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 HasFPARMv8 : Predicate<"Subtarget->hasFPARMv8()">,
18 AssemblerPredicate<"FeatureFPARMv8", "fp-armv8">;
19 def HasNEON : Predicate<"Subtarget->hasNEON()">,
20 AssemblerPredicate<"FeatureNEON", "neon">;
21 def HasCrypto : Predicate<"Subtarget->hasCrypto()">,
22 AssemblerPredicate<"FeatureCrypto", "crypto">;
23 def HasCRC : Predicate<"Subtarget->hasCRC()">,
24 AssemblerPredicate<"FeatureCRC", "crc">;
25 def IsLE : Predicate<"Subtarget->isLittleEndian()">;
26 def IsBE : Predicate<"!Subtarget->isLittleEndian()">;
27 def IsCyclone : Predicate<"Subtarget->isCyclone()">;
29 //===----------------------------------------------------------------------===//
30 // AArch64-specific DAG Nodes.
33 // SDTBinaryArithWithFlagsOut - RES1, FLAGS = op LHS, RHS
34 def SDTBinaryArithWithFlagsOut : SDTypeProfile<2, 2,
37 SDTCisInt<0>, SDTCisVT<1, i32>]>;
39 // SDTBinaryArithWithFlagsIn - RES1, FLAGS = op LHS, RHS, FLAGS
40 def SDTBinaryArithWithFlagsIn : SDTypeProfile<1, 3,
46 // SDTBinaryArithWithFlagsInOut - RES1, FLAGS = op LHS, RHS, FLAGS
47 def SDTBinaryArithWithFlagsInOut : SDTypeProfile<2, 3,
54 def SDT_AArch64Brcond : SDTypeProfile<0, 3,
55 [SDTCisVT<0, OtherVT>, SDTCisVT<1, i32>,
57 def SDT_AArch64cbz : SDTypeProfile<0, 2, [SDTCisInt<0>, SDTCisVT<1, OtherVT>]>;
58 def SDT_AArch64tbz : SDTypeProfile<0, 3, [SDTCisInt<0>, SDTCisInt<1>,
59 SDTCisVT<2, OtherVT>]>;
62 def SDT_AArch64CSel : SDTypeProfile<1, 4,
67 def SDT_AArch64FCmp : SDTypeProfile<0, 2,
70 def SDT_AArch64Dup : SDTypeProfile<1, 1, [SDTCisVec<0>]>;
71 def SDT_AArch64DupLane : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisInt<2>]>;
72 def SDT_AArch64Zip : SDTypeProfile<1, 2, [SDTCisVec<0>,
75 def SDT_AArch64MOVIedit : SDTypeProfile<1, 1, [SDTCisInt<1>]>;
76 def SDT_AArch64MOVIshift : SDTypeProfile<1, 2, [SDTCisInt<1>, SDTCisInt<2>]>;
77 def SDT_AArch64vecimm : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0,1>,
78 SDTCisInt<2>, SDTCisInt<3>]>;
79 def SDT_AArch64UnaryVec: SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0,1>]>;
80 def SDT_AArch64ExtVec: SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0,1>,
81 SDTCisSameAs<0,2>, SDTCisInt<3>]>;
82 def SDT_AArch64vshift : SDTypeProfile<1, 2, [SDTCisSameAs<0,1>, SDTCisInt<2>]>;
84 def SDT_AArch64unvec : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0,1>]>;
85 def SDT_AArch64fcmpz : SDTypeProfile<1, 1, []>;
86 def SDT_AArch64fcmp : SDTypeProfile<1, 2, [SDTCisSameAs<1,2>]>;
87 def SDT_AArch64binvec : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisSameAs<0,1>,
89 def SDT_AArch64trivec : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0,1>,
92 def SDT_AArch64TCRET : SDTypeProfile<0, 2, [SDTCisPtrTy<0>]>;
93 def SDT_AArch64PREFETCH : SDTypeProfile<0, 2, [SDTCisVT<0, i32>, SDTCisPtrTy<1>]>;
95 def SDT_AArch64ITOF : SDTypeProfile<1, 1, [SDTCisFP<0>, SDTCisSameAs<0,1>]>;
97 def SDT_AArch64TLSDescCall : SDTypeProfile<0, -2, [SDTCisPtrTy<0>,
100 // Generates the general dynamic sequences, i.e.
101 // adrp x0, :tlsdesc:var
102 // ldr x1, [x0, #:tlsdesc_lo12:var]
103 // add x0, x0, #:tlsdesc_lo12:var
107 // (the TPIDR_EL0 offset is put directly in X0, hence no "result" here)
108 // number of operands (the variable)
109 def SDT_AArch64TLSDescCallSeq : SDTypeProfile<0,1,
112 def SDT_AArch64WrapperLarge : SDTypeProfile<1, 4,
113 [SDTCisVT<0, i64>, SDTCisVT<1, i32>,
114 SDTCisSameAs<1, 2>, SDTCisSameAs<1, 3>,
115 SDTCisSameAs<1, 4>]>;
119 def AArch64adrp : SDNode<"AArch64ISD::ADRP", SDTIntUnaryOp, []>;
120 def AArch64addlow : SDNode<"AArch64ISD::ADDlow", SDTIntBinOp, []>;
121 def AArch64LOADgot : SDNode<"AArch64ISD::LOADgot", SDTIntUnaryOp>;
122 def AArch64callseq_start : SDNode<"ISD::CALLSEQ_START",
123 SDCallSeqStart<[ SDTCisVT<0, i32> ]>,
124 [SDNPHasChain, SDNPOutGlue]>;
125 def AArch64callseq_end : SDNode<"ISD::CALLSEQ_END",
126 SDCallSeqEnd<[ SDTCisVT<0, i32>,
128 [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue]>;
129 def AArch64call : SDNode<"AArch64ISD::CALL",
130 SDTypeProfile<0, -1, [SDTCisPtrTy<0>]>,
131 [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue,
133 def AArch64brcond : SDNode<"AArch64ISD::BRCOND", SDT_AArch64Brcond,
135 def AArch64cbz : SDNode<"AArch64ISD::CBZ", SDT_AArch64cbz,
137 def AArch64cbnz : SDNode<"AArch64ISD::CBNZ", SDT_AArch64cbz,
139 def AArch64tbz : SDNode<"AArch64ISD::TBZ", SDT_AArch64tbz,
141 def AArch64tbnz : SDNode<"AArch64ISD::TBNZ", SDT_AArch64tbz,
145 def AArch64csel : SDNode<"AArch64ISD::CSEL", SDT_AArch64CSel>;
146 def AArch64csinv : SDNode<"AArch64ISD::CSINV", SDT_AArch64CSel>;
147 def AArch64csneg : SDNode<"AArch64ISD::CSNEG", SDT_AArch64CSel>;
148 def AArch64csinc : SDNode<"AArch64ISD::CSINC", SDT_AArch64CSel>;
149 def AArch64retflag : SDNode<"AArch64ISD::RET_FLAG", SDTNone,
150 [SDNPHasChain, SDNPOptInGlue, SDNPVariadic]>;
151 def AArch64adc : SDNode<"AArch64ISD::ADC", SDTBinaryArithWithFlagsIn >;
152 def AArch64sbc : SDNode<"AArch64ISD::SBC", SDTBinaryArithWithFlagsIn>;
153 def AArch64add_flag : SDNode<"AArch64ISD::ADDS", SDTBinaryArithWithFlagsOut,
155 def AArch64sub_flag : SDNode<"AArch64ISD::SUBS", SDTBinaryArithWithFlagsOut>;
156 def AArch64and_flag : SDNode<"AArch64ISD::ANDS", SDTBinaryArithWithFlagsOut,
158 def AArch64adc_flag : SDNode<"AArch64ISD::ADCS", SDTBinaryArithWithFlagsInOut>;
159 def AArch64sbc_flag : SDNode<"AArch64ISD::SBCS", SDTBinaryArithWithFlagsInOut>;
161 def AArch64threadpointer : SDNode<"AArch64ISD::THREAD_POINTER", SDTPtrLeaf>;
163 def AArch64fcmp : SDNode<"AArch64ISD::FCMP", SDT_AArch64FCmp>;
165 def AArch64fmax : SDNode<"AArch64ISD::FMAX", SDTFPBinOp>;
166 def AArch64fmin : SDNode<"AArch64ISD::FMIN", SDTFPBinOp>;
168 def AArch64dup : SDNode<"AArch64ISD::DUP", SDT_AArch64Dup>;
169 def AArch64duplane8 : SDNode<"AArch64ISD::DUPLANE8", SDT_AArch64DupLane>;
170 def AArch64duplane16 : SDNode<"AArch64ISD::DUPLANE16", SDT_AArch64DupLane>;
171 def AArch64duplane32 : SDNode<"AArch64ISD::DUPLANE32", SDT_AArch64DupLane>;
172 def AArch64duplane64 : SDNode<"AArch64ISD::DUPLANE64", SDT_AArch64DupLane>;
174 def AArch64zip1 : SDNode<"AArch64ISD::ZIP1", SDT_AArch64Zip>;
175 def AArch64zip2 : SDNode<"AArch64ISD::ZIP2", SDT_AArch64Zip>;
176 def AArch64uzp1 : SDNode<"AArch64ISD::UZP1", SDT_AArch64Zip>;
177 def AArch64uzp2 : SDNode<"AArch64ISD::UZP2", SDT_AArch64Zip>;
178 def AArch64trn1 : SDNode<"AArch64ISD::TRN1", SDT_AArch64Zip>;
179 def AArch64trn2 : SDNode<"AArch64ISD::TRN2", SDT_AArch64Zip>;
181 def AArch64movi_edit : SDNode<"AArch64ISD::MOVIedit", SDT_AArch64MOVIedit>;
182 def AArch64movi_shift : SDNode<"AArch64ISD::MOVIshift", SDT_AArch64MOVIshift>;
183 def AArch64movi_msl : SDNode<"AArch64ISD::MOVImsl", SDT_AArch64MOVIshift>;
184 def AArch64mvni_shift : SDNode<"AArch64ISD::MVNIshift", SDT_AArch64MOVIshift>;
185 def AArch64mvni_msl : SDNode<"AArch64ISD::MVNImsl", SDT_AArch64MOVIshift>;
186 def AArch64movi : SDNode<"AArch64ISD::MOVI", SDT_AArch64MOVIedit>;
187 def AArch64fmov : SDNode<"AArch64ISD::FMOV", SDT_AArch64MOVIedit>;
189 def AArch64rev16 : SDNode<"AArch64ISD::REV16", SDT_AArch64UnaryVec>;
190 def AArch64rev32 : SDNode<"AArch64ISD::REV32", SDT_AArch64UnaryVec>;
191 def AArch64rev64 : SDNode<"AArch64ISD::REV64", SDT_AArch64UnaryVec>;
192 def AArch64ext : SDNode<"AArch64ISD::EXT", SDT_AArch64ExtVec>;
194 def AArch64vashr : SDNode<"AArch64ISD::VASHR", SDT_AArch64vshift>;
195 def AArch64vlshr : SDNode<"AArch64ISD::VLSHR", SDT_AArch64vshift>;
196 def AArch64vshl : SDNode<"AArch64ISD::VSHL", SDT_AArch64vshift>;
197 def AArch64sqshli : SDNode<"AArch64ISD::SQSHL_I", SDT_AArch64vshift>;
198 def AArch64uqshli : SDNode<"AArch64ISD::UQSHL_I", SDT_AArch64vshift>;
199 def AArch64sqshlui : SDNode<"AArch64ISD::SQSHLU_I", SDT_AArch64vshift>;
200 def AArch64srshri : SDNode<"AArch64ISD::SRSHR_I", SDT_AArch64vshift>;
201 def AArch64urshri : SDNode<"AArch64ISD::URSHR_I", SDT_AArch64vshift>;
203 def AArch64not: SDNode<"AArch64ISD::NOT", SDT_AArch64unvec>;
204 def AArch64bit: SDNode<"AArch64ISD::BIT", SDT_AArch64trivec>;
205 def AArch64bsl: SDNode<"AArch64ISD::BSL", SDT_AArch64trivec>;
207 def AArch64cmeq: SDNode<"AArch64ISD::CMEQ", SDT_AArch64binvec>;
208 def AArch64cmge: SDNode<"AArch64ISD::CMGE", SDT_AArch64binvec>;
209 def AArch64cmgt: SDNode<"AArch64ISD::CMGT", SDT_AArch64binvec>;
210 def AArch64cmhi: SDNode<"AArch64ISD::CMHI", SDT_AArch64binvec>;
211 def AArch64cmhs: SDNode<"AArch64ISD::CMHS", SDT_AArch64binvec>;
213 def AArch64fcmeq: SDNode<"AArch64ISD::FCMEQ", SDT_AArch64fcmp>;
214 def AArch64fcmge: SDNode<"AArch64ISD::FCMGE", SDT_AArch64fcmp>;
215 def AArch64fcmgt: SDNode<"AArch64ISD::FCMGT", SDT_AArch64fcmp>;
217 def AArch64cmeqz: SDNode<"AArch64ISD::CMEQz", SDT_AArch64unvec>;
218 def AArch64cmgez: SDNode<"AArch64ISD::CMGEz", SDT_AArch64unvec>;
219 def AArch64cmgtz: SDNode<"AArch64ISD::CMGTz", SDT_AArch64unvec>;
220 def AArch64cmlez: SDNode<"AArch64ISD::CMLEz", SDT_AArch64unvec>;
221 def AArch64cmltz: SDNode<"AArch64ISD::CMLTz", SDT_AArch64unvec>;
222 def AArch64cmtst : PatFrag<(ops node:$LHS, node:$RHS),
223 (AArch64not (AArch64cmeqz (and node:$LHS, node:$RHS)))>;
225 def AArch64fcmeqz: SDNode<"AArch64ISD::FCMEQz", SDT_AArch64fcmpz>;
226 def AArch64fcmgez: SDNode<"AArch64ISD::FCMGEz", SDT_AArch64fcmpz>;
227 def AArch64fcmgtz: SDNode<"AArch64ISD::FCMGTz", SDT_AArch64fcmpz>;
228 def AArch64fcmlez: SDNode<"AArch64ISD::FCMLEz", SDT_AArch64fcmpz>;
229 def AArch64fcmltz: SDNode<"AArch64ISD::FCMLTz", SDT_AArch64fcmpz>;
231 def AArch64bici: SDNode<"AArch64ISD::BICi", SDT_AArch64vecimm>;
232 def AArch64orri: SDNode<"AArch64ISD::ORRi", SDT_AArch64vecimm>;
234 def AArch64neg : SDNode<"AArch64ISD::NEG", SDT_AArch64unvec>;
236 def AArch64tcret: SDNode<"AArch64ISD::TC_RETURN", SDT_AArch64TCRET,
237 [SDNPHasChain, SDNPOptInGlue, SDNPVariadic]>;
239 def AArch64Prefetch : SDNode<"AArch64ISD::PREFETCH", SDT_AArch64PREFETCH,
240 [SDNPHasChain, SDNPSideEffect]>;
242 def AArch64sitof: SDNode<"AArch64ISD::SITOF", SDT_AArch64ITOF>;
243 def AArch64uitof: SDNode<"AArch64ISD::UITOF", SDT_AArch64ITOF>;
245 def AArch64tlsdesc_callseq : SDNode<"AArch64ISD::TLSDESC_CALLSEQ",
246 SDT_AArch64TLSDescCallSeq,
247 [SDNPInGlue, SDNPOutGlue, SDNPHasChain,
251 def AArch64WrapperLarge : SDNode<"AArch64ISD::WrapperLarge",
252 SDT_AArch64WrapperLarge>;
254 def AArch64NvCast : SDNode<"AArch64ISD::NVCAST", SDTUnaryOp>;
256 def SDT_AArch64mull : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
257 SDTCisSameAs<1, 2>]>;
258 def AArch64smull : SDNode<"AArch64ISD::SMULL", SDT_AArch64mull>;
259 def AArch64umull : SDNode<"AArch64ISD::UMULL", SDT_AArch64mull>;
261 //===----------------------------------------------------------------------===//
263 //===----------------------------------------------------------------------===//
265 // AArch64 Instruction Predicate Definitions.
267 def HasZCZ : Predicate<"Subtarget->hasZeroCycleZeroing()">;
268 def NoZCZ : Predicate<"!Subtarget->hasZeroCycleZeroing()">;
269 def IsDarwin : Predicate<"Subtarget->isTargetDarwin()">;
270 def IsNotDarwin: Predicate<"!Subtarget->isTargetDarwin()">;
271 def ForCodeSize : Predicate<"ForCodeSize">;
272 def NotForCodeSize : Predicate<"!ForCodeSize">;
274 include "AArch64InstrFormats.td"
276 //===----------------------------------------------------------------------===//
278 //===----------------------------------------------------------------------===//
279 // Miscellaneous instructions.
280 //===----------------------------------------------------------------------===//
282 let Defs = [SP], Uses = [SP], hasSideEffects = 1, isCodeGenOnly = 1 in {
283 def ADJCALLSTACKDOWN : Pseudo<(outs), (ins i32imm:$amt),
284 [(AArch64callseq_start timm:$amt)]>;
285 def ADJCALLSTACKUP : Pseudo<(outs), (ins i32imm:$amt1, i32imm:$amt2),
286 [(AArch64callseq_end timm:$amt1, timm:$amt2)]>;
287 } // Defs = [SP], Uses = [SP], hasSideEffects = 1, isCodeGenOnly = 1
289 let isReMaterializable = 1, isCodeGenOnly = 1 in {
290 // FIXME: The following pseudo instructions are only needed because remat
291 // cannot handle multiple instructions. When that changes, they can be
292 // removed, along with the AArch64Wrapper node.
294 let AddedComplexity = 10 in
295 def LOADgot : Pseudo<(outs GPR64:$dst), (ins i64imm:$addr),
296 [(set GPR64:$dst, (AArch64LOADgot tglobaladdr:$addr))]>,
299 // The MOVaddr instruction should match only when the add is not folded
300 // into a load or store address.
302 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
303 [(set GPR64:$dst, (AArch64addlow (AArch64adrp tglobaladdr:$hi),
304 tglobaladdr:$low))]>,
305 Sched<[WriteAdrAdr]>;
307 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
308 [(set GPR64:$dst, (AArch64addlow (AArch64adrp tjumptable:$hi),
310 Sched<[WriteAdrAdr]>;
312 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
313 [(set GPR64:$dst, (AArch64addlow (AArch64adrp tconstpool:$hi),
315 Sched<[WriteAdrAdr]>;
317 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
318 [(set GPR64:$dst, (AArch64addlow (AArch64adrp tblockaddress:$hi),
319 tblockaddress:$low))]>,
320 Sched<[WriteAdrAdr]>;
322 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
323 [(set GPR64:$dst, (AArch64addlow (AArch64adrp tglobaltlsaddr:$hi),
324 tglobaltlsaddr:$low))]>,
325 Sched<[WriteAdrAdr]>;
327 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
328 [(set GPR64:$dst, (AArch64addlow (AArch64adrp texternalsym:$hi),
329 texternalsym:$low))]>,
330 Sched<[WriteAdrAdr]>;
332 } // isReMaterializable, isCodeGenOnly
334 def : Pat<(AArch64LOADgot tglobaltlsaddr:$addr),
335 (LOADgot tglobaltlsaddr:$addr)>;
337 def : Pat<(AArch64LOADgot texternalsym:$addr),
338 (LOADgot texternalsym:$addr)>;
340 def : Pat<(AArch64LOADgot tconstpool:$addr),
341 (LOADgot tconstpool:$addr)>;
343 //===----------------------------------------------------------------------===//
344 // System instructions.
345 //===----------------------------------------------------------------------===//
347 def HINT : HintI<"hint">;
348 def : InstAlias<"nop", (HINT 0b000)>;
349 def : InstAlias<"yield",(HINT 0b001)>;
350 def : InstAlias<"wfe", (HINT 0b010)>;
351 def : InstAlias<"wfi", (HINT 0b011)>;
352 def : InstAlias<"sev", (HINT 0b100)>;
353 def : InstAlias<"sevl", (HINT 0b101)>;
355 // As far as LLVM is concerned this writes to the system's exclusive monitors.
356 let mayLoad = 1, mayStore = 1 in
357 def CLREX : CRmSystemI<imm0_15, 0b010, "clrex">;
359 // NOTE: ideally, this would have mayStore = 0, mayLoad = 0, but we cannot
360 // model patterns with sufficiently fine granularity.
361 let mayLoad = ?, mayStore = ? in {
362 def DMB : CRmSystemI<barrier_op, 0b101, "dmb",
363 [(int_aarch64_dmb (i32 imm32_0_15:$CRm))]>;
365 def DSB : CRmSystemI<barrier_op, 0b100, "dsb",
366 [(int_aarch64_dsb (i32 imm32_0_15:$CRm))]>;
368 def ISB : CRmSystemI<barrier_op, 0b110, "isb",
369 [(int_aarch64_isb (i32 imm32_0_15:$CRm))]>;
372 def : InstAlias<"clrex", (CLREX 0xf)>;
373 def : InstAlias<"isb", (ISB 0xf)>;
377 def MSRpstate: MSRpstateI;
379 // The thread pointer (on Linux, at least, where this has been implemented) is
381 def : Pat<(AArch64threadpointer), (MRS 0xde82)>;
383 // Generic system instructions
384 def SYSxt : SystemXtI<0, "sys">;
385 def SYSLxt : SystemLXtI<1, "sysl">;
387 def : InstAlias<"sys $op1, $Cn, $Cm, $op2",
388 (SYSxt imm0_7:$op1, sys_cr_op:$Cn,
389 sys_cr_op:$Cm, imm0_7:$op2, XZR)>;
391 //===----------------------------------------------------------------------===//
392 // Move immediate instructions.
393 //===----------------------------------------------------------------------===//
395 defm MOVK : InsertImmediate<0b11, "movk">;
396 defm MOVN : MoveImmediate<0b00, "movn">;
398 let PostEncoderMethod = "fixMOVZ" in
399 defm MOVZ : MoveImmediate<0b10, "movz">;
401 // First group of aliases covers an implicit "lsl #0".
402 def : InstAlias<"movk $dst, $imm", (MOVKWi GPR32:$dst, imm0_65535:$imm, 0)>;
403 def : InstAlias<"movk $dst, $imm", (MOVKXi GPR64:$dst, imm0_65535:$imm, 0)>;
404 def : InstAlias<"movn $dst, $imm", (MOVNWi GPR32:$dst, imm0_65535:$imm, 0)>;
405 def : InstAlias<"movn $dst, $imm", (MOVNXi GPR64:$dst, imm0_65535:$imm, 0)>;
406 def : InstAlias<"movz $dst, $imm", (MOVZWi GPR32:$dst, imm0_65535:$imm, 0)>;
407 def : InstAlias<"movz $dst, $imm", (MOVZXi GPR64:$dst, imm0_65535:$imm, 0)>;
409 // Next, we have various ELF relocations with the ":XYZ_g0:sym" syntax.
410 def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movz_symbol_g3:$sym, 48)>;
411 def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movz_symbol_g2:$sym, 32)>;
412 def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movz_symbol_g1:$sym, 16)>;
413 def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movz_symbol_g0:$sym, 0)>;
415 def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movz_symbol_g3:$sym, 48)>;
416 def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movz_symbol_g2:$sym, 32)>;
417 def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movz_symbol_g1:$sym, 16)>;
418 def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movz_symbol_g0:$sym, 0)>;
420 def : InstAlias<"movk $Rd, $sym", (MOVKXi GPR64:$Rd, movk_symbol_g3:$sym, 48)>;
421 def : InstAlias<"movk $Rd, $sym", (MOVKXi GPR64:$Rd, movk_symbol_g2:$sym, 32)>;
422 def : InstAlias<"movk $Rd, $sym", (MOVKXi GPR64:$Rd, movk_symbol_g1:$sym, 16)>;
423 def : InstAlias<"movk $Rd, $sym", (MOVKXi GPR64:$Rd, movk_symbol_g0:$sym, 0)>;
425 def : InstAlias<"movz $Rd, $sym", (MOVZWi GPR32:$Rd, movz_symbol_g1:$sym, 16)>;
426 def : InstAlias<"movz $Rd, $sym", (MOVZWi GPR32:$Rd, movz_symbol_g0:$sym, 0)>;
428 def : InstAlias<"movn $Rd, $sym", (MOVNWi GPR32:$Rd, movz_symbol_g1:$sym, 16)>;
429 def : InstAlias<"movn $Rd, $sym", (MOVNWi GPR32:$Rd, movz_symbol_g0:$sym, 0)>;
431 def : InstAlias<"movk $Rd, $sym", (MOVKWi GPR32:$Rd, movk_symbol_g1:$sym, 16)>;
432 def : InstAlias<"movk $Rd, $sym", (MOVKWi GPR32:$Rd, movk_symbol_g0:$sym, 0)>;
434 // Final group of aliases covers true "mov $Rd, $imm" cases.
435 multiclass movw_mov_alias<string basename,Instruction INST, RegisterClass GPR,
436 int width, int shift> {
437 def _asmoperand : AsmOperandClass {
438 let Name = basename # width # "_lsl" # shift # "MovAlias";
439 let PredicateMethod = "is" # basename # "MovAlias<" # width # ", "
441 let RenderMethod = "add" # basename # "MovAliasOperands<" # shift # ">";
444 def _movimm : Operand<i32> {
445 let ParserMatchClass = !cast<AsmOperandClass>(NAME # "_asmoperand");
448 def : InstAlias<"mov $Rd, $imm",
449 (INST GPR:$Rd, !cast<Operand>(NAME # "_movimm"):$imm, shift)>;
452 defm : movw_mov_alias<"MOVZ", MOVZWi, GPR32, 32, 0>;
453 defm : movw_mov_alias<"MOVZ", MOVZWi, GPR32, 32, 16>;
455 defm : movw_mov_alias<"MOVZ", MOVZXi, GPR64, 64, 0>;
456 defm : movw_mov_alias<"MOVZ", MOVZXi, GPR64, 64, 16>;
457 defm : movw_mov_alias<"MOVZ", MOVZXi, GPR64, 64, 32>;
458 defm : movw_mov_alias<"MOVZ", MOVZXi, GPR64, 64, 48>;
460 defm : movw_mov_alias<"MOVN", MOVNWi, GPR32, 32, 0>;
461 defm : movw_mov_alias<"MOVN", MOVNWi, GPR32, 32, 16>;
463 defm : movw_mov_alias<"MOVN", MOVNXi, GPR64, 64, 0>;
464 defm : movw_mov_alias<"MOVN", MOVNXi, GPR64, 64, 16>;
465 defm : movw_mov_alias<"MOVN", MOVNXi, GPR64, 64, 32>;
466 defm : movw_mov_alias<"MOVN", MOVNXi, GPR64, 64, 48>;
468 let isReMaterializable = 1, isCodeGenOnly = 1, isMoveImm = 1,
469 isAsCheapAsAMove = 1 in {
470 // FIXME: The following pseudo instructions are only needed because remat
471 // cannot handle multiple instructions. When that changes, we can select
472 // directly to the real instructions and get rid of these pseudos.
475 : Pseudo<(outs GPR32:$dst), (ins i32imm:$src),
476 [(set GPR32:$dst, imm:$src)]>,
479 : Pseudo<(outs GPR64:$dst), (ins i64imm:$src),
480 [(set GPR64:$dst, imm:$src)]>,
482 } // isReMaterializable, isCodeGenOnly
484 // If possible, we want to use MOVi32imm even for 64-bit moves. This gives the
485 // eventual expansion code fewer bits to worry about getting right. Marshalling
486 // the types is a little tricky though:
487 def i64imm_32bit : ImmLeaf<i64, [{
488 return (Imm & 0xffffffffULL) == static_cast<uint64_t>(Imm);
491 def trunc_imm : SDNodeXForm<imm, [{
492 return CurDAG->getTargetConstant(N->getZExtValue(), MVT::i32);
495 def : Pat<(i64 i64imm_32bit:$src),
496 (SUBREG_TO_REG (i64 0), (MOVi32imm (trunc_imm imm:$src)), sub_32)>;
498 // Materialize FP constants via MOVi32imm/MOVi64imm (MachO large code model).
499 def bitcast_fpimm_to_i32 : SDNodeXForm<fpimm, [{
500 return CurDAG->getTargetConstant(
501 N->getValueAPF().bitcastToAPInt().getZExtValue(), MVT::i32);
504 def bitcast_fpimm_to_i64 : SDNodeXForm<fpimm, [{
505 return CurDAG->getTargetConstant(
506 N->getValueAPF().bitcastToAPInt().getZExtValue(), MVT::i64);
510 def : Pat<(f32 fpimm:$in),
511 (COPY_TO_REGCLASS (MOVi32imm (bitcast_fpimm_to_i32 f32:$in)), FPR32)>;
512 def : Pat<(f64 fpimm:$in),
513 (COPY_TO_REGCLASS (MOVi64imm (bitcast_fpimm_to_i64 f64:$in)), FPR64)>;
516 // Deal with the various forms of (ELF) large addressing with MOVZ/MOVK
518 def : Pat<(AArch64WrapperLarge tglobaladdr:$g3, tglobaladdr:$g2,
519 tglobaladdr:$g1, tglobaladdr:$g0),
520 (MOVKXi (MOVKXi (MOVKXi (MOVZXi tglobaladdr:$g3, 48),
521 tglobaladdr:$g2, 32),
522 tglobaladdr:$g1, 16),
523 tglobaladdr:$g0, 0)>;
525 def : Pat<(AArch64WrapperLarge tblockaddress:$g3, tblockaddress:$g2,
526 tblockaddress:$g1, tblockaddress:$g0),
527 (MOVKXi (MOVKXi (MOVKXi (MOVZXi tblockaddress:$g3, 48),
528 tblockaddress:$g2, 32),
529 tblockaddress:$g1, 16),
530 tblockaddress:$g0, 0)>;
532 def : Pat<(AArch64WrapperLarge tconstpool:$g3, tconstpool:$g2,
533 tconstpool:$g1, tconstpool:$g0),
534 (MOVKXi (MOVKXi (MOVKXi (MOVZXi tconstpool:$g3, 48),
539 def : Pat<(AArch64WrapperLarge tjumptable:$g3, tjumptable:$g2,
540 tjumptable:$g1, tjumptable:$g0),
541 (MOVKXi (MOVKXi (MOVKXi (MOVZXi tjumptable:$g3, 48),
547 //===----------------------------------------------------------------------===//
548 // Arithmetic instructions.
549 //===----------------------------------------------------------------------===//
551 // Add/subtract with carry.
552 defm ADC : AddSubCarry<0, "adc", "adcs", AArch64adc, AArch64adc_flag>;
553 defm SBC : AddSubCarry<1, "sbc", "sbcs", AArch64sbc, AArch64sbc_flag>;
555 def : InstAlias<"ngc $dst, $src", (SBCWr GPR32:$dst, WZR, GPR32:$src)>;
556 def : InstAlias<"ngc $dst, $src", (SBCXr GPR64:$dst, XZR, GPR64:$src)>;
557 def : InstAlias<"ngcs $dst, $src", (SBCSWr GPR32:$dst, WZR, GPR32:$src)>;
558 def : InstAlias<"ngcs $dst, $src", (SBCSXr GPR64:$dst, XZR, GPR64:$src)>;
561 defm ADD : AddSub<0, "add", add>;
562 defm SUB : AddSub<1, "sub">;
564 def : InstAlias<"mov $dst, $src",
565 (ADDWri GPR32sponly:$dst, GPR32sp:$src, 0, 0)>;
566 def : InstAlias<"mov $dst, $src",
567 (ADDWri GPR32sp:$dst, GPR32sponly:$src, 0, 0)>;
568 def : InstAlias<"mov $dst, $src",
569 (ADDXri GPR64sponly:$dst, GPR64sp:$src, 0, 0)>;
570 def : InstAlias<"mov $dst, $src",
571 (ADDXri GPR64sp:$dst, GPR64sponly:$src, 0, 0)>;
573 defm ADDS : AddSubS<0, "adds", AArch64add_flag, "cmn">;
574 defm SUBS : AddSubS<1, "subs", AArch64sub_flag, "cmp">;
576 // Use SUBS instead of SUB to enable CSE between SUBS and SUB.
577 def : Pat<(sub GPR32sp:$Rn, addsub_shifted_imm32:$imm),
578 (SUBSWri GPR32sp:$Rn, addsub_shifted_imm32:$imm)>;
579 def : Pat<(sub GPR64sp:$Rn, addsub_shifted_imm64:$imm),
580 (SUBSXri GPR64sp:$Rn, addsub_shifted_imm64:$imm)>;
581 def : Pat<(sub GPR32:$Rn, GPR32:$Rm),
582 (SUBSWrr GPR32:$Rn, GPR32:$Rm)>;
583 def : Pat<(sub GPR64:$Rn, GPR64:$Rm),
584 (SUBSXrr GPR64:$Rn, GPR64:$Rm)>;
585 def : Pat<(sub GPR32:$Rn, arith_shifted_reg32:$Rm),
586 (SUBSWrs GPR32:$Rn, arith_shifted_reg32:$Rm)>;
587 def : Pat<(sub GPR64:$Rn, arith_shifted_reg64:$Rm),
588 (SUBSXrs GPR64:$Rn, arith_shifted_reg64:$Rm)>;
589 def : Pat<(sub GPR32sp:$R2, arith_extended_reg32<i32>:$R3),
590 (SUBSWrx GPR32sp:$R2, arith_extended_reg32<i32>:$R3)>;
591 def : Pat<(sub GPR64sp:$R2, arith_extended_reg32to64<i64>:$R3),
592 (SUBSXrx GPR64sp:$R2, arith_extended_reg32to64<i64>:$R3)>;
594 // Because of the immediate format for add/sub-imm instructions, the
595 // expression (add x, -1) must be transformed to (SUB{W,X}ri x, 1).
596 // These patterns capture that transformation.
597 let AddedComplexity = 1 in {
598 def : Pat<(add GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
599 (SUBSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
600 def : Pat<(add GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
601 (SUBSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
602 def : Pat<(sub GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
603 (ADDWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
604 def : Pat<(sub GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
605 (ADDXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
608 // Because of the immediate format for add/sub-imm instructions, the
609 // expression (add x, -1) must be transformed to (SUB{W,X}ri x, 1).
610 // These patterns capture that transformation.
611 let AddedComplexity = 1 in {
612 def : Pat<(AArch64add_flag GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
613 (SUBSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
614 def : Pat<(AArch64add_flag GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
615 (SUBSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
616 def : Pat<(AArch64sub_flag GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
617 (ADDSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
618 def : Pat<(AArch64sub_flag GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
619 (ADDSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
622 def : InstAlias<"neg $dst, $src", (SUBWrs GPR32:$dst, WZR, GPR32:$src, 0), 3>;
623 def : InstAlias<"neg $dst, $src", (SUBXrs GPR64:$dst, XZR, GPR64:$src, 0), 3>;
624 def : InstAlias<"neg $dst, $src$shift",
625 (SUBWrs GPR32:$dst, WZR, GPR32:$src, arith_shift32:$shift), 2>;
626 def : InstAlias<"neg $dst, $src$shift",
627 (SUBXrs GPR64:$dst, XZR, GPR64:$src, arith_shift64:$shift), 2>;
629 def : InstAlias<"negs $dst, $src", (SUBSWrs GPR32:$dst, WZR, GPR32:$src, 0), 3>;
630 def : InstAlias<"negs $dst, $src", (SUBSXrs GPR64:$dst, XZR, GPR64:$src, 0), 3>;
631 def : InstAlias<"negs $dst, $src$shift",
632 (SUBSWrs GPR32:$dst, WZR, GPR32:$src, arith_shift32:$shift), 2>;
633 def : InstAlias<"negs $dst, $src$shift",
634 (SUBSXrs GPR64:$dst, XZR, GPR64:$src, arith_shift64:$shift), 2>;
637 // Unsigned/Signed divide
638 defm UDIV : Div<0, "udiv", udiv>;
639 defm SDIV : Div<1, "sdiv", sdiv>;
640 let isCodeGenOnly = 1 in {
641 defm UDIV_Int : Div<0, "udiv", int_aarch64_udiv>;
642 defm SDIV_Int : Div<1, "sdiv", int_aarch64_sdiv>;
646 defm ASRV : Shift<0b10, "asr", sra>;
647 defm LSLV : Shift<0b00, "lsl", shl>;
648 defm LSRV : Shift<0b01, "lsr", srl>;
649 defm RORV : Shift<0b11, "ror", rotr>;
651 def : ShiftAlias<"asrv", ASRVWr, GPR32>;
652 def : ShiftAlias<"asrv", ASRVXr, GPR64>;
653 def : ShiftAlias<"lslv", LSLVWr, GPR32>;
654 def : ShiftAlias<"lslv", LSLVXr, GPR64>;
655 def : ShiftAlias<"lsrv", LSRVWr, GPR32>;
656 def : ShiftAlias<"lsrv", LSRVXr, GPR64>;
657 def : ShiftAlias<"rorv", RORVWr, GPR32>;
658 def : ShiftAlias<"rorv", RORVXr, GPR64>;
661 let AddedComplexity = 7 in {
662 defm MADD : MulAccum<0, "madd", add>;
663 defm MSUB : MulAccum<1, "msub", sub>;
665 def : Pat<(i32 (mul GPR32:$Rn, GPR32:$Rm)),
666 (MADDWrrr GPR32:$Rn, GPR32:$Rm, WZR)>;
667 def : Pat<(i64 (mul GPR64:$Rn, GPR64:$Rm)),
668 (MADDXrrr GPR64:$Rn, GPR64:$Rm, XZR)>;
670 def : Pat<(i32 (ineg (mul GPR32:$Rn, GPR32:$Rm))),
671 (MSUBWrrr GPR32:$Rn, GPR32:$Rm, WZR)>;
672 def : Pat<(i64 (ineg (mul GPR64:$Rn, GPR64:$Rm))),
673 (MSUBXrrr GPR64:$Rn, GPR64:$Rm, XZR)>;
674 def : Pat<(i32 (mul (ineg GPR32:$Rn), GPR32:$Rm)),
675 (MSUBWrrr GPR32:$Rn, GPR32:$Rm, WZR)>;
676 def : Pat<(i64 (mul (ineg GPR64:$Rn), GPR64:$Rm)),
677 (MSUBXrrr GPR64:$Rn, GPR64:$Rm, XZR)>;
678 } // AddedComplexity = 7
680 let AddedComplexity = 5 in {
681 def SMADDLrrr : WideMulAccum<0, 0b001, "smaddl", add, sext>;
682 def SMSUBLrrr : WideMulAccum<1, 0b001, "smsubl", sub, sext>;
683 def UMADDLrrr : WideMulAccum<0, 0b101, "umaddl", add, zext>;
684 def UMSUBLrrr : WideMulAccum<1, 0b101, "umsubl", sub, zext>;
686 def : Pat<(i64 (mul (sext GPR32:$Rn), (sext GPR32:$Rm))),
687 (SMADDLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
688 def : Pat<(i64 (mul (zext GPR32:$Rn), (zext GPR32:$Rm))),
689 (UMADDLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
691 def : Pat<(i64 (ineg (mul (sext GPR32:$Rn), (sext GPR32:$Rm)))),
692 (SMSUBLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
693 def : Pat<(i64 (ineg (mul (zext GPR32:$Rn), (zext GPR32:$Rm)))),
694 (UMSUBLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
695 } // AddedComplexity = 5
697 def : MulAccumWAlias<"mul", MADDWrrr>;
698 def : MulAccumXAlias<"mul", MADDXrrr>;
699 def : MulAccumWAlias<"mneg", MSUBWrrr>;
700 def : MulAccumXAlias<"mneg", MSUBXrrr>;
701 def : WideMulAccumAlias<"smull", SMADDLrrr>;
702 def : WideMulAccumAlias<"smnegl", SMSUBLrrr>;
703 def : WideMulAccumAlias<"umull", UMADDLrrr>;
704 def : WideMulAccumAlias<"umnegl", UMSUBLrrr>;
707 def SMULHrr : MulHi<0b010, "smulh", mulhs>;
708 def UMULHrr : MulHi<0b110, "umulh", mulhu>;
711 def CRC32Brr : BaseCRC32<0, 0b00, 0, GPR32, int_aarch64_crc32b, "crc32b">;
712 def CRC32Hrr : BaseCRC32<0, 0b01, 0, GPR32, int_aarch64_crc32h, "crc32h">;
713 def CRC32Wrr : BaseCRC32<0, 0b10, 0, GPR32, int_aarch64_crc32w, "crc32w">;
714 def CRC32Xrr : BaseCRC32<1, 0b11, 0, GPR64, int_aarch64_crc32x, "crc32x">;
716 def CRC32CBrr : BaseCRC32<0, 0b00, 1, GPR32, int_aarch64_crc32cb, "crc32cb">;
717 def CRC32CHrr : BaseCRC32<0, 0b01, 1, GPR32, int_aarch64_crc32ch, "crc32ch">;
718 def CRC32CWrr : BaseCRC32<0, 0b10, 1, GPR32, int_aarch64_crc32cw, "crc32cw">;
719 def CRC32CXrr : BaseCRC32<1, 0b11, 1, GPR64, int_aarch64_crc32cx, "crc32cx">;
722 //===----------------------------------------------------------------------===//
723 // Logical instructions.
724 //===----------------------------------------------------------------------===//
727 defm ANDS : LogicalImmS<0b11, "ands", AArch64and_flag, "bics">;
728 defm AND : LogicalImm<0b00, "and", and, "bic">;
729 defm EOR : LogicalImm<0b10, "eor", xor, "eon">;
730 defm ORR : LogicalImm<0b01, "orr", or, "orn">;
732 // FIXME: these aliases *are* canonical sometimes (when movz can't be
733 // used). Actually, it seems to be working right now, but putting logical_immXX
734 // here is a bit dodgy on the AsmParser side too.
735 def : InstAlias<"mov $dst, $imm", (ORRWri GPR32sp:$dst, WZR,
736 logical_imm32:$imm), 0>;
737 def : InstAlias<"mov $dst, $imm", (ORRXri GPR64sp:$dst, XZR,
738 logical_imm64:$imm), 0>;
742 defm ANDS : LogicalRegS<0b11, 0, "ands", AArch64and_flag>;
743 defm BICS : LogicalRegS<0b11, 1, "bics",
744 BinOpFrag<(AArch64and_flag node:$LHS, (not node:$RHS))>>;
745 defm AND : LogicalReg<0b00, 0, "and", and>;
746 defm BIC : LogicalReg<0b00, 1, "bic",
747 BinOpFrag<(and node:$LHS, (not node:$RHS))>>;
748 defm EON : LogicalReg<0b10, 1, "eon",
749 BinOpFrag<(xor node:$LHS, (not node:$RHS))>>;
750 defm EOR : LogicalReg<0b10, 0, "eor", xor>;
751 defm ORN : LogicalReg<0b01, 1, "orn",
752 BinOpFrag<(or node:$LHS, (not node:$RHS))>>;
753 defm ORR : LogicalReg<0b01, 0, "orr", or>;
755 def : InstAlias<"mov $dst, $src", (ORRWrs GPR32:$dst, WZR, GPR32:$src, 0), 2>;
756 def : InstAlias<"mov $dst, $src", (ORRXrs GPR64:$dst, XZR, GPR64:$src, 0), 2>;
758 def : InstAlias<"mvn $Wd, $Wm", (ORNWrs GPR32:$Wd, WZR, GPR32:$Wm, 0), 3>;
759 def : InstAlias<"mvn $Xd, $Xm", (ORNXrs GPR64:$Xd, XZR, GPR64:$Xm, 0), 3>;
761 def : InstAlias<"mvn $Wd, $Wm$sh",
762 (ORNWrs GPR32:$Wd, WZR, GPR32:$Wm, logical_shift32:$sh), 2>;
763 def : InstAlias<"mvn $Xd, $Xm$sh",
764 (ORNXrs GPR64:$Xd, XZR, GPR64:$Xm, logical_shift64:$sh), 2>;
766 def : InstAlias<"tst $src1, $src2",
767 (ANDSWri WZR, GPR32:$src1, logical_imm32:$src2), 2>;
768 def : InstAlias<"tst $src1, $src2",
769 (ANDSXri XZR, GPR64:$src1, logical_imm64:$src2), 2>;
771 def : InstAlias<"tst $src1, $src2",
772 (ANDSWrs WZR, GPR32:$src1, GPR32:$src2, 0), 3>;
773 def : InstAlias<"tst $src1, $src2",
774 (ANDSXrs XZR, GPR64:$src1, GPR64:$src2, 0), 3>;
776 def : InstAlias<"tst $src1, $src2$sh",
777 (ANDSWrs WZR, GPR32:$src1, GPR32:$src2, logical_shift32:$sh), 2>;
778 def : InstAlias<"tst $src1, $src2$sh",
779 (ANDSXrs XZR, GPR64:$src1, GPR64:$src2, logical_shift64:$sh), 2>;
782 def : Pat<(not GPR32:$Wm), (ORNWrr WZR, GPR32:$Wm)>;
783 def : Pat<(not GPR64:$Xm), (ORNXrr XZR, GPR64:$Xm)>;
786 //===----------------------------------------------------------------------===//
787 // One operand data processing instructions.
788 //===----------------------------------------------------------------------===//
790 defm CLS : OneOperandData<0b101, "cls">;
791 defm CLZ : OneOperandData<0b100, "clz", ctlz>;
792 defm RBIT : OneOperandData<0b000, "rbit">;
794 def : Pat<(int_aarch64_rbit GPR32:$Rn), (RBITWr $Rn)>;
795 def : Pat<(int_aarch64_rbit GPR64:$Rn), (RBITXr $Rn)>;
797 def REV16Wr : OneWRegData<0b001, "rev16",
798 UnOpFrag<(rotr (bswap node:$LHS), (i64 16))>>;
799 def REV16Xr : OneXRegData<0b001, "rev16", null_frag>;
801 def : Pat<(cttz GPR32:$Rn),
802 (CLZWr (RBITWr GPR32:$Rn))>;
803 def : Pat<(cttz GPR64:$Rn),
804 (CLZXr (RBITXr GPR64:$Rn))>;
805 def : Pat<(ctlz (or (shl (xor (sra GPR32:$Rn, (i64 31)), GPR32:$Rn), (i64 1)),
808 def : Pat<(ctlz (or (shl (xor (sra GPR64:$Rn, (i64 63)), GPR64:$Rn), (i64 1)),
812 // Unlike the other one operand instructions, the instructions with the "rev"
813 // mnemonic do *not* just different in the size bit, but actually use different
814 // opcode bits for the different sizes.
815 def REVWr : OneWRegData<0b010, "rev", bswap>;
816 def REVXr : OneXRegData<0b011, "rev", bswap>;
817 def REV32Xr : OneXRegData<0b010, "rev32",
818 UnOpFrag<(rotr (bswap node:$LHS), (i64 32))>>;
820 // The bswap commutes with the rotr so we want a pattern for both possible
822 def : Pat<(bswap (rotr GPR32:$Rn, (i64 16))), (REV16Wr GPR32:$Rn)>;
823 def : Pat<(bswap (rotr GPR64:$Rn, (i64 32))), (REV32Xr GPR64:$Rn)>;
825 //===----------------------------------------------------------------------===//
826 // Bitfield immediate extraction instruction.
827 //===----------------------------------------------------------------------===//
828 let hasSideEffects = 0 in
829 defm EXTR : ExtractImm<"extr">;
830 def : InstAlias<"ror $dst, $src, $shift",
831 (EXTRWrri GPR32:$dst, GPR32:$src, GPR32:$src, imm0_31:$shift)>;
832 def : InstAlias<"ror $dst, $src, $shift",
833 (EXTRXrri GPR64:$dst, GPR64:$src, GPR64:$src, imm0_63:$shift)>;
835 def : Pat<(rotr GPR32:$Rn, (i64 imm0_31:$imm)),
836 (EXTRWrri GPR32:$Rn, GPR32:$Rn, imm0_31:$imm)>;
837 def : Pat<(rotr GPR64:$Rn, (i64 imm0_63:$imm)),
838 (EXTRXrri GPR64:$Rn, GPR64:$Rn, imm0_63:$imm)>;
840 //===----------------------------------------------------------------------===//
841 // Other bitfield immediate instructions.
842 //===----------------------------------------------------------------------===//
843 let hasSideEffects = 0 in {
844 defm BFM : BitfieldImmWith2RegArgs<0b01, "bfm">;
845 defm SBFM : BitfieldImm<0b00, "sbfm">;
846 defm UBFM : BitfieldImm<0b10, "ubfm">;
849 def i32shift_a : Operand<i64>, SDNodeXForm<imm, [{
850 uint64_t enc = (32 - N->getZExtValue()) & 0x1f;
851 return CurDAG->getTargetConstant(enc, MVT::i64);
854 def i32shift_b : Operand<i64>, SDNodeXForm<imm, [{
855 uint64_t enc = 31 - N->getZExtValue();
856 return CurDAG->getTargetConstant(enc, MVT::i64);
859 // min(7, 31 - shift_amt)
860 def i32shift_sext_i8 : Operand<i64>, SDNodeXForm<imm, [{
861 uint64_t enc = 31 - N->getZExtValue();
862 enc = enc > 7 ? 7 : enc;
863 return CurDAG->getTargetConstant(enc, MVT::i64);
866 // min(15, 31 - shift_amt)
867 def i32shift_sext_i16 : Operand<i64>, SDNodeXForm<imm, [{
868 uint64_t enc = 31 - N->getZExtValue();
869 enc = enc > 15 ? 15 : enc;
870 return CurDAG->getTargetConstant(enc, MVT::i64);
873 def i64shift_a : Operand<i64>, SDNodeXForm<imm, [{
874 uint64_t enc = (64 - N->getZExtValue()) & 0x3f;
875 return CurDAG->getTargetConstant(enc, MVT::i64);
878 def i64shift_b : Operand<i64>, SDNodeXForm<imm, [{
879 uint64_t enc = 63 - N->getZExtValue();
880 return CurDAG->getTargetConstant(enc, MVT::i64);
883 // min(7, 63 - shift_amt)
884 def i64shift_sext_i8 : Operand<i64>, SDNodeXForm<imm, [{
885 uint64_t enc = 63 - N->getZExtValue();
886 enc = enc > 7 ? 7 : enc;
887 return CurDAG->getTargetConstant(enc, MVT::i64);
890 // min(15, 63 - shift_amt)
891 def i64shift_sext_i16 : Operand<i64>, SDNodeXForm<imm, [{
892 uint64_t enc = 63 - N->getZExtValue();
893 enc = enc > 15 ? 15 : enc;
894 return CurDAG->getTargetConstant(enc, MVT::i64);
897 // min(31, 63 - shift_amt)
898 def i64shift_sext_i32 : Operand<i64>, SDNodeXForm<imm, [{
899 uint64_t enc = 63 - N->getZExtValue();
900 enc = enc > 31 ? 31 : enc;
901 return CurDAG->getTargetConstant(enc, MVT::i64);
904 def : Pat<(shl GPR32:$Rn, (i64 imm0_31:$imm)),
905 (UBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
906 (i64 (i32shift_b imm0_31:$imm)))>;
907 def : Pat<(shl GPR64:$Rn, (i64 imm0_63:$imm)),
908 (UBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
909 (i64 (i64shift_b imm0_63:$imm)))>;
911 let AddedComplexity = 10 in {
912 def : Pat<(sra GPR32:$Rn, (i64 imm0_31:$imm)),
913 (SBFMWri GPR32:$Rn, imm0_31:$imm, 31)>;
914 def : Pat<(sra GPR64:$Rn, (i64 imm0_63:$imm)),
915 (SBFMXri GPR64:$Rn, imm0_63:$imm, 63)>;
918 def : InstAlias<"asr $dst, $src, $shift",
919 (SBFMWri GPR32:$dst, GPR32:$src, imm0_31:$shift, 31)>;
920 def : InstAlias<"asr $dst, $src, $shift",
921 (SBFMXri GPR64:$dst, GPR64:$src, imm0_63:$shift, 63)>;
922 def : InstAlias<"sxtb $dst, $src", (SBFMWri GPR32:$dst, GPR32:$src, 0, 7)>;
923 def : InstAlias<"sxtb $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 7)>;
924 def : InstAlias<"sxth $dst, $src", (SBFMWri GPR32:$dst, GPR32:$src, 0, 15)>;
925 def : InstAlias<"sxth $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 15)>;
926 def : InstAlias<"sxtw $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 31)>;
928 def : Pat<(srl GPR32:$Rn, (i64 imm0_31:$imm)),
929 (UBFMWri GPR32:$Rn, imm0_31:$imm, 31)>;
930 def : Pat<(srl GPR64:$Rn, (i64 imm0_63:$imm)),
931 (UBFMXri GPR64:$Rn, imm0_63:$imm, 63)>;
933 def : InstAlias<"lsr $dst, $src, $shift",
934 (UBFMWri GPR32:$dst, GPR32:$src, imm0_31:$shift, 31)>;
935 def : InstAlias<"lsr $dst, $src, $shift",
936 (UBFMXri GPR64:$dst, GPR64:$src, imm0_63:$shift, 63)>;
937 def : InstAlias<"uxtb $dst, $src", (UBFMWri GPR32:$dst, GPR32:$src, 0, 7)>;
938 def : InstAlias<"uxtb $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 7)>;
939 def : InstAlias<"uxth $dst, $src", (UBFMWri GPR32:$dst, GPR32:$src, 0, 15)>;
940 def : InstAlias<"uxth $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 15)>;
941 def : InstAlias<"uxtw $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 31)>;
943 //===----------------------------------------------------------------------===//
944 // Conditionally set flags instructions.
945 //===----------------------------------------------------------------------===//
946 defm CCMN : CondSetFlagsImm<0, "ccmn">;
947 defm CCMP : CondSetFlagsImm<1, "ccmp">;
949 defm CCMN : CondSetFlagsReg<0, "ccmn">;
950 defm CCMP : CondSetFlagsReg<1, "ccmp">;
952 //===----------------------------------------------------------------------===//
953 // Conditional select instructions.
954 //===----------------------------------------------------------------------===//
955 defm CSEL : CondSelect<0, 0b00, "csel">;
957 def inc : PatFrag<(ops node:$in), (add node:$in, 1)>;
958 defm CSINC : CondSelectOp<0, 0b01, "csinc", inc>;
959 defm CSINV : CondSelectOp<1, 0b00, "csinv", not>;
960 defm CSNEG : CondSelectOp<1, 0b01, "csneg", ineg>;
962 def : Pat<(AArch64csinv GPR32:$tval, GPR32:$fval, (i32 imm:$cc), NZCV),
963 (CSINVWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
964 def : Pat<(AArch64csinv GPR64:$tval, GPR64:$fval, (i32 imm:$cc), NZCV),
965 (CSINVXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
966 def : Pat<(AArch64csneg GPR32:$tval, GPR32:$fval, (i32 imm:$cc), NZCV),
967 (CSNEGWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
968 def : Pat<(AArch64csneg GPR64:$tval, GPR64:$fval, (i32 imm:$cc), NZCV),
969 (CSNEGXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
970 def : Pat<(AArch64csinc GPR32:$tval, GPR32:$fval, (i32 imm:$cc), NZCV),
971 (CSINCWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
972 def : Pat<(AArch64csinc GPR64:$tval, GPR64:$fval, (i32 imm:$cc), NZCV),
973 (CSINCXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
975 def : Pat<(AArch64csel (i32 0), (i32 1), (i32 imm:$cc), NZCV),
976 (CSINCWr WZR, WZR, (i32 imm:$cc))>;
977 def : Pat<(AArch64csel (i64 0), (i64 1), (i32 imm:$cc), NZCV),
978 (CSINCXr XZR, XZR, (i32 imm:$cc))>;
979 def : Pat<(AArch64csel (i32 0), (i32 -1), (i32 imm:$cc), NZCV),
980 (CSINVWr WZR, WZR, (i32 imm:$cc))>;
981 def : Pat<(AArch64csel (i64 0), (i64 -1), (i32 imm:$cc), NZCV),
982 (CSINVXr XZR, XZR, (i32 imm:$cc))>;
984 // The inverse of the condition code from the alias instruction is what is used
985 // in the aliased instruction. The parser all ready inverts the condition code
986 // for these aliases.
987 def : InstAlias<"cset $dst, $cc",
988 (CSINCWr GPR32:$dst, WZR, WZR, inv_ccode:$cc)>;
989 def : InstAlias<"cset $dst, $cc",
990 (CSINCXr GPR64:$dst, XZR, XZR, inv_ccode:$cc)>;
992 def : InstAlias<"csetm $dst, $cc",
993 (CSINVWr GPR32:$dst, WZR, WZR, inv_ccode:$cc)>;
994 def : InstAlias<"csetm $dst, $cc",
995 (CSINVXr GPR64:$dst, XZR, XZR, inv_ccode:$cc)>;
997 def : InstAlias<"cinc $dst, $src, $cc",
998 (CSINCWr GPR32:$dst, GPR32:$src, GPR32:$src, inv_ccode:$cc)>;
999 def : InstAlias<"cinc $dst, $src, $cc",
1000 (CSINCXr GPR64:$dst, GPR64:$src, GPR64:$src, inv_ccode:$cc)>;
1002 def : InstAlias<"cinv $dst, $src, $cc",
1003 (CSINVWr GPR32:$dst, GPR32:$src, GPR32:$src, inv_ccode:$cc)>;
1004 def : InstAlias<"cinv $dst, $src, $cc",
1005 (CSINVXr GPR64:$dst, GPR64:$src, GPR64:$src, inv_ccode:$cc)>;
1007 def : InstAlias<"cneg $dst, $src, $cc",
1008 (CSNEGWr GPR32:$dst, GPR32:$src, GPR32:$src, inv_ccode:$cc)>;
1009 def : InstAlias<"cneg $dst, $src, $cc",
1010 (CSNEGXr GPR64:$dst, GPR64:$src, GPR64:$src, inv_ccode:$cc)>;
1012 //===----------------------------------------------------------------------===//
1013 // PC-relative instructions.
1014 //===----------------------------------------------------------------------===//
1015 let isReMaterializable = 1 in {
1016 let hasSideEffects = 0, mayStore = 0, mayLoad = 0 in {
1017 def ADR : ADRI<0, "adr", adrlabel, []>;
1018 } // hasSideEffects = 0
1020 def ADRP : ADRI<1, "adrp", adrplabel,
1021 [(set GPR64:$Xd, (AArch64adrp tglobaladdr:$label))]>;
1022 } // isReMaterializable = 1
1024 // page address of a constant pool entry, block address
1025 def : Pat<(AArch64adrp tconstpool:$cp), (ADRP tconstpool:$cp)>;
1026 def : Pat<(AArch64adrp tblockaddress:$cp), (ADRP tblockaddress:$cp)>;
1028 //===----------------------------------------------------------------------===//
1029 // Unconditional branch (register) instructions.
1030 //===----------------------------------------------------------------------===//
1032 let isReturn = 1, isTerminator = 1, isBarrier = 1 in {
1033 def RET : BranchReg<0b0010, "ret", []>;
1034 def DRPS : SpecialReturn<0b0101, "drps">;
1035 def ERET : SpecialReturn<0b0100, "eret">;
1036 } // isReturn = 1, isTerminator = 1, isBarrier = 1
1038 // Default to the LR register.
1039 def : InstAlias<"ret", (RET LR)>;
1041 let isCall = 1, Defs = [LR], Uses = [SP] in {
1042 def BLR : BranchReg<0b0001, "blr", [(AArch64call GPR64:$Rn)]>;
1045 let isBranch = 1, isTerminator = 1, isBarrier = 1, isIndirectBranch = 1 in {
1046 def BR : BranchReg<0b0000, "br", [(brind GPR64:$Rn)]>;
1047 } // isBranch, isTerminator, isBarrier, isIndirectBranch
1049 // Create a separate pseudo-instruction for codegen to use so that we don't
1050 // flag lr as used in every function. It'll be restored before the RET by the
1051 // epilogue if it's legitimately used.
1052 def RET_ReallyLR : Pseudo<(outs), (ins), [(AArch64retflag)]> {
1053 let isTerminator = 1;
1058 // This is a directive-like pseudo-instruction. The purpose is to insert an
1059 // R_AARCH64_TLSDESC_CALL relocation at the offset of the following instruction
1060 // (which in the usual case is a BLR).
1061 let hasSideEffects = 1 in
1062 def TLSDESCCALL : Pseudo<(outs), (ins i64imm:$sym), []> {
1063 let AsmString = ".tlsdesccall $sym";
1066 // FIXME: maybe the scratch register used shouldn't be fixed to X1?
1067 // FIXME: can "hasSideEffects be dropped?
1068 let isCall = 1, Defs = [LR, X0, X1], hasSideEffects = 1,
1069 isCodeGenOnly = 1 in
1071 : Pseudo<(outs), (ins i64imm:$sym),
1072 [(AArch64tlsdesc_callseq tglobaltlsaddr:$sym)]>;
1073 def : Pat<(AArch64tlsdesc_callseq texternalsym:$sym),
1074 (TLSDESC_CALLSEQ texternalsym:$sym)>;
1076 //===----------------------------------------------------------------------===//
1077 // Conditional branch (immediate) instruction.
1078 //===----------------------------------------------------------------------===//
1079 def Bcc : BranchCond;
1081 //===----------------------------------------------------------------------===//
1082 // Compare-and-branch instructions.
1083 //===----------------------------------------------------------------------===//
1084 defm CBZ : CmpBranch<0, "cbz", AArch64cbz>;
1085 defm CBNZ : CmpBranch<1, "cbnz", AArch64cbnz>;
1087 //===----------------------------------------------------------------------===//
1088 // Test-bit-and-branch instructions.
1089 //===----------------------------------------------------------------------===//
1090 defm TBZ : TestBranch<0, "tbz", AArch64tbz>;
1091 defm TBNZ : TestBranch<1, "tbnz", AArch64tbnz>;
1093 //===----------------------------------------------------------------------===//
1094 // Unconditional branch (immediate) instructions.
1095 //===----------------------------------------------------------------------===//
1096 let isBranch = 1, isTerminator = 1, isBarrier = 1 in {
1097 def B : BranchImm<0, "b", [(br bb:$addr)]>;
1098 } // isBranch, isTerminator, isBarrier
1100 let isCall = 1, Defs = [LR], Uses = [SP] in {
1101 def BL : CallImm<1, "bl", [(AArch64call tglobaladdr:$addr)]>;
1103 def : Pat<(AArch64call texternalsym:$func), (BL texternalsym:$func)>;
1105 //===----------------------------------------------------------------------===//
1106 // Exception generation instructions.
1107 //===----------------------------------------------------------------------===//
1108 def BRK : ExceptionGeneration<0b001, 0b00, "brk">;
1109 def DCPS1 : ExceptionGeneration<0b101, 0b01, "dcps1">;
1110 def DCPS2 : ExceptionGeneration<0b101, 0b10, "dcps2">;
1111 def DCPS3 : ExceptionGeneration<0b101, 0b11, "dcps3">;
1112 def HLT : ExceptionGeneration<0b010, 0b00, "hlt">;
1113 def HVC : ExceptionGeneration<0b000, 0b10, "hvc">;
1114 def SMC : ExceptionGeneration<0b000, 0b11, "smc">;
1115 def SVC : ExceptionGeneration<0b000, 0b01, "svc">;
1117 // DCPSn defaults to an immediate operand of zero if unspecified.
1118 def : InstAlias<"dcps1", (DCPS1 0)>;
1119 def : InstAlias<"dcps2", (DCPS2 0)>;
1120 def : InstAlias<"dcps3", (DCPS3 0)>;
1122 //===----------------------------------------------------------------------===//
1123 // Load instructions.
1124 //===----------------------------------------------------------------------===//
1126 // Pair (indexed, offset)
1127 defm LDPW : LoadPairOffset<0b00, 0, GPR32, simm7s4, "ldp">;
1128 defm LDPX : LoadPairOffset<0b10, 0, GPR64, simm7s8, "ldp">;
1129 defm LDPS : LoadPairOffset<0b00, 1, FPR32, simm7s4, "ldp">;
1130 defm LDPD : LoadPairOffset<0b01, 1, FPR64, simm7s8, "ldp">;
1131 defm LDPQ : LoadPairOffset<0b10, 1, FPR128, simm7s16, "ldp">;
1133 defm LDPSW : LoadPairOffset<0b01, 0, GPR64, simm7s4, "ldpsw">;
1135 // Pair (pre-indexed)
1136 def LDPWpre : LoadPairPreIdx<0b00, 0, GPR32, simm7s4, "ldp">;
1137 def LDPXpre : LoadPairPreIdx<0b10, 0, GPR64, simm7s8, "ldp">;
1138 def LDPSpre : LoadPairPreIdx<0b00, 1, FPR32, simm7s4, "ldp">;
1139 def LDPDpre : LoadPairPreIdx<0b01, 1, FPR64, simm7s8, "ldp">;
1140 def LDPQpre : LoadPairPreIdx<0b10, 1, FPR128, simm7s16, "ldp">;
1142 def LDPSWpre : LoadPairPreIdx<0b01, 0, GPR64, simm7s4, "ldpsw">;
1144 // Pair (post-indexed)
1145 def LDPWpost : LoadPairPostIdx<0b00, 0, GPR32, simm7s4, "ldp">;
1146 def LDPXpost : LoadPairPostIdx<0b10, 0, GPR64, simm7s8, "ldp">;
1147 def LDPSpost : LoadPairPostIdx<0b00, 1, FPR32, simm7s4, "ldp">;
1148 def LDPDpost : LoadPairPostIdx<0b01, 1, FPR64, simm7s8, "ldp">;
1149 def LDPQpost : LoadPairPostIdx<0b10, 1, FPR128, simm7s16, "ldp">;
1151 def LDPSWpost : LoadPairPostIdx<0b01, 0, GPR64, simm7s4, "ldpsw">;
1154 // Pair (no allocate)
1155 defm LDNPW : LoadPairNoAlloc<0b00, 0, GPR32, simm7s4, "ldnp">;
1156 defm LDNPX : LoadPairNoAlloc<0b10, 0, GPR64, simm7s8, "ldnp">;
1157 defm LDNPS : LoadPairNoAlloc<0b00, 1, FPR32, simm7s4, "ldnp">;
1158 defm LDNPD : LoadPairNoAlloc<0b01, 1, FPR64, simm7s8, "ldnp">;
1159 defm LDNPQ : LoadPairNoAlloc<0b10, 1, FPR128, simm7s16, "ldnp">;
1162 // (register offset)
1166 defm LDRBB : Load8RO<0b00, 0, 0b01, GPR32, "ldrb", i32, zextloadi8>;
1167 defm LDRHH : Load16RO<0b01, 0, 0b01, GPR32, "ldrh", i32, zextloadi16>;
1168 defm LDRW : Load32RO<0b10, 0, 0b01, GPR32, "ldr", i32, load>;
1169 defm LDRX : Load64RO<0b11, 0, 0b01, GPR64, "ldr", i64, load>;
1172 defm LDRB : Load8RO<0b00, 1, 0b01, FPR8, "ldr", untyped, load>;
1173 defm LDRH : Load16RO<0b01, 1, 0b01, FPR16, "ldr", f16, load>;
1174 defm LDRS : Load32RO<0b10, 1, 0b01, FPR32, "ldr", f32, load>;
1175 defm LDRD : Load64RO<0b11, 1, 0b01, FPR64, "ldr", f64, load>;
1176 defm LDRQ : Load128RO<0b00, 1, 0b11, FPR128, "ldr", f128, load>;
1178 // Load sign-extended half-word
1179 defm LDRSHW : Load16RO<0b01, 0, 0b11, GPR32, "ldrsh", i32, sextloadi16>;
1180 defm LDRSHX : Load16RO<0b01, 0, 0b10, GPR64, "ldrsh", i64, sextloadi16>;
1182 // Load sign-extended byte
1183 defm LDRSBW : Load8RO<0b00, 0, 0b11, GPR32, "ldrsb", i32, sextloadi8>;
1184 defm LDRSBX : Load8RO<0b00, 0, 0b10, GPR64, "ldrsb", i64, sextloadi8>;
1186 // Load sign-extended word
1187 defm LDRSW : Load32RO<0b10, 0, 0b10, GPR64, "ldrsw", i64, sextloadi32>;
1190 defm PRFM : PrefetchRO<0b11, 0, 0b10, "prfm">;
1192 // For regular load, we do not have any alignment requirement.
1193 // Thus, it is safe to directly map the vector loads with interesting
1194 // addressing modes.
1195 // FIXME: We could do the same for bitconvert to floating point vectors.
1196 multiclass ScalToVecROLoadPat<ROAddrMode ro, SDPatternOperator loadop,
1197 ValueType ScalTy, ValueType VecTy,
1198 Instruction LOADW, Instruction LOADX,
1200 def : Pat<(VecTy (scalar_to_vector (ScalTy
1201 (loadop (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$offset))))),
1202 (INSERT_SUBREG (VecTy (IMPLICIT_DEF)),
1203 (LOADW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$offset),
1206 def : Pat<(VecTy (scalar_to_vector (ScalTy
1207 (loadop (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$offset))))),
1208 (INSERT_SUBREG (VecTy (IMPLICIT_DEF)),
1209 (LOADX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$offset),
1213 let AddedComplexity = 10 in {
1214 defm : ScalToVecROLoadPat<ro8, extloadi8, i32, v8i8, LDRBroW, LDRBroX, bsub>;
1215 defm : ScalToVecROLoadPat<ro8, extloadi8, i32, v16i8, LDRBroW, LDRBroX, bsub>;
1217 defm : ScalToVecROLoadPat<ro16, extloadi16, i32, v4i16, LDRHroW, LDRHroX, hsub>;
1218 defm : ScalToVecROLoadPat<ro16, extloadi16, i32, v8i16, LDRHroW, LDRHroX, hsub>;
1220 defm : ScalToVecROLoadPat<ro16, load, i32, v4f16, LDRHroW, LDRHroX, hsub>;
1221 defm : ScalToVecROLoadPat<ro16, load, i32, v8f16, LDRHroW, LDRHroX, hsub>;
1223 defm : ScalToVecROLoadPat<ro32, load, i32, v2i32, LDRSroW, LDRSroX, ssub>;
1224 defm : ScalToVecROLoadPat<ro32, load, i32, v4i32, LDRSroW, LDRSroX, ssub>;
1226 defm : ScalToVecROLoadPat<ro32, load, f32, v2f32, LDRSroW, LDRSroX, ssub>;
1227 defm : ScalToVecROLoadPat<ro32, load, f32, v4f32, LDRSroW, LDRSroX, ssub>;
1229 defm : ScalToVecROLoadPat<ro64, load, i64, v2i64, LDRDroW, LDRDroX, dsub>;
1231 defm : ScalToVecROLoadPat<ro64, load, f64, v2f64, LDRDroW, LDRDroX, dsub>;
1234 def : Pat <(v1i64 (scalar_to_vector (i64
1235 (load (ro_Windexed64 GPR64sp:$Rn, GPR32:$Rm,
1236 ro_Wextend64:$extend))))),
1237 (LDRDroW GPR64sp:$Rn, GPR32:$Rm, ro_Wextend64:$extend)>;
1239 def : Pat <(v1i64 (scalar_to_vector (i64
1240 (load (ro_Xindexed64 GPR64sp:$Rn, GPR64:$Rm,
1241 ro_Xextend64:$extend))))),
1242 (LDRDroX GPR64sp:$Rn, GPR64:$Rm, ro_Xextend64:$extend)>;
1245 // Match all load 64 bits width whose type is compatible with FPR64
1246 multiclass VecROLoadPat<ROAddrMode ro, ValueType VecTy,
1247 Instruction LOADW, Instruction LOADX> {
1249 def : Pat<(VecTy (load (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend))),
1250 (LOADW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
1252 def : Pat<(VecTy (load (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend))),
1253 (LOADX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
1256 let AddedComplexity = 10 in {
1257 let Predicates = [IsLE] in {
1258 // We must do vector loads with LD1 in big-endian.
1259 defm : VecROLoadPat<ro64, v2i32, LDRDroW, LDRDroX>;
1260 defm : VecROLoadPat<ro64, v2f32, LDRDroW, LDRDroX>;
1261 defm : VecROLoadPat<ro64, v8i8, LDRDroW, LDRDroX>;
1262 defm : VecROLoadPat<ro64, v4i16, LDRDroW, LDRDroX>;
1263 defm : VecROLoadPat<ro64, v4f16, LDRDroW, LDRDroX>;
1266 defm : VecROLoadPat<ro64, v1i64, LDRDroW, LDRDroX>;
1267 defm : VecROLoadPat<ro64, v1f64, LDRDroW, LDRDroX>;
1269 // Match all load 128 bits width whose type is compatible with FPR128
1270 let Predicates = [IsLE] in {
1271 // We must do vector loads with LD1 in big-endian.
1272 defm : VecROLoadPat<ro128, v2i64, LDRQroW, LDRQroX>;
1273 defm : VecROLoadPat<ro128, v2f64, LDRQroW, LDRQroX>;
1274 defm : VecROLoadPat<ro128, v4i32, LDRQroW, LDRQroX>;
1275 defm : VecROLoadPat<ro128, v4f32, LDRQroW, LDRQroX>;
1276 defm : VecROLoadPat<ro128, v8i16, LDRQroW, LDRQroX>;
1277 defm : VecROLoadPat<ro128, v8f16, LDRQroW, LDRQroX>;
1278 defm : VecROLoadPat<ro128, v16i8, LDRQroW, LDRQroX>;
1280 } // AddedComplexity = 10
1283 multiclass ExtLoadTo64ROPat<ROAddrMode ro, SDPatternOperator loadop,
1284 Instruction INSTW, Instruction INSTX> {
1285 def : Pat<(i64 (loadop (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend))),
1286 (SUBREG_TO_REG (i64 0),
1287 (INSTW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend),
1290 def : Pat<(i64 (loadop (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend))),
1291 (SUBREG_TO_REG (i64 0),
1292 (INSTX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend),
1296 let AddedComplexity = 10 in {
1297 defm : ExtLoadTo64ROPat<ro8, zextloadi8, LDRBBroW, LDRBBroX>;
1298 defm : ExtLoadTo64ROPat<ro16, zextloadi16, LDRHHroW, LDRHHroX>;
1299 defm : ExtLoadTo64ROPat<ro32, zextloadi32, LDRWroW, LDRWroX>;
1301 // zextloadi1 -> zextloadi8
1302 defm : ExtLoadTo64ROPat<ro8, zextloadi1, LDRBBroW, LDRBBroX>;
1304 // extload -> zextload
1305 defm : ExtLoadTo64ROPat<ro8, extloadi8, LDRBBroW, LDRBBroX>;
1306 defm : ExtLoadTo64ROPat<ro16, extloadi16, LDRHHroW, LDRHHroX>;
1307 defm : ExtLoadTo64ROPat<ro32, extloadi32, LDRWroW, LDRWroX>;
1309 // extloadi1 -> zextloadi8
1310 defm : ExtLoadTo64ROPat<ro8, extloadi1, LDRBBroW, LDRBBroX>;
1315 multiclass ExtLoadTo32ROPat<ROAddrMode ro, SDPatternOperator loadop,
1316 Instruction INSTW, Instruction INSTX> {
1317 def : Pat<(i32 (loadop (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend))),
1318 (INSTW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
1320 def : Pat<(i32 (loadop (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend))),
1321 (INSTX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
1325 let AddedComplexity = 10 in {
1326 // extload -> zextload
1327 defm : ExtLoadTo32ROPat<ro8, extloadi8, LDRBBroW, LDRBBroX>;
1328 defm : ExtLoadTo32ROPat<ro16, extloadi16, LDRHHroW, LDRHHroX>;
1329 defm : ExtLoadTo32ROPat<ro32, extloadi32, LDRWroW, LDRWroX>;
1331 // zextloadi1 -> zextloadi8
1332 defm : ExtLoadTo32ROPat<ro8, zextloadi1, LDRBBroW, LDRBBroX>;
1336 // (unsigned immediate)
1338 defm LDRX : LoadUI<0b11, 0, 0b01, GPR64, uimm12s8, "ldr",
1340 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)))]>;
1341 defm LDRW : LoadUI<0b10, 0, 0b01, GPR32, uimm12s4, "ldr",
1343 (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset)))]>;
1344 defm LDRB : LoadUI<0b00, 1, 0b01, FPR8, uimm12s1, "ldr",
1346 (load (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset)))]>;
1347 defm LDRH : LoadUI<0b01, 1, 0b01, FPR16, uimm12s2, "ldr",
1348 [(set (f16 FPR16:$Rt),
1349 (load (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset)))]>;
1350 defm LDRS : LoadUI<0b10, 1, 0b01, FPR32, uimm12s4, "ldr",
1351 [(set (f32 FPR32:$Rt),
1352 (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset)))]>;
1353 defm LDRD : LoadUI<0b11, 1, 0b01, FPR64, uimm12s8, "ldr",
1354 [(set (f64 FPR64:$Rt),
1355 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)))]>;
1356 defm LDRQ : LoadUI<0b00, 1, 0b11, FPR128, uimm12s16, "ldr",
1357 [(set (f128 FPR128:$Rt),
1358 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)))]>;
1360 // For regular load, we do not have any alignment requirement.
1361 // Thus, it is safe to directly map the vector loads with interesting
1362 // addressing modes.
1363 // FIXME: We could do the same for bitconvert to floating point vectors.
1364 def : Pat <(v8i8 (scalar_to_vector (i32
1365 (extloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))))),
1366 (INSERT_SUBREG (v8i8 (IMPLICIT_DEF)),
1367 (LDRBui GPR64sp:$Rn, uimm12s1:$offset), bsub)>;
1368 def : Pat <(v16i8 (scalar_to_vector (i32
1369 (extloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))))),
1370 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
1371 (LDRBui GPR64sp:$Rn, uimm12s1:$offset), bsub)>;
1372 def : Pat <(v4i16 (scalar_to_vector (i32
1373 (extloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))))),
1374 (INSERT_SUBREG (v4i16 (IMPLICIT_DEF)),
1375 (LDRHui GPR64sp:$Rn, uimm12s2:$offset), hsub)>;
1376 def : Pat <(v8i16 (scalar_to_vector (i32
1377 (extloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))))),
1378 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
1379 (LDRHui GPR64sp:$Rn, uimm12s2:$offset), hsub)>;
1380 def : Pat <(v2i32 (scalar_to_vector (i32
1381 (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))))),
1382 (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)),
1383 (LDRSui GPR64sp:$Rn, uimm12s4:$offset), ssub)>;
1384 def : Pat <(v4i32 (scalar_to_vector (i32
1385 (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))))),
1386 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
1387 (LDRSui GPR64sp:$Rn, uimm12s4:$offset), ssub)>;
1388 def : Pat <(v1i64 (scalar_to_vector (i64
1389 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))))),
1390 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1391 def : Pat <(v2i64 (scalar_to_vector (i64
1392 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))))),
1393 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
1394 (LDRDui GPR64sp:$Rn, uimm12s8:$offset), dsub)>;
1396 // Match all load 64 bits width whose type is compatible with FPR64
1397 let Predicates = [IsLE] in {
1398 // We must use LD1 to perform vector loads in big-endian.
1399 def : Pat<(v2f32 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1400 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1401 def : Pat<(v8i8 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1402 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1403 def : Pat<(v4i16 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1404 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1405 def : Pat<(v2i32 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1406 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1407 def : Pat<(v4f16 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1408 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1410 def : Pat<(v1f64 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1411 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1412 def : Pat<(v1i64 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1413 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1415 // Match all load 128 bits width whose type is compatible with FPR128
1416 let Predicates = [IsLE] in {
1417 // We must use LD1 to perform vector loads in big-endian.
1418 def : Pat<(v4f32 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1419 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1420 def : Pat<(v2f64 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1421 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1422 def : Pat<(v16i8 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1423 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1424 def : Pat<(v8i16 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1425 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1426 def : Pat<(v4i32 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1427 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1428 def : Pat<(v2i64 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1429 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1430 def : Pat<(v8f16 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1431 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1433 def : Pat<(f128 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1434 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1436 defm LDRHH : LoadUI<0b01, 0, 0b01, GPR32, uimm12s2, "ldrh",
1438 (zextloadi16 (am_indexed16 GPR64sp:$Rn,
1439 uimm12s2:$offset)))]>;
1440 defm LDRBB : LoadUI<0b00, 0, 0b01, GPR32, uimm12s1, "ldrb",
1442 (zextloadi8 (am_indexed8 GPR64sp:$Rn,
1443 uimm12s1:$offset)))]>;
1445 def : Pat<(i64 (zextloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1446 (SUBREG_TO_REG (i64 0), (LDRBBui GPR64sp:$Rn, uimm12s1:$offset), sub_32)>;
1447 def : Pat<(i64 (zextloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))),
1448 (SUBREG_TO_REG (i64 0), (LDRHHui GPR64sp:$Rn, uimm12s2:$offset), sub_32)>;
1450 // zextloadi1 -> zextloadi8
1451 def : Pat<(i32 (zextloadi1 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1452 (LDRBBui GPR64sp:$Rn, uimm12s1:$offset)>;
1453 def : Pat<(i64 (zextloadi1 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1454 (SUBREG_TO_REG (i64 0), (LDRBBui GPR64sp:$Rn, uimm12s1:$offset), sub_32)>;
1456 // extload -> zextload
1457 def : Pat<(i32 (extloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))),
1458 (LDRHHui GPR64sp:$Rn, uimm12s2:$offset)>;
1459 def : Pat<(i32 (extloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1460 (LDRBBui GPR64sp:$Rn, uimm12s1:$offset)>;
1461 def : Pat<(i32 (extloadi1 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1462 (LDRBBui GPR64sp:$Rn, uimm12s1:$offset)>;
1463 def : Pat<(i64 (extloadi32 (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))),
1464 (SUBREG_TO_REG (i64 0), (LDRWui GPR64sp:$Rn, uimm12s4:$offset), sub_32)>;
1465 def : Pat<(i64 (extloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))),
1466 (SUBREG_TO_REG (i64 0), (LDRHHui GPR64sp:$Rn, uimm12s2:$offset), sub_32)>;
1467 def : Pat<(i64 (extloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1468 (SUBREG_TO_REG (i64 0), (LDRBBui GPR64sp:$Rn, uimm12s1:$offset), sub_32)>;
1469 def : Pat<(i64 (extloadi1 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1470 (SUBREG_TO_REG (i64 0), (LDRBBui GPR64sp:$Rn, uimm12s1:$offset), sub_32)>;
1472 // load sign-extended half-word
1473 defm LDRSHW : LoadUI<0b01, 0, 0b11, GPR32, uimm12s2, "ldrsh",
1475 (sextloadi16 (am_indexed16 GPR64sp:$Rn,
1476 uimm12s2:$offset)))]>;
1477 defm LDRSHX : LoadUI<0b01, 0, 0b10, GPR64, uimm12s2, "ldrsh",
1479 (sextloadi16 (am_indexed16 GPR64sp:$Rn,
1480 uimm12s2:$offset)))]>;
1482 // load sign-extended byte
1483 defm LDRSBW : LoadUI<0b00, 0, 0b11, GPR32, uimm12s1, "ldrsb",
1485 (sextloadi8 (am_indexed8 GPR64sp:$Rn,
1486 uimm12s1:$offset)))]>;
1487 defm LDRSBX : LoadUI<0b00, 0, 0b10, GPR64, uimm12s1, "ldrsb",
1489 (sextloadi8 (am_indexed8 GPR64sp:$Rn,
1490 uimm12s1:$offset)))]>;
1492 // load sign-extended word
1493 defm LDRSW : LoadUI<0b10, 0, 0b10, GPR64, uimm12s4, "ldrsw",
1495 (sextloadi32 (am_indexed32 GPR64sp:$Rn,
1496 uimm12s4:$offset)))]>;
1498 // load zero-extended word
1499 def : Pat<(i64 (zextloadi32 (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))),
1500 (SUBREG_TO_REG (i64 0), (LDRWui GPR64sp:$Rn, uimm12s4:$offset), sub_32)>;
1503 def PRFMui : PrefetchUI<0b11, 0, 0b10, "prfm",
1504 [(AArch64Prefetch imm:$Rt,
1505 (am_indexed64 GPR64sp:$Rn,
1506 uimm12s8:$offset))]>;
1508 def : InstAlias<"prfm $Rt, [$Rn]", (PRFMui prfop:$Rt, GPR64sp:$Rn, 0)>;
1512 def LDRWl : LoadLiteral<0b00, 0, GPR32, "ldr">;
1513 def LDRXl : LoadLiteral<0b01, 0, GPR64, "ldr">;
1514 def LDRSl : LoadLiteral<0b00, 1, FPR32, "ldr">;
1515 def LDRDl : LoadLiteral<0b01, 1, FPR64, "ldr">;
1516 def LDRQl : LoadLiteral<0b10, 1, FPR128, "ldr">;
1518 // load sign-extended word
1519 def LDRSWl : LoadLiteral<0b10, 0, GPR64, "ldrsw">;
1522 def PRFMl : PrefetchLiteral<0b11, 0, "prfm", []>;
1523 // [(AArch64Prefetch imm:$Rt, tglobaladdr:$label)]>;
1526 // (unscaled immediate)
1527 defm LDURX : LoadUnscaled<0b11, 0, 0b01, GPR64, "ldur",
1529 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset)))]>;
1530 defm LDURW : LoadUnscaled<0b10, 0, 0b01, GPR32, "ldur",
1532 (load (am_unscaled32 GPR64sp:$Rn, simm9:$offset)))]>;
1533 defm LDURB : LoadUnscaled<0b00, 1, 0b01, FPR8, "ldur",
1535 (load (am_unscaled8 GPR64sp:$Rn, simm9:$offset)))]>;
1536 defm LDURH : LoadUnscaled<0b01, 1, 0b01, FPR16, "ldur",
1538 (load (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
1539 defm LDURS : LoadUnscaled<0b10, 1, 0b01, FPR32, "ldur",
1540 [(set (f32 FPR32:$Rt),
1541 (load (am_unscaled32 GPR64sp:$Rn, simm9:$offset)))]>;
1542 defm LDURD : LoadUnscaled<0b11, 1, 0b01, FPR64, "ldur",
1543 [(set (f64 FPR64:$Rt),
1544 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset)))]>;
1545 defm LDURQ : LoadUnscaled<0b00, 1, 0b11, FPR128, "ldur",
1546 [(set (f128 FPR128:$Rt),
1547 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset)))]>;
1550 : LoadUnscaled<0b01, 0, 0b01, GPR32, "ldurh",
1552 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
1554 : LoadUnscaled<0b00, 0, 0b01, GPR32, "ldurb",
1556 (zextloadi8 (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
1558 // Match all load 64 bits width whose type is compatible with FPR64
1559 let Predicates = [IsLE] in {
1560 def : Pat<(v2f32 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1561 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1562 def : Pat<(v2i32 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1563 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1564 def : Pat<(v4i16 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1565 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1566 def : Pat<(v8i8 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1567 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1568 def : Pat<(v4f16 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1569 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1571 def : Pat<(v1f64 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1572 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1573 def : Pat<(v1i64 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1574 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1576 // Match all load 128 bits width whose type is compatible with FPR128
1577 let Predicates = [IsLE] in {
1578 def : Pat<(v2f64 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1579 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1580 def : Pat<(v2i64 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1581 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1582 def : Pat<(v4f32 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1583 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1584 def : Pat<(v4i32 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1585 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1586 def : Pat<(v8i16 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1587 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1588 def : Pat<(v16i8 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1589 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1590 def : Pat<(v8f16 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1591 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1595 def : Pat<(i32 (extloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
1596 (LDURHHi GPR64sp:$Rn, simm9:$offset)>;
1597 def : Pat<(i32 (extloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1598 (LDURBBi GPR64sp:$Rn, simm9:$offset)>;
1599 def : Pat<(i32 (extloadi1 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1600 (LDURBBi GPR64sp:$Rn, simm9:$offset)>;
1601 def : Pat<(i64 (extloadi32 (am_unscaled32 GPR64sp:$Rn, simm9:$offset))),
1602 (SUBREG_TO_REG (i64 0), (LDURWi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1603 def : Pat<(i64 (extloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
1604 (SUBREG_TO_REG (i64 0), (LDURHHi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1605 def : Pat<(i64 (extloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1606 (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1607 def : Pat<(i64 (extloadi1 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1608 (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1610 def : Pat<(i32 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
1611 (LDURHHi GPR64sp:$Rn, simm9:$offset)>;
1612 def : Pat<(i32 (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1613 (LDURBBi GPR64sp:$Rn, simm9:$offset)>;
1614 def : Pat<(i32 (zextloadi1 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1615 (LDURBBi GPR64sp:$Rn, simm9:$offset)>;
1616 def : Pat<(i64 (zextloadi32 (am_unscaled32 GPR64sp:$Rn, simm9:$offset))),
1617 (SUBREG_TO_REG (i64 0), (LDURWi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1618 def : Pat<(i64 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
1619 (SUBREG_TO_REG (i64 0), (LDURHHi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1620 def : Pat<(i64 (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1621 (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1622 def : Pat<(i64 (zextloadi1 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1623 (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1627 // LDR mnemonics fall back to LDUR for negative or unaligned offsets.
1629 // Define new assembler match classes as we want to only match these when
1630 // the don't otherwise match the scaled addressing mode for LDR/STR. Don't
1631 // associate a DiagnosticType either, as we want the diagnostic for the
1632 // canonical form (the scaled operand) to take precedence.
1633 class SImm9OffsetOperand<int Width> : AsmOperandClass {
1634 let Name = "SImm9OffsetFB" # Width;
1635 let PredicateMethod = "isSImm9OffsetFB<" # Width # ">";
1636 let RenderMethod = "addImmOperands";
1639 def SImm9OffsetFB8Operand : SImm9OffsetOperand<8>;
1640 def SImm9OffsetFB16Operand : SImm9OffsetOperand<16>;
1641 def SImm9OffsetFB32Operand : SImm9OffsetOperand<32>;
1642 def SImm9OffsetFB64Operand : SImm9OffsetOperand<64>;
1643 def SImm9OffsetFB128Operand : SImm9OffsetOperand<128>;
1645 def simm9_offset_fb8 : Operand<i64> {
1646 let ParserMatchClass = SImm9OffsetFB8Operand;
1648 def simm9_offset_fb16 : Operand<i64> {
1649 let ParserMatchClass = SImm9OffsetFB16Operand;
1651 def simm9_offset_fb32 : Operand<i64> {
1652 let ParserMatchClass = SImm9OffsetFB32Operand;
1654 def simm9_offset_fb64 : Operand<i64> {
1655 let ParserMatchClass = SImm9OffsetFB64Operand;
1657 def simm9_offset_fb128 : Operand<i64> {
1658 let ParserMatchClass = SImm9OffsetFB128Operand;
1661 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1662 (LDURXi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb64:$offset), 0>;
1663 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1664 (LDURWi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
1665 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1666 (LDURBi FPR8:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
1667 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1668 (LDURHi FPR16:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
1669 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1670 (LDURSi FPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
1671 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1672 (LDURDi FPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb64:$offset), 0>;
1673 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1674 (LDURQi FPR128:$Rt, GPR64sp:$Rn, simm9_offset_fb128:$offset), 0>;
1677 def : Pat<(i64 (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1678 (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1679 def : Pat<(i64 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
1680 (SUBREG_TO_REG (i64 0), (LDURHHi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1682 // load sign-extended half-word
1684 : LoadUnscaled<0b01, 0, 0b11, GPR32, "ldursh",
1686 (sextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
1688 : LoadUnscaled<0b01, 0, 0b10, GPR64, "ldursh",
1690 (sextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
1692 // load sign-extended byte
1694 : LoadUnscaled<0b00, 0, 0b11, GPR32, "ldursb",
1696 (sextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset)))]>;
1698 : LoadUnscaled<0b00, 0, 0b10, GPR64, "ldursb",
1700 (sextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset)))]>;
1702 // load sign-extended word
1704 : LoadUnscaled<0b10, 0, 0b10, GPR64, "ldursw",
1706 (sextloadi32 (am_unscaled32 GPR64sp:$Rn, simm9:$offset)))]>;
1708 // zero and sign extending aliases from generic LDR* mnemonics to LDUR*.
1709 def : InstAlias<"ldrb $Rt, [$Rn, $offset]",
1710 (LDURBBi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
1711 def : InstAlias<"ldrh $Rt, [$Rn, $offset]",
1712 (LDURHHi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
1713 def : InstAlias<"ldrsb $Rt, [$Rn, $offset]",
1714 (LDURSBWi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
1715 def : InstAlias<"ldrsb $Rt, [$Rn, $offset]",
1716 (LDURSBXi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
1717 def : InstAlias<"ldrsh $Rt, [$Rn, $offset]",
1718 (LDURSHWi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
1719 def : InstAlias<"ldrsh $Rt, [$Rn, $offset]",
1720 (LDURSHXi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
1721 def : InstAlias<"ldrsw $Rt, [$Rn, $offset]",
1722 (LDURSWi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
1725 defm PRFUM : PrefetchUnscaled<0b11, 0, 0b10, "prfum",
1726 [(AArch64Prefetch imm:$Rt,
1727 (am_unscaled64 GPR64sp:$Rn, simm9:$offset))]>;
1730 // (unscaled immediate, unprivileged)
1731 defm LDTRX : LoadUnprivileged<0b11, 0, 0b01, GPR64, "ldtr">;
1732 defm LDTRW : LoadUnprivileged<0b10, 0, 0b01, GPR32, "ldtr">;
1734 defm LDTRH : LoadUnprivileged<0b01, 0, 0b01, GPR32, "ldtrh">;
1735 defm LDTRB : LoadUnprivileged<0b00, 0, 0b01, GPR32, "ldtrb">;
1737 // load sign-extended half-word
1738 defm LDTRSHW : LoadUnprivileged<0b01, 0, 0b11, GPR32, "ldtrsh">;
1739 defm LDTRSHX : LoadUnprivileged<0b01, 0, 0b10, GPR64, "ldtrsh">;
1741 // load sign-extended byte
1742 defm LDTRSBW : LoadUnprivileged<0b00, 0, 0b11, GPR32, "ldtrsb">;
1743 defm LDTRSBX : LoadUnprivileged<0b00, 0, 0b10, GPR64, "ldtrsb">;
1745 // load sign-extended word
1746 defm LDTRSW : LoadUnprivileged<0b10, 0, 0b10, GPR64, "ldtrsw">;
1749 // (immediate pre-indexed)
1750 def LDRWpre : LoadPreIdx<0b10, 0, 0b01, GPR32, "ldr">;
1751 def LDRXpre : LoadPreIdx<0b11, 0, 0b01, GPR64, "ldr">;
1752 def LDRBpre : LoadPreIdx<0b00, 1, 0b01, FPR8, "ldr">;
1753 def LDRHpre : LoadPreIdx<0b01, 1, 0b01, FPR16, "ldr">;
1754 def LDRSpre : LoadPreIdx<0b10, 1, 0b01, FPR32, "ldr">;
1755 def LDRDpre : LoadPreIdx<0b11, 1, 0b01, FPR64, "ldr">;
1756 def LDRQpre : LoadPreIdx<0b00, 1, 0b11, FPR128, "ldr">;
1758 // load sign-extended half-word
1759 def LDRSHWpre : LoadPreIdx<0b01, 0, 0b11, GPR32, "ldrsh">;
1760 def LDRSHXpre : LoadPreIdx<0b01, 0, 0b10, GPR64, "ldrsh">;
1762 // load sign-extended byte
1763 def LDRSBWpre : LoadPreIdx<0b00, 0, 0b11, GPR32, "ldrsb">;
1764 def LDRSBXpre : LoadPreIdx<0b00, 0, 0b10, GPR64, "ldrsb">;
1766 // load zero-extended byte
1767 def LDRBBpre : LoadPreIdx<0b00, 0, 0b01, GPR32, "ldrb">;
1768 def LDRHHpre : LoadPreIdx<0b01, 0, 0b01, GPR32, "ldrh">;
1770 // load sign-extended word
1771 def LDRSWpre : LoadPreIdx<0b10, 0, 0b10, GPR64, "ldrsw">;
1774 // (immediate post-indexed)
1775 def LDRWpost : LoadPostIdx<0b10, 0, 0b01, GPR32, "ldr">;
1776 def LDRXpost : LoadPostIdx<0b11, 0, 0b01, GPR64, "ldr">;
1777 def LDRBpost : LoadPostIdx<0b00, 1, 0b01, FPR8, "ldr">;
1778 def LDRHpost : LoadPostIdx<0b01, 1, 0b01, FPR16, "ldr">;
1779 def LDRSpost : LoadPostIdx<0b10, 1, 0b01, FPR32, "ldr">;
1780 def LDRDpost : LoadPostIdx<0b11, 1, 0b01, FPR64, "ldr">;
1781 def LDRQpost : LoadPostIdx<0b00, 1, 0b11, FPR128, "ldr">;
1783 // load sign-extended half-word
1784 def LDRSHWpost : LoadPostIdx<0b01, 0, 0b11, GPR32, "ldrsh">;
1785 def LDRSHXpost : LoadPostIdx<0b01, 0, 0b10, GPR64, "ldrsh">;
1787 // load sign-extended byte
1788 def LDRSBWpost : LoadPostIdx<0b00, 0, 0b11, GPR32, "ldrsb">;
1789 def LDRSBXpost : LoadPostIdx<0b00, 0, 0b10, GPR64, "ldrsb">;
1791 // load zero-extended byte
1792 def LDRBBpost : LoadPostIdx<0b00, 0, 0b01, GPR32, "ldrb">;
1793 def LDRHHpost : LoadPostIdx<0b01, 0, 0b01, GPR32, "ldrh">;
1795 // load sign-extended word
1796 def LDRSWpost : LoadPostIdx<0b10, 0, 0b10, GPR64, "ldrsw">;
1798 //===----------------------------------------------------------------------===//
1799 // Store instructions.
1800 //===----------------------------------------------------------------------===//
1802 // Pair (indexed, offset)
1803 // FIXME: Use dedicated range-checked addressing mode operand here.
1804 defm STPW : StorePairOffset<0b00, 0, GPR32, simm7s4, "stp">;
1805 defm STPX : StorePairOffset<0b10, 0, GPR64, simm7s8, "stp">;
1806 defm STPS : StorePairOffset<0b00, 1, FPR32, simm7s4, "stp">;
1807 defm STPD : StorePairOffset<0b01, 1, FPR64, simm7s8, "stp">;
1808 defm STPQ : StorePairOffset<0b10, 1, FPR128, simm7s16, "stp">;
1810 // Pair (pre-indexed)
1811 def STPWpre : StorePairPreIdx<0b00, 0, GPR32, simm7s4, "stp">;
1812 def STPXpre : StorePairPreIdx<0b10, 0, GPR64, simm7s8, "stp">;
1813 def STPSpre : StorePairPreIdx<0b00, 1, FPR32, simm7s4, "stp">;
1814 def STPDpre : StorePairPreIdx<0b01, 1, FPR64, simm7s8, "stp">;
1815 def STPQpre : StorePairPreIdx<0b10, 1, FPR128, simm7s16, "stp">;
1817 // Pair (pre-indexed)
1818 def STPWpost : StorePairPostIdx<0b00, 0, GPR32, simm7s4, "stp">;
1819 def STPXpost : StorePairPostIdx<0b10, 0, GPR64, simm7s8, "stp">;
1820 def STPSpost : StorePairPostIdx<0b00, 1, FPR32, simm7s4, "stp">;
1821 def STPDpost : StorePairPostIdx<0b01, 1, FPR64, simm7s8, "stp">;
1822 def STPQpost : StorePairPostIdx<0b10, 1, FPR128, simm7s16, "stp">;
1824 // Pair (no allocate)
1825 defm STNPW : StorePairNoAlloc<0b00, 0, GPR32, simm7s4, "stnp">;
1826 defm STNPX : StorePairNoAlloc<0b10, 0, GPR64, simm7s8, "stnp">;
1827 defm STNPS : StorePairNoAlloc<0b00, 1, FPR32, simm7s4, "stnp">;
1828 defm STNPD : StorePairNoAlloc<0b01, 1, FPR64, simm7s8, "stnp">;
1829 defm STNPQ : StorePairNoAlloc<0b10, 1, FPR128, simm7s16, "stnp">;
1832 // (Register offset)
1835 defm STRBB : Store8RO< 0b00, 0, 0b00, GPR32, "strb", i32, truncstorei8>;
1836 defm STRHH : Store16RO<0b01, 0, 0b00, GPR32, "strh", i32, truncstorei16>;
1837 defm STRW : Store32RO<0b10, 0, 0b00, GPR32, "str", i32, store>;
1838 defm STRX : Store64RO<0b11, 0, 0b00, GPR64, "str", i64, store>;
1842 defm STRB : Store8RO< 0b00, 1, 0b00, FPR8, "str", untyped, store>;
1843 defm STRH : Store16RO<0b01, 1, 0b00, FPR16, "str", f16, store>;
1844 defm STRS : Store32RO<0b10, 1, 0b00, FPR32, "str", f32, store>;
1845 defm STRD : Store64RO<0b11, 1, 0b00, FPR64, "str", f64, store>;
1846 defm STRQ : Store128RO<0b00, 1, 0b10, FPR128, "str", f128, store>;
1848 multiclass TruncStoreFrom64ROPat<ROAddrMode ro, SDPatternOperator storeop,
1849 Instruction STRW, Instruction STRX> {
1851 def : Pat<(storeop GPR64:$Rt,
1852 (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)),
1853 (STRW (EXTRACT_SUBREG GPR64:$Rt, sub_32),
1854 GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
1856 def : Pat<(storeop GPR64:$Rt,
1857 (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)),
1858 (STRX (EXTRACT_SUBREG GPR64:$Rt, sub_32),
1859 GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
1862 let AddedComplexity = 10 in {
1864 defm : TruncStoreFrom64ROPat<ro8, truncstorei8, STRBBroW, STRBBroX>;
1865 defm : TruncStoreFrom64ROPat<ro16, truncstorei16, STRHHroW, STRHHroX>;
1866 defm : TruncStoreFrom64ROPat<ro32, truncstorei32, STRWroW, STRWroX>;
1869 multiclass VecROStorePat<ROAddrMode ro, ValueType VecTy, RegisterClass FPR,
1870 Instruction STRW, Instruction STRX> {
1871 def : Pat<(store (VecTy FPR:$Rt),
1872 (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)),
1873 (STRW FPR:$Rt, GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
1875 def : Pat<(store (VecTy FPR:$Rt),
1876 (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)),
1877 (STRX FPR:$Rt, GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
1880 let AddedComplexity = 10 in {
1881 // Match all store 64 bits width whose type is compatible with FPR64
1882 let Predicates = [IsLE] in {
1883 // We must use ST1 to store vectors in big-endian.
1884 defm : VecROStorePat<ro64, v2i32, FPR64, STRDroW, STRDroX>;
1885 defm : VecROStorePat<ro64, v2f32, FPR64, STRDroW, STRDroX>;
1886 defm : VecROStorePat<ro64, v4i16, FPR64, STRDroW, STRDroX>;
1887 defm : VecROStorePat<ro64, v8i8, FPR64, STRDroW, STRDroX>;
1888 defm : VecROStorePat<ro64, v4f16, FPR64, STRDroW, STRDroX>;
1891 defm : VecROStorePat<ro64, v1i64, FPR64, STRDroW, STRDroX>;
1892 defm : VecROStorePat<ro64, v1f64, FPR64, STRDroW, STRDroX>;
1894 // Match all store 128 bits width whose type is compatible with FPR128
1895 let Predicates = [IsLE] in {
1896 // We must use ST1 to store vectors in big-endian.
1897 defm : VecROStorePat<ro128, v2i64, FPR128, STRQroW, STRQroX>;
1898 defm : VecROStorePat<ro128, v2f64, FPR128, STRQroW, STRQroX>;
1899 defm : VecROStorePat<ro128, v4i32, FPR128, STRQroW, STRQroX>;
1900 defm : VecROStorePat<ro128, v4f32, FPR128, STRQroW, STRQroX>;
1901 defm : VecROStorePat<ro128, v8i16, FPR128, STRQroW, STRQroX>;
1902 defm : VecROStorePat<ro128, v16i8, FPR128, STRQroW, STRQroX>;
1903 defm : VecROStorePat<ro128, v8f16, FPR128, STRQroW, STRQroX>;
1905 } // AddedComplexity = 10
1907 // Match stores from lane 0 to the appropriate subreg's store.
1908 multiclass VecROStoreLane0Pat<ROAddrMode ro, SDPatternOperator storeop,
1909 ValueType VecTy, ValueType STy,
1910 SubRegIndex SubRegIdx,
1911 Instruction STRW, Instruction STRX> {
1913 def : Pat<(storeop (STy (vector_extract (VecTy VecListOne128:$Vt), 0)),
1914 (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)),
1915 (STRW (EXTRACT_SUBREG VecListOne128:$Vt, SubRegIdx),
1916 GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
1918 def : Pat<(storeop (STy (vector_extract (VecTy VecListOne128:$Vt), 0)),
1919 (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)),
1920 (STRX (EXTRACT_SUBREG VecListOne128:$Vt, SubRegIdx),
1921 GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
1924 let AddedComplexity = 19 in {
1925 defm : VecROStoreLane0Pat<ro16, truncstorei16, v8i16, i32, hsub, STRHroW, STRHroX>;
1926 defm : VecROStoreLane0Pat<ro16, store , v8i16, i16, hsub, STRHroW, STRHroX>;
1927 defm : VecROStoreLane0Pat<ro32, truncstorei32, v4i32, i32, ssub, STRSroW, STRSroX>;
1928 defm : VecROStoreLane0Pat<ro32, store , v4i32, i32, ssub, STRSroW, STRSroX>;
1929 defm : VecROStoreLane0Pat<ro32, store , v4f32, f32, ssub, STRSroW, STRSroX>;
1930 defm : VecROStoreLane0Pat<ro64, store , v2i64, i64, dsub, STRDroW, STRDroX>;
1931 defm : VecROStoreLane0Pat<ro64, store , v2f64, f64, dsub, STRDroW, STRDroX>;
1935 // (unsigned immediate)
1936 defm STRX : StoreUI<0b11, 0, 0b00, GPR64, uimm12s8, "str",
1938 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))]>;
1939 defm STRW : StoreUI<0b10, 0, 0b00, GPR32, uimm12s4, "str",
1941 (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))]>;
1942 defm STRB : StoreUI<0b00, 1, 0b00, FPR8, uimm12s1, "str",
1944 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))]>;
1945 defm STRH : StoreUI<0b01, 1, 0b00, FPR16, uimm12s2, "str",
1946 [(store (f16 FPR16:$Rt),
1947 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))]>;
1948 defm STRS : StoreUI<0b10, 1, 0b00, FPR32, uimm12s4, "str",
1949 [(store (f32 FPR32:$Rt),
1950 (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))]>;
1951 defm STRD : StoreUI<0b11, 1, 0b00, FPR64, uimm12s8, "str",
1952 [(store (f64 FPR64:$Rt),
1953 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))]>;
1954 defm STRQ : StoreUI<0b00, 1, 0b10, FPR128, uimm12s16, "str", []>;
1956 defm STRHH : StoreUI<0b01, 0, 0b00, GPR32, uimm12s2, "strh",
1957 [(truncstorei16 GPR32:$Rt,
1958 (am_indexed16 GPR64sp:$Rn,
1959 uimm12s2:$offset))]>;
1960 defm STRBB : StoreUI<0b00, 0, 0b00, GPR32, uimm12s1, "strb",
1961 [(truncstorei8 GPR32:$Rt,
1962 (am_indexed8 GPR64sp:$Rn,
1963 uimm12s1:$offset))]>;
1965 // Match all store 64 bits width whose type is compatible with FPR64
1966 let AddedComplexity = 10 in {
1967 let Predicates = [IsLE] in {
1968 // We must use ST1 to store vectors in big-endian.
1969 def : Pat<(store (v2f32 FPR64:$Rt),
1970 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
1971 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
1972 def : Pat<(store (v8i8 FPR64:$Rt),
1973 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
1974 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
1975 def : Pat<(store (v4i16 FPR64:$Rt),
1976 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
1977 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
1978 def : Pat<(store (v2i32 FPR64:$Rt),
1979 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
1980 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
1981 def : Pat<(store (v4f16 FPR64:$Rt),
1982 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
1983 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
1985 def : Pat<(store (v1f64 FPR64:$Rt),
1986 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
1987 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
1988 def : Pat<(store (v1i64 FPR64:$Rt),
1989 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
1990 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
1992 // Match all store 128 bits width whose type is compatible with FPR128
1993 let Predicates = [IsLE] in {
1994 // We must use ST1 to store vectors in big-endian.
1995 def : Pat<(store (v4f32 FPR128:$Rt),
1996 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
1997 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
1998 def : Pat<(store (v2f64 FPR128:$Rt),
1999 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
2000 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
2001 def : Pat<(store (v16i8 FPR128:$Rt),
2002 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
2003 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
2004 def : Pat<(store (v8i16 FPR128:$Rt),
2005 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
2006 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
2007 def : Pat<(store (v4i32 FPR128:$Rt),
2008 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
2009 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
2010 def : Pat<(store (v2i64 FPR128:$Rt),
2011 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
2012 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
2013 def : Pat<(store (v8f16 FPR128:$Rt),
2014 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
2015 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
2017 def : Pat<(store (f128 FPR128:$Rt),
2018 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
2019 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
2022 def : Pat<(truncstorei32 GPR64:$Rt,
2023 (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset)),
2024 (STRWui (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, uimm12s4:$offset)>;
2025 def : Pat<(truncstorei16 GPR64:$Rt,
2026 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset)),
2027 (STRHHui (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, uimm12s2:$offset)>;
2028 def : Pat<(truncstorei8 GPR64:$Rt, (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset)),
2029 (STRBBui (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, uimm12s1:$offset)>;
2031 } // AddedComplexity = 10
2034 // (unscaled immediate)
2035 defm STURX : StoreUnscaled<0b11, 0, 0b00, GPR64, "stur",
2037 (am_unscaled64 GPR64sp:$Rn, simm9:$offset))]>;
2038 defm STURW : StoreUnscaled<0b10, 0, 0b00, GPR32, "stur",
2040 (am_unscaled32 GPR64sp:$Rn, simm9:$offset))]>;
2041 defm STURB : StoreUnscaled<0b00, 1, 0b00, FPR8, "stur",
2043 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))]>;
2044 defm STURH : StoreUnscaled<0b01, 1, 0b00, FPR16, "stur",
2045 [(store (f16 FPR16:$Rt),
2046 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))]>;
2047 defm STURS : StoreUnscaled<0b10, 1, 0b00, FPR32, "stur",
2048 [(store (f32 FPR32:$Rt),
2049 (am_unscaled32 GPR64sp:$Rn, simm9:$offset))]>;
2050 defm STURD : StoreUnscaled<0b11, 1, 0b00, FPR64, "stur",
2051 [(store (f64 FPR64:$Rt),
2052 (am_unscaled64 GPR64sp:$Rn, simm9:$offset))]>;
2053 defm STURQ : StoreUnscaled<0b00, 1, 0b10, FPR128, "stur",
2054 [(store (f128 FPR128:$Rt),
2055 (am_unscaled128 GPR64sp:$Rn, simm9:$offset))]>;
2056 defm STURHH : StoreUnscaled<0b01, 0, 0b00, GPR32, "sturh",
2057 [(truncstorei16 GPR32:$Rt,
2058 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))]>;
2059 defm STURBB : StoreUnscaled<0b00, 0, 0b00, GPR32, "sturb",
2060 [(truncstorei8 GPR32:$Rt,
2061 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))]>;
2063 // Match all store 64 bits width whose type is compatible with FPR64
2064 let Predicates = [IsLE] in {
2065 // We must use ST1 to store vectors in big-endian.
2066 def : Pat<(store (v2f32 FPR64:$Rt),
2067 (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2068 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2069 def : Pat<(store (v8i8 FPR64:$Rt),
2070 (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2071 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2072 def : Pat<(store (v4i16 FPR64:$Rt),
2073 (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2074 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2075 def : Pat<(store (v2i32 FPR64:$Rt),
2076 (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2077 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2078 def : Pat<(store (v4f16 FPR64:$Rt),
2079 (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2080 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2082 def : Pat<(store (v1f64 FPR64:$Rt), (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2083 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2084 def : Pat<(store (v1i64 FPR64:$Rt), (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2085 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2087 // Match all store 128 bits width whose type is compatible with FPR128
2088 let Predicates = [IsLE] in {
2089 // We must use ST1 to store vectors in big-endian.
2090 def : Pat<(store (v4f32 FPR128:$Rt),
2091 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2092 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2093 def : Pat<(store (v2f64 FPR128:$Rt),
2094 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2095 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2096 def : Pat<(store (v16i8 FPR128:$Rt),
2097 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2098 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2099 def : Pat<(store (v8i16 FPR128:$Rt),
2100 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2101 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2102 def : Pat<(store (v4i32 FPR128:$Rt),
2103 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2104 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2105 def : Pat<(store (v2i64 FPR128:$Rt),
2106 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2107 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2108 def : Pat<(store (v2f64 FPR128:$Rt),
2109 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2110 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2111 def : Pat<(store (v8f16 FPR128:$Rt),
2112 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2113 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2116 // unscaled i64 truncating stores
2117 def : Pat<(truncstorei32 GPR64:$Rt, (am_unscaled32 GPR64sp:$Rn, simm9:$offset)),
2118 (STURWi (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, simm9:$offset)>;
2119 def : Pat<(truncstorei16 GPR64:$Rt, (am_unscaled16 GPR64sp:$Rn, simm9:$offset)),
2120 (STURHHi (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, simm9:$offset)>;
2121 def : Pat<(truncstorei8 GPR64:$Rt, (am_unscaled8 GPR64sp:$Rn, simm9:$offset)),
2122 (STURBBi (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, simm9:$offset)>;
2125 // STR mnemonics fall back to STUR for negative or unaligned offsets.
2126 def : InstAlias<"str $Rt, [$Rn, $offset]",
2127 (STURXi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb64:$offset), 0>;
2128 def : InstAlias<"str $Rt, [$Rn, $offset]",
2129 (STURWi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
2130 def : InstAlias<"str $Rt, [$Rn, $offset]",
2131 (STURBi FPR8:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
2132 def : InstAlias<"str $Rt, [$Rn, $offset]",
2133 (STURHi FPR16:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
2134 def : InstAlias<"str $Rt, [$Rn, $offset]",
2135 (STURSi FPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
2136 def : InstAlias<"str $Rt, [$Rn, $offset]",
2137 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb64:$offset), 0>;
2138 def : InstAlias<"str $Rt, [$Rn, $offset]",
2139 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9_offset_fb128:$offset), 0>;
2141 def : InstAlias<"strb $Rt, [$Rn, $offset]",
2142 (STURBBi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
2143 def : InstAlias<"strh $Rt, [$Rn, $offset]",
2144 (STURHHi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
2147 // (unscaled immediate, unprivileged)
2148 defm STTRW : StoreUnprivileged<0b10, 0, 0b00, GPR32, "sttr">;
2149 defm STTRX : StoreUnprivileged<0b11, 0, 0b00, GPR64, "sttr">;
2151 defm STTRH : StoreUnprivileged<0b01, 0, 0b00, GPR32, "sttrh">;
2152 defm STTRB : StoreUnprivileged<0b00, 0, 0b00, GPR32, "sttrb">;
2155 // (immediate pre-indexed)
2156 def STRWpre : StorePreIdx<0b10, 0, 0b00, GPR32, "str", pre_store, i32>;
2157 def STRXpre : StorePreIdx<0b11, 0, 0b00, GPR64, "str", pre_store, i64>;
2158 def STRBpre : StorePreIdx<0b00, 1, 0b00, FPR8, "str", pre_store, untyped>;
2159 def STRHpre : StorePreIdx<0b01, 1, 0b00, FPR16, "str", pre_store, f16>;
2160 def STRSpre : StorePreIdx<0b10, 1, 0b00, FPR32, "str", pre_store, f32>;
2161 def STRDpre : StorePreIdx<0b11, 1, 0b00, FPR64, "str", pre_store, f64>;
2162 def STRQpre : StorePreIdx<0b00, 1, 0b10, FPR128, "str", pre_store, f128>;
2164 def STRBBpre : StorePreIdx<0b00, 0, 0b00, GPR32, "strb", pre_truncsti8, i32>;
2165 def STRHHpre : StorePreIdx<0b01, 0, 0b00, GPR32, "strh", pre_truncsti16, i32>;
2168 def : Pat<(pre_truncsti32 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
2169 (STRWpre (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
2171 def : Pat<(pre_truncsti16 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
2172 (STRHHpre (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
2174 def : Pat<(pre_truncsti8 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
2175 (STRBBpre (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
2178 def : Pat<(pre_store (v8i8 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2179 (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2180 def : Pat<(pre_store (v4i16 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2181 (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2182 def : Pat<(pre_store (v2i32 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2183 (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2184 def : Pat<(pre_store (v2f32 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2185 (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2186 def : Pat<(pre_store (v1i64 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2187 (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2188 def : Pat<(pre_store (v1f64 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2189 (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2190 def : Pat<(pre_store (v4f16 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2191 (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2193 def : Pat<(pre_store (v16i8 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2194 (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2195 def : Pat<(pre_store (v8i16 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2196 (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2197 def : Pat<(pre_store (v4i32 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2198 (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2199 def : Pat<(pre_store (v4f32 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2200 (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2201 def : Pat<(pre_store (v2i64 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2202 (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2203 def : Pat<(pre_store (v2f64 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2204 (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2205 def : Pat<(pre_store (v8f16 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2206 (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2209 // (immediate post-indexed)
2210 def STRWpost : StorePostIdx<0b10, 0, 0b00, GPR32, "str", post_store, i32>;
2211 def STRXpost : StorePostIdx<0b11, 0, 0b00, GPR64, "str", post_store, i64>;
2212 def STRBpost : StorePostIdx<0b00, 1, 0b00, FPR8, "str", post_store, untyped>;
2213 def STRHpost : StorePostIdx<0b01, 1, 0b00, FPR16, "str", post_store, f16>;
2214 def STRSpost : StorePostIdx<0b10, 1, 0b00, FPR32, "str", post_store, f32>;
2215 def STRDpost : StorePostIdx<0b11, 1, 0b00, FPR64, "str", post_store, f64>;
2216 def STRQpost : StorePostIdx<0b00, 1, 0b10, FPR128, "str", post_store, f128>;
2218 def STRBBpost : StorePostIdx<0b00, 0, 0b00, GPR32, "strb", post_truncsti8, i32>;
2219 def STRHHpost : StorePostIdx<0b01, 0, 0b00, GPR32, "strh", post_truncsti16, i32>;
2222 def : Pat<(post_truncsti32 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
2223 (STRWpost (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
2225 def : Pat<(post_truncsti16 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
2226 (STRHHpost (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
2228 def : Pat<(post_truncsti8 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
2229 (STRBBpost (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
2232 def : Pat<(post_store (v8i8 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2233 (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2234 def : Pat<(post_store (v4i16 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2235 (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2236 def : Pat<(post_store (v2i32 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2237 (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2238 def : Pat<(post_store (v2f32 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2239 (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2240 def : Pat<(post_store (v1i64 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2241 (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2242 def : Pat<(post_store (v1f64 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2243 (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2244 def : Pat<(post_store (v4f16 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2245 (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2247 def : Pat<(post_store (v16i8 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2248 (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2249 def : Pat<(post_store (v8i16 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2250 (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2251 def : Pat<(post_store (v4i32 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2252 (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2253 def : Pat<(post_store (v4f32 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2254 (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2255 def : Pat<(post_store (v2i64 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2256 (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2257 def : Pat<(post_store (v2f64 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2258 (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2259 def : Pat<(post_store (v8f16 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2260 (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2262 //===----------------------------------------------------------------------===//
2263 // Load/store exclusive instructions.
2264 //===----------------------------------------------------------------------===//
2266 def LDARW : LoadAcquire <0b10, 1, 1, 0, 1, GPR32, "ldar">;
2267 def LDARX : LoadAcquire <0b11, 1, 1, 0, 1, GPR64, "ldar">;
2268 def LDARB : LoadAcquire <0b00, 1, 1, 0, 1, GPR32, "ldarb">;
2269 def LDARH : LoadAcquire <0b01, 1, 1, 0, 1, GPR32, "ldarh">;
2271 def LDAXRW : LoadExclusive <0b10, 0, 1, 0, 1, GPR32, "ldaxr">;
2272 def LDAXRX : LoadExclusive <0b11, 0, 1, 0, 1, GPR64, "ldaxr">;
2273 def LDAXRB : LoadExclusive <0b00, 0, 1, 0, 1, GPR32, "ldaxrb">;
2274 def LDAXRH : LoadExclusive <0b01, 0, 1, 0, 1, GPR32, "ldaxrh">;
2276 def LDXRW : LoadExclusive <0b10, 0, 1, 0, 0, GPR32, "ldxr">;
2277 def LDXRX : LoadExclusive <0b11, 0, 1, 0, 0, GPR64, "ldxr">;
2278 def LDXRB : LoadExclusive <0b00, 0, 1, 0, 0, GPR32, "ldxrb">;
2279 def LDXRH : LoadExclusive <0b01, 0, 1, 0, 0, GPR32, "ldxrh">;
2281 def STLRW : StoreRelease <0b10, 1, 0, 0, 1, GPR32, "stlr">;
2282 def STLRX : StoreRelease <0b11, 1, 0, 0, 1, GPR64, "stlr">;
2283 def STLRB : StoreRelease <0b00, 1, 0, 0, 1, GPR32, "stlrb">;
2284 def STLRH : StoreRelease <0b01, 1, 0, 0, 1, GPR32, "stlrh">;
2286 def STLXRW : StoreExclusive<0b10, 0, 0, 0, 1, GPR32, "stlxr">;
2287 def STLXRX : StoreExclusive<0b11, 0, 0, 0, 1, GPR64, "stlxr">;
2288 def STLXRB : StoreExclusive<0b00, 0, 0, 0, 1, GPR32, "stlxrb">;
2289 def STLXRH : StoreExclusive<0b01, 0, 0, 0, 1, GPR32, "stlxrh">;
2291 def STXRW : StoreExclusive<0b10, 0, 0, 0, 0, GPR32, "stxr">;
2292 def STXRX : StoreExclusive<0b11, 0, 0, 0, 0, GPR64, "stxr">;
2293 def STXRB : StoreExclusive<0b00, 0, 0, 0, 0, GPR32, "stxrb">;
2294 def STXRH : StoreExclusive<0b01, 0, 0, 0, 0, GPR32, "stxrh">;
2296 def LDAXPW : LoadExclusivePair<0b10, 0, 1, 1, 1, GPR32, "ldaxp">;
2297 def LDAXPX : LoadExclusivePair<0b11, 0, 1, 1, 1, GPR64, "ldaxp">;
2299 def LDXPW : LoadExclusivePair<0b10, 0, 1, 1, 0, GPR32, "ldxp">;
2300 def LDXPX : LoadExclusivePair<0b11, 0, 1, 1, 0, GPR64, "ldxp">;
2302 def STLXPW : StoreExclusivePair<0b10, 0, 0, 1, 1, GPR32, "stlxp">;
2303 def STLXPX : StoreExclusivePair<0b11, 0, 0, 1, 1, GPR64, "stlxp">;
2305 def STXPW : StoreExclusivePair<0b10, 0, 0, 1, 0, GPR32, "stxp">;
2306 def STXPX : StoreExclusivePair<0b11, 0, 0, 1, 0, GPR64, "stxp">;
2308 //===----------------------------------------------------------------------===//
2309 // Scaled floating point to integer conversion instructions.
2310 //===----------------------------------------------------------------------===//
2312 defm FCVTAS : FPToIntegerUnscaled<0b00, 0b100, "fcvtas", int_aarch64_neon_fcvtas>;
2313 defm FCVTAU : FPToIntegerUnscaled<0b00, 0b101, "fcvtau", int_aarch64_neon_fcvtau>;
2314 defm FCVTMS : FPToIntegerUnscaled<0b10, 0b000, "fcvtms", int_aarch64_neon_fcvtms>;
2315 defm FCVTMU : FPToIntegerUnscaled<0b10, 0b001, "fcvtmu", int_aarch64_neon_fcvtmu>;
2316 defm FCVTNS : FPToIntegerUnscaled<0b00, 0b000, "fcvtns", int_aarch64_neon_fcvtns>;
2317 defm FCVTNU : FPToIntegerUnscaled<0b00, 0b001, "fcvtnu", int_aarch64_neon_fcvtnu>;
2318 defm FCVTPS : FPToIntegerUnscaled<0b01, 0b000, "fcvtps", int_aarch64_neon_fcvtps>;
2319 defm FCVTPU : FPToIntegerUnscaled<0b01, 0b001, "fcvtpu", int_aarch64_neon_fcvtpu>;
2320 defm FCVTZS : FPToIntegerUnscaled<0b11, 0b000, "fcvtzs", fp_to_sint>;
2321 defm FCVTZU : FPToIntegerUnscaled<0b11, 0b001, "fcvtzu", fp_to_uint>;
2322 defm FCVTZS : FPToIntegerScaled<0b11, 0b000, "fcvtzs", fp_to_sint>;
2323 defm FCVTZU : FPToIntegerScaled<0b11, 0b001, "fcvtzu", fp_to_uint>;
2324 let isCodeGenOnly = 1 in {
2325 defm FCVTZS_Int : FPToIntegerUnscaled<0b11, 0b000, "fcvtzs", int_aarch64_neon_fcvtzs>;
2326 defm FCVTZU_Int : FPToIntegerUnscaled<0b11, 0b001, "fcvtzu", int_aarch64_neon_fcvtzu>;
2327 defm FCVTZS_Int : FPToIntegerScaled<0b11, 0b000, "fcvtzs", int_aarch64_neon_fcvtzs>;
2328 defm FCVTZU_Int : FPToIntegerScaled<0b11, 0b001, "fcvtzu", int_aarch64_neon_fcvtzu>;
2331 //===----------------------------------------------------------------------===//
2332 // Scaled integer to floating point conversion instructions.
2333 //===----------------------------------------------------------------------===//
2335 defm SCVTF : IntegerToFP<0, "scvtf", sint_to_fp>;
2336 defm UCVTF : IntegerToFP<1, "ucvtf", uint_to_fp>;
2338 //===----------------------------------------------------------------------===//
2339 // Unscaled integer to floating point conversion instruction.
2340 //===----------------------------------------------------------------------===//
2342 defm FMOV : UnscaledConversion<"fmov">;
2344 def : Pat<(f32 (fpimm0)), (FMOVWSr WZR)>, Requires<[NoZCZ]>;
2345 def : Pat<(f64 (fpimm0)), (FMOVXDr XZR)>, Requires<[NoZCZ]>;
2347 //===----------------------------------------------------------------------===//
2348 // Floating point conversion instruction.
2349 //===----------------------------------------------------------------------===//
2351 defm FCVT : FPConversion<"fcvt">;
2353 //===----------------------------------------------------------------------===//
2354 // Floating point single operand instructions.
2355 //===----------------------------------------------------------------------===//
2357 defm FABS : SingleOperandFPData<0b0001, "fabs", fabs>;
2358 defm FMOV : SingleOperandFPData<0b0000, "fmov">;
2359 defm FNEG : SingleOperandFPData<0b0010, "fneg", fneg>;
2360 defm FRINTA : SingleOperandFPData<0b1100, "frinta", frnd>;
2361 defm FRINTI : SingleOperandFPData<0b1111, "frinti", fnearbyint>;
2362 defm FRINTM : SingleOperandFPData<0b1010, "frintm", ffloor>;
2363 defm FRINTN : SingleOperandFPData<0b1000, "frintn", int_aarch64_neon_frintn>;
2364 defm FRINTP : SingleOperandFPData<0b1001, "frintp", fceil>;
2366 def : Pat<(v1f64 (int_aarch64_neon_frintn (v1f64 FPR64:$Rn))),
2367 (FRINTNDr FPR64:$Rn)>;
2369 // FRINTX is inserted to set the flags as required by FENV_ACCESS ON behavior
2370 // in the C spec. Setting hasSideEffects ensures it is not DCE'd.
2371 // <rdar://problem/13715968>
2372 // TODO: We should really model the FPSR flags correctly. This is really ugly.
2373 let hasSideEffects = 1 in {
2374 defm FRINTX : SingleOperandFPData<0b1110, "frintx", frint>;
2377 defm FRINTZ : SingleOperandFPData<0b1011, "frintz", ftrunc>;
2379 let SchedRW = [WriteFDiv] in {
2380 defm FSQRT : SingleOperandFPData<0b0011, "fsqrt", fsqrt>;
2383 //===----------------------------------------------------------------------===//
2384 // Floating point two operand instructions.
2385 //===----------------------------------------------------------------------===//
2387 defm FADD : TwoOperandFPData<0b0010, "fadd", fadd>;
2388 let SchedRW = [WriteFDiv] in {
2389 defm FDIV : TwoOperandFPData<0b0001, "fdiv", fdiv>;
2391 defm FMAXNM : TwoOperandFPData<0b0110, "fmaxnm", int_aarch64_neon_fmaxnm>;
2392 defm FMAX : TwoOperandFPData<0b0100, "fmax", AArch64fmax>;
2393 defm FMINNM : TwoOperandFPData<0b0111, "fminnm", int_aarch64_neon_fminnm>;
2394 defm FMIN : TwoOperandFPData<0b0101, "fmin", AArch64fmin>;
2395 let SchedRW = [WriteFMul] in {
2396 defm FMUL : TwoOperandFPData<0b0000, "fmul", fmul>;
2397 defm FNMUL : TwoOperandFPDataNeg<0b1000, "fnmul", fmul>;
2399 defm FSUB : TwoOperandFPData<0b0011, "fsub", fsub>;
2401 def : Pat<(v1f64 (AArch64fmax (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2402 (FMAXDrr FPR64:$Rn, FPR64:$Rm)>;
2403 def : Pat<(v1f64 (AArch64fmin (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2404 (FMINDrr FPR64:$Rn, FPR64:$Rm)>;
2405 def : Pat<(v1f64 (int_aarch64_neon_fmaxnm (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2406 (FMAXNMDrr FPR64:$Rn, FPR64:$Rm)>;
2407 def : Pat<(v1f64 (int_aarch64_neon_fminnm (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2408 (FMINNMDrr FPR64:$Rn, FPR64:$Rm)>;
2410 //===----------------------------------------------------------------------===//
2411 // Floating point three operand instructions.
2412 //===----------------------------------------------------------------------===//
2414 defm FMADD : ThreeOperandFPData<0, 0, "fmadd", fma>;
2415 defm FMSUB : ThreeOperandFPData<0, 1, "fmsub",
2416 TriOpFrag<(fma node:$LHS, (fneg node:$MHS), node:$RHS)> >;
2417 defm FNMADD : ThreeOperandFPData<1, 0, "fnmadd",
2418 TriOpFrag<(fneg (fma node:$LHS, node:$MHS, node:$RHS))> >;
2419 defm FNMSUB : ThreeOperandFPData<1, 1, "fnmsub",
2420 TriOpFrag<(fma node:$LHS, node:$MHS, (fneg node:$RHS))> >;
2422 // The following def pats catch the case where the LHS of an FMA is negated.
2423 // The TriOpFrag above catches the case where the middle operand is negated.
2425 // N.b. FMSUB etc have the accumulator at the *end* of (outs), unlike
2426 // the NEON variant.
2427 def : Pat<(f32 (fma (fneg FPR32:$Rn), FPR32:$Rm, FPR32:$Ra)),
2428 (FMSUBSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
2430 def : Pat<(f64 (fma (fneg FPR64:$Rn), FPR64:$Rm, FPR64:$Ra)),
2431 (FMSUBDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
2433 // We handled -(a + b*c) for FNMADD above, now it's time for "(-a) + (-b)*c" and
2435 def : Pat<(f32 (fma (fneg FPR32:$Rn), FPR32:$Rm, (fneg FPR32:$Ra))),
2436 (FNMADDSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
2438 def : Pat<(f64 (fma (fneg FPR64:$Rn), FPR64:$Rm, (fneg FPR64:$Ra))),
2439 (FNMADDDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
2441 def : Pat<(f32 (fma FPR32:$Rn, (fneg FPR32:$Rm), (fneg FPR32:$Ra))),
2442 (FNMADDSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
2444 def : Pat<(f64 (fma FPR64:$Rn, (fneg FPR64:$Rm), (fneg FPR64:$Ra))),
2445 (FNMADDDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
2447 //===----------------------------------------------------------------------===//
2448 // Floating point comparison instructions.
2449 //===----------------------------------------------------------------------===//
2451 defm FCMPE : FPComparison<1, "fcmpe">;
2452 defm FCMP : FPComparison<0, "fcmp", AArch64fcmp>;
2454 //===----------------------------------------------------------------------===//
2455 // Floating point conditional comparison instructions.
2456 //===----------------------------------------------------------------------===//
2458 defm FCCMPE : FPCondComparison<1, "fccmpe">;
2459 defm FCCMP : FPCondComparison<0, "fccmp">;
2461 //===----------------------------------------------------------------------===//
2462 // Floating point conditional select instruction.
2463 //===----------------------------------------------------------------------===//
2465 defm FCSEL : FPCondSelect<"fcsel">;
2467 // CSEL instructions providing f128 types need to be handled by a
2468 // pseudo-instruction since the eventual code will need to introduce basic
2469 // blocks and control flow.
2470 def F128CSEL : Pseudo<(outs FPR128:$Rd),
2471 (ins FPR128:$Rn, FPR128:$Rm, ccode:$cond),
2472 [(set (f128 FPR128:$Rd),
2473 (AArch64csel FPR128:$Rn, FPR128:$Rm,
2474 (i32 imm:$cond), NZCV))]> {
2476 let usesCustomInserter = 1;
2480 //===----------------------------------------------------------------------===//
2481 // Floating point immediate move.
2482 //===----------------------------------------------------------------------===//
2484 let isReMaterializable = 1 in {
2485 defm FMOV : FPMoveImmediate<"fmov">;
2488 //===----------------------------------------------------------------------===//
2489 // Advanced SIMD two vector instructions.
2490 //===----------------------------------------------------------------------===//
2492 defm ABS : SIMDTwoVectorBHSD<0, 0b01011, "abs", int_aarch64_neon_abs>;
2493 def : Pat<(xor (v8i8 (AArch64vashr V64:$src, (i32 7))),
2494 (v8i8 (add V64:$src, (AArch64vashr V64:$src, (i32 7))))),
2495 (ABSv8i8 V64:$src)>;
2496 def : Pat<(xor (v4i16 (AArch64vashr V64:$src, (i32 15))),
2497 (v4i16 (add V64:$src, (AArch64vashr V64:$src, (i32 15))))),
2498 (ABSv4i16 V64:$src)>;
2499 def : Pat<(xor (v2i32 (AArch64vashr V64:$src, (i32 31))),
2500 (v2i32 (add V64:$src, (AArch64vashr V64:$src, (i32 31))))),
2501 (ABSv2i32 V64:$src)>;
2502 def : Pat<(xor (v16i8 (AArch64vashr V128:$src, (i32 7))),
2503 (v16i8 (add V128:$src, (AArch64vashr V128:$src, (i32 7))))),
2504 (ABSv16i8 V128:$src)>;
2505 def : Pat<(xor (v8i16 (AArch64vashr V128:$src, (i32 15))),
2506 (v8i16 (add V128:$src, (AArch64vashr V128:$src, (i32 15))))),
2507 (ABSv8i16 V128:$src)>;
2508 def : Pat<(xor (v4i32 (AArch64vashr V128:$src, (i32 31))),
2509 (v4i32 (add V128:$src, (AArch64vashr V128:$src, (i32 31))))),
2510 (ABSv4i32 V128:$src)>;
2511 def : Pat<(xor (v2i64 (AArch64vashr V128:$src, (i32 63))),
2512 (v2i64 (add V128:$src, (AArch64vashr V128:$src, (i32 63))))),
2513 (ABSv2i64 V128:$src)>;
2515 defm CLS : SIMDTwoVectorBHS<0, 0b00100, "cls", int_aarch64_neon_cls>;
2516 defm CLZ : SIMDTwoVectorBHS<1, 0b00100, "clz", ctlz>;
2517 defm CMEQ : SIMDCmpTwoVector<0, 0b01001, "cmeq", AArch64cmeqz>;
2518 defm CMGE : SIMDCmpTwoVector<1, 0b01000, "cmge", AArch64cmgez>;
2519 defm CMGT : SIMDCmpTwoVector<0, 0b01000, "cmgt", AArch64cmgtz>;
2520 defm CMLE : SIMDCmpTwoVector<1, 0b01001, "cmle", AArch64cmlez>;
2521 defm CMLT : SIMDCmpTwoVector<0, 0b01010, "cmlt", AArch64cmltz>;
2522 defm CNT : SIMDTwoVectorB<0, 0b00, 0b00101, "cnt", ctpop>;
2523 defm FABS : SIMDTwoVectorFP<0, 1, 0b01111, "fabs", fabs>;
2525 defm FCMEQ : SIMDFPCmpTwoVector<0, 1, 0b01101, "fcmeq", AArch64fcmeqz>;
2526 defm FCMGE : SIMDFPCmpTwoVector<1, 1, 0b01100, "fcmge", AArch64fcmgez>;
2527 defm FCMGT : SIMDFPCmpTwoVector<0, 1, 0b01100, "fcmgt", AArch64fcmgtz>;
2528 defm FCMLE : SIMDFPCmpTwoVector<1, 1, 0b01101, "fcmle", AArch64fcmlez>;
2529 defm FCMLT : SIMDFPCmpTwoVector<0, 1, 0b01110, "fcmlt", AArch64fcmltz>;
2530 defm FCVTAS : SIMDTwoVectorFPToInt<0,0,0b11100, "fcvtas",int_aarch64_neon_fcvtas>;
2531 defm FCVTAU : SIMDTwoVectorFPToInt<1,0,0b11100, "fcvtau",int_aarch64_neon_fcvtau>;
2532 defm FCVTL : SIMDFPWidenTwoVector<0, 0, 0b10111, "fcvtl">;
2533 def : Pat<(v4f32 (int_aarch64_neon_vcvthf2fp (v4i16 V64:$Rn))),
2534 (FCVTLv4i16 V64:$Rn)>;
2535 def : Pat<(v4f32 (int_aarch64_neon_vcvthf2fp (extract_subvector (v8i16 V128:$Rn),
2537 (FCVTLv8i16 V128:$Rn)>;
2538 def : Pat<(v2f64 (fextend (v2f32 V64:$Rn))), (FCVTLv2i32 V64:$Rn)>;
2539 def : Pat<(v2f64 (fextend (v2f32 (extract_subvector (v4f32 V128:$Rn),
2541 (FCVTLv4i32 V128:$Rn)>;
2543 def : Pat<(v4f32 (fextend (v4f16 V64:$Rn))), (FCVTLv4i16 V64:$Rn)>;
2544 def : Pat<(v4f32 (fextend (v4f16 (extract_subvector (v8f16 V128:$Rn),
2546 (FCVTLv8i16 V128:$Rn)>;
2548 defm FCVTMS : SIMDTwoVectorFPToInt<0,0,0b11011, "fcvtms",int_aarch64_neon_fcvtms>;
2549 defm FCVTMU : SIMDTwoVectorFPToInt<1,0,0b11011, "fcvtmu",int_aarch64_neon_fcvtmu>;
2550 defm FCVTNS : SIMDTwoVectorFPToInt<0,0,0b11010, "fcvtns",int_aarch64_neon_fcvtns>;
2551 defm FCVTNU : SIMDTwoVectorFPToInt<1,0,0b11010, "fcvtnu",int_aarch64_neon_fcvtnu>;
2552 defm FCVTN : SIMDFPNarrowTwoVector<0, 0, 0b10110, "fcvtn">;
2553 def : Pat<(v4i16 (int_aarch64_neon_vcvtfp2hf (v4f32 V128:$Rn))),
2554 (FCVTNv4i16 V128:$Rn)>;
2555 def : Pat<(concat_vectors V64:$Rd,
2556 (v4i16 (int_aarch64_neon_vcvtfp2hf (v4f32 V128:$Rn)))),
2557 (FCVTNv8i16 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
2558 def : Pat<(v2f32 (fround (v2f64 V128:$Rn))), (FCVTNv2i32 V128:$Rn)>;
2559 def : Pat<(v4f16 (fround (v4f32 V128:$Rn))), (FCVTNv4i16 V128:$Rn)>;
2560 def : Pat<(concat_vectors V64:$Rd, (v2f32 (fround (v2f64 V128:$Rn)))),
2561 (FCVTNv4i32 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
2562 defm FCVTPS : SIMDTwoVectorFPToInt<0,1,0b11010, "fcvtps",int_aarch64_neon_fcvtps>;
2563 defm FCVTPU : SIMDTwoVectorFPToInt<1,1,0b11010, "fcvtpu",int_aarch64_neon_fcvtpu>;
2564 defm FCVTXN : SIMDFPInexactCvtTwoVector<1, 0, 0b10110, "fcvtxn",
2565 int_aarch64_neon_fcvtxn>;
2566 defm FCVTZS : SIMDTwoVectorFPToInt<0, 1, 0b11011, "fcvtzs", fp_to_sint>;
2567 defm FCVTZU : SIMDTwoVectorFPToInt<1, 1, 0b11011, "fcvtzu", fp_to_uint>;
2568 let isCodeGenOnly = 1 in {
2569 defm FCVTZS_Int : SIMDTwoVectorFPToInt<0, 1, 0b11011, "fcvtzs",
2570 int_aarch64_neon_fcvtzs>;
2571 defm FCVTZU_Int : SIMDTwoVectorFPToInt<1, 1, 0b11011, "fcvtzu",
2572 int_aarch64_neon_fcvtzu>;
2574 defm FNEG : SIMDTwoVectorFP<1, 1, 0b01111, "fneg", fneg>;
2575 defm FRECPE : SIMDTwoVectorFP<0, 1, 0b11101, "frecpe", int_aarch64_neon_frecpe>;
2576 defm FRINTA : SIMDTwoVectorFP<1, 0, 0b11000, "frinta", frnd>;
2577 defm FRINTI : SIMDTwoVectorFP<1, 1, 0b11001, "frinti", fnearbyint>;
2578 defm FRINTM : SIMDTwoVectorFP<0, 0, 0b11001, "frintm", ffloor>;
2579 defm FRINTN : SIMDTwoVectorFP<0, 0, 0b11000, "frintn", int_aarch64_neon_frintn>;
2580 defm FRINTP : SIMDTwoVectorFP<0, 1, 0b11000, "frintp", fceil>;
2581 defm FRINTX : SIMDTwoVectorFP<1, 0, 0b11001, "frintx", frint>;
2582 defm FRINTZ : SIMDTwoVectorFP<0, 1, 0b11001, "frintz", ftrunc>;
2583 defm FRSQRTE: SIMDTwoVectorFP<1, 1, 0b11101, "frsqrte", int_aarch64_neon_frsqrte>;
2584 defm FSQRT : SIMDTwoVectorFP<1, 1, 0b11111, "fsqrt", fsqrt>;
2585 defm NEG : SIMDTwoVectorBHSD<1, 0b01011, "neg",
2586 UnOpFrag<(sub immAllZerosV, node:$LHS)> >;
2587 defm NOT : SIMDTwoVectorB<1, 0b00, 0b00101, "not", vnot>;
2588 // Aliases for MVN -> NOT.
2589 def : InstAlias<"mvn{ $Vd.8b, $Vn.8b|.8b $Vd, $Vn}",
2590 (NOTv8i8 V64:$Vd, V64:$Vn)>;
2591 def : InstAlias<"mvn{ $Vd.16b, $Vn.16b|.16b $Vd, $Vn}",
2592 (NOTv16i8 V128:$Vd, V128:$Vn)>;
2594 def : Pat<(AArch64neg (v8i8 V64:$Rn)), (NEGv8i8 V64:$Rn)>;
2595 def : Pat<(AArch64neg (v16i8 V128:$Rn)), (NEGv16i8 V128:$Rn)>;
2596 def : Pat<(AArch64neg (v4i16 V64:$Rn)), (NEGv4i16 V64:$Rn)>;
2597 def : Pat<(AArch64neg (v8i16 V128:$Rn)), (NEGv8i16 V128:$Rn)>;
2598 def : Pat<(AArch64neg (v2i32 V64:$Rn)), (NEGv2i32 V64:$Rn)>;
2599 def : Pat<(AArch64neg (v4i32 V128:$Rn)), (NEGv4i32 V128:$Rn)>;
2600 def : Pat<(AArch64neg (v2i64 V128:$Rn)), (NEGv2i64 V128:$Rn)>;
2602 def : Pat<(AArch64not (v8i8 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2603 def : Pat<(AArch64not (v16i8 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2604 def : Pat<(AArch64not (v4i16 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2605 def : Pat<(AArch64not (v8i16 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2606 def : Pat<(AArch64not (v2i32 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2607 def : Pat<(AArch64not (v1i64 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2608 def : Pat<(AArch64not (v4i32 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2609 def : Pat<(AArch64not (v2i64 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2611 def : Pat<(vnot (v4i16 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2612 def : Pat<(vnot (v8i16 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2613 def : Pat<(vnot (v2i32 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2614 def : Pat<(vnot (v4i32 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2615 def : Pat<(vnot (v2i64 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2617 defm RBIT : SIMDTwoVectorB<1, 0b01, 0b00101, "rbit", int_aarch64_neon_rbit>;
2618 defm REV16 : SIMDTwoVectorB<0, 0b00, 0b00001, "rev16", AArch64rev16>;
2619 defm REV32 : SIMDTwoVectorBH<1, 0b00000, "rev32", AArch64rev32>;
2620 defm REV64 : SIMDTwoVectorBHS<0, 0b00000, "rev64", AArch64rev64>;
2621 defm SADALP : SIMDLongTwoVectorTied<0, 0b00110, "sadalp",
2622 BinOpFrag<(add node:$LHS, (int_aarch64_neon_saddlp node:$RHS))> >;
2623 defm SADDLP : SIMDLongTwoVector<0, 0b00010, "saddlp", int_aarch64_neon_saddlp>;
2624 defm SCVTF : SIMDTwoVectorIntToFP<0, 0, 0b11101, "scvtf", sint_to_fp>;
2625 defm SHLL : SIMDVectorLShiftLongBySizeBHS;
2626 defm SQABS : SIMDTwoVectorBHSD<0, 0b00111, "sqabs", int_aarch64_neon_sqabs>;
2627 defm SQNEG : SIMDTwoVectorBHSD<1, 0b00111, "sqneg", int_aarch64_neon_sqneg>;
2628 defm SQXTN : SIMDMixedTwoVector<0, 0b10100, "sqxtn", int_aarch64_neon_sqxtn>;
2629 defm SQXTUN : SIMDMixedTwoVector<1, 0b10010, "sqxtun", int_aarch64_neon_sqxtun>;
2630 defm SUQADD : SIMDTwoVectorBHSDTied<0, 0b00011, "suqadd",int_aarch64_neon_suqadd>;
2631 defm UADALP : SIMDLongTwoVectorTied<1, 0b00110, "uadalp",
2632 BinOpFrag<(add node:$LHS, (int_aarch64_neon_uaddlp node:$RHS))> >;
2633 defm UADDLP : SIMDLongTwoVector<1, 0b00010, "uaddlp",
2634 int_aarch64_neon_uaddlp>;
2635 defm UCVTF : SIMDTwoVectorIntToFP<1, 0, 0b11101, "ucvtf", uint_to_fp>;
2636 defm UQXTN : SIMDMixedTwoVector<1, 0b10100, "uqxtn", int_aarch64_neon_uqxtn>;
2637 defm URECPE : SIMDTwoVectorS<0, 1, 0b11100, "urecpe", int_aarch64_neon_urecpe>;
2638 defm URSQRTE: SIMDTwoVectorS<1, 1, 0b11100, "ursqrte", int_aarch64_neon_ursqrte>;
2639 defm USQADD : SIMDTwoVectorBHSDTied<1, 0b00011, "usqadd",int_aarch64_neon_usqadd>;
2640 defm XTN : SIMDMixedTwoVector<0, 0b10010, "xtn", trunc>;
2642 def : Pat<(v4f16 (AArch64rev32 V64:$Rn)), (REV32v4i16 V64:$Rn)>;
2643 def : Pat<(v4f16 (AArch64rev64 V64:$Rn)), (REV64v4i16 V64:$Rn)>;
2644 def : Pat<(v8f16 (AArch64rev32 V128:$Rn)), (REV32v8i16 V128:$Rn)>;
2645 def : Pat<(v8f16 (AArch64rev64 V128:$Rn)), (REV64v8i16 V128:$Rn)>;
2646 def : Pat<(v2f32 (AArch64rev64 V64:$Rn)), (REV64v2i32 V64:$Rn)>;
2647 def : Pat<(v4f32 (AArch64rev64 V128:$Rn)), (REV64v4i32 V128:$Rn)>;
2649 // Patterns for vector long shift (by element width). These need to match all
2650 // three of zext, sext and anyext so it's easier to pull the patterns out of the
2652 multiclass SIMDVectorLShiftLongBySizeBHSPats<SDPatternOperator ext> {
2653 def : Pat<(AArch64vshl (v8i16 (ext (v8i8 V64:$Rn))), (i32 8)),
2654 (SHLLv8i8 V64:$Rn)>;
2655 def : Pat<(AArch64vshl (v8i16 (ext (extract_high_v16i8 V128:$Rn))), (i32 8)),
2656 (SHLLv16i8 V128:$Rn)>;
2657 def : Pat<(AArch64vshl (v4i32 (ext (v4i16 V64:$Rn))), (i32 16)),
2658 (SHLLv4i16 V64:$Rn)>;
2659 def : Pat<(AArch64vshl (v4i32 (ext (extract_high_v8i16 V128:$Rn))), (i32 16)),
2660 (SHLLv8i16 V128:$Rn)>;
2661 def : Pat<(AArch64vshl (v2i64 (ext (v2i32 V64:$Rn))), (i32 32)),
2662 (SHLLv2i32 V64:$Rn)>;
2663 def : Pat<(AArch64vshl (v2i64 (ext (extract_high_v4i32 V128:$Rn))), (i32 32)),
2664 (SHLLv4i32 V128:$Rn)>;
2667 defm : SIMDVectorLShiftLongBySizeBHSPats<anyext>;
2668 defm : SIMDVectorLShiftLongBySizeBHSPats<zext>;
2669 defm : SIMDVectorLShiftLongBySizeBHSPats<sext>;
2671 //===----------------------------------------------------------------------===//
2672 // Advanced SIMD three vector instructions.
2673 //===----------------------------------------------------------------------===//
2675 defm ADD : SIMDThreeSameVector<0, 0b10000, "add", add>;
2676 defm ADDP : SIMDThreeSameVector<0, 0b10111, "addp", int_aarch64_neon_addp>;
2677 defm CMEQ : SIMDThreeSameVector<1, 0b10001, "cmeq", AArch64cmeq>;
2678 defm CMGE : SIMDThreeSameVector<0, 0b00111, "cmge", AArch64cmge>;
2679 defm CMGT : SIMDThreeSameVector<0, 0b00110, "cmgt", AArch64cmgt>;
2680 defm CMHI : SIMDThreeSameVector<1, 0b00110, "cmhi", AArch64cmhi>;
2681 defm CMHS : SIMDThreeSameVector<1, 0b00111, "cmhs", AArch64cmhs>;
2682 defm CMTST : SIMDThreeSameVector<0, 0b10001, "cmtst", AArch64cmtst>;
2683 defm FABD : SIMDThreeSameVectorFP<1,1,0b11010,"fabd", int_aarch64_neon_fabd>;
2684 defm FACGE : SIMDThreeSameVectorFPCmp<1,0,0b11101,"facge",int_aarch64_neon_facge>;
2685 defm FACGT : SIMDThreeSameVectorFPCmp<1,1,0b11101,"facgt",int_aarch64_neon_facgt>;
2686 defm FADDP : SIMDThreeSameVectorFP<1,0,0b11010,"faddp",int_aarch64_neon_addp>;
2687 defm FADD : SIMDThreeSameVectorFP<0,0,0b11010,"fadd", fadd>;
2688 defm FCMEQ : SIMDThreeSameVectorFPCmp<0, 0, 0b11100, "fcmeq", AArch64fcmeq>;
2689 defm FCMGE : SIMDThreeSameVectorFPCmp<1, 0, 0b11100, "fcmge", AArch64fcmge>;
2690 defm FCMGT : SIMDThreeSameVectorFPCmp<1, 1, 0b11100, "fcmgt", AArch64fcmgt>;
2691 defm FDIV : SIMDThreeSameVectorFP<1,0,0b11111,"fdiv", fdiv>;
2692 defm FMAXNMP : SIMDThreeSameVectorFP<1,0,0b11000,"fmaxnmp", int_aarch64_neon_fmaxnmp>;
2693 defm FMAXNM : SIMDThreeSameVectorFP<0,0,0b11000,"fmaxnm", int_aarch64_neon_fmaxnm>;
2694 defm FMAXP : SIMDThreeSameVectorFP<1,0,0b11110,"fmaxp", int_aarch64_neon_fmaxp>;
2695 defm FMAX : SIMDThreeSameVectorFP<0,0,0b11110,"fmax", AArch64fmax>;
2696 defm FMINNMP : SIMDThreeSameVectorFP<1,1,0b11000,"fminnmp", int_aarch64_neon_fminnmp>;
2697 defm FMINNM : SIMDThreeSameVectorFP<0,1,0b11000,"fminnm", int_aarch64_neon_fminnm>;
2698 defm FMINP : SIMDThreeSameVectorFP<1,1,0b11110,"fminp", int_aarch64_neon_fminp>;
2699 defm FMIN : SIMDThreeSameVectorFP<0,1,0b11110,"fmin", AArch64fmin>;
2701 // NOTE: The operands of the PatFrag are reordered on FMLA/FMLS because the
2702 // instruction expects the addend first, while the fma intrinsic puts it last.
2703 defm FMLA : SIMDThreeSameVectorFPTied<0, 0, 0b11001, "fmla",
2704 TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)> >;
2705 defm FMLS : SIMDThreeSameVectorFPTied<0, 1, 0b11001, "fmls",
2706 TriOpFrag<(fma node:$MHS, (fneg node:$RHS), node:$LHS)> >;
2708 // The following def pats catch the case where the LHS of an FMA is negated.
2709 // The TriOpFrag above catches the case where the middle operand is negated.
2710 def : Pat<(v2f32 (fma (fneg V64:$Rn), V64:$Rm, V64:$Rd)),
2711 (FMLSv2f32 V64:$Rd, V64:$Rn, V64:$Rm)>;
2713 def : Pat<(v4f32 (fma (fneg V128:$Rn), V128:$Rm, V128:$Rd)),
2714 (FMLSv4f32 V128:$Rd, V128:$Rn, V128:$Rm)>;
2716 def : Pat<(v2f64 (fma (fneg V128:$Rn), V128:$Rm, V128:$Rd)),
2717 (FMLSv2f64 V128:$Rd, V128:$Rn, V128:$Rm)>;
2719 defm FMULX : SIMDThreeSameVectorFP<0,0,0b11011,"fmulx", int_aarch64_neon_fmulx>;
2720 defm FMUL : SIMDThreeSameVectorFP<1,0,0b11011,"fmul", fmul>;
2721 defm FRECPS : SIMDThreeSameVectorFP<0,0,0b11111,"frecps", int_aarch64_neon_frecps>;
2722 defm FRSQRTS : SIMDThreeSameVectorFP<0,1,0b11111,"frsqrts", int_aarch64_neon_frsqrts>;
2723 defm FSUB : SIMDThreeSameVectorFP<0,1,0b11010,"fsub", fsub>;
2724 defm MLA : SIMDThreeSameVectorBHSTied<0, 0b10010, "mla",
2725 TriOpFrag<(add node:$LHS, (mul node:$MHS, node:$RHS))> >;
2726 defm MLS : SIMDThreeSameVectorBHSTied<1, 0b10010, "mls",
2727 TriOpFrag<(sub node:$LHS, (mul node:$MHS, node:$RHS))> >;
2728 defm MUL : SIMDThreeSameVectorBHS<0, 0b10011, "mul", mul>;
2729 defm PMUL : SIMDThreeSameVectorB<1, 0b10011, "pmul", int_aarch64_neon_pmul>;
2730 defm SABA : SIMDThreeSameVectorBHSTied<0, 0b01111, "saba",
2731 TriOpFrag<(add node:$LHS, (int_aarch64_neon_sabd node:$MHS, node:$RHS))> >;
2732 defm SABD : SIMDThreeSameVectorBHS<0,0b01110,"sabd", int_aarch64_neon_sabd>;
2733 defm SHADD : SIMDThreeSameVectorBHS<0,0b00000,"shadd", int_aarch64_neon_shadd>;
2734 defm SHSUB : SIMDThreeSameVectorBHS<0,0b00100,"shsub", int_aarch64_neon_shsub>;
2735 defm SMAXP : SIMDThreeSameVectorBHS<0,0b10100,"smaxp", int_aarch64_neon_smaxp>;
2736 defm SMAX : SIMDThreeSameVectorBHS<0,0b01100,"smax", int_aarch64_neon_smax>;
2737 defm SMINP : SIMDThreeSameVectorBHS<0,0b10101,"sminp", int_aarch64_neon_sminp>;
2738 defm SMIN : SIMDThreeSameVectorBHS<0,0b01101,"smin", int_aarch64_neon_smin>;
2739 defm SQADD : SIMDThreeSameVector<0,0b00001,"sqadd", int_aarch64_neon_sqadd>;
2740 defm SQDMULH : SIMDThreeSameVectorHS<0,0b10110,"sqdmulh",int_aarch64_neon_sqdmulh>;
2741 defm SQRDMULH : SIMDThreeSameVectorHS<1,0b10110,"sqrdmulh",int_aarch64_neon_sqrdmulh>;
2742 defm SQRSHL : SIMDThreeSameVector<0,0b01011,"sqrshl", int_aarch64_neon_sqrshl>;
2743 defm SQSHL : SIMDThreeSameVector<0,0b01001,"sqshl", int_aarch64_neon_sqshl>;
2744 defm SQSUB : SIMDThreeSameVector<0,0b00101,"sqsub", int_aarch64_neon_sqsub>;
2745 defm SRHADD : SIMDThreeSameVectorBHS<0,0b00010,"srhadd",int_aarch64_neon_srhadd>;
2746 defm SRSHL : SIMDThreeSameVector<0,0b01010,"srshl", int_aarch64_neon_srshl>;
2747 defm SSHL : SIMDThreeSameVector<0,0b01000,"sshl", int_aarch64_neon_sshl>;
2748 defm SUB : SIMDThreeSameVector<1,0b10000,"sub", sub>;
2749 defm UABA : SIMDThreeSameVectorBHSTied<1, 0b01111, "uaba",
2750 TriOpFrag<(add node:$LHS, (int_aarch64_neon_uabd node:$MHS, node:$RHS))> >;
2751 defm UABD : SIMDThreeSameVectorBHS<1,0b01110,"uabd", int_aarch64_neon_uabd>;
2752 defm UHADD : SIMDThreeSameVectorBHS<1,0b00000,"uhadd", int_aarch64_neon_uhadd>;
2753 defm UHSUB : SIMDThreeSameVectorBHS<1,0b00100,"uhsub", int_aarch64_neon_uhsub>;
2754 defm UMAXP : SIMDThreeSameVectorBHS<1,0b10100,"umaxp", int_aarch64_neon_umaxp>;
2755 defm UMAX : SIMDThreeSameVectorBHS<1,0b01100,"umax", int_aarch64_neon_umax>;
2756 defm UMINP : SIMDThreeSameVectorBHS<1,0b10101,"uminp", int_aarch64_neon_uminp>;
2757 defm UMIN : SIMDThreeSameVectorBHS<1,0b01101,"umin", int_aarch64_neon_umin>;
2758 defm UQADD : SIMDThreeSameVector<1,0b00001,"uqadd", int_aarch64_neon_uqadd>;
2759 defm UQRSHL : SIMDThreeSameVector<1,0b01011,"uqrshl", int_aarch64_neon_uqrshl>;
2760 defm UQSHL : SIMDThreeSameVector<1,0b01001,"uqshl", int_aarch64_neon_uqshl>;
2761 defm UQSUB : SIMDThreeSameVector<1,0b00101,"uqsub", int_aarch64_neon_uqsub>;
2762 defm URHADD : SIMDThreeSameVectorBHS<1,0b00010,"urhadd", int_aarch64_neon_urhadd>;
2763 defm URSHL : SIMDThreeSameVector<1,0b01010,"urshl", int_aarch64_neon_urshl>;
2764 defm USHL : SIMDThreeSameVector<1,0b01000,"ushl", int_aarch64_neon_ushl>;
2766 defm AND : SIMDLogicalThreeVector<0, 0b00, "and", and>;
2767 defm BIC : SIMDLogicalThreeVector<0, 0b01, "bic",
2768 BinOpFrag<(and node:$LHS, (vnot node:$RHS))> >;
2769 defm BIF : SIMDLogicalThreeVector<1, 0b11, "bif">;
2770 defm BIT : SIMDLogicalThreeVectorTied<1, 0b10, "bit", AArch64bit>;
2771 defm BSL : SIMDLogicalThreeVectorTied<1, 0b01, "bsl",
2772 TriOpFrag<(or (and node:$LHS, node:$MHS), (and (vnot node:$LHS), node:$RHS))>>;
2773 defm EOR : SIMDLogicalThreeVector<1, 0b00, "eor", xor>;
2774 defm ORN : SIMDLogicalThreeVector<0, 0b11, "orn",
2775 BinOpFrag<(or node:$LHS, (vnot node:$RHS))> >;
2776 defm ORR : SIMDLogicalThreeVector<0, 0b10, "orr", or>;
2778 def : Pat<(AArch64bsl (v8i8 V64:$Rd), V64:$Rn, V64:$Rm),
2779 (BSLv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
2780 def : Pat<(AArch64bsl (v4i16 V64:$Rd), V64:$Rn, V64:$Rm),
2781 (BSLv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
2782 def : Pat<(AArch64bsl (v2i32 V64:$Rd), V64:$Rn, V64:$Rm),
2783 (BSLv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
2784 def : Pat<(AArch64bsl (v1i64 V64:$Rd), V64:$Rn, V64:$Rm),
2785 (BSLv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
2787 def : Pat<(AArch64bsl (v16i8 V128:$Rd), V128:$Rn, V128:$Rm),
2788 (BSLv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
2789 def : Pat<(AArch64bsl (v8i16 V128:$Rd), V128:$Rn, V128:$Rm),
2790 (BSLv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
2791 def : Pat<(AArch64bsl (v4i32 V128:$Rd), V128:$Rn, V128:$Rm),
2792 (BSLv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
2793 def : Pat<(AArch64bsl (v2i64 V128:$Rd), V128:$Rn, V128:$Rm),
2794 (BSLv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
2796 def : InstAlias<"mov{\t$dst.16b, $src.16b|.16b\t$dst, $src}",
2797 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 1>;
2798 def : InstAlias<"mov{\t$dst.8h, $src.8h|.8h\t$dst, $src}",
2799 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2800 def : InstAlias<"mov{\t$dst.4s, $src.4s|.4s\t$dst, $src}",
2801 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2802 def : InstAlias<"mov{\t$dst.2d, $src.2d|.2d\t$dst, $src}",
2803 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2805 def : InstAlias<"mov{\t$dst.8b, $src.8b|.8b\t$dst, $src}",
2806 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 1>;
2807 def : InstAlias<"mov{\t$dst.4h, $src.4h|.4h\t$dst, $src}",
2808 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2809 def : InstAlias<"mov{\t$dst.2s, $src.2s|.2s\t$dst, $src}",
2810 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2811 def : InstAlias<"mov{\t$dst.1d, $src.1d|.1d\t$dst, $src}",
2812 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2814 def : InstAlias<"{cmls\t$dst.8b, $src1.8b, $src2.8b" #
2815 "|cmls.8b\t$dst, $src1, $src2}",
2816 (CMHSv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2817 def : InstAlias<"{cmls\t$dst.16b, $src1.16b, $src2.16b" #
2818 "|cmls.16b\t$dst, $src1, $src2}",
2819 (CMHSv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2820 def : InstAlias<"{cmls\t$dst.4h, $src1.4h, $src2.4h" #
2821 "|cmls.4h\t$dst, $src1, $src2}",
2822 (CMHSv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2823 def : InstAlias<"{cmls\t$dst.8h, $src1.8h, $src2.8h" #
2824 "|cmls.8h\t$dst, $src1, $src2}",
2825 (CMHSv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2826 def : InstAlias<"{cmls\t$dst.2s, $src1.2s, $src2.2s" #
2827 "|cmls.2s\t$dst, $src1, $src2}",
2828 (CMHSv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2829 def : InstAlias<"{cmls\t$dst.4s, $src1.4s, $src2.4s" #
2830 "|cmls.4s\t$dst, $src1, $src2}",
2831 (CMHSv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2832 def : InstAlias<"{cmls\t$dst.2d, $src1.2d, $src2.2d" #
2833 "|cmls.2d\t$dst, $src1, $src2}",
2834 (CMHSv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2836 def : InstAlias<"{cmlo\t$dst.8b, $src1.8b, $src2.8b" #
2837 "|cmlo.8b\t$dst, $src1, $src2}",
2838 (CMHIv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2839 def : InstAlias<"{cmlo\t$dst.16b, $src1.16b, $src2.16b" #
2840 "|cmlo.16b\t$dst, $src1, $src2}",
2841 (CMHIv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2842 def : InstAlias<"{cmlo\t$dst.4h, $src1.4h, $src2.4h" #
2843 "|cmlo.4h\t$dst, $src1, $src2}",
2844 (CMHIv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2845 def : InstAlias<"{cmlo\t$dst.8h, $src1.8h, $src2.8h" #
2846 "|cmlo.8h\t$dst, $src1, $src2}",
2847 (CMHIv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2848 def : InstAlias<"{cmlo\t$dst.2s, $src1.2s, $src2.2s" #
2849 "|cmlo.2s\t$dst, $src1, $src2}",
2850 (CMHIv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2851 def : InstAlias<"{cmlo\t$dst.4s, $src1.4s, $src2.4s" #
2852 "|cmlo.4s\t$dst, $src1, $src2}",
2853 (CMHIv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2854 def : InstAlias<"{cmlo\t$dst.2d, $src1.2d, $src2.2d" #
2855 "|cmlo.2d\t$dst, $src1, $src2}",
2856 (CMHIv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2858 def : InstAlias<"{cmle\t$dst.8b, $src1.8b, $src2.8b" #
2859 "|cmle.8b\t$dst, $src1, $src2}",
2860 (CMGEv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2861 def : InstAlias<"{cmle\t$dst.16b, $src1.16b, $src2.16b" #
2862 "|cmle.16b\t$dst, $src1, $src2}",
2863 (CMGEv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2864 def : InstAlias<"{cmle\t$dst.4h, $src1.4h, $src2.4h" #
2865 "|cmle.4h\t$dst, $src1, $src2}",
2866 (CMGEv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2867 def : InstAlias<"{cmle\t$dst.8h, $src1.8h, $src2.8h" #
2868 "|cmle.8h\t$dst, $src1, $src2}",
2869 (CMGEv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2870 def : InstAlias<"{cmle\t$dst.2s, $src1.2s, $src2.2s" #
2871 "|cmle.2s\t$dst, $src1, $src2}",
2872 (CMGEv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2873 def : InstAlias<"{cmle\t$dst.4s, $src1.4s, $src2.4s" #
2874 "|cmle.4s\t$dst, $src1, $src2}",
2875 (CMGEv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2876 def : InstAlias<"{cmle\t$dst.2d, $src1.2d, $src2.2d" #
2877 "|cmle.2d\t$dst, $src1, $src2}",
2878 (CMGEv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2880 def : InstAlias<"{cmlt\t$dst.8b, $src1.8b, $src2.8b" #
2881 "|cmlt.8b\t$dst, $src1, $src2}",
2882 (CMGTv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2883 def : InstAlias<"{cmlt\t$dst.16b, $src1.16b, $src2.16b" #
2884 "|cmlt.16b\t$dst, $src1, $src2}",
2885 (CMGTv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2886 def : InstAlias<"{cmlt\t$dst.4h, $src1.4h, $src2.4h" #
2887 "|cmlt.4h\t$dst, $src1, $src2}",
2888 (CMGTv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2889 def : InstAlias<"{cmlt\t$dst.8h, $src1.8h, $src2.8h" #
2890 "|cmlt.8h\t$dst, $src1, $src2}",
2891 (CMGTv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2892 def : InstAlias<"{cmlt\t$dst.2s, $src1.2s, $src2.2s" #
2893 "|cmlt.2s\t$dst, $src1, $src2}",
2894 (CMGTv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2895 def : InstAlias<"{cmlt\t$dst.4s, $src1.4s, $src2.4s" #
2896 "|cmlt.4s\t$dst, $src1, $src2}",
2897 (CMGTv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2898 def : InstAlias<"{cmlt\t$dst.2d, $src1.2d, $src2.2d" #
2899 "|cmlt.2d\t$dst, $src1, $src2}",
2900 (CMGTv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2902 def : InstAlias<"{fcmle\t$dst.2s, $src1.2s, $src2.2s" #
2903 "|fcmle.2s\t$dst, $src1, $src2}",
2904 (FCMGEv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
2905 def : InstAlias<"{fcmle\t$dst.4s, $src1.4s, $src2.4s" #
2906 "|fcmle.4s\t$dst, $src1, $src2}",
2907 (FCMGEv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
2908 def : InstAlias<"{fcmle\t$dst.2d, $src1.2d, $src2.2d" #
2909 "|fcmle.2d\t$dst, $src1, $src2}",
2910 (FCMGEv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
2912 def : InstAlias<"{fcmlt\t$dst.2s, $src1.2s, $src2.2s" #
2913 "|fcmlt.2s\t$dst, $src1, $src2}",
2914 (FCMGTv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
2915 def : InstAlias<"{fcmlt\t$dst.4s, $src1.4s, $src2.4s" #
2916 "|fcmlt.4s\t$dst, $src1, $src2}",
2917 (FCMGTv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
2918 def : InstAlias<"{fcmlt\t$dst.2d, $src1.2d, $src2.2d" #
2919 "|fcmlt.2d\t$dst, $src1, $src2}",
2920 (FCMGTv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
2922 def : InstAlias<"{facle\t$dst.2s, $src1.2s, $src2.2s" #
2923 "|facle.2s\t$dst, $src1, $src2}",
2924 (FACGEv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
2925 def : InstAlias<"{facle\t$dst.4s, $src1.4s, $src2.4s" #
2926 "|facle.4s\t$dst, $src1, $src2}",
2927 (FACGEv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
2928 def : InstAlias<"{facle\t$dst.2d, $src1.2d, $src2.2d" #
2929 "|facle.2d\t$dst, $src1, $src2}",
2930 (FACGEv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
2932 def : InstAlias<"{faclt\t$dst.2s, $src1.2s, $src2.2s" #
2933 "|faclt.2s\t$dst, $src1, $src2}",
2934 (FACGTv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
2935 def : InstAlias<"{faclt\t$dst.4s, $src1.4s, $src2.4s" #
2936 "|faclt.4s\t$dst, $src1, $src2}",
2937 (FACGTv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
2938 def : InstAlias<"{faclt\t$dst.2d, $src1.2d, $src2.2d" #
2939 "|faclt.2d\t$dst, $src1, $src2}",
2940 (FACGTv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
2942 //===----------------------------------------------------------------------===//
2943 // Advanced SIMD three scalar instructions.
2944 //===----------------------------------------------------------------------===//
2946 defm ADD : SIMDThreeScalarD<0, 0b10000, "add", add>;
2947 defm CMEQ : SIMDThreeScalarD<1, 0b10001, "cmeq", AArch64cmeq>;
2948 defm CMGE : SIMDThreeScalarD<0, 0b00111, "cmge", AArch64cmge>;
2949 defm CMGT : SIMDThreeScalarD<0, 0b00110, "cmgt", AArch64cmgt>;
2950 defm CMHI : SIMDThreeScalarD<1, 0b00110, "cmhi", AArch64cmhi>;
2951 defm CMHS : SIMDThreeScalarD<1, 0b00111, "cmhs", AArch64cmhs>;
2952 defm CMTST : SIMDThreeScalarD<0, 0b10001, "cmtst", AArch64cmtst>;
2953 defm FABD : SIMDThreeScalarSD<1, 1, 0b11010, "fabd", int_aarch64_sisd_fabd>;
2954 def : Pat<(v1f64 (int_aarch64_neon_fabd (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2955 (FABD64 FPR64:$Rn, FPR64:$Rm)>;
2956 defm FACGE : SIMDThreeScalarFPCmp<1, 0, 0b11101, "facge",
2957 int_aarch64_neon_facge>;
2958 defm FACGT : SIMDThreeScalarFPCmp<1, 1, 0b11101, "facgt",
2959 int_aarch64_neon_facgt>;
2960 defm FCMEQ : SIMDThreeScalarFPCmp<0, 0, 0b11100, "fcmeq", AArch64fcmeq>;
2961 defm FCMGE : SIMDThreeScalarFPCmp<1, 0, 0b11100, "fcmge", AArch64fcmge>;
2962 defm FCMGT : SIMDThreeScalarFPCmp<1, 1, 0b11100, "fcmgt", AArch64fcmgt>;
2963 defm FMULX : SIMDThreeScalarSD<0, 0, 0b11011, "fmulx", int_aarch64_neon_fmulx>;
2964 defm FRECPS : SIMDThreeScalarSD<0, 0, 0b11111, "frecps", int_aarch64_neon_frecps>;
2965 defm FRSQRTS : SIMDThreeScalarSD<0, 1, 0b11111, "frsqrts", int_aarch64_neon_frsqrts>;
2966 defm SQADD : SIMDThreeScalarBHSD<0, 0b00001, "sqadd", int_aarch64_neon_sqadd>;
2967 defm SQDMULH : SIMDThreeScalarHS< 0, 0b10110, "sqdmulh", int_aarch64_neon_sqdmulh>;
2968 defm SQRDMULH : SIMDThreeScalarHS< 1, 0b10110, "sqrdmulh", int_aarch64_neon_sqrdmulh>;
2969 defm SQRSHL : SIMDThreeScalarBHSD<0, 0b01011, "sqrshl",int_aarch64_neon_sqrshl>;
2970 defm SQSHL : SIMDThreeScalarBHSD<0, 0b01001, "sqshl", int_aarch64_neon_sqshl>;
2971 defm SQSUB : SIMDThreeScalarBHSD<0, 0b00101, "sqsub", int_aarch64_neon_sqsub>;
2972 defm SRSHL : SIMDThreeScalarD< 0, 0b01010, "srshl", int_aarch64_neon_srshl>;
2973 defm SSHL : SIMDThreeScalarD< 0, 0b01000, "sshl", int_aarch64_neon_sshl>;
2974 defm SUB : SIMDThreeScalarD< 1, 0b10000, "sub", sub>;
2975 defm UQADD : SIMDThreeScalarBHSD<1, 0b00001, "uqadd", int_aarch64_neon_uqadd>;
2976 defm UQRSHL : SIMDThreeScalarBHSD<1, 0b01011, "uqrshl",int_aarch64_neon_uqrshl>;
2977 defm UQSHL : SIMDThreeScalarBHSD<1, 0b01001, "uqshl", int_aarch64_neon_uqshl>;
2978 defm UQSUB : SIMDThreeScalarBHSD<1, 0b00101, "uqsub", int_aarch64_neon_uqsub>;
2979 defm URSHL : SIMDThreeScalarD< 1, 0b01010, "urshl", int_aarch64_neon_urshl>;
2980 defm USHL : SIMDThreeScalarD< 1, 0b01000, "ushl", int_aarch64_neon_ushl>;
2982 def : InstAlias<"cmls $dst, $src1, $src2",
2983 (CMHSv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
2984 def : InstAlias<"cmle $dst, $src1, $src2",
2985 (CMGEv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
2986 def : InstAlias<"cmlo $dst, $src1, $src2",
2987 (CMHIv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
2988 def : InstAlias<"cmlt $dst, $src1, $src2",
2989 (CMGTv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
2990 def : InstAlias<"fcmle $dst, $src1, $src2",
2991 (FCMGE32 FPR32:$dst, FPR32:$src2, FPR32:$src1), 0>;
2992 def : InstAlias<"fcmle $dst, $src1, $src2",
2993 (FCMGE64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
2994 def : InstAlias<"fcmlt $dst, $src1, $src2",
2995 (FCMGT32 FPR32:$dst, FPR32:$src2, FPR32:$src1), 0>;
2996 def : InstAlias<"fcmlt $dst, $src1, $src2",
2997 (FCMGT64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
2998 def : InstAlias<"facle $dst, $src1, $src2",
2999 (FACGE32 FPR32:$dst, FPR32:$src2, FPR32:$src1), 0>;
3000 def : InstAlias<"facle $dst, $src1, $src2",
3001 (FACGE64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
3002 def : InstAlias<"faclt $dst, $src1, $src2",
3003 (FACGT32 FPR32:$dst, FPR32:$src2, FPR32:$src1), 0>;
3004 def : InstAlias<"faclt $dst, $src1, $src2",
3005 (FACGT64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
3007 //===----------------------------------------------------------------------===//
3008 // Advanced SIMD three scalar instructions (mixed operands).
3009 //===----------------------------------------------------------------------===//
3010 defm SQDMULL : SIMDThreeScalarMixedHS<0, 0b11010, "sqdmull",
3011 int_aarch64_neon_sqdmulls_scalar>;
3012 defm SQDMLAL : SIMDThreeScalarMixedTiedHS<0, 0b10010, "sqdmlal">;
3013 defm SQDMLSL : SIMDThreeScalarMixedTiedHS<0, 0b10110, "sqdmlsl">;
3015 def : Pat<(i64 (int_aarch64_neon_sqadd (i64 FPR64:$Rd),
3016 (i64 (int_aarch64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
3017 (i32 FPR32:$Rm))))),
3018 (SQDMLALi32 FPR64:$Rd, FPR32:$Rn, FPR32:$Rm)>;
3019 def : Pat<(i64 (int_aarch64_neon_sqsub (i64 FPR64:$Rd),
3020 (i64 (int_aarch64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
3021 (i32 FPR32:$Rm))))),
3022 (SQDMLSLi32 FPR64:$Rd, FPR32:$Rn, FPR32:$Rm)>;
3024 //===----------------------------------------------------------------------===//
3025 // Advanced SIMD two scalar instructions.
3026 //===----------------------------------------------------------------------===//
3028 defm ABS : SIMDTwoScalarD< 0, 0b01011, "abs", int_aarch64_neon_abs>;
3029 defm CMEQ : SIMDCmpTwoScalarD< 0, 0b01001, "cmeq", AArch64cmeqz>;
3030 defm CMGE : SIMDCmpTwoScalarD< 1, 0b01000, "cmge", AArch64cmgez>;
3031 defm CMGT : SIMDCmpTwoScalarD< 0, 0b01000, "cmgt", AArch64cmgtz>;
3032 defm CMLE : SIMDCmpTwoScalarD< 1, 0b01001, "cmle", AArch64cmlez>;
3033 defm CMLT : SIMDCmpTwoScalarD< 0, 0b01010, "cmlt", AArch64cmltz>;
3034 defm FCMEQ : SIMDCmpTwoScalarSD<0, 1, 0b01101, "fcmeq", AArch64fcmeqz>;
3035 defm FCMGE : SIMDCmpTwoScalarSD<1, 1, 0b01100, "fcmge", AArch64fcmgez>;
3036 defm FCMGT : SIMDCmpTwoScalarSD<0, 1, 0b01100, "fcmgt", AArch64fcmgtz>;
3037 defm FCMLE : SIMDCmpTwoScalarSD<1, 1, 0b01101, "fcmle", AArch64fcmlez>;
3038 defm FCMLT : SIMDCmpTwoScalarSD<0, 1, 0b01110, "fcmlt", AArch64fcmltz>;
3039 defm FCVTAS : SIMDTwoScalarSD< 0, 0, 0b11100, "fcvtas">;
3040 defm FCVTAU : SIMDTwoScalarSD< 1, 0, 0b11100, "fcvtau">;
3041 defm FCVTMS : SIMDTwoScalarSD< 0, 0, 0b11011, "fcvtms">;
3042 defm FCVTMU : SIMDTwoScalarSD< 1, 0, 0b11011, "fcvtmu">;
3043 defm FCVTNS : SIMDTwoScalarSD< 0, 0, 0b11010, "fcvtns">;
3044 defm FCVTNU : SIMDTwoScalarSD< 1, 0, 0b11010, "fcvtnu">;
3045 defm FCVTPS : SIMDTwoScalarSD< 0, 1, 0b11010, "fcvtps">;
3046 defm FCVTPU : SIMDTwoScalarSD< 1, 1, 0b11010, "fcvtpu">;
3047 def FCVTXNv1i64 : SIMDInexactCvtTwoScalar<0b10110, "fcvtxn">;
3048 defm FCVTZS : SIMDTwoScalarSD< 0, 1, 0b11011, "fcvtzs">;
3049 defm FCVTZU : SIMDTwoScalarSD< 1, 1, 0b11011, "fcvtzu">;
3050 defm FRECPE : SIMDTwoScalarSD< 0, 1, 0b11101, "frecpe">;
3051 defm FRECPX : SIMDTwoScalarSD< 0, 1, 0b11111, "frecpx">;
3052 defm FRSQRTE : SIMDTwoScalarSD< 1, 1, 0b11101, "frsqrte">;
3053 defm NEG : SIMDTwoScalarD< 1, 0b01011, "neg",
3054 UnOpFrag<(sub immAllZerosV, node:$LHS)> >;
3055 defm SCVTF : SIMDTwoScalarCVTSD< 0, 0, 0b11101, "scvtf", AArch64sitof>;
3056 defm SQABS : SIMDTwoScalarBHSD< 0, 0b00111, "sqabs", int_aarch64_neon_sqabs>;
3057 defm SQNEG : SIMDTwoScalarBHSD< 1, 0b00111, "sqneg", int_aarch64_neon_sqneg>;
3058 defm SQXTN : SIMDTwoScalarMixedBHS< 0, 0b10100, "sqxtn", int_aarch64_neon_scalar_sqxtn>;
3059 defm SQXTUN : SIMDTwoScalarMixedBHS< 1, 0b10010, "sqxtun", int_aarch64_neon_scalar_sqxtun>;
3060 defm SUQADD : SIMDTwoScalarBHSDTied< 0, 0b00011, "suqadd",
3061 int_aarch64_neon_suqadd>;
3062 defm UCVTF : SIMDTwoScalarCVTSD< 1, 0, 0b11101, "ucvtf", AArch64uitof>;
3063 defm UQXTN : SIMDTwoScalarMixedBHS<1, 0b10100, "uqxtn", int_aarch64_neon_scalar_uqxtn>;
3064 defm USQADD : SIMDTwoScalarBHSDTied< 1, 0b00011, "usqadd",
3065 int_aarch64_neon_usqadd>;
3067 def : Pat<(AArch64neg (v1i64 V64:$Rn)), (NEGv1i64 V64:$Rn)>;
3069 def : Pat<(v1i64 (int_aarch64_neon_fcvtas (v1f64 FPR64:$Rn))),
3070 (FCVTASv1i64 FPR64:$Rn)>;
3071 def : Pat<(v1i64 (int_aarch64_neon_fcvtau (v1f64 FPR64:$Rn))),
3072 (FCVTAUv1i64 FPR64:$Rn)>;
3073 def : Pat<(v1i64 (int_aarch64_neon_fcvtms (v1f64 FPR64:$Rn))),
3074 (FCVTMSv1i64 FPR64:$Rn)>;
3075 def : Pat<(v1i64 (int_aarch64_neon_fcvtmu (v1f64 FPR64:$Rn))),
3076 (FCVTMUv1i64 FPR64:$Rn)>;
3077 def : Pat<(v1i64 (int_aarch64_neon_fcvtns (v1f64 FPR64:$Rn))),
3078 (FCVTNSv1i64 FPR64:$Rn)>;
3079 def : Pat<(v1i64 (int_aarch64_neon_fcvtnu (v1f64 FPR64:$Rn))),
3080 (FCVTNUv1i64 FPR64:$Rn)>;
3081 def : Pat<(v1i64 (int_aarch64_neon_fcvtps (v1f64 FPR64:$Rn))),
3082 (FCVTPSv1i64 FPR64:$Rn)>;
3083 def : Pat<(v1i64 (int_aarch64_neon_fcvtpu (v1f64 FPR64:$Rn))),
3084 (FCVTPUv1i64 FPR64:$Rn)>;
3086 def : Pat<(f32 (int_aarch64_neon_frecpe (f32 FPR32:$Rn))),
3087 (FRECPEv1i32 FPR32:$Rn)>;
3088 def : Pat<(f64 (int_aarch64_neon_frecpe (f64 FPR64:$Rn))),
3089 (FRECPEv1i64 FPR64:$Rn)>;
3090 def : Pat<(v1f64 (int_aarch64_neon_frecpe (v1f64 FPR64:$Rn))),
3091 (FRECPEv1i64 FPR64:$Rn)>;
3093 def : Pat<(f32 (int_aarch64_neon_frecpx (f32 FPR32:$Rn))),
3094 (FRECPXv1i32 FPR32:$Rn)>;
3095 def : Pat<(f64 (int_aarch64_neon_frecpx (f64 FPR64:$Rn))),
3096 (FRECPXv1i64 FPR64:$Rn)>;
3098 def : Pat<(f32 (int_aarch64_neon_frsqrte (f32 FPR32:$Rn))),
3099 (FRSQRTEv1i32 FPR32:$Rn)>;
3100 def : Pat<(f64 (int_aarch64_neon_frsqrte (f64 FPR64:$Rn))),
3101 (FRSQRTEv1i64 FPR64:$Rn)>;
3102 def : Pat<(v1f64 (int_aarch64_neon_frsqrte (v1f64 FPR64:$Rn))),
3103 (FRSQRTEv1i64 FPR64:$Rn)>;
3105 // If an integer is about to be converted to a floating point value,
3106 // just load it on the floating point unit.
3107 // Here are the patterns for 8 and 16-bits to float.
3109 multiclass UIntToFPROLoadPat<ValueType DstTy, ValueType SrcTy,
3110 SDPatternOperator loadop, Instruction UCVTF,
3111 ROAddrMode ro, Instruction LDRW, Instruction LDRX,
3113 def : Pat<(DstTy (uint_to_fp (SrcTy
3114 (loadop (ro.Wpat GPR64sp:$Rn, GPR32:$Rm,
3115 ro.Wext:$extend))))),
3116 (UCVTF (INSERT_SUBREG (DstTy (IMPLICIT_DEF)),
3117 (LDRW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend),
3120 def : Pat<(DstTy (uint_to_fp (SrcTy
3121 (loadop (ro.Xpat GPR64sp:$Rn, GPR64:$Rm,
3122 ro.Wext:$extend))))),
3123 (UCVTF (INSERT_SUBREG (DstTy (IMPLICIT_DEF)),
3124 (LDRX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend),
3128 defm : UIntToFPROLoadPat<f32, i32, zextloadi8,
3129 UCVTFv1i32, ro8, LDRBroW, LDRBroX, bsub>;
3130 def : Pat <(f32 (uint_to_fp (i32
3131 (zextloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))))),
3132 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
3133 (LDRBui GPR64sp:$Rn, uimm12s1:$offset), bsub))>;
3134 def : Pat <(f32 (uint_to_fp (i32
3135 (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))))),
3136 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
3137 (LDURBi GPR64sp:$Rn, simm9:$offset), bsub))>;
3138 // 16-bits -> float.
3139 defm : UIntToFPROLoadPat<f32, i32, zextloadi16,
3140 UCVTFv1i32, ro16, LDRHroW, LDRHroX, hsub>;
3141 def : Pat <(f32 (uint_to_fp (i32
3142 (zextloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))))),
3143 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
3144 (LDRHui GPR64sp:$Rn, uimm12s2:$offset), hsub))>;
3145 def : Pat <(f32 (uint_to_fp (i32
3146 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))))),
3147 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
3148 (LDURHi GPR64sp:$Rn, simm9:$offset), hsub))>;
3149 // 32-bits are handled in target specific dag combine:
3150 // performIntToFpCombine.
3151 // 64-bits integer to 32-bits floating point, not possible with
3152 // UCVTF on floating point registers (both source and destination
3153 // must have the same size).
3155 // Here are the patterns for 8, 16, 32, and 64-bits to double.
3156 // 8-bits -> double.
3157 defm : UIntToFPROLoadPat<f64, i32, zextloadi8,
3158 UCVTFv1i64, ro8, LDRBroW, LDRBroX, bsub>;
3159 def : Pat <(f64 (uint_to_fp (i32
3160 (zextloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))))),
3161 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3162 (LDRBui GPR64sp:$Rn, uimm12s1:$offset), bsub))>;
3163 def : Pat <(f64 (uint_to_fp (i32
3164 (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))))),
3165 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3166 (LDURBi GPR64sp:$Rn, simm9:$offset), bsub))>;
3167 // 16-bits -> double.
3168 defm : UIntToFPROLoadPat<f64, i32, zextloadi16,
3169 UCVTFv1i64, ro16, LDRHroW, LDRHroX, hsub>;
3170 def : Pat <(f64 (uint_to_fp (i32
3171 (zextloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))))),
3172 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3173 (LDRHui GPR64sp:$Rn, uimm12s2:$offset), hsub))>;
3174 def : Pat <(f64 (uint_to_fp (i32
3175 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))))),
3176 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3177 (LDURHi GPR64sp:$Rn, simm9:$offset), hsub))>;
3178 // 32-bits -> double.
3179 defm : UIntToFPROLoadPat<f64, i32, load,
3180 UCVTFv1i64, ro32, LDRSroW, LDRSroX, ssub>;
3181 def : Pat <(f64 (uint_to_fp (i32
3182 (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))))),
3183 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3184 (LDRSui GPR64sp:$Rn, uimm12s4:$offset), ssub))>;
3185 def : Pat <(f64 (uint_to_fp (i32
3186 (load (am_unscaled32 GPR64sp:$Rn, simm9:$offset))))),
3187 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3188 (LDURSi GPR64sp:$Rn, simm9:$offset), ssub))>;
3189 // 64-bits -> double are handled in target specific dag combine:
3190 // performIntToFpCombine.
3192 //===----------------------------------------------------------------------===//
3193 // Advanced SIMD three different-sized vector instructions.
3194 //===----------------------------------------------------------------------===//
3196 defm ADDHN : SIMDNarrowThreeVectorBHS<0,0b0100,"addhn", int_aarch64_neon_addhn>;
3197 defm SUBHN : SIMDNarrowThreeVectorBHS<0,0b0110,"subhn", int_aarch64_neon_subhn>;
3198 defm RADDHN : SIMDNarrowThreeVectorBHS<1,0b0100,"raddhn",int_aarch64_neon_raddhn>;
3199 defm RSUBHN : SIMDNarrowThreeVectorBHS<1,0b0110,"rsubhn",int_aarch64_neon_rsubhn>;
3200 defm PMULL : SIMDDifferentThreeVectorBD<0,0b1110,"pmull",int_aarch64_neon_pmull>;
3201 defm SABAL : SIMDLongThreeVectorTiedBHSabal<0,0b0101,"sabal",
3202 int_aarch64_neon_sabd>;
3203 defm SABDL : SIMDLongThreeVectorBHSabdl<0, 0b0111, "sabdl",
3204 int_aarch64_neon_sabd>;
3205 defm SADDL : SIMDLongThreeVectorBHS< 0, 0b0000, "saddl",
3206 BinOpFrag<(add (sext node:$LHS), (sext node:$RHS))>>;
3207 defm SADDW : SIMDWideThreeVectorBHS< 0, 0b0001, "saddw",
3208 BinOpFrag<(add node:$LHS, (sext node:$RHS))>>;
3209 defm SMLAL : SIMDLongThreeVectorTiedBHS<0, 0b1000, "smlal",
3210 TriOpFrag<(add node:$LHS, (int_aarch64_neon_smull node:$MHS, node:$RHS))>>;
3211 defm SMLSL : SIMDLongThreeVectorTiedBHS<0, 0b1010, "smlsl",
3212 TriOpFrag<(sub node:$LHS, (int_aarch64_neon_smull node:$MHS, node:$RHS))>>;
3213 defm SMULL : SIMDLongThreeVectorBHS<0, 0b1100, "smull", int_aarch64_neon_smull>;
3214 defm SQDMLAL : SIMDLongThreeVectorSQDMLXTiedHS<0, 0b1001, "sqdmlal",
3215 int_aarch64_neon_sqadd>;
3216 defm SQDMLSL : SIMDLongThreeVectorSQDMLXTiedHS<0, 0b1011, "sqdmlsl",
3217 int_aarch64_neon_sqsub>;
3218 defm SQDMULL : SIMDLongThreeVectorHS<0, 0b1101, "sqdmull",
3219 int_aarch64_neon_sqdmull>;
3220 defm SSUBL : SIMDLongThreeVectorBHS<0, 0b0010, "ssubl",
3221 BinOpFrag<(sub (sext node:$LHS), (sext node:$RHS))>>;
3222 defm SSUBW : SIMDWideThreeVectorBHS<0, 0b0011, "ssubw",
3223 BinOpFrag<(sub node:$LHS, (sext node:$RHS))>>;
3224 defm UABAL : SIMDLongThreeVectorTiedBHSabal<1, 0b0101, "uabal",
3225 int_aarch64_neon_uabd>;
3226 defm UABDL : SIMDLongThreeVectorBHSabdl<1, 0b0111, "uabdl",
3227 int_aarch64_neon_uabd>;
3228 defm UADDL : SIMDLongThreeVectorBHS<1, 0b0000, "uaddl",
3229 BinOpFrag<(add (zext node:$LHS), (zext node:$RHS))>>;
3230 defm UADDW : SIMDWideThreeVectorBHS<1, 0b0001, "uaddw",
3231 BinOpFrag<(add node:$LHS, (zext node:$RHS))>>;
3232 defm UMLAL : SIMDLongThreeVectorTiedBHS<1, 0b1000, "umlal",
3233 TriOpFrag<(add node:$LHS, (int_aarch64_neon_umull node:$MHS, node:$RHS))>>;
3234 defm UMLSL : SIMDLongThreeVectorTiedBHS<1, 0b1010, "umlsl",
3235 TriOpFrag<(sub node:$LHS, (int_aarch64_neon_umull node:$MHS, node:$RHS))>>;
3236 defm UMULL : SIMDLongThreeVectorBHS<1, 0b1100, "umull", int_aarch64_neon_umull>;
3237 defm USUBL : SIMDLongThreeVectorBHS<1, 0b0010, "usubl",
3238 BinOpFrag<(sub (zext node:$LHS), (zext node:$RHS))>>;
3239 defm USUBW : SIMDWideThreeVectorBHS< 1, 0b0011, "usubw",
3240 BinOpFrag<(sub node:$LHS, (zext node:$RHS))>>;
3242 // Additional patterns for SMULL and UMULL
3243 multiclass Neon_mul_widen_patterns<SDPatternOperator opnode,
3244 Instruction INST8B, Instruction INST4H, Instruction INST2S> {
3245 def : Pat<(v8i16 (opnode (v8i8 V64:$Rn), (v8i8 V64:$Rm))),
3246 (INST8B V64:$Rn, V64:$Rm)>;
3247 def : Pat<(v4i32 (opnode (v4i16 V64:$Rn), (v4i16 V64:$Rm))),
3248 (INST4H V64:$Rn, V64:$Rm)>;
3249 def : Pat<(v2i64 (opnode (v2i32 V64:$Rn), (v2i32 V64:$Rm))),
3250 (INST2S V64:$Rn, V64:$Rm)>;
3253 defm : Neon_mul_widen_patterns<AArch64smull, SMULLv8i8_v8i16,
3254 SMULLv4i16_v4i32, SMULLv2i32_v2i64>;
3255 defm : Neon_mul_widen_patterns<AArch64umull, UMULLv8i8_v8i16,
3256 UMULLv4i16_v4i32, UMULLv2i32_v2i64>;
3258 // Additional patterns for SMLAL/SMLSL and UMLAL/UMLSL
3259 multiclass Neon_mulacc_widen_patterns<SDPatternOperator opnode,
3260 Instruction INST8B, Instruction INST4H, Instruction INST2S> {
3261 def : Pat<(v8i16 (opnode (v8i16 V128:$Rd), (v8i8 V64:$Rn), (v8i8 V64:$Rm))),
3262 (INST8B V128:$Rd, V64:$Rn, V64:$Rm)>;
3263 def : Pat<(v4i32 (opnode (v4i32 V128:$Rd), (v4i16 V64:$Rn), (v4i16 V64:$Rm))),
3264 (INST4H V128:$Rd, V64:$Rn, V64:$Rm)>;
3265 def : Pat<(v2i64 (opnode (v2i64 V128:$Rd), (v2i32 V64:$Rn), (v2i32 V64:$Rm))),
3266 (INST2S V128:$Rd, V64:$Rn, V64:$Rm)>;
3269 defm : Neon_mulacc_widen_patterns<
3270 TriOpFrag<(add node:$LHS, (AArch64smull node:$MHS, node:$RHS))>,
3271 SMLALv8i8_v8i16, SMLALv4i16_v4i32, SMLALv2i32_v2i64>;
3272 defm : Neon_mulacc_widen_patterns<
3273 TriOpFrag<(add node:$LHS, (AArch64umull node:$MHS, node:$RHS))>,
3274 UMLALv8i8_v8i16, UMLALv4i16_v4i32, UMLALv2i32_v2i64>;
3275 defm : Neon_mulacc_widen_patterns<
3276 TriOpFrag<(sub node:$LHS, (AArch64smull node:$MHS, node:$RHS))>,
3277 SMLSLv8i8_v8i16, SMLSLv4i16_v4i32, SMLSLv2i32_v2i64>;
3278 defm : Neon_mulacc_widen_patterns<
3279 TriOpFrag<(sub node:$LHS, (AArch64umull node:$MHS, node:$RHS))>,
3280 UMLSLv8i8_v8i16, UMLSLv4i16_v4i32, UMLSLv2i32_v2i64>;
3282 // Patterns for 64-bit pmull
3283 def : Pat<(int_aarch64_neon_pmull64 V64:$Rn, V64:$Rm),
3284 (PMULLv1i64 V64:$Rn, V64:$Rm)>;
3285 def : Pat<(int_aarch64_neon_pmull64 (vector_extract (v2i64 V128:$Rn), (i64 1)),
3286 (vector_extract (v2i64 V128:$Rm), (i64 1))),
3287 (PMULLv2i64 V128:$Rn, V128:$Rm)>;
3289 // CodeGen patterns for addhn and subhn instructions, which can actually be
3290 // written in LLVM IR without too much difficulty.
3293 def : Pat<(v8i8 (trunc (v8i16 (AArch64vlshr (add V128:$Rn, V128:$Rm), (i32 8))))),
3294 (ADDHNv8i16_v8i8 V128:$Rn, V128:$Rm)>;
3295 def : Pat<(v4i16 (trunc (v4i32 (AArch64vlshr (add V128:$Rn, V128:$Rm),
3297 (ADDHNv4i32_v4i16 V128:$Rn, V128:$Rm)>;
3298 def : Pat<(v2i32 (trunc (v2i64 (AArch64vlshr (add V128:$Rn, V128:$Rm),
3300 (ADDHNv2i64_v2i32 V128:$Rn, V128:$Rm)>;
3301 def : Pat<(concat_vectors (v8i8 V64:$Rd),
3302 (trunc (v8i16 (AArch64vlshr (add V128:$Rn, V128:$Rm),
3304 (ADDHNv8i16_v16i8 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3305 V128:$Rn, V128:$Rm)>;
3306 def : Pat<(concat_vectors (v4i16 V64:$Rd),
3307 (trunc (v4i32 (AArch64vlshr (add V128:$Rn, V128:$Rm),
3309 (ADDHNv4i32_v8i16 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3310 V128:$Rn, V128:$Rm)>;
3311 def : Pat<(concat_vectors (v2i32 V64:$Rd),
3312 (trunc (v2i64 (AArch64vlshr (add V128:$Rn, V128:$Rm),
3314 (ADDHNv2i64_v4i32 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3315 V128:$Rn, V128:$Rm)>;
3318 def : Pat<(v8i8 (trunc (v8i16 (AArch64vlshr (sub V128:$Rn, V128:$Rm), (i32 8))))),
3319 (SUBHNv8i16_v8i8 V128:$Rn, V128:$Rm)>;
3320 def : Pat<(v4i16 (trunc (v4i32 (AArch64vlshr (sub V128:$Rn, V128:$Rm),
3322 (SUBHNv4i32_v4i16 V128:$Rn, V128:$Rm)>;
3323 def : Pat<(v2i32 (trunc (v2i64 (AArch64vlshr (sub V128:$Rn, V128:$Rm),
3325 (SUBHNv2i64_v2i32 V128:$Rn, V128:$Rm)>;
3326 def : Pat<(concat_vectors (v8i8 V64:$Rd),
3327 (trunc (v8i16 (AArch64vlshr (sub V128:$Rn, V128:$Rm),
3329 (SUBHNv8i16_v16i8 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3330 V128:$Rn, V128:$Rm)>;
3331 def : Pat<(concat_vectors (v4i16 V64:$Rd),
3332 (trunc (v4i32 (AArch64vlshr (sub V128:$Rn, V128:$Rm),
3334 (SUBHNv4i32_v8i16 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3335 V128:$Rn, V128:$Rm)>;
3336 def : Pat<(concat_vectors (v2i32 V64:$Rd),
3337 (trunc (v2i64 (AArch64vlshr (sub V128:$Rn, V128:$Rm),
3339 (SUBHNv2i64_v4i32 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3340 V128:$Rn, V128:$Rm)>;
3342 //----------------------------------------------------------------------------
3343 // AdvSIMD bitwise extract from vector instruction.
3344 //----------------------------------------------------------------------------
3346 defm EXT : SIMDBitwiseExtract<"ext">;
3348 def : Pat<(v4i16 (AArch64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
3349 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
3350 def : Pat<(v8i16 (AArch64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3351 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3352 def : Pat<(v2i32 (AArch64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
3353 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
3354 def : Pat<(v2f32 (AArch64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
3355 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
3356 def : Pat<(v4i32 (AArch64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3357 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3358 def : Pat<(v4f32 (AArch64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3359 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3360 def : Pat<(v2i64 (AArch64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3361 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3362 def : Pat<(v2f64 (AArch64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3363 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3364 def : Pat<(v4f16 (AArch64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
3365 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
3366 def : Pat<(v8f16 (AArch64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3367 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3369 // We use EXT to handle extract_subvector to copy the upper 64-bits of a
3371 def : Pat<(v8i8 (extract_subvector V128:$Rn, (i64 8))),
3372 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3373 def : Pat<(v4i16 (extract_subvector V128:$Rn, (i64 4))),
3374 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3375 def : Pat<(v2i32 (extract_subvector V128:$Rn, (i64 2))),
3376 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3377 def : Pat<(v1i64 (extract_subvector V128:$Rn, (i64 1))),
3378 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3379 def : Pat<(v4f16 (extract_subvector V128:$Rn, (i64 4))),
3380 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3381 def : Pat<(v2f32 (extract_subvector V128:$Rn, (i64 2))),
3382 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3383 def : Pat<(v1f64 (extract_subvector V128:$Rn, (i64 1))),
3384 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3387 //----------------------------------------------------------------------------
3388 // AdvSIMD zip vector
3389 //----------------------------------------------------------------------------
3391 defm TRN1 : SIMDZipVector<0b010, "trn1", AArch64trn1>;
3392 defm TRN2 : SIMDZipVector<0b110, "trn2", AArch64trn2>;
3393 defm UZP1 : SIMDZipVector<0b001, "uzp1", AArch64uzp1>;
3394 defm UZP2 : SIMDZipVector<0b101, "uzp2", AArch64uzp2>;
3395 defm ZIP1 : SIMDZipVector<0b011, "zip1", AArch64zip1>;
3396 defm ZIP2 : SIMDZipVector<0b111, "zip2", AArch64zip2>;
3398 //----------------------------------------------------------------------------
3399 // AdvSIMD TBL/TBX instructions
3400 //----------------------------------------------------------------------------
3402 defm TBL : SIMDTableLookup< 0, "tbl">;
3403 defm TBX : SIMDTableLookupTied<1, "tbx">;
3405 def : Pat<(v8i8 (int_aarch64_neon_tbl1 (v16i8 VecListOne128:$Rn), (v8i8 V64:$Ri))),
3406 (TBLv8i8One VecListOne128:$Rn, V64:$Ri)>;
3407 def : Pat<(v16i8 (int_aarch64_neon_tbl1 (v16i8 V128:$Ri), (v16i8 V128:$Rn))),
3408 (TBLv16i8One V128:$Ri, V128:$Rn)>;
3410 def : Pat<(v8i8 (int_aarch64_neon_tbx1 (v8i8 V64:$Rd),
3411 (v16i8 VecListOne128:$Rn), (v8i8 V64:$Ri))),
3412 (TBXv8i8One V64:$Rd, VecListOne128:$Rn, V64:$Ri)>;
3413 def : Pat<(v16i8 (int_aarch64_neon_tbx1 (v16i8 V128:$Rd),
3414 (v16i8 V128:$Ri), (v16i8 V128:$Rn))),
3415 (TBXv16i8One V128:$Rd, V128:$Ri, V128:$Rn)>;
3418 //----------------------------------------------------------------------------
3419 // AdvSIMD scalar CPY instruction
3420 //----------------------------------------------------------------------------
3422 defm CPY : SIMDScalarCPY<"cpy">;
3424 //----------------------------------------------------------------------------
3425 // AdvSIMD scalar pairwise instructions
3426 //----------------------------------------------------------------------------
3428 defm ADDP : SIMDPairwiseScalarD<0, 0b11011, "addp">;
3429 defm FADDP : SIMDPairwiseScalarSD<1, 0, 0b01101, "faddp">;
3430 defm FMAXNMP : SIMDPairwiseScalarSD<1, 0, 0b01100, "fmaxnmp">;
3431 defm FMAXP : SIMDPairwiseScalarSD<1, 0, 0b01111, "fmaxp">;
3432 defm FMINNMP : SIMDPairwiseScalarSD<1, 1, 0b01100, "fminnmp">;
3433 defm FMINP : SIMDPairwiseScalarSD<1, 1, 0b01111, "fminp">;
3434 def : Pat<(i64 (int_aarch64_neon_saddv (v2i64 V128:$Rn))),
3435 (ADDPv2i64p V128:$Rn)>;
3436 def : Pat<(i64 (int_aarch64_neon_uaddv (v2i64 V128:$Rn))),
3437 (ADDPv2i64p V128:$Rn)>;
3438 def : Pat<(f32 (int_aarch64_neon_faddv (v2f32 V64:$Rn))),
3439 (FADDPv2i32p V64:$Rn)>;
3440 def : Pat<(f32 (int_aarch64_neon_faddv (v4f32 V128:$Rn))),
3441 (FADDPv2i32p (EXTRACT_SUBREG (FADDPv4f32 V128:$Rn, V128:$Rn), dsub))>;
3442 def : Pat<(f64 (int_aarch64_neon_faddv (v2f64 V128:$Rn))),
3443 (FADDPv2i64p V128:$Rn)>;
3444 def : Pat<(f32 (int_aarch64_neon_fmaxnmv (v2f32 V64:$Rn))),
3445 (FMAXNMPv2i32p V64:$Rn)>;
3446 def : Pat<(f64 (int_aarch64_neon_fmaxnmv (v2f64 V128:$Rn))),
3447 (FMAXNMPv2i64p V128:$Rn)>;
3448 def : Pat<(f32 (int_aarch64_neon_fmaxv (v2f32 V64:$Rn))),
3449 (FMAXPv2i32p V64:$Rn)>;
3450 def : Pat<(f64 (int_aarch64_neon_fmaxv (v2f64 V128:$Rn))),
3451 (FMAXPv2i64p V128:$Rn)>;
3452 def : Pat<(f32 (int_aarch64_neon_fminnmv (v2f32 V64:$Rn))),
3453 (FMINNMPv2i32p V64:$Rn)>;
3454 def : Pat<(f64 (int_aarch64_neon_fminnmv (v2f64 V128:$Rn))),
3455 (FMINNMPv2i64p V128:$Rn)>;
3456 def : Pat<(f32 (int_aarch64_neon_fminv (v2f32 V64:$Rn))),
3457 (FMINPv2i32p V64:$Rn)>;
3458 def : Pat<(f64 (int_aarch64_neon_fminv (v2f64 V128:$Rn))),
3459 (FMINPv2i64p V128:$Rn)>;
3461 //----------------------------------------------------------------------------
3462 // AdvSIMD INS/DUP instructions
3463 //----------------------------------------------------------------------------
3465 def DUPv8i8gpr : SIMDDupFromMain<0, 0b00001, ".8b", v8i8, V64, GPR32>;
3466 def DUPv16i8gpr : SIMDDupFromMain<1, 0b00001, ".16b", v16i8, V128, GPR32>;
3467 def DUPv4i16gpr : SIMDDupFromMain<0, 0b00010, ".4h", v4i16, V64, GPR32>;
3468 def DUPv8i16gpr : SIMDDupFromMain<1, 0b00010, ".8h", v8i16, V128, GPR32>;
3469 def DUPv2i32gpr : SIMDDupFromMain<0, 0b00100, ".2s", v2i32, V64, GPR32>;
3470 def DUPv4i32gpr : SIMDDupFromMain<1, 0b00100, ".4s", v4i32, V128, GPR32>;
3471 def DUPv2i64gpr : SIMDDupFromMain<1, 0b01000, ".2d", v2i64, V128, GPR64>;
3473 def DUPv2i64lane : SIMDDup64FromElement;
3474 def DUPv2i32lane : SIMDDup32FromElement<0, ".2s", v2i32, V64>;
3475 def DUPv4i32lane : SIMDDup32FromElement<1, ".4s", v4i32, V128>;
3476 def DUPv4i16lane : SIMDDup16FromElement<0, ".4h", v4i16, V64>;
3477 def DUPv8i16lane : SIMDDup16FromElement<1, ".8h", v8i16, V128>;
3478 def DUPv8i8lane : SIMDDup8FromElement <0, ".8b", v8i8, V64>;
3479 def DUPv16i8lane : SIMDDup8FromElement <1, ".16b", v16i8, V128>;
3481 def : Pat<(v2f32 (AArch64dup (f32 FPR32:$Rn))),
3482 (v2f32 (DUPv2i32lane
3483 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rn, ssub),
3485 def : Pat<(v4f32 (AArch64dup (f32 FPR32:$Rn))),
3486 (v4f32 (DUPv4i32lane
3487 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rn, ssub),
3489 def : Pat<(v2f64 (AArch64dup (f64 FPR64:$Rn))),
3490 (v2f64 (DUPv2i64lane
3491 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$Rn, dsub),
3493 def : Pat<(v4f16 (AArch64dup (f16 FPR16:$Rn))),
3494 (v4f16 (DUPv4i16lane
3495 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), FPR16:$Rn, hsub),
3497 def : Pat<(v8f16 (AArch64dup (f16 FPR16:$Rn))),
3498 (v8f16 (DUPv8i16lane
3499 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), FPR16:$Rn, hsub),
3502 def : Pat<(v4f16 (AArch64duplane16 (v8f16 V128:$Rn), VectorIndexH:$imm)),
3503 (DUPv4i16lane V128:$Rn, VectorIndexH:$imm)>;
3504 def : Pat<(v8f16 (AArch64duplane16 (v8f16 V128:$Rn), VectorIndexH:$imm)),
3505 (DUPv8i16lane V128:$Rn, VectorIndexH:$imm)>;
3507 def : Pat<(v2f32 (AArch64duplane32 (v4f32 V128:$Rn), VectorIndexS:$imm)),
3508 (DUPv2i32lane V128:$Rn, VectorIndexS:$imm)>;
3509 def : Pat<(v4f32 (AArch64duplane32 (v4f32 V128:$Rn), VectorIndexS:$imm)),
3510 (DUPv4i32lane V128:$Rn, VectorIndexS:$imm)>;
3511 def : Pat<(v2f64 (AArch64duplane64 (v2f64 V128:$Rn), VectorIndexD:$imm)),
3512 (DUPv2i64lane V128:$Rn, VectorIndexD:$imm)>;
3514 // If there's an (AArch64dup (vector_extract ...) ...), we can use a duplane
3515 // instruction even if the types don't match: we just have to remap the lane
3516 // carefully. N.b. this trick only applies to truncations.
3517 def VecIndex_x2 : SDNodeXForm<imm, [{
3518 return CurDAG->getTargetConstant(2 * N->getZExtValue(), MVT::i64);
3520 def VecIndex_x4 : SDNodeXForm<imm, [{
3521 return CurDAG->getTargetConstant(4 * N->getZExtValue(), MVT::i64);
3523 def VecIndex_x8 : SDNodeXForm<imm, [{
3524 return CurDAG->getTargetConstant(8 * N->getZExtValue(), MVT::i64);
3527 multiclass DUPWithTruncPats<ValueType ResVT, ValueType Src64VT,
3528 ValueType Src128VT, ValueType ScalVT,
3529 Instruction DUP, SDNodeXForm IdxXFORM> {
3530 def : Pat<(ResVT (AArch64dup (ScalVT (vector_extract (Src128VT V128:$Rn),
3532 (DUP V128:$Rn, (IdxXFORM imm:$idx))>;
3534 def : Pat<(ResVT (AArch64dup (ScalVT (vector_extract (Src64VT V64:$Rn),
3536 (DUP (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), (IdxXFORM imm:$idx))>;
3539 defm : DUPWithTruncPats<v8i8, v4i16, v8i16, i32, DUPv8i8lane, VecIndex_x2>;
3540 defm : DUPWithTruncPats<v8i8, v2i32, v4i32, i32, DUPv8i8lane, VecIndex_x4>;
3541 defm : DUPWithTruncPats<v4i16, v2i32, v4i32, i32, DUPv4i16lane, VecIndex_x2>;
3543 defm : DUPWithTruncPats<v16i8, v4i16, v8i16, i32, DUPv16i8lane, VecIndex_x2>;
3544 defm : DUPWithTruncPats<v16i8, v2i32, v4i32, i32, DUPv16i8lane, VecIndex_x4>;
3545 defm : DUPWithTruncPats<v8i16, v2i32, v4i32, i32, DUPv8i16lane, VecIndex_x2>;
3547 multiclass DUPWithTrunci64Pats<ValueType ResVT, Instruction DUP,
3548 SDNodeXForm IdxXFORM> {
3549 def : Pat<(ResVT (AArch64dup (i32 (trunc (vector_extract (v2i64 V128:$Rn),
3551 (DUP V128:$Rn, (IdxXFORM imm:$idx))>;
3553 def : Pat<(ResVT (AArch64dup (i32 (trunc (vector_extract (v1i64 V64:$Rn),
3555 (DUP (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), (IdxXFORM imm:$idx))>;
3558 defm : DUPWithTrunci64Pats<v8i8, DUPv8i8lane, VecIndex_x8>;
3559 defm : DUPWithTrunci64Pats<v4i16, DUPv4i16lane, VecIndex_x4>;
3560 defm : DUPWithTrunci64Pats<v2i32, DUPv2i32lane, VecIndex_x2>;
3562 defm : DUPWithTrunci64Pats<v16i8, DUPv16i8lane, VecIndex_x8>;
3563 defm : DUPWithTrunci64Pats<v8i16, DUPv8i16lane, VecIndex_x4>;
3564 defm : DUPWithTrunci64Pats<v4i32, DUPv4i32lane, VecIndex_x2>;
3566 // SMOV and UMOV definitions, with some extra patterns for convenience
3570 def : Pat<(sext_inreg (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx), i8),
3571 (i32 (SMOVvi8to32 V128:$Rn, VectorIndexB:$idx))>;
3572 def : Pat<(sext_inreg (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx), i8),
3573 (i64 (SMOVvi8to64 V128:$Rn, VectorIndexB:$idx))>;
3574 def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
3575 (i32 (SMOVvi16to32 V128:$Rn, VectorIndexH:$idx))>;
3576 def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
3577 (i64 (SMOVvi16to64 V128:$Rn, VectorIndexH:$idx))>;
3578 def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
3579 (i32 (SMOVvi16to32 V128:$Rn, VectorIndexH:$idx))>;
3580 def : Pat<(sext (i32 (vector_extract (v4i32 V128:$Rn), VectorIndexS:$idx))),
3581 (i64 (SMOVvi32to64 V128:$Rn, VectorIndexS:$idx))>;
3583 // Extracting i8 or i16 elements will have the zero-extend transformed to
3584 // an 'and' mask by type legalization since neither i8 nor i16 are legal types
3585 // for AArch64. Match these patterns here since UMOV already zeroes out the high
3586 // bits of the destination register.
3587 def : Pat<(and (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx),
3589 (i32 (UMOVvi8 V128:$Rn, VectorIndexB:$idx))>;
3590 def : Pat<(and (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),
3592 (i32 (UMOVvi16 V128:$Rn, VectorIndexH:$idx))>;
3596 def : Pat<(v16i8 (scalar_to_vector GPR32:$Rn)),
3597 (SUBREG_TO_REG (i32 0),
3598 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3599 def : Pat<(v8i8 (scalar_to_vector GPR32:$Rn)),
3600 (SUBREG_TO_REG (i32 0),
3601 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3603 def : Pat<(v8i16 (scalar_to_vector GPR32:$Rn)),
3604 (SUBREG_TO_REG (i32 0),
3605 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3606 def : Pat<(v4i16 (scalar_to_vector GPR32:$Rn)),
3607 (SUBREG_TO_REG (i32 0),
3608 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3610 def : Pat<(v2i32 (scalar_to_vector (i32 FPR32:$Rn))),
3611 (v2i32 (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)),
3612 (i32 FPR32:$Rn), ssub))>;
3613 def : Pat<(v4i32 (scalar_to_vector (i32 FPR32:$Rn))),
3614 (v4i32 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
3615 (i32 FPR32:$Rn), ssub))>;
3616 def : Pat<(v2i64 (scalar_to_vector (i64 FPR64:$Rn))),
3617 (v2i64 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
3618 (i64 FPR64:$Rn), dsub))>;
3620 def : Pat<(v4f32 (scalar_to_vector (f32 FPR32:$Rn))),
3621 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rn, ssub)>;
3622 def : Pat<(v2f32 (scalar_to_vector (f32 FPR32:$Rn))),
3623 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), FPR32:$Rn, ssub)>;
3624 def : Pat<(v2f64 (scalar_to_vector (f64 FPR64:$Rn))),
3625 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$Rn, dsub)>;
3627 def : Pat<(v4f16 (vector_insert (v4f16 V64:$Rn),
3628 (f16 FPR16:$Rm), (i64 VectorIndexS:$imm))),
3631 (v8f16 (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), V64:$Rn, dsub)),
3633 (v8f16 (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), FPR16:$Rm, hsub)),
3637 def : Pat<(v8f16 (vector_insert (v8f16 V128:$Rn),
3638 (f16 FPR16:$Rm), (i64 VectorIndexH:$imm))),
3640 V128:$Rn, VectorIndexH:$imm,
3641 (v8f16 (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), FPR16:$Rm, hsub)),
3644 def : Pat<(v2f32 (vector_insert (v2f32 V64:$Rn),
3645 (f32 FPR32:$Rm), (i64 VectorIndexS:$imm))),
3648 (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), V64:$Rn, dsub)),
3650 (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rm, ssub)),
3653 def : Pat<(v4f32 (vector_insert (v4f32 V128:$Rn),
3654 (f32 FPR32:$Rm), (i64 VectorIndexS:$imm))),
3656 V128:$Rn, VectorIndexS:$imm,
3657 (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rm, ssub)),
3659 def : Pat<(v2f64 (vector_insert (v2f64 V128:$Rn),
3660 (f64 FPR64:$Rm), (i64 VectorIndexD:$imm))),
3662 V128:$Rn, VectorIndexD:$imm,
3663 (v2f64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$Rm, dsub)),
3666 // Copy an element at a constant index in one vector into a constant indexed
3667 // element of another.
3668 // FIXME refactor to a shared class/dev parameterized on vector type, vector
3669 // index type and INS extension
3670 def : Pat<(v16i8 (int_aarch64_neon_vcopy_lane
3671 (v16i8 V128:$Vd), VectorIndexB:$idx, (v16i8 V128:$Vs),
3672 VectorIndexB:$idx2)),
3674 V128:$Vd, VectorIndexB:$idx, V128:$Vs, VectorIndexB:$idx2)
3676 def : Pat<(v8i16 (int_aarch64_neon_vcopy_lane
3677 (v8i16 V128:$Vd), VectorIndexH:$idx, (v8i16 V128:$Vs),
3678 VectorIndexH:$idx2)),
3680 V128:$Vd, VectorIndexH:$idx, V128:$Vs, VectorIndexH:$idx2)
3682 def : Pat<(v4i32 (int_aarch64_neon_vcopy_lane
3683 (v4i32 V128:$Vd), VectorIndexS:$idx, (v4i32 V128:$Vs),
3684 VectorIndexS:$idx2)),
3686 V128:$Vd, VectorIndexS:$idx, V128:$Vs, VectorIndexS:$idx2)
3688 def : Pat<(v2i64 (int_aarch64_neon_vcopy_lane
3689 (v2i64 V128:$Vd), VectorIndexD:$idx, (v2i64 V128:$Vs),
3690 VectorIndexD:$idx2)),
3692 V128:$Vd, VectorIndexD:$idx, V128:$Vs, VectorIndexD:$idx2)
3695 multiclass Neon_INS_elt_pattern<ValueType VT128, ValueType VT64,
3696 ValueType VTScal, Instruction INS> {
3697 def : Pat<(VT128 (vector_insert V128:$src,
3698 (VTScal (vector_extract (VT128 V128:$Rn), imm:$Immn)),
3700 (INS V128:$src, imm:$Immd, V128:$Rn, imm:$Immn)>;
3702 def : Pat<(VT128 (vector_insert V128:$src,
3703 (VTScal (vector_extract (VT64 V64:$Rn), imm:$Immn)),
3705 (INS V128:$src, imm:$Immd,
3706 (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), imm:$Immn)>;
3708 def : Pat<(VT64 (vector_insert V64:$src,
3709 (VTScal (vector_extract (VT128 V128:$Rn), imm:$Immn)),
3711 (EXTRACT_SUBREG (INS (SUBREG_TO_REG (i64 0), V64:$src, dsub),
3712 imm:$Immd, V128:$Rn, imm:$Immn),
3715 def : Pat<(VT64 (vector_insert V64:$src,
3716 (VTScal (vector_extract (VT64 V64:$Rn), imm:$Immn)),
3719 (INS (SUBREG_TO_REG (i64 0), V64:$src, dsub), imm:$Immd,
3720 (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), imm:$Immn),
3724 defm : Neon_INS_elt_pattern<v8f16, v4f16, f16, INSvi16lane>;
3725 defm : Neon_INS_elt_pattern<v4f32, v2f32, f32, INSvi32lane>;
3726 defm : Neon_INS_elt_pattern<v2f64, v1f64, f64, INSvi64lane>;
3727 defm : Neon_INS_elt_pattern<v16i8, v8i8, i32, INSvi8lane>;
3728 defm : Neon_INS_elt_pattern<v8i16, v4i16, i32, INSvi16lane>;
3729 defm : Neon_INS_elt_pattern<v4i32, v2i32, i32, INSvi32lane>;
3730 defm : Neon_INS_elt_pattern<v2i64, v1i64, i64, INSvi32lane>;
3733 // Floating point vector extractions are codegen'd as either a sequence of
3734 // subregister extractions, or a MOV (aka CPY here, alias for DUP) if
3735 // the lane number is anything other than zero.
3736 def : Pat<(vector_extract (v2f64 V128:$Rn), 0),
3737 (f64 (EXTRACT_SUBREG V128:$Rn, dsub))>;
3738 def : Pat<(vector_extract (v4f32 V128:$Rn), 0),
3739 (f32 (EXTRACT_SUBREG V128:$Rn, ssub))>;
3740 def : Pat<(vector_extract (v8f16 V128:$Rn), 0),
3741 (f16 (EXTRACT_SUBREG V128:$Rn, hsub))>;
3743 def : Pat<(vector_extract (v2f64 V128:$Rn), VectorIndexD:$idx),
3744 (f64 (CPYi64 V128:$Rn, VectorIndexD:$idx))>;
3745 def : Pat<(vector_extract (v4f32 V128:$Rn), VectorIndexS:$idx),
3746 (f32 (CPYi32 V128:$Rn, VectorIndexS:$idx))>;
3747 def : Pat<(vector_extract (v8f16 V128:$Rn), VectorIndexH:$idx),
3748 (f16 (CPYi16 V128:$Rn, VectorIndexH:$idx))>;
3750 // All concat_vectors operations are canonicalised to act on i64 vectors for
3751 // AArch64. In the general case we need an instruction, which had just as well be
3753 class ConcatPat<ValueType DstTy, ValueType SrcTy>
3754 : Pat<(DstTy (concat_vectors (SrcTy V64:$Rd), V64:$Rn)),
3755 (INSvi64lane (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), 1,
3756 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rn, dsub), 0)>;
3758 def : ConcatPat<v2i64, v1i64>;
3759 def : ConcatPat<v2f64, v1f64>;
3760 def : ConcatPat<v4i32, v2i32>;
3761 def : ConcatPat<v4f32, v2f32>;
3762 def : ConcatPat<v8i16, v4i16>;
3763 def : ConcatPat<v8f16, v4f16>;
3764 def : ConcatPat<v16i8, v8i8>;
3766 // If the high lanes are undef, though, we can just ignore them:
3767 class ConcatUndefPat<ValueType DstTy, ValueType SrcTy>
3768 : Pat<(DstTy (concat_vectors (SrcTy V64:$Rn), undef)),
3769 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rn, dsub)>;
3771 def : ConcatUndefPat<v2i64, v1i64>;
3772 def : ConcatUndefPat<v2f64, v1f64>;
3773 def : ConcatUndefPat<v4i32, v2i32>;
3774 def : ConcatUndefPat<v4f32, v2f32>;
3775 def : ConcatUndefPat<v8i16, v4i16>;
3776 def : ConcatUndefPat<v16i8, v8i8>;
3778 //----------------------------------------------------------------------------
3779 // AdvSIMD across lanes instructions
3780 //----------------------------------------------------------------------------
3782 defm ADDV : SIMDAcrossLanesBHS<0, 0b11011, "addv">;
3783 defm SMAXV : SIMDAcrossLanesBHS<0, 0b01010, "smaxv">;
3784 defm SMINV : SIMDAcrossLanesBHS<0, 0b11010, "sminv">;
3785 defm UMAXV : SIMDAcrossLanesBHS<1, 0b01010, "umaxv">;
3786 defm UMINV : SIMDAcrossLanesBHS<1, 0b11010, "uminv">;
3787 defm SADDLV : SIMDAcrossLanesHSD<0, 0b00011, "saddlv">;
3788 defm UADDLV : SIMDAcrossLanesHSD<1, 0b00011, "uaddlv">;
3789 defm FMAXNMV : SIMDAcrossLanesS<0b01100, 0, "fmaxnmv", int_aarch64_neon_fmaxnmv>;
3790 defm FMAXV : SIMDAcrossLanesS<0b01111, 0, "fmaxv", int_aarch64_neon_fmaxv>;
3791 defm FMINNMV : SIMDAcrossLanesS<0b01100, 1, "fminnmv", int_aarch64_neon_fminnmv>;
3792 defm FMINV : SIMDAcrossLanesS<0b01111, 1, "fminv", int_aarch64_neon_fminv>;
3794 multiclass SIMDAcrossLanesSignedIntrinsic<string baseOpc, Intrinsic intOp> {
3795 // If there is a sign extension after this intrinsic, consume it as smov already
3797 def : Pat<(i32 (sext_inreg (i32 (intOp (v8i8 V64:$Rn))), i8)),
3799 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3800 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3802 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
3804 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3805 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3807 // If there is a sign extension after this intrinsic, consume it as smov already
3809 def : Pat<(i32 (sext_inreg (i32 (intOp (v16i8 V128:$Rn))), i8)),
3811 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3812 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3814 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
3816 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3817 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3819 // If there is a sign extension after this intrinsic, consume it as smov already
3821 def : Pat<(i32 (sext_inreg (i32 (intOp (v4i16 V64:$Rn))), i16)),
3823 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3824 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3826 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
3828 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3829 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3831 // If there is a sign extension after this intrinsic, consume it as smov already
3833 def : Pat<(i32 (sext_inreg (i32 (intOp (v8i16 V128:$Rn))), i16)),
3835 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3836 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3838 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
3840 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3841 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3844 def : Pat<(i32 (intOp (v4i32 V128:$Rn))),
3845 (i32 (EXTRACT_SUBREG
3846 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3847 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), ssub),
3851 multiclass SIMDAcrossLanesUnsignedIntrinsic<string baseOpc, Intrinsic intOp> {
3852 // If there is a masking operation keeping only what has been actually
3853 // generated, consume it.
3854 def : Pat<(i32 (and (i32 (intOp (v8i8 V64:$Rn))), maski8_or_more)),
3855 (i32 (EXTRACT_SUBREG
3856 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3857 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3859 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
3860 (i32 (EXTRACT_SUBREG
3861 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3862 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3864 // If there is a masking operation keeping only what has been actually
3865 // generated, consume it.
3866 def : Pat<(i32 (and (i32 (intOp (v16i8 V128:$Rn))), maski8_or_more)),
3867 (i32 (EXTRACT_SUBREG
3868 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3869 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3871 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
3872 (i32 (EXTRACT_SUBREG
3873 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3874 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3877 // If there is a masking operation keeping only what has been actually
3878 // generated, consume it.
3879 def : Pat<(i32 (and (i32 (intOp (v4i16 V64:$Rn))), maski16_or_more)),
3880 (i32 (EXTRACT_SUBREG
3881 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3882 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3884 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
3885 (i32 (EXTRACT_SUBREG
3886 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3887 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3889 // If there is a masking operation keeping only what has been actually
3890 // generated, consume it.
3891 def : Pat<(i32 (and (i32 (intOp (v8i16 V128:$Rn))), maski16_or_more)),
3892 (i32 (EXTRACT_SUBREG
3893 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3894 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3896 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
3897 (i32 (EXTRACT_SUBREG
3898 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3899 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3902 def : Pat<(i32 (intOp (v4i32 V128:$Rn))),
3903 (i32 (EXTRACT_SUBREG
3904 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3905 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), ssub),
3910 multiclass SIMDAcrossLanesSignedLongIntrinsic<string baseOpc, Intrinsic intOp> {
3911 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
3913 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3914 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), hsub),
3916 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
3918 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3919 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), hsub),
3922 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
3923 (i32 (EXTRACT_SUBREG
3924 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3925 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), ssub),
3927 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
3928 (i32 (EXTRACT_SUBREG
3929 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3930 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), ssub),
3933 def : Pat<(i64 (intOp (v4i32 V128:$Rn))),
3934 (i64 (EXTRACT_SUBREG
3935 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3936 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), dsub),
3940 multiclass SIMDAcrossLanesUnsignedLongIntrinsic<string baseOpc,
3942 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
3943 (i32 (EXTRACT_SUBREG
3944 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3945 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), hsub),
3947 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
3948 (i32 (EXTRACT_SUBREG
3949 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3950 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), hsub),
3953 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
3954 (i32 (EXTRACT_SUBREG
3955 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3956 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), ssub),
3958 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
3959 (i32 (EXTRACT_SUBREG
3960 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3961 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), ssub),
3964 def : Pat<(i64 (intOp (v4i32 V128:$Rn))),
3965 (i64 (EXTRACT_SUBREG
3966 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3967 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), dsub),
3971 defm : SIMDAcrossLanesSignedIntrinsic<"ADDV", int_aarch64_neon_saddv>;
3972 // vaddv_[su]32 is special; -> ADDP Vd.2S,Vn.2S,Vm.2S; return Vd.s[0];Vn==Vm
3973 def : Pat<(i32 (int_aarch64_neon_saddv (v2i32 V64:$Rn))),
3974 (EXTRACT_SUBREG (ADDPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3976 defm : SIMDAcrossLanesUnsignedIntrinsic<"ADDV", int_aarch64_neon_uaddv>;
3977 // vaddv_[su]32 is special; -> ADDP Vd.2S,Vn.2S,Vm.2S; return Vd.s[0];Vn==Vm
3978 def : Pat<(i32 (int_aarch64_neon_uaddv (v2i32 V64:$Rn))),
3979 (EXTRACT_SUBREG (ADDPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3981 defm : SIMDAcrossLanesSignedIntrinsic<"SMAXV", int_aarch64_neon_smaxv>;
3982 def : Pat<(i32 (int_aarch64_neon_smaxv (v2i32 V64:$Rn))),
3983 (EXTRACT_SUBREG (SMAXPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3985 defm : SIMDAcrossLanesSignedIntrinsic<"SMINV", int_aarch64_neon_sminv>;
3986 def : Pat<(i32 (int_aarch64_neon_sminv (v2i32 V64:$Rn))),
3987 (EXTRACT_SUBREG (SMINPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3989 defm : SIMDAcrossLanesUnsignedIntrinsic<"UMAXV", int_aarch64_neon_umaxv>;
3990 def : Pat<(i32 (int_aarch64_neon_umaxv (v2i32 V64:$Rn))),
3991 (EXTRACT_SUBREG (UMAXPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3993 defm : SIMDAcrossLanesUnsignedIntrinsic<"UMINV", int_aarch64_neon_uminv>;
3994 def : Pat<(i32 (int_aarch64_neon_uminv (v2i32 V64:$Rn))),
3995 (EXTRACT_SUBREG (UMINPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3997 defm : SIMDAcrossLanesSignedLongIntrinsic<"SADDLV", int_aarch64_neon_saddlv>;
3998 defm : SIMDAcrossLanesUnsignedLongIntrinsic<"UADDLV", int_aarch64_neon_uaddlv>;
4000 // The vaddlv_s32 intrinsic gets mapped to SADDLP.
4001 def : Pat<(i64 (int_aarch64_neon_saddlv (v2i32 V64:$Rn))),
4002 (i64 (EXTRACT_SUBREG
4003 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4004 (SADDLPv2i32_v1i64 V64:$Rn), dsub),
4006 // The vaddlv_u32 intrinsic gets mapped to UADDLP.
4007 def : Pat<(i64 (int_aarch64_neon_uaddlv (v2i32 V64:$Rn))),
4008 (i64 (EXTRACT_SUBREG
4009 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4010 (UADDLPv2i32_v1i64 V64:$Rn), dsub),
4013 //------------------------------------------------------------------------------
4014 // AdvSIMD modified immediate instructions
4015 //------------------------------------------------------------------------------
4018 defm BIC : SIMDModifiedImmVectorShiftTied<1, 0b11, 0b01, "bic", AArch64bici>;
4020 defm ORR : SIMDModifiedImmVectorShiftTied<0, 0b11, 0b01, "orr", AArch64orri>;
4022 def : InstAlias<"bic $Vd.4h, $imm", (BICv4i16 V64:$Vd, imm0_255:$imm, 0)>;
4023 def : InstAlias<"bic $Vd.8h, $imm", (BICv8i16 V128:$Vd, imm0_255:$imm, 0)>;
4024 def : InstAlias<"bic $Vd.2s, $imm", (BICv2i32 V64:$Vd, imm0_255:$imm, 0)>;
4025 def : InstAlias<"bic $Vd.4s, $imm", (BICv4i32 V128:$Vd, imm0_255:$imm, 0)>;
4027 def : InstAlias<"bic.4h $Vd, $imm", (BICv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
4028 def : InstAlias<"bic.8h $Vd, $imm", (BICv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
4029 def : InstAlias<"bic.2s $Vd, $imm", (BICv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
4030 def : InstAlias<"bic.4s $Vd, $imm", (BICv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
4032 def : InstAlias<"orr $Vd.4h, $imm", (ORRv4i16 V64:$Vd, imm0_255:$imm, 0)>;
4033 def : InstAlias<"orr $Vd.8h, $imm", (ORRv8i16 V128:$Vd, imm0_255:$imm, 0)>;
4034 def : InstAlias<"orr $Vd.2s, $imm", (ORRv2i32 V64:$Vd, imm0_255:$imm, 0)>;
4035 def : InstAlias<"orr $Vd.4s, $imm", (ORRv4i32 V128:$Vd, imm0_255:$imm, 0)>;
4037 def : InstAlias<"orr.4h $Vd, $imm", (ORRv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
4038 def : InstAlias<"orr.8h $Vd, $imm", (ORRv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
4039 def : InstAlias<"orr.2s $Vd, $imm", (ORRv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
4040 def : InstAlias<"orr.4s $Vd, $imm", (ORRv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
4043 def FMOVv2f64_ns : SIMDModifiedImmVectorNoShift<1, 1, 0b1111, V128, fpimm8,
4045 [(set (v2f64 V128:$Rd), (AArch64fmov imm0_255:$imm8))]>;
4046 def FMOVv2f32_ns : SIMDModifiedImmVectorNoShift<0, 0, 0b1111, V64, fpimm8,
4048 [(set (v2f32 V64:$Rd), (AArch64fmov imm0_255:$imm8))]>;
4049 def FMOVv4f32_ns : SIMDModifiedImmVectorNoShift<1, 0, 0b1111, V128, fpimm8,
4051 [(set (v4f32 V128:$Rd), (AArch64fmov imm0_255:$imm8))]>;
4055 // EDIT byte mask: scalar
4056 let isReMaterializable = 1, isAsCheapAsAMove = 1 in
4057 def MOVID : SIMDModifiedImmScalarNoShift<0, 1, 0b1110, "movi",
4058 [(set FPR64:$Rd, simdimmtype10:$imm8)]>;
4059 // The movi_edit node has the immediate value already encoded, so we use
4060 // a plain imm0_255 here.
4061 def : Pat<(f64 (AArch64movi_edit imm0_255:$shift)),
4062 (MOVID imm0_255:$shift)>;
4064 def : Pat<(v1i64 immAllZerosV), (MOVID (i32 0))>;
4065 def : Pat<(v2i32 immAllZerosV), (MOVID (i32 0))>;
4066 def : Pat<(v4i16 immAllZerosV), (MOVID (i32 0))>;
4067 def : Pat<(v8i8 immAllZerosV), (MOVID (i32 0))>;
4069 def : Pat<(v1i64 immAllOnesV), (MOVID (i32 255))>;
4070 def : Pat<(v2i32 immAllOnesV), (MOVID (i32 255))>;
4071 def : Pat<(v4i16 immAllOnesV), (MOVID (i32 255))>;
4072 def : Pat<(v8i8 immAllOnesV), (MOVID (i32 255))>;
4074 // EDIT byte mask: 2d
4076 // The movi_edit node has the immediate value already encoded, so we use
4077 // a plain imm0_255 in the pattern
4078 let isReMaterializable = 1, isAsCheapAsAMove = 1 in
4079 def MOVIv2d_ns : SIMDModifiedImmVectorNoShift<1, 1, 0b1110, V128,
4082 [(set (v2i64 V128:$Rd), (AArch64movi_edit imm0_255:$imm8))]>;
4085 // Use movi.2d to materialize 0.0 if the HW does zero-cycle zeroing.
4086 // Complexity is added to break a tie with a plain MOVI.
4087 let AddedComplexity = 1 in {
4088 def : Pat<(f32 fpimm0),
4089 (f32 (EXTRACT_SUBREG (v2i64 (MOVIv2d_ns (i32 0))), ssub))>,
4091 def : Pat<(f64 fpimm0),
4092 (f64 (EXTRACT_SUBREG (v2i64 (MOVIv2d_ns (i32 0))), dsub))>,
4096 def : Pat<(v2i64 immAllZerosV), (MOVIv2d_ns (i32 0))>;
4097 def : Pat<(v4i32 immAllZerosV), (MOVIv2d_ns (i32 0))>;
4098 def : Pat<(v8i16 immAllZerosV), (MOVIv2d_ns (i32 0))>;
4099 def : Pat<(v16i8 immAllZerosV), (MOVIv2d_ns (i32 0))>;
4101 def : Pat<(v2i64 immAllOnesV), (MOVIv2d_ns (i32 255))>;
4102 def : Pat<(v4i32 immAllOnesV), (MOVIv2d_ns (i32 255))>;
4103 def : Pat<(v8i16 immAllOnesV), (MOVIv2d_ns (i32 255))>;
4104 def : Pat<(v16i8 immAllOnesV), (MOVIv2d_ns (i32 255))>;
4106 def : Pat<(v2f64 (AArch64dup (f64 fpimm0))), (MOVIv2d_ns (i32 0))>;
4107 def : Pat<(v4f32 (AArch64dup (f32 fpimm0))), (MOVIv2d_ns (i32 0))>;
4109 // EDIT per word & halfword: 2s, 4h, 4s, & 8h
4110 defm MOVI : SIMDModifiedImmVectorShift<0, 0b10, 0b00, "movi">;
4112 def : InstAlias<"movi $Vd.4h, $imm", (MOVIv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
4113 def : InstAlias<"movi $Vd.8h, $imm", (MOVIv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
4114 def : InstAlias<"movi $Vd.2s, $imm", (MOVIv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
4115 def : InstAlias<"movi $Vd.4s, $imm", (MOVIv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
4117 def : InstAlias<"movi.4h $Vd, $imm", (MOVIv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
4118 def : InstAlias<"movi.8h $Vd, $imm", (MOVIv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
4119 def : InstAlias<"movi.2s $Vd, $imm", (MOVIv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
4120 def : InstAlias<"movi.4s $Vd, $imm", (MOVIv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
4122 def : Pat<(v2i32 (AArch64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
4123 (MOVIv2i32 imm0_255:$imm8, imm:$shift)>;
4124 def : Pat<(v4i32 (AArch64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
4125 (MOVIv4i32 imm0_255:$imm8, imm:$shift)>;
4126 def : Pat<(v4i16 (AArch64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
4127 (MOVIv4i16 imm0_255:$imm8, imm:$shift)>;
4128 def : Pat<(v8i16 (AArch64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
4129 (MOVIv8i16 imm0_255:$imm8, imm:$shift)>;
4131 // EDIT per word: 2s & 4s with MSL shifter
4132 def MOVIv2s_msl : SIMDModifiedImmMoveMSL<0, 0, {1,1,0,?}, V64, "movi", ".2s",
4133 [(set (v2i32 V64:$Rd),
4134 (AArch64movi_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
4135 def MOVIv4s_msl : SIMDModifiedImmMoveMSL<1, 0, {1,1,0,?}, V128, "movi", ".4s",
4136 [(set (v4i32 V128:$Rd),
4137 (AArch64movi_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
4139 // Per byte: 8b & 16b
4140 def MOVIv8b_ns : SIMDModifiedImmVectorNoShift<0, 0, 0b1110, V64, imm0_255,
4142 [(set (v8i8 V64:$Rd), (AArch64movi imm0_255:$imm8))]>;
4143 def MOVIv16b_ns : SIMDModifiedImmVectorNoShift<1, 0, 0b1110, V128, imm0_255,
4145 [(set (v16i8 V128:$Rd), (AArch64movi imm0_255:$imm8))]>;
4149 // EDIT per word & halfword: 2s, 4h, 4s, & 8h
4150 defm MVNI : SIMDModifiedImmVectorShift<1, 0b10, 0b00, "mvni">;
4152 def : InstAlias<"mvni $Vd.4h, $imm", (MVNIv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
4153 def : InstAlias<"mvni $Vd.8h, $imm", (MVNIv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
4154 def : InstAlias<"mvni $Vd.2s, $imm", (MVNIv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
4155 def : InstAlias<"mvni $Vd.4s, $imm", (MVNIv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
4157 def : InstAlias<"mvni.4h $Vd, $imm", (MVNIv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
4158 def : InstAlias<"mvni.8h $Vd, $imm", (MVNIv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
4159 def : InstAlias<"mvni.2s $Vd, $imm", (MVNIv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
4160 def : InstAlias<"mvni.4s $Vd, $imm", (MVNIv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
4162 def : Pat<(v2i32 (AArch64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
4163 (MVNIv2i32 imm0_255:$imm8, imm:$shift)>;
4164 def : Pat<(v4i32 (AArch64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
4165 (MVNIv4i32 imm0_255:$imm8, imm:$shift)>;
4166 def : Pat<(v4i16 (AArch64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
4167 (MVNIv4i16 imm0_255:$imm8, imm:$shift)>;
4168 def : Pat<(v8i16 (AArch64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
4169 (MVNIv8i16 imm0_255:$imm8, imm:$shift)>;
4171 // EDIT per word: 2s & 4s with MSL shifter
4172 def MVNIv2s_msl : SIMDModifiedImmMoveMSL<0, 1, {1,1,0,?}, V64, "mvni", ".2s",
4173 [(set (v2i32 V64:$Rd),
4174 (AArch64mvni_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
4175 def MVNIv4s_msl : SIMDModifiedImmMoveMSL<1, 1, {1,1,0,?}, V128, "mvni", ".4s",
4176 [(set (v4i32 V128:$Rd),
4177 (AArch64mvni_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
4179 //----------------------------------------------------------------------------
4180 // AdvSIMD indexed element
4181 //----------------------------------------------------------------------------
4183 let hasSideEffects = 0 in {
4184 defm FMLA : SIMDFPIndexedSDTied<0, 0b0001, "fmla">;
4185 defm FMLS : SIMDFPIndexedSDTied<0, 0b0101, "fmls">;
4188 // NOTE: Operands are reordered in the FMLA/FMLS PatFrags because the
4189 // instruction expects the addend first, while the intrinsic expects it last.
4191 // On the other hand, there are quite a few valid combinatorial options due to
4192 // the commutativity of multiplication and the fact that (-x) * y = x * (-y).
4193 defm : SIMDFPIndexedSDTiedPatterns<"FMLA",
4194 TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)>>;
4195 defm : SIMDFPIndexedSDTiedPatterns<"FMLA",
4196 TriOpFrag<(fma node:$MHS, node:$RHS, node:$LHS)>>;
4198 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
4199 TriOpFrag<(fma node:$MHS, (fneg node:$RHS), node:$LHS)> >;
4200 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
4201 TriOpFrag<(fma node:$RHS, (fneg node:$MHS), node:$LHS)> >;
4202 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
4203 TriOpFrag<(fma (fneg node:$RHS), node:$MHS, node:$LHS)> >;
4204 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
4205 TriOpFrag<(fma (fneg node:$MHS), node:$RHS, node:$LHS)> >;
4207 multiclass FMLSIndexedAfterNegPatterns<SDPatternOperator OpNode> {
4208 // 3 variants for the .2s version: DUPLANE from 128-bit, DUPLANE from 64-bit
4210 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
4211 (AArch64duplane32 (v4f32 (fneg V128:$Rm)),
4212 VectorIndexS:$idx))),
4213 (FMLSv2i32_indexed V64:$Rd, V64:$Rn, V128:$Rm, VectorIndexS:$idx)>;
4214 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
4215 (v2f32 (AArch64duplane32
4216 (v4f32 (insert_subvector undef,
4217 (v2f32 (fneg V64:$Rm)),
4219 VectorIndexS:$idx)))),
4220 (FMLSv2i32_indexed V64:$Rd, V64:$Rn,
4221 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub),
4222 VectorIndexS:$idx)>;
4223 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
4224 (AArch64dup (f32 (fneg FPR32Op:$Rm))))),
4225 (FMLSv2i32_indexed V64:$Rd, V64:$Rn,
4226 (SUBREG_TO_REG (i32 0), FPR32Op:$Rm, ssub), (i64 0))>;
4228 // 3 variants for the .4s version: DUPLANE from 128-bit, DUPLANE from 64-bit
4230 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
4231 (AArch64duplane32 (v4f32 (fneg V128:$Rm)),
4232 VectorIndexS:$idx))),
4233 (FMLSv4i32_indexed V128:$Rd, V128:$Rn, V128:$Rm,
4234 VectorIndexS:$idx)>;
4235 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
4236 (v4f32 (AArch64duplane32
4237 (v4f32 (insert_subvector undef,
4238 (v2f32 (fneg V64:$Rm)),
4240 VectorIndexS:$idx)))),
4241 (FMLSv4i32_indexed V128:$Rd, V128:$Rn,
4242 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub),
4243 VectorIndexS:$idx)>;
4244 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
4245 (AArch64dup (f32 (fneg FPR32Op:$Rm))))),
4246 (FMLSv4i32_indexed V128:$Rd, V128:$Rn,
4247 (SUBREG_TO_REG (i32 0), FPR32Op:$Rm, ssub), (i64 0))>;
4249 // 2 variants for the .2d version: DUPLANE from 128-bit, and DUP scalar
4250 // (DUPLANE from 64-bit would be trivial).
4251 def : Pat<(v2f64 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn),
4252 (AArch64duplane64 (v2f64 (fneg V128:$Rm)),
4253 VectorIndexD:$idx))),
4255 V128:$Rd, V128:$Rn, V128:$Rm, VectorIndexS:$idx)>;
4256 def : Pat<(v2f64 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn),
4257 (AArch64dup (f64 (fneg FPR64Op:$Rm))))),
4258 (FMLSv2i64_indexed V128:$Rd, V128:$Rn,
4259 (SUBREG_TO_REG (i32 0), FPR64Op:$Rm, dsub), (i64 0))>;
4261 // 2 variants for 32-bit scalar version: extract from .2s or from .4s
4262 def : Pat<(f32 (OpNode (f32 FPR32:$Rd), (f32 FPR32:$Rn),
4263 (vector_extract (v4f32 (fneg V128:$Rm)),
4264 VectorIndexS:$idx))),
4265 (FMLSv1i32_indexed FPR32:$Rd, FPR32:$Rn,
4266 V128:$Rm, VectorIndexS:$idx)>;
4267 def : Pat<(f32 (OpNode (f32 FPR32:$Rd), (f32 FPR32:$Rn),
4268 (vector_extract (v2f32 (fneg V64:$Rm)),
4269 VectorIndexS:$idx))),
4270 (FMLSv1i32_indexed FPR32:$Rd, FPR32:$Rn,
4271 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub), VectorIndexS:$idx)>;
4273 // 1 variant for 64-bit scalar version: extract from .1d or from .2d
4274 def : Pat<(f64 (OpNode (f64 FPR64:$Rd), (f64 FPR64:$Rn),
4275 (vector_extract (v2f64 (fneg V128:$Rm)),
4276 VectorIndexS:$idx))),
4277 (FMLSv1i64_indexed FPR64:$Rd, FPR64:$Rn,
4278 V128:$Rm, VectorIndexS:$idx)>;
4281 defm : FMLSIndexedAfterNegPatterns<
4282 TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)> >;
4283 defm : FMLSIndexedAfterNegPatterns<
4284 TriOpFrag<(fma node:$MHS, node:$RHS, node:$LHS)> >;
4286 defm FMULX : SIMDFPIndexedSD<1, 0b1001, "fmulx", int_aarch64_neon_fmulx>;
4287 defm FMUL : SIMDFPIndexedSD<0, 0b1001, "fmul", fmul>;
4289 def : Pat<(v2f32 (fmul V64:$Rn, (AArch64dup (f32 FPR32:$Rm)))),
4290 (FMULv2i32_indexed V64:$Rn,
4291 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rm, ssub),
4293 def : Pat<(v4f32 (fmul V128:$Rn, (AArch64dup (f32 FPR32:$Rm)))),
4294 (FMULv4i32_indexed V128:$Rn,
4295 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rm, ssub),
4297 def : Pat<(v2f64 (fmul V128:$Rn, (AArch64dup (f64 FPR64:$Rm)))),
4298 (FMULv2i64_indexed V128:$Rn,
4299 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$Rm, dsub),
4302 defm SQDMULH : SIMDIndexedHS<0, 0b1100, "sqdmulh", int_aarch64_neon_sqdmulh>;
4303 defm SQRDMULH : SIMDIndexedHS<0, 0b1101, "sqrdmulh", int_aarch64_neon_sqrdmulh>;
4304 defm MLA : SIMDVectorIndexedHSTied<1, 0b0000, "mla",
4305 TriOpFrag<(add node:$LHS, (mul node:$MHS, node:$RHS))>>;
4306 defm MLS : SIMDVectorIndexedHSTied<1, 0b0100, "mls",
4307 TriOpFrag<(sub node:$LHS, (mul node:$MHS, node:$RHS))>>;
4308 defm MUL : SIMDVectorIndexedHS<0, 0b1000, "mul", mul>;
4309 defm SMLAL : SIMDVectorIndexedLongSDTied<0, 0b0010, "smlal",
4310 TriOpFrag<(add node:$LHS, (int_aarch64_neon_smull node:$MHS, node:$RHS))>>;
4311 defm SMLSL : SIMDVectorIndexedLongSDTied<0, 0b0110, "smlsl",
4312 TriOpFrag<(sub node:$LHS, (int_aarch64_neon_smull node:$MHS, node:$RHS))>>;
4313 defm SMULL : SIMDVectorIndexedLongSD<0, 0b1010, "smull",
4314 int_aarch64_neon_smull>;
4315 defm SQDMLAL : SIMDIndexedLongSQDMLXSDTied<0, 0b0011, "sqdmlal",
4316 int_aarch64_neon_sqadd>;
4317 defm SQDMLSL : SIMDIndexedLongSQDMLXSDTied<0, 0b0111, "sqdmlsl",
4318 int_aarch64_neon_sqsub>;
4319 defm SQDMULL : SIMDIndexedLongSD<0, 0b1011, "sqdmull", int_aarch64_neon_sqdmull>;
4320 defm UMLAL : SIMDVectorIndexedLongSDTied<1, 0b0010, "umlal",
4321 TriOpFrag<(add node:$LHS, (int_aarch64_neon_umull node:$MHS, node:$RHS))>>;
4322 defm UMLSL : SIMDVectorIndexedLongSDTied<1, 0b0110, "umlsl",
4323 TriOpFrag<(sub node:$LHS, (int_aarch64_neon_umull node:$MHS, node:$RHS))>>;
4324 defm UMULL : SIMDVectorIndexedLongSD<1, 0b1010, "umull",
4325 int_aarch64_neon_umull>;
4327 // A scalar sqdmull with the second operand being a vector lane can be
4328 // handled directly with the indexed instruction encoding.
4329 def : Pat<(int_aarch64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
4330 (vector_extract (v4i32 V128:$Vm),
4331 VectorIndexS:$idx)),
4332 (SQDMULLv1i64_indexed FPR32:$Rn, V128:$Vm, VectorIndexS:$idx)>;
4334 //----------------------------------------------------------------------------
4335 // AdvSIMD scalar shift instructions
4336 //----------------------------------------------------------------------------
4337 defm FCVTZS : SIMDScalarRShiftSD<0, 0b11111, "fcvtzs">;
4338 defm FCVTZU : SIMDScalarRShiftSD<1, 0b11111, "fcvtzu">;
4339 defm SCVTF : SIMDScalarRShiftSD<0, 0b11100, "scvtf">;
4340 defm UCVTF : SIMDScalarRShiftSD<1, 0b11100, "ucvtf">;
4341 // Codegen patterns for the above. We don't put these directly on the
4342 // instructions because TableGen's type inference can't handle the truth.
4343 // Having the same base pattern for fp <--> int totally freaks it out.
4344 def : Pat<(int_aarch64_neon_vcvtfp2fxs FPR32:$Rn, vecshiftR32:$imm),
4345 (FCVTZSs FPR32:$Rn, vecshiftR32:$imm)>;
4346 def : Pat<(int_aarch64_neon_vcvtfp2fxu FPR32:$Rn, vecshiftR32:$imm),
4347 (FCVTZUs FPR32:$Rn, vecshiftR32:$imm)>;
4348 def : Pat<(i64 (int_aarch64_neon_vcvtfp2fxs (f64 FPR64:$Rn), vecshiftR64:$imm)),
4349 (FCVTZSd FPR64:$Rn, vecshiftR64:$imm)>;
4350 def : Pat<(i64 (int_aarch64_neon_vcvtfp2fxu (f64 FPR64:$Rn), vecshiftR64:$imm)),
4351 (FCVTZUd FPR64:$Rn, vecshiftR64:$imm)>;
4352 def : Pat<(v1i64 (int_aarch64_neon_vcvtfp2fxs (v1f64 FPR64:$Rn),
4354 (FCVTZSd FPR64:$Rn, vecshiftR64:$imm)>;
4355 def : Pat<(v1i64 (int_aarch64_neon_vcvtfp2fxu (v1f64 FPR64:$Rn),
4357 (FCVTZUd FPR64:$Rn, vecshiftR64:$imm)>;
4358 def : Pat<(int_aarch64_neon_vcvtfxs2fp FPR32:$Rn, vecshiftR32:$imm),
4359 (SCVTFs FPR32:$Rn, vecshiftR32:$imm)>;
4360 def : Pat<(int_aarch64_neon_vcvtfxu2fp FPR32:$Rn, vecshiftR32:$imm),
4361 (UCVTFs FPR32:$Rn, vecshiftR32:$imm)>;
4362 def : Pat<(f64 (int_aarch64_neon_vcvtfxs2fp (i64 FPR64:$Rn), vecshiftR64:$imm)),
4363 (SCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
4364 def : Pat<(f64 (int_aarch64_neon_vcvtfxu2fp (i64 FPR64:$Rn), vecshiftR64:$imm)),
4365 (UCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
4366 def : Pat<(v1f64 (int_aarch64_neon_vcvtfxs2fp (v1i64 FPR64:$Rn),
4368 (SCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
4369 def : Pat<(v1f64 (int_aarch64_neon_vcvtfxu2fp (v1i64 FPR64:$Rn),
4371 (UCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
4373 defm SHL : SIMDScalarLShiftD< 0, 0b01010, "shl", AArch64vshl>;
4374 defm SLI : SIMDScalarLShiftDTied<1, 0b01010, "sli">;
4375 defm SQRSHRN : SIMDScalarRShiftBHS< 0, 0b10011, "sqrshrn",
4376 int_aarch64_neon_sqrshrn>;
4377 defm SQRSHRUN : SIMDScalarRShiftBHS< 1, 0b10001, "sqrshrun",
4378 int_aarch64_neon_sqrshrun>;
4379 defm SQSHLU : SIMDScalarLShiftBHSD<1, 0b01100, "sqshlu", AArch64sqshlui>;
4380 defm SQSHL : SIMDScalarLShiftBHSD<0, 0b01110, "sqshl", AArch64sqshli>;
4381 defm SQSHRN : SIMDScalarRShiftBHS< 0, 0b10010, "sqshrn",
4382 int_aarch64_neon_sqshrn>;
4383 defm SQSHRUN : SIMDScalarRShiftBHS< 1, 0b10000, "sqshrun",
4384 int_aarch64_neon_sqshrun>;
4385 defm SRI : SIMDScalarRShiftDTied< 1, 0b01000, "sri">;
4386 defm SRSHR : SIMDScalarRShiftD< 0, 0b00100, "srshr", AArch64srshri>;
4387 defm SRSRA : SIMDScalarRShiftDTied< 0, 0b00110, "srsra",
4388 TriOpFrag<(add node:$LHS,
4389 (AArch64srshri node:$MHS, node:$RHS))>>;
4390 defm SSHR : SIMDScalarRShiftD< 0, 0b00000, "sshr", AArch64vashr>;
4391 defm SSRA : SIMDScalarRShiftDTied< 0, 0b00010, "ssra",
4392 TriOpFrag<(add node:$LHS,
4393 (AArch64vashr node:$MHS, node:$RHS))>>;
4394 defm UQRSHRN : SIMDScalarRShiftBHS< 1, 0b10011, "uqrshrn",
4395 int_aarch64_neon_uqrshrn>;
4396 defm UQSHL : SIMDScalarLShiftBHSD<1, 0b01110, "uqshl", AArch64uqshli>;
4397 defm UQSHRN : SIMDScalarRShiftBHS< 1, 0b10010, "uqshrn",
4398 int_aarch64_neon_uqshrn>;
4399 defm URSHR : SIMDScalarRShiftD< 1, 0b00100, "urshr", AArch64urshri>;
4400 defm URSRA : SIMDScalarRShiftDTied< 1, 0b00110, "ursra",
4401 TriOpFrag<(add node:$LHS,
4402 (AArch64urshri node:$MHS, node:$RHS))>>;
4403 defm USHR : SIMDScalarRShiftD< 1, 0b00000, "ushr", AArch64vlshr>;
4404 defm USRA : SIMDScalarRShiftDTied< 1, 0b00010, "usra",
4405 TriOpFrag<(add node:$LHS,
4406 (AArch64vlshr node:$MHS, node:$RHS))>>;
4408 //----------------------------------------------------------------------------
4409 // AdvSIMD vector shift instructions
4410 //----------------------------------------------------------------------------
4411 defm FCVTZS:SIMDVectorRShiftSD<0, 0b11111, "fcvtzs", int_aarch64_neon_vcvtfp2fxs>;
4412 defm FCVTZU:SIMDVectorRShiftSD<1, 0b11111, "fcvtzu", int_aarch64_neon_vcvtfp2fxu>;
4413 defm SCVTF: SIMDVectorRShiftSDToFP<0, 0b11100, "scvtf",
4414 int_aarch64_neon_vcvtfxs2fp>;
4415 defm RSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10001, "rshrn",
4416 int_aarch64_neon_rshrn>;
4417 defm SHL : SIMDVectorLShiftBHSD<0, 0b01010, "shl", AArch64vshl>;
4418 defm SHRN : SIMDVectorRShiftNarrowBHS<0, 0b10000, "shrn",
4419 BinOpFrag<(trunc (AArch64vashr node:$LHS, node:$RHS))>>;
4420 defm SLI : SIMDVectorLShiftBHSDTied<1, 0b01010, "sli", int_aarch64_neon_vsli>;
4421 def : Pat<(v1i64 (int_aarch64_neon_vsli (v1i64 FPR64:$Rd), (v1i64 FPR64:$Rn),
4422 (i32 vecshiftL64:$imm))),
4423 (SLId FPR64:$Rd, FPR64:$Rn, vecshiftL64:$imm)>;
4424 defm SQRSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10011, "sqrshrn",
4425 int_aarch64_neon_sqrshrn>;
4426 defm SQRSHRUN: SIMDVectorRShiftNarrowBHS<1, 0b10001, "sqrshrun",
4427 int_aarch64_neon_sqrshrun>;
4428 defm SQSHLU : SIMDVectorLShiftBHSD<1, 0b01100, "sqshlu", AArch64sqshlui>;
4429 defm SQSHL : SIMDVectorLShiftBHSD<0, 0b01110, "sqshl", AArch64sqshli>;
4430 defm SQSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10010, "sqshrn",
4431 int_aarch64_neon_sqshrn>;
4432 defm SQSHRUN : SIMDVectorRShiftNarrowBHS<1, 0b10000, "sqshrun",
4433 int_aarch64_neon_sqshrun>;
4434 defm SRI : SIMDVectorRShiftBHSDTied<1, 0b01000, "sri", int_aarch64_neon_vsri>;
4435 def : Pat<(v1i64 (int_aarch64_neon_vsri (v1i64 FPR64:$Rd), (v1i64 FPR64:$Rn),
4436 (i32 vecshiftR64:$imm))),
4437 (SRId FPR64:$Rd, FPR64:$Rn, vecshiftR64:$imm)>;
4438 defm SRSHR : SIMDVectorRShiftBHSD<0, 0b00100, "srshr", AArch64srshri>;
4439 defm SRSRA : SIMDVectorRShiftBHSDTied<0, 0b00110, "srsra",
4440 TriOpFrag<(add node:$LHS,
4441 (AArch64srshri node:$MHS, node:$RHS))> >;
4442 defm SSHLL : SIMDVectorLShiftLongBHSD<0, 0b10100, "sshll",
4443 BinOpFrag<(AArch64vshl (sext node:$LHS), node:$RHS)>>;
4445 defm SSHR : SIMDVectorRShiftBHSD<0, 0b00000, "sshr", AArch64vashr>;
4446 defm SSRA : SIMDVectorRShiftBHSDTied<0, 0b00010, "ssra",
4447 TriOpFrag<(add node:$LHS, (AArch64vashr node:$MHS, node:$RHS))>>;
4448 defm UCVTF : SIMDVectorRShiftSDToFP<1, 0b11100, "ucvtf",
4449 int_aarch64_neon_vcvtfxu2fp>;
4450 defm UQRSHRN : SIMDVectorRShiftNarrowBHS<1, 0b10011, "uqrshrn",
4451 int_aarch64_neon_uqrshrn>;
4452 defm UQSHL : SIMDVectorLShiftBHSD<1, 0b01110, "uqshl", AArch64uqshli>;
4453 defm UQSHRN : SIMDVectorRShiftNarrowBHS<1, 0b10010, "uqshrn",
4454 int_aarch64_neon_uqshrn>;
4455 defm URSHR : SIMDVectorRShiftBHSD<1, 0b00100, "urshr", AArch64urshri>;
4456 defm URSRA : SIMDVectorRShiftBHSDTied<1, 0b00110, "ursra",
4457 TriOpFrag<(add node:$LHS,
4458 (AArch64urshri node:$MHS, node:$RHS))> >;
4459 defm USHLL : SIMDVectorLShiftLongBHSD<1, 0b10100, "ushll",
4460 BinOpFrag<(AArch64vshl (zext node:$LHS), node:$RHS)>>;
4461 defm USHR : SIMDVectorRShiftBHSD<1, 0b00000, "ushr", AArch64vlshr>;
4462 defm USRA : SIMDVectorRShiftBHSDTied<1, 0b00010, "usra",
4463 TriOpFrag<(add node:$LHS, (AArch64vlshr node:$MHS, node:$RHS))> >;
4465 // SHRN patterns for when a logical right shift was used instead of arithmetic
4466 // (the immediate guarantees no sign bits actually end up in the result so it
4468 def : Pat<(v8i8 (trunc (AArch64vlshr (v8i16 V128:$Rn), vecshiftR16Narrow:$imm))),
4469 (SHRNv8i8_shift V128:$Rn, vecshiftR16Narrow:$imm)>;
4470 def : Pat<(v4i16 (trunc (AArch64vlshr (v4i32 V128:$Rn), vecshiftR32Narrow:$imm))),
4471 (SHRNv4i16_shift V128:$Rn, vecshiftR32Narrow:$imm)>;
4472 def : Pat<(v2i32 (trunc (AArch64vlshr (v2i64 V128:$Rn), vecshiftR64Narrow:$imm))),
4473 (SHRNv2i32_shift V128:$Rn, vecshiftR64Narrow:$imm)>;
4475 def : Pat<(v16i8 (concat_vectors (v8i8 V64:$Rd),
4476 (trunc (AArch64vlshr (v8i16 V128:$Rn),
4477 vecshiftR16Narrow:$imm)))),
4478 (SHRNv16i8_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
4479 V128:$Rn, vecshiftR16Narrow:$imm)>;
4480 def : Pat<(v8i16 (concat_vectors (v4i16 V64:$Rd),
4481 (trunc (AArch64vlshr (v4i32 V128:$Rn),
4482 vecshiftR32Narrow:$imm)))),
4483 (SHRNv8i16_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
4484 V128:$Rn, vecshiftR32Narrow:$imm)>;
4485 def : Pat<(v4i32 (concat_vectors (v2i32 V64:$Rd),
4486 (trunc (AArch64vlshr (v2i64 V128:$Rn),
4487 vecshiftR64Narrow:$imm)))),
4488 (SHRNv4i32_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
4489 V128:$Rn, vecshiftR32Narrow:$imm)>;
4491 // Vector sign and zero extensions are implemented with SSHLL and USSHLL.
4492 // Anyexts are implemented as zexts.
4493 def : Pat<(v8i16 (sext (v8i8 V64:$Rn))), (SSHLLv8i8_shift V64:$Rn, (i32 0))>;
4494 def : Pat<(v8i16 (zext (v8i8 V64:$Rn))), (USHLLv8i8_shift V64:$Rn, (i32 0))>;
4495 def : Pat<(v8i16 (anyext (v8i8 V64:$Rn))), (USHLLv8i8_shift V64:$Rn, (i32 0))>;
4496 def : Pat<(v4i32 (sext (v4i16 V64:$Rn))), (SSHLLv4i16_shift V64:$Rn, (i32 0))>;
4497 def : Pat<(v4i32 (zext (v4i16 V64:$Rn))), (USHLLv4i16_shift V64:$Rn, (i32 0))>;
4498 def : Pat<(v4i32 (anyext (v4i16 V64:$Rn))), (USHLLv4i16_shift V64:$Rn, (i32 0))>;
4499 def : Pat<(v2i64 (sext (v2i32 V64:$Rn))), (SSHLLv2i32_shift V64:$Rn, (i32 0))>;
4500 def : Pat<(v2i64 (zext (v2i32 V64:$Rn))), (USHLLv2i32_shift V64:$Rn, (i32 0))>;
4501 def : Pat<(v2i64 (anyext (v2i32 V64:$Rn))), (USHLLv2i32_shift V64:$Rn, (i32 0))>;
4502 // Also match an extend from the upper half of a 128 bit source register.
4503 def : Pat<(v8i16 (anyext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
4504 (USHLLv16i8_shift V128:$Rn, (i32 0))>;
4505 def : Pat<(v8i16 (zext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
4506 (USHLLv16i8_shift V128:$Rn, (i32 0))>;
4507 def : Pat<(v8i16 (sext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
4508 (SSHLLv16i8_shift V128:$Rn, (i32 0))>;
4509 def : Pat<(v4i32 (anyext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
4510 (USHLLv8i16_shift V128:$Rn, (i32 0))>;
4511 def : Pat<(v4i32 (zext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
4512 (USHLLv8i16_shift V128:$Rn, (i32 0))>;
4513 def : Pat<(v4i32 (sext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
4514 (SSHLLv8i16_shift V128:$Rn, (i32 0))>;
4515 def : Pat<(v2i64 (anyext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
4516 (USHLLv4i32_shift V128:$Rn, (i32 0))>;
4517 def : Pat<(v2i64 (zext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
4518 (USHLLv4i32_shift V128:$Rn, (i32 0))>;
4519 def : Pat<(v2i64 (sext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
4520 (SSHLLv4i32_shift V128:$Rn, (i32 0))>;
4522 // Vector shift sxtl aliases
4523 def : InstAlias<"sxtl.8h $dst, $src1",
4524 (SSHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
4525 def : InstAlias<"sxtl $dst.8h, $src1.8b",
4526 (SSHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
4527 def : InstAlias<"sxtl.4s $dst, $src1",
4528 (SSHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
4529 def : InstAlias<"sxtl $dst.4s, $src1.4h",
4530 (SSHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
4531 def : InstAlias<"sxtl.2d $dst, $src1",
4532 (SSHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
4533 def : InstAlias<"sxtl $dst.2d, $src1.2s",
4534 (SSHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
4536 // Vector shift sxtl2 aliases
4537 def : InstAlias<"sxtl2.8h $dst, $src1",
4538 (SSHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
4539 def : InstAlias<"sxtl2 $dst.8h, $src1.16b",
4540 (SSHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
4541 def : InstAlias<"sxtl2.4s $dst, $src1",
4542 (SSHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
4543 def : InstAlias<"sxtl2 $dst.4s, $src1.8h",
4544 (SSHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
4545 def : InstAlias<"sxtl2.2d $dst, $src1",
4546 (SSHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
4547 def : InstAlias<"sxtl2 $dst.2d, $src1.4s",
4548 (SSHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
4550 // Vector shift uxtl aliases
4551 def : InstAlias<"uxtl.8h $dst, $src1",
4552 (USHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
4553 def : InstAlias<"uxtl $dst.8h, $src1.8b",
4554 (USHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
4555 def : InstAlias<"uxtl.4s $dst, $src1",
4556 (USHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
4557 def : InstAlias<"uxtl $dst.4s, $src1.4h",
4558 (USHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
4559 def : InstAlias<"uxtl.2d $dst, $src1",
4560 (USHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
4561 def : InstAlias<"uxtl $dst.2d, $src1.2s",
4562 (USHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
4564 // Vector shift uxtl2 aliases
4565 def : InstAlias<"uxtl2.8h $dst, $src1",
4566 (USHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
4567 def : InstAlias<"uxtl2 $dst.8h, $src1.16b",
4568 (USHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
4569 def : InstAlias<"uxtl2.4s $dst, $src1",
4570 (USHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
4571 def : InstAlias<"uxtl2 $dst.4s, $src1.8h",
4572 (USHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
4573 def : InstAlias<"uxtl2.2d $dst, $src1",
4574 (USHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
4575 def : InstAlias<"uxtl2 $dst.2d, $src1.4s",
4576 (USHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
4578 // If an integer is about to be converted to a floating point value,
4579 // just load it on the floating point unit.
4580 // These patterns are more complex because floating point loads do not
4581 // support sign extension.
4582 // The sign extension has to be explicitly added and is only supported for
4583 // one step: byte-to-half, half-to-word, word-to-doubleword.
4584 // SCVTF GPR -> FPR is 9 cycles.
4585 // SCVTF FPR -> FPR is 4 cyclces.
4586 // (sign extension with lengthen) SXTL FPR -> FPR is 2 cycles.
4587 // Therefore, we can do 2 sign extensions and one SCVTF FPR -> FPR
4588 // and still being faster.
4589 // However, this is not good for code size.
4590 // 8-bits -> float. 2 sizes step-up.
4591 class SExtLoadi8CVTf32Pat<dag addrmode, dag INST>
4592 : Pat<(f32 (sint_to_fp (i32 (sextloadi8 addrmode)))),
4593 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
4598 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4604 ssub)))>, Requires<[NotForCodeSize, IsCyclone]>;
4606 def : SExtLoadi8CVTf32Pat<(ro8.Wpat GPR64sp:$Rn, GPR32:$Rm, ro8.Wext:$ext),
4607 (LDRBroW GPR64sp:$Rn, GPR32:$Rm, ro8.Wext:$ext)>;
4608 def : SExtLoadi8CVTf32Pat<(ro8.Xpat GPR64sp:$Rn, GPR64:$Rm, ro8.Xext:$ext),
4609 (LDRBroX GPR64sp:$Rn, GPR64:$Rm, ro8.Xext:$ext)>;
4610 def : SExtLoadi8CVTf32Pat<(am_indexed8 GPR64sp:$Rn, uimm12s1:$offset),
4611 (LDRBui GPR64sp:$Rn, uimm12s1:$offset)>;
4612 def : SExtLoadi8CVTf32Pat<(am_unscaled8 GPR64sp:$Rn, simm9:$offset),
4613 (LDURBi GPR64sp:$Rn, simm9:$offset)>;
4615 // 16-bits -> float. 1 size step-up.
4616 class SExtLoadi16CVTf32Pat<dag addrmode, dag INST>
4617 : Pat<(f32 (sint_to_fp (i32 (sextloadi16 addrmode)))),
4618 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
4620 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4624 ssub)))>, Requires<[NotForCodeSize]>;
4626 def : SExtLoadi16CVTf32Pat<(ro16.Wpat GPR64sp:$Rn, GPR32:$Rm, ro16.Wext:$ext),
4627 (LDRHroW GPR64sp:$Rn, GPR32:$Rm, ro16.Wext:$ext)>;
4628 def : SExtLoadi16CVTf32Pat<(ro16.Xpat GPR64sp:$Rn, GPR64:$Rm, ro16.Xext:$ext),
4629 (LDRHroX GPR64sp:$Rn, GPR64:$Rm, ro16.Xext:$ext)>;
4630 def : SExtLoadi16CVTf32Pat<(am_indexed16 GPR64sp:$Rn, uimm12s2:$offset),
4631 (LDRHui GPR64sp:$Rn, uimm12s2:$offset)>;
4632 def : SExtLoadi16CVTf32Pat<(am_unscaled16 GPR64sp:$Rn, simm9:$offset),
4633 (LDURHi GPR64sp:$Rn, simm9:$offset)>;
4635 // 32-bits to 32-bits are handled in target specific dag combine:
4636 // performIntToFpCombine.
4637 // 64-bits integer to 32-bits floating point, not possible with
4638 // SCVTF on floating point registers (both source and destination
4639 // must have the same size).
4641 // Here are the patterns for 8, 16, 32, and 64-bits to double.
4642 // 8-bits -> double. 3 size step-up: give up.
4643 // 16-bits -> double. 2 size step.
4644 class SExtLoadi16CVTf64Pat<dag addrmode, dag INST>
4645 : Pat <(f64 (sint_to_fp (i32 (sextloadi16 addrmode)))),
4646 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4651 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4657 dsub)))>, Requires<[NotForCodeSize, IsCyclone]>;
4659 def : SExtLoadi16CVTf64Pat<(ro16.Wpat GPR64sp:$Rn, GPR32:$Rm, ro16.Wext:$ext),
4660 (LDRHroW GPR64sp:$Rn, GPR32:$Rm, ro16.Wext:$ext)>;
4661 def : SExtLoadi16CVTf64Pat<(ro16.Xpat GPR64sp:$Rn, GPR64:$Rm, ro16.Xext:$ext),
4662 (LDRHroX GPR64sp:$Rn, GPR64:$Rm, ro16.Xext:$ext)>;
4663 def : SExtLoadi16CVTf64Pat<(am_indexed16 GPR64sp:$Rn, uimm12s2:$offset),
4664 (LDRHui GPR64sp:$Rn, uimm12s2:$offset)>;
4665 def : SExtLoadi16CVTf64Pat<(am_unscaled16 GPR64sp:$Rn, simm9:$offset),
4666 (LDURHi GPR64sp:$Rn, simm9:$offset)>;
4667 // 32-bits -> double. 1 size step-up.
4668 class SExtLoadi32CVTf64Pat<dag addrmode, dag INST>
4669 : Pat <(f64 (sint_to_fp (i32 (load addrmode)))),
4670 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4672 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4676 dsub)))>, Requires<[NotForCodeSize]>;
4678 def : SExtLoadi32CVTf64Pat<(ro32.Wpat GPR64sp:$Rn, GPR32:$Rm, ro32.Wext:$ext),
4679 (LDRSroW GPR64sp:$Rn, GPR32:$Rm, ro32.Wext:$ext)>;
4680 def : SExtLoadi32CVTf64Pat<(ro32.Xpat GPR64sp:$Rn, GPR64:$Rm, ro32.Xext:$ext),
4681 (LDRSroX GPR64sp:$Rn, GPR64:$Rm, ro32.Xext:$ext)>;
4682 def : SExtLoadi32CVTf64Pat<(am_indexed32 GPR64sp:$Rn, uimm12s4:$offset),
4683 (LDRSui GPR64sp:$Rn, uimm12s4:$offset)>;
4684 def : SExtLoadi32CVTf64Pat<(am_unscaled32 GPR64sp:$Rn, simm9:$offset),
4685 (LDURSi GPR64sp:$Rn, simm9:$offset)>;
4687 // 64-bits -> double are handled in target specific dag combine:
4688 // performIntToFpCombine.
4691 //----------------------------------------------------------------------------
4692 // AdvSIMD Load-Store Structure
4693 //----------------------------------------------------------------------------
4694 defm LD1 : SIMDLd1Multiple<"ld1">;
4695 defm LD2 : SIMDLd2Multiple<"ld2">;
4696 defm LD3 : SIMDLd3Multiple<"ld3">;
4697 defm LD4 : SIMDLd4Multiple<"ld4">;
4699 defm ST1 : SIMDSt1Multiple<"st1">;
4700 defm ST2 : SIMDSt2Multiple<"st2">;
4701 defm ST3 : SIMDSt3Multiple<"st3">;
4702 defm ST4 : SIMDSt4Multiple<"st4">;
4704 class Ld1Pat<ValueType ty, Instruction INST>
4705 : Pat<(ty (load GPR64sp:$Rn)), (INST GPR64sp:$Rn)>;
4707 def : Ld1Pat<v16i8, LD1Onev16b>;
4708 def : Ld1Pat<v8i16, LD1Onev8h>;
4709 def : Ld1Pat<v4i32, LD1Onev4s>;
4710 def : Ld1Pat<v2i64, LD1Onev2d>;
4711 def : Ld1Pat<v8i8, LD1Onev8b>;
4712 def : Ld1Pat<v4i16, LD1Onev4h>;
4713 def : Ld1Pat<v2i32, LD1Onev2s>;
4714 def : Ld1Pat<v1i64, LD1Onev1d>;
4716 class St1Pat<ValueType ty, Instruction INST>
4717 : Pat<(store ty:$Vt, GPR64sp:$Rn),
4718 (INST ty:$Vt, GPR64sp:$Rn)>;
4720 def : St1Pat<v16i8, ST1Onev16b>;
4721 def : St1Pat<v8i16, ST1Onev8h>;
4722 def : St1Pat<v4i32, ST1Onev4s>;
4723 def : St1Pat<v2i64, ST1Onev2d>;
4724 def : St1Pat<v8i8, ST1Onev8b>;
4725 def : St1Pat<v4i16, ST1Onev4h>;
4726 def : St1Pat<v2i32, ST1Onev2s>;
4727 def : St1Pat<v1i64, ST1Onev1d>;
4733 defm LD1R : SIMDLdR<0, 0b110, 0, "ld1r", "One", 1, 2, 4, 8>;
4734 defm LD2R : SIMDLdR<1, 0b110, 0, "ld2r", "Two", 2, 4, 8, 16>;
4735 defm LD3R : SIMDLdR<0, 0b111, 0, "ld3r", "Three", 3, 6, 12, 24>;
4736 defm LD4R : SIMDLdR<1, 0b111, 0, "ld4r", "Four", 4, 8, 16, 32>;
4737 let mayLoad = 1, hasSideEffects = 0 in {
4738 defm LD1 : SIMDLdSingleBTied<0, 0b000, "ld1", VecListOneb, GPR64pi1>;
4739 defm LD1 : SIMDLdSingleHTied<0, 0b010, 0, "ld1", VecListOneh, GPR64pi2>;
4740 defm LD1 : SIMDLdSingleSTied<0, 0b100, 0b00, "ld1", VecListOnes, GPR64pi4>;
4741 defm LD1 : SIMDLdSingleDTied<0, 0b100, 0b01, "ld1", VecListOned, GPR64pi8>;
4742 defm LD2 : SIMDLdSingleBTied<1, 0b000, "ld2", VecListTwob, GPR64pi2>;
4743 defm LD2 : SIMDLdSingleHTied<1, 0b010, 0, "ld2", VecListTwoh, GPR64pi4>;
4744 defm LD2 : SIMDLdSingleSTied<1, 0b100, 0b00, "ld2", VecListTwos, GPR64pi8>;
4745 defm LD2 : SIMDLdSingleDTied<1, 0b100, 0b01, "ld2", VecListTwod, GPR64pi16>;
4746 defm LD3 : SIMDLdSingleBTied<0, 0b001, "ld3", VecListThreeb, GPR64pi3>;
4747 defm LD3 : SIMDLdSingleHTied<0, 0b011, 0, "ld3", VecListThreeh, GPR64pi6>;
4748 defm LD3 : SIMDLdSingleSTied<0, 0b101, 0b00, "ld3", VecListThrees, GPR64pi12>;
4749 defm LD3 : SIMDLdSingleDTied<0, 0b101, 0b01, "ld3", VecListThreed, GPR64pi24>;
4750 defm LD4 : SIMDLdSingleBTied<1, 0b001, "ld4", VecListFourb, GPR64pi4>;
4751 defm LD4 : SIMDLdSingleHTied<1, 0b011, 0, "ld4", VecListFourh, GPR64pi8>;
4752 defm LD4 : SIMDLdSingleSTied<1, 0b101, 0b00, "ld4", VecListFours, GPR64pi16>;
4753 defm LD4 : SIMDLdSingleDTied<1, 0b101, 0b01, "ld4", VecListFourd, GPR64pi32>;
4756 def : Pat<(v8i8 (AArch64dup (i32 (extloadi8 GPR64sp:$Rn)))),
4757 (LD1Rv8b GPR64sp:$Rn)>;
4758 def : Pat<(v16i8 (AArch64dup (i32 (extloadi8 GPR64sp:$Rn)))),
4759 (LD1Rv16b GPR64sp:$Rn)>;
4760 def : Pat<(v4i16 (AArch64dup (i32 (extloadi16 GPR64sp:$Rn)))),
4761 (LD1Rv4h GPR64sp:$Rn)>;
4762 def : Pat<(v8i16 (AArch64dup (i32 (extloadi16 GPR64sp:$Rn)))),
4763 (LD1Rv8h GPR64sp:$Rn)>;
4764 def : Pat<(v2i32 (AArch64dup (i32 (load GPR64sp:$Rn)))),
4765 (LD1Rv2s GPR64sp:$Rn)>;
4766 def : Pat<(v4i32 (AArch64dup (i32 (load GPR64sp:$Rn)))),
4767 (LD1Rv4s GPR64sp:$Rn)>;
4768 def : Pat<(v2i64 (AArch64dup (i64 (load GPR64sp:$Rn)))),
4769 (LD1Rv2d GPR64sp:$Rn)>;
4770 def : Pat<(v1i64 (AArch64dup (i64 (load GPR64sp:$Rn)))),
4771 (LD1Rv1d GPR64sp:$Rn)>;
4772 // Grab the floating point version too
4773 def : Pat<(v2f32 (AArch64dup (f32 (load GPR64sp:$Rn)))),
4774 (LD1Rv2s GPR64sp:$Rn)>;
4775 def : Pat<(v4f32 (AArch64dup (f32 (load GPR64sp:$Rn)))),
4776 (LD1Rv4s GPR64sp:$Rn)>;
4777 def : Pat<(v2f64 (AArch64dup (f64 (load GPR64sp:$Rn)))),
4778 (LD1Rv2d GPR64sp:$Rn)>;
4779 def : Pat<(v1f64 (AArch64dup (f64 (load GPR64sp:$Rn)))),
4780 (LD1Rv1d GPR64sp:$Rn)>;
4781 def : Pat<(v4f16 (AArch64dup (f16 (load GPR64sp:$Rn)))),
4782 (LD1Rv4h GPR64sp:$Rn)>;
4783 def : Pat<(v8f16 (AArch64dup (f16 (load GPR64sp:$Rn)))),
4784 (LD1Rv8h GPR64sp:$Rn)>;
4786 class Ld1Lane128Pat<SDPatternOperator scalar_load, Operand VecIndex,
4787 ValueType VTy, ValueType STy, Instruction LD1>
4788 : Pat<(vector_insert (VTy VecListOne128:$Rd),
4789 (STy (scalar_load GPR64sp:$Rn)), VecIndex:$idx),
4790 (LD1 VecListOne128:$Rd, VecIndex:$idx, GPR64sp:$Rn)>;
4792 def : Ld1Lane128Pat<extloadi8, VectorIndexB, v16i8, i32, LD1i8>;
4793 def : Ld1Lane128Pat<extloadi16, VectorIndexH, v8i16, i32, LD1i16>;
4794 def : Ld1Lane128Pat<load, VectorIndexS, v4i32, i32, LD1i32>;
4795 def : Ld1Lane128Pat<load, VectorIndexS, v4f32, f32, LD1i32>;
4796 def : Ld1Lane128Pat<load, VectorIndexD, v2i64, i64, LD1i64>;
4797 def : Ld1Lane128Pat<load, VectorIndexD, v2f64, f64, LD1i64>;
4798 def : Ld1Lane128Pat<load, VectorIndexH, v8f16, f16, LD1i16>;
4800 class Ld1Lane64Pat<SDPatternOperator scalar_load, Operand VecIndex,
4801 ValueType VTy, ValueType STy, Instruction LD1>
4802 : Pat<(vector_insert (VTy VecListOne64:$Rd),
4803 (STy (scalar_load GPR64sp:$Rn)), VecIndex:$idx),
4805 (LD1 (SUBREG_TO_REG (i32 0), VecListOne64:$Rd, dsub),
4806 VecIndex:$idx, GPR64sp:$Rn),
4809 def : Ld1Lane64Pat<extloadi8, VectorIndexB, v8i8, i32, LD1i8>;
4810 def : Ld1Lane64Pat<extloadi16, VectorIndexH, v4i16, i32, LD1i16>;
4811 def : Ld1Lane64Pat<load, VectorIndexS, v2i32, i32, LD1i32>;
4812 def : Ld1Lane64Pat<load, VectorIndexS, v2f32, f32, LD1i32>;
4813 def : Ld1Lane64Pat<load, VectorIndexH, v4f16, f16, LD1i16>;
4816 defm LD1 : SIMDLdSt1SingleAliases<"ld1">;
4817 defm LD2 : SIMDLdSt2SingleAliases<"ld2">;
4818 defm LD3 : SIMDLdSt3SingleAliases<"ld3">;
4819 defm LD4 : SIMDLdSt4SingleAliases<"ld4">;
4822 defm ST1 : SIMDStSingleB<0, 0b000, "st1", VecListOneb, GPR64pi1>;
4823 defm ST1 : SIMDStSingleH<0, 0b010, 0, "st1", VecListOneh, GPR64pi2>;
4824 defm ST1 : SIMDStSingleS<0, 0b100, 0b00, "st1", VecListOnes, GPR64pi4>;
4825 defm ST1 : SIMDStSingleD<0, 0b100, 0b01, "st1", VecListOned, GPR64pi8>;
4827 let AddedComplexity = 19 in
4828 class St1Lane128Pat<SDPatternOperator scalar_store, Operand VecIndex,
4829 ValueType VTy, ValueType STy, Instruction ST1>
4831 (STy (vector_extract (VTy VecListOne128:$Vt), VecIndex:$idx)),
4833 (ST1 VecListOne128:$Vt, VecIndex:$idx, GPR64sp:$Rn)>;
4835 def : St1Lane128Pat<truncstorei8, VectorIndexB, v16i8, i32, ST1i8>;
4836 def : St1Lane128Pat<truncstorei16, VectorIndexH, v8i16, i32, ST1i16>;
4837 def : St1Lane128Pat<store, VectorIndexS, v4i32, i32, ST1i32>;
4838 def : St1Lane128Pat<store, VectorIndexS, v4f32, f32, ST1i32>;
4839 def : St1Lane128Pat<store, VectorIndexD, v2i64, i64, ST1i64>;
4840 def : St1Lane128Pat<store, VectorIndexD, v2f64, f64, ST1i64>;
4841 def : St1Lane128Pat<store, VectorIndexH, v8f16, f16, ST1i16>;
4843 let AddedComplexity = 19 in
4844 class St1Lane64Pat<SDPatternOperator scalar_store, Operand VecIndex,
4845 ValueType VTy, ValueType STy, Instruction ST1>
4847 (STy (vector_extract (VTy VecListOne64:$Vt), VecIndex:$idx)),
4849 (ST1 (SUBREG_TO_REG (i32 0), VecListOne64:$Vt, dsub),
4850 VecIndex:$idx, GPR64sp:$Rn)>;
4852 def : St1Lane64Pat<truncstorei8, VectorIndexB, v8i8, i32, ST1i8>;
4853 def : St1Lane64Pat<truncstorei16, VectorIndexH, v4i16, i32, ST1i16>;
4854 def : St1Lane64Pat<store, VectorIndexS, v2i32, i32, ST1i32>;
4855 def : St1Lane64Pat<store, VectorIndexS, v2f32, f32, ST1i32>;
4856 def : St1Lane64Pat<store, VectorIndexH, v4f16, f16, ST1i16>;
4858 multiclass St1LanePost64Pat<SDPatternOperator scalar_store, Operand VecIndex,
4859 ValueType VTy, ValueType STy, Instruction ST1,
4861 def : Pat<(scalar_store
4862 (STy (vector_extract (VTy VecListOne64:$Vt), VecIndex:$idx)),
4863 GPR64sp:$Rn, offset),
4864 (ST1 (SUBREG_TO_REG (i32 0), VecListOne64:$Vt, dsub),
4865 VecIndex:$idx, GPR64sp:$Rn, XZR)>;
4867 def : Pat<(scalar_store
4868 (STy (vector_extract (VTy VecListOne64:$Vt), VecIndex:$idx)),
4869 GPR64sp:$Rn, GPR64:$Rm),
4870 (ST1 (SUBREG_TO_REG (i32 0), VecListOne64:$Vt, dsub),
4871 VecIndex:$idx, GPR64sp:$Rn, $Rm)>;
4874 defm : St1LanePost64Pat<post_truncsti8, VectorIndexB, v8i8, i32, ST1i8_POST, 1>;
4875 defm : St1LanePost64Pat<post_truncsti16, VectorIndexH, v4i16, i32, ST1i16_POST,
4877 defm : St1LanePost64Pat<post_store, VectorIndexS, v2i32, i32, ST1i32_POST, 4>;
4878 defm : St1LanePost64Pat<post_store, VectorIndexS, v2f32, f32, ST1i32_POST, 4>;
4879 defm : St1LanePost64Pat<post_store, VectorIndexD, v1i64, i64, ST1i64_POST, 8>;
4880 defm : St1LanePost64Pat<post_store, VectorIndexD, v1f64, f64, ST1i64_POST, 8>;
4881 defm : St1LanePost64Pat<post_store, VectorIndexH, v4f16, f16, ST1i16_POST, 2>;
4883 multiclass St1LanePost128Pat<SDPatternOperator scalar_store, Operand VecIndex,
4884 ValueType VTy, ValueType STy, Instruction ST1,
4886 def : Pat<(scalar_store
4887 (STy (vector_extract (VTy VecListOne128:$Vt), VecIndex:$idx)),
4888 GPR64sp:$Rn, offset),
4889 (ST1 VecListOne128:$Vt, VecIndex:$idx, GPR64sp:$Rn, XZR)>;
4891 def : Pat<(scalar_store
4892 (STy (vector_extract (VTy VecListOne128:$Vt), VecIndex:$idx)),
4893 GPR64sp:$Rn, GPR64:$Rm),
4894 (ST1 VecListOne128:$Vt, VecIndex:$idx, GPR64sp:$Rn, $Rm)>;
4897 defm : St1LanePost128Pat<post_truncsti8, VectorIndexB, v16i8, i32, ST1i8_POST,
4899 defm : St1LanePost128Pat<post_truncsti16, VectorIndexH, v8i16, i32, ST1i16_POST,
4901 defm : St1LanePost128Pat<post_store, VectorIndexS, v4i32, i32, ST1i32_POST, 4>;
4902 defm : St1LanePost128Pat<post_store, VectorIndexS, v4f32, f32, ST1i32_POST, 4>;
4903 defm : St1LanePost128Pat<post_store, VectorIndexD, v2i64, i64, ST1i64_POST, 8>;
4904 defm : St1LanePost128Pat<post_store, VectorIndexD, v2f64, f64, ST1i64_POST, 8>;
4905 defm : St1LanePost128Pat<post_store, VectorIndexH, v8f16, f16, ST1i16_POST, 2>;
4907 let mayStore = 1, hasSideEffects = 0 in {
4908 defm ST2 : SIMDStSingleB<1, 0b000, "st2", VecListTwob, GPR64pi2>;
4909 defm ST2 : SIMDStSingleH<1, 0b010, 0, "st2", VecListTwoh, GPR64pi4>;
4910 defm ST2 : SIMDStSingleS<1, 0b100, 0b00, "st2", VecListTwos, GPR64pi8>;
4911 defm ST2 : SIMDStSingleD<1, 0b100, 0b01, "st2", VecListTwod, GPR64pi16>;
4912 defm ST3 : SIMDStSingleB<0, 0b001, "st3", VecListThreeb, GPR64pi3>;
4913 defm ST3 : SIMDStSingleH<0, 0b011, 0, "st3", VecListThreeh, GPR64pi6>;
4914 defm ST3 : SIMDStSingleS<0, 0b101, 0b00, "st3", VecListThrees, GPR64pi12>;
4915 defm ST3 : SIMDStSingleD<0, 0b101, 0b01, "st3", VecListThreed, GPR64pi24>;
4916 defm ST4 : SIMDStSingleB<1, 0b001, "st4", VecListFourb, GPR64pi4>;
4917 defm ST4 : SIMDStSingleH<1, 0b011, 0, "st4", VecListFourh, GPR64pi8>;
4918 defm ST4 : SIMDStSingleS<1, 0b101, 0b00, "st4", VecListFours, GPR64pi16>;
4919 defm ST4 : SIMDStSingleD<1, 0b101, 0b01, "st4", VecListFourd, GPR64pi32>;
4922 defm ST1 : SIMDLdSt1SingleAliases<"st1">;
4923 defm ST2 : SIMDLdSt2SingleAliases<"st2">;
4924 defm ST3 : SIMDLdSt3SingleAliases<"st3">;
4925 defm ST4 : SIMDLdSt4SingleAliases<"st4">;
4927 //----------------------------------------------------------------------------
4928 // Crypto extensions
4929 //----------------------------------------------------------------------------
4931 def AESErr : AESTiedInst<0b0100, "aese", int_aarch64_crypto_aese>;
4932 def AESDrr : AESTiedInst<0b0101, "aesd", int_aarch64_crypto_aesd>;
4933 def AESMCrr : AESInst< 0b0110, "aesmc", int_aarch64_crypto_aesmc>;
4934 def AESIMCrr : AESInst< 0b0111, "aesimc", int_aarch64_crypto_aesimc>;
4936 def SHA1Crrr : SHATiedInstQSV<0b000, "sha1c", int_aarch64_crypto_sha1c>;
4937 def SHA1Prrr : SHATiedInstQSV<0b001, "sha1p", int_aarch64_crypto_sha1p>;
4938 def SHA1Mrrr : SHATiedInstQSV<0b010, "sha1m", int_aarch64_crypto_sha1m>;
4939 def SHA1SU0rrr : SHATiedInstVVV<0b011, "sha1su0", int_aarch64_crypto_sha1su0>;
4940 def SHA256Hrrr : SHATiedInstQQV<0b100, "sha256h", int_aarch64_crypto_sha256h>;
4941 def SHA256H2rrr : SHATiedInstQQV<0b101, "sha256h2",int_aarch64_crypto_sha256h2>;
4942 def SHA256SU1rrr :SHATiedInstVVV<0b110, "sha256su1",int_aarch64_crypto_sha256su1>;
4944 def SHA1Hrr : SHAInstSS< 0b0000, "sha1h", int_aarch64_crypto_sha1h>;
4945 def SHA1SU1rr : SHATiedInstVV<0b0001, "sha1su1", int_aarch64_crypto_sha1su1>;
4946 def SHA256SU0rr : SHATiedInstVV<0b0010, "sha256su0",int_aarch64_crypto_sha256su0>;
4948 //----------------------------------------------------------------------------
4950 //----------------------------------------------------------------------------
4951 // FIXME: Like for X86, these should go in their own separate .td file.
4953 // Any instruction that defines a 32-bit result leaves the high half of the
4954 // register. Truncate can be lowered to EXTRACT_SUBREG. CopyFromReg may
4955 // be copying from a truncate. But any other 32-bit operation will zero-extend
4957 // FIXME: X86 also checks for CMOV here. Do we need something similar?
4958 def def32 : PatLeaf<(i32 GPR32:$src), [{
4959 return N->getOpcode() != ISD::TRUNCATE &&
4960 N->getOpcode() != TargetOpcode::EXTRACT_SUBREG &&
4961 N->getOpcode() != ISD::CopyFromReg;
4964 // In the case of a 32-bit def that is known to implicitly zero-extend,
4965 // we can use a SUBREG_TO_REG.
4966 def : Pat<(i64 (zext def32:$src)), (SUBREG_TO_REG (i64 0), GPR32:$src, sub_32)>;
4968 // For an anyext, we don't care what the high bits are, so we can perform an
4969 // INSERT_SUBREF into an IMPLICIT_DEF.
4970 def : Pat<(i64 (anyext GPR32:$src)),
4971 (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32)>;
4973 // When we need to explicitly zero-extend, we use an unsigned bitfield move
4974 // instruction (UBFM) on the enclosing super-reg.
4975 def : Pat<(i64 (zext GPR32:$src)),
4976 (UBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32), 0, 31)>;
4978 // To sign extend, we use a signed bitfield move instruction (SBFM) on the
4979 // containing super-reg.
4980 def : Pat<(i64 (sext GPR32:$src)),
4981 (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32), 0, 31)>;
4982 def : Pat<(i64 (sext_inreg GPR64:$src, i32)), (SBFMXri GPR64:$src, 0, 31)>;
4983 def : Pat<(i64 (sext_inreg GPR64:$src, i16)), (SBFMXri GPR64:$src, 0, 15)>;
4984 def : Pat<(i64 (sext_inreg GPR64:$src, i8)), (SBFMXri GPR64:$src, 0, 7)>;
4985 def : Pat<(i64 (sext_inreg GPR64:$src, i1)), (SBFMXri GPR64:$src, 0, 0)>;
4986 def : Pat<(i32 (sext_inreg GPR32:$src, i16)), (SBFMWri GPR32:$src, 0, 15)>;
4987 def : Pat<(i32 (sext_inreg GPR32:$src, i8)), (SBFMWri GPR32:$src, 0, 7)>;
4988 def : Pat<(i32 (sext_inreg GPR32:$src, i1)), (SBFMWri GPR32:$src, 0, 0)>;
4990 def : Pat<(shl (sext_inreg GPR32:$Rn, i8), (i64 imm0_31:$imm)),
4991 (SBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
4992 (i64 (i32shift_sext_i8 imm0_31:$imm)))>;
4993 def : Pat<(shl (sext_inreg GPR64:$Rn, i8), (i64 imm0_63:$imm)),
4994 (SBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
4995 (i64 (i64shift_sext_i8 imm0_63:$imm)))>;
4997 def : Pat<(shl (sext_inreg GPR32:$Rn, i16), (i64 imm0_31:$imm)),
4998 (SBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
4999 (i64 (i32shift_sext_i16 imm0_31:$imm)))>;
5000 def : Pat<(shl (sext_inreg GPR64:$Rn, i16), (i64 imm0_63:$imm)),
5001 (SBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
5002 (i64 (i64shift_sext_i16 imm0_63:$imm)))>;
5004 def : Pat<(shl (i64 (sext GPR32:$Rn)), (i64 imm0_63:$imm)),
5005 (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$Rn, sub_32),
5006 (i64 (i64shift_a imm0_63:$imm)),
5007 (i64 (i64shift_sext_i32 imm0_63:$imm)))>;
5009 // sra patterns have an AddedComplexity of 10, so make sure we have a higher
5010 // AddedComplexity for the following patterns since we want to match sext + sra
5011 // patterns before we attempt to match a single sra node.
5012 let AddedComplexity = 20 in {
5013 // We support all sext + sra combinations which preserve at least one bit of the
5014 // original value which is to be sign extended. E.g. we support shifts up to
5016 def : Pat<(sra (sext_inreg GPR32:$Rn, i8), (i64 imm0_7:$imm)),
5017 (SBFMWri GPR32:$Rn, (i64 imm0_7:$imm), 7)>;
5018 def : Pat<(sra (sext_inreg GPR64:$Rn, i8), (i64 imm0_7:$imm)),
5019 (SBFMXri GPR64:$Rn, (i64 imm0_7:$imm), 7)>;
5021 def : Pat<(sra (sext_inreg GPR32:$Rn, i16), (i64 imm0_15:$imm)),
5022 (SBFMWri GPR32:$Rn, (i64 imm0_15:$imm), 15)>;
5023 def : Pat<(sra (sext_inreg GPR64:$Rn, i16), (i64 imm0_15:$imm)),
5024 (SBFMXri GPR64:$Rn, (i64 imm0_15:$imm), 15)>;
5026 def : Pat<(sra (i64 (sext GPR32:$Rn)), (i64 imm0_31:$imm)),
5027 (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$Rn, sub_32),
5028 (i64 imm0_31:$imm), 31)>;
5029 } // AddedComplexity = 20
5031 // To truncate, we can simply extract from a subregister.
5032 def : Pat<(i32 (trunc GPR64sp:$src)),
5033 (i32 (EXTRACT_SUBREG GPR64sp:$src, sub_32))>;
5035 // __builtin_trap() uses the BRK instruction on AArch64.
5036 def : Pat<(trap), (BRK 1)>;
5038 // Conversions within AdvSIMD types in the same register size are free.
5039 // But because we need a consistent lane ordering, in big endian many
5040 // conversions require one or more REV instructions.
5042 // Consider a simple memory load followed by a bitconvert then a store.
5044 // v1 = BITCAST v2i32 v0 to v4i16
5047 // In big endian mode every memory access has an implicit byte swap. LDR and
5048 // STR do a 64-bit byte swap, whereas LD1/ST1 do a byte swap per lane - that
5049 // is, they treat the vector as a sequence of elements to be byte-swapped.
5050 // The two pairs of instructions are fundamentally incompatible. We've decided
5051 // to use LD1/ST1 only to simplify compiler implementation.
5053 // LD1/ST1 perform the equivalent of a sequence of LDR/STR + REV. This makes
5054 // the original code sequence:
5056 // v1 = REV v2i32 (implicit)
5057 // v2 = BITCAST v2i32 v1 to v4i16
5058 // v3 = REV v4i16 v2 (implicit)
5061 // But this is now broken - the value stored is different to the value loaded
5062 // due to lane reordering. To fix this, on every BITCAST we must perform two
5065 // v1 = REV v2i32 (implicit)
5067 // v3 = BITCAST v2i32 v2 to v4i16
5069 // v5 = REV v4i16 v4 (implicit)
5072 // This means an extra two instructions, but actually in most cases the two REV
5073 // instructions can be combined into one. For example:
5074 // (REV64_2s (REV64_4h X)) === (REV32_4h X)
5076 // There is also no 128-bit REV instruction. This must be synthesized with an
5079 // Most bitconverts require some sort of conversion. The only exceptions are:
5080 // a) Identity conversions - vNfX <-> vNiX
5081 // b) Single-lane-to-scalar - v1fX <-> fX or v1iX <-> iX
5084 // Natural vector casts (64 bit)
5085 def : Pat<(v8i8 (AArch64NvCast (v2i32 FPR64:$src))), (v8i8 FPR64:$src)>;
5086 def : Pat<(v4i16 (AArch64NvCast (v2i32 FPR64:$src))), (v4i16 FPR64:$src)>;
5087 def : Pat<(v2i32 (AArch64NvCast (v2i32 FPR64:$src))), (v2i32 FPR64:$src)>;
5088 def : Pat<(v2f32 (AArch64NvCast (v2i32 FPR64:$src))), (v2f32 FPR64:$src)>;
5089 def : Pat<(v1i64 (AArch64NvCast (v2i32 FPR64:$src))), (v1i64 FPR64:$src)>;
5091 def : Pat<(v8i8 (AArch64NvCast (v4i16 FPR64:$src))), (v8i8 FPR64:$src)>;
5092 def : Pat<(v4i16 (AArch64NvCast (v4i16 FPR64:$src))), (v4i16 FPR64:$src)>;
5093 def : Pat<(v2i32 (AArch64NvCast (v4i16 FPR64:$src))), (v2i32 FPR64:$src)>;
5094 def : Pat<(v1i64 (AArch64NvCast (v4i16 FPR64:$src))), (v1i64 FPR64:$src)>;
5096 def : Pat<(v8i8 (AArch64NvCast (v8i8 FPR64:$src))), (v8i8 FPR64:$src)>;
5097 def : Pat<(v4i16 (AArch64NvCast (v8i8 FPR64:$src))), (v4i16 FPR64:$src)>;
5098 def : Pat<(v2i32 (AArch64NvCast (v8i8 FPR64:$src))), (v2i32 FPR64:$src)>;
5099 def : Pat<(v1i64 (AArch64NvCast (v8i8 FPR64:$src))), (v1i64 FPR64:$src)>;
5101 def : Pat<(v8i8 (AArch64NvCast (f64 FPR64:$src))), (v8i8 FPR64:$src)>;
5102 def : Pat<(v4i16 (AArch64NvCast (f64 FPR64:$src))), (v4i16 FPR64:$src)>;
5103 def : Pat<(v2i32 (AArch64NvCast (f64 FPR64:$src))), (v2i32 FPR64:$src)>;
5104 def : Pat<(v2f32 (AArch64NvCast (f64 FPR64:$src))), (v2f32 FPR64:$src)>;
5105 def : Pat<(v1i64 (AArch64NvCast (f64 FPR64:$src))), (v1i64 FPR64:$src)>;
5106 def : Pat<(v1f64 (AArch64NvCast (f64 FPR64:$src))), (v1f64 FPR64:$src)>;
5108 def : Pat<(v8i8 (AArch64NvCast (v2f32 FPR64:$src))), (v8i8 FPR64:$src)>;
5109 def : Pat<(v4i16 (AArch64NvCast (v2f32 FPR64:$src))), (v4i16 FPR64:$src)>;
5110 def : Pat<(v2i32 (AArch64NvCast (v2f32 FPR64:$src))), (v2i32 FPR64:$src)>;
5111 def : Pat<(v2f32 (AArch64NvCast (v2f32 FPR64:$src))), (v2f32 FPR64:$src)>;
5112 def : Pat<(v1i64 (AArch64NvCast (v2f32 FPR64:$src))), (v1i64 FPR64:$src)>;
5114 // Natural vector casts (128 bit)
5115 def : Pat<(v16i8 (AArch64NvCast (v4i32 FPR128:$src))), (v16i8 FPR128:$src)>;
5116 def : Pat<(v8i16 (AArch64NvCast (v4i32 FPR128:$src))), (v8i16 FPR128:$src)>;
5117 def : Pat<(v4i32 (AArch64NvCast (v4i32 FPR128:$src))), (v4i32 FPR128:$src)>;
5118 def : Pat<(v4f32 (AArch64NvCast (v4i32 FPR128:$src))), (v4f32 FPR128:$src)>;
5119 def : Pat<(v2i64 (AArch64NvCast (v4i32 FPR128:$src))), (v2i64 FPR128:$src)>;
5121 def : Pat<(v16i8 (AArch64NvCast (v8i16 FPR128:$src))), (v16i8 FPR128:$src)>;
5122 def : Pat<(v8i16 (AArch64NvCast (v8i16 FPR128:$src))), (v8i16 FPR128:$src)>;
5123 def : Pat<(v4i32 (AArch64NvCast (v8i16 FPR128:$src))), (v4i32 FPR128:$src)>;
5124 def : Pat<(v2i64 (AArch64NvCast (v8i16 FPR128:$src))), (v2i64 FPR128:$src)>;
5126 def : Pat<(v16i8 (AArch64NvCast (v16i8 FPR128:$src))), (v16i8 FPR128:$src)>;
5127 def : Pat<(v8i16 (AArch64NvCast (v16i8 FPR128:$src))), (v8i16 FPR128:$src)>;
5128 def : Pat<(v4i32 (AArch64NvCast (v16i8 FPR128:$src))), (v4i32 FPR128:$src)>;
5129 def : Pat<(v2i64 (AArch64NvCast (v16i8 FPR128:$src))), (v2i64 FPR128:$src)>;
5131 def : Pat<(v16i8 (AArch64NvCast (v2i64 FPR128:$src))), (v16i8 FPR128:$src)>;
5132 def : Pat<(v8i16 (AArch64NvCast (v2i64 FPR128:$src))), (v8i16 FPR128:$src)>;
5133 def : Pat<(v4i32 (AArch64NvCast (v2i64 FPR128:$src))), (v4i32 FPR128:$src)>;
5134 def : Pat<(v2i64 (AArch64NvCast (v2i64 FPR128:$src))), (v2i64 FPR128:$src)>;
5135 def : Pat<(v4f32 (AArch64NvCast (v2i64 FPR128:$src))), (v4f32 FPR128:$src)>;
5136 def : Pat<(v2f64 (AArch64NvCast (v2i64 FPR128:$src))), (v2f64 FPR128:$src)>;
5138 def : Pat<(v16i8 (AArch64NvCast (v4f32 FPR128:$src))), (v16i8 FPR128:$src)>;
5139 def : Pat<(v8i16 (AArch64NvCast (v4f32 FPR128:$src))), (v8i16 FPR128:$src)>;
5140 def : Pat<(v4i32 (AArch64NvCast (v4f32 FPR128:$src))), (v4i32 FPR128:$src)>;
5141 def : Pat<(v4f32 (AArch64NvCast (v4f32 FPR128:$src))), (v4f32 FPR128:$src)>;
5142 def : Pat<(v2i64 (AArch64NvCast (v4f32 FPR128:$src))), (v2i64 FPR128:$src)>;
5144 def : Pat<(v16i8 (AArch64NvCast (v2f64 FPR128:$src))), (v16i8 FPR128:$src)>;
5145 def : Pat<(v8i16 (AArch64NvCast (v2f64 FPR128:$src))), (v8i16 FPR128:$src)>;
5146 def : Pat<(v4i32 (AArch64NvCast (v2f64 FPR128:$src))), (v4i32 FPR128:$src)>;
5147 def : Pat<(v2i64 (AArch64NvCast (v2f64 FPR128:$src))), (v2i64 FPR128:$src)>;
5148 def : Pat<(v2f64 (AArch64NvCast (v2f64 FPR128:$src))), (v2f64 FPR128:$src)>;
5150 let Predicates = [IsLE] in {
5151 def : Pat<(v8i8 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5152 def : Pat<(v4i16 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5153 def : Pat<(v2i32 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5154 def : Pat<(v4f16 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5155 def : Pat<(v2f32 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5157 def : Pat<(i64 (bitconvert (v8i8 V64:$Vn))),
5158 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5159 def : Pat<(i64 (bitconvert (v4i16 V64:$Vn))),
5160 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5161 def : Pat<(i64 (bitconvert (v2i32 V64:$Vn))),
5162 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5163 def : Pat<(i64 (bitconvert (v4f16 V64:$Vn))),
5164 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5165 def : Pat<(i64 (bitconvert (v2f32 V64:$Vn))),
5166 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5167 def : Pat<(i64 (bitconvert (v1f64 V64:$Vn))),
5168 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5170 let Predicates = [IsBE] in {
5171 def : Pat<(v8i8 (bitconvert GPR64:$Xn)),
5172 (REV64v8i8 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
5173 def : Pat<(v4i16 (bitconvert GPR64:$Xn)),
5174 (REV64v4i16 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
5175 def : Pat<(v2i32 (bitconvert GPR64:$Xn)),
5176 (REV64v2i32 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
5177 def : Pat<(v4f16 (bitconvert GPR64:$Xn)),
5178 (REV64v4i16 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
5179 def : Pat<(v2f32 (bitconvert GPR64:$Xn)),
5180 (REV64v2i32 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
5182 def : Pat<(i64 (bitconvert (v8i8 V64:$Vn))),
5183 (REV64v8i8 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
5184 def : Pat<(i64 (bitconvert (v4i16 V64:$Vn))),
5185 (REV64v4i16 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
5186 def : Pat<(i64 (bitconvert (v2i32 V64:$Vn))),
5187 (REV64v2i32 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
5188 def : Pat<(i64 (bitconvert (v4f16 V64:$Vn))),
5189 (REV64v4i16 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
5190 def : Pat<(i64 (bitconvert (v2f32 V64:$Vn))),
5191 (REV64v2i32 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
5193 def : Pat<(v1i64 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5194 def : Pat<(v1f64 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5195 def : Pat<(i64 (bitconvert (v1i64 V64:$Vn))),
5196 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5197 def : Pat<(v1i64 (scalar_to_vector GPR64:$Xn)),
5198 (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5199 def : Pat<(v1f64 (scalar_to_vector GPR64:$Xn)),
5200 (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5201 def : Pat<(v1f64 (scalar_to_vector (f64 FPR64:$Xn))), (v1f64 FPR64:$Xn)>;
5203 def : Pat<(f32 (bitconvert (i32 GPR32:$Xn))),
5204 (COPY_TO_REGCLASS GPR32:$Xn, FPR32)>;
5205 def : Pat<(i32 (bitconvert (f32 FPR32:$Xn))),
5206 (COPY_TO_REGCLASS FPR32:$Xn, GPR32)>;
5207 def : Pat<(f64 (bitconvert (i64 GPR64:$Xn))),
5208 (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5209 def : Pat<(i64 (bitconvert (f64 FPR64:$Xn))),
5210 (COPY_TO_REGCLASS FPR64:$Xn, GPR64)>;
5211 def : Pat<(i64 (bitconvert (v1f64 V64:$Vn))),
5212 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5214 let Predicates = [IsLE] in {
5215 def : Pat<(v1i64 (bitconvert (v2i32 FPR64:$src))), (v1i64 FPR64:$src)>;
5216 def : Pat<(v1i64 (bitconvert (v4i16 FPR64:$src))), (v1i64 FPR64:$src)>;
5217 def : Pat<(v1i64 (bitconvert (v8i8 FPR64:$src))), (v1i64 FPR64:$src)>;
5218 def : Pat<(v1i64 (bitconvert (v4f16 FPR64:$src))), (v1i64 FPR64:$src)>;
5219 def : Pat<(v1i64 (bitconvert (v2f32 FPR64:$src))), (v1i64 FPR64:$src)>;
5221 let Predicates = [IsBE] in {
5222 def : Pat<(v1i64 (bitconvert (v2i32 FPR64:$src))),
5223 (v1i64 (REV64v2i32 FPR64:$src))>;
5224 def : Pat<(v1i64 (bitconvert (v4i16 FPR64:$src))),
5225 (v1i64 (REV64v4i16 FPR64:$src))>;
5226 def : Pat<(v1i64 (bitconvert (v8i8 FPR64:$src))),
5227 (v1i64 (REV64v8i8 FPR64:$src))>;
5228 def : Pat<(v1i64 (bitconvert (v4f16 FPR64:$src))),
5229 (v1i64 (REV64v4i16 FPR64:$src))>;
5230 def : Pat<(v1i64 (bitconvert (v2f32 FPR64:$src))),
5231 (v1i64 (REV64v2i32 FPR64:$src))>;
5233 def : Pat<(v1i64 (bitconvert (v1f64 FPR64:$src))), (v1i64 FPR64:$src)>;
5234 def : Pat<(v1i64 (bitconvert (f64 FPR64:$src))), (v1i64 FPR64:$src)>;
5236 let Predicates = [IsLE] in {
5237 def : Pat<(v2i32 (bitconvert (v1i64 FPR64:$src))), (v2i32 FPR64:$src)>;
5238 def : Pat<(v2i32 (bitconvert (v4i16 FPR64:$src))), (v2i32 FPR64:$src)>;
5239 def : Pat<(v2i32 (bitconvert (v8i8 FPR64:$src))), (v2i32 FPR64:$src)>;
5240 def : Pat<(v2i32 (bitconvert (f64 FPR64:$src))), (v2i32 FPR64:$src)>;
5241 def : Pat<(v2i32 (bitconvert (v1f64 FPR64:$src))), (v2i32 FPR64:$src)>;
5242 def : Pat<(v2i32 (bitconvert (v4f16 FPR64:$src))), (v2i32 FPR64:$src)>;
5244 let Predicates = [IsBE] in {
5245 def : Pat<(v2i32 (bitconvert (v1i64 FPR64:$src))),
5246 (v2i32 (REV64v2i32 FPR64:$src))>;
5247 def : Pat<(v2i32 (bitconvert (v4i16 FPR64:$src))),
5248 (v2i32 (REV32v4i16 FPR64:$src))>;
5249 def : Pat<(v2i32 (bitconvert (v8i8 FPR64:$src))),
5250 (v2i32 (REV32v8i8 FPR64:$src))>;
5251 def : Pat<(v2i32 (bitconvert (f64 FPR64:$src))),
5252 (v2i32 (REV64v2i32 FPR64:$src))>;
5253 def : Pat<(v2i32 (bitconvert (v1f64 FPR64:$src))),
5254 (v2i32 (REV64v2i32 FPR64:$src))>;
5255 def : Pat<(v2i32 (bitconvert (v4f16 FPR64:$src))),
5256 (v2i32 (REV64v4i16 FPR64:$src))>;
5258 def : Pat<(v2i32 (bitconvert (v2f32 FPR64:$src))), (v2i32 FPR64:$src)>;
5260 let Predicates = [IsLE] in {
5261 def : Pat<(v4i16 (bitconvert (v1i64 FPR64:$src))), (v4i16 FPR64:$src)>;
5262 def : Pat<(v4i16 (bitconvert (v2i32 FPR64:$src))), (v4i16 FPR64:$src)>;
5263 def : Pat<(v4i16 (bitconvert (v8i8 FPR64:$src))), (v4i16 FPR64:$src)>;
5264 def : Pat<(v4i16 (bitconvert (f64 FPR64:$src))), (v4i16 FPR64:$src)>;
5265 def : Pat<(v4i16 (bitconvert (v4f16 FPR64:$src))), (v4i16 FPR64:$src)>;
5266 def : Pat<(v4i16 (bitconvert (v2f32 FPR64:$src))), (v4i16 FPR64:$src)>;
5267 def : Pat<(v4i16 (bitconvert (v1f64 FPR64:$src))), (v4i16 FPR64:$src)>;
5269 let Predicates = [IsBE] in {
5270 def : Pat<(v4i16 (bitconvert (v1i64 FPR64:$src))),
5271 (v4i16 (REV64v4i16 FPR64:$src))>;
5272 def : Pat<(v4i16 (bitconvert (v2i32 FPR64:$src))),
5273 (v4i16 (REV32v4i16 FPR64:$src))>;
5274 def : Pat<(v4i16 (bitconvert (v8i8 FPR64:$src))),
5275 (v4i16 (REV16v8i8 FPR64:$src))>;
5276 def : Pat<(v4i16 (bitconvert (f64 FPR64:$src))),
5277 (v4i16 (REV64v4i16 FPR64:$src))>;
5278 def : Pat<(v4i16 (bitconvert (v4f16 FPR64:$src))),
5279 (v4i16 (REV32v4i16 FPR64:$src))>;
5280 def : Pat<(v4i16 (bitconvert (v2f32 FPR64:$src))),
5281 (v4i16 (REV32v4i16 FPR64:$src))>;
5282 def : Pat<(v4i16 (bitconvert (v1f64 FPR64:$src))),
5283 (v4i16 (REV64v4i16 FPR64:$src))>;
5286 let Predicates = [IsLE] in {
5287 def : Pat<(v4f16 (bitconvert (v1i64 FPR64:$src))), (v4f16 FPR64:$src)>;
5288 def : Pat<(v4f16 (bitconvert (v2i32 FPR64:$src))), (v4f16 FPR64:$src)>;
5289 def : Pat<(v4f16 (bitconvert (v4i16 FPR64:$src))), (v4f16 FPR64:$src)>;
5290 def : Pat<(v4f16 (bitconvert (v8i8 FPR64:$src))), (v4f16 FPR64:$src)>;
5291 def : Pat<(v4f16 (bitconvert (f64 FPR64:$src))), (v4f16 FPR64:$src)>;
5292 def : Pat<(v4f16 (bitconvert (v2f32 FPR64:$src))), (v4f16 FPR64:$src)>;
5293 def : Pat<(v4f16 (bitconvert (v1f64 FPR64:$src))), (v4f16 FPR64:$src)>;
5295 let Predicates = [IsBE] in {
5296 def : Pat<(v4f16 (bitconvert (v1i64 FPR64:$src))),
5297 (v4f16 (REV64v4i16 FPR64:$src))>;
5298 def : Pat<(v4f16 (bitconvert (v2i32 FPR64:$src))),
5299 (v4f16 (REV64v4i16 FPR64:$src))>;
5300 def : Pat<(v4f16 (bitconvert (v4i16 FPR64:$src))),
5301 (v4f16 (REV64v4i16 FPR64:$src))>;
5302 def : Pat<(v4f16 (bitconvert (v8i8 FPR64:$src))),
5303 (v4f16 (REV16v8i8 FPR64:$src))>;
5304 def : Pat<(v4f16 (bitconvert (f64 FPR64:$src))),
5305 (v4f16 (REV64v4i16 FPR64:$src))>;
5306 def : Pat<(v4f16 (bitconvert (v2f32 FPR64:$src))),
5307 (v4f16 (REV64v4i16 FPR64:$src))>;
5308 def : Pat<(v4f16 (bitconvert (v1f64 FPR64:$src))),
5309 (v4f16 (REV64v4i16 FPR64:$src))>;
5314 let Predicates = [IsLE] in {
5315 def : Pat<(v8i8 (bitconvert (v1i64 FPR64:$src))), (v8i8 FPR64:$src)>;
5316 def : Pat<(v8i8 (bitconvert (v2i32 FPR64:$src))), (v8i8 FPR64:$src)>;
5317 def : Pat<(v8i8 (bitconvert (v4i16 FPR64:$src))), (v8i8 FPR64:$src)>;
5318 def : Pat<(v8i8 (bitconvert (f64 FPR64:$src))), (v8i8 FPR64:$src)>;
5319 def : Pat<(v8i8 (bitconvert (v2f32 FPR64:$src))), (v8i8 FPR64:$src)>;
5320 def : Pat<(v8i8 (bitconvert (v1f64 FPR64:$src))), (v8i8 FPR64:$src)>;
5321 def : Pat<(v8i8 (bitconvert (v4f16 FPR64:$src))), (v8i8 FPR64:$src)>;
5323 let Predicates = [IsBE] in {
5324 def : Pat<(v8i8 (bitconvert (v1i64 FPR64:$src))),
5325 (v8i8 (REV64v8i8 FPR64:$src))>;
5326 def : Pat<(v8i8 (bitconvert (v2i32 FPR64:$src))),
5327 (v8i8 (REV32v8i8 FPR64:$src))>;
5328 def : Pat<(v8i8 (bitconvert (v4i16 FPR64:$src))),
5329 (v8i8 (REV16v8i8 FPR64:$src))>;
5330 def : Pat<(v8i8 (bitconvert (f64 FPR64:$src))),
5331 (v8i8 (REV64v8i8 FPR64:$src))>;
5332 def : Pat<(v8i8 (bitconvert (v2f32 FPR64:$src))),
5333 (v8i8 (REV32v8i8 FPR64:$src))>;
5334 def : Pat<(v8i8 (bitconvert (v1f64 FPR64:$src))),
5335 (v8i8 (REV64v8i8 FPR64:$src))>;
5336 def : Pat<(v8i8 (bitconvert (v4f16 FPR64:$src))),
5337 (v8i8 (REV16v8i8 FPR64:$src))>;
5340 let Predicates = [IsLE] in {
5341 def : Pat<(f64 (bitconvert (v2i32 FPR64:$src))), (f64 FPR64:$src)>;
5342 def : Pat<(f64 (bitconvert (v4i16 FPR64:$src))), (f64 FPR64:$src)>;
5343 def : Pat<(f64 (bitconvert (v2f32 FPR64:$src))), (f64 FPR64:$src)>;
5344 def : Pat<(f64 (bitconvert (v8i8 FPR64:$src))), (f64 FPR64:$src)>;
5345 def : Pat<(f64 (bitconvert (v4f16 FPR64:$src))), (f64 FPR64:$src)>;
5347 let Predicates = [IsBE] in {
5348 def : Pat<(f64 (bitconvert (v2i32 FPR64:$src))),
5349 (f64 (REV64v2i32 FPR64:$src))>;
5350 def : Pat<(f64 (bitconvert (v4i16 FPR64:$src))),
5351 (f64 (REV64v4i16 FPR64:$src))>;
5352 def : Pat<(f64 (bitconvert (v2f32 FPR64:$src))),
5353 (f64 (REV64v2i32 FPR64:$src))>;
5354 def : Pat<(f64 (bitconvert (v8i8 FPR64:$src))),
5355 (f64 (REV64v8i8 FPR64:$src))>;
5356 def : Pat<(f64 (bitconvert (v4f16 FPR64:$src))),
5357 (f64 (REV64v4i16 FPR64:$src))>;
5359 def : Pat<(f64 (bitconvert (v1i64 FPR64:$src))), (f64 FPR64:$src)>;
5360 def : Pat<(f64 (bitconvert (v1f64 FPR64:$src))), (f64 FPR64:$src)>;
5362 let Predicates = [IsLE] in {
5363 def : Pat<(v1f64 (bitconvert (v2i32 FPR64:$src))), (v1f64 FPR64:$src)>;
5364 def : Pat<(v1f64 (bitconvert (v4i16 FPR64:$src))), (v1f64 FPR64:$src)>;
5365 def : Pat<(v1f64 (bitconvert (v8i8 FPR64:$src))), (v1f64 FPR64:$src)>;
5366 def : Pat<(v1f64 (bitconvert (v2f32 FPR64:$src))), (v1f64 FPR64:$src)>;
5367 def : Pat<(v1f64 (bitconvert (v4f16 FPR64:$src))), (v1f64 FPR64:$src)>;
5369 let Predicates = [IsBE] in {
5370 def : Pat<(v1f64 (bitconvert (v2i32 FPR64:$src))),
5371 (v1f64 (REV64v2i32 FPR64:$src))>;
5372 def : Pat<(v1f64 (bitconvert (v4i16 FPR64:$src))),
5373 (v1f64 (REV64v4i16 FPR64:$src))>;
5374 def : Pat<(v1f64 (bitconvert (v8i8 FPR64:$src))),
5375 (v1f64 (REV64v8i8 FPR64:$src))>;
5376 def : Pat<(v1f64 (bitconvert (v2f32 FPR64:$src))),
5377 (v1f64 (REV64v2i32 FPR64:$src))>;
5378 def : Pat<(v1f64 (bitconvert (v4f16 FPR64:$src))),
5379 (v1f64 (REV64v4i16 FPR64:$src))>;
5381 def : Pat<(v1f64 (bitconvert (v1i64 FPR64:$src))), (v1f64 FPR64:$src)>;
5382 def : Pat<(v1f64 (bitconvert (f64 FPR64:$src))), (v1f64 FPR64:$src)>;
5384 let Predicates = [IsLE] in {
5385 def : Pat<(v2f32 (bitconvert (v1i64 FPR64:$src))), (v2f32 FPR64:$src)>;
5386 def : Pat<(v2f32 (bitconvert (v4i16 FPR64:$src))), (v2f32 FPR64:$src)>;
5387 def : Pat<(v2f32 (bitconvert (v8i8 FPR64:$src))), (v2f32 FPR64:$src)>;
5388 def : Pat<(v2f32 (bitconvert (v1f64 FPR64:$src))), (v2f32 FPR64:$src)>;
5389 def : Pat<(v2f32 (bitconvert (f64 FPR64:$src))), (v2f32 FPR64:$src)>;
5390 def : Pat<(v2f32 (bitconvert (v4f16 FPR64:$src))), (v2f32 FPR64:$src)>;
5392 let Predicates = [IsBE] in {
5393 def : Pat<(v2f32 (bitconvert (v1i64 FPR64:$src))),
5394 (v2f32 (REV64v2i32 FPR64:$src))>;
5395 def : Pat<(v2f32 (bitconvert (v4i16 FPR64:$src))),
5396 (v2f32 (REV32v4i16 FPR64:$src))>;
5397 def : Pat<(v2f32 (bitconvert (v8i8 FPR64:$src))),
5398 (v2f32 (REV32v8i8 FPR64:$src))>;
5399 def : Pat<(v2f32 (bitconvert (v1f64 FPR64:$src))),
5400 (v2f32 (REV64v2i32 FPR64:$src))>;
5401 def : Pat<(v2f32 (bitconvert (f64 FPR64:$src))),
5402 (v2f32 (REV64v2i32 FPR64:$src))>;
5403 def : Pat<(v2f32 (bitconvert (v4f16 FPR64:$src))),
5404 (v2f32 (REV64v4i16 FPR64:$src))>;
5406 def : Pat<(v2f32 (bitconvert (v2i32 FPR64:$src))), (v2f32 FPR64:$src)>;
5408 let Predicates = [IsLE] in {
5409 def : Pat<(f128 (bitconvert (v2i64 FPR128:$src))), (f128 FPR128:$src)>;
5410 def : Pat<(f128 (bitconvert (v4i32 FPR128:$src))), (f128 FPR128:$src)>;
5411 def : Pat<(f128 (bitconvert (v8i16 FPR128:$src))), (f128 FPR128:$src)>;
5412 def : Pat<(f128 (bitconvert (v2f64 FPR128:$src))), (f128 FPR128:$src)>;
5413 def : Pat<(f128 (bitconvert (v4f32 FPR128:$src))), (f128 FPR128:$src)>;
5414 def : Pat<(f128 (bitconvert (v8f16 FPR128:$src))), (f128 FPR128:$src)>;
5415 def : Pat<(f128 (bitconvert (v16i8 FPR128:$src))), (f128 FPR128:$src)>;
5417 let Predicates = [IsBE] in {
5418 def : Pat<(f128 (bitconvert (v2i64 FPR128:$src))),
5419 (f128 (EXTv16i8 FPR128:$src, FPR128:$src, (i32 8)))>;
5420 def : Pat<(f128 (bitconvert (v4i32 FPR128:$src))),
5421 (f128 (EXTv16i8 (REV64v4i32 FPR128:$src),
5422 (REV64v4i32 FPR128:$src), (i32 8)))>;
5423 def : Pat<(f128 (bitconvert (v8i16 FPR128:$src))),
5424 (f128 (EXTv16i8 (REV64v8i16 FPR128:$src),
5425 (REV64v8i16 FPR128:$src), (i32 8)))>;
5426 def : Pat<(f128 (bitconvert (v8f16 FPR128:$src))),
5427 (f128 (EXTv16i8 (REV64v8i16 FPR128:$src),
5428 (REV64v8i16 FPR128:$src), (i32 8)))>;
5429 def : Pat<(f128 (bitconvert (v2f64 FPR128:$src))),
5430 (f128 (EXTv16i8 FPR128:$src, FPR128:$src, (i32 8)))>;
5431 def : Pat<(f128 (bitconvert (v4f32 FPR128:$src))),
5432 (f128 (EXTv16i8 (REV64v4i32 FPR128:$src),
5433 (REV64v4i32 FPR128:$src), (i32 8)))>;
5434 def : Pat<(f128 (bitconvert (v16i8 FPR128:$src))),
5435 (f128 (EXTv16i8 (REV64v16i8 FPR128:$src),
5436 (REV64v16i8 FPR128:$src), (i32 8)))>;
5439 let Predicates = [IsLE] in {
5440 def : Pat<(v2f64 (bitconvert (f128 FPR128:$src))), (v2f64 FPR128:$src)>;
5441 def : Pat<(v2f64 (bitconvert (v4i32 FPR128:$src))), (v2f64 FPR128:$src)>;
5442 def : Pat<(v2f64 (bitconvert (v8i16 FPR128:$src))), (v2f64 FPR128:$src)>;
5443 def : Pat<(v2f64 (bitconvert (v8f16 FPR128:$src))), (v2f64 FPR128:$src)>;
5444 def : Pat<(v2f64 (bitconvert (v16i8 FPR128:$src))), (v2f64 FPR128:$src)>;
5445 def : Pat<(v2f64 (bitconvert (v4f32 FPR128:$src))), (v2f64 FPR128:$src)>;
5447 let Predicates = [IsBE] in {
5448 def : Pat<(v2f64 (bitconvert (f128 FPR128:$src))),
5449 (v2f64 (EXTv16i8 FPR128:$src,
5450 FPR128:$src, (i32 8)))>;
5451 def : Pat<(v2f64 (bitconvert (v4i32 FPR128:$src))),
5452 (v2f64 (REV64v4i32 FPR128:$src))>;
5453 def : Pat<(v2f64 (bitconvert (v8i16 FPR128:$src))),
5454 (v2f64 (REV64v8i16 FPR128:$src))>;
5455 def : Pat<(v2f64 (bitconvert (v8f16 FPR128:$src))),
5456 (v2f64 (REV64v8i16 FPR128:$src))>;
5457 def : Pat<(v2f64 (bitconvert (v16i8 FPR128:$src))),
5458 (v2f64 (REV64v16i8 FPR128:$src))>;
5459 def : Pat<(v2f64 (bitconvert (v4f32 FPR128:$src))),
5460 (v2f64 (REV64v4i32 FPR128:$src))>;
5462 def : Pat<(v2f64 (bitconvert (v2i64 FPR128:$src))), (v2f64 FPR128:$src)>;
5464 let Predicates = [IsLE] in {
5465 def : Pat<(v4f32 (bitconvert (f128 FPR128:$src))), (v4f32 FPR128:$src)>;
5466 def : Pat<(v4f32 (bitconvert (v8i16 FPR128:$src))), (v4f32 FPR128:$src)>;
5467 def : Pat<(v4f32 (bitconvert (v8f16 FPR128:$src))), (v4f32 FPR128:$src)>;
5468 def : Pat<(v4f32 (bitconvert (v16i8 FPR128:$src))), (v4f32 FPR128:$src)>;
5469 def : Pat<(v4f32 (bitconvert (v2i64 FPR128:$src))), (v4f32 FPR128:$src)>;
5470 def : Pat<(v4f32 (bitconvert (v2f64 FPR128:$src))), (v4f32 FPR128:$src)>;
5472 let Predicates = [IsBE] in {
5473 def : Pat<(v4f32 (bitconvert (f128 FPR128:$src))),
5474 (v4f32 (EXTv16i8 (REV64v4i32 FPR128:$src),
5475 (REV64v4i32 FPR128:$src), (i32 8)))>;
5476 def : Pat<(v4f32 (bitconvert (v8i16 FPR128:$src))),
5477 (v4f32 (REV32v8i16 FPR128:$src))>;
5478 def : Pat<(v4f32 (bitconvert (v8f16 FPR128:$src))),
5479 (v4f32 (REV32v8i16 FPR128:$src))>;
5480 def : Pat<(v4f32 (bitconvert (v16i8 FPR128:$src))),
5481 (v4f32 (REV32v16i8 FPR128:$src))>;
5482 def : Pat<(v4f32 (bitconvert (v2i64 FPR128:$src))),
5483 (v4f32 (REV64v4i32 FPR128:$src))>;
5484 def : Pat<(v4f32 (bitconvert (v2f64 FPR128:$src))),
5485 (v4f32 (REV64v4i32 FPR128:$src))>;
5487 def : Pat<(v4f32 (bitconvert (v4i32 FPR128:$src))), (v4f32 FPR128:$src)>;
5489 let Predicates = [IsLE] in {
5490 def : Pat<(v2i64 (bitconvert (f128 FPR128:$src))), (v2i64 FPR128:$src)>;
5491 def : Pat<(v2i64 (bitconvert (v4i32 FPR128:$src))), (v2i64 FPR128:$src)>;
5492 def : Pat<(v2i64 (bitconvert (v8i16 FPR128:$src))), (v2i64 FPR128:$src)>;
5493 def : Pat<(v2i64 (bitconvert (v16i8 FPR128:$src))), (v2i64 FPR128:$src)>;
5494 def : Pat<(v2i64 (bitconvert (v4f32 FPR128:$src))), (v2i64 FPR128:$src)>;
5495 def : Pat<(v2i64 (bitconvert (v8f16 FPR128:$src))), (v2i64 FPR128:$src)>;
5497 let Predicates = [IsBE] in {
5498 def : Pat<(v2i64 (bitconvert (f128 FPR128:$src))),
5499 (v2i64 (EXTv16i8 FPR128:$src,
5500 FPR128:$src, (i32 8)))>;
5501 def : Pat<(v2i64 (bitconvert (v4i32 FPR128:$src))),
5502 (v2i64 (REV64v4i32 FPR128:$src))>;
5503 def : Pat<(v2i64 (bitconvert (v8i16 FPR128:$src))),
5504 (v2i64 (REV64v8i16 FPR128:$src))>;
5505 def : Pat<(v2i64 (bitconvert (v16i8 FPR128:$src))),
5506 (v2i64 (REV64v16i8 FPR128:$src))>;
5507 def : Pat<(v2i64 (bitconvert (v4f32 FPR128:$src))),
5508 (v2i64 (REV64v4i32 FPR128:$src))>;
5509 def : Pat<(v2i64 (bitconvert (v8f16 FPR128:$src))),
5510 (v2i64 (REV64v8i16 FPR128:$src))>;
5512 def : Pat<(v2i64 (bitconvert (v2f64 FPR128:$src))), (v2i64 FPR128:$src)>;
5514 let Predicates = [IsLE] in {
5515 def : Pat<(v4i32 (bitconvert (f128 FPR128:$src))), (v4i32 FPR128:$src)>;
5516 def : Pat<(v4i32 (bitconvert (v2i64 FPR128:$src))), (v4i32 FPR128:$src)>;
5517 def : Pat<(v4i32 (bitconvert (v8i16 FPR128:$src))), (v4i32 FPR128:$src)>;
5518 def : Pat<(v4i32 (bitconvert (v16i8 FPR128:$src))), (v4i32 FPR128:$src)>;
5519 def : Pat<(v4i32 (bitconvert (v2f64 FPR128:$src))), (v4i32 FPR128:$src)>;
5520 def : Pat<(v4i32 (bitconvert (v8f16 FPR128:$src))), (v4i32 FPR128:$src)>;
5522 let Predicates = [IsBE] in {
5523 def : Pat<(v4i32 (bitconvert (f128 FPR128:$src))),
5524 (v4i32 (EXTv16i8 (REV64v4i32 FPR128:$src),
5525 (REV64v4i32 FPR128:$src),
5527 def : Pat<(v4i32 (bitconvert (v2i64 FPR128:$src))),
5528 (v4i32 (REV64v4i32 FPR128:$src))>;
5529 def : Pat<(v4i32 (bitconvert (v8i16 FPR128:$src))),
5530 (v4i32 (REV32v8i16 FPR128:$src))>;
5531 def : Pat<(v4i32 (bitconvert (v16i8 FPR128:$src))),
5532 (v4i32 (REV32v16i8 FPR128:$src))>;
5533 def : Pat<(v4i32 (bitconvert (v2f64 FPR128:$src))),
5534 (v4i32 (REV64v4i32 FPR128:$src))>;
5535 def : Pat<(v4i32 (bitconvert (v8f16 FPR128:$src))),
5536 (v4i32 (REV32v8i16 FPR128:$src))>;
5538 def : Pat<(v4i32 (bitconvert (v4f32 FPR128:$src))), (v4i32 FPR128:$src)>;
5540 let Predicates = [IsLE] in {
5541 def : Pat<(v8i16 (bitconvert (f128 FPR128:$src))), (v8i16 FPR128:$src)>;
5542 def : Pat<(v8i16 (bitconvert (v2i64 FPR128:$src))), (v8i16 FPR128:$src)>;
5543 def : Pat<(v8i16 (bitconvert (v4i32 FPR128:$src))), (v8i16 FPR128:$src)>;
5544 def : Pat<(v8i16 (bitconvert (v16i8 FPR128:$src))), (v8i16 FPR128:$src)>;
5545 def : Pat<(v8i16 (bitconvert (v2f64 FPR128:$src))), (v8i16 FPR128:$src)>;
5546 def : Pat<(v8i16 (bitconvert (v4f32 FPR128:$src))), (v8i16 FPR128:$src)>;
5547 def : Pat<(v8i16 (bitconvert (v8f16 FPR128:$src))), (v8i16 FPR128:$src)>;
5549 let Predicates = [IsBE] in {
5550 def : Pat<(v8i16 (bitconvert (f128 FPR128:$src))),
5551 (v8i16 (EXTv16i8 (REV64v8i16 FPR128:$src),
5552 (REV64v8i16 FPR128:$src),
5554 def : Pat<(v8i16 (bitconvert (v2i64 FPR128:$src))),
5555 (v8i16 (REV64v8i16 FPR128:$src))>;
5556 def : Pat<(v8i16 (bitconvert (v4i32 FPR128:$src))),
5557 (v8i16 (REV32v8i16 FPR128:$src))>;
5558 def : Pat<(v8i16 (bitconvert (v16i8 FPR128:$src))),
5559 (v8i16 (REV16v16i8 FPR128:$src))>;
5560 def : Pat<(v8i16 (bitconvert (v2f64 FPR128:$src))),
5561 (v8i16 (REV64v8i16 FPR128:$src))>;
5562 def : Pat<(v8i16 (bitconvert (v4f32 FPR128:$src))),
5563 (v8i16 (REV32v8i16 FPR128:$src))>;
5564 def : Pat<(v8i16 (bitconvert (v8f16 FPR128:$src))),
5565 (v8i16 (REV32v8i16 FPR128:$src))>;
5568 let Predicates = [IsLE] in {
5569 def : Pat<(v8f16 (bitconvert (f128 FPR128:$src))), (v8f16 FPR128:$src)>;
5570 def : Pat<(v8f16 (bitconvert (v2i64 FPR128:$src))), (v8f16 FPR128:$src)>;
5571 def : Pat<(v8f16 (bitconvert (v4i32 FPR128:$src))), (v8f16 FPR128:$src)>;
5572 def : Pat<(v8f16 (bitconvert (v8i16 FPR128:$src))), (v8f16 FPR128:$src)>;
5573 def : Pat<(v8f16 (bitconvert (v16i8 FPR128:$src))), (v8f16 FPR128:$src)>;
5574 def : Pat<(v8f16 (bitconvert (v2f64 FPR128:$src))), (v8f16 FPR128:$src)>;
5575 def : Pat<(v8f16 (bitconvert (v4f32 FPR128:$src))), (v8f16 FPR128:$src)>;
5577 let Predicates = [IsBE] in {
5578 def : Pat<(v8f16 (bitconvert (f128 FPR128:$src))),
5579 (v8f16 (EXTv16i8 (REV64v8i16 FPR128:$src),
5580 (REV64v8i16 FPR128:$src),
5582 def : Pat<(v8f16 (bitconvert (v2i64 FPR128:$src))),
5583 (v8f16 (REV64v8i16 FPR128:$src))>;
5584 def : Pat<(v8f16 (bitconvert (v4i32 FPR128:$src))),
5585 (v8f16 (REV32v8i16 FPR128:$src))>;
5586 def : Pat<(v8f16 (bitconvert (v8i16 FPR128:$src))),
5587 (v8f16 (REV64v8i16 FPR128:$src))>;
5588 def : Pat<(v8f16 (bitconvert (v16i8 FPR128:$src))),
5589 (v8f16 (REV16v16i8 FPR128:$src))>;
5590 def : Pat<(v8f16 (bitconvert (v2f64 FPR128:$src))),
5591 (v8f16 (REV64v8i16 FPR128:$src))>;
5592 def : Pat<(v8f16 (bitconvert (v4f32 FPR128:$src))),
5593 (v8f16 (REV32v8i16 FPR128:$src))>;
5596 let Predicates = [IsLE] in {
5597 def : Pat<(v16i8 (bitconvert (f128 FPR128:$src))), (v16i8 FPR128:$src)>;
5598 def : Pat<(v16i8 (bitconvert (v2i64 FPR128:$src))), (v16i8 FPR128:$src)>;
5599 def : Pat<(v16i8 (bitconvert (v4i32 FPR128:$src))), (v16i8 FPR128:$src)>;
5600 def : Pat<(v16i8 (bitconvert (v8i16 FPR128:$src))), (v16i8 FPR128:$src)>;
5601 def : Pat<(v16i8 (bitconvert (v2f64 FPR128:$src))), (v16i8 FPR128:$src)>;
5602 def : Pat<(v16i8 (bitconvert (v4f32 FPR128:$src))), (v16i8 FPR128:$src)>;
5603 def : Pat<(v16i8 (bitconvert (v8f16 FPR128:$src))), (v16i8 FPR128:$src)>;
5605 let Predicates = [IsBE] in {
5606 def : Pat<(v16i8 (bitconvert (f128 FPR128:$src))),
5607 (v16i8 (EXTv16i8 (REV64v16i8 FPR128:$src),
5608 (REV64v16i8 FPR128:$src),
5610 def : Pat<(v16i8 (bitconvert (v2i64 FPR128:$src))),
5611 (v16i8 (REV64v16i8 FPR128:$src))>;
5612 def : Pat<(v16i8 (bitconvert (v4i32 FPR128:$src))),
5613 (v16i8 (REV32v16i8 FPR128:$src))>;
5614 def : Pat<(v16i8 (bitconvert (v8i16 FPR128:$src))),
5615 (v16i8 (REV16v16i8 FPR128:$src))>;
5616 def : Pat<(v16i8 (bitconvert (v2f64 FPR128:$src))),
5617 (v16i8 (REV64v16i8 FPR128:$src))>;
5618 def : Pat<(v16i8 (bitconvert (v4f32 FPR128:$src))),
5619 (v16i8 (REV32v16i8 FPR128:$src))>;
5620 def : Pat<(v16i8 (bitconvert (v8f16 FPR128:$src))),
5621 (v16i8 (REV16v16i8 FPR128:$src))>;
5624 def : Pat<(v8i8 (extract_subvector (v16i8 FPR128:$Rn), (i64 1))),
5625 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
5626 def : Pat<(v4i16 (extract_subvector (v8i16 FPR128:$Rn), (i64 1))),
5627 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
5628 def : Pat<(v2i32 (extract_subvector (v4i32 FPR128:$Rn), (i64 1))),
5629 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
5630 def : Pat<(v1i64 (extract_subvector (v2i64 FPR128:$Rn), (i64 1))),
5631 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
5633 // A 64-bit subvector insert to the first 128-bit vector position
5634 // is a subregister copy that needs no instruction.
5635 def : Pat<(insert_subvector undef, (v1i64 FPR64:$src), (i32 0)),
5636 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5637 def : Pat<(insert_subvector undef, (v1f64 FPR64:$src), (i32 0)),
5638 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5639 def : Pat<(insert_subvector undef, (v2i32 FPR64:$src), (i32 0)),
5640 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5641 def : Pat<(insert_subvector undef, (v2f32 FPR64:$src), (i32 0)),
5642 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5643 def : Pat<(insert_subvector undef, (v4i16 FPR64:$src), (i32 0)),
5644 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5645 def : Pat<(insert_subvector undef, (v4f16 FPR64:$src), (i32 0)),
5646 (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5647 def : Pat<(insert_subvector undef, (v8i8 FPR64:$src), (i32 0)),
5648 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5650 // Use pair-wise add instructions when summing up the lanes for v2f64, v2i64
5652 def : Pat<(i64 (add (vector_extract (v2i64 FPR128:$Rn), (i64 0)),
5653 (vector_extract (v2i64 FPR128:$Rn), (i64 1)))),
5654 (i64 (ADDPv2i64p (v2i64 FPR128:$Rn)))>;
5655 def : Pat<(f64 (fadd (vector_extract (v2f64 FPR128:$Rn), (i64 0)),
5656 (vector_extract (v2f64 FPR128:$Rn), (i64 1)))),
5657 (f64 (FADDPv2i64p (v2f64 FPR128:$Rn)))>;
5658 // vector_extract on 64-bit vectors gets promoted to a 128 bit vector,
5659 // so we match on v4f32 here, not v2f32. This will also catch adding
5660 // the low two lanes of a true v4f32 vector.
5661 def : Pat<(fadd (vector_extract (v4f32 FPR128:$Rn), (i64 0)),
5662 (vector_extract (v4f32 FPR128:$Rn), (i64 1))),
5663 (f32 (FADDPv2i32p (EXTRACT_SUBREG FPR128:$Rn, dsub)))>;
5665 // Scalar 64-bit shifts in FPR64 registers.
5666 def : Pat<(i64 (int_aarch64_neon_sshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
5667 (SSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
5668 def : Pat<(i64 (int_aarch64_neon_ushl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
5669 (USHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
5670 def : Pat<(i64 (int_aarch64_neon_srshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
5671 (SRSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
5672 def : Pat<(i64 (int_aarch64_neon_urshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
5673 (URSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
5675 // Tail call return handling. These are all compiler pseudo-instructions,
5676 // so no encoding information or anything like that.
5677 let isCall = 1, isTerminator = 1, isReturn = 1, isBarrier = 1, Uses = [SP] in {
5678 def TCRETURNdi : Pseudo<(outs), (ins i64imm:$dst, i32imm:$FPDiff),[]>;
5679 def TCRETURNri : Pseudo<(outs), (ins tcGPR64:$dst, i32imm:$FPDiff), []>;
5682 def : Pat<(AArch64tcret tcGPR64:$dst, (i32 timm:$FPDiff)),
5683 (TCRETURNri tcGPR64:$dst, imm:$FPDiff)>;
5684 def : Pat<(AArch64tcret tglobaladdr:$dst, (i32 timm:$FPDiff)),
5685 (TCRETURNdi texternalsym:$dst, imm:$FPDiff)>;
5686 def : Pat<(AArch64tcret texternalsym:$dst, (i32 timm:$FPDiff)),
5687 (TCRETURNdi texternalsym:$dst, imm:$FPDiff)>;
5689 include "AArch64InstrAtomics.td"