1 //=- AArch64InstrInfo.td - Describe the AArch64 Instructions -*- tablegen -*-=//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // AArch64 Instruction definitions.
12 //===----------------------------------------------------------------------===//
14 //===----------------------------------------------------------------------===//
15 // ARM Instruction Predicate Definitions.
17 def HasV8_1a : Predicate<"Subtarget->hasV8_1aOps()">,
18 AssemblerPredicate<"HasV8_1aOps", "armv8.1a">;
19 def HasV8_2a : Predicate<"Subtarget->hasV8_2aOps()">,
20 AssemblerPredicate<"HasV8_2aOps", "armv8.2a">;
21 def HasFPARMv8 : Predicate<"Subtarget->hasFPARMv8()">,
22 AssemblerPredicate<"FeatureFPARMv8", "fp-armv8">;
23 def HasNEON : Predicate<"Subtarget->hasNEON()">,
24 AssemblerPredicate<"FeatureNEON", "neon">;
25 def HasCrypto : Predicate<"Subtarget->hasCrypto()">,
26 AssemblerPredicate<"FeatureCrypto", "crypto">;
27 def HasCRC : Predicate<"Subtarget->hasCRC()">,
28 AssemblerPredicate<"FeatureCRC", "crc">;
29 def HasLSE : Predicate<"Subtarget->hasLSE()">,
30 AssemblerPredicate<"FeatureLSE", "lse">;
31 def HasRAS : Predicate<"Subtarget->hasRAS()">,
32 AssemblerPredicate<"FeatureRAS", "ras">;
33 def HasRDM : Predicate<"Subtarget->hasRDM()">,
34 AssemblerPredicate<"FeatureRDM", "rdm">;
35 def HasPerfMon : Predicate<"Subtarget->hasPerfMon()">;
36 def HasFullFP16 : Predicate<"Subtarget->hasFullFP16()">,
37 AssemblerPredicate<"FeatureFullFP16", "fullfp16">;
38 def HasSPE : Predicate<"Subtarget->hasSPE()">,
39 AssemblerPredicate<"FeatureSPE", "spe">;
41 def IsLE : Predicate<"Subtarget->isLittleEndian()">;
42 def IsBE : Predicate<"!Subtarget->isLittleEndian()">;
43 def UseAlternateSExtLoadCVTF32
44 : Predicate<"Subtarget->useAlternateSExtLoadCVTF32Pattern()">;
46 def UseNegativeImmediates
47 : Predicate<"false">, AssemblerPredicate<"!FeatureNoNegativeImmediates",
48 "NegativeImmediates">;
51 //===----------------------------------------------------------------------===//
52 // AArch64-specific DAG Nodes.
55 // SDTBinaryArithWithFlagsOut - RES1, FLAGS = op LHS, RHS
56 def SDTBinaryArithWithFlagsOut : SDTypeProfile<2, 2,
59 SDTCisInt<0>, SDTCisVT<1, i32>]>;
61 // SDTBinaryArithWithFlagsIn - RES1, FLAGS = op LHS, RHS, FLAGS
62 def SDTBinaryArithWithFlagsIn : SDTypeProfile<1, 3,
68 // SDTBinaryArithWithFlagsInOut - RES1, FLAGS = op LHS, RHS, FLAGS
69 def SDTBinaryArithWithFlagsInOut : SDTypeProfile<2, 3,
76 def SDT_AArch64Brcond : SDTypeProfile<0, 3,
77 [SDTCisVT<0, OtherVT>, SDTCisVT<1, i32>,
79 def SDT_AArch64cbz : SDTypeProfile<0, 2, [SDTCisInt<0>, SDTCisVT<1, OtherVT>]>;
80 def SDT_AArch64tbz : SDTypeProfile<0, 3, [SDTCisInt<0>, SDTCisInt<1>,
81 SDTCisVT<2, OtherVT>]>;
84 def SDT_AArch64CSel : SDTypeProfile<1, 4,
89 def SDT_AArch64CCMP : SDTypeProfile<1, 5,
96 def SDT_AArch64FCCMP : SDTypeProfile<1, 5,
103 def SDT_AArch64FCmp : SDTypeProfile<0, 2,
105 SDTCisSameAs<0, 1>]>;
106 def SDT_AArch64Dup : SDTypeProfile<1, 1, [SDTCisVec<0>]>;
107 def SDT_AArch64DupLane : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisInt<2>]>;
108 def SDT_AArch64Zip : SDTypeProfile<1, 2, [SDTCisVec<0>,
110 SDTCisSameAs<0, 2>]>;
111 def SDT_AArch64MOVIedit : SDTypeProfile<1, 1, [SDTCisInt<1>]>;
112 def SDT_AArch64MOVIshift : SDTypeProfile<1, 2, [SDTCisInt<1>, SDTCisInt<2>]>;
113 def SDT_AArch64vecimm : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0,1>,
114 SDTCisInt<2>, SDTCisInt<3>]>;
115 def SDT_AArch64UnaryVec: SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0,1>]>;
116 def SDT_AArch64ExtVec: SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0,1>,
117 SDTCisSameAs<0,2>, SDTCisInt<3>]>;
118 def SDT_AArch64vshift : SDTypeProfile<1, 2, [SDTCisSameAs<0,1>, SDTCisInt<2>]>;
120 def SDT_AArch64unvec : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0,1>]>;
121 def SDT_AArch64fcmpz : SDTypeProfile<1, 1, []>;
122 def SDT_AArch64fcmp : SDTypeProfile<1, 2, [SDTCisSameAs<1,2>]>;
123 def SDT_AArch64binvec : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisSameAs<0,1>,
125 def SDT_AArch64trivec : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0,1>,
128 def SDT_AArch64TCRET : SDTypeProfile<0, 2, [SDTCisPtrTy<0>]>;
129 def SDT_AArch64PREFETCH : SDTypeProfile<0, 2, [SDTCisVT<0, i32>, SDTCisPtrTy<1>]>;
131 def SDT_AArch64ITOF : SDTypeProfile<1, 1, [SDTCisFP<0>, SDTCisSameAs<0,1>]>;
133 def SDT_AArch64TLSDescCall : SDTypeProfile<0, -2, [SDTCisPtrTy<0>,
136 // Generates the general dynamic sequences, i.e.
137 // adrp x0, :tlsdesc:var
138 // ldr x1, [x0, #:tlsdesc_lo12:var]
139 // add x0, x0, #:tlsdesc_lo12:var
143 // (the TPIDR_EL0 offset is put directly in X0, hence no "result" here)
144 // number of operands (the variable)
145 def SDT_AArch64TLSDescCallSeq : SDTypeProfile<0,1,
148 def SDT_AArch64WrapperLarge : SDTypeProfile<1, 4,
149 [SDTCisVT<0, i64>, SDTCisVT<1, i32>,
150 SDTCisSameAs<1, 2>, SDTCisSameAs<1, 3>,
151 SDTCisSameAs<1, 4>]>;
155 def AArch64adrp : SDNode<"AArch64ISD::ADRP", SDTIntUnaryOp, []>;
156 def AArch64addlow : SDNode<"AArch64ISD::ADDlow", SDTIntBinOp, []>;
157 def AArch64LOADgot : SDNode<"AArch64ISD::LOADgot", SDTIntUnaryOp>;
158 def AArch64callseq_start : SDNode<"ISD::CALLSEQ_START",
159 SDCallSeqStart<[ SDTCisVT<0, i32>,
161 [SDNPHasChain, SDNPOutGlue]>;
162 def AArch64callseq_end : SDNode<"ISD::CALLSEQ_END",
163 SDCallSeqEnd<[ SDTCisVT<0, i32>,
165 [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue]>;
166 def AArch64call : SDNode<"AArch64ISD::CALL",
167 SDTypeProfile<0, -1, [SDTCisPtrTy<0>]>,
168 [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue,
170 def AArch64brcond : SDNode<"AArch64ISD::BRCOND", SDT_AArch64Brcond,
172 def AArch64cbz : SDNode<"AArch64ISD::CBZ", SDT_AArch64cbz,
174 def AArch64cbnz : SDNode<"AArch64ISD::CBNZ", SDT_AArch64cbz,
176 def AArch64tbz : SDNode<"AArch64ISD::TBZ", SDT_AArch64tbz,
178 def AArch64tbnz : SDNode<"AArch64ISD::TBNZ", SDT_AArch64tbz,
182 def AArch64csel : SDNode<"AArch64ISD::CSEL", SDT_AArch64CSel>;
183 def AArch64csinv : SDNode<"AArch64ISD::CSINV", SDT_AArch64CSel>;
184 def AArch64csneg : SDNode<"AArch64ISD::CSNEG", SDT_AArch64CSel>;
185 def AArch64csinc : SDNode<"AArch64ISD::CSINC", SDT_AArch64CSel>;
186 def AArch64retflag : SDNode<"AArch64ISD::RET_FLAG", SDTNone,
187 [SDNPHasChain, SDNPOptInGlue, SDNPVariadic]>;
188 def AArch64adc : SDNode<"AArch64ISD::ADC", SDTBinaryArithWithFlagsIn >;
189 def AArch64sbc : SDNode<"AArch64ISD::SBC", SDTBinaryArithWithFlagsIn>;
190 def AArch64add_flag : SDNode<"AArch64ISD::ADDS", SDTBinaryArithWithFlagsOut,
192 def AArch64sub_flag : SDNode<"AArch64ISD::SUBS", SDTBinaryArithWithFlagsOut>;
193 def AArch64and_flag : SDNode<"AArch64ISD::ANDS", SDTBinaryArithWithFlagsOut,
195 def AArch64adc_flag : SDNode<"AArch64ISD::ADCS", SDTBinaryArithWithFlagsInOut>;
196 def AArch64sbc_flag : SDNode<"AArch64ISD::SBCS", SDTBinaryArithWithFlagsInOut>;
198 def AArch64ccmp : SDNode<"AArch64ISD::CCMP", SDT_AArch64CCMP>;
199 def AArch64ccmn : SDNode<"AArch64ISD::CCMN", SDT_AArch64CCMP>;
200 def AArch64fccmp : SDNode<"AArch64ISD::FCCMP", SDT_AArch64FCCMP>;
202 def AArch64threadpointer : SDNode<"AArch64ISD::THREAD_POINTER", SDTPtrLeaf>;
204 def AArch64fcmp : SDNode<"AArch64ISD::FCMP", SDT_AArch64FCmp>;
206 def AArch64dup : SDNode<"AArch64ISD::DUP", SDT_AArch64Dup>;
207 def AArch64duplane8 : SDNode<"AArch64ISD::DUPLANE8", SDT_AArch64DupLane>;
208 def AArch64duplane16 : SDNode<"AArch64ISD::DUPLANE16", SDT_AArch64DupLane>;
209 def AArch64duplane32 : SDNode<"AArch64ISD::DUPLANE32", SDT_AArch64DupLane>;
210 def AArch64duplane64 : SDNode<"AArch64ISD::DUPLANE64", SDT_AArch64DupLane>;
212 def AArch64zip1 : SDNode<"AArch64ISD::ZIP1", SDT_AArch64Zip>;
213 def AArch64zip2 : SDNode<"AArch64ISD::ZIP2", SDT_AArch64Zip>;
214 def AArch64uzp1 : SDNode<"AArch64ISD::UZP1", SDT_AArch64Zip>;
215 def AArch64uzp2 : SDNode<"AArch64ISD::UZP2", SDT_AArch64Zip>;
216 def AArch64trn1 : SDNode<"AArch64ISD::TRN1", SDT_AArch64Zip>;
217 def AArch64trn2 : SDNode<"AArch64ISD::TRN2", SDT_AArch64Zip>;
219 def AArch64movi_edit : SDNode<"AArch64ISD::MOVIedit", SDT_AArch64MOVIedit>;
220 def AArch64movi_shift : SDNode<"AArch64ISD::MOVIshift", SDT_AArch64MOVIshift>;
221 def AArch64movi_msl : SDNode<"AArch64ISD::MOVImsl", SDT_AArch64MOVIshift>;
222 def AArch64mvni_shift : SDNode<"AArch64ISD::MVNIshift", SDT_AArch64MOVIshift>;
223 def AArch64mvni_msl : SDNode<"AArch64ISD::MVNImsl", SDT_AArch64MOVIshift>;
224 def AArch64movi : SDNode<"AArch64ISD::MOVI", SDT_AArch64MOVIedit>;
225 def AArch64fmov : SDNode<"AArch64ISD::FMOV", SDT_AArch64MOVIedit>;
227 def AArch64rev16 : SDNode<"AArch64ISD::REV16", SDT_AArch64UnaryVec>;
228 def AArch64rev32 : SDNode<"AArch64ISD::REV32", SDT_AArch64UnaryVec>;
229 def AArch64rev64 : SDNode<"AArch64ISD::REV64", SDT_AArch64UnaryVec>;
230 def AArch64ext : SDNode<"AArch64ISD::EXT", SDT_AArch64ExtVec>;
232 def AArch64vashr : SDNode<"AArch64ISD::VASHR", SDT_AArch64vshift>;
233 def AArch64vlshr : SDNode<"AArch64ISD::VLSHR", SDT_AArch64vshift>;
234 def AArch64vshl : SDNode<"AArch64ISD::VSHL", SDT_AArch64vshift>;
235 def AArch64sqshli : SDNode<"AArch64ISD::SQSHL_I", SDT_AArch64vshift>;
236 def AArch64uqshli : SDNode<"AArch64ISD::UQSHL_I", SDT_AArch64vshift>;
237 def AArch64sqshlui : SDNode<"AArch64ISD::SQSHLU_I", SDT_AArch64vshift>;
238 def AArch64srshri : SDNode<"AArch64ISD::SRSHR_I", SDT_AArch64vshift>;
239 def AArch64urshri : SDNode<"AArch64ISD::URSHR_I", SDT_AArch64vshift>;
241 def AArch64not: SDNode<"AArch64ISD::NOT", SDT_AArch64unvec>;
242 def AArch64bit: SDNode<"AArch64ISD::BIT", SDT_AArch64trivec>;
243 def AArch64bsl: SDNode<"AArch64ISD::BSL", SDT_AArch64trivec>;
245 def AArch64cmeq: SDNode<"AArch64ISD::CMEQ", SDT_AArch64binvec>;
246 def AArch64cmge: SDNode<"AArch64ISD::CMGE", SDT_AArch64binvec>;
247 def AArch64cmgt: SDNode<"AArch64ISD::CMGT", SDT_AArch64binvec>;
248 def AArch64cmhi: SDNode<"AArch64ISD::CMHI", SDT_AArch64binvec>;
249 def AArch64cmhs: SDNode<"AArch64ISD::CMHS", SDT_AArch64binvec>;
251 def AArch64fcmeq: SDNode<"AArch64ISD::FCMEQ", SDT_AArch64fcmp>;
252 def AArch64fcmge: SDNode<"AArch64ISD::FCMGE", SDT_AArch64fcmp>;
253 def AArch64fcmgt: SDNode<"AArch64ISD::FCMGT", SDT_AArch64fcmp>;
255 def AArch64cmeqz: SDNode<"AArch64ISD::CMEQz", SDT_AArch64unvec>;
256 def AArch64cmgez: SDNode<"AArch64ISD::CMGEz", SDT_AArch64unvec>;
257 def AArch64cmgtz: SDNode<"AArch64ISD::CMGTz", SDT_AArch64unvec>;
258 def AArch64cmlez: SDNode<"AArch64ISD::CMLEz", SDT_AArch64unvec>;
259 def AArch64cmltz: SDNode<"AArch64ISD::CMLTz", SDT_AArch64unvec>;
260 def AArch64cmtst : PatFrag<(ops node:$LHS, node:$RHS),
261 (AArch64not (AArch64cmeqz (and node:$LHS, node:$RHS)))>;
263 def AArch64fcmeqz: SDNode<"AArch64ISD::FCMEQz", SDT_AArch64fcmpz>;
264 def AArch64fcmgez: SDNode<"AArch64ISD::FCMGEz", SDT_AArch64fcmpz>;
265 def AArch64fcmgtz: SDNode<"AArch64ISD::FCMGTz", SDT_AArch64fcmpz>;
266 def AArch64fcmlez: SDNode<"AArch64ISD::FCMLEz", SDT_AArch64fcmpz>;
267 def AArch64fcmltz: SDNode<"AArch64ISD::FCMLTz", SDT_AArch64fcmpz>;
269 def AArch64bici: SDNode<"AArch64ISD::BICi", SDT_AArch64vecimm>;
270 def AArch64orri: SDNode<"AArch64ISD::ORRi", SDT_AArch64vecimm>;
272 def AArch64neg : SDNode<"AArch64ISD::NEG", SDT_AArch64unvec>;
274 def AArch64tcret: SDNode<"AArch64ISD::TC_RETURN", SDT_AArch64TCRET,
275 [SDNPHasChain, SDNPOptInGlue, SDNPVariadic]>;
277 def AArch64Prefetch : SDNode<"AArch64ISD::PREFETCH", SDT_AArch64PREFETCH,
278 [SDNPHasChain, SDNPSideEffect]>;
280 def AArch64sitof: SDNode<"AArch64ISD::SITOF", SDT_AArch64ITOF>;
281 def AArch64uitof: SDNode<"AArch64ISD::UITOF", SDT_AArch64ITOF>;
283 def AArch64tlsdesc_callseq : SDNode<"AArch64ISD::TLSDESC_CALLSEQ",
284 SDT_AArch64TLSDescCallSeq,
285 [SDNPInGlue, SDNPOutGlue, SDNPHasChain,
289 def AArch64WrapperLarge : SDNode<"AArch64ISD::WrapperLarge",
290 SDT_AArch64WrapperLarge>;
292 def AArch64NvCast : SDNode<"AArch64ISD::NVCAST", SDTUnaryOp>;
294 def SDT_AArch64mull : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
295 SDTCisSameAs<1, 2>]>;
296 def AArch64smull : SDNode<"AArch64ISD::SMULL", SDT_AArch64mull>;
297 def AArch64umull : SDNode<"AArch64ISD::UMULL", SDT_AArch64mull>;
299 def AArch64frecpe : SDNode<"AArch64ISD::FRECPE", SDTFPUnaryOp>;
300 def AArch64frecps : SDNode<"AArch64ISD::FRECPS", SDTFPBinOp>;
301 def AArch64frsqrte : SDNode<"AArch64ISD::FRSQRTE", SDTFPUnaryOp>;
302 def AArch64frsqrts : SDNode<"AArch64ISD::FRSQRTS", SDTFPBinOp>;
304 def AArch64saddv : SDNode<"AArch64ISD::SADDV", SDT_AArch64UnaryVec>;
305 def AArch64uaddv : SDNode<"AArch64ISD::UADDV", SDT_AArch64UnaryVec>;
306 def AArch64sminv : SDNode<"AArch64ISD::SMINV", SDT_AArch64UnaryVec>;
307 def AArch64uminv : SDNode<"AArch64ISD::UMINV", SDT_AArch64UnaryVec>;
308 def AArch64smaxv : SDNode<"AArch64ISD::SMAXV", SDT_AArch64UnaryVec>;
309 def AArch64umaxv : SDNode<"AArch64ISD::UMAXV", SDT_AArch64UnaryVec>;
311 //===----------------------------------------------------------------------===//
313 //===----------------------------------------------------------------------===//
315 // AArch64 Instruction Predicate Definitions.
316 def IsDarwin : Predicate<"Subtarget->isTargetDarwin()">;
317 def IsNotDarwin: Predicate<"!Subtarget->isTargetDarwin()">;
319 // We could compute these on a per-module basis but doing so requires accessing
320 // the Function object through the <Target>Subtarget and objections were raised
321 // to that (see post-commit review comments for r301750).
322 let RecomputePerFunction = 1 in {
323 def ForCodeSize : Predicate<"MF->getFunction()->optForSize()">;
324 def NotForCodeSize : Predicate<"!MF->getFunction()->optForSize()">;
327 include "AArch64InstrFormats.td"
329 //===----------------------------------------------------------------------===//
331 //===----------------------------------------------------------------------===//
332 // Miscellaneous instructions.
333 //===----------------------------------------------------------------------===//
335 let Defs = [SP], Uses = [SP], hasSideEffects = 1, isCodeGenOnly = 1 in {
336 // We set Sched to empty list because we expect these instructions to simply get
337 // removed in most cases.
338 def ADJCALLSTACKDOWN : Pseudo<(outs), (ins i32imm:$amt1, i32imm:$amt2),
339 [(AArch64callseq_start timm:$amt1, timm:$amt2)]>,
341 def ADJCALLSTACKUP : Pseudo<(outs), (ins i32imm:$amt1, i32imm:$amt2),
342 [(AArch64callseq_end timm:$amt1, timm:$amt2)]>,
344 } // Defs = [SP], Uses = [SP], hasSideEffects = 1, isCodeGenOnly = 1
346 let isReMaterializable = 1, isCodeGenOnly = 1 in {
347 // FIXME: The following pseudo instructions are only needed because remat
348 // cannot handle multiple instructions. When that changes, they can be
349 // removed, along with the AArch64Wrapper node.
351 let AddedComplexity = 10 in
352 def LOADgot : Pseudo<(outs GPR64:$dst), (ins i64imm:$addr),
353 [(set GPR64:$dst, (AArch64LOADgot tglobaladdr:$addr))]>,
356 // The MOVaddr instruction should match only when the add is not folded
357 // into a load or store address.
359 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
360 [(set GPR64:$dst, (AArch64addlow (AArch64adrp tglobaladdr:$hi),
361 tglobaladdr:$low))]>,
362 Sched<[WriteAdrAdr]>;
364 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
365 [(set GPR64:$dst, (AArch64addlow (AArch64adrp tjumptable:$hi),
367 Sched<[WriteAdrAdr]>;
369 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
370 [(set GPR64:$dst, (AArch64addlow (AArch64adrp tconstpool:$hi),
372 Sched<[WriteAdrAdr]>;
374 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
375 [(set GPR64:$dst, (AArch64addlow (AArch64adrp tblockaddress:$hi),
376 tblockaddress:$low))]>,
377 Sched<[WriteAdrAdr]>;
379 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
380 [(set GPR64:$dst, (AArch64addlow (AArch64adrp tglobaltlsaddr:$hi),
381 tglobaltlsaddr:$low))]>,
382 Sched<[WriteAdrAdr]>;
384 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
385 [(set GPR64:$dst, (AArch64addlow (AArch64adrp texternalsym:$hi),
386 texternalsym:$low))]>,
387 Sched<[WriteAdrAdr]>;
389 } // isReMaterializable, isCodeGenOnly
391 def : Pat<(AArch64LOADgot tglobaltlsaddr:$addr),
392 (LOADgot tglobaltlsaddr:$addr)>;
394 def : Pat<(AArch64LOADgot texternalsym:$addr),
395 (LOADgot texternalsym:$addr)>;
397 def : Pat<(AArch64LOADgot tconstpool:$addr),
398 (LOADgot tconstpool:$addr)>;
400 //===----------------------------------------------------------------------===//
401 // System instructions.
402 //===----------------------------------------------------------------------===//
404 def HINT : HintI<"hint">;
405 def : InstAlias<"nop", (HINT 0b000)>;
406 def : InstAlias<"yield",(HINT 0b001)>;
407 def : InstAlias<"wfe", (HINT 0b010)>;
408 def : InstAlias<"wfi", (HINT 0b011)>;
409 def : InstAlias<"sev", (HINT 0b100)>;
410 def : InstAlias<"sevl", (HINT 0b101)>;
411 def : InstAlias<"esb", (HINT 0b10000)>, Requires<[HasRAS]>;
413 // v8.2a Statistical Profiling extension
414 def : InstAlias<"psb $op", (HINT psbhint_op:$op)>, Requires<[HasSPE]>;
416 // As far as LLVM is concerned this writes to the system's exclusive monitors.
417 let mayLoad = 1, mayStore = 1 in
418 def CLREX : CRmSystemI<imm0_15, 0b010, "clrex">;
420 // NOTE: ideally, this would have mayStore = 0, mayLoad = 0, but we cannot
421 // model patterns with sufficiently fine granularity.
422 let mayLoad = ?, mayStore = ? in {
423 def DMB : CRmSystemI<barrier_op, 0b101, "dmb",
424 [(int_aarch64_dmb (i32 imm32_0_15:$CRm))]>;
426 def DSB : CRmSystemI<barrier_op, 0b100, "dsb",
427 [(int_aarch64_dsb (i32 imm32_0_15:$CRm))]>;
429 def ISB : CRmSystemI<barrier_op, 0b110, "isb",
430 [(int_aarch64_isb (i32 imm32_0_15:$CRm))]>;
433 def : InstAlias<"clrex", (CLREX 0xf)>;
434 def : InstAlias<"isb", (ISB 0xf)>;
438 def MSRpstateImm1 : MSRpstateImm0_1;
439 def MSRpstateImm4 : MSRpstateImm0_15;
441 // The thread pointer (on Linux, at least, where this has been implemented) is
442 // TPIDR_EL0. Add pseudo op so we can mark it as not having any side effects.
443 let hasSideEffects = 0 in
444 def MOVbaseTLS : Pseudo<(outs GPR64:$dst), (ins),
445 [(set GPR64:$dst, AArch64threadpointer)]>, Sched<[WriteSys]>;
447 // The cycle counter PMC register is PMCCNTR_EL0.
448 let Predicates = [HasPerfMon] in
449 def : Pat<(readcyclecounter), (MRS 0xdce8)>;
451 // Generic system instructions
452 def SYSxt : SystemXtI<0, "sys">;
453 def SYSLxt : SystemLXtI<1, "sysl">;
455 def : InstAlias<"sys $op1, $Cn, $Cm, $op2",
456 (SYSxt imm0_7:$op1, sys_cr_op:$Cn,
457 sys_cr_op:$Cm, imm0_7:$op2, XZR)>;
459 //===----------------------------------------------------------------------===//
460 // Move immediate instructions.
461 //===----------------------------------------------------------------------===//
463 defm MOVK : InsertImmediate<0b11, "movk">;
464 defm MOVN : MoveImmediate<0b00, "movn">;
466 let PostEncoderMethod = "fixMOVZ" in
467 defm MOVZ : MoveImmediate<0b10, "movz">;
469 // First group of aliases covers an implicit "lsl #0".
470 def : InstAlias<"movk $dst, $imm", (MOVKWi GPR32:$dst, imm0_65535:$imm, 0)>;
471 def : InstAlias<"movk $dst, $imm", (MOVKXi GPR64:$dst, imm0_65535:$imm, 0)>;
472 def : InstAlias<"movn $dst, $imm", (MOVNWi GPR32:$dst, imm0_65535:$imm, 0)>;
473 def : InstAlias<"movn $dst, $imm", (MOVNXi GPR64:$dst, imm0_65535:$imm, 0)>;
474 def : InstAlias<"movz $dst, $imm", (MOVZWi GPR32:$dst, imm0_65535:$imm, 0)>;
475 def : InstAlias<"movz $dst, $imm", (MOVZXi GPR64:$dst, imm0_65535:$imm, 0)>;
477 // Next, we have various ELF relocations with the ":XYZ_g0:sym" syntax.
478 def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movz_symbol_g3:$sym, 48)>;
479 def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movz_symbol_g2:$sym, 32)>;
480 def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movz_symbol_g1:$sym, 16)>;
481 def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movz_symbol_g0:$sym, 0)>;
483 def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movz_symbol_g3:$sym, 48)>;
484 def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movz_symbol_g2:$sym, 32)>;
485 def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movz_symbol_g1:$sym, 16)>;
486 def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movz_symbol_g0:$sym, 0)>;
488 def : InstAlias<"movk $Rd, $sym", (MOVKXi GPR64:$Rd, movk_symbol_g3:$sym, 48)>;
489 def : InstAlias<"movk $Rd, $sym", (MOVKXi GPR64:$Rd, movk_symbol_g2:$sym, 32)>;
490 def : InstAlias<"movk $Rd, $sym", (MOVKXi GPR64:$Rd, movk_symbol_g1:$sym, 16)>;
491 def : InstAlias<"movk $Rd, $sym", (MOVKXi GPR64:$Rd, movk_symbol_g0:$sym, 0)>;
493 def : InstAlias<"movz $Rd, $sym", (MOVZWi GPR32:$Rd, movz_symbol_g1:$sym, 16)>;
494 def : InstAlias<"movz $Rd, $sym", (MOVZWi GPR32:$Rd, movz_symbol_g0:$sym, 0)>;
496 def : InstAlias<"movn $Rd, $sym", (MOVNWi GPR32:$Rd, movz_symbol_g1:$sym, 16)>;
497 def : InstAlias<"movn $Rd, $sym", (MOVNWi GPR32:$Rd, movz_symbol_g0:$sym, 0)>;
499 def : InstAlias<"movk $Rd, $sym", (MOVKWi GPR32:$Rd, movk_symbol_g1:$sym, 16)>;
500 def : InstAlias<"movk $Rd, $sym", (MOVKWi GPR32:$Rd, movk_symbol_g0:$sym, 0)>;
502 // Final group of aliases covers true "mov $Rd, $imm" cases.
503 multiclass movw_mov_alias<string basename,Instruction INST, RegisterClass GPR,
504 int width, int shift> {
505 def _asmoperand : AsmOperandClass {
506 let Name = basename # width # "_lsl" # shift # "MovAlias";
507 let PredicateMethod = "is" # basename # "MovAlias<" # width # ", "
509 let RenderMethod = "add" # basename # "MovAliasOperands<" # shift # ">";
512 def _movimm : Operand<i32> {
513 let ParserMatchClass = !cast<AsmOperandClass>(NAME # "_asmoperand");
516 def : InstAlias<"mov $Rd, $imm",
517 (INST GPR:$Rd, !cast<Operand>(NAME # "_movimm"):$imm, shift)>;
520 defm : movw_mov_alias<"MOVZ", MOVZWi, GPR32, 32, 0>;
521 defm : movw_mov_alias<"MOVZ", MOVZWi, GPR32, 32, 16>;
523 defm : movw_mov_alias<"MOVZ", MOVZXi, GPR64, 64, 0>;
524 defm : movw_mov_alias<"MOVZ", MOVZXi, GPR64, 64, 16>;
525 defm : movw_mov_alias<"MOVZ", MOVZXi, GPR64, 64, 32>;
526 defm : movw_mov_alias<"MOVZ", MOVZXi, GPR64, 64, 48>;
528 defm : movw_mov_alias<"MOVN", MOVNWi, GPR32, 32, 0>;
529 defm : movw_mov_alias<"MOVN", MOVNWi, GPR32, 32, 16>;
531 defm : movw_mov_alias<"MOVN", MOVNXi, GPR64, 64, 0>;
532 defm : movw_mov_alias<"MOVN", MOVNXi, GPR64, 64, 16>;
533 defm : movw_mov_alias<"MOVN", MOVNXi, GPR64, 64, 32>;
534 defm : movw_mov_alias<"MOVN", MOVNXi, GPR64, 64, 48>;
536 let isReMaterializable = 1, isCodeGenOnly = 1, isMoveImm = 1,
537 isAsCheapAsAMove = 1 in {
538 // FIXME: The following pseudo instructions are only needed because remat
539 // cannot handle multiple instructions. When that changes, we can select
540 // directly to the real instructions and get rid of these pseudos.
543 : Pseudo<(outs GPR32:$dst), (ins i32imm:$src),
544 [(set GPR32:$dst, imm:$src)]>,
547 : Pseudo<(outs GPR64:$dst), (ins i64imm:$src),
548 [(set GPR64:$dst, imm:$src)]>,
550 } // isReMaterializable, isCodeGenOnly
552 // If possible, we want to use MOVi32imm even for 64-bit moves. This gives the
553 // eventual expansion code fewer bits to worry about getting right. Marshalling
554 // the types is a little tricky though:
555 def i64imm_32bit : ImmLeaf<i64, [{
556 return (Imm & 0xffffffffULL) == static_cast<uint64_t>(Imm);
559 def s64imm_32bit : ImmLeaf<i64, [{
560 int64_t Imm64 = static_cast<int64_t>(Imm);
561 return Imm64 >= std::numeric_limits<int32_t>::min() &&
562 Imm64 <= std::numeric_limits<int32_t>::max();
565 def trunc_imm : SDNodeXForm<imm, [{
566 return CurDAG->getTargetConstant(N->getZExtValue(), SDLoc(N), MVT::i32);
569 def : Pat<(i64 i64imm_32bit:$src),
570 (SUBREG_TO_REG (i64 0), (MOVi32imm (trunc_imm imm:$src)), sub_32)>;
572 // Materialize FP constants via MOVi32imm/MOVi64imm (MachO large code model).
573 def bitcast_fpimm_to_i32 : SDNodeXForm<fpimm, [{
574 return CurDAG->getTargetConstant(
575 N->getValueAPF().bitcastToAPInt().getZExtValue(), SDLoc(N), MVT::i32);
578 def bitcast_fpimm_to_i64 : SDNodeXForm<fpimm, [{
579 return CurDAG->getTargetConstant(
580 N->getValueAPF().bitcastToAPInt().getZExtValue(), SDLoc(N), MVT::i64);
584 def : Pat<(f32 fpimm:$in),
585 (COPY_TO_REGCLASS (MOVi32imm (bitcast_fpimm_to_i32 f32:$in)), FPR32)>;
586 def : Pat<(f64 fpimm:$in),
587 (COPY_TO_REGCLASS (MOVi64imm (bitcast_fpimm_to_i64 f64:$in)), FPR64)>;
590 // Deal with the various forms of (ELF) large addressing with MOVZ/MOVK
592 def : Pat<(AArch64WrapperLarge tglobaladdr:$g3, tglobaladdr:$g2,
593 tglobaladdr:$g1, tglobaladdr:$g0),
594 (MOVKXi (MOVKXi (MOVKXi (MOVZXi tglobaladdr:$g0, 0),
595 tglobaladdr:$g1, 16),
596 tglobaladdr:$g2, 32),
597 tglobaladdr:$g3, 48)>;
599 def : Pat<(AArch64WrapperLarge tblockaddress:$g3, tblockaddress:$g2,
600 tblockaddress:$g1, tblockaddress:$g0),
601 (MOVKXi (MOVKXi (MOVKXi (MOVZXi tblockaddress:$g0, 0),
602 tblockaddress:$g1, 16),
603 tblockaddress:$g2, 32),
604 tblockaddress:$g3, 48)>;
606 def : Pat<(AArch64WrapperLarge tconstpool:$g3, tconstpool:$g2,
607 tconstpool:$g1, tconstpool:$g0),
608 (MOVKXi (MOVKXi (MOVKXi (MOVZXi tconstpool:$g0, 0),
611 tconstpool:$g3, 48)>;
613 def : Pat<(AArch64WrapperLarge tjumptable:$g3, tjumptable:$g2,
614 tjumptable:$g1, tjumptable:$g0),
615 (MOVKXi (MOVKXi (MOVKXi (MOVZXi tjumptable:$g0, 0),
618 tjumptable:$g3, 48)>;
621 //===----------------------------------------------------------------------===//
622 // Arithmetic instructions.
623 //===----------------------------------------------------------------------===//
625 // Add/subtract with carry.
626 defm ADC : AddSubCarry<0, "adc", "adcs", AArch64adc, AArch64adc_flag>;
627 defm SBC : AddSubCarry<1, "sbc", "sbcs", AArch64sbc, AArch64sbc_flag>;
629 def : InstAlias<"ngc $dst, $src", (SBCWr GPR32:$dst, WZR, GPR32:$src)>;
630 def : InstAlias<"ngc $dst, $src", (SBCXr GPR64:$dst, XZR, GPR64:$src)>;
631 def : InstAlias<"ngcs $dst, $src", (SBCSWr GPR32:$dst, WZR, GPR32:$src)>;
632 def : InstAlias<"ngcs $dst, $src", (SBCSXr GPR64:$dst, XZR, GPR64:$src)>;
635 defm ADD : AddSub<0, "add", "sub", add>;
636 defm SUB : AddSub<1, "sub", "add">;
638 def : InstAlias<"mov $dst, $src",
639 (ADDWri GPR32sponly:$dst, GPR32sp:$src, 0, 0)>;
640 def : InstAlias<"mov $dst, $src",
641 (ADDWri GPR32sp:$dst, GPR32sponly:$src, 0, 0)>;
642 def : InstAlias<"mov $dst, $src",
643 (ADDXri GPR64sponly:$dst, GPR64sp:$src, 0, 0)>;
644 def : InstAlias<"mov $dst, $src",
645 (ADDXri GPR64sp:$dst, GPR64sponly:$src, 0, 0)>;
647 defm ADDS : AddSubS<0, "adds", AArch64add_flag, "cmn", "subs", "cmp">;
648 defm SUBS : AddSubS<1, "subs", AArch64sub_flag, "cmp", "adds", "cmn">;
650 // Use SUBS instead of SUB to enable CSE between SUBS and SUB.
651 def : Pat<(sub GPR32sp:$Rn, addsub_shifted_imm32:$imm),
652 (SUBSWri GPR32sp:$Rn, addsub_shifted_imm32:$imm)>;
653 def : Pat<(sub GPR64sp:$Rn, addsub_shifted_imm64:$imm),
654 (SUBSXri GPR64sp:$Rn, addsub_shifted_imm64:$imm)>;
655 def : Pat<(sub GPR32:$Rn, GPR32:$Rm),
656 (SUBSWrr GPR32:$Rn, GPR32:$Rm)>;
657 def : Pat<(sub GPR64:$Rn, GPR64:$Rm),
658 (SUBSXrr GPR64:$Rn, GPR64:$Rm)>;
659 def : Pat<(sub GPR32:$Rn, arith_shifted_reg32:$Rm),
660 (SUBSWrs GPR32:$Rn, arith_shifted_reg32:$Rm)>;
661 def : Pat<(sub GPR64:$Rn, arith_shifted_reg64:$Rm),
662 (SUBSXrs GPR64:$Rn, arith_shifted_reg64:$Rm)>;
663 let AddedComplexity = 1 in {
664 def : Pat<(sub GPR32sp:$R2, arith_extended_reg32<i32>:$R3),
665 (SUBSWrx GPR32sp:$R2, arith_extended_reg32<i32>:$R3)>;
666 def : Pat<(sub GPR64sp:$R2, arith_extended_reg32to64<i64>:$R3),
667 (SUBSXrx GPR64sp:$R2, arith_extended_reg32to64<i64>:$R3)>;
670 // Because of the immediate format for add/sub-imm instructions, the
671 // expression (add x, -1) must be transformed to (SUB{W,X}ri x, 1).
672 // These patterns capture that transformation.
673 let AddedComplexity = 1 in {
674 def : Pat<(add GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
675 (SUBSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
676 def : Pat<(add GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
677 (SUBSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
678 def : Pat<(sub GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
679 (ADDWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
680 def : Pat<(sub GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
681 (ADDXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
684 // Because of the immediate format for add/sub-imm instructions, the
685 // expression (add x, -1) must be transformed to (SUB{W,X}ri x, 1).
686 // These patterns capture that transformation.
687 let AddedComplexity = 1 in {
688 def : Pat<(AArch64add_flag GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
689 (SUBSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
690 def : Pat<(AArch64add_flag GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
691 (SUBSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
692 def : Pat<(AArch64sub_flag GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
693 (ADDSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
694 def : Pat<(AArch64sub_flag GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
695 (ADDSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
698 def : InstAlias<"neg $dst, $src", (SUBWrs GPR32:$dst, WZR, GPR32:$src, 0), 3>;
699 def : InstAlias<"neg $dst, $src", (SUBXrs GPR64:$dst, XZR, GPR64:$src, 0), 3>;
700 def : InstAlias<"neg $dst, $src$shift",
701 (SUBWrs GPR32:$dst, WZR, GPR32:$src, arith_shift32:$shift), 2>;
702 def : InstAlias<"neg $dst, $src$shift",
703 (SUBXrs GPR64:$dst, XZR, GPR64:$src, arith_shift64:$shift), 2>;
705 def : InstAlias<"negs $dst, $src", (SUBSWrs GPR32:$dst, WZR, GPR32:$src, 0), 3>;
706 def : InstAlias<"negs $dst, $src", (SUBSXrs GPR64:$dst, XZR, GPR64:$src, 0), 3>;
707 def : InstAlias<"negs $dst, $src$shift",
708 (SUBSWrs GPR32:$dst, WZR, GPR32:$src, arith_shift32:$shift), 2>;
709 def : InstAlias<"negs $dst, $src$shift",
710 (SUBSXrs GPR64:$dst, XZR, GPR64:$src, arith_shift64:$shift), 2>;
713 // Unsigned/Signed divide
714 defm UDIV : Div<0, "udiv", udiv>;
715 defm SDIV : Div<1, "sdiv", sdiv>;
717 def : Pat<(int_aarch64_udiv GPR32:$Rn, GPR32:$Rm), (UDIVWr $Rn, $Rm)>;
718 def : Pat<(int_aarch64_udiv GPR64:$Rn, GPR64:$Rm), (UDIVXr $Rn, $Rm)>;
719 def : Pat<(int_aarch64_sdiv GPR32:$Rn, GPR32:$Rm), (SDIVWr $Rn, $Rm)>;
720 def : Pat<(int_aarch64_sdiv GPR64:$Rn, GPR64:$Rm), (SDIVXr $Rn, $Rm)>;
723 defm ASRV : Shift<0b10, "asr", sra>;
724 defm LSLV : Shift<0b00, "lsl", shl>;
725 defm LSRV : Shift<0b01, "lsr", srl>;
726 defm RORV : Shift<0b11, "ror", rotr>;
728 def : ShiftAlias<"asrv", ASRVWr, GPR32>;
729 def : ShiftAlias<"asrv", ASRVXr, GPR64>;
730 def : ShiftAlias<"lslv", LSLVWr, GPR32>;
731 def : ShiftAlias<"lslv", LSLVXr, GPR64>;
732 def : ShiftAlias<"lsrv", LSRVWr, GPR32>;
733 def : ShiftAlias<"lsrv", LSRVXr, GPR64>;
734 def : ShiftAlias<"rorv", RORVWr, GPR32>;
735 def : ShiftAlias<"rorv", RORVXr, GPR64>;
738 let AddedComplexity = 7 in {
739 defm MADD : MulAccum<0, "madd", add>;
740 defm MSUB : MulAccum<1, "msub", sub>;
742 def : Pat<(i32 (mul GPR32:$Rn, GPR32:$Rm)),
743 (MADDWrrr GPR32:$Rn, GPR32:$Rm, WZR)>;
744 def : Pat<(i64 (mul GPR64:$Rn, GPR64:$Rm)),
745 (MADDXrrr GPR64:$Rn, GPR64:$Rm, XZR)>;
747 def : Pat<(i32 (ineg (mul GPR32:$Rn, GPR32:$Rm))),
748 (MSUBWrrr GPR32:$Rn, GPR32:$Rm, WZR)>;
749 def : Pat<(i64 (ineg (mul GPR64:$Rn, GPR64:$Rm))),
750 (MSUBXrrr GPR64:$Rn, GPR64:$Rm, XZR)>;
751 def : Pat<(i32 (mul (ineg GPR32:$Rn), GPR32:$Rm)),
752 (MSUBWrrr GPR32:$Rn, GPR32:$Rm, WZR)>;
753 def : Pat<(i64 (mul (ineg GPR64:$Rn), GPR64:$Rm)),
754 (MSUBXrrr GPR64:$Rn, GPR64:$Rm, XZR)>;
755 } // AddedComplexity = 7
757 let AddedComplexity = 5 in {
758 def SMADDLrrr : WideMulAccum<0, 0b001, "smaddl", add, sext>;
759 def SMSUBLrrr : WideMulAccum<1, 0b001, "smsubl", sub, sext>;
760 def UMADDLrrr : WideMulAccum<0, 0b101, "umaddl", add, zext>;
761 def UMSUBLrrr : WideMulAccum<1, 0b101, "umsubl", sub, zext>;
763 def : Pat<(i64 (mul (sext GPR32:$Rn), (sext GPR32:$Rm))),
764 (SMADDLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
765 def : Pat<(i64 (mul (zext GPR32:$Rn), (zext GPR32:$Rm))),
766 (UMADDLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
768 def : Pat<(i64 (ineg (mul (sext GPR32:$Rn), (sext GPR32:$Rm)))),
769 (SMSUBLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
770 def : Pat<(i64 (ineg (mul (zext GPR32:$Rn), (zext GPR32:$Rm)))),
771 (UMSUBLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
773 def : Pat<(i64 (mul (sext GPR32:$Rn), (s64imm_32bit:$C))),
774 (SMADDLrrr GPR32:$Rn, (MOVi32imm (trunc_imm imm:$C)), XZR)>;
775 def : Pat<(i64 (mul (zext GPR32:$Rn), (i64imm_32bit:$C))),
776 (UMADDLrrr GPR32:$Rn, (MOVi32imm (trunc_imm imm:$C)), XZR)>;
777 def : Pat<(i64 (mul (sext_inreg GPR64:$Rn, i32), (s64imm_32bit:$C))),
778 (SMADDLrrr (i32 (EXTRACT_SUBREG GPR64:$Rn, sub_32)),
779 (MOVi32imm (trunc_imm imm:$C)), XZR)>;
781 def : Pat<(i64 (ineg (mul (sext GPR32:$Rn), (s64imm_32bit:$C)))),
782 (SMSUBLrrr GPR32:$Rn, (MOVi32imm (trunc_imm imm:$C)), XZR)>;
783 def : Pat<(i64 (ineg (mul (zext GPR32:$Rn), (i64imm_32bit:$C)))),
784 (UMSUBLrrr GPR32:$Rn, (MOVi32imm (trunc_imm imm:$C)), XZR)>;
785 def : Pat<(i64 (ineg (mul (sext_inreg GPR64:$Rn, i32), (s64imm_32bit:$C)))),
786 (SMSUBLrrr (i32 (EXTRACT_SUBREG GPR64:$Rn, sub_32)),
787 (MOVi32imm (trunc_imm imm:$C)), XZR)>;
789 def : Pat<(i64 (add (mul (sext GPR32:$Rn), (s64imm_32bit:$C)), GPR64:$Ra)),
790 (SMADDLrrr GPR32:$Rn, (MOVi32imm (trunc_imm imm:$C)), GPR64:$Ra)>;
791 def : Pat<(i64 (add (mul (zext GPR32:$Rn), (i64imm_32bit:$C)), GPR64:$Ra)),
792 (UMADDLrrr GPR32:$Rn, (MOVi32imm (trunc_imm imm:$C)), GPR64:$Ra)>;
793 def : Pat<(i64 (add (mul (sext_inreg GPR64:$Rn, i32), (s64imm_32bit:$C)),
795 (SMADDLrrr (i32 (EXTRACT_SUBREG GPR64:$Rn, sub_32)),
796 (MOVi32imm (trunc_imm imm:$C)), GPR64:$Ra)>;
798 def : Pat<(i64 (sub GPR64:$Ra, (mul (sext GPR32:$Rn), (s64imm_32bit:$C)))),
799 (SMSUBLrrr GPR32:$Rn, (MOVi32imm (trunc_imm imm:$C)), GPR64:$Ra)>;
800 def : Pat<(i64 (sub GPR64:$Ra, (mul (zext GPR32:$Rn), (i64imm_32bit:$C)))),
801 (UMSUBLrrr GPR32:$Rn, (MOVi32imm (trunc_imm imm:$C)), GPR64:$Ra)>;
802 def : Pat<(i64 (sub GPR64:$Ra, (mul (sext_inreg GPR64:$Rn, i32),
803 (s64imm_32bit:$C)))),
804 (SMSUBLrrr (i32 (EXTRACT_SUBREG GPR64:$Rn, sub_32)),
805 (MOVi32imm (trunc_imm imm:$C)), GPR64:$Ra)>;
806 } // AddedComplexity = 5
808 def : MulAccumWAlias<"mul", MADDWrrr>;
809 def : MulAccumXAlias<"mul", MADDXrrr>;
810 def : MulAccumWAlias<"mneg", MSUBWrrr>;
811 def : MulAccumXAlias<"mneg", MSUBXrrr>;
812 def : WideMulAccumAlias<"smull", SMADDLrrr>;
813 def : WideMulAccumAlias<"smnegl", SMSUBLrrr>;
814 def : WideMulAccumAlias<"umull", UMADDLrrr>;
815 def : WideMulAccumAlias<"umnegl", UMSUBLrrr>;
818 def SMULHrr : MulHi<0b010, "smulh", mulhs>;
819 def UMULHrr : MulHi<0b110, "umulh", mulhu>;
822 def CRC32Brr : BaseCRC32<0, 0b00, 0, GPR32, int_aarch64_crc32b, "crc32b">;
823 def CRC32Hrr : BaseCRC32<0, 0b01, 0, GPR32, int_aarch64_crc32h, "crc32h">;
824 def CRC32Wrr : BaseCRC32<0, 0b10, 0, GPR32, int_aarch64_crc32w, "crc32w">;
825 def CRC32Xrr : BaseCRC32<1, 0b11, 0, GPR64, int_aarch64_crc32x, "crc32x">;
827 def CRC32CBrr : BaseCRC32<0, 0b00, 1, GPR32, int_aarch64_crc32cb, "crc32cb">;
828 def CRC32CHrr : BaseCRC32<0, 0b01, 1, GPR32, int_aarch64_crc32ch, "crc32ch">;
829 def CRC32CWrr : BaseCRC32<0, 0b10, 1, GPR32, int_aarch64_crc32cw, "crc32cw">;
830 def CRC32CXrr : BaseCRC32<1, 0b11, 1, GPR64, int_aarch64_crc32cx, "crc32cx">;
833 defm CAS : CompareAndSwap<0, 0, "">;
834 defm CASA : CompareAndSwap<1, 0, "a">;
835 defm CASL : CompareAndSwap<0, 1, "l">;
836 defm CASAL : CompareAndSwap<1, 1, "al">;
839 defm CASP : CompareAndSwapPair<0, 0, "">;
840 defm CASPA : CompareAndSwapPair<1, 0, "a">;
841 defm CASPL : CompareAndSwapPair<0, 1, "l">;
842 defm CASPAL : CompareAndSwapPair<1, 1, "al">;
845 defm SWP : Swap<0, 0, "">;
846 defm SWPA : Swap<1, 0, "a">;
847 defm SWPL : Swap<0, 1, "l">;
848 defm SWPAL : Swap<1, 1, "al">;
850 // v8.1 atomic LD<OP>(register). Performs load and then ST<OP>(register)
851 defm LDADD : LDOPregister<0b000, "add", 0, 0, "">;
852 defm LDADDA : LDOPregister<0b000, "add", 1, 0, "a">;
853 defm LDADDL : LDOPregister<0b000, "add", 0, 1, "l">;
854 defm LDADDAL : LDOPregister<0b000, "add", 1, 1, "al">;
856 defm LDCLR : LDOPregister<0b001, "clr", 0, 0, "">;
857 defm LDCLRA : LDOPregister<0b001, "clr", 1, 0, "a">;
858 defm LDCLRL : LDOPregister<0b001, "clr", 0, 1, "l">;
859 defm LDCLRAL : LDOPregister<0b001, "clr", 1, 1, "al">;
861 defm LDEOR : LDOPregister<0b010, "eor", 0, 0, "">;
862 defm LDEORA : LDOPregister<0b010, "eor", 1, 0, "a">;
863 defm LDEORL : LDOPregister<0b010, "eor", 0, 1, "l">;
864 defm LDEORAL : LDOPregister<0b010, "eor", 1, 1, "al">;
866 defm LDSET : LDOPregister<0b011, "set", 0, 0, "">;
867 defm LDSETA : LDOPregister<0b011, "set", 1, 0, "a">;
868 defm LDSETL : LDOPregister<0b011, "set", 0, 1, "l">;
869 defm LDSETAL : LDOPregister<0b011, "set", 1, 1, "al">;
871 defm LDSMAX : LDOPregister<0b100, "smax", 0, 0, "">;
872 defm LDSMAXA : LDOPregister<0b100, "smax", 1, 0, "a">;
873 defm LDSMAXL : LDOPregister<0b100, "smax", 0, 1, "l">;
874 defm LDSMAXAL : LDOPregister<0b100, "smax", 1, 1, "al">;
876 defm LDSMIN : LDOPregister<0b101, "smin", 0, 0, "">;
877 defm LDSMINA : LDOPregister<0b101, "smin", 1, 0, "a">;
878 defm LDSMINL : LDOPregister<0b101, "smin", 0, 1, "l">;
879 defm LDSMINAL : LDOPregister<0b101, "smin", 1, 1, "al">;
881 defm LDUMAX : LDOPregister<0b110, "umax", 0, 0, "">;
882 defm LDUMAXA : LDOPregister<0b110, "umax", 1, 0, "a">;
883 defm LDUMAXL : LDOPregister<0b110, "umax", 0, 1, "l">;
884 defm LDUMAXAL : LDOPregister<0b110, "umax", 1, 1, "al">;
886 defm LDUMIN : LDOPregister<0b111, "umin", 0, 0, "">;
887 defm LDUMINA : LDOPregister<0b111, "umin", 1, 0, "a">;
888 defm LDUMINL : LDOPregister<0b111, "umin", 0, 1, "l">;
889 defm LDUMINAL : LDOPregister<0b111, "umin", 1, 1, "al">;
891 // v8.1 atomic ST<OP>(register) as aliases to "LD<OP>(register) when Rt=xZR"
892 defm : STOPregister<"stadd","LDADD">; // STADDx
893 defm : STOPregister<"stclr","LDCLR">; // STCLRx
894 defm : STOPregister<"steor","LDEOR">; // STEORx
895 defm : STOPregister<"stset","LDSET">; // STSETx
896 defm : STOPregister<"stsmax","LDSMAX">;// STSMAXx
897 defm : STOPregister<"stsmin","LDSMIN">;// STSMINx
898 defm : STOPregister<"stumax","LDUMAX">;// STUMAXx
899 defm : STOPregister<"stumin","LDUMIN">;// STUMINx
901 //===----------------------------------------------------------------------===//
902 // Logical instructions.
903 //===----------------------------------------------------------------------===//
906 defm ANDS : LogicalImmS<0b11, "ands", AArch64and_flag, "bics">;
907 defm AND : LogicalImm<0b00, "and", and, "bic">;
908 defm EOR : LogicalImm<0b10, "eor", xor, "eon">;
909 defm ORR : LogicalImm<0b01, "orr", or, "orn">;
911 // FIXME: these aliases *are* canonical sometimes (when movz can't be
912 // used). Actually, it seems to be working right now, but putting logical_immXX
913 // here is a bit dodgy on the AsmParser side too.
914 def : InstAlias<"mov $dst, $imm", (ORRWri GPR32sp:$dst, WZR,
915 logical_imm32:$imm), 0>;
916 def : InstAlias<"mov $dst, $imm", (ORRXri GPR64sp:$dst, XZR,
917 logical_imm64:$imm), 0>;
921 defm ANDS : LogicalRegS<0b11, 0, "ands", AArch64and_flag>;
922 defm BICS : LogicalRegS<0b11, 1, "bics",
923 BinOpFrag<(AArch64and_flag node:$LHS, (not node:$RHS))>>;
924 defm AND : LogicalReg<0b00, 0, "and", and>;
925 defm BIC : LogicalReg<0b00, 1, "bic",
926 BinOpFrag<(and node:$LHS, (not node:$RHS))>>;
927 defm EON : LogicalReg<0b10, 1, "eon",
928 BinOpFrag<(not (xor node:$LHS, node:$RHS))>>;
929 defm EOR : LogicalReg<0b10, 0, "eor", xor>;
930 defm ORN : LogicalReg<0b01, 1, "orn",
931 BinOpFrag<(or node:$LHS, (not node:$RHS))>>;
932 defm ORR : LogicalReg<0b01, 0, "orr", or>;
934 def : InstAlias<"mov $dst, $src", (ORRWrs GPR32:$dst, WZR, GPR32:$src, 0), 2>;
935 def : InstAlias<"mov $dst, $src", (ORRXrs GPR64:$dst, XZR, GPR64:$src, 0), 2>;
937 def : InstAlias<"mvn $Wd, $Wm", (ORNWrs GPR32:$Wd, WZR, GPR32:$Wm, 0), 3>;
938 def : InstAlias<"mvn $Xd, $Xm", (ORNXrs GPR64:$Xd, XZR, GPR64:$Xm, 0), 3>;
940 def : InstAlias<"mvn $Wd, $Wm$sh",
941 (ORNWrs GPR32:$Wd, WZR, GPR32:$Wm, logical_shift32:$sh), 2>;
942 def : InstAlias<"mvn $Xd, $Xm$sh",
943 (ORNXrs GPR64:$Xd, XZR, GPR64:$Xm, logical_shift64:$sh), 2>;
945 def : InstAlias<"tst $src1, $src2",
946 (ANDSWri WZR, GPR32:$src1, logical_imm32:$src2), 2>;
947 def : InstAlias<"tst $src1, $src2",
948 (ANDSXri XZR, GPR64:$src1, logical_imm64:$src2), 2>;
950 def : InstAlias<"tst $src1, $src2",
951 (ANDSWrs WZR, GPR32:$src1, GPR32:$src2, 0), 3>;
952 def : InstAlias<"tst $src1, $src2",
953 (ANDSXrs XZR, GPR64:$src1, GPR64:$src2, 0), 3>;
955 def : InstAlias<"tst $src1, $src2$sh",
956 (ANDSWrs WZR, GPR32:$src1, GPR32:$src2, logical_shift32:$sh), 2>;
957 def : InstAlias<"tst $src1, $src2$sh",
958 (ANDSXrs XZR, GPR64:$src1, GPR64:$src2, logical_shift64:$sh), 2>;
961 def : Pat<(not GPR32:$Wm), (ORNWrr WZR, GPR32:$Wm)>;
962 def : Pat<(not GPR64:$Xm), (ORNXrr XZR, GPR64:$Xm)>;
965 //===----------------------------------------------------------------------===//
966 // One operand data processing instructions.
967 //===----------------------------------------------------------------------===//
969 defm CLS : OneOperandData<0b101, "cls">;
970 defm CLZ : OneOperandData<0b100, "clz", ctlz>;
971 defm RBIT : OneOperandData<0b000, "rbit", bitreverse>;
973 def REV16Wr : OneWRegData<0b001, "rev16",
974 UnOpFrag<(rotr (bswap node:$LHS), (i64 16))>>;
975 def REV16Xr : OneXRegData<0b001, "rev16", null_frag>;
977 def : Pat<(cttz GPR32:$Rn),
978 (CLZWr (RBITWr GPR32:$Rn))>;
979 def : Pat<(cttz GPR64:$Rn),
980 (CLZXr (RBITXr GPR64:$Rn))>;
981 def : Pat<(ctlz (or (shl (xor (sra GPR32:$Rn, (i64 31)), GPR32:$Rn), (i64 1)),
984 def : Pat<(ctlz (or (shl (xor (sra GPR64:$Rn, (i64 63)), GPR64:$Rn), (i64 1)),
988 // Unlike the other one operand instructions, the instructions with the "rev"
989 // mnemonic do *not* just different in the size bit, but actually use different
990 // opcode bits for the different sizes.
991 def REVWr : OneWRegData<0b010, "rev", bswap>;
992 def REVXr : OneXRegData<0b011, "rev", bswap>;
993 def REV32Xr : OneXRegData<0b010, "rev32",
994 UnOpFrag<(rotr (bswap node:$LHS), (i64 32))>>;
996 def : InstAlias<"rev64 $Rd, $Rn", (REVXr GPR64:$Rd, GPR64:$Rn), 0>;
998 // The bswap commutes with the rotr so we want a pattern for both possible
1000 def : Pat<(bswap (rotr GPR32:$Rn, (i64 16))), (REV16Wr GPR32:$Rn)>;
1001 def : Pat<(bswap (rotr GPR64:$Rn, (i64 32))), (REV32Xr GPR64:$Rn)>;
1003 //===----------------------------------------------------------------------===//
1004 // Bitfield immediate extraction instruction.
1005 //===----------------------------------------------------------------------===//
1006 let hasSideEffects = 0 in
1007 defm EXTR : ExtractImm<"extr">;
1008 def : InstAlias<"ror $dst, $src, $shift",
1009 (EXTRWrri GPR32:$dst, GPR32:$src, GPR32:$src, imm0_31:$shift)>;
1010 def : InstAlias<"ror $dst, $src, $shift",
1011 (EXTRXrri GPR64:$dst, GPR64:$src, GPR64:$src, imm0_63:$shift)>;
1013 def : Pat<(rotr GPR32:$Rn, (i64 imm0_31:$imm)),
1014 (EXTRWrri GPR32:$Rn, GPR32:$Rn, imm0_31:$imm)>;
1015 def : Pat<(rotr GPR64:$Rn, (i64 imm0_63:$imm)),
1016 (EXTRXrri GPR64:$Rn, GPR64:$Rn, imm0_63:$imm)>;
1018 //===----------------------------------------------------------------------===//
1019 // Other bitfield immediate instructions.
1020 //===----------------------------------------------------------------------===//
1021 let hasSideEffects = 0 in {
1022 defm BFM : BitfieldImmWith2RegArgs<0b01, "bfm">;
1023 defm SBFM : BitfieldImm<0b00, "sbfm">;
1024 defm UBFM : BitfieldImm<0b10, "ubfm">;
1027 def i32shift_a : Operand<i64>, SDNodeXForm<imm, [{
1028 uint64_t enc = (32 - N->getZExtValue()) & 0x1f;
1029 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
1032 def i32shift_b : Operand<i64>, SDNodeXForm<imm, [{
1033 uint64_t enc = 31 - N->getZExtValue();
1034 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
1037 // min(7, 31 - shift_amt)
1038 def i32shift_sext_i8 : Operand<i64>, SDNodeXForm<imm, [{
1039 uint64_t enc = 31 - N->getZExtValue();
1040 enc = enc > 7 ? 7 : enc;
1041 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
1044 // min(15, 31 - shift_amt)
1045 def i32shift_sext_i16 : Operand<i64>, SDNodeXForm<imm, [{
1046 uint64_t enc = 31 - N->getZExtValue();
1047 enc = enc > 15 ? 15 : enc;
1048 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
1051 def i64shift_a : Operand<i64>, SDNodeXForm<imm, [{
1052 uint64_t enc = (64 - N->getZExtValue()) & 0x3f;
1053 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
1056 def i64shift_b : Operand<i64>, SDNodeXForm<imm, [{
1057 uint64_t enc = 63 - N->getZExtValue();
1058 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
1061 // min(7, 63 - shift_amt)
1062 def i64shift_sext_i8 : Operand<i64>, SDNodeXForm<imm, [{
1063 uint64_t enc = 63 - N->getZExtValue();
1064 enc = enc > 7 ? 7 : enc;
1065 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
1068 // min(15, 63 - shift_amt)
1069 def i64shift_sext_i16 : Operand<i64>, SDNodeXForm<imm, [{
1070 uint64_t enc = 63 - N->getZExtValue();
1071 enc = enc > 15 ? 15 : enc;
1072 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
1075 // min(31, 63 - shift_amt)
1076 def i64shift_sext_i32 : Operand<i64>, SDNodeXForm<imm, [{
1077 uint64_t enc = 63 - N->getZExtValue();
1078 enc = enc > 31 ? 31 : enc;
1079 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
1082 def : Pat<(shl GPR32:$Rn, (i64 imm0_31:$imm)),
1083 (UBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
1084 (i64 (i32shift_b imm0_31:$imm)))>;
1085 def : Pat<(shl GPR64:$Rn, (i64 imm0_63:$imm)),
1086 (UBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
1087 (i64 (i64shift_b imm0_63:$imm)))>;
1089 let AddedComplexity = 10 in {
1090 def : Pat<(sra GPR32:$Rn, (i64 imm0_31:$imm)),
1091 (SBFMWri GPR32:$Rn, imm0_31:$imm, 31)>;
1092 def : Pat<(sra GPR64:$Rn, (i64 imm0_63:$imm)),
1093 (SBFMXri GPR64:$Rn, imm0_63:$imm, 63)>;
1096 def : InstAlias<"asr $dst, $src, $shift",
1097 (SBFMWri GPR32:$dst, GPR32:$src, imm0_31:$shift, 31)>;
1098 def : InstAlias<"asr $dst, $src, $shift",
1099 (SBFMXri GPR64:$dst, GPR64:$src, imm0_63:$shift, 63)>;
1100 def : InstAlias<"sxtb $dst, $src", (SBFMWri GPR32:$dst, GPR32:$src, 0, 7)>;
1101 def : InstAlias<"sxtb $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 7)>;
1102 def : InstAlias<"sxth $dst, $src", (SBFMWri GPR32:$dst, GPR32:$src, 0, 15)>;
1103 def : InstAlias<"sxth $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 15)>;
1104 def : InstAlias<"sxtw $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 31)>;
1106 def : Pat<(srl GPR32:$Rn, (i64 imm0_31:$imm)),
1107 (UBFMWri GPR32:$Rn, imm0_31:$imm, 31)>;
1108 def : Pat<(srl GPR64:$Rn, (i64 imm0_63:$imm)),
1109 (UBFMXri GPR64:$Rn, imm0_63:$imm, 63)>;
1111 def : InstAlias<"lsr $dst, $src, $shift",
1112 (UBFMWri GPR32:$dst, GPR32:$src, imm0_31:$shift, 31)>;
1113 def : InstAlias<"lsr $dst, $src, $shift",
1114 (UBFMXri GPR64:$dst, GPR64:$src, imm0_63:$shift, 63)>;
1115 def : InstAlias<"uxtb $dst, $src", (UBFMWri GPR32:$dst, GPR32:$src, 0, 7)>;
1116 def : InstAlias<"uxtb $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 7)>;
1117 def : InstAlias<"uxth $dst, $src", (UBFMWri GPR32:$dst, GPR32:$src, 0, 15)>;
1118 def : InstAlias<"uxth $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 15)>;
1119 def : InstAlias<"uxtw $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 31)>;
1121 //===----------------------------------------------------------------------===//
1122 // Conditional comparison instructions.
1123 //===----------------------------------------------------------------------===//
1124 defm CCMN : CondComparison<0, "ccmn", AArch64ccmn>;
1125 defm CCMP : CondComparison<1, "ccmp", AArch64ccmp>;
1127 //===----------------------------------------------------------------------===//
1128 // Conditional select instructions.
1129 //===----------------------------------------------------------------------===//
1130 defm CSEL : CondSelect<0, 0b00, "csel">;
1132 def inc : PatFrag<(ops node:$in), (add node:$in, 1)>;
1133 defm CSINC : CondSelectOp<0, 0b01, "csinc", inc>;
1134 defm CSINV : CondSelectOp<1, 0b00, "csinv", not>;
1135 defm CSNEG : CondSelectOp<1, 0b01, "csneg", ineg>;
1137 def : Pat<(AArch64csinv GPR32:$tval, GPR32:$fval, (i32 imm:$cc), NZCV),
1138 (CSINVWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
1139 def : Pat<(AArch64csinv GPR64:$tval, GPR64:$fval, (i32 imm:$cc), NZCV),
1140 (CSINVXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
1141 def : Pat<(AArch64csneg GPR32:$tval, GPR32:$fval, (i32 imm:$cc), NZCV),
1142 (CSNEGWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
1143 def : Pat<(AArch64csneg GPR64:$tval, GPR64:$fval, (i32 imm:$cc), NZCV),
1144 (CSNEGXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
1145 def : Pat<(AArch64csinc GPR32:$tval, GPR32:$fval, (i32 imm:$cc), NZCV),
1146 (CSINCWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
1147 def : Pat<(AArch64csinc GPR64:$tval, GPR64:$fval, (i32 imm:$cc), NZCV),
1148 (CSINCXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
1150 def : Pat<(AArch64csel (i32 0), (i32 1), (i32 imm:$cc), NZCV),
1151 (CSINCWr WZR, WZR, (i32 imm:$cc))>;
1152 def : Pat<(AArch64csel (i64 0), (i64 1), (i32 imm:$cc), NZCV),
1153 (CSINCXr XZR, XZR, (i32 imm:$cc))>;
1154 def : Pat<(AArch64csel GPR32:$tval, (i32 1), (i32 imm:$cc), NZCV),
1155 (CSINCWr GPR32:$tval, WZR, (i32 imm:$cc))>;
1156 def : Pat<(AArch64csel GPR64:$tval, (i64 1), (i32 imm:$cc), NZCV),
1157 (CSINCXr GPR64:$tval, XZR, (i32 imm:$cc))>;
1158 def : Pat<(AArch64csel (i32 1), GPR32:$fval, (i32 imm:$cc), NZCV),
1159 (CSINCWr GPR32:$fval, WZR, (i32 (inv_cond_XFORM imm:$cc)))>;
1160 def : Pat<(AArch64csel (i64 1), GPR64:$fval, (i32 imm:$cc), NZCV),
1161 (CSINCXr GPR64:$fval, XZR, (i32 (inv_cond_XFORM imm:$cc)))>;
1162 def : Pat<(AArch64csel (i32 0), (i32 -1), (i32 imm:$cc), NZCV),
1163 (CSINVWr WZR, WZR, (i32 imm:$cc))>;
1164 def : Pat<(AArch64csel (i64 0), (i64 -1), (i32 imm:$cc), NZCV),
1165 (CSINVXr XZR, XZR, (i32 imm:$cc))>;
1166 def : Pat<(AArch64csel GPR32:$tval, (i32 -1), (i32 imm:$cc), NZCV),
1167 (CSINVWr GPR32:$tval, WZR, (i32 imm:$cc))>;
1168 def : Pat<(AArch64csel GPR64:$tval, (i64 -1), (i32 imm:$cc), NZCV),
1169 (CSINVXr GPR64:$tval, XZR, (i32 imm:$cc))>;
1170 def : Pat<(AArch64csel (i32 -1), GPR32:$fval, (i32 imm:$cc), NZCV),
1171 (CSINVWr GPR32:$fval, WZR, (i32 (inv_cond_XFORM imm:$cc)))>;
1172 def : Pat<(AArch64csel (i64 -1), GPR64:$fval, (i32 imm:$cc), NZCV),
1173 (CSINVXr GPR64:$fval, XZR, (i32 (inv_cond_XFORM imm:$cc)))>;
1175 // The inverse of the condition code from the alias instruction is what is used
1176 // in the aliased instruction. The parser all ready inverts the condition code
1177 // for these aliases.
1178 def : InstAlias<"cset $dst, $cc",
1179 (CSINCWr GPR32:$dst, WZR, WZR, inv_ccode:$cc)>;
1180 def : InstAlias<"cset $dst, $cc",
1181 (CSINCXr GPR64:$dst, XZR, XZR, inv_ccode:$cc)>;
1183 def : InstAlias<"csetm $dst, $cc",
1184 (CSINVWr GPR32:$dst, WZR, WZR, inv_ccode:$cc)>;
1185 def : InstAlias<"csetm $dst, $cc",
1186 (CSINVXr GPR64:$dst, XZR, XZR, inv_ccode:$cc)>;
1188 def : InstAlias<"cinc $dst, $src, $cc",
1189 (CSINCWr GPR32:$dst, GPR32:$src, GPR32:$src, inv_ccode:$cc)>;
1190 def : InstAlias<"cinc $dst, $src, $cc",
1191 (CSINCXr GPR64:$dst, GPR64:$src, GPR64:$src, inv_ccode:$cc)>;
1193 def : InstAlias<"cinv $dst, $src, $cc",
1194 (CSINVWr GPR32:$dst, GPR32:$src, GPR32:$src, inv_ccode:$cc)>;
1195 def : InstAlias<"cinv $dst, $src, $cc",
1196 (CSINVXr GPR64:$dst, GPR64:$src, GPR64:$src, inv_ccode:$cc)>;
1198 def : InstAlias<"cneg $dst, $src, $cc",
1199 (CSNEGWr GPR32:$dst, GPR32:$src, GPR32:$src, inv_ccode:$cc)>;
1200 def : InstAlias<"cneg $dst, $src, $cc",
1201 (CSNEGXr GPR64:$dst, GPR64:$src, GPR64:$src, inv_ccode:$cc)>;
1203 //===----------------------------------------------------------------------===//
1204 // PC-relative instructions.
1205 //===----------------------------------------------------------------------===//
1206 let isReMaterializable = 1 in {
1207 let hasSideEffects = 0, mayStore = 0, mayLoad = 0 in {
1208 def ADR : ADRI<0, "adr", adrlabel, []>;
1209 } // hasSideEffects = 0
1211 def ADRP : ADRI<1, "adrp", adrplabel,
1212 [(set GPR64:$Xd, (AArch64adrp tglobaladdr:$label))]>;
1213 } // isReMaterializable = 1
1215 // page address of a constant pool entry, block address
1216 def : Pat<(AArch64adrp tconstpool:$cp), (ADRP tconstpool:$cp)>;
1217 def : Pat<(AArch64adrp tblockaddress:$cp), (ADRP tblockaddress:$cp)>;
1219 //===----------------------------------------------------------------------===//
1220 // Unconditional branch (register) instructions.
1221 //===----------------------------------------------------------------------===//
1223 let isReturn = 1, isTerminator = 1, isBarrier = 1 in {
1224 def RET : BranchReg<0b0010, "ret", []>;
1225 def DRPS : SpecialReturn<0b0101, "drps">;
1226 def ERET : SpecialReturn<0b0100, "eret">;
1227 } // isReturn = 1, isTerminator = 1, isBarrier = 1
1229 // Default to the LR register.
1230 def : InstAlias<"ret", (RET LR)>;
1232 let isCall = 1, Defs = [LR], Uses = [SP] in {
1233 def BLR : BranchReg<0b0001, "blr", [(AArch64call GPR64:$Rn)]>;
1236 let isBranch = 1, isTerminator = 1, isBarrier = 1, isIndirectBranch = 1 in {
1237 def BR : BranchReg<0b0000, "br", [(brind GPR64:$Rn)]>;
1238 } // isBranch, isTerminator, isBarrier, isIndirectBranch
1240 // Create a separate pseudo-instruction for codegen to use so that we don't
1241 // flag lr as used in every function. It'll be restored before the RET by the
1242 // epilogue if it's legitimately used.
1243 def RET_ReallyLR : Pseudo<(outs), (ins), [(AArch64retflag)]>,
1244 Sched<[WriteBrReg]> {
1245 let isTerminator = 1;
1250 // This is a directive-like pseudo-instruction. The purpose is to insert an
1251 // R_AARCH64_TLSDESC_CALL relocation at the offset of the following instruction
1252 // (which in the usual case is a BLR).
1253 let hasSideEffects = 1 in
1254 def TLSDESCCALL : Pseudo<(outs), (ins i64imm:$sym), []>, Sched<[]> {
1255 let AsmString = ".tlsdesccall $sym";
1258 // FIXME: maybe the scratch register used shouldn't be fixed to X1?
1259 // FIXME: can "hasSideEffects be dropped?
1260 let isCall = 1, Defs = [LR, X0, X1], hasSideEffects = 1,
1261 isCodeGenOnly = 1 in
1263 : Pseudo<(outs), (ins i64imm:$sym),
1264 [(AArch64tlsdesc_callseq tglobaltlsaddr:$sym)]>,
1265 Sched<[WriteI, WriteLD, WriteI, WriteBrReg]>;
1266 def : Pat<(AArch64tlsdesc_callseq texternalsym:$sym),
1267 (TLSDESC_CALLSEQ texternalsym:$sym)>;
1269 //===----------------------------------------------------------------------===//
1270 // Conditional branch (immediate) instruction.
1271 //===----------------------------------------------------------------------===//
1272 def Bcc : BranchCond;
1274 //===----------------------------------------------------------------------===//
1275 // Compare-and-branch instructions.
1276 //===----------------------------------------------------------------------===//
1277 defm CBZ : CmpBranch<0, "cbz", AArch64cbz>;
1278 defm CBNZ : CmpBranch<1, "cbnz", AArch64cbnz>;
1280 //===----------------------------------------------------------------------===//
1281 // Test-bit-and-branch instructions.
1282 //===----------------------------------------------------------------------===//
1283 defm TBZ : TestBranch<0, "tbz", AArch64tbz>;
1284 defm TBNZ : TestBranch<1, "tbnz", AArch64tbnz>;
1286 //===----------------------------------------------------------------------===//
1287 // Unconditional branch (immediate) instructions.
1288 //===----------------------------------------------------------------------===//
1289 let isBranch = 1, isTerminator = 1, isBarrier = 1 in {
1290 def B : BranchImm<0, "b", [(br bb:$addr)]>;
1291 } // isBranch, isTerminator, isBarrier
1293 let isCall = 1, Defs = [LR], Uses = [SP] in {
1294 def BL : CallImm<1, "bl", [(AArch64call tglobaladdr:$addr)]>;
1296 def : Pat<(AArch64call texternalsym:$func), (BL texternalsym:$func)>;
1298 //===----------------------------------------------------------------------===//
1299 // Exception generation instructions.
1300 //===----------------------------------------------------------------------===//
1301 def BRK : ExceptionGeneration<0b001, 0b00, "brk">;
1302 def DCPS1 : ExceptionGeneration<0b101, 0b01, "dcps1">;
1303 def DCPS2 : ExceptionGeneration<0b101, 0b10, "dcps2">;
1304 def DCPS3 : ExceptionGeneration<0b101, 0b11, "dcps3">;
1305 def HLT : ExceptionGeneration<0b010, 0b00, "hlt">;
1306 def HVC : ExceptionGeneration<0b000, 0b10, "hvc">;
1307 def SMC : ExceptionGeneration<0b000, 0b11, "smc">;
1308 def SVC : ExceptionGeneration<0b000, 0b01, "svc">;
1310 // DCPSn defaults to an immediate operand of zero if unspecified.
1311 def : InstAlias<"dcps1", (DCPS1 0)>;
1312 def : InstAlias<"dcps2", (DCPS2 0)>;
1313 def : InstAlias<"dcps3", (DCPS3 0)>;
1315 //===----------------------------------------------------------------------===//
1316 // Load instructions.
1317 //===----------------------------------------------------------------------===//
1319 // Pair (indexed, offset)
1320 defm LDPW : LoadPairOffset<0b00, 0, GPR32, simm7s4, "ldp">;
1321 defm LDPX : LoadPairOffset<0b10, 0, GPR64, simm7s8, "ldp">;
1322 defm LDPS : LoadPairOffset<0b00, 1, FPR32, simm7s4, "ldp">;
1323 defm LDPD : LoadPairOffset<0b01, 1, FPR64, simm7s8, "ldp">;
1324 defm LDPQ : LoadPairOffset<0b10, 1, FPR128, simm7s16, "ldp">;
1326 defm LDPSW : LoadPairOffset<0b01, 0, GPR64, simm7s4, "ldpsw">;
1328 // Pair (pre-indexed)
1329 def LDPWpre : LoadPairPreIdx<0b00, 0, GPR32, simm7s4, "ldp">;
1330 def LDPXpre : LoadPairPreIdx<0b10, 0, GPR64, simm7s8, "ldp">;
1331 def LDPSpre : LoadPairPreIdx<0b00, 1, FPR32, simm7s4, "ldp">;
1332 def LDPDpre : LoadPairPreIdx<0b01, 1, FPR64, simm7s8, "ldp">;
1333 def LDPQpre : LoadPairPreIdx<0b10, 1, FPR128, simm7s16, "ldp">;
1335 def LDPSWpre : LoadPairPreIdx<0b01, 0, GPR64, simm7s4, "ldpsw">;
1337 // Pair (post-indexed)
1338 def LDPWpost : LoadPairPostIdx<0b00, 0, GPR32, simm7s4, "ldp">;
1339 def LDPXpost : LoadPairPostIdx<0b10, 0, GPR64, simm7s8, "ldp">;
1340 def LDPSpost : LoadPairPostIdx<0b00, 1, FPR32, simm7s4, "ldp">;
1341 def LDPDpost : LoadPairPostIdx<0b01, 1, FPR64, simm7s8, "ldp">;
1342 def LDPQpost : LoadPairPostIdx<0b10, 1, FPR128, simm7s16, "ldp">;
1344 def LDPSWpost : LoadPairPostIdx<0b01, 0, GPR64, simm7s4, "ldpsw">;
1347 // Pair (no allocate)
1348 defm LDNPW : LoadPairNoAlloc<0b00, 0, GPR32, simm7s4, "ldnp">;
1349 defm LDNPX : LoadPairNoAlloc<0b10, 0, GPR64, simm7s8, "ldnp">;
1350 defm LDNPS : LoadPairNoAlloc<0b00, 1, FPR32, simm7s4, "ldnp">;
1351 defm LDNPD : LoadPairNoAlloc<0b01, 1, FPR64, simm7s8, "ldnp">;
1352 defm LDNPQ : LoadPairNoAlloc<0b10, 1, FPR128, simm7s16, "ldnp">;
1355 // (register offset)
1359 defm LDRBB : Load8RO<0b00, 0, 0b01, GPR32, "ldrb", i32, zextloadi8>;
1360 defm LDRHH : Load16RO<0b01, 0, 0b01, GPR32, "ldrh", i32, zextloadi16>;
1361 defm LDRW : Load32RO<0b10, 0, 0b01, GPR32, "ldr", i32, load>;
1362 defm LDRX : Load64RO<0b11, 0, 0b01, GPR64, "ldr", i64, load>;
1365 defm LDRB : Load8RO<0b00, 1, 0b01, FPR8, "ldr", untyped, load>;
1366 defm LDRH : Load16RO<0b01, 1, 0b01, FPR16, "ldr", f16, load>;
1367 defm LDRS : Load32RO<0b10, 1, 0b01, FPR32, "ldr", f32, load>;
1368 defm LDRD : Load64RO<0b11, 1, 0b01, FPR64, "ldr", f64, load>;
1369 defm LDRQ : Load128RO<0b00, 1, 0b11, FPR128, "ldr", f128, load>;
1371 // Load sign-extended half-word
1372 defm LDRSHW : Load16RO<0b01, 0, 0b11, GPR32, "ldrsh", i32, sextloadi16>;
1373 defm LDRSHX : Load16RO<0b01, 0, 0b10, GPR64, "ldrsh", i64, sextloadi16>;
1375 // Load sign-extended byte
1376 defm LDRSBW : Load8RO<0b00, 0, 0b11, GPR32, "ldrsb", i32, sextloadi8>;
1377 defm LDRSBX : Load8RO<0b00, 0, 0b10, GPR64, "ldrsb", i64, sextloadi8>;
1379 // Load sign-extended word
1380 defm LDRSW : Load32RO<0b10, 0, 0b10, GPR64, "ldrsw", i64, sextloadi32>;
1383 defm PRFM : PrefetchRO<0b11, 0, 0b10, "prfm">;
1385 // For regular load, we do not have any alignment requirement.
1386 // Thus, it is safe to directly map the vector loads with interesting
1387 // addressing modes.
1388 // FIXME: We could do the same for bitconvert to floating point vectors.
1389 multiclass ScalToVecROLoadPat<ROAddrMode ro, SDPatternOperator loadop,
1390 ValueType ScalTy, ValueType VecTy,
1391 Instruction LOADW, Instruction LOADX,
1393 def : Pat<(VecTy (scalar_to_vector (ScalTy
1394 (loadop (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$offset))))),
1395 (INSERT_SUBREG (VecTy (IMPLICIT_DEF)),
1396 (LOADW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$offset),
1399 def : Pat<(VecTy (scalar_to_vector (ScalTy
1400 (loadop (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$offset))))),
1401 (INSERT_SUBREG (VecTy (IMPLICIT_DEF)),
1402 (LOADX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$offset),
1406 let AddedComplexity = 10 in {
1407 defm : ScalToVecROLoadPat<ro8, extloadi8, i32, v8i8, LDRBroW, LDRBroX, bsub>;
1408 defm : ScalToVecROLoadPat<ro8, extloadi8, i32, v16i8, LDRBroW, LDRBroX, bsub>;
1410 defm : ScalToVecROLoadPat<ro16, extloadi16, i32, v4i16, LDRHroW, LDRHroX, hsub>;
1411 defm : ScalToVecROLoadPat<ro16, extloadi16, i32, v8i16, LDRHroW, LDRHroX, hsub>;
1413 defm : ScalToVecROLoadPat<ro16, load, i32, v4f16, LDRHroW, LDRHroX, hsub>;
1414 defm : ScalToVecROLoadPat<ro16, load, i32, v8f16, LDRHroW, LDRHroX, hsub>;
1416 defm : ScalToVecROLoadPat<ro32, load, i32, v2i32, LDRSroW, LDRSroX, ssub>;
1417 defm : ScalToVecROLoadPat<ro32, load, i32, v4i32, LDRSroW, LDRSroX, ssub>;
1419 defm : ScalToVecROLoadPat<ro32, load, f32, v2f32, LDRSroW, LDRSroX, ssub>;
1420 defm : ScalToVecROLoadPat<ro32, load, f32, v4f32, LDRSroW, LDRSroX, ssub>;
1422 defm : ScalToVecROLoadPat<ro64, load, i64, v2i64, LDRDroW, LDRDroX, dsub>;
1424 defm : ScalToVecROLoadPat<ro64, load, f64, v2f64, LDRDroW, LDRDroX, dsub>;
1427 def : Pat <(v1i64 (scalar_to_vector (i64
1428 (load (ro_Windexed64 GPR64sp:$Rn, GPR32:$Rm,
1429 ro_Wextend64:$extend))))),
1430 (LDRDroW GPR64sp:$Rn, GPR32:$Rm, ro_Wextend64:$extend)>;
1432 def : Pat <(v1i64 (scalar_to_vector (i64
1433 (load (ro_Xindexed64 GPR64sp:$Rn, GPR64:$Rm,
1434 ro_Xextend64:$extend))))),
1435 (LDRDroX GPR64sp:$Rn, GPR64:$Rm, ro_Xextend64:$extend)>;
1438 // Match all load 64 bits width whose type is compatible with FPR64
1439 multiclass VecROLoadPat<ROAddrMode ro, ValueType VecTy,
1440 Instruction LOADW, Instruction LOADX> {
1442 def : Pat<(VecTy (load (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend))),
1443 (LOADW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
1445 def : Pat<(VecTy (load (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend))),
1446 (LOADX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
1449 let AddedComplexity = 10 in {
1450 let Predicates = [IsLE] in {
1451 // We must do vector loads with LD1 in big-endian.
1452 defm : VecROLoadPat<ro64, v2i32, LDRDroW, LDRDroX>;
1453 defm : VecROLoadPat<ro64, v2f32, LDRDroW, LDRDroX>;
1454 defm : VecROLoadPat<ro64, v8i8, LDRDroW, LDRDroX>;
1455 defm : VecROLoadPat<ro64, v4i16, LDRDroW, LDRDroX>;
1456 defm : VecROLoadPat<ro64, v4f16, LDRDroW, LDRDroX>;
1459 defm : VecROLoadPat<ro64, v1i64, LDRDroW, LDRDroX>;
1460 defm : VecROLoadPat<ro64, v1f64, LDRDroW, LDRDroX>;
1462 // Match all load 128 bits width whose type is compatible with FPR128
1463 let Predicates = [IsLE] in {
1464 // We must do vector loads with LD1 in big-endian.
1465 defm : VecROLoadPat<ro128, v2i64, LDRQroW, LDRQroX>;
1466 defm : VecROLoadPat<ro128, v2f64, LDRQroW, LDRQroX>;
1467 defm : VecROLoadPat<ro128, v4i32, LDRQroW, LDRQroX>;
1468 defm : VecROLoadPat<ro128, v4f32, LDRQroW, LDRQroX>;
1469 defm : VecROLoadPat<ro128, v8i16, LDRQroW, LDRQroX>;
1470 defm : VecROLoadPat<ro128, v8f16, LDRQroW, LDRQroX>;
1471 defm : VecROLoadPat<ro128, v16i8, LDRQroW, LDRQroX>;
1473 } // AddedComplexity = 10
1476 multiclass ExtLoadTo64ROPat<ROAddrMode ro, SDPatternOperator loadop,
1477 Instruction INSTW, Instruction INSTX> {
1478 def : Pat<(i64 (loadop (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend))),
1479 (SUBREG_TO_REG (i64 0),
1480 (INSTW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend),
1483 def : Pat<(i64 (loadop (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend))),
1484 (SUBREG_TO_REG (i64 0),
1485 (INSTX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend),
1489 let AddedComplexity = 10 in {
1490 defm : ExtLoadTo64ROPat<ro8, zextloadi8, LDRBBroW, LDRBBroX>;
1491 defm : ExtLoadTo64ROPat<ro16, zextloadi16, LDRHHroW, LDRHHroX>;
1492 defm : ExtLoadTo64ROPat<ro32, zextloadi32, LDRWroW, LDRWroX>;
1494 // zextloadi1 -> zextloadi8
1495 defm : ExtLoadTo64ROPat<ro8, zextloadi1, LDRBBroW, LDRBBroX>;
1497 // extload -> zextload
1498 defm : ExtLoadTo64ROPat<ro8, extloadi8, LDRBBroW, LDRBBroX>;
1499 defm : ExtLoadTo64ROPat<ro16, extloadi16, LDRHHroW, LDRHHroX>;
1500 defm : ExtLoadTo64ROPat<ro32, extloadi32, LDRWroW, LDRWroX>;
1502 // extloadi1 -> zextloadi8
1503 defm : ExtLoadTo64ROPat<ro8, extloadi1, LDRBBroW, LDRBBroX>;
1508 multiclass ExtLoadTo32ROPat<ROAddrMode ro, SDPatternOperator loadop,
1509 Instruction INSTW, Instruction INSTX> {
1510 def : Pat<(i32 (loadop (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend))),
1511 (INSTW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
1513 def : Pat<(i32 (loadop (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend))),
1514 (INSTX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
1518 let AddedComplexity = 10 in {
1519 // extload -> zextload
1520 defm : ExtLoadTo32ROPat<ro8, extloadi8, LDRBBroW, LDRBBroX>;
1521 defm : ExtLoadTo32ROPat<ro16, extloadi16, LDRHHroW, LDRHHroX>;
1522 defm : ExtLoadTo32ROPat<ro32, extloadi32, LDRWroW, LDRWroX>;
1524 // zextloadi1 -> zextloadi8
1525 defm : ExtLoadTo32ROPat<ro8, zextloadi1, LDRBBroW, LDRBBroX>;
1529 // (unsigned immediate)
1531 defm LDRX : LoadUI<0b11, 0, 0b01, GPR64, uimm12s8, "ldr",
1533 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)))]>;
1534 defm LDRW : LoadUI<0b10, 0, 0b01, GPR32, uimm12s4, "ldr",
1536 (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset)))]>;
1537 defm LDRB : LoadUI<0b00, 1, 0b01, FPR8, uimm12s1, "ldr",
1539 (load (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset)))]>;
1540 defm LDRH : LoadUI<0b01, 1, 0b01, FPR16, uimm12s2, "ldr",
1541 [(set (f16 FPR16:$Rt),
1542 (load (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset)))]>;
1543 defm LDRS : LoadUI<0b10, 1, 0b01, FPR32, uimm12s4, "ldr",
1544 [(set (f32 FPR32:$Rt),
1545 (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset)))]>;
1546 defm LDRD : LoadUI<0b11, 1, 0b01, FPR64, uimm12s8, "ldr",
1547 [(set (f64 FPR64:$Rt),
1548 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)))]>;
1549 defm LDRQ : LoadUI<0b00, 1, 0b11, FPR128, uimm12s16, "ldr",
1550 [(set (f128 FPR128:$Rt),
1551 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)))]>;
1553 // For regular load, we do not have any alignment requirement.
1554 // Thus, it is safe to directly map the vector loads with interesting
1555 // addressing modes.
1556 // FIXME: We could do the same for bitconvert to floating point vectors.
1557 def : Pat <(v8i8 (scalar_to_vector (i32
1558 (extloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))))),
1559 (INSERT_SUBREG (v8i8 (IMPLICIT_DEF)),
1560 (LDRBui GPR64sp:$Rn, uimm12s1:$offset), bsub)>;
1561 def : Pat <(v16i8 (scalar_to_vector (i32
1562 (extloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))))),
1563 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
1564 (LDRBui GPR64sp:$Rn, uimm12s1:$offset), bsub)>;
1565 def : Pat <(v4i16 (scalar_to_vector (i32
1566 (extloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))))),
1567 (INSERT_SUBREG (v4i16 (IMPLICIT_DEF)),
1568 (LDRHui GPR64sp:$Rn, uimm12s2:$offset), hsub)>;
1569 def : Pat <(v8i16 (scalar_to_vector (i32
1570 (extloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))))),
1571 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
1572 (LDRHui GPR64sp:$Rn, uimm12s2:$offset), hsub)>;
1573 def : Pat <(v2i32 (scalar_to_vector (i32
1574 (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))))),
1575 (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)),
1576 (LDRSui GPR64sp:$Rn, uimm12s4:$offset), ssub)>;
1577 def : Pat <(v4i32 (scalar_to_vector (i32
1578 (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))))),
1579 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
1580 (LDRSui GPR64sp:$Rn, uimm12s4:$offset), ssub)>;
1581 def : Pat <(v1i64 (scalar_to_vector (i64
1582 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))))),
1583 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1584 def : Pat <(v2i64 (scalar_to_vector (i64
1585 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))))),
1586 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
1587 (LDRDui GPR64sp:$Rn, uimm12s8:$offset), dsub)>;
1589 // Match all load 64 bits width whose type is compatible with FPR64
1590 let Predicates = [IsLE] in {
1591 // We must use LD1 to perform vector loads in big-endian.
1592 def : Pat<(v2f32 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1593 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1594 def : Pat<(v8i8 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1595 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1596 def : Pat<(v4i16 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1597 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1598 def : Pat<(v2i32 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1599 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1600 def : Pat<(v4f16 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1601 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1603 def : Pat<(v1f64 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1604 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1605 def : Pat<(v1i64 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1606 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1608 // Match all load 128 bits width whose type is compatible with FPR128
1609 let Predicates = [IsLE] in {
1610 // We must use LD1 to perform vector loads in big-endian.
1611 def : Pat<(v4f32 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1612 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1613 def : Pat<(v2f64 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1614 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1615 def : Pat<(v16i8 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1616 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1617 def : Pat<(v8i16 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1618 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1619 def : Pat<(v4i32 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1620 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1621 def : Pat<(v2i64 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1622 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1623 def : Pat<(v8f16 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1624 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1626 def : Pat<(f128 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1627 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1629 defm LDRHH : LoadUI<0b01, 0, 0b01, GPR32, uimm12s2, "ldrh",
1631 (zextloadi16 (am_indexed16 GPR64sp:$Rn,
1632 uimm12s2:$offset)))]>;
1633 defm LDRBB : LoadUI<0b00, 0, 0b01, GPR32, uimm12s1, "ldrb",
1635 (zextloadi8 (am_indexed8 GPR64sp:$Rn,
1636 uimm12s1:$offset)))]>;
1638 def : Pat<(i64 (zextloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1639 (SUBREG_TO_REG (i64 0), (LDRBBui GPR64sp:$Rn, uimm12s1:$offset), sub_32)>;
1640 def : Pat<(i64 (zextloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))),
1641 (SUBREG_TO_REG (i64 0), (LDRHHui GPR64sp:$Rn, uimm12s2:$offset), sub_32)>;
1643 // zextloadi1 -> zextloadi8
1644 def : Pat<(i32 (zextloadi1 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1645 (LDRBBui GPR64sp:$Rn, uimm12s1:$offset)>;
1646 def : Pat<(i64 (zextloadi1 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1647 (SUBREG_TO_REG (i64 0), (LDRBBui GPR64sp:$Rn, uimm12s1:$offset), sub_32)>;
1649 // extload -> zextload
1650 def : Pat<(i32 (extloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))),
1651 (LDRHHui GPR64sp:$Rn, uimm12s2:$offset)>;
1652 def : Pat<(i32 (extloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1653 (LDRBBui GPR64sp:$Rn, uimm12s1:$offset)>;
1654 def : Pat<(i32 (extloadi1 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1655 (LDRBBui GPR64sp:$Rn, uimm12s1:$offset)>;
1656 def : Pat<(i64 (extloadi32 (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))),
1657 (SUBREG_TO_REG (i64 0), (LDRWui GPR64sp:$Rn, uimm12s4:$offset), sub_32)>;
1658 def : Pat<(i64 (extloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))),
1659 (SUBREG_TO_REG (i64 0), (LDRHHui GPR64sp:$Rn, uimm12s2:$offset), sub_32)>;
1660 def : Pat<(i64 (extloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1661 (SUBREG_TO_REG (i64 0), (LDRBBui GPR64sp:$Rn, uimm12s1:$offset), sub_32)>;
1662 def : Pat<(i64 (extloadi1 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1663 (SUBREG_TO_REG (i64 0), (LDRBBui GPR64sp:$Rn, uimm12s1:$offset), sub_32)>;
1665 // load sign-extended half-word
1666 defm LDRSHW : LoadUI<0b01, 0, 0b11, GPR32, uimm12s2, "ldrsh",
1668 (sextloadi16 (am_indexed16 GPR64sp:$Rn,
1669 uimm12s2:$offset)))]>;
1670 defm LDRSHX : LoadUI<0b01, 0, 0b10, GPR64, uimm12s2, "ldrsh",
1672 (sextloadi16 (am_indexed16 GPR64sp:$Rn,
1673 uimm12s2:$offset)))]>;
1675 // load sign-extended byte
1676 defm LDRSBW : LoadUI<0b00, 0, 0b11, GPR32, uimm12s1, "ldrsb",
1678 (sextloadi8 (am_indexed8 GPR64sp:$Rn,
1679 uimm12s1:$offset)))]>;
1680 defm LDRSBX : LoadUI<0b00, 0, 0b10, GPR64, uimm12s1, "ldrsb",
1682 (sextloadi8 (am_indexed8 GPR64sp:$Rn,
1683 uimm12s1:$offset)))]>;
1685 // load sign-extended word
1686 defm LDRSW : LoadUI<0b10, 0, 0b10, GPR64, uimm12s4, "ldrsw",
1688 (sextloadi32 (am_indexed32 GPR64sp:$Rn,
1689 uimm12s4:$offset)))]>;
1691 // load zero-extended word
1692 def : Pat<(i64 (zextloadi32 (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))),
1693 (SUBREG_TO_REG (i64 0), (LDRWui GPR64sp:$Rn, uimm12s4:$offset), sub_32)>;
1696 def PRFMui : PrefetchUI<0b11, 0, 0b10, "prfm",
1697 [(AArch64Prefetch imm:$Rt,
1698 (am_indexed64 GPR64sp:$Rn,
1699 uimm12s8:$offset))]>;
1701 def : InstAlias<"prfm $Rt, [$Rn]", (PRFMui prfop:$Rt, GPR64sp:$Rn, 0)>;
1705 def LDRWl : LoadLiteral<0b00, 0, GPR32, "ldr">;
1706 def LDRXl : LoadLiteral<0b01, 0, GPR64, "ldr">;
1707 def LDRSl : LoadLiteral<0b00, 1, FPR32, "ldr">;
1708 def LDRDl : LoadLiteral<0b01, 1, FPR64, "ldr">;
1709 def LDRQl : LoadLiteral<0b10, 1, FPR128, "ldr">;
1711 // load sign-extended word
1712 def LDRSWl : LoadLiteral<0b10, 0, GPR64, "ldrsw">;
1715 def PRFMl : PrefetchLiteral<0b11, 0, "prfm", []>;
1716 // [(AArch64Prefetch imm:$Rt, tglobaladdr:$label)]>;
1719 // (unscaled immediate)
1720 defm LDURX : LoadUnscaled<0b11, 0, 0b01, GPR64, "ldur",
1722 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset)))]>;
1723 defm LDURW : LoadUnscaled<0b10, 0, 0b01, GPR32, "ldur",
1725 (load (am_unscaled32 GPR64sp:$Rn, simm9:$offset)))]>;
1726 defm LDURB : LoadUnscaled<0b00, 1, 0b01, FPR8, "ldur",
1728 (load (am_unscaled8 GPR64sp:$Rn, simm9:$offset)))]>;
1729 defm LDURH : LoadUnscaled<0b01, 1, 0b01, FPR16, "ldur",
1731 (load (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
1732 defm LDURS : LoadUnscaled<0b10, 1, 0b01, FPR32, "ldur",
1733 [(set (f32 FPR32:$Rt),
1734 (load (am_unscaled32 GPR64sp:$Rn, simm9:$offset)))]>;
1735 defm LDURD : LoadUnscaled<0b11, 1, 0b01, FPR64, "ldur",
1736 [(set (f64 FPR64:$Rt),
1737 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset)))]>;
1738 defm LDURQ : LoadUnscaled<0b00, 1, 0b11, FPR128, "ldur",
1739 [(set (f128 FPR128:$Rt),
1740 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset)))]>;
1743 : LoadUnscaled<0b01, 0, 0b01, GPR32, "ldurh",
1745 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
1747 : LoadUnscaled<0b00, 0, 0b01, GPR32, "ldurb",
1749 (zextloadi8 (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
1751 // Match all load 64 bits width whose type is compatible with FPR64
1752 let Predicates = [IsLE] in {
1753 def : Pat<(v2f32 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1754 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1755 def : Pat<(v2i32 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1756 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1757 def : Pat<(v4i16 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1758 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1759 def : Pat<(v8i8 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1760 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1761 def : Pat<(v4f16 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1762 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1764 def : Pat<(v1f64 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1765 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1766 def : Pat<(v1i64 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1767 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1769 // Match all load 128 bits width whose type is compatible with FPR128
1770 let Predicates = [IsLE] in {
1771 def : Pat<(v2f64 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1772 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1773 def : Pat<(v2i64 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1774 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1775 def : Pat<(v4f32 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1776 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1777 def : Pat<(v4i32 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1778 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1779 def : Pat<(v8i16 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1780 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1781 def : Pat<(v16i8 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1782 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1783 def : Pat<(v8f16 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1784 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1788 def : Pat<(i32 (extloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
1789 (LDURHHi GPR64sp:$Rn, simm9:$offset)>;
1790 def : Pat<(i32 (extloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1791 (LDURBBi GPR64sp:$Rn, simm9:$offset)>;
1792 def : Pat<(i32 (extloadi1 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1793 (LDURBBi GPR64sp:$Rn, simm9:$offset)>;
1794 def : Pat<(i64 (extloadi32 (am_unscaled32 GPR64sp:$Rn, simm9:$offset))),
1795 (SUBREG_TO_REG (i64 0), (LDURWi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1796 def : Pat<(i64 (extloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
1797 (SUBREG_TO_REG (i64 0), (LDURHHi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1798 def : Pat<(i64 (extloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1799 (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1800 def : Pat<(i64 (extloadi1 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1801 (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1803 def : Pat<(i32 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
1804 (LDURHHi GPR64sp:$Rn, simm9:$offset)>;
1805 def : Pat<(i32 (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1806 (LDURBBi GPR64sp:$Rn, simm9:$offset)>;
1807 def : Pat<(i32 (zextloadi1 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1808 (LDURBBi GPR64sp:$Rn, simm9:$offset)>;
1809 def : Pat<(i64 (zextloadi32 (am_unscaled32 GPR64sp:$Rn, simm9:$offset))),
1810 (SUBREG_TO_REG (i64 0), (LDURWi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1811 def : Pat<(i64 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
1812 (SUBREG_TO_REG (i64 0), (LDURHHi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1813 def : Pat<(i64 (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1814 (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1815 def : Pat<(i64 (zextloadi1 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1816 (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1820 // LDR mnemonics fall back to LDUR for negative or unaligned offsets.
1822 // Define new assembler match classes as we want to only match these when
1823 // the don't otherwise match the scaled addressing mode for LDR/STR. Don't
1824 // associate a DiagnosticType either, as we want the diagnostic for the
1825 // canonical form (the scaled operand) to take precedence.
1826 class SImm9OffsetOperand<int Width> : AsmOperandClass {
1827 let Name = "SImm9OffsetFB" # Width;
1828 let PredicateMethod = "isSImm9OffsetFB<" # Width # ">";
1829 let RenderMethod = "addImmOperands";
1832 def SImm9OffsetFB8Operand : SImm9OffsetOperand<8>;
1833 def SImm9OffsetFB16Operand : SImm9OffsetOperand<16>;
1834 def SImm9OffsetFB32Operand : SImm9OffsetOperand<32>;
1835 def SImm9OffsetFB64Operand : SImm9OffsetOperand<64>;
1836 def SImm9OffsetFB128Operand : SImm9OffsetOperand<128>;
1838 def simm9_offset_fb8 : Operand<i64> {
1839 let ParserMatchClass = SImm9OffsetFB8Operand;
1841 def simm9_offset_fb16 : Operand<i64> {
1842 let ParserMatchClass = SImm9OffsetFB16Operand;
1844 def simm9_offset_fb32 : Operand<i64> {
1845 let ParserMatchClass = SImm9OffsetFB32Operand;
1847 def simm9_offset_fb64 : Operand<i64> {
1848 let ParserMatchClass = SImm9OffsetFB64Operand;
1850 def simm9_offset_fb128 : Operand<i64> {
1851 let ParserMatchClass = SImm9OffsetFB128Operand;
1854 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1855 (LDURXi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb64:$offset), 0>;
1856 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1857 (LDURWi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
1858 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1859 (LDURBi FPR8:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
1860 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1861 (LDURHi FPR16:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
1862 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1863 (LDURSi FPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
1864 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1865 (LDURDi FPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb64:$offset), 0>;
1866 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1867 (LDURQi FPR128:$Rt, GPR64sp:$Rn, simm9_offset_fb128:$offset), 0>;
1870 def : Pat<(i64 (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1871 (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1872 def : Pat<(i64 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
1873 (SUBREG_TO_REG (i64 0), (LDURHHi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1875 // load sign-extended half-word
1877 : LoadUnscaled<0b01, 0, 0b11, GPR32, "ldursh",
1879 (sextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
1881 : LoadUnscaled<0b01, 0, 0b10, GPR64, "ldursh",
1883 (sextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
1885 // load sign-extended byte
1887 : LoadUnscaled<0b00, 0, 0b11, GPR32, "ldursb",
1889 (sextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset)))]>;
1891 : LoadUnscaled<0b00, 0, 0b10, GPR64, "ldursb",
1893 (sextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset)))]>;
1895 // load sign-extended word
1897 : LoadUnscaled<0b10, 0, 0b10, GPR64, "ldursw",
1899 (sextloadi32 (am_unscaled32 GPR64sp:$Rn, simm9:$offset)))]>;
1901 // zero and sign extending aliases from generic LDR* mnemonics to LDUR*.
1902 def : InstAlias<"ldrb $Rt, [$Rn, $offset]",
1903 (LDURBBi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
1904 def : InstAlias<"ldrh $Rt, [$Rn, $offset]",
1905 (LDURHHi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
1906 def : InstAlias<"ldrsb $Rt, [$Rn, $offset]",
1907 (LDURSBWi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
1908 def : InstAlias<"ldrsb $Rt, [$Rn, $offset]",
1909 (LDURSBXi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
1910 def : InstAlias<"ldrsh $Rt, [$Rn, $offset]",
1911 (LDURSHWi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
1912 def : InstAlias<"ldrsh $Rt, [$Rn, $offset]",
1913 (LDURSHXi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
1914 def : InstAlias<"ldrsw $Rt, [$Rn, $offset]",
1915 (LDURSWi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
1918 defm PRFUM : PrefetchUnscaled<0b11, 0, 0b10, "prfum",
1919 [(AArch64Prefetch imm:$Rt,
1920 (am_unscaled64 GPR64sp:$Rn, simm9:$offset))]>;
1923 // (unscaled immediate, unprivileged)
1924 defm LDTRX : LoadUnprivileged<0b11, 0, 0b01, GPR64, "ldtr">;
1925 defm LDTRW : LoadUnprivileged<0b10, 0, 0b01, GPR32, "ldtr">;
1927 defm LDTRH : LoadUnprivileged<0b01, 0, 0b01, GPR32, "ldtrh">;
1928 defm LDTRB : LoadUnprivileged<0b00, 0, 0b01, GPR32, "ldtrb">;
1930 // load sign-extended half-word
1931 defm LDTRSHW : LoadUnprivileged<0b01, 0, 0b11, GPR32, "ldtrsh">;
1932 defm LDTRSHX : LoadUnprivileged<0b01, 0, 0b10, GPR64, "ldtrsh">;
1934 // load sign-extended byte
1935 defm LDTRSBW : LoadUnprivileged<0b00, 0, 0b11, GPR32, "ldtrsb">;
1936 defm LDTRSBX : LoadUnprivileged<0b00, 0, 0b10, GPR64, "ldtrsb">;
1938 // load sign-extended word
1939 defm LDTRSW : LoadUnprivileged<0b10, 0, 0b10, GPR64, "ldtrsw">;
1942 // (immediate pre-indexed)
1943 def LDRWpre : LoadPreIdx<0b10, 0, 0b01, GPR32, "ldr">;
1944 def LDRXpre : LoadPreIdx<0b11, 0, 0b01, GPR64, "ldr">;
1945 def LDRBpre : LoadPreIdx<0b00, 1, 0b01, FPR8, "ldr">;
1946 def LDRHpre : LoadPreIdx<0b01, 1, 0b01, FPR16, "ldr">;
1947 def LDRSpre : LoadPreIdx<0b10, 1, 0b01, FPR32, "ldr">;
1948 def LDRDpre : LoadPreIdx<0b11, 1, 0b01, FPR64, "ldr">;
1949 def LDRQpre : LoadPreIdx<0b00, 1, 0b11, FPR128, "ldr">;
1951 // load sign-extended half-word
1952 def LDRSHWpre : LoadPreIdx<0b01, 0, 0b11, GPR32, "ldrsh">;
1953 def LDRSHXpre : LoadPreIdx<0b01, 0, 0b10, GPR64, "ldrsh">;
1955 // load sign-extended byte
1956 def LDRSBWpre : LoadPreIdx<0b00, 0, 0b11, GPR32, "ldrsb">;
1957 def LDRSBXpre : LoadPreIdx<0b00, 0, 0b10, GPR64, "ldrsb">;
1959 // load zero-extended byte
1960 def LDRBBpre : LoadPreIdx<0b00, 0, 0b01, GPR32, "ldrb">;
1961 def LDRHHpre : LoadPreIdx<0b01, 0, 0b01, GPR32, "ldrh">;
1963 // load sign-extended word
1964 def LDRSWpre : LoadPreIdx<0b10, 0, 0b10, GPR64, "ldrsw">;
1967 // (immediate post-indexed)
1968 def LDRWpost : LoadPostIdx<0b10, 0, 0b01, GPR32, "ldr">;
1969 def LDRXpost : LoadPostIdx<0b11, 0, 0b01, GPR64, "ldr">;
1970 def LDRBpost : LoadPostIdx<0b00, 1, 0b01, FPR8, "ldr">;
1971 def LDRHpost : LoadPostIdx<0b01, 1, 0b01, FPR16, "ldr">;
1972 def LDRSpost : LoadPostIdx<0b10, 1, 0b01, FPR32, "ldr">;
1973 def LDRDpost : LoadPostIdx<0b11, 1, 0b01, FPR64, "ldr">;
1974 def LDRQpost : LoadPostIdx<0b00, 1, 0b11, FPR128, "ldr">;
1976 // load sign-extended half-word
1977 def LDRSHWpost : LoadPostIdx<0b01, 0, 0b11, GPR32, "ldrsh">;
1978 def LDRSHXpost : LoadPostIdx<0b01, 0, 0b10, GPR64, "ldrsh">;
1980 // load sign-extended byte
1981 def LDRSBWpost : LoadPostIdx<0b00, 0, 0b11, GPR32, "ldrsb">;
1982 def LDRSBXpost : LoadPostIdx<0b00, 0, 0b10, GPR64, "ldrsb">;
1984 // load zero-extended byte
1985 def LDRBBpost : LoadPostIdx<0b00, 0, 0b01, GPR32, "ldrb">;
1986 def LDRHHpost : LoadPostIdx<0b01, 0, 0b01, GPR32, "ldrh">;
1988 // load sign-extended word
1989 def LDRSWpost : LoadPostIdx<0b10, 0, 0b10, GPR64, "ldrsw">;
1991 //===----------------------------------------------------------------------===//
1992 // Store instructions.
1993 //===----------------------------------------------------------------------===//
1995 // Pair (indexed, offset)
1996 // FIXME: Use dedicated range-checked addressing mode operand here.
1997 defm STPW : StorePairOffset<0b00, 0, GPR32, simm7s4, "stp">;
1998 defm STPX : StorePairOffset<0b10, 0, GPR64, simm7s8, "stp">;
1999 defm STPS : StorePairOffset<0b00, 1, FPR32, simm7s4, "stp">;
2000 defm STPD : StorePairOffset<0b01, 1, FPR64, simm7s8, "stp">;
2001 defm STPQ : StorePairOffset<0b10, 1, FPR128, simm7s16, "stp">;
2003 // Pair (pre-indexed)
2004 def STPWpre : StorePairPreIdx<0b00, 0, GPR32, simm7s4, "stp">;
2005 def STPXpre : StorePairPreIdx<0b10, 0, GPR64, simm7s8, "stp">;
2006 def STPSpre : StorePairPreIdx<0b00, 1, FPR32, simm7s4, "stp">;
2007 def STPDpre : StorePairPreIdx<0b01, 1, FPR64, simm7s8, "stp">;
2008 def STPQpre : StorePairPreIdx<0b10, 1, FPR128, simm7s16, "stp">;
2010 // Pair (pre-indexed)
2011 def STPWpost : StorePairPostIdx<0b00, 0, GPR32, simm7s4, "stp">;
2012 def STPXpost : StorePairPostIdx<0b10, 0, GPR64, simm7s8, "stp">;
2013 def STPSpost : StorePairPostIdx<0b00, 1, FPR32, simm7s4, "stp">;
2014 def STPDpost : StorePairPostIdx<0b01, 1, FPR64, simm7s8, "stp">;
2015 def STPQpost : StorePairPostIdx<0b10, 1, FPR128, simm7s16, "stp">;
2017 // Pair (no allocate)
2018 defm STNPW : StorePairNoAlloc<0b00, 0, GPR32, simm7s4, "stnp">;
2019 defm STNPX : StorePairNoAlloc<0b10, 0, GPR64, simm7s8, "stnp">;
2020 defm STNPS : StorePairNoAlloc<0b00, 1, FPR32, simm7s4, "stnp">;
2021 defm STNPD : StorePairNoAlloc<0b01, 1, FPR64, simm7s8, "stnp">;
2022 defm STNPQ : StorePairNoAlloc<0b10, 1, FPR128, simm7s16, "stnp">;
2025 // (Register offset)
2028 defm STRBB : Store8RO< 0b00, 0, 0b00, GPR32, "strb", i32, truncstorei8>;
2029 defm STRHH : Store16RO<0b01, 0, 0b00, GPR32, "strh", i32, truncstorei16>;
2030 defm STRW : Store32RO<0b10, 0, 0b00, GPR32, "str", i32, store>;
2031 defm STRX : Store64RO<0b11, 0, 0b00, GPR64, "str", i64, store>;
2035 defm STRB : Store8RO< 0b00, 1, 0b00, FPR8, "str", untyped, store>;
2036 defm STRH : Store16RO<0b01, 1, 0b00, FPR16, "str", f16, store>;
2037 defm STRS : Store32RO<0b10, 1, 0b00, FPR32, "str", f32, store>;
2038 defm STRD : Store64RO<0b11, 1, 0b00, FPR64, "str", f64, store>;
2039 defm STRQ : Store128RO<0b00, 1, 0b10, FPR128, "str", f128, store>;
2041 multiclass TruncStoreFrom64ROPat<ROAddrMode ro, SDPatternOperator storeop,
2042 Instruction STRW, Instruction STRX> {
2044 def : Pat<(storeop GPR64:$Rt,
2045 (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)),
2046 (STRW (EXTRACT_SUBREG GPR64:$Rt, sub_32),
2047 GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
2049 def : Pat<(storeop GPR64:$Rt,
2050 (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)),
2051 (STRX (EXTRACT_SUBREG GPR64:$Rt, sub_32),
2052 GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
2055 let AddedComplexity = 10 in {
2057 defm : TruncStoreFrom64ROPat<ro8, truncstorei8, STRBBroW, STRBBroX>;
2058 defm : TruncStoreFrom64ROPat<ro16, truncstorei16, STRHHroW, STRHHroX>;
2059 defm : TruncStoreFrom64ROPat<ro32, truncstorei32, STRWroW, STRWroX>;
2062 multiclass VecROStorePat<ROAddrMode ro, ValueType VecTy, RegisterClass FPR,
2063 Instruction STRW, Instruction STRX> {
2064 def : Pat<(store (VecTy FPR:$Rt),
2065 (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)),
2066 (STRW FPR:$Rt, GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
2068 def : Pat<(store (VecTy FPR:$Rt),
2069 (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)),
2070 (STRX FPR:$Rt, GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
2073 let AddedComplexity = 10 in {
2074 // Match all store 64 bits width whose type is compatible with FPR64
2075 let Predicates = [IsLE] in {
2076 // We must use ST1 to store vectors in big-endian.
2077 defm : VecROStorePat<ro64, v2i32, FPR64, STRDroW, STRDroX>;
2078 defm : VecROStorePat<ro64, v2f32, FPR64, STRDroW, STRDroX>;
2079 defm : VecROStorePat<ro64, v4i16, FPR64, STRDroW, STRDroX>;
2080 defm : VecROStorePat<ro64, v8i8, FPR64, STRDroW, STRDroX>;
2081 defm : VecROStorePat<ro64, v4f16, FPR64, STRDroW, STRDroX>;
2084 defm : VecROStorePat<ro64, v1i64, FPR64, STRDroW, STRDroX>;
2085 defm : VecROStorePat<ro64, v1f64, FPR64, STRDroW, STRDroX>;
2087 // Match all store 128 bits width whose type is compatible with FPR128
2088 let Predicates = [IsLE] in {
2089 // We must use ST1 to store vectors in big-endian.
2090 defm : VecROStorePat<ro128, v2i64, FPR128, STRQroW, STRQroX>;
2091 defm : VecROStorePat<ro128, v2f64, FPR128, STRQroW, STRQroX>;
2092 defm : VecROStorePat<ro128, v4i32, FPR128, STRQroW, STRQroX>;
2093 defm : VecROStorePat<ro128, v4f32, FPR128, STRQroW, STRQroX>;
2094 defm : VecROStorePat<ro128, v8i16, FPR128, STRQroW, STRQroX>;
2095 defm : VecROStorePat<ro128, v16i8, FPR128, STRQroW, STRQroX>;
2096 defm : VecROStorePat<ro128, v8f16, FPR128, STRQroW, STRQroX>;
2098 } // AddedComplexity = 10
2100 // Match stores from lane 0 to the appropriate subreg's store.
2101 multiclass VecROStoreLane0Pat<ROAddrMode ro, SDPatternOperator storeop,
2102 ValueType VecTy, ValueType STy,
2103 SubRegIndex SubRegIdx,
2104 Instruction STRW, Instruction STRX> {
2106 def : Pat<(storeop (STy (vector_extract (VecTy VecListOne128:$Vt), 0)),
2107 (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)),
2108 (STRW (EXTRACT_SUBREG VecListOne128:$Vt, SubRegIdx),
2109 GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
2111 def : Pat<(storeop (STy (vector_extract (VecTy VecListOne128:$Vt), 0)),
2112 (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)),
2113 (STRX (EXTRACT_SUBREG VecListOne128:$Vt, SubRegIdx),
2114 GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
2117 let AddedComplexity = 19 in {
2118 defm : VecROStoreLane0Pat<ro16, truncstorei16, v8i16, i32, hsub, STRHroW, STRHroX>;
2119 defm : VecROStoreLane0Pat<ro16, store , v8i16, i16, hsub, STRHroW, STRHroX>;
2120 defm : VecROStoreLane0Pat<ro32, truncstorei32, v4i32, i32, ssub, STRSroW, STRSroX>;
2121 defm : VecROStoreLane0Pat<ro32, store , v4i32, i32, ssub, STRSroW, STRSroX>;
2122 defm : VecROStoreLane0Pat<ro32, store , v4f32, f32, ssub, STRSroW, STRSroX>;
2123 defm : VecROStoreLane0Pat<ro64, store , v2i64, i64, dsub, STRDroW, STRDroX>;
2124 defm : VecROStoreLane0Pat<ro64, store , v2f64, f64, dsub, STRDroW, STRDroX>;
2128 // (unsigned immediate)
2129 defm STRX : StoreUI<0b11, 0, 0b00, GPR64, uimm12s8, "str",
2131 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))]>;
2132 defm STRW : StoreUI<0b10, 0, 0b00, GPR32, uimm12s4, "str",
2134 (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))]>;
2135 defm STRB : StoreUI<0b00, 1, 0b00, FPR8, uimm12s1, "str",
2137 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))]>;
2138 defm STRH : StoreUI<0b01, 1, 0b00, FPR16, uimm12s2, "str",
2139 [(store (f16 FPR16:$Rt),
2140 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))]>;
2141 defm STRS : StoreUI<0b10, 1, 0b00, FPR32, uimm12s4, "str",
2142 [(store (f32 FPR32:$Rt),
2143 (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))]>;
2144 defm STRD : StoreUI<0b11, 1, 0b00, FPR64, uimm12s8, "str",
2145 [(store (f64 FPR64:$Rt),
2146 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))]>;
2147 defm STRQ : StoreUI<0b00, 1, 0b10, FPR128, uimm12s16, "str", []>;
2149 defm STRHH : StoreUI<0b01, 0, 0b00, GPR32, uimm12s2, "strh",
2150 [(truncstorei16 GPR32:$Rt,
2151 (am_indexed16 GPR64sp:$Rn,
2152 uimm12s2:$offset))]>;
2153 defm STRBB : StoreUI<0b00, 0, 0b00, GPR32, uimm12s1, "strb",
2154 [(truncstorei8 GPR32:$Rt,
2155 (am_indexed8 GPR64sp:$Rn,
2156 uimm12s1:$offset))]>;
2158 // Match all store 64 bits width whose type is compatible with FPR64
2159 let AddedComplexity = 10 in {
2160 let Predicates = [IsLE] in {
2161 // We must use ST1 to store vectors in big-endian.
2162 def : Pat<(store (v2f32 FPR64:$Rt),
2163 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
2164 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
2165 def : Pat<(store (v8i8 FPR64:$Rt),
2166 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
2167 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
2168 def : Pat<(store (v4i16 FPR64:$Rt),
2169 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
2170 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
2171 def : Pat<(store (v2i32 FPR64:$Rt),
2172 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
2173 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
2174 def : Pat<(store (v4f16 FPR64:$Rt),
2175 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
2176 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
2178 def : Pat<(store (v1f64 FPR64:$Rt),
2179 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
2180 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
2181 def : Pat<(store (v1i64 FPR64:$Rt),
2182 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
2183 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
2185 // Match all store 128 bits width whose type is compatible with FPR128
2186 let Predicates = [IsLE] in {
2187 // We must use ST1 to store vectors in big-endian.
2188 def : Pat<(store (v4f32 FPR128:$Rt),
2189 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
2190 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
2191 def : Pat<(store (v2f64 FPR128:$Rt),
2192 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
2193 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
2194 def : Pat<(store (v16i8 FPR128:$Rt),
2195 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
2196 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
2197 def : Pat<(store (v8i16 FPR128:$Rt),
2198 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
2199 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
2200 def : Pat<(store (v4i32 FPR128:$Rt),
2201 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
2202 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
2203 def : Pat<(store (v2i64 FPR128:$Rt),
2204 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
2205 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
2206 def : Pat<(store (v8f16 FPR128:$Rt),
2207 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
2208 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
2210 def : Pat<(store (f128 FPR128:$Rt),
2211 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
2212 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
2215 def : Pat<(truncstorei32 GPR64:$Rt,
2216 (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset)),
2217 (STRWui (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, uimm12s4:$offset)>;
2218 def : Pat<(truncstorei16 GPR64:$Rt,
2219 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset)),
2220 (STRHHui (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, uimm12s2:$offset)>;
2221 def : Pat<(truncstorei8 GPR64:$Rt, (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset)),
2222 (STRBBui (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, uimm12s1:$offset)>;
2224 } // AddedComplexity = 10
2227 // (unscaled immediate)
2228 defm STURX : StoreUnscaled<0b11, 0, 0b00, GPR64, "stur",
2230 (am_unscaled64 GPR64sp:$Rn, simm9:$offset))]>;
2231 defm STURW : StoreUnscaled<0b10, 0, 0b00, GPR32, "stur",
2233 (am_unscaled32 GPR64sp:$Rn, simm9:$offset))]>;
2234 defm STURB : StoreUnscaled<0b00, 1, 0b00, FPR8, "stur",
2236 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))]>;
2237 defm STURH : StoreUnscaled<0b01, 1, 0b00, FPR16, "stur",
2238 [(store (f16 FPR16:$Rt),
2239 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))]>;
2240 defm STURS : StoreUnscaled<0b10, 1, 0b00, FPR32, "stur",
2241 [(store (f32 FPR32:$Rt),
2242 (am_unscaled32 GPR64sp:$Rn, simm9:$offset))]>;
2243 defm STURD : StoreUnscaled<0b11, 1, 0b00, FPR64, "stur",
2244 [(store (f64 FPR64:$Rt),
2245 (am_unscaled64 GPR64sp:$Rn, simm9:$offset))]>;
2246 defm STURQ : StoreUnscaled<0b00, 1, 0b10, FPR128, "stur",
2247 [(store (f128 FPR128:$Rt),
2248 (am_unscaled128 GPR64sp:$Rn, simm9:$offset))]>;
2249 defm STURHH : StoreUnscaled<0b01, 0, 0b00, GPR32, "sturh",
2250 [(truncstorei16 GPR32:$Rt,
2251 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))]>;
2252 defm STURBB : StoreUnscaled<0b00, 0, 0b00, GPR32, "sturb",
2253 [(truncstorei8 GPR32:$Rt,
2254 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))]>;
2256 // Match all store 64 bits width whose type is compatible with FPR64
2257 let Predicates = [IsLE] in {
2258 // We must use ST1 to store vectors in big-endian.
2259 def : Pat<(store (v2f32 FPR64:$Rt),
2260 (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2261 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2262 def : Pat<(store (v8i8 FPR64:$Rt),
2263 (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2264 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2265 def : Pat<(store (v4i16 FPR64:$Rt),
2266 (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2267 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2268 def : Pat<(store (v2i32 FPR64:$Rt),
2269 (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2270 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2271 def : Pat<(store (v4f16 FPR64:$Rt),
2272 (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2273 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2275 def : Pat<(store (v1f64 FPR64:$Rt), (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2276 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2277 def : Pat<(store (v1i64 FPR64:$Rt), (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2278 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2280 // Match all store 128 bits width whose type is compatible with FPR128
2281 let Predicates = [IsLE] in {
2282 // We must use ST1 to store vectors in big-endian.
2283 def : Pat<(store (v4f32 FPR128:$Rt),
2284 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2285 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2286 def : Pat<(store (v2f64 FPR128:$Rt),
2287 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2288 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2289 def : Pat<(store (v16i8 FPR128:$Rt),
2290 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2291 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2292 def : Pat<(store (v8i16 FPR128:$Rt),
2293 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2294 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2295 def : Pat<(store (v4i32 FPR128:$Rt),
2296 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2297 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2298 def : Pat<(store (v2i64 FPR128:$Rt),
2299 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2300 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2301 def : Pat<(store (v2f64 FPR128:$Rt),
2302 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2303 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2304 def : Pat<(store (v8f16 FPR128:$Rt),
2305 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2306 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2309 // unscaled i64 truncating stores
2310 def : Pat<(truncstorei32 GPR64:$Rt, (am_unscaled32 GPR64sp:$Rn, simm9:$offset)),
2311 (STURWi (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, simm9:$offset)>;
2312 def : Pat<(truncstorei16 GPR64:$Rt, (am_unscaled16 GPR64sp:$Rn, simm9:$offset)),
2313 (STURHHi (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, simm9:$offset)>;
2314 def : Pat<(truncstorei8 GPR64:$Rt, (am_unscaled8 GPR64sp:$Rn, simm9:$offset)),
2315 (STURBBi (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, simm9:$offset)>;
2318 // STR mnemonics fall back to STUR for negative or unaligned offsets.
2319 def : InstAlias<"str $Rt, [$Rn, $offset]",
2320 (STURXi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb64:$offset), 0>;
2321 def : InstAlias<"str $Rt, [$Rn, $offset]",
2322 (STURWi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
2323 def : InstAlias<"str $Rt, [$Rn, $offset]",
2324 (STURBi FPR8:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
2325 def : InstAlias<"str $Rt, [$Rn, $offset]",
2326 (STURHi FPR16:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
2327 def : InstAlias<"str $Rt, [$Rn, $offset]",
2328 (STURSi FPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
2329 def : InstAlias<"str $Rt, [$Rn, $offset]",
2330 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb64:$offset), 0>;
2331 def : InstAlias<"str $Rt, [$Rn, $offset]",
2332 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9_offset_fb128:$offset), 0>;
2334 def : InstAlias<"strb $Rt, [$Rn, $offset]",
2335 (STURBBi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
2336 def : InstAlias<"strh $Rt, [$Rn, $offset]",
2337 (STURHHi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
2340 // (unscaled immediate, unprivileged)
2341 defm STTRW : StoreUnprivileged<0b10, 0, 0b00, GPR32, "sttr">;
2342 defm STTRX : StoreUnprivileged<0b11, 0, 0b00, GPR64, "sttr">;
2344 defm STTRH : StoreUnprivileged<0b01, 0, 0b00, GPR32, "sttrh">;
2345 defm STTRB : StoreUnprivileged<0b00, 0, 0b00, GPR32, "sttrb">;
2348 // (immediate pre-indexed)
2349 def STRWpre : StorePreIdx<0b10, 0, 0b00, GPR32, "str", pre_store, i32>;
2350 def STRXpre : StorePreIdx<0b11, 0, 0b00, GPR64, "str", pre_store, i64>;
2351 def STRBpre : StorePreIdx<0b00, 1, 0b00, FPR8, "str", pre_store, untyped>;
2352 def STRHpre : StorePreIdx<0b01, 1, 0b00, FPR16, "str", pre_store, f16>;
2353 def STRSpre : StorePreIdx<0b10, 1, 0b00, FPR32, "str", pre_store, f32>;
2354 def STRDpre : StorePreIdx<0b11, 1, 0b00, FPR64, "str", pre_store, f64>;
2355 def STRQpre : StorePreIdx<0b00, 1, 0b10, FPR128, "str", pre_store, f128>;
2357 def STRBBpre : StorePreIdx<0b00, 0, 0b00, GPR32, "strb", pre_truncsti8, i32>;
2358 def STRHHpre : StorePreIdx<0b01, 0, 0b00, GPR32, "strh", pre_truncsti16, i32>;
2361 def : Pat<(pre_truncsti32 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
2362 (STRWpre (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
2364 def : Pat<(pre_truncsti16 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
2365 (STRHHpre (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
2367 def : Pat<(pre_truncsti8 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
2368 (STRBBpre (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
2371 def : Pat<(pre_store (v8i8 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2372 (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2373 def : Pat<(pre_store (v4i16 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2374 (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2375 def : Pat<(pre_store (v2i32 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2376 (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2377 def : Pat<(pre_store (v2f32 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2378 (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2379 def : Pat<(pre_store (v1i64 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2380 (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2381 def : Pat<(pre_store (v1f64 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2382 (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2383 def : Pat<(pre_store (v4f16 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2384 (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2386 def : Pat<(pre_store (v16i8 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2387 (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2388 def : Pat<(pre_store (v8i16 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2389 (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2390 def : Pat<(pre_store (v4i32 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2391 (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2392 def : Pat<(pre_store (v4f32 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2393 (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2394 def : Pat<(pre_store (v2i64 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2395 (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2396 def : Pat<(pre_store (v2f64 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2397 (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2398 def : Pat<(pre_store (v8f16 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2399 (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2402 // (immediate post-indexed)
2403 def STRWpost : StorePostIdx<0b10, 0, 0b00, GPR32, "str", post_store, i32>;
2404 def STRXpost : StorePostIdx<0b11, 0, 0b00, GPR64, "str", post_store, i64>;
2405 def STRBpost : StorePostIdx<0b00, 1, 0b00, FPR8, "str", post_store, untyped>;
2406 def STRHpost : StorePostIdx<0b01, 1, 0b00, FPR16, "str", post_store, f16>;
2407 def STRSpost : StorePostIdx<0b10, 1, 0b00, FPR32, "str", post_store, f32>;
2408 def STRDpost : StorePostIdx<0b11, 1, 0b00, FPR64, "str", post_store, f64>;
2409 def STRQpost : StorePostIdx<0b00, 1, 0b10, FPR128, "str", post_store, f128>;
2411 def STRBBpost : StorePostIdx<0b00, 0, 0b00, GPR32, "strb", post_truncsti8, i32>;
2412 def STRHHpost : StorePostIdx<0b01, 0, 0b00, GPR32, "strh", post_truncsti16, i32>;
2415 def : Pat<(post_truncsti32 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
2416 (STRWpost (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
2418 def : Pat<(post_truncsti16 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
2419 (STRHHpost (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
2421 def : Pat<(post_truncsti8 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
2422 (STRBBpost (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
2425 def : Pat<(post_store (v8i8 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2426 (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2427 def : Pat<(post_store (v4i16 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2428 (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2429 def : Pat<(post_store (v2i32 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2430 (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2431 def : Pat<(post_store (v2f32 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2432 (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2433 def : Pat<(post_store (v1i64 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2434 (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2435 def : Pat<(post_store (v1f64 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2436 (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2437 def : Pat<(post_store (v4f16 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2438 (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2440 def : Pat<(post_store (v16i8 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2441 (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2442 def : Pat<(post_store (v8i16 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2443 (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2444 def : Pat<(post_store (v4i32 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2445 (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2446 def : Pat<(post_store (v4f32 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2447 (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2448 def : Pat<(post_store (v2i64 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2449 (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2450 def : Pat<(post_store (v2f64 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2451 (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2452 def : Pat<(post_store (v8f16 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2453 (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2455 //===----------------------------------------------------------------------===//
2456 // Load/store exclusive instructions.
2457 //===----------------------------------------------------------------------===//
2459 def LDARW : LoadAcquire <0b10, 1, 1, 0, 1, GPR32, "ldar">;
2460 def LDARX : LoadAcquire <0b11, 1, 1, 0, 1, GPR64, "ldar">;
2461 def LDARB : LoadAcquire <0b00, 1, 1, 0, 1, GPR32, "ldarb">;
2462 def LDARH : LoadAcquire <0b01, 1, 1, 0, 1, GPR32, "ldarh">;
2464 def LDAXRW : LoadExclusive <0b10, 0, 1, 0, 1, GPR32, "ldaxr">;
2465 def LDAXRX : LoadExclusive <0b11, 0, 1, 0, 1, GPR64, "ldaxr">;
2466 def LDAXRB : LoadExclusive <0b00, 0, 1, 0, 1, GPR32, "ldaxrb">;
2467 def LDAXRH : LoadExclusive <0b01, 0, 1, 0, 1, GPR32, "ldaxrh">;
2469 def LDXRW : LoadExclusive <0b10, 0, 1, 0, 0, GPR32, "ldxr">;
2470 def LDXRX : LoadExclusive <0b11, 0, 1, 0, 0, GPR64, "ldxr">;
2471 def LDXRB : LoadExclusive <0b00, 0, 1, 0, 0, GPR32, "ldxrb">;
2472 def LDXRH : LoadExclusive <0b01, 0, 1, 0, 0, GPR32, "ldxrh">;
2474 def STLRW : StoreRelease <0b10, 1, 0, 0, 1, GPR32, "stlr">;
2475 def STLRX : StoreRelease <0b11, 1, 0, 0, 1, GPR64, "stlr">;
2476 def STLRB : StoreRelease <0b00, 1, 0, 0, 1, GPR32, "stlrb">;
2477 def STLRH : StoreRelease <0b01, 1, 0, 0, 1, GPR32, "stlrh">;
2479 def STLXRW : StoreExclusive<0b10, 0, 0, 0, 1, GPR32, "stlxr">;
2480 def STLXRX : StoreExclusive<0b11, 0, 0, 0, 1, GPR64, "stlxr">;
2481 def STLXRB : StoreExclusive<0b00, 0, 0, 0, 1, GPR32, "stlxrb">;
2482 def STLXRH : StoreExclusive<0b01, 0, 0, 0, 1, GPR32, "stlxrh">;
2484 def STXRW : StoreExclusive<0b10, 0, 0, 0, 0, GPR32, "stxr">;
2485 def STXRX : StoreExclusive<0b11, 0, 0, 0, 0, GPR64, "stxr">;
2486 def STXRB : StoreExclusive<0b00, 0, 0, 0, 0, GPR32, "stxrb">;
2487 def STXRH : StoreExclusive<0b01, 0, 0, 0, 0, GPR32, "stxrh">;
2489 def LDAXPW : LoadExclusivePair<0b10, 0, 1, 1, 1, GPR32, "ldaxp">;
2490 def LDAXPX : LoadExclusivePair<0b11, 0, 1, 1, 1, GPR64, "ldaxp">;
2492 def LDXPW : LoadExclusivePair<0b10, 0, 1, 1, 0, GPR32, "ldxp">;
2493 def LDXPX : LoadExclusivePair<0b11, 0, 1, 1, 0, GPR64, "ldxp">;
2495 def STLXPW : StoreExclusivePair<0b10, 0, 0, 1, 1, GPR32, "stlxp">;
2496 def STLXPX : StoreExclusivePair<0b11, 0, 0, 1, 1, GPR64, "stlxp">;
2498 def STXPW : StoreExclusivePair<0b10, 0, 0, 1, 0, GPR32, "stxp">;
2499 def STXPX : StoreExclusivePair<0b11, 0, 0, 1, 0, GPR64, "stxp">;
2501 let Predicates = [HasV8_1a] in {
2502 // v8.1a "Limited Order Region" extension load-acquire instructions
2503 def LDLARW : LoadAcquire <0b10, 1, 1, 0, 0, GPR32, "ldlar">;
2504 def LDLARX : LoadAcquire <0b11, 1, 1, 0, 0, GPR64, "ldlar">;
2505 def LDLARB : LoadAcquire <0b00, 1, 1, 0, 0, GPR32, "ldlarb">;
2506 def LDLARH : LoadAcquire <0b01, 1, 1, 0, 0, GPR32, "ldlarh">;
2508 // v8.1a "Limited Order Region" extension store-release instructions
2509 def STLLRW : StoreRelease <0b10, 1, 0, 0, 0, GPR32, "stllr">;
2510 def STLLRX : StoreRelease <0b11, 1, 0, 0, 0, GPR64, "stllr">;
2511 def STLLRB : StoreRelease <0b00, 1, 0, 0, 0, GPR32, "stllrb">;
2512 def STLLRH : StoreRelease <0b01, 1, 0, 0, 0, GPR32, "stllrh">;
2515 //===----------------------------------------------------------------------===//
2516 // Scaled floating point to integer conversion instructions.
2517 //===----------------------------------------------------------------------===//
2519 defm FCVTAS : FPToIntegerUnscaled<0b00, 0b100, "fcvtas", int_aarch64_neon_fcvtas>;
2520 defm FCVTAU : FPToIntegerUnscaled<0b00, 0b101, "fcvtau", int_aarch64_neon_fcvtau>;
2521 defm FCVTMS : FPToIntegerUnscaled<0b10, 0b000, "fcvtms", int_aarch64_neon_fcvtms>;
2522 defm FCVTMU : FPToIntegerUnscaled<0b10, 0b001, "fcvtmu", int_aarch64_neon_fcvtmu>;
2523 defm FCVTNS : FPToIntegerUnscaled<0b00, 0b000, "fcvtns", int_aarch64_neon_fcvtns>;
2524 defm FCVTNU : FPToIntegerUnscaled<0b00, 0b001, "fcvtnu", int_aarch64_neon_fcvtnu>;
2525 defm FCVTPS : FPToIntegerUnscaled<0b01, 0b000, "fcvtps", int_aarch64_neon_fcvtps>;
2526 defm FCVTPU : FPToIntegerUnscaled<0b01, 0b001, "fcvtpu", int_aarch64_neon_fcvtpu>;
2527 defm FCVTZS : FPToIntegerUnscaled<0b11, 0b000, "fcvtzs", fp_to_sint>;
2528 defm FCVTZU : FPToIntegerUnscaled<0b11, 0b001, "fcvtzu", fp_to_uint>;
2529 defm FCVTZS : FPToIntegerScaled<0b11, 0b000, "fcvtzs", fp_to_sint>;
2530 defm FCVTZU : FPToIntegerScaled<0b11, 0b001, "fcvtzu", fp_to_uint>;
2532 multiclass FPToIntegerIntPats<Intrinsic round, string INST> {
2533 def : Pat<(i32 (round f16:$Rn)), (!cast<Instruction>(INST # UWHr) $Rn)>;
2534 def : Pat<(i64 (round f16:$Rn)), (!cast<Instruction>(INST # UXHr) $Rn)>;
2535 def : Pat<(i32 (round f32:$Rn)), (!cast<Instruction>(INST # UWSr) $Rn)>;
2536 def : Pat<(i64 (round f32:$Rn)), (!cast<Instruction>(INST # UXSr) $Rn)>;
2537 def : Pat<(i32 (round f64:$Rn)), (!cast<Instruction>(INST # UWDr) $Rn)>;
2538 def : Pat<(i64 (round f64:$Rn)), (!cast<Instruction>(INST # UXDr) $Rn)>;
2540 def : Pat<(i32 (round (fmul f16:$Rn, fixedpoint_f16_i32:$scale))),
2541 (!cast<Instruction>(INST # SWHri) $Rn, $scale)>;
2542 def : Pat<(i64 (round (fmul f16:$Rn, fixedpoint_f16_i64:$scale))),
2543 (!cast<Instruction>(INST # SXHri) $Rn, $scale)>;
2544 def : Pat<(i32 (round (fmul f32:$Rn, fixedpoint_f32_i32:$scale))),
2545 (!cast<Instruction>(INST # SWSri) $Rn, $scale)>;
2546 def : Pat<(i64 (round (fmul f32:$Rn, fixedpoint_f32_i64:$scale))),
2547 (!cast<Instruction>(INST # SXSri) $Rn, $scale)>;
2548 def : Pat<(i32 (round (fmul f64:$Rn, fixedpoint_f64_i32:$scale))),
2549 (!cast<Instruction>(INST # SWDri) $Rn, $scale)>;
2550 def : Pat<(i64 (round (fmul f64:$Rn, fixedpoint_f64_i64:$scale))),
2551 (!cast<Instruction>(INST # SXDri) $Rn, $scale)>;
2554 defm : FPToIntegerIntPats<int_aarch64_neon_fcvtzs, "FCVTZS">;
2555 defm : FPToIntegerIntPats<int_aarch64_neon_fcvtzu, "FCVTZU">;
2557 multiclass FPToIntegerPats<SDNode to_int, SDNode round, string INST> {
2558 def : Pat<(i32 (to_int (round f32:$Rn))),
2559 (!cast<Instruction>(INST # UWSr) f32:$Rn)>;
2560 def : Pat<(i64 (to_int (round f32:$Rn))),
2561 (!cast<Instruction>(INST # UXSr) f32:$Rn)>;
2562 def : Pat<(i32 (to_int (round f64:$Rn))),
2563 (!cast<Instruction>(INST # UWDr) f64:$Rn)>;
2564 def : Pat<(i64 (to_int (round f64:$Rn))),
2565 (!cast<Instruction>(INST # UXDr) f64:$Rn)>;
2568 defm : FPToIntegerPats<fp_to_sint, fceil, "FCVTPS">;
2569 defm : FPToIntegerPats<fp_to_uint, fceil, "FCVTPU">;
2570 defm : FPToIntegerPats<fp_to_sint, ffloor, "FCVTMS">;
2571 defm : FPToIntegerPats<fp_to_uint, ffloor, "FCVTMU">;
2572 defm : FPToIntegerPats<fp_to_sint, ftrunc, "FCVTZS">;
2573 defm : FPToIntegerPats<fp_to_uint, ftrunc, "FCVTZU">;
2574 defm : FPToIntegerPats<fp_to_sint, fround, "FCVTAS">;
2575 defm : FPToIntegerPats<fp_to_uint, fround, "FCVTAU">;
2577 //===----------------------------------------------------------------------===//
2578 // Scaled integer to floating point conversion instructions.
2579 //===----------------------------------------------------------------------===//
2581 defm SCVTF : IntegerToFP<0, "scvtf", sint_to_fp>;
2582 defm UCVTF : IntegerToFP<1, "ucvtf", uint_to_fp>;
2584 //===----------------------------------------------------------------------===//
2585 // Unscaled integer to floating point conversion instruction.
2586 //===----------------------------------------------------------------------===//
2588 defm FMOV : UnscaledConversion<"fmov">;
2590 // Add pseudo ops for FMOV 0 so we can mark them as isReMaterializable
2591 let isReMaterializable = 1, isCodeGenOnly = 1, isAsCheapAsAMove = 1 in {
2592 def FMOVS0 : Pseudo<(outs FPR32:$Rd), (ins), [(set f32:$Rd, (fpimm0))]>,
2594 def FMOVD0 : Pseudo<(outs FPR64:$Rd), (ins), [(set f64:$Rd, (fpimm0))]>,
2597 // Similarly add aliases
2598 def : InstAlias<"fmov $Rd, #0.0", (FMOVWHr FPR16:$Rd, WZR), 0>,
2599 Requires<[HasFullFP16]>;
2600 def : InstAlias<"fmov $Rd, #0.0", (FMOVWSr FPR32:$Rd, WZR), 0>;
2601 def : InstAlias<"fmov $Rd, #0.0", (FMOVXDr FPR64:$Rd, XZR), 0>;
2603 //===----------------------------------------------------------------------===//
2604 // Floating point conversion instruction.
2605 //===----------------------------------------------------------------------===//
2607 defm FCVT : FPConversion<"fcvt">;
2609 //===----------------------------------------------------------------------===//
2610 // Floating point single operand instructions.
2611 //===----------------------------------------------------------------------===//
2613 defm FABS : SingleOperandFPData<0b0001, "fabs", fabs>;
2614 defm FMOV : SingleOperandFPData<0b0000, "fmov">;
2615 defm FNEG : SingleOperandFPData<0b0010, "fneg", fneg>;
2616 defm FRINTA : SingleOperandFPData<0b1100, "frinta", fround>;
2617 defm FRINTI : SingleOperandFPData<0b1111, "frinti", fnearbyint>;
2618 defm FRINTM : SingleOperandFPData<0b1010, "frintm", ffloor>;
2619 defm FRINTN : SingleOperandFPData<0b1000, "frintn", int_aarch64_neon_frintn>;
2620 defm FRINTP : SingleOperandFPData<0b1001, "frintp", fceil>;
2622 def : Pat<(v1f64 (int_aarch64_neon_frintn (v1f64 FPR64:$Rn))),
2623 (FRINTNDr FPR64:$Rn)>;
2625 defm FRINTX : SingleOperandFPData<0b1110, "frintx", frint>;
2626 defm FRINTZ : SingleOperandFPData<0b1011, "frintz", ftrunc>;
2628 let SchedRW = [WriteFDiv] in {
2629 defm FSQRT : SingleOperandFPData<0b0011, "fsqrt", fsqrt>;
2632 //===----------------------------------------------------------------------===//
2633 // Floating point two operand instructions.
2634 //===----------------------------------------------------------------------===//
2636 defm FADD : TwoOperandFPData<0b0010, "fadd", fadd>;
2637 let SchedRW = [WriteFDiv] in {
2638 defm FDIV : TwoOperandFPData<0b0001, "fdiv", fdiv>;
2640 defm FMAXNM : TwoOperandFPData<0b0110, "fmaxnm", fmaxnum>;
2641 defm FMAX : TwoOperandFPData<0b0100, "fmax", fmaxnan>;
2642 defm FMINNM : TwoOperandFPData<0b0111, "fminnm", fminnum>;
2643 defm FMIN : TwoOperandFPData<0b0101, "fmin", fminnan>;
2644 let SchedRW = [WriteFMul] in {
2645 defm FMUL : TwoOperandFPData<0b0000, "fmul", fmul>;
2646 defm FNMUL : TwoOperandFPDataNeg<0b1000, "fnmul", fmul>;
2648 defm FSUB : TwoOperandFPData<0b0011, "fsub", fsub>;
2650 def : Pat<(v1f64 (fmaxnan (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2651 (FMAXDrr FPR64:$Rn, FPR64:$Rm)>;
2652 def : Pat<(v1f64 (fminnan (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2653 (FMINDrr FPR64:$Rn, FPR64:$Rm)>;
2654 def : Pat<(v1f64 (fmaxnum (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2655 (FMAXNMDrr FPR64:$Rn, FPR64:$Rm)>;
2656 def : Pat<(v1f64 (fminnum (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2657 (FMINNMDrr FPR64:$Rn, FPR64:$Rm)>;
2659 //===----------------------------------------------------------------------===//
2660 // Floating point three operand instructions.
2661 //===----------------------------------------------------------------------===//
2663 defm FMADD : ThreeOperandFPData<0, 0, "fmadd", fma>;
2664 defm FMSUB : ThreeOperandFPData<0, 1, "fmsub",
2665 TriOpFrag<(fma node:$LHS, (fneg node:$MHS), node:$RHS)> >;
2666 defm FNMADD : ThreeOperandFPData<1, 0, "fnmadd",
2667 TriOpFrag<(fneg (fma node:$LHS, node:$MHS, node:$RHS))> >;
2668 defm FNMSUB : ThreeOperandFPData<1, 1, "fnmsub",
2669 TriOpFrag<(fma node:$LHS, node:$MHS, (fneg node:$RHS))> >;
2671 // The following def pats catch the case where the LHS of an FMA is negated.
2672 // The TriOpFrag above catches the case where the middle operand is negated.
2674 // N.b. FMSUB etc have the accumulator at the *end* of (outs), unlike
2675 // the NEON variant.
2676 def : Pat<(f32 (fma (fneg FPR32:$Rn), FPR32:$Rm, FPR32:$Ra)),
2677 (FMSUBSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
2679 def : Pat<(f64 (fma (fneg FPR64:$Rn), FPR64:$Rm, FPR64:$Ra)),
2680 (FMSUBDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
2682 // We handled -(a + b*c) for FNMADD above, now it's time for "(-a) + (-b)*c" and
2684 def : Pat<(f32 (fma (fneg FPR32:$Rn), FPR32:$Rm, (fneg FPR32:$Ra))),
2685 (FNMADDSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
2687 def : Pat<(f64 (fma (fneg FPR64:$Rn), FPR64:$Rm, (fneg FPR64:$Ra))),
2688 (FNMADDDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
2690 def : Pat<(f32 (fma FPR32:$Rn, (fneg FPR32:$Rm), (fneg FPR32:$Ra))),
2691 (FNMADDSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
2693 def : Pat<(f64 (fma FPR64:$Rn, (fneg FPR64:$Rm), (fneg FPR64:$Ra))),
2694 (FNMADDDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
2696 //===----------------------------------------------------------------------===//
2697 // Floating point comparison instructions.
2698 //===----------------------------------------------------------------------===//
2700 defm FCMPE : FPComparison<1, "fcmpe">;
2701 defm FCMP : FPComparison<0, "fcmp", AArch64fcmp>;
2703 //===----------------------------------------------------------------------===//
2704 // Floating point conditional comparison instructions.
2705 //===----------------------------------------------------------------------===//
2707 defm FCCMPE : FPCondComparison<1, "fccmpe">;
2708 defm FCCMP : FPCondComparison<0, "fccmp", AArch64fccmp>;
2710 //===----------------------------------------------------------------------===//
2711 // Floating point conditional select instruction.
2712 //===----------------------------------------------------------------------===//
2714 defm FCSEL : FPCondSelect<"fcsel">;
2716 // CSEL instructions providing f128 types need to be handled by a
2717 // pseudo-instruction since the eventual code will need to introduce basic
2718 // blocks and control flow.
2719 def F128CSEL : Pseudo<(outs FPR128:$Rd),
2720 (ins FPR128:$Rn, FPR128:$Rm, ccode:$cond),
2721 [(set (f128 FPR128:$Rd),
2722 (AArch64csel FPR128:$Rn, FPR128:$Rm,
2723 (i32 imm:$cond), NZCV))]> {
2725 let usesCustomInserter = 1;
2726 let hasNoSchedulingInfo = 1;
2730 //===----------------------------------------------------------------------===//
2731 // Floating point immediate move.
2732 //===----------------------------------------------------------------------===//
2734 let isReMaterializable = 1 in {
2735 defm FMOV : FPMoveImmediate<"fmov">;
2738 //===----------------------------------------------------------------------===//
2739 // Advanced SIMD two vector instructions.
2740 //===----------------------------------------------------------------------===//
2742 defm UABDL : SIMDLongThreeVectorBHSabdl<1, 0b0111, "uabdl",
2743 int_aarch64_neon_uabd>;
2744 // Match UABDL in log2-shuffle patterns.
2745 def : Pat<(abs (v8i16 (sub (zext (v8i8 V64:$opA)),
2746 (zext (v8i8 V64:$opB))))),
2747 (UABDLv8i8_v8i16 V64:$opA, V64:$opB)>;
2748 def : Pat<(xor (v8i16 (AArch64vashr v8i16:$src, (i32 15))),
2749 (v8i16 (add (sub (zext (v8i8 V64:$opA)),
2750 (zext (v8i8 V64:$opB))),
2751 (AArch64vashr v8i16:$src, (i32 15))))),
2752 (UABDLv8i8_v8i16 V64:$opA, V64:$opB)>;
2753 def : Pat<(abs (v8i16 (sub (zext (extract_high_v16i8 V128:$opA)),
2754 (zext (extract_high_v16i8 V128:$opB))))),
2755 (UABDLv16i8_v8i16 V128:$opA, V128:$opB)>;
2756 def : Pat<(xor (v8i16 (AArch64vashr v8i16:$src, (i32 15))),
2757 (v8i16 (add (sub (zext (extract_high_v16i8 V128:$opA)),
2758 (zext (extract_high_v16i8 V128:$opB))),
2759 (AArch64vashr v8i16:$src, (i32 15))))),
2760 (UABDLv16i8_v8i16 V128:$opA, V128:$opB)>;
2761 def : Pat<(abs (v4i32 (sub (zext (v4i16 V64:$opA)),
2762 (zext (v4i16 V64:$opB))))),
2763 (UABDLv4i16_v4i32 V64:$opA, V64:$opB)>;
2764 def : Pat<(abs (v4i32 (sub (zext (extract_high_v8i16 V128:$opA)),
2765 (zext (extract_high_v8i16 V128:$opB))))),
2766 (UABDLv8i16_v4i32 V128:$opA, V128:$opB)>;
2767 def : Pat<(abs (v2i64 (sub (zext (v2i32 V64:$opA)),
2768 (zext (v2i32 V64:$opB))))),
2769 (UABDLv2i32_v2i64 V64:$opA, V64:$opB)>;
2770 def : Pat<(abs (v2i64 (sub (zext (extract_high_v4i32 V128:$opA)),
2771 (zext (extract_high_v4i32 V128:$opB))))),
2772 (UABDLv4i32_v2i64 V128:$opA, V128:$opB)>;
2774 defm ABS : SIMDTwoVectorBHSD<0, 0b01011, "abs", abs>;
2775 defm CLS : SIMDTwoVectorBHS<0, 0b00100, "cls", int_aarch64_neon_cls>;
2776 defm CLZ : SIMDTwoVectorBHS<1, 0b00100, "clz", ctlz>;
2777 defm CMEQ : SIMDCmpTwoVector<0, 0b01001, "cmeq", AArch64cmeqz>;
2778 defm CMGE : SIMDCmpTwoVector<1, 0b01000, "cmge", AArch64cmgez>;
2779 defm CMGT : SIMDCmpTwoVector<0, 0b01000, "cmgt", AArch64cmgtz>;
2780 defm CMLE : SIMDCmpTwoVector<1, 0b01001, "cmle", AArch64cmlez>;
2781 defm CMLT : SIMDCmpTwoVector<0, 0b01010, "cmlt", AArch64cmltz>;
2782 defm CNT : SIMDTwoVectorB<0, 0b00, 0b00101, "cnt", ctpop>;
2783 defm FABS : SIMDTwoVectorFP<0, 1, 0b01111, "fabs", fabs>;
2785 defm FCMEQ : SIMDFPCmpTwoVector<0, 1, 0b01101, "fcmeq", AArch64fcmeqz>;
2786 defm FCMGE : SIMDFPCmpTwoVector<1, 1, 0b01100, "fcmge", AArch64fcmgez>;
2787 defm FCMGT : SIMDFPCmpTwoVector<0, 1, 0b01100, "fcmgt", AArch64fcmgtz>;
2788 defm FCMLE : SIMDFPCmpTwoVector<1, 1, 0b01101, "fcmle", AArch64fcmlez>;
2789 defm FCMLT : SIMDFPCmpTwoVector<0, 1, 0b01110, "fcmlt", AArch64fcmltz>;
2790 defm FCVTAS : SIMDTwoVectorFPToInt<0,0,0b11100, "fcvtas",int_aarch64_neon_fcvtas>;
2791 defm FCVTAU : SIMDTwoVectorFPToInt<1,0,0b11100, "fcvtau",int_aarch64_neon_fcvtau>;
2792 defm FCVTL : SIMDFPWidenTwoVector<0, 0, 0b10111, "fcvtl">;
2793 def : Pat<(v4f32 (int_aarch64_neon_vcvthf2fp (v4i16 V64:$Rn))),
2794 (FCVTLv4i16 V64:$Rn)>;
2795 def : Pat<(v4f32 (int_aarch64_neon_vcvthf2fp (extract_subvector (v8i16 V128:$Rn),
2797 (FCVTLv8i16 V128:$Rn)>;
2798 def : Pat<(v2f64 (fpextend (v2f32 V64:$Rn))), (FCVTLv2i32 V64:$Rn)>;
2799 def : Pat<(v2f64 (fpextend (v2f32 (extract_subvector (v4f32 V128:$Rn),
2801 (FCVTLv4i32 V128:$Rn)>;
2803 def : Pat<(v4f32 (fpextend (v4f16 V64:$Rn))), (FCVTLv4i16 V64:$Rn)>;
2804 def : Pat<(v4f32 (fpextend (v4f16 (extract_subvector (v8f16 V128:$Rn),
2806 (FCVTLv8i16 V128:$Rn)>;
2808 defm FCVTMS : SIMDTwoVectorFPToInt<0,0,0b11011, "fcvtms",int_aarch64_neon_fcvtms>;
2809 defm FCVTMU : SIMDTwoVectorFPToInt<1,0,0b11011, "fcvtmu",int_aarch64_neon_fcvtmu>;
2810 defm FCVTNS : SIMDTwoVectorFPToInt<0,0,0b11010, "fcvtns",int_aarch64_neon_fcvtns>;
2811 defm FCVTNU : SIMDTwoVectorFPToInt<1,0,0b11010, "fcvtnu",int_aarch64_neon_fcvtnu>;
2812 defm FCVTN : SIMDFPNarrowTwoVector<0, 0, 0b10110, "fcvtn">;
2813 def : Pat<(v4i16 (int_aarch64_neon_vcvtfp2hf (v4f32 V128:$Rn))),
2814 (FCVTNv4i16 V128:$Rn)>;
2815 def : Pat<(concat_vectors V64:$Rd,
2816 (v4i16 (int_aarch64_neon_vcvtfp2hf (v4f32 V128:$Rn)))),
2817 (FCVTNv8i16 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
2818 def : Pat<(v2f32 (fpround (v2f64 V128:$Rn))), (FCVTNv2i32 V128:$Rn)>;
2819 def : Pat<(v4f16 (fpround (v4f32 V128:$Rn))), (FCVTNv4i16 V128:$Rn)>;
2820 def : Pat<(concat_vectors V64:$Rd, (v2f32 (fpround (v2f64 V128:$Rn)))),
2821 (FCVTNv4i32 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
2822 defm FCVTPS : SIMDTwoVectorFPToInt<0,1,0b11010, "fcvtps",int_aarch64_neon_fcvtps>;
2823 defm FCVTPU : SIMDTwoVectorFPToInt<1,1,0b11010, "fcvtpu",int_aarch64_neon_fcvtpu>;
2824 defm FCVTXN : SIMDFPInexactCvtTwoVector<1, 0, 0b10110, "fcvtxn",
2825 int_aarch64_neon_fcvtxn>;
2826 defm FCVTZS : SIMDTwoVectorFPToInt<0, 1, 0b11011, "fcvtzs", fp_to_sint>;
2827 defm FCVTZU : SIMDTwoVectorFPToInt<1, 1, 0b11011, "fcvtzu", fp_to_uint>;
2829 def : Pat<(v4i16 (int_aarch64_neon_fcvtzs v4f16:$Rn)), (FCVTZSv4f16 $Rn)>;
2830 def : Pat<(v8i16 (int_aarch64_neon_fcvtzs v8f16:$Rn)), (FCVTZSv8f16 $Rn)>;
2831 def : Pat<(v2i32 (int_aarch64_neon_fcvtzs v2f32:$Rn)), (FCVTZSv2f32 $Rn)>;
2832 def : Pat<(v4i32 (int_aarch64_neon_fcvtzs v4f32:$Rn)), (FCVTZSv4f32 $Rn)>;
2833 def : Pat<(v2i64 (int_aarch64_neon_fcvtzs v2f64:$Rn)), (FCVTZSv2f64 $Rn)>;
2835 def : Pat<(v4i16 (int_aarch64_neon_fcvtzu v4f16:$Rn)), (FCVTZUv4f16 $Rn)>;
2836 def : Pat<(v8i16 (int_aarch64_neon_fcvtzu v8f16:$Rn)), (FCVTZUv8f16 $Rn)>;
2837 def : Pat<(v2i32 (int_aarch64_neon_fcvtzu v2f32:$Rn)), (FCVTZUv2f32 $Rn)>;
2838 def : Pat<(v4i32 (int_aarch64_neon_fcvtzu v4f32:$Rn)), (FCVTZUv4f32 $Rn)>;
2839 def : Pat<(v2i64 (int_aarch64_neon_fcvtzu v2f64:$Rn)), (FCVTZUv2f64 $Rn)>;
2841 defm FNEG : SIMDTwoVectorFP<1, 1, 0b01111, "fneg", fneg>;
2842 defm FRECPE : SIMDTwoVectorFP<0, 1, 0b11101, "frecpe", int_aarch64_neon_frecpe>;
2843 defm FRINTA : SIMDTwoVectorFP<1, 0, 0b11000, "frinta", fround>;
2844 defm FRINTI : SIMDTwoVectorFP<1, 1, 0b11001, "frinti", fnearbyint>;
2845 defm FRINTM : SIMDTwoVectorFP<0, 0, 0b11001, "frintm", ffloor>;
2846 defm FRINTN : SIMDTwoVectorFP<0, 0, 0b11000, "frintn", int_aarch64_neon_frintn>;
2847 defm FRINTP : SIMDTwoVectorFP<0, 1, 0b11000, "frintp", fceil>;
2848 defm FRINTX : SIMDTwoVectorFP<1, 0, 0b11001, "frintx", frint>;
2849 defm FRINTZ : SIMDTwoVectorFP<0, 1, 0b11001, "frintz", ftrunc>;
2850 defm FRSQRTE: SIMDTwoVectorFP<1, 1, 0b11101, "frsqrte", int_aarch64_neon_frsqrte>;
2851 defm FSQRT : SIMDTwoVectorFP<1, 1, 0b11111, "fsqrt", fsqrt>;
2852 defm NEG : SIMDTwoVectorBHSD<1, 0b01011, "neg",
2853 UnOpFrag<(sub immAllZerosV, node:$LHS)> >;
2854 defm NOT : SIMDTwoVectorB<1, 0b00, 0b00101, "not", vnot>;
2855 // Aliases for MVN -> NOT.
2856 def : InstAlias<"mvn{ $Vd.8b, $Vn.8b|.8b $Vd, $Vn}",
2857 (NOTv8i8 V64:$Vd, V64:$Vn)>;
2858 def : InstAlias<"mvn{ $Vd.16b, $Vn.16b|.16b $Vd, $Vn}",
2859 (NOTv16i8 V128:$Vd, V128:$Vn)>;
2861 def : Pat<(AArch64neg (v8i8 V64:$Rn)), (NEGv8i8 V64:$Rn)>;
2862 def : Pat<(AArch64neg (v16i8 V128:$Rn)), (NEGv16i8 V128:$Rn)>;
2863 def : Pat<(AArch64neg (v4i16 V64:$Rn)), (NEGv4i16 V64:$Rn)>;
2864 def : Pat<(AArch64neg (v8i16 V128:$Rn)), (NEGv8i16 V128:$Rn)>;
2865 def : Pat<(AArch64neg (v2i32 V64:$Rn)), (NEGv2i32 V64:$Rn)>;
2866 def : Pat<(AArch64neg (v4i32 V128:$Rn)), (NEGv4i32 V128:$Rn)>;
2867 def : Pat<(AArch64neg (v2i64 V128:$Rn)), (NEGv2i64 V128:$Rn)>;
2869 def : Pat<(AArch64not (v8i8 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2870 def : Pat<(AArch64not (v16i8 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2871 def : Pat<(AArch64not (v4i16 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2872 def : Pat<(AArch64not (v8i16 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2873 def : Pat<(AArch64not (v2i32 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2874 def : Pat<(AArch64not (v1i64 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2875 def : Pat<(AArch64not (v4i32 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2876 def : Pat<(AArch64not (v2i64 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2878 def : Pat<(vnot (v4i16 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2879 def : Pat<(vnot (v8i16 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2880 def : Pat<(vnot (v2i32 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2881 def : Pat<(vnot (v4i32 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2882 def : Pat<(vnot (v2i64 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2884 defm RBIT : SIMDTwoVectorB<1, 0b01, 0b00101, "rbit", int_aarch64_neon_rbit>;
2885 defm REV16 : SIMDTwoVectorB<0, 0b00, 0b00001, "rev16", AArch64rev16>;
2886 defm REV32 : SIMDTwoVectorBH<1, 0b00000, "rev32", AArch64rev32>;
2887 defm REV64 : SIMDTwoVectorBHS<0, 0b00000, "rev64", AArch64rev64>;
2888 defm SADALP : SIMDLongTwoVectorTied<0, 0b00110, "sadalp",
2889 BinOpFrag<(add node:$LHS, (int_aarch64_neon_saddlp node:$RHS))> >;
2890 defm SADDLP : SIMDLongTwoVector<0, 0b00010, "saddlp", int_aarch64_neon_saddlp>;
2891 defm SCVTF : SIMDTwoVectorIntToFP<0, 0, 0b11101, "scvtf", sint_to_fp>;
2892 defm SHLL : SIMDVectorLShiftLongBySizeBHS;
2893 defm SQABS : SIMDTwoVectorBHSD<0, 0b00111, "sqabs", int_aarch64_neon_sqabs>;
2894 defm SQNEG : SIMDTwoVectorBHSD<1, 0b00111, "sqneg", int_aarch64_neon_sqneg>;
2895 defm SQXTN : SIMDMixedTwoVector<0, 0b10100, "sqxtn", int_aarch64_neon_sqxtn>;
2896 defm SQXTUN : SIMDMixedTwoVector<1, 0b10010, "sqxtun", int_aarch64_neon_sqxtun>;
2897 defm SUQADD : SIMDTwoVectorBHSDTied<0, 0b00011, "suqadd",int_aarch64_neon_suqadd>;
2898 defm UADALP : SIMDLongTwoVectorTied<1, 0b00110, "uadalp",
2899 BinOpFrag<(add node:$LHS, (int_aarch64_neon_uaddlp node:$RHS))> >;
2900 defm UADDLP : SIMDLongTwoVector<1, 0b00010, "uaddlp",
2901 int_aarch64_neon_uaddlp>;
2902 defm UCVTF : SIMDTwoVectorIntToFP<1, 0, 0b11101, "ucvtf", uint_to_fp>;
2903 defm UQXTN : SIMDMixedTwoVector<1, 0b10100, "uqxtn", int_aarch64_neon_uqxtn>;
2904 defm URECPE : SIMDTwoVectorS<0, 1, 0b11100, "urecpe", int_aarch64_neon_urecpe>;
2905 defm URSQRTE: SIMDTwoVectorS<1, 1, 0b11100, "ursqrte", int_aarch64_neon_ursqrte>;
2906 defm USQADD : SIMDTwoVectorBHSDTied<1, 0b00011, "usqadd",int_aarch64_neon_usqadd>;
2907 defm XTN : SIMDMixedTwoVector<0, 0b10010, "xtn", trunc>;
2909 def : Pat<(v4f16 (AArch64rev32 V64:$Rn)), (REV32v4i16 V64:$Rn)>;
2910 def : Pat<(v4f16 (AArch64rev64 V64:$Rn)), (REV64v4i16 V64:$Rn)>;
2911 def : Pat<(v8f16 (AArch64rev32 V128:$Rn)), (REV32v8i16 V128:$Rn)>;
2912 def : Pat<(v8f16 (AArch64rev64 V128:$Rn)), (REV64v8i16 V128:$Rn)>;
2913 def : Pat<(v2f32 (AArch64rev64 V64:$Rn)), (REV64v2i32 V64:$Rn)>;
2914 def : Pat<(v4f32 (AArch64rev64 V128:$Rn)), (REV64v4i32 V128:$Rn)>;
2916 // Patterns for vector long shift (by element width). These need to match all
2917 // three of zext, sext and anyext so it's easier to pull the patterns out of the
2919 multiclass SIMDVectorLShiftLongBySizeBHSPats<SDPatternOperator ext> {
2920 def : Pat<(AArch64vshl (v8i16 (ext (v8i8 V64:$Rn))), (i32 8)),
2921 (SHLLv8i8 V64:$Rn)>;
2922 def : Pat<(AArch64vshl (v8i16 (ext (extract_high_v16i8 V128:$Rn))), (i32 8)),
2923 (SHLLv16i8 V128:$Rn)>;
2924 def : Pat<(AArch64vshl (v4i32 (ext (v4i16 V64:$Rn))), (i32 16)),
2925 (SHLLv4i16 V64:$Rn)>;
2926 def : Pat<(AArch64vshl (v4i32 (ext (extract_high_v8i16 V128:$Rn))), (i32 16)),
2927 (SHLLv8i16 V128:$Rn)>;
2928 def : Pat<(AArch64vshl (v2i64 (ext (v2i32 V64:$Rn))), (i32 32)),
2929 (SHLLv2i32 V64:$Rn)>;
2930 def : Pat<(AArch64vshl (v2i64 (ext (extract_high_v4i32 V128:$Rn))), (i32 32)),
2931 (SHLLv4i32 V128:$Rn)>;
2934 defm : SIMDVectorLShiftLongBySizeBHSPats<anyext>;
2935 defm : SIMDVectorLShiftLongBySizeBHSPats<zext>;
2936 defm : SIMDVectorLShiftLongBySizeBHSPats<sext>;
2938 //===----------------------------------------------------------------------===//
2939 // Advanced SIMD three vector instructions.
2940 //===----------------------------------------------------------------------===//
2942 defm ADD : SIMDThreeSameVector<0, 0b10000, "add", add>;
2943 defm ADDP : SIMDThreeSameVector<0, 0b10111, "addp", int_aarch64_neon_addp>;
2944 defm CMEQ : SIMDThreeSameVector<1, 0b10001, "cmeq", AArch64cmeq>;
2945 defm CMGE : SIMDThreeSameVector<0, 0b00111, "cmge", AArch64cmge>;
2946 defm CMGT : SIMDThreeSameVector<0, 0b00110, "cmgt", AArch64cmgt>;
2947 defm CMHI : SIMDThreeSameVector<1, 0b00110, "cmhi", AArch64cmhi>;
2948 defm CMHS : SIMDThreeSameVector<1, 0b00111, "cmhs", AArch64cmhs>;
2949 defm CMTST : SIMDThreeSameVector<0, 0b10001, "cmtst", AArch64cmtst>;
2950 defm FABD : SIMDThreeSameVectorFP<1,1,0b010,"fabd", int_aarch64_neon_fabd>;
2951 defm FACGE : SIMDThreeSameVectorFPCmp<1,0,0b101,"facge",int_aarch64_neon_facge>;
2952 defm FACGT : SIMDThreeSameVectorFPCmp<1,1,0b101,"facgt",int_aarch64_neon_facgt>;
2953 defm FADDP : SIMDThreeSameVectorFP<1,0,0b010,"faddp",int_aarch64_neon_addp>;
2954 defm FADD : SIMDThreeSameVectorFP<0,0,0b010,"fadd", fadd>;
2955 defm FCMEQ : SIMDThreeSameVectorFPCmp<0, 0, 0b100, "fcmeq", AArch64fcmeq>;
2956 defm FCMGE : SIMDThreeSameVectorFPCmp<1, 0, 0b100, "fcmge", AArch64fcmge>;
2957 defm FCMGT : SIMDThreeSameVectorFPCmp<1, 1, 0b100, "fcmgt", AArch64fcmgt>;
2958 defm FDIV : SIMDThreeSameVectorFP<1,0,0b111,"fdiv", fdiv>;
2959 defm FMAXNMP : SIMDThreeSameVectorFP<1,0,0b000,"fmaxnmp", int_aarch64_neon_fmaxnmp>;
2960 defm FMAXNM : SIMDThreeSameVectorFP<0,0,0b000,"fmaxnm", fmaxnum>;
2961 defm FMAXP : SIMDThreeSameVectorFP<1,0,0b110,"fmaxp", int_aarch64_neon_fmaxp>;
2962 defm FMAX : SIMDThreeSameVectorFP<0,0,0b110,"fmax", fmaxnan>;
2963 defm FMINNMP : SIMDThreeSameVectorFP<1,1,0b000,"fminnmp", int_aarch64_neon_fminnmp>;
2964 defm FMINNM : SIMDThreeSameVectorFP<0,1,0b000,"fminnm", fminnum>;
2965 defm FMINP : SIMDThreeSameVectorFP<1,1,0b110,"fminp", int_aarch64_neon_fminp>;
2966 defm FMIN : SIMDThreeSameVectorFP<0,1,0b110,"fmin", fminnan>;
2968 // NOTE: The operands of the PatFrag are reordered on FMLA/FMLS because the
2969 // instruction expects the addend first, while the fma intrinsic puts it last.
2970 defm FMLA : SIMDThreeSameVectorFPTied<0, 0, 0b001, "fmla",
2971 TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)> >;
2972 defm FMLS : SIMDThreeSameVectorFPTied<0, 1, 0b001, "fmls",
2973 TriOpFrag<(fma node:$MHS, (fneg node:$RHS), node:$LHS)> >;
2975 // The following def pats catch the case where the LHS of an FMA is negated.
2976 // The TriOpFrag above catches the case where the middle operand is negated.
2977 def : Pat<(v2f32 (fma (fneg V64:$Rn), V64:$Rm, V64:$Rd)),
2978 (FMLSv2f32 V64:$Rd, V64:$Rn, V64:$Rm)>;
2980 def : Pat<(v4f32 (fma (fneg V128:$Rn), V128:$Rm, V128:$Rd)),
2981 (FMLSv4f32 V128:$Rd, V128:$Rn, V128:$Rm)>;
2983 def : Pat<(v2f64 (fma (fneg V128:$Rn), V128:$Rm, V128:$Rd)),
2984 (FMLSv2f64 V128:$Rd, V128:$Rn, V128:$Rm)>;
2986 defm FMULX : SIMDThreeSameVectorFP<0,0,0b011,"fmulx", int_aarch64_neon_fmulx>;
2987 defm FMUL : SIMDThreeSameVectorFP<1,0,0b011,"fmul", fmul>;
2988 defm FRECPS : SIMDThreeSameVectorFP<0,0,0b111,"frecps", int_aarch64_neon_frecps>;
2989 defm FRSQRTS : SIMDThreeSameVectorFP<0,1,0b111,"frsqrts", int_aarch64_neon_frsqrts>;
2990 defm FSUB : SIMDThreeSameVectorFP<0,1,0b010,"fsub", fsub>;
2991 defm MLA : SIMDThreeSameVectorBHSTied<0, 0b10010, "mla",
2992 TriOpFrag<(add node:$LHS, (mul node:$MHS, node:$RHS))> >;
2993 defm MLS : SIMDThreeSameVectorBHSTied<1, 0b10010, "mls",
2994 TriOpFrag<(sub node:$LHS, (mul node:$MHS, node:$RHS))> >;
2995 defm MUL : SIMDThreeSameVectorBHS<0, 0b10011, "mul", mul>;
2996 defm PMUL : SIMDThreeSameVectorB<1, 0b10011, "pmul", int_aarch64_neon_pmul>;
2997 defm SABA : SIMDThreeSameVectorBHSTied<0, 0b01111, "saba",
2998 TriOpFrag<(add node:$LHS, (int_aarch64_neon_sabd node:$MHS, node:$RHS))> >;
2999 defm SABD : SIMDThreeSameVectorBHS<0,0b01110,"sabd", int_aarch64_neon_sabd>;
3000 defm SHADD : SIMDThreeSameVectorBHS<0,0b00000,"shadd", int_aarch64_neon_shadd>;
3001 defm SHSUB : SIMDThreeSameVectorBHS<0,0b00100,"shsub", int_aarch64_neon_shsub>;
3002 defm SMAXP : SIMDThreeSameVectorBHS<0,0b10100,"smaxp", int_aarch64_neon_smaxp>;
3003 defm SMAX : SIMDThreeSameVectorBHS<0,0b01100,"smax", smax>;
3004 defm SMINP : SIMDThreeSameVectorBHS<0,0b10101,"sminp", int_aarch64_neon_sminp>;
3005 defm SMIN : SIMDThreeSameVectorBHS<0,0b01101,"smin", smin>;
3006 defm SQADD : SIMDThreeSameVector<0,0b00001,"sqadd", int_aarch64_neon_sqadd>;
3007 defm SQDMULH : SIMDThreeSameVectorHS<0,0b10110,"sqdmulh",int_aarch64_neon_sqdmulh>;
3008 defm SQRDMULH : SIMDThreeSameVectorHS<1,0b10110,"sqrdmulh",int_aarch64_neon_sqrdmulh>;
3009 defm SQRSHL : SIMDThreeSameVector<0,0b01011,"sqrshl", int_aarch64_neon_sqrshl>;
3010 defm SQSHL : SIMDThreeSameVector<0,0b01001,"sqshl", int_aarch64_neon_sqshl>;
3011 defm SQSUB : SIMDThreeSameVector<0,0b00101,"sqsub", int_aarch64_neon_sqsub>;
3012 defm SRHADD : SIMDThreeSameVectorBHS<0,0b00010,"srhadd",int_aarch64_neon_srhadd>;
3013 defm SRSHL : SIMDThreeSameVector<0,0b01010,"srshl", int_aarch64_neon_srshl>;
3014 defm SSHL : SIMDThreeSameVector<0,0b01000,"sshl", int_aarch64_neon_sshl>;
3015 defm SUB : SIMDThreeSameVector<1,0b10000,"sub", sub>;
3016 defm UABA : SIMDThreeSameVectorBHSTied<1, 0b01111, "uaba",
3017 TriOpFrag<(add node:$LHS, (int_aarch64_neon_uabd node:$MHS, node:$RHS))> >;
3018 defm UABD : SIMDThreeSameVectorBHS<1,0b01110,"uabd", int_aarch64_neon_uabd>;
3019 defm UHADD : SIMDThreeSameVectorBHS<1,0b00000,"uhadd", int_aarch64_neon_uhadd>;
3020 defm UHSUB : SIMDThreeSameVectorBHS<1,0b00100,"uhsub", int_aarch64_neon_uhsub>;
3021 defm UMAXP : SIMDThreeSameVectorBHS<1,0b10100,"umaxp", int_aarch64_neon_umaxp>;
3022 defm UMAX : SIMDThreeSameVectorBHS<1,0b01100,"umax", umax>;
3023 defm UMINP : SIMDThreeSameVectorBHS<1,0b10101,"uminp", int_aarch64_neon_uminp>;
3024 defm UMIN : SIMDThreeSameVectorBHS<1,0b01101,"umin", umin>;
3025 defm UQADD : SIMDThreeSameVector<1,0b00001,"uqadd", int_aarch64_neon_uqadd>;
3026 defm UQRSHL : SIMDThreeSameVector<1,0b01011,"uqrshl", int_aarch64_neon_uqrshl>;
3027 defm UQSHL : SIMDThreeSameVector<1,0b01001,"uqshl", int_aarch64_neon_uqshl>;
3028 defm UQSUB : SIMDThreeSameVector<1,0b00101,"uqsub", int_aarch64_neon_uqsub>;
3029 defm URHADD : SIMDThreeSameVectorBHS<1,0b00010,"urhadd", int_aarch64_neon_urhadd>;
3030 defm URSHL : SIMDThreeSameVector<1,0b01010,"urshl", int_aarch64_neon_urshl>;
3031 defm USHL : SIMDThreeSameVector<1,0b01000,"ushl", int_aarch64_neon_ushl>;
3032 defm SQRDMLAH : SIMDThreeSameVectorSQRDMLxHTiedHS<1,0b10000,"sqrdmlah",
3033 int_aarch64_neon_sqadd>;
3034 defm SQRDMLSH : SIMDThreeSameVectorSQRDMLxHTiedHS<1,0b10001,"sqrdmlsh",
3035 int_aarch64_neon_sqsub>;
3037 defm AND : SIMDLogicalThreeVector<0, 0b00, "and", and>;
3038 defm BIC : SIMDLogicalThreeVector<0, 0b01, "bic",
3039 BinOpFrag<(and node:$LHS, (vnot node:$RHS))> >;
3040 defm BIF : SIMDLogicalThreeVector<1, 0b11, "bif">;
3041 defm BIT : SIMDLogicalThreeVectorTied<1, 0b10, "bit", AArch64bit>;
3042 defm BSL : SIMDLogicalThreeVectorTied<1, 0b01, "bsl",
3043 TriOpFrag<(or (and node:$LHS, node:$MHS), (and (vnot node:$LHS), node:$RHS))>>;
3044 defm EOR : SIMDLogicalThreeVector<1, 0b00, "eor", xor>;
3045 defm ORN : SIMDLogicalThreeVector<0, 0b11, "orn",
3046 BinOpFrag<(or node:$LHS, (vnot node:$RHS))> >;
3047 defm ORR : SIMDLogicalThreeVector<0, 0b10, "orr", or>;
3050 def : Pat<(AArch64bsl (v8i8 V64:$Rd), V64:$Rn, V64:$Rm),
3051 (BSLv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
3052 def : Pat<(AArch64bsl (v4i16 V64:$Rd), V64:$Rn, V64:$Rm),
3053 (BSLv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
3054 def : Pat<(AArch64bsl (v2i32 V64:$Rd), V64:$Rn, V64:$Rm),
3055 (BSLv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
3056 def : Pat<(AArch64bsl (v1i64 V64:$Rd), V64:$Rn, V64:$Rm),
3057 (BSLv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
3059 def : Pat<(AArch64bsl (v16i8 V128:$Rd), V128:$Rn, V128:$Rm),
3060 (BSLv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
3061 def : Pat<(AArch64bsl (v8i16 V128:$Rd), V128:$Rn, V128:$Rm),
3062 (BSLv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
3063 def : Pat<(AArch64bsl (v4i32 V128:$Rd), V128:$Rn, V128:$Rm),
3064 (BSLv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
3065 def : Pat<(AArch64bsl (v2i64 V128:$Rd), V128:$Rn, V128:$Rm),
3066 (BSLv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
3068 def : InstAlias<"mov{\t$dst.16b, $src.16b|.16b\t$dst, $src}",
3069 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 1>;
3070 def : InstAlias<"mov{\t$dst.8h, $src.8h|.8h\t$dst, $src}",
3071 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
3072 def : InstAlias<"mov{\t$dst.4s, $src.4s|.4s\t$dst, $src}",
3073 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
3074 def : InstAlias<"mov{\t$dst.2d, $src.2d|.2d\t$dst, $src}",
3075 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
3077 def : InstAlias<"mov{\t$dst.8b, $src.8b|.8b\t$dst, $src}",
3078 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 1>;
3079 def : InstAlias<"mov{\t$dst.4h, $src.4h|.4h\t$dst, $src}",
3080 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
3081 def : InstAlias<"mov{\t$dst.2s, $src.2s|.2s\t$dst, $src}",
3082 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
3083 def : InstAlias<"mov{\t$dst.1d, $src.1d|.1d\t$dst, $src}",
3084 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
3086 def : InstAlias<"{cmls\t$dst.8b, $src1.8b, $src2.8b" #
3087 "|cmls.8b\t$dst, $src1, $src2}",
3088 (CMHSv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
3089 def : InstAlias<"{cmls\t$dst.16b, $src1.16b, $src2.16b" #
3090 "|cmls.16b\t$dst, $src1, $src2}",
3091 (CMHSv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
3092 def : InstAlias<"{cmls\t$dst.4h, $src1.4h, $src2.4h" #
3093 "|cmls.4h\t$dst, $src1, $src2}",
3094 (CMHSv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
3095 def : InstAlias<"{cmls\t$dst.8h, $src1.8h, $src2.8h" #
3096 "|cmls.8h\t$dst, $src1, $src2}",
3097 (CMHSv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
3098 def : InstAlias<"{cmls\t$dst.2s, $src1.2s, $src2.2s" #
3099 "|cmls.2s\t$dst, $src1, $src2}",
3100 (CMHSv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
3101 def : InstAlias<"{cmls\t$dst.4s, $src1.4s, $src2.4s" #
3102 "|cmls.4s\t$dst, $src1, $src2}",
3103 (CMHSv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
3104 def : InstAlias<"{cmls\t$dst.2d, $src1.2d, $src2.2d" #
3105 "|cmls.2d\t$dst, $src1, $src2}",
3106 (CMHSv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
3108 def : InstAlias<"{cmlo\t$dst.8b, $src1.8b, $src2.8b" #
3109 "|cmlo.8b\t$dst, $src1, $src2}",
3110 (CMHIv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
3111 def : InstAlias<"{cmlo\t$dst.16b, $src1.16b, $src2.16b" #
3112 "|cmlo.16b\t$dst, $src1, $src2}",
3113 (CMHIv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
3114 def : InstAlias<"{cmlo\t$dst.4h, $src1.4h, $src2.4h" #
3115 "|cmlo.4h\t$dst, $src1, $src2}",
3116 (CMHIv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
3117 def : InstAlias<"{cmlo\t$dst.8h, $src1.8h, $src2.8h" #
3118 "|cmlo.8h\t$dst, $src1, $src2}",
3119 (CMHIv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
3120 def : InstAlias<"{cmlo\t$dst.2s, $src1.2s, $src2.2s" #
3121 "|cmlo.2s\t$dst, $src1, $src2}",
3122 (CMHIv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
3123 def : InstAlias<"{cmlo\t$dst.4s, $src1.4s, $src2.4s" #
3124 "|cmlo.4s\t$dst, $src1, $src2}",
3125 (CMHIv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
3126 def : InstAlias<"{cmlo\t$dst.2d, $src1.2d, $src2.2d" #
3127 "|cmlo.2d\t$dst, $src1, $src2}",
3128 (CMHIv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
3130 def : InstAlias<"{cmle\t$dst.8b, $src1.8b, $src2.8b" #
3131 "|cmle.8b\t$dst, $src1, $src2}",
3132 (CMGEv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
3133 def : InstAlias<"{cmle\t$dst.16b, $src1.16b, $src2.16b" #
3134 "|cmle.16b\t$dst, $src1, $src2}",
3135 (CMGEv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
3136 def : InstAlias<"{cmle\t$dst.4h, $src1.4h, $src2.4h" #
3137 "|cmle.4h\t$dst, $src1, $src2}",
3138 (CMGEv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
3139 def : InstAlias<"{cmle\t$dst.8h, $src1.8h, $src2.8h" #
3140 "|cmle.8h\t$dst, $src1, $src2}",
3141 (CMGEv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
3142 def : InstAlias<"{cmle\t$dst.2s, $src1.2s, $src2.2s" #
3143 "|cmle.2s\t$dst, $src1, $src2}",
3144 (CMGEv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
3145 def : InstAlias<"{cmle\t$dst.4s, $src1.4s, $src2.4s" #
3146 "|cmle.4s\t$dst, $src1, $src2}",
3147 (CMGEv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
3148 def : InstAlias<"{cmle\t$dst.2d, $src1.2d, $src2.2d" #
3149 "|cmle.2d\t$dst, $src1, $src2}",
3150 (CMGEv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
3152 def : InstAlias<"{cmlt\t$dst.8b, $src1.8b, $src2.8b" #
3153 "|cmlt.8b\t$dst, $src1, $src2}",
3154 (CMGTv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
3155 def : InstAlias<"{cmlt\t$dst.16b, $src1.16b, $src2.16b" #
3156 "|cmlt.16b\t$dst, $src1, $src2}",
3157 (CMGTv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
3158 def : InstAlias<"{cmlt\t$dst.4h, $src1.4h, $src2.4h" #
3159 "|cmlt.4h\t$dst, $src1, $src2}",
3160 (CMGTv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
3161 def : InstAlias<"{cmlt\t$dst.8h, $src1.8h, $src2.8h" #
3162 "|cmlt.8h\t$dst, $src1, $src2}",
3163 (CMGTv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
3164 def : InstAlias<"{cmlt\t$dst.2s, $src1.2s, $src2.2s" #
3165 "|cmlt.2s\t$dst, $src1, $src2}",
3166 (CMGTv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
3167 def : InstAlias<"{cmlt\t$dst.4s, $src1.4s, $src2.4s" #
3168 "|cmlt.4s\t$dst, $src1, $src2}",
3169 (CMGTv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
3170 def : InstAlias<"{cmlt\t$dst.2d, $src1.2d, $src2.2d" #
3171 "|cmlt.2d\t$dst, $src1, $src2}",
3172 (CMGTv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
3174 let Predicates = [HasNEON, HasFullFP16] in {
3175 def : InstAlias<"{fcmle\t$dst.4h, $src1.4h, $src2.4h" #
3176 "|fcmle.4h\t$dst, $src1, $src2}",
3177 (FCMGEv4f16 V64:$dst, V64:$src2, V64:$src1), 0>;
3178 def : InstAlias<"{fcmle\t$dst.8h, $src1.8h, $src2.8h" #
3179 "|fcmle.8h\t$dst, $src1, $src2}",
3180 (FCMGEv8f16 V128:$dst, V128:$src2, V128:$src1), 0>;
3182 def : InstAlias<"{fcmle\t$dst.2s, $src1.2s, $src2.2s" #
3183 "|fcmle.2s\t$dst, $src1, $src2}",
3184 (FCMGEv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
3185 def : InstAlias<"{fcmle\t$dst.4s, $src1.4s, $src2.4s" #
3186 "|fcmle.4s\t$dst, $src1, $src2}",
3187 (FCMGEv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
3188 def : InstAlias<"{fcmle\t$dst.2d, $src1.2d, $src2.2d" #
3189 "|fcmle.2d\t$dst, $src1, $src2}",
3190 (FCMGEv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
3192 let Predicates = [HasNEON, HasFullFP16] in {
3193 def : InstAlias<"{fcmlt\t$dst.4h, $src1.4h, $src2.4h" #
3194 "|fcmlt.4h\t$dst, $src1, $src2}",
3195 (FCMGTv4f16 V64:$dst, V64:$src2, V64:$src1), 0>;
3196 def : InstAlias<"{fcmlt\t$dst.8h, $src1.8h, $src2.8h" #
3197 "|fcmlt.8h\t$dst, $src1, $src2}",
3198 (FCMGTv8f16 V128:$dst, V128:$src2, V128:$src1), 0>;
3200 def : InstAlias<"{fcmlt\t$dst.2s, $src1.2s, $src2.2s" #
3201 "|fcmlt.2s\t$dst, $src1, $src2}",
3202 (FCMGTv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
3203 def : InstAlias<"{fcmlt\t$dst.4s, $src1.4s, $src2.4s" #
3204 "|fcmlt.4s\t$dst, $src1, $src2}",
3205 (FCMGTv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
3206 def : InstAlias<"{fcmlt\t$dst.2d, $src1.2d, $src2.2d" #
3207 "|fcmlt.2d\t$dst, $src1, $src2}",
3208 (FCMGTv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
3210 let Predicates = [HasNEON, HasFullFP16] in {
3211 def : InstAlias<"{facle\t$dst.4h, $src1.4h, $src2.4h" #
3212 "|facle.4h\t$dst, $src1, $src2}",
3213 (FACGEv4f16 V64:$dst, V64:$src2, V64:$src1), 0>;
3214 def : InstAlias<"{facle\t$dst.8h, $src1.8h, $src2.8h" #
3215 "|facle.8h\t$dst, $src1, $src2}",
3216 (FACGEv8f16 V128:$dst, V128:$src2, V128:$src1), 0>;
3218 def : InstAlias<"{facle\t$dst.2s, $src1.2s, $src2.2s" #
3219 "|facle.2s\t$dst, $src1, $src2}",
3220 (FACGEv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
3221 def : InstAlias<"{facle\t$dst.4s, $src1.4s, $src2.4s" #
3222 "|facle.4s\t$dst, $src1, $src2}",
3223 (FACGEv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
3224 def : InstAlias<"{facle\t$dst.2d, $src1.2d, $src2.2d" #
3225 "|facle.2d\t$dst, $src1, $src2}",
3226 (FACGEv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
3228 let Predicates = [HasNEON, HasFullFP16] in {
3229 def : InstAlias<"{faclt\t$dst.4h, $src1.4h, $src2.4h" #
3230 "|faclt.4h\t$dst, $src1, $src2}",
3231 (FACGTv4f16 V64:$dst, V64:$src2, V64:$src1), 0>;
3232 def : InstAlias<"{faclt\t$dst.8h, $src1.8h, $src2.8h" #
3233 "|faclt.8h\t$dst, $src1, $src2}",
3234 (FACGTv8f16 V128:$dst, V128:$src2, V128:$src1), 0>;
3236 def : InstAlias<"{faclt\t$dst.2s, $src1.2s, $src2.2s" #
3237 "|faclt.2s\t$dst, $src1, $src2}",
3238 (FACGTv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
3239 def : InstAlias<"{faclt\t$dst.4s, $src1.4s, $src2.4s" #
3240 "|faclt.4s\t$dst, $src1, $src2}",
3241 (FACGTv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
3242 def : InstAlias<"{faclt\t$dst.2d, $src1.2d, $src2.2d" #
3243 "|faclt.2d\t$dst, $src1, $src2}",
3244 (FACGTv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
3246 //===----------------------------------------------------------------------===//
3247 // Advanced SIMD three scalar instructions.
3248 //===----------------------------------------------------------------------===//
3250 defm ADD : SIMDThreeScalarD<0, 0b10000, "add", add>;
3251 defm CMEQ : SIMDThreeScalarD<1, 0b10001, "cmeq", AArch64cmeq>;
3252 defm CMGE : SIMDThreeScalarD<0, 0b00111, "cmge", AArch64cmge>;
3253 defm CMGT : SIMDThreeScalarD<0, 0b00110, "cmgt", AArch64cmgt>;
3254 defm CMHI : SIMDThreeScalarD<1, 0b00110, "cmhi", AArch64cmhi>;
3255 defm CMHS : SIMDThreeScalarD<1, 0b00111, "cmhs", AArch64cmhs>;
3256 defm CMTST : SIMDThreeScalarD<0, 0b10001, "cmtst", AArch64cmtst>;
3257 defm FABD : SIMDFPThreeScalar<1, 1, 0b010, "fabd", int_aarch64_sisd_fabd>;
3258 def : Pat<(v1f64 (int_aarch64_neon_fabd (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
3259 (FABD64 FPR64:$Rn, FPR64:$Rm)>;
3260 defm FACGE : SIMDThreeScalarFPCmp<1, 0, 0b101, "facge",
3261 int_aarch64_neon_facge>;
3262 defm FACGT : SIMDThreeScalarFPCmp<1, 1, 0b101, "facgt",
3263 int_aarch64_neon_facgt>;
3264 defm FCMEQ : SIMDThreeScalarFPCmp<0, 0, 0b100, "fcmeq", AArch64fcmeq>;
3265 defm FCMGE : SIMDThreeScalarFPCmp<1, 0, 0b100, "fcmge", AArch64fcmge>;
3266 defm FCMGT : SIMDThreeScalarFPCmp<1, 1, 0b100, "fcmgt", AArch64fcmgt>;
3267 defm FMULX : SIMDFPThreeScalar<0, 0, 0b011, "fmulx", int_aarch64_neon_fmulx>;
3268 defm FRECPS : SIMDFPThreeScalar<0, 0, 0b111, "frecps", int_aarch64_neon_frecps>;
3269 defm FRSQRTS : SIMDFPThreeScalar<0, 1, 0b111, "frsqrts", int_aarch64_neon_frsqrts>;
3270 defm SQADD : SIMDThreeScalarBHSD<0, 0b00001, "sqadd", int_aarch64_neon_sqadd>;
3271 defm SQDMULH : SIMDThreeScalarHS< 0, 0b10110, "sqdmulh", int_aarch64_neon_sqdmulh>;
3272 defm SQRDMULH : SIMDThreeScalarHS< 1, 0b10110, "sqrdmulh", int_aarch64_neon_sqrdmulh>;
3273 defm SQRSHL : SIMDThreeScalarBHSD<0, 0b01011, "sqrshl",int_aarch64_neon_sqrshl>;
3274 defm SQSHL : SIMDThreeScalarBHSD<0, 0b01001, "sqshl", int_aarch64_neon_sqshl>;
3275 defm SQSUB : SIMDThreeScalarBHSD<0, 0b00101, "sqsub", int_aarch64_neon_sqsub>;
3276 defm SRSHL : SIMDThreeScalarD< 0, 0b01010, "srshl", int_aarch64_neon_srshl>;
3277 defm SSHL : SIMDThreeScalarD< 0, 0b01000, "sshl", int_aarch64_neon_sshl>;
3278 defm SUB : SIMDThreeScalarD< 1, 0b10000, "sub", sub>;
3279 defm UQADD : SIMDThreeScalarBHSD<1, 0b00001, "uqadd", int_aarch64_neon_uqadd>;
3280 defm UQRSHL : SIMDThreeScalarBHSD<1, 0b01011, "uqrshl",int_aarch64_neon_uqrshl>;
3281 defm UQSHL : SIMDThreeScalarBHSD<1, 0b01001, "uqshl", int_aarch64_neon_uqshl>;
3282 defm UQSUB : SIMDThreeScalarBHSD<1, 0b00101, "uqsub", int_aarch64_neon_uqsub>;
3283 defm URSHL : SIMDThreeScalarD< 1, 0b01010, "urshl", int_aarch64_neon_urshl>;
3284 defm USHL : SIMDThreeScalarD< 1, 0b01000, "ushl", int_aarch64_neon_ushl>;
3285 let Predicates = [HasRDM] in {
3286 defm SQRDMLAH : SIMDThreeScalarHSTied<1, 0, 0b10000, "sqrdmlah">;
3287 defm SQRDMLSH : SIMDThreeScalarHSTied<1, 0, 0b10001, "sqrdmlsh">;
3288 def : Pat<(i32 (int_aarch64_neon_sqadd
3290 (i32 (int_aarch64_neon_sqrdmulh (i32 FPR32:$Rn),
3291 (i32 FPR32:$Rm))))),
3292 (SQRDMLAHv1i32 FPR32:$Rd, FPR32:$Rn, FPR32:$Rm)>;
3293 def : Pat<(i32 (int_aarch64_neon_sqsub
3295 (i32 (int_aarch64_neon_sqrdmulh (i32 FPR32:$Rn),
3296 (i32 FPR32:$Rm))))),
3297 (SQRDMLSHv1i32 FPR32:$Rd, FPR32:$Rn, FPR32:$Rm)>;
3300 def : InstAlias<"cmls $dst, $src1, $src2",
3301 (CMHSv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
3302 def : InstAlias<"cmle $dst, $src1, $src2",
3303 (CMGEv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
3304 def : InstAlias<"cmlo $dst, $src1, $src2",
3305 (CMHIv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
3306 def : InstAlias<"cmlt $dst, $src1, $src2",
3307 (CMGTv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
3308 def : InstAlias<"fcmle $dst, $src1, $src2",
3309 (FCMGE32 FPR32:$dst, FPR32:$src2, FPR32:$src1), 0>;
3310 def : InstAlias<"fcmle $dst, $src1, $src2",
3311 (FCMGE64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
3312 def : InstAlias<"fcmlt $dst, $src1, $src2",
3313 (FCMGT32 FPR32:$dst, FPR32:$src2, FPR32:$src1), 0>;
3314 def : InstAlias<"fcmlt $dst, $src1, $src2",
3315 (FCMGT64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
3316 def : InstAlias<"facle $dst, $src1, $src2",
3317 (FACGE32 FPR32:$dst, FPR32:$src2, FPR32:$src1), 0>;
3318 def : InstAlias<"facle $dst, $src1, $src2",
3319 (FACGE64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
3320 def : InstAlias<"faclt $dst, $src1, $src2",
3321 (FACGT32 FPR32:$dst, FPR32:$src2, FPR32:$src1), 0>;
3322 def : InstAlias<"faclt $dst, $src1, $src2",
3323 (FACGT64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
3325 //===----------------------------------------------------------------------===//
3326 // Advanced SIMD three scalar instructions (mixed operands).
3327 //===----------------------------------------------------------------------===//
3328 defm SQDMULL : SIMDThreeScalarMixedHS<0, 0b11010, "sqdmull",
3329 int_aarch64_neon_sqdmulls_scalar>;
3330 defm SQDMLAL : SIMDThreeScalarMixedTiedHS<0, 0b10010, "sqdmlal">;
3331 defm SQDMLSL : SIMDThreeScalarMixedTiedHS<0, 0b10110, "sqdmlsl">;
3333 def : Pat<(i64 (int_aarch64_neon_sqadd (i64 FPR64:$Rd),
3334 (i64 (int_aarch64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
3335 (i32 FPR32:$Rm))))),
3336 (SQDMLALi32 FPR64:$Rd, FPR32:$Rn, FPR32:$Rm)>;
3337 def : Pat<(i64 (int_aarch64_neon_sqsub (i64 FPR64:$Rd),
3338 (i64 (int_aarch64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
3339 (i32 FPR32:$Rm))))),
3340 (SQDMLSLi32 FPR64:$Rd, FPR32:$Rn, FPR32:$Rm)>;
3342 //===----------------------------------------------------------------------===//
3343 // Advanced SIMD two scalar instructions.
3344 //===----------------------------------------------------------------------===//
3346 defm ABS : SIMDTwoScalarD< 0, 0b01011, "abs", abs>;
3347 defm CMEQ : SIMDCmpTwoScalarD< 0, 0b01001, "cmeq", AArch64cmeqz>;
3348 defm CMGE : SIMDCmpTwoScalarD< 1, 0b01000, "cmge", AArch64cmgez>;
3349 defm CMGT : SIMDCmpTwoScalarD< 0, 0b01000, "cmgt", AArch64cmgtz>;
3350 defm CMLE : SIMDCmpTwoScalarD< 1, 0b01001, "cmle", AArch64cmlez>;
3351 defm CMLT : SIMDCmpTwoScalarD< 0, 0b01010, "cmlt", AArch64cmltz>;
3352 defm FCMEQ : SIMDFPCmpTwoScalar<0, 1, 0b01101, "fcmeq", AArch64fcmeqz>;
3353 defm FCMGE : SIMDFPCmpTwoScalar<1, 1, 0b01100, "fcmge", AArch64fcmgez>;
3354 defm FCMGT : SIMDFPCmpTwoScalar<0, 1, 0b01100, "fcmgt", AArch64fcmgtz>;
3355 defm FCMLE : SIMDFPCmpTwoScalar<1, 1, 0b01101, "fcmle", AArch64fcmlez>;
3356 defm FCMLT : SIMDFPCmpTwoScalar<0, 1, 0b01110, "fcmlt", AArch64fcmltz>;
3357 defm FCVTAS : SIMDFPTwoScalar< 0, 0, 0b11100, "fcvtas">;
3358 defm FCVTAU : SIMDFPTwoScalar< 1, 0, 0b11100, "fcvtau">;
3359 defm FCVTMS : SIMDFPTwoScalar< 0, 0, 0b11011, "fcvtms">;
3360 defm FCVTMU : SIMDFPTwoScalar< 1, 0, 0b11011, "fcvtmu">;
3361 defm FCVTNS : SIMDFPTwoScalar< 0, 0, 0b11010, "fcvtns">;
3362 defm FCVTNU : SIMDFPTwoScalar< 1, 0, 0b11010, "fcvtnu">;
3363 defm FCVTPS : SIMDFPTwoScalar< 0, 1, 0b11010, "fcvtps">;
3364 defm FCVTPU : SIMDFPTwoScalar< 1, 1, 0b11010, "fcvtpu">;
3365 def FCVTXNv1i64 : SIMDInexactCvtTwoScalar<0b10110, "fcvtxn">;
3366 defm FCVTZS : SIMDFPTwoScalar< 0, 1, 0b11011, "fcvtzs">;
3367 defm FCVTZU : SIMDFPTwoScalar< 1, 1, 0b11011, "fcvtzu">;
3368 defm FRECPE : SIMDFPTwoScalar< 0, 1, 0b11101, "frecpe">;
3369 defm FRECPX : SIMDFPTwoScalar< 0, 1, 0b11111, "frecpx">;
3370 defm FRSQRTE : SIMDFPTwoScalar< 1, 1, 0b11101, "frsqrte">;
3371 defm NEG : SIMDTwoScalarD< 1, 0b01011, "neg",
3372 UnOpFrag<(sub immAllZerosV, node:$LHS)> >;
3373 defm SCVTF : SIMDFPTwoScalarCVT< 0, 0, 0b11101, "scvtf", AArch64sitof>;
3374 defm SQABS : SIMDTwoScalarBHSD< 0, 0b00111, "sqabs", int_aarch64_neon_sqabs>;
3375 defm SQNEG : SIMDTwoScalarBHSD< 1, 0b00111, "sqneg", int_aarch64_neon_sqneg>;
3376 defm SQXTN : SIMDTwoScalarMixedBHS< 0, 0b10100, "sqxtn", int_aarch64_neon_scalar_sqxtn>;
3377 defm SQXTUN : SIMDTwoScalarMixedBHS< 1, 0b10010, "sqxtun", int_aarch64_neon_scalar_sqxtun>;
3378 defm SUQADD : SIMDTwoScalarBHSDTied< 0, 0b00011, "suqadd",
3379 int_aarch64_neon_suqadd>;
3380 defm UCVTF : SIMDFPTwoScalarCVT< 1, 0, 0b11101, "ucvtf", AArch64uitof>;
3381 defm UQXTN : SIMDTwoScalarMixedBHS<1, 0b10100, "uqxtn", int_aarch64_neon_scalar_uqxtn>;
3382 defm USQADD : SIMDTwoScalarBHSDTied< 1, 0b00011, "usqadd",
3383 int_aarch64_neon_usqadd>;
3385 def : Pat<(AArch64neg (v1i64 V64:$Rn)), (NEGv1i64 V64:$Rn)>;
3387 def : Pat<(v1i64 (int_aarch64_neon_fcvtas (v1f64 FPR64:$Rn))),
3388 (FCVTASv1i64 FPR64:$Rn)>;
3389 def : Pat<(v1i64 (int_aarch64_neon_fcvtau (v1f64 FPR64:$Rn))),
3390 (FCVTAUv1i64 FPR64:$Rn)>;
3391 def : Pat<(v1i64 (int_aarch64_neon_fcvtms (v1f64 FPR64:$Rn))),
3392 (FCVTMSv1i64 FPR64:$Rn)>;
3393 def : Pat<(v1i64 (int_aarch64_neon_fcvtmu (v1f64 FPR64:$Rn))),
3394 (FCVTMUv1i64 FPR64:$Rn)>;
3395 def : Pat<(v1i64 (int_aarch64_neon_fcvtns (v1f64 FPR64:$Rn))),
3396 (FCVTNSv1i64 FPR64:$Rn)>;
3397 def : Pat<(v1i64 (int_aarch64_neon_fcvtnu (v1f64 FPR64:$Rn))),
3398 (FCVTNUv1i64 FPR64:$Rn)>;
3399 def : Pat<(v1i64 (int_aarch64_neon_fcvtps (v1f64 FPR64:$Rn))),
3400 (FCVTPSv1i64 FPR64:$Rn)>;
3401 def : Pat<(v1i64 (int_aarch64_neon_fcvtpu (v1f64 FPR64:$Rn))),
3402 (FCVTPUv1i64 FPR64:$Rn)>;
3404 def : Pat<(f32 (int_aarch64_neon_frecpe (f32 FPR32:$Rn))),
3405 (FRECPEv1i32 FPR32:$Rn)>;
3406 def : Pat<(f64 (int_aarch64_neon_frecpe (f64 FPR64:$Rn))),
3407 (FRECPEv1i64 FPR64:$Rn)>;
3408 def : Pat<(v1f64 (int_aarch64_neon_frecpe (v1f64 FPR64:$Rn))),
3409 (FRECPEv1i64 FPR64:$Rn)>;
3411 def : Pat<(f32 (AArch64frecpe (f32 FPR32:$Rn))),
3412 (FRECPEv1i32 FPR32:$Rn)>;
3413 def : Pat<(v2f32 (AArch64frecpe (v2f32 V64:$Rn))),
3414 (FRECPEv2f32 V64:$Rn)>;
3415 def : Pat<(v4f32 (AArch64frecpe (v4f32 FPR128:$Rn))),
3416 (FRECPEv4f32 FPR128:$Rn)>;
3417 def : Pat<(f64 (AArch64frecpe (f64 FPR64:$Rn))),
3418 (FRECPEv1i64 FPR64:$Rn)>;
3419 def : Pat<(v1f64 (AArch64frecpe (v1f64 FPR64:$Rn))),
3420 (FRECPEv1i64 FPR64:$Rn)>;
3421 def : Pat<(v2f64 (AArch64frecpe (v2f64 FPR128:$Rn))),
3422 (FRECPEv2f64 FPR128:$Rn)>;
3424 def : Pat<(f32 (AArch64frecps (f32 FPR32:$Rn), (f32 FPR32:$Rm))),
3425 (FRECPS32 FPR32:$Rn, FPR32:$Rm)>;
3426 def : Pat<(v2f32 (AArch64frecps (v2f32 V64:$Rn), (v2f32 V64:$Rm))),
3427 (FRECPSv2f32 V64:$Rn, V64:$Rm)>;
3428 def : Pat<(v4f32 (AArch64frecps (v4f32 FPR128:$Rn), (v4f32 FPR128:$Rm))),
3429 (FRECPSv4f32 FPR128:$Rn, FPR128:$Rm)>;
3430 def : Pat<(f64 (AArch64frecps (f64 FPR64:$Rn), (f64 FPR64:$Rm))),
3431 (FRECPS64 FPR64:$Rn, FPR64:$Rm)>;
3432 def : Pat<(v2f64 (AArch64frecps (v2f64 FPR128:$Rn), (v2f64 FPR128:$Rm))),
3433 (FRECPSv2f64 FPR128:$Rn, FPR128:$Rm)>;
3435 def : Pat<(f32 (int_aarch64_neon_frecpx (f32 FPR32:$Rn))),
3436 (FRECPXv1i32 FPR32:$Rn)>;
3437 def : Pat<(f64 (int_aarch64_neon_frecpx (f64 FPR64:$Rn))),
3438 (FRECPXv1i64 FPR64:$Rn)>;
3440 def : Pat<(f32 (int_aarch64_neon_frsqrte (f32 FPR32:$Rn))),
3441 (FRSQRTEv1i32 FPR32:$Rn)>;
3442 def : Pat<(f64 (int_aarch64_neon_frsqrte (f64 FPR64:$Rn))),
3443 (FRSQRTEv1i64 FPR64:$Rn)>;
3444 def : Pat<(v1f64 (int_aarch64_neon_frsqrte (v1f64 FPR64:$Rn))),
3445 (FRSQRTEv1i64 FPR64:$Rn)>;
3447 def : Pat<(f32 (AArch64frsqrte (f32 FPR32:$Rn))),
3448 (FRSQRTEv1i32 FPR32:$Rn)>;
3449 def : Pat<(v2f32 (AArch64frsqrte (v2f32 V64:$Rn))),
3450 (FRSQRTEv2f32 V64:$Rn)>;
3451 def : Pat<(v4f32 (AArch64frsqrte (v4f32 FPR128:$Rn))),
3452 (FRSQRTEv4f32 FPR128:$Rn)>;
3453 def : Pat<(f64 (AArch64frsqrte (f64 FPR64:$Rn))),
3454 (FRSQRTEv1i64 FPR64:$Rn)>;
3455 def : Pat<(v1f64 (AArch64frsqrte (v1f64 FPR64:$Rn))),
3456 (FRSQRTEv1i64 FPR64:$Rn)>;
3457 def : Pat<(v2f64 (AArch64frsqrte (v2f64 FPR128:$Rn))),
3458 (FRSQRTEv2f64 FPR128:$Rn)>;
3460 def : Pat<(f32 (AArch64frsqrts (f32 FPR32:$Rn), (f32 FPR32:$Rm))),
3461 (FRSQRTS32 FPR32:$Rn, FPR32:$Rm)>;
3462 def : Pat<(v2f32 (AArch64frsqrts (v2f32 V64:$Rn), (v2f32 V64:$Rm))),
3463 (FRSQRTSv2f32 V64:$Rn, V64:$Rm)>;
3464 def : Pat<(v4f32 (AArch64frsqrts (v4f32 FPR128:$Rn), (v4f32 FPR128:$Rm))),
3465 (FRSQRTSv4f32 FPR128:$Rn, FPR128:$Rm)>;
3466 def : Pat<(f64 (AArch64frsqrts (f64 FPR64:$Rn), (f64 FPR64:$Rm))),
3467 (FRSQRTS64 FPR64:$Rn, FPR64:$Rm)>;
3468 def : Pat<(v2f64 (AArch64frsqrts (v2f64 FPR128:$Rn), (v2f64 FPR128:$Rm))),
3469 (FRSQRTSv2f64 FPR128:$Rn, FPR128:$Rm)>;
3471 // If an integer is about to be converted to a floating point value,
3472 // just load it on the floating point unit.
3473 // Here are the patterns for 8 and 16-bits to float.
3475 multiclass UIntToFPROLoadPat<ValueType DstTy, ValueType SrcTy,
3476 SDPatternOperator loadop, Instruction UCVTF,
3477 ROAddrMode ro, Instruction LDRW, Instruction LDRX,
3479 def : Pat<(DstTy (uint_to_fp (SrcTy
3480 (loadop (ro.Wpat GPR64sp:$Rn, GPR32:$Rm,
3481 ro.Wext:$extend))))),
3482 (UCVTF (INSERT_SUBREG (DstTy (IMPLICIT_DEF)),
3483 (LDRW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend),
3486 def : Pat<(DstTy (uint_to_fp (SrcTy
3487 (loadop (ro.Xpat GPR64sp:$Rn, GPR64:$Rm,
3488 ro.Wext:$extend))))),
3489 (UCVTF (INSERT_SUBREG (DstTy (IMPLICIT_DEF)),
3490 (LDRX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend),
3494 defm : UIntToFPROLoadPat<f32, i32, zextloadi8,
3495 UCVTFv1i32, ro8, LDRBroW, LDRBroX, bsub>;
3496 def : Pat <(f32 (uint_to_fp (i32
3497 (zextloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))))),
3498 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
3499 (LDRBui GPR64sp:$Rn, uimm12s1:$offset), bsub))>;
3500 def : Pat <(f32 (uint_to_fp (i32
3501 (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))))),
3502 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
3503 (LDURBi GPR64sp:$Rn, simm9:$offset), bsub))>;
3504 // 16-bits -> float.
3505 defm : UIntToFPROLoadPat<f32, i32, zextloadi16,
3506 UCVTFv1i32, ro16, LDRHroW, LDRHroX, hsub>;
3507 def : Pat <(f32 (uint_to_fp (i32
3508 (zextloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))))),
3509 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
3510 (LDRHui GPR64sp:$Rn, uimm12s2:$offset), hsub))>;
3511 def : Pat <(f32 (uint_to_fp (i32
3512 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))))),
3513 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
3514 (LDURHi GPR64sp:$Rn, simm9:$offset), hsub))>;
3515 // 32-bits are handled in target specific dag combine:
3516 // performIntToFpCombine.
3517 // 64-bits integer to 32-bits floating point, not possible with
3518 // UCVTF on floating point registers (both source and destination
3519 // must have the same size).
3521 // Here are the patterns for 8, 16, 32, and 64-bits to double.
3522 // 8-bits -> double.
3523 defm : UIntToFPROLoadPat<f64, i32, zextloadi8,
3524 UCVTFv1i64, ro8, LDRBroW, LDRBroX, bsub>;
3525 def : Pat <(f64 (uint_to_fp (i32
3526 (zextloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))))),
3527 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3528 (LDRBui GPR64sp:$Rn, uimm12s1:$offset), bsub))>;
3529 def : Pat <(f64 (uint_to_fp (i32
3530 (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))))),
3531 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3532 (LDURBi GPR64sp:$Rn, simm9:$offset), bsub))>;
3533 // 16-bits -> double.
3534 defm : UIntToFPROLoadPat<f64, i32, zextloadi16,
3535 UCVTFv1i64, ro16, LDRHroW, LDRHroX, hsub>;
3536 def : Pat <(f64 (uint_to_fp (i32
3537 (zextloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))))),
3538 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3539 (LDRHui GPR64sp:$Rn, uimm12s2:$offset), hsub))>;
3540 def : Pat <(f64 (uint_to_fp (i32
3541 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))))),
3542 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3543 (LDURHi GPR64sp:$Rn, simm9:$offset), hsub))>;
3544 // 32-bits -> double.
3545 defm : UIntToFPROLoadPat<f64, i32, load,
3546 UCVTFv1i64, ro32, LDRSroW, LDRSroX, ssub>;
3547 def : Pat <(f64 (uint_to_fp (i32
3548 (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))))),
3549 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3550 (LDRSui GPR64sp:$Rn, uimm12s4:$offset), ssub))>;
3551 def : Pat <(f64 (uint_to_fp (i32
3552 (load (am_unscaled32 GPR64sp:$Rn, simm9:$offset))))),
3553 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3554 (LDURSi GPR64sp:$Rn, simm9:$offset), ssub))>;
3555 // 64-bits -> double are handled in target specific dag combine:
3556 // performIntToFpCombine.
3558 //===----------------------------------------------------------------------===//
3559 // Advanced SIMD three different-sized vector instructions.
3560 //===----------------------------------------------------------------------===//
3562 defm ADDHN : SIMDNarrowThreeVectorBHS<0,0b0100,"addhn", int_aarch64_neon_addhn>;
3563 defm SUBHN : SIMDNarrowThreeVectorBHS<0,0b0110,"subhn", int_aarch64_neon_subhn>;
3564 defm RADDHN : SIMDNarrowThreeVectorBHS<1,0b0100,"raddhn",int_aarch64_neon_raddhn>;
3565 defm RSUBHN : SIMDNarrowThreeVectorBHS<1,0b0110,"rsubhn",int_aarch64_neon_rsubhn>;
3566 defm PMULL : SIMDDifferentThreeVectorBD<0,0b1110,"pmull",int_aarch64_neon_pmull>;
3567 defm SABAL : SIMDLongThreeVectorTiedBHSabal<0,0b0101,"sabal",
3568 int_aarch64_neon_sabd>;
3569 defm SABDL : SIMDLongThreeVectorBHSabdl<0, 0b0111, "sabdl",
3570 int_aarch64_neon_sabd>;
3571 defm SADDL : SIMDLongThreeVectorBHS< 0, 0b0000, "saddl",
3572 BinOpFrag<(add (sext node:$LHS), (sext node:$RHS))>>;
3573 defm SADDW : SIMDWideThreeVectorBHS< 0, 0b0001, "saddw",
3574 BinOpFrag<(add node:$LHS, (sext node:$RHS))>>;
3575 defm SMLAL : SIMDLongThreeVectorTiedBHS<0, 0b1000, "smlal",
3576 TriOpFrag<(add node:$LHS, (int_aarch64_neon_smull node:$MHS, node:$RHS))>>;
3577 defm SMLSL : SIMDLongThreeVectorTiedBHS<0, 0b1010, "smlsl",
3578 TriOpFrag<(sub node:$LHS, (int_aarch64_neon_smull node:$MHS, node:$RHS))>>;
3579 defm SMULL : SIMDLongThreeVectorBHS<0, 0b1100, "smull", int_aarch64_neon_smull>;
3580 defm SQDMLAL : SIMDLongThreeVectorSQDMLXTiedHS<0, 0b1001, "sqdmlal",
3581 int_aarch64_neon_sqadd>;
3582 defm SQDMLSL : SIMDLongThreeVectorSQDMLXTiedHS<0, 0b1011, "sqdmlsl",
3583 int_aarch64_neon_sqsub>;
3584 defm SQDMULL : SIMDLongThreeVectorHS<0, 0b1101, "sqdmull",
3585 int_aarch64_neon_sqdmull>;
3586 defm SSUBL : SIMDLongThreeVectorBHS<0, 0b0010, "ssubl",
3587 BinOpFrag<(sub (sext node:$LHS), (sext node:$RHS))>>;
3588 defm SSUBW : SIMDWideThreeVectorBHS<0, 0b0011, "ssubw",
3589 BinOpFrag<(sub node:$LHS, (sext node:$RHS))>>;
3590 defm UABAL : SIMDLongThreeVectorTiedBHSabal<1, 0b0101, "uabal",
3591 int_aarch64_neon_uabd>;
3592 defm UADDL : SIMDLongThreeVectorBHS<1, 0b0000, "uaddl",
3593 BinOpFrag<(add (zext node:$LHS), (zext node:$RHS))>>;
3594 defm UADDW : SIMDWideThreeVectorBHS<1, 0b0001, "uaddw",
3595 BinOpFrag<(add node:$LHS, (zext node:$RHS))>>;
3596 defm UMLAL : SIMDLongThreeVectorTiedBHS<1, 0b1000, "umlal",
3597 TriOpFrag<(add node:$LHS, (int_aarch64_neon_umull node:$MHS, node:$RHS))>>;
3598 defm UMLSL : SIMDLongThreeVectorTiedBHS<1, 0b1010, "umlsl",
3599 TriOpFrag<(sub node:$LHS, (int_aarch64_neon_umull node:$MHS, node:$RHS))>>;
3600 defm UMULL : SIMDLongThreeVectorBHS<1, 0b1100, "umull", int_aarch64_neon_umull>;
3601 defm USUBL : SIMDLongThreeVectorBHS<1, 0b0010, "usubl",
3602 BinOpFrag<(sub (zext node:$LHS), (zext node:$RHS))>>;
3603 defm USUBW : SIMDWideThreeVectorBHS< 1, 0b0011, "usubw",
3604 BinOpFrag<(sub node:$LHS, (zext node:$RHS))>>;
3606 // Additional patterns for SMULL and UMULL
3607 multiclass Neon_mul_widen_patterns<SDPatternOperator opnode,
3608 Instruction INST8B, Instruction INST4H, Instruction INST2S> {
3609 def : Pat<(v8i16 (opnode (v8i8 V64:$Rn), (v8i8 V64:$Rm))),
3610 (INST8B V64:$Rn, V64:$Rm)>;
3611 def : Pat<(v4i32 (opnode (v4i16 V64:$Rn), (v4i16 V64:$Rm))),
3612 (INST4H V64:$Rn, V64:$Rm)>;
3613 def : Pat<(v2i64 (opnode (v2i32 V64:$Rn), (v2i32 V64:$Rm))),
3614 (INST2S V64:$Rn, V64:$Rm)>;
3617 defm : Neon_mul_widen_patterns<AArch64smull, SMULLv8i8_v8i16,
3618 SMULLv4i16_v4i32, SMULLv2i32_v2i64>;
3619 defm : Neon_mul_widen_patterns<AArch64umull, UMULLv8i8_v8i16,
3620 UMULLv4i16_v4i32, UMULLv2i32_v2i64>;
3622 // Additional patterns for SMLAL/SMLSL and UMLAL/UMLSL
3623 multiclass Neon_mulacc_widen_patterns<SDPatternOperator opnode,
3624 Instruction INST8B, Instruction INST4H, Instruction INST2S> {
3625 def : Pat<(v8i16 (opnode (v8i16 V128:$Rd), (v8i8 V64:$Rn), (v8i8 V64:$Rm))),
3626 (INST8B V128:$Rd, V64:$Rn, V64:$Rm)>;
3627 def : Pat<(v4i32 (opnode (v4i32 V128:$Rd), (v4i16 V64:$Rn), (v4i16 V64:$Rm))),
3628 (INST4H V128:$Rd, V64:$Rn, V64:$Rm)>;
3629 def : Pat<(v2i64 (opnode (v2i64 V128:$Rd), (v2i32 V64:$Rn), (v2i32 V64:$Rm))),
3630 (INST2S V128:$Rd, V64:$Rn, V64:$Rm)>;
3633 defm : Neon_mulacc_widen_patterns<
3634 TriOpFrag<(add node:$LHS, (AArch64smull node:$MHS, node:$RHS))>,
3635 SMLALv8i8_v8i16, SMLALv4i16_v4i32, SMLALv2i32_v2i64>;
3636 defm : Neon_mulacc_widen_patterns<
3637 TriOpFrag<(add node:$LHS, (AArch64umull node:$MHS, node:$RHS))>,
3638 UMLALv8i8_v8i16, UMLALv4i16_v4i32, UMLALv2i32_v2i64>;
3639 defm : Neon_mulacc_widen_patterns<
3640 TriOpFrag<(sub node:$LHS, (AArch64smull node:$MHS, node:$RHS))>,
3641 SMLSLv8i8_v8i16, SMLSLv4i16_v4i32, SMLSLv2i32_v2i64>;
3642 defm : Neon_mulacc_widen_patterns<
3643 TriOpFrag<(sub node:$LHS, (AArch64umull node:$MHS, node:$RHS))>,
3644 UMLSLv8i8_v8i16, UMLSLv4i16_v4i32, UMLSLv2i32_v2i64>;
3646 // Patterns for 64-bit pmull
3647 def : Pat<(int_aarch64_neon_pmull64 V64:$Rn, V64:$Rm),
3648 (PMULLv1i64 V64:$Rn, V64:$Rm)>;
3649 def : Pat<(int_aarch64_neon_pmull64 (extractelt (v2i64 V128:$Rn), (i64 1)),
3650 (extractelt (v2i64 V128:$Rm), (i64 1))),
3651 (PMULLv2i64 V128:$Rn, V128:$Rm)>;
3653 // CodeGen patterns for addhn and subhn instructions, which can actually be
3654 // written in LLVM IR without too much difficulty.
3657 def : Pat<(v8i8 (trunc (v8i16 (AArch64vlshr (add V128:$Rn, V128:$Rm), (i32 8))))),
3658 (ADDHNv8i16_v8i8 V128:$Rn, V128:$Rm)>;
3659 def : Pat<(v4i16 (trunc (v4i32 (AArch64vlshr (add V128:$Rn, V128:$Rm),
3661 (ADDHNv4i32_v4i16 V128:$Rn, V128:$Rm)>;
3662 def : Pat<(v2i32 (trunc (v2i64 (AArch64vlshr (add V128:$Rn, V128:$Rm),
3664 (ADDHNv2i64_v2i32 V128:$Rn, V128:$Rm)>;
3665 def : Pat<(concat_vectors (v8i8 V64:$Rd),
3666 (trunc (v8i16 (AArch64vlshr (add V128:$Rn, V128:$Rm),
3668 (ADDHNv8i16_v16i8 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3669 V128:$Rn, V128:$Rm)>;
3670 def : Pat<(concat_vectors (v4i16 V64:$Rd),
3671 (trunc (v4i32 (AArch64vlshr (add V128:$Rn, V128:$Rm),
3673 (ADDHNv4i32_v8i16 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3674 V128:$Rn, V128:$Rm)>;
3675 def : Pat<(concat_vectors (v2i32 V64:$Rd),
3676 (trunc (v2i64 (AArch64vlshr (add V128:$Rn, V128:$Rm),
3678 (ADDHNv2i64_v4i32 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3679 V128:$Rn, V128:$Rm)>;
3682 def : Pat<(v8i8 (trunc (v8i16 (AArch64vlshr (sub V128:$Rn, V128:$Rm), (i32 8))))),
3683 (SUBHNv8i16_v8i8 V128:$Rn, V128:$Rm)>;
3684 def : Pat<(v4i16 (trunc (v4i32 (AArch64vlshr (sub V128:$Rn, V128:$Rm),
3686 (SUBHNv4i32_v4i16 V128:$Rn, V128:$Rm)>;
3687 def : Pat<(v2i32 (trunc (v2i64 (AArch64vlshr (sub V128:$Rn, V128:$Rm),
3689 (SUBHNv2i64_v2i32 V128:$Rn, V128:$Rm)>;
3690 def : Pat<(concat_vectors (v8i8 V64:$Rd),
3691 (trunc (v8i16 (AArch64vlshr (sub V128:$Rn, V128:$Rm),
3693 (SUBHNv8i16_v16i8 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3694 V128:$Rn, V128:$Rm)>;
3695 def : Pat<(concat_vectors (v4i16 V64:$Rd),
3696 (trunc (v4i32 (AArch64vlshr (sub V128:$Rn, V128:$Rm),
3698 (SUBHNv4i32_v8i16 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3699 V128:$Rn, V128:$Rm)>;
3700 def : Pat<(concat_vectors (v2i32 V64:$Rd),
3701 (trunc (v2i64 (AArch64vlshr (sub V128:$Rn, V128:$Rm),
3703 (SUBHNv2i64_v4i32 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3704 V128:$Rn, V128:$Rm)>;
3706 //----------------------------------------------------------------------------
3707 // AdvSIMD bitwise extract from vector instruction.
3708 //----------------------------------------------------------------------------
3710 defm EXT : SIMDBitwiseExtract<"ext">;
3712 def : Pat<(v4i16 (AArch64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
3713 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
3714 def : Pat<(v8i16 (AArch64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3715 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3716 def : Pat<(v2i32 (AArch64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
3717 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
3718 def : Pat<(v2f32 (AArch64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
3719 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
3720 def : Pat<(v4i32 (AArch64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3721 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3722 def : Pat<(v4f32 (AArch64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3723 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3724 def : Pat<(v2i64 (AArch64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3725 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3726 def : Pat<(v2f64 (AArch64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3727 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3728 def : Pat<(v4f16 (AArch64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
3729 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
3730 def : Pat<(v8f16 (AArch64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3731 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3733 // We use EXT to handle extract_subvector to copy the upper 64-bits of a
3735 def : Pat<(v8i8 (extract_subvector V128:$Rn, (i64 8))),
3736 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3737 def : Pat<(v4i16 (extract_subvector V128:$Rn, (i64 4))),
3738 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3739 def : Pat<(v2i32 (extract_subvector V128:$Rn, (i64 2))),
3740 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3741 def : Pat<(v1i64 (extract_subvector V128:$Rn, (i64 1))),
3742 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3743 def : Pat<(v4f16 (extract_subvector V128:$Rn, (i64 4))),
3744 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3745 def : Pat<(v2f32 (extract_subvector V128:$Rn, (i64 2))),
3746 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3747 def : Pat<(v1f64 (extract_subvector V128:$Rn, (i64 1))),
3748 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3751 //----------------------------------------------------------------------------
3752 // AdvSIMD zip vector
3753 //----------------------------------------------------------------------------
3755 defm TRN1 : SIMDZipVector<0b010, "trn1", AArch64trn1>;
3756 defm TRN2 : SIMDZipVector<0b110, "trn2", AArch64trn2>;
3757 defm UZP1 : SIMDZipVector<0b001, "uzp1", AArch64uzp1>;
3758 defm UZP2 : SIMDZipVector<0b101, "uzp2", AArch64uzp2>;
3759 defm ZIP1 : SIMDZipVector<0b011, "zip1", AArch64zip1>;
3760 defm ZIP2 : SIMDZipVector<0b111, "zip2", AArch64zip2>;
3762 //----------------------------------------------------------------------------
3763 // AdvSIMD TBL/TBX instructions
3764 //----------------------------------------------------------------------------
3766 defm TBL : SIMDTableLookup< 0, "tbl">;
3767 defm TBX : SIMDTableLookupTied<1, "tbx">;
3769 def : Pat<(v8i8 (int_aarch64_neon_tbl1 (v16i8 VecListOne128:$Rn), (v8i8 V64:$Ri))),
3770 (TBLv8i8One VecListOne128:$Rn, V64:$Ri)>;
3771 def : Pat<(v16i8 (int_aarch64_neon_tbl1 (v16i8 V128:$Ri), (v16i8 V128:$Rn))),
3772 (TBLv16i8One V128:$Ri, V128:$Rn)>;
3774 def : Pat<(v8i8 (int_aarch64_neon_tbx1 (v8i8 V64:$Rd),
3775 (v16i8 VecListOne128:$Rn), (v8i8 V64:$Ri))),
3776 (TBXv8i8One V64:$Rd, VecListOne128:$Rn, V64:$Ri)>;
3777 def : Pat<(v16i8 (int_aarch64_neon_tbx1 (v16i8 V128:$Rd),
3778 (v16i8 V128:$Ri), (v16i8 V128:$Rn))),
3779 (TBXv16i8One V128:$Rd, V128:$Ri, V128:$Rn)>;
3782 //----------------------------------------------------------------------------
3783 // AdvSIMD scalar CPY instruction
3784 //----------------------------------------------------------------------------
3786 defm CPY : SIMDScalarCPY<"cpy">;
3788 //----------------------------------------------------------------------------
3789 // AdvSIMD scalar pairwise instructions
3790 //----------------------------------------------------------------------------
3792 defm ADDP : SIMDPairwiseScalarD<0, 0b11011, "addp">;
3793 defm FADDP : SIMDFPPairwiseScalar<0, 0b01101, "faddp">;
3794 defm FMAXNMP : SIMDFPPairwiseScalar<0, 0b01100, "fmaxnmp">;
3795 defm FMAXP : SIMDFPPairwiseScalar<0, 0b01111, "fmaxp">;
3796 defm FMINNMP : SIMDFPPairwiseScalar<1, 0b01100, "fminnmp">;
3797 defm FMINP : SIMDFPPairwiseScalar<1, 0b01111, "fminp">;
3798 def : Pat<(v2i64 (AArch64saddv V128:$Rn)),
3799 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), (ADDPv2i64p V128:$Rn), dsub)>;
3800 def : Pat<(v2i64 (AArch64uaddv V128:$Rn)),
3801 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), (ADDPv2i64p V128:$Rn), dsub)>;
3802 def : Pat<(f32 (int_aarch64_neon_faddv (v2f32 V64:$Rn))),
3803 (FADDPv2i32p V64:$Rn)>;
3804 def : Pat<(f32 (int_aarch64_neon_faddv (v4f32 V128:$Rn))),
3805 (FADDPv2i32p (EXTRACT_SUBREG (FADDPv4f32 V128:$Rn, V128:$Rn), dsub))>;
3806 def : Pat<(f64 (int_aarch64_neon_faddv (v2f64 V128:$Rn))),
3807 (FADDPv2i64p V128:$Rn)>;
3808 def : Pat<(f32 (int_aarch64_neon_fmaxnmv (v2f32 V64:$Rn))),
3809 (FMAXNMPv2i32p V64:$Rn)>;
3810 def : Pat<(f64 (int_aarch64_neon_fmaxnmv (v2f64 V128:$Rn))),
3811 (FMAXNMPv2i64p V128:$Rn)>;
3812 def : Pat<(f32 (int_aarch64_neon_fmaxv (v2f32 V64:$Rn))),
3813 (FMAXPv2i32p V64:$Rn)>;
3814 def : Pat<(f64 (int_aarch64_neon_fmaxv (v2f64 V128:$Rn))),
3815 (FMAXPv2i64p V128:$Rn)>;
3816 def : Pat<(f32 (int_aarch64_neon_fminnmv (v2f32 V64:$Rn))),
3817 (FMINNMPv2i32p V64:$Rn)>;
3818 def : Pat<(f64 (int_aarch64_neon_fminnmv (v2f64 V128:$Rn))),
3819 (FMINNMPv2i64p V128:$Rn)>;
3820 def : Pat<(f32 (int_aarch64_neon_fminv (v2f32 V64:$Rn))),
3821 (FMINPv2i32p V64:$Rn)>;
3822 def : Pat<(f64 (int_aarch64_neon_fminv (v2f64 V128:$Rn))),
3823 (FMINPv2i64p V128:$Rn)>;
3825 //----------------------------------------------------------------------------
3826 // AdvSIMD INS/DUP instructions
3827 //----------------------------------------------------------------------------
3829 def DUPv8i8gpr : SIMDDupFromMain<0, {?,?,?,?,1}, ".8b", v8i8, V64, GPR32>;
3830 def DUPv16i8gpr : SIMDDupFromMain<1, {?,?,?,?,1}, ".16b", v16i8, V128, GPR32>;
3831 def DUPv4i16gpr : SIMDDupFromMain<0, {?,?,?,1,0}, ".4h", v4i16, V64, GPR32>;
3832 def DUPv8i16gpr : SIMDDupFromMain<1, {?,?,?,1,0}, ".8h", v8i16, V128, GPR32>;
3833 def DUPv2i32gpr : SIMDDupFromMain<0, {?,?,1,0,0}, ".2s", v2i32, V64, GPR32>;
3834 def DUPv4i32gpr : SIMDDupFromMain<1, {?,?,1,0,0}, ".4s", v4i32, V128, GPR32>;
3835 def DUPv2i64gpr : SIMDDupFromMain<1, {?,1,0,0,0}, ".2d", v2i64, V128, GPR64>;
3837 def DUPv2i64lane : SIMDDup64FromElement;
3838 def DUPv2i32lane : SIMDDup32FromElement<0, ".2s", v2i32, V64>;
3839 def DUPv4i32lane : SIMDDup32FromElement<1, ".4s", v4i32, V128>;
3840 def DUPv4i16lane : SIMDDup16FromElement<0, ".4h", v4i16, V64>;
3841 def DUPv8i16lane : SIMDDup16FromElement<1, ".8h", v8i16, V128>;
3842 def DUPv8i8lane : SIMDDup8FromElement <0, ".8b", v8i8, V64>;
3843 def DUPv16i8lane : SIMDDup8FromElement <1, ".16b", v16i8, V128>;
3845 def : Pat<(v2f32 (AArch64dup (f32 FPR32:$Rn))),
3846 (v2f32 (DUPv2i32lane
3847 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rn, ssub),
3849 def : Pat<(v4f32 (AArch64dup (f32 FPR32:$Rn))),
3850 (v4f32 (DUPv4i32lane
3851 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rn, ssub),
3853 def : Pat<(v2f64 (AArch64dup (f64 FPR64:$Rn))),
3854 (v2f64 (DUPv2i64lane
3855 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$Rn, dsub),
3857 def : Pat<(v4f16 (AArch64dup (f16 FPR16:$Rn))),
3858 (v4f16 (DUPv4i16lane
3859 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), FPR16:$Rn, hsub),
3861 def : Pat<(v8f16 (AArch64dup (f16 FPR16:$Rn))),
3862 (v8f16 (DUPv8i16lane
3863 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), FPR16:$Rn, hsub),
3866 def : Pat<(v4f16 (AArch64duplane16 (v8f16 V128:$Rn), VectorIndexH:$imm)),
3867 (DUPv4i16lane V128:$Rn, VectorIndexH:$imm)>;
3868 def : Pat<(v8f16 (AArch64duplane16 (v8f16 V128:$Rn), VectorIndexH:$imm)),
3869 (DUPv8i16lane V128:$Rn, VectorIndexH:$imm)>;
3871 def : Pat<(v2f32 (AArch64duplane32 (v4f32 V128:$Rn), VectorIndexS:$imm)),
3872 (DUPv2i32lane V128:$Rn, VectorIndexS:$imm)>;
3873 def : Pat<(v4f32 (AArch64duplane32 (v4f32 V128:$Rn), VectorIndexS:$imm)),
3874 (DUPv4i32lane V128:$Rn, VectorIndexS:$imm)>;
3875 def : Pat<(v2f64 (AArch64duplane64 (v2f64 V128:$Rn), VectorIndexD:$imm)),
3876 (DUPv2i64lane V128:$Rn, VectorIndexD:$imm)>;
3878 // If there's an (AArch64dup (vector_extract ...) ...), we can use a duplane
3879 // instruction even if the types don't match: we just have to remap the lane
3880 // carefully. N.b. this trick only applies to truncations.
3881 def VecIndex_x2 : SDNodeXForm<imm, [{
3882 return CurDAG->getTargetConstant(2 * N->getZExtValue(), SDLoc(N), MVT::i64);
3884 def VecIndex_x4 : SDNodeXForm<imm, [{
3885 return CurDAG->getTargetConstant(4 * N->getZExtValue(), SDLoc(N), MVT::i64);
3887 def VecIndex_x8 : SDNodeXForm<imm, [{
3888 return CurDAG->getTargetConstant(8 * N->getZExtValue(), SDLoc(N), MVT::i64);
3891 multiclass DUPWithTruncPats<ValueType ResVT, ValueType Src64VT,
3892 ValueType Src128VT, ValueType ScalVT,
3893 Instruction DUP, SDNodeXForm IdxXFORM> {
3894 def : Pat<(ResVT (AArch64dup (ScalVT (vector_extract (Src128VT V128:$Rn),
3896 (DUP V128:$Rn, (IdxXFORM imm:$idx))>;
3898 def : Pat<(ResVT (AArch64dup (ScalVT (vector_extract (Src64VT V64:$Rn),
3900 (DUP (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), (IdxXFORM imm:$idx))>;
3903 defm : DUPWithTruncPats<v8i8, v4i16, v8i16, i32, DUPv8i8lane, VecIndex_x2>;
3904 defm : DUPWithTruncPats<v8i8, v2i32, v4i32, i32, DUPv8i8lane, VecIndex_x4>;
3905 defm : DUPWithTruncPats<v4i16, v2i32, v4i32, i32, DUPv4i16lane, VecIndex_x2>;
3907 defm : DUPWithTruncPats<v16i8, v4i16, v8i16, i32, DUPv16i8lane, VecIndex_x2>;
3908 defm : DUPWithTruncPats<v16i8, v2i32, v4i32, i32, DUPv16i8lane, VecIndex_x4>;
3909 defm : DUPWithTruncPats<v8i16, v2i32, v4i32, i32, DUPv8i16lane, VecIndex_x2>;
3911 multiclass DUPWithTrunci64Pats<ValueType ResVT, Instruction DUP,
3912 SDNodeXForm IdxXFORM> {
3913 def : Pat<(ResVT (AArch64dup (i32 (trunc (extractelt (v2i64 V128:$Rn),
3915 (DUP V128:$Rn, (IdxXFORM imm:$idx))>;
3917 def : Pat<(ResVT (AArch64dup (i32 (trunc (extractelt (v1i64 V64:$Rn),
3919 (DUP (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), (IdxXFORM imm:$idx))>;
3922 defm : DUPWithTrunci64Pats<v8i8, DUPv8i8lane, VecIndex_x8>;
3923 defm : DUPWithTrunci64Pats<v4i16, DUPv4i16lane, VecIndex_x4>;
3924 defm : DUPWithTrunci64Pats<v2i32, DUPv2i32lane, VecIndex_x2>;
3926 defm : DUPWithTrunci64Pats<v16i8, DUPv16i8lane, VecIndex_x8>;
3927 defm : DUPWithTrunci64Pats<v8i16, DUPv8i16lane, VecIndex_x4>;
3928 defm : DUPWithTrunci64Pats<v4i32, DUPv4i32lane, VecIndex_x2>;
3930 // SMOV and UMOV definitions, with some extra patterns for convenience
3934 def : Pat<(sext_inreg (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx), i8),
3935 (i32 (SMOVvi8to32 V128:$Rn, VectorIndexB:$idx))>;
3936 def : Pat<(sext_inreg (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx), i8),
3937 (i64 (SMOVvi8to64 V128:$Rn, VectorIndexB:$idx))>;
3938 def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
3939 (i32 (SMOVvi16to32 V128:$Rn, VectorIndexH:$idx))>;
3940 def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
3941 (i64 (SMOVvi16to64 V128:$Rn, VectorIndexH:$idx))>;
3942 def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
3943 (i32 (SMOVvi16to32 V128:$Rn, VectorIndexH:$idx))>;
3944 def : Pat<(sext (i32 (vector_extract (v4i32 V128:$Rn), VectorIndexS:$idx))),
3945 (i64 (SMOVvi32to64 V128:$Rn, VectorIndexS:$idx))>;
3947 def : Pat<(sext_inreg (i64 (anyext (i32 (vector_extract (v16i8 V128:$Rn),
3948 VectorIndexB:$idx)))), i8),
3949 (i64 (SMOVvi8to64 V128:$Rn, VectorIndexB:$idx))>;
3950 def : Pat<(sext_inreg (i64 (anyext (i32 (vector_extract (v8i16 V128:$Rn),
3951 VectorIndexH:$idx)))), i16),
3952 (i64 (SMOVvi16to64 V128:$Rn, VectorIndexH:$idx))>;
3954 // Extracting i8 or i16 elements will have the zero-extend transformed to
3955 // an 'and' mask by type legalization since neither i8 nor i16 are legal types
3956 // for AArch64. Match these patterns here since UMOV already zeroes out the high
3957 // bits of the destination register.
3958 def : Pat<(and (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx),
3960 (i32 (UMOVvi8 V128:$Rn, VectorIndexB:$idx))>;
3961 def : Pat<(and (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),
3963 (i32 (UMOVvi16 V128:$Rn, VectorIndexH:$idx))>;
3967 def : Pat<(v16i8 (scalar_to_vector GPR32:$Rn)),
3968 (SUBREG_TO_REG (i32 0),
3969 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3970 def : Pat<(v8i8 (scalar_to_vector GPR32:$Rn)),
3971 (SUBREG_TO_REG (i32 0),
3972 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3974 def : Pat<(v8i16 (scalar_to_vector GPR32:$Rn)),
3975 (SUBREG_TO_REG (i32 0),
3976 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3977 def : Pat<(v4i16 (scalar_to_vector GPR32:$Rn)),
3978 (SUBREG_TO_REG (i32 0),
3979 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3981 def : Pat<(v2i32 (scalar_to_vector (i32 FPR32:$Rn))),
3982 (v2i32 (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)),
3983 (i32 FPR32:$Rn), ssub))>;
3984 def : Pat<(v4i32 (scalar_to_vector (i32 FPR32:$Rn))),
3985 (v4i32 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
3986 (i32 FPR32:$Rn), ssub))>;
3987 def : Pat<(v2i64 (scalar_to_vector (i64 FPR64:$Rn))),
3988 (v2i64 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
3989 (i64 FPR64:$Rn), dsub))>;
3991 def : Pat<(v4f16 (scalar_to_vector (f16 FPR16:$Rn))),
3992 (INSERT_SUBREG (v4f16 (IMPLICIT_DEF)), FPR16:$Rn, hsub)>;
3993 def : Pat<(v8f16 (scalar_to_vector (f16 FPR16:$Rn))),
3994 (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), FPR16:$Rn, hsub)>;
3996 def : Pat<(v4f32 (scalar_to_vector (f32 FPR32:$Rn))),
3997 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rn, ssub)>;
3998 def : Pat<(v2f32 (scalar_to_vector (f32 FPR32:$Rn))),
3999 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), FPR32:$Rn, ssub)>;
4000 def : Pat<(v2f64 (scalar_to_vector (f64 FPR64:$Rn))),
4001 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$Rn, dsub)>;
4003 def : Pat<(v4f16 (vector_insert (v4f16 V64:$Rn),
4004 (f16 FPR16:$Rm), (i64 VectorIndexS:$imm))),
4007 (v8f16 (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), V64:$Rn, dsub)),
4009 (v8f16 (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), FPR16:$Rm, hsub)),
4013 def : Pat<(v8f16 (vector_insert (v8f16 V128:$Rn),
4014 (f16 FPR16:$Rm), (i64 VectorIndexH:$imm))),
4016 V128:$Rn, VectorIndexH:$imm,
4017 (v8f16 (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), FPR16:$Rm, hsub)),
4020 def : Pat<(v2f32 (vector_insert (v2f32 V64:$Rn),
4021 (f32 FPR32:$Rm), (i64 VectorIndexS:$imm))),
4024 (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), V64:$Rn, dsub)),
4026 (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rm, ssub)),
4029 def : Pat<(v4f32 (vector_insert (v4f32 V128:$Rn),
4030 (f32 FPR32:$Rm), (i64 VectorIndexS:$imm))),
4032 V128:$Rn, VectorIndexS:$imm,
4033 (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rm, ssub)),
4035 def : Pat<(v2f64 (vector_insert (v2f64 V128:$Rn),
4036 (f64 FPR64:$Rm), (i64 VectorIndexD:$imm))),
4038 V128:$Rn, VectorIndexD:$imm,
4039 (v2f64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$Rm, dsub)),
4042 // Copy an element at a constant index in one vector into a constant indexed
4043 // element of another.
4044 // FIXME refactor to a shared class/dev parameterized on vector type, vector
4045 // index type and INS extension
4046 def : Pat<(v16i8 (int_aarch64_neon_vcopy_lane
4047 (v16i8 V128:$Vd), VectorIndexB:$idx, (v16i8 V128:$Vs),
4048 VectorIndexB:$idx2)),
4050 V128:$Vd, VectorIndexB:$idx, V128:$Vs, VectorIndexB:$idx2)
4052 def : Pat<(v8i16 (int_aarch64_neon_vcopy_lane
4053 (v8i16 V128:$Vd), VectorIndexH:$idx, (v8i16 V128:$Vs),
4054 VectorIndexH:$idx2)),
4056 V128:$Vd, VectorIndexH:$idx, V128:$Vs, VectorIndexH:$idx2)
4058 def : Pat<(v4i32 (int_aarch64_neon_vcopy_lane
4059 (v4i32 V128:$Vd), VectorIndexS:$idx, (v4i32 V128:$Vs),
4060 VectorIndexS:$idx2)),
4062 V128:$Vd, VectorIndexS:$idx, V128:$Vs, VectorIndexS:$idx2)
4064 def : Pat<(v2i64 (int_aarch64_neon_vcopy_lane
4065 (v2i64 V128:$Vd), VectorIndexD:$idx, (v2i64 V128:$Vs),
4066 VectorIndexD:$idx2)),
4068 V128:$Vd, VectorIndexD:$idx, V128:$Vs, VectorIndexD:$idx2)
4071 multiclass Neon_INS_elt_pattern<ValueType VT128, ValueType VT64,
4072 ValueType VTScal, Instruction INS> {
4073 def : Pat<(VT128 (vector_insert V128:$src,
4074 (VTScal (vector_extract (VT128 V128:$Rn), imm:$Immn)),
4076 (INS V128:$src, imm:$Immd, V128:$Rn, imm:$Immn)>;
4078 def : Pat<(VT128 (vector_insert V128:$src,
4079 (VTScal (vector_extract (VT64 V64:$Rn), imm:$Immn)),
4081 (INS V128:$src, imm:$Immd,
4082 (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), imm:$Immn)>;
4084 def : Pat<(VT64 (vector_insert V64:$src,
4085 (VTScal (vector_extract (VT128 V128:$Rn), imm:$Immn)),
4087 (EXTRACT_SUBREG (INS (SUBREG_TO_REG (i64 0), V64:$src, dsub),
4088 imm:$Immd, V128:$Rn, imm:$Immn),
4091 def : Pat<(VT64 (vector_insert V64:$src,
4092 (VTScal (vector_extract (VT64 V64:$Rn), imm:$Immn)),
4095 (INS (SUBREG_TO_REG (i64 0), V64:$src, dsub), imm:$Immd,
4096 (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), imm:$Immn),
4100 defm : Neon_INS_elt_pattern<v8f16, v4f16, f16, INSvi16lane>;
4101 defm : Neon_INS_elt_pattern<v4f32, v2f32, f32, INSvi32lane>;
4102 defm : Neon_INS_elt_pattern<v2f64, v1f64, f64, INSvi64lane>;
4105 // Floating point vector extractions are codegen'd as either a sequence of
4106 // subregister extractions, or a MOV (aka CPY here, alias for DUP) if
4107 // the lane number is anything other than zero.
4108 def : Pat<(vector_extract (v2f64 V128:$Rn), 0),
4109 (f64 (EXTRACT_SUBREG V128:$Rn, dsub))>;
4110 def : Pat<(vector_extract (v4f32 V128:$Rn), 0),
4111 (f32 (EXTRACT_SUBREG V128:$Rn, ssub))>;
4112 def : Pat<(vector_extract (v8f16 V128:$Rn), 0),
4113 (f16 (EXTRACT_SUBREG V128:$Rn, hsub))>;
4115 def : Pat<(vector_extract (v2f64 V128:$Rn), VectorIndexD:$idx),
4116 (f64 (CPYi64 V128:$Rn, VectorIndexD:$idx))>;
4117 def : Pat<(vector_extract (v4f32 V128:$Rn), VectorIndexS:$idx),
4118 (f32 (CPYi32 V128:$Rn, VectorIndexS:$idx))>;
4119 def : Pat<(vector_extract (v8f16 V128:$Rn), VectorIndexH:$idx),
4120 (f16 (CPYi16 V128:$Rn, VectorIndexH:$idx))>;
4122 // All concat_vectors operations are canonicalised to act on i64 vectors for
4123 // AArch64. In the general case we need an instruction, which had just as well be
4125 class ConcatPat<ValueType DstTy, ValueType SrcTy>
4126 : Pat<(DstTy (concat_vectors (SrcTy V64:$Rd), V64:$Rn)),
4127 (INSvi64lane (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), 1,
4128 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rn, dsub), 0)>;
4130 def : ConcatPat<v2i64, v1i64>;
4131 def : ConcatPat<v2f64, v1f64>;
4132 def : ConcatPat<v4i32, v2i32>;
4133 def : ConcatPat<v4f32, v2f32>;
4134 def : ConcatPat<v8i16, v4i16>;
4135 def : ConcatPat<v8f16, v4f16>;
4136 def : ConcatPat<v16i8, v8i8>;
4138 // If the high lanes are undef, though, we can just ignore them:
4139 class ConcatUndefPat<ValueType DstTy, ValueType SrcTy>
4140 : Pat<(DstTy (concat_vectors (SrcTy V64:$Rn), undef)),
4141 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rn, dsub)>;
4143 def : ConcatUndefPat<v2i64, v1i64>;
4144 def : ConcatUndefPat<v2f64, v1f64>;
4145 def : ConcatUndefPat<v4i32, v2i32>;
4146 def : ConcatUndefPat<v4f32, v2f32>;
4147 def : ConcatUndefPat<v8i16, v4i16>;
4148 def : ConcatUndefPat<v16i8, v8i8>;
4150 //----------------------------------------------------------------------------
4151 // AdvSIMD across lanes instructions
4152 //----------------------------------------------------------------------------
4154 defm ADDV : SIMDAcrossLanesBHS<0, 0b11011, "addv">;
4155 defm SMAXV : SIMDAcrossLanesBHS<0, 0b01010, "smaxv">;
4156 defm SMINV : SIMDAcrossLanesBHS<0, 0b11010, "sminv">;
4157 defm UMAXV : SIMDAcrossLanesBHS<1, 0b01010, "umaxv">;
4158 defm UMINV : SIMDAcrossLanesBHS<1, 0b11010, "uminv">;
4159 defm SADDLV : SIMDAcrossLanesHSD<0, 0b00011, "saddlv">;
4160 defm UADDLV : SIMDAcrossLanesHSD<1, 0b00011, "uaddlv">;
4161 defm FMAXNMV : SIMDFPAcrossLanes<0b01100, 0, "fmaxnmv", int_aarch64_neon_fmaxnmv>;
4162 defm FMAXV : SIMDFPAcrossLanes<0b01111, 0, "fmaxv", int_aarch64_neon_fmaxv>;
4163 defm FMINNMV : SIMDFPAcrossLanes<0b01100, 1, "fminnmv", int_aarch64_neon_fminnmv>;
4164 defm FMINV : SIMDFPAcrossLanes<0b01111, 1, "fminv", int_aarch64_neon_fminv>;
4166 // Patterns for across-vector intrinsics, that have a node equivalent, that
4167 // returns a vector (with only the low lane defined) instead of a scalar.
4168 // In effect, opNode is the same as (scalar_to_vector (IntNode)).
4169 multiclass SIMDAcrossLanesIntrinsic<string baseOpc,
4170 SDPatternOperator opNode> {
4171 // If a lane instruction caught the vector_extract around opNode, we can
4172 // directly match the latter to the instruction.
4173 def : Pat<(v8i8 (opNode V64:$Rn)),
4174 (INSERT_SUBREG (v8i8 (IMPLICIT_DEF)),
4175 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub)>;
4176 def : Pat<(v16i8 (opNode V128:$Rn)),
4177 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4178 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub)>;
4179 def : Pat<(v4i16 (opNode V64:$Rn)),
4180 (INSERT_SUBREG (v4i16 (IMPLICIT_DEF)),
4181 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub)>;
4182 def : Pat<(v8i16 (opNode V128:$Rn)),
4183 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
4184 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub)>;
4185 def : Pat<(v4i32 (opNode V128:$Rn)),
4186 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
4187 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), ssub)>;
4190 // If none did, fallback to the explicit patterns, consuming the vector_extract.
4191 def : Pat<(i32 (vector_extract (insert_subvector undef, (v8i8 (opNode V64:$Rn)),
4192 (i32 0)), (i64 0))),
4193 (EXTRACT_SUBREG (INSERT_SUBREG (v8i8 (IMPLICIT_DEF)),
4194 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn),
4196 def : Pat<(i32 (vector_extract (v16i8 (opNode V128:$Rn)), (i64 0))),
4197 (EXTRACT_SUBREG (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4198 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn),
4200 def : Pat<(i32 (vector_extract (insert_subvector undef,
4201 (v4i16 (opNode V64:$Rn)), (i32 0)), (i64 0))),
4202 (EXTRACT_SUBREG (INSERT_SUBREG (v4i16 (IMPLICIT_DEF)),
4203 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn),
4205 def : Pat<(i32 (vector_extract (v8i16 (opNode V128:$Rn)), (i64 0))),
4206 (EXTRACT_SUBREG (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
4207 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn),
4209 def : Pat<(i32 (vector_extract (v4i32 (opNode V128:$Rn)), (i64 0))),
4210 (EXTRACT_SUBREG (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
4211 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn),
4216 multiclass SIMDAcrossLanesSignedIntrinsic<string baseOpc,
4217 SDPatternOperator opNode>
4218 : SIMDAcrossLanesIntrinsic<baseOpc, opNode> {
4219 // If there is a sign extension after this intrinsic, consume it as smov already
4221 def : Pat<(i32 (sext_inreg (i32 (vector_extract (insert_subvector undef,
4222 (opNode (v8i8 V64:$Rn)), (i32 0)), (i64 0))), i8)),
4224 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4225 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
4227 def : Pat<(i32 (sext_inreg (i32 (vector_extract
4228 (opNode (v16i8 V128:$Rn)), (i64 0))), i8)),
4230 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4231 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
4233 def : Pat<(i32 (sext_inreg (i32 (vector_extract (insert_subvector undef,
4234 (opNode (v4i16 V64:$Rn)), (i32 0)), (i64 0))), i16)),
4236 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4237 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
4239 def : Pat<(i32 (sext_inreg (i32 (vector_extract
4240 (opNode (v8i16 V128:$Rn)), (i64 0))), i16)),
4242 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4243 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
4247 multiclass SIMDAcrossLanesUnsignedIntrinsic<string baseOpc,
4248 SDPatternOperator opNode>
4249 : SIMDAcrossLanesIntrinsic<baseOpc, opNode> {
4250 // If there is a masking operation keeping only what has been actually
4251 // generated, consume it.
4252 def : Pat<(i32 (and (i32 (vector_extract (insert_subvector undef,
4253 (opNode (v8i8 V64:$Rn)), (i32 0)), (i64 0))), maski8_or_more)),
4254 (i32 (EXTRACT_SUBREG
4255 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4256 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
4258 def : Pat<(i32 (and (i32 (vector_extract (opNode (v16i8 V128:$Rn)), (i64 0))),
4260 (i32 (EXTRACT_SUBREG
4261 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4262 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
4264 def : Pat<(i32 (and (i32 (vector_extract (insert_subvector undef,
4265 (opNode (v4i16 V64:$Rn)), (i32 0)), (i64 0))), maski16_or_more)),
4266 (i32 (EXTRACT_SUBREG
4267 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4268 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
4270 def : Pat<(i32 (and (i32 (vector_extract (opNode (v8i16 V128:$Rn)), (i64 0))),
4272 (i32 (EXTRACT_SUBREG
4273 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4274 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
4278 defm : SIMDAcrossLanesSignedIntrinsic<"ADDV", AArch64saddv>;
4279 // vaddv_[su]32 is special; -> ADDP Vd.2S,Vn.2S,Vm.2S; return Vd.s[0];Vn==Vm
4280 def : Pat<(v2i32 (AArch64saddv (v2i32 V64:$Rn))),
4281 (ADDPv2i32 V64:$Rn, V64:$Rn)>;
4283 defm : SIMDAcrossLanesUnsignedIntrinsic<"ADDV", AArch64uaddv>;
4284 // vaddv_[su]32 is special; -> ADDP Vd.2S,Vn.2S,Vm.2S; return Vd.s[0];Vn==Vm
4285 def : Pat<(v2i32 (AArch64uaddv (v2i32 V64:$Rn))),
4286 (ADDPv2i32 V64:$Rn, V64:$Rn)>;
4288 defm : SIMDAcrossLanesSignedIntrinsic<"SMAXV", AArch64smaxv>;
4289 def : Pat<(v2i32 (AArch64smaxv (v2i32 V64:$Rn))),
4290 (SMAXPv2i32 V64:$Rn, V64:$Rn)>;
4292 defm : SIMDAcrossLanesSignedIntrinsic<"SMINV", AArch64sminv>;
4293 def : Pat<(v2i32 (AArch64sminv (v2i32 V64:$Rn))),
4294 (SMINPv2i32 V64:$Rn, V64:$Rn)>;
4296 defm : SIMDAcrossLanesUnsignedIntrinsic<"UMAXV", AArch64umaxv>;
4297 def : Pat<(v2i32 (AArch64umaxv (v2i32 V64:$Rn))),
4298 (UMAXPv2i32 V64:$Rn, V64:$Rn)>;
4300 defm : SIMDAcrossLanesUnsignedIntrinsic<"UMINV", AArch64uminv>;
4301 def : Pat<(v2i32 (AArch64uminv (v2i32 V64:$Rn))),
4302 (UMINPv2i32 V64:$Rn, V64:$Rn)>;
4304 multiclass SIMDAcrossLanesSignedLongIntrinsic<string baseOpc, Intrinsic intOp> {
4305 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
4307 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4308 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), hsub),
4310 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
4312 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4313 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), hsub),
4316 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
4317 (i32 (EXTRACT_SUBREG
4318 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4319 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), ssub),
4321 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
4322 (i32 (EXTRACT_SUBREG
4323 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4324 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), ssub),
4327 def : Pat<(i64 (intOp (v4i32 V128:$Rn))),
4328 (i64 (EXTRACT_SUBREG
4329 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4330 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), dsub),
4334 multiclass SIMDAcrossLanesUnsignedLongIntrinsic<string baseOpc,
4336 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
4337 (i32 (EXTRACT_SUBREG
4338 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4339 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), hsub),
4341 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
4342 (i32 (EXTRACT_SUBREG
4343 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4344 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), hsub),
4347 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
4348 (i32 (EXTRACT_SUBREG
4349 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4350 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), ssub),
4352 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
4353 (i32 (EXTRACT_SUBREG
4354 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4355 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), ssub),
4358 def : Pat<(i64 (intOp (v4i32 V128:$Rn))),
4359 (i64 (EXTRACT_SUBREG
4360 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4361 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), dsub),
4365 defm : SIMDAcrossLanesSignedLongIntrinsic<"SADDLV", int_aarch64_neon_saddlv>;
4366 defm : SIMDAcrossLanesUnsignedLongIntrinsic<"UADDLV", int_aarch64_neon_uaddlv>;
4368 // The vaddlv_s32 intrinsic gets mapped to SADDLP.
4369 def : Pat<(i64 (int_aarch64_neon_saddlv (v2i32 V64:$Rn))),
4370 (i64 (EXTRACT_SUBREG
4371 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4372 (SADDLPv2i32_v1i64 V64:$Rn), dsub),
4374 // The vaddlv_u32 intrinsic gets mapped to UADDLP.
4375 def : Pat<(i64 (int_aarch64_neon_uaddlv (v2i32 V64:$Rn))),
4376 (i64 (EXTRACT_SUBREG
4377 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4378 (UADDLPv2i32_v1i64 V64:$Rn), dsub),
4381 //------------------------------------------------------------------------------
4382 // AdvSIMD modified immediate instructions
4383 //------------------------------------------------------------------------------
4386 defm BIC : SIMDModifiedImmVectorShiftTied<1, 0b11, 0b01, "bic", AArch64bici>;
4388 defm ORR : SIMDModifiedImmVectorShiftTied<0, 0b11, 0b01, "orr", AArch64orri>;
4390 def : InstAlias<"bic $Vd.4h, $imm", (BICv4i16 V64:$Vd, imm0_255:$imm, 0)>;
4391 def : InstAlias<"bic $Vd.8h, $imm", (BICv8i16 V128:$Vd, imm0_255:$imm, 0)>;
4392 def : InstAlias<"bic $Vd.2s, $imm", (BICv2i32 V64:$Vd, imm0_255:$imm, 0)>;
4393 def : InstAlias<"bic $Vd.4s, $imm", (BICv4i32 V128:$Vd, imm0_255:$imm, 0)>;
4395 def : InstAlias<"bic.4h $Vd, $imm", (BICv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
4396 def : InstAlias<"bic.8h $Vd, $imm", (BICv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
4397 def : InstAlias<"bic.2s $Vd, $imm", (BICv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
4398 def : InstAlias<"bic.4s $Vd, $imm", (BICv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
4400 def : InstAlias<"orr $Vd.4h, $imm", (ORRv4i16 V64:$Vd, imm0_255:$imm, 0)>;
4401 def : InstAlias<"orr $Vd.8h, $imm", (ORRv8i16 V128:$Vd, imm0_255:$imm, 0)>;
4402 def : InstAlias<"orr $Vd.2s, $imm", (ORRv2i32 V64:$Vd, imm0_255:$imm, 0)>;
4403 def : InstAlias<"orr $Vd.4s, $imm", (ORRv4i32 V128:$Vd, imm0_255:$imm, 0)>;
4405 def : InstAlias<"orr.4h $Vd, $imm", (ORRv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
4406 def : InstAlias<"orr.8h $Vd, $imm", (ORRv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
4407 def : InstAlias<"orr.2s $Vd, $imm", (ORRv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
4408 def : InstAlias<"orr.4s $Vd, $imm", (ORRv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
4411 def FMOVv2f64_ns : SIMDModifiedImmVectorNoShift<1, 1, 0, 0b1111, V128, fpimm8,
4413 [(set (v2f64 V128:$Rd), (AArch64fmov imm0_255:$imm8))]>;
4414 def FMOVv2f32_ns : SIMDModifiedImmVectorNoShift<0, 0, 0, 0b1111, V64, fpimm8,
4416 [(set (v2f32 V64:$Rd), (AArch64fmov imm0_255:$imm8))]>;
4417 def FMOVv4f32_ns : SIMDModifiedImmVectorNoShift<1, 0, 0, 0b1111, V128, fpimm8,
4419 [(set (v4f32 V128:$Rd), (AArch64fmov imm0_255:$imm8))]>;
4420 let Predicates = [HasNEON, HasFullFP16] in {
4421 def FMOVv4f16_ns : SIMDModifiedImmVectorNoShift<0, 0, 1, 0b1111, V64, fpimm8,
4423 [(set (v4f16 V64:$Rd), (AArch64fmov imm0_255:$imm8))]>;
4424 def FMOVv8f16_ns : SIMDModifiedImmVectorNoShift<1, 0, 1, 0b1111, V128, fpimm8,
4426 [(set (v8f16 V128:$Rd), (AArch64fmov imm0_255:$imm8))]>;
4427 } // Predicates = [HasNEON, HasFullFP16]
4431 // EDIT byte mask: scalar
4432 let isReMaterializable = 1, isAsCheapAsAMove = 1 in
4433 def MOVID : SIMDModifiedImmScalarNoShift<0, 1, 0b1110, "movi",
4434 [(set FPR64:$Rd, simdimmtype10:$imm8)]>;
4435 // The movi_edit node has the immediate value already encoded, so we use
4436 // a plain imm0_255 here.
4437 def : Pat<(f64 (AArch64movi_edit imm0_255:$shift)),
4438 (MOVID imm0_255:$shift)>;
4440 def : Pat<(v1i64 immAllZerosV), (MOVID (i32 0))>;
4441 def : Pat<(v2i32 immAllZerosV), (MOVID (i32 0))>;
4442 def : Pat<(v4i16 immAllZerosV), (MOVID (i32 0))>;
4443 def : Pat<(v8i8 immAllZerosV), (MOVID (i32 0))>;
4445 def : Pat<(v1i64 immAllOnesV), (MOVID (i32 255))>;
4446 def : Pat<(v2i32 immAllOnesV), (MOVID (i32 255))>;
4447 def : Pat<(v4i16 immAllOnesV), (MOVID (i32 255))>;
4448 def : Pat<(v8i8 immAllOnesV), (MOVID (i32 255))>;
4450 // EDIT byte mask: 2d
4452 // The movi_edit node has the immediate value already encoded, so we use
4453 // a plain imm0_255 in the pattern
4454 let isReMaterializable = 1, isAsCheapAsAMove = 1 in
4455 def MOVIv2d_ns : SIMDModifiedImmVectorNoShift<1, 1, 0, 0b1110, V128,
4458 [(set (v2i64 V128:$Rd), (AArch64movi_edit imm0_255:$imm8))]>;
4460 def : Pat<(v2i64 immAllZerosV), (MOVIv2d_ns (i32 0))>;
4461 def : Pat<(v4i32 immAllZerosV), (MOVIv2d_ns (i32 0))>;
4462 def : Pat<(v8i16 immAllZerosV), (MOVIv2d_ns (i32 0))>;
4463 def : Pat<(v16i8 immAllZerosV), (MOVIv2d_ns (i32 0))>;
4465 def : Pat<(v2i64 immAllOnesV), (MOVIv2d_ns (i32 255))>;
4466 def : Pat<(v4i32 immAllOnesV), (MOVIv2d_ns (i32 255))>;
4467 def : Pat<(v8i16 immAllOnesV), (MOVIv2d_ns (i32 255))>;
4468 def : Pat<(v16i8 immAllOnesV), (MOVIv2d_ns (i32 255))>;
4470 def : Pat<(v2f64 (AArch64dup (f64 fpimm0))), (MOVIv2d_ns (i32 0))>;
4471 def : Pat<(v4f32 (AArch64dup (f32 fpimm0))), (MOVIv2d_ns (i32 0))>;
4473 // EDIT per word & halfword: 2s, 4h, 4s, & 8h
4474 defm MOVI : SIMDModifiedImmVectorShift<0, 0b10, 0b00, "movi">;
4476 def : InstAlias<"movi $Vd.4h, $imm", (MOVIv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
4477 def : InstAlias<"movi $Vd.8h, $imm", (MOVIv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
4478 def : InstAlias<"movi $Vd.2s, $imm", (MOVIv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
4479 def : InstAlias<"movi $Vd.4s, $imm", (MOVIv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
4481 def : InstAlias<"movi.4h $Vd, $imm", (MOVIv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
4482 def : InstAlias<"movi.8h $Vd, $imm", (MOVIv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
4483 def : InstAlias<"movi.2s $Vd, $imm", (MOVIv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
4484 def : InstAlias<"movi.4s $Vd, $imm", (MOVIv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
4486 def : Pat<(v2i32 (AArch64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
4487 (MOVIv2i32 imm0_255:$imm8, imm:$shift)>;
4488 def : Pat<(v4i32 (AArch64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
4489 (MOVIv4i32 imm0_255:$imm8, imm:$shift)>;
4490 def : Pat<(v4i16 (AArch64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
4491 (MOVIv4i16 imm0_255:$imm8, imm:$shift)>;
4492 def : Pat<(v8i16 (AArch64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
4493 (MOVIv8i16 imm0_255:$imm8, imm:$shift)>;
4495 // EDIT per word: 2s & 4s with MSL shifter
4496 def MOVIv2s_msl : SIMDModifiedImmMoveMSL<0, 0, {1,1,0,?}, V64, "movi", ".2s",
4497 [(set (v2i32 V64:$Rd),
4498 (AArch64movi_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
4499 def MOVIv4s_msl : SIMDModifiedImmMoveMSL<1, 0, {1,1,0,?}, V128, "movi", ".4s",
4500 [(set (v4i32 V128:$Rd),
4501 (AArch64movi_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
4503 // Per byte: 8b & 16b
4504 def MOVIv8b_ns : SIMDModifiedImmVectorNoShift<0, 0, 0, 0b1110, V64, imm0_255,
4506 [(set (v8i8 V64:$Rd), (AArch64movi imm0_255:$imm8))]>;
4507 def MOVIv16b_ns : SIMDModifiedImmVectorNoShift<1, 0, 0, 0b1110, V128, imm0_255,
4509 [(set (v16i8 V128:$Rd), (AArch64movi imm0_255:$imm8))]>;
4513 // EDIT per word & halfword: 2s, 4h, 4s, & 8h
4514 defm MVNI : SIMDModifiedImmVectorShift<1, 0b10, 0b00, "mvni">;
4516 def : InstAlias<"mvni $Vd.4h, $imm", (MVNIv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
4517 def : InstAlias<"mvni $Vd.8h, $imm", (MVNIv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
4518 def : InstAlias<"mvni $Vd.2s, $imm", (MVNIv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
4519 def : InstAlias<"mvni $Vd.4s, $imm", (MVNIv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
4521 def : InstAlias<"mvni.4h $Vd, $imm", (MVNIv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
4522 def : InstAlias<"mvni.8h $Vd, $imm", (MVNIv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
4523 def : InstAlias<"mvni.2s $Vd, $imm", (MVNIv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
4524 def : InstAlias<"mvni.4s $Vd, $imm", (MVNIv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
4526 def : Pat<(v2i32 (AArch64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
4527 (MVNIv2i32 imm0_255:$imm8, imm:$shift)>;
4528 def : Pat<(v4i32 (AArch64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
4529 (MVNIv4i32 imm0_255:$imm8, imm:$shift)>;
4530 def : Pat<(v4i16 (AArch64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
4531 (MVNIv4i16 imm0_255:$imm8, imm:$shift)>;
4532 def : Pat<(v8i16 (AArch64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
4533 (MVNIv8i16 imm0_255:$imm8, imm:$shift)>;
4535 // EDIT per word: 2s & 4s with MSL shifter
4536 def MVNIv2s_msl : SIMDModifiedImmMoveMSL<0, 1, {1,1,0,?}, V64, "mvni", ".2s",
4537 [(set (v2i32 V64:$Rd),
4538 (AArch64mvni_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
4539 def MVNIv4s_msl : SIMDModifiedImmMoveMSL<1, 1, {1,1,0,?}, V128, "mvni", ".4s",
4540 [(set (v4i32 V128:$Rd),
4541 (AArch64mvni_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
4543 //----------------------------------------------------------------------------
4544 // AdvSIMD indexed element
4545 //----------------------------------------------------------------------------
4547 let hasSideEffects = 0 in {
4548 defm FMLA : SIMDFPIndexedTied<0, 0b0001, "fmla">;
4549 defm FMLS : SIMDFPIndexedTied<0, 0b0101, "fmls">;
4552 // NOTE: Operands are reordered in the FMLA/FMLS PatFrags because the
4553 // instruction expects the addend first, while the intrinsic expects it last.
4555 // On the other hand, there are quite a few valid combinatorial options due to
4556 // the commutativity of multiplication and the fact that (-x) * y = x * (-y).
4557 defm : SIMDFPIndexedTiedPatterns<"FMLA",
4558 TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)>>;
4559 defm : SIMDFPIndexedTiedPatterns<"FMLA",
4560 TriOpFrag<(fma node:$MHS, node:$RHS, node:$LHS)>>;
4562 defm : SIMDFPIndexedTiedPatterns<"FMLS",
4563 TriOpFrag<(fma node:$MHS, (fneg node:$RHS), node:$LHS)> >;
4564 defm : SIMDFPIndexedTiedPatterns<"FMLS",
4565 TriOpFrag<(fma node:$RHS, (fneg node:$MHS), node:$LHS)> >;
4566 defm : SIMDFPIndexedTiedPatterns<"FMLS",
4567 TriOpFrag<(fma (fneg node:$RHS), node:$MHS, node:$LHS)> >;
4568 defm : SIMDFPIndexedTiedPatterns<"FMLS",
4569 TriOpFrag<(fma (fneg node:$MHS), node:$RHS, node:$LHS)> >;
4571 multiclass FMLSIndexedAfterNegPatterns<SDPatternOperator OpNode> {
4572 // 3 variants for the .2s version: DUPLANE from 128-bit, DUPLANE from 64-bit
4574 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
4575 (AArch64duplane32 (v4f32 (fneg V128:$Rm)),
4576 VectorIndexS:$idx))),
4577 (FMLSv2i32_indexed V64:$Rd, V64:$Rn, V128:$Rm, VectorIndexS:$idx)>;
4578 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
4579 (v2f32 (AArch64duplane32
4580 (v4f32 (insert_subvector undef,
4581 (v2f32 (fneg V64:$Rm)),
4583 VectorIndexS:$idx)))),
4584 (FMLSv2i32_indexed V64:$Rd, V64:$Rn,
4585 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub),
4586 VectorIndexS:$idx)>;
4587 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
4588 (AArch64dup (f32 (fneg FPR32Op:$Rm))))),
4589 (FMLSv2i32_indexed V64:$Rd, V64:$Rn,
4590 (SUBREG_TO_REG (i32 0), FPR32Op:$Rm, ssub), (i64 0))>;
4592 // 3 variants for the .4s version: DUPLANE from 128-bit, DUPLANE from 64-bit
4594 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
4595 (AArch64duplane32 (v4f32 (fneg V128:$Rm)),
4596 VectorIndexS:$idx))),
4597 (FMLSv4i32_indexed V128:$Rd, V128:$Rn, V128:$Rm,
4598 VectorIndexS:$idx)>;
4599 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
4600 (v4f32 (AArch64duplane32
4601 (v4f32 (insert_subvector undef,
4602 (v2f32 (fneg V64:$Rm)),
4604 VectorIndexS:$idx)))),
4605 (FMLSv4i32_indexed V128:$Rd, V128:$Rn,
4606 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub),
4607 VectorIndexS:$idx)>;
4608 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
4609 (AArch64dup (f32 (fneg FPR32Op:$Rm))))),
4610 (FMLSv4i32_indexed V128:$Rd, V128:$Rn,
4611 (SUBREG_TO_REG (i32 0), FPR32Op:$Rm, ssub), (i64 0))>;
4613 // 2 variants for the .2d version: DUPLANE from 128-bit, and DUP scalar
4614 // (DUPLANE from 64-bit would be trivial).
4615 def : Pat<(v2f64 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn),
4616 (AArch64duplane64 (v2f64 (fneg V128:$Rm)),
4617 VectorIndexD:$idx))),
4619 V128:$Rd, V128:$Rn, V128:$Rm, VectorIndexS:$idx)>;
4620 def : Pat<(v2f64 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn),
4621 (AArch64dup (f64 (fneg FPR64Op:$Rm))))),
4622 (FMLSv2i64_indexed V128:$Rd, V128:$Rn,
4623 (SUBREG_TO_REG (i32 0), FPR64Op:$Rm, dsub), (i64 0))>;
4625 // 2 variants for 32-bit scalar version: extract from .2s or from .4s
4626 def : Pat<(f32 (OpNode (f32 FPR32:$Rd), (f32 FPR32:$Rn),
4627 (vector_extract (v4f32 (fneg V128:$Rm)),
4628 VectorIndexS:$idx))),
4629 (FMLSv1i32_indexed FPR32:$Rd, FPR32:$Rn,
4630 V128:$Rm, VectorIndexS:$idx)>;
4631 def : Pat<(f32 (OpNode (f32 FPR32:$Rd), (f32 FPR32:$Rn),
4632 (vector_extract (v4f32 (insert_subvector undef,
4633 (v2f32 (fneg V64:$Rm)),
4635 VectorIndexS:$idx))),
4636 (FMLSv1i32_indexed FPR32:$Rd, FPR32:$Rn,
4637 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub), VectorIndexS:$idx)>;
4639 // 1 variant for 64-bit scalar version: extract from .1d or from .2d
4640 def : Pat<(f64 (OpNode (f64 FPR64:$Rd), (f64 FPR64:$Rn),
4641 (vector_extract (v2f64 (fneg V128:$Rm)),
4642 VectorIndexS:$idx))),
4643 (FMLSv1i64_indexed FPR64:$Rd, FPR64:$Rn,
4644 V128:$Rm, VectorIndexS:$idx)>;
4647 defm : FMLSIndexedAfterNegPatterns<
4648 TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)> >;
4649 defm : FMLSIndexedAfterNegPatterns<
4650 TriOpFrag<(fma node:$MHS, node:$RHS, node:$LHS)> >;
4652 defm FMULX : SIMDFPIndexed<1, 0b1001, "fmulx", int_aarch64_neon_fmulx>;
4653 defm FMUL : SIMDFPIndexed<0, 0b1001, "fmul", fmul>;
4655 def : Pat<(v2f32 (fmul V64:$Rn, (AArch64dup (f32 FPR32:$Rm)))),
4656 (FMULv2i32_indexed V64:$Rn,
4657 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rm, ssub),
4659 def : Pat<(v4f32 (fmul V128:$Rn, (AArch64dup (f32 FPR32:$Rm)))),
4660 (FMULv4i32_indexed V128:$Rn,
4661 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rm, ssub),
4663 def : Pat<(v2f64 (fmul V128:$Rn, (AArch64dup (f64 FPR64:$Rm)))),
4664 (FMULv2i64_indexed V128:$Rn,
4665 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$Rm, dsub),
4668 defm SQDMULH : SIMDIndexedHS<0, 0b1100, "sqdmulh", int_aarch64_neon_sqdmulh>;
4669 defm SQRDMULH : SIMDIndexedHS<0, 0b1101, "sqrdmulh", int_aarch64_neon_sqrdmulh>;
4670 defm MLA : SIMDVectorIndexedHSTied<1, 0b0000, "mla",
4671 TriOpFrag<(add node:$LHS, (mul node:$MHS, node:$RHS))>>;
4672 defm MLS : SIMDVectorIndexedHSTied<1, 0b0100, "mls",
4673 TriOpFrag<(sub node:$LHS, (mul node:$MHS, node:$RHS))>>;
4674 defm MUL : SIMDVectorIndexedHS<0, 0b1000, "mul", mul>;
4675 defm SMLAL : SIMDVectorIndexedLongSDTied<0, 0b0010, "smlal",
4676 TriOpFrag<(add node:$LHS, (int_aarch64_neon_smull node:$MHS, node:$RHS))>>;
4677 defm SMLSL : SIMDVectorIndexedLongSDTied<0, 0b0110, "smlsl",
4678 TriOpFrag<(sub node:$LHS, (int_aarch64_neon_smull node:$MHS, node:$RHS))>>;
4679 defm SMULL : SIMDVectorIndexedLongSD<0, 0b1010, "smull",
4680 int_aarch64_neon_smull>;
4681 defm SQDMLAL : SIMDIndexedLongSQDMLXSDTied<0, 0b0011, "sqdmlal",
4682 int_aarch64_neon_sqadd>;
4683 defm SQDMLSL : SIMDIndexedLongSQDMLXSDTied<0, 0b0111, "sqdmlsl",
4684 int_aarch64_neon_sqsub>;
4685 defm SQRDMLAH : SIMDIndexedSQRDMLxHSDTied<1, 0b1101, "sqrdmlah",
4686 int_aarch64_neon_sqadd>;
4687 defm SQRDMLSH : SIMDIndexedSQRDMLxHSDTied<1, 0b1111, "sqrdmlsh",
4688 int_aarch64_neon_sqsub>;
4689 defm SQDMULL : SIMDIndexedLongSD<0, 0b1011, "sqdmull", int_aarch64_neon_sqdmull>;
4690 defm UMLAL : SIMDVectorIndexedLongSDTied<1, 0b0010, "umlal",
4691 TriOpFrag<(add node:$LHS, (int_aarch64_neon_umull node:$MHS, node:$RHS))>>;
4692 defm UMLSL : SIMDVectorIndexedLongSDTied<1, 0b0110, "umlsl",
4693 TriOpFrag<(sub node:$LHS, (int_aarch64_neon_umull node:$MHS, node:$RHS))>>;
4694 defm UMULL : SIMDVectorIndexedLongSD<1, 0b1010, "umull",
4695 int_aarch64_neon_umull>;
4697 // A scalar sqdmull with the second operand being a vector lane can be
4698 // handled directly with the indexed instruction encoding.
4699 def : Pat<(int_aarch64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
4700 (vector_extract (v4i32 V128:$Vm),
4701 VectorIndexS:$idx)),
4702 (SQDMULLv1i64_indexed FPR32:$Rn, V128:$Vm, VectorIndexS:$idx)>;
4704 //----------------------------------------------------------------------------
4705 // AdvSIMD scalar shift instructions
4706 //----------------------------------------------------------------------------
4707 defm FCVTZS : SIMDFPScalarRShift<0, 0b11111, "fcvtzs">;
4708 defm FCVTZU : SIMDFPScalarRShift<1, 0b11111, "fcvtzu">;
4709 defm SCVTF : SIMDFPScalarRShift<0, 0b11100, "scvtf">;
4710 defm UCVTF : SIMDFPScalarRShift<1, 0b11100, "ucvtf">;
4711 // Codegen patterns for the above. We don't put these directly on the
4712 // instructions because TableGen's type inference can't handle the truth.
4713 // Having the same base pattern for fp <--> int totally freaks it out.
4714 def : Pat<(int_aarch64_neon_vcvtfp2fxs FPR32:$Rn, vecshiftR32:$imm),
4715 (FCVTZSs FPR32:$Rn, vecshiftR32:$imm)>;
4716 def : Pat<(int_aarch64_neon_vcvtfp2fxu FPR32:$Rn, vecshiftR32:$imm),
4717 (FCVTZUs FPR32:$Rn, vecshiftR32:$imm)>;
4718 def : Pat<(i64 (int_aarch64_neon_vcvtfp2fxs (f64 FPR64:$Rn), vecshiftR64:$imm)),
4719 (FCVTZSd FPR64:$Rn, vecshiftR64:$imm)>;
4720 def : Pat<(i64 (int_aarch64_neon_vcvtfp2fxu (f64 FPR64:$Rn), vecshiftR64:$imm)),
4721 (FCVTZUd FPR64:$Rn, vecshiftR64:$imm)>;
4722 def : Pat<(v1i64 (int_aarch64_neon_vcvtfp2fxs (v1f64 FPR64:$Rn),
4724 (FCVTZSd FPR64:$Rn, vecshiftR64:$imm)>;
4725 def : Pat<(v1i64 (int_aarch64_neon_vcvtfp2fxu (v1f64 FPR64:$Rn),
4727 (FCVTZUd FPR64:$Rn, vecshiftR64:$imm)>;
4728 def : Pat<(int_aarch64_neon_vcvtfxs2fp FPR32:$Rn, vecshiftR32:$imm),
4729 (SCVTFs FPR32:$Rn, vecshiftR32:$imm)>;
4730 def : Pat<(int_aarch64_neon_vcvtfxu2fp FPR32:$Rn, vecshiftR32:$imm),
4731 (UCVTFs FPR32:$Rn, vecshiftR32:$imm)>;
4732 def : Pat<(f64 (int_aarch64_neon_vcvtfxs2fp (i64 FPR64:$Rn), vecshiftR64:$imm)),
4733 (SCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
4734 def : Pat<(f64 (int_aarch64_neon_vcvtfxu2fp (i64 FPR64:$Rn), vecshiftR64:$imm)),
4735 (UCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
4736 def : Pat<(v1f64 (int_aarch64_neon_vcvtfxs2fp (v1i64 FPR64:$Rn),
4738 (SCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
4739 def : Pat<(v1f64 (int_aarch64_neon_vcvtfxu2fp (v1i64 FPR64:$Rn),
4741 (UCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
4743 defm SHL : SIMDScalarLShiftD< 0, 0b01010, "shl", AArch64vshl>;
4744 defm SLI : SIMDScalarLShiftDTied<1, 0b01010, "sli">;
4745 defm SQRSHRN : SIMDScalarRShiftBHS< 0, 0b10011, "sqrshrn",
4746 int_aarch64_neon_sqrshrn>;
4747 defm SQRSHRUN : SIMDScalarRShiftBHS< 1, 0b10001, "sqrshrun",
4748 int_aarch64_neon_sqrshrun>;
4749 defm SQSHLU : SIMDScalarLShiftBHSD<1, 0b01100, "sqshlu", AArch64sqshlui>;
4750 defm SQSHL : SIMDScalarLShiftBHSD<0, 0b01110, "sqshl", AArch64sqshli>;
4751 defm SQSHRN : SIMDScalarRShiftBHS< 0, 0b10010, "sqshrn",
4752 int_aarch64_neon_sqshrn>;
4753 defm SQSHRUN : SIMDScalarRShiftBHS< 1, 0b10000, "sqshrun",
4754 int_aarch64_neon_sqshrun>;
4755 defm SRI : SIMDScalarRShiftDTied< 1, 0b01000, "sri">;
4756 defm SRSHR : SIMDScalarRShiftD< 0, 0b00100, "srshr", AArch64srshri>;
4757 defm SRSRA : SIMDScalarRShiftDTied< 0, 0b00110, "srsra",
4758 TriOpFrag<(add node:$LHS,
4759 (AArch64srshri node:$MHS, node:$RHS))>>;
4760 defm SSHR : SIMDScalarRShiftD< 0, 0b00000, "sshr", AArch64vashr>;
4761 defm SSRA : SIMDScalarRShiftDTied< 0, 0b00010, "ssra",
4762 TriOpFrag<(add node:$LHS,
4763 (AArch64vashr node:$MHS, node:$RHS))>>;
4764 defm UQRSHRN : SIMDScalarRShiftBHS< 1, 0b10011, "uqrshrn",
4765 int_aarch64_neon_uqrshrn>;
4766 defm UQSHL : SIMDScalarLShiftBHSD<1, 0b01110, "uqshl", AArch64uqshli>;
4767 defm UQSHRN : SIMDScalarRShiftBHS< 1, 0b10010, "uqshrn",
4768 int_aarch64_neon_uqshrn>;
4769 defm URSHR : SIMDScalarRShiftD< 1, 0b00100, "urshr", AArch64urshri>;
4770 defm URSRA : SIMDScalarRShiftDTied< 1, 0b00110, "ursra",
4771 TriOpFrag<(add node:$LHS,
4772 (AArch64urshri node:$MHS, node:$RHS))>>;
4773 defm USHR : SIMDScalarRShiftD< 1, 0b00000, "ushr", AArch64vlshr>;
4774 defm USRA : SIMDScalarRShiftDTied< 1, 0b00010, "usra",
4775 TriOpFrag<(add node:$LHS,
4776 (AArch64vlshr node:$MHS, node:$RHS))>>;
4778 //----------------------------------------------------------------------------
4779 // AdvSIMD vector shift instructions
4780 //----------------------------------------------------------------------------
4781 defm FCVTZS:SIMDVectorRShiftSD<0, 0b11111, "fcvtzs", int_aarch64_neon_vcvtfp2fxs>;
4782 defm FCVTZU:SIMDVectorRShiftSD<1, 0b11111, "fcvtzu", int_aarch64_neon_vcvtfp2fxu>;
4783 defm SCVTF: SIMDVectorRShiftToFP<0, 0b11100, "scvtf",
4784 int_aarch64_neon_vcvtfxs2fp>;
4785 defm RSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10001, "rshrn",
4786 int_aarch64_neon_rshrn>;
4787 defm SHL : SIMDVectorLShiftBHSD<0, 0b01010, "shl", AArch64vshl>;
4788 defm SHRN : SIMDVectorRShiftNarrowBHS<0, 0b10000, "shrn",
4789 BinOpFrag<(trunc (AArch64vashr node:$LHS, node:$RHS))>>;
4790 defm SLI : SIMDVectorLShiftBHSDTied<1, 0b01010, "sli", int_aarch64_neon_vsli>;
4791 def : Pat<(v1i64 (int_aarch64_neon_vsli (v1i64 FPR64:$Rd), (v1i64 FPR64:$Rn),
4792 (i32 vecshiftL64:$imm))),
4793 (SLId FPR64:$Rd, FPR64:$Rn, vecshiftL64:$imm)>;
4794 defm SQRSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10011, "sqrshrn",
4795 int_aarch64_neon_sqrshrn>;
4796 defm SQRSHRUN: SIMDVectorRShiftNarrowBHS<1, 0b10001, "sqrshrun",
4797 int_aarch64_neon_sqrshrun>;
4798 defm SQSHLU : SIMDVectorLShiftBHSD<1, 0b01100, "sqshlu", AArch64sqshlui>;
4799 defm SQSHL : SIMDVectorLShiftBHSD<0, 0b01110, "sqshl", AArch64sqshli>;
4800 defm SQSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10010, "sqshrn",
4801 int_aarch64_neon_sqshrn>;
4802 defm SQSHRUN : SIMDVectorRShiftNarrowBHS<1, 0b10000, "sqshrun",
4803 int_aarch64_neon_sqshrun>;
4804 defm SRI : SIMDVectorRShiftBHSDTied<1, 0b01000, "sri", int_aarch64_neon_vsri>;
4805 def : Pat<(v1i64 (int_aarch64_neon_vsri (v1i64 FPR64:$Rd), (v1i64 FPR64:$Rn),
4806 (i32 vecshiftR64:$imm))),
4807 (SRId FPR64:$Rd, FPR64:$Rn, vecshiftR64:$imm)>;
4808 defm SRSHR : SIMDVectorRShiftBHSD<0, 0b00100, "srshr", AArch64srshri>;
4809 defm SRSRA : SIMDVectorRShiftBHSDTied<0, 0b00110, "srsra",
4810 TriOpFrag<(add node:$LHS,
4811 (AArch64srshri node:$MHS, node:$RHS))> >;
4812 defm SSHLL : SIMDVectorLShiftLongBHSD<0, 0b10100, "sshll",
4813 BinOpFrag<(AArch64vshl (sext node:$LHS), node:$RHS)>>;
4815 defm SSHR : SIMDVectorRShiftBHSD<0, 0b00000, "sshr", AArch64vashr>;
4816 defm SSRA : SIMDVectorRShiftBHSDTied<0, 0b00010, "ssra",
4817 TriOpFrag<(add node:$LHS, (AArch64vashr node:$MHS, node:$RHS))>>;
4818 defm UCVTF : SIMDVectorRShiftToFP<1, 0b11100, "ucvtf",
4819 int_aarch64_neon_vcvtfxu2fp>;
4820 defm UQRSHRN : SIMDVectorRShiftNarrowBHS<1, 0b10011, "uqrshrn",
4821 int_aarch64_neon_uqrshrn>;
4822 defm UQSHL : SIMDVectorLShiftBHSD<1, 0b01110, "uqshl", AArch64uqshli>;
4823 defm UQSHRN : SIMDVectorRShiftNarrowBHS<1, 0b10010, "uqshrn",
4824 int_aarch64_neon_uqshrn>;
4825 defm URSHR : SIMDVectorRShiftBHSD<1, 0b00100, "urshr", AArch64urshri>;
4826 defm URSRA : SIMDVectorRShiftBHSDTied<1, 0b00110, "ursra",
4827 TriOpFrag<(add node:$LHS,
4828 (AArch64urshri node:$MHS, node:$RHS))> >;
4829 defm USHLL : SIMDVectorLShiftLongBHSD<1, 0b10100, "ushll",
4830 BinOpFrag<(AArch64vshl (zext node:$LHS), node:$RHS)>>;
4831 defm USHR : SIMDVectorRShiftBHSD<1, 0b00000, "ushr", AArch64vlshr>;
4832 defm USRA : SIMDVectorRShiftBHSDTied<1, 0b00010, "usra",
4833 TriOpFrag<(add node:$LHS, (AArch64vlshr node:$MHS, node:$RHS))> >;
4835 // SHRN patterns for when a logical right shift was used instead of arithmetic
4836 // (the immediate guarantees no sign bits actually end up in the result so it
4838 def : Pat<(v8i8 (trunc (AArch64vlshr (v8i16 V128:$Rn), vecshiftR16Narrow:$imm))),
4839 (SHRNv8i8_shift V128:$Rn, vecshiftR16Narrow:$imm)>;
4840 def : Pat<(v4i16 (trunc (AArch64vlshr (v4i32 V128:$Rn), vecshiftR32Narrow:$imm))),
4841 (SHRNv4i16_shift V128:$Rn, vecshiftR32Narrow:$imm)>;
4842 def : Pat<(v2i32 (trunc (AArch64vlshr (v2i64 V128:$Rn), vecshiftR64Narrow:$imm))),
4843 (SHRNv2i32_shift V128:$Rn, vecshiftR64Narrow:$imm)>;
4845 def : Pat<(v16i8 (concat_vectors (v8i8 V64:$Rd),
4846 (trunc (AArch64vlshr (v8i16 V128:$Rn),
4847 vecshiftR16Narrow:$imm)))),
4848 (SHRNv16i8_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
4849 V128:$Rn, vecshiftR16Narrow:$imm)>;
4850 def : Pat<(v8i16 (concat_vectors (v4i16 V64:$Rd),
4851 (trunc (AArch64vlshr (v4i32 V128:$Rn),
4852 vecshiftR32Narrow:$imm)))),
4853 (SHRNv8i16_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
4854 V128:$Rn, vecshiftR32Narrow:$imm)>;
4855 def : Pat<(v4i32 (concat_vectors (v2i32 V64:$Rd),
4856 (trunc (AArch64vlshr (v2i64 V128:$Rn),
4857 vecshiftR64Narrow:$imm)))),
4858 (SHRNv4i32_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
4859 V128:$Rn, vecshiftR32Narrow:$imm)>;
4861 // Vector sign and zero extensions are implemented with SSHLL and USSHLL.
4862 // Anyexts are implemented as zexts.
4863 def : Pat<(v8i16 (sext (v8i8 V64:$Rn))), (SSHLLv8i8_shift V64:$Rn, (i32 0))>;
4864 def : Pat<(v8i16 (zext (v8i8 V64:$Rn))), (USHLLv8i8_shift V64:$Rn, (i32 0))>;
4865 def : Pat<(v8i16 (anyext (v8i8 V64:$Rn))), (USHLLv8i8_shift V64:$Rn, (i32 0))>;
4866 def : Pat<(v4i32 (sext (v4i16 V64:$Rn))), (SSHLLv4i16_shift V64:$Rn, (i32 0))>;
4867 def : Pat<(v4i32 (zext (v4i16 V64:$Rn))), (USHLLv4i16_shift V64:$Rn, (i32 0))>;
4868 def : Pat<(v4i32 (anyext (v4i16 V64:$Rn))), (USHLLv4i16_shift V64:$Rn, (i32 0))>;
4869 def : Pat<(v2i64 (sext (v2i32 V64:$Rn))), (SSHLLv2i32_shift V64:$Rn, (i32 0))>;
4870 def : Pat<(v2i64 (zext (v2i32 V64:$Rn))), (USHLLv2i32_shift V64:$Rn, (i32 0))>;
4871 def : Pat<(v2i64 (anyext (v2i32 V64:$Rn))), (USHLLv2i32_shift V64:$Rn, (i32 0))>;
4872 // Also match an extend from the upper half of a 128 bit source register.
4873 def : Pat<(v8i16 (anyext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
4874 (USHLLv16i8_shift V128:$Rn, (i32 0))>;
4875 def : Pat<(v8i16 (zext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
4876 (USHLLv16i8_shift V128:$Rn, (i32 0))>;
4877 def : Pat<(v8i16 (sext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
4878 (SSHLLv16i8_shift V128:$Rn, (i32 0))>;
4879 def : Pat<(v4i32 (anyext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
4880 (USHLLv8i16_shift V128:$Rn, (i32 0))>;
4881 def : Pat<(v4i32 (zext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
4882 (USHLLv8i16_shift V128:$Rn, (i32 0))>;
4883 def : Pat<(v4i32 (sext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
4884 (SSHLLv8i16_shift V128:$Rn, (i32 0))>;
4885 def : Pat<(v2i64 (anyext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
4886 (USHLLv4i32_shift V128:$Rn, (i32 0))>;
4887 def : Pat<(v2i64 (zext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
4888 (USHLLv4i32_shift V128:$Rn, (i32 0))>;
4889 def : Pat<(v2i64 (sext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
4890 (SSHLLv4i32_shift V128:$Rn, (i32 0))>;
4892 // Vector shift sxtl aliases
4893 def : InstAlias<"sxtl.8h $dst, $src1",
4894 (SSHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
4895 def : InstAlias<"sxtl $dst.8h, $src1.8b",
4896 (SSHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
4897 def : InstAlias<"sxtl.4s $dst, $src1",
4898 (SSHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
4899 def : InstAlias<"sxtl $dst.4s, $src1.4h",
4900 (SSHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
4901 def : InstAlias<"sxtl.2d $dst, $src1",
4902 (SSHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
4903 def : InstAlias<"sxtl $dst.2d, $src1.2s",
4904 (SSHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
4906 // Vector shift sxtl2 aliases
4907 def : InstAlias<"sxtl2.8h $dst, $src1",
4908 (SSHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
4909 def : InstAlias<"sxtl2 $dst.8h, $src1.16b",
4910 (SSHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
4911 def : InstAlias<"sxtl2.4s $dst, $src1",
4912 (SSHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
4913 def : InstAlias<"sxtl2 $dst.4s, $src1.8h",
4914 (SSHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
4915 def : InstAlias<"sxtl2.2d $dst, $src1",
4916 (SSHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
4917 def : InstAlias<"sxtl2 $dst.2d, $src1.4s",
4918 (SSHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
4920 // Vector shift uxtl aliases
4921 def : InstAlias<"uxtl.8h $dst, $src1",
4922 (USHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
4923 def : InstAlias<"uxtl $dst.8h, $src1.8b",
4924 (USHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
4925 def : InstAlias<"uxtl.4s $dst, $src1",
4926 (USHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
4927 def : InstAlias<"uxtl $dst.4s, $src1.4h",
4928 (USHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
4929 def : InstAlias<"uxtl.2d $dst, $src1",
4930 (USHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
4931 def : InstAlias<"uxtl $dst.2d, $src1.2s",
4932 (USHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
4934 // Vector shift uxtl2 aliases
4935 def : InstAlias<"uxtl2.8h $dst, $src1",
4936 (USHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
4937 def : InstAlias<"uxtl2 $dst.8h, $src1.16b",
4938 (USHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
4939 def : InstAlias<"uxtl2.4s $dst, $src1",
4940 (USHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
4941 def : InstAlias<"uxtl2 $dst.4s, $src1.8h",
4942 (USHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
4943 def : InstAlias<"uxtl2.2d $dst, $src1",
4944 (USHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
4945 def : InstAlias<"uxtl2 $dst.2d, $src1.4s",
4946 (USHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
4948 // If an integer is about to be converted to a floating point value,
4949 // just load it on the floating point unit.
4950 // These patterns are more complex because floating point loads do not
4951 // support sign extension.
4952 // The sign extension has to be explicitly added and is only supported for
4953 // one step: byte-to-half, half-to-word, word-to-doubleword.
4954 // SCVTF GPR -> FPR is 9 cycles.
4955 // SCVTF FPR -> FPR is 4 cyclces.
4956 // (sign extension with lengthen) SXTL FPR -> FPR is 2 cycles.
4957 // Therefore, we can do 2 sign extensions and one SCVTF FPR -> FPR
4958 // and still being faster.
4959 // However, this is not good for code size.
4960 // 8-bits -> float. 2 sizes step-up.
4961 class SExtLoadi8CVTf32Pat<dag addrmode, dag INST>
4962 : Pat<(f32 (sint_to_fp (i32 (sextloadi8 addrmode)))),
4963 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
4968 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4975 Requires<[NotForCodeSize, UseAlternateSExtLoadCVTF32]>;
4977 def : SExtLoadi8CVTf32Pat<(ro8.Wpat GPR64sp:$Rn, GPR32:$Rm, ro8.Wext:$ext),
4978 (LDRBroW GPR64sp:$Rn, GPR32:$Rm, ro8.Wext:$ext)>;
4979 def : SExtLoadi8CVTf32Pat<(ro8.Xpat GPR64sp:$Rn, GPR64:$Rm, ro8.Xext:$ext),
4980 (LDRBroX GPR64sp:$Rn, GPR64:$Rm, ro8.Xext:$ext)>;
4981 def : SExtLoadi8CVTf32Pat<(am_indexed8 GPR64sp:$Rn, uimm12s1:$offset),
4982 (LDRBui GPR64sp:$Rn, uimm12s1:$offset)>;
4983 def : SExtLoadi8CVTf32Pat<(am_unscaled8 GPR64sp:$Rn, simm9:$offset),
4984 (LDURBi GPR64sp:$Rn, simm9:$offset)>;
4986 // 16-bits -> float. 1 size step-up.
4987 class SExtLoadi16CVTf32Pat<dag addrmode, dag INST>
4988 : Pat<(f32 (sint_to_fp (i32 (sextloadi16 addrmode)))),
4989 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
4991 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4995 ssub)))>, Requires<[NotForCodeSize]>;
4997 def : SExtLoadi16CVTf32Pat<(ro16.Wpat GPR64sp:$Rn, GPR32:$Rm, ro16.Wext:$ext),
4998 (LDRHroW GPR64sp:$Rn, GPR32:$Rm, ro16.Wext:$ext)>;
4999 def : SExtLoadi16CVTf32Pat<(ro16.Xpat GPR64sp:$Rn, GPR64:$Rm, ro16.Xext:$ext),
5000 (LDRHroX GPR64sp:$Rn, GPR64:$Rm, ro16.Xext:$ext)>;
5001 def : SExtLoadi16CVTf32Pat<(am_indexed16 GPR64sp:$Rn, uimm12s2:$offset),
5002 (LDRHui GPR64sp:$Rn, uimm12s2:$offset)>;
5003 def : SExtLoadi16CVTf32Pat<(am_unscaled16 GPR64sp:$Rn, simm9:$offset),
5004 (LDURHi GPR64sp:$Rn, simm9:$offset)>;
5006 // 32-bits to 32-bits are handled in target specific dag combine:
5007 // performIntToFpCombine.
5008 // 64-bits integer to 32-bits floating point, not possible with
5009 // SCVTF on floating point registers (both source and destination
5010 // must have the same size).
5012 // Here are the patterns for 8, 16, 32, and 64-bits to double.
5013 // 8-bits -> double. 3 size step-up: give up.
5014 // 16-bits -> double. 2 size step.
5015 class SExtLoadi16CVTf64Pat<dag addrmode, dag INST>
5016 : Pat <(f64 (sint_to_fp (i32 (sextloadi16 addrmode)))),
5017 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
5022 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
5029 Requires<[NotForCodeSize, UseAlternateSExtLoadCVTF32]>;
5031 def : SExtLoadi16CVTf64Pat<(ro16.Wpat GPR64sp:$Rn, GPR32:$Rm, ro16.Wext:$ext),
5032 (LDRHroW GPR64sp:$Rn, GPR32:$Rm, ro16.Wext:$ext)>;
5033 def : SExtLoadi16CVTf64Pat<(ro16.Xpat GPR64sp:$Rn, GPR64:$Rm, ro16.Xext:$ext),
5034 (LDRHroX GPR64sp:$Rn, GPR64:$Rm, ro16.Xext:$ext)>;
5035 def : SExtLoadi16CVTf64Pat<(am_indexed16 GPR64sp:$Rn, uimm12s2:$offset),
5036 (LDRHui GPR64sp:$Rn, uimm12s2:$offset)>;
5037 def : SExtLoadi16CVTf64Pat<(am_unscaled16 GPR64sp:$Rn, simm9:$offset),
5038 (LDURHi GPR64sp:$Rn, simm9:$offset)>;
5039 // 32-bits -> double. 1 size step-up.
5040 class SExtLoadi32CVTf64Pat<dag addrmode, dag INST>
5041 : Pat <(f64 (sint_to_fp (i32 (load addrmode)))),
5042 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
5044 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
5048 dsub)))>, Requires<[NotForCodeSize]>;
5050 def : SExtLoadi32CVTf64Pat<(ro32.Wpat GPR64sp:$Rn, GPR32:$Rm, ro32.Wext:$ext),
5051 (LDRSroW GPR64sp:$Rn, GPR32:$Rm, ro32.Wext:$ext)>;
5052 def : SExtLoadi32CVTf64Pat<(ro32.Xpat GPR64sp:$Rn, GPR64:$Rm, ro32.Xext:$ext),
5053 (LDRSroX GPR64sp:$Rn, GPR64:$Rm, ro32.Xext:$ext)>;
5054 def : SExtLoadi32CVTf64Pat<(am_indexed32 GPR64sp:$Rn, uimm12s4:$offset),
5055 (LDRSui GPR64sp:$Rn, uimm12s4:$offset)>;
5056 def : SExtLoadi32CVTf64Pat<(am_unscaled32 GPR64sp:$Rn, simm9:$offset),
5057 (LDURSi GPR64sp:$Rn, simm9:$offset)>;
5059 // 64-bits -> double are handled in target specific dag combine:
5060 // performIntToFpCombine.
5063 //----------------------------------------------------------------------------
5064 // AdvSIMD Load-Store Structure
5065 //----------------------------------------------------------------------------
5066 defm LD1 : SIMDLd1Multiple<"ld1">;
5067 defm LD2 : SIMDLd2Multiple<"ld2">;
5068 defm LD3 : SIMDLd3Multiple<"ld3">;
5069 defm LD4 : SIMDLd4Multiple<"ld4">;
5071 defm ST1 : SIMDSt1Multiple<"st1">;
5072 defm ST2 : SIMDSt2Multiple<"st2">;
5073 defm ST3 : SIMDSt3Multiple<"st3">;
5074 defm ST4 : SIMDSt4Multiple<"st4">;
5076 class Ld1Pat<ValueType ty, Instruction INST>
5077 : Pat<(ty (load GPR64sp:$Rn)), (INST GPR64sp:$Rn)>;
5079 def : Ld1Pat<v16i8, LD1Onev16b>;
5080 def : Ld1Pat<v8i16, LD1Onev8h>;
5081 def : Ld1Pat<v4i32, LD1Onev4s>;
5082 def : Ld1Pat<v2i64, LD1Onev2d>;
5083 def : Ld1Pat<v8i8, LD1Onev8b>;
5084 def : Ld1Pat<v4i16, LD1Onev4h>;
5085 def : Ld1Pat<v2i32, LD1Onev2s>;
5086 def : Ld1Pat<v1i64, LD1Onev1d>;
5088 class St1Pat<ValueType ty, Instruction INST>
5089 : Pat<(store ty:$Vt, GPR64sp:$Rn),
5090 (INST ty:$Vt, GPR64sp:$Rn)>;
5092 def : St1Pat<v16i8, ST1Onev16b>;
5093 def : St1Pat<v8i16, ST1Onev8h>;
5094 def : St1Pat<v4i32, ST1Onev4s>;
5095 def : St1Pat<v2i64, ST1Onev2d>;
5096 def : St1Pat<v8i8, ST1Onev8b>;
5097 def : St1Pat<v4i16, ST1Onev4h>;
5098 def : St1Pat<v2i32, ST1Onev2s>;
5099 def : St1Pat<v1i64, ST1Onev1d>;
5105 defm LD1R : SIMDLdR<0, 0b110, 0, "ld1r", "One", 1, 2, 4, 8>;
5106 defm LD2R : SIMDLdR<1, 0b110, 0, "ld2r", "Two", 2, 4, 8, 16>;
5107 defm LD3R : SIMDLdR<0, 0b111, 0, "ld3r", "Three", 3, 6, 12, 24>;
5108 defm LD4R : SIMDLdR<1, 0b111, 0, "ld4r", "Four", 4, 8, 16, 32>;
5109 let mayLoad = 1, hasSideEffects = 0 in {
5110 defm LD1 : SIMDLdSingleBTied<0, 0b000, "ld1", VecListOneb, GPR64pi1>;
5111 defm LD1 : SIMDLdSingleHTied<0, 0b010, 0, "ld1", VecListOneh, GPR64pi2>;
5112 defm LD1 : SIMDLdSingleSTied<0, 0b100, 0b00, "ld1", VecListOnes, GPR64pi4>;
5113 defm LD1 : SIMDLdSingleDTied<0, 0b100, 0b01, "ld1", VecListOned, GPR64pi8>;
5114 defm LD2 : SIMDLdSingleBTied<1, 0b000, "ld2", VecListTwob, GPR64pi2>;
5115 defm LD2 : SIMDLdSingleHTied<1, 0b010, 0, "ld2", VecListTwoh, GPR64pi4>;
5116 defm LD2 : SIMDLdSingleSTied<1, 0b100, 0b00, "ld2", VecListTwos, GPR64pi8>;
5117 defm LD2 : SIMDLdSingleDTied<1, 0b100, 0b01, "ld2", VecListTwod, GPR64pi16>;
5118 defm LD3 : SIMDLdSingleBTied<0, 0b001, "ld3", VecListThreeb, GPR64pi3>;
5119 defm LD3 : SIMDLdSingleHTied<0, 0b011, 0, "ld3", VecListThreeh, GPR64pi6>;
5120 defm LD3 : SIMDLdSingleSTied<0, 0b101, 0b00, "ld3", VecListThrees, GPR64pi12>;
5121 defm LD3 : SIMDLdSingleDTied<0, 0b101, 0b01, "ld3", VecListThreed, GPR64pi24>;
5122 defm LD4 : SIMDLdSingleBTied<1, 0b001, "ld4", VecListFourb, GPR64pi4>;
5123 defm LD4 : SIMDLdSingleHTied<1, 0b011, 0, "ld4", VecListFourh, GPR64pi8>;
5124 defm LD4 : SIMDLdSingleSTied<1, 0b101, 0b00, "ld4", VecListFours, GPR64pi16>;
5125 defm LD4 : SIMDLdSingleDTied<1, 0b101, 0b01, "ld4", VecListFourd, GPR64pi32>;
5128 def : Pat<(v8i8 (AArch64dup (i32 (extloadi8 GPR64sp:$Rn)))),
5129 (LD1Rv8b GPR64sp:$Rn)>;
5130 def : Pat<(v16i8 (AArch64dup (i32 (extloadi8 GPR64sp:$Rn)))),
5131 (LD1Rv16b GPR64sp:$Rn)>;
5132 def : Pat<(v4i16 (AArch64dup (i32 (extloadi16 GPR64sp:$Rn)))),
5133 (LD1Rv4h GPR64sp:$Rn)>;
5134 def : Pat<(v8i16 (AArch64dup (i32 (extloadi16 GPR64sp:$Rn)))),
5135 (LD1Rv8h GPR64sp:$Rn)>;
5136 def : Pat<(v2i32 (AArch64dup (i32 (load GPR64sp:$Rn)))),
5137 (LD1Rv2s GPR64sp:$Rn)>;
5138 def : Pat<(v4i32 (AArch64dup (i32 (load GPR64sp:$Rn)))),
5139 (LD1Rv4s GPR64sp:$Rn)>;
5140 def : Pat<(v2i64 (AArch64dup (i64 (load GPR64sp:$Rn)))),
5141 (LD1Rv2d GPR64sp:$Rn)>;
5142 def : Pat<(v1i64 (AArch64dup (i64 (load GPR64sp:$Rn)))),
5143 (LD1Rv1d GPR64sp:$Rn)>;
5144 // Grab the floating point version too
5145 def : Pat<(v2f32 (AArch64dup (f32 (load GPR64sp:$Rn)))),
5146 (LD1Rv2s GPR64sp:$Rn)>;
5147 def : Pat<(v4f32 (AArch64dup (f32 (load GPR64sp:$Rn)))),
5148 (LD1Rv4s GPR64sp:$Rn)>;
5149 def : Pat<(v2f64 (AArch64dup (f64 (load GPR64sp:$Rn)))),
5150 (LD1Rv2d GPR64sp:$Rn)>;
5151 def : Pat<(v1f64 (AArch64dup (f64 (load GPR64sp:$Rn)))),
5152 (LD1Rv1d GPR64sp:$Rn)>;
5153 def : Pat<(v4f16 (AArch64dup (f16 (load GPR64sp:$Rn)))),
5154 (LD1Rv4h GPR64sp:$Rn)>;
5155 def : Pat<(v8f16 (AArch64dup (f16 (load GPR64sp:$Rn)))),
5156 (LD1Rv8h GPR64sp:$Rn)>;
5158 class Ld1Lane128Pat<SDPatternOperator scalar_load, Operand VecIndex,
5159 ValueType VTy, ValueType STy, Instruction LD1>
5160 : Pat<(vector_insert (VTy VecListOne128:$Rd),
5161 (STy (scalar_load GPR64sp:$Rn)), VecIndex:$idx),
5162 (LD1 VecListOne128:$Rd, VecIndex:$idx, GPR64sp:$Rn)>;
5164 def : Ld1Lane128Pat<extloadi8, VectorIndexB, v16i8, i32, LD1i8>;
5165 def : Ld1Lane128Pat<extloadi16, VectorIndexH, v8i16, i32, LD1i16>;
5166 def : Ld1Lane128Pat<load, VectorIndexS, v4i32, i32, LD1i32>;
5167 def : Ld1Lane128Pat<load, VectorIndexS, v4f32, f32, LD1i32>;
5168 def : Ld1Lane128Pat<load, VectorIndexD, v2i64, i64, LD1i64>;
5169 def : Ld1Lane128Pat<load, VectorIndexD, v2f64, f64, LD1i64>;
5170 def : Ld1Lane128Pat<load, VectorIndexH, v8f16, f16, LD1i16>;
5172 class Ld1Lane64Pat<SDPatternOperator scalar_load, Operand VecIndex,
5173 ValueType VTy, ValueType STy, Instruction LD1>
5174 : Pat<(vector_insert (VTy VecListOne64:$Rd),
5175 (STy (scalar_load GPR64sp:$Rn)), VecIndex:$idx),
5177 (LD1 (SUBREG_TO_REG (i32 0), VecListOne64:$Rd, dsub),
5178 VecIndex:$idx, GPR64sp:$Rn),
5181 def : Ld1Lane64Pat<extloadi8, VectorIndexB, v8i8, i32, LD1i8>;
5182 def : Ld1Lane64Pat<extloadi16, VectorIndexH, v4i16, i32, LD1i16>;
5183 def : Ld1Lane64Pat<load, VectorIndexS, v2i32, i32, LD1i32>;
5184 def : Ld1Lane64Pat<load, VectorIndexS, v2f32, f32, LD1i32>;
5185 def : Ld1Lane64Pat<load, VectorIndexH, v4f16, f16, LD1i16>;
5188 defm LD1 : SIMDLdSt1SingleAliases<"ld1">;
5189 defm LD2 : SIMDLdSt2SingleAliases<"ld2">;
5190 defm LD3 : SIMDLdSt3SingleAliases<"ld3">;
5191 defm LD4 : SIMDLdSt4SingleAliases<"ld4">;
5194 defm ST1 : SIMDStSingleB<0, 0b000, "st1", VecListOneb, GPR64pi1>;
5195 defm ST1 : SIMDStSingleH<0, 0b010, 0, "st1", VecListOneh, GPR64pi2>;
5196 defm ST1 : SIMDStSingleS<0, 0b100, 0b00, "st1", VecListOnes, GPR64pi4>;
5197 defm ST1 : SIMDStSingleD<0, 0b100, 0b01, "st1", VecListOned, GPR64pi8>;
5199 let AddedComplexity = 19 in
5200 class St1Lane128Pat<SDPatternOperator scalar_store, Operand VecIndex,
5201 ValueType VTy, ValueType STy, Instruction ST1>
5203 (STy (vector_extract (VTy VecListOne128:$Vt), VecIndex:$idx)),
5205 (ST1 VecListOne128:$Vt, VecIndex:$idx, GPR64sp:$Rn)>;
5207 def : St1Lane128Pat<truncstorei8, VectorIndexB, v16i8, i32, ST1i8>;
5208 def : St1Lane128Pat<truncstorei16, VectorIndexH, v8i16, i32, ST1i16>;
5209 def : St1Lane128Pat<store, VectorIndexS, v4i32, i32, ST1i32>;
5210 def : St1Lane128Pat<store, VectorIndexS, v4f32, f32, ST1i32>;
5211 def : St1Lane128Pat<store, VectorIndexD, v2i64, i64, ST1i64>;
5212 def : St1Lane128Pat<store, VectorIndexD, v2f64, f64, ST1i64>;
5213 def : St1Lane128Pat<store, VectorIndexH, v8f16, f16, ST1i16>;
5215 let AddedComplexity = 19 in
5216 class St1Lane64Pat<SDPatternOperator scalar_store, Operand VecIndex,
5217 ValueType VTy, ValueType STy, Instruction ST1>
5219 (STy (vector_extract (VTy VecListOne64:$Vt), VecIndex:$idx)),
5221 (ST1 (SUBREG_TO_REG (i32 0), VecListOne64:$Vt, dsub),
5222 VecIndex:$idx, GPR64sp:$Rn)>;
5224 def : St1Lane64Pat<truncstorei8, VectorIndexB, v8i8, i32, ST1i8>;
5225 def : St1Lane64Pat<truncstorei16, VectorIndexH, v4i16, i32, ST1i16>;
5226 def : St1Lane64Pat<store, VectorIndexS, v2i32, i32, ST1i32>;
5227 def : St1Lane64Pat<store, VectorIndexS, v2f32, f32, ST1i32>;
5228 def : St1Lane64Pat<store, VectorIndexH, v4f16, f16, ST1i16>;
5230 multiclass St1LanePost64Pat<SDPatternOperator scalar_store, Operand VecIndex,
5231 ValueType VTy, ValueType STy, Instruction ST1,
5233 def : Pat<(scalar_store
5234 (STy (vector_extract (VTy VecListOne64:$Vt), VecIndex:$idx)),
5235 GPR64sp:$Rn, offset),
5236 (ST1 (SUBREG_TO_REG (i32 0), VecListOne64:$Vt, dsub),
5237 VecIndex:$idx, GPR64sp:$Rn, XZR)>;
5239 def : Pat<(scalar_store
5240 (STy (vector_extract (VTy VecListOne64:$Vt), VecIndex:$idx)),
5241 GPR64sp:$Rn, GPR64:$Rm),
5242 (ST1 (SUBREG_TO_REG (i32 0), VecListOne64:$Vt, dsub),
5243 VecIndex:$idx, GPR64sp:$Rn, $Rm)>;
5246 defm : St1LanePost64Pat<post_truncsti8, VectorIndexB, v8i8, i32, ST1i8_POST, 1>;
5247 defm : St1LanePost64Pat<post_truncsti16, VectorIndexH, v4i16, i32, ST1i16_POST,
5249 defm : St1LanePost64Pat<post_store, VectorIndexS, v2i32, i32, ST1i32_POST, 4>;
5250 defm : St1LanePost64Pat<post_store, VectorIndexS, v2f32, f32, ST1i32_POST, 4>;
5251 defm : St1LanePost64Pat<post_store, VectorIndexD, v1i64, i64, ST1i64_POST, 8>;
5252 defm : St1LanePost64Pat<post_store, VectorIndexD, v1f64, f64, ST1i64_POST, 8>;
5253 defm : St1LanePost64Pat<post_store, VectorIndexH, v4f16, f16, ST1i16_POST, 2>;
5255 multiclass St1LanePost128Pat<SDPatternOperator scalar_store, Operand VecIndex,
5256 ValueType VTy, ValueType STy, Instruction ST1,
5258 def : Pat<(scalar_store
5259 (STy (vector_extract (VTy VecListOne128:$Vt), VecIndex:$idx)),
5260 GPR64sp:$Rn, offset),
5261 (ST1 VecListOne128:$Vt, VecIndex:$idx, GPR64sp:$Rn, XZR)>;
5263 def : Pat<(scalar_store
5264 (STy (vector_extract (VTy VecListOne128:$Vt), VecIndex:$idx)),
5265 GPR64sp:$Rn, GPR64:$Rm),
5266 (ST1 VecListOne128:$Vt, VecIndex:$idx, GPR64sp:$Rn, $Rm)>;
5269 defm : St1LanePost128Pat<post_truncsti8, VectorIndexB, v16i8, i32, ST1i8_POST,
5271 defm : St1LanePost128Pat<post_truncsti16, VectorIndexH, v8i16, i32, ST1i16_POST,
5273 defm : St1LanePost128Pat<post_store, VectorIndexS, v4i32, i32, ST1i32_POST, 4>;
5274 defm : St1LanePost128Pat<post_store, VectorIndexS, v4f32, f32, ST1i32_POST, 4>;
5275 defm : St1LanePost128Pat<post_store, VectorIndexD, v2i64, i64, ST1i64_POST, 8>;
5276 defm : St1LanePost128Pat<post_store, VectorIndexD, v2f64, f64, ST1i64_POST, 8>;
5277 defm : St1LanePost128Pat<post_store, VectorIndexH, v8f16, f16, ST1i16_POST, 2>;
5279 let mayStore = 1, hasSideEffects = 0 in {
5280 defm ST2 : SIMDStSingleB<1, 0b000, "st2", VecListTwob, GPR64pi2>;
5281 defm ST2 : SIMDStSingleH<1, 0b010, 0, "st2", VecListTwoh, GPR64pi4>;
5282 defm ST2 : SIMDStSingleS<1, 0b100, 0b00, "st2", VecListTwos, GPR64pi8>;
5283 defm ST2 : SIMDStSingleD<1, 0b100, 0b01, "st2", VecListTwod, GPR64pi16>;
5284 defm ST3 : SIMDStSingleB<0, 0b001, "st3", VecListThreeb, GPR64pi3>;
5285 defm ST3 : SIMDStSingleH<0, 0b011, 0, "st3", VecListThreeh, GPR64pi6>;
5286 defm ST3 : SIMDStSingleS<0, 0b101, 0b00, "st3", VecListThrees, GPR64pi12>;
5287 defm ST3 : SIMDStSingleD<0, 0b101, 0b01, "st3", VecListThreed, GPR64pi24>;
5288 defm ST4 : SIMDStSingleB<1, 0b001, "st4", VecListFourb, GPR64pi4>;
5289 defm ST4 : SIMDStSingleH<1, 0b011, 0, "st4", VecListFourh, GPR64pi8>;
5290 defm ST4 : SIMDStSingleS<1, 0b101, 0b00, "st4", VecListFours, GPR64pi16>;
5291 defm ST4 : SIMDStSingleD<1, 0b101, 0b01, "st4", VecListFourd, GPR64pi32>;
5294 defm ST1 : SIMDLdSt1SingleAliases<"st1">;
5295 defm ST2 : SIMDLdSt2SingleAliases<"st2">;
5296 defm ST3 : SIMDLdSt3SingleAliases<"st3">;
5297 defm ST4 : SIMDLdSt4SingleAliases<"st4">;
5299 //----------------------------------------------------------------------------
5300 // Crypto extensions
5301 //----------------------------------------------------------------------------
5303 def AESErr : AESTiedInst<0b0100, "aese", int_aarch64_crypto_aese>;
5304 def AESDrr : AESTiedInst<0b0101, "aesd", int_aarch64_crypto_aesd>;
5305 def AESMCrr : AESInst< 0b0110, "aesmc", int_aarch64_crypto_aesmc>;
5306 def AESIMCrr : AESInst< 0b0111, "aesimc", int_aarch64_crypto_aesimc>;
5308 def SHA1Crrr : SHATiedInstQSV<0b000, "sha1c", int_aarch64_crypto_sha1c>;
5309 def SHA1Prrr : SHATiedInstQSV<0b001, "sha1p", int_aarch64_crypto_sha1p>;
5310 def SHA1Mrrr : SHATiedInstQSV<0b010, "sha1m", int_aarch64_crypto_sha1m>;
5311 def SHA1SU0rrr : SHATiedInstVVV<0b011, "sha1su0", int_aarch64_crypto_sha1su0>;
5312 def SHA256Hrrr : SHATiedInstQQV<0b100, "sha256h", int_aarch64_crypto_sha256h>;
5313 def SHA256H2rrr : SHATiedInstQQV<0b101, "sha256h2",int_aarch64_crypto_sha256h2>;
5314 def SHA256SU1rrr :SHATiedInstVVV<0b110, "sha256su1",int_aarch64_crypto_sha256su1>;
5316 def SHA1Hrr : SHAInstSS< 0b0000, "sha1h", int_aarch64_crypto_sha1h>;
5317 def SHA1SU1rr : SHATiedInstVV<0b0001, "sha1su1", int_aarch64_crypto_sha1su1>;
5318 def SHA256SU0rr : SHATiedInstVV<0b0010, "sha256su0",int_aarch64_crypto_sha256su0>;
5320 //----------------------------------------------------------------------------
5322 //----------------------------------------------------------------------------
5323 // FIXME: Like for X86, these should go in their own separate .td file.
5325 def def32 : PatLeaf<(i32 GPR32:$src), [{
5329 // In the case of a 32-bit def that is known to implicitly zero-extend,
5330 // we can use a SUBREG_TO_REG.
5331 def : Pat<(i64 (zext def32:$src)), (SUBREG_TO_REG (i64 0), GPR32:$src, sub_32)>;
5333 // For an anyext, we don't care what the high bits are, so we can perform an
5334 // INSERT_SUBREF into an IMPLICIT_DEF.
5335 def : Pat<(i64 (anyext GPR32:$src)),
5336 (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32)>;
5338 // When we need to explicitly zero-extend, we use a 32-bit MOV instruction and
5339 // then assert the extension has happened.
5340 def : Pat<(i64 (zext GPR32:$src)),
5341 (SUBREG_TO_REG (i32 0), (ORRWrs WZR, GPR32:$src, 0), sub_32)>;
5343 // To sign extend, we use a signed bitfield move instruction (SBFM) on the
5344 // containing super-reg.
5345 def : Pat<(i64 (sext GPR32:$src)),
5346 (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32), 0, 31)>;
5347 def : Pat<(i64 (sext_inreg GPR64:$src, i32)), (SBFMXri GPR64:$src, 0, 31)>;
5348 def : Pat<(i64 (sext_inreg GPR64:$src, i16)), (SBFMXri GPR64:$src, 0, 15)>;
5349 def : Pat<(i64 (sext_inreg GPR64:$src, i8)), (SBFMXri GPR64:$src, 0, 7)>;
5350 def : Pat<(i64 (sext_inreg GPR64:$src, i1)), (SBFMXri GPR64:$src, 0, 0)>;
5351 def : Pat<(i32 (sext_inreg GPR32:$src, i16)), (SBFMWri GPR32:$src, 0, 15)>;
5352 def : Pat<(i32 (sext_inreg GPR32:$src, i8)), (SBFMWri GPR32:$src, 0, 7)>;
5353 def : Pat<(i32 (sext_inreg GPR32:$src, i1)), (SBFMWri GPR32:$src, 0, 0)>;
5355 def : Pat<(shl (sext_inreg GPR32:$Rn, i8), (i64 imm0_31:$imm)),
5356 (SBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
5357 (i64 (i32shift_sext_i8 imm0_31:$imm)))>;
5358 def : Pat<(shl (sext_inreg GPR64:$Rn, i8), (i64 imm0_63:$imm)),
5359 (SBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
5360 (i64 (i64shift_sext_i8 imm0_63:$imm)))>;
5362 def : Pat<(shl (sext_inreg GPR32:$Rn, i16), (i64 imm0_31:$imm)),
5363 (SBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
5364 (i64 (i32shift_sext_i16 imm0_31:$imm)))>;
5365 def : Pat<(shl (sext_inreg GPR64:$Rn, i16), (i64 imm0_63:$imm)),
5366 (SBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
5367 (i64 (i64shift_sext_i16 imm0_63:$imm)))>;
5369 def : Pat<(shl (i64 (sext GPR32:$Rn)), (i64 imm0_63:$imm)),
5370 (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$Rn, sub_32),
5371 (i64 (i64shift_a imm0_63:$imm)),
5372 (i64 (i64shift_sext_i32 imm0_63:$imm)))>;
5374 // sra patterns have an AddedComplexity of 10, so make sure we have a higher
5375 // AddedComplexity for the following patterns since we want to match sext + sra
5376 // patterns before we attempt to match a single sra node.
5377 let AddedComplexity = 20 in {
5378 // We support all sext + sra combinations which preserve at least one bit of the
5379 // original value which is to be sign extended. E.g. we support shifts up to
5381 def : Pat<(sra (sext_inreg GPR32:$Rn, i8), (i64 imm0_7:$imm)),
5382 (SBFMWri GPR32:$Rn, (i64 imm0_7:$imm), 7)>;
5383 def : Pat<(sra (sext_inreg GPR64:$Rn, i8), (i64 imm0_7:$imm)),
5384 (SBFMXri GPR64:$Rn, (i64 imm0_7:$imm), 7)>;
5386 def : Pat<(sra (sext_inreg GPR32:$Rn, i16), (i64 imm0_15:$imm)),
5387 (SBFMWri GPR32:$Rn, (i64 imm0_15:$imm), 15)>;
5388 def : Pat<(sra (sext_inreg GPR64:$Rn, i16), (i64 imm0_15:$imm)),
5389 (SBFMXri GPR64:$Rn, (i64 imm0_15:$imm), 15)>;
5391 def : Pat<(sra (i64 (sext GPR32:$Rn)), (i64 imm0_31:$imm)),
5392 (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$Rn, sub_32),
5393 (i64 imm0_31:$imm), 31)>;
5394 } // AddedComplexity = 20
5396 // To truncate, we can simply extract from a subregister.
5397 def : Pat<(i32 (trunc GPR64sp:$src)),
5398 (i32 (EXTRACT_SUBREG GPR64sp:$src, sub_32))>;
5400 // __builtin_trap() uses the BRK instruction on AArch64.
5401 def : Pat<(trap), (BRK 1)>;
5403 // Conversions within AdvSIMD types in the same register size are free.
5404 // But because we need a consistent lane ordering, in big endian many
5405 // conversions require one or more REV instructions.
5407 // Consider a simple memory load followed by a bitconvert then a store.
5409 // v1 = BITCAST v2i32 v0 to v4i16
5412 // In big endian mode every memory access has an implicit byte swap. LDR and
5413 // STR do a 64-bit byte swap, whereas LD1/ST1 do a byte swap per lane - that
5414 // is, they treat the vector as a sequence of elements to be byte-swapped.
5415 // The two pairs of instructions are fundamentally incompatible. We've decided
5416 // to use LD1/ST1 only to simplify compiler implementation.
5418 // LD1/ST1 perform the equivalent of a sequence of LDR/STR + REV. This makes
5419 // the original code sequence:
5421 // v1 = REV v2i32 (implicit)
5422 // v2 = BITCAST v2i32 v1 to v4i16
5423 // v3 = REV v4i16 v2 (implicit)
5426 // But this is now broken - the value stored is different to the value loaded
5427 // due to lane reordering. To fix this, on every BITCAST we must perform two
5430 // v1 = REV v2i32 (implicit)
5432 // v3 = BITCAST v2i32 v2 to v4i16
5434 // v5 = REV v4i16 v4 (implicit)
5437 // This means an extra two instructions, but actually in most cases the two REV
5438 // instructions can be combined into one. For example:
5439 // (REV64_2s (REV64_4h X)) === (REV32_4h X)
5441 // There is also no 128-bit REV instruction. This must be synthesized with an
5444 // Most bitconverts require some sort of conversion. The only exceptions are:
5445 // a) Identity conversions - vNfX <-> vNiX
5446 // b) Single-lane-to-scalar - v1fX <-> fX or v1iX <-> iX
5449 // Natural vector casts (64 bit)
5450 def : Pat<(v8i8 (AArch64NvCast (v2i32 FPR64:$src))), (v8i8 FPR64:$src)>;
5451 def : Pat<(v4i16 (AArch64NvCast (v2i32 FPR64:$src))), (v4i16 FPR64:$src)>;
5452 def : Pat<(v4f16 (AArch64NvCast (v2i32 FPR64:$src))), (v4f16 FPR64:$src)>;
5453 def : Pat<(v2i32 (AArch64NvCast (v2i32 FPR64:$src))), (v2i32 FPR64:$src)>;
5454 def : Pat<(v2f32 (AArch64NvCast (v2i32 FPR64:$src))), (v2f32 FPR64:$src)>;
5455 def : Pat<(v1i64 (AArch64NvCast (v2i32 FPR64:$src))), (v1i64 FPR64:$src)>;
5457 def : Pat<(v8i8 (AArch64NvCast (v4i16 FPR64:$src))), (v8i8 FPR64:$src)>;
5458 def : Pat<(v4i16 (AArch64NvCast (v4i16 FPR64:$src))), (v4i16 FPR64:$src)>;
5459 def : Pat<(v4f16 (AArch64NvCast (v4i16 FPR64:$src))), (v4f16 FPR64:$src)>;
5460 def : Pat<(v2i32 (AArch64NvCast (v4i16 FPR64:$src))), (v2i32 FPR64:$src)>;
5461 def : Pat<(v1i64 (AArch64NvCast (v4i16 FPR64:$src))), (v1i64 FPR64:$src)>;
5463 def : Pat<(v8i8 (AArch64NvCast (v8i8 FPR64:$src))), (v8i8 FPR64:$src)>;
5464 def : Pat<(v4i16 (AArch64NvCast (v8i8 FPR64:$src))), (v4i16 FPR64:$src)>;
5465 def : Pat<(v4f16 (AArch64NvCast (v8i8 FPR64:$src))), (v4f16 FPR64:$src)>;
5466 def : Pat<(v2i32 (AArch64NvCast (v8i8 FPR64:$src))), (v2i32 FPR64:$src)>;
5467 def : Pat<(v1i64 (AArch64NvCast (v8i8 FPR64:$src))), (v1i64 FPR64:$src)>;
5469 def : Pat<(v8i8 (AArch64NvCast (f64 FPR64:$src))), (v8i8 FPR64:$src)>;
5470 def : Pat<(v4i16 (AArch64NvCast (f64 FPR64:$src))), (v4i16 FPR64:$src)>;
5471 def : Pat<(v4f16 (AArch64NvCast (f64 FPR64:$src))), (v4f16 FPR64:$src)>;
5472 def : Pat<(v2i32 (AArch64NvCast (f64 FPR64:$src))), (v2i32 FPR64:$src)>;
5473 def : Pat<(v2f32 (AArch64NvCast (f64 FPR64:$src))), (v2f32 FPR64:$src)>;
5474 def : Pat<(v1i64 (AArch64NvCast (f64 FPR64:$src))), (v1i64 FPR64:$src)>;
5475 def : Pat<(v1f64 (AArch64NvCast (f64 FPR64:$src))), (v1f64 FPR64:$src)>;
5477 def : Pat<(v8i8 (AArch64NvCast (v2f32 FPR64:$src))), (v8i8 FPR64:$src)>;
5478 def : Pat<(v4i16 (AArch64NvCast (v2f32 FPR64:$src))), (v4i16 FPR64:$src)>;
5479 def : Pat<(v2i32 (AArch64NvCast (v2f32 FPR64:$src))), (v2i32 FPR64:$src)>;
5480 def : Pat<(v2f32 (AArch64NvCast (v2f32 FPR64:$src))), (v2f32 FPR64:$src)>;
5481 def : Pat<(v1i64 (AArch64NvCast (v2f32 FPR64:$src))), (v1i64 FPR64:$src)>;
5483 // Natural vector casts (128 bit)
5484 def : Pat<(v16i8 (AArch64NvCast (v4i32 FPR128:$src))), (v16i8 FPR128:$src)>;
5485 def : Pat<(v8i16 (AArch64NvCast (v4i32 FPR128:$src))), (v8i16 FPR128:$src)>;
5486 def : Pat<(v8f16 (AArch64NvCast (v4i32 FPR128:$src))), (v8f16 FPR128:$src)>;
5487 def : Pat<(v4i32 (AArch64NvCast (v4i32 FPR128:$src))), (v4i32 FPR128:$src)>;
5488 def : Pat<(v4f32 (AArch64NvCast (v4i32 FPR128:$src))), (v4f32 FPR128:$src)>;
5489 def : Pat<(v2i64 (AArch64NvCast (v4i32 FPR128:$src))), (v2i64 FPR128:$src)>;
5490 def : Pat<(v2f64 (AArch64NvCast (v4i32 FPR128:$src))), (v2f64 FPR128:$src)>;
5492 def : Pat<(v16i8 (AArch64NvCast (v8i16 FPR128:$src))), (v16i8 FPR128:$src)>;
5493 def : Pat<(v8i16 (AArch64NvCast (v8i16 FPR128:$src))), (v8i16 FPR128:$src)>;
5494 def : Pat<(v8f16 (AArch64NvCast (v8i16 FPR128:$src))), (v8f16 FPR128:$src)>;
5495 def : Pat<(v4i32 (AArch64NvCast (v8i16 FPR128:$src))), (v4i32 FPR128:$src)>;
5496 def : Pat<(v2i64 (AArch64NvCast (v8i16 FPR128:$src))), (v2i64 FPR128:$src)>;
5497 def : Pat<(v4f32 (AArch64NvCast (v8i16 FPR128:$src))), (v4f32 FPR128:$src)>;
5498 def : Pat<(v2f64 (AArch64NvCast (v8i16 FPR128:$src))), (v2f64 FPR128:$src)>;
5500 def : Pat<(v16i8 (AArch64NvCast (v16i8 FPR128:$src))), (v16i8 FPR128:$src)>;
5501 def : Pat<(v8i16 (AArch64NvCast (v16i8 FPR128:$src))), (v8i16 FPR128:$src)>;
5502 def : Pat<(v8f16 (AArch64NvCast (v16i8 FPR128:$src))), (v8f16 FPR128:$src)>;
5503 def : Pat<(v4i32 (AArch64NvCast (v16i8 FPR128:$src))), (v4i32 FPR128:$src)>;
5504 def : Pat<(v2i64 (AArch64NvCast (v16i8 FPR128:$src))), (v2i64 FPR128:$src)>;
5505 def : Pat<(v4f32 (AArch64NvCast (v16i8 FPR128:$src))), (v4f32 FPR128:$src)>;
5506 def : Pat<(v2f64 (AArch64NvCast (v16i8 FPR128:$src))), (v2f64 FPR128:$src)>;
5508 def : Pat<(v16i8 (AArch64NvCast (v2i64 FPR128:$src))), (v16i8 FPR128:$src)>;
5509 def : Pat<(v8i16 (AArch64NvCast (v2i64 FPR128:$src))), (v8i16 FPR128:$src)>;
5510 def : Pat<(v8f16 (AArch64NvCast (v2i64 FPR128:$src))), (v8f16 FPR128:$src)>;
5511 def : Pat<(v4i32 (AArch64NvCast (v2i64 FPR128:$src))), (v4i32 FPR128:$src)>;
5512 def : Pat<(v2i64 (AArch64NvCast (v2i64 FPR128:$src))), (v2i64 FPR128:$src)>;
5513 def : Pat<(v4f32 (AArch64NvCast (v2i64 FPR128:$src))), (v4f32 FPR128:$src)>;
5514 def : Pat<(v2f64 (AArch64NvCast (v2i64 FPR128:$src))), (v2f64 FPR128:$src)>;
5516 def : Pat<(v16i8 (AArch64NvCast (v4f32 FPR128:$src))), (v16i8 FPR128:$src)>;
5517 def : Pat<(v8i16 (AArch64NvCast (v4f32 FPR128:$src))), (v8i16 FPR128:$src)>;
5518 def : Pat<(v4i32 (AArch64NvCast (v4f32 FPR128:$src))), (v4i32 FPR128:$src)>;
5519 def : Pat<(v4f32 (AArch64NvCast (v4f32 FPR128:$src))), (v4f32 FPR128:$src)>;
5520 def : Pat<(v2i64 (AArch64NvCast (v4f32 FPR128:$src))), (v2i64 FPR128:$src)>;
5521 def : Pat<(v8f16 (AArch64NvCast (v4f32 FPR128:$src))), (v8f16 FPR128:$src)>;
5522 def : Pat<(v2f64 (AArch64NvCast (v4f32 FPR128:$src))), (v2f64 FPR128:$src)>;
5524 def : Pat<(v16i8 (AArch64NvCast (v2f64 FPR128:$src))), (v16i8 FPR128:$src)>;
5525 def : Pat<(v8i16 (AArch64NvCast (v2f64 FPR128:$src))), (v8i16 FPR128:$src)>;
5526 def : Pat<(v4i32 (AArch64NvCast (v2f64 FPR128:$src))), (v4i32 FPR128:$src)>;
5527 def : Pat<(v2i64 (AArch64NvCast (v2f64 FPR128:$src))), (v2i64 FPR128:$src)>;
5528 def : Pat<(v2f64 (AArch64NvCast (v2f64 FPR128:$src))), (v2f64 FPR128:$src)>;
5529 def : Pat<(v8f16 (AArch64NvCast (v2f64 FPR128:$src))), (v8f16 FPR128:$src)>;
5530 def : Pat<(v4f32 (AArch64NvCast (v2f64 FPR128:$src))), (v4f32 FPR128:$src)>;
5532 let Predicates = [IsLE] in {
5533 def : Pat<(v8i8 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5534 def : Pat<(v4i16 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5535 def : Pat<(v2i32 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5536 def : Pat<(v4f16 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5537 def : Pat<(v2f32 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5539 def : Pat<(i64 (bitconvert (v8i8 V64:$Vn))),
5540 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5541 def : Pat<(i64 (bitconvert (v4i16 V64:$Vn))),
5542 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5543 def : Pat<(i64 (bitconvert (v2i32 V64:$Vn))),
5544 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5545 def : Pat<(i64 (bitconvert (v4f16 V64:$Vn))),
5546 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5547 def : Pat<(i64 (bitconvert (v2f32 V64:$Vn))),
5548 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5549 def : Pat<(i64 (bitconvert (v1f64 V64:$Vn))),
5550 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5552 let Predicates = [IsBE] in {
5553 def : Pat<(v8i8 (bitconvert GPR64:$Xn)),
5554 (REV64v8i8 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
5555 def : Pat<(v4i16 (bitconvert GPR64:$Xn)),
5556 (REV64v4i16 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
5557 def : Pat<(v2i32 (bitconvert GPR64:$Xn)),
5558 (REV64v2i32 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
5559 def : Pat<(v4f16 (bitconvert GPR64:$Xn)),
5560 (REV64v4i16 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
5561 def : Pat<(v2f32 (bitconvert GPR64:$Xn)),
5562 (REV64v2i32 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
5564 def : Pat<(i64 (bitconvert (v8i8 V64:$Vn))),
5565 (REV64v8i8 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
5566 def : Pat<(i64 (bitconvert (v4i16 V64:$Vn))),
5567 (REV64v4i16 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
5568 def : Pat<(i64 (bitconvert (v2i32 V64:$Vn))),
5569 (REV64v2i32 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
5570 def : Pat<(i64 (bitconvert (v4f16 V64:$Vn))),
5571 (REV64v4i16 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
5572 def : Pat<(i64 (bitconvert (v2f32 V64:$Vn))),
5573 (REV64v2i32 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
5575 def : Pat<(v1i64 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5576 def : Pat<(v1f64 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5577 def : Pat<(i64 (bitconvert (v1i64 V64:$Vn))),
5578 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5579 def : Pat<(v1i64 (scalar_to_vector GPR64:$Xn)),
5580 (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5581 def : Pat<(v1f64 (scalar_to_vector GPR64:$Xn)),
5582 (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5583 def : Pat<(v1f64 (scalar_to_vector (f64 FPR64:$Xn))), (v1f64 FPR64:$Xn)>;
5585 def : Pat<(f32 (bitconvert (i32 GPR32:$Xn))),
5586 (COPY_TO_REGCLASS GPR32:$Xn, FPR32)>;
5587 def : Pat<(i32 (bitconvert (f32 FPR32:$Xn))),
5588 (COPY_TO_REGCLASS FPR32:$Xn, GPR32)>;
5589 def : Pat<(f64 (bitconvert (i64 GPR64:$Xn))),
5590 (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5591 def : Pat<(i64 (bitconvert (f64 FPR64:$Xn))),
5592 (COPY_TO_REGCLASS FPR64:$Xn, GPR64)>;
5593 def : Pat<(i64 (bitconvert (v1f64 V64:$Vn))),
5594 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5596 let Predicates = [IsLE] in {
5597 def : Pat<(v1i64 (bitconvert (v2i32 FPR64:$src))), (v1i64 FPR64:$src)>;
5598 def : Pat<(v1i64 (bitconvert (v4i16 FPR64:$src))), (v1i64 FPR64:$src)>;
5599 def : Pat<(v1i64 (bitconvert (v8i8 FPR64:$src))), (v1i64 FPR64:$src)>;
5600 def : Pat<(v1i64 (bitconvert (v4f16 FPR64:$src))), (v1i64 FPR64:$src)>;
5601 def : Pat<(v1i64 (bitconvert (v2f32 FPR64:$src))), (v1i64 FPR64:$src)>;
5603 let Predicates = [IsBE] in {
5604 def : Pat<(v1i64 (bitconvert (v2i32 FPR64:$src))),
5605 (v1i64 (REV64v2i32 FPR64:$src))>;
5606 def : Pat<(v1i64 (bitconvert (v4i16 FPR64:$src))),
5607 (v1i64 (REV64v4i16 FPR64:$src))>;
5608 def : Pat<(v1i64 (bitconvert (v8i8 FPR64:$src))),
5609 (v1i64 (REV64v8i8 FPR64:$src))>;
5610 def : Pat<(v1i64 (bitconvert (v4f16 FPR64:$src))),
5611 (v1i64 (REV64v4i16 FPR64:$src))>;
5612 def : Pat<(v1i64 (bitconvert (v2f32 FPR64:$src))),
5613 (v1i64 (REV64v2i32 FPR64:$src))>;
5615 def : Pat<(v1i64 (bitconvert (v1f64 FPR64:$src))), (v1i64 FPR64:$src)>;
5616 def : Pat<(v1i64 (bitconvert (f64 FPR64:$src))), (v1i64 FPR64:$src)>;
5618 let Predicates = [IsLE] in {
5619 def : Pat<(v2i32 (bitconvert (v1i64 FPR64:$src))), (v2i32 FPR64:$src)>;
5620 def : Pat<(v2i32 (bitconvert (v4i16 FPR64:$src))), (v2i32 FPR64:$src)>;
5621 def : Pat<(v2i32 (bitconvert (v8i8 FPR64:$src))), (v2i32 FPR64:$src)>;
5622 def : Pat<(v2i32 (bitconvert (f64 FPR64:$src))), (v2i32 FPR64:$src)>;
5623 def : Pat<(v2i32 (bitconvert (v1f64 FPR64:$src))), (v2i32 FPR64:$src)>;
5624 def : Pat<(v2i32 (bitconvert (v4f16 FPR64:$src))), (v2i32 FPR64:$src)>;
5626 let Predicates = [IsBE] in {
5627 def : Pat<(v2i32 (bitconvert (v1i64 FPR64:$src))),
5628 (v2i32 (REV64v2i32 FPR64:$src))>;
5629 def : Pat<(v2i32 (bitconvert (v4i16 FPR64:$src))),
5630 (v2i32 (REV32v4i16 FPR64:$src))>;
5631 def : Pat<(v2i32 (bitconvert (v8i8 FPR64:$src))),
5632 (v2i32 (REV32v8i8 FPR64:$src))>;
5633 def : Pat<(v2i32 (bitconvert (f64 FPR64:$src))),
5634 (v2i32 (REV64v2i32 FPR64:$src))>;
5635 def : Pat<(v2i32 (bitconvert (v1f64 FPR64:$src))),
5636 (v2i32 (REV64v2i32 FPR64:$src))>;
5637 def : Pat<(v2i32 (bitconvert (v4f16 FPR64:$src))),
5638 (v2i32 (REV64v4i16 FPR64:$src))>;
5640 def : Pat<(v2i32 (bitconvert (v2f32 FPR64:$src))), (v2i32 FPR64:$src)>;
5642 let Predicates = [IsLE] in {
5643 def : Pat<(v4i16 (bitconvert (v1i64 FPR64:$src))), (v4i16 FPR64:$src)>;
5644 def : Pat<(v4i16 (bitconvert (v2i32 FPR64:$src))), (v4i16 FPR64:$src)>;
5645 def : Pat<(v4i16 (bitconvert (v8i8 FPR64:$src))), (v4i16 FPR64:$src)>;
5646 def : Pat<(v4i16 (bitconvert (f64 FPR64:$src))), (v4i16 FPR64:$src)>;
5647 def : Pat<(v4i16 (bitconvert (v4f16 FPR64:$src))), (v4i16 FPR64:$src)>;
5648 def : Pat<(v4i16 (bitconvert (v2f32 FPR64:$src))), (v4i16 FPR64:$src)>;
5649 def : Pat<(v4i16 (bitconvert (v1f64 FPR64:$src))), (v4i16 FPR64:$src)>;
5651 let Predicates = [IsBE] in {
5652 def : Pat<(v4i16 (bitconvert (v1i64 FPR64:$src))),
5653 (v4i16 (REV64v4i16 FPR64:$src))>;
5654 def : Pat<(v4i16 (bitconvert (v2i32 FPR64:$src))),
5655 (v4i16 (REV32v4i16 FPR64:$src))>;
5656 def : Pat<(v4i16 (bitconvert (v8i8 FPR64:$src))),
5657 (v4i16 (REV16v8i8 FPR64:$src))>;
5658 def : Pat<(v4i16 (bitconvert (f64 FPR64:$src))),
5659 (v4i16 (REV64v4i16 FPR64:$src))>;
5660 def : Pat<(v4i16 (bitconvert (v4f16 FPR64:$src))),
5661 (v4i16 (REV32v4i16 FPR64:$src))>;
5662 def : Pat<(v4i16 (bitconvert (v2f32 FPR64:$src))),
5663 (v4i16 (REV32v4i16 FPR64:$src))>;
5664 def : Pat<(v4i16 (bitconvert (v1f64 FPR64:$src))),
5665 (v4i16 (REV64v4i16 FPR64:$src))>;
5668 let Predicates = [IsLE] in {
5669 def : Pat<(v4f16 (bitconvert (v1i64 FPR64:$src))), (v4f16 FPR64:$src)>;
5670 def : Pat<(v4f16 (bitconvert (v2i32 FPR64:$src))), (v4f16 FPR64:$src)>;
5671 def : Pat<(v4f16 (bitconvert (v4i16 FPR64:$src))), (v4f16 FPR64:$src)>;
5672 def : Pat<(v4f16 (bitconvert (v8i8 FPR64:$src))), (v4f16 FPR64:$src)>;
5673 def : Pat<(v4f16 (bitconvert (f64 FPR64:$src))), (v4f16 FPR64:$src)>;
5674 def : Pat<(v4f16 (bitconvert (v2f32 FPR64:$src))), (v4f16 FPR64:$src)>;
5675 def : Pat<(v4f16 (bitconvert (v1f64 FPR64:$src))), (v4f16 FPR64:$src)>;
5677 let Predicates = [IsBE] in {
5678 def : Pat<(v4f16 (bitconvert (v1i64 FPR64:$src))),
5679 (v4f16 (REV64v4i16 FPR64:$src))>;
5680 def : Pat<(v4f16 (bitconvert (v2i32 FPR64:$src))),
5681 (v4f16 (REV64v4i16 FPR64:$src))>;
5682 def : Pat<(v4f16 (bitconvert (v4i16 FPR64:$src))),
5683 (v4f16 (REV64v4i16 FPR64:$src))>;
5684 def : Pat<(v4f16 (bitconvert (v8i8 FPR64:$src))),
5685 (v4f16 (REV16v8i8 FPR64:$src))>;
5686 def : Pat<(v4f16 (bitconvert (f64 FPR64:$src))),
5687 (v4f16 (REV64v4i16 FPR64:$src))>;
5688 def : Pat<(v4f16 (bitconvert (v2f32 FPR64:$src))),
5689 (v4f16 (REV64v4i16 FPR64:$src))>;
5690 def : Pat<(v4f16 (bitconvert (v1f64 FPR64:$src))),
5691 (v4f16 (REV64v4i16 FPR64:$src))>;
5696 let Predicates = [IsLE] in {
5697 def : Pat<(v8i8 (bitconvert (v1i64 FPR64:$src))), (v8i8 FPR64:$src)>;
5698 def : Pat<(v8i8 (bitconvert (v2i32 FPR64:$src))), (v8i8 FPR64:$src)>;
5699 def : Pat<(v8i8 (bitconvert (v4i16 FPR64:$src))), (v8i8 FPR64:$src)>;
5700 def : Pat<(v8i8 (bitconvert (f64 FPR64:$src))), (v8i8 FPR64:$src)>;
5701 def : Pat<(v8i8 (bitconvert (v2f32 FPR64:$src))), (v8i8 FPR64:$src)>;
5702 def : Pat<(v8i8 (bitconvert (v1f64 FPR64:$src))), (v8i8 FPR64:$src)>;
5703 def : Pat<(v8i8 (bitconvert (v4f16 FPR64:$src))), (v8i8 FPR64:$src)>;
5705 let Predicates = [IsBE] in {
5706 def : Pat<(v8i8 (bitconvert (v1i64 FPR64:$src))),
5707 (v8i8 (REV64v8i8 FPR64:$src))>;
5708 def : Pat<(v8i8 (bitconvert (v2i32 FPR64:$src))),
5709 (v8i8 (REV32v8i8 FPR64:$src))>;
5710 def : Pat<(v8i8 (bitconvert (v4i16 FPR64:$src))),
5711 (v8i8 (REV16v8i8 FPR64:$src))>;
5712 def : Pat<(v8i8 (bitconvert (f64 FPR64:$src))),
5713 (v8i8 (REV64v8i8 FPR64:$src))>;
5714 def : Pat<(v8i8 (bitconvert (v2f32 FPR64:$src))),
5715 (v8i8 (REV32v8i8 FPR64:$src))>;
5716 def : Pat<(v8i8 (bitconvert (v1f64 FPR64:$src))),
5717 (v8i8 (REV64v8i8 FPR64:$src))>;
5718 def : Pat<(v8i8 (bitconvert (v4f16 FPR64:$src))),
5719 (v8i8 (REV16v8i8 FPR64:$src))>;
5722 let Predicates = [IsLE] in {
5723 def : Pat<(f64 (bitconvert (v2i32 FPR64:$src))), (f64 FPR64:$src)>;
5724 def : Pat<(f64 (bitconvert (v4i16 FPR64:$src))), (f64 FPR64:$src)>;
5725 def : Pat<(f64 (bitconvert (v2f32 FPR64:$src))), (f64 FPR64:$src)>;
5726 def : Pat<(f64 (bitconvert (v8i8 FPR64:$src))), (f64 FPR64:$src)>;
5727 def : Pat<(f64 (bitconvert (v4f16 FPR64:$src))), (f64 FPR64:$src)>;
5729 let Predicates = [IsBE] in {
5730 def : Pat<(f64 (bitconvert (v2i32 FPR64:$src))),
5731 (f64 (REV64v2i32 FPR64:$src))>;
5732 def : Pat<(f64 (bitconvert (v4i16 FPR64:$src))),
5733 (f64 (REV64v4i16 FPR64:$src))>;
5734 def : Pat<(f64 (bitconvert (v2f32 FPR64:$src))),
5735 (f64 (REV64v2i32 FPR64:$src))>;
5736 def : Pat<(f64 (bitconvert (v8i8 FPR64:$src))),
5737 (f64 (REV64v8i8 FPR64:$src))>;
5738 def : Pat<(f64 (bitconvert (v4f16 FPR64:$src))),
5739 (f64 (REV64v4i16 FPR64:$src))>;
5741 def : Pat<(f64 (bitconvert (v1i64 FPR64:$src))), (f64 FPR64:$src)>;
5742 def : Pat<(f64 (bitconvert (v1f64 FPR64:$src))), (f64 FPR64:$src)>;
5744 let Predicates = [IsLE] in {
5745 def : Pat<(v1f64 (bitconvert (v2i32 FPR64:$src))), (v1f64 FPR64:$src)>;
5746 def : Pat<(v1f64 (bitconvert (v4i16 FPR64:$src))), (v1f64 FPR64:$src)>;
5747 def : Pat<(v1f64 (bitconvert (v8i8 FPR64:$src))), (v1f64 FPR64:$src)>;
5748 def : Pat<(v1f64 (bitconvert (v2f32 FPR64:$src))), (v1f64 FPR64:$src)>;
5749 def : Pat<(v1f64 (bitconvert (v4f16 FPR64:$src))), (v1f64 FPR64:$src)>;
5751 let Predicates = [IsBE] in {
5752 def : Pat<(v1f64 (bitconvert (v2i32 FPR64:$src))),
5753 (v1f64 (REV64v2i32 FPR64:$src))>;
5754 def : Pat<(v1f64 (bitconvert (v4i16 FPR64:$src))),
5755 (v1f64 (REV64v4i16 FPR64:$src))>;
5756 def : Pat<(v1f64 (bitconvert (v8i8 FPR64:$src))),
5757 (v1f64 (REV64v8i8 FPR64:$src))>;
5758 def : Pat<(v1f64 (bitconvert (v2f32 FPR64:$src))),
5759 (v1f64 (REV64v2i32 FPR64:$src))>;
5760 def : Pat<(v1f64 (bitconvert (v4f16 FPR64:$src))),
5761 (v1f64 (REV64v4i16 FPR64:$src))>;
5763 def : Pat<(v1f64 (bitconvert (v1i64 FPR64:$src))), (v1f64 FPR64:$src)>;
5764 def : Pat<(v1f64 (bitconvert (f64 FPR64:$src))), (v1f64 FPR64:$src)>;
5766 let Predicates = [IsLE] in {
5767 def : Pat<(v2f32 (bitconvert (v1i64 FPR64:$src))), (v2f32 FPR64:$src)>;
5768 def : Pat<(v2f32 (bitconvert (v4i16 FPR64:$src))), (v2f32 FPR64:$src)>;
5769 def : Pat<(v2f32 (bitconvert (v8i8 FPR64:$src))), (v2f32 FPR64:$src)>;
5770 def : Pat<(v2f32 (bitconvert (v1f64 FPR64:$src))), (v2f32 FPR64:$src)>;
5771 def : Pat<(v2f32 (bitconvert (f64 FPR64:$src))), (v2f32 FPR64:$src)>;
5772 def : Pat<(v2f32 (bitconvert (v4f16 FPR64:$src))), (v2f32 FPR64:$src)>;
5774 let Predicates = [IsBE] in {
5775 def : Pat<(v2f32 (bitconvert (v1i64 FPR64:$src))),
5776 (v2f32 (REV64v2i32 FPR64:$src))>;
5777 def : Pat<(v2f32 (bitconvert (v4i16 FPR64:$src))),
5778 (v2f32 (REV32v4i16 FPR64:$src))>;
5779 def : Pat<(v2f32 (bitconvert (v8i8 FPR64:$src))),
5780 (v2f32 (REV32v8i8 FPR64:$src))>;
5781 def : Pat<(v2f32 (bitconvert (v1f64 FPR64:$src))),
5782 (v2f32 (REV64v2i32 FPR64:$src))>;
5783 def : Pat<(v2f32 (bitconvert (f64 FPR64:$src))),
5784 (v2f32 (REV64v2i32 FPR64:$src))>;
5785 def : Pat<(v2f32 (bitconvert (v4f16 FPR64:$src))),
5786 (v2f32 (REV64v4i16 FPR64:$src))>;
5788 def : Pat<(v2f32 (bitconvert (v2i32 FPR64:$src))), (v2f32 FPR64:$src)>;
5790 let Predicates = [IsLE] in {
5791 def : Pat<(f128 (bitconvert (v2i64 FPR128:$src))), (f128 FPR128:$src)>;
5792 def : Pat<(f128 (bitconvert (v4i32 FPR128:$src))), (f128 FPR128:$src)>;
5793 def : Pat<(f128 (bitconvert (v8i16 FPR128:$src))), (f128 FPR128:$src)>;
5794 def : Pat<(f128 (bitconvert (v2f64 FPR128:$src))), (f128 FPR128:$src)>;
5795 def : Pat<(f128 (bitconvert (v4f32 FPR128:$src))), (f128 FPR128:$src)>;
5796 def : Pat<(f128 (bitconvert (v8f16 FPR128:$src))), (f128 FPR128:$src)>;
5797 def : Pat<(f128 (bitconvert (v16i8 FPR128:$src))), (f128 FPR128:$src)>;
5799 let Predicates = [IsBE] in {
5800 def : Pat<(f128 (bitconvert (v2i64 FPR128:$src))),
5801 (f128 (EXTv16i8 FPR128:$src, FPR128:$src, (i32 8)))>;
5802 def : Pat<(f128 (bitconvert (v4i32 FPR128:$src))),
5803 (f128 (EXTv16i8 (REV64v4i32 FPR128:$src),
5804 (REV64v4i32 FPR128:$src), (i32 8)))>;
5805 def : Pat<(f128 (bitconvert (v8i16 FPR128:$src))),
5806 (f128 (EXTv16i8 (REV64v8i16 FPR128:$src),
5807 (REV64v8i16 FPR128:$src), (i32 8)))>;
5808 def : Pat<(f128 (bitconvert (v8f16 FPR128:$src))),
5809 (f128 (EXTv16i8 (REV64v8i16 FPR128:$src),
5810 (REV64v8i16 FPR128:$src), (i32 8)))>;
5811 def : Pat<(f128 (bitconvert (v2f64 FPR128:$src))),
5812 (f128 (EXTv16i8 FPR128:$src, FPR128:$src, (i32 8)))>;
5813 def : Pat<(f128 (bitconvert (v4f32 FPR128:$src))),
5814 (f128 (EXTv16i8 (REV64v4i32 FPR128:$src),
5815 (REV64v4i32 FPR128:$src), (i32 8)))>;
5816 def : Pat<(f128 (bitconvert (v16i8 FPR128:$src))),
5817 (f128 (EXTv16i8 (REV64v16i8 FPR128:$src),
5818 (REV64v16i8 FPR128:$src), (i32 8)))>;
5821 let Predicates = [IsLE] in {
5822 def : Pat<(v2f64 (bitconvert (f128 FPR128:$src))), (v2f64 FPR128:$src)>;
5823 def : Pat<(v2f64 (bitconvert (v4i32 FPR128:$src))), (v2f64 FPR128:$src)>;
5824 def : Pat<(v2f64 (bitconvert (v8i16 FPR128:$src))), (v2f64 FPR128:$src)>;
5825 def : Pat<(v2f64 (bitconvert (v8f16 FPR128:$src))), (v2f64 FPR128:$src)>;
5826 def : Pat<(v2f64 (bitconvert (v16i8 FPR128:$src))), (v2f64 FPR128:$src)>;
5827 def : Pat<(v2f64 (bitconvert (v4f32 FPR128:$src))), (v2f64 FPR128:$src)>;
5829 let Predicates = [IsBE] in {
5830 def : Pat<(v2f64 (bitconvert (f128 FPR128:$src))),
5831 (v2f64 (EXTv16i8 FPR128:$src,
5832 FPR128:$src, (i32 8)))>;
5833 def : Pat<(v2f64 (bitconvert (v4i32 FPR128:$src))),
5834 (v2f64 (REV64v4i32 FPR128:$src))>;
5835 def : Pat<(v2f64 (bitconvert (v8i16 FPR128:$src))),
5836 (v2f64 (REV64v8i16 FPR128:$src))>;
5837 def : Pat<(v2f64 (bitconvert (v8f16 FPR128:$src))),
5838 (v2f64 (REV64v8i16 FPR128:$src))>;
5839 def : Pat<(v2f64 (bitconvert (v16i8 FPR128:$src))),
5840 (v2f64 (REV64v16i8 FPR128:$src))>;
5841 def : Pat<(v2f64 (bitconvert (v4f32 FPR128:$src))),
5842 (v2f64 (REV64v4i32 FPR128:$src))>;
5844 def : Pat<(v2f64 (bitconvert (v2i64 FPR128:$src))), (v2f64 FPR128:$src)>;
5846 let Predicates = [IsLE] in {
5847 def : Pat<(v4f32 (bitconvert (f128 FPR128:$src))), (v4f32 FPR128:$src)>;
5848 def : Pat<(v4f32 (bitconvert (v8i16 FPR128:$src))), (v4f32 FPR128:$src)>;
5849 def : Pat<(v4f32 (bitconvert (v8f16 FPR128:$src))), (v4f32 FPR128:$src)>;
5850 def : Pat<(v4f32 (bitconvert (v16i8 FPR128:$src))), (v4f32 FPR128:$src)>;
5851 def : Pat<(v4f32 (bitconvert (v2i64 FPR128:$src))), (v4f32 FPR128:$src)>;
5852 def : Pat<(v4f32 (bitconvert (v2f64 FPR128:$src))), (v4f32 FPR128:$src)>;
5854 let Predicates = [IsBE] in {
5855 def : Pat<(v4f32 (bitconvert (f128 FPR128:$src))),
5856 (v4f32 (EXTv16i8 (REV64v4i32 FPR128:$src),
5857 (REV64v4i32 FPR128:$src), (i32 8)))>;
5858 def : Pat<(v4f32 (bitconvert (v8i16 FPR128:$src))),
5859 (v4f32 (REV32v8i16 FPR128:$src))>;
5860 def : Pat<(v4f32 (bitconvert (v8f16 FPR128:$src))),
5861 (v4f32 (REV32v8i16 FPR128:$src))>;
5862 def : Pat<(v4f32 (bitconvert (v16i8 FPR128:$src))),
5863 (v4f32 (REV32v16i8 FPR128:$src))>;
5864 def : Pat<(v4f32 (bitconvert (v2i64 FPR128:$src))),
5865 (v4f32 (REV64v4i32 FPR128:$src))>;
5866 def : Pat<(v4f32 (bitconvert (v2f64 FPR128:$src))),
5867 (v4f32 (REV64v4i32 FPR128:$src))>;
5869 def : Pat<(v4f32 (bitconvert (v4i32 FPR128:$src))), (v4f32 FPR128:$src)>;
5871 let Predicates = [IsLE] in {
5872 def : Pat<(v2i64 (bitconvert (f128 FPR128:$src))), (v2i64 FPR128:$src)>;
5873 def : Pat<(v2i64 (bitconvert (v4i32 FPR128:$src))), (v2i64 FPR128:$src)>;
5874 def : Pat<(v2i64 (bitconvert (v8i16 FPR128:$src))), (v2i64 FPR128:$src)>;
5875 def : Pat<(v2i64 (bitconvert (v16i8 FPR128:$src))), (v2i64 FPR128:$src)>;
5876 def : Pat<(v2i64 (bitconvert (v4f32 FPR128:$src))), (v2i64 FPR128:$src)>;
5877 def : Pat<(v2i64 (bitconvert (v8f16 FPR128:$src))), (v2i64 FPR128:$src)>;
5879 let Predicates = [IsBE] in {
5880 def : Pat<(v2i64 (bitconvert (f128 FPR128:$src))),
5881 (v2i64 (EXTv16i8 FPR128:$src,
5882 FPR128:$src, (i32 8)))>;
5883 def : Pat<(v2i64 (bitconvert (v4i32 FPR128:$src))),
5884 (v2i64 (REV64v4i32 FPR128:$src))>;
5885 def : Pat<(v2i64 (bitconvert (v8i16 FPR128:$src))),
5886 (v2i64 (REV64v8i16 FPR128:$src))>;
5887 def : Pat<(v2i64 (bitconvert (v16i8 FPR128:$src))),
5888 (v2i64 (REV64v16i8 FPR128:$src))>;
5889 def : Pat<(v2i64 (bitconvert (v4f32 FPR128:$src))),
5890 (v2i64 (REV64v4i32 FPR128:$src))>;
5891 def : Pat<(v2i64 (bitconvert (v8f16 FPR128:$src))),
5892 (v2i64 (REV64v8i16 FPR128:$src))>;
5894 def : Pat<(v2i64 (bitconvert (v2f64 FPR128:$src))), (v2i64 FPR128:$src)>;
5896 let Predicates = [IsLE] in {
5897 def : Pat<(v4i32 (bitconvert (f128 FPR128:$src))), (v4i32 FPR128:$src)>;
5898 def : Pat<(v4i32 (bitconvert (v2i64 FPR128:$src))), (v4i32 FPR128:$src)>;
5899 def : Pat<(v4i32 (bitconvert (v8i16 FPR128:$src))), (v4i32 FPR128:$src)>;
5900 def : Pat<(v4i32 (bitconvert (v16i8 FPR128:$src))), (v4i32 FPR128:$src)>;
5901 def : Pat<(v4i32 (bitconvert (v2f64 FPR128:$src))), (v4i32 FPR128:$src)>;
5902 def : Pat<(v4i32 (bitconvert (v8f16 FPR128:$src))), (v4i32 FPR128:$src)>;
5904 let Predicates = [IsBE] in {
5905 def : Pat<(v4i32 (bitconvert (f128 FPR128:$src))),
5906 (v4i32 (EXTv16i8 (REV64v4i32 FPR128:$src),
5907 (REV64v4i32 FPR128:$src),
5909 def : Pat<(v4i32 (bitconvert (v2i64 FPR128:$src))),
5910 (v4i32 (REV64v4i32 FPR128:$src))>;
5911 def : Pat<(v4i32 (bitconvert (v8i16 FPR128:$src))),
5912 (v4i32 (REV32v8i16 FPR128:$src))>;
5913 def : Pat<(v4i32 (bitconvert (v16i8 FPR128:$src))),
5914 (v4i32 (REV32v16i8 FPR128:$src))>;
5915 def : Pat<(v4i32 (bitconvert (v2f64 FPR128:$src))),
5916 (v4i32 (REV64v4i32 FPR128:$src))>;
5917 def : Pat<(v4i32 (bitconvert (v8f16 FPR128:$src))),
5918 (v4i32 (REV32v8i16 FPR128:$src))>;
5920 def : Pat<(v4i32 (bitconvert (v4f32 FPR128:$src))), (v4i32 FPR128:$src)>;
5922 let Predicates = [IsLE] in {
5923 def : Pat<(v8i16 (bitconvert (f128 FPR128:$src))), (v8i16 FPR128:$src)>;
5924 def : Pat<(v8i16 (bitconvert (v2i64 FPR128:$src))), (v8i16 FPR128:$src)>;
5925 def : Pat<(v8i16 (bitconvert (v4i32 FPR128:$src))), (v8i16 FPR128:$src)>;
5926 def : Pat<(v8i16 (bitconvert (v16i8 FPR128:$src))), (v8i16 FPR128:$src)>;
5927 def : Pat<(v8i16 (bitconvert (v2f64 FPR128:$src))), (v8i16 FPR128:$src)>;
5928 def : Pat<(v8i16 (bitconvert (v4f32 FPR128:$src))), (v8i16 FPR128:$src)>;
5929 def : Pat<(v8i16 (bitconvert (v8f16 FPR128:$src))), (v8i16 FPR128:$src)>;
5931 let Predicates = [IsBE] in {
5932 def : Pat<(v8i16 (bitconvert (f128 FPR128:$src))),
5933 (v8i16 (EXTv16i8 (REV64v8i16 FPR128:$src),
5934 (REV64v8i16 FPR128:$src),
5936 def : Pat<(v8i16 (bitconvert (v2i64 FPR128:$src))),
5937 (v8i16 (REV64v8i16 FPR128:$src))>;
5938 def : Pat<(v8i16 (bitconvert (v4i32 FPR128:$src))),
5939 (v8i16 (REV32v8i16 FPR128:$src))>;
5940 def : Pat<(v8i16 (bitconvert (v16i8 FPR128:$src))),
5941 (v8i16 (REV16v16i8 FPR128:$src))>;
5942 def : Pat<(v8i16 (bitconvert (v2f64 FPR128:$src))),
5943 (v8i16 (REV64v8i16 FPR128:$src))>;
5944 def : Pat<(v8i16 (bitconvert (v4f32 FPR128:$src))),
5945 (v8i16 (REV32v8i16 FPR128:$src))>;
5946 def : Pat<(v8i16 (bitconvert (v8f16 FPR128:$src))),
5947 (v8i16 (REV32v8i16 FPR128:$src))>;
5950 let Predicates = [IsLE] in {
5951 def : Pat<(v8f16 (bitconvert (f128 FPR128:$src))), (v8f16 FPR128:$src)>;
5952 def : Pat<(v8f16 (bitconvert (v2i64 FPR128:$src))), (v8f16 FPR128:$src)>;
5953 def : Pat<(v8f16 (bitconvert (v4i32 FPR128:$src))), (v8f16 FPR128:$src)>;
5954 def : Pat<(v8f16 (bitconvert (v8i16 FPR128:$src))), (v8f16 FPR128:$src)>;
5955 def : Pat<(v8f16 (bitconvert (v16i8 FPR128:$src))), (v8f16 FPR128:$src)>;
5956 def : Pat<(v8f16 (bitconvert (v2f64 FPR128:$src))), (v8f16 FPR128:$src)>;
5957 def : Pat<(v8f16 (bitconvert (v4f32 FPR128:$src))), (v8f16 FPR128:$src)>;
5959 let Predicates = [IsBE] in {
5960 def : Pat<(v8f16 (bitconvert (f128 FPR128:$src))),
5961 (v8f16 (EXTv16i8 (REV64v8i16 FPR128:$src),
5962 (REV64v8i16 FPR128:$src),
5964 def : Pat<(v8f16 (bitconvert (v2i64 FPR128:$src))),
5965 (v8f16 (REV64v8i16 FPR128:$src))>;
5966 def : Pat<(v8f16 (bitconvert (v4i32 FPR128:$src))),
5967 (v8f16 (REV32v8i16 FPR128:$src))>;
5968 def : Pat<(v8f16 (bitconvert (v8i16 FPR128:$src))),
5969 (v8f16 (REV64v8i16 FPR128:$src))>;
5970 def : Pat<(v8f16 (bitconvert (v16i8 FPR128:$src))),
5971 (v8f16 (REV16v16i8 FPR128:$src))>;
5972 def : Pat<(v8f16 (bitconvert (v2f64 FPR128:$src))),
5973 (v8f16 (REV64v8i16 FPR128:$src))>;
5974 def : Pat<(v8f16 (bitconvert (v4f32 FPR128:$src))),
5975 (v8f16 (REV32v8i16 FPR128:$src))>;
5978 let Predicates = [IsLE] in {
5979 def : Pat<(v16i8 (bitconvert (f128 FPR128:$src))), (v16i8 FPR128:$src)>;
5980 def : Pat<(v16i8 (bitconvert (v2i64 FPR128:$src))), (v16i8 FPR128:$src)>;
5981 def : Pat<(v16i8 (bitconvert (v4i32 FPR128:$src))), (v16i8 FPR128:$src)>;
5982 def : Pat<(v16i8 (bitconvert (v8i16 FPR128:$src))), (v16i8 FPR128:$src)>;
5983 def : Pat<(v16i8 (bitconvert (v2f64 FPR128:$src))), (v16i8 FPR128:$src)>;
5984 def : Pat<(v16i8 (bitconvert (v4f32 FPR128:$src))), (v16i8 FPR128:$src)>;
5985 def : Pat<(v16i8 (bitconvert (v8f16 FPR128:$src))), (v16i8 FPR128:$src)>;
5987 let Predicates = [IsBE] in {
5988 def : Pat<(v16i8 (bitconvert (f128 FPR128:$src))),
5989 (v16i8 (EXTv16i8 (REV64v16i8 FPR128:$src),
5990 (REV64v16i8 FPR128:$src),
5992 def : Pat<(v16i8 (bitconvert (v2i64 FPR128:$src))),
5993 (v16i8 (REV64v16i8 FPR128:$src))>;
5994 def : Pat<(v16i8 (bitconvert (v4i32 FPR128:$src))),
5995 (v16i8 (REV32v16i8 FPR128:$src))>;
5996 def : Pat<(v16i8 (bitconvert (v8i16 FPR128:$src))),
5997 (v16i8 (REV16v16i8 FPR128:$src))>;
5998 def : Pat<(v16i8 (bitconvert (v2f64 FPR128:$src))),
5999 (v16i8 (REV64v16i8 FPR128:$src))>;
6000 def : Pat<(v16i8 (bitconvert (v4f32 FPR128:$src))),
6001 (v16i8 (REV32v16i8 FPR128:$src))>;
6002 def : Pat<(v16i8 (bitconvert (v8f16 FPR128:$src))),
6003 (v16i8 (REV16v16i8 FPR128:$src))>;
6006 def : Pat<(v4i16 (extract_subvector V128:$Rn, (i64 0))),
6007 (EXTRACT_SUBREG V128:$Rn, dsub)>;
6008 def : Pat<(v8i8 (extract_subvector V128:$Rn, (i64 0))),
6009 (EXTRACT_SUBREG V128:$Rn, dsub)>;
6010 def : Pat<(v2f32 (extract_subvector V128:$Rn, (i64 0))),
6011 (EXTRACT_SUBREG V128:$Rn, dsub)>;
6012 def : Pat<(v4f16 (extract_subvector V128:$Rn, (i64 0))),
6013 (EXTRACT_SUBREG V128:$Rn, dsub)>;
6014 def : Pat<(v2i32 (extract_subvector V128:$Rn, (i64 0))),
6015 (EXTRACT_SUBREG V128:$Rn, dsub)>;
6016 def : Pat<(v1i64 (extract_subvector V128:$Rn, (i64 0))),
6017 (EXTRACT_SUBREG V128:$Rn, dsub)>;
6018 def : Pat<(v1f64 (extract_subvector V128:$Rn, (i64 0))),
6019 (EXTRACT_SUBREG V128:$Rn, dsub)>;
6021 def : Pat<(v8i8 (extract_subvector (v16i8 FPR128:$Rn), (i64 1))),
6022 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
6023 def : Pat<(v4i16 (extract_subvector (v8i16 FPR128:$Rn), (i64 1))),
6024 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
6025 def : Pat<(v2i32 (extract_subvector (v4i32 FPR128:$Rn), (i64 1))),
6026 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
6027 def : Pat<(v1i64 (extract_subvector (v2i64 FPR128:$Rn), (i64 1))),
6028 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
6030 // A 64-bit subvector insert to the first 128-bit vector position
6031 // is a subregister copy that needs no instruction.
6032 def : Pat<(insert_subvector undef, (v1i64 FPR64:$src), (i32 0)),
6033 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
6034 def : Pat<(insert_subvector undef, (v1f64 FPR64:$src), (i32 0)),
6035 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
6036 def : Pat<(insert_subvector undef, (v2i32 FPR64:$src), (i32 0)),
6037 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
6038 def : Pat<(insert_subvector undef, (v2f32 FPR64:$src), (i32 0)),
6039 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
6040 def : Pat<(insert_subvector undef, (v4i16 FPR64:$src), (i32 0)),
6041 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
6042 def : Pat<(insert_subvector undef, (v4f16 FPR64:$src), (i32 0)),
6043 (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
6044 def : Pat<(insert_subvector undef, (v8i8 FPR64:$src), (i32 0)),
6045 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
6047 // Use pair-wise add instructions when summing up the lanes for v2f64, v2i64
6049 def : Pat<(i64 (add (vector_extract (v2i64 FPR128:$Rn), (i64 0)),
6050 (vector_extract (v2i64 FPR128:$Rn), (i64 1)))),
6051 (i64 (ADDPv2i64p (v2i64 FPR128:$Rn)))>;
6052 def : Pat<(f64 (fadd (vector_extract (v2f64 FPR128:$Rn), (i64 0)),
6053 (vector_extract (v2f64 FPR128:$Rn), (i64 1)))),
6054 (f64 (FADDPv2i64p (v2f64 FPR128:$Rn)))>;
6055 // vector_extract on 64-bit vectors gets promoted to a 128 bit vector,
6056 // so we match on v4f32 here, not v2f32. This will also catch adding
6057 // the low two lanes of a true v4f32 vector.
6058 def : Pat<(fadd (vector_extract (v4f32 FPR128:$Rn), (i64 0)),
6059 (vector_extract (v4f32 FPR128:$Rn), (i64 1))),
6060 (f32 (FADDPv2i32p (EXTRACT_SUBREG FPR128:$Rn, dsub)))>;
6062 // Scalar 64-bit shifts in FPR64 registers.
6063 def : Pat<(i64 (int_aarch64_neon_sshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
6064 (SSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
6065 def : Pat<(i64 (int_aarch64_neon_ushl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
6066 (USHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
6067 def : Pat<(i64 (int_aarch64_neon_srshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
6068 (SRSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
6069 def : Pat<(i64 (int_aarch64_neon_urshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
6070 (URSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
6072 // Patterns for nontemporal/no-allocate stores.
6073 // We have to resort to tricks to turn a single-input store into a store pair,
6074 // because there is no single-input nontemporal store, only STNP.
6075 let Predicates = [IsLE] in {
6076 let AddedComplexity = 15 in {
6077 class NTStore128Pat<ValueType VT> :
6078 Pat<(nontemporalstore (VT FPR128:$Rt),
6079 (am_indexed7s64 GPR64sp:$Rn, simm7s8:$offset)),
6080 (STNPDi (EXTRACT_SUBREG FPR128:$Rt, dsub),
6081 (CPYi64 FPR128:$Rt, (i64 1)),
6082 GPR64sp:$Rn, simm7s8:$offset)>;
6084 def : NTStore128Pat<v2i64>;
6085 def : NTStore128Pat<v4i32>;
6086 def : NTStore128Pat<v8i16>;
6087 def : NTStore128Pat<v16i8>;
6089 class NTStore64Pat<ValueType VT> :
6090 Pat<(nontemporalstore (VT FPR64:$Rt),
6091 (am_indexed7s32 GPR64sp:$Rn, simm7s4:$offset)),
6092 (STNPSi (EXTRACT_SUBREG FPR64:$Rt, ssub),
6093 (CPYi32 (SUBREG_TO_REG (i64 0), FPR64:$Rt, dsub), (i64 1)),
6094 GPR64sp:$Rn, simm7s4:$offset)>;
6096 // FIXME: Shouldn't v1f64 loads/stores be promoted to v1i64?
6097 def : NTStore64Pat<v1f64>;
6098 def : NTStore64Pat<v1i64>;
6099 def : NTStore64Pat<v2i32>;
6100 def : NTStore64Pat<v4i16>;
6101 def : NTStore64Pat<v8i8>;
6103 def : Pat<(nontemporalstore GPR64:$Rt,
6104 (am_indexed7s32 GPR64sp:$Rn, simm7s4:$offset)),
6105 (STNPWi (EXTRACT_SUBREG GPR64:$Rt, sub_32),
6106 (EXTRACT_SUBREG (UBFMXri GPR64:$Rt, 32, 63), sub_32),
6107 GPR64sp:$Rn, simm7s4:$offset)>;
6108 } // AddedComplexity=10
6109 } // Predicates = [IsLE]
6111 // Tail call return handling. These are all compiler pseudo-instructions,
6112 // so no encoding information or anything like that.
6113 let isCall = 1, isTerminator = 1, isReturn = 1, isBarrier = 1, Uses = [SP] in {
6114 def TCRETURNdi : Pseudo<(outs), (ins i64imm:$dst, i32imm:$FPDiff), []>,
6115 Sched<[WriteBrReg]>;
6116 def TCRETURNri : Pseudo<(outs), (ins tcGPR64:$dst, i32imm:$FPDiff), []>,
6117 Sched<[WriteBrReg]>;
6120 def : Pat<(AArch64tcret tcGPR64:$dst, (i32 timm:$FPDiff)),
6121 (TCRETURNri tcGPR64:$dst, imm:$FPDiff)>;
6122 def : Pat<(AArch64tcret tglobaladdr:$dst, (i32 timm:$FPDiff)),
6123 (TCRETURNdi texternalsym:$dst, imm:$FPDiff)>;
6124 def : Pat<(AArch64tcret texternalsym:$dst, (i32 timm:$FPDiff)),
6125 (TCRETURNdi texternalsym:$dst, imm:$FPDiff)>;
6127 include "AArch64InstrAtomics.td"