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", "sub", add>;
589 defm SUB : AddSub<1, "sub", "add">;
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:$src, 0, 0)>;
600 defm ADDS : AddSubS<0, "adds", AArch64add_flag, "cmn", "subs", "cmp">;
601 defm SUBS : AddSubS<1, "subs", AArch64sub_flag, "cmp", "adds", "cmn">;
603 // Use SUBS instead of SUB to enable CSE between SUBS and SUB.
604 def : Pat<(sub GPR32sp:$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:$Rm)>;
616 let AddedComplexity = 1 in {
617 def : Pat<(sub GPR32sp:$R2, arith_extended_reg32<i32>:$R3),
618 (SUBSWrx GPR32sp:$R2, arith_extended_reg32<i32>:$R3)>;
619 def : Pat<(sub GPR64sp:$R2, arith_extended_reg32to64<i64>:$R3),
620 (SUBSXrx GPR64sp:$R2, arith_extended_reg32to64<i64>:$R3)>;
623 // Because of the immediate format for add/sub-imm instructions, the
624 // expression (add x, -1) must be transformed to (SUB{W,X}ri x, 1).
625 // These patterns capture that transformation.
626 let AddedComplexity = 1 in {
627 def : Pat<(add GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
628 (SUBSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
629 def : Pat<(add GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
630 (SUBSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
631 def : Pat<(sub GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
632 (ADDWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
633 def : Pat<(sub GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
634 (ADDXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
637 // Because of the immediate format for add/sub-imm instructions, the
638 // expression (add x, -1) must be transformed to (SUB{W,X}ri x, 1).
639 // These patterns capture that transformation.
640 let AddedComplexity = 1 in {
641 def : Pat<(AArch64add_flag GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
642 (SUBSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
643 def : Pat<(AArch64add_flag GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
644 (SUBSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
645 def : Pat<(AArch64sub_flag GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
646 (ADDSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
647 def : Pat<(AArch64sub_flag GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
648 (ADDSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
651 def : InstAlias<"neg $dst, $src", (SUBWrs GPR32:$dst, WZR, GPR32:$src, 0), 3>;
652 def : InstAlias<"neg $dst, $src", (SUBXrs GPR64:$dst, XZR, GPR64:$src, 0), 3>;
653 def : InstAlias<"neg $dst, $src$shift",
654 (SUBWrs GPR32:$dst, WZR, GPR32:$src, arith_shift32:$shift), 2>;
655 def : InstAlias<"neg $dst, $src$shift",
656 (SUBXrs GPR64:$dst, XZR, GPR64:$src, arith_shift64:$shift), 2>;
658 def : InstAlias<"negs $dst, $src", (SUBSWrs GPR32:$dst, WZR, GPR32:$src, 0), 3>;
659 def : InstAlias<"negs $dst, $src", (SUBSXrs GPR64:$dst, XZR, GPR64:$src, 0), 3>;
660 def : InstAlias<"negs $dst, $src$shift",
661 (SUBSWrs GPR32:$dst, WZR, GPR32:$src, arith_shift32:$shift), 2>;
662 def : InstAlias<"negs $dst, $src$shift",
663 (SUBSXrs GPR64:$dst, XZR, GPR64:$src, arith_shift64:$shift), 2>;
666 // Unsigned/Signed divide
667 defm UDIV : Div<0, "udiv", udiv>;
668 defm SDIV : Div<1, "sdiv", sdiv>;
669 let isCodeGenOnly = 1 in {
670 defm UDIV_Int : Div<0, "udiv", int_aarch64_udiv>;
671 defm SDIV_Int : Div<1, "sdiv", int_aarch64_sdiv>;
675 defm ASRV : Shift<0b10, "asr", sra>;
676 defm LSLV : Shift<0b00, "lsl", shl>;
677 defm LSRV : Shift<0b01, "lsr", srl>;
678 defm RORV : Shift<0b11, "ror", rotr>;
680 def : ShiftAlias<"asrv", ASRVWr, GPR32>;
681 def : ShiftAlias<"asrv", ASRVXr, GPR64>;
682 def : ShiftAlias<"lslv", LSLVWr, GPR32>;
683 def : ShiftAlias<"lslv", LSLVXr, GPR64>;
684 def : ShiftAlias<"lsrv", LSRVWr, GPR32>;
685 def : ShiftAlias<"lsrv", LSRVXr, GPR64>;
686 def : ShiftAlias<"rorv", RORVWr, GPR32>;
687 def : ShiftAlias<"rorv", RORVXr, GPR64>;
690 let AddedComplexity = 7 in {
691 defm MADD : MulAccum<0, "madd", add>;
692 defm MSUB : MulAccum<1, "msub", sub>;
694 def : Pat<(i32 (mul GPR32:$Rn, GPR32:$Rm)),
695 (MADDWrrr GPR32:$Rn, GPR32:$Rm, WZR)>;
696 def : Pat<(i64 (mul GPR64:$Rn, GPR64:$Rm)),
697 (MADDXrrr GPR64:$Rn, GPR64:$Rm, XZR)>;
699 def : Pat<(i32 (ineg (mul GPR32:$Rn, GPR32:$Rm))),
700 (MSUBWrrr GPR32:$Rn, GPR32:$Rm, WZR)>;
701 def : Pat<(i64 (ineg (mul GPR64:$Rn, GPR64:$Rm))),
702 (MSUBXrrr GPR64:$Rn, GPR64:$Rm, XZR)>;
703 def : Pat<(i32 (mul (ineg GPR32:$Rn), GPR32:$Rm)),
704 (MSUBWrrr GPR32:$Rn, GPR32:$Rm, WZR)>;
705 def : Pat<(i64 (mul (ineg GPR64:$Rn), GPR64:$Rm)),
706 (MSUBXrrr GPR64:$Rn, GPR64:$Rm, XZR)>;
707 } // AddedComplexity = 7
709 let AddedComplexity = 5 in {
710 def SMADDLrrr : WideMulAccum<0, 0b001, "smaddl", add, sext>;
711 def SMSUBLrrr : WideMulAccum<1, 0b001, "smsubl", sub, sext>;
712 def UMADDLrrr : WideMulAccum<0, 0b101, "umaddl", add, zext>;
713 def UMSUBLrrr : WideMulAccum<1, 0b101, "umsubl", sub, zext>;
715 def : Pat<(i64 (mul (sext GPR32:$Rn), (sext GPR32:$Rm))),
716 (SMADDLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
717 def : Pat<(i64 (mul (zext GPR32:$Rn), (zext GPR32:$Rm))),
718 (UMADDLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
720 def : Pat<(i64 (ineg (mul (sext GPR32:$Rn), (sext GPR32:$Rm)))),
721 (SMSUBLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
722 def : Pat<(i64 (ineg (mul (zext GPR32:$Rn), (zext GPR32:$Rm)))),
723 (UMSUBLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
724 } // AddedComplexity = 5
726 def : MulAccumWAlias<"mul", MADDWrrr>;
727 def : MulAccumXAlias<"mul", MADDXrrr>;
728 def : MulAccumWAlias<"mneg", MSUBWrrr>;
729 def : MulAccumXAlias<"mneg", MSUBXrrr>;
730 def : WideMulAccumAlias<"smull", SMADDLrrr>;
731 def : WideMulAccumAlias<"smnegl", SMSUBLrrr>;
732 def : WideMulAccumAlias<"umull", UMADDLrrr>;
733 def : WideMulAccumAlias<"umnegl", UMSUBLrrr>;
736 def SMULHrr : MulHi<0b010, "smulh", mulhs>;
737 def UMULHrr : MulHi<0b110, "umulh", mulhu>;
740 def CRC32Brr : BaseCRC32<0, 0b00, 0, GPR32, int_aarch64_crc32b, "crc32b">;
741 def CRC32Hrr : BaseCRC32<0, 0b01, 0, GPR32, int_aarch64_crc32h, "crc32h">;
742 def CRC32Wrr : BaseCRC32<0, 0b10, 0, GPR32, int_aarch64_crc32w, "crc32w">;
743 def CRC32Xrr : BaseCRC32<1, 0b11, 0, GPR64, int_aarch64_crc32x, "crc32x">;
745 def CRC32CBrr : BaseCRC32<0, 0b00, 1, GPR32, int_aarch64_crc32cb, "crc32cb">;
746 def CRC32CHrr : BaseCRC32<0, 0b01, 1, GPR32, int_aarch64_crc32ch, "crc32ch">;
747 def CRC32CWrr : BaseCRC32<0, 0b10, 1, GPR32, int_aarch64_crc32cw, "crc32cw">;
748 def CRC32CXrr : BaseCRC32<1, 0b11, 1, GPR64, int_aarch64_crc32cx, "crc32cx">;
751 defm CAS : CompareAndSwap<0, 0, "">;
752 defm CASA : CompareAndSwap<1, 0, "a">;
753 defm CASL : CompareAndSwap<0, 1, "l">;
754 defm CASAL : CompareAndSwap<1, 1, "al">;
757 defm CASP : CompareAndSwapPair<0, 0, "">;
758 defm CASPA : CompareAndSwapPair<1, 0, "a">;
759 defm CASPL : CompareAndSwapPair<0, 1, "l">;
760 defm CASPAL : CompareAndSwapPair<1, 1, "al">;
763 defm SWP : Swap<0, 0, "">;
764 defm SWPA : Swap<1, 0, "a">;
765 defm SWPL : Swap<0, 1, "l">;
766 defm SWPAL : Swap<1, 1, "al">;
768 // v8.1 atomic LD<OP>(register). Performs load and then ST<OP>(register)
769 defm LDADD : LDOPregister<0b000, "add", 0, 0, "">;
770 defm LDADDA : LDOPregister<0b000, "add", 1, 0, "a">;
771 defm LDADDL : LDOPregister<0b000, "add", 0, 1, "l">;
772 defm LDADDAL : LDOPregister<0b000, "add", 1, 1, "al">;
774 defm LDCLR : LDOPregister<0b001, "clr", 0, 0, "">;
775 defm LDCLRA : LDOPregister<0b001, "clr", 1, 0, "a">;
776 defm LDCLRL : LDOPregister<0b001, "clr", 0, 1, "l">;
777 defm LDCLRAL : LDOPregister<0b001, "clr", 1, 1, "al">;
779 defm LDEOR : LDOPregister<0b010, "eor", 0, 0, "">;
780 defm LDEORA : LDOPregister<0b010, "eor", 1, 0, "a">;
781 defm LDEORL : LDOPregister<0b010, "eor", 0, 1, "l">;
782 defm LDEORAL : LDOPregister<0b010, "eor", 1, 1, "al">;
784 defm LDSET : LDOPregister<0b011, "set", 0, 0, "">;
785 defm LDSETA : LDOPregister<0b011, "set", 1, 0, "a">;
786 defm LDSETL : LDOPregister<0b011, "set", 0, 1, "l">;
787 defm LDSETAL : LDOPregister<0b011, "set", 1, 1, "al">;
789 defm LDSMAX : LDOPregister<0b100, "smax", 0, 0, "">;
790 defm LDSMAXA : LDOPregister<0b100, "smax", 1, 0, "a">;
791 defm LDSMAXL : LDOPregister<0b100, "smax", 0, 1, "l">;
792 defm LDSMAXAL : LDOPregister<0b100, "smax", 1, 1, "al">;
794 defm LDSMIN : LDOPregister<0b101, "smin", 0, 0, "">;
795 defm LDSMINA : LDOPregister<0b101, "smin", 1, 0, "a">;
796 defm LDSMINL : LDOPregister<0b101, "smin", 0, 1, "l">;
797 defm LDSMINAL : LDOPregister<0b101, "smin", 1, 1, "al">;
799 defm LDUMAX : LDOPregister<0b110, "umax", 0, 0, "">;
800 defm LDUMAXA : LDOPregister<0b110, "umax", 1, 0, "a">;
801 defm LDUMAXL : LDOPregister<0b110, "umax", 0, 1, "l">;
802 defm LDUMAXAL : LDOPregister<0b110, "umax", 1, 1, "al">;
804 defm LDUMIN : LDOPregister<0b111, "umin", 0, 0, "">;
805 defm LDUMINA : LDOPregister<0b111, "umin", 1, 0, "a">;
806 defm LDUMINL : LDOPregister<0b111, "umin", 0, 1, "l">;
807 defm LDUMINAL : LDOPregister<0b111, "umin", 1, 1, "al">;
809 // v8.1 atomic ST<OP>(register) as aliases to "LD<OP>(register) when Rt=xZR"
810 defm : STOPregister<"stadd","LDADD">; // STADDx
811 defm : STOPregister<"stclr","LDCLR">; // STCLRx
812 defm : STOPregister<"steor","LDEOR">; // STEORx
813 defm : STOPregister<"stset","LDSET">; // STSETx
814 defm : STOPregister<"stsmax","LDSMAX">;// STSMAXx
815 defm : STOPregister<"stsmin","LDSMIN">;// STSMINx
816 defm : STOPregister<"stumax","LDUMAX">;// STUMAXx
817 defm : STOPregister<"stumin","LDUMIN">;// STUMINx
819 //===----------------------------------------------------------------------===//
820 // Logical instructions.
821 //===----------------------------------------------------------------------===//
824 defm ANDS : LogicalImmS<0b11, "ands", AArch64and_flag, "bics">;
825 defm AND : LogicalImm<0b00, "and", and, "bic">;
826 defm EOR : LogicalImm<0b10, "eor", xor, "eon">;
827 defm ORR : LogicalImm<0b01, "orr", or, "orn">;
829 // FIXME: these aliases *are* canonical sometimes (when movz can't be
830 // used). Actually, it seems to be working right now, but putting logical_immXX
831 // here is a bit dodgy on the AsmParser side too.
832 def : InstAlias<"mov $dst, $imm", (ORRWri GPR32sp:$dst, WZR,
833 logical_imm32:$imm), 0>;
834 def : InstAlias<"mov $dst, $imm", (ORRXri GPR64sp:$dst, XZR,
835 logical_imm64:$imm), 0>;
839 defm ANDS : LogicalRegS<0b11, 0, "ands", AArch64and_flag>;
840 defm BICS : LogicalRegS<0b11, 1, "bics",
841 BinOpFrag<(AArch64and_flag node:$LHS, (not node:$RHS))>>;
842 defm AND : LogicalReg<0b00, 0, "and", and>;
843 defm BIC : LogicalReg<0b00, 1, "bic",
844 BinOpFrag<(and node:$LHS, (not node:$RHS))>>;
845 defm EON : LogicalReg<0b10, 1, "eon",
846 BinOpFrag<(not (xor node:$LHS, node:$RHS))>>;
847 defm EOR : LogicalReg<0b10, 0, "eor", xor>;
848 defm ORN : LogicalReg<0b01, 1, "orn",
849 BinOpFrag<(or node:$LHS, (not node:$RHS))>>;
850 defm ORR : LogicalReg<0b01, 0, "orr", or>;
852 def : InstAlias<"mov $dst, $src", (ORRWrs GPR32:$dst, WZR, GPR32:$src, 0), 2>;
853 def : InstAlias<"mov $dst, $src", (ORRXrs GPR64:$dst, XZR, GPR64:$src, 0), 2>;
855 def : InstAlias<"mvn $Wd, $Wm", (ORNWrs GPR32:$Wd, WZR, GPR32:$Wm, 0), 3>;
856 def : InstAlias<"mvn $Xd, $Xm", (ORNXrs GPR64:$Xd, XZR, GPR64:$Xm, 0), 3>;
858 def : InstAlias<"mvn $Wd, $Wm$sh",
859 (ORNWrs GPR32:$Wd, WZR, GPR32:$Wm, logical_shift32:$sh), 2>;
860 def : InstAlias<"mvn $Xd, $Xm$sh",
861 (ORNXrs GPR64:$Xd, XZR, GPR64:$Xm, logical_shift64:$sh), 2>;
863 def : InstAlias<"tst $src1, $src2",
864 (ANDSWri WZR, GPR32:$src1, logical_imm32:$src2), 2>;
865 def : InstAlias<"tst $src1, $src2",
866 (ANDSXri XZR, GPR64:$src1, logical_imm64:$src2), 2>;
868 def : InstAlias<"tst $src1, $src2",
869 (ANDSWrs WZR, GPR32:$src1, GPR32:$src2, 0), 3>;
870 def : InstAlias<"tst $src1, $src2",
871 (ANDSXrs XZR, GPR64:$src1, GPR64:$src2, 0), 3>;
873 def : InstAlias<"tst $src1, $src2$sh",
874 (ANDSWrs WZR, GPR32:$src1, GPR32:$src2, logical_shift32:$sh), 2>;
875 def : InstAlias<"tst $src1, $src2$sh",
876 (ANDSXrs XZR, GPR64:$src1, GPR64:$src2, logical_shift64:$sh), 2>;
879 def : Pat<(not GPR32:$Wm), (ORNWrr WZR, GPR32:$Wm)>;
880 def : Pat<(not GPR64:$Xm), (ORNXrr XZR, GPR64:$Xm)>;
883 //===----------------------------------------------------------------------===//
884 // One operand data processing instructions.
885 //===----------------------------------------------------------------------===//
887 defm CLS : OneOperandData<0b101, "cls">;
888 defm CLZ : OneOperandData<0b100, "clz", ctlz>;
889 defm RBIT : OneOperandData<0b000, "rbit">;
891 def : Pat<(int_aarch64_rbit GPR32:$Rn), (RBITWr $Rn)>;
892 def : Pat<(int_aarch64_rbit GPR64:$Rn), (RBITXr $Rn)>;
894 def REV16Wr : OneWRegData<0b001, "rev16",
895 UnOpFrag<(rotr (bswap node:$LHS), (i64 16))>>;
896 def REV16Xr : OneXRegData<0b001, "rev16", null_frag>;
898 def : Pat<(cttz GPR32:$Rn),
899 (CLZWr (RBITWr GPR32:$Rn))>;
900 def : Pat<(cttz GPR64:$Rn),
901 (CLZXr (RBITXr GPR64:$Rn))>;
902 def : Pat<(ctlz (or (shl (xor (sra GPR32:$Rn, (i64 31)), GPR32:$Rn), (i64 1)),
905 def : Pat<(ctlz (or (shl (xor (sra GPR64:$Rn, (i64 63)), GPR64:$Rn), (i64 1)),
909 // Unlike the other one operand instructions, the instructions with the "rev"
910 // mnemonic do *not* just different in the size bit, but actually use different
911 // opcode bits for the different sizes.
912 def REVWr : OneWRegData<0b010, "rev", bswap>;
913 def REVXr : OneXRegData<0b011, "rev", bswap>;
914 def REV32Xr : OneXRegData<0b010, "rev32",
915 UnOpFrag<(rotr (bswap node:$LHS), (i64 32))>>;
917 def : InstAlias<"rev64 $Rd, $Rn", (REVXr GPR64:$Rd, GPR64:$Rn), 0>;
919 // The bswap commutes with the rotr so we want a pattern for both possible
921 def : Pat<(bswap (rotr GPR32:$Rn, (i64 16))), (REV16Wr GPR32:$Rn)>;
922 def : Pat<(bswap (rotr GPR64:$Rn, (i64 32))), (REV32Xr GPR64:$Rn)>;
924 //===----------------------------------------------------------------------===//
925 // Bitfield immediate extraction instruction.
926 //===----------------------------------------------------------------------===//
927 let hasSideEffects = 0 in
928 defm EXTR : ExtractImm<"extr">;
929 def : InstAlias<"ror $dst, $src, $shift",
930 (EXTRWrri GPR32:$dst, GPR32:$src, GPR32:$src, imm0_31:$shift)>;
931 def : InstAlias<"ror $dst, $src, $shift",
932 (EXTRXrri GPR64:$dst, GPR64:$src, GPR64:$src, imm0_63:$shift)>;
934 def : Pat<(rotr GPR32:$Rn, (i64 imm0_31:$imm)),
935 (EXTRWrri GPR32:$Rn, GPR32:$Rn, imm0_31:$imm)>;
936 def : Pat<(rotr GPR64:$Rn, (i64 imm0_63:$imm)),
937 (EXTRXrri GPR64:$Rn, GPR64:$Rn, imm0_63:$imm)>;
939 //===----------------------------------------------------------------------===//
940 // Other bitfield immediate instructions.
941 //===----------------------------------------------------------------------===//
942 let hasSideEffects = 0 in {
943 defm BFM : BitfieldImmWith2RegArgs<0b01, "bfm">;
944 defm SBFM : BitfieldImm<0b00, "sbfm">;
945 defm UBFM : BitfieldImm<0b10, "ubfm">;
948 def i32shift_a : Operand<i64>, SDNodeXForm<imm, [{
949 uint64_t enc = (32 - N->getZExtValue()) & 0x1f;
950 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
953 def i32shift_b : Operand<i64>, SDNodeXForm<imm, [{
954 uint64_t enc = 31 - N->getZExtValue();
955 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
958 // min(7, 31 - shift_amt)
959 def i32shift_sext_i8 : Operand<i64>, SDNodeXForm<imm, [{
960 uint64_t enc = 31 - N->getZExtValue();
961 enc = enc > 7 ? 7 : enc;
962 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
965 // min(15, 31 - shift_amt)
966 def i32shift_sext_i16 : Operand<i64>, SDNodeXForm<imm, [{
967 uint64_t enc = 31 - N->getZExtValue();
968 enc = enc > 15 ? 15 : enc;
969 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
972 def i64shift_a : Operand<i64>, SDNodeXForm<imm, [{
973 uint64_t enc = (64 - N->getZExtValue()) & 0x3f;
974 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
977 def i64shift_b : Operand<i64>, SDNodeXForm<imm, [{
978 uint64_t enc = 63 - N->getZExtValue();
979 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
982 // min(7, 63 - shift_amt)
983 def i64shift_sext_i8 : Operand<i64>, SDNodeXForm<imm, [{
984 uint64_t enc = 63 - N->getZExtValue();
985 enc = enc > 7 ? 7 : enc;
986 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
989 // min(15, 63 - shift_amt)
990 def i64shift_sext_i16 : Operand<i64>, SDNodeXForm<imm, [{
991 uint64_t enc = 63 - N->getZExtValue();
992 enc = enc > 15 ? 15 : enc;
993 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
996 // min(31, 63 - shift_amt)
997 def i64shift_sext_i32 : Operand<i64>, SDNodeXForm<imm, [{
998 uint64_t enc = 63 - N->getZExtValue();
999 enc = enc > 31 ? 31 : enc;
1000 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
1003 def : Pat<(shl GPR32:$Rn, (i64 imm0_31:$imm)),
1004 (UBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
1005 (i64 (i32shift_b imm0_31:$imm)))>;
1006 def : Pat<(shl GPR64:$Rn, (i64 imm0_63:$imm)),
1007 (UBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
1008 (i64 (i64shift_b imm0_63:$imm)))>;
1010 let AddedComplexity = 10 in {
1011 def : Pat<(sra GPR32:$Rn, (i64 imm0_31:$imm)),
1012 (SBFMWri GPR32:$Rn, imm0_31:$imm, 31)>;
1013 def : Pat<(sra GPR64:$Rn, (i64 imm0_63:$imm)),
1014 (SBFMXri GPR64:$Rn, imm0_63:$imm, 63)>;
1017 def : InstAlias<"asr $dst, $src, $shift",
1018 (SBFMWri GPR32:$dst, GPR32:$src, imm0_31:$shift, 31)>;
1019 def : InstAlias<"asr $dst, $src, $shift",
1020 (SBFMXri GPR64:$dst, GPR64:$src, imm0_63:$shift, 63)>;
1021 def : InstAlias<"sxtb $dst, $src", (SBFMWri GPR32:$dst, GPR32:$src, 0, 7)>;
1022 def : InstAlias<"sxtb $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 7)>;
1023 def : InstAlias<"sxth $dst, $src", (SBFMWri GPR32:$dst, GPR32:$src, 0, 15)>;
1024 def : InstAlias<"sxth $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 15)>;
1025 def : InstAlias<"sxtw $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 31)>;
1027 def : Pat<(srl GPR32:$Rn, (i64 imm0_31:$imm)),
1028 (UBFMWri GPR32:$Rn, imm0_31:$imm, 31)>;
1029 def : Pat<(srl GPR64:$Rn, (i64 imm0_63:$imm)),
1030 (UBFMXri GPR64:$Rn, imm0_63:$imm, 63)>;
1032 def : InstAlias<"lsr $dst, $src, $shift",
1033 (UBFMWri GPR32:$dst, GPR32:$src, imm0_31:$shift, 31)>;
1034 def : InstAlias<"lsr $dst, $src, $shift",
1035 (UBFMXri GPR64:$dst, GPR64:$src, imm0_63:$shift, 63)>;
1036 def : InstAlias<"uxtb $dst, $src", (UBFMWri GPR32:$dst, GPR32:$src, 0, 7)>;
1037 def : InstAlias<"uxtb $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 7)>;
1038 def : InstAlias<"uxth $dst, $src", (UBFMWri GPR32:$dst, GPR32:$src, 0, 15)>;
1039 def : InstAlias<"uxth $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 15)>;
1040 def : InstAlias<"uxtw $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 31)>;
1042 //===----------------------------------------------------------------------===//
1043 // Conditional comparison instructions.
1044 //===----------------------------------------------------------------------===//
1045 defm CCMN : CondComparison<0, "ccmn", AArch64ccmn>;
1046 defm CCMP : CondComparison<1, "ccmp", AArch64ccmp>;
1048 //===----------------------------------------------------------------------===//
1049 // Conditional select instructions.
1050 //===----------------------------------------------------------------------===//
1051 defm CSEL : CondSelect<0, 0b00, "csel">;
1053 def inc : PatFrag<(ops node:$in), (add node:$in, 1)>;
1054 defm CSINC : CondSelectOp<0, 0b01, "csinc", inc>;
1055 defm CSINV : CondSelectOp<1, 0b00, "csinv", not>;
1056 defm CSNEG : CondSelectOp<1, 0b01, "csneg", ineg>;
1058 def : Pat<(AArch64csinv GPR32:$tval, GPR32:$fval, (i32 imm:$cc), NZCV),
1059 (CSINVWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
1060 def : Pat<(AArch64csinv GPR64:$tval, GPR64:$fval, (i32 imm:$cc), NZCV),
1061 (CSINVXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
1062 def : Pat<(AArch64csneg GPR32:$tval, GPR32:$fval, (i32 imm:$cc), NZCV),
1063 (CSNEGWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
1064 def : Pat<(AArch64csneg GPR64:$tval, GPR64:$fval, (i32 imm:$cc), NZCV),
1065 (CSNEGXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
1066 def : Pat<(AArch64csinc GPR32:$tval, GPR32:$fval, (i32 imm:$cc), NZCV),
1067 (CSINCWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
1068 def : Pat<(AArch64csinc GPR64:$tval, GPR64:$fval, (i32 imm:$cc), NZCV),
1069 (CSINCXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
1071 def : Pat<(AArch64csel (i32 0), (i32 1), (i32 imm:$cc), NZCV),
1072 (CSINCWr WZR, WZR, (i32 imm:$cc))>;
1073 def : Pat<(AArch64csel (i64 0), (i64 1), (i32 imm:$cc), NZCV),
1074 (CSINCXr XZR, XZR, (i32 imm:$cc))>;
1075 def : Pat<(AArch64csel (i32 0), (i32 -1), (i32 imm:$cc), NZCV),
1076 (CSINVWr WZR, WZR, (i32 imm:$cc))>;
1077 def : Pat<(AArch64csel (i64 0), (i64 -1), (i32 imm:$cc), NZCV),
1078 (CSINVXr XZR, XZR, (i32 imm:$cc))>;
1080 // The inverse of the condition code from the alias instruction is what is used
1081 // in the aliased instruction. The parser all ready inverts the condition code
1082 // for these aliases.
1083 def : InstAlias<"cset $dst, $cc",
1084 (CSINCWr GPR32:$dst, WZR, WZR, inv_ccode:$cc)>;
1085 def : InstAlias<"cset $dst, $cc",
1086 (CSINCXr GPR64:$dst, XZR, XZR, inv_ccode:$cc)>;
1088 def : InstAlias<"csetm $dst, $cc",
1089 (CSINVWr GPR32:$dst, WZR, WZR, inv_ccode:$cc)>;
1090 def : InstAlias<"csetm $dst, $cc",
1091 (CSINVXr GPR64:$dst, XZR, XZR, inv_ccode:$cc)>;
1093 def : InstAlias<"cinc $dst, $src, $cc",
1094 (CSINCWr GPR32:$dst, GPR32:$src, GPR32:$src, inv_ccode:$cc)>;
1095 def : InstAlias<"cinc $dst, $src, $cc",
1096 (CSINCXr GPR64:$dst, GPR64:$src, GPR64:$src, inv_ccode:$cc)>;
1098 def : InstAlias<"cinv $dst, $src, $cc",
1099 (CSINVWr GPR32:$dst, GPR32:$src, GPR32:$src, inv_ccode:$cc)>;
1100 def : InstAlias<"cinv $dst, $src, $cc",
1101 (CSINVXr GPR64:$dst, GPR64:$src, GPR64:$src, inv_ccode:$cc)>;
1103 def : InstAlias<"cneg $dst, $src, $cc",
1104 (CSNEGWr GPR32:$dst, GPR32:$src, GPR32:$src, inv_ccode:$cc)>;
1105 def : InstAlias<"cneg $dst, $src, $cc",
1106 (CSNEGXr GPR64:$dst, GPR64:$src, GPR64:$src, inv_ccode:$cc)>;
1108 //===----------------------------------------------------------------------===//
1109 // PC-relative instructions.
1110 //===----------------------------------------------------------------------===//
1111 let isReMaterializable = 1 in {
1112 let hasSideEffects = 0, mayStore = 0, mayLoad = 0 in {
1113 def ADR : ADRI<0, "adr", adrlabel, []>;
1114 } // hasSideEffects = 0
1116 def ADRP : ADRI<1, "adrp", adrplabel,
1117 [(set GPR64:$Xd, (AArch64adrp tglobaladdr:$label))]>;
1118 } // isReMaterializable = 1
1120 // page address of a constant pool entry, block address
1121 def : Pat<(AArch64adrp tconstpool:$cp), (ADRP tconstpool:$cp)>;
1122 def : Pat<(AArch64adrp tblockaddress:$cp), (ADRP tblockaddress:$cp)>;
1124 //===----------------------------------------------------------------------===//
1125 // Unconditional branch (register) instructions.
1126 //===----------------------------------------------------------------------===//
1128 let isReturn = 1, isTerminator = 1, isBarrier = 1 in {
1129 def RET : BranchReg<0b0010, "ret", []>;
1130 def DRPS : SpecialReturn<0b0101, "drps">;
1131 def ERET : SpecialReturn<0b0100, "eret">;
1132 } // isReturn = 1, isTerminator = 1, isBarrier = 1
1134 // Default to the LR register.
1135 def : InstAlias<"ret", (RET LR)>;
1137 let isCall = 1, Defs = [LR], Uses = [SP] in {
1138 def BLR : BranchReg<0b0001, "blr", [(AArch64call GPR64:$Rn)]>;
1141 let isBranch = 1, isTerminator = 1, isBarrier = 1, isIndirectBranch = 1 in {
1142 def BR : BranchReg<0b0000, "br", [(brind GPR64:$Rn)]>;
1143 } // isBranch, isTerminator, isBarrier, isIndirectBranch
1145 // Create a separate pseudo-instruction for codegen to use so that we don't
1146 // flag lr as used in every function. It'll be restored before the RET by the
1147 // epilogue if it's legitimately used.
1148 def RET_ReallyLR : Pseudo<(outs), (ins), [(AArch64retflag)]> {
1149 let isTerminator = 1;
1154 // This is a directive-like pseudo-instruction. The purpose is to insert an
1155 // R_AARCH64_TLSDESC_CALL relocation at the offset of the following instruction
1156 // (which in the usual case is a BLR).
1157 let hasSideEffects = 1 in
1158 def TLSDESCCALL : Pseudo<(outs), (ins i64imm:$sym), []> {
1159 let AsmString = ".tlsdesccall $sym";
1162 // FIXME: maybe the scratch register used shouldn't be fixed to X1?
1163 // FIXME: can "hasSideEffects be dropped?
1164 let isCall = 1, Defs = [LR, X0, X1], hasSideEffects = 1,
1165 isCodeGenOnly = 1 in
1167 : Pseudo<(outs), (ins i64imm:$sym),
1168 [(AArch64tlsdesc_callseq tglobaltlsaddr:$sym)]>;
1169 def : Pat<(AArch64tlsdesc_callseq texternalsym:$sym),
1170 (TLSDESC_CALLSEQ texternalsym:$sym)>;
1172 //===----------------------------------------------------------------------===//
1173 // Conditional branch (immediate) instruction.
1174 //===----------------------------------------------------------------------===//
1175 def Bcc : BranchCond;
1177 //===----------------------------------------------------------------------===//
1178 // Compare-and-branch instructions.
1179 //===----------------------------------------------------------------------===//
1180 defm CBZ : CmpBranch<0, "cbz", AArch64cbz>;
1181 defm CBNZ : CmpBranch<1, "cbnz", AArch64cbnz>;
1183 //===----------------------------------------------------------------------===//
1184 // Test-bit-and-branch instructions.
1185 //===----------------------------------------------------------------------===//
1186 defm TBZ : TestBranch<0, "tbz", AArch64tbz>;
1187 defm TBNZ : TestBranch<1, "tbnz", AArch64tbnz>;
1189 //===----------------------------------------------------------------------===//
1190 // Unconditional branch (immediate) instructions.
1191 //===----------------------------------------------------------------------===//
1192 let isBranch = 1, isTerminator = 1, isBarrier = 1 in {
1193 def B : BranchImm<0, "b", [(br bb:$addr)]>;
1194 } // isBranch, isTerminator, isBarrier
1196 let isCall = 1, Defs = [LR], Uses = [SP] in {
1197 def BL : CallImm<1, "bl", [(AArch64call tglobaladdr:$addr)]>;
1199 def : Pat<(AArch64call texternalsym:$func), (BL texternalsym:$func)>;
1201 //===----------------------------------------------------------------------===//
1202 // Exception generation instructions.
1203 //===----------------------------------------------------------------------===//
1204 def BRK : ExceptionGeneration<0b001, 0b00, "brk">;
1205 def DCPS1 : ExceptionGeneration<0b101, 0b01, "dcps1">;
1206 def DCPS2 : ExceptionGeneration<0b101, 0b10, "dcps2">;
1207 def DCPS3 : ExceptionGeneration<0b101, 0b11, "dcps3">;
1208 def HLT : ExceptionGeneration<0b010, 0b00, "hlt">;
1209 def HVC : ExceptionGeneration<0b000, 0b10, "hvc">;
1210 def SMC : ExceptionGeneration<0b000, 0b11, "smc">;
1211 def SVC : ExceptionGeneration<0b000, 0b01, "svc">;
1213 // DCPSn defaults to an immediate operand of zero if unspecified.
1214 def : InstAlias<"dcps1", (DCPS1 0)>;
1215 def : InstAlias<"dcps2", (DCPS2 0)>;
1216 def : InstAlias<"dcps3", (DCPS3 0)>;
1218 //===----------------------------------------------------------------------===//
1219 // Load instructions.
1220 //===----------------------------------------------------------------------===//
1222 // Pair (indexed, offset)
1223 defm LDPW : LoadPairOffset<0b00, 0, GPR32, simm7s4, "ldp">;
1224 defm LDPX : LoadPairOffset<0b10, 0, GPR64, simm7s8, "ldp">;
1225 defm LDPS : LoadPairOffset<0b00, 1, FPR32, simm7s4, "ldp">;
1226 defm LDPD : LoadPairOffset<0b01, 1, FPR64, simm7s8, "ldp">;
1227 defm LDPQ : LoadPairOffset<0b10, 1, FPR128, simm7s16, "ldp">;
1229 defm LDPSW : LoadPairOffset<0b01, 0, GPR64, simm7s4, "ldpsw">;
1231 // Pair (pre-indexed)
1232 def LDPWpre : LoadPairPreIdx<0b00, 0, GPR32, simm7s4, "ldp">;
1233 def LDPXpre : LoadPairPreIdx<0b10, 0, GPR64, simm7s8, "ldp">;
1234 def LDPSpre : LoadPairPreIdx<0b00, 1, FPR32, simm7s4, "ldp">;
1235 def LDPDpre : LoadPairPreIdx<0b01, 1, FPR64, simm7s8, "ldp">;
1236 def LDPQpre : LoadPairPreIdx<0b10, 1, FPR128, simm7s16, "ldp">;
1238 def LDPSWpre : LoadPairPreIdx<0b01, 0, GPR64, simm7s4, "ldpsw">;
1240 // Pair (post-indexed)
1241 def LDPWpost : LoadPairPostIdx<0b00, 0, GPR32, simm7s4, "ldp">;
1242 def LDPXpost : LoadPairPostIdx<0b10, 0, GPR64, simm7s8, "ldp">;
1243 def LDPSpost : LoadPairPostIdx<0b00, 1, FPR32, simm7s4, "ldp">;
1244 def LDPDpost : LoadPairPostIdx<0b01, 1, FPR64, simm7s8, "ldp">;
1245 def LDPQpost : LoadPairPostIdx<0b10, 1, FPR128, simm7s16, "ldp">;
1247 def LDPSWpost : LoadPairPostIdx<0b01, 0, GPR64, simm7s4, "ldpsw">;
1250 // Pair (no allocate)
1251 defm LDNPW : LoadPairNoAlloc<0b00, 0, GPR32, simm7s4, "ldnp">;
1252 defm LDNPX : LoadPairNoAlloc<0b10, 0, GPR64, simm7s8, "ldnp">;
1253 defm LDNPS : LoadPairNoAlloc<0b00, 1, FPR32, simm7s4, "ldnp">;
1254 defm LDNPD : LoadPairNoAlloc<0b01, 1, FPR64, simm7s8, "ldnp">;
1255 defm LDNPQ : LoadPairNoAlloc<0b10, 1, FPR128, simm7s16, "ldnp">;
1258 // (register offset)
1262 defm LDRBB : Load8RO<0b00, 0, 0b01, GPR32, "ldrb", i32, zextloadi8>;
1263 defm LDRHH : Load16RO<0b01, 0, 0b01, GPR32, "ldrh", i32, zextloadi16>;
1264 defm LDRW : Load32RO<0b10, 0, 0b01, GPR32, "ldr", i32, load>;
1265 defm LDRX : Load64RO<0b11, 0, 0b01, GPR64, "ldr", i64, load>;
1268 defm LDRB : Load8RO<0b00, 1, 0b01, FPR8, "ldr", untyped, load>;
1269 defm LDRH : Load16RO<0b01, 1, 0b01, FPR16, "ldr", f16, load>;
1270 defm LDRS : Load32RO<0b10, 1, 0b01, FPR32, "ldr", f32, load>;
1271 defm LDRD : Load64RO<0b11, 1, 0b01, FPR64, "ldr", f64, load>;
1272 defm LDRQ : Load128RO<0b00, 1, 0b11, FPR128, "ldr", f128, load>;
1274 // Load sign-extended half-word
1275 defm LDRSHW : Load16RO<0b01, 0, 0b11, GPR32, "ldrsh", i32, sextloadi16>;
1276 defm LDRSHX : Load16RO<0b01, 0, 0b10, GPR64, "ldrsh", i64, sextloadi16>;
1278 // Load sign-extended byte
1279 defm LDRSBW : Load8RO<0b00, 0, 0b11, GPR32, "ldrsb", i32, sextloadi8>;
1280 defm LDRSBX : Load8RO<0b00, 0, 0b10, GPR64, "ldrsb", i64, sextloadi8>;
1282 // Load sign-extended word
1283 defm LDRSW : Load32RO<0b10, 0, 0b10, GPR64, "ldrsw", i64, sextloadi32>;
1286 defm PRFM : PrefetchRO<0b11, 0, 0b10, "prfm">;
1288 // For regular load, we do not have any alignment requirement.
1289 // Thus, it is safe to directly map the vector loads with interesting
1290 // addressing modes.
1291 // FIXME: We could do the same for bitconvert to floating point vectors.
1292 multiclass ScalToVecROLoadPat<ROAddrMode ro, SDPatternOperator loadop,
1293 ValueType ScalTy, ValueType VecTy,
1294 Instruction LOADW, Instruction LOADX,
1296 def : Pat<(VecTy (scalar_to_vector (ScalTy
1297 (loadop (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$offset))))),
1298 (INSERT_SUBREG (VecTy (IMPLICIT_DEF)),
1299 (LOADW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$offset),
1302 def : Pat<(VecTy (scalar_to_vector (ScalTy
1303 (loadop (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$offset))))),
1304 (INSERT_SUBREG (VecTy (IMPLICIT_DEF)),
1305 (LOADX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$offset),
1309 let AddedComplexity = 10 in {
1310 defm : ScalToVecROLoadPat<ro8, extloadi8, i32, v8i8, LDRBroW, LDRBroX, bsub>;
1311 defm : ScalToVecROLoadPat<ro8, extloadi8, i32, v16i8, LDRBroW, LDRBroX, bsub>;
1313 defm : ScalToVecROLoadPat<ro16, extloadi16, i32, v4i16, LDRHroW, LDRHroX, hsub>;
1314 defm : ScalToVecROLoadPat<ro16, extloadi16, i32, v8i16, LDRHroW, LDRHroX, hsub>;
1316 defm : ScalToVecROLoadPat<ro16, load, i32, v4f16, LDRHroW, LDRHroX, hsub>;
1317 defm : ScalToVecROLoadPat<ro16, load, i32, v8f16, LDRHroW, LDRHroX, hsub>;
1319 defm : ScalToVecROLoadPat<ro32, load, i32, v2i32, LDRSroW, LDRSroX, ssub>;
1320 defm : ScalToVecROLoadPat<ro32, load, i32, v4i32, LDRSroW, LDRSroX, ssub>;
1322 defm : ScalToVecROLoadPat<ro32, load, f32, v2f32, LDRSroW, LDRSroX, ssub>;
1323 defm : ScalToVecROLoadPat<ro32, load, f32, v4f32, LDRSroW, LDRSroX, ssub>;
1325 defm : ScalToVecROLoadPat<ro64, load, i64, v2i64, LDRDroW, LDRDroX, dsub>;
1327 defm : ScalToVecROLoadPat<ro64, load, f64, v2f64, LDRDroW, LDRDroX, dsub>;
1330 def : Pat <(v1i64 (scalar_to_vector (i64
1331 (load (ro_Windexed64 GPR64sp:$Rn, GPR32:$Rm,
1332 ro_Wextend64:$extend))))),
1333 (LDRDroW GPR64sp:$Rn, GPR32:$Rm, ro_Wextend64:$extend)>;
1335 def : Pat <(v1i64 (scalar_to_vector (i64
1336 (load (ro_Xindexed64 GPR64sp:$Rn, GPR64:$Rm,
1337 ro_Xextend64:$extend))))),
1338 (LDRDroX GPR64sp:$Rn, GPR64:$Rm, ro_Xextend64:$extend)>;
1341 // Match all load 64 bits width whose type is compatible with FPR64
1342 multiclass VecROLoadPat<ROAddrMode ro, ValueType VecTy,
1343 Instruction LOADW, Instruction LOADX> {
1345 def : Pat<(VecTy (load (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend))),
1346 (LOADW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
1348 def : Pat<(VecTy (load (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend))),
1349 (LOADX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
1352 let AddedComplexity = 10 in {
1353 let Predicates = [IsLE] in {
1354 // We must do vector loads with LD1 in big-endian.
1355 defm : VecROLoadPat<ro64, v2i32, LDRDroW, LDRDroX>;
1356 defm : VecROLoadPat<ro64, v2f32, LDRDroW, LDRDroX>;
1357 defm : VecROLoadPat<ro64, v8i8, LDRDroW, LDRDroX>;
1358 defm : VecROLoadPat<ro64, v4i16, LDRDroW, LDRDroX>;
1359 defm : VecROLoadPat<ro64, v4f16, LDRDroW, LDRDroX>;
1362 defm : VecROLoadPat<ro64, v1i64, LDRDroW, LDRDroX>;
1363 defm : VecROLoadPat<ro64, v1f64, LDRDroW, LDRDroX>;
1365 // Match all load 128 bits width whose type is compatible with FPR128
1366 let Predicates = [IsLE] in {
1367 // We must do vector loads with LD1 in big-endian.
1368 defm : VecROLoadPat<ro128, v2i64, LDRQroW, LDRQroX>;
1369 defm : VecROLoadPat<ro128, v2f64, LDRQroW, LDRQroX>;
1370 defm : VecROLoadPat<ro128, v4i32, LDRQroW, LDRQroX>;
1371 defm : VecROLoadPat<ro128, v4f32, LDRQroW, LDRQroX>;
1372 defm : VecROLoadPat<ro128, v8i16, LDRQroW, LDRQroX>;
1373 defm : VecROLoadPat<ro128, v8f16, LDRQroW, LDRQroX>;
1374 defm : VecROLoadPat<ro128, v16i8, LDRQroW, LDRQroX>;
1376 } // AddedComplexity = 10
1379 multiclass ExtLoadTo64ROPat<ROAddrMode ro, SDPatternOperator loadop,
1380 Instruction INSTW, Instruction INSTX> {
1381 def : Pat<(i64 (loadop (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend))),
1382 (SUBREG_TO_REG (i64 0),
1383 (INSTW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend),
1386 def : Pat<(i64 (loadop (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend))),
1387 (SUBREG_TO_REG (i64 0),
1388 (INSTX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend),
1392 let AddedComplexity = 10 in {
1393 defm : ExtLoadTo64ROPat<ro8, zextloadi8, LDRBBroW, LDRBBroX>;
1394 defm : ExtLoadTo64ROPat<ro16, zextloadi16, LDRHHroW, LDRHHroX>;
1395 defm : ExtLoadTo64ROPat<ro32, zextloadi32, LDRWroW, LDRWroX>;
1397 // zextloadi1 -> zextloadi8
1398 defm : ExtLoadTo64ROPat<ro8, zextloadi1, LDRBBroW, LDRBBroX>;
1400 // extload -> zextload
1401 defm : ExtLoadTo64ROPat<ro8, extloadi8, LDRBBroW, LDRBBroX>;
1402 defm : ExtLoadTo64ROPat<ro16, extloadi16, LDRHHroW, LDRHHroX>;
1403 defm : ExtLoadTo64ROPat<ro32, extloadi32, LDRWroW, LDRWroX>;
1405 // extloadi1 -> zextloadi8
1406 defm : ExtLoadTo64ROPat<ro8, extloadi1, LDRBBroW, LDRBBroX>;
1411 multiclass ExtLoadTo32ROPat<ROAddrMode ro, SDPatternOperator loadop,
1412 Instruction INSTW, Instruction INSTX> {
1413 def : Pat<(i32 (loadop (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend))),
1414 (INSTW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
1416 def : Pat<(i32 (loadop (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend))),
1417 (INSTX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
1421 let AddedComplexity = 10 in {
1422 // extload -> zextload
1423 defm : ExtLoadTo32ROPat<ro8, extloadi8, LDRBBroW, LDRBBroX>;
1424 defm : ExtLoadTo32ROPat<ro16, extloadi16, LDRHHroW, LDRHHroX>;
1425 defm : ExtLoadTo32ROPat<ro32, extloadi32, LDRWroW, LDRWroX>;
1427 // zextloadi1 -> zextloadi8
1428 defm : ExtLoadTo32ROPat<ro8, zextloadi1, LDRBBroW, LDRBBroX>;
1432 // (unsigned immediate)
1434 defm LDRX : LoadUI<0b11, 0, 0b01, GPR64, uimm12s8, "ldr",
1436 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)))]>;
1437 defm LDRW : LoadUI<0b10, 0, 0b01, GPR32, uimm12s4, "ldr",
1439 (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset)))]>;
1440 defm LDRB : LoadUI<0b00, 1, 0b01, FPR8, uimm12s1, "ldr",
1442 (load (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset)))]>;
1443 defm LDRH : LoadUI<0b01, 1, 0b01, FPR16, uimm12s2, "ldr",
1444 [(set (f16 FPR16:$Rt),
1445 (load (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset)))]>;
1446 defm LDRS : LoadUI<0b10, 1, 0b01, FPR32, uimm12s4, "ldr",
1447 [(set (f32 FPR32:$Rt),
1448 (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset)))]>;
1449 defm LDRD : LoadUI<0b11, 1, 0b01, FPR64, uimm12s8, "ldr",
1450 [(set (f64 FPR64:$Rt),
1451 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)))]>;
1452 defm LDRQ : LoadUI<0b00, 1, 0b11, FPR128, uimm12s16, "ldr",
1453 [(set (f128 FPR128:$Rt),
1454 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)))]>;
1456 // For regular load, we do not have any alignment requirement.
1457 // Thus, it is safe to directly map the vector loads with interesting
1458 // addressing modes.
1459 // FIXME: We could do the same for bitconvert to floating point vectors.
1460 def : Pat <(v8i8 (scalar_to_vector (i32
1461 (extloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))))),
1462 (INSERT_SUBREG (v8i8 (IMPLICIT_DEF)),
1463 (LDRBui GPR64sp:$Rn, uimm12s1:$offset), bsub)>;
1464 def : Pat <(v16i8 (scalar_to_vector (i32
1465 (extloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))))),
1466 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
1467 (LDRBui GPR64sp:$Rn, uimm12s1:$offset), bsub)>;
1468 def : Pat <(v4i16 (scalar_to_vector (i32
1469 (extloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))))),
1470 (INSERT_SUBREG (v4i16 (IMPLICIT_DEF)),
1471 (LDRHui GPR64sp:$Rn, uimm12s2:$offset), hsub)>;
1472 def : Pat <(v8i16 (scalar_to_vector (i32
1473 (extloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))))),
1474 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
1475 (LDRHui GPR64sp:$Rn, uimm12s2:$offset), hsub)>;
1476 def : Pat <(v2i32 (scalar_to_vector (i32
1477 (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))))),
1478 (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)),
1479 (LDRSui GPR64sp:$Rn, uimm12s4:$offset), ssub)>;
1480 def : Pat <(v4i32 (scalar_to_vector (i32
1481 (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))))),
1482 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
1483 (LDRSui GPR64sp:$Rn, uimm12s4:$offset), ssub)>;
1484 def : Pat <(v1i64 (scalar_to_vector (i64
1485 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))))),
1486 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1487 def : Pat <(v2i64 (scalar_to_vector (i64
1488 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))))),
1489 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
1490 (LDRDui GPR64sp:$Rn, uimm12s8:$offset), dsub)>;
1492 // Match all load 64 bits width whose type is compatible with FPR64
1493 let Predicates = [IsLE] in {
1494 // We must use LD1 to perform vector loads in big-endian.
1495 def : Pat<(v2f32 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1496 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1497 def : Pat<(v8i8 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1498 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1499 def : Pat<(v4i16 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1500 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1501 def : Pat<(v2i32 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1502 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1503 def : Pat<(v4f16 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1504 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1506 def : Pat<(v1f64 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1507 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1508 def : Pat<(v1i64 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1509 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1511 // Match all load 128 bits width whose type is compatible with FPR128
1512 let Predicates = [IsLE] in {
1513 // We must use LD1 to perform vector loads in big-endian.
1514 def : Pat<(v4f32 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1515 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1516 def : Pat<(v2f64 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1517 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1518 def : Pat<(v16i8 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1519 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1520 def : Pat<(v8i16 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1521 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1522 def : Pat<(v4i32 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1523 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1524 def : Pat<(v2i64 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1525 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1526 def : Pat<(v8f16 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1527 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1529 def : Pat<(f128 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1530 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1532 defm LDRHH : LoadUI<0b01, 0, 0b01, GPR32, uimm12s2, "ldrh",
1534 (zextloadi16 (am_indexed16 GPR64sp:$Rn,
1535 uimm12s2:$offset)))]>;
1536 defm LDRBB : LoadUI<0b00, 0, 0b01, GPR32, uimm12s1, "ldrb",
1538 (zextloadi8 (am_indexed8 GPR64sp:$Rn,
1539 uimm12s1:$offset)))]>;
1541 def : Pat<(i64 (zextloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1542 (SUBREG_TO_REG (i64 0), (LDRBBui GPR64sp:$Rn, uimm12s1:$offset), sub_32)>;
1543 def : Pat<(i64 (zextloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))),
1544 (SUBREG_TO_REG (i64 0), (LDRHHui GPR64sp:$Rn, uimm12s2:$offset), sub_32)>;
1546 // zextloadi1 -> zextloadi8
1547 def : Pat<(i32 (zextloadi1 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1548 (LDRBBui GPR64sp:$Rn, uimm12s1:$offset)>;
1549 def : Pat<(i64 (zextloadi1 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1550 (SUBREG_TO_REG (i64 0), (LDRBBui GPR64sp:$Rn, uimm12s1:$offset), sub_32)>;
1552 // extload -> zextload
1553 def : Pat<(i32 (extloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))),
1554 (LDRHHui GPR64sp:$Rn, uimm12s2:$offset)>;
1555 def : Pat<(i32 (extloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1556 (LDRBBui GPR64sp:$Rn, uimm12s1:$offset)>;
1557 def : Pat<(i32 (extloadi1 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1558 (LDRBBui GPR64sp:$Rn, uimm12s1:$offset)>;
1559 def : Pat<(i64 (extloadi32 (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))),
1560 (SUBREG_TO_REG (i64 0), (LDRWui GPR64sp:$Rn, uimm12s4:$offset), sub_32)>;
1561 def : Pat<(i64 (extloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))),
1562 (SUBREG_TO_REG (i64 0), (LDRHHui GPR64sp:$Rn, uimm12s2:$offset), sub_32)>;
1563 def : Pat<(i64 (extloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1564 (SUBREG_TO_REG (i64 0), (LDRBBui GPR64sp:$Rn, uimm12s1:$offset), sub_32)>;
1565 def : Pat<(i64 (extloadi1 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1566 (SUBREG_TO_REG (i64 0), (LDRBBui GPR64sp:$Rn, uimm12s1:$offset), sub_32)>;
1568 // load sign-extended half-word
1569 defm LDRSHW : LoadUI<0b01, 0, 0b11, GPR32, uimm12s2, "ldrsh",
1571 (sextloadi16 (am_indexed16 GPR64sp:$Rn,
1572 uimm12s2:$offset)))]>;
1573 defm LDRSHX : LoadUI<0b01, 0, 0b10, GPR64, uimm12s2, "ldrsh",
1575 (sextloadi16 (am_indexed16 GPR64sp:$Rn,
1576 uimm12s2:$offset)))]>;
1578 // load sign-extended byte
1579 defm LDRSBW : LoadUI<0b00, 0, 0b11, GPR32, uimm12s1, "ldrsb",
1581 (sextloadi8 (am_indexed8 GPR64sp:$Rn,
1582 uimm12s1:$offset)))]>;
1583 defm LDRSBX : LoadUI<0b00, 0, 0b10, GPR64, uimm12s1, "ldrsb",
1585 (sextloadi8 (am_indexed8 GPR64sp:$Rn,
1586 uimm12s1:$offset)))]>;
1588 // load sign-extended word
1589 defm LDRSW : LoadUI<0b10, 0, 0b10, GPR64, uimm12s4, "ldrsw",
1591 (sextloadi32 (am_indexed32 GPR64sp:$Rn,
1592 uimm12s4:$offset)))]>;
1594 // load zero-extended word
1595 def : Pat<(i64 (zextloadi32 (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))),
1596 (SUBREG_TO_REG (i64 0), (LDRWui GPR64sp:$Rn, uimm12s4:$offset), sub_32)>;
1599 def PRFMui : PrefetchUI<0b11, 0, 0b10, "prfm",
1600 [(AArch64Prefetch imm:$Rt,
1601 (am_indexed64 GPR64sp:$Rn,
1602 uimm12s8:$offset))]>;
1604 def : InstAlias<"prfm $Rt, [$Rn]", (PRFMui prfop:$Rt, GPR64sp:$Rn, 0)>;
1608 def LDRWl : LoadLiteral<0b00, 0, GPR32, "ldr">;
1609 def LDRXl : LoadLiteral<0b01, 0, GPR64, "ldr">;
1610 def LDRSl : LoadLiteral<0b00, 1, FPR32, "ldr">;
1611 def LDRDl : LoadLiteral<0b01, 1, FPR64, "ldr">;
1612 def LDRQl : LoadLiteral<0b10, 1, FPR128, "ldr">;
1614 // load sign-extended word
1615 def LDRSWl : LoadLiteral<0b10, 0, GPR64, "ldrsw">;
1618 def PRFMl : PrefetchLiteral<0b11, 0, "prfm", []>;
1619 // [(AArch64Prefetch imm:$Rt, tglobaladdr:$label)]>;
1622 // (unscaled immediate)
1623 defm LDURX : LoadUnscaled<0b11, 0, 0b01, GPR64, "ldur",
1625 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset)))]>;
1626 defm LDURW : LoadUnscaled<0b10, 0, 0b01, GPR32, "ldur",
1628 (load (am_unscaled32 GPR64sp:$Rn, simm9:$offset)))]>;
1629 defm LDURB : LoadUnscaled<0b00, 1, 0b01, FPR8, "ldur",
1631 (load (am_unscaled8 GPR64sp:$Rn, simm9:$offset)))]>;
1632 defm LDURH : LoadUnscaled<0b01, 1, 0b01, FPR16, "ldur",
1634 (load (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
1635 defm LDURS : LoadUnscaled<0b10, 1, 0b01, FPR32, "ldur",
1636 [(set (f32 FPR32:$Rt),
1637 (load (am_unscaled32 GPR64sp:$Rn, simm9:$offset)))]>;
1638 defm LDURD : LoadUnscaled<0b11, 1, 0b01, FPR64, "ldur",
1639 [(set (f64 FPR64:$Rt),
1640 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset)))]>;
1641 defm LDURQ : LoadUnscaled<0b00, 1, 0b11, FPR128, "ldur",
1642 [(set (f128 FPR128:$Rt),
1643 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset)))]>;
1646 : LoadUnscaled<0b01, 0, 0b01, GPR32, "ldurh",
1648 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
1650 : LoadUnscaled<0b00, 0, 0b01, GPR32, "ldurb",
1652 (zextloadi8 (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
1654 // Match all load 64 bits width whose type is compatible with FPR64
1655 let Predicates = [IsLE] in {
1656 def : Pat<(v2f32 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1657 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1658 def : Pat<(v2i32 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1659 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1660 def : Pat<(v4i16 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1661 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1662 def : Pat<(v8i8 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1663 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1664 def : Pat<(v4f16 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1665 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1667 def : Pat<(v1f64 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1668 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1669 def : Pat<(v1i64 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1670 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1672 // Match all load 128 bits width whose type is compatible with FPR128
1673 let Predicates = [IsLE] in {
1674 def : Pat<(v2f64 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1675 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1676 def : Pat<(v2i64 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1677 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1678 def : Pat<(v4f32 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1679 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1680 def : Pat<(v4i32 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1681 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1682 def : Pat<(v8i16 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1683 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1684 def : Pat<(v16i8 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1685 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1686 def : Pat<(v8f16 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1687 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1691 def : Pat<(i32 (extloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
1692 (LDURHHi GPR64sp:$Rn, simm9:$offset)>;
1693 def : Pat<(i32 (extloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1694 (LDURBBi GPR64sp:$Rn, simm9:$offset)>;
1695 def : Pat<(i32 (extloadi1 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1696 (LDURBBi GPR64sp:$Rn, simm9:$offset)>;
1697 def : Pat<(i64 (extloadi32 (am_unscaled32 GPR64sp:$Rn, simm9:$offset))),
1698 (SUBREG_TO_REG (i64 0), (LDURWi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1699 def : Pat<(i64 (extloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
1700 (SUBREG_TO_REG (i64 0), (LDURHHi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1701 def : Pat<(i64 (extloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1702 (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1703 def : Pat<(i64 (extloadi1 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1704 (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1706 def : Pat<(i32 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
1707 (LDURHHi GPR64sp:$Rn, simm9:$offset)>;
1708 def : Pat<(i32 (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1709 (LDURBBi GPR64sp:$Rn, simm9:$offset)>;
1710 def : Pat<(i32 (zextloadi1 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1711 (LDURBBi GPR64sp:$Rn, simm9:$offset)>;
1712 def : Pat<(i64 (zextloadi32 (am_unscaled32 GPR64sp:$Rn, simm9:$offset))),
1713 (SUBREG_TO_REG (i64 0), (LDURWi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1714 def : Pat<(i64 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
1715 (SUBREG_TO_REG (i64 0), (LDURHHi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1716 def : Pat<(i64 (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1717 (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1718 def : Pat<(i64 (zextloadi1 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1719 (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1723 // LDR mnemonics fall back to LDUR for negative or unaligned offsets.
1725 // Define new assembler match classes as we want to only match these when
1726 // the don't otherwise match the scaled addressing mode for LDR/STR. Don't
1727 // associate a DiagnosticType either, as we want the diagnostic for the
1728 // canonical form (the scaled operand) to take precedence.
1729 class SImm9OffsetOperand<int Width> : AsmOperandClass {
1730 let Name = "SImm9OffsetFB" # Width;
1731 let PredicateMethod = "isSImm9OffsetFB<" # Width # ">";
1732 let RenderMethod = "addImmOperands";
1735 def SImm9OffsetFB8Operand : SImm9OffsetOperand<8>;
1736 def SImm9OffsetFB16Operand : SImm9OffsetOperand<16>;
1737 def SImm9OffsetFB32Operand : SImm9OffsetOperand<32>;
1738 def SImm9OffsetFB64Operand : SImm9OffsetOperand<64>;
1739 def SImm9OffsetFB128Operand : SImm9OffsetOperand<128>;
1741 def simm9_offset_fb8 : Operand<i64> {
1742 let ParserMatchClass = SImm9OffsetFB8Operand;
1744 def simm9_offset_fb16 : Operand<i64> {
1745 let ParserMatchClass = SImm9OffsetFB16Operand;
1747 def simm9_offset_fb32 : Operand<i64> {
1748 let ParserMatchClass = SImm9OffsetFB32Operand;
1750 def simm9_offset_fb64 : Operand<i64> {
1751 let ParserMatchClass = SImm9OffsetFB64Operand;
1753 def simm9_offset_fb128 : Operand<i64> {
1754 let ParserMatchClass = SImm9OffsetFB128Operand;
1757 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1758 (LDURXi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb64:$offset), 0>;
1759 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1760 (LDURWi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
1761 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1762 (LDURBi FPR8:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
1763 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1764 (LDURHi FPR16:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
1765 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1766 (LDURSi FPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
1767 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1768 (LDURDi FPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb64:$offset), 0>;
1769 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1770 (LDURQi FPR128:$Rt, GPR64sp:$Rn, simm9_offset_fb128:$offset), 0>;
1773 def : Pat<(i64 (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1774 (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1775 def : Pat<(i64 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
1776 (SUBREG_TO_REG (i64 0), (LDURHHi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1778 // load sign-extended half-word
1780 : LoadUnscaled<0b01, 0, 0b11, GPR32, "ldursh",
1782 (sextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
1784 : LoadUnscaled<0b01, 0, 0b10, GPR64, "ldursh",
1786 (sextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
1788 // load sign-extended byte
1790 : LoadUnscaled<0b00, 0, 0b11, GPR32, "ldursb",
1792 (sextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset)))]>;
1794 : LoadUnscaled<0b00, 0, 0b10, GPR64, "ldursb",
1796 (sextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset)))]>;
1798 // load sign-extended word
1800 : LoadUnscaled<0b10, 0, 0b10, GPR64, "ldursw",
1802 (sextloadi32 (am_unscaled32 GPR64sp:$Rn, simm9:$offset)))]>;
1804 // zero and sign extending aliases from generic LDR* mnemonics to LDUR*.
1805 def : InstAlias<"ldrb $Rt, [$Rn, $offset]",
1806 (LDURBBi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
1807 def : InstAlias<"ldrh $Rt, [$Rn, $offset]",
1808 (LDURHHi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
1809 def : InstAlias<"ldrsb $Rt, [$Rn, $offset]",
1810 (LDURSBWi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
1811 def : InstAlias<"ldrsb $Rt, [$Rn, $offset]",
1812 (LDURSBXi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
1813 def : InstAlias<"ldrsh $Rt, [$Rn, $offset]",
1814 (LDURSHWi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
1815 def : InstAlias<"ldrsh $Rt, [$Rn, $offset]",
1816 (LDURSHXi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
1817 def : InstAlias<"ldrsw $Rt, [$Rn, $offset]",
1818 (LDURSWi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
1821 defm PRFUM : PrefetchUnscaled<0b11, 0, 0b10, "prfum",
1822 [(AArch64Prefetch imm:$Rt,
1823 (am_unscaled64 GPR64sp:$Rn, simm9:$offset))]>;
1826 // (unscaled immediate, unprivileged)
1827 defm LDTRX : LoadUnprivileged<0b11, 0, 0b01, GPR64, "ldtr">;
1828 defm LDTRW : LoadUnprivileged<0b10, 0, 0b01, GPR32, "ldtr">;
1830 defm LDTRH : LoadUnprivileged<0b01, 0, 0b01, GPR32, "ldtrh">;
1831 defm LDTRB : LoadUnprivileged<0b00, 0, 0b01, GPR32, "ldtrb">;
1833 // load sign-extended half-word
1834 defm LDTRSHW : LoadUnprivileged<0b01, 0, 0b11, GPR32, "ldtrsh">;
1835 defm LDTRSHX : LoadUnprivileged<0b01, 0, 0b10, GPR64, "ldtrsh">;
1837 // load sign-extended byte
1838 defm LDTRSBW : LoadUnprivileged<0b00, 0, 0b11, GPR32, "ldtrsb">;
1839 defm LDTRSBX : LoadUnprivileged<0b00, 0, 0b10, GPR64, "ldtrsb">;
1841 // load sign-extended word
1842 defm LDTRSW : LoadUnprivileged<0b10, 0, 0b10, GPR64, "ldtrsw">;
1845 // (immediate pre-indexed)
1846 def LDRWpre : LoadPreIdx<0b10, 0, 0b01, GPR32, "ldr">;
1847 def LDRXpre : LoadPreIdx<0b11, 0, 0b01, GPR64, "ldr">;
1848 def LDRBpre : LoadPreIdx<0b00, 1, 0b01, FPR8, "ldr">;
1849 def LDRHpre : LoadPreIdx<0b01, 1, 0b01, FPR16, "ldr">;
1850 def LDRSpre : LoadPreIdx<0b10, 1, 0b01, FPR32, "ldr">;
1851 def LDRDpre : LoadPreIdx<0b11, 1, 0b01, FPR64, "ldr">;
1852 def LDRQpre : LoadPreIdx<0b00, 1, 0b11, FPR128, "ldr">;
1854 // load sign-extended half-word
1855 def LDRSHWpre : LoadPreIdx<0b01, 0, 0b11, GPR32, "ldrsh">;
1856 def LDRSHXpre : LoadPreIdx<0b01, 0, 0b10, GPR64, "ldrsh">;
1858 // load sign-extended byte
1859 def LDRSBWpre : LoadPreIdx<0b00, 0, 0b11, GPR32, "ldrsb">;
1860 def LDRSBXpre : LoadPreIdx<0b00, 0, 0b10, GPR64, "ldrsb">;
1862 // load zero-extended byte
1863 def LDRBBpre : LoadPreIdx<0b00, 0, 0b01, GPR32, "ldrb">;
1864 def LDRHHpre : LoadPreIdx<0b01, 0, 0b01, GPR32, "ldrh">;
1866 // load sign-extended word
1867 def LDRSWpre : LoadPreIdx<0b10, 0, 0b10, GPR64, "ldrsw">;
1870 // (immediate post-indexed)
1871 def LDRWpost : LoadPostIdx<0b10, 0, 0b01, GPR32, "ldr">;
1872 def LDRXpost : LoadPostIdx<0b11, 0, 0b01, GPR64, "ldr">;
1873 def LDRBpost : LoadPostIdx<0b00, 1, 0b01, FPR8, "ldr">;
1874 def LDRHpost : LoadPostIdx<0b01, 1, 0b01, FPR16, "ldr">;
1875 def LDRSpost : LoadPostIdx<0b10, 1, 0b01, FPR32, "ldr">;
1876 def LDRDpost : LoadPostIdx<0b11, 1, 0b01, FPR64, "ldr">;
1877 def LDRQpost : LoadPostIdx<0b00, 1, 0b11, FPR128, "ldr">;
1879 // load sign-extended half-word
1880 def LDRSHWpost : LoadPostIdx<0b01, 0, 0b11, GPR32, "ldrsh">;
1881 def LDRSHXpost : LoadPostIdx<0b01, 0, 0b10, GPR64, "ldrsh">;
1883 // load sign-extended byte
1884 def LDRSBWpost : LoadPostIdx<0b00, 0, 0b11, GPR32, "ldrsb">;
1885 def LDRSBXpost : LoadPostIdx<0b00, 0, 0b10, GPR64, "ldrsb">;
1887 // load zero-extended byte
1888 def LDRBBpost : LoadPostIdx<0b00, 0, 0b01, GPR32, "ldrb">;
1889 def LDRHHpost : LoadPostIdx<0b01, 0, 0b01, GPR32, "ldrh">;
1891 // load sign-extended word
1892 def LDRSWpost : LoadPostIdx<0b10, 0, 0b10, GPR64, "ldrsw">;
1894 //===----------------------------------------------------------------------===//
1895 // Store instructions.
1896 //===----------------------------------------------------------------------===//
1898 // Pair (indexed, offset)
1899 // FIXME: Use dedicated range-checked addressing mode operand here.
1900 defm STPW : StorePairOffset<0b00, 0, GPR32, simm7s4, "stp">;
1901 defm STPX : StorePairOffset<0b10, 0, GPR64, simm7s8, "stp">;
1902 defm STPS : StorePairOffset<0b00, 1, FPR32, simm7s4, "stp">;
1903 defm STPD : StorePairOffset<0b01, 1, FPR64, simm7s8, "stp">;
1904 defm STPQ : StorePairOffset<0b10, 1, FPR128, simm7s16, "stp">;
1906 // Pair (pre-indexed)
1907 def STPWpre : StorePairPreIdx<0b00, 0, GPR32, simm7s4, "stp">;
1908 def STPXpre : StorePairPreIdx<0b10, 0, GPR64, simm7s8, "stp">;
1909 def STPSpre : StorePairPreIdx<0b00, 1, FPR32, simm7s4, "stp">;
1910 def STPDpre : StorePairPreIdx<0b01, 1, FPR64, simm7s8, "stp">;
1911 def STPQpre : StorePairPreIdx<0b10, 1, FPR128, simm7s16, "stp">;
1913 // Pair (pre-indexed)
1914 def STPWpost : StorePairPostIdx<0b00, 0, GPR32, simm7s4, "stp">;
1915 def STPXpost : StorePairPostIdx<0b10, 0, GPR64, simm7s8, "stp">;
1916 def STPSpost : StorePairPostIdx<0b00, 1, FPR32, simm7s4, "stp">;
1917 def STPDpost : StorePairPostIdx<0b01, 1, FPR64, simm7s8, "stp">;
1918 def STPQpost : StorePairPostIdx<0b10, 1, FPR128, simm7s16, "stp">;
1920 // Pair (no allocate)
1921 defm STNPW : StorePairNoAlloc<0b00, 0, GPR32, simm7s4, "stnp">;
1922 defm STNPX : StorePairNoAlloc<0b10, 0, GPR64, simm7s8, "stnp">;
1923 defm STNPS : StorePairNoAlloc<0b00, 1, FPR32, simm7s4, "stnp">;
1924 defm STNPD : StorePairNoAlloc<0b01, 1, FPR64, simm7s8, "stnp">;
1925 defm STNPQ : StorePairNoAlloc<0b10, 1, FPR128, simm7s16, "stnp">;
1928 // (Register offset)
1931 defm STRBB : Store8RO< 0b00, 0, 0b00, GPR32, "strb", i32, truncstorei8>;
1932 defm STRHH : Store16RO<0b01, 0, 0b00, GPR32, "strh", i32, truncstorei16>;
1933 defm STRW : Store32RO<0b10, 0, 0b00, GPR32, "str", i32, store>;
1934 defm STRX : Store64RO<0b11, 0, 0b00, GPR64, "str", i64, store>;
1938 defm STRB : Store8RO< 0b00, 1, 0b00, FPR8, "str", untyped, store>;
1939 defm STRH : Store16RO<0b01, 1, 0b00, FPR16, "str", f16, store>;
1940 defm STRS : Store32RO<0b10, 1, 0b00, FPR32, "str", f32, store>;
1941 defm STRD : Store64RO<0b11, 1, 0b00, FPR64, "str", f64, store>;
1942 defm STRQ : Store128RO<0b00, 1, 0b10, FPR128, "str", f128, store>;
1944 multiclass TruncStoreFrom64ROPat<ROAddrMode ro, SDPatternOperator storeop,
1945 Instruction STRW, Instruction STRX> {
1947 def : Pat<(storeop GPR64:$Rt,
1948 (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)),
1949 (STRW (EXTRACT_SUBREG GPR64:$Rt, sub_32),
1950 GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
1952 def : Pat<(storeop GPR64:$Rt,
1953 (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)),
1954 (STRX (EXTRACT_SUBREG GPR64:$Rt, sub_32),
1955 GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
1958 let AddedComplexity = 10 in {
1960 defm : TruncStoreFrom64ROPat<ro8, truncstorei8, STRBBroW, STRBBroX>;
1961 defm : TruncStoreFrom64ROPat<ro16, truncstorei16, STRHHroW, STRHHroX>;
1962 defm : TruncStoreFrom64ROPat<ro32, truncstorei32, STRWroW, STRWroX>;
1965 multiclass VecROStorePat<ROAddrMode ro, ValueType VecTy, RegisterClass FPR,
1966 Instruction STRW, Instruction STRX> {
1967 def : Pat<(store (VecTy FPR:$Rt),
1968 (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)),
1969 (STRW FPR:$Rt, GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
1971 def : Pat<(store (VecTy FPR:$Rt),
1972 (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)),
1973 (STRX FPR:$Rt, GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
1976 let AddedComplexity = 10 in {
1977 // Match all store 64 bits width whose type is compatible with FPR64
1978 let Predicates = [IsLE] in {
1979 // We must use ST1 to store vectors in big-endian.
1980 defm : VecROStorePat<ro64, v2i32, FPR64, STRDroW, STRDroX>;
1981 defm : VecROStorePat<ro64, v2f32, FPR64, STRDroW, STRDroX>;
1982 defm : VecROStorePat<ro64, v4i16, FPR64, STRDroW, STRDroX>;
1983 defm : VecROStorePat<ro64, v8i8, FPR64, STRDroW, STRDroX>;
1984 defm : VecROStorePat<ro64, v4f16, FPR64, STRDroW, STRDroX>;
1987 defm : VecROStorePat<ro64, v1i64, FPR64, STRDroW, STRDroX>;
1988 defm : VecROStorePat<ro64, v1f64, FPR64, STRDroW, STRDroX>;
1990 // Match all store 128 bits width whose type is compatible with FPR128
1991 let Predicates = [IsLE] in {
1992 // We must use ST1 to store vectors in big-endian.
1993 defm : VecROStorePat<ro128, v2i64, FPR128, STRQroW, STRQroX>;
1994 defm : VecROStorePat<ro128, v2f64, FPR128, STRQroW, STRQroX>;
1995 defm : VecROStorePat<ro128, v4i32, FPR128, STRQroW, STRQroX>;
1996 defm : VecROStorePat<ro128, v4f32, FPR128, STRQroW, STRQroX>;
1997 defm : VecROStorePat<ro128, v8i16, FPR128, STRQroW, STRQroX>;
1998 defm : VecROStorePat<ro128, v16i8, FPR128, STRQroW, STRQroX>;
1999 defm : VecROStorePat<ro128, v8f16, FPR128, STRQroW, STRQroX>;
2001 } // AddedComplexity = 10
2003 // Match stores from lane 0 to the appropriate subreg's store.
2004 multiclass VecROStoreLane0Pat<ROAddrMode ro, SDPatternOperator storeop,
2005 ValueType VecTy, ValueType STy,
2006 SubRegIndex SubRegIdx,
2007 Instruction STRW, Instruction STRX> {
2009 def : Pat<(storeop (STy (vector_extract (VecTy VecListOne128:$Vt), 0)),
2010 (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)),
2011 (STRW (EXTRACT_SUBREG VecListOne128:$Vt, SubRegIdx),
2012 GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
2014 def : Pat<(storeop (STy (vector_extract (VecTy VecListOne128:$Vt), 0)),
2015 (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)),
2016 (STRX (EXTRACT_SUBREG VecListOne128:$Vt, SubRegIdx),
2017 GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
2020 let AddedComplexity = 19 in {
2021 defm : VecROStoreLane0Pat<ro16, truncstorei16, v8i16, i32, hsub, STRHroW, STRHroX>;
2022 defm : VecROStoreLane0Pat<ro16, store , v8i16, i16, hsub, STRHroW, STRHroX>;
2023 defm : VecROStoreLane0Pat<ro32, truncstorei32, v4i32, i32, ssub, STRSroW, STRSroX>;
2024 defm : VecROStoreLane0Pat<ro32, store , v4i32, i32, ssub, STRSroW, STRSroX>;
2025 defm : VecROStoreLane0Pat<ro32, store , v4f32, f32, ssub, STRSroW, STRSroX>;
2026 defm : VecROStoreLane0Pat<ro64, store , v2i64, i64, dsub, STRDroW, STRDroX>;
2027 defm : VecROStoreLane0Pat<ro64, store , v2f64, f64, dsub, STRDroW, STRDroX>;
2031 // (unsigned immediate)
2032 defm STRX : StoreUI<0b11, 0, 0b00, GPR64, uimm12s8, "str",
2034 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))]>;
2035 defm STRW : StoreUI<0b10, 0, 0b00, GPR32, uimm12s4, "str",
2037 (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))]>;
2038 defm STRB : StoreUI<0b00, 1, 0b00, FPR8, uimm12s1, "str",
2040 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))]>;
2041 defm STRH : StoreUI<0b01, 1, 0b00, FPR16, uimm12s2, "str",
2042 [(store (f16 FPR16:$Rt),
2043 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))]>;
2044 defm STRS : StoreUI<0b10, 1, 0b00, FPR32, uimm12s4, "str",
2045 [(store (f32 FPR32:$Rt),
2046 (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))]>;
2047 defm STRD : StoreUI<0b11, 1, 0b00, FPR64, uimm12s8, "str",
2048 [(store (f64 FPR64:$Rt),
2049 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))]>;
2050 defm STRQ : StoreUI<0b00, 1, 0b10, FPR128, uimm12s16, "str", []>;
2052 defm STRHH : StoreUI<0b01, 0, 0b00, GPR32, uimm12s2, "strh",
2053 [(truncstorei16 GPR32:$Rt,
2054 (am_indexed16 GPR64sp:$Rn,
2055 uimm12s2:$offset))]>;
2056 defm STRBB : StoreUI<0b00, 0, 0b00, GPR32, uimm12s1, "strb",
2057 [(truncstorei8 GPR32:$Rt,
2058 (am_indexed8 GPR64sp:$Rn,
2059 uimm12s1:$offset))]>;
2061 // Match all store 64 bits width whose type is compatible with FPR64
2062 let AddedComplexity = 10 in {
2063 let Predicates = [IsLE] in {
2064 // We must use ST1 to store vectors in big-endian.
2065 def : Pat<(store (v2f32 FPR64:$Rt),
2066 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
2067 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
2068 def : Pat<(store (v8i8 FPR64:$Rt),
2069 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
2070 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
2071 def : Pat<(store (v4i16 FPR64:$Rt),
2072 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
2073 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
2074 def : Pat<(store (v2i32 FPR64:$Rt),
2075 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
2076 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
2077 def : Pat<(store (v4f16 FPR64:$Rt),
2078 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
2079 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
2081 def : Pat<(store (v1f64 FPR64:$Rt),
2082 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
2083 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
2084 def : Pat<(store (v1i64 FPR64:$Rt),
2085 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
2086 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
2088 // Match all store 128 bits width whose type is compatible with FPR128
2089 let Predicates = [IsLE] in {
2090 // We must use ST1 to store vectors in big-endian.
2091 def : Pat<(store (v4f32 FPR128:$Rt),
2092 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
2093 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
2094 def : Pat<(store (v2f64 FPR128:$Rt),
2095 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
2096 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
2097 def : Pat<(store (v16i8 FPR128:$Rt),
2098 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
2099 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
2100 def : Pat<(store (v8i16 FPR128:$Rt),
2101 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
2102 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
2103 def : Pat<(store (v4i32 FPR128:$Rt),
2104 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
2105 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
2106 def : Pat<(store (v2i64 FPR128:$Rt),
2107 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
2108 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
2109 def : Pat<(store (v8f16 FPR128:$Rt),
2110 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
2111 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
2113 def : Pat<(store (f128 FPR128:$Rt),
2114 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
2115 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
2118 def : Pat<(truncstorei32 GPR64:$Rt,
2119 (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset)),
2120 (STRWui (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, uimm12s4:$offset)>;
2121 def : Pat<(truncstorei16 GPR64:$Rt,
2122 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset)),
2123 (STRHHui (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, uimm12s2:$offset)>;
2124 def : Pat<(truncstorei8 GPR64:$Rt, (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset)),
2125 (STRBBui (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, uimm12s1:$offset)>;
2127 } // AddedComplexity = 10
2130 // (unscaled immediate)
2131 defm STURX : StoreUnscaled<0b11, 0, 0b00, GPR64, "stur",
2133 (am_unscaled64 GPR64sp:$Rn, simm9:$offset))]>;
2134 defm STURW : StoreUnscaled<0b10, 0, 0b00, GPR32, "stur",
2136 (am_unscaled32 GPR64sp:$Rn, simm9:$offset))]>;
2137 defm STURB : StoreUnscaled<0b00, 1, 0b00, FPR8, "stur",
2139 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))]>;
2140 defm STURH : StoreUnscaled<0b01, 1, 0b00, FPR16, "stur",
2141 [(store (f16 FPR16:$Rt),
2142 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))]>;
2143 defm STURS : StoreUnscaled<0b10, 1, 0b00, FPR32, "stur",
2144 [(store (f32 FPR32:$Rt),
2145 (am_unscaled32 GPR64sp:$Rn, simm9:$offset))]>;
2146 defm STURD : StoreUnscaled<0b11, 1, 0b00, FPR64, "stur",
2147 [(store (f64 FPR64:$Rt),
2148 (am_unscaled64 GPR64sp:$Rn, simm9:$offset))]>;
2149 defm STURQ : StoreUnscaled<0b00, 1, 0b10, FPR128, "stur",
2150 [(store (f128 FPR128:$Rt),
2151 (am_unscaled128 GPR64sp:$Rn, simm9:$offset))]>;
2152 defm STURHH : StoreUnscaled<0b01, 0, 0b00, GPR32, "sturh",
2153 [(truncstorei16 GPR32:$Rt,
2154 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))]>;
2155 defm STURBB : StoreUnscaled<0b00, 0, 0b00, GPR32, "sturb",
2156 [(truncstorei8 GPR32:$Rt,
2157 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))]>;
2159 // Match all store 64 bits width whose type is compatible with FPR64
2160 let Predicates = [IsLE] in {
2161 // We must use ST1 to store vectors in big-endian.
2162 def : Pat<(store (v2f32 FPR64:$Rt),
2163 (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2164 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2165 def : Pat<(store (v8i8 FPR64:$Rt),
2166 (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2167 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2168 def : Pat<(store (v4i16 FPR64:$Rt),
2169 (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2170 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2171 def : Pat<(store (v2i32 FPR64:$Rt),
2172 (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2173 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2174 def : Pat<(store (v4f16 FPR64:$Rt),
2175 (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2176 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2178 def : Pat<(store (v1f64 FPR64:$Rt), (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2179 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2180 def : Pat<(store (v1i64 FPR64:$Rt), (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2181 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2183 // Match all store 128 bits width whose type is compatible with FPR128
2184 let Predicates = [IsLE] in {
2185 // We must use ST1 to store vectors in big-endian.
2186 def : Pat<(store (v4f32 FPR128:$Rt),
2187 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2188 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2189 def : Pat<(store (v2f64 FPR128:$Rt),
2190 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2191 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2192 def : Pat<(store (v16i8 FPR128:$Rt),
2193 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2194 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2195 def : Pat<(store (v8i16 FPR128:$Rt),
2196 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2197 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2198 def : Pat<(store (v4i32 FPR128:$Rt),
2199 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2200 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2201 def : Pat<(store (v2i64 FPR128:$Rt),
2202 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2203 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2204 def : Pat<(store (v2f64 FPR128:$Rt),
2205 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2206 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2207 def : Pat<(store (v8f16 FPR128:$Rt),
2208 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2209 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2212 // unscaled i64 truncating stores
2213 def : Pat<(truncstorei32 GPR64:$Rt, (am_unscaled32 GPR64sp:$Rn, simm9:$offset)),
2214 (STURWi (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, simm9:$offset)>;
2215 def : Pat<(truncstorei16 GPR64:$Rt, (am_unscaled16 GPR64sp:$Rn, simm9:$offset)),
2216 (STURHHi (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, simm9:$offset)>;
2217 def : Pat<(truncstorei8 GPR64:$Rt, (am_unscaled8 GPR64sp:$Rn, simm9:$offset)),
2218 (STURBBi (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, simm9:$offset)>;
2221 // STR mnemonics fall back to STUR for negative or unaligned offsets.
2222 def : InstAlias<"str $Rt, [$Rn, $offset]",
2223 (STURXi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb64:$offset), 0>;
2224 def : InstAlias<"str $Rt, [$Rn, $offset]",
2225 (STURWi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
2226 def : InstAlias<"str $Rt, [$Rn, $offset]",
2227 (STURBi FPR8:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
2228 def : InstAlias<"str $Rt, [$Rn, $offset]",
2229 (STURHi FPR16:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
2230 def : InstAlias<"str $Rt, [$Rn, $offset]",
2231 (STURSi FPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
2232 def : InstAlias<"str $Rt, [$Rn, $offset]",
2233 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb64:$offset), 0>;
2234 def : InstAlias<"str $Rt, [$Rn, $offset]",
2235 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9_offset_fb128:$offset), 0>;
2237 def : InstAlias<"strb $Rt, [$Rn, $offset]",
2238 (STURBBi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
2239 def : InstAlias<"strh $Rt, [$Rn, $offset]",
2240 (STURHHi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
2243 // (unscaled immediate, unprivileged)
2244 defm STTRW : StoreUnprivileged<0b10, 0, 0b00, GPR32, "sttr">;
2245 defm STTRX : StoreUnprivileged<0b11, 0, 0b00, GPR64, "sttr">;
2247 defm STTRH : StoreUnprivileged<0b01, 0, 0b00, GPR32, "sttrh">;
2248 defm STTRB : StoreUnprivileged<0b00, 0, 0b00, GPR32, "sttrb">;
2251 // (immediate pre-indexed)
2252 def STRWpre : StorePreIdx<0b10, 0, 0b00, GPR32, "str", pre_store, i32>;
2253 def STRXpre : StorePreIdx<0b11, 0, 0b00, GPR64, "str", pre_store, i64>;
2254 def STRBpre : StorePreIdx<0b00, 1, 0b00, FPR8, "str", pre_store, untyped>;
2255 def STRHpre : StorePreIdx<0b01, 1, 0b00, FPR16, "str", pre_store, f16>;
2256 def STRSpre : StorePreIdx<0b10, 1, 0b00, FPR32, "str", pre_store, f32>;
2257 def STRDpre : StorePreIdx<0b11, 1, 0b00, FPR64, "str", pre_store, f64>;
2258 def STRQpre : StorePreIdx<0b00, 1, 0b10, FPR128, "str", pre_store, f128>;
2260 def STRBBpre : StorePreIdx<0b00, 0, 0b00, GPR32, "strb", pre_truncsti8, i32>;
2261 def STRHHpre : StorePreIdx<0b01, 0, 0b00, GPR32, "strh", pre_truncsti16, i32>;
2264 def : Pat<(pre_truncsti32 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
2265 (STRWpre (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
2267 def : Pat<(pre_truncsti16 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
2268 (STRHHpre (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
2270 def : Pat<(pre_truncsti8 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
2271 (STRBBpre (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
2274 def : Pat<(pre_store (v8i8 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2275 (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2276 def : Pat<(pre_store (v4i16 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2277 (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2278 def : Pat<(pre_store (v2i32 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2279 (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2280 def : Pat<(pre_store (v2f32 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2281 (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2282 def : Pat<(pre_store (v1i64 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2283 (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2284 def : Pat<(pre_store (v1f64 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2285 (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2286 def : Pat<(pre_store (v4f16 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2287 (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2289 def : Pat<(pre_store (v16i8 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2290 (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2291 def : Pat<(pre_store (v8i16 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2292 (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2293 def : Pat<(pre_store (v4i32 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2294 (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2295 def : Pat<(pre_store (v4f32 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2296 (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2297 def : Pat<(pre_store (v2i64 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2298 (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2299 def : Pat<(pre_store (v2f64 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2300 (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2301 def : Pat<(pre_store (v8f16 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2302 (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2305 // (immediate post-indexed)
2306 def STRWpost : StorePostIdx<0b10, 0, 0b00, GPR32, "str", post_store, i32>;
2307 def STRXpost : StorePostIdx<0b11, 0, 0b00, GPR64, "str", post_store, i64>;
2308 def STRBpost : StorePostIdx<0b00, 1, 0b00, FPR8, "str", post_store, untyped>;
2309 def STRHpost : StorePostIdx<0b01, 1, 0b00, FPR16, "str", post_store, f16>;
2310 def STRSpost : StorePostIdx<0b10, 1, 0b00, FPR32, "str", post_store, f32>;
2311 def STRDpost : StorePostIdx<0b11, 1, 0b00, FPR64, "str", post_store, f64>;
2312 def STRQpost : StorePostIdx<0b00, 1, 0b10, FPR128, "str", post_store, f128>;
2314 def STRBBpost : StorePostIdx<0b00, 0, 0b00, GPR32, "strb", post_truncsti8, i32>;
2315 def STRHHpost : StorePostIdx<0b01, 0, 0b00, GPR32, "strh", post_truncsti16, i32>;
2318 def : Pat<(post_truncsti32 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
2319 (STRWpost (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
2321 def : Pat<(post_truncsti16 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
2322 (STRHHpost (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
2324 def : Pat<(post_truncsti8 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
2325 (STRBBpost (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
2328 def : Pat<(post_store (v8i8 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2329 (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2330 def : Pat<(post_store (v4i16 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2331 (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2332 def : Pat<(post_store (v2i32 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2333 (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2334 def : Pat<(post_store (v2f32 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2335 (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2336 def : Pat<(post_store (v1i64 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2337 (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2338 def : Pat<(post_store (v1f64 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2339 (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2340 def : Pat<(post_store (v4f16 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2341 (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2343 def : Pat<(post_store (v16i8 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2344 (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2345 def : Pat<(post_store (v8i16 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2346 (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2347 def : Pat<(post_store (v4i32 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2348 (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2349 def : Pat<(post_store (v4f32 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2350 (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2351 def : Pat<(post_store (v2i64 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2352 (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2353 def : Pat<(post_store (v2f64 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2354 (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2355 def : Pat<(post_store (v8f16 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2356 (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2358 //===----------------------------------------------------------------------===//
2359 // Load/store exclusive instructions.
2360 //===----------------------------------------------------------------------===//
2362 def LDARW : LoadAcquire <0b10, 1, 1, 0, 1, GPR32, "ldar">;
2363 def LDARX : LoadAcquire <0b11, 1, 1, 0, 1, GPR64, "ldar">;
2364 def LDARB : LoadAcquire <0b00, 1, 1, 0, 1, GPR32, "ldarb">;
2365 def LDARH : LoadAcquire <0b01, 1, 1, 0, 1, GPR32, "ldarh">;
2367 def LDAXRW : LoadExclusive <0b10, 0, 1, 0, 1, GPR32, "ldaxr">;
2368 def LDAXRX : LoadExclusive <0b11, 0, 1, 0, 1, GPR64, "ldaxr">;
2369 def LDAXRB : LoadExclusive <0b00, 0, 1, 0, 1, GPR32, "ldaxrb">;
2370 def LDAXRH : LoadExclusive <0b01, 0, 1, 0, 1, GPR32, "ldaxrh">;
2372 def LDXRW : LoadExclusive <0b10, 0, 1, 0, 0, GPR32, "ldxr">;
2373 def LDXRX : LoadExclusive <0b11, 0, 1, 0, 0, GPR64, "ldxr">;
2374 def LDXRB : LoadExclusive <0b00, 0, 1, 0, 0, GPR32, "ldxrb">;
2375 def LDXRH : LoadExclusive <0b01, 0, 1, 0, 0, GPR32, "ldxrh">;
2377 def STLRW : StoreRelease <0b10, 1, 0, 0, 1, GPR32, "stlr">;
2378 def STLRX : StoreRelease <0b11, 1, 0, 0, 1, GPR64, "stlr">;
2379 def STLRB : StoreRelease <0b00, 1, 0, 0, 1, GPR32, "stlrb">;
2380 def STLRH : StoreRelease <0b01, 1, 0, 0, 1, GPR32, "stlrh">;
2382 def STLXRW : StoreExclusive<0b10, 0, 0, 0, 1, GPR32, "stlxr">;
2383 def STLXRX : StoreExclusive<0b11, 0, 0, 0, 1, GPR64, "stlxr">;
2384 def STLXRB : StoreExclusive<0b00, 0, 0, 0, 1, GPR32, "stlxrb">;
2385 def STLXRH : StoreExclusive<0b01, 0, 0, 0, 1, GPR32, "stlxrh">;
2387 def STXRW : StoreExclusive<0b10, 0, 0, 0, 0, GPR32, "stxr">;
2388 def STXRX : StoreExclusive<0b11, 0, 0, 0, 0, GPR64, "stxr">;
2389 def STXRB : StoreExclusive<0b00, 0, 0, 0, 0, GPR32, "stxrb">;
2390 def STXRH : StoreExclusive<0b01, 0, 0, 0, 0, GPR32, "stxrh">;
2392 def LDAXPW : LoadExclusivePair<0b10, 0, 1, 1, 1, GPR32, "ldaxp">;
2393 def LDAXPX : LoadExclusivePair<0b11, 0, 1, 1, 1, GPR64, "ldaxp">;
2395 def LDXPW : LoadExclusivePair<0b10, 0, 1, 1, 0, GPR32, "ldxp">;
2396 def LDXPX : LoadExclusivePair<0b11, 0, 1, 1, 0, GPR64, "ldxp">;
2398 def STLXPW : StoreExclusivePair<0b10, 0, 0, 1, 1, GPR32, "stlxp">;
2399 def STLXPX : StoreExclusivePair<0b11, 0, 0, 1, 1, GPR64, "stlxp">;
2401 def STXPW : StoreExclusivePair<0b10, 0, 0, 1, 0, GPR32, "stxp">;
2402 def STXPX : StoreExclusivePair<0b11, 0, 0, 1, 0, GPR64, "stxp">;
2404 let Predicates = [HasV8_1a] in {
2405 // v8.1a "Limited Order Region" extension load-acquire instructions
2406 def LDLARW : LoadAcquire <0b10, 1, 1, 0, 0, GPR32, "ldlar">;
2407 def LDLARX : LoadAcquire <0b11, 1, 1, 0, 0, GPR64, "ldlar">;
2408 def LDLARB : LoadAcquire <0b00, 1, 1, 0, 0, GPR32, "ldlarb">;
2409 def LDLARH : LoadAcquire <0b01, 1, 1, 0, 0, GPR32, "ldlarh">;
2411 // v8.1a "Limited Order Region" extension store-release instructions
2412 def STLLRW : StoreRelease <0b10, 1, 0, 0, 0, GPR32, "stllr">;
2413 def STLLRX : StoreRelease <0b11, 1, 0, 0, 0, GPR64, "stllr">;
2414 def STLLRB : StoreRelease <0b00, 1, 0, 0, 0, GPR32, "stllrb">;
2415 def STLLRH : StoreRelease <0b01, 1, 0, 0, 0, GPR32, "stllrh">;
2418 //===----------------------------------------------------------------------===//
2419 // Scaled floating point to integer conversion instructions.
2420 //===----------------------------------------------------------------------===//
2422 defm FCVTAS : FPToIntegerUnscaled<0b00, 0b100, "fcvtas", int_aarch64_neon_fcvtas>;
2423 defm FCVTAU : FPToIntegerUnscaled<0b00, 0b101, "fcvtau", int_aarch64_neon_fcvtau>;
2424 defm FCVTMS : FPToIntegerUnscaled<0b10, 0b000, "fcvtms", int_aarch64_neon_fcvtms>;
2425 defm FCVTMU : FPToIntegerUnscaled<0b10, 0b001, "fcvtmu", int_aarch64_neon_fcvtmu>;
2426 defm FCVTNS : FPToIntegerUnscaled<0b00, 0b000, "fcvtns", int_aarch64_neon_fcvtns>;
2427 defm FCVTNU : FPToIntegerUnscaled<0b00, 0b001, "fcvtnu", int_aarch64_neon_fcvtnu>;
2428 defm FCVTPS : FPToIntegerUnscaled<0b01, 0b000, "fcvtps", int_aarch64_neon_fcvtps>;
2429 defm FCVTPU : FPToIntegerUnscaled<0b01, 0b001, "fcvtpu", int_aarch64_neon_fcvtpu>;
2430 defm FCVTZS : FPToIntegerUnscaled<0b11, 0b000, "fcvtzs", fp_to_sint>;
2431 defm FCVTZU : FPToIntegerUnscaled<0b11, 0b001, "fcvtzu", fp_to_uint>;
2432 defm FCVTZS : FPToIntegerScaled<0b11, 0b000, "fcvtzs", fp_to_sint>;
2433 defm FCVTZU : FPToIntegerScaled<0b11, 0b001, "fcvtzu", fp_to_uint>;
2434 let isCodeGenOnly = 1 in {
2435 defm FCVTZS_Int : FPToIntegerUnscaled<0b11, 0b000, "fcvtzs", int_aarch64_neon_fcvtzs>;
2436 defm FCVTZU_Int : FPToIntegerUnscaled<0b11, 0b001, "fcvtzu", int_aarch64_neon_fcvtzu>;
2437 defm FCVTZS_Int : FPToIntegerScaled<0b11, 0b000, "fcvtzs", int_aarch64_neon_fcvtzs>;
2438 defm FCVTZU_Int : FPToIntegerScaled<0b11, 0b001, "fcvtzu", int_aarch64_neon_fcvtzu>;
2441 //===----------------------------------------------------------------------===//
2442 // Scaled integer to floating point conversion instructions.
2443 //===----------------------------------------------------------------------===//
2445 defm SCVTF : IntegerToFP<0, "scvtf", sint_to_fp>;
2446 defm UCVTF : IntegerToFP<1, "ucvtf", uint_to_fp>;
2448 //===----------------------------------------------------------------------===//
2449 // Unscaled integer to floating point conversion instruction.
2450 //===----------------------------------------------------------------------===//
2452 defm FMOV : UnscaledConversion<"fmov">;
2454 // Add pseudo ops for FMOV 0 so we can mark them as isReMaterializable
2455 let isReMaterializable = 1, isCodeGenOnly = 1 in {
2456 def FMOVS0 : Pseudo<(outs FPR32:$Rd), (ins), [(set f32:$Rd, (fpimm0))]>,
2457 PseudoInstExpansion<(FMOVWSr FPR32:$Rd, WZR)>,
2459 def FMOVD0 : Pseudo<(outs FPR64:$Rd), (ins), [(set f64:$Rd, (fpimm0))]>,
2460 PseudoInstExpansion<(FMOVXDr FPR64:$Rd, XZR)>,
2464 //===----------------------------------------------------------------------===//
2465 // Floating point conversion instruction.
2466 //===----------------------------------------------------------------------===//
2468 defm FCVT : FPConversion<"fcvt">;
2470 //===----------------------------------------------------------------------===//
2471 // Floating point single operand instructions.
2472 //===----------------------------------------------------------------------===//
2474 defm FABS : SingleOperandFPData<0b0001, "fabs", fabs>;
2475 defm FMOV : SingleOperandFPData<0b0000, "fmov">;
2476 defm FNEG : SingleOperandFPData<0b0010, "fneg", fneg>;
2477 defm FRINTA : SingleOperandFPData<0b1100, "frinta", frnd>;
2478 defm FRINTI : SingleOperandFPData<0b1111, "frinti", fnearbyint>;
2479 defm FRINTM : SingleOperandFPData<0b1010, "frintm", ffloor>;
2480 defm FRINTN : SingleOperandFPData<0b1000, "frintn", int_aarch64_neon_frintn>;
2481 defm FRINTP : SingleOperandFPData<0b1001, "frintp", fceil>;
2483 def : Pat<(v1f64 (int_aarch64_neon_frintn (v1f64 FPR64:$Rn))),
2484 (FRINTNDr FPR64:$Rn)>;
2486 // FRINTX is inserted to set the flags as required by FENV_ACCESS ON behavior
2487 // in the C spec. Setting hasSideEffects ensures it is not DCE'd.
2488 // <rdar://problem/13715968>
2489 // TODO: We should really model the FPSR flags correctly. This is really ugly.
2490 let hasSideEffects = 1 in {
2491 defm FRINTX : SingleOperandFPData<0b1110, "frintx", frint>;
2494 defm FRINTZ : SingleOperandFPData<0b1011, "frintz", ftrunc>;
2496 let SchedRW = [WriteFDiv] in {
2497 defm FSQRT : SingleOperandFPData<0b0011, "fsqrt", fsqrt>;
2500 //===----------------------------------------------------------------------===//
2501 // Floating point two operand instructions.
2502 //===----------------------------------------------------------------------===//
2504 defm FADD : TwoOperandFPData<0b0010, "fadd", fadd>;
2505 let SchedRW = [WriteFDiv] in {
2506 defm FDIV : TwoOperandFPData<0b0001, "fdiv", fdiv>;
2508 defm FMAXNM : TwoOperandFPData<0b0110, "fmaxnm", int_aarch64_neon_fmaxnm>;
2509 defm FMAX : TwoOperandFPData<0b0100, "fmax", AArch64fmax>;
2510 defm FMINNM : TwoOperandFPData<0b0111, "fminnm", int_aarch64_neon_fminnm>;
2511 defm FMIN : TwoOperandFPData<0b0101, "fmin", AArch64fmin>;
2512 let SchedRW = [WriteFMul] in {
2513 defm FMUL : TwoOperandFPData<0b0000, "fmul", fmul>;
2514 defm FNMUL : TwoOperandFPDataNeg<0b1000, "fnmul", fmul>;
2516 defm FSUB : TwoOperandFPData<0b0011, "fsub", fsub>;
2518 def : Pat<(v1f64 (AArch64fmax (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2519 (FMAXDrr FPR64:$Rn, FPR64:$Rm)>;
2520 def : Pat<(v1f64 (AArch64fmin (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2521 (FMINDrr FPR64:$Rn, FPR64:$Rm)>;
2522 def : Pat<(v1f64 (int_aarch64_neon_fmaxnm (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2523 (FMAXNMDrr FPR64:$Rn, FPR64:$Rm)>;
2524 def : Pat<(v1f64 (int_aarch64_neon_fminnm (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2525 (FMINNMDrr FPR64:$Rn, FPR64:$Rm)>;
2527 //===----------------------------------------------------------------------===//
2528 // Floating point three operand instructions.
2529 //===----------------------------------------------------------------------===//
2531 defm FMADD : ThreeOperandFPData<0, 0, "fmadd", fma>;
2532 defm FMSUB : ThreeOperandFPData<0, 1, "fmsub",
2533 TriOpFrag<(fma node:$LHS, (fneg node:$MHS), node:$RHS)> >;
2534 defm FNMADD : ThreeOperandFPData<1, 0, "fnmadd",
2535 TriOpFrag<(fneg (fma node:$LHS, node:$MHS, node:$RHS))> >;
2536 defm FNMSUB : ThreeOperandFPData<1, 1, "fnmsub",
2537 TriOpFrag<(fma node:$LHS, node:$MHS, (fneg node:$RHS))> >;
2539 // The following def pats catch the case where the LHS of an FMA is negated.
2540 // The TriOpFrag above catches the case where the middle operand is negated.
2542 // N.b. FMSUB etc have the accumulator at the *end* of (outs), unlike
2543 // the NEON variant.
2544 def : Pat<(f32 (fma (fneg FPR32:$Rn), FPR32:$Rm, FPR32:$Ra)),
2545 (FMSUBSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
2547 def : Pat<(f64 (fma (fneg FPR64:$Rn), FPR64:$Rm, FPR64:$Ra)),
2548 (FMSUBDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
2550 // We handled -(a + b*c) for FNMADD above, now it's time for "(-a) + (-b)*c" and
2552 def : Pat<(f32 (fma (fneg FPR32:$Rn), FPR32:$Rm, (fneg FPR32:$Ra))),
2553 (FNMADDSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
2555 def : Pat<(f64 (fma (fneg FPR64:$Rn), FPR64:$Rm, (fneg FPR64:$Ra))),
2556 (FNMADDDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
2558 def : Pat<(f32 (fma FPR32:$Rn, (fneg FPR32:$Rm), (fneg FPR32:$Ra))),
2559 (FNMADDSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
2561 def : Pat<(f64 (fma FPR64:$Rn, (fneg FPR64:$Rm), (fneg FPR64:$Ra))),
2562 (FNMADDDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
2564 //===----------------------------------------------------------------------===//
2565 // Floating point comparison instructions.
2566 //===----------------------------------------------------------------------===//
2568 defm FCMPE : FPComparison<1, "fcmpe">;
2569 defm FCMP : FPComparison<0, "fcmp", AArch64fcmp>;
2571 //===----------------------------------------------------------------------===//
2572 // Floating point conditional comparison instructions.
2573 //===----------------------------------------------------------------------===//
2575 defm FCCMPE : FPCondComparison<1, "fccmpe">;
2576 defm FCCMP : FPCondComparison<0, "fccmp", AArch64fccmp>;
2578 //===----------------------------------------------------------------------===//
2579 // Floating point conditional select instruction.
2580 //===----------------------------------------------------------------------===//
2582 defm FCSEL : FPCondSelect<"fcsel">;
2584 // CSEL instructions providing f128 types need to be handled by a
2585 // pseudo-instruction since the eventual code will need to introduce basic
2586 // blocks and control flow.
2587 def F128CSEL : Pseudo<(outs FPR128:$Rd),
2588 (ins FPR128:$Rn, FPR128:$Rm, ccode:$cond),
2589 [(set (f128 FPR128:$Rd),
2590 (AArch64csel FPR128:$Rn, FPR128:$Rm,
2591 (i32 imm:$cond), NZCV))]> {
2593 let usesCustomInserter = 1;
2597 //===----------------------------------------------------------------------===//
2598 // Floating point immediate move.
2599 //===----------------------------------------------------------------------===//
2601 let isReMaterializable = 1 in {
2602 defm FMOV : FPMoveImmediate<"fmov">;
2605 //===----------------------------------------------------------------------===//
2606 // Advanced SIMD two vector instructions.
2607 //===----------------------------------------------------------------------===//
2609 defm ABS : SIMDTwoVectorBHSD<0, 0b01011, "abs", int_aarch64_neon_abs>;
2610 def : Pat<(xor (v8i8 (AArch64vashr V64:$src, (i32 7))),
2611 (v8i8 (add V64:$src, (AArch64vashr V64:$src, (i32 7))))),
2612 (ABSv8i8 V64:$src)>;
2613 def : Pat<(xor (v4i16 (AArch64vashr V64:$src, (i32 15))),
2614 (v4i16 (add V64:$src, (AArch64vashr V64:$src, (i32 15))))),
2615 (ABSv4i16 V64:$src)>;
2616 def : Pat<(xor (v2i32 (AArch64vashr V64:$src, (i32 31))),
2617 (v2i32 (add V64:$src, (AArch64vashr V64:$src, (i32 31))))),
2618 (ABSv2i32 V64:$src)>;
2619 def : Pat<(xor (v16i8 (AArch64vashr V128:$src, (i32 7))),
2620 (v16i8 (add V128:$src, (AArch64vashr V128:$src, (i32 7))))),
2621 (ABSv16i8 V128:$src)>;
2622 def : Pat<(xor (v8i16 (AArch64vashr V128:$src, (i32 15))),
2623 (v8i16 (add V128:$src, (AArch64vashr V128:$src, (i32 15))))),
2624 (ABSv8i16 V128:$src)>;
2625 def : Pat<(xor (v4i32 (AArch64vashr V128:$src, (i32 31))),
2626 (v4i32 (add V128:$src, (AArch64vashr V128:$src, (i32 31))))),
2627 (ABSv4i32 V128:$src)>;
2628 def : Pat<(xor (v2i64 (AArch64vashr V128:$src, (i32 63))),
2629 (v2i64 (add V128:$src, (AArch64vashr V128:$src, (i32 63))))),
2630 (ABSv2i64 V128:$src)>;
2632 defm CLS : SIMDTwoVectorBHS<0, 0b00100, "cls", int_aarch64_neon_cls>;
2633 defm CLZ : SIMDTwoVectorBHS<1, 0b00100, "clz", ctlz>;
2634 defm CMEQ : SIMDCmpTwoVector<0, 0b01001, "cmeq", AArch64cmeqz>;
2635 defm CMGE : SIMDCmpTwoVector<1, 0b01000, "cmge", AArch64cmgez>;
2636 defm CMGT : SIMDCmpTwoVector<0, 0b01000, "cmgt", AArch64cmgtz>;
2637 defm CMLE : SIMDCmpTwoVector<1, 0b01001, "cmle", AArch64cmlez>;
2638 defm CMLT : SIMDCmpTwoVector<0, 0b01010, "cmlt", AArch64cmltz>;
2639 defm CNT : SIMDTwoVectorB<0, 0b00, 0b00101, "cnt", ctpop>;
2640 defm FABS : SIMDTwoVectorFP<0, 1, 0b01111, "fabs", fabs>;
2642 defm FCMEQ : SIMDFPCmpTwoVector<0, 1, 0b01101, "fcmeq", AArch64fcmeqz>;
2643 defm FCMGE : SIMDFPCmpTwoVector<1, 1, 0b01100, "fcmge", AArch64fcmgez>;
2644 defm FCMGT : SIMDFPCmpTwoVector<0, 1, 0b01100, "fcmgt", AArch64fcmgtz>;
2645 defm FCMLE : SIMDFPCmpTwoVector<1, 1, 0b01101, "fcmle", AArch64fcmlez>;
2646 defm FCMLT : SIMDFPCmpTwoVector<0, 1, 0b01110, "fcmlt", AArch64fcmltz>;
2647 defm FCVTAS : SIMDTwoVectorFPToInt<0,0,0b11100, "fcvtas",int_aarch64_neon_fcvtas>;
2648 defm FCVTAU : SIMDTwoVectorFPToInt<1,0,0b11100, "fcvtau",int_aarch64_neon_fcvtau>;
2649 defm FCVTL : SIMDFPWidenTwoVector<0, 0, 0b10111, "fcvtl">;
2650 def : Pat<(v4f32 (int_aarch64_neon_vcvthf2fp (v4i16 V64:$Rn))),
2651 (FCVTLv4i16 V64:$Rn)>;
2652 def : Pat<(v4f32 (int_aarch64_neon_vcvthf2fp (extract_subvector (v8i16 V128:$Rn),
2654 (FCVTLv8i16 V128:$Rn)>;
2655 def : Pat<(v2f64 (fextend (v2f32 V64:$Rn))), (FCVTLv2i32 V64:$Rn)>;
2656 def : Pat<(v2f64 (fextend (v2f32 (extract_subvector (v4f32 V128:$Rn),
2658 (FCVTLv4i32 V128:$Rn)>;
2660 def : Pat<(v4f32 (fextend (v4f16 V64:$Rn))), (FCVTLv4i16 V64:$Rn)>;
2661 def : Pat<(v4f32 (fextend (v4f16 (extract_subvector (v8f16 V128:$Rn),
2663 (FCVTLv8i16 V128:$Rn)>;
2665 defm FCVTMS : SIMDTwoVectorFPToInt<0,0,0b11011, "fcvtms",int_aarch64_neon_fcvtms>;
2666 defm FCVTMU : SIMDTwoVectorFPToInt<1,0,0b11011, "fcvtmu",int_aarch64_neon_fcvtmu>;
2667 defm FCVTNS : SIMDTwoVectorFPToInt<0,0,0b11010, "fcvtns",int_aarch64_neon_fcvtns>;
2668 defm FCVTNU : SIMDTwoVectorFPToInt<1,0,0b11010, "fcvtnu",int_aarch64_neon_fcvtnu>;
2669 defm FCVTN : SIMDFPNarrowTwoVector<0, 0, 0b10110, "fcvtn">;
2670 def : Pat<(v4i16 (int_aarch64_neon_vcvtfp2hf (v4f32 V128:$Rn))),
2671 (FCVTNv4i16 V128:$Rn)>;
2672 def : Pat<(concat_vectors V64:$Rd,
2673 (v4i16 (int_aarch64_neon_vcvtfp2hf (v4f32 V128:$Rn)))),
2674 (FCVTNv8i16 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
2675 def : Pat<(v2f32 (fround (v2f64 V128:$Rn))), (FCVTNv2i32 V128:$Rn)>;
2676 def : Pat<(v4f16 (fround (v4f32 V128:$Rn))), (FCVTNv4i16 V128:$Rn)>;
2677 def : Pat<(concat_vectors V64:$Rd, (v2f32 (fround (v2f64 V128:$Rn)))),
2678 (FCVTNv4i32 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
2679 defm FCVTPS : SIMDTwoVectorFPToInt<0,1,0b11010, "fcvtps",int_aarch64_neon_fcvtps>;
2680 defm FCVTPU : SIMDTwoVectorFPToInt<1,1,0b11010, "fcvtpu",int_aarch64_neon_fcvtpu>;
2681 defm FCVTXN : SIMDFPInexactCvtTwoVector<1, 0, 0b10110, "fcvtxn",
2682 int_aarch64_neon_fcvtxn>;
2683 defm FCVTZS : SIMDTwoVectorFPToInt<0, 1, 0b11011, "fcvtzs", fp_to_sint>;
2684 defm FCVTZU : SIMDTwoVectorFPToInt<1, 1, 0b11011, "fcvtzu", fp_to_uint>;
2685 let isCodeGenOnly = 1 in {
2686 defm FCVTZS_Int : SIMDTwoVectorFPToInt<0, 1, 0b11011, "fcvtzs",
2687 int_aarch64_neon_fcvtzs>;
2688 defm FCVTZU_Int : SIMDTwoVectorFPToInt<1, 1, 0b11011, "fcvtzu",
2689 int_aarch64_neon_fcvtzu>;
2691 defm FNEG : SIMDTwoVectorFP<1, 1, 0b01111, "fneg", fneg>;
2692 defm FRECPE : SIMDTwoVectorFP<0, 1, 0b11101, "frecpe", int_aarch64_neon_frecpe>;
2693 defm FRINTA : SIMDTwoVectorFP<1, 0, 0b11000, "frinta", frnd>;
2694 defm FRINTI : SIMDTwoVectorFP<1, 1, 0b11001, "frinti", fnearbyint>;
2695 defm FRINTM : SIMDTwoVectorFP<0, 0, 0b11001, "frintm", ffloor>;
2696 defm FRINTN : SIMDTwoVectorFP<0, 0, 0b11000, "frintn", int_aarch64_neon_frintn>;
2697 defm FRINTP : SIMDTwoVectorFP<0, 1, 0b11000, "frintp", fceil>;
2698 defm FRINTX : SIMDTwoVectorFP<1, 0, 0b11001, "frintx", frint>;
2699 defm FRINTZ : SIMDTwoVectorFP<0, 1, 0b11001, "frintz", ftrunc>;
2700 defm FRSQRTE: SIMDTwoVectorFP<1, 1, 0b11101, "frsqrte", int_aarch64_neon_frsqrte>;
2701 defm FSQRT : SIMDTwoVectorFP<1, 1, 0b11111, "fsqrt", fsqrt>;
2702 defm NEG : SIMDTwoVectorBHSD<1, 0b01011, "neg",
2703 UnOpFrag<(sub immAllZerosV, node:$LHS)> >;
2704 defm NOT : SIMDTwoVectorB<1, 0b00, 0b00101, "not", vnot>;
2705 // Aliases for MVN -> NOT.
2706 def : InstAlias<"mvn{ $Vd.8b, $Vn.8b|.8b $Vd, $Vn}",
2707 (NOTv8i8 V64:$Vd, V64:$Vn)>;
2708 def : InstAlias<"mvn{ $Vd.16b, $Vn.16b|.16b $Vd, $Vn}",
2709 (NOTv16i8 V128:$Vd, V128:$Vn)>;
2711 def : Pat<(AArch64neg (v8i8 V64:$Rn)), (NEGv8i8 V64:$Rn)>;
2712 def : Pat<(AArch64neg (v16i8 V128:$Rn)), (NEGv16i8 V128:$Rn)>;
2713 def : Pat<(AArch64neg (v4i16 V64:$Rn)), (NEGv4i16 V64:$Rn)>;
2714 def : Pat<(AArch64neg (v8i16 V128:$Rn)), (NEGv8i16 V128:$Rn)>;
2715 def : Pat<(AArch64neg (v2i32 V64:$Rn)), (NEGv2i32 V64:$Rn)>;
2716 def : Pat<(AArch64neg (v4i32 V128:$Rn)), (NEGv4i32 V128:$Rn)>;
2717 def : Pat<(AArch64neg (v2i64 V128:$Rn)), (NEGv2i64 V128:$Rn)>;
2719 def : Pat<(AArch64not (v8i8 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2720 def : Pat<(AArch64not (v16i8 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2721 def : Pat<(AArch64not (v4i16 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2722 def : Pat<(AArch64not (v8i16 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2723 def : Pat<(AArch64not (v2i32 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2724 def : Pat<(AArch64not (v1i64 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2725 def : Pat<(AArch64not (v4i32 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2726 def : Pat<(AArch64not (v2i64 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2728 def : Pat<(vnot (v4i16 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2729 def : Pat<(vnot (v8i16 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2730 def : Pat<(vnot (v2i32 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2731 def : Pat<(vnot (v4i32 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2732 def : Pat<(vnot (v2i64 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2734 defm RBIT : SIMDTwoVectorB<1, 0b01, 0b00101, "rbit", int_aarch64_neon_rbit>;
2735 defm REV16 : SIMDTwoVectorB<0, 0b00, 0b00001, "rev16", AArch64rev16>;
2736 defm REV32 : SIMDTwoVectorBH<1, 0b00000, "rev32", AArch64rev32>;
2737 defm REV64 : SIMDTwoVectorBHS<0, 0b00000, "rev64", AArch64rev64>;
2738 defm SADALP : SIMDLongTwoVectorTied<0, 0b00110, "sadalp",
2739 BinOpFrag<(add node:$LHS, (int_aarch64_neon_saddlp node:$RHS))> >;
2740 defm SADDLP : SIMDLongTwoVector<0, 0b00010, "saddlp", int_aarch64_neon_saddlp>;
2741 defm SCVTF : SIMDTwoVectorIntToFP<0, 0, 0b11101, "scvtf", sint_to_fp>;
2742 defm SHLL : SIMDVectorLShiftLongBySizeBHS;
2743 defm SQABS : SIMDTwoVectorBHSD<0, 0b00111, "sqabs", int_aarch64_neon_sqabs>;
2744 defm SQNEG : SIMDTwoVectorBHSD<1, 0b00111, "sqneg", int_aarch64_neon_sqneg>;
2745 defm SQXTN : SIMDMixedTwoVector<0, 0b10100, "sqxtn", int_aarch64_neon_sqxtn>;
2746 defm SQXTUN : SIMDMixedTwoVector<1, 0b10010, "sqxtun", int_aarch64_neon_sqxtun>;
2747 defm SUQADD : SIMDTwoVectorBHSDTied<0, 0b00011, "suqadd",int_aarch64_neon_suqadd>;
2748 defm UADALP : SIMDLongTwoVectorTied<1, 0b00110, "uadalp",
2749 BinOpFrag<(add node:$LHS, (int_aarch64_neon_uaddlp node:$RHS))> >;
2750 defm UADDLP : SIMDLongTwoVector<1, 0b00010, "uaddlp",
2751 int_aarch64_neon_uaddlp>;
2752 defm UCVTF : SIMDTwoVectorIntToFP<1, 0, 0b11101, "ucvtf", uint_to_fp>;
2753 defm UQXTN : SIMDMixedTwoVector<1, 0b10100, "uqxtn", int_aarch64_neon_uqxtn>;
2754 defm URECPE : SIMDTwoVectorS<0, 1, 0b11100, "urecpe", int_aarch64_neon_urecpe>;
2755 defm URSQRTE: SIMDTwoVectorS<1, 1, 0b11100, "ursqrte", int_aarch64_neon_ursqrte>;
2756 defm USQADD : SIMDTwoVectorBHSDTied<1, 0b00011, "usqadd",int_aarch64_neon_usqadd>;
2757 defm XTN : SIMDMixedTwoVector<0, 0b10010, "xtn", trunc>;
2759 def : Pat<(v4f16 (AArch64rev32 V64:$Rn)), (REV32v4i16 V64:$Rn)>;
2760 def : Pat<(v4f16 (AArch64rev64 V64:$Rn)), (REV64v4i16 V64:$Rn)>;
2761 def : Pat<(v8f16 (AArch64rev32 V128:$Rn)), (REV32v8i16 V128:$Rn)>;
2762 def : Pat<(v8f16 (AArch64rev64 V128:$Rn)), (REV64v8i16 V128:$Rn)>;
2763 def : Pat<(v2f32 (AArch64rev64 V64:$Rn)), (REV64v2i32 V64:$Rn)>;
2764 def : Pat<(v4f32 (AArch64rev64 V128:$Rn)), (REV64v4i32 V128:$Rn)>;
2766 // Patterns for vector long shift (by element width). These need to match all
2767 // three of zext, sext and anyext so it's easier to pull the patterns out of the
2769 multiclass SIMDVectorLShiftLongBySizeBHSPats<SDPatternOperator ext> {
2770 def : Pat<(AArch64vshl (v8i16 (ext (v8i8 V64:$Rn))), (i32 8)),
2771 (SHLLv8i8 V64:$Rn)>;
2772 def : Pat<(AArch64vshl (v8i16 (ext (extract_high_v16i8 V128:$Rn))), (i32 8)),
2773 (SHLLv16i8 V128:$Rn)>;
2774 def : Pat<(AArch64vshl (v4i32 (ext (v4i16 V64:$Rn))), (i32 16)),
2775 (SHLLv4i16 V64:$Rn)>;
2776 def : Pat<(AArch64vshl (v4i32 (ext (extract_high_v8i16 V128:$Rn))), (i32 16)),
2777 (SHLLv8i16 V128:$Rn)>;
2778 def : Pat<(AArch64vshl (v2i64 (ext (v2i32 V64:$Rn))), (i32 32)),
2779 (SHLLv2i32 V64:$Rn)>;
2780 def : Pat<(AArch64vshl (v2i64 (ext (extract_high_v4i32 V128:$Rn))), (i32 32)),
2781 (SHLLv4i32 V128:$Rn)>;
2784 defm : SIMDVectorLShiftLongBySizeBHSPats<anyext>;
2785 defm : SIMDVectorLShiftLongBySizeBHSPats<zext>;
2786 defm : SIMDVectorLShiftLongBySizeBHSPats<sext>;
2788 //===----------------------------------------------------------------------===//
2789 // Advanced SIMD three vector instructions.
2790 //===----------------------------------------------------------------------===//
2792 defm ADD : SIMDThreeSameVector<0, 0b10000, "add", add>;
2793 defm ADDP : SIMDThreeSameVector<0, 0b10111, "addp", int_aarch64_neon_addp>;
2794 defm CMEQ : SIMDThreeSameVector<1, 0b10001, "cmeq", AArch64cmeq>;
2795 defm CMGE : SIMDThreeSameVector<0, 0b00111, "cmge", AArch64cmge>;
2796 defm CMGT : SIMDThreeSameVector<0, 0b00110, "cmgt", AArch64cmgt>;
2797 defm CMHI : SIMDThreeSameVector<1, 0b00110, "cmhi", AArch64cmhi>;
2798 defm CMHS : SIMDThreeSameVector<1, 0b00111, "cmhs", AArch64cmhs>;
2799 defm CMTST : SIMDThreeSameVector<0, 0b10001, "cmtst", AArch64cmtst>;
2800 defm FABD : SIMDThreeSameVectorFP<1,1,0b11010,"fabd", int_aarch64_neon_fabd>;
2801 defm FACGE : SIMDThreeSameVectorFPCmp<1,0,0b11101,"facge",int_aarch64_neon_facge>;
2802 defm FACGT : SIMDThreeSameVectorFPCmp<1,1,0b11101,"facgt",int_aarch64_neon_facgt>;
2803 defm FADDP : SIMDThreeSameVectorFP<1,0,0b11010,"faddp",int_aarch64_neon_addp>;
2804 defm FADD : SIMDThreeSameVectorFP<0,0,0b11010,"fadd", fadd>;
2805 defm FCMEQ : SIMDThreeSameVectorFPCmp<0, 0, 0b11100, "fcmeq", AArch64fcmeq>;
2806 defm FCMGE : SIMDThreeSameVectorFPCmp<1, 0, 0b11100, "fcmge", AArch64fcmge>;
2807 defm FCMGT : SIMDThreeSameVectorFPCmp<1, 1, 0b11100, "fcmgt", AArch64fcmgt>;
2808 defm FDIV : SIMDThreeSameVectorFP<1,0,0b11111,"fdiv", fdiv>;
2809 defm FMAXNMP : SIMDThreeSameVectorFP<1,0,0b11000,"fmaxnmp", int_aarch64_neon_fmaxnmp>;
2810 defm FMAXNM : SIMDThreeSameVectorFP<0,0,0b11000,"fmaxnm", int_aarch64_neon_fmaxnm>;
2811 defm FMAXP : SIMDThreeSameVectorFP<1,0,0b11110,"fmaxp", int_aarch64_neon_fmaxp>;
2812 defm FMAX : SIMDThreeSameVectorFP<0,0,0b11110,"fmax", AArch64fmax>;
2813 defm FMINNMP : SIMDThreeSameVectorFP<1,1,0b11000,"fminnmp", int_aarch64_neon_fminnmp>;
2814 defm FMINNM : SIMDThreeSameVectorFP<0,1,0b11000,"fminnm", int_aarch64_neon_fminnm>;
2815 defm FMINP : SIMDThreeSameVectorFP<1,1,0b11110,"fminp", int_aarch64_neon_fminp>;
2816 defm FMIN : SIMDThreeSameVectorFP<0,1,0b11110,"fmin", AArch64fmin>;
2818 // NOTE: The operands of the PatFrag are reordered on FMLA/FMLS because the
2819 // instruction expects the addend first, while the fma intrinsic puts it last.
2820 defm FMLA : SIMDThreeSameVectorFPTied<0, 0, 0b11001, "fmla",
2821 TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)> >;
2822 defm FMLS : SIMDThreeSameVectorFPTied<0, 1, 0b11001, "fmls",
2823 TriOpFrag<(fma node:$MHS, (fneg node:$RHS), node:$LHS)> >;
2825 // The following def pats catch the case where the LHS of an FMA is negated.
2826 // The TriOpFrag above catches the case where the middle operand is negated.
2827 def : Pat<(v2f32 (fma (fneg V64:$Rn), V64:$Rm, V64:$Rd)),
2828 (FMLSv2f32 V64:$Rd, V64:$Rn, V64:$Rm)>;
2830 def : Pat<(v4f32 (fma (fneg V128:$Rn), V128:$Rm, V128:$Rd)),
2831 (FMLSv4f32 V128:$Rd, V128:$Rn, V128:$Rm)>;
2833 def : Pat<(v2f64 (fma (fneg V128:$Rn), V128:$Rm, V128:$Rd)),
2834 (FMLSv2f64 V128:$Rd, V128:$Rn, V128:$Rm)>;
2836 defm FMULX : SIMDThreeSameVectorFP<0,0,0b11011,"fmulx", int_aarch64_neon_fmulx>;
2837 defm FMUL : SIMDThreeSameVectorFP<1,0,0b11011,"fmul", fmul>;
2838 defm FRECPS : SIMDThreeSameVectorFP<0,0,0b11111,"frecps", int_aarch64_neon_frecps>;
2839 defm FRSQRTS : SIMDThreeSameVectorFP<0,1,0b11111,"frsqrts", int_aarch64_neon_frsqrts>;
2840 defm FSUB : SIMDThreeSameVectorFP<0,1,0b11010,"fsub", fsub>;
2841 defm MLA : SIMDThreeSameVectorBHSTied<0, 0b10010, "mla",
2842 TriOpFrag<(add node:$LHS, (mul node:$MHS, node:$RHS))> >;
2843 defm MLS : SIMDThreeSameVectorBHSTied<1, 0b10010, "mls",
2844 TriOpFrag<(sub node:$LHS, (mul node:$MHS, node:$RHS))> >;
2845 defm MUL : SIMDThreeSameVectorBHS<0, 0b10011, "mul", mul>;
2846 defm PMUL : SIMDThreeSameVectorB<1, 0b10011, "pmul", int_aarch64_neon_pmul>;
2847 defm SABA : SIMDThreeSameVectorBHSTied<0, 0b01111, "saba",
2848 TriOpFrag<(add node:$LHS, (sabsdiff node:$MHS, node:$RHS))> >;
2849 defm SABD : SIMDThreeSameVectorBHS<0,0b01110,"sabd", sabsdiff>;
2850 defm SHADD : SIMDThreeSameVectorBHS<0,0b00000,"shadd", int_aarch64_neon_shadd>;
2851 defm SHSUB : SIMDThreeSameVectorBHS<0,0b00100,"shsub", int_aarch64_neon_shsub>;
2852 defm SMAXP : SIMDThreeSameVectorBHS<0,0b10100,"smaxp", int_aarch64_neon_smaxp>;
2853 defm SMAX : SIMDThreeSameVectorBHS<0,0b01100,"smax", int_aarch64_neon_smax>;
2854 defm SMINP : SIMDThreeSameVectorBHS<0,0b10101,"sminp", int_aarch64_neon_sminp>;
2855 defm SMIN : SIMDThreeSameVectorBHS<0,0b01101,"smin", int_aarch64_neon_smin>;
2856 defm SQADD : SIMDThreeSameVector<0,0b00001,"sqadd", int_aarch64_neon_sqadd>;
2857 defm SQDMULH : SIMDThreeSameVectorHS<0,0b10110,"sqdmulh",int_aarch64_neon_sqdmulh>;
2858 defm SQRDMULH : SIMDThreeSameVectorHS<1,0b10110,"sqrdmulh",int_aarch64_neon_sqrdmulh>;
2859 defm SQRSHL : SIMDThreeSameVector<0,0b01011,"sqrshl", int_aarch64_neon_sqrshl>;
2860 defm SQSHL : SIMDThreeSameVector<0,0b01001,"sqshl", int_aarch64_neon_sqshl>;
2861 defm SQSUB : SIMDThreeSameVector<0,0b00101,"sqsub", int_aarch64_neon_sqsub>;
2862 defm SRHADD : SIMDThreeSameVectorBHS<0,0b00010,"srhadd",int_aarch64_neon_srhadd>;
2863 defm SRSHL : SIMDThreeSameVector<0,0b01010,"srshl", int_aarch64_neon_srshl>;
2864 defm SSHL : SIMDThreeSameVector<0,0b01000,"sshl", int_aarch64_neon_sshl>;
2865 defm SUB : SIMDThreeSameVector<1,0b10000,"sub", sub>;
2866 defm UABA : SIMDThreeSameVectorBHSTied<1, 0b01111, "uaba",
2867 TriOpFrag<(add node:$LHS, (uabsdiff node:$MHS, node:$RHS))> >;
2868 defm UABD : SIMDThreeSameVectorBHS<1,0b01110,"uabd", uabsdiff>;
2869 defm UHADD : SIMDThreeSameVectorBHS<1,0b00000,"uhadd", int_aarch64_neon_uhadd>;
2870 defm UHSUB : SIMDThreeSameVectorBHS<1,0b00100,"uhsub", int_aarch64_neon_uhsub>;
2871 defm UMAXP : SIMDThreeSameVectorBHS<1,0b10100,"umaxp", int_aarch64_neon_umaxp>;
2872 defm UMAX : SIMDThreeSameVectorBHS<1,0b01100,"umax", int_aarch64_neon_umax>;
2873 defm UMINP : SIMDThreeSameVectorBHS<1,0b10101,"uminp", int_aarch64_neon_uminp>;
2874 defm UMIN : SIMDThreeSameVectorBHS<1,0b01101,"umin", int_aarch64_neon_umin>;
2875 defm UQADD : SIMDThreeSameVector<1,0b00001,"uqadd", int_aarch64_neon_uqadd>;
2876 defm UQRSHL : SIMDThreeSameVector<1,0b01011,"uqrshl", int_aarch64_neon_uqrshl>;
2877 defm UQSHL : SIMDThreeSameVector<1,0b01001,"uqshl", int_aarch64_neon_uqshl>;
2878 defm UQSUB : SIMDThreeSameVector<1,0b00101,"uqsub", int_aarch64_neon_uqsub>;
2879 defm URHADD : SIMDThreeSameVectorBHS<1,0b00010,"urhadd", int_aarch64_neon_urhadd>;
2880 defm URSHL : SIMDThreeSameVector<1,0b01010,"urshl", int_aarch64_neon_urshl>;
2881 defm USHL : SIMDThreeSameVector<1,0b01000,"ushl", int_aarch64_neon_ushl>;
2882 defm SQRDMLAH : SIMDThreeSameVectorSQRDMLxHTiedHS<1,0b10000,"sqrdmlah",
2883 int_aarch64_neon_sqadd>;
2884 defm SQRDMLSH : SIMDThreeSameVectorSQRDMLxHTiedHS<1,0b10001,"sqrdmlsh",
2885 int_aarch64_neon_sqsub>;
2887 defm AND : SIMDLogicalThreeVector<0, 0b00, "and", and>;
2888 defm BIC : SIMDLogicalThreeVector<0, 0b01, "bic",
2889 BinOpFrag<(and node:$LHS, (vnot node:$RHS))> >;
2890 defm BIF : SIMDLogicalThreeVector<1, 0b11, "bif">;
2891 defm BIT : SIMDLogicalThreeVectorTied<1, 0b10, "bit", AArch64bit>;
2892 defm BSL : SIMDLogicalThreeVectorTied<1, 0b01, "bsl",
2893 TriOpFrag<(or (and node:$LHS, node:$MHS), (and (vnot node:$LHS), node:$RHS))>>;
2894 defm EOR : SIMDLogicalThreeVector<1, 0b00, "eor", xor>;
2895 defm ORN : SIMDLogicalThreeVector<0, 0b11, "orn",
2896 BinOpFrag<(or node:$LHS, (vnot node:$RHS))> >;
2897 defm ORR : SIMDLogicalThreeVector<0, 0b10, "orr", or>;
2899 def : Pat<(v8i8 (smin V64:$Rn, V64:$Rm)),
2900 (SMINv8i8 V64:$Rn, V64:$Rm)>;
2901 def : Pat<(v4i16 (smin V64:$Rn, V64:$Rm)),
2902 (SMINv4i16 V64:$Rn, V64:$Rm)>;
2903 def : Pat<(v2i32 (smin V64:$Rn, V64:$Rm)),
2904 (SMINv2i32 V64:$Rn, V64:$Rm)>;
2905 def : Pat<(v16i8 (smin V128:$Rn, V128:$Rm)),
2906 (SMINv16i8 V128:$Rn, V128:$Rm)>;
2907 def : Pat<(v8i16 (smin V128:$Rn, V128:$Rm)),
2908 (SMINv8i16 V128:$Rn, V128:$Rm)>;
2909 def : Pat<(v4i32 (smin V128:$Rn, V128:$Rm)),
2910 (SMINv4i32 V128:$Rn, V128:$Rm)>;
2911 def : Pat<(v8i8 (smax V64:$Rn, V64:$Rm)),
2912 (SMAXv8i8 V64:$Rn, V64:$Rm)>;
2913 def : Pat<(v4i16 (smax V64:$Rn, V64:$Rm)),
2914 (SMAXv4i16 V64:$Rn, V64:$Rm)>;
2915 def : Pat<(v2i32 (smax V64:$Rn, V64:$Rm)),
2916 (SMAXv2i32 V64:$Rn, V64:$Rm)>;
2917 def : Pat<(v16i8 (smax V128:$Rn, V128:$Rm)),
2918 (SMAXv16i8 V128:$Rn, V128:$Rm)>;
2919 def : Pat<(v8i16 (smax V128:$Rn, V128:$Rm)),
2920 (SMAXv8i16 V128:$Rn, V128:$Rm)>;
2921 def : Pat<(v4i32 (smax V128:$Rn, V128:$Rm)),
2922 (SMAXv4i32 V128:$Rn, V128:$Rm)>;
2923 def : Pat<(v8i8 (umin V64:$Rn, V64:$Rm)),
2924 (UMINv8i8 V64:$Rn, V64:$Rm)>;
2925 def : Pat<(v4i16 (umin V64:$Rn, V64:$Rm)),
2926 (UMINv4i16 V64:$Rn, V64:$Rm)>;
2927 def : Pat<(v2i32 (umin V64:$Rn, V64:$Rm)),
2928 (UMINv2i32 V64:$Rn, V64:$Rm)>;
2929 def : Pat<(v16i8 (umin V128:$Rn, V128:$Rm)),
2930 (UMINv16i8 V128:$Rn, V128:$Rm)>;
2931 def : Pat<(v8i16 (umin V128:$Rn, V128:$Rm)),
2932 (UMINv8i16 V128:$Rn, V128:$Rm)>;
2933 def : Pat<(v4i32 (umin V128:$Rn, V128:$Rm)),
2934 (UMINv4i32 V128:$Rn, V128:$Rm)>;
2935 def : Pat<(v8i8 (umax V64:$Rn, V64:$Rm)),
2936 (UMAXv8i8 V64:$Rn, V64:$Rm)>;
2937 def : Pat<(v4i16 (umax V64:$Rn, V64:$Rm)),
2938 (UMAXv4i16 V64:$Rn, V64:$Rm)>;
2939 def : Pat<(v2i32 (umax V64:$Rn, V64:$Rm)),
2940 (UMAXv2i32 V64:$Rn, V64:$Rm)>;
2941 def : Pat<(v16i8 (umax V128:$Rn, V128:$Rm)),
2942 (UMAXv16i8 V128:$Rn, V128:$Rm)>;
2943 def : Pat<(v8i16 (umax V128:$Rn, V128:$Rm)),
2944 (UMAXv8i16 V128:$Rn, V128:$Rm)>;
2945 def : Pat<(v4i32 (umax V128:$Rn, V128:$Rm)),
2946 (UMAXv4i32 V128:$Rn, V128:$Rm)>;
2948 def : Pat<(AArch64bsl (v8i8 V64:$Rd), V64:$Rn, V64:$Rm),
2949 (BSLv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
2950 def : Pat<(AArch64bsl (v4i16 V64:$Rd), V64:$Rn, V64:$Rm),
2951 (BSLv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
2952 def : Pat<(AArch64bsl (v2i32 V64:$Rd), V64:$Rn, V64:$Rm),
2953 (BSLv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
2954 def : Pat<(AArch64bsl (v1i64 V64:$Rd), V64:$Rn, V64:$Rm),
2955 (BSLv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
2957 def : Pat<(AArch64bsl (v16i8 V128:$Rd), V128:$Rn, V128:$Rm),
2958 (BSLv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
2959 def : Pat<(AArch64bsl (v8i16 V128:$Rd), V128:$Rn, V128:$Rm),
2960 (BSLv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
2961 def : Pat<(AArch64bsl (v4i32 V128:$Rd), V128:$Rn, V128:$Rm),
2962 (BSLv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
2963 def : Pat<(AArch64bsl (v2i64 V128:$Rd), V128:$Rn, V128:$Rm),
2964 (BSLv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
2966 def : InstAlias<"mov{\t$dst.16b, $src.16b|.16b\t$dst, $src}",
2967 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 1>;
2968 def : InstAlias<"mov{\t$dst.8h, $src.8h|.8h\t$dst, $src}",
2969 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2970 def : InstAlias<"mov{\t$dst.4s, $src.4s|.4s\t$dst, $src}",
2971 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2972 def : InstAlias<"mov{\t$dst.2d, $src.2d|.2d\t$dst, $src}",
2973 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2975 def : InstAlias<"mov{\t$dst.8b, $src.8b|.8b\t$dst, $src}",
2976 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 1>;
2977 def : InstAlias<"mov{\t$dst.4h, $src.4h|.4h\t$dst, $src}",
2978 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2979 def : InstAlias<"mov{\t$dst.2s, $src.2s|.2s\t$dst, $src}",
2980 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2981 def : InstAlias<"mov{\t$dst.1d, $src.1d|.1d\t$dst, $src}",
2982 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2984 def : InstAlias<"{cmls\t$dst.8b, $src1.8b, $src2.8b" #
2985 "|cmls.8b\t$dst, $src1, $src2}",
2986 (CMHSv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2987 def : InstAlias<"{cmls\t$dst.16b, $src1.16b, $src2.16b" #
2988 "|cmls.16b\t$dst, $src1, $src2}",
2989 (CMHSv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2990 def : InstAlias<"{cmls\t$dst.4h, $src1.4h, $src2.4h" #
2991 "|cmls.4h\t$dst, $src1, $src2}",
2992 (CMHSv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2993 def : InstAlias<"{cmls\t$dst.8h, $src1.8h, $src2.8h" #
2994 "|cmls.8h\t$dst, $src1, $src2}",
2995 (CMHSv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2996 def : InstAlias<"{cmls\t$dst.2s, $src1.2s, $src2.2s" #
2997 "|cmls.2s\t$dst, $src1, $src2}",
2998 (CMHSv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2999 def : InstAlias<"{cmls\t$dst.4s, $src1.4s, $src2.4s" #
3000 "|cmls.4s\t$dst, $src1, $src2}",
3001 (CMHSv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
3002 def : InstAlias<"{cmls\t$dst.2d, $src1.2d, $src2.2d" #
3003 "|cmls.2d\t$dst, $src1, $src2}",
3004 (CMHSv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
3006 def : InstAlias<"{cmlo\t$dst.8b, $src1.8b, $src2.8b" #
3007 "|cmlo.8b\t$dst, $src1, $src2}",
3008 (CMHIv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
3009 def : InstAlias<"{cmlo\t$dst.16b, $src1.16b, $src2.16b" #
3010 "|cmlo.16b\t$dst, $src1, $src2}",
3011 (CMHIv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
3012 def : InstAlias<"{cmlo\t$dst.4h, $src1.4h, $src2.4h" #
3013 "|cmlo.4h\t$dst, $src1, $src2}",
3014 (CMHIv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
3015 def : InstAlias<"{cmlo\t$dst.8h, $src1.8h, $src2.8h" #
3016 "|cmlo.8h\t$dst, $src1, $src2}",
3017 (CMHIv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
3018 def : InstAlias<"{cmlo\t$dst.2s, $src1.2s, $src2.2s" #
3019 "|cmlo.2s\t$dst, $src1, $src2}",
3020 (CMHIv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
3021 def : InstAlias<"{cmlo\t$dst.4s, $src1.4s, $src2.4s" #
3022 "|cmlo.4s\t$dst, $src1, $src2}",
3023 (CMHIv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
3024 def : InstAlias<"{cmlo\t$dst.2d, $src1.2d, $src2.2d" #
3025 "|cmlo.2d\t$dst, $src1, $src2}",
3026 (CMHIv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
3028 def : InstAlias<"{cmle\t$dst.8b, $src1.8b, $src2.8b" #
3029 "|cmle.8b\t$dst, $src1, $src2}",
3030 (CMGEv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
3031 def : InstAlias<"{cmle\t$dst.16b, $src1.16b, $src2.16b" #
3032 "|cmle.16b\t$dst, $src1, $src2}",
3033 (CMGEv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
3034 def : InstAlias<"{cmle\t$dst.4h, $src1.4h, $src2.4h" #
3035 "|cmle.4h\t$dst, $src1, $src2}",
3036 (CMGEv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
3037 def : InstAlias<"{cmle\t$dst.8h, $src1.8h, $src2.8h" #
3038 "|cmle.8h\t$dst, $src1, $src2}",
3039 (CMGEv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
3040 def : InstAlias<"{cmle\t$dst.2s, $src1.2s, $src2.2s" #
3041 "|cmle.2s\t$dst, $src1, $src2}",
3042 (CMGEv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
3043 def : InstAlias<"{cmle\t$dst.4s, $src1.4s, $src2.4s" #
3044 "|cmle.4s\t$dst, $src1, $src2}",
3045 (CMGEv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
3046 def : InstAlias<"{cmle\t$dst.2d, $src1.2d, $src2.2d" #
3047 "|cmle.2d\t$dst, $src1, $src2}",
3048 (CMGEv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
3050 def : InstAlias<"{cmlt\t$dst.8b, $src1.8b, $src2.8b" #
3051 "|cmlt.8b\t$dst, $src1, $src2}",
3052 (CMGTv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
3053 def : InstAlias<"{cmlt\t$dst.16b, $src1.16b, $src2.16b" #
3054 "|cmlt.16b\t$dst, $src1, $src2}",
3055 (CMGTv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
3056 def : InstAlias<"{cmlt\t$dst.4h, $src1.4h, $src2.4h" #
3057 "|cmlt.4h\t$dst, $src1, $src2}",
3058 (CMGTv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
3059 def : InstAlias<"{cmlt\t$dst.8h, $src1.8h, $src2.8h" #
3060 "|cmlt.8h\t$dst, $src1, $src2}",
3061 (CMGTv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
3062 def : InstAlias<"{cmlt\t$dst.2s, $src1.2s, $src2.2s" #
3063 "|cmlt.2s\t$dst, $src1, $src2}",
3064 (CMGTv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
3065 def : InstAlias<"{cmlt\t$dst.4s, $src1.4s, $src2.4s" #
3066 "|cmlt.4s\t$dst, $src1, $src2}",
3067 (CMGTv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
3068 def : InstAlias<"{cmlt\t$dst.2d, $src1.2d, $src2.2d" #
3069 "|cmlt.2d\t$dst, $src1, $src2}",
3070 (CMGTv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
3072 def : InstAlias<"{fcmle\t$dst.2s, $src1.2s, $src2.2s" #
3073 "|fcmle.2s\t$dst, $src1, $src2}",
3074 (FCMGEv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
3075 def : InstAlias<"{fcmle\t$dst.4s, $src1.4s, $src2.4s" #
3076 "|fcmle.4s\t$dst, $src1, $src2}",
3077 (FCMGEv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
3078 def : InstAlias<"{fcmle\t$dst.2d, $src1.2d, $src2.2d" #
3079 "|fcmle.2d\t$dst, $src1, $src2}",
3080 (FCMGEv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
3082 def : InstAlias<"{fcmlt\t$dst.2s, $src1.2s, $src2.2s" #
3083 "|fcmlt.2s\t$dst, $src1, $src2}",
3084 (FCMGTv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
3085 def : InstAlias<"{fcmlt\t$dst.4s, $src1.4s, $src2.4s" #
3086 "|fcmlt.4s\t$dst, $src1, $src2}",
3087 (FCMGTv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
3088 def : InstAlias<"{fcmlt\t$dst.2d, $src1.2d, $src2.2d" #
3089 "|fcmlt.2d\t$dst, $src1, $src2}",
3090 (FCMGTv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
3092 def : InstAlias<"{facle\t$dst.2s, $src1.2s, $src2.2s" #
3093 "|facle.2s\t$dst, $src1, $src2}",
3094 (FACGEv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
3095 def : InstAlias<"{facle\t$dst.4s, $src1.4s, $src2.4s" #
3096 "|facle.4s\t$dst, $src1, $src2}",
3097 (FACGEv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
3098 def : InstAlias<"{facle\t$dst.2d, $src1.2d, $src2.2d" #
3099 "|facle.2d\t$dst, $src1, $src2}",
3100 (FACGEv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
3102 def : InstAlias<"{faclt\t$dst.2s, $src1.2s, $src2.2s" #
3103 "|faclt.2s\t$dst, $src1, $src2}",
3104 (FACGTv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
3105 def : InstAlias<"{faclt\t$dst.4s, $src1.4s, $src2.4s" #
3106 "|faclt.4s\t$dst, $src1, $src2}",
3107 (FACGTv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
3108 def : InstAlias<"{faclt\t$dst.2d, $src1.2d, $src2.2d" #
3109 "|faclt.2d\t$dst, $src1, $src2}",
3110 (FACGTv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
3112 //===----------------------------------------------------------------------===//
3113 // Advanced SIMD three scalar instructions.
3114 //===----------------------------------------------------------------------===//
3116 defm ADD : SIMDThreeScalarD<0, 0b10000, "add", add>;
3117 defm CMEQ : SIMDThreeScalarD<1, 0b10001, "cmeq", AArch64cmeq>;
3118 defm CMGE : SIMDThreeScalarD<0, 0b00111, "cmge", AArch64cmge>;
3119 defm CMGT : SIMDThreeScalarD<0, 0b00110, "cmgt", AArch64cmgt>;
3120 defm CMHI : SIMDThreeScalarD<1, 0b00110, "cmhi", AArch64cmhi>;
3121 defm CMHS : SIMDThreeScalarD<1, 0b00111, "cmhs", AArch64cmhs>;
3122 defm CMTST : SIMDThreeScalarD<0, 0b10001, "cmtst", AArch64cmtst>;
3123 defm FABD : SIMDThreeScalarSD<1, 1, 0b11010, "fabd", int_aarch64_sisd_fabd>;
3124 def : Pat<(v1f64 (int_aarch64_neon_fabd (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
3125 (FABD64 FPR64:$Rn, FPR64:$Rm)>;
3126 defm FACGE : SIMDThreeScalarFPCmp<1, 0, 0b11101, "facge",
3127 int_aarch64_neon_facge>;
3128 defm FACGT : SIMDThreeScalarFPCmp<1, 1, 0b11101, "facgt",
3129 int_aarch64_neon_facgt>;
3130 defm FCMEQ : SIMDThreeScalarFPCmp<0, 0, 0b11100, "fcmeq", AArch64fcmeq>;
3131 defm FCMGE : SIMDThreeScalarFPCmp<1, 0, 0b11100, "fcmge", AArch64fcmge>;
3132 defm FCMGT : SIMDThreeScalarFPCmp<1, 1, 0b11100, "fcmgt", AArch64fcmgt>;
3133 defm FMULX : SIMDThreeScalarSD<0, 0, 0b11011, "fmulx", int_aarch64_neon_fmulx>;
3134 defm FRECPS : SIMDThreeScalarSD<0, 0, 0b11111, "frecps", int_aarch64_neon_frecps>;
3135 defm FRSQRTS : SIMDThreeScalarSD<0, 1, 0b11111, "frsqrts", int_aarch64_neon_frsqrts>;
3136 defm SQADD : SIMDThreeScalarBHSD<0, 0b00001, "sqadd", int_aarch64_neon_sqadd>;
3137 defm SQDMULH : SIMDThreeScalarHS< 0, 0b10110, "sqdmulh", int_aarch64_neon_sqdmulh>;
3138 defm SQRDMULH : SIMDThreeScalarHS< 1, 0b10110, "sqrdmulh", int_aarch64_neon_sqrdmulh>;
3139 defm SQRSHL : SIMDThreeScalarBHSD<0, 0b01011, "sqrshl",int_aarch64_neon_sqrshl>;
3140 defm SQSHL : SIMDThreeScalarBHSD<0, 0b01001, "sqshl", int_aarch64_neon_sqshl>;
3141 defm SQSUB : SIMDThreeScalarBHSD<0, 0b00101, "sqsub", int_aarch64_neon_sqsub>;
3142 defm SRSHL : SIMDThreeScalarD< 0, 0b01010, "srshl", int_aarch64_neon_srshl>;
3143 defm SSHL : SIMDThreeScalarD< 0, 0b01000, "sshl", int_aarch64_neon_sshl>;
3144 defm SUB : SIMDThreeScalarD< 1, 0b10000, "sub", sub>;
3145 defm UQADD : SIMDThreeScalarBHSD<1, 0b00001, "uqadd", int_aarch64_neon_uqadd>;
3146 defm UQRSHL : SIMDThreeScalarBHSD<1, 0b01011, "uqrshl",int_aarch64_neon_uqrshl>;
3147 defm UQSHL : SIMDThreeScalarBHSD<1, 0b01001, "uqshl", int_aarch64_neon_uqshl>;
3148 defm UQSUB : SIMDThreeScalarBHSD<1, 0b00101, "uqsub", int_aarch64_neon_uqsub>;
3149 defm URSHL : SIMDThreeScalarD< 1, 0b01010, "urshl", int_aarch64_neon_urshl>;
3150 defm USHL : SIMDThreeScalarD< 1, 0b01000, "ushl", int_aarch64_neon_ushl>;
3151 let Predicates = [HasV8_1a] in {
3152 defm SQRDMLAH : SIMDThreeScalarHSTied<1, 0, 0b10000, "sqrdmlah">;
3153 defm SQRDMLSH : SIMDThreeScalarHSTied<1, 0, 0b10001, "sqrdmlsh">;
3154 def : Pat<(i32 (int_aarch64_neon_sqadd
3156 (i32 (int_aarch64_neon_sqrdmulh (i32 FPR32:$Rn),
3157 (i32 FPR32:$Rm))))),
3158 (SQRDMLAHv1i32 FPR32:$Rd, FPR32:$Rn, FPR32:$Rm)>;
3159 def : Pat<(i32 (int_aarch64_neon_sqsub
3161 (i32 (int_aarch64_neon_sqrdmulh (i32 FPR32:$Rn),
3162 (i32 FPR32:$Rm))))),
3163 (SQRDMLSHv1i32 FPR32:$Rd, FPR32:$Rn, FPR32:$Rm)>;
3166 def : InstAlias<"cmls $dst, $src1, $src2",
3167 (CMHSv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
3168 def : InstAlias<"cmle $dst, $src1, $src2",
3169 (CMGEv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
3170 def : InstAlias<"cmlo $dst, $src1, $src2",
3171 (CMHIv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
3172 def : InstAlias<"cmlt $dst, $src1, $src2",
3173 (CMGTv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
3174 def : InstAlias<"fcmle $dst, $src1, $src2",
3175 (FCMGE32 FPR32:$dst, FPR32:$src2, FPR32:$src1), 0>;
3176 def : InstAlias<"fcmle $dst, $src1, $src2",
3177 (FCMGE64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
3178 def : InstAlias<"fcmlt $dst, $src1, $src2",
3179 (FCMGT32 FPR32:$dst, FPR32:$src2, FPR32:$src1), 0>;
3180 def : InstAlias<"fcmlt $dst, $src1, $src2",
3181 (FCMGT64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
3182 def : InstAlias<"facle $dst, $src1, $src2",
3183 (FACGE32 FPR32:$dst, FPR32:$src2, FPR32:$src1), 0>;
3184 def : InstAlias<"facle $dst, $src1, $src2",
3185 (FACGE64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
3186 def : InstAlias<"faclt $dst, $src1, $src2",
3187 (FACGT32 FPR32:$dst, FPR32:$src2, FPR32:$src1), 0>;
3188 def : InstAlias<"faclt $dst, $src1, $src2",
3189 (FACGT64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
3191 //===----------------------------------------------------------------------===//
3192 // Advanced SIMD three scalar instructions (mixed operands).
3193 //===----------------------------------------------------------------------===//
3194 defm SQDMULL : SIMDThreeScalarMixedHS<0, 0b11010, "sqdmull",
3195 int_aarch64_neon_sqdmulls_scalar>;
3196 defm SQDMLAL : SIMDThreeScalarMixedTiedHS<0, 0b10010, "sqdmlal">;
3197 defm SQDMLSL : SIMDThreeScalarMixedTiedHS<0, 0b10110, "sqdmlsl">;
3199 def : Pat<(i64 (int_aarch64_neon_sqadd (i64 FPR64:$Rd),
3200 (i64 (int_aarch64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
3201 (i32 FPR32:$Rm))))),
3202 (SQDMLALi32 FPR64:$Rd, FPR32:$Rn, FPR32:$Rm)>;
3203 def : Pat<(i64 (int_aarch64_neon_sqsub (i64 FPR64:$Rd),
3204 (i64 (int_aarch64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
3205 (i32 FPR32:$Rm))))),
3206 (SQDMLSLi32 FPR64:$Rd, FPR32:$Rn, FPR32:$Rm)>;
3208 //===----------------------------------------------------------------------===//
3209 // Advanced SIMD two scalar instructions.
3210 //===----------------------------------------------------------------------===//
3212 defm ABS : SIMDTwoScalarD< 0, 0b01011, "abs", int_aarch64_neon_abs>;
3213 defm CMEQ : SIMDCmpTwoScalarD< 0, 0b01001, "cmeq", AArch64cmeqz>;
3214 defm CMGE : SIMDCmpTwoScalarD< 1, 0b01000, "cmge", AArch64cmgez>;
3215 defm CMGT : SIMDCmpTwoScalarD< 0, 0b01000, "cmgt", AArch64cmgtz>;
3216 defm CMLE : SIMDCmpTwoScalarD< 1, 0b01001, "cmle", AArch64cmlez>;
3217 defm CMLT : SIMDCmpTwoScalarD< 0, 0b01010, "cmlt", AArch64cmltz>;
3218 defm FCMEQ : SIMDFPCmpTwoScalar<0, 1, 0b01101, "fcmeq", AArch64fcmeqz>;
3219 defm FCMGE : SIMDFPCmpTwoScalar<1, 1, 0b01100, "fcmge", AArch64fcmgez>;
3220 defm FCMGT : SIMDFPCmpTwoScalar<0, 1, 0b01100, "fcmgt", AArch64fcmgtz>;
3221 defm FCMLE : SIMDFPCmpTwoScalar<1, 1, 0b01101, "fcmle", AArch64fcmlez>;
3222 defm FCMLT : SIMDFPCmpTwoScalar<0, 1, 0b01110, "fcmlt", AArch64fcmltz>;
3223 defm FCVTAS : SIMDFPTwoScalar< 0, 0, 0b11100, "fcvtas">;
3224 defm FCVTAU : SIMDFPTwoScalar< 1, 0, 0b11100, "fcvtau">;
3225 defm FCVTMS : SIMDFPTwoScalar< 0, 0, 0b11011, "fcvtms">;
3226 defm FCVTMU : SIMDFPTwoScalar< 1, 0, 0b11011, "fcvtmu">;
3227 defm FCVTNS : SIMDFPTwoScalar< 0, 0, 0b11010, "fcvtns">;
3228 defm FCVTNU : SIMDFPTwoScalar< 1, 0, 0b11010, "fcvtnu">;
3229 defm FCVTPS : SIMDFPTwoScalar< 0, 1, 0b11010, "fcvtps">;
3230 defm FCVTPU : SIMDFPTwoScalar< 1, 1, 0b11010, "fcvtpu">;
3231 def FCVTXNv1i64 : SIMDInexactCvtTwoScalar<0b10110, "fcvtxn">;
3232 defm FCVTZS : SIMDFPTwoScalar< 0, 1, 0b11011, "fcvtzs">;
3233 defm FCVTZU : SIMDFPTwoScalar< 1, 1, 0b11011, "fcvtzu">;
3234 defm FRECPE : SIMDFPTwoScalar< 0, 1, 0b11101, "frecpe">;
3235 defm FRECPX : SIMDFPTwoScalar< 0, 1, 0b11111, "frecpx">;
3236 defm FRSQRTE : SIMDFPTwoScalar< 1, 1, 0b11101, "frsqrte">;
3237 defm NEG : SIMDTwoScalarD< 1, 0b01011, "neg",
3238 UnOpFrag<(sub immAllZerosV, node:$LHS)> >;
3239 defm SCVTF : SIMDTwoScalarCVTSD< 0, 0, 0b11101, "scvtf", AArch64sitof>;
3240 defm SQABS : SIMDTwoScalarBHSD< 0, 0b00111, "sqabs", int_aarch64_neon_sqabs>;
3241 defm SQNEG : SIMDTwoScalarBHSD< 1, 0b00111, "sqneg", int_aarch64_neon_sqneg>;
3242 defm SQXTN : SIMDTwoScalarMixedBHS< 0, 0b10100, "sqxtn", int_aarch64_neon_scalar_sqxtn>;
3243 defm SQXTUN : SIMDTwoScalarMixedBHS< 1, 0b10010, "sqxtun", int_aarch64_neon_scalar_sqxtun>;
3244 defm SUQADD : SIMDTwoScalarBHSDTied< 0, 0b00011, "suqadd",
3245 int_aarch64_neon_suqadd>;
3246 defm UCVTF : SIMDTwoScalarCVTSD< 1, 0, 0b11101, "ucvtf", AArch64uitof>;
3247 defm UQXTN : SIMDTwoScalarMixedBHS<1, 0b10100, "uqxtn", int_aarch64_neon_scalar_uqxtn>;
3248 defm USQADD : SIMDTwoScalarBHSDTied< 1, 0b00011, "usqadd",
3249 int_aarch64_neon_usqadd>;
3251 def : Pat<(AArch64neg (v1i64 V64:$Rn)), (NEGv1i64 V64:$Rn)>;
3253 def : Pat<(v1i64 (int_aarch64_neon_fcvtas (v1f64 FPR64:$Rn))),
3254 (FCVTASv1i64 FPR64:$Rn)>;
3255 def : Pat<(v1i64 (int_aarch64_neon_fcvtau (v1f64 FPR64:$Rn))),
3256 (FCVTAUv1i64 FPR64:$Rn)>;
3257 def : Pat<(v1i64 (int_aarch64_neon_fcvtms (v1f64 FPR64:$Rn))),
3258 (FCVTMSv1i64 FPR64:$Rn)>;
3259 def : Pat<(v1i64 (int_aarch64_neon_fcvtmu (v1f64 FPR64:$Rn))),
3260 (FCVTMUv1i64 FPR64:$Rn)>;
3261 def : Pat<(v1i64 (int_aarch64_neon_fcvtns (v1f64 FPR64:$Rn))),
3262 (FCVTNSv1i64 FPR64:$Rn)>;
3263 def : Pat<(v1i64 (int_aarch64_neon_fcvtnu (v1f64 FPR64:$Rn))),
3264 (FCVTNUv1i64 FPR64:$Rn)>;
3265 def : Pat<(v1i64 (int_aarch64_neon_fcvtps (v1f64 FPR64:$Rn))),
3266 (FCVTPSv1i64 FPR64:$Rn)>;
3267 def : Pat<(v1i64 (int_aarch64_neon_fcvtpu (v1f64 FPR64:$Rn))),
3268 (FCVTPUv1i64 FPR64:$Rn)>;
3270 def : Pat<(f32 (int_aarch64_neon_frecpe (f32 FPR32:$Rn))),
3271 (FRECPEv1i32 FPR32:$Rn)>;
3272 def : Pat<(f64 (int_aarch64_neon_frecpe (f64 FPR64:$Rn))),
3273 (FRECPEv1i64 FPR64:$Rn)>;
3274 def : Pat<(v1f64 (int_aarch64_neon_frecpe (v1f64 FPR64:$Rn))),
3275 (FRECPEv1i64 FPR64:$Rn)>;
3277 def : Pat<(f32 (int_aarch64_neon_frecpx (f32 FPR32:$Rn))),
3278 (FRECPXv1i32 FPR32:$Rn)>;
3279 def : Pat<(f64 (int_aarch64_neon_frecpx (f64 FPR64:$Rn))),
3280 (FRECPXv1i64 FPR64:$Rn)>;
3282 def : Pat<(f32 (int_aarch64_neon_frsqrte (f32 FPR32:$Rn))),
3283 (FRSQRTEv1i32 FPR32:$Rn)>;
3284 def : Pat<(f64 (int_aarch64_neon_frsqrte (f64 FPR64:$Rn))),
3285 (FRSQRTEv1i64 FPR64:$Rn)>;
3286 def : Pat<(v1f64 (int_aarch64_neon_frsqrte (v1f64 FPR64:$Rn))),
3287 (FRSQRTEv1i64 FPR64:$Rn)>;
3289 // If an integer is about to be converted to a floating point value,
3290 // just load it on the floating point unit.
3291 // Here are the patterns for 8 and 16-bits to float.
3293 multiclass UIntToFPROLoadPat<ValueType DstTy, ValueType SrcTy,
3294 SDPatternOperator loadop, Instruction UCVTF,
3295 ROAddrMode ro, Instruction LDRW, Instruction LDRX,
3297 def : Pat<(DstTy (uint_to_fp (SrcTy
3298 (loadop (ro.Wpat GPR64sp:$Rn, GPR32:$Rm,
3299 ro.Wext:$extend))))),
3300 (UCVTF (INSERT_SUBREG (DstTy (IMPLICIT_DEF)),
3301 (LDRW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend),
3304 def : Pat<(DstTy (uint_to_fp (SrcTy
3305 (loadop (ro.Xpat GPR64sp:$Rn, GPR64:$Rm,
3306 ro.Wext:$extend))))),
3307 (UCVTF (INSERT_SUBREG (DstTy (IMPLICIT_DEF)),
3308 (LDRX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend),
3312 defm : UIntToFPROLoadPat<f32, i32, zextloadi8,
3313 UCVTFv1i32, ro8, LDRBroW, LDRBroX, bsub>;
3314 def : Pat <(f32 (uint_to_fp (i32
3315 (zextloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))))),
3316 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
3317 (LDRBui GPR64sp:$Rn, uimm12s1:$offset), bsub))>;
3318 def : Pat <(f32 (uint_to_fp (i32
3319 (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))))),
3320 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
3321 (LDURBi GPR64sp:$Rn, simm9:$offset), bsub))>;
3322 // 16-bits -> float.
3323 defm : UIntToFPROLoadPat<f32, i32, zextloadi16,
3324 UCVTFv1i32, ro16, LDRHroW, LDRHroX, hsub>;
3325 def : Pat <(f32 (uint_to_fp (i32
3326 (zextloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))))),
3327 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
3328 (LDRHui GPR64sp:$Rn, uimm12s2:$offset), hsub))>;
3329 def : Pat <(f32 (uint_to_fp (i32
3330 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))))),
3331 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
3332 (LDURHi GPR64sp:$Rn, simm9:$offset), hsub))>;
3333 // 32-bits are handled in target specific dag combine:
3334 // performIntToFpCombine.
3335 // 64-bits integer to 32-bits floating point, not possible with
3336 // UCVTF on floating point registers (both source and destination
3337 // must have the same size).
3339 // Here are the patterns for 8, 16, 32, and 64-bits to double.
3340 // 8-bits -> double.
3341 defm : UIntToFPROLoadPat<f64, i32, zextloadi8,
3342 UCVTFv1i64, ro8, LDRBroW, LDRBroX, bsub>;
3343 def : Pat <(f64 (uint_to_fp (i32
3344 (zextloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))))),
3345 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3346 (LDRBui GPR64sp:$Rn, uimm12s1:$offset), bsub))>;
3347 def : Pat <(f64 (uint_to_fp (i32
3348 (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))))),
3349 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3350 (LDURBi GPR64sp:$Rn, simm9:$offset), bsub))>;
3351 // 16-bits -> double.
3352 defm : UIntToFPROLoadPat<f64, i32, zextloadi16,
3353 UCVTFv1i64, ro16, LDRHroW, LDRHroX, hsub>;
3354 def : Pat <(f64 (uint_to_fp (i32
3355 (zextloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))))),
3356 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3357 (LDRHui GPR64sp:$Rn, uimm12s2:$offset), hsub))>;
3358 def : Pat <(f64 (uint_to_fp (i32
3359 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))))),
3360 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3361 (LDURHi GPR64sp:$Rn, simm9:$offset), hsub))>;
3362 // 32-bits -> double.
3363 defm : UIntToFPROLoadPat<f64, i32, load,
3364 UCVTFv1i64, ro32, LDRSroW, LDRSroX, ssub>;
3365 def : Pat <(f64 (uint_to_fp (i32
3366 (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))))),
3367 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3368 (LDRSui GPR64sp:$Rn, uimm12s4:$offset), ssub))>;
3369 def : Pat <(f64 (uint_to_fp (i32
3370 (load (am_unscaled32 GPR64sp:$Rn, simm9:$offset))))),
3371 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3372 (LDURSi GPR64sp:$Rn, simm9:$offset), ssub))>;
3373 // 64-bits -> double are handled in target specific dag combine:
3374 // performIntToFpCombine.
3376 //===----------------------------------------------------------------------===//
3377 // Advanced SIMD three different-sized vector instructions.
3378 //===----------------------------------------------------------------------===//
3380 defm ADDHN : SIMDNarrowThreeVectorBHS<0,0b0100,"addhn", int_aarch64_neon_addhn>;
3381 defm SUBHN : SIMDNarrowThreeVectorBHS<0,0b0110,"subhn", int_aarch64_neon_subhn>;
3382 defm RADDHN : SIMDNarrowThreeVectorBHS<1,0b0100,"raddhn",int_aarch64_neon_raddhn>;
3383 defm RSUBHN : SIMDNarrowThreeVectorBHS<1,0b0110,"rsubhn",int_aarch64_neon_rsubhn>;
3384 defm PMULL : SIMDDifferentThreeVectorBD<0,0b1110,"pmull",int_aarch64_neon_pmull>;
3385 defm SABAL : SIMDLongThreeVectorTiedBHSabal<0,0b0101,"sabal",
3387 defm SABDL : SIMDLongThreeVectorBHSabdl<0, 0b0111, "sabdl",
3389 defm SADDL : SIMDLongThreeVectorBHS< 0, 0b0000, "saddl",
3390 BinOpFrag<(add (sext node:$LHS), (sext node:$RHS))>>;
3391 defm SADDW : SIMDWideThreeVectorBHS< 0, 0b0001, "saddw",
3392 BinOpFrag<(add node:$LHS, (sext node:$RHS))>>;
3393 defm SMLAL : SIMDLongThreeVectorTiedBHS<0, 0b1000, "smlal",
3394 TriOpFrag<(add node:$LHS, (int_aarch64_neon_smull node:$MHS, node:$RHS))>>;
3395 defm SMLSL : SIMDLongThreeVectorTiedBHS<0, 0b1010, "smlsl",
3396 TriOpFrag<(sub node:$LHS, (int_aarch64_neon_smull node:$MHS, node:$RHS))>>;
3397 defm SMULL : SIMDLongThreeVectorBHS<0, 0b1100, "smull", int_aarch64_neon_smull>;
3398 defm SQDMLAL : SIMDLongThreeVectorSQDMLXTiedHS<0, 0b1001, "sqdmlal",
3399 int_aarch64_neon_sqadd>;
3400 defm SQDMLSL : SIMDLongThreeVectorSQDMLXTiedHS<0, 0b1011, "sqdmlsl",
3401 int_aarch64_neon_sqsub>;
3402 defm SQDMULL : SIMDLongThreeVectorHS<0, 0b1101, "sqdmull",
3403 int_aarch64_neon_sqdmull>;
3404 defm SSUBL : SIMDLongThreeVectorBHS<0, 0b0010, "ssubl",
3405 BinOpFrag<(sub (sext node:$LHS), (sext node:$RHS))>>;
3406 defm SSUBW : SIMDWideThreeVectorBHS<0, 0b0011, "ssubw",
3407 BinOpFrag<(sub node:$LHS, (sext node:$RHS))>>;
3408 defm UABAL : SIMDLongThreeVectorTiedBHSabal<1, 0b0101, "uabal",
3410 defm UABDL : SIMDLongThreeVectorBHSabdl<1, 0b0111, "uabdl",
3412 defm UADDL : SIMDLongThreeVectorBHS<1, 0b0000, "uaddl",
3413 BinOpFrag<(add (zext node:$LHS), (zext node:$RHS))>>;
3414 defm UADDW : SIMDWideThreeVectorBHS<1, 0b0001, "uaddw",
3415 BinOpFrag<(add node:$LHS, (zext node:$RHS))>>;
3416 defm UMLAL : SIMDLongThreeVectorTiedBHS<1, 0b1000, "umlal",
3417 TriOpFrag<(add node:$LHS, (int_aarch64_neon_umull node:$MHS, node:$RHS))>>;
3418 defm UMLSL : SIMDLongThreeVectorTiedBHS<1, 0b1010, "umlsl",
3419 TriOpFrag<(sub node:$LHS, (int_aarch64_neon_umull node:$MHS, node:$RHS))>>;
3420 defm UMULL : SIMDLongThreeVectorBHS<1, 0b1100, "umull", int_aarch64_neon_umull>;
3421 defm USUBL : SIMDLongThreeVectorBHS<1, 0b0010, "usubl",
3422 BinOpFrag<(sub (zext node:$LHS), (zext node:$RHS))>>;
3423 defm USUBW : SIMDWideThreeVectorBHS< 1, 0b0011, "usubw",
3424 BinOpFrag<(sub node:$LHS, (zext node:$RHS))>>;
3426 // Additional patterns for SMULL and UMULL
3427 multiclass Neon_mul_widen_patterns<SDPatternOperator opnode,
3428 Instruction INST8B, Instruction INST4H, Instruction INST2S> {
3429 def : Pat<(v8i16 (opnode (v8i8 V64:$Rn), (v8i8 V64:$Rm))),
3430 (INST8B V64:$Rn, V64:$Rm)>;
3431 def : Pat<(v4i32 (opnode (v4i16 V64:$Rn), (v4i16 V64:$Rm))),
3432 (INST4H V64:$Rn, V64:$Rm)>;
3433 def : Pat<(v2i64 (opnode (v2i32 V64:$Rn), (v2i32 V64:$Rm))),
3434 (INST2S V64:$Rn, V64:$Rm)>;
3437 defm : Neon_mul_widen_patterns<AArch64smull, SMULLv8i8_v8i16,
3438 SMULLv4i16_v4i32, SMULLv2i32_v2i64>;
3439 defm : Neon_mul_widen_patterns<AArch64umull, UMULLv8i8_v8i16,
3440 UMULLv4i16_v4i32, UMULLv2i32_v2i64>;
3442 // Additional patterns for SMLAL/SMLSL and UMLAL/UMLSL
3443 multiclass Neon_mulacc_widen_patterns<SDPatternOperator opnode,
3444 Instruction INST8B, Instruction INST4H, Instruction INST2S> {
3445 def : Pat<(v8i16 (opnode (v8i16 V128:$Rd), (v8i8 V64:$Rn), (v8i8 V64:$Rm))),
3446 (INST8B V128:$Rd, V64:$Rn, V64:$Rm)>;
3447 def : Pat<(v4i32 (opnode (v4i32 V128:$Rd), (v4i16 V64:$Rn), (v4i16 V64:$Rm))),
3448 (INST4H V128:$Rd, V64:$Rn, V64:$Rm)>;
3449 def : Pat<(v2i64 (opnode (v2i64 V128:$Rd), (v2i32 V64:$Rn), (v2i32 V64:$Rm))),
3450 (INST2S V128:$Rd, V64:$Rn, V64:$Rm)>;
3453 defm : Neon_mulacc_widen_patterns<
3454 TriOpFrag<(add node:$LHS, (AArch64smull node:$MHS, node:$RHS))>,
3455 SMLALv8i8_v8i16, SMLALv4i16_v4i32, SMLALv2i32_v2i64>;
3456 defm : Neon_mulacc_widen_patterns<
3457 TriOpFrag<(add node:$LHS, (AArch64umull node:$MHS, node:$RHS))>,
3458 UMLALv8i8_v8i16, UMLALv4i16_v4i32, UMLALv2i32_v2i64>;
3459 defm : Neon_mulacc_widen_patterns<
3460 TriOpFrag<(sub node:$LHS, (AArch64smull node:$MHS, node:$RHS))>,
3461 SMLSLv8i8_v8i16, SMLSLv4i16_v4i32, SMLSLv2i32_v2i64>;
3462 defm : Neon_mulacc_widen_patterns<
3463 TriOpFrag<(sub node:$LHS, (AArch64umull node:$MHS, node:$RHS))>,
3464 UMLSLv8i8_v8i16, UMLSLv4i16_v4i32, UMLSLv2i32_v2i64>;
3466 // Patterns for 64-bit pmull
3467 def : Pat<(int_aarch64_neon_pmull64 V64:$Rn, V64:$Rm),
3468 (PMULLv1i64 V64:$Rn, V64:$Rm)>;
3469 def : Pat<(int_aarch64_neon_pmull64 (vector_extract (v2i64 V128:$Rn), (i64 1)),
3470 (vector_extract (v2i64 V128:$Rm), (i64 1))),
3471 (PMULLv2i64 V128:$Rn, V128:$Rm)>;
3473 // CodeGen patterns for addhn and subhn instructions, which can actually be
3474 // written in LLVM IR without too much difficulty.
3477 def : Pat<(v8i8 (trunc (v8i16 (AArch64vlshr (add V128:$Rn, V128:$Rm), (i32 8))))),
3478 (ADDHNv8i16_v8i8 V128:$Rn, V128:$Rm)>;
3479 def : Pat<(v4i16 (trunc (v4i32 (AArch64vlshr (add V128:$Rn, V128:$Rm),
3481 (ADDHNv4i32_v4i16 V128:$Rn, V128:$Rm)>;
3482 def : Pat<(v2i32 (trunc (v2i64 (AArch64vlshr (add V128:$Rn, V128:$Rm),
3484 (ADDHNv2i64_v2i32 V128:$Rn, V128:$Rm)>;
3485 def : Pat<(concat_vectors (v8i8 V64:$Rd),
3486 (trunc (v8i16 (AArch64vlshr (add V128:$Rn, V128:$Rm),
3488 (ADDHNv8i16_v16i8 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3489 V128:$Rn, V128:$Rm)>;
3490 def : Pat<(concat_vectors (v4i16 V64:$Rd),
3491 (trunc (v4i32 (AArch64vlshr (add V128:$Rn, V128:$Rm),
3493 (ADDHNv4i32_v8i16 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3494 V128:$Rn, V128:$Rm)>;
3495 def : Pat<(concat_vectors (v2i32 V64:$Rd),
3496 (trunc (v2i64 (AArch64vlshr (add V128:$Rn, V128:$Rm),
3498 (ADDHNv2i64_v4i32 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3499 V128:$Rn, V128:$Rm)>;
3502 def : Pat<(v8i8 (trunc (v8i16 (AArch64vlshr (sub V128:$Rn, V128:$Rm), (i32 8))))),
3503 (SUBHNv8i16_v8i8 V128:$Rn, V128:$Rm)>;
3504 def : Pat<(v4i16 (trunc (v4i32 (AArch64vlshr (sub V128:$Rn, V128:$Rm),
3506 (SUBHNv4i32_v4i16 V128:$Rn, V128:$Rm)>;
3507 def : Pat<(v2i32 (trunc (v2i64 (AArch64vlshr (sub V128:$Rn, V128:$Rm),
3509 (SUBHNv2i64_v2i32 V128:$Rn, V128:$Rm)>;
3510 def : Pat<(concat_vectors (v8i8 V64:$Rd),
3511 (trunc (v8i16 (AArch64vlshr (sub V128:$Rn, V128:$Rm),
3513 (SUBHNv8i16_v16i8 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3514 V128:$Rn, V128:$Rm)>;
3515 def : Pat<(concat_vectors (v4i16 V64:$Rd),
3516 (trunc (v4i32 (AArch64vlshr (sub V128:$Rn, V128:$Rm),
3518 (SUBHNv4i32_v8i16 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3519 V128:$Rn, V128:$Rm)>;
3520 def : Pat<(concat_vectors (v2i32 V64:$Rd),
3521 (trunc (v2i64 (AArch64vlshr (sub V128:$Rn, V128:$Rm),
3523 (SUBHNv2i64_v4i32 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3524 V128:$Rn, V128:$Rm)>;
3526 //----------------------------------------------------------------------------
3527 // AdvSIMD bitwise extract from vector instruction.
3528 //----------------------------------------------------------------------------
3530 defm EXT : SIMDBitwiseExtract<"ext">;
3532 def : Pat<(v4i16 (AArch64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
3533 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
3534 def : Pat<(v8i16 (AArch64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3535 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3536 def : Pat<(v2i32 (AArch64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
3537 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
3538 def : Pat<(v2f32 (AArch64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
3539 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
3540 def : Pat<(v4i32 (AArch64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3541 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3542 def : Pat<(v4f32 (AArch64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3543 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3544 def : Pat<(v2i64 (AArch64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3545 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3546 def : Pat<(v2f64 (AArch64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3547 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3548 def : Pat<(v4f16 (AArch64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
3549 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
3550 def : Pat<(v8f16 (AArch64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3551 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3553 // We use EXT to handle extract_subvector to copy the upper 64-bits of a
3555 def : Pat<(v8i8 (extract_subvector V128:$Rn, (i64 8))),
3556 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3557 def : Pat<(v4i16 (extract_subvector V128:$Rn, (i64 4))),
3558 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3559 def : Pat<(v2i32 (extract_subvector V128:$Rn, (i64 2))),
3560 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3561 def : Pat<(v1i64 (extract_subvector V128:$Rn, (i64 1))),
3562 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3563 def : Pat<(v4f16 (extract_subvector V128:$Rn, (i64 4))),
3564 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3565 def : Pat<(v2f32 (extract_subvector V128:$Rn, (i64 2))),
3566 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3567 def : Pat<(v1f64 (extract_subvector V128:$Rn, (i64 1))),
3568 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3571 //----------------------------------------------------------------------------
3572 // AdvSIMD zip vector
3573 //----------------------------------------------------------------------------
3575 defm TRN1 : SIMDZipVector<0b010, "trn1", AArch64trn1>;
3576 defm TRN2 : SIMDZipVector<0b110, "trn2", AArch64trn2>;
3577 defm UZP1 : SIMDZipVector<0b001, "uzp1", AArch64uzp1>;
3578 defm UZP2 : SIMDZipVector<0b101, "uzp2", AArch64uzp2>;
3579 defm ZIP1 : SIMDZipVector<0b011, "zip1", AArch64zip1>;
3580 defm ZIP2 : SIMDZipVector<0b111, "zip2", AArch64zip2>;
3582 //----------------------------------------------------------------------------
3583 // AdvSIMD TBL/TBX instructions
3584 //----------------------------------------------------------------------------
3586 defm TBL : SIMDTableLookup< 0, "tbl">;
3587 defm TBX : SIMDTableLookupTied<1, "tbx">;
3589 def : Pat<(v8i8 (int_aarch64_neon_tbl1 (v16i8 VecListOne128:$Rn), (v8i8 V64:$Ri))),
3590 (TBLv8i8One VecListOne128:$Rn, V64:$Ri)>;
3591 def : Pat<(v16i8 (int_aarch64_neon_tbl1 (v16i8 V128:$Ri), (v16i8 V128:$Rn))),
3592 (TBLv16i8One V128:$Ri, V128:$Rn)>;
3594 def : Pat<(v8i8 (int_aarch64_neon_tbx1 (v8i8 V64:$Rd),
3595 (v16i8 VecListOne128:$Rn), (v8i8 V64:$Ri))),
3596 (TBXv8i8One V64:$Rd, VecListOne128:$Rn, V64:$Ri)>;
3597 def : Pat<(v16i8 (int_aarch64_neon_tbx1 (v16i8 V128:$Rd),
3598 (v16i8 V128:$Ri), (v16i8 V128:$Rn))),
3599 (TBXv16i8One V128:$Rd, V128:$Ri, V128:$Rn)>;
3602 //----------------------------------------------------------------------------
3603 // AdvSIMD scalar CPY instruction
3604 //----------------------------------------------------------------------------
3606 defm CPY : SIMDScalarCPY<"cpy">;
3608 //----------------------------------------------------------------------------
3609 // AdvSIMD scalar pairwise instructions
3610 //----------------------------------------------------------------------------
3612 defm ADDP : SIMDPairwiseScalarD<0, 0b11011, "addp">;
3613 defm FADDP : SIMDFPPairwiseScalar<1, 0, 0b01101, "faddp">;
3614 defm FMAXNMP : SIMDFPPairwiseScalar<1, 0, 0b01100, "fmaxnmp">;
3615 defm FMAXP : SIMDFPPairwiseScalar<1, 0, 0b01111, "fmaxp">;
3616 defm FMINNMP : SIMDFPPairwiseScalar<1, 1, 0b01100, "fminnmp">;
3617 defm FMINP : SIMDFPPairwiseScalar<1, 1, 0b01111, "fminp">;
3618 def : Pat<(v2i64 (AArch64saddv V128:$Rn)),
3619 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), (ADDPv2i64p V128:$Rn), dsub)>;
3620 def : Pat<(v2i64 (AArch64uaddv V128:$Rn)),
3621 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), (ADDPv2i64p V128:$Rn), dsub)>;
3622 def : Pat<(f32 (int_aarch64_neon_faddv (v2f32 V64:$Rn))),
3623 (FADDPv2i32p V64:$Rn)>;
3624 def : Pat<(f32 (int_aarch64_neon_faddv (v4f32 V128:$Rn))),
3625 (FADDPv2i32p (EXTRACT_SUBREG (FADDPv4f32 V128:$Rn, V128:$Rn), dsub))>;
3626 def : Pat<(f64 (int_aarch64_neon_faddv (v2f64 V128:$Rn))),
3627 (FADDPv2i64p V128:$Rn)>;
3628 def : Pat<(f32 (int_aarch64_neon_fmaxnmv (v2f32 V64:$Rn))),
3629 (FMAXNMPv2i32p V64:$Rn)>;
3630 def : Pat<(f64 (int_aarch64_neon_fmaxnmv (v2f64 V128:$Rn))),
3631 (FMAXNMPv2i64p V128:$Rn)>;
3632 def : Pat<(f32 (int_aarch64_neon_fmaxv (v2f32 V64:$Rn))),
3633 (FMAXPv2i32p V64:$Rn)>;
3634 def : Pat<(f64 (int_aarch64_neon_fmaxv (v2f64 V128:$Rn))),
3635 (FMAXPv2i64p V128:$Rn)>;
3636 def : Pat<(f32 (int_aarch64_neon_fminnmv (v2f32 V64:$Rn))),
3637 (FMINNMPv2i32p V64:$Rn)>;
3638 def : Pat<(f64 (int_aarch64_neon_fminnmv (v2f64 V128:$Rn))),
3639 (FMINNMPv2i64p V128:$Rn)>;
3640 def : Pat<(f32 (int_aarch64_neon_fminv (v2f32 V64:$Rn))),
3641 (FMINPv2i32p V64:$Rn)>;
3642 def : Pat<(f64 (int_aarch64_neon_fminv (v2f64 V128:$Rn))),
3643 (FMINPv2i64p V128:$Rn)>;
3645 //----------------------------------------------------------------------------
3646 // AdvSIMD INS/DUP instructions
3647 //----------------------------------------------------------------------------
3649 def DUPv8i8gpr : SIMDDupFromMain<0, {?,?,?,?,1}, ".8b", v8i8, V64, GPR32>;
3650 def DUPv16i8gpr : SIMDDupFromMain<1, {?,?,?,?,1}, ".16b", v16i8, V128, GPR32>;
3651 def DUPv4i16gpr : SIMDDupFromMain<0, {?,?,?,1,0}, ".4h", v4i16, V64, GPR32>;
3652 def DUPv8i16gpr : SIMDDupFromMain<1, {?,?,?,1,0}, ".8h", v8i16, V128, GPR32>;
3653 def DUPv2i32gpr : SIMDDupFromMain<0, {?,?,1,0,0}, ".2s", v2i32, V64, GPR32>;
3654 def DUPv4i32gpr : SIMDDupFromMain<1, {?,?,1,0,0}, ".4s", v4i32, V128, GPR32>;
3655 def DUPv2i64gpr : SIMDDupFromMain<1, {?,1,0,0,0}, ".2d", v2i64, V128, GPR64>;
3657 def DUPv2i64lane : SIMDDup64FromElement;
3658 def DUPv2i32lane : SIMDDup32FromElement<0, ".2s", v2i32, V64>;
3659 def DUPv4i32lane : SIMDDup32FromElement<1, ".4s", v4i32, V128>;
3660 def DUPv4i16lane : SIMDDup16FromElement<0, ".4h", v4i16, V64>;
3661 def DUPv8i16lane : SIMDDup16FromElement<1, ".8h", v8i16, V128>;
3662 def DUPv8i8lane : SIMDDup8FromElement <0, ".8b", v8i8, V64>;
3663 def DUPv16i8lane : SIMDDup8FromElement <1, ".16b", v16i8, V128>;
3665 def : Pat<(v2f32 (AArch64dup (f32 FPR32:$Rn))),
3666 (v2f32 (DUPv2i32lane
3667 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rn, ssub),
3669 def : Pat<(v4f32 (AArch64dup (f32 FPR32:$Rn))),
3670 (v4f32 (DUPv4i32lane
3671 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rn, ssub),
3673 def : Pat<(v2f64 (AArch64dup (f64 FPR64:$Rn))),
3674 (v2f64 (DUPv2i64lane
3675 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$Rn, dsub),
3677 def : Pat<(v4f16 (AArch64dup (f16 FPR16:$Rn))),
3678 (v4f16 (DUPv4i16lane
3679 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), FPR16:$Rn, hsub),
3681 def : Pat<(v8f16 (AArch64dup (f16 FPR16:$Rn))),
3682 (v8f16 (DUPv8i16lane
3683 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), FPR16:$Rn, hsub),
3686 def : Pat<(v4f16 (AArch64duplane16 (v8f16 V128:$Rn), VectorIndexH:$imm)),
3687 (DUPv4i16lane V128:$Rn, VectorIndexH:$imm)>;
3688 def : Pat<(v8f16 (AArch64duplane16 (v8f16 V128:$Rn), VectorIndexH:$imm)),
3689 (DUPv8i16lane V128:$Rn, VectorIndexH:$imm)>;
3691 def : Pat<(v2f32 (AArch64duplane32 (v4f32 V128:$Rn), VectorIndexS:$imm)),
3692 (DUPv2i32lane V128:$Rn, VectorIndexS:$imm)>;
3693 def : Pat<(v4f32 (AArch64duplane32 (v4f32 V128:$Rn), VectorIndexS:$imm)),
3694 (DUPv4i32lane V128:$Rn, VectorIndexS:$imm)>;
3695 def : Pat<(v2f64 (AArch64duplane64 (v2f64 V128:$Rn), VectorIndexD:$imm)),
3696 (DUPv2i64lane V128:$Rn, VectorIndexD:$imm)>;
3698 // If there's an (AArch64dup (vector_extract ...) ...), we can use a duplane
3699 // instruction even if the types don't match: we just have to remap the lane
3700 // carefully. N.b. this trick only applies to truncations.
3701 def VecIndex_x2 : SDNodeXForm<imm, [{
3702 return CurDAG->getTargetConstant(2 * N->getZExtValue(), SDLoc(N), MVT::i64);
3704 def VecIndex_x4 : SDNodeXForm<imm, [{
3705 return CurDAG->getTargetConstant(4 * N->getZExtValue(), SDLoc(N), MVT::i64);
3707 def VecIndex_x8 : SDNodeXForm<imm, [{
3708 return CurDAG->getTargetConstant(8 * N->getZExtValue(), SDLoc(N), MVT::i64);
3711 multiclass DUPWithTruncPats<ValueType ResVT, ValueType Src64VT,
3712 ValueType Src128VT, ValueType ScalVT,
3713 Instruction DUP, SDNodeXForm IdxXFORM> {
3714 def : Pat<(ResVT (AArch64dup (ScalVT (vector_extract (Src128VT V128:$Rn),
3716 (DUP V128:$Rn, (IdxXFORM imm:$idx))>;
3718 def : Pat<(ResVT (AArch64dup (ScalVT (vector_extract (Src64VT V64:$Rn),
3720 (DUP (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), (IdxXFORM imm:$idx))>;
3723 defm : DUPWithTruncPats<v8i8, v4i16, v8i16, i32, DUPv8i8lane, VecIndex_x2>;
3724 defm : DUPWithTruncPats<v8i8, v2i32, v4i32, i32, DUPv8i8lane, VecIndex_x4>;
3725 defm : DUPWithTruncPats<v4i16, v2i32, v4i32, i32, DUPv4i16lane, VecIndex_x2>;
3727 defm : DUPWithTruncPats<v16i8, v4i16, v8i16, i32, DUPv16i8lane, VecIndex_x2>;
3728 defm : DUPWithTruncPats<v16i8, v2i32, v4i32, i32, DUPv16i8lane, VecIndex_x4>;
3729 defm : DUPWithTruncPats<v8i16, v2i32, v4i32, i32, DUPv8i16lane, VecIndex_x2>;
3731 multiclass DUPWithTrunci64Pats<ValueType ResVT, Instruction DUP,
3732 SDNodeXForm IdxXFORM> {
3733 def : Pat<(ResVT (AArch64dup (i32 (trunc (vector_extract (v2i64 V128:$Rn),
3735 (DUP V128:$Rn, (IdxXFORM imm:$idx))>;
3737 def : Pat<(ResVT (AArch64dup (i32 (trunc (vector_extract (v1i64 V64:$Rn),
3739 (DUP (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), (IdxXFORM imm:$idx))>;
3742 defm : DUPWithTrunci64Pats<v8i8, DUPv8i8lane, VecIndex_x8>;
3743 defm : DUPWithTrunci64Pats<v4i16, DUPv4i16lane, VecIndex_x4>;
3744 defm : DUPWithTrunci64Pats<v2i32, DUPv2i32lane, VecIndex_x2>;
3746 defm : DUPWithTrunci64Pats<v16i8, DUPv16i8lane, VecIndex_x8>;
3747 defm : DUPWithTrunci64Pats<v8i16, DUPv8i16lane, VecIndex_x4>;
3748 defm : DUPWithTrunci64Pats<v4i32, DUPv4i32lane, VecIndex_x2>;
3750 // SMOV and UMOV definitions, with some extra patterns for convenience
3754 def : Pat<(sext_inreg (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx), i8),
3755 (i32 (SMOVvi8to32 V128:$Rn, VectorIndexB:$idx))>;
3756 def : Pat<(sext_inreg (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx), i8),
3757 (i64 (SMOVvi8to64 V128:$Rn, VectorIndexB:$idx))>;
3758 def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
3759 (i32 (SMOVvi16to32 V128:$Rn, VectorIndexH:$idx))>;
3760 def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
3761 (i64 (SMOVvi16to64 V128:$Rn, VectorIndexH:$idx))>;
3762 def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
3763 (i32 (SMOVvi16to32 V128:$Rn, VectorIndexH:$idx))>;
3764 def : Pat<(sext (i32 (vector_extract (v4i32 V128:$Rn), VectorIndexS:$idx))),
3765 (i64 (SMOVvi32to64 V128:$Rn, VectorIndexS:$idx))>;
3767 // Extracting i8 or i16 elements will have the zero-extend transformed to
3768 // an 'and' mask by type legalization since neither i8 nor i16 are legal types
3769 // for AArch64. Match these patterns here since UMOV already zeroes out the high
3770 // bits of the destination register.
3771 def : Pat<(and (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx),
3773 (i32 (UMOVvi8 V128:$Rn, VectorIndexB:$idx))>;
3774 def : Pat<(and (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),
3776 (i32 (UMOVvi16 V128:$Rn, VectorIndexH:$idx))>;
3780 def : Pat<(v16i8 (scalar_to_vector GPR32:$Rn)),
3781 (SUBREG_TO_REG (i32 0),
3782 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3783 def : Pat<(v8i8 (scalar_to_vector GPR32:$Rn)),
3784 (SUBREG_TO_REG (i32 0),
3785 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3787 def : Pat<(v8i16 (scalar_to_vector GPR32:$Rn)),
3788 (SUBREG_TO_REG (i32 0),
3789 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3790 def : Pat<(v4i16 (scalar_to_vector GPR32:$Rn)),
3791 (SUBREG_TO_REG (i32 0),
3792 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3794 def : Pat<(v2i32 (scalar_to_vector (i32 FPR32:$Rn))),
3795 (v2i32 (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)),
3796 (i32 FPR32:$Rn), ssub))>;
3797 def : Pat<(v4i32 (scalar_to_vector (i32 FPR32:$Rn))),
3798 (v4i32 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
3799 (i32 FPR32:$Rn), ssub))>;
3800 def : Pat<(v2i64 (scalar_to_vector (i64 FPR64:$Rn))),
3801 (v2i64 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
3802 (i64 FPR64:$Rn), dsub))>;
3804 def : Pat<(v4f32 (scalar_to_vector (f32 FPR32:$Rn))),
3805 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rn, ssub)>;
3806 def : Pat<(v2f32 (scalar_to_vector (f32 FPR32:$Rn))),
3807 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), FPR32:$Rn, ssub)>;
3808 def : Pat<(v2f64 (scalar_to_vector (f64 FPR64:$Rn))),
3809 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$Rn, dsub)>;
3811 def : Pat<(v4f16 (vector_insert (v4f16 V64:$Rn),
3812 (f16 FPR16:$Rm), (i64 VectorIndexS:$imm))),
3815 (v8f16 (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), V64:$Rn, dsub)),
3817 (v8f16 (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), FPR16:$Rm, hsub)),
3821 def : Pat<(v8f16 (vector_insert (v8f16 V128:$Rn),
3822 (f16 FPR16:$Rm), (i64 VectorIndexH:$imm))),
3824 V128:$Rn, VectorIndexH:$imm,
3825 (v8f16 (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), FPR16:$Rm, hsub)),
3828 def : Pat<(v2f32 (vector_insert (v2f32 V64:$Rn),
3829 (f32 FPR32:$Rm), (i64 VectorIndexS:$imm))),
3832 (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), V64:$Rn, dsub)),
3834 (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rm, ssub)),
3837 def : Pat<(v4f32 (vector_insert (v4f32 V128:$Rn),
3838 (f32 FPR32:$Rm), (i64 VectorIndexS:$imm))),
3840 V128:$Rn, VectorIndexS:$imm,
3841 (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rm, ssub)),
3843 def : Pat<(v2f64 (vector_insert (v2f64 V128:$Rn),
3844 (f64 FPR64:$Rm), (i64 VectorIndexD:$imm))),
3846 V128:$Rn, VectorIndexD:$imm,
3847 (v2f64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$Rm, dsub)),
3850 // Copy an element at a constant index in one vector into a constant indexed
3851 // element of another.
3852 // FIXME refactor to a shared class/dev parameterized on vector type, vector
3853 // index type and INS extension
3854 def : Pat<(v16i8 (int_aarch64_neon_vcopy_lane
3855 (v16i8 V128:$Vd), VectorIndexB:$idx, (v16i8 V128:$Vs),
3856 VectorIndexB:$idx2)),
3858 V128:$Vd, VectorIndexB:$idx, V128:$Vs, VectorIndexB:$idx2)
3860 def : Pat<(v8i16 (int_aarch64_neon_vcopy_lane
3861 (v8i16 V128:$Vd), VectorIndexH:$idx, (v8i16 V128:$Vs),
3862 VectorIndexH:$idx2)),
3864 V128:$Vd, VectorIndexH:$idx, V128:$Vs, VectorIndexH:$idx2)
3866 def : Pat<(v4i32 (int_aarch64_neon_vcopy_lane
3867 (v4i32 V128:$Vd), VectorIndexS:$idx, (v4i32 V128:$Vs),
3868 VectorIndexS:$idx2)),
3870 V128:$Vd, VectorIndexS:$idx, V128:$Vs, VectorIndexS:$idx2)
3872 def : Pat<(v2i64 (int_aarch64_neon_vcopy_lane
3873 (v2i64 V128:$Vd), VectorIndexD:$idx, (v2i64 V128:$Vs),
3874 VectorIndexD:$idx2)),
3876 V128:$Vd, VectorIndexD:$idx, V128:$Vs, VectorIndexD:$idx2)
3879 multiclass Neon_INS_elt_pattern<ValueType VT128, ValueType VT64,
3880 ValueType VTScal, Instruction INS> {
3881 def : Pat<(VT128 (vector_insert V128:$src,
3882 (VTScal (vector_extract (VT128 V128:$Rn), imm:$Immn)),
3884 (INS V128:$src, imm:$Immd, V128:$Rn, imm:$Immn)>;
3886 def : Pat<(VT128 (vector_insert V128:$src,
3887 (VTScal (vector_extract (VT64 V64:$Rn), imm:$Immn)),
3889 (INS V128:$src, imm:$Immd,
3890 (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), imm:$Immn)>;
3892 def : Pat<(VT64 (vector_insert V64:$src,
3893 (VTScal (vector_extract (VT128 V128:$Rn), imm:$Immn)),
3895 (EXTRACT_SUBREG (INS (SUBREG_TO_REG (i64 0), V64:$src, dsub),
3896 imm:$Immd, V128:$Rn, imm:$Immn),
3899 def : Pat<(VT64 (vector_insert V64:$src,
3900 (VTScal (vector_extract (VT64 V64:$Rn), imm:$Immn)),
3903 (INS (SUBREG_TO_REG (i64 0), V64:$src, dsub), imm:$Immd,
3904 (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), imm:$Immn),
3908 defm : Neon_INS_elt_pattern<v8f16, v4f16, f16, INSvi16lane>;
3909 defm : Neon_INS_elt_pattern<v4f32, v2f32, f32, INSvi32lane>;
3910 defm : Neon_INS_elt_pattern<v2f64, v1f64, f64, INSvi64lane>;
3913 // Floating point vector extractions are codegen'd as either a sequence of
3914 // subregister extractions, or a MOV (aka CPY here, alias for DUP) if
3915 // the lane number is anything other than zero.
3916 def : Pat<(vector_extract (v2f64 V128:$Rn), 0),
3917 (f64 (EXTRACT_SUBREG V128:$Rn, dsub))>;
3918 def : Pat<(vector_extract (v4f32 V128:$Rn), 0),
3919 (f32 (EXTRACT_SUBREG V128:$Rn, ssub))>;
3920 def : Pat<(vector_extract (v8f16 V128:$Rn), 0),
3921 (f16 (EXTRACT_SUBREG V128:$Rn, hsub))>;
3923 def : Pat<(vector_extract (v2f64 V128:$Rn), VectorIndexD:$idx),
3924 (f64 (CPYi64 V128:$Rn, VectorIndexD:$idx))>;
3925 def : Pat<(vector_extract (v4f32 V128:$Rn), VectorIndexS:$idx),
3926 (f32 (CPYi32 V128:$Rn, VectorIndexS:$idx))>;
3927 def : Pat<(vector_extract (v8f16 V128:$Rn), VectorIndexH:$idx),
3928 (f16 (CPYi16 V128:$Rn, VectorIndexH:$idx))>;
3930 // All concat_vectors operations are canonicalised to act on i64 vectors for
3931 // AArch64. In the general case we need an instruction, which had just as well be
3933 class ConcatPat<ValueType DstTy, ValueType SrcTy>
3934 : Pat<(DstTy (concat_vectors (SrcTy V64:$Rd), V64:$Rn)),
3935 (INSvi64lane (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), 1,
3936 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rn, dsub), 0)>;
3938 def : ConcatPat<v2i64, v1i64>;
3939 def : ConcatPat<v2f64, v1f64>;
3940 def : ConcatPat<v4i32, v2i32>;
3941 def : ConcatPat<v4f32, v2f32>;
3942 def : ConcatPat<v8i16, v4i16>;
3943 def : ConcatPat<v8f16, v4f16>;
3944 def : ConcatPat<v16i8, v8i8>;
3946 // If the high lanes are undef, though, we can just ignore them:
3947 class ConcatUndefPat<ValueType DstTy, ValueType SrcTy>
3948 : Pat<(DstTy (concat_vectors (SrcTy V64:$Rn), undef)),
3949 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rn, dsub)>;
3951 def : ConcatUndefPat<v2i64, v1i64>;
3952 def : ConcatUndefPat<v2f64, v1f64>;
3953 def : ConcatUndefPat<v4i32, v2i32>;
3954 def : ConcatUndefPat<v4f32, v2f32>;
3955 def : ConcatUndefPat<v8i16, v4i16>;
3956 def : ConcatUndefPat<v16i8, v8i8>;
3958 //----------------------------------------------------------------------------
3959 // AdvSIMD across lanes instructions
3960 //----------------------------------------------------------------------------
3962 defm ADDV : SIMDAcrossLanesBHS<0, 0b11011, "addv">;
3963 defm SMAXV : SIMDAcrossLanesBHS<0, 0b01010, "smaxv">;
3964 defm SMINV : SIMDAcrossLanesBHS<0, 0b11010, "sminv">;
3965 defm UMAXV : SIMDAcrossLanesBHS<1, 0b01010, "umaxv">;
3966 defm UMINV : SIMDAcrossLanesBHS<1, 0b11010, "uminv">;
3967 defm SADDLV : SIMDAcrossLanesHSD<0, 0b00011, "saddlv">;
3968 defm UADDLV : SIMDAcrossLanesHSD<1, 0b00011, "uaddlv">;
3969 defm FMAXNMV : SIMDAcrossLanesS<0b01100, 0, "fmaxnmv", int_aarch64_neon_fmaxnmv>;
3970 defm FMAXV : SIMDAcrossLanesS<0b01111, 0, "fmaxv", int_aarch64_neon_fmaxv>;
3971 defm FMINNMV : SIMDAcrossLanesS<0b01100, 1, "fminnmv", int_aarch64_neon_fminnmv>;
3972 defm FMINV : SIMDAcrossLanesS<0b01111, 1, "fminv", int_aarch64_neon_fminv>;
3974 // Patterns for across-vector intrinsics, that have a node equivalent, that
3975 // returns a vector (with only the low lane defined) instead of a scalar.
3976 // In effect, opNode is the same as (scalar_to_vector (IntNode)).
3977 multiclass SIMDAcrossLanesIntrinsic<string baseOpc,
3978 SDPatternOperator opNode> {
3979 // If a lane instruction caught the vector_extract around opNode, we can
3980 // directly match the latter to the instruction.
3981 def : Pat<(v8i8 (opNode V64:$Rn)),
3982 (INSERT_SUBREG (v8i8 (IMPLICIT_DEF)),
3983 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub)>;
3984 def : Pat<(v16i8 (opNode V128:$Rn)),
3985 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3986 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub)>;
3987 def : Pat<(v4i16 (opNode V64:$Rn)),
3988 (INSERT_SUBREG (v4i16 (IMPLICIT_DEF)),
3989 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub)>;
3990 def : Pat<(v8i16 (opNode V128:$Rn)),
3991 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
3992 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub)>;
3993 def : Pat<(v4i32 (opNode V128:$Rn)),
3994 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
3995 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), ssub)>;
3998 // If none did, fallback to the explicit patterns, consuming the vector_extract.
3999 def : Pat<(i32 (vector_extract (insert_subvector undef, (v8i8 (opNode V64:$Rn)),
4000 (i32 0)), (i64 0))),
4001 (EXTRACT_SUBREG (INSERT_SUBREG (v8i8 (IMPLICIT_DEF)),
4002 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn),
4004 def : Pat<(i32 (vector_extract (v16i8 (opNode V128:$Rn)), (i64 0))),
4005 (EXTRACT_SUBREG (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4006 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn),
4008 def : Pat<(i32 (vector_extract (insert_subvector undef,
4009 (v4i16 (opNode V64:$Rn)), (i32 0)), (i64 0))),
4010 (EXTRACT_SUBREG (INSERT_SUBREG (v4i16 (IMPLICIT_DEF)),
4011 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn),
4013 def : Pat<(i32 (vector_extract (v8i16 (opNode V128:$Rn)), (i64 0))),
4014 (EXTRACT_SUBREG (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
4015 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn),
4017 def : Pat<(i32 (vector_extract (v4i32 (opNode V128:$Rn)), (i64 0))),
4018 (EXTRACT_SUBREG (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
4019 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn),
4024 multiclass SIMDAcrossLanesSignedIntrinsic<string baseOpc,
4025 SDPatternOperator opNode>
4026 : SIMDAcrossLanesIntrinsic<baseOpc, opNode> {
4027 // If there is a sign extension after this intrinsic, consume it as smov already
4029 def : Pat<(i32 (sext_inreg (i32 (vector_extract (insert_subvector undef,
4030 (opNode (v8i8 V64:$Rn)), (i32 0)), (i64 0))), i8)),
4032 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4033 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
4035 def : Pat<(i32 (sext_inreg (i32 (vector_extract
4036 (opNode (v16i8 V128:$Rn)), (i64 0))), i8)),
4038 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4039 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
4041 def : Pat<(i32 (sext_inreg (i32 (vector_extract (insert_subvector undef,
4042 (opNode (v4i16 V64:$Rn)), (i32 0)), (i64 0))), i16)),
4044 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4045 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
4047 def : Pat<(i32 (sext_inreg (i32 (vector_extract
4048 (opNode (v8i16 V128:$Rn)), (i64 0))), i16)),
4050 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4051 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
4055 multiclass SIMDAcrossLanesUnsignedIntrinsic<string baseOpc,
4056 SDPatternOperator opNode>
4057 : SIMDAcrossLanesIntrinsic<baseOpc, opNode> {
4058 // If there is a masking operation keeping only what has been actually
4059 // generated, consume it.
4060 def : Pat<(i32 (and (i32 (vector_extract (insert_subvector undef,
4061 (opNode (v8i8 V64:$Rn)), (i32 0)), (i64 0))), maski8_or_more)),
4062 (i32 (EXTRACT_SUBREG
4063 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4064 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
4066 def : Pat<(i32 (and (i32 (vector_extract (opNode (v16i8 V128:$Rn)), (i64 0))),
4068 (i32 (EXTRACT_SUBREG
4069 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4070 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
4072 def : Pat<(i32 (and (i32 (vector_extract (insert_subvector undef,
4073 (opNode (v4i16 V64:$Rn)), (i32 0)), (i64 0))), maski16_or_more)),
4074 (i32 (EXTRACT_SUBREG
4075 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4076 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
4078 def : Pat<(i32 (and (i32 (vector_extract (opNode (v8i16 V128:$Rn)), (i64 0))),
4080 (i32 (EXTRACT_SUBREG
4081 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4082 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
4086 defm : SIMDAcrossLanesSignedIntrinsic<"ADDV", AArch64saddv>;
4087 // vaddv_[su]32 is special; -> ADDP Vd.2S,Vn.2S,Vm.2S; return Vd.s[0];Vn==Vm
4088 def : Pat<(v2i32 (AArch64saddv (v2i32 V64:$Rn))),
4089 (ADDPv2i32 V64:$Rn, V64:$Rn)>;
4091 defm : SIMDAcrossLanesUnsignedIntrinsic<"ADDV", AArch64uaddv>;
4092 // vaddv_[su]32 is special; -> ADDP Vd.2S,Vn.2S,Vm.2S; return Vd.s[0];Vn==Vm
4093 def : Pat<(v2i32 (AArch64uaddv (v2i32 V64:$Rn))),
4094 (ADDPv2i32 V64:$Rn, V64:$Rn)>;
4096 defm : SIMDAcrossLanesSignedIntrinsic<"SMAXV", AArch64smaxv>;
4097 def : Pat<(v2i32 (AArch64smaxv (v2i32 V64:$Rn))),
4098 (SMAXPv2i32 V64:$Rn, V64:$Rn)>;
4100 defm : SIMDAcrossLanesSignedIntrinsic<"SMINV", AArch64sminv>;
4101 def : Pat<(v2i32 (AArch64sminv (v2i32 V64:$Rn))),
4102 (SMINPv2i32 V64:$Rn, V64:$Rn)>;
4104 defm : SIMDAcrossLanesUnsignedIntrinsic<"UMAXV", AArch64umaxv>;
4105 def : Pat<(v2i32 (AArch64umaxv (v2i32 V64:$Rn))),
4106 (UMAXPv2i32 V64:$Rn, V64:$Rn)>;
4108 defm : SIMDAcrossLanesUnsignedIntrinsic<"UMINV", AArch64uminv>;
4109 def : Pat<(v2i32 (AArch64uminv (v2i32 V64:$Rn))),
4110 (UMINPv2i32 V64:$Rn, V64:$Rn)>;
4112 multiclass SIMDAcrossLanesSignedLongIntrinsic<string baseOpc, Intrinsic intOp> {
4113 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
4115 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4116 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), hsub),
4118 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
4120 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4121 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), hsub),
4124 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
4125 (i32 (EXTRACT_SUBREG
4126 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4127 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), ssub),
4129 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
4130 (i32 (EXTRACT_SUBREG
4131 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4132 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), ssub),
4135 def : Pat<(i64 (intOp (v4i32 V128:$Rn))),
4136 (i64 (EXTRACT_SUBREG
4137 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4138 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), dsub),
4142 multiclass SIMDAcrossLanesUnsignedLongIntrinsic<string baseOpc,
4144 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
4145 (i32 (EXTRACT_SUBREG
4146 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4147 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), hsub),
4149 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
4150 (i32 (EXTRACT_SUBREG
4151 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4152 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), hsub),
4155 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
4156 (i32 (EXTRACT_SUBREG
4157 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4158 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), ssub),
4160 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
4161 (i32 (EXTRACT_SUBREG
4162 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4163 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), ssub),
4166 def : Pat<(i64 (intOp (v4i32 V128:$Rn))),
4167 (i64 (EXTRACT_SUBREG
4168 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4169 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), dsub),
4173 defm : SIMDAcrossLanesSignedLongIntrinsic<"SADDLV", int_aarch64_neon_saddlv>;
4174 defm : SIMDAcrossLanesUnsignedLongIntrinsic<"UADDLV", int_aarch64_neon_uaddlv>;
4176 // The vaddlv_s32 intrinsic gets mapped to SADDLP.
4177 def : Pat<(i64 (int_aarch64_neon_saddlv (v2i32 V64:$Rn))),
4178 (i64 (EXTRACT_SUBREG
4179 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4180 (SADDLPv2i32_v1i64 V64:$Rn), dsub),
4182 // The vaddlv_u32 intrinsic gets mapped to UADDLP.
4183 def : Pat<(i64 (int_aarch64_neon_uaddlv (v2i32 V64:$Rn))),
4184 (i64 (EXTRACT_SUBREG
4185 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4186 (UADDLPv2i32_v1i64 V64:$Rn), dsub),
4189 //------------------------------------------------------------------------------
4190 // AdvSIMD modified immediate instructions
4191 //------------------------------------------------------------------------------
4194 defm BIC : SIMDModifiedImmVectorShiftTied<1, 0b11, 0b01, "bic", AArch64bici>;
4196 defm ORR : SIMDModifiedImmVectorShiftTied<0, 0b11, 0b01, "orr", AArch64orri>;
4198 def : InstAlias<"bic $Vd.4h, $imm", (BICv4i16 V64:$Vd, imm0_255:$imm, 0)>;
4199 def : InstAlias<"bic $Vd.8h, $imm", (BICv8i16 V128:$Vd, imm0_255:$imm, 0)>;
4200 def : InstAlias<"bic $Vd.2s, $imm", (BICv2i32 V64:$Vd, imm0_255:$imm, 0)>;
4201 def : InstAlias<"bic $Vd.4s, $imm", (BICv4i32 V128:$Vd, imm0_255:$imm, 0)>;
4203 def : InstAlias<"bic.4h $Vd, $imm", (BICv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
4204 def : InstAlias<"bic.8h $Vd, $imm", (BICv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
4205 def : InstAlias<"bic.2s $Vd, $imm", (BICv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
4206 def : InstAlias<"bic.4s $Vd, $imm", (BICv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
4208 def : InstAlias<"orr $Vd.4h, $imm", (ORRv4i16 V64:$Vd, imm0_255:$imm, 0)>;
4209 def : InstAlias<"orr $Vd.8h, $imm", (ORRv8i16 V128:$Vd, imm0_255:$imm, 0)>;
4210 def : InstAlias<"orr $Vd.2s, $imm", (ORRv2i32 V64:$Vd, imm0_255:$imm, 0)>;
4211 def : InstAlias<"orr $Vd.4s, $imm", (ORRv4i32 V128:$Vd, imm0_255:$imm, 0)>;
4213 def : InstAlias<"orr.4h $Vd, $imm", (ORRv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
4214 def : InstAlias<"orr.8h $Vd, $imm", (ORRv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
4215 def : InstAlias<"orr.2s $Vd, $imm", (ORRv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
4216 def : InstAlias<"orr.4s $Vd, $imm", (ORRv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
4219 def FMOVv2f64_ns : SIMDModifiedImmVectorNoShift<1, 1, 0b1111, V128, fpimm8,
4221 [(set (v2f64 V128:$Rd), (AArch64fmov imm0_255:$imm8))]>;
4222 def FMOVv2f32_ns : SIMDModifiedImmVectorNoShift<0, 0, 0b1111, V64, fpimm8,
4224 [(set (v2f32 V64:$Rd), (AArch64fmov imm0_255:$imm8))]>;
4225 def FMOVv4f32_ns : SIMDModifiedImmVectorNoShift<1, 0, 0b1111, V128, fpimm8,
4227 [(set (v4f32 V128:$Rd), (AArch64fmov imm0_255:$imm8))]>;
4231 // EDIT byte mask: scalar
4232 let isReMaterializable = 1, isAsCheapAsAMove = 1 in
4233 def MOVID : SIMDModifiedImmScalarNoShift<0, 1, 0b1110, "movi",
4234 [(set FPR64:$Rd, simdimmtype10:$imm8)]>;
4235 // The movi_edit node has the immediate value already encoded, so we use
4236 // a plain imm0_255 here.
4237 def : Pat<(f64 (AArch64movi_edit imm0_255:$shift)),
4238 (MOVID imm0_255:$shift)>;
4240 def : Pat<(v1i64 immAllZerosV), (MOVID (i32 0))>;
4241 def : Pat<(v2i32 immAllZerosV), (MOVID (i32 0))>;
4242 def : Pat<(v4i16 immAllZerosV), (MOVID (i32 0))>;
4243 def : Pat<(v8i8 immAllZerosV), (MOVID (i32 0))>;
4245 def : Pat<(v1i64 immAllOnesV), (MOVID (i32 255))>;
4246 def : Pat<(v2i32 immAllOnesV), (MOVID (i32 255))>;
4247 def : Pat<(v4i16 immAllOnesV), (MOVID (i32 255))>;
4248 def : Pat<(v8i8 immAllOnesV), (MOVID (i32 255))>;
4250 // EDIT byte mask: 2d
4252 // The movi_edit node has the immediate value already encoded, so we use
4253 // a plain imm0_255 in the pattern
4254 let isReMaterializable = 1, isAsCheapAsAMove = 1 in
4255 def MOVIv2d_ns : SIMDModifiedImmVectorNoShift<1, 1, 0b1110, V128,
4258 [(set (v2i64 V128:$Rd), (AArch64movi_edit imm0_255:$imm8))]>;
4261 // Use movi.2d to materialize 0.0 if the HW does zero-cycle zeroing.
4262 // Complexity is added to break a tie with a plain MOVI.
4263 let AddedComplexity = 1 in {
4264 def : Pat<(f32 fpimm0),
4265 (f32 (EXTRACT_SUBREG (v2i64 (MOVIv2d_ns (i32 0))), ssub))>,
4267 def : Pat<(f64 fpimm0),
4268 (f64 (EXTRACT_SUBREG (v2i64 (MOVIv2d_ns (i32 0))), dsub))>,
4272 def : Pat<(v2i64 immAllZerosV), (MOVIv2d_ns (i32 0))>;
4273 def : Pat<(v4i32 immAllZerosV), (MOVIv2d_ns (i32 0))>;
4274 def : Pat<(v8i16 immAllZerosV), (MOVIv2d_ns (i32 0))>;
4275 def : Pat<(v16i8 immAllZerosV), (MOVIv2d_ns (i32 0))>;
4277 def : Pat<(v2i64 immAllOnesV), (MOVIv2d_ns (i32 255))>;
4278 def : Pat<(v4i32 immAllOnesV), (MOVIv2d_ns (i32 255))>;
4279 def : Pat<(v8i16 immAllOnesV), (MOVIv2d_ns (i32 255))>;
4280 def : Pat<(v16i8 immAllOnesV), (MOVIv2d_ns (i32 255))>;
4282 def : Pat<(v2f64 (AArch64dup (f64 fpimm0))), (MOVIv2d_ns (i32 0))>;
4283 def : Pat<(v4f32 (AArch64dup (f32 fpimm0))), (MOVIv2d_ns (i32 0))>;
4285 // EDIT per word & halfword: 2s, 4h, 4s, & 8h
4286 defm MOVI : SIMDModifiedImmVectorShift<0, 0b10, 0b00, "movi">;
4288 def : InstAlias<"movi $Vd.4h, $imm", (MOVIv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
4289 def : InstAlias<"movi $Vd.8h, $imm", (MOVIv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
4290 def : InstAlias<"movi $Vd.2s, $imm", (MOVIv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
4291 def : InstAlias<"movi $Vd.4s, $imm", (MOVIv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
4293 def : InstAlias<"movi.4h $Vd, $imm", (MOVIv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
4294 def : InstAlias<"movi.8h $Vd, $imm", (MOVIv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
4295 def : InstAlias<"movi.2s $Vd, $imm", (MOVIv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
4296 def : InstAlias<"movi.4s $Vd, $imm", (MOVIv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
4298 def : Pat<(v2i32 (AArch64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
4299 (MOVIv2i32 imm0_255:$imm8, imm:$shift)>;
4300 def : Pat<(v4i32 (AArch64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
4301 (MOVIv4i32 imm0_255:$imm8, imm:$shift)>;
4302 def : Pat<(v4i16 (AArch64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
4303 (MOVIv4i16 imm0_255:$imm8, imm:$shift)>;
4304 def : Pat<(v8i16 (AArch64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
4305 (MOVIv8i16 imm0_255:$imm8, imm:$shift)>;
4307 // EDIT per word: 2s & 4s with MSL shifter
4308 def MOVIv2s_msl : SIMDModifiedImmMoveMSL<0, 0, {1,1,0,?}, V64, "movi", ".2s",
4309 [(set (v2i32 V64:$Rd),
4310 (AArch64movi_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
4311 def MOVIv4s_msl : SIMDModifiedImmMoveMSL<1, 0, {1,1,0,?}, V128, "movi", ".4s",
4312 [(set (v4i32 V128:$Rd),
4313 (AArch64movi_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
4315 // Per byte: 8b & 16b
4316 def MOVIv8b_ns : SIMDModifiedImmVectorNoShift<0, 0, 0b1110, V64, imm0_255,
4318 [(set (v8i8 V64:$Rd), (AArch64movi imm0_255:$imm8))]>;
4319 def MOVIv16b_ns : SIMDModifiedImmVectorNoShift<1, 0, 0b1110, V128, imm0_255,
4321 [(set (v16i8 V128:$Rd), (AArch64movi imm0_255:$imm8))]>;
4325 // EDIT per word & halfword: 2s, 4h, 4s, & 8h
4326 defm MVNI : SIMDModifiedImmVectorShift<1, 0b10, 0b00, "mvni">;
4328 def : InstAlias<"mvni $Vd.4h, $imm", (MVNIv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
4329 def : InstAlias<"mvni $Vd.8h, $imm", (MVNIv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
4330 def : InstAlias<"mvni $Vd.2s, $imm", (MVNIv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
4331 def : InstAlias<"mvni $Vd.4s, $imm", (MVNIv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
4333 def : InstAlias<"mvni.4h $Vd, $imm", (MVNIv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
4334 def : InstAlias<"mvni.8h $Vd, $imm", (MVNIv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
4335 def : InstAlias<"mvni.2s $Vd, $imm", (MVNIv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
4336 def : InstAlias<"mvni.4s $Vd, $imm", (MVNIv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
4338 def : Pat<(v2i32 (AArch64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
4339 (MVNIv2i32 imm0_255:$imm8, imm:$shift)>;
4340 def : Pat<(v4i32 (AArch64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
4341 (MVNIv4i32 imm0_255:$imm8, imm:$shift)>;
4342 def : Pat<(v4i16 (AArch64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
4343 (MVNIv4i16 imm0_255:$imm8, imm:$shift)>;
4344 def : Pat<(v8i16 (AArch64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
4345 (MVNIv8i16 imm0_255:$imm8, imm:$shift)>;
4347 // EDIT per word: 2s & 4s with MSL shifter
4348 def MVNIv2s_msl : SIMDModifiedImmMoveMSL<0, 1, {1,1,0,?}, V64, "mvni", ".2s",
4349 [(set (v2i32 V64:$Rd),
4350 (AArch64mvni_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
4351 def MVNIv4s_msl : SIMDModifiedImmMoveMSL<1, 1, {1,1,0,?}, V128, "mvni", ".4s",
4352 [(set (v4i32 V128:$Rd),
4353 (AArch64mvni_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
4355 //----------------------------------------------------------------------------
4356 // AdvSIMD indexed element
4357 //----------------------------------------------------------------------------
4359 let hasSideEffects = 0 in {
4360 defm FMLA : SIMDFPIndexedTied<0, 0b0001, "fmla">;
4361 defm FMLS : SIMDFPIndexedTied<0, 0b0101, "fmls">;
4364 // NOTE: Operands are reordered in the FMLA/FMLS PatFrags because the
4365 // instruction expects the addend first, while the intrinsic expects it last.
4367 // On the other hand, there are quite a few valid combinatorial options due to
4368 // the commutativity of multiplication and the fact that (-x) * y = x * (-y).
4369 defm : SIMDFPIndexedTiedPatterns<"FMLA",
4370 TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)>>;
4371 defm : SIMDFPIndexedTiedPatterns<"FMLA",
4372 TriOpFrag<(fma node:$MHS, node:$RHS, node:$LHS)>>;
4374 defm : SIMDFPIndexedTiedPatterns<"FMLS",
4375 TriOpFrag<(fma node:$MHS, (fneg node:$RHS), node:$LHS)> >;
4376 defm : SIMDFPIndexedTiedPatterns<"FMLS",
4377 TriOpFrag<(fma node:$RHS, (fneg node:$MHS), node:$LHS)> >;
4378 defm : SIMDFPIndexedTiedPatterns<"FMLS",
4379 TriOpFrag<(fma (fneg node:$RHS), node:$MHS, node:$LHS)> >;
4380 defm : SIMDFPIndexedTiedPatterns<"FMLS",
4381 TriOpFrag<(fma (fneg node:$MHS), node:$RHS, node:$LHS)> >;
4383 multiclass FMLSIndexedAfterNegPatterns<SDPatternOperator OpNode> {
4384 // 3 variants for the .2s version: DUPLANE from 128-bit, DUPLANE from 64-bit
4386 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
4387 (AArch64duplane32 (v4f32 (fneg V128:$Rm)),
4388 VectorIndexS:$idx))),
4389 (FMLSv2i32_indexed V64:$Rd, V64:$Rn, V128:$Rm, VectorIndexS:$idx)>;
4390 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
4391 (v2f32 (AArch64duplane32
4392 (v4f32 (insert_subvector undef,
4393 (v2f32 (fneg V64:$Rm)),
4395 VectorIndexS:$idx)))),
4396 (FMLSv2i32_indexed V64:$Rd, V64:$Rn,
4397 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub),
4398 VectorIndexS:$idx)>;
4399 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
4400 (AArch64dup (f32 (fneg FPR32Op:$Rm))))),
4401 (FMLSv2i32_indexed V64:$Rd, V64:$Rn,
4402 (SUBREG_TO_REG (i32 0), FPR32Op:$Rm, ssub), (i64 0))>;
4404 // 3 variants for the .4s version: DUPLANE from 128-bit, DUPLANE from 64-bit
4406 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
4407 (AArch64duplane32 (v4f32 (fneg V128:$Rm)),
4408 VectorIndexS:$idx))),
4409 (FMLSv4i32_indexed V128:$Rd, V128:$Rn, V128:$Rm,
4410 VectorIndexS:$idx)>;
4411 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
4412 (v4f32 (AArch64duplane32
4413 (v4f32 (insert_subvector undef,
4414 (v2f32 (fneg V64:$Rm)),
4416 VectorIndexS:$idx)))),
4417 (FMLSv4i32_indexed V128:$Rd, V128:$Rn,
4418 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub),
4419 VectorIndexS:$idx)>;
4420 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
4421 (AArch64dup (f32 (fneg FPR32Op:$Rm))))),
4422 (FMLSv4i32_indexed V128:$Rd, V128:$Rn,
4423 (SUBREG_TO_REG (i32 0), FPR32Op:$Rm, ssub), (i64 0))>;
4425 // 2 variants for the .2d version: DUPLANE from 128-bit, and DUP scalar
4426 // (DUPLANE from 64-bit would be trivial).
4427 def : Pat<(v2f64 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn),
4428 (AArch64duplane64 (v2f64 (fneg V128:$Rm)),
4429 VectorIndexD:$idx))),
4431 V128:$Rd, V128:$Rn, V128:$Rm, VectorIndexS:$idx)>;
4432 def : Pat<(v2f64 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn),
4433 (AArch64dup (f64 (fneg FPR64Op:$Rm))))),
4434 (FMLSv2i64_indexed V128:$Rd, V128:$Rn,
4435 (SUBREG_TO_REG (i32 0), FPR64Op:$Rm, dsub), (i64 0))>;
4437 // 2 variants for 32-bit scalar version: extract from .2s or from .4s
4438 def : Pat<(f32 (OpNode (f32 FPR32:$Rd), (f32 FPR32:$Rn),
4439 (vector_extract (v4f32 (fneg V128:$Rm)),
4440 VectorIndexS:$idx))),
4441 (FMLSv1i32_indexed FPR32:$Rd, FPR32:$Rn,
4442 V128:$Rm, VectorIndexS:$idx)>;
4443 def : Pat<(f32 (OpNode (f32 FPR32:$Rd), (f32 FPR32:$Rn),
4444 (vector_extract (v2f32 (fneg V64:$Rm)),
4445 VectorIndexS:$idx))),
4446 (FMLSv1i32_indexed FPR32:$Rd, FPR32:$Rn,
4447 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub), VectorIndexS:$idx)>;
4449 // 1 variant for 64-bit scalar version: extract from .1d or from .2d
4450 def : Pat<(f64 (OpNode (f64 FPR64:$Rd), (f64 FPR64:$Rn),
4451 (vector_extract (v2f64 (fneg V128:$Rm)),
4452 VectorIndexS:$idx))),
4453 (FMLSv1i64_indexed FPR64:$Rd, FPR64:$Rn,
4454 V128:$Rm, VectorIndexS:$idx)>;
4457 defm : FMLSIndexedAfterNegPatterns<
4458 TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)> >;
4459 defm : FMLSIndexedAfterNegPatterns<
4460 TriOpFrag<(fma node:$MHS, node:$RHS, node:$LHS)> >;
4462 defm FMULX : SIMDFPIndexed<1, 0b1001, "fmulx", int_aarch64_neon_fmulx>;
4463 defm FMUL : SIMDFPIndexed<0, 0b1001, "fmul", fmul>;
4465 def : Pat<(v2f32 (fmul V64:$Rn, (AArch64dup (f32 FPR32:$Rm)))),
4466 (FMULv2i32_indexed V64:$Rn,
4467 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rm, ssub),
4469 def : Pat<(v4f32 (fmul V128:$Rn, (AArch64dup (f32 FPR32:$Rm)))),
4470 (FMULv4i32_indexed V128:$Rn,
4471 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rm, ssub),
4473 def : Pat<(v2f64 (fmul V128:$Rn, (AArch64dup (f64 FPR64:$Rm)))),
4474 (FMULv2i64_indexed V128:$Rn,
4475 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$Rm, dsub),
4478 defm SQDMULH : SIMDIndexedHS<0, 0b1100, "sqdmulh", int_aarch64_neon_sqdmulh>;
4479 defm SQRDMULH : SIMDIndexedHS<0, 0b1101, "sqrdmulh", int_aarch64_neon_sqrdmulh>;
4480 defm MLA : SIMDVectorIndexedHSTied<1, 0b0000, "mla",
4481 TriOpFrag<(add node:$LHS, (mul node:$MHS, node:$RHS))>>;
4482 defm MLS : SIMDVectorIndexedHSTied<1, 0b0100, "mls",
4483 TriOpFrag<(sub node:$LHS, (mul node:$MHS, node:$RHS))>>;
4484 defm MUL : SIMDVectorIndexedHS<0, 0b1000, "mul", mul>;
4485 defm SMLAL : SIMDVectorIndexedLongSDTied<0, 0b0010, "smlal",
4486 TriOpFrag<(add node:$LHS, (int_aarch64_neon_smull node:$MHS, node:$RHS))>>;
4487 defm SMLSL : SIMDVectorIndexedLongSDTied<0, 0b0110, "smlsl",
4488 TriOpFrag<(sub node:$LHS, (int_aarch64_neon_smull node:$MHS, node:$RHS))>>;
4489 defm SMULL : SIMDVectorIndexedLongSD<0, 0b1010, "smull",
4490 int_aarch64_neon_smull>;
4491 defm SQDMLAL : SIMDIndexedLongSQDMLXSDTied<0, 0b0011, "sqdmlal",
4492 int_aarch64_neon_sqadd>;
4493 defm SQDMLSL : SIMDIndexedLongSQDMLXSDTied<0, 0b0111, "sqdmlsl",
4494 int_aarch64_neon_sqsub>;
4495 defm SQRDMLAH : SIMDIndexedSQRDMLxHSDTied<1, 0b1101, "sqrdmlah",
4496 int_aarch64_neon_sqadd>;
4497 defm SQRDMLSH : SIMDIndexedSQRDMLxHSDTied<1, 0b1111, "sqrdmlsh",
4498 int_aarch64_neon_sqsub>;
4499 defm SQDMULL : SIMDIndexedLongSD<0, 0b1011, "sqdmull", int_aarch64_neon_sqdmull>;
4500 defm UMLAL : SIMDVectorIndexedLongSDTied<1, 0b0010, "umlal",
4501 TriOpFrag<(add node:$LHS, (int_aarch64_neon_umull node:$MHS, node:$RHS))>>;
4502 defm UMLSL : SIMDVectorIndexedLongSDTied<1, 0b0110, "umlsl",
4503 TriOpFrag<(sub node:$LHS, (int_aarch64_neon_umull node:$MHS, node:$RHS))>>;
4504 defm UMULL : SIMDVectorIndexedLongSD<1, 0b1010, "umull",
4505 int_aarch64_neon_umull>;
4507 // A scalar sqdmull with the second operand being a vector lane can be
4508 // handled directly with the indexed instruction encoding.
4509 def : Pat<(int_aarch64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
4510 (vector_extract (v4i32 V128:$Vm),
4511 VectorIndexS:$idx)),
4512 (SQDMULLv1i64_indexed FPR32:$Rn, V128:$Vm, VectorIndexS:$idx)>;
4514 //----------------------------------------------------------------------------
4515 // AdvSIMD scalar shift instructions
4516 //----------------------------------------------------------------------------
4517 defm FCVTZS : SIMDScalarRShiftSD<0, 0b11111, "fcvtzs">;
4518 defm FCVTZU : SIMDScalarRShiftSD<1, 0b11111, "fcvtzu">;
4519 defm SCVTF : SIMDScalarRShiftSD<0, 0b11100, "scvtf">;
4520 defm UCVTF : SIMDScalarRShiftSD<1, 0b11100, "ucvtf">;
4521 // Codegen patterns for the above. We don't put these directly on the
4522 // instructions because TableGen's type inference can't handle the truth.
4523 // Having the same base pattern for fp <--> int totally freaks it out.
4524 def : Pat<(int_aarch64_neon_vcvtfp2fxs FPR32:$Rn, vecshiftR32:$imm),
4525 (FCVTZSs FPR32:$Rn, vecshiftR32:$imm)>;
4526 def : Pat<(int_aarch64_neon_vcvtfp2fxu FPR32:$Rn, vecshiftR32:$imm),
4527 (FCVTZUs FPR32:$Rn, vecshiftR32:$imm)>;
4528 def : Pat<(i64 (int_aarch64_neon_vcvtfp2fxs (f64 FPR64:$Rn), vecshiftR64:$imm)),
4529 (FCVTZSd FPR64:$Rn, vecshiftR64:$imm)>;
4530 def : Pat<(i64 (int_aarch64_neon_vcvtfp2fxu (f64 FPR64:$Rn), vecshiftR64:$imm)),
4531 (FCVTZUd FPR64:$Rn, vecshiftR64:$imm)>;
4532 def : Pat<(v1i64 (int_aarch64_neon_vcvtfp2fxs (v1f64 FPR64:$Rn),
4534 (FCVTZSd FPR64:$Rn, vecshiftR64:$imm)>;
4535 def : Pat<(v1i64 (int_aarch64_neon_vcvtfp2fxu (v1f64 FPR64:$Rn),
4537 (FCVTZUd FPR64:$Rn, vecshiftR64:$imm)>;
4538 def : Pat<(int_aarch64_neon_vcvtfxs2fp FPR32:$Rn, vecshiftR32:$imm),
4539 (SCVTFs FPR32:$Rn, vecshiftR32:$imm)>;
4540 def : Pat<(int_aarch64_neon_vcvtfxu2fp FPR32:$Rn, vecshiftR32:$imm),
4541 (UCVTFs FPR32:$Rn, vecshiftR32:$imm)>;
4542 def : Pat<(f64 (int_aarch64_neon_vcvtfxs2fp (i64 FPR64:$Rn), vecshiftR64:$imm)),
4543 (SCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
4544 def : Pat<(f64 (int_aarch64_neon_vcvtfxu2fp (i64 FPR64:$Rn), vecshiftR64:$imm)),
4545 (UCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
4546 def : Pat<(v1f64 (int_aarch64_neon_vcvtfxs2fp (v1i64 FPR64:$Rn),
4548 (SCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
4549 def : Pat<(v1f64 (int_aarch64_neon_vcvtfxu2fp (v1i64 FPR64:$Rn),
4551 (UCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
4553 defm SHL : SIMDScalarLShiftD< 0, 0b01010, "shl", AArch64vshl>;
4554 defm SLI : SIMDScalarLShiftDTied<1, 0b01010, "sli">;
4555 defm SQRSHRN : SIMDScalarRShiftBHS< 0, 0b10011, "sqrshrn",
4556 int_aarch64_neon_sqrshrn>;
4557 defm SQRSHRUN : SIMDScalarRShiftBHS< 1, 0b10001, "sqrshrun",
4558 int_aarch64_neon_sqrshrun>;
4559 defm SQSHLU : SIMDScalarLShiftBHSD<1, 0b01100, "sqshlu", AArch64sqshlui>;
4560 defm SQSHL : SIMDScalarLShiftBHSD<0, 0b01110, "sqshl", AArch64sqshli>;
4561 defm SQSHRN : SIMDScalarRShiftBHS< 0, 0b10010, "sqshrn",
4562 int_aarch64_neon_sqshrn>;
4563 defm SQSHRUN : SIMDScalarRShiftBHS< 1, 0b10000, "sqshrun",
4564 int_aarch64_neon_sqshrun>;
4565 defm SRI : SIMDScalarRShiftDTied< 1, 0b01000, "sri">;
4566 defm SRSHR : SIMDScalarRShiftD< 0, 0b00100, "srshr", AArch64srshri>;
4567 defm SRSRA : SIMDScalarRShiftDTied< 0, 0b00110, "srsra",
4568 TriOpFrag<(add node:$LHS,
4569 (AArch64srshri node:$MHS, node:$RHS))>>;
4570 defm SSHR : SIMDScalarRShiftD< 0, 0b00000, "sshr", AArch64vashr>;
4571 defm SSRA : SIMDScalarRShiftDTied< 0, 0b00010, "ssra",
4572 TriOpFrag<(add node:$LHS,
4573 (AArch64vashr node:$MHS, node:$RHS))>>;
4574 defm UQRSHRN : SIMDScalarRShiftBHS< 1, 0b10011, "uqrshrn",
4575 int_aarch64_neon_uqrshrn>;
4576 defm UQSHL : SIMDScalarLShiftBHSD<1, 0b01110, "uqshl", AArch64uqshli>;
4577 defm UQSHRN : SIMDScalarRShiftBHS< 1, 0b10010, "uqshrn",
4578 int_aarch64_neon_uqshrn>;
4579 defm URSHR : SIMDScalarRShiftD< 1, 0b00100, "urshr", AArch64urshri>;
4580 defm URSRA : SIMDScalarRShiftDTied< 1, 0b00110, "ursra",
4581 TriOpFrag<(add node:$LHS,
4582 (AArch64urshri node:$MHS, node:$RHS))>>;
4583 defm USHR : SIMDScalarRShiftD< 1, 0b00000, "ushr", AArch64vlshr>;
4584 defm USRA : SIMDScalarRShiftDTied< 1, 0b00010, "usra",
4585 TriOpFrag<(add node:$LHS,
4586 (AArch64vlshr node:$MHS, node:$RHS))>>;
4588 //----------------------------------------------------------------------------
4589 // AdvSIMD vector shift instructions
4590 //----------------------------------------------------------------------------
4591 defm FCVTZS:SIMDVectorRShiftSD<0, 0b11111, "fcvtzs", int_aarch64_neon_vcvtfp2fxs>;
4592 defm FCVTZU:SIMDVectorRShiftSD<1, 0b11111, "fcvtzu", int_aarch64_neon_vcvtfp2fxu>;
4593 defm SCVTF: SIMDVectorRShiftSDToFP<0, 0b11100, "scvtf",
4594 int_aarch64_neon_vcvtfxs2fp>;
4595 defm RSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10001, "rshrn",
4596 int_aarch64_neon_rshrn>;
4597 defm SHL : SIMDVectorLShiftBHSD<0, 0b01010, "shl", AArch64vshl>;
4598 defm SHRN : SIMDVectorRShiftNarrowBHS<0, 0b10000, "shrn",
4599 BinOpFrag<(trunc (AArch64vashr node:$LHS, node:$RHS))>>;
4600 defm SLI : SIMDVectorLShiftBHSDTied<1, 0b01010, "sli", int_aarch64_neon_vsli>;
4601 def : Pat<(v1i64 (int_aarch64_neon_vsli (v1i64 FPR64:$Rd), (v1i64 FPR64:$Rn),
4602 (i32 vecshiftL64:$imm))),
4603 (SLId FPR64:$Rd, FPR64:$Rn, vecshiftL64:$imm)>;
4604 defm SQRSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10011, "sqrshrn",
4605 int_aarch64_neon_sqrshrn>;
4606 defm SQRSHRUN: SIMDVectorRShiftNarrowBHS<1, 0b10001, "sqrshrun",
4607 int_aarch64_neon_sqrshrun>;
4608 defm SQSHLU : SIMDVectorLShiftBHSD<1, 0b01100, "sqshlu", AArch64sqshlui>;
4609 defm SQSHL : SIMDVectorLShiftBHSD<0, 0b01110, "sqshl", AArch64sqshli>;
4610 defm SQSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10010, "sqshrn",
4611 int_aarch64_neon_sqshrn>;
4612 defm SQSHRUN : SIMDVectorRShiftNarrowBHS<1, 0b10000, "sqshrun",
4613 int_aarch64_neon_sqshrun>;
4614 defm SRI : SIMDVectorRShiftBHSDTied<1, 0b01000, "sri", int_aarch64_neon_vsri>;
4615 def : Pat<(v1i64 (int_aarch64_neon_vsri (v1i64 FPR64:$Rd), (v1i64 FPR64:$Rn),
4616 (i32 vecshiftR64:$imm))),
4617 (SRId FPR64:$Rd, FPR64:$Rn, vecshiftR64:$imm)>;
4618 defm SRSHR : SIMDVectorRShiftBHSD<0, 0b00100, "srshr", AArch64srshri>;
4619 defm SRSRA : SIMDVectorRShiftBHSDTied<0, 0b00110, "srsra",
4620 TriOpFrag<(add node:$LHS,
4621 (AArch64srshri node:$MHS, node:$RHS))> >;
4622 defm SSHLL : SIMDVectorLShiftLongBHSD<0, 0b10100, "sshll",
4623 BinOpFrag<(AArch64vshl (sext node:$LHS), node:$RHS)>>;
4625 defm SSHR : SIMDVectorRShiftBHSD<0, 0b00000, "sshr", AArch64vashr>;
4626 defm SSRA : SIMDVectorRShiftBHSDTied<0, 0b00010, "ssra",
4627 TriOpFrag<(add node:$LHS, (AArch64vashr node:$MHS, node:$RHS))>>;
4628 defm UCVTF : SIMDVectorRShiftSDToFP<1, 0b11100, "ucvtf",
4629 int_aarch64_neon_vcvtfxu2fp>;
4630 defm UQRSHRN : SIMDVectorRShiftNarrowBHS<1, 0b10011, "uqrshrn",
4631 int_aarch64_neon_uqrshrn>;
4632 defm UQSHL : SIMDVectorLShiftBHSD<1, 0b01110, "uqshl", AArch64uqshli>;
4633 defm UQSHRN : SIMDVectorRShiftNarrowBHS<1, 0b10010, "uqshrn",
4634 int_aarch64_neon_uqshrn>;
4635 defm URSHR : SIMDVectorRShiftBHSD<1, 0b00100, "urshr", AArch64urshri>;
4636 defm URSRA : SIMDVectorRShiftBHSDTied<1, 0b00110, "ursra",
4637 TriOpFrag<(add node:$LHS,
4638 (AArch64urshri node:$MHS, node:$RHS))> >;
4639 defm USHLL : SIMDVectorLShiftLongBHSD<1, 0b10100, "ushll",
4640 BinOpFrag<(AArch64vshl (zext node:$LHS), node:$RHS)>>;
4641 defm USHR : SIMDVectorRShiftBHSD<1, 0b00000, "ushr", AArch64vlshr>;
4642 defm USRA : SIMDVectorRShiftBHSDTied<1, 0b00010, "usra",
4643 TriOpFrag<(add node:$LHS, (AArch64vlshr node:$MHS, node:$RHS))> >;
4645 // SHRN patterns for when a logical right shift was used instead of arithmetic
4646 // (the immediate guarantees no sign bits actually end up in the result so it
4648 def : Pat<(v8i8 (trunc (AArch64vlshr (v8i16 V128:$Rn), vecshiftR16Narrow:$imm))),
4649 (SHRNv8i8_shift V128:$Rn, vecshiftR16Narrow:$imm)>;
4650 def : Pat<(v4i16 (trunc (AArch64vlshr (v4i32 V128:$Rn), vecshiftR32Narrow:$imm))),
4651 (SHRNv4i16_shift V128:$Rn, vecshiftR32Narrow:$imm)>;
4652 def : Pat<(v2i32 (trunc (AArch64vlshr (v2i64 V128:$Rn), vecshiftR64Narrow:$imm))),
4653 (SHRNv2i32_shift V128:$Rn, vecshiftR64Narrow:$imm)>;
4655 def : Pat<(v16i8 (concat_vectors (v8i8 V64:$Rd),
4656 (trunc (AArch64vlshr (v8i16 V128:$Rn),
4657 vecshiftR16Narrow:$imm)))),
4658 (SHRNv16i8_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
4659 V128:$Rn, vecshiftR16Narrow:$imm)>;
4660 def : Pat<(v8i16 (concat_vectors (v4i16 V64:$Rd),
4661 (trunc (AArch64vlshr (v4i32 V128:$Rn),
4662 vecshiftR32Narrow:$imm)))),
4663 (SHRNv8i16_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
4664 V128:$Rn, vecshiftR32Narrow:$imm)>;
4665 def : Pat<(v4i32 (concat_vectors (v2i32 V64:$Rd),
4666 (trunc (AArch64vlshr (v2i64 V128:$Rn),
4667 vecshiftR64Narrow:$imm)))),
4668 (SHRNv4i32_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
4669 V128:$Rn, vecshiftR32Narrow:$imm)>;
4671 // Vector sign and zero extensions are implemented with SSHLL and USSHLL.
4672 // Anyexts are implemented as zexts.
4673 def : Pat<(v8i16 (sext (v8i8 V64:$Rn))), (SSHLLv8i8_shift V64:$Rn, (i32 0))>;
4674 def : Pat<(v8i16 (zext (v8i8 V64:$Rn))), (USHLLv8i8_shift V64:$Rn, (i32 0))>;
4675 def : Pat<(v8i16 (anyext (v8i8 V64:$Rn))), (USHLLv8i8_shift V64:$Rn, (i32 0))>;
4676 def : Pat<(v4i32 (sext (v4i16 V64:$Rn))), (SSHLLv4i16_shift V64:$Rn, (i32 0))>;
4677 def : Pat<(v4i32 (zext (v4i16 V64:$Rn))), (USHLLv4i16_shift V64:$Rn, (i32 0))>;
4678 def : Pat<(v4i32 (anyext (v4i16 V64:$Rn))), (USHLLv4i16_shift V64:$Rn, (i32 0))>;
4679 def : Pat<(v2i64 (sext (v2i32 V64:$Rn))), (SSHLLv2i32_shift V64:$Rn, (i32 0))>;
4680 def : Pat<(v2i64 (zext (v2i32 V64:$Rn))), (USHLLv2i32_shift V64:$Rn, (i32 0))>;
4681 def : Pat<(v2i64 (anyext (v2i32 V64:$Rn))), (USHLLv2i32_shift V64:$Rn, (i32 0))>;
4682 // Also match an extend from the upper half of a 128 bit source register.
4683 def : Pat<(v8i16 (anyext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
4684 (USHLLv16i8_shift V128:$Rn, (i32 0))>;
4685 def : Pat<(v8i16 (zext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
4686 (USHLLv16i8_shift V128:$Rn, (i32 0))>;
4687 def : Pat<(v8i16 (sext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
4688 (SSHLLv16i8_shift V128:$Rn, (i32 0))>;
4689 def : Pat<(v4i32 (anyext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
4690 (USHLLv8i16_shift V128:$Rn, (i32 0))>;
4691 def : Pat<(v4i32 (zext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
4692 (USHLLv8i16_shift V128:$Rn, (i32 0))>;
4693 def : Pat<(v4i32 (sext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
4694 (SSHLLv8i16_shift V128:$Rn, (i32 0))>;
4695 def : Pat<(v2i64 (anyext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
4696 (USHLLv4i32_shift V128:$Rn, (i32 0))>;
4697 def : Pat<(v2i64 (zext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
4698 (USHLLv4i32_shift V128:$Rn, (i32 0))>;
4699 def : Pat<(v2i64 (sext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
4700 (SSHLLv4i32_shift V128:$Rn, (i32 0))>;
4702 // Vector shift sxtl aliases
4703 def : InstAlias<"sxtl.8h $dst, $src1",
4704 (SSHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
4705 def : InstAlias<"sxtl $dst.8h, $src1.8b",
4706 (SSHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
4707 def : InstAlias<"sxtl.4s $dst, $src1",
4708 (SSHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
4709 def : InstAlias<"sxtl $dst.4s, $src1.4h",
4710 (SSHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
4711 def : InstAlias<"sxtl.2d $dst, $src1",
4712 (SSHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
4713 def : InstAlias<"sxtl $dst.2d, $src1.2s",
4714 (SSHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
4716 // Vector shift sxtl2 aliases
4717 def : InstAlias<"sxtl2.8h $dst, $src1",
4718 (SSHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
4719 def : InstAlias<"sxtl2 $dst.8h, $src1.16b",
4720 (SSHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
4721 def : InstAlias<"sxtl2.4s $dst, $src1",
4722 (SSHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
4723 def : InstAlias<"sxtl2 $dst.4s, $src1.8h",
4724 (SSHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
4725 def : InstAlias<"sxtl2.2d $dst, $src1",
4726 (SSHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
4727 def : InstAlias<"sxtl2 $dst.2d, $src1.4s",
4728 (SSHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
4730 // Vector shift uxtl aliases
4731 def : InstAlias<"uxtl.8h $dst, $src1",
4732 (USHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
4733 def : InstAlias<"uxtl $dst.8h, $src1.8b",
4734 (USHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
4735 def : InstAlias<"uxtl.4s $dst, $src1",
4736 (USHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
4737 def : InstAlias<"uxtl $dst.4s, $src1.4h",
4738 (USHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
4739 def : InstAlias<"uxtl.2d $dst, $src1",
4740 (USHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
4741 def : InstAlias<"uxtl $dst.2d, $src1.2s",
4742 (USHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
4744 // Vector shift uxtl2 aliases
4745 def : InstAlias<"uxtl2.8h $dst, $src1",
4746 (USHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
4747 def : InstAlias<"uxtl2 $dst.8h, $src1.16b",
4748 (USHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
4749 def : InstAlias<"uxtl2.4s $dst, $src1",
4750 (USHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
4751 def : InstAlias<"uxtl2 $dst.4s, $src1.8h",
4752 (USHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
4753 def : InstAlias<"uxtl2.2d $dst, $src1",
4754 (USHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
4755 def : InstAlias<"uxtl2 $dst.2d, $src1.4s",
4756 (USHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
4758 // If an integer is about to be converted to a floating point value,
4759 // just load it on the floating point unit.
4760 // These patterns are more complex because floating point loads do not
4761 // support sign extension.
4762 // The sign extension has to be explicitly added and is only supported for
4763 // one step: byte-to-half, half-to-word, word-to-doubleword.
4764 // SCVTF GPR -> FPR is 9 cycles.
4765 // SCVTF FPR -> FPR is 4 cyclces.
4766 // (sign extension with lengthen) SXTL FPR -> FPR is 2 cycles.
4767 // Therefore, we can do 2 sign extensions and one SCVTF FPR -> FPR
4768 // and still being faster.
4769 // However, this is not good for code size.
4770 // 8-bits -> float. 2 sizes step-up.
4771 class SExtLoadi8CVTf32Pat<dag addrmode, dag INST>
4772 : Pat<(f32 (sint_to_fp (i32 (sextloadi8 addrmode)))),
4773 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
4778 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4784 ssub)))>, Requires<[NotForCodeSize, IsCyclone]>;
4786 def : SExtLoadi8CVTf32Pat<(ro8.Wpat GPR64sp:$Rn, GPR32:$Rm, ro8.Wext:$ext),
4787 (LDRBroW GPR64sp:$Rn, GPR32:$Rm, ro8.Wext:$ext)>;
4788 def : SExtLoadi8CVTf32Pat<(ro8.Xpat GPR64sp:$Rn, GPR64:$Rm, ro8.Xext:$ext),
4789 (LDRBroX GPR64sp:$Rn, GPR64:$Rm, ro8.Xext:$ext)>;
4790 def : SExtLoadi8CVTf32Pat<(am_indexed8 GPR64sp:$Rn, uimm12s1:$offset),
4791 (LDRBui GPR64sp:$Rn, uimm12s1:$offset)>;
4792 def : SExtLoadi8CVTf32Pat<(am_unscaled8 GPR64sp:$Rn, simm9:$offset),
4793 (LDURBi GPR64sp:$Rn, simm9:$offset)>;
4795 // 16-bits -> float. 1 size step-up.
4796 class SExtLoadi16CVTf32Pat<dag addrmode, dag INST>
4797 : Pat<(f32 (sint_to_fp (i32 (sextloadi16 addrmode)))),
4798 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
4800 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4804 ssub)))>, Requires<[NotForCodeSize]>;
4806 def : SExtLoadi16CVTf32Pat<(ro16.Wpat GPR64sp:$Rn, GPR32:$Rm, ro16.Wext:$ext),
4807 (LDRHroW GPR64sp:$Rn, GPR32:$Rm, ro16.Wext:$ext)>;
4808 def : SExtLoadi16CVTf32Pat<(ro16.Xpat GPR64sp:$Rn, GPR64:$Rm, ro16.Xext:$ext),
4809 (LDRHroX GPR64sp:$Rn, GPR64:$Rm, ro16.Xext:$ext)>;
4810 def : SExtLoadi16CVTf32Pat<(am_indexed16 GPR64sp:$Rn, uimm12s2:$offset),
4811 (LDRHui GPR64sp:$Rn, uimm12s2:$offset)>;
4812 def : SExtLoadi16CVTf32Pat<(am_unscaled16 GPR64sp:$Rn, simm9:$offset),
4813 (LDURHi GPR64sp:$Rn, simm9:$offset)>;
4815 // 32-bits to 32-bits are handled in target specific dag combine:
4816 // performIntToFpCombine.
4817 // 64-bits integer to 32-bits floating point, not possible with
4818 // SCVTF on floating point registers (both source and destination
4819 // must have the same size).
4821 // Here are the patterns for 8, 16, 32, and 64-bits to double.
4822 // 8-bits -> double. 3 size step-up: give up.
4823 // 16-bits -> double. 2 size step.
4824 class SExtLoadi16CVTf64Pat<dag addrmode, dag INST>
4825 : Pat <(f64 (sint_to_fp (i32 (sextloadi16 addrmode)))),
4826 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4831 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4837 dsub)))>, Requires<[NotForCodeSize, IsCyclone]>;
4839 def : SExtLoadi16CVTf64Pat<(ro16.Wpat GPR64sp:$Rn, GPR32:$Rm, ro16.Wext:$ext),
4840 (LDRHroW GPR64sp:$Rn, GPR32:$Rm, ro16.Wext:$ext)>;
4841 def : SExtLoadi16CVTf64Pat<(ro16.Xpat GPR64sp:$Rn, GPR64:$Rm, ro16.Xext:$ext),
4842 (LDRHroX GPR64sp:$Rn, GPR64:$Rm, ro16.Xext:$ext)>;
4843 def : SExtLoadi16CVTf64Pat<(am_indexed16 GPR64sp:$Rn, uimm12s2:$offset),
4844 (LDRHui GPR64sp:$Rn, uimm12s2:$offset)>;
4845 def : SExtLoadi16CVTf64Pat<(am_unscaled16 GPR64sp:$Rn, simm9:$offset),
4846 (LDURHi GPR64sp:$Rn, simm9:$offset)>;
4847 // 32-bits -> double. 1 size step-up.
4848 class SExtLoadi32CVTf64Pat<dag addrmode, dag INST>
4849 : Pat <(f64 (sint_to_fp (i32 (load addrmode)))),
4850 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4852 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4856 dsub)))>, Requires<[NotForCodeSize]>;
4858 def : SExtLoadi32CVTf64Pat<(ro32.Wpat GPR64sp:$Rn, GPR32:$Rm, ro32.Wext:$ext),
4859 (LDRSroW GPR64sp:$Rn, GPR32:$Rm, ro32.Wext:$ext)>;
4860 def : SExtLoadi32CVTf64Pat<(ro32.Xpat GPR64sp:$Rn, GPR64:$Rm, ro32.Xext:$ext),
4861 (LDRSroX GPR64sp:$Rn, GPR64:$Rm, ro32.Xext:$ext)>;
4862 def : SExtLoadi32CVTf64Pat<(am_indexed32 GPR64sp:$Rn, uimm12s4:$offset),
4863 (LDRSui GPR64sp:$Rn, uimm12s4:$offset)>;
4864 def : SExtLoadi32CVTf64Pat<(am_unscaled32 GPR64sp:$Rn, simm9:$offset),
4865 (LDURSi GPR64sp:$Rn, simm9:$offset)>;
4867 // 64-bits -> double are handled in target specific dag combine:
4868 // performIntToFpCombine.
4871 //----------------------------------------------------------------------------
4872 // AdvSIMD Load-Store Structure
4873 //----------------------------------------------------------------------------
4874 defm LD1 : SIMDLd1Multiple<"ld1">;
4875 defm LD2 : SIMDLd2Multiple<"ld2">;
4876 defm LD3 : SIMDLd3Multiple<"ld3">;
4877 defm LD4 : SIMDLd4Multiple<"ld4">;
4879 defm ST1 : SIMDSt1Multiple<"st1">;
4880 defm ST2 : SIMDSt2Multiple<"st2">;
4881 defm ST3 : SIMDSt3Multiple<"st3">;
4882 defm ST4 : SIMDSt4Multiple<"st4">;
4884 class Ld1Pat<ValueType ty, Instruction INST>
4885 : Pat<(ty (load GPR64sp:$Rn)), (INST GPR64sp:$Rn)>;
4887 def : Ld1Pat<v16i8, LD1Onev16b>;
4888 def : Ld1Pat<v8i16, LD1Onev8h>;
4889 def : Ld1Pat<v4i32, LD1Onev4s>;
4890 def : Ld1Pat<v2i64, LD1Onev2d>;
4891 def : Ld1Pat<v8i8, LD1Onev8b>;
4892 def : Ld1Pat<v4i16, LD1Onev4h>;
4893 def : Ld1Pat<v2i32, LD1Onev2s>;
4894 def : Ld1Pat<v1i64, LD1Onev1d>;
4896 class St1Pat<ValueType ty, Instruction INST>
4897 : Pat<(store ty:$Vt, GPR64sp:$Rn),
4898 (INST ty:$Vt, GPR64sp:$Rn)>;
4900 def : St1Pat<v16i8, ST1Onev16b>;
4901 def : St1Pat<v8i16, ST1Onev8h>;
4902 def : St1Pat<v4i32, ST1Onev4s>;
4903 def : St1Pat<v2i64, ST1Onev2d>;
4904 def : St1Pat<v8i8, ST1Onev8b>;
4905 def : St1Pat<v4i16, ST1Onev4h>;
4906 def : St1Pat<v2i32, ST1Onev2s>;
4907 def : St1Pat<v1i64, ST1Onev1d>;
4913 defm LD1R : SIMDLdR<0, 0b110, 0, "ld1r", "One", 1, 2, 4, 8>;
4914 defm LD2R : SIMDLdR<1, 0b110, 0, "ld2r", "Two", 2, 4, 8, 16>;
4915 defm LD3R : SIMDLdR<0, 0b111, 0, "ld3r", "Three", 3, 6, 12, 24>;
4916 defm LD4R : SIMDLdR<1, 0b111, 0, "ld4r", "Four", 4, 8, 16, 32>;
4917 let mayLoad = 1, hasSideEffects = 0 in {
4918 defm LD1 : SIMDLdSingleBTied<0, 0b000, "ld1", VecListOneb, GPR64pi1>;
4919 defm LD1 : SIMDLdSingleHTied<0, 0b010, 0, "ld1", VecListOneh, GPR64pi2>;
4920 defm LD1 : SIMDLdSingleSTied<0, 0b100, 0b00, "ld1", VecListOnes, GPR64pi4>;
4921 defm LD1 : SIMDLdSingleDTied<0, 0b100, 0b01, "ld1", VecListOned, GPR64pi8>;
4922 defm LD2 : SIMDLdSingleBTied<1, 0b000, "ld2", VecListTwob, GPR64pi2>;
4923 defm LD2 : SIMDLdSingleHTied<1, 0b010, 0, "ld2", VecListTwoh, GPR64pi4>;
4924 defm LD2 : SIMDLdSingleSTied<1, 0b100, 0b00, "ld2", VecListTwos, GPR64pi8>;
4925 defm LD2 : SIMDLdSingleDTied<1, 0b100, 0b01, "ld2", VecListTwod, GPR64pi16>;
4926 defm LD3 : SIMDLdSingleBTied<0, 0b001, "ld3", VecListThreeb, GPR64pi3>;
4927 defm LD3 : SIMDLdSingleHTied<0, 0b011, 0, "ld3", VecListThreeh, GPR64pi6>;
4928 defm LD3 : SIMDLdSingleSTied<0, 0b101, 0b00, "ld3", VecListThrees, GPR64pi12>;
4929 defm LD3 : SIMDLdSingleDTied<0, 0b101, 0b01, "ld3", VecListThreed, GPR64pi24>;
4930 defm LD4 : SIMDLdSingleBTied<1, 0b001, "ld4", VecListFourb, GPR64pi4>;
4931 defm LD4 : SIMDLdSingleHTied<1, 0b011, 0, "ld4", VecListFourh, GPR64pi8>;
4932 defm LD4 : SIMDLdSingleSTied<1, 0b101, 0b00, "ld4", VecListFours, GPR64pi16>;
4933 defm LD4 : SIMDLdSingleDTied<1, 0b101, 0b01, "ld4", VecListFourd, GPR64pi32>;
4936 def : Pat<(v8i8 (AArch64dup (i32 (extloadi8 GPR64sp:$Rn)))),
4937 (LD1Rv8b GPR64sp:$Rn)>;
4938 def : Pat<(v16i8 (AArch64dup (i32 (extloadi8 GPR64sp:$Rn)))),
4939 (LD1Rv16b GPR64sp:$Rn)>;
4940 def : Pat<(v4i16 (AArch64dup (i32 (extloadi16 GPR64sp:$Rn)))),
4941 (LD1Rv4h GPR64sp:$Rn)>;
4942 def : Pat<(v8i16 (AArch64dup (i32 (extloadi16 GPR64sp:$Rn)))),
4943 (LD1Rv8h GPR64sp:$Rn)>;
4944 def : Pat<(v2i32 (AArch64dup (i32 (load GPR64sp:$Rn)))),
4945 (LD1Rv2s GPR64sp:$Rn)>;
4946 def : Pat<(v4i32 (AArch64dup (i32 (load GPR64sp:$Rn)))),
4947 (LD1Rv4s GPR64sp:$Rn)>;
4948 def : Pat<(v2i64 (AArch64dup (i64 (load GPR64sp:$Rn)))),
4949 (LD1Rv2d GPR64sp:$Rn)>;
4950 def : Pat<(v1i64 (AArch64dup (i64 (load GPR64sp:$Rn)))),
4951 (LD1Rv1d GPR64sp:$Rn)>;
4952 // Grab the floating point version too
4953 def : Pat<(v2f32 (AArch64dup (f32 (load GPR64sp:$Rn)))),
4954 (LD1Rv2s GPR64sp:$Rn)>;
4955 def : Pat<(v4f32 (AArch64dup (f32 (load GPR64sp:$Rn)))),
4956 (LD1Rv4s GPR64sp:$Rn)>;
4957 def : Pat<(v2f64 (AArch64dup (f64 (load GPR64sp:$Rn)))),
4958 (LD1Rv2d GPR64sp:$Rn)>;
4959 def : Pat<(v1f64 (AArch64dup (f64 (load GPR64sp:$Rn)))),
4960 (LD1Rv1d GPR64sp:$Rn)>;
4961 def : Pat<(v4f16 (AArch64dup (f16 (load GPR64sp:$Rn)))),
4962 (LD1Rv4h GPR64sp:$Rn)>;
4963 def : Pat<(v8f16 (AArch64dup (f16 (load GPR64sp:$Rn)))),
4964 (LD1Rv8h GPR64sp:$Rn)>;
4966 class Ld1Lane128Pat<SDPatternOperator scalar_load, Operand VecIndex,
4967 ValueType VTy, ValueType STy, Instruction LD1>
4968 : Pat<(vector_insert (VTy VecListOne128:$Rd),
4969 (STy (scalar_load GPR64sp:$Rn)), VecIndex:$idx),
4970 (LD1 VecListOne128:$Rd, VecIndex:$idx, GPR64sp:$Rn)>;
4972 def : Ld1Lane128Pat<extloadi8, VectorIndexB, v16i8, i32, LD1i8>;
4973 def : Ld1Lane128Pat<extloadi16, VectorIndexH, v8i16, i32, LD1i16>;
4974 def : Ld1Lane128Pat<load, VectorIndexS, v4i32, i32, LD1i32>;
4975 def : Ld1Lane128Pat<load, VectorIndexS, v4f32, f32, LD1i32>;
4976 def : Ld1Lane128Pat<load, VectorIndexD, v2i64, i64, LD1i64>;
4977 def : Ld1Lane128Pat<load, VectorIndexD, v2f64, f64, LD1i64>;
4978 def : Ld1Lane128Pat<load, VectorIndexH, v8f16, f16, LD1i16>;
4980 class Ld1Lane64Pat<SDPatternOperator scalar_load, Operand VecIndex,
4981 ValueType VTy, ValueType STy, Instruction LD1>
4982 : Pat<(vector_insert (VTy VecListOne64:$Rd),
4983 (STy (scalar_load GPR64sp:$Rn)), VecIndex:$idx),
4985 (LD1 (SUBREG_TO_REG (i32 0), VecListOne64:$Rd, dsub),
4986 VecIndex:$idx, GPR64sp:$Rn),
4989 def : Ld1Lane64Pat<extloadi8, VectorIndexB, v8i8, i32, LD1i8>;
4990 def : Ld1Lane64Pat<extloadi16, VectorIndexH, v4i16, i32, LD1i16>;
4991 def : Ld1Lane64Pat<load, VectorIndexS, v2i32, i32, LD1i32>;
4992 def : Ld1Lane64Pat<load, VectorIndexS, v2f32, f32, LD1i32>;
4993 def : Ld1Lane64Pat<load, VectorIndexH, v4f16, f16, LD1i16>;
4996 defm LD1 : SIMDLdSt1SingleAliases<"ld1">;
4997 defm LD2 : SIMDLdSt2SingleAliases<"ld2">;
4998 defm LD3 : SIMDLdSt3SingleAliases<"ld3">;
4999 defm LD4 : SIMDLdSt4SingleAliases<"ld4">;
5002 defm ST1 : SIMDStSingleB<0, 0b000, "st1", VecListOneb, GPR64pi1>;
5003 defm ST1 : SIMDStSingleH<0, 0b010, 0, "st1", VecListOneh, GPR64pi2>;
5004 defm ST1 : SIMDStSingleS<0, 0b100, 0b00, "st1", VecListOnes, GPR64pi4>;
5005 defm ST1 : SIMDStSingleD<0, 0b100, 0b01, "st1", VecListOned, GPR64pi8>;
5007 let AddedComplexity = 19 in
5008 class St1Lane128Pat<SDPatternOperator scalar_store, Operand VecIndex,
5009 ValueType VTy, ValueType STy, Instruction ST1>
5011 (STy (vector_extract (VTy VecListOne128:$Vt), VecIndex:$idx)),
5013 (ST1 VecListOne128:$Vt, VecIndex:$idx, GPR64sp:$Rn)>;
5015 def : St1Lane128Pat<truncstorei8, VectorIndexB, v16i8, i32, ST1i8>;
5016 def : St1Lane128Pat<truncstorei16, VectorIndexH, v8i16, i32, ST1i16>;
5017 def : St1Lane128Pat<store, VectorIndexS, v4i32, i32, ST1i32>;
5018 def : St1Lane128Pat<store, VectorIndexS, v4f32, f32, ST1i32>;
5019 def : St1Lane128Pat<store, VectorIndexD, v2i64, i64, ST1i64>;
5020 def : St1Lane128Pat<store, VectorIndexD, v2f64, f64, ST1i64>;
5021 def : St1Lane128Pat<store, VectorIndexH, v8f16, f16, ST1i16>;
5023 let AddedComplexity = 19 in
5024 class St1Lane64Pat<SDPatternOperator scalar_store, Operand VecIndex,
5025 ValueType VTy, ValueType STy, Instruction ST1>
5027 (STy (vector_extract (VTy VecListOne64:$Vt), VecIndex:$idx)),
5029 (ST1 (SUBREG_TO_REG (i32 0), VecListOne64:$Vt, dsub),
5030 VecIndex:$idx, GPR64sp:$Rn)>;
5032 def : St1Lane64Pat<truncstorei8, VectorIndexB, v8i8, i32, ST1i8>;
5033 def : St1Lane64Pat<truncstorei16, VectorIndexH, v4i16, i32, ST1i16>;
5034 def : St1Lane64Pat<store, VectorIndexS, v2i32, i32, ST1i32>;
5035 def : St1Lane64Pat<store, VectorIndexS, v2f32, f32, ST1i32>;
5036 def : St1Lane64Pat<store, VectorIndexH, v4f16, f16, ST1i16>;
5038 multiclass St1LanePost64Pat<SDPatternOperator scalar_store, Operand VecIndex,
5039 ValueType VTy, ValueType STy, Instruction ST1,
5041 def : Pat<(scalar_store
5042 (STy (vector_extract (VTy VecListOne64:$Vt), VecIndex:$idx)),
5043 GPR64sp:$Rn, offset),
5044 (ST1 (SUBREG_TO_REG (i32 0), VecListOne64:$Vt, dsub),
5045 VecIndex:$idx, GPR64sp:$Rn, XZR)>;
5047 def : Pat<(scalar_store
5048 (STy (vector_extract (VTy VecListOne64:$Vt), VecIndex:$idx)),
5049 GPR64sp:$Rn, GPR64:$Rm),
5050 (ST1 (SUBREG_TO_REG (i32 0), VecListOne64:$Vt, dsub),
5051 VecIndex:$idx, GPR64sp:$Rn, $Rm)>;
5054 defm : St1LanePost64Pat<post_truncsti8, VectorIndexB, v8i8, i32, ST1i8_POST, 1>;
5055 defm : St1LanePost64Pat<post_truncsti16, VectorIndexH, v4i16, i32, ST1i16_POST,
5057 defm : St1LanePost64Pat<post_store, VectorIndexS, v2i32, i32, ST1i32_POST, 4>;
5058 defm : St1LanePost64Pat<post_store, VectorIndexS, v2f32, f32, ST1i32_POST, 4>;
5059 defm : St1LanePost64Pat<post_store, VectorIndexD, v1i64, i64, ST1i64_POST, 8>;
5060 defm : St1LanePost64Pat<post_store, VectorIndexD, v1f64, f64, ST1i64_POST, 8>;
5061 defm : St1LanePost64Pat<post_store, VectorIndexH, v4f16, f16, ST1i16_POST, 2>;
5063 multiclass St1LanePost128Pat<SDPatternOperator scalar_store, Operand VecIndex,
5064 ValueType VTy, ValueType STy, Instruction ST1,
5066 def : Pat<(scalar_store
5067 (STy (vector_extract (VTy VecListOne128:$Vt), VecIndex:$idx)),
5068 GPR64sp:$Rn, offset),
5069 (ST1 VecListOne128:$Vt, VecIndex:$idx, GPR64sp:$Rn, XZR)>;
5071 def : Pat<(scalar_store
5072 (STy (vector_extract (VTy VecListOne128:$Vt), VecIndex:$idx)),
5073 GPR64sp:$Rn, GPR64:$Rm),
5074 (ST1 VecListOne128:$Vt, VecIndex:$idx, GPR64sp:$Rn, $Rm)>;
5077 defm : St1LanePost128Pat<post_truncsti8, VectorIndexB, v16i8, i32, ST1i8_POST,
5079 defm : St1LanePost128Pat<post_truncsti16, VectorIndexH, v8i16, i32, ST1i16_POST,
5081 defm : St1LanePost128Pat<post_store, VectorIndexS, v4i32, i32, ST1i32_POST, 4>;
5082 defm : St1LanePost128Pat<post_store, VectorIndexS, v4f32, f32, ST1i32_POST, 4>;
5083 defm : St1LanePost128Pat<post_store, VectorIndexD, v2i64, i64, ST1i64_POST, 8>;
5084 defm : St1LanePost128Pat<post_store, VectorIndexD, v2f64, f64, ST1i64_POST, 8>;
5085 defm : St1LanePost128Pat<post_store, VectorIndexH, v8f16, f16, ST1i16_POST, 2>;
5087 let mayStore = 1, hasSideEffects = 0 in {
5088 defm ST2 : SIMDStSingleB<1, 0b000, "st2", VecListTwob, GPR64pi2>;
5089 defm ST2 : SIMDStSingleH<1, 0b010, 0, "st2", VecListTwoh, GPR64pi4>;
5090 defm ST2 : SIMDStSingleS<1, 0b100, 0b00, "st2", VecListTwos, GPR64pi8>;
5091 defm ST2 : SIMDStSingleD<1, 0b100, 0b01, "st2", VecListTwod, GPR64pi16>;
5092 defm ST3 : SIMDStSingleB<0, 0b001, "st3", VecListThreeb, GPR64pi3>;
5093 defm ST3 : SIMDStSingleH<0, 0b011, 0, "st3", VecListThreeh, GPR64pi6>;
5094 defm ST3 : SIMDStSingleS<0, 0b101, 0b00, "st3", VecListThrees, GPR64pi12>;
5095 defm ST3 : SIMDStSingleD<0, 0b101, 0b01, "st3", VecListThreed, GPR64pi24>;
5096 defm ST4 : SIMDStSingleB<1, 0b001, "st4", VecListFourb, GPR64pi4>;
5097 defm ST4 : SIMDStSingleH<1, 0b011, 0, "st4", VecListFourh, GPR64pi8>;
5098 defm ST4 : SIMDStSingleS<1, 0b101, 0b00, "st4", VecListFours, GPR64pi16>;
5099 defm ST4 : SIMDStSingleD<1, 0b101, 0b01, "st4", VecListFourd, GPR64pi32>;
5102 defm ST1 : SIMDLdSt1SingleAliases<"st1">;
5103 defm ST2 : SIMDLdSt2SingleAliases<"st2">;
5104 defm ST3 : SIMDLdSt3SingleAliases<"st3">;
5105 defm ST4 : SIMDLdSt4SingleAliases<"st4">;
5107 //----------------------------------------------------------------------------
5108 // Crypto extensions
5109 //----------------------------------------------------------------------------
5111 def AESErr : AESTiedInst<0b0100, "aese", int_aarch64_crypto_aese>;
5112 def AESDrr : AESTiedInst<0b0101, "aesd", int_aarch64_crypto_aesd>;
5113 def AESMCrr : AESInst< 0b0110, "aesmc", int_aarch64_crypto_aesmc>;
5114 def AESIMCrr : AESInst< 0b0111, "aesimc", int_aarch64_crypto_aesimc>;
5116 def SHA1Crrr : SHATiedInstQSV<0b000, "sha1c", int_aarch64_crypto_sha1c>;
5117 def SHA1Prrr : SHATiedInstQSV<0b001, "sha1p", int_aarch64_crypto_sha1p>;
5118 def SHA1Mrrr : SHATiedInstQSV<0b010, "sha1m", int_aarch64_crypto_sha1m>;
5119 def SHA1SU0rrr : SHATiedInstVVV<0b011, "sha1su0", int_aarch64_crypto_sha1su0>;
5120 def SHA256Hrrr : SHATiedInstQQV<0b100, "sha256h", int_aarch64_crypto_sha256h>;
5121 def SHA256H2rrr : SHATiedInstQQV<0b101, "sha256h2",int_aarch64_crypto_sha256h2>;
5122 def SHA256SU1rrr :SHATiedInstVVV<0b110, "sha256su1",int_aarch64_crypto_sha256su1>;
5124 def SHA1Hrr : SHAInstSS< 0b0000, "sha1h", int_aarch64_crypto_sha1h>;
5125 def SHA1SU1rr : SHATiedInstVV<0b0001, "sha1su1", int_aarch64_crypto_sha1su1>;
5126 def SHA256SU0rr : SHATiedInstVV<0b0010, "sha256su0",int_aarch64_crypto_sha256su0>;
5128 //----------------------------------------------------------------------------
5130 //----------------------------------------------------------------------------
5131 // FIXME: Like for X86, these should go in their own separate .td file.
5133 // Any instruction that defines a 32-bit result leaves the high half of the
5134 // register. Truncate can be lowered to EXTRACT_SUBREG. CopyFromReg may
5135 // be copying from a truncate. But any other 32-bit operation will zero-extend
5137 // FIXME: X86 also checks for CMOV here. Do we need something similar?
5138 def def32 : PatLeaf<(i32 GPR32:$src), [{
5139 return N->getOpcode() != ISD::TRUNCATE &&
5140 N->getOpcode() != TargetOpcode::EXTRACT_SUBREG &&
5141 N->getOpcode() != ISD::CopyFromReg;
5144 // In the case of a 32-bit def that is known to implicitly zero-extend,
5145 // we can use a SUBREG_TO_REG.
5146 def : Pat<(i64 (zext def32:$src)), (SUBREG_TO_REG (i64 0), GPR32:$src, sub_32)>;
5148 // For an anyext, we don't care what the high bits are, so we can perform an
5149 // INSERT_SUBREF into an IMPLICIT_DEF.
5150 def : Pat<(i64 (anyext GPR32:$src)),
5151 (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32)>;
5153 // When we need to explicitly zero-extend, we use a 32-bit MOV instruction and
5154 // then assert the extension has happened.
5155 def : Pat<(i64 (zext GPR32:$src)),
5156 (SUBREG_TO_REG (i32 0), (ORRWrs WZR, GPR32:$src, 0), sub_32)>;
5158 // To sign extend, we use a signed bitfield move instruction (SBFM) on the
5159 // containing super-reg.
5160 def : Pat<(i64 (sext GPR32:$src)),
5161 (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32), 0, 31)>;
5162 def : Pat<(i64 (sext_inreg GPR64:$src, i32)), (SBFMXri GPR64:$src, 0, 31)>;
5163 def : Pat<(i64 (sext_inreg GPR64:$src, i16)), (SBFMXri GPR64:$src, 0, 15)>;
5164 def : Pat<(i64 (sext_inreg GPR64:$src, i8)), (SBFMXri GPR64:$src, 0, 7)>;
5165 def : Pat<(i64 (sext_inreg GPR64:$src, i1)), (SBFMXri GPR64:$src, 0, 0)>;
5166 def : Pat<(i32 (sext_inreg GPR32:$src, i16)), (SBFMWri GPR32:$src, 0, 15)>;
5167 def : Pat<(i32 (sext_inreg GPR32:$src, i8)), (SBFMWri GPR32:$src, 0, 7)>;
5168 def : Pat<(i32 (sext_inreg GPR32:$src, i1)), (SBFMWri GPR32:$src, 0, 0)>;
5170 def : Pat<(shl (sext_inreg GPR32:$Rn, i8), (i64 imm0_31:$imm)),
5171 (SBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
5172 (i64 (i32shift_sext_i8 imm0_31:$imm)))>;
5173 def : Pat<(shl (sext_inreg GPR64:$Rn, i8), (i64 imm0_63:$imm)),
5174 (SBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
5175 (i64 (i64shift_sext_i8 imm0_63:$imm)))>;
5177 def : Pat<(shl (sext_inreg GPR32:$Rn, i16), (i64 imm0_31:$imm)),
5178 (SBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
5179 (i64 (i32shift_sext_i16 imm0_31:$imm)))>;
5180 def : Pat<(shl (sext_inreg GPR64:$Rn, i16), (i64 imm0_63:$imm)),
5181 (SBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
5182 (i64 (i64shift_sext_i16 imm0_63:$imm)))>;
5184 def : Pat<(shl (i64 (sext GPR32:$Rn)), (i64 imm0_63:$imm)),
5185 (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$Rn, sub_32),
5186 (i64 (i64shift_a imm0_63:$imm)),
5187 (i64 (i64shift_sext_i32 imm0_63:$imm)))>;
5189 // sra patterns have an AddedComplexity of 10, so make sure we have a higher
5190 // AddedComplexity for the following patterns since we want to match sext + sra
5191 // patterns before we attempt to match a single sra node.
5192 let AddedComplexity = 20 in {
5193 // We support all sext + sra combinations which preserve at least one bit of the
5194 // original value which is to be sign extended. E.g. we support shifts up to
5196 def : Pat<(sra (sext_inreg GPR32:$Rn, i8), (i64 imm0_7:$imm)),
5197 (SBFMWri GPR32:$Rn, (i64 imm0_7:$imm), 7)>;
5198 def : Pat<(sra (sext_inreg GPR64:$Rn, i8), (i64 imm0_7:$imm)),
5199 (SBFMXri GPR64:$Rn, (i64 imm0_7:$imm), 7)>;
5201 def : Pat<(sra (sext_inreg GPR32:$Rn, i16), (i64 imm0_15:$imm)),
5202 (SBFMWri GPR32:$Rn, (i64 imm0_15:$imm), 15)>;
5203 def : Pat<(sra (sext_inreg GPR64:$Rn, i16), (i64 imm0_15:$imm)),
5204 (SBFMXri GPR64:$Rn, (i64 imm0_15:$imm), 15)>;
5206 def : Pat<(sra (i64 (sext GPR32:$Rn)), (i64 imm0_31:$imm)),
5207 (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$Rn, sub_32),
5208 (i64 imm0_31:$imm), 31)>;
5209 } // AddedComplexity = 20
5211 // To truncate, we can simply extract from a subregister.
5212 def : Pat<(i32 (trunc GPR64sp:$src)),
5213 (i32 (EXTRACT_SUBREG GPR64sp:$src, sub_32))>;
5215 // __builtin_trap() uses the BRK instruction on AArch64.
5216 def : Pat<(trap), (BRK 1)>;
5218 // Conversions within AdvSIMD types in the same register size are free.
5219 // But because we need a consistent lane ordering, in big endian many
5220 // conversions require one or more REV instructions.
5222 // Consider a simple memory load followed by a bitconvert then a store.
5224 // v1 = BITCAST v2i32 v0 to v4i16
5227 // In big endian mode every memory access has an implicit byte swap. LDR and
5228 // STR do a 64-bit byte swap, whereas LD1/ST1 do a byte swap per lane - that
5229 // is, they treat the vector as a sequence of elements to be byte-swapped.
5230 // The two pairs of instructions are fundamentally incompatible. We've decided
5231 // to use LD1/ST1 only to simplify compiler implementation.
5233 // LD1/ST1 perform the equivalent of a sequence of LDR/STR + REV. This makes
5234 // the original code sequence:
5236 // v1 = REV v2i32 (implicit)
5237 // v2 = BITCAST v2i32 v1 to v4i16
5238 // v3 = REV v4i16 v2 (implicit)
5241 // But this is now broken - the value stored is different to the value loaded
5242 // due to lane reordering. To fix this, on every BITCAST we must perform two
5245 // v1 = REV v2i32 (implicit)
5247 // v3 = BITCAST v2i32 v2 to v4i16
5249 // v5 = REV v4i16 v4 (implicit)
5252 // This means an extra two instructions, but actually in most cases the two REV
5253 // instructions can be combined into one. For example:
5254 // (REV64_2s (REV64_4h X)) === (REV32_4h X)
5256 // There is also no 128-bit REV instruction. This must be synthesized with an
5259 // Most bitconverts require some sort of conversion. The only exceptions are:
5260 // a) Identity conversions - vNfX <-> vNiX
5261 // b) Single-lane-to-scalar - v1fX <-> fX or v1iX <-> iX
5264 // Natural vector casts (64 bit)
5265 def : Pat<(v8i8 (AArch64NvCast (v2i32 FPR64:$src))), (v8i8 FPR64:$src)>;
5266 def : Pat<(v4i16 (AArch64NvCast (v2i32 FPR64:$src))), (v4i16 FPR64:$src)>;
5267 def : Pat<(v4f16 (AArch64NvCast (v2i32 FPR64:$src))), (v4f16 FPR64:$src)>;
5268 def : Pat<(v2i32 (AArch64NvCast (v2i32 FPR64:$src))), (v2i32 FPR64:$src)>;
5269 def : Pat<(v2f32 (AArch64NvCast (v2i32 FPR64:$src))), (v2f32 FPR64:$src)>;
5270 def : Pat<(v1i64 (AArch64NvCast (v2i32 FPR64:$src))), (v1i64 FPR64:$src)>;
5272 def : Pat<(v8i8 (AArch64NvCast (v4i16 FPR64:$src))), (v8i8 FPR64:$src)>;
5273 def : Pat<(v4i16 (AArch64NvCast (v4i16 FPR64:$src))), (v4i16 FPR64:$src)>;
5274 def : Pat<(v4f16 (AArch64NvCast (v4i16 FPR64:$src))), (v4f16 FPR64:$src)>;
5275 def : Pat<(v2i32 (AArch64NvCast (v4i16 FPR64:$src))), (v2i32 FPR64:$src)>;
5276 def : Pat<(v1i64 (AArch64NvCast (v4i16 FPR64:$src))), (v1i64 FPR64:$src)>;
5278 def : Pat<(v8i8 (AArch64NvCast (v8i8 FPR64:$src))), (v8i8 FPR64:$src)>;
5279 def : Pat<(v4i16 (AArch64NvCast (v8i8 FPR64:$src))), (v4i16 FPR64:$src)>;
5280 def : Pat<(v4f16 (AArch64NvCast (v8i8 FPR64:$src))), (v4f16 FPR64:$src)>;
5281 def : Pat<(v2i32 (AArch64NvCast (v8i8 FPR64:$src))), (v2i32 FPR64:$src)>;
5282 def : Pat<(v1i64 (AArch64NvCast (v8i8 FPR64:$src))), (v1i64 FPR64:$src)>;
5284 def : Pat<(v8i8 (AArch64NvCast (f64 FPR64:$src))), (v8i8 FPR64:$src)>;
5285 def : Pat<(v4i16 (AArch64NvCast (f64 FPR64:$src))), (v4i16 FPR64:$src)>;
5286 def : Pat<(v4f16 (AArch64NvCast (f64 FPR64:$src))), (v4f16 FPR64:$src)>;
5287 def : Pat<(v2i32 (AArch64NvCast (f64 FPR64:$src))), (v2i32 FPR64:$src)>;
5288 def : Pat<(v2f32 (AArch64NvCast (f64 FPR64:$src))), (v2f32 FPR64:$src)>;
5289 def : Pat<(v1i64 (AArch64NvCast (f64 FPR64:$src))), (v1i64 FPR64:$src)>;
5290 def : Pat<(v1f64 (AArch64NvCast (f64 FPR64:$src))), (v1f64 FPR64:$src)>;
5292 def : Pat<(v8i8 (AArch64NvCast (v2f32 FPR64:$src))), (v8i8 FPR64:$src)>;
5293 def : Pat<(v4i16 (AArch64NvCast (v2f32 FPR64:$src))), (v4i16 FPR64:$src)>;
5294 def : Pat<(v2i32 (AArch64NvCast (v2f32 FPR64:$src))), (v2i32 FPR64:$src)>;
5295 def : Pat<(v2f32 (AArch64NvCast (v2f32 FPR64:$src))), (v2f32 FPR64:$src)>;
5296 def : Pat<(v1i64 (AArch64NvCast (v2f32 FPR64:$src))), (v1i64 FPR64:$src)>;
5298 // Natural vector casts (128 bit)
5299 def : Pat<(v16i8 (AArch64NvCast (v4i32 FPR128:$src))), (v16i8 FPR128:$src)>;
5300 def : Pat<(v8i16 (AArch64NvCast (v4i32 FPR128:$src))), (v8i16 FPR128:$src)>;
5301 def : Pat<(v8f16 (AArch64NvCast (v4i32 FPR128:$src))), (v8f16 FPR128:$src)>;
5302 def : Pat<(v4i32 (AArch64NvCast (v4i32 FPR128:$src))), (v4i32 FPR128:$src)>;
5303 def : Pat<(v4f32 (AArch64NvCast (v4i32 FPR128:$src))), (v4f32 FPR128:$src)>;
5304 def : Pat<(v2i64 (AArch64NvCast (v4i32 FPR128:$src))), (v2i64 FPR128:$src)>;
5305 def : Pat<(v2f64 (AArch64NvCast (v4i32 FPR128:$src))), (v2f64 FPR128:$src)>;
5307 def : Pat<(v16i8 (AArch64NvCast (v8i16 FPR128:$src))), (v16i8 FPR128:$src)>;
5308 def : Pat<(v8i16 (AArch64NvCast (v8i16 FPR128:$src))), (v8i16 FPR128:$src)>;
5309 def : Pat<(v8f16 (AArch64NvCast (v8i16 FPR128:$src))), (v8f16 FPR128:$src)>;
5310 def : Pat<(v4i32 (AArch64NvCast (v8i16 FPR128:$src))), (v4i32 FPR128:$src)>;
5311 def : Pat<(v2i64 (AArch64NvCast (v8i16 FPR128:$src))), (v2i64 FPR128:$src)>;
5312 def : Pat<(v4f32 (AArch64NvCast (v8i16 FPR128:$src))), (v4f32 FPR128:$src)>;
5313 def : Pat<(v2f64 (AArch64NvCast (v8i16 FPR128:$src))), (v2f64 FPR128:$src)>;
5315 def : Pat<(v16i8 (AArch64NvCast (v16i8 FPR128:$src))), (v16i8 FPR128:$src)>;
5316 def : Pat<(v8i16 (AArch64NvCast (v16i8 FPR128:$src))), (v8i16 FPR128:$src)>;
5317 def : Pat<(v8f16 (AArch64NvCast (v16i8 FPR128:$src))), (v8f16 FPR128:$src)>;
5318 def : Pat<(v4i32 (AArch64NvCast (v16i8 FPR128:$src))), (v4i32 FPR128:$src)>;
5319 def : Pat<(v2i64 (AArch64NvCast (v16i8 FPR128:$src))), (v2i64 FPR128:$src)>;
5320 def : Pat<(v4f32 (AArch64NvCast (v16i8 FPR128:$src))), (v4f32 FPR128:$src)>;
5321 def : Pat<(v2f64 (AArch64NvCast (v16i8 FPR128:$src))), (v2f64 FPR128:$src)>;
5323 def : Pat<(v16i8 (AArch64NvCast (v2i64 FPR128:$src))), (v16i8 FPR128:$src)>;
5324 def : Pat<(v8i16 (AArch64NvCast (v2i64 FPR128:$src))), (v8i16 FPR128:$src)>;
5325 def : Pat<(v8f16 (AArch64NvCast (v2i64 FPR128:$src))), (v8f16 FPR128:$src)>;
5326 def : Pat<(v4i32 (AArch64NvCast (v2i64 FPR128:$src))), (v4i32 FPR128:$src)>;
5327 def : Pat<(v2i64 (AArch64NvCast (v2i64 FPR128:$src))), (v2i64 FPR128:$src)>;
5328 def : Pat<(v4f32 (AArch64NvCast (v2i64 FPR128:$src))), (v4f32 FPR128:$src)>;
5329 def : Pat<(v2f64 (AArch64NvCast (v2i64 FPR128:$src))), (v2f64 FPR128:$src)>;
5331 def : Pat<(v16i8 (AArch64NvCast (v4f32 FPR128:$src))), (v16i8 FPR128:$src)>;
5332 def : Pat<(v8i16 (AArch64NvCast (v4f32 FPR128:$src))), (v8i16 FPR128:$src)>;
5333 def : Pat<(v4i32 (AArch64NvCast (v4f32 FPR128:$src))), (v4i32 FPR128:$src)>;
5334 def : Pat<(v4f32 (AArch64NvCast (v4f32 FPR128:$src))), (v4f32 FPR128:$src)>;
5335 def : Pat<(v2i64 (AArch64NvCast (v4f32 FPR128:$src))), (v2i64 FPR128:$src)>;
5336 def : Pat<(v8f16 (AArch64NvCast (v4f32 FPR128:$src))), (v8f16 FPR128:$src)>;
5337 def : Pat<(v2f64 (AArch64NvCast (v4f32 FPR128:$src))), (v2f64 FPR128:$src)>;
5339 def : Pat<(v16i8 (AArch64NvCast (v2f64 FPR128:$src))), (v16i8 FPR128:$src)>;
5340 def : Pat<(v8i16 (AArch64NvCast (v2f64 FPR128:$src))), (v8i16 FPR128:$src)>;
5341 def : Pat<(v4i32 (AArch64NvCast (v2f64 FPR128:$src))), (v4i32 FPR128:$src)>;
5342 def : Pat<(v2i64 (AArch64NvCast (v2f64 FPR128:$src))), (v2i64 FPR128:$src)>;
5343 def : Pat<(v2f64 (AArch64NvCast (v2f64 FPR128:$src))), (v2f64 FPR128:$src)>;
5344 def : Pat<(v8f16 (AArch64NvCast (v2f64 FPR128:$src))), (v8f16 FPR128:$src)>;
5345 def : Pat<(v4f32 (AArch64NvCast (v2f64 FPR128:$src))), (v4f32 FPR128:$src)>;
5347 let Predicates = [IsLE] in {
5348 def : Pat<(v8i8 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5349 def : Pat<(v4i16 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5350 def : Pat<(v2i32 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5351 def : Pat<(v4f16 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5352 def : Pat<(v2f32 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5354 def : Pat<(i64 (bitconvert (v8i8 V64:$Vn))),
5355 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5356 def : Pat<(i64 (bitconvert (v4i16 V64:$Vn))),
5357 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5358 def : Pat<(i64 (bitconvert (v2i32 V64:$Vn))),
5359 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5360 def : Pat<(i64 (bitconvert (v4f16 V64:$Vn))),
5361 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5362 def : Pat<(i64 (bitconvert (v2f32 V64:$Vn))),
5363 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5364 def : Pat<(i64 (bitconvert (v1f64 V64:$Vn))),
5365 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5367 let Predicates = [IsBE] in {
5368 def : Pat<(v8i8 (bitconvert GPR64:$Xn)),
5369 (REV64v8i8 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
5370 def : Pat<(v4i16 (bitconvert GPR64:$Xn)),
5371 (REV64v4i16 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
5372 def : Pat<(v2i32 (bitconvert GPR64:$Xn)),
5373 (REV64v2i32 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
5374 def : Pat<(v4f16 (bitconvert GPR64:$Xn)),
5375 (REV64v4i16 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
5376 def : Pat<(v2f32 (bitconvert GPR64:$Xn)),
5377 (REV64v2i32 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
5379 def : Pat<(i64 (bitconvert (v8i8 V64:$Vn))),
5380 (REV64v8i8 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
5381 def : Pat<(i64 (bitconvert (v4i16 V64:$Vn))),
5382 (REV64v4i16 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
5383 def : Pat<(i64 (bitconvert (v2i32 V64:$Vn))),
5384 (REV64v2i32 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
5385 def : Pat<(i64 (bitconvert (v4f16 V64:$Vn))),
5386 (REV64v4i16 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
5387 def : Pat<(i64 (bitconvert (v2f32 V64:$Vn))),
5388 (REV64v2i32 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
5390 def : Pat<(v1i64 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5391 def : Pat<(v1f64 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5392 def : Pat<(i64 (bitconvert (v1i64 V64:$Vn))),
5393 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5394 def : Pat<(v1i64 (scalar_to_vector GPR64:$Xn)),
5395 (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5396 def : Pat<(v1f64 (scalar_to_vector GPR64:$Xn)),
5397 (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5398 def : Pat<(v1f64 (scalar_to_vector (f64 FPR64:$Xn))), (v1f64 FPR64:$Xn)>;
5400 def : Pat<(f32 (bitconvert (i32 GPR32:$Xn))),
5401 (COPY_TO_REGCLASS GPR32:$Xn, FPR32)>;
5402 def : Pat<(i32 (bitconvert (f32 FPR32:$Xn))),
5403 (COPY_TO_REGCLASS FPR32:$Xn, GPR32)>;
5404 def : Pat<(f64 (bitconvert (i64 GPR64:$Xn))),
5405 (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5406 def : Pat<(i64 (bitconvert (f64 FPR64:$Xn))),
5407 (COPY_TO_REGCLASS FPR64:$Xn, GPR64)>;
5408 def : Pat<(i64 (bitconvert (v1f64 V64:$Vn))),
5409 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5411 let Predicates = [IsLE] in {
5412 def : Pat<(v1i64 (bitconvert (v2i32 FPR64:$src))), (v1i64 FPR64:$src)>;
5413 def : Pat<(v1i64 (bitconvert (v4i16 FPR64:$src))), (v1i64 FPR64:$src)>;
5414 def : Pat<(v1i64 (bitconvert (v8i8 FPR64:$src))), (v1i64 FPR64:$src)>;
5415 def : Pat<(v1i64 (bitconvert (v4f16 FPR64:$src))), (v1i64 FPR64:$src)>;
5416 def : Pat<(v1i64 (bitconvert (v2f32 FPR64:$src))), (v1i64 FPR64:$src)>;
5418 let Predicates = [IsBE] in {
5419 def : Pat<(v1i64 (bitconvert (v2i32 FPR64:$src))),
5420 (v1i64 (REV64v2i32 FPR64:$src))>;
5421 def : Pat<(v1i64 (bitconvert (v4i16 FPR64:$src))),
5422 (v1i64 (REV64v4i16 FPR64:$src))>;
5423 def : Pat<(v1i64 (bitconvert (v8i8 FPR64:$src))),
5424 (v1i64 (REV64v8i8 FPR64:$src))>;
5425 def : Pat<(v1i64 (bitconvert (v4f16 FPR64:$src))),
5426 (v1i64 (REV64v4i16 FPR64:$src))>;
5427 def : Pat<(v1i64 (bitconvert (v2f32 FPR64:$src))),
5428 (v1i64 (REV64v2i32 FPR64:$src))>;
5430 def : Pat<(v1i64 (bitconvert (v1f64 FPR64:$src))), (v1i64 FPR64:$src)>;
5431 def : Pat<(v1i64 (bitconvert (f64 FPR64:$src))), (v1i64 FPR64:$src)>;
5433 let Predicates = [IsLE] in {
5434 def : Pat<(v2i32 (bitconvert (v1i64 FPR64:$src))), (v2i32 FPR64:$src)>;
5435 def : Pat<(v2i32 (bitconvert (v4i16 FPR64:$src))), (v2i32 FPR64:$src)>;
5436 def : Pat<(v2i32 (bitconvert (v8i8 FPR64:$src))), (v2i32 FPR64:$src)>;
5437 def : Pat<(v2i32 (bitconvert (f64 FPR64:$src))), (v2i32 FPR64:$src)>;
5438 def : Pat<(v2i32 (bitconvert (v1f64 FPR64:$src))), (v2i32 FPR64:$src)>;
5439 def : Pat<(v2i32 (bitconvert (v4f16 FPR64:$src))), (v2i32 FPR64:$src)>;
5441 let Predicates = [IsBE] in {
5442 def : Pat<(v2i32 (bitconvert (v1i64 FPR64:$src))),
5443 (v2i32 (REV64v2i32 FPR64:$src))>;
5444 def : Pat<(v2i32 (bitconvert (v4i16 FPR64:$src))),
5445 (v2i32 (REV32v4i16 FPR64:$src))>;
5446 def : Pat<(v2i32 (bitconvert (v8i8 FPR64:$src))),
5447 (v2i32 (REV32v8i8 FPR64:$src))>;
5448 def : Pat<(v2i32 (bitconvert (f64 FPR64:$src))),
5449 (v2i32 (REV64v2i32 FPR64:$src))>;
5450 def : Pat<(v2i32 (bitconvert (v1f64 FPR64:$src))),
5451 (v2i32 (REV64v2i32 FPR64:$src))>;
5452 def : Pat<(v2i32 (bitconvert (v4f16 FPR64:$src))),
5453 (v2i32 (REV64v4i16 FPR64:$src))>;
5455 def : Pat<(v2i32 (bitconvert (v2f32 FPR64:$src))), (v2i32 FPR64:$src)>;
5457 let Predicates = [IsLE] in {
5458 def : Pat<(v4i16 (bitconvert (v1i64 FPR64:$src))), (v4i16 FPR64:$src)>;
5459 def : Pat<(v4i16 (bitconvert (v2i32 FPR64:$src))), (v4i16 FPR64:$src)>;
5460 def : Pat<(v4i16 (bitconvert (v8i8 FPR64:$src))), (v4i16 FPR64:$src)>;
5461 def : Pat<(v4i16 (bitconvert (f64 FPR64:$src))), (v4i16 FPR64:$src)>;
5462 def : Pat<(v4i16 (bitconvert (v4f16 FPR64:$src))), (v4i16 FPR64:$src)>;
5463 def : Pat<(v4i16 (bitconvert (v2f32 FPR64:$src))), (v4i16 FPR64:$src)>;
5464 def : Pat<(v4i16 (bitconvert (v1f64 FPR64:$src))), (v4i16 FPR64:$src)>;
5466 let Predicates = [IsBE] in {
5467 def : Pat<(v4i16 (bitconvert (v1i64 FPR64:$src))),
5468 (v4i16 (REV64v4i16 FPR64:$src))>;
5469 def : Pat<(v4i16 (bitconvert (v2i32 FPR64:$src))),
5470 (v4i16 (REV32v4i16 FPR64:$src))>;
5471 def : Pat<(v4i16 (bitconvert (v8i8 FPR64:$src))),
5472 (v4i16 (REV16v8i8 FPR64:$src))>;
5473 def : Pat<(v4i16 (bitconvert (f64 FPR64:$src))),
5474 (v4i16 (REV64v4i16 FPR64:$src))>;
5475 def : Pat<(v4i16 (bitconvert (v4f16 FPR64:$src))),
5476 (v4i16 (REV32v4i16 FPR64:$src))>;
5477 def : Pat<(v4i16 (bitconvert (v2f32 FPR64:$src))),
5478 (v4i16 (REV32v4i16 FPR64:$src))>;
5479 def : Pat<(v4i16 (bitconvert (v1f64 FPR64:$src))),
5480 (v4i16 (REV64v4i16 FPR64:$src))>;
5483 let Predicates = [IsLE] in {
5484 def : Pat<(v4f16 (bitconvert (v1i64 FPR64:$src))), (v4f16 FPR64:$src)>;
5485 def : Pat<(v4f16 (bitconvert (v2i32 FPR64:$src))), (v4f16 FPR64:$src)>;
5486 def : Pat<(v4f16 (bitconvert (v4i16 FPR64:$src))), (v4f16 FPR64:$src)>;
5487 def : Pat<(v4f16 (bitconvert (v8i8 FPR64:$src))), (v4f16 FPR64:$src)>;
5488 def : Pat<(v4f16 (bitconvert (f64 FPR64:$src))), (v4f16 FPR64:$src)>;
5489 def : Pat<(v4f16 (bitconvert (v2f32 FPR64:$src))), (v4f16 FPR64:$src)>;
5490 def : Pat<(v4f16 (bitconvert (v1f64 FPR64:$src))), (v4f16 FPR64:$src)>;
5492 let Predicates = [IsBE] in {
5493 def : Pat<(v4f16 (bitconvert (v1i64 FPR64:$src))),
5494 (v4f16 (REV64v4i16 FPR64:$src))>;
5495 def : Pat<(v4f16 (bitconvert (v2i32 FPR64:$src))),
5496 (v4f16 (REV64v4i16 FPR64:$src))>;
5497 def : Pat<(v4f16 (bitconvert (v4i16 FPR64:$src))),
5498 (v4f16 (REV64v4i16 FPR64:$src))>;
5499 def : Pat<(v4f16 (bitconvert (v8i8 FPR64:$src))),
5500 (v4f16 (REV16v8i8 FPR64:$src))>;
5501 def : Pat<(v4f16 (bitconvert (f64 FPR64:$src))),
5502 (v4f16 (REV64v4i16 FPR64:$src))>;
5503 def : Pat<(v4f16 (bitconvert (v2f32 FPR64:$src))),
5504 (v4f16 (REV64v4i16 FPR64:$src))>;
5505 def : Pat<(v4f16 (bitconvert (v1f64 FPR64:$src))),
5506 (v4f16 (REV64v4i16 FPR64:$src))>;
5511 let Predicates = [IsLE] in {
5512 def : Pat<(v8i8 (bitconvert (v1i64 FPR64:$src))), (v8i8 FPR64:$src)>;
5513 def : Pat<(v8i8 (bitconvert (v2i32 FPR64:$src))), (v8i8 FPR64:$src)>;
5514 def : Pat<(v8i8 (bitconvert (v4i16 FPR64:$src))), (v8i8 FPR64:$src)>;
5515 def : Pat<(v8i8 (bitconvert (f64 FPR64:$src))), (v8i8 FPR64:$src)>;
5516 def : Pat<(v8i8 (bitconvert (v2f32 FPR64:$src))), (v8i8 FPR64:$src)>;
5517 def : Pat<(v8i8 (bitconvert (v1f64 FPR64:$src))), (v8i8 FPR64:$src)>;
5518 def : Pat<(v8i8 (bitconvert (v4f16 FPR64:$src))), (v8i8 FPR64:$src)>;
5520 let Predicates = [IsBE] in {
5521 def : Pat<(v8i8 (bitconvert (v1i64 FPR64:$src))),
5522 (v8i8 (REV64v8i8 FPR64:$src))>;
5523 def : Pat<(v8i8 (bitconvert (v2i32 FPR64:$src))),
5524 (v8i8 (REV32v8i8 FPR64:$src))>;
5525 def : Pat<(v8i8 (bitconvert (v4i16 FPR64:$src))),
5526 (v8i8 (REV16v8i8 FPR64:$src))>;
5527 def : Pat<(v8i8 (bitconvert (f64 FPR64:$src))),
5528 (v8i8 (REV64v8i8 FPR64:$src))>;
5529 def : Pat<(v8i8 (bitconvert (v2f32 FPR64:$src))),
5530 (v8i8 (REV32v8i8 FPR64:$src))>;
5531 def : Pat<(v8i8 (bitconvert (v1f64 FPR64:$src))),
5532 (v8i8 (REV64v8i8 FPR64:$src))>;
5533 def : Pat<(v8i8 (bitconvert (v4f16 FPR64:$src))),
5534 (v8i8 (REV16v8i8 FPR64:$src))>;
5537 let Predicates = [IsLE] in {
5538 def : Pat<(f64 (bitconvert (v2i32 FPR64:$src))), (f64 FPR64:$src)>;
5539 def : Pat<(f64 (bitconvert (v4i16 FPR64:$src))), (f64 FPR64:$src)>;
5540 def : Pat<(f64 (bitconvert (v2f32 FPR64:$src))), (f64 FPR64:$src)>;
5541 def : Pat<(f64 (bitconvert (v8i8 FPR64:$src))), (f64 FPR64:$src)>;
5542 def : Pat<(f64 (bitconvert (v4f16 FPR64:$src))), (f64 FPR64:$src)>;
5544 let Predicates = [IsBE] in {
5545 def : Pat<(f64 (bitconvert (v2i32 FPR64:$src))),
5546 (f64 (REV64v2i32 FPR64:$src))>;
5547 def : Pat<(f64 (bitconvert (v4i16 FPR64:$src))),
5548 (f64 (REV64v4i16 FPR64:$src))>;
5549 def : Pat<(f64 (bitconvert (v2f32 FPR64:$src))),
5550 (f64 (REV64v2i32 FPR64:$src))>;
5551 def : Pat<(f64 (bitconvert (v8i8 FPR64:$src))),
5552 (f64 (REV64v8i8 FPR64:$src))>;
5553 def : Pat<(f64 (bitconvert (v4f16 FPR64:$src))),
5554 (f64 (REV64v4i16 FPR64:$src))>;
5556 def : Pat<(f64 (bitconvert (v1i64 FPR64:$src))), (f64 FPR64:$src)>;
5557 def : Pat<(f64 (bitconvert (v1f64 FPR64:$src))), (f64 FPR64:$src)>;
5559 let Predicates = [IsLE] in {
5560 def : Pat<(v1f64 (bitconvert (v2i32 FPR64:$src))), (v1f64 FPR64:$src)>;
5561 def : Pat<(v1f64 (bitconvert (v4i16 FPR64:$src))), (v1f64 FPR64:$src)>;
5562 def : Pat<(v1f64 (bitconvert (v8i8 FPR64:$src))), (v1f64 FPR64:$src)>;
5563 def : Pat<(v1f64 (bitconvert (v2f32 FPR64:$src))), (v1f64 FPR64:$src)>;
5564 def : Pat<(v1f64 (bitconvert (v4f16 FPR64:$src))), (v1f64 FPR64:$src)>;
5566 let Predicates = [IsBE] in {
5567 def : Pat<(v1f64 (bitconvert (v2i32 FPR64:$src))),
5568 (v1f64 (REV64v2i32 FPR64:$src))>;
5569 def : Pat<(v1f64 (bitconvert (v4i16 FPR64:$src))),
5570 (v1f64 (REV64v4i16 FPR64:$src))>;
5571 def : Pat<(v1f64 (bitconvert (v8i8 FPR64:$src))),
5572 (v1f64 (REV64v8i8 FPR64:$src))>;
5573 def : Pat<(v1f64 (bitconvert (v2f32 FPR64:$src))),
5574 (v1f64 (REV64v2i32 FPR64:$src))>;
5575 def : Pat<(v1f64 (bitconvert (v4f16 FPR64:$src))),
5576 (v1f64 (REV64v4i16 FPR64:$src))>;
5578 def : Pat<(v1f64 (bitconvert (v1i64 FPR64:$src))), (v1f64 FPR64:$src)>;
5579 def : Pat<(v1f64 (bitconvert (f64 FPR64:$src))), (v1f64 FPR64:$src)>;
5581 let Predicates = [IsLE] in {
5582 def : Pat<(v2f32 (bitconvert (v1i64 FPR64:$src))), (v2f32 FPR64:$src)>;
5583 def : Pat<(v2f32 (bitconvert (v4i16 FPR64:$src))), (v2f32 FPR64:$src)>;
5584 def : Pat<(v2f32 (bitconvert (v8i8 FPR64:$src))), (v2f32 FPR64:$src)>;
5585 def : Pat<(v2f32 (bitconvert (v1f64 FPR64:$src))), (v2f32 FPR64:$src)>;
5586 def : Pat<(v2f32 (bitconvert (f64 FPR64:$src))), (v2f32 FPR64:$src)>;
5587 def : Pat<(v2f32 (bitconvert (v4f16 FPR64:$src))), (v2f32 FPR64:$src)>;
5589 let Predicates = [IsBE] in {
5590 def : Pat<(v2f32 (bitconvert (v1i64 FPR64:$src))),
5591 (v2f32 (REV64v2i32 FPR64:$src))>;
5592 def : Pat<(v2f32 (bitconvert (v4i16 FPR64:$src))),
5593 (v2f32 (REV32v4i16 FPR64:$src))>;
5594 def : Pat<(v2f32 (bitconvert (v8i8 FPR64:$src))),
5595 (v2f32 (REV32v8i8 FPR64:$src))>;
5596 def : Pat<(v2f32 (bitconvert (v1f64 FPR64:$src))),
5597 (v2f32 (REV64v2i32 FPR64:$src))>;
5598 def : Pat<(v2f32 (bitconvert (f64 FPR64:$src))),
5599 (v2f32 (REV64v2i32 FPR64:$src))>;
5600 def : Pat<(v2f32 (bitconvert (v4f16 FPR64:$src))),
5601 (v2f32 (REV64v4i16 FPR64:$src))>;
5603 def : Pat<(v2f32 (bitconvert (v2i32 FPR64:$src))), (v2f32 FPR64:$src)>;
5605 let Predicates = [IsLE] in {
5606 def : Pat<(f128 (bitconvert (v2i64 FPR128:$src))), (f128 FPR128:$src)>;
5607 def : Pat<(f128 (bitconvert (v4i32 FPR128:$src))), (f128 FPR128:$src)>;
5608 def : Pat<(f128 (bitconvert (v8i16 FPR128:$src))), (f128 FPR128:$src)>;
5609 def : Pat<(f128 (bitconvert (v2f64 FPR128:$src))), (f128 FPR128:$src)>;
5610 def : Pat<(f128 (bitconvert (v4f32 FPR128:$src))), (f128 FPR128:$src)>;
5611 def : Pat<(f128 (bitconvert (v8f16 FPR128:$src))), (f128 FPR128:$src)>;
5612 def : Pat<(f128 (bitconvert (v16i8 FPR128:$src))), (f128 FPR128:$src)>;
5614 let Predicates = [IsBE] in {
5615 def : Pat<(f128 (bitconvert (v2i64 FPR128:$src))),
5616 (f128 (EXTv16i8 FPR128:$src, FPR128:$src, (i32 8)))>;
5617 def : Pat<(f128 (bitconvert (v4i32 FPR128:$src))),
5618 (f128 (EXTv16i8 (REV64v4i32 FPR128:$src),
5619 (REV64v4i32 FPR128:$src), (i32 8)))>;
5620 def : Pat<(f128 (bitconvert (v8i16 FPR128:$src))),
5621 (f128 (EXTv16i8 (REV64v8i16 FPR128:$src),
5622 (REV64v8i16 FPR128:$src), (i32 8)))>;
5623 def : Pat<(f128 (bitconvert (v8f16 FPR128:$src))),
5624 (f128 (EXTv16i8 (REV64v8i16 FPR128:$src),
5625 (REV64v8i16 FPR128:$src), (i32 8)))>;
5626 def : Pat<(f128 (bitconvert (v2f64 FPR128:$src))),
5627 (f128 (EXTv16i8 FPR128:$src, FPR128:$src, (i32 8)))>;
5628 def : Pat<(f128 (bitconvert (v4f32 FPR128:$src))),
5629 (f128 (EXTv16i8 (REV64v4i32 FPR128:$src),
5630 (REV64v4i32 FPR128:$src), (i32 8)))>;
5631 def : Pat<(f128 (bitconvert (v16i8 FPR128:$src))),
5632 (f128 (EXTv16i8 (REV64v16i8 FPR128:$src),
5633 (REV64v16i8 FPR128:$src), (i32 8)))>;
5636 let Predicates = [IsLE] in {
5637 def : Pat<(v2f64 (bitconvert (f128 FPR128:$src))), (v2f64 FPR128:$src)>;
5638 def : Pat<(v2f64 (bitconvert (v4i32 FPR128:$src))), (v2f64 FPR128:$src)>;
5639 def : Pat<(v2f64 (bitconvert (v8i16 FPR128:$src))), (v2f64 FPR128:$src)>;
5640 def : Pat<(v2f64 (bitconvert (v8f16 FPR128:$src))), (v2f64 FPR128:$src)>;
5641 def : Pat<(v2f64 (bitconvert (v16i8 FPR128:$src))), (v2f64 FPR128:$src)>;
5642 def : Pat<(v2f64 (bitconvert (v4f32 FPR128:$src))), (v2f64 FPR128:$src)>;
5644 let Predicates = [IsBE] in {
5645 def : Pat<(v2f64 (bitconvert (f128 FPR128:$src))),
5646 (v2f64 (EXTv16i8 FPR128:$src,
5647 FPR128:$src, (i32 8)))>;
5648 def : Pat<(v2f64 (bitconvert (v4i32 FPR128:$src))),
5649 (v2f64 (REV64v4i32 FPR128:$src))>;
5650 def : Pat<(v2f64 (bitconvert (v8i16 FPR128:$src))),
5651 (v2f64 (REV64v8i16 FPR128:$src))>;
5652 def : Pat<(v2f64 (bitconvert (v8f16 FPR128:$src))),
5653 (v2f64 (REV64v8i16 FPR128:$src))>;
5654 def : Pat<(v2f64 (bitconvert (v16i8 FPR128:$src))),
5655 (v2f64 (REV64v16i8 FPR128:$src))>;
5656 def : Pat<(v2f64 (bitconvert (v4f32 FPR128:$src))),
5657 (v2f64 (REV64v4i32 FPR128:$src))>;
5659 def : Pat<(v2f64 (bitconvert (v2i64 FPR128:$src))), (v2f64 FPR128:$src)>;
5661 let Predicates = [IsLE] in {
5662 def : Pat<(v4f32 (bitconvert (f128 FPR128:$src))), (v4f32 FPR128:$src)>;
5663 def : Pat<(v4f32 (bitconvert (v8i16 FPR128:$src))), (v4f32 FPR128:$src)>;
5664 def : Pat<(v4f32 (bitconvert (v8f16 FPR128:$src))), (v4f32 FPR128:$src)>;
5665 def : Pat<(v4f32 (bitconvert (v16i8 FPR128:$src))), (v4f32 FPR128:$src)>;
5666 def : Pat<(v4f32 (bitconvert (v2i64 FPR128:$src))), (v4f32 FPR128:$src)>;
5667 def : Pat<(v4f32 (bitconvert (v2f64 FPR128:$src))), (v4f32 FPR128:$src)>;
5669 let Predicates = [IsBE] in {
5670 def : Pat<(v4f32 (bitconvert (f128 FPR128:$src))),
5671 (v4f32 (EXTv16i8 (REV64v4i32 FPR128:$src),
5672 (REV64v4i32 FPR128:$src), (i32 8)))>;
5673 def : Pat<(v4f32 (bitconvert (v8i16 FPR128:$src))),
5674 (v4f32 (REV32v8i16 FPR128:$src))>;
5675 def : Pat<(v4f32 (bitconvert (v8f16 FPR128:$src))),
5676 (v4f32 (REV32v8i16 FPR128:$src))>;
5677 def : Pat<(v4f32 (bitconvert (v16i8 FPR128:$src))),
5678 (v4f32 (REV32v16i8 FPR128:$src))>;
5679 def : Pat<(v4f32 (bitconvert (v2i64 FPR128:$src))),
5680 (v4f32 (REV64v4i32 FPR128:$src))>;
5681 def : Pat<(v4f32 (bitconvert (v2f64 FPR128:$src))),
5682 (v4f32 (REV64v4i32 FPR128:$src))>;
5684 def : Pat<(v4f32 (bitconvert (v4i32 FPR128:$src))), (v4f32 FPR128:$src)>;
5686 let Predicates = [IsLE] in {
5687 def : Pat<(v2i64 (bitconvert (f128 FPR128:$src))), (v2i64 FPR128:$src)>;
5688 def : Pat<(v2i64 (bitconvert (v4i32 FPR128:$src))), (v2i64 FPR128:$src)>;
5689 def : Pat<(v2i64 (bitconvert (v8i16 FPR128:$src))), (v2i64 FPR128:$src)>;
5690 def : Pat<(v2i64 (bitconvert (v16i8 FPR128:$src))), (v2i64 FPR128:$src)>;
5691 def : Pat<(v2i64 (bitconvert (v4f32 FPR128:$src))), (v2i64 FPR128:$src)>;
5692 def : Pat<(v2i64 (bitconvert (v8f16 FPR128:$src))), (v2i64 FPR128:$src)>;
5694 let Predicates = [IsBE] in {
5695 def : Pat<(v2i64 (bitconvert (f128 FPR128:$src))),
5696 (v2i64 (EXTv16i8 FPR128:$src,
5697 FPR128:$src, (i32 8)))>;
5698 def : Pat<(v2i64 (bitconvert (v4i32 FPR128:$src))),
5699 (v2i64 (REV64v4i32 FPR128:$src))>;
5700 def : Pat<(v2i64 (bitconvert (v8i16 FPR128:$src))),
5701 (v2i64 (REV64v8i16 FPR128:$src))>;
5702 def : Pat<(v2i64 (bitconvert (v16i8 FPR128:$src))),
5703 (v2i64 (REV64v16i8 FPR128:$src))>;
5704 def : Pat<(v2i64 (bitconvert (v4f32 FPR128:$src))),
5705 (v2i64 (REV64v4i32 FPR128:$src))>;
5706 def : Pat<(v2i64 (bitconvert (v8f16 FPR128:$src))),
5707 (v2i64 (REV64v8i16 FPR128:$src))>;
5709 def : Pat<(v2i64 (bitconvert (v2f64 FPR128:$src))), (v2i64 FPR128:$src)>;
5711 let Predicates = [IsLE] in {
5712 def : Pat<(v4i32 (bitconvert (f128 FPR128:$src))), (v4i32 FPR128:$src)>;
5713 def : Pat<(v4i32 (bitconvert (v2i64 FPR128:$src))), (v4i32 FPR128:$src)>;
5714 def : Pat<(v4i32 (bitconvert (v8i16 FPR128:$src))), (v4i32 FPR128:$src)>;
5715 def : Pat<(v4i32 (bitconvert (v16i8 FPR128:$src))), (v4i32 FPR128:$src)>;
5716 def : Pat<(v4i32 (bitconvert (v2f64 FPR128:$src))), (v4i32 FPR128:$src)>;
5717 def : Pat<(v4i32 (bitconvert (v8f16 FPR128:$src))), (v4i32 FPR128:$src)>;
5719 let Predicates = [IsBE] in {
5720 def : Pat<(v4i32 (bitconvert (f128 FPR128:$src))),
5721 (v4i32 (EXTv16i8 (REV64v4i32 FPR128:$src),
5722 (REV64v4i32 FPR128:$src),
5724 def : Pat<(v4i32 (bitconvert (v2i64 FPR128:$src))),
5725 (v4i32 (REV64v4i32 FPR128:$src))>;
5726 def : Pat<(v4i32 (bitconvert (v8i16 FPR128:$src))),
5727 (v4i32 (REV32v8i16 FPR128:$src))>;
5728 def : Pat<(v4i32 (bitconvert (v16i8 FPR128:$src))),
5729 (v4i32 (REV32v16i8 FPR128:$src))>;
5730 def : Pat<(v4i32 (bitconvert (v2f64 FPR128:$src))),
5731 (v4i32 (REV64v4i32 FPR128:$src))>;
5732 def : Pat<(v4i32 (bitconvert (v8f16 FPR128:$src))),
5733 (v4i32 (REV32v8i16 FPR128:$src))>;
5735 def : Pat<(v4i32 (bitconvert (v4f32 FPR128:$src))), (v4i32 FPR128:$src)>;
5737 let Predicates = [IsLE] in {
5738 def : Pat<(v8i16 (bitconvert (f128 FPR128:$src))), (v8i16 FPR128:$src)>;
5739 def : Pat<(v8i16 (bitconvert (v2i64 FPR128:$src))), (v8i16 FPR128:$src)>;
5740 def : Pat<(v8i16 (bitconvert (v4i32 FPR128:$src))), (v8i16 FPR128:$src)>;
5741 def : Pat<(v8i16 (bitconvert (v16i8 FPR128:$src))), (v8i16 FPR128:$src)>;
5742 def : Pat<(v8i16 (bitconvert (v2f64 FPR128:$src))), (v8i16 FPR128:$src)>;
5743 def : Pat<(v8i16 (bitconvert (v4f32 FPR128:$src))), (v8i16 FPR128:$src)>;
5744 def : Pat<(v8i16 (bitconvert (v8f16 FPR128:$src))), (v8i16 FPR128:$src)>;
5746 let Predicates = [IsBE] in {
5747 def : Pat<(v8i16 (bitconvert (f128 FPR128:$src))),
5748 (v8i16 (EXTv16i8 (REV64v8i16 FPR128:$src),
5749 (REV64v8i16 FPR128:$src),
5751 def : Pat<(v8i16 (bitconvert (v2i64 FPR128:$src))),
5752 (v8i16 (REV64v8i16 FPR128:$src))>;
5753 def : Pat<(v8i16 (bitconvert (v4i32 FPR128:$src))),
5754 (v8i16 (REV32v8i16 FPR128:$src))>;
5755 def : Pat<(v8i16 (bitconvert (v16i8 FPR128:$src))),
5756 (v8i16 (REV16v16i8 FPR128:$src))>;
5757 def : Pat<(v8i16 (bitconvert (v2f64 FPR128:$src))),
5758 (v8i16 (REV64v8i16 FPR128:$src))>;
5759 def : Pat<(v8i16 (bitconvert (v4f32 FPR128:$src))),
5760 (v8i16 (REV32v8i16 FPR128:$src))>;
5761 def : Pat<(v8i16 (bitconvert (v8f16 FPR128:$src))),
5762 (v8i16 (REV32v8i16 FPR128:$src))>;
5765 let Predicates = [IsLE] in {
5766 def : Pat<(v8f16 (bitconvert (f128 FPR128:$src))), (v8f16 FPR128:$src)>;
5767 def : Pat<(v8f16 (bitconvert (v2i64 FPR128:$src))), (v8f16 FPR128:$src)>;
5768 def : Pat<(v8f16 (bitconvert (v4i32 FPR128:$src))), (v8f16 FPR128:$src)>;
5769 def : Pat<(v8f16 (bitconvert (v8i16 FPR128:$src))), (v8f16 FPR128:$src)>;
5770 def : Pat<(v8f16 (bitconvert (v16i8 FPR128:$src))), (v8f16 FPR128:$src)>;
5771 def : Pat<(v8f16 (bitconvert (v2f64 FPR128:$src))), (v8f16 FPR128:$src)>;
5772 def : Pat<(v8f16 (bitconvert (v4f32 FPR128:$src))), (v8f16 FPR128:$src)>;
5774 let Predicates = [IsBE] in {
5775 def : Pat<(v8f16 (bitconvert (f128 FPR128:$src))),
5776 (v8f16 (EXTv16i8 (REV64v8i16 FPR128:$src),
5777 (REV64v8i16 FPR128:$src),
5779 def : Pat<(v8f16 (bitconvert (v2i64 FPR128:$src))),
5780 (v8f16 (REV64v8i16 FPR128:$src))>;
5781 def : Pat<(v8f16 (bitconvert (v4i32 FPR128:$src))),
5782 (v8f16 (REV32v8i16 FPR128:$src))>;
5783 def : Pat<(v8f16 (bitconvert (v8i16 FPR128:$src))),
5784 (v8f16 (REV64v8i16 FPR128:$src))>;
5785 def : Pat<(v8f16 (bitconvert (v16i8 FPR128:$src))),
5786 (v8f16 (REV16v16i8 FPR128:$src))>;
5787 def : Pat<(v8f16 (bitconvert (v2f64 FPR128:$src))),
5788 (v8f16 (REV64v8i16 FPR128:$src))>;
5789 def : Pat<(v8f16 (bitconvert (v4f32 FPR128:$src))),
5790 (v8f16 (REV32v8i16 FPR128:$src))>;
5793 let Predicates = [IsLE] in {
5794 def : Pat<(v16i8 (bitconvert (f128 FPR128:$src))), (v16i8 FPR128:$src)>;
5795 def : Pat<(v16i8 (bitconvert (v2i64 FPR128:$src))), (v16i8 FPR128:$src)>;
5796 def : Pat<(v16i8 (bitconvert (v4i32 FPR128:$src))), (v16i8 FPR128:$src)>;
5797 def : Pat<(v16i8 (bitconvert (v8i16 FPR128:$src))), (v16i8 FPR128:$src)>;
5798 def : Pat<(v16i8 (bitconvert (v2f64 FPR128:$src))), (v16i8 FPR128:$src)>;
5799 def : Pat<(v16i8 (bitconvert (v4f32 FPR128:$src))), (v16i8 FPR128:$src)>;
5800 def : Pat<(v16i8 (bitconvert (v8f16 FPR128:$src))), (v16i8 FPR128:$src)>;
5802 let Predicates = [IsBE] in {
5803 def : Pat<(v16i8 (bitconvert (f128 FPR128:$src))),
5804 (v16i8 (EXTv16i8 (REV64v16i8 FPR128:$src),
5805 (REV64v16i8 FPR128:$src),
5807 def : Pat<(v16i8 (bitconvert (v2i64 FPR128:$src))),
5808 (v16i8 (REV64v16i8 FPR128:$src))>;
5809 def : Pat<(v16i8 (bitconvert (v4i32 FPR128:$src))),
5810 (v16i8 (REV32v16i8 FPR128:$src))>;
5811 def : Pat<(v16i8 (bitconvert (v8i16 FPR128:$src))),
5812 (v16i8 (REV16v16i8 FPR128:$src))>;
5813 def : Pat<(v16i8 (bitconvert (v2f64 FPR128:$src))),
5814 (v16i8 (REV64v16i8 FPR128:$src))>;
5815 def : Pat<(v16i8 (bitconvert (v4f32 FPR128:$src))),
5816 (v16i8 (REV32v16i8 FPR128:$src))>;
5817 def : Pat<(v16i8 (bitconvert (v8f16 FPR128:$src))),
5818 (v16i8 (REV16v16i8 FPR128:$src))>;
5821 def : Pat<(v8i8 (extract_subvector (v16i8 FPR128:$Rn), (i64 1))),
5822 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
5823 def : Pat<(v4i16 (extract_subvector (v8i16 FPR128:$Rn), (i64 1))),
5824 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
5825 def : Pat<(v2i32 (extract_subvector (v4i32 FPR128:$Rn), (i64 1))),
5826 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
5827 def : Pat<(v1i64 (extract_subvector (v2i64 FPR128:$Rn), (i64 1))),
5828 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
5830 // A 64-bit subvector insert to the first 128-bit vector position
5831 // is a subregister copy that needs no instruction.
5832 def : Pat<(insert_subvector undef, (v1i64 FPR64:$src), (i32 0)),
5833 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5834 def : Pat<(insert_subvector undef, (v1f64 FPR64:$src), (i32 0)),
5835 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5836 def : Pat<(insert_subvector undef, (v2i32 FPR64:$src), (i32 0)),
5837 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5838 def : Pat<(insert_subvector undef, (v2f32 FPR64:$src), (i32 0)),
5839 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5840 def : Pat<(insert_subvector undef, (v4i16 FPR64:$src), (i32 0)),
5841 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5842 def : Pat<(insert_subvector undef, (v4f16 FPR64:$src), (i32 0)),
5843 (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5844 def : Pat<(insert_subvector undef, (v8i8 FPR64:$src), (i32 0)),
5845 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5847 // Use pair-wise add instructions when summing up the lanes for v2f64, v2i64
5849 def : Pat<(i64 (add (vector_extract (v2i64 FPR128:$Rn), (i64 0)),
5850 (vector_extract (v2i64 FPR128:$Rn), (i64 1)))),
5851 (i64 (ADDPv2i64p (v2i64 FPR128:$Rn)))>;
5852 def : Pat<(f64 (fadd (vector_extract (v2f64 FPR128:$Rn), (i64 0)),
5853 (vector_extract (v2f64 FPR128:$Rn), (i64 1)))),
5854 (f64 (FADDPv2i64p (v2f64 FPR128:$Rn)))>;
5855 // vector_extract on 64-bit vectors gets promoted to a 128 bit vector,
5856 // so we match on v4f32 here, not v2f32. This will also catch adding
5857 // the low two lanes of a true v4f32 vector.
5858 def : Pat<(fadd (vector_extract (v4f32 FPR128:$Rn), (i64 0)),
5859 (vector_extract (v4f32 FPR128:$Rn), (i64 1))),
5860 (f32 (FADDPv2i32p (EXTRACT_SUBREG FPR128:$Rn, dsub)))>;
5862 // Scalar 64-bit shifts in FPR64 registers.
5863 def : Pat<(i64 (int_aarch64_neon_sshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
5864 (SSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
5865 def : Pat<(i64 (int_aarch64_neon_ushl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
5866 (USHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
5867 def : Pat<(i64 (int_aarch64_neon_srshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
5868 (SRSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
5869 def : Pat<(i64 (int_aarch64_neon_urshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
5870 (URSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
5872 // Tail call return handling. These are all compiler pseudo-instructions,
5873 // so no encoding information or anything like that.
5874 let isCall = 1, isTerminator = 1, isReturn = 1, isBarrier = 1, Uses = [SP] in {
5875 def TCRETURNdi : Pseudo<(outs), (ins i64imm:$dst, i32imm:$FPDiff),[]>;
5876 def TCRETURNri : Pseudo<(outs), (ins tcGPR64:$dst, i32imm:$FPDiff), []>;
5879 def : Pat<(AArch64tcret tcGPR64:$dst, (i32 timm:$FPDiff)),
5880 (TCRETURNri tcGPR64:$dst, imm:$FPDiff)>;
5881 def : Pat<(AArch64tcret tglobaladdr:$dst, (i32 timm:$FPDiff)),
5882 (TCRETURNdi texternalsym:$dst, imm:$FPDiff)>;
5883 def : Pat<(AArch64tcret texternalsym:$dst, (i32 timm:$FPDiff)),
5884 (TCRETURNdi texternalsym:$dst, imm:$FPDiff)>;
5886 include "AArch64InstrAtomics.td"