xref: /freebsd/contrib/llvm-project/llvm/lib/Target/AArch64/AArch64InstrInfo.td (revision 1db9f3b21e39176dd5b67cf8ac378633b172463e)
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 HasPAuthLR       : Predicate<"Subtarget->hasPAuthLR()">,
65                       AssemblerPredicateWithAll<(all_of FeaturePAuthLR), "pauth-lr">;
66
67def HasJS            : Predicate<"Subtarget->hasJS()">,
68                       AssemblerPredicateWithAll<(all_of FeatureJS), "jsconv">;
69
70def HasCCIDX         : Predicate<"Subtarget->hasCCIDX()">,
71                       AssemblerPredicateWithAll<(all_of FeatureCCIDX), "ccidx">;
72
73def HasComplxNum      : Predicate<"Subtarget->hasComplxNum()">,
74                       AssemblerPredicateWithAll<(all_of FeatureComplxNum), "complxnum">;
75
76def HasNV            : Predicate<"Subtarget->hasNV()">,
77                       AssemblerPredicateWithAll<(all_of FeatureNV), "nv">;
78
79def HasMPAM          : Predicate<"Subtarget->hasMPAM()">,
80                       AssemblerPredicateWithAll<(all_of FeatureMPAM), "mpam">;
81
82def HasDIT           : Predicate<"Subtarget->hasDIT()">,
83                       AssemblerPredicateWithAll<(all_of FeatureDIT), "dit">;
84
85def HasTRACEV8_4         : Predicate<"Subtarget->hasTRACEV8_4()">,
86                       AssemblerPredicateWithAll<(all_of FeatureTRACEV8_4), "tracev8.4">;
87
88def HasAM            : Predicate<"Subtarget->hasAM()">,
89                       AssemblerPredicateWithAll<(all_of FeatureAM), "am">;
90
91def HasSEL2          : Predicate<"Subtarget->hasSEL2()">,
92                       AssemblerPredicateWithAll<(all_of FeatureSEL2), "sel2">;
93
94def HasTLB_RMI          : Predicate<"Subtarget->hasTLB_RMI()">,
95                       AssemblerPredicateWithAll<(all_of FeatureTLB_RMI), "tlb-rmi">;
96
97def HasFlagM         : Predicate<"Subtarget->hasFlagM()">,
98                       AssemblerPredicateWithAll<(all_of FeatureFlagM), "flagm">;
99
100def HasRCPC_IMMO      : Predicate<"Subtarget->hasRCPC_IMMO()">,
101                       AssemblerPredicateWithAll<(all_of FeatureRCPC_IMMO), "rcpc-immo">;
102
103def HasFPARMv8       : Predicate<"Subtarget->hasFPARMv8()">,
104                               AssemblerPredicateWithAll<(all_of FeatureFPARMv8), "fp-armv8">;
105def HasNEON          : Predicate<"Subtarget->hasNEON()">,
106                                 AssemblerPredicateWithAll<(all_of FeatureNEON), "neon">;
107def HasSM4           : Predicate<"Subtarget->hasSM4()">,
108                                 AssemblerPredicateWithAll<(all_of FeatureSM4), "sm4">;
109def HasSHA3          : Predicate<"Subtarget->hasSHA3()">,
110                                 AssemblerPredicateWithAll<(all_of FeatureSHA3), "sha3">;
111def HasSHA2          : Predicate<"Subtarget->hasSHA2()">,
112                                 AssemblerPredicateWithAll<(all_of FeatureSHA2), "sha2">;
113def HasAES           : Predicate<"Subtarget->hasAES()">,
114                                 AssemblerPredicateWithAll<(all_of FeatureAES), "aes">;
115def HasDotProd       : Predicate<"Subtarget->hasDotProd()">,
116                                 AssemblerPredicateWithAll<(all_of FeatureDotProd), "dotprod">;
117def HasCRC           : Predicate<"Subtarget->hasCRC()">,
118                                 AssemblerPredicateWithAll<(all_of FeatureCRC), "crc">;
119def HasCSSC          : Predicate<"Subtarget->hasCSSC()">,
120                                 AssemblerPredicateWithAll<(all_of FeatureCSSC), "cssc">;
121def HasNoCSSC        : Predicate<"!Subtarget->hasCSSC()">;
122def HasLSE           : Predicate<"Subtarget->hasLSE()">,
123                                 AssemblerPredicateWithAll<(all_of FeatureLSE), "lse">;
124def HasNoLSE         : Predicate<"!Subtarget->hasLSE()">;
125def HasRAS           : Predicate<"Subtarget->hasRAS()">,
126                                 AssemblerPredicateWithAll<(all_of FeatureRAS), "ras">;
127def HasRDM           : Predicate<"Subtarget->hasRDM()">,
128                                 AssemblerPredicateWithAll<(all_of FeatureRDM), "rdm">;
129def HasFullFP16      : Predicate<"Subtarget->hasFullFP16()">,
130                                 AssemblerPredicateWithAll<(all_of FeatureFullFP16), "fullfp16">;
131def HasFP16FML       : Predicate<"Subtarget->hasFP16FML()">,
132                                 AssemblerPredicateWithAll<(all_of FeatureFP16FML), "fp16fml">;
133def HasSPE           : Predicate<"Subtarget->hasSPE()">,
134                                 AssemblerPredicateWithAll<(all_of FeatureSPE), "spe">;
135def HasFuseAES       : Predicate<"Subtarget->hasFuseAES()">,
136                                 AssemblerPredicateWithAll<(all_of FeatureFuseAES),
137                                 "fuse-aes">;
138def HasSVE           : Predicate<"Subtarget->hasSVE()">,
139                                 AssemblerPredicateWithAll<(all_of FeatureSVE), "sve">;
140def HasSVE2          : Predicate<"Subtarget->hasSVE2()">,
141                                 AssemblerPredicateWithAll<(all_of FeatureSVE2), "sve2">;
142def HasSVE2p1        : Predicate<"Subtarget->hasSVE2p1()">,
143                                 AssemblerPredicateWithAll<(all_of FeatureSVE2p1), "sve2p1">;
144def HasSVE2AES       : Predicate<"Subtarget->hasSVE2AES()">,
145                                 AssemblerPredicateWithAll<(all_of FeatureSVE2AES), "sve2-aes">;
146def HasSVE2SM4       : Predicate<"Subtarget->hasSVE2SM4()">,
147                                 AssemblerPredicateWithAll<(all_of FeatureSVE2SM4), "sve2-sm4">;
148def HasSVE2SHA3      : Predicate<"Subtarget->hasSVE2SHA3()">,
149                                 AssemblerPredicateWithAll<(all_of FeatureSVE2SHA3), "sve2-sha3">;
150def HasSVE2BitPerm   : Predicate<"Subtarget->hasSVE2BitPerm()">,
151                                 AssemblerPredicateWithAll<(all_of FeatureSVE2BitPerm), "sve2-bitperm">;
152def HasB16B16        : Predicate<"Subtarget->hasB16B16()">,
153                                 AssemblerPredicateWithAll<(all_of FeatureB16B16), "b16b16">;
154def HasSME           : Predicate<"Subtarget->hasSME()">,
155                                 AssemblerPredicateWithAll<(all_of FeatureSME), "sme">;
156def HasSMEF64F64     : Predicate<"Subtarget->hasSMEF64F64()">,
157                                 AssemblerPredicateWithAll<(all_of FeatureSMEF64F64), "sme-f64f64">;
158def HasSMEF16F16     : Predicate<"Subtarget->hasSMEF16F16()">,
159                                 AssemblerPredicateWithAll<(all_of FeatureSMEF16F16), "sme-f16f16">;
160def HasSMEFA64       : Predicate<"Subtarget->hasSMEFA64()">,
161                                 AssemblerPredicateWithAll<(all_of FeatureSMEFA64), "sme-fa64">;
162def HasSMEI16I64     : Predicate<"Subtarget->hasSMEI16I64()">,
163                                 AssemblerPredicateWithAll<(all_of FeatureSMEI16I64), "sme-i16i64">;
164def HasSME2          : Predicate<"Subtarget->hasSME2()">,
165                                 AssemblerPredicateWithAll<(all_of FeatureSME2), "sme2">;
166def HasSME2p1        : Predicate<"Subtarget->hasSME2p1()">,
167                                 AssemblerPredicateWithAll<(all_of FeatureSME2p1), "sme2p1">;
168def HasFPMR          : Predicate<"Subtarget->hasFPMR()">,
169                                 AssemblerPredicateWithAll<(all_of FeatureFPMR), "fpmr">;
170def HasFP8           : Predicate<"Subtarget->hasFP8()">,
171                                 AssemblerPredicateWithAll<(all_of FeatureFP8), "fp8">;
172def HasFAMINMAX      : Predicate<"Subtarget->hasFAMINMAX()">,
173                                 AssemblerPredicateWithAll<(all_of FeatureFAMINMAX), "faminmax">;
174def HasFP8FMA        : Predicate<"Subtarget->hasFP8FMA()">,
175                                 AssemblerPredicateWithAll<(all_of FeatureFP8FMA), "fp8fma">;
176def HasSSVE_FP8FMA   : Predicate<"Subtarget->SSVE_FP8FMA() || "
177                                 "(Subtarget->hasSVE2() && Subtarget->hasFP8FMA())">,
178                                 AssemblerPredicateWithAll<(any_of FeatureSSVE_FP8FMA,
179                                                           (all_of FeatureSVE2, FeatureFP8FMA)),
180                                                           "ssve-fp8fma or (sve2 and fp8fma)">;
181def HasFP8DOT2       : Predicate<"Subtarget->hasFP8DOT2()">,
182                                 AssemblerPredicateWithAll<(all_of FeatureFP8DOT2), "fp8dot2">;
183def HasSSVE_FP8DOT2  : Predicate<"Subtarget->hasSSVE_FP8DOT2() || "
184                                 "(Subtarget->hasSVE2() && Subtarget->hasFP8DOT2())">,
185                                 AssemblerPredicateWithAll<(any_of FeatureSSVE_FP8DOT2,
186                                                           (all_of FeatureSVE2, FeatureFP8DOT2)),
187                                "ssve-fp8dot2 or (sve2 and fp8dot2)">;
188def HasFP8DOT4       : Predicate<"Subtarget->hasFP8DOT4()">,
189                                 AssemblerPredicateWithAll<(all_of FeatureFP8DOT4), "fp8dot4">;
190def HasSSVE_FP8DOT4  : Predicate<"Subtarget->hasSSVE_FP8DOT4() || "
191                                 "(Subtarget->hasSVE2() && Subtarget->hasFP8DOT4())">,
192                                 AssemblerPredicateWithAll<(any_of FeatureSSVE_FP8DOT4,
193                                                           (all_of FeatureSVE2, FeatureFP8DOT4)),
194                                 "ssve-fp8dot4 or (sve2 and fp8dot4)">;
195def HasLUT          : Predicate<"Subtarget->hasLUT()">,
196                                 AssemblerPredicateWithAll<(all_of FeatureLUT), "lut">;
197def HasSME_LUTv2     : Predicate<"Subtarget->hasSME_LUTv2()">,
198                                 AssemblerPredicateWithAll<(all_of FeatureSME_LUTv2), "sme-lutv2">;
199def HasSMEF8F16     : Predicate<"Subtarget->hasSMEF8F16()">,
200                                 AssemblerPredicateWithAll<(all_of FeatureSMEF8F16), "sme-f8f16">;
201def HasSMEF8F32     : Predicate<"Subtarget->hasSMEF8F32()">,
202                                 AssemblerPredicateWithAll<(all_of FeatureSMEF8F32), "sme-f8f32">;
203
204// A subset of SVE(2) instructions are legal in Streaming SVE execution mode,
205// they should be enabled if either has been specified.
206def HasSVEorSME
207    : Predicate<"Subtarget->hasSVEorSME()">,
208                AssemblerPredicateWithAll<(any_of FeatureSVE, FeatureSME),
209                "sve or sme">;
210def HasSVE2orSME
211    : Predicate<"Subtarget->hasSVE2() || Subtarget->hasSME()">,
212                AssemblerPredicateWithAll<(any_of FeatureSVE2, FeatureSME),
213                "sve2 or sme">;
214def HasSVE2orSME2
215    : Predicate<"Subtarget->hasSVE2() || Subtarget->hasSME2()">,
216                AssemblerPredicateWithAll<(any_of FeatureSVE2, FeatureSME2),
217                "sve2 or sme2">;
218def HasSVE2p1_or_HasSME
219    : Predicate<"Subtarget->hasSVE2p1() || Subtarget->hasSME()">,
220                 AssemblerPredicateWithAll<(any_of FeatureSME, FeatureSVE2p1), "sme or sve2p1">;
221def HasSVE2p1_or_HasSME2
222    : Predicate<"Subtarget->hasSVE2p1() || Subtarget->hasSME2()">,
223                 AssemblerPredicateWithAll<(any_of FeatureSME2, FeatureSVE2p1), "sme2 or sve2p1">;
224def HasSVE2p1_or_HasSME2p1
225    : Predicate<"Subtarget->hasSVE2p1() || Subtarget->hasSME2p1()">,
226                 AssemblerPredicateWithAll<(any_of FeatureSME2p1, FeatureSVE2p1), "sme2p1 or sve2p1">;
227// A subset of NEON instructions are legal in Streaming SVE execution mode,
228// they should be enabled if either has been specified.
229def HasNEONorSME
230    : Predicate<"Subtarget->hasNEON() || Subtarget->hasSME()">,
231                AssemblerPredicateWithAll<(any_of FeatureNEON, FeatureSME),
232                "neon or sme">;
233def HasRCPC          : Predicate<"Subtarget->hasRCPC()">,
234                                 AssemblerPredicateWithAll<(all_of FeatureRCPC), "rcpc">;
235def HasAltNZCV       : Predicate<"Subtarget->hasAlternativeNZCV()">,
236                       AssemblerPredicateWithAll<(all_of FeatureAltFPCmp), "altnzcv">;
237def HasFRInt3264     : Predicate<"Subtarget->hasFRInt3264()">,
238                       AssemblerPredicateWithAll<(all_of FeatureFRInt3264), "frint3264">;
239def HasSB            : Predicate<"Subtarget->hasSB()">,
240                       AssemblerPredicateWithAll<(all_of FeatureSB), "sb">;
241def HasPredRes      : Predicate<"Subtarget->hasPredRes()">,
242                       AssemblerPredicateWithAll<(all_of FeaturePredRes), "predres">;
243def HasCCDP          : Predicate<"Subtarget->hasCCDP()">,
244                       AssemblerPredicateWithAll<(all_of FeatureCacheDeepPersist), "ccdp">;
245def HasBTI           : Predicate<"Subtarget->hasBTI()">,
246                       AssemblerPredicateWithAll<(all_of FeatureBranchTargetId), "bti">;
247def HasMTE           : Predicate<"Subtarget->hasMTE()">,
248                       AssemblerPredicateWithAll<(all_of FeatureMTE), "mte">;
249def HasTME           : Predicate<"Subtarget->hasTME()">,
250                       AssemblerPredicateWithAll<(all_of FeatureTME), "tme">;
251def HasETE           : Predicate<"Subtarget->hasETE()">,
252                       AssemblerPredicateWithAll<(all_of FeatureETE), "ete">;
253def HasTRBE          : Predicate<"Subtarget->hasTRBE()">,
254                       AssemblerPredicateWithAll<(all_of FeatureTRBE), "trbe">;
255def HasBF16          : Predicate<"Subtarget->hasBF16()">,
256                       AssemblerPredicateWithAll<(all_of FeatureBF16), "bf16">;
257def HasMatMulInt8    : Predicate<"Subtarget->hasMatMulInt8()">,
258                       AssemblerPredicateWithAll<(all_of FeatureMatMulInt8), "i8mm">;
259def HasMatMulFP32    : Predicate<"Subtarget->hasMatMulFP32()">,
260                       AssemblerPredicateWithAll<(all_of FeatureMatMulFP32), "f32mm">;
261def HasMatMulFP64    : Predicate<"Subtarget->hasMatMulFP64()">,
262                       AssemblerPredicateWithAll<(all_of FeatureMatMulFP64), "f64mm">;
263def HasXS            : Predicate<"Subtarget->hasXS()">,
264                       AssemblerPredicateWithAll<(all_of FeatureXS), "xs">;
265def HasWFxT          : Predicate<"Subtarget->hasWFxT()">,
266                       AssemblerPredicateWithAll<(all_of FeatureWFxT), "wfxt">;
267def HasLS64          : Predicate<"Subtarget->hasLS64()">,
268                       AssemblerPredicateWithAll<(all_of FeatureLS64), "ls64">;
269def HasBRBE          : Predicate<"Subtarget->hasBRBE()">,
270                       AssemblerPredicateWithAll<(all_of FeatureBRBE), "brbe">;
271def HasSPE_EEF       : Predicate<"Subtarget->hasSPE_EEF()">,
272                       AssemblerPredicateWithAll<(all_of FeatureSPE_EEF), "spe-eef">;
273def HasHBC           : Predicate<"Subtarget->hasHBC()">,
274                       AssemblerPredicateWithAll<(all_of FeatureHBC), "hbc">;
275def HasMOPS          : Predicate<"Subtarget->hasMOPS()">,
276                       AssemblerPredicateWithAll<(all_of FeatureMOPS), "mops">;
277def HasCLRBHB        : Predicate<"Subtarget->hasCLRBHB()">,
278                       AssemblerPredicateWithAll<(all_of FeatureCLRBHB), "clrbhb">;
279def HasSPECRES2      : Predicate<"Subtarget->hasSPECRES2()">,
280                       AssemblerPredicateWithAll<(all_of FeatureSPECRES2), "specres2">;
281def HasITE           : Predicate<"Subtarget->hasITE()">,
282                       AssemblerPredicateWithAll<(all_of FeatureITE), "ite">;
283def HasTHE           : Predicate<"Subtarget->hasTHE()">,
284                       AssemblerPredicateWithAll<(all_of FeatureTHE), "the">;
285def HasRCPC3         : Predicate<"Subtarget->hasRCPC3()">,
286                       AssemblerPredicateWithAll<(all_of FeatureRCPC3), "rcpc3">;
287def HasLSE128        : Predicate<"Subtarget->hasLSE128()">,
288                       AssemblerPredicateWithAll<(all_of FeatureLSE128), "lse128">;
289def HasD128          : Predicate<"Subtarget->hasD128()">,
290                       AssemblerPredicateWithAll<(all_of FeatureD128), "d128">;
291def HasCHK           : Predicate<"Subtarget->hasCHK()">,
292                       AssemblerPredicateWithAll<(all_of FeatureCHK), "chk">;
293def HasGCS           : Predicate<"Subtarget->hasGCS()">,
294                       AssemblerPredicateWithAll<(all_of FeatureGCS), "gcs">;
295def HasCPA           : Predicate<"Subtarget->hasCPA()">,
296                       AssemblerPredicateWithAll<(all_of FeatureCPA), "cpa">;
297def IsLE             : Predicate<"Subtarget->isLittleEndian()">;
298def IsBE             : Predicate<"!Subtarget->isLittleEndian()">;
299def IsWindows        : Predicate<"Subtarget->isTargetWindows()">;
300def UseExperimentalZeroingPseudos
301    : Predicate<"Subtarget->useExperimentalZeroingPseudos()">;
302def UseAlternateSExtLoadCVTF32
303    : Predicate<"Subtarget->useAlternateSExtLoadCVTF32Pattern()">;
304
305def UseNegativeImmediates
306    : Predicate<"false">, AssemblerPredicate<(all_of (not FeatureNoNegativeImmediates)),
307                                             "NegativeImmediates">;
308
309def UseScalarIncVL : Predicate<"Subtarget->useScalarIncVL()">;
310
311def UseSVEFPLD1R : Predicate<"!Subtarget->noSVEFPLD1R()">;
312
313def IsNeonAvailable : Predicate<"Subtarget->isNeonAvailable()">;
314
315def AArch64LocalRecover : SDNode<"ISD::LOCAL_RECOVER",
316                                  SDTypeProfile<1, 1, [SDTCisSameAs<0, 1>,
317                                                       SDTCisInt<1>]>>;
318
319
320//===----------------------------------------------------------------------===//
321// AArch64-specific DAG Nodes.
322//
323
324// SDTBinaryArithWithFlagsOut - RES1, FLAGS = op LHS, RHS
325def SDTBinaryArithWithFlagsOut : SDTypeProfile<2, 2,
326                                              [SDTCisSameAs<0, 2>,
327                                               SDTCisSameAs<0, 3>,
328                                               SDTCisInt<0>, SDTCisVT<1, i32>]>;
329
330// SDTBinaryArithWithFlagsIn - RES1, FLAGS = op LHS, RHS, FLAGS
331def SDTBinaryArithWithFlagsIn : SDTypeProfile<1, 3,
332                                            [SDTCisSameAs<0, 1>,
333                                             SDTCisSameAs<0, 2>,
334                                             SDTCisInt<0>,
335                                             SDTCisVT<3, i32>]>;
336
337// SDTBinaryArithWithFlagsInOut - RES1, FLAGS = op LHS, RHS, FLAGS
338def SDTBinaryArithWithFlagsInOut : SDTypeProfile<2, 3,
339                                            [SDTCisSameAs<0, 2>,
340                                             SDTCisSameAs<0, 3>,
341                                             SDTCisInt<0>,
342                                             SDTCisVT<1, i32>,
343                                             SDTCisVT<4, i32>]>;
344
345def SDT_AArch64Brcond  : SDTypeProfile<0, 3,
346                                     [SDTCisVT<0, OtherVT>, SDTCisVT<1, i32>,
347                                      SDTCisVT<2, i32>]>;
348def SDT_AArch64cbz : SDTypeProfile<0, 2, [SDTCisInt<0>, SDTCisVT<1, OtherVT>]>;
349def SDT_AArch64tbz : SDTypeProfile<0, 3, [SDTCisInt<0>, SDTCisInt<1>,
350                                        SDTCisVT<2, OtherVT>]>;
351
352
353def SDT_AArch64CSel  : SDTypeProfile<1, 4,
354                                   [SDTCisSameAs<0, 1>,
355                                    SDTCisSameAs<0, 2>,
356                                    SDTCisInt<3>,
357                                    SDTCisVT<4, i32>]>;
358def SDT_AArch64CCMP : SDTypeProfile<1, 5,
359                                    [SDTCisVT<0, i32>,
360                                     SDTCisInt<1>,
361                                     SDTCisSameAs<1, 2>,
362                                     SDTCisInt<3>,
363                                     SDTCisInt<4>,
364                                     SDTCisVT<5, i32>]>;
365def SDT_AArch64FCCMP : SDTypeProfile<1, 5,
366                                     [SDTCisVT<0, i32>,
367                                      SDTCisFP<1>,
368                                      SDTCisSameAs<1, 2>,
369                                      SDTCisInt<3>,
370                                      SDTCisInt<4>,
371                                      SDTCisVT<5, i32>]>;
372def SDT_AArch64FCmp   : SDTypeProfile<0, 2,
373                                   [SDTCisFP<0>,
374                                    SDTCisSameAs<0, 1>]>;
375def SDT_AArch64Dup   : SDTypeProfile<1, 1, [SDTCisVec<0>]>;
376def SDT_AArch64DupLane   : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisInt<2>]>;
377def SDT_AArch64Insr  : SDTypeProfile<1, 2, [SDTCisVec<0>]>;
378def SDT_AArch64Zip   : SDTypeProfile<1, 2, [SDTCisVec<0>,
379                                          SDTCisSameAs<0, 1>,
380                                          SDTCisSameAs<0, 2>]>;
381def SDT_AArch64MOVIedit : SDTypeProfile<1, 1, [SDTCisInt<1>]>;
382def SDT_AArch64MOVIshift : SDTypeProfile<1, 2, [SDTCisInt<1>, SDTCisInt<2>]>;
383def SDT_AArch64vecimm : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0,1>,
384                                           SDTCisInt<2>, SDTCisInt<3>]>;
385def SDT_AArch64UnaryVec: SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0,1>]>;
386def SDT_AArch64ExtVec: SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0,1>,
387                                          SDTCisSameAs<0,2>, SDTCisInt<3>]>;
388def SDT_AArch64vshift : SDTypeProfile<1, 2, [SDTCisSameAs<0,1>, SDTCisInt<2>]>;
389def SDT_AArch64Dot: SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0,1>,
390                                         SDTCisVec<2>, SDTCisSameAs<2,3>]>;
391
392def SDT_AArch64vshiftinsert : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisInt<3>,
393                                                 SDTCisSameAs<0,1>,
394                                                 SDTCisSameAs<0,2>]>;
395
396def SDT_AArch64unvec : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0,1>]>;
397def SDT_AArch64fcmpz : SDTypeProfile<1, 1, []>;
398def SDT_AArch64fcmp  : SDTypeProfile<1, 2, [SDTCisSameAs<1,2>]>;
399def SDT_AArch64binvec : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisSameAs<0,1>,
400                                           SDTCisSameAs<0,2>]>;
401def SDT_AArch64trivec : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0,1>,
402                                           SDTCisSameAs<0,2>,
403                                           SDTCisSameAs<0,3>]>;
404def SDT_AArch64TCRET : SDTypeProfile<0, 2, [SDTCisPtrTy<0>]>;
405def SDT_AArch64PREFETCH : SDTypeProfile<0, 2, [SDTCisVT<0, i32>, SDTCisPtrTy<1>]>;
406
407def SDT_AArch64ITOF  : SDTypeProfile<1, 1, [SDTCisFP<0>, SDTCisSameAs<0,1>]>;
408
409def SDT_AArch64TLSDescCall : SDTypeProfile<0, -2, [SDTCisPtrTy<0>,
410                                                 SDTCisPtrTy<1>]>;
411
412def SDT_AArch64uaddlp : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisVec<1>]>;
413
414def SDT_AArch64ldp : SDTypeProfile<2, 1, [SDTCisVT<0, i64>, SDTCisSameAs<0, 1>, SDTCisPtrTy<2>]>;
415def SDT_AArch64ldiapp : SDTypeProfile<2, 1, [SDTCisVT<0, i64>, SDTCisSameAs<0, 1>, SDTCisPtrTy<2>]>;
416def SDT_AArch64ldnp : SDTypeProfile<2, 1, [SDTCisVT<0, v4i32>, SDTCisSameAs<0, 1>, SDTCisPtrTy<2>]>;
417def SDT_AArch64stp : SDTypeProfile<0, 3, [SDTCisVT<0, i64>, SDTCisSameAs<0, 1>, SDTCisPtrTy<2>]>;
418def SDT_AArch64stilp : SDTypeProfile<0, 3, [SDTCisVT<0, i64>, SDTCisSameAs<0, 1>, SDTCisPtrTy<2>]>;
419def SDT_AArch64stnp : SDTypeProfile<0, 3, [SDTCisVT<0, v4i32>, SDTCisSameAs<0, 1>, SDTCisPtrTy<2>]>;
420
421// Generates the general dynamic sequences, i.e.
422//  adrp  x0, :tlsdesc:var
423//  ldr   x1, [x0, #:tlsdesc_lo12:var]
424//  add   x0, x0, #:tlsdesc_lo12:var
425//  .tlsdesccall var
426//  blr   x1
427
428// (the TPIDR_EL0 offset is put directly in X0, hence no "result" here)
429// number of operands (the variable)
430def SDT_AArch64TLSDescCallSeq : SDTypeProfile<0,1,
431                                          [SDTCisPtrTy<0>]>;
432
433def SDT_AArch64WrapperLarge : SDTypeProfile<1, 4,
434                                        [SDTCisVT<0, i64>, SDTCisVT<1, i32>,
435                                         SDTCisSameAs<1, 2>, SDTCisSameAs<1, 3>,
436                                         SDTCisSameAs<1, 4>]>;
437
438def SDT_AArch64TBL : SDTypeProfile<1, 2, [
439  SDTCisVec<0>, SDTCisSameAs<0, 1>, SDTCisInt<2>
440]>;
441
442// non-extending masked load fragment.
443def nonext_masked_load :
444  PatFrag<(ops node:$ptr, node:$pred, node:$def),
445          (masked_ld node:$ptr, undef, node:$pred, node:$def), [{
446  return cast<MaskedLoadSDNode>(N)->getExtensionType() == ISD::NON_EXTLOAD &&
447         cast<MaskedLoadSDNode>(N)->isUnindexed() &&
448         !cast<MaskedLoadSDNode>(N)->isNonTemporal();
449}]>;
450// Any/Zero extending masked load fragments.
451def azext_masked_load :
452  PatFrag<(ops node:$ptr, node:$pred, node:$def),
453          (masked_ld node:$ptr, undef, node:$pred, node:$def),[{
454  return (cast<MaskedLoadSDNode>(N)->getExtensionType() == ISD::EXTLOAD ||
455          cast<MaskedLoadSDNode>(N)->getExtensionType() == ISD::ZEXTLOAD) &&
456         cast<MaskedLoadSDNode>(N)->isUnindexed();
457}]>;
458def azext_masked_load_i8 :
459  PatFrag<(ops node:$ptr, node:$pred, node:$def),
460          (azext_masked_load node:$ptr, node:$pred, node:$def), [{
461  return cast<MaskedLoadSDNode>(N)->getMemoryVT().getScalarType() == MVT::i8;
462}]>;
463def azext_masked_load_i16 :
464  PatFrag<(ops node:$ptr, node:$pred, node:$def),
465          (azext_masked_load node:$ptr, node:$pred, node:$def), [{
466  return cast<MaskedLoadSDNode>(N)->getMemoryVT().getScalarType() == MVT::i16;
467}]>;
468def azext_masked_load_i32 :
469  PatFrag<(ops node:$ptr, node:$pred, node:$def),
470          (azext_masked_load node:$ptr, node:$pred, node:$def), [{
471  return cast<MaskedLoadSDNode>(N)->getMemoryVT().getScalarType() == MVT::i32;
472}]>;
473// Sign extending masked load fragments.
474def sext_masked_load :
475  PatFrag<(ops node:$ptr, node:$pred, node:$def),
476          (masked_ld node:$ptr, undef, node:$pred, node:$def), [{
477  return cast<MaskedLoadSDNode>(N)->getExtensionType() == ISD::SEXTLOAD &&
478         cast<MaskedLoadSDNode>(N)->isUnindexed();
479}]>;
480def sext_masked_load_i8 :
481  PatFrag<(ops node:$ptr, node:$pred, node:$def),
482          (sext_masked_load node:$ptr, node:$pred, node:$def), [{
483  return cast<MaskedLoadSDNode>(N)->getMemoryVT().getScalarType() == MVT::i8;
484}]>;
485def sext_masked_load_i16 :
486  PatFrag<(ops node:$ptr, node:$pred, node:$def),
487          (sext_masked_load node:$ptr, node:$pred, node:$def), [{
488  return cast<MaskedLoadSDNode>(N)->getMemoryVT().getScalarType() == MVT::i16;
489}]>;
490def sext_masked_load_i32 :
491  PatFrag<(ops node:$ptr, node:$pred, node:$def),
492          (sext_masked_load node:$ptr, node:$pred, node:$def), [{
493  return cast<MaskedLoadSDNode>(N)->getMemoryVT().getScalarType() == MVT::i32;
494}]>;
495
496def non_temporal_load :
497   PatFrag<(ops node:$ptr, node:$pred, node:$def),
498           (masked_ld node:$ptr, undef, node:$pred, node:$def), [{
499   return cast<MaskedLoadSDNode>(N)->getExtensionType() == ISD::NON_EXTLOAD &&
500          cast<MaskedLoadSDNode>(N)->isUnindexed() &&
501          cast<MaskedLoadSDNode>(N)->isNonTemporal();
502}]>;
503
504// non-truncating masked store fragment.
505def nontrunc_masked_store :
506  PatFrag<(ops node:$val, node:$ptr, node:$pred),
507          (masked_st node:$val, node:$ptr, undef, node:$pred), [{
508  return !cast<MaskedStoreSDNode>(N)->isTruncatingStore() &&
509         cast<MaskedStoreSDNode>(N)->isUnindexed() &&
510         !cast<MaskedStoreSDNode>(N)->isNonTemporal();
511}]>;
512// truncating masked store fragments.
513def trunc_masked_store :
514  PatFrag<(ops node:$val, node:$ptr, node:$pred),
515          (masked_st node:$val, node:$ptr, undef, node:$pred), [{
516  return cast<MaskedStoreSDNode>(N)->isTruncatingStore() &&
517         cast<MaskedStoreSDNode>(N)->isUnindexed();
518}]>;
519def trunc_masked_store_i8 :
520  PatFrag<(ops node:$val, node:$ptr, node:$pred),
521          (trunc_masked_store node:$val, node:$ptr, node:$pred), [{
522  return cast<MaskedStoreSDNode>(N)->getMemoryVT().getScalarType() == MVT::i8;
523}]>;
524def trunc_masked_store_i16 :
525  PatFrag<(ops node:$val, node:$ptr, node:$pred),
526          (trunc_masked_store node:$val, node:$ptr, node:$pred), [{
527  return cast<MaskedStoreSDNode>(N)->getMemoryVT().getScalarType() == MVT::i16;
528}]>;
529def trunc_masked_store_i32 :
530  PatFrag<(ops node:$val, node:$ptr, node:$pred),
531          (trunc_masked_store node:$val, node:$ptr, node:$pred), [{
532  return cast<MaskedStoreSDNode>(N)->getMemoryVT().getScalarType() == MVT::i32;
533}]>;
534
535def non_temporal_store :
536  PatFrag<(ops node:$val, node:$ptr, node:$pred),
537          (masked_st node:$val, node:$ptr, undef, node:$pred), [{
538  return !cast<MaskedStoreSDNode>(N)->isTruncatingStore() &&
539         cast<MaskedStoreSDNode>(N)->isUnindexed() &&
540         cast<MaskedStoreSDNode>(N)->isNonTemporal();
541}]>;
542
543multiclass masked_gather_scatter<PatFrags GatherScatterOp> {
544  // offsets = (signed)Index << sizeof(elt)
545  def NAME#_signed_scaled :
546    PatFrag<(ops node:$val, node:$pred, node:$ptr, node:$idx),
547            (GatherScatterOp node:$val, node:$pred, node:$ptr, node:$idx),[{
548    auto MGS = cast<MaskedGatherScatterSDNode>(N);
549    bool Signed = MGS->isIndexSigned() ||
550        MGS->getIndex().getValueType().getVectorElementType() == MVT::i64;
551    return Signed && MGS->isIndexScaled();
552  }]>;
553  // offsets = (signed)Index
554  def NAME#_signed_unscaled :
555    PatFrag<(ops node:$val, node:$pred, node:$ptr, node:$idx),
556            (GatherScatterOp node:$val, node:$pred, node:$ptr, node:$idx),[{
557    auto MGS = cast<MaskedGatherScatterSDNode>(N);
558    bool Signed = MGS->isIndexSigned() ||
559        MGS->getIndex().getValueType().getVectorElementType() == MVT::i64;
560    return Signed && !MGS->isIndexScaled();
561  }]>;
562  // offsets = (unsigned)Index << sizeof(elt)
563  def NAME#_unsigned_scaled :
564    PatFrag<(ops node:$val, node:$pred, node:$ptr, node:$idx),
565            (GatherScatterOp node:$val, node:$pred, node:$ptr, node:$idx),[{
566    auto MGS = cast<MaskedGatherScatterSDNode>(N);
567    bool Signed = MGS->isIndexSigned() ||
568        MGS->getIndex().getValueType().getVectorElementType() == MVT::i64;
569    return !Signed && MGS->isIndexScaled();
570  }]>;
571  // offsets = (unsigned)Index
572  def NAME#_unsigned_unscaled :
573    PatFrag<(ops node:$val, node:$pred, node:$ptr, node:$idx),
574            (GatherScatterOp node:$val, node:$pred, node:$ptr, node:$idx),[{
575    auto MGS = cast<MaskedGatherScatterSDNode>(N);
576    bool Signed = MGS->isIndexSigned() ||
577        MGS->getIndex().getValueType().getVectorElementType() == MVT::i64;
578    return !Signed && !MGS->isIndexScaled();
579  }]>;
580}
581
582defm nonext_masked_gather    : masked_gather_scatter<nonext_masked_gather>;
583defm azext_masked_gather_i8  : masked_gather_scatter<azext_masked_gather_i8>;
584defm azext_masked_gather_i16 : masked_gather_scatter<azext_masked_gather_i16>;
585defm azext_masked_gather_i32 : masked_gather_scatter<azext_masked_gather_i32>;
586defm sext_masked_gather_i8   : masked_gather_scatter<sext_masked_gather_i8>;
587defm sext_masked_gather_i16  : masked_gather_scatter<sext_masked_gather_i16>;
588defm sext_masked_gather_i32  : masked_gather_scatter<sext_masked_gather_i32>;
589
590defm nontrunc_masked_scatter  : masked_gather_scatter<nontrunc_masked_scatter>;
591defm trunc_masked_scatter_i8  : masked_gather_scatter<trunc_masked_scatter_i8>;
592defm trunc_masked_scatter_i16 : masked_gather_scatter<trunc_masked_scatter_i16>;
593defm trunc_masked_scatter_i32 : masked_gather_scatter<trunc_masked_scatter_i32>;
594
595// top16Zero - answer true if the upper 16 bits of $src are 0, false otherwise
596def top16Zero: PatLeaf<(i32 GPR32:$src), [{
597  return SDValue(N,0)->getValueType(0) == MVT::i32 &&
598         CurDAG->MaskedValueIsZero(SDValue(N,0), APInt::getHighBitsSet(32, 16));
599  }]>;
600
601// top32Zero - answer true if the upper 32 bits of $src are 0, false otherwise
602def top32Zero: PatLeaf<(i64 GPR64:$src), [{
603  return SDValue(N,0)->getValueType(0) == MVT::i64 &&
604         CurDAG->MaskedValueIsZero(SDValue(N,0), APInt::getHighBitsSet(64, 32));
605  }]>;
606
607// topbitsallzero - Return true if all bits except the lowest bit are known zero
608def topbitsallzero32: PatLeaf<(i32 GPR32:$src), [{
609  return SDValue(N,0)->getValueType(0) == MVT::i32 &&
610         CurDAG->MaskedValueIsZero(SDValue(N,0), APInt::getHighBitsSet(32, 31));
611  }]>;
612def topbitsallzero64: PatLeaf<(i64 GPR64:$src), [{
613  return SDValue(N,0)->getValueType(0) == MVT::i64 &&
614         CurDAG->MaskedValueIsZero(SDValue(N,0), APInt::getHighBitsSet(64, 63));
615  }]>;
616
617// Node definitions.
618def AArch64adrp          : SDNode<"AArch64ISD::ADRP", SDTIntUnaryOp, []>;
619def AArch64adr           : SDNode<"AArch64ISD::ADR", SDTIntUnaryOp, []>;
620def AArch64addlow        : SDNode<"AArch64ISD::ADDlow", SDTIntBinOp, []>;
621def AArch64LOADgot       : SDNode<"AArch64ISD::LOADgot", SDTIntUnaryOp>;
622def AArch64callseq_start : SDNode<"ISD::CALLSEQ_START",
623                                SDCallSeqStart<[ SDTCisVT<0, i32>,
624                                                 SDTCisVT<1, i32> ]>,
625                                [SDNPHasChain, SDNPOutGlue]>;
626def AArch64callseq_end   : SDNode<"ISD::CALLSEQ_END",
627                                SDCallSeqEnd<[ SDTCisVT<0, i32>,
628                                               SDTCisVT<1, i32> ]>,
629                                [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue]>;
630def AArch64call          : SDNode<"AArch64ISD::CALL",
631                                SDTypeProfile<0, -1, [SDTCisPtrTy<0>]>,
632                                [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue,
633                                 SDNPVariadic]>;
634
635def AArch64call_bti      : SDNode<"AArch64ISD::CALL_BTI",
636                                SDTypeProfile<0, -1, [SDTCisPtrTy<0>]>,
637                                [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue,
638                                 SDNPVariadic]>;
639
640def AArch64call_rvmarker: SDNode<"AArch64ISD::CALL_RVMARKER",
641                             SDTypeProfile<0, -1, [SDTCisPtrTy<0>]>,
642                             [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue,
643                              SDNPVariadic]>;
644
645def AArch64brcond        : SDNode<"AArch64ISD::BRCOND", SDT_AArch64Brcond,
646                                [SDNPHasChain]>;
647def AArch64cbz           : SDNode<"AArch64ISD::CBZ", SDT_AArch64cbz,
648                                [SDNPHasChain]>;
649def AArch64cbnz           : SDNode<"AArch64ISD::CBNZ", SDT_AArch64cbz,
650                                [SDNPHasChain]>;
651def AArch64tbz           : SDNode<"AArch64ISD::TBZ", SDT_AArch64tbz,
652                                [SDNPHasChain]>;
653def AArch64tbnz           : SDNode<"AArch64ISD::TBNZ", SDT_AArch64tbz,
654                                [SDNPHasChain]>;
655
656
657def AArch64csel          : SDNode<"AArch64ISD::CSEL", SDT_AArch64CSel>;
658def AArch64csinv         : SDNode<"AArch64ISD::CSINV", SDT_AArch64CSel>;
659def AArch64csneg         : SDNode<"AArch64ISD::CSNEG", SDT_AArch64CSel>;
660def AArch64csinc         : SDNode<"AArch64ISD::CSINC", SDT_AArch64CSel>;
661def AArch64retglue       : SDNode<"AArch64ISD::RET_GLUE", SDTNone,
662                                [SDNPHasChain, SDNPOptInGlue, SDNPVariadic]>;
663def AArch64adc       : SDNode<"AArch64ISD::ADC",  SDTBinaryArithWithFlagsIn >;
664def AArch64sbc       : SDNode<"AArch64ISD::SBC",  SDTBinaryArithWithFlagsIn>;
665def AArch64add_flag  : SDNode<"AArch64ISD::ADDS",  SDTBinaryArithWithFlagsOut,
666                            [SDNPCommutative]>;
667def AArch64sub_flag  : SDNode<"AArch64ISD::SUBS",  SDTBinaryArithWithFlagsOut>;
668def AArch64and_flag  : SDNode<"AArch64ISD::ANDS",  SDTBinaryArithWithFlagsOut,
669                            [SDNPCommutative]>;
670def AArch64adc_flag  : SDNode<"AArch64ISD::ADCS",  SDTBinaryArithWithFlagsInOut>;
671def AArch64sbc_flag  : SDNode<"AArch64ISD::SBCS",  SDTBinaryArithWithFlagsInOut>;
672
673def AArch64ccmp      : SDNode<"AArch64ISD::CCMP",  SDT_AArch64CCMP>;
674def AArch64ccmn      : SDNode<"AArch64ISD::CCMN",  SDT_AArch64CCMP>;
675def AArch64fccmp     : SDNode<"AArch64ISD::FCCMP", SDT_AArch64FCCMP>;
676
677def AArch64threadpointer : SDNode<"AArch64ISD::THREAD_POINTER", SDTPtrLeaf>;
678
679def AArch64fcmp         : SDNode<"AArch64ISD::FCMP", SDT_AArch64FCmp>;
680def AArch64strict_fcmp  : SDNode<"AArch64ISD::STRICT_FCMP", SDT_AArch64FCmp,
681                                 [SDNPHasChain]>;
682def AArch64strict_fcmpe : SDNode<"AArch64ISD::STRICT_FCMPE", SDT_AArch64FCmp,
683                                 [SDNPHasChain]>;
684def AArch64any_fcmp     : PatFrags<(ops node:$lhs, node:$rhs),
685                                   [(AArch64strict_fcmp node:$lhs, node:$rhs),
686                                    (AArch64fcmp node:$lhs, node:$rhs)]>;
687
688def AArch64dup       : SDNode<"AArch64ISD::DUP", SDT_AArch64Dup>;
689def AArch64duplane8  : SDNode<"AArch64ISD::DUPLANE8", SDT_AArch64DupLane>;
690def AArch64duplane16 : SDNode<"AArch64ISD::DUPLANE16", SDT_AArch64DupLane>;
691def AArch64duplane32 : SDNode<"AArch64ISD::DUPLANE32", SDT_AArch64DupLane>;
692def AArch64duplane64 : SDNode<"AArch64ISD::DUPLANE64", SDT_AArch64DupLane>;
693def AArch64duplane128 : SDNode<"AArch64ISD::DUPLANE128", SDT_AArch64DupLane>;
694
695def AArch64insr      : SDNode<"AArch64ISD::INSR", SDT_AArch64Insr>;
696
697def AArch64zip1      : SDNode<"AArch64ISD::ZIP1", SDT_AArch64Zip>;
698def AArch64zip2      : SDNode<"AArch64ISD::ZIP2", SDT_AArch64Zip>;
699def AArch64uzp1      : SDNode<"AArch64ISD::UZP1", SDT_AArch64Zip>;
700def AArch64uzp2      : SDNode<"AArch64ISD::UZP2", SDT_AArch64Zip>;
701def AArch64trn1      : SDNode<"AArch64ISD::TRN1", SDT_AArch64Zip>;
702def AArch64trn2      : SDNode<"AArch64ISD::TRN2", SDT_AArch64Zip>;
703
704def AArch64movi_edit : SDNode<"AArch64ISD::MOVIedit", SDT_AArch64MOVIedit>;
705def AArch64movi_shift : SDNode<"AArch64ISD::MOVIshift", SDT_AArch64MOVIshift>;
706def AArch64movi_msl : SDNode<"AArch64ISD::MOVImsl", SDT_AArch64MOVIshift>;
707def AArch64mvni_shift : SDNode<"AArch64ISD::MVNIshift", SDT_AArch64MOVIshift>;
708def AArch64mvni_msl : SDNode<"AArch64ISD::MVNImsl", SDT_AArch64MOVIshift>;
709def AArch64movi : SDNode<"AArch64ISD::MOVI", SDT_AArch64MOVIedit>;
710def AArch64fmov : SDNode<"AArch64ISD::FMOV", SDT_AArch64MOVIedit>;
711
712def AArch64rev16 : SDNode<"AArch64ISD::REV16", SDT_AArch64UnaryVec>;
713def AArch64rev32 : SDNode<"AArch64ISD::REV32", SDT_AArch64UnaryVec>;
714def AArch64rev64 : SDNode<"AArch64ISD::REV64", SDT_AArch64UnaryVec>;
715def AArch64ext : SDNode<"AArch64ISD::EXT", SDT_AArch64ExtVec>;
716
717def AArch64vashr : SDNode<"AArch64ISD::VASHR", SDT_AArch64vshift>;
718def AArch64vlshr : SDNode<"AArch64ISD::VLSHR", SDT_AArch64vshift>;
719def AArch64vshl : SDNode<"AArch64ISD::VSHL", SDT_AArch64vshift>;
720def AArch64sqshli : SDNode<"AArch64ISD::SQSHL_I", SDT_AArch64vshift>;
721def AArch64uqshli : SDNode<"AArch64ISD::UQSHL_I", SDT_AArch64vshift>;
722def AArch64sqshlui : SDNode<"AArch64ISD::SQSHLU_I", SDT_AArch64vshift>;
723def AArch64srshri : SDNode<"AArch64ISD::SRSHR_I", SDT_AArch64vshift>;
724def AArch64urshri : SDNode<"AArch64ISD::URSHR_I", SDT_AArch64vshift>;
725def AArch64vsli : SDNode<"AArch64ISD::VSLI", SDT_AArch64vshiftinsert>;
726def AArch64vsri : SDNode<"AArch64ISD::VSRI", SDT_AArch64vshiftinsert>;
727
728def AArch64bit: SDNode<"AArch64ISD::BIT", SDT_AArch64trivec>;
729def AArch64bsp: SDNode<"AArch64ISD::BSP", SDT_AArch64trivec>;
730
731def AArch64cmeq: SDNode<"AArch64ISD::CMEQ", SDT_AArch64binvec>;
732def AArch64cmge: SDNode<"AArch64ISD::CMGE", SDT_AArch64binvec>;
733def AArch64cmgt: SDNode<"AArch64ISD::CMGT", SDT_AArch64binvec>;
734def AArch64cmhi: SDNode<"AArch64ISD::CMHI", SDT_AArch64binvec>;
735def AArch64cmhs: SDNode<"AArch64ISD::CMHS", SDT_AArch64binvec>;
736
737def AArch64fcmeq: SDNode<"AArch64ISD::FCMEQ", SDT_AArch64fcmp>;
738def AArch64fcmge: SDNode<"AArch64ISD::FCMGE", SDT_AArch64fcmp>;
739def AArch64fcmgt: SDNode<"AArch64ISD::FCMGT", SDT_AArch64fcmp>;
740
741def AArch64cmeqz: SDNode<"AArch64ISD::CMEQz", SDT_AArch64unvec>;
742def AArch64cmgez: SDNode<"AArch64ISD::CMGEz", SDT_AArch64unvec>;
743def AArch64cmgtz: SDNode<"AArch64ISD::CMGTz", SDT_AArch64unvec>;
744def AArch64cmlez: SDNode<"AArch64ISD::CMLEz", SDT_AArch64unvec>;
745def AArch64cmltz: SDNode<"AArch64ISD::CMLTz", SDT_AArch64unvec>;
746def AArch64cmtst : PatFrag<(ops node:$LHS, node:$RHS),
747                        (vnot (AArch64cmeqz (and node:$LHS, node:$RHS)))>;
748
749def AArch64fcmeqz: SDNode<"AArch64ISD::FCMEQz", SDT_AArch64fcmpz>;
750def AArch64fcmgez: SDNode<"AArch64ISD::FCMGEz", SDT_AArch64fcmpz>;
751def AArch64fcmgtz: SDNode<"AArch64ISD::FCMGTz", SDT_AArch64fcmpz>;
752def AArch64fcmlez: SDNode<"AArch64ISD::FCMLEz", SDT_AArch64fcmpz>;
753def AArch64fcmltz: SDNode<"AArch64ISD::FCMLTz", SDT_AArch64fcmpz>;
754
755def AArch64bici: SDNode<"AArch64ISD::BICi", SDT_AArch64vecimm>;
756def AArch64orri: SDNode<"AArch64ISD::ORRi", SDT_AArch64vecimm>;
757
758def AArch64tcret: SDNode<"AArch64ISD::TC_RETURN", SDT_AArch64TCRET,
759                  [SDNPHasChain,  SDNPOptInGlue, SDNPVariadic]>;
760
761def AArch64Prefetch        : SDNode<"AArch64ISD::PREFETCH", SDT_AArch64PREFETCH,
762                               [SDNPHasChain, SDNPSideEffect]>;
763
764def AArch64sitof: SDNode<"AArch64ISD::SITOF", SDT_AArch64ITOF>;
765def AArch64uitof: SDNode<"AArch64ISD::UITOF", SDT_AArch64ITOF>;
766
767def AArch64tlsdesc_callseq : SDNode<"AArch64ISD::TLSDESC_CALLSEQ",
768                                    SDT_AArch64TLSDescCallSeq,
769                                    [SDNPInGlue, SDNPOutGlue, SDNPHasChain,
770                                     SDNPVariadic]>;
771
772
773def AArch64WrapperLarge : SDNode<"AArch64ISD::WrapperLarge",
774                                 SDT_AArch64WrapperLarge>;
775
776def AArch64NvCast : SDNode<"AArch64ISD::NVCAST", SDTUnaryOp>;
777
778def SDT_AArch64mull : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
779                                    SDTCisSameAs<1, 2>]>;
780def AArch64pmull    : SDNode<"AArch64ISD::PMULL", SDT_AArch64mull,
781                             [SDNPCommutative]>;
782def AArch64smull    : SDNode<"AArch64ISD::SMULL", SDT_AArch64mull,
783                             [SDNPCommutative]>;
784def AArch64umull    : SDNode<"AArch64ISD::UMULL", SDT_AArch64mull,
785                             [SDNPCommutative]>;
786
787def AArch64frecpe   : SDNode<"AArch64ISD::FRECPE", SDTFPUnaryOp>;
788def AArch64frecps   : SDNode<"AArch64ISD::FRECPS", SDTFPBinOp>;
789def AArch64frsqrte  : SDNode<"AArch64ISD::FRSQRTE", SDTFPUnaryOp>;
790def AArch64frsqrts  : SDNode<"AArch64ISD::FRSQRTS", SDTFPBinOp>;
791
792def AArch64sdot     : SDNode<"AArch64ISD::SDOT", SDT_AArch64Dot>;
793def AArch64udot     : SDNode<"AArch64ISD::UDOT", SDT_AArch64Dot>;
794
795def AArch64saddv    : SDNode<"AArch64ISD::SADDV", SDT_AArch64UnaryVec>;
796def AArch64uaddv    : SDNode<"AArch64ISD::UADDV", SDT_AArch64UnaryVec>;
797def AArch64sminv    : SDNode<"AArch64ISD::SMINV", SDT_AArch64UnaryVec>;
798def AArch64uminv    : SDNode<"AArch64ISD::UMINV", SDT_AArch64UnaryVec>;
799def AArch64smaxv    : SDNode<"AArch64ISD::SMAXV", SDT_AArch64UnaryVec>;
800def AArch64umaxv    : SDNode<"AArch64ISD::UMAXV", SDT_AArch64UnaryVec>;
801def AArch64uaddlv   : SDNode<"AArch64ISD::UADDLV", SDT_AArch64uaddlp>;
802
803def AArch64uabd     : PatFrags<(ops node:$lhs, node:$rhs),
804                               [(abdu node:$lhs, node:$rhs),
805                                (int_aarch64_neon_uabd node:$lhs, node:$rhs)]>;
806def AArch64sabd     : PatFrags<(ops node:$lhs, node:$rhs),
807                               [(abds node:$lhs, node:$rhs),
808                                (int_aarch64_neon_sabd node:$lhs, node:$rhs)]>;
809
810def AArch64addp_n   : SDNode<"AArch64ISD::ADDP", SDT_AArch64Zip>;
811def AArch64uaddlp_n : SDNode<"AArch64ISD::UADDLP", SDT_AArch64uaddlp>;
812def AArch64saddlp_n : SDNode<"AArch64ISD::SADDLP", SDT_AArch64uaddlp>;
813def AArch64addp     : PatFrags<(ops node:$Rn, node:$Rm),
814                               [(AArch64addp_n node:$Rn, node:$Rm),
815                                (int_aarch64_neon_addp node:$Rn, node:$Rm)]>;
816def AArch64uaddlp   : PatFrags<(ops node:$src),
817                               [(AArch64uaddlp_n node:$src),
818                                (int_aarch64_neon_uaddlp node:$src)]>;
819def AArch64saddlp   : PatFrags<(ops node:$src),
820                               [(AArch64saddlp_n node:$src),
821                                (int_aarch64_neon_saddlp node:$src)]>;
822def AArch64faddp     : PatFrags<(ops node:$Rn, node:$Rm),
823                                [(AArch64addp_n node:$Rn, node:$Rm),
824                                 (int_aarch64_neon_faddp node:$Rn, node:$Rm)]>;
825def AArch64roundingvlshr : ComplexPattern<vAny, 2, "SelectRoundingVLShr", [AArch64vlshr]>;
826def AArch64rshrn : PatFrags<(ops node:$LHS, node:$RHS),
827                            [(trunc (AArch64roundingvlshr node:$LHS, node:$RHS)),
828                             (int_aarch64_neon_rshrn node:$LHS, node:$RHS)]>;
829def AArch64facge     : PatFrags<(ops node:$Rn, node:$Rm),
830                                [(AArch64fcmge (fabs node:$Rn), (fabs node:$Rm)),
831                                 (int_aarch64_neon_facge node:$Rn, node:$Rm)]>;
832def AArch64facgt     : PatFrags<(ops node:$Rn, node:$Rm),
833                                [(AArch64fcmgt (fabs node:$Rn), (fabs node:$Rm)),
834                                 (int_aarch64_neon_facgt node:$Rn, node:$Rm)]>;
835
836def AArch64fmaxnmv : PatFrags<(ops node:$Rn),
837                              [(vecreduce_fmax node:$Rn),
838                               (int_aarch64_neon_fmaxnmv node:$Rn)]>;
839def AArch64fminnmv : PatFrags<(ops node:$Rn),
840                              [(vecreduce_fmin node:$Rn),
841                               (int_aarch64_neon_fminnmv node:$Rn)]>;
842def AArch64fmaxv : PatFrags<(ops node:$Rn),
843                            [(vecreduce_fmaximum node:$Rn),
844                             (int_aarch64_neon_fmaxv node:$Rn)]>;
845def AArch64fminv : PatFrags<(ops node:$Rn),
846                            [(vecreduce_fminimum node:$Rn),
847                             (int_aarch64_neon_fminv node:$Rn)]>;
848
849def SDT_AArch64SETTAG : SDTypeProfile<0, 2, [SDTCisPtrTy<0>, SDTCisPtrTy<1>]>;
850def AArch64stg : SDNode<"AArch64ISD::STG", SDT_AArch64SETTAG, [SDNPHasChain, SDNPMayStore, SDNPMemOperand]>;
851def AArch64stzg : SDNode<"AArch64ISD::STZG", SDT_AArch64SETTAG, [SDNPHasChain, SDNPMayStore, SDNPMemOperand]>;
852def AArch64st2g : SDNode<"AArch64ISD::ST2G", SDT_AArch64SETTAG, [SDNPHasChain, SDNPMayStore, SDNPMemOperand]>;
853def AArch64stz2g : SDNode<"AArch64ISD::STZ2G", SDT_AArch64SETTAG, [SDNPHasChain, SDNPMayStore, SDNPMemOperand]>;
854
855def SDT_AArch64unpk : SDTypeProfile<1, 1, [
856    SDTCisInt<0>, SDTCisInt<1>, SDTCisOpSmallerThanOp<1, 0>
857]>;
858def AArch64sunpkhi : SDNode<"AArch64ISD::SUNPKHI", SDT_AArch64unpk>;
859def AArch64sunpklo : SDNode<"AArch64ISD::SUNPKLO", SDT_AArch64unpk>;
860def AArch64uunpkhi : SDNode<"AArch64ISD::UUNPKHI", SDT_AArch64unpk>;
861def AArch64uunpklo : SDNode<"AArch64ISD::UUNPKLO", SDT_AArch64unpk>;
862
863def AArch64ldp : SDNode<"AArch64ISD::LDP", SDT_AArch64ldp, [SDNPHasChain, SDNPMayLoad, SDNPMemOperand]>;
864def AArch64ldiapp : SDNode<"AArch64ISD::LDIAPP", SDT_AArch64ldiapp, [SDNPHasChain, SDNPMayLoad, SDNPMemOperand]>;
865def AArch64ldnp : SDNode<"AArch64ISD::LDNP", SDT_AArch64ldnp, [SDNPHasChain, SDNPMayLoad, SDNPMemOperand]>;
866def AArch64stp : SDNode<"AArch64ISD::STP", SDT_AArch64stp, [SDNPHasChain, SDNPMayStore, SDNPMemOperand]>;
867def AArch64stilp : SDNode<"AArch64ISD::STILP", SDT_AArch64stilp, [SDNPHasChain, SDNPMayStore, SDNPMemOperand]>;
868def AArch64stnp : SDNode<"AArch64ISD::STNP", SDT_AArch64stnp, [SDNPHasChain, SDNPMayStore, SDNPMemOperand]>;
869
870def AArch64tbl : SDNode<"AArch64ISD::TBL", SDT_AArch64TBL>;
871
872def AArch64probedalloca
873    : SDNode<"AArch64ISD::PROBED_ALLOCA",
874             SDTypeProfile<0, 1, [SDTCisPtrTy<0>]>,
875             [SDNPHasChain, SDNPMayStore]>;
876
877def AArch64mrs : SDNode<"AArch64ISD::MRS",
878                        SDTypeProfile<1, 1, [SDTCisVT<0, i64>, SDTCisVT<1, i32>]>,
879                        [SDNPHasChain, SDNPOutGlue]>;
880
881def SD_AArch64rshrnb : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisVec<1>, SDTCisInt<2>]>;
882def AArch64rshrnb : SDNode<"AArch64ISD::RSHRNB_I", SD_AArch64rshrnb>;
883def AArch64rshrnb_pf : PatFrags<(ops node:$rs, node:$i),
884                            [(AArch64rshrnb node:$rs, node:$i),
885                            (int_aarch64_sve_rshrnb node:$rs, node:$i)]>;
886
887def AArch64CttzElts : SDNode<"AArch64ISD::CTTZ_ELTS", SDTypeProfile<1, 1,
888                             [SDTCisInt<0>, SDTCisVec<1>]>, []>;
889
890// Match add node and also treat an 'or' node is as an 'add' if the or'ed operands
891// have no common bits.
892def add_and_or_is_add : PatFrags<(ops node:$lhs, node:$rhs),
893                         [(add node:$lhs, node:$rhs), (or node:$lhs, node:$rhs)],[{
894   if (N->getOpcode() == ISD::ADD)
895     return true;
896   return CurDAG->haveNoCommonBitsSet(N->getOperand(0), N->getOperand(1));
897}]> {
898  let GISelPredicateCode = [{
899     // Only handle G_ADD for now. FIXME. build capability to compute whether
900     // operands of G_OR have common bits set or not.
901     return MI.getOpcode() == TargetOpcode::G_ADD;
902  }];
903}
904
905// Match mul with enough sign-bits. Can be reduced to a smaller mul operand.
906def smullwithsignbits : PatFrag<(ops node:$l, node:$r), (mul node:$l, node:$r), [{
907  return CurDAG->ComputeNumSignBits(N->getOperand(0)) > 32 &&
908         CurDAG->ComputeNumSignBits(N->getOperand(1)) > 32;
909}]>;
910
911//===----------------------------------------------------------------------===//
912
913//===----------------------------------------------------------------------===//
914
915// AArch64 Instruction Predicate Definitions.
916// We could compute these on a per-module basis but doing so requires accessing
917// the Function object through the <Target>Subtarget and objections were raised
918// to that (see post-commit review comments for r301750).
919let RecomputePerFunction = 1 in {
920  def ForCodeSize   : Predicate<"shouldOptForSize(MF)">;
921  def NotForCodeSize   : Predicate<"!shouldOptForSize(MF)">;
922  // Avoid generating STRQro if it is slow, unless we're optimizing for code size.
923  def UseSTRQro : Predicate<"!Subtarget->isSTRQroSlow() || shouldOptForSize(MF)">;
924
925  def UseBTI : Predicate<[{ MF->getInfo<AArch64FunctionInfo>()->branchTargetEnforcement() }]>;
926  def NotUseBTI : Predicate<[{ !MF->getInfo<AArch64FunctionInfo>()->branchTargetEnforcement() }]>;
927
928  def SLSBLRMitigation : Predicate<[{ MF->getSubtarget<AArch64Subtarget>().hardenSlsBlr() }]>;
929  def NoSLSBLRMitigation : Predicate<[{ !MF->getSubtarget<AArch64Subtarget>().hardenSlsBlr() }]>;
930  // Toggles patterns which aren't beneficial in GlobalISel when we aren't
931  // optimizing. This allows us to selectively use patterns without impacting
932  // SelectionDAG's behaviour.
933  // FIXME: One day there will probably be a nicer way to check for this, but
934  // today is not that day.
935  def OptimizedGISelOrOtherSelector : Predicate<"!MF->getFunction().hasOptNone() || MF->getProperties().hasProperty(MachineFunctionProperties::Property::FailedISel) || !MF->getProperties().hasProperty(MachineFunctionProperties::Property::Legalized)">;
936}
937
938include "AArch64InstrFormats.td"
939include "SVEInstrFormats.td"
940include "SMEInstrFormats.td"
941
942//===----------------------------------------------------------------------===//
943
944//===----------------------------------------------------------------------===//
945// Miscellaneous instructions.
946//===----------------------------------------------------------------------===//
947
948let hasSideEffects = 1, isCodeGenOnly = 1 in {
949let Defs = [SP], Uses = [SP] in {
950// We set Sched to empty list because we expect these instructions to simply get
951// removed in most cases.
952def ADJCALLSTACKDOWN : Pseudo<(outs), (ins i32imm:$amt1, i32imm:$amt2),
953                              [(AArch64callseq_start timm:$amt1, timm:$amt2)]>,
954                              Sched<[]>;
955def ADJCALLSTACKUP : Pseudo<(outs), (ins i32imm:$amt1, i32imm:$amt2),
956                            [(AArch64callseq_end timm:$amt1, timm:$amt2)]>,
957                            Sched<[]>;
958
959}
960
961let Defs = [SP, NZCV], Uses = [SP] in {
962// Probed stack allocation of a constant size, used in function prologues when
963// stack-clash protection is enabled.
964def PROBED_STACKALLOC : Pseudo<(outs GPR64:$scratch),
965                               (ins i64imm:$stacksize, i64imm:$fixed_offset,
966                                i64imm:$scalable_offset),
967                               []>,
968                               Sched<[]>;
969
970// Probed stack allocation of a variable size, used in function prologues when
971// stack-clash protection is enabled.
972def PROBED_STACKALLOC_VAR : Pseudo<(outs),
973                                   (ins GPR64sp:$target),
974                                   []>,
975                                   Sched<[]>;
976
977// Probed stack allocations of a variable size, used for allocas of unknown size
978// when stack-clash protection is enabled.
979let usesCustomInserter = 1 in
980def PROBED_STACKALLOC_DYN : Pseudo<(outs),
981                                   (ins GPR64common:$target),
982                                   [(AArch64probedalloca GPR64common:$target)]>,
983                                   Sched<[]>;
984
985} // Defs = [SP, NZCV], Uses = [SP] in
986} // hasSideEffects = 1, isCodeGenOnly = 1
987
988let isReMaterializable = 1, isCodeGenOnly = 1 in {
989// FIXME: The following pseudo instructions are only needed because remat
990// cannot handle multiple instructions.  When that changes, they can be
991// removed, along with the AArch64Wrapper node.
992
993let AddedComplexity = 10 in
994def LOADgot : Pseudo<(outs GPR64common:$dst), (ins i64imm:$addr),
995                     [(set GPR64common:$dst, (AArch64LOADgot tglobaladdr:$addr))]>,
996              Sched<[WriteLDAdr]>;
997
998// The MOVaddr instruction should match only when the add is not folded
999// into a load or store address.
1000def MOVaddr
1001    : Pseudo<(outs GPR64common:$dst), (ins i64imm:$hi, i64imm:$low),
1002             [(set GPR64common:$dst, (AArch64addlow (AArch64adrp tglobaladdr:$hi),
1003                                            tglobaladdr:$low))]>,
1004      Sched<[WriteAdrAdr]>;
1005def MOVaddrJT
1006    : Pseudo<(outs GPR64common:$dst), (ins i64imm:$hi, i64imm:$low),
1007             [(set GPR64common:$dst, (AArch64addlow (AArch64adrp tjumptable:$hi),
1008                                             tjumptable:$low))]>,
1009      Sched<[WriteAdrAdr]>;
1010def MOVaddrCP
1011    : Pseudo<(outs GPR64common:$dst), (ins i64imm:$hi, i64imm:$low),
1012             [(set GPR64common:$dst, (AArch64addlow (AArch64adrp tconstpool:$hi),
1013                                             tconstpool:$low))]>,
1014      Sched<[WriteAdrAdr]>;
1015def MOVaddrBA
1016    : Pseudo<(outs GPR64common:$dst), (ins i64imm:$hi, i64imm:$low),
1017             [(set GPR64common:$dst, (AArch64addlow (AArch64adrp tblockaddress:$hi),
1018                                             tblockaddress:$low))]>,
1019      Sched<[WriteAdrAdr]>;
1020def MOVaddrTLS
1021    : Pseudo<(outs GPR64common:$dst), (ins i64imm:$hi, i64imm:$low),
1022             [(set GPR64common:$dst, (AArch64addlow (AArch64adrp tglobaltlsaddr:$hi),
1023                                            tglobaltlsaddr:$low))]>,
1024      Sched<[WriteAdrAdr]>;
1025def MOVaddrEXT
1026    : Pseudo<(outs GPR64common:$dst), (ins i64imm:$hi, i64imm:$low),
1027             [(set GPR64common:$dst, (AArch64addlow (AArch64adrp texternalsym:$hi),
1028                                            texternalsym:$low))]>,
1029      Sched<[WriteAdrAdr]>;
1030// Normally AArch64addlow either gets folded into a following ldr/str,
1031// or together with an adrp into MOVaddr above. For cases with TLS, it
1032// might appear without either of them, so allow lowering it into a plain
1033// add.
1034def ADDlowTLS
1035    : Pseudo<(outs GPR64sp:$dst), (ins GPR64sp:$src, i64imm:$low),
1036             [(set GPR64sp:$dst, (AArch64addlow GPR64sp:$src,
1037                                            tglobaltlsaddr:$low))]>,
1038      Sched<[WriteAdr]>;
1039
1040} // isReMaterializable, isCodeGenOnly
1041
1042def : Pat<(AArch64LOADgot tglobaltlsaddr:$addr),
1043          (LOADgot tglobaltlsaddr:$addr)>;
1044
1045def : Pat<(AArch64LOADgot texternalsym:$addr),
1046          (LOADgot texternalsym:$addr)>;
1047
1048def : Pat<(AArch64LOADgot tconstpool:$addr),
1049          (LOADgot tconstpool:$addr)>;
1050
1051// In general these get lowered into a sequence of three 4-byte instructions.
1052// 32-bit jump table destination is actually only 2 instructions since we can
1053// use the table itself as a PC-relative base. But optimization occurs after
1054// branch relaxation so be pessimistic.
1055let Size = 12, Constraints = "@earlyclobber $dst,@earlyclobber $scratch",
1056    isNotDuplicable = 1 in {
1057def JumpTableDest32 : Pseudo<(outs GPR64:$dst, GPR64sp:$scratch),
1058                             (ins GPR64:$table, GPR64:$entry, i32imm:$jti), []>,
1059                      Sched<[]>;
1060def JumpTableDest16 : Pseudo<(outs GPR64:$dst, GPR64sp:$scratch),
1061                             (ins GPR64:$table, GPR64:$entry, i32imm:$jti), []>,
1062                      Sched<[]>;
1063def JumpTableDest8 : Pseudo<(outs GPR64:$dst, GPR64sp:$scratch),
1064                            (ins GPR64:$table, GPR64:$entry, i32imm:$jti), []>,
1065                     Sched<[]>;
1066}
1067
1068// Space-consuming pseudo to aid testing of placement and reachability
1069// algorithms. Immediate operand is the number of bytes this "instruction"
1070// occupies; register operands can be used to enforce dependency and constrain
1071// the scheduler.
1072let hasSideEffects = 1, mayLoad = 1, mayStore = 1 in
1073def SPACE : Pseudo<(outs GPR64:$Rd), (ins i32imm:$size, GPR64:$Rn),
1074                   [(set GPR64:$Rd, (int_aarch64_space imm:$size, GPR64:$Rn))]>,
1075            Sched<[]>;
1076
1077let hasSideEffects = 1, isCodeGenOnly = 1 in {
1078  def SpeculationSafeValueX
1079      : Pseudo<(outs GPR64:$dst), (ins GPR64:$src), []>, Sched<[]>;
1080  def SpeculationSafeValueW
1081      : Pseudo<(outs GPR32:$dst), (ins GPR32:$src), []>, Sched<[]>;
1082}
1083
1084// SpeculationBarrierEndBB must only be used after an unconditional control
1085// flow, i.e. after a terminator for which isBarrier is True.
1086let hasSideEffects = 1, isCodeGenOnly = 1, isTerminator = 1, isBarrier = 1 in {
1087  // This gets lowered to a pair of 4-byte instructions.
1088  let Size = 8 in
1089  def SpeculationBarrierISBDSBEndBB
1090      : Pseudo<(outs), (ins), []>, Sched<[]>;
1091  // This gets lowered to a 4-byte instruction.
1092  let Size = 4 in
1093  def SpeculationBarrierSBEndBB
1094      : Pseudo<(outs), (ins), []>, Sched<[]>;
1095}
1096
1097//===----------------------------------------------------------------------===//
1098// System instructions.
1099//===----------------------------------------------------------------------===//
1100
1101def HINT : HintI<"hint">;
1102def : InstAlias<"nop",  (HINT 0b000)>;
1103def : InstAlias<"yield",(HINT 0b001)>;
1104def : InstAlias<"wfe",  (HINT 0b010)>;
1105def : InstAlias<"wfi",  (HINT 0b011)>;
1106def : InstAlias<"sev",  (HINT 0b100)>;
1107def : InstAlias<"sevl", (HINT 0b101)>;
1108def : InstAlias<"dgh",  (HINT 0b110)>;
1109def : InstAlias<"esb",  (HINT 0b10000)>, Requires<[HasRAS]>;
1110def : InstAlias<"csdb", (HINT 20)>;
1111// In order to be able to write readable assembly, LLVM should accept assembly
1112// inputs that use Branch Target Indentification mnemonics, even with BTI disabled.
1113// However, in order to be compatible with other assemblers (e.g. GAS), LLVM
1114// should not emit these mnemonics unless BTI is enabled.
1115def : InstAlias<"bti",  (HINT 32), 0>;
1116def : InstAlias<"bti $op", (HINT btihint_op:$op), 0>;
1117def : InstAlias<"bti",  (HINT 32)>, Requires<[HasBTI]>;
1118def : InstAlias<"bti $op", (HINT btihint_op:$op)>, Requires<[HasBTI]>;
1119
1120// v8.2a Statistical Profiling extension
1121def : InstAlias<"psb $op",  (HINT psbhint_op:$op)>, Requires<[HasSPE]>;
1122
1123// As far as LLVM is concerned this writes to the system's exclusive monitors.
1124let mayLoad = 1, mayStore = 1 in
1125def CLREX : CRmSystemI<imm0_15, 0b010, "clrex">;
1126
1127// NOTE: ideally, this would have mayStore = 0, mayLoad = 0, but we cannot
1128// model patterns with sufficiently fine granularity.
1129let mayLoad = ?, mayStore = ? in {
1130def DMB   : CRmSystemI<barrier_op, 0b101, "dmb",
1131                       [(int_aarch64_dmb (i32 imm32_0_15:$CRm))]>;
1132
1133def DSB   : CRmSystemI<barrier_op, 0b100, "dsb",
1134                       [(int_aarch64_dsb (i32 imm32_0_15:$CRm))]>;
1135
1136def ISB   : CRmSystemI<barrier_op, 0b110, "isb",
1137                       [(int_aarch64_isb (i32 imm32_0_15:$CRm))]>;
1138
1139def TSB   : CRmSystemI<barrier_op, 0b010, "tsb", []> {
1140  let CRm        = 0b0010;
1141  let Inst{12}   = 0;
1142  let Predicates = [HasTRACEV8_4];
1143}
1144
1145def DSBnXS  : CRmSystemI<barrier_nxs_op, 0b001, "dsb"> {
1146  let CRm{1-0}   = 0b11;
1147  let Inst{9-8}  = 0b10;
1148  let Predicates = [HasXS];
1149}
1150
1151let Predicates = [HasWFxT] in {
1152def WFET : RegInputSystemI<0b0000, 0b000, "wfet">;
1153def WFIT : RegInputSystemI<0b0000, 0b001, "wfit">;
1154}
1155
1156// Branch Record Buffer two-word mnemonic instructions
1157class BRBEI<bits<3> op2, string keyword>
1158    : SimpleSystemI<0, (ins), "brb", keyword>, Sched<[WriteSys]> {
1159  let Inst{31-8} = 0b110101010000100101110010;
1160  let Inst{7-5} = op2;
1161  let Predicates = [HasBRBE];
1162}
1163def BRB_IALL: BRBEI<0b100, "\tiall">;
1164def BRB_INJ:  BRBEI<0b101, "\tinj">;
1165
1166}
1167
1168// Allow uppercase and lowercase keyword arguments for BRB IALL and BRB INJ
1169def : TokenAlias<"INJ", "inj">;
1170def : TokenAlias<"IALL", "iall">;
1171
1172
1173// ARMv9.4-A Guarded Control Stack
1174class GCSNoOp<bits<3> op2, string mnemonic>
1175    : SimpleSystemI<0, (ins), mnemonic, "">, Sched<[]> {
1176  let Inst{20-8} = 0b0100001110111;
1177  let Inst{7-5} = op2;
1178  let Predicates = [HasGCS];
1179}
1180def GCSPUSHX : GCSNoOp<0b100, "gcspushx">;
1181def GCSPOPCX : GCSNoOp<0b101, "gcspopcx">;
1182def GCSPOPX  : GCSNoOp<0b110, "gcspopx">;
1183
1184class GCSRtIn<bits<3> op1, bits<3> op2, string mnemonic,
1185            list<dag> pattern = []>
1186    : RtSystemI<0, (outs), (ins GPR64:$Rt), mnemonic, "\t$Rt", pattern> {
1187  let Inst{20-19} = 0b01;
1188  let Inst{18-16} = op1;
1189  let Inst{15-8} = 0b01110111;
1190  let Inst{7-5} = op2;
1191  let Predicates = [HasGCS];
1192}
1193
1194def GCSSS1   : GCSRtIn<0b011, 0b010, "gcsss1">;
1195def GCSPUSHM : GCSRtIn<0b011, 0b000, "gcspushm">;
1196
1197class GCSRtOut<bits<3> op1, bits<3> op2, string mnemonic,
1198            list<dag> pattern = []>
1199    : RtSystemI<1, (outs GPR64:$Rt), (ins), mnemonic, "\t$Rt", pattern> {
1200  let Inst{20-19} = 0b01;
1201  let Inst{18-16} = op1;
1202  let Inst{15-8} = 0b01110111;
1203  let Inst{7-5} = op2;
1204  let Predicates = [HasGCS];
1205}
1206
1207def GCSSS2  : GCSRtOut<0b011, 0b011, "gcsss2">;
1208def GCSPOPM : GCSRtOut<0b011, 0b001, "gcspopm">;
1209def GCSPOPM_NoOp : InstAlias<"gcspopm", (GCSPOPM XZR)>, Requires<[HasGCS]>; // Rt defaults to XZR if absent
1210
1211def GCSB_DSYNC_disable : InstAlias<"gcsb\tdsync", (HINT 19), 0>;
1212def GCSB_DSYNC         : InstAlias<"gcsb\tdsync", (HINT 19), 1>, Requires<[HasGCS]>;
1213
1214def : TokenAlias<"DSYNC", "dsync">;
1215
1216let Uses = [X16], Defs = [X16], CRm = 0b0101 in {
1217  def CHKFEAT   : SystemNoOperands<0b000, "hint\t#40">;
1218}
1219def : InstAlias<"chkfeat\tx16", (CHKFEAT), 0>;
1220def : InstAlias<"chkfeat\tx16", (CHKFEAT), 1>, Requires<[HasCHK]>;
1221
1222class GCSSt<string mnemonic, bits<3> op>
1223    : I<(outs), (ins GPR64:$Rt, GPR64sp:$Rn), mnemonic, "\t$Rt, $Rn", "", []>, Sched<[]> {
1224  bits<5> Rt;
1225  bits<5> Rn;
1226  let Inst{31-15} = 0b11011001000111110;
1227  let Inst{14-12} = op;
1228  let Inst{11-10} = 0b11;
1229  let Inst{9-5} = Rn;
1230  let Inst{4-0} = Rt;
1231  let Predicates = [HasGCS];
1232}
1233def GCSSTR  : GCSSt<"gcsstr",  0b000>;
1234def GCSSTTR : GCSSt<"gcssttr", 0b001>;
1235
1236
1237// ARMv8.2-A Dot Product
1238let Predicates = [HasDotProd] in {
1239defm SDOT : SIMDThreeSameVectorDot<0, 0, "sdot", AArch64sdot>;
1240defm UDOT : SIMDThreeSameVectorDot<1, 0, "udot", AArch64udot>;
1241defm SDOTlane : SIMDThreeSameVectorDotIndex<0, 0, 0b10, "sdot", AArch64sdot>;
1242defm UDOTlane : SIMDThreeSameVectorDotIndex<1, 0, 0b10, "udot", AArch64udot>;
1243}
1244
1245// ARMv8.6-A BFloat
1246let Predicates = [HasNEON, HasBF16] in {
1247defm BFDOT       : SIMDThreeSameVectorBFDot<1, "bfdot">;
1248defm BF16DOTlane : SIMDThreeSameVectorBF16DotI<0, "bfdot">;
1249def BFMMLA       : SIMDThreeSameVectorBF16MatrixMul<"bfmmla">;
1250def BFMLALB      : SIMDBF16MLAL<0, "bfmlalb", int_aarch64_neon_bfmlalb>;
1251def BFMLALT      : SIMDBF16MLAL<1, "bfmlalt", int_aarch64_neon_bfmlalt>;
1252def BFMLALBIdx   : SIMDBF16MLALIndex<0, "bfmlalb", int_aarch64_neon_bfmlalb>;
1253def BFMLALTIdx   : SIMDBF16MLALIndex<1, "bfmlalt", int_aarch64_neon_bfmlalt>;
1254def BFCVTN       : SIMD_BFCVTN;
1255def BFCVTN2      : SIMD_BFCVTN2;
1256
1257// Vector-scalar BFDOT:
1258// The second source operand of the 64-bit variant of BF16DOTlane is a 128-bit
1259// register (the instruction uses a single 32-bit lane from it), so the pattern
1260// is a bit tricky.
1261def : Pat<(v2f32 (int_aarch64_neon_bfdot
1262                    (v2f32 V64:$Rd), (v4bf16 V64:$Rn),
1263                    (v4bf16 (bitconvert
1264                      (v2i32 (AArch64duplane32
1265                        (v4i32 (bitconvert
1266                          (v8bf16 (insert_subvector undef,
1267                            (v4bf16 V64:$Rm),
1268                            (i64 0))))),
1269                        VectorIndexS:$idx)))))),
1270          (BF16DOTlanev4bf16 (v2f32 V64:$Rd), (v4bf16 V64:$Rn),
1271                             (SUBREG_TO_REG (i32 0), V64:$Rm, dsub),
1272                             VectorIndexS:$idx)>;
1273}
1274
1275let Predicates = [HasNEONorSME, HasBF16] in {
1276def BFCVT : BF16ToSinglePrecision<"bfcvt">;
1277}
1278
1279// ARMv8.6A AArch64 matrix multiplication
1280let Predicates = [HasMatMulInt8] in {
1281def  SMMLA : SIMDThreeSameVectorMatMul<0, 0, "smmla", int_aarch64_neon_smmla>;
1282def  UMMLA : SIMDThreeSameVectorMatMul<0, 1, "ummla", int_aarch64_neon_ummla>;
1283def USMMLA : SIMDThreeSameVectorMatMul<1, 0, "usmmla", int_aarch64_neon_usmmla>;
1284defm USDOT : SIMDThreeSameVectorDot<0, 1, "usdot", int_aarch64_neon_usdot>;
1285defm USDOTlane : SIMDThreeSameVectorDotIndex<0, 1, 0b10, "usdot", int_aarch64_neon_usdot>;
1286
1287// sudot lane has a pattern where usdot is expected (there is no sudot).
1288// The second operand is used in the dup operation to repeat the indexed
1289// element.
1290class BaseSIMDSUDOTIndex<bit Q, string dst_kind, string lhs_kind,
1291                         string rhs_kind, RegisterOperand RegType,
1292                         ValueType AccumType, ValueType InputType>
1293      : BaseSIMDThreeSameVectorIndexS<Q, 0, 0b00, 0b1111, "sudot", dst_kind,
1294                                        lhs_kind, rhs_kind, RegType, AccumType,
1295                                        InputType, null_frag> {
1296  let Pattern = [(set (AccumType RegType:$dst),
1297                      (AccumType (int_aarch64_neon_usdot (AccumType RegType:$Rd),
1298                                 (InputType (bitconvert (AccumType
1299                                    (AArch64duplane32 (v4i32 V128:$Rm),
1300                                        VectorIndexS:$idx)))),
1301                                 (InputType RegType:$Rn))))];
1302}
1303
1304multiclass SIMDSUDOTIndex {
1305  def v8i8  : BaseSIMDSUDOTIndex<0, ".2s", ".8b", ".4b", V64, v2i32, v8i8>;
1306  def v16i8 : BaseSIMDSUDOTIndex<1, ".4s", ".16b", ".4b", V128, v4i32, v16i8>;
1307}
1308
1309defm SUDOTlane : SIMDSUDOTIndex;
1310
1311}
1312
1313// ARMv8.2-A FP16 Fused Multiply-Add Long
1314let Predicates = [HasNEON, HasFP16FML] in {
1315defm FMLAL      : SIMDThreeSameVectorFML<0, 1, 0b001, "fmlal", int_aarch64_neon_fmlal>;
1316defm FMLSL      : SIMDThreeSameVectorFML<0, 1, 0b101, "fmlsl", int_aarch64_neon_fmlsl>;
1317defm FMLAL2     : SIMDThreeSameVectorFML<1, 0, 0b001, "fmlal2", int_aarch64_neon_fmlal2>;
1318defm FMLSL2     : SIMDThreeSameVectorFML<1, 0, 0b101, "fmlsl2", int_aarch64_neon_fmlsl2>;
1319defm FMLALlane  : SIMDThreeSameVectorFMLIndex<0, 0b0000, "fmlal", int_aarch64_neon_fmlal>;
1320defm FMLSLlane  : SIMDThreeSameVectorFMLIndex<0, 0b0100, "fmlsl", int_aarch64_neon_fmlsl>;
1321defm FMLAL2lane : SIMDThreeSameVectorFMLIndex<1, 0b1000, "fmlal2", int_aarch64_neon_fmlal2>;
1322defm FMLSL2lane : SIMDThreeSameVectorFMLIndex<1, 0b1100, "fmlsl2", int_aarch64_neon_fmlsl2>;
1323}
1324
1325// Armv8.2-A Crypto extensions
1326let Predicates = [HasSHA3] in {
1327def SHA512H   : CryptoRRRTied<0b0, 0b00, "sha512h">;
1328def SHA512H2  : CryptoRRRTied<0b0, 0b01, "sha512h2">;
1329def SHA512SU0 : CryptoRRTied_2D<0b0, 0b00, "sha512su0">;
1330def SHA512SU1 : CryptoRRRTied_2D<0b0, 0b10, "sha512su1">;
1331def RAX1      : CryptoRRR_2D<0b0,0b11, "rax1">;
1332def EOR3      : CryptoRRRR_16B<0b00, "eor3">;
1333def BCAX      : CryptoRRRR_16B<0b01, "bcax">;
1334def XAR       : CryptoRRRi6<"xar">;
1335
1336class SHA3_pattern<Instruction INST, Intrinsic OpNode, ValueType VecTy>
1337  : Pat<(VecTy (OpNode (VecTy V128:$Vd), (VecTy V128:$Vn), (VecTy V128:$Vm))),
1338        (INST (VecTy V128:$Vd), (VecTy V128:$Vn), (VecTy V128:$Vm))>;
1339
1340def : Pat<(v2i64 (int_aarch64_crypto_sha512su0 (v2i64 V128:$Vn), (v2i64 V128:$Vm))),
1341          (SHA512SU0 (v2i64 V128:$Vn), (v2i64 V128:$Vm))>;
1342
1343def : SHA3_pattern<SHA512H, int_aarch64_crypto_sha512h, v2i64>;
1344def : SHA3_pattern<SHA512H2, int_aarch64_crypto_sha512h2, v2i64>;
1345def : SHA3_pattern<SHA512SU1, int_aarch64_crypto_sha512su1, v2i64>;
1346
1347def : SHA3_pattern<EOR3, int_aarch64_crypto_eor3u, v16i8>;
1348def : SHA3_pattern<EOR3, int_aarch64_crypto_eor3u, v8i16>;
1349def : SHA3_pattern<EOR3, int_aarch64_crypto_eor3u, v4i32>;
1350def : SHA3_pattern<EOR3, int_aarch64_crypto_eor3u, v2i64>;
1351
1352class EOR3_pattern<ValueType VecTy>
1353  : Pat<(xor (xor (VecTy V128:$Vn), (VecTy V128:$Vm)), (VecTy V128:$Va)),
1354        (EOR3 (VecTy V128:$Vn), (VecTy V128:$Vm), (VecTy V128:$Va))>;
1355
1356def : EOR3_pattern<v16i8>;
1357def : EOR3_pattern<v8i16>;
1358def : EOR3_pattern<v4i32>;
1359def : EOR3_pattern<v2i64>;
1360
1361class BCAX_pattern<ValueType VecTy>
1362  : Pat<(xor (VecTy V128:$Vn), (and (VecTy V128:$Vm), (vnot (VecTy V128:$Va)))),
1363        (BCAX (VecTy V128:$Vn), (VecTy V128:$Vm), (VecTy V128:$Va))>;
1364
1365def : BCAX_pattern<v16i8>;
1366def : BCAX_pattern<v8i16>;
1367def : BCAX_pattern<v4i32>;
1368def : BCAX_pattern<v2i64>;
1369
1370def : SHA3_pattern<BCAX, int_aarch64_crypto_bcaxu, v16i8>;
1371def : SHA3_pattern<BCAX, int_aarch64_crypto_bcaxu, v8i16>;
1372def : SHA3_pattern<BCAX, int_aarch64_crypto_bcaxu, v4i32>;
1373def : SHA3_pattern<BCAX, int_aarch64_crypto_bcaxu, v2i64>;
1374
1375def : SHA3_pattern<EOR3, int_aarch64_crypto_eor3s, v16i8>;
1376def : SHA3_pattern<EOR3, int_aarch64_crypto_eor3s, v8i16>;
1377def : SHA3_pattern<EOR3, int_aarch64_crypto_eor3s, v4i32>;
1378def : SHA3_pattern<EOR3, int_aarch64_crypto_eor3s, v2i64>;
1379
1380def : SHA3_pattern<BCAX, int_aarch64_crypto_bcaxs, v16i8>;
1381def : SHA3_pattern<BCAX, int_aarch64_crypto_bcaxs, v8i16>;
1382def : SHA3_pattern<BCAX, int_aarch64_crypto_bcaxs, v4i32>;
1383def : SHA3_pattern<BCAX, int_aarch64_crypto_bcaxs, v2i64>;
1384
1385def : Pat<(v2i64 (int_aarch64_crypto_rax1 (v2i64 V128:$Vn), (v2i64 V128:$Vm))),
1386          (RAX1 (v2i64 V128:$Vn), (v2i64 V128:$Vm))>;
1387
1388def : Pat<(v2i64 (int_aarch64_crypto_xar (v2i64 V128:$Vn), (v2i64 V128:$Vm), (i64 timm0_63:$imm))),
1389          (XAR (v2i64 V128:$Vn), (v2i64 V128:$Vm), (timm0_63:$imm))>;
1390
1391def : Pat<(xor  (v2i64 V128:$Vn), (or (AArch64vlshr (v2i64 V128:$Vm), (i32 63)), (AArch64vshl (v2i64 V128:$Vm), (i32 1)))),
1392          (RAX1 (v2i64 V128:$Vn), (v2i64 V128:$Vm))>;
1393
1394} // HasSHA3
1395
1396let Predicates = [HasSM4] in {
1397def SM3TT1A   : CryptoRRRi2Tied<0b0, 0b00, "sm3tt1a">;
1398def SM3TT1B   : CryptoRRRi2Tied<0b0, 0b01, "sm3tt1b">;
1399def SM3TT2A   : CryptoRRRi2Tied<0b0, 0b10, "sm3tt2a">;
1400def SM3TT2B   : CryptoRRRi2Tied<0b0, 0b11, "sm3tt2b">;
1401def SM3SS1    : CryptoRRRR_4S<0b10, "sm3ss1">;
1402def SM3PARTW1 : CryptoRRRTied_4S<0b1, 0b00, "sm3partw1">;
1403def SM3PARTW2 : CryptoRRRTied_4S<0b1, 0b01, "sm3partw2">;
1404def SM4ENCKEY : CryptoRRR_4S<0b1, 0b10, "sm4ekey">;
1405def SM4E      : CryptoRRTied_4S<0b0, 0b01, "sm4e">;
1406
1407def : Pat<(v4i32 (int_aarch64_crypto_sm3ss1 (v4i32 V128:$Vn), (v4i32 V128:$Vm), (v4i32 V128:$Va))),
1408          (SM3SS1 (v4i32 V128:$Vn), (v4i32 V128:$Vm), (v4i32 V128:$Va))>;
1409
1410class SM3PARTW_pattern<Instruction INST, Intrinsic OpNode>
1411  : Pat<(v4i32 (OpNode (v4i32 V128:$Vd), (v4i32 V128:$Vn), (v4i32 V128:$Vm))),
1412        (INST (v4i32 V128:$Vd), (v4i32 V128:$Vn), (v4i32 V128:$Vm))>;
1413
1414class SM3TT_pattern<Instruction INST, Intrinsic OpNode>
1415  : Pat<(v4i32 (OpNode (v4i32 V128:$Vd), (v4i32 V128:$Vn), (v4i32 V128:$Vm), (i64 VectorIndexS_timm:$imm) )),
1416        (INST (v4i32 V128:$Vd), (v4i32 V128:$Vn), (v4i32 V128:$Vm), (VectorIndexS_timm:$imm))>;
1417
1418class SM4_pattern<Instruction INST, Intrinsic OpNode>
1419  : Pat<(v4i32 (OpNode (v4i32 V128:$Vn), (v4i32 V128:$Vm))),
1420        (INST (v4i32 V128:$Vn), (v4i32 V128:$Vm))>;
1421
1422def : SM3PARTW_pattern<SM3PARTW1, int_aarch64_crypto_sm3partw1>;
1423def : SM3PARTW_pattern<SM3PARTW2, int_aarch64_crypto_sm3partw2>;
1424
1425def : SM3TT_pattern<SM3TT1A, int_aarch64_crypto_sm3tt1a>;
1426def : SM3TT_pattern<SM3TT1B, int_aarch64_crypto_sm3tt1b>;
1427def : SM3TT_pattern<SM3TT2A, int_aarch64_crypto_sm3tt2a>;
1428def : SM3TT_pattern<SM3TT2B, int_aarch64_crypto_sm3tt2b>;
1429
1430def : SM4_pattern<SM4ENCKEY, int_aarch64_crypto_sm4ekey>;
1431def : SM4_pattern<SM4E, int_aarch64_crypto_sm4e>;
1432} // HasSM4
1433
1434let Predicates = [HasRCPC] in {
1435  // v8.3 Release Consistent Processor Consistent support, optional in v8.2.
1436  def LDAPRB  : RCPCLoad<0b00, "ldaprb", GPR32>;
1437  def LDAPRH  : RCPCLoad<0b01, "ldaprh", GPR32>;
1438  def LDAPRW  : RCPCLoad<0b10, "ldapr", GPR32>;
1439  def LDAPRX  : RCPCLoad<0b11, "ldapr", GPR64>;
1440}
1441
1442// v8.3a complex add and multiply-accumulate. No predicate here, that is done
1443// inside the multiclass as the FP16 versions need different predicates.
1444defm FCMLA : SIMDThreeSameVectorTiedComplexHSD<1, 0b110, complexrotateop,
1445                                               "fcmla", null_frag>;
1446defm FCADD : SIMDThreeSameVectorComplexHSD<1, 0b111, complexrotateopodd,
1447                                           "fcadd", null_frag>;
1448defm FCMLA : SIMDIndexedTiedComplexHSD<0, 1, complexrotateop, "fcmla">;
1449
1450let Predicates = [HasComplxNum, HasNEON, HasFullFP16] in {
1451  def : Pat<(v4f16 (int_aarch64_neon_vcadd_rot90 (v4f16 V64:$Rn), (v4f16 V64:$Rm))),
1452            (FCADDv4f16 (v4f16 V64:$Rn), (v4f16 V64:$Rm), (i32 0))>;
1453  def : Pat<(v4f16 (int_aarch64_neon_vcadd_rot270 (v4f16 V64:$Rn), (v4f16 V64:$Rm))),
1454            (FCADDv4f16 (v4f16 V64:$Rn), (v4f16 V64:$Rm), (i32 1))>;
1455  def : Pat<(v8f16 (int_aarch64_neon_vcadd_rot90 (v8f16 V128:$Rn), (v8f16 V128:$Rm))),
1456            (FCADDv8f16 (v8f16 V128:$Rn), (v8f16 V128:$Rm), (i32 0))>;
1457  def : Pat<(v8f16 (int_aarch64_neon_vcadd_rot270 (v8f16 V128:$Rn), (v8f16 V128:$Rm))),
1458            (FCADDv8f16 (v8f16 V128:$Rn), (v8f16 V128:$Rm), (i32 1))>;
1459}
1460
1461let Predicates = [HasComplxNum, HasNEON] in {
1462  def : Pat<(v2f32 (int_aarch64_neon_vcadd_rot90 (v2f32 V64:$Rn), (v2f32 V64:$Rm))),
1463            (FCADDv2f32 (v2f32 V64:$Rn), (v2f32 V64:$Rm), (i32 0))>;
1464  def : Pat<(v2f32 (int_aarch64_neon_vcadd_rot270 (v2f32 V64:$Rn), (v2f32 V64:$Rm))),
1465            (FCADDv2f32 (v2f32 V64:$Rn), (v2f32 V64:$Rm), (i32 1))>;
1466  foreach Ty = [v4f32, v2f64] in {
1467    def : Pat<(Ty (int_aarch64_neon_vcadd_rot90 (Ty V128:$Rn), (Ty V128:$Rm))),
1468              (!cast<Instruction>("FCADD"#Ty) (Ty V128:$Rn), (Ty V128:$Rm), (i32 0))>;
1469    def : Pat<(Ty (int_aarch64_neon_vcadd_rot270 (Ty V128:$Rn), (Ty V128:$Rm))),
1470              (!cast<Instruction>("FCADD"#Ty) (Ty V128:$Rn), (Ty V128:$Rm), (i32 1))>;
1471  }
1472}
1473
1474multiclass FCMLA_PATS<ValueType ty, DAGOperand Reg> {
1475  def : Pat<(ty (int_aarch64_neon_vcmla_rot0 (ty Reg:$Rd), (ty Reg:$Rn), (ty Reg:$Rm))),
1476            (!cast<Instruction>("FCMLA" # ty) $Rd, $Rn, $Rm, 0)>;
1477  def : Pat<(ty (int_aarch64_neon_vcmla_rot90 (ty Reg:$Rd), (ty Reg:$Rn), (ty Reg:$Rm))),
1478            (!cast<Instruction>("FCMLA" # ty) $Rd, $Rn, $Rm, 1)>;
1479  def : Pat<(ty (int_aarch64_neon_vcmla_rot180 (ty Reg:$Rd), (ty Reg:$Rn), (ty Reg:$Rm))),
1480            (!cast<Instruction>("FCMLA" # ty) $Rd, $Rn, $Rm, 2)>;
1481  def : Pat<(ty (int_aarch64_neon_vcmla_rot270 (ty Reg:$Rd), (ty Reg:$Rn), (ty Reg:$Rm))),
1482            (!cast<Instruction>("FCMLA" # ty) $Rd, $Rn, $Rm, 3)>;
1483}
1484
1485multiclass FCMLA_LANE_PATS<ValueType ty, DAGOperand Reg, dag RHSDup> {
1486  def : Pat<(ty (int_aarch64_neon_vcmla_rot0 (ty Reg:$Rd), (ty Reg:$Rn), RHSDup)),
1487            (!cast<Instruction>("FCMLA" # ty # "_indexed") $Rd, $Rn, $Rm, VectorIndexS:$idx, 0)>;
1488  def : Pat<(ty (int_aarch64_neon_vcmla_rot90 (ty Reg:$Rd), (ty Reg:$Rn), RHSDup)),
1489            (!cast<Instruction>("FCMLA" # ty # "_indexed") $Rd, $Rn, $Rm, VectorIndexS:$idx, 1)>;
1490  def : Pat<(ty (int_aarch64_neon_vcmla_rot180 (ty Reg:$Rd), (ty Reg:$Rn), RHSDup)),
1491            (!cast<Instruction>("FCMLA" # ty # "_indexed") $Rd, $Rn, $Rm, VectorIndexS:$idx, 2)>;
1492  def : Pat<(ty (int_aarch64_neon_vcmla_rot270 (ty Reg:$Rd), (ty Reg:$Rn), RHSDup)),
1493            (!cast<Instruction>("FCMLA" # ty # "_indexed") $Rd, $Rn, $Rm, VectorIndexS:$idx, 3)>;
1494}
1495
1496
1497let Predicates = [HasComplxNum, HasNEON, HasFullFP16] in {
1498  defm : FCMLA_PATS<v4f16, V64>;
1499  defm : FCMLA_PATS<v8f16, V128>;
1500
1501  defm : FCMLA_LANE_PATS<v4f16, V64,
1502                         (v4f16 (bitconvert (v2i32 (AArch64duplane32 (v4i32 V128:$Rm), VectorIndexD:$idx))))>;
1503  defm : FCMLA_LANE_PATS<v8f16, V128,
1504                         (v8f16 (bitconvert (v4i32 (AArch64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))>;
1505}
1506let Predicates = [HasComplxNum, HasNEON] in {
1507  defm : FCMLA_PATS<v2f32, V64>;
1508  defm : FCMLA_PATS<v4f32, V128>;
1509  defm : FCMLA_PATS<v2f64, V128>;
1510
1511  defm : FCMLA_LANE_PATS<v4f32, V128,
1512                         (v4f32 (bitconvert (v2i64 (AArch64duplane64 (v2i64 V128:$Rm), VectorIndexD:$idx))))>;
1513}
1514
1515// v8.3a Pointer Authentication
1516// These instructions inhabit part of the hint space and so can be used for
1517// armv8 targets. Keeping the old HINT mnemonic when compiling without PA is
1518// important for compatibility with other assemblers (e.g. GAS) when building
1519// software compatible with both CPUs that do or don't implement PA.
1520let Uses = [LR], Defs = [LR] in {
1521  def PACIAZ   : SystemNoOperands<0b000, "hint\t#24">;
1522  def PACIBZ   : SystemNoOperands<0b010, "hint\t#26">;
1523  let isAuthenticated = 1 in {
1524    def AUTIAZ   : SystemNoOperands<0b100, "hint\t#28">;
1525    def AUTIBZ   : SystemNoOperands<0b110, "hint\t#30">;
1526  }
1527}
1528let Uses = [LR, SP], Defs = [LR] in {
1529  def PACIASP  : SystemNoOperands<0b001, "hint\t#25">;
1530  def PACIBSP  : SystemNoOperands<0b011, "hint\t#27">;
1531  let isAuthenticated = 1 in {
1532    def AUTIASP  : SystemNoOperands<0b101, "hint\t#29">;
1533    def AUTIBSP  : SystemNoOperands<0b111, "hint\t#31">;
1534  }
1535}
1536let Uses = [X16, X17], Defs = [X17], CRm = 0b0001 in {
1537  def PACIA1716  : SystemNoOperands<0b000, "hint\t#8">;
1538  def PACIB1716  : SystemNoOperands<0b010, "hint\t#10">;
1539  let isAuthenticated = 1 in {
1540    def AUTIA1716  : SystemNoOperands<0b100, "hint\t#12">;
1541    def AUTIB1716  : SystemNoOperands<0b110, "hint\t#14">;
1542  }
1543}
1544
1545let Uses = [LR], Defs = [LR], CRm = 0b0000 in {
1546  def XPACLRI   : SystemNoOperands<0b111, "hint\t#7">;
1547}
1548
1549// In order to be able to write readable assembly, LLVM should accept assembly
1550// inputs that use pointer authentication mnemonics, even with PA disabled.
1551// However, in order to be compatible with other assemblers (e.g. GAS), LLVM
1552// should not emit these mnemonics unless PA is enabled.
1553def : InstAlias<"paciaz", (PACIAZ), 0>;
1554def : InstAlias<"pacibz", (PACIBZ), 0>;
1555def : InstAlias<"autiaz", (AUTIAZ), 0>;
1556def : InstAlias<"autibz", (AUTIBZ), 0>;
1557def : InstAlias<"paciasp", (PACIASP), 0>;
1558def : InstAlias<"pacibsp", (PACIBSP), 0>;
1559def : InstAlias<"autiasp", (AUTIASP), 0>;
1560def : InstAlias<"autibsp", (AUTIBSP), 0>;
1561def : InstAlias<"pacia1716", (PACIA1716), 0>;
1562def : InstAlias<"pacib1716", (PACIB1716), 0>;
1563def : InstAlias<"autia1716", (AUTIA1716), 0>;
1564def : InstAlias<"autib1716", (AUTIB1716), 0>;
1565def : InstAlias<"xpaclri", (XPACLRI), 0>;
1566
1567// Pseudos
1568
1569let Uses = [LR, SP], Defs = [LR] in {
1570// Insertion point of LR signing code.
1571def PAUTH_PROLOGUE : Pseudo<(outs), (ins), []>, Sched<[]>;
1572// Insertion point of LR authentication code.
1573// The RET terminator of the containing machine basic block may be replaced
1574// with a combined RETA(A|B) instruction when rewriting this Pseudo.
1575def PAUTH_EPILOGUE : Pseudo<(outs), (ins), []>, Sched<[]>;
1576}
1577
1578// These pointer authentication instructions require armv8.3a
1579let Predicates = [HasPAuth] in {
1580
1581  // When PA is enabled, a better mnemonic should be emitted.
1582  def : InstAlias<"paciaz", (PACIAZ), 1>;
1583  def : InstAlias<"pacibz", (PACIBZ), 1>;
1584  def : InstAlias<"autiaz", (AUTIAZ), 1>;
1585  def : InstAlias<"autibz", (AUTIBZ), 1>;
1586  def : InstAlias<"paciasp", (PACIASP), 1>;
1587  def : InstAlias<"pacibsp", (PACIBSP), 1>;
1588  def : InstAlias<"autiasp", (AUTIASP), 1>;
1589  def : InstAlias<"autibsp", (AUTIBSP), 1>;
1590  def : InstAlias<"pacia1716", (PACIA1716), 1>;
1591  def : InstAlias<"pacib1716", (PACIB1716), 1>;
1592  def : InstAlias<"autia1716", (AUTIA1716), 1>;
1593  def : InstAlias<"autib1716", (AUTIB1716), 1>;
1594  def : InstAlias<"xpaclri", (XPACLRI), 1>;
1595
1596  multiclass SignAuth<bits<3> prefix, bits<3> prefix_z, string asm,
1597                      SDPatternOperator op> {
1598    def IA   : SignAuthOneData<prefix, 0b00, !strconcat(asm,  "ia"), op>;
1599    def IB   : SignAuthOneData<prefix, 0b01, !strconcat(asm,  "ib"), op>;
1600    def DA   : SignAuthOneData<prefix, 0b10, !strconcat(asm,  "da"), op>;
1601    def DB   : SignAuthOneData<prefix, 0b11, !strconcat(asm,  "db"), op>;
1602    def IZA  : SignAuthZero<prefix_z,  0b00, !strconcat(asm, "iza"), op>;
1603    def DZA  : SignAuthZero<prefix_z,  0b10, !strconcat(asm, "dza"), op>;
1604    def IZB  : SignAuthZero<prefix_z,  0b01, !strconcat(asm, "izb"), op>;
1605    def DZB  : SignAuthZero<prefix_z,  0b11, !strconcat(asm, "dzb"), op>;
1606  }
1607
1608  defm PAC : SignAuth<0b000, 0b010, "pac", int_ptrauth_sign>;
1609  defm AUT : SignAuth<0b001, 0b011, "aut", null_frag>;
1610
1611  def XPACI : ClearAuth<0, "xpaci">;
1612  def : Pat<(int_ptrauth_strip GPR64:$Rd, 0), (XPACI GPR64:$Rd)>;
1613  def : Pat<(int_ptrauth_strip GPR64:$Rd, 1), (XPACI GPR64:$Rd)>;
1614
1615  def XPACD : ClearAuth<1, "xpacd">;
1616  def : Pat<(int_ptrauth_strip GPR64:$Rd, 2), (XPACD GPR64:$Rd)>;
1617  def : Pat<(int_ptrauth_strip GPR64:$Rd, 3), (XPACD GPR64:$Rd)>;
1618
1619  def PACGA : SignAuthTwoOperand<0b1100, "pacga", int_ptrauth_sign_generic>;
1620
1621  // Combined Instructions
1622  let isBranch = 1, isTerminator = 1, isBarrier = 1, isIndirectBranch = 1  in {
1623    def BRAA    : AuthBranchTwoOperands<0, 0, "braa">;
1624    def BRAB    : AuthBranchTwoOperands<0, 1, "brab">;
1625  }
1626  let isCall = 1, Defs = [LR], Uses = [SP] in {
1627    def BLRAA   : AuthBranchTwoOperands<1, 0, "blraa">;
1628    def BLRAB   : AuthBranchTwoOperands<1, 1, "blrab">;
1629  }
1630
1631  let isBranch = 1, isTerminator = 1, isBarrier = 1, isIndirectBranch = 1  in {
1632    def BRAAZ   : AuthOneOperand<0b000, 0, "braaz">;
1633    def BRABZ   : AuthOneOperand<0b000, 1, "brabz">;
1634  }
1635  let isCall = 1, Defs = [LR], Uses = [SP] in {
1636    def BLRAAZ  : AuthOneOperand<0b001, 0, "blraaz">;
1637    def BLRABZ  : AuthOneOperand<0b001, 1, "blrabz">;
1638  }
1639
1640  let isReturn = 1, isTerminator = 1, isBarrier = 1 in {
1641    def RETAA   : AuthReturn<0b010, 0, "retaa">;
1642    def RETAB   : AuthReturn<0b010, 1, "retab">;
1643    def ERETAA  : AuthReturn<0b100, 0, "eretaa">;
1644    def ERETAB  : AuthReturn<0b100, 1, "eretab">;
1645  }
1646
1647  defm LDRAA  : AuthLoad<0, "ldraa", simm10Scaled>;
1648  defm LDRAB  : AuthLoad<1, "ldrab", simm10Scaled>;
1649
1650}
1651
1652// v9.5-A pointer authentication extensions
1653
1654// Always accept "pacm" as an alias for "hint #39", but don't emit it when
1655// disassembling if we don't have the pauth-lr feature.
1656let CRm = 0b0100 in {
1657  def PACM : SystemNoOperands<0b111, "hint\t#39">;
1658}
1659def : InstAlias<"pacm", (PACM), 0>;
1660
1661let Predicates = [HasPAuthLR] in {
1662  let Defs = [LR], Uses = [LR, SP] in {
1663    //                                opcode2, opcode,   asm
1664    def PACIASPPC : SignAuthFixedRegs<0b00001, 0b101000, "paciasppc">;
1665    def PACIBSPPC : SignAuthFixedRegs<0b00001, 0b101001, "pacibsppc">;
1666    def PACNBIASPPC : SignAuthFixedRegs<0b00001, 0b100000, "pacnbiasppc">;
1667    def PACNBIBSPPC : SignAuthFixedRegs<0b00001, 0b100001, "pacnbibsppc">;
1668    //                             opc,  asm
1669    def AUTIASPPCi : SignAuthPCRel<0b00, "autiasppc">;
1670    def AUTIBSPPCi : SignAuthPCRel<0b01, "autibsppc">;
1671    //                              opcode2, opcode,   asm
1672    def AUTIASPPCr : SignAuthOneReg<0b00001, 0b100100, "autiasppc">;
1673    def AUTIBSPPCr : SignAuthOneReg<0b00001, 0b100101, "autibsppc">;
1674    //                                  opcode2, opcode,   asm
1675    def PACIA171615 : SignAuthFixedRegs<0b00001, 0b100010, "pacia171615">;
1676    def PACIB171615 : SignAuthFixedRegs<0b00001, 0b100011, "pacib171615">;
1677    def AUTIA171615 : SignAuthFixedRegs<0b00001, 0b101110, "autia171615">;
1678    def AUTIB171615 : SignAuthFixedRegs<0b00001, 0b101111, "autib171615">;
1679  }
1680
1681  let Uses = [LR, SP], isReturn = 1, isTerminator = 1, isBarrier = 1 in {
1682    //                                   opc,   op2,     asm
1683    def RETAASPPCi : SignAuthReturnPCRel<0b000, 0b11111, "retaasppc">;
1684    def RETABSPPCi : SignAuthReturnPCRel<0b001, 0b11111, "retabsppc">;
1685    //                                 op3,      asm
1686    def RETAASPPCr : SignAuthReturnReg<0b000010, "retaasppc">;
1687    def RETABSPPCr : SignAuthReturnReg<0b000011, "retabsppc">;
1688  }
1689  def : InstAlias<"pacm", (PACM), 1>;
1690}
1691
1692
1693// v8.3a floating point conversion for javascript
1694let Predicates = [HasJS, HasFPARMv8], Defs = [NZCV] in
1695def FJCVTZS  : BaseFPToIntegerUnscaled<0b01, 0b11, 0b110, FPR64, GPR32,
1696                                      "fjcvtzs",
1697                                      [(set GPR32:$Rd,
1698                                         (int_aarch64_fjcvtzs FPR64:$Rn))]> {
1699  let Inst{31} = 0;
1700} // HasJS, HasFPARMv8
1701
1702// v8.4 Flag manipulation instructions
1703let Predicates = [HasFlagM], Defs = [NZCV], Uses = [NZCV] in {
1704def CFINV : SimpleSystemI<0, (ins), "cfinv", "">, Sched<[WriteSys]> {
1705  let Inst{20-5} = 0b0000001000000000;
1706}
1707def SETF8  : BaseFlagManipulation<0, 0, (ins GPR32:$Rn), "setf8", "{\t$Rn}">;
1708def SETF16 : BaseFlagManipulation<0, 1, (ins GPR32:$Rn), "setf16", "{\t$Rn}">;
1709def RMIF   : FlagRotate<(ins GPR64:$Rn, uimm6:$imm, imm0_15:$mask), "rmif",
1710                        "{\t$Rn, $imm, $mask}">;
1711} // HasFlagM
1712
1713// v8.5 flag manipulation instructions
1714let Predicates = [HasAltNZCV], Uses = [NZCV], Defs = [NZCV] in {
1715
1716def XAFLAG : PstateWriteSimple<(ins), "xaflag", "">, Sched<[WriteSys]> {
1717  let Inst{18-16} = 0b000;
1718  let Inst{11-8} = 0b0000;
1719  let Unpredictable{11-8} = 0b1111;
1720  let Inst{7-5} = 0b001;
1721}
1722
1723def AXFLAG : PstateWriteSimple<(ins), "axflag", "">, Sched<[WriteSys]> {
1724  let Inst{18-16} = 0b000;
1725  let Inst{11-8} = 0b0000;
1726  let Unpredictable{11-8} = 0b1111;
1727  let Inst{7-5} = 0b010;
1728}
1729} // HasAltNZCV
1730
1731
1732// Armv8.5-A speculation barrier
1733def SB : SimpleSystemI<0, (ins), "sb", "">, Sched<[]> {
1734  let Inst{20-5} = 0b0001100110000111;
1735  let Unpredictable{11-8} = 0b1111;
1736  let Predicates = [HasSB];
1737  let hasSideEffects = 1;
1738}
1739
1740def : InstAlias<"clrex", (CLREX 0xf)>;
1741def : InstAlias<"isb", (ISB 0xf)>;
1742def : InstAlias<"ssbb", (DSB 0)>;
1743def : InstAlias<"pssbb", (DSB 4)>;
1744def : InstAlias<"dfb", (DSB 0b1100)>, Requires<[HasV8_0r]>;
1745
1746def MRS    : MRSI;
1747def MSR    : MSRI;
1748def MSRpstateImm1 : MSRpstateImm0_1;
1749def MSRpstateImm4 : MSRpstateImm0_15;
1750
1751def : Pat<(AArch64mrs imm:$id),
1752          (MRS imm:$id)>;
1753
1754// The thread pointer (on Linux, at least, where this has been implemented) is
1755// TPIDR_EL0.
1756def MOVbaseTLS : Pseudo<(outs GPR64:$dst), (ins),
1757                       [(set GPR64:$dst, AArch64threadpointer)]>, Sched<[WriteSys]>;
1758
1759// This gets lowered into a 24-byte instruction sequence
1760let Defs = [ X9, X16, X17, NZCV ], Size = 24 in {
1761def KCFI_CHECK : Pseudo<
1762  (outs), (ins GPR64:$ptr, i32imm:$type), []>, Sched<[]>;
1763}
1764
1765let Uses = [ X9 ], Defs = [ X16, X17, LR, NZCV ] in {
1766def HWASAN_CHECK_MEMACCESS : Pseudo<
1767  (outs), (ins GPR64noip:$ptr, i32imm:$accessinfo),
1768  [(int_hwasan_check_memaccess X9, GPR64noip:$ptr, (i32 timm:$accessinfo))]>,
1769  Sched<[]>;
1770}
1771
1772let Uses = [ X20 ], Defs = [ X16, X17, LR, NZCV ] in {
1773def HWASAN_CHECK_MEMACCESS_SHORTGRANULES : Pseudo<
1774  (outs), (ins GPR64noip:$ptr, i32imm:$accessinfo),
1775  [(int_hwasan_check_memaccess_shortgranules X20, GPR64noip:$ptr, (i32 timm:$accessinfo))]>,
1776  Sched<[]>;
1777}
1778
1779// The virtual cycle counter register is CNTVCT_EL0.
1780def : Pat<(readcyclecounter), (MRS 0xdf02)>;
1781
1782// FPCR register
1783let Uses = [FPCR] in
1784def MRS_FPCR : Pseudo<(outs GPR64:$dst), (ins),
1785                      [(set GPR64:$dst, (int_aarch64_get_fpcr))]>,
1786               PseudoInstExpansion<(MRS GPR64:$dst, 0xda20)>,
1787               Sched<[WriteSys]>;
1788let Defs = [FPCR] in
1789def MSR_FPCR : Pseudo<(outs), (ins GPR64:$val),
1790                      [(int_aarch64_set_fpcr i64:$val)]>,
1791               PseudoInstExpansion<(MSR 0xda20, GPR64:$val)>,
1792               Sched<[WriteSys]>;
1793
1794// Generic system instructions
1795def SYSxt  : SystemXtI<0, "sys">;
1796def SYSLxt : SystemLXtI<1, "sysl">;
1797
1798def : InstAlias<"sys $op1, $Cn, $Cm, $op2",
1799                (SYSxt imm0_7:$op1, sys_cr_op:$Cn,
1800                 sys_cr_op:$Cm, imm0_7:$op2, XZR)>;
1801
1802
1803let Predicates = [HasTME] in {
1804
1805def TSTART : TMSystemI<0b0000, "tstart",
1806                      [(set GPR64:$Rt, (int_aarch64_tstart))]>;
1807
1808def TCOMMIT : TMSystemINoOperand<0b0000, "tcommit", [(int_aarch64_tcommit)]>;
1809
1810def TCANCEL : TMSystemException<0b011, "tcancel",
1811                                [(int_aarch64_tcancel timm64_0_65535:$imm)]>;
1812
1813def TTEST : TMSystemI<0b0001, "ttest", [(set GPR64:$Rt, (int_aarch64_ttest))]> {
1814  let mayLoad = 0;
1815  let mayStore = 0;
1816}
1817} // HasTME
1818
1819//===----------------------------------------------------------------------===//
1820// Move immediate instructions.
1821//===----------------------------------------------------------------------===//
1822
1823defm MOVK : InsertImmediate<0b11, "movk">;
1824defm MOVN : MoveImmediate<0b00, "movn">;
1825
1826let PostEncoderMethod = "fixMOVZ" in
1827defm MOVZ : MoveImmediate<0b10, "movz">;
1828
1829// First group of aliases covers an implicit "lsl #0".
1830def : InstAlias<"movk $dst, $imm", (MOVKWi GPR32:$dst, timm32_0_65535:$imm, 0), 0>;
1831def : InstAlias<"movk $dst, $imm", (MOVKXi GPR64:$dst, timm32_0_65535:$imm, 0), 0>;
1832def : InstAlias<"movn $dst, $imm", (MOVNWi GPR32:$dst, timm32_0_65535:$imm, 0)>;
1833def : InstAlias<"movn $dst, $imm", (MOVNXi GPR64:$dst, timm32_0_65535:$imm, 0)>;
1834def : InstAlias<"movz $dst, $imm", (MOVZWi GPR32:$dst, timm32_0_65535:$imm, 0)>;
1835def : InstAlias<"movz $dst, $imm", (MOVZXi GPR64:$dst, timm32_0_65535:$imm, 0)>;
1836
1837// Next, we have various ELF relocations with the ":XYZ_g0:sym" syntax.
1838def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movw_symbol_g3:$sym, 48)>;
1839def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movw_symbol_g2:$sym, 32)>;
1840def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movw_symbol_g1:$sym, 16)>;
1841def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movw_symbol_g0:$sym, 0)>;
1842
1843def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movw_symbol_g3:$sym, 48)>;
1844def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movw_symbol_g2:$sym, 32)>;
1845def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movw_symbol_g1:$sym, 16)>;
1846def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movw_symbol_g0:$sym, 0)>;
1847
1848def : InstAlias<"movk $Rd, $sym", (MOVKXi GPR64:$Rd, movw_symbol_g3:$sym, 48), 0>;
1849def : InstAlias<"movk $Rd, $sym", (MOVKXi GPR64:$Rd, movw_symbol_g2:$sym, 32), 0>;
1850def : InstAlias<"movk $Rd, $sym", (MOVKXi GPR64:$Rd, movw_symbol_g1:$sym, 16), 0>;
1851def : InstAlias<"movk $Rd, $sym", (MOVKXi GPR64:$Rd, movw_symbol_g0:$sym, 0), 0>;
1852
1853def : InstAlias<"movz $Rd, $sym", (MOVZWi GPR32:$Rd, movw_symbol_g1:$sym, 16)>;
1854def : InstAlias<"movz $Rd, $sym", (MOVZWi GPR32:$Rd, movw_symbol_g0:$sym, 0)>;
1855
1856def : InstAlias<"movn $Rd, $sym", (MOVNWi GPR32:$Rd, movw_symbol_g1:$sym, 16)>;
1857def : InstAlias<"movn $Rd, $sym", (MOVNWi GPR32:$Rd, movw_symbol_g0:$sym, 0)>;
1858
1859def : InstAlias<"movk $Rd, $sym", (MOVKWi GPR32:$Rd, movw_symbol_g1:$sym, 16), 0>;
1860def : InstAlias<"movk $Rd, $sym", (MOVKWi GPR32:$Rd, movw_symbol_g0:$sym, 0), 0>;
1861
1862// Final group of aliases covers true "mov $Rd, $imm" cases.
1863multiclass movw_mov_alias<string basename,Instruction INST, RegisterClass GPR,
1864                          int width, int shift> {
1865  def _asmoperand : AsmOperandClass {
1866    let Name = basename # width # "_lsl" # shift # "MovAlias";
1867    let PredicateMethod = "is" # basename # "MovAlias<" # width # ", "
1868                               # shift # ">";
1869    let RenderMethod = "add" # basename # "MovAliasOperands<" # shift # ">";
1870  }
1871
1872  def _movimm : Operand<i32> {
1873    let ParserMatchClass = !cast<AsmOperandClass>(NAME # "_asmoperand");
1874  }
1875
1876  def : InstAlias<"mov $Rd, $imm",
1877                  (INST GPR:$Rd, !cast<Operand>(NAME # "_movimm"):$imm, shift)>;
1878}
1879
1880defm : movw_mov_alias<"MOVZ", MOVZWi, GPR32, 32, 0>;
1881defm : movw_mov_alias<"MOVZ", MOVZWi, GPR32, 32, 16>;
1882
1883defm : movw_mov_alias<"MOVZ", MOVZXi, GPR64, 64, 0>;
1884defm : movw_mov_alias<"MOVZ", MOVZXi, GPR64, 64, 16>;
1885defm : movw_mov_alias<"MOVZ", MOVZXi, GPR64, 64, 32>;
1886defm : movw_mov_alias<"MOVZ", MOVZXi, GPR64, 64, 48>;
1887
1888defm : movw_mov_alias<"MOVN", MOVNWi, GPR32, 32, 0>;
1889defm : movw_mov_alias<"MOVN", MOVNWi, GPR32, 32, 16>;
1890
1891defm : movw_mov_alias<"MOVN", MOVNXi, GPR64, 64, 0>;
1892defm : movw_mov_alias<"MOVN", MOVNXi, GPR64, 64, 16>;
1893defm : movw_mov_alias<"MOVN", MOVNXi, GPR64, 64, 32>;
1894defm : movw_mov_alias<"MOVN", MOVNXi, GPR64, 64, 48>;
1895
1896let isReMaterializable = 1, isCodeGenOnly = 1, isMoveImm = 1,
1897    isAsCheapAsAMove = 1 in {
1898// FIXME: The following pseudo instructions are only needed because remat
1899// cannot handle multiple instructions.  When that changes, we can select
1900// directly to the real instructions and get rid of these pseudos.
1901
1902def MOVi32imm
1903    : Pseudo<(outs GPR32:$dst), (ins i32imm:$src),
1904             [(set GPR32:$dst, imm:$src)]>,
1905      Sched<[WriteImm]>;
1906def MOVi64imm
1907    : Pseudo<(outs GPR64:$dst), (ins i64imm:$src),
1908             [(set GPR64:$dst, imm:$src)]>,
1909      Sched<[WriteImm]>;
1910} // isReMaterializable, isCodeGenOnly
1911
1912// If possible, we want to use MOVi32imm even for 64-bit moves. This gives the
1913// eventual expansion code fewer bits to worry about getting right. Marshalling
1914// the types is a little tricky though:
1915def i64imm_32bit : ImmLeaf<i64, [{
1916  return (Imm & 0xffffffffULL) == static_cast<uint64_t>(Imm);
1917}]>;
1918
1919def s64imm_32bit : ImmLeaf<i64, [{
1920  int64_t Imm64 = static_cast<int64_t>(Imm);
1921  return Imm64 >= std::numeric_limits<int32_t>::min() &&
1922         Imm64 <= std::numeric_limits<int32_t>::max();
1923}]>;
1924
1925def trunc_imm : SDNodeXForm<imm, [{
1926  return CurDAG->getTargetConstant(N->getZExtValue(), SDLoc(N), MVT::i32);
1927}]>;
1928
1929def gi_trunc_imm : GICustomOperandRenderer<"renderTruncImm">,
1930  GISDNodeXFormEquiv<trunc_imm>;
1931
1932let Predicates = [OptimizedGISelOrOtherSelector] in {
1933// The SUBREG_TO_REG isn't eliminated at -O0, which can result in pointless
1934// copies.
1935def : Pat<(i64 i64imm_32bit:$src),
1936          (SUBREG_TO_REG (i64 0), (MOVi32imm (trunc_imm imm:$src)), sub_32)>;
1937}
1938
1939// Materialize FP constants via MOVi32imm/MOVi64imm (MachO large code model).
1940def bitcast_fpimm_to_i32 : SDNodeXForm<fpimm, [{
1941return CurDAG->getTargetConstant(
1942  N->getValueAPF().bitcastToAPInt().getZExtValue(), SDLoc(N), MVT::i32);
1943}]>;
1944
1945def bitcast_fpimm_to_i64 : SDNodeXForm<fpimm, [{
1946return CurDAG->getTargetConstant(
1947  N->getValueAPF().bitcastToAPInt().getZExtValue(), SDLoc(N), MVT::i64);
1948}]>;
1949
1950
1951def : Pat<(f32 fpimm:$in),
1952  (COPY_TO_REGCLASS (MOVi32imm (bitcast_fpimm_to_i32 f32:$in)), FPR32)>;
1953def : Pat<(f64 fpimm:$in),
1954  (COPY_TO_REGCLASS (MOVi64imm (bitcast_fpimm_to_i64 f64:$in)), FPR64)>;
1955
1956
1957// Deal with the various forms of (ELF) large addressing with MOVZ/MOVK
1958// sequences.
1959def : Pat<(AArch64WrapperLarge tglobaladdr:$g3, tglobaladdr:$g2,
1960                             tglobaladdr:$g1, tglobaladdr:$g0),
1961          (MOVKXi (MOVKXi (MOVKXi (MOVZXi tglobaladdr:$g0, 0),
1962                                  tglobaladdr:$g1, 16),
1963                          tglobaladdr:$g2, 32),
1964                  tglobaladdr:$g3, 48)>;
1965
1966def : Pat<(AArch64WrapperLarge tblockaddress:$g3, tblockaddress:$g2,
1967                             tblockaddress:$g1, tblockaddress:$g0),
1968          (MOVKXi (MOVKXi (MOVKXi (MOVZXi tblockaddress:$g0, 0),
1969                                  tblockaddress:$g1, 16),
1970                          tblockaddress:$g2, 32),
1971                  tblockaddress:$g3, 48)>;
1972
1973def : Pat<(AArch64WrapperLarge tconstpool:$g3, tconstpool:$g2,
1974                             tconstpool:$g1, tconstpool:$g0),
1975          (MOVKXi (MOVKXi (MOVKXi (MOVZXi tconstpool:$g0, 0),
1976                                  tconstpool:$g1, 16),
1977                          tconstpool:$g2, 32),
1978                  tconstpool:$g3, 48)>;
1979
1980def : Pat<(AArch64WrapperLarge tjumptable:$g3, tjumptable:$g2,
1981                             tjumptable:$g1, tjumptable:$g0),
1982          (MOVKXi (MOVKXi (MOVKXi (MOVZXi tjumptable:$g0, 0),
1983                                  tjumptable:$g1, 16),
1984                          tjumptable:$g2, 32),
1985                  tjumptable:$g3, 48)>;
1986
1987
1988//===----------------------------------------------------------------------===//
1989// Arithmetic instructions.
1990//===----------------------------------------------------------------------===//
1991
1992// Add/subtract with carry.
1993defm ADC : AddSubCarry<0, "adc", "adcs", AArch64adc, AArch64adc_flag>;
1994defm SBC : AddSubCarry<1, "sbc", "sbcs", AArch64sbc, AArch64sbc_flag>;
1995
1996def : InstAlias<"ngc $dst, $src",  (SBCWr  GPR32:$dst, WZR, GPR32:$src)>;
1997def : InstAlias<"ngc $dst, $src",  (SBCXr  GPR64:$dst, XZR, GPR64:$src)>;
1998def : InstAlias<"ngcs $dst, $src", (SBCSWr GPR32:$dst, WZR, GPR32:$src)>;
1999def : InstAlias<"ngcs $dst, $src", (SBCSXr GPR64:$dst, XZR, GPR64:$src)>;
2000
2001// Add/subtract
2002defm ADD : AddSub<0, "add", "sub", add>;
2003defm SUB : AddSub<1, "sub", "add">;
2004
2005def : InstAlias<"mov $dst, $src",
2006                (ADDWri GPR32sponly:$dst, GPR32sp:$src, 0, 0)>;
2007def : InstAlias<"mov $dst, $src",
2008                (ADDWri GPR32sp:$dst, GPR32sponly:$src, 0, 0)>;
2009def : InstAlias<"mov $dst, $src",
2010                (ADDXri GPR64sponly:$dst, GPR64sp:$src, 0, 0)>;
2011def : InstAlias<"mov $dst, $src",
2012                (ADDXri GPR64sp:$dst, GPR64sponly:$src, 0, 0)>;
2013
2014defm ADDS : AddSubS<0, "adds", AArch64add_flag, "cmn", "subs", "cmp">;
2015defm SUBS : AddSubS<1, "subs", AArch64sub_flag, "cmp", "adds", "cmn">;
2016
2017def copyFromSP: PatLeaf<(i64 GPR64:$src), [{
2018  return N->getOpcode() == ISD::CopyFromReg &&
2019         cast<RegisterSDNode>(N->getOperand(1))->getReg() == AArch64::SP;
2020}]>;
2021
2022// Use SUBS instead of SUB to enable CSE between SUBS and SUB.
2023def : Pat<(sub GPR32sp:$Rn, addsub_shifted_imm32:$imm),
2024          (SUBSWri GPR32sp:$Rn, addsub_shifted_imm32:$imm)>;
2025def : Pat<(sub GPR64sp:$Rn, addsub_shifted_imm64:$imm),
2026          (SUBSXri GPR64sp:$Rn, addsub_shifted_imm64:$imm)>;
2027def : Pat<(sub GPR32:$Rn, GPR32:$Rm),
2028          (SUBSWrr GPR32:$Rn, GPR32:$Rm)>;
2029def : Pat<(sub GPR64:$Rn, GPR64:$Rm),
2030          (SUBSXrr GPR64:$Rn, GPR64:$Rm)>;
2031def : Pat<(sub GPR32:$Rn, arith_shifted_reg32:$Rm),
2032          (SUBSWrs GPR32:$Rn, arith_shifted_reg32:$Rm)>;
2033def : Pat<(sub GPR64:$Rn, arith_shifted_reg64:$Rm),
2034          (SUBSXrs GPR64:$Rn, arith_shifted_reg64:$Rm)>;
2035let AddedComplexity = 1 in {
2036def : Pat<(sub GPR32sp:$R2, arith_extended_reg32_i32:$R3),
2037          (SUBSWrx GPR32sp:$R2, arith_extended_reg32_i32:$R3)>;
2038def : Pat<(sub GPR64sp:$R2, arith_extended_reg32to64_i64:$R3),
2039          (SUBSXrx GPR64sp:$R2, arith_extended_reg32to64_i64:$R3)>;
2040def : Pat<(sub copyFromSP:$R2, (arith_uxtx GPR64:$R3, arith_extendlsl64:$imm)),
2041          (SUBXrx64 GPR64sp:$R2, GPR64:$R3, arith_extendlsl64:$imm)>;
2042}
2043
2044// Because of the immediate format for add/sub-imm instructions, the
2045// expression (add x, -1) must be transformed to (SUB{W,X}ri x, 1).
2046//  These patterns capture that transformation.
2047let AddedComplexity = 1 in {
2048def : Pat<(add GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
2049          (SUBSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
2050def : Pat<(add GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
2051          (SUBSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
2052def : Pat<(sub GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
2053          (ADDWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
2054def : Pat<(sub GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
2055          (ADDXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
2056}
2057
2058// Because of the immediate format for add/sub-imm instructions, the
2059// expression (add x, -1) must be transformed to (SUB{W,X}ri x, 1).
2060//  These patterns capture that transformation.
2061let AddedComplexity = 1 in {
2062def : Pat<(AArch64add_flag GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
2063          (SUBSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
2064def : Pat<(AArch64add_flag GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
2065          (SUBSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
2066def : Pat<(AArch64sub_flag GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
2067          (ADDSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
2068def : Pat<(AArch64sub_flag GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
2069          (ADDSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
2070}
2071
2072def : InstAlias<"neg $dst, $src", (SUBWrs GPR32:$dst, WZR, GPR32:$src, 0), 3>;
2073def : InstAlias<"neg $dst, $src", (SUBXrs GPR64:$dst, XZR, GPR64:$src, 0), 3>;
2074def : InstAlias<"neg $dst, $src$shift",
2075                (SUBWrs GPR32:$dst, WZR, GPR32:$src, arith_shift32:$shift), 2>;
2076def : InstAlias<"neg $dst, $src$shift",
2077                (SUBXrs GPR64:$dst, XZR, GPR64:$src, arith_shift64:$shift), 2>;
2078
2079def : InstAlias<"negs $dst, $src", (SUBSWrs GPR32:$dst, WZR, GPR32:$src, 0), 3>;
2080def : InstAlias<"negs $dst, $src", (SUBSXrs GPR64:$dst, XZR, GPR64:$src, 0), 3>;
2081def : InstAlias<"negs $dst, $src$shift",
2082                (SUBSWrs GPR32:$dst, WZR, GPR32:$src, arith_shift32:$shift), 2>;
2083def : InstAlias<"negs $dst, $src$shift",
2084                (SUBSXrs GPR64:$dst, XZR, GPR64:$src, arith_shift64:$shift), 2>;
2085
2086
2087// Unsigned/Signed divide
2088defm UDIV : Div<0, "udiv", udiv>;
2089defm SDIV : Div<1, "sdiv", sdiv>;
2090
2091def : Pat<(int_aarch64_udiv GPR32:$Rn, GPR32:$Rm), (UDIVWr GPR32:$Rn, GPR32:$Rm)>;
2092def : Pat<(int_aarch64_udiv GPR64:$Rn, GPR64:$Rm), (UDIVXr GPR64:$Rn, GPR64:$Rm)>;
2093def : Pat<(int_aarch64_sdiv GPR32:$Rn, GPR32:$Rm), (SDIVWr GPR32:$Rn, GPR32:$Rm)>;
2094def : Pat<(int_aarch64_sdiv GPR64:$Rn, GPR64:$Rm), (SDIVXr GPR64:$Rn, GPR64:$Rm)>;
2095
2096// Variable shift
2097defm ASRV : Shift<0b10, "asr", sra>;
2098defm LSLV : Shift<0b00, "lsl", shl>;
2099defm LSRV : Shift<0b01, "lsr", srl>;
2100defm RORV : Shift<0b11, "ror", rotr>;
2101
2102def : ShiftAlias<"asrv", ASRVWr, GPR32>;
2103def : ShiftAlias<"asrv", ASRVXr, GPR64>;
2104def : ShiftAlias<"lslv", LSLVWr, GPR32>;
2105def : ShiftAlias<"lslv", LSLVXr, GPR64>;
2106def : ShiftAlias<"lsrv", LSRVWr, GPR32>;
2107def : ShiftAlias<"lsrv", LSRVXr, GPR64>;
2108def : ShiftAlias<"rorv", RORVWr, GPR32>;
2109def : ShiftAlias<"rorv", RORVXr, GPR64>;
2110
2111// Multiply-add
2112let AddedComplexity = 5 in {
2113defm MADD : MulAccum<0, "madd">;
2114defm MSUB : MulAccum<1, "msub">;
2115
2116def : Pat<(i32 (mul GPR32:$Rn, GPR32:$Rm)),
2117          (MADDWrrr GPR32:$Rn, GPR32:$Rm, WZR)>;
2118def : Pat<(i64 (mul GPR64:$Rn, GPR64:$Rm)),
2119          (MADDXrrr GPR64:$Rn, GPR64:$Rm, XZR)>;
2120
2121def : Pat<(i32 (ineg (mul GPR32:$Rn, GPR32:$Rm))),
2122          (MSUBWrrr GPR32:$Rn, GPR32:$Rm, WZR)>;
2123def : Pat<(i64 (ineg (mul GPR64:$Rn, GPR64:$Rm))),
2124          (MSUBXrrr GPR64:$Rn, GPR64:$Rm, XZR)>;
2125def : Pat<(i32 (mul (ineg GPR32:$Rn), GPR32:$Rm)),
2126          (MSUBWrrr GPR32:$Rn, GPR32:$Rm, WZR)>;
2127def : Pat<(i64 (mul (ineg GPR64:$Rn), GPR64:$Rm)),
2128          (MSUBXrrr GPR64:$Rn, GPR64:$Rm, XZR)>;
2129} // AddedComplexity = 5
2130
2131let AddedComplexity = 5 in {
2132def SMADDLrrr : WideMulAccum<0, 0b001, "smaddl", add, sext>;
2133def SMSUBLrrr : WideMulAccum<1, 0b001, "smsubl", sub, sext>;
2134def UMADDLrrr : WideMulAccum<0, 0b101, "umaddl", add, zext>;
2135def UMSUBLrrr : WideMulAccum<1, 0b101, "umsubl", sub, zext>;
2136
2137def : Pat<(i64 (mul (sext_inreg GPR64:$Rn, i32), (sext_inreg GPR64:$Rm, i32))),
2138          (SMADDLrrr (EXTRACT_SUBREG $Rn, sub_32), (EXTRACT_SUBREG $Rm, sub_32), XZR)>;
2139def : Pat<(i64 (mul (sext_inreg GPR64:$Rn, i32), (sext GPR32:$Rm))),
2140          (SMADDLrrr (EXTRACT_SUBREG $Rn, sub_32), $Rm, XZR)>;
2141def : Pat<(i64 (mul (sext GPR32:$Rn), (sext GPR32:$Rm))),
2142          (SMADDLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
2143def : Pat<(i64 (mul (and GPR64:$Rn, 0xFFFFFFFF), (and GPR64:$Rm, 0xFFFFFFFF))),
2144          (UMADDLrrr (EXTRACT_SUBREG $Rn, sub_32), (EXTRACT_SUBREG $Rm, sub_32), XZR)>;
2145def : Pat<(i64 (mul (and GPR64:$Rn, 0xFFFFFFFF), (zext GPR32:$Rm))),
2146          (UMADDLrrr (EXTRACT_SUBREG $Rn, sub_32), $Rm, XZR)>;
2147def : Pat<(i64 (mul (zext GPR32:$Rn), (zext GPR32:$Rm))),
2148          (UMADDLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
2149
2150def : Pat<(i64 (ineg (mul (sext GPR32:$Rn), (sext GPR32:$Rm)))),
2151          (SMSUBLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
2152def : Pat<(i64 (ineg (mul (zext GPR32:$Rn), (zext GPR32:$Rm)))),
2153          (UMSUBLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
2154
2155def : Pat<(i64 (mul (sext GPR32:$Rn), (s64imm_32bit:$C))),
2156          (SMADDLrrr GPR32:$Rn, (MOVi32imm (trunc_imm imm:$C)), XZR)>;
2157def : Pat<(i64 (mul (zext GPR32:$Rn), (i64imm_32bit:$C))),
2158          (UMADDLrrr GPR32:$Rn, (MOVi32imm (trunc_imm imm:$C)), XZR)>;
2159def : Pat<(i64 (mul (sext_inreg GPR64:$Rn, i32), (s64imm_32bit:$C))),
2160          (SMADDLrrr (i32 (EXTRACT_SUBREG GPR64:$Rn, sub_32)),
2161                     (MOVi32imm (trunc_imm imm:$C)), XZR)>;
2162
2163def : Pat<(i64 (ineg (mul (sext GPR32:$Rn), (s64imm_32bit:$C)))),
2164          (SMSUBLrrr GPR32:$Rn, (MOVi32imm (trunc_imm imm:$C)), XZR)>;
2165def : Pat<(i64 (ineg (mul (zext GPR32:$Rn), (i64imm_32bit:$C)))),
2166          (UMSUBLrrr GPR32:$Rn, (MOVi32imm (trunc_imm imm:$C)), XZR)>;
2167def : Pat<(i64 (ineg (mul (sext_inreg GPR64:$Rn, i32), (s64imm_32bit:$C)))),
2168          (SMSUBLrrr (i32 (EXTRACT_SUBREG GPR64:$Rn, sub_32)),
2169                     (MOVi32imm (trunc_imm imm:$C)), XZR)>;
2170
2171def : Pat<(i64 (add (mul (sext GPR32:$Rn), (s64imm_32bit:$C)), GPR64:$Ra)),
2172          (SMADDLrrr GPR32:$Rn, (MOVi32imm (trunc_imm imm:$C)), GPR64:$Ra)>;
2173def : Pat<(i64 (add (mul (zext GPR32:$Rn), (i64imm_32bit:$C)), GPR64:$Ra)),
2174          (UMADDLrrr GPR32:$Rn, (MOVi32imm (trunc_imm imm:$C)), GPR64:$Ra)>;
2175def : Pat<(i64 (add (mul (sext_inreg GPR64:$Rn, i32), (s64imm_32bit:$C)),
2176                    GPR64:$Ra)),
2177          (SMADDLrrr (i32 (EXTRACT_SUBREG GPR64:$Rn, sub_32)),
2178                     (MOVi32imm (trunc_imm imm:$C)), GPR64:$Ra)>;
2179
2180def : Pat<(i64 (sub GPR64:$Ra, (mul (sext GPR32:$Rn), (s64imm_32bit:$C)))),
2181          (SMSUBLrrr GPR32:$Rn, (MOVi32imm (trunc_imm imm:$C)), GPR64:$Ra)>;
2182def : Pat<(i64 (sub GPR64:$Ra, (mul (zext GPR32:$Rn), (i64imm_32bit:$C)))),
2183          (UMSUBLrrr GPR32:$Rn, (MOVi32imm (trunc_imm imm:$C)), GPR64:$Ra)>;
2184def : Pat<(i64 (sub GPR64:$Ra, (mul (sext_inreg GPR64:$Rn, i32),
2185                                    (s64imm_32bit:$C)))),
2186          (SMSUBLrrr (i32 (EXTRACT_SUBREG GPR64:$Rn, sub_32)),
2187                     (MOVi32imm (trunc_imm imm:$C)), GPR64:$Ra)>;
2188
2189def : Pat<(i64 (smullwithsignbits GPR64:$Rn, GPR64:$Rm)),
2190          (SMADDLrrr (EXTRACT_SUBREG $Rn, sub_32), (EXTRACT_SUBREG $Rm, sub_32), XZR)>;
2191def : Pat<(i64 (smullwithsignbits GPR64:$Rn, (sext GPR32:$Rm))),
2192          (SMADDLrrr (EXTRACT_SUBREG $Rn, sub_32), $Rm, XZR)>;
2193
2194def : Pat<(i64 (add (smullwithsignbits GPR64:$Rn, GPR64:$Rm), GPR64:$Ra)),
2195          (SMADDLrrr (EXTRACT_SUBREG $Rn, sub_32), (EXTRACT_SUBREG $Rm, sub_32), GPR64:$Ra)>;
2196def : Pat<(i64 (add (smullwithsignbits GPR64:$Rn, (sext GPR32:$Rm)), GPR64:$Ra)),
2197          (SMADDLrrr (EXTRACT_SUBREG $Rn, sub_32), $Rm, GPR64:$Ra)>;
2198
2199def : Pat<(i64 (ineg (smullwithsignbits GPR64:$Rn, GPR64:$Rm))),
2200          (SMSUBLrrr (EXTRACT_SUBREG $Rn, sub_32), (EXTRACT_SUBREG $Rm, sub_32), XZR)>;
2201def : Pat<(i64 (ineg (smullwithsignbits GPR64:$Rn, (sext GPR32:$Rm)))),
2202          (SMSUBLrrr (EXTRACT_SUBREG $Rn, sub_32), $Rm, XZR)>;
2203
2204def : Pat<(i64 (sub GPR64:$Ra, (smullwithsignbits GPR64:$Rn, GPR64:$Rm))),
2205          (SMSUBLrrr (EXTRACT_SUBREG $Rn, sub_32), (EXTRACT_SUBREG $Rm, sub_32), GPR64:$Ra)>;
2206def : Pat<(i64 (sub GPR64:$Ra, (smullwithsignbits GPR64:$Rn, (sext GPR32:$Rm)))),
2207          (SMSUBLrrr (EXTRACT_SUBREG $Rn, sub_32), $Rm, GPR64:$Ra)>;
2208
2209def : Pat<(i64 (mul top32Zero:$Rn, top32Zero:$Rm)),
2210          (UMADDLrrr (EXTRACT_SUBREG $Rn, sub_32), (EXTRACT_SUBREG $Rm, sub_32), XZR)>;
2211def : Pat<(i64 (mul top32Zero:$Rn, (zext GPR32:$Rm))),
2212          (UMADDLrrr (EXTRACT_SUBREG $Rn, sub_32), $Rm, XZR)>;
2213
2214def : Pat<(i64 (add (mul top32Zero:$Rn, top32Zero:$Rm), GPR64:$Ra)),
2215          (UMADDLrrr (EXTRACT_SUBREG $Rn, sub_32), (EXTRACT_SUBREG $Rm, sub_32), GPR64:$Ra)>;
2216def : Pat<(i64 (add (mul top32Zero:$Rn, (zext GPR32:$Rm)), GPR64:$Ra)),
2217          (UMADDLrrr (EXTRACT_SUBREG $Rn, sub_32), $Rm, GPR64:$Ra)>;
2218
2219def : Pat<(i64 (ineg (mul top32Zero:$Rn, top32Zero:$Rm))),
2220          (UMSUBLrrr (EXTRACT_SUBREG $Rn, sub_32), (EXTRACT_SUBREG $Rm, sub_32), XZR)>;
2221def : Pat<(i64 (ineg (mul top32Zero:$Rn, (zext GPR32:$Rm)))),
2222          (UMSUBLrrr (EXTRACT_SUBREG $Rn, sub_32), $Rm, XZR)>;
2223
2224def : Pat<(i64 (sub GPR64:$Ra, (mul top32Zero:$Rn, top32Zero:$Rm))),
2225          (UMSUBLrrr (EXTRACT_SUBREG $Rn, sub_32), (EXTRACT_SUBREG $Rm, sub_32), GPR64:$Ra)>;
2226def : Pat<(i64 (sub GPR64:$Ra, (mul top32Zero:$Rn, (zext GPR32:$Rm)))),
2227          (UMSUBLrrr (EXTRACT_SUBREG $Rn, sub_32), $Rm, GPR64:$Ra)>;
2228} // AddedComplexity = 5
2229
2230def : MulAccumWAlias<"mul", MADDWrrr>;
2231def : MulAccumXAlias<"mul", MADDXrrr>;
2232def : MulAccumWAlias<"mneg", MSUBWrrr>;
2233def : MulAccumXAlias<"mneg", MSUBXrrr>;
2234def : WideMulAccumAlias<"smull", SMADDLrrr>;
2235def : WideMulAccumAlias<"smnegl", SMSUBLrrr>;
2236def : WideMulAccumAlias<"umull", UMADDLrrr>;
2237def : WideMulAccumAlias<"umnegl", UMSUBLrrr>;
2238
2239// Multiply-high
2240def SMULHrr : MulHi<0b010, "smulh", mulhs>;
2241def UMULHrr : MulHi<0b110, "umulh", mulhu>;
2242
2243// CRC32
2244def CRC32Brr : BaseCRC32<0, 0b00, 0, GPR32, int_aarch64_crc32b, "crc32b">;
2245def CRC32Hrr : BaseCRC32<0, 0b01, 0, GPR32, int_aarch64_crc32h, "crc32h">;
2246def CRC32Wrr : BaseCRC32<0, 0b10, 0, GPR32, int_aarch64_crc32w, "crc32w">;
2247def CRC32Xrr : BaseCRC32<1, 0b11, 0, GPR64, int_aarch64_crc32x, "crc32x">;
2248
2249def CRC32CBrr : BaseCRC32<0, 0b00, 1, GPR32, int_aarch64_crc32cb, "crc32cb">;
2250def CRC32CHrr : BaseCRC32<0, 0b01, 1, GPR32, int_aarch64_crc32ch, "crc32ch">;
2251def CRC32CWrr : BaseCRC32<0, 0b10, 1, GPR32, int_aarch64_crc32cw, "crc32cw">;
2252def CRC32CXrr : BaseCRC32<1, 0b11, 1, GPR64, int_aarch64_crc32cx, "crc32cx">;
2253
2254// v8.1 atomic CAS
2255defm CAS   : CompareAndSwap<0, 0, "">;
2256defm CASA  : CompareAndSwap<1, 0, "a">;
2257defm CASL  : CompareAndSwap<0, 1, "l">;
2258defm CASAL : CompareAndSwap<1, 1, "al">;
2259
2260// v8.1 atomic CASP
2261defm CASP   : CompareAndSwapPair<0, 0, "">;
2262defm CASPA  : CompareAndSwapPair<1, 0, "a">;
2263defm CASPL  : CompareAndSwapPair<0, 1, "l">;
2264defm CASPAL : CompareAndSwapPair<1, 1, "al">;
2265
2266// v8.1 atomic SWP
2267defm SWP   : Swap<0, 0, "">;
2268defm SWPA  : Swap<1, 0, "a">;
2269defm SWPL  : Swap<0, 1, "l">;
2270defm SWPAL : Swap<1, 1, "al">;
2271
2272// v8.1 atomic LD<OP>(register). Performs load and then ST<OP>(register)
2273defm LDADD   : LDOPregister<0b000, "add", 0, 0, "">;
2274defm LDADDA  : LDOPregister<0b000, "add", 1, 0, "a">;
2275defm LDADDL  : LDOPregister<0b000, "add", 0, 1, "l">;
2276defm LDADDAL : LDOPregister<0b000, "add", 1, 1, "al">;
2277
2278defm LDCLR   : LDOPregister<0b001, "clr", 0, 0, "">;
2279defm LDCLRA  : LDOPregister<0b001, "clr", 1, 0, "a">;
2280defm LDCLRL  : LDOPregister<0b001, "clr", 0, 1, "l">;
2281defm LDCLRAL : LDOPregister<0b001, "clr", 1, 1, "al">;
2282
2283defm LDEOR   : LDOPregister<0b010, "eor", 0, 0, "">;
2284defm LDEORA  : LDOPregister<0b010, "eor", 1, 0, "a">;
2285defm LDEORL  : LDOPregister<0b010, "eor", 0, 1, "l">;
2286defm LDEORAL : LDOPregister<0b010, "eor", 1, 1, "al">;
2287
2288defm LDSET   : LDOPregister<0b011, "set", 0, 0, "">;
2289defm LDSETA  : LDOPregister<0b011, "set", 1, 0, "a">;
2290defm LDSETL  : LDOPregister<0b011, "set", 0, 1, "l">;
2291defm LDSETAL : LDOPregister<0b011, "set", 1, 1, "al">;
2292
2293defm LDSMAX   : LDOPregister<0b100, "smax", 0, 0, "">;
2294defm LDSMAXA  : LDOPregister<0b100, "smax", 1, 0, "a">;
2295defm LDSMAXL  : LDOPregister<0b100, "smax", 0, 1, "l">;
2296defm LDSMAXAL : LDOPregister<0b100, "smax", 1, 1, "al">;
2297
2298defm LDSMIN   : LDOPregister<0b101, "smin", 0, 0, "">;
2299defm LDSMINA  : LDOPregister<0b101, "smin", 1, 0, "a">;
2300defm LDSMINL  : LDOPregister<0b101, "smin", 0, 1, "l">;
2301defm LDSMINAL : LDOPregister<0b101, "smin", 1, 1, "al">;
2302
2303defm LDUMAX   : LDOPregister<0b110, "umax", 0, 0, "">;
2304defm LDUMAXA  : LDOPregister<0b110, "umax", 1, 0, "a">;
2305defm LDUMAXL  : LDOPregister<0b110, "umax", 0, 1, "l">;
2306defm LDUMAXAL : LDOPregister<0b110, "umax", 1, 1, "al">;
2307
2308defm LDUMIN   : LDOPregister<0b111, "umin", 0, 0, "">;
2309defm LDUMINA  : LDOPregister<0b111, "umin", 1, 0, "a">;
2310defm LDUMINL  : LDOPregister<0b111, "umin", 0, 1, "l">;
2311defm LDUMINAL : LDOPregister<0b111, "umin", 1, 1, "al">;
2312
2313// v8.1 atomic ST<OP>(register) as aliases to "LD<OP>(register) when Rt=xZR"
2314defm : STOPregister<"stadd","LDADD">; // STADDx
2315defm : STOPregister<"stclr","LDCLR">; // STCLRx
2316defm : STOPregister<"steor","LDEOR">; // STEORx
2317defm : STOPregister<"stset","LDSET">; // STSETx
2318defm : STOPregister<"stsmax","LDSMAX">;// STSMAXx
2319defm : STOPregister<"stsmin","LDSMIN">;// STSMINx
2320defm : STOPregister<"stumax","LDUMAX">;// STUMAXx
2321defm : STOPregister<"stumin","LDUMIN">;// STUMINx
2322
2323// v8.5 Memory Tagging Extension
2324let Predicates = [HasMTE] in {
2325
2326def IRG   : BaseTwoOperandRegReg<0b1, 0b0, 0b000100, GPR64sp, "irg",
2327                                 int_aarch64_irg, GPR64sp, GPR64>, Sched<[]>;
2328
2329def GMI   : BaseTwoOperandRegReg<0b1, 0b0, 0b000101, GPR64, "gmi",
2330                                 int_aarch64_gmi, GPR64sp>, Sched<[]> {
2331  let isNotDuplicable = 1;
2332}
2333def ADDG  : AddSubG<0, "addg", null_frag>;
2334def SUBG  : AddSubG<1, "subg", null_frag>;
2335
2336def : InstAlias<"irg $dst, $src", (IRG GPR64sp:$dst, GPR64sp:$src, XZR), 1>;
2337
2338def SUBP : SUBP<0, "subp", int_aarch64_subp>, Sched<[]>;
2339def SUBPS : SUBP<1, "subps", null_frag>, Sched<[]>{
2340  let Defs = [NZCV];
2341}
2342
2343def : InstAlias<"cmpp $lhs, $rhs", (SUBPS XZR, GPR64sp:$lhs, GPR64sp:$rhs), 0>;
2344
2345def LDG : MemTagLoad<"ldg", "\t$Rt, [$Rn, $offset]">;
2346
2347def : Pat<(int_aarch64_addg (am_indexedu6s128 GPR64sp:$Rn, uimm6s16:$imm6), imm0_15:$imm4),
2348          (ADDG GPR64sp:$Rn, imm0_63:$imm6, imm0_15:$imm4)>;
2349def : Pat<(int_aarch64_ldg GPR64:$Rt, (am_indexeds9s128 GPR64sp:$Rn,  simm9s16:$offset)),
2350          (LDG GPR64:$Rt, GPR64sp:$Rn,  simm9s16:$offset)>;
2351
2352def : InstAlias<"ldg $Rt, [$Rn]", (LDG GPR64:$Rt, GPR64sp:$Rn, 0), 1>;
2353
2354def LDGM : MemTagVector<1, "ldgm", "\t$Rt, [$Rn]",
2355                   (outs GPR64:$Rt), (ins GPR64sp:$Rn)>;
2356def STGM : MemTagVector<0, "stgm", "\t$Rt, [$Rn]",
2357                   (outs), (ins GPR64:$Rt, GPR64sp:$Rn)>;
2358def STZGM : MemTagVector<0, "stzgm", "\t$Rt, [$Rn]",
2359                   (outs), (ins GPR64:$Rt, GPR64sp:$Rn)> {
2360  let Inst{23} = 0;
2361}
2362
2363defm STG   : MemTagStore<0b00, "stg">;
2364defm STZG  : MemTagStore<0b01, "stzg">;
2365defm ST2G  : MemTagStore<0b10, "st2g">;
2366defm STZ2G : MemTagStore<0b11, "stz2g">;
2367
2368def : Pat<(AArch64stg GPR64sp:$Rn, (am_indexeds9s128 GPR64sp:$Rm, simm9s16:$imm)),
2369          (STGi $Rn, $Rm, $imm)>;
2370def : Pat<(AArch64stzg GPR64sp:$Rn, (am_indexeds9s128 GPR64sp:$Rm, simm9s16:$imm)),
2371          (STZGi $Rn, $Rm, $imm)>;
2372def : Pat<(AArch64st2g GPR64sp:$Rn, (am_indexeds9s128 GPR64sp:$Rm, simm9s16:$imm)),
2373          (ST2Gi $Rn, $Rm, $imm)>;
2374def : Pat<(AArch64stz2g GPR64sp:$Rn, (am_indexeds9s128 GPR64sp:$Rm, simm9s16:$imm)),
2375          (STZ2Gi $Rn, $Rm, $imm)>;
2376
2377defm STGP     : StorePairOffset <0b01, 0, GPR64z, simm7s16, "stgp">;
2378def  STGPpre  : StorePairPreIdx <0b01, 0, GPR64z, simm7s16, "stgp">;
2379def  STGPpost : StorePairPostIdx<0b01, 0, GPR64z, simm7s16, "stgp">;
2380
2381def : Pat<(int_aarch64_stg GPR64:$Rt, (am_indexeds9s128 GPR64sp:$Rn, simm9s16:$offset)),
2382          (STGi GPR64:$Rt, GPR64sp:$Rn,  simm9s16:$offset)>;
2383
2384def : Pat<(int_aarch64_stgp (am_indexed7s128 GPR64sp:$Rn, simm7s16:$imm), GPR64:$Rt, GPR64:$Rt2),
2385          (STGPi $Rt, $Rt2, $Rn, $imm)>;
2386
2387def IRGstack
2388    : Pseudo<(outs GPR64sp:$Rd), (ins GPR64sp:$Rsp, GPR64:$Rm), []>,
2389      Sched<[]>;
2390def TAGPstack
2391    : Pseudo<(outs GPR64sp:$Rd), (ins GPR64sp:$Rn, uimm6s16:$imm6, GPR64sp:$Rm, imm0_15:$imm4), []>,
2392      Sched<[]>;
2393
2394// Explicit SP in the first operand prevents ShrinkWrap optimization
2395// from leaving this instruction out of the stack frame. When IRGstack
2396// is transformed into IRG, this operand is replaced with the actual
2397// register / expression for the tagged base pointer of the current function.
2398def : Pat<(int_aarch64_irg_sp i64:$Rm), (IRGstack SP, i64:$Rm)>;
2399
2400// Large STG to be expanded into a loop. $sz is the size, $Rn is start address.
2401// $Rn_wback is one past the end of the range. $Rm is the loop counter.
2402let isCodeGenOnly=1, mayStore=1, Defs=[NZCV] in {
2403def STGloop_wback
2404    : Pseudo<(outs GPR64common:$Rm, GPR64sp:$Rn_wback), (ins i64imm:$sz, GPR64sp:$Rn),
2405             [], "$Rn = $Rn_wback,@earlyclobber $Rn_wback,@earlyclobber $Rm" >,
2406      Sched<[WriteAdr, WriteST]>;
2407
2408def STZGloop_wback
2409    : Pseudo<(outs GPR64common:$Rm, GPR64sp:$Rn_wback), (ins i64imm:$sz, GPR64sp:$Rn),
2410             [], "$Rn = $Rn_wback,@earlyclobber $Rn_wback,@earlyclobber $Rm" >,
2411      Sched<[WriteAdr, WriteST]>;
2412
2413// A variant of the above where $Rn2 is an independent register not tied to the input register $Rn.
2414// Their purpose is to use a FrameIndex operand as $Rn (which of course can not be written back).
2415def STGloop
2416    : Pseudo<(outs GPR64common:$Rm, GPR64sp:$Rn2), (ins i64imm:$sz, GPR64sp:$Rn),
2417             [], "@earlyclobber $Rn2,@earlyclobber $Rm" >,
2418      Sched<[WriteAdr, WriteST]>;
2419
2420def STZGloop
2421    : Pseudo<(outs GPR64common:$Rm, GPR64sp:$Rn2), (ins i64imm:$sz, GPR64sp:$Rn),
2422             [], "@earlyclobber $Rn2,@earlyclobber $Rm" >,
2423      Sched<[WriteAdr, WriteST]>;
2424}
2425
2426} // Predicates = [HasMTE]
2427
2428//===----------------------------------------------------------------------===//
2429// Logical instructions.
2430//===----------------------------------------------------------------------===//
2431
2432// (immediate)
2433defm ANDS : LogicalImmS<0b11, "ands", AArch64and_flag, "bics">;
2434defm AND  : LogicalImm<0b00, "and", and, "bic">;
2435defm EOR  : LogicalImm<0b10, "eor", xor, "eon">;
2436defm ORR  : LogicalImm<0b01, "orr", or, "orn">;
2437
2438// FIXME: these aliases *are* canonical sometimes (when movz can't be
2439// used). Actually, it seems to be working right now, but putting logical_immXX
2440// here is a bit dodgy on the AsmParser side too.
2441def : InstAlias<"mov $dst, $imm", (ORRWri GPR32sp:$dst, WZR,
2442                                          logical_imm32:$imm), 0>;
2443def : InstAlias<"mov $dst, $imm", (ORRXri GPR64sp:$dst, XZR,
2444                                          logical_imm64:$imm), 0>;
2445
2446
2447// (register)
2448defm ANDS : LogicalRegS<0b11, 0, "ands", AArch64and_flag>;
2449defm BICS : LogicalRegS<0b11, 1, "bics",
2450                        BinOpFrag<(AArch64and_flag node:$LHS, (not node:$RHS))>>;
2451defm AND  : LogicalReg<0b00, 0, "and", and>;
2452defm BIC  : LogicalReg<0b00, 1, "bic",
2453                       BinOpFrag<(and node:$LHS, (not node:$RHS))>, 3>;
2454defm EON  : LogicalReg<0b10, 1, "eon",
2455                       BinOpFrag<(not (xor node:$LHS, node:$RHS))>>;
2456defm EOR  : LogicalReg<0b10, 0, "eor", xor>;
2457defm ORN  : LogicalReg<0b01, 1, "orn",
2458                       BinOpFrag<(or node:$LHS, (not node:$RHS))>>;
2459defm ORR  : LogicalReg<0b01, 0, "orr", or>;
2460
2461def : InstAlias<"mov $dst, $src", (ORRWrs GPR32:$dst, WZR, GPR32:$src, 0), 2>;
2462def : InstAlias<"mov $dst, $src", (ORRXrs GPR64:$dst, XZR, GPR64:$src, 0), 2>;
2463
2464def : InstAlias<"mvn $Wd, $Wm", (ORNWrs GPR32:$Wd, WZR, GPR32:$Wm, 0), 3>;
2465def : InstAlias<"mvn $Xd, $Xm", (ORNXrs GPR64:$Xd, XZR, GPR64:$Xm, 0), 3>;
2466
2467def : InstAlias<"mvn $Wd, $Wm$sh",
2468                (ORNWrs GPR32:$Wd, WZR, GPR32:$Wm, logical_shift32:$sh), 2>;
2469def : InstAlias<"mvn $Xd, $Xm$sh",
2470                (ORNXrs GPR64:$Xd, XZR, GPR64:$Xm, logical_shift64:$sh), 2>;
2471
2472def : InstAlias<"tst $src1, $src2",
2473                (ANDSWri WZR, GPR32:$src1, logical_imm32:$src2), 2>;
2474def : InstAlias<"tst $src1, $src2",
2475                (ANDSXri XZR, GPR64:$src1, logical_imm64:$src2), 2>;
2476
2477def : InstAlias<"tst $src1, $src2",
2478                        (ANDSWrs WZR, GPR32:$src1, GPR32:$src2, 0), 3>;
2479def : InstAlias<"tst $src1, $src2",
2480                        (ANDSXrs XZR, GPR64:$src1, GPR64:$src2, 0), 3>;
2481
2482def : InstAlias<"tst $src1, $src2$sh",
2483               (ANDSWrs WZR, GPR32:$src1, GPR32:$src2, logical_shift32:$sh), 2>;
2484def : InstAlias<"tst $src1, $src2$sh",
2485               (ANDSXrs XZR, GPR64:$src1, GPR64:$src2, logical_shift64:$sh), 2>;
2486
2487
2488def : Pat<(not GPR32:$Wm), (ORNWrr WZR, GPR32:$Wm)>;
2489def : Pat<(not GPR64:$Xm), (ORNXrr XZR, GPR64:$Xm)>;
2490
2491
2492//===----------------------------------------------------------------------===//
2493// One operand data processing instructions.
2494//===----------------------------------------------------------------------===//
2495
2496defm CLS    : OneOperandData<0b000101, "cls">;
2497defm CLZ    : OneOperandData<0b000100, "clz", ctlz>;
2498defm RBIT   : OneOperandData<0b000000, "rbit", bitreverse>;
2499
2500def  REV16Wr : OneWRegData<0b000001, "rev16",
2501                                     UnOpFrag<(rotr (bswap node:$LHS), (i64 16))>>;
2502def  REV16Xr : OneXRegData<0b000001, "rev16", null_frag>;
2503
2504def : Pat<(cttz GPR32:$Rn),
2505          (CLZWr (RBITWr GPR32:$Rn))>;
2506def : Pat<(cttz GPR64:$Rn),
2507          (CLZXr (RBITXr GPR64:$Rn))>;
2508def : Pat<(ctlz (or (shl (xor (sra GPR32:$Rn, (i64 31)), GPR32:$Rn), (i64 1)),
2509                (i32 1))),
2510          (CLSWr GPR32:$Rn)>;
2511def : Pat<(ctlz (or (shl (xor (sra GPR64:$Rn, (i64 63)), GPR64:$Rn), (i64 1)),
2512                (i64 1))),
2513          (CLSXr GPR64:$Rn)>;
2514def : Pat<(int_aarch64_cls GPR32:$Rn), (CLSWr GPR32:$Rn)>;
2515def : Pat<(int_aarch64_cls64 GPR64:$Rm), (EXTRACT_SUBREG (CLSXr GPR64:$Rm), sub_32)>;
2516
2517// Unlike the other one operand instructions, the instructions with the "rev"
2518// mnemonic do *not* just different in the size bit, but actually use different
2519// opcode bits for the different sizes.
2520def REVWr   : OneWRegData<0b000010, "rev", bswap>;
2521def REVXr   : OneXRegData<0b000011, "rev", bswap>;
2522def REV32Xr : OneXRegData<0b000010, "rev32",
2523                                    UnOpFrag<(rotr (bswap node:$LHS), (i64 32))>>;
2524
2525def : InstAlias<"rev64 $Rd, $Rn", (REVXr GPR64:$Rd, GPR64:$Rn), 0>;
2526
2527// The bswap commutes with the rotr so we want a pattern for both possible
2528// orders.
2529def : Pat<(bswap (rotr GPR32:$Rn, (i64 16))), (REV16Wr GPR32:$Rn)>;
2530def : Pat<(bswap (rotr GPR64:$Rn, (i64 32))), (REV32Xr GPR64:$Rn)>;
2531
2532// Match (srl (bswap x), C) -> revC if the upper bswap bits are known zero.
2533def : Pat<(srl (bswap top16Zero:$Rn), (i64 16)), (REV16Wr GPR32:$Rn)>;
2534def : Pat<(srl (bswap top32Zero:$Rn), (i64 32)), (REV32Xr GPR64:$Rn)>;
2535
2536def : Pat<(or (and (srl GPR64:$Rn, (i64 8)), (i64 0x00ff00ff00ff00ff)),
2537              (and (shl GPR64:$Rn, (i64 8)), (i64 0xff00ff00ff00ff00))),
2538          (REV16Xr GPR64:$Rn)>;
2539
2540//===----------------------------------------------------------------------===//
2541// Bitfield immediate extraction instruction.
2542//===----------------------------------------------------------------------===//
2543let hasSideEffects = 0 in
2544defm EXTR : ExtractImm<"extr">;
2545def : InstAlias<"ror $dst, $src, $shift",
2546            (EXTRWrri GPR32:$dst, GPR32:$src, GPR32:$src, imm0_31:$shift)>;
2547def : InstAlias<"ror $dst, $src, $shift",
2548            (EXTRXrri GPR64:$dst, GPR64:$src, GPR64:$src, imm0_63:$shift)>;
2549
2550def : Pat<(rotr GPR32:$Rn, (i64 imm0_31:$imm)),
2551          (EXTRWrri GPR32:$Rn, GPR32:$Rn, imm0_31:$imm)>;
2552def : Pat<(rotr GPR64:$Rn, (i64 imm0_63:$imm)),
2553          (EXTRXrri GPR64:$Rn, GPR64:$Rn, imm0_63:$imm)>;
2554
2555//===----------------------------------------------------------------------===//
2556// Other bitfield immediate instructions.
2557//===----------------------------------------------------------------------===//
2558let hasSideEffects = 0 in {
2559defm BFM  : BitfieldImmWith2RegArgs<0b01, "bfm">;
2560defm SBFM : BitfieldImm<0b00, "sbfm">;
2561defm UBFM : BitfieldImm<0b10, "ubfm">;
2562}
2563
2564def i32shift_a : Operand<i64>, SDNodeXForm<imm, [{
2565  uint64_t enc = (32 - N->getZExtValue()) & 0x1f;
2566  return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
2567}]>;
2568
2569def i32shift_b : Operand<i64>, SDNodeXForm<imm, [{
2570  uint64_t enc = 31 - N->getZExtValue();
2571  return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
2572}]>;
2573
2574// min(7, 31 - shift_amt)
2575def i32shift_sext_i8 : Operand<i64>, SDNodeXForm<imm, [{
2576  uint64_t enc = 31 - N->getZExtValue();
2577  enc = enc > 7 ? 7 : enc;
2578  return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
2579}]>;
2580
2581// min(15, 31 - shift_amt)
2582def i32shift_sext_i16 : Operand<i64>, SDNodeXForm<imm, [{
2583  uint64_t enc = 31 - N->getZExtValue();
2584  enc = enc > 15 ? 15 : enc;
2585  return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
2586}]>;
2587
2588def i64shift_a : Operand<i64>, SDNodeXForm<imm, [{
2589  uint64_t enc = (64 - N->getZExtValue()) & 0x3f;
2590  return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
2591}]>;
2592
2593def i64shift_b : Operand<i64>, SDNodeXForm<imm, [{
2594  uint64_t enc = 63 - N->getZExtValue();
2595  return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
2596}]>;
2597
2598// min(7, 63 - shift_amt)
2599def i64shift_sext_i8 : Operand<i64>, SDNodeXForm<imm, [{
2600  uint64_t enc = 63 - N->getZExtValue();
2601  enc = enc > 7 ? 7 : enc;
2602  return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
2603}]>;
2604
2605// min(15, 63 - shift_amt)
2606def i64shift_sext_i16 : Operand<i64>, SDNodeXForm<imm, [{
2607  uint64_t enc = 63 - N->getZExtValue();
2608  enc = enc > 15 ? 15 : enc;
2609  return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
2610}]>;
2611
2612// min(31, 63 - shift_amt)
2613def i64shift_sext_i32 : Operand<i64>, SDNodeXForm<imm, [{
2614  uint64_t enc = 63 - N->getZExtValue();
2615  enc = enc > 31 ? 31 : enc;
2616  return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
2617}]>;
2618
2619def : Pat<(shl GPR32:$Rn, (i64 imm0_31:$imm)),
2620          (UBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
2621                              (i64 (i32shift_b imm0_31:$imm)))>;
2622def : Pat<(shl GPR64:$Rn, (i64 imm0_63:$imm)),
2623          (UBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
2624                              (i64 (i64shift_b imm0_63:$imm)))>;
2625
2626let AddedComplexity = 10 in {
2627def : Pat<(sra GPR32:$Rn, (i64 imm0_31:$imm)),
2628          (SBFMWri GPR32:$Rn, imm0_31:$imm, 31)>;
2629def : Pat<(sra GPR64:$Rn, (i64 imm0_63:$imm)),
2630          (SBFMXri GPR64:$Rn, imm0_63:$imm, 63)>;
2631}
2632
2633def : InstAlias<"asr $dst, $src, $shift",
2634                (SBFMWri GPR32:$dst, GPR32:$src, imm0_31:$shift, 31)>;
2635def : InstAlias<"asr $dst, $src, $shift",
2636                (SBFMXri GPR64:$dst, GPR64:$src, imm0_63:$shift, 63)>;
2637def : InstAlias<"sxtb $dst, $src", (SBFMWri GPR32:$dst, GPR32:$src, 0, 7)>;
2638def : InstAlias<"sxtb $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 7)>;
2639def : InstAlias<"sxth $dst, $src", (SBFMWri GPR32:$dst, GPR32:$src, 0, 15)>;
2640def : InstAlias<"sxth $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 15)>;
2641def : InstAlias<"sxtw $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 31)>;
2642
2643def : Pat<(srl GPR32:$Rn, (i64 imm0_31:$imm)),
2644          (UBFMWri GPR32:$Rn, imm0_31:$imm, 31)>;
2645def : Pat<(srl GPR64:$Rn, (i64 imm0_63:$imm)),
2646          (UBFMXri GPR64:$Rn, imm0_63:$imm, 63)>;
2647
2648def : InstAlias<"lsr $dst, $src, $shift",
2649                (UBFMWri GPR32:$dst, GPR32:$src, imm0_31:$shift, 31)>;
2650def : InstAlias<"lsr $dst, $src, $shift",
2651                (UBFMXri GPR64:$dst, GPR64:$src, imm0_63:$shift, 63)>;
2652def : InstAlias<"uxtb $dst, $src", (UBFMWri GPR32:$dst, GPR32:$src, 0, 7)>;
2653def : InstAlias<"uxtb $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 7)>;
2654def : InstAlias<"uxth $dst, $src", (UBFMWri GPR32:$dst, GPR32:$src, 0, 15)>;
2655def : InstAlias<"uxth $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 15)>;
2656def : InstAlias<"uxtw $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 31)>;
2657
2658//===----------------------------------------------------------------------===//
2659// Conditional comparison instructions.
2660//===----------------------------------------------------------------------===//
2661defm CCMN : CondComparison<0, "ccmn", AArch64ccmn>;
2662defm CCMP : CondComparison<1, "ccmp", AArch64ccmp>;
2663
2664//===----------------------------------------------------------------------===//
2665// Conditional select instructions.
2666//===----------------------------------------------------------------------===//
2667defm CSEL  : CondSelect<0, 0b00, "csel">;
2668
2669def inc : PatFrag<(ops node:$in), (add node:$in, 1)>;
2670defm CSINC : CondSelectOp<0, 0b01, "csinc", inc>;
2671defm CSINV : CondSelectOp<1, 0b00, "csinv", not>;
2672defm CSNEG : CondSelectOp<1, 0b01, "csneg", ineg>;
2673
2674def : Pat<(AArch64csinv GPR32:$tval, GPR32:$fval, (i32 imm:$cc), NZCV),
2675          (CSINVWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
2676def : Pat<(AArch64csinv GPR64:$tval, GPR64:$fval, (i32 imm:$cc), NZCV),
2677          (CSINVXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
2678def : Pat<(AArch64csneg GPR32:$tval, GPR32:$fval, (i32 imm:$cc), NZCV),
2679          (CSNEGWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
2680def : Pat<(AArch64csneg GPR64:$tval, GPR64:$fval, (i32 imm:$cc), NZCV),
2681          (CSNEGXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
2682def : Pat<(AArch64csinc GPR32:$tval, GPR32:$fval, (i32 imm:$cc), NZCV),
2683          (CSINCWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
2684def : Pat<(AArch64csinc GPR64:$tval, GPR64:$fval, (i32 imm:$cc), NZCV),
2685          (CSINCXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
2686
2687def : Pat<(AArch64csel (i32 0), (i32 1), (i32 imm:$cc), NZCV),
2688          (CSINCWr WZR, WZR, (i32 imm:$cc))>;
2689def : Pat<(AArch64csel (i64 0), (i64 1), (i32 imm:$cc), NZCV),
2690          (CSINCXr XZR, XZR, (i32 imm:$cc))>;
2691def : Pat<(AArch64csel GPR32:$tval, (i32 1), (i32 imm:$cc), NZCV),
2692          (CSINCWr GPR32:$tval, WZR, (i32 imm:$cc))>;
2693def : Pat<(AArch64csel GPR64:$tval, (i64 1), (i32 imm:$cc), NZCV),
2694          (CSINCXr GPR64:$tval, XZR, (i32 imm:$cc))>;
2695def : Pat<(AArch64csel (i32 1), GPR32:$fval, (i32 imm:$cc), NZCV),
2696          (CSINCWr GPR32:$fval, WZR, (i32 (inv_cond_XFORM imm:$cc)))>;
2697def : Pat<(AArch64csel (i64 1), GPR64:$fval, (i32 imm:$cc), NZCV),
2698          (CSINCXr GPR64:$fval, XZR, (i32 (inv_cond_XFORM imm:$cc)))>;
2699def : Pat<(AArch64csel (i32 0), (i32 -1), (i32 imm:$cc), NZCV),
2700          (CSINVWr WZR, WZR, (i32 imm:$cc))>;
2701def : Pat<(AArch64csel (i64 0), (i64 -1), (i32 imm:$cc), NZCV),
2702          (CSINVXr XZR, XZR, (i32 imm:$cc))>;
2703def : Pat<(AArch64csel GPR32:$tval, (i32 -1), (i32 imm:$cc), NZCV),
2704          (CSINVWr GPR32:$tval, WZR, (i32 imm:$cc))>;
2705def : Pat<(AArch64csel GPR64:$tval, (i64 -1), (i32 imm:$cc), NZCV),
2706          (CSINVXr GPR64:$tval, XZR, (i32 imm:$cc))>;
2707def : Pat<(AArch64csel (i32 -1), GPR32:$fval, (i32 imm:$cc), NZCV),
2708          (CSINVWr GPR32:$fval, WZR, (i32 (inv_cond_XFORM imm:$cc)))>;
2709def : Pat<(AArch64csel (i64 -1), GPR64:$fval, (i32 imm:$cc), NZCV),
2710          (CSINVXr GPR64:$fval, XZR, (i32 (inv_cond_XFORM imm:$cc)))>;
2711
2712def : Pat<(add GPR32:$val, (AArch64csel (i32 0), (i32 1), (i32 imm:$cc), NZCV)),
2713          (CSINCWr GPR32:$val, GPR32:$val, (i32 imm:$cc))>;
2714def : Pat<(add GPR64:$val, (zext (AArch64csel (i32 0), (i32 1), (i32 imm:$cc), NZCV))),
2715          (CSINCXr GPR64:$val, GPR64:$val, (i32 imm:$cc))>;
2716
2717def : Pat<(or (topbitsallzero32:$val), (AArch64csel (i32 0), (i32 1), (i32 imm:$cc), NZCV)),
2718          (CSINCWr GPR32:$val, WZR, imm:$cc)>;
2719def : Pat<(or (topbitsallzero64:$val), (AArch64csel (i64 0), (i64 1), (i32 imm:$cc), NZCV)),
2720          (CSINCXr GPR64:$val, XZR, imm:$cc)>;
2721def : Pat<(or (topbitsallzero64:$val), (zext (AArch64csel (i32 0), (i32 1), (i32 imm:$cc), NZCV))),
2722          (CSINCXr GPR64:$val, XZR, imm:$cc)>;
2723
2724def : Pat<(and (topbitsallzero32:$val), (AArch64csel (i32 0), (i32 1), (i32 imm:$cc), NZCV)),
2725          (CSELWr WZR, GPR32:$val, imm:$cc)>;
2726def : Pat<(and (topbitsallzero64:$val), (AArch64csel (i64 0), (i64 1), (i32 imm:$cc), NZCV)),
2727          (CSELXr XZR, GPR64:$val, imm:$cc)>;
2728def : Pat<(and (topbitsallzero64:$val), (zext (AArch64csel (i32 0), (i32 1), (i32 imm:$cc), NZCV))),
2729          (CSELXr XZR, GPR64:$val, imm:$cc)>;
2730
2731// The inverse of the condition code from the alias instruction is what is used
2732// in the aliased instruction. The parser all ready inverts the condition code
2733// for these aliases.
2734def : InstAlias<"cset $dst, $cc",
2735                (CSINCWr GPR32:$dst, WZR, WZR, inv_ccode:$cc)>;
2736def : InstAlias<"cset $dst, $cc",
2737                (CSINCXr GPR64:$dst, XZR, XZR, inv_ccode:$cc)>;
2738
2739def : InstAlias<"csetm $dst, $cc",
2740                (CSINVWr GPR32:$dst, WZR, WZR, inv_ccode:$cc)>;
2741def : InstAlias<"csetm $dst, $cc",
2742                (CSINVXr GPR64:$dst, XZR, XZR, inv_ccode:$cc)>;
2743
2744def : InstAlias<"cinc $dst, $src, $cc",
2745                (CSINCWr GPR32:$dst, GPR32:$src, GPR32:$src, inv_ccode:$cc)>;
2746def : InstAlias<"cinc $dst, $src, $cc",
2747                (CSINCXr GPR64:$dst, GPR64:$src, GPR64:$src, inv_ccode:$cc)>;
2748
2749def : InstAlias<"cinv $dst, $src, $cc",
2750                (CSINVWr GPR32:$dst, GPR32:$src, GPR32:$src, inv_ccode:$cc)>;
2751def : InstAlias<"cinv $dst, $src, $cc",
2752                (CSINVXr GPR64:$dst, GPR64:$src, GPR64:$src, inv_ccode:$cc)>;
2753
2754def : InstAlias<"cneg $dst, $src, $cc",
2755                (CSNEGWr GPR32:$dst, GPR32:$src, GPR32:$src, inv_ccode:$cc)>;
2756def : InstAlias<"cneg $dst, $src, $cc",
2757                (CSNEGXr GPR64:$dst, GPR64:$src, GPR64:$src, inv_ccode:$cc)>;
2758
2759//===----------------------------------------------------------------------===//
2760// PC-relative instructions.
2761//===----------------------------------------------------------------------===//
2762let isReMaterializable = 1 in {
2763let hasSideEffects = 0, mayStore = 0, mayLoad = 0 in {
2764def ADR  : ADRI<0, "adr", adrlabel,
2765                [(set GPR64:$Xd, (AArch64adr tglobaladdr:$label))]>;
2766} // hasSideEffects = 0
2767
2768def ADRP : ADRI<1, "adrp", adrplabel,
2769                [(set GPR64:$Xd, (AArch64adrp tglobaladdr:$label))]>;
2770} // isReMaterializable = 1
2771
2772// page address of a constant pool entry, block address
2773def : Pat<(AArch64adr tconstpool:$cp), (ADR tconstpool:$cp)>;
2774def : Pat<(AArch64adr tblockaddress:$cp), (ADR tblockaddress:$cp)>;
2775def : Pat<(AArch64adr texternalsym:$sym), (ADR texternalsym:$sym)>;
2776def : Pat<(AArch64adr tjumptable:$sym), (ADR tjumptable:$sym)>;
2777def : Pat<(AArch64adrp tconstpool:$cp), (ADRP tconstpool:$cp)>;
2778def : Pat<(AArch64adrp tblockaddress:$cp), (ADRP tblockaddress:$cp)>;
2779def : Pat<(AArch64adrp texternalsym:$sym), (ADRP texternalsym:$sym)>;
2780
2781//===----------------------------------------------------------------------===//
2782// Unconditional branch (register) instructions.
2783//===----------------------------------------------------------------------===//
2784
2785let isReturn = 1, isTerminator = 1, isBarrier = 1 in {
2786def RET  : BranchReg<0b0010, "ret", []>;
2787def DRPS : SpecialReturn<0b0101, "drps">;
2788def ERET : SpecialReturn<0b0100, "eret">;
2789} // isReturn = 1, isTerminator = 1, isBarrier = 1
2790
2791// Default to the LR register.
2792def : InstAlias<"ret", (RET LR)>;
2793
2794let isCall = 1, Defs = [LR], Uses = [SP] in {
2795  def BLR : BranchReg<0b0001, "blr", []>;
2796  def BLRNoIP : Pseudo<(outs), (ins GPR64noip:$Rn), []>,
2797                Sched<[WriteBrReg]>,
2798                PseudoInstExpansion<(BLR GPR64:$Rn)>;
2799  def BLR_RVMARKER : Pseudo<(outs), (ins variable_ops), []>,
2800                     Sched<[WriteBrReg]>;
2801  def BLR_BTI : Pseudo<(outs), (ins variable_ops), []>,
2802                Sched<[WriteBrReg]>;
2803} // isCall
2804
2805def : Pat<(AArch64call GPR64:$Rn),
2806          (BLR GPR64:$Rn)>,
2807      Requires<[NoSLSBLRMitigation]>;
2808def : Pat<(AArch64call GPR64noip:$Rn),
2809          (BLRNoIP GPR64noip:$Rn)>,
2810      Requires<[SLSBLRMitigation]>;
2811
2812def : Pat<(AArch64call_rvmarker (i64 tglobaladdr:$rvfunc), GPR64:$Rn),
2813          (BLR_RVMARKER tglobaladdr:$rvfunc, GPR64:$Rn)>,
2814      Requires<[NoSLSBLRMitigation]>;
2815
2816def : Pat<(AArch64call_bti GPR64:$Rn),
2817          (BLR_BTI GPR64:$Rn)>,
2818      Requires<[NoSLSBLRMitigation]>;
2819def : Pat<(AArch64call_bti GPR64noip:$Rn),
2820          (BLR_BTI GPR64noip:$Rn)>,
2821      Requires<[SLSBLRMitigation]>;
2822
2823let isBranch = 1, isTerminator = 1, isBarrier = 1, isIndirectBranch = 1 in {
2824def BR  : BranchReg<0b0000, "br", [(brind GPR64:$Rn)]>;
2825} // isBranch, isTerminator, isBarrier, isIndirectBranch
2826
2827// Create a separate pseudo-instruction for codegen to use so that we don't
2828// flag lr as used in every function. It'll be restored before the RET by the
2829// epilogue if it's legitimately used.
2830def RET_ReallyLR : Pseudo<(outs), (ins), [(AArch64retglue)]>,
2831                   Sched<[WriteBrReg]> {
2832  let isTerminator = 1;
2833  let isBarrier = 1;
2834  let isReturn = 1;
2835}
2836
2837// This is a directive-like pseudo-instruction. The purpose is to insert an
2838// R_AARCH64_TLSDESC_CALL relocation at the offset of the following instruction
2839// (which in the usual case is a BLR).
2840let hasSideEffects = 1 in
2841def TLSDESCCALL : Pseudo<(outs), (ins i64imm:$sym), []>, Sched<[]> {
2842  let AsmString = ".tlsdesccall $sym";
2843}
2844
2845// Pseudo instruction to tell the streamer to emit a 'B' character into the
2846// augmentation string.
2847def EMITBKEY : Pseudo<(outs), (ins), []>, Sched<[]> {}
2848
2849// Pseudo instruction to tell the streamer to emit a 'G' character into the
2850// augmentation string.
2851def EMITMTETAGGED : Pseudo<(outs), (ins), []>, Sched<[]> {}
2852
2853// FIXME: maybe the scratch register used shouldn't be fixed to X1?
2854// FIXME: can "hasSideEffects be dropped?
2855// This gets lowered to an instruction sequence which takes 16 bytes
2856let isCall = 1, Defs = [NZCV, LR, X0, X1], hasSideEffects = 1, Size = 16,
2857    isCodeGenOnly = 1 in
2858def TLSDESC_CALLSEQ
2859    : Pseudo<(outs), (ins i64imm:$sym),
2860             [(AArch64tlsdesc_callseq tglobaltlsaddr:$sym)]>,
2861      Sched<[WriteI, WriteLD, WriteI, WriteBrReg]>;
2862def : Pat<(AArch64tlsdesc_callseq texternalsym:$sym),
2863          (TLSDESC_CALLSEQ texternalsym:$sym)>;
2864
2865//===----------------------------------------------------------------------===//
2866// Conditional branch (immediate) instruction.
2867//===----------------------------------------------------------------------===//
2868def Bcc : BranchCond<0, "b">;
2869
2870// Armv8.8-A variant form which hints to the branch predictor that
2871// this branch is very likely to go the same way nearly all the time
2872// (even though it is not known at compile time _which_ way that is).
2873def BCcc : BranchCond<1, "bc">, Requires<[HasHBC]>;
2874
2875//===----------------------------------------------------------------------===//
2876// Compare-and-branch instructions.
2877//===----------------------------------------------------------------------===//
2878defm CBZ  : CmpBranch<0, "cbz", AArch64cbz>;
2879defm CBNZ : CmpBranch<1, "cbnz", AArch64cbnz>;
2880
2881//===----------------------------------------------------------------------===//
2882// Test-bit-and-branch instructions.
2883//===----------------------------------------------------------------------===//
2884defm TBZ  : TestBranch<0, "tbz", AArch64tbz>;
2885defm TBNZ : TestBranch<1, "tbnz", AArch64tbnz>;
2886
2887//===----------------------------------------------------------------------===//
2888// Unconditional branch (immediate) instructions.
2889//===----------------------------------------------------------------------===//
2890let isBranch = 1, isTerminator = 1, isBarrier = 1 in {
2891def B  : BranchImm<0, "b", [(br bb:$addr)]>;
2892} // isBranch, isTerminator, isBarrier
2893
2894let isCall = 1, Defs = [LR], Uses = [SP] in {
2895def BL : CallImm<1, "bl", [(AArch64call tglobaladdr:$addr)]>;
2896} // isCall
2897def : Pat<(AArch64call texternalsym:$func), (BL texternalsym:$func)>;
2898
2899//===----------------------------------------------------------------------===//
2900// Exception generation instructions.
2901//===----------------------------------------------------------------------===//
2902let isTrap = 1 in {
2903def BRK   : ExceptionGeneration<0b001, 0b00, "brk",
2904                                [(int_aarch64_break timm32_0_65535:$imm)]>;
2905}
2906def DCPS1 : ExceptionGeneration<0b101, 0b01, "dcps1">;
2907def DCPS2 : ExceptionGeneration<0b101, 0b10, "dcps2">;
2908def DCPS3 : ExceptionGeneration<0b101, 0b11, "dcps3">, Requires<[HasEL3]>;
2909def HLT   : ExceptionGeneration<0b010, 0b00, "hlt">;
2910def HVC   : ExceptionGeneration<0b000, 0b10, "hvc">;
2911def SMC   : ExceptionGeneration<0b000, 0b11, "smc">, Requires<[HasEL3]>;
2912def SVC   : ExceptionGeneration<0b000, 0b01, "svc">;
2913
2914// DCPSn defaults to an immediate operand of zero if unspecified.
2915def : InstAlias<"dcps1", (DCPS1 0)>;
2916def : InstAlias<"dcps2", (DCPS2 0)>;
2917def : InstAlias<"dcps3", (DCPS3 0)>, Requires<[HasEL3]>;
2918
2919def UDF : UDFType<0, "udf">;
2920
2921//===----------------------------------------------------------------------===//
2922// Load instructions.
2923//===----------------------------------------------------------------------===//
2924
2925// Pair (indexed, offset)
2926defm LDPW : LoadPairOffset<0b00, 0, GPR32z, simm7s4, "ldp">;
2927defm LDPX : LoadPairOffset<0b10, 0, GPR64z, simm7s8, "ldp">;
2928defm LDPS : LoadPairOffset<0b00, 1, FPR32Op, simm7s4, "ldp">;
2929defm LDPD : LoadPairOffset<0b01, 1, FPR64Op, simm7s8, "ldp">;
2930defm LDPQ : LoadPairOffset<0b10, 1, FPR128Op, simm7s16, "ldp">;
2931
2932defm LDPSW : LoadPairOffset<0b01, 0, GPR64z, simm7s4, "ldpsw">;
2933
2934// Pair (pre-indexed)
2935def LDPWpre : LoadPairPreIdx<0b00, 0, GPR32z, simm7s4, "ldp">;
2936def LDPXpre : LoadPairPreIdx<0b10, 0, GPR64z, simm7s8, "ldp">;
2937def LDPSpre : LoadPairPreIdx<0b00, 1, FPR32Op, simm7s4, "ldp">;
2938def LDPDpre : LoadPairPreIdx<0b01, 1, FPR64Op, simm7s8, "ldp">;
2939def LDPQpre : LoadPairPreIdx<0b10, 1, FPR128Op, simm7s16, "ldp">;
2940
2941def LDPSWpre : LoadPairPreIdx<0b01, 0, GPR64z, simm7s4, "ldpsw">;
2942
2943// Pair (post-indexed)
2944def LDPWpost : LoadPairPostIdx<0b00, 0, GPR32z, simm7s4, "ldp">;
2945def LDPXpost : LoadPairPostIdx<0b10, 0, GPR64z, simm7s8, "ldp">;
2946def LDPSpost : LoadPairPostIdx<0b00, 1, FPR32Op, simm7s4, "ldp">;
2947def LDPDpost : LoadPairPostIdx<0b01, 1, FPR64Op, simm7s8, "ldp">;
2948def LDPQpost : LoadPairPostIdx<0b10, 1, FPR128Op, simm7s16, "ldp">;
2949
2950def LDPSWpost : LoadPairPostIdx<0b01, 0, GPR64z, simm7s4, "ldpsw">;
2951
2952
2953// Pair (no allocate)
2954defm LDNPW : LoadPairNoAlloc<0b00, 0, GPR32z, simm7s4, "ldnp">;
2955defm LDNPX : LoadPairNoAlloc<0b10, 0, GPR64z, simm7s8, "ldnp">;
2956defm LDNPS : LoadPairNoAlloc<0b00, 1, FPR32Op, simm7s4, "ldnp">;
2957defm LDNPD : LoadPairNoAlloc<0b01, 1, FPR64Op, simm7s8, "ldnp">;
2958defm LDNPQ : LoadPairNoAlloc<0b10, 1, FPR128Op, simm7s16, "ldnp">;
2959
2960def : Pat<(AArch64ldp (am_indexed7s64 GPR64sp:$Rn, simm7s8:$offset)),
2961          (LDPXi GPR64sp:$Rn, simm7s8:$offset)>;
2962
2963def : Pat<(AArch64ldnp (am_indexed7s128 GPR64sp:$Rn, simm7s16:$offset)),
2964          (LDNPQi GPR64sp:$Rn, simm7s16:$offset)>;
2965//---
2966// (register offset)
2967//---
2968
2969// Integer
2970defm LDRBB : Load8RO<0b00,  0, 0b01, GPR32, "ldrb", i32, zextloadi8>;
2971defm LDRHH : Load16RO<0b01, 0, 0b01, GPR32, "ldrh", i32, zextloadi16>;
2972defm LDRW  : Load32RO<0b10, 0, 0b01, GPR32, "ldr", i32, load>;
2973defm LDRX  : Load64RO<0b11, 0, 0b01, GPR64, "ldr", i64, load>;
2974
2975// Floating-point
2976defm LDRB : Load8RO<0b00,   1, 0b01, FPR8Op,   "ldr", i8, load>;
2977defm LDRH : Load16RO<0b01,  1, 0b01, FPR16Op,  "ldr", f16, load>;
2978defm LDRS : Load32RO<0b10,  1, 0b01, FPR32Op,  "ldr", f32, load>;
2979defm LDRD : Load64RO<0b11,  1, 0b01, FPR64Op,  "ldr", f64, load>;
2980defm LDRQ : Load128RO<0b00, 1, 0b11, FPR128Op, "ldr", f128, load>;
2981
2982// Load sign-extended half-word
2983defm LDRSHW : Load16RO<0b01, 0, 0b11, GPR32, "ldrsh", i32, sextloadi16>;
2984defm LDRSHX : Load16RO<0b01, 0, 0b10, GPR64, "ldrsh", i64, sextloadi16>;
2985
2986// Load sign-extended byte
2987defm LDRSBW : Load8RO<0b00, 0, 0b11, GPR32, "ldrsb", i32, sextloadi8>;
2988defm LDRSBX : Load8RO<0b00, 0, 0b10, GPR64, "ldrsb", i64, sextloadi8>;
2989
2990// Load sign-extended word
2991defm LDRSW  : Load32RO<0b10, 0, 0b10, GPR64, "ldrsw", i64, sextloadi32>;
2992
2993// Pre-fetch.
2994defm PRFM : PrefetchRO<0b11, 0, 0b10, "prfm">;
2995
2996// For regular load, we do not have any alignment requirement.
2997// Thus, it is safe to directly map the vector loads with interesting
2998// addressing modes.
2999// FIXME: We could do the same for bitconvert to floating point vectors.
3000multiclass ScalToVecROLoadPat<ROAddrMode ro, SDPatternOperator loadop,
3001                              ValueType ScalTy, ValueType VecTy,
3002                              Instruction LOADW, Instruction LOADX,
3003                              SubRegIndex sub> {
3004  def : Pat<(VecTy (scalar_to_vector (ScalTy
3005              (loadop (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$offset))))),
3006            (INSERT_SUBREG (VecTy (IMPLICIT_DEF)),
3007                           (LOADW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$offset),
3008                           sub)>;
3009
3010  def : Pat<(VecTy (scalar_to_vector (ScalTy
3011              (loadop (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$offset))))),
3012            (INSERT_SUBREG (VecTy (IMPLICIT_DEF)),
3013                           (LOADX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$offset),
3014                           sub)>;
3015}
3016
3017let AddedComplexity = 10 in {
3018defm : ScalToVecROLoadPat<ro8,  extloadi8,  i32, v8i8,  LDRBroW, LDRBroX, bsub>;
3019defm : ScalToVecROLoadPat<ro8,  extloadi8,  i32, v16i8, LDRBroW, LDRBroX, bsub>;
3020
3021defm : ScalToVecROLoadPat<ro16, extloadi16, i32, v4i16, LDRHroW, LDRHroX, hsub>;
3022defm : ScalToVecROLoadPat<ro16, extloadi16, i32, v8i16, LDRHroW, LDRHroX, hsub>;
3023
3024defm : ScalToVecROLoadPat<ro16, load,       i32, v4f16, LDRHroW, LDRHroX, hsub>;
3025defm : ScalToVecROLoadPat<ro16, load,       i32, v8f16, LDRHroW, LDRHroX, hsub>;
3026
3027defm : ScalToVecROLoadPat<ro32, load,       i32, v2i32, LDRSroW, LDRSroX, ssub>;
3028defm : ScalToVecROLoadPat<ro32, load,       i32, v4i32, LDRSroW, LDRSroX, ssub>;
3029
3030defm : ScalToVecROLoadPat<ro32, load,       f32, v2f32, LDRSroW, LDRSroX, ssub>;
3031defm : ScalToVecROLoadPat<ro32, load,       f32, v4f32, LDRSroW, LDRSroX, ssub>;
3032
3033defm : ScalToVecROLoadPat<ro64, load,       i64, v2i64, LDRDroW, LDRDroX, dsub>;
3034
3035defm : ScalToVecROLoadPat<ro64, load,       f64, v2f64, LDRDroW, LDRDroX, dsub>;
3036
3037
3038def : Pat <(v1i64 (scalar_to_vector (i64
3039                      (load (ro_Windexed64 GPR64sp:$Rn, GPR32:$Rm,
3040                                           ro_Wextend64:$extend))))),
3041           (LDRDroW GPR64sp:$Rn, GPR32:$Rm, ro_Wextend64:$extend)>;
3042
3043def : Pat <(v1i64 (scalar_to_vector (i64
3044                      (load (ro_Xindexed64 GPR64sp:$Rn, GPR64:$Rm,
3045                                           ro_Xextend64:$extend))))),
3046           (LDRDroX GPR64sp:$Rn, GPR64:$Rm, ro_Xextend64:$extend)>;
3047}
3048
3049// Match all load 64 bits width whose type is compatible with FPR64
3050multiclass VecROLoadPat<ROAddrMode ro, ValueType VecTy,
3051                        Instruction LOADW, Instruction LOADX> {
3052
3053  def : Pat<(VecTy (load (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend))),
3054            (LOADW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
3055
3056  def : Pat<(VecTy (load (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend))),
3057            (LOADX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
3058}
3059
3060let AddedComplexity = 10 in {
3061let Predicates = [IsLE] in {
3062  // We must do vector loads with LD1 in big-endian.
3063  defm : VecROLoadPat<ro64, v2i32, LDRDroW, LDRDroX>;
3064  defm : VecROLoadPat<ro64, v2f32, LDRDroW, LDRDroX>;
3065  defm : VecROLoadPat<ro64, v8i8,  LDRDroW, LDRDroX>;
3066  defm : VecROLoadPat<ro64, v4i16, LDRDroW, LDRDroX>;
3067  defm : VecROLoadPat<ro64, v4f16, LDRDroW, LDRDroX>;
3068  defm : VecROLoadPat<ro64, v4bf16, LDRDroW, LDRDroX>;
3069}
3070
3071defm : VecROLoadPat<ro64, v1i64,  LDRDroW, LDRDroX>;
3072defm : VecROLoadPat<ro64, v1f64,  LDRDroW, LDRDroX>;
3073
3074// Match all load 128 bits width whose type is compatible with FPR128
3075let Predicates = [IsLE] in {
3076  // We must do vector loads with LD1 in big-endian.
3077  defm : VecROLoadPat<ro128, v2i64,  LDRQroW, LDRQroX>;
3078  defm : VecROLoadPat<ro128, v2f64,  LDRQroW, LDRQroX>;
3079  defm : VecROLoadPat<ro128, v4i32,  LDRQroW, LDRQroX>;
3080  defm : VecROLoadPat<ro128, v4f32,  LDRQroW, LDRQroX>;
3081  defm : VecROLoadPat<ro128, v8i16,  LDRQroW, LDRQroX>;
3082  defm : VecROLoadPat<ro128, v8f16,  LDRQroW, LDRQroX>;
3083  defm : VecROLoadPat<ro128, v8bf16,  LDRQroW, LDRQroX>;
3084  defm : VecROLoadPat<ro128, v16i8,  LDRQroW, LDRQroX>;
3085}
3086} // AddedComplexity = 10
3087
3088// zextload -> i64
3089multiclass ExtLoadTo64ROPat<ROAddrMode ro, SDPatternOperator loadop,
3090                            Instruction INSTW, Instruction INSTX> {
3091  def : Pat<(i64 (loadop (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend))),
3092            (SUBREG_TO_REG (i64 0),
3093                           (INSTW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend),
3094                           sub_32)>;
3095
3096  def : Pat<(i64 (loadop (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend))),
3097            (SUBREG_TO_REG (i64 0),
3098                           (INSTX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend),
3099                           sub_32)>;
3100}
3101
3102let AddedComplexity = 10 in {
3103  defm : ExtLoadTo64ROPat<ro8,  zextloadi8,  LDRBBroW, LDRBBroX>;
3104  defm : ExtLoadTo64ROPat<ro16, zextloadi16, LDRHHroW, LDRHHroX>;
3105  defm : ExtLoadTo64ROPat<ro32, zextloadi32, LDRWroW,  LDRWroX>;
3106
3107  // zextloadi1 -> zextloadi8
3108  defm : ExtLoadTo64ROPat<ro8,  zextloadi1,  LDRBBroW, LDRBBroX>;
3109
3110  // extload -> zextload
3111  defm : ExtLoadTo64ROPat<ro8,  extloadi8,   LDRBBroW, LDRBBroX>;
3112  defm : ExtLoadTo64ROPat<ro16, extloadi16,  LDRHHroW, LDRHHroX>;
3113  defm : ExtLoadTo64ROPat<ro32, extloadi32,  LDRWroW,  LDRWroX>;
3114
3115  // extloadi1 -> zextloadi8
3116  defm : ExtLoadTo64ROPat<ro8,  extloadi1,   LDRBBroW, LDRBBroX>;
3117}
3118
3119
3120// zextload -> i64
3121multiclass ExtLoadTo32ROPat<ROAddrMode ro, SDPatternOperator loadop,
3122                            Instruction INSTW, Instruction INSTX> {
3123  def : Pat<(i32 (loadop (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend))),
3124            (INSTW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
3125
3126  def : Pat<(i32 (loadop (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend))),
3127            (INSTX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
3128
3129}
3130
3131let AddedComplexity = 10 in {
3132  // extload -> zextload
3133  defm : ExtLoadTo32ROPat<ro8,  extloadi8,   LDRBBroW, LDRBBroX>;
3134  defm : ExtLoadTo32ROPat<ro16, extloadi16,  LDRHHroW, LDRHHroX>;
3135  defm : ExtLoadTo32ROPat<ro32, extloadi32,  LDRWroW,  LDRWroX>;
3136
3137  // zextloadi1 -> zextloadi8
3138  defm : ExtLoadTo32ROPat<ro8, zextloadi1, LDRBBroW, LDRBBroX>;
3139}
3140
3141//---
3142// (unsigned immediate)
3143//---
3144defm LDRX : LoadUI<0b11, 0, 0b01, GPR64z, uimm12s8, "ldr",
3145                   [(set GPR64z:$Rt,
3146                         (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)))]>;
3147defm LDRW : LoadUI<0b10, 0, 0b01, GPR32z, uimm12s4, "ldr",
3148                   [(set GPR32z:$Rt,
3149                         (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset)))]>;
3150defm LDRB : LoadUI<0b00, 1, 0b01, FPR8Op, uimm12s1, "ldr",
3151                   [(set FPR8Op:$Rt,
3152                         (load (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset)))]>;
3153defm LDRH : LoadUI<0b01, 1, 0b01, FPR16Op, uimm12s2, "ldr",
3154                   [(set (f16 FPR16Op:$Rt),
3155                         (load (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset)))]>;
3156defm LDRS : LoadUI<0b10, 1, 0b01, FPR32Op, uimm12s4, "ldr",
3157                   [(set (f32 FPR32Op:$Rt),
3158                         (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset)))]>;
3159defm LDRD : LoadUI<0b11, 1, 0b01, FPR64Op, uimm12s8, "ldr",
3160                   [(set (f64 FPR64Op:$Rt),
3161                         (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)))]>;
3162defm LDRQ : LoadUI<0b00, 1, 0b11, FPR128Op, uimm12s16, "ldr",
3163                 [(set (f128 FPR128Op:$Rt),
3164                       (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)))]>;
3165
3166// bf16 load pattern
3167def : Pat <(bf16 (load (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))),
3168           (LDRHui GPR64sp:$Rn, uimm12s2:$offset)>;
3169
3170// For regular load, we do not have any alignment requirement.
3171// Thus, it is safe to directly map the vector loads with interesting
3172// addressing modes.
3173// FIXME: We could do the same for bitconvert to floating point vectors.
3174def : Pat <(v8i8 (scalar_to_vector (i32
3175               (extloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))))),
3176           (INSERT_SUBREG (v8i8 (IMPLICIT_DEF)),
3177                          (LDRBui GPR64sp:$Rn, uimm12s1:$offset), bsub)>;
3178def : Pat <(v16i8 (scalar_to_vector (i32
3179               (extloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))))),
3180           (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3181                          (LDRBui GPR64sp:$Rn, uimm12s1:$offset), bsub)>;
3182def : Pat <(v4i16 (scalar_to_vector (i32
3183               (extloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))))),
3184           (INSERT_SUBREG (v4i16 (IMPLICIT_DEF)),
3185                          (LDRHui GPR64sp:$Rn, uimm12s2:$offset), hsub)>;
3186def : Pat <(v8i16 (scalar_to_vector (i32
3187               (extloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))))),
3188           (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
3189                          (LDRHui GPR64sp:$Rn, uimm12s2:$offset), hsub)>;
3190def : Pat <(v2i32 (scalar_to_vector (i32
3191               (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))))),
3192           (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)),
3193                          (LDRSui GPR64sp:$Rn, uimm12s4:$offset), ssub)>;
3194def : Pat <(v4i32 (scalar_to_vector (i32
3195               (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))))),
3196           (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
3197                          (LDRSui GPR64sp:$Rn, uimm12s4:$offset), ssub)>;
3198def : Pat <(v1i64 (scalar_to_vector (i64
3199               (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))))),
3200           (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
3201def : Pat <(v2i64 (scalar_to_vector (i64
3202               (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))))),
3203           (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
3204                          (LDRDui GPR64sp:$Rn, uimm12s8:$offset), dsub)>;
3205
3206// Match all load 64 bits width whose type is compatible with FPR64
3207let Predicates = [IsLE] in {
3208  // We must use LD1 to perform vector loads in big-endian.
3209  def : Pat<(v2f32 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
3210            (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
3211  def : Pat<(v8i8 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
3212            (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
3213  def : Pat<(v4i16 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
3214            (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
3215  def : Pat<(v2i32 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
3216            (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
3217  def : Pat<(v4f16 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
3218            (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
3219  def : Pat<(v4bf16 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
3220            (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
3221}
3222def : Pat<(v1f64 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
3223          (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
3224def : Pat<(v1i64 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
3225          (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
3226
3227// Match all load 128 bits width whose type is compatible with FPR128
3228let Predicates = [IsLE] in {
3229  // We must use LD1 to perform vector loads in big-endian.
3230  def : Pat<(v4f32 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
3231            (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
3232  def : Pat<(v2f64 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
3233            (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
3234  def : Pat<(v16i8 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
3235            (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
3236  def : Pat<(v8i16 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
3237            (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
3238  def : Pat<(v4i32 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
3239            (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
3240  def : Pat<(v2i64 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
3241            (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
3242  def : Pat<(v8f16 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
3243            (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
3244  def : Pat<(v8bf16 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
3245            (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
3246}
3247def : Pat<(f128  (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
3248          (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
3249
3250defm LDRHH : LoadUI<0b01, 0, 0b01, GPR32, uimm12s2, "ldrh",
3251                    [(set GPR32:$Rt,
3252                          (zextloadi16 (am_indexed16 GPR64sp:$Rn,
3253                                                     uimm12s2:$offset)))]>;
3254defm LDRBB : LoadUI<0b00, 0, 0b01, GPR32, uimm12s1, "ldrb",
3255                    [(set GPR32:$Rt,
3256                          (zextloadi8 (am_indexed8 GPR64sp:$Rn,
3257                                                   uimm12s1:$offset)))]>;
3258// zextload -> i64
3259def : Pat<(i64 (zextloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
3260    (SUBREG_TO_REG (i64 0), (LDRBBui GPR64sp:$Rn, uimm12s1:$offset), sub_32)>;
3261def : Pat<(i64 (zextloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))),
3262    (SUBREG_TO_REG (i64 0), (LDRHHui GPR64sp:$Rn, uimm12s2:$offset), sub_32)>;
3263
3264// zextloadi1 -> zextloadi8
3265def : Pat<(i32 (zextloadi1 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
3266          (LDRBBui GPR64sp:$Rn, uimm12s1:$offset)>;
3267def : Pat<(i64 (zextloadi1 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
3268    (SUBREG_TO_REG (i64 0), (LDRBBui GPR64sp:$Rn, uimm12s1:$offset), sub_32)>;
3269
3270// extload -> zextload
3271def : Pat<(i32 (extloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))),
3272          (LDRHHui GPR64sp:$Rn, uimm12s2:$offset)>;
3273def : Pat<(i32 (extloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
3274          (LDRBBui GPR64sp:$Rn, uimm12s1:$offset)>;
3275def : Pat<(i32 (extloadi1 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
3276          (LDRBBui GPR64sp:$Rn, uimm12s1:$offset)>;
3277def : Pat<(i64 (extloadi32 (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))),
3278    (SUBREG_TO_REG (i64 0), (LDRWui GPR64sp:$Rn, uimm12s4:$offset), sub_32)>;
3279def : Pat<(i64 (extloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))),
3280    (SUBREG_TO_REG (i64 0), (LDRHHui GPR64sp:$Rn, uimm12s2:$offset), sub_32)>;
3281def : Pat<(i64 (extloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
3282    (SUBREG_TO_REG (i64 0), (LDRBBui GPR64sp:$Rn, uimm12s1:$offset), sub_32)>;
3283def : Pat<(i64 (extloadi1 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
3284    (SUBREG_TO_REG (i64 0), (LDRBBui GPR64sp:$Rn, uimm12s1:$offset), sub_32)>;
3285
3286// load sign-extended half-word
3287defm LDRSHW : LoadUI<0b01, 0, 0b11, GPR32, uimm12s2, "ldrsh",
3288                     [(set GPR32:$Rt,
3289                           (sextloadi16 (am_indexed16 GPR64sp:$Rn,
3290                                                      uimm12s2:$offset)))]>;
3291defm LDRSHX : LoadUI<0b01, 0, 0b10, GPR64, uimm12s2, "ldrsh",
3292                     [(set GPR64:$Rt,
3293                           (sextloadi16 (am_indexed16 GPR64sp:$Rn,
3294                                                      uimm12s2:$offset)))]>;
3295
3296// load sign-extended byte
3297defm LDRSBW : LoadUI<0b00, 0, 0b11, GPR32, uimm12s1, "ldrsb",
3298                     [(set GPR32:$Rt,
3299                           (sextloadi8 (am_indexed8 GPR64sp:$Rn,
3300                                                    uimm12s1:$offset)))]>;
3301defm LDRSBX : LoadUI<0b00, 0, 0b10, GPR64, uimm12s1, "ldrsb",
3302                     [(set GPR64:$Rt,
3303                           (sextloadi8 (am_indexed8 GPR64sp:$Rn,
3304                                                    uimm12s1:$offset)))]>;
3305
3306// load sign-extended word
3307defm LDRSW  : LoadUI<0b10, 0, 0b10, GPR64, uimm12s4, "ldrsw",
3308                     [(set GPR64:$Rt,
3309                           (sextloadi32 (am_indexed32 GPR64sp:$Rn,
3310                                                      uimm12s4:$offset)))]>;
3311
3312// load zero-extended word
3313def : Pat<(i64 (zextloadi32 (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))),
3314      (SUBREG_TO_REG (i64 0), (LDRWui GPR64sp:$Rn, uimm12s4:$offset), sub_32)>;
3315
3316// Pre-fetch.
3317def PRFMui : PrefetchUI<0b11, 0, 0b10, "prfm",
3318                        [(AArch64Prefetch timm:$Rt,
3319                                        (am_indexed64 GPR64sp:$Rn,
3320                                                      uimm12s8:$offset))]>;
3321
3322def : InstAlias<"prfm $Rt, [$Rn]", (PRFMui prfop:$Rt, GPR64sp:$Rn, 0)>;
3323
3324//---
3325// (literal)
3326
3327def alignedglobal : PatLeaf<(iPTR iPTR:$label), [{
3328  if (auto *G = dyn_cast<GlobalAddressSDNode>(N)) {
3329    const DataLayout &DL = MF->getDataLayout();
3330    Align Align = G->getGlobal()->getPointerAlignment(DL);
3331    return Align >= 4 && G->getOffset() % 4 == 0;
3332  }
3333  if (auto *C = dyn_cast<ConstantPoolSDNode>(N))
3334    return C->getAlign() >= 4 && C->getOffset() % 4 == 0;
3335  return false;
3336}]>;
3337
3338def LDRWl : LoadLiteral<0b00, 0, GPR32z, "ldr",
3339  [(set GPR32z:$Rt, (load (AArch64adr alignedglobal:$label)))]>;
3340def LDRXl : LoadLiteral<0b01, 0, GPR64z, "ldr",
3341  [(set GPR64z:$Rt, (load (AArch64adr alignedglobal:$label)))]>;
3342def LDRSl : LoadLiteral<0b00, 1, FPR32Op, "ldr",
3343  [(set (f32 FPR32Op:$Rt), (load (AArch64adr alignedglobal:$label)))]>;
3344def LDRDl : LoadLiteral<0b01, 1, FPR64Op, "ldr",
3345  [(set (f64 FPR64Op:$Rt), (load (AArch64adr alignedglobal:$label)))]>;
3346def LDRQl : LoadLiteral<0b10, 1, FPR128Op, "ldr",
3347  [(set (f128 FPR128Op:$Rt), (load (AArch64adr alignedglobal:$label)))]>;
3348
3349// load sign-extended word
3350def LDRSWl : LoadLiteral<0b10, 0, GPR64z, "ldrsw",
3351  [(set GPR64z:$Rt, (sextloadi32 (AArch64adr alignedglobal:$label)))]>;
3352
3353let AddedComplexity = 20 in {
3354def : Pat<(i64 (zextloadi32 (AArch64adr alignedglobal:$label))),
3355        (SUBREG_TO_REG (i64 0), (LDRWl $label), sub_32)>;
3356}
3357
3358// prefetch
3359def PRFMl : PrefetchLiteral<0b11, 0, "prfm", []>;
3360//                   [(AArch64Prefetch imm:$Rt, tglobaladdr:$label)]>;
3361
3362//---
3363// (unscaled immediate)
3364defm LDURX : LoadUnscaled<0b11, 0, 0b01, GPR64z, "ldur",
3365                    [(set GPR64z:$Rt,
3366                          (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset)))]>;
3367defm LDURW : LoadUnscaled<0b10, 0, 0b01, GPR32z, "ldur",
3368                    [(set GPR32z:$Rt,
3369                          (load (am_unscaled32 GPR64sp:$Rn, simm9:$offset)))]>;
3370defm LDURB : LoadUnscaled<0b00, 1, 0b01, FPR8Op, "ldur",
3371                    [(set FPR8Op:$Rt,
3372                          (load (am_unscaled8 GPR64sp:$Rn, simm9:$offset)))]>;
3373defm LDURH : LoadUnscaled<0b01, 1, 0b01, FPR16Op, "ldur",
3374                    [(set (f16 FPR16Op:$Rt),
3375                          (load (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
3376defm LDURS : LoadUnscaled<0b10, 1, 0b01, FPR32Op, "ldur",
3377                    [(set (f32 FPR32Op:$Rt),
3378                          (load (am_unscaled32 GPR64sp:$Rn, simm9:$offset)))]>;
3379defm LDURD : LoadUnscaled<0b11, 1, 0b01, FPR64Op, "ldur",
3380                    [(set (f64 FPR64Op:$Rt),
3381                          (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset)))]>;
3382defm LDURQ : LoadUnscaled<0b00, 1, 0b11, FPR128Op, "ldur",
3383                    [(set (f128 FPR128Op:$Rt),
3384                          (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset)))]>;
3385
3386defm LDURHH
3387    : LoadUnscaled<0b01, 0, 0b01, GPR32, "ldurh",
3388             [(set GPR32:$Rt,
3389                    (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
3390defm LDURBB
3391    : LoadUnscaled<0b00, 0, 0b01, GPR32, "ldurb",
3392             [(set GPR32:$Rt,
3393                    (zextloadi8 (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
3394
3395// bf16 load pattern
3396def : Pat <(bf16 (load (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
3397           (LDURHi GPR64sp:$Rn, simm9:$offset)>;
3398
3399// Match all load 64 bits width whose type is compatible with FPR64
3400let Predicates = [IsLE] in {
3401  def : Pat<(v2f32 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
3402            (LDURDi GPR64sp:$Rn, simm9:$offset)>;
3403  def : Pat<(v2i32 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
3404            (LDURDi GPR64sp:$Rn, simm9:$offset)>;
3405  def : Pat<(v4i16 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
3406            (LDURDi GPR64sp:$Rn, simm9:$offset)>;
3407  def : Pat<(v8i8 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
3408            (LDURDi GPR64sp:$Rn, simm9:$offset)>;
3409  def : Pat<(v4f16 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
3410            (LDURDi GPR64sp:$Rn, simm9:$offset)>;
3411}
3412def : Pat<(v1f64 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
3413          (LDURDi GPR64sp:$Rn, simm9:$offset)>;
3414def : Pat<(v1i64 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
3415          (LDURDi GPR64sp:$Rn, simm9:$offset)>;
3416
3417// Match all load 128 bits width whose type is compatible with FPR128
3418let Predicates = [IsLE] in {
3419  def : Pat<(v2f64 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
3420            (LDURQi GPR64sp:$Rn, simm9:$offset)>;
3421  def : Pat<(v2i64 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
3422            (LDURQi GPR64sp:$Rn, simm9:$offset)>;
3423  def : Pat<(v4f32 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
3424            (LDURQi GPR64sp:$Rn, simm9:$offset)>;
3425  def : Pat<(v4i32 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
3426            (LDURQi GPR64sp:$Rn, simm9:$offset)>;
3427  def : Pat<(v8i16 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
3428            (LDURQi GPR64sp:$Rn, simm9:$offset)>;
3429  def : Pat<(v16i8 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
3430            (LDURQi GPR64sp:$Rn, simm9:$offset)>;
3431  def : Pat<(v8f16 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
3432            (LDURQi GPR64sp:$Rn, simm9:$offset)>;
3433}
3434
3435//  anyext -> zext
3436def : Pat<(i32 (extloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
3437          (LDURHHi GPR64sp:$Rn, simm9:$offset)>;
3438def : Pat<(i32 (extloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
3439          (LDURBBi GPR64sp:$Rn, simm9:$offset)>;
3440def : Pat<(i32 (extloadi1 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
3441          (LDURBBi GPR64sp:$Rn, simm9:$offset)>;
3442def : Pat<(i64 (extloadi32 (am_unscaled32 GPR64sp:$Rn, simm9:$offset))),
3443    (SUBREG_TO_REG (i64 0), (LDURWi GPR64sp:$Rn, simm9:$offset), sub_32)>;
3444def : Pat<(i64 (extloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
3445    (SUBREG_TO_REG (i64 0), (LDURHHi GPR64sp:$Rn, simm9:$offset), sub_32)>;
3446def : Pat<(i64 (extloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
3447    (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
3448def : Pat<(i64 (extloadi1 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
3449    (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
3450// unscaled zext
3451def : Pat<(i32 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
3452          (LDURHHi GPR64sp:$Rn, simm9:$offset)>;
3453def : Pat<(i32 (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
3454          (LDURBBi GPR64sp:$Rn, simm9:$offset)>;
3455def : Pat<(i32 (zextloadi1 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
3456          (LDURBBi GPR64sp:$Rn, simm9:$offset)>;
3457def : Pat<(i64 (zextloadi32 (am_unscaled32 GPR64sp:$Rn, simm9:$offset))),
3458    (SUBREG_TO_REG (i64 0), (LDURWi GPR64sp:$Rn, simm9:$offset), sub_32)>;
3459def : Pat<(i64 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
3460    (SUBREG_TO_REG (i64 0), (LDURHHi GPR64sp:$Rn, simm9:$offset), sub_32)>;
3461def : Pat<(i64 (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
3462    (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
3463def : Pat<(i64 (zextloadi1 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
3464    (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
3465
3466
3467//---
3468// LDR mnemonics fall back to LDUR for negative or unaligned offsets.
3469
3470// Define new assembler match classes as we want to only match these when
3471// the don't otherwise match the scaled addressing mode for LDR/STR. Don't
3472// associate a DiagnosticType either, as we want the diagnostic for the
3473// canonical form (the scaled operand) to take precedence.
3474class SImm9OffsetOperand<int Width> : AsmOperandClass {
3475  let Name = "SImm9OffsetFB" # Width;
3476  let PredicateMethod = "isSImm9OffsetFB<" # Width # ">";
3477  let RenderMethod = "addImmOperands";
3478}
3479
3480def SImm9OffsetFB8Operand : SImm9OffsetOperand<8>;
3481def SImm9OffsetFB16Operand : SImm9OffsetOperand<16>;
3482def SImm9OffsetFB32Operand : SImm9OffsetOperand<32>;
3483def SImm9OffsetFB64Operand : SImm9OffsetOperand<64>;
3484def SImm9OffsetFB128Operand : SImm9OffsetOperand<128>;
3485
3486def simm9_offset_fb8 : Operand<i64> {
3487  let ParserMatchClass = SImm9OffsetFB8Operand;
3488}
3489def simm9_offset_fb16 : Operand<i64> {
3490  let ParserMatchClass = SImm9OffsetFB16Operand;
3491}
3492def simm9_offset_fb32 : Operand<i64> {
3493  let ParserMatchClass = SImm9OffsetFB32Operand;
3494}
3495def simm9_offset_fb64 : Operand<i64> {
3496  let ParserMatchClass = SImm9OffsetFB64Operand;
3497}
3498def simm9_offset_fb128 : Operand<i64> {
3499  let ParserMatchClass = SImm9OffsetFB128Operand;
3500}
3501
3502def : InstAlias<"ldr $Rt, [$Rn, $offset]",
3503                (LDURXi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb64:$offset), 0>;
3504def : InstAlias<"ldr $Rt, [$Rn, $offset]",
3505                (LDURWi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
3506def : InstAlias<"ldr $Rt, [$Rn, $offset]",
3507                (LDURBi FPR8Op:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
3508def : InstAlias<"ldr $Rt, [$Rn, $offset]",
3509                (LDURHi FPR16Op:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
3510def : InstAlias<"ldr $Rt, [$Rn, $offset]",
3511                (LDURSi FPR32Op:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
3512def : InstAlias<"ldr $Rt, [$Rn, $offset]",
3513                (LDURDi FPR64Op:$Rt, GPR64sp:$Rn, simm9_offset_fb64:$offset), 0>;
3514def : InstAlias<"ldr $Rt, [$Rn, $offset]",
3515               (LDURQi FPR128Op:$Rt, GPR64sp:$Rn, simm9_offset_fb128:$offset), 0>;
3516
3517// zextload -> i64
3518def : Pat<(i64 (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
3519  (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
3520def : Pat<(i64 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
3521  (SUBREG_TO_REG (i64 0), (LDURHHi GPR64sp:$Rn, simm9:$offset), sub_32)>;
3522
3523// load sign-extended half-word
3524defm LDURSHW
3525    : LoadUnscaled<0b01, 0, 0b11, GPR32, "ldursh",
3526               [(set GPR32:$Rt,
3527                    (sextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
3528defm LDURSHX
3529    : LoadUnscaled<0b01, 0, 0b10, GPR64, "ldursh",
3530              [(set GPR64:$Rt,
3531                    (sextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
3532
3533// load sign-extended byte
3534defm LDURSBW
3535    : LoadUnscaled<0b00, 0, 0b11, GPR32, "ldursb",
3536                [(set GPR32:$Rt,
3537                      (sextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset)))]>;
3538defm LDURSBX
3539    : LoadUnscaled<0b00, 0, 0b10, GPR64, "ldursb",
3540                [(set GPR64:$Rt,
3541                      (sextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset)))]>;
3542
3543// load sign-extended word
3544defm LDURSW
3545    : LoadUnscaled<0b10, 0, 0b10, GPR64, "ldursw",
3546              [(set GPR64:$Rt,
3547                    (sextloadi32 (am_unscaled32 GPR64sp:$Rn, simm9:$offset)))]>;
3548
3549// zero and sign extending aliases from generic LDR* mnemonics to LDUR*.
3550def : InstAlias<"ldrb $Rt, [$Rn, $offset]",
3551                (LDURBBi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
3552def : InstAlias<"ldrh $Rt, [$Rn, $offset]",
3553                (LDURHHi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
3554def : InstAlias<"ldrsb $Rt, [$Rn, $offset]",
3555                (LDURSBWi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
3556def : InstAlias<"ldrsb $Rt, [$Rn, $offset]",
3557                (LDURSBXi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
3558def : InstAlias<"ldrsh $Rt, [$Rn, $offset]",
3559                (LDURSHWi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
3560def : InstAlias<"ldrsh $Rt, [$Rn, $offset]",
3561                (LDURSHXi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
3562def : InstAlias<"ldrsw $Rt, [$Rn, $offset]",
3563                (LDURSWi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
3564
3565// A LDR will implicitly zero the rest of the vector, so vector_insert(zeros,
3566// load, 0) can use a single load.
3567multiclass LoadInsertZeroPatterns<SDPatternOperator LoadOp, ValueType VT, ValueType HVT, ValueType SVT,
3568                                  ValueType ScalarVT, Instruction LoadInst, Instruction UnscaledLoadInst,
3569                                  ComplexPattern Addr, ComplexPattern UnscaledAddr, Operand AddrImm,
3570                                  SubRegIndex SubReg> {
3571  // Scaled
3572  def : Pat <(vector_insert (VT immAllZerosV),
3573                (ScalarVT (LoadOp (Addr GPR64sp:$Rn, AddrImm:$offset))), (i64 0)),
3574            (SUBREG_TO_REG (i64 0), (LoadInst GPR64sp:$Rn, AddrImm:$offset), SubReg)>;
3575  // Unscaled
3576  def : Pat <(vector_insert (VT immAllZerosV),
3577                 (ScalarVT (LoadOp (UnscaledAddr GPR64sp:$Rn, simm9:$offset))), (i64 0)),
3578             (SUBREG_TO_REG (i64 0), (UnscaledLoadInst GPR64sp:$Rn, simm9:$offset), SubReg)>;
3579
3580  // Half-vector patterns
3581  def : Pat <(vector_insert (HVT immAllZerosV),
3582                 (ScalarVT (LoadOp (Addr GPR64sp:$Rn, AddrImm:$offset))), (i64 0)),
3583             (SUBREG_TO_REG (i64 0), (LoadInst GPR64sp:$Rn, AddrImm:$offset), SubReg)>;
3584  // Unscaled
3585  def : Pat <(vector_insert (HVT immAllZerosV),
3586                 (ScalarVT (LoadOp (UnscaledAddr GPR64sp:$Rn, simm9:$offset))), (i64 0)),
3587             (SUBREG_TO_REG (i64 0), (UnscaledLoadInst GPR64sp:$Rn, simm9:$offset), SubReg)>;
3588
3589  // SVE patterns
3590  def : Pat <(vector_insert (SVT immAllZerosV),
3591                 (ScalarVT (LoadOp (Addr GPR64sp:$Rn, AddrImm:$offset))), (i64 0)),
3592             (SUBREG_TO_REG (i64 0), (LoadInst GPR64sp:$Rn, AddrImm:$offset), SubReg)>;
3593  // Unscaled
3594  def : Pat <(vector_insert (SVT immAllZerosV),
3595                 (ScalarVT (LoadOp (UnscaledAddr GPR64sp:$Rn, simm9:$offset))), (i64 0)),
3596             (SUBREG_TO_REG (i64 0), (UnscaledLoadInst GPR64sp:$Rn, simm9:$offset), SubReg)>;
3597}
3598
3599defm : LoadInsertZeroPatterns<extloadi8,  v16i8,  v8i8,   nxv16i8,  i32,  LDRBui, LDURBi,
3600                              am_indexed8,  am_unscaled8,  uimm12s1, bsub>;
3601defm : LoadInsertZeroPatterns<extloadi16, v8i16,  v4i16,  nxv8i16,  i32,  LDRHui, LDURHi,
3602                              am_indexed16, am_unscaled16, uimm12s2, hsub>;
3603defm : LoadInsertZeroPatterns<load,       v4i32,  v2i32,  nxv4i32,  i32,  LDRSui, LDURSi,
3604                              am_indexed32, am_unscaled32, uimm12s4, ssub>;
3605defm : LoadInsertZeroPatterns<load,       v2i64,  v1i64,  nxv2i64,  i64,  LDRDui, LDURDi,
3606                              am_indexed64, am_unscaled64, uimm12s8, dsub>;
3607defm : LoadInsertZeroPatterns<load,       v8f16,  v4f16,  nxv8f16,  f16,  LDRHui, LDURHi,
3608                              am_indexed16, am_unscaled16, uimm12s2, hsub>;
3609defm : LoadInsertZeroPatterns<load,       v8bf16, v4bf16, nxv8bf16, bf16, LDRHui, LDURHi,
3610                              am_indexed16, am_unscaled16, uimm12s2, hsub>;
3611defm : LoadInsertZeroPatterns<load,       v4f32,  v2f32,  nxv4f32,  f32,  LDRSui, LDURSi,
3612                              am_indexed32, am_unscaled32, uimm12s4, ssub>;
3613defm : LoadInsertZeroPatterns<load,       v2f64,  v1f64,  nxv2f64,  f64,  LDRDui, LDURDi,
3614                              am_indexed64, am_unscaled64, uimm12s8, dsub>;
3615
3616// Pre-fetch.
3617defm PRFUM : PrefetchUnscaled<0b11, 0, 0b10, "prfum",
3618                  [(AArch64Prefetch timm:$Rt,
3619                                  (am_unscaled64 GPR64sp:$Rn, simm9:$offset))]>;
3620
3621//---
3622// (unscaled immediate, unprivileged)
3623defm LDTRX : LoadUnprivileged<0b11, 0, 0b01, GPR64, "ldtr">;
3624defm LDTRW : LoadUnprivileged<0b10, 0, 0b01, GPR32, "ldtr">;
3625
3626defm LDTRH : LoadUnprivileged<0b01, 0, 0b01, GPR32, "ldtrh">;
3627defm LDTRB : LoadUnprivileged<0b00, 0, 0b01, GPR32, "ldtrb">;
3628
3629// load sign-extended half-word
3630defm LDTRSHW : LoadUnprivileged<0b01, 0, 0b11, GPR32, "ldtrsh">;
3631defm LDTRSHX : LoadUnprivileged<0b01, 0, 0b10, GPR64, "ldtrsh">;
3632
3633// load sign-extended byte
3634defm LDTRSBW : LoadUnprivileged<0b00, 0, 0b11, GPR32, "ldtrsb">;
3635defm LDTRSBX : LoadUnprivileged<0b00, 0, 0b10, GPR64, "ldtrsb">;
3636
3637// load sign-extended word
3638defm LDTRSW  : LoadUnprivileged<0b10, 0, 0b10, GPR64, "ldtrsw">;
3639
3640//---
3641// (immediate pre-indexed)
3642def LDRWpre : LoadPreIdx<0b10, 0, 0b01, GPR32z, "ldr">;
3643def LDRXpre : LoadPreIdx<0b11, 0, 0b01, GPR64z, "ldr">;
3644def LDRBpre : LoadPreIdx<0b00, 1, 0b01, FPR8Op,  "ldr">;
3645def LDRHpre : LoadPreIdx<0b01, 1, 0b01, FPR16Op, "ldr">;
3646def LDRSpre : LoadPreIdx<0b10, 1, 0b01, FPR32Op, "ldr">;
3647def LDRDpre : LoadPreIdx<0b11, 1, 0b01, FPR64Op, "ldr">;
3648def LDRQpre : LoadPreIdx<0b00, 1, 0b11, FPR128Op, "ldr">;
3649
3650// load sign-extended half-word
3651def LDRSHWpre : LoadPreIdx<0b01, 0, 0b11, GPR32z, "ldrsh">;
3652def LDRSHXpre : LoadPreIdx<0b01, 0, 0b10, GPR64z, "ldrsh">;
3653
3654// load sign-extended byte
3655def LDRSBWpre : LoadPreIdx<0b00, 0, 0b11, GPR32z, "ldrsb">;
3656def LDRSBXpre : LoadPreIdx<0b00, 0, 0b10, GPR64z, "ldrsb">;
3657
3658// load zero-extended byte
3659def LDRBBpre : LoadPreIdx<0b00, 0, 0b01, GPR32z, "ldrb">;
3660def LDRHHpre : LoadPreIdx<0b01, 0, 0b01, GPR32z, "ldrh">;
3661
3662// load sign-extended word
3663def LDRSWpre : LoadPreIdx<0b10, 0, 0b10, GPR64z, "ldrsw">;
3664
3665//---
3666// (immediate post-indexed)
3667def LDRWpost : LoadPostIdx<0b10, 0, 0b01, GPR32z, "ldr">;
3668def LDRXpost : LoadPostIdx<0b11, 0, 0b01, GPR64z, "ldr">;
3669def LDRBpost : LoadPostIdx<0b00, 1, 0b01, FPR8Op,  "ldr">;
3670def LDRHpost : LoadPostIdx<0b01, 1, 0b01, FPR16Op, "ldr">;
3671def LDRSpost : LoadPostIdx<0b10, 1, 0b01, FPR32Op, "ldr">;
3672def LDRDpost : LoadPostIdx<0b11, 1, 0b01, FPR64Op, "ldr">;
3673def LDRQpost : LoadPostIdx<0b00, 1, 0b11, FPR128Op, "ldr">;
3674
3675// load sign-extended half-word
3676def LDRSHWpost : LoadPostIdx<0b01, 0, 0b11, GPR32z, "ldrsh">;
3677def LDRSHXpost : LoadPostIdx<0b01, 0, 0b10, GPR64z, "ldrsh">;
3678
3679// load sign-extended byte
3680def LDRSBWpost : LoadPostIdx<0b00, 0, 0b11, GPR32z, "ldrsb">;
3681def LDRSBXpost : LoadPostIdx<0b00, 0, 0b10, GPR64z, "ldrsb">;
3682
3683// load zero-extended byte
3684def LDRBBpost : LoadPostIdx<0b00, 0, 0b01, GPR32z, "ldrb">;
3685def LDRHHpost : LoadPostIdx<0b01, 0, 0b01, GPR32z, "ldrh">;
3686
3687// load sign-extended word
3688def LDRSWpost : LoadPostIdx<0b10, 0, 0b10, GPR64z, "ldrsw">;
3689
3690//===----------------------------------------------------------------------===//
3691// Store instructions.
3692//===----------------------------------------------------------------------===//
3693
3694// Pair (indexed, offset)
3695// FIXME: Use dedicated range-checked addressing mode operand here.
3696defm STPW : StorePairOffset<0b00, 0, GPR32z, simm7s4, "stp">;
3697defm STPX : StorePairOffset<0b10, 0, GPR64z, simm7s8, "stp">;
3698defm STPS : StorePairOffset<0b00, 1, FPR32Op, simm7s4, "stp">;
3699defm STPD : StorePairOffset<0b01, 1, FPR64Op, simm7s8, "stp">;
3700defm STPQ : StorePairOffset<0b10, 1, FPR128Op, simm7s16, "stp">;
3701
3702// Pair (pre-indexed)
3703def STPWpre : StorePairPreIdx<0b00, 0, GPR32z, simm7s4, "stp">;
3704def STPXpre : StorePairPreIdx<0b10, 0, GPR64z, simm7s8, "stp">;
3705def STPSpre : StorePairPreIdx<0b00, 1, FPR32Op, simm7s4, "stp">;
3706def STPDpre : StorePairPreIdx<0b01, 1, FPR64Op, simm7s8, "stp">;
3707def STPQpre : StorePairPreIdx<0b10, 1, FPR128Op, simm7s16, "stp">;
3708
3709// Pair (post-indexed)
3710def STPWpost : StorePairPostIdx<0b00, 0, GPR32z, simm7s4, "stp">;
3711def STPXpost : StorePairPostIdx<0b10, 0, GPR64z, simm7s8, "stp">;
3712def STPSpost : StorePairPostIdx<0b00, 1, FPR32Op, simm7s4, "stp">;
3713def STPDpost : StorePairPostIdx<0b01, 1, FPR64Op, simm7s8, "stp">;
3714def STPQpost : StorePairPostIdx<0b10, 1, FPR128Op, simm7s16, "stp">;
3715
3716// Pair (no allocate)
3717defm STNPW : StorePairNoAlloc<0b00, 0, GPR32z, simm7s4, "stnp">;
3718defm STNPX : StorePairNoAlloc<0b10, 0, GPR64z, simm7s8, "stnp">;
3719defm STNPS : StorePairNoAlloc<0b00, 1, FPR32Op, simm7s4, "stnp">;
3720defm STNPD : StorePairNoAlloc<0b01, 1, FPR64Op, simm7s8, "stnp">;
3721defm STNPQ : StorePairNoAlloc<0b10, 1, FPR128Op, simm7s16, "stnp">;
3722
3723def : Pat<(AArch64stp GPR64z:$Rt, GPR64z:$Rt2, (am_indexed7s64 GPR64sp:$Rn, simm7s8:$offset)),
3724          (STPXi GPR64z:$Rt, GPR64z:$Rt2, GPR64sp:$Rn, simm7s8:$offset)>;
3725
3726def : Pat<(AArch64stnp FPR128:$Rt, FPR128:$Rt2, (am_indexed7s128 GPR64sp:$Rn, simm7s16:$offset)),
3727          (STNPQi FPR128:$Rt, FPR128:$Rt2, GPR64sp:$Rn, simm7s16:$offset)>;
3728
3729
3730//---
3731// (Register offset)
3732
3733// Integer
3734defm STRBB : Store8RO< 0b00, 0, 0b00, GPR32, "strb", i32, truncstorei8>;
3735defm STRHH : Store16RO<0b01, 0, 0b00, GPR32, "strh", i32, truncstorei16>;
3736defm STRW  : Store32RO<0b10, 0, 0b00, GPR32, "str",  i32, store>;
3737defm STRX  : Store64RO<0b11, 0, 0b00, GPR64, "str",  i64, store>;
3738
3739
3740// Floating-point
3741defm STRB : Store8RO< 0b00,  1, 0b00, FPR8Op,   "str", i8, store>;
3742defm STRH : Store16RO<0b01,  1, 0b00, FPR16Op,  "str", f16,     store>;
3743defm STRS : Store32RO<0b10,  1, 0b00, FPR32Op,  "str", f32,     store>;
3744defm STRD : Store64RO<0b11,  1, 0b00, FPR64Op,  "str", f64,     store>;
3745defm STRQ : Store128RO<0b00, 1, 0b10, FPR128Op, "str">;
3746
3747let Predicates = [UseSTRQro], AddedComplexity = 10 in {
3748  def : Pat<(store (f128 FPR128:$Rt),
3749                        (ro_Windexed128 GPR64sp:$Rn, GPR32:$Rm,
3750                                        ro_Wextend128:$extend)),
3751            (STRQroW FPR128:$Rt, GPR64sp:$Rn, GPR32:$Rm, ro_Wextend128:$extend)>;
3752  def : Pat<(store (f128 FPR128:$Rt),
3753                        (ro_Xindexed128 GPR64sp:$Rn, GPR64:$Rm,
3754                                        ro_Xextend128:$extend)),
3755            (STRQroX FPR128:$Rt, GPR64sp:$Rn, GPR64:$Rm, ro_Wextend128:$extend)>;
3756}
3757
3758multiclass TruncStoreFrom64ROPat<ROAddrMode ro, SDPatternOperator storeop,
3759                                 Instruction STRW, Instruction STRX> {
3760
3761  def : Pat<(storeop GPR64:$Rt,
3762                     (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)),
3763            (STRW (EXTRACT_SUBREG GPR64:$Rt, sub_32),
3764                  GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
3765
3766  def : Pat<(storeop GPR64:$Rt,
3767                     (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)),
3768            (STRX (EXTRACT_SUBREG GPR64:$Rt, sub_32),
3769                  GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
3770}
3771
3772let AddedComplexity = 10 in {
3773  // truncstore i64
3774  defm : TruncStoreFrom64ROPat<ro8,  truncstorei8,  STRBBroW, STRBBroX>;
3775  defm : TruncStoreFrom64ROPat<ro16, truncstorei16, STRHHroW, STRHHroX>;
3776  defm : TruncStoreFrom64ROPat<ro32, truncstorei32, STRWroW,  STRWroX>;
3777}
3778
3779multiclass VecROStorePat<ROAddrMode ro, ValueType VecTy, RegisterClass FPR,
3780                         Instruction STRW, Instruction STRX> {
3781  def : Pat<(store (VecTy FPR:$Rt),
3782                   (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)),
3783            (STRW FPR:$Rt, GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
3784
3785  def : Pat<(store (VecTy FPR:$Rt),
3786                   (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)),
3787            (STRX FPR:$Rt, GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
3788}
3789
3790let AddedComplexity = 10 in {
3791// Match all store 64 bits width whose type is compatible with FPR64
3792let Predicates = [IsLE] in {
3793  // We must use ST1 to store vectors in big-endian.
3794  defm : VecROStorePat<ro64, v2i32, FPR64, STRDroW, STRDroX>;
3795  defm : VecROStorePat<ro64, v2f32, FPR64, STRDroW, STRDroX>;
3796  defm : VecROStorePat<ro64, v4i16, FPR64, STRDroW, STRDroX>;
3797  defm : VecROStorePat<ro64, v8i8, FPR64, STRDroW, STRDroX>;
3798  defm : VecROStorePat<ro64, v4f16, FPR64, STRDroW, STRDroX>;
3799  defm : VecROStorePat<ro64, v4bf16, FPR64, STRDroW, STRDroX>;
3800}
3801
3802defm : VecROStorePat<ro64, v1i64, FPR64, STRDroW, STRDroX>;
3803defm : VecROStorePat<ro64, v1f64, FPR64, STRDroW, STRDroX>;
3804
3805// Match all store 128 bits width whose type is compatible with FPR128
3806let Predicates = [IsLE, UseSTRQro] in {
3807  // We must use ST1 to store vectors in big-endian.
3808  defm : VecROStorePat<ro128, v2i64, FPR128, STRQroW, STRQroX>;
3809  defm : VecROStorePat<ro128, v2f64, FPR128, STRQroW, STRQroX>;
3810  defm : VecROStorePat<ro128, v4i32, FPR128, STRQroW, STRQroX>;
3811  defm : VecROStorePat<ro128, v4f32, FPR128, STRQroW, STRQroX>;
3812  defm : VecROStorePat<ro128, v8i16, FPR128, STRQroW, STRQroX>;
3813  defm : VecROStorePat<ro128, v16i8, FPR128, STRQroW, STRQroX>;
3814  defm : VecROStorePat<ro128, v8f16, FPR128, STRQroW, STRQroX>;
3815  defm : VecROStorePat<ro128, v8bf16, FPR128, STRQroW, STRQroX>;
3816}
3817} // AddedComplexity = 10
3818
3819// Match stores from lane 0 to the appropriate subreg's store.
3820multiclass VecROStoreLane0Pat<ROAddrMode ro, SDPatternOperator storeop,
3821                              ValueType VecTy, ValueType STy,
3822                              ValueType SubRegTy,
3823                              SubRegIndex SubRegIdx,
3824                              Instruction STRW, Instruction STRX> {
3825
3826  def : Pat<(storeop (STy (vector_extract (VecTy VecListOne128:$Vt), (i64 0))),
3827                     (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)),
3828            (STRW (SubRegTy (EXTRACT_SUBREG VecListOne128:$Vt, SubRegIdx)),
3829                  GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
3830
3831  def : Pat<(storeop (STy (vector_extract (VecTy VecListOne128:$Vt), (i64 0))),
3832                     (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)),
3833            (STRX (SubRegTy (EXTRACT_SUBREG VecListOne128:$Vt, SubRegIdx)),
3834                  GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
3835}
3836
3837let AddedComplexity = 19 in {
3838  defm : VecROStoreLane0Pat<ro16, truncstorei16, v8i16, i32, f16, hsub, STRHroW, STRHroX>;
3839  defm : VecROStoreLane0Pat<ro16,         store, v8f16, f16, f16, hsub, STRHroW, STRHroX>;
3840  defm : VecROStoreLane0Pat<ro32,         store, v4i32, i32, i32, ssub, STRSroW, STRSroX>;
3841  defm : VecROStoreLane0Pat<ro32,         store, v4f32, f32, i32, ssub, STRSroW, STRSroX>;
3842  defm : VecROStoreLane0Pat<ro64,         store, v2i64, i64, i64, dsub, STRDroW, STRDroX>;
3843  defm : VecROStoreLane0Pat<ro64,         store, v2f64, f64, i64, dsub, STRDroW, STRDroX>;
3844}
3845
3846//---
3847// (unsigned immediate)
3848defm STRX : StoreUIz<0b11, 0, 0b00, GPR64z, uimm12s8, "str",
3849                   [(store GPR64z:$Rt,
3850                            (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))]>;
3851defm STRW : StoreUIz<0b10, 0, 0b00, GPR32z, uimm12s4, "str",
3852                    [(store GPR32z:$Rt,
3853                            (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))]>;
3854defm STRB : StoreUI<0b00, 1, 0b00, FPR8Op, uimm12s1, "str",
3855                    [(store FPR8Op:$Rt,
3856                            (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))]>;
3857defm STRH : StoreUI<0b01, 1, 0b00, FPR16Op, uimm12s2, "str",
3858                    [(store (f16 FPR16Op:$Rt),
3859                            (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))]>;
3860defm STRS : StoreUI<0b10, 1, 0b00, FPR32Op, uimm12s4, "str",
3861                    [(store (f32 FPR32Op:$Rt),
3862                            (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))]>;
3863defm STRD : StoreUI<0b11, 1, 0b00, FPR64Op, uimm12s8, "str",
3864                    [(store (f64 FPR64Op:$Rt),
3865                            (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))]>;
3866defm STRQ : StoreUI<0b00, 1, 0b10, FPR128Op, uimm12s16, "str", []>;
3867
3868defm STRHH : StoreUIz<0b01, 0, 0b00, GPR32z, uimm12s2, "strh",
3869                     [(truncstorei16 GPR32z:$Rt,
3870                                     (am_indexed16 GPR64sp:$Rn,
3871                                                   uimm12s2:$offset))]>;
3872defm STRBB : StoreUIz<0b00, 0, 0b00, GPR32z, uimm12s1,  "strb",
3873                     [(truncstorei8 GPR32z:$Rt,
3874                                    (am_indexed8 GPR64sp:$Rn,
3875                                                 uimm12s1:$offset))]>;
3876
3877// bf16 store pattern
3878def : Pat<(store (bf16 FPR16Op:$Rt),
3879                 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset)),
3880          (STRHui FPR16:$Rt, GPR64sp:$Rn, uimm12s2:$offset)>;
3881
3882let AddedComplexity = 10 in {
3883
3884// Match all store 64 bits width whose type is compatible with FPR64
3885def : Pat<(store (v1i64 FPR64:$Rt),
3886                 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
3887          (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
3888def : Pat<(store (v1f64 FPR64:$Rt),
3889                 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
3890          (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
3891
3892let Predicates = [IsLE] in {
3893  // We must use ST1 to store vectors in big-endian.
3894  def : Pat<(store (v2f32 FPR64:$Rt),
3895                   (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
3896            (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
3897  def : Pat<(store (v8i8 FPR64:$Rt),
3898                   (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
3899            (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
3900  def : Pat<(store (v4i16 FPR64:$Rt),
3901                   (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
3902            (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
3903  def : Pat<(store (v2i32 FPR64:$Rt),
3904                   (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
3905            (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
3906  def : Pat<(store (v4f16 FPR64:$Rt),
3907                   (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
3908            (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
3909  def : Pat<(store (v4bf16 FPR64:$Rt),
3910                   (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
3911            (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
3912}
3913
3914// Match all store 128 bits width whose type is compatible with FPR128
3915def : Pat<(store (f128  FPR128:$Rt),
3916                 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
3917          (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
3918
3919let Predicates = [IsLE] in {
3920  // We must use ST1 to store vectors in big-endian.
3921  def : Pat<(store (v4f32 FPR128:$Rt),
3922                   (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
3923            (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
3924  def : Pat<(store (v2f64 FPR128:$Rt),
3925                   (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
3926            (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
3927  def : Pat<(store (v16i8 FPR128:$Rt),
3928                   (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
3929            (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
3930  def : Pat<(store (v8i16 FPR128:$Rt),
3931                   (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
3932            (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
3933  def : Pat<(store (v4i32 FPR128:$Rt),
3934                   (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
3935            (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
3936  def : Pat<(store (v2i64 FPR128:$Rt),
3937                   (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
3938            (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
3939  def : Pat<(store (v8f16 FPR128:$Rt),
3940                   (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
3941            (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
3942  def : Pat<(store (v8bf16 FPR128:$Rt),
3943                   (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
3944            (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
3945}
3946
3947// truncstore i64
3948def : Pat<(truncstorei32 GPR64:$Rt,
3949                         (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset)),
3950  (STRWui (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, uimm12s4:$offset)>;
3951def : Pat<(truncstorei16 GPR64:$Rt,
3952                         (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset)),
3953  (STRHHui (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, uimm12s2:$offset)>;
3954def : Pat<(truncstorei8 GPR64:$Rt, (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset)),
3955  (STRBBui (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, uimm12s1:$offset)>;
3956
3957} // AddedComplexity = 10
3958
3959// Match stores from lane 0 to the appropriate subreg's store.
3960multiclass VecStoreLane0Pat<ComplexPattern UIAddrMode, SDPatternOperator storeop,
3961                            ValueType VTy, ValueType STy,
3962                            ValueType SubRegTy,
3963                            SubRegIndex SubRegIdx, Operand IndexType,
3964                            Instruction STR> {
3965  def : Pat<(storeop (STy (vector_extract (VTy VecListOne128:$Vt), (i64 0))),
3966                     (UIAddrMode GPR64sp:$Rn, IndexType:$offset)),
3967            (STR (SubRegTy (EXTRACT_SUBREG VecListOne128:$Vt, SubRegIdx)),
3968                 GPR64sp:$Rn, IndexType:$offset)>;
3969}
3970
3971let AddedComplexity = 19 in {
3972  defm : VecStoreLane0Pat<am_indexed16, truncstorei16, v8i16, i32, f16, hsub, uimm12s2, STRHui>;
3973  defm : VecStoreLane0Pat<am_indexed16,         store, v8f16, f16, f16, hsub, uimm12s2, STRHui>;
3974  defm : VecStoreLane0Pat<am_indexed32,         store, v4i32, i32, i32, ssub, uimm12s4, STRSui>;
3975  defm : VecStoreLane0Pat<am_indexed32,         store, v4f32, f32, i32, ssub, uimm12s4, STRSui>;
3976  defm : VecStoreLane0Pat<am_indexed64,         store, v2i64, i64, i64, dsub, uimm12s8, STRDui>;
3977  defm : VecStoreLane0Pat<am_indexed64,         store, v2f64, f64, i64, dsub, uimm12s8, STRDui>;
3978}
3979
3980//---
3981// (unscaled immediate)
3982defm STURX : StoreUnscaled<0b11, 0, 0b00, GPR64z, "stur",
3983                         [(store GPR64z:$Rt,
3984                                 (am_unscaled64 GPR64sp:$Rn, simm9:$offset))]>;
3985defm STURW : StoreUnscaled<0b10, 0, 0b00, GPR32z, "stur",
3986                         [(store GPR32z:$Rt,
3987                                 (am_unscaled32 GPR64sp:$Rn, simm9:$offset))]>;
3988defm STURB : StoreUnscaled<0b00, 1, 0b00, FPR8Op, "stur",
3989                         [(store FPR8Op:$Rt,
3990                                 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))]>;
3991defm STURH : StoreUnscaled<0b01, 1, 0b00, FPR16Op, "stur",
3992                         [(store (f16 FPR16Op:$Rt),
3993                                 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))]>;
3994defm STURS : StoreUnscaled<0b10, 1, 0b00, FPR32Op, "stur",
3995                         [(store (f32 FPR32Op:$Rt),
3996                                 (am_unscaled32 GPR64sp:$Rn, simm9:$offset))]>;
3997defm STURD : StoreUnscaled<0b11, 1, 0b00, FPR64Op, "stur",
3998                         [(store (f64 FPR64Op:$Rt),
3999                                 (am_unscaled64 GPR64sp:$Rn, simm9:$offset))]>;
4000defm STURQ : StoreUnscaled<0b00, 1, 0b10, FPR128Op, "stur",
4001                         [(store (f128 FPR128Op:$Rt),
4002                                 (am_unscaled128 GPR64sp:$Rn, simm9:$offset))]>;
4003defm STURHH : StoreUnscaled<0b01, 0, 0b00, GPR32z, "sturh",
4004                         [(truncstorei16 GPR32z:$Rt,
4005                                 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))]>;
4006defm STURBB : StoreUnscaled<0b00, 0, 0b00, GPR32z, "sturb",
4007                         [(truncstorei8 GPR32z:$Rt,
4008                                  (am_unscaled8 GPR64sp:$Rn, simm9:$offset))]>;
4009
4010// bf16 store pattern
4011def : Pat<(store (bf16 FPR16Op:$Rt),
4012                 (am_unscaled16 GPR64sp:$Rn, simm9:$offset)),
4013          (STURHi FPR16:$Rt, GPR64sp:$Rn, simm9:$offset)>;
4014
4015// Armv8.4 Weaker Release Consistency enhancements
4016//         LDAPR & STLR with Immediate Offset instructions
4017let Predicates = [HasRCPC_IMMO] in {
4018defm STLURB     : BaseStoreUnscaleV84<"stlurb",  0b00, 0b00, GPR32>;
4019defm STLURH     : BaseStoreUnscaleV84<"stlurh",  0b01, 0b00, GPR32>;
4020defm STLURW     : BaseStoreUnscaleV84<"stlur",   0b10, 0b00, GPR32>;
4021defm STLURX     : BaseStoreUnscaleV84<"stlur",   0b11, 0b00, GPR64>;
4022defm LDAPURB    : BaseLoadUnscaleV84<"ldapurb",  0b00, 0b01, GPR32>;
4023defm LDAPURSBW  : BaseLoadUnscaleV84<"ldapursb", 0b00, 0b11, GPR32>;
4024defm LDAPURSBX  : BaseLoadUnscaleV84<"ldapursb", 0b00, 0b10, GPR64>;
4025defm LDAPURH    : BaseLoadUnscaleV84<"ldapurh",  0b01, 0b01, GPR32>;
4026defm LDAPURSHW  : BaseLoadUnscaleV84<"ldapursh", 0b01, 0b11, GPR32>;
4027defm LDAPURSHX  : BaseLoadUnscaleV84<"ldapursh", 0b01, 0b10, GPR64>;
4028defm LDAPUR     : BaseLoadUnscaleV84<"ldapur",   0b10, 0b01, GPR32>;
4029defm LDAPURSW   : BaseLoadUnscaleV84<"ldapursw", 0b10, 0b10, GPR64>;
4030defm LDAPURX    : BaseLoadUnscaleV84<"ldapur",   0b11, 0b01, GPR64>;
4031}
4032
4033// Match all store 64 bits width whose type is compatible with FPR64
4034def : Pat<(store (v1f64 FPR64:$Rt), (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
4035          (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
4036def : Pat<(store (v1i64 FPR64:$Rt), (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
4037          (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
4038
4039let AddedComplexity = 10 in {
4040
4041let Predicates = [IsLE] in {
4042  // We must use ST1 to store vectors in big-endian.
4043  def : Pat<(store (v2f32 FPR64:$Rt),
4044                   (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
4045            (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
4046  def : Pat<(store (v8i8 FPR64:$Rt),
4047                   (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
4048            (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
4049  def : Pat<(store (v4i16 FPR64:$Rt),
4050                   (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
4051            (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
4052  def : Pat<(store (v2i32 FPR64:$Rt),
4053                   (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
4054            (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
4055  def : Pat<(store (v4f16 FPR64:$Rt),
4056                   (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
4057            (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
4058  def : Pat<(store (v4bf16 FPR64:$Rt),
4059                   (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
4060            (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
4061}
4062
4063// Match all store 128 bits width whose type is compatible with FPR128
4064def : Pat<(store (f128 FPR128:$Rt), (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
4065          (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
4066
4067let Predicates = [IsLE] in {
4068  // We must use ST1 to store vectors in big-endian.
4069  def : Pat<(store (v4f32 FPR128:$Rt),
4070                   (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
4071            (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
4072  def : Pat<(store (v2f64 FPR128:$Rt),
4073                   (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
4074            (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
4075  def : Pat<(store (v16i8 FPR128:$Rt),
4076                   (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
4077            (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
4078  def : Pat<(store (v8i16 FPR128:$Rt),
4079                   (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
4080            (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
4081  def : Pat<(store (v4i32 FPR128:$Rt),
4082                   (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
4083            (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
4084  def : Pat<(store (v2i64 FPR128:$Rt),
4085                   (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
4086            (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
4087  def : Pat<(store (v2f64 FPR128:$Rt),
4088                   (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
4089            (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
4090  def : Pat<(store (v8f16 FPR128:$Rt),
4091                   (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
4092            (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
4093  def : Pat<(store (v8bf16 FPR128:$Rt),
4094                   (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
4095            (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
4096}
4097
4098} // AddedComplexity = 10
4099
4100// unscaled i64 truncating stores
4101def : Pat<(truncstorei32 GPR64:$Rt, (am_unscaled32 GPR64sp:$Rn, simm9:$offset)),
4102  (STURWi (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, simm9:$offset)>;
4103def : Pat<(truncstorei16 GPR64:$Rt, (am_unscaled16 GPR64sp:$Rn, simm9:$offset)),
4104  (STURHHi (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, simm9:$offset)>;
4105def : Pat<(truncstorei8 GPR64:$Rt, (am_unscaled8 GPR64sp:$Rn, simm9:$offset)),
4106  (STURBBi (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, simm9:$offset)>;
4107
4108// Match stores from lane 0 to the appropriate subreg's store.
4109multiclass VecStoreULane0Pat<SDPatternOperator StoreOp,
4110                             ValueType VTy, ValueType STy,
4111                             ValueType SubRegTy,
4112                             SubRegIndex SubRegIdx, Instruction STR> {
4113  defm : VecStoreLane0Pat<am_unscaled128, StoreOp, VTy, STy, SubRegTy, SubRegIdx, simm9, STR>;
4114}
4115
4116let AddedComplexity = 19 in {
4117  defm : VecStoreULane0Pat<truncstorei16, v8i16, i32, f16, hsub, STURHi>;
4118  defm : VecStoreULane0Pat<store,         v8f16, f16, f16, hsub, STURHi>;
4119  defm : VecStoreULane0Pat<store,         v4i32, i32, i32, ssub, STURSi>;
4120  defm : VecStoreULane0Pat<store,         v4f32, f32, i32, ssub, STURSi>;
4121  defm : VecStoreULane0Pat<store,         v2i64, i64, i64, dsub, STURDi>;
4122  defm : VecStoreULane0Pat<store,         v2f64, f64, i64, dsub, STURDi>;
4123}
4124
4125//---
4126// STR mnemonics fall back to STUR for negative or unaligned offsets.
4127def : InstAlias<"str $Rt, [$Rn, $offset]",
4128                (STURXi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb64:$offset), 0>;
4129def : InstAlias<"str $Rt, [$Rn, $offset]",
4130                (STURWi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
4131def : InstAlias<"str $Rt, [$Rn, $offset]",
4132                (STURBi FPR8Op:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
4133def : InstAlias<"str $Rt, [$Rn, $offset]",
4134                (STURHi FPR16Op:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
4135def : InstAlias<"str $Rt, [$Rn, $offset]",
4136                (STURSi FPR32Op:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
4137def : InstAlias<"str $Rt, [$Rn, $offset]",
4138                (STURDi FPR64Op:$Rt, GPR64sp:$Rn, simm9_offset_fb64:$offset), 0>;
4139def : InstAlias<"str $Rt, [$Rn, $offset]",
4140                (STURQi FPR128Op:$Rt, GPR64sp:$Rn, simm9_offset_fb128:$offset), 0>;
4141
4142def : InstAlias<"strb $Rt, [$Rn, $offset]",
4143                (STURBBi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
4144def : InstAlias<"strh $Rt, [$Rn, $offset]",
4145                (STURHHi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
4146
4147//---
4148// (unscaled immediate, unprivileged)
4149defm STTRW : StoreUnprivileged<0b10, 0, 0b00, GPR32, "sttr">;
4150defm STTRX : StoreUnprivileged<0b11, 0, 0b00, GPR64, "sttr">;
4151
4152defm STTRH : StoreUnprivileged<0b01, 0, 0b00, GPR32, "sttrh">;
4153defm STTRB : StoreUnprivileged<0b00, 0, 0b00, GPR32, "sttrb">;
4154
4155//---
4156// (immediate pre-indexed)
4157def STRWpre : StorePreIdx<0b10, 0, 0b00, GPR32z, "str",  pre_store, i32>;
4158def STRXpre : StorePreIdx<0b11, 0, 0b00, GPR64z, "str",  pre_store, i64>;
4159def STRBpre : StorePreIdx<0b00, 1, 0b00, FPR8Op,  "str",  pre_store, i8>;
4160def STRHpre : StorePreIdx<0b01, 1, 0b00, FPR16Op, "str",  pre_store, f16>;
4161def STRSpre : StorePreIdx<0b10, 1, 0b00, FPR32Op, "str",  pre_store, f32>;
4162def STRDpre : StorePreIdx<0b11, 1, 0b00, FPR64Op, "str",  pre_store, f64>;
4163def STRQpre : StorePreIdx<0b00, 1, 0b10, FPR128Op, "str", pre_store, f128>;
4164
4165def STRBBpre : StorePreIdx<0b00, 0, 0b00, GPR32z, "strb", pre_truncsti8,  i32>;
4166def STRHHpre : StorePreIdx<0b01, 0, 0b00, GPR32z, "strh", pre_truncsti16, i32>;
4167
4168// truncstore i64
4169def : Pat<(pre_truncsti32 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
4170  (STRWpre (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
4171           simm9:$off)>;
4172def : Pat<(pre_truncsti16 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
4173  (STRHHpre (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
4174            simm9:$off)>;
4175def : Pat<(pre_truncsti8 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
4176  (STRBBpre (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
4177            simm9:$off)>;
4178
4179def : Pat<(pre_store (v8i8 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
4180          (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
4181def : Pat<(pre_store (v4i16 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
4182          (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
4183def : Pat<(pre_store (v2i32 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
4184          (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
4185def : Pat<(pre_store (v2f32 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
4186          (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
4187def : Pat<(pre_store (v1i64 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
4188          (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
4189def : Pat<(pre_store (v1f64 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
4190          (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
4191def : Pat<(pre_store (v4f16 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
4192          (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
4193
4194def : Pat<(pre_store (v16i8 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
4195          (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
4196def : Pat<(pre_store (v8i16 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
4197          (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
4198def : Pat<(pre_store (v4i32 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
4199          (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
4200def : Pat<(pre_store (v4f32 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
4201          (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
4202def : Pat<(pre_store (v2i64 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
4203          (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
4204def : Pat<(pre_store (v2f64 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
4205          (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
4206def : Pat<(pre_store (v8f16 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
4207          (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
4208
4209//---
4210// (immediate post-indexed)
4211def STRWpost : StorePostIdx<0b10, 0, 0b00, GPR32z,  "str", post_store, i32>;
4212def STRXpost : StorePostIdx<0b11, 0, 0b00, GPR64z,  "str", post_store, i64>;
4213def STRBpost : StorePostIdx<0b00, 1, 0b00, FPR8Op,   "str", post_store, i8>;
4214def STRHpost : StorePostIdx<0b01, 1, 0b00, FPR16Op,  "str", post_store, f16>;
4215def STRSpost : StorePostIdx<0b10, 1, 0b00, FPR32Op,  "str", post_store, f32>;
4216def STRDpost : StorePostIdx<0b11, 1, 0b00, FPR64Op,  "str", post_store, f64>;
4217def STRQpost : StorePostIdx<0b00, 1, 0b10, FPR128Op, "str", post_store, f128>;
4218
4219def STRBBpost : StorePostIdx<0b00, 0, 0b00, GPR32z, "strb", post_truncsti8, i32>;
4220def STRHHpost : StorePostIdx<0b01, 0, 0b00, GPR32z, "strh", post_truncsti16, i32>;
4221
4222// truncstore i64
4223def : Pat<(post_truncsti32 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
4224  (STRWpost (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
4225            simm9:$off)>;
4226def : Pat<(post_truncsti16 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
4227  (STRHHpost (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
4228             simm9:$off)>;
4229def : Pat<(post_truncsti8 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
4230  (STRBBpost (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
4231             simm9:$off)>;
4232
4233def : Pat<(post_store (bf16 FPR16:$Rt), GPR64sp:$addr, simm9:$off),
4234          (STRHpost FPR16:$Rt, GPR64sp:$addr, simm9:$off)>;
4235
4236def : Pat<(post_store (v8i8 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
4237          (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
4238def : Pat<(post_store (v4i16 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
4239          (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
4240def : Pat<(post_store (v2i32 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
4241          (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
4242def : Pat<(post_store (v2f32 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
4243          (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
4244def : Pat<(post_store (v1i64 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
4245          (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
4246def : Pat<(post_store (v1f64 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
4247          (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
4248def : Pat<(post_store (v4f16 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
4249          (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
4250def : Pat<(post_store (v4bf16 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
4251          (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
4252
4253def : Pat<(post_store (v16i8 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
4254          (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
4255def : Pat<(post_store (v8i16 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
4256          (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
4257def : Pat<(post_store (v4i32 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
4258          (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
4259def : Pat<(post_store (v4f32 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
4260          (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
4261def : Pat<(post_store (v2i64 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
4262          (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
4263def : Pat<(post_store (v2f64 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
4264          (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
4265def : Pat<(post_store (v8f16 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
4266          (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
4267def : Pat<(post_store (v8bf16 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
4268          (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
4269
4270//===----------------------------------------------------------------------===//
4271// Load/store exclusive instructions.
4272//===----------------------------------------------------------------------===//
4273
4274def LDARW  : LoadAcquire   <0b10, 1, 1, 0, 1, GPR32, "ldar">;
4275def LDARX  : LoadAcquire   <0b11, 1, 1, 0, 1, GPR64, "ldar">;
4276def LDARB  : LoadAcquire   <0b00, 1, 1, 0, 1, GPR32, "ldarb">;
4277def LDARH  : LoadAcquire   <0b01, 1, 1, 0, 1, GPR32, "ldarh">;
4278
4279def LDAXRW : LoadExclusive <0b10, 0, 1, 0, 1, GPR32, "ldaxr">;
4280def LDAXRX : LoadExclusive <0b11, 0, 1, 0, 1, GPR64, "ldaxr">;
4281def LDAXRB : LoadExclusive <0b00, 0, 1, 0, 1, GPR32, "ldaxrb">;
4282def LDAXRH : LoadExclusive <0b01, 0, 1, 0, 1, GPR32, "ldaxrh">;
4283
4284def LDXRW  : LoadExclusive <0b10, 0, 1, 0, 0, GPR32, "ldxr">;
4285def LDXRX  : LoadExclusive <0b11, 0, 1, 0, 0, GPR64, "ldxr">;
4286def LDXRB  : LoadExclusive <0b00, 0, 1, 0, 0, GPR32, "ldxrb">;
4287def LDXRH  : LoadExclusive <0b01, 0, 1, 0, 0, GPR32, "ldxrh">;
4288
4289def STLRW  : StoreRelease  <0b10, 1, 0, 0, 1, GPR32, "stlr">;
4290def STLRX  : StoreRelease  <0b11, 1, 0, 0, 1, GPR64, "stlr">;
4291def STLRB  : StoreRelease  <0b00, 1, 0, 0, 1, GPR32, "stlrb">;
4292def STLRH  : StoreRelease  <0b01, 1, 0, 0, 1, GPR32, "stlrh">;
4293
4294/*
4295Aliases for when offset=0. Note that in contrast to LoadAcquire which has a $Rn
4296of type GPR64sp0, we deliberately choose to make $Rn of type GPR64sp and add an
4297alias for the case of immediate #0. This is because new STLR versions (from
4298LRCPC3 extension) do have a non-zero immediate value, so GPR64sp0 is not
4299appropriate anymore (it parses and discards the optional zero). This is not the
4300case for LoadAcquire because the new LRCPC3 LDAR instructions are post-indexed,
4301and the immediate values are not inside the [] brackets and thus not accepted
4302by GPR64sp0 parser.
4303*/
4304def STLRW0  : InstAlias<"stlr\t$Rt, [$Rn, #0]" , (STLRW   GPR32: $Rt, GPR64sp:$Rn)>;
4305def STLRX0  : InstAlias<"stlr\t$Rt, [$Rn, #0]" , (STLRX   GPR64: $Rt, GPR64sp:$Rn)>;
4306def STLRB0  : InstAlias<"stlrb\t$Rt, [$Rn, #0]", (STLRB   GPR32: $Rt, GPR64sp:$Rn)>;
4307def STLRH0  : InstAlias<"stlrh\t$Rt, [$Rn, #0]", (STLRH   GPR32: $Rt, GPR64sp:$Rn)>;
4308
4309def STLXRW : StoreExclusive<0b10, 0, 0, 0, 1, GPR32, "stlxr">;
4310def STLXRX : StoreExclusive<0b11, 0, 0, 0, 1, GPR64, "stlxr">;
4311def STLXRB : StoreExclusive<0b00, 0, 0, 0, 1, GPR32, "stlxrb">;
4312def STLXRH : StoreExclusive<0b01, 0, 0, 0, 1, GPR32, "stlxrh">;
4313
4314def STXRW  : StoreExclusive<0b10, 0, 0, 0, 0, GPR32, "stxr">;
4315def STXRX  : StoreExclusive<0b11, 0, 0, 0, 0, GPR64, "stxr">;
4316def STXRB  : StoreExclusive<0b00, 0, 0, 0, 0, GPR32, "stxrb">;
4317def STXRH  : StoreExclusive<0b01, 0, 0, 0, 0, GPR32, "stxrh">;
4318
4319def LDAXPW : LoadExclusivePair<0b10, 0, 1, 1, 1, GPR32, "ldaxp">;
4320def LDAXPX : LoadExclusivePair<0b11, 0, 1, 1, 1, GPR64, "ldaxp">;
4321
4322def LDXPW  : LoadExclusivePair<0b10, 0, 1, 1, 0, GPR32, "ldxp">;
4323def LDXPX  : LoadExclusivePair<0b11, 0, 1, 1, 0, GPR64, "ldxp">;
4324
4325def STLXPW : StoreExclusivePair<0b10, 0, 0, 1, 1, GPR32, "stlxp">;
4326def STLXPX : StoreExclusivePair<0b11, 0, 0, 1, 1, GPR64, "stlxp">;
4327
4328def STXPW  : StoreExclusivePair<0b10, 0, 0, 1, 0, GPR32, "stxp">;
4329def STXPX  : StoreExclusivePair<0b11, 0, 0, 1, 0, GPR64, "stxp">;
4330
4331let Predicates = [HasLOR] in {
4332  // v8.1a "Limited Order Region" extension load-acquire instructions
4333  def LDLARW  : LoadAcquire   <0b10, 1, 1, 0, 0, GPR32, "ldlar">;
4334  def LDLARX  : LoadAcquire   <0b11, 1, 1, 0, 0, GPR64, "ldlar">;
4335  def LDLARB  : LoadAcquire   <0b00, 1, 1, 0, 0, GPR32, "ldlarb">;
4336  def LDLARH  : LoadAcquire   <0b01, 1, 1, 0, 0, GPR32, "ldlarh">;
4337
4338  // v8.1a "Limited Order Region" extension store-release instructions
4339  def STLLRW  : StoreRelease   <0b10, 1, 0, 0, 0, GPR32, "stllr">;
4340  def STLLRX  : StoreRelease   <0b11, 1, 0, 0, 0, GPR64, "stllr">;
4341  def STLLRB  : StoreRelease   <0b00, 1, 0, 0, 0, GPR32, "stllrb">;
4342  def STLLRH  : StoreRelease   <0b01, 1, 0, 0, 0, GPR32, "stllrh">;
4343
4344  // Aliases for when offset=0
4345  def STLLRW0 : InstAlias<"stllr\t$Rt,  [$Rn, #0]",  (STLLRW   GPR32: $Rt, GPR64sp:$Rn)>;
4346  def STLLRX0 : InstAlias<"stllr\t$Rt,  [$Rn, #0]",  (STLLRX   GPR64: $Rt, GPR64sp:$Rn)>;
4347  def STLLRB0 : InstAlias<"stllrb\t$Rt, [$Rn, #0]",  (STLLRB   GPR32: $Rt, GPR64sp:$Rn)>;
4348  def STLLRH0 : InstAlias<"stllrh\t$Rt, [$Rn, #0]",  (STLLRH   GPR32: $Rt, GPR64sp:$Rn)>;
4349}
4350
4351//===----------------------------------------------------------------------===//
4352// Scaled floating point to integer conversion instructions.
4353//===----------------------------------------------------------------------===//
4354
4355defm FCVTAS : FPToIntegerUnscaled<0b00, 0b100, "fcvtas", int_aarch64_neon_fcvtas>;
4356defm FCVTAU : FPToIntegerUnscaled<0b00, 0b101, "fcvtau", int_aarch64_neon_fcvtau>;
4357defm FCVTMS : FPToIntegerUnscaled<0b10, 0b000, "fcvtms", int_aarch64_neon_fcvtms>;
4358defm FCVTMU : FPToIntegerUnscaled<0b10, 0b001, "fcvtmu", int_aarch64_neon_fcvtmu>;
4359defm FCVTNS : FPToIntegerUnscaled<0b00, 0b000, "fcvtns", int_aarch64_neon_fcvtns>;
4360defm FCVTNU : FPToIntegerUnscaled<0b00, 0b001, "fcvtnu", int_aarch64_neon_fcvtnu>;
4361defm FCVTPS : FPToIntegerUnscaled<0b01, 0b000, "fcvtps", int_aarch64_neon_fcvtps>;
4362defm FCVTPU : FPToIntegerUnscaled<0b01, 0b001, "fcvtpu", int_aarch64_neon_fcvtpu>;
4363defm FCVTZS : FPToIntegerUnscaled<0b11, 0b000, "fcvtzs", any_fp_to_sint>;
4364defm FCVTZU : FPToIntegerUnscaled<0b11, 0b001, "fcvtzu", any_fp_to_uint>;
4365defm FCVTZS : FPToIntegerScaled<0b11, 0b000, "fcvtzs", any_fp_to_sint>;
4366defm FCVTZU : FPToIntegerScaled<0b11, 0b001, "fcvtzu", any_fp_to_uint>;
4367
4368// AArch64's FCVT instructions saturate when out of range.
4369multiclass FPToIntegerSatPats<SDNode to_int_sat, string INST> {
4370  let Predicates = [HasFullFP16] in {
4371  def : Pat<(i32 (to_int_sat f16:$Rn, i32)),
4372            (!cast<Instruction>(INST # UWHr) f16:$Rn)>;
4373  def : Pat<(i64 (to_int_sat f16:$Rn, i64)),
4374            (!cast<Instruction>(INST # UXHr) f16:$Rn)>;
4375  }
4376  def : Pat<(i32 (to_int_sat f32:$Rn, i32)),
4377            (!cast<Instruction>(INST # UWSr) f32:$Rn)>;
4378  def : Pat<(i64 (to_int_sat f32:$Rn, i64)),
4379            (!cast<Instruction>(INST # UXSr) f32:$Rn)>;
4380  def : Pat<(i32 (to_int_sat f64:$Rn, i32)),
4381            (!cast<Instruction>(INST # UWDr) f64:$Rn)>;
4382  def : Pat<(i64 (to_int_sat f64:$Rn, i64)),
4383            (!cast<Instruction>(INST # UXDr) f64:$Rn)>;
4384
4385  let Predicates = [HasFullFP16] in {
4386  def : Pat<(i32 (to_int_sat (fmul f16:$Rn, fixedpoint_f16_i32:$scale), i32)),
4387            (!cast<Instruction>(INST # SWHri) $Rn, $scale)>;
4388  def : Pat<(i64 (to_int_sat (fmul f16:$Rn, fixedpoint_f16_i64:$scale), i64)),
4389            (!cast<Instruction>(INST # SXHri) $Rn, $scale)>;
4390  }
4391  def : Pat<(i32 (to_int_sat (fmul f32:$Rn, fixedpoint_f32_i32:$scale), i32)),
4392            (!cast<Instruction>(INST # SWSri) $Rn, $scale)>;
4393  def : Pat<(i64 (to_int_sat (fmul f32:$Rn, fixedpoint_f32_i64:$scale), i64)),
4394            (!cast<Instruction>(INST # SXSri) $Rn, $scale)>;
4395  def : Pat<(i32 (to_int_sat (fmul f64:$Rn, fixedpoint_f64_i32:$scale), i32)),
4396            (!cast<Instruction>(INST # SWDri) $Rn, $scale)>;
4397  def : Pat<(i64 (to_int_sat (fmul f64:$Rn, fixedpoint_f64_i64:$scale), i64)),
4398            (!cast<Instruction>(INST # SXDri) $Rn, $scale)>;
4399}
4400
4401defm : FPToIntegerSatPats<fp_to_sint_sat, "FCVTZS">;
4402defm : FPToIntegerSatPats<fp_to_uint_sat, "FCVTZU">;
4403
4404multiclass FPToIntegerIntPats<Intrinsic round, string INST> {
4405  let Predicates = [HasFullFP16] in {
4406  def : Pat<(i32 (round f16:$Rn)), (!cast<Instruction>(INST # UWHr) $Rn)>;
4407  def : Pat<(i64 (round f16:$Rn)), (!cast<Instruction>(INST # UXHr) $Rn)>;
4408  }
4409  def : Pat<(i32 (round f32:$Rn)), (!cast<Instruction>(INST # UWSr) $Rn)>;
4410  def : Pat<(i64 (round f32:$Rn)), (!cast<Instruction>(INST # UXSr) $Rn)>;
4411  def : Pat<(i32 (round f64:$Rn)), (!cast<Instruction>(INST # UWDr) $Rn)>;
4412  def : Pat<(i64 (round f64:$Rn)), (!cast<Instruction>(INST # UXDr) $Rn)>;
4413
4414  let Predicates = [HasFullFP16] in {
4415  def : Pat<(i32 (round (fmul f16:$Rn, fixedpoint_f16_i32:$scale))),
4416            (!cast<Instruction>(INST # SWHri) $Rn, $scale)>;
4417  def : Pat<(i64 (round (fmul f16:$Rn, fixedpoint_f16_i64:$scale))),
4418            (!cast<Instruction>(INST # SXHri) $Rn, $scale)>;
4419  }
4420  def : Pat<(i32 (round (fmul f32:$Rn, fixedpoint_f32_i32:$scale))),
4421            (!cast<Instruction>(INST # SWSri) $Rn, $scale)>;
4422  def : Pat<(i64 (round (fmul f32:$Rn, fixedpoint_f32_i64:$scale))),
4423            (!cast<Instruction>(INST # SXSri) $Rn, $scale)>;
4424  def : Pat<(i32 (round (fmul f64:$Rn, fixedpoint_f64_i32:$scale))),
4425            (!cast<Instruction>(INST # SWDri) $Rn, $scale)>;
4426  def : Pat<(i64 (round (fmul f64:$Rn, fixedpoint_f64_i64:$scale))),
4427            (!cast<Instruction>(INST # SXDri) $Rn, $scale)>;
4428}
4429
4430defm : FPToIntegerIntPats<int_aarch64_neon_fcvtzs, "FCVTZS">;
4431defm : FPToIntegerIntPats<int_aarch64_neon_fcvtzu, "FCVTZU">;
4432
4433multiclass FPToIntegerPats<SDNode to_int, SDNode to_int_sat, SDNode round, string INST> {
4434  def : Pat<(i32 (to_int (round f32:$Rn))),
4435            (!cast<Instruction>(INST # UWSr) f32:$Rn)>;
4436  def : Pat<(i64 (to_int (round f32:$Rn))),
4437            (!cast<Instruction>(INST # UXSr) f32:$Rn)>;
4438  def : Pat<(i32 (to_int (round f64:$Rn))),
4439            (!cast<Instruction>(INST # UWDr) f64:$Rn)>;
4440  def : Pat<(i64 (to_int (round f64:$Rn))),
4441            (!cast<Instruction>(INST # UXDr) f64:$Rn)>;
4442
4443  // These instructions saturate like fp_to_[su]int_sat.
4444  let Predicates = [HasFullFP16] in {
4445  def : Pat<(i32 (to_int_sat (round f16:$Rn), i32)),
4446            (!cast<Instruction>(INST # UWHr) f16:$Rn)>;
4447  def : Pat<(i64 (to_int_sat (round f16:$Rn), i64)),
4448            (!cast<Instruction>(INST # UXHr) f16:$Rn)>;
4449  }
4450  def : Pat<(i32 (to_int_sat (round f32:$Rn), i32)),
4451            (!cast<Instruction>(INST # UWSr) f32:$Rn)>;
4452  def : Pat<(i64 (to_int_sat (round f32:$Rn), i64)),
4453            (!cast<Instruction>(INST # UXSr) f32:$Rn)>;
4454  def : Pat<(i32 (to_int_sat (round f64:$Rn), i32)),
4455            (!cast<Instruction>(INST # UWDr) f64:$Rn)>;
4456  def : Pat<(i64 (to_int_sat (round f64:$Rn), i64)),
4457            (!cast<Instruction>(INST # UXDr) f64:$Rn)>;
4458}
4459
4460defm : FPToIntegerPats<fp_to_sint, fp_to_sint_sat, fceil,  "FCVTPS">;
4461defm : FPToIntegerPats<fp_to_uint, fp_to_uint_sat, fceil,  "FCVTPU">;
4462defm : FPToIntegerPats<fp_to_sint, fp_to_sint_sat, ffloor, "FCVTMS">;
4463defm : FPToIntegerPats<fp_to_uint, fp_to_uint_sat, ffloor, "FCVTMU">;
4464defm : FPToIntegerPats<fp_to_sint, fp_to_sint_sat, ftrunc, "FCVTZS">;
4465defm : FPToIntegerPats<fp_to_uint, fp_to_uint_sat, ftrunc, "FCVTZU">;
4466defm : FPToIntegerPats<fp_to_sint, fp_to_sint_sat, fround, "FCVTAS">;
4467defm : FPToIntegerPats<fp_to_uint, fp_to_uint_sat, fround, "FCVTAU">;
4468
4469
4470
4471let Predicates = [HasFullFP16] in {
4472  def : Pat<(i32 (any_lround f16:$Rn)),
4473            (FCVTASUWHr f16:$Rn)>;
4474  def : Pat<(i64 (any_lround f16:$Rn)),
4475            (FCVTASUXHr f16:$Rn)>;
4476  def : Pat<(i64 (any_llround f16:$Rn)),
4477            (FCVTASUXHr f16:$Rn)>;
4478}
4479def : Pat<(i32 (any_lround f32:$Rn)),
4480          (FCVTASUWSr f32:$Rn)>;
4481def : Pat<(i32 (any_lround f64:$Rn)),
4482          (FCVTASUWDr f64:$Rn)>;
4483def : Pat<(i64 (any_lround f32:$Rn)),
4484          (FCVTASUXSr f32:$Rn)>;
4485def : Pat<(i64 (any_lround f64:$Rn)),
4486          (FCVTASUXDr f64:$Rn)>;
4487def : Pat<(i64 (any_llround f32:$Rn)),
4488          (FCVTASUXSr f32:$Rn)>;
4489def : Pat<(i64 (any_llround f64:$Rn)),
4490          (FCVTASUXDr f64:$Rn)>;
4491
4492//===----------------------------------------------------------------------===//
4493// Scaled integer to floating point conversion instructions.
4494//===----------------------------------------------------------------------===//
4495
4496defm SCVTF : IntegerToFP<0, "scvtf", any_sint_to_fp>;
4497defm UCVTF : IntegerToFP<1, "ucvtf", any_uint_to_fp>;
4498
4499def : Pat<(f16 (fdiv (f16 (any_sint_to_fp (i32 GPR32:$Rn))), fixedpoint_f16_i32:$scale)),
4500          (SCVTFSWHri GPR32:$Rn, fixedpoint_f16_i32:$scale)>;
4501def : Pat<(f32 (fdiv (f32 (any_sint_to_fp (i32 GPR32:$Rn))), fixedpoint_f32_i32:$scale)),
4502          (SCVTFSWSri GPR32:$Rn, fixedpoint_f32_i32:$scale)>;
4503def : Pat<(f64 (fdiv (f64 (any_sint_to_fp (i32 GPR32:$Rn))), fixedpoint_f64_i32:$scale)),
4504          (SCVTFSWDri GPR32:$Rn, fixedpoint_f64_i32:$scale)>;
4505
4506def : Pat<(f16 (fdiv (f16 (any_sint_to_fp (i64 GPR64:$Rn))), fixedpoint_f16_i64:$scale)),
4507          (SCVTFSXHri GPR64:$Rn, fixedpoint_f16_i64:$scale)>;
4508def : Pat<(f32 (fdiv (f32 (any_sint_to_fp (i64 GPR64:$Rn))), fixedpoint_f32_i64:$scale)),
4509          (SCVTFSXSri GPR64:$Rn, fixedpoint_f32_i64:$scale)>;
4510def : Pat<(f64 (fdiv (f64 (any_sint_to_fp (i64 GPR64:$Rn))), fixedpoint_f64_i64:$scale)),
4511          (SCVTFSXDri GPR64:$Rn, fixedpoint_f64_i64:$scale)>;
4512
4513def : Pat<(f16 (fdiv (f16 (any_uint_to_fp (i64 GPR64:$Rn))), fixedpoint_f16_i64:$scale)),
4514          (UCVTFSXHri GPR64:$Rn, fixedpoint_f16_i64:$scale)>;
4515def : Pat<(f32 (fdiv (f32 (any_uint_to_fp (i64 GPR64:$Rn))), fixedpoint_f32_i64:$scale)),
4516          (UCVTFSXSri GPR64:$Rn, fixedpoint_f32_i64:$scale)>;
4517def : Pat<(f64 (fdiv (f64 (any_uint_to_fp (i64 GPR64:$Rn))), fixedpoint_f64_i64:$scale)),
4518          (UCVTFSXDri GPR64:$Rn, fixedpoint_f64_i64:$scale)>;
4519
4520def : Pat<(f16 (fdiv (f16 (any_uint_to_fp (i32 GPR32:$Rn))), fixedpoint_f16_i32:$scale)),
4521          (UCVTFSWHri GPR32:$Rn, fixedpoint_f16_i32:$scale)>;
4522def : Pat<(f32 (fdiv (f32 (any_uint_to_fp (i32 GPR32:$Rn))), fixedpoint_f32_i32:$scale)),
4523          (UCVTFSWSri GPR32:$Rn, fixedpoint_f32_i32:$scale)>;
4524def : Pat<(f64 (fdiv (f64 (any_uint_to_fp (i32 GPR32:$Rn))), fixedpoint_f64_i32:$scale)),
4525          (UCVTFSWDri GPR32:$Rn, fixedpoint_f64_i32:$scale)>;
4526
4527//===----------------------------------------------------------------------===//
4528// Unscaled integer to floating point conversion instruction.
4529//===----------------------------------------------------------------------===//
4530
4531defm FMOV : UnscaledConversion<"fmov">;
4532
4533// Add pseudo ops for FMOV 0 so we can mark them as isReMaterializable
4534let isReMaterializable = 1, isCodeGenOnly = 1, isAsCheapAsAMove = 1 in {
4535def FMOVH0 : Pseudo<(outs FPR16:$Rd), (ins), [(set f16:$Rd, (fpimm0))]>,
4536    Sched<[WriteF]>;
4537def FMOVS0 : Pseudo<(outs FPR32:$Rd), (ins), [(set f32:$Rd, (fpimm0))]>,
4538    Sched<[WriteF]>;
4539def FMOVD0 : Pseudo<(outs FPR64:$Rd), (ins), [(set f64:$Rd, (fpimm0))]>,
4540    Sched<[WriteF]>;
4541}
4542
4543// Similarly add aliases
4544def : InstAlias<"fmov $Rd, #0.0", (FMOVWHr FPR16:$Rd, WZR), 0>,
4545    Requires<[HasFullFP16]>;
4546def : InstAlias<"fmov $Rd, #0.0", (FMOVWSr FPR32:$Rd, WZR), 0>;
4547def : InstAlias<"fmov $Rd, #0.0", (FMOVXDr FPR64:$Rd, XZR), 0>;
4548
4549def : Pat<(bf16 fpimm0),
4550          (FMOVH0)>;
4551
4552// Pattern for FP16 and BF16 immediates
4553let Predicates = [HasFullFP16] in {
4554  def : Pat<(f16 fpimm:$in),
4555            (FMOVWHr (MOVi32imm (bitcast_fpimm_to_i32 f16:$in)))>;
4556
4557  def : Pat<(bf16 fpimm:$in),
4558            (FMOVWHr (MOVi32imm (bitcast_fpimm_to_i32 bf16:$in)))>;
4559}
4560
4561//===----------------------------------------------------------------------===//
4562// Floating point conversion instruction.
4563//===----------------------------------------------------------------------===//
4564
4565defm FCVT : FPConversion<"fcvt">;
4566
4567//===----------------------------------------------------------------------===//
4568// Floating point single operand instructions.
4569//===----------------------------------------------------------------------===//
4570
4571defm FABS   : SingleOperandFPDataNoException<0b0001, "fabs", fabs>;
4572defm FMOV   : SingleOperandFPDataNoException<0b0000, "fmov">;
4573defm FNEG   : SingleOperandFPDataNoException<0b0010, "fneg", fneg>;
4574defm FRINTA : SingleOperandFPData<0b1100, "frinta", any_fround>;
4575defm FRINTI : SingleOperandFPData<0b1111, "frinti", any_fnearbyint>;
4576defm FRINTM : SingleOperandFPData<0b1010, "frintm", any_ffloor>;
4577defm FRINTN : SingleOperandFPData<0b1000, "frintn", any_froundeven>;
4578defm FRINTP : SingleOperandFPData<0b1001, "frintp", any_fceil>;
4579
4580defm FRINTX : SingleOperandFPData<0b1110, "frintx", any_frint>;
4581defm FRINTZ : SingleOperandFPData<0b1011, "frintz", any_ftrunc>;
4582
4583let SchedRW = [WriteFDiv] in {
4584defm FSQRT  : SingleOperandFPData<0b0011, "fsqrt", any_fsqrt>;
4585}
4586
4587let Predicates = [HasFRInt3264] in {
4588  defm FRINT32Z : FRIntNNT<0b00, "frint32z", int_aarch64_frint32z>;
4589  defm FRINT64Z : FRIntNNT<0b10, "frint64z", int_aarch64_frint64z>;
4590  defm FRINT32X : FRIntNNT<0b01, "frint32x", int_aarch64_frint32x>;
4591  defm FRINT64X : FRIntNNT<0b11, "frint64x", int_aarch64_frint64x>;
4592} // HasFRInt3264
4593
4594// Pattern to convert 1x64 vector intrinsics to equivalent scalar instructions
4595def : Pat<(v1f64 (int_aarch64_neon_frint32z (v1f64 FPR64:$Rn))),
4596          (FRINT32ZDr FPR64:$Rn)>;
4597def : Pat<(v1f64 (int_aarch64_neon_frint64z (v1f64 FPR64:$Rn))),
4598          (FRINT64ZDr FPR64:$Rn)>;
4599def : Pat<(v1f64 (int_aarch64_neon_frint32x (v1f64 FPR64:$Rn))),
4600          (FRINT32XDr FPR64:$Rn)>;
4601def : Pat<(v1f64 (int_aarch64_neon_frint64x (v1f64 FPR64:$Rn))),
4602          (FRINT64XDr FPR64:$Rn)>;
4603
4604// Emitting strict_lrint as two instructions is valid as any exceptions that
4605// occur will happen in exactly one of the instructions (e.g. if the input is
4606// not an integer the inexact exception will happen in the FRINTX but not then
4607// in the FCVTZS as the output of FRINTX is an integer).
4608let Predicates = [HasFullFP16] in {
4609  def : Pat<(i32 (any_lrint f16:$Rn)),
4610            (FCVTZSUWHr (FRINTXHr f16:$Rn))>;
4611  def : Pat<(i64 (any_lrint f16:$Rn)),
4612            (FCVTZSUXHr (FRINTXHr f16:$Rn))>;
4613  def : Pat<(i64 (any_llrint f16:$Rn)),
4614            (FCVTZSUXHr (FRINTXHr f16:$Rn))>;
4615}
4616def : Pat<(i32 (any_lrint f32:$Rn)),
4617          (FCVTZSUWSr (FRINTXSr f32:$Rn))>;
4618def : Pat<(i32 (any_lrint f64:$Rn)),
4619          (FCVTZSUWDr (FRINTXDr f64:$Rn))>;
4620def : Pat<(i64 (any_lrint f32:$Rn)),
4621          (FCVTZSUXSr (FRINTXSr f32:$Rn))>;
4622def : Pat<(i64 (any_lrint f64:$Rn)),
4623          (FCVTZSUXDr (FRINTXDr f64:$Rn))>;
4624def : Pat<(i64 (any_llrint f32:$Rn)),
4625          (FCVTZSUXSr (FRINTXSr f32:$Rn))>;
4626def : Pat<(i64 (any_llrint f64:$Rn)),
4627          (FCVTZSUXDr (FRINTXDr f64:$Rn))>;
4628
4629//===----------------------------------------------------------------------===//
4630// Floating point two operand instructions.
4631//===----------------------------------------------------------------------===//
4632
4633defm FADD   : TwoOperandFPData<0b0010, "fadd", any_fadd>;
4634let SchedRW = [WriteFDiv] in {
4635defm FDIV   : TwoOperandFPData<0b0001, "fdiv", any_fdiv>;
4636}
4637defm FMAXNM : TwoOperandFPData<0b0110, "fmaxnm", any_fmaxnum>;
4638defm FMAX   : TwoOperandFPData<0b0100, "fmax", any_fmaximum>;
4639defm FMINNM : TwoOperandFPData<0b0111, "fminnm", any_fminnum>;
4640defm FMIN   : TwoOperandFPData<0b0101, "fmin", any_fminimum>;
4641let SchedRW = [WriteFMul] in {
4642defm FMUL   : TwoOperandFPData<0b0000, "fmul", any_fmul>;
4643defm FNMUL  : TwoOperandFPDataNeg<0b1000, "fnmul", any_fmul>;
4644}
4645defm FSUB   : TwoOperandFPData<0b0011, "fsub", any_fsub>;
4646
4647multiclass FMULScalarFromIndexedLane0Patterns<string inst,
4648                                              string inst_f16_suffix,
4649                                              string inst_f32_suffix,
4650                                              string inst_f64_suffix,
4651                                              SDPatternOperator OpNode,
4652                                              list<Predicate> preds = []> {
4653  let Predicates = !listconcat(preds, [HasFullFP16]) in {
4654  def : Pat<(f16 (OpNode (f16 FPR16:$Rn),
4655                         (f16 (vector_extract (v8f16 V128:$Rm), (i64 0))))),
4656            (!cast<Instruction>(inst # inst_f16_suffix)
4657              FPR16:$Rn, (f16 (EXTRACT_SUBREG V128:$Rm, hsub)))>;
4658  }
4659  let Predicates = preds in {
4660  def : Pat<(f32 (OpNode (f32 FPR32:$Rn),
4661                         (f32 (vector_extract (v4f32 V128:$Rm), (i64 0))))),
4662            (!cast<Instruction>(inst # inst_f32_suffix)
4663              FPR32:$Rn, (EXTRACT_SUBREG V128:$Rm, ssub))>;
4664  def : Pat<(f64 (OpNode (f64 FPR64:$Rn),
4665                         (f64 (vector_extract (v2f64 V128:$Rm), (i64 0))))),
4666            (!cast<Instruction>(inst # inst_f64_suffix)
4667              FPR64:$Rn, (EXTRACT_SUBREG V128:$Rm, dsub))>;
4668  }
4669}
4670
4671defm : FMULScalarFromIndexedLane0Patterns<"FMUL", "Hrr", "Srr", "Drr",
4672                                          any_fmul>;
4673
4674// Match reassociated forms of FNMUL.
4675def : Pat<(fmul (fneg FPR16:$a), (f16 FPR16:$b)),
4676          (FNMULHrr FPR16:$a, FPR16:$b)>,
4677          Requires<[HasFullFP16]>;
4678def : Pat<(fmul (fneg FPR32:$a), (f32 FPR32:$b)),
4679          (FNMULSrr FPR32:$a, FPR32:$b)>;
4680def : Pat<(fmul (fneg FPR64:$a), (f64 FPR64:$b)),
4681          (FNMULDrr FPR64:$a, FPR64:$b)>;
4682
4683def : Pat<(v1f64 (fmaximum (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
4684          (FMAXDrr FPR64:$Rn, FPR64:$Rm)>;
4685def : Pat<(v1f64 (fminimum (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
4686          (FMINDrr FPR64:$Rn, FPR64:$Rm)>;
4687def : Pat<(v1f64 (fmaxnum (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
4688          (FMAXNMDrr FPR64:$Rn, FPR64:$Rm)>;
4689def : Pat<(v1f64 (fminnum (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
4690          (FMINNMDrr FPR64:$Rn, FPR64:$Rm)>;
4691
4692//===----------------------------------------------------------------------===//
4693// Floating point three operand instructions.
4694//===----------------------------------------------------------------------===//
4695
4696defm FMADD  : ThreeOperandFPData<0, 0, "fmadd", any_fma>;
4697defm FMSUB  : ThreeOperandFPData<0, 1, "fmsub",
4698     TriOpFrag<(any_fma node:$LHS, (fneg node:$MHS), node:$RHS)> >;
4699defm FNMADD : ThreeOperandFPData<1, 0, "fnmadd",
4700     TriOpFrag<(fneg (any_fma node:$LHS, node:$MHS, node:$RHS))> >;
4701defm FNMSUB : ThreeOperandFPData<1, 1, "fnmsub",
4702     TriOpFrag<(any_fma node:$LHS, node:$MHS, (fneg node:$RHS))> >;
4703
4704// The following def pats catch the case where the LHS of an FMA is negated.
4705// The TriOpFrag above catches the case where the middle operand is negated.
4706
4707// N.b. FMSUB etc have the accumulator at the *end* of (outs), unlike
4708// the NEON variant.
4709
4710// Here we handle first -(a + b*c) for FNMADD:
4711
4712let Predicates = [HasNEON, HasFullFP16] in
4713def : Pat<(f16 (fma (fneg FPR16:$Rn), FPR16:$Rm, FPR16:$Ra)),
4714          (FMSUBHrrr FPR16:$Rn, FPR16:$Rm, FPR16:$Ra)>;
4715
4716def : Pat<(f32 (fma (fneg FPR32:$Rn), FPR32:$Rm, FPR32:$Ra)),
4717          (FMSUBSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
4718
4719def : Pat<(f64 (fma (fneg FPR64:$Rn), FPR64:$Rm, FPR64:$Ra)),
4720          (FMSUBDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
4721
4722// Now it's time for "(-a) + (-b)*c"
4723
4724let Predicates = [HasNEON, HasFullFP16] in
4725def : Pat<(f16 (fma (fneg FPR16:$Rn), FPR16:$Rm, (fneg FPR16:$Ra))),
4726          (FNMADDHrrr FPR16:$Rn, FPR16:$Rm, FPR16:$Ra)>;
4727
4728def : Pat<(f32 (fma (fneg FPR32:$Rn), FPR32:$Rm, (fneg FPR32:$Ra))),
4729          (FNMADDSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
4730
4731def : Pat<(f64 (fma (fneg FPR64:$Rn), FPR64:$Rm, (fneg FPR64:$Ra))),
4732          (FNMADDDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
4733
4734//===----------------------------------------------------------------------===//
4735// Floating point comparison instructions.
4736//===----------------------------------------------------------------------===//
4737
4738defm FCMPE : FPComparison<1, "fcmpe", AArch64strict_fcmpe>;
4739defm FCMP  : FPComparison<0, "fcmp", AArch64any_fcmp>;
4740
4741//===----------------------------------------------------------------------===//
4742// Floating point conditional comparison instructions.
4743//===----------------------------------------------------------------------===//
4744
4745defm FCCMPE : FPCondComparison<1, "fccmpe">;
4746defm FCCMP  : FPCondComparison<0, "fccmp", AArch64fccmp>;
4747
4748//===----------------------------------------------------------------------===//
4749// Floating point conditional select instruction.
4750//===----------------------------------------------------------------------===//
4751
4752defm FCSEL : FPCondSelect<"fcsel">;
4753
4754let Predicates = [HasFullFP16] in
4755def : Pat<(bf16 (AArch64csel (bf16 FPR16:$Rn), (bf16 FPR16:$Rm), (i32 imm:$cond), NZCV)),
4756          (FCSELHrrr FPR16:$Rn, FPR16:$Rm, imm:$cond)>;
4757
4758// CSEL instructions providing f128 types need to be handled by a
4759// pseudo-instruction since the eventual code will need to introduce basic
4760// blocks and control flow.
4761def F128CSEL : Pseudo<(outs FPR128:$Rd),
4762                      (ins FPR128:$Rn, FPR128:$Rm, ccode:$cond),
4763                      [(set (f128 FPR128:$Rd),
4764                            (AArch64csel FPR128:$Rn, FPR128:$Rm,
4765                                       (i32 imm:$cond), NZCV))]> {
4766  let Uses = [NZCV];
4767  let usesCustomInserter = 1;
4768  let hasNoSchedulingInfo = 1;
4769}
4770
4771//===----------------------------------------------------------------------===//
4772// Instructions used for emitting unwind opcodes on ARM64 Windows.
4773//===----------------------------------------------------------------------===//
4774let isPseudo = 1 in {
4775  def SEH_StackAlloc : Pseudo<(outs), (ins i32imm:$size), []>, Sched<[]>;
4776  def SEH_SaveFPLR : Pseudo<(outs), (ins i32imm:$offs), []>, Sched<[]>;
4777  def SEH_SaveFPLR_X : Pseudo<(outs), (ins i32imm:$offs), []>, Sched<[]>;
4778  def SEH_SaveReg : Pseudo<(outs), (ins i32imm:$reg, i32imm:$offs), []>, Sched<[]>;
4779  def SEH_SaveReg_X : Pseudo<(outs), (ins i32imm:$reg, i32imm:$offs), []>, Sched<[]>;
4780  def SEH_SaveRegP : Pseudo<(outs), (ins i32imm:$reg0, i32imm:$reg1, i32imm:$offs), []>, Sched<[]>;
4781  def SEH_SaveRegP_X : Pseudo<(outs), (ins i32imm:$reg0, i32imm:$reg1, i32imm:$offs), []>, Sched<[]>;
4782  def SEH_SaveFReg : Pseudo<(outs), (ins i32imm:$reg, i32imm:$offs), []>, Sched<[]>;
4783  def SEH_SaveFReg_X :  Pseudo<(outs), (ins i32imm:$reg, i32imm:$offs), []>, Sched<[]>;
4784  def SEH_SaveFRegP : Pseudo<(outs), (ins i32imm:$reg0, i32imm:$reg1, i32imm:$offs), []>, Sched<[]>;
4785  def SEH_SaveFRegP_X : Pseudo<(outs), (ins i32imm:$reg0, i32imm:$reg1, i32imm:$offs), []>, Sched<[]>;
4786  def SEH_SetFP : Pseudo<(outs), (ins), []>, Sched<[]>;
4787  def SEH_AddFP : Pseudo<(outs), (ins i32imm:$offs), []>, Sched<[]>;
4788  def SEH_Nop : Pseudo<(outs), (ins), []>, Sched<[]>;
4789  def SEH_PrologEnd : Pseudo<(outs), (ins), []>, Sched<[]>;
4790  def SEH_EpilogStart : Pseudo<(outs), (ins), []>, Sched<[]>;
4791  def SEH_EpilogEnd : Pseudo<(outs), (ins), []>, Sched<[]>;
4792  def SEH_PACSignLR : Pseudo<(outs), (ins), []>, Sched<[]>;
4793}
4794
4795// Pseudo instructions for Windows EH
4796//===----------------------------------------------------------------------===//
4797let isTerminator = 1, hasSideEffects = 1, isBarrier = 1, hasCtrlDep = 1,
4798    isCodeGenOnly = 1, isReturn = 1, isEHScopeReturn = 1, isPseudo = 1 in {
4799   def CLEANUPRET : Pseudo<(outs), (ins), [(cleanupret)]>, Sched<[]>;
4800   let usesCustomInserter = 1 in
4801     def CATCHRET : Pseudo<(outs), (ins am_brcond:$dst, am_brcond:$src), [(catchret bb:$dst, bb:$src)]>,
4802                    Sched<[]>;
4803}
4804
4805// Pseudo instructions for homogeneous prolog/epilog
4806let isPseudo = 1 in {
4807  // Save CSRs in order, {FPOffset}
4808  def HOM_Prolog : Pseudo<(outs), (ins variable_ops), []>, Sched<[]>;
4809  // Restore CSRs in order
4810  def HOM_Epilog : Pseudo<(outs), (ins variable_ops), []>, Sched<[]>;
4811}
4812
4813//===----------------------------------------------------------------------===//
4814// Floating point immediate move.
4815//===----------------------------------------------------------------------===//
4816
4817let isReMaterializable = 1, isAsCheapAsAMove = 1 in {
4818defm FMOV : FPMoveImmediate<"fmov">;
4819}
4820
4821let Predicates = [HasFullFP16] in {
4822  def : Pat<(bf16 fpimmbf16:$in),
4823            (FMOVHi (fpimm16XForm bf16:$in))>;
4824}
4825
4826//===----------------------------------------------------------------------===//
4827// Advanced SIMD two vector instructions.
4828//===----------------------------------------------------------------------===//
4829
4830defm UABDL   : SIMDLongThreeVectorBHSabdl<1, 0b0111, "uabdl",
4831                                          AArch64uabd>;
4832// Match UABDL in log2-shuffle patterns.
4833def : Pat<(abs (v8i16 (sub (zext (v8i8 V64:$opA)),
4834                           (zext (v8i8 V64:$opB))))),
4835          (UABDLv8i8_v8i16 V64:$opA, V64:$opB)>;
4836def : Pat<(xor (v8i16 (AArch64vashr v8i16:$src, (i32 15))),
4837               (v8i16 (add (sub (zext (v8i8 V64:$opA)),
4838                                (zext (v8i8 V64:$opB))),
4839                           (AArch64vashr v8i16:$src, (i32 15))))),
4840          (UABDLv8i8_v8i16 V64:$opA, V64:$opB)>;
4841def : Pat<(abs (v8i16 (sub (zext (extract_high_v16i8 (v16i8 V128:$opA))),
4842                           (zext (extract_high_v16i8 (v16i8 V128:$opB)))))),
4843          (UABDLv16i8_v8i16 V128:$opA, V128:$opB)>;
4844def : Pat<(xor (v8i16 (AArch64vashr v8i16:$src, (i32 15))),
4845               (v8i16 (add (sub (zext (extract_high_v16i8 (v16i8 V128:$opA))),
4846                                (zext (extract_high_v16i8 (v16i8 V128:$opB)))),
4847                           (AArch64vashr v8i16:$src, (i32 15))))),
4848          (UABDLv16i8_v8i16 V128:$opA, V128:$opB)>;
4849def : Pat<(abs (v4i32 (sub (zext (v4i16 V64:$opA)),
4850                           (zext (v4i16 V64:$opB))))),
4851          (UABDLv4i16_v4i32 V64:$opA, V64:$opB)>;
4852def : Pat<(abs (v4i32 (sub (zext (extract_high_v8i16 (v8i16 V128:$opA))),
4853                           (zext (extract_high_v8i16 (v8i16 V128:$opB)))))),
4854          (UABDLv8i16_v4i32 V128:$opA, V128:$opB)>;
4855def : Pat<(abs (v2i64 (sub (zext (v2i32 V64:$opA)),
4856                           (zext (v2i32 V64:$opB))))),
4857          (UABDLv2i32_v2i64 V64:$opA, V64:$opB)>;
4858def : Pat<(abs (v2i64 (sub (zext (extract_high_v4i32 (v4i32 V128:$opA))),
4859                           (zext (extract_high_v4i32 (v4i32 V128:$opB)))))),
4860          (UABDLv4i32_v2i64 V128:$opA, V128:$opB)>;
4861
4862defm ABS    : SIMDTwoVectorBHSD<0, 0b01011, "abs", abs>;
4863defm CLS    : SIMDTwoVectorBHS<0, 0b00100, "cls", int_aarch64_neon_cls>;
4864defm CLZ    : SIMDTwoVectorBHS<1, 0b00100, "clz", ctlz>;
4865defm CMEQ   : SIMDCmpTwoVector<0, 0b01001, "cmeq", AArch64cmeqz>;
4866defm CMGE   : SIMDCmpTwoVector<1, 0b01000, "cmge", AArch64cmgez>;
4867defm CMGT   : SIMDCmpTwoVector<0, 0b01000, "cmgt", AArch64cmgtz>;
4868defm CMLE   : SIMDCmpTwoVector<1, 0b01001, "cmle", AArch64cmlez>;
4869defm CMLT   : SIMDCmpTwoVector<0, 0b01010, "cmlt", AArch64cmltz>;
4870defm CNT    : SIMDTwoVectorB<0, 0b00, 0b00101, "cnt", ctpop>;
4871defm FABS   : SIMDTwoVectorFPNoException<0, 1, 0b01111, "fabs", fabs>;
4872
4873def : Pat<(v8i8 (AArch64vashr (v8i8 V64:$Rn), (i32 7))),
4874          (CMLTv8i8rz V64:$Rn)>;
4875def : Pat<(v4i16 (AArch64vashr (v4i16 V64:$Rn), (i32 15))),
4876          (CMLTv4i16rz V64:$Rn)>;
4877def : Pat<(v2i32 (AArch64vashr (v2i32 V64:$Rn), (i32 31))),
4878          (CMLTv2i32rz V64:$Rn)>;
4879def : Pat<(v16i8 (AArch64vashr (v16i8 V128:$Rn), (i32 7))),
4880          (CMLTv16i8rz V128:$Rn)>;
4881def : Pat<(v8i16 (AArch64vashr (v8i16 V128:$Rn), (i32 15))),
4882          (CMLTv8i16rz V128:$Rn)>;
4883def : Pat<(v4i32 (AArch64vashr (v4i32 V128:$Rn), (i32 31))),
4884          (CMLTv4i32rz V128:$Rn)>;
4885def : Pat<(v2i64 (AArch64vashr (v2i64 V128:$Rn), (i32 63))),
4886          (CMLTv2i64rz V128:$Rn)>;
4887
4888defm FCMEQ  : SIMDFPCmpTwoVector<0, 1, 0b01101, "fcmeq", AArch64fcmeqz>;
4889defm FCMGE  : SIMDFPCmpTwoVector<1, 1, 0b01100, "fcmge", AArch64fcmgez>;
4890defm FCMGT  : SIMDFPCmpTwoVector<0, 1, 0b01100, "fcmgt", AArch64fcmgtz>;
4891defm FCMLE  : SIMDFPCmpTwoVector<1, 1, 0b01101, "fcmle", AArch64fcmlez>;
4892defm FCMLT  : SIMDFPCmpTwoVector<0, 1, 0b01110, "fcmlt", AArch64fcmltz>;
4893defm FCVTAS : SIMDTwoVectorFPToInt<0,0,0b11100, "fcvtas",int_aarch64_neon_fcvtas>;
4894defm FCVTAU : SIMDTwoVectorFPToInt<1,0,0b11100, "fcvtau",int_aarch64_neon_fcvtau>;
4895defm FCVTL  : SIMDFPWidenTwoVector<0, 0, 0b10111, "fcvtl">;
4896def : Pat<(v4f32 (int_aarch64_neon_vcvthf2fp (v4i16 V64:$Rn))),
4897          (FCVTLv4i16 V64:$Rn)>;
4898def : Pat<(v4f32 (int_aarch64_neon_vcvthf2fp (extract_subvector (v8i16 V128:$Rn),
4899                                                                (i64 4)))),
4900          (FCVTLv8i16 V128:$Rn)>;
4901def : Pat<(v2f64 (any_fpextend (v2f32 V64:$Rn))),
4902          (FCVTLv2i32 V64:$Rn)>;
4903def : Pat<(v2f64 (any_fpextend (v2f32 (extract_high_v4f32 (v4f32 V128:$Rn))))),
4904          (FCVTLv4i32 V128:$Rn)>;
4905def : Pat<(v4f32 (any_fpextend (v4f16 V64:$Rn))),
4906          (FCVTLv4i16 V64:$Rn)>;
4907def : Pat<(v4f32 (any_fpextend (v4f16 (extract_high_v8f16 (v8f16 V128:$Rn))))),
4908          (FCVTLv8i16 V128:$Rn)>;
4909
4910defm FCVTMS : SIMDTwoVectorFPToInt<0,0,0b11011, "fcvtms",int_aarch64_neon_fcvtms>;
4911defm FCVTMU : SIMDTwoVectorFPToInt<1,0,0b11011, "fcvtmu",int_aarch64_neon_fcvtmu>;
4912defm FCVTNS : SIMDTwoVectorFPToInt<0,0,0b11010, "fcvtns",int_aarch64_neon_fcvtns>;
4913defm FCVTNU : SIMDTwoVectorFPToInt<1,0,0b11010, "fcvtnu",int_aarch64_neon_fcvtnu>;
4914defm FCVTN  : SIMDFPNarrowTwoVector<0, 0, 0b10110, "fcvtn">;
4915def : Pat<(v4i16 (int_aarch64_neon_vcvtfp2hf (v4f32 V128:$Rn))),
4916          (FCVTNv4i16 V128:$Rn)>;
4917def : Pat<(concat_vectors V64:$Rd,
4918                          (v4i16 (int_aarch64_neon_vcvtfp2hf (v4f32 V128:$Rn)))),
4919          (FCVTNv8i16 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
4920def : Pat<(v2f32 (any_fpround (v2f64 V128:$Rn))),
4921          (FCVTNv2i32 V128:$Rn)>;
4922def : Pat<(v4f16 (any_fpround (v4f32 V128:$Rn))),
4923          (FCVTNv4i16 V128:$Rn)>;
4924def : Pat<(concat_vectors V64:$Rd, (v2f32 (any_fpround (v2f64 V128:$Rn)))),
4925          (FCVTNv4i32 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
4926def : Pat<(concat_vectors V64:$Rd, (v4f16 (any_fpround (v4f32 V128:$Rn)))),
4927          (FCVTNv8i16 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
4928defm FCVTPS : SIMDTwoVectorFPToInt<0,1,0b11010, "fcvtps",int_aarch64_neon_fcvtps>;
4929defm FCVTPU : SIMDTwoVectorFPToInt<1,1,0b11010, "fcvtpu",int_aarch64_neon_fcvtpu>;
4930defm FCVTXN : SIMDFPInexactCvtTwoVector<1, 0, 0b10110, "fcvtxn",
4931                                        int_aarch64_neon_fcvtxn>;
4932defm FCVTZS : SIMDTwoVectorFPToInt<0, 1, 0b11011, "fcvtzs", any_fp_to_sint>;
4933defm FCVTZU : SIMDTwoVectorFPToInt<1, 1, 0b11011, "fcvtzu", any_fp_to_uint>;
4934
4935// AArch64's FCVT instructions saturate when out of range.
4936multiclass SIMDTwoVectorFPToIntSatPats<SDNode to_int_sat, string INST> {
4937  let Predicates = [HasFullFP16] in {
4938  def : Pat<(v4i16 (to_int_sat v4f16:$Rn, i16)),
4939            (!cast<Instruction>(INST # v4f16) v4f16:$Rn)>;
4940  def : Pat<(v8i16 (to_int_sat v8f16:$Rn, i16)),
4941            (!cast<Instruction>(INST # v8f16) v8f16:$Rn)>;
4942  }
4943  def : Pat<(v2i32 (to_int_sat v2f32:$Rn, i32)),
4944            (!cast<Instruction>(INST # v2f32) v2f32:$Rn)>;
4945  def : Pat<(v4i32 (to_int_sat v4f32:$Rn, i32)),
4946            (!cast<Instruction>(INST # v4f32) v4f32:$Rn)>;
4947  def : Pat<(v2i64 (to_int_sat v2f64:$Rn, i64)),
4948            (!cast<Instruction>(INST # v2f64) v2f64:$Rn)>;
4949}
4950defm : SIMDTwoVectorFPToIntSatPats<fp_to_sint_sat, "FCVTZS">;
4951defm : SIMDTwoVectorFPToIntSatPats<fp_to_uint_sat, "FCVTZU">;
4952
4953def : Pat<(v4i16 (int_aarch64_neon_fcvtzs v4f16:$Rn)), (FCVTZSv4f16 $Rn)>;
4954def : Pat<(v8i16 (int_aarch64_neon_fcvtzs v8f16:$Rn)), (FCVTZSv8f16 $Rn)>;
4955def : Pat<(v2i32 (int_aarch64_neon_fcvtzs v2f32:$Rn)), (FCVTZSv2f32 $Rn)>;
4956def : Pat<(v4i32 (int_aarch64_neon_fcvtzs v4f32:$Rn)), (FCVTZSv4f32 $Rn)>;
4957def : Pat<(v2i64 (int_aarch64_neon_fcvtzs v2f64:$Rn)), (FCVTZSv2f64 $Rn)>;
4958
4959def : Pat<(v4i16 (int_aarch64_neon_fcvtzu v4f16:$Rn)), (FCVTZUv4f16 $Rn)>;
4960def : Pat<(v8i16 (int_aarch64_neon_fcvtzu v8f16:$Rn)), (FCVTZUv8f16 $Rn)>;
4961def : Pat<(v2i32 (int_aarch64_neon_fcvtzu v2f32:$Rn)), (FCVTZUv2f32 $Rn)>;
4962def : Pat<(v4i32 (int_aarch64_neon_fcvtzu v4f32:$Rn)), (FCVTZUv4f32 $Rn)>;
4963def : Pat<(v2i64 (int_aarch64_neon_fcvtzu v2f64:$Rn)), (FCVTZUv2f64 $Rn)>;
4964
4965defm FNEG   : SIMDTwoVectorFPNoException<1, 1, 0b01111, "fneg", fneg>;
4966defm FRECPE : SIMDTwoVectorFP<0, 1, 0b11101, "frecpe", int_aarch64_neon_frecpe>;
4967defm FRINTA : SIMDTwoVectorFP<1, 0, 0b11000, "frinta", any_fround>;
4968defm FRINTI : SIMDTwoVectorFP<1, 1, 0b11001, "frinti", any_fnearbyint>;
4969defm FRINTM : SIMDTwoVectorFP<0, 0, 0b11001, "frintm", any_ffloor>;
4970defm FRINTN : SIMDTwoVectorFP<0, 0, 0b11000, "frintn", any_froundeven>;
4971defm FRINTP : SIMDTwoVectorFP<0, 1, 0b11000, "frintp", any_fceil>;
4972defm FRINTX : SIMDTwoVectorFP<1, 0, 0b11001, "frintx", any_frint>;
4973defm FRINTZ : SIMDTwoVectorFP<0, 1, 0b11001, "frintz", any_ftrunc>;
4974
4975let Predicates = [HasFRInt3264] in {
4976  defm FRINT32Z : FRIntNNTVector<0, 0, "frint32z", int_aarch64_neon_frint32z>;
4977  defm FRINT64Z : FRIntNNTVector<0, 1, "frint64z", int_aarch64_neon_frint64z>;
4978  defm FRINT32X : FRIntNNTVector<1, 0, "frint32x", int_aarch64_neon_frint32x>;
4979  defm FRINT64X : FRIntNNTVector<1, 1, "frint64x", int_aarch64_neon_frint64x>;
4980} // HasFRInt3264
4981
4982defm FRSQRTE: SIMDTwoVectorFP<1, 1, 0b11101, "frsqrte", int_aarch64_neon_frsqrte>;
4983defm FSQRT  : SIMDTwoVectorFP<1, 1, 0b11111, "fsqrt", any_fsqrt>;
4984defm NEG    : SIMDTwoVectorBHSD<1, 0b01011, "neg",
4985                               UnOpFrag<(sub immAllZerosV, node:$LHS)> >;
4986defm NOT    : SIMDTwoVectorB<1, 0b00, 0b00101, "not", vnot>;
4987// Aliases for MVN -> NOT.
4988def : InstAlias<"mvn{ $Vd.8b, $Vn.8b|.8b $Vd, $Vn}",
4989                (NOTv8i8 V64:$Vd, V64:$Vn)>;
4990def : InstAlias<"mvn{ $Vd.16b, $Vn.16b|.16b $Vd, $Vn}",
4991                (NOTv16i8 V128:$Vd, V128:$Vn)>;
4992
4993def : Pat<(vnot (v4i16 V64:$Rn)),  (NOTv8i8  V64:$Rn)>;
4994def : Pat<(vnot (v8i16 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
4995def : Pat<(vnot (v2i32 V64:$Rn)),  (NOTv8i8  V64:$Rn)>;
4996def : Pat<(vnot (v4i32 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
4997def : Pat<(vnot (v1i64 V64:$Rn)),  (NOTv8i8  V64:$Rn)>;
4998def : Pat<(vnot (v2i64 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
4999
5000defm RBIT   : SIMDTwoVectorB<1, 0b01, 0b00101, "rbit", bitreverse>;
5001defm REV16  : SIMDTwoVectorB<0, 0b00, 0b00001, "rev16", AArch64rev16>;
5002defm REV32  : SIMDTwoVectorBH<1, 0b00000, "rev32", AArch64rev32>;
5003defm REV64  : SIMDTwoVectorBHS<0, 0b00000, "rev64", AArch64rev64>;
5004defm SADALP : SIMDLongTwoVectorTied<0, 0b00110, "sadalp",
5005       BinOpFrag<(add node:$LHS, (AArch64saddlp node:$RHS))> >;
5006defm SADDLP : SIMDLongTwoVector<0, 0b00010, "saddlp", AArch64saddlp>;
5007defm SCVTF  : SIMDTwoVectorIntToFP<0, 0, 0b11101, "scvtf", any_sint_to_fp>;
5008defm SHLL   : SIMDVectorLShiftLongBySizeBHS;
5009defm SQABS  : SIMDTwoVectorBHSD<0, 0b00111, "sqabs", int_aarch64_neon_sqabs>;
5010defm SQNEG  : SIMDTwoVectorBHSD<1, 0b00111, "sqneg", int_aarch64_neon_sqneg>;
5011defm SQXTN  : SIMDMixedTwoVector<0, 0b10100, "sqxtn", int_aarch64_neon_sqxtn>;
5012defm SQXTUN : SIMDMixedTwoVector<1, 0b10010, "sqxtun", int_aarch64_neon_sqxtun>;
5013defm SUQADD : SIMDTwoVectorBHSDTied<0, 0b00011, "suqadd",int_aarch64_neon_suqadd>;
5014defm UADALP : SIMDLongTwoVectorTied<1, 0b00110, "uadalp",
5015       BinOpFrag<(add node:$LHS, (AArch64uaddlp node:$RHS))> >;
5016defm UADDLP : SIMDLongTwoVector<1, 0b00010, "uaddlp", AArch64uaddlp>;
5017defm UCVTF  : SIMDTwoVectorIntToFP<1, 0, 0b11101, "ucvtf", any_uint_to_fp>;
5018defm UQXTN  : SIMDMixedTwoVector<1, 0b10100, "uqxtn", int_aarch64_neon_uqxtn>;
5019defm URECPE : SIMDTwoVectorS<0, 1, 0b11100, "urecpe", int_aarch64_neon_urecpe>;
5020defm URSQRTE: SIMDTwoVectorS<1, 1, 0b11100, "ursqrte", int_aarch64_neon_ursqrte>;
5021defm USQADD : SIMDTwoVectorBHSDTied<1, 0b00011, "usqadd",int_aarch64_neon_usqadd>;
5022defm XTN    : SIMDMixedTwoVector<0, 0b10010, "xtn", trunc>;
5023
5024def : Pat<(v4f16  (AArch64rev32 V64:$Rn)),  (REV32v4i16 V64:$Rn)>;
5025def : Pat<(v4f16  (AArch64rev64 V64:$Rn)),  (REV64v4i16 V64:$Rn)>;
5026def : Pat<(v4bf16 (AArch64rev32 V64:$Rn)),  (REV32v4i16 V64:$Rn)>;
5027def : Pat<(v4bf16 (AArch64rev64 V64:$Rn)),  (REV64v4i16 V64:$Rn)>;
5028def : Pat<(v8f16  (AArch64rev32 V128:$Rn)), (REV32v8i16 V128:$Rn)>;
5029def : Pat<(v8f16  (AArch64rev64 V128:$Rn)), (REV64v8i16 V128:$Rn)>;
5030def : Pat<(v8bf16 (AArch64rev32 V128:$Rn)), (REV32v8i16 V128:$Rn)>;
5031def : Pat<(v8bf16 (AArch64rev64 V128:$Rn)), (REV64v8i16 V128:$Rn)>;
5032def : Pat<(v2f32  (AArch64rev64 V64:$Rn)),  (REV64v2i32 V64:$Rn)>;
5033def : Pat<(v4f32  (AArch64rev64 V128:$Rn)), (REV64v4i32 V128:$Rn)>;
5034
5035// Patterns for vector long shift (by element width). These need to match all
5036// three of zext, sext and anyext so it's easier to pull the patterns out of the
5037// definition.
5038multiclass SIMDVectorLShiftLongBySizeBHSPats<SDPatternOperator ext> {
5039  def : Pat<(AArch64vshl (v8i16 (ext (v8i8 V64:$Rn))), (i32 8)),
5040            (SHLLv8i8 V64:$Rn)>;
5041  def : Pat<(AArch64vshl (v8i16 (ext (extract_high_v16i8 (v16i8 V128:$Rn)))), (i32 8)),
5042            (SHLLv16i8 V128:$Rn)>;
5043  def : Pat<(AArch64vshl (v4i32 (ext (v4i16 V64:$Rn))), (i32 16)),
5044            (SHLLv4i16 V64:$Rn)>;
5045  def : Pat<(AArch64vshl (v4i32 (ext (extract_high_v8i16 (v8i16 V128:$Rn)))), (i32 16)),
5046            (SHLLv8i16 V128:$Rn)>;
5047  def : Pat<(AArch64vshl (v2i64 (ext (v2i32 V64:$Rn))), (i32 32)),
5048            (SHLLv2i32 V64:$Rn)>;
5049  def : Pat<(AArch64vshl (v2i64 (ext (extract_high_v4i32 (v4i32 V128:$Rn)))), (i32 32)),
5050            (SHLLv4i32 V128:$Rn)>;
5051}
5052
5053defm : SIMDVectorLShiftLongBySizeBHSPats<anyext>;
5054defm : SIMDVectorLShiftLongBySizeBHSPats<zext>;
5055defm : SIMDVectorLShiftLongBySizeBHSPats<sext>;
5056
5057// Constant vector values, used in the S/UQXTN patterns below.
5058def VImmFF:   PatLeaf<(AArch64NvCast (v2i64 (AArch64movi_edit (i32 85))))>;
5059def VImmFFFF: PatLeaf<(AArch64NvCast (v2i64 (AArch64movi_edit (i32 51))))>;
5060def VImm7F:   PatLeaf<(AArch64movi_shift (i32 127), (i32 0))>;
5061def VImm80:   PatLeaf<(AArch64mvni_shift (i32 127), (i32 0))>;
5062def VImm7FFF: PatLeaf<(AArch64movi_msl (i32 127), (i32 264))>;
5063def VImm8000: PatLeaf<(AArch64mvni_msl (i32 127), (i32 264))>;
5064
5065// trunc(umin(X, 255)) -> UQXTRN v8i8
5066def : Pat<(v8i8 (trunc (umin (v8i16 V128:$Vn), (v8i16 VImmFF)))),
5067          (UQXTNv8i8 V128:$Vn)>;
5068// trunc(umin(X, 65535)) -> UQXTRN v4i16
5069def : Pat<(v4i16 (trunc (umin (v4i32 V128:$Vn), (v4i32 VImmFFFF)))),
5070          (UQXTNv4i16 V128:$Vn)>;
5071// trunc(smin(smax(X, -128), 128)) -> SQXTRN
5072//  with reversed min/max
5073def : Pat<(v8i8 (trunc (smin (smax (v8i16 V128:$Vn), (v8i16 VImm80)),
5074                             (v8i16 VImm7F)))),
5075          (SQXTNv8i8 V128:$Vn)>;
5076def : Pat<(v8i8 (trunc (smax (smin (v8i16 V128:$Vn), (v8i16 VImm7F)),
5077                             (v8i16 VImm80)))),
5078          (SQXTNv8i8 V128:$Vn)>;
5079// trunc(smin(smax(X, -32768), 32767)) -> SQXTRN
5080//  with reversed min/max
5081def : Pat<(v4i16 (trunc (smin (smax (v4i32 V128:$Vn), (v4i32 VImm8000)),
5082                              (v4i32 VImm7FFF)))),
5083          (SQXTNv4i16 V128:$Vn)>;
5084def : Pat<(v4i16 (trunc (smax (smin (v4i32 V128:$Vn), (v4i32 VImm7FFF)),
5085                              (v4i32 VImm8000)))),
5086          (SQXTNv4i16 V128:$Vn)>;
5087
5088// concat_vectors(Vd, trunc(smin(smax Vm, -128), 127) ~> SQXTN2(Vd, Vn)
5089// with reversed min/max
5090def : Pat<(v16i8 (concat_vectors
5091                 (v8i8 V64:$Vd),
5092                 (v8i8 (trunc (smin (smax (v8i16 V128:$Vn), (v8i16 VImm80)),
5093                                          (v8i16 VImm7F)))))),
5094          (SQXTNv16i8 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Vd, dsub), V128:$Vn)>;
5095def : Pat<(v16i8 (concat_vectors
5096                 (v8i8 V64:$Vd),
5097                 (v8i8 (trunc (smax (smin (v8i16 V128:$Vn), (v8i16 VImm7F)),
5098                                          (v8i16 VImm80)))))),
5099          (SQXTNv16i8 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Vd, dsub), V128:$Vn)>;
5100
5101// concat_vectors(Vd, trunc(smin(smax Vm, -32768), 32767) ~> SQXTN2(Vd, Vn)
5102// with reversed min/max
5103def : Pat<(v8i16 (concat_vectors
5104                 (v4i16 V64:$Vd),
5105                 (v4i16 (trunc (smin (smax (v4i32 V128:$Vn), (v4i32 VImm8000)),
5106                                           (v4i32 VImm7FFF)))))),
5107          (SQXTNv8i16 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Vd, dsub), V128:$Vn)>;
5108def : Pat<(v8i16 (concat_vectors
5109                 (v4i16 V64:$Vd),
5110                 (v4i16 (trunc (smax (smin (v4i32 V128:$Vn), (v4i32 VImm7FFF)),
5111                                           (v4i32 VImm8000)))))),
5112          (SQXTNv8i16 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Vd, dsub), V128:$Vn)>;
5113
5114//===----------------------------------------------------------------------===//
5115// Advanced SIMD three vector instructions.
5116//===----------------------------------------------------------------------===//
5117
5118defm ADD     : SIMDThreeSameVector<0, 0b10000, "add", add>;
5119defm ADDP    : SIMDThreeSameVector<0, 0b10111, "addp", AArch64addp>;
5120defm CMEQ    : SIMDThreeSameVector<1, 0b10001, "cmeq", AArch64cmeq>;
5121defm CMGE    : SIMDThreeSameVector<0, 0b00111, "cmge", AArch64cmge>;
5122defm CMGT    : SIMDThreeSameVector<0, 0b00110, "cmgt", AArch64cmgt>;
5123defm CMHI    : SIMDThreeSameVector<1, 0b00110, "cmhi", AArch64cmhi>;
5124defm CMHS    : SIMDThreeSameVector<1, 0b00111, "cmhs", AArch64cmhs>;
5125defm CMTST   : SIMDThreeSameVector<0, 0b10001, "cmtst", AArch64cmtst>;
5126foreach VT = [ v8i8, v16i8, v4i16, v8i16, v2i32, v4i32, v2i64 ] in {
5127def : Pat<(vnot (AArch64cmeqz VT:$Rn)), (!cast<Instruction>("CMTST"#VT) VT:$Rn, VT:$Rn)>;
5128}
5129defm FABD    : SIMDThreeSameVectorFP<1,1,0b010,"fabd", int_aarch64_neon_fabd>;
5130let Predicates = [HasNEON] in {
5131foreach VT = [ v2f32, v4f32, v2f64 ] in
5132def : Pat<(fabs (fsub VT:$Rn, VT:$Rm)), (!cast<Instruction>("FABD"#VT) VT:$Rn, VT:$Rm)>;
5133}
5134let Predicates = [HasNEON, HasFullFP16] in {
5135foreach VT = [ v4f16, v8f16 ] in
5136def : Pat<(fabs (fsub VT:$Rn, VT:$Rm)), (!cast<Instruction>("FABD"#VT) VT:$Rn, VT:$Rm)>;
5137}
5138defm FACGE   : SIMDThreeSameVectorFPCmp<1,0,0b101,"facge",AArch64facge>;
5139defm FACGT   : SIMDThreeSameVectorFPCmp<1,1,0b101,"facgt",AArch64facgt>;
5140defm FADDP   : SIMDThreeSameVectorFP<1,0,0b010,"faddp", AArch64faddp>;
5141defm FADD    : SIMDThreeSameVectorFP<0,0,0b010,"fadd", any_fadd>;
5142defm FCMEQ   : SIMDThreeSameVectorFPCmp<0, 0, 0b100, "fcmeq", AArch64fcmeq>;
5143defm FCMGE   : SIMDThreeSameVectorFPCmp<1, 0, 0b100, "fcmge", AArch64fcmge>;
5144defm FCMGT   : SIMDThreeSameVectorFPCmp<1, 1, 0b100, "fcmgt", AArch64fcmgt>;
5145defm FDIV    : SIMDThreeSameVectorFP<1,0,0b111,"fdiv", any_fdiv>;
5146defm FMAXNMP : SIMDThreeSameVectorFP<1,0,0b000,"fmaxnmp", int_aarch64_neon_fmaxnmp>;
5147defm FMAXNM  : SIMDThreeSameVectorFP<0,0,0b000,"fmaxnm", any_fmaxnum>;
5148defm FMAXP   : SIMDThreeSameVectorFP<1,0,0b110,"fmaxp", int_aarch64_neon_fmaxp>;
5149defm FMAX    : SIMDThreeSameVectorFP<0,0,0b110,"fmax", any_fmaximum>;
5150defm FMINNMP : SIMDThreeSameVectorFP<1,1,0b000,"fminnmp", int_aarch64_neon_fminnmp>;
5151defm FMINNM  : SIMDThreeSameVectorFP<0,1,0b000,"fminnm", any_fminnum>;
5152defm FMINP   : SIMDThreeSameVectorFP<1,1,0b110,"fminp", int_aarch64_neon_fminp>;
5153defm FMIN    : SIMDThreeSameVectorFP<0,1,0b110,"fmin", any_fminimum>;
5154
5155// NOTE: The operands of the PatFrag are reordered on FMLA/FMLS because the
5156// instruction expects the addend first, while the fma intrinsic puts it last.
5157defm FMLA     : SIMDThreeSameVectorFPTied<0, 0, 0b001, "fmla",
5158            TriOpFrag<(any_fma node:$RHS, node:$MHS, node:$LHS)> >;
5159defm FMLS     : SIMDThreeSameVectorFPTied<0, 1, 0b001, "fmls",
5160            TriOpFrag<(any_fma node:$MHS, (fneg node:$RHS), node:$LHS)> >;
5161
5162defm FMULX    : SIMDThreeSameVectorFP<0,0,0b011,"fmulx", int_aarch64_neon_fmulx>;
5163defm FMUL     : SIMDThreeSameVectorFP<1,0,0b011,"fmul", any_fmul>;
5164defm FRECPS   : SIMDThreeSameVectorFP<0,0,0b111,"frecps", int_aarch64_neon_frecps>;
5165defm FRSQRTS  : SIMDThreeSameVectorFP<0,1,0b111,"frsqrts", int_aarch64_neon_frsqrts>;
5166defm FSUB     : SIMDThreeSameVectorFP<0,1,0b010,"fsub", any_fsub>;
5167
5168// MLA and MLS are generated in MachineCombine
5169defm MLA      : SIMDThreeSameVectorBHSTied<0, 0b10010, "mla", null_frag>;
5170defm MLS      : SIMDThreeSameVectorBHSTied<1, 0b10010, "mls", null_frag>;
5171
5172defm MUL      : SIMDThreeSameVectorBHS<0, 0b10011, "mul", mul>;
5173defm PMUL     : SIMDThreeSameVectorB<1, 0b10011, "pmul", int_aarch64_neon_pmul>;
5174defm SABA     : SIMDThreeSameVectorBHSTied<0, 0b01111, "saba",
5175      TriOpFrag<(add node:$LHS, (AArch64sabd node:$MHS, node:$RHS))> >;
5176defm SABD     : SIMDThreeSameVectorBHS<0,0b01110,"sabd", AArch64sabd>;
5177defm SHADD    : SIMDThreeSameVectorBHS<0,0b00000,"shadd", avgfloors>;
5178defm SHSUB    : SIMDThreeSameVectorBHS<0,0b00100,"shsub", int_aarch64_neon_shsub>;
5179defm SMAXP    : SIMDThreeSameVectorBHS<0,0b10100,"smaxp", int_aarch64_neon_smaxp>;
5180defm SMAX     : SIMDThreeSameVectorBHS<0,0b01100,"smax", smax>;
5181defm SMINP    : SIMDThreeSameVectorBHS<0,0b10101,"sminp", int_aarch64_neon_sminp>;
5182defm SMIN     : SIMDThreeSameVectorBHS<0,0b01101,"smin", smin>;
5183defm SQADD    : SIMDThreeSameVector<0,0b00001,"sqadd", int_aarch64_neon_sqadd>;
5184defm SQDMULH  : SIMDThreeSameVectorHS<0,0b10110,"sqdmulh",int_aarch64_neon_sqdmulh>;
5185defm SQRDMULH : SIMDThreeSameVectorHS<1,0b10110,"sqrdmulh",int_aarch64_neon_sqrdmulh>;
5186defm SQRSHL   : SIMDThreeSameVector<0,0b01011,"sqrshl", int_aarch64_neon_sqrshl>;
5187defm SQSHL    : SIMDThreeSameVector<0,0b01001,"sqshl", int_aarch64_neon_sqshl>;
5188defm SQSUB    : SIMDThreeSameVector<0,0b00101,"sqsub", int_aarch64_neon_sqsub>;
5189defm SRHADD   : SIMDThreeSameVectorBHS<0,0b00010,"srhadd", avgceils>;
5190defm SRSHL    : SIMDThreeSameVector<0,0b01010,"srshl", int_aarch64_neon_srshl>;
5191defm SSHL     : SIMDThreeSameVector<0,0b01000,"sshl", int_aarch64_neon_sshl>;
5192defm SUB      : SIMDThreeSameVector<1,0b10000,"sub", sub>;
5193defm UABA     : SIMDThreeSameVectorBHSTied<1, 0b01111, "uaba",
5194      TriOpFrag<(add node:$LHS, (AArch64uabd node:$MHS, node:$RHS))> >;
5195defm UABD     : SIMDThreeSameVectorBHS<1,0b01110,"uabd", AArch64uabd>;
5196defm UHADD    : SIMDThreeSameVectorBHS<1,0b00000,"uhadd", avgflooru>;
5197defm UHSUB    : SIMDThreeSameVectorBHS<1,0b00100,"uhsub", int_aarch64_neon_uhsub>;
5198defm UMAXP    : SIMDThreeSameVectorBHS<1,0b10100,"umaxp", int_aarch64_neon_umaxp>;
5199defm UMAX     : SIMDThreeSameVectorBHS<1,0b01100,"umax", umax>;
5200defm UMINP    : SIMDThreeSameVectorBHS<1,0b10101,"uminp", int_aarch64_neon_uminp>;
5201defm UMIN     : SIMDThreeSameVectorBHS<1,0b01101,"umin", umin>;
5202defm UQADD    : SIMDThreeSameVector<1,0b00001,"uqadd", int_aarch64_neon_uqadd>;
5203defm UQRSHL   : SIMDThreeSameVector<1,0b01011,"uqrshl", int_aarch64_neon_uqrshl>;
5204defm UQSHL    : SIMDThreeSameVector<1,0b01001,"uqshl", int_aarch64_neon_uqshl>;
5205defm UQSUB    : SIMDThreeSameVector<1,0b00101,"uqsub", int_aarch64_neon_uqsub>;
5206defm URHADD   : SIMDThreeSameVectorBHS<1,0b00010,"urhadd", avgceilu>;
5207defm URSHL    : SIMDThreeSameVector<1,0b01010,"urshl", int_aarch64_neon_urshl>;
5208defm USHL     : SIMDThreeSameVector<1,0b01000,"ushl", int_aarch64_neon_ushl>;
5209defm SQRDMLAH : SIMDThreeSameVectorSQRDMLxHTiedHS<1,0b10000,"sqrdmlah",
5210                                                  int_aarch64_neon_sqrdmlah>;
5211defm SQRDMLSH : SIMDThreeSameVectorSQRDMLxHTiedHS<1,0b10001,"sqrdmlsh",
5212                                                    int_aarch64_neon_sqrdmlsh>;
5213
5214// Extra saturate patterns, other than the intrinsics matches above
5215defm : SIMDThreeSameVectorExtraPatterns<"SQADD", saddsat>;
5216defm : SIMDThreeSameVectorExtraPatterns<"UQADD", uaddsat>;
5217defm : SIMDThreeSameVectorExtraPatterns<"SQSUB", ssubsat>;
5218defm : SIMDThreeSameVectorExtraPatterns<"UQSUB", usubsat>;
5219
5220defm AND : SIMDLogicalThreeVector<0, 0b00, "and", and>;
5221defm BIC : SIMDLogicalThreeVector<0, 0b01, "bic",
5222                                  BinOpFrag<(and node:$LHS, (vnot node:$RHS))> >;
5223defm EOR : SIMDLogicalThreeVector<1, 0b00, "eor", xor>;
5224defm ORN : SIMDLogicalThreeVector<0, 0b11, "orn",
5225                                  BinOpFrag<(or node:$LHS, (vnot node:$RHS))> >;
5226defm ORR : SIMDLogicalThreeVector<0, 0b10, "orr", or>;
5227
5228// Pseudo bitwise select pattern BSP.
5229// It is expanded into BSL/BIT/BIF after register allocation.
5230defm BSP : SIMDLogicalThreeVectorPseudo<TriOpFrag<(or (and node:$LHS, node:$MHS),
5231                                                      (and (vnot node:$LHS), node:$RHS))>>;
5232defm BSL : SIMDLogicalThreeVectorTied<1, 0b01, "bsl">;
5233defm BIT : SIMDLogicalThreeVectorTied<1, 0b10, "bit", AArch64bit>;
5234defm BIF : SIMDLogicalThreeVectorTied<1, 0b11, "bif">;
5235
5236def : Pat<(AArch64bsp (v8i8 V64:$Rd), V64:$Rn, V64:$Rm),
5237          (BSPv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
5238def : Pat<(AArch64bsp (v4i16 V64:$Rd), V64:$Rn, V64:$Rm),
5239          (BSPv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
5240def : Pat<(AArch64bsp (v2i32 V64:$Rd), V64:$Rn, V64:$Rm),
5241          (BSPv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
5242def : Pat<(AArch64bsp (v1i64 V64:$Rd), V64:$Rn, V64:$Rm),
5243          (BSPv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
5244
5245def : Pat<(AArch64bsp (v16i8 V128:$Rd), V128:$Rn, V128:$Rm),
5246          (BSPv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
5247def : Pat<(AArch64bsp (v8i16 V128:$Rd), V128:$Rn, V128:$Rm),
5248          (BSPv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
5249def : Pat<(AArch64bsp (v4i32 V128:$Rd), V128:$Rn, V128:$Rm),
5250          (BSPv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
5251def : Pat<(AArch64bsp (v2i64 V128:$Rd), V128:$Rn, V128:$Rm),
5252          (BSPv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
5253
5254def : InstAlias<"mov{\t$dst.16b, $src.16b|.16b\t$dst, $src}",
5255                (ORRv16i8 V128:$dst, V128:$src, V128:$src), 1>;
5256def : InstAlias<"mov{\t$dst.8h, $src.8h|.8h\t$dst, $src}",
5257                (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
5258def : InstAlias<"mov{\t$dst.4s, $src.4s|.4s\t$dst, $src}",
5259                (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
5260def : InstAlias<"mov{\t$dst.2d, $src.2d|.2d\t$dst, $src}",
5261                (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
5262
5263def : InstAlias<"mov{\t$dst.8b, $src.8b|.8b\t$dst, $src}",
5264                (ORRv8i8 V64:$dst, V64:$src, V64:$src), 1>;
5265def : InstAlias<"mov{\t$dst.4h, $src.4h|.4h\t$dst, $src}",
5266                (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
5267def : InstAlias<"mov{\t$dst.2s, $src.2s|.2s\t$dst, $src}",
5268                (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
5269def : InstAlias<"mov{\t$dst.1d, $src.1d|.1d\t$dst, $src}",
5270                (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
5271
5272def : InstAlias<"{cmls\t$dst.8b, $src1.8b, $src2.8b" #
5273                "|cmls.8b\t$dst, $src1, $src2}",
5274                (CMHSv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
5275def : InstAlias<"{cmls\t$dst.16b, $src1.16b, $src2.16b" #
5276                "|cmls.16b\t$dst, $src1, $src2}",
5277                (CMHSv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
5278def : InstAlias<"{cmls\t$dst.4h, $src1.4h, $src2.4h" #
5279                "|cmls.4h\t$dst, $src1, $src2}",
5280                (CMHSv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
5281def : InstAlias<"{cmls\t$dst.8h, $src1.8h, $src2.8h" #
5282                "|cmls.8h\t$dst, $src1, $src2}",
5283                (CMHSv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
5284def : InstAlias<"{cmls\t$dst.2s, $src1.2s, $src2.2s" #
5285                "|cmls.2s\t$dst, $src1, $src2}",
5286                (CMHSv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
5287def : InstAlias<"{cmls\t$dst.4s, $src1.4s, $src2.4s" #
5288                "|cmls.4s\t$dst, $src1, $src2}",
5289                (CMHSv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
5290def : InstAlias<"{cmls\t$dst.2d, $src1.2d, $src2.2d" #
5291                "|cmls.2d\t$dst, $src1, $src2}",
5292                (CMHSv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
5293
5294def : InstAlias<"{cmlo\t$dst.8b, $src1.8b, $src2.8b" #
5295                "|cmlo.8b\t$dst, $src1, $src2}",
5296                (CMHIv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
5297def : InstAlias<"{cmlo\t$dst.16b, $src1.16b, $src2.16b" #
5298                "|cmlo.16b\t$dst, $src1, $src2}",
5299                (CMHIv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
5300def : InstAlias<"{cmlo\t$dst.4h, $src1.4h, $src2.4h" #
5301                "|cmlo.4h\t$dst, $src1, $src2}",
5302                (CMHIv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
5303def : InstAlias<"{cmlo\t$dst.8h, $src1.8h, $src2.8h" #
5304                "|cmlo.8h\t$dst, $src1, $src2}",
5305                (CMHIv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
5306def : InstAlias<"{cmlo\t$dst.2s, $src1.2s, $src2.2s" #
5307                "|cmlo.2s\t$dst, $src1, $src2}",
5308                (CMHIv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
5309def : InstAlias<"{cmlo\t$dst.4s, $src1.4s, $src2.4s" #
5310                "|cmlo.4s\t$dst, $src1, $src2}",
5311                (CMHIv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
5312def : InstAlias<"{cmlo\t$dst.2d, $src1.2d, $src2.2d" #
5313                "|cmlo.2d\t$dst, $src1, $src2}",
5314                (CMHIv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
5315
5316def : InstAlias<"{cmle\t$dst.8b, $src1.8b, $src2.8b" #
5317                "|cmle.8b\t$dst, $src1, $src2}",
5318                (CMGEv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
5319def : InstAlias<"{cmle\t$dst.16b, $src1.16b, $src2.16b" #
5320                "|cmle.16b\t$dst, $src1, $src2}",
5321                (CMGEv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
5322def : InstAlias<"{cmle\t$dst.4h, $src1.4h, $src2.4h" #
5323                "|cmle.4h\t$dst, $src1, $src2}",
5324                (CMGEv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
5325def : InstAlias<"{cmle\t$dst.8h, $src1.8h, $src2.8h" #
5326                "|cmle.8h\t$dst, $src1, $src2}",
5327                (CMGEv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
5328def : InstAlias<"{cmle\t$dst.2s, $src1.2s, $src2.2s" #
5329                "|cmle.2s\t$dst, $src1, $src2}",
5330                (CMGEv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
5331def : InstAlias<"{cmle\t$dst.4s, $src1.4s, $src2.4s" #
5332                "|cmle.4s\t$dst, $src1, $src2}",
5333                (CMGEv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
5334def : InstAlias<"{cmle\t$dst.2d, $src1.2d, $src2.2d" #
5335                "|cmle.2d\t$dst, $src1, $src2}",
5336                (CMGEv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
5337
5338def : InstAlias<"{cmlt\t$dst.8b, $src1.8b, $src2.8b" #
5339                "|cmlt.8b\t$dst, $src1, $src2}",
5340                (CMGTv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
5341def : InstAlias<"{cmlt\t$dst.16b, $src1.16b, $src2.16b" #
5342                "|cmlt.16b\t$dst, $src1, $src2}",
5343                (CMGTv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
5344def : InstAlias<"{cmlt\t$dst.4h, $src1.4h, $src2.4h" #
5345                "|cmlt.4h\t$dst, $src1, $src2}",
5346                (CMGTv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
5347def : InstAlias<"{cmlt\t$dst.8h, $src1.8h, $src2.8h" #
5348                "|cmlt.8h\t$dst, $src1, $src2}",
5349                (CMGTv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
5350def : InstAlias<"{cmlt\t$dst.2s, $src1.2s, $src2.2s" #
5351                "|cmlt.2s\t$dst, $src1, $src2}",
5352                (CMGTv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
5353def : InstAlias<"{cmlt\t$dst.4s, $src1.4s, $src2.4s" #
5354                "|cmlt.4s\t$dst, $src1, $src2}",
5355                (CMGTv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
5356def : InstAlias<"{cmlt\t$dst.2d, $src1.2d, $src2.2d" #
5357                "|cmlt.2d\t$dst, $src1, $src2}",
5358                (CMGTv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
5359
5360let Predicates = [HasNEON, HasFullFP16] in {
5361def : InstAlias<"{fcmle\t$dst.4h, $src1.4h, $src2.4h" #
5362                "|fcmle.4h\t$dst, $src1, $src2}",
5363                (FCMGEv4f16 V64:$dst, V64:$src2, V64:$src1), 0>;
5364def : InstAlias<"{fcmle\t$dst.8h, $src1.8h, $src2.8h" #
5365                "|fcmle.8h\t$dst, $src1, $src2}",
5366                (FCMGEv8f16 V128:$dst, V128:$src2, V128:$src1), 0>;
5367}
5368def : InstAlias<"{fcmle\t$dst.2s, $src1.2s, $src2.2s" #
5369                "|fcmle.2s\t$dst, $src1, $src2}",
5370                (FCMGEv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
5371def : InstAlias<"{fcmle\t$dst.4s, $src1.4s, $src2.4s" #
5372                "|fcmle.4s\t$dst, $src1, $src2}",
5373                (FCMGEv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
5374def : InstAlias<"{fcmle\t$dst.2d, $src1.2d, $src2.2d" #
5375                "|fcmle.2d\t$dst, $src1, $src2}",
5376                (FCMGEv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
5377
5378let Predicates = [HasNEON, HasFullFP16] in {
5379def : InstAlias<"{fcmlt\t$dst.4h, $src1.4h, $src2.4h" #
5380                "|fcmlt.4h\t$dst, $src1, $src2}",
5381                (FCMGTv4f16 V64:$dst, V64:$src2, V64:$src1), 0>;
5382def : InstAlias<"{fcmlt\t$dst.8h, $src1.8h, $src2.8h" #
5383                "|fcmlt.8h\t$dst, $src1, $src2}",
5384                (FCMGTv8f16 V128:$dst, V128:$src2, V128:$src1), 0>;
5385}
5386def : InstAlias<"{fcmlt\t$dst.2s, $src1.2s, $src2.2s" #
5387                "|fcmlt.2s\t$dst, $src1, $src2}",
5388                (FCMGTv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
5389def : InstAlias<"{fcmlt\t$dst.4s, $src1.4s, $src2.4s" #
5390                "|fcmlt.4s\t$dst, $src1, $src2}",
5391                (FCMGTv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
5392def : InstAlias<"{fcmlt\t$dst.2d, $src1.2d, $src2.2d" #
5393                "|fcmlt.2d\t$dst, $src1, $src2}",
5394                (FCMGTv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
5395
5396let Predicates = [HasNEON, HasFullFP16] in {
5397def : InstAlias<"{facle\t$dst.4h, $src1.4h, $src2.4h" #
5398                "|facle.4h\t$dst, $src1, $src2}",
5399                (FACGEv4f16 V64:$dst, V64:$src2, V64:$src1), 0>;
5400def : InstAlias<"{facle\t$dst.8h, $src1.8h, $src2.8h" #
5401                "|facle.8h\t$dst, $src1, $src2}",
5402                (FACGEv8f16 V128:$dst, V128:$src2, V128:$src1), 0>;
5403}
5404def : InstAlias<"{facle\t$dst.2s, $src1.2s, $src2.2s" #
5405                "|facle.2s\t$dst, $src1, $src2}",
5406                (FACGEv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
5407def : InstAlias<"{facle\t$dst.4s, $src1.4s, $src2.4s" #
5408                "|facle.4s\t$dst, $src1, $src2}",
5409                (FACGEv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
5410def : InstAlias<"{facle\t$dst.2d, $src1.2d, $src2.2d" #
5411                "|facle.2d\t$dst, $src1, $src2}",
5412                (FACGEv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
5413
5414let Predicates = [HasNEON, HasFullFP16] in {
5415def : InstAlias<"{faclt\t$dst.4h, $src1.4h, $src2.4h" #
5416                "|faclt.4h\t$dst, $src1, $src2}",
5417                (FACGTv4f16 V64:$dst, V64:$src2, V64:$src1), 0>;
5418def : InstAlias<"{faclt\t$dst.8h, $src1.8h, $src2.8h" #
5419                "|faclt.8h\t$dst, $src1, $src2}",
5420                (FACGTv8f16 V128:$dst, V128:$src2, V128:$src1), 0>;
5421}
5422def : InstAlias<"{faclt\t$dst.2s, $src1.2s, $src2.2s" #
5423                "|faclt.2s\t$dst, $src1, $src2}",
5424                (FACGTv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
5425def : InstAlias<"{faclt\t$dst.4s, $src1.4s, $src2.4s" #
5426                "|faclt.4s\t$dst, $src1, $src2}",
5427                (FACGTv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
5428def : InstAlias<"{faclt\t$dst.2d, $src1.2d, $src2.2d" #
5429                "|faclt.2d\t$dst, $src1, $src2}",
5430                (FACGTv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
5431
5432//===----------------------------------------------------------------------===//
5433// Advanced SIMD three scalar instructions.
5434//===----------------------------------------------------------------------===//
5435
5436defm ADD      : SIMDThreeScalarD<0, 0b10000, "add", add>;
5437defm CMEQ     : SIMDThreeScalarD<1, 0b10001, "cmeq", AArch64cmeq>;
5438defm CMGE     : SIMDThreeScalarD<0, 0b00111, "cmge", AArch64cmge>;
5439defm CMGT     : SIMDThreeScalarD<0, 0b00110, "cmgt", AArch64cmgt>;
5440defm CMHI     : SIMDThreeScalarD<1, 0b00110, "cmhi", AArch64cmhi>;
5441defm CMHS     : SIMDThreeScalarD<1, 0b00111, "cmhs", AArch64cmhs>;
5442defm CMTST    : SIMDThreeScalarD<0, 0b10001, "cmtst", AArch64cmtst>;
5443defm FABD     : SIMDFPThreeScalar<1, 1, 0b010, "fabd", int_aarch64_sisd_fabd>;
5444def : Pat<(v1f64 (int_aarch64_neon_fabd (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
5445          (FABD64 FPR64:$Rn, FPR64:$Rm)>;
5446let Predicates = [HasNEON, HasFullFP16] in {
5447def : Pat<(fabs (fsub f16:$Rn, f16:$Rm)), (FABD16 f16:$Rn, f16:$Rm)>;
5448}
5449let Predicates = [HasNEON] in {
5450def : Pat<(fabs (fsub f32:$Rn, f32:$Rm)), (FABD32 f32:$Rn, f32:$Rm)>;
5451def : Pat<(fabs (fsub f64:$Rn, f64:$Rm)), (FABD64 f64:$Rn, f64:$Rm)>;
5452}
5453defm FACGE    : SIMDThreeScalarFPCmp<1, 0, 0b101, "facge",
5454                                     int_aarch64_neon_facge>;
5455defm FACGT    : SIMDThreeScalarFPCmp<1, 1, 0b101, "facgt",
5456                                     int_aarch64_neon_facgt>;
5457defm FCMEQ    : SIMDThreeScalarFPCmp<0, 0, 0b100, "fcmeq", AArch64fcmeq>;
5458defm FCMGE    : SIMDThreeScalarFPCmp<1, 0, 0b100, "fcmge", AArch64fcmge>;
5459defm FCMGT    : SIMDThreeScalarFPCmp<1, 1, 0b100, "fcmgt", AArch64fcmgt>;
5460defm FMULX    : SIMDFPThreeScalar<0, 0, 0b011, "fmulx", int_aarch64_neon_fmulx, HasNEONorSME>;
5461defm FRECPS   : SIMDFPThreeScalar<0, 0, 0b111, "frecps", int_aarch64_neon_frecps, HasNEONorSME>;
5462defm FRSQRTS  : SIMDFPThreeScalar<0, 1, 0b111, "frsqrts", int_aarch64_neon_frsqrts, HasNEONorSME>;
5463defm SQADD    : SIMDThreeScalarBHSD<0, 0b00001, "sqadd", int_aarch64_neon_sqadd>;
5464defm SQDMULH  : SIMDThreeScalarHS<  0, 0b10110, "sqdmulh", int_aarch64_neon_sqdmulh>;
5465defm SQRDMULH : SIMDThreeScalarHS<  1, 0b10110, "sqrdmulh", int_aarch64_neon_sqrdmulh>;
5466defm SQRSHL   : SIMDThreeScalarBHSD<0, 0b01011, "sqrshl",int_aarch64_neon_sqrshl>;
5467defm SQSHL    : SIMDThreeScalarBHSD<0, 0b01001, "sqshl", int_aarch64_neon_sqshl>;
5468defm SQSUB    : SIMDThreeScalarBHSD<0, 0b00101, "sqsub", int_aarch64_neon_sqsub>;
5469defm SRSHL    : SIMDThreeScalarD<   0, 0b01010, "srshl", int_aarch64_neon_srshl>;
5470defm SSHL     : SIMDThreeScalarD<   0, 0b01000, "sshl", int_aarch64_neon_sshl>;
5471defm SUB      : SIMDThreeScalarD<   1, 0b10000, "sub", sub>;
5472defm UQADD    : SIMDThreeScalarBHSD<1, 0b00001, "uqadd", int_aarch64_neon_uqadd>;
5473defm UQRSHL   : SIMDThreeScalarBHSD<1, 0b01011, "uqrshl",int_aarch64_neon_uqrshl>;
5474defm UQSHL    : SIMDThreeScalarBHSD<1, 0b01001, "uqshl", int_aarch64_neon_uqshl>;
5475defm UQSUB    : SIMDThreeScalarBHSD<1, 0b00101, "uqsub", int_aarch64_neon_uqsub>;
5476defm URSHL    : SIMDThreeScalarD<   1, 0b01010, "urshl", int_aarch64_neon_urshl>;
5477defm USHL     : SIMDThreeScalarD<   1, 0b01000, "ushl", int_aarch64_neon_ushl>;
5478let Predicates = [HasRDM] in {
5479  defm SQRDMLAH : SIMDThreeScalarHSTied<1, 0, 0b10000, "sqrdmlah">;
5480  defm SQRDMLSH : SIMDThreeScalarHSTied<1, 0, 0b10001, "sqrdmlsh">;
5481  def : Pat<(i32 (int_aarch64_neon_sqrdmlah (i32 FPR32:$Rd), (i32 FPR32:$Rn),
5482                                            (i32 FPR32:$Rm))),
5483            (SQRDMLAHv1i32 FPR32:$Rd, FPR32:$Rn, FPR32:$Rm)>;
5484  def : Pat<(i32 (int_aarch64_neon_sqrdmlsh (i32 FPR32:$Rd), (i32 FPR32:$Rn),
5485                                            (i32 FPR32:$Rm))),
5486            (SQRDMLSHv1i32 FPR32:$Rd, FPR32:$Rn, FPR32:$Rm)>;
5487}
5488
5489defm : FMULScalarFromIndexedLane0Patterns<"FMULX", "16", "32", "64",
5490                                          int_aarch64_neon_fmulx,
5491                                          [HasNEONorSME]>;
5492
5493def : InstAlias<"cmls $dst, $src1, $src2",
5494                (CMHSv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
5495def : InstAlias<"cmle $dst, $src1, $src2",
5496                (CMGEv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
5497def : InstAlias<"cmlo $dst, $src1, $src2",
5498                (CMHIv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
5499def : InstAlias<"cmlt $dst, $src1, $src2",
5500                (CMGTv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
5501def : InstAlias<"fcmle $dst, $src1, $src2",
5502                (FCMGE32 FPR32:$dst, FPR32:$src2, FPR32:$src1), 0>;
5503def : InstAlias<"fcmle $dst, $src1, $src2",
5504                (FCMGE64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
5505def : InstAlias<"fcmlt $dst, $src1, $src2",
5506                (FCMGT32 FPR32:$dst, FPR32:$src2, FPR32:$src1), 0>;
5507def : InstAlias<"fcmlt $dst, $src1, $src2",
5508                (FCMGT64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
5509def : InstAlias<"facle $dst, $src1, $src2",
5510                (FACGE32 FPR32:$dst, FPR32:$src2, FPR32:$src1), 0>;
5511def : InstAlias<"facle $dst, $src1, $src2",
5512                (FACGE64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
5513def : InstAlias<"faclt $dst, $src1, $src2",
5514                (FACGT32 FPR32:$dst, FPR32:$src2, FPR32:$src1), 0>;
5515def : InstAlias<"faclt $dst, $src1, $src2",
5516                (FACGT64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
5517
5518//===----------------------------------------------------------------------===//
5519// Advanced SIMD three scalar instructions (mixed operands).
5520//===----------------------------------------------------------------------===//
5521defm SQDMULL  : SIMDThreeScalarMixedHS<0, 0b11010, "sqdmull",
5522                                       int_aarch64_neon_sqdmulls_scalar>;
5523defm SQDMLAL  : SIMDThreeScalarMixedTiedHS<0, 0b10010, "sqdmlal">;
5524defm SQDMLSL  : SIMDThreeScalarMixedTiedHS<0, 0b10110, "sqdmlsl">;
5525
5526def : Pat<(i64 (int_aarch64_neon_sqadd (i64 FPR64:$Rd),
5527                   (i64 (int_aarch64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
5528                                                        (i32 FPR32:$Rm))))),
5529          (SQDMLALi32 FPR64:$Rd, FPR32:$Rn, FPR32:$Rm)>;
5530def : Pat<(i64 (int_aarch64_neon_sqsub (i64 FPR64:$Rd),
5531                   (i64 (int_aarch64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
5532                                                        (i32 FPR32:$Rm))))),
5533          (SQDMLSLi32 FPR64:$Rd, FPR32:$Rn, FPR32:$Rm)>;
5534
5535//===----------------------------------------------------------------------===//
5536// Advanced SIMD two scalar instructions.
5537//===----------------------------------------------------------------------===//
5538
5539defm ABS    : SIMDTwoScalarD<    0, 0b01011, "abs", abs, [HasNoCSSC]>;
5540defm CMEQ   : SIMDCmpTwoScalarD< 0, 0b01001, "cmeq", AArch64cmeqz>;
5541defm CMGE   : SIMDCmpTwoScalarD< 1, 0b01000, "cmge", AArch64cmgez>;
5542defm CMGT   : SIMDCmpTwoScalarD< 0, 0b01000, "cmgt", AArch64cmgtz>;
5543defm CMLE   : SIMDCmpTwoScalarD< 1, 0b01001, "cmle", AArch64cmlez>;
5544defm CMLT   : SIMDCmpTwoScalarD< 0, 0b01010, "cmlt", AArch64cmltz>;
5545defm FCMEQ  : SIMDFPCmpTwoScalar<0, 1, 0b01101, "fcmeq", AArch64fcmeqz>;
5546defm FCMGE  : SIMDFPCmpTwoScalar<1, 1, 0b01100, "fcmge", AArch64fcmgez>;
5547defm FCMGT  : SIMDFPCmpTwoScalar<0, 1, 0b01100, "fcmgt", AArch64fcmgtz>;
5548defm FCMLE  : SIMDFPCmpTwoScalar<1, 1, 0b01101, "fcmle", AArch64fcmlez>;
5549defm FCMLT  : SIMDFPCmpTwoScalar<0, 1, 0b01110, "fcmlt", AArch64fcmltz>;
5550defm FCVTAS : SIMDFPTwoScalar<   0, 0, 0b11100, "fcvtas">;
5551defm FCVTAU : SIMDFPTwoScalar<   1, 0, 0b11100, "fcvtau">;
5552defm FCVTMS : SIMDFPTwoScalar<   0, 0, 0b11011, "fcvtms">;
5553defm FCVTMU : SIMDFPTwoScalar<   1, 0, 0b11011, "fcvtmu">;
5554defm FCVTNS : SIMDFPTwoScalar<   0, 0, 0b11010, "fcvtns">;
5555defm FCVTNU : SIMDFPTwoScalar<   1, 0, 0b11010, "fcvtnu">;
5556defm FCVTPS : SIMDFPTwoScalar<   0, 1, 0b11010, "fcvtps">;
5557defm FCVTPU : SIMDFPTwoScalar<   1, 1, 0b11010, "fcvtpu">;
5558def  FCVTXNv1i64 : SIMDInexactCvtTwoScalar<0b10110, "fcvtxn">;
5559defm FCVTZS : SIMDFPTwoScalar<   0, 1, 0b11011, "fcvtzs">;
5560defm FCVTZU : SIMDFPTwoScalar<   1, 1, 0b11011, "fcvtzu">;
5561defm FRECPE : SIMDFPTwoScalar<   0, 1, 0b11101, "frecpe", HasNEONorSME>;
5562defm FRECPX : SIMDFPTwoScalar<   0, 1, 0b11111, "frecpx", HasNEONorSME>;
5563defm FRSQRTE : SIMDFPTwoScalar<  1, 1, 0b11101, "frsqrte", HasNEONorSME>;
5564defm NEG    : SIMDTwoScalarD<    1, 0b01011, "neg",
5565                                 UnOpFrag<(sub immAllZerosV, node:$LHS)> >;
5566defm SCVTF  : SIMDFPTwoScalarCVT<   0, 0, 0b11101, "scvtf", AArch64sitof>;
5567defm SQABS  : SIMDTwoScalarBHSD< 0, 0b00111, "sqabs", int_aarch64_neon_sqabs>;
5568defm SQNEG  : SIMDTwoScalarBHSD< 1, 0b00111, "sqneg", int_aarch64_neon_sqneg>;
5569defm SQXTN  : SIMDTwoScalarMixedBHS< 0, 0b10100, "sqxtn", int_aarch64_neon_scalar_sqxtn>;
5570defm SQXTUN : SIMDTwoScalarMixedBHS< 1, 0b10010, "sqxtun", int_aarch64_neon_scalar_sqxtun>;
5571defm SUQADD : SIMDTwoScalarBHSDTied< 0, 0b00011, "suqadd",
5572                                     int_aarch64_neon_suqadd>;
5573defm UCVTF  : SIMDFPTwoScalarCVT<   1, 0, 0b11101, "ucvtf", AArch64uitof>;
5574defm UQXTN  : SIMDTwoScalarMixedBHS<1, 0b10100, "uqxtn", int_aarch64_neon_scalar_uqxtn>;
5575defm USQADD : SIMDTwoScalarBHSDTied< 1, 0b00011, "usqadd",
5576                                    int_aarch64_neon_usqadd>;
5577
5578def : Pat<(v1i64 (AArch64vashr (v1i64 V64:$Rn), (i32 63))),
5579          (CMLTv1i64rz V64:$Rn)>;
5580
5581def : Pat<(v1i64 (int_aarch64_neon_fcvtas (v1f64 FPR64:$Rn))),
5582          (FCVTASv1i64 FPR64:$Rn)>;
5583def : Pat<(v1i64 (int_aarch64_neon_fcvtau (v1f64 FPR64:$Rn))),
5584          (FCVTAUv1i64 FPR64:$Rn)>;
5585def : Pat<(v1i64 (int_aarch64_neon_fcvtms (v1f64 FPR64:$Rn))),
5586          (FCVTMSv1i64 FPR64:$Rn)>;
5587def : Pat<(v1i64 (int_aarch64_neon_fcvtmu (v1f64 FPR64:$Rn))),
5588          (FCVTMUv1i64 FPR64:$Rn)>;
5589def : Pat<(v1i64 (int_aarch64_neon_fcvtns (v1f64 FPR64:$Rn))),
5590          (FCVTNSv1i64 FPR64:$Rn)>;
5591def : Pat<(v1i64 (int_aarch64_neon_fcvtnu (v1f64 FPR64:$Rn))),
5592          (FCVTNUv1i64 FPR64:$Rn)>;
5593def : Pat<(v1i64 (int_aarch64_neon_fcvtps (v1f64 FPR64:$Rn))),
5594          (FCVTPSv1i64 FPR64:$Rn)>;
5595def : Pat<(v1i64 (int_aarch64_neon_fcvtpu (v1f64 FPR64:$Rn))),
5596          (FCVTPUv1i64 FPR64:$Rn)>;
5597def : Pat<(v1i64 (int_aarch64_neon_fcvtzs (v1f64 FPR64:$Rn))),
5598          (FCVTZSv1i64 FPR64:$Rn)>;
5599def : Pat<(v1i64 (int_aarch64_neon_fcvtzu (v1f64 FPR64:$Rn))),
5600          (FCVTZUv1i64 FPR64:$Rn)>;
5601
5602def : Pat<(f16 (int_aarch64_neon_frecpe (f16 FPR16:$Rn))),
5603          (FRECPEv1f16 FPR16:$Rn)>;
5604def : Pat<(f32 (int_aarch64_neon_frecpe (f32 FPR32:$Rn))),
5605          (FRECPEv1i32 FPR32:$Rn)>;
5606def : Pat<(f64 (int_aarch64_neon_frecpe (f64 FPR64:$Rn))),
5607          (FRECPEv1i64 FPR64:$Rn)>;
5608def : Pat<(v1f64 (int_aarch64_neon_frecpe (v1f64 FPR64:$Rn))),
5609          (FRECPEv1i64 FPR64:$Rn)>;
5610
5611def : Pat<(f32 (AArch64frecpe (f32 FPR32:$Rn))),
5612          (FRECPEv1i32 FPR32:$Rn)>;
5613def : Pat<(v2f32 (AArch64frecpe (v2f32 V64:$Rn))),
5614          (FRECPEv2f32 V64:$Rn)>;
5615def : Pat<(v4f32 (AArch64frecpe (v4f32 FPR128:$Rn))),
5616          (FRECPEv4f32 FPR128:$Rn)>;
5617def : Pat<(f64 (AArch64frecpe (f64 FPR64:$Rn))),
5618          (FRECPEv1i64 FPR64:$Rn)>;
5619def : Pat<(v1f64 (AArch64frecpe (v1f64 FPR64:$Rn))),
5620          (FRECPEv1i64 FPR64:$Rn)>;
5621def : Pat<(v2f64 (AArch64frecpe (v2f64 FPR128:$Rn))),
5622          (FRECPEv2f64 FPR128:$Rn)>;
5623
5624def : Pat<(f32 (AArch64frecps (f32 FPR32:$Rn), (f32 FPR32:$Rm))),
5625          (FRECPS32 FPR32:$Rn, FPR32:$Rm)>;
5626def : Pat<(v2f32 (AArch64frecps (v2f32 V64:$Rn), (v2f32 V64:$Rm))),
5627          (FRECPSv2f32 V64:$Rn, V64:$Rm)>;
5628def : Pat<(v4f32 (AArch64frecps (v4f32 FPR128:$Rn), (v4f32 FPR128:$Rm))),
5629          (FRECPSv4f32 FPR128:$Rn, FPR128:$Rm)>;
5630def : Pat<(f64 (AArch64frecps (f64 FPR64:$Rn), (f64 FPR64:$Rm))),
5631          (FRECPS64 FPR64:$Rn, FPR64:$Rm)>;
5632def : Pat<(v2f64 (AArch64frecps (v2f64 FPR128:$Rn), (v2f64 FPR128:$Rm))),
5633          (FRECPSv2f64 FPR128:$Rn, FPR128:$Rm)>;
5634
5635def : Pat<(f16 (int_aarch64_neon_frecpx (f16 FPR16:$Rn))),
5636          (FRECPXv1f16 FPR16:$Rn)>;
5637def : Pat<(f32 (int_aarch64_neon_frecpx (f32 FPR32:$Rn))),
5638          (FRECPXv1i32 FPR32:$Rn)>;
5639def : Pat<(f64 (int_aarch64_neon_frecpx (f64 FPR64:$Rn))),
5640          (FRECPXv1i64 FPR64:$Rn)>;
5641
5642def : Pat<(f16 (int_aarch64_neon_frsqrte (f16 FPR16:$Rn))),
5643          (FRSQRTEv1f16 FPR16:$Rn)>;
5644def : Pat<(f32 (int_aarch64_neon_frsqrte (f32 FPR32:$Rn))),
5645          (FRSQRTEv1i32 FPR32:$Rn)>;
5646def : Pat<(f64 (int_aarch64_neon_frsqrte (f64 FPR64:$Rn))),
5647          (FRSQRTEv1i64 FPR64:$Rn)>;
5648def : Pat<(v1f64 (int_aarch64_neon_frsqrte (v1f64 FPR64:$Rn))),
5649          (FRSQRTEv1i64 FPR64:$Rn)>;
5650
5651def : Pat<(f32 (AArch64frsqrte (f32 FPR32:$Rn))),
5652          (FRSQRTEv1i32 FPR32:$Rn)>;
5653def : Pat<(v2f32 (AArch64frsqrte (v2f32 V64:$Rn))),
5654          (FRSQRTEv2f32 V64:$Rn)>;
5655def : Pat<(v4f32 (AArch64frsqrte (v4f32 FPR128:$Rn))),
5656          (FRSQRTEv4f32 FPR128:$Rn)>;
5657def : Pat<(f64 (AArch64frsqrte (f64 FPR64:$Rn))),
5658          (FRSQRTEv1i64 FPR64:$Rn)>;
5659def : Pat<(v1f64 (AArch64frsqrte (v1f64 FPR64:$Rn))),
5660          (FRSQRTEv1i64 FPR64:$Rn)>;
5661def : Pat<(v2f64 (AArch64frsqrte (v2f64 FPR128:$Rn))),
5662          (FRSQRTEv2f64 FPR128:$Rn)>;
5663
5664def : Pat<(f32 (AArch64frsqrts (f32 FPR32:$Rn), (f32 FPR32:$Rm))),
5665          (FRSQRTS32 FPR32:$Rn, FPR32:$Rm)>;
5666def : Pat<(v2f32 (AArch64frsqrts (v2f32 V64:$Rn), (v2f32 V64:$Rm))),
5667          (FRSQRTSv2f32 V64:$Rn, V64:$Rm)>;
5668def : Pat<(v4f32 (AArch64frsqrts (v4f32 FPR128:$Rn), (v4f32 FPR128:$Rm))),
5669          (FRSQRTSv4f32 FPR128:$Rn, FPR128:$Rm)>;
5670def : Pat<(f64 (AArch64frsqrts (f64 FPR64:$Rn), (f64 FPR64:$Rm))),
5671          (FRSQRTS64 FPR64:$Rn, FPR64:$Rm)>;
5672def : Pat<(v2f64 (AArch64frsqrts (v2f64 FPR128:$Rn), (v2f64 FPR128:$Rm))),
5673          (FRSQRTSv2f64 FPR128:$Rn, FPR128:$Rm)>;
5674
5675// Some float -> int -> float conversion patterns for which we want to keep the
5676// int values in FP registers using the corresponding NEON instructions to
5677// avoid more costly int <-> fp register transfers.
5678let Predicates = [HasNEON] in {
5679def : Pat<(f64 (any_sint_to_fp (i64 (any_fp_to_sint f64:$Rn)))),
5680          (SCVTFv1i64 (i64 (FCVTZSv1i64 f64:$Rn)))>;
5681def : Pat<(f32 (any_sint_to_fp (i32 (any_fp_to_sint f32:$Rn)))),
5682          (SCVTFv1i32 (i32 (FCVTZSv1i32 f32:$Rn)))>;
5683def : Pat<(f64 (any_uint_to_fp (i64 (any_fp_to_uint f64:$Rn)))),
5684          (UCVTFv1i64 (i64 (FCVTZUv1i64 f64:$Rn)))>;
5685def : Pat<(f32 (any_uint_to_fp (i32 (any_fp_to_uint f32:$Rn)))),
5686          (UCVTFv1i32 (i32 (FCVTZUv1i32 f32:$Rn)))>;
5687
5688let Predicates = [HasFullFP16] in {
5689def : Pat<(f16 (any_sint_to_fp (i32 (any_fp_to_sint f16:$Rn)))),
5690          (SCVTFv1i16 (f16 (FCVTZSv1f16 f16:$Rn)))>;
5691def : Pat<(f16 (any_uint_to_fp (i32 (any_fp_to_uint f16:$Rn)))),
5692          (UCVTFv1i16 (f16 (FCVTZUv1f16 f16:$Rn)))>;
5693}
5694// If an integer is about to be converted to a floating point value,
5695// just load it on the floating point unit.
5696// Here are the patterns for 8 and 16-bits to float.
5697// 8-bits -> float.
5698multiclass UIntToFPROLoadPat<ValueType DstTy, ValueType SrcTy,
5699                             SDPatternOperator loadop, Instruction UCVTF,
5700                             ROAddrMode ro, Instruction LDRW, Instruction LDRX,
5701                             SubRegIndex sub> {
5702  def : Pat<(DstTy (uint_to_fp (SrcTy
5703                     (loadop (ro.Wpat GPR64sp:$Rn, GPR32:$Rm,
5704                                      ro.Wext:$extend))))),
5705           (UCVTF (INSERT_SUBREG (DstTy (IMPLICIT_DEF)),
5706                                 (LDRW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend),
5707                                 sub))>;
5708
5709  def : Pat<(DstTy (uint_to_fp (SrcTy
5710                     (loadop (ro.Xpat GPR64sp:$Rn, GPR64:$Rm,
5711                                      ro.Wext:$extend))))),
5712           (UCVTF (INSERT_SUBREG (DstTy (IMPLICIT_DEF)),
5713                                 (LDRX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend),
5714                                 sub))>;
5715}
5716
5717defm : UIntToFPROLoadPat<f32, i32, zextloadi8,
5718                         UCVTFv1i32, ro8, LDRBroW, LDRBroX, bsub>;
5719def : Pat <(f32 (uint_to_fp (i32
5720               (zextloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))))),
5721           (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
5722                          (LDRBui GPR64sp:$Rn, uimm12s1:$offset), bsub))>;
5723def : Pat <(f32 (uint_to_fp (i32
5724                     (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))))),
5725           (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
5726                          (LDURBi GPR64sp:$Rn, simm9:$offset), bsub))>;
5727// 16-bits -> float.
5728defm : UIntToFPROLoadPat<f32, i32, zextloadi16,
5729                         UCVTFv1i32, ro16, LDRHroW, LDRHroX, hsub>;
5730def : Pat <(f32 (uint_to_fp (i32
5731                  (zextloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))))),
5732           (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
5733                          (LDRHui GPR64sp:$Rn, uimm12s2:$offset), hsub))>;
5734def : Pat <(f32 (uint_to_fp (i32
5735                  (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))))),
5736           (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
5737                          (LDURHi GPR64sp:$Rn, simm9:$offset), hsub))>;
5738// 32-bits are handled in target specific dag combine:
5739// performIntToFpCombine.
5740// 64-bits integer to 32-bits floating point, not possible with
5741// UCVTF on floating point registers (both source and destination
5742// must have the same size).
5743
5744// Here are the patterns for 8, 16, 32, and 64-bits to double.
5745// 8-bits -> double.
5746defm : UIntToFPROLoadPat<f64, i32, zextloadi8,
5747                         UCVTFv1i64, ro8, LDRBroW, LDRBroX, bsub>;
5748def : Pat <(f64 (uint_to_fp (i32
5749                    (zextloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))))),
5750           (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
5751                          (LDRBui GPR64sp:$Rn, uimm12s1:$offset), bsub))>;
5752def : Pat <(f64 (uint_to_fp (i32
5753                  (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))))),
5754           (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
5755                          (LDURBi GPR64sp:$Rn, simm9:$offset), bsub))>;
5756// 16-bits -> double.
5757defm : UIntToFPROLoadPat<f64, i32, zextloadi16,
5758                         UCVTFv1i64, ro16, LDRHroW, LDRHroX, hsub>;
5759def : Pat <(f64 (uint_to_fp (i32
5760                  (zextloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))))),
5761           (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
5762                          (LDRHui GPR64sp:$Rn, uimm12s2:$offset), hsub))>;
5763def : Pat <(f64 (uint_to_fp (i32
5764                  (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))))),
5765           (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
5766                          (LDURHi GPR64sp:$Rn, simm9:$offset), hsub))>;
5767// 32-bits -> double.
5768defm : UIntToFPROLoadPat<f64, i32, load,
5769                         UCVTFv1i64, ro32, LDRSroW, LDRSroX, ssub>;
5770def : Pat <(f64 (uint_to_fp (i32
5771                  (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))))),
5772           (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
5773                          (LDRSui GPR64sp:$Rn, uimm12s4:$offset), ssub))>;
5774def : Pat <(f64 (uint_to_fp (i32
5775                  (load (am_unscaled32 GPR64sp:$Rn, simm9:$offset))))),
5776           (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
5777                          (LDURSi GPR64sp:$Rn, simm9:$offset), ssub))>;
5778// 64-bits -> double are handled in target specific dag combine:
5779// performIntToFpCombine.
5780} // let Predicates = [HasNEON]
5781
5782//===----------------------------------------------------------------------===//
5783// Advanced SIMD three different-sized vector instructions.
5784//===----------------------------------------------------------------------===//
5785
5786defm ADDHN  : SIMDNarrowThreeVectorBHS<0,0b0100,"addhn", int_aarch64_neon_addhn>;
5787defm SUBHN  : SIMDNarrowThreeVectorBHS<0,0b0110,"subhn", int_aarch64_neon_subhn>;
5788defm RADDHN : SIMDNarrowThreeVectorBHS<1,0b0100,"raddhn",int_aarch64_neon_raddhn>;
5789defm RSUBHN : SIMDNarrowThreeVectorBHS<1,0b0110,"rsubhn",int_aarch64_neon_rsubhn>;
5790defm PMULL  : SIMDDifferentThreeVectorBD<0,0b1110,"pmull", AArch64pmull>;
5791defm SABAL  : SIMDLongThreeVectorTiedBHSabal<0,0b0101,"sabal",
5792                                             AArch64sabd>;
5793defm SABDL   : SIMDLongThreeVectorBHSabdl<0, 0b0111, "sabdl",
5794                                          AArch64sabd>;
5795defm SADDL   : SIMDLongThreeVectorBHS<   0, 0b0000, "saddl",
5796            BinOpFrag<(add (sext node:$LHS), (sext node:$RHS))>>;
5797defm SADDW   : SIMDWideThreeVectorBHS<   0, 0b0001, "saddw",
5798                 BinOpFrag<(add node:$LHS, (sext node:$RHS))>>;
5799defm SMLAL   : SIMDLongThreeVectorTiedBHS<0, 0b1000, "smlal",
5800    TriOpFrag<(add node:$LHS, (AArch64smull node:$MHS, node:$RHS))>>;
5801defm SMLSL   : SIMDLongThreeVectorTiedBHS<0, 0b1010, "smlsl",
5802    TriOpFrag<(sub node:$LHS, (AArch64smull node:$MHS, node:$RHS))>>;
5803defm SMULL   : SIMDLongThreeVectorBHS<0, 0b1100, "smull", AArch64smull>;
5804defm SQDMLAL : SIMDLongThreeVectorSQDMLXTiedHS<0, 0b1001, "sqdmlal",
5805                                               int_aarch64_neon_sqadd>;
5806defm SQDMLSL : SIMDLongThreeVectorSQDMLXTiedHS<0, 0b1011, "sqdmlsl",
5807                                               int_aarch64_neon_sqsub>;
5808defm SQDMULL : SIMDLongThreeVectorHS<0, 0b1101, "sqdmull",
5809                                     int_aarch64_neon_sqdmull>;
5810defm SSUBL   : SIMDLongThreeVectorBHS<0, 0b0010, "ssubl",
5811                 BinOpFrag<(sub (sext node:$LHS), (sext node:$RHS))>>;
5812defm SSUBW   : SIMDWideThreeVectorBHS<0, 0b0011, "ssubw",
5813                 BinOpFrag<(sub node:$LHS, (sext node:$RHS))>>;
5814defm UABAL   : SIMDLongThreeVectorTiedBHSabal<1, 0b0101, "uabal",
5815                                              AArch64uabd>;
5816defm UADDL   : SIMDLongThreeVectorBHS<1, 0b0000, "uaddl",
5817                 BinOpFrag<(add (zanyext node:$LHS), (zanyext node:$RHS))>>;
5818defm UADDW   : SIMDWideThreeVectorBHS<1, 0b0001, "uaddw",
5819                 BinOpFrag<(add node:$LHS, (zanyext node:$RHS))>>;
5820defm UMLAL   : SIMDLongThreeVectorTiedBHS<1, 0b1000, "umlal",
5821    TriOpFrag<(add node:$LHS, (AArch64umull node:$MHS, node:$RHS))>>;
5822defm UMLSL   : SIMDLongThreeVectorTiedBHS<1, 0b1010, "umlsl",
5823    TriOpFrag<(sub node:$LHS, (AArch64umull node:$MHS, node:$RHS))>>;
5824defm UMULL   : SIMDLongThreeVectorBHS<1, 0b1100, "umull", AArch64umull>;
5825defm USUBL   : SIMDLongThreeVectorBHS<1, 0b0010, "usubl",
5826                 BinOpFrag<(sub (zanyext node:$LHS), (zanyext node:$RHS))>>;
5827defm USUBW   : SIMDWideThreeVectorBHS<   1, 0b0011, "usubw",
5828                 BinOpFrag<(sub node:$LHS, (zanyext node:$RHS))>>;
5829
5830// Additional patterns for [SU]ML[AS]L
5831multiclass Neon_mul_acc_widen_patterns<SDPatternOperator opnode, SDPatternOperator vecopnode,
5832  Instruction INST8B, Instruction INST4H, Instruction INST2S> {
5833  def : Pat<(v4i16 (opnode
5834                    V64:$Ra,
5835                    (v4i16 (extract_subvector
5836                            (vecopnode (v8i8 V64:$Rn),(v8i8 V64:$Rm)),
5837                            (i64 0))))),
5838             (EXTRACT_SUBREG (v8i16 (INST8B
5839                                     (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), V64:$Ra, dsub),
5840                                     V64:$Rn, V64:$Rm)), dsub)>;
5841  def : Pat<(v2i32 (opnode
5842                    V64:$Ra,
5843                    (v2i32 (extract_subvector
5844                            (vecopnode (v4i16 V64:$Rn),(v4i16 V64:$Rm)),
5845                            (i64 0))))),
5846             (EXTRACT_SUBREG (v4i32 (INST4H
5847                                     (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), V64:$Ra, dsub),
5848                                     V64:$Rn, V64:$Rm)), dsub)>;
5849  def : Pat<(v1i64 (opnode
5850                    V64:$Ra,
5851                    (v1i64 (extract_subvector
5852                            (vecopnode (v2i32 V64:$Rn),(v2i32 V64:$Rm)),
5853                            (i64 0))))),
5854             (EXTRACT_SUBREG (v2i64 (INST2S
5855                                     (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), V64:$Ra, dsub),
5856                                     V64:$Rn, V64:$Rm)), dsub)>;
5857}
5858
5859defm : Neon_mul_acc_widen_patterns<add, AArch64umull,
5860     UMLALv8i8_v8i16, UMLALv4i16_v4i32, UMLALv2i32_v2i64>;
5861defm : Neon_mul_acc_widen_patterns<add, AArch64smull,
5862     SMLALv8i8_v8i16, SMLALv4i16_v4i32, SMLALv2i32_v2i64>;
5863defm : Neon_mul_acc_widen_patterns<sub, AArch64umull,
5864     UMLSLv8i8_v8i16, UMLSLv4i16_v4i32, UMLSLv2i32_v2i64>;
5865defm : Neon_mul_acc_widen_patterns<sub, AArch64smull,
5866     SMLSLv8i8_v8i16, SMLSLv4i16_v4i32, SMLSLv2i32_v2i64>;
5867
5868
5869multiclass Neon_addl_extract_patterns<SDPatternOperator opnode, SDPatternOperator ext, string Inst> {
5870  def : Pat<(v4i16 (opnode (extract_subvector (ext (v8i8 V64:$Rn)), (i64 0)),
5871                           (extract_subvector (ext (v8i8 V64:$Rm)), (i64 0)))),
5872            (EXTRACT_SUBREG (v8i16 (!cast<Instruction>(Inst#"Lv8i8_v8i16") V64:$Rn, V64:$Rm)), dsub)>;
5873  def : Pat<(v2i32 (opnode (extract_subvector (ext (v4i16 V64:$Rn)), (i64 0)),
5874                           (extract_subvector (ext (v4i16 V64:$Rm)), (i64 0)))),
5875            (EXTRACT_SUBREG (v4i32 (!cast<Instruction>(Inst#"Lv4i16_v4i32") V64:$Rn, V64:$Rm)), dsub)>;
5876  def : Pat<(v1i64 (opnode (extract_subvector (ext (v2i32 V64:$Rn)), (i64 0)),
5877                           (extract_subvector (ext (v2i32 V64:$Rm)), (i64 0)))),
5878            (EXTRACT_SUBREG (v2i64 (!cast<Instruction>(Inst#"Lv2i32_v2i64") V64:$Rn, V64:$Rm)), dsub)>;
5879
5880  def : Pat<(v4i16 (opnode (v4i16 V64:$Rn),
5881                           (extract_subvector (ext (v8i8 V64:$Rm)), (i64 0)))),
5882            (EXTRACT_SUBREG (v8i16 (!cast<Instruction>(Inst#"Wv8i8_v8i16") (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), V64:$Rn, dsub), V64:$Rm)), dsub)>;
5883  def : Pat<(v2i32 (opnode (v2i32 V64:$Rn),
5884                           (extract_subvector (ext (v4i16 V64:$Rm)), (i64 0)))),
5885            (EXTRACT_SUBREG (v4i32 (!cast<Instruction>(Inst#"Wv4i16_v4i32") (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), V64:$Rn, dsub), V64:$Rm)), dsub)>;
5886  def : Pat<(v1i64 (opnode (v1i64 V64:$Rn),
5887                           (extract_subvector (ext (v2i32 V64:$Rm)), (i64 0)))),
5888            (EXTRACT_SUBREG (v2i64 (!cast<Instruction>(Inst#"Wv2i32_v2i64") (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), V64:$Rn, dsub), V64:$Rm)), dsub)>;
5889}
5890
5891defm : Neon_addl_extract_patterns<add, zanyext, "UADD">;
5892defm : Neon_addl_extract_patterns<add, sext, "SADD">;
5893defm : Neon_addl_extract_patterns<sub, zanyext, "USUB">;
5894defm : Neon_addl_extract_patterns<sub, sext, "SSUB">;
5895
5896// CodeGen patterns for addhn and subhn instructions, which can actually be
5897// written in LLVM IR without too much difficulty.
5898
5899// Prioritize ADDHN and SUBHN over UZP2.
5900let AddedComplexity = 10 in {
5901
5902// ADDHN
5903def : Pat<(v8i8 (trunc (v8i16 (AArch64vlshr (add V128:$Rn, V128:$Rm), (i32 8))))),
5904          (ADDHNv8i16_v8i8 V128:$Rn, V128:$Rm)>;
5905def : Pat<(v4i16 (trunc (v4i32 (AArch64vlshr (add V128:$Rn, V128:$Rm),
5906                                           (i32 16))))),
5907          (ADDHNv4i32_v4i16 V128:$Rn, V128:$Rm)>;
5908def : Pat<(v2i32 (trunc (v2i64 (AArch64vlshr (add V128:$Rn, V128:$Rm),
5909                                           (i32 32))))),
5910          (ADDHNv2i64_v2i32 V128:$Rn, V128:$Rm)>;
5911def : Pat<(concat_vectors (v8i8 V64:$Rd),
5912                          (trunc (v8i16 (AArch64vlshr (add V128:$Rn, V128:$Rm),
5913                                                    (i32 8))))),
5914          (ADDHNv8i16_v16i8 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
5915                            V128:$Rn, V128:$Rm)>;
5916def : Pat<(concat_vectors (v4i16 V64:$Rd),
5917                          (trunc (v4i32 (AArch64vlshr (add V128:$Rn, V128:$Rm),
5918                                                    (i32 16))))),
5919          (ADDHNv4i32_v8i16 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
5920                            V128:$Rn, V128:$Rm)>;
5921def : Pat<(concat_vectors (v2i32 V64:$Rd),
5922                          (trunc (v2i64 (AArch64vlshr (add V128:$Rn, V128:$Rm),
5923                                                    (i32 32))))),
5924          (ADDHNv2i64_v4i32 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
5925                            V128:$Rn, V128:$Rm)>;
5926
5927// SUBHN
5928def : Pat<(v8i8 (trunc (v8i16 (AArch64vlshr (sub V128:$Rn, V128:$Rm), (i32 8))))),
5929          (SUBHNv8i16_v8i8 V128:$Rn, V128:$Rm)>;
5930def : Pat<(v4i16 (trunc (v4i32 (AArch64vlshr (sub V128:$Rn, V128:$Rm),
5931                                           (i32 16))))),
5932          (SUBHNv4i32_v4i16 V128:$Rn, V128:$Rm)>;
5933def : Pat<(v2i32 (trunc (v2i64 (AArch64vlshr (sub V128:$Rn, V128:$Rm),
5934                                           (i32 32))))),
5935          (SUBHNv2i64_v2i32 V128:$Rn, V128:$Rm)>;
5936def : Pat<(concat_vectors (v8i8 V64:$Rd),
5937                          (trunc (v8i16 (AArch64vlshr (sub V128:$Rn, V128:$Rm),
5938                                                    (i32 8))))),
5939          (SUBHNv8i16_v16i8 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
5940                            V128:$Rn, V128:$Rm)>;
5941def : Pat<(concat_vectors (v4i16 V64:$Rd),
5942                          (trunc (v4i32 (AArch64vlshr (sub V128:$Rn, V128:$Rm),
5943                                                    (i32 16))))),
5944          (SUBHNv4i32_v8i16 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
5945                            V128:$Rn, V128:$Rm)>;
5946def : Pat<(concat_vectors (v2i32 V64:$Rd),
5947                          (trunc (v2i64 (AArch64vlshr (sub V128:$Rn, V128:$Rm),
5948                                                    (i32 32))))),
5949          (SUBHNv2i64_v4i32 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
5950                            V128:$Rn, V128:$Rm)>;
5951
5952} // AddedComplexity = 10
5953
5954//----------------------------------------------------------------------------
5955// AdvSIMD bitwise extract from vector instruction.
5956//----------------------------------------------------------------------------
5957
5958defm EXT : SIMDBitwiseExtract<"ext">;
5959
5960def AdjustExtImm : SDNodeXForm<imm, [{
5961  return CurDAG->getTargetConstant(8 + N->getZExtValue(), SDLoc(N), MVT::i32);
5962}]>;
5963multiclass ExtPat<ValueType VT64, ValueType VT128, int N> {
5964  def : Pat<(VT64 (AArch64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
5965            (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
5966  def : Pat<(VT128 (AArch64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
5967            (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
5968  // We use EXT to handle extract_subvector to copy the upper 64-bits of a
5969  // 128-bit vector.
5970  def : Pat<(VT64 (extract_subvector V128:$Rn, (i64 N))),
5971            (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
5972  // A 64-bit EXT of two halves of the same 128-bit register can be done as a
5973  // single 128-bit EXT.
5974  def : Pat<(VT64 (AArch64ext (extract_subvector V128:$Rn, (i64 0)),
5975                              (extract_subvector V128:$Rn, (i64 N)),
5976                              (i32 imm:$imm))),
5977            (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, imm:$imm), dsub)>;
5978  // A 64-bit EXT of the high half of a 128-bit register can be done using a
5979  // 128-bit EXT of the whole register with an adjustment to the immediate. The
5980  // top half of the other operand will be unset, but that doesn't matter as it
5981  // will not be used.
5982  def : Pat<(VT64 (AArch64ext (extract_subvector V128:$Rn, (i64 N)),
5983                              V64:$Rm,
5984                              (i32 imm:$imm))),
5985            (EXTRACT_SUBREG (EXTv16i8 V128:$Rn,
5986                                      (SUBREG_TO_REG (i32 0), V64:$Rm, dsub),
5987                                      (AdjustExtImm imm:$imm)), dsub)>;
5988}
5989
5990defm : ExtPat<v8i8, v16i8, 8>;
5991defm : ExtPat<v4i16, v8i16, 4>;
5992defm : ExtPat<v4f16, v8f16, 4>;
5993defm : ExtPat<v4bf16, v8bf16, 4>;
5994defm : ExtPat<v2i32, v4i32, 2>;
5995defm : ExtPat<v2f32, v4f32, 2>;
5996defm : ExtPat<v1i64, v2i64, 1>;
5997defm : ExtPat<v1f64, v2f64, 1>;
5998
5999//----------------------------------------------------------------------------
6000// AdvSIMD zip vector
6001//----------------------------------------------------------------------------
6002
6003defm TRN1 : SIMDZipVector<0b010, "trn1", AArch64trn1>;
6004defm TRN2 : SIMDZipVector<0b110, "trn2", AArch64trn2>;
6005defm UZP1 : SIMDZipVector<0b001, "uzp1", AArch64uzp1>;
6006defm UZP2 : SIMDZipVector<0b101, "uzp2", AArch64uzp2>;
6007defm ZIP1 : SIMDZipVector<0b011, "zip1", AArch64zip1>;
6008defm ZIP2 : SIMDZipVector<0b111, "zip2", AArch64zip2>;
6009
6010def : Pat<(v16i8 (concat_vectors (v8i8 (trunc (v8i16 V128:$Vn))),
6011                                 (v8i8 (trunc (v8i16 V128:$Vm))))),
6012          (UZP1v16i8 V128:$Vn, V128:$Vm)>;
6013def : Pat<(v8i16 (concat_vectors (v4i16 (trunc (v4i32 V128:$Vn))),
6014                                 (v4i16 (trunc (v4i32 V128:$Vm))))),
6015          (UZP1v8i16 V128:$Vn, V128:$Vm)>;
6016def : Pat<(v4i32 (concat_vectors (v2i32 (trunc (v2i64 V128:$Vn))),
6017                                 (v2i32 (trunc (v2i64 V128:$Vm))))),
6018          (UZP1v4i32 V128:$Vn, V128:$Vm)>;
6019// These are the same as above, with an optional assertzext node that can be
6020// generated from fptoi lowering.
6021def : Pat<(v16i8 (concat_vectors (v8i8 (assertzext (trunc (v8i16 V128:$Vn)))),
6022                                 (v8i8 (assertzext (trunc (v8i16 V128:$Vm)))))),
6023          (UZP1v16i8 V128:$Vn, V128:$Vm)>;
6024def : Pat<(v8i16 (concat_vectors (v4i16 (assertzext (trunc (v4i32 V128:$Vn)))),
6025                                 (v4i16 (assertzext (trunc (v4i32 V128:$Vm)))))),
6026          (UZP1v8i16 V128:$Vn, V128:$Vm)>;
6027def : Pat<(v4i32 (concat_vectors (v2i32 (assertzext (trunc (v2i64 V128:$Vn)))),
6028                                 (v2i32 (assertzext (trunc (v2i64 V128:$Vm)))))),
6029          (UZP1v4i32 V128:$Vn, V128:$Vm)>;
6030
6031def : Pat<(v16i8 (concat_vectors
6032                 (v8i8 (trunc (AArch64vlshr (v8i16 V128:$Vn), (i32 8)))),
6033                 (v8i8 (trunc (AArch64vlshr (v8i16 V128:$Vm), (i32 8)))))),
6034          (UZP2v16i8 V128:$Vn, V128:$Vm)>;
6035def : Pat<(v8i16 (concat_vectors
6036                 (v4i16 (trunc (AArch64vlshr (v4i32 V128:$Vn), (i32 16)))),
6037                 (v4i16 (trunc (AArch64vlshr (v4i32 V128:$Vm), (i32 16)))))),
6038          (UZP2v8i16 V128:$Vn, V128:$Vm)>;
6039def : Pat<(v4i32 (concat_vectors
6040                 (v2i32 (trunc (AArch64vlshr (v2i64 V128:$Vn), (i32 32)))),
6041                 (v2i32 (trunc (AArch64vlshr (v2i64 V128:$Vm), (i32 32)))))),
6042          (UZP2v4i32 V128:$Vn, V128:$Vm)>;
6043
6044//----------------------------------------------------------------------------
6045// AdvSIMD TBL/TBX instructions
6046//----------------------------------------------------------------------------
6047
6048defm TBL : SIMDTableLookup<    0, "tbl">;
6049defm TBX : SIMDTableLookupTied<1, "tbx">;
6050
6051def : Pat<(v8i8 (int_aarch64_neon_tbl1 (v16i8 VecListOne128:$Rn), (v8i8 V64:$Ri))),
6052          (TBLv8i8One VecListOne128:$Rn, V64:$Ri)>;
6053def : Pat<(v16i8 (int_aarch64_neon_tbl1 (v16i8 V128:$Ri), (v16i8 V128:$Rn))),
6054          (TBLv16i8One V128:$Ri, V128:$Rn)>;
6055
6056def : Pat<(v8i8 (int_aarch64_neon_tbx1 (v8i8 V64:$Rd),
6057                  (v16i8 VecListOne128:$Rn), (v8i8 V64:$Ri))),
6058          (TBXv8i8One V64:$Rd, VecListOne128:$Rn, V64:$Ri)>;
6059def : Pat<(v16i8 (int_aarch64_neon_tbx1 (v16i8 V128:$Rd),
6060                   (v16i8 V128:$Ri), (v16i8 V128:$Rn))),
6061          (TBXv16i8One V128:$Rd, V128:$Ri, V128:$Rn)>;
6062
6063//----------------------------------------------------------------------------
6064// AdvSIMD LUT instructions
6065//----------------------------------------------------------------------------
6066let Predicates = [HasLUT] in {
6067  defm LUT2 : BaseSIMDTableLookupIndexed2<"luti2">;
6068  defm LUT4 : BaseSIMDTableLookupIndexed4<"luti4">;
6069}
6070
6071//----------------------------------------------------------------------------
6072// AdvSIMD scalar DUP instruction
6073//----------------------------------------------------------------------------
6074
6075defm DUP : SIMDScalarDUP<"mov">;
6076
6077//----------------------------------------------------------------------------
6078// AdvSIMD scalar pairwise instructions
6079//----------------------------------------------------------------------------
6080
6081defm ADDP    : SIMDPairwiseScalarD<0, 0b11011, "addp">;
6082defm FADDP   : SIMDFPPairwiseScalar<0, 0b01101, "faddp">;
6083defm FMAXNMP : SIMDFPPairwiseScalar<0, 0b01100, "fmaxnmp">;
6084defm FMAXP   : SIMDFPPairwiseScalar<0, 0b01111, "fmaxp">;
6085defm FMINNMP : SIMDFPPairwiseScalar<1, 0b01100, "fminnmp">;
6086defm FMINP   : SIMDFPPairwiseScalar<1, 0b01111, "fminp">;
6087
6088// Only the lower half of the result of the inner FADDP is used in the patterns
6089// below, so the second operand does not matter. Re-use the first input
6090// operand, so no additional dependencies need to be introduced.
6091let Predicates = [HasFullFP16] in {
6092def : Pat<(f16 (vecreduce_fadd (v8f16 V128:$Rn))),
6093            (FADDPv2i16p
6094              (EXTRACT_SUBREG
6095                 (FADDPv8f16 (FADDPv8f16 V128:$Rn, V128:$Rn), V128:$Rn),
6096               dsub))>;
6097def : Pat<(f16 (vecreduce_fadd (v4f16 V64:$Rn))),
6098          (FADDPv2i16p (FADDPv4f16 V64:$Rn, V64:$Rn))>;
6099}
6100def : Pat<(f32 (vecreduce_fadd (v4f32 V128:$Rn))),
6101          (FADDPv2i32p
6102            (EXTRACT_SUBREG
6103              (FADDPv4f32 V128:$Rn, V128:$Rn),
6104             dsub))>;
6105def : Pat<(f32 (vecreduce_fadd (v2f32 V64:$Rn))),
6106          (FADDPv2i32p V64:$Rn)>;
6107def : Pat<(f64 (vecreduce_fadd (v2f64 V128:$Rn))),
6108          (FADDPv2i64p V128:$Rn)>;
6109
6110def : Pat<(v2i64 (AArch64saddv V128:$Rn)),
6111          (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), (ADDPv2i64p V128:$Rn), dsub)>;
6112def : Pat<(v2i64 (AArch64uaddv V128:$Rn)),
6113          (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), (ADDPv2i64p V128:$Rn), dsub)>;
6114def : Pat<(f32 (int_aarch64_neon_faddv (v2f32 V64:$Rn))),
6115          (FADDPv2i32p V64:$Rn)>;
6116def : Pat<(f32 (int_aarch64_neon_faddv (v4f32 V128:$Rn))),
6117          (FADDPv2i32p (EXTRACT_SUBREG (FADDPv4f32 V128:$Rn, V128:$Rn), dsub))>;
6118def : Pat<(f64 (int_aarch64_neon_faddv (v2f64 V128:$Rn))),
6119          (FADDPv2i64p V128:$Rn)>;
6120def : Pat<(f32 (AArch64fmaxnmv (v2f32 V64:$Rn))),
6121          (FMAXNMPv2i32p V64:$Rn)>;
6122def : Pat<(f64 (AArch64fmaxnmv (v2f64 V128:$Rn))),
6123          (FMAXNMPv2i64p V128:$Rn)>;
6124def : Pat<(f32 (AArch64fmaxv (v2f32 V64:$Rn))),
6125          (FMAXPv2i32p V64:$Rn)>;
6126def : Pat<(f64 (AArch64fmaxv (v2f64 V128:$Rn))),
6127          (FMAXPv2i64p V128:$Rn)>;
6128def : Pat<(f32 (AArch64fminnmv (v2f32 V64:$Rn))),
6129          (FMINNMPv2i32p V64:$Rn)>;
6130def : Pat<(f64 (AArch64fminnmv (v2f64 V128:$Rn))),
6131          (FMINNMPv2i64p V128:$Rn)>;
6132def : Pat<(f32 (AArch64fminv (v2f32 V64:$Rn))),
6133          (FMINPv2i32p V64:$Rn)>;
6134def : Pat<(f64 (AArch64fminv (v2f64 V128:$Rn))),
6135          (FMINPv2i64p V128:$Rn)>;
6136
6137//----------------------------------------------------------------------------
6138// AdvSIMD INS/DUP instructions
6139//----------------------------------------------------------------------------
6140
6141def DUPv8i8gpr  : SIMDDupFromMain<0, {?,?,?,?,1}, ".8b", v8i8, V64, GPR32>;
6142def DUPv16i8gpr : SIMDDupFromMain<1, {?,?,?,?,1}, ".16b", v16i8, V128, GPR32>;
6143def DUPv4i16gpr : SIMDDupFromMain<0, {?,?,?,1,0}, ".4h", v4i16, V64, GPR32>;
6144def DUPv8i16gpr : SIMDDupFromMain<1, {?,?,?,1,0}, ".8h", v8i16, V128, GPR32>;
6145def DUPv2i32gpr : SIMDDupFromMain<0, {?,?,1,0,0}, ".2s", v2i32, V64, GPR32>;
6146def DUPv4i32gpr : SIMDDupFromMain<1, {?,?,1,0,0}, ".4s", v4i32, V128, GPR32>;
6147def DUPv2i64gpr : SIMDDupFromMain<1, {?,1,0,0,0}, ".2d", v2i64, V128, GPR64>;
6148
6149def DUPv2i64lane : SIMDDup64FromElement;
6150def DUPv2i32lane : SIMDDup32FromElement<0, ".2s", v2i32, V64>;
6151def DUPv4i32lane : SIMDDup32FromElement<1, ".4s", v4i32, V128>;
6152def DUPv4i16lane : SIMDDup16FromElement<0, ".4h", v4i16, V64>;
6153def DUPv8i16lane : SIMDDup16FromElement<1, ".8h", v8i16, V128>;
6154def DUPv8i8lane  : SIMDDup8FromElement <0, ".8b", v8i8, V64>;
6155def DUPv16i8lane : SIMDDup8FromElement <1, ".16b", v16i8, V128>;
6156
6157// DUP from a 64-bit register to a 64-bit register is just a copy
6158def : Pat<(v1i64 (AArch64dup (i64 GPR64:$Rn))),
6159          (COPY_TO_REGCLASS GPR64:$Rn, FPR64)>;
6160def : Pat<(v1f64 (AArch64dup (f64 FPR64:$Rn))),
6161          (COPY_TO_REGCLASS FPR64:$Rn, FPR64)>;
6162
6163def : Pat<(v2f32 (AArch64dup (f32 FPR32:$Rn))),
6164          (v2f32 (DUPv2i32lane
6165            (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rn, ssub),
6166            (i64 0)))>;
6167def : Pat<(v4f32 (AArch64dup (f32 FPR32:$Rn))),
6168          (v4f32 (DUPv4i32lane
6169            (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rn, ssub),
6170            (i64 0)))>;
6171def : Pat<(v2f64 (AArch64dup (f64 FPR64:$Rn))),
6172          (v2f64 (DUPv2i64lane
6173            (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$Rn, dsub),
6174            (i64 0)))>;
6175def : Pat<(v4f16 (AArch64dup (f16 FPR16:$Rn))),
6176          (v4f16 (DUPv4i16lane
6177            (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), FPR16:$Rn, hsub),
6178            (i64 0)))>;
6179def : Pat<(v4bf16 (AArch64dup (bf16 FPR16:$Rn))),
6180          (v4bf16 (DUPv4i16lane
6181            (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), FPR16:$Rn, hsub),
6182            (i64 0)))>;
6183def : Pat<(v8f16 (AArch64dup (f16 FPR16:$Rn))),
6184          (v8f16 (DUPv8i16lane
6185            (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), FPR16:$Rn, hsub),
6186            (i64 0)))>;
6187def : Pat<(v8bf16 (AArch64dup (bf16 FPR16:$Rn))),
6188          (v8bf16 (DUPv8i16lane
6189            (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), FPR16:$Rn, hsub),
6190            (i64 0)))>;
6191
6192def : Pat<(v4f16 (AArch64duplane16 (v8f16 V128:$Rn), VectorIndexH:$imm)),
6193          (DUPv4i16lane V128:$Rn, VectorIndexH:$imm)>;
6194def : Pat<(v8f16 (AArch64duplane16 (v8f16 V128:$Rn), VectorIndexH:$imm)),
6195          (DUPv8i16lane V128:$Rn, VectorIndexH:$imm)>;
6196
6197def : Pat<(v4bf16 (AArch64duplane16 (v8bf16 V128:$Rn), VectorIndexH:$imm)),
6198          (DUPv4i16lane V128:$Rn, VectorIndexH:$imm)>;
6199def : Pat<(v8bf16 (AArch64duplane16 (v8bf16 V128:$Rn), VectorIndexH:$imm)),
6200          (DUPv8i16lane V128:$Rn, VectorIndexH:$imm)>;
6201
6202def : Pat<(v2f32 (AArch64duplane32 (v4f32 V128:$Rn), VectorIndexS:$imm)),
6203          (DUPv2i32lane V128:$Rn, VectorIndexS:$imm)>;
6204def : Pat<(v4f32 (AArch64duplane32 (v4f32 V128:$Rn), VectorIndexS:$imm)),
6205         (DUPv4i32lane V128:$Rn, VectorIndexS:$imm)>;
6206def : Pat<(v2f64 (AArch64duplane64 (v2f64 V128:$Rn), VectorIndexD:$imm)),
6207          (DUPv2i64lane V128:$Rn, VectorIndexD:$imm)>;
6208
6209// If there's an (AArch64dup (vector_extract ...) ...), we can use a duplane
6210// instruction even if the types don't match: we just have to remap the lane
6211// carefully. N.b. this trick only applies to truncations.
6212def VecIndex_x2 : SDNodeXForm<imm, [{
6213  return CurDAG->getTargetConstant(2 * N->getZExtValue(), SDLoc(N), MVT::i64);
6214}]>;
6215def VecIndex_x4 : SDNodeXForm<imm, [{
6216  return CurDAG->getTargetConstant(4 * N->getZExtValue(), SDLoc(N), MVT::i64);
6217}]>;
6218def VecIndex_x8 : SDNodeXForm<imm, [{
6219  return CurDAG->getTargetConstant(8 * N->getZExtValue(), SDLoc(N), MVT::i64);
6220}]>;
6221
6222multiclass DUPWithTruncPats<ValueType ResVT, ValueType Src64VT,
6223                            ValueType Src128VT, ValueType ScalVT,
6224                            Instruction DUP, SDNodeXForm IdxXFORM> {
6225  def : Pat<(ResVT (AArch64dup (ScalVT (vector_extract (Src128VT V128:$Rn),
6226                                                     imm:$idx)))),
6227            (DUP V128:$Rn, (IdxXFORM imm:$idx))>;
6228
6229  def : Pat<(ResVT (AArch64dup (ScalVT (vector_extract (Src64VT V64:$Rn),
6230                                                     imm:$idx)))),
6231            (DUP (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), (IdxXFORM imm:$idx))>;
6232}
6233
6234defm : DUPWithTruncPats<v8i8,   v4i16, v8i16, i32, DUPv8i8lane,  VecIndex_x2>;
6235defm : DUPWithTruncPats<v8i8,   v2i32, v4i32, i32, DUPv8i8lane,  VecIndex_x4>;
6236defm : DUPWithTruncPats<v4i16,  v2i32, v4i32, i32, DUPv4i16lane, VecIndex_x2>;
6237
6238defm : DUPWithTruncPats<v16i8,  v4i16, v8i16, i32, DUPv16i8lane, VecIndex_x2>;
6239defm : DUPWithTruncPats<v16i8,  v2i32, v4i32, i32, DUPv16i8lane, VecIndex_x4>;
6240defm : DUPWithTruncPats<v8i16,  v2i32, v4i32, i32, DUPv8i16lane, VecIndex_x2>;
6241
6242multiclass DUPWithTrunci64Pats<ValueType ResVT, Instruction DUP,
6243                               SDNodeXForm IdxXFORM> {
6244  def : Pat<(ResVT (AArch64dup (i32 (trunc (extractelt (v2i64 V128:$Rn),
6245                                                         imm:$idx))))),
6246            (DUP V128:$Rn, (IdxXFORM imm:$idx))>;
6247
6248  def : Pat<(ResVT (AArch64dup (i32 (trunc (extractelt (v1i64 V64:$Rn),
6249                                                       imm:$idx))))),
6250            (DUP (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), (IdxXFORM imm:$idx))>;
6251}
6252
6253defm : DUPWithTrunci64Pats<v8i8,  DUPv8i8lane,   VecIndex_x8>;
6254defm : DUPWithTrunci64Pats<v4i16, DUPv4i16lane,  VecIndex_x4>;
6255defm : DUPWithTrunci64Pats<v2i32, DUPv2i32lane,  VecIndex_x2>;
6256
6257defm : DUPWithTrunci64Pats<v16i8, DUPv16i8lane, VecIndex_x8>;
6258defm : DUPWithTrunci64Pats<v8i16, DUPv8i16lane, VecIndex_x4>;
6259defm : DUPWithTrunci64Pats<v4i32, DUPv4i32lane, VecIndex_x2>;
6260
6261// SMOV and UMOV definitions, with some extra patterns for convenience
6262defm SMOV : SMov;
6263defm UMOV : UMov;
6264
6265def : Pat<(sext_inreg (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx), i8),
6266          (i32 (SMOVvi8to32 V128:$Rn, VectorIndexB:$idx))>;
6267def : Pat<(sext_inreg (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx), i8),
6268          (i64 (SMOVvi8to64 V128:$Rn, VectorIndexB:$idx))>;
6269def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
6270          (i32 (SMOVvi16to32 V128:$Rn, VectorIndexH:$idx))>;
6271def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
6272          (i64 (SMOVvi16to64 V128:$Rn, VectorIndexH:$idx))>;
6273def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
6274          (i32 (SMOVvi16to32 V128:$Rn, VectorIndexH:$idx))>;
6275def : Pat<(sext (i32 (vector_extract (v4i32 V128:$Rn), VectorIndexS:$idx))),
6276          (i64 (SMOVvi32to64 V128:$Rn, VectorIndexS:$idx))>;
6277
6278def : Pat<(sext_inreg (i64 (anyext (i32 (vector_extract (v16i8 V128:$Rn),
6279            VectorIndexB:$idx)))), i8),
6280          (i64 (SMOVvi8to64 V128:$Rn, VectorIndexB:$idx))>;
6281def : Pat<(sext_inreg (i64 (anyext (i32 (vector_extract (v8i16 V128:$Rn),
6282            VectorIndexH:$idx)))), i16),
6283          (i64 (SMOVvi16to64 V128:$Rn, VectorIndexH:$idx))>;
6284
6285// Extracting i8 or i16 elements will have the zero-extend transformed to
6286// an 'and' mask by type legalization since neither i8 nor i16 are legal types
6287// for AArch64. Match these patterns here since UMOV already zeroes out the high
6288// bits of the destination register.
6289def : Pat<(and (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx),
6290               (i32 0xff)),
6291          (i32 (UMOVvi8 V128:$Rn, VectorIndexB:$idx))>;
6292def : Pat<(and (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),
6293               (i32 0xffff)),
6294          (i32 (UMOVvi16 V128:$Rn, VectorIndexH:$idx))>;
6295
6296def : Pat<(i64 (and (i64 (anyext (i32 (vector_extract (v16i8 V128:$Rn),
6297            VectorIndexB:$idx)))), (i64 0xff))),
6298          (SUBREG_TO_REG (i64 0), (i32 (UMOVvi8 V128:$Rn, VectorIndexB:$idx)), sub_32)>;
6299def : Pat<(i64 (and (i64 (anyext (i32 (vector_extract (v8i16 V128:$Rn),
6300            VectorIndexH:$idx)))), (i64 0xffff))),
6301          (SUBREG_TO_REG (i64 0), (i32 (UMOVvi16 V128:$Rn, VectorIndexH:$idx)), sub_32)>;
6302
6303defm INS : SIMDIns;
6304
6305def : Pat<(v16i8 (scalar_to_vector GPR32:$Rn)),
6306          (SUBREG_TO_REG (i32 0),
6307                         (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
6308def : Pat<(v8i8 (scalar_to_vector GPR32:$Rn)),
6309          (SUBREG_TO_REG (i32 0),
6310                         (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
6311
6312// The top bits will be zero from the FMOVWSr
6313def : Pat<(v8i8 (bitconvert (i64 (zext GPR32:$Rn)))),
6314          (SUBREG_TO_REG (i32 0), (f32 (FMOVWSr GPR32:$Rn)), ssub)>;
6315
6316def : Pat<(v8i16 (scalar_to_vector GPR32:$Rn)),
6317          (SUBREG_TO_REG (i32 0),
6318                         (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
6319def : Pat<(v4i16 (scalar_to_vector GPR32:$Rn)),
6320          (SUBREG_TO_REG (i32 0),
6321                         (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
6322
6323def : Pat<(v4f16 (scalar_to_vector (f16 FPR16:$Rn))),
6324          (INSERT_SUBREG (v4f16 (IMPLICIT_DEF)), FPR16:$Rn, hsub)>;
6325def : Pat<(v8f16 (scalar_to_vector (f16 FPR16:$Rn))),
6326          (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), FPR16:$Rn, hsub)>;
6327
6328def : Pat<(v4bf16 (scalar_to_vector (bf16 FPR16:$Rn))),
6329          (INSERT_SUBREG (v4bf16 (IMPLICIT_DEF)), FPR16:$Rn, hsub)>;
6330def : Pat<(v8bf16 (scalar_to_vector (bf16 FPR16:$Rn))),
6331          (INSERT_SUBREG (v8bf16 (IMPLICIT_DEF)), FPR16:$Rn, hsub)>;
6332
6333def : Pat<(v2i32 (scalar_to_vector (i32 FPR32:$Rn))),
6334            (v2i32 (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)),
6335                                  (i32 FPR32:$Rn), ssub))>;
6336def : Pat<(v4i32 (scalar_to_vector (i32 FPR32:$Rn))),
6337            (v4i32 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
6338                                  (i32 FPR32:$Rn), ssub))>;
6339
6340def : Pat<(v2i64 (scalar_to_vector (i64 FPR64:$Rn))),
6341            (v2i64 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
6342                                  (i64 FPR64:$Rn), dsub))>;
6343
6344def : Pat<(v4f16 (scalar_to_vector (f16 FPR16:$Rn))),
6345          (INSERT_SUBREG (v4f16 (IMPLICIT_DEF)), FPR16:$Rn, hsub)>;
6346def : Pat<(v8f16 (scalar_to_vector (f16 FPR16:$Rn))),
6347          (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), FPR16:$Rn, hsub)>;
6348
6349def : Pat<(v4bf16 (scalar_to_vector (bf16 FPR16:$Rn))),
6350          (INSERT_SUBREG (v4bf16 (IMPLICIT_DEF)), FPR16:$Rn, hsub)>;
6351def : Pat<(v8bf16 (scalar_to_vector (bf16 FPR16:$Rn))),
6352          (INSERT_SUBREG (v8bf16 (IMPLICIT_DEF)), FPR16:$Rn, hsub)>;
6353
6354def : Pat<(v4f32 (scalar_to_vector (f32 FPR32:$Rn))),
6355          (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rn, ssub)>;
6356def : Pat<(v2f32 (scalar_to_vector (f32 FPR32:$Rn))),
6357          (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), FPR32:$Rn, ssub)>;
6358
6359def : Pat<(v2f64 (scalar_to_vector (f64 FPR64:$Rn))),
6360          (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$Rn, dsub)>;
6361
6362def : Pat<(v4f16 (vector_insert (v4f16 V64:$Rn),
6363            (f16 FPR16:$Rm), (i64 VectorIndexS:$imm))),
6364          (EXTRACT_SUBREG
6365            (INSvi16lane
6366              (v8f16 (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), V64:$Rn, dsub)),
6367              VectorIndexS:$imm,
6368              (v8f16 (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), FPR16:$Rm, hsub)),
6369              (i64 0)),
6370            dsub)>;
6371
6372def : Pat<(vector_insert (v8f16 V128:$Rn), (f16 fpimm0), (i64 VectorIndexH:$imm)),
6373          (INSvi16gpr V128:$Rn, VectorIndexH:$imm, WZR)>;
6374def : Pat<(vector_insert (v4f16 V64:$Rn), (f16 fpimm0), (i64 VectorIndexH:$imm)),
6375          (EXTRACT_SUBREG (INSvi16gpr (v8f16 (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), V64:$Rn, dsub)), VectorIndexH:$imm, WZR), dsub)>;
6376def : Pat<(vector_insert (v4f32 V128:$Rn), (f32 fpimm0), (i64 VectorIndexS:$imm)),
6377          (INSvi32gpr V128:$Rn, VectorIndexS:$imm, WZR)>;
6378def : Pat<(vector_insert (v2f32 V64:$Rn), (f32 fpimm0), (i64 VectorIndexS:$imm)),
6379          (EXTRACT_SUBREG (INSvi32gpr (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), V64:$Rn, dsub)), VectorIndexS:$imm, WZR), dsub)>;
6380def : Pat<(vector_insert v2f64:$Rn, (f64 fpimm0), (i64 VectorIndexD:$imm)),
6381          (INSvi64gpr V128:$Rn, VectorIndexS:$imm, XZR)>;
6382
6383def : Pat<(v8f16 (vector_insert (v8f16 V128:$Rn),
6384            (f16 FPR16:$Rm), (i64 VectorIndexH:$imm))),
6385          (INSvi16lane
6386            V128:$Rn, VectorIndexH:$imm,
6387            (v8f16 (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), FPR16:$Rm, hsub)),
6388            (i64 0))>;
6389
6390def : Pat<(v4bf16 (vector_insert (v4bf16 V64:$Rn),
6391            (bf16 FPR16:$Rm), (i64 VectorIndexS:$imm))),
6392          (EXTRACT_SUBREG
6393            (INSvi16lane
6394              (v8bf16 (INSERT_SUBREG (v8bf16 (IMPLICIT_DEF)), V64:$Rn, dsub)),
6395              VectorIndexS:$imm,
6396              (v8bf16 (INSERT_SUBREG (v8bf16 (IMPLICIT_DEF)), FPR16:$Rm, hsub)),
6397              (i64 0)),
6398            dsub)>;
6399
6400def : Pat<(v8bf16 (vector_insert (v8bf16 V128:$Rn),
6401            (bf16 FPR16:$Rm), (i64 VectorIndexH:$imm))),
6402          (INSvi16lane
6403            V128:$Rn, VectorIndexH:$imm,
6404            (v8bf16 (INSERT_SUBREG (v8bf16 (IMPLICIT_DEF)), FPR16:$Rm, hsub)),
6405            (i64 0))>;
6406
6407def : Pat<(v2f32 (vector_insert (v2f32 V64:$Rn),
6408            (f32 FPR32:$Rm), (i64 VectorIndexS:$imm))),
6409          (EXTRACT_SUBREG
6410            (INSvi32lane
6411              (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), V64:$Rn, dsub)),
6412              VectorIndexS:$imm,
6413              (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rm, ssub)),
6414              (i64 0)),
6415            dsub)>;
6416def : Pat<(v4f32 (vector_insert (v4f32 V128:$Rn),
6417            (f32 FPR32:$Rm), (i64 VectorIndexS:$imm))),
6418          (INSvi32lane
6419            V128:$Rn, VectorIndexS:$imm,
6420            (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rm, ssub)),
6421            (i64 0))>;
6422def : Pat<(v2f64 (vector_insert (v2f64 V128:$Rn),
6423            (f64 FPR64:$Rm), (i64 VectorIndexD:$imm))),
6424          (INSvi64lane
6425            V128:$Rn, VectorIndexD:$imm,
6426            (v2f64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$Rm, dsub)),
6427            (i64 0))>;
6428
6429def : Pat<(v2i32 (vector_insert (v2i32 V64:$Rn), (i32 GPR32:$Rm), (i64 VectorIndexS:$imm))),
6430          (EXTRACT_SUBREG
6431            (INSvi32gpr (v4i32 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), V64:$Rn, dsub)),
6432                        VectorIndexS:$imm, GPR32:$Rm),
6433            dsub)>;
6434def : Pat<(v4i16 (vector_insert (v4i16 V64:$Rn), (i32 GPR32:$Rm), (i64 VectorIndexH:$imm))),
6435          (EXTRACT_SUBREG
6436            (INSvi16gpr (v8i16 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), V64:$Rn, dsub)),
6437                        VectorIndexH:$imm, GPR32:$Rm),
6438            dsub)>;
6439def : Pat<(v8i8 (vector_insert (v8i8 V64:$Rn), (i32 GPR32:$Rm), (i64 VectorIndexB:$imm))),
6440          (EXTRACT_SUBREG
6441            (INSvi8gpr (v16i8 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)), V64:$Rn, dsub)),
6442                       VectorIndexB:$imm, GPR32:$Rm),
6443            dsub)>;
6444
6445// Copy an element at a constant index in one vector into a constant indexed
6446// element of another.
6447// FIXME refactor to a shared class/dev parameterized on vector type, vector
6448// index type and INS extension
6449def : Pat<(v16i8 (int_aarch64_neon_vcopy_lane
6450                   (v16i8 V128:$Vd), VectorIndexB:$idx, (v16i8 V128:$Vs),
6451                   VectorIndexB:$idx2)),
6452          (v16i8 (INSvi8lane
6453                   V128:$Vd, VectorIndexB:$idx, V128:$Vs, VectorIndexB:$idx2)
6454          )>;
6455def : Pat<(v8i16 (int_aarch64_neon_vcopy_lane
6456                   (v8i16 V128:$Vd), VectorIndexH:$idx, (v8i16 V128:$Vs),
6457                   VectorIndexH:$idx2)),
6458          (v8i16 (INSvi16lane
6459                   V128:$Vd, VectorIndexH:$idx, V128:$Vs, VectorIndexH:$idx2)
6460          )>;
6461def : Pat<(v4i32 (int_aarch64_neon_vcopy_lane
6462                   (v4i32 V128:$Vd), VectorIndexS:$idx, (v4i32 V128:$Vs),
6463                   VectorIndexS:$idx2)),
6464          (v4i32 (INSvi32lane
6465                   V128:$Vd, VectorIndexS:$idx, V128:$Vs, VectorIndexS:$idx2)
6466          )>;
6467def : Pat<(v2i64 (int_aarch64_neon_vcopy_lane
6468                   (v2i64 V128:$Vd), VectorIndexD:$idx, (v2i64 V128:$Vs),
6469                   VectorIndexD:$idx2)),
6470          (v2i64 (INSvi64lane
6471                   V128:$Vd, VectorIndexD:$idx, V128:$Vs, VectorIndexD:$idx2)
6472          )>;
6473
6474multiclass Neon_INS_elt_pattern<ValueType VT128, ValueType VT64,
6475                                ValueType VTScal, Instruction INS> {
6476  def : Pat<(VT128 (vector_insert V128:$src,
6477                        (VTScal (vector_extract (VT128 V128:$Rn), imm:$Immn)),
6478                        imm:$Immd)),
6479            (INS V128:$src, imm:$Immd, V128:$Rn, imm:$Immn)>;
6480
6481  def : Pat<(VT128 (vector_insert V128:$src,
6482                        (VTScal (vector_extract (VT64 V64:$Rn), imm:$Immn)),
6483                        imm:$Immd)),
6484            (INS V128:$src, imm:$Immd,
6485                 (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), imm:$Immn)>;
6486
6487  def : Pat<(VT64 (vector_insert V64:$src,
6488                        (VTScal (vector_extract (VT128 V128:$Rn), imm:$Immn)),
6489                        imm:$Immd)),
6490            (EXTRACT_SUBREG (INS (SUBREG_TO_REG (i64 0), V64:$src, dsub),
6491                                 imm:$Immd, V128:$Rn, imm:$Immn),
6492                            dsub)>;
6493
6494  def : Pat<(VT64 (vector_insert V64:$src,
6495                        (VTScal (vector_extract (VT64 V64:$Rn), imm:$Immn)),
6496                        imm:$Immd)),
6497            (EXTRACT_SUBREG
6498                (INS (SUBREG_TO_REG (i64 0), V64:$src, dsub), imm:$Immd,
6499                     (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), imm:$Immn),
6500                dsub)>;
6501}
6502
6503defm : Neon_INS_elt_pattern<v8f16, v4f16, f16, INSvi16lane>;
6504defm : Neon_INS_elt_pattern<v8bf16, v4bf16, bf16, INSvi16lane>;
6505defm : Neon_INS_elt_pattern<v4f32, v2f32, f32, INSvi32lane>;
6506defm : Neon_INS_elt_pattern<v2f64, v1f64, f64, INSvi64lane>;
6507
6508defm : Neon_INS_elt_pattern<v16i8, v8i8, i32, INSvi8lane>;
6509defm : Neon_INS_elt_pattern<v8i16, v4i16, i32, INSvi16lane>;
6510defm : Neon_INS_elt_pattern<v4i32, v2i32, i32, INSvi32lane>;
6511defm : Neon_INS_elt_pattern<v2i64, v1i64, i64, INSvi64lane>;
6512
6513// Insert from bitcast
6514// vector_insert(bitcast(f32 src), n, lane) -> INSvi32lane(src, lane, INSERT_SUBREG(-, n), 0)
6515def : Pat<(v4i32 (vector_insert v4i32:$src, (i32 (bitconvert (f32 FPR32:$Sn))), imm:$Immd)),
6516          (INSvi32lane V128:$src, imm:$Immd, (INSERT_SUBREG (IMPLICIT_DEF), FPR32:$Sn, ssub), 0)>;
6517def : Pat<(v2i32 (vector_insert v2i32:$src, (i32 (bitconvert (f32 FPR32:$Sn))), imm:$Immd)),
6518          (EXTRACT_SUBREG
6519            (INSvi32lane (v4i32 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), V64:$src, dsub)),
6520                         imm:$Immd, (INSERT_SUBREG (IMPLICIT_DEF), FPR32:$Sn, ssub), 0),
6521            dsub)>;
6522def : Pat<(v2i64 (vector_insert v2i64:$src, (i64 (bitconvert (f64 FPR64:$Sn))), imm:$Immd)),
6523          (INSvi64lane V128:$src, imm:$Immd, (INSERT_SUBREG (IMPLICIT_DEF), FPR64:$Sn, dsub), 0)>;
6524
6525// bitcast of an extract
6526// f32 bitcast(vector_extract(v4i32 src, lane)) -> EXTRACT_SUBREG(INSvi32lane(-, 0, src, lane))
6527def : Pat<(f32 (bitconvert (i32 (vector_extract v4i32:$src, imm:$Immd)))),
6528          (EXTRACT_SUBREG (INSvi32lane (IMPLICIT_DEF), 0, V128:$src, imm:$Immd), ssub)>;
6529def : Pat<(f32 (bitconvert (i32 (vector_extract v4i32:$src, (i64 0))))),
6530          (EXTRACT_SUBREG V128:$src, ssub)>;
6531def : Pat<(f64 (bitconvert (i64 (vector_extract v2i64:$src, imm:$Immd)))),
6532          (EXTRACT_SUBREG (INSvi64lane (IMPLICIT_DEF), 0, V128:$src, imm:$Immd), dsub)>;
6533def : Pat<(f64 (bitconvert (i64 (vector_extract v2i64:$src, (i64 0))))),
6534          (EXTRACT_SUBREG V128:$src, dsub)>;
6535
6536// Floating point vector extractions are codegen'd as either a sequence of
6537// subregister extractions, or a MOV (aka DUP here) if
6538// the lane number is anything other than zero.
6539def : Pat<(f64 (vector_extract (v2f64 V128:$Rn), (i64 0))),
6540          (f64 (EXTRACT_SUBREG V128:$Rn, dsub))>;
6541def : Pat<(f32 (vector_extract (v4f32 V128:$Rn), (i64 0))),
6542          (f32 (EXTRACT_SUBREG V128:$Rn, ssub))>;
6543def : Pat<(f16 (vector_extract (v8f16 V128:$Rn), (i64 0))),
6544          (f16 (EXTRACT_SUBREG V128:$Rn, hsub))>;
6545def : Pat<(bf16 (vector_extract (v8bf16 V128:$Rn), (i64 0))),
6546          (bf16 (EXTRACT_SUBREG V128:$Rn, hsub))>;
6547
6548
6549def : Pat<(vector_extract (v2f64 V128:$Rn), VectorIndexD:$idx),
6550          (f64 (DUPi64 V128:$Rn, VectorIndexD:$idx))>;
6551def : Pat<(vector_extract (v4f32 V128:$Rn), VectorIndexS:$idx),
6552          (f32 (DUPi32 V128:$Rn, VectorIndexS:$idx))>;
6553def : Pat<(vector_extract (v8f16 V128:$Rn), VectorIndexH:$idx),
6554          (f16 (DUPi16 V128:$Rn, VectorIndexH:$idx))>;
6555def : Pat<(vector_extract (v8bf16 V128:$Rn), VectorIndexH:$idx),
6556          (bf16 (DUPi16 V128:$Rn, VectorIndexH:$idx))>;
6557
6558// All concat_vectors operations are canonicalised to act on i64 vectors for
6559// AArch64. In the general case we need an instruction, which had just as well be
6560// INS.
6561class ConcatPat<ValueType DstTy, ValueType SrcTy>
6562  : Pat<(DstTy (concat_vectors (SrcTy V64:$Rd), V64:$Rn)),
6563        (INSvi64lane (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), 1,
6564                     (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rn, dsub), 0)>;
6565
6566def : ConcatPat<v2i64, v1i64>;
6567def : ConcatPat<v2f64, v1f64>;
6568def : ConcatPat<v4i32, v2i32>;
6569def : ConcatPat<v4f32, v2f32>;
6570def : ConcatPat<v8i16, v4i16>;
6571def : ConcatPat<v8f16, v4f16>;
6572def : ConcatPat<v8bf16, v4bf16>;
6573def : ConcatPat<v16i8, v8i8>;
6574
6575// If the high lanes are undef, though, we can just ignore them:
6576class ConcatUndefPat<ValueType DstTy, ValueType SrcTy>
6577  : Pat<(DstTy (concat_vectors (SrcTy V64:$Rn), undef)),
6578        (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rn, dsub)>;
6579
6580def : ConcatUndefPat<v2i64, v1i64>;
6581def : ConcatUndefPat<v2f64, v1f64>;
6582def : ConcatUndefPat<v4i32, v2i32>;
6583def : ConcatUndefPat<v4f32, v2f32>;
6584def : ConcatUndefPat<v8i16, v4i16>;
6585def : ConcatUndefPat<v16i8, v8i8>;
6586
6587//----------------------------------------------------------------------------
6588// AdvSIMD across lanes instructions
6589//----------------------------------------------------------------------------
6590
6591defm ADDV    : SIMDAcrossLanesBHS<0, 0b11011, "addv">;
6592defm SMAXV   : SIMDAcrossLanesBHS<0, 0b01010, "smaxv">;
6593defm SMINV   : SIMDAcrossLanesBHS<0, 0b11010, "sminv">;
6594defm UMAXV   : SIMDAcrossLanesBHS<1, 0b01010, "umaxv">;
6595defm UMINV   : SIMDAcrossLanesBHS<1, 0b11010, "uminv">;
6596defm SADDLV  : SIMDAcrossLanesHSD<0, 0b00011, "saddlv">;
6597defm UADDLV  : SIMDAcrossLanesHSD<1, 0b00011, "uaddlv">;
6598defm FMAXNMV : SIMDFPAcrossLanes<0b01100, 0, "fmaxnmv", AArch64fmaxnmv>;
6599defm FMAXV   : SIMDFPAcrossLanes<0b01111, 0, "fmaxv", AArch64fmaxv>;
6600defm FMINNMV : SIMDFPAcrossLanes<0b01100, 1, "fminnmv", AArch64fminnmv>;
6601defm FMINV   : SIMDFPAcrossLanes<0b01111, 1, "fminv", AArch64fminv>;
6602
6603multiclass SIMDAcrossLaneLongPairIntrinsic<string Opc, SDPatternOperator addlp> {
6604  // Patterns for addv(addlp(x)) ==> addlv
6605  def : Pat<(i32 (vector_extract (v8i16 (insert_subvector undef,
6606              (v4i16 (AArch64uaddv (v4i16 (addlp (v8i8 V64:$op))))),
6607              (i64 0))), (i64 0))),
6608            (EXTRACT_SUBREG (INSERT_SUBREG (v4i16 (IMPLICIT_DEF)),
6609              (!cast<Instruction>(Opc#"v8i8v") V64:$op), hsub), ssub)>;
6610  def : Pat<(i32 (vector_extract (v8i16 (AArch64uaddv (v8i16 (addlp (v16i8 V128:$op))))), (i64 0))),
6611            (EXTRACT_SUBREG (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
6612              (!cast<Instruction>(Opc#"v16i8v") V128:$op), hsub), ssub)>;
6613  def : Pat<(v4i32 (AArch64uaddv (v4i32 (addlp (v8i16 V128:$op))))),
6614            (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), (!cast<Instruction>(Opc#"v8i16v") V128:$op), ssub)>;
6615
6616  // Patterns for addp(addlp(x))) ==> addlv
6617  def : Pat<(v2i32 (AArch64uaddv (v2i32 (addlp (v4i16 V64:$op))))),
6618            (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)), (!cast<Instruction>(Opc#"v4i16v") V64:$op), ssub)>;
6619  def : Pat<(v2i64 (AArch64uaddv (v2i64 (addlp (v4i32 V128:$op))))),
6620            (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), (!cast<Instruction>(Opc#"v4i32v") V128:$op), dsub)>;
6621}
6622
6623defm : SIMDAcrossLaneLongPairIntrinsic<"UADDLV", AArch64uaddlp>;
6624defm : SIMDAcrossLaneLongPairIntrinsic<"SADDLV", AArch64saddlp>;
6625
6626// Patterns for uaddlv(uaddlp(x)) ==> uaddlv
6627def : Pat<(i64 (int_aarch64_neon_uaddlv (v4i32 (AArch64uaddlp (v8i16 V128:$op))))),
6628          (i64 (EXTRACT_SUBREG
6629            (v4i32 (SUBREG_TO_REG (i64 0), (UADDLVv8i16v V128:$op), ssub)),
6630            dsub))>;
6631
6632def : Pat<(i32 (int_aarch64_neon_uaddlv (v8i16 (AArch64uaddlp (v16i8 V128:$op))))),
6633          (i32 (EXTRACT_SUBREG
6634            (v8i16 (SUBREG_TO_REG (i64 0), (UADDLVv16i8v V128:$op), hsub)),
6635            ssub))>;
6636
6637def : Pat<(v4i32 (AArch64uaddlv (v8i16 (AArch64uaddlp (v16i8 V128:$op))))),
6638          (v4i32 (SUBREG_TO_REG (i64 0), (UADDLVv16i8v V128:$op), hsub))>;
6639
6640def : Pat<(v4i32 (AArch64uaddlv (v4i16 (AArch64uaddlp (v8i8 V64:$op))))),
6641          (v4i32 (SUBREG_TO_REG (i64 0), (UADDLVv8i8v V64:$op), hsub))>;
6642
6643def : Pat<(v4i32 (AArch64uaddlv (v8i8 V64:$Rn))),
6644          (v4i32 (SUBREG_TO_REG (i64 0), (UADDLVv8i8v V64:$Rn), hsub))>;
6645
6646def : Pat<(v4i32 (AArch64uaddlv (v4i16 V64:$Rn))),
6647          (v4i32 (SUBREG_TO_REG (i64 0), (UADDLVv4i16v V64:$Rn), ssub))>;
6648
6649def : Pat<(v4i32 (AArch64uaddlv (v16i8 V128:$Rn))),
6650          (v4i32 (SUBREG_TO_REG (i64 0), (UADDLVv16i8v V128:$Rn), hsub))>;
6651
6652def : Pat<(v4i32 (AArch64uaddlv (v8i16 V128:$Rn))),
6653          (v4i32 (SUBREG_TO_REG (i64 0), (UADDLVv8i16v V128:$Rn), ssub))>;
6654
6655// Patterns for across-vector intrinsics, that have a node equivalent, that
6656// returns a vector (with only the low lane defined) instead of a scalar.
6657// In effect, opNode is the same as (scalar_to_vector (IntNode)).
6658multiclass SIMDAcrossLanesIntrinsic<string baseOpc,
6659                                    SDPatternOperator opNode> {
6660// If a lane instruction caught the vector_extract around opNode, we can
6661// directly match the latter to the instruction.
6662def : Pat<(v8i8 (opNode V64:$Rn)),
6663          (INSERT_SUBREG (v8i8 (IMPLICIT_DEF)),
6664           (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub)>;
6665def : Pat<(v16i8 (opNode V128:$Rn)),
6666          (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
6667           (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub)>;
6668def : Pat<(v4i16 (opNode V64:$Rn)),
6669          (INSERT_SUBREG (v4i16 (IMPLICIT_DEF)),
6670           (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub)>;
6671def : Pat<(v8i16 (opNode V128:$Rn)),
6672          (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
6673           (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub)>;
6674def : Pat<(v4i32 (opNode V128:$Rn)),
6675          (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
6676           (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), ssub)>;
6677
6678
6679// If none did, fallback to the explicit patterns, consuming the vector_extract.
6680def : Pat<(i32 (vector_extract (insert_subvector undef, (v8i8 (opNode V64:$Rn)),
6681            (i64 0)), (i64 0))),
6682          (EXTRACT_SUBREG (INSERT_SUBREG (v8i8 (IMPLICIT_DEF)),
6683            (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn),
6684            bsub), ssub)>;
6685def : Pat<(i32 (vector_extract (v16i8 (opNode V128:$Rn)), (i64 0))),
6686          (EXTRACT_SUBREG (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
6687            (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn),
6688            bsub), ssub)>;
6689def : Pat<(i32 (vector_extract (insert_subvector undef,
6690            (v4i16 (opNode V64:$Rn)), (i64 0)), (i64 0))),
6691          (EXTRACT_SUBREG (INSERT_SUBREG (v4i16 (IMPLICIT_DEF)),
6692            (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn),
6693            hsub), ssub)>;
6694def : Pat<(i32 (vector_extract (v8i16 (opNode V128:$Rn)), (i64 0))),
6695          (EXTRACT_SUBREG (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
6696            (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn),
6697            hsub), ssub)>;
6698def : Pat<(i32 (vector_extract (v4i32 (opNode V128:$Rn)), (i64 0))),
6699          (EXTRACT_SUBREG (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
6700            (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn),
6701            ssub), ssub)>;
6702
6703}
6704
6705multiclass SIMDAcrossLanesSignedIntrinsic<string baseOpc,
6706                                          SDPatternOperator opNode>
6707    : SIMDAcrossLanesIntrinsic<baseOpc, opNode> {
6708// If there is a sign extension after this intrinsic, consume it as smov already
6709// performed it
6710def : Pat<(i32 (sext_inreg (i32 (vector_extract (insert_subvector undef,
6711            (opNode (v8i8 V64:$Rn)), (i64 0)), (i64 0))), i8)),
6712          (i32 (SMOVvi8to32
6713            (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
6714              (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
6715            (i64 0)))>;
6716def : Pat<(i32 (sext_inreg (i32 (vector_extract
6717            (opNode (v16i8 V128:$Rn)), (i64 0))), i8)),
6718          (i32 (SMOVvi8to32
6719            (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
6720             (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
6721            (i64 0)))>;
6722def : Pat<(i32 (sext_inreg (i32 (vector_extract (insert_subvector undef,
6723            (opNode (v4i16 V64:$Rn)), (i64 0)), (i64 0))), i16)),
6724          (i32 (SMOVvi16to32
6725           (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
6726            (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
6727           (i64 0)))>;
6728def : Pat<(i32 (sext_inreg (i32 (vector_extract
6729            (opNode (v8i16 V128:$Rn)), (i64 0))), i16)),
6730          (i32 (SMOVvi16to32
6731            (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
6732             (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
6733            (i64 0)))>;
6734}
6735
6736multiclass SIMDAcrossLanesUnsignedIntrinsic<string baseOpc,
6737                                            SDPatternOperator opNode>
6738    : SIMDAcrossLanesIntrinsic<baseOpc, opNode> {
6739// If there is a masking operation keeping only what has been actually
6740// generated, consume it.
6741def : Pat<(i32 (and (i32 (vector_extract (insert_subvector undef,
6742            (opNode (v8i8 V64:$Rn)), (i64 0)), (i64 0))), maski8_or_more)),
6743      (i32 (EXTRACT_SUBREG
6744        (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
6745          (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
6746        ssub))>;
6747def : Pat<(i32 (and (i32 (vector_extract (opNode (v16i8 V128:$Rn)), (i64 0))),
6748            maski8_or_more)),
6749        (i32 (EXTRACT_SUBREG
6750          (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
6751            (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
6752          ssub))>;
6753def : Pat<(i32 (and (i32 (vector_extract (insert_subvector undef,
6754            (opNode (v4i16 V64:$Rn)), (i64 0)), (i64 0))), maski16_or_more)),
6755          (i32 (EXTRACT_SUBREG
6756            (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
6757              (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
6758            ssub))>;
6759def : Pat<(i32 (and (i32 (vector_extract (opNode (v8i16 V128:$Rn)), (i64 0))),
6760            maski16_or_more)),
6761        (i32 (EXTRACT_SUBREG
6762          (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
6763            (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
6764          ssub))>;
6765}
6766
6767// For vecreduce_add, used by GlobalISel not SDAG
6768def : Pat<(i8 (vecreduce_add (v8i8 V64:$Rn))),
6769          (i8 (ADDVv8i8v V64:$Rn))>;
6770def : Pat<(i8 (vecreduce_add (v16i8 V128:$Rn))),
6771          (i8 (ADDVv16i8v V128:$Rn))>;
6772def : Pat<(i16 (vecreduce_add (v4i16 V64:$Rn))),
6773          (i16 (ADDVv4i16v V64:$Rn))>;
6774def : Pat<(i16 (vecreduce_add (v8i16 V128:$Rn))),
6775          (i16 (ADDVv8i16v V128:$Rn))>;
6776def : Pat<(i32 (vecreduce_add (v2i32 V64:$Rn))),
6777          (i32 (EXTRACT_SUBREG (ADDPv2i32 V64:$Rn, V64:$Rn), ssub))>;
6778def : Pat<(i32 (vecreduce_add (v4i32 V128:$Rn))),
6779          (i32 (ADDVv4i32v V128:$Rn))>;
6780def : Pat<(i64 (vecreduce_add (v2i64 V128:$Rn))),
6781          (i64 (ADDPv2i64p V128:$Rn))>;
6782
6783defm : SIMDAcrossLanesSignedIntrinsic<"ADDV",  AArch64saddv>;
6784// vaddv_[su]32 is special; -> ADDP Vd.2S,Vn.2S,Vm.2S; return Vd.s[0];Vn==Vm
6785def : Pat<(v2i32 (AArch64saddv (v2i32 V64:$Rn))),
6786          (ADDPv2i32 V64:$Rn, V64:$Rn)>;
6787
6788defm : SIMDAcrossLanesUnsignedIntrinsic<"ADDV", AArch64uaddv>;
6789// vaddv_[su]32 is special; -> ADDP Vd.2S,Vn.2S,Vm.2S; return Vd.s[0];Vn==Vm
6790def : Pat<(v2i32 (AArch64uaddv (v2i32 V64:$Rn))),
6791          (ADDPv2i32 V64:$Rn, V64:$Rn)>;
6792
6793defm : SIMDAcrossLanesSignedIntrinsic<"SMAXV", AArch64smaxv>;
6794def : Pat<(v2i32 (AArch64smaxv (v2i32 V64:$Rn))),
6795          (SMAXPv2i32 V64:$Rn, V64:$Rn)>;
6796
6797defm : SIMDAcrossLanesSignedIntrinsic<"SMINV", AArch64sminv>;
6798def : Pat<(v2i32 (AArch64sminv (v2i32 V64:$Rn))),
6799          (SMINPv2i32 V64:$Rn, V64:$Rn)>;
6800
6801defm : SIMDAcrossLanesUnsignedIntrinsic<"UMAXV", AArch64umaxv>;
6802def : Pat<(v2i32 (AArch64umaxv (v2i32 V64:$Rn))),
6803          (UMAXPv2i32 V64:$Rn, V64:$Rn)>;
6804
6805defm : SIMDAcrossLanesUnsignedIntrinsic<"UMINV", AArch64uminv>;
6806def : Pat<(v2i32 (AArch64uminv (v2i32 V64:$Rn))),
6807          (UMINPv2i32 V64:$Rn, V64:$Rn)>;
6808
6809// For vecreduce_{opc} used by GlobalISel, not SDAG at the moment
6810// because GlobalISel allows us to specify the return register to be a FPR
6811multiclass SIMDAcrossLanesVecReductionIntrinsic<string baseOpc,
6812                                               SDPatternOperator opNode> {
6813def : Pat<(i8 (opNode (v8i8 FPR64:$Rn))),
6814          (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) FPR64:$Rn)>;
6815
6816def : Pat<(i8 (opNode (v16i8 FPR128:$Rn))),
6817          (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) FPR128:$Rn)>;
6818
6819def : Pat<(i16 (opNode (v4i16 FPR64:$Rn))),
6820          (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) FPR64:$Rn)>;
6821
6822def : Pat<(i16 (opNode (v8i16 FPR128:$Rn))),
6823          (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) FPR128:$Rn)>;
6824
6825def : Pat<(i32 (opNode (v4i32 V128:$Rn))),
6826          (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn)>;
6827}
6828
6829// For v2i32 source type, the pairwise instruction can be used instead
6830defm : SIMDAcrossLanesVecReductionIntrinsic<"UMINV", vecreduce_umin>;
6831def : Pat<(i32 (vecreduce_umin (v2i32 V64:$Rn))),
6832          (i32 (EXTRACT_SUBREG (UMINPv2i32 V64:$Rn, V64:$Rn), ssub))>;
6833
6834defm : SIMDAcrossLanesVecReductionIntrinsic<"UMAXV", vecreduce_umax>;
6835def : Pat<(i32 (vecreduce_umax (v2i32 V64:$Rn))),
6836          (i32 (EXTRACT_SUBREG (UMAXPv2i32 V64:$Rn, V64:$Rn), ssub))>;
6837
6838defm : SIMDAcrossLanesVecReductionIntrinsic<"SMINV", vecreduce_smin>;
6839def : Pat<(i32 (vecreduce_smin (v2i32 V64:$Rn))),
6840          (i32 (EXTRACT_SUBREG (SMINPv2i32 V64:$Rn, V64:$Rn), ssub))>;
6841
6842defm : SIMDAcrossLanesVecReductionIntrinsic<"SMAXV", vecreduce_smax>;
6843def : Pat<(i32 (vecreduce_smax (v2i32 V64:$Rn))),
6844          (i32 (EXTRACT_SUBREG (SMAXPv2i32 V64:$Rn, V64:$Rn), ssub))>;
6845
6846multiclass SIMDAcrossLanesSignedLongIntrinsic<string baseOpc, Intrinsic intOp> {
6847  def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
6848        (i32 (SMOVvi16to32
6849          (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
6850            (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), hsub),
6851          (i64 0)))>;
6852def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
6853        (i32 (SMOVvi16to32
6854          (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
6855           (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), hsub),
6856          (i64 0)))>;
6857
6858def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
6859          (i32 (EXTRACT_SUBREG
6860           (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
6861            (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), ssub),
6862           ssub))>;
6863def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
6864        (i32 (EXTRACT_SUBREG
6865          (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
6866           (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), ssub),
6867          ssub))>;
6868
6869def : Pat<(i64 (intOp (v4i32 V128:$Rn))),
6870        (i64 (EXTRACT_SUBREG
6871          (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
6872           (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), dsub),
6873          dsub))>;
6874}
6875
6876multiclass SIMDAcrossLanesUnsignedLongIntrinsic<string baseOpc,
6877                                                Intrinsic intOp> {
6878  def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
6879        (i32 (EXTRACT_SUBREG
6880          (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
6881            (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), hsub),
6882          ssub))>;
6883def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
6884        (i32 (EXTRACT_SUBREG
6885          (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
6886            (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), hsub),
6887          ssub))>;
6888
6889def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
6890          (i32 (EXTRACT_SUBREG
6891            (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
6892              (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), ssub),
6893            ssub))>;
6894def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
6895        (i32 (EXTRACT_SUBREG
6896          (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
6897            (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), ssub),
6898          ssub))>;
6899
6900def : Pat<(i64 (intOp (v4i32 V128:$Rn))),
6901        (i64 (EXTRACT_SUBREG
6902          (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
6903            (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), dsub),
6904          dsub))>;
6905}
6906
6907defm : SIMDAcrossLanesSignedLongIntrinsic<"SADDLV", int_aarch64_neon_saddlv>;
6908defm : SIMDAcrossLanesUnsignedLongIntrinsic<"UADDLV", int_aarch64_neon_uaddlv>;
6909
6910// The vaddlv_s32 intrinsic gets mapped to SADDLP.
6911def : Pat<(i64 (int_aarch64_neon_saddlv (v2i32 V64:$Rn))),
6912          (i64 (EXTRACT_SUBREG
6913            (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
6914              (SADDLPv2i32_v1i64 V64:$Rn), dsub),
6915            dsub))>;
6916// The vaddlv_u32 intrinsic gets mapped to UADDLP.
6917def : Pat<(i64 (int_aarch64_neon_uaddlv (v2i32 V64:$Rn))),
6918          (i64 (EXTRACT_SUBREG
6919            (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
6920              (UADDLPv2i32_v1i64 V64:$Rn), dsub),
6921            dsub))>;
6922
6923//------------------------------------------------------------------------------
6924// AdvSIMD modified immediate instructions
6925//------------------------------------------------------------------------------
6926
6927// AdvSIMD BIC
6928defm BIC : SIMDModifiedImmVectorShiftTied<1, 0b11, 0b01, "bic", AArch64bici>;
6929// AdvSIMD ORR
6930defm ORR : SIMDModifiedImmVectorShiftTied<0, 0b11, 0b01, "orr", AArch64orri>;
6931
6932def : InstAlias<"bic $Vd.4h, $imm", (BICv4i16 V64:$Vd,  imm0_255:$imm, 0)>;
6933def : InstAlias<"bic $Vd.8h, $imm", (BICv8i16 V128:$Vd, imm0_255:$imm, 0)>;
6934def : InstAlias<"bic $Vd.2s, $imm", (BICv2i32 V64:$Vd,  imm0_255:$imm, 0)>;
6935def : InstAlias<"bic $Vd.4s, $imm", (BICv4i32 V128:$Vd, imm0_255:$imm, 0)>;
6936
6937def : InstAlias<"bic.4h $Vd, $imm", (BICv4i16 V64:$Vd,  imm0_255:$imm, 0)>;
6938def : InstAlias<"bic.8h $Vd, $imm", (BICv8i16 V128:$Vd, imm0_255:$imm, 0)>;
6939def : InstAlias<"bic.2s $Vd, $imm", (BICv2i32 V64:$Vd,  imm0_255:$imm, 0)>;
6940def : InstAlias<"bic.4s $Vd, $imm", (BICv4i32 V128:$Vd, imm0_255:$imm, 0)>;
6941
6942def : InstAlias<"orr $Vd.4h, $imm", (ORRv4i16 V64:$Vd,  imm0_255:$imm, 0)>;
6943def : InstAlias<"orr $Vd.8h, $imm", (ORRv8i16 V128:$Vd, imm0_255:$imm, 0)>;
6944def : InstAlias<"orr $Vd.2s, $imm", (ORRv2i32 V64:$Vd,  imm0_255:$imm, 0)>;
6945def : InstAlias<"orr $Vd.4s, $imm", (ORRv4i32 V128:$Vd, imm0_255:$imm, 0)>;
6946
6947def : InstAlias<"orr.4h $Vd, $imm", (ORRv4i16 V64:$Vd,  imm0_255:$imm, 0)>;
6948def : InstAlias<"orr.8h $Vd, $imm", (ORRv8i16 V128:$Vd, imm0_255:$imm, 0)>;
6949def : InstAlias<"orr.2s $Vd, $imm", (ORRv2i32 V64:$Vd,  imm0_255:$imm, 0)>;
6950def : InstAlias<"orr.4s $Vd, $imm", (ORRv4i32 V128:$Vd, imm0_255:$imm, 0)>;
6951
6952// AdvSIMD FMOV
6953def FMOVv2f64_ns : SIMDModifiedImmVectorNoShift<1, 1, 0, 0b1111, V128, fpimm8,
6954                                              "fmov", ".2d",
6955                       [(set (v2f64 V128:$Rd), (AArch64fmov imm0_255:$imm8))]>;
6956def FMOVv2f32_ns : SIMDModifiedImmVectorNoShift<0, 0, 0, 0b1111, V64,  fpimm8,
6957                                              "fmov", ".2s",
6958                       [(set (v2f32 V64:$Rd), (AArch64fmov imm0_255:$imm8))]>;
6959def FMOVv4f32_ns : SIMDModifiedImmVectorNoShift<1, 0, 0, 0b1111, V128, fpimm8,
6960                                              "fmov", ".4s",
6961                       [(set (v4f32 V128:$Rd), (AArch64fmov imm0_255:$imm8))]>;
6962let Predicates = [HasNEON, HasFullFP16] in {
6963def FMOVv4f16_ns : SIMDModifiedImmVectorNoShift<0, 0, 1, 0b1111, V64,  fpimm8,
6964                                              "fmov", ".4h",
6965                       [(set (v4f16 V64:$Rd), (AArch64fmov imm0_255:$imm8))]>;
6966def FMOVv8f16_ns : SIMDModifiedImmVectorNoShift<1, 0, 1, 0b1111, V128, fpimm8,
6967                                              "fmov", ".8h",
6968                       [(set (v8f16 V128:$Rd), (AArch64fmov imm0_255:$imm8))]>;
6969} // Predicates = [HasNEON, HasFullFP16]
6970
6971// AdvSIMD MOVI
6972
6973// EDIT byte mask: scalar
6974let isReMaterializable = 1, isAsCheapAsAMove = 1 in
6975def MOVID      : SIMDModifiedImmScalarNoShift<0, 1, 0b1110, "movi",
6976                    [(set FPR64:$Rd, simdimmtype10:$imm8)]>;
6977// The movi_edit node has the immediate value already encoded, so we use
6978// a plain imm0_255 here.
6979def : Pat<(f64 (AArch64movi_edit imm0_255:$shift)),
6980          (MOVID imm0_255:$shift)>;
6981
6982// EDIT byte mask: 2d
6983
6984// The movi_edit node has the immediate value already encoded, so we use
6985// a plain imm0_255 in the pattern
6986let isReMaterializable = 1, isAsCheapAsAMove = 1 in
6987def MOVIv2d_ns   : SIMDModifiedImmVectorNoShift<1, 1, 0, 0b1110, V128,
6988                                                simdimmtype10,
6989                                                "movi", ".2d",
6990                   [(set (v2i64 V128:$Rd), (AArch64movi_edit imm0_255:$imm8))]>;
6991
6992def : Pat<(v2i64 immAllZerosV), (MOVIv2d_ns (i32 0))>;
6993def : Pat<(v4i32 immAllZerosV), (MOVIv2d_ns (i32 0))>;
6994def : Pat<(v8i16 immAllZerosV), (MOVIv2d_ns (i32 0))>;
6995def : Pat<(v16i8 immAllZerosV), (MOVIv2d_ns (i32 0))>;
6996def : Pat<(v2f64 immAllZerosV), (MOVIv2d_ns (i32 0))>;
6997def : Pat<(v4f32 immAllZerosV), (MOVIv2d_ns (i32 0))>;
6998def : Pat<(v8f16 immAllZerosV), (MOVIv2d_ns (i32 0))>;
6999def : Pat<(v8bf16 immAllZerosV), (MOVIv2d_ns (i32 0))>;
7000
7001def : Pat<(v2i64 immAllOnesV), (MOVIv2d_ns (i32 255))>;
7002def : Pat<(v4i32 immAllOnesV), (MOVIv2d_ns (i32 255))>;
7003def : Pat<(v8i16 immAllOnesV), (MOVIv2d_ns (i32 255))>;
7004def : Pat<(v16i8 immAllOnesV), (MOVIv2d_ns (i32 255))>;
7005
7006// Set 64-bit vectors to all 0/1 by extracting from a 128-bit register as the
7007// extract is free and this gives better MachineCSE results.
7008def : Pat<(v1i64 immAllZerosV), (EXTRACT_SUBREG (MOVIv2d_ns (i32 0)), dsub)>;
7009def : Pat<(v2i32 immAllZerosV), (EXTRACT_SUBREG (MOVIv2d_ns (i32 0)), dsub)>;
7010def : Pat<(v4i16 immAllZerosV), (EXTRACT_SUBREG (MOVIv2d_ns (i32 0)), dsub)>;
7011def : Pat<(v8i8  immAllZerosV), (EXTRACT_SUBREG (MOVIv2d_ns (i32 0)), dsub)>;
7012def : Pat<(v1f64 immAllZerosV), (MOVID (i32 0))>;
7013def : Pat<(v2f32 immAllZerosV), (MOVID (i32 0))>;
7014def : Pat<(v4f16 immAllZerosV), (MOVID (i32 0))>;
7015def : Pat<(v4bf16 immAllZerosV), (MOVID (i32 0))>;
7016
7017def : Pat<(v1i64 immAllOnesV), (EXTRACT_SUBREG (MOVIv2d_ns (i32 255)), dsub)>;
7018def : Pat<(v2i32 immAllOnesV), (EXTRACT_SUBREG (MOVIv2d_ns (i32 255)), dsub)>;
7019def : Pat<(v4i16 immAllOnesV), (EXTRACT_SUBREG (MOVIv2d_ns (i32 255)), dsub)>;
7020def : Pat<(v8i8  immAllOnesV), (EXTRACT_SUBREG (MOVIv2d_ns (i32 255)), dsub)>;
7021
7022// EDIT per word & halfword: 2s, 4h, 4s, & 8h
7023let isReMaterializable = 1, isAsCheapAsAMove = 1 in
7024defm MOVI      : SIMDModifiedImmVectorShift<0, 0b10, 0b00, "movi">;
7025
7026let Predicates = [HasNEON] in {
7027  // Using the MOVI to materialize fp constants.
7028  def : Pat<(f32 fpimm32SIMDModImmType4:$in),
7029            (EXTRACT_SUBREG (MOVIv2i32 (fpimm32SIMDModImmType4XForm f32:$in),
7030                                       (i32 24)),
7031                            ssub)>;
7032}
7033
7034def : InstAlias<"movi $Vd.4h, $imm", (MOVIv4i16 V64:$Vd,  imm0_255:$imm, 0), 0>;
7035def : InstAlias<"movi $Vd.8h, $imm", (MOVIv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
7036def : InstAlias<"movi $Vd.2s, $imm", (MOVIv2i32 V64:$Vd,  imm0_255:$imm, 0), 0>;
7037def : InstAlias<"movi $Vd.4s, $imm", (MOVIv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
7038
7039def : InstAlias<"movi.4h $Vd, $imm", (MOVIv4i16 V64:$Vd,  imm0_255:$imm, 0), 0>;
7040def : InstAlias<"movi.8h $Vd, $imm", (MOVIv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
7041def : InstAlias<"movi.2s $Vd, $imm", (MOVIv2i32 V64:$Vd,  imm0_255:$imm, 0), 0>;
7042def : InstAlias<"movi.4s $Vd, $imm", (MOVIv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
7043
7044def : Pat<(v2i32 (AArch64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
7045          (MOVIv2i32 imm0_255:$imm8, imm:$shift)>;
7046def : Pat<(v4i32 (AArch64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
7047          (MOVIv4i32 imm0_255:$imm8, imm:$shift)>;
7048def : Pat<(v4i16 (AArch64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
7049          (MOVIv4i16 imm0_255:$imm8, imm:$shift)>;
7050def : Pat<(v8i16 (AArch64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
7051          (MOVIv8i16 imm0_255:$imm8, imm:$shift)>;
7052
7053let isReMaterializable = 1, isAsCheapAsAMove = 1 in {
7054// EDIT per word: 2s & 4s with MSL shifter
7055def MOVIv2s_msl  : SIMDModifiedImmMoveMSL<0, 0, {1,1,0,?}, V64, "movi", ".2s",
7056                      [(set (v2i32 V64:$Rd),
7057                            (AArch64movi_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
7058def MOVIv4s_msl  : SIMDModifiedImmMoveMSL<1, 0, {1,1,0,?}, V128, "movi", ".4s",
7059                      [(set (v4i32 V128:$Rd),
7060                            (AArch64movi_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
7061
7062// Per byte: 8b & 16b
7063def MOVIv8b_ns   : SIMDModifiedImmVectorNoShift<0, 0, 0, 0b1110, V64,  imm0_255,
7064                                                 "movi", ".8b",
7065                       [(set (v8i8 V64:$Rd), (AArch64movi imm0_255:$imm8))]>;
7066
7067def MOVIv16b_ns  : SIMDModifiedImmVectorNoShift<1, 0, 0, 0b1110, V128, imm0_255,
7068                                                 "movi", ".16b",
7069                       [(set (v16i8 V128:$Rd), (AArch64movi imm0_255:$imm8))]>;
7070}
7071
7072// AdvSIMD MVNI
7073
7074// EDIT per word & halfword: 2s, 4h, 4s, & 8h
7075let isReMaterializable = 1, isAsCheapAsAMove = 1 in
7076defm MVNI      : SIMDModifiedImmVectorShift<1, 0b10, 0b00, "mvni">;
7077
7078def : InstAlias<"mvni $Vd.4h, $imm", (MVNIv4i16 V64:$Vd,  imm0_255:$imm, 0), 0>;
7079def : InstAlias<"mvni $Vd.8h, $imm", (MVNIv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
7080def : InstAlias<"mvni $Vd.2s, $imm", (MVNIv2i32 V64:$Vd,  imm0_255:$imm, 0), 0>;
7081def : InstAlias<"mvni $Vd.4s, $imm", (MVNIv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
7082
7083def : InstAlias<"mvni.4h $Vd, $imm", (MVNIv4i16 V64:$Vd,  imm0_255:$imm, 0), 0>;
7084def : InstAlias<"mvni.8h $Vd, $imm", (MVNIv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
7085def : InstAlias<"mvni.2s $Vd, $imm", (MVNIv2i32 V64:$Vd,  imm0_255:$imm, 0), 0>;
7086def : InstAlias<"mvni.4s $Vd, $imm", (MVNIv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
7087
7088def : Pat<(v2i32 (AArch64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
7089          (MVNIv2i32 imm0_255:$imm8, imm:$shift)>;
7090def : Pat<(v4i32 (AArch64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
7091          (MVNIv4i32 imm0_255:$imm8, imm:$shift)>;
7092def : Pat<(v4i16 (AArch64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
7093          (MVNIv4i16 imm0_255:$imm8, imm:$shift)>;
7094def : Pat<(v8i16 (AArch64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
7095          (MVNIv8i16 imm0_255:$imm8, imm:$shift)>;
7096
7097// EDIT per word: 2s & 4s with MSL shifter
7098let isReMaterializable = 1, isAsCheapAsAMove = 1 in {
7099def MVNIv2s_msl   : SIMDModifiedImmMoveMSL<0, 1, {1,1,0,?}, V64, "mvni", ".2s",
7100                      [(set (v2i32 V64:$Rd),
7101                            (AArch64mvni_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
7102def MVNIv4s_msl   : SIMDModifiedImmMoveMSL<1, 1, {1,1,0,?}, V128, "mvni", ".4s",
7103                      [(set (v4i32 V128:$Rd),
7104                            (AArch64mvni_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
7105}
7106
7107//----------------------------------------------------------------------------
7108// AdvSIMD indexed element
7109//----------------------------------------------------------------------------
7110
7111let hasSideEffects = 0 in {
7112  defm FMLA  : SIMDFPIndexedTied<0, 0b0001, "fmla">;
7113  defm FMLS  : SIMDFPIndexedTied<0, 0b0101, "fmls">;
7114}
7115
7116// NOTE: Operands are reordered in the FMLA/FMLS PatFrags because the
7117// instruction expects the addend first, while the intrinsic expects it last.
7118
7119// On the other hand, there are quite a few valid combinatorial options due to
7120// the commutativity of multiplication and the fact that (-x) * y = x * (-y).
7121defm : SIMDFPIndexedTiedPatterns<"FMLA",
7122           TriOpFrag<(any_fma node:$RHS, node:$MHS, node:$LHS)>>;
7123defm : SIMDFPIndexedTiedPatterns<"FMLA",
7124           TriOpFrag<(any_fma node:$MHS, node:$RHS, node:$LHS)>>;
7125
7126defm : SIMDFPIndexedTiedPatterns<"FMLS",
7127           TriOpFrag<(any_fma node:$MHS, (fneg node:$RHS), node:$LHS)> >;
7128defm : SIMDFPIndexedTiedPatterns<"FMLS",
7129           TriOpFrag<(any_fma node:$RHS, (fneg node:$MHS), node:$LHS)> >;
7130defm : SIMDFPIndexedTiedPatterns<"FMLS",
7131           TriOpFrag<(any_fma (fneg node:$RHS), node:$MHS, node:$LHS)> >;
7132defm : SIMDFPIndexedTiedPatterns<"FMLS",
7133           TriOpFrag<(any_fma (fneg node:$MHS), node:$RHS, node:$LHS)> >;
7134
7135multiclass FMLSIndexedAfterNegPatterns<SDPatternOperator OpNode> {
7136  // 3 variants for the .2s version: DUPLANE from 128-bit, DUPLANE from 64-bit
7137  // and DUP scalar.
7138  def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
7139                           (AArch64duplane32 (v4f32 (fneg V128:$Rm)),
7140                                           VectorIndexS:$idx))),
7141            (FMLSv2i32_indexed V64:$Rd, V64:$Rn, V128:$Rm, VectorIndexS:$idx)>;
7142  def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
7143                           (v2f32 (AArch64duplane32
7144                                      (v4f32 (insert_subvector undef,
7145                                                 (v2f32 (fneg V64:$Rm)),
7146                                                 (i64 0))),
7147                                      VectorIndexS:$idx)))),
7148            (FMLSv2i32_indexed V64:$Rd, V64:$Rn,
7149                               (SUBREG_TO_REG (i32 0), V64:$Rm, dsub),
7150                               VectorIndexS:$idx)>;
7151  def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
7152                           (AArch64dup (f32 (fneg FPR32Op:$Rm))))),
7153            (FMLSv2i32_indexed V64:$Rd, V64:$Rn,
7154                (SUBREG_TO_REG (i32 0), FPR32Op:$Rm, ssub), (i64 0))>;
7155
7156  // 3 variants for the .4s version: DUPLANE from 128-bit, DUPLANE from 64-bit
7157  // and DUP scalar.
7158  def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
7159                           (AArch64duplane32 (v4f32 (fneg V128:$Rm)),
7160                                           VectorIndexS:$idx))),
7161            (FMLSv4i32_indexed V128:$Rd, V128:$Rn, V128:$Rm,
7162                               VectorIndexS:$idx)>;
7163  def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
7164                           (v4f32 (AArch64duplane32
7165                                      (v4f32 (insert_subvector undef,
7166                                                 (v2f32 (fneg V64:$Rm)),
7167                                                 (i64 0))),
7168                                      VectorIndexS:$idx)))),
7169            (FMLSv4i32_indexed V128:$Rd, V128:$Rn,
7170                               (SUBREG_TO_REG (i32 0), V64:$Rm, dsub),
7171                               VectorIndexS:$idx)>;
7172  def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
7173                           (AArch64dup (f32 (fneg FPR32Op:$Rm))))),
7174            (FMLSv4i32_indexed V128:$Rd, V128:$Rn,
7175                (SUBREG_TO_REG (i32 0), FPR32Op:$Rm, ssub), (i64 0))>;
7176
7177  // 2 variants for the .2d version: DUPLANE from 128-bit, and DUP scalar
7178  // (DUPLANE from 64-bit would be trivial).
7179  def : Pat<(v2f64 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn),
7180                           (AArch64duplane64 (v2f64 (fneg V128:$Rm)),
7181                                           VectorIndexD:$idx))),
7182            (FMLSv2i64_indexed
7183                V128:$Rd, V128:$Rn, V128:$Rm, VectorIndexS:$idx)>;
7184  def : Pat<(v2f64 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn),
7185                           (AArch64dup (f64 (fneg FPR64Op:$Rm))))),
7186            (FMLSv2i64_indexed V128:$Rd, V128:$Rn,
7187                (SUBREG_TO_REG (i32 0), FPR64Op:$Rm, dsub), (i64 0))>;
7188
7189  // 2 variants for 32-bit scalar version: extract from .2s or from .4s
7190  def : Pat<(f32 (OpNode (f32 FPR32:$Rd), (f32 FPR32:$Rn),
7191                         (vector_extract (v4f32 (fneg V128:$Rm)),
7192                                         VectorIndexS:$idx))),
7193            (FMLSv1i32_indexed FPR32:$Rd, FPR32:$Rn,
7194                V128:$Rm, VectorIndexS:$idx)>;
7195  def : Pat<(f32 (OpNode (f32 FPR32:$Rd), (f32 FPR32:$Rn),
7196                         (vector_extract (v4f32 (insert_subvector undef,
7197                                                    (v2f32 (fneg V64:$Rm)),
7198                                                    (i64 0))),
7199                                         VectorIndexS:$idx))),
7200            (FMLSv1i32_indexed FPR32:$Rd, FPR32:$Rn,
7201                (SUBREG_TO_REG (i32 0), V64:$Rm, dsub), VectorIndexS:$idx)>;
7202
7203  // 1 variant for 64-bit scalar version: extract from .1d or from .2d
7204  def : Pat<(f64 (OpNode (f64 FPR64:$Rd), (f64 FPR64:$Rn),
7205                         (vector_extract (v2f64 (fneg V128:$Rm)),
7206                                         VectorIndexS:$idx))),
7207            (FMLSv1i64_indexed FPR64:$Rd, FPR64:$Rn,
7208                V128:$Rm, VectorIndexS:$idx)>;
7209}
7210
7211defm : FMLSIndexedAfterNegPatterns<
7212           TriOpFrag<(any_fma node:$RHS, node:$MHS, node:$LHS)> >;
7213defm : FMLSIndexedAfterNegPatterns<
7214           TriOpFrag<(any_fma node:$MHS, node:$RHS, node:$LHS)> >;
7215
7216defm FMULX : SIMDFPIndexed<1, 0b1001, "fmulx", int_aarch64_neon_fmulx>;
7217defm FMUL  : SIMDFPIndexed<0, 0b1001, "fmul", any_fmul>;
7218
7219def : Pat<(v2f32 (any_fmul V64:$Rn, (AArch64dup (f32 FPR32:$Rm)))),
7220          (FMULv2i32_indexed V64:$Rn,
7221            (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rm, ssub),
7222            (i64 0))>;
7223def : Pat<(v4f32 (any_fmul V128:$Rn, (AArch64dup (f32 FPR32:$Rm)))),
7224          (FMULv4i32_indexed V128:$Rn,
7225            (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rm, ssub),
7226            (i64 0))>;
7227def : Pat<(v2f64 (any_fmul V128:$Rn, (AArch64dup (f64 FPR64:$Rm)))),
7228          (FMULv2i64_indexed V128:$Rn,
7229            (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$Rm, dsub),
7230            (i64 0))>;
7231
7232defm SQDMULH : SIMDIndexedHS<0, 0b1100, "sqdmulh", int_aarch64_neon_sqdmulh>;
7233defm SQRDMULH : SIMDIndexedHS<0, 0b1101, "sqrdmulh", int_aarch64_neon_sqrdmulh>;
7234
7235defm SQDMULH : SIMDIndexedHSPatterns<int_aarch64_neon_sqdmulh_lane,
7236                                     int_aarch64_neon_sqdmulh_laneq>;
7237defm SQRDMULH : SIMDIndexedHSPatterns<int_aarch64_neon_sqrdmulh_lane,
7238                                      int_aarch64_neon_sqrdmulh_laneq>;
7239
7240// Generated by MachineCombine
7241defm MLA   : SIMDVectorIndexedHSTied<1, 0b0000, "mla", null_frag>;
7242defm MLS   : SIMDVectorIndexedHSTied<1, 0b0100, "mls", null_frag>;
7243
7244defm MUL   : SIMDVectorIndexedHS<0, 0b1000, "mul", mul>;
7245defm SMLAL : SIMDVectorIndexedLongSDTied<0, 0b0010, "smlal",
7246    TriOpFrag<(add node:$LHS, (AArch64smull node:$MHS, node:$RHS))>>;
7247defm SMLSL : SIMDVectorIndexedLongSDTied<0, 0b0110, "smlsl",
7248    TriOpFrag<(sub node:$LHS, (AArch64smull node:$MHS, node:$RHS))>>;
7249defm SMULL : SIMDVectorIndexedLongSD<0, 0b1010, "smull", AArch64smull>;
7250defm SQDMLAL : SIMDIndexedLongSQDMLXSDTied<0, 0b0011, "sqdmlal",
7251                                           int_aarch64_neon_sqadd>;
7252defm SQDMLSL : SIMDIndexedLongSQDMLXSDTied<0, 0b0111, "sqdmlsl",
7253                                           int_aarch64_neon_sqsub>;
7254defm SQRDMLAH : SIMDIndexedSQRDMLxHSDTied<1, 0b1101, "sqrdmlah",
7255                                          int_aarch64_neon_sqrdmlah>;
7256defm SQRDMLSH : SIMDIndexedSQRDMLxHSDTied<1, 0b1111, "sqrdmlsh",
7257                                          int_aarch64_neon_sqrdmlsh>;
7258defm SQDMULL : SIMDIndexedLongSD<0, 0b1011, "sqdmull", int_aarch64_neon_sqdmull>;
7259defm UMLAL   : SIMDVectorIndexedLongSDTied<1, 0b0010, "umlal",
7260    TriOpFrag<(add node:$LHS, (AArch64umull node:$MHS, node:$RHS))>>;
7261defm UMLSL   : SIMDVectorIndexedLongSDTied<1, 0b0110, "umlsl",
7262    TriOpFrag<(sub node:$LHS, (AArch64umull node:$MHS, node:$RHS))>>;
7263defm UMULL   : SIMDVectorIndexedLongSD<1, 0b1010, "umull", AArch64umull>;
7264
7265// A scalar sqdmull with the second operand being a vector lane can be
7266// handled directly with the indexed instruction encoding.
7267def : Pat<(int_aarch64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
7268                                          (vector_extract (v4i32 V128:$Vm),
7269                                                           VectorIndexS:$idx)),
7270          (SQDMULLv1i64_indexed FPR32:$Rn, V128:$Vm, VectorIndexS:$idx)>;
7271
7272//----------------------------------------------------------------------------
7273// AdvSIMD scalar shift instructions
7274//----------------------------------------------------------------------------
7275defm FCVTZS : SIMDFPScalarRShift<0, 0b11111, "fcvtzs">;
7276defm FCVTZU : SIMDFPScalarRShift<1, 0b11111, "fcvtzu">;
7277defm SCVTF  : SIMDFPScalarRShift<0, 0b11100, "scvtf">;
7278defm UCVTF  : SIMDFPScalarRShift<1, 0b11100, "ucvtf">;
7279// Codegen patterns for the above. We don't put these directly on the
7280// instructions because TableGen's type inference can't handle the truth.
7281// Having the same base pattern for fp <--> int totally freaks it out.
7282def : Pat<(int_aarch64_neon_vcvtfp2fxs FPR32:$Rn, vecshiftR32:$imm),
7283          (FCVTZSs FPR32:$Rn, vecshiftR32:$imm)>;
7284def : Pat<(int_aarch64_neon_vcvtfp2fxu FPR32:$Rn, vecshiftR32:$imm),
7285          (FCVTZUs FPR32:$Rn, vecshiftR32:$imm)>;
7286def : Pat<(i64 (int_aarch64_neon_vcvtfp2fxs (f64 FPR64:$Rn), vecshiftR64:$imm)),
7287          (FCVTZSd FPR64:$Rn, vecshiftR64:$imm)>;
7288def : Pat<(i64 (int_aarch64_neon_vcvtfp2fxu (f64 FPR64:$Rn), vecshiftR64:$imm)),
7289          (FCVTZUd FPR64:$Rn, vecshiftR64:$imm)>;
7290def : Pat<(v1i64 (int_aarch64_neon_vcvtfp2fxs (v1f64 FPR64:$Rn),
7291                                            vecshiftR64:$imm)),
7292          (FCVTZSd FPR64:$Rn, vecshiftR64:$imm)>;
7293def : Pat<(v1i64 (int_aarch64_neon_vcvtfp2fxu (v1f64 FPR64:$Rn),
7294                                            vecshiftR64:$imm)),
7295          (FCVTZUd FPR64:$Rn, vecshiftR64:$imm)>;
7296def : Pat<(int_aarch64_neon_vcvtfxu2fp FPR32:$Rn, vecshiftR32:$imm),
7297          (UCVTFs FPR32:$Rn, vecshiftR32:$imm)>;
7298def : Pat<(f64 (int_aarch64_neon_vcvtfxu2fp (i64 FPR64:$Rn), vecshiftR64:$imm)),
7299          (UCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
7300def : Pat<(v1f64 (int_aarch64_neon_vcvtfxs2fp (v1i64 FPR64:$Rn),
7301                                            vecshiftR64:$imm)),
7302          (SCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
7303def : Pat<(f64 (int_aarch64_neon_vcvtfxs2fp (i64 FPR64:$Rn), vecshiftR64:$imm)),
7304          (SCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
7305def : Pat<(v1f64 (int_aarch64_neon_vcvtfxu2fp (v1i64 FPR64:$Rn),
7306                                            vecshiftR64:$imm)),
7307          (UCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
7308def : Pat<(int_aarch64_neon_vcvtfxs2fp FPR32:$Rn, vecshiftR32:$imm),
7309          (SCVTFs FPR32:$Rn, vecshiftR32:$imm)>;
7310
7311// Patterns for FP16 Intrinsics - requires reg copy to/from as i16s not supported.
7312
7313def : Pat<(f16 (int_aarch64_neon_vcvtfxs2fp (i32 (sext_inreg FPR32:$Rn, i16)), vecshiftR16:$imm)),
7314          (SCVTFh (f16 (EXTRACT_SUBREG FPR32:$Rn, hsub)), vecshiftR16:$imm)>;
7315def : Pat<(f16 (int_aarch64_neon_vcvtfxs2fp (i32 FPR32:$Rn), vecshiftR16:$imm)),
7316          (SCVTFh (f16 (EXTRACT_SUBREG FPR32:$Rn, hsub)), vecshiftR16:$imm)>;
7317def : Pat<(f16 (int_aarch64_neon_vcvtfxs2fp (i64 FPR64:$Rn), vecshiftR16:$imm)),
7318          (SCVTFh (f16 (EXTRACT_SUBREG FPR64:$Rn, hsub)), vecshiftR16:$imm)>;
7319def : Pat<(f16 (int_aarch64_neon_vcvtfxu2fp
7320            (and FPR32:$Rn, (i32 65535)),
7321            vecshiftR16:$imm)),
7322          (UCVTFh (f16 (EXTRACT_SUBREG FPR32:$Rn, hsub)), vecshiftR16:$imm)>;
7323def : Pat<(f16 (int_aarch64_neon_vcvtfxu2fp FPR32:$Rn, vecshiftR16:$imm)),
7324          (UCVTFh (f16 (EXTRACT_SUBREG FPR32:$Rn, hsub)), vecshiftR16:$imm)>;
7325def : Pat<(f16 (int_aarch64_neon_vcvtfxu2fp (i64 FPR64:$Rn), vecshiftR16:$imm)),
7326          (UCVTFh (f16 (EXTRACT_SUBREG FPR64:$Rn, hsub)), vecshiftR16:$imm)>;
7327def : Pat<(i32 (int_aarch64_neon_vcvtfp2fxs (f16 FPR16:$Rn), vecshiftR32:$imm)),
7328          (i32 (INSERT_SUBREG
7329            (i32 (IMPLICIT_DEF)),
7330            (FCVTZSh FPR16:$Rn, vecshiftR32:$imm),
7331            hsub))>;
7332def : Pat<(i64 (int_aarch64_neon_vcvtfp2fxs (f16 FPR16:$Rn), vecshiftR64:$imm)),
7333          (i64 (INSERT_SUBREG
7334            (i64 (IMPLICIT_DEF)),
7335            (FCVTZSh FPR16:$Rn, vecshiftR64:$imm),
7336            hsub))>;
7337def : Pat<(i32 (int_aarch64_neon_vcvtfp2fxu (f16 FPR16:$Rn), vecshiftR32:$imm)),
7338          (i32 (INSERT_SUBREG
7339            (i32 (IMPLICIT_DEF)),
7340            (FCVTZUh FPR16:$Rn, vecshiftR32:$imm),
7341            hsub))>;
7342def : Pat<(i64 (int_aarch64_neon_vcvtfp2fxu (f16 FPR16:$Rn), vecshiftR64:$imm)),
7343          (i64 (INSERT_SUBREG
7344            (i64 (IMPLICIT_DEF)),
7345            (FCVTZUh FPR16:$Rn, vecshiftR64:$imm),
7346            hsub))>;
7347def : Pat<(i32 (int_aarch64_neon_facge (f16 FPR16:$Rn), (f16 FPR16:$Rm))),
7348          (i32 (INSERT_SUBREG
7349            (i32 (IMPLICIT_DEF)),
7350            (FACGE16 FPR16:$Rn, FPR16:$Rm),
7351            hsub))>;
7352def : Pat<(i32 (int_aarch64_neon_facgt (f16 FPR16:$Rn), (f16 FPR16:$Rm))),
7353          (i32 (INSERT_SUBREG
7354            (i32 (IMPLICIT_DEF)),
7355            (FACGT16 FPR16:$Rn, FPR16:$Rm),
7356            hsub))>;
7357
7358defm SHL      : SIMDScalarLShiftD<   0, 0b01010, "shl", AArch64vshl>;
7359defm SLI      : SIMDScalarLShiftDTied<1, 0b01010, "sli">;
7360defm SQRSHRN  : SIMDScalarRShiftBHS< 0, 0b10011, "sqrshrn",
7361                                     int_aarch64_neon_sqrshrn>;
7362defm SQRSHRUN : SIMDScalarRShiftBHS< 1, 0b10001, "sqrshrun",
7363                                     int_aarch64_neon_sqrshrun>;
7364defm SQSHLU   : SIMDScalarLShiftBHSD<1, 0b01100, "sqshlu", AArch64sqshlui>;
7365defm SQSHL    : SIMDScalarLShiftBHSD<0, 0b01110, "sqshl", AArch64sqshli>;
7366defm SQSHRN   : SIMDScalarRShiftBHS< 0, 0b10010, "sqshrn",
7367                                     int_aarch64_neon_sqshrn>;
7368defm SQSHRUN  : SIMDScalarRShiftBHS< 1, 0b10000, "sqshrun",
7369                                     int_aarch64_neon_sqshrun>;
7370defm SRI      : SIMDScalarRShiftDTied<   1, 0b01000, "sri">;
7371defm SRSHR    : SIMDScalarRShiftD<   0, 0b00100, "srshr", AArch64srshri>;
7372defm SRSRA    : SIMDScalarRShiftDTied<   0, 0b00110, "srsra",
7373    TriOpFrag<(add node:$LHS,
7374                   (AArch64srshri node:$MHS, node:$RHS))>>;
7375defm SSHR     : SIMDScalarRShiftD<   0, 0b00000, "sshr", AArch64vashr>;
7376defm SSRA     : SIMDScalarRShiftDTied<   0, 0b00010, "ssra",
7377    TriOpFrag<(add_and_or_is_add node:$LHS,
7378                   (AArch64vashr node:$MHS, node:$RHS))>>;
7379defm UQRSHRN  : SIMDScalarRShiftBHS< 1, 0b10011, "uqrshrn",
7380                                     int_aarch64_neon_uqrshrn>;
7381defm UQSHL    : SIMDScalarLShiftBHSD<1, 0b01110, "uqshl", AArch64uqshli>;
7382defm UQSHRN   : SIMDScalarRShiftBHS< 1, 0b10010, "uqshrn",
7383                                     int_aarch64_neon_uqshrn>;
7384defm URSHR    : SIMDScalarRShiftD<   1, 0b00100, "urshr", AArch64urshri>;
7385defm URSRA    : SIMDScalarRShiftDTied<   1, 0b00110, "ursra",
7386    TriOpFrag<(add node:$LHS,
7387                   (AArch64urshri node:$MHS, node:$RHS))>>;
7388defm USHR     : SIMDScalarRShiftD<   1, 0b00000, "ushr", AArch64vlshr>;
7389defm USRA     : SIMDScalarRShiftDTied<   1, 0b00010, "usra",
7390    TriOpFrag<(add_and_or_is_add node:$LHS,
7391                   (AArch64vlshr node:$MHS, node:$RHS))>>;
7392
7393//----------------------------------------------------------------------------
7394// AdvSIMD vector shift instructions
7395//----------------------------------------------------------------------------
7396defm FCVTZS:SIMDVectorRShiftSD<0, 0b11111, "fcvtzs", int_aarch64_neon_vcvtfp2fxs>;
7397defm FCVTZU:SIMDVectorRShiftSD<1, 0b11111, "fcvtzu", int_aarch64_neon_vcvtfp2fxu>;
7398defm SCVTF: SIMDVectorRShiftToFP<0, 0b11100, "scvtf",
7399                                   int_aarch64_neon_vcvtfxs2fp>;
7400defm RSHRN   : SIMDVectorRShiftNarrowBHS<0, 0b10001, "rshrn", AArch64rshrn>;
7401defm SHL     : SIMDVectorLShiftBHSD<0, 0b01010, "shl", AArch64vshl>;
7402
7403// X << 1 ==> X + X
7404class SHLToADDPat<ValueType ty, RegisterClass regtype>
7405  : Pat<(ty (AArch64vshl (ty regtype:$Rn), (i32 1))),
7406            (!cast<Instruction>("ADD"#ty) regtype:$Rn, regtype:$Rn)>;
7407
7408def : SHLToADDPat<v16i8, FPR128>;
7409def : SHLToADDPat<v8i16, FPR128>;
7410def : SHLToADDPat<v4i32, FPR128>;
7411def : SHLToADDPat<v2i64, FPR128>;
7412def : SHLToADDPat<v8i8,  FPR64>;
7413def : SHLToADDPat<v4i16, FPR64>;
7414def : SHLToADDPat<v2i32, FPR64>;
7415
7416defm SHRN    : SIMDVectorRShiftNarrowBHS<0, 0b10000, "shrn",
7417                          BinOpFrag<(trunc (AArch64vashr node:$LHS, node:$RHS))>>;
7418defm SLI     : SIMDVectorLShiftBHSDTied<1, 0b01010, "sli", AArch64vsli>;
7419def : Pat<(v1i64 (AArch64vsli (v1i64 FPR64:$Rd), (v1i64 FPR64:$Rn),
7420                                      (i32 vecshiftL64:$imm))),
7421          (SLId FPR64:$Rd, FPR64:$Rn, vecshiftL64:$imm)>;
7422defm SQRSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10011, "sqrshrn",
7423                                         int_aarch64_neon_sqrshrn>;
7424defm SQRSHRUN: SIMDVectorRShiftNarrowBHS<1, 0b10001, "sqrshrun",
7425                                         int_aarch64_neon_sqrshrun>;
7426defm SQSHLU : SIMDVectorLShiftBHSD<1, 0b01100, "sqshlu", AArch64sqshlui>;
7427defm SQSHL  : SIMDVectorLShiftBHSD<0, 0b01110, "sqshl", AArch64sqshli>;
7428defm SQSHRN  : SIMDVectorRShiftNarrowBHS<0, 0b10010, "sqshrn",
7429                                         int_aarch64_neon_sqshrn>;
7430defm SQSHRUN : SIMDVectorRShiftNarrowBHS<1, 0b10000, "sqshrun",
7431                                         int_aarch64_neon_sqshrun>;
7432defm SRI     : SIMDVectorRShiftBHSDTied<1, 0b01000, "sri", AArch64vsri>;
7433def : Pat<(v1i64 (AArch64vsri (v1i64 FPR64:$Rd), (v1i64 FPR64:$Rn),
7434                                      (i32 vecshiftR64:$imm))),
7435          (SRId FPR64:$Rd, FPR64:$Rn, vecshiftR64:$imm)>;
7436defm SRSHR   : SIMDVectorRShiftBHSD<0, 0b00100, "srshr", AArch64srshri>;
7437defm SRSRA   : SIMDVectorRShiftBHSDTied<0, 0b00110, "srsra",
7438                 TriOpFrag<(add node:$LHS,
7439                                (AArch64srshri node:$MHS, node:$RHS))> >;
7440defm SSHLL   : SIMDVectorLShiftLongBHSD<0, 0b10100, "sshll",
7441                BinOpFrag<(AArch64vshl (sext node:$LHS), node:$RHS)>>;
7442
7443defm SSHR    : SIMDVectorRShiftBHSD<0, 0b00000, "sshr", AArch64vashr>;
7444defm SSRA    : SIMDVectorRShiftBHSDTied<0, 0b00010, "ssra",
7445                TriOpFrag<(add_and_or_is_add node:$LHS, (AArch64vashr node:$MHS, node:$RHS))>>;
7446defm UCVTF   : SIMDVectorRShiftToFP<1, 0b11100, "ucvtf",
7447                        int_aarch64_neon_vcvtfxu2fp>;
7448defm UQRSHRN : SIMDVectorRShiftNarrowBHS<1, 0b10011, "uqrshrn",
7449                                         int_aarch64_neon_uqrshrn>;
7450defm UQSHL   : SIMDVectorLShiftBHSD<1, 0b01110, "uqshl", AArch64uqshli>;
7451defm UQSHRN  : SIMDVectorRShiftNarrowBHS<1, 0b10010, "uqshrn",
7452                                         int_aarch64_neon_uqshrn>;
7453defm URSHR   : SIMDVectorRShiftBHSD<1, 0b00100, "urshr", AArch64urshri>;
7454defm URSRA   : SIMDVectorRShiftBHSDTied<1, 0b00110, "ursra",
7455                TriOpFrag<(add node:$LHS,
7456                               (AArch64urshri node:$MHS, node:$RHS))> >;
7457defm USHLL   : SIMDVectorLShiftLongBHSD<1, 0b10100, "ushll",
7458                BinOpFrag<(AArch64vshl (zext node:$LHS), node:$RHS)>>;
7459defm USHR    : SIMDVectorRShiftBHSD<1, 0b00000, "ushr", AArch64vlshr>;
7460defm USRA    : SIMDVectorRShiftBHSDTied<1, 0b00010, "usra",
7461                TriOpFrag<(add_and_or_is_add node:$LHS, (AArch64vlshr node:$MHS, node:$RHS))> >;
7462
7463// RADDHN patterns for when RSHRN shifts by half the size of the vector element
7464def : Pat<(v8i8 (trunc (AArch64vlshr (add (v8i16 V128:$Vn), (AArch64movi_shift (i32 128), (i32 0))), (i32 8)))),
7465          (RADDHNv8i16_v8i8 V128:$Vn, (v8i16 (MOVIv2d_ns (i32 0))))>;
7466def : Pat<(v4i16 (trunc (AArch64vlshr (add (v4i32 V128:$Vn), (AArch64movi_shift (i32 128), (i32 8))), (i32 16)))),
7467          (RADDHNv4i32_v4i16 V128:$Vn, (v4i32 (MOVIv2d_ns (i32 0))))>;
7468let AddedComplexity = 5 in
7469def : Pat<(v2i32 (trunc (AArch64vlshr (add (v2i64 V128:$Vn), (AArch64dup (i64 2147483648))), (i32 32)))),
7470          (RADDHNv2i64_v2i32 V128:$Vn, (v2i64 (MOVIv2d_ns (i32 0))))>;
7471def : Pat<(v8i8 (int_aarch64_neon_rshrn (v8i16 V128:$Vn), (i32 8))),
7472          (RADDHNv8i16_v8i8 V128:$Vn, (v8i16 (MOVIv2d_ns (i32 0))))>;
7473def : Pat<(v4i16 (int_aarch64_neon_rshrn (v4i32 V128:$Vn), (i32 16))),
7474          (RADDHNv4i32_v4i16 V128:$Vn, (v4i32 (MOVIv2d_ns (i32 0))))>;
7475def : Pat<(v2i32 (int_aarch64_neon_rshrn (v2i64 V128:$Vn), (i32 32))),
7476          (RADDHNv2i64_v2i32 V128:$Vn, (v2i64 (MOVIv2d_ns (i32 0))))>;
7477
7478// RADDHN2 patterns for when RSHRN shifts by half the size of the vector element
7479def : Pat<(v16i8 (concat_vectors
7480                 (v8i8 V64:$Vd),
7481                 (v8i8 (trunc (AArch64vlshr (add (v8i16 V128:$Vn), (AArch64movi_shift (i32 128), (i32 0))), (i32 8)))))),
7482          (RADDHNv8i16_v16i8
7483                 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Vd, dsub), V128:$Vn,
7484                 (v8i16 (MOVIv2d_ns (i32 0))))>;
7485def : Pat<(v8i16 (concat_vectors
7486                 (v4i16 V64:$Vd),
7487                 (v4i16 (trunc (AArch64vlshr (add (v4i32 V128:$Vn), (AArch64movi_shift (i32 128), (i32 8))), (i32 16)))))),
7488          (RADDHNv4i32_v8i16
7489                 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Vd, dsub), V128:$Vn,
7490                 (v4i32 (MOVIv2d_ns (i32 0))))>;
7491let AddedComplexity = 5 in
7492def : Pat<(v4i32 (concat_vectors
7493                 (v2i32 V64:$Vd),
7494                 (v2i32 (trunc (AArch64vlshr (add (v2i64 V128:$Vn), (AArch64dup (i64 2147483648))), (i32 32)))))),
7495          (RADDHNv2i64_v4i32
7496                 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Vd, dsub), V128:$Vn,
7497                 (v2i64 (MOVIv2d_ns (i32 0))))>;
7498def : Pat<(v16i8 (concat_vectors
7499                 (v8i8 V64:$Vd),
7500                 (v8i8 (int_aarch64_neon_rshrn (v8i16 V128:$Vn), (i32 8))))),
7501          (RADDHNv8i16_v16i8
7502                 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Vd, dsub), V128:$Vn,
7503                 (v8i16 (MOVIv2d_ns (i32 0))))>;
7504def : Pat<(v8i16 (concat_vectors
7505                 (v4i16 V64:$Vd),
7506                 (v4i16 (int_aarch64_neon_rshrn (v4i32 V128:$Vn), (i32 16))))),
7507          (RADDHNv4i32_v8i16
7508                 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Vd, dsub), V128:$Vn,
7509                 (v4i32 (MOVIv2d_ns (i32 0))))>;
7510def : Pat<(v4i32 (concat_vectors
7511                 (v2i32 V64:$Vd),
7512                 (v2i32 (int_aarch64_neon_rshrn (v2i64 V128:$Vn), (i32 32))))),
7513          (RADDHNv2i64_v4i32
7514                 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Vd, dsub), V128:$Vn,
7515                 (v2i64 (MOVIv2d_ns (i32 0))))>;
7516
7517// SHRN patterns for when a logical right shift was used instead of arithmetic
7518// (the immediate guarantees no sign bits actually end up in the result so it
7519// doesn't matter).
7520def : Pat<(v8i8 (trunc (AArch64vlshr (v8i16 V128:$Rn), vecshiftR16Narrow:$imm))),
7521          (SHRNv8i8_shift V128:$Rn, vecshiftR16Narrow:$imm)>;
7522def : Pat<(v4i16 (trunc (AArch64vlshr (v4i32 V128:$Rn), vecshiftR32Narrow:$imm))),
7523          (SHRNv4i16_shift V128:$Rn, vecshiftR32Narrow:$imm)>;
7524def : Pat<(v2i32 (trunc (AArch64vlshr (v2i64 V128:$Rn), vecshiftR64Narrow:$imm))),
7525          (SHRNv2i32_shift V128:$Rn, vecshiftR64Narrow:$imm)>;
7526
7527def : Pat<(v16i8 (concat_vectors (v8i8 V64:$Rd),
7528                                 (trunc (AArch64vlshr (v8i16 V128:$Rn),
7529                                                    vecshiftR16Narrow:$imm)))),
7530          (SHRNv16i8_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
7531                           V128:$Rn, vecshiftR16Narrow:$imm)>;
7532def : Pat<(v8i16 (concat_vectors (v4i16 V64:$Rd),
7533                                 (trunc (AArch64vlshr (v4i32 V128:$Rn),
7534                                                    vecshiftR32Narrow:$imm)))),
7535          (SHRNv8i16_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
7536                           V128:$Rn, vecshiftR32Narrow:$imm)>;
7537def : Pat<(v4i32 (concat_vectors (v2i32 V64:$Rd),
7538                                 (trunc (AArch64vlshr (v2i64 V128:$Rn),
7539                                                    vecshiftR64Narrow:$imm)))),
7540          (SHRNv4i32_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
7541                           V128:$Rn, vecshiftR32Narrow:$imm)>;
7542
7543// Vector sign and zero extensions are implemented with SSHLL and USSHLL.
7544// Anyexts are implemented as zexts.
7545def : Pat<(v8i16 (sext   (v8i8 V64:$Rn))),  (SSHLLv8i8_shift  V64:$Rn, (i32 0))>;
7546def : Pat<(v8i16 (zext   (v8i8 V64:$Rn))),  (USHLLv8i8_shift  V64:$Rn, (i32 0))>;
7547def : Pat<(v8i16 (anyext (v8i8 V64:$Rn))),  (USHLLv8i8_shift  V64:$Rn, (i32 0))>;
7548def : Pat<(v4i32 (sext   (v4i16 V64:$Rn))), (SSHLLv4i16_shift V64:$Rn, (i32 0))>;
7549def : Pat<(v4i32 (zext   (v4i16 V64:$Rn))), (USHLLv4i16_shift V64:$Rn, (i32 0))>;
7550def : Pat<(v4i32 (anyext (v4i16 V64:$Rn))), (USHLLv4i16_shift V64:$Rn, (i32 0))>;
7551def : Pat<(v2i64 (sext   (v2i32 V64:$Rn))), (SSHLLv2i32_shift V64:$Rn, (i32 0))>;
7552def : Pat<(v2i64 (zext   (v2i32 V64:$Rn))), (USHLLv2i32_shift V64:$Rn, (i32 0))>;
7553def : Pat<(v2i64 (anyext (v2i32 V64:$Rn))), (USHLLv2i32_shift V64:$Rn, (i32 0))>;
7554// Also match an extend from the upper half of a 128 bit source register.
7555def : Pat<(v8i16 (anyext (v8i8 (extract_high_v16i8 (v16i8 V128:$Rn)) ))),
7556          (USHLLv16i8_shift V128:$Rn, (i32 0))>;
7557def : Pat<(v8i16 (zext   (v8i8 (extract_high_v16i8 (v16i8 V128:$Rn)) ))),
7558          (USHLLv16i8_shift V128:$Rn, (i32 0))>;
7559def : Pat<(v8i16 (sext   (v8i8 (extract_high_v16i8 (v16i8 V128:$Rn)) ))),
7560          (SSHLLv16i8_shift V128:$Rn, (i32 0))>;
7561def : Pat<(v4i32 (anyext (v4i16 (extract_high_v8i16 (v8i16 V128:$Rn)) ))),
7562          (USHLLv8i16_shift V128:$Rn, (i32 0))>;
7563def : Pat<(v4i32 (zext   (v4i16 (extract_high_v8i16 (v8i16 V128:$Rn)) ))),
7564          (USHLLv8i16_shift V128:$Rn, (i32 0))>;
7565def : Pat<(v4i32 (sext   (v4i16 (extract_high_v8i16 (v8i16 V128:$Rn)) ))),
7566          (SSHLLv8i16_shift V128:$Rn, (i32 0))>;
7567def : Pat<(v2i64 (anyext (v2i32 (extract_high_v4i32 (v4i32 V128:$Rn)) ))),
7568          (USHLLv4i32_shift V128:$Rn, (i32 0))>;
7569def : Pat<(v2i64 (zext   (v2i32 (extract_high_v4i32 (v4i32 V128:$Rn)) ))),
7570          (USHLLv4i32_shift V128:$Rn, (i32 0))>;
7571def : Pat<(v2i64 (sext   (v2i32 (extract_high_v4i32 (v4i32 V128:$Rn)) ))),
7572          (SSHLLv4i32_shift V128:$Rn, (i32 0))>;
7573
7574// Vector shift sxtl aliases
7575def : InstAlias<"sxtl.8h $dst, $src1",
7576                (SSHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
7577def : InstAlias<"sxtl $dst.8h, $src1.8b",
7578                (SSHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
7579def : InstAlias<"sxtl.4s $dst, $src1",
7580                (SSHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
7581def : InstAlias<"sxtl $dst.4s, $src1.4h",
7582                (SSHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
7583def : InstAlias<"sxtl.2d $dst, $src1",
7584                (SSHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
7585def : InstAlias<"sxtl $dst.2d, $src1.2s",
7586                (SSHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
7587
7588// Vector shift sxtl2 aliases
7589def : InstAlias<"sxtl2.8h $dst, $src1",
7590                (SSHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
7591def : InstAlias<"sxtl2 $dst.8h, $src1.16b",
7592                (SSHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
7593def : InstAlias<"sxtl2.4s $dst, $src1",
7594                (SSHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
7595def : InstAlias<"sxtl2 $dst.4s, $src1.8h",
7596                (SSHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
7597def : InstAlias<"sxtl2.2d $dst, $src1",
7598                (SSHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
7599def : InstAlias<"sxtl2 $dst.2d, $src1.4s",
7600                (SSHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
7601
7602// Vector shift uxtl aliases
7603def : InstAlias<"uxtl.8h $dst, $src1",
7604                (USHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
7605def : InstAlias<"uxtl $dst.8h, $src1.8b",
7606                (USHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
7607def : InstAlias<"uxtl.4s $dst, $src1",
7608                (USHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
7609def : InstAlias<"uxtl $dst.4s, $src1.4h",
7610                (USHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
7611def : InstAlias<"uxtl.2d $dst, $src1",
7612                (USHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
7613def : InstAlias<"uxtl $dst.2d, $src1.2s",
7614                (USHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
7615
7616// Vector shift uxtl2 aliases
7617def : InstAlias<"uxtl2.8h $dst, $src1",
7618                (USHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
7619def : InstAlias<"uxtl2 $dst.8h, $src1.16b",
7620                (USHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
7621def : InstAlias<"uxtl2.4s $dst, $src1",
7622                (USHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
7623def : InstAlias<"uxtl2 $dst.4s, $src1.8h",
7624                (USHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
7625def : InstAlias<"uxtl2.2d $dst, $src1",
7626                (USHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
7627def : InstAlias<"uxtl2 $dst.2d, $src1.4s",
7628                (USHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
7629
7630// If an integer is about to be converted to a floating point value,
7631// just load it on the floating point unit.
7632// These patterns are more complex because floating point loads do not
7633// support sign extension.
7634// The sign extension has to be explicitly added and is only supported for
7635// one step: byte-to-half, half-to-word, word-to-doubleword.
7636// SCVTF GPR -> FPR is 9 cycles.
7637// SCVTF FPR -> FPR is 4 cyclces.
7638// (sign extension with lengthen) SXTL FPR -> FPR is 2 cycles.
7639// Therefore, we can do 2 sign extensions and one SCVTF FPR -> FPR
7640// and still being faster.
7641// However, this is not good for code size.
7642// 8-bits -> float. 2 sizes step-up.
7643class SExtLoadi8CVTf32Pat<dag addrmode, dag INST>
7644  : Pat<(f32 (sint_to_fp (i32 (sextloadi8 addrmode)))),
7645        (SCVTFv1i32 (f32 (EXTRACT_SUBREG
7646                            (SSHLLv4i16_shift
7647                              (f64
7648                                (EXTRACT_SUBREG
7649                                  (SSHLLv8i8_shift
7650                                    (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
7651                                        INST,
7652                                        bsub),
7653                                    0),
7654                                  dsub)),
7655                               0),
7656                             ssub)))>,
7657    Requires<[NotForCodeSize, UseAlternateSExtLoadCVTF32, HasNEON]>;
7658
7659def : SExtLoadi8CVTf32Pat<(ro8.Wpat GPR64sp:$Rn, GPR32:$Rm, ro8.Wext:$ext),
7660                          (LDRBroW  GPR64sp:$Rn, GPR32:$Rm, ro8.Wext:$ext)>;
7661def : SExtLoadi8CVTf32Pat<(ro8.Xpat GPR64sp:$Rn, GPR64:$Rm, ro8.Xext:$ext),
7662                          (LDRBroX  GPR64sp:$Rn, GPR64:$Rm, ro8.Xext:$ext)>;
7663def : SExtLoadi8CVTf32Pat<(am_indexed8 GPR64sp:$Rn, uimm12s1:$offset),
7664                          (LDRBui GPR64sp:$Rn, uimm12s1:$offset)>;
7665def : SExtLoadi8CVTf32Pat<(am_unscaled8 GPR64sp:$Rn, simm9:$offset),
7666                          (LDURBi GPR64sp:$Rn, simm9:$offset)>;
7667
7668// 16-bits -> float. 1 size step-up.
7669class SExtLoadi16CVTf32Pat<dag addrmode, dag INST>
7670  : Pat<(f32 (sint_to_fp (i32 (sextloadi16 addrmode)))),
7671        (SCVTFv1i32 (f32 (EXTRACT_SUBREG
7672                            (SSHLLv4i16_shift
7673                                (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
7674                                  INST,
7675                                  hsub),
7676                                0),
7677                            ssub)))>,
7678    Requires<[NotForCodeSize, UseAlternateSExtLoadCVTF32, HasNEON]>;
7679
7680def : SExtLoadi16CVTf32Pat<(ro16.Wpat GPR64sp:$Rn, GPR32:$Rm, ro16.Wext:$ext),
7681                           (LDRHroW   GPR64sp:$Rn, GPR32:$Rm, ro16.Wext:$ext)>;
7682def : SExtLoadi16CVTf32Pat<(ro16.Xpat GPR64sp:$Rn, GPR64:$Rm, ro16.Xext:$ext),
7683                           (LDRHroX   GPR64sp:$Rn, GPR64:$Rm, ro16.Xext:$ext)>;
7684def : SExtLoadi16CVTf32Pat<(am_indexed16 GPR64sp:$Rn, uimm12s2:$offset),
7685                           (LDRHui GPR64sp:$Rn, uimm12s2:$offset)>;
7686def : SExtLoadi16CVTf32Pat<(am_unscaled16 GPR64sp:$Rn, simm9:$offset),
7687                           (LDURHi GPR64sp:$Rn, simm9:$offset)>;
7688
7689// 32-bits to 32-bits are handled in target specific dag combine:
7690// performIntToFpCombine.
7691// 64-bits integer to 32-bits floating point, not possible with
7692// SCVTF on floating point registers (both source and destination
7693// must have the same size).
7694
7695// Here are the patterns for 8, 16, 32, and 64-bits to double.
7696// 8-bits -> double. 3 size step-up: give up.
7697// 16-bits -> double. 2 size step.
7698class SExtLoadi16CVTf64Pat<dag addrmode, dag INST>
7699  : Pat <(f64 (sint_to_fp (i32 (sextloadi16 addrmode)))),
7700           (SCVTFv1i64 (f64 (EXTRACT_SUBREG
7701                              (SSHLLv2i32_shift
7702                                 (f64
7703                                  (EXTRACT_SUBREG
7704                                    (SSHLLv4i16_shift
7705                                      (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
7706                                        INST,
7707                                        hsub),
7708                                     0),
7709                                   dsub)),
7710                               0),
7711                             dsub)))>,
7712    Requires<[NotForCodeSize, UseAlternateSExtLoadCVTF32, HasNEON]>;
7713
7714def : SExtLoadi16CVTf64Pat<(ro16.Wpat GPR64sp:$Rn, GPR32:$Rm, ro16.Wext:$ext),
7715                           (LDRHroW GPR64sp:$Rn, GPR32:$Rm, ro16.Wext:$ext)>;
7716def : SExtLoadi16CVTf64Pat<(ro16.Xpat GPR64sp:$Rn, GPR64:$Rm, ro16.Xext:$ext),
7717                           (LDRHroX GPR64sp:$Rn, GPR64:$Rm, ro16.Xext:$ext)>;
7718def : SExtLoadi16CVTf64Pat<(am_indexed16 GPR64sp:$Rn, uimm12s2:$offset),
7719                           (LDRHui GPR64sp:$Rn, uimm12s2:$offset)>;
7720def : SExtLoadi16CVTf64Pat<(am_unscaled16 GPR64sp:$Rn, simm9:$offset),
7721                           (LDURHi GPR64sp:$Rn, simm9:$offset)>;
7722// 32-bits -> double. 1 size step-up.
7723class SExtLoadi32CVTf64Pat<dag addrmode, dag INST>
7724  : Pat <(f64 (sint_to_fp (i32 (load addrmode)))),
7725           (SCVTFv1i64 (f64 (EXTRACT_SUBREG
7726                              (SSHLLv2i32_shift
7727                                (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
7728                                  INST,
7729                                  ssub),
7730                               0),
7731                             dsub)))>,
7732    Requires<[NotForCodeSize, UseAlternateSExtLoadCVTF32, HasNEON]>;
7733
7734def : SExtLoadi32CVTf64Pat<(ro32.Wpat GPR64sp:$Rn, GPR32:$Rm, ro32.Wext:$ext),
7735                           (LDRSroW GPR64sp:$Rn, GPR32:$Rm, ro32.Wext:$ext)>;
7736def : SExtLoadi32CVTf64Pat<(ro32.Xpat GPR64sp:$Rn, GPR64:$Rm, ro32.Xext:$ext),
7737                           (LDRSroX GPR64sp:$Rn, GPR64:$Rm, ro32.Xext:$ext)>;
7738def : SExtLoadi32CVTf64Pat<(am_indexed32 GPR64sp:$Rn, uimm12s4:$offset),
7739                           (LDRSui GPR64sp:$Rn, uimm12s4:$offset)>;
7740def : SExtLoadi32CVTf64Pat<(am_unscaled32 GPR64sp:$Rn, simm9:$offset),
7741                           (LDURSi GPR64sp:$Rn, simm9:$offset)>;
7742
7743// 64-bits -> double are handled in target specific dag combine:
7744// performIntToFpCombine.
7745
7746
7747//----------------------------------------------------------------------------
7748// AdvSIMD Load-Store Structure
7749//----------------------------------------------------------------------------
7750defm LD1 : SIMDLd1Multiple<"ld1">;
7751defm LD2 : SIMDLd2Multiple<"ld2">;
7752defm LD3 : SIMDLd3Multiple<"ld3">;
7753defm LD4 : SIMDLd4Multiple<"ld4">;
7754
7755defm ST1 : SIMDSt1Multiple<"st1">;
7756defm ST2 : SIMDSt2Multiple<"st2">;
7757defm ST3 : SIMDSt3Multiple<"st3">;
7758defm ST4 : SIMDSt4Multiple<"st4">;
7759
7760class Ld1Pat<ValueType ty, Instruction INST>
7761  : Pat<(ty (load GPR64sp:$Rn)), (INST GPR64sp:$Rn)>;
7762
7763def : Ld1Pat<v16i8, LD1Onev16b>;
7764def : Ld1Pat<v8i16, LD1Onev8h>;
7765def : Ld1Pat<v4i32, LD1Onev4s>;
7766def : Ld1Pat<v2i64, LD1Onev2d>;
7767def : Ld1Pat<v8i8,  LD1Onev8b>;
7768def : Ld1Pat<v4i16, LD1Onev4h>;
7769def : Ld1Pat<v2i32, LD1Onev2s>;
7770def : Ld1Pat<v1i64, LD1Onev1d>;
7771
7772class St1Pat<ValueType ty, Instruction INST>
7773  : Pat<(store ty:$Vt, GPR64sp:$Rn),
7774        (INST ty:$Vt, GPR64sp:$Rn)>;
7775
7776def : St1Pat<v16i8, ST1Onev16b>;
7777def : St1Pat<v8i16, ST1Onev8h>;
7778def : St1Pat<v4i32, ST1Onev4s>;
7779def : St1Pat<v2i64, ST1Onev2d>;
7780def : St1Pat<v8i8,  ST1Onev8b>;
7781def : St1Pat<v4i16, ST1Onev4h>;
7782def : St1Pat<v2i32, ST1Onev2s>;
7783def : St1Pat<v1i64, ST1Onev1d>;
7784
7785//---
7786// Single-element
7787//---
7788
7789defm LD1R          : SIMDLdR<0, 0b110, 0, "ld1r", "One", 1, 2, 4, 8>;
7790defm LD2R          : SIMDLdR<1, 0b110, 0, "ld2r", "Two", 2, 4, 8, 16>;
7791defm LD3R          : SIMDLdR<0, 0b111, 0, "ld3r", "Three", 3, 6, 12, 24>;
7792defm LD4R          : SIMDLdR<1, 0b111, 0, "ld4r", "Four", 4, 8, 16, 32>;
7793let mayLoad = 1, hasSideEffects = 0 in {
7794defm LD1 : SIMDLdSingleBTied<0, 0b000,       "ld1", VecListOneb,   GPR64pi1>;
7795defm LD1 : SIMDLdSingleHTied<0, 0b010, 0,    "ld1", VecListOneh,   GPR64pi2>;
7796defm LD1 : SIMDLdSingleSTied<0, 0b100, 0b00, "ld1", VecListOnes,   GPR64pi4>;
7797defm LD1 : SIMDLdSingleDTied<0, 0b100, 0b01, "ld1", VecListOned,   GPR64pi8>;
7798defm LD2 : SIMDLdSingleBTied<1, 0b000,       "ld2", VecListTwob,   GPR64pi2>;
7799defm LD2 : SIMDLdSingleHTied<1, 0b010, 0,    "ld2", VecListTwoh,   GPR64pi4>;
7800defm LD2 : SIMDLdSingleSTied<1, 0b100, 0b00, "ld2", VecListTwos,   GPR64pi8>;
7801defm LD2 : SIMDLdSingleDTied<1, 0b100, 0b01, "ld2", VecListTwod,   GPR64pi16>;
7802defm LD3 : SIMDLdSingleBTied<0, 0b001,       "ld3", VecListThreeb, GPR64pi3>;
7803defm LD3 : SIMDLdSingleHTied<0, 0b011, 0,    "ld3", VecListThreeh, GPR64pi6>;
7804defm LD3 : SIMDLdSingleSTied<0, 0b101, 0b00, "ld3", VecListThrees, GPR64pi12>;
7805defm LD3 : SIMDLdSingleDTied<0, 0b101, 0b01, "ld3", VecListThreed, GPR64pi24>;
7806defm LD4 : SIMDLdSingleBTied<1, 0b001,       "ld4", VecListFourb,  GPR64pi4>;
7807defm LD4 : SIMDLdSingleHTied<1, 0b011, 0,    "ld4", VecListFourh,  GPR64pi8>;
7808defm LD4 : SIMDLdSingleSTied<1, 0b101, 0b00, "ld4", VecListFours,  GPR64pi16>;
7809defm LD4 : SIMDLdSingleDTied<1, 0b101, 0b01, "ld4", VecListFourd,  GPR64pi32>;
7810}
7811
7812def : Pat<(v8i8 (AArch64dup (i32 (extloadi8 GPR64sp:$Rn)))),
7813          (LD1Rv8b GPR64sp:$Rn)>;
7814def : Pat<(v16i8 (AArch64dup (i32 (extloadi8 GPR64sp:$Rn)))),
7815          (LD1Rv16b GPR64sp:$Rn)>;
7816def : Pat<(v4i16 (AArch64dup (i32 (extloadi16 GPR64sp:$Rn)))),
7817          (LD1Rv4h GPR64sp:$Rn)>;
7818def : Pat<(v8i16 (AArch64dup (i32 (extloadi16 GPR64sp:$Rn)))),
7819          (LD1Rv8h GPR64sp:$Rn)>;
7820def : Pat<(v2i32 (AArch64dup (i32 (load GPR64sp:$Rn)))),
7821          (LD1Rv2s GPR64sp:$Rn)>;
7822def : Pat<(v4i32 (AArch64dup (i32 (load GPR64sp:$Rn)))),
7823          (LD1Rv4s GPR64sp:$Rn)>;
7824def : Pat<(v2i64 (AArch64dup (i64 (load GPR64sp:$Rn)))),
7825          (LD1Rv2d GPR64sp:$Rn)>;
7826def : Pat<(v1i64 (AArch64dup (i64 (load GPR64sp:$Rn)))),
7827          (LD1Rv1d GPR64sp:$Rn)>;
7828
7829def : Pat<(v8i8 (AArch64duplane8 (v16i8 (insert_subvector undef, (v8i8 (load GPR64sp:$Rn)), (i64 0))), (i64 0))),
7830          (LD1Rv8b GPR64sp:$Rn)>;
7831def : Pat<(v16i8 (AArch64duplane8 (v16i8 (load GPR64sp:$Rn)), (i64 0))),
7832          (LD1Rv16b GPR64sp:$Rn)>;
7833def : Pat<(v4i16 (AArch64duplane16 (v8i16 (insert_subvector undef, (v4i16 (load GPR64sp:$Rn)), (i64 0))), (i64 0))),
7834          (LD1Rv4h GPR64sp:$Rn)>;
7835def : Pat<(v8i16 (AArch64duplane16 (v8i16 (load GPR64sp:$Rn)), (i64 0))),
7836          (LD1Rv8h GPR64sp:$Rn)>;
7837def : Pat<(v2i32 (AArch64duplane32 (v4i32 (insert_subvector undef, (v2i32 (load GPR64sp:$Rn)), (i64 0))), (i64 0))),
7838          (LD1Rv2s GPR64sp:$Rn)>;
7839def : Pat<(v4i32 (AArch64duplane32 (v4i32 (load GPR64sp:$Rn)), (i64 0))),
7840          (LD1Rv4s GPR64sp:$Rn)>;
7841def : Pat<(v2i64 (AArch64duplane64 (v2i64 (load GPR64sp:$Rn)), (i64 0))),
7842          (LD1Rv2d GPR64sp:$Rn)>;
7843
7844// Grab the floating point version too
7845def : Pat<(v2f32 (AArch64dup (f32 (load GPR64sp:$Rn)))),
7846          (LD1Rv2s GPR64sp:$Rn)>;
7847def : Pat<(v4f32 (AArch64dup (f32 (load GPR64sp:$Rn)))),
7848          (LD1Rv4s GPR64sp:$Rn)>;
7849def : Pat<(v2f64 (AArch64dup (f64 (load GPR64sp:$Rn)))),
7850          (LD1Rv2d GPR64sp:$Rn)>;
7851def : Pat<(v1f64 (AArch64dup (f64 (load GPR64sp:$Rn)))),
7852          (LD1Rv1d GPR64sp:$Rn)>;
7853def : Pat<(v4f16 (AArch64dup (f16 (load GPR64sp:$Rn)))),
7854          (LD1Rv4h GPR64sp:$Rn)>;
7855def : Pat<(v8f16 (AArch64dup (f16 (load GPR64sp:$Rn)))),
7856          (LD1Rv8h GPR64sp:$Rn)>;
7857def : Pat<(v4bf16 (AArch64dup (bf16 (load GPR64sp:$Rn)))),
7858          (LD1Rv4h GPR64sp:$Rn)>;
7859def : Pat<(v8bf16 (AArch64dup (bf16 (load GPR64sp:$Rn)))),
7860          (LD1Rv8h GPR64sp:$Rn)>;
7861
7862class Ld1Lane128Pat<SDPatternOperator scalar_load, Operand VecIndex,
7863                    ValueType VTy, ValueType STy, Instruction LD1>
7864  : Pat<(vector_insert (VTy VecListOne128:$Rd),
7865           (STy (scalar_load GPR64sp:$Rn)), VecIndex:$idx),
7866        (LD1 VecListOne128:$Rd, VecIndex:$idx, GPR64sp:$Rn)>;
7867
7868def : Ld1Lane128Pat<extloadi8,  VectorIndexB, v16i8, i32, LD1i8>;
7869def : Ld1Lane128Pat<extloadi16, VectorIndexH, v8i16, i32, LD1i16>;
7870def : Ld1Lane128Pat<load,       VectorIndexS, v4i32, i32, LD1i32>;
7871def : Ld1Lane128Pat<load,       VectorIndexS, v4f32, f32, LD1i32>;
7872def : Ld1Lane128Pat<load,       VectorIndexD, v2i64, i64, LD1i64>;
7873def : Ld1Lane128Pat<load,       VectorIndexD, v2f64, f64, LD1i64>;
7874def : Ld1Lane128Pat<load,       VectorIndexH, v8f16, f16, LD1i16>;
7875def : Ld1Lane128Pat<load,       VectorIndexH, v8bf16, bf16, LD1i16>;
7876
7877// Generate LD1 for extload if memory type does not match the
7878// destination type, for example:
7879//
7880//   (v4i32 (insert_vector_elt (load anyext from i8) idx))
7881//
7882// In this case, the index must be adjusted to match LD1 type.
7883//
7884class Ld1Lane128IdxOpPat<SDPatternOperator scalar_load, Operand
7885                         VecIndex, ValueType VTy, ValueType STy,
7886                         Instruction LD1, SDNodeXForm IdxOp>
7887  : Pat<(vector_insert (VTy VecListOne128:$Rd),
7888                       (STy (scalar_load GPR64sp:$Rn)), VecIndex:$idx),
7889        (LD1 VecListOne128:$Rd, (IdxOp VecIndex:$idx), GPR64sp:$Rn)>;
7890
7891class Ld1Lane64IdxOpPat<SDPatternOperator scalar_load, Operand VecIndex,
7892                        ValueType VTy, ValueType STy, Instruction LD1,
7893                        SDNodeXForm IdxOp>
7894  : Pat<(vector_insert (VTy VecListOne64:$Rd),
7895                       (STy (scalar_load GPR64sp:$Rn)), VecIndex:$idx),
7896        (EXTRACT_SUBREG
7897            (LD1 (SUBREG_TO_REG (i32 0), VecListOne64:$Rd, dsub),
7898                (IdxOp VecIndex:$idx), GPR64sp:$Rn),
7899            dsub)>;
7900
7901def VectorIndexStoH : SDNodeXForm<imm, [{
7902  return CurDAG->getTargetConstant(N->getZExtValue() * 2, SDLoc(N), MVT::i64);
7903}]>;
7904def VectorIndexStoB : SDNodeXForm<imm, [{
7905  return CurDAG->getTargetConstant(N->getZExtValue() * 4, SDLoc(N), MVT::i64);
7906}]>;
7907def VectorIndexHtoB : SDNodeXForm<imm, [{
7908  return CurDAG->getTargetConstant(N->getZExtValue() * 2, SDLoc(N), MVT::i64);
7909}]>;
7910
7911def : Ld1Lane128IdxOpPat<extloadi16, VectorIndexS, v4i32, i32, LD1i16, VectorIndexStoH>;
7912def : Ld1Lane128IdxOpPat<extloadi8, VectorIndexS, v4i32, i32, LD1i8, VectorIndexStoB>;
7913def : Ld1Lane128IdxOpPat<extloadi8, VectorIndexH, v8i16, i32, LD1i8, VectorIndexHtoB>;
7914
7915def : Ld1Lane64IdxOpPat<extloadi16, VectorIndexS, v2i32, i32, LD1i16, VectorIndexStoH>;
7916def : Ld1Lane64IdxOpPat<extloadi8, VectorIndexS, v2i32, i32, LD1i8, VectorIndexStoB>;
7917def : Ld1Lane64IdxOpPat<extloadi8, VectorIndexH, v4i16, i32, LD1i8, VectorIndexHtoB>;
7918
7919// Same as above, but the first element is populated using
7920// scalar_to_vector + insert_subvector instead of insert_vector_elt.
7921let Predicates = [IsNeonAvailable] in {
7922  class Ld1Lane128FirstElm<ValueType ResultTy, ValueType VecTy,
7923                          SDPatternOperator ExtLoad, Instruction LD1>
7924    : Pat<(ResultTy (scalar_to_vector (i32 (ExtLoad GPR64sp:$Rn)))),
7925            (ResultTy (EXTRACT_SUBREG
7926              (LD1 (VecTy (IMPLICIT_DEF)), 0, GPR64sp:$Rn), dsub))>;
7927
7928  def : Ld1Lane128FirstElm<v2i32, v8i16, extloadi16, LD1i16>;
7929  def : Ld1Lane128FirstElm<v2i32, v16i8, extloadi8, LD1i8>;
7930  def : Ld1Lane128FirstElm<v4i16, v16i8, extloadi8, LD1i8>;
7931}
7932class Ld1Lane64Pat<SDPatternOperator scalar_load, Operand VecIndex,
7933                   ValueType VTy, ValueType STy, Instruction LD1>
7934  : Pat<(vector_insert (VTy VecListOne64:$Rd),
7935           (STy (scalar_load GPR64sp:$Rn)), VecIndex:$idx),
7936        (EXTRACT_SUBREG
7937            (LD1 (SUBREG_TO_REG (i32 0), VecListOne64:$Rd, dsub),
7938                          VecIndex:$idx, GPR64sp:$Rn),
7939            dsub)>;
7940
7941def : Ld1Lane64Pat<extloadi8,  VectorIndexB, v8i8,  i32, LD1i8>;
7942def : Ld1Lane64Pat<extloadi16, VectorIndexH, v4i16, i32, LD1i16>;
7943def : Ld1Lane64Pat<load,       VectorIndexS, v2i32, i32, LD1i32>;
7944def : Ld1Lane64Pat<load,       VectorIndexS, v2f32, f32, LD1i32>;
7945def : Ld1Lane64Pat<load,       VectorIndexH, v4f16, f16, LD1i16>;
7946def : Ld1Lane64Pat<load,       VectorIndexH, v4bf16, bf16, LD1i16>;
7947
7948
7949defm LD1 : SIMDLdSt1SingleAliases<"ld1">;
7950defm LD2 : SIMDLdSt2SingleAliases<"ld2">;
7951defm LD3 : SIMDLdSt3SingleAliases<"ld3">;
7952defm LD4 : SIMDLdSt4SingleAliases<"ld4">;
7953
7954// Stores
7955defm ST1 : SIMDStSingleB<0, 0b000,       "st1", VecListOneb, GPR64pi1>;
7956defm ST1 : SIMDStSingleH<0, 0b010, 0,    "st1", VecListOneh, GPR64pi2>;
7957defm ST1 : SIMDStSingleS<0, 0b100, 0b00, "st1", VecListOnes, GPR64pi4>;
7958defm ST1 : SIMDStSingleD<0, 0b100, 0b01, "st1", VecListOned, GPR64pi8>;
7959
7960let AddedComplexity = 19 in
7961class St1Lane128Pat<SDPatternOperator scalar_store, Operand VecIndex,
7962                    ValueType VTy, ValueType STy, Instruction ST1>
7963  : Pat<(scalar_store
7964             (STy (vector_extract (VTy VecListOne128:$Vt), VecIndex:$idx)),
7965             GPR64sp:$Rn),
7966        (ST1 VecListOne128:$Vt, VecIndex:$idx, GPR64sp:$Rn)>;
7967
7968def : St1Lane128Pat<truncstorei8,  VectorIndexB, v16i8, i32, ST1i8>;
7969def : St1Lane128Pat<truncstorei16, VectorIndexH, v8i16, i32, ST1i16>;
7970def : St1Lane128Pat<store,         VectorIndexS, v4i32, i32, ST1i32>;
7971def : St1Lane128Pat<store,         VectorIndexS, v4f32, f32, ST1i32>;
7972def : St1Lane128Pat<store,         VectorIndexD, v2i64, i64, ST1i64>;
7973def : St1Lane128Pat<store,         VectorIndexD, v2f64, f64, ST1i64>;
7974def : St1Lane128Pat<store,         VectorIndexH, v8f16, f16, ST1i16>;
7975def : St1Lane128Pat<store,         VectorIndexH, v8bf16, bf16, ST1i16>;
7976
7977let AddedComplexity = 19 in
7978class St1Lane64Pat<SDPatternOperator scalar_store, Operand VecIndex,
7979                   ValueType VTy, ValueType STy, Instruction ST1>
7980  : Pat<(scalar_store
7981             (STy (vector_extract (VTy VecListOne64:$Vt), VecIndex:$idx)),
7982             GPR64sp:$Rn),
7983        (ST1 (SUBREG_TO_REG (i32 0), VecListOne64:$Vt, dsub),
7984             VecIndex:$idx, GPR64sp:$Rn)>;
7985
7986def : St1Lane64Pat<truncstorei8,  VectorIndexB, v8i8, i32, ST1i8>;
7987def : St1Lane64Pat<truncstorei16, VectorIndexH, v4i16, i32, ST1i16>;
7988def : St1Lane64Pat<store,         VectorIndexS, v2i32, i32, ST1i32>;
7989def : St1Lane64Pat<store,         VectorIndexS, v2f32, f32, ST1i32>;
7990def : St1Lane64Pat<store,         VectorIndexH, v4f16, f16, ST1i16>;
7991def : St1Lane64Pat<store,         VectorIndexH, v4bf16, bf16, ST1i16>;
7992
7993multiclass St1LanePost64Pat<SDPatternOperator scalar_store, Operand VecIndex,
7994                             ValueType VTy, ValueType STy, Instruction ST1,
7995                             int offset> {
7996  def : Pat<(scalar_store
7997              (STy (vector_extract (VTy VecListOne64:$Vt), VecIndex:$idx)),
7998              GPR64sp:$Rn, offset),
7999        (ST1 (SUBREG_TO_REG (i32 0), VecListOne64:$Vt, dsub),
8000             VecIndex:$idx, GPR64sp:$Rn, XZR)>;
8001
8002  def : Pat<(scalar_store
8003              (STy (vector_extract (VTy VecListOne64:$Vt), VecIndex:$idx)),
8004              GPR64sp:$Rn, GPR64:$Rm),
8005        (ST1 (SUBREG_TO_REG (i32 0), VecListOne64:$Vt, dsub),
8006             VecIndex:$idx, GPR64sp:$Rn, $Rm)>;
8007}
8008
8009defm : St1LanePost64Pat<post_truncsti8, VectorIndexB, v8i8, i32, ST1i8_POST, 1>;
8010defm : St1LanePost64Pat<post_truncsti16, VectorIndexH, v4i16, i32, ST1i16_POST,
8011                        2>;
8012defm : St1LanePost64Pat<post_store, VectorIndexS, v2i32, i32, ST1i32_POST, 4>;
8013defm : St1LanePost64Pat<post_store, VectorIndexS, v2f32, f32, ST1i32_POST, 4>;
8014defm : St1LanePost64Pat<post_store, VectorIndexD, v1i64, i64, ST1i64_POST, 8>;
8015defm : St1LanePost64Pat<post_store, VectorIndexD, v1f64, f64, ST1i64_POST, 8>;
8016defm : St1LanePost64Pat<post_store, VectorIndexH, v4f16, f16, ST1i16_POST, 2>;
8017defm : St1LanePost64Pat<post_store, VectorIndexH, v4bf16, bf16, ST1i16_POST, 2>;
8018
8019multiclass St1LanePost128Pat<SDPatternOperator scalar_store, Operand VecIndex,
8020                             ValueType VTy, ValueType STy, Instruction ST1,
8021                             int offset> {
8022  def : Pat<(scalar_store
8023              (STy (vector_extract (VTy VecListOne128:$Vt), VecIndex:$idx)),
8024              GPR64sp:$Rn, offset),
8025        (ST1 VecListOne128:$Vt, VecIndex:$idx, GPR64sp:$Rn, XZR)>;
8026
8027  def : Pat<(scalar_store
8028              (STy (vector_extract (VTy VecListOne128:$Vt), VecIndex:$idx)),
8029              GPR64sp:$Rn, GPR64:$Rm),
8030        (ST1 VecListOne128:$Vt, VecIndex:$idx, GPR64sp:$Rn, $Rm)>;
8031}
8032
8033defm : St1LanePost128Pat<post_truncsti8, VectorIndexB, v16i8, i32, ST1i8_POST,
8034                         1>;
8035defm : St1LanePost128Pat<post_truncsti16, VectorIndexH, v8i16, i32, ST1i16_POST,
8036                         2>;
8037defm : St1LanePost128Pat<post_store, VectorIndexS, v4i32, i32, ST1i32_POST, 4>;
8038defm : St1LanePost128Pat<post_store, VectorIndexS, v4f32, f32, ST1i32_POST, 4>;
8039defm : St1LanePost128Pat<post_store, VectorIndexD, v2i64, i64, ST1i64_POST, 8>;
8040defm : St1LanePost128Pat<post_store, VectorIndexD, v2f64, f64, ST1i64_POST, 8>;
8041defm : St1LanePost128Pat<post_store, VectorIndexH, v8f16, f16, ST1i16_POST, 2>;
8042defm : St1LanePost128Pat<post_store, VectorIndexH, v8bf16, bf16, ST1i16_POST, 2>;
8043
8044let mayStore = 1, hasSideEffects = 0 in {
8045defm ST2 : SIMDStSingleB<1, 0b000,       "st2", VecListTwob,   GPR64pi2>;
8046defm ST2 : SIMDStSingleH<1, 0b010, 0,    "st2", VecListTwoh,   GPR64pi4>;
8047defm ST2 : SIMDStSingleS<1, 0b100, 0b00, "st2", VecListTwos,   GPR64pi8>;
8048defm ST2 : SIMDStSingleD<1, 0b100, 0b01, "st2", VecListTwod,   GPR64pi16>;
8049defm ST3 : SIMDStSingleB<0, 0b001,       "st3", VecListThreeb, GPR64pi3>;
8050defm ST3 : SIMDStSingleH<0, 0b011, 0,    "st3", VecListThreeh, GPR64pi6>;
8051defm ST3 : SIMDStSingleS<0, 0b101, 0b00, "st3", VecListThrees, GPR64pi12>;
8052defm ST3 : SIMDStSingleD<0, 0b101, 0b01, "st3", VecListThreed, GPR64pi24>;
8053defm ST4 : SIMDStSingleB<1, 0b001,       "st4", VecListFourb,  GPR64pi4>;
8054defm ST4 : SIMDStSingleH<1, 0b011, 0,    "st4", VecListFourh,  GPR64pi8>;
8055defm ST4 : SIMDStSingleS<1, 0b101, 0b00, "st4", VecListFours,  GPR64pi16>;
8056defm ST4 : SIMDStSingleD<1, 0b101, 0b01, "st4", VecListFourd,  GPR64pi32>;
8057}
8058
8059defm ST1 : SIMDLdSt1SingleAliases<"st1">;
8060defm ST2 : SIMDLdSt2SingleAliases<"st2">;
8061defm ST3 : SIMDLdSt3SingleAliases<"st3">;
8062defm ST4 : SIMDLdSt4SingleAliases<"st4">;
8063
8064//----------------------------------------------------------------------------
8065// Crypto extensions
8066//----------------------------------------------------------------------------
8067
8068let Predicates = [HasAES] in {
8069def AESErr   : AESTiedInst<0b0100, "aese",   int_aarch64_crypto_aese>;
8070def AESDrr   : AESTiedInst<0b0101, "aesd",   int_aarch64_crypto_aesd>;
8071def AESMCrr  : AESInst<    0b0110, "aesmc",  int_aarch64_crypto_aesmc>;
8072def AESIMCrr : AESInst<    0b0111, "aesimc", int_aarch64_crypto_aesimc>;
8073}
8074
8075// Pseudo instructions for AESMCrr/AESIMCrr with a register constraint required
8076// for AES fusion on some CPUs.
8077let hasSideEffects = 0, mayStore = 0, mayLoad = 0 in {
8078def AESMCrrTied: Pseudo<(outs V128:$Rd), (ins V128:$Rn), [], "$Rn = $Rd">,
8079                        Sched<[WriteVq]>;
8080def AESIMCrrTied: Pseudo<(outs V128:$Rd), (ins V128:$Rn), [], "$Rn = $Rd">,
8081                         Sched<[WriteVq]>;
8082}
8083
8084// Only use constrained versions of AES(I)MC instructions if they are paired with
8085// AESE/AESD.
8086def : Pat<(v16i8 (int_aarch64_crypto_aesmc
8087            (v16i8 (int_aarch64_crypto_aese (v16i8 V128:$src1),
8088                                            (v16i8 V128:$src2))))),
8089          (v16i8 (AESMCrrTied (v16i8 (AESErr (v16i8 V128:$src1),
8090                                             (v16i8 V128:$src2)))))>,
8091          Requires<[HasFuseAES]>;
8092
8093def : Pat<(v16i8 (int_aarch64_crypto_aesimc
8094            (v16i8 (int_aarch64_crypto_aesd (v16i8 V128:$src1),
8095                                            (v16i8 V128:$src2))))),
8096          (v16i8 (AESIMCrrTied (v16i8 (AESDrr (v16i8 V128:$src1),
8097                                              (v16i8 V128:$src2)))))>,
8098          Requires<[HasFuseAES]>;
8099
8100let Predicates = [HasSHA2] in {
8101def SHA1Crrr     : SHATiedInstQSV<0b000, "sha1c",   int_aarch64_crypto_sha1c>;
8102def SHA1Prrr     : SHATiedInstQSV<0b001, "sha1p",   int_aarch64_crypto_sha1p>;
8103def SHA1Mrrr     : SHATiedInstQSV<0b010, "sha1m",   int_aarch64_crypto_sha1m>;
8104def SHA1SU0rrr   : SHATiedInstVVV<0b011, "sha1su0", int_aarch64_crypto_sha1su0>;
8105def SHA256Hrrr   : SHATiedInstQQV<0b100, "sha256h", int_aarch64_crypto_sha256h>;
8106def SHA256H2rrr  : SHATiedInstQQV<0b101, "sha256h2",int_aarch64_crypto_sha256h2>;
8107def SHA256SU1rrr :SHATiedInstVVV<0b110, "sha256su1",int_aarch64_crypto_sha256su1>;
8108
8109def SHA1Hrr     : SHAInstSS<    0b0000, "sha1h",    int_aarch64_crypto_sha1h>;
8110def SHA1SU1rr   : SHATiedInstVV<0b0001, "sha1su1",  int_aarch64_crypto_sha1su1>;
8111def SHA256SU0rr : SHATiedInstVV<0b0010, "sha256su0",int_aarch64_crypto_sha256su0>;
8112}
8113
8114//----------------------------------------------------------------------------
8115// Compiler-pseudos
8116//----------------------------------------------------------------------------
8117// FIXME: Like for X86, these should go in their own separate .td file.
8118
8119// For an anyext, we don't care what the high bits are, so we can perform an
8120// INSERT_SUBREF into an IMPLICIT_DEF.
8121def : Pat<(i64 (anyext GPR32:$src)),
8122          (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32)>;
8123
8124// When we need to explicitly zero-extend, we use a 32-bit MOV instruction and
8125// then assert the extension has happened.
8126def : Pat<(i64 (zext GPR32:$src)),
8127          (SUBREG_TO_REG (i32 0), (ORRWrs WZR, GPR32:$src, 0), sub_32)>;
8128
8129// To sign extend, we use a signed bitfield move instruction (SBFM) on the
8130// containing super-reg.
8131def : Pat<(i64 (sext GPR32:$src)),
8132   (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32), 0, 31)>;
8133def : Pat<(i64 (sext_inreg GPR64:$src, i32)), (SBFMXri GPR64:$src, 0, 31)>;
8134def : Pat<(i64 (sext_inreg GPR64:$src, i16)), (SBFMXri GPR64:$src, 0, 15)>;
8135def : Pat<(i64 (sext_inreg GPR64:$src, i8)),  (SBFMXri GPR64:$src, 0, 7)>;
8136def : Pat<(i64 (sext_inreg GPR64:$src, i1)),  (SBFMXri GPR64:$src, 0, 0)>;
8137def : Pat<(i32 (sext_inreg GPR32:$src, i16)), (SBFMWri GPR32:$src, 0, 15)>;
8138def : Pat<(i32 (sext_inreg GPR32:$src, i8)),  (SBFMWri GPR32:$src, 0, 7)>;
8139def : Pat<(i32 (sext_inreg GPR32:$src, i1)),  (SBFMWri GPR32:$src, 0, 0)>;
8140
8141def : Pat<(shl (sext_inreg GPR32:$Rn, i8), (i64 imm0_31:$imm)),
8142          (SBFMWri GPR32:$Rn, (i64 (i32shift_a       imm0_31:$imm)),
8143                              (i64 (i32shift_sext_i8 imm0_31:$imm)))>;
8144def : Pat<(shl (sext_inreg GPR64:$Rn, i8), (i64 imm0_63:$imm)),
8145          (SBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
8146                              (i64 (i64shift_sext_i8 imm0_63:$imm)))>;
8147
8148def : Pat<(shl (sext_inreg GPR32:$Rn, i16), (i64 imm0_31:$imm)),
8149          (SBFMWri GPR32:$Rn, (i64 (i32shift_a        imm0_31:$imm)),
8150                              (i64 (i32shift_sext_i16 imm0_31:$imm)))>;
8151def : Pat<(shl (sext_inreg GPR64:$Rn, i16), (i64 imm0_63:$imm)),
8152          (SBFMXri GPR64:$Rn, (i64 (i64shift_a        imm0_63:$imm)),
8153                              (i64 (i64shift_sext_i16 imm0_63:$imm)))>;
8154
8155def : Pat<(shl (i64 (sext GPR32:$Rn)), (i64 imm0_63:$imm)),
8156          (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$Rn, sub_32),
8157                   (i64 (i64shift_a        imm0_63:$imm)),
8158                   (i64 (i64shift_sext_i32 imm0_63:$imm)))>;
8159
8160def : Pat<(shl (i64 (zext GPR32:$Rn)), (i64 imm0_63:$imm)),
8161          (UBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$Rn, sub_32),
8162                   (i64 (i64shift_a        imm0_63:$imm)),
8163                   (i64 (i64shift_sext_i32 imm0_63:$imm)))>;
8164
8165// sra patterns have an AddedComplexity of 10, so make sure we have a higher
8166// AddedComplexity for the following patterns since we want to match sext + sra
8167// patterns before we attempt to match a single sra node.
8168let AddedComplexity = 20 in {
8169// We support all sext + sra combinations which preserve at least one bit of the
8170// original value which is to be sign extended. E.g. we support shifts up to
8171// bitwidth-1 bits.
8172def : Pat<(sra (sext_inreg GPR32:$Rn, i8), (i64 imm0_7:$imm)),
8173          (SBFMWri GPR32:$Rn, (i64 imm0_7:$imm), 7)>;
8174def : Pat<(sra (sext_inreg GPR64:$Rn, i8), (i64 imm0_7:$imm)),
8175          (SBFMXri GPR64:$Rn, (i64 imm0_7:$imm), 7)>;
8176
8177def : Pat<(sra (sext_inreg GPR32:$Rn, i16), (i64 imm0_15:$imm)),
8178          (SBFMWri GPR32:$Rn, (i64 imm0_15:$imm), 15)>;
8179def : Pat<(sra (sext_inreg GPR64:$Rn, i16), (i64 imm0_15:$imm)),
8180          (SBFMXri GPR64:$Rn, (i64 imm0_15:$imm), 15)>;
8181
8182def : Pat<(sra (i64 (sext GPR32:$Rn)), (i64 imm0_31:$imm)),
8183          (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$Rn, sub_32),
8184                   (i64 imm0_31:$imm), 31)>;
8185} // AddedComplexity = 20
8186
8187// To truncate, we can simply extract from a subregister.
8188def : Pat<(i32 (trunc GPR64sp:$src)),
8189          (i32 (EXTRACT_SUBREG GPR64sp:$src, sub_32))>;
8190
8191// __builtin_trap() uses the BRK instruction on AArch64.
8192def : Pat<(trap), (BRK 1)>;
8193def : Pat<(debugtrap), (BRK 0xF000)>;
8194
8195def ubsan_trap_xform : SDNodeXForm<timm, [{
8196  return CurDAG->getTargetConstant(N->getZExtValue() | ('U' << 8), SDLoc(N), MVT::i32);
8197}]>;
8198
8199def ubsan_trap_imm : TImmLeaf<i32, [{
8200  return isUInt<8>(Imm);
8201}], ubsan_trap_xform>;
8202
8203def : Pat<(ubsantrap ubsan_trap_imm:$kind), (BRK ubsan_trap_imm:$kind)>;
8204
8205// Multiply high patterns which multiply the lower subvector using smull/umull
8206// and the upper subvector with smull2/umull2. Then shuffle the high the high
8207// part of both results together.
8208def : Pat<(v16i8 (mulhs V128:$Rn, V128:$Rm)),
8209          (UZP2v16i8
8210           (SMULLv8i8_v8i16 (EXTRACT_SUBREG V128:$Rn, dsub),
8211                            (EXTRACT_SUBREG V128:$Rm, dsub)),
8212           (SMULLv16i8_v8i16 V128:$Rn, V128:$Rm))>;
8213def : Pat<(v8i16 (mulhs V128:$Rn, V128:$Rm)),
8214          (UZP2v8i16
8215           (SMULLv4i16_v4i32 (EXTRACT_SUBREG V128:$Rn, dsub),
8216                             (EXTRACT_SUBREG V128:$Rm, dsub)),
8217           (SMULLv8i16_v4i32 V128:$Rn, V128:$Rm))>;
8218def : Pat<(v4i32 (mulhs V128:$Rn, V128:$Rm)),
8219          (UZP2v4i32
8220           (SMULLv2i32_v2i64 (EXTRACT_SUBREG V128:$Rn, dsub),
8221                             (EXTRACT_SUBREG V128:$Rm, dsub)),
8222           (SMULLv4i32_v2i64 V128:$Rn, V128:$Rm))>;
8223
8224def : Pat<(v16i8 (mulhu V128:$Rn, V128:$Rm)),
8225          (UZP2v16i8
8226           (UMULLv8i8_v8i16 (EXTRACT_SUBREG V128:$Rn, dsub),
8227                            (EXTRACT_SUBREG V128:$Rm, dsub)),
8228           (UMULLv16i8_v8i16 V128:$Rn, V128:$Rm))>;
8229def : Pat<(v8i16 (mulhu V128:$Rn, V128:$Rm)),
8230          (UZP2v8i16
8231           (UMULLv4i16_v4i32 (EXTRACT_SUBREG V128:$Rn, dsub),
8232                             (EXTRACT_SUBREG V128:$Rm, dsub)),
8233           (UMULLv8i16_v4i32 V128:$Rn, V128:$Rm))>;
8234def : Pat<(v4i32 (mulhu V128:$Rn, V128:$Rm)),
8235          (UZP2v4i32
8236           (UMULLv2i32_v2i64 (EXTRACT_SUBREG V128:$Rn, dsub),
8237                             (EXTRACT_SUBREG V128:$Rm, dsub)),
8238           (UMULLv4i32_v2i64 V128:$Rn, V128:$Rm))>;
8239
8240// Conversions within AdvSIMD types in the same register size are free.
8241// But because we need a consistent lane ordering, in big endian many
8242// conversions require one or more REV instructions.
8243//
8244// Consider a simple memory load followed by a bitconvert then a store.
8245//   v0 = load v2i32
8246//   v1 = BITCAST v2i32 v0 to v4i16
8247//        store v4i16 v2
8248//
8249// In big endian mode every memory access has an implicit byte swap. LDR and
8250// STR do a 64-bit byte swap, whereas LD1/ST1 do a byte swap per lane - that
8251// is, they treat the vector as a sequence of elements to be byte-swapped.
8252// The two pairs of instructions are fundamentally incompatible. We've decided
8253// to use LD1/ST1 only to simplify compiler implementation.
8254//
8255// LD1/ST1 perform the equivalent of a sequence of LDR/STR + REV. This makes
8256// the original code sequence:
8257//   v0 = load v2i32
8258//   v1 = REV v2i32                  (implicit)
8259//   v2 = BITCAST v2i32 v1 to v4i16
8260//   v3 = REV v4i16 v2               (implicit)
8261//        store v4i16 v3
8262//
8263// But this is now broken - the value stored is different to the value loaded
8264// due to lane reordering. To fix this, on every BITCAST we must perform two
8265// other REVs:
8266//   v0 = load v2i32
8267//   v1 = REV v2i32                  (implicit)
8268//   v2 = REV v2i32
8269//   v3 = BITCAST v2i32 v2 to v4i16
8270//   v4 = REV v4i16
8271//   v5 = REV v4i16 v4               (implicit)
8272//        store v4i16 v5
8273//
8274// This means an extra two instructions, but actually in most cases the two REV
8275// instructions can be combined into one. For example:
8276//   (REV64_2s (REV64_4h X)) === (REV32_4h X)
8277//
8278// There is also no 128-bit REV instruction. This must be synthesized with an
8279// EXT instruction.
8280//
8281// Most bitconverts require some sort of conversion. The only exceptions are:
8282//   a) Identity conversions -  vNfX <-> vNiX
8283//   b) Single-lane-to-scalar - v1fX <-> fX or v1iX <-> iX
8284//
8285
8286// Natural vector casts (64 bit)
8287foreach VT = [ v8i8, v4i16, v4f16, v4bf16, v2i32, v2f32, v1i64, v1f64, f64 ] in
8288  foreach VT2 = [ v8i8, v4i16, v4f16, v4bf16, v2i32, v2f32, v1i64, v1f64, f64 ] in
8289    def : Pat<(VT (AArch64NvCast (VT2 FPR64:$src))),
8290              (VT FPR64:$src)>;
8291
8292// Natural vector casts (128 bit)
8293foreach VT = [ v16i8, v8i16, v8f16, v8bf16, v4i32, v4f32, v2i64, v2f64 ] in
8294  foreach VT2 = [ v16i8, v8i16, v8f16, v8bf16, v4i32, v4f32, v2i64, v2f64 ] in
8295    def : Pat<(VT (AArch64NvCast (VT2 FPR128:$src))),
8296              (VT FPR128:$src)>;
8297
8298let Predicates = [IsLE] in {
8299def : Pat<(v8i8  (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
8300def : Pat<(v4i16 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
8301def : Pat<(v2i32 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
8302def : Pat<(v4f16 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
8303def : Pat<(v4bf16 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
8304def : Pat<(v2f32 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
8305
8306def : Pat<(i64 (bitconvert (v8i8  V64:$Vn))),
8307          (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
8308def : Pat<(i64 (bitconvert (v4i16 V64:$Vn))),
8309          (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
8310def : Pat<(i64 (bitconvert (v2i32 V64:$Vn))),
8311          (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
8312def : Pat<(i64 (bitconvert (v4f16 V64:$Vn))),
8313          (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
8314def : Pat<(i64 (bitconvert (v4bf16 V64:$Vn))),
8315          (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
8316def : Pat<(i64 (bitconvert (v2f32 V64:$Vn))),
8317          (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
8318def : Pat<(i64 (bitconvert (v1f64 V64:$Vn))),
8319          (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
8320}
8321let Predicates = [IsBE] in {
8322def : Pat<(v8i8  (bitconvert GPR64:$Xn)),
8323                 (REV64v8i8 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
8324def : Pat<(v4i16 (bitconvert GPR64:$Xn)),
8325                 (REV64v4i16 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
8326def : Pat<(v2i32 (bitconvert GPR64:$Xn)),
8327                 (REV64v2i32 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
8328def : Pat<(v4f16 (bitconvert GPR64:$Xn)),
8329                 (REV64v4i16 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
8330def : Pat<(v4bf16 (bitconvert GPR64:$Xn)),
8331                  (REV64v4i16 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
8332def : Pat<(v2f32 (bitconvert GPR64:$Xn)),
8333                 (REV64v2i32 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
8334
8335def : Pat<(i64 (bitconvert (v8i8  V64:$Vn))),
8336          (REV64v8i8 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
8337def : Pat<(i64 (bitconvert (v4i16 V64:$Vn))),
8338          (REV64v4i16 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
8339def : Pat<(i64 (bitconvert (v2i32 V64:$Vn))),
8340          (REV64v2i32 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
8341def : Pat<(i64 (bitconvert (v4f16 V64:$Vn))),
8342          (REV64v4i16 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
8343def : Pat<(i64 (bitconvert (v4bf16 V64:$Vn))),
8344          (REV64v4i16 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
8345def : Pat<(i64 (bitconvert (v2f32 V64:$Vn))),
8346          (REV64v2i32 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
8347}
8348def : Pat<(v1i64 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
8349def : Pat<(v1f64 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
8350def : Pat<(i64 (bitconvert (v1i64 V64:$Vn))),
8351          (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
8352def : Pat<(v1i64 (scalar_to_vector GPR64:$Xn)),
8353          (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
8354def : Pat<(v1f64 (scalar_to_vector GPR64:$Xn)),
8355          (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
8356def : Pat<(v1f64 (scalar_to_vector (f64 FPR64:$Xn))), (v1f64 FPR64:$Xn)>;
8357
8358def : Pat<(f32 (bitconvert (i32 GPR32:$Xn))),
8359          (COPY_TO_REGCLASS GPR32:$Xn, FPR32)>;
8360def : Pat<(i32 (bitconvert (f32 FPR32:$Xn))),
8361          (COPY_TO_REGCLASS FPR32:$Xn, GPR32)>;
8362def : Pat<(f64 (bitconvert (i64 GPR64:$Xn))),
8363          (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
8364def : Pat<(i64 (bitconvert (f64 FPR64:$Xn))),
8365          (COPY_TO_REGCLASS FPR64:$Xn, GPR64)>;
8366def : Pat<(i64 (bitconvert (v1f64 V64:$Vn))),
8367          (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
8368
8369def : Pat<(f16 (bitconvert (bf16 FPR16:$src))), (f16 FPR16:$src)>;
8370def : Pat<(bf16 (bitconvert (f16 FPR16:$src))), (bf16 FPR16:$src)>;
8371
8372let Predicates = [IsLE] in {
8373def : Pat<(v1i64 (bitconvert (v2i32 FPR64:$src))), (v1i64 FPR64:$src)>;
8374def : Pat<(v1i64 (bitconvert (v4i16 FPR64:$src))), (v1i64 FPR64:$src)>;
8375def : Pat<(v1i64 (bitconvert (v8i8  FPR64:$src))), (v1i64 FPR64:$src)>;
8376def : Pat<(v1i64 (bitconvert (v4f16 FPR64:$src))), (v1i64 FPR64:$src)>;
8377def : Pat<(v1i64 (bitconvert (v4bf16 FPR64:$src))), (v1i64 FPR64:$src)>;
8378def : Pat<(v1i64 (bitconvert (v2f32 FPR64:$src))), (v1i64 FPR64:$src)>;
8379}
8380let Predicates = [IsBE] in {
8381def : Pat<(v1i64 (bitconvert (v2i32 FPR64:$src))),
8382                             (v1i64 (REV64v2i32 FPR64:$src))>;
8383def : Pat<(v1i64 (bitconvert (v4i16 FPR64:$src))),
8384                             (v1i64 (REV64v4i16 FPR64:$src))>;
8385def : Pat<(v1i64 (bitconvert (v8i8  FPR64:$src))),
8386                             (v1i64 (REV64v8i8 FPR64:$src))>;
8387def : Pat<(v1i64 (bitconvert (v4f16 FPR64:$src))),
8388                             (v1i64 (REV64v4i16 FPR64:$src))>;
8389def : Pat<(v1i64 (bitconvert (v4bf16 FPR64:$src))),
8390                             (v1i64 (REV64v4i16 FPR64:$src))>;
8391def : Pat<(v1i64 (bitconvert (v2f32 FPR64:$src))),
8392                             (v1i64 (REV64v2i32 FPR64:$src))>;
8393}
8394def : Pat<(v1i64 (bitconvert (v1f64 FPR64:$src))), (v1i64 FPR64:$src)>;
8395def : Pat<(v1i64 (bitconvert (f64   FPR64:$src))), (v1i64 FPR64:$src)>;
8396
8397let Predicates = [IsLE] in {
8398def : Pat<(v2i32 (bitconvert (v1i64 FPR64:$src))), (v2i32 FPR64:$src)>;
8399def : Pat<(v2i32 (bitconvert (v4i16 FPR64:$src))), (v2i32 FPR64:$src)>;
8400def : Pat<(v2i32 (bitconvert (v8i8  FPR64:$src))), (v2i32 FPR64:$src)>;
8401def : Pat<(v2i32 (bitconvert (f64   FPR64:$src))), (v2i32 FPR64:$src)>;
8402def : Pat<(v2i32 (bitconvert (v1f64 FPR64:$src))), (v2i32 FPR64:$src)>;
8403def : Pat<(v2i32 (bitconvert (v4f16 FPR64:$src))), (v2i32 FPR64:$src)>;
8404def : Pat<(v2i32 (bitconvert (v4bf16 FPR64:$src))), (v2i32 FPR64:$src)>;
8405}
8406let Predicates = [IsBE] in {
8407def : Pat<(v2i32 (bitconvert (v1i64 FPR64:$src))),
8408                             (v2i32 (REV64v2i32 FPR64:$src))>;
8409def : Pat<(v2i32 (bitconvert (v4i16 FPR64:$src))),
8410                             (v2i32 (REV32v4i16 FPR64:$src))>;
8411def : Pat<(v2i32 (bitconvert (v8i8  FPR64:$src))),
8412                             (v2i32 (REV32v8i8 FPR64:$src))>;
8413def : Pat<(v2i32 (bitconvert (f64   FPR64:$src))),
8414                             (v2i32 (REV64v2i32 FPR64:$src))>;
8415def : Pat<(v2i32 (bitconvert (v1f64 FPR64:$src))),
8416                             (v2i32 (REV64v2i32 FPR64:$src))>;
8417def : Pat<(v2i32 (bitconvert (v4f16 FPR64:$src))),
8418                             (v2i32 (REV32v4i16 FPR64:$src))>;
8419def : Pat<(v2i32 (bitconvert (v4bf16 FPR64:$src))),
8420                             (v2i32 (REV32v4i16 FPR64:$src))>;
8421}
8422def : Pat<(v2i32 (bitconvert (v2f32 FPR64:$src))), (v2i32 FPR64:$src)>;
8423
8424let Predicates = [IsLE] in {
8425def : Pat<(v4i16 (bitconvert (v1i64 FPR64:$src))), (v4i16 FPR64:$src)>;
8426def : Pat<(v4i16 (bitconvert (v2i32 FPR64:$src))), (v4i16 FPR64:$src)>;
8427def : Pat<(v4i16 (bitconvert (v8i8  FPR64:$src))), (v4i16 FPR64:$src)>;
8428def : Pat<(v4i16 (bitconvert (f64   FPR64:$src))), (v4i16 FPR64:$src)>;
8429def : Pat<(v4i16 (bitconvert (v2f32 FPR64:$src))), (v4i16 FPR64:$src)>;
8430def : Pat<(v4i16 (bitconvert (v1f64 FPR64:$src))), (v4i16 FPR64:$src)>;
8431}
8432let Predicates = [IsBE] in {
8433def : Pat<(v4i16 (bitconvert (v1i64 FPR64:$src))),
8434                             (v4i16 (REV64v4i16 FPR64:$src))>;
8435def : Pat<(v4i16 (bitconvert (v2i32 FPR64:$src))),
8436                             (v4i16 (REV32v4i16 FPR64:$src))>;
8437def : Pat<(v4i16 (bitconvert (v8i8  FPR64:$src))),
8438                             (v4i16 (REV16v8i8 FPR64:$src))>;
8439def : Pat<(v4i16 (bitconvert (f64   FPR64:$src))),
8440                             (v4i16 (REV64v4i16 FPR64:$src))>;
8441def : Pat<(v4i16 (bitconvert (v2f32 FPR64:$src))),
8442                             (v4i16 (REV32v4i16 FPR64:$src))>;
8443def : Pat<(v4i16 (bitconvert (v1f64 FPR64:$src))),
8444                             (v4i16 (REV64v4i16 FPR64:$src))>;
8445}
8446def : Pat<(v4i16 (bitconvert (v4f16 FPR64:$src))), (v4i16 FPR64:$src)>;
8447def : Pat<(v4i16 (bitconvert (v4bf16 FPR64:$src))), (v4i16 FPR64:$src)>;
8448
8449let Predicates = [IsLE] in {
8450def : Pat<(v4f16 (bitconvert (v1i64 FPR64:$src))), (v4f16 FPR64:$src)>;
8451def : Pat<(v4f16 (bitconvert (v2i32 FPR64:$src))), (v4f16 FPR64:$src)>;
8452def : Pat<(v4f16 (bitconvert (v8i8  FPR64:$src))), (v4f16 FPR64:$src)>;
8453def : Pat<(v4f16 (bitconvert (f64   FPR64:$src))), (v4f16 FPR64:$src)>;
8454def : Pat<(v4f16 (bitconvert (v2f32 FPR64:$src))), (v4f16 FPR64:$src)>;
8455def : Pat<(v4f16 (bitconvert (v1f64 FPR64:$src))), (v4f16 FPR64:$src)>;
8456
8457def : Pat<(v4bf16 (bitconvert (v1i64 FPR64:$src))), (v4bf16 FPR64:$src)>;
8458def : Pat<(v4bf16 (bitconvert (v2i32 FPR64:$src))), (v4bf16 FPR64:$src)>;
8459def : Pat<(v4bf16 (bitconvert (v8i8  FPR64:$src))), (v4bf16 FPR64:$src)>;
8460def : Pat<(v4bf16 (bitconvert (f64   FPR64:$src))), (v4bf16 FPR64:$src)>;
8461def : Pat<(v4bf16 (bitconvert (v2f32 FPR64:$src))), (v4bf16 FPR64:$src)>;
8462def : Pat<(v4bf16 (bitconvert (v1f64 FPR64:$src))), (v4bf16 FPR64:$src)>;
8463}
8464let Predicates = [IsBE] in {
8465def : Pat<(v4f16 (bitconvert (v1i64 FPR64:$src))),
8466                             (v4f16 (REV64v4i16 FPR64:$src))>;
8467def : Pat<(v4f16 (bitconvert (v2i32 FPR64:$src))),
8468                             (v4f16 (REV32v4i16 FPR64:$src))>;
8469def : Pat<(v4f16 (bitconvert (v8i8  FPR64:$src))),
8470                             (v4f16 (REV16v8i8 FPR64:$src))>;
8471def : Pat<(v4f16 (bitconvert (f64   FPR64:$src))),
8472                             (v4f16 (REV64v4i16 FPR64:$src))>;
8473def : Pat<(v4f16 (bitconvert (v2f32 FPR64:$src))),
8474                             (v4f16 (REV32v4i16 FPR64:$src))>;
8475def : Pat<(v4f16 (bitconvert (v1f64 FPR64:$src))),
8476                             (v4f16 (REV64v4i16 FPR64:$src))>;
8477
8478def : Pat<(v4bf16 (bitconvert (v1i64 FPR64:$src))),
8479                             (v4bf16 (REV64v4i16 FPR64:$src))>;
8480def : Pat<(v4bf16 (bitconvert (v2i32 FPR64:$src))),
8481                             (v4bf16 (REV32v4i16 FPR64:$src))>;
8482def : Pat<(v4bf16 (bitconvert (v8i8  FPR64:$src))),
8483                             (v4bf16 (REV16v8i8 FPR64:$src))>;
8484def : Pat<(v4bf16 (bitconvert (f64   FPR64:$src))),
8485                             (v4bf16 (REV64v4i16 FPR64:$src))>;
8486def : Pat<(v4bf16 (bitconvert (v2f32 FPR64:$src))),
8487                             (v4bf16 (REV32v4i16 FPR64:$src))>;
8488def : Pat<(v4bf16 (bitconvert (v1f64 FPR64:$src))),
8489                             (v4bf16 (REV64v4i16 FPR64:$src))>;
8490}
8491def : Pat<(v4f16 (bitconvert (v4i16 FPR64:$src))), (v4f16 FPR64:$src)>;
8492def : Pat<(v4bf16 (bitconvert (v4i16 FPR64:$src))), (v4bf16 FPR64:$src)>;
8493
8494let Predicates = [IsLE] in {
8495def : Pat<(v8i8  (bitconvert (v1i64 FPR64:$src))), (v8i8  FPR64:$src)>;
8496def : Pat<(v8i8  (bitconvert (v2i32 FPR64:$src))), (v8i8  FPR64:$src)>;
8497def : Pat<(v8i8  (bitconvert (v4i16 FPR64:$src))), (v8i8  FPR64:$src)>;
8498def : Pat<(v8i8  (bitconvert (f64   FPR64:$src))), (v8i8  FPR64:$src)>;
8499def : Pat<(v8i8  (bitconvert (v2f32 FPR64:$src))), (v8i8  FPR64:$src)>;
8500def : Pat<(v8i8  (bitconvert (v1f64 FPR64:$src))), (v8i8  FPR64:$src)>;
8501def : Pat<(v8i8  (bitconvert (v4f16 FPR64:$src))), (v8i8  FPR64:$src)>;
8502def : Pat<(v8i8  (bitconvert (v4bf16 FPR64:$src))), (v8i8  FPR64:$src)>;
8503}
8504let Predicates = [IsBE] in {
8505def : Pat<(v8i8  (bitconvert (v1i64 FPR64:$src))),
8506                             (v8i8 (REV64v8i8 FPR64:$src))>;
8507def : Pat<(v8i8  (bitconvert (v2i32 FPR64:$src))),
8508                             (v8i8 (REV32v8i8 FPR64:$src))>;
8509def : Pat<(v8i8  (bitconvert (v4i16 FPR64:$src))),
8510                             (v8i8 (REV16v8i8 FPR64:$src))>;
8511def : Pat<(v8i8  (bitconvert (f64   FPR64:$src))),
8512                             (v8i8 (REV64v8i8 FPR64:$src))>;
8513def : Pat<(v8i8  (bitconvert (v2f32 FPR64:$src))),
8514                             (v8i8 (REV32v8i8 FPR64:$src))>;
8515def : Pat<(v8i8  (bitconvert (v1f64 FPR64:$src))),
8516                             (v8i8 (REV64v8i8 FPR64:$src))>;
8517def : Pat<(v8i8  (bitconvert (v4f16 FPR64:$src))),
8518                             (v8i8 (REV16v8i8 FPR64:$src))>;
8519def : Pat<(v8i8  (bitconvert (v4bf16 FPR64:$src))),
8520                             (v8i8 (REV16v8i8 FPR64:$src))>;
8521}
8522
8523let Predicates = [IsLE] in {
8524def : Pat<(f64   (bitconvert (v2i32 FPR64:$src))), (f64   FPR64:$src)>;
8525def : Pat<(f64   (bitconvert (v4i16 FPR64:$src))), (f64   FPR64:$src)>;
8526def : Pat<(f64   (bitconvert (v2f32 FPR64:$src))), (f64   FPR64:$src)>;
8527def : Pat<(f64   (bitconvert (v8i8  FPR64:$src))), (f64   FPR64:$src)>;
8528def : Pat<(f64   (bitconvert (v4f16 FPR64:$src))), (f64   FPR64:$src)>;
8529def : Pat<(f64   (bitconvert (v4bf16 FPR64:$src))), (f64   FPR64:$src)>;
8530}
8531let Predicates = [IsBE] in {
8532def : Pat<(f64   (bitconvert (v2i32 FPR64:$src))),
8533                             (f64 (REV64v2i32 FPR64:$src))>;
8534def : Pat<(f64   (bitconvert (v4i16 FPR64:$src))),
8535                             (f64 (REV64v4i16 FPR64:$src))>;
8536def : Pat<(f64   (bitconvert (v2f32 FPR64:$src))),
8537                             (f64 (REV64v2i32 FPR64:$src))>;
8538def : Pat<(f64   (bitconvert (v8i8  FPR64:$src))),
8539                             (f64 (REV64v8i8 FPR64:$src))>;
8540def : Pat<(f64   (bitconvert (v4f16 FPR64:$src))),
8541                             (f64 (REV64v4i16 FPR64:$src))>;
8542def : Pat<(f64   (bitconvert (v4bf16 FPR64:$src))),
8543                             (f64 (REV64v4i16 FPR64:$src))>;
8544}
8545def : Pat<(f64   (bitconvert (v1i64 FPR64:$src))), (f64   FPR64:$src)>;
8546def : Pat<(f64   (bitconvert (v1f64 FPR64:$src))), (f64   FPR64:$src)>;
8547
8548let Predicates = [IsLE] in {
8549def : Pat<(v1f64 (bitconvert (v2i32 FPR64:$src))), (v1f64 FPR64:$src)>;
8550def : Pat<(v1f64 (bitconvert (v4i16 FPR64:$src))), (v1f64 FPR64:$src)>;
8551def : Pat<(v1f64 (bitconvert (v8i8  FPR64:$src))), (v1f64 FPR64:$src)>;
8552def : Pat<(v1f64 (bitconvert (v2f32 FPR64:$src))), (v1f64 FPR64:$src)>;
8553def : Pat<(v1f64 (bitconvert (v4f16 FPR64:$src))), (v1f64 FPR64:$src)>;
8554def : Pat<(v1f64 (bitconvert (v4bf16 FPR64:$src))), (v1f64 FPR64:$src)>;
8555}
8556let Predicates = [IsBE] in {
8557def : Pat<(v1f64 (bitconvert (v2i32 FPR64:$src))),
8558                             (v1f64 (REV64v2i32 FPR64:$src))>;
8559def : Pat<(v1f64 (bitconvert (v4i16 FPR64:$src))),
8560                             (v1f64 (REV64v4i16 FPR64:$src))>;
8561def : Pat<(v1f64 (bitconvert (v8i8  FPR64:$src))),
8562                             (v1f64 (REV64v8i8 FPR64:$src))>;
8563def : Pat<(v1f64 (bitconvert (v2f32 FPR64:$src))),
8564                             (v1f64 (REV64v2i32 FPR64:$src))>;
8565def : Pat<(v1f64 (bitconvert (v4f16 FPR64:$src))),
8566                             (v1f64 (REV64v4i16 FPR64:$src))>;
8567def : Pat<(v1f64 (bitconvert (v4bf16 FPR64:$src))),
8568                             (v1f64 (REV64v4i16 FPR64:$src))>;
8569}
8570def : Pat<(v1f64 (bitconvert (v1i64 FPR64:$src))), (v1f64 FPR64:$src)>;
8571def : Pat<(v1f64 (bitconvert (f64   FPR64:$src))), (v1f64 FPR64:$src)>;
8572
8573let Predicates = [IsLE] in {
8574def : Pat<(v2f32 (bitconvert (v1i64 FPR64:$src))), (v2f32 FPR64:$src)>;
8575def : Pat<(v2f32 (bitconvert (v4i16 FPR64:$src))), (v2f32 FPR64:$src)>;
8576def : Pat<(v2f32 (bitconvert (v8i8  FPR64:$src))), (v2f32 FPR64:$src)>;
8577def : Pat<(v2f32 (bitconvert (v1f64 FPR64:$src))), (v2f32 FPR64:$src)>;
8578def : Pat<(v2f32 (bitconvert (f64   FPR64:$src))), (v2f32 FPR64:$src)>;
8579def : Pat<(v2f32 (bitconvert (v4f16 FPR64:$src))), (v2f32 FPR64:$src)>;
8580def : Pat<(v2f32 (bitconvert (v4bf16 FPR64:$src))), (v2f32 FPR64:$src)>;
8581}
8582let Predicates = [IsBE] in {
8583def : Pat<(v2f32 (bitconvert (v1i64 FPR64:$src))),
8584                             (v2f32 (REV64v2i32 FPR64:$src))>;
8585def : Pat<(v2f32 (bitconvert (v4i16 FPR64:$src))),
8586                             (v2f32 (REV32v4i16 FPR64:$src))>;
8587def : Pat<(v2f32 (bitconvert (v8i8  FPR64:$src))),
8588                             (v2f32 (REV32v8i8 FPR64:$src))>;
8589def : Pat<(v2f32 (bitconvert (v1f64 FPR64:$src))),
8590                             (v2f32 (REV64v2i32 FPR64:$src))>;
8591def : Pat<(v2f32 (bitconvert (f64   FPR64:$src))),
8592                             (v2f32 (REV64v2i32 FPR64:$src))>;
8593def : Pat<(v2f32 (bitconvert (v4f16 FPR64:$src))),
8594                             (v2f32 (REV32v4i16 FPR64:$src))>;
8595def : Pat<(v2f32 (bitconvert (v4bf16 FPR64:$src))),
8596                             (v2f32 (REV32v4i16 FPR64:$src))>;
8597}
8598def : Pat<(v2f32 (bitconvert (v2i32 FPR64:$src))), (v2f32 FPR64:$src)>;
8599
8600let Predicates = [IsLE] in {
8601def : Pat<(f128 (bitconvert (v2i64 FPR128:$src))), (f128 FPR128:$src)>;
8602def : Pat<(f128 (bitconvert (v4i32 FPR128:$src))), (f128 FPR128:$src)>;
8603def : Pat<(f128 (bitconvert (v8i16 FPR128:$src))), (f128 FPR128:$src)>;
8604def : Pat<(f128 (bitconvert (v2f64 FPR128:$src))), (f128 FPR128:$src)>;
8605def : Pat<(f128 (bitconvert (v4f32 FPR128:$src))), (f128 FPR128:$src)>;
8606def : Pat<(f128 (bitconvert (v8f16 FPR128:$src))), (f128 FPR128:$src)>;
8607def : Pat<(f128 (bitconvert (v8bf16 FPR128:$src))), (f128 FPR128:$src)>;
8608def : Pat<(f128 (bitconvert (v16i8 FPR128:$src))), (f128 FPR128:$src)>;
8609}
8610let Predicates = [IsBE] in {
8611def : Pat<(f128 (bitconvert (v2i64 FPR128:$src))),
8612                            (f128 (EXTv16i8 FPR128:$src, FPR128:$src, (i32 8)))>;
8613def : Pat<(f128 (bitconvert (v4i32 FPR128:$src))),
8614                            (f128 (EXTv16i8 (REV64v4i32 FPR128:$src),
8615                                            (REV64v4i32 FPR128:$src), (i32 8)))>;
8616def : Pat<(f128 (bitconvert (v8i16 FPR128:$src))),
8617                            (f128 (EXTv16i8 (REV64v8i16 FPR128:$src),
8618                                            (REV64v8i16 FPR128:$src), (i32 8)))>;
8619def : Pat<(f128 (bitconvert (v8f16 FPR128:$src))),
8620                            (f128 (EXTv16i8 (REV64v8i16 FPR128:$src),
8621                                            (REV64v8i16 FPR128:$src), (i32 8)))>;
8622def : Pat<(f128 (bitconvert (v8bf16 FPR128:$src))),
8623                            (f128 (EXTv16i8 (REV64v8i16 FPR128:$src),
8624                                            (REV64v8i16 FPR128:$src), (i32 8)))>;
8625def : Pat<(f128 (bitconvert (v2f64 FPR128:$src))),
8626                            (f128 (EXTv16i8 FPR128:$src, FPR128:$src, (i32 8)))>;
8627def : Pat<(f128 (bitconvert (v4f32 FPR128:$src))),
8628                            (f128 (EXTv16i8 (REV64v4i32 FPR128:$src),
8629                                            (REV64v4i32 FPR128:$src), (i32 8)))>;
8630def : Pat<(f128 (bitconvert (v16i8 FPR128:$src))),
8631                            (f128 (EXTv16i8 (REV64v16i8 FPR128:$src),
8632                                            (REV64v16i8 FPR128:$src), (i32 8)))>;
8633}
8634
8635let Predicates = [IsLE] in {
8636def : Pat<(v2f64 (bitconvert (f128  FPR128:$src))), (v2f64 FPR128:$src)>;
8637def : Pat<(v2f64 (bitconvert (v4i32 FPR128:$src))), (v2f64 FPR128:$src)>;
8638def : Pat<(v2f64 (bitconvert (v8i16 FPR128:$src))), (v2f64 FPR128:$src)>;
8639def : Pat<(v2f64 (bitconvert (v8f16 FPR128:$src))), (v2f64 FPR128:$src)>;
8640def : Pat<(v2f64 (bitconvert (v8bf16 FPR128:$src))), (v2f64 FPR128:$src)>;
8641def : Pat<(v2f64 (bitconvert (v16i8 FPR128:$src))), (v2f64 FPR128:$src)>;
8642def : Pat<(v2f64 (bitconvert (v4f32 FPR128:$src))), (v2f64 FPR128:$src)>;
8643}
8644let Predicates = [IsBE] in {
8645def : Pat<(v2f64 (bitconvert (f128  FPR128:$src))),
8646                             (v2f64 (EXTv16i8 FPR128:$src,
8647                                              FPR128:$src, (i32 8)))>;
8648def : Pat<(v2f64 (bitconvert (v4i32 FPR128:$src))),
8649                             (v2f64 (REV64v4i32 FPR128:$src))>;
8650def : Pat<(v2f64 (bitconvert (v8i16 FPR128:$src))),
8651                             (v2f64 (REV64v8i16 FPR128:$src))>;
8652def : Pat<(v2f64 (bitconvert (v8f16 FPR128:$src))),
8653                             (v2f64 (REV64v8i16 FPR128:$src))>;
8654def : Pat<(v2f64 (bitconvert (v8bf16 FPR128:$src))),
8655                             (v2f64 (REV64v8i16 FPR128:$src))>;
8656def : Pat<(v2f64 (bitconvert (v16i8 FPR128:$src))),
8657                             (v2f64 (REV64v16i8 FPR128:$src))>;
8658def : Pat<(v2f64 (bitconvert (v4f32 FPR128:$src))),
8659                             (v2f64 (REV64v4i32 FPR128:$src))>;
8660}
8661def : Pat<(v2f64 (bitconvert (v2i64 FPR128:$src))), (v2f64 FPR128:$src)>;
8662
8663let Predicates = [IsLE] in {
8664def : Pat<(v4f32 (bitconvert (f128  FPR128:$src))), (v4f32 FPR128:$src)>;
8665def : Pat<(v4f32 (bitconvert (v8i16 FPR128:$src))), (v4f32 FPR128:$src)>;
8666def : Pat<(v4f32 (bitconvert (v8f16 FPR128:$src))), (v4f32 FPR128:$src)>;
8667def : Pat<(v4f32 (bitconvert (v8bf16 FPR128:$src))), (v4f32 FPR128:$src)>;
8668def : Pat<(v4f32 (bitconvert (v16i8 FPR128:$src))), (v4f32 FPR128:$src)>;
8669def : Pat<(v4f32 (bitconvert (v2i64 FPR128:$src))), (v4f32 FPR128:$src)>;
8670def : Pat<(v4f32 (bitconvert (v2f64 FPR128:$src))), (v4f32 FPR128:$src)>;
8671}
8672let Predicates = [IsBE] in {
8673def : Pat<(v4f32 (bitconvert (f128  FPR128:$src))),
8674                             (v4f32 (EXTv16i8 (REV64v4i32 FPR128:$src),
8675                                    (REV64v4i32 FPR128:$src), (i32 8)))>;
8676def : Pat<(v4f32 (bitconvert (v8i16 FPR128:$src))),
8677                             (v4f32 (REV32v8i16 FPR128:$src))>;
8678def : Pat<(v4f32 (bitconvert (v8f16 FPR128:$src))),
8679                             (v4f32 (REV32v8i16 FPR128:$src))>;
8680def : Pat<(v4f32 (bitconvert (v8bf16 FPR128:$src))),
8681                             (v4f32 (REV32v8i16 FPR128:$src))>;
8682def : Pat<(v4f32 (bitconvert (v16i8 FPR128:$src))),
8683                             (v4f32 (REV32v16i8 FPR128:$src))>;
8684def : Pat<(v4f32 (bitconvert (v2i64 FPR128:$src))),
8685                             (v4f32 (REV64v4i32 FPR128:$src))>;
8686def : Pat<(v4f32 (bitconvert (v2f64 FPR128:$src))),
8687                             (v4f32 (REV64v4i32 FPR128:$src))>;
8688}
8689def : Pat<(v4f32 (bitconvert (v4i32 FPR128:$src))), (v4f32 FPR128:$src)>;
8690
8691let Predicates = [IsLE] in {
8692def : Pat<(v2i64 (bitconvert (f128  FPR128:$src))), (v2i64 FPR128:$src)>;
8693def : Pat<(v2i64 (bitconvert (v4i32 FPR128:$src))), (v2i64 FPR128:$src)>;
8694def : Pat<(v2i64 (bitconvert (v8i16 FPR128:$src))), (v2i64 FPR128:$src)>;
8695def : Pat<(v2i64 (bitconvert (v16i8 FPR128:$src))), (v2i64 FPR128:$src)>;
8696def : Pat<(v2i64 (bitconvert (v4f32 FPR128:$src))), (v2i64 FPR128:$src)>;
8697def : Pat<(v2i64 (bitconvert (v8f16 FPR128:$src))), (v2i64 FPR128:$src)>;
8698def : Pat<(v2i64 (bitconvert (v8bf16 FPR128:$src))), (v2i64 FPR128:$src)>;
8699}
8700let Predicates = [IsBE] in {
8701def : Pat<(v2i64 (bitconvert (f128  FPR128:$src))),
8702                             (v2i64 (EXTv16i8 FPR128:$src,
8703                                              FPR128:$src, (i32 8)))>;
8704def : Pat<(v2i64 (bitconvert (v4i32 FPR128:$src))),
8705                             (v2i64 (REV64v4i32 FPR128:$src))>;
8706def : Pat<(v2i64 (bitconvert (v8i16 FPR128:$src))),
8707                             (v2i64 (REV64v8i16 FPR128:$src))>;
8708def : Pat<(v2i64 (bitconvert (v16i8 FPR128:$src))),
8709                             (v2i64 (REV64v16i8 FPR128:$src))>;
8710def : Pat<(v2i64 (bitconvert (v4f32 FPR128:$src))),
8711                             (v2i64 (REV64v4i32 FPR128:$src))>;
8712def : Pat<(v2i64 (bitconvert (v8f16 FPR128:$src))),
8713                             (v2i64 (REV64v8i16 FPR128:$src))>;
8714def : Pat<(v2i64 (bitconvert (v8bf16 FPR128:$src))),
8715                             (v2i64 (REV64v8i16 FPR128:$src))>;
8716}
8717def : Pat<(v2i64 (bitconvert (v2f64 FPR128:$src))), (v2i64 FPR128:$src)>;
8718
8719let Predicates = [IsLE] in {
8720def : Pat<(v4i32 (bitconvert (f128  FPR128:$src))), (v4i32 FPR128:$src)>;
8721def : Pat<(v4i32 (bitconvert (v2i64 FPR128:$src))), (v4i32 FPR128:$src)>;
8722def : Pat<(v4i32 (bitconvert (v8i16 FPR128:$src))), (v4i32 FPR128:$src)>;
8723def : Pat<(v4i32 (bitconvert (v16i8 FPR128:$src))), (v4i32 FPR128:$src)>;
8724def : Pat<(v4i32 (bitconvert (v2f64 FPR128:$src))), (v4i32 FPR128:$src)>;
8725def : Pat<(v4i32 (bitconvert (v8f16 FPR128:$src))), (v4i32 FPR128:$src)>;
8726def : Pat<(v4i32 (bitconvert (v8bf16 FPR128:$src))), (v4i32 FPR128:$src)>;
8727}
8728let Predicates = [IsBE] in {
8729def : Pat<(v4i32 (bitconvert (f128  FPR128:$src))),
8730                             (v4i32 (EXTv16i8 (REV64v4i32 FPR128:$src),
8731                                              (REV64v4i32 FPR128:$src),
8732                                              (i32 8)))>;
8733def : Pat<(v4i32 (bitconvert (v2i64 FPR128:$src))),
8734                             (v4i32 (REV64v4i32 FPR128:$src))>;
8735def : Pat<(v4i32 (bitconvert (v8i16 FPR128:$src))),
8736                             (v4i32 (REV32v8i16 FPR128:$src))>;
8737def : Pat<(v4i32 (bitconvert (v16i8 FPR128:$src))),
8738                             (v4i32 (REV32v16i8 FPR128:$src))>;
8739def : Pat<(v4i32 (bitconvert (v2f64 FPR128:$src))),
8740                             (v4i32 (REV64v4i32 FPR128:$src))>;
8741def : Pat<(v4i32 (bitconvert (v8f16 FPR128:$src))),
8742                             (v4i32 (REV32v8i16 FPR128:$src))>;
8743def : Pat<(v4i32 (bitconvert (v8bf16 FPR128:$src))),
8744                             (v4i32 (REV32v8i16 FPR128:$src))>;
8745}
8746def : Pat<(v4i32 (bitconvert (v4f32 FPR128:$src))), (v4i32 FPR128:$src)>;
8747
8748let Predicates = [IsLE] in {
8749def : Pat<(v8i16 (bitconvert (f128  FPR128:$src))), (v8i16 FPR128:$src)>;
8750def : Pat<(v8i16 (bitconvert (v2i64 FPR128:$src))), (v8i16 FPR128:$src)>;
8751def : Pat<(v8i16 (bitconvert (v4i32 FPR128:$src))), (v8i16 FPR128:$src)>;
8752def : Pat<(v8i16 (bitconvert (v16i8 FPR128:$src))), (v8i16 FPR128:$src)>;
8753def : Pat<(v8i16 (bitconvert (v2f64 FPR128:$src))), (v8i16 FPR128:$src)>;
8754def : Pat<(v8i16 (bitconvert (v4f32 FPR128:$src))), (v8i16 FPR128:$src)>;
8755}
8756let Predicates = [IsBE] in {
8757def : Pat<(v8i16 (bitconvert (f128  FPR128:$src))),
8758                             (v8i16 (EXTv16i8 (REV64v8i16 FPR128:$src),
8759                                              (REV64v8i16 FPR128:$src),
8760                                              (i32 8)))>;
8761def : Pat<(v8i16 (bitconvert (v2i64 FPR128:$src))),
8762                             (v8i16 (REV64v8i16 FPR128:$src))>;
8763def : Pat<(v8i16 (bitconvert (v4i32 FPR128:$src))),
8764                             (v8i16 (REV32v8i16 FPR128:$src))>;
8765def : Pat<(v8i16 (bitconvert (v16i8 FPR128:$src))),
8766                             (v8i16 (REV16v16i8 FPR128:$src))>;
8767def : Pat<(v8i16 (bitconvert (v2f64 FPR128:$src))),
8768                             (v8i16 (REV64v8i16 FPR128:$src))>;
8769def : Pat<(v8i16 (bitconvert (v4f32 FPR128:$src))),
8770                             (v8i16 (REV32v8i16 FPR128:$src))>;
8771}
8772def : Pat<(v8i16 (bitconvert (v8f16 FPR128:$src))), (v8i16 FPR128:$src)>;
8773def : Pat<(v8i16 (bitconvert (v8bf16 FPR128:$src))), (v8i16 FPR128:$src)>;
8774
8775let Predicates = [IsLE] in {
8776def : Pat<(v8f16 (bitconvert (f128  FPR128:$src))), (v8f16 FPR128:$src)>;
8777def : Pat<(v8f16 (bitconvert (v2i64 FPR128:$src))), (v8f16 FPR128:$src)>;
8778def : Pat<(v8f16 (bitconvert (v4i32 FPR128:$src))), (v8f16 FPR128:$src)>;
8779def : Pat<(v8f16 (bitconvert (v16i8 FPR128:$src))), (v8f16 FPR128:$src)>;
8780def : Pat<(v8f16 (bitconvert (v2f64 FPR128:$src))), (v8f16 FPR128:$src)>;
8781def : Pat<(v8f16 (bitconvert (v4f32 FPR128:$src))), (v8f16 FPR128:$src)>;
8782
8783def : Pat<(v8bf16 (bitconvert (f128  FPR128:$src))), (v8bf16 FPR128:$src)>;
8784def : Pat<(v8bf16 (bitconvert (v2i64 FPR128:$src))), (v8bf16 FPR128:$src)>;
8785def : Pat<(v8bf16 (bitconvert (v4i32 FPR128:$src))), (v8bf16 FPR128:$src)>;
8786def : Pat<(v8bf16 (bitconvert (v16i8 FPR128:$src))), (v8bf16 FPR128:$src)>;
8787def : Pat<(v8bf16 (bitconvert (v2f64 FPR128:$src))), (v8bf16 FPR128:$src)>;
8788def : Pat<(v8bf16 (bitconvert (v4f32 FPR128:$src))), (v8bf16 FPR128:$src)>;
8789}
8790let Predicates = [IsBE] in {
8791def : Pat<(v8f16 (bitconvert (f128  FPR128:$src))),
8792                             (v8f16 (EXTv16i8 (REV64v8i16 FPR128:$src),
8793                                              (REV64v8i16 FPR128:$src),
8794                                              (i32 8)))>;
8795def : Pat<(v8f16 (bitconvert (v2i64 FPR128:$src))),
8796                             (v8f16 (REV64v8i16 FPR128:$src))>;
8797def : Pat<(v8f16 (bitconvert (v4i32 FPR128:$src))),
8798                             (v8f16 (REV32v8i16 FPR128:$src))>;
8799def : Pat<(v8f16 (bitconvert (v16i8 FPR128:$src))),
8800                             (v8f16 (REV16v16i8 FPR128:$src))>;
8801def : Pat<(v8f16 (bitconvert (v2f64 FPR128:$src))),
8802                             (v8f16 (REV64v8i16 FPR128:$src))>;
8803def : Pat<(v8f16 (bitconvert (v4f32 FPR128:$src))),
8804                             (v8f16 (REV32v8i16 FPR128:$src))>;
8805
8806def : Pat<(v8bf16 (bitconvert (f128  FPR128:$src))),
8807                             (v8bf16 (EXTv16i8 (REV64v8i16 FPR128:$src),
8808                                              (REV64v8i16 FPR128:$src),
8809                                              (i32 8)))>;
8810def : Pat<(v8bf16 (bitconvert (v2i64 FPR128:$src))),
8811                             (v8bf16 (REV64v8i16 FPR128:$src))>;
8812def : Pat<(v8bf16 (bitconvert (v4i32 FPR128:$src))),
8813                             (v8bf16 (REV32v8i16 FPR128:$src))>;
8814def : Pat<(v8bf16 (bitconvert (v16i8 FPR128:$src))),
8815                             (v8bf16 (REV16v16i8 FPR128:$src))>;
8816def : Pat<(v8bf16 (bitconvert (v2f64 FPR128:$src))),
8817                             (v8bf16 (REV64v8i16 FPR128:$src))>;
8818def : Pat<(v8bf16 (bitconvert (v4f32 FPR128:$src))),
8819                             (v8bf16 (REV32v8i16 FPR128:$src))>;
8820}
8821def : Pat<(v8f16 (bitconvert (v8i16 FPR128:$src))), (v8f16 FPR128:$src)>;
8822def : Pat<(v8bf16 (bitconvert (v8i16 FPR128:$src))), (v8bf16 FPR128:$src)>;
8823
8824let Predicates = [IsLE] in {
8825def : Pat<(v16i8 (bitconvert (f128  FPR128:$src))), (v16i8 FPR128:$src)>;
8826def : Pat<(v16i8 (bitconvert (v2i64 FPR128:$src))), (v16i8 FPR128:$src)>;
8827def : Pat<(v16i8 (bitconvert (v4i32 FPR128:$src))), (v16i8 FPR128:$src)>;
8828def : Pat<(v16i8 (bitconvert (v8i16 FPR128:$src))), (v16i8 FPR128:$src)>;
8829def : Pat<(v16i8 (bitconvert (v2f64 FPR128:$src))), (v16i8 FPR128:$src)>;
8830def : Pat<(v16i8 (bitconvert (v4f32 FPR128:$src))), (v16i8 FPR128:$src)>;
8831def : Pat<(v16i8 (bitconvert (v8f16 FPR128:$src))), (v16i8 FPR128:$src)>;
8832def : Pat<(v16i8 (bitconvert (v8bf16 FPR128:$src))), (v16i8 FPR128:$src)>;
8833}
8834let Predicates = [IsBE] in {
8835def : Pat<(v16i8 (bitconvert (f128  FPR128:$src))),
8836                             (v16i8 (EXTv16i8 (REV64v16i8 FPR128:$src),
8837                                              (REV64v16i8 FPR128:$src),
8838                                              (i32 8)))>;
8839def : Pat<(v16i8 (bitconvert (v2i64 FPR128:$src))),
8840                             (v16i8 (REV64v16i8 FPR128:$src))>;
8841def : Pat<(v16i8 (bitconvert (v4i32 FPR128:$src))),
8842                             (v16i8 (REV32v16i8 FPR128:$src))>;
8843def : Pat<(v16i8 (bitconvert (v8i16 FPR128:$src))),
8844                             (v16i8 (REV16v16i8 FPR128:$src))>;
8845def : Pat<(v16i8 (bitconvert (v2f64 FPR128:$src))),
8846                             (v16i8 (REV64v16i8 FPR128:$src))>;
8847def : Pat<(v16i8 (bitconvert (v4f32 FPR128:$src))),
8848                             (v16i8 (REV32v16i8 FPR128:$src))>;
8849def : Pat<(v16i8 (bitconvert (v8f16 FPR128:$src))),
8850                             (v16i8 (REV16v16i8 FPR128:$src))>;
8851def : Pat<(v16i8 (bitconvert (v8bf16 FPR128:$src))),
8852                             (v16i8 (REV16v16i8 FPR128:$src))>;
8853}
8854
8855def : Pat<(v4i16 (extract_subvector V128:$Rn, (i64 0))),
8856           (EXTRACT_SUBREG V128:$Rn, dsub)>;
8857def : Pat<(v8i8 (extract_subvector V128:$Rn, (i64 0))),
8858           (EXTRACT_SUBREG V128:$Rn, dsub)>;
8859def : Pat<(v2f32 (extract_subvector V128:$Rn, (i64 0))),
8860           (EXTRACT_SUBREG V128:$Rn, dsub)>;
8861def : Pat<(v4f16 (extract_subvector V128:$Rn, (i64 0))),
8862           (EXTRACT_SUBREG V128:$Rn, dsub)>;
8863def : Pat<(v4bf16 (extract_subvector V128:$Rn, (i64 0))),
8864           (EXTRACT_SUBREG V128:$Rn, dsub)>;
8865def : Pat<(v2i32 (extract_subvector V128:$Rn, (i64 0))),
8866           (EXTRACT_SUBREG V128:$Rn, dsub)>;
8867def : Pat<(v1i64 (extract_subvector V128:$Rn, (i64 0))),
8868           (EXTRACT_SUBREG V128:$Rn, dsub)>;
8869def : Pat<(v1f64 (extract_subvector V128:$Rn, (i64 0))),
8870           (EXTRACT_SUBREG V128:$Rn, dsub)>;
8871
8872def : Pat<(v8i8 (extract_subvector (v16i8 FPR128:$Rn), (i64 1))),
8873          (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
8874def : Pat<(v4i16 (extract_subvector (v8i16 FPR128:$Rn), (i64 1))),
8875          (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
8876def : Pat<(v2i32 (extract_subvector (v4i32 FPR128:$Rn), (i64 1))),
8877          (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
8878def : Pat<(v1i64 (extract_subvector (v2i64 FPR128:$Rn), (i64 1))),
8879          (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
8880
8881// A 64-bit subvector insert to the first 128-bit vector position
8882// is a subregister copy that needs no instruction.
8883multiclass InsertSubvectorUndef<ValueType Ty> {
8884  def : Pat<(insert_subvector undef, (v1i64 FPR64:$src), (Ty 0)),
8885            (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
8886  def : Pat<(insert_subvector undef, (v1f64 FPR64:$src), (Ty 0)),
8887            (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
8888  def : Pat<(insert_subvector undef, (v2i32 FPR64:$src), (Ty 0)),
8889            (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
8890  def : Pat<(insert_subvector undef, (v2f32 FPR64:$src), (Ty 0)),
8891            (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
8892  def : Pat<(insert_subvector undef, (v4i16 FPR64:$src), (Ty 0)),
8893            (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
8894  def : Pat<(insert_subvector undef, (v4f16 FPR64:$src), (Ty 0)),
8895            (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
8896  def : Pat<(insert_subvector undef, (v4bf16 FPR64:$src), (Ty 0)),
8897            (INSERT_SUBREG (v8bf16 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
8898  def : Pat<(insert_subvector undef, (v8i8 FPR64:$src), (Ty 0)),
8899            (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
8900}
8901
8902defm : InsertSubvectorUndef<i32>;
8903defm : InsertSubvectorUndef<i64>;
8904
8905// Use pair-wise add instructions when summing up the lanes for v2f64, v2i64
8906// or v2f32.
8907def : Pat<(i64 (add (vector_extract (v2i64 FPR128:$Rn), (i64 0)),
8908                    (vector_extract (v2i64 FPR128:$Rn), (i64 1)))),
8909           (i64 (ADDPv2i64p (v2i64 FPR128:$Rn)))>;
8910def : Pat<(f64 (any_fadd (vector_extract (v2f64 FPR128:$Rn), (i64 0)),
8911                         (vector_extract (v2f64 FPR128:$Rn), (i64 1)))),
8912           (f64 (FADDPv2i64p (v2f64 FPR128:$Rn)))>;
8913    // vector_extract on 64-bit vectors gets promoted to a 128 bit vector,
8914    // so we match on v4f32 here, not v2f32. This will also catch adding
8915    // the low two lanes of a true v4f32 vector.
8916def : Pat<(any_fadd (vector_extract (v4f32 FPR128:$Rn), (i64 0)),
8917                    (vector_extract (v4f32 FPR128:$Rn), (i64 1))),
8918          (f32 (FADDPv2i32p (EXTRACT_SUBREG FPR128:$Rn, dsub)))>;
8919def : Pat<(any_fadd (vector_extract (v8f16 FPR128:$Rn), (i64 0)),
8920                    (vector_extract (v8f16 FPR128:$Rn), (i64 1))),
8921          (f16 (FADDPv2i16p (EXTRACT_SUBREG FPR128:$Rn, dsub)))>;
8922
8923// Prefer using the bottom lanes of addp Rn, Rn compared to
8924// addp extractlow(Rn), extracthigh(Rn)
8925def : Pat<(AArch64addp (v2i32 (extract_subvector (v4i32 FPR128:$Rn), (i64 0))),
8926                       (v2i32 (extract_subvector (v4i32 FPR128:$Rn), (i64 2)))),
8927          (v2i32 (EXTRACT_SUBREG (ADDPv4i32 $Rn, $Rn), dsub))>;
8928def : Pat<(AArch64addp (v4i16 (extract_subvector (v8i16 FPR128:$Rn), (i64 0))),
8929                       (v4i16 (extract_subvector (v8i16 FPR128:$Rn), (i64 4)))),
8930          (v4i16 (EXTRACT_SUBREG (ADDPv8i16 $Rn, $Rn), dsub))>;
8931def : Pat<(AArch64addp (v8i8 (extract_subvector (v16i8 FPR128:$Rn), (i64 0))),
8932                       (v8i8 (extract_subvector (v16i8 FPR128:$Rn), (i64 8)))),
8933          (v8i8 (EXTRACT_SUBREG (ADDPv16i8 $Rn, $Rn), dsub))>;
8934
8935def : Pat<(AArch64faddp (v2f32 (extract_subvector (v4f32 FPR128:$Rn), (i64 0))),
8936                        (v2f32 (extract_subvector (v4f32 FPR128:$Rn), (i64 2)))),
8937          (v2f32 (EXTRACT_SUBREG (FADDPv4f32 $Rn, $Rn), dsub))>;
8938def : Pat<(AArch64faddp (v4f16 (extract_subvector (v8f16 FPR128:$Rn), (i64 0))),
8939                        (v4f16 (extract_subvector (v8f16 FPR128:$Rn), (i64 4)))),
8940          (v4f16 (EXTRACT_SUBREG (FADDPv8f16 $Rn, $Rn), dsub))>;
8941
8942// Scalar 64-bit shifts in FPR64 registers.
8943def : Pat<(i64 (int_aarch64_neon_sshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
8944          (SSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
8945def : Pat<(i64 (int_aarch64_neon_ushl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
8946          (USHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
8947def : Pat<(i64 (int_aarch64_neon_srshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
8948          (SRSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
8949def : Pat<(i64 (int_aarch64_neon_urshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
8950          (URSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
8951
8952// Patterns for nontemporal/no-allocate stores.
8953// We have to resort to tricks to turn a single-input store into a store pair,
8954// because there is no single-input nontemporal store, only STNP.
8955let Predicates = [IsLE] in {
8956let AddedComplexity = 15 in {
8957class NTStore128Pat<ValueType VT> :
8958  Pat<(nontemporalstore (VT FPR128:$Rt),
8959        (am_indexed7s64 GPR64sp:$Rn, simm7s8:$offset)),
8960      (STNPDi (EXTRACT_SUBREG FPR128:$Rt, dsub),
8961              (DUPi64 FPR128:$Rt, (i64 1)),
8962              GPR64sp:$Rn, simm7s8:$offset)>;
8963
8964def : NTStore128Pat<v2i64>;
8965def : NTStore128Pat<v4i32>;
8966def : NTStore128Pat<v8i16>;
8967def : NTStore128Pat<v16i8>;
8968
8969class NTStore64Pat<ValueType VT> :
8970  Pat<(nontemporalstore (VT FPR64:$Rt),
8971        (am_indexed7s32 GPR64sp:$Rn, simm7s4:$offset)),
8972      (STNPSi (EXTRACT_SUBREG FPR64:$Rt, ssub),
8973              (DUPi32 (SUBREG_TO_REG (i64 0), FPR64:$Rt, dsub), (i64 1)),
8974              GPR64sp:$Rn, simm7s4:$offset)>;
8975
8976// FIXME: Shouldn't v1f64 loads/stores be promoted to v1i64?
8977def : NTStore64Pat<v1f64>;
8978def : NTStore64Pat<v1i64>;
8979def : NTStore64Pat<v2i32>;
8980def : NTStore64Pat<v4i16>;
8981def : NTStore64Pat<v8i8>;
8982
8983def : Pat<(nontemporalstore GPR64:$Rt,
8984            (am_indexed7s32 GPR64sp:$Rn, simm7s4:$offset)),
8985          (STNPWi (EXTRACT_SUBREG GPR64:$Rt, sub_32),
8986                  (EXTRACT_SUBREG (UBFMXri GPR64:$Rt, 32, 63), sub_32),
8987                  GPR64sp:$Rn, simm7s4:$offset)>;
8988} // AddedComplexity=10
8989} // Predicates = [IsLE]
8990
8991// Tail call return handling. These are all compiler pseudo-instructions,
8992// so no encoding information or anything like that.
8993let isCall = 1, isTerminator = 1, isReturn = 1, isBarrier = 1, Uses = [SP] in {
8994  def TCRETURNdi : Pseudo<(outs), (ins i64imm:$dst, i32imm:$FPDiff), []>,
8995                   Sched<[WriteBrReg]>;
8996  def TCRETURNri : Pseudo<(outs), (ins tcGPR64:$dst, i32imm:$FPDiff), []>,
8997                   Sched<[WriteBrReg]>;
8998  // Indirect tail-call with any register allowed, used by MachineOutliner when
8999  // this is proven safe.
9000  // FIXME: If we have to add any more hacks like this, we should instead relax
9001  // some verifier checks for outlined functions.
9002  def TCRETURNriALL : Pseudo<(outs), (ins GPR64:$dst, i32imm:$FPDiff), []>,
9003                      Sched<[WriteBrReg]>;
9004  // Indirect tail-call limited to only use registers (x16 and x17) which are
9005  // allowed to tail-call a "BTI c" instruction.
9006  def TCRETURNriBTI : Pseudo<(outs), (ins rtcGPR64:$dst, i32imm:$FPDiff), []>,
9007                      Sched<[WriteBrReg]>;
9008}
9009
9010def : Pat<(AArch64tcret tcGPR64:$dst, (i32 timm:$FPDiff)),
9011          (TCRETURNri tcGPR64:$dst, imm:$FPDiff)>,
9012      Requires<[NotUseBTI]>;
9013def : Pat<(AArch64tcret rtcGPR64:$dst, (i32 timm:$FPDiff)),
9014          (TCRETURNriBTI rtcGPR64:$dst, imm:$FPDiff)>,
9015      Requires<[UseBTI]>;
9016def : Pat<(AArch64tcret tglobaladdr:$dst, (i32 timm:$FPDiff)),
9017          (TCRETURNdi texternalsym:$dst, imm:$FPDiff)>;
9018def : Pat<(AArch64tcret texternalsym:$dst, (i32 timm:$FPDiff)),
9019          (TCRETURNdi texternalsym:$dst, imm:$FPDiff)>;
9020
9021def MOVMCSym : Pseudo<(outs GPR64:$dst), (ins i64imm:$sym), []>, Sched<[]>;
9022def : Pat<(i64 (AArch64LocalRecover mcsym:$sym)), (MOVMCSym mcsym:$sym)>;
9023
9024// Extracting lane zero is a special case where we can just use a plain
9025// EXTRACT_SUBREG instruction, which will become FMOV. This is easier for the
9026// rest of the compiler, especially the register allocator and copy propagation,
9027// to reason about, so is preferred when it's possible to use it.
9028let AddedComplexity = 10 in {
9029  def : Pat<(i64 (extractelt (v2i64 V128:$V), (i64 0))), (EXTRACT_SUBREG V128:$V, dsub)>;
9030  def : Pat<(i32 (extractelt (v4i32 V128:$V), (i64 0))), (EXTRACT_SUBREG V128:$V, ssub)>;
9031  def : Pat<(i32 (extractelt (v2i32 V64:$V), (i64 0))), (EXTRACT_SUBREG V64:$V, ssub)>;
9032}
9033
9034// dot_v4i8
9035class mul_v4i8<SDPatternOperator ldop> :
9036  PatFrag<(ops node:$Rn, node:$Rm, node:$offset),
9037          (mul (ldop (add node:$Rn, node:$offset)),
9038               (ldop (add node:$Rm, node:$offset)))>;
9039class mulz_v4i8<SDPatternOperator ldop> :
9040  PatFrag<(ops node:$Rn, node:$Rm),
9041          (mul (ldop node:$Rn), (ldop node:$Rm))>;
9042
9043def load_v4i8 :
9044  OutPatFrag<(ops node:$R),
9045             (INSERT_SUBREG
9046              (v2i32 (IMPLICIT_DEF)),
9047               (i32 (COPY_TO_REGCLASS (LDRWui node:$R, (i64 0)), FPR32)),
9048              ssub)>;
9049
9050class dot_v4i8<Instruction DOT, SDPatternOperator ldop> :
9051  Pat<(i32 (add (mul_v4i8<ldop> GPR64sp:$Rn, GPR64sp:$Rm, (i64 3)),
9052           (add (mul_v4i8<ldop> GPR64sp:$Rn, GPR64sp:$Rm, (i64 2)),
9053           (add (mul_v4i8<ldop> GPR64sp:$Rn, GPR64sp:$Rm, (i64 1)),
9054                (mulz_v4i8<ldop> GPR64sp:$Rn, GPR64sp:$Rm))))),
9055      (EXTRACT_SUBREG (i64 (DOT (DUPv2i32gpr WZR),
9056                                (load_v4i8 GPR64sp:$Rn),
9057                                (load_v4i8 GPR64sp:$Rm))),
9058                      sub_32)>, Requires<[HasDotProd]>;
9059
9060// dot_v8i8
9061class ee_v8i8<SDPatternOperator extend> :
9062  PatFrag<(ops node:$V, node:$K),
9063          (v4i16 (extract_subvector (v8i16 (extend node:$V)), node:$K))>;
9064
9065class mul_v8i8<SDPatternOperator mulop, SDPatternOperator extend> :
9066  PatFrag<(ops node:$M, node:$N, node:$K),
9067          (mulop (v4i16 (ee_v8i8<extend> node:$M, node:$K)),
9068                 (v4i16 (ee_v8i8<extend> node:$N, node:$K)))>;
9069
9070class idot_v8i8<SDPatternOperator mulop, SDPatternOperator extend> :
9071  PatFrag<(ops node:$M, node:$N),
9072          (i32 (extractelt
9073           (v4i32 (AArch64uaddv
9074            (add (mul_v8i8<mulop, extend> node:$M, node:$N, (i64 0)),
9075                 (mul_v8i8<mulop, extend> node:$M, node:$N, (i64 4))))),
9076           (i64 0)))>;
9077
9078// vaddv_[su]32 is special; -> ADDP Vd.2S,Vn.2S,Vm.2S; return Vd.s[0];Vn==Vm
9079def VADDV_32 : OutPatFrag<(ops node:$R), (ADDPv2i32 node:$R, node:$R)>;
9080
9081class odot_v8i8<Instruction DOT> :
9082  OutPatFrag<(ops node:$Vm, node:$Vn),
9083             (EXTRACT_SUBREG
9084              (VADDV_32
9085               (i64 (DOT (DUPv2i32gpr WZR),
9086                         (v8i8 node:$Vm),
9087                         (v8i8 node:$Vn)))),
9088              sub_32)>;
9089
9090class dot_v8i8<Instruction DOT, SDPatternOperator mulop,
9091                    SDPatternOperator extend> :
9092  Pat<(idot_v8i8<mulop, extend> V64:$Vm, V64:$Vn),
9093      (odot_v8i8<DOT> V64:$Vm, V64:$Vn)>,
9094  Requires<[HasDotProd]>;
9095
9096// dot_v16i8
9097class ee_v16i8<SDPatternOperator extend> :
9098  PatFrag<(ops node:$V, node:$K1, node:$K2),
9099          (v4i16 (extract_subvector
9100           (v8i16 (extend
9101            (v8i8 (extract_subvector node:$V, node:$K1)))), node:$K2))>;
9102
9103class mul_v16i8<SDPatternOperator mulop, SDPatternOperator extend> :
9104  PatFrag<(ops node:$M, node:$N, node:$K1, node:$K2),
9105          (v4i32
9106           (mulop (v4i16 (ee_v16i8<extend> node:$M, node:$K1, node:$K2)),
9107                  (v4i16 (ee_v16i8<extend> node:$N, node:$K1, node:$K2))))>;
9108
9109class idot_v16i8<SDPatternOperator m, SDPatternOperator x> :
9110  PatFrag<(ops node:$M, node:$N),
9111          (i32 (extractelt
9112           (v4i32 (AArch64uaddv
9113            (add
9114             (add (mul_v16i8<m, x> node:$M, node:$N, (i64 0), (i64 0)),
9115                  (mul_v16i8<m, x> node:$M, node:$N, (i64 8), (i64 0))),
9116             (add (mul_v16i8<m, x> node:$M, node:$N, (i64 0), (i64 4)),
9117                  (mul_v16i8<m, x> node:$M, node:$N, (i64 8), (i64 4)))))),
9118           (i64 0)))>;
9119
9120class odot_v16i8<Instruction DOT> :
9121  OutPatFrag<(ops node:$Vm, node:$Vn),
9122             (i32 (ADDVv4i32v
9123              (DOT (DUPv4i32gpr WZR), node:$Vm, node:$Vn)))>;
9124
9125class dot_v16i8<Instruction DOT, SDPatternOperator mulop,
9126                SDPatternOperator extend> :
9127  Pat<(idot_v16i8<mulop, extend> V128:$Vm, V128:$Vn),
9128      (odot_v16i8<DOT> V128:$Vm, V128:$Vn)>,
9129  Requires<[HasDotProd]>;
9130
9131let AddedComplexity = 10 in {
9132  def : dot_v4i8<SDOTv8i8, sextloadi8>;
9133  def : dot_v4i8<UDOTv8i8, zextloadi8>;
9134  def : dot_v8i8<SDOTv8i8, AArch64smull, sext>;
9135  def : dot_v8i8<UDOTv8i8, AArch64umull, zext>;
9136  def : dot_v16i8<SDOTv16i8, AArch64smull, sext>;
9137  def : dot_v16i8<UDOTv16i8, AArch64umull, zext>;
9138
9139  // FIXME: add patterns to generate vector by element dot product.
9140  // FIXME: add SVE dot-product patterns.
9141}
9142
9143// Custom DAG nodes and isel rules to make a 64-byte block out of eight GPRs,
9144// so that it can be used as input to inline asm, and vice versa.
9145def LS64_BUILD : SDNode<"AArch64ISD::LS64_BUILD", SDTypeProfile<1, 8, []>>;
9146def LS64_EXTRACT : SDNode<"AArch64ISD::LS64_EXTRACT", SDTypeProfile<1, 2, []>>;
9147def : Pat<(i64x8 (LS64_BUILD GPR64:$x0, GPR64:$x1, GPR64:$x2, GPR64:$x3,
9148                             GPR64:$x4, GPR64:$x5, GPR64:$x6, GPR64:$x7)),
9149          (REG_SEQUENCE GPR64x8Class,
9150              $x0, x8sub_0, $x1, x8sub_1, $x2, x8sub_2, $x3, x8sub_3,
9151              $x4, x8sub_4, $x5, x8sub_5, $x6, x8sub_6, $x7, x8sub_7)>;
9152foreach i = 0-7 in {
9153  def : Pat<(i64 (LS64_EXTRACT (i64x8 GPR64x8:$val), (i32 i))),
9154            (EXTRACT_SUBREG $val, !cast<SubRegIndex>("x8sub_"#i))>;
9155}
9156
9157let Predicates = [HasLS64] in {
9158  def LD64B: LoadStore64B<0b101, "ld64b", (ins GPR64sp:$Rn),
9159                                          (outs GPR64x8:$Rt)>;
9160  def ST64B: LoadStore64B<0b001, "st64b", (ins GPR64x8:$Rt, GPR64sp:$Rn),
9161                                          (outs)>;
9162  def ST64BV:   Store64BV<0b011, "st64bv">;
9163  def ST64BV0:  Store64BV<0b010, "st64bv0">;
9164
9165  class ST64BPattern<Intrinsic intrinsic, Instruction instruction>
9166    : Pat<(intrinsic GPR64sp:$addr, GPR64:$x0, GPR64:$x1, GPR64:$x2, GPR64:$x3, GPR64:$x4, GPR64:$x5, GPR64:$x6, GPR64:$x7),
9167          (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)>;
9168
9169  def : ST64BPattern<int_aarch64_st64b, ST64B>;
9170  def : ST64BPattern<int_aarch64_st64bv, ST64BV>;
9171  def : ST64BPattern<int_aarch64_st64bv0, ST64BV0>;
9172}
9173
9174let Predicates = [HasMOPS] in {
9175  let Defs = [NZCV] in {
9176    defm CPYFP : MOPSMemoryCopyInsns<0b00, "cpyfp">;
9177
9178    defm CPYP : MOPSMemoryMoveInsns<0b00, "cpyp">;
9179
9180    defm SETP : MOPSMemorySetInsns<0b00, "setp">;
9181  }
9182  let Uses = [NZCV] in {
9183    defm CPYFM : MOPSMemoryCopyInsns<0b01, "cpyfm">;
9184    defm CPYFE : MOPSMemoryCopyInsns<0b10, "cpyfe">;
9185
9186    defm CPYM : MOPSMemoryMoveInsns<0b01, "cpym">;
9187    defm CPYE : MOPSMemoryMoveInsns<0b10, "cpye">;
9188
9189    defm SETM : MOPSMemorySetInsns<0b01, "setm">;
9190    defm SETE : MOPSMemorySetInsns<0b10, "sete">;
9191  }
9192}
9193let Predicates = [HasMOPS, HasMTE] in {
9194  let Defs = [NZCV] in {
9195    defm SETGP     : MOPSMemorySetTaggingInsns<0b00, "setgp">;
9196  }
9197  let Uses = [NZCV] in {
9198    defm SETGM     : MOPSMemorySetTaggingInsns<0b01, "setgm">;
9199    // Can't use SETGE because it's a reserved name in TargetSelectionDAG.td
9200    defm MOPSSETGE : MOPSMemorySetTaggingInsns<0b10, "setge">;
9201  }
9202}
9203
9204// MOPS Node operands: 0: Dst, 1: Src or Value, 2: Size, 3: Chain
9205// MOPS Node results: 0: Dst writeback, 1: Size writeback, 2: Chain
9206def SDT_AArch64mops : SDTypeProfile<2, 3, [ SDTCisInt<0>, SDTCisInt<1>, SDTCisInt<2> ]>;
9207def AArch64mops_memset : SDNode<"AArch64ISD::MOPS_MEMSET", SDT_AArch64mops>;
9208def AArch64mops_memset_tagging : SDNode<"AArch64ISD::MOPS_MEMSET_TAGGING", SDT_AArch64mops>;
9209def AArch64mops_memcopy : SDNode<"AArch64ISD::MOPS_MEMCOPY", SDT_AArch64mops>;
9210def AArch64mops_memmove : SDNode<"AArch64ISD::MOPS_MEMMOVE", SDT_AArch64mops>;
9211
9212// MOPS operations always contain three 4-byte instructions
9213let Predicates = [HasMOPS], Defs = [NZCV], Size = 12, mayStore = 1 in {
9214  let mayLoad = 1 in {
9215    def MOPSMemoryCopyPseudo : Pseudo<(outs GPR64common:$Rd_wb, GPR64common:$Rs_wb, GPR64:$Rn_wb),
9216                                      (ins GPR64common:$Rd, GPR64common:$Rs, GPR64:$Rn),
9217                                      [], "$Rd = $Rd_wb,$Rs = $Rs_wb,$Rn = $Rn_wb">, Sched<[]>;
9218    def MOPSMemoryMovePseudo : Pseudo<(outs GPR64common:$Rd_wb, GPR64common:$Rs_wb, GPR64:$Rn_wb),
9219                                      (ins GPR64common:$Rd, GPR64common:$Rs, GPR64:$Rn),
9220                                      [], "$Rd = $Rd_wb,$Rs = $Rs_wb,$Rn = $Rn_wb">, Sched<[]>;
9221  }
9222  let mayLoad = 0 in {
9223    def MOPSMemorySetPseudo  : Pseudo<(outs GPR64common:$Rd_wb, GPR64:$Rn_wb),
9224                                      (ins GPR64common:$Rd, GPR64:$Rn, GPR64:$Rm),
9225                                      [], "$Rd = $Rd_wb,$Rn = $Rn_wb">, Sched<[]>;
9226  }
9227}
9228let Predicates = [HasMOPS, HasMTE], Defs = [NZCV], Size = 12, mayLoad = 0, mayStore = 1 in {
9229  def MOPSMemorySetTaggingPseudo : Pseudo<(outs GPR64common:$Rd_wb, GPR64:$Rn_wb),
9230                                          (ins GPR64common:$Rd, GPR64:$Rn, GPR64:$Rm),
9231                                          [], "$Rd = $Rd_wb,$Rn = $Rn_wb">, Sched<[]>;
9232}
9233
9234//-----------------------------------------------------------------------------
9235// v8.3 Pointer Authentication late patterns
9236
9237let Predicates = [HasPAuth] in {
9238def : Pat<(int_ptrauth_blend GPR64:$Rd, imm64_0_65535:$imm),
9239          (MOVKXi GPR64:$Rd, (trunc_imm imm64_0_65535:$imm), 48)>;
9240def : Pat<(int_ptrauth_blend GPR64:$Rd, GPR64:$Rn),
9241          (BFMXri GPR64:$Rd, GPR64:$Rn, 16, 15)>;
9242}
9243
9244//-----------------------------------------------------------------------------
9245
9246// This gets lowered into an instruction sequence of 20 bytes
9247let Defs = [X16, X17], mayStore = 1, isCodeGenOnly = 1, Size = 20 in
9248def StoreSwiftAsyncContext
9249      : Pseudo<(outs), (ins GPR64:$ctx, GPR64sp:$base, simm9:$offset),
9250               []>, Sched<[]>;
9251
9252def AArch64AssertZExtBool : SDNode<"AArch64ISD::ASSERT_ZEXT_BOOL", SDT_assert>;
9253def : Pat<(AArch64AssertZExtBool GPR32:$op),
9254          (i32 GPR32:$op)>;
9255
9256//===----------------------------===//
9257// 2022 Architecture Extensions:
9258//===----------------------------===//
9259
9260def : InstAlias<"clrbhb",  (HINT 22), 0>;
9261let Predicates = [HasCLRBHB] in {
9262  def : InstAlias<"clrbhb",  (HINT 22), 1>;
9263}
9264
9265//===----------------------------------------------------------------------===//
9266// Translation Hardening Extension (FEAT_THE)
9267//===----------------------------------------------------------------------===//
9268defm RCW     : ReadCheckWriteCompareAndSwap;
9269
9270defm RCWCLR  : ReadCheckWriteOperation<0b001, "clr">;
9271defm RCWSET  : ReadCheckWriteOperation<0b011, "set">;
9272defm RCWSWP  : ReadCheckWriteOperation<0b010, "swp">;
9273
9274//===----------------------------------------------------------------------===//
9275// General Data-Processing Instructions (FEAT_V94_DP)
9276//===----------------------------------------------------------------------===//
9277defm ABS : OneOperandData<0b001000, "abs", abs>, Requires<[HasCSSC]>;
9278defm CNT : OneOperandData<0b000111, "cnt", ctpop>, Requires<[HasCSSC]>;
9279defm CTZ : OneOperandData<0b000110, "ctz", cttz>, Requires<[HasCSSC]>;
9280
9281defm SMAX : ComparisonOp<0, 0, "smax", smax>, Requires<[HasCSSC]>;
9282defm SMIN : ComparisonOp<0, 1, "smin", smin>, Requires<[HasCSSC]>;
9283defm UMAX : ComparisonOp<1, 0, "umax", umax>, Requires<[HasCSSC]>;
9284defm UMIN : ComparisonOp<1, 1, "umin", umin>, Requires<[HasCSSC]>;
9285
9286def RPRFM:
9287    I<(outs), (ins rprfop:$Rt, GPR64:$Rm, GPR64sp:$Rn),
9288      "rprfm", "\t$Rt, $Rm, [$Rn]", "", []>,
9289    Sched<[]> {
9290  bits<6> Rt;
9291  bits<5> Rn;
9292  bits<5> Rm;
9293  let Inst{2-0} = Rt{2-0};
9294  let Inst{4-3} = 0b11;
9295  let Inst{9-5} = Rn;
9296  let Inst{11-10} = 0b10;
9297  let Inst{13-12} = Rt{4-3};
9298  let Inst{14} = 0b1;
9299  let Inst{15} = Rt{5};
9300  let Inst{20-16} = Rm;
9301  let Inst{31-21} = 0b11111000101;
9302  let mayLoad = 0;
9303  let mayStore = 0;
9304  let hasSideEffects = 1;
9305  // RPRFM overlaps with PRFM (reg), when the decoder method of PRFM returns
9306  // Fail, the decoder should attempt to decode RPRFM. This requires setting
9307  // the decoder namespace to "Fallback".
9308  let DecoderNamespace = "Fallback";
9309}
9310
9311//===----------------------------------------------------------------------===//
9312// 128-bit Atomics (FEAT_LSE128)
9313//===----------------------------------------------------------------------===//
9314let Predicates = [HasLSE128] in {
9315  def SWPP     : LSE128Base<0b000, 0b00, 0b1, "swpp">;
9316  def SWPPA    : LSE128Base<0b000, 0b10, 0b1, "swppa">;
9317  def SWPPAL   : LSE128Base<0b000, 0b11, 0b1, "swppal">;
9318  def SWPPL    : LSE128Base<0b000, 0b01, 0b1, "swppl">;
9319  def LDCLRP   : LSE128Base<0b001, 0b00, 0b0, "ldclrp">;
9320  def LDCLRPA  : LSE128Base<0b001, 0b10, 0b0, "ldclrpa">;
9321  def LDCLRPAL : LSE128Base<0b001, 0b11, 0b0, "ldclrpal">;
9322  def LDCLRPL  : LSE128Base<0b001, 0b01, 0b0, "ldclrpl">;
9323  def LDSETP   : LSE128Base<0b011, 0b00, 0b0, "ldsetp">;
9324  def LDSETPA  : LSE128Base<0b011, 0b10, 0b0, "ldsetpa">;
9325  def LDSETPAL : LSE128Base<0b011, 0b11, 0b0, "ldsetpal">;
9326  def LDSETPL  : LSE128Base<0b011, 0b01, 0b0, "ldsetpl">;
9327}
9328
9329//===----------------------------------------------------------------------===//
9330// RCPC Instructions (FEAT_LRCPC3)
9331//===----------------------------------------------------------------------===//
9332
9333let Predicates = [HasRCPC3] in {
9334  //                                             size   opc    opc2
9335  def STILPWpre:  BaseLRCPC3IntegerLoadStorePair<0b10, 0b00, 0b0000, (outs GPR64sp:$wback), (ins GPR32:$Rt, GPR32:$Rt2, GPR64sp:$Rn), "stilp", "\t$Rt, $Rt2, [$Rn, #-8]!", "$Rn = $wback">;
9336  def STILPXpre:  BaseLRCPC3IntegerLoadStorePair<0b11, 0b00, 0b0000, (outs GPR64sp:$wback), (ins GPR64:$Rt, GPR64:$Rt2, GPR64sp:$Rn), "stilp", "\t$Rt, $Rt2, [$Rn, #-16]!", "$Rn = $wback">;
9337  def STILPW:     BaseLRCPC3IntegerLoadStorePair<0b10, 0b00, 0b0001, (outs), (ins GPR32:$Rt, GPR32:$Rt2, GPR64sp:$Rn), "stilp", "\t$Rt, $Rt2, [$Rn]", "">;
9338  def STILPX:     BaseLRCPC3IntegerLoadStorePair<0b11, 0b00, 0b0001, (outs), (ins GPR64:$Rt, GPR64:$Rt2, GPR64sp:$Rn), "stilp", "\t$Rt, $Rt2, [$Rn]", "">;
9339  def LDIAPPWpre: BaseLRCPC3IntegerLoadStorePair<0b10, 0b01, 0b0000, (outs GPR64sp:$wback, GPR32:$Rt, GPR32:$Rt2), (ins GPR64sp:$Rn), "ldiapp", "\t$Rt, $Rt2, [$Rn], #8", "$Rn = $wback">;
9340  def LDIAPPXpre: BaseLRCPC3IntegerLoadStorePair<0b11, 0b01, 0b0000, (outs GPR64sp:$wback, GPR64:$Rt, GPR64:$Rt2), (ins GPR64sp:$Rn), "ldiapp", "\t$Rt, $Rt2, [$Rn], #16", "$Rn = $wback">;
9341  def LDIAPPW:    BaseLRCPC3IntegerLoadStorePair<0b10, 0b01, 0b0001, (outs GPR32:$Rt, GPR32:$Rt2), (ins GPR64sp0:$Rn), "ldiapp", "\t$Rt, $Rt2, [$Rn]", "">;
9342  def LDIAPPX:    BaseLRCPC3IntegerLoadStorePair<0b11, 0b01, 0b0001, (outs GPR64:$Rt, GPR64:$Rt2), (ins GPR64sp0:$Rn), "ldiapp", "\t$Rt, $Rt2, [$Rn]", "">;
9343
9344  def : Pat<(AArch64ldiapp GPR64sp:$Rn), (LDIAPPX GPR64sp:$Rn)>;
9345  def : Pat<(AArch64stilp GPR64:$Rt, GPR64:$Rt2, GPR64sp:$Rn), (STILPX GPR64:$Rt, GPR64:$Rt2, GPR64sp:$Rn)>;
9346
9347  // Aliases for when offset=0
9348  def : InstAlias<"stilp\t$Rt, $Rt2, [$Rn, #0]", (STILPW GPR32: $Rt, GPR32: $Rt2, GPR64sp:$Rn)>;
9349  def : InstAlias<"stilp\t$Rt, $Rt2, [$Rn, #0]", (STILPX GPR64: $Rt, GPR64: $Rt2, GPR64sp:$Rn)>;
9350
9351  //                                        size   opc
9352  def STLRWpre:  BaseLRCPC3IntegerLoadStore<0b10, 0b10, (outs GPR64sp:$wback),            (ins GPR32:$Rt, GPR64sp:$Rn), "stlr",  "\t$Rt, [$Rn, #-4]!", "$Rn = $wback">;
9353  def STLRXpre:  BaseLRCPC3IntegerLoadStore<0b11, 0b10, (outs GPR64sp:$wback),            (ins GPR64:$Rt, GPR64sp:$Rn), "stlr",  "\t$Rt, [$Rn, #-8]!", "$Rn = $wback">;
9354  def LDAPRWpre: BaseLRCPC3IntegerLoadStore<0b10, 0b11, (outs GPR64sp:$wback, GPR32:$Rt), (ins GPR64sp:$Rn),            "ldapr", "\t$Rt, [$Rn], #4",   "$Rn = $wback">;
9355  def LDAPRXpre: BaseLRCPC3IntegerLoadStore<0b11, 0b11, (outs GPR64sp:$wback, GPR64:$Rt), (ins GPR64sp:$Rn),            "ldapr", "\t$Rt, [$Rn], #8",   "$Rn = $wback">;
9356}
9357
9358let Predicates = [HasRCPC3, HasNEON] in {
9359  //                                              size   opc regtype
9360  defm STLURb:  LRCPC3NEONLoadStoreUnscaledOffset<0b00, 0b00, FPR8  , (outs), (ins FPR8  :$Rt, GPR64sp:$Rn, simm9:$simm), "stlur">;
9361  defm STLURh:  LRCPC3NEONLoadStoreUnscaledOffset<0b01, 0b00, FPR16 , (outs), (ins FPR16 :$Rt, GPR64sp:$Rn, simm9:$simm), "stlur">;
9362  defm STLURs:  LRCPC3NEONLoadStoreUnscaledOffset<0b10, 0b00, FPR32 , (outs), (ins FPR32 :$Rt, GPR64sp:$Rn, simm9:$simm), "stlur">;
9363  defm STLURd:  LRCPC3NEONLoadStoreUnscaledOffset<0b11, 0b00, FPR64 , (outs), (ins FPR64 :$Rt, GPR64sp:$Rn, simm9:$simm), "stlur">;
9364  defm STLURq:  LRCPC3NEONLoadStoreUnscaledOffset<0b00, 0b10, FPR128, (outs), (ins FPR128:$Rt, GPR64sp:$Rn, simm9:$simm), "stlur">;
9365  defm LDAPURb: LRCPC3NEONLoadStoreUnscaledOffset<0b00, 0b01, FPR8  , (outs FPR8  :$Rt), (ins GPR64sp:$Rn, simm9:$simm), "ldapur">;
9366  defm LDAPURh: LRCPC3NEONLoadStoreUnscaledOffset<0b01, 0b01, FPR16 , (outs FPR16 :$Rt), (ins GPR64sp:$Rn, simm9:$simm), "ldapur">;
9367  defm LDAPURs: LRCPC3NEONLoadStoreUnscaledOffset<0b10, 0b01, FPR32 , (outs FPR32 :$Rt), (ins GPR64sp:$Rn, simm9:$simm), "ldapur">;
9368  defm LDAPURd: LRCPC3NEONLoadStoreUnscaledOffset<0b11, 0b01, FPR64 , (outs FPR64 :$Rt), (ins GPR64sp:$Rn, simm9:$simm), "ldapur">;
9369  defm LDAPURq: LRCPC3NEONLoadStoreUnscaledOffset<0b00, 0b11, FPR128, (outs FPR128:$Rt), (ins GPR64sp:$Rn, simm9:$simm), "ldapur">;
9370
9371  //                                L
9372  def STL1:  LRCPC3NEONLdStSingle<0b0, (outs), (ins VecListOned:$Vt, VectorIndexD:$Q, GPR64sp:$Rn) , "stl1", "">;
9373  def LDAP1: LRCPC3NEONLdStSingle<0b1, (outs VecListOned:$dst), (ins VecListOned:$Vt, VectorIndexD:$Q, GPR64sp0:$Rn), "ldap1", "$Vt = $dst">;
9374
9375  // Aliases for when offset=0
9376  def : InstAlias<"stl1\t$Vt$Q, [$Rn, #0]", (STL1 VecListOned:$Vt, VectorIndexD:$Q, GPR64sp:$Rn)>;
9377}
9378
9379//===----------------------------------------------------------------------===//
9380// 128-bit System Instructions (FEAT_SYSINSTR128)
9381//===----------------------------------------------------------------------===//
9382let Predicates = [HasD128] in {
9383  def SYSPxt  : SystemPXtI<0, "sysp">;
9384
9385  def SYSPxt_XZR
9386    : BaseSystemI<0, (outs),
9387        (ins imm0_7:$op1, sys_cr_op:$Cn, sys_cr_op:$Cm, imm0_7:$op2, SyspXzrPairOperand:$xzr_pair),
9388        "sysp", "\t$op1, $Cn, $Cm, $op2, $xzr_pair">,
9389      Sched<[WriteSys]>
9390  {
9391    // Had to use a custom decoder because tablegen interprets this as having 4 fields (why?)
9392    // and therefore autogenerates a decoder that builds an MC representation that has 4 fields
9393    // (decodeToMCInst), but when printing we expect the MC representation to have 5 fields (one
9394    // extra for the XZR) because AArch64InstPrinter::printInstruction in AArch64GenAsmWriter.inc
9395    // is based off of the asm template (maybe) and therefore wants to print 5 operands.
9396    // I could add a bits<5> xzr_pair. But without a way to constrain it to 0b11111 here it would
9397    // overlap with the main SYSP instruction.
9398    let DecoderMethod = "DecodeSyspXzrInstruction";
9399    bits<3> op1;
9400    bits<4> Cn;
9401    bits<4> Cm;
9402    bits<3> op2;
9403    let Inst{22}    = 0b1; // override BaseSystemI
9404    let Inst{20-19} = 0b01;
9405    let Inst{18-16} = op1;
9406    let Inst{15-12} = Cn;
9407    let Inst{11-8}  = Cm;
9408    let Inst{7-5}   = op2;
9409    let Inst{4-0}   = 0b11111;
9410  }
9411
9412  def : InstAlias<"sysp $op1, $Cn, $Cm, $op2",
9413                  (SYSPxt_XZR imm0_7:$op1, sys_cr_op:$Cn, sys_cr_op:$Cm, imm0_7:$op2, XZR)>;
9414}
9415
9416//---
9417// 128-bit System Registers (FEAT_SYSREG128)
9418//---
9419
9420// Instruction encoding:
9421//
9422//          31       22|21|20|19|18 16|15 12|11 8|7 5|4 0
9423// MRRS      1101010101| 1| 1|o0|  op1|   Cn|  Cm|op2| Rt
9424// MSRR      1101010101| 0| 1|o0|  op1|   Cn|  Cm|op2| Rt
9425
9426// Instruction syntax:
9427//
9428// MRRS <Xt>, <Xt+1>, <sysreg|S<op0>_<op1>_<Cn>_<Cm>_<op2>>
9429// MSRR <sysreg|S<op0>_<op1>_<Cn>_<Cm>_<op2>>, <Xt>, <Xt+1>
9430//
9431// ...where t is even (X0, X2, etc).
9432
9433let Predicates = [HasD128] in {
9434  def MRRS : RtSystemI128<1,
9435    (outs MrrsMssrPairClassOperand:$Rt), (ins mrs_sysreg_op:$systemreg),
9436    "mrrs", "\t$Rt, $systemreg">
9437  {
9438    bits<16> systemreg;
9439    let Inst{20-5} = systemreg;
9440  }
9441
9442  def MSRR : RtSystemI128<0,
9443    (outs),  (ins msr_sysreg_op:$systemreg, MrrsMssrPairClassOperand:$Rt),
9444    "msrr", "\t$systemreg, $Rt">
9445  {
9446    bits<16> systemreg;
9447    let Inst{20-5} = systemreg;
9448  }
9449}
9450
9451//===----------------------------===//
9452// 2023 Architecture Extensions:
9453//===----------------------------===//
9454
9455let Predicates = [HasFP8] in {
9456  defm F1CVTL  : SIMDMixedTwoVectorFP8<0b00, "f1cvtl">;
9457  defm F2CVTL  : SIMDMixedTwoVectorFP8<0b01, "f2cvtl">;
9458  defm BF1CVTL : SIMDMixedTwoVectorFP8<0b10, "bf1cvtl">;
9459  defm BF2CVTL : SIMDMixedTwoVectorFP8<0b11, "bf2cvtl">;
9460  defm FCVTN_F16_F8 : SIMDThreeSameSizeVectorCvt<"fcvtn">;
9461  defm FCVTN_F32_F8 : SIMDThreeVectorCvt<"fcvtn">;
9462  defm FSCALE : SIMDThreeSameVectorFP<0b1, 0b1, 0b111, "fscale", null_frag>;
9463} // End let Predicates = [HasFP8]
9464
9465let Predicates = [HasFAMINMAX] in {
9466 defm FAMAX : SIMDThreeSameVectorFP<0b0, 0b1, 0b011, "famax", null_frag>;
9467 defm FAMIN : SIMDThreeSameVectorFP<0b1, 0b1, 0b011, "famin", null_frag>;
9468} // End let Predicates = [HasFAMAXMIN]
9469
9470let Predicates = [HasFP8FMA] in {
9471 defm FMLALBlane : SIMDThreeSameVectorMLAIndex<0b0, "fmlalb">;
9472 defm FMLALTlane : SIMDThreeSameVectorMLAIndex<0b1, "fmlalt">;
9473 defm FMLALLBBlane : SIMDThreeSameVectorMLALIndex<0b0, 0b00, "fmlallbb">;
9474 defm FMLALLBTlane : SIMDThreeSameVectorMLALIndex<0b0, 0b01, "fmlallbt">;
9475 defm FMLALLTBlane : SIMDThreeSameVectorMLALIndex<0b1, 0b00, "fmlalltb">;
9476 defm FMLALLTTlane : SIMDThreeSameVectorMLALIndex<0b1, 0b01, "fmlalltt">;
9477
9478 defm FMLALB : SIMDThreeSameVectorMLA<0b0, "fmlalb">;
9479 defm FMLALT : SIMDThreeSameVectorMLA<0b1, "fmlalt">;
9480 defm FMLALLBB : SIMDThreeSameVectorMLAL<0b0, 0b00, "fmlallbb">;
9481 defm FMLALLBT : SIMDThreeSameVectorMLAL<0b0, 0b01, "fmlallbt">;
9482 defm FMLALLTB : SIMDThreeSameVectorMLAL<0b1, 0b00, "fmlalltb">;
9483 defm FMLALLTT : SIMDThreeSameVectorMLAL<0b1, 0b01, "fmlalltt">;
9484} // End let Predicates = [HasFP8FMA]
9485
9486let Predicates = [HasFP8DOT2] in {
9487 defm FDOTlane : SIMDThreeSameVectorFP8DOT2Index<"fdot">;
9488 defm FDOT : SIMDThreeSameVectorDOT2<"fdot">;
9489} // End let Predicates = [HasFP8DOT2]
9490
9491let Predicates = [HasFP8DOT4] in {
9492 defm FDOTlane : SIMDThreeSameVectorFP8DOT4Index<"fdot">;
9493 defm FDOT : SIMDThreeSameVectorDOT4<"fdot">;
9494} // End let Predicates = [HasFP8DOT4]
9495
9496//===----------------------------------------------------------------------===//
9497// Checked Pointer Arithmetic (FEAT_CPA)
9498//===----------------------------------------------------------------------===//
9499let Predicates = [HasCPA] in {
9500  // Scalar add/subtract
9501  defm ADDPT : AddSubCPA<0, "addpt">;
9502  defm SUBPT : AddSubCPA<1, "subpt">;
9503
9504  // Scalar multiply-add/subtract
9505  def MADDPT : MulAccumCPA<0, "maddpt">;
9506  def MSUBPT : MulAccumCPA<1, "msubpt">;
9507}
9508
9509include "AArch64InstrAtomics.td"
9510include "AArch64SVEInstrInfo.td"
9511include "AArch64SMEInstrInfo.td"
9512include "AArch64InstrGISel.td"
9513