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>,
99 def SDT_AArch64WrapperLarge : SDTypeProfile<1, 4,
100 [SDTCisVT<0, i64>, SDTCisVT<1, i32>,
101 SDTCisSameAs<1, 2>, SDTCisSameAs<1, 3>,
102 SDTCisSameAs<1, 4>]>;
106 def AArch64adrp : SDNode<"AArch64ISD::ADRP", SDTIntUnaryOp, []>;
107 def AArch64addlow : SDNode<"AArch64ISD::ADDlow", SDTIntBinOp, []>;
108 def AArch64LOADgot : SDNode<"AArch64ISD::LOADgot", SDTIntUnaryOp>;
109 def AArch64callseq_start : SDNode<"ISD::CALLSEQ_START",
110 SDCallSeqStart<[ SDTCisVT<0, i32> ]>,
111 [SDNPHasChain, SDNPOutGlue]>;
112 def AArch64callseq_end : SDNode<"ISD::CALLSEQ_END",
113 SDCallSeqEnd<[ SDTCisVT<0, i32>,
115 [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue]>;
116 def AArch64call : SDNode<"AArch64ISD::CALL",
117 SDTypeProfile<0, -1, [SDTCisPtrTy<0>]>,
118 [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue,
120 def AArch64brcond : SDNode<"AArch64ISD::BRCOND", SDT_AArch64Brcond,
122 def AArch64cbz : SDNode<"AArch64ISD::CBZ", SDT_AArch64cbz,
124 def AArch64cbnz : SDNode<"AArch64ISD::CBNZ", SDT_AArch64cbz,
126 def AArch64tbz : SDNode<"AArch64ISD::TBZ", SDT_AArch64tbz,
128 def AArch64tbnz : SDNode<"AArch64ISD::TBNZ", SDT_AArch64tbz,
132 def AArch64csel : SDNode<"AArch64ISD::CSEL", SDT_AArch64CSel>;
133 def AArch64csinv : SDNode<"AArch64ISD::CSINV", SDT_AArch64CSel>;
134 def AArch64csneg : SDNode<"AArch64ISD::CSNEG", SDT_AArch64CSel>;
135 def AArch64csinc : SDNode<"AArch64ISD::CSINC", SDT_AArch64CSel>;
136 def AArch64retflag : SDNode<"AArch64ISD::RET_FLAG", SDTNone,
137 [SDNPHasChain, SDNPOptInGlue, SDNPVariadic]>;
138 def AArch64adc : SDNode<"AArch64ISD::ADC", SDTBinaryArithWithFlagsIn >;
139 def AArch64sbc : SDNode<"AArch64ISD::SBC", SDTBinaryArithWithFlagsIn>;
140 def AArch64add_flag : SDNode<"AArch64ISD::ADDS", SDTBinaryArithWithFlagsOut,
142 def AArch64sub_flag : SDNode<"AArch64ISD::SUBS", SDTBinaryArithWithFlagsOut>;
143 def AArch64and_flag : SDNode<"AArch64ISD::ANDS", SDTBinaryArithWithFlagsOut,
145 def AArch64adc_flag : SDNode<"AArch64ISD::ADCS", SDTBinaryArithWithFlagsInOut>;
146 def AArch64sbc_flag : SDNode<"AArch64ISD::SBCS", SDTBinaryArithWithFlagsInOut>;
148 def AArch64threadpointer : SDNode<"AArch64ISD::THREAD_POINTER", SDTPtrLeaf>;
150 def AArch64fcmp : SDNode<"AArch64ISD::FCMP", SDT_AArch64FCmp>;
152 def AArch64fmax : SDNode<"AArch64ISD::FMAX", SDTFPBinOp>;
153 def AArch64fmin : SDNode<"AArch64ISD::FMIN", SDTFPBinOp>;
155 def AArch64dup : SDNode<"AArch64ISD::DUP", SDT_AArch64Dup>;
156 def AArch64duplane8 : SDNode<"AArch64ISD::DUPLANE8", SDT_AArch64DupLane>;
157 def AArch64duplane16 : SDNode<"AArch64ISD::DUPLANE16", SDT_AArch64DupLane>;
158 def AArch64duplane32 : SDNode<"AArch64ISD::DUPLANE32", SDT_AArch64DupLane>;
159 def AArch64duplane64 : SDNode<"AArch64ISD::DUPLANE64", SDT_AArch64DupLane>;
161 def AArch64zip1 : SDNode<"AArch64ISD::ZIP1", SDT_AArch64Zip>;
162 def AArch64zip2 : SDNode<"AArch64ISD::ZIP2", SDT_AArch64Zip>;
163 def AArch64uzp1 : SDNode<"AArch64ISD::UZP1", SDT_AArch64Zip>;
164 def AArch64uzp2 : SDNode<"AArch64ISD::UZP2", SDT_AArch64Zip>;
165 def AArch64trn1 : SDNode<"AArch64ISD::TRN1", SDT_AArch64Zip>;
166 def AArch64trn2 : SDNode<"AArch64ISD::TRN2", SDT_AArch64Zip>;
168 def AArch64movi_edit : SDNode<"AArch64ISD::MOVIedit", SDT_AArch64MOVIedit>;
169 def AArch64movi_shift : SDNode<"AArch64ISD::MOVIshift", SDT_AArch64MOVIshift>;
170 def AArch64movi_msl : SDNode<"AArch64ISD::MOVImsl", SDT_AArch64MOVIshift>;
171 def AArch64mvni_shift : SDNode<"AArch64ISD::MVNIshift", SDT_AArch64MOVIshift>;
172 def AArch64mvni_msl : SDNode<"AArch64ISD::MVNImsl", SDT_AArch64MOVIshift>;
173 def AArch64movi : SDNode<"AArch64ISD::MOVI", SDT_AArch64MOVIedit>;
174 def AArch64fmov : SDNode<"AArch64ISD::FMOV", SDT_AArch64MOVIedit>;
176 def AArch64rev16 : SDNode<"AArch64ISD::REV16", SDT_AArch64UnaryVec>;
177 def AArch64rev32 : SDNode<"AArch64ISD::REV32", SDT_AArch64UnaryVec>;
178 def AArch64rev64 : SDNode<"AArch64ISD::REV64", SDT_AArch64UnaryVec>;
179 def AArch64ext : SDNode<"AArch64ISD::EXT", SDT_AArch64ExtVec>;
181 def AArch64vashr : SDNode<"AArch64ISD::VASHR", SDT_AArch64vshift>;
182 def AArch64vlshr : SDNode<"AArch64ISD::VLSHR", SDT_AArch64vshift>;
183 def AArch64vshl : SDNode<"AArch64ISD::VSHL", SDT_AArch64vshift>;
184 def AArch64sqshli : SDNode<"AArch64ISD::SQSHL_I", SDT_AArch64vshift>;
185 def AArch64uqshli : SDNode<"AArch64ISD::UQSHL_I", SDT_AArch64vshift>;
186 def AArch64sqshlui : SDNode<"AArch64ISD::SQSHLU_I", SDT_AArch64vshift>;
187 def AArch64srshri : SDNode<"AArch64ISD::SRSHR_I", SDT_AArch64vshift>;
188 def AArch64urshri : SDNode<"AArch64ISD::URSHR_I", SDT_AArch64vshift>;
190 def AArch64not: SDNode<"AArch64ISD::NOT", SDT_AArch64unvec>;
191 def AArch64bit: SDNode<"AArch64ISD::BIT", SDT_AArch64trivec>;
192 def AArch64bsl: SDNode<"AArch64ISD::BSL", SDT_AArch64trivec>;
194 def AArch64cmeq: SDNode<"AArch64ISD::CMEQ", SDT_AArch64binvec>;
195 def AArch64cmge: SDNode<"AArch64ISD::CMGE", SDT_AArch64binvec>;
196 def AArch64cmgt: SDNode<"AArch64ISD::CMGT", SDT_AArch64binvec>;
197 def AArch64cmhi: SDNode<"AArch64ISD::CMHI", SDT_AArch64binvec>;
198 def AArch64cmhs: SDNode<"AArch64ISD::CMHS", SDT_AArch64binvec>;
200 def AArch64fcmeq: SDNode<"AArch64ISD::FCMEQ", SDT_AArch64fcmp>;
201 def AArch64fcmge: SDNode<"AArch64ISD::FCMGE", SDT_AArch64fcmp>;
202 def AArch64fcmgt: SDNode<"AArch64ISD::FCMGT", SDT_AArch64fcmp>;
204 def AArch64cmeqz: SDNode<"AArch64ISD::CMEQz", SDT_AArch64unvec>;
205 def AArch64cmgez: SDNode<"AArch64ISD::CMGEz", SDT_AArch64unvec>;
206 def AArch64cmgtz: SDNode<"AArch64ISD::CMGTz", SDT_AArch64unvec>;
207 def AArch64cmlez: SDNode<"AArch64ISD::CMLEz", SDT_AArch64unvec>;
208 def AArch64cmltz: SDNode<"AArch64ISD::CMLTz", SDT_AArch64unvec>;
209 def AArch64cmtst : PatFrag<(ops node:$LHS, node:$RHS),
210 (AArch64not (AArch64cmeqz (and node:$LHS, node:$RHS)))>;
212 def AArch64fcmeqz: SDNode<"AArch64ISD::FCMEQz", SDT_AArch64fcmpz>;
213 def AArch64fcmgez: SDNode<"AArch64ISD::FCMGEz", SDT_AArch64fcmpz>;
214 def AArch64fcmgtz: SDNode<"AArch64ISD::FCMGTz", SDT_AArch64fcmpz>;
215 def AArch64fcmlez: SDNode<"AArch64ISD::FCMLEz", SDT_AArch64fcmpz>;
216 def AArch64fcmltz: SDNode<"AArch64ISD::FCMLTz", SDT_AArch64fcmpz>;
218 def AArch64bici: SDNode<"AArch64ISD::BICi", SDT_AArch64vecimm>;
219 def AArch64orri: SDNode<"AArch64ISD::ORRi", SDT_AArch64vecimm>;
221 def AArch64neg : SDNode<"AArch64ISD::NEG", SDT_AArch64unvec>;
223 def AArch64tcret: SDNode<"AArch64ISD::TC_RETURN", SDT_AArch64TCRET,
224 [SDNPHasChain, SDNPOptInGlue, SDNPVariadic]>;
226 def AArch64Prefetch : SDNode<"AArch64ISD::PREFETCH", SDT_AArch64PREFETCH,
227 [SDNPHasChain, SDNPSideEffect]>;
229 def AArch64sitof: SDNode<"AArch64ISD::SITOF", SDT_AArch64ITOF>;
230 def AArch64uitof: SDNode<"AArch64ISD::UITOF", SDT_AArch64ITOF>;
232 def AArch64tlsdesc_call : SDNode<"AArch64ISD::TLSDESC_CALL",
233 SDT_AArch64TLSDescCall,
234 [SDNPInGlue, SDNPOutGlue, SDNPHasChain,
237 def AArch64WrapperLarge : SDNode<"AArch64ISD::WrapperLarge",
238 SDT_AArch64WrapperLarge>;
240 def AArch64NvCast : SDNode<"AArch64ISD::NVCAST", SDTUnaryOp>;
242 def SDT_AArch64mull : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
243 SDTCisSameAs<1, 2>]>;
244 def AArch64smull : SDNode<"AArch64ISD::SMULL", SDT_AArch64mull>;
245 def AArch64umull : SDNode<"AArch64ISD::UMULL", SDT_AArch64mull>;
247 //===----------------------------------------------------------------------===//
249 //===----------------------------------------------------------------------===//
251 // AArch64 Instruction Predicate Definitions.
253 def HasZCZ : Predicate<"Subtarget->hasZeroCycleZeroing()">;
254 def NoZCZ : Predicate<"!Subtarget->hasZeroCycleZeroing()">;
255 def IsDarwin : Predicate<"Subtarget->isTargetDarwin()">;
256 def IsNotDarwin: Predicate<"!Subtarget->isTargetDarwin()">;
257 def ForCodeSize : Predicate<"ForCodeSize">;
258 def NotForCodeSize : Predicate<"!ForCodeSize">;
260 include "AArch64InstrFormats.td"
262 //===----------------------------------------------------------------------===//
264 //===----------------------------------------------------------------------===//
265 // Miscellaneous instructions.
266 //===----------------------------------------------------------------------===//
268 let Defs = [SP], Uses = [SP], hasSideEffects = 1, isCodeGenOnly = 1 in {
269 def ADJCALLSTACKDOWN : Pseudo<(outs), (ins i32imm:$amt),
270 [(AArch64callseq_start timm:$amt)]>;
271 def ADJCALLSTACKUP : Pseudo<(outs), (ins i32imm:$amt1, i32imm:$amt2),
272 [(AArch64callseq_end timm:$amt1, timm:$amt2)]>;
273 } // Defs = [SP], Uses = [SP], hasSideEffects = 1, isCodeGenOnly = 1
275 let isReMaterializable = 1, isCodeGenOnly = 1 in {
276 // FIXME: The following pseudo instructions are only needed because remat
277 // cannot handle multiple instructions. When that changes, they can be
278 // removed, along with the AArch64Wrapper node.
280 let AddedComplexity = 10 in
281 def LOADgot : Pseudo<(outs GPR64:$dst), (ins i64imm:$addr),
282 [(set GPR64:$dst, (AArch64LOADgot tglobaladdr:$addr))]>,
285 // The MOVaddr instruction should match only when the add is not folded
286 // into a load or store address.
288 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
289 [(set GPR64:$dst, (AArch64addlow (AArch64adrp tglobaladdr:$hi),
290 tglobaladdr:$low))]>,
291 Sched<[WriteAdrAdr]>;
293 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
294 [(set GPR64:$dst, (AArch64addlow (AArch64adrp tjumptable:$hi),
296 Sched<[WriteAdrAdr]>;
298 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
299 [(set GPR64:$dst, (AArch64addlow (AArch64adrp tconstpool:$hi),
301 Sched<[WriteAdrAdr]>;
303 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
304 [(set GPR64:$dst, (AArch64addlow (AArch64adrp tblockaddress:$hi),
305 tblockaddress:$low))]>,
306 Sched<[WriteAdrAdr]>;
308 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
309 [(set GPR64:$dst, (AArch64addlow (AArch64adrp tglobaltlsaddr:$hi),
310 tglobaltlsaddr:$low))]>,
311 Sched<[WriteAdrAdr]>;
313 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
314 [(set GPR64:$dst, (AArch64addlow (AArch64adrp texternalsym:$hi),
315 texternalsym:$low))]>,
316 Sched<[WriteAdrAdr]>;
318 } // isReMaterializable, isCodeGenOnly
320 def : Pat<(AArch64LOADgot tglobaltlsaddr:$addr),
321 (LOADgot tglobaltlsaddr:$addr)>;
323 def : Pat<(AArch64LOADgot texternalsym:$addr),
324 (LOADgot texternalsym:$addr)>;
326 def : Pat<(AArch64LOADgot tconstpool:$addr),
327 (LOADgot tconstpool:$addr)>;
329 //===----------------------------------------------------------------------===//
330 // System instructions.
331 //===----------------------------------------------------------------------===//
333 def HINT : HintI<"hint">;
334 def : InstAlias<"nop", (HINT 0b000)>;
335 def : InstAlias<"yield",(HINT 0b001)>;
336 def : InstAlias<"wfe", (HINT 0b010)>;
337 def : InstAlias<"wfi", (HINT 0b011)>;
338 def : InstAlias<"sev", (HINT 0b100)>;
339 def : InstAlias<"sevl", (HINT 0b101)>;
341 // As far as LLVM is concerned this writes to the system's exclusive monitors.
342 let mayLoad = 1, mayStore = 1 in
343 def CLREX : CRmSystemI<imm0_15, 0b010, "clrex">;
345 // NOTE: ideally, this would have mayStore = 0, mayLoad = 0, but we cannot
346 // model patterns with sufficiently fine granularity.
347 let mayLoad = ?, mayStore = ? in {
348 def DMB : CRmSystemI<barrier_op, 0b101, "dmb",
349 [(int_aarch64_dmb (i32 imm32_0_15:$CRm))]>;
351 def DSB : CRmSystemI<barrier_op, 0b100, "dsb",
352 [(int_aarch64_dsb (i32 imm32_0_15:$CRm))]>;
354 def ISB : CRmSystemI<barrier_op, 0b110, "isb",
355 [(int_aarch64_isb (i32 imm32_0_15:$CRm))]>;
358 def : InstAlias<"clrex", (CLREX 0xf)>;
359 def : InstAlias<"isb", (ISB 0xf)>;
363 def MSRpstate: MSRpstateI;
365 // The thread pointer (on Linux, at least, where this has been implemented) is
367 def : Pat<(AArch64threadpointer), (MRS 0xde82)>;
369 // Generic system instructions
370 def SYSxt : SystemXtI<0, "sys">;
371 def SYSLxt : SystemLXtI<1, "sysl">;
373 def : InstAlias<"sys $op1, $Cn, $Cm, $op2",
374 (SYSxt imm0_7:$op1, sys_cr_op:$Cn,
375 sys_cr_op:$Cm, imm0_7:$op2, XZR)>;
377 //===----------------------------------------------------------------------===//
378 // Move immediate instructions.
379 //===----------------------------------------------------------------------===//
381 defm MOVK : InsertImmediate<0b11, "movk">;
382 defm MOVN : MoveImmediate<0b00, "movn">;
384 let PostEncoderMethod = "fixMOVZ" in
385 defm MOVZ : MoveImmediate<0b10, "movz">;
387 // First group of aliases covers an implicit "lsl #0".
388 def : InstAlias<"movk $dst, $imm", (MOVKWi GPR32:$dst, imm0_65535:$imm, 0)>;
389 def : InstAlias<"movk $dst, $imm", (MOVKXi GPR64:$dst, imm0_65535:$imm, 0)>;
390 def : InstAlias<"movn $dst, $imm", (MOVNWi GPR32:$dst, imm0_65535:$imm, 0)>;
391 def : InstAlias<"movn $dst, $imm", (MOVNXi GPR64:$dst, imm0_65535:$imm, 0)>;
392 def : InstAlias<"movz $dst, $imm", (MOVZWi GPR32:$dst, imm0_65535:$imm, 0)>;
393 def : InstAlias<"movz $dst, $imm", (MOVZXi GPR64:$dst, imm0_65535:$imm, 0)>;
395 // Next, we have various ELF relocations with the ":XYZ_g0:sym" syntax.
396 def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movz_symbol_g3:$sym, 48)>;
397 def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movz_symbol_g2:$sym, 32)>;
398 def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movz_symbol_g1:$sym, 16)>;
399 def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movz_symbol_g0:$sym, 0)>;
401 def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movz_symbol_g3:$sym, 48)>;
402 def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movz_symbol_g2:$sym, 32)>;
403 def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movz_symbol_g1:$sym, 16)>;
404 def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movz_symbol_g0:$sym, 0)>;
406 def : InstAlias<"movk $Rd, $sym", (MOVKXi GPR64:$Rd, movk_symbol_g3:$sym, 48)>;
407 def : InstAlias<"movk $Rd, $sym", (MOVKXi GPR64:$Rd, movk_symbol_g2:$sym, 32)>;
408 def : InstAlias<"movk $Rd, $sym", (MOVKXi GPR64:$Rd, movk_symbol_g1:$sym, 16)>;
409 def : InstAlias<"movk $Rd, $sym", (MOVKXi GPR64:$Rd, movk_symbol_g0:$sym, 0)>;
411 def : InstAlias<"movz $Rd, $sym", (MOVZWi GPR32:$Rd, movz_symbol_g1:$sym, 16)>;
412 def : InstAlias<"movz $Rd, $sym", (MOVZWi GPR32:$Rd, movz_symbol_g0:$sym, 0)>;
414 def : InstAlias<"movn $Rd, $sym", (MOVNWi GPR32:$Rd, movz_symbol_g1:$sym, 16)>;
415 def : InstAlias<"movn $Rd, $sym", (MOVNWi GPR32:$Rd, movz_symbol_g0:$sym, 0)>;
417 def : InstAlias<"movk $Rd, $sym", (MOVKWi GPR32:$Rd, movk_symbol_g1:$sym, 16)>;
418 def : InstAlias<"movk $Rd, $sym", (MOVKWi GPR32:$Rd, movk_symbol_g0:$sym, 0)>;
420 // Final group of aliases covers true "mov $Rd, $imm" cases.
421 multiclass movw_mov_alias<string basename,Instruction INST, RegisterClass GPR,
422 int width, int shift> {
423 def _asmoperand : AsmOperandClass {
424 let Name = basename # width # "_lsl" # shift # "MovAlias";
425 let PredicateMethod = "is" # basename # "MovAlias<" # width # ", "
427 let RenderMethod = "add" # basename # "MovAliasOperands<" # shift # ">";
430 def _movimm : Operand<i32> {
431 let ParserMatchClass = !cast<AsmOperandClass>(NAME # "_asmoperand");
434 def : InstAlias<"mov $Rd, $imm",
435 (INST GPR:$Rd, !cast<Operand>(NAME # "_movimm"):$imm, shift)>;
438 defm : movw_mov_alias<"MOVZ", MOVZWi, GPR32, 32, 0>;
439 defm : movw_mov_alias<"MOVZ", MOVZWi, GPR32, 32, 16>;
441 defm : movw_mov_alias<"MOVZ", MOVZXi, GPR64, 64, 0>;
442 defm : movw_mov_alias<"MOVZ", MOVZXi, GPR64, 64, 16>;
443 defm : movw_mov_alias<"MOVZ", MOVZXi, GPR64, 64, 32>;
444 defm : movw_mov_alias<"MOVZ", MOVZXi, GPR64, 64, 48>;
446 defm : movw_mov_alias<"MOVN", MOVNWi, GPR32, 32, 0>;
447 defm : movw_mov_alias<"MOVN", MOVNWi, GPR32, 32, 16>;
449 defm : movw_mov_alias<"MOVN", MOVNXi, GPR64, 64, 0>;
450 defm : movw_mov_alias<"MOVN", MOVNXi, GPR64, 64, 16>;
451 defm : movw_mov_alias<"MOVN", MOVNXi, GPR64, 64, 32>;
452 defm : movw_mov_alias<"MOVN", MOVNXi, GPR64, 64, 48>;
454 let isReMaterializable = 1, isCodeGenOnly = 1, isMoveImm = 1,
455 isAsCheapAsAMove = 1 in {
456 // FIXME: The following pseudo instructions are only needed because remat
457 // cannot handle multiple instructions. When that changes, we can select
458 // directly to the real instructions and get rid of these pseudos.
461 : Pseudo<(outs GPR32:$dst), (ins i32imm:$src),
462 [(set GPR32:$dst, imm:$src)]>,
465 : Pseudo<(outs GPR64:$dst), (ins i64imm:$src),
466 [(set GPR64:$dst, imm:$src)]>,
468 } // isReMaterializable, isCodeGenOnly
470 // If possible, we want to use MOVi32imm even for 64-bit moves. This gives the
471 // eventual expansion code fewer bits to worry about getting right. Marshalling
472 // the types is a little tricky though:
473 def i64imm_32bit : ImmLeaf<i64, [{
474 return (Imm & 0xffffffffULL) == static_cast<uint64_t>(Imm);
477 def trunc_imm : SDNodeXForm<imm, [{
478 return CurDAG->getTargetConstant(N->getZExtValue(), MVT::i32);
481 def : Pat<(i64 i64imm_32bit:$src),
482 (SUBREG_TO_REG (i64 0), (MOVi32imm (trunc_imm imm:$src)), sub_32)>;
484 // Deal with the various forms of (ELF) large addressing with MOVZ/MOVK
486 def : Pat<(AArch64WrapperLarge tglobaladdr:$g3, tglobaladdr:$g2,
487 tglobaladdr:$g1, tglobaladdr:$g0),
488 (MOVKXi (MOVKXi (MOVKXi (MOVZXi tglobaladdr:$g3, 48),
489 tglobaladdr:$g2, 32),
490 tglobaladdr:$g1, 16),
491 tglobaladdr:$g0, 0)>;
493 def : Pat<(AArch64WrapperLarge tblockaddress:$g3, tblockaddress:$g2,
494 tblockaddress:$g1, tblockaddress:$g0),
495 (MOVKXi (MOVKXi (MOVKXi (MOVZXi tblockaddress:$g3, 48),
496 tblockaddress:$g2, 32),
497 tblockaddress:$g1, 16),
498 tblockaddress:$g0, 0)>;
500 def : Pat<(AArch64WrapperLarge tconstpool:$g3, tconstpool:$g2,
501 tconstpool:$g1, tconstpool:$g0),
502 (MOVKXi (MOVKXi (MOVKXi (MOVZXi tconstpool:$g3, 48),
507 def : Pat<(AArch64WrapperLarge tjumptable:$g3, tjumptable:$g2,
508 tjumptable:$g1, tjumptable:$g0),
509 (MOVKXi (MOVKXi (MOVKXi (MOVZXi tjumptable:$g3, 48),
515 //===----------------------------------------------------------------------===//
516 // Arithmetic instructions.
517 //===----------------------------------------------------------------------===//
519 // Add/subtract with carry.
520 defm ADC : AddSubCarry<0, "adc", "adcs", AArch64adc, AArch64adc_flag>;
521 defm SBC : AddSubCarry<1, "sbc", "sbcs", AArch64sbc, AArch64sbc_flag>;
523 def : InstAlias<"ngc $dst, $src", (SBCWr GPR32:$dst, WZR, GPR32:$src)>;
524 def : InstAlias<"ngc $dst, $src", (SBCXr GPR64:$dst, XZR, GPR64:$src)>;
525 def : InstAlias<"ngcs $dst, $src", (SBCSWr GPR32:$dst, WZR, GPR32:$src)>;
526 def : InstAlias<"ngcs $dst, $src", (SBCSXr GPR64:$dst, XZR, GPR64:$src)>;
529 defm ADD : AddSub<0, "add", add>;
530 defm SUB : AddSub<1, "sub">;
532 def : InstAlias<"mov $dst, $src",
533 (ADDWri GPR32sponly:$dst, GPR32sp:$src, 0, 0)>;
534 def : InstAlias<"mov $dst, $src",
535 (ADDWri GPR32sp:$dst, GPR32sponly:$src, 0, 0)>;
536 def : InstAlias<"mov $dst, $src",
537 (ADDXri GPR64sponly:$dst, GPR64sp:$src, 0, 0)>;
538 def : InstAlias<"mov $dst, $src",
539 (ADDXri GPR64sp:$dst, GPR64sponly:$src, 0, 0)>;
541 defm ADDS : AddSubS<0, "adds", AArch64add_flag, "cmn">;
542 defm SUBS : AddSubS<1, "subs", AArch64sub_flag, "cmp">;
544 // Use SUBS instead of SUB to enable CSE between SUBS and SUB.
545 def : Pat<(sub GPR32sp:$Rn, addsub_shifted_imm32:$imm),
546 (SUBSWri GPR32sp:$Rn, addsub_shifted_imm32:$imm)>;
547 def : Pat<(sub GPR64sp:$Rn, addsub_shifted_imm64:$imm),
548 (SUBSXri GPR64sp:$Rn, addsub_shifted_imm64:$imm)>;
549 def : Pat<(sub GPR32:$Rn, GPR32:$Rm),
550 (SUBSWrr GPR32:$Rn, GPR32:$Rm)>;
551 def : Pat<(sub GPR64:$Rn, GPR64:$Rm),
552 (SUBSXrr GPR64:$Rn, GPR64:$Rm)>;
553 def : Pat<(sub GPR32:$Rn, arith_shifted_reg32:$Rm),
554 (SUBSWrs GPR32:$Rn, arith_shifted_reg32:$Rm)>;
555 def : Pat<(sub GPR64:$Rn, arith_shifted_reg64:$Rm),
556 (SUBSXrs GPR64:$Rn, arith_shifted_reg64:$Rm)>;
557 def : Pat<(sub GPR32sp:$R2, arith_extended_reg32<i32>:$R3),
558 (SUBSWrx GPR32sp:$R2, arith_extended_reg32<i32>:$R3)>;
559 def : Pat<(sub GPR64sp:$R2, arith_extended_reg32to64<i64>:$R3),
560 (SUBSXrx GPR64sp:$R2, arith_extended_reg32to64<i64>:$R3)>;
562 // Because of the immediate format for add/sub-imm instructions, the
563 // expression (add x, -1) must be transformed to (SUB{W,X}ri x, 1).
564 // These patterns capture that transformation.
565 let AddedComplexity = 1 in {
566 def : Pat<(add GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
567 (SUBSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
568 def : Pat<(add GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
569 (SUBSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
570 def : Pat<(sub GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
571 (ADDWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
572 def : Pat<(sub GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
573 (ADDXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
576 // Because of the immediate format for add/sub-imm instructions, the
577 // expression (add x, -1) must be transformed to (SUB{W,X}ri x, 1).
578 // These patterns capture that transformation.
579 let AddedComplexity = 1 in {
580 def : Pat<(AArch64add_flag GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
581 (SUBSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
582 def : Pat<(AArch64add_flag GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
583 (SUBSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
584 def : Pat<(AArch64sub_flag GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
585 (ADDSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
586 def : Pat<(AArch64sub_flag GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
587 (ADDSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
590 def : InstAlias<"neg $dst, $src", (SUBWrs GPR32:$dst, WZR, GPR32:$src, 0), 3>;
591 def : InstAlias<"neg $dst, $src", (SUBXrs GPR64:$dst, XZR, GPR64:$src, 0), 3>;
592 def : InstAlias<"neg $dst, $src$shift",
593 (SUBWrs GPR32:$dst, WZR, GPR32:$src, arith_shift32:$shift), 2>;
594 def : InstAlias<"neg $dst, $src$shift",
595 (SUBXrs GPR64:$dst, XZR, GPR64:$src, arith_shift64:$shift), 2>;
597 def : InstAlias<"negs $dst, $src", (SUBSWrs GPR32:$dst, WZR, GPR32:$src, 0), 3>;
598 def : InstAlias<"negs $dst, $src", (SUBSXrs GPR64:$dst, XZR, GPR64:$src, 0), 3>;
599 def : InstAlias<"negs $dst, $src$shift",
600 (SUBSWrs GPR32:$dst, WZR, GPR32:$src, arith_shift32:$shift), 2>;
601 def : InstAlias<"negs $dst, $src$shift",
602 (SUBSXrs GPR64:$dst, XZR, GPR64:$src, arith_shift64:$shift), 2>;
605 // Unsigned/Signed divide
606 defm UDIV : Div<0, "udiv", udiv>;
607 defm SDIV : Div<1, "sdiv", sdiv>;
608 let isCodeGenOnly = 1 in {
609 defm UDIV_Int : Div<0, "udiv", int_aarch64_udiv>;
610 defm SDIV_Int : Div<1, "sdiv", int_aarch64_sdiv>;
614 defm ASRV : Shift<0b10, "asr", sra>;
615 defm LSLV : Shift<0b00, "lsl", shl>;
616 defm LSRV : Shift<0b01, "lsr", srl>;
617 defm RORV : Shift<0b11, "ror", rotr>;
619 def : ShiftAlias<"asrv", ASRVWr, GPR32>;
620 def : ShiftAlias<"asrv", ASRVXr, GPR64>;
621 def : ShiftAlias<"lslv", LSLVWr, GPR32>;
622 def : ShiftAlias<"lslv", LSLVXr, GPR64>;
623 def : ShiftAlias<"lsrv", LSRVWr, GPR32>;
624 def : ShiftAlias<"lsrv", LSRVXr, GPR64>;
625 def : ShiftAlias<"rorv", RORVWr, GPR32>;
626 def : ShiftAlias<"rorv", RORVXr, GPR64>;
629 let AddedComplexity = 7 in {
630 defm MADD : MulAccum<0, "madd", add>;
631 defm MSUB : MulAccum<1, "msub", sub>;
633 def : Pat<(i32 (mul GPR32:$Rn, GPR32:$Rm)),
634 (MADDWrrr GPR32:$Rn, GPR32:$Rm, WZR)>;
635 def : Pat<(i64 (mul GPR64:$Rn, GPR64:$Rm)),
636 (MADDXrrr GPR64:$Rn, GPR64:$Rm, XZR)>;
638 def : Pat<(i32 (ineg (mul GPR32:$Rn, GPR32:$Rm))),
639 (MSUBWrrr GPR32:$Rn, GPR32:$Rm, WZR)>;
640 def : Pat<(i64 (ineg (mul GPR64:$Rn, GPR64:$Rm))),
641 (MSUBXrrr GPR64:$Rn, GPR64:$Rm, XZR)>;
642 } // AddedComplexity = 7
644 let AddedComplexity = 5 in {
645 def SMADDLrrr : WideMulAccum<0, 0b001, "smaddl", add, sext>;
646 def SMSUBLrrr : WideMulAccum<1, 0b001, "smsubl", sub, sext>;
647 def UMADDLrrr : WideMulAccum<0, 0b101, "umaddl", add, zext>;
648 def UMSUBLrrr : WideMulAccum<1, 0b101, "umsubl", sub, zext>;
650 def : Pat<(i64 (mul (sext GPR32:$Rn), (sext GPR32:$Rm))),
651 (SMADDLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
652 def : Pat<(i64 (mul (zext GPR32:$Rn), (zext GPR32:$Rm))),
653 (UMADDLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
655 def : Pat<(i64 (ineg (mul (sext GPR32:$Rn), (sext GPR32:$Rm)))),
656 (SMSUBLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
657 def : Pat<(i64 (ineg (mul (zext GPR32:$Rn), (zext GPR32:$Rm)))),
658 (UMSUBLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
659 } // AddedComplexity = 5
661 def : MulAccumWAlias<"mul", MADDWrrr>;
662 def : MulAccumXAlias<"mul", MADDXrrr>;
663 def : MulAccumWAlias<"mneg", MSUBWrrr>;
664 def : MulAccumXAlias<"mneg", MSUBXrrr>;
665 def : WideMulAccumAlias<"smull", SMADDLrrr>;
666 def : WideMulAccumAlias<"smnegl", SMSUBLrrr>;
667 def : WideMulAccumAlias<"umull", UMADDLrrr>;
668 def : WideMulAccumAlias<"umnegl", UMSUBLrrr>;
671 def SMULHrr : MulHi<0b010, "smulh", mulhs>;
672 def UMULHrr : MulHi<0b110, "umulh", mulhu>;
675 def CRC32Brr : BaseCRC32<0, 0b00, 0, GPR32, int_aarch64_crc32b, "crc32b">;
676 def CRC32Hrr : BaseCRC32<0, 0b01, 0, GPR32, int_aarch64_crc32h, "crc32h">;
677 def CRC32Wrr : BaseCRC32<0, 0b10, 0, GPR32, int_aarch64_crc32w, "crc32w">;
678 def CRC32Xrr : BaseCRC32<1, 0b11, 0, GPR64, int_aarch64_crc32x, "crc32x">;
680 def CRC32CBrr : BaseCRC32<0, 0b00, 1, GPR32, int_aarch64_crc32cb, "crc32cb">;
681 def CRC32CHrr : BaseCRC32<0, 0b01, 1, GPR32, int_aarch64_crc32ch, "crc32ch">;
682 def CRC32CWrr : BaseCRC32<0, 0b10, 1, GPR32, int_aarch64_crc32cw, "crc32cw">;
683 def CRC32CXrr : BaseCRC32<1, 0b11, 1, GPR64, int_aarch64_crc32cx, "crc32cx">;
686 //===----------------------------------------------------------------------===//
687 // Logical instructions.
688 //===----------------------------------------------------------------------===//
691 defm ANDS : LogicalImmS<0b11, "ands", AArch64and_flag, "bics">;
692 defm AND : LogicalImm<0b00, "and", and, "bic">;
693 defm EOR : LogicalImm<0b10, "eor", xor, "eon">;
694 defm ORR : LogicalImm<0b01, "orr", or, "orn">;
696 // FIXME: these aliases *are* canonical sometimes (when movz can't be
697 // used). Actually, it seems to be working right now, but putting logical_immXX
698 // here is a bit dodgy on the AsmParser side too.
699 def : InstAlias<"mov $dst, $imm", (ORRWri GPR32sp:$dst, WZR,
700 logical_imm32:$imm), 0>;
701 def : InstAlias<"mov $dst, $imm", (ORRXri GPR64sp:$dst, XZR,
702 logical_imm64:$imm), 0>;
706 defm ANDS : LogicalRegS<0b11, 0, "ands", AArch64and_flag>;
707 defm BICS : LogicalRegS<0b11, 1, "bics",
708 BinOpFrag<(AArch64and_flag node:$LHS, (not node:$RHS))>>;
709 defm AND : LogicalReg<0b00, 0, "and", and>;
710 defm BIC : LogicalReg<0b00, 1, "bic",
711 BinOpFrag<(and node:$LHS, (not node:$RHS))>>;
712 defm EON : LogicalReg<0b10, 1, "eon",
713 BinOpFrag<(xor node:$LHS, (not node:$RHS))>>;
714 defm EOR : LogicalReg<0b10, 0, "eor", xor>;
715 defm ORN : LogicalReg<0b01, 1, "orn",
716 BinOpFrag<(or node:$LHS, (not node:$RHS))>>;
717 defm ORR : LogicalReg<0b01, 0, "orr", or>;
719 def : InstAlias<"mov $dst, $src", (ORRWrs GPR32:$dst, WZR, GPR32:$src, 0), 2>;
720 def : InstAlias<"mov $dst, $src", (ORRXrs GPR64:$dst, XZR, GPR64:$src, 0), 2>;
722 def : InstAlias<"mvn $Wd, $Wm", (ORNWrs GPR32:$Wd, WZR, GPR32:$Wm, 0), 3>;
723 def : InstAlias<"mvn $Xd, $Xm", (ORNXrs GPR64:$Xd, XZR, GPR64:$Xm, 0), 3>;
725 def : InstAlias<"mvn $Wd, $Wm$sh",
726 (ORNWrs GPR32:$Wd, WZR, GPR32:$Wm, logical_shift32:$sh), 2>;
727 def : InstAlias<"mvn $Xd, $Xm$sh",
728 (ORNXrs GPR64:$Xd, XZR, GPR64:$Xm, logical_shift64:$sh), 2>;
730 def : InstAlias<"tst $src1, $src2",
731 (ANDSWri WZR, GPR32:$src1, logical_imm32:$src2), 2>;
732 def : InstAlias<"tst $src1, $src2",
733 (ANDSXri XZR, GPR64:$src1, logical_imm64:$src2), 2>;
735 def : InstAlias<"tst $src1, $src2",
736 (ANDSWrs WZR, GPR32:$src1, GPR32:$src2, 0), 3>;
737 def : InstAlias<"tst $src1, $src2",
738 (ANDSXrs XZR, GPR64:$src1, GPR64:$src2, 0), 3>;
740 def : InstAlias<"tst $src1, $src2$sh",
741 (ANDSWrs WZR, GPR32:$src1, GPR32:$src2, logical_shift32:$sh), 2>;
742 def : InstAlias<"tst $src1, $src2$sh",
743 (ANDSXrs XZR, GPR64:$src1, GPR64:$src2, logical_shift64:$sh), 2>;
746 def : Pat<(not GPR32:$Wm), (ORNWrr WZR, GPR32:$Wm)>;
747 def : Pat<(not GPR64:$Xm), (ORNXrr XZR, GPR64:$Xm)>;
750 //===----------------------------------------------------------------------===//
751 // One operand data processing instructions.
752 //===----------------------------------------------------------------------===//
754 defm CLS : OneOperandData<0b101, "cls">;
755 defm CLZ : OneOperandData<0b100, "clz", ctlz>;
756 defm RBIT : OneOperandData<0b000, "rbit">;
758 def : Pat<(int_aarch64_rbit GPR32:$Rn), (RBITWr $Rn)>;
759 def : Pat<(int_aarch64_rbit GPR64:$Rn), (RBITXr $Rn)>;
761 def REV16Wr : OneWRegData<0b001, "rev16",
762 UnOpFrag<(rotr (bswap node:$LHS), (i64 16))>>;
763 def REV16Xr : OneXRegData<0b001, "rev16", null_frag>;
765 def : Pat<(cttz GPR32:$Rn),
766 (CLZWr (RBITWr GPR32:$Rn))>;
767 def : Pat<(cttz GPR64:$Rn),
768 (CLZXr (RBITXr GPR64:$Rn))>;
769 def : Pat<(ctlz (or (shl (xor (sra GPR32:$Rn, (i64 31)), GPR32:$Rn), (i64 1)),
772 def : Pat<(ctlz (or (shl (xor (sra GPR64:$Rn, (i64 63)), GPR64:$Rn), (i64 1)),
776 // Unlike the other one operand instructions, the instructions with the "rev"
777 // mnemonic do *not* just different in the size bit, but actually use different
778 // opcode bits for the different sizes.
779 def REVWr : OneWRegData<0b010, "rev", bswap>;
780 def REVXr : OneXRegData<0b011, "rev", bswap>;
781 def REV32Xr : OneXRegData<0b010, "rev32",
782 UnOpFrag<(rotr (bswap node:$LHS), (i64 32))>>;
784 // The bswap commutes with the rotr so we want a pattern for both possible
786 def : Pat<(bswap (rotr GPR32:$Rn, (i64 16))), (REV16Wr GPR32:$Rn)>;
787 def : Pat<(bswap (rotr GPR64:$Rn, (i64 32))), (REV32Xr GPR64:$Rn)>;
789 //===----------------------------------------------------------------------===//
790 // Bitfield immediate extraction instruction.
791 //===----------------------------------------------------------------------===//
792 let neverHasSideEffects = 1 in
793 defm EXTR : ExtractImm<"extr">;
794 def : InstAlias<"ror $dst, $src, $shift",
795 (EXTRWrri GPR32:$dst, GPR32:$src, GPR32:$src, imm0_31:$shift)>;
796 def : InstAlias<"ror $dst, $src, $shift",
797 (EXTRXrri GPR64:$dst, GPR64:$src, GPR64:$src, imm0_63:$shift)>;
799 def : Pat<(rotr GPR32:$Rn, (i64 imm0_31:$imm)),
800 (EXTRWrri GPR32:$Rn, GPR32:$Rn, imm0_31:$imm)>;
801 def : Pat<(rotr GPR64:$Rn, (i64 imm0_63:$imm)),
802 (EXTRXrri GPR64:$Rn, GPR64:$Rn, imm0_63:$imm)>;
804 //===----------------------------------------------------------------------===//
805 // Other bitfield immediate instructions.
806 //===----------------------------------------------------------------------===//
807 let neverHasSideEffects = 1 in {
808 defm BFM : BitfieldImmWith2RegArgs<0b01, "bfm">;
809 defm SBFM : BitfieldImm<0b00, "sbfm">;
810 defm UBFM : BitfieldImm<0b10, "ubfm">;
813 def i32shift_a : Operand<i64>, SDNodeXForm<imm, [{
814 uint64_t enc = (32 - N->getZExtValue()) & 0x1f;
815 return CurDAG->getTargetConstant(enc, MVT::i64);
818 def i32shift_b : Operand<i64>, SDNodeXForm<imm, [{
819 uint64_t enc = 31 - N->getZExtValue();
820 return CurDAG->getTargetConstant(enc, MVT::i64);
823 // min(7, 31 - shift_amt)
824 def i32shift_sext_i8 : Operand<i64>, SDNodeXForm<imm, [{
825 uint64_t enc = 31 - N->getZExtValue();
826 enc = enc > 7 ? 7 : enc;
827 return CurDAG->getTargetConstant(enc, MVT::i64);
830 // min(15, 31 - shift_amt)
831 def i32shift_sext_i16 : Operand<i64>, SDNodeXForm<imm, [{
832 uint64_t enc = 31 - N->getZExtValue();
833 enc = enc > 15 ? 15 : enc;
834 return CurDAG->getTargetConstant(enc, MVT::i64);
837 def i64shift_a : Operand<i64>, SDNodeXForm<imm, [{
838 uint64_t enc = (64 - N->getZExtValue()) & 0x3f;
839 return CurDAG->getTargetConstant(enc, MVT::i64);
842 def i64shift_b : Operand<i64>, SDNodeXForm<imm, [{
843 uint64_t enc = 63 - N->getZExtValue();
844 return CurDAG->getTargetConstant(enc, MVT::i64);
847 // min(7, 63 - shift_amt)
848 def i64shift_sext_i8 : Operand<i64>, SDNodeXForm<imm, [{
849 uint64_t enc = 63 - N->getZExtValue();
850 enc = enc > 7 ? 7 : enc;
851 return CurDAG->getTargetConstant(enc, MVT::i64);
854 // min(15, 63 - shift_amt)
855 def i64shift_sext_i16 : Operand<i64>, SDNodeXForm<imm, [{
856 uint64_t enc = 63 - N->getZExtValue();
857 enc = enc > 15 ? 15 : enc;
858 return CurDAG->getTargetConstant(enc, MVT::i64);
861 // min(31, 63 - shift_amt)
862 def i64shift_sext_i32 : Operand<i64>, SDNodeXForm<imm, [{
863 uint64_t enc = 63 - N->getZExtValue();
864 enc = enc > 31 ? 31 : enc;
865 return CurDAG->getTargetConstant(enc, MVT::i64);
868 def : Pat<(shl GPR32:$Rn, (i64 imm0_31:$imm)),
869 (UBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
870 (i64 (i32shift_b imm0_31:$imm)))>;
871 def : Pat<(shl GPR64:$Rn, (i64 imm0_63:$imm)),
872 (UBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
873 (i64 (i64shift_b imm0_63:$imm)))>;
875 let AddedComplexity = 10 in {
876 def : Pat<(sra GPR32:$Rn, (i64 imm0_31:$imm)),
877 (SBFMWri GPR32:$Rn, imm0_31:$imm, 31)>;
878 def : Pat<(sra GPR64:$Rn, (i64 imm0_63:$imm)),
879 (SBFMXri GPR64:$Rn, imm0_63:$imm, 63)>;
882 def : InstAlias<"asr $dst, $src, $shift",
883 (SBFMWri GPR32:$dst, GPR32:$src, imm0_31:$shift, 31)>;
884 def : InstAlias<"asr $dst, $src, $shift",
885 (SBFMXri GPR64:$dst, GPR64:$src, imm0_63:$shift, 63)>;
886 def : InstAlias<"sxtb $dst, $src", (SBFMWri GPR32:$dst, GPR32:$src, 0, 7)>;
887 def : InstAlias<"sxtb $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 7)>;
888 def : InstAlias<"sxth $dst, $src", (SBFMWri GPR32:$dst, GPR32:$src, 0, 15)>;
889 def : InstAlias<"sxth $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 15)>;
890 def : InstAlias<"sxtw $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 31)>;
892 def : Pat<(srl GPR32:$Rn, (i64 imm0_31:$imm)),
893 (UBFMWri GPR32:$Rn, imm0_31:$imm, 31)>;
894 def : Pat<(srl GPR64:$Rn, (i64 imm0_63:$imm)),
895 (UBFMXri GPR64:$Rn, imm0_63:$imm, 63)>;
897 def : InstAlias<"lsr $dst, $src, $shift",
898 (UBFMWri GPR32:$dst, GPR32:$src, imm0_31:$shift, 31)>;
899 def : InstAlias<"lsr $dst, $src, $shift",
900 (UBFMXri GPR64:$dst, GPR64:$src, imm0_63:$shift, 63)>;
901 def : InstAlias<"uxtb $dst, $src", (UBFMWri GPR32:$dst, GPR32:$src, 0, 7)>;
902 def : InstAlias<"uxtb $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 7)>;
903 def : InstAlias<"uxth $dst, $src", (UBFMWri GPR32:$dst, GPR32:$src, 0, 15)>;
904 def : InstAlias<"uxth $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 15)>;
905 def : InstAlias<"uxtw $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 31)>;
907 //===----------------------------------------------------------------------===//
908 // Conditionally set flags instructions.
909 //===----------------------------------------------------------------------===//
910 defm CCMN : CondSetFlagsImm<0, "ccmn">;
911 defm CCMP : CondSetFlagsImm<1, "ccmp">;
913 defm CCMN : CondSetFlagsReg<0, "ccmn">;
914 defm CCMP : CondSetFlagsReg<1, "ccmp">;
916 //===----------------------------------------------------------------------===//
917 // Conditional select instructions.
918 //===----------------------------------------------------------------------===//
919 defm CSEL : CondSelect<0, 0b00, "csel">;
921 def inc : PatFrag<(ops node:$in), (add node:$in, 1)>;
922 defm CSINC : CondSelectOp<0, 0b01, "csinc", inc>;
923 defm CSINV : CondSelectOp<1, 0b00, "csinv", not>;
924 defm CSNEG : CondSelectOp<1, 0b01, "csneg", ineg>;
926 def : Pat<(AArch64csinv GPR32:$tval, GPR32:$fval, (i32 imm:$cc), NZCV),
927 (CSINVWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
928 def : Pat<(AArch64csinv GPR64:$tval, GPR64:$fval, (i32 imm:$cc), NZCV),
929 (CSINVXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
930 def : Pat<(AArch64csneg GPR32:$tval, GPR32:$fval, (i32 imm:$cc), NZCV),
931 (CSNEGWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
932 def : Pat<(AArch64csneg GPR64:$tval, GPR64:$fval, (i32 imm:$cc), NZCV),
933 (CSNEGXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
934 def : Pat<(AArch64csinc GPR32:$tval, GPR32:$fval, (i32 imm:$cc), NZCV),
935 (CSINCWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
936 def : Pat<(AArch64csinc GPR64:$tval, GPR64:$fval, (i32 imm:$cc), NZCV),
937 (CSINCXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
939 def : Pat<(AArch64csel (i32 0), (i32 1), (i32 imm:$cc), NZCV),
940 (CSINCWr WZR, WZR, (i32 imm:$cc))>;
941 def : Pat<(AArch64csel (i64 0), (i64 1), (i32 imm:$cc), NZCV),
942 (CSINCXr XZR, XZR, (i32 imm:$cc))>;
943 def : Pat<(AArch64csel (i32 0), (i32 -1), (i32 imm:$cc), NZCV),
944 (CSINVWr WZR, WZR, (i32 imm:$cc))>;
945 def : Pat<(AArch64csel (i64 0), (i64 -1), (i32 imm:$cc), NZCV),
946 (CSINVXr XZR, XZR, (i32 imm:$cc))>;
948 // The inverse of the condition code from the alias instruction is what is used
949 // in the aliased instruction. The parser all ready inverts the condition code
950 // for these aliases.
951 def : InstAlias<"cset $dst, $cc",
952 (CSINCWr GPR32:$dst, WZR, WZR, inv_ccode:$cc)>;
953 def : InstAlias<"cset $dst, $cc",
954 (CSINCXr GPR64:$dst, XZR, XZR, inv_ccode:$cc)>;
956 def : InstAlias<"csetm $dst, $cc",
957 (CSINVWr GPR32:$dst, WZR, WZR, inv_ccode:$cc)>;
958 def : InstAlias<"csetm $dst, $cc",
959 (CSINVXr GPR64:$dst, XZR, XZR, inv_ccode:$cc)>;
961 def : InstAlias<"cinc $dst, $src, $cc",
962 (CSINCWr GPR32:$dst, GPR32:$src, GPR32:$src, inv_ccode:$cc)>;
963 def : InstAlias<"cinc $dst, $src, $cc",
964 (CSINCXr GPR64:$dst, GPR64:$src, GPR64:$src, inv_ccode:$cc)>;
966 def : InstAlias<"cinv $dst, $src, $cc",
967 (CSINVWr GPR32:$dst, GPR32:$src, GPR32:$src, inv_ccode:$cc)>;
968 def : InstAlias<"cinv $dst, $src, $cc",
969 (CSINVXr GPR64:$dst, GPR64:$src, GPR64:$src, inv_ccode:$cc)>;
971 def : InstAlias<"cneg $dst, $src, $cc",
972 (CSNEGWr GPR32:$dst, GPR32:$src, GPR32:$src, inv_ccode:$cc)>;
973 def : InstAlias<"cneg $dst, $src, $cc",
974 (CSNEGXr GPR64:$dst, GPR64:$src, GPR64:$src, inv_ccode:$cc)>;
976 //===----------------------------------------------------------------------===//
977 // PC-relative instructions.
978 //===----------------------------------------------------------------------===//
979 let isReMaterializable = 1 in {
980 let neverHasSideEffects = 1, mayStore = 0, mayLoad = 0 in {
981 def ADR : ADRI<0, "adr", adrlabel, []>;
982 } // neverHasSideEffects = 1
984 def ADRP : ADRI<1, "adrp", adrplabel,
985 [(set GPR64:$Xd, (AArch64adrp tglobaladdr:$label))]>;
986 } // isReMaterializable = 1
988 // page address of a constant pool entry, block address
989 def : Pat<(AArch64adrp tconstpool:$cp), (ADRP tconstpool:$cp)>;
990 def : Pat<(AArch64adrp tblockaddress:$cp), (ADRP tblockaddress:$cp)>;
992 //===----------------------------------------------------------------------===//
993 // Unconditional branch (register) instructions.
994 //===----------------------------------------------------------------------===//
996 let isReturn = 1, isTerminator = 1, isBarrier = 1 in {
997 def RET : BranchReg<0b0010, "ret", []>;
998 def DRPS : SpecialReturn<0b0101, "drps">;
999 def ERET : SpecialReturn<0b0100, "eret">;
1000 } // isReturn = 1, isTerminator = 1, isBarrier = 1
1002 // Default to the LR register.
1003 def : InstAlias<"ret", (RET LR)>;
1005 let isCall = 1, Defs = [LR], Uses = [SP] in {
1006 def BLR : BranchReg<0b0001, "blr", [(AArch64call GPR64:$Rn)]>;
1009 let isBranch = 1, isTerminator = 1, isBarrier = 1, isIndirectBranch = 1 in {
1010 def BR : BranchReg<0b0000, "br", [(brind GPR64:$Rn)]>;
1011 } // isBranch, isTerminator, isBarrier, isIndirectBranch
1013 // Create a separate pseudo-instruction for codegen to use so that we don't
1014 // flag lr as used in every function. It'll be restored before the RET by the
1015 // epilogue if it's legitimately used.
1016 def RET_ReallyLR : Pseudo<(outs), (ins), [(AArch64retflag)]> {
1017 let isTerminator = 1;
1022 // This is a directive-like pseudo-instruction. The purpose is to insert an
1023 // R_AARCH64_TLSDESC_CALL relocation at the offset of the following instruction
1024 // (which in the usual case is a BLR).
1025 let hasSideEffects = 1 in
1026 def TLSDESCCALL : Pseudo<(outs), (ins i64imm:$sym), []> {
1027 let AsmString = ".tlsdesccall $sym";
1030 // Pseudo-instruction representing a BLR with attached TLSDESC relocation. It
1031 // gets expanded to two MCInsts during lowering.
1032 let isCall = 1, Defs = [LR] in
1034 : Pseudo<(outs), (ins GPR64:$dest, i64imm:$sym),
1035 [(AArch64tlsdesc_call GPR64:$dest, tglobaltlsaddr:$sym)]>;
1037 def : Pat<(AArch64tlsdesc_call GPR64:$dest, texternalsym:$sym),
1038 (TLSDESC_BLR GPR64:$dest, texternalsym:$sym)>;
1039 //===----------------------------------------------------------------------===//
1040 // Conditional branch (immediate) instruction.
1041 //===----------------------------------------------------------------------===//
1042 def Bcc : BranchCond;
1044 //===----------------------------------------------------------------------===//
1045 // Compare-and-branch instructions.
1046 //===----------------------------------------------------------------------===//
1047 defm CBZ : CmpBranch<0, "cbz", AArch64cbz>;
1048 defm CBNZ : CmpBranch<1, "cbnz", AArch64cbnz>;
1050 //===----------------------------------------------------------------------===//
1051 // Test-bit-and-branch instructions.
1052 //===----------------------------------------------------------------------===//
1053 defm TBZ : TestBranch<0, "tbz", AArch64tbz>;
1054 defm TBNZ : TestBranch<1, "tbnz", AArch64tbnz>;
1056 //===----------------------------------------------------------------------===//
1057 // Unconditional branch (immediate) instructions.
1058 //===----------------------------------------------------------------------===//
1059 let isBranch = 1, isTerminator = 1, isBarrier = 1 in {
1060 def B : BranchImm<0, "b", [(br bb:$addr)]>;
1061 } // isBranch, isTerminator, isBarrier
1063 let isCall = 1, Defs = [LR], Uses = [SP] in {
1064 def BL : CallImm<1, "bl", [(AArch64call tglobaladdr:$addr)]>;
1066 def : Pat<(AArch64call texternalsym:$func), (BL texternalsym:$func)>;
1068 //===----------------------------------------------------------------------===//
1069 // Exception generation instructions.
1070 //===----------------------------------------------------------------------===//
1071 def BRK : ExceptionGeneration<0b001, 0b00, "brk">;
1072 def DCPS1 : ExceptionGeneration<0b101, 0b01, "dcps1">;
1073 def DCPS2 : ExceptionGeneration<0b101, 0b10, "dcps2">;
1074 def DCPS3 : ExceptionGeneration<0b101, 0b11, "dcps3">;
1075 def HLT : ExceptionGeneration<0b010, 0b00, "hlt">;
1076 def HVC : ExceptionGeneration<0b000, 0b10, "hvc">;
1077 def SMC : ExceptionGeneration<0b000, 0b11, "smc">;
1078 def SVC : ExceptionGeneration<0b000, 0b01, "svc">;
1080 // DCPSn defaults to an immediate operand of zero if unspecified.
1081 def : InstAlias<"dcps1", (DCPS1 0)>;
1082 def : InstAlias<"dcps2", (DCPS2 0)>;
1083 def : InstAlias<"dcps3", (DCPS3 0)>;
1085 //===----------------------------------------------------------------------===//
1086 // Load instructions.
1087 //===----------------------------------------------------------------------===//
1089 // Pair (indexed, offset)
1090 defm LDPW : LoadPairOffset<0b00, 0, GPR32, simm7s4, "ldp">;
1091 defm LDPX : LoadPairOffset<0b10, 0, GPR64, simm7s8, "ldp">;
1092 defm LDPS : LoadPairOffset<0b00, 1, FPR32, simm7s4, "ldp">;
1093 defm LDPD : LoadPairOffset<0b01, 1, FPR64, simm7s8, "ldp">;
1094 defm LDPQ : LoadPairOffset<0b10, 1, FPR128, simm7s16, "ldp">;
1096 defm LDPSW : LoadPairOffset<0b01, 0, GPR64, simm7s4, "ldpsw">;
1098 // Pair (pre-indexed)
1099 def LDPWpre : LoadPairPreIdx<0b00, 0, GPR32, simm7s4, "ldp">;
1100 def LDPXpre : LoadPairPreIdx<0b10, 0, GPR64, simm7s8, "ldp">;
1101 def LDPSpre : LoadPairPreIdx<0b00, 1, FPR32, simm7s4, "ldp">;
1102 def LDPDpre : LoadPairPreIdx<0b01, 1, FPR64, simm7s8, "ldp">;
1103 def LDPQpre : LoadPairPreIdx<0b10, 1, FPR128, simm7s16, "ldp">;
1105 def LDPSWpre : LoadPairPreIdx<0b01, 0, GPR64, simm7s4, "ldpsw">;
1107 // Pair (post-indexed)
1108 def LDPWpost : LoadPairPostIdx<0b00, 0, GPR32, simm7s4, "ldp">;
1109 def LDPXpost : LoadPairPostIdx<0b10, 0, GPR64, simm7s8, "ldp">;
1110 def LDPSpost : LoadPairPostIdx<0b00, 1, FPR32, simm7s4, "ldp">;
1111 def LDPDpost : LoadPairPostIdx<0b01, 1, FPR64, simm7s8, "ldp">;
1112 def LDPQpost : LoadPairPostIdx<0b10, 1, FPR128, simm7s16, "ldp">;
1114 def LDPSWpost : LoadPairPostIdx<0b01, 0, GPR64, simm7s4, "ldpsw">;
1117 // Pair (no allocate)
1118 defm LDNPW : LoadPairNoAlloc<0b00, 0, GPR32, simm7s4, "ldnp">;
1119 defm LDNPX : LoadPairNoAlloc<0b10, 0, GPR64, simm7s8, "ldnp">;
1120 defm LDNPS : LoadPairNoAlloc<0b00, 1, FPR32, simm7s4, "ldnp">;
1121 defm LDNPD : LoadPairNoAlloc<0b01, 1, FPR64, simm7s8, "ldnp">;
1122 defm LDNPQ : LoadPairNoAlloc<0b10, 1, FPR128, simm7s16, "ldnp">;
1125 // (register offset)
1129 defm LDRBB : Load8RO<0b00, 0, 0b01, GPR32, "ldrb", i32, zextloadi8>;
1130 defm LDRHH : Load16RO<0b01, 0, 0b01, GPR32, "ldrh", i32, zextloadi16>;
1131 defm LDRW : Load32RO<0b10, 0, 0b01, GPR32, "ldr", i32, load>;
1132 defm LDRX : Load64RO<0b11, 0, 0b01, GPR64, "ldr", i64, load>;
1135 defm LDRB : Load8RO<0b00, 1, 0b01, FPR8, "ldr", untyped, load>;
1136 defm LDRH : Load16RO<0b01, 1, 0b01, FPR16, "ldr", f16, load>;
1137 defm LDRS : Load32RO<0b10, 1, 0b01, FPR32, "ldr", f32, load>;
1138 defm LDRD : Load64RO<0b11, 1, 0b01, FPR64, "ldr", f64, load>;
1139 defm LDRQ : Load128RO<0b00, 1, 0b11, FPR128, "ldr", f128, load>;
1141 // Load sign-extended half-word
1142 defm LDRSHW : Load16RO<0b01, 0, 0b11, GPR32, "ldrsh", i32, sextloadi16>;
1143 defm LDRSHX : Load16RO<0b01, 0, 0b10, GPR64, "ldrsh", i64, sextloadi16>;
1145 // Load sign-extended byte
1146 defm LDRSBW : Load8RO<0b00, 0, 0b11, GPR32, "ldrsb", i32, sextloadi8>;
1147 defm LDRSBX : Load8RO<0b00, 0, 0b10, GPR64, "ldrsb", i64, sextloadi8>;
1149 // Load sign-extended word
1150 defm LDRSW : Load32RO<0b10, 0, 0b10, GPR64, "ldrsw", i64, sextloadi32>;
1153 defm PRFM : PrefetchRO<0b11, 0, 0b10, "prfm">;
1155 // For regular load, we do not have any alignment requirement.
1156 // Thus, it is safe to directly map the vector loads with interesting
1157 // addressing modes.
1158 // FIXME: We could do the same for bitconvert to floating point vectors.
1159 multiclass ScalToVecROLoadPat<ROAddrMode ro, SDPatternOperator loadop,
1160 ValueType ScalTy, ValueType VecTy,
1161 Instruction LOADW, Instruction LOADX,
1163 def : Pat<(VecTy (scalar_to_vector (ScalTy
1164 (loadop (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$offset))))),
1165 (INSERT_SUBREG (VecTy (IMPLICIT_DEF)),
1166 (LOADW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$offset),
1169 def : Pat<(VecTy (scalar_to_vector (ScalTy
1170 (loadop (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$offset))))),
1171 (INSERT_SUBREG (VecTy (IMPLICIT_DEF)),
1172 (LOADX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$offset),
1176 let AddedComplexity = 10 in {
1177 defm : ScalToVecROLoadPat<ro8, extloadi8, i32, v8i8, LDRBroW, LDRBroX, bsub>;
1178 defm : ScalToVecROLoadPat<ro8, extloadi8, i32, v16i8, LDRBroW, LDRBroX, bsub>;
1180 defm : ScalToVecROLoadPat<ro16, extloadi16, i32, v4i16, LDRHroW, LDRHroX, hsub>;
1181 defm : ScalToVecROLoadPat<ro16, extloadi16, i32, v8i16, LDRHroW, LDRHroX, hsub>;
1183 defm : ScalToVecROLoadPat<ro16, load, i32, v4f16, LDRHroW, LDRHroX, hsub>;
1184 defm : ScalToVecROLoadPat<ro16, load, i32, v8f16, LDRHroW, LDRHroX, hsub>;
1186 defm : ScalToVecROLoadPat<ro32, load, i32, v2i32, LDRSroW, LDRSroX, ssub>;
1187 defm : ScalToVecROLoadPat<ro32, load, i32, v4i32, LDRSroW, LDRSroX, ssub>;
1189 defm : ScalToVecROLoadPat<ro32, load, f32, v2f32, LDRSroW, LDRSroX, ssub>;
1190 defm : ScalToVecROLoadPat<ro32, load, f32, v4f32, LDRSroW, LDRSroX, ssub>;
1192 defm : ScalToVecROLoadPat<ro64, load, i64, v2i64, LDRDroW, LDRDroX, dsub>;
1194 defm : ScalToVecROLoadPat<ro64, load, f64, v2f64, LDRDroW, LDRDroX, dsub>;
1197 def : Pat <(v1i64 (scalar_to_vector (i64
1198 (load (ro_Windexed64 GPR64sp:$Rn, GPR32:$Rm,
1199 ro_Wextend64:$extend))))),
1200 (LDRDroW GPR64sp:$Rn, GPR32:$Rm, ro_Wextend64:$extend)>;
1202 def : Pat <(v1i64 (scalar_to_vector (i64
1203 (load (ro_Xindexed64 GPR64sp:$Rn, GPR64:$Rm,
1204 ro_Xextend64:$extend))))),
1205 (LDRDroX GPR64sp:$Rn, GPR64:$Rm, ro_Xextend64:$extend)>;
1208 // Match all load 64 bits width whose type is compatible with FPR64
1209 multiclass VecROLoadPat<ROAddrMode ro, ValueType VecTy,
1210 Instruction LOADW, Instruction LOADX> {
1212 def : Pat<(VecTy (load (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend))),
1213 (LOADW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
1215 def : Pat<(VecTy (load (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend))),
1216 (LOADX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
1219 let AddedComplexity = 10 in {
1220 let Predicates = [IsLE] in {
1221 // We must do vector loads with LD1 in big-endian.
1222 defm : VecROLoadPat<ro64, v2i32, LDRDroW, LDRDroX>;
1223 defm : VecROLoadPat<ro64, v2f32, LDRDroW, LDRDroX>;
1224 defm : VecROLoadPat<ro64, v8i8, LDRDroW, LDRDroX>;
1225 defm : VecROLoadPat<ro64, v4i16, LDRDroW, LDRDroX>;
1226 defm : VecROLoadPat<ro64, v4f16, LDRDroW, LDRDroX>;
1229 defm : VecROLoadPat<ro64, v1i64, LDRDroW, LDRDroX>;
1230 defm : VecROLoadPat<ro64, v1f64, LDRDroW, LDRDroX>;
1232 // Match all load 128 bits width whose type is compatible with FPR128
1233 let Predicates = [IsLE] in {
1234 // We must do vector loads with LD1 in big-endian.
1235 defm : VecROLoadPat<ro128, v2i64, LDRQroW, LDRQroX>;
1236 defm : VecROLoadPat<ro128, v2f64, LDRQroW, LDRQroX>;
1237 defm : VecROLoadPat<ro128, v4i32, LDRQroW, LDRQroX>;
1238 defm : VecROLoadPat<ro128, v4f32, LDRQroW, LDRQroX>;
1239 defm : VecROLoadPat<ro128, v8i16, LDRQroW, LDRQroX>;
1240 defm : VecROLoadPat<ro128, v8f16, LDRQroW, LDRQroX>;
1241 defm : VecROLoadPat<ro128, v16i8, LDRQroW, LDRQroX>;
1243 } // AddedComplexity = 10
1246 multiclass ExtLoadTo64ROPat<ROAddrMode ro, SDPatternOperator loadop,
1247 Instruction INSTW, Instruction INSTX> {
1248 def : Pat<(i64 (loadop (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend))),
1249 (SUBREG_TO_REG (i64 0),
1250 (INSTW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend),
1253 def : Pat<(i64 (loadop (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend))),
1254 (SUBREG_TO_REG (i64 0),
1255 (INSTX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend),
1259 let AddedComplexity = 10 in {
1260 defm : ExtLoadTo64ROPat<ro8, zextloadi8, LDRBBroW, LDRBBroX>;
1261 defm : ExtLoadTo64ROPat<ro16, zextloadi16, LDRHHroW, LDRHHroX>;
1262 defm : ExtLoadTo64ROPat<ro32, zextloadi32, LDRWroW, LDRWroX>;
1264 // zextloadi1 -> zextloadi8
1265 defm : ExtLoadTo64ROPat<ro8, zextloadi1, LDRBBroW, LDRBBroX>;
1267 // extload -> zextload
1268 defm : ExtLoadTo64ROPat<ro8, extloadi8, LDRBBroW, LDRBBroX>;
1269 defm : ExtLoadTo64ROPat<ro16, extloadi16, LDRHHroW, LDRHHroX>;
1270 defm : ExtLoadTo64ROPat<ro32, extloadi32, LDRWroW, LDRWroX>;
1272 // extloadi1 -> zextloadi8
1273 defm : ExtLoadTo64ROPat<ro8, extloadi1, LDRBBroW, LDRBBroX>;
1278 multiclass ExtLoadTo32ROPat<ROAddrMode ro, SDPatternOperator loadop,
1279 Instruction INSTW, Instruction INSTX> {
1280 def : Pat<(i32 (loadop (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend))),
1281 (INSTW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
1283 def : Pat<(i32 (loadop (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend))),
1284 (INSTX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
1288 let AddedComplexity = 10 in {
1289 // extload -> zextload
1290 defm : ExtLoadTo32ROPat<ro8, extloadi8, LDRBBroW, LDRBBroX>;
1291 defm : ExtLoadTo32ROPat<ro16, extloadi16, LDRHHroW, LDRHHroX>;
1292 defm : ExtLoadTo32ROPat<ro32, extloadi32, LDRWroW, LDRWroX>;
1294 // zextloadi1 -> zextloadi8
1295 defm : ExtLoadTo32ROPat<ro8, zextloadi1, LDRBBroW, LDRBBroX>;
1299 // (unsigned immediate)
1301 defm LDRX : LoadUI<0b11, 0, 0b01, GPR64, uimm12s8, "ldr",
1303 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)))]>;
1304 defm LDRW : LoadUI<0b10, 0, 0b01, GPR32, uimm12s4, "ldr",
1306 (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset)))]>;
1307 defm LDRB : LoadUI<0b00, 1, 0b01, FPR8, uimm12s1, "ldr",
1309 (load (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset)))]>;
1310 defm LDRH : LoadUI<0b01, 1, 0b01, FPR16, uimm12s2, "ldr",
1311 [(set (f16 FPR16:$Rt),
1312 (load (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset)))]>;
1313 defm LDRS : LoadUI<0b10, 1, 0b01, FPR32, uimm12s4, "ldr",
1314 [(set (f32 FPR32:$Rt),
1315 (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset)))]>;
1316 defm LDRD : LoadUI<0b11, 1, 0b01, FPR64, uimm12s8, "ldr",
1317 [(set (f64 FPR64:$Rt),
1318 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)))]>;
1319 defm LDRQ : LoadUI<0b00, 1, 0b11, FPR128, uimm12s16, "ldr",
1320 [(set (f128 FPR128:$Rt),
1321 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)))]>;
1323 // For regular load, we do not have any alignment requirement.
1324 // Thus, it is safe to directly map the vector loads with interesting
1325 // addressing modes.
1326 // FIXME: We could do the same for bitconvert to floating point vectors.
1327 def : Pat <(v8i8 (scalar_to_vector (i32
1328 (extloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))))),
1329 (INSERT_SUBREG (v8i8 (IMPLICIT_DEF)),
1330 (LDRBui GPR64sp:$Rn, uimm12s1:$offset), bsub)>;
1331 def : Pat <(v16i8 (scalar_to_vector (i32
1332 (extloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))))),
1333 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
1334 (LDRBui GPR64sp:$Rn, uimm12s1:$offset), bsub)>;
1335 def : Pat <(v4i16 (scalar_to_vector (i32
1336 (extloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))))),
1337 (INSERT_SUBREG (v4i16 (IMPLICIT_DEF)),
1338 (LDRHui GPR64sp:$Rn, uimm12s2:$offset), hsub)>;
1339 def : Pat <(v8i16 (scalar_to_vector (i32
1340 (extloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))))),
1341 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
1342 (LDRHui GPR64sp:$Rn, uimm12s2:$offset), hsub)>;
1343 def : Pat <(v2i32 (scalar_to_vector (i32
1344 (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))))),
1345 (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)),
1346 (LDRSui GPR64sp:$Rn, uimm12s4:$offset), ssub)>;
1347 def : Pat <(v4i32 (scalar_to_vector (i32
1348 (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))))),
1349 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
1350 (LDRSui GPR64sp:$Rn, uimm12s4:$offset), ssub)>;
1351 def : Pat <(v1i64 (scalar_to_vector (i64
1352 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))))),
1353 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1354 def : Pat <(v2i64 (scalar_to_vector (i64
1355 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))))),
1356 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
1357 (LDRDui GPR64sp:$Rn, uimm12s8:$offset), dsub)>;
1359 // Match all load 64 bits width whose type is compatible with FPR64
1360 let Predicates = [IsLE] in {
1361 // We must use LD1 to perform vector loads in big-endian.
1362 def : Pat<(v2f32 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1363 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1364 def : Pat<(v8i8 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1365 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1366 def : Pat<(v4i16 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1367 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1368 def : Pat<(v2i32 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1369 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1370 def : Pat<(v4f16 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1371 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1373 def : Pat<(v1f64 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1374 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1375 def : Pat<(v1i64 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1376 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1378 // Match all load 128 bits width whose type is compatible with FPR128
1379 let Predicates = [IsLE] in {
1380 // We must use LD1 to perform vector loads in big-endian.
1381 def : Pat<(v4f32 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1382 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1383 def : Pat<(v2f64 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1384 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1385 def : Pat<(v16i8 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1386 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1387 def : Pat<(v8i16 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1388 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1389 def : Pat<(v4i32 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1390 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1391 def : Pat<(v2i64 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1392 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1393 def : Pat<(v8f16 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1394 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1396 def : Pat<(f128 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1397 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1399 defm LDRHH : LoadUI<0b01, 0, 0b01, GPR32, uimm12s2, "ldrh",
1401 (zextloadi16 (am_indexed16 GPR64sp:$Rn,
1402 uimm12s2:$offset)))]>;
1403 defm LDRBB : LoadUI<0b00, 0, 0b01, GPR32, uimm12s1, "ldrb",
1405 (zextloadi8 (am_indexed8 GPR64sp:$Rn,
1406 uimm12s1:$offset)))]>;
1408 def : Pat<(i64 (zextloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1409 (SUBREG_TO_REG (i64 0), (LDRBBui GPR64sp:$Rn, uimm12s1:$offset), sub_32)>;
1410 def : Pat<(i64 (zextloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))),
1411 (SUBREG_TO_REG (i64 0), (LDRHHui GPR64sp:$Rn, uimm12s2:$offset), sub_32)>;
1413 // zextloadi1 -> zextloadi8
1414 def : Pat<(i32 (zextloadi1 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1415 (LDRBBui GPR64sp:$Rn, uimm12s1:$offset)>;
1416 def : Pat<(i64 (zextloadi1 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1417 (SUBREG_TO_REG (i64 0), (LDRBBui GPR64sp:$Rn, uimm12s1:$offset), sub_32)>;
1419 // extload -> zextload
1420 def : Pat<(i32 (extloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))),
1421 (LDRHHui GPR64sp:$Rn, uimm12s2:$offset)>;
1422 def : Pat<(i32 (extloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1423 (LDRBBui GPR64sp:$Rn, uimm12s1:$offset)>;
1424 def : Pat<(i32 (extloadi1 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1425 (LDRBBui GPR64sp:$Rn, uimm12s1:$offset)>;
1426 def : Pat<(i64 (extloadi32 (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))),
1427 (SUBREG_TO_REG (i64 0), (LDRWui GPR64sp:$Rn, uimm12s4:$offset), sub_32)>;
1428 def : Pat<(i64 (extloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))),
1429 (SUBREG_TO_REG (i64 0), (LDRHHui GPR64sp:$Rn, uimm12s2:$offset), sub_32)>;
1430 def : Pat<(i64 (extloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1431 (SUBREG_TO_REG (i64 0), (LDRBBui GPR64sp:$Rn, uimm12s1:$offset), sub_32)>;
1432 def : Pat<(i64 (extloadi1 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1433 (SUBREG_TO_REG (i64 0), (LDRBBui GPR64sp:$Rn, uimm12s1:$offset), sub_32)>;
1435 // load sign-extended half-word
1436 defm LDRSHW : LoadUI<0b01, 0, 0b11, GPR32, uimm12s2, "ldrsh",
1438 (sextloadi16 (am_indexed16 GPR64sp:$Rn,
1439 uimm12s2:$offset)))]>;
1440 defm LDRSHX : LoadUI<0b01, 0, 0b10, GPR64, uimm12s2, "ldrsh",
1442 (sextloadi16 (am_indexed16 GPR64sp:$Rn,
1443 uimm12s2:$offset)))]>;
1445 // load sign-extended byte
1446 defm LDRSBW : LoadUI<0b00, 0, 0b11, GPR32, uimm12s1, "ldrsb",
1448 (sextloadi8 (am_indexed8 GPR64sp:$Rn,
1449 uimm12s1:$offset)))]>;
1450 defm LDRSBX : LoadUI<0b00, 0, 0b10, GPR64, uimm12s1, "ldrsb",
1452 (sextloadi8 (am_indexed8 GPR64sp:$Rn,
1453 uimm12s1:$offset)))]>;
1455 // load sign-extended word
1456 defm LDRSW : LoadUI<0b10, 0, 0b10, GPR64, uimm12s4, "ldrsw",
1458 (sextloadi32 (am_indexed32 GPR64sp:$Rn,
1459 uimm12s4:$offset)))]>;
1461 // load zero-extended word
1462 def : Pat<(i64 (zextloadi32 (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))),
1463 (SUBREG_TO_REG (i64 0), (LDRWui GPR64sp:$Rn, uimm12s4:$offset), sub_32)>;
1466 def PRFMui : PrefetchUI<0b11, 0, 0b10, "prfm",
1467 [(AArch64Prefetch imm:$Rt,
1468 (am_indexed64 GPR64sp:$Rn,
1469 uimm12s8:$offset))]>;
1471 def : InstAlias<"prfm $Rt, [$Rn]", (PRFMui prfop:$Rt, GPR64sp:$Rn, 0)>;
1475 def LDRWl : LoadLiteral<0b00, 0, GPR32, "ldr">;
1476 def LDRXl : LoadLiteral<0b01, 0, GPR64, "ldr">;
1477 def LDRSl : LoadLiteral<0b00, 1, FPR32, "ldr">;
1478 def LDRDl : LoadLiteral<0b01, 1, FPR64, "ldr">;
1479 def LDRQl : LoadLiteral<0b10, 1, FPR128, "ldr">;
1481 // load sign-extended word
1482 def LDRSWl : LoadLiteral<0b10, 0, GPR64, "ldrsw">;
1485 def PRFMl : PrefetchLiteral<0b11, 0, "prfm", []>;
1486 // [(AArch64Prefetch imm:$Rt, tglobaladdr:$label)]>;
1489 // (unscaled immediate)
1490 defm LDURX : LoadUnscaled<0b11, 0, 0b01, GPR64, "ldur",
1492 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset)))]>;
1493 defm LDURW : LoadUnscaled<0b10, 0, 0b01, GPR32, "ldur",
1495 (load (am_unscaled32 GPR64sp:$Rn, simm9:$offset)))]>;
1496 defm LDURB : LoadUnscaled<0b00, 1, 0b01, FPR8, "ldur",
1498 (load (am_unscaled8 GPR64sp:$Rn, simm9:$offset)))]>;
1499 defm LDURH : LoadUnscaled<0b01, 1, 0b01, FPR16, "ldur",
1501 (load (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
1502 defm LDURS : LoadUnscaled<0b10, 1, 0b01, FPR32, "ldur",
1503 [(set (f32 FPR32:$Rt),
1504 (load (am_unscaled32 GPR64sp:$Rn, simm9:$offset)))]>;
1505 defm LDURD : LoadUnscaled<0b11, 1, 0b01, FPR64, "ldur",
1506 [(set (f64 FPR64:$Rt),
1507 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset)))]>;
1508 defm LDURQ : LoadUnscaled<0b00, 1, 0b11, FPR128, "ldur",
1509 [(set (f128 FPR128:$Rt),
1510 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset)))]>;
1513 : LoadUnscaled<0b01, 0, 0b01, GPR32, "ldurh",
1515 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
1517 : LoadUnscaled<0b00, 0, 0b01, GPR32, "ldurb",
1519 (zextloadi8 (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
1521 // Match all load 64 bits width whose type is compatible with FPR64
1522 let Predicates = [IsLE] in {
1523 def : Pat<(v2f32 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1524 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1525 def : Pat<(v2i32 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1526 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1527 def : Pat<(v4i16 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1528 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1529 def : Pat<(v8i8 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1530 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1531 def : Pat<(v4f16 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1532 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1534 def : Pat<(v1f64 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1535 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1536 def : Pat<(v1i64 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1537 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1539 // Match all load 128 bits width whose type is compatible with FPR128
1540 let Predicates = [IsLE] in {
1541 def : Pat<(v2f64 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1542 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1543 def : Pat<(v2i64 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1544 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1545 def : Pat<(v4f32 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1546 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1547 def : Pat<(v4i32 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1548 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1549 def : Pat<(v8i16 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1550 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1551 def : Pat<(v16i8 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1552 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1553 def : Pat<(v8f16 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1554 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1558 def : Pat<(i32 (extloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
1559 (LDURHHi GPR64sp:$Rn, simm9:$offset)>;
1560 def : Pat<(i32 (extloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1561 (LDURBBi GPR64sp:$Rn, simm9:$offset)>;
1562 def : Pat<(i32 (extloadi1 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1563 (LDURBBi GPR64sp:$Rn, simm9:$offset)>;
1564 def : Pat<(i64 (extloadi32 (am_unscaled32 GPR64sp:$Rn, simm9:$offset))),
1565 (SUBREG_TO_REG (i64 0), (LDURWi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1566 def : Pat<(i64 (extloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
1567 (SUBREG_TO_REG (i64 0), (LDURHHi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1568 def : Pat<(i64 (extloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1569 (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1570 def : Pat<(i64 (extloadi1 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1571 (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1573 def : Pat<(i32 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
1574 (LDURHHi GPR64sp:$Rn, simm9:$offset)>;
1575 def : Pat<(i32 (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1576 (LDURBBi GPR64sp:$Rn, simm9:$offset)>;
1577 def : Pat<(i32 (zextloadi1 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1578 (LDURBBi GPR64sp:$Rn, simm9:$offset)>;
1579 def : Pat<(i64 (zextloadi32 (am_unscaled32 GPR64sp:$Rn, simm9:$offset))),
1580 (SUBREG_TO_REG (i64 0), (LDURWi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1581 def : Pat<(i64 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
1582 (SUBREG_TO_REG (i64 0), (LDURHHi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1583 def : Pat<(i64 (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1584 (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1585 def : Pat<(i64 (zextloadi1 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1586 (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1590 // LDR mnemonics fall back to LDUR for negative or unaligned offsets.
1592 // Define new assembler match classes as we want to only match these when
1593 // the don't otherwise match the scaled addressing mode for LDR/STR. Don't
1594 // associate a DiagnosticType either, as we want the diagnostic for the
1595 // canonical form (the scaled operand) to take precedence.
1596 class SImm9OffsetOperand<int Width> : AsmOperandClass {
1597 let Name = "SImm9OffsetFB" # Width;
1598 let PredicateMethod = "isSImm9OffsetFB<" # Width # ">";
1599 let RenderMethod = "addImmOperands";
1602 def SImm9OffsetFB8Operand : SImm9OffsetOperand<8>;
1603 def SImm9OffsetFB16Operand : SImm9OffsetOperand<16>;
1604 def SImm9OffsetFB32Operand : SImm9OffsetOperand<32>;
1605 def SImm9OffsetFB64Operand : SImm9OffsetOperand<64>;
1606 def SImm9OffsetFB128Operand : SImm9OffsetOperand<128>;
1608 def simm9_offset_fb8 : Operand<i64> {
1609 let ParserMatchClass = SImm9OffsetFB8Operand;
1611 def simm9_offset_fb16 : Operand<i64> {
1612 let ParserMatchClass = SImm9OffsetFB16Operand;
1614 def simm9_offset_fb32 : Operand<i64> {
1615 let ParserMatchClass = SImm9OffsetFB32Operand;
1617 def simm9_offset_fb64 : Operand<i64> {
1618 let ParserMatchClass = SImm9OffsetFB64Operand;
1620 def simm9_offset_fb128 : Operand<i64> {
1621 let ParserMatchClass = SImm9OffsetFB128Operand;
1624 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1625 (LDURXi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb64:$offset), 0>;
1626 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1627 (LDURWi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
1628 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1629 (LDURBi FPR8:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
1630 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1631 (LDURHi FPR16:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
1632 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1633 (LDURSi FPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
1634 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1635 (LDURDi FPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb64:$offset), 0>;
1636 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1637 (LDURQi FPR128:$Rt, GPR64sp:$Rn, simm9_offset_fb128:$offset), 0>;
1640 def : Pat<(i64 (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1641 (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1642 def : Pat<(i64 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
1643 (SUBREG_TO_REG (i64 0), (LDURHHi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1645 // load sign-extended half-word
1647 : LoadUnscaled<0b01, 0, 0b11, GPR32, "ldursh",
1649 (sextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
1651 : LoadUnscaled<0b01, 0, 0b10, GPR64, "ldursh",
1653 (sextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
1655 // load sign-extended byte
1657 : LoadUnscaled<0b00, 0, 0b11, GPR32, "ldursb",
1659 (sextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset)))]>;
1661 : LoadUnscaled<0b00, 0, 0b10, GPR64, "ldursb",
1663 (sextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset)))]>;
1665 // load sign-extended word
1667 : LoadUnscaled<0b10, 0, 0b10, GPR64, "ldursw",
1669 (sextloadi32 (am_unscaled32 GPR64sp:$Rn, simm9:$offset)))]>;
1671 // zero and sign extending aliases from generic LDR* mnemonics to LDUR*.
1672 def : InstAlias<"ldrb $Rt, [$Rn, $offset]",
1673 (LDURBBi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
1674 def : InstAlias<"ldrh $Rt, [$Rn, $offset]",
1675 (LDURHHi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
1676 def : InstAlias<"ldrsb $Rt, [$Rn, $offset]",
1677 (LDURSBWi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
1678 def : InstAlias<"ldrsb $Rt, [$Rn, $offset]",
1679 (LDURSBXi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
1680 def : InstAlias<"ldrsh $Rt, [$Rn, $offset]",
1681 (LDURSHWi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
1682 def : InstAlias<"ldrsh $Rt, [$Rn, $offset]",
1683 (LDURSHXi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
1684 def : InstAlias<"ldrsw $Rt, [$Rn, $offset]",
1685 (LDURSWi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
1688 defm PRFUM : PrefetchUnscaled<0b11, 0, 0b10, "prfum",
1689 [(AArch64Prefetch imm:$Rt,
1690 (am_unscaled64 GPR64sp:$Rn, simm9:$offset))]>;
1693 // (unscaled immediate, unprivileged)
1694 defm LDTRX : LoadUnprivileged<0b11, 0, 0b01, GPR64, "ldtr">;
1695 defm LDTRW : LoadUnprivileged<0b10, 0, 0b01, GPR32, "ldtr">;
1697 defm LDTRH : LoadUnprivileged<0b01, 0, 0b01, GPR32, "ldtrh">;
1698 defm LDTRB : LoadUnprivileged<0b00, 0, 0b01, GPR32, "ldtrb">;
1700 // load sign-extended half-word
1701 defm LDTRSHW : LoadUnprivileged<0b01, 0, 0b11, GPR32, "ldtrsh">;
1702 defm LDTRSHX : LoadUnprivileged<0b01, 0, 0b10, GPR64, "ldtrsh">;
1704 // load sign-extended byte
1705 defm LDTRSBW : LoadUnprivileged<0b00, 0, 0b11, GPR32, "ldtrsb">;
1706 defm LDTRSBX : LoadUnprivileged<0b00, 0, 0b10, GPR64, "ldtrsb">;
1708 // load sign-extended word
1709 defm LDTRSW : LoadUnprivileged<0b10, 0, 0b10, GPR64, "ldtrsw">;
1712 // (immediate pre-indexed)
1713 def LDRWpre : LoadPreIdx<0b10, 0, 0b01, GPR32, "ldr">;
1714 def LDRXpre : LoadPreIdx<0b11, 0, 0b01, GPR64, "ldr">;
1715 def LDRBpre : LoadPreIdx<0b00, 1, 0b01, FPR8, "ldr">;
1716 def LDRHpre : LoadPreIdx<0b01, 1, 0b01, FPR16, "ldr">;
1717 def LDRSpre : LoadPreIdx<0b10, 1, 0b01, FPR32, "ldr">;
1718 def LDRDpre : LoadPreIdx<0b11, 1, 0b01, FPR64, "ldr">;
1719 def LDRQpre : LoadPreIdx<0b00, 1, 0b11, FPR128, "ldr">;
1721 // load sign-extended half-word
1722 def LDRSHWpre : LoadPreIdx<0b01, 0, 0b11, GPR32, "ldrsh">;
1723 def LDRSHXpre : LoadPreIdx<0b01, 0, 0b10, GPR64, "ldrsh">;
1725 // load sign-extended byte
1726 def LDRSBWpre : LoadPreIdx<0b00, 0, 0b11, GPR32, "ldrsb">;
1727 def LDRSBXpre : LoadPreIdx<0b00, 0, 0b10, GPR64, "ldrsb">;
1729 // load zero-extended byte
1730 def LDRBBpre : LoadPreIdx<0b00, 0, 0b01, GPR32, "ldrb">;
1731 def LDRHHpre : LoadPreIdx<0b01, 0, 0b01, GPR32, "ldrh">;
1733 // load sign-extended word
1734 def LDRSWpre : LoadPreIdx<0b10, 0, 0b10, GPR64, "ldrsw">;
1737 // (immediate post-indexed)
1738 def LDRWpost : LoadPostIdx<0b10, 0, 0b01, GPR32, "ldr">;
1739 def LDRXpost : LoadPostIdx<0b11, 0, 0b01, GPR64, "ldr">;
1740 def LDRBpost : LoadPostIdx<0b00, 1, 0b01, FPR8, "ldr">;
1741 def LDRHpost : LoadPostIdx<0b01, 1, 0b01, FPR16, "ldr">;
1742 def LDRSpost : LoadPostIdx<0b10, 1, 0b01, FPR32, "ldr">;
1743 def LDRDpost : LoadPostIdx<0b11, 1, 0b01, FPR64, "ldr">;
1744 def LDRQpost : LoadPostIdx<0b00, 1, 0b11, FPR128, "ldr">;
1746 // load sign-extended half-word
1747 def LDRSHWpost : LoadPostIdx<0b01, 0, 0b11, GPR32, "ldrsh">;
1748 def LDRSHXpost : LoadPostIdx<0b01, 0, 0b10, GPR64, "ldrsh">;
1750 // load sign-extended byte
1751 def LDRSBWpost : LoadPostIdx<0b00, 0, 0b11, GPR32, "ldrsb">;
1752 def LDRSBXpost : LoadPostIdx<0b00, 0, 0b10, GPR64, "ldrsb">;
1754 // load zero-extended byte
1755 def LDRBBpost : LoadPostIdx<0b00, 0, 0b01, GPR32, "ldrb">;
1756 def LDRHHpost : LoadPostIdx<0b01, 0, 0b01, GPR32, "ldrh">;
1758 // load sign-extended word
1759 def LDRSWpost : LoadPostIdx<0b10, 0, 0b10, GPR64, "ldrsw">;
1761 //===----------------------------------------------------------------------===//
1762 // Store instructions.
1763 //===----------------------------------------------------------------------===//
1765 // Pair (indexed, offset)
1766 // FIXME: Use dedicated range-checked addressing mode operand here.
1767 defm STPW : StorePairOffset<0b00, 0, GPR32, simm7s4, "stp">;
1768 defm STPX : StorePairOffset<0b10, 0, GPR64, simm7s8, "stp">;
1769 defm STPS : StorePairOffset<0b00, 1, FPR32, simm7s4, "stp">;
1770 defm STPD : StorePairOffset<0b01, 1, FPR64, simm7s8, "stp">;
1771 defm STPQ : StorePairOffset<0b10, 1, FPR128, simm7s16, "stp">;
1773 // Pair (pre-indexed)
1774 def STPWpre : StorePairPreIdx<0b00, 0, GPR32, simm7s4, "stp">;
1775 def STPXpre : StorePairPreIdx<0b10, 0, GPR64, simm7s8, "stp">;
1776 def STPSpre : StorePairPreIdx<0b00, 1, FPR32, simm7s4, "stp">;
1777 def STPDpre : StorePairPreIdx<0b01, 1, FPR64, simm7s8, "stp">;
1778 def STPQpre : StorePairPreIdx<0b10, 1, FPR128, simm7s16, "stp">;
1780 // Pair (pre-indexed)
1781 def STPWpost : StorePairPostIdx<0b00, 0, GPR32, simm7s4, "stp">;
1782 def STPXpost : StorePairPostIdx<0b10, 0, GPR64, simm7s8, "stp">;
1783 def STPSpost : StorePairPostIdx<0b00, 1, FPR32, simm7s4, "stp">;
1784 def STPDpost : StorePairPostIdx<0b01, 1, FPR64, simm7s8, "stp">;
1785 def STPQpost : StorePairPostIdx<0b10, 1, FPR128, simm7s16, "stp">;
1787 // Pair (no allocate)
1788 defm STNPW : StorePairNoAlloc<0b00, 0, GPR32, simm7s4, "stnp">;
1789 defm STNPX : StorePairNoAlloc<0b10, 0, GPR64, simm7s8, "stnp">;
1790 defm STNPS : StorePairNoAlloc<0b00, 1, FPR32, simm7s4, "stnp">;
1791 defm STNPD : StorePairNoAlloc<0b01, 1, FPR64, simm7s8, "stnp">;
1792 defm STNPQ : StorePairNoAlloc<0b10, 1, FPR128, simm7s16, "stnp">;
1795 // (Register offset)
1798 defm STRBB : Store8RO< 0b00, 0, 0b00, GPR32, "strb", i32, truncstorei8>;
1799 defm STRHH : Store16RO<0b01, 0, 0b00, GPR32, "strh", i32, truncstorei16>;
1800 defm STRW : Store32RO<0b10, 0, 0b00, GPR32, "str", i32, store>;
1801 defm STRX : Store64RO<0b11, 0, 0b00, GPR64, "str", i64, store>;
1805 defm STRB : Store8RO< 0b00, 1, 0b00, FPR8, "str", untyped, store>;
1806 defm STRH : Store16RO<0b01, 1, 0b00, FPR16, "str", f16, store>;
1807 defm STRS : Store32RO<0b10, 1, 0b00, FPR32, "str", f32, store>;
1808 defm STRD : Store64RO<0b11, 1, 0b00, FPR64, "str", f64, store>;
1809 defm STRQ : Store128RO<0b00, 1, 0b10, FPR128, "str", f128, store>;
1811 multiclass TruncStoreFrom64ROPat<ROAddrMode ro, SDPatternOperator storeop,
1812 Instruction STRW, Instruction STRX> {
1814 def : Pat<(storeop GPR64:$Rt,
1815 (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)),
1816 (STRW (EXTRACT_SUBREG GPR64:$Rt, sub_32),
1817 GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
1819 def : Pat<(storeop GPR64:$Rt,
1820 (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)),
1821 (STRX (EXTRACT_SUBREG GPR64:$Rt, sub_32),
1822 GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
1825 let AddedComplexity = 10 in {
1827 defm : TruncStoreFrom64ROPat<ro8, truncstorei8, STRBBroW, STRBBroX>;
1828 defm : TruncStoreFrom64ROPat<ro16, truncstorei16, STRHHroW, STRHHroX>;
1829 defm : TruncStoreFrom64ROPat<ro32, truncstorei32, STRWroW, STRWroX>;
1832 multiclass VecROStorePat<ROAddrMode ro, ValueType VecTy, RegisterClass FPR,
1833 Instruction STRW, Instruction STRX> {
1834 def : Pat<(store (VecTy FPR:$Rt),
1835 (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)),
1836 (STRW FPR:$Rt, GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
1838 def : Pat<(store (VecTy FPR:$Rt),
1839 (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)),
1840 (STRX FPR:$Rt, GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
1843 let AddedComplexity = 10 in {
1844 // Match all store 64 bits width whose type is compatible with FPR64
1845 let Predicates = [IsLE] in {
1846 // We must use ST1 to store vectors in big-endian.
1847 defm : VecROStorePat<ro64, v2i32, FPR64, STRDroW, STRDroX>;
1848 defm : VecROStorePat<ro64, v2f32, FPR64, STRDroW, STRDroX>;
1849 defm : VecROStorePat<ro64, v4i16, FPR64, STRDroW, STRDroX>;
1850 defm : VecROStorePat<ro64, v8i8, FPR64, STRDroW, STRDroX>;
1851 defm : VecROStorePat<ro64, v4f16, FPR64, STRDroW, STRDroX>;
1854 defm : VecROStorePat<ro64, v1i64, FPR64, STRDroW, STRDroX>;
1855 defm : VecROStorePat<ro64, v1f64, FPR64, STRDroW, STRDroX>;
1857 // Match all store 128 bits width whose type is compatible with FPR128
1858 let Predicates = [IsLE] in {
1859 // We must use ST1 to store vectors in big-endian.
1860 defm : VecROStorePat<ro128, v2i64, FPR128, STRQroW, STRQroX>;
1861 defm : VecROStorePat<ro128, v2f64, FPR128, STRQroW, STRQroX>;
1862 defm : VecROStorePat<ro128, v4i32, FPR128, STRQroW, STRQroX>;
1863 defm : VecROStorePat<ro128, v4f32, FPR128, STRQroW, STRQroX>;
1864 defm : VecROStorePat<ro128, v8i16, FPR128, STRQroW, STRQroX>;
1865 defm : VecROStorePat<ro128, v16i8, FPR128, STRQroW, STRQroX>;
1866 defm : VecROStorePat<ro128, v8f16, FPR128, STRQroW, STRQroX>;
1868 } // AddedComplexity = 10
1871 // (unsigned immediate)
1872 defm STRX : StoreUI<0b11, 0, 0b00, GPR64, uimm12s8, "str",
1874 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))]>;
1875 defm STRW : StoreUI<0b10, 0, 0b00, GPR32, uimm12s4, "str",
1877 (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))]>;
1878 defm STRB : StoreUI<0b00, 1, 0b00, FPR8, uimm12s1, "str",
1880 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))]>;
1881 defm STRH : StoreUI<0b01, 1, 0b00, FPR16, uimm12s2, "str",
1882 [(store (f16 FPR16:$Rt),
1883 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))]>;
1884 defm STRS : StoreUI<0b10, 1, 0b00, FPR32, uimm12s4, "str",
1885 [(store (f32 FPR32:$Rt),
1886 (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))]>;
1887 defm STRD : StoreUI<0b11, 1, 0b00, FPR64, uimm12s8, "str",
1888 [(store (f64 FPR64:$Rt),
1889 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))]>;
1890 defm STRQ : StoreUI<0b00, 1, 0b10, FPR128, uimm12s16, "str", []>;
1892 defm STRHH : StoreUI<0b01, 0, 0b00, GPR32, uimm12s2, "strh",
1893 [(truncstorei16 GPR32:$Rt,
1894 (am_indexed16 GPR64sp:$Rn,
1895 uimm12s2:$offset))]>;
1896 defm STRBB : StoreUI<0b00, 0, 0b00, GPR32, uimm12s1, "strb",
1897 [(truncstorei8 GPR32:$Rt,
1898 (am_indexed8 GPR64sp:$Rn,
1899 uimm12s1:$offset))]>;
1901 // Match all store 64 bits width whose type is compatible with FPR64
1902 let AddedComplexity = 10 in {
1903 let Predicates = [IsLE] in {
1904 // We must use ST1 to store vectors in big-endian.
1905 def : Pat<(store (v2f32 FPR64:$Rt),
1906 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
1907 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
1908 def : Pat<(store (v8i8 FPR64:$Rt),
1909 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
1910 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
1911 def : Pat<(store (v4i16 FPR64:$Rt),
1912 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
1913 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
1914 def : Pat<(store (v2i32 FPR64:$Rt),
1915 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
1916 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
1917 def : Pat<(store (v4f16 FPR64:$Rt),
1918 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
1919 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
1921 def : Pat<(store (v1f64 FPR64:$Rt),
1922 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
1923 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
1924 def : Pat<(store (v1i64 FPR64:$Rt),
1925 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
1926 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
1928 // Match all store 128 bits width whose type is compatible with FPR128
1929 let Predicates = [IsLE] in {
1930 // We must use ST1 to store vectors in big-endian.
1931 def : Pat<(store (v4f32 FPR128:$Rt),
1932 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
1933 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
1934 def : Pat<(store (v2f64 FPR128:$Rt),
1935 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
1936 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
1937 def : Pat<(store (v16i8 FPR128:$Rt),
1938 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
1939 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
1940 def : Pat<(store (v8i16 FPR128:$Rt),
1941 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
1942 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
1943 def : Pat<(store (v4i32 FPR128:$Rt),
1944 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
1945 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
1946 def : Pat<(store (v2i64 FPR128:$Rt),
1947 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
1948 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
1949 def : Pat<(store (v8f16 FPR128:$Rt),
1950 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
1951 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
1953 def : Pat<(store (f128 FPR128:$Rt),
1954 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
1955 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
1958 def : Pat<(truncstorei32 GPR64:$Rt,
1959 (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset)),
1960 (STRWui (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, uimm12s4:$offset)>;
1961 def : Pat<(truncstorei16 GPR64:$Rt,
1962 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset)),
1963 (STRHHui (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, uimm12s2:$offset)>;
1964 def : Pat<(truncstorei8 GPR64:$Rt, (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset)),
1965 (STRBBui (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, uimm12s1:$offset)>;
1967 } // AddedComplexity = 10
1970 // (unscaled immediate)
1971 defm STURX : StoreUnscaled<0b11, 0, 0b00, GPR64, "stur",
1973 (am_unscaled64 GPR64sp:$Rn, simm9:$offset))]>;
1974 defm STURW : StoreUnscaled<0b10, 0, 0b00, GPR32, "stur",
1976 (am_unscaled32 GPR64sp:$Rn, simm9:$offset))]>;
1977 defm STURB : StoreUnscaled<0b00, 1, 0b00, FPR8, "stur",
1979 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))]>;
1980 defm STURH : StoreUnscaled<0b01, 1, 0b00, FPR16, "stur",
1981 [(store (f16 FPR16:$Rt),
1982 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))]>;
1983 defm STURS : StoreUnscaled<0b10, 1, 0b00, FPR32, "stur",
1984 [(store (f32 FPR32:$Rt),
1985 (am_unscaled32 GPR64sp:$Rn, simm9:$offset))]>;
1986 defm STURD : StoreUnscaled<0b11, 1, 0b00, FPR64, "stur",
1987 [(store (f64 FPR64:$Rt),
1988 (am_unscaled64 GPR64sp:$Rn, simm9:$offset))]>;
1989 defm STURQ : StoreUnscaled<0b00, 1, 0b10, FPR128, "stur",
1990 [(store (f128 FPR128:$Rt),
1991 (am_unscaled128 GPR64sp:$Rn, simm9:$offset))]>;
1992 defm STURHH : StoreUnscaled<0b01, 0, 0b00, GPR32, "sturh",
1993 [(truncstorei16 GPR32:$Rt,
1994 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))]>;
1995 defm STURBB : StoreUnscaled<0b00, 0, 0b00, GPR32, "sturb",
1996 [(truncstorei8 GPR32:$Rt,
1997 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))]>;
1999 // Match all store 64 bits width whose type is compatible with FPR64
2000 let Predicates = [IsLE] in {
2001 // We must use ST1 to store vectors in big-endian.
2002 def : Pat<(store (v2f32 FPR64:$Rt),
2003 (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2004 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2005 def : Pat<(store (v8i8 FPR64:$Rt),
2006 (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2007 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2008 def : Pat<(store (v4i16 FPR64:$Rt),
2009 (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2010 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2011 def : Pat<(store (v2i32 FPR64:$Rt),
2012 (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2013 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2014 def : Pat<(store (v4f16 FPR64:$Rt),
2015 (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2016 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2018 def : Pat<(store (v1f64 FPR64:$Rt), (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2019 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2020 def : Pat<(store (v1i64 FPR64:$Rt), (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2021 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2023 // Match all store 128 bits width whose type is compatible with FPR128
2024 let Predicates = [IsLE] in {
2025 // We must use ST1 to store vectors in big-endian.
2026 def : Pat<(store (v4f32 FPR128:$Rt),
2027 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2028 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2029 def : Pat<(store (v2f64 FPR128:$Rt),
2030 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2031 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2032 def : Pat<(store (v16i8 FPR128:$Rt),
2033 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2034 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2035 def : Pat<(store (v8i16 FPR128:$Rt),
2036 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2037 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2038 def : Pat<(store (v4i32 FPR128:$Rt),
2039 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2040 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2041 def : Pat<(store (v2i64 FPR128:$Rt),
2042 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2043 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2044 def : Pat<(store (v2f64 FPR128:$Rt),
2045 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2046 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2047 def : Pat<(store (v8f16 FPR128:$Rt),
2048 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2049 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2052 // unscaled i64 truncating stores
2053 def : Pat<(truncstorei32 GPR64:$Rt, (am_unscaled32 GPR64sp:$Rn, simm9:$offset)),
2054 (STURWi (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, simm9:$offset)>;
2055 def : Pat<(truncstorei16 GPR64:$Rt, (am_unscaled16 GPR64sp:$Rn, simm9:$offset)),
2056 (STURHHi (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, simm9:$offset)>;
2057 def : Pat<(truncstorei8 GPR64:$Rt, (am_unscaled8 GPR64sp:$Rn, simm9:$offset)),
2058 (STURBBi (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, simm9:$offset)>;
2061 // STR mnemonics fall back to STUR for negative or unaligned offsets.
2062 def : InstAlias<"str $Rt, [$Rn, $offset]",
2063 (STURXi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb64:$offset), 0>;
2064 def : InstAlias<"str $Rt, [$Rn, $offset]",
2065 (STURWi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
2066 def : InstAlias<"str $Rt, [$Rn, $offset]",
2067 (STURBi FPR8:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
2068 def : InstAlias<"str $Rt, [$Rn, $offset]",
2069 (STURHi FPR16:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
2070 def : InstAlias<"str $Rt, [$Rn, $offset]",
2071 (STURSi FPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
2072 def : InstAlias<"str $Rt, [$Rn, $offset]",
2073 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb64:$offset), 0>;
2074 def : InstAlias<"str $Rt, [$Rn, $offset]",
2075 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9_offset_fb128:$offset), 0>;
2077 def : InstAlias<"strb $Rt, [$Rn, $offset]",
2078 (STURBBi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
2079 def : InstAlias<"strh $Rt, [$Rn, $offset]",
2080 (STURHHi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
2083 // (unscaled immediate, unprivileged)
2084 defm STTRW : StoreUnprivileged<0b10, 0, 0b00, GPR32, "sttr">;
2085 defm STTRX : StoreUnprivileged<0b11, 0, 0b00, GPR64, "sttr">;
2087 defm STTRH : StoreUnprivileged<0b01, 0, 0b00, GPR32, "sttrh">;
2088 defm STTRB : StoreUnprivileged<0b00, 0, 0b00, GPR32, "sttrb">;
2091 // (immediate pre-indexed)
2092 def STRWpre : StorePreIdx<0b10, 0, 0b00, GPR32, "str", pre_store, i32>;
2093 def STRXpre : StorePreIdx<0b11, 0, 0b00, GPR64, "str", pre_store, i64>;
2094 def STRBpre : StorePreIdx<0b00, 1, 0b00, FPR8, "str", pre_store, untyped>;
2095 def STRHpre : StorePreIdx<0b01, 1, 0b00, FPR16, "str", pre_store, f16>;
2096 def STRSpre : StorePreIdx<0b10, 1, 0b00, FPR32, "str", pre_store, f32>;
2097 def STRDpre : StorePreIdx<0b11, 1, 0b00, FPR64, "str", pre_store, f64>;
2098 def STRQpre : StorePreIdx<0b00, 1, 0b10, FPR128, "str", pre_store, f128>;
2100 def STRBBpre : StorePreIdx<0b00, 0, 0b00, GPR32, "strb", pre_truncsti8, i32>;
2101 def STRHHpre : StorePreIdx<0b01, 0, 0b00, GPR32, "strh", pre_truncsti16, i32>;
2104 def : Pat<(pre_truncsti32 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
2105 (STRWpre (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
2107 def : Pat<(pre_truncsti16 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
2108 (STRHHpre (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
2110 def : Pat<(pre_truncsti8 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
2111 (STRBBpre (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
2114 def : Pat<(pre_store (v8i8 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2115 (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2116 def : Pat<(pre_store (v4i16 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2117 (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2118 def : Pat<(pre_store (v2i32 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2119 (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2120 def : Pat<(pre_store (v2f32 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2121 (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2122 def : Pat<(pre_store (v1i64 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2123 (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2124 def : Pat<(pre_store (v1f64 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2125 (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2126 def : Pat<(pre_store (v4f16 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2127 (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2129 def : Pat<(pre_store (v16i8 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2130 (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2131 def : Pat<(pre_store (v8i16 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2132 (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2133 def : Pat<(pre_store (v4i32 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2134 (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2135 def : Pat<(pre_store (v4f32 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2136 (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2137 def : Pat<(pre_store (v2i64 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2138 (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2139 def : Pat<(pre_store (v2f64 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2140 (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2141 def : Pat<(pre_store (v8f16 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2142 (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2145 // (immediate post-indexed)
2146 def STRWpost : StorePostIdx<0b10, 0, 0b00, GPR32, "str", post_store, i32>;
2147 def STRXpost : StorePostIdx<0b11, 0, 0b00, GPR64, "str", post_store, i64>;
2148 def STRBpost : StorePostIdx<0b00, 1, 0b00, FPR8, "str", post_store, untyped>;
2149 def STRHpost : StorePostIdx<0b01, 1, 0b00, FPR16, "str", post_store, f16>;
2150 def STRSpost : StorePostIdx<0b10, 1, 0b00, FPR32, "str", post_store, f32>;
2151 def STRDpost : StorePostIdx<0b11, 1, 0b00, FPR64, "str", post_store, f64>;
2152 def STRQpost : StorePostIdx<0b00, 1, 0b10, FPR128, "str", post_store, f128>;
2154 def STRBBpost : StorePostIdx<0b00, 0, 0b00, GPR32, "strb", post_truncsti8, i32>;
2155 def STRHHpost : StorePostIdx<0b01, 0, 0b00, GPR32, "strh", post_truncsti16, i32>;
2158 def : Pat<(post_truncsti32 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
2159 (STRWpost (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
2161 def : Pat<(post_truncsti16 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
2162 (STRHHpost (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
2164 def : Pat<(post_truncsti8 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
2165 (STRBBpost (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
2168 def : Pat<(post_store (v8i8 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2169 (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2170 def : Pat<(post_store (v4i16 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2171 (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2172 def : Pat<(post_store (v2i32 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2173 (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2174 def : Pat<(post_store (v2f32 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2175 (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2176 def : Pat<(post_store (v1i64 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2177 (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2178 def : Pat<(post_store (v1f64 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2179 (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2180 def : Pat<(post_store (v4f16 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2181 (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2183 def : Pat<(post_store (v16i8 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2184 (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2185 def : Pat<(post_store (v8i16 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2186 (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2187 def : Pat<(post_store (v4i32 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2188 (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2189 def : Pat<(post_store (v4f32 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2190 (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2191 def : Pat<(post_store (v2i64 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2192 (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2193 def : Pat<(post_store (v2f64 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2194 (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2195 def : Pat<(post_store (v8f16 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2196 (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2198 //===----------------------------------------------------------------------===//
2199 // Load/store exclusive instructions.
2200 //===----------------------------------------------------------------------===//
2202 def LDARW : LoadAcquire <0b10, 1, 1, 0, 1, GPR32, "ldar">;
2203 def LDARX : LoadAcquire <0b11, 1, 1, 0, 1, GPR64, "ldar">;
2204 def LDARB : LoadAcquire <0b00, 1, 1, 0, 1, GPR32, "ldarb">;
2205 def LDARH : LoadAcquire <0b01, 1, 1, 0, 1, GPR32, "ldarh">;
2207 def LDAXRW : LoadExclusive <0b10, 0, 1, 0, 1, GPR32, "ldaxr">;
2208 def LDAXRX : LoadExclusive <0b11, 0, 1, 0, 1, GPR64, "ldaxr">;
2209 def LDAXRB : LoadExclusive <0b00, 0, 1, 0, 1, GPR32, "ldaxrb">;
2210 def LDAXRH : LoadExclusive <0b01, 0, 1, 0, 1, GPR32, "ldaxrh">;
2212 def LDXRW : LoadExclusive <0b10, 0, 1, 0, 0, GPR32, "ldxr">;
2213 def LDXRX : LoadExclusive <0b11, 0, 1, 0, 0, GPR64, "ldxr">;
2214 def LDXRB : LoadExclusive <0b00, 0, 1, 0, 0, GPR32, "ldxrb">;
2215 def LDXRH : LoadExclusive <0b01, 0, 1, 0, 0, GPR32, "ldxrh">;
2217 def STLRW : StoreRelease <0b10, 1, 0, 0, 1, GPR32, "stlr">;
2218 def STLRX : StoreRelease <0b11, 1, 0, 0, 1, GPR64, "stlr">;
2219 def STLRB : StoreRelease <0b00, 1, 0, 0, 1, GPR32, "stlrb">;
2220 def STLRH : StoreRelease <0b01, 1, 0, 0, 1, GPR32, "stlrh">;
2222 def STLXRW : StoreExclusive<0b10, 0, 0, 0, 1, GPR32, "stlxr">;
2223 def STLXRX : StoreExclusive<0b11, 0, 0, 0, 1, GPR64, "stlxr">;
2224 def STLXRB : StoreExclusive<0b00, 0, 0, 0, 1, GPR32, "stlxrb">;
2225 def STLXRH : StoreExclusive<0b01, 0, 0, 0, 1, GPR32, "stlxrh">;
2227 def STXRW : StoreExclusive<0b10, 0, 0, 0, 0, GPR32, "stxr">;
2228 def STXRX : StoreExclusive<0b11, 0, 0, 0, 0, GPR64, "stxr">;
2229 def STXRB : StoreExclusive<0b00, 0, 0, 0, 0, GPR32, "stxrb">;
2230 def STXRH : StoreExclusive<0b01, 0, 0, 0, 0, GPR32, "stxrh">;
2232 def LDAXPW : LoadExclusivePair<0b10, 0, 1, 1, 1, GPR32, "ldaxp">;
2233 def LDAXPX : LoadExclusivePair<0b11, 0, 1, 1, 1, GPR64, "ldaxp">;
2235 def LDXPW : LoadExclusivePair<0b10, 0, 1, 1, 0, GPR32, "ldxp">;
2236 def LDXPX : LoadExclusivePair<0b11, 0, 1, 1, 0, GPR64, "ldxp">;
2238 def STLXPW : StoreExclusivePair<0b10, 0, 0, 1, 1, GPR32, "stlxp">;
2239 def STLXPX : StoreExclusivePair<0b11, 0, 0, 1, 1, GPR64, "stlxp">;
2241 def STXPW : StoreExclusivePair<0b10, 0, 0, 1, 0, GPR32, "stxp">;
2242 def STXPX : StoreExclusivePair<0b11, 0, 0, 1, 0, GPR64, "stxp">;
2244 //===----------------------------------------------------------------------===//
2245 // Scaled floating point to integer conversion instructions.
2246 //===----------------------------------------------------------------------===//
2248 defm FCVTAS : FPToIntegerUnscaled<0b00, 0b100, "fcvtas", int_aarch64_neon_fcvtas>;
2249 defm FCVTAU : FPToIntegerUnscaled<0b00, 0b101, "fcvtau", int_aarch64_neon_fcvtau>;
2250 defm FCVTMS : FPToIntegerUnscaled<0b10, 0b000, "fcvtms", int_aarch64_neon_fcvtms>;
2251 defm FCVTMU : FPToIntegerUnscaled<0b10, 0b001, "fcvtmu", int_aarch64_neon_fcvtmu>;
2252 defm FCVTNS : FPToIntegerUnscaled<0b00, 0b000, "fcvtns", int_aarch64_neon_fcvtns>;
2253 defm FCVTNU : FPToIntegerUnscaled<0b00, 0b001, "fcvtnu", int_aarch64_neon_fcvtnu>;
2254 defm FCVTPS : FPToIntegerUnscaled<0b01, 0b000, "fcvtps", int_aarch64_neon_fcvtps>;
2255 defm FCVTPU : FPToIntegerUnscaled<0b01, 0b001, "fcvtpu", int_aarch64_neon_fcvtpu>;
2256 defm FCVTZS : FPToIntegerUnscaled<0b11, 0b000, "fcvtzs", fp_to_sint>;
2257 defm FCVTZU : FPToIntegerUnscaled<0b11, 0b001, "fcvtzu", fp_to_uint>;
2258 defm FCVTZS : FPToIntegerScaled<0b11, 0b000, "fcvtzs", fp_to_sint>;
2259 defm FCVTZU : FPToIntegerScaled<0b11, 0b001, "fcvtzu", fp_to_uint>;
2260 let isCodeGenOnly = 1 in {
2261 defm FCVTZS_Int : FPToIntegerUnscaled<0b11, 0b000, "fcvtzs", int_aarch64_neon_fcvtzs>;
2262 defm FCVTZU_Int : FPToIntegerUnscaled<0b11, 0b001, "fcvtzu", int_aarch64_neon_fcvtzu>;
2263 defm FCVTZS_Int : FPToIntegerScaled<0b11, 0b000, "fcvtzs", int_aarch64_neon_fcvtzs>;
2264 defm FCVTZU_Int : FPToIntegerScaled<0b11, 0b001, "fcvtzu", int_aarch64_neon_fcvtzu>;
2267 //===----------------------------------------------------------------------===//
2268 // Scaled integer to floating point conversion instructions.
2269 //===----------------------------------------------------------------------===//
2271 defm SCVTF : IntegerToFP<0, "scvtf", sint_to_fp>;
2272 defm UCVTF : IntegerToFP<1, "ucvtf", uint_to_fp>;
2274 //===----------------------------------------------------------------------===//
2275 // Unscaled integer to floating point conversion instruction.
2276 //===----------------------------------------------------------------------===//
2278 defm FMOV : UnscaledConversion<"fmov">;
2280 def : Pat<(f32 (fpimm0)), (FMOVWSr WZR)>, Requires<[NoZCZ]>;
2281 def : Pat<(f64 (fpimm0)), (FMOVXDr XZR)>, Requires<[NoZCZ]>;
2283 //===----------------------------------------------------------------------===//
2284 // Floating point conversion instruction.
2285 //===----------------------------------------------------------------------===//
2287 defm FCVT : FPConversion<"fcvt">;
2289 //===----------------------------------------------------------------------===//
2290 // Floating point single operand instructions.
2291 //===----------------------------------------------------------------------===//
2293 defm FABS : SingleOperandFPData<0b0001, "fabs", fabs>;
2294 defm FMOV : SingleOperandFPData<0b0000, "fmov">;
2295 defm FNEG : SingleOperandFPData<0b0010, "fneg", fneg>;
2296 defm FRINTA : SingleOperandFPData<0b1100, "frinta", frnd>;
2297 defm FRINTI : SingleOperandFPData<0b1111, "frinti", fnearbyint>;
2298 defm FRINTM : SingleOperandFPData<0b1010, "frintm", ffloor>;
2299 defm FRINTN : SingleOperandFPData<0b1000, "frintn", int_aarch64_neon_frintn>;
2300 defm FRINTP : SingleOperandFPData<0b1001, "frintp", fceil>;
2302 def : Pat<(v1f64 (int_aarch64_neon_frintn (v1f64 FPR64:$Rn))),
2303 (FRINTNDr FPR64:$Rn)>;
2305 // FRINTX is inserted to set the flags as required by FENV_ACCESS ON behavior
2306 // in the C spec. Setting hasSideEffects ensures it is not DCE'd.
2307 // <rdar://problem/13715968>
2308 // TODO: We should really model the FPSR flags correctly. This is really ugly.
2309 let hasSideEffects = 1 in {
2310 defm FRINTX : SingleOperandFPData<0b1110, "frintx", frint>;
2313 defm FRINTZ : SingleOperandFPData<0b1011, "frintz", ftrunc>;
2315 let SchedRW = [WriteFDiv] in {
2316 defm FSQRT : SingleOperandFPData<0b0011, "fsqrt", fsqrt>;
2319 //===----------------------------------------------------------------------===//
2320 // Floating point two operand instructions.
2321 //===----------------------------------------------------------------------===//
2323 defm FADD : TwoOperandFPData<0b0010, "fadd", fadd>;
2324 let SchedRW = [WriteFDiv] in {
2325 defm FDIV : TwoOperandFPData<0b0001, "fdiv", fdiv>;
2327 defm FMAXNM : TwoOperandFPData<0b0110, "fmaxnm", int_aarch64_neon_fmaxnm>;
2328 defm FMAX : TwoOperandFPData<0b0100, "fmax", AArch64fmax>;
2329 defm FMINNM : TwoOperandFPData<0b0111, "fminnm", int_aarch64_neon_fminnm>;
2330 defm FMIN : TwoOperandFPData<0b0101, "fmin", AArch64fmin>;
2331 let SchedRW = [WriteFMul] in {
2332 defm FMUL : TwoOperandFPData<0b0000, "fmul", fmul>;
2333 defm FNMUL : TwoOperandFPDataNeg<0b1000, "fnmul", fmul>;
2335 defm FSUB : TwoOperandFPData<0b0011, "fsub", fsub>;
2337 def : Pat<(v1f64 (AArch64fmax (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2338 (FMAXDrr FPR64:$Rn, FPR64:$Rm)>;
2339 def : Pat<(v1f64 (AArch64fmin (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2340 (FMINDrr FPR64:$Rn, FPR64:$Rm)>;
2341 def : Pat<(v1f64 (int_aarch64_neon_fmaxnm (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2342 (FMAXNMDrr FPR64:$Rn, FPR64:$Rm)>;
2343 def : Pat<(v1f64 (int_aarch64_neon_fminnm (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2344 (FMINNMDrr FPR64:$Rn, FPR64:$Rm)>;
2346 //===----------------------------------------------------------------------===//
2347 // Floating point three operand instructions.
2348 //===----------------------------------------------------------------------===//
2350 defm FMADD : ThreeOperandFPData<0, 0, "fmadd", fma>;
2351 defm FMSUB : ThreeOperandFPData<0, 1, "fmsub",
2352 TriOpFrag<(fma node:$LHS, (fneg node:$MHS), node:$RHS)> >;
2353 defm FNMADD : ThreeOperandFPData<1, 0, "fnmadd",
2354 TriOpFrag<(fneg (fma node:$LHS, node:$MHS, node:$RHS))> >;
2355 defm FNMSUB : ThreeOperandFPData<1, 1, "fnmsub",
2356 TriOpFrag<(fma node:$LHS, node:$MHS, (fneg node:$RHS))> >;
2358 // The following def pats catch the case where the LHS of an FMA is negated.
2359 // The TriOpFrag above catches the case where the middle operand is negated.
2361 // N.b. FMSUB etc have the accumulator at the *end* of (outs), unlike
2362 // the NEON variant.
2363 def : Pat<(f32 (fma (fneg FPR32:$Rn), FPR32:$Rm, FPR32:$Ra)),
2364 (FMSUBSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
2366 def : Pat<(f64 (fma (fneg FPR64:$Rn), FPR64:$Rm, FPR64:$Ra)),
2367 (FMSUBDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
2369 // We handled -(a + b*c) for FNMADD above, now it's time for "(-a) + (-b)*c" and
2371 def : Pat<(f32 (fma (fneg FPR32:$Rn), FPR32:$Rm, (fneg FPR32:$Ra))),
2372 (FNMADDSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
2374 def : Pat<(f64 (fma (fneg FPR64:$Rn), FPR64:$Rm, (fneg FPR64:$Ra))),
2375 (FNMADDDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
2377 def : Pat<(f32 (fma FPR32:$Rn, (fneg FPR32:$Rm), (fneg FPR32:$Ra))),
2378 (FNMADDSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
2380 def : Pat<(f64 (fma FPR64:$Rn, (fneg FPR64:$Rm), (fneg FPR64:$Ra))),
2381 (FNMADDDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
2383 //===----------------------------------------------------------------------===//
2384 // Floating point comparison instructions.
2385 //===----------------------------------------------------------------------===//
2387 defm FCMPE : FPComparison<1, "fcmpe">;
2388 defm FCMP : FPComparison<0, "fcmp", AArch64fcmp>;
2390 //===----------------------------------------------------------------------===//
2391 // Floating point conditional comparison instructions.
2392 //===----------------------------------------------------------------------===//
2394 defm FCCMPE : FPCondComparison<1, "fccmpe">;
2395 defm FCCMP : FPCondComparison<0, "fccmp">;
2397 //===----------------------------------------------------------------------===//
2398 // Floating point conditional select instruction.
2399 //===----------------------------------------------------------------------===//
2401 defm FCSEL : FPCondSelect<"fcsel">;
2403 // CSEL instructions providing f128 types need to be handled by a
2404 // pseudo-instruction since the eventual code will need to introduce basic
2405 // blocks and control flow.
2406 def F128CSEL : Pseudo<(outs FPR128:$Rd),
2407 (ins FPR128:$Rn, FPR128:$Rm, ccode:$cond),
2408 [(set (f128 FPR128:$Rd),
2409 (AArch64csel FPR128:$Rn, FPR128:$Rm,
2410 (i32 imm:$cond), NZCV))]> {
2412 let usesCustomInserter = 1;
2416 //===----------------------------------------------------------------------===//
2417 // Floating point immediate move.
2418 //===----------------------------------------------------------------------===//
2420 let isReMaterializable = 1 in {
2421 defm FMOV : FPMoveImmediate<"fmov">;
2424 //===----------------------------------------------------------------------===//
2425 // Advanced SIMD two vector instructions.
2426 //===----------------------------------------------------------------------===//
2428 defm ABS : SIMDTwoVectorBHSD<0, 0b01011, "abs", int_aarch64_neon_abs>;
2429 defm CLS : SIMDTwoVectorBHS<0, 0b00100, "cls", int_aarch64_neon_cls>;
2430 defm CLZ : SIMDTwoVectorBHS<1, 0b00100, "clz", ctlz>;
2431 defm CMEQ : SIMDCmpTwoVector<0, 0b01001, "cmeq", AArch64cmeqz>;
2432 defm CMGE : SIMDCmpTwoVector<1, 0b01000, "cmge", AArch64cmgez>;
2433 defm CMGT : SIMDCmpTwoVector<0, 0b01000, "cmgt", AArch64cmgtz>;
2434 defm CMLE : SIMDCmpTwoVector<1, 0b01001, "cmle", AArch64cmlez>;
2435 defm CMLT : SIMDCmpTwoVector<0, 0b01010, "cmlt", AArch64cmltz>;
2436 defm CNT : SIMDTwoVectorB<0, 0b00, 0b00101, "cnt", ctpop>;
2437 defm FABS : SIMDTwoVectorFP<0, 1, 0b01111, "fabs", fabs>;
2439 defm FCMEQ : SIMDFPCmpTwoVector<0, 1, 0b01101, "fcmeq", AArch64fcmeqz>;
2440 defm FCMGE : SIMDFPCmpTwoVector<1, 1, 0b01100, "fcmge", AArch64fcmgez>;
2441 defm FCMGT : SIMDFPCmpTwoVector<0, 1, 0b01100, "fcmgt", AArch64fcmgtz>;
2442 defm FCMLE : SIMDFPCmpTwoVector<1, 1, 0b01101, "fcmle", AArch64fcmlez>;
2443 defm FCMLT : SIMDFPCmpTwoVector<0, 1, 0b01110, "fcmlt", AArch64fcmltz>;
2444 defm FCVTAS : SIMDTwoVectorFPToInt<0,0,0b11100, "fcvtas",int_aarch64_neon_fcvtas>;
2445 defm FCVTAU : SIMDTwoVectorFPToInt<1,0,0b11100, "fcvtau",int_aarch64_neon_fcvtau>;
2446 defm FCVTL : SIMDFPWidenTwoVector<0, 0, 0b10111, "fcvtl">;
2447 def : Pat<(v4f32 (int_aarch64_neon_vcvthf2fp (v4i16 V64:$Rn))),
2448 (FCVTLv4i16 V64:$Rn)>;
2449 def : Pat<(v4f32 (int_aarch64_neon_vcvthf2fp (extract_subvector (v8i16 V128:$Rn),
2451 (FCVTLv8i16 V128:$Rn)>;
2452 def : Pat<(v2f64 (fextend (v2f32 V64:$Rn))), (FCVTLv2i32 V64:$Rn)>;
2453 def : Pat<(v2f64 (fextend (v2f32 (extract_subvector (v4f32 V128:$Rn),
2455 (FCVTLv4i32 V128:$Rn)>;
2457 def : Pat<(v4f32 (fextend (v4f16 V64:$Rn))), (FCVTLv4i16 V64:$Rn)>;
2458 def : Pat<(v4f32 (fextend (v4f16 (extract_subvector (v8f16 V128:$Rn),
2460 (FCVTLv8i16 V128:$Rn)>;
2462 defm FCVTMS : SIMDTwoVectorFPToInt<0,0,0b11011, "fcvtms",int_aarch64_neon_fcvtms>;
2463 defm FCVTMU : SIMDTwoVectorFPToInt<1,0,0b11011, "fcvtmu",int_aarch64_neon_fcvtmu>;
2464 defm FCVTNS : SIMDTwoVectorFPToInt<0,0,0b11010, "fcvtns",int_aarch64_neon_fcvtns>;
2465 defm FCVTNU : SIMDTwoVectorFPToInt<1,0,0b11010, "fcvtnu",int_aarch64_neon_fcvtnu>;
2466 defm FCVTN : SIMDFPNarrowTwoVector<0, 0, 0b10110, "fcvtn">;
2467 def : Pat<(v4i16 (int_aarch64_neon_vcvtfp2hf (v4f32 V128:$Rn))),
2468 (FCVTNv4i16 V128:$Rn)>;
2469 def : Pat<(concat_vectors V64:$Rd,
2470 (v4i16 (int_aarch64_neon_vcvtfp2hf (v4f32 V128:$Rn)))),
2471 (FCVTNv8i16 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
2472 def : Pat<(v2f32 (fround (v2f64 V128:$Rn))), (FCVTNv2i32 V128:$Rn)>;
2473 def : Pat<(v4f16 (fround (v4f32 V128:$Rn))), (FCVTNv4i16 V128:$Rn)>;
2474 def : Pat<(concat_vectors V64:$Rd, (v2f32 (fround (v2f64 V128:$Rn)))),
2475 (FCVTNv4i32 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
2476 defm FCVTPS : SIMDTwoVectorFPToInt<0,1,0b11010, "fcvtps",int_aarch64_neon_fcvtps>;
2477 defm FCVTPU : SIMDTwoVectorFPToInt<1,1,0b11010, "fcvtpu",int_aarch64_neon_fcvtpu>;
2478 defm FCVTXN : SIMDFPInexactCvtTwoVector<1, 0, 0b10110, "fcvtxn",
2479 int_aarch64_neon_fcvtxn>;
2480 defm FCVTZS : SIMDTwoVectorFPToInt<0, 1, 0b11011, "fcvtzs", fp_to_sint>;
2481 defm FCVTZU : SIMDTwoVectorFPToInt<1, 1, 0b11011, "fcvtzu", fp_to_uint>;
2482 let isCodeGenOnly = 1 in {
2483 defm FCVTZS_Int : SIMDTwoVectorFPToInt<0, 1, 0b11011, "fcvtzs",
2484 int_aarch64_neon_fcvtzs>;
2485 defm FCVTZU_Int : SIMDTwoVectorFPToInt<1, 1, 0b11011, "fcvtzu",
2486 int_aarch64_neon_fcvtzu>;
2488 defm FNEG : SIMDTwoVectorFP<1, 1, 0b01111, "fneg", fneg>;
2489 defm FRECPE : SIMDTwoVectorFP<0, 1, 0b11101, "frecpe", int_aarch64_neon_frecpe>;
2490 defm FRINTA : SIMDTwoVectorFP<1, 0, 0b11000, "frinta", frnd>;
2491 defm FRINTI : SIMDTwoVectorFP<1, 1, 0b11001, "frinti", fnearbyint>;
2492 defm FRINTM : SIMDTwoVectorFP<0, 0, 0b11001, "frintm", ffloor>;
2493 defm FRINTN : SIMDTwoVectorFP<0, 0, 0b11000, "frintn", int_aarch64_neon_frintn>;
2494 defm FRINTP : SIMDTwoVectorFP<0, 1, 0b11000, "frintp", fceil>;
2495 defm FRINTX : SIMDTwoVectorFP<1, 0, 0b11001, "frintx", frint>;
2496 defm FRINTZ : SIMDTwoVectorFP<0, 1, 0b11001, "frintz", ftrunc>;
2497 defm FRSQRTE: SIMDTwoVectorFP<1, 1, 0b11101, "frsqrte", int_aarch64_neon_frsqrte>;
2498 defm FSQRT : SIMDTwoVectorFP<1, 1, 0b11111, "fsqrt", fsqrt>;
2499 defm NEG : SIMDTwoVectorBHSD<1, 0b01011, "neg",
2500 UnOpFrag<(sub immAllZerosV, node:$LHS)> >;
2501 defm NOT : SIMDTwoVectorB<1, 0b00, 0b00101, "not", vnot>;
2502 // Aliases for MVN -> NOT.
2503 def : InstAlias<"mvn{ $Vd.8b, $Vn.8b|.8b $Vd, $Vn}",
2504 (NOTv8i8 V64:$Vd, V64:$Vn)>;
2505 def : InstAlias<"mvn{ $Vd.16b, $Vn.16b|.16b $Vd, $Vn}",
2506 (NOTv16i8 V128:$Vd, V128:$Vn)>;
2508 def : Pat<(AArch64neg (v8i8 V64:$Rn)), (NEGv8i8 V64:$Rn)>;
2509 def : Pat<(AArch64neg (v16i8 V128:$Rn)), (NEGv16i8 V128:$Rn)>;
2510 def : Pat<(AArch64neg (v4i16 V64:$Rn)), (NEGv4i16 V64:$Rn)>;
2511 def : Pat<(AArch64neg (v8i16 V128:$Rn)), (NEGv8i16 V128:$Rn)>;
2512 def : Pat<(AArch64neg (v2i32 V64:$Rn)), (NEGv2i32 V64:$Rn)>;
2513 def : Pat<(AArch64neg (v4i32 V128:$Rn)), (NEGv4i32 V128:$Rn)>;
2514 def : Pat<(AArch64neg (v2i64 V128:$Rn)), (NEGv2i64 V128:$Rn)>;
2516 def : Pat<(AArch64not (v8i8 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2517 def : Pat<(AArch64not (v16i8 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2518 def : Pat<(AArch64not (v4i16 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2519 def : Pat<(AArch64not (v8i16 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2520 def : Pat<(AArch64not (v2i32 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2521 def : Pat<(AArch64not (v1i64 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2522 def : Pat<(AArch64not (v4i32 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2523 def : Pat<(AArch64not (v2i64 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2525 def : Pat<(vnot (v4i16 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2526 def : Pat<(vnot (v8i16 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2527 def : Pat<(vnot (v2i32 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2528 def : Pat<(vnot (v4i32 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2529 def : Pat<(vnot (v2i64 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2531 defm RBIT : SIMDTwoVectorB<1, 0b01, 0b00101, "rbit", int_aarch64_neon_rbit>;
2532 defm REV16 : SIMDTwoVectorB<0, 0b00, 0b00001, "rev16", AArch64rev16>;
2533 defm REV32 : SIMDTwoVectorBH<1, 0b00000, "rev32", AArch64rev32>;
2534 defm REV64 : SIMDTwoVectorBHS<0, 0b00000, "rev64", AArch64rev64>;
2535 defm SADALP : SIMDLongTwoVectorTied<0, 0b00110, "sadalp",
2536 BinOpFrag<(add node:$LHS, (int_aarch64_neon_saddlp node:$RHS))> >;
2537 defm SADDLP : SIMDLongTwoVector<0, 0b00010, "saddlp", int_aarch64_neon_saddlp>;
2538 defm SCVTF : SIMDTwoVectorIntToFP<0, 0, 0b11101, "scvtf", sint_to_fp>;
2539 defm SHLL : SIMDVectorLShiftLongBySizeBHS;
2540 defm SQABS : SIMDTwoVectorBHSD<0, 0b00111, "sqabs", int_aarch64_neon_sqabs>;
2541 defm SQNEG : SIMDTwoVectorBHSD<1, 0b00111, "sqneg", int_aarch64_neon_sqneg>;
2542 defm SQXTN : SIMDMixedTwoVector<0, 0b10100, "sqxtn", int_aarch64_neon_sqxtn>;
2543 defm SQXTUN : SIMDMixedTwoVector<1, 0b10010, "sqxtun", int_aarch64_neon_sqxtun>;
2544 defm SUQADD : SIMDTwoVectorBHSDTied<0, 0b00011, "suqadd",int_aarch64_neon_suqadd>;
2545 defm UADALP : SIMDLongTwoVectorTied<1, 0b00110, "uadalp",
2546 BinOpFrag<(add node:$LHS, (int_aarch64_neon_uaddlp node:$RHS))> >;
2547 defm UADDLP : SIMDLongTwoVector<1, 0b00010, "uaddlp",
2548 int_aarch64_neon_uaddlp>;
2549 defm UCVTF : SIMDTwoVectorIntToFP<1, 0, 0b11101, "ucvtf", uint_to_fp>;
2550 defm UQXTN : SIMDMixedTwoVector<1, 0b10100, "uqxtn", int_aarch64_neon_uqxtn>;
2551 defm URECPE : SIMDTwoVectorS<0, 1, 0b11100, "urecpe", int_aarch64_neon_urecpe>;
2552 defm URSQRTE: SIMDTwoVectorS<1, 1, 0b11100, "ursqrte", int_aarch64_neon_ursqrte>;
2553 defm USQADD : SIMDTwoVectorBHSDTied<1, 0b00011, "usqadd",int_aarch64_neon_usqadd>;
2554 defm XTN : SIMDMixedTwoVector<0, 0b10010, "xtn", trunc>;
2556 def : Pat<(v4f16 (AArch64rev32 V64:$Rn)), (REV32v4i16 V64:$Rn)>;
2557 def : Pat<(v4f16 (AArch64rev64 V64:$Rn)), (REV64v4i16 V64:$Rn)>;
2558 def : Pat<(v8f16 (AArch64rev32 V128:$Rn)), (REV32v8i16 V128:$Rn)>;
2559 def : Pat<(v8f16 (AArch64rev64 V128:$Rn)), (REV64v8i16 V128:$Rn)>;
2560 def : Pat<(v2f32 (AArch64rev64 V64:$Rn)), (REV64v2i32 V64:$Rn)>;
2561 def : Pat<(v4f32 (AArch64rev64 V128:$Rn)), (REV64v4i32 V128:$Rn)>;
2563 // Patterns for vector long shift (by element width). These need to match all
2564 // three of zext, sext and anyext so it's easier to pull the patterns out of the
2566 multiclass SIMDVectorLShiftLongBySizeBHSPats<SDPatternOperator ext> {
2567 def : Pat<(AArch64vshl (v8i16 (ext (v8i8 V64:$Rn))), (i32 8)),
2568 (SHLLv8i8 V64:$Rn)>;
2569 def : Pat<(AArch64vshl (v8i16 (ext (extract_high_v16i8 V128:$Rn))), (i32 8)),
2570 (SHLLv16i8 V128:$Rn)>;
2571 def : Pat<(AArch64vshl (v4i32 (ext (v4i16 V64:$Rn))), (i32 16)),
2572 (SHLLv4i16 V64:$Rn)>;
2573 def : Pat<(AArch64vshl (v4i32 (ext (extract_high_v8i16 V128:$Rn))), (i32 16)),
2574 (SHLLv8i16 V128:$Rn)>;
2575 def : Pat<(AArch64vshl (v2i64 (ext (v2i32 V64:$Rn))), (i32 32)),
2576 (SHLLv2i32 V64:$Rn)>;
2577 def : Pat<(AArch64vshl (v2i64 (ext (extract_high_v4i32 V128:$Rn))), (i32 32)),
2578 (SHLLv4i32 V128:$Rn)>;
2581 defm : SIMDVectorLShiftLongBySizeBHSPats<anyext>;
2582 defm : SIMDVectorLShiftLongBySizeBHSPats<zext>;
2583 defm : SIMDVectorLShiftLongBySizeBHSPats<sext>;
2585 //===----------------------------------------------------------------------===//
2586 // Advanced SIMD three vector instructions.
2587 //===----------------------------------------------------------------------===//
2589 defm ADD : SIMDThreeSameVector<0, 0b10000, "add", add>;
2590 defm ADDP : SIMDThreeSameVector<0, 0b10111, "addp", int_aarch64_neon_addp>;
2591 defm CMEQ : SIMDThreeSameVector<1, 0b10001, "cmeq", AArch64cmeq>;
2592 defm CMGE : SIMDThreeSameVector<0, 0b00111, "cmge", AArch64cmge>;
2593 defm CMGT : SIMDThreeSameVector<0, 0b00110, "cmgt", AArch64cmgt>;
2594 defm CMHI : SIMDThreeSameVector<1, 0b00110, "cmhi", AArch64cmhi>;
2595 defm CMHS : SIMDThreeSameVector<1, 0b00111, "cmhs", AArch64cmhs>;
2596 defm CMTST : SIMDThreeSameVector<0, 0b10001, "cmtst", AArch64cmtst>;
2597 defm FABD : SIMDThreeSameVectorFP<1,1,0b11010,"fabd", int_aarch64_neon_fabd>;
2598 defm FACGE : SIMDThreeSameVectorFPCmp<1,0,0b11101,"facge",int_aarch64_neon_facge>;
2599 defm FACGT : SIMDThreeSameVectorFPCmp<1,1,0b11101,"facgt",int_aarch64_neon_facgt>;
2600 defm FADDP : SIMDThreeSameVectorFP<1,0,0b11010,"faddp",int_aarch64_neon_addp>;
2601 defm FADD : SIMDThreeSameVectorFP<0,0,0b11010,"fadd", fadd>;
2602 defm FCMEQ : SIMDThreeSameVectorFPCmp<0, 0, 0b11100, "fcmeq", AArch64fcmeq>;
2603 defm FCMGE : SIMDThreeSameVectorFPCmp<1, 0, 0b11100, "fcmge", AArch64fcmge>;
2604 defm FCMGT : SIMDThreeSameVectorFPCmp<1, 1, 0b11100, "fcmgt", AArch64fcmgt>;
2605 defm FDIV : SIMDThreeSameVectorFP<1,0,0b11111,"fdiv", fdiv>;
2606 defm FMAXNMP : SIMDThreeSameVectorFP<1,0,0b11000,"fmaxnmp", int_aarch64_neon_fmaxnmp>;
2607 defm FMAXNM : SIMDThreeSameVectorFP<0,0,0b11000,"fmaxnm", int_aarch64_neon_fmaxnm>;
2608 defm FMAXP : SIMDThreeSameVectorFP<1,0,0b11110,"fmaxp", int_aarch64_neon_fmaxp>;
2609 defm FMAX : SIMDThreeSameVectorFP<0,0,0b11110,"fmax", AArch64fmax>;
2610 defm FMINNMP : SIMDThreeSameVectorFP<1,1,0b11000,"fminnmp", int_aarch64_neon_fminnmp>;
2611 defm FMINNM : SIMDThreeSameVectorFP<0,1,0b11000,"fminnm", int_aarch64_neon_fminnm>;
2612 defm FMINP : SIMDThreeSameVectorFP<1,1,0b11110,"fminp", int_aarch64_neon_fminp>;
2613 defm FMIN : SIMDThreeSameVectorFP<0,1,0b11110,"fmin", AArch64fmin>;
2615 // NOTE: The operands of the PatFrag are reordered on FMLA/FMLS because the
2616 // instruction expects the addend first, while the fma intrinsic puts it last.
2617 defm FMLA : SIMDThreeSameVectorFPTied<0, 0, 0b11001, "fmla",
2618 TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)> >;
2619 defm FMLS : SIMDThreeSameVectorFPTied<0, 1, 0b11001, "fmls",
2620 TriOpFrag<(fma node:$MHS, (fneg node:$RHS), node:$LHS)> >;
2622 // The following def pats catch the case where the LHS of an FMA is negated.
2623 // The TriOpFrag above catches the case where the middle operand is negated.
2624 def : Pat<(v2f32 (fma (fneg V64:$Rn), V64:$Rm, V64:$Rd)),
2625 (FMLSv2f32 V64:$Rd, V64:$Rn, V64:$Rm)>;
2627 def : Pat<(v4f32 (fma (fneg V128:$Rn), V128:$Rm, V128:$Rd)),
2628 (FMLSv4f32 V128:$Rd, V128:$Rn, V128:$Rm)>;
2630 def : Pat<(v2f64 (fma (fneg V128:$Rn), V128:$Rm, V128:$Rd)),
2631 (FMLSv2f64 V128:$Rd, V128:$Rn, V128:$Rm)>;
2633 defm FMULX : SIMDThreeSameVectorFP<0,0,0b11011,"fmulx", int_aarch64_neon_fmulx>;
2634 defm FMUL : SIMDThreeSameVectorFP<1,0,0b11011,"fmul", fmul>;
2635 defm FRECPS : SIMDThreeSameVectorFP<0,0,0b11111,"frecps", int_aarch64_neon_frecps>;
2636 defm FRSQRTS : SIMDThreeSameVectorFP<0,1,0b11111,"frsqrts", int_aarch64_neon_frsqrts>;
2637 defm FSUB : SIMDThreeSameVectorFP<0,1,0b11010,"fsub", fsub>;
2638 defm MLA : SIMDThreeSameVectorBHSTied<0, 0b10010, "mla",
2639 TriOpFrag<(add node:$LHS, (mul node:$MHS, node:$RHS))> >;
2640 defm MLS : SIMDThreeSameVectorBHSTied<1, 0b10010, "mls",
2641 TriOpFrag<(sub node:$LHS, (mul node:$MHS, node:$RHS))> >;
2642 defm MUL : SIMDThreeSameVectorBHS<0, 0b10011, "mul", mul>;
2643 defm PMUL : SIMDThreeSameVectorB<1, 0b10011, "pmul", int_aarch64_neon_pmul>;
2644 defm SABA : SIMDThreeSameVectorBHSTied<0, 0b01111, "saba",
2645 TriOpFrag<(add node:$LHS, (int_aarch64_neon_sabd node:$MHS, node:$RHS))> >;
2646 defm SABD : SIMDThreeSameVectorBHS<0,0b01110,"sabd", int_aarch64_neon_sabd>;
2647 defm SHADD : SIMDThreeSameVectorBHS<0,0b00000,"shadd", int_aarch64_neon_shadd>;
2648 defm SHSUB : SIMDThreeSameVectorBHS<0,0b00100,"shsub", int_aarch64_neon_shsub>;
2649 defm SMAXP : SIMDThreeSameVectorBHS<0,0b10100,"smaxp", int_aarch64_neon_smaxp>;
2650 defm SMAX : SIMDThreeSameVectorBHS<0,0b01100,"smax", int_aarch64_neon_smax>;
2651 defm SMINP : SIMDThreeSameVectorBHS<0,0b10101,"sminp", int_aarch64_neon_sminp>;
2652 defm SMIN : SIMDThreeSameVectorBHS<0,0b01101,"smin", int_aarch64_neon_smin>;
2653 defm SQADD : SIMDThreeSameVector<0,0b00001,"sqadd", int_aarch64_neon_sqadd>;
2654 defm SQDMULH : SIMDThreeSameVectorHS<0,0b10110,"sqdmulh",int_aarch64_neon_sqdmulh>;
2655 defm SQRDMULH : SIMDThreeSameVectorHS<1,0b10110,"sqrdmulh",int_aarch64_neon_sqrdmulh>;
2656 defm SQRSHL : SIMDThreeSameVector<0,0b01011,"sqrshl", int_aarch64_neon_sqrshl>;
2657 defm SQSHL : SIMDThreeSameVector<0,0b01001,"sqshl", int_aarch64_neon_sqshl>;
2658 defm SQSUB : SIMDThreeSameVector<0,0b00101,"sqsub", int_aarch64_neon_sqsub>;
2659 defm SRHADD : SIMDThreeSameVectorBHS<0,0b00010,"srhadd",int_aarch64_neon_srhadd>;
2660 defm SRSHL : SIMDThreeSameVector<0,0b01010,"srshl", int_aarch64_neon_srshl>;
2661 defm SSHL : SIMDThreeSameVector<0,0b01000,"sshl", int_aarch64_neon_sshl>;
2662 defm SUB : SIMDThreeSameVector<1,0b10000,"sub", sub>;
2663 defm UABA : SIMDThreeSameVectorBHSTied<1, 0b01111, "uaba",
2664 TriOpFrag<(add node:$LHS, (int_aarch64_neon_uabd node:$MHS, node:$RHS))> >;
2665 defm UABD : SIMDThreeSameVectorBHS<1,0b01110,"uabd", int_aarch64_neon_uabd>;
2666 defm UHADD : SIMDThreeSameVectorBHS<1,0b00000,"uhadd", int_aarch64_neon_uhadd>;
2667 defm UHSUB : SIMDThreeSameVectorBHS<1,0b00100,"uhsub", int_aarch64_neon_uhsub>;
2668 defm UMAXP : SIMDThreeSameVectorBHS<1,0b10100,"umaxp", int_aarch64_neon_umaxp>;
2669 defm UMAX : SIMDThreeSameVectorBHS<1,0b01100,"umax", int_aarch64_neon_umax>;
2670 defm UMINP : SIMDThreeSameVectorBHS<1,0b10101,"uminp", int_aarch64_neon_uminp>;
2671 defm UMIN : SIMDThreeSameVectorBHS<1,0b01101,"umin", int_aarch64_neon_umin>;
2672 defm UQADD : SIMDThreeSameVector<1,0b00001,"uqadd", int_aarch64_neon_uqadd>;
2673 defm UQRSHL : SIMDThreeSameVector<1,0b01011,"uqrshl", int_aarch64_neon_uqrshl>;
2674 defm UQSHL : SIMDThreeSameVector<1,0b01001,"uqshl", int_aarch64_neon_uqshl>;
2675 defm UQSUB : SIMDThreeSameVector<1,0b00101,"uqsub", int_aarch64_neon_uqsub>;
2676 defm URHADD : SIMDThreeSameVectorBHS<1,0b00010,"urhadd", int_aarch64_neon_urhadd>;
2677 defm URSHL : SIMDThreeSameVector<1,0b01010,"urshl", int_aarch64_neon_urshl>;
2678 defm USHL : SIMDThreeSameVector<1,0b01000,"ushl", int_aarch64_neon_ushl>;
2680 defm AND : SIMDLogicalThreeVector<0, 0b00, "and", and>;
2681 defm BIC : SIMDLogicalThreeVector<0, 0b01, "bic",
2682 BinOpFrag<(and node:$LHS, (vnot node:$RHS))> >;
2683 defm BIF : SIMDLogicalThreeVector<1, 0b11, "bif">;
2684 defm BIT : SIMDLogicalThreeVectorTied<1, 0b10, "bit", AArch64bit>;
2685 defm BSL : SIMDLogicalThreeVectorTied<1, 0b01, "bsl",
2686 TriOpFrag<(or (and node:$LHS, node:$MHS), (and (vnot node:$LHS), node:$RHS))>>;
2687 defm EOR : SIMDLogicalThreeVector<1, 0b00, "eor", xor>;
2688 defm ORN : SIMDLogicalThreeVector<0, 0b11, "orn",
2689 BinOpFrag<(or node:$LHS, (vnot node:$RHS))> >;
2690 defm ORR : SIMDLogicalThreeVector<0, 0b10, "orr", or>;
2692 def : Pat<(AArch64bsl (v8i8 V64:$Rd), V64:$Rn, V64:$Rm),
2693 (BSLv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
2694 def : Pat<(AArch64bsl (v4i16 V64:$Rd), V64:$Rn, V64:$Rm),
2695 (BSLv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
2696 def : Pat<(AArch64bsl (v2i32 V64:$Rd), V64:$Rn, V64:$Rm),
2697 (BSLv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
2698 def : Pat<(AArch64bsl (v1i64 V64:$Rd), V64:$Rn, V64:$Rm),
2699 (BSLv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
2701 def : Pat<(AArch64bsl (v16i8 V128:$Rd), V128:$Rn, V128:$Rm),
2702 (BSLv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
2703 def : Pat<(AArch64bsl (v8i16 V128:$Rd), V128:$Rn, V128:$Rm),
2704 (BSLv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
2705 def : Pat<(AArch64bsl (v4i32 V128:$Rd), V128:$Rn, V128:$Rm),
2706 (BSLv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
2707 def : Pat<(AArch64bsl (v2i64 V128:$Rd), V128:$Rn, V128:$Rm),
2708 (BSLv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
2710 def : InstAlias<"mov{\t$dst.16b, $src.16b|.16b\t$dst, $src}",
2711 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 1>;
2712 def : InstAlias<"mov{\t$dst.8h, $src.8h|.8h\t$dst, $src}",
2713 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2714 def : InstAlias<"mov{\t$dst.4s, $src.4s|.4s\t$dst, $src}",
2715 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2716 def : InstAlias<"mov{\t$dst.2d, $src.2d|.2d\t$dst, $src}",
2717 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2719 def : InstAlias<"mov{\t$dst.8b, $src.8b|.8b\t$dst, $src}",
2720 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 1>;
2721 def : InstAlias<"mov{\t$dst.4h, $src.4h|.4h\t$dst, $src}",
2722 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2723 def : InstAlias<"mov{\t$dst.2s, $src.2s|.2s\t$dst, $src}",
2724 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2725 def : InstAlias<"mov{\t$dst.1d, $src.1d|.1d\t$dst, $src}",
2726 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2728 def : InstAlias<"{cmls\t$dst.8b, $src1.8b, $src2.8b" #
2729 "|cmls.8b\t$dst, $src1, $src2}",
2730 (CMHSv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2731 def : InstAlias<"{cmls\t$dst.16b, $src1.16b, $src2.16b" #
2732 "|cmls.16b\t$dst, $src1, $src2}",
2733 (CMHSv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2734 def : InstAlias<"{cmls\t$dst.4h, $src1.4h, $src2.4h" #
2735 "|cmls.4h\t$dst, $src1, $src2}",
2736 (CMHSv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2737 def : InstAlias<"{cmls\t$dst.8h, $src1.8h, $src2.8h" #
2738 "|cmls.8h\t$dst, $src1, $src2}",
2739 (CMHSv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2740 def : InstAlias<"{cmls\t$dst.2s, $src1.2s, $src2.2s" #
2741 "|cmls.2s\t$dst, $src1, $src2}",
2742 (CMHSv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2743 def : InstAlias<"{cmls\t$dst.4s, $src1.4s, $src2.4s" #
2744 "|cmls.4s\t$dst, $src1, $src2}",
2745 (CMHSv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2746 def : InstAlias<"{cmls\t$dst.2d, $src1.2d, $src2.2d" #
2747 "|cmls.2d\t$dst, $src1, $src2}",
2748 (CMHSv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2750 def : InstAlias<"{cmlo\t$dst.8b, $src1.8b, $src2.8b" #
2751 "|cmlo.8b\t$dst, $src1, $src2}",
2752 (CMHIv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2753 def : InstAlias<"{cmlo\t$dst.16b, $src1.16b, $src2.16b" #
2754 "|cmlo.16b\t$dst, $src1, $src2}",
2755 (CMHIv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2756 def : InstAlias<"{cmlo\t$dst.4h, $src1.4h, $src2.4h" #
2757 "|cmlo.4h\t$dst, $src1, $src2}",
2758 (CMHIv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2759 def : InstAlias<"{cmlo\t$dst.8h, $src1.8h, $src2.8h" #
2760 "|cmlo.8h\t$dst, $src1, $src2}",
2761 (CMHIv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2762 def : InstAlias<"{cmlo\t$dst.2s, $src1.2s, $src2.2s" #
2763 "|cmlo.2s\t$dst, $src1, $src2}",
2764 (CMHIv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2765 def : InstAlias<"{cmlo\t$dst.4s, $src1.4s, $src2.4s" #
2766 "|cmlo.4s\t$dst, $src1, $src2}",
2767 (CMHIv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2768 def : InstAlias<"{cmlo\t$dst.2d, $src1.2d, $src2.2d" #
2769 "|cmlo.2d\t$dst, $src1, $src2}",
2770 (CMHIv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2772 def : InstAlias<"{cmle\t$dst.8b, $src1.8b, $src2.8b" #
2773 "|cmle.8b\t$dst, $src1, $src2}",
2774 (CMGEv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2775 def : InstAlias<"{cmle\t$dst.16b, $src1.16b, $src2.16b" #
2776 "|cmle.16b\t$dst, $src1, $src2}",
2777 (CMGEv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2778 def : InstAlias<"{cmle\t$dst.4h, $src1.4h, $src2.4h" #
2779 "|cmle.4h\t$dst, $src1, $src2}",
2780 (CMGEv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2781 def : InstAlias<"{cmle\t$dst.8h, $src1.8h, $src2.8h" #
2782 "|cmle.8h\t$dst, $src1, $src2}",
2783 (CMGEv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2784 def : InstAlias<"{cmle\t$dst.2s, $src1.2s, $src2.2s" #
2785 "|cmle.2s\t$dst, $src1, $src2}",
2786 (CMGEv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2787 def : InstAlias<"{cmle\t$dst.4s, $src1.4s, $src2.4s" #
2788 "|cmle.4s\t$dst, $src1, $src2}",
2789 (CMGEv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2790 def : InstAlias<"{cmle\t$dst.2d, $src1.2d, $src2.2d" #
2791 "|cmle.2d\t$dst, $src1, $src2}",
2792 (CMGEv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2794 def : InstAlias<"{cmlt\t$dst.8b, $src1.8b, $src2.8b" #
2795 "|cmlt.8b\t$dst, $src1, $src2}",
2796 (CMGTv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2797 def : InstAlias<"{cmlt\t$dst.16b, $src1.16b, $src2.16b" #
2798 "|cmlt.16b\t$dst, $src1, $src2}",
2799 (CMGTv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2800 def : InstAlias<"{cmlt\t$dst.4h, $src1.4h, $src2.4h" #
2801 "|cmlt.4h\t$dst, $src1, $src2}",
2802 (CMGTv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2803 def : InstAlias<"{cmlt\t$dst.8h, $src1.8h, $src2.8h" #
2804 "|cmlt.8h\t$dst, $src1, $src2}",
2805 (CMGTv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2806 def : InstAlias<"{cmlt\t$dst.2s, $src1.2s, $src2.2s" #
2807 "|cmlt.2s\t$dst, $src1, $src2}",
2808 (CMGTv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2809 def : InstAlias<"{cmlt\t$dst.4s, $src1.4s, $src2.4s" #
2810 "|cmlt.4s\t$dst, $src1, $src2}",
2811 (CMGTv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2812 def : InstAlias<"{cmlt\t$dst.2d, $src1.2d, $src2.2d" #
2813 "|cmlt.2d\t$dst, $src1, $src2}",
2814 (CMGTv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2816 def : InstAlias<"{fcmle\t$dst.2s, $src1.2s, $src2.2s" #
2817 "|fcmle.2s\t$dst, $src1, $src2}",
2818 (FCMGEv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
2819 def : InstAlias<"{fcmle\t$dst.4s, $src1.4s, $src2.4s" #
2820 "|fcmle.4s\t$dst, $src1, $src2}",
2821 (FCMGEv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
2822 def : InstAlias<"{fcmle\t$dst.2d, $src1.2d, $src2.2d" #
2823 "|fcmle.2d\t$dst, $src1, $src2}",
2824 (FCMGEv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
2826 def : InstAlias<"{fcmlt\t$dst.2s, $src1.2s, $src2.2s" #
2827 "|fcmlt.2s\t$dst, $src1, $src2}",
2828 (FCMGTv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
2829 def : InstAlias<"{fcmlt\t$dst.4s, $src1.4s, $src2.4s" #
2830 "|fcmlt.4s\t$dst, $src1, $src2}",
2831 (FCMGTv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
2832 def : InstAlias<"{fcmlt\t$dst.2d, $src1.2d, $src2.2d" #
2833 "|fcmlt.2d\t$dst, $src1, $src2}",
2834 (FCMGTv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
2836 def : InstAlias<"{facle\t$dst.2s, $src1.2s, $src2.2s" #
2837 "|facle.2s\t$dst, $src1, $src2}",
2838 (FACGEv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
2839 def : InstAlias<"{facle\t$dst.4s, $src1.4s, $src2.4s" #
2840 "|facle.4s\t$dst, $src1, $src2}",
2841 (FACGEv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
2842 def : InstAlias<"{facle\t$dst.2d, $src1.2d, $src2.2d" #
2843 "|facle.2d\t$dst, $src1, $src2}",
2844 (FACGEv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
2846 def : InstAlias<"{faclt\t$dst.2s, $src1.2s, $src2.2s" #
2847 "|faclt.2s\t$dst, $src1, $src2}",
2848 (FACGTv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
2849 def : InstAlias<"{faclt\t$dst.4s, $src1.4s, $src2.4s" #
2850 "|faclt.4s\t$dst, $src1, $src2}",
2851 (FACGTv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
2852 def : InstAlias<"{faclt\t$dst.2d, $src1.2d, $src2.2d" #
2853 "|faclt.2d\t$dst, $src1, $src2}",
2854 (FACGTv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
2856 //===----------------------------------------------------------------------===//
2857 // Advanced SIMD three scalar instructions.
2858 //===----------------------------------------------------------------------===//
2860 defm ADD : SIMDThreeScalarD<0, 0b10000, "add", add>;
2861 defm CMEQ : SIMDThreeScalarD<1, 0b10001, "cmeq", AArch64cmeq>;
2862 defm CMGE : SIMDThreeScalarD<0, 0b00111, "cmge", AArch64cmge>;
2863 defm CMGT : SIMDThreeScalarD<0, 0b00110, "cmgt", AArch64cmgt>;
2864 defm CMHI : SIMDThreeScalarD<1, 0b00110, "cmhi", AArch64cmhi>;
2865 defm CMHS : SIMDThreeScalarD<1, 0b00111, "cmhs", AArch64cmhs>;
2866 defm CMTST : SIMDThreeScalarD<0, 0b10001, "cmtst", AArch64cmtst>;
2867 defm FABD : SIMDThreeScalarSD<1, 1, 0b11010, "fabd", int_aarch64_sisd_fabd>;
2868 def : Pat<(v1f64 (int_aarch64_neon_fabd (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2869 (FABD64 FPR64:$Rn, FPR64:$Rm)>;
2870 defm FACGE : SIMDThreeScalarFPCmp<1, 0, 0b11101, "facge",
2871 int_aarch64_neon_facge>;
2872 defm FACGT : SIMDThreeScalarFPCmp<1, 1, 0b11101, "facgt",
2873 int_aarch64_neon_facgt>;
2874 defm FCMEQ : SIMDThreeScalarFPCmp<0, 0, 0b11100, "fcmeq", AArch64fcmeq>;
2875 defm FCMGE : SIMDThreeScalarFPCmp<1, 0, 0b11100, "fcmge", AArch64fcmge>;
2876 defm FCMGT : SIMDThreeScalarFPCmp<1, 1, 0b11100, "fcmgt", AArch64fcmgt>;
2877 defm FMULX : SIMDThreeScalarSD<0, 0, 0b11011, "fmulx", int_aarch64_neon_fmulx>;
2878 defm FRECPS : SIMDThreeScalarSD<0, 0, 0b11111, "frecps", int_aarch64_neon_frecps>;
2879 defm FRSQRTS : SIMDThreeScalarSD<0, 1, 0b11111, "frsqrts", int_aarch64_neon_frsqrts>;
2880 defm SQADD : SIMDThreeScalarBHSD<0, 0b00001, "sqadd", int_aarch64_neon_sqadd>;
2881 defm SQDMULH : SIMDThreeScalarHS< 0, 0b10110, "sqdmulh", int_aarch64_neon_sqdmulh>;
2882 defm SQRDMULH : SIMDThreeScalarHS< 1, 0b10110, "sqrdmulh", int_aarch64_neon_sqrdmulh>;
2883 defm SQRSHL : SIMDThreeScalarBHSD<0, 0b01011, "sqrshl",int_aarch64_neon_sqrshl>;
2884 defm SQSHL : SIMDThreeScalarBHSD<0, 0b01001, "sqshl", int_aarch64_neon_sqshl>;
2885 defm SQSUB : SIMDThreeScalarBHSD<0, 0b00101, "sqsub", int_aarch64_neon_sqsub>;
2886 defm SRSHL : SIMDThreeScalarD< 0, 0b01010, "srshl", int_aarch64_neon_srshl>;
2887 defm SSHL : SIMDThreeScalarD< 0, 0b01000, "sshl", int_aarch64_neon_sshl>;
2888 defm SUB : SIMDThreeScalarD< 1, 0b10000, "sub", sub>;
2889 defm UQADD : SIMDThreeScalarBHSD<1, 0b00001, "uqadd", int_aarch64_neon_uqadd>;
2890 defm UQRSHL : SIMDThreeScalarBHSD<1, 0b01011, "uqrshl",int_aarch64_neon_uqrshl>;
2891 defm UQSHL : SIMDThreeScalarBHSD<1, 0b01001, "uqshl", int_aarch64_neon_uqshl>;
2892 defm UQSUB : SIMDThreeScalarBHSD<1, 0b00101, "uqsub", int_aarch64_neon_uqsub>;
2893 defm URSHL : SIMDThreeScalarD< 1, 0b01010, "urshl", int_aarch64_neon_urshl>;
2894 defm USHL : SIMDThreeScalarD< 1, 0b01000, "ushl", int_aarch64_neon_ushl>;
2896 def : InstAlias<"cmls $dst, $src1, $src2",
2897 (CMHSv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
2898 def : InstAlias<"cmle $dst, $src1, $src2",
2899 (CMGEv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
2900 def : InstAlias<"cmlo $dst, $src1, $src2",
2901 (CMHIv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
2902 def : InstAlias<"cmlt $dst, $src1, $src2",
2903 (CMGTv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
2904 def : InstAlias<"fcmle $dst, $src1, $src2",
2905 (FCMGE32 FPR32:$dst, FPR32:$src2, FPR32:$src1), 0>;
2906 def : InstAlias<"fcmle $dst, $src1, $src2",
2907 (FCMGE64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
2908 def : InstAlias<"fcmlt $dst, $src1, $src2",
2909 (FCMGT32 FPR32:$dst, FPR32:$src2, FPR32:$src1), 0>;
2910 def : InstAlias<"fcmlt $dst, $src1, $src2",
2911 (FCMGT64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
2912 def : InstAlias<"facle $dst, $src1, $src2",
2913 (FACGE32 FPR32:$dst, FPR32:$src2, FPR32:$src1), 0>;
2914 def : InstAlias<"facle $dst, $src1, $src2",
2915 (FACGE64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
2916 def : InstAlias<"faclt $dst, $src1, $src2",
2917 (FACGT32 FPR32:$dst, FPR32:$src2, FPR32:$src1), 0>;
2918 def : InstAlias<"faclt $dst, $src1, $src2",
2919 (FACGT64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
2921 //===----------------------------------------------------------------------===//
2922 // Advanced SIMD three scalar instructions (mixed operands).
2923 //===----------------------------------------------------------------------===//
2924 defm SQDMULL : SIMDThreeScalarMixedHS<0, 0b11010, "sqdmull",
2925 int_aarch64_neon_sqdmulls_scalar>;
2926 defm SQDMLAL : SIMDThreeScalarMixedTiedHS<0, 0b10010, "sqdmlal">;
2927 defm SQDMLSL : SIMDThreeScalarMixedTiedHS<0, 0b10110, "sqdmlsl">;
2929 def : Pat<(i64 (int_aarch64_neon_sqadd (i64 FPR64:$Rd),
2930 (i64 (int_aarch64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
2931 (i32 FPR32:$Rm))))),
2932 (SQDMLALi32 FPR64:$Rd, FPR32:$Rn, FPR32:$Rm)>;
2933 def : Pat<(i64 (int_aarch64_neon_sqsub (i64 FPR64:$Rd),
2934 (i64 (int_aarch64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
2935 (i32 FPR32:$Rm))))),
2936 (SQDMLSLi32 FPR64:$Rd, FPR32:$Rn, FPR32:$Rm)>;
2938 //===----------------------------------------------------------------------===//
2939 // Advanced SIMD two scalar instructions.
2940 //===----------------------------------------------------------------------===//
2942 defm ABS : SIMDTwoScalarD< 0, 0b01011, "abs", int_aarch64_neon_abs>;
2943 defm CMEQ : SIMDCmpTwoScalarD< 0, 0b01001, "cmeq", AArch64cmeqz>;
2944 defm CMGE : SIMDCmpTwoScalarD< 1, 0b01000, "cmge", AArch64cmgez>;
2945 defm CMGT : SIMDCmpTwoScalarD< 0, 0b01000, "cmgt", AArch64cmgtz>;
2946 defm CMLE : SIMDCmpTwoScalarD< 1, 0b01001, "cmle", AArch64cmlez>;
2947 defm CMLT : SIMDCmpTwoScalarD< 0, 0b01010, "cmlt", AArch64cmltz>;
2948 defm FCMEQ : SIMDCmpTwoScalarSD<0, 1, 0b01101, "fcmeq", AArch64fcmeqz>;
2949 defm FCMGE : SIMDCmpTwoScalarSD<1, 1, 0b01100, "fcmge", AArch64fcmgez>;
2950 defm FCMGT : SIMDCmpTwoScalarSD<0, 1, 0b01100, "fcmgt", AArch64fcmgtz>;
2951 defm FCMLE : SIMDCmpTwoScalarSD<1, 1, 0b01101, "fcmle", AArch64fcmlez>;
2952 defm FCMLT : SIMDCmpTwoScalarSD<0, 1, 0b01110, "fcmlt", AArch64fcmltz>;
2953 defm FCVTAS : SIMDTwoScalarSD< 0, 0, 0b11100, "fcvtas">;
2954 defm FCVTAU : SIMDTwoScalarSD< 1, 0, 0b11100, "fcvtau">;
2955 defm FCVTMS : SIMDTwoScalarSD< 0, 0, 0b11011, "fcvtms">;
2956 defm FCVTMU : SIMDTwoScalarSD< 1, 0, 0b11011, "fcvtmu">;
2957 defm FCVTNS : SIMDTwoScalarSD< 0, 0, 0b11010, "fcvtns">;
2958 defm FCVTNU : SIMDTwoScalarSD< 1, 0, 0b11010, "fcvtnu">;
2959 defm FCVTPS : SIMDTwoScalarSD< 0, 1, 0b11010, "fcvtps">;
2960 defm FCVTPU : SIMDTwoScalarSD< 1, 1, 0b11010, "fcvtpu">;
2961 def FCVTXNv1i64 : SIMDInexactCvtTwoScalar<0b10110, "fcvtxn">;
2962 defm FCVTZS : SIMDTwoScalarSD< 0, 1, 0b11011, "fcvtzs">;
2963 defm FCVTZU : SIMDTwoScalarSD< 1, 1, 0b11011, "fcvtzu">;
2964 defm FRECPE : SIMDTwoScalarSD< 0, 1, 0b11101, "frecpe">;
2965 defm FRECPX : SIMDTwoScalarSD< 0, 1, 0b11111, "frecpx">;
2966 defm FRSQRTE : SIMDTwoScalarSD< 1, 1, 0b11101, "frsqrte">;
2967 defm NEG : SIMDTwoScalarD< 1, 0b01011, "neg",
2968 UnOpFrag<(sub immAllZerosV, node:$LHS)> >;
2969 defm SCVTF : SIMDTwoScalarCVTSD< 0, 0, 0b11101, "scvtf", AArch64sitof>;
2970 defm SQABS : SIMDTwoScalarBHSD< 0, 0b00111, "sqabs", int_aarch64_neon_sqabs>;
2971 defm SQNEG : SIMDTwoScalarBHSD< 1, 0b00111, "sqneg", int_aarch64_neon_sqneg>;
2972 defm SQXTN : SIMDTwoScalarMixedBHS< 0, 0b10100, "sqxtn", int_aarch64_neon_scalar_sqxtn>;
2973 defm SQXTUN : SIMDTwoScalarMixedBHS< 1, 0b10010, "sqxtun", int_aarch64_neon_scalar_sqxtun>;
2974 defm SUQADD : SIMDTwoScalarBHSDTied< 0, 0b00011, "suqadd",
2975 int_aarch64_neon_suqadd>;
2976 defm UCVTF : SIMDTwoScalarCVTSD< 1, 0, 0b11101, "ucvtf", AArch64uitof>;
2977 defm UQXTN : SIMDTwoScalarMixedBHS<1, 0b10100, "uqxtn", int_aarch64_neon_scalar_uqxtn>;
2978 defm USQADD : SIMDTwoScalarBHSDTied< 1, 0b00011, "usqadd",
2979 int_aarch64_neon_usqadd>;
2981 def : Pat<(AArch64neg (v1i64 V64:$Rn)), (NEGv1i64 V64:$Rn)>;
2983 def : Pat<(v1i64 (int_aarch64_neon_fcvtas (v1f64 FPR64:$Rn))),
2984 (FCVTASv1i64 FPR64:$Rn)>;
2985 def : Pat<(v1i64 (int_aarch64_neon_fcvtau (v1f64 FPR64:$Rn))),
2986 (FCVTAUv1i64 FPR64:$Rn)>;
2987 def : Pat<(v1i64 (int_aarch64_neon_fcvtms (v1f64 FPR64:$Rn))),
2988 (FCVTMSv1i64 FPR64:$Rn)>;
2989 def : Pat<(v1i64 (int_aarch64_neon_fcvtmu (v1f64 FPR64:$Rn))),
2990 (FCVTMUv1i64 FPR64:$Rn)>;
2991 def : Pat<(v1i64 (int_aarch64_neon_fcvtns (v1f64 FPR64:$Rn))),
2992 (FCVTNSv1i64 FPR64:$Rn)>;
2993 def : Pat<(v1i64 (int_aarch64_neon_fcvtnu (v1f64 FPR64:$Rn))),
2994 (FCVTNUv1i64 FPR64:$Rn)>;
2995 def : Pat<(v1i64 (int_aarch64_neon_fcvtps (v1f64 FPR64:$Rn))),
2996 (FCVTPSv1i64 FPR64:$Rn)>;
2997 def : Pat<(v1i64 (int_aarch64_neon_fcvtpu (v1f64 FPR64:$Rn))),
2998 (FCVTPUv1i64 FPR64:$Rn)>;
3000 def : Pat<(f32 (int_aarch64_neon_frecpe (f32 FPR32:$Rn))),
3001 (FRECPEv1i32 FPR32:$Rn)>;
3002 def : Pat<(f64 (int_aarch64_neon_frecpe (f64 FPR64:$Rn))),
3003 (FRECPEv1i64 FPR64:$Rn)>;
3004 def : Pat<(v1f64 (int_aarch64_neon_frecpe (v1f64 FPR64:$Rn))),
3005 (FRECPEv1i64 FPR64:$Rn)>;
3007 def : Pat<(f32 (int_aarch64_neon_frecpx (f32 FPR32:$Rn))),
3008 (FRECPXv1i32 FPR32:$Rn)>;
3009 def : Pat<(f64 (int_aarch64_neon_frecpx (f64 FPR64:$Rn))),
3010 (FRECPXv1i64 FPR64:$Rn)>;
3012 def : Pat<(f32 (int_aarch64_neon_frsqrte (f32 FPR32:$Rn))),
3013 (FRSQRTEv1i32 FPR32:$Rn)>;
3014 def : Pat<(f64 (int_aarch64_neon_frsqrte (f64 FPR64:$Rn))),
3015 (FRSQRTEv1i64 FPR64:$Rn)>;
3016 def : Pat<(v1f64 (int_aarch64_neon_frsqrte (v1f64 FPR64:$Rn))),
3017 (FRSQRTEv1i64 FPR64:$Rn)>;
3019 // If an integer is about to be converted to a floating point value,
3020 // just load it on the floating point unit.
3021 // Here are the patterns for 8 and 16-bits to float.
3023 multiclass UIntToFPROLoadPat<ValueType DstTy, ValueType SrcTy,
3024 SDPatternOperator loadop, Instruction UCVTF,
3025 ROAddrMode ro, Instruction LDRW, Instruction LDRX,
3027 def : Pat<(DstTy (uint_to_fp (SrcTy
3028 (loadop (ro.Wpat GPR64sp:$Rn, GPR32:$Rm,
3029 ro.Wext:$extend))))),
3030 (UCVTF (INSERT_SUBREG (DstTy (IMPLICIT_DEF)),
3031 (LDRW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend),
3034 def : Pat<(DstTy (uint_to_fp (SrcTy
3035 (loadop (ro.Xpat GPR64sp:$Rn, GPR64:$Rm,
3036 ro.Wext:$extend))))),
3037 (UCVTF (INSERT_SUBREG (DstTy (IMPLICIT_DEF)),
3038 (LDRX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend),
3042 defm : UIntToFPROLoadPat<f32, i32, zextloadi8,
3043 UCVTFv1i32, ro8, LDRBroW, LDRBroX, bsub>;
3044 def : Pat <(f32 (uint_to_fp (i32
3045 (zextloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))))),
3046 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
3047 (LDRBui GPR64sp:$Rn, uimm12s1:$offset), bsub))>;
3048 def : Pat <(f32 (uint_to_fp (i32
3049 (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))))),
3050 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
3051 (LDURBi GPR64sp:$Rn, simm9:$offset), bsub))>;
3052 // 16-bits -> float.
3053 defm : UIntToFPROLoadPat<f32, i32, zextloadi16,
3054 UCVTFv1i32, ro16, LDRHroW, LDRHroX, hsub>;
3055 def : Pat <(f32 (uint_to_fp (i32
3056 (zextloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))))),
3057 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
3058 (LDRHui GPR64sp:$Rn, uimm12s2:$offset), hsub))>;
3059 def : Pat <(f32 (uint_to_fp (i32
3060 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))))),
3061 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
3062 (LDURHi GPR64sp:$Rn, simm9:$offset), hsub))>;
3063 // 32-bits are handled in target specific dag combine:
3064 // performIntToFpCombine.
3065 // 64-bits integer to 32-bits floating point, not possible with
3066 // UCVTF on floating point registers (both source and destination
3067 // must have the same size).
3069 // Here are the patterns for 8, 16, 32, and 64-bits to double.
3070 // 8-bits -> double.
3071 defm : UIntToFPROLoadPat<f64, i32, zextloadi8,
3072 UCVTFv1i64, ro8, LDRBroW, LDRBroX, bsub>;
3073 def : Pat <(f64 (uint_to_fp (i32
3074 (zextloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))))),
3075 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3076 (LDRBui GPR64sp:$Rn, uimm12s1:$offset), bsub))>;
3077 def : Pat <(f64 (uint_to_fp (i32
3078 (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))))),
3079 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3080 (LDURBi GPR64sp:$Rn, simm9:$offset), bsub))>;
3081 // 16-bits -> double.
3082 defm : UIntToFPROLoadPat<f64, i32, zextloadi16,
3083 UCVTFv1i64, ro16, LDRHroW, LDRHroX, hsub>;
3084 def : Pat <(f64 (uint_to_fp (i32
3085 (zextloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))))),
3086 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3087 (LDRHui GPR64sp:$Rn, uimm12s2:$offset), hsub))>;
3088 def : Pat <(f64 (uint_to_fp (i32
3089 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))))),
3090 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3091 (LDURHi GPR64sp:$Rn, simm9:$offset), hsub))>;
3092 // 32-bits -> double.
3093 defm : UIntToFPROLoadPat<f64, i32, load,
3094 UCVTFv1i64, ro32, LDRSroW, LDRSroX, ssub>;
3095 def : Pat <(f64 (uint_to_fp (i32
3096 (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))))),
3097 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3098 (LDRSui GPR64sp:$Rn, uimm12s4:$offset), ssub))>;
3099 def : Pat <(f64 (uint_to_fp (i32
3100 (load (am_unscaled32 GPR64sp:$Rn, simm9:$offset))))),
3101 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3102 (LDURSi GPR64sp:$Rn, simm9:$offset), ssub))>;
3103 // 64-bits -> double are handled in target specific dag combine:
3104 // performIntToFpCombine.
3106 //===----------------------------------------------------------------------===//
3107 // Advanced SIMD three different-sized vector instructions.
3108 //===----------------------------------------------------------------------===//
3110 defm ADDHN : SIMDNarrowThreeVectorBHS<0,0b0100,"addhn", int_aarch64_neon_addhn>;
3111 defm SUBHN : SIMDNarrowThreeVectorBHS<0,0b0110,"subhn", int_aarch64_neon_subhn>;
3112 defm RADDHN : SIMDNarrowThreeVectorBHS<1,0b0100,"raddhn",int_aarch64_neon_raddhn>;
3113 defm RSUBHN : SIMDNarrowThreeVectorBHS<1,0b0110,"rsubhn",int_aarch64_neon_rsubhn>;
3114 defm PMULL : SIMDDifferentThreeVectorBD<0,0b1110,"pmull",int_aarch64_neon_pmull>;
3115 defm SABAL : SIMDLongThreeVectorTiedBHSabal<0,0b0101,"sabal",
3116 int_aarch64_neon_sabd>;
3117 defm SABDL : SIMDLongThreeVectorBHSabdl<0, 0b0111, "sabdl",
3118 int_aarch64_neon_sabd>;
3119 defm SADDL : SIMDLongThreeVectorBHS< 0, 0b0000, "saddl",
3120 BinOpFrag<(add (sext node:$LHS), (sext node:$RHS))>>;
3121 defm SADDW : SIMDWideThreeVectorBHS< 0, 0b0001, "saddw",
3122 BinOpFrag<(add node:$LHS, (sext node:$RHS))>>;
3123 defm SMLAL : SIMDLongThreeVectorTiedBHS<0, 0b1000, "smlal",
3124 TriOpFrag<(add node:$LHS, (int_aarch64_neon_smull node:$MHS, node:$RHS))>>;
3125 defm SMLSL : SIMDLongThreeVectorTiedBHS<0, 0b1010, "smlsl",
3126 TriOpFrag<(sub node:$LHS, (int_aarch64_neon_smull node:$MHS, node:$RHS))>>;
3127 defm SMULL : SIMDLongThreeVectorBHS<0, 0b1100, "smull", int_aarch64_neon_smull>;
3128 defm SQDMLAL : SIMDLongThreeVectorSQDMLXTiedHS<0, 0b1001, "sqdmlal",
3129 int_aarch64_neon_sqadd>;
3130 defm SQDMLSL : SIMDLongThreeVectorSQDMLXTiedHS<0, 0b1011, "sqdmlsl",
3131 int_aarch64_neon_sqsub>;
3132 defm SQDMULL : SIMDLongThreeVectorHS<0, 0b1101, "sqdmull",
3133 int_aarch64_neon_sqdmull>;
3134 defm SSUBL : SIMDLongThreeVectorBHS<0, 0b0010, "ssubl",
3135 BinOpFrag<(sub (sext node:$LHS), (sext node:$RHS))>>;
3136 defm SSUBW : SIMDWideThreeVectorBHS<0, 0b0011, "ssubw",
3137 BinOpFrag<(sub node:$LHS, (sext node:$RHS))>>;
3138 defm UABAL : SIMDLongThreeVectorTiedBHSabal<1, 0b0101, "uabal",
3139 int_aarch64_neon_uabd>;
3140 defm UABDL : SIMDLongThreeVectorBHSabdl<1, 0b0111, "uabdl",
3141 int_aarch64_neon_uabd>;
3142 defm UADDL : SIMDLongThreeVectorBHS<1, 0b0000, "uaddl",
3143 BinOpFrag<(add (zext node:$LHS), (zext node:$RHS))>>;
3144 defm UADDW : SIMDWideThreeVectorBHS<1, 0b0001, "uaddw",
3145 BinOpFrag<(add node:$LHS, (zext node:$RHS))>>;
3146 defm UMLAL : SIMDLongThreeVectorTiedBHS<1, 0b1000, "umlal",
3147 TriOpFrag<(add node:$LHS, (int_aarch64_neon_umull node:$MHS, node:$RHS))>>;
3148 defm UMLSL : SIMDLongThreeVectorTiedBHS<1, 0b1010, "umlsl",
3149 TriOpFrag<(sub node:$LHS, (int_aarch64_neon_umull node:$MHS, node:$RHS))>>;
3150 defm UMULL : SIMDLongThreeVectorBHS<1, 0b1100, "umull", int_aarch64_neon_umull>;
3151 defm USUBL : SIMDLongThreeVectorBHS<1, 0b0010, "usubl",
3152 BinOpFrag<(sub (zext node:$LHS), (zext node:$RHS))>>;
3153 defm USUBW : SIMDWideThreeVectorBHS< 1, 0b0011, "usubw",
3154 BinOpFrag<(sub node:$LHS, (zext node:$RHS))>>;
3156 // Additional patterns for SMULL and UMULL
3157 multiclass Neon_mul_widen_patterns<SDPatternOperator opnode,
3158 Instruction INST8B, Instruction INST4H, Instruction INST2S> {
3159 def : Pat<(v8i16 (opnode (v8i8 V64:$Rn), (v8i8 V64:$Rm))),
3160 (INST8B V64:$Rn, V64:$Rm)>;
3161 def : Pat<(v4i32 (opnode (v4i16 V64:$Rn), (v4i16 V64:$Rm))),
3162 (INST4H V64:$Rn, V64:$Rm)>;
3163 def : Pat<(v2i64 (opnode (v2i32 V64:$Rn), (v2i32 V64:$Rm))),
3164 (INST2S V64:$Rn, V64:$Rm)>;
3167 defm : Neon_mul_widen_patterns<AArch64smull, SMULLv8i8_v8i16,
3168 SMULLv4i16_v4i32, SMULLv2i32_v2i64>;
3169 defm : Neon_mul_widen_patterns<AArch64umull, UMULLv8i8_v8i16,
3170 UMULLv4i16_v4i32, UMULLv2i32_v2i64>;
3172 // Additional patterns for SMLAL/SMLSL and UMLAL/UMLSL
3173 multiclass Neon_mulacc_widen_patterns<SDPatternOperator opnode,
3174 Instruction INST8B, Instruction INST4H, Instruction INST2S> {
3175 def : Pat<(v8i16 (opnode (v8i16 V128:$Rd), (v8i8 V64:$Rn), (v8i8 V64:$Rm))),
3176 (INST8B V128:$Rd, V64:$Rn, V64:$Rm)>;
3177 def : Pat<(v4i32 (opnode (v4i32 V128:$Rd), (v4i16 V64:$Rn), (v4i16 V64:$Rm))),
3178 (INST4H V128:$Rd, V64:$Rn, V64:$Rm)>;
3179 def : Pat<(v2i64 (opnode (v2i64 V128:$Rd), (v2i32 V64:$Rn), (v2i32 V64:$Rm))),
3180 (INST2S V128:$Rd, V64:$Rn, V64:$Rm)>;
3183 defm : Neon_mulacc_widen_patterns<
3184 TriOpFrag<(add node:$LHS, (AArch64smull node:$MHS, node:$RHS))>,
3185 SMLALv8i8_v8i16, SMLALv4i16_v4i32, SMLALv2i32_v2i64>;
3186 defm : Neon_mulacc_widen_patterns<
3187 TriOpFrag<(add node:$LHS, (AArch64umull node:$MHS, node:$RHS))>,
3188 UMLALv8i8_v8i16, UMLALv4i16_v4i32, UMLALv2i32_v2i64>;
3189 defm : Neon_mulacc_widen_patterns<
3190 TriOpFrag<(sub node:$LHS, (AArch64smull node:$MHS, node:$RHS))>,
3191 SMLSLv8i8_v8i16, SMLSLv4i16_v4i32, SMLSLv2i32_v2i64>;
3192 defm : Neon_mulacc_widen_patterns<
3193 TriOpFrag<(sub node:$LHS, (AArch64umull node:$MHS, node:$RHS))>,
3194 UMLSLv8i8_v8i16, UMLSLv4i16_v4i32, UMLSLv2i32_v2i64>;
3196 // Patterns for 64-bit pmull
3197 def : Pat<(int_aarch64_neon_pmull64 V64:$Rn, V64:$Rm),
3198 (PMULLv1i64 V64:$Rn, V64:$Rm)>;
3199 def : Pat<(int_aarch64_neon_pmull64 (vector_extract (v2i64 V128:$Rn), (i64 1)),
3200 (vector_extract (v2i64 V128:$Rm), (i64 1))),
3201 (PMULLv2i64 V128:$Rn, V128:$Rm)>;
3203 // CodeGen patterns for addhn and subhn instructions, which can actually be
3204 // written in LLVM IR without too much difficulty.
3207 def : Pat<(v8i8 (trunc (v8i16 (AArch64vlshr (add V128:$Rn, V128:$Rm), (i32 8))))),
3208 (ADDHNv8i16_v8i8 V128:$Rn, V128:$Rm)>;
3209 def : Pat<(v4i16 (trunc (v4i32 (AArch64vlshr (add V128:$Rn, V128:$Rm),
3211 (ADDHNv4i32_v4i16 V128:$Rn, V128:$Rm)>;
3212 def : Pat<(v2i32 (trunc (v2i64 (AArch64vlshr (add V128:$Rn, V128:$Rm),
3214 (ADDHNv2i64_v2i32 V128:$Rn, V128:$Rm)>;
3215 def : Pat<(concat_vectors (v8i8 V64:$Rd),
3216 (trunc (v8i16 (AArch64vlshr (add V128:$Rn, V128:$Rm),
3218 (ADDHNv8i16_v16i8 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3219 V128:$Rn, V128:$Rm)>;
3220 def : Pat<(concat_vectors (v4i16 V64:$Rd),
3221 (trunc (v4i32 (AArch64vlshr (add V128:$Rn, V128:$Rm),
3223 (ADDHNv4i32_v8i16 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3224 V128:$Rn, V128:$Rm)>;
3225 def : Pat<(concat_vectors (v2i32 V64:$Rd),
3226 (trunc (v2i64 (AArch64vlshr (add V128:$Rn, V128:$Rm),
3228 (ADDHNv2i64_v4i32 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3229 V128:$Rn, V128:$Rm)>;
3232 def : Pat<(v8i8 (trunc (v8i16 (AArch64vlshr (sub V128:$Rn, V128:$Rm), (i32 8))))),
3233 (SUBHNv8i16_v8i8 V128:$Rn, V128:$Rm)>;
3234 def : Pat<(v4i16 (trunc (v4i32 (AArch64vlshr (sub V128:$Rn, V128:$Rm),
3236 (SUBHNv4i32_v4i16 V128:$Rn, V128:$Rm)>;
3237 def : Pat<(v2i32 (trunc (v2i64 (AArch64vlshr (sub V128:$Rn, V128:$Rm),
3239 (SUBHNv2i64_v2i32 V128:$Rn, V128:$Rm)>;
3240 def : Pat<(concat_vectors (v8i8 V64:$Rd),
3241 (trunc (v8i16 (AArch64vlshr (sub V128:$Rn, V128:$Rm),
3243 (SUBHNv8i16_v16i8 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3244 V128:$Rn, V128:$Rm)>;
3245 def : Pat<(concat_vectors (v4i16 V64:$Rd),
3246 (trunc (v4i32 (AArch64vlshr (sub V128:$Rn, V128:$Rm),
3248 (SUBHNv4i32_v8i16 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3249 V128:$Rn, V128:$Rm)>;
3250 def : Pat<(concat_vectors (v2i32 V64:$Rd),
3251 (trunc (v2i64 (AArch64vlshr (sub V128:$Rn, V128:$Rm),
3253 (SUBHNv2i64_v4i32 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3254 V128:$Rn, V128:$Rm)>;
3256 //----------------------------------------------------------------------------
3257 // AdvSIMD bitwise extract from vector instruction.
3258 //----------------------------------------------------------------------------
3260 defm EXT : SIMDBitwiseExtract<"ext">;
3262 def : Pat<(v4i16 (AArch64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
3263 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
3264 def : Pat<(v8i16 (AArch64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3265 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3266 def : Pat<(v2i32 (AArch64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
3267 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
3268 def : Pat<(v2f32 (AArch64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
3269 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
3270 def : Pat<(v4i32 (AArch64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3271 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3272 def : Pat<(v4f32 (AArch64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3273 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3274 def : Pat<(v2i64 (AArch64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3275 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3276 def : Pat<(v2f64 (AArch64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3277 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3278 def : Pat<(v4f16 (AArch64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
3279 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
3280 def : Pat<(v8f16 (AArch64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3281 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3283 // We use EXT to handle extract_subvector to copy the upper 64-bits of a
3285 def : Pat<(v8i8 (extract_subvector V128:$Rn, (i64 8))),
3286 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3287 def : Pat<(v4i16 (extract_subvector V128:$Rn, (i64 4))),
3288 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3289 def : Pat<(v2i32 (extract_subvector V128:$Rn, (i64 2))),
3290 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3291 def : Pat<(v1i64 (extract_subvector V128:$Rn, (i64 1))),
3292 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3293 def : Pat<(v4f16 (extract_subvector V128:$Rn, (i64 4))),
3294 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3295 def : Pat<(v2f32 (extract_subvector V128:$Rn, (i64 2))),
3296 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3297 def : Pat<(v1f64 (extract_subvector V128:$Rn, (i64 1))),
3298 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3301 //----------------------------------------------------------------------------
3302 // AdvSIMD zip vector
3303 //----------------------------------------------------------------------------
3305 defm TRN1 : SIMDZipVector<0b010, "trn1", AArch64trn1>;
3306 defm TRN2 : SIMDZipVector<0b110, "trn2", AArch64trn2>;
3307 defm UZP1 : SIMDZipVector<0b001, "uzp1", AArch64uzp1>;
3308 defm UZP2 : SIMDZipVector<0b101, "uzp2", AArch64uzp2>;
3309 defm ZIP1 : SIMDZipVector<0b011, "zip1", AArch64zip1>;
3310 defm ZIP2 : SIMDZipVector<0b111, "zip2", AArch64zip2>;
3312 //----------------------------------------------------------------------------
3313 // AdvSIMD TBL/TBX instructions
3314 //----------------------------------------------------------------------------
3316 defm TBL : SIMDTableLookup< 0, "tbl">;
3317 defm TBX : SIMDTableLookupTied<1, "tbx">;
3319 def : Pat<(v8i8 (int_aarch64_neon_tbl1 (v16i8 VecListOne128:$Rn), (v8i8 V64:$Ri))),
3320 (TBLv8i8One VecListOne128:$Rn, V64:$Ri)>;
3321 def : Pat<(v16i8 (int_aarch64_neon_tbl1 (v16i8 V128:$Ri), (v16i8 V128:$Rn))),
3322 (TBLv16i8One V128:$Ri, V128:$Rn)>;
3324 def : Pat<(v8i8 (int_aarch64_neon_tbx1 (v8i8 V64:$Rd),
3325 (v16i8 VecListOne128:$Rn), (v8i8 V64:$Ri))),
3326 (TBXv8i8One V64:$Rd, VecListOne128:$Rn, V64:$Ri)>;
3327 def : Pat<(v16i8 (int_aarch64_neon_tbx1 (v16i8 V128:$Rd),
3328 (v16i8 V128:$Ri), (v16i8 V128:$Rn))),
3329 (TBXv16i8One V128:$Rd, V128:$Ri, V128:$Rn)>;
3332 //----------------------------------------------------------------------------
3333 // AdvSIMD scalar CPY instruction
3334 //----------------------------------------------------------------------------
3336 defm CPY : SIMDScalarCPY<"cpy">;
3338 //----------------------------------------------------------------------------
3339 // AdvSIMD scalar pairwise instructions
3340 //----------------------------------------------------------------------------
3342 defm ADDP : SIMDPairwiseScalarD<0, 0b11011, "addp">;
3343 defm FADDP : SIMDPairwiseScalarSD<1, 0, 0b01101, "faddp">;
3344 defm FMAXNMP : SIMDPairwiseScalarSD<1, 0, 0b01100, "fmaxnmp">;
3345 defm FMAXP : SIMDPairwiseScalarSD<1, 0, 0b01111, "fmaxp">;
3346 defm FMINNMP : SIMDPairwiseScalarSD<1, 1, 0b01100, "fminnmp">;
3347 defm FMINP : SIMDPairwiseScalarSD<1, 1, 0b01111, "fminp">;
3348 def : Pat<(i64 (int_aarch64_neon_saddv (v2i64 V128:$Rn))),
3349 (ADDPv2i64p V128:$Rn)>;
3350 def : Pat<(i64 (int_aarch64_neon_uaddv (v2i64 V128:$Rn))),
3351 (ADDPv2i64p V128:$Rn)>;
3352 def : Pat<(f32 (int_aarch64_neon_faddv (v2f32 V64:$Rn))),
3353 (FADDPv2i32p V64:$Rn)>;
3354 def : Pat<(f32 (int_aarch64_neon_faddv (v4f32 V128:$Rn))),
3355 (FADDPv2i32p (EXTRACT_SUBREG (FADDPv4f32 V128:$Rn, V128:$Rn), dsub))>;
3356 def : Pat<(f64 (int_aarch64_neon_faddv (v2f64 V128:$Rn))),
3357 (FADDPv2i64p V128:$Rn)>;
3358 def : Pat<(f32 (int_aarch64_neon_fmaxnmv (v2f32 V64:$Rn))),
3359 (FMAXNMPv2i32p V64:$Rn)>;
3360 def : Pat<(f64 (int_aarch64_neon_fmaxnmv (v2f64 V128:$Rn))),
3361 (FMAXNMPv2i64p V128:$Rn)>;
3362 def : Pat<(f32 (int_aarch64_neon_fmaxv (v2f32 V64:$Rn))),
3363 (FMAXPv2i32p V64:$Rn)>;
3364 def : Pat<(f64 (int_aarch64_neon_fmaxv (v2f64 V128:$Rn))),
3365 (FMAXPv2i64p V128:$Rn)>;
3366 def : Pat<(f32 (int_aarch64_neon_fminnmv (v2f32 V64:$Rn))),
3367 (FMINNMPv2i32p V64:$Rn)>;
3368 def : Pat<(f64 (int_aarch64_neon_fminnmv (v2f64 V128:$Rn))),
3369 (FMINNMPv2i64p V128:$Rn)>;
3370 def : Pat<(f32 (int_aarch64_neon_fminv (v2f32 V64:$Rn))),
3371 (FMINPv2i32p V64:$Rn)>;
3372 def : Pat<(f64 (int_aarch64_neon_fminv (v2f64 V128:$Rn))),
3373 (FMINPv2i64p V128:$Rn)>;
3375 //----------------------------------------------------------------------------
3376 // AdvSIMD INS/DUP instructions
3377 //----------------------------------------------------------------------------
3379 def DUPv8i8gpr : SIMDDupFromMain<0, 0b00001, ".8b", v8i8, V64, GPR32>;
3380 def DUPv16i8gpr : SIMDDupFromMain<1, 0b00001, ".16b", v16i8, V128, GPR32>;
3381 def DUPv4i16gpr : SIMDDupFromMain<0, 0b00010, ".4h", v4i16, V64, GPR32>;
3382 def DUPv8i16gpr : SIMDDupFromMain<1, 0b00010, ".8h", v8i16, V128, GPR32>;
3383 def DUPv2i32gpr : SIMDDupFromMain<0, 0b00100, ".2s", v2i32, V64, GPR32>;
3384 def DUPv4i32gpr : SIMDDupFromMain<1, 0b00100, ".4s", v4i32, V128, GPR32>;
3385 def DUPv2i64gpr : SIMDDupFromMain<1, 0b01000, ".2d", v2i64, V128, GPR64>;
3387 def DUPv2i64lane : SIMDDup64FromElement;
3388 def DUPv2i32lane : SIMDDup32FromElement<0, ".2s", v2i32, V64>;
3389 def DUPv4i32lane : SIMDDup32FromElement<1, ".4s", v4i32, V128>;
3390 def DUPv4i16lane : SIMDDup16FromElement<0, ".4h", v4i16, V64>;
3391 def DUPv8i16lane : SIMDDup16FromElement<1, ".8h", v8i16, V128>;
3392 def DUPv8i8lane : SIMDDup8FromElement <0, ".8b", v8i8, V64>;
3393 def DUPv16i8lane : SIMDDup8FromElement <1, ".16b", v16i8, V128>;
3395 def : Pat<(v2f32 (AArch64dup (f32 FPR32:$Rn))),
3396 (v2f32 (DUPv2i32lane
3397 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rn, ssub),
3399 def : Pat<(v4f32 (AArch64dup (f32 FPR32:$Rn))),
3400 (v4f32 (DUPv4i32lane
3401 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rn, ssub),
3403 def : Pat<(v2f64 (AArch64dup (f64 FPR64:$Rn))),
3404 (v2f64 (DUPv2i64lane
3405 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$Rn, dsub),
3407 def : Pat<(v4f16 (AArch64dup (f16 FPR16:$Rn))),
3408 (v4f16 (DUPv4i16lane
3409 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), FPR16:$Rn, hsub),
3411 def : Pat<(v8f16 (AArch64dup (f16 FPR16:$Rn))),
3412 (v8f16 (DUPv8i16lane
3413 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), FPR16:$Rn, hsub),
3416 def : Pat<(v4f16 (AArch64duplane16 (v8f16 V128:$Rn), VectorIndexH:$imm)),
3417 (DUPv4i16lane V128:$Rn, VectorIndexH:$imm)>;
3418 def : Pat<(v8f16 (AArch64duplane16 (v8f16 V128:$Rn), VectorIndexH:$imm)),
3419 (DUPv8i16lane V128:$Rn, VectorIndexH:$imm)>;
3421 def : Pat<(v2f32 (AArch64duplane32 (v4f32 V128:$Rn), VectorIndexS:$imm)),
3422 (DUPv2i32lane V128:$Rn, VectorIndexS:$imm)>;
3423 def : Pat<(v4f32 (AArch64duplane32 (v4f32 V128:$Rn), VectorIndexS:$imm)),
3424 (DUPv4i32lane V128:$Rn, VectorIndexS:$imm)>;
3425 def : Pat<(v2f64 (AArch64duplane64 (v2f64 V128:$Rn), VectorIndexD:$imm)),
3426 (DUPv2i64lane V128:$Rn, VectorIndexD:$imm)>;
3428 // If there's an (AArch64dup (vector_extract ...) ...), we can use a duplane
3429 // instruction even if the types don't match: we just have to remap the lane
3430 // carefully. N.b. this trick only applies to truncations.
3431 def VecIndex_x2 : SDNodeXForm<imm, [{
3432 return CurDAG->getTargetConstant(2 * N->getZExtValue(), MVT::i64);
3434 def VecIndex_x4 : SDNodeXForm<imm, [{
3435 return CurDAG->getTargetConstant(4 * N->getZExtValue(), MVT::i64);
3437 def VecIndex_x8 : SDNodeXForm<imm, [{
3438 return CurDAG->getTargetConstant(8 * N->getZExtValue(), MVT::i64);
3441 multiclass DUPWithTruncPats<ValueType ResVT, ValueType Src64VT,
3442 ValueType Src128VT, ValueType ScalVT,
3443 Instruction DUP, SDNodeXForm IdxXFORM> {
3444 def : Pat<(ResVT (AArch64dup (ScalVT (vector_extract (Src128VT V128:$Rn),
3446 (DUP V128:$Rn, (IdxXFORM imm:$idx))>;
3448 def : Pat<(ResVT (AArch64dup (ScalVT (vector_extract (Src64VT V64:$Rn),
3450 (DUP (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), (IdxXFORM imm:$idx))>;
3453 defm : DUPWithTruncPats<v8i8, v4i16, v8i16, i32, DUPv8i8lane, VecIndex_x2>;
3454 defm : DUPWithTruncPats<v8i8, v2i32, v4i32, i32, DUPv8i8lane, VecIndex_x4>;
3455 defm : DUPWithTruncPats<v4i16, v2i32, v4i32, i32, DUPv4i16lane, VecIndex_x2>;
3457 defm : DUPWithTruncPats<v16i8, v4i16, v8i16, i32, DUPv16i8lane, VecIndex_x2>;
3458 defm : DUPWithTruncPats<v16i8, v2i32, v4i32, i32, DUPv16i8lane, VecIndex_x4>;
3459 defm : DUPWithTruncPats<v8i16, v2i32, v4i32, i32, DUPv8i16lane, VecIndex_x2>;
3461 multiclass DUPWithTrunci64Pats<ValueType ResVT, Instruction DUP,
3462 SDNodeXForm IdxXFORM> {
3463 def : Pat<(ResVT (AArch64dup (i32 (trunc (vector_extract (v2i64 V128:$Rn),
3465 (DUP V128:$Rn, (IdxXFORM imm:$idx))>;
3467 def : Pat<(ResVT (AArch64dup (i32 (trunc (vector_extract (v1i64 V64:$Rn),
3469 (DUP (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), (IdxXFORM imm:$idx))>;
3472 defm : DUPWithTrunci64Pats<v8i8, DUPv8i8lane, VecIndex_x8>;
3473 defm : DUPWithTrunci64Pats<v4i16, DUPv4i16lane, VecIndex_x4>;
3474 defm : DUPWithTrunci64Pats<v2i32, DUPv2i32lane, VecIndex_x2>;
3476 defm : DUPWithTrunci64Pats<v16i8, DUPv16i8lane, VecIndex_x8>;
3477 defm : DUPWithTrunci64Pats<v8i16, DUPv8i16lane, VecIndex_x4>;
3478 defm : DUPWithTrunci64Pats<v4i32, DUPv4i32lane, VecIndex_x2>;
3480 // SMOV and UMOV definitions, with some extra patterns for convenience
3484 def : Pat<(sext_inreg (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx), i8),
3485 (i32 (SMOVvi8to32 V128:$Rn, VectorIndexB:$idx))>;
3486 def : Pat<(sext_inreg (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx), i8),
3487 (i64 (SMOVvi8to64 V128:$Rn, VectorIndexB:$idx))>;
3488 def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
3489 (i32 (SMOVvi16to32 V128:$Rn, VectorIndexH:$idx))>;
3490 def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
3491 (i64 (SMOVvi16to64 V128:$Rn, VectorIndexH:$idx))>;
3492 def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
3493 (i32 (SMOVvi16to32 V128:$Rn, VectorIndexH:$idx))>;
3494 def : Pat<(sext (i32 (vector_extract (v4i32 V128:$Rn), VectorIndexS:$idx))),
3495 (i64 (SMOVvi32to64 V128:$Rn, VectorIndexS:$idx))>;
3497 // Extracting i8 or i16 elements will have the zero-extend transformed to
3498 // an 'and' mask by type legalization since neither i8 nor i16 are legal types
3499 // for AArch64. Match these patterns here since UMOV already zeroes out the high
3500 // bits of the destination register.
3501 def : Pat<(and (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx),
3503 (i32 (UMOVvi8 V128:$Rn, VectorIndexB:$idx))>;
3504 def : Pat<(and (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),
3506 (i32 (UMOVvi16 V128:$Rn, VectorIndexH:$idx))>;
3510 def : Pat<(v16i8 (scalar_to_vector GPR32:$Rn)),
3511 (SUBREG_TO_REG (i32 0),
3512 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3513 def : Pat<(v8i8 (scalar_to_vector GPR32:$Rn)),
3514 (SUBREG_TO_REG (i32 0),
3515 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3517 def : Pat<(v8i16 (scalar_to_vector GPR32:$Rn)),
3518 (SUBREG_TO_REG (i32 0),
3519 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3520 def : Pat<(v4i16 (scalar_to_vector GPR32:$Rn)),
3521 (SUBREG_TO_REG (i32 0),
3522 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3524 def : Pat<(v2i32 (scalar_to_vector (i32 FPR32:$Rn))),
3525 (v2i32 (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)),
3526 (i32 FPR32:$Rn), ssub))>;
3527 def : Pat<(v4i32 (scalar_to_vector (i32 FPR32:$Rn))),
3528 (v4i32 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
3529 (i32 FPR32:$Rn), ssub))>;
3530 def : Pat<(v2i64 (scalar_to_vector (i64 FPR64:$Rn))),
3531 (v2i64 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
3532 (i64 FPR64:$Rn), dsub))>;
3534 def : Pat<(v4f32 (scalar_to_vector (f32 FPR32:$Rn))),
3535 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rn, ssub)>;
3536 def : Pat<(v2f32 (scalar_to_vector (f32 FPR32:$Rn))),
3537 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), FPR32:$Rn, ssub)>;
3538 def : Pat<(v2f64 (scalar_to_vector (f64 FPR64:$Rn))),
3539 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$Rn, dsub)>;
3541 def : Pat<(v4f16 (vector_insert (v4f16 V64:$Rn),
3542 (f16 FPR16:$Rm), (i64 VectorIndexS:$imm))),
3545 (v8f16 (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), V64:$Rn, dsub)),
3547 (v8f16 (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), FPR16:$Rm, hsub)),
3551 def : Pat<(v8f16 (vector_insert (v8f16 V128:$Rn),
3552 (f16 FPR16:$Rm), (i64 VectorIndexH:$imm))),
3554 V128:$Rn, VectorIndexH:$imm,
3555 (v8f16 (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), FPR16:$Rm, hsub)),
3558 def : Pat<(v2f32 (vector_insert (v2f32 V64:$Rn),
3559 (f32 FPR32:$Rm), (i64 VectorIndexS:$imm))),
3562 (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), V64:$Rn, dsub)),
3564 (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rm, ssub)),
3567 def : Pat<(v4f32 (vector_insert (v4f32 V128:$Rn),
3568 (f32 FPR32:$Rm), (i64 VectorIndexS:$imm))),
3570 V128:$Rn, VectorIndexS:$imm,
3571 (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rm, ssub)),
3573 def : Pat<(v2f64 (vector_insert (v2f64 V128:$Rn),
3574 (f64 FPR64:$Rm), (i64 VectorIndexD:$imm))),
3576 V128:$Rn, VectorIndexD:$imm,
3577 (v2f64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$Rm, dsub)),
3580 // Copy an element at a constant index in one vector into a constant indexed
3581 // element of another.
3582 // FIXME refactor to a shared class/dev parameterized on vector type, vector
3583 // index type and INS extension
3584 def : Pat<(v16i8 (int_aarch64_neon_vcopy_lane
3585 (v16i8 V128:$Vd), VectorIndexB:$idx, (v16i8 V128:$Vs),
3586 VectorIndexB:$idx2)),
3588 V128:$Vd, VectorIndexB:$idx, V128:$Vs, VectorIndexB:$idx2)
3590 def : Pat<(v8i16 (int_aarch64_neon_vcopy_lane
3591 (v8i16 V128:$Vd), VectorIndexH:$idx, (v8i16 V128:$Vs),
3592 VectorIndexH:$idx2)),
3594 V128:$Vd, VectorIndexH:$idx, V128:$Vs, VectorIndexH:$idx2)
3596 def : Pat<(v4i32 (int_aarch64_neon_vcopy_lane
3597 (v4i32 V128:$Vd), VectorIndexS:$idx, (v4i32 V128:$Vs),
3598 VectorIndexS:$idx2)),
3600 V128:$Vd, VectorIndexS:$idx, V128:$Vs, VectorIndexS:$idx2)
3602 def : Pat<(v2i64 (int_aarch64_neon_vcopy_lane
3603 (v2i64 V128:$Vd), VectorIndexD:$idx, (v2i64 V128:$Vs),
3604 VectorIndexD:$idx2)),
3606 V128:$Vd, VectorIndexD:$idx, V128:$Vs, VectorIndexD:$idx2)
3609 multiclass Neon_INS_elt_pattern<ValueType VT128, ValueType VT64,
3610 ValueType VTScal, Instruction INS> {
3611 def : Pat<(VT128 (vector_insert V128:$src,
3612 (VTScal (vector_extract (VT128 V128:$Rn), imm:$Immn)),
3614 (INS V128:$src, imm:$Immd, V128:$Rn, imm:$Immn)>;
3616 def : Pat<(VT128 (vector_insert V128:$src,
3617 (VTScal (vector_extract (VT64 V64:$Rn), imm:$Immn)),
3619 (INS V128:$src, imm:$Immd,
3620 (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), imm:$Immn)>;
3622 def : Pat<(VT64 (vector_insert V64:$src,
3623 (VTScal (vector_extract (VT128 V128:$Rn), imm:$Immn)),
3625 (EXTRACT_SUBREG (INS (SUBREG_TO_REG (i64 0), V64:$src, dsub),
3626 imm:$Immd, V128:$Rn, imm:$Immn),
3629 def : Pat<(VT64 (vector_insert V64:$src,
3630 (VTScal (vector_extract (VT64 V64:$Rn), imm:$Immn)),
3633 (INS (SUBREG_TO_REG (i64 0), V64:$src, dsub), imm:$Immd,
3634 (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), imm:$Immn),
3638 defm : Neon_INS_elt_pattern<v8f16, v4f16, f16, INSvi16lane>;
3639 defm : Neon_INS_elt_pattern<v4f32, v2f32, f32, INSvi32lane>;
3640 defm : Neon_INS_elt_pattern<v2f64, v1f64, f64, INSvi64lane>;
3641 defm : Neon_INS_elt_pattern<v16i8, v8i8, i32, INSvi8lane>;
3642 defm : Neon_INS_elt_pattern<v8i16, v4i16, i32, INSvi16lane>;
3643 defm : Neon_INS_elt_pattern<v4i32, v2i32, i32, INSvi32lane>;
3644 defm : Neon_INS_elt_pattern<v2i64, v1i64, i64, INSvi32lane>;
3647 // Floating point vector extractions are codegen'd as either a sequence of
3648 // subregister extractions, possibly fed by an INS if the lane number is
3649 // anything other than zero.
3650 def : Pat<(vector_extract (v2f64 V128:$Rn), 0),
3651 (f64 (EXTRACT_SUBREG V128:$Rn, dsub))>;
3652 def : Pat<(vector_extract (v4f32 V128:$Rn), 0),
3653 (f32 (EXTRACT_SUBREG V128:$Rn, ssub))>;
3654 def : Pat<(vector_extract (v8f16 V128:$Rn), 0),
3655 (f16 (EXTRACT_SUBREG V128:$Rn, hsub))>;
3656 def : Pat<(vector_extract (v2f64 V128:$Rn), VectorIndexD:$idx),
3657 (f64 (EXTRACT_SUBREG
3658 (INSvi64lane (v2f64 (IMPLICIT_DEF)), 0,
3659 V128:$Rn, VectorIndexD:$idx),
3661 def : Pat<(vector_extract (v4f32 V128:$Rn), VectorIndexS:$idx),
3662 (f32 (EXTRACT_SUBREG
3663 (INSvi32lane (v4f32 (IMPLICIT_DEF)), 0,
3664 V128:$Rn, VectorIndexS:$idx),
3666 def : Pat<(vector_extract (v8f16 V128:$Rn), VectorIndexH:$idx),
3667 (f16 (EXTRACT_SUBREG
3668 (INSvi16lane (v8f16 (IMPLICIT_DEF)), 0,
3669 V128:$Rn, VectorIndexH:$idx),
3672 // All concat_vectors operations are canonicalised to act on i64 vectors for
3673 // AArch64. In the general case we need an instruction, which had just as well be
3675 class ConcatPat<ValueType DstTy, ValueType SrcTy>
3676 : Pat<(DstTy (concat_vectors (SrcTy V64:$Rd), V64:$Rn)),
3677 (INSvi64lane (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), 1,
3678 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rn, dsub), 0)>;
3680 def : ConcatPat<v2i64, v1i64>;
3681 def : ConcatPat<v2f64, v1f64>;
3682 def : ConcatPat<v4i32, v2i32>;
3683 def : ConcatPat<v4f32, v2f32>;
3684 def : ConcatPat<v8i16, v4i16>;
3685 def : ConcatPat<v8f16, v4f16>;
3686 def : ConcatPat<v16i8, v8i8>;
3688 // If the high lanes are undef, though, we can just ignore them:
3689 class ConcatUndefPat<ValueType DstTy, ValueType SrcTy>
3690 : Pat<(DstTy (concat_vectors (SrcTy V64:$Rn), undef)),
3691 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rn, dsub)>;
3693 def : ConcatUndefPat<v2i64, v1i64>;
3694 def : ConcatUndefPat<v2f64, v1f64>;
3695 def : ConcatUndefPat<v4i32, v2i32>;
3696 def : ConcatUndefPat<v4f32, v2f32>;
3697 def : ConcatUndefPat<v8i16, v4i16>;
3698 def : ConcatUndefPat<v16i8, v8i8>;
3700 //----------------------------------------------------------------------------
3701 // AdvSIMD across lanes instructions
3702 //----------------------------------------------------------------------------
3704 defm ADDV : SIMDAcrossLanesBHS<0, 0b11011, "addv">;
3705 defm SMAXV : SIMDAcrossLanesBHS<0, 0b01010, "smaxv">;
3706 defm SMINV : SIMDAcrossLanesBHS<0, 0b11010, "sminv">;
3707 defm UMAXV : SIMDAcrossLanesBHS<1, 0b01010, "umaxv">;
3708 defm UMINV : SIMDAcrossLanesBHS<1, 0b11010, "uminv">;
3709 defm SADDLV : SIMDAcrossLanesHSD<0, 0b00011, "saddlv">;
3710 defm UADDLV : SIMDAcrossLanesHSD<1, 0b00011, "uaddlv">;
3711 defm FMAXNMV : SIMDAcrossLanesS<0b01100, 0, "fmaxnmv", int_aarch64_neon_fmaxnmv>;
3712 defm FMAXV : SIMDAcrossLanesS<0b01111, 0, "fmaxv", int_aarch64_neon_fmaxv>;
3713 defm FMINNMV : SIMDAcrossLanesS<0b01100, 1, "fminnmv", int_aarch64_neon_fminnmv>;
3714 defm FMINV : SIMDAcrossLanesS<0b01111, 1, "fminv", int_aarch64_neon_fminv>;
3716 multiclass SIMDAcrossLanesSignedIntrinsic<string baseOpc, Intrinsic intOp> {
3717 // If there is a sign extension after this intrinsic, consume it as smov already
3719 def : Pat<(i32 (sext_inreg (i32 (intOp (v8i8 V64:$Rn))), i8)),
3721 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3722 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3724 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
3726 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3727 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3729 // If there is a sign extension after this intrinsic, consume it as smov already
3731 def : Pat<(i32 (sext_inreg (i32 (intOp (v16i8 V128:$Rn))), i8)),
3733 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3734 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3736 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
3738 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3739 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3741 // If there is a sign extension after this intrinsic, consume it as smov already
3743 def : Pat<(i32 (sext_inreg (i32 (intOp (v4i16 V64:$Rn))), i16)),
3745 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3746 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3748 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
3750 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3751 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3753 // If there is a sign extension after this intrinsic, consume it as smov already
3755 def : Pat<(i32 (sext_inreg (i32 (intOp (v8i16 V128:$Rn))), i16)),
3757 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3758 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3760 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
3762 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3763 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3766 def : Pat<(i32 (intOp (v4i32 V128:$Rn))),
3767 (i32 (EXTRACT_SUBREG
3768 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3769 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), ssub),
3773 multiclass SIMDAcrossLanesUnsignedIntrinsic<string baseOpc, Intrinsic intOp> {
3774 // If there is a masking operation keeping only what has been actually
3775 // generated, consume it.
3776 def : Pat<(i32 (and (i32 (intOp (v8i8 V64:$Rn))), maski8_or_more)),
3777 (i32 (EXTRACT_SUBREG
3778 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3779 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3781 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
3782 (i32 (EXTRACT_SUBREG
3783 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3784 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3786 // If there is a masking operation keeping only what has been actually
3787 // generated, consume it.
3788 def : Pat<(i32 (and (i32 (intOp (v16i8 V128:$Rn))), maski8_or_more)),
3789 (i32 (EXTRACT_SUBREG
3790 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3791 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3793 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
3794 (i32 (EXTRACT_SUBREG
3795 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3796 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3799 // If there is a masking operation keeping only what has been actually
3800 // generated, consume it.
3801 def : Pat<(i32 (and (i32 (intOp (v4i16 V64:$Rn))), maski16_or_more)),
3802 (i32 (EXTRACT_SUBREG
3803 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3804 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3806 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
3807 (i32 (EXTRACT_SUBREG
3808 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3809 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3811 // If there is a masking operation keeping only what has been actually
3812 // generated, consume it.
3813 def : Pat<(i32 (and (i32 (intOp (v8i16 V128:$Rn))), maski16_or_more)),
3814 (i32 (EXTRACT_SUBREG
3815 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3816 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3818 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
3819 (i32 (EXTRACT_SUBREG
3820 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3821 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3824 def : Pat<(i32 (intOp (v4i32 V128:$Rn))),
3825 (i32 (EXTRACT_SUBREG
3826 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3827 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), ssub),
3832 multiclass SIMDAcrossLanesSignedLongIntrinsic<string baseOpc, Intrinsic intOp> {
3833 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
3835 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3836 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), hsub),
3838 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
3840 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3841 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), hsub),
3844 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
3845 (i32 (EXTRACT_SUBREG
3846 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3847 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), ssub),
3849 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
3850 (i32 (EXTRACT_SUBREG
3851 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3852 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), ssub),
3855 def : Pat<(i64 (intOp (v4i32 V128:$Rn))),
3856 (i64 (EXTRACT_SUBREG
3857 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3858 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), dsub),
3862 multiclass SIMDAcrossLanesUnsignedLongIntrinsic<string baseOpc,
3864 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
3865 (i32 (EXTRACT_SUBREG
3866 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3867 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), hsub),
3869 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
3870 (i32 (EXTRACT_SUBREG
3871 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3872 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), hsub),
3875 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
3876 (i32 (EXTRACT_SUBREG
3877 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3878 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), ssub),
3880 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
3881 (i32 (EXTRACT_SUBREG
3882 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3883 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), ssub),
3886 def : Pat<(i64 (intOp (v4i32 V128:$Rn))),
3887 (i64 (EXTRACT_SUBREG
3888 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3889 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), dsub),
3893 defm : SIMDAcrossLanesSignedIntrinsic<"ADDV", int_aarch64_neon_saddv>;
3894 // vaddv_[su]32 is special; -> ADDP Vd.2S,Vn.2S,Vm.2S; return Vd.s[0];Vn==Vm
3895 def : Pat<(i32 (int_aarch64_neon_saddv (v2i32 V64:$Rn))),
3896 (EXTRACT_SUBREG (ADDPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3898 defm : SIMDAcrossLanesUnsignedIntrinsic<"ADDV", int_aarch64_neon_uaddv>;
3899 // vaddv_[su]32 is special; -> ADDP Vd.2S,Vn.2S,Vm.2S; return Vd.s[0];Vn==Vm
3900 def : Pat<(i32 (int_aarch64_neon_uaddv (v2i32 V64:$Rn))),
3901 (EXTRACT_SUBREG (ADDPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3903 defm : SIMDAcrossLanesSignedIntrinsic<"SMAXV", int_aarch64_neon_smaxv>;
3904 def : Pat<(i32 (int_aarch64_neon_smaxv (v2i32 V64:$Rn))),
3905 (EXTRACT_SUBREG (SMAXPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3907 defm : SIMDAcrossLanesSignedIntrinsic<"SMINV", int_aarch64_neon_sminv>;
3908 def : Pat<(i32 (int_aarch64_neon_sminv (v2i32 V64:$Rn))),
3909 (EXTRACT_SUBREG (SMINPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3911 defm : SIMDAcrossLanesUnsignedIntrinsic<"UMAXV", int_aarch64_neon_umaxv>;
3912 def : Pat<(i32 (int_aarch64_neon_umaxv (v2i32 V64:$Rn))),
3913 (EXTRACT_SUBREG (UMAXPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3915 defm : SIMDAcrossLanesUnsignedIntrinsic<"UMINV", int_aarch64_neon_uminv>;
3916 def : Pat<(i32 (int_aarch64_neon_uminv (v2i32 V64:$Rn))),
3917 (EXTRACT_SUBREG (UMINPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3919 defm : SIMDAcrossLanesSignedLongIntrinsic<"SADDLV", int_aarch64_neon_saddlv>;
3920 defm : SIMDAcrossLanesUnsignedLongIntrinsic<"UADDLV", int_aarch64_neon_uaddlv>;
3922 // The vaddlv_s32 intrinsic gets mapped to SADDLP.
3923 def : Pat<(i64 (int_aarch64_neon_saddlv (v2i32 V64:$Rn))),
3924 (i64 (EXTRACT_SUBREG
3925 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3926 (SADDLPv2i32_v1i64 V64:$Rn), dsub),
3928 // The vaddlv_u32 intrinsic gets mapped to UADDLP.
3929 def : Pat<(i64 (int_aarch64_neon_uaddlv (v2i32 V64:$Rn))),
3930 (i64 (EXTRACT_SUBREG
3931 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3932 (UADDLPv2i32_v1i64 V64:$Rn), dsub),
3935 //------------------------------------------------------------------------------
3936 // AdvSIMD modified immediate instructions
3937 //------------------------------------------------------------------------------
3940 defm BIC : SIMDModifiedImmVectorShiftTied<1, 0b11, 0b01, "bic", AArch64bici>;
3942 defm ORR : SIMDModifiedImmVectorShiftTied<0, 0b11, 0b01, "orr", AArch64orri>;
3944 def : InstAlias<"bic $Vd.4h, $imm", (BICv4i16 V64:$Vd, imm0_255:$imm, 0)>;
3945 def : InstAlias<"bic $Vd.8h, $imm", (BICv8i16 V128:$Vd, imm0_255:$imm, 0)>;
3946 def : InstAlias<"bic $Vd.2s, $imm", (BICv2i32 V64:$Vd, imm0_255:$imm, 0)>;
3947 def : InstAlias<"bic $Vd.4s, $imm", (BICv4i32 V128:$Vd, imm0_255:$imm, 0)>;
3949 def : InstAlias<"bic.4h $Vd, $imm", (BICv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
3950 def : InstAlias<"bic.8h $Vd, $imm", (BICv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
3951 def : InstAlias<"bic.2s $Vd, $imm", (BICv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
3952 def : InstAlias<"bic.4s $Vd, $imm", (BICv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
3954 def : InstAlias<"orr $Vd.4h, $imm", (ORRv4i16 V64:$Vd, imm0_255:$imm, 0)>;
3955 def : InstAlias<"orr $Vd.8h, $imm", (ORRv8i16 V128:$Vd, imm0_255:$imm, 0)>;
3956 def : InstAlias<"orr $Vd.2s, $imm", (ORRv2i32 V64:$Vd, imm0_255:$imm, 0)>;
3957 def : InstAlias<"orr $Vd.4s, $imm", (ORRv4i32 V128:$Vd, imm0_255:$imm, 0)>;
3959 def : InstAlias<"orr.4h $Vd, $imm", (ORRv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
3960 def : InstAlias<"orr.8h $Vd, $imm", (ORRv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
3961 def : InstAlias<"orr.2s $Vd, $imm", (ORRv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
3962 def : InstAlias<"orr.4s $Vd, $imm", (ORRv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
3965 def FMOVv2f64_ns : SIMDModifiedImmVectorNoShift<1, 1, 0b1111, V128, fpimm8,
3967 [(set (v2f64 V128:$Rd), (AArch64fmov imm0_255:$imm8))]>;
3968 def FMOVv2f32_ns : SIMDModifiedImmVectorNoShift<0, 0, 0b1111, V64, fpimm8,
3970 [(set (v2f32 V64:$Rd), (AArch64fmov imm0_255:$imm8))]>;
3971 def FMOVv4f32_ns : SIMDModifiedImmVectorNoShift<1, 0, 0b1111, V128, fpimm8,
3973 [(set (v4f32 V128:$Rd), (AArch64fmov imm0_255:$imm8))]>;
3977 // EDIT byte mask: scalar
3978 let isReMaterializable = 1, isAsCheapAsAMove = 1 in
3979 def MOVID : SIMDModifiedImmScalarNoShift<0, 1, 0b1110, "movi",
3980 [(set FPR64:$Rd, simdimmtype10:$imm8)]>;
3981 // The movi_edit node has the immediate value already encoded, so we use
3982 // a plain imm0_255 here.
3983 def : Pat<(f64 (AArch64movi_edit imm0_255:$shift)),
3984 (MOVID imm0_255:$shift)>;
3986 def : Pat<(v1i64 immAllZerosV), (MOVID (i32 0))>;
3987 def : Pat<(v2i32 immAllZerosV), (MOVID (i32 0))>;
3988 def : Pat<(v4i16 immAllZerosV), (MOVID (i32 0))>;
3989 def : Pat<(v8i8 immAllZerosV), (MOVID (i32 0))>;
3991 def : Pat<(v1i64 immAllOnesV), (MOVID (i32 255))>;
3992 def : Pat<(v2i32 immAllOnesV), (MOVID (i32 255))>;
3993 def : Pat<(v4i16 immAllOnesV), (MOVID (i32 255))>;
3994 def : Pat<(v8i8 immAllOnesV), (MOVID (i32 255))>;
3996 // EDIT byte mask: 2d
3998 // The movi_edit node has the immediate value already encoded, so we use
3999 // a plain imm0_255 in the pattern
4000 let isReMaterializable = 1, isAsCheapAsAMove = 1 in
4001 def MOVIv2d_ns : SIMDModifiedImmVectorNoShift<1, 1, 0b1110, V128,
4004 [(set (v2i64 V128:$Rd), (AArch64movi_edit imm0_255:$imm8))]>;
4007 // Use movi.2d to materialize 0.0 if the HW does zero-cycle zeroing.
4008 // Complexity is added to break a tie with a plain MOVI.
4009 let AddedComplexity = 1 in {
4010 def : Pat<(f32 fpimm0),
4011 (f32 (EXTRACT_SUBREG (v2i64 (MOVIv2d_ns (i32 0))), ssub))>,
4013 def : Pat<(f64 fpimm0),
4014 (f64 (EXTRACT_SUBREG (v2i64 (MOVIv2d_ns (i32 0))), dsub))>,
4018 def : Pat<(v2i64 immAllZerosV), (MOVIv2d_ns (i32 0))>;
4019 def : Pat<(v4i32 immAllZerosV), (MOVIv2d_ns (i32 0))>;
4020 def : Pat<(v8i16 immAllZerosV), (MOVIv2d_ns (i32 0))>;
4021 def : Pat<(v16i8 immAllZerosV), (MOVIv2d_ns (i32 0))>;
4023 def : Pat<(v2i64 immAllOnesV), (MOVIv2d_ns (i32 255))>;
4024 def : Pat<(v4i32 immAllOnesV), (MOVIv2d_ns (i32 255))>;
4025 def : Pat<(v8i16 immAllOnesV), (MOVIv2d_ns (i32 255))>;
4026 def : Pat<(v16i8 immAllOnesV), (MOVIv2d_ns (i32 255))>;
4028 def : Pat<(v2f64 (AArch64dup (f64 fpimm0))), (MOVIv2d_ns (i32 0))>;
4029 def : Pat<(v4f32 (AArch64dup (f32 fpimm0))), (MOVIv2d_ns (i32 0))>;
4031 // EDIT per word & halfword: 2s, 4h, 4s, & 8h
4032 defm MOVI : SIMDModifiedImmVectorShift<0, 0b10, 0b00, "movi">;
4034 def : InstAlias<"movi $Vd.4h, $imm", (MOVIv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
4035 def : InstAlias<"movi $Vd.8h, $imm", (MOVIv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
4036 def : InstAlias<"movi $Vd.2s, $imm", (MOVIv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
4037 def : InstAlias<"movi $Vd.4s, $imm", (MOVIv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
4039 def : InstAlias<"movi.4h $Vd, $imm", (MOVIv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
4040 def : InstAlias<"movi.8h $Vd, $imm", (MOVIv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
4041 def : InstAlias<"movi.2s $Vd, $imm", (MOVIv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
4042 def : InstAlias<"movi.4s $Vd, $imm", (MOVIv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
4044 def : Pat<(v2i32 (AArch64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
4045 (MOVIv2i32 imm0_255:$imm8, imm:$shift)>;
4046 def : Pat<(v4i32 (AArch64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
4047 (MOVIv4i32 imm0_255:$imm8, imm:$shift)>;
4048 def : Pat<(v4i16 (AArch64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
4049 (MOVIv4i16 imm0_255:$imm8, imm:$shift)>;
4050 def : Pat<(v8i16 (AArch64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
4051 (MOVIv8i16 imm0_255:$imm8, imm:$shift)>;
4053 // EDIT per word: 2s & 4s with MSL shifter
4054 def MOVIv2s_msl : SIMDModifiedImmMoveMSL<0, 0, {1,1,0,?}, V64, "movi", ".2s",
4055 [(set (v2i32 V64:$Rd),
4056 (AArch64movi_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
4057 def MOVIv4s_msl : SIMDModifiedImmMoveMSL<1, 0, {1,1,0,?}, V128, "movi", ".4s",
4058 [(set (v4i32 V128:$Rd),
4059 (AArch64movi_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
4061 // Per byte: 8b & 16b
4062 def MOVIv8b_ns : SIMDModifiedImmVectorNoShift<0, 0, 0b1110, V64, imm0_255,
4064 [(set (v8i8 V64:$Rd), (AArch64movi imm0_255:$imm8))]>;
4065 def MOVIv16b_ns : SIMDModifiedImmVectorNoShift<1, 0, 0b1110, V128, imm0_255,
4067 [(set (v16i8 V128:$Rd), (AArch64movi imm0_255:$imm8))]>;
4071 // EDIT per word & halfword: 2s, 4h, 4s, & 8h
4072 defm MVNI : SIMDModifiedImmVectorShift<1, 0b10, 0b00, "mvni">;
4074 def : InstAlias<"mvni $Vd.4h, $imm", (MVNIv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
4075 def : InstAlias<"mvni $Vd.8h, $imm", (MVNIv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
4076 def : InstAlias<"mvni $Vd.2s, $imm", (MVNIv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
4077 def : InstAlias<"mvni $Vd.4s, $imm", (MVNIv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
4079 def : InstAlias<"mvni.4h $Vd, $imm", (MVNIv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
4080 def : InstAlias<"mvni.8h $Vd, $imm", (MVNIv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
4081 def : InstAlias<"mvni.2s $Vd, $imm", (MVNIv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
4082 def : InstAlias<"mvni.4s $Vd, $imm", (MVNIv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
4084 def : Pat<(v2i32 (AArch64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
4085 (MVNIv2i32 imm0_255:$imm8, imm:$shift)>;
4086 def : Pat<(v4i32 (AArch64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
4087 (MVNIv4i32 imm0_255:$imm8, imm:$shift)>;
4088 def : Pat<(v4i16 (AArch64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
4089 (MVNIv4i16 imm0_255:$imm8, imm:$shift)>;
4090 def : Pat<(v8i16 (AArch64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
4091 (MVNIv8i16 imm0_255:$imm8, imm:$shift)>;
4093 // EDIT per word: 2s & 4s with MSL shifter
4094 def MVNIv2s_msl : SIMDModifiedImmMoveMSL<0, 1, {1,1,0,?}, V64, "mvni", ".2s",
4095 [(set (v2i32 V64:$Rd),
4096 (AArch64mvni_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
4097 def MVNIv4s_msl : SIMDModifiedImmMoveMSL<1, 1, {1,1,0,?}, V128, "mvni", ".4s",
4098 [(set (v4i32 V128:$Rd),
4099 (AArch64mvni_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
4101 //----------------------------------------------------------------------------
4102 // AdvSIMD indexed element
4103 //----------------------------------------------------------------------------
4105 let neverHasSideEffects = 1 in {
4106 defm FMLA : SIMDFPIndexedSDTied<0, 0b0001, "fmla">;
4107 defm FMLS : SIMDFPIndexedSDTied<0, 0b0101, "fmls">;
4110 // NOTE: Operands are reordered in the FMLA/FMLS PatFrags because the
4111 // instruction expects the addend first, while the intrinsic expects it last.
4113 // On the other hand, there are quite a few valid combinatorial options due to
4114 // the commutativity of multiplication and the fact that (-x) * y = x * (-y).
4115 defm : SIMDFPIndexedSDTiedPatterns<"FMLA",
4116 TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)>>;
4117 defm : SIMDFPIndexedSDTiedPatterns<"FMLA",
4118 TriOpFrag<(fma node:$MHS, node:$RHS, node:$LHS)>>;
4120 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
4121 TriOpFrag<(fma node:$MHS, (fneg node:$RHS), node:$LHS)> >;
4122 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
4123 TriOpFrag<(fma node:$RHS, (fneg node:$MHS), node:$LHS)> >;
4124 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
4125 TriOpFrag<(fma (fneg node:$RHS), node:$MHS, node:$LHS)> >;
4126 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
4127 TriOpFrag<(fma (fneg node:$MHS), node:$RHS, node:$LHS)> >;
4129 multiclass FMLSIndexedAfterNegPatterns<SDPatternOperator OpNode> {
4130 // 3 variants for the .2s version: DUPLANE from 128-bit, DUPLANE from 64-bit
4132 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
4133 (AArch64duplane32 (v4f32 (fneg V128:$Rm)),
4134 VectorIndexS:$idx))),
4135 (FMLSv2i32_indexed V64:$Rd, V64:$Rn, V128:$Rm, VectorIndexS:$idx)>;
4136 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
4137 (v2f32 (AArch64duplane32
4138 (v4f32 (insert_subvector undef,
4139 (v2f32 (fneg V64:$Rm)),
4141 VectorIndexS:$idx)))),
4142 (FMLSv2i32_indexed V64:$Rd, V64:$Rn,
4143 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub),
4144 VectorIndexS:$idx)>;
4145 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
4146 (AArch64dup (f32 (fneg FPR32Op:$Rm))))),
4147 (FMLSv2i32_indexed V64:$Rd, V64:$Rn,
4148 (SUBREG_TO_REG (i32 0), FPR32Op:$Rm, ssub), (i64 0))>;
4150 // 3 variants for the .4s version: DUPLANE from 128-bit, DUPLANE from 64-bit
4152 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
4153 (AArch64duplane32 (v4f32 (fneg V128:$Rm)),
4154 VectorIndexS:$idx))),
4155 (FMLSv4i32_indexed V128:$Rd, V128:$Rn, V128:$Rm,
4156 VectorIndexS:$idx)>;
4157 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
4158 (v4f32 (AArch64duplane32
4159 (v4f32 (insert_subvector undef,
4160 (v2f32 (fneg V64:$Rm)),
4162 VectorIndexS:$idx)))),
4163 (FMLSv4i32_indexed V128:$Rd, V128:$Rn,
4164 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub),
4165 VectorIndexS:$idx)>;
4166 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
4167 (AArch64dup (f32 (fneg FPR32Op:$Rm))))),
4168 (FMLSv4i32_indexed V128:$Rd, V128:$Rn,
4169 (SUBREG_TO_REG (i32 0), FPR32Op:$Rm, ssub), (i64 0))>;
4171 // 2 variants for the .2d version: DUPLANE from 128-bit, and DUP scalar
4172 // (DUPLANE from 64-bit would be trivial).
4173 def : Pat<(v2f64 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn),
4174 (AArch64duplane64 (v2f64 (fneg V128:$Rm)),
4175 VectorIndexD:$idx))),
4177 V128:$Rd, V128:$Rn, V128:$Rm, VectorIndexS:$idx)>;
4178 def : Pat<(v2f64 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn),
4179 (AArch64dup (f64 (fneg FPR64Op:$Rm))))),
4180 (FMLSv2i64_indexed V128:$Rd, V128:$Rn,
4181 (SUBREG_TO_REG (i32 0), FPR64Op:$Rm, dsub), (i64 0))>;
4183 // 2 variants for 32-bit scalar version: extract from .2s or from .4s
4184 def : Pat<(f32 (OpNode (f32 FPR32:$Rd), (f32 FPR32:$Rn),
4185 (vector_extract (v4f32 (fneg V128:$Rm)),
4186 VectorIndexS:$idx))),
4187 (FMLSv1i32_indexed FPR32:$Rd, FPR32:$Rn,
4188 V128:$Rm, VectorIndexS:$idx)>;
4189 def : Pat<(f32 (OpNode (f32 FPR32:$Rd), (f32 FPR32:$Rn),
4190 (vector_extract (v2f32 (fneg V64:$Rm)),
4191 VectorIndexS:$idx))),
4192 (FMLSv1i32_indexed FPR32:$Rd, FPR32:$Rn,
4193 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub), VectorIndexS:$idx)>;
4195 // 1 variant for 64-bit scalar version: extract from .1d or from .2d
4196 def : Pat<(f64 (OpNode (f64 FPR64:$Rd), (f64 FPR64:$Rn),
4197 (vector_extract (v2f64 (fneg V128:$Rm)),
4198 VectorIndexS:$idx))),
4199 (FMLSv1i64_indexed FPR64:$Rd, FPR64:$Rn,
4200 V128:$Rm, VectorIndexS:$idx)>;
4203 defm : FMLSIndexedAfterNegPatterns<
4204 TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)> >;
4205 defm : FMLSIndexedAfterNegPatterns<
4206 TriOpFrag<(fma node:$MHS, node:$RHS, node:$LHS)> >;
4208 defm FMULX : SIMDFPIndexedSD<1, 0b1001, "fmulx", int_aarch64_neon_fmulx>;
4209 defm FMUL : SIMDFPIndexedSD<0, 0b1001, "fmul", fmul>;
4211 def : Pat<(v2f32 (fmul V64:$Rn, (AArch64dup (f32 FPR32:$Rm)))),
4212 (FMULv2i32_indexed V64:$Rn,
4213 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rm, ssub),
4215 def : Pat<(v4f32 (fmul V128:$Rn, (AArch64dup (f32 FPR32:$Rm)))),
4216 (FMULv4i32_indexed V128:$Rn,
4217 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rm, ssub),
4219 def : Pat<(v2f64 (fmul V128:$Rn, (AArch64dup (f64 FPR64:$Rm)))),
4220 (FMULv2i64_indexed V128:$Rn,
4221 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$Rm, dsub),
4224 defm SQDMULH : SIMDIndexedHS<0, 0b1100, "sqdmulh", int_aarch64_neon_sqdmulh>;
4225 defm SQRDMULH : SIMDIndexedHS<0, 0b1101, "sqrdmulh", int_aarch64_neon_sqrdmulh>;
4226 defm MLA : SIMDVectorIndexedHSTied<1, 0b0000, "mla",
4227 TriOpFrag<(add node:$LHS, (mul node:$MHS, node:$RHS))>>;
4228 defm MLS : SIMDVectorIndexedHSTied<1, 0b0100, "mls",
4229 TriOpFrag<(sub node:$LHS, (mul node:$MHS, node:$RHS))>>;
4230 defm MUL : SIMDVectorIndexedHS<0, 0b1000, "mul", mul>;
4231 defm SMLAL : SIMDVectorIndexedLongSDTied<0, 0b0010, "smlal",
4232 TriOpFrag<(add node:$LHS, (int_aarch64_neon_smull node:$MHS, node:$RHS))>>;
4233 defm SMLSL : SIMDVectorIndexedLongSDTied<0, 0b0110, "smlsl",
4234 TriOpFrag<(sub node:$LHS, (int_aarch64_neon_smull node:$MHS, node:$RHS))>>;
4235 defm SMULL : SIMDVectorIndexedLongSD<0, 0b1010, "smull",
4236 int_aarch64_neon_smull>;
4237 defm SQDMLAL : SIMDIndexedLongSQDMLXSDTied<0, 0b0011, "sqdmlal",
4238 int_aarch64_neon_sqadd>;
4239 defm SQDMLSL : SIMDIndexedLongSQDMLXSDTied<0, 0b0111, "sqdmlsl",
4240 int_aarch64_neon_sqsub>;
4241 defm SQDMULL : SIMDIndexedLongSD<0, 0b1011, "sqdmull", int_aarch64_neon_sqdmull>;
4242 defm UMLAL : SIMDVectorIndexedLongSDTied<1, 0b0010, "umlal",
4243 TriOpFrag<(add node:$LHS, (int_aarch64_neon_umull node:$MHS, node:$RHS))>>;
4244 defm UMLSL : SIMDVectorIndexedLongSDTied<1, 0b0110, "umlsl",
4245 TriOpFrag<(sub node:$LHS, (int_aarch64_neon_umull node:$MHS, node:$RHS))>>;
4246 defm UMULL : SIMDVectorIndexedLongSD<1, 0b1010, "umull",
4247 int_aarch64_neon_umull>;
4249 // A scalar sqdmull with the second operand being a vector lane can be
4250 // handled directly with the indexed instruction encoding.
4251 def : Pat<(int_aarch64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
4252 (vector_extract (v4i32 V128:$Vm),
4253 VectorIndexS:$idx)),
4254 (SQDMULLv1i64_indexed FPR32:$Rn, V128:$Vm, VectorIndexS:$idx)>;
4256 //----------------------------------------------------------------------------
4257 // AdvSIMD scalar shift instructions
4258 //----------------------------------------------------------------------------
4259 defm FCVTZS : SIMDScalarRShiftSD<0, 0b11111, "fcvtzs">;
4260 defm FCVTZU : SIMDScalarRShiftSD<1, 0b11111, "fcvtzu">;
4261 defm SCVTF : SIMDScalarRShiftSD<0, 0b11100, "scvtf">;
4262 defm UCVTF : SIMDScalarRShiftSD<1, 0b11100, "ucvtf">;
4263 // Codegen patterns for the above. We don't put these directly on the
4264 // instructions because TableGen's type inference can't handle the truth.
4265 // Having the same base pattern for fp <--> int totally freaks it out.
4266 def : Pat<(int_aarch64_neon_vcvtfp2fxs FPR32:$Rn, vecshiftR32:$imm),
4267 (FCVTZSs FPR32:$Rn, vecshiftR32:$imm)>;
4268 def : Pat<(int_aarch64_neon_vcvtfp2fxu FPR32:$Rn, vecshiftR32:$imm),
4269 (FCVTZUs FPR32:$Rn, vecshiftR32:$imm)>;
4270 def : Pat<(i64 (int_aarch64_neon_vcvtfp2fxs (f64 FPR64:$Rn), vecshiftR64:$imm)),
4271 (FCVTZSd FPR64:$Rn, vecshiftR64:$imm)>;
4272 def : Pat<(i64 (int_aarch64_neon_vcvtfp2fxu (f64 FPR64:$Rn), vecshiftR64:$imm)),
4273 (FCVTZUd FPR64:$Rn, vecshiftR64:$imm)>;
4274 def : Pat<(v1i64 (int_aarch64_neon_vcvtfp2fxs (v1f64 FPR64:$Rn),
4276 (FCVTZSd FPR64:$Rn, vecshiftR64:$imm)>;
4277 def : Pat<(v1i64 (int_aarch64_neon_vcvtfp2fxu (v1f64 FPR64:$Rn),
4279 (FCVTZUd FPR64:$Rn, vecshiftR64:$imm)>;
4280 def : Pat<(int_aarch64_neon_vcvtfxs2fp FPR32:$Rn, vecshiftR32:$imm),
4281 (SCVTFs FPR32:$Rn, vecshiftR32:$imm)>;
4282 def : Pat<(int_aarch64_neon_vcvtfxu2fp FPR32:$Rn, vecshiftR32:$imm),
4283 (UCVTFs FPR32:$Rn, vecshiftR32:$imm)>;
4284 def : Pat<(f64 (int_aarch64_neon_vcvtfxs2fp (i64 FPR64:$Rn), vecshiftR64:$imm)),
4285 (SCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
4286 def : Pat<(f64 (int_aarch64_neon_vcvtfxu2fp (i64 FPR64:$Rn), vecshiftR64:$imm)),
4287 (UCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
4288 def : Pat<(v1f64 (int_aarch64_neon_vcvtfxs2fp (v1i64 FPR64:$Rn),
4290 (SCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
4291 def : Pat<(v1f64 (int_aarch64_neon_vcvtfxu2fp (v1i64 FPR64:$Rn),
4293 (UCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
4295 defm SHL : SIMDScalarLShiftD< 0, 0b01010, "shl", AArch64vshl>;
4296 defm SLI : SIMDScalarLShiftDTied<1, 0b01010, "sli">;
4297 defm SQRSHRN : SIMDScalarRShiftBHS< 0, 0b10011, "sqrshrn",
4298 int_aarch64_neon_sqrshrn>;
4299 defm SQRSHRUN : SIMDScalarRShiftBHS< 1, 0b10001, "sqrshrun",
4300 int_aarch64_neon_sqrshrun>;
4301 defm SQSHLU : SIMDScalarLShiftBHSD<1, 0b01100, "sqshlu", AArch64sqshlui>;
4302 defm SQSHL : SIMDScalarLShiftBHSD<0, 0b01110, "sqshl", AArch64sqshli>;
4303 defm SQSHRN : SIMDScalarRShiftBHS< 0, 0b10010, "sqshrn",
4304 int_aarch64_neon_sqshrn>;
4305 defm SQSHRUN : SIMDScalarRShiftBHS< 1, 0b10000, "sqshrun",
4306 int_aarch64_neon_sqshrun>;
4307 defm SRI : SIMDScalarRShiftDTied< 1, 0b01000, "sri">;
4308 defm SRSHR : SIMDScalarRShiftD< 0, 0b00100, "srshr", AArch64srshri>;
4309 defm SRSRA : SIMDScalarRShiftDTied< 0, 0b00110, "srsra",
4310 TriOpFrag<(add node:$LHS,
4311 (AArch64srshri node:$MHS, node:$RHS))>>;
4312 defm SSHR : SIMDScalarRShiftD< 0, 0b00000, "sshr", AArch64vashr>;
4313 defm SSRA : SIMDScalarRShiftDTied< 0, 0b00010, "ssra",
4314 TriOpFrag<(add node:$LHS,
4315 (AArch64vashr node:$MHS, node:$RHS))>>;
4316 defm UQRSHRN : SIMDScalarRShiftBHS< 1, 0b10011, "uqrshrn",
4317 int_aarch64_neon_uqrshrn>;
4318 defm UQSHL : SIMDScalarLShiftBHSD<1, 0b01110, "uqshl", AArch64uqshli>;
4319 defm UQSHRN : SIMDScalarRShiftBHS< 1, 0b10010, "uqshrn",
4320 int_aarch64_neon_uqshrn>;
4321 defm URSHR : SIMDScalarRShiftD< 1, 0b00100, "urshr", AArch64urshri>;
4322 defm URSRA : SIMDScalarRShiftDTied< 1, 0b00110, "ursra",
4323 TriOpFrag<(add node:$LHS,
4324 (AArch64urshri node:$MHS, node:$RHS))>>;
4325 defm USHR : SIMDScalarRShiftD< 1, 0b00000, "ushr", AArch64vlshr>;
4326 defm USRA : SIMDScalarRShiftDTied< 1, 0b00010, "usra",
4327 TriOpFrag<(add node:$LHS,
4328 (AArch64vlshr node:$MHS, node:$RHS))>>;
4330 //----------------------------------------------------------------------------
4331 // AdvSIMD vector shift instructions
4332 //----------------------------------------------------------------------------
4333 defm FCVTZS:SIMDVectorRShiftSD<0, 0b11111, "fcvtzs", int_aarch64_neon_vcvtfp2fxs>;
4334 defm FCVTZU:SIMDVectorRShiftSD<1, 0b11111, "fcvtzu", int_aarch64_neon_vcvtfp2fxu>;
4335 defm SCVTF: SIMDVectorRShiftSDToFP<0, 0b11100, "scvtf",
4336 int_aarch64_neon_vcvtfxs2fp>;
4337 defm RSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10001, "rshrn",
4338 int_aarch64_neon_rshrn>;
4339 defm SHL : SIMDVectorLShiftBHSD<0, 0b01010, "shl", AArch64vshl>;
4340 defm SHRN : SIMDVectorRShiftNarrowBHS<0, 0b10000, "shrn",
4341 BinOpFrag<(trunc (AArch64vashr node:$LHS, node:$RHS))>>;
4342 defm SLI : SIMDVectorLShiftBHSDTied<1, 0b01010, "sli", int_aarch64_neon_vsli>;
4343 def : Pat<(v1i64 (int_aarch64_neon_vsli (v1i64 FPR64:$Rd), (v1i64 FPR64:$Rn),
4344 (i32 vecshiftL64:$imm))),
4345 (SLId FPR64:$Rd, FPR64:$Rn, vecshiftL64:$imm)>;
4346 defm SQRSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10011, "sqrshrn",
4347 int_aarch64_neon_sqrshrn>;
4348 defm SQRSHRUN: SIMDVectorRShiftNarrowBHS<1, 0b10001, "sqrshrun",
4349 int_aarch64_neon_sqrshrun>;
4350 defm SQSHLU : SIMDVectorLShiftBHSD<1, 0b01100, "sqshlu", AArch64sqshlui>;
4351 defm SQSHL : SIMDVectorLShiftBHSD<0, 0b01110, "sqshl", AArch64sqshli>;
4352 defm SQSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10010, "sqshrn",
4353 int_aarch64_neon_sqshrn>;
4354 defm SQSHRUN : SIMDVectorRShiftNarrowBHS<1, 0b10000, "sqshrun",
4355 int_aarch64_neon_sqshrun>;
4356 defm SRI : SIMDVectorRShiftBHSDTied<1, 0b01000, "sri", int_aarch64_neon_vsri>;
4357 def : Pat<(v1i64 (int_aarch64_neon_vsri (v1i64 FPR64:$Rd), (v1i64 FPR64:$Rn),
4358 (i32 vecshiftR64:$imm))),
4359 (SRId FPR64:$Rd, FPR64:$Rn, vecshiftR64:$imm)>;
4360 defm SRSHR : SIMDVectorRShiftBHSD<0, 0b00100, "srshr", AArch64srshri>;
4361 defm SRSRA : SIMDVectorRShiftBHSDTied<0, 0b00110, "srsra",
4362 TriOpFrag<(add node:$LHS,
4363 (AArch64srshri node:$MHS, node:$RHS))> >;
4364 defm SSHLL : SIMDVectorLShiftLongBHSD<0, 0b10100, "sshll",
4365 BinOpFrag<(AArch64vshl (sext node:$LHS), node:$RHS)>>;
4367 defm SSHR : SIMDVectorRShiftBHSD<0, 0b00000, "sshr", AArch64vashr>;
4368 defm SSRA : SIMDVectorRShiftBHSDTied<0, 0b00010, "ssra",
4369 TriOpFrag<(add node:$LHS, (AArch64vashr node:$MHS, node:$RHS))>>;
4370 defm UCVTF : SIMDVectorRShiftSDToFP<1, 0b11100, "ucvtf",
4371 int_aarch64_neon_vcvtfxu2fp>;
4372 defm UQRSHRN : SIMDVectorRShiftNarrowBHS<1, 0b10011, "uqrshrn",
4373 int_aarch64_neon_uqrshrn>;
4374 defm UQSHL : SIMDVectorLShiftBHSD<1, 0b01110, "uqshl", AArch64uqshli>;
4375 defm UQSHRN : SIMDVectorRShiftNarrowBHS<1, 0b10010, "uqshrn",
4376 int_aarch64_neon_uqshrn>;
4377 defm URSHR : SIMDVectorRShiftBHSD<1, 0b00100, "urshr", AArch64urshri>;
4378 defm URSRA : SIMDVectorRShiftBHSDTied<1, 0b00110, "ursra",
4379 TriOpFrag<(add node:$LHS,
4380 (AArch64urshri node:$MHS, node:$RHS))> >;
4381 defm USHLL : SIMDVectorLShiftLongBHSD<1, 0b10100, "ushll",
4382 BinOpFrag<(AArch64vshl (zext node:$LHS), node:$RHS)>>;
4383 defm USHR : SIMDVectorRShiftBHSD<1, 0b00000, "ushr", AArch64vlshr>;
4384 defm USRA : SIMDVectorRShiftBHSDTied<1, 0b00010, "usra",
4385 TriOpFrag<(add node:$LHS, (AArch64vlshr node:$MHS, node:$RHS))> >;
4387 // SHRN patterns for when a logical right shift was used instead of arithmetic
4388 // (the immediate guarantees no sign bits actually end up in the result so it
4390 def : Pat<(v8i8 (trunc (AArch64vlshr (v8i16 V128:$Rn), vecshiftR16Narrow:$imm))),
4391 (SHRNv8i8_shift V128:$Rn, vecshiftR16Narrow:$imm)>;
4392 def : Pat<(v4i16 (trunc (AArch64vlshr (v4i32 V128:$Rn), vecshiftR32Narrow:$imm))),
4393 (SHRNv4i16_shift V128:$Rn, vecshiftR32Narrow:$imm)>;
4394 def : Pat<(v2i32 (trunc (AArch64vlshr (v2i64 V128:$Rn), vecshiftR64Narrow:$imm))),
4395 (SHRNv2i32_shift V128:$Rn, vecshiftR64Narrow:$imm)>;
4397 def : Pat<(v16i8 (concat_vectors (v8i8 V64:$Rd),
4398 (trunc (AArch64vlshr (v8i16 V128:$Rn),
4399 vecshiftR16Narrow:$imm)))),
4400 (SHRNv16i8_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
4401 V128:$Rn, vecshiftR16Narrow:$imm)>;
4402 def : Pat<(v8i16 (concat_vectors (v4i16 V64:$Rd),
4403 (trunc (AArch64vlshr (v4i32 V128:$Rn),
4404 vecshiftR32Narrow:$imm)))),
4405 (SHRNv8i16_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
4406 V128:$Rn, vecshiftR32Narrow:$imm)>;
4407 def : Pat<(v4i32 (concat_vectors (v2i32 V64:$Rd),
4408 (trunc (AArch64vlshr (v2i64 V128:$Rn),
4409 vecshiftR64Narrow:$imm)))),
4410 (SHRNv4i32_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
4411 V128:$Rn, vecshiftR32Narrow:$imm)>;
4413 // Vector sign and zero extensions are implemented with SSHLL and USSHLL.
4414 // Anyexts are implemented as zexts.
4415 def : Pat<(v8i16 (sext (v8i8 V64:$Rn))), (SSHLLv8i8_shift V64:$Rn, (i32 0))>;
4416 def : Pat<(v8i16 (zext (v8i8 V64:$Rn))), (USHLLv8i8_shift V64:$Rn, (i32 0))>;
4417 def : Pat<(v8i16 (anyext (v8i8 V64:$Rn))), (USHLLv8i8_shift V64:$Rn, (i32 0))>;
4418 def : Pat<(v4i32 (sext (v4i16 V64:$Rn))), (SSHLLv4i16_shift V64:$Rn, (i32 0))>;
4419 def : Pat<(v4i32 (zext (v4i16 V64:$Rn))), (USHLLv4i16_shift V64:$Rn, (i32 0))>;
4420 def : Pat<(v4i32 (anyext (v4i16 V64:$Rn))), (USHLLv4i16_shift V64:$Rn, (i32 0))>;
4421 def : Pat<(v2i64 (sext (v2i32 V64:$Rn))), (SSHLLv2i32_shift V64:$Rn, (i32 0))>;
4422 def : Pat<(v2i64 (zext (v2i32 V64:$Rn))), (USHLLv2i32_shift V64:$Rn, (i32 0))>;
4423 def : Pat<(v2i64 (anyext (v2i32 V64:$Rn))), (USHLLv2i32_shift V64:$Rn, (i32 0))>;
4424 // Also match an extend from the upper half of a 128 bit source register.
4425 def : Pat<(v8i16 (anyext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
4426 (USHLLv16i8_shift V128:$Rn, (i32 0))>;
4427 def : Pat<(v8i16 (zext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
4428 (USHLLv16i8_shift V128:$Rn, (i32 0))>;
4429 def : Pat<(v8i16 (sext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
4430 (SSHLLv16i8_shift V128:$Rn, (i32 0))>;
4431 def : Pat<(v4i32 (anyext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
4432 (USHLLv8i16_shift V128:$Rn, (i32 0))>;
4433 def : Pat<(v4i32 (zext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
4434 (USHLLv8i16_shift V128:$Rn, (i32 0))>;
4435 def : Pat<(v4i32 (sext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
4436 (SSHLLv8i16_shift V128:$Rn, (i32 0))>;
4437 def : Pat<(v2i64 (anyext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
4438 (USHLLv4i32_shift V128:$Rn, (i32 0))>;
4439 def : Pat<(v2i64 (zext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
4440 (USHLLv4i32_shift V128:$Rn, (i32 0))>;
4441 def : Pat<(v2i64 (sext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
4442 (SSHLLv4i32_shift V128:$Rn, (i32 0))>;
4444 // Vector shift sxtl aliases
4445 def : InstAlias<"sxtl.8h $dst, $src1",
4446 (SSHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
4447 def : InstAlias<"sxtl $dst.8h, $src1.8b",
4448 (SSHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
4449 def : InstAlias<"sxtl.4s $dst, $src1",
4450 (SSHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
4451 def : InstAlias<"sxtl $dst.4s, $src1.4h",
4452 (SSHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
4453 def : InstAlias<"sxtl.2d $dst, $src1",
4454 (SSHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
4455 def : InstAlias<"sxtl $dst.2d, $src1.2s",
4456 (SSHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
4458 // Vector shift sxtl2 aliases
4459 def : InstAlias<"sxtl2.8h $dst, $src1",
4460 (SSHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
4461 def : InstAlias<"sxtl2 $dst.8h, $src1.16b",
4462 (SSHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
4463 def : InstAlias<"sxtl2.4s $dst, $src1",
4464 (SSHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
4465 def : InstAlias<"sxtl2 $dst.4s, $src1.8h",
4466 (SSHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
4467 def : InstAlias<"sxtl2.2d $dst, $src1",
4468 (SSHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
4469 def : InstAlias<"sxtl2 $dst.2d, $src1.4s",
4470 (SSHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
4472 // Vector shift uxtl aliases
4473 def : InstAlias<"uxtl.8h $dst, $src1",
4474 (USHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
4475 def : InstAlias<"uxtl $dst.8h, $src1.8b",
4476 (USHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
4477 def : InstAlias<"uxtl.4s $dst, $src1",
4478 (USHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
4479 def : InstAlias<"uxtl $dst.4s, $src1.4h",
4480 (USHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
4481 def : InstAlias<"uxtl.2d $dst, $src1",
4482 (USHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
4483 def : InstAlias<"uxtl $dst.2d, $src1.2s",
4484 (USHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
4486 // Vector shift uxtl2 aliases
4487 def : InstAlias<"uxtl2.8h $dst, $src1",
4488 (USHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
4489 def : InstAlias<"uxtl2 $dst.8h, $src1.16b",
4490 (USHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
4491 def : InstAlias<"uxtl2.4s $dst, $src1",
4492 (USHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
4493 def : InstAlias<"uxtl2 $dst.4s, $src1.8h",
4494 (USHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
4495 def : InstAlias<"uxtl2.2d $dst, $src1",
4496 (USHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
4497 def : InstAlias<"uxtl2 $dst.2d, $src1.4s",
4498 (USHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
4500 // If an integer is about to be converted to a floating point value,
4501 // just load it on the floating point unit.
4502 // These patterns are more complex because floating point loads do not
4503 // support sign extension.
4504 // The sign extension has to be explicitly added and is only supported for
4505 // one step: byte-to-half, half-to-word, word-to-doubleword.
4506 // SCVTF GPR -> FPR is 9 cycles.
4507 // SCVTF FPR -> FPR is 4 cyclces.
4508 // (sign extension with lengthen) SXTL FPR -> FPR is 2 cycles.
4509 // Therefore, we can do 2 sign extensions and one SCVTF FPR -> FPR
4510 // and still being faster.
4511 // However, this is not good for code size.
4512 // 8-bits -> float. 2 sizes step-up.
4513 class SExtLoadi8CVTf32Pat<dag addrmode, dag INST>
4514 : Pat<(f32 (sint_to_fp (i32 (sextloadi8 addrmode)))),
4515 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
4520 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4526 ssub)))>, Requires<[NotForCodeSize, IsCyclone]>;
4528 def : SExtLoadi8CVTf32Pat<(ro8.Wpat GPR64sp:$Rn, GPR32:$Rm, ro8.Wext:$ext),
4529 (LDRBroW GPR64sp:$Rn, GPR32:$Rm, ro8.Wext:$ext)>;
4530 def : SExtLoadi8CVTf32Pat<(ro8.Xpat GPR64sp:$Rn, GPR64:$Rm, ro8.Xext:$ext),
4531 (LDRBroX GPR64sp:$Rn, GPR64:$Rm, ro8.Xext:$ext)>;
4532 def : SExtLoadi8CVTf32Pat<(am_indexed8 GPR64sp:$Rn, uimm12s1:$offset),
4533 (LDRBui GPR64sp:$Rn, uimm12s1:$offset)>;
4534 def : SExtLoadi8CVTf32Pat<(am_unscaled8 GPR64sp:$Rn, simm9:$offset),
4535 (LDURBi GPR64sp:$Rn, simm9:$offset)>;
4537 // 16-bits -> float. 1 size step-up.
4538 class SExtLoadi16CVTf32Pat<dag addrmode, dag INST>
4539 : Pat<(f32 (sint_to_fp (i32 (sextloadi16 addrmode)))),
4540 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
4542 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4546 ssub)))>, Requires<[NotForCodeSize]>;
4548 def : SExtLoadi16CVTf32Pat<(ro16.Wpat GPR64sp:$Rn, GPR32:$Rm, ro16.Wext:$ext),
4549 (LDRHroW GPR64sp:$Rn, GPR32:$Rm, ro16.Wext:$ext)>;
4550 def : SExtLoadi16CVTf32Pat<(ro16.Xpat GPR64sp:$Rn, GPR64:$Rm, ro16.Xext:$ext),
4551 (LDRHroX GPR64sp:$Rn, GPR64:$Rm, ro16.Xext:$ext)>;
4552 def : SExtLoadi16CVTf32Pat<(am_indexed16 GPR64sp:$Rn, uimm12s2:$offset),
4553 (LDRHui GPR64sp:$Rn, uimm12s2:$offset)>;
4554 def : SExtLoadi16CVTf32Pat<(am_unscaled16 GPR64sp:$Rn, simm9:$offset),
4555 (LDURHi GPR64sp:$Rn, simm9:$offset)>;
4557 // 32-bits to 32-bits are handled in target specific dag combine:
4558 // performIntToFpCombine.
4559 // 64-bits integer to 32-bits floating point, not possible with
4560 // SCVTF on floating point registers (both source and destination
4561 // must have the same size).
4563 // Here are the patterns for 8, 16, 32, and 64-bits to double.
4564 // 8-bits -> double. 3 size step-up: give up.
4565 // 16-bits -> double. 2 size step.
4566 class SExtLoadi16CVTf64Pat<dag addrmode, dag INST>
4567 : Pat <(f64 (sint_to_fp (i32 (sextloadi16 addrmode)))),
4568 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4573 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4579 dsub)))>, Requires<[NotForCodeSize, IsCyclone]>;
4581 def : SExtLoadi16CVTf64Pat<(ro16.Wpat GPR64sp:$Rn, GPR32:$Rm, ro16.Wext:$ext),
4582 (LDRHroW GPR64sp:$Rn, GPR32:$Rm, ro16.Wext:$ext)>;
4583 def : SExtLoadi16CVTf64Pat<(ro16.Xpat GPR64sp:$Rn, GPR64:$Rm, ro16.Xext:$ext),
4584 (LDRHroX GPR64sp:$Rn, GPR64:$Rm, ro16.Xext:$ext)>;
4585 def : SExtLoadi16CVTf64Pat<(am_indexed16 GPR64sp:$Rn, uimm12s2:$offset),
4586 (LDRHui GPR64sp:$Rn, uimm12s2:$offset)>;
4587 def : SExtLoadi16CVTf64Pat<(am_unscaled16 GPR64sp:$Rn, simm9:$offset),
4588 (LDURHi GPR64sp:$Rn, simm9:$offset)>;
4589 // 32-bits -> double. 1 size step-up.
4590 class SExtLoadi32CVTf64Pat<dag addrmode, dag INST>
4591 : Pat <(f64 (sint_to_fp (i32 (load addrmode)))),
4592 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4594 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4598 dsub)))>, Requires<[NotForCodeSize]>;
4600 def : SExtLoadi32CVTf64Pat<(ro32.Wpat GPR64sp:$Rn, GPR32:$Rm, ro32.Wext:$ext),
4601 (LDRSroW GPR64sp:$Rn, GPR32:$Rm, ro32.Wext:$ext)>;
4602 def : SExtLoadi32CVTf64Pat<(ro32.Xpat GPR64sp:$Rn, GPR64:$Rm, ro32.Xext:$ext),
4603 (LDRSroX GPR64sp:$Rn, GPR64:$Rm, ro32.Xext:$ext)>;
4604 def : SExtLoadi32CVTf64Pat<(am_indexed32 GPR64sp:$Rn, uimm12s4:$offset),
4605 (LDRSui GPR64sp:$Rn, uimm12s4:$offset)>;
4606 def : SExtLoadi32CVTf64Pat<(am_unscaled32 GPR64sp:$Rn, simm9:$offset),
4607 (LDURSi GPR64sp:$Rn, simm9:$offset)>;
4609 // 64-bits -> double are handled in target specific dag combine:
4610 // performIntToFpCombine.
4613 //----------------------------------------------------------------------------
4614 // AdvSIMD Load-Store Structure
4615 //----------------------------------------------------------------------------
4616 defm LD1 : SIMDLd1Multiple<"ld1">;
4617 defm LD2 : SIMDLd2Multiple<"ld2">;
4618 defm LD3 : SIMDLd3Multiple<"ld3">;
4619 defm LD4 : SIMDLd4Multiple<"ld4">;
4621 defm ST1 : SIMDSt1Multiple<"st1">;
4622 defm ST2 : SIMDSt2Multiple<"st2">;
4623 defm ST3 : SIMDSt3Multiple<"st3">;
4624 defm ST4 : SIMDSt4Multiple<"st4">;
4626 class Ld1Pat<ValueType ty, Instruction INST>
4627 : Pat<(ty (load GPR64sp:$Rn)), (INST GPR64sp:$Rn)>;
4629 def : Ld1Pat<v16i8, LD1Onev16b>;
4630 def : Ld1Pat<v8i16, LD1Onev8h>;
4631 def : Ld1Pat<v4i32, LD1Onev4s>;
4632 def : Ld1Pat<v2i64, LD1Onev2d>;
4633 def : Ld1Pat<v8i8, LD1Onev8b>;
4634 def : Ld1Pat<v4i16, LD1Onev4h>;
4635 def : Ld1Pat<v2i32, LD1Onev2s>;
4636 def : Ld1Pat<v1i64, LD1Onev1d>;
4638 class St1Pat<ValueType ty, Instruction INST>
4639 : Pat<(store ty:$Vt, GPR64sp:$Rn),
4640 (INST ty:$Vt, GPR64sp:$Rn)>;
4642 def : St1Pat<v16i8, ST1Onev16b>;
4643 def : St1Pat<v8i16, ST1Onev8h>;
4644 def : St1Pat<v4i32, ST1Onev4s>;
4645 def : St1Pat<v2i64, ST1Onev2d>;
4646 def : St1Pat<v8i8, ST1Onev8b>;
4647 def : St1Pat<v4i16, ST1Onev4h>;
4648 def : St1Pat<v2i32, ST1Onev2s>;
4649 def : St1Pat<v1i64, ST1Onev1d>;
4655 defm LD1R : SIMDLdR<0, 0b110, 0, "ld1r", "One", 1, 2, 4, 8>;
4656 defm LD2R : SIMDLdR<1, 0b110, 0, "ld2r", "Two", 2, 4, 8, 16>;
4657 defm LD3R : SIMDLdR<0, 0b111, 0, "ld3r", "Three", 3, 6, 12, 24>;
4658 defm LD4R : SIMDLdR<1, 0b111, 0, "ld4r", "Four", 4, 8, 16, 32>;
4659 let mayLoad = 1, neverHasSideEffects = 1 in {
4660 defm LD1 : SIMDLdSingleBTied<0, 0b000, "ld1", VecListOneb, GPR64pi1>;
4661 defm LD1 : SIMDLdSingleHTied<0, 0b010, 0, "ld1", VecListOneh, GPR64pi2>;
4662 defm LD1 : SIMDLdSingleSTied<0, 0b100, 0b00, "ld1", VecListOnes, GPR64pi4>;
4663 defm LD1 : SIMDLdSingleDTied<0, 0b100, 0b01, "ld1", VecListOned, GPR64pi8>;
4664 defm LD2 : SIMDLdSingleBTied<1, 0b000, "ld2", VecListTwob, GPR64pi2>;
4665 defm LD2 : SIMDLdSingleHTied<1, 0b010, 0, "ld2", VecListTwoh, GPR64pi4>;
4666 defm LD2 : SIMDLdSingleSTied<1, 0b100, 0b00, "ld2", VecListTwos, GPR64pi8>;
4667 defm LD2 : SIMDLdSingleDTied<1, 0b100, 0b01, "ld2", VecListTwod, GPR64pi16>;
4668 defm LD3 : SIMDLdSingleBTied<0, 0b001, "ld3", VecListThreeb, GPR64pi3>;
4669 defm LD3 : SIMDLdSingleHTied<0, 0b011, 0, "ld3", VecListThreeh, GPR64pi6>;
4670 defm LD3 : SIMDLdSingleSTied<0, 0b101, 0b00, "ld3", VecListThrees, GPR64pi12>;
4671 defm LD3 : SIMDLdSingleDTied<0, 0b101, 0b01, "ld3", VecListThreed, GPR64pi24>;
4672 defm LD4 : SIMDLdSingleBTied<1, 0b001, "ld4", VecListFourb, GPR64pi4>;
4673 defm LD4 : SIMDLdSingleHTied<1, 0b011, 0, "ld4", VecListFourh, GPR64pi8>;
4674 defm LD4 : SIMDLdSingleSTied<1, 0b101, 0b00, "ld4", VecListFours, GPR64pi16>;
4675 defm LD4 : SIMDLdSingleDTied<1, 0b101, 0b01, "ld4", VecListFourd, GPR64pi32>;
4678 def : Pat<(v8i8 (AArch64dup (i32 (extloadi8 GPR64sp:$Rn)))),
4679 (LD1Rv8b GPR64sp:$Rn)>;
4680 def : Pat<(v16i8 (AArch64dup (i32 (extloadi8 GPR64sp:$Rn)))),
4681 (LD1Rv16b GPR64sp:$Rn)>;
4682 def : Pat<(v4i16 (AArch64dup (i32 (extloadi16 GPR64sp:$Rn)))),
4683 (LD1Rv4h GPR64sp:$Rn)>;
4684 def : Pat<(v8i16 (AArch64dup (i32 (extloadi16 GPR64sp:$Rn)))),
4685 (LD1Rv8h GPR64sp:$Rn)>;
4686 def : Pat<(v2i32 (AArch64dup (i32 (load GPR64sp:$Rn)))),
4687 (LD1Rv2s GPR64sp:$Rn)>;
4688 def : Pat<(v4i32 (AArch64dup (i32 (load GPR64sp:$Rn)))),
4689 (LD1Rv4s GPR64sp:$Rn)>;
4690 def : Pat<(v2i64 (AArch64dup (i64 (load GPR64sp:$Rn)))),
4691 (LD1Rv2d GPR64sp:$Rn)>;
4692 def : Pat<(v1i64 (AArch64dup (i64 (load GPR64sp:$Rn)))),
4693 (LD1Rv1d GPR64sp:$Rn)>;
4694 // Grab the floating point version too
4695 def : Pat<(v2f32 (AArch64dup (f32 (load GPR64sp:$Rn)))),
4696 (LD1Rv2s GPR64sp:$Rn)>;
4697 def : Pat<(v4f32 (AArch64dup (f32 (load GPR64sp:$Rn)))),
4698 (LD1Rv4s GPR64sp:$Rn)>;
4699 def : Pat<(v2f64 (AArch64dup (f64 (load GPR64sp:$Rn)))),
4700 (LD1Rv2d GPR64sp:$Rn)>;
4701 def : Pat<(v1f64 (AArch64dup (f64 (load GPR64sp:$Rn)))),
4702 (LD1Rv1d GPR64sp:$Rn)>;
4703 def : Pat<(v4f16 (AArch64dup (f16 (load GPR64sp:$Rn)))),
4704 (LD1Rv4h GPR64sp:$Rn)>;
4705 def : Pat<(v8f16 (AArch64dup (f16 (load GPR64sp:$Rn)))),
4706 (LD1Rv8h GPR64sp:$Rn)>;
4708 class Ld1Lane128Pat<SDPatternOperator scalar_load, Operand VecIndex,
4709 ValueType VTy, ValueType STy, Instruction LD1>
4710 : Pat<(vector_insert (VTy VecListOne128:$Rd),
4711 (STy (scalar_load GPR64sp:$Rn)), VecIndex:$idx),
4712 (LD1 VecListOne128:$Rd, VecIndex:$idx, GPR64sp:$Rn)>;
4714 def : Ld1Lane128Pat<extloadi8, VectorIndexB, v16i8, i32, LD1i8>;
4715 def : Ld1Lane128Pat<extloadi16, VectorIndexH, v8i16, i32, LD1i16>;
4716 def : Ld1Lane128Pat<load, VectorIndexS, v4i32, i32, LD1i32>;
4717 def : Ld1Lane128Pat<load, VectorIndexS, v4f32, f32, LD1i32>;
4718 def : Ld1Lane128Pat<load, VectorIndexD, v2i64, i64, LD1i64>;
4719 def : Ld1Lane128Pat<load, VectorIndexD, v2f64, f64, LD1i64>;
4720 def : Ld1Lane128Pat<load, VectorIndexH, v8f16, f16, LD1i16>;
4722 class Ld1Lane64Pat<SDPatternOperator scalar_load, Operand VecIndex,
4723 ValueType VTy, ValueType STy, Instruction LD1>
4724 : Pat<(vector_insert (VTy VecListOne64:$Rd),
4725 (STy (scalar_load GPR64sp:$Rn)), VecIndex:$idx),
4727 (LD1 (SUBREG_TO_REG (i32 0), VecListOne64:$Rd, dsub),
4728 VecIndex:$idx, GPR64sp:$Rn),
4731 def : Ld1Lane64Pat<extloadi8, VectorIndexB, v8i8, i32, LD1i8>;
4732 def : Ld1Lane64Pat<extloadi16, VectorIndexH, v4i16, i32, LD1i16>;
4733 def : Ld1Lane64Pat<load, VectorIndexS, v2i32, i32, LD1i32>;
4734 def : Ld1Lane64Pat<load, VectorIndexS, v2f32, f32, LD1i32>;
4735 def : Ld1Lane64Pat<load, VectorIndexH, v4f16, f16, LD1i16>;
4738 defm LD1 : SIMDLdSt1SingleAliases<"ld1">;
4739 defm LD2 : SIMDLdSt2SingleAliases<"ld2">;
4740 defm LD3 : SIMDLdSt3SingleAliases<"ld3">;
4741 defm LD4 : SIMDLdSt4SingleAliases<"ld4">;
4744 defm ST1 : SIMDStSingleB<0, 0b000, "st1", VecListOneb, GPR64pi1>;
4745 defm ST1 : SIMDStSingleH<0, 0b010, 0, "st1", VecListOneh, GPR64pi2>;
4746 defm ST1 : SIMDStSingleS<0, 0b100, 0b00, "st1", VecListOnes, GPR64pi4>;
4747 defm ST1 : SIMDStSingleD<0, 0b100, 0b01, "st1", VecListOned, GPR64pi8>;
4749 let AddedComplexity = 15 in
4750 class St1Lane128Pat<SDPatternOperator scalar_store, Operand VecIndex,
4751 ValueType VTy, ValueType STy, Instruction ST1>
4753 (STy (vector_extract (VTy VecListOne128:$Vt), VecIndex:$idx)),
4755 (ST1 VecListOne128:$Vt, VecIndex:$idx, GPR64sp:$Rn)>;
4757 def : St1Lane128Pat<truncstorei8, VectorIndexB, v16i8, i32, ST1i8>;
4758 def : St1Lane128Pat<truncstorei16, VectorIndexH, v8i16, i32, ST1i16>;
4759 def : St1Lane128Pat<store, VectorIndexS, v4i32, i32, ST1i32>;
4760 def : St1Lane128Pat<store, VectorIndexS, v4f32, f32, ST1i32>;
4761 def : St1Lane128Pat<store, VectorIndexD, v2i64, i64, ST1i64>;
4762 def : St1Lane128Pat<store, VectorIndexD, v2f64, f64, ST1i64>;
4763 def : St1Lane128Pat<store, VectorIndexH, v8f16, f16, ST1i16>;
4765 let AddedComplexity = 15 in
4766 class St1Lane64Pat<SDPatternOperator scalar_store, Operand VecIndex,
4767 ValueType VTy, ValueType STy, Instruction ST1>
4769 (STy (vector_extract (VTy VecListOne64:$Vt), VecIndex:$idx)),
4771 (ST1 (SUBREG_TO_REG (i32 0), VecListOne64:$Vt, dsub),
4772 VecIndex:$idx, GPR64sp:$Rn)>;
4774 def : St1Lane64Pat<truncstorei8, VectorIndexB, v8i8, i32, ST1i8>;
4775 def : St1Lane64Pat<truncstorei16, VectorIndexH, v4i16, i32, ST1i16>;
4776 def : St1Lane64Pat<store, VectorIndexS, v2i32, i32, ST1i32>;
4777 def : St1Lane64Pat<store, VectorIndexS, v2f32, f32, ST1i32>;
4778 def : St1Lane64Pat<store, VectorIndexH, v4f16, f16, ST1i16>;
4780 multiclass St1LanePost64Pat<SDPatternOperator scalar_store, Operand VecIndex,
4781 ValueType VTy, ValueType STy, Instruction ST1,
4783 def : Pat<(scalar_store
4784 (STy (vector_extract (VTy VecListOne64:$Vt), VecIndex:$idx)),
4785 GPR64sp:$Rn, offset),
4786 (ST1 (SUBREG_TO_REG (i32 0), VecListOne64:$Vt, dsub),
4787 VecIndex:$idx, GPR64sp:$Rn, XZR)>;
4789 def : Pat<(scalar_store
4790 (STy (vector_extract (VTy VecListOne64:$Vt), VecIndex:$idx)),
4791 GPR64sp:$Rn, GPR64:$Rm),
4792 (ST1 (SUBREG_TO_REG (i32 0), VecListOne64:$Vt, dsub),
4793 VecIndex:$idx, GPR64sp:$Rn, $Rm)>;
4796 defm : St1LanePost64Pat<post_truncsti8, VectorIndexB, v8i8, i32, ST1i8_POST, 1>;
4797 defm : St1LanePost64Pat<post_truncsti16, VectorIndexH, v4i16, i32, ST1i16_POST,
4799 defm : St1LanePost64Pat<post_store, VectorIndexS, v2i32, i32, ST1i32_POST, 4>;
4800 defm : St1LanePost64Pat<post_store, VectorIndexS, v2f32, f32, ST1i32_POST, 4>;
4801 defm : St1LanePost64Pat<post_store, VectorIndexD, v1i64, i64, ST1i64_POST, 8>;
4802 defm : St1LanePost64Pat<post_store, VectorIndexD, v1f64, f64, ST1i64_POST, 8>;
4803 defm : St1LanePost64Pat<post_store, VectorIndexH, v4f16, f16, ST1i16_POST, 2>;
4805 multiclass St1LanePost128Pat<SDPatternOperator scalar_store, Operand VecIndex,
4806 ValueType VTy, ValueType STy, Instruction ST1,
4808 def : Pat<(scalar_store
4809 (STy (vector_extract (VTy VecListOne128:$Vt), VecIndex:$idx)),
4810 GPR64sp:$Rn, offset),
4811 (ST1 VecListOne128:$Vt, VecIndex:$idx, GPR64sp:$Rn, XZR)>;
4813 def : Pat<(scalar_store
4814 (STy (vector_extract (VTy VecListOne128:$Vt), VecIndex:$idx)),
4815 GPR64sp:$Rn, GPR64:$Rm),
4816 (ST1 VecListOne128:$Vt, VecIndex:$idx, GPR64sp:$Rn, $Rm)>;
4819 defm : St1LanePost128Pat<post_truncsti8, VectorIndexB, v16i8, i32, ST1i8_POST,
4821 defm : St1LanePost128Pat<post_truncsti16, VectorIndexH, v8i16, i32, ST1i16_POST,
4823 defm : St1LanePost128Pat<post_store, VectorIndexS, v4i32, i32, ST1i32_POST, 4>;
4824 defm : St1LanePost128Pat<post_store, VectorIndexS, v4f32, f32, ST1i32_POST, 4>;
4825 defm : St1LanePost128Pat<post_store, VectorIndexD, v2i64, i64, ST1i64_POST, 8>;
4826 defm : St1LanePost128Pat<post_store, VectorIndexD, v2f64, f64, ST1i64_POST, 8>;
4827 defm : St1LanePost128Pat<post_store, VectorIndexH, v8f16, f16, ST1i16_POST, 2>;
4829 let mayStore = 1, neverHasSideEffects = 1 in {
4830 defm ST2 : SIMDStSingleB<1, 0b000, "st2", VecListTwob, GPR64pi2>;
4831 defm ST2 : SIMDStSingleH<1, 0b010, 0, "st2", VecListTwoh, GPR64pi4>;
4832 defm ST2 : SIMDStSingleS<1, 0b100, 0b00, "st2", VecListTwos, GPR64pi8>;
4833 defm ST2 : SIMDStSingleD<1, 0b100, 0b01, "st2", VecListTwod, GPR64pi16>;
4834 defm ST3 : SIMDStSingleB<0, 0b001, "st3", VecListThreeb, GPR64pi3>;
4835 defm ST3 : SIMDStSingleH<0, 0b011, 0, "st3", VecListThreeh, GPR64pi6>;
4836 defm ST3 : SIMDStSingleS<0, 0b101, 0b00, "st3", VecListThrees, GPR64pi12>;
4837 defm ST3 : SIMDStSingleD<0, 0b101, 0b01, "st3", VecListThreed, GPR64pi24>;
4838 defm ST4 : SIMDStSingleB<1, 0b001, "st4", VecListFourb, GPR64pi4>;
4839 defm ST4 : SIMDStSingleH<1, 0b011, 0, "st4", VecListFourh, GPR64pi8>;
4840 defm ST4 : SIMDStSingleS<1, 0b101, 0b00, "st4", VecListFours, GPR64pi16>;
4841 defm ST4 : SIMDStSingleD<1, 0b101, 0b01, "st4", VecListFourd, GPR64pi32>;
4844 defm ST1 : SIMDLdSt1SingleAliases<"st1">;
4845 defm ST2 : SIMDLdSt2SingleAliases<"st2">;
4846 defm ST3 : SIMDLdSt3SingleAliases<"st3">;
4847 defm ST4 : SIMDLdSt4SingleAliases<"st4">;
4849 //----------------------------------------------------------------------------
4850 // Crypto extensions
4851 //----------------------------------------------------------------------------
4853 def AESErr : AESTiedInst<0b0100, "aese", int_aarch64_crypto_aese>;
4854 def AESDrr : AESTiedInst<0b0101, "aesd", int_aarch64_crypto_aesd>;
4855 def AESMCrr : AESInst< 0b0110, "aesmc", int_aarch64_crypto_aesmc>;
4856 def AESIMCrr : AESInst< 0b0111, "aesimc", int_aarch64_crypto_aesimc>;
4858 def SHA1Crrr : SHATiedInstQSV<0b000, "sha1c", int_aarch64_crypto_sha1c>;
4859 def SHA1Prrr : SHATiedInstQSV<0b001, "sha1p", int_aarch64_crypto_sha1p>;
4860 def SHA1Mrrr : SHATiedInstQSV<0b010, "sha1m", int_aarch64_crypto_sha1m>;
4861 def SHA1SU0rrr : SHATiedInstVVV<0b011, "sha1su0", int_aarch64_crypto_sha1su0>;
4862 def SHA256Hrrr : SHATiedInstQQV<0b100, "sha256h", int_aarch64_crypto_sha256h>;
4863 def SHA256H2rrr : SHATiedInstQQV<0b101, "sha256h2",int_aarch64_crypto_sha256h2>;
4864 def SHA256SU1rrr :SHATiedInstVVV<0b110, "sha256su1",int_aarch64_crypto_sha256su1>;
4866 def SHA1Hrr : SHAInstSS< 0b0000, "sha1h", int_aarch64_crypto_sha1h>;
4867 def SHA1SU1rr : SHATiedInstVV<0b0001, "sha1su1", int_aarch64_crypto_sha1su1>;
4868 def SHA256SU0rr : SHATiedInstVV<0b0010, "sha256su0",int_aarch64_crypto_sha256su0>;
4870 //----------------------------------------------------------------------------
4872 //----------------------------------------------------------------------------
4873 // FIXME: Like for X86, these should go in their own separate .td file.
4875 // Any instruction that defines a 32-bit result leaves the high half of the
4876 // register. Truncate can be lowered to EXTRACT_SUBREG. CopyFromReg may
4877 // be copying from a truncate. But any other 32-bit operation will zero-extend
4879 // FIXME: X86 also checks for CMOV here. Do we need something similar?
4880 def def32 : PatLeaf<(i32 GPR32:$src), [{
4881 return N->getOpcode() != ISD::TRUNCATE &&
4882 N->getOpcode() != TargetOpcode::EXTRACT_SUBREG &&
4883 N->getOpcode() != ISD::CopyFromReg;
4886 // In the case of a 32-bit def that is known to implicitly zero-extend,
4887 // we can use a SUBREG_TO_REG.
4888 def : Pat<(i64 (zext def32:$src)), (SUBREG_TO_REG (i64 0), GPR32:$src, sub_32)>;
4890 // For an anyext, we don't care what the high bits are, so we can perform an
4891 // INSERT_SUBREF into an IMPLICIT_DEF.
4892 def : Pat<(i64 (anyext GPR32:$src)),
4893 (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32)>;
4895 // When we need to explicitly zero-extend, we use an unsigned bitfield move
4896 // instruction (UBFM) on the enclosing super-reg.
4897 def : Pat<(i64 (zext GPR32:$src)),
4898 (UBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32), 0, 31)>;
4900 // To sign extend, we use a signed bitfield move instruction (SBFM) on the
4901 // containing super-reg.
4902 def : Pat<(i64 (sext GPR32:$src)),
4903 (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32), 0, 31)>;
4904 def : Pat<(i64 (sext_inreg GPR64:$src, i32)), (SBFMXri GPR64:$src, 0, 31)>;
4905 def : Pat<(i64 (sext_inreg GPR64:$src, i16)), (SBFMXri GPR64:$src, 0, 15)>;
4906 def : Pat<(i64 (sext_inreg GPR64:$src, i8)), (SBFMXri GPR64:$src, 0, 7)>;
4907 def : Pat<(i64 (sext_inreg GPR64:$src, i1)), (SBFMXri GPR64:$src, 0, 0)>;
4908 def : Pat<(i32 (sext_inreg GPR32:$src, i16)), (SBFMWri GPR32:$src, 0, 15)>;
4909 def : Pat<(i32 (sext_inreg GPR32:$src, i8)), (SBFMWri GPR32:$src, 0, 7)>;
4910 def : Pat<(i32 (sext_inreg GPR32:$src, i1)), (SBFMWri GPR32:$src, 0, 0)>;
4912 def : Pat<(shl (sext_inreg GPR32:$Rn, i8), (i64 imm0_31:$imm)),
4913 (SBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
4914 (i64 (i32shift_sext_i8 imm0_31:$imm)))>;
4915 def : Pat<(shl (sext_inreg GPR64:$Rn, i8), (i64 imm0_63:$imm)),
4916 (SBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
4917 (i64 (i64shift_sext_i8 imm0_63:$imm)))>;
4919 def : Pat<(shl (sext_inreg GPR32:$Rn, i16), (i64 imm0_31:$imm)),
4920 (SBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
4921 (i64 (i32shift_sext_i16 imm0_31:$imm)))>;
4922 def : Pat<(shl (sext_inreg GPR64:$Rn, i16), (i64 imm0_63:$imm)),
4923 (SBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
4924 (i64 (i64shift_sext_i16 imm0_63:$imm)))>;
4926 def : Pat<(shl (i64 (sext GPR32:$Rn)), (i64 imm0_63:$imm)),
4927 (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$Rn, sub_32),
4928 (i64 (i64shift_a imm0_63:$imm)),
4929 (i64 (i64shift_sext_i32 imm0_63:$imm)))>;
4931 // sra patterns have an AddedComplexity of 10, so make sure we have a higher
4932 // AddedComplexity for the following patterns since we want to match sext + sra
4933 // patterns before we attempt to match a single sra node.
4934 let AddedComplexity = 20 in {
4935 // We support all sext + sra combinations which preserve at least one bit of the
4936 // original value which is to be sign extended. E.g. we support shifts up to
4938 def : Pat<(sra (sext_inreg GPR32:$Rn, i8), (i64 imm0_7:$imm)),
4939 (SBFMWri GPR32:$Rn, (i64 imm0_7:$imm), 7)>;
4940 def : Pat<(sra (sext_inreg GPR64:$Rn, i8), (i64 imm0_7:$imm)),
4941 (SBFMXri GPR64:$Rn, (i64 imm0_7:$imm), 7)>;
4943 def : Pat<(sra (sext_inreg GPR32:$Rn, i16), (i64 imm0_15:$imm)),
4944 (SBFMWri GPR32:$Rn, (i64 imm0_15:$imm), 15)>;
4945 def : Pat<(sra (sext_inreg GPR64:$Rn, i16), (i64 imm0_15:$imm)),
4946 (SBFMXri GPR64:$Rn, (i64 imm0_15:$imm), 15)>;
4948 def : Pat<(sra (i64 (sext GPR32:$Rn)), (i64 imm0_31:$imm)),
4949 (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$Rn, sub_32),
4950 (i64 imm0_31:$imm), 31)>;
4951 } // AddedComplexity = 20
4953 // To truncate, we can simply extract from a subregister.
4954 def : Pat<(i32 (trunc GPR64sp:$src)),
4955 (i32 (EXTRACT_SUBREG GPR64sp:$src, sub_32))>;
4957 // __builtin_trap() uses the BRK instruction on AArch64.
4958 def : Pat<(trap), (BRK 1)>;
4960 // Conversions within AdvSIMD types in the same register size are free.
4961 // But because we need a consistent lane ordering, in big endian many
4962 // conversions require one or more REV instructions.
4964 // Consider a simple memory load followed by a bitconvert then a store.
4966 // v1 = BITCAST v2i32 v0 to v4i16
4969 // In big endian mode every memory access has an implicit byte swap. LDR and
4970 // STR do a 64-bit byte swap, whereas LD1/ST1 do a byte swap per lane - that
4971 // is, they treat the vector as a sequence of elements to be byte-swapped.
4972 // The two pairs of instructions are fundamentally incompatible. We've decided
4973 // to use LD1/ST1 only to simplify compiler implementation.
4975 // LD1/ST1 perform the equivalent of a sequence of LDR/STR + REV. This makes
4976 // the original code sequence:
4978 // v1 = REV v2i32 (implicit)
4979 // v2 = BITCAST v2i32 v1 to v4i16
4980 // v3 = REV v4i16 v2 (implicit)
4983 // But this is now broken - the value stored is different to the value loaded
4984 // due to lane reordering. To fix this, on every BITCAST we must perform two
4987 // v1 = REV v2i32 (implicit)
4989 // v3 = BITCAST v2i32 v2 to v4i16
4991 // v5 = REV v4i16 v4 (implicit)
4994 // This means an extra two instructions, but actually in most cases the two REV
4995 // instructions can be combined into one. For example:
4996 // (REV64_2s (REV64_4h X)) === (REV32_4h X)
4998 // There is also no 128-bit REV instruction. This must be synthesized with an
5001 // Most bitconverts require some sort of conversion. The only exceptions are:
5002 // a) Identity conversions - vNfX <-> vNiX
5003 // b) Single-lane-to-scalar - v1fX <-> fX or v1iX <-> iX
5006 // Natural vector casts (64 bit)
5007 def : Pat<(v8i8 (AArch64NvCast (v2i32 FPR64:$src))), (v8i8 FPR64:$src)>;
5008 def : Pat<(v4i16 (AArch64NvCast (v2i32 FPR64:$src))), (v4i16 FPR64:$src)>;
5009 def : Pat<(v2i32 (AArch64NvCast (v2i32 FPR64:$src))), (v2i32 FPR64:$src)>;
5010 def : Pat<(v2f32 (AArch64NvCast (v2i32 FPR64:$src))), (v2f32 FPR64:$src)>;
5011 def : Pat<(v1i64 (AArch64NvCast (v2i32 FPR64:$src))), (v1i64 FPR64:$src)>;
5013 def : Pat<(v8i8 (AArch64NvCast (v4i16 FPR64:$src))), (v8i8 FPR64:$src)>;
5014 def : Pat<(v4i16 (AArch64NvCast (v4i16 FPR64:$src))), (v4i16 FPR64:$src)>;
5015 def : Pat<(v2i32 (AArch64NvCast (v4i16 FPR64:$src))), (v2i32 FPR64:$src)>;
5016 def : Pat<(v1i64 (AArch64NvCast (v4i16 FPR64:$src))), (v1i64 FPR64:$src)>;
5018 def : Pat<(v8i8 (AArch64NvCast (v8i8 FPR64:$src))), (v8i8 FPR64:$src)>;
5019 def : Pat<(v4i16 (AArch64NvCast (v8i8 FPR64:$src))), (v4i16 FPR64:$src)>;
5020 def : Pat<(v2i32 (AArch64NvCast (v8i8 FPR64:$src))), (v2i32 FPR64:$src)>;
5021 def : Pat<(v1i64 (AArch64NvCast (v8i8 FPR64:$src))), (v1i64 FPR64:$src)>;
5023 def : Pat<(v8i8 (AArch64NvCast (f64 FPR64:$src))), (v8i8 FPR64:$src)>;
5024 def : Pat<(v4i16 (AArch64NvCast (f64 FPR64:$src))), (v4i16 FPR64:$src)>;
5025 def : Pat<(v2i32 (AArch64NvCast (f64 FPR64:$src))), (v2i32 FPR64:$src)>;
5026 def : Pat<(v2f32 (AArch64NvCast (f64 FPR64:$src))), (v2f32 FPR64:$src)>;
5027 def : Pat<(v1i64 (AArch64NvCast (f64 FPR64:$src))), (v1i64 FPR64:$src)>;
5028 def : Pat<(v1f64 (AArch64NvCast (f64 FPR64:$src))), (v1f64 FPR64:$src)>;
5030 def : Pat<(v8i8 (AArch64NvCast (v2f32 FPR64:$src))), (v8i8 FPR64:$src)>;
5031 def : Pat<(v4i16 (AArch64NvCast (v2f32 FPR64:$src))), (v4i16 FPR64:$src)>;
5032 def : Pat<(v2i32 (AArch64NvCast (v2f32 FPR64:$src))), (v2i32 FPR64:$src)>;
5033 def : Pat<(v2f32 (AArch64NvCast (v2f32 FPR64:$src))), (v2f32 FPR64:$src)>;
5034 def : Pat<(v1i64 (AArch64NvCast (v2f32 FPR64:$src))), (v1i64 FPR64:$src)>;
5036 // Natural vector casts (128 bit)
5037 def : Pat<(v16i8 (AArch64NvCast (v4i32 FPR128:$src))), (v16i8 FPR128:$src)>;
5038 def : Pat<(v8i16 (AArch64NvCast (v4i32 FPR128:$src))), (v8i16 FPR128:$src)>;
5039 def : Pat<(v4i32 (AArch64NvCast (v4i32 FPR128:$src))), (v4i32 FPR128:$src)>;
5040 def : Pat<(v4f32 (AArch64NvCast (v4i32 FPR128:$src))), (v4f32 FPR128:$src)>;
5041 def : Pat<(v2i64 (AArch64NvCast (v4i32 FPR128:$src))), (v2i64 FPR128:$src)>;
5043 def : Pat<(v16i8 (AArch64NvCast (v8i16 FPR128:$src))), (v16i8 FPR128:$src)>;
5044 def : Pat<(v8i16 (AArch64NvCast (v8i16 FPR128:$src))), (v8i16 FPR128:$src)>;
5045 def : Pat<(v4i32 (AArch64NvCast (v8i16 FPR128:$src))), (v4i32 FPR128:$src)>;
5046 def : Pat<(v2i64 (AArch64NvCast (v8i16 FPR128:$src))), (v2i64 FPR128:$src)>;
5048 def : Pat<(v16i8 (AArch64NvCast (v16i8 FPR128:$src))), (v16i8 FPR128:$src)>;
5049 def : Pat<(v8i16 (AArch64NvCast (v16i8 FPR128:$src))), (v8i16 FPR128:$src)>;
5050 def : Pat<(v4i32 (AArch64NvCast (v16i8 FPR128:$src))), (v4i32 FPR128:$src)>;
5051 def : Pat<(v2i64 (AArch64NvCast (v16i8 FPR128:$src))), (v2i64 FPR128:$src)>;
5053 def : Pat<(v16i8 (AArch64NvCast (v2i64 FPR128:$src))), (v16i8 FPR128:$src)>;
5054 def : Pat<(v8i16 (AArch64NvCast (v2i64 FPR128:$src))), (v8i16 FPR128:$src)>;
5055 def : Pat<(v4i32 (AArch64NvCast (v2i64 FPR128:$src))), (v4i32 FPR128:$src)>;
5056 def : Pat<(v2i64 (AArch64NvCast (v2i64 FPR128:$src))), (v2i64 FPR128:$src)>;
5057 def : Pat<(v4f32 (AArch64NvCast (v2i64 FPR128:$src))), (v4f32 FPR128:$src)>;
5058 def : Pat<(v2f64 (AArch64NvCast (v2i64 FPR128:$src))), (v2f64 FPR128:$src)>;
5060 def : Pat<(v16i8 (AArch64NvCast (v4f32 FPR128:$src))), (v16i8 FPR128:$src)>;
5061 def : Pat<(v8i16 (AArch64NvCast (v4f32 FPR128:$src))), (v8i16 FPR128:$src)>;
5062 def : Pat<(v4i32 (AArch64NvCast (v4f32 FPR128:$src))), (v4i32 FPR128:$src)>;
5063 def : Pat<(v4f32 (AArch64NvCast (v4f32 FPR128:$src))), (v4f32 FPR128:$src)>;
5064 def : Pat<(v2i64 (AArch64NvCast (v4f32 FPR128:$src))), (v2i64 FPR128:$src)>;
5066 def : Pat<(v16i8 (AArch64NvCast (v2f64 FPR128:$src))), (v16i8 FPR128:$src)>;
5067 def : Pat<(v8i16 (AArch64NvCast (v2f64 FPR128:$src))), (v8i16 FPR128:$src)>;
5068 def : Pat<(v4i32 (AArch64NvCast (v2f64 FPR128:$src))), (v4i32 FPR128:$src)>;
5069 def : Pat<(v2i64 (AArch64NvCast (v2f64 FPR128:$src))), (v2i64 FPR128:$src)>;
5070 def : Pat<(v2f64 (AArch64NvCast (v2f64 FPR128:$src))), (v2f64 FPR128:$src)>;
5072 let Predicates = [IsLE] in {
5073 def : Pat<(v8i8 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5074 def : Pat<(v4i16 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5075 def : Pat<(v2i32 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5076 def : Pat<(v4f16 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5077 def : Pat<(v2f32 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5079 def : Pat<(i64 (bitconvert (v8i8 V64:$Vn))),
5080 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5081 def : Pat<(i64 (bitconvert (v4i16 V64:$Vn))),
5082 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5083 def : Pat<(i64 (bitconvert (v2i32 V64:$Vn))),
5084 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5085 def : Pat<(i64 (bitconvert (v4f16 V64:$Vn))),
5086 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5087 def : Pat<(i64 (bitconvert (v2f32 V64:$Vn))),
5088 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5089 def : Pat<(i64 (bitconvert (v1f64 V64:$Vn))),
5090 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5092 let Predicates = [IsBE] in {
5093 def : Pat<(v8i8 (bitconvert GPR64:$Xn)),
5094 (REV64v8i8 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
5095 def : Pat<(v4i16 (bitconvert GPR64:$Xn)),
5096 (REV64v4i16 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
5097 def : Pat<(v2i32 (bitconvert GPR64:$Xn)),
5098 (REV64v2i32 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
5099 def : Pat<(v4f16 (bitconvert GPR64:$Xn)),
5100 (REV64v4i16 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
5101 def : Pat<(v2f32 (bitconvert GPR64:$Xn)),
5102 (REV64v2i32 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
5104 def : Pat<(i64 (bitconvert (v8i8 V64:$Vn))),
5105 (REV64v8i8 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
5106 def : Pat<(i64 (bitconvert (v4i16 V64:$Vn))),
5107 (REV64v4i16 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
5108 def : Pat<(i64 (bitconvert (v2i32 V64:$Vn))),
5109 (REV64v2i32 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
5110 def : Pat<(i64 (bitconvert (v4f16 V64:$Vn))),
5111 (REV64v4i16 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
5112 def : Pat<(i64 (bitconvert (v2f32 V64:$Vn))),
5113 (REV64v2i32 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
5115 def : Pat<(v1i64 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5116 def : Pat<(v1f64 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5117 def : Pat<(i64 (bitconvert (v1i64 V64:$Vn))),
5118 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5119 def : Pat<(v1i64 (scalar_to_vector GPR64:$Xn)),
5120 (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5121 def : Pat<(v1f64 (scalar_to_vector GPR64:$Xn)),
5122 (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5123 def : Pat<(v1f64 (scalar_to_vector (f64 FPR64:$Xn))), (v1f64 FPR64:$Xn)>;
5125 def : Pat<(f32 (bitconvert (i32 GPR32:$Xn))),
5126 (COPY_TO_REGCLASS GPR32:$Xn, FPR32)>;
5127 def : Pat<(i32 (bitconvert (f32 FPR32:$Xn))),
5128 (COPY_TO_REGCLASS FPR32:$Xn, GPR32)>;
5129 def : Pat<(f64 (bitconvert (i64 GPR64:$Xn))),
5130 (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5131 def : Pat<(i64 (bitconvert (f64 FPR64:$Xn))),
5132 (COPY_TO_REGCLASS FPR64:$Xn, GPR64)>;
5133 def : Pat<(i64 (bitconvert (v1f64 V64:$Vn))),
5134 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5136 let Predicates = [IsLE] in {
5137 def : Pat<(v1i64 (bitconvert (v2i32 FPR64:$src))), (v1i64 FPR64:$src)>;
5138 def : Pat<(v1i64 (bitconvert (v4i16 FPR64:$src))), (v1i64 FPR64:$src)>;
5139 def : Pat<(v1i64 (bitconvert (v8i8 FPR64:$src))), (v1i64 FPR64:$src)>;
5140 def : Pat<(v1i64 (bitconvert (v4f16 FPR64:$src))), (v1i64 FPR64:$src)>;
5141 def : Pat<(v1i64 (bitconvert (v2f32 FPR64:$src))), (v1i64 FPR64:$src)>;
5143 let Predicates = [IsBE] in {
5144 def : Pat<(v1i64 (bitconvert (v2i32 FPR64:$src))),
5145 (v1i64 (REV64v2i32 FPR64:$src))>;
5146 def : Pat<(v1i64 (bitconvert (v4i16 FPR64:$src))),
5147 (v1i64 (REV64v4i16 FPR64:$src))>;
5148 def : Pat<(v1i64 (bitconvert (v8i8 FPR64:$src))),
5149 (v1i64 (REV64v8i8 FPR64:$src))>;
5150 def : Pat<(v1i64 (bitconvert (v4f16 FPR64:$src))),
5151 (v1i64 (REV64v4i16 FPR64:$src))>;
5152 def : Pat<(v1i64 (bitconvert (v2f32 FPR64:$src))),
5153 (v1i64 (REV64v2i32 FPR64:$src))>;
5155 def : Pat<(v1i64 (bitconvert (v1f64 FPR64:$src))), (v1i64 FPR64:$src)>;
5156 def : Pat<(v1i64 (bitconvert (f64 FPR64:$src))), (v1i64 FPR64:$src)>;
5158 let Predicates = [IsLE] in {
5159 def : Pat<(v2i32 (bitconvert (v1i64 FPR64:$src))), (v2i32 FPR64:$src)>;
5160 def : Pat<(v2i32 (bitconvert (v4i16 FPR64:$src))), (v2i32 FPR64:$src)>;
5161 def : Pat<(v2i32 (bitconvert (v8i8 FPR64:$src))), (v2i32 FPR64:$src)>;
5162 def : Pat<(v2i32 (bitconvert (f64 FPR64:$src))), (v2i32 FPR64:$src)>;
5163 def : Pat<(v2i32 (bitconvert (v1f64 FPR64:$src))), (v2i32 FPR64:$src)>;
5164 def : Pat<(v2i32 (bitconvert (v4f16 FPR64:$src))), (v2i32 FPR64:$src)>;
5166 let Predicates = [IsBE] in {
5167 def : Pat<(v2i32 (bitconvert (v1i64 FPR64:$src))),
5168 (v2i32 (REV64v2i32 FPR64:$src))>;
5169 def : Pat<(v2i32 (bitconvert (v4i16 FPR64:$src))),
5170 (v2i32 (REV32v4i16 FPR64:$src))>;
5171 def : Pat<(v2i32 (bitconvert (v8i8 FPR64:$src))),
5172 (v2i32 (REV32v8i8 FPR64:$src))>;
5173 def : Pat<(v2i32 (bitconvert (f64 FPR64:$src))),
5174 (v2i32 (REV64v2i32 FPR64:$src))>;
5175 def : Pat<(v2i32 (bitconvert (v1f64 FPR64:$src))),
5176 (v2i32 (REV64v2i32 FPR64:$src))>;
5177 def : Pat<(v2i32 (bitconvert (v4f16 FPR64:$src))),
5178 (v2i32 (REV64v4i16 FPR64:$src))>;
5180 def : Pat<(v2i32 (bitconvert (v2f32 FPR64:$src))), (v2i32 FPR64:$src)>;
5182 let Predicates = [IsLE] in {
5183 def : Pat<(v4i16 (bitconvert (v1i64 FPR64:$src))), (v4i16 FPR64:$src)>;
5184 def : Pat<(v4i16 (bitconvert (v2i32 FPR64:$src))), (v4i16 FPR64:$src)>;
5185 def : Pat<(v4i16 (bitconvert (v8i8 FPR64:$src))), (v4i16 FPR64:$src)>;
5186 def : Pat<(v4i16 (bitconvert (f64 FPR64:$src))), (v4i16 FPR64:$src)>;
5187 def : Pat<(v4i16 (bitconvert (v4f16 FPR64:$src))), (v4i16 FPR64:$src)>;
5188 def : Pat<(v4i16 (bitconvert (v2f32 FPR64:$src))), (v4i16 FPR64:$src)>;
5189 def : Pat<(v4i16 (bitconvert (v1f64 FPR64:$src))), (v4i16 FPR64:$src)>;
5191 let Predicates = [IsBE] in {
5192 def : Pat<(v4i16 (bitconvert (v1i64 FPR64:$src))),
5193 (v4i16 (REV64v4i16 FPR64:$src))>;
5194 def : Pat<(v4i16 (bitconvert (v2i32 FPR64:$src))),
5195 (v4i16 (REV32v4i16 FPR64:$src))>;
5196 def : Pat<(v4i16 (bitconvert (v8i8 FPR64:$src))),
5197 (v4i16 (REV16v8i8 FPR64:$src))>;
5198 def : Pat<(v4i16 (bitconvert (f64 FPR64:$src))),
5199 (v4i16 (REV64v4i16 FPR64:$src))>;
5200 def : Pat<(v4i16 (bitconvert (v4f16 FPR64:$src))),
5201 (v4i16 (REV32v4i16 FPR64:$src))>;
5202 def : Pat<(v4i16 (bitconvert (v2f32 FPR64:$src))),
5203 (v4i16 (REV32v4i16 FPR64:$src))>;
5204 def : Pat<(v4i16 (bitconvert (v1f64 FPR64:$src))),
5205 (v4i16 (REV64v4i16 FPR64:$src))>;
5208 let Predicates = [IsLE] in {
5209 def : Pat<(v4f16 (bitconvert (v1i64 FPR64:$src))), (v4f16 FPR64:$src)>;
5210 def : Pat<(v4f16 (bitconvert (v2i32 FPR64:$src))), (v4f16 FPR64:$src)>;
5211 def : Pat<(v4f16 (bitconvert (v4i16 FPR64:$src))), (v4f16 FPR64:$src)>;
5212 def : Pat<(v4f16 (bitconvert (v8i8 FPR64:$src))), (v4f16 FPR64:$src)>;
5213 def : Pat<(v4f16 (bitconvert (f64 FPR64:$src))), (v4f16 FPR64:$src)>;
5214 def : Pat<(v4f16 (bitconvert (v2f32 FPR64:$src))), (v4f16 FPR64:$src)>;
5215 def : Pat<(v4f16 (bitconvert (v1f64 FPR64:$src))), (v4f16 FPR64:$src)>;
5217 let Predicates = [IsBE] in {
5218 def : Pat<(v4f16 (bitconvert (v1i64 FPR64:$src))),
5219 (v4f16 (REV64v4i16 FPR64:$src))>;
5220 def : Pat<(v4f16 (bitconvert (v2i32 FPR64:$src))),
5221 (v4f16 (REV64v4i16 FPR64:$src))>;
5222 def : Pat<(v4f16 (bitconvert (v4i16 FPR64:$src))),
5223 (v4f16 (REV64v4i16 FPR64:$src))>;
5224 def : Pat<(v4f16 (bitconvert (v8i8 FPR64:$src))),
5225 (v4f16 (REV16v8i8 FPR64:$src))>;
5226 def : Pat<(v4f16 (bitconvert (f64 FPR64:$src))),
5227 (v4f16 (REV64v4i16 FPR64:$src))>;
5228 def : Pat<(v4f16 (bitconvert (v2f32 FPR64:$src))),
5229 (v4f16 (REV64v4i16 FPR64:$src))>;
5230 def : Pat<(v4f16 (bitconvert (v1f64 FPR64:$src))),
5231 (v4f16 (REV64v4i16 FPR64:$src))>;
5236 let Predicates = [IsLE] in {
5237 def : Pat<(v8i8 (bitconvert (v1i64 FPR64:$src))), (v8i8 FPR64:$src)>;
5238 def : Pat<(v8i8 (bitconvert (v2i32 FPR64:$src))), (v8i8 FPR64:$src)>;
5239 def : Pat<(v8i8 (bitconvert (v4i16 FPR64:$src))), (v8i8 FPR64:$src)>;
5240 def : Pat<(v8i8 (bitconvert (f64 FPR64:$src))), (v8i8 FPR64:$src)>;
5241 def : Pat<(v8i8 (bitconvert (v2f32 FPR64:$src))), (v8i8 FPR64:$src)>;
5242 def : Pat<(v8i8 (bitconvert (v1f64 FPR64:$src))), (v8i8 FPR64:$src)>;
5243 def : Pat<(v8i8 (bitconvert (v4f16 FPR64:$src))), (v8i8 FPR64:$src)>;
5245 let Predicates = [IsBE] in {
5246 def : Pat<(v8i8 (bitconvert (v1i64 FPR64:$src))),
5247 (v8i8 (REV64v8i8 FPR64:$src))>;
5248 def : Pat<(v8i8 (bitconvert (v2i32 FPR64:$src))),
5249 (v8i8 (REV32v8i8 FPR64:$src))>;
5250 def : Pat<(v8i8 (bitconvert (v4i16 FPR64:$src))),
5251 (v8i8 (REV16v8i8 FPR64:$src))>;
5252 def : Pat<(v8i8 (bitconvert (f64 FPR64:$src))),
5253 (v8i8 (REV64v8i8 FPR64:$src))>;
5254 def : Pat<(v8i8 (bitconvert (v2f32 FPR64:$src))),
5255 (v8i8 (REV32v8i8 FPR64:$src))>;
5256 def : Pat<(v8i8 (bitconvert (v1f64 FPR64:$src))),
5257 (v8i8 (REV64v8i8 FPR64:$src))>;
5258 def : Pat<(v8i8 (bitconvert (v4f16 FPR64:$src))),
5259 (v8i8 (REV16v8i8 FPR64:$src))>;
5262 let Predicates = [IsLE] in {
5263 def : Pat<(f64 (bitconvert (v2i32 FPR64:$src))), (f64 FPR64:$src)>;
5264 def : Pat<(f64 (bitconvert (v4i16 FPR64:$src))), (f64 FPR64:$src)>;
5265 def : Pat<(f64 (bitconvert (v2f32 FPR64:$src))), (f64 FPR64:$src)>;
5266 def : Pat<(f64 (bitconvert (v8i8 FPR64:$src))), (f64 FPR64:$src)>;
5267 def : Pat<(f64 (bitconvert (v4f16 FPR64:$src))), (f64 FPR64:$src)>;
5269 let Predicates = [IsBE] in {
5270 def : Pat<(f64 (bitconvert (v2i32 FPR64:$src))),
5271 (f64 (REV64v2i32 FPR64:$src))>;
5272 def : Pat<(f64 (bitconvert (v4i16 FPR64:$src))),
5273 (f64 (REV64v4i16 FPR64:$src))>;
5274 def : Pat<(f64 (bitconvert (v2f32 FPR64:$src))),
5275 (f64 (REV64v2i32 FPR64:$src))>;
5276 def : Pat<(f64 (bitconvert (v8i8 FPR64:$src))),
5277 (f64 (REV64v8i8 FPR64:$src))>;
5278 def : Pat<(f64 (bitconvert (v4f16 FPR64:$src))),
5279 (f64 (REV64v4i16 FPR64:$src))>;
5281 def : Pat<(f64 (bitconvert (v1i64 FPR64:$src))), (f64 FPR64:$src)>;
5282 def : Pat<(f64 (bitconvert (v1f64 FPR64:$src))), (f64 FPR64:$src)>;
5284 let Predicates = [IsLE] in {
5285 def : Pat<(v1f64 (bitconvert (v2i32 FPR64:$src))), (v1f64 FPR64:$src)>;
5286 def : Pat<(v1f64 (bitconvert (v4i16 FPR64:$src))), (v1f64 FPR64:$src)>;
5287 def : Pat<(v1f64 (bitconvert (v8i8 FPR64:$src))), (v1f64 FPR64:$src)>;
5288 def : Pat<(v1f64 (bitconvert (v2f32 FPR64:$src))), (v1f64 FPR64:$src)>;
5289 def : Pat<(v1f64 (bitconvert (v4f16 FPR64:$src))), (v1f64 FPR64:$src)>;
5291 let Predicates = [IsBE] in {
5292 def : Pat<(v1f64 (bitconvert (v2i32 FPR64:$src))),
5293 (v1f64 (REV64v2i32 FPR64:$src))>;
5294 def : Pat<(v1f64 (bitconvert (v4i16 FPR64:$src))),
5295 (v1f64 (REV64v4i16 FPR64:$src))>;
5296 def : Pat<(v1f64 (bitconvert (v8i8 FPR64:$src))),
5297 (v1f64 (REV64v8i8 FPR64:$src))>;
5298 def : Pat<(v1f64 (bitconvert (v2f32 FPR64:$src))),
5299 (v1f64 (REV64v2i32 FPR64:$src))>;
5300 def : Pat<(v1f64 (bitconvert (v4f16 FPR64:$src))),
5301 (v1f64 (REV64v4i16 FPR64:$src))>;
5303 def : Pat<(v1f64 (bitconvert (v1i64 FPR64:$src))), (v1f64 FPR64:$src)>;
5304 def : Pat<(v1f64 (bitconvert (f64 FPR64:$src))), (v1f64 FPR64:$src)>;
5306 let Predicates = [IsLE] in {
5307 def : Pat<(v2f32 (bitconvert (v1i64 FPR64:$src))), (v2f32 FPR64:$src)>;
5308 def : Pat<(v2f32 (bitconvert (v4i16 FPR64:$src))), (v2f32 FPR64:$src)>;
5309 def : Pat<(v2f32 (bitconvert (v8i8 FPR64:$src))), (v2f32 FPR64:$src)>;
5310 def : Pat<(v2f32 (bitconvert (v1f64 FPR64:$src))), (v2f32 FPR64:$src)>;
5311 def : Pat<(v2f32 (bitconvert (f64 FPR64:$src))), (v2f32 FPR64:$src)>;
5312 def : Pat<(v2f32 (bitconvert (v4f16 FPR64:$src))), (v2f32 FPR64:$src)>;
5314 let Predicates = [IsBE] in {
5315 def : Pat<(v2f32 (bitconvert (v1i64 FPR64:$src))),
5316 (v2f32 (REV64v2i32 FPR64:$src))>;
5317 def : Pat<(v2f32 (bitconvert (v4i16 FPR64:$src))),
5318 (v2f32 (REV32v4i16 FPR64:$src))>;
5319 def : Pat<(v2f32 (bitconvert (v8i8 FPR64:$src))),
5320 (v2f32 (REV32v8i8 FPR64:$src))>;
5321 def : Pat<(v2f32 (bitconvert (v1f64 FPR64:$src))),
5322 (v2f32 (REV64v2i32 FPR64:$src))>;
5323 def : Pat<(v2f32 (bitconvert (f64 FPR64:$src))),
5324 (v2f32 (REV64v2i32 FPR64:$src))>;
5325 def : Pat<(v2f32 (bitconvert (v4f16 FPR64:$src))),
5326 (v2f32 (REV64v4i16 FPR64:$src))>;
5328 def : Pat<(v2f32 (bitconvert (v2i32 FPR64:$src))), (v2f32 FPR64:$src)>;
5330 let Predicates = [IsLE] in {
5331 def : Pat<(f128 (bitconvert (v2i64 FPR128:$src))), (f128 FPR128:$src)>;
5332 def : Pat<(f128 (bitconvert (v4i32 FPR128:$src))), (f128 FPR128:$src)>;
5333 def : Pat<(f128 (bitconvert (v8i16 FPR128:$src))), (f128 FPR128:$src)>;
5334 def : Pat<(f128 (bitconvert (v2f64 FPR128:$src))), (f128 FPR128:$src)>;
5335 def : Pat<(f128 (bitconvert (v4f32 FPR128:$src))), (f128 FPR128:$src)>;
5336 def : Pat<(f128 (bitconvert (v8f16 FPR128:$src))), (f128 FPR128:$src)>;
5337 def : Pat<(f128 (bitconvert (v16i8 FPR128:$src))), (f128 FPR128:$src)>;
5339 let Predicates = [IsBE] in {
5340 def : Pat<(f128 (bitconvert (v2i64 FPR128:$src))),
5341 (f128 (EXTv16i8 FPR128:$src, FPR128:$src, (i32 8)))>;
5342 def : Pat<(f128 (bitconvert (v4i32 FPR128:$src))),
5343 (f128 (EXTv16i8 (REV64v4i32 FPR128:$src),
5344 (REV64v4i32 FPR128:$src), (i32 8)))>;
5345 def : Pat<(f128 (bitconvert (v8i16 FPR128:$src))),
5346 (f128 (EXTv16i8 (REV64v8i16 FPR128:$src),
5347 (REV64v8i16 FPR128:$src), (i32 8)))>;
5348 def : Pat<(f128 (bitconvert (v8f16 FPR128:$src))),
5349 (f128 (EXTv16i8 (REV64v8i16 FPR128:$src),
5350 (REV64v8i16 FPR128:$src), (i32 8)))>;
5351 def : Pat<(f128 (bitconvert (v2f64 FPR128:$src))),
5352 (f128 (EXTv16i8 FPR128:$src, FPR128:$src, (i32 8)))>;
5353 def : Pat<(f128 (bitconvert (v4f32 FPR128:$src))),
5354 (f128 (EXTv16i8 (REV64v4i32 FPR128:$src),
5355 (REV64v4i32 FPR128:$src), (i32 8)))>;
5356 def : Pat<(f128 (bitconvert (v16i8 FPR128:$src))),
5357 (f128 (EXTv16i8 (REV64v16i8 FPR128:$src),
5358 (REV64v16i8 FPR128:$src), (i32 8)))>;
5361 let Predicates = [IsLE] in {
5362 def : Pat<(v2f64 (bitconvert (f128 FPR128:$src))), (v2f64 FPR128:$src)>;
5363 def : Pat<(v2f64 (bitconvert (v4i32 FPR128:$src))), (v2f64 FPR128:$src)>;
5364 def : Pat<(v2f64 (bitconvert (v8i16 FPR128:$src))), (v2f64 FPR128:$src)>;
5365 def : Pat<(v2f64 (bitconvert (v8f16 FPR128:$src))), (v2f64 FPR128:$src)>;
5366 def : Pat<(v2f64 (bitconvert (v16i8 FPR128:$src))), (v2f64 FPR128:$src)>;
5367 def : Pat<(v2f64 (bitconvert (v4f32 FPR128:$src))), (v2f64 FPR128:$src)>;
5369 let Predicates = [IsBE] in {
5370 def : Pat<(v2f64 (bitconvert (f128 FPR128:$src))),
5371 (v2f64 (EXTv16i8 FPR128:$src,
5372 FPR128:$src, (i32 8)))>;
5373 def : Pat<(v2f64 (bitconvert (v4i32 FPR128:$src))),
5374 (v2f64 (REV64v4i32 FPR128:$src))>;
5375 def : Pat<(v2f64 (bitconvert (v8i16 FPR128:$src))),
5376 (v2f64 (REV64v8i16 FPR128:$src))>;
5377 def : Pat<(v2f64 (bitconvert (v8f16 FPR128:$src))),
5378 (v2f64 (REV64v8i16 FPR128:$src))>;
5379 def : Pat<(v2f64 (bitconvert (v16i8 FPR128:$src))),
5380 (v2f64 (REV64v16i8 FPR128:$src))>;
5381 def : Pat<(v2f64 (bitconvert (v4f32 FPR128:$src))),
5382 (v2f64 (REV64v4i32 FPR128:$src))>;
5384 def : Pat<(v2f64 (bitconvert (v2i64 FPR128:$src))), (v2f64 FPR128:$src)>;
5386 let Predicates = [IsLE] in {
5387 def : Pat<(v4f32 (bitconvert (f128 FPR128:$src))), (v4f32 FPR128:$src)>;
5388 def : Pat<(v4f32 (bitconvert (v8i16 FPR128:$src))), (v4f32 FPR128:$src)>;
5389 def : Pat<(v4f32 (bitconvert (v8f16 FPR128:$src))), (v4f32 FPR128:$src)>;
5390 def : Pat<(v4f32 (bitconvert (v16i8 FPR128:$src))), (v4f32 FPR128:$src)>;
5391 def : Pat<(v4f32 (bitconvert (v2i64 FPR128:$src))), (v4f32 FPR128:$src)>;
5392 def : Pat<(v4f32 (bitconvert (v2f64 FPR128:$src))), (v4f32 FPR128:$src)>;
5394 let Predicates = [IsBE] in {
5395 def : Pat<(v4f32 (bitconvert (f128 FPR128:$src))),
5396 (v4f32 (EXTv16i8 (REV64v4i32 FPR128:$src),
5397 (REV64v4i32 FPR128:$src), (i32 8)))>;
5398 def : Pat<(v4f32 (bitconvert (v8i16 FPR128:$src))),
5399 (v4f32 (REV32v8i16 FPR128:$src))>;
5400 def : Pat<(v4f32 (bitconvert (v8f16 FPR128:$src))),
5401 (v4f32 (REV32v8i16 FPR128:$src))>;
5402 def : Pat<(v4f32 (bitconvert (v16i8 FPR128:$src))),
5403 (v4f32 (REV32v16i8 FPR128:$src))>;
5404 def : Pat<(v4f32 (bitconvert (v2i64 FPR128:$src))),
5405 (v4f32 (REV64v4i32 FPR128:$src))>;
5406 def : Pat<(v4f32 (bitconvert (v2f64 FPR128:$src))),
5407 (v4f32 (REV64v4i32 FPR128:$src))>;
5409 def : Pat<(v4f32 (bitconvert (v4i32 FPR128:$src))), (v4f32 FPR128:$src)>;
5411 let Predicates = [IsLE] in {
5412 def : Pat<(v2i64 (bitconvert (f128 FPR128:$src))), (v2i64 FPR128:$src)>;
5413 def : Pat<(v2i64 (bitconvert (v4i32 FPR128:$src))), (v2i64 FPR128:$src)>;
5414 def : Pat<(v2i64 (bitconvert (v8i16 FPR128:$src))), (v2i64 FPR128:$src)>;
5415 def : Pat<(v2i64 (bitconvert (v16i8 FPR128:$src))), (v2i64 FPR128:$src)>;
5416 def : Pat<(v2i64 (bitconvert (v4f32 FPR128:$src))), (v2i64 FPR128:$src)>;
5417 def : Pat<(v2i64 (bitconvert (v8f16 FPR128:$src))), (v2i64 FPR128:$src)>;
5419 let Predicates = [IsBE] in {
5420 def : Pat<(v2i64 (bitconvert (f128 FPR128:$src))),
5421 (v2i64 (EXTv16i8 FPR128:$src,
5422 FPR128:$src, (i32 8)))>;
5423 def : Pat<(v2i64 (bitconvert (v4i32 FPR128:$src))),
5424 (v2i64 (REV64v4i32 FPR128:$src))>;
5425 def : Pat<(v2i64 (bitconvert (v8i16 FPR128:$src))),
5426 (v2i64 (REV64v8i16 FPR128:$src))>;
5427 def : Pat<(v2i64 (bitconvert (v16i8 FPR128:$src))),
5428 (v2i64 (REV64v16i8 FPR128:$src))>;
5429 def : Pat<(v2i64 (bitconvert (v4f32 FPR128:$src))),
5430 (v2i64 (REV64v4i32 FPR128:$src))>;
5431 def : Pat<(v2i64 (bitconvert (v8f16 FPR128:$src))),
5432 (v2i64 (REV64v8i16 FPR128:$src))>;
5434 def : Pat<(v2i64 (bitconvert (v2f64 FPR128:$src))), (v2i64 FPR128:$src)>;
5436 let Predicates = [IsLE] in {
5437 def : Pat<(v4i32 (bitconvert (f128 FPR128:$src))), (v4i32 FPR128:$src)>;
5438 def : Pat<(v4i32 (bitconvert (v2i64 FPR128:$src))), (v4i32 FPR128:$src)>;
5439 def : Pat<(v4i32 (bitconvert (v8i16 FPR128:$src))), (v4i32 FPR128:$src)>;
5440 def : Pat<(v4i32 (bitconvert (v16i8 FPR128:$src))), (v4i32 FPR128:$src)>;
5441 def : Pat<(v4i32 (bitconvert (v2f64 FPR128:$src))), (v4i32 FPR128:$src)>;
5442 def : Pat<(v4i32 (bitconvert (v8f16 FPR128:$src))), (v4i32 FPR128:$src)>;
5444 let Predicates = [IsBE] in {
5445 def : Pat<(v4i32 (bitconvert (f128 FPR128:$src))),
5446 (v4i32 (EXTv16i8 (REV64v4i32 FPR128:$src),
5447 (REV64v4i32 FPR128:$src),
5449 def : Pat<(v4i32 (bitconvert (v2i64 FPR128:$src))),
5450 (v4i32 (REV64v4i32 FPR128:$src))>;
5451 def : Pat<(v4i32 (bitconvert (v8i16 FPR128:$src))),
5452 (v4i32 (REV32v8i16 FPR128:$src))>;
5453 def : Pat<(v4i32 (bitconvert (v16i8 FPR128:$src))),
5454 (v4i32 (REV32v16i8 FPR128:$src))>;
5455 def : Pat<(v4i32 (bitconvert (v2f64 FPR128:$src))),
5456 (v4i32 (REV64v4i32 FPR128:$src))>;
5457 def : Pat<(v4i32 (bitconvert (v8f16 FPR128:$src))),
5458 (v4i32 (REV32v8i16 FPR128:$src))>;
5460 def : Pat<(v4i32 (bitconvert (v4f32 FPR128:$src))), (v4i32 FPR128:$src)>;
5462 let Predicates = [IsLE] in {
5463 def : Pat<(v8i16 (bitconvert (f128 FPR128:$src))), (v8i16 FPR128:$src)>;
5464 def : Pat<(v8i16 (bitconvert (v2i64 FPR128:$src))), (v8i16 FPR128:$src)>;
5465 def : Pat<(v8i16 (bitconvert (v4i32 FPR128:$src))), (v8i16 FPR128:$src)>;
5466 def : Pat<(v8i16 (bitconvert (v16i8 FPR128:$src))), (v8i16 FPR128:$src)>;
5467 def : Pat<(v8i16 (bitconvert (v2f64 FPR128:$src))), (v8i16 FPR128:$src)>;
5468 def : Pat<(v8i16 (bitconvert (v4f32 FPR128:$src))), (v8i16 FPR128:$src)>;
5469 def : Pat<(v8i16 (bitconvert (v8f16 FPR128:$src))), (v8i16 FPR128:$src)>;
5471 let Predicates = [IsBE] in {
5472 def : Pat<(v8i16 (bitconvert (f128 FPR128:$src))),
5473 (v8i16 (EXTv16i8 (REV64v8i16 FPR128:$src),
5474 (REV64v8i16 FPR128:$src),
5476 def : Pat<(v8i16 (bitconvert (v2i64 FPR128:$src))),
5477 (v8i16 (REV64v8i16 FPR128:$src))>;
5478 def : Pat<(v8i16 (bitconvert (v4i32 FPR128:$src))),
5479 (v8i16 (REV32v8i16 FPR128:$src))>;
5480 def : Pat<(v8i16 (bitconvert (v16i8 FPR128:$src))),
5481 (v8i16 (REV16v16i8 FPR128:$src))>;
5482 def : Pat<(v8i16 (bitconvert (v2f64 FPR128:$src))),
5483 (v8i16 (REV64v8i16 FPR128:$src))>;
5484 def : Pat<(v8i16 (bitconvert (v4f32 FPR128:$src))),
5485 (v8i16 (REV32v8i16 FPR128:$src))>;
5486 def : Pat<(v8i16 (bitconvert (v8f16 FPR128:$src))),
5487 (v8i16 (REV32v8i16 FPR128:$src))>;
5490 let Predicates = [IsLE] in {
5491 def : Pat<(v8f16 (bitconvert (f128 FPR128:$src))), (v8f16 FPR128:$src)>;
5492 def : Pat<(v8f16 (bitconvert (v2i64 FPR128:$src))), (v8f16 FPR128:$src)>;
5493 def : Pat<(v8f16 (bitconvert (v4i32 FPR128:$src))), (v8f16 FPR128:$src)>;
5494 def : Pat<(v8f16 (bitconvert (v8i16 FPR128:$src))), (v8f16 FPR128:$src)>;
5495 def : Pat<(v8f16 (bitconvert (v16i8 FPR128:$src))), (v8f16 FPR128:$src)>;
5496 def : Pat<(v8f16 (bitconvert (v2f64 FPR128:$src))), (v8f16 FPR128:$src)>;
5497 def : Pat<(v8f16 (bitconvert (v4f32 FPR128:$src))), (v8f16 FPR128:$src)>;
5499 let Predicates = [IsBE] in {
5500 def : Pat<(v8f16 (bitconvert (f128 FPR128:$src))),
5501 (v8f16 (EXTv16i8 (REV64v8i16 FPR128:$src),
5502 (REV64v8i16 FPR128:$src),
5504 def : Pat<(v8f16 (bitconvert (v2i64 FPR128:$src))),
5505 (v8f16 (REV64v8i16 FPR128:$src))>;
5506 def : Pat<(v8f16 (bitconvert (v4i32 FPR128:$src))),
5507 (v8f16 (REV32v8i16 FPR128:$src))>;
5508 def : Pat<(v8f16 (bitconvert (v8i16 FPR128:$src))),
5509 (v8f16 (REV64v8i16 FPR128:$src))>;
5510 def : Pat<(v8f16 (bitconvert (v16i8 FPR128:$src))),
5511 (v8f16 (REV16v16i8 FPR128:$src))>;
5512 def : Pat<(v8f16 (bitconvert (v2f64 FPR128:$src))),
5513 (v8f16 (REV64v8i16 FPR128:$src))>;
5514 def : Pat<(v8f16 (bitconvert (v4f32 FPR128:$src))),
5515 (v8f16 (REV32v8i16 FPR128:$src))>;
5518 let Predicates = [IsLE] in {
5519 def : Pat<(v16i8 (bitconvert (f128 FPR128:$src))), (v16i8 FPR128:$src)>;
5520 def : Pat<(v16i8 (bitconvert (v2i64 FPR128:$src))), (v16i8 FPR128:$src)>;
5521 def : Pat<(v16i8 (bitconvert (v4i32 FPR128:$src))), (v16i8 FPR128:$src)>;
5522 def : Pat<(v16i8 (bitconvert (v8i16 FPR128:$src))), (v16i8 FPR128:$src)>;
5523 def : Pat<(v16i8 (bitconvert (v2f64 FPR128:$src))), (v16i8 FPR128:$src)>;
5524 def : Pat<(v16i8 (bitconvert (v4f32 FPR128:$src))), (v16i8 FPR128:$src)>;
5525 def : Pat<(v16i8 (bitconvert (v8f16 FPR128:$src))), (v16i8 FPR128:$src)>;
5527 let Predicates = [IsBE] in {
5528 def : Pat<(v16i8 (bitconvert (f128 FPR128:$src))),
5529 (v16i8 (EXTv16i8 (REV64v16i8 FPR128:$src),
5530 (REV64v16i8 FPR128:$src),
5532 def : Pat<(v16i8 (bitconvert (v2i64 FPR128:$src))),
5533 (v16i8 (REV64v16i8 FPR128:$src))>;
5534 def : Pat<(v16i8 (bitconvert (v4i32 FPR128:$src))),
5535 (v16i8 (REV32v16i8 FPR128:$src))>;
5536 def : Pat<(v16i8 (bitconvert (v8i16 FPR128:$src))),
5537 (v16i8 (REV16v16i8 FPR128:$src))>;
5538 def : Pat<(v16i8 (bitconvert (v2f64 FPR128:$src))),
5539 (v16i8 (REV64v16i8 FPR128:$src))>;
5540 def : Pat<(v16i8 (bitconvert (v4f32 FPR128:$src))),
5541 (v16i8 (REV32v16i8 FPR128:$src))>;
5542 def : Pat<(v16i8 (bitconvert (v8f16 FPR128:$src))),
5543 (v16i8 (REV16v16i8 FPR128:$src))>;
5546 def : Pat<(v8i8 (extract_subvector (v16i8 FPR128:$Rn), (i64 1))),
5547 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
5548 def : Pat<(v4i16 (extract_subvector (v8i16 FPR128:$Rn), (i64 1))),
5549 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
5550 def : Pat<(v2i32 (extract_subvector (v4i32 FPR128:$Rn), (i64 1))),
5551 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
5552 def : Pat<(v1i64 (extract_subvector (v2i64 FPR128:$Rn), (i64 1))),
5553 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
5555 // A 64-bit subvector insert to the first 128-bit vector position
5556 // is a subregister copy that needs no instruction.
5557 def : Pat<(insert_subvector undef, (v1i64 FPR64:$src), (i32 0)),
5558 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5559 def : Pat<(insert_subvector undef, (v1f64 FPR64:$src), (i32 0)),
5560 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5561 def : Pat<(insert_subvector undef, (v2i32 FPR64:$src), (i32 0)),
5562 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5563 def : Pat<(insert_subvector undef, (v2f32 FPR64:$src), (i32 0)),
5564 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5565 def : Pat<(insert_subvector undef, (v4i16 FPR64:$src), (i32 0)),
5566 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5567 def : Pat<(insert_subvector undef, (v4f16 FPR64:$src), (i32 0)),
5568 (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5569 def : Pat<(insert_subvector undef, (v8i8 FPR64:$src), (i32 0)),
5570 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5572 // Use pair-wise add instructions when summing up the lanes for v2f64, v2i64
5574 def : Pat<(i64 (add (vector_extract (v2i64 FPR128:$Rn), (i64 0)),
5575 (vector_extract (v2i64 FPR128:$Rn), (i64 1)))),
5576 (i64 (ADDPv2i64p (v2i64 FPR128:$Rn)))>;
5577 def : Pat<(f64 (fadd (vector_extract (v2f64 FPR128:$Rn), (i64 0)),
5578 (vector_extract (v2f64 FPR128:$Rn), (i64 1)))),
5579 (f64 (FADDPv2i64p (v2f64 FPR128:$Rn)))>;
5580 // vector_extract on 64-bit vectors gets promoted to a 128 bit vector,
5581 // so we match on v4f32 here, not v2f32. This will also catch adding
5582 // the low two lanes of a true v4f32 vector.
5583 def : Pat<(fadd (vector_extract (v4f32 FPR128:$Rn), (i64 0)),
5584 (vector_extract (v4f32 FPR128:$Rn), (i64 1))),
5585 (f32 (FADDPv2i32p (EXTRACT_SUBREG FPR128:$Rn, dsub)))>;
5587 // Scalar 64-bit shifts in FPR64 registers.
5588 def : Pat<(i64 (int_aarch64_neon_sshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
5589 (SSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
5590 def : Pat<(i64 (int_aarch64_neon_ushl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
5591 (USHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
5592 def : Pat<(i64 (int_aarch64_neon_srshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
5593 (SRSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
5594 def : Pat<(i64 (int_aarch64_neon_urshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
5595 (URSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
5597 // Tail call return handling. These are all compiler pseudo-instructions,
5598 // so no encoding information or anything like that.
5599 let isCall = 1, isTerminator = 1, isReturn = 1, isBarrier = 1, Uses = [SP] in {
5600 def TCRETURNdi : Pseudo<(outs), (ins i64imm:$dst, i32imm:$FPDiff),[]>;
5601 def TCRETURNri : Pseudo<(outs), (ins tcGPR64:$dst, i32imm:$FPDiff), []>;
5604 def : Pat<(AArch64tcret tcGPR64:$dst, (i32 timm:$FPDiff)),
5605 (TCRETURNri tcGPR64:$dst, imm:$FPDiff)>;
5606 def : Pat<(AArch64tcret tglobaladdr:$dst, (i32 timm:$FPDiff)),
5607 (TCRETURNdi texternalsym:$dst, imm:$FPDiff)>;
5608 def : Pat<(AArch64tcret texternalsym:$dst, (i32 timm:$FPDiff)),
5609 (TCRETURNdi texternalsym:$dst, imm:$FPDiff)>;
5611 include "AArch64InstrAtomics.td"