (v4f16 FPR64:$src))), (v2f32 FPR64:$src)>;
7954def : Pat<(v2f32 (bitconvert (v4bf16 FPR64:$src))), (v2f32 FPR64:$src)>;
7955}
7956let Predicates = [IsBE] in {
7957def : Pat<(v2f32 (bitconvert (v1i64 FPR64:$src))),
7958                             (v2f32 (REV64v2i32 FPR64:$src))>;
7959def : Pat<(v2f32 (bitconvert (v4i16 FPR64:$src))),
7960                             (v2f32 (REV32v4i16 FPR64:$src))>;
7961def : Pat<(v2f32 (bitconvert (v8i8  FPR64:$src))),
7962                             (v2f32 (REV32v8i8 FPR64:$src))>;
7963def : Pat<(v2f32 (bitconvert (v1f64 FPR64:$src))),
7964                             (v2f32 (REV64v2i32 FPR64:$src))>;
7965def : Pat<(v2f32 (bitconvert (f64   FPR64:$src))),
7966                             (v2f32 (REV64v2i32 FPR64:$src))>;
7967def : Pat<(v2f32 (bitconvert (v4f16 FPR64:$src))),
7968                             (v2f32 (REV32v4i16 FPR64:$src))>;
7969def : Pat<(v2f32 (bitconvert (v4bf16 FPR64:$src))),
7970                             (v2f32 (REV32v4i16 FPR64:$src))>;
7971}
7972def : Pat<(v2f32 (bitconvert (v2i32 FPR64:$src))), (v2f32 FPR64:$src)>;
7973
7974let Predicates = [IsLE] in {
7975def : Pat<(f128 (bitconvert (v2i64 FPR128:$src))), (f128 FPR128:$src)>;
7976def : Pat<(f128 (bitconvert (v4i32 FPR128:$src))), (f128 FPR128:$src)>;
7977def : Pat<(f128 (bitconvert (v8i16 FPR128:$src))), (f128 FPR128:$src)>;
7978def : Pat<(f128 (bitconvert (v2f64 FPR128:$src))), (f128 FPR128:$src)>;
7979def : Pat<(f128 (bitconvert (v4f32 FPR128:$src))), (f128 FPR128:$src)>;
7980def : Pat<(f128 (bitconvert (v8f16 FPR128:$src))), (f128 FPR128:$src)>;
7981def : Pat<(f128 (bitconvert (v8bf16 FPR128:$src))), (f128 FPR128:$src)>;
7982def : Pat<(f128 (bitconvert (v16i8 FPR128:$src))), (f128 FPR128:$src)>;
7983}
7984let Predicates = [IsBE] in {
7985def : Pat<(f128 (bitconvert (v2i64 FPR128:$src))),
7986                            (f128 (EXTv16i8 FPR128:$src, FPR128:$src, (i32 8)))>;
7987def : Pat<(f128 (bitconvert (v4i32 FPR128:$src))),
7988                            (f128 (EXTv16i8 (REV64v4i32 FPR128:$src),
7989                                            (REV64v4i32 FPR128:$src), (i32 8)))>;
7990def : Pat<(f128 (bitconvert (v8i16 FPR128:$src))),
7991                            (f128 (EXTv16i8 (REV64v8i16 FPR128:$src),
7992                                            (REV64v8i16 FPR128:$src), (i32 8)))>;
7993def : Pat<(f128 (bitconvert (v8f16 FPR128:$src))),
7994                            (f128 (EXTv16i8 (REV64v8i16 FPR128:$src),
7995                                            (REV64v8i16 FPR128:$src), (i32 8)))>;
7996def : Pat<(f128 (bitconvert (v8bf16 FPR128:$src))),
7997                            (f128 (EXTv16i8 (REV64v8i16 FPR128:$src),
7998                                            (REV64v8i16 FPR128:$src), (i32 8)))>;
7999def : Pat<(f128 (bitconvert (v2f64 FPR128:$src))),
8000                            (f128 (EXTv16i8 FPR128:$src, FPR128:$src, (i32 8)))>;
8001def : Pat<(f128 (bitconvert (v4f32 FPR128:$src))),
8002                            (f128 (EXTv16i8 (REV64v4i32 FPR128:$src),
8003                                            (REV64v4i32 FPR128:$src), (i32 8)))>;
8004def : Pat<(f128 (bitconvert (v16i8 FPR128:$src))),
8005                            (f128 (EXTv16i8 (REV64v16i8 FPR128:$src),
8006                                            (REV64v16i8 FPR128:$src), (i32 8)))>;
8007}
8008
8009let Predicates = [IsLE] in {
8010def : Pat<(v2f64 (bitconvert (f128  FPR128:$src))), (v2f64 FPR128:$src)>;
8011def : Pat<(v2f64 (bitconvert (v4i32 FPR128:$src))), (v2f64 FPR128:$src)>;
8012def : Pat<(v2f64 (bitconvert (v8i16 FPR128:$src))), (v2f64 FPR128:$src)>;
8013def : Pat<(v2f64 (bitconvert (v8f16 FPR128:$src))), (v2f64 FPR128:$src)>;
8014def : Pat<(v2f64 (bitconvert (v8bf16 FPR128:$src))), (v2f64 FPR128:$src)>;
8015def : Pat<(v2f64 (bitconvert (v16i8 FPR128:$src))), (v2f64 FPR128:$src)>;
8016def : Pat<(v2f64 (bitconvert (v4f32 FPR128:$src))), (v2f64 FPR128:$src)>;
8017}
8018let Predicates = [IsBE] in {
8019def : Pat<(v2f64 (bitconvert (f128  FPR128:$src))),
8020                             (v2f64 (EXTv16i8 FPR128:$src,
8021                                              FPR128:$src, (i32 8)))>;
8022def : Pat<(v2f64 (bitconvert (v4i32 FPR128:$src))),
8023                             (v2f64 (REV64v4i32 FPR128:$src))>;
8024def : Pat<(v2f64 (bitconvert (v8i16 FPR128:$src))),
8025                             (v2f64 (REV64v8i16 FPR128:$src))>;
8026def : Pat<(v2f64 (bitconvert (v8f16 FPR128:$src))),
8027                             (v2f64 (REV64v8i16 FPR128:$src))>;
8028def : Pat<(v2f64 (bitconvert (v8bf16 FPR128:$src))),
8029                             (v2f64 (REV64v8i16 FPR128:$src))>;
8030def : Pat<(v2f64 (bitconvert (v16i8 FPR128:$src))),
8031                             (v2f64 (REV64v16i8 FPR128:$src))>;
8032def : Pat<(v2f64 (bitconvert (v4f32 FPR128:$src))),
8033                             (v2f64 (REV64v4i32 FPR128:$src))>;
8034}
8035def : Pat<(v2f64 (bitconvert (v2i64 FPR128:$src))), (v2f64 FPR128:$src)>;
8036
8037let Predicates = [IsLE] in {
8038def : Pat<(v4f32 (bitconvert (f128  FPR128:$src))), (v4f32 FPR128:$src)>;
8039def : Pat<(v4f32 (bitconvert (v8i16 FPR128:$src))), (v4f32 FPR128:$src)>;
8040def : Pat<(v4f32 (bitconvert (v8f16 FPR128:$src))), (v4f32 FPR128:$src)>;
8041def : Pat<(v4f32 (bitconvert (v8bf16 FPR128:$src))), (v4f32 FPR128:$src)>;
8042def : Pat<(v4f32 (bitconvert (v16i8 FPR128:$src))), (v4f32 FPR128:$src)>;
8043def : Pat<(v4f32 (bitconvert (v2i64 FPR128:$src))), (v4f32 FPR128:$src)>;
8044def : Pat<(v4f32 (bitconvert (v2f64 FPR128:$src))), (v4f32 FPR128:$src)>;
8045}
8046let Predicates = [IsBE] in {
8047def : Pat<(v4f32 (bitconvert (f128  FPR128:$src))),
8048                             (v4f32 (EXTv16i8 (REV64v4i32 FPR128:$src),
8049                                    (REV64v4i32 FPR128:$src), (i32 8)))>;
8050def : Pat<(v4f32 (bitconvert (v8i16 FPR128:$src))),
8051                             (v4f32 (REV32v8i16 FPR128:$src))>;
8052def : Pat<(v4f32 (bitconvert (v8f16 FPR128:$src))),
8053                             (v4f32 (REV32v8i16 FPR128:$src))>;
8054def : Pat<(v4f32 (bitconvert (v8bf16 FPR128:$src))),
8055                             (v4f32 (REV32v8i16 FPR128:$src))>;
8056def : Pat<(v4f32 (bitconvert (v16i8 FPR128:$src))),
8057                             (v4f32 (REV32v16i8 FPR128:$src))>;
8058def : Pat<(v4f32 (bitconvert (v2i64 FPR128:$src))),
8059                             (v4f32 (REV64v4i32 FPR128:$src))>;
8060def : Pat<(v4f32 (bitconvert (v2f64 FPR128:$src))),
8061                             (v4f32 (REV64v4i32 FPR128:$src))>;
8062}
8063def : Pat<(v4f32 (bitconvert (v4i32 FPR128:$src))), (v4f32 FPR128:$src)>;
8064
8065let Predicates = [IsLE] in {
8066def : Pat<(v2i64 (bitconvert (f128  FPR128:$src))), (v2i64 FPR128:$src)>;
8067def : Pat<(v2i64 (bitconvert (v4i32 FPR128:$src))), (v2i64 FPR128:$src)>;
8068def : Pat<(v2i64 (bitconvert (v8i16 FPR128:$src))), (v2i64 FPR128:$src)>;
8069def : Pat<(v2i64 (bitconvert (v16i8 FPR128:$src))), (v2i64 FPR128:$src)>;
8070def : Pat<(v2i64 (bitconvert (v4f32 FPR128:$src))), (v2i64 FPR128:$src)>;
8071def : Pat<(v2i64 (bitconvert (v8f16 FPR128:$src))), (v2i64 FPR128:$src)>;
8072def : Pat<(v2i64 (bitconvert (v8bf16 FPR128:$src))), (v2i64 FPR128:$src)>;
8073}
8074let Predicates = [IsBE] in {
8075def : Pat<(v2i64 (bitconvert (f128  FPR128:$src))),
8076                             (v2i64 (EXTv16i8 FPR128:$src,
8077                                              FPR128:$src, (i32 8)))>;
8078def : Pat<(v2i64 (bitconvert (v4i32 FPR128:$src))),
8079                             (v2i64 (REV64v4i32 FPR128:$src))>;
8080def : Pat<(v2i64 (bitconvert (v8i16 FPR128:$src))),
8081                             (v2i64 (REV64v8i16 FPR128:$src))>;
8082def : Pat<(v2i64 (bitconvert (v16i8 FPR128:$src))),
8083                             (v2i64 (REV64v16i8 FPR128:$src))>;
8084def : Pat<(v2i64 (bitconvert (v4f32 FPR128:$src))),
8085                             (v2i64 (REV64v4i32 FPR128:$src))>;
8086def : Pat<(v2i64 (bitconvert (v8f16 FPR128:$src))),
8087                             (v2i64 (REV64v8i16 FPR128:$src))>;
8088def : Pat<(v2i64 (bitconvert (v8bf16 FPR128:$src))),
8089                             (v2i64 (REV64v8i16 FPR128:$src))>;
8090}
8091def : Pat<(v2i64 (bitconvert (v2f64 FPR128:$src))), (v2i64 FPR128:$src)>;
8092
8093let Predicates = [IsLE] in {
8094def : Pat<(v4i32 (bitconvert (f128  FPR128:$src))), (v4i32 FPR128:$src)>;
8095def : Pat<(v4i32 (bitconvert (v2i64 FPR128:$src))), (v4i32 FPR128:$src)>;
8096def : Pat<(v4i32 (bitconvert (v8i16 FPR128:$src))), (v4i32 FPR128:$src)>;
8097def : Pat<(v4i32 (bitconvert (v16i8 FPR128:$src))), (v4i32 FPR128:$src)>;
8098def : Pat<(v4i32 (bitconvert (v2f64 FPR128:$src))), (v4i32 FPR128:$src)>;
8099def : Pat<(v4i32 (bitconvert (v8f16 FPR128:$src))), (v4i32 FPR128:$src)>;
8100def : Pat<(v4i32 (bitconvert (v8bf16 FPR128:$src))), (v4i32 FPR128:$src)>;
8101}
8102let Predicates = [IsBE] in {
8103def : Pat<(v4i32 (bitconvert (f128  FPR128:$src))),
8104                             (v4i32 (EXTv16i8 (REV64v4i32 FPR128:$src),
8105                                              (REV64v4i32 FPR128:$src),
8106                                              (i32 8)))>;
8107def : Pat<(v4i32 (bitconvert (v2i64 FPR128:$src))),
8108                             (v4i32 (REV64v4i32 FPR128:$src))>;
8109def : Pat<(v4i32 (bitconvert (v8i16 FPR128:$src))),
8110                             (v4i32 (REV32v8i16 FPR128:$src))>;
8111def : Pat<(v4i32 (bitconvert (v16i8 FPR128:$src))),
8112                             (v4i32 (REV32v16i8 FPR128:$src))>;
8113def : Pat<(v4i32 (bitconvert (v2f64 FPR128:$src))),
8114                             (v4i32 (REV64v4i32 FPR128:$src))>;
8115def : Pat<(v4i32 (bitconvert (v8f16 FPR128:$src))),
8116                             (v4i32 (REV32v8i16 FPR128:$src))>;
8117def : Pat<(v4i32 (bitconvert (v8bf16 FPR128:$src))),
8118                             (v4i32 (REV32v8i16 FPR128:$src))>;
8119}
8120def : Pat<(v4i32 (bitconvert (v4f32 FPR128:$src))), (v4i32 FPR128:$src)>;
8121
8122let Predicates = [IsLE] in {
8123def : Pat<(v8i16 (bitconvert (f128  FPR128:$src))), (v8i16 FPR128:$src)>;
8124def : Pat<(v8i16 (bitconvert (v2i64 FPR128:$src))), (v8i16 FPR128:$src)>;
8125def : Pat<(v8i16 (bitconvert (v4i32 FPR128:$src))), (v8i16 FPR128:$src)>;
8126def : Pat<(v8i16 (bitconvert (v16i8 FPR128:$src))), (v8i16 FPR128:$src)>;
8127def : Pat<(v8i16 (bitconvert (v2f64 FPR128:$src))), (v8i16 FPR128:$src)>;
8128def : Pat<(v8i16 (bitconvert (v4f32 FPR128:$src))), (v8i16 FPR128:$src)>;
8129}
8130let Predicates = [IsBE] in {
8131def : Pat<(v8i16 (bitconvert (f128  FPR128:$src))),
8132                             (v8i16 (EXTv16i8 (REV64v8i16 FPR128:$src),
8133                                              (REV64v8i16 FPR128:$src),
8134                                              (i32 8)))>;
8135def : Pat<(v8i16 (bitconvert (v2i64 FPR128:$src))),
8136                             (v8i16 (REV64v8i16 FPR128:$src))>;
8137def : Pat<(v8i16 (bitconvert (v4i32 FPR128:$src))),
8138                             (v8i16 (REV32v8i16 FPR128:$src))>;
8139def : Pat<(v8i16 (bitconvert (v16i8 FPR128:$src))),
8140                             (v8i16 (REV16v16i8 FPR128:$src))>;
8141def : Pat<(v8i16 (bitconvert (v2f64 FPR128:$src))),
8142                             (v8i16 (REV64v8i16 FPR128:$src))>;
8143def : Pat<(v8i16 (bitconvert (v4f32 FPR128:$src))),
8144                             (v8i16 (REV32v8i16 FPR128:$src))>;
8145}
8146def : Pat<(v8i16 (bitconvert (v8f16 FPR128:$src))), (v8i16 FPR128:$src)>;
8147def : Pat<(v8i16 (bitconvert (v8bf16 FPR128:$src))), (v8i16 FPR128:$src)>;
8148
8149let Predicates = [IsLE] in {
8150def : Pat<(v8f16 (bitconvert (f128  FPR128:$src))), (v8f16 FPR128:$src)>;
8151def : Pat<(v8f16 (bitconvert (v2i64 FPR128:$src))), (v8f16 FPR128:$src)>;
8152def : Pat<(v8f16 (bitconvert (v4i32 FPR128:$src))), (v8f16 FPR128:$src)>;
8153def : Pat<(v8f16 (bitconvert (v16i8 FPR128:$src))), (v8f16 FPR128:$src)>;
8154def : Pat<(v8f16 (bitconvert (v2f64 FPR128:$src))), (v8f16 FPR128:$src)>;
8155def : Pat<(v8f16 (bitconvert (v4f32 FPR128:$src))), (v8f16 FPR128:$src)>;
8156
8157def : Pat<(v8bf16 (bitconvert (f128  FPR128:$src))), (v8bf16 FPR128:$src)>;
8158def : Pat<(v8bf16 (bitconvert (v2i64 FPR128:$src))), (v8bf16 FPR128:$src)>;
8159def : Pat<(v8bf16 (bitconvert (v4i32 FPR128:$src))), (v8bf16 FPR128:$src)>;
8160def : Pat<(v8bf16 (bitconvert (v16i8 FPR128:$src))), (v8bf16 FPR128:$src)>;
8161def : Pat<(v8bf16 (bitconvert (v2f64 FPR128:$src))), (v8bf16 FPR128:$src)>;
8162def : Pat<(v8bf16 (bitconvert (v4f32 FPR128:$src))), (v8bf16 FPR128:$src)>;
8163}
8164let Predicates = [IsBE] in {
8165def : Pat<(v8f16 (bitconvert (f128  FPR128:$src))),
8166                             (v8f16 (EXTv16i8 (REV64v8i16 FPR128:$src),
8167                                              (REV64v8i16 FPR128:$src),
8168                                              (i32 8)))>;
8169def : Pat<(v8f16 (bitconvert (v2i64 FPR128:$src))),
8170                             (v8f16 (REV64v8i16 FPR128:$src))>;
8171def : Pat<(v8f16 (bitconvert (v4i32 FPR128:$src))),
8172                             (v8f16 (REV32v8i16 FPR128:$src))>;
8173def : Pat<(v8f16 (bitconvert (v16i8 FPR128:$src))),
8174                             (v8f16 (REV16v16i8 FPR128:$src))>;
8175def : Pat<(v8f16 (bitconvert (v2f64 FPR128:$src))),
8176                             (v8f16 (REV64v8i16 FPR128:$src))>;
8177def : Pat<(v8f16 (bitconvert (v4f32 FPR128:$src))),
8178                             (v8f16 (REV32v8i16 FPR128:$src))>;
8179
8180def : Pat<(v8bf16 (bitconvert (f128  FPR128:$src))),
8181                             (v8bf16 (EXTv16i8 (REV64v8i16 FPR128:$src),
8182                                              (REV64v8i16 FPR128:$src),
8183                                              (i32 8)))>;
8184def : Pat<(v8bf16 (bitconvert (v2i64 FPR128:$src))),
8185                             (v8bf16 (REV64v8i16 FPR128:$src))>;
8186def : Pat<(v8bf16 (bitconvert (v4i32 FPR128:$src))),
8187                             (v8bf16 (REV32v8i16 FPR128:$src))>;
8188def : Pat<(v8bf16 (bitconvert (v16i8 FPR128:$src))),
8189                             (v8bf16 (REV16v16i8 FPR128:$src))>;
8190def : Pat<(v8bf16 (bitconvert (v2f64 FPR128:$src))),
8191                             (v8bf16 (REV64v8i16 FPR128:$src))>;
8192def : Pat<(v8bf16 (bitconvert (v4f32 FPR128:$src))),
8193                             (v8bf16 (REV32v8i16 FPR128:$src))>;
8194}
8195def : Pat<(v8f16 (bitconvert (v8i16 FPR128:$src))), (v8f16 FPR128:$src)>;
8196def : Pat<(v8bf16 (bitconvert (v8i16 FPR128:$src))), (v8bf16 FPR128:$src)>;
8197
8198let Predicates = [IsLE] in {
8199def : Pat<(v16i8 (bitconvert (f128  FPR128:$src))), (v16i8 FPR128:$src)>;
8200def : Pat<(v16i8 (bitconvert (v2i64 FPR128:$src))), (v16i8 FPR128:$src)>;
8201def : Pat<(v16i8 (bitconvert (v4i32 FPR128:$src))), (v16i8 FPR128:$src)>;
8202def : Pat<(v16i8 (bitconvert (v8i16 FPR128:$src))), (v16i8 FPR128:$src)>;
8203def : Pat<(v16i8 (bitconvert (v2f64 FPR128:$src))), (v16i8 FPR128:$src)>;
8204def : Pat<(v16i8 (bitconvert (v4f32 FPR128:$src))), (v16i8 FPR128:$src)>;
8205def : Pat<(v16i8 (bitconvert (v8f16 FPR128:$src))), (v16i8 FPR128:$src)>;
8206def : Pat<(v16i8 (bitconvert (v8bf16 FPR128:$src))), (v16i8 FPR128:$src)>;
8207}
8208let Predicates = [IsBE] in {
8209def : Pat<(v16i8 (bitconvert (f128  FPR128:$src))),
8210                             (v16i8 (EXTv16i8 (REV64v16i8 FPR128:$src),
8211                                              (REV64v16i8 FPR128:$src),
8212                                              (i32 8)))>;
8213def : Pat<(v16i8 (bitconvert (v2i64 FPR128:$src))),
8214                             (v16i8 (REV64v16i8 FPR128:$src))>;
8215def : Pat<(v16i8 (bitconvert (v4i32 FPR128:$src))),
8216                             (v16i8 (REV32v16i8 FPR128:$src))>;
8217def : Pat<(v16i8 (bitconvert (v8i16 FPR128:$src))),
8218                             (v16i8 (REV16v16i8 FPR128:$src))>;
8219def : Pat<(v16i8 (bitconvert (v2f64 FPR128:$src))),
8220                             (v16i8 (REV64v16i8 FPR128:$src))>;
8221def : Pat<(v16i8 (bitconvert (v4f32 FPR128:$src))),
8222                             (v16i8 (REV32v16i8 FPR128:$src))>;
8223def : Pat<(v16i8 (bitconvert (v8f16 FPR128:$src))),
8224                             (v16i8 (REV16v16i8 FPR128:$src))>;
8225def : Pat<(v16i8 (bitconvert (v8bf16 FPR128:$src))),
8226                             (v16i8 (REV16v16i8 FPR128:$src))>;
8227}
8228
8229def : Pat<(v4i16 (extract_subvector V128:$Rn, (i64 0))),
8230           (EXTRACT_SUBREG V128:$Rn, dsub)>;
8231def : Pat<(v8i8 (extract_subvector V128:$Rn, (i64 0))),
8232           (EXTRACT_SUBREG V128:$Rn, dsub)>;
8233def : Pat<(v2f32 (extract_subvector V128:$Rn, (i64 0))),
8234           (EXTRACT_SUBREG V128:$Rn, dsub)>;
8235def : Pat<(v4f16 (extract_subvector V128:$Rn, (i64 0))),
8236           (EXTRACT_SUBREG V128:$Rn, dsub)>;
8237def : Pat<(v4bf16 (extract_subvector V128:$Rn, (i64 0))),
8238           (EXTRACT_SUBREG V128:$Rn, dsub)>;
8239def : Pat<(v2i32 (extract_subvector V128:$Rn, (i64 0))),
8240           (EXTRACT_SUBREG V128:$Rn, dsub)>;
8241def : Pat<(v1i64 (extract_subvector V128:$Rn, (i64 0))),
8242           (EXTRACT_SUBREG V128:$Rn, dsub)>;
8243def : Pat<(v1f64 (extract_subvector V128:$Rn, (i64 0))),
8244           (EXTRACT_SUBREG V128:$Rn, dsub)>;
8245
8246def : Pat<(v8i8 (extract_subvector (v16i8 FPR128:$Rn), (i64 1))),
8247          (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
8248def : Pat<(v4i16 (extract_subvector (v8i16 FPR128:$Rn), (i64 1))),
8249          (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
8250def : Pat<(v2i32 (extract_subvector (v4i32 FPR128:$Rn), (i64 1))),
8251          (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
8252def : Pat<(v1i64 (extract_subvector (v2i64 FPR128:$Rn), (i64 1))),
8253          (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
8254
8255// A 64-bit subvector insert to the first 128-bit vector position
8256// is a subregister copy that needs no instruction.
8257multiclass InsertSubvectorUndef<ValueType Ty> {
8258  def : Pat<(insert_subvector undef, (v1i64 FPR64:$src), (Ty 0)),
8259            (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
8260  def : Pat<(insert_subvector undef, (v1f64 FPR64:$src), (Ty 0)),
8261            (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
8262  def : Pat<(insert_subvector undef, (v2i32 FPR64:$src), (Ty 0)),
8263            (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
8264  def : Pat<(insert_subvector undef, (v2f32 FPR64:$src), (Ty 0)),
8265            (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
8266  def : Pat<(insert_subvector undef, (v4i16 FPR64:$src), (Ty 0)),
8267            (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
8268  def : Pat<(insert_subvector undef, (v4f16 FPR64:$src), (Ty 0)),
8269            (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
8270  def : Pat<(insert_subvector undef, (v4bf16 FPR64:$src), (Ty 0)),
8271            (INSERT_SUBREG (v8bf16 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
8272  def : Pat<(insert_subvector undef, (v8i8 FPR64:$src), (Ty 0)),
8273            (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
8274}
8275
8276defm : InsertSubvectorUndef<i32>;
8277defm : InsertSubvectorUndef<i64>;
8278
8279// Use pair-wise add instructions when summing up the lanes for v2f64, v2i64
8280// or v2f32.
8281def : Pat<(i64 (add (vector_extract (v2i64 FPR128:$Rn), (i64 0)),
8282                    (vector_extract (v2i64 FPR128:$Rn), (i64 1)))),
8283           (i64 (ADDPv2i64p (v2i64 FPR128:$Rn)))>;
8284def : Pat<(f64 (any_fadd (vector_extract (v2f64 FPR128:$Rn), (i64 0)),
8285                         (vector_extract (v2f64 FPR128:$Rn), (i64 1)))),
8286           (f64 (FADDPv2i64p (v2f64 FPR128:$Rn)))>;
8287    // vector_extract on 64-bit vectors gets promoted to a 128 bit vector,
8288    // so we match on v4f32 here, not v2f32. This will also catch adding
8289    // the low two lanes of a true v4f32 vector.
8290def : Pat<(any_fadd (vector_extract (v4f32 FPR128:$Rn), (i64 0)),
8291                    (vector_extract (v4f32 FPR128:$Rn), (i64 1))),
8292          (f32 (FADDPv2i32p (EXTRACT_SUBREG FPR128:$Rn, dsub)))>;
8293def : Pat<(any_fadd (vector_extract (v8f16 FPR128:$Rn), (i64 0)),
8294                    (vector_extract (v8f16 FPR128:$Rn), (i64 1))),
8295          (f16 (FADDPv2i16p (EXTRACT_SUBREG FPR128:$Rn, dsub)))>;
8296
8297// Scalar 64-bit shifts in FPR64 registers.
8298def : Pat<(i64 (int_aarch64_neon_sshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
8299          (SSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
8300def : Pat<(i64 (int_aarch64_neon_ushl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
8301          (USHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
8302def : Pat<(i64 (int_aarch64_neon_srshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
8303          (SRSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
8304def : Pat<(i64 (int_aarch64_neon_urshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
8305          (URSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
8306
8307// Patterns for nontemporal/no-allocate stores.
8308// We have to resort to tricks to turn a single-input store into a store pair,
8309// because there is no single-input nontemporal store, only STNP.
8310let Predicates = [IsLE] in {
8311let AddedComplexity = 15 in {
8312class NTStore128Pat<ValueType VT> :
8313  Pat<(nontemporalstore (VT FPR128:$Rt),
8314        (am_indexed7s64 GPR64sp:$Rn, simm7s8:$offset)),
8315      (STNPDi (EXTRACT_SUBREG FPR128:$Rt, dsub),
8316              (DUPi64 FPR128:$Rt, (i64 1)),
8317              GPR64sp:$Rn, simm7s8:$offset)>;
8318
8319def : NTStore128Pat<v2i64>;
8320def : NTStore128Pat<v4i32>;
8321def : NTStore128Pat<v8i16>;
8322def : NTStore128Pat<v16i8>;
8323
8324class NTStore64Pat<ValueType VT> :
8325  Pat<(nontemporalstore (VT FPR64:$Rt),
8326        (am_indexed7s32 GPR64sp:$Rn, simm7s4:$offset)),
8327      (STNPSi (EXTRACT_SUBREG FPR64:$Rt, ssub),
8328              (DUPi32 (SUBREG_TO_REG (i64 0), FPR64:$Rt, dsub), (i64 1)),
8329              GPR64sp:$Rn, simm7s4:$offset)>;
8330
8331// FIXME: Shouldn't v1f64 loads/stores be promoted to v1i64?
8332def : NTStore64Pat<v1f64>;
8333def : NTStore64Pat<v1i64>;
8334def : NTStore64Pat<v2i32>;
8335def : NTStore64Pat<v4i16>;
8336def : NTStore64Pat<v8i8>;
8337
8338def : Pat<(nontemporalstore GPR64:$Rt,
8339            (am_indexed7s32 GPR64sp:$Rn, simm7s4:$offset)),
8340          (STNPWi (EXTRACT_SUBREG GPR64:$Rt, sub_32),
8341                  (EXTRACT_SUBREG (UBFMXri GPR64:$Rt, 32, 63), sub_32),
8342                  GPR64sp:$Rn, simm7s4:$offset)>;
8343} // AddedComplexity=10
8344} // Predicates = [IsLE]
8345
8346// Tail call return handling. These are all compiler pseudo-instructions,
8347// so no encoding information or anything like that.
8348let isCall = 1, isTerminator = 1, isReturn = 1, isBarrier = 1, Uses = [SP] in {
8349  def TCRETURNdi : Pseudo<(outs), (ins i64imm:$dst, i32imm:$FPDiff), []>,
8350                   Sched<[WriteBrReg]>;
8351  def TCRETURNri : Pseudo<(outs), (ins tcGPR64:$dst, i32imm:$FPDiff), []>,
8352                   Sched<[WriteBrReg]>;
8353  // Indirect tail-call with any register allowed, used by MachineOutliner when
8354  // this is proven safe.
8355  // FIXME: If we have to add any more hacks like this, we should instead relax
8356  // some verifier checks for outlined functions.
8357  def TCRETURNriALL : Pseudo<(outs), (ins GPR64:$dst, i32imm:$FPDiff), []>,
8358                      Sched<[WriteBrReg]>;
8359  // Indirect tail-call limited to only use registers (x16 and x17) which are
8360  // allowed to tail-call a "BTI c" instruction.
8361  def TCRETURNriBTI : Pseudo<(outs), (ins rtcGPR64:$dst, i32imm:$FPDiff), []>,
8362                      Sched<[WriteBrReg]>;
8363}
8364
8365def : Pat<(AArch64tcret tcGPR64:$dst, (i32 timm:$FPDiff)),
8366          (TCRETURNri tcGPR64:$dst, imm:$FPDiff)>,
8367      Requires<[NotUseBTI]>;
8368def : Pat<(AArch64tcret rtcGPR64:$dst, (i32 timm:$FPDiff)),
8369          (TCRETURNriBTI rtcGPR64:$dst, imm:$FPDiff)>,
8370      Requires<[UseBTI]>;
8371def : Pat<(AArch64tcret tglobaladdr:$dst, (i32 timm:$FPDiff)),
8372          (TCRETURNdi texternalsym:$dst, imm:$FPDiff)>;
8373def : Pat<(AArch64tcret texternalsym:$dst, (i32 timm:$FPDiff)),
8374          (TCRETURNdi texternalsym:$dst, imm:$FPDiff)>;
8375
8376def MOVMCSym : Pseudo<(outs GPR64:$dst), (ins i64imm:$sym), []>, Sched<[]>;
8377def : Pat<(i64 (AArch64LocalRecover mcsym:$sym)), (MOVMCSym mcsym:$sym)>;
8378
8379// Extracting lane zero is a special case where we can just use a plain
8380// EXTRACT_SUBREG instruction, which will become FMOV. This is easier for the
8381// rest of the compiler, especially the register allocator and copy propagation,
8382// to reason about, so is preferred when it's possible to use it.
8383let AddedComplexity = 10 in {
8384  def : Pat<(i64 (extractelt (v2i64 V128:$V), (i64 0))), (EXTRACT_SUBREG V128:$V, dsub)>;
8385  def : Pat<(i32 (extractelt (v4i32 V128:$V), (i64 0))), (EXTRACT_SUBREG V128:$V, ssub)>;
8386  def : Pat<(i32 (extractelt (v2i32 V64:$V), (i64 0))), (EXTRACT_SUBREG V64:$V, ssub)>;
8387}
8388
8389// dot_v4i8
8390class mul_v4i8<SDPatternOperator ldop> :
8391  PatFrag<(ops node:$Rn, node:$Rm, node:$offset),
8392          (mul (ldop (add node:$Rn, node:$offset)),
8393               (ldop (add node:$Rm, node:$offset)))>;
8394class mulz_v4i8<SDPatternOperator ldop> :
8395  PatFrag<(ops node:$Rn, node:$Rm),
8396          (mul (ldop node:$Rn), (ldop node:$Rm))>;
8397
8398def load_v4i8 :
8399  OutPatFrag<(ops node:$R),
8400             (INSERT_SUBREG
8401              (v2i32 (IMPLICIT_DEF)),
8402               (i32 (COPY_TO_REGCLASS (LDRWui node:$R, (i64 0)), FPR32)),
8403              ssub)>;
8404
8405class dot_v4i8<Instruction DOT, SDPatternOperator ldop> :
8406  Pat<(i32 (add (mul_v4i8<ldop> GPR64sp:$Rn, GPR64sp:$Rm, (i64 3)),
8407           (add (mul_v4i8<ldop> GPR64sp:$Rn, GPR64sp:$Rm, (i64 2)),
8408           (add (mul_v4i8<ldop> GPR64sp:$Rn, GPR64sp:$Rm, (i64 1)),
8409                (mulz_v4i8<ldop> GPR64sp:$Rn, GPR64sp:$Rm))))),
8410      (EXTRACT_SUBREG (i64 (DOT (DUPv2i32gpr WZR),
8411                                (load_v4i8 GPR64sp:$Rn),
8412                                (load_v4i8 GPR64sp:$Rm))),
8413                      sub_32)>, Requires<[HasDotProd]>;
8414
8415// dot_v8i8
8416class ee_v8i8<SDPatternOperator extend> :
8417  PatFrag<(ops node:$V, node:$K),
8418          (v4i16 (extract_subvector (v8i16 (extend node:$V)), node:$K))>;
8419
8420class mul_v8i8<SDPatternOperator mulop, SDPatternOperator extend> :
8421  PatFrag<(ops node:$M, node:$N, node:$K),
8422          (mulop (v4i16 (ee_v8i8<extend> node:$M, node:$K)),
8423                 (v4i16 (ee_v8i8<extend> node:$N, node:$K)))>;
8424
8425class idot_v8i8<SDPatternOperator mulop, SDPatternOperator extend> :
8426  PatFrag<(ops node:$M, node:$N),
8427          (i32 (extractelt
8428           (v4i32 (AArch64uaddv
8429            (add (mul_v8i8<mulop, extend> node:$M, node:$N, (i64 0)),
8430                 (mul_v8i8<mulop, extend> node:$M, node:$N, (i64 4))))),
8431           (i64 0)))>;
8432
8433// vaddv_[su]32 is special; -> ADDP Vd.2S,Vn.2S,Vm.2S; return Vd.s[0];Vn==Vm
8434def VADDV_32 : OutPatFrag<(ops node:$R), (ADDPv2i32 node:$R, node:$R)>;
8435
8436class odot_v8i8<Instruction DOT> :
8437  OutPatFrag<(ops node:$Vm, node:$Vn),
8438             (EXTRACT_SUBREG
8439              (VADDV_32
8440               (i64 (DOT (DUPv2i32gpr WZR),
8441                         (v8i8 node:$Vm),
8442                         (v8i8 node:$Vn)))),
8443              sub_32)>;
8444
8445class dot_v8i8<Instruction DOT, SDPatternOperator mulop,
8446                    SDPatternOperator extend> :
8447  Pat<(idot_v8i8<mulop, extend> V64:$Vm, V64:$Vn),
8448      (odot_v8i8<DOT> V64:$Vm, V64:$Vn)>,
8449  Requires<[HasDotProd]>;
8450
8451// dot_v16i8
8452class ee_v16i8<SDPatternOperator extend> :
8453  PatFrag<(ops node:$V, node:$K1, node:$K2),
8454          (v4i16 (extract_subvector
8455           (v8i16 (extend
8456            (v8i8 (extract_subvector node:$V, node:$K1)))), node:$K2))>;
8457
8458class mul_v16i8<SDPatternOperator mulop, SDPatternOperator extend> :
8459  PatFrag<(ops node:$M, node:$N, node:$K1, node:$K2),
8460          (v4i32
8461           (mulop (v4i16 (ee_v16i8<extend> node:$M, node:$K1, node:$K2)),
8462                  (v4i16 (ee_v16i8<extend> node:$N, node:$K1, node:$K2))))>;
8463
8464class idot_v16i8<SDPatternOperator m, SDPatternOperator x> :
8465  PatFrag<(ops node:$M, node:$N),
8466          (i32 (extractelt
8467           (v4i32 (AArch64uaddv
8468            (add
8469             (add (mul_v16i8<m, x> node:$M, node:$N, (i64 0), (i64 0)),
8470                  (mul_v16i8<m, x> node:$M, node:$N, (i64 8), (i64 0))),
8471             (add (mul_v16i8<m, x> node:$M, node:$N, (i64 0), (i64 4)),
8472                  (mul_v16i8<m, x> node:$M, node:$N, (i64 8), (i64 4)))))),
8473           (i64 0)))>;
8474
8475class odot_v16i8<Instruction DOT> :
8476  OutPatFrag<(ops node:$Vm, node:$Vn),
8477             (i32 (ADDVv4i32v
8478              (DOT (DUPv4i32gpr WZR), node:$Vm, node:$Vn)))>;
8479
8480class dot_v16i8<Instruction DOT, SDPatternOperator mulop,
8481                SDPatternOperator extend> :
8482  Pat<(idot_v16i8<mulop, extend> V128:$Vm, V128:$Vn),
8483      (odot_v16i8<DOT> V128:$Vm, V128:$Vn)>,
8484  Requires<[HasDotProd]>;
8485
8486let AddedComplexity = 10 in {
8487  def : dot_v4i8<SDOTv8i8, sextloadi8>;
8488  def : dot_v4i8<UDOTv8i8, zextloadi8>;
8489  def : dot_v8i8<SDOTv8i8, AArch64smull, sext>;
8490  def : dot_v8i8<UDOTv8i8, AArch64umull, zext>;
8491  def : dot_v16i8<SDOTv16i8, AArch64smull, sext>;
8492  def : dot_v16i8<UDOTv16i8, AArch64umull, zext>;
8493
8494  // FIXME: add patterns to generate vector by element dot product.
8495  // FIXME: add SVE dot-product patterns.
8496}
8497
8498// Custom DAG nodes and isel rules to make a 64-byte block out of eight GPRs,
8499// so that it can be used as input to inline asm, and vice versa.
8500def LS64_BUILD : SDNode<"AArch64ISD::LS64_BUILD", SDTypeProfile<1, 8, []>>;
8501def LS64_EXTRACT : SDNode<"AArch64ISD::LS64_EXTRACT", SDTypeProfile<1, 2, []>>;
8502def : Pat<(i64x8 (LS64_BUILD GPR64:$x0, GPR64:$x1, GPR64:$x2, GPR64:$x3,
8503                             GPR64:$x4, GPR64:$x5, GPR64:$x6, GPR64:$x7)),
8504          (REG_SEQUENCE GPR64x8Class,
8505              $x0, x8sub_0, $x1, x8sub_1, $x2, x8sub_2, $x3, x8sub_3,
8506              $x4, x8sub_4, $x5, x8sub_5, $x6, x8sub_6, $x7, x8sub_7)>;
8507foreach i = 0-7 in {
8508  def : Pat<(i64 (LS64_EXTRACT (i64x8 GPR64x8:$val), (i32 i))),
8509            (EXTRACT_SUBREG $val, !cast<SubRegIndex>("x8sub_"#i))>;
8510}
8511
8512let Predicates = [HasLS64] in {
8513  def LD64B: LoadStore64B<0b101, "ld64b", (ins GPR64sp:$Rn),
8514                                          (outs GPR64x8:$Rt)>;
8515  def ST64B: LoadStore64B<0b001, "st64b", (ins GPR64x8:$Rt, GPR64sp:$Rn),
8516                                          (outs)>;
8517  def ST64BV:   Store64BV<0b011, "st64bv">;
8518  def ST64BV0:  Store64BV<0b010, "st64bv0">;
8519
8520  class ST64BPattern<Intrinsic intrinsic, Instruction instruction>
8521    : Pat<(intrinsic GPR64sp:$addr, GPR64:$x0, GPR64:$x1, GPR64:$x2, GPR64:$x3, GPR64:$x4, GPR64:$x5, GPR64:$x6, GPR64:$x7),
8522          (instruction (REG_SEQUENCE GPR64x8Class, $x0, x8sub_0, $x1, x8sub_1, $x2, x8sub_2, $x3, x8sub_3, $x4, x8sub_4, $x5, x8sub_5, $x6, x8sub_6, $x7, x8sub_7), $addr)>;
8523
8524  def : ST64BPattern<int_aarch64_st64b, ST64B>;
8525  def : ST64BPattern<int_aarch64_st64bv, ST64BV>;
8526  def : ST64BPattern<int_aarch64_st64bv0, ST64BV0>;
8527}
8528
8529let Predicates = [HasMOPS] in {
8530  let Defs = [NZCV] in {
8531    defm CPYFP : MOPSMemoryCopyInsns<0b00, "cpyfp">;
8532
8533    defm CPYP : MOPSMemoryMoveInsns<0b00, "cpyp">;
8534
8535    defm SETP : MOPSMemorySetInsns<0b00, "setp">;
8536  }
8537  let Uses = [NZCV] in {
8538    defm CPYFM : MOPSMemoryCopyInsns<0b01, "cpyfm">;
8539    defm CPYFE : MOPSMemoryCopyInsns<0b10, "cpyfe">;
8540
8541    defm CPYM : MOPSMemoryMoveInsns<0b01, "cpym">;
8542    defm CPYE : MOPSMemoryMoveInsns<0b10, "cpye">;
8543
8544    defm SETM : MOPSMemorySetInsns<0b01, "setm">;
8545    defm SETE : MOPSMemorySetInsns<0b10, "sete">;
8546  }
8547}
8548let Predicates = [HasMOPS, HasMTE] in {
8549  let Defs = [NZCV] in {
8550    defm SETGP     : MOPSMemorySetTaggingInsns<0b00, "setgp">;
8551  }
8552  let Uses = [NZCV] in {
8553    defm SETGM     : MOPSMemorySetTaggingInsns<0b01, "setgm">;
8554    // Can't use SETGE because it's a reserved name in TargetSelectionDAG.td
8555    defm MOPSSETGE : MOPSMemorySetTaggingInsns<0b10, "setge">;
8556  }
8557}
8558
8559// MOPS Node operands: 0: Dst, 1: Src or Value, 2: Size, 3: Chain
8560// MOPS Node results: 0: Dst writeback, 1: Size writeback, 2: Chain
8561def SDT_AArch64mops : SDTypeProfile<2, 3, [ SDTCisInt<0>, SDTCisInt<1>, SDTCisInt<2> ]>;
8562def AArch64mops_memset : SDNode<"AArch64ISD::MOPS_MEMSET", SDT_AArch64mops>;
8563def AArch64mops_memset_tagging : SDNode<"AArch64ISD::MOPS_MEMSET_TAGGING", SDT_AArch64mops>;
8564def AArch64mops_memcopy : SDNode<"AArch64ISD::MOPS_MEMCOPY", SDT_AArch64mops>;
8565def AArch64mops_memmove : SDNode<"AArch64ISD::MOPS_MEMMOVE", SDT_AArch64mops>;
8566
8567// MOPS operations always contain three 4-byte instructions
8568let Predicates = [HasMOPS], Defs = [NZCV], Size = 12, mayStore = 1 in {
8569  let mayLoad = 1 in {
8570    def MOPSMemoryCopyPseudo : Pseudo<(outs GPR64common:$Rd_wb, GPR64common:$Rs_wb, GPR64:$Rn_wb),
8571                                      (ins GPR64common:$Rd, GPR64common:$Rs, GPR64:$Rn),
8572                                      [], "$Rd = $Rd_wb,$Rs = $Rs_wb,$Rn = $Rn_wb">, Sched<[]>;
8573    def MOPSMemoryMovePseudo : Pseudo<(outs GPR64common:$Rd_wb, GPR64common:$Rs_wb, GPR64:$Rn_wb),
8574                                      (ins GPR64common:$Rd, GPR64common:$Rs, GPR64:$Rn),
8575                                      [], "$Rd = $Rd_wb,$Rs = $Rs_wb,$Rn = $Rn_wb">, Sched<[]>;
8576  }
8577  let mayLoad = 0 in {
8578    def MOPSMemorySetPseudo  : Pseudo<(outs GPR64common:$Rd_wb, GPR64:$Rn_wb),
8579                                      (ins GPR64common:$Rd, GPR64:$Rn, GPR64:$Rm),
8580                                      [], "$Rd = $Rd_wb,$Rn = $Rn_wb">, Sched<[]>;
8581  }
8582}
8583let Predicates = [HasMOPS, HasMTE], Defs = [NZCV], Size = 12, mayLoad = 0, mayStore = 1 in {
8584  def MOPSMemorySetTaggingPseudo : Pseudo<(outs GPR64common:$Rd_wb, GPR64:$Rn_wb),
8585                                          (ins GPR64common:$Rd, GPR64:$Rn, GPR64:$Rm),
8586                                          [], "$Rd = $Rd_wb,$Rn = $Rn_wb">, Sched<[]>;
8587}
8588
8589// This gets lowered into an instruction sequence of 20 bytes
8590let Defs = [X16, X17], mayStore = 1, isCodeGenOnly = 1, Size = 20 in
8591def StoreSwiftAsyncContext
8592      : Pseudo<(outs), (ins GPR64:$ctx, GPR64sp:$base, simm9:$offset),
8593               []>, Sched<[]>;
8594
8595def AArch64AssertZExtBool : SDNode<"AArch64ISD::ASSERT_ZEXT_BOOL", SDT_assert>;
8596def : Pat<(AArch64AssertZExtBool GPR32:$op),
8597          (i32 GPR32:$op)>;
8598
8599//===----------------------------===//
8600// 2022 Architecture Extensions:
8601//===----------------------------===//
8602
8603def : InstAlias<"clrbhb",  (HINT 22), 0>;
8604let Predicates = [HasCLRBHB] in {
8605  def : InstAlias<"clrbhb",  (HINT 22), 1>;
8606}
8607
8608//===----------------------------------------------------------------------===//
8609// Translation Hardening Extension (FEAT_THE)
8610//===----------------------------------------------------------------------===//
8611defm RCW     : ReadCheckWriteCompareAndSwap;
8612
8613defm RCWCLR  : ReadCheckWriteOperation<0b001, "clr">;
8614defm RCWSET  : ReadCheckWriteOperation<0b011, "set">;
8615defm RCWSWP  : ReadCheckWriteOperation<0b010, "swp">;
8616
8617//===----------------------------------------------------------------------===//
8618// General Data-Processing Instructions (FEAT_V94_DP)
8619//===----------------------------------------------------------------------===//
8620defm ABS : OneOperandData<0b001000, "abs", abs>, Requires<[HasCSSC]>;
8621defm CNT : OneOperandData<0b000111, "cnt", ctpop>, Requires<[HasCSSC]>;
8622defm CTZ : OneOperandData<0b000110, "ctz", cttz>, Requires<[HasCSSC]>;
8623
8624defm SMAX : ComparisonOp<0, 0, "smax", smax>, Requires<[HasCSSC]>;
8625defm SMIN : ComparisonOp<0, 1, "smin", smin>, Requires<[HasCSSC]>;
8626defm UMAX : ComparisonOp<1, 0, "umax", umax>, Requires<[HasCSSC]>;
8627defm UMIN : ComparisonOp<1, 1, "umin", umin>, Requires<[HasCSSC]>;
8628
8629def RPRFM:
8630    I<(outs), (ins rprfop:$Rt, GPR64:$Rm, GPR64sp:$Rn),
8631      "rprfm", "\t$Rt, $Rm, [$Rn]", "", []>,
8632    Sched<[]> {
8633  bits<6> Rt;
8634  bits<5> Rn;
8635  bits<5> Rm;
8636  let Inst{2-0} = Rt{2-0};
8637  let Inst{4-3} = 0b11;
8638  let Inst{9-5} = Rn;
8639  let Inst{11-10} = 0b10;
8640  let Inst{13-12} = Rt{4-3};
8641  let Inst{14} = 0b1;
8642  let Inst{15} = Rt{5};
8643  let Inst{20-16} = Rm;
8644  let Inst{31-21} = 0b11111000101;
8645  let mayLoad = 0;
8646  let mayStore = 0;
8647  let hasSideEffects = 1;
8648  // RPRFM overlaps with PRFM (reg), when the decoder method of PRFM returns
8649  // Fail, the decoder should attempt to decode RPRFM. This requires setting
8650  // the decoder namespace to "Fallback".
8651  let DecoderNamespace = "Fallback";
8652}
8653
8654//===----------------------------------------------------------------------===//
8655// 128-bit Atomics (FEAT_LSE128)
8656//===----------------------------------------------------------------------===//
8657let Predicates = [HasLSE128] in {
8658  def SWPP     : LSE128Base<0b000, 0b00, 0b1, "swpp">;
8659  def SWPPA    : LSE128Base<0b000, 0b10, 0b1, "swppa">;
8660  def SWPPAL   : LSE128Base<0b000, 0b11, 0b1, "swppal">;
8661  def SWPPL    : LSE128Base<0b000, 0b01, 0b1, "swppl">;
8662  def LDCLRP   : LSE128Base<0b001, 0b00, 0b0, "ldclrp">;
8663  def LDCLRPA  : LSE128Base<0b001, 0b10, 0b0, "ldclrpa">;
8664  def LDCLRPAL : LSE128Base<0b001, 0b11, 0b0, "ldclrpal">;
8665  def LDCLRPL  : LSE128Base<0b001, 0b01, 0b0, "ldclrpl">;
8666  def LDSETP   : LSE128Base<0b011, 0b00, 0b0, "ldsetp">;
8667  def LDSETPA  : LSE128Base<0b011, 0b10, 0b0, "ldsetpa">;
8668  def LDSETPAL : LSE128Base<0b011, 0b11, 0b0, "ldsetpal">;
8669  def LDSETPL  : LSE128Base<0b011, 0b01, 0b0, "ldsetpl">;
8670}
8671
8672//===----------------------------------------------------------------------===//
8673// RCPC Instructions (FEAT_LRCPC3)
8674//===----------------------------------------------------------------------===//
8675
8676let Predicates = [HasRCPC3] in {
8677  //                                             size   opc    opc2
8678  def STILPWpre:  BaseLRCPC3IntegerLoadStorePair<0b10, 0b00, 0b0000, (outs GPR64sp:$wback), (ins GPR32:$Rt, GPR32:$Rt2, GPR64sp:$Rn), "stilp", "\t$Rt, $Rt2, [$Rn, #-8]!", "$Rn = $wback">;
8679  def STILPXpre:  BaseLRCPC3IntegerLoadStorePair<0b11, 0b00, 0b0000, (outs GPR64sp:$wback), (ins GPR64:$Rt, GPR64:$Rt2, GPR64sp:$Rn), "stilp", "\t$Rt, $Rt2, [$Rn, #-16]!", "$Rn = $wback">;
8680  def STILPW:     BaseLRCPC3IntegerLoadStorePair<0b10, 0b00, 0b0001, (outs), (ins GPR32:$Rt, GPR32:$Rt2, GPR64sp:$Rn), "stilp", "\t$Rt, $Rt2, [$Rn]", "">;
8681  def STILPX:     BaseLRCPC3IntegerLoadStorePair<0b11, 0b00, 0b0001, (outs), (ins GPR64:$Rt, GPR64:$Rt2, GPR64sp:$Rn), "stilp", "\t$Rt, $Rt2, [$Rn]", "">;
8682  def LDIAPPWpre: BaseLRCPC3IntegerLoadStorePair<0b10, 0b01, 0b0000, (outs GPR64sp:$wback, GPR32:$Rt, GPR32:$Rt2), (ins GPR64sp:$Rn), "ldiapp", "\t$Rt, $Rt2, [$Rn], #8", "$Rn = $wback">;
8683  def LDIAPPXpre: BaseLRCPC3IntegerLoadStorePair<0b11, 0b01, 0b0000, (outs GPR64sp:$wback, GPR64:$Rt, GPR64:$Rt2), (ins GPR64sp:$Rn), "ldiapp", "\t$Rt, $Rt2, [$Rn], #16", "$Rn = $wback">;
8684  def LDIAPPW:    BaseLRCPC3IntegerLoadStorePair<0b10, 0b01, 0b0001, (outs GPR32:$Rt, GPR32:$Rt2), (ins GPR64sp0:$Rn), "ldiapp", "\t$Rt, $Rt2, [$Rn]", "">;
8685  def LDIAPPX:    BaseLRCPC3IntegerLoadStorePair<0b11, 0b01, 0b0001, (outs GPR64:$Rt, GPR64:$Rt2), (ins GPR64sp0:$Rn), "ldiapp", "\t$Rt, $Rt2, [$Rn]", "">;
8686
8687  // Aliases for when offset=0
8688  def : InstAlias<"stilp\t$Rt, $Rt2, [$Rn, #0]", (STILPW GPR32: $Rt, GPR32: $Rt2, GPR64sp:$Rn)>;
8689  def : InstAlias<"stilp\t$Rt, $Rt2, [$Rn, #0]", (STILPX GPR64: $Rt, GPR64: $Rt2, GPR64sp:$Rn)>;
8690
8691  //                                        size   opc
8692  def STLRWpre:  BaseLRCPC3IntegerLoadStore<0b10, 0b10, (outs GPR64sp:$wback),            (ins GPR32:$Rt, GPR64sp:$Rn), "stlr",  "\t$Rt, [$Rn, #-4]!", "$Rn = $wback">;
8693  def STLRXpre:  BaseLRCPC3IntegerLoadStore<0b11, 0b10, (outs GPR64sp:$wback),            (ins GPR64:$Rt, GPR64sp:$Rn), "stlr",  "\t$Rt, [$Rn, #-8]!", "$Rn = $wback">;
8694  def LDAPRWpre: BaseLRCPC3IntegerLoadStore<0b10, 0b11, (outs GPR64sp:$wback, GPR32:$Rt), (ins GPR64sp:$Rn),            "ldapr", "\t$Rt, [$Rn], #4",   "$Rn = $wback">;
8695  def LDAPRXpre: BaseLRCPC3IntegerLoadStore<0b11, 0b11, (outs GPR64sp:$wback, GPR64:$Rt), (ins GPR64sp:$Rn),            "ldapr", "\t$Rt, [$Rn], #8",   "$Rn = $wback">;
8696}
8697
8698let Predicates = [HasRCPC3, HasNEON] in {
8699  //                                              size   opc regtype
8700  defm STLURb:  LRCPC3NEONLoadStoreUnscaledOffset<0b00, 0b00, FPR8  , (outs), (ins FPR8  :$Rt, GPR64sp:$Rn, simm9:$simm), "stlur">;
8701  defm STLURh:  LRCPC3NEONLoadStoreUnscaledOffset<0b01, 0b00, FPR16 , (outs), (ins FPR16 :$Rt, GPR64sp:$Rn, simm9:$simm), "stlur">;
8702  defm STLURs:  LRCPC3NEONLoadStoreUnscaledOffset<0b10, 0b00, FPR32 , (outs), (ins FPR32 :$Rt, GPR64sp:$Rn, simm9:$simm), "stlur">;
8703  defm STLURd:  LRCPC3NEONLoadStoreUnscaledOffset<0b11, 0b00, FPR64 , (outs), (ins FPR64 :$Rt, GPR64sp:$Rn, simm9:$simm), "stlur">;
8704  defm STLURq:  LRCPC3NEONLoadStoreUnscaledOffset<0b00, 0b10, FPR128, (outs), (ins FPR128:$Rt, GPR64sp:$Rn, simm9:$simm), "stlur">;
8705  defm LDAPURb: LRCPC3NEONLoadStoreUnscaledOffset<0b00, 0b01, FPR8  , (outs FPR8  :$Rt), (ins GPR64sp:$Rn, simm9:$simm), "ldapur">;
8706  defm LDAPURh: LRCPC3NEONLoadStoreUnscaledOffset<0b01, 0b01, FPR16 , (outs FPR16 :$Rt), (ins GPR64sp:$Rn, simm9:$simm), "ldapur">;
8707  defm LDAPURs: LRCPC3NEONLoadStoreUnscaledOffset<0b10, 0b01, FPR32 , (outs FPR32 :$Rt), (ins GPR64sp:$Rn, simm9:$simm), "ldapur">;
8708  defm LDAPURd: LRCPC3NEONLoadStoreUnscaledOffset<0b11, 0b01, FPR64 , (outs FPR64 :$Rt), (ins GPR64sp:$Rn, simm9:$simm), "ldapur">;
8709  defm LDAPURq: LRCPC3NEONLoadStoreUnscaledOffset<0b00, 0b11, FPR128, (outs FPR128:$Rt), (ins GPR64sp:$Rn, simm9:$simm), "ldapur">;
8710
8711  //                                L
8712  def STL1:  LRCPC3NEONLdStSingle<0b0, (outs), (ins VecListOned:$Vt, VectorIndexD:$Q, GPR64sp:$Rn) , "stl1", "">;
8713  def LDAP1: LRCPC3NEONLdStSingle<0b1, (outs VecListOned:$dst), (ins VecListOned:$Vt, VectorIndexD:$Q, GPR64sp0:$Rn), "ldap1", "$Vt = $dst">;
8714
8715  // Aliases for when offset=0
8716  def : InstAlias<"stl1\t$Vt$Q, [$Rn, #0]", (STL1 VecListOned:$Vt, VectorIndexD:$Q, GPR64sp:$Rn)>;
8717}
8718
8719//===----------------------------------------------------------------------===//
8720// 128-bit System Instructions (FEAT_SYSINSTR128)
8721//===----------------------------------------------------------------------===//
8722let Predicates = [HasD128] in {
8723  def SYSPxt  : SystemPXtI<0, "sysp">;
8724
8725  def SYSPxt_XZR
8726    : BaseSystemI<0, (outs),
8727        (ins imm0_7:$op1, sys_cr_op:$Cn, sys_cr_op:$Cm, imm0_7:$op2, SyspXzrPairOperand:$xzr_pair),
8728        "sysp", "\t$op1, $Cn, $Cm, $op2, $xzr_pair">,
8729      Sched<[WriteSys]>
8730  {
8731    // Had to use a custom decoder because tablegen interprets this as having 4 fields (why?)
8732    // and therefore autogenerates a decoder that builds an MC representation that has 4 fields
8733    // (decodeToMCInst), but when printing we expect the MC representation to have 5 fields (one
8734    // extra for the XZR) because AArch64InstPrinter::printInstruction in AArch64GenAsmWriter.inc
8735    // is based off of the asm template (maybe) and therefore wants to print 5 operands.
8736    // I could add a bits<5> xzr_pair. But without a way to constrain it to 0b11111 here it would
8737    // overlap with the main SYSP instruction.
8738    let DecoderMethod = "DecodeSyspXzrInstruction";
8739    bits<3> op1;
8740    bits<4> Cn;
8741    bits<4> Cm;
8742    bits<3> op2;
8743    let Inst{22}    = 0b1; // override BaseSystemI
8744    let Inst{20-19} = 0b01;
8745    let Inst{18-16} = op1;
8746    let Inst{15-12} = Cn;
8747    let Inst{11-8}  = Cm;
8748    let Inst{7-5}   = op2;
8749    let Inst{4-0}   = 0b11111;
8750  }
8751
8752  def : InstAlias<"sysp $op1, $Cn, $Cm, $op2",
8753                  (SYSPxt_XZR imm0_7:$op1, sys_cr_op:$Cn, sys_cr_op:$Cm, imm0_7:$op2, XZR)>;
8754}
8755
8756//---
8757// 128-bit System Registers (FEAT_SYSREG128)
8758//---
8759
8760// Instruction encoding:
8761//
8762//          31       22|21|20|19|18 16|15 12|11 8|7 5|4 0
8763// MRRS      1101010101| 1| 1|o0|  op1|   Cn|  Cm|op2| Rt
8764// MSRR      1101010101| 0| 1|o0|  op1|   Cn|  Cm|op2| Rt
8765
8766// Instruction syntax:
8767//
8768// MRRS <Xt>, <Xt+1>, <sysreg|S<op0>_<op1>_<Cn>_<Cm>_<op2>>
8769// MSRR <sysreg|S<op0>_<op1>_<Cn>_<Cm>_<op2>>, <Xt>, <Xt+1>
8770//
8771// ...where t is even (X0, X2, etc).
8772
8773let Predicates = [HasD128] in {
8774  def MRRS : RtSystemI128<1,
8775    (outs MrrsMssrPairClassOperand:$Rt), (ins mrs_sysreg_op:$systemreg),
8776    "mrrs", "\t$Rt, $systemreg">
8777  {
8778    bits<16> systemreg;
8779    let Inst{20-5} = systemreg;
8780  }
8781
8782  def MSRR : RtSystemI128<0,
8783    (outs),  (ins msr_sysreg_op:$systemreg, MrrsMssrPairClassOperand:$Rt),
8784    "msrr", "\t$systemreg, $Rt">
8785  {
8786    bits<16> systemreg;
8787    let Inst{20-5} = systemreg;
8788  }
8789}
8790
8791
8792include "AArch64InstrAtomics.td"
8793include "AArch64SVEInstrInfo.td"
8794include "AArch64SMEInstrInfo.td"
8795include "AArch64InstrGISel.td"
8796