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 HasFPARMv8 : Predicate<"Subtarget->hasFPARMv8()">,
20 AssemblerPredicate<"FeatureFPARMv8", "fp-armv8">;
21 def HasNEON : Predicate<"Subtarget->hasNEON()">,
22 AssemblerPredicate<"FeatureNEON", "neon">;
23 def HasCrypto : Predicate<"Subtarget->hasCrypto()">,
24 AssemblerPredicate<"FeatureCrypto", "crypto">;
25 def HasCRC : Predicate<"Subtarget->hasCRC()">,
26 AssemblerPredicate<"FeatureCRC", "crc">;
27 def IsLE : Predicate<"Subtarget->isLittleEndian()">;
28 def IsBE : Predicate<"!Subtarget->isLittleEndian()">;
29 def IsCyclone : Predicate<"Subtarget->isCyclone()">;
31 //===----------------------------------------------------------------------===//
32 // AArch64-specific DAG Nodes.
35 // SDTBinaryArithWithFlagsOut - RES1, FLAGS = op LHS, RHS
36 def SDTBinaryArithWithFlagsOut : SDTypeProfile<2, 2,
39 SDTCisInt<0>, SDTCisVT<1, i32>]>;
41 // SDTBinaryArithWithFlagsIn - RES1, FLAGS = op LHS, RHS, FLAGS
42 def SDTBinaryArithWithFlagsIn : SDTypeProfile<1, 3,
48 // SDTBinaryArithWithFlagsInOut - RES1, FLAGS = op LHS, RHS, FLAGS
49 def SDTBinaryArithWithFlagsInOut : SDTypeProfile<2, 3,
56 def SDT_AArch64Brcond : SDTypeProfile<0, 3,
57 [SDTCisVT<0, OtherVT>, SDTCisVT<1, i32>,
59 def SDT_AArch64cbz : SDTypeProfile<0, 2, [SDTCisInt<0>, SDTCisVT<1, OtherVT>]>;
60 def SDT_AArch64tbz : SDTypeProfile<0, 3, [SDTCisInt<0>, SDTCisInt<1>,
61 SDTCisVT<2, OtherVT>]>;
64 def SDT_AArch64CSel : SDTypeProfile<1, 4,
69 def SDT_AArch64CCMP : SDTypeProfile<1, 5,
76 def SDT_AArch64FCCMP : SDTypeProfile<1, 5,
83 def SDT_AArch64FCmp : SDTypeProfile<0, 2,
86 def SDT_AArch64Dup : SDTypeProfile<1, 1, [SDTCisVec<0>]>;
87 def SDT_AArch64DupLane : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisInt<2>]>;
88 def SDT_AArch64Zip : SDTypeProfile<1, 2, [SDTCisVec<0>,
91 def SDT_AArch64MOVIedit : SDTypeProfile<1, 1, [SDTCisInt<1>]>;
92 def SDT_AArch64MOVIshift : SDTypeProfile<1, 2, [SDTCisInt<1>, SDTCisInt<2>]>;
93 def SDT_AArch64vecimm : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0,1>,
94 SDTCisInt<2>, SDTCisInt<3>]>;
95 def SDT_AArch64UnaryVec: SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0,1>]>;
96 def SDT_AArch64ExtVec: SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0,1>,
97 SDTCisSameAs<0,2>, SDTCisInt<3>]>;
98 def SDT_AArch64vshift : SDTypeProfile<1, 2, [SDTCisSameAs<0,1>, SDTCisInt<2>]>;
100 def SDT_AArch64unvec : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0,1>]>;
101 def SDT_AArch64fcmpz : SDTypeProfile<1, 1, []>;
102 def SDT_AArch64fcmp : SDTypeProfile<1, 2, [SDTCisSameAs<1,2>]>;
103 def SDT_AArch64binvec : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisSameAs<0,1>,
105 def SDT_AArch64trivec : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0,1>,
108 def SDT_AArch64TCRET : SDTypeProfile<0, 2, [SDTCisPtrTy<0>]>;
109 def SDT_AArch64PREFETCH : SDTypeProfile<0, 2, [SDTCisVT<0, i32>, SDTCisPtrTy<1>]>;
111 def SDT_AArch64ITOF : SDTypeProfile<1, 1, [SDTCisFP<0>, SDTCisSameAs<0,1>]>;
113 def SDT_AArch64TLSDescCall : SDTypeProfile<0, -2, [SDTCisPtrTy<0>,
116 // Generates the general dynamic sequences, i.e.
117 // adrp x0, :tlsdesc:var
118 // ldr x1, [x0, #:tlsdesc_lo12:var]
119 // add x0, x0, #:tlsdesc_lo12:var
123 // (the TPIDR_EL0 offset is put directly in X0, hence no "result" here)
124 // number of operands (the variable)
125 def SDT_AArch64TLSDescCallSeq : SDTypeProfile<0,1,
128 def SDT_AArch64WrapperLarge : SDTypeProfile<1, 4,
129 [SDTCisVT<0, i64>, SDTCisVT<1, i32>,
130 SDTCisSameAs<1, 2>, SDTCisSameAs<1, 3>,
131 SDTCisSameAs<1, 4>]>;
135 def AArch64adrp : SDNode<"AArch64ISD::ADRP", SDTIntUnaryOp, []>;
136 def AArch64addlow : SDNode<"AArch64ISD::ADDlow", SDTIntBinOp, []>;
137 def AArch64LOADgot : SDNode<"AArch64ISD::LOADgot", SDTIntUnaryOp>;
138 def AArch64callseq_start : SDNode<"ISD::CALLSEQ_START",
139 SDCallSeqStart<[ SDTCisVT<0, i32> ]>,
140 [SDNPHasChain, SDNPOutGlue]>;
141 def AArch64callseq_end : SDNode<"ISD::CALLSEQ_END",
142 SDCallSeqEnd<[ SDTCisVT<0, i32>,
144 [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue]>;
145 def AArch64call : SDNode<"AArch64ISD::CALL",
146 SDTypeProfile<0, -1, [SDTCisPtrTy<0>]>,
147 [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue,
149 def AArch64brcond : SDNode<"AArch64ISD::BRCOND", SDT_AArch64Brcond,
151 def AArch64cbz : SDNode<"AArch64ISD::CBZ", SDT_AArch64cbz,
153 def AArch64cbnz : SDNode<"AArch64ISD::CBNZ", SDT_AArch64cbz,
155 def AArch64tbz : SDNode<"AArch64ISD::TBZ", SDT_AArch64tbz,
157 def AArch64tbnz : SDNode<"AArch64ISD::TBNZ", SDT_AArch64tbz,
161 def AArch64csel : SDNode<"AArch64ISD::CSEL", SDT_AArch64CSel>;
162 def AArch64csinv : SDNode<"AArch64ISD::CSINV", SDT_AArch64CSel>;
163 def AArch64csneg : SDNode<"AArch64ISD::CSNEG", SDT_AArch64CSel>;
164 def AArch64csinc : SDNode<"AArch64ISD::CSINC", SDT_AArch64CSel>;
165 def AArch64retflag : SDNode<"AArch64ISD::RET_FLAG", SDTNone,
166 [SDNPHasChain, SDNPOptInGlue, SDNPVariadic]>;
167 def AArch64adc : SDNode<"AArch64ISD::ADC", SDTBinaryArithWithFlagsIn >;
168 def AArch64sbc : SDNode<"AArch64ISD::SBC", SDTBinaryArithWithFlagsIn>;
169 def AArch64add_flag : SDNode<"AArch64ISD::ADDS", SDTBinaryArithWithFlagsOut,
171 def AArch64sub_flag : SDNode<"AArch64ISD::SUBS", SDTBinaryArithWithFlagsOut>;
172 def AArch64and_flag : SDNode<"AArch64ISD::ANDS", SDTBinaryArithWithFlagsOut,
174 def AArch64adc_flag : SDNode<"AArch64ISD::ADCS", SDTBinaryArithWithFlagsInOut>;
175 def AArch64sbc_flag : SDNode<"AArch64ISD::SBCS", SDTBinaryArithWithFlagsInOut>;
177 def AArch64ccmp : SDNode<"AArch64ISD::CCMP", SDT_AArch64CCMP>;
178 def AArch64ccmn : SDNode<"AArch64ISD::CCMN", SDT_AArch64CCMP>;
179 def AArch64fccmp : SDNode<"AArch64ISD::FCCMP", SDT_AArch64FCCMP>;
181 def AArch64threadpointer : SDNode<"AArch64ISD::THREAD_POINTER", SDTPtrLeaf>;
183 def AArch64fcmp : SDNode<"AArch64ISD::FCMP", SDT_AArch64FCmp>;
185 def AArch64fmax : SDNode<"AArch64ISD::FMAX", SDTFPBinOp>;
186 def AArch64fmin : SDNode<"AArch64ISD::FMIN", SDTFPBinOp>;
188 def AArch64dup : SDNode<"AArch64ISD::DUP", SDT_AArch64Dup>;
189 def AArch64duplane8 : SDNode<"AArch64ISD::DUPLANE8", SDT_AArch64DupLane>;
190 def AArch64duplane16 : SDNode<"AArch64ISD::DUPLANE16", SDT_AArch64DupLane>;
191 def AArch64duplane32 : SDNode<"AArch64ISD::DUPLANE32", SDT_AArch64DupLane>;
192 def AArch64duplane64 : SDNode<"AArch64ISD::DUPLANE64", SDT_AArch64DupLane>;
194 def AArch64zip1 : SDNode<"AArch64ISD::ZIP1", SDT_AArch64Zip>;
195 def AArch64zip2 : SDNode<"AArch64ISD::ZIP2", SDT_AArch64Zip>;
196 def AArch64uzp1 : SDNode<"AArch64ISD::UZP1", SDT_AArch64Zip>;
197 def AArch64uzp2 : SDNode<"AArch64ISD::UZP2", SDT_AArch64Zip>;
198 def AArch64trn1 : SDNode<"AArch64ISD::TRN1", SDT_AArch64Zip>;
199 def AArch64trn2 : SDNode<"AArch64ISD::TRN2", SDT_AArch64Zip>;
201 def AArch64movi_edit : SDNode<"AArch64ISD::MOVIedit", SDT_AArch64MOVIedit>;
202 def AArch64movi_shift : SDNode<"AArch64ISD::MOVIshift", SDT_AArch64MOVIshift>;
203 def AArch64movi_msl : SDNode<"AArch64ISD::MOVImsl", SDT_AArch64MOVIshift>;
204 def AArch64mvni_shift : SDNode<"AArch64ISD::MVNIshift", SDT_AArch64MOVIshift>;
205 def AArch64mvni_msl : SDNode<"AArch64ISD::MVNImsl", SDT_AArch64MOVIshift>;
206 def AArch64movi : SDNode<"AArch64ISD::MOVI", SDT_AArch64MOVIedit>;
207 def AArch64fmov : SDNode<"AArch64ISD::FMOV", SDT_AArch64MOVIedit>;
209 def AArch64rev16 : SDNode<"AArch64ISD::REV16", SDT_AArch64UnaryVec>;
210 def AArch64rev32 : SDNode<"AArch64ISD::REV32", SDT_AArch64UnaryVec>;
211 def AArch64rev64 : SDNode<"AArch64ISD::REV64", SDT_AArch64UnaryVec>;
212 def AArch64ext : SDNode<"AArch64ISD::EXT", SDT_AArch64ExtVec>;
214 def AArch64vashr : SDNode<"AArch64ISD::VASHR", SDT_AArch64vshift>;
215 def AArch64vlshr : SDNode<"AArch64ISD::VLSHR", SDT_AArch64vshift>;
216 def AArch64vshl : SDNode<"AArch64ISD::VSHL", SDT_AArch64vshift>;
217 def AArch64sqshli : SDNode<"AArch64ISD::SQSHL_I", SDT_AArch64vshift>;
218 def AArch64uqshli : SDNode<"AArch64ISD::UQSHL_I", SDT_AArch64vshift>;
219 def AArch64sqshlui : SDNode<"AArch64ISD::SQSHLU_I", SDT_AArch64vshift>;
220 def AArch64srshri : SDNode<"AArch64ISD::SRSHR_I", SDT_AArch64vshift>;
221 def AArch64urshri : SDNode<"AArch64ISD::URSHR_I", SDT_AArch64vshift>;
223 def AArch64not: SDNode<"AArch64ISD::NOT", SDT_AArch64unvec>;
224 def AArch64bit: SDNode<"AArch64ISD::BIT", SDT_AArch64trivec>;
225 def AArch64bsl: SDNode<"AArch64ISD::BSL", SDT_AArch64trivec>;
227 def AArch64cmeq: SDNode<"AArch64ISD::CMEQ", SDT_AArch64binvec>;
228 def AArch64cmge: SDNode<"AArch64ISD::CMGE", SDT_AArch64binvec>;
229 def AArch64cmgt: SDNode<"AArch64ISD::CMGT", SDT_AArch64binvec>;
230 def AArch64cmhi: SDNode<"AArch64ISD::CMHI", SDT_AArch64binvec>;
231 def AArch64cmhs: SDNode<"AArch64ISD::CMHS", SDT_AArch64binvec>;
233 def AArch64fcmeq: SDNode<"AArch64ISD::FCMEQ", SDT_AArch64fcmp>;
234 def AArch64fcmge: SDNode<"AArch64ISD::FCMGE", SDT_AArch64fcmp>;
235 def AArch64fcmgt: SDNode<"AArch64ISD::FCMGT", SDT_AArch64fcmp>;
237 def AArch64cmeqz: SDNode<"AArch64ISD::CMEQz", SDT_AArch64unvec>;
238 def AArch64cmgez: SDNode<"AArch64ISD::CMGEz", SDT_AArch64unvec>;
239 def AArch64cmgtz: SDNode<"AArch64ISD::CMGTz", SDT_AArch64unvec>;
240 def AArch64cmlez: SDNode<"AArch64ISD::CMLEz", SDT_AArch64unvec>;
241 def AArch64cmltz: SDNode<"AArch64ISD::CMLTz", SDT_AArch64unvec>;
242 def AArch64cmtst : PatFrag<(ops node:$LHS, node:$RHS),
243 (AArch64not (AArch64cmeqz (and node:$LHS, node:$RHS)))>;
245 def AArch64fcmeqz: SDNode<"AArch64ISD::FCMEQz", SDT_AArch64fcmpz>;
246 def AArch64fcmgez: SDNode<"AArch64ISD::FCMGEz", SDT_AArch64fcmpz>;
247 def AArch64fcmgtz: SDNode<"AArch64ISD::FCMGTz", SDT_AArch64fcmpz>;
248 def AArch64fcmlez: SDNode<"AArch64ISD::FCMLEz", SDT_AArch64fcmpz>;
249 def AArch64fcmltz: SDNode<"AArch64ISD::FCMLTz", SDT_AArch64fcmpz>;
251 def AArch64bici: SDNode<"AArch64ISD::BICi", SDT_AArch64vecimm>;
252 def AArch64orri: SDNode<"AArch64ISD::ORRi", SDT_AArch64vecimm>;
254 def AArch64neg : SDNode<"AArch64ISD::NEG", SDT_AArch64unvec>;
256 def AArch64tcret: SDNode<"AArch64ISD::TC_RETURN", SDT_AArch64TCRET,
257 [SDNPHasChain, SDNPOptInGlue, SDNPVariadic]>;
259 def AArch64Prefetch : SDNode<"AArch64ISD::PREFETCH", SDT_AArch64PREFETCH,
260 [SDNPHasChain, SDNPSideEffect]>;
262 def AArch64sitof: SDNode<"AArch64ISD::SITOF", SDT_AArch64ITOF>;
263 def AArch64uitof: SDNode<"AArch64ISD::UITOF", SDT_AArch64ITOF>;
265 def AArch64tlsdesc_callseq : SDNode<"AArch64ISD::TLSDESC_CALLSEQ",
266 SDT_AArch64TLSDescCallSeq,
267 [SDNPInGlue, SDNPOutGlue, SDNPHasChain,
271 def AArch64WrapperLarge : SDNode<"AArch64ISD::WrapperLarge",
272 SDT_AArch64WrapperLarge>;
274 def AArch64NvCast : SDNode<"AArch64ISD::NVCAST", SDTUnaryOp>;
276 def SDT_AArch64mull : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
277 SDTCisSameAs<1, 2>]>;
278 def AArch64smull : SDNode<"AArch64ISD::SMULL", SDT_AArch64mull>;
279 def AArch64umull : SDNode<"AArch64ISD::UMULL", SDT_AArch64mull>;
281 def AArch64saddv : SDNode<"AArch64ISD::SADDV", SDT_AArch64UnaryVec>;
282 def AArch64uaddv : SDNode<"AArch64ISD::UADDV", SDT_AArch64UnaryVec>;
283 def AArch64sminv : SDNode<"AArch64ISD::SMINV", SDT_AArch64UnaryVec>;
284 def AArch64uminv : SDNode<"AArch64ISD::UMINV", SDT_AArch64UnaryVec>;
285 def AArch64smaxv : SDNode<"AArch64ISD::SMAXV", SDT_AArch64UnaryVec>;
286 def AArch64umaxv : SDNode<"AArch64ISD::UMAXV", SDT_AArch64UnaryVec>;
288 //===----------------------------------------------------------------------===//
290 //===----------------------------------------------------------------------===//
292 // AArch64 Instruction Predicate Definitions.
294 def HasZCZ : Predicate<"Subtarget->hasZeroCycleZeroing()">;
295 def NoZCZ : Predicate<"!Subtarget->hasZeroCycleZeroing()">;
296 def IsDarwin : Predicate<"Subtarget->isTargetDarwin()">;
297 def IsNotDarwin: Predicate<"!Subtarget->isTargetDarwin()">;
298 def ForCodeSize : Predicate<"ForCodeSize">;
299 def NotForCodeSize : Predicate<"!ForCodeSize">;
301 include "AArch64InstrFormats.td"
303 //===----------------------------------------------------------------------===//
305 //===----------------------------------------------------------------------===//
306 // Miscellaneous instructions.
307 //===----------------------------------------------------------------------===//
309 let Defs = [SP], Uses = [SP], hasSideEffects = 1, isCodeGenOnly = 1 in {
310 def ADJCALLSTACKDOWN : Pseudo<(outs), (ins i32imm:$amt),
311 [(AArch64callseq_start timm:$amt)]>;
312 def ADJCALLSTACKUP : Pseudo<(outs), (ins i32imm:$amt1, i32imm:$amt2),
313 [(AArch64callseq_end timm:$amt1, timm:$amt2)]>;
314 } // Defs = [SP], Uses = [SP], hasSideEffects = 1, isCodeGenOnly = 1
316 let isReMaterializable = 1, isCodeGenOnly = 1 in {
317 // FIXME: The following pseudo instructions are only needed because remat
318 // cannot handle multiple instructions. When that changes, they can be
319 // removed, along with the AArch64Wrapper node.
321 let AddedComplexity = 10 in
322 def LOADgot : Pseudo<(outs GPR64:$dst), (ins i64imm:$addr),
323 [(set GPR64:$dst, (AArch64LOADgot tglobaladdr:$addr))]>,
326 // The MOVaddr instruction should match only when the add is not folded
327 // into a load or store address.
329 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
330 [(set GPR64:$dst, (AArch64addlow (AArch64adrp tglobaladdr:$hi),
331 tglobaladdr:$low))]>,
332 Sched<[WriteAdrAdr]>;
334 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
335 [(set GPR64:$dst, (AArch64addlow (AArch64adrp tjumptable:$hi),
337 Sched<[WriteAdrAdr]>;
339 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
340 [(set GPR64:$dst, (AArch64addlow (AArch64adrp tconstpool:$hi),
342 Sched<[WriteAdrAdr]>;
344 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
345 [(set GPR64:$dst, (AArch64addlow (AArch64adrp tblockaddress:$hi),
346 tblockaddress:$low))]>,
347 Sched<[WriteAdrAdr]>;
349 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
350 [(set GPR64:$dst, (AArch64addlow (AArch64adrp tglobaltlsaddr:$hi),
351 tglobaltlsaddr:$low))]>,
352 Sched<[WriteAdrAdr]>;
354 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
355 [(set GPR64:$dst, (AArch64addlow (AArch64adrp texternalsym:$hi),
356 texternalsym:$low))]>,
357 Sched<[WriteAdrAdr]>;
359 } // isReMaterializable, isCodeGenOnly
361 def : Pat<(AArch64LOADgot tglobaltlsaddr:$addr),
362 (LOADgot tglobaltlsaddr:$addr)>;
364 def : Pat<(AArch64LOADgot texternalsym:$addr),
365 (LOADgot texternalsym:$addr)>;
367 def : Pat<(AArch64LOADgot tconstpool:$addr),
368 (LOADgot tconstpool:$addr)>;
370 //===----------------------------------------------------------------------===//
371 // System instructions.
372 //===----------------------------------------------------------------------===//
374 def HINT : HintI<"hint">;
375 def : InstAlias<"nop", (HINT 0b000)>;
376 def : InstAlias<"yield",(HINT 0b001)>;
377 def : InstAlias<"wfe", (HINT 0b010)>;
378 def : InstAlias<"wfi", (HINT 0b011)>;
379 def : InstAlias<"sev", (HINT 0b100)>;
380 def : InstAlias<"sevl", (HINT 0b101)>;
382 // As far as LLVM is concerned this writes to the system's exclusive monitors.
383 let mayLoad = 1, mayStore = 1 in
384 def CLREX : CRmSystemI<imm0_15, 0b010, "clrex">;
386 // NOTE: ideally, this would have mayStore = 0, mayLoad = 0, but we cannot
387 // model patterns with sufficiently fine granularity.
388 let mayLoad = ?, mayStore = ? in {
389 def DMB : CRmSystemI<barrier_op, 0b101, "dmb",
390 [(int_aarch64_dmb (i32 imm32_0_15:$CRm))]>;
392 def DSB : CRmSystemI<barrier_op, 0b100, "dsb",
393 [(int_aarch64_dsb (i32 imm32_0_15:$CRm))]>;
395 def ISB : CRmSystemI<barrier_op, 0b110, "isb",
396 [(int_aarch64_isb (i32 imm32_0_15:$CRm))]>;
399 def : InstAlias<"clrex", (CLREX 0xf)>;
400 def : InstAlias<"isb", (ISB 0xf)>;
404 def MSRpstate: MSRpstateI;
406 // The thread pointer (on Linux, at least, where this has been implemented) is
408 def : Pat<(AArch64threadpointer), (MRS 0xde82)>;
410 // Generic system instructions
411 def SYSxt : SystemXtI<0, "sys">;
412 def SYSLxt : SystemLXtI<1, "sysl">;
414 def : InstAlias<"sys $op1, $Cn, $Cm, $op2",
415 (SYSxt imm0_7:$op1, sys_cr_op:$Cn,
416 sys_cr_op:$Cm, imm0_7:$op2, XZR)>;
418 //===----------------------------------------------------------------------===//
419 // Move immediate instructions.
420 //===----------------------------------------------------------------------===//
422 defm MOVK : InsertImmediate<0b11, "movk">;
423 defm MOVN : MoveImmediate<0b00, "movn">;
425 let PostEncoderMethod = "fixMOVZ" in
426 defm MOVZ : MoveImmediate<0b10, "movz">;
428 // First group of aliases covers an implicit "lsl #0".
429 def : InstAlias<"movk $dst, $imm", (MOVKWi GPR32:$dst, imm0_65535:$imm, 0)>;
430 def : InstAlias<"movk $dst, $imm", (MOVKXi GPR64:$dst, imm0_65535:$imm, 0)>;
431 def : InstAlias<"movn $dst, $imm", (MOVNWi GPR32:$dst, imm0_65535:$imm, 0)>;
432 def : InstAlias<"movn $dst, $imm", (MOVNXi GPR64:$dst, imm0_65535:$imm, 0)>;
433 def : InstAlias<"movz $dst, $imm", (MOVZWi GPR32:$dst, imm0_65535:$imm, 0)>;
434 def : InstAlias<"movz $dst, $imm", (MOVZXi GPR64:$dst, imm0_65535:$imm, 0)>;
436 // Next, we have various ELF relocations with the ":XYZ_g0:sym" syntax.
437 def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movz_symbol_g3:$sym, 48)>;
438 def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movz_symbol_g2:$sym, 32)>;
439 def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movz_symbol_g1:$sym, 16)>;
440 def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movz_symbol_g0:$sym, 0)>;
442 def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movz_symbol_g3:$sym, 48)>;
443 def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movz_symbol_g2:$sym, 32)>;
444 def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movz_symbol_g1:$sym, 16)>;
445 def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movz_symbol_g0:$sym, 0)>;
447 def : InstAlias<"movk $Rd, $sym", (MOVKXi GPR64:$Rd, movk_symbol_g3:$sym, 48)>;
448 def : InstAlias<"movk $Rd, $sym", (MOVKXi GPR64:$Rd, movk_symbol_g2:$sym, 32)>;
449 def : InstAlias<"movk $Rd, $sym", (MOVKXi GPR64:$Rd, movk_symbol_g1:$sym, 16)>;
450 def : InstAlias<"movk $Rd, $sym", (MOVKXi GPR64:$Rd, movk_symbol_g0:$sym, 0)>;
452 def : InstAlias<"movz $Rd, $sym", (MOVZWi GPR32:$Rd, movz_symbol_g1:$sym, 16)>;
453 def : InstAlias<"movz $Rd, $sym", (MOVZWi GPR32:$Rd, movz_symbol_g0:$sym, 0)>;
455 def : InstAlias<"movn $Rd, $sym", (MOVNWi GPR32:$Rd, movz_symbol_g1:$sym, 16)>;
456 def : InstAlias<"movn $Rd, $sym", (MOVNWi GPR32:$Rd, movz_symbol_g0:$sym, 0)>;
458 def : InstAlias<"movk $Rd, $sym", (MOVKWi GPR32:$Rd, movk_symbol_g1:$sym, 16)>;
459 def : InstAlias<"movk $Rd, $sym", (MOVKWi GPR32:$Rd, movk_symbol_g0:$sym, 0)>;
461 // Final group of aliases covers true "mov $Rd, $imm" cases.
462 multiclass movw_mov_alias<string basename,Instruction INST, RegisterClass GPR,
463 int width, int shift> {
464 def _asmoperand : AsmOperandClass {
465 let Name = basename # width # "_lsl" # shift # "MovAlias";
466 let PredicateMethod = "is" # basename # "MovAlias<" # width # ", "
468 let RenderMethod = "add" # basename # "MovAliasOperands<" # shift # ">";
471 def _movimm : Operand<i32> {
472 let ParserMatchClass = !cast<AsmOperandClass>(NAME # "_asmoperand");
475 def : InstAlias<"mov $Rd, $imm",
476 (INST GPR:$Rd, !cast<Operand>(NAME # "_movimm"):$imm, shift)>;
479 defm : movw_mov_alias<"MOVZ", MOVZWi, GPR32, 32, 0>;
480 defm : movw_mov_alias<"MOVZ", MOVZWi, GPR32, 32, 16>;
482 defm : movw_mov_alias<"MOVZ", MOVZXi, GPR64, 64, 0>;
483 defm : movw_mov_alias<"MOVZ", MOVZXi, GPR64, 64, 16>;
484 defm : movw_mov_alias<"MOVZ", MOVZXi, GPR64, 64, 32>;
485 defm : movw_mov_alias<"MOVZ", MOVZXi, GPR64, 64, 48>;
487 defm : movw_mov_alias<"MOVN", MOVNWi, GPR32, 32, 0>;
488 defm : movw_mov_alias<"MOVN", MOVNWi, GPR32, 32, 16>;
490 defm : movw_mov_alias<"MOVN", MOVNXi, GPR64, 64, 0>;
491 defm : movw_mov_alias<"MOVN", MOVNXi, GPR64, 64, 16>;
492 defm : movw_mov_alias<"MOVN", MOVNXi, GPR64, 64, 32>;
493 defm : movw_mov_alias<"MOVN", MOVNXi, GPR64, 64, 48>;
495 let isReMaterializable = 1, isCodeGenOnly = 1, isMoveImm = 1,
496 isAsCheapAsAMove = 1 in {
497 // FIXME: The following pseudo instructions are only needed because remat
498 // cannot handle multiple instructions. When that changes, we can select
499 // directly to the real instructions and get rid of these pseudos.
502 : Pseudo<(outs GPR32:$dst), (ins i32imm:$src),
503 [(set GPR32:$dst, imm:$src)]>,
506 : Pseudo<(outs GPR64:$dst), (ins i64imm:$src),
507 [(set GPR64:$dst, imm:$src)]>,
509 } // isReMaterializable, isCodeGenOnly
511 // If possible, we want to use MOVi32imm even for 64-bit moves. This gives the
512 // eventual expansion code fewer bits to worry about getting right. Marshalling
513 // the types is a little tricky though:
514 def i64imm_32bit : ImmLeaf<i64, [{
515 return (Imm & 0xffffffffULL) == static_cast<uint64_t>(Imm);
518 def trunc_imm : SDNodeXForm<imm, [{
519 return CurDAG->getTargetConstant(N->getZExtValue(), SDLoc(N), MVT::i32);
522 def : Pat<(i64 i64imm_32bit:$src),
523 (SUBREG_TO_REG (i64 0), (MOVi32imm (trunc_imm imm:$src)), sub_32)>;
525 // Materialize FP constants via MOVi32imm/MOVi64imm (MachO large code model).
526 def bitcast_fpimm_to_i32 : SDNodeXForm<fpimm, [{
527 return CurDAG->getTargetConstant(
528 N->getValueAPF().bitcastToAPInt().getZExtValue(), SDLoc(N), MVT::i32);
531 def bitcast_fpimm_to_i64 : SDNodeXForm<fpimm, [{
532 return CurDAG->getTargetConstant(
533 N->getValueAPF().bitcastToAPInt().getZExtValue(), SDLoc(N), MVT::i64);
537 def : Pat<(f32 fpimm:$in),
538 (COPY_TO_REGCLASS (MOVi32imm (bitcast_fpimm_to_i32 f32:$in)), FPR32)>;
539 def : Pat<(f64 fpimm:$in),
540 (COPY_TO_REGCLASS (MOVi64imm (bitcast_fpimm_to_i64 f64:$in)), FPR64)>;
543 // Deal with the various forms of (ELF) large addressing with MOVZ/MOVK
545 def : Pat<(AArch64WrapperLarge tglobaladdr:$g3, tglobaladdr:$g2,
546 tglobaladdr:$g1, tglobaladdr:$g0),
547 (MOVKXi (MOVKXi (MOVKXi (MOVZXi tglobaladdr:$g3, 48),
548 tglobaladdr:$g2, 32),
549 tglobaladdr:$g1, 16),
550 tglobaladdr:$g0, 0)>;
552 def : Pat<(AArch64WrapperLarge tblockaddress:$g3, tblockaddress:$g2,
553 tblockaddress:$g1, tblockaddress:$g0),
554 (MOVKXi (MOVKXi (MOVKXi (MOVZXi tblockaddress:$g3, 48),
555 tblockaddress:$g2, 32),
556 tblockaddress:$g1, 16),
557 tblockaddress:$g0, 0)>;
559 def : Pat<(AArch64WrapperLarge tconstpool:$g3, tconstpool:$g2,
560 tconstpool:$g1, tconstpool:$g0),
561 (MOVKXi (MOVKXi (MOVKXi (MOVZXi tconstpool:$g3, 48),
566 def : Pat<(AArch64WrapperLarge tjumptable:$g3, tjumptable:$g2,
567 tjumptable:$g1, tjumptable:$g0),
568 (MOVKXi (MOVKXi (MOVKXi (MOVZXi tjumptable:$g3, 48),
574 //===----------------------------------------------------------------------===//
575 // Arithmetic instructions.
576 //===----------------------------------------------------------------------===//
578 // Add/subtract with carry.
579 defm ADC : AddSubCarry<0, "adc", "adcs", AArch64adc, AArch64adc_flag>;
580 defm SBC : AddSubCarry<1, "sbc", "sbcs", AArch64sbc, AArch64sbc_flag>;
582 def : InstAlias<"ngc $dst, $src", (SBCWr GPR32:$dst, WZR, GPR32:$src)>;
583 def : InstAlias<"ngc $dst, $src", (SBCXr GPR64:$dst, XZR, GPR64:$src)>;
584 def : InstAlias<"ngcs $dst, $src", (SBCSWr GPR32:$dst, WZR, GPR32:$src)>;
585 def : InstAlias<"ngcs $dst, $src", (SBCSXr GPR64:$dst, XZR, GPR64:$src)>;
588 defm ADD : AddSub<0, "add", add>;
589 defm SUB : AddSub<1, "sub">;
591 def : InstAlias<"mov $dst, $src",
592 (ADDWri GPR32sponly:$dst, GPR32sp:$src, 0, 0)>;
593 def : InstAlias<"mov $dst, $src",
594 (ADDWri GPR32sp:$dst, GPR32sponly:$src, 0, 0)>;
595 def : InstAlias<"mov $dst, $src",
596 (ADDXri GPR64sponly:$dst, GPR64sp:$src, 0, 0)>;
597 def : InstAlias<"mov $dst, $src",
598 (ADDXri GPR64sp:$dst, GPR64sponly:$src, 0, 0)>;
600 defm ADDS : AddSubS<0, "adds", AArch64add_flag, "cmn">;
601 defm SUBS : AddSubS<1, "subs", AArch64sub_flag, "cmp">;
603 // Use SUBS instead of SUB to enable CSE between SUBS and SUB.
604 def : Pat<(sub GPR32sp:$Rn, addsub_shifted_imm32:$imm),
605 (SUBSWri GPR32sp:$Rn, addsub_shifted_imm32:$imm)>;
606 def : Pat<(sub GPR64sp:$Rn, addsub_shifted_imm64:$imm),
607 (SUBSXri GPR64sp:$Rn, addsub_shifted_imm64:$imm)>;
608 def : Pat<(sub GPR32:$Rn, GPR32:$Rm),
609 (SUBSWrr GPR32:$Rn, GPR32:$Rm)>;
610 def : Pat<(sub GPR64:$Rn, GPR64:$Rm),
611 (SUBSXrr GPR64:$Rn, GPR64:$Rm)>;
612 def : Pat<(sub GPR32:$Rn, arith_shifted_reg32:$Rm),
613 (SUBSWrs GPR32:$Rn, arith_shifted_reg32:$Rm)>;
614 def : Pat<(sub GPR64:$Rn, arith_shifted_reg64:$Rm),
615 (SUBSXrs GPR64:$Rn, arith_shifted_reg64:$Rm)>;
616 def : Pat<(sub GPR32sp:$R2, arith_extended_reg32<i32>:$R3),
617 (SUBSWrx GPR32sp:$R2, arith_extended_reg32<i32>:$R3)>;
618 def : Pat<(sub GPR64sp:$R2, arith_extended_reg32to64<i64>:$R3),
619 (SUBSXrx GPR64sp:$R2, arith_extended_reg32to64<i64>:$R3)>;
621 // Because of the immediate format for add/sub-imm instructions, the
622 // expression (add x, -1) must be transformed to (SUB{W,X}ri x, 1).
623 // These patterns capture that transformation.
624 let AddedComplexity = 1 in {
625 def : Pat<(add GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
626 (SUBSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
627 def : Pat<(add GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
628 (SUBSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
629 def : Pat<(sub GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
630 (ADDWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
631 def : Pat<(sub GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
632 (ADDXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
635 // Because of the immediate format for add/sub-imm instructions, the
636 // expression (add x, -1) must be transformed to (SUB{W,X}ri x, 1).
637 // These patterns capture that transformation.
638 let AddedComplexity = 1 in {
639 def : Pat<(AArch64add_flag GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
640 (SUBSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
641 def : Pat<(AArch64add_flag GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
642 (SUBSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
643 def : Pat<(AArch64sub_flag GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
644 (ADDSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
645 def : Pat<(AArch64sub_flag GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
646 (ADDSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
649 def : InstAlias<"neg $dst, $src", (SUBWrs GPR32:$dst, WZR, GPR32:$src, 0), 3>;
650 def : InstAlias<"neg $dst, $src", (SUBXrs GPR64:$dst, XZR, GPR64:$src, 0), 3>;
651 def : InstAlias<"neg $dst, $src$shift",
652 (SUBWrs GPR32:$dst, WZR, GPR32:$src, arith_shift32:$shift), 2>;
653 def : InstAlias<"neg $dst, $src$shift",
654 (SUBXrs GPR64:$dst, XZR, GPR64:$src, arith_shift64:$shift), 2>;
656 def : InstAlias<"negs $dst, $src", (SUBSWrs GPR32:$dst, WZR, GPR32:$src, 0), 3>;
657 def : InstAlias<"negs $dst, $src", (SUBSXrs GPR64:$dst, XZR, GPR64:$src, 0), 3>;
658 def : InstAlias<"negs $dst, $src$shift",
659 (SUBSWrs GPR32:$dst, WZR, GPR32:$src, arith_shift32:$shift), 2>;
660 def : InstAlias<"negs $dst, $src$shift",
661 (SUBSXrs GPR64:$dst, XZR, GPR64:$src, arith_shift64:$shift), 2>;
664 // Unsigned/Signed divide
665 defm UDIV : Div<0, "udiv", udiv>;
666 defm SDIV : Div<1, "sdiv", sdiv>;
667 let isCodeGenOnly = 1 in {
668 defm UDIV_Int : Div<0, "udiv", int_aarch64_udiv>;
669 defm SDIV_Int : Div<1, "sdiv", int_aarch64_sdiv>;
673 defm ASRV : Shift<0b10, "asr", sra>;
674 defm LSLV : Shift<0b00, "lsl", shl>;
675 defm LSRV : Shift<0b01, "lsr", srl>;
676 defm RORV : Shift<0b11, "ror", rotr>;
678 def : ShiftAlias<"asrv", ASRVWr, GPR32>;
679 def : ShiftAlias<"asrv", ASRVXr, GPR64>;
680 def : ShiftAlias<"lslv", LSLVWr, GPR32>;
681 def : ShiftAlias<"lslv", LSLVXr, GPR64>;
682 def : ShiftAlias<"lsrv", LSRVWr, GPR32>;
683 def : ShiftAlias<"lsrv", LSRVXr, GPR64>;
684 def : ShiftAlias<"rorv", RORVWr, GPR32>;
685 def : ShiftAlias<"rorv", RORVXr, GPR64>;
688 let AddedComplexity = 7 in {
689 defm MADD : MulAccum<0, "madd", add>;
690 defm MSUB : MulAccum<1, "msub", sub>;
692 def : Pat<(i32 (mul GPR32:$Rn, GPR32:$Rm)),
693 (MADDWrrr GPR32:$Rn, GPR32:$Rm, WZR)>;
694 def : Pat<(i64 (mul GPR64:$Rn, GPR64:$Rm)),
695 (MADDXrrr GPR64:$Rn, GPR64:$Rm, XZR)>;
697 def : Pat<(i32 (ineg (mul GPR32:$Rn, GPR32:$Rm))),
698 (MSUBWrrr GPR32:$Rn, GPR32:$Rm, WZR)>;
699 def : Pat<(i64 (ineg (mul GPR64:$Rn, GPR64:$Rm))),
700 (MSUBXrrr GPR64:$Rn, GPR64:$Rm, XZR)>;
701 def : Pat<(i32 (mul (ineg GPR32:$Rn), GPR32:$Rm)),
702 (MSUBWrrr GPR32:$Rn, GPR32:$Rm, WZR)>;
703 def : Pat<(i64 (mul (ineg GPR64:$Rn), GPR64:$Rm)),
704 (MSUBXrrr GPR64:$Rn, GPR64:$Rm, XZR)>;
705 } // AddedComplexity = 7
707 let AddedComplexity = 5 in {
708 def SMADDLrrr : WideMulAccum<0, 0b001, "smaddl", add, sext>;
709 def SMSUBLrrr : WideMulAccum<1, 0b001, "smsubl", sub, sext>;
710 def UMADDLrrr : WideMulAccum<0, 0b101, "umaddl", add, zext>;
711 def UMSUBLrrr : WideMulAccum<1, 0b101, "umsubl", sub, zext>;
713 def : Pat<(i64 (mul (sext GPR32:$Rn), (sext GPR32:$Rm))),
714 (SMADDLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
715 def : Pat<(i64 (mul (zext GPR32:$Rn), (zext GPR32:$Rm))),
716 (UMADDLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
718 def : Pat<(i64 (ineg (mul (sext GPR32:$Rn), (sext GPR32:$Rm)))),
719 (SMSUBLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
720 def : Pat<(i64 (ineg (mul (zext GPR32:$Rn), (zext GPR32:$Rm)))),
721 (UMSUBLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
722 } // AddedComplexity = 5
724 def : MulAccumWAlias<"mul", MADDWrrr>;
725 def : MulAccumXAlias<"mul", MADDXrrr>;
726 def : MulAccumWAlias<"mneg", MSUBWrrr>;
727 def : MulAccumXAlias<"mneg", MSUBXrrr>;
728 def : WideMulAccumAlias<"smull", SMADDLrrr>;
729 def : WideMulAccumAlias<"smnegl", SMSUBLrrr>;
730 def : WideMulAccumAlias<"umull", UMADDLrrr>;
731 def : WideMulAccumAlias<"umnegl", UMSUBLrrr>;
734 def SMULHrr : MulHi<0b010, "smulh", mulhs>;
735 def UMULHrr : MulHi<0b110, "umulh", mulhu>;
738 def CRC32Brr : BaseCRC32<0, 0b00, 0, GPR32, int_aarch64_crc32b, "crc32b">;
739 def CRC32Hrr : BaseCRC32<0, 0b01, 0, GPR32, int_aarch64_crc32h, "crc32h">;
740 def CRC32Wrr : BaseCRC32<0, 0b10, 0, GPR32, int_aarch64_crc32w, "crc32w">;
741 def CRC32Xrr : BaseCRC32<1, 0b11, 0, GPR64, int_aarch64_crc32x, "crc32x">;
743 def CRC32CBrr : BaseCRC32<0, 0b00, 1, GPR32, int_aarch64_crc32cb, "crc32cb">;
744 def CRC32CHrr : BaseCRC32<0, 0b01, 1, GPR32, int_aarch64_crc32ch, "crc32ch">;
745 def CRC32CWrr : BaseCRC32<0, 0b10, 1, GPR32, int_aarch64_crc32cw, "crc32cw">;
746 def CRC32CXrr : BaseCRC32<1, 0b11, 1, GPR64, int_aarch64_crc32cx, "crc32cx">;
749 //===----------------------------------------------------------------------===//
750 // Logical instructions.
751 //===----------------------------------------------------------------------===//
754 defm ANDS : LogicalImmS<0b11, "ands", AArch64and_flag, "bics">;
755 defm AND : LogicalImm<0b00, "and", and, "bic">;
756 defm EOR : LogicalImm<0b10, "eor", xor, "eon">;
757 defm ORR : LogicalImm<0b01, "orr", or, "orn">;
759 // FIXME: these aliases *are* canonical sometimes (when movz can't be
760 // used). Actually, it seems to be working right now, but putting logical_immXX
761 // here is a bit dodgy on the AsmParser side too.
762 def : InstAlias<"mov $dst, $imm", (ORRWri GPR32sp:$dst, WZR,
763 logical_imm32:$imm), 0>;
764 def : InstAlias<"mov $dst, $imm", (ORRXri GPR64sp:$dst, XZR,
765 logical_imm64:$imm), 0>;
769 defm ANDS : LogicalRegS<0b11, 0, "ands", AArch64and_flag>;
770 defm BICS : LogicalRegS<0b11, 1, "bics",
771 BinOpFrag<(AArch64and_flag node:$LHS, (not node:$RHS))>>;
772 defm AND : LogicalReg<0b00, 0, "and", and>;
773 defm BIC : LogicalReg<0b00, 1, "bic",
774 BinOpFrag<(and node:$LHS, (not node:$RHS))>>;
775 defm EON : LogicalReg<0b10, 1, "eon",
776 BinOpFrag<(xor node:$LHS, (not node:$RHS))>>;
777 defm EOR : LogicalReg<0b10, 0, "eor", xor>;
778 defm ORN : LogicalReg<0b01, 1, "orn",
779 BinOpFrag<(or node:$LHS, (not node:$RHS))>>;
780 defm ORR : LogicalReg<0b01, 0, "orr", or>;
782 def : InstAlias<"mov $dst, $src", (ORRWrs GPR32:$dst, WZR, GPR32:$src, 0), 2>;
783 def : InstAlias<"mov $dst, $src", (ORRXrs GPR64:$dst, XZR, GPR64:$src, 0), 2>;
785 def : InstAlias<"mvn $Wd, $Wm", (ORNWrs GPR32:$Wd, WZR, GPR32:$Wm, 0), 3>;
786 def : InstAlias<"mvn $Xd, $Xm", (ORNXrs GPR64:$Xd, XZR, GPR64:$Xm, 0), 3>;
788 def : InstAlias<"mvn $Wd, $Wm$sh",
789 (ORNWrs GPR32:$Wd, WZR, GPR32:$Wm, logical_shift32:$sh), 2>;
790 def : InstAlias<"mvn $Xd, $Xm$sh",
791 (ORNXrs GPR64:$Xd, XZR, GPR64:$Xm, logical_shift64:$sh), 2>;
793 def : InstAlias<"tst $src1, $src2",
794 (ANDSWri WZR, GPR32:$src1, logical_imm32:$src2), 2>;
795 def : InstAlias<"tst $src1, $src2",
796 (ANDSXri XZR, GPR64:$src1, logical_imm64:$src2), 2>;
798 def : InstAlias<"tst $src1, $src2",
799 (ANDSWrs WZR, GPR32:$src1, GPR32:$src2, 0), 3>;
800 def : InstAlias<"tst $src1, $src2",
801 (ANDSXrs XZR, GPR64:$src1, GPR64:$src2, 0), 3>;
803 def : InstAlias<"tst $src1, $src2$sh",
804 (ANDSWrs WZR, GPR32:$src1, GPR32:$src2, logical_shift32:$sh), 2>;
805 def : InstAlias<"tst $src1, $src2$sh",
806 (ANDSXrs XZR, GPR64:$src1, GPR64:$src2, logical_shift64:$sh), 2>;
809 def : Pat<(not GPR32:$Wm), (ORNWrr WZR, GPR32:$Wm)>;
810 def : Pat<(not GPR64:$Xm), (ORNXrr XZR, GPR64:$Xm)>;
813 //===----------------------------------------------------------------------===//
814 // One operand data processing instructions.
815 //===----------------------------------------------------------------------===//
817 defm CLS : OneOperandData<0b101, "cls">;
818 defm CLZ : OneOperandData<0b100, "clz", ctlz>;
819 defm RBIT : OneOperandData<0b000, "rbit">;
821 def : Pat<(int_aarch64_rbit GPR32:$Rn), (RBITWr $Rn)>;
822 def : Pat<(int_aarch64_rbit GPR64:$Rn), (RBITXr $Rn)>;
824 def REV16Wr : OneWRegData<0b001, "rev16",
825 UnOpFrag<(rotr (bswap node:$LHS), (i64 16))>>;
826 def REV16Xr : OneXRegData<0b001, "rev16", null_frag>;
828 def : Pat<(cttz GPR32:$Rn),
829 (CLZWr (RBITWr GPR32:$Rn))>;
830 def : Pat<(cttz GPR64:$Rn),
831 (CLZXr (RBITXr GPR64:$Rn))>;
832 def : Pat<(ctlz (or (shl (xor (sra GPR32:$Rn, (i64 31)), GPR32:$Rn), (i64 1)),
835 def : Pat<(ctlz (or (shl (xor (sra GPR64:$Rn, (i64 63)), GPR64:$Rn), (i64 1)),
839 // Unlike the other one operand instructions, the instructions with the "rev"
840 // mnemonic do *not* just different in the size bit, but actually use different
841 // opcode bits for the different sizes.
842 def REVWr : OneWRegData<0b010, "rev", bswap>;
843 def REVXr : OneXRegData<0b011, "rev", bswap>;
844 def REV32Xr : OneXRegData<0b010, "rev32",
845 UnOpFrag<(rotr (bswap node:$LHS), (i64 32))>>;
847 // The bswap commutes with the rotr so we want a pattern for both possible
849 def : Pat<(bswap (rotr GPR32:$Rn, (i64 16))), (REV16Wr GPR32:$Rn)>;
850 def : Pat<(bswap (rotr GPR64:$Rn, (i64 32))), (REV32Xr GPR64:$Rn)>;
852 //===----------------------------------------------------------------------===//
853 // Bitfield immediate extraction instruction.
854 //===----------------------------------------------------------------------===//
855 let hasSideEffects = 0 in
856 defm EXTR : ExtractImm<"extr">;
857 def : InstAlias<"ror $dst, $src, $shift",
858 (EXTRWrri GPR32:$dst, GPR32:$src, GPR32:$src, imm0_31:$shift)>;
859 def : InstAlias<"ror $dst, $src, $shift",
860 (EXTRXrri GPR64:$dst, GPR64:$src, GPR64:$src, imm0_63:$shift)>;
862 def : Pat<(rotr GPR32:$Rn, (i64 imm0_31:$imm)),
863 (EXTRWrri GPR32:$Rn, GPR32:$Rn, imm0_31:$imm)>;
864 def : Pat<(rotr GPR64:$Rn, (i64 imm0_63:$imm)),
865 (EXTRXrri GPR64:$Rn, GPR64:$Rn, imm0_63:$imm)>;
867 //===----------------------------------------------------------------------===//
868 // Other bitfield immediate instructions.
869 //===----------------------------------------------------------------------===//
870 let hasSideEffects = 0 in {
871 defm BFM : BitfieldImmWith2RegArgs<0b01, "bfm">;
872 defm SBFM : BitfieldImm<0b00, "sbfm">;
873 defm UBFM : BitfieldImm<0b10, "ubfm">;
876 def i32shift_a : Operand<i64>, SDNodeXForm<imm, [{
877 uint64_t enc = (32 - N->getZExtValue()) & 0x1f;
878 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
881 def i32shift_b : Operand<i64>, SDNodeXForm<imm, [{
882 uint64_t enc = 31 - N->getZExtValue();
883 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
886 // min(7, 31 - shift_amt)
887 def i32shift_sext_i8 : Operand<i64>, SDNodeXForm<imm, [{
888 uint64_t enc = 31 - N->getZExtValue();
889 enc = enc > 7 ? 7 : enc;
890 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
893 // min(15, 31 - shift_amt)
894 def i32shift_sext_i16 : Operand<i64>, SDNodeXForm<imm, [{
895 uint64_t enc = 31 - N->getZExtValue();
896 enc = enc > 15 ? 15 : enc;
897 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
900 def i64shift_a : Operand<i64>, SDNodeXForm<imm, [{
901 uint64_t enc = (64 - N->getZExtValue()) & 0x3f;
902 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
905 def i64shift_b : Operand<i64>, SDNodeXForm<imm, [{
906 uint64_t enc = 63 - N->getZExtValue();
907 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
910 // min(7, 63 - shift_amt)
911 def i64shift_sext_i8 : Operand<i64>, SDNodeXForm<imm, [{
912 uint64_t enc = 63 - N->getZExtValue();
913 enc = enc > 7 ? 7 : enc;
914 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
917 // min(15, 63 - shift_amt)
918 def i64shift_sext_i16 : Operand<i64>, SDNodeXForm<imm, [{
919 uint64_t enc = 63 - N->getZExtValue();
920 enc = enc > 15 ? 15 : enc;
921 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
924 // min(31, 63 - shift_amt)
925 def i64shift_sext_i32 : Operand<i64>, SDNodeXForm<imm, [{
926 uint64_t enc = 63 - N->getZExtValue();
927 enc = enc > 31 ? 31 : enc;
928 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
931 def : Pat<(shl GPR32:$Rn, (i64 imm0_31:$imm)),
932 (UBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
933 (i64 (i32shift_b imm0_31:$imm)))>;
934 def : Pat<(shl GPR64:$Rn, (i64 imm0_63:$imm)),
935 (UBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
936 (i64 (i64shift_b imm0_63:$imm)))>;
938 let AddedComplexity = 10 in {
939 def : Pat<(sra GPR32:$Rn, (i64 imm0_31:$imm)),
940 (SBFMWri GPR32:$Rn, imm0_31:$imm, 31)>;
941 def : Pat<(sra GPR64:$Rn, (i64 imm0_63:$imm)),
942 (SBFMXri GPR64:$Rn, imm0_63:$imm, 63)>;
945 def : InstAlias<"asr $dst, $src, $shift",
946 (SBFMWri GPR32:$dst, GPR32:$src, imm0_31:$shift, 31)>;
947 def : InstAlias<"asr $dst, $src, $shift",
948 (SBFMXri GPR64:$dst, GPR64:$src, imm0_63:$shift, 63)>;
949 def : InstAlias<"sxtb $dst, $src", (SBFMWri GPR32:$dst, GPR32:$src, 0, 7)>;
950 def : InstAlias<"sxtb $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 7)>;
951 def : InstAlias<"sxth $dst, $src", (SBFMWri GPR32:$dst, GPR32:$src, 0, 15)>;
952 def : InstAlias<"sxth $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 15)>;
953 def : InstAlias<"sxtw $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 31)>;
955 def : Pat<(srl GPR32:$Rn, (i64 imm0_31:$imm)),
956 (UBFMWri GPR32:$Rn, imm0_31:$imm, 31)>;
957 def : Pat<(srl GPR64:$Rn, (i64 imm0_63:$imm)),
958 (UBFMXri GPR64:$Rn, imm0_63:$imm, 63)>;
960 def : InstAlias<"lsr $dst, $src, $shift",
961 (UBFMWri GPR32:$dst, GPR32:$src, imm0_31:$shift, 31)>;
962 def : InstAlias<"lsr $dst, $src, $shift",
963 (UBFMXri GPR64:$dst, GPR64:$src, imm0_63:$shift, 63)>;
964 def : InstAlias<"uxtb $dst, $src", (UBFMWri GPR32:$dst, GPR32:$src, 0, 7)>;
965 def : InstAlias<"uxtb $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 7)>;
966 def : InstAlias<"uxth $dst, $src", (UBFMWri GPR32:$dst, GPR32:$src, 0, 15)>;
967 def : InstAlias<"uxth $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 15)>;
968 def : InstAlias<"uxtw $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 31)>;
970 //===----------------------------------------------------------------------===//
971 // Conditional comparison instructions.
972 //===----------------------------------------------------------------------===//
973 defm CCMN : CondComparison<0, "ccmn", AArch64ccmn>;
974 defm CCMP : CondComparison<1, "ccmp", AArch64ccmp>;
976 //===----------------------------------------------------------------------===//
977 // Conditional select instructions.
978 //===----------------------------------------------------------------------===//
979 defm CSEL : CondSelect<0, 0b00, "csel">;
981 def inc : PatFrag<(ops node:$in), (add node:$in, 1)>;
982 defm CSINC : CondSelectOp<0, 0b01, "csinc", inc>;
983 defm CSINV : CondSelectOp<1, 0b00, "csinv", not>;
984 defm CSNEG : CondSelectOp<1, 0b01, "csneg", ineg>;
986 def : Pat<(AArch64csinv GPR32:$tval, GPR32:$fval, (i32 imm:$cc), NZCV),
987 (CSINVWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
988 def : Pat<(AArch64csinv GPR64:$tval, GPR64:$fval, (i32 imm:$cc), NZCV),
989 (CSINVXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
990 def : Pat<(AArch64csneg GPR32:$tval, GPR32:$fval, (i32 imm:$cc), NZCV),
991 (CSNEGWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
992 def : Pat<(AArch64csneg GPR64:$tval, GPR64:$fval, (i32 imm:$cc), NZCV),
993 (CSNEGXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
994 def : Pat<(AArch64csinc GPR32:$tval, GPR32:$fval, (i32 imm:$cc), NZCV),
995 (CSINCWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
996 def : Pat<(AArch64csinc GPR64:$tval, GPR64:$fval, (i32 imm:$cc), NZCV),
997 (CSINCXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
999 def : Pat<(AArch64csel (i32 0), (i32 1), (i32 imm:$cc), NZCV),
1000 (CSINCWr WZR, WZR, (i32 imm:$cc))>;
1001 def : Pat<(AArch64csel (i64 0), (i64 1), (i32 imm:$cc), NZCV),
1002 (CSINCXr XZR, XZR, (i32 imm:$cc))>;
1003 def : Pat<(AArch64csel (i32 0), (i32 -1), (i32 imm:$cc), NZCV),
1004 (CSINVWr WZR, WZR, (i32 imm:$cc))>;
1005 def : Pat<(AArch64csel (i64 0), (i64 -1), (i32 imm:$cc), NZCV),
1006 (CSINVXr XZR, XZR, (i32 imm:$cc))>;
1008 // The inverse of the condition code from the alias instruction is what is used
1009 // in the aliased instruction. The parser all ready inverts the condition code
1010 // for these aliases.
1011 def : InstAlias<"cset $dst, $cc",
1012 (CSINCWr GPR32:$dst, WZR, WZR, inv_ccode:$cc)>;
1013 def : InstAlias<"cset $dst, $cc",
1014 (CSINCXr GPR64:$dst, XZR, XZR, inv_ccode:$cc)>;
1016 def : InstAlias<"csetm $dst, $cc",
1017 (CSINVWr GPR32:$dst, WZR, WZR, inv_ccode:$cc)>;
1018 def : InstAlias<"csetm $dst, $cc",
1019 (CSINVXr GPR64:$dst, XZR, XZR, inv_ccode:$cc)>;
1021 def : InstAlias<"cinc $dst, $src, $cc",
1022 (CSINCWr GPR32:$dst, GPR32:$src, GPR32:$src, inv_ccode:$cc)>;
1023 def : InstAlias<"cinc $dst, $src, $cc",
1024 (CSINCXr GPR64:$dst, GPR64:$src, GPR64:$src, inv_ccode:$cc)>;
1026 def : InstAlias<"cinv $dst, $src, $cc",
1027 (CSINVWr GPR32:$dst, GPR32:$src, GPR32:$src, inv_ccode:$cc)>;
1028 def : InstAlias<"cinv $dst, $src, $cc",
1029 (CSINVXr GPR64:$dst, GPR64:$src, GPR64:$src, inv_ccode:$cc)>;
1031 def : InstAlias<"cneg $dst, $src, $cc",
1032 (CSNEGWr GPR32:$dst, GPR32:$src, GPR32:$src, inv_ccode:$cc)>;
1033 def : InstAlias<"cneg $dst, $src, $cc",
1034 (CSNEGXr GPR64:$dst, GPR64:$src, GPR64:$src, inv_ccode:$cc)>;
1036 //===----------------------------------------------------------------------===//
1037 // PC-relative instructions.
1038 //===----------------------------------------------------------------------===//
1039 let isReMaterializable = 1 in {
1040 let hasSideEffects = 0, mayStore = 0, mayLoad = 0 in {
1041 def ADR : ADRI<0, "adr", adrlabel, []>;
1042 } // hasSideEffects = 0
1044 def ADRP : ADRI<1, "adrp", adrplabel,
1045 [(set GPR64:$Xd, (AArch64adrp tglobaladdr:$label))]>;
1046 } // isReMaterializable = 1
1048 // page address of a constant pool entry, block address
1049 def : Pat<(AArch64adrp tconstpool:$cp), (ADRP tconstpool:$cp)>;
1050 def : Pat<(AArch64adrp tblockaddress:$cp), (ADRP tblockaddress:$cp)>;
1052 //===----------------------------------------------------------------------===//
1053 // Unconditional branch (register) instructions.
1054 //===----------------------------------------------------------------------===//
1056 let isReturn = 1, isTerminator = 1, isBarrier = 1 in {
1057 def RET : BranchReg<0b0010, "ret", []>;
1058 def DRPS : SpecialReturn<0b0101, "drps">;
1059 def ERET : SpecialReturn<0b0100, "eret">;
1060 } // isReturn = 1, isTerminator = 1, isBarrier = 1
1062 // Default to the LR register.
1063 def : InstAlias<"ret", (RET LR)>;
1065 let isCall = 1, Defs = [LR], Uses = [SP] in {
1066 def BLR : BranchReg<0b0001, "blr", [(AArch64call GPR64:$Rn)]>;
1069 let isBranch = 1, isTerminator = 1, isBarrier = 1, isIndirectBranch = 1 in {
1070 def BR : BranchReg<0b0000, "br", [(brind GPR64:$Rn)]>;
1071 } // isBranch, isTerminator, isBarrier, isIndirectBranch
1073 // Create a separate pseudo-instruction for codegen to use so that we don't
1074 // flag lr as used in every function. It'll be restored before the RET by the
1075 // epilogue if it's legitimately used.
1076 def RET_ReallyLR : Pseudo<(outs), (ins), [(AArch64retflag)]> {
1077 let isTerminator = 1;
1082 // This is a directive-like pseudo-instruction. The purpose is to insert an
1083 // R_AARCH64_TLSDESC_CALL relocation at the offset of the following instruction
1084 // (which in the usual case is a BLR).
1085 let hasSideEffects = 1 in
1086 def TLSDESCCALL : Pseudo<(outs), (ins i64imm:$sym), []> {
1087 let AsmString = ".tlsdesccall $sym";
1090 // FIXME: maybe the scratch register used shouldn't be fixed to X1?
1091 // FIXME: can "hasSideEffects be dropped?
1092 let isCall = 1, Defs = [LR, X0, X1], hasSideEffects = 1,
1093 isCodeGenOnly = 1 in
1095 : Pseudo<(outs), (ins i64imm:$sym),
1096 [(AArch64tlsdesc_callseq tglobaltlsaddr:$sym)]>;
1097 def : Pat<(AArch64tlsdesc_callseq texternalsym:$sym),
1098 (TLSDESC_CALLSEQ texternalsym:$sym)>;
1100 //===----------------------------------------------------------------------===//
1101 // Conditional branch (immediate) instruction.
1102 //===----------------------------------------------------------------------===//
1103 def Bcc : BranchCond;
1105 //===----------------------------------------------------------------------===//
1106 // Compare-and-branch instructions.
1107 //===----------------------------------------------------------------------===//
1108 defm CBZ : CmpBranch<0, "cbz", AArch64cbz>;
1109 defm CBNZ : CmpBranch<1, "cbnz", AArch64cbnz>;
1111 //===----------------------------------------------------------------------===//
1112 // Test-bit-and-branch instructions.
1113 //===----------------------------------------------------------------------===//
1114 defm TBZ : TestBranch<0, "tbz", AArch64tbz>;
1115 defm TBNZ : TestBranch<1, "tbnz", AArch64tbnz>;
1117 //===----------------------------------------------------------------------===//
1118 // Unconditional branch (immediate) instructions.
1119 //===----------------------------------------------------------------------===//
1120 let isBranch = 1, isTerminator = 1, isBarrier = 1 in {
1121 def B : BranchImm<0, "b", [(br bb:$addr)]>;
1122 } // isBranch, isTerminator, isBarrier
1124 let isCall = 1, Defs = [LR], Uses = [SP] in {
1125 def BL : CallImm<1, "bl", [(AArch64call tglobaladdr:$addr)]>;
1127 def : Pat<(AArch64call texternalsym:$func), (BL texternalsym:$func)>;
1129 //===----------------------------------------------------------------------===//
1130 // Exception generation instructions.
1131 //===----------------------------------------------------------------------===//
1132 def BRK : ExceptionGeneration<0b001, 0b00, "brk">;
1133 def DCPS1 : ExceptionGeneration<0b101, 0b01, "dcps1">;
1134 def DCPS2 : ExceptionGeneration<0b101, 0b10, "dcps2">;
1135 def DCPS3 : ExceptionGeneration<0b101, 0b11, "dcps3">;
1136 def HLT : ExceptionGeneration<0b010, 0b00, "hlt">;
1137 def HVC : ExceptionGeneration<0b000, 0b10, "hvc">;
1138 def SMC : ExceptionGeneration<0b000, 0b11, "smc">;
1139 def SVC : ExceptionGeneration<0b000, 0b01, "svc">;
1141 // DCPSn defaults to an immediate operand of zero if unspecified.
1142 def : InstAlias<"dcps1", (DCPS1 0)>;
1143 def : InstAlias<"dcps2", (DCPS2 0)>;
1144 def : InstAlias<"dcps3", (DCPS3 0)>;
1146 //===----------------------------------------------------------------------===//
1147 // Load instructions.
1148 //===----------------------------------------------------------------------===//
1150 // Pair (indexed, offset)
1151 defm LDPW : LoadPairOffset<0b00, 0, GPR32, simm7s4, "ldp">;
1152 defm LDPX : LoadPairOffset<0b10, 0, GPR64, simm7s8, "ldp">;
1153 defm LDPS : LoadPairOffset<0b00, 1, FPR32, simm7s4, "ldp">;
1154 defm LDPD : LoadPairOffset<0b01, 1, FPR64, simm7s8, "ldp">;
1155 defm LDPQ : LoadPairOffset<0b10, 1, FPR128, simm7s16, "ldp">;
1157 defm LDPSW : LoadPairOffset<0b01, 0, GPR64, simm7s4, "ldpsw">;
1159 // Pair (pre-indexed)
1160 def LDPWpre : LoadPairPreIdx<0b00, 0, GPR32, simm7s4, "ldp">;
1161 def LDPXpre : LoadPairPreIdx<0b10, 0, GPR64, simm7s8, "ldp">;
1162 def LDPSpre : LoadPairPreIdx<0b00, 1, FPR32, simm7s4, "ldp">;
1163 def LDPDpre : LoadPairPreIdx<0b01, 1, FPR64, simm7s8, "ldp">;
1164 def LDPQpre : LoadPairPreIdx<0b10, 1, FPR128, simm7s16, "ldp">;
1166 def LDPSWpre : LoadPairPreIdx<0b01, 0, GPR64, simm7s4, "ldpsw">;
1168 // Pair (post-indexed)
1169 def LDPWpost : LoadPairPostIdx<0b00, 0, GPR32, simm7s4, "ldp">;
1170 def LDPXpost : LoadPairPostIdx<0b10, 0, GPR64, simm7s8, "ldp">;
1171 def LDPSpost : LoadPairPostIdx<0b00, 1, FPR32, simm7s4, "ldp">;
1172 def LDPDpost : LoadPairPostIdx<0b01, 1, FPR64, simm7s8, "ldp">;
1173 def LDPQpost : LoadPairPostIdx<0b10, 1, FPR128, simm7s16, "ldp">;
1175 def LDPSWpost : LoadPairPostIdx<0b01, 0, GPR64, simm7s4, "ldpsw">;
1178 // Pair (no allocate)
1179 defm LDNPW : LoadPairNoAlloc<0b00, 0, GPR32, simm7s4, "ldnp">;
1180 defm LDNPX : LoadPairNoAlloc<0b10, 0, GPR64, simm7s8, "ldnp">;
1181 defm LDNPS : LoadPairNoAlloc<0b00, 1, FPR32, simm7s4, "ldnp">;
1182 defm LDNPD : LoadPairNoAlloc<0b01, 1, FPR64, simm7s8, "ldnp">;
1183 defm LDNPQ : LoadPairNoAlloc<0b10, 1, FPR128, simm7s16, "ldnp">;
1186 // (register offset)
1190 defm LDRBB : Load8RO<0b00, 0, 0b01, GPR32, "ldrb", i32, zextloadi8>;
1191 defm LDRHH : Load16RO<0b01, 0, 0b01, GPR32, "ldrh", i32, zextloadi16>;
1192 defm LDRW : Load32RO<0b10, 0, 0b01, GPR32, "ldr", i32, load>;
1193 defm LDRX : Load64RO<0b11, 0, 0b01, GPR64, "ldr", i64, load>;
1196 defm LDRB : Load8RO<0b00, 1, 0b01, FPR8, "ldr", untyped, load>;
1197 defm LDRH : Load16RO<0b01, 1, 0b01, FPR16, "ldr", f16, load>;
1198 defm LDRS : Load32RO<0b10, 1, 0b01, FPR32, "ldr", f32, load>;
1199 defm LDRD : Load64RO<0b11, 1, 0b01, FPR64, "ldr", f64, load>;
1200 defm LDRQ : Load128RO<0b00, 1, 0b11, FPR128, "ldr", f128, load>;
1202 // Load sign-extended half-word
1203 defm LDRSHW : Load16RO<0b01, 0, 0b11, GPR32, "ldrsh", i32, sextloadi16>;
1204 defm LDRSHX : Load16RO<0b01, 0, 0b10, GPR64, "ldrsh", i64, sextloadi16>;
1206 // Load sign-extended byte
1207 defm LDRSBW : Load8RO<0b00, 0, 0b11, GPR32, "ldrsb", i32, sextloadi8>;
1208 defm LDRSBX : Load8RO<0b00, 0, 0b10, GPR64, "ldrsb", i64, sextloadi8>;
1210 // Load sign-extended word
1211 defm LDRSW : Load32RO<0b10, 0, 0b10, GPR64, "ldrsw", i64, sextloadi32>;
1214 defm PRFM : PrefetchRO<0b11, 0, 0b10, "prfm">;
1216 // For regular load, we do not have any alignment requirement.
1217 // Thus, it is safe to directly map the vector loads with interesting
1218 // addressing modes.
1219 // FIXME: We could do the same for bitconvert to floating point vectors.
1220 multiclass ScalToVecROLoadPat<ROAddrMode ro, SDPatternOperator loadop,
1221 ValueType ScalTy, ValueType VecTy,
1222 Instruction LOADW, Instruction LOADX,
1224 def : Pat<(VecTy (scalar_to_vector (ScalTy
1225 (loadop (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$offset))))),
1226 (INSERT_SUBREG (VecTy (IMPLICIT_DEF)),
1227 (LOADW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$offset),
1230 def : Pat<(VecTy (scalar_to_vector (ScalTy
1231 (loadop (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$offset))))),
1232 (INSERT_SUBREG (VecTy (IMPLICIT_DEF)),
1233 (LOADX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$offset),
1237 let AddedComplexity = 10 in {
1238 defm : ScalToVecROLoadPat<ro8, extloadi8, i32, v8i8, LDRBroW, LDRBroX, bsub>;
1239 defm : ScalToVecROLoadPat<ro8, extloadi8, i32, v16i8, LDRBroW, LDRBroX, bsub>;
1241 defm : ScalToVecROLoadPat<ro16, extloadi16, i32, v4i16, LDRHroW, LDRHroX, hsub>;
1242 defm : ScalToVecROLoadPat<ro16, extloadi16, i32, v8i16, LDRHroW, LDRHroX, hsub>;
1244 defm : ScalToVecROLoadPat<ro16, load, i32, v4f16, LDRHroW, LDRHroX, hsub>;
1245 defm : ScalToVecROLoadPat<ro16, load, i32, v8f16, LDRHroW, LDRHroX, hsub>;
1247 defm : ScalToVecROLoadPat<ro32, load, i32, v2i32, LDRSroW, LDRSroX, ssub>;
1248 defm : ScalToVecROLoadPat<ro32, load, i32, v4i32, LDRSroW, LDRSroX, ssub>;
1250 defm : ScalToVecROLoadPat<ro32, load, f32, v2f32, LDRSroW, LDRSroX, ssub>;
1251 defm : ScalToVecROLoadPat<ro32, load, f32, v4f32, LDRSroW, LDRSroX, ssub>;
1253 defm : ScalToVecROLoadPat<ro64, load, i64, v2i64, LDRDroW, LDRDroX, dsub>;
1255 defm : ScalToVecROLoadPat<ro64, load, f64, v2f64, LDRDroW, LDRDroX, dsub>;
1258 def : Pat <(v1i64 (scalar_to_vector (i64
1259 (load (ro_Windexed64 GPR64sp:$Rn, GPR32:$Rm,
1260 ro_Wextend64:$extend))))),
1261 (LDRDroW GPR64sp:$Rn, GPR32:$Rm, ro_Wextend64:$extend)>;
1263 def : Pat <(v1i64 (scalar_to_vector (i64
1264 (load (ro_Xindexed64 GPR64sp:$Rn, GPR64:$Rm,
1265 ro_Xextend64:$extend))))),
1266 (LDRDroX GPR64sp:$Rn, GPR64:$Rm, ro_Xextend64:$extend)>;
1269 // Match all load 64 bits width whose type is compatible with FPR64
1270 multiclass VecROLoadPat<ROAddrMode ro, ValueType VecTy,
1271 Instruction LOADW, Instruction LOADX> {
1273 def : Pat<(VecTy (load (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend))),
1274 (LOADW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
1276 def : Pat<(VecTy (load (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend))),
1277 (LOADX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
1280 let AddedComplexity = 10 in {
1281 let Predicates = [IsLE] in {
1282 // We must do vector loads with LD1 in big-endian.
1283 defm : VecROLoadPat<ro64, v2i32, LDRDroW, LDRDroX>;
1284 defm : VecROLoadPat<ro64, v2f32, LDRDroW, LDRDroX>;
1285 defm : VecROLoadPat<ro64, v8i8, LDRDroW, LDRDroX>;
1286 defm : VecROLoadPat<ro64, v4i16, LDRDroW, LDRDroX>;
1287 defm : VecROLoadPat<ro64, v4f16, LDRDroW, LDRDroX>;
1290 defm : VecROLoadPat<ro64, v1i64, LDRDroW, LDRDroX>;
1291 defm : VecROLoadPat<ro64, v1f64, LDRDroW, LDRDroX>;
1293 // Match all load 128 bits width whose type is compatible with FPR128
1294 let Predicates = [IsLE] in {
1295 // We must do vector loads with LD1 in big-endian.
1296 defm : VecROLoadPat<ro128, v2i64, LDRQroW, LDRQroX>;
1297 defm : VecROLoadPat<ro128, v2f64, LDRQroW, LDRQroX>;
1298 defm : VecROLoadPat<ro128, v4i32, LDRQroW, LDRQroX>;
1299 defm : VecROLoadPat<ro128, v4f32, LDRQroW, LDRQroX>;
1300 defm : VecROLoadPat<ro128, v8i16, LDRQroW, LDRQroX>;
1301 defm : VecROLoadPat<ro128, v8f16, LDRQroW, LDRQroX>;
1302 defm : VecROLoadPat<ro128, v16i8, LDRQroW, LDRQroX>;
1304 } // AddedComplexity = 10
1307 multiclass ExtLoadTo64ROPat<ROAddrMode ro, SDPatternOperator loadop,
1308 Instruction INSTW, Instruction INSTX> {
1309 def : Pat<(i64 (loadop (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend))),
1310 (SUBREG_TO_REG (i64 0),
1311 (INSTW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend),
1314 def : Pat<(i64 (loadop (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend))),
1315 (SUBREG_TO_REG (i64 0),
1316 (INSTX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend),
1320 let AddedComplexity = 10 in {
1321 defm : ExtLoadTo64ROPat<ro8, zextloadi8, LDRBBroW, LDRBBroX>;
1322 defm : ExtLoadTo64ROPat<ro16, zextloadi16, LDRHHroW, LDRHHroX>;
1323 defm : ExtLoadTo64ROPat<ro32, zextloadi32, LDRWroW, LDRWroX>;
1325 // zextloadi1 -> zextloadi8
1326 defm : ExtLoadTo64ROPat<ro8, zextloadi1, LDRBBroW, LDRBBroX>;
1328 // extload -> zextload
1329 defm : ExtLoadTo64ROPat<ro8, extloadi8, LDRBBroW, LDRBBroX>;
1330 defm : ExtLoadTo64ROPat<ro16, extloadi16, LDRHHroW, LDRHHroX>;
1331 defm : ExtLoadTo64ROPat<ro32, extloadi32, LDRWroW, LDRWroX>;
1333 // extloadi1 -> zextloadi8
1334 defm : ExtLoadTo64ROPat<ro8, extloadi1, LDRBBroW, LDRBBroX>;
1339 multiclass ExtLoadTo32ROPat<ROAddrMode ro, SDPatternOperator loadop,
1340 Instruction INSTW, Instruction INSTX> {
1341 def : Pat<(i32 (loadop (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend))),
1342 (INSTW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
1344 def : Pat<(i32 (loadop (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend))),
1345 (INSTX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
1349 let AddedComplexity = 10 in {
1350 // extload -> zextload
1351 defm : ExtLoadTo32ROPat<ro8, extloadi8, LDRBBroW, LDRBBroX>;
1352 defm : ExtLoadTo32ROPat<ro16, extloadi16, LDRHHroW, LDRHHroX>;
1353 defm : ExtLoadTo32ROPat<ro32, extloadi32, LDRWroW, LDRWroX>;
1355 // zextloadi1 -> zextloadi8
1356 defm : ExtLoadTo32ROPat<ro8, zextloadi1, LDRBBroW, LDRBBroX>;
1360 // (unsigned immediate)
1362 defm LDRX : LoadUI<0b11, 0, 0b01, GPR64, uimm12s8, "ldr",
1364 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)))]>;
1365 defm LDRW : LoadUI<0b10, 0, 0b01, GPR32, uimm12s4, "ldr",
1367 (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset)))]>;
1368 defm LDRB : LoadUI<0b00, 1, 0b01, FPR8, uimm12s1, "ldr",
1370 (load (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset)))]>;
1371 defm LDRH : LoadUI<0b01, 1, 0b01, FPR16, uimm12s2, "ldr",
1372 [(set (f16 FPR16:$Rt),
1373 (load (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset)))]>;
1374 defm LDRS : LoadUI<0b10, 1, 0b01, FPR32, uimm12s4, "ldr",
1375 [(set (f32 FPR32:$Rt),
1376 (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset)))]>;
1377 defm LDRD : LoadUI<0b11, 1, 0b01, FPR64, uimm12s8, "ldr",
1378 [(set (f64 FPR64:$Rt),
1379 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)))]>;
1380 defm LDRQ : LoadUI<0b00, 1, 0b11, FPR128, uimm12s16, "ldr",
1381 [(set (f128 FPR128:$Rt),
1382 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)))]>;
1384 // For regular load, we do not have any alignment requirement.
1385 // Thus, it is safe to directly map the vector loads with interesting
1386 // addressing modes.
1387 // FIXME: We could do the same for bitconvert to floating point vectors.
1388 def : Pat <(v8i8 (scalar_to_vector (i32
1389 (extloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))))),
1390 (INSERT_SUBREG (v8i8 (IMPLICIT_DEF)),
1391 (LDRBui GPR64sp:$Rn, uimm12s1:$offset), bsub)>;
1392 def : Pat <(v16i8 (scalar_to_vector (i32
1393 (extloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))))),
1394 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
1395 (LDRBui GPR64sp:$Rn, uimm12s1:$offset), bsub)>;
1396 def : Pat <(v4i16 (scalar_to_vector (i32
1397 (extloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))))),
1398 (INSERT_SUBREG (v4i16 (IMPLICIT_DEF)),
1399 (LDRHui GPR64sp:$Rn, uimm12s2:$offset), hsub)>;
1400 def : Pat <(v8i16 (scalar_to_vector (i32
1401 (extloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))))),
1402 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
1403 (LDRHui GPR64sp:$Rn, uimm12s2:$offset), hsub)>;
1404 def : Pat <(v2i32 (scalar_to_vector (i32
1405 (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))))),
1406 (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)),
1407 (LDRSui GPR64sp:$Rn, uimm12s4:$offset), ssub)>;
1408 def : Pat <(v4i32 (scalar_to_vector (i32
1409 (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))))),
1410 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
1411 (LDRSui GPR64sp:$Rn, uimm12s4:$offset), ssub)>;
1412 def : Pat <(v1i64 (scalar_to_vector (i64
1413 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))))),
1414 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1415 def : Pat <(v2i64 (scalar_to_vector (i64
1416 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))))),
1417 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
1418 (LDRDui GPR64sp:$Rn, uimm12s8:$offset), dsub)>;
1420 // Match all load 64 bits width whose type is compatible with FPR64
1421 let Predicates = [IsLE] in {
1422 // We must use LD1 to perform vector loads in big-endian.
1423 def : Pat<(v2f32 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1424 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1425 def : Pat<(v8i8 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1426 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1427 def : Pat<(v4i16 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1428 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1429 def : Pat<(v2i32 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1430 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1431 def : Pat<(v4f16 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1432 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1434 def : Pat<(v1f64 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1435 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1436 def : Pat<(v1i64 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1437 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1439 // Match all load 128 bits width whose type is compatible with FPR128
1440 let Predicates = [IsLE] in {
1441 // We must use LD1 to perform vector loads in big-endian.
1442 def : Pat<(v4f32 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1443 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1444 def : Pat<(v2f64 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1445 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1446 def : Pat<(v16i8 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1447 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1448 def : Pat<(v8i16 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1449 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1450 def : Pat<(v4i32 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1451 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1452 def : Pat<(v2i64 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1453 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1454 def : Pat<(v8f16 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1455 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1457 def : Pat<(f128 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1458 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1460 defm LDRHH : LoadUI<0b01, 0, 0b01, GPR32, uimm12s2, "ldrh",
1462 (zextloadi16 (am_indexed16 GPR64sp:$Rn,
1463 uimm12s2:$offset)))]>;
1464 defm LDRBB : LoadUI<0b00, 0, 0b01, GPR32, uimm12s1, "ldrb",
1466 (zextloadi8 (am_indexed8 GPR64sp:$Rn,
1467 uimm12s1:$offset)))]>;
1469 def : Pat<(i64 (zextloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1470 (SUBREG_TO_REG (i64 0), (LDRBBui GPR64sp:$Rn, uimm12s1:$offset), sub_32)>;
1471 def : Pat<(i64 (zextloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))),
1472 (SUBREG_TO_REG (i64 0), (LDRHHui GPR64sp:$Rn, uimm12s2:$offset), sub_32)>;
1474 // zextloadi1 -> zextloadi8
1475 def : Pat<(i32 (zextloadi1 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1476 (LDRBBui GPR64sp:$Rn, uimm12s1:$offset)>;
1477 def : Pat<(i64 (zextloadi1 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1478 (SUBREG_TO_REG (i64 0), (LDRBBui GPR64sp:$Rn, uimm12s1:$offset), sub_32)>;
1480 // extload -> zextload
1481 def : Pat<(i32 (extloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))),
1482 (LDRHHui GPR64sp:$Rn, uimm12s2:$offset)>;
1483 def : Pat<(i32 (extloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1484 (LDRBBui GPR64sp:$Rn, uimm12s1:$offset)>;
1485 def : Pat<(i32 (extloadi1 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1486 (LDRBBui GPR64sp:$Rn, uimm12s1:$offset)>;
1487 def : Pat<(i64 (extloadi32 (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))),
1488 (SUBREG_TO_REG (i64 0), (LDRWui GPR64sp:$Rn, uimm12s4:$offset), sub_32)>;
1489 def : Pat<(i64 (extloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))),
1490 (SUBREG_TO_REG (i64 0), (LDRHHui GPR64sp:$Rn, uimm12s2:$offset), sub_32)>;
1491 def : Pat<(i64 (extloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1492 (SUBREG_TO_REG (i64 0), (LDRBBui GPR64sp:$Rn, uimm12s1:$offset), sub_32)>;
1493 def : Pat<(i64 (extloadi1 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1494 (SUBREG_TO_REG (i64 0), (LDRBBui GPR64sp:$Rn, uimm12s1:$offset), sub_32)>;
1496 // load sign-extended half-word
1497 defm LDRSHW : LoadUI<0b01, 0, 0b11, GPR32, uimm12s2, "ldrsh",
1499 (sextloadi16 (am_indexed16 GPR64sp:$Rn,
1500 uimm12s2:$offset)))]>;
1501 defm LDRSHX : LoadUI<0b01, 0, 0b10, GPR64, uimm12s2, "ldrsh",
1503 (sextloadi16 (am_indexed16 GPR64sp:$Rn,
1504 uimm12s2:$offset)))]>;
1506 // load sign-extended byte
1507 defm LDRSBW : LoadUI<0b00, 0, 0b11, GPR32, uimm12s1, "ldrsb",
1509 (sextloadi8 (am_indexed8 GPR64sp:$Rn,
1510 uimm12s1:$offset)))]>;
1511 defm LDRSBX : LoadUI<0b00, 0, 0b10, GPR64, uimm12s1, "ldrsb",
1513 (sextloadi8 (am_indexed8 GPR64sp:$Rn,
1514 uimm12s1:$offset)))]>;
1516 // load sign-extended word
1517 defm LDRSW : LoadUI<0b10, 0, 0b10, GPR64, uimm12s4, "ldrsw",
1519 (sextloadi32 (am_indexed32 GPR64sp:$Rn,
1520 uimm12s4:$offset)))]>;
1522 // load zero-extended word
1523 def : Pat<(i64 (zextloadi32 (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))),
1524 (SUBREG_TO_REG (i64 0), (LDRWui GPR64sp:$Rn, uimm12s4:$offset), sub_32)>;
1527 def PRFMui : PrefetchUI<0b11, 0, 0b10, "prfm",
1528 [(AArch64Prefetch imm:$Rt,
1529 (am_indexed64 GPR64sp:$Rn,
1530 uimm12s8:$offset))]>;
1532 def : InstAlias<"prfm $Rt, [$Rn]", (PRFMui prfop:$Rt, GPR64sp:$Rn, 0)>;
1536 def LDRWl : LoadLiteral<0b00, 0, GPR32, "ldr">;
1537 def LDRXl : LoadLiteral<0b01, 0, GPR64, "ldr">;
1538 def LDRSl : LoadLiteral<0b00, 1, FPR32, "ldr">;
1539 def LDRDl : LoadLiteral<0b01, 1, FPR64, "ldr">;
1540 def LDRQl : LoadLiteral<0b10, 1, FPR128, "ldr">;
1542 // load sign-extended word
1543 def LDRSWl : LoadLiteral<0b10, 0, GPR64, "ldrsw">;
1546 def PRFMl : PrefetchLiteral<0b11, 0, "prfm", []>;
1547 // [(AArch64Prefetch imm:$Rt, tglobaladdr:$label)]>;
1550 // (unscaled immediate)
1551 defm LDURX : LoadUnscaled<0b11, 0, 0b01, GPR64, "ldur",
1553 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset)))]>;
1554 defm LDURW : LoadUnscaled<0b10, 0, 0b01, GPR32, "ldur",
1556 (load (am_unscaled32 GPR64sp:$Rn, simm9:$offset)))]>;
1557 defm LDURB : LoadUnscaled<0b00, 1, 0b01, FPR8, "ldur",
1559 (load (am_unscaled8 GPR64sp:$Rn, simm9:$offset)))]>;
1560 defm LDURH : LoadUnscaled<0b01, 1, 0b01, FPR16, "ldur",
1562 (load (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
1563 defm LDURS : LoadUnscaled<0b10, 1, 0b01, FPR32, "ldur",
1564 [(set (f32 FPR32:$Rt),
1565 (load (am_unscaled32 GPR64sp:$Rn, simm9:$offset)))]>;
1566 defm LDURD : LoadUnscaled<0b11, 1, 0b01, FPR64, "ldur",
1567 [(set (f64 FPR64:$Rt),
1568 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset)))]>;
1569 defm LDURQ : LoadUnscaled<0b00, 1, 0b11, FPR128, "ldur",
1570 [(set (f128 FPR128:$Rt),
1571 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset)))]>;
1574 : LoadUnscaled<0b01, 0, 0b01, GPR32, "ldurh",
1576 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
1578 : LoadUnscaled<0b00, 0, 0b01, GPR32, "ldurb",
1580 (zextloadi8 (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
1582 // Match all load 64 bits width whose type is compatible with FPR64
1583 let Predicates = [IsLE] in {
1584 def : Pat<(v2f32 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1585 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1586 def : Pat<(v2i32 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1587 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1588 def : Pat<(v4i16 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1589 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1590 def : Pat<(v8i8 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1591 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1592 def : Pat<(v4f16 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1593 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1595 def : Pat<(v1f64 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1596 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1597 def : Pat<(v1i64 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1598 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1600 // Match all load 128 bits width whose type is compatible with FPR128
1601 let Predicates = [IsLE] in {
1602 def : Pat<(v2f64 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1603 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1604 def : Pat<(v2i64 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1605 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1606 def : Pat<(v4f32 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1607 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1608 def : Pat<(v4i32 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1609 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1610 def : Pat<(v8i16 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1611 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1612 def : Pat<(v16i8 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1613 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1614 def : Pat<(v8f16 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1615 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1619 def : Pat<(i32 (extloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
1620 (LDURHHi GPR64sp:$Rn, simm9:$offset)>;
1621 def : Pat<(i32 (extloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1622 (LDURBBi GPR64sp:$Rn, simm9:$offset)>;
1623 def : Pat<(i32 (extloadi1 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1624 (LDURBBi GPR64sp:$Rn, simm9:$offset)>;
1625 def : Pat<(i64 (extloadi32 (am_unscaled32 GPR64sp:$Rn, simm9:$offset))),
1626 (SUBREG_TO_REG (i64 0), (LDURWi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1627 def : Pat<(i64 (extloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
1628 (SUBREG_TO_REG (i64 0), (LDURHHi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1629 def : Pat<(i64 (extloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1630 (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1631 def : Pat<(i64 (extloadi1 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1632 (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1634 def : Pat<(i32 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
1635 (LDURHHi GPR64sp:$Rn, simm9:$offset)>;
1636 def : Pat<(i32 (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1637 (LDURBBi GPR64sp:$Rn, simm9:$offset)>;
1638 def : Pat<(i32 (zextloadi1 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1639 (LDURBBi GPR64sp:$Rn, simm9:$offset)>;
1640 def : Pat<(i64 (zextloadi32 (am_unscaled32 GPR64sp:$Rn, simm9:$offset))),
1641 (SUBREG_TO_REG (i64 0), (LDURWi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1642 def : Pat<(i64 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
1643 (SUBREG_TO_REG (i64 0), (LDURHHi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1644 def : Pat<(i64 (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1645 (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1646 def : Pat<(i64 (zextloadi1 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1647 (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1651 // LDR mnemonics fall back to LDUR for negative or unaligned offsets.
1653 // Define new assembler match classes as we want to only match these when
1654 // the don't otherwise match the scaled addressing mode for LDR/STR. Don't
1655 // associate a DiagnosticType either, as we want the diagnostic for the
1656 // canonical form (the scaled operand) to take precedence.
1657 class SImm9OffsetOperand<int Width> : AsmOperandClass {
1658 let Name = "SImm9OffsetFB" # Width;
1659 let PredicateMethod = "isSImm9OffsetFB<" # Width # ">";
1660 let RenderMethod = "addImmOperands";
1663 def SImm9OffsetFB8Operand : SImm9OffsetOperand<8>;
1664 def SImm9OffsetFB16Operand : SImm9OffsetOperand<16>;
1665 def SImm9OffsetFB32Operand : SImm9OffsetOperand<32>;
1666 def SImm9OffsetFB64Operand : SImm9OffsetOperand<64>;
1667 def SImm9OffsetFB128Operand : SImm9OffsetOperand<128>;
1669 def simm9_offset_fb8 : Operand<i64> {
1670 let ParserMatchClass = SImm9OffsetFB8Operand;
1672 def simm9_offset_fb16 : Operand<i64> {
1673 let ParserMatchClass = SImm9OffsetFB16Operand;
1675 def simm9_offset_fb32 : Operand<i64> {
1676 let ParserMatchClass = SImm9OffsetFB32Operand;
1678 def simm9_offset_fb64 : Operand<i64> {
1679 let ParserMatchClass = SImm9OffsetFB64Operand;
1681 def simm9_offset_fb128 : Operand<i64> {
1682 let ParserMatchClass = SImm9OffsetFB128Operand;
1685 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1686 (LDURXi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb64:$offset), 0>;
1687 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1688 (LDURWi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
1689 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1690 (LDURBi FPR8:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
1691 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1692 (LDURHi FPR16:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
1693 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1694 (LDURSi FPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
1695 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1696 (LDURDi FPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb64:$offset), 0>;
1697 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1698 (LDURQi FPR128:$Rt, GPR64sp:$Rn, simm9_offset_fb128:$offset), 0>;
1701 def : Pat<(i64 (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1702 (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1703 def : Pat<(i64 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
1704 (SUBREG_TO_REG (i64 0), (LDURHHi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1706 // load sign-extended half-word
1708 : LoadUnscaled<0b01, 0, 0b11, GPR32, "ldursh",
1710 (sextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
1712 : LoadUnscaled<0b01, 0, 0b10, GPR64, "ldursh",
1714 (sextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
1716 // load sign-extended byte
1718 : LoadUnscaled<0b00, 0, 0b11, GPR32, "ldursb",
1720 (sextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset)))]>;
1722 : LoadUnscaled<0b00, 0, 0b10, GPR64, "ldursb",
1724 (sextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset)))]>;
1726 // load sign-extended word
1728 : LoadUnscaled<0b10, 0, 0b10, GPR64, "ldursw",
1730 (sextloadi32 (am_unscaled32 GPR64sp:$Rn, simm9:$offset)))]>;
1732 // zero and sign extending aliases from generic LDR* mnemonics to LDUR*.
1733 def : InstAlias<"ldrb $Rt, [$Rn, $offset]",
1734 (LDURBBi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
1735 def : InstAlias<"ldrh $Rt, [$Rn, $offset]",
1736 (LDURHHi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
1737 def : InstAlias<"ldrsb $Rt, [$Rn, $offset]",
1738 (LDURSBWi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
1739 def : InstAlias<"ldrsb $Rt, [$Rn, $offset]",
1740 (LDURSBXi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
1741 def : InstAlias<"ldrsh $Rt, [$Rn, $offset]",
1742 (LDURSHWi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
1743 def : InstAlias<"ldrsh $Rt, [$Rn, $offset]",
1744 (LDURSHXi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
1745 def : InstAlias<"ldrsw $Rt, [$Rn, $offset]",
1746 (LDURSWi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
1749 defm PRFUM : PrefetchUnscaled<0b11, 0, 0b10, "prfum",
1750 [(AArch64Prefetch imm:$Rt,
1751 (am_unscaled64 GPR64sp:$Rn, simm9:$offset))]>;
1754 // (unscaled immediate, unprivileged)
1755 defm LDTRX : LoadUnprivileged<0b11, 0, 0b01, GPR64, "ldtr">;
1756 defm LDTRW : LoadUnprivileged<0b10, 0, 0b01, GPR32, "ldtr">;
1758 defm LDTRH : LoadUnprivileged<0b01, 0, 0b01, GPR32, "ldtrh">;
1759 defm LDTRB : LoadUnprivileged<0b00, 0, 0b01, GPR32, "ldtrb">;
1761 // load sign-extended half-word
1762 defm LDTRSHW : LoadUnprivileged<0b01, 0, 0b11, GPR32, "ldtrsh">;
1763 defm LDTRSHX : LoadUnprivileged<0b01, 0, 0b10, GPR64, "ldtrsh">;
1765 // load sign-extended byte
1766 defm LDTRSBW : LoadUnprivileged<0b00, 0, 0b11, GPR32, "ldtrsb">;
1767 defm LDTRSBX : LoadUnprivileged<0b00, 0, 0b10, GPR64, "ldtrsb">;
1769 // load sign-extended word
1770 defm LDTRSW : LoadUnprivileged<0b10, 0, 0b10, GPR64, "ldtrsw">;
1773 // (immediate pre-indexed)
1774 def LDRWpre : LoadPreIdx<0b10, 0, 0b01, GPR32, "ldr">;
1775 def LDRXpre : LoadPreIdx<0b11, 0, 0b01, GPR64, "ldr">;
1776 def LDRBpre : LoadPreIdx<0b00, 1, 0b01, FPR8, "ldr">;
1777 def LDRHpre : LoadPreIdx<0b01, 1, 0b01, FPR16, "ldr">;
1778 def LDRSpre : LoadPreIdx<0b10, 1, 0b01, FPR32, "ldr">;
1779 def LDRDpre : LoadPreIdx<0b11, 1, 0b01, FPR64, "ldr">;
1780 def LDRQpre : LoadPreIdx<0b00, 1, 0b11, FPR128, "ldr">;
1782 // load sign-extended half-word
1783 def LDRSHWpre : LoadPreIdx<0b01, 0, 0b11, GPR32, "ldrsh">;
1784 def LDRSHXpre : LoadPreIdx<0b01, 0, 0b10, GPR64, "ldrsh">;
1786 // load sign-extended byte
1787 def LDRSBWpre : LoadPreIdx<0b00, 0, 0b11, GPR32, "ldrsb">;
1788 def LDRSBXpre : LoadPreIdx<0b00, 0, 0b10, GPR64, "ldrsb">;
1790 // load zero-extended byte
1791 def LDRBBpre : LoadPreIdx<0b00, 0, 0b01, GPR32, "ldrb">;
1792 def LDRHHpre : LoadPreIdx<0b01, 0, 0b01, GPR32, "ldrh">;
1794 // load sign-extended word
1795 def LDRSWpre : LoadPreIdx<0b10, 0, 0b10, GPR64, "ldrsw">;
1798 // (immediate post-indexed)
1799 def LDRWpost : LoadPostIdx<0b10, 0, 0b01, GPR32, "ldr">;
1800 def LDRXpost : LoadPostIdx<0b11, 0, 0b01, GPR64, "ldr">;
1801 def LDRBpost : LoadPostIdx<0b00, 1, 0b01, FPR8, "ldr">;
1802 def LDRHpost : LoadPostIdx<0b01, 1, 0b01, FPR16, "ldr">;
1803 def LDRSpost : LoadPostIdx<0b10, 1, 0b01, FPR32, "ldr">;
1804 def LDRDpost : LoadPostIdx<0b11, 1, 0b01, FPR64, "ldr">;
1805 def LDRQpost : LoadPostIdx<0b00, 1, 0b11, FPR128, "ldr">;
1807 // load sign-extended half-word
1808 def LDRSHWpost : LoadPostIdx<0b01, 0, 0b11, GPR32, "ldrsh">;
1809 def LDRSHXpost : LoadPostIdx<0b01, 0, 0b10, GPR64, "ldrsh">;
1811 // load sign-extended byte
1812 def LDRSBWpost : LoadPostIdx<0b00, 0, 0b11, GPR32, "ldrsb">;
1813 def LDRSBXpost : LoadPostIdx<0b00, 0, 0b10, GPR64, "ldrsb">;
1815 // load zero-extended byte
1816 def LDRBBpost : LoadPostIdx<0b00, 0, 0b01, GPR32, "ldrb">;
1817 def LDRHHpost : LoadPostIdx<0b01, 0, 0b01, GPR32, "ldrh">;
1819 // load sign-extended word
1820 def LDRSWpost : LoadPostIdx<0b10, 0, 0b10, GPR64, "ldrsw">;
1822 //===----------------------------------------------------------------------===//
1823 // Store instructions.
1824 //===----------------------------------------------------------------------===//
1826 // Pair (indexed, offset)
1827 // FIXME: Use dedicated range-checked addressing mode operand here.
1828 defm STPW : StorePairOffset<0b00, 0, GPR32, simm7s4, "stp">;
1829 defm STPX : StorePairOffset<0b10, 0, GPR64, simm7s8, "stp">;
1830 defm STPS : StorePairOffset<0b00, 1, FPR32, simm7s4, "stp">;
1831 defm STPD : StorePairOffset<0b01, 1, FPR64, simm7s8, "stp">;
1832 defm STPQ : StorePairOffset<0b10, 1, FPR128, simm7s16, "stp">;
1834 // Pair (pre-indexed)
1835 def STPWpre : StorePairPreIdx<0b00, 0, GPR32, simm7s4, "stp">;
1836 def STPXpre : StorePairPreIdx<0b10, 0, GPR64, simm7s8, "stp">;
1837 def STPSpre : StorePairPreIdx<0b00, 1, FPR32, simm7s4, "stp">;
1838 def STPDpre : StorePairPreIdx<0b01, 1, FPR64, simm7s8, "stp">;
1839 def STPQpre : StorePairPreIdx<0b10, 1, FPR128, simm7s16, "stp">;
1841 // Pair (pre-indexed)
1842 def STPWpost : StorePairPostIdx<0b00, 0, GPR32, simm7s4, "stp">;
1843 def STPXpost : StorePairPostIdx<0b10, 0, GPR64, simm7s8, "stp">;
1844 def STPSpost : StorePairPostIdx<0b00, 1, FPR32, simm7s4, "stp">;
1845 def STPDpost : StorePairPostIdx<0b01, 1, FPR64, simm7s8, "stp">;
1846 def STPQpost : StorePairPostIdx<0b10, 1, FPR128, simm7s16, "stp">;
1848 // Pair (no allocate)
1849 defm STNPW : StorePairNoAlloc<0b00, 0, GPR32, simm7s4, "stnp">;
1850 defm STNPX : StorePairNoAlloc<0b10, 0, GPR64, simm7s8, "stnp">;
1851 defm STNPS : StorePairNoAlloc<0b00, 1, FPR32, simm7s4, "stnp">;
1852 defm STNPD : StorePairNoAlloc<0b01, 1, FPR64, simm7s8, "stnp">;
1853 defm STNPQ : StorePairNoAlloc<0b10, 1, FPR128, simm7s16, "stnp">;
1856 // (Register offset)
1859 defm STRBB : Store8RO< 0b00, 0, 0b00, GPR32, "strb", i32, truncstorei8>;
1860 defm STRHH : Store16RO<0b01, 0, 0b00, GPR32, "strh", i32, truncstorei16>;
1861 defm STRW : Store32RO<0b10, 0, 0b00, GPR32, "str", i32, store>;
1862 defm STRX : Store64RO<0b11, 0, 0b00, GPR64, "str", i64, store>;
1866 defm STRB : Store8RO< 0b00, 1, 0b00, FPR8, "str", untyped, store>;
1867 defm STRH : Store16RO<0b01, 1, 0b00, FPR16, "str", f16, store>;
1868 defm STRS : Store32RO<0b10, 1, 0b00, FPR32, "str", f32, store>;
1869 defm STRD : Store64RO<0b11, 1, 0b00, FPR64, "str", f64, store>;
1870 defm STRQ : Store128RO<0b00, 1, 0b10, FPR128, "str", f128, store>;
1872 multiclass TruncStoreFrom64ROPat<ROAddrMode ro, SDPatternOperator storeop,
1873 Instruction STRW, Instruction STRX> {
1875 def : Pat<(storeop GPR64:$Rt,
1876 (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)),
1877 (STRW (EXTRACT_SUBREG GPR64:$Rt, sub_32),
1878 GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
1880 def : Pat<(storeop GPR64:$Rt,
1881 (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)),
1882 (STRX (EXTRACT_SUBREG GPR64:$Rt, sub_32),
1883 GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
1886 let AddedComplexity = 10 in {
1888 defm : TruncStoreFrom64ROPat<ro8, truncstorei8, STRBBroW, STRBBroX>;
1889 defm : TruncStoreFrom64ROPat<ro16, truncstorei16, STRHHroW, STRHHroX>;
1890 defm : TruncStoreFrom64ROPat<ro32, truncstorei32, STRWroW, STRWroX>;
1893 multiclass VecROStorePat<ROAddrMode ro, ValueType VecTy, RegisterClass FPR,
1894 Instruction STRW, Instruction STRX> {
1895 def : Pat<(store (VecTy FPR:$Rt),
1896 (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)),
1897 (STRW FPR:$Rt, GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
1899 def : Pat<(store (VecTy FPR:$Rt),
1900 (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)),
1901 (STRX FPR:$Rt, GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
1904 let AddedComplexity = 10 in {
1905 // Match all store 64 bits width whose type is compatible with FPR64
1906 let Predicates = [IsLE] in {
1907 // We must use ST1 to store vectors in big-endian.
1908 defm : VecROStorePat<ro64, v2i32, FPR64, STRDroW, STRDroX>;
1909 defm : VecROStorePat<ro64, v2f32, FPR64, STRDroW, STRDroX>;
1910 defm : VecROStorePat<ro64, v4i16, FPR64, STRDroW, STRDroX>;
1911 defm : VecROStorePat<ro64, v8i8, FPR64, STRDroW, STRDroX>;
1912 defm : VecROStorePat<ro64, v4f16, FPR64, STRDroW, STRDroX>;
1915 defm : VecROStorePat<ro64, v1i64, FPR64, STRDroW, STRDroX>;
1916 defm : VecROStorePat<ro64, v1f64, FPR64, STRDroW, STRDroX>;
1918 // Match all store 128 bits width whose type is compatible with FPR128
1919 let Predicates = [IsLE] in {
1920 // We must use ST1 to store vectors in big-endian.
1921 defm : VecROStorePat<ro128, v2i64, FPR128, STRQroW, STRQroX>;
1922 defm : VecROStorePat<ro128, v2f64, FPR128, STRQroW, STRQroX>;
1923 defm : VecROStorePat<ro128, v4i32, FPR128, STRQroW, STRQroX>;
1924 defm : VecROStorePat<ro128, v4f32, FPR128, STRQroW, STRQroX>;
1925 defm : VecROStorePat<ro128, v8i16, FPR128, STRQroW, STRQroX>;
1926 defm : VecROStorePat<ro128, v16i8, FPR128, STRQroW, STRQroX>;
1927 defm : VecROStorePat<ro128, v8f16, FPR128, STRQroW, STRQroX>;
1929 } // AddedComplexity = 10
1931 // Match stores from lane 0 to the appropriate subreg's store.
1932 multiclass VecROStoreLane0Pat<ROAddrMode ro, SDPatternOperator storeop,
1933 ValueType VecTy, ValueType STy,
1934 SubRegIndex SubRegIdx,
1935 Instruction STRW, Instruction STRX> {
1937 def : Pat<(storeop (STy (vector_extract (VecTy VecListOne128:$Vt), 0)),
1938 (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)),
1939 (STRW (EXTRACT_SUBREG VecListOne128:$Vt, SubRegIdx),
1940 GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
1942 def : Pat<(storeop (STy (vector_extract (VecTy VecListOne128:$Vt), 0)),
1943 (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)),
1944 (STRX (EXTRACT_SUBREG VecListOne128:$Vt, SubRegIdx),
1945 GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
1948 let AddedComplexity = 19 in {
1949 defm : VecROStoreLane0Pat<ro16, truncstorei16, v8i16, i32, hsub, STRHroW, STRHroX>;
1950 defm : VecROStoreLane0Pat<ro16, store , v8i16, i16, hsub, STRHroW, STRHroX>;
1951 defm : VecROStoreLane0Pat<ro32, truncstorei32, v4i32, i32, ssub, STRSroW, STRSroX>;
1952 defm : VecROStoreLane0Pat<ro32, store , v4i32, i32, ssub, STRSroW, STRSroX>;
1953 defm : VecROStoreLane0Pat<ro32, store , v4f32, f32, ssub, STRSroW, STRSroX>;
1954 defm : VecROStoreLane0Pat<ro64, store , v2i64, i64, dsub, STRDroW, STRDroX>;
1955 defm : VecROStoreLane0Pat<ro64, store , v2f64, f64, dsub, STRDroW, STRDroX>;
1959 // (unsigned immediate)
1960 defm STRX : StoreUI<0b11, 0, 0b00, GPR64, uimm12s8, "str",
1962 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))]>;
1963 defm STRW : StoreUI<0b10, 0, 0b00, GPR32, uimm12s4, "str",
1965 (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))]>;
1966 defm STRB : StoreUI<0b00, 1, 0b00, FPR8, uimm12s1, "str",
1968 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))]>;
1969 defm STRH : StoreUI<0b01, 1, 0b00, FPR16, uimm12s2, "str",
1970 [(store (f16 FPR16:$Rt),
1971 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))]>;
1972 defm STRS : StoreUI<0b10, 1, 0b00, FPR32, uimm12s4, "str",
1973 [(store (f32 FPR32:$Rt),
1974 (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))]>;
1975 defm STRD : StoreUI<0b11, 1, 0b00, FPR64, uimm12s8, "str",
1976 [(store (f64 FPR64:$Rt),
1977 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))]>;
1978 defm STRQ : StoreUI<0b00, 1, 0b10, FPR128, uimm12s16, "str", []>;
1980 defm STRHH : StoreUI<0b01, 0, 0b00, GPR32, uimm12s2, "strh",
1981 [(truncstorei16 GPR32:$Rt,
1982 (am_indexed16 GPR64sp:$Rn,
1983 uimm12s2:$offset))]>;
1984 defm STRBB : StoreUI<0b00, 0, 0b00, GPR32, uimm12s1, "strb",
1985 [(truncstorei8 GPR32:$Rt,
1986 (am_indexed8 GPR64sp:$Rn,
1987 uimm12s1:$offset))]>;
1989 // Match all store 64 bits width whose type is compatible with FPR64
1990 let AddedComplexity = 10 in {
1991 let Predicates = [IsLE] in {
1992 // We must use ST1 to store vectors in big-endian.
1993 def : Pat<(store (v2f32 FPR64:$Rt),
1994 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
1995 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
1996 def : Pat<(store (v8i8 FPR64:$Rt),
1997 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
1998 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
1999 def : Pat<(store (v4i16 FPR64:$Rt),
2000 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
2001 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
2002 def : Pat<(store (v2i32 FPR64:$Rt),
2003 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
2004 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
2005 def : Pat<(store (v4f16 FPR64:$Rt),
2006 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
2007 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
2009 def : Pat<(store (v1f64 FPR64:$Rt),
2010 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
2011 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
2012 def : Pat<(store (v1i64 FPR64:$Rt),
2013 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
2014 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
2016 // Match all store 128 bits width whose type is compatible with FPR128
2017 let Predicates = [IsLE] in {
2018 // We must use ST1 to store vectors in big-endian.
2019 def : Pat<(store (v4f32 FPR128:$Rt),
2020 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
2021 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
2022 def : Pat<(store (v2f64 FPR128:$Rt),
2023 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
2024 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
2025 def : Pat<(store (v16i8 FPR128:$Rt),
2026 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
2027 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
2028 def : Pat<(store (v8i16 FPR128:$Rt),
2029 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
2030 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
2031 def : Pat<(store (v4i32 FPR128:$Rt),
2032 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
2033 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
2034 def : Pat<(store (v2i64 FPR128:$Rt),
2035 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
2036 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
2037 def : Pat<(store (v8f16 FPR128:$Rt),
2038 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
2039 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
2041 def : Pat<(store (f128 FPR128:$Rt),
2042 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
2043 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
2046 def : Pat<(truncstorei32 GPR64:$Rt,
2047 (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset)),
2048 (STRWui (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, uimm12s4:$offset)>;
2049 def : Pat<(truncstorei16 GPR64:$Rt,
2050 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset)),
2051 (STRHHui (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, uimm12s2:$offset)>;
2052 def : Pat<(truncstorei8 GPR64:$Rt, (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset)),
2053 (STRBBui (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, uimm12s1:$offset)>;
2055 } // AddedComplexity = 10
2058 // (unscaled immediate)
2059 defm STURX : StoreUnscaled<0b11, 0, 0b00, GPR64, "stur",
2061 (am_unscaled64 GPR64sp:$Rn, simm9:$offset))]>;
2062 defm STURW : StoreUnscaled<0b10, 0, 0b00, GPR32, "stur",
2064 (am_unscaled32 GPR64sp:$Rn, simm9:$offset))]>;
2065 defm STURB : StoreUnscaled<0b00, 1, 0b00, FPR8, "stur",
2067 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))]>;
2068 defm STURH : StoreUnscaled<0b01, 1, 0b00, FPR16, "stur",
2069 [(store (f16 FPR16:$Rt),
2070 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))]>;
2071 defm STURS : StoreUnscaled<0b10, 1, 0b00, FPR32, "stur",
2072 [(store (f32 FPR32:$Rt),
2073 (am_unscaled32 GPR64sp:$Rn, simm9:$offset))]>;
2074 defm STURD : StoreUnscaled<0b11, 1, 0b00, FPR64, "stur",
2075 [(store (f64 FPR64:$Rt),
2076 (am_unscaled64 GPR64sp:$Rn, simm9:$offset))]>;
2077 defm STURQ : StoreUnscaled<0b00, 1, 0b10, FPR128, "stur",
2078 [(store (f128 FPR128:$Rt),
2079 (am_unscaled128 GPR64sp:$Rn, simm9:$offset))]>;
2080 defm STURHH : StoreUnscaled<0b01, 0, 0b00, GPR32, "sturh",
2081 [(truncstorei16 GPR32:$Rt,
2082 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))]>;
2083 defm STURBB : StoreUnscaled<0b00, 0, 0b00, GPR32, "sturb",
2084 [(truncstorei8 GPR32:$Rt,
2085 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))]>;
2087 // Match all store 64 bits width whose type is compatible with FPR64
2088 let Predicates = [IsLE] in {
2089 // We must use ST1 to store vectors in big-endian.
2090 def : Pat<(store (v2f32 FPR64:$Rt),
2091 (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2092 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2093 def : Pat<(store (v8i8 FPR64:$Rt),
2094 (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2095 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2096 def : Pat<(store (v4i16 FPR64:$Rt),
2097 (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2098 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2099 def : Pat<(store (v2i32 FPR64:$Rt),
2100 (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2101 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2102 def : Pat<(store (v4f16 FPR64:$Rt),
2103 (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2104 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2106 def : Pat<(store (v1f64 FPR64:$Rt), (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2107 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2108 def : Pat<(store (v1i64 FPR64:$Rt), (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2109 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2111 // Match all store 128 bits width whose type is compatible with FPR128
2112 let Predicates = [IsLE] in {
2113 // We must use ST1 to store vectors in big-endian.
2114 def : Pat<(store (v4f32 FPR128:$Rt),
2115 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2116 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2117 def : Pat<(store (v2f64 FPR128:$Rt),
2118 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2119 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2120 def : Pat<(store (v16i8 FPR128:$Rt),
2121 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2122 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2123 def : Pat<(store (v8i16 FPR128:$Rt),
2124 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2125 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2126 def : Pat<(store (v4i32 FPR128:$Rt),
2127 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2128 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2129 def : Pat<(store (v2i64 FPR128:$Rt),
2130 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2131 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2132 def : Pat<(store (v2f64 FPR128:$Rt),
2133 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2134 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2135 def : Pat<(store (v8f16 FPR128:$Rt),
2136 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2137 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2140 // unscaled i64 truncating stores
2141 def : Pat<(truncstorei32 GPR64:$Rt, (am_unscaled32 GPR64sp:$Rn, simm9:$offset)),
2142 (STURWi (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, simm9:$offset)>;
2143 def : Pat<(truncstorei16 GPR64:$Rt, (am_unscaled16 GPR64sp:$Rn, simm9:$offset)),
2144 (STURHHi (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, simm9:$offset)>;
2145 def : Pat<(truncstorei8 GPR64:$Rt, (am_unscaled8 GPR64sp:$Rn, simm9:$offset)),
2146 (STURBBi (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, simm9:$offset)>;
2149 // STR mnemonics fall back to STUR for negative or unaligned offsets.
2150 def : InstAlias<"str $Rt, [$Rn, $offset]",
2151 (STURXi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb64:$offset), 0>;
2152 def : InstAlias<"str $Rt, [$Rn, $offset]",
2153 (STURWi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
2154 def : InstAlias<"str $Rt, [$Rn, $offset]",
2155 (STURBi FPR8:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
2156 def : InstAlias<"str $Rt, [$Rn, $offset]",
2157 (STURHi FPR16:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
2158 def : InstAlias<"str $Rt, [$Rn, $offset]",
2159 (STURSi FPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
2160 def : InstAlias<"str $Rt, [$Rn, $offset]",
2161 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb64:$offset), 0>;
2162 def : InstAlias<"str $Rt, [$Rn, $offset]",
2163 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9_offset_fb128:$offset), 0>;
2165 def : InstAlias<"strb $Rt, [$Rn, $offset]",
2166 (STURBBi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
2167 def : InstAlias<"strh $Rt, [$Rn, $offset]",
2168 (STURHHi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
2171 // (unscaled immediate, unprivileged)
2172 defm STTRW : StoreUnprivileged<0b10, 0, 0b00, GPR32, "sttr">;
2173 defm STTRX : StoreUnprivileged<0b11, 0, 0b00, GPR64, "sttr">;
2175 defm STTRH : StoreUnprivileged<0b01, 0, 0b00, GPR32, "sttrh">;
2176 defm STTRB : StoreUnprivileged<0b00, 0, 0b00, GPR32, "sttrb">;
2179 // (immediate pre-indexed)
2180 def STRWpre : StorePreIdx<0b10, 0, 0b00, GPR32, "str", pre_store, i32>;
2181 def STRXpre : StorePreIdx<0b11, 0, 0b00, GPR64, "str", pre_store, i64>;
2182 def STRBpre : StorePreIdx<0b00, 1, 0b00, FPR8, "str", pre_store, untyped>;
2183 def STRHpre : StorePreIdx<0b01, 1, 0b00, FPR16, "str", pre_store, f16>;
2184 def STRSpre : StorePreIdx<0b10, 1, 0b00, FPR32, "str", pre_store, f32>;
2185 def STRDpre : StorePreIdx<0b11, 1, 0b00, FPR64, "str", pre_store, f64>;
2186 def STRQpre : StorePreIdx<0b00, 1, 0b10, FPR128, "str", pre_store, f128>;
2188 def STRBBpre : StorePreIdx<0b00, 0, 0b00, GPR32, "strb", pre_truncsti8, i32>;
2189 def STRHHpre : StorePreIdx<0b01, 0, 0b00, GPR32, "strh", pre_truncsti16, i32>;
2192 def : Pat<(pre_truncsti32 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
2193 (STRWpre (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
2195 def : Pat<(pre_truncsti16 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
2196 (STRHHpre (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
2198 def : Pat<(pre_truncsti8 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
2199 (STRBBpre (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
2202 def : Pat<(pre_store (v8i8 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2203 (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2204 def : Pat<(pre_store (v4i16 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2205 (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2206 def : Pat<(pre_store (v2i32 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2207 (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2208 def : Pat<(pre_store (v2f32 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2209 (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2210 def : Pat<(pre_store (v1i64 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2211 (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2212 def : Pat<(pre_store (v1f64 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2213 (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2214 def : Pat<(pre_store (v4f16 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2215 (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2217 def : Pat<(pre_store (v16i8 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2218 (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2219 def : Pat<(pre_store (v8i16 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2220 (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2221 def : Pat<(pre_store (v4i32 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2222 (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2223 def : Pat<(pre_store (v4f32 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2224 (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2225 def : Pat<(pre_store (v2i64 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2226 (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2227 def : Pat<(pre_store (v2f64 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2228 (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2229 def : Pat<(pre_store (v8f16 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2230 (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2233 // (immediate post-indexed)
2234 def STRWpost : StorePostIdx<0b10, 0, 0b00, GPR32, "str", post_store, i32>;
2235 def STRXpost : StorePostIdx<0b11, 0, 0b00, GPR64, "str", post_store, i64>;
2236 def STRBpost : StorePostIdx<0b00, 1, 0b00, FPR8, "str", post_store, untyped>;
2237 def STRHpost : StorePostIdx<0b01, 1, 0b00, FPR16, "str", post_store, f16>;
2238 def STRSpost : StorePostIdx<0b10, 1, 0b00, FPR32, "str", post_store, f32>;
2239 def STRDpost : StorePostIdx<0b11, 1, 0b00, FPR64, "str", post_store, f64>;
2240 def STRQpost : StorePostIdx<0b00, 1, 0b10, FPR128, "str", post_store, f128>;
2242 def STRBBpost : StorePostIdx<0b00, 0, 0b00, GPR32, "strb", post_truncsti8, i32>;
2243 def STRHHpost : StorePostIdx<0b01, 0, 0b00, GPR32, "strh", post_truncsti16, i32>;
2246 def : Pat<(post_truncsti32 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
2247 (STRWpost (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
2249 def : Pat<(post_truncsti16 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
2250 (STRHHpost (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
2252 def : Pat<(post_truncsti8 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
2253 (STRBBpost (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
2256 def : Pat<(post_store (v8i8 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2257 (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2258 def : Pat<(post_store (v4i16 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2259 (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2260 def : Pat<(post_store (v2i32 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2261 (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2262 def : Pat<(post_store (v2f32 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2263 (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2264 def : Pat<(post_store (v1i64 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2265 (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2266 def : Pat<(post_store (v1f64 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2267 (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2268 def : Pat<(post_store (v4f16 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2269 (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2271 def : Pat<(post_store (v16i8 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2272 (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2273 def : Pat<(post_store (v8i16 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2274 (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2275 def : Pat<(post_store (v4i32 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2276 (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2277 def : Pat<(post_store (v4f32 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2278 (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2279 def : Pat<(post_store (v2i64 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2280 (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2281 def : Pat<(post_store (v2f64 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2282 (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2283 def : Pat<(post_store (v8f16 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2284 (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2286 //===----------------------------------------------------------------------===//
2287 // Load/store exclusive instructions.
2288 //===----------------------------------------------------------------------===//
2290 def LDARW : LoadAcquire <0b10, 1, 1, 0, 1, GPR32, "ldar">;
2291 def LDARX : LoadAcquire <0b11, 1, 1, 0, 1, GPR64, "ldar">;
2292 def LDARB : LoadAcquire <0b00, 1, 1, 0, 1, GPR32, "ldarb">;
2293 def LDARH : LoadAcquire <0b01, 1, 1, 0, 1, GPR32, "ldarh">;
2295 def LDAXRW : LoadExclusive <0b10, 0, 1, 0, 1, GPR32, "ldaxr">;
2296 def LDAXRX : LoadExclusive <0b11, 0, 1, 0, 1, GPR64, "ldaxr">;
2297 def LDAXRB : LoadExclusive <0b00, 0, 1, 0, 1, GPR32, "ldaxrb">;
2298 def LDAXRH : LoadExclusive <0b01, 0, 1, 0, 1, GPR32, "ldaxrh">;
2300 def LDXRW : LoadExclusive <0b10, 0, 1, 0, 0, GPR32, "ldxr">;
2301 def LDXRX : LoadExclusive <0b11, 0, 1, 0, 0, GPR64, "ldxr">;
2302 def LDXRB : LoadExclusive <0b00, 0, 1, 0, 0, GPR32, "ldxrb">;
2303 def LDXRH : LoadExclusive <0b01, 0, 1, 0, 0, GPR32, "ldxrh">;
2305 def STLRW : StoreRelease <0b10, 1, 0, 0, 1, GPR32, "stlr">;
2306 def STLRX : StoreRelease <0b11, 1, 0, 0, 1, GPR64, "stlr">;
2307 def STLRB : StoreRelease <0b00, 1, 0, 0, 1, GPR32, "stlrb">;
2308 def STLRH : StoreRelease <0b01, 1, 0, 0, 1, GPR32, "stlrh">;
2310 def STLXRW : StoreExclusive<0b10, 0, 0, 0, 1, GPR32, "stlxr">;
2311 def STLXRX : StoreExclusive<0b11, 0, 0, 0, 1, GPR64, "stlxr">;
2312 def STLXRB : StoreExclusive<0b00, 0, 0, 0, 1, GPR32, "stlxrb">;
2313 def STLXRH : StoreExclusive<0b01, 0, 0, 0, 1, GPR32, "stlxrh">;
2315 def STXRW : StoreExclusive<0b10, 0, 0, 0, 0, GPR32, "stxr">;
2316 def STXRX : StoreExclusive<0b11, 0, 0, 0, 0, GPR64, "stxr">;
2317 def STXRB : StoreExclusive<0b00, 0, 0, 0, 0, GPR32, "stxrb">;
2318 def STXRH : StoreExclusive<0b01, 0, 0, 0, 0, GPR32, "stxrh">;
2320 def LDAXPW : LoadExclusivePair<0b10, 0, 1, 1, 1, GPR32, "ldaxp">;
2321 def LDAXPX : LoadExclusivePair<0b11, 0, 1, 1, 1, GPR64, "ldaxp">;
2323 def LDXPW : LoadExclusivePair<0b10, 0, 1, 1, 0, GPR32, "ldxp">;
2324 def LDXPX : LoadExclusivePair<0b11, 0, 1, 1, 0, GPR64, "ldxp">;
2326 def STLXPW : StoreExclusivePair<0b10, 0, 0, 1, 1, GPR32, "stlxp">;
2327 def STLXPX : StoreExclusivePair<0b11, 0, 0, 1, 1, GPR64, "stlxp">;
2329 def STXPW : StoreExclusivePair<0b10, 0, 0, 1, 0, GPR32, "stxp">;
2330 def STXPX : StoreExclusivePair<0b11, 0, 0, 1, 0, GPR64, "stxp">;
2332 let Predicates = [HasV8_1a] in {
2333 // v8.1a "Limited Order Region" extension load-acquire instructions
2334 def LDLARW : LoadAcquire <0b10, 1, 1, 0, 0, GPR32, "ldlar">;
2335 def LDLARX : LoadAcquire <0b11, 1, 1, 0, 0, GPR64, "ldlar">;
2336 def LDLARB : LoadAcquire <0b00, 1, 1, 0, 0, GPR32, "ldlarb">;
2337 def LDLARH : LoadAcquire <0b01, 1, 1, 0, 0, GPR32, "ldlarh">;
2339 // v8.1a "Limited Order Region" extension store-release instructions
2340 def STLLRW : StoreRelease <0b10, 1, 0, 0, 0, GPR32, "stllr">;
2341 def STLLRX : StoreRelease <0b11, 1, 0, 0, 0, GPR64, "stllr">;
2342 def STLLRB : StoreRelease <0b00, 1, 0, 0, 0, GPR32, "stllrb">;
2343 def STLLRH : StoreRelease <0b01, 1, 0, 0, 0, GPR32, "stllrh">;
2346 //===----------------------------------------------------------------------===//
2347 // Scaled floating point to integer conversion instructions.
2348 //===----------------------------------------------------------------------===//
2350 defm FCVTAS : FPToIntegerUnscaled<0b00, 0b100, "fcvtas", int_aarch64_neon_fcvtas>;
2351 defm FCVTAU : FPToIntegerUnscaled<0b00, 0b101, "fcvtau", int_aarch64_neon_fcvtau>;
2352 defm FCVTMS : FPToIntegerUnscaled<0b10, 0b000, "fcvtms", int_aarch64_neon_fcvtms>;
2353 defm FCVTMU : FPToIntegerUnscaled<0b10, 0b001, "fcvtmu", int_aarch64_neon_fcvtmu>;
2354 defm FCVTNS : FPToIntegerUnscaled<0b00, 0b000, "fcvtns", int_aarch64_neon_fcvtns>;
2355 defm FCVTNU : FPToIntegerUnscaled<0b00, 0b001, "fcvtnu", int_aarch64_neon_fcvtnu>;
2356 defm FCVTPS : FPToIntegerUnscaled<0b01, 0b000, "fcvtps", int_aarch64_neon_fcvtps>;
2357 defm FCVTPU : FPToIntegerUnscaled<0b01, 0b001, "fcvtpu", int_aarch64_neon_fcvtpu>;
2358 defm FCVTZS : FPToIntegerUnscaled<0b11, 0b000, "fcvtzs", fp_to_sint>;
2359 defm FCVTZU : FPToIntegerUnscaled<0b11, 0b001, "fcvtzu", fp_to_uint>;
2360 defm FCVTZS : FPToIntegerScaled<0b11, 0b000, "fcvtzs", fp_to_sint>;
2361 defm FCVTZU : FPToIntegerScaled<0b11, 0b001, "fcvtzu", fp_to_uint>;
2362 let isCodeGenOnly = 1 in {
2363 defm FCVTZS_Int : FPToIntegerUnscaled<0b11, 0b000, "fcvtzs", int_aarch64_neon_fcvtzs>;
2364 defm FCVTZU_Int : FPToIntegerUnscaled<0b11, 0b001, "fcvtzu", int_aarch64_neon_fcvtzu>;
2365 defm FCVTZS_Int : FPToIntegerScaled<0b11, 0b000, "fcvtzs", int_aarch64_neon_fcvtzs>;
2366 defm FCVTZU_Int : FPToIntegerScaled<0b11, 0b001, "fcvtzu", int_aarch64_neon_fcvtzu>;
2369 //===----------------------------------------------------------------------===//
2370 // Scaled integer to floating point conversion instructions.
2371 //===----------------------------------------------------------------------===//
2373 defm SCVTF : IntegerToFP<0, "scvtf", sint_to_fp>;
2374 defm UCVTF : IntegerToFP<1, "ucvtf", uint_to_fp>;
2376 //===----------------------------------------------------------------------===//
2377 // Unscaled integer to floating point conversion instruction.
2378 //===----------------------------------------------------------------------===//
2380 defm FMOV : UnscaledConversion<"fmov">;
2382 // Add pseudo ops for FMOV 0 so we can mark them as isReMaterializable
2383 let isReMaterializable = 1, isCodeGenOnly = 1 in {
2384 def FMOVS0 : Pseudo<(outs FPR32:$Rd), (ins), [(set f32:$Rd, (fpimm0))]>,
2385 PseudoInstExpansion<(FMOVWSr FPR32:$Rd, WZR)>,
2387 def FMOVD0 : Pseudo<(outs FPR64:$Rd), (ins), [(set f64:$Rd, (fpimm0))]>,
2388 PseudoInstExpansion<(FMOVXDr FPR64:$Rd, XZR)>,
2392 //===----------------------------------------------------------------------===//
2393 // Floating point conversion instruction.
2394 //===----------------------------------------------------------------------===//
2396 defm FCVT : FPConversion<"fcvt">;
2398 //===----------------------------------------------------------------------===//
2399 // Floating point single operand instructions.
2400 //===----------------------------------------------------------------------===//
2402 defm FABS : SingleOperandFPData<0b0001, "fabs", fabs>;
2403 defm FMOV : SingleOperandFPData<0b0000, "fmov">;
2404 defm FNEG : SingleOperandFPData<0b0010, "fneg", fneg>;
2405 defm FRINTA : SingleOperandFPData<0b1100, "frinta", frnd>;
2406 defm FRINTI : SingleOperandFPData<0b1111, "frinti", fnearbyint>;
2407 defm FRINTM : SingleOperandFPData<0b1010, "frintm", ffloor>;
2408 defm FRINTN : SingleOperandFPData<0b1000, "frintn", int_aarch64_neon_frintn>;
2409 defm FRINTP : SingleOperandFPData<0b1001, "frintp", fceil>;
2411 def : Pat<(v1f64 (int_aarch64_neon_frintn (v1f64 FPR64:$Rn))),
2412 (FRINTNDr FPR64:$Rn)>;
2414 // FRINTX is inserted to set the flags as required by FENV_ACCESS ON behavior
2415 // in the C spec. Setting hasSideEffects ensures it is not DCE'd.
2416 // <rdar://problem/13715968>
2417 // TODO: We should really model the FPSR flags correctly. This is really ugly.
2418 let hasSideEffects = 1 in {
2419 defm FRINTX : SingleOperandFPData<0b1110, "frintx", frint>;
2422 defm FRINTZ : SingleOperandFPData<0b1011, "frintz", ftrunc>;
2424 let SchedRW = [WriteFDiv] in {
2425 defm FSQRT : SingleOperandFPData<0b0011, "fsqrt", fsqrt>;
2428 //===----------------------------------------------------------------------===//
2429 // Floating point two operand instructions.
2430 //===----------------------------------------------------------------------===//
2432 defm FADD : TwoOperandFPData<0b0010, "fadd", fadd>;
2433 let SchedRW = [WriteFDiv] in {
2434 defm FDIV : TwoOperandFPData<0b0001, "fdiv", fdiv>;
2436 defm FMAXNM : TwoOperandFPData<0b0110, "fmaxnm", int_aarch64_neon_fmaxnm>;
2437 defm FMAX : TwoOperandFPData<0b0100, "fmax", AArch64fmax>;
2438 defm FMINNM : TwoOperandFPData<0b0111, "fminnm", int_aarch64_neon_fminnm>;
2439 defm FMIN : TwoOperandFPData<0b0101, "fmin", AArch64fmin>;
2440 let SchedRW = [WriteFMul] in {
2441 defm FMUL : TwoOperandFPData<0b0000, "fmul", fmul>;
2442 defm FNMUL : TwoOperandFPDataNeg<0b1000, "fnmul", fmul>;
2444 defm FSUB : TwoOperandFPData<0b0011, "fsub", fsub>;
2446 def : Pat<(v1f64 (AArch64fmax (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2447 (FMAXDrr FPR64:$Rn, FPR64:$Rm)>;
2448 def : Pat<(v1f64 (AArch64fmin (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2449 (FMINDrr FPR64:$Rn, FPR64:$Rm)>;
2450 def : Pat<(v1f64 (int_aarch64_neon_fmaxnm (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2451 (FMAXNMDrr FPR64:$Rn, FPR64:$Rm)>;
2452 def : Pat<(v1f64 (int_aarch64_neon_fminnm (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2453 (FMINNMDrr FPR64:$Rn, FPR64:$Rm)>;
2455 //===----------------------------------------------------------------------===//
2456 // Floating point three operand instructions.
2457 //===----------------------------------------------------------------------===//
2459 defm FMADD : ThreeOperandFPData<0, 0, "fmadd", fma>;
2460 defm FMSUB : ThreeOperandFPData<0, 1, "fmsub",
2461 TriOpFrag<(fma node:$LHS, (fneg node:$MHS), node:$RHS)> >;
2462 defm FNMADD : ThreeOperandFPData<1, 0, "fnmadd",
2463 TriOpFrag<(fneg (fma node:$LHS, node:$MHS, node:$RHS))> >;
2464 defm FNMSUB : ThreeOperandFPData<1, 1, "fnmsub",
2465 TriOpFrag<(fma node:$LHS, node:$MHS, (fneg node:$RHS))> >;
2467 // The following def pats catch the case where the LHS of an FMA is negated.
2468 // The TriOpFrag above catches the case where the middle operand is negated.
2470 // N.b. FMSUB etc have the accumulator at the *end* of (outs), unlike
2471 // the NEON variant.
2472 def : Pat<(f32 (fma (fneg FPR32:$Rn), FPR32:$Rm, FPR32:$Ra)),
2473 (FMSUBSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
2475 def : Pat<(f64 (fma (fneg FPR64:$Rn), FPR64:$Rm, FPR64:$Ra)),
2476 (FMSUBDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
2478 // We handled -(a + b*c) for FNMADD above, now it's time for "(-a) + (-b)*c" and
2480 def : Pat<(f32 (fma (fneg FPR32:$Rn), FPR32:$Rm, (fneg FPR32:$Ra))),
2481 (FNMADDSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
2483 def : Pat<(f64 (fma (fneg FPR64:$Rn), FPR64:$Rm, (fneg FPR64:$Ra))),
2484 (FNMADDDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
2486 def : Pat<(f32 (fma FPR32:$Rn, (fneg FPR32:$Rm), (fneg FPR32:$Ra))),
2487 (FNMADDSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
2489 def : Pat<(f64 (fma FPR64:$Rn, (fneg FPR64:$Rm), (fneg FPR64:$Ra))),
2490 (FNMADDDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
2492 //===----------------------------------------------------------------------===//
2493 // Floating point comparison instructions.
2494 //===----------------------------------------------------------------------===//
2496 defm FCMPE : FPComparison<1, "fcmpe">;
2497 defm FCMP : FPComparison<0, "fcmp", AArch64fcmp>;
2499 //===----------------------------------------------------------------------===//
2500 // Floating point conditional comparison instructions.
2501 //===----------------------------------------------------------------------===//
2503 defm FCCMPE : FPCondComparison<1, "fccmpe">;
2504 defm FCCMP : FPCondComparison<0, "fccmp", AArch64fccmp>;
2506 //===----------------------------------------------------------------------===//
2507 // Floating point conditional select instruction.
2508 //===----------------------------------------------------------------------===//
2510 defm FCSEL : FPCondSelect<"fcsel">;
2512 // CSEL instructions providing f128 types need to be handled by a
2513 // pseudo-instruction since the eventual code will need to introduce basic
2514 // blocks and control flow.
2515 def F128CSEL : Pseudo<(outs FPR128:$Rd),
2516 (ins FPR128:$Rn, FPR128:$Rm, ccode:$cond),
2517 [(set (f128 FPR128:$Rd),
2518 (AArch64csel FPR128:$Rn, FPR128:$Rm,
2519 (i32 imm:$cond), NZCV))]> {
2521 let usesCustomInserter = 1;
2525 //===----------------------------------------------------------------------===//
2526 // Floating point immediate move.
2527 //===----------------------------------------------------------------------===//
2529 let isReMaterializable = 1 in {
2530 defm FMOV : FPMoveImmediate<"fmov">;
2533 //===----------------------------------------------------------------------===//
2534 // Advanced SIMD two vector instructions.
2535 //===----------------------------------------------------------------------===//
2537 defm ABS : SIMDTwoVectorBHSD<0, 0b01011, "abs", int_aarch64_neon_abs>;
2538 def : Pat<(xor (v8i8 (AArch64vashr V64:$src, (i32 7))),
2539 (v8i8 (add V64:$src, (AArch64vashr V64:$src, (i32 7))))),
2540 (ABSv8i8 V64:$src)>;
2541 def : Pat<(xor (v4i16 (AArch64vashr V64:$src, (i32 15))),
2542 (v4i16 (add V64:$src, (AArch64vashr V64:$src, (i32 15))))),
2543 (ABSv4i16 V64:$src)>;
2544 def : Pat<(xor (v2i32 (AArch64vashr V64:$src, (i32 31))),
2545 (v2i32 (add V64:$src, (AArch64vashr V64:$src, (i32 31))))),
2546 (ABSv2i32 V64:$src)>;
2547 def : Pat<(xor (v16i8 (AArch64vashr V128:$src, (i32 7))),
2548 (v16i8 (add V128:$src, (AArch64vashr V128:$src, (i32 7))))),
2549 (ABSv16i8 V128:$src)>;
2550 def : Pat<(xor (v8i16 (AArch64vashr V128:$src, (i32 15))),
2551 (v8i16 (add V128:$src, (AArch64vashr V128:$src, (i32 15))))),
2552 (ABSv8i16 V128:$src)>;
2553 def : Pat<(xor (v4i32 (AArch64vashr V128:$src, (i32 31))),
2554 (v4i32 (add V128:$src, (AArch64vashr V128:$src, (i32 31))))),
2555 (ABSv4i32 V128:$src)>;
2556 def : Pat<(xor (v2i64 (AArch64vashr V128:$src, (i32 63))),
2557 (v2i64 (add V128:$src, (AArch64vashr V128:$src, (i32 63))))),
2558 (ABSv2i64 V128:$src)>;
2560 defm CLS : SIMDTwoVectorBHS<0, 0b00100, "cls", int_aarch64_neon_cls>;
2561 defm CLZ : SIMDTwoVectorBHS<1, 0b00100, "clz", ctlz>;
2562 defm CMEQ : SIMDCmpTwoVector<0, 0b01001, "cmeq", AArch64cmeqz>;
2563 defm CMGE : SIMDCmpTwoVector<1, 0b01000, "cmge", AArch64cmgez>;
2564 defm CMGT : SIMDCmpTwoVector<0, 0b01000, "cmgt", AArch64cmgtz>;
2565 defm CMLE : SIMDCmpTwoVector<1, 0b01001, "cmle", AArch64cmlez>;
2566 defm CMLT : SIMDCmpTwoVector<0, 0b01010, "cmlt", AArch64cmltz>;
2567 defm CNT : SIMDTwoVectorB<0, 0b00, 0b00101, "cnt", ctpop>;
2568 defm FABS : SIMDTwoVectorFP<0, 1, 0b01111, "fabs", fabs>;
2570 defm FCMEQ : SIMDFPCmpTwoVector<0, 1, 0b01101, "fcmeq", AArch64fcmeqz>;
2571 defm FCMGE : SIMDFPCmpTwoVector<1, 1, 0b01100, "fcmge", AArch64fcmgez>;
2572 defm FCMGT : SIMDFPCmpTwoVector<0, 1, 0b01100, "fcmgt", AArch64fcmgtz>;
2573 defm FCMLE : SIMDFPCmpTwoVector<1, 1, 0b01101, "fcmle", AArch64fcmlez>;
2574 defm FCMLT : SIMDFPCmpTwoVector<0, 1, 0b01110, "fcmlt", AArch64fcmltz>;
2575 defm FCVTAS : SIMDTwoVectorFPToInt<0,0,0b11100, "fcvtas",int_aarch64_neon_fcvtas>;
2576 defm FCVTAU : SIMDTwoVectorFPToInt<1,0,0b11100, "fcvtau",int_aarch64_neon_fcvtau>;
2577 defm FCVTL : SIMDFPWidenTwoVector<0, 0, 0b10111, "fcvtl">;
2578 def : Pat<(v4f32 (int_aarch64_neon_vcvthf2fp (v4i16 V64:$Rn))),
2579 (FCVTLv4i16 V64:$Rn)>;
2580 def : Pat<(v4f32 (int_aarch64_neon_vcvthf2fp (extract_subvector (v8i16 V128:$Rn),
2582 (FCVTLv8i16 V128:$Rn)>;
2583 def : Pat<(v2f64 (fextend (v2f32 V64:$Rn))), (FCVTLv2i32 V64:$Rn)>;
2584 def : Pat<(v2f64 (fextend (v2f32 (extract_subvector (v4f32 V128:$Rn),
2586 (FCVTLv4i32 V128:$Rn)>;
2588 def : Pat<(v4f32 (fextend (v4f16 V64:$Rn))), (FCVTLv4i16 V64:$Rn)>;
2589 def : Pat<(v4f32 (fextend (v4f16 (extract_subvector (v8f16 V128:$Rn),
2591 (FCVTLv8i16 V128:$Rn)>;
2593 defm FCVTMS : SIMDTwoVectorFPToInt<0,0,0b11011, "fcvtms",int_aarch64_neon_fcvtms>;
2594 defm FCVTMU : SIMDTwoVectorFPToInt<1,0,0b11011, "fcvtmu",int_aarch64_neon_fcvtmu>;
2595 defm FCVTNS : SIMDTwoVectorFPToInt<0,0,0b11010, "fcvtns",int_aarch64_neon_fcvtns>;
2596 defm FCVTNU : SIMDTwoVectorFPToInt<1,0,0b11010, "fcvtnu",int_aarch64_neon_fcvtnu>;
2597 defm FCVTN : SIMDFPNarrowTwoVector<0, 0, 0b10110, "fcvtn">;
2598 def : Pat<(v4i16 (int_aarch64_neon_vcvtfp2hf (v4f32 V128:$Rn))),
2599 (FCVTNv4i16 V128:$Rn)>;
2600 def : Pat<(concat_vectors V64:$Rd,
2601 (v4i16 (int_aarch64_neon_vcvtfp2hf (v4f32 V128:$Rn)))),
2602 (FCVTNv8i16 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
2603 def : Pat<(v2f32 (fround (v2f64 V128:$Rn))), (FCVTNv2i32 V128:$Rn)>;
2604 def : Pat<(v4f16 (fround (v4f32 V128:$Rn))), (FCVTNv4i16 V128:$Rn)>;
2605 def : Pat<(concat_vectors V64:$Rd, (v2f32 (fround (v2f64 V128:$Rn)))),
2606 (FCVTNv4i32 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
2607 defm FCVTPS : SIMDTwoVectorFPToInt<0,1,0b11010, "fcvtps",int_aarch64_neon_fcvtps>;
2608 defm FCVTPU : SIMDTwoVectorFPToInt<1,1,0b11010, "fcvtpu",int_aarch64_neon_fcvtpu>;
2609 defm FCVTXN : SIMDFPInexactCvtTwoVector<1, 0, 0b10110, "fcvtxn",
2610 int_aarch64_neon_fcvtxn>;
2611 defm FCVTZS : SIMDTwoVectorFPToInt<0, 1, 0b11011, "fcvtzs", fp_to_sint>;
2612 defm FCVTZU : SIMDTwoVectorFPToInt<1, 1, 0b11011, "fcvtzu", fp_to_uint>;
2613 let isCodeGenOnly = 1 in {
2614 defm FCVTZS_Int : SIMDTwoVectorFPToInt<0, 1, 0b11011, "fcvtzs",
2615 int_aarch64_neon_fcvtzs>;
2616 defm FCVTZU_Int : SIMDTwoVectorFPToInt<1, 1, 0b11011, "fcvtzu",
2617 int_aarch64_neon_fcvtzu>;
2619 defm FNEG : SIMDTwoVectorFP<1, 1, 0b01111, "fneg", fneg>;
2620 defm FRECPE : SIMDTwoVectorFP<0, 1, 0b11101, "frecpe", int_aarch64_neon_frecpe>;
2621 defm FRINTA : SIMDTwoVectorFP<1, 0, 0b11000, "frinta", frnd>;
2622 defm FRINTI : SIMDTwoVectorFP<1, 1, 0b11001, "frinti", fnearbyint>;
2623 defm FRINTM : SIMDTwoVectorFP<0, 0, 0b11001, "frintm", ffloor>;
2624 defm FRINTN : SIMDTwoVectorFP<0, 0, 0b11000, "frintn", int_aarch64_neon_frintn>;
2625 defm FRINTP : SIMDTwoVectorFP<0, 1, 0b11000, "frintp", fceil>;
2626 defm FRINTX : SIMDTwoVectorFP<1, 0, 0b11001, "frintx", frint>;
2627 defm FRINTZ : SIMDTwoVectorFP<0, 1, 0b11001, "frintz", ftrunc>;
2628 defm FRSQRTE: SIMDTwoVectorFP<1, 1, 0b11101, "frsqrte", int_aarch64_neon_frsqrte>;
2629 defm FSQRT : SIMDTwoVectorFP<1, 1, 0b11111, "fsqrt", fsqrt>;
2630 defm NEG : SIMDTwoVectorBHSD<1, 0b01011, "neg",
2631 UnOpFrag<(sub immAllZerosV, node:$LHS)> >;
2632 defm NOT : SIMDTwoVectorB<1, 0b00, 0b00101, "not", vnot>;
2633 // Aliases for MVN -> NOT.
2634 def : InstAlias<"mvn{ $Vd.8b, $Vn.8b|.8b $Vd, $Vn}",
2635 (NOTv8i8 V64:$Vd, V64:$Vn)>;
2636 def : InstAlias<"mvn{ $Vd.16b, $Vn.16b|.16b $Vd, $Vn}",
2637 (NOTv16i8 V128:$Vd, V128:$Vn)>;
2639 def : Pat<(AArch64neg (v8i8 V64:$Rn)), (NEGv8i8 V64:$Rn)>;
2640 def : Pat<(AArch64neg (v16i8 V128:$Rn)), (NEGv16i8 V128:$Rn)>;
2641 def : Pat<(AArch64neg (v4i16 V64:$Rn)), (NEGv4i16 V64:$Rn)>;
2642 def : Pat<(AArch64neg (v8i16 V128:$Rn)), (NEGv8i16 V128:$Rn)>;
2643 def : Pat<(AArch64neg (v2i32 V64:$Rn)), (NEGv2i32 V64:$Rn)>;
2644 def : Pat<(AArch64neg (v4i32 V128:$Rn)), (NEGv4i32 V128:$Rn)>;
2645 def : Pat<(AArch64neg (v2i64 V128:$Rn)), (NEGv2i64 V128:$Rn)>;
2647 def : Pat<(AArch64not (v8i8 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2648 def : Pat<(AArch64not (v16i8 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2649 def : Pat<(AArch64not (v4i16 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2650 def : Pat<(AArch64not (v8i16 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2651 def : Pat<(AArch64not (v2i32 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2652 def : Pat<(AArch64not (v1i64 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2653 def : Pat<(AArch64not (v4i32 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2654 def : Pat<(AArch64not (v2i64 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2656 def : Pat<(vnot (v4i16 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2657 def : Pat<(vnot (v8i16 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2658 def : Pat<(vnot (v2i32 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2659 def : Pat<(vnot (v4i32 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2660 def : Pat<(vnot (v2i64 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2662 defm RBIT : SIMDTwoVectorB<1, 0b01, 0b00101, "rbit", int_aarch64_neon_rbit>;
2663 defm REV16 : SIMDTwoVectorB<0, 0b00, 0b00001, "rev16", AArch64rev16>;
2664 defm REV32 : SIMDTwoVectorBH<1, 0b00000, "rev32", AArch64rev32>;
2665 defm REV64 : SIMDTwoVectorBHS<0, 0b00000, "rev64", AArch64rev64>;
2666 defm SADALP : SIMDLongTwoVectorTied<0, 0b00110, "sadalp",
2667 BinOpFrag<(add node:$LHS, (int_aarch64_neon_saddlp node:$RHS))> >;
2668 defm SADDLP : SIMDLongTwoVector<0, 0b00010, "saddlp", int_aarch64_neon_saddlp>;
2669 defm SCVTF : SIMDTwoVectorIntToFP<0, 0, 0b11101, "scvtf", sint_to_fp>;
2670 defm SHLL : SIMDVectorLShiftLongBySizeBHS;
2671 defm SQABS : SIMDTwoVectorBHSD<0, 0b00111, "sqabs", int_aarch64_neon_sqabs>;
2672 defm SQNEG : SIMDTwoVectorBHSD<1, 0b00111, "sqneg", int_aarch64_neon_sqneg>;
2673 defm SQXTN : SIMDMixedTwoVector<0, 0b10100, "sqxtn", int_aarch64_neon_sqxtn>;
2674 defm SQXTUN : SIMDMixedTwoVector<1, 0b10010, "sqxtun", int_aarch64_neon_sqxtun>;
2675 defm SUQADD : SIMDTwoVectorBHSDTied<0, 0b00011, "suqadd",int_aarch64_neon_suqadd>;
2676 defm UADALP : SIMDLongTwoVectorTied<1, 0b00110, "uadalp",
2677 BinOpFrag<(add node:$LHS, (int_aarch64_neon_uaddlp node:$RHS))> >;
2678 defm UADDLP : SIMDLongTwoVector<1, 0b00010, "uaddlp",
2679 int_aarch64_neon_uaddlp>;
2680 defm UCVTF : SIMDTwoVectorIntToFP<1, 0, 0b11101, "ucvtf", uint_to_fp>;
2681 defm UQXTN : SIMDMixedTwoVector<1, 0b10100, "uqxtn", int_aarch64_neon_uqxtn>;
2682 defm URECPE : SIMDTwoVectorS<0, 1, 0b11100, "urecpe", int_aarch64_neon_urecpe>;
2683 defm URSQRTE: SIMDTwoVectorS<1, 1, 0b11100, "ursqrte", int_aarch64_neon_ursqrte>;
2684 defm USQADD : SIMDTwoVectorBHSDTied<1, 0b00011, "usqadd",int_aarch64_neon_usqadd>;
2685 defm XTN : SIMDMixedTwoVector<0, 0b10010, "xtn", trunc>;
2687 def : Pat<(v4f16 (AArch64rev32 V64:$Rn)), (REV32v4i16 V64:$Rn)>;
2688 def : Pat<(v4f16 (AArch64rev64 V64:$Rn)), (REV64v4i16 V64:$Rn)>;
2689 def : Pat<(v8f16 (AArch64rev32 V128:$Rn)), (REV32v8i16 V128:$Rn)>;
2690 def : Pat<(v8f16 (AArch64rev64 V128:$Rn)), (REV64v8i16 V128:$Rn)>;
2691 def : Pat<(v2f32 (AArch64rev64 V64:$Rn)), (REV64v2i32 V64:$Rn)>;
2692 def : Pat<(v4f32 (AArch64rev64 V128:$Rn)), (REV64v4i32 V128:$Rn)>;
2694 // Patterns for vector long shift (by element width). These need to match all
2695 // three of zext, sext and anyext so it's easier to pull the patterns out of the
2697 multiclass SIMDVectorLShiftLongBySizeBHSPats<SDPatternOperator ext> {
2698 def : Pat<(AArch64vshl (v8i16 (ext (v8i8 V64:$Rn))), (i32 8)),
2699 (SHLLv8i8 V64:$Rn)>;
2700 def : Pat<(AArch64vshl (v8i16 (ext (extract_high_v16i8 V128:$Rn))), (i32 8)),
2701 (SHLLv16i8 V128:$Rn)>;
2702 def : Pat<(AArch64vshl (v4i32 (ext (v4i16 V64:$Rn))), (i32 16)),
2703 (SHLLv4i16 V64:$Rn)>;
2704 def : Pat<(AArch64vshl (v4i32 (ext (extract_high_v8i16 V128:$Rn))), (i32 16)),
2705 (SHLLv8i16 V128:$Rn)>;
2706 def : Pat<(AArch64vshl (v2i64 (ext (v2i32 V64:$Rn))), (i32 32)),
2707 (SHLLv2i32 V64:$Rn)>;
2708 def : Pat<(AArch64vshl (v2i64 (ext (extract_high_v4i32 V128:$Rn))), (i32 32)),
2709 (SHLLv4i32 V128:$Rn)>;
2712 defm : SIMDVectorLShiftLongBySizeBHSPats<anyext>;
2713 defm : SIMDVectorLShiftLongBySizeBHSPats<zext>;
2714 defm : SIMDVectorLShiftLongBySizeBHSPats<sext>;
2716 //===----------------------------------------------------------------------===//
2717 // Advanced SIMD three vector instructions.
2718 //===----------------------------------------------------------------------===//
2720 defm ADD : SIMDThreeSameVector<0, 0b10000, "add", add>;
2721 defm ADDP : SIMDThreeSameVector<0, 0b10111, "addp", int_aarch64_neon_addp>;
2722 defm CMEQ : SIMDThreeSameVector<1, 0b10001, "cmeq", AArch64cmeq>;
2723 defm CMGE : SIMDThreeSameVector<0, 0b00111, "cmge", AArch64cmge>;
2724 defm CMGT : SIMDThreeSameVector<0, 0b00110, "cmgt", AArch64cmgt>;
2725 defm CMHI : SIMDThreeSameVector<1, 0b00110, "cmhi", AArch64cmhi>;
2726 defm CMHS : SIMDThreeSameVector<1, 0b00111, "cmhs", AArch64cmhs>;
2727 defm CMTST : SIMDThreeSameVector<0, 0b10001, "cmtst", AArch64cmtst>;
2728 defm FABD : SIMDThreeSameVectorFP<1,1,0b11010,"fabd", int_aarch64_neon_fabd>;
2729 defm FACGE : SIMDThreeSameVectorFPCmp<1,0,0b11101,"facge",int_aarch64_neon_facge>;
2730 defm FACGT : SIMDThreeSameVectorFPCmp<1,1,0b11101,"facgt",int_aarch64_neon_facgt>;
2731 defm FADDP : SIMDThreeSameVectorFP<1,0,0b11010,"faddp",int_aarch64_neon_addp>;
2732 defm FADD : SIMDThreeSameVectorFP<0,0,0b11010,"fadd", fadd>;
2733 defm FCMEQ : SIMDThreeSameVectorFPCmp<0, 0, 0b11100, "fcmeq", AArch64fcmeq>;
2734 defm FCMGE : SIMDThreeSameVectorFPCmp<1, 0, 0b11100, "fcmge", AArch64fcmge>;
2735 defm FCMGT : SIMDThreeSameVectorFPCmp<1, 1, 0b11100, "fcmgt", AArch64fcmgt>;
2736 defm FDIV : SIMDThreeSameVectorFP<1,0,0b11111,"fdiv", fdiv>;
2737 defm FMAXNMP : SIMDThreeSameVectorFP<1,0,0b11000,"fmaxnmp", int_aarch64_neon_fmaxnmp>;
2738 defm FMAXNM : SIMDThreeSameVectorFP<0,0,0b11000,"fmaxnm", int_aarch64_neon_fmaxnm>;
2739 defm FMAXP : SIMDThreeSameVectorFP<1,0,0b11110,"fmaxp", int_aarch64_neon_fmaxp>;
2740 defm FMAX : SIMDThreeSameVectorFP<0,0,0b11110,"fmax", AArch64fmax>;
2741 defm FMINNMP : SIMDThreeSameVectorFP<1,1,0b11000,"fminnmp", int_aarch64_neon_fminnmp>;
2742 defm FMINNM : SIMDThreeSameVectorFP<0,1,0b11000,"fminnm", int_aarch64_neon_fminnm>;
2743 defm FMINP : SIMDThreeSameVectorFP<1,1,0b11110,"fminp", int_aarch64_neon_fminp>;
2744 defm FMIN : SIMDThreeSameVectorFP<0,1,0b11110,"fmin", AArch64fmin>;
2746 // NOTE: The operands of the PatFrag are reordered on FMLA/FMLS because the
2747 // instruction expects the addend first, while the fma intrinsic puts it last.
2748 defm FMLA : SIMDThreeSameVectorFPTied<0, 0, 0b11001, "fmla",
2749 TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)> >;
2750 defm FMLS : SIMDThreeSameVectorFPTied<0, 1, 0b11001, "fmls",
2751 TriOpFrag<(fma node:$MHS, (fneg node:$RHS), node:$LHS)> >;
2753 // The following def pats catch the case where the LHS of an FMA is negated.
2754 // The TriOpFrag above catches the case where the middle operand is negated.
2755 def : Pat<(v2f32 (fma (fneg V64:$Rn), V64:$Rm, V64:$Rd)),
2756 (FMLSv2f32 V64:$Rd, V64:$Rn, V64:$Rm)>;
2758 def : Pat<(v4f32 (fma (fneg V128:$Rn), V128:$Rm, V128:$Rd)),
2759 (FMLSv4f32 V128:$Rd, V128:$Rn, V128:$Rm)>;
2761 def : Pat<(v2f64 (fma (fneg V128:$Rn), V128:$Rm, V128:$Rd)),
2762 (FMLSv2f64 V128:$Rd, V128:$Rn, V128:$Rm)>;
2764 defm FMULX : SIMDThreeSameVectorFP<0,0,0b11011,"fmulx", int_aarch64_neon_fmulx>;
2765 defm FMUL : SIMDThreeSameVectorFP<1,0,0b11011,"fmul", fmul>;
2766 defm FRECPS : SIMDThreeSameVectorFP<0,0,0b11111,"frecps", int_aarch64_neon_frecps>;
2767 defm FRSQRTS : SIMDThreeSameVectorFP<0,1,0b11111,"frsqrts", int_aarch64_neon_frsqrts>;
2768 defm FSUB : SIMDThreeSameVectorFP<0,1,0b11010,"fsub", fsub>;
2769 defm MLA : SIMDThreeSameVectorBHSTied<0, 0b10010, "mla",
2770 TriOpFrag<(add node:$LHS, (mul node:$MHS, node:$RHS))> >;
2771 defm MLS : SIMDThreeSameVectorBHSTied<1, 0b10010, "mls",
2772 TriOpFrag<(sub node:$LHS, (mul node:$MHS, node:$RHS))> >;
2773 defm MUL : SIMDThreeSameVectorBHS<0, 0b10011, "mul", mul>;
2774 defm PMUL : SIMDThreeSameVectorB<1, 0b10011, "pmul", int_aarch64_neon_pmul>;
2775 defm SABA : SIMDThreeSameVectorBHSTied<0, 0b01111, "saba",
2776 TriOpFrag<(add node:$LHS, (int_aarch64_neon_sabd node:$MHS, node:$RHS))> >;
2777 defm SABD : SIMDThreeSameVectorBHS<0,0b01110,"sabd", int_aarch64_neon_sabd>;
2778 defm SHADD : SIMDThreeSameVectorBHS<0,0b00000,"shadd", int_aarch64_neon_shadd>;
2779 defm SHSUB : SIMDThreeSameVectorBHS<0,0b00100,"shsub", int_aarch64_neon_shsub>;
2780 defm SMAXP : SIMDThreeSameVectorBHS<0,0b10100,"smaxp", int_aarch64_neon_smaxp>;
2781 defm SMAX : SIMDThreeSameVectorBHS<0,0b01100,"smax", int_aarch64_neon_smax>;
2782 defm SMINP : SIMDThreeSameVectorBHS<0,0b10101,"sminp", int_aarch64_neon_sminp>;
2783 defm SMIN : SIMDThreeSameVectorBHS<0,0b01101,"smin", int_aarch64_neon_smin>;
2784 defm SQADD : SIMDThreeSameVector<0,0b00001,"sqadd", int_aarch64_neon_sqadd>;
2785 defm SQDMULH : SIMDThreeSameVectorHS<0,0b10110,"sqdmulh",int_aarch64_neon_sqdmulh>;
2786 defm SQRDMULH : SIMDThreeSameVectorHS<1,0b10110,"sqrdmulh",int_aarch64_neon_sqrdmulh>;
2787 defm SQRSHL : SIMDThreeSameVector<0,0b01011,"sqrshl", int_aarch64_neon_sqrshl>;
2788 defm SQSHL : SIMDThreeSameVector<0,0b01001,"sqshl", int_aarch64_neon_sqshl>;
2789 defm SQSUB : SIMDThreeSameVector<0,0b00101,"sqsub", int_aarch64_neon_sqsub>;
2790 defm SRHADD : SIMDThreeSameVectorBHS<0,0b00010,"srhadd",int_aarch64_neon_srhadd>;
2791 defm SRSHL : SIMDThreeSameVector<0,0b01010,"srshl", int_aarch64_neon_srshl>;
2792 defm SSHL : SIMDThreeSameVector<0,0b01000,"sshl", int_aarch64_neon_sshl>;
2793 defm SUB : SIMDThreeSameVector<1,0b10000,"sub", sub>;
2794 defm UABA : SIMDThreeSameVectorBHSTied<1, 0b01111, "uaba",
2795 TriOpFrag<(add node:$LHS, (int_aarch64_neon_uabd node:$MHS, node:$RHS))> >;
2796 defm UABD : SIMDThreeSameVectorBHS<1,0b01110,"uabd", int_aarch64_neon_uabd>;
2797 defm UHADD : SIMDThreeSameVectorBHS<1,0b00000,"uhadd", int_aarch64_neon_uhadd>;
2798 defm UHSUB : SIMDThreeSameVectorBHS<1,0b00100,"uhsub", int_aarch64_neon_uhsub>;
2799 defm UMAXP : SIMDThreeSameVectorBHS<1,0b10100,"umaxp", int_aarch64_neon_umaxp>;
2800 defm UMAX : SIMDThreeSameVectorBHS<1,0b01100,"umax", int_aarch64_neon_umax>;
2801 defm UMINP : SIMDThreeSameVectorBHS<1,0b10101,"uminp", int_aarch64_neon_uminp>;
2802 defm UMIN : SIMDThreeSameVectorBHS<1,0b01101,"umin", int_aarch64_neon_umin>;
2803 defm UQADD : SIMDThreeSameVector<1,0b00001,"uqadd", int_aarch64_neon_uqadd>;
2804 defm UQRSHL : SIMDThreeSameVector<1,0b01011,"uqrshl", int_aarch64_neon_uqrshl>;
2805 defm UQSHL : SIMDThreeSameVector<1,0b01001,"uqshl", int_aarch64_neon_uqshl>;
2806 defm UQSUB : SIMDThreeSameVector<1,0b00101,"uqsub", int_aarch64_neon_uqsub>;
2807 defm URHADD : SIMDThreeSameVectorBHS<1,0b00010,"urhadd", int_aarch64_neon_urhadd>;
2808 defm URSHL : SIMDThreeSameVector<1,0b01010,"urshl", int_aarch64_neon_urshl>;
2809 defm USHL : SIMDThreeSameVector<1,0b01000,"ushl", int_aarch64_neon_ushl>;
2810 defm SQRDMLAH : SIMDThreeSameVectorSQRDMLxHTiedHS<1,0b10000,"sqrdmlah",
2811 int_aarch64_neon_sqadd>;
2812 defm SQRDMLSH : SIMDThreeSameVectorSQRDMLxHTiedHS<1,0b10001,"sqrdmlsh",
2813 int_aarch64_neon_sqsub>;
2815 defm AND : SIMDLogicalThreeVector<0, 0b00, "and", and>;
2816 defm BIC : SIMDLogicalThreeVector<0, 0b01, "bic",
2817 BinOpFrag<(and node:$LHS, (vnot node:$RHS))> >;
2818 defm BIF : SIMDLogicalThreeVector<1, 0b11, "bif">;
2819 defm BIT : SIMDLogicalThreeVectorTied<1, 0b10, "bit", AArch64bit>;
2820 defm BSL : SIMDLogicalThreeVectorTied<1, 0b01, "bsl",
2821 TriOpFrag<(or (and node:$LHS, node:$MHS), (and (vnot node:$LHS), node:$RHS))>>;
2822 defm EOR : SIMDLogicalThreeVector<1, 0b00, "eor", xor>;
2823 defm ORN : SIMDLogicalThreeVector<0, 0b11, "orn",
2824 BinOpFrag<(or node:$LHS, (vnot node:$RHS))> >;
2825 defm ORR : SIMDLogicalThreeVector<0, 0b10, "orr", or>;
2827 def : Pat<(v8i8 (smin V64:$Rn, V64:$Rm)),
2828 (SMINv8i8 V64:$Rn, V64:$Rm)>;
2829 def : Pat<(v4i16 (smin V64:$Rn, V64:$Rm)),
2830 (SMINv4i16 V64:$Rn, V64:$Rm)>;
2831 def : Pat<(v2i32 (smin V64:$Rn, V64:$Rm)),
2832 (SMINv2i32 V64:$Rn, V64:$Rm)>;
2833 def : Pat<(v16i8 (smin V128:$Rn, V128:$Rm)),
2834 (SMINv16i8 V128:$Rn, V128:$Rm)>;
2835 def : Pat<(v8i16 (smin V128:$Rn, V128:$Rm)),
2836 (SMINv8i16 V128:$Rn, V128:$Rm)>;
2837 def : Pat<(v4i32 (smin V128:$Rn, V128:$Rm)),
2838 (SMINv4i32 V128:$Rn, V128:$Rm)>;
2839 def : Pat<(v8i8 (smax V64:$Rn, V64:$Rm)),
2840 (SMAXv8i8 V64:$Rn, V64:$Rm)>;
2841 def : Pat<(v4i16 (smax V64:$Rn, V64:$Rm)),
2842 (SMAXv4i16 V64:$Rn, V64:$Rm)>;
2843 def : Pat<(v2i32 (smax V64:$Rn, V64:$Rm)),
2844 (SMAXv2i32 V64:$Rn, V64:$Rm)>;
2845 def : Pat<(v16i8 (smax V128:$Rn, V128:$Rm)),
2846 (SMAXv16i8 V128:$Rn, V128:$Rm)>;
2847 def : Pat<(v8i16 (smax V128:$Rn, V128:$Rm)),
2848 (SMAXv8i16 V128:$Rn, V128:$Rm)>;
2849 def : Pat<(v4i32 (smax V128:$Rn, V128:$Rm)),
2850 (SMAXv4i32 V128:$Rn, V128:$Rm)>;
2851 def : Pat<(v8i8 (umin V64:$Rn, V64:$Rm)),
2852 (UMINv8i8 V64:$Rn, V64:$Rm)>;
2853 def : Pat<(v4i16 (umin V64:$Rn, V64:$Rm)),
2854 (UMINv4i16 V64:$Rn, V64:$Rm)>;
2855 def : Pat<(v2i32 (umin V64:$Rn, V64:$Rm)),
2856 (UMINv2i32 V64:$Rn, V64:$Rm)>;
2857 def : Pat<(v16i8 (umin V128:$Rn, V128:$Rm)),
2858 (UMINv16i8 V128:$Rn, V128:$Rm)>;
2859 def : Pat<(v8i16 (umin V128:$Rn, V128:$Rm)),
2860 (UMINv8i16 V128:$Rn, V128:$Rm)>;
2861 def : Pat<(v4i32 (umin V128:$Rn, V128:$Rm)),
2862 (UMINv4i32 V128:$Rn, V128:$Rm)>;
2863 def : Pat<(v8i8 (umax V64:$Rn, V64:$Rm)),
2864 (UMAXv8i8 V64:$Rn, V64:$Rm)>;
2865 def : Pat<(v4i16 (umax V64:$Rn, V64:$Rm)),
2866 (UMAXv4i16 V64:$Rn, V64:$Rm)>;
2867 def : Pat<(v2i32 (umax V64:$Rn, V64:$Rm)),
2868 (UMAXv2i32 V64:$Rn, V64:$Rm)>;
2869 def : Pat<(v16i8 (umax V128:$Rn, V128:$Rm)),
2870 (UMAXv16i8 V128:$Rn, V128:$Rm)>;
2871 def : Pat<(v8i16 (umax V128:$Rn, V128:$Rm)),
2872 (UMAXv8i16 V128:$Rn, V128:$Rm)>;
2873 def : Pat<(v4i32 (umax V128:$Rn, V128:$Rm)),
2874 (UMAXv4i32 V128:$Rn, V128:$Rm)>;
2876 def : Pat<(AArch64bsl (v8i8 V64:$Rd), V64:$Rn, V64:$Rm),
2877 (BSLv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
2878 def : Pat<(AArch64bsl (v4i16 V64:$Rd), V64:$Rn, V64:$Rm),
2879 (BSLv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
2880 def : Pat<(AArch64bsl (v2i32 V64:$Rd), V64:$Rn, V64:$Rm),
2881 (BSLv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
2882 def : Pat<(AArch64bsl (v1i64 V64:$Rd), V64:$Rn, V64:$Rm),
2883 (BSLv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
2885 def : Pat<(AArch64bsl (v16i8 V128:$Rd), V128:$Rn, V128:$Rm),
2886 (BSLv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
2887 def : Pat<(AArch64bsl (v8i16 V128:$Rd), V128:$Rn, V128:$Rm),
2888 (BSLv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
2889 def : Pat<(AArch64bsl (v4i32 V128:$Rd), V128:$Rn, V128:$Rm),
2890 (BSLv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
2891 def : Pat<(AArch64bsl (v2i64 V128:$Rd), V128:$Rn, V128:$Rm),
2892 (BSLv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
2894 def : InstAlias<"mov{\t$dst.16b, $src.16b|.16b\t$dst, $src}",
2895 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 1>;
2896 def : InstAlias<"mov{\t$dst.8h, $src.8h|.8h\t$dst, $src}",
2897 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2898 def : InstAlias<"mov{\t$dst.4s, $src.4s|.4s\t$dst, $src}",
2899 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2900 def : InstAlias<"mov{\t$dst.2d, $src.2d|.2d\t$dst, $src}",
2901 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2903 def : InstAlias<"mov{\t$dst.8b, $src.8b|.8b\t$dst, $src}",
2904 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 1>;
2905 def : InstAlias<"mov{\t$dst.4h, $src.4h|.4h\t$dst, $src}",
2906 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2907 def : InstAlias<"mov{\t$dst.2s, $src.2s|.2s\t$dst, $src}",
2908 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2909 def : InstAlias<"mov{\t$dst.1d, $src.1d|.1d\t$dst, $src}",
2910 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2912 def : InstAlias<"{cmls\t$dst.8b, $src1.8b, $src2.8b" #
2913 "|cmls.8b\t$dst, $src1, $src2}",
2914 (CMHSv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2915 def : InstAlias<"{cmls\t$dst.16b, $src1.16b, $src2.16b" #
2916 "|cmls.16b\t$dst, $src1, $src2}",
2917 (CMHSv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2918 def : InstAlias<"{cmls\t$dst.4h, $src1.4h, $src2.4h" #
2919 "|cmls.4h\t$dst, $src1, $src2}",
2920 (CMHSv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2921 def : InstAlias<"{cmls\t$dst.8h, $src1.8h, $src2.8h" #
2922 "|cmls.8h\t$dst, $src1, $src2}",
2923 (CMHSv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2924 def : InstAlias<"{cmls\t$dst.2s, $src1.2s, $src2.2s" #
2925 "|cmls.2s\t$dst, $src1, $src2}",
2926 (CMHSv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2927 def : InstAlias<"{cmls\t$dst.4s, $src1.4s, $src2.4s" #
2928 "|cmls.4s\t$dst, $src1, $src2}",
2929 (CMHSv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2930 def : InstAlias<"{cmls\t$dst.2d, $src1.2d, $src2.2d" #
2931 "|cmls.2d\t$dst, $src1, $src2}",
2932 (CMHSv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2934 def : InstAlias<"{cmlo\t$dst.8b, $src1.8b, $src2.8b" #
2935 "|cmlo.8b\t$dst, $src1, $src2}",
2936 (CMHIv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2937 def : InstAlias<"{cmlo\t$dst.16b, $src1.16b, $src2.16b" #
2938 "|cmlo.16b\t$dst, $src1, $src2}",
2939 (CMHIv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2940 def : InstAlias<"{cmlo\t$dst.4h, $src1.4h, $src2.4h" #
2941 "|cmlo.4h\t$dst, $src1, $src2}",
2942 (CMHIv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2943 def : InstAlias<"{cmlo\t$dst.8h, $src1.8h, $src2.8h" #
2944 "|cmlo.8h\t$dst, $src1, $src2}",
2945 (CMHIv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2946 def : InstAlias<"{cmlo\t$dst.2s, $src1.2s, $src2.2s" #
2947 "|cmlo.2s\t$dst, $src1, $src2}",
2948 (CMHIv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2949 def : InstAlias<"{cmlo\t$dst.4s, $src1.4s, $src2.4s" #
2950 "|cmlo.4s\t$dst, $src1, $src2}",
2951 (CMHIv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2952 def : InstAlias<"{cmlo\t$dst.2d, $src1.2d, $src2.2d" #
2953 "|cmlo.2d\t$dst, $src1, $src2}",
2954 (CMHIv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2956 def : InstAlias<"{cmle\t$dst.8b, $src1.8b, $src2.8b" #
2957 "|cmle.8b\t$dst, $src1, $src2}",
2958 (CMGEv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2959 def : InstAlias<"{cmle\t$dst.16b, $src1.16b, $src2.16b" #
2960 "|cmle.16b\t$dst, $src1, $src2}",
2961 (CMGEv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2962 def : InstAlias<"{cmle\t$dst.4h, $src1.4h, $src2.4h" #
2963 "|cmle.4h\t$dst, $src1, $src2}",
2964 (CMGEv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2965 def : InstAlias<"{cmle\t$dst.8h, $src1.8h, $src2.8h" #
2966 "|cmle.8h\t$dst, $src1, $src2}",
2967 (CMGEv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2968 def : InstAlias<"{cmle\t$dst.2s, $src1.2s, $src2.2s" #
2969 "|cmle.2s\t$dst, $src1, $src2}",
2970 (CMGEv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2971 def : InstAlias<"{cmle\t$dst.4s, $src1.4s, $src2.4s" #
2972 "|cmle.4s\t$dst, $src1, $src2}",
2973 (CMGEv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2974 def : InstAlias<"{cmle\t$dst.2d, $src1.2d, $src2.2d" #
2975 "|cmle.2d\t$dst, $src1, $src2}",
2976 (CMGEv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2978 def : InstAlias<"{cmlt\t$dst.8b, $src1.8b, $src2.8b" #
2979 "|cmlt.8b\t$dst, $src1, $src2}",
2980 (CMGTv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2981 def : InstAlias<"{cmlt\t$dst.16b, $src1.16b, $src2.16b" #
2982 "|cmlt.16b\t$dst, $src1, $src2}",
2983 (CMGTv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2984 def : InstAlias<"{cmlt\t$dst.4h, $src1.4h, $src2.4h" #
2985 "|cmlt.4h\t$dst, $src1, $src2}",
2986 (CMGTv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2987 def : InstAlias<"{cmlt\t$dst.8h, $src1.8h, $src2.8h" #
2988 "|cmlt.8h\t$dst, $src1, $src2}",
2989 (CMGTv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2990 def : InstAlias<"{cmlt\t$dst.2s, $src1.2s, $src2.2s" #
2991 "|cmlt.2s\t$dst, $src1, $src2}",
2992 (CMGTv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2993 def : InstAlias<"{cmlt\t$dst.4s, $src1.4s, $src2.4s" #
2994 "|cmlt.4s\t$dst, $src1, $src2}",
2995 (CMGTv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2996 def : InstAlias<"{cmlt\t$dst.2d, $src1.2d, $src2.2d" #
2997 "|cmlt.2d\t$dst, $src1, $src2}",
2998 (CMGTv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
3000 def : InstAlias<"{fcmle\t$dst.2s, $src1.2s, $src2.2s" #
3001 "|fcmle.2s\t$dst, $src1, $src2}",
3002 (FCMGEv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
3003 def : InstAlias<"{fcmle\t$dst.4s, $src1.4s, $src2.4s" #
3004 "|fcmle.4s\t$dst, $src1, $src2}",
3005 (FCMGEv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
3006 def : InstAlias<"{fcmle\t$dst.2d, $src1.2d, $src2.2d" #
3007 "|fcmle.2d\t$dst, $src1, $src2}",
3008 (FCMGEv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
3010 def : InstAlias<"{fcmlt\t$dst.2s, $src1.2s, $src2.2s" #
3011 "|fcmlt.2s\t$dst, $src1, $src2}",
3012 (FCMGTv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
3013 def : InstAlias<"{fcmlt\t$dst.4s, $src1.4s, $src2.4s" #
3014 "|fcmlt.4s\t$dst, $src1, $src2}",
3015 (FCMGTv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
3016 def : InstAlias<"{fcmlt\t$dst.2d, $src1.2d, $src2.2d" #
3017 "|fcmlt.2d\t$dst, $src1, $src2}",
3018 (FCMGTv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
3020 def : InstAlias<"{facle\t$dst.2s, $src1.2s, $src2.2s" #
3021 "|facle.2s\t$dst, $src1, $src2}",
3022 (FACGEv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
3023 def : InstAlias<"{facle\t$dst.4s, $src1.4s, $src2.4s" #
3024 "|facle.4s\t$dst, $src1, $src2}",
3025 (FACGEv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
3026 def : InstAlias<"{facle\t$dst.2d, $src1.2d, $src2.2d" #
3027 "|facle.2d\t$dst, $src1, $src2}",
3028 (FACGEv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
3030 def : InstAlias<"{faclt\t$dst.2s, $src1.2s, $src2.2s" #
3031 "|faclt.2s\t$dst, $src1, $src2}",
3032 (FACGTv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
3033 def : InstAlias<"{faclt\t$dst.4s, $src1.4s, $src2.4s" #
3034 "|faclt.4s\t$dst, $src1, $src2}",
3035 (FACGTv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
3036 def : InstAlias<"{faclt\t$dst.2d, $src1.2d, $src2.2d" #
3037 "|faclt.2d\t$dst, $src1, $src2}",
3038 (FACGTv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
3040 //===----------------------------------------------------------------------===//
3041 // Advanced SIMD three scalar instructions.
3042 //===----------------------------------------------------------------------===//
3044 defm ADD : SIMDThreeScalarD<0, 0b10000, "add", add>;
3045 defm CMEQ : SIMDThreeScalarD<1, 0b10001, "cmeq", AArch64cmeq>;
3046 defm CMGE : SIMDThreeScalarD<0, 0b00111, "cmge", AArch64cmge>;
3047 defm CMGT : SIMDThreeScalarD<0, 0b00110, "cmgt", AArch64cmgt>;
3048 defm CMHI : SIMDThreeScalarD<1, 0b00110, "cmhi", AArch64cmhi>;
3049 defm CMHS : SIMDThreeScalarD<1, 0b00111, "cmhs", AArch64cmhs>;
3050 defm CMTST : SIMDThreeScalarD<0, 0b10001, "cmtst", AArch64cmtst>;
3051 defm FABD : SIMDThreeScalarSD<1, 1, 0b11010, "fabd", int_aarch64_sisd_fabd>;
3052 def : Pat<(v1f64 (int_aarch64_neon_fabd (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
3053 (FABD64 FPR64:$Rn, FPR64:$Rm)>;
3054 defm FACGE : SIMDThreeScalarFPCmp<1, 0, 0b11101, "facge",
3055 int_aarch64_neon_facge>;
3056 defm FACGT : SIMDThreeScalarFPCmp<1, 1, 0b11101, "facgt",
3057 int_aarch64_neon_facgt>;
3058 defm FCMEQ : SIMDThreeScalarFPCmp<0, 0, 0b11100, "fcmeq", AArch64fcmeq>;
3059 defm FCMGE : SIMDThreeScalarFPCmp<1, 0, 0b11100, "fcmge", AArch64fcmge>;
3060 defm FCMGT : SIMDThreeScalarFPCmp<1, 1, 0b11100, "fcmgt", AArch64fcmgt>;
3061 defm FMULX : SIMDThreeScalarSD<0, 0, 0b11011, "fmulx", int_aarch64_neon_fmulx>;
3062 defm FRECPS : SIMDThreeScalarSD<0, 0, 0b11111, "frecps", int_aarch64_neon_frecps>;
3063 defm FRSQRTS : SIMDThreeScalarSD<0, 1, 0b11111, "frsqrts", int_aarch64_neon_frsqrts>;
3064 defm SQADD : SIMDThreeScalarBHSD<0, 0b00001, "sqadd", int_aarch64_neon_sqadd>;
3065 defm SQDMULH : SIMDThreeScalarHS< 0, 0b10110, "sqdmulh", int_aarch64_neon_sqdmulh>;
3066 defm SQRDMULH : SIMDThreeScalarHS< 1, 0b10110, "sqrdmulh", int_aarch64_neon_sqrdmulh>;
3067 defm SQRSHL : SIMDThreeScalarBHSD<0, 0b01011, "sqrshl",int_aarch64_neon_sqrshl>;
3068 defm SQSHL : SIMDThreeScalarBHSD<0, 0b01001, "sqshl", int_aarch64_neon_sqshl>;
3069 defm SQSUB : SIMDThreeScalarBHSD<0, 0b00101, "sqsub", int_aarch64_neon_sqsub>;
3070 defm SRSHL : SIMDThreeScalarD< 0, 0b01010, "srshl", int_aarch64_neon_srshl>;
3071 defm SSHL : SIMDThreeScalarD< 0, 0b01000, "sshl", int_aarch64_neon_sshl>;
3072 defm SUB : SIMDThreeScalarD< 1, 0b10000, "sub", sub>;
3073 defm UQADD : SIMDThreeScalarBHSD<1, 0b00001, "uqadd", int_aarch64_neon_uqadd>;
3074 defm UQRSHL : SIMDThreeScalarBHSD<1, 0b01011, "uqrshl",int_aarch64_neon_uqrshl>;
3075 defm UQSHL : SIMDThreeScalarBHSD<1, 0b01001, "uqshl", int_aarch64_neon_uqshl>;
3076 defm UQSUB : SIMDThreeScalarBHSD<1, 0b00101, "uqsub", int_aarch64_neon_uqsub>;
3077 defm URSHL : SIMDThreeScalarD< 1, 0b01010, "urshl", int_aarch64_neon_urshl>;
3078 defm USHL : SIMDThreeScalarD< 1, 0b01000, "ushl", int_aarch64_neon_ushl>;
3079 let Predicates = [HasV8_1a] in {
3080 defm SQRDMLAH : SIMDThreeScalarHSTied<1, 0, 0b10000, "sqrdmlah">;
3081 defm SQRDMLSH : SIMDThreeScalarHSTied<1, 0, 0b10001, "sqrdmlsh">;
3082 def : Pat<(i32 (int_aarch64_neon_sqadd
3084 (i32 (int_aarch64_neon_sqrdmulh (i32 FPR32:$Rn),
3085 (i32 FPR32:$Rm))))),
3086 (SQRDMLAHv1i32 FPR32:$Rd, FPR32:$Rn, FPR32:$Rm)>;
3087 def : Pat<(i32 (int_aarch64_neon_sqsub
3089 (i32 (int_aarch64_neon_sqrdmulh (i32 FPR32:$Rn),
3090 (i32 FPR32:$Rm))))),
3091 (SQRDMLSHv1i32 FPR32:$Rd, FPR32:$Rn, FPR32:$Rm)>;
3094 def : InstAlias<"cmls $dst, $src1, $src2",
3095 (CMHSv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
3096 def : InstAlias<"cmle $dst, $src1, $src2",
3097 (CMGEv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
3098 def : InstAlias<"cmlo $dst, $src1, $src2",
3099 (CMHIv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
3100 def : InstAlias<"cmlt $dst, $src1, $src2",
3101 (CMGTv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
3102 def : InstAlias<"fcmle $dst, $src1, $src2",
3103 (FCMGE32 FPR32:$dst, FPR32:$src2, FPR32:$src1), 0>;
3104 def : InstAlias<"fcmle $dst, $src1, $src2",
3105 (FCMGE64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
3106 def : InstAlias<"fcmlt $dst, $src1, $src2",
3107 (FCMGT32 FPR32:$dst, FPR32:$src2, FPR32:$src1), 0>;
3108 def : InstAlias<"fcmlt $dst, $src1, $src2",
3109 (FCMGT64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
3110 def : InstAlias<"facle $dst, $src1, $src2",
3111 (FACGE32 FPR32:$dst, FPR32:$src2, FPR32:$src1), 0>;
3112 def : InstAlias<"facle $dst, $src1, $src2",
3113 (FACGE64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
3114 def : InstAlias<"faclt $dst, $src1, $src2",
3115 (FACGT32 FPR32:$dst, FPR32:$src2, FPR32:$src1), 0>;
3116 def : InstAlias<"faclt $dst, $src1, $src2",
3117 (FACGT64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
3119 //===----------------------------------------------------------------------===//
3120 // Advanced SIMD three scalar instructions (mixed operands).
3121 //===----------------------------------------------------------------------===//
3122 defm SQDMULL : SIMDThreeScalarMixedHS<0, 0b11010, "sqdmull",
3123 int_aarch64_neon_sqdmulls_scalar>;
3124 defm SQDMLAL : SIMDThreeScalarMixedTiedHS<0, 0b10010, "sqdmlal">;
3125 defm SQDMLSL : SIMDThreeScalarMixedTiedHS<0, 0b10110, "sqdmlsl">;
3127 def : Pat<(i64 (int_aarch64_neon_sqadd (i64 FPR64:$Rd),
3128 (i64 (int_aarch64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
3129 (i32 FPR32:$Rm))))),
3130 (SQDMLALi32 FPR64:$Rd, FPR32:$Rn, FPR32:$Rm)>;
3131 def : Pat<(i64 (int_aarch64_neon_sqsub (i64 FPR64:$Rd),
3132 (i64 (int_aarch64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
3133 (i32 FPR32:$Rm))))),
3134 (SQDMLSLi32 FPR64:$Rd, FPR32:$Rn, FPR32:$Rm)>;
3136 //===----------------------------------------------------------------------===//
3137 // Advanced SIMD two scalar instructions.
3138 //===----------------------------------------------------------------------===//
3140 defm ABS : SIMDTwoScalarD< 0, 0b01011, "abs", int_aarch64_neon_abs>;
3141 defm CMEQ : SIMDCmpTwoScalarD< 0, 0b01001, "cmeq", AArch64cmeqz>;
3142 defm CMGE : SIMDCmpTwoScalarD< 1, 0b01000, "cmge", AArch64cmgez>;
3143 defm CMGT : SIMDCmpTwoScalarD< 0, 0b01000, "cmgt", AArch64cmgtz>;
3144 defm CMLE : SIMDCmpTwoScalarD< 1, 0b01001, "cmle", AArch64cmlez>;
3145 defm CMLT : SIMDCmpTwoScalarD< 0, 0b01010, "cmlt", AArch64cmltz>;
3146 defm FCMEQ : SIMDCmpTwoScalarSD<0, 1, 0b01101, "fcmeq", AArch64fcmeqz>;
3147 defm FCMGE : SIMDCmpTwoScalarSD<1, 1, 0b01100, "fcmge", AArch64fcmgez>;
3148 defm FCMGT : SIMDCmpTwoScalarSD<0, 1, 0b01100, "fcmgt", AArch64fcmgtz>;
3149 defm FCMLE : SIMDCmpTwoScalarSD<1, 1, 0b01101, "fcmle", AArch64fcmlez>;
3150 defm FCMLT : SIMDCmpTwoScalarSD<0, 1, 0b01110, "fcmlt", AArch64fcmltz>;
3151 defm FCVTAS : SIMDTwoScalarSD< 0, 0, 0b11100, "fcvtas">;
3152 defm FCVTAU : SIMDTwoScalarSD< 1, 0, 0b11100, "fcvtau">;
3153 defm FCVTMS : SIMDTwoScalarSD< 0, 0, 0b11011, "fcvtms">;
3154 defm FCVTMU : SIMDTwoScalarSD< 1, 0, 0b11011, "fcvtmu">;
3155 defm FCVTNS : SIMDTwoScalarSD< 0, 0, 0b11010, "fcvtns">;
3156 defm FCVTNU : SIMDTwoScalarSD< 1, 0, 0b11010, "fcvtnu">;
3157 defm FCVTPS : SIMDTwoScalarSD< 0, 1, 0b11010, "fcvtps">;
3158 defm FCVTPU : SIMDTwoScalarSD< 1, 1, 0b11010, "fcvtpu">;
3159 def FCVTXNv1i64 : SIMDInexactCvtTwoScalar<0b10110, "fcvtxn">;
3160 defm FCVTZS : SIMDTwoScalarSD< 0, 1, 0b11011, "fcvtzs">;
3161 defm FCVTZU : SIMDTwoScalarSD< 1, 1, 0b11011, "fcvtzu">;
3162 defm FRECPE : SIMDTwoScalarSD< 0, 1, 0b11101, "frecpe">;
3163 defm FRECPX : SIMDTwoScalarSD< 0, 1, 0b11111, "frecpx">;
3164 defm FRSQRTE : SIMDTwoScalarSD< 1, 1, 0b11101, "frsqrte">;
3165 defm NEG : SIMDTwoScalarD< 1, 0b01011, "neg",
3166 UnOpFrag<(sub immAllZerosV, node:$LHS)> >;
3167 defm SCVTF : SIMDTwoScalarCVTSD< 0, 0, 0b11101, "scvtf", AArch64sitof>;
3168 defm SQABS : SIMDTwoScalarBHSD< 0, 0b00111, "sqabs", int_aarch64_neon_sqabs>;
3169 defm SQNEG : SIMDTwoScalarBHSD< 1, 0b00111, "sqneg", int_aarch64_neon_sqneg>;
3170 defm SQXTN : SIMDTwoScalarMixedBHS< 0, 0b10100, "sqxtn", int_aarch64_neon_scalar_sqxtn>;
3171 defm SQXTUN : SIMDTwoScalarMixedBHS< 1, 0b10010, "sqxtun", int_aarch64_neon_scalar_sqxtun>;
3172 defm SUQADD : SIMDTwoScalarBHSDTied< 0, 0b00011, "suqadd",
3173 int_aarch64_neon_suqadd>;
3174 defm UCVTF : SIMDTwoScalarCVTSD< 1, 0, 0b11101, "ucvtf", AArch64uitof>;
3175 defm UQXTN : SIMDTwoScalarMixedBHS<1, 0b10100, "uqxtn", int_aarch64_neon_scalar_uqxtn>;
3176 defm USQADD : SIMDTwoScalarBHSDTied< 1, 0b00011, "usqadd",
3177 int_aarch64_neon_usqadd>;
3179 def : Pat<(AArch64neg (v1i64 V64:$Rn)), (NEGv1i64 V64:$Rn)>;
3181 def : Pat<(v1i64 (int_aarch64_neon_fcvtas (v1f64 FPR64:$Rn))),
3182 (FCVTASv1i64 FPR64:$Rn)>;
3183 def : Pat<(v1i64 (int_aarch64_neon_fcvtau (v1f64 FPR64:$Rn))),
3184 (FCVTAUv1i64 FPR64:$Rn)>;
3185 def : Pat<(v1i64 (int_aarch64_neon_fcvtms (v1f64 FPR64:$Rn))),
3186 (FCVTMSv1i64 FPR64:$Rn)>;
3187 def : Pat<(v1i64 (int_aarch64_neon_fcvtmu (v1f64 FPR64:$Rn))),
3188 (FCVTMUv1i64 FPR64:$Rn)>;
3189 def : Pat<(v1i64 (int_aarch64_neon_fcvtns (v1f64 FPR64:$Rn))),
3190 (FCVTNSv1i64 FPR64:$Rn)>;
3191 def : Pat<(v1i64 (int_aarch64_neon_fcvtnu (v1f64 FPR64:$Rn))),
3192 (FCVTNUv1i64 FPR64:$Rn)>;
3193 def : Pat<(v1i64 (int_aarch64_neon_fcvtps (v1f64 FPR64:$Rn))),
3194 (FCVTPSv1i64 FPR64:$Rn)>;
3195 def : Pat<(v1i64 (int_aarch64_neon_fcvtpu (v1f64 FPR64:$Rn))),
3196 (FCVTPUv1i64 FPR64:$Rn)>;
3198 def : Pat<(f32 (int_aarch64_neon_frecpe (f32 FPR32:$Rn))),
3199 (FRECPEv1i32 FPR32:$Rn)>;
3200 def : Pat<(f64 (int_aarch64_neon_frecpe (f64 FPR64:$Rn))),
3201 (FRECPEv1i64 FPR64:$Rn)>;
3202 def : Pat<(v1f64 (int_aarch64_neon_frecpe (v1f64 FPR64:$Rn))),
3203 (FRECPEv1i64 FPR64:$Rn)>;
3205 def : Pat<(f32 (int_aarch64_neon_frecpx (f32 FPR32:$Rn))),
3206 (FRECPXv1i32 FPR32:$Rn)>;
3207 def : Pat<(f64 (int_aarch64_neon_frecpx (f64 FPR64:$Rn))),
3208 (FRECPXv1i64 FPR64:$Rn)>;
3210 def : Pat<(f32 (int_aarch64_neon_frsqrte (f32 FPR32:$Rn))),
3211 (FRSQRTEv1i32 FPR32:$Rn)>;
3212 def : Pat<(f64 (int_aarch64_neon_frsqrte (f64 FPR64:$Rn))),
3213 (FRSQRTEv1i64 FPR64:$Rn)>;
3214 def : Pat<(v1f64 (int_aarch64_neon_frsqrte (v1f64 FPR64:$Rn))),
3215 (FRSQRTEv1i64 FPR64:$Rn)>;
3217 // If an integer is about to be converted to a floating point value,
3218 // just load it on the floating point unit.
3219 // Here are the patterns for 8 and 16-bits to float.
3221 multiclass UIntToFPROLoadPat<ValueType DstTy, ValueType SrcTy,
3222 SDPatternOperator loadop, Instruction UCVTF,
3223 ROAddrMode ro, Instruction LDRW, Instruction LDRX,
3225 def : Pat<(DstTy (uint_to_fp (SrcTy
3226 (loadop (ro.Wpat GPR64sp:$Rn, GPR32:$Rm,
3227 ro.Wext:$extend))))),
3228 (UCVTF (INSERT_SUBREG (DstTy (IMPLICIT_DEF)),
3229 (LDRW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend),
3232 def : Pat<(DstTy (uint_to_fp (SrcTy
3233 (loadop (ro.Xpat GPR64sp:$Rn, GPR64:$Rm,
3234 ro.Wext:$extend))))),
3235 (UCVTF (INSERT_SUBREG (DstTy (IMPLICIT_DEF)),
3236 (LDRX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend),
3240 defm : UIntToFPROLoadPat<f32, i32, zextloadi8,
3241 UCVTFv1i32, ro8, LDRBroW, LDRBroX, bsub>;
3242 def : Pat <(f32 (uint_to_fp (i32
3243 (zextloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))))),
3244 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
3245 (LDRBui GPR64sp:$Rn, uimm12s1:$offset), bsub))>;
3246 def : Pat <(f32 (uint_to_fp (i32
3247 (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))))),
3248 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
3249 (LDURBi GPR64sp:$Rn, simm9:$offset), bsub))>;
3250 // 16-bits -> float.
3251 defm : UIntToFPROLoadPat<f32, i32, zextloadi16,
3252 UCVTFv1i32, ro16, LDRHroW, LDRHroX, hsub>;
3253 def : Pat <(f32 (uint_to_fp (i32
3254 (zextloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))))),
3255 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
3256 (LDRHui GPR64sp:$Rn, uimm12s2:$offset), hsub))>;
3257 def : Pat <(f32 (uint_to_fp (i32
3258 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))))),
3259 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
3260 (LDURHi GPR64sp:$Rn, simm9:$offset), hsub))>;
3261 // 32-bits are handled in target specific dag combine:
3262 // performIntToFpCombine.
3263 // 64-bits integer to 32-bits floating point, not possible with
3264 // UCVTF on floating point registers (both source and destination
3265 // must have the same size).
3267 // Here are the patterns for 8, 16, 32, and 64-bits to double.
3268 // 8-bits -> double.
3269 defm : UIntToFPROLoadPat<f64, i32, zextloadi8,
3270 UCVTFv1i64, ro8, LDRBroW, LDRBroX, bsub>;
3271 def : Pat <(f64 (uint_to_fp (i32
3272 (zextloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))))),
3273 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3274 (LDRBui GPR64sp:$Rn, uimm12s1:$offset), bsub))>;
3275 def : Pat <(f64 (uint_to_fp (i32
3276 (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))))),
3277 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3278 (LDURBi GPR64sp:$Rn, simm9:$offset), bsub))>;
3279 // 16-bits -> double.
3280 defm : UIntToFPROLoadPat<f64, i32, zextloadi16,
3281 UCVTFv1i64, ro16, LDRHroW, LDRHroX, hsub>;
3282 def : Pat <(f64 (uint_to_fp (i32
3283 (zextloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))))),
3284 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3285 (LDRHui GPR64sp:$Rn, uimm12s2:$offset), hsub))>;
3286 def : Pat <(f64 (uint_to_fp (i32
3287 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))))),
3288 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3289 (LDURHi GPR64sp:$Rn, simm9:$offset), hsub))>;
3290 // 32-bits -> double.
3291 defm : UIntToFPROLoadPat<f64, i32, load,
3292 UCVTFv1i64, ro32, LDRSroW, LDRSroX, ssub>;
3293 def : Pat <(f64 (uint_to_fp (i32
3294 (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))))),
3295 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3296 (LDRSui GPR64sp:$Rn, uimm12s4:$offset), ssub))>;
3297 def : Pat <(f64 (uint_to_fp (i32
3298 (load (am_unscaled32 GPR64sp:$Rn, simm9:$offset))))),
3299 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3300 (LDURSi GPR64sp:$Rn, simm9:$offset), ssub))>;
3301 // 64-bits -> double are handled in target specific dag combine:
3302 // performIntToFpCombine.
3304 //===----------------------------------------------------------------------===//
3305 // Advanced SIMD three different-sized vector instructions.
3306 //===----------------------------------------------------------------------===//
3308 defm ADDHN : SIMDNarrowThreeVectorBHS<0,0b0100,"addhn", int_aarch64_neon_addhn>;
3309 defm SUBHN : SIMDNarrowThreeVectorBHS<0,0b0110,"subhn", int_aarch64_neon_subhn>;
3310 defm RADDHN : SIMDNarrowThreeVectorBHS<1,0b0100,"raddhn",int_aarch64_neon_raddhn>;
3311 defm RSUBHN : SIMDNarrowThreeVectorBHS<1,0b0110,"rsubhn",int_aarch64_neon_rsubhn>;
3312 defm PMULL : SIMDDifferentThreeVectorBD<0,0b1110,"pmull",int_aarch64_neon_pmull>;
3313 defm SABAL : SIMDLongThreeVectorTiedBHSabal<0,0b0101,"sabal",
3314 int_aarch64_neon_sabd>;
3315 defm SABDL : SIMDLongThreeVectorBHSabdl<0, 0b0111, "sabdl",
3316 int_aarch64_neon_sabd>;
3317 defm SADDL : SIMDLongThreeVectorBHS< 0, 0b0000, "saddl",
3318 BinOpFrag<(add (sext node:$LHS), (sext node:$RHS))>>;
3319 defm SADDW : SIMDWideThreeVectorBHS< 0, 0b0001, "saddw",
3320 BinOpFrag<(add node:$LHS, (sext node:$RHS))>>;
3321 defm SMLAL : SIMDLongThreeVectorTiedBHS<0, 0b1000, "smlal",
3322 TriOpFrag<(add node:$LHS, (int_aarch64_neon_smull node:$MHS, node:$RHS))>>;
3323 defm SMLSL : SIMDLongThreeVectorTiedBHS<0, 0b1010, "smlsl",
3324 TriOpFrag<(sub node:$LHS, (int_aarch64_neon_smull node:$MHS, node:$RHS))>>;
3325 defm SMULL : SIMDLongThreeVectorBHS<0, 0b1100, "smull", int_aarch64_neon_smull>;
3326 defm SQDMLAL : SIMDLongThreeVectorSQDMLXTiedHS<0, 0b1001, "sqdmlal",
3327 int_aarch64_neon_sqadd>;
3328 defm SQDMLSL : SIMDLongThreeVectorSQDMLXTiedHS<0, 0b1011, "sqdmlsl",
3329 int_aarch64_neon_sqsub>;
3330 defm SQDMULL : SIMDLongThreeVectorHS<0, 0b1101, "sqdmull",
3331 int_aarch64_neon_sqdmull>;
3332 defm SSUBL : SIMDLongThreeVectorBHS<0, 0b0010, "ssubl",
3333 BinOpFrag<(sub (sext node:$LHS), (sext node:$RHS))>>;
3334 defm SSUBW : SIMDWideThreeVectorBHS<0, 0b0011, "ssubw",
3335 BinOpFrag<(sub node:$LHS, (sext node:$RHS))>>;
3336 defm UABAL : SIMDLongThreeVectorTiedBHSabal<1, 0b0101, "uabal",
3337 int_aarch64_neon_uabd>;
3338 defm UABDL : SIMDLongThreeVectorBHSabdl<1, 0b0111, "uabdl",
3339 int_aarch64_neon_uabd>;
3340 defm UADDL : SIMDLongThreeVectorBHS<1, 0b0000, "uaddl",
3341 BinOpFrag<(add (zext node:$LHS), (zext node:$RHS))>>;
3342 defm UADDW : SIMDWideThreeVectorBHS<1, 0b0001, "uaddw",
3343 BinOpFrag<(add node:$LHS, (zext node:$RHS))>>;
3344 defm UMLAL : SIMDLongThreeVectorTiedBHS<1, 0b1000, "umlal",
3345 TriOpFrag<(add node:$LHS, (int_aarch64_neon_umull node:$MHS, node:$RHS))>>;
3346 defm UMLSL : SIMDLongThreeVectorTiedBHS<1, 0b1010, "umlsl",
3347 TriOpFrag<(sub node:$LHS, (int_aarch64_neon_umull node:$MHS, node:$RHS))>>;
3348 defm UMULL : SIMDLongThreeVectorBHS<1, 0b1100, "umull", int_aarch64_neon_umull>;
3349 defm USUBL : SIMDLongThreeVectorBHS<1, 0b0010, "usubl",
3350 BinOpFrag<(sub (zext node:$LHS), (zext node:$RHS))>>;
3351 defm USUBW : SIMDWideThreeVectorBHS< 1, 0b0011, "usubw",
3352 BinOpFrag<(sub node:$LHS, (zext node:$RHS))>>;
3354 // Additional patterns for SMULL and UMULL
3355 multiclass Neon_mul_widen_patterns<SDPatternOperator opnode,
3356 Instruction INST8B, Instruction INST4H, Instruction INST2S> {
3357 def : Pat<(v8i16 (opnode (v8i8 V64:$Rn), (v8i8 V64:$Rm))),
3358 (INST8B V64:$Rn, V64:$Rm)>;
3359 def : Pat<(v4i32 (opnode (v4i16 V64:$Rn), (v4i16 V64:$Rm))),
3360 (INST4H V64:$Rn, V64:$Rm)>;
3361 def : Pat<(v2i64 (opnode (v2i32 V64:$Rn), (v2i32 V64:$Rm))),
3362 (INST2S V64:$Rn, V64:$Rm)>;
3365 defm : Neon_mul_widen_patterns<AArch64smull, SMULLv8i8_v8i16,
3366 SMULLv4i16_v4i32, SMULLv2i32_v2i64>;
3367 defm : Neon_mul_widen_patterns<AArch64umull, UMULLv8i8_v8i16,
3368 UMULLv4i16_v4i32, UMULLv2i32_v2i64>;
3370 // Additional patterns for SMLAL/SMLSL and UMLAL/UMLSL
3371 multiclass Neon_mulacc_widen_patterns<SDPatternOperator opnode,
3372 Instruction INST8B, Instruction INST4H, Instruction INST2S> {
3373 def : Pat<(v8i16 (opnode (v8i16 V128:$Rd), (v8i8 V64:$Rn), (v8i8 V64:$Rm))),
3374 (INST8B V128:$Rd, V64:$Rn, V64:$Rm)>;
3375 def : Pat<(v4i32 (opnode (v4i32 V128:$Rd), (v4i16 V64:$Rn), (v4i16 V64:$Rm))),
3376 (INST4H V128:$Rd, V64:$Rn, V64:$Rm)>;
3377 def : Pat<(v2i64 (opnode (v2i64 V128:$Rd), (v2i32 V64:$Rn), (v2i32 V64:$Rm))),
3378 (INST2S V128:$Rd, V64:$Rn, V64:$Rm)>;
3381 defm : Neon_mulacc_widen_patterns<
3382 TriOpFrag<(add node:$LHS, (AArch64smull node:$MHS, node:$RHS))>,
3383 SMLALv8i8_v8i16, SMLALv4i16_v4i32, SMLALv2i32_v2i64>;
3384 defm : Neon_mulacc_widen_patterns<
3385 TriOpFrag<(add node:$LHS, (AArch64umull node:$MHS, node:$RHS))>,
3386 UMLALv8i8_v8i16, UMLALv4i16_v4i32, UMLALv2i32_v2i64>;
3387 defm : Neon_mulacc_widen_patterns<
3388 TriOpFrag<(sub node:$LHS, (AArch64smull node:$MHS, node:$RHS))>,
3389 SMLSLv8i8_v8i16, SMLSLv4i16_v4i32, SMLSLv2i32_v2i64>;
3390 defm : Neon_mulacc_widen_patterns<
3391 TriOpFrag<(sub node:$LHS, (AArch64umull node:$MHS, node:$RHS))>,
3392 UMLSLv8i8_v8i16, UMLSLv4i16_v4i32, UMLSLv2i32_v2i64>;
3394 // Patterns for 64-bit pmull
3395 def : Pat<(int_aarch64_neon_pmull64 V64:$Rn, V64:$Rm),
3396 (PMULLv1i64 V64:$Rn, V64:$Rm)>;
3397 def : Pat<(int_aarch64_neon_pmull64 (vector_extract (v2i64 V128:$Rn), (i64 1)),
3398 (vector_extract (v2i64 V128:$Rm), (i64 1))),
3399 (PMULLv2i64 V128:$Rn, V128:$Rm)>;
3401 // CodeGen patterns for addhn and subhn instructions, which can actually be
3402 // written in LLVM IR without too much difficulty.
3405 def : Pat<(v8i8 (trunc (v8i16 (AArch64vlshr (add V128:$Rn, V128:$Rm), (i32 8))))),
3406 (ADDHNv8i16_v8i8 V128:$Rn, V128:$Rm)>;
3407 def : Pat<(v4i16 (trunc (v4i32 (AArch64vlshr (add V128:$Rn, V128:$Rm),
3409 (ADDHNv4i32_v4i16 V128:$Rn, V128:$Rm)>;
3410 def : Pat<(v2i32 (trunc (v2i64 (AArch64vlshr (add V128:$Rn, V128:$Rm),
3412 (ADDHNv2i64_v2i32 V128:$Rn, V128:$Rm)>;
3413 def : Pat<(concat_vectors (v8i8 V64:$Rd),
3414 (trunc (v8i16 (AArch64vlshr (add V128:$Rn, V128:$Rm),
3416 (ADDHNv8i16_v16i8 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3417 V128:$Rn, V128:$Rm)>;
3418 def : Pat<(concat_vectors (v4i16 V64:$Rd),
3419 (trunc (v4i32 (AArch64vlshr (add V128:$Rn, V128:$Rm),
3421 (ADDHNv4i32_v8i16 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3422 V128:$Rn, V128:$Rm)>;
3423 def : Pat<(concat_vectors (v2i32 V64:$Rd),
3424 (trunc (v2i64 (AArch64vlshr (add V128:$Rn, V128:$Rm),
3426 (ADDHNv2i64_v4i32 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3427 V128:$Rn, V128:$Rm)>;
3430 def : Pat<(v8i8 (trunc (v8i16 (AArch64vlshr (sub V128:$Rn, V128:$Rm), (i32 8))))),
3431 (SUBHNv8i16_v8i8 V128:$Rn, V128:$Rm)>;
3432 def : Pat<(v4i16 (trunc (v4i32 (AArch64vlshr (sub V128:$Rn, V128:$Rm),
3434 (SUBHNv4i32_v4i16 V128:$Rn, V128:$Rm)>;
3435 def : Pat<(v2i32 (trunc (v2i64 (AArch64vlshr (sub V128:$Rn, V128:$Rm),
3437 (SUBHNv2i64_v2i32 V128:$Rn, V128:$Rm)>;
3438 def : Pat<(concat_vectors (v8i8 V64:$Rd),
3439 (trunc (v8i16 (AArch64vlshr (sub V128:$Rn, V128:$Rm),
3441 (SUBHNv8i16_v16i8 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3442 V128:$Rn, V128:$Rm)>;
3443 def : Pat<(concat_vectors (v4i16 V64:$Rd),
3444 (trunc (v4i32 (AArch64vlshr (sub V128:$Rn, V128:$Rm),
3446 (SUBHNv4i32_v8i16 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3447 V128:$Rn, V128:$Rm)>;
3448 def : Pat<(concat_vectors (v2i32 V64:$Rd),
3449 (trunc (v2i64 (AArch64vlshr (sub V128:$Rn, V128:$Rm),
3451 (SUBHNv2i64_v4i32 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3452 V128:$Rn, V128:$Rm)>;
3454 //----------------------------------------------------------------------------
3455 // AdvSIMD bitwise extract from vector instruction.
3456 //----------------------------------------------------------------------------
3458 defm EXT : SIMDBitwiseExtract<"ext">;
3460 def : Pat<(v4i16 (AArch64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
3461 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
3462 def : Pat<(v8i16 (AArch64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3463 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3464 def : Pat<(v2i32 (AArch64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
3465 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
3466 def : Pat<(v2f32 (AArch64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
3467 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
3468 def : Pat<(v4i32 (AArch64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3469 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3470 def : Pat<(v4f32 (AArch64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3471 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3472 def : Pat<(v2i64 (AArch64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3473 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3474 def : Pat<(v2f64 (AArch64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3475 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3476 def : Pat<(v4f16 (AArch64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
3477 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
3478 def : Pat<(v8f16 (AArch64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3479 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3481 // We use EXT to handle extract_subvector to copy the upper 64-bits of a
3483 def : Pat<(v8i8 (extract_subvector V128:$Rn, (i64 8))),
3484 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3485 def : Pat<(v4i16 (extract_subvector V128:$Rn, (i64 4))),
3486 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3487 def : Pat<(v2i32 (extract_subvector V128:$Rn, (i64 2))),
3488 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3489 def : Pat<(v1i64 (extract_subvector V128:$Rn, (i64 1))),
3490 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3491 def : Pat<(v4f16 (extract_subvector V128:$Rn, (i64 4))),
3492 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3493 def : Pat<(v2f32 (extract_subvector V128:$Rn, (i64 2))),
3494 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3495 def : Pat<(v1f64 (extract_subvector V128:$Rn, (i64 1))),
3496 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3499 //----------------------------------------------------------------------------
3500 // AdvSIMD zip vector
3501 //----------------------------------------------------------------------------
3503 defm TRN1 : SIMDZipVector<0b010, "trn1", AArch64trn1>;
3504 defm TRN2 : SIMDZipVector<0b110, "trn2", AArch64trn2>;
3505 defm UZP1 : SIMDZipVector<0b001, "uzp1", AArch64uzp1>;
3506 defm UZP2 : SIMDZipVector<0b101, "uzp2", AArch64uzp2>;
3507 defm ZIP1 : SIMDZipVector<0b011, "zip1", AArch64zip1>;
3508 defm ZIP2 : SIMDZipVector<0b111, "zip2", AArch64zip2>;
3510 //----------------------------------------------------------------------------
3511 // AdvSIMD TBL/TBX instructions
3512 //----------------------------------------------------------------------------
3514 defm TBL : SIMDTableLookup< 0, "tbl">;
3515 defm TBX : SIMDTableLookupTied<1, "tbx">;
3517 def : Pat<(v8i8 (int_aarch64_neon_tbl1 (v16i8 VecListOne128:$Rn), (v8i8 V64:$Ri))),
3518 (TBLv8i8One VecListOne128:$Rn, V64:$Ri)>;
3519 def : Pat<(v16i8 (int_aarch64_neon_tbl1 (v16i8 V128:$Ri), (v16i8 V128:$Rn))),
3520 (TBLv16i8One V128:$Ri, V128:$Rn)>;
3522 def : Pat<(v8i8 (int_aarch64_neon_tbx1 (v8i8 V64:$Rd),
3523 (v16i8 VecListOne128:$Rn), (v8i8 V64:$Ri))),
3524 (TBXv8i8One V64:$Rd, VecListOne128:$Rn, V64:$Ri)>;
3525 def : Pat<(v16i8 (int_aarch64_neon_tbx1 (v16i8 V128:$Rd),
3526 (v16i8 V128:$Ri), (v16i8 V128:$Rn))),
3527 (TBXv16i8One V128:$Rd, V128:$Ri, V128:$Rn)>;
3530 //----------------------------------------------------------------------------
3531 // AdvSIMD scalar CPY instruction
3532 //----------------------------------------------------------------------------
3534 defm CPY : SIMDScalarCPY<"cpy">;
3536 //----------------------------------------------------------------------------
3537 // AdvSIMD scalar pairwise instructions
3538 //----------------------------------------------------------------------------
3540 defm ADDP : SIMDPairwiseScalarD<0, 0b11011, "addp">;
3541 defm FADDP : SIMDPairwiseScalarSD<1, 0, 0b01101, "faddp">;
3542 defm FMAXNMP : SIMDPairwiseScalarSD<1, 0, 0b01100, "fmaxnmp">;
3543 defm FMAXP : SIMDPairwiseScalarSD<1, 0, 0b01111, "fmaxp">;
3544 defm FMINNMP : SIMDPairwiseScalarSD<1, 1, 0b01100, "fminnmp">;
3545 defm FMINP : SIMDPairwiseScalarSD<1, 1, 0b01111, "fminp">;
3546 def : Pat<(v2i64 (AArch64saddv V128:$Rn)),
3547 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), (ADDPv2i64p V128:$Rn), dsub)>;
3548 def : Pat<(v2i64 (AArch64uaddv V128:$Rn)),
3549 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), (ADDPv2i64p V128:$Rn), dsub)>;
3550 def : Pat<(f32 (int_aarch64_neon_faddv (v2f32 V64:$Rn))),
3551 (FADDPv2i32p V64:$Rn)>;
3552 def : Pat<(f32 (int_aarch64_neon_faddv (v4f32 V128:$Rn))),
3553 (FADDPv2i32p (EXTRACT_SUBREG (FADDPv4f32 V128:$Rn, V128:$Rn), dsub))>;
3554 def : Pat<(f64 (int_aarch64_neon_faddv (v2f64 V128:$Rn))),
3555 (FADDPv2i64p V128:$Rn)>;
3556 def : Pat<(f32 (int_aarch64_neon_fmaxnmv (v2f32 V64:$Rn))),
3557 (FMAXNMPv2i32p V64:$Rn)>;
3558 def : Pat<(f64 (int_aarch64_neon_fmaxnmv (v2f64 V128:$Rn))),
3559 (FMAXNMPv2i64p V128:$Rn)>;
3560 def : Pat<(f32 (int_aarch64_neon_fmaxv (v2f32 V64:$Rn))),
3561 (FMAXPv2i32p V64:$Rn)>;
3562 def : Pat<(f64 (int_aarch64_neon_fmaxv (v2f64 V128:$Rn))),
3563 (FMAXPv2i64p V128:$Rn)>;
3564 def : Pat<(f32 (int_aarch64_neon_fminnmv (v2f32 V64:$Rn))),
3565 (FMINNMPv2i32p V64:$Rn)>;
3566 def : Pat<(f64 (int_aarch64_neon_fminnmv (v2f64 V128:$Rn))),
3567 (FMINNMPv2i64p V128:$Rn)>;
3568 def : Pat<(f32 (int_aarch64_neon_fminv (v2f32 V64:$Rn))),
3569 (FMINPv2i32p V64:$Rn)>;
3570 def : Pat<(f64 (int_aarch64_neon_fminv (v2f64 V128:$Rn))),
3571 (FMINPv2i64p V128:$Rn)>;
3573 //----------------------------------------------------------------------------
3574 // AdvSIMD INS/DUP instructions
3575 //----------------------------------------------------------------------------
3577 def DUPv8i8gpr : SIMDDupFromMain<0, {?,?,?,?,1}, ".8b", v8i8, V64, GPR32>;
3578 def DUPv16i8gpr : SIMDDupFromMain<1, {?,?,?,?,1}, ".16b", v16i8, V128, GPR32>;
3579 def DUPv4i16gpr : SIMDDupFromMain<0, {?,?,?,1,0}, ".4h", v4i16, V64, GPR32>;
3580 def DUPv8i16gpr : SIMDDupFromMain<1, {?,?,?,1,0}, ".8h", v8i16, V128, GPR32>;
3581 def DUPv2i32gpr : SIMDDupFromMain<0, {?,?,1,0,0}, ".2s", v2i32, V64, GPR32>;
3582 def DUPv4i32gpr : SIMDDupFromMain<1, {?,?,1,0,0}, ".4s", v4i32, V128, GPR32>;
3583 def DUPv2i64gpr : SIMDDupFromMain<1, {?,1,0,0,0}, ".2d", v2i64, V128, GPR64>;
3585 def DUPv2i64lane : SIMDDup64FromElement;
3586 def DUPv2i32lane : SIMDDup32FromElement<0, ".2s", v2i32, V64>;
3587 def DUPv4i32lane : SIMDDup32FromElement<1, ".4s", v4i32, V128>;
3588 def DUPv4i16lane : SIMDDup16FromElement<0, ".4h", v4i16, V64>;
3589 def DUPv8i16lane : SIMDDup16FromElement<1, ".8h", v8i16, V128>;
3590 def DUPv8i8lane : SIMDDup8FromElement <0, ".8b", v8i8, V64>;
3591 def DUPv16i8lane : SIMDDup8FromElement <1, ".16b", v16i8, V128>;
3593 def : Pat<(v2f32 (AArch64dup (f32 FPR32:$Rn))),
3594 (v2f32 (DUPv2i32lane
3595 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rn, ssub),
3597 def : Pat<(v4f32 (AArch64dup (f32 FPR32:$Rn))),
3598 (v4f32 (DUPv4i32lane
3599 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rn, ssub),
3601 def : Pat<(v2f64 (AArch64dup (f64 FPR64:$Rn))),
3602 (v2f64 (DUPv2i64lane
3603 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$Rn, dsub),
3605 def : Pat<(v4f16 (AArch64dup (f16 FPR16:$Rn))),
3606 (v4f16 (DUPv4i16lane
3607 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), FPR16:$Rn, hsub),
3609 def : Pat<(v8f16 (AArch64dup (f16 FPR16:$Rn))),
3610 (v8f16 (DUPv8i16lane
3611 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), FPR16:$Rn, hsub),
3614 def : Pat<(v4f16 (AArch64duplane16 (v8f16 V128:$Rn), VectorIndexH:$imm)),
3615 (DUPv4i16lane V128:$Rn, VectorIndexH:$imm)>;
3616 def : Pat<(v8f16 (AArch64duplane16 (v8f16 V128:$Rn), VectorIndexH:$imm)),
3617 (DUPv8i16lane V128:$Rn, VectorIndexH:$imm)>;
3619 def : Pat<(v2f32 (AArch64duplane32 (v4f32 V128:$Rn), VectorIndexS:$imm)),
3620 (DUPv2i32lane V128:$Rn, VectorIndexS:$imm)>;
3621 def : Pat<(v4f32 (AArch64duplane32 (v4f32 V128:$Rn), VectorIndexS:$imm)),
3622 (DUPv4i32lane V128:$Rn, VectorIndexS:$imm)>;
3623 def : Pat<(v2f64 (AArch64duplane64 (v2f64 V128:$Rn), VectorIndexD:$imm)),
3624 (DUPv2i64lane V128:$Rn, VectorIndexD:$imm)>;
3626 // If there's an (AArch64dup (vector_extract ...) ...), we can use a duplane
3627 // instruction even if the types don't match: we just have to remap the lane
3628 // carefully. N.b. this trick only applies to truncations.
3629 def VecIndex_x2 : SDNodeXForm<imm, [{
3630 return CurDAG->getTargetConstant(2 * N->getZExtValue(), SDLoc(N), MVT::i64);
3632 def VecIndex_x4 : SDNodeXForm<imm, [{
3633 return CurDAG->getTargetConstant(4 * N->getZExtValue(), SDLoc(N), MVT::i64);
3635 def VecIndex_x8 : SDNodeXForm<imm, [{
3636 return CurDAG->getTargetConstant(8 * N->getZExtValue(), SDLoc(N), MVT::i64);
3639 multiclass DUPWithTruncPats<ValueType ResVT, ValueType Src64VT,
3640 ValueType Src128VT, ValueType ScalVT,
3641 Instruction DUP, SDNodeXForm IdxXFORM> {
3642 def : Pat<(ResVT (AArch64dup (ScalVT (vector_extract (Src128VT V128:$Rn),
3644 (DUP V128:$Rn, (IdxXFORM imm:$idx))>;
3646 def : Pat<(ResVT (AArch64dup (ScalVT (vector_extract (Src64VT V64:$Rn),
3648 (DUP (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), (IdxXFORM imm:$idx))>;
3651 defm : DUPWithTruncPats<v8i8, v4i16, v8i16, i32, DUPv8i8lane, VecIndex_x2>;
3652 defm : DUPWithTruncPats<v8i8, v2i32, v4i32, i32, DUPv8i8lane, VecIndex_x4>;
3653 defm : DUPWithTruncPats<v4i16, v2i32, v4i32, i32, DUPv4i16lane, VecIndex_x2>;
3655 defm : DUPWithTruncPats<v16i8, v4i16, v8i16, i32, DUPv16i8lane, VecIndex_x2>;
3656 defm : DUPWithTruncPats<v16i8, v2i32, v4i32, i32, DUPv16i8lane, VecIndex_x4>;
3657 defm : DUPWithTruncPats<v8i16, v2i32, v4i32, i32, DUPv8i16lane, VecIndex_x2>;
3659 multiclass DUPWithTrunci64Pats<ValueType ResVT, Instruction DUP,
3660 SDNodeXForm IdxXFORM> {
3661 def : Pat<(ResVT (AArch64dup (i32 (trunc (vector_extract (v2i64 V128:$Rn),
3663 (DUP V128:$Rn, (IdxXFORM imm:$idx))>;
3665 def : Pat<(ResVT (AArch64dup (i32 (trunc (vector_extract (v1i64 V64:$Rn),
3667 (DUP (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), (IdxXFORM imm:$idx))>;
3670 defm : DUPWithTrunci64Pats<v8i8, DUPv8i8lane, VecIndex_x8>;
3671 defm : DUPWithTrunci64Pats<v4i16, DUPv4i16lane, VecIndex_x4>;
3672 defm : DUPWithTrunci64Pats<v2i32, DUPv2i32lane, VecIndex_x2>;
3674 defm : DUPWithTrunci64Pats<v16i8, DUPv16i8lane, VecIndex_x8>;
3675 defm : DUPWithTrunci64Pats<v8i16, DUPv8i16lane, VecIndex_x4>;
3676 defm : DUPWithTrunci64Pats<v4i32, DUPv4i32lane, VecIndex_x2>;
3678 // SMOV and UMOV definitions, with some extra patterns for convenience
3682 def : Pat<(sext_inreg (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx), i8),
3683 (i32 (SMOVvi8to32 V128:$Rn, VectorIndexB:$idx))>;
3684 def : Pat<(sext_inreg (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx), i8),
3685 (i64 (SMOVvi8to64 V128:$Rn, VectorIndexB:$idx))>;
3686 def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
3687 (i32 (SMOVvi16to32 V128:$Rn, VectorIndexH:$idx))>;
3688 def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
3689 (i64 (SMOVvi16to64 V128:$Rn, VectorIndexH:$idx))>;
3690 def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
3691 (i32 (SMOVvi16to32 V128:$Rn, VectorIndexH:$idx))>;
3692 def : Pat<(sext (i32 (vector_extract (v4i32 V128:$Rn), VectorIndexS:$idx))),
3693 (i64 (SMOVvi32to64 V128:$Rn, VectorIndexS:$idx))>;
3695 // Extracting i8 or i16 elements will have the zero-extend transformed to
3696 // an 'and' mask by type legalization since neither i8 nor i16 are legal types
3697 // for AArch64. Match these patterns here since UMOV already zeroes out the high
3698 // bits of the destination register.
3699 def : Pat<(and (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx),
3701 (i32 (UMOVvi8 V128:$Rn, VectorIndexB:$idx))>;
3702 def : Pat<(and (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),
3704 (i32 (UMOVvi16 V128:$Rn, VectorIndexH:$idx))>;
3708 def : Pat<(v16i8 (scalar_to_vector GPR32:$Rn)),
3709 (SUBREG_TO_REG (i32 0),
3710 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3711 def : Pat<(v8i8 (scalar_to_vector GPR32:$Rn)),
3712 (SUBREG_TO_REG (i32 0),
3713 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3715 def : Pat<(v8i16 (scalar_to_vector GPR32:$Rn)),
3716 (SUBREG_TO_REG (i32 0),
3717 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3718 def : Pat<(v4i16 (scalar_to_vector GPR32:$Rn)),
3719 (SUBREG_TO_REG (i32 0),
3720 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3722 def : Pat<(v2i32 (scalar_to_vector (i32 FPR32:$Rn))),
3723 (v2i32 (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)),
3724 (i32 FPR32:$Rn), ssub))>;
3725 def : Pat<(v4i32 (scalar_to_vector (i32 FPR32:$Rn))),
3726 (v4i32 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
3727 (i32 FPR32:$Rn), ssub))>;
3728 def : Pat<(v2i64 (scalar_to_vector (i64 FPR64:$Rn))),
3729 (v2i64 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
3730 (i64 FPR64:$Rn), dsub))>;
3732 def : Pat<(v4f32 (scalar_to_vector (f32 FPR32:$Rn))),
3733 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rn, ssub)>;
3734 def : Pat<(v2f32 (scalar_to_vector (f32 FPR32:$Rn))),
3735 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), FPR32:$Rn, ssub)>;
3736 def : Pat<(v2f64 (scalar_to_vector (f64 FPR64:$Rn))),
3737 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$Rn, dsub)>;
3739 def : Pat<(v4f16 (vector_insert (v4f16 V64:$Rn),
3740 (f16 FPR16:$Rm), (i64 VectorIndexS:$imm))),
3743 (v8f16 (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), V64:$Rn, dsub)),
3745 (v8f16 (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), FPR16:$Rm, hsub)),
3749 def : Pat<(v8f16 (vector_insert (v8f16 V128:$Rn),
3750 (f16 FPR16:$Rm), (i64 VectorIndexH:$imm))),
3752 V128:$Rn, VectorIndexH:$imm,
3753 (v8f16 (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), FPR16:$Rm, hsub)),
3756 def : Pat<(v2f32 (vector_insert (v2f32 V64:$Rn),
3757 (f32 FPR32:$Rm), (i64 VectorIndexS:$imm))),
3760 (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), V64:$Rn, dsub)),
3762 (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rm, ssub)),
3765 def : Pat<(v4f32 (vector_insert (v4f32 V128:$Rn),
3766 (f32 FPR32:$Rm), (i64 VectorIndexS:$imm))),
3768 V128:$Rn, VectorIndexS:$imm,
3769 (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rm, ssub)),
3771 def : Pat<(v2f64 (vector_insert (v2f64 V128:$Rn),
3772 (f64 FPR64:$Rm), (i64 VectorIndexD:$imm))),
3774 V128:$Rn, VectorIndexD:$imm,
3775 (v2f64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$Rm, dsub)),
3778 // Copy an element at a constant index in one vector into a constant indexed
3779 // element of another.
3780 // FIXME refactor to a shared class/dev parameterized on vector type, vector
3781 // index type and INS extension
3782 def : Pat<(v16i8 (int_aarch64_neon_vcopy_lane
3783 (v16i8 V128:$Vd), VectorIndexB:$idx, (v16i8 V128:$Vs),
3784 VectorIndexB:$idx2)),
3786 V128:$Vd, VectorIndexB:$idx, V128:$Vs, VectorIndexB:$idx2)
3788 def : Pat<(v8i16 (int_aarch64_neon_vcopy_lane
3789 (v8i16 V128:$Vd), VectorIndexH:$idx, (v8i16 V128:$Vs),
3790 VectorIndexH:$idx2)),
3792 V128:$Vd, VectorIndexH:$idx, V128:$Vs, VectorIndexH:$idx2)
3794 def : Pat<(v4i32 (int_aarch64_neon_vcopy_lane
3795 (v4i32 V128:$Vd), VectorIndexS:$idx, (v4i32 V128:$Vs),
3796 VectorIndexS:$idx2)),
3798 V128:$Vd, VectorIndexS:$idx, V128:$Vs, VectorIndexS:$idx2)
3800 def : Pat<(v2i64 (int_aarch64_neon_vcopy_lane
3801 (v2i64 V128:$Vd), VectorIndexD:$idx, (v2i64 V128:$Vs),
3802 VectorIndexD:$idx2)),
3804 V128:$Vd, VectorIndexD:$idx, V128:$Vs, VectorIndexD:$idx2)
3807 multiclass Neon_INS_elt_pattern<ValueType VT128, ValueType VT64,
3808 ValueType VTScal, Instruction INS> {
3809 def : Pat<(VT128 (vector_insert V128:$src,
3810 (VTScal (vector_extract (VT128 V128:$Rn), imm:$Immn)),
3812 (INS V128:$src, imm:$Immd, V128:$Rn, imm:$Immn)>;
3814 def : Pat<(VT128 (vector_insert V128:$src,
3815 (VTScal (vector_extract (VT64 V64:$Rn), imm:$Immn)),
3817 (INS V128:$src, imm:$Immd,
3818 (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), imm:$Immn)>;
3820 def : Pat<(VT64 (vector_insert V64:$src,
3821 (VTScal (vector_extract (VT128 V128:$Rn), imm:$Immn)),
3823 (EXTRACT_SUBREG (INS (SUBREG_TO_REG (i64 0), V64:$src, dsub),
3824 imm:$Immd, V128:$Rn, imm:$Immn),
3827 def : Pat<(VT64 (vector_insert V64:$src,
3828 (VTScal (vector_extract (VT64 V64:$Rn), imm:$Immn)),
3831 (INS (SUBREG_TO_REG (i64 0), V64:$src, dsub), imm:$Immd,
3832 (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), imm:$Immn),
3836 defm : Neon_INS_elt_pattern<v8f16, v4f16, f16, INSvi16lane>;
3837 defm : Neon_INS_elt_pattern<v4f32, v2f32, f32, INSvi32lane>;
3838 defm : Neon_INS_elt_pattern<v2f64, v1f64, f64, INSvi64lane>;
3841 // Floating point vector extractions are codegen'd as either a sequence of
3842 // subregister extractions, or a MOV (aka CPY here, alias for DUP) if
3843 // the lane number is anything other than zero.
3844 def : Pat<(vector_extract (v2f64 V128:$Rn), 0),
3845 (f64 (EXTRACT_SUBREG V128:$Rn, dsub))>;
3846 def : Pat<(vector_extract (v4f32 V128:$Rn), 0),
3847 (f32 (EXTRACT_SUBREG V128:$Rn, ssub))>;
3848 def : Pat<(vector_extract (v8f16 V128:$Rn), 0),
3849 (f16 (EXTRACT_SUBREG V128:$Rn, hsub))>;
3851 def : Pat<(vector_extract (v2f64 V128:$Rn), VectorIndexD:$idx),
3852 (f64 (CPYi64 V128:$Rn, VectorIndexD:$idx))>;
3853 def : Pat<(vector_extract (v4f32 V128:$Rn), VectorIndexS:$idx),
3854 (f32 (CPYi32 V128:$Rn, VectorIndexS:$idx))>;
3855 def : Pat<(vector_extract (v8f16 V128:$Rn), VectorIndexH:$idx),
3856 (f16 (CPYi16 V128:$Rn, VectorIndexH:$idx))>;
3858 // All concat_vectors operations are canonicalised to act on i64 vectors for
3859 // AArch64. In the general case we need an instruction, which had just as well be
3861 class ConcatPat<ValueType DstTy, ValueType SrcTy>
3862 : Pat<(DstTy (concat_vectors (SrcTy V64:$Rd), V64:$Rn)),
3863 (INSvi64lane (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), 1,
3864 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rn, dsub), 0)>;
3866 def : ConcatPat<v2i64, v1i64>;
3867 def : ConcatPat<v2f64, v1f64>;
3868 def : ConcatPat<v4i32, v2i32>;
3869 def : ConcatPat<v4f32, v2f32>;
3870 def : ConcatPat<v8i16, v4i16>;
3871 def : ConcatPat<v8f16, v4f16>;
3872 def : ConcatPat<v16i8, v8i8>;
3874 // If the high lanes are undef, though, we can just ignore them:
3875 class ConcatUndefPat<ValueType DstTy, ValueType SrcTy>
3876 : Pat<(DstTy (concat_vectors (SrcTy V64:$Rn), undef)),
3877 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rn, dsub)>;
3879 def : ConcatUndefPat<v2i64, v1i64>;
3880 def : ConcatUndefPat<v2f64, v1f64>;
3881 def : ConcatUndefPat<v4i32, v2i32>;
3882 def : ConcatUndefPat<v4f32, v2f32>;
3883 def : ConcatUndefPat<v8i16, v4i16>;
3884 def : ConcatUndefPat<v16i8, v8i8>;
3886 //----------------------------------------------------------------------------
3887 // AdvSIMD across lanes instructions
3888 //----------------------------------------------------------------------------
3890 defm ADDV : SIMDAcrossLanesBHS<0, 0b11011, "addv">;
3891 defm SMAXV : SIMDAcrossLanesBHS<0, 0b01010, "smaxv">;
3892 defm SMINV : SIMDAcrossLanesBHS<0, 0b11010, "sminv">;
3893 defm UMAXV : SIMDAcrossLanesBHS<1, 0b01010, "umaxv">;
3894 defm UMINV : SIMDAcrossLanesBHS<1, 0b11010, "uminv">;
3895 defm SADDLV : SIMDAcrossLanesHSD<0, 0b00011, "saddlv">;
3896 defm UADDLV : SIMDAcrossLanesHSD<1, 0b00011, "uaddlv">;
3897 defm FMAXNMV : SIMDAcrossLanesS<0b01100, 0, "fmaxnmv", int_aarch64_neon_fmaxnmv>;
3898 defm FMAXV : SIMDAcrossLanesS<0b01111, 0, "fmaxv", int_aarch64_neon_fmaxv>;
3899 defm FMINNMV : SIMDAcrossLanesS<0b01100, 1, "fminnmv", int_aarch64_neon_fminnmv>;
3900 defm FMINV : SIMDAcrossLanesS<0b01111, 1, "fminv", int_aarch64_neon_fminv>;
3902 // Patterns for across-vector intrinsics, that have a node equivalent, that
3903 // returns a vector (with only the low lane defined) instead of a scalar.
3904 // In effect, opNode is the same as (scalar_to_vector (IntNode)).
3905 multiclass SIMDAcrossLanesIntrinsic<string baseOpc,
3906 SDPatternOperator opNode> {
3907 // If a lane instruction caught the vector_extract around opNode, we can
3908 // directly match the latter to the instruction.
3909 def : Pat<(v8i8 (opNode V64:$Rn)),
3910 (INSERT_SUBREG (v8i8 (IMPLICIT_DEF)),
3911 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub)>;
3912 def : Pat<(v16i8 (opNode V128:$Rn)),
3913 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3914 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub)>;
3915 def : Pat<(v4i16 (opNode V64:$Rn)),
3916 (INSERT_SUBREG (v4i16 (IMPLICIT_DEF)),
3917 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub)>;
3918 def : Pat<(v8i16 (opNode V128:$Rn)),
3919 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
3920 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub)>;
3921 def : Pat<(v4i32 (opNode V128:$Rn)),
3922 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
3923 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), ssub)>;
3926 // If none did, fallback to the explicit patterns, consuming the vector_extract.
3927 def : Pat<(i32 (vector_extract (insert_subvector undef, (v8i8 (opNode V64:$Rn)),
3928 (i32 0)), (i64 0))),
3929 (EXTRACT_SUBREG (INSERT_SUBREG (v8i8 (IMPLICIT_DEF)),
3930 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn),
3932 def : Pat<(i32 (vector_extract (v16i8 (opNode V128:$Rn)), (i64 0))),
3933 (EXTRACT_SUBREG (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3934 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn),
3936 def : Pat<(i32 (vector_extract (insert_subvector undef,
3937 (v4i16 (opNode V64:$Rn)), (i32 0)), (i64 0))),
3938 (EXTRACT_SUBREG (INSERT_SUBREG (v4i16 (IMPLICIT_DEF)),
3939 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn),
3941 def : Pat<(i32 (vector_extract (v8i16 (opNode V128:$Rn)), (i64 0))),
3942 (EXTRACT_SUBREG (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
3943 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn),
3945 def : Pat<(i32 (vector_extract (v4i32 (opNode V128:$Rn)), (i64 0))),
3946 (EXTRACT_SUBREG (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
3947 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn),
3952 multiclass SIMDAcrossLanesSignedIntrinsic<string baseOpc,
3953 SDPatternOperator opNode>
3954 : SIMDAcrossLanesIntrinsic<baseOpc, opNode> {
3955 // If there is a sign extension after this intrinsic, consume it as smov already
3957 def : Pat<(i32 (sext_inreg (i32 (vector_extract (insert_subvector undef,
3958 (opNode (v8i8 V64:$Rn)), (i32 0)), (i64 0))), i8)),
3960 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3961 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3963 def : Pat<(i32 (sext_inreg (i32 (vector_extract
3964 (opNode (v16i8 V128:$Rn)), (i64 0))), i8)),
3966 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3967 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3969 def : Pat<(i32 (sext_inreg (i32 (vector_extract (insert_subvector undef,
3970 (opNode (v4i16 V64:$Rn)), (i32 0)), (i64 0))), i16)),
3972 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3973 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3975 def : Pat<(i32 (sext_inreg (i32 (vector_extract
3976 (opNode (v8i16 V128:$Rn)), (i64 0))), i16)),
3978 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3979 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3983 multiclass SIMDAcrossLanesUnsignedIntrinsic<string baseOpc,
3984 SDPatternOperator opNode>
3985 : SIMDAcrossLanesIntrinsic<baseOpc, opNode> {
3986 // If there is a masking operation keeping only what has been actually
3987 // generated, consume it.
3988 def : Pat<(i32 (and (i32 (vector_extract (insert_subvector undef,
3989 (opNode (v8i8 V64:$Rn)), (i32 0)), (i64 0))), maski8_or_more)),
3990 (i32 (EXTRACT_SUBREG
3991 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3992 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3994 def : Pat<(i32 (and (i32 (vector_extract (opNode (v16i8 V128:$Rn)), (i64 0))),
3996 (i32 (EXTRACT_SUBREG
3997 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3998 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
4000 def : Pat<(i32 (and (i32 (vector_extract (insert_subvector undef,
4001 (opNode (v4i16 V64:$Rn)), (i32 0)), (i64 0))), maski16_or_more)),
4002 (i32 (EXTRACT_SUBREG
4003 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4004 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
4006 def : Pat<(i32 (and (i32 (vector_extract (opNode (v8i16 V128:$Rn)), (i64 0))),
4008 (i32 (EXTRACT_SUBREG
4009 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4010 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
4014 defm : SIMDAcrossLanesSignedIntrinsic<"ADDV", AArch64saddv>;
4015 // vaddv_[su]32 is special; -> ADDP Vd.2S,Vn.2S,Vm.2S; return Vd.s[0];Vn==Vm
4016 def : Pat<(v2i32 (AArch64saddv (v2i32 V64:$Rn))),
4017 (ADDPv2i32 V64:$Rn, V64:$Rn)>;
4019 defm : SIMDAcrossLanesUnsignedIntrinsic<"ADDV", AArch64uaddv>;
4020 // vaddv_[su]32 is special; -> ADDP Vd.2S,Vn.2S,Vm.2S; return Vd.s[0];Vn==Vm
4021 def : Pat<(v2i32 (AArch64uaddv (v2i32 V64:$Rn))),
4022 (ADDPv2i32 V64:$Rn, V64:$Rn)>;
4024 defm : SIMDAcrossLanesSignedIntrinsic<"SMAXV", AArch64smaxv>;
4025 def : Pat<(v2i32 (AArch64smaxv (v2i32 V64:$Rn))),
4026 (SMAXPv2i32 V64:$Rn, V64:$Rn)>;
4028 defm : SIMDAcrossLanesSignedIntrinsic<"SMINV", AArch64sminv>;
4029 def : Pat<(v2i32 (AArch64sminv (v2i32 V64:$Rn))),
4030 (SMINPv2i32 V64:$Rn, V64:$Rn)>;
4032 defm : SIMDAcrossLanesUnsignedIntrinsic<"UMAXV", AArch64umaxv>;
4033 def : Pat<(v2i32 (AArch64umaxv (v2i32 V64:$Rn))),
4034 (UMAXPv2i32 V64:$Rn, V64:$Rn)>;
4036 defm : SIMDAcrossLanesUnsignedIntrinsic<"UMINV", AArch64uminv>;
4037 def : Pat<(v2i32 (AArch64uminv (v2i32 V64:$Rn))),
4038 (UMINPv2i32 V64:$Rn, V64:$Rn)>;
4040 multiclass SIMDAcrossLanesSignedLongIntrinsic<string baseOpc, Intrinsic intOp> {
4041 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
4043 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4044 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), hsub),
4046 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
4048 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4049 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), hsub),
4052 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
4053 (i32 (EXTRACT_SUBREG
4054 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4055 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), ssub),
4057 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
4058 (i32 (EXTRACT_SUBREG
4059 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4060 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), ssub),
4063 def : Pat<(i64 (intOp (v4i32 V128:$Rn))),
4064 (i64 (EXTRACT_SUBREG
4065 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4066 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), dsub),
4070 multiclass SIMDAcrossLanesUnsignedLongIntrinsic<string baseOpc,
4072 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
4073 (i32 (EXTRACT_SUBREG
4074 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4075 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), hsub),
4077 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
4078 (i32 (EXTRACT_SUBREG
4079 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4080 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), hsub),
4083 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
4084 (i32 (EXTRACT_SUBREG
4085 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4086 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), ssub),
4088 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
4089 (i32 (EXTRACT_SUBREG
4090 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4091 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), ssub),
4094 def : Pat<(i64 (intOp (v4i32 V128:$Rn))),
4095 (i64 (EXTRACT_SUBREG
4096 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4097 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), dsub),
4101 defm : SIMDAcrossLanesSignedLongIntrinsic<"SADDLV", int_aarch64_neon_saddlv>;
4102 defm : SIMDAcrossLanesUnsignedLongIntrinsic<"UADDLV", int_aarch64_neon_uaddlv>;
4104 // The vaddlv_s32 intrinsic gets mapped to SADDLP.
4105 def : Pat<(i64 (int_aarch64_neon_saddlv (v2i32 V64:$Rn))),
4106 (i64 (EXTRACT_SUBREG
4107 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4108 (SADDLPv2i32_v1i64 V64:$Rn), dsub),
4110 // The vaddlv_u32 intrinsic gets mapped to UADDLP.
4111 def : Pat<(i64 (int_aarch64_neon_uaddlv (v2i32 V64:$Rn))),
4112 (i64 (EXTRACT_SUBREG
4113 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4114 (UADDLPv2i32_v1i64 V64:$Rn), dsub),
4117 //------------------------------------------------------------------------------
4118 // AdvSIMD modified immediate instructions
4119 //------------------------------------------------------------------------------
4122 defm BIC : SIMDModifiedImmVectorShiftTied<1, 0b11, 0b01, "bic", AArch64bici>;
4124 defm ORR : SIMDModifiedImmVectorShiftTied<0, 0b11, 0b01, "orr", AArch64orri>;
4126 def : InstAlias<"bic $Vd.4h, $imm", (BICv4i16 V64:$Vd, imm0_255:$imm, 0)>;
4127 def : InstAlias<"bic $Vd.8h, $imm", (BICv8i16 V128:$Vd, imm0_255:$imm, 0)>;
4128 def : InstAlias<"bic $Vd.2s, $imm", (BICv2i32 V64:$Vd, imm0_255:$imm, 0)>;
4129 def : InstAlias<"bic $Vd.4s, $imm", (BICv4i32 V128:$Vd, imm0_255:$imm, 0)>;
4131 def : InstAlias<"bic.4h $Vd, $imm", (BICv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
4132 def : InstAlias<"bic.8h $Vd, $imm", (BICv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
4133 def : InstAlias<"bic.2s $Vd, $imm", (BICv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
4134 def : InstAlias<"bic.4s $Vd, $imm", (BICv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
4136 def : InstAlias<"orr $Vd.4h, $imm", (ORRv4i16 V64:$Vd, imm0_255:$imm, 0)>;
4137 def : InstAlias<"orr $Vd.8h, $imm", (ORRv8i16 V128:$Vd, imm0_255:$imm, 0)>;
4138 def : InstAlias<"orr $Vd.2s, $imm", (ORRv2i32 V64:$Vd, imm0_255:$imm, 0)>;
4139 def : InstAlias<"orr $Vd.4s, $imm", (ORRv4i32 V128:$Vd, imm0_255:$imm, 0)>;
4141 def : InstAlias<"orr.4h $Vd, $imm", (ORRv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
4142 def : InstAlias<"orr.8h $Vd, $imm", (ORRv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
4143 def : InstAlias<"orr.2s $Vd, $imm", (ORRv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
4144 def : InstAlias<"orr.4s $Vd, $imm", (ORRv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
4147 def FMOVv2f64_ns : SIMDModifiedImmVectorNoShift<1, 1, 0b1111, V128, fpimm8,
4149 [(set (v2f64 V128:$Rd), (AArch64fmov imm0_255:$imm8))]>;
4150 def FMOVv2f32_ns : SIMDModifiedImmVectorNoShift<0, 0, 0b1111, V64, fpimm8,
4152 [(set (v2f32 V64:$Rd), (AArch64fmov imm0_255:$imm8))]>;
4153 def FMOVv4f32_ns : SIMDModifiedImmVectorNoShift<1, 0, 0b1111, V128, fpimm8,
4155 [(set (v4f32 V128:$Rd), (AArch64fmov imm0_255:$imm8))]>;
4159 // EDIT byte mask: scalar
4160 let isReMaterializable = 1, isAsCheapAsAMove = 1 in
4161 def MOVID : SIMDModifiedImmScalarNoShift<0, 1, 0b1110, "movi",
4162 [(set FPR64:$Rd, simdimmtype10:$imm8)]>;
4163 // The movi_edit node has the immediate value already encoded, so we use
4164 // a plain imm0_255 here.
4165 def : Pat<(f64 (AArch64movi_edit imm0_255:$shift)),
4166 (MOVID imm0_255:$shift)>;
4168 def : Pat<(v1i64 immAllZerosV), (MOVID (i32 0))>;
4169 def : Pat<(v2i32 immAllZerosV), (MOVID (i32 0))>;
4170 def : Pat<(v4i16 immAllZerosV), (MOVID (i32 0))>;
4171 def : Pat<(v8i8 immAllZerosV), (MOVID (i32 0))>;
4173 def : Pat<(v1i64 immAllOnesV), (MOVID (i32 255))>;
4174 def : Pat<(v2i32 immAllOnesV), (MOVID (i32 255))>;
4175 def : Pat<(v4i16 immAllOnesV), (MOVID (i32 255))>;
4176 def : Pat<(v8i8 immAllOnesV), (MOVID (i32 255))>;
4178 // EDIT byte mask: 2d
4180 // The movi_edit node has the immediate value already encoded, so we use
4181 // a plain imm0_255 in the pattern
4182 let isReMaterializable = 1, isAsCheapAsAMove = 1 in
4183 def MOVIv2d_ns : SIMDModifiedImmVectorNoShift<1, 1, 0b1110, V128,
4186 [(set (v2i64 V128:$Rd), (AArch64movi_edit imm0_255:$imm8))]>;
4189 // Use movi.2d to materialize 0.0 if the HW does zero-cycle zeroing.
4190 // Complexity is added to break a tie with a plain MOVI.
4191 let AddedComplexity = 1 in {
4192 def : Pat<(f32 fpimm0),
4193 (f32 (EXTRACT_SUBREG (v2i64 (MOVIv2d_ns (i32 0))), ssub))>,
4195 def : Pat<(f64 fpimm0),
4196 (f64 (EXTRACT_SUBREG (v2i64 (MOVIv2d_ns (i32 0))), dsub))>,
4200 def : Pat<(v2i64 immAllZerosV), (MOVIv2d_ns (i32 0))>;
4201 def : Pat<(v4i32 immAllZerosV), (MOVIv2d_ns (i32 0))>;
4202 def : Pat<(v8i16 immAllZerosV), (MOVIv2d_ns (i32 0))>;
4203 def : Pat<(v16i8 immAllZerosV), (MOVIv2d_ns (i32 0))>;
4205 def : Pat<(v2i64 immAllOnesV), (MOVIv2d_ns (i32 255))>;
4206 def : Pat<(v4i32 immAllOnesV), (MOVIv2d_ns (i32 255))>;
4207 def : Pat<(v8i16 immAllOnesV), (MOVIv2d_ns (i32 255))>;
4208 def : Pat<(v16i8 immAllOnesV), (MOVIv2d_ns (i32 255))>;
4210 def : Pat<(v2f64 (AArch64dup (f64 fpimm0))), (MOVIv2d_ns (i32 0))>;
4211 def : Pat<(v4f32 (AArch64dup (f32 fpimm0))), (MOVIv2d_ns (i32 0))>;
4213 // EDIT per word & halfword: 2s, 4h, 4s, & 8h
4214 defm MOVI : SIMDModifiedImmVectorShift<0, 0b10, 0b00, "movi">;
4216 def : InstAlias<"movi $Vd.4h, $imm", (MOVIv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
4217 def : InstAlias<"movi $Vd.8h, $imm", (MOVIv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
4218 def : InstAlias<"movi $Vd.2s, $imm", (MOVIv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
4219 def : InstAlias<"movi $Vd.4s, $imm", (MOVIv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
4221 def : InstAlias<"movi.4h $Vd, $imm", (MOVIv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
4222 def : InstAlias<"movi.8h $Vd, $imm", (MOVIv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
4223 def : InstAlias<"movi.2s $Vd, $imm", (MOVIv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
4224 def : InstAlias<"movi.4s $Vd, $imm", (MOVIv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
4226 def : Pat<(v2i32 (AArch64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
4227 (MOVIv2i32 imm0_255:$imm8, imm:$shift)>;
4228 def : Pat<(v4i32 (AArch64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
4229 (MOVIv4i32 imm0_255:$imm8, imm:$shift)>;
4230 def : Pat<(v4i16 (AArch64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
4231 (MOVIv4i16 imm0_255:$imm8, imm:$shift)>;
4232 def : Pat<(v8i16 (AArch64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
4233 (MOVIv8i16 imm0_255:$imm8, imm:$shift)>;
4235 // EDIT per word: 2s & 4s with MSL shifter
4236 def MOVIv2s_msl : SIMDModifiedImmMoveMSL<0, 0, {1,1,0,?}, V64, "movi", ".2s",
4237 [(set (v2i32 V64:$Rd),
4238 (AArch64movi_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
4239 def MOVIv4s_msl : SIMDModifiedImmMoveMSL<1, 0, {1,1,0,?}, V128, "movi", ".4s",
4240 [(set (v4i32 V128:$Rd),
4241 (AArch64movi_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
4243 // Per byte: 8b & 16b
4244 def MOVIv8b_ns : SIMDModifiedImmVectorNoShift<0, 0, 0b1110, V64, imm0_255,
4246 [(set (v8i8 V64:$Rd), (AArch64movi imm0_255:$imm8))]>;
4247 def MOVIv16b_ns : SIMDModifiedImmVectorNoShift<1, 0, 0b1110, V128, imm0_255,
4249 [(set (v16i8 V128:$Rd), (AArch64movi imm0_255:$imm8))]>;
4253 // EDIT per word & halfword: 2s, 4h, 4s, & 8h
4254 defm MVNI : SIMDModifiedImmVectorShift<1, 0b10, 0b00, "mvni">;
4256 def : InstAlias<"mvni $Vd.4h, $imm", (MVNIv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
4257 def : InstAlias<"mvni $Vd.8h, $imm", (MVNIv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
4258 def : InstAlias<"mvni $Vd.2s, $imm", (MVNIv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
4259 def : InstAlias<"mvni $Vd.4s, $imm", (MVNIv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
4261 def : InstAlias<"mvni.4h $Vd, $imm", (MVNIv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
4262 def : InstAlias<"mvni.8h $Vd, $imm", (MVNIv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
4263 def : InstAlias<"mvni.2s $Vd, $imm", (MVNIv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
4264 def : InstAlias<"mvni.4s $Vd, $imm", (MVNIv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
4266 def : Pat<(v2i32 (AArch64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
4267 (MVNIv2i32 imm0_255:$imm8, imm:$shift)>;
4268 def : Pat<(v4i32 (AArch64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
4269 (MVNIv4i32 imm0_255:$imm8, imm:$shift)>;
4270 def : Pat<(v4i16 (AArch64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
4271 (MVNIv4i16 imm0_255:$imm8, imm:$shift)>;
4272 def : Pat<(v8i16 (AArch64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
4273 (MVNIv8i16 imm0_255:$imm8, imm:$shift)>;
4275 // EDIT per word: 2s & 4s with MSL shifter
4276 def MVNIv2s_msl : SIMDModifiedImmMoveMSL<0, 1, {1,1,0,?}, V64, "mvni", ".2s",
4277 [(set (v2i32 V64:$Rd),
4278 (AArch64mvni_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
4279 def MVNIv4s_msl : SIMDModifiedImmMoveMSL<1, 1, {1,1,0,?}, V128, "mvni", ".4s",
4280 [(set (v4i32 V128:$Rd),
4281 (AArch64mvni_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
4283 //----------------------------------------------------------------------------
4284 // AdvSIMD indexed element
4285 //----------------------------------------------------------------------------
4287 let hasSideEffects = 0 in {
4288 defm FMLA : SIMDFPIndexedSDTied<0, 0b0001, "fmla">;
4289 defm FMLS : SIMDFPIndexedSDTied<0, 0b0101, "fmls">;
4292 // NOTE: Operands are reordered in the FMLA/FMLS PatFrags because the
4293 // instruction expects the addend first, while the intrinsic expects it last.
4295 // On the other hand, there are quite a few valid combinatorial options due to
4296 // the commutativity of multiplication and the fact that (-x) * y = x * (-y).
4297 defm : SIMDFPIndexedSDTiedPatterns<"FMLA",
4298 TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)>>;
4299 defm : SIMDFPIndexedSDTiedPatterns<"FMLA",
4300 TriOpFrag<(fma node:$MHS, node:$RHS, node:$LHS)>>;
4302 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
4303 TriOpFrag<(fma node:$MHS, (fneg node:$RHS), node:$LHS)> >;
4304 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
4305 TriOpFrag<(fma node:$RHS, (fneg node:$MHS), node:$LHS)> >;
4306 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
4307 TriOpFrag<(fma (fneg node:$RHS), node:$MHS, node:$LHS)> >;
4308 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
4309 TriOpFrag<(fma (fneg node:$MHS), node:$RHS, node:$LHS)> >;
4311 multiclass FMLSIndexedAfterNegPatterns<SDPatternOperator OpNode> {
4312 // 3 variants for the .2s version: DUPLANE from 128-bit, DUPLANE from 64-bit
4314 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
4315 (AArch64duplane32 (v4f32 (fneg V128:$Rm)),
4316 VectorIndexS:$idx))),
4317 (FMLSv2i32_indexed V64:$Rd, V64:$Rn, V128:$Rm, VectorIndexS:$idx)>;
4318 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
4319 (v2f32 (AArch64duplane32
4320 (v4f32 (insert_subvector undef,
4321 (v2f32 (fneg V64:$Rm)),
4323 VectorIndexS:$idx)))),
4324 (FMLSv2i32_indexed V64:$Rd, V64:$Rn,
4325 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub),
4326 VectorIndexS:$idx)>;
4327 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
4328 (AArch64dup (f32 (fneg FPR32Op:$Rm))))),
4329 (FMLSv2i32_indexed V64:$Rd, V64:$Rn,
4330 (SUBREG_TO_REG (i32 0), FPR32Op:$Rm, ssub), (i64 0))>;
4332 // 3 variants for the .4s version: DUPLANE from 128-bit, DUPLANE from 64-bit
4334 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
4335 (AArch64duplane32 (v4f32 (fneg V128:$Rm)),
4336 VectorIndexS:$idx))),
4337 (FMLSv4i32_indexed V128:$Rd, V128:$Rn, V128:$Rm,
4338 VectorIndexS:$idx)>;
4339 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
4340 (v4f32 (AArch64duplane32
4341 (v4f32 (insert_subvector undef,
4342 (v2f32 (fneg V64:$Rm)),
4344 VectorIndexS:$idx)))),
4345 (FMLSv4i32_indexed V128:$Rd, V128:$Rn,
4346 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub),
4347 VectorIndexS:$idx)>;
4348 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
4349 (AArch64dup (f32 (fneg FPR32Op:$Rm))))),
4350 (FMLSv4i32_indexed V128:$Rd, V128:$Rn,
4351 (SUBREG_TO_REG (i32 0), FPR32Op:$Rm, ssub), (i64 0))>;
4353 // 2 variants for the .2d version: DUPLANE from 128-bit, and DUP scalar
4354 // (DUPLANE from 64-bit would be trivial).
4355 def : Pat<(v2f64 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn),
4356 (AArch64duplane64 (v2f64 (fneg V128:$Rm)),
4357 VectorIndexD:$idx))),
4359 V128:$Rd, V128:$Rn, V128:$Rm, VectorIndexS:$idx)>;
4360 def : Pat<(v2f64 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn),
4361 (AArch64dup (f64 (fneg FPR64Op:$Rm))))),
4362 (FMLSv2i64_indexed V128:$Rd, V128:$Rn,
4363 (SUBREG_TO_REG (i32 0), FPR64Op:$Rm, dsub), (i64 0))>;
4365 // 2 variants for 32-bit scalar version: extract from .2s or from .4s
4366 def : Pat<(f32 (OpNode (f32 FPR32:$Rd), (f32 FPR32:$Rn),
4367 (vector_extract (v4f32 (fneg V128:$Rm)),
4368 VectorIndexS:$idx))),
4369 (FMLSv1i32_indexed FPR32:$Rd, FPR32:$Rn,
4370 V128:$Rm, VectorIndexS:$idx)>;
4371 def : Pat<(f32 (OpNode (f32 FPR32:$Rd), (f32 FPR32:$Rn),
4372 (vector_extract (v2f32 (fneg V64:$Rm)),
4373 VectorIndexS:$idx))),
4374 (FMLSv1i32_indexed FPR32:$Rd, FPR32:$Rn,
4375 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub), VectorIndexS:$idx)>;
4377 // 1 variant for 64-bit scalar version: extract from .1d or from .2d
4378 def : Pat<(f64 (OpNode (f64 FPR64:$Rd), (f64 FPR64:$Rn),
4379 (vector_extract (v2f64 (fneg V128:$Rm)),
4380 VectorIndexS:$idx))),
4381 (FMLSv1i64_indexed FPR64:$Rd, FPR64:$Rn,
4382 V128:$Rm, VectorIndexS:$idx)>;
4385 defm : FMLSIndexedAfterNegPatterns<
4386 TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)> >;
4387 defm : FMLSIndexedAfterNegPatterns<
4388 TriOpFrag<(fma node:$MHS, node:$RHS, node:$LHS)> >;
4390 defm FMULX : SIMDFPIndexedSD<1, 0b1001, "fmulx", int_aarch64_neon_fmulx>;
4391 defm FMUL : SIMDFPIndexedSD<0, 0b1001, "fmul", fmul>;
4393 def : Pat<(v2f32 (fmul V64:$Rn, (AArch64dup (f32 FPR32:$Rm)))),
4394 (FMULv2i32_indexed V64:$Rn,
4395 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rm, ssub),
4397 def : Pat<(v4f32 (fmul V128:$Rn, (AArch64dup (f32 FPR32:$Rm)))),
4398 (FMULv4i32_indexed V128:$Rn,
4399 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rm, ssub),
4401 def : Pat<(v2f64 (fmul V128:$Rn, (AArch64dup (f64 FPR64:$Rm)))),
4402 (FMULv2i64_indexed V128:$Rn,
4403 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$Rm, dsub),
4406 defm SQDMULH : SIMDIndexedHS<0, 0b1100, "sqdmulh", int_aarch64_neon_sqdmulh>;
4407 defm SQRDMULH : SIMDIndexedHS<0, 0b1101, "sqrdmulh", int_aarch64_neon_sqrdmulh>;
4408 defm MLA : SIMDVectorIndexedHSTied<1, 0b0000, "mla",
4409 TriOpFrag<(add node:$LHS, (mul node:$MHS, node:$RHS))>>;
4410 defm MLS : SIMDVectorIndexedHSTied<1, 0b0100, "mls",
4411 TriOpFrag<(sub node:$LHS, (mul node:$MHS, node:$RHS))>>;
4412 defm MUL : SIMDVectorIndexedHS<0, 0b1000, "mul", mul>;
4413 defm SMLAL : SIMDVectorIndexedLongSDTied<0, 0b0010, "smlal",
4414 TriOpFrag<(add node:$LHS, (int_aarch64_neon_smull node:$MHS, node:$RHS))>>;
4415 defm SMLSL : SIMDVectorIndexedLongSDTied<0, 0b0110, "smlsl",
4416 TriOpFrag<(sub node:$LHS, (int_aarch64_neon_smull node:$MHS, node:$RHS))>>;
4417 defm SMULL : SIMDVectorIndexedLongSD<0, 0b1010, "smull",
4418 int_aarch64_neon_smull>;
4419 defm SQDMLAL : SIMDIndexedLongSQDMLXSDTied<0, 0b0011, "sqdmlal",
4420 int_aarch64_neon_sqadd>;
4421 defm SQDMLSL : SIMDIndexedLongSQDMLXSDTied<0, 0b0111, "sqdmlsl",
4422 int_aarch64_neon_sqsub>;
4423 defm SQRDMLAH : SIMDIndexedSQRDMLxHSDTied<1, 0b1101, "sqrdmlah",
4424 int_aarch64_neon_sqadd>;
4425 defm SQRDMLSH : SIMDIndexedSQRDMLxHSDTied<1, 0b1111, "sqrdmlsh",
4426 int_aarch64_neon_sqsub>;
4427 defm SQDMULL : SIMDIndexedLongSD<0, 0b1011, "sqdmull", int_aarch64_neon_sqdmull>;
4428 defm UMLAL : SIMDVectorIndexedLongSDTied<1, 0b0010, "umlal",
4429 TriOpFrag<(add node:$LHS, (int_aarch64_neon_umull node:$MHS, node:$RHS))>>;
4430 defm UMLSL : SIMDVectorIndexedLongSDTied<1, 0b0110, "umlsl",
4431 TriOpFrag<(sub node:$LHS, (int_aarch64_neon_umull node:$MHS, node:$RHS))>>;
4432 defm UMULL : SIMDVectorIndexedLongSD<1, 0b1010, "umull",
4433 int_aarch64_neon_umull>;
4435 // A scalar sqdmull with the second operand being a vector lane can be
4436 // handled directly with the indexed instruction encoding.
4437 def : Pat<(int_aarch64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
4438 (vector_extract (v4i32 V128:$Vm),
4439 VectorIndexS:$idx)),
4440 (SQDMULLv1i64_indexed FPR32:$Rn, V128:$Vm, VectorIndexS:$idx)>;
4442 //----------------------------------------------------------------------------
4443 // AdvSIMD scalar shift instructions
4444 //----------------------------------------------------------------------------
4445 defm FCVTZS : SIMDScalarRShiftSD<0, 0b11111, "fcvtzs">;
4446 defm FCVTZU : SIMDScalarRShiftSD<1, 0b11111, "fcvtzu">;
4447 defm SCVTF : SIMDScalarRShiftSD<0, 0b11100, "scvtf">;
4448 defm UCVTF : SIMDScalarRShiftSD<1, 0b11100, "ucvtf">;
4449 // Codegen patterns for the above. We don't put these directly on the
4450 // instructions because TableGen's type inference can't handle the truth.
4451 // Having the same base pattern for fp <--> int totally freaks it out.
4452 def : Pat<(int_aarch64_neon_vcvtfp2fxs FPR32:$Rn, vecshiftR32:$imm),
4453 (FCVTZSs FPR32:$Rn, vecshiftR32:$imm)>;
4454 def : Pat<(int_aarch64_neon_vcvtfp2fxu FPR32:$Rn, vecshiftR32:$imm),
4455 (FCVTZUs FPR32:$Rn, vecshiftR32:$imm)>;
4456 def : Pat<(i64 (int_aarch64_neon_vcvtfp2fxs (f64 FPR64:$Rn), vecshiftR64:$imm)),
4457 (FCVTZSd FPR64:$Rn, vecshiftR64:$imm)>;
4458 def : Pat<(i64 (int_aarch64_neon_vcvtfp2fxu (f64 FPR64:$Rn), vecshiftR64:$imm)),
4459 (FCVTZUd FPR64:$Rn, vecshiftR64:$imm)>;
4460 def : Pat<(v1i64 (int_aarch64_neon_vcvtfp2fxs (v1f64 FPR64:$Rn),
4462 (FCVTZSd FPR64:$Rn, vecshiftR64:$imm)>;
4463 def : Pat<(v1i64 (int_aarch64_neon_vcvtfp2fxu (v1f64 FPR64:$Rn),
4465 (FCVTZUd FPR64:$Rn, vecshiftR64:$imm)>;
4466 def : Pat<(int_aarch64_neon_vcvtfxs2fp FPR32:$Rn, vecshiftR32:$imm),
4467 (SCVTFs FPR32:$Rn, vecshiftR32:$imm)>;
4468 def : Pat<(int_aarch64_neon_vcvtfxu2fp FPR32:$Rn, vecshiftR32:$imm),
4469 (UCVTFs FPR32:$Rn, vecshiftR32:$imm)>;
4470 def : Pat<(f64 (int_aarch64_neon_vcvtfxs2fp (i64 FPR64:$Rn), vecshiftR64:$imm)),
4471 (SCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
4472 def : Pat<(f64 (int_aarch64_neon_vcvtfxu2fp (i64 FPR64:$Rn), vecshiftR64:$imm)),
4473 (UCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
4474 def : Pat<(v1f64 (int_aarch64_neon_vcvtfxs2fp (v1i64 FPR64:$Rn),
4476 (SCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
4477 def : Pat<(v1f64 (int_aarch64_neon_vcvtfxu2fp (v1i64 FPR64:$Rn),
4479 (UCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
4481 defm SHL : SIMDScalarLShiftD< 0, 0b01010, "shl", AArch64vshl>;
4482 defm SLI : SIMDScalarLShiftDTied<1, 0b01010, "sli">;
4483 defm SQRSHRN : SIMDScalarRShiftBHS< 0, 0b10011, "sqrshrn",
4484 int_aarch64_neon_sqrshrn>;
4485 defm SQRSHRUN : SIMDScalarRShiftBHS< 1, 0b10001, "sqrshrun",
4486 int_aarch64_neon_sqrshrun>;
4487 defm SQSHLU : SIMDScalarLShiftBHSD<1, 0b01100, "sqshlu", AArch64sqshlui>;
4488 defm SQSHL : SIMDScalarLShiftBHSD<0, 0b01110, "sqshl", AArch64sqshli>;
4489 defm SQSHRN : SIMDScalarRShiftBHS< 0, 0b10010, "sqshrn",
4490 int_aarch64_neon_sqshrn>;
4491 defm SQSHRUN : SIMDScalarRShiftBHS< 1, 0b10000, "sqshrun",
4492 int_aarch64_neon_sqshrun>;
4493 defm SRI : SIMDScalarRShiftDTied< 1, 0b01000, "sri">;
4494 defm SRSHR : SIMDScalarRShiftD< 0, 0b00100, "srshr", AArch64srshri>;
4495 defm SRSRA : SIMDScalarRShiftDTied< 0, 0b00110, "srsra",
4496 TriOpFrag<(add node:$LHS,
4497 (AArch64srshri node:$MHS, node:$RHS))>>;
4498 defm SSHR : SIMDScalarRShiftD< 0, 0b00000, "sshr", AArch64vashr>;
4499 defm SSRA : SIMDScalarRShiftDTied< 0, 0b00010, "ssra",
4500 TriOpFrag<(add node:$LHS,
4501 (AArch64vashr node:$MHS, node:$RHS))>>;
4502 defm UQRSHRN : SIMDScalarRShiftBHS< 1, 0b10011, "uqrshrn",
4503 int_aarch64_neon_uqrshrn>;
4504 defm UQSHL : SIMDScalarLShiftBHSD<1, 0b01110, "uqshl", AArch64uqshli>;
4505 defm UQSHRN : SIMDScalarRShiftBHS< 1, 0b10010, "uqshrn",
4506 int_aarch64_neon_uqshrn>;
4507 defm URSHR : SIMDScalarRShiftD< 1, 0b00100, "urshr", AArch64urshri>;
4508 defm URSRA : SIMDScalarRShiftDTied< 1, 0b00110, "ursra",
4509 TriOpFrag<(add node:$LHS,
4510 (AArch64urshri node:$MHS, node:$RHS))>>;
4511 defm USHR : SIMDScalarRShiftD< 1, 0b00000, "ushr", AArch64vlshr>;
4512 defm USRA : SIMDScalarRShiftDTied< 1, 0b00010, "usra",
4513 TriOpFrag<(add node:$LHS,
4514 (AArch64vlshr node:$MHS, node:$RHS))>>;
4516 //----------------------------------------------------------------------------
4517 // AdvSIMD vector shift instructions
4518 //----------------------------------------------------------------------------
4519 defm FCVTZS:SIMDVectorRShiftSD<0, 0b11111, "fcvtzs", int_aarch64_neon_vcvtfp2fxs>;
4520 defm FCVTZU:SIMDVectorRShiftSD<1, 0b11111, "fcvtzu", int_aarch64_neon_vcvtfp2fxu>;
4521 defm SCVTF: SIMDVectorRShiftSDToFP<0, 0b11100, "scvtf",
4522 int_aarch64_neon_vcvtfxs2fp>;
4523 defm RSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10001, "rshrn",
4524 int_aarch64_neon_rshrn>;
4525 defm SHL : SIMDVectorLShiftBHSD<0, 0b01010, "shl", AArch64vshl>;
4526 defm SHRN : SIMDVectorRShiftNarrowBHS<0, 0b10000, "shrn",
4527 BinOpFrag<(trunc (AArch64vashr node:$LHS, node:$RHS))>>;
4528 defm SLI : SIMDVectorLShiftBHSDTied<1, 0b01010, "sli", int_aarch64_neon_vsli>;
4529 def : Pat<(v1i64 (int_aarch64_neon_vsli (v1i64 FPR64:$Rd), (v1i64 FPR64:$Rn),
4530 (i32 vecshiftL64:$imm))),
4531 (SLId FPR64:$Rd, FPR64:$Rn, vecshiftL64:$imm)>;
4532 defm SQRSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10011, "sqrshrn",
4533 int_aarch64_neon_sqrshrn>;
4534 defm SQRSHRUN: SIMDVectorRShiftNarrowBHS<1, 0b10001, "sqrshrun",
4535 int_aarch64_neon_sqrshrun>;
4536 defm SQSHLU : SIMDVectorLShiftBHSD<1, 0b01100, "sqshlu", AArch64sqshlui>;
4537 defm SQSHL : SIMDVectorLShiftBHSD<0, 0b01110, "sqshl", AArch64sqshli>;
4538 defm SQSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10010, "sqshrn",
4539 int_aarch64_neon_sqshrn>;
4540 defm SQSHRUN : SIMDVectorRShiftNarrowBHS<1, 0b10000, "sqshrun",
4541 int_aarch64_neon_sqshrun>;
4542 defm SRI : SIMDVectorRShiftBHSDTied<1, 0b01000, "sri", int_aarch64_neon_vsri>;
4543 def : Pat<(v1i64 (int_aarch64_neon_vsri (v1i64 FPR64:$Rd), (v1i64 FPR64:$Rn),
4544 (i32 vecshiftR64:$imm))),
4545 (SRId FPR64:$Rd, FPR64:$Rn, vecshiftR64:$imm)>;
4546 defm SRSHR : SIMDVectorRShiftBHSD<0, 0b00100, "srshr", AArch64srshri>;
4547 defm SRSRA : SIMDVectorRShiftBHSDTied<0, 0b00110, "srsra",
4548 TriOpFrag<(add node:$LHS,
4549 (AArch64srshri node:$MHS, node:$RHS))> >;
4550 defm SSHLL : SIMDVectorLShiftLongBHSD<0, 0b10100, "sshll",
4551 BinOpFrag<(AArch64vshl (sext node:$LHS), node:$RHS)>>;
4553 defm SSHR : SIMDVectorRShiftBHSD<0, 0b00000, "sshr", AArch64vashr>;
4554 defm SSRA : SIMDVectorRShiftBHSDTied<0, 0b00010, "ssra",
4555 TriOpFrag<(add node:$LHS, (AArch64vashr node:$MHS, node:$RHS))>>;
4556 defm UCVTF : SIMDVectorRShiftSDToFP<1, 0b11100, "ucvtf",
4557 int_aarch64_neon_vcvtfxu2fp>;
4558 defm UQRSHRN : SIMDVectorRShiftNarrowBHS<1, 0b10011, "uqrshrn",
4559 int_aarch64_neon_uqrshrn>;
4560 defm UQSHL : SIMDVectorLShiftBHSD<1, 0b01110, "uqshl", AArch64uqshli>;
4561 defm UQSHRN : SIMDVectorRShiftNarrowBHS<1, 0b10010, "uqshrn",
4562 int_aarch64_neon_uqshrn>;
4563 defm URSHR : SIMDVectorRShiftBHSD<1, 0b00100, "urshr", AArch64urshri>;
4564 defm URSRA : SIMDVectorRShiftBHSDTied<1, 0b00110, "ursra",
4565 TriOpFrag<(add node:$LHS,
4566 (AArch64urshri node:$MHS, node:$RHS))> >;
4567 defm USHLL : SIMDVectorLShiftLongBHSD<1, 0b10100, "ushll",
4568 BinOpFrag<(AArch64vshl (zext node:$LHS), node:$RHS)>>;
4569 defm USHR : SIMDVectorRShiftBHSD<1, 0b00000, "ushr", AArch64vlshr>;
4570 defm USRA : SIMDVectorRShiftBHSDTied<1, 0b00010, "usra",
4571 TriOpFrag<(add node:$LHS, (AArch64vlshr node:$MHS, node:$RHS))> >;
4573 // SHRN patterns for when a logical right shift was used instead of arithmetic
4574 // (the immediate guarantees no sign bits actually end up in the result so it
4576 def : Pat<(v8i8 (trunc (AArch64vlshr (v8i16 V128:$Rn), vecshiftR16Narrow:$imm))),
4577 (SHRNv8i8_shift V128:$Rn, vecshiftR16Narrow:$imm)>;
4578 def : Pat<(v4i16 (trunc (AArch64vlshr (v4i32 V128:$Rn), vecshiftR32Narrow:$imm))),
4579 (SHRNv4i16_shift V128:$Rn, vecshiftR32Narrow:$imm)>;
4580 def : Pat<(v2i32 (trunc (AArch64vlshr (v2i64 V128:$Rn), vecshiftR64Narrow:$imm))),
4581 (SHRNv2i32_shift V128:$Rn, vecshiftR64Narrow:$imm)>;
4583 def : Pat<(v16i8 (concat_vectors (v8i8 V64:$Rd),
4584 (trunc (AArch64vlshr (v8i16 V128:$Rn),
4585 vecshiftR16Narrow:$imm)))),
4586 (SHRNv16i8_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
4587 V128:$Rn, vecshiftR16Narrow:$imm)>;
4588 def : Pat<(v8i16 (concat_vectors (v4i16 V64:$Rd),
4589 (trunc (AArch64vlshr (v4i32 V128:$Rn),
4590 vecshiftR32Narrow:$imm)))),
4591 (SHRNv8i16_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
4592 V128:$Rn, vecshiftR32Narrow:$imm)>;
4593 def : Pat<(v4i32 (concat_vectors (v2i32 V64:$Rd),
4594 (trunc (AArch64vlshr (v2i64 V128:$Rn),
4595 vecshiftR64Narrow:$imm)))),
4596 (SHRNv4i32_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
4597 V128:$Rn, vecshiftR32Narrow:$imm)>;
4599 // Vector sign and zero extensions are implemented with SSHLL and USSHLL.
4600 // Anyexts are implemented as zexts.
4601 def : Pat<(v8i16 (sext (v8i8 V64:$Rn))), (SSHLLv8i8_shift V64:$Rn, (i32 0))>;
4602 def : Pat<(v8i16 (zext (v8i8 V64:$Rn))), (USHLLv8i8_shift V64:$Rn, (i32 0))>;
4603 def : Pat<(v8i16 (anyext (v8i8 V64:$Rn))), (USHLLv8i8_shift V64:$Rn, (i32 0))>;
4604 def : Pat<(v4i32 (sext (v4i16 V64:$Rn))), (SSHLLv4i16_shift V64:$Rn, (i32 0))>;
4605 def : Pat<(v4i32 (zext (v4i16 V64:$Rn))), (USHLLv4i16_shift V64:$Rn, (i32 0))>;
4606 def : Pat<(v4i32 (anyext (v4i16 V64:$Rn))), (USHLLv4i16_shift V64:$Rn, (i32 0))>;
4607 def : Pat<(v2i64 (sext (v2i32 V64:$Rn))), (SSHLLv2i32_shift V64:$Rn, (i32 0))>;
4608 def : Pat<(v2i64 (zext (v2i32 V64:$Rn))), (USHLLv2i32_shift V64:$Rn, (i32 0))>;
4609 def : Pat<(v2i64 (anyext (v2i32 V64:$Rn))), (USHLLv2i32_shift V64:$Rn, (i32 0))>;
4610 // Also match an extend from the upper half of a 128 bit source register.
4611 def : Pat<(v8i16 (anyext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
4612 (USHLLv16i8_shift V128:$Rn, (i32 0))>;
4613 def : Pat<(v8i16 (zext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
4614 (USHLLv16i8_shift V128:$Rn, (i32 0))>;
4615 def : Pat<(v8i16 (sext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
4616 (SSHLLv16i8_shift V128:$Rn, (i32 0))>;
4617 def : Pat<(v4i32 (anyext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
4618 (USHLLv8i16_shift V128:$Rn, (i32 0))>;
4619 def : Pat<(v4i32 (zext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
4620 (USHLLv8i16_shift V128:$Rn, (i32 0))>;
4621 def : Pat<(v4i32 (sext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
4622 (SSHLLv8i16_shift V128:$Rn, (i32 0))>;
4623 def : Pat<(v2i64 (anyext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
4624 (USHLLv4i32_shift V128:$Rn, (i32 0))>;
4625 def : Pat<(v2i64 (zext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
4626 (USHLLv4i32_shift V128:$Rn, (i32 0))>;
4627 def : Pat<(v2i64 (sext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
4628 (SSHLLv4i32_shift V128:$Rn, (i32 0))>;
4630 // Vector shift sxtl aliases
4631 def : InstAlias<"sxtl.8h $dst, $src1",
4632 (SSHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
4633 def : InstAlias<"sxtl $dst.8h, $src1.8b",
4634 (SSHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
4635 def : InstAlias<"sxtl.4s $dst, $src1",
4636 (SSHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
4637 def : InstAlias<"sxtl $dst.4s, $src1.4h",
4638 (SSHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
4639 def : InstAlias<"sxtl.2d $dst, $src1",
4640 (SSHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
4641 def : InstAlias<"sxtl $dst.2d, $src1.2s",
4642 (SSHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
4644 // Vector shift sxtl2 aliases
4645 def : InstAlias<"sxtl2.8h $dst, $src1",
4646 (SSHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
4647 def : InstAlias<"sxtl2 $dst.8h, $src1.16b",
4648 (SSHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
4649 def : InstAlias<"sxtl2.4s $dst, $src1",
4650 (SSHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
4651 def : InstAlias<"sxtl2 $dst.4s, $src1.8h",
4652 (SSHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
4653 def : InstAlias<"sxtl2.2d $dst, $src1",
4654 (SSHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
4655 def : InstAlias<"sxtl2 $dst.2d, $src1.4s",
4656 (SSHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
4658 // Vector shift uxtl aliases
4659 def : InstAlias<"uxtl.8h $dst, $src1",
4660 (USHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
4661 def : InstAlias<"uxtl $dst.8h, $src1.8b",
4662 (USHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
4663 def : InstAlias<"uxtl.4s $dst, $src1",
4664 (USHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
4665 def : InstAlias<"uxtl $dst.4s, $src1.4h",
4666 (USHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
4667 def : InstAlias<"uxtl.2d $dst, $src1",
4668 (USHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
4669 def : InstAlias<"uxtl $dst.2d, $src1.2s",
4670 (USHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
4672 // Vector shift uxtl2 aliases
4673 def : InstAlias<"uxtl2.8h $dst, $src1",
4674 (USHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
4675 def : InstAlias<"uxtl2 $dst.8h, $src1.16b",
4676 (USHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
4677 def : InstAlias<"uxtl2.4s $dst, $src1",
4678 (USHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
4679 def : InstAlias<"uxtl2 $dst.4s, $src1.8h",
4680 (USHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
4681 def : InstAlias<"uxtl2.2d $dst, $src1",
4682 (USHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
4683 def : InstAlias<"uxtl2 $dst.2d, $src1.4s",
4684 (USHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
4686 // If an integer is about to be converted to a floating point value,
4687 // just load it on the floating point unit.
4688 // These patterns are more complex because floating point loads do not
4689 // support sign extension.
4690 // The sign extension has to be explicitly added and is only supported for
4691 // one step: byte-to-half, half-to-word, word-to-doubleword.
4692 // SCVTF GPR -> FPR is 9 cycles.
4693 // SCVTF FPR -> FPR is 4 cyclces.
4694 // (sign extension with lengthen) SXTL FPR -> FPR is 2 cycles.
4695 // Therefore, we can do 2 sign extensions and one SCVTF FPR -> FPR
4696 // and still being faster.
4697 // However, this is not good for code size.
4698 // 8-bits -> float. 2 sizes step-up.
4699 class SExtLoadi8CVTf32Pat<dag addrmode, dag INST>
4700 : Pat<(f32 (sint_to_fp (i32 (sextloadi8 addrmode)))),
4701 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
4706 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4712 ssub)))>, Requires<[NotForCodeSize, IsCyclone]>;
4714 def : SExtLoadi8CVTf32Pat<(ro8.Wpat GPR64sp:$Rn, GPR32:$Rm, ro8.Wext:$ext),
4715 (LDRBroW GPR64sp:$Rn, GPR32:$Rm, ro8.Wext:$ext)>;
4716 def : SExtLoadi8CVTf32Pat<(ro8.Xpat GPR64sp:$Rn, GPR64:$Rm, ro8.Xext:$ext),
4717 (LDRBroX GPR64sp:$Rn, GPR64:$Rm, ro8.Xext:$ext)>;
4718 def : SExtLoadi8CVTf32Pat<(am_indexed8 GPR64sp:$Rn, uimm12s1:$offset),
4719 (LDRBui GPR64sp:$Rn, uimm12s1:$offset)>;
4720 def : SExtLoadi8CVTf32Pat<(am_unscaled8 GPR64sp:$Rn, simm9:$offset),
4721 (LDURBi GPR64sp:$Rn, simm9:$offset)>;
4723 // 16-bits -> float. 1 size step-up.
4724 class SExtLoadi16CVTf32Pat<dag addrmode, dag INST>
4725 : Pat<(f32 (sint_to_fp (i32 (sextloadi16 addrmode)))),
4726 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
4728 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4732 ssub)))>, Requires<[NotForCodeSize]>;
4734 def : SExtLoadi16CVTf32Pat<(ro16.Wpat GPR64sp:$Rn, GPR32:$Rm, ro16.Wext:$ext),
4735 (LDRHroW GPR64sp:$Rn, GPR32:$Rm, ro16.Wext:$ext)>;
4736 def : SExtLoadi16CVTf32Pat<(ro16.Xpat GPR64sp:$Rn, GPR64:$Rm, ro16.Xext:$ext),
4737 (LDRHroX GPR64sp:$Rn, GPR64:$Rm, ro16.Xext:$ext)>;
4738 def : SExtLoadi16CVTf32Pat<(am_indexed16 GPR64sp:$Rn, uimm12s2:$offset),
4739 (LDRHui GPR64sp:$Rn, uimm12s2:$offset)>;
4740 def : SExtLoadi16CVTf32Pat<(am_unscaled16 GPR64sp:$Rn, simm9:$offset),
4741 (LDURHi GPR64sp:$Rn, simm9:$offset)>;
4743 // 32-bits to 32-bits are handled in target specific dag combine:
4744 // performIntToFpCombine.
4745 // 64-bits integer to 32-bits floating point, not possible with
4746 // SCVTF on floating point registers (both source and destination
4747 // must have the same size).
4749 // Here are the patterns for 8, 16, 32, and 64-bits to double.
4750 // 8-bits -> double. 3 size step-up: give up.
4751 // 16-bits -> double. 2 size step.
4752 class SExtLoadi16CVTf64Pat<dag addrmode, dag INST>
4753 : Pat <(f64 (sint_to_fp (i32 (sextloadi16 addrmode)))),
4754 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4759 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4765 dsub)))>, Requires<[NotForCodeSize, IsCyclone]>;
4767 def : SExtLoadi16CVTf64Pat<(ro16.Wpat GPR64sp:$Rn, GPR32:$Rm, ro16.Wext:$ext),
4768 (LDRHroW GPR64sp:$Rn, GPR32:$Rm, ro16.Wext:$ext)>;
4769 def : SExtLoadi16CVTf64Pat<(ro16.Xpat GPR64sp:$Rn, GPR64:$Rm, ro16.Xext:$ext),
4770 (LDRHroX GPR64sp:$Rn, GPR64:$Rm, ro16.Xext:$ext)>;
4771 def : SExtLoadi16CVTf64Pat<(am_indexed16 GPR64sp:$Rn, uimm12s2:$offset),
4772 (LDRHui GPR64sp:$Rn, uimm12s2:$offset)>;
4773 def : SExtLoadi16CVTf64Pat<(am_unscaled16 GPR64sp:$Rn, simm9:$offset),
4774 (LDURHi GPR64sp:$Rn, simm9:$offset)>;
4775 // 32-bits -> double. 1 size step-up.
4776 class SExtLoadi32CVTf64Pat<dag addrmode, dag INST>
4777 : Pat <(f64 (sint_to_fp (i32 (load addrmode)))),
4778 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4780 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4784 dsub)))>, Requires<[NotForCodeSize]>;
4786 def : SExtLoadi32CVTf64Pat<(ro32.Wpat GPR64sp:$Rn, GPR32:$Rm, ro32.Wext:$ext),
4787 (LDRSroW GPR64sp:$Rn, GPR32:$Rm, ro32.Wext:$ext)>;
4788 def : SExtLoadi32CVTf64Pat<(ro32.Xpat GPR64sp:$Rn, GPR64:$Rm, ro32.Xext:$ext),
4789 (LDRSroX GPR64sp:$Rn, GPR64:$Rm, ro32.Xext:$ext)>;
4790 def : SExtLoadi32CVTf64Pat<(am_indexed32 GPR64sp:$Rn, uimm12s4:$offset),
4791 (LDRSui GPR64sp:$Rn, uimm12s4:$offset)>;
4792 def : SExtLoadi32CVTf64Pat<(am_unscaled32 GPR64sp:$Rn, simm9:$offset),
4793 (LDURSi GPR64sp:$Rn, simm9:$offset)>;
4795 // 64-bits -> double are handled in target specific dag combine:
4796 // performIntToFpCombine.
4799 //----------------------------------------------------------------------------
4800 // AdvSIMD Load-Store Structure
4801 //----------------------------------------------------------------------------
4802 defm LD1 : SIMDLd1Multiple<"ld1">;
4803 defm LD2 : SIMDLd2Multiple<"ld2">;
4804 defm LD3 : SIMDLd3Multiple<"ld3">;
4805 defm LD4 : SIMDLd4Multiple<"ld4">;
4807 defm ST1 : SIMDSt1Multiple<"st1">;
4808 defm ST2 : SIMDSt2Multiple<"st2">;
4809 defm ST3 : SIMDSt3Multiple<"st3">;
4810 defm ST4 : SIMDSt4Multiple<"st4">;
4812 class Ld1Pat<ValueType ty, Instruction INST>
4813 : Pat<(ty (load GPR64sp:$Rn)), (INST GPR64sp:$Rn)>;
4815 def : Ld1Pat<v16i8, LD1Onev16b>;
4816 def : Ld1Pat<v8i16, LD1Onev8h>;
4817 def : Ld1Pat<v4i32, LD1Onev4s>;
4818 def : Ld1Pat<v2i64, LD1Onev2d>;
4819 def : Ld1Pat<v8i8, LD1Onev8b>;
4820 def : Ld1Pat<v4i16, LD1Onev4h>;
4821 def : Ld1Pat<v2i32, LD1Onev2s>;
4822 def : Ld1Pat<v1i64, LD1Onev1d>;
4824 class St1Pat<ValueType ty, Instruction INST>
4825 : Pat<(store ty:$Vt, GPR64sp:$Rn),
4826 (INST ty:$Vt, GPR64sp:$Rn)>;
4828 def : St1Pat<v16i8, ST1Onev16b>;
4829 def : St1Pat<v8i16, ST1Onev8h>;
4830 def : St1Pat<v4i32, ST1Onev4s>;
4831 def : St1Pat<v2i64, ST1Onev2d>;
4832 def : St1Pat<v8i8, ST1Onev8b>;
4833 def : St1Pat<v4i16, ST1Onev4h>;
4834 def : St1Pat<v2i32, ST1Onev2s>;
4835 def : St1Pat<v1i64, ST1Onev1d>;
4841 defm LD1R : SIMDLdR<0, 0b110, 0, "ld1r", "One", 1, 2, 4, 8>;
4842 defm LD2R : SIMDLdR<1, 0b110, 0, "ld2r", "Two", 2, 4, 8, 16>;
4843 defm LD3R : SIMDLdR<0, 0b111, 0, "ld3r", "Three", 3, 6, 12, 24>;
4844 defm LD4R : SIMDLdR<1, 0b111, 0, "ld4r", "Four", 4, 8, 16, 32>;
4845 let mayLoad = 1, hasSideEffects = 0 in {
4846 defm LD1 : SIMDLdSingleBTied<0, 0b000, "ld1", VecListOneb, GPR64pi1>;
4847 defm LD1 : SIMDLdSingleHTied<0, 0b010, 0, "ld1", VecListOneh, GPR64pi2>;
4848 defm LD1 : SIMDLdSingleSTied<0, 0b100, 0b00, "ld1", VecListOnes, GPR64pi4>;
4849 defm LD1 : SIMDLdSingleDTied<0, 0b100, 0b01, "ld1", VecListOned, GPR64pi8>;
4850 defm LD2 : SIMDLdSingleBTied<1, 0b000, "ld2", VecListTwob, GPR64pi2>;
4851 defm LD2 : SIMDLdSingleHTied<1, 0b010, 0, "ld2", VecListTwoh, GPR64pi4>;
4852 defm LD2 : SIMDLdSingleSTied<1, 0b100, 0b00, "ld2", VecListTwos, GPR64pi8>;
4853 defm LD2 : SIMDLdSingleDTied<1, 0b100, 0b01, "ld2", VecListTwod, GPR64pi16>;
4854 defm LD3 : SIMDLdSingleBTied<0, 0b001, "ld3", VecListThreeb, GPR64pi3>;
4855 defm LD3 : SIMDLdSingleHTied<0, 0b011, 0, "ld3", VecListThreeh, GPR64pi6>;
4856 defm LD3 : SIMDLdSingleSTied<0, 0b101, 0b00, "ld3", VecListThrees, GPR64pi12>;
4857 defm LD3 : SIMDLdSingleDTied<0, 0b101, 0b01, "ld3", VecListThreed, GPR64pi24>;
4858 defm LD4 : SIMDLdSingleBTied<1, 0b001, "ld4", VecListFourb, GPR64pi4>;
4859 defm LD4 : SIMDLdSingleHTied<1, 0b011, 0, "ld4", VecListFourh, GPR64pi8>;
4860 defm LD4 : SIMDLdSingleSTied<1, 0b101, 0b00, "ld4", VecListFours, GPR64pi16>;
4861 defm LD4 : SIMDLdSingleDTied<1, 0b101, 0b01, "ld4", VecListFourd, GPR64pi32>;
4864 def : Pat<(v8i8 (AArch64dup (i32 (extloadi8 GPR64sp:$Rn)))),
4865 (LD1Rv8b GPR64sp:$Rn)>;
4866 def : Pat<(v16i8 (AArch64dup (i32 (extloadi8 GPR64sp:$Rn)))),
4867 (LD1Rv16b GPR64sp:$Rn)>;
4868 def : Pat<(v4i16 (AArch64dup (i32 (extloadi16 GPR64sp:$Rn)))),
4869 (LD1Rv4h GPR64sp:$Rn)>;
4870 def : Pat<(v8i16 (AArch64dup (i32 (extloadi16 GPR64sp:$Rn)))),
4871 (LD1Rv8h GPR64sp:$Rn)>;
4872 def : Pat<(v2i32 (AArch64dup (i32 (load GPR64sp:$Rn)))),
4873 (LD1Rv2s GPR64sp:$Rn)>;
4874 def : Pat<(v4i32 (AArch64dup (i32 (load GPR64sp:$Rn)))),
4875 (LD1Rv4s GPR64sp:$Rn)>;
4876 def : Pat<(v2i64 (AArch64dup (i64 (load GPR64sp:$Rn)))),
4877 (LD1Rv2d GPR64sp:$Rn)>;
4878 def : Pat<(v1i64 (AArch64dup (i64 (load GPR64sp:$Rn)))),
4879 (LD1Rv1d GPR64sp:$Rn)>;
4880 // Grab the floating point version too
4881 def : Pat<(v2f32 (AArch64dup (f32 (load GPR64sp:$Rn)))),
4882 (LD1Rv2s GPR64sp:$Rn)>;
4883 def : Pat<(v4f32 (AArch64dup (f32 (load GPR64sp:$Rn)))),
4884 (LD1Rv4s GPR64sp:$Rn)>;
4885 def : Pat<(v2f64 (AArch64dup (f64 (load GPR64sp:$Rn)))),
4886 (LD1Rv2d GPR64sp:$Rn)>;
4887 def : Pat<(v1f64 (AArch64dup (f64 (load GPR64sp:$Rn)))),
4888 (LD1Rv1d GPR64sp:$Rn)>;
4889 def : Pat<(v4f16 (AArch64dup (f16 (load GPR64sp:$Rn)))),
4890 (LD1Rv4h GPR64sp:$Rn)>;
4891 def : Pat<(v8f16 (AArch64dup (f16 (load GPR64sp:$Rn)))),
4892 (LD1Rv8h GPR64sp:$Rn)>;
4894 class Ld1Lane128Pat<SDPatternOperator scalar_load, Operand VecIndex,
4895 ValueType VTy, ValueType STy, Instruction LD1>
4896 : Pat<(vector_insert (VTy VecListOne128:$Rd),
4897 (STy (scalar_load GPR64sp:$Rn)), VecIndex:$idx),
4898 (LD1 VecListOne128:$Rd, VecIndex:$idx, GPR64sp:$Rn)>;
4900 def : Ld1Lane128Pat<extloadi8, VectorIndexB, v16i8, i32, LD1i8>;
4901 def : Ld1Lane128Pat<extloadi16, VectorIndexH, v8i16, i32, LD1i16>;
4902 def : Ld1Lane128Pat<load, VectorIndexS, v4i32, i32, LD1i32>;
4903 def : Ld1Lane128Pat<load, VectorIndexS, v4f32, f32, LD1i32>;
4904 def : Ld1Lane128Pat<load, VectorIndexD, v2i64, i64, LD1i64>;
4905 def : Ld1Lane128Pat<load, VectorIndexD, v2f64, f64, LD1i64>;
4906 def : Ld1Lane128Pat<load, VectorIndexH, v8f16, f16, LD1i16>;
4908 class Ld1Lane64Pat<SDPatternOperator scalar_load, Operand VecIndex,
4909 ValueType VTy, ValueType STy, Instruction LD1>
4910 : Pat<(vector_insert (VTy VecListOne64:$Rd),
4911 (STy (scalar_load GPR64sp:$Rn)), VecIndex:$idx),
4913 (LD1 (SUBREG_TO_REG (i32 0), VecListOne64:$Rd, dsub),
4914 VecIndex:$idx, GPR64sp:$Rn),
4917 def : Ld1Lane64Pat<extloadi8, VectorIndexB, v8i8, i32, LD1i8>;
4918 def : Ld1Lane64Pat<extloadi16, VectorIndexH, v4i16, i32, LD1i16>;
4919 def : Ld1Lane64Pat<load, VectorIndexS, v2i32, i32, LD1i32>;
4920 def : Ld1Lane64Pat<load, VectorIndexS, v2f32, f32, LD1i32>;
4921 def : Ld1Lane64Pat<load, VectorIndexH, v4f16, f16, LD1i16>;
4924 defm LD1 : SIMDLdSt1SingleAliases<"ld1">;
4925 defm LD2 : SIMDLdSt2SingleAliases<"ld2">;
4926 defm LD3 : SIMDLdSt3SingleAliases<"ld3">;
4927 defm LD4 : SIMDLdSt4SingleAliases<"ld4">;
4930 defm ST1 : SIMDStSingleB<0, 0b000, "st1", VecListOneb, GPR64pi1>;
4931 defm ST1 : SIMDStSingleH<0, 0b010, 0, "st1", VecListOneh, GPR64pi2>;
4932 defm ST1 : SIMDStSingleS<0, 0b100, 0b00, "st1", VecListOnes, GPR64pi4>;
4933 defm ST1 : SIMDStSingleD<0, 0b100, 0b01, "st1", VecListOned, GPR64pi8>;
4935 let AddedComplexity = 19 in
4936 class St1Lane128Pat<SDPatternOperator scalar_store, Operand VecIndex,
4937 ValueType VTy, ValueType STy, Instruction ST1>
4939 (STy (vector_extract (VTy VecListOne128:$Vt), VecIndex:$idx)),
4941 (ST1 VecListOne128:$Vt, VecIndex:$idx, GPR64sp:$Rn)>;
4943 def : St1Lane128Pat<truncstorei8, VectorIndexB, v16i8, i32, ST1i8>;
4944 def : St1Lane128Pat<truncstorei16, VectorIndexH, v8i16, i32, ST1i16>;
4945 def : St1Lane128Pat<store, VectorIndexS, v4i32, i32, ST1i32>;
4946 def : St1Lane128Pat<store, VectorIndexS, v4f32, f32, ST1i32>;
4947 def : St1Lane128Pat<store, VectorIndexD, v2i64, i64, ST1i64>;
4948 def : St1Lane128Pat<store, VectorIndexD, v2f64, f64, ST1i64>;
4949 def : St1Lane128Pat<store, VectorIndexH, v8f16, f16, ST1i16>;
4951 let AddedComplexity = 19 in
4952 class St1Lane64Pat<SDPatternOperator scalar_store, Operand VecIndex,
4953 ValueType VTy, ValueType STy, Instruction ST1>
4955 (STy (vector_extract (VTy VecListOne64:$Vt), VecIndex:$idx)),
4957 (ST1 (SUBREG_TO_REG (i32 0), VecListOne64:$Vt, dsub),
4958 VecIndex:$idx, GPR64sp:$Rn)>;
4960 def : St1Lane64Pat<truncstorei8, VectorIndexB, v8i8, i32, ST1i8>;
4961 def : St1Lane64Pat<truncstorei16, VectorIndexH, v4i16, i32, ST1i16>;
4962 def : St1Lane64Pat<store, VectorIndexS, v2i32, i32, ST1i32>;
4963 def : St1Lane64Pat<store, VectorIndexS, v2f32, f32, ST1i32>;
4964 def : St1Lane64Pat<store, VectorIndexH, v4f16, f16, ST1i16>;
4966 multiclass St1LanePost64Pat<SDPatternOperator scalar_store, Operand VecIndex,
4967 ValueType VTy, ValueType STy, Instruction ST1,
4969 def : Pat<(scalar_store
4970 (STy (vector_extract (VTy VecListOne64:$Vt), VecIndex:$idx)),
4971 GPR64sp:$Rn, offset),
4972 (ST1 (SUBREG_TO_REG (i32 0), VecListOne64:$Vt, dsub),
4973 VecIndex:$idx, GPR64sp:$Rn, XZR)>;
4975 def : Pat<(scalar_store
4976 (STy (vector_extract (VTy VecListOne64:$Vt), VecIndex:$idx)),
4977 GPR64sp:$Rn, GPR64:$Rm),
4978 (ST1 (SUBREG_TO_REG (i32 0), VecListOne64:$Vt, dsub),
4979 VecIndex:$idx, GPR64sp:$Rn, $Rm)>;
4982 defm : St1LanePost64Pat<post_truncsti8, VectorIndexB, v8i8, i32, ST1i8_POST, 1>;
4983 defm : St1LanePost64Pat<post_truncsti16, VectorIndexH, v4i16, i32, ST1i16_POST,
4985 defm : St1LanePost64Pat<post_store, VectorIndexS, v2i32, i32, ST1i32_POST, 4>;
4986 defm : St1LanePost64Pat<post_store, VectorIndexS, v2f32, f32, ST1i32_POST, 4>;
4987 defm : St1LanePost64Pat<post_store, VectorIndexD, v1i64, i64, ST1i64_POST, 8>;
4988 defm : St1LanePost64Pat<post_store, VectorIndexD, v1f64, f64, ST1i64_POST, 8>;
4989 defm : St1LanePost64Pat<post_store, VectorIndexH, v4f16, f16, ST1i16_POST, 2>;
4991 multiclass St1LanePost128Pat<SDPatternOperator scalar_store, Operand VecIndex,
4992 ValueType VTy, ValueType STy, Instruction ST1,
4994 def : Pat<(scalar_store
4995 (STy (vector_extract (VTy VecListOne128:$Vt), VecIndex:$idx)),
4996 GPR64sp:$Rn, offset),
4997 (ST1 VecListOne128:$Vt, VecIndex:$idx, GPR64sp:$Rn, XZR)>;
4999 def : Pat<(scalar_store
5000 (STy (vector_extract (VTy VecListOne128:$Vt), VecIndex:$idx)),
5001 GPR64sp:$Rn, GPR64:$Rm),
5002 (ST1 VecListOne128:$Vt, VecIndex:$idx, GPR64sp:$Rn, $Rm)>;
5005 defm : St1LanePost128Pat<post_truncsti8, VectorIndexB, v16i8, i32, ST1i8_POST,
5007 defm : St1LanePost128Pat<post_truncsti16, VectorIndexH, v8i16, i32, ST1i16_POST,
5009 defm : St1LanePost128Pat<post_store, VectorIndexS, v4i32, i32, ST1i32_POST, 4>;
5010 defm : St1LanePost128Pat<post_store, VectorIndexS, v4f32, f32, ST1i32_POST, 4>;
5011 defm : St1LanePost128Pat<post_store, VectorIndexD, v2i64, i64, ST1i64_POST, 8>;
5012 defm : St1LanePost128Pat<post_store, VectorIndexD, v2f64, f64, ST1i64_POST, 8>;
5013 defm : St1LanePost128Pat<post_store, VectorIndexH, v8f16, f16, ST1i16_POST, 2>;
5015 let mayStore = 1, hasSideEffects = 0 in {
5016 defm ST2 : SIMDStSingleB<1, 0b000, "st2", VecListTwob, GPR64pi2>;
5017 defm ST2 : SIMDStSingleH<1, 0b010, 0, "st2", VecListTwoh, GPR64pi4>;
5018 defm ST2 : SIMDStSingleS<1, 0b100, 0b00, "st2", VecListTwos, GPR64pi8>;
5019 defm ST2 : SIMDStSingleD<1, 0b100, 0b01, "st2", VecListTwod, GPR64pi16>;
5020 defm ST3 : SIMDStSingleB<0, 0b001, "st3", VecListThreeb, GPR64pi3>;
5021 defm ST3 : SIMDStSingleH<0, 0b011, 0, "st3", VecListThreeh, GPR64pi6>;
5022 defm ST3 : SIMDStSingleS<0, 0b101, 0b00, "st3", VecListThrees, GPR64pi12>;
5023 defm ST3 : SIMDStSingleD<0, 0b101, 0b01, "st3", VecListThreed, GPR64pi24>;
5024 defm ST4 : SIMDStSingleB<1, 0b001, "st4", VecListFourb, GPR64pi4>;
5025 defm ST4 : SIMDStSingleH<1, 0b011, 0, "st4", VecListFourh, GPR64pi8>;
5026 defm ST4 : SIMDStSingleS<1, 0b101, 0b00, "st4", VecListFours, GPR64pi16>;
5027 defm ST4 : SIMDStSingleD<1, 0b101, 0b01, "st4", VecListFourd, GPR64pi32>;
5030 defm ST1 : SIMDLdSt1SingleAliases<"st1">;
5031 defm ST2 : SIMDLdSt2SingleAliases<"st2">;
5032 defm ST3 : SIMDLdSt3SingleAliases<"st3">;
5033 defm ST4 : SIMDLdSt4SingleAliases<"st4">;
5035 //----------------------------------------------------------------------------
5036 // Crypto extensions
5037 //----------------------------------------------------------------------------
5039 def AESErr : AESTiedInst<0b0100, "aese", int_aarch64_crypto_aese>;
5040 def AESDrr : AESTiedInst<0b0101, "aesd", int_aarch64_crypto_aesd>;
5041 def AESMCrr : AESInst< 0b0110, "aesmc", int_aarch64_crypto_aesmc>;
5042 def AESIMCrr : AESInst< 0b0111, "aesimc", int_aarch64_crypto_aesimc>;
5044 def SHA1Crrr : SHATiedInstQSV<0b000, "sha1c", int_aarch64_crypto_sha1c>;
5045 def SHA1Prrr : SHATiedInstQSV<0b001, "sha1p", int_aarch64_crypto_sha1p>;
5046 def SHA1Mrrr : SHATiedInstQSV<0b010, "sha1m", int_aarch64_crypto_sha1m>;
5047 def SHA1SU0rrr : SHATiedInstVVV<0b011, "sha1su0", int_aarch64_crypto_sha1su0>;
5048 def SHA256Hrrr : SHATiedInstQQV<0b100, "sha256h", int_aarch64_crypto_sha256h>;
5049 def SHA256H2rrr : SHATiedInstQQV<0b101, "sha256h2",int_aarch64_crypto_sha256h2>;
5050 def SHA256SU1rrr :SHATiedInstVVV<0b110, "sha256su1",int_aarch64_crypto_sha256su1>;
5052 def SHA1Hrr : SHAInstSS< 0b0000, "sha1h", int_aarch64_crypto_sha1h>;
5053 def SHA1SU1rr : SHATiedInstVV<0b0001, "sha1su1", int_aarch64_crypto_sha1su1>;
5054 def SHA256SU0rr : SHATiedInstVV<0b0010, "sha256su0",int_aarch64_crypto_sha256su0>;
5056 //----------------------------------------------------------------------------
5058 //----------------------------------------------------------------------------
5059 // FIXME: Like for X86, these should go in their own separate .td file.
5061 // Any instruction that defines a 32-bit result leaves the high half of the
5062 // register. Truncate can be lowered to EXTRACT_SUBREG. CopyFromReg may
5063 // be copying from a truncate. But any other 32-bit operation will zero-extend
5065 // FIXME: X86 also checks for CMOV here. Do we need something similar?
5066 def def32 : PatLeaf<(i32 GPR32:$src), [{
5067 return N->getOpcode() != ISD::TRUNCATE &&
5068 N->getOpcode() != TargetOpcode::EXTRACT_SUBREG &&
5069 N->getOpcode() != ISD::CopyFromReg;
5072 // In the case of a 32-bit def that is known to implicitly zero-extend,
5073 // we can use a SUBREG_TO_REG.
5074 def : Pat<(i64 (zext def32:$src)), (SUBREG_TO_REG (i64 0), GPR32:$src, sub_32)>;
5076 // For an anyext, we don't care what the high bits are, so we can perform an
5077 // INSERT_SUBREF into an IMPLICIT_DEF.
5078 def : Pat<(i64 (anyext GPR32:$src)),
5079 (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32)>;
5081 // When we need to explicitly zero-extend, we use an unsigned bitfield move
5082 // instruction (UBFM) on the enclosing super-reg.
5083 def : Pat<(i64 (zext GPR32:$src)),
5084 (UBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32), 0, 31)>;
5086 // To sign extend, we use a signed bitfield move instruction (SBFM) on the
5087 // containing super-reg.
5088 def : Pat<(i64 (sext GPR32:$src)),
5089 (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32), 0, 31)>;
5090 def : Pat<(i64 (sext_inreg GPR64:$src, i32)), (SBFMXri GPR64:$src, 0, 31)>;
5091 def : Pat<(i64 (sext_inreg GPR64:$src, i16)), (SBFMXri GPR64:$src, 0, 15)>;
5092 def : Pat<(i64 (sext_inreg GPR64:$src, i8)), (SBFMXri GPR64:$src, 0, 7)>;
5093 def : Pat<(i64 (sext_inreg GPR64:$src, i1)), (SBFMXri GPR64:$src, 0, 0)>;
5094 def : Pat<(i32 (sext_inreg GPR32:$src, i16)), (SBFMWri GPR32:$src, 0, 15)>;
5095 def : Pat<(i32 (sext_inreg GPR32:$src, i8)), (SBFMWri GPR32:$src, 0, 7)>;
5096 def : Pat<(i32 (sext_inreg GPR32:$src, i1)), (SBFMWri GPR32:$src, 0, 0)>;
5098 def : Pat<(shl (sext_inreg GPR32:$Rn, i8), (i64 imm0_31:$imm)),
5099 (SBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
5100 (i64 (i32shift_sext_i8 imm0_31:$imm)))>;
5101 def : Pat<(shl (sext_inreg GPR64:$Rn, i8), (i64 imm0_63:$imm)),
5102 (SBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
5103 (i64 (i64shift_sext_i8 imm0_63:$imm)))>;
5105 def : Pat<(shl (sext_inreg GPR32:$Rn, i16), (i64 imm0_31:$imm)),
5106 (SBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
5107 (i64 (i32shift_sext_i16 imm0_31:$imm)))>;
5108 def : Pat<(shl (sext_inreg GPR64:$Rn, i16), (i64 imm0_63:$imm)),
5109 (SBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
5110 (i64 (i64shift_sext_i16 imm0_63:$imm)))>;
5112 def : Pat<(shl (i64 (sext GPR32:$Rn)), (i64 imm0_63:$imm)),
5113 (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$Rn, sub_32),
5114 (i64 (i64shift_a imm0_63:$imm)),
5115 (i64 (i64shift_sext_i32 imm0_63:$imm)))>;
5117 // sra patterns have an AddedComplexity of 10, so make sure we have a higher
5118 // AddedComplexity for the following patterns since we want to match sext + sra
5119 // patterns before we attempt to match a single sra node.
5120 let AddedComplexity = 20 in {
5121 // We support all sext + sra combinations which preserve at least one bit of the
5122 // original value which is to be sign extended. E.g. we support shifts up to
5124 def : Pat<(sra (sext_inreg GPR32:$Rn, i8), (i64 imm0_7:$imm)),
5125 (SBFMWri GPR32:$Rn, (i64 imm0_7:$imm), 7)>;
5126 def : Pat<(sra (sext_inreg GPR64:$Rn, i8), (i64 imm0_7:$imm)),
5127 (SBFMXri GPR64:$Rn, (i64 imm0_7:$imm), 7)>;
5129 def : Pat<(sra (sext_inreg GPR32:$Rn, i16), (i64 imm0_15:$imm)),
5130 (SBFMWri GPR32:$Rn, (i64 imm0_15:$imm), 15)>;
5131 def : Pat<(sra (sext_inreg GPR64:$Rn, i16), (i64 imm0_15:$imm)),
5132 (SBFMXri GPR64:$Rn, (i64 imm0_15:$imm), 15)>;
5134 def : Pat<(sra (i64 (sext GPR32:$Rn)), (i64 imm0_31:$imm)),
5135 (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$Rn, sub_32),
5136 (i64 imm0_31:$imm), 31)>;
5137 } // AddedComplexity = 20
5139 // To truncate, we can simply extract from a subregister.
5140 def : Pat<(i32 (trunc GPR64sp:$src)),
5141 (i32 (EXTRACT_SUBREG GPR64sp:$src, sub_32))>;
5143 // __builtin_trap() uses the BRK instruction on AArch64.
5144 def : Pat<(trap), (BRK 1)>;
5146 // Conversions within AdvSIMD types in the same register size are free.
5147 // But because we need a consistent lane ordering, in big endian many
5148 // conversions require one or more REV instructions.
5150 // Consider a simple memory load followed by a bitconvert then a store.
5152 // v1 = BITCAST v2i32 v0 to v4i16
5155 // In big endian mode every memory access has an implicit byte swap. LDR and
5156 // STR do a 64-bit byte swap, whereas LD1/ST1 do a byte swap per lane - that
5157 // is, they treat the vector as a sequence of elements to be byte-swapped.
5158 // The two pairs of instructions are fundamentally incompatible. We've decided
5159 // to use LD1/ST1 only to simplify compiler implementation.
5161 // LD1/ST1 perform the equivalent of a sequence of LDR/STR + REV. This makes
5162 // the original code sequence:
5164 // v1 = REV v2i32 (implicit)
5165 // v2 = BITCAST v2i32 v1 to v4i16
5166 // v3 = REV v4i16 v2 (implicit)
5169 // But this is now broken - the value stored is different to the value loaded
5170 // due to lane reordering. To fix this, on every BITCAST we must perform two
5173 // v1 = REV v2i32 (implicit)
5175 // v3 = BITCAST v2i32 v2 to v4i16
5177 // v5 = REV v4i16 v4 (implicit)
5180 // This means an extra two instructions, but actually in most cases the two REV
5181 // instructions can be combined into one. For example:
5182 // (REV64_2s (REV64_4h X)) === (REV32_4h X)
5184 // There is also no 128-bit REV instruction. This must be synthesized with an
5187 // Most bitconverts require some sort of conversion. The only exceptions are:
5188 // a) Identity conversions - vNfX <-> vNiX
5189 // b) Single-lane-to-scalar - v1fX <-> fX or v1iX <-> iX
5192 // Natural vector casts (64 bit)
5193 def : Pat<(v8i8 (AArch64NvCast (v2i32 FPR64:$src))), (v8i8 FPR64:$src)>;
5194 def : Pat<(v4i16 (AArch64NvCast (v2i32 FPR64:$src))), (v4i16 FPR64:$src)>;
5195 def : Pat<(v4f16 (AArch64NvCast (v2i32 FPR64:$src))), (v4f16 FPR64:$src)>;
5196 def : Pat<(v2i32 (AArch64NvCast (v2i32 FPR64:$src))), (v2i32 FPR64:$src)>;
5197 def : Pat<(v2f32 (AArch64NvCast (v2i32 FPR64:$src))), (v2f32 FPR64:$src)>;
5198 def : Pat<(v1i64 (AArch64NvCast (v2i32 FPR64:$src))), (v1i64 FPR64:$src)>;
5200 def : Pat<(v8i8 (AArch64NvCast (v4i16 FPR64:$src))), (v8i8 FPR64:$src)>;
5201 def : Pat<(v4i16 (AArch64NvCast (v4i16 FPR64:$src))), (v4i16 FPR64:$src)>;
5202 def : Pat<(v4f16 (AArch64NvCast (v4i16 FPR64:$src))), (v4f16 FPR64:$src)>;
5203 def : Pat<(v2i32 (AArch64NvCast (v4i16 FPR64:$src))), (v2i32 FPR64:$src)>;
5204 def : Pat<(v1i64 (AArch64NvCast (v4i16 FPR64:$src))), (v1i64 FPR64:$src)>;
5206 def : Pat<(v8i8 (AArch64NvCast (v8i8 FPR64:$src))), (v8i8 FPR64:$src)>;
5207 def : Pat<(v4i16 (AArch64NvCast (v8i8 FPR64:$src))), (v4i16 FPR64:$src)>;
5208 def : Pat<(v4f16 (AArch64NvCast (v8i8 FPR64:$src))), (v4f16 FPR64:$src)>;
5209 def : Pat<(v2i32 (AArch64NvCast (v8i8 FPR64:$src))), (v2i32 FPR64:$src)>;
5210 def : Pat<(v1i64 (AArch64NvCast (v8i8 FPR64:$src))), (v1i64 FPR64:$src)>;
5212 def : Pat<(v8i8 (AArch64NvCast (f64 FPR64:$src))), (v8i8 FPR64:$src)>;
5213 def : Pat<(v4i16 (AArch64NvCast (f64 FPR64:$src))), (v4i16 FPR64:$src)>;
5214 def : Pat<(v4f16 (AArch64NvCast (f64 FPR64:$src))), (v4f16 FPR64:$src)>;
5215 def : Pat<(v2i32 (AArch64NvCast (f64 FPR64:$src))), (v2i32 FPR64:$src)>;
5216 def : Pat<(v2f32 (AArch64NvCast (f64 FPR64:$src))), (v2f32 FPR64:$src)>;
5217 def : Pat<(v1i64 (AArch64NvCast (f64 FPR64:$src))), (v1i64 FPR64:$src)>;
5218 def : Pat<(v1f64 (AArch64NvCast (f64 FPR64:$src))), (v1f64 FPR64:$src)>;
5220 def : Pat<(v8i8 (AArch64NvCast (v2f32 FPR64:$src))), (v8i8 FPR64:$src)>;
5221 def : Pat<(v4i16 (AArch64NvCast (v2f32 FPR64:$src))), (v4i16 FPR64:$src)>;
5222 def : Pat<(v2i32 (AArch64NvCast (v2f32 FPR64:$src))), (v2i32 FPR64:$src)>;
5223 def : Pat<(v2f32 (AArch64NvCast (v2f32 FPR64:$src))), (v2f32 FPR64:$src)>;
5224 def : Pat<(v1i64 (AArch64NvCast (v2f32 FPR64:$src))), (v1i64 FPR64:$src)>;
5226 // Natural vector casts (128 bit)
5227 def : Pat<(v16i8 (AArch64NvCast (v4i32 FPR128:$src))), (v16i8 FPR128:$src)>;
5228 def : Pat<(v8i16 (AArch64NvCast (v4i32 FPR128:$src))), (v8i16 FPR128:$src)>;
5229 def : Pat<(v8f16 (AArch64NvCast (v4i32 FPR128:$src))), (v8f16 FPR128:$src)>;
5230 def : Pat<(v4i32 (AArch64NvCast (v4i32 FPR128:$src))), (v4i32 FPR128:$src)>;
5231 def : Pat<(v4f32 (AArch64NvCast (v4i32 FPR128:$src))), (v4f32 FPR128:$src)>;
5232 def : Pat<(v2i64 (AArch64NvCast (v4i32 FPR128:$src))), (v2i64 FPR128:$src)>;
5234 def : Pat<(v16i8 (AArch64NvCast (v8i16 FPR128:$src))), (v16i8 FPR128:$src)>;
5235 def : Pat<(v8i16 (AArch64NvCast (v8i16 FPR128:$src))), (v8i16 FPR128:$src)>;
5236 def : Pat<(v8f16 (AArch64NvCast (v8i16 FPR128:$src))), (v8f16 FPR128:$src)>;
5237 def : Pat<(v4i32 (AArch64NvCast (v8i16 FPR128:$src))), (v4i32 FPR128:$src)>;
5238 def : Pat<(v2i64 (AArch64NvCast (v8i16 FPR128:$src))), (v2i64 FPR128:$src)>;
5240 def : Pat<(v16i8 (AArch64NvCast (v16i8 FPR128:$src))), (v16i8 FPR128:$src)>;
5241 def : Pat<(v8i16 (AArch64NvCast (v16i8 FPR128:$src))), (v8i16 FPR128:$src)>;
5242 def : Pat<(v8f16 (AArch64NvCast (v16i8 FPR128:$src))), (v8f16 FPR128:$src)>;
5243 def : Pat<(v4i32 (AArch64NvCast (v16i8 FPR128:$src))), (v4i32 FPR128:$src)>;
5244 def : Pat<(v2i64 (AArch64NvCast (v16i8 FPR128:$src))), (v2i64 FPR128:$src)>;
5246 def : Pat<(v16i8 (AArch64NvCast (v2i64 FPR128:$src))), (v16i8 FPR128:$src)>;
5247 def : Pat<(v8i16 (AArch64NvCast (v2i64 FPR128:$src))), (v8i16 FPR128:$src)>;
5248 def : Pat<(v8f16 (AArch64NvCast (v2i64 FPR128:$src))), (v8f16 FPR128:$src)>;
5249 def : Pat<(v4i32 (AArch64NvCast (v2i64 FPR128:$src))), (v4i32 FPR128:$src)>;
5250 def : Pat<(v2i64 (AArch64NvCast (v2i64 FPR128:$src))), (v2i64 FPR128:$src)>;
5251 def : Pat<(v4f32 (AArch64NvCast (v2i64 FPR128:$src))), (v4f32 FPR128:$src)>;
5252 def : Pat<(v2f64 (AArch64NvCast (v2i64 FPR128:$src))), (v2f64 FPR128:$src)>;
5254 def : Pat<(v16i8 (AArch64NvCast (v4f32 FPR128:$src))), (v16i8 FPR128:$src)>;
5255 def : Pat<(v8i16 (AArch64NvCast (v4f32 FPR128:$src))), (v8i16 FPR128:$src)>;
5256 def : Pat<(v4i32 (AArch64NvCast (v4f32 FPR128:$src))), (v4i32 FPR128:$src)>;
5257 def : Pat<(v4f32 (AArch64NvCast (v4f32 FPR128:$src))), (v4f32 FPR128:$src)>;
5258 def : Pat<(v2i64 (AArch64NvCast (v4f32 FPR128:$src))), (v2i64 FPR128:$src)>;
5260 def : Pat<(v16i8 (AArch64NvCast (v2f64 FPR128:$src))), (v16i8 FPR128:$src)>;
5261 def : Pat<(v8i16 (AArch64NvCast (v2f64 FPR128:$src))), (v8i16 FPR128:$src)>;
5262 def : Pat<(v4i32 (AArch64NvCast (v2f64 FPR128:$src))), (v4i32 FPR128:$src)>;
5263 def : Pat<(v2i64 (AArch64NvCast (v2f64 FPR128:$src))), (v2i64 FPR128:$src)>;
5264 def : Pat<(v2f64 (AArch64NvCast (v2f64 FPR128:$src))), (v2f64 FPR128:$src)>;
5266 let Predicates = [IsLE] in {
5267 def : Pat<(v8i8 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5268 def : Pat<(v4i16 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5269 def : Pat<(v2i32 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5270 def : Pat<(v4f16 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5271 def : Pat<(v2f32 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5273 def : Pat<(i64 (bitconvert (v8i8 V64:$Vn))),
5274 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5275 def : Pat<(i64 (bitconvert (v4i16 V64:$Vn))),
5276 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5277 def : Pat<(i64 (bitconvert (v2i32 V64:$Vn))),
5278 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5279 def : Pat<(i64 (bitconvert (v4f16 V64:$Vn))),
5280 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5281 def : Pat<(i64 (bitconvert (v2f32 V64:$Vn))),
5282 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5283 def : Pat<(i64 (bitconvert (v1f64 V64:$Vn))),
5284 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5286 let Predicates = [IsBE] in {
5287 def : Pat<(v8i8 (bitconvert GPR64:$Xn)),
5288 (REV64v8i8 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
5289 def : Pat<(v4i16 (bitconvert GPR64:$Xn)),
5290 (REV64v4i16 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
5291 def : Pat<(v2i32 (bitconvert GPR64:$Xn)),
5292 (REV64v2i32 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
5293 def : Pat<(v4f16 (bitconvert GPR64:$Xn)),
5294 (REV64v4i16 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
5295 def : Pat<(v2f32 (bitconvert GPR64:$Xn)),
5296 (REV64v2i32 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
5298 def : Pat<(i64 (bitconvert (v8i8 V64:$Vn))),
5299 (REV64v8i8 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
5300 def : Pat<(i64 (bitconvert (v4i16 V64:$Vn))),
5301 (REV64v4i16 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
5302 def : Pat<(i64 (bitconvert (v2i32 V64:$Vn))),
5303 (REV64v2i32 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
5304 def : Pat<(i64 (bitconvert (v4f16 V64:$Vn))),
5305 (REV64v4i16 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
5306 def : Pat<(i64 (bitconvert (v2f32 V64:$Vn))),
5307 (REV64v2i32 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
5309 def : Pat<(v1i64 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5310 def : Pat<(v1f64 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5311 def : Pat<(i64 (bitconvert (v1i64 V64:$Vn))),
5312 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5313 def : Pat<(v1i64 (scalar_to_vector GPR64:$Xn)),
5314 (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5315 def : Pat<(v1f64 (scalar_to_vector GPR64:$Xn)),
5316 (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5317 def : Pat<(v1f64 (scalar_to_vector (f64 FPR64:$Xn))), (v1f64 FPR64:$Xn)>;
5319 def : Pat<(f32 (bitconvert (i32 GPR32:$Xn))),
5320 (COPY_TO_REGCLASS GPR32:$Xn, FPR32)>;
5321 def : Pat<(i32 (bitconvert (f32 FPR32:$Xn))),
5322 (COPY_TO_REGCLASS FPR32:$Xn, GPR32)>;
5323 def : Pat<(f64 (bitconvert (i64 GPR64:$Xn))),
5324 (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5325 def : Pat<(i64 (bitconvert (f64 FPR64:$Xn))),
5326 (COPY_TO_REGCLASS FPR64:$Xn, GPR64)>;
5327 def : Pat<(i64 (bitconvert (v1f64 V64:$Vn))),
5328 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5330 let Predicates = [IsLE] in {
5331 def : Pat<(v1i64 (bitconvert (v2i32 FPR64:$src))), (v1i64 FPR64:$src)>;
5332 def : Pat<(v1i64 (bitconvert (v4i16 FPR64:$src))), (v1i64 FPR64:$src)>;
5333 def : Pat<(v1i64 (bitconvert (v8i8 FPR64:$src))), (v1i64 FPR64:$src)>;
5334 def : Pat<(v1i64 (bitconvert (v4f16 FPR64:$src))), (v1i64 FPR64:$src)>;
5335 def : Pat<(v1i64 (bitconvert (v2f32 FPR64:$src))), (v1i64 FPR64:$src)>;
5337 let Predicates = [IsBE] in {
5338 def : Pat<(v1i64 (bitconvert (v2i32 FPR64:$src))),
5339 (v1i64 (REV64v2i32 FPR64:$src))>;
5340 def : Pat<(v1i64 (bitconvert (v4i16 FPR64:$src))),
5341 (v1i64 (REV64v4i16 FPR64:$src))>;
5342 def : Pat<(v1i64 (bitconvert (v8i8 FPR64:$src))),
5343 (v1i64 (REV64v8i8 FPR64:$src))>;
5344 def : Pat<(v1i64 (bitconvert (v4f16 FPR64:$src))),
5345 (v1i64 (REV64v4i16 FPR64:$src))>;
5346 def : Pat<(v1i64 (bitconvert (v2f32 FPR64:$src))),
5347 (v1i64 (REV64v2i32 FPR64:$src))>;
5349 def : Pat<(v1i64 (bitconvert (v1f64 FPR64:$src))), (v1i64 FPR64:$src)>;
5350 def : Pat<(v1i64 (bitconvert (f64 FPR64:$src))), (v1i64 FPR64:$src)>;
5352 let Predicates = [IsLE] in {
5353 def : Pat<(v2i32 (bitconvert (v1i64 FPR64:$src))), (v2i32 FPR64:$src)>;
5354 def : Pat<(v2i32 (bitconvert (v4i16 FPR64:$src))), (v2i32 FPR64:$src)>;
5355 def : Pat<(v2i32 (bitconvert (v8i8 FPR64:$src))), (v2i32 FPR64:$src)>;
5356 def : Pat<(v2i32 (bitconvert (f64 FPR64:$src))), (v2i32 FPR64:$src)>;
5357 def : Pat<(v2i32 (bitconvert (v1f64 FPR64:$src))), (v2i32 FPR64:$src)>;
5358 def : Pat<(v2i32 (bitconvert (v4f16 FPR64:$src))), (v2i32 FPR64:$src)>;
5360 let Predicates = [IsBE] in {
5361 def : Pat<(v2i32 (bitconvert (v1i64 FPR64:$src))),
5362 (v2i32 (REV64v2i32 FPR64:$src))>;
5363 def : Pat<(v2i32 (bitconvert (v4i16 FPR64:$src))),
5364 (v2i32 (REV32v4i16 FPR64:$src))>;
5365 def : Pat<(v2i32 (bitconvert (v8i8 FPR64:$src))),
5366 (v2i32 (REV32v8i8 FPR64:$src))>;
5367 def : Pat<(v2i32 (bitconvert (f64 FPR64:$src))),
5368 (v2i32 (REV64v2i32 FPR64:$src))>;
5369 def : Pat<(v2i32 (bitconvert (v1f64 FPR64:$src))),
5370 (v2i32 (REV64v2i32 FPR64:$src))>;
5371 def : Pat<(v2i32 (bitconvert (v4f16 FPR64:$src))),
5372 (v2i32 (REV64v4i16 FPR64:$src))>;
5374 def : Pat<(v2i32 (bitconvert (v2f32 FPR64:$src))), (v2i32 FPR64:$src)>;
5376 let Predicates = [IsLE] in {
5377 def : Pat<(v4i16 (bitconvert (v1i64 FPR64:$src))), (v4i16 FPR64:$src)>;
5378 def : Pat<(v4i16 (bitconvert (v2i32 FPR64:$src))), (v4i16 FPR64:$src)>;
5379 def : Pat<(v4i16 (bitconvert (v8i8 FPR64:$src))), (v4i16 FPR64:$src)>;
5380 def : Pat<(v4i16 (bitconvert (f64 FPR64:$src))), (v4i16 FPR64:$src)>;
5381 def : Pat<(v4i16 (bitconvert (v4f16 FPR64:$src))), (v4i16 FPR64:$src)>;
5382 def : Pat<(v4i16 (bitconvert (v2f32 FPR64:$src))), (v4i16 FPR64:$src)>;
5383 def : Pat<(v4i16 (bitconvert (v1f64 FPR64:$src))), (v4i16 FPR64:$src)>;
5385 let Predicates = [IsBE] in {
5386 def : Pat<(v4i16 (bitconvert (v1i64 FPR64:$src))),
5387 (v4i16 (REV64v4i16 FPR64:$src))>;
5388 def : Pat<(v4i16 (bitconvert (v2i32 FPR64:$src))),
5389 (v4i16 (REV32v4i16 FPR64:$src))>;
5390 def : Pat<(v4i16 (bitconvert (v8i8 FPR64:$src))),
5391 (v4i16 (REV16v8i8 FPR64:$src))>;
5392 def : Pat<(v4i16 (bitconvert (f64 FPR64:$src))),
5393 (v4i16 (REV64v4i16 FPR64:$src))>;
5394 def : Pat<(v4i16 (bitconvert (v4f16 FPR64:$src))),
5395 (v4i16 (REV32v4i16 FPR64:$src))>;
5396 def : Pat<(v4i16 (bitconvert (v2f32 FPR64:$src))),
5397 (v4i16 (REV32v4i16 FPR64:$src))>;
5398 def : Pat<(v4i16 (bitconvert (v1f64 FPR64:$src))),
5399 (v4i16 (REV64v4i16 FPR64:$src))>;
5402 let Predicates = [IsLE] in {
5403 def : Pat<(v4f16 (bitconvert (v1i64 FPR64:$src))), (v4f16 FPR64:$src)>;
5404 def : Pat<(v4f16 (bitconvert (v2i32 FPR64:$src))), (v4f16 FPR64:$src)>;
5405 def : Pat<(v4f16 (bitconvert (v4i16 FPR64:$src))), (v4f16 FPR64:$src)>;
5406 def : Pat<(v4f16 (bitconvert (v8i8 FPR64:$src))), (v4f16 FPR64:$src)>;
5407 def : Pat<(v4f16 (bitconvert (f64 FPR64:$src))), (v4f16 FPR64:$src)>;
5408 def : Pat<(v4f16 (bitconvert (v2f32 FPR64:$src))), (v4f16 FPR64:$src)>;
5409 def : Pat<(v4f16 (bitconvert (v1f64 FPR64:$src))), (v4f16 FPR64:$src)>;
5411 let Predicates = [IsBE] in {
5412 def : Pat<(v4f16 (bitconvert (v1i64 FPR64:$src))),
5413 (v4f16 (REV64v4i16 FPR64:$src))>;
5414 def : Pat<(v4f16 (bitconvert (v2i32 FPR64:$src))),
5415 (v4f16 (REV64v4i16 FPR64:$src))>;
5416 def : Pat<(v4f16 (bitconvert (v4i16 FPR64:$src))),
5417 (v4f16 (REV64v4i16 FPR64:$src))>;
5418 def : Pat<(v4f16 (bitconvert (v8i8 FPR64:$src))),
5419 (v4f16 (REV16v8i8 FPR64:$src))>;
5420 def : Pat<(v4f16 (bitconvert (f64 FPR64:$src))),
5421 (v4f16 (REV64v4i16 FPR64:$src))>;
5422 def : Pat<(v4f16 (bitconvert (v2f32 FPR64:$src))),
5423 (v4f16 (REV64v4i16 FPR64:$src))>;
5424 def : Pat<(v4f16 (bitconvert (v1f64 FPR64:$src))),
5425 (v4f16 (REV64v4i16 FPR64:$src))>;
5430 let Predicates = [IsLE] in {
5431 def : Pat<(v8i8 (bitconvert (v1i64 FPR64:$src))), (v8i8 FPR64:$src)>;
5432 def : Pat<(v8i8 (bitconvert (v2i32 FPR64:$src))), (v8i8 FPR64:$src)>;
5433 def : Pat<(v8i8 (bitconvert (v4i16 FPR64:$src))), (v8i8 FPR64:$src)>;
5434 def : Pat<(v8i8 (bitconvert (f64 FPR64:$src))), (v8i8 FPR64:$src)>;
5435 def : Pat<(v8i8 (bitconvert (v2f32 FPR64:$src))), (v8i8 FPR64:$src)>;
5436 def : Pat<(v8i8 (bitconvert (v1f64 FPR64:$src))), (v8i8 FPR64:$src)>;
5437 def : Pat<(v8i8 (bitconvert (v4f16 FPR64:$src))), (v8i8 FPR64:$src)>;
5439 let Predicates = [IsBE] in {
5440 def : Pat<(v8i8 (bitconvert (v1i64 FPR64:$src))),
5441 (v8i8 (REV64v8i8 FPR64:$src))>;
5442 def : Pat<(v8i8 (bitconvert (v2i32 FPR64:$src))),
5443 (v8i8 (REV32v8i8 FPR64:$src))>;
5444 def : Pat<(v8i8 (bitconvert (v4i16 FPR64:$src))),
5445 (v8i8 (REV16v8i8 FPR64:$src))>;
5446 def : Pat<(v8i8 (bitconvert (f64 FPR64:$src))),
5447 (v8i8 (REV64v8i8 FPR64:$src))>;
5448 def : Pat<(v8i8 (bitconvert (v2f32 FPR64:$src))),
5449 (v8i8 (REV32v8i8 FPR64:$src))>;
5450 def : Pat<(v8i8 (bitconvert (v1f64 FPR64:$src))),
5451 (v8i8 (REV64v8i8 FPR64:$src))>;
5452 def : Pat<(v8i8 (bitconvert (v4f16 FPR64:$src))),
5453 (v8i8 (REV16v8i8 FPR64:$src))>;
5456 let Predicates = [IsLE] in {
5457 def : Pat<(f64 (bitconvert (v2i32 FPR64:$src))), (f64 FPR64:$src)>;
5458 def : Pat<(f64 (bitconvert (v4i16 FPR64:$src))), (f64 FPR64:$src)>;
5459 def : Pat<(f64 (bitconvert (v2f32 FPR64:$src))), (f64 FPR64:$src)>;
5460 def : Pat<(f64 (bitconvert (v8i8 FPR64:$src))), (f64 FPR64:$src)>;
5461 def : Pat<(f64 (bitconvert (v4f16 FPR64:$src))), (f64 FPR64:$src)>;
5463 let Predicates = [IsBE] in {
5464 def : Pat<(f64 (bitconvert (v2i32 FPR64:$src))),
5465 (f64 (REV64v2i32 FPR64:$src))>;
5466 def : Pat<(f64 (bitconvert (v4i16 FPR64:$src))),
5467 (f64 (REV64v4i16 FPR64:$src))>;
5468 def : Pat<(f64 (bitconvert (v2f32 FPR64:$src))),
5469 (f64 (REV64v2i32 FPR64:$src))>;
5470 def : Pat<(f64 (bitconvert (v8i8 FPR64:$src))),
5471 (f64 (REV64v8i8 FPR64:$src))>;
5472 def : Pat<(f64 (bitconvert (v4f16 FPR64:$src))),
5473 (f64 (REV64v4i16 FPR64:$src))>;
5475 def : Pat<(f64 (bitconvert (v1i64 FPR64:$src))), (f64 FPR64:$src)>;
5476 def : Pat<(f64 (bitconvert (v1f64 FPR64:$src))), (f64 FPR64:$src)>;
5478 let Predicates = [IsLE] in {
5479 def : Pat<(v1f64 (bitconvert (v2i32 FPR64:$src))), (v1f64 FPR64:$src)>;
5480 def : Pat<(v1f64 (bitconvert (v4i16 FPR64:$src))), (v1f64 FPR64:$src)>;
5481 def : Pat<(v1f64 (bitconvert (v8i8 FPR64:$src))), (v1f64 FPR64:$src)>;
5482 def : Pat<(v1f64 (bitconvert (v2f32 FPR64:$src))), (v1f64 FPR64:$src)>;
5483 def : Pat<(v1f64 (bitconvert (v4f16 FPR64:$src))), (v1f64 FPR64:$src)>;
5485 let Predicates = [IsBE] in {
5486 def : Pat<(v1f64 (bitconvert (v2i32 FPR64:$src))),
5487 (v1f64 (REV64v2i32 FPR64:$src))>;
5488 def : Pat<(v1f64 (bitconvert (v4i16 FPR64:$src))),
5489 (v1f64 (REV64v4i16 FPR64:$src))>;
5490 def : Pat<(v1f64 (bitconvert (v8i8 FPR64:$src))),
5491 (v1f64 (REV64v8i8 FPR64:$src))>;
5492 def : Pat<(v1f64 (bitconvert (v2f32 FPR64:$src))),
5493 (v1f64 (REV64v2i32 FPR64:$src))>;
5494 def : Pat<(v1f64 (bitconvert (v4f16 FPR64:$src))),
5495 (v1f64 (REV64v4i16 FPR64:$src))>;
5497 def : Pat<(v1f64 (bitconvert (v1i64 FPR64:$src))), (v1f64 FPR64:$src)>;
5498 def : Pat<(v1f64 (bitconvert (f64 FPR64:$src))), (v1f64 FPR64:$src)>;
5500 let Predicates = [IsLE] in {
5501 def : Pat<(v2f32 (bitconvert (v1i64 FPR64:$src))), (v2f32 FPR64:$src)>;
5502 def : Pat<(v2f32 (bitconvert (v4i16 FPR64:$src))), (v2f32 FPR64:$src)>;
5503 def : Pat<(v2f32 (bitconvert (v8i8 FPR64:$src))), (v2f32 FPR64:$src)>;
5504 def : Pat<(v2f32 (bitconvert (v1f64 FPR64:$src))), (v2f32 FPR64:$src)>;
5505 def : Pat<(v2f32 (bitconvert (f64 FPR64:$src))), (v2f32 FPR64:$src)>;
5506 def : Pat<(v2f32 (bitconvert (v4f16 FPR64:$src))), (v2f32 FPR64:$src)>;
5508 let Predicates = [IsBE] in {
5509 def : Pat<(v2f32 (bitconvert (v1i64 FPR64:$src))),
5510 (v2f32 (REV64v2i32 FPR64:$src))>;
5511 def : Pat<(v2f32 (bitconvert (v4i16 FPR64:$src))),
5512 (v2f32 (REV32v4i16 FPR64:$src))>;
5513 def : Pat<(v2f32 (bitconvert (v8i8 FPR64:$src))),
5514 (v2f32 (REV32v8i8 FPR64:$src))>;
5515 def : Pat<(v2f32 (bitconvert (v1f64 FPR64:$src))),
5516 (v2f32 (REV64v2i32 FPR64:$src))>;
5517 def : Pat<(v2f32 (bitconvert (f64 FPR64:$src))),
5518 (v2f32 (REV64v2i32 FPR64:$src))>;
5519 def : Pat<(v2f32 (bitconvert (v4f16 FPR64:$src))),
5520 (v2f32 (REV64v4i16 FPR64:$src))>;
5522 def : Pat<(v2f32 (bitconvert (v2i32 FPR64:$src))), (v2f32 FPR64:$src)>;
5524 let Predicates = [IsLE] in {
5525 def : Pat<(f128 (bitconvert (v2i64 FPR128:$src))), (f128 FPR128:$src)>;
5526 def : Pat<(f128 (bitconvert (v4i32 FPR128:$src))), (f128 FPR128:$src)>;
5527 def : Pat<(f128 (bitconvert (v8i16 FPR128:$src))), (f128 FPR128:$src)>;
5528 def : Pat<(f128 (bitconvert (v2f64 FPR128:$src))), (f128 FPR128:$src)>;
5529 def : Pat<(f128 (bitconvert (v4f32 FPR128:$src))), (f128 FPR128:$src)>;
5530 def : Pat<(f128 (bitconvert (v8f16 FPR128:$src))), (f128 FPR128:$src)>;
5531 def : Pat<(f128 (bitconvert (v16i8 FPR128:$src))), (f128 FPR128:$src)>;
5533 let Predicates = [IsBE] in {
5534 def : Pat<(f128 (bitconvert (v2i64 FPR128:$src))),
5535 (f128 (EXTv16i8 FPR128:$src, FPR128:$src, (i32 8)))>;
5536 def : Pat<(f128 (bitconvert (v4i32 FPR128:$src))),
5537 (f128 (EXTv16i8 (REV64v4i32 FPR128:$src),
5538 (REV64v4i32 FPR128:$src), (i32 8)))>;
5539 def : Pat<(f128 (bitconvert (v8i16 FPR128:$src))),
5540 (f128 (EXTv16i8 (REV64v8i16 FPR128:$src),
5541 (REV64v8i16 FPR128:$src), (i32 8)))>;
5542 def : Pat<(f128 (bitconvert (v8f16 FPR128:$src))),
5543 (f128 (EXTv16i8 (REV64v8i16 FPR128:$src),
5544 (REV64v8i16 FPR128:$src), (i32 8)))>;
5545 def : Pat<(f128 (bitconvert (v2f64 FPR128:$src))),
5546 (f128 (EXTv16i8 FPR128:$src, FPR128:$src, (i32 8)))>;
5547 def : Pat<(f128 (bitconvert (v4f32 FPR128:$src))),
5548 (f128 (EXTv16i8 (REV64v4i32 FPR128:$src),
5549 (REV64v4i32 FPR128:$src), (i32 8)))>;
5550 def : Pat<(f128 (bitconvert (v16i8 FPR128:$src))),
5551 (f128 (EXTv16i8 (REV64v16i8 FPR128:$src),
5552 (REV64v16i8 FPR128:$src), (i32 8)))>;
5555 let Predicates = [IsLE] in {
5556 def : Pat<(v2f64 (bitconvert (f128 FPR128:$src))), (v2f64 FPR128:$src)>;
5557 def : Pat<(v2f64 (bitconvert (v4i32 FPR128:$src))), (v2f64 FPR128:$src)>;
5558 def : Pat<(v2f64 (bitconvert (v8i16 FPR128:$src))), (v2f64 FPR128:$src)>;
5559 def : Pat<(v2f64 (bitconvert (v8f16 FPR128:$src))), (v2f64 FPR128:$src)>;
5560 def : Pat<(v2f64 (bitconvert (v16i8 FPR128:$src))), (v2f64 FPR128:$src)>;
5561 def : Pat<(v2f64 (bitconvert (v4f32 FPR128:$src))), (v2f64 FPR128:$src)>;
5563 let Predicates = [IsBE] in {
5564 def : Pat<(v2f64 (bitconvert (f128 FPR128:$src))),
5565 (v2f64 (EXTv16i8 FPR128:$src,
5566 FPR128:$src, (i32 8)))>;
5567 def : Pat<(v2f64 (bitconvert (v4i32 FPR128:$src))),
5568 (v2f64 (REV64v4i32 FPR128:$src))>;
5569 def : Pat<(v2f64 (bitconvert (v8i16 FPR128:$src))),
5570 (v2f64 (REV64v8i16 FPR128:$src))>;
5571 def : Pat<(v2f64 (bitconvert (v8f16 FPR128:$src))),
5572 (v2f64 (REV64v8i16 FPR128:$src))>;
5573 def : Pat<(v2f64 (bitconvert (v16i8 FPR128:$src))),
5574 (v2f64 (REV64v16i8 FPR128:$src))>;
5575 def : Pat<(v2f64 (bitconvert (v4f32 FPR128:$src))),
5576 (v2f64 (REV64v4i32 FPR128:$src))>;
5578 def : Pat<(v2f64 (bitconvert (v2i64 FPR128:$src))), (v2f64 FPR128:$src)>;
5580 let Predicates = [IsLE] in {
5581 def : Pat<(v4f32 (bitconvert (f128 FPR128:$src))), (v4f32 FPR128:$src)>;
5582 def : Pat<(v4f32 (bitconvert (v8i16 FPR128:$src))), (v4f32 FPR128:$src)>;
5583 def : Pat<(v4f32 (bitconvert (v8f16 FPR128:$src))), (v4f32 FPR128:$src)>;
5584 def : Pat<(v4f32 (bitconvert (v16i8 FPR128:$src))), (v4f32 FPR128:$src)>;
5585 def : Pat<(v4f32 (bitconvert (v2i64 FPR128:$src))), (v4f32 FPR128:$src)>;
5586 def : Pat<(v4f32 (bitconvert (v2f64 FPR128:$src))), (v4f32 FPR128:$src)>;
5588 let Predicates = [IsBE] in {
5589 def : Pat<(v4f32 (bitconvert (f128 FPR128:$src))),
5590 (v4f32 (EXTv16i8 (REV64v4i32 FPR128:$src),
5591 (REV64v4i32 FPR128:$src), (i32 8)))>;
5592 def : Pat<(v4f32 (bitconvert (v8i16 FPR128:$src))),
5593 (v4f32 (REV32v8i16 FPR128:$src))>;
5594 def : Pat<(v4f32 (bitconvert (v8f16 FPR128:$src))),
5595 (v4f32 (REV32v8i16 FPR128:$src))>;
5596 def : Pat<(v4f32 (bitconvert (v16i8 FPR128:$src))),
5597 (v4f32 (REV32v16i8 FPR128:$src))>;
5598 def : Pat<(v4f32 (bitconvert (v2i64 FPR128:$src))),
5599 (v4f32 (REV64v4i32 FPR128:$src))>;
5600 def : Pat<(v4f32 (bitconvert (v2f64 FPR128:$src))),
5601 (v4f32 (REV64v4i32 FPR128:$src))>;
5603 def : Pat<(v4f32 (bitconvert (v4i32 FPR128:$src))), (v4f32 FPR128:$src)>;
5605 let Predicates = [IsLE] in {
5606 def : Pat<(v2i64 (bitconvert (f128 FPR128:$src))), (v2i64 FPR128:$src)>;
5607 def : Pat<(v2i64 (bitconvert (v4i32 FPR128:$src))), (v2i64 FPR128:$src)>;
5608 def : Pat<(v2i64 (bitconvert (v8i16 FPR128:$src))), (v2i64 FPR128:$src)>;
5609 def : Pat<(v2i64 (bitconvert (v16i8 FPR128:$src))), (v2i64 FPR128:$src)>;
5610 def : Pat<(v2i64 (bitconvert (v4f32 FPR128:$src))), (v2i64 FPR128:$src)>;
5611 def : Pat<(v2i64 (bitconvert (v8f16 FPR128:$src))), (v2i64 FPR128:$src)>;
5613 let Predicates = [IsBE] in {
5614 def : Pat<(v2i64 (bitconvert (f128 FPR128:$src))),
5615 (v2i64 (EXTv16i8 FPR128:$src,
5616 FPR128:$src, (i32 8)))>;
5617 def : Pat<(v2i64 (bitconvert (v4i32 FPR128:$src))),
5618 (v2i64 (REV64v4i32 FPR128:$src))>;
5619 def : Pat<(v2i64 (bitconvert (v8i16 FPR128:$src))),
5620 (v2i64 (REV64v8i16 FPR128:$src))>;
5621 def : Pat<(v2i64 (bitconvert (v16i8 FPR128:$src))),
5622 (v2i64 (REV64v16i8 FPR128:$src))>;
5623 def : Pat<(v2i64 (bitconvert (v4f32 FPR128:$src))),
5624 (v2i64 (REV64v4i32 FPR128:$src))>;
5625 def : Pat<(v2i64 (bitconvert (v8f16 FPR128:$src))),
5626 (v2i64 (REV64v8i16 FPR128:$src))>;
5628 def : Pat<(v2i64 (bitconvert (v2f64 FPR128:$src))), (v2i64 FPR128:$src)>;
5630 let Predicates = [IsLE] in {
5631 def : Pat<(v4i32 (bitconvert (f128 FPR128:$src))), (v4i32 FPR128:$src)>;
5632 def : Pat<(v4i32 (bitconvert (v2i64 FPR128:$src))), (v4i32 FPR128:$src)>;
5633 def : Pat<(v4i32 (bitconvert (v8i16 FPR128:$src))), (v4i32 FPR128:$src)>;
5634 def : Pat<(v4i32 (bitconvert (v16i8 FPR128:$src))), (v4i32 FPR128:$src)>;
5635 def : Pat<(v4i32 (bitconvert (v2f64 FPR128:$src))), (v4i32 FPR128:$src)>;
5636 def : Pat<(v4i32 (bitconvert (v8f16 FPR128:$src))), (v4i32 FPR128:$src)>;
5638 let Predicates = [IsBE] in {
5639 def : Pat<(v4i32 (bitconvert (f128 FPR128:$src))),
5640 (v4i32 (EXTv16i8 (REV64v4i32 FPR128:$src),
5641 (REV64v4i32 FPR128:$src),
5643 def : Pat<(v4i32 (bitconvert (v2i64 FPR128:$src))),
5644 (v4i32 (REV64v4i32 FPR128:$src))>;
5645 def : Pat<(v4i32 (bitconvert (v8i16 FPR128:$src))),
5646 (v4i32 (REV32v8i16 FPR128:$src))>;
5647 def : Pat<(v4i32 (bitconvert (v16i8 FPR128:$src))),
5648 (v4i32 (REV32v16i8 FPR128:$src))>;
5649 def : Pat<(v4i32 (bitconvert (v2f64 FPR128:$src))),
5650 (v4i32 (REV64v4i32 FPR128:$src))>;
5651 def : Pat<(v4i32 (bitconvert (v8f16 FPR128:$src))),
5652 (v4i32 (REV32v8i16 FPR128:$src))>;
5654 def : Pat<(v4i32 (bitconvert (v4f32 FPR128:$src))), (v4i32 FPR128:$src)>;
5656 let Predicates = [IsLE] in {
5657 def : Pat<(v8i16 (bitconvert (f128 FPR128:$src))), (v8i16 FPR128:$src)>;
5658 def : Pat<(v8i16 (bitconvert (v2i64 FPR128:$src))), (v8i16 FPR128:$src)>;
5659 def : Pat<(v8i16 (bitconvert (v4i32 FPR128:$src))), (v8i16 FPR128:$src)>;
5660 def : Pat<(v8i16 (bitconvert (v16i8 FPR128:$src))), (v8i16 FPR128:$src)>;
5661 def : Pat<(v8i16 (bitconvert (v2f64 FPR128:$src))), (v8i16 FPR128:$src)>;
5662 def : Pat<(v8i16 (bitconvert (v4f32 FPR128:$src))), (v8i16 FPR128:$src)>;
5663 def : Pat<(v8i16 (bitconvert (v8f16 FPR128:$src))), (v8i16 FPR128:$src)>;
5665 let Predicates = [IsBE] in {
5666 def : Pat<(v8i16 (bitconvert (f128 FPR128:$src))),
5667 (v8i16 (EXTv16i8 (REV64v8i16 FPR128:$src),
5668 (REV64v8i16 FPR128:$src),
5670 def : Pat<(v8i16 (bitconvert (v2i64 FPR128:$src))),
5671 (v8i16 (REV64v8i16 FPR128:$src))>;
5672 def : Pat<(v8i16 (bitconvert (v4i32 FPR128:$src))),
5673 (v8i16 (REV32v8i16 FPR128:$src))>;
5674 def : Pat<(v8i16 (bitconvert (v16i8 FPR128:$src))),
5675 (v8i16 (REV16v16i8 FPR128:$src))>;
5676 def : Pat<(v8i16 (bitconvert (v2f64 FPR128:$src))),
5677 (v8i16 (REV64v8i16 FPR128:$src))>;
5678 def : Pat<(v8i16 (bitconvert (v4f32 FPR128:$src))),
5679 (v8i16 (REV32v8i16 FPR128:$src))>;
5680 def : Pat<(v8i16 (bitconvert (v8f16 FPR128:$src))),
5681 (v8i16 (REV32v8i16 FPR128:$src))>;
5684 let Predicates = [IsLE] in {
5685 def : Pat<(v8f16 (bitconvert (f128 FPR128:$src))), (v8f16 FPR128:$src)>;
5686 def : Pat<(v8f16 (bitconvert (v2i64 FPR128:$src))), (v8f16 FPR128:$src)>;
5687 def : Pat<(v8f16 (bitconvert (v4i32 FPR128:$src))), (v8f16 FPR128:$src)>;
5688 def : Pat<(v8f16 (bitconvert (v8i16 FPR128:$src))), (v8f16 FPR128:$src)>;
5689 def : Pat<(v8f16 (bitconvert (v16i8 FPR128:$src))), (v8f16 FPR128:$src)>;
5690 def : Pat<(v8f16 (bitconvert (v2f64 FPR128:$src))), (v8f16 FPR128:$src)>;
5691 def : Pat<(v8f16 (bitconvert (v4f32 FPR128:$src))), (v8f16 FPR128:$src)>;
5693 let Predicates = [IsBE] in {
5694 def : Pat<(v8f16 (bitconvert (f128 FPR128:$src))),
5695 (v8f16 (EXTv16i8 (REV64v8i16 FPR128:$src),
5696 (REV64v8i16 FPR128:$src),
5698 def : Pat<(v8f16 (bitconvert (v2i64 FPR128:$src))),
5699 (v8f16 (REV64v8i16 FPR128:$src))>;
5700 def : Pat<(v8f16 (bitconvert (v4i32 FPR128:$src))),
5701 (v8f16 (REV32v8i16 FPR128:$src))>;
5702 def : Pat<(v8f16 (bitconvert (v8i16 FPR128:$src))),
5703 (v8f16 (REV64v8i16 FPR128:$src))>;
5704 def : Pat<(v8f16 (bitconvert (v16i8 FPR128:$src))),
5705 (v8f16 (REV16v16i8 FPR128:$src))>;
5706 def : Pat<(v8f16 (bitconvert (v2f64 FPR128:$src))),
5707 (v8f16 (REV64v8i16 FPR128:$src))>;
5708 def : Pat<(v8f16 (bitconvert (v4f32 FPR128:$src))),
5709 (v8f16 (REV32v8i16 FPR128:$src))>;
5712 let Predicates = [IsLE] in {
5713 def : Pat<(v16i8 (bitconvert (f128 FPR128:$src))), (v16i8 FPR128:$src)>;
5714 def : Pat<(v16i8 (bitconvert (v2i64 FPR128:$src))), (v16i8 FPR128:$src)>;
5715 def : Pat<(v16i8 (bitconvert (v4i32 FPR128:$src))), (v16i8 FPR128:$src)>;
5716 def : Pat<(v16i8 (bitconvert (v8i16 FPR128:$src))), (v16i8 FPR128:$src)>;
5717 def : Pat<(v16i8 (bitconvert (v2f64 FPR128:$src))), (v16i8 FPR128:$src)>;
5718 def : Pat<(v16i8 (bitconvert (v4f32 FPR128:$src))), (v16i8 FPR128:$src)>;
5719 def : Pat<(v16i8 (bitconvert (v8f16 FPR128:$src))), (v16i8 FPR128:$src)>;
5721 let Predicates = [IsBE] in {
5722 def : Pat<(v16i8 (bitconvert (f128 FPR128:$src))),
5723 (v16i8 (EXTv16i8 (REV64v16i8 FPR128:$src),
5724 (REV64v16i8 FPR128:$src),
5726 def : Pat<(v16i8 (bitconvert (v2i64 FPR128:$src))),
5727 (v16i8 (REV64v16i8 FPR128:$src))>;
5728 def : Pat<(v16i8 (bitconvert (v4i32 FPR128:$src))),
5729 (v16i8 (REV32v16i8 FPR128:$src))>;
5730 def : Pat<(v16i8 (bitconvert (v8i16 FPR128:$src))),
5731 (v16i8 (REV16v16i8 FPR128:$src))>;
5732 def : Pat<(v16i8 (bitconvert (v2f64 FPR128:$src))),
5733 (v16i8 (REV64v16i8 FPR128:$src))>;
5734 def : Pat<(v16i8 (bitconvert (v4f32 FPR128:$src))),
5735 (v16i8 (REV32v16i8 FPR128:$src))>;
5736 def : Pat<(v16i8 (bitconvert (v8f16 FPR128:$src))),
5737 (v16i8 (REV16v16i8 FPR128:$src))>;
5740 def : Pat<(v8i8 (extract_subvector (v16i8 FPR128:$Rn), (i64 1))),
5741 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
5742 def : Pat<(v4i16 (extract_subvector (v8i16 FPR128:$Rn), (i64 1))),
5743 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
5744 def : Pat<(v2i32 (extract_subvector (v4i32 FPR128:$Rn), (i64 1))),
5745 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
5746 def : Pat<(v1i64 (extract_subvector (v2i64 FPR128:$Rn), (i64 1))),
5747 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
5749 // A 64-bit subvector insert to the first 128-bit vector position
5750 // is a subregister copy that needs no instruction.
5751 def : Pat<(insert_subvector undef, (v1i64 FPR64:$src), (i32 0)),
5752 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5753 def : Pat<(insert_subvector undef, (v1f64 FPR64:$src), (i32 0)),
5754 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5755 def : Pat<(insert_subvector undef, (v2i32 FPR64:$src), (i32 0)),
5756 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5757 def : Pat<(insert_subvector undef, (v2f32 FPR64:$src), (i32 0)),
5758 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5759 def : Pat<(insert_subvector undef, (v4i16 FPR64:$src), (i32 0)),
5760 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5761 def : Pat<(insert_subvector undef, (v4f16 FPR64:$src), (i32 0)),
5762 (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5763 def : Pat<(insert_subvector undef, (v8i8 FPR64:$src), (i32 0)),
5764 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5766 // Use pair-wise add instructions when summing up the lanes for v2f64, v2i64
5768 def : Pat<(i64 (add (vector_extract (v2i64 FPR128:$Rn), (i64 0)),
5769 (vector_extract (v2i64 FPR128:$Rn), (i64 1)))),
5770 (i64 (ADDPv2i64p (v2i64 FPR128:$Rn)))>;
5771 def : Pat<(f64 (fadd (vector_extract (v2f64 FPR128:$Rn), (i64 0)),
5772 (vector_extract (v2f64 FPR128:$Rn), (i64 1)))),
5773 (f64 (FADDPv2i64p (v2f64 FPR128:$Rn)))>;
5774 // vector_extract on 64-bit vectors gets promoted to a 128 bit vector,
5775 // so we match on v4f32 here, not v2f32. This will also catch adding
5776 // the low two lanes of a true v4f32 vector.
5777 def : Pat<(fadd (vector_extract (v4f32 FPR128:$Rn), (i64 0)),
5778 (vector_extract (v4f32 FPR128:$Rn), (i64 1))),
5779 (f32 (FADDPv2i32p (EXTRACT_SUBREG FPR128:$Rn, dsub)))>;
5781 // Scalar 64-bit shifts in FPR64 registers.
5782 def : Pat<(i64 (int_aarch64_neon_sshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
5783 (SSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
5784 def : Pat<(i64 (int_aarch64_neon_ushl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
5785 (USHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
5786 def : Pat<(i64 (int_aarch64_neon_srshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
5787 (SRSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
5788 def : Pat<(i64 (int_aarch64_neon_urshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
5789 (URSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
5791 // Tail call return handling. These are all compiler pseudo-instructions,
5792 // so no encoding information or anything like that.
5793 let isCall = 1, isTerminator = 1, isReturn = 1, isBarrier = 1, Uses = [SP] in {
5794 def TCRETURNdi : Pseudo<(outs), (ins i64imm:$dst, i32imm:$FPDiff),[]>;
5795 def TCRETURNri : Pseudo<(outs), (ins tcGPR64:$dst, i32imm:$FPDiff), []>;
5798 def : Pat<(AArch64tcret tcGPR64:$dst, (i32 timm:$FPDiff)),
5799 (TCRETURNri tcGPR64:$dst, imm:$FPDiff)>;
5800 def : Pat<(AArch64tcret tglobaladdr:$dst, (i32 timm:$FPDiff)),
5801 (TCRETURNdi texternalsym:$dst, imm:$FPDiff)>;
5802 def : Pat<(AArch64tcret texternalsym:$dst, (i32 timm:$FPDiff)),
5803 (TCRETURNdi texternalsym:$dst, imm:$FPDiff)>;
5805 include "AArch64InstrAtomics.td"