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:$src)), sub_32)>;
484 // Materialize FP constants via MOVi32imm/MOVi64imm (MachO large code model).
485 def bitcast_fpimm_to_i32 : SDNodeXForm<fpimm, [{
486 return CurDAG->getTargetConstant(
487 N->getValueAPF().bitcastToAPInt().getZExtValue(), MVT::i32);
490 def bitcast_fpimm_to_i64 : SDNodeXForm<fpimm, [{
491 return CurDAG->getTargetConstant(
492 N->getValueAPF().bitcastToAPInt().getZExtValue(), MVT::i64);
496 def : Pat<(f32 fpimm:$in),
497 (COPY_TO_REGCLASS (MOVi32imm (bitcast_fpimm_to_i32 f32:$in)), FPR32)>;
498 def : Pat<(f64 fpimm:$in),
499 (COPY_TO_REGCLASS (MOVi64imm (bitcast_fpimm_to_i64 f64:$in)), FPR64)>;
502 // Deal with the various forms of (ELF) large addressing with MOVZ/MOVK
504 def : Pat<(AArch64WrapperLarge tglobaladdr:$g3, tglobaladdr:$g2,
505 tglobaladdr:$g1, tglobaladdr:$g0),
506 (MOVKXi (MOVKXi (MOVKXi (MOVZXi tglobaladdr:$g3, 48),
507 tglobaladdr:$g2, 32),
508 tglobaladdr:$g1, 16),
509 tglobaladdr:$g0, 0)>;
511 def : Pat<(AArch64WrapperLarge tblockaddress:$g3, tblockaddress:$g2,
512 tblockaddress:$g1, tblockaddress:$g0),
513 (MOVKXi (MOVKXi (MOVKXi (MOVZXi tblockaddress:$g3, 48),
514 tblockaddress:$g2, 32),
515 tblockaddress:$g1, 16),
516 tblockaddress:$g0, 0)>;
518 def : Pat<(AArch64WrapperLarge tconstpool:$g3, tconstpool:$g2,
519 tconstpool:$g1, tconstpool:$g0),
520 (MOVKXi (MOVKXi (MOVKXi (MOVZXi tconstpool:$g3, 48),
525 def : Pat<(AArch64WrapperLarge tjumptable:$g3, tjumptable:$g2,
526 tjumptable:$g1, tjumptable:$g0),
527 (MOVKXi (MOVKXi (MOVKXi (MOVZXi tjumptable:$g3, 48),
533 //===----------------------------------------------------------------------===//
534 // Arithmetic instructions.
535 //===----------------------------------------------------------------------===//
537 // Add/subtract with carry.
538 defm ADC : AddSubCarry<0, "adc", "adcs", AArch64adc, AArch64adc_flag>;
539 defm SBC : AddSubCarry<1, "sbc", "sbcs", AArch64sbc, AArch64sbc_flag>;
541 def : InstAlias<"ngc $dst, $src", (SBCWr GPR32:$dst, WZR, GPR32:$src)>;
542 def : InstAlias<"ngc $dst, $src", (SBCXr GPR64:$dst, XZR, GPR64:$src)>;
543 def : InstAlias<"ngcs $dst, $src", (SBCSWr GPR32:$dst, WZR, GPR32:$src)>;
544 def : InstAlias<"ngcs $dst, $src", (SBCSXr GPR64:$dst, XZR, GPR64:$src)>;
547 defm ADD : AddSub<0, "add", add>;
548 defm SUB : AddSub<1, "sub">;
550 def : InstAlias<"mov $dst, $src",
551 (ADDWri GPR32sponly:$dst, GPR32sp:$src, 0, 0)>;
552 def : InstAlias<"mov $dst, $src",
553 (ADDWri GPR32sp:$dst, GPR32sponly:$src, 0, 0)>;
554 def : InstAlias<"mov $dst, $src",
555 (ADDXri GPR64sponly:$dst, GPR64sp:$src, 0, 0)>;
556 def : InstAlias<"mov $dst, $src",
557 (ADDXri GPR64sp:$dst, GPR64sponly:$src, 0, 0)>;
559 defm ADDS : AddSubS<0, "adds", AArch64add_flag, "cmn">;
560 defm SUBS : AddSubS<1, "subs", AArch64sub_flag, "cmp">;
562 // Use SUBS instead of SUB to enable CSE between SUBS and SUB.
563 def : Pat<(sub GPR32sp:$Rn, addsub_shifted_imm32:$imm),
564 (SUBSWri GPR32sp:$Rn, addsub_shifted_imm32:$imm)>;
565 def : Pat<(sub GPR64sp:$Rn, addsub_shifted_imm64:$imm),
566 (SUBSXri GPR64sp:$Rn, addsub_shifted_imm64:$imm)>;
567 def : Pat<(sub GPR32:$Rn, GPR32:$Rm),
568 (SUBSWrr GPR32:$Rn, GPR32:$Rm)>;
569 def : Pat<(sub GPR64:$Rn, GPR64:$Rm),
570 (SUBSXrr GPR64:$Rn, GPR64:$Rm)>;
571 def : Pat<(sub GPR32:$Rn, arith_shifted_reg32:$Rm),
572 (SUBSWrs GPR32:$Rn, arith_shifted_reg32:$Rm)>;
573 def : Pat<(sub GPR64:$Rn, arith_shifted_reg64:$Rm),
574 (SUBSXrs GPR64:$Rn, arith_shifted_reg64:$Rm)>;
575 def : Pat<(sub GPR32sp:$R2, arith_extended_reg32<i32>:$R3),
576 (SUBSWrx GPR32sp:$R2, arith_extended_reg32<i32>:$R3)>;
577 def : Pat<(sub GPR64sp:$R2, arith_extended_reg32to64<i64>:$R3),
578 (SUBSXrx GPR64sp:$R2, arith_extended_reg32to64<i64>:$R3)>;
580 // Because of the immediate format for add/sub-imm instructions, the
581 // expression (add x, -1) must be transformed to (SUB{W,X}ri x, 1).
582 // These patterns capture that transformation.
583 let AddedComplexity = 1 in {
584 def : Pat<(add GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
585 (SUBSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
586 def : Pat<(add GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
587 (SUBSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
588 def : Pat<(sub GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
589 (ADDWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
590 def : Pat<(sub GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
591 (ADDXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
594 // Because of the immediate format for add/sub-imm instructions, the
595 // expression (add x, -1) must be transformed to (SUB{W,X}ri x, 1).
596 // These patterns capture that transformation.
597 let AddedComplexity = 1 in {
598 def : Pat<(AArch64add_flag GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
599 (SUBSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
600 def : Pat<(AArch64add_flag GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
601 (SUBSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
602 def : Pat<(AArch64sub_flag GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
603 (ADDSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
604 def : Pat<(AArch64sub_flag GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
605 (ADDSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
608 def : InstAlias<"neg $dst, $src", (SUBWrs GPR32:$dst, WZR, GPR32:$src, 0), 3>;
609 def : InstAlias<"neg $dst, $src", (SUBXrs GPR64:$dst, XZR, GPR64:$src, 0), 3>;
610 def : InstAlias<"neg $dst, $src$shift",
611 (SUBWrs GPR32:$dst, WZR, GPR32:$src, arith_shift32:$shift), 2>;
612 def : InstAlias<"neg $dst, $src$shift",
613 (SUBXrs GPR64:$dst, XZR, GPR64:$src, arith_shift64:$shift), 2>;
615 def : InstAlias<"negs $dst, $src", (SUBSWrs GPR32:$dst, WZR, GPR32:$src, 0), 3>;
616 def : InstAlias<"negs $dst, $src", (SUBSXrs GPR64:$dst, XZR, GPR64:$src, 0), 3>;
617 def : InstAlias<"negs $dst, $src$shift",
618 (SUBSWrs GPR32:$dst, WZR, GPR32:$src, arith_shift32:$shift), 2>;
619 def : InstAlias<"negs $dst, $src$shift",
620 (SUBSXrs GPR64:$dst, XZR, GPR64:$src, arith_shift64:$shift), 2>;
623 // Unsigned/Signed divide
624 defm UDIV : Div<0, "udiv", udiv>;
625 defm SDIV : Div<1, "sdiv", sdiv>;
626 let isCodeGenOnly = 1 in {
627 defm UDIV_Int : Div<0, "udiv", int_aarch64_udiv>;
628 defm SDIV_Int : Div<1, "sdiv", int_aarch64_sdiv>;
632 defm ASRV : Shift<0b10, "asr", sra>;
633 defm LSLV : Shift<0b00, "lsl", shl>;
634 defm LSRV : Shift<0b01, "lsr", srl>;
635 defm RORV : Shift<0b11, "ror", rotr>;
637 def : ShiftAlias<"asrv", ASRVWr, GPR32>;
638 def : ShiftAlias<"asrv", ASRVXr, GPR64>;
639 def : ShiftAlias<"lslv", LSLVWr, GPR32>;
640 def : ShiftAlias<"lslv", LSLVXr, GPR64>;
641 def : ShiftAlias<"lsrv", LSRVWr, GPR32>;
642 def : ShiftAlias<"lsrv", LSRVXr, GPR64>;
643 def : ShiftAlias<"rorv", RORVWr, GPR32>;
644 def : ShiftAlias<"rorv", RORVXr, GPR64>;
647 let AddedComplexity = 7 in {
648 defm MADD : MulAccum<0, "madd", add>;
649 defm MSUB : MulAccum<1, "msub", sub>;
651 def : Pat<(i32 (mul GPR32:$Rn, GPR32:$Rm)),
652 (MADDWrrr GPR32:$Rn, GPR32:$Rm, WZR)>;
653 def : Pat<(i64 (mul GPR64:$Rn, GPR64:$Rm)),
654 (MADDXrrr GPR64:$Rn, GPR64:$Rm, XZR)>;
656 def : Pat<(i32 (ineg (mul GPR32:$Rn, GPR32:$Rm))),
657 (MSUBWrrr GPR32:$Rn, GPR32:$Rm, WZR)>;
658 def : Pat<(i64 (ineg (mul GPR64:$Rn, GPR64:$Rm))),
659 (MSUBXrrr GPR64:$Rn, GPR64:$Rm, XZR)>;
660 def : Pat<(i32 (mul (ineg GPR32:$Rn), GPR32:$Rm)),
661 (MSUBWrrr GPR32:$Rn, GPR32:$Rm, WZR)>;
662 def : Pat<(i64 (mul (ineg GPR64:$Rn), GPR64:$Rm)),
663 (MSUBXrrr GPR64:$Rn, GPR64:$Rm, XZR)>;
664 } // AddedComplexity = 7
666 let AddedComplexity = 5 in {
667 def SMADDLrrr : WideMulAccum<0, 0b001, "smaddl", add, sext>;
668 def SMSUBLrrr : WideMulAccum<1, 0b001, "smsubl", sub, sext>;
669 def UMADDLrrr : WideMulAccum<0, 0b101, "umaddl", add, zext>;
670 def UMSUBLrrr : WideMulAccum<1, 0b101, "umsubl", sub, zext>;
672 def : Pat<(i64 (mul (sext GPR32:$Rn), (sext GPR32:$Rm))),
673 (SMADDLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
674 def : Pat<(i64 (mul (zext GPR32:$Rn), (zext GPR32:$Rm))),
675 (UMADDLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
677 def : Pat<(i64 (ineg (mul (sext GPR32:$Rn), (sext GPR32:$Rm)))),
678 (SMSUBLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
679 def : Pat<(i64 (ineg (mul (zext GPR32:$Rn), (zext GPR32:$Rm)))),
680 (UMSUBLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
681 } // AddedComplexity = 5
683 def : MulAccumWAlias<"mul", MADDWrrr>;
684 def : MulAccumXAlias<"mul", MADDXrrr>;
685 def : MulAccumWAlias<"mneg", MSUBWrrr>;
686 def : MulAccumXAlias<"mneg", MSUBXrrr>;
687 def : WideMulAccumAlias<"smull", SMADDLrrr>;
688 def : WideMulAccumAlias<"smnegl", SMSUBLrrr>;
689 def : WideMulAccumAlias<"umull", UMADDLrrr>;
690 def : WideMulAccumAlias<"umnegl", UMSUBLrrr>;
693 def SMULHrr : MulHi<0b010, "smulh", mulhs>;
694 def UMULHrr : MulHi<0b110, "umulh", mulhu>;
697 def CRC32Brr : BaseCRC32<0, 0b00, 0, GPR32, int_aarch64_crc32b, "crc32b">;
698 def CRC32Hrr : BaseCRC32<0, 0b01, 0, GPR32, int_aarch64_crc32h, "crc32h">;
699 def CRC32Wrr : BaseCRC32<0, 0b10, 0, GPR32, int_aarch64_crc32w, "crc32w">;
700 def CRC32Xrr : BaseCRC32<1, 0b11, 0, GPR64, int_aarch64_crc32x, "crc32x">;
702 def CRC32CBrr : BaseCRC32<0, 0b00, 1, GPR32, int_aarch64_crc32cb, "crc32cb">;
703 def CRC32CHrr : BaseCRC32<0, 0b01, 1, GPR32, int_aarch64_crc32ch, "crc32ch">;
704 def CRC32CWrr : BaseCRC32<0, 0b10, 1, GPR32, int_aarch64_crc32cw, "crc32cw">;
705 def CRC32CXrr : BaseCRC32<1, 0b11, 1, GPR64, int_aarch64_crc32cx, "crc32cx">;
708 //===----------------------------------------------------------------------===//
709 // Logical instructions.
710 //===----------------------------------------------------------------------===//
713 defm ANDS : LogicalImmS<0b11, "ands", AArch64and_flag, "bics">;
714 defm AND : LogicalImm<0b00, "and", and, "bic">;
715 defm EOR : LogicalImm<0b10, "eor", xor, "eon">;
716 defm ORR : LogicalImm<0b01, "orr", or, "orn">;
718 // FIXME: these aliases *are* canonical sometimes (when movz can't be
719 // used). Actually, it seems to be working right now, but putting logical_immXX
720 // here is a bit dodgy on the AsmParser side too.
721 def : InstAlias<"mov $dst, $imm", (ORRWri GPR32sp:$dst, WZR,
722 logical_imm32:$imm), 0>;
723 def : InstAlias<"mov $dst, $imm", (ORRXri GPR64sp:$dst, XZR,
724 logical_imm64:$imm), 0>;
728 defm ANDS : LogicalRegS<0b11, 0, "ands", AArch64and_flag>;
729 defm BICS : LogicalRegS<0b11, 1, "bics",
730 BinOpFrag<(AArch64and_flag node:$LHS, (not node:$RHS))>>;
731 defm AND : LogicalReg<0b00, 0, "and", and>;
732 defm BIC : LogicalReg<0b00, 1, "bic",
733 BinOpFrag<(and node:$LHS, (not node:$RHS))>>;
734 defm EON : LogicalReg<0b10, 1, "eon",
735 BinOpFrag<(xor node:$LHS, (not node:$RHS))>>;
736 defm EOR : LogicalReg<0b10, 0, "eor", xor>;
737 defm ORN : LogicalReg<0b01, 1, "orn",
738 BinOpFrag<(or node:$LHS, (not node:$RHS))>>;
739 defm ORR : LogicalReg<0b01, 0, "orr", or>;
741 def : InstAlias<"mov $dst, $src", (ORRWrs GPR32:$dst, WZR, GPR32:$src, 0), 2>;
742 def : InstAlias<"mov $dst, $src", (ORRXrs GPR64:$dst, XZR, GPR64:$src, 0), 2>;
744 def : InstAlias<"mvn $Wd, $Wm", (ORNWrs GPR32:$Wd, WZR, GPR32:$Wm, 0), 3>;
745 def : InstAlias<"mvn $Xd, $Xm", (ORNXrs GPR64:$Xd, XZR, GPR64:$Xm, 0), 3>;
747 def : InstAlias<"mvn $Wd, $Wm$sh",
748 (ORNWrs GPR32:$Wd, WZR, GPR32:$Wm, logical_shift32:$sh), 2>;
749 def : InstAlias<"mvn $Xd, $Xm$sh",
750 (ORNXrs GPR64:$Xd, XZR, GPR64:$Xm, logical_shift64:$sh), 2>;
752 def : InstAlias<"tst $src1, $src2",
753 (ANDSWri WZR, GPR32:$src1, logical_imm32:$src2), 2>;
754 def : InstAlias<"tst $src1, $src2",
755 (ANDSXri XZR, GPR64:$src1, logical_imm64:$src2), 2>;
757 def : InstAlias<"tst $src1, $src2",
758 (ANDSWrs WZR, GPR32:$src1, GPR32:$src2, 0), 3>;
759 def : InstAlias<"tst $src1, $src2",
760 (ANDSXrs XZR, GPR64:$src1, GPR64:$src2, 0), 3>;
762 def : InstAlias<"tst $src1, $src2$sh",
763 (ANDSWrs WZR, GPR32:$src1, GPR32:$src2, logical_shift32:$sh), 2>;
764 def : InstAlias<"tst $src1, $src2$sh",
765 (ANDSXrs XZR, GPR64:$src1, GPR64:$src2, logical_shift64:$sh), 2>;
768 def : Pat<(not GPR32:$Wm), (ORNWrr WZR, GPR32:$Wm)>;
769 def : Pat<(not GPR64:$Xm), (ORNXrr XZR, GPR64:$Xm)>;
772 //===----------------------------------------------------------------------===//
773 // One operand data processing instructions.
774 //===----------------------------------------------------------------------===//
776 defm CLS : OneOperandData<0b101, "cls">;
777 defm CLZ : OneOperandData<0b100, "clz", ctlz>;
778 defm RBIT : OneOperandData<0b000, "rbit">;
780 def : Pat<(int_aarch64_rbit GPR32:$Rn), (RBITWr $Rn)>;
781 def : Pat<(int_aarch64_rbit GPR64:$Rn), (RBITXr $Rn)>;
783 def REV16Wr : OneWRegData<0b001, "rev16",
784 UnOpFrag<(rotr (bswap node:$LHS), (i64 16))>>;
785 def REV16Xr : OneXRegData<0b001, "rev16", null_frag>;
787 def : Pat<(cttz GPR32:$Rn),
788 (CLZWr (RBITWr GPR32:$Rn))>;
789 def : Pat<(cttz GPR64:$Rn),
790 (CLZXr (RBITXr GPR64:$Rn))>;
791 def : Pat<(ctlz (or (shl (xor (sra GPR32:$Rn, (i64 31)), GPR32:$Rn), (i64 1)),
794 def : Pat<(ctlz (or (shl (xor (sra GPR64:$Rn, (i64 63)), GPR64:$Rn), (i64 1)),
798 // Unlike the other one operand instructions, the instructions with the "rev"
799 // mnemonic do *not* just different in the size bit, but actually use different
800 // opcode bits for the different sizes.
801 def REVWr : OneWRegData<0b010, "rev", bswap>;
802 def REVXr : OneXRegData<0b011, "rev", bswap>;
803 def REV32Xr : OneXRegData<0b010, "rev32",
804 UnOpFrag<(rotr (bswap node:$LHS), (i64 32))>>;
806 // The bswap commutes with the rotr so we want a pattern for both possible
808 def : Pat<(bswap (rotr GPR32:$Rn, (i64 16))), (REV16Wr GPR32:$Rn)>;
809 def : Pat<(bswap (rotr GPR64:$Rn, (i64 32))), (REV32Xr GPR64:$Rn)>;
811 //===----------------------------------------------------------------------===//
812 // Bitfield immediate extraction instruction.
813 //===----------------------------------------------------------------------===//
814 let hasSideEffects = 0 in
815 defm EXTR : ExtractImm<"extr">;
816 def : InstAlias<"ror $dst, $src, $shift",
817 (EXTRWrri GPR32:$dst, GPR32:$src, GPR32:$src, imm0_31:$shift)>;
818 def : InstAlias<"ror $dst, $src, $shift",
819 (EXTRXrri GPR64:$dst, GPR64:$src, GPR64:$src, imm0_63:$shift)>;
821 def : Pat<(rotr GPR32:$Rn, (i64 imm0_31:$imm)),
822 (EXTRWrri GPR32:$Rn, GPR32:$Rn, imm0_31:$imm)>;
823 def : Pat<(rotr GPR64:$Rn, (i64 imm0_63:$imm)),
824 (EXTRXrri GPR64:$Rn, GPR64:$Rn, imm0_63:$imm)>;
826 //===----------------------------------------------------------------------===//
827 // Other bitfield immediate instructions.
828 //===----------------------------------------------------------------------===//
829 let hasSideEffects = 0 in {
830 defm BFM : BitfieldImmWith2RegArgs<0b01, "bfm">;
831 defm SBFM : BitfieldImm<0b00, "sbfm">;
832 defm UBFM : BitfieldImm<0b10, "ubfm">;
835 def i32shift_a : Operand<i64>, SDNodeXForm<imm, [{
836 uint64_t enc = (32 - N->getZExtValue()) & 0x1f;
837 return CurDAG->getTargetConstant(enc, MVT::i64);
840 def i32shift_b : Operand<i64>, SDNodeXForm<imm, [{
841 uint64_t enc = 31 - N->getZExtValue();
842 return CurDAG->getTargetConstant(enc, MVT::i64);
845 // min(7, 31 - shift_amt)
846 def i32shift_sext_i8 : Operand<i64>, SDNodeXForm<imm, [{
847 uint64_t enc = 31 - N->getZExtValue();
848 enc = enc > 7 ? 7 : enc;
849 return CurDAG->getTargetConstant(enc, MVT::i64);
852 // min(15, 31 - shift_amt)
853 def i32shift_sext_i16 : Operand<i64>, SDNodeXForm<imm, [{
854 uint64_t enc = 31 - N->getZExtValue();
855 enc = enc > 15 ? 15 : enc;
856 return CurDAG->getTargetConstant(enc, MVT::i64);
859 def i64shift_a : Operand<i64>, SDNodeXForm<imm, [{
860 uint64_t enc = (64 - N->getZExtValue()) & 0x3f;
861 return CurDAG->getTargetConstant(enc, MVT::i64);
864 def i64shift_b : Operand<i64>, SDNodeXForm<imm, [{
865 uint64_t enc = 63 - N->getZExtValue();
866 return CurDAG->getTargetConstant(enc, MVT::i64);
869 // min(7, 63 - shift_amt)
870 def i64shift_sext_i8 : Operand<i64>, SDNodeXForm<imm, [{
871 uint64_t enc = 63 - N->getZExtValue();
872 enc = enc > 7 ? 7 : enc;
873 return CurDAG->getTargetConstant(enc, MVT::i64);
876 // min(15, 63 - shift_amt)
877 def i64shift_sext_i16 : Operand<i64>, SDNodeXForm<imm, [{
878 uint64_t enc = 63 - N->getZExtValue();
879 enc = enc > 15 ? 15 : enc;
880 return CurDAG->getTargetConstant(enc, MVT::i64);
883 // min(31, 63 - shift_amt)
884 def i64shift_sext_i32 : Operand<i64>, SDNodeXForm<imm, [{
885 uint64_t enc = 63 - N->getZExtValue();
886 enc = enc > 31 ? 31 : enc;
887 return CurDAG->getTargetConstant(enc, MVT::i64);
890 def : Pat<(shl GPR32:$Rn, (i64 imm0_31:$imm)),
891 (UBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
892 (i64 (i32shift_b imm0_31:$imm)))>;
893 def : Pat<(shl GPR64:$Rn, (i64 imm0_63:$imm)),
894 (UBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
895 (i64 (i64shift_b imm0_63:$imm)))>;
897 let AddedComplexity = 10 in {
898 def : Pat<(sra GPR32:$Rn, (i64 imm0_31:$imm)),
899 (SBFMWri GPR32:$Rn, imm0_31:$imm, 31)>;
900 def : Pat<(sra GPR64:$Rn, (i64 imm0_63:$imm)),
901 (SBFMXri GPR64:$Rn, imm0_63:$imm, 63)>;
904 def : InstAlias<"asr $dst, $src, $shift",
905 (SBFMWri GPR32:$dst, GPR32:$src, imm0_31:$shift, 31)>;
906 def : InstAlias<"asr $dst, $src, $shift",
907 (SBFMXri GPR64:$dst, GPR64:$src, imm0_63:$shift, 63)>;
908 def : InstAlias<"sxtb $dst, $src", (SBFMWri GPR32:$dst, GPR32:$src, 0, 7)>;
909 def : InstAlias<"sxtb $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 7)>;
910 def : InstAlias<"sxth $dst, $src", (SBFMWri GPR32:$dst, GPR32:$src, 0, 15)>;
911 def : InstAlias<"sxth $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 15)>;
912 def : InstAlias<"sxtw $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 31)>;
914 def : Pat<(srl GPR32:$Rn, (i64 imm0_31:$imm)),
915 (UBFMWri GPR32:$Rn, imm0_31:$imm, 31)>;
916 def : Pat<(srl GPR64:$Rn, (i64 imm0_63:$imm)),
917 (UBFMXri GPR64:$Rn, imm0_63:$imm, 63)>;
919 def : InstAlias<"lsr $dst, $src, $shift",
920 (UBFMWri GPR32:$dst, GPR32:$src, imm0_31:$shift, 31)>;
921 def : InstAlias<"lsr $dst, $src, $shift",
922 (UBFMXri GPR64:$dst, GPR64:$src, imm0_63:$shift, 63)>;
923 def : InstAlias<"uxtb $dst, $src", (UBFMWri GPR32:$dst, GPR32:$src, 0, 7)>;
924 def : InstAlias<"uxtb $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 7)>;
925 def : InstAlias<"uxth $dst, $src", (UBFMWri GPR32:$dst, GPR32:$src, 0, 15)>;
926 def : InstAlias<"uxth $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 15)>;
927 def : InstAlias<"uxtw $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 31)>;
929 //===----------------------------------------------------------------------===//
930 // Conditionally set flags instructions.
931 //===----------------------------------------------------------------------===//
932 defm CCMN : CondSetFlagsImm<0, "ccmn">;
933 defm CCMP : CondSetFlagsImm<1, "ccmp">;
935 defm CCMN : CondSetFlagsReg<0, "ccmn">;
936 defm CCMP : CondSetFlagsReg<1, "ccmp">;
938 //===----------------------------------------------------------------------===//
939 // Conditional select instructions.
940 //===----------------------------------------------------------------------===//
941 defm CSEL : CondSelect<0, 0b00, "csel">;
943 def inc : PatFrag<(ops node:$in), (add node:$in, 1)>;
944 defm CSINC : CondSelectOp<0, 0b01, "csinc", inc>;
945 defm CSINV : CondSelectOp<1, 0b00, "csinv", not>;
946 defm CSNEG : CondSelectOp<1, 0b01, "csneg", ineg>;
948 def : Pat<(AArch64csinv GPR32:$tval, GPR32:$fval, (i32 imm:$cc), NZCV),
949 (CSINVWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
950 def : Pat<(AArch64csinv GPR64:$tval, GPR64:$fval, (i32 imm:$cc), NZCV),
951 (CSINVXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
952 def : Pat<(AArch64csneg GPR32:$tval, GPR32:$fval, (i32 imm:$cc), NZCV),
953 (CSNEGWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
954 def : Pat<(AArch64csneg GPR64:$tval, GPR64:$fval, (i32 imm:$cc), NZCV),
955 (CSNEGXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
956 def : Pat<(AArch64csinc GPR32:$tval, GPR32:$fval, (i32 imm:$cc), NZCV),
957 (CSINCWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
958 def : Pat<(AArch64csinc GPR64:$tval, GPR64:$fval, (i32 imm:$cc), NZCV),
959 (CSINCXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
961 def : Pat<(AArch64csel (i32 0), (i32 1), (i32 imm:$cc), NZCV),
962 (CSINCWr WZR, WZR, (i32 imm:$cc))>;
963 def : Pat<(AArch64csel (i64 0), (i64 1), (i32 imm:$cc), NZCV),
964 (CSINCXr XZR, XZR, (i32 imm:$cc))>;
965 def : Pat<(AArch64csel (i32 0), (i32 -1), (i32 imm:$cc), NZCV),
966 (CSINVWr WZR, WZR, (i32 imm:$cc))>;
967 def : Pat<(AArch64csel (i64 0), (i64 -1), (i32 imm:$cc), NZCV),
968 (CSINVXr XZR, XZR, (i32 imm:$cc))>;
970 // The inverse of the condition code from the alias instruction is what is used
971 // in the aliased instruction. The parser all ready inverts the condition code
972 // for these aliases.
973 def : InstAlias<"cset $dst, $cc",
974 (CSINCWr GPR32:$dst, WZR, WZR, inv_ccode:$cc)>;
975 def : InstAlias<"cset $dst, $cc",
976 (CSINCXr GPR64:$dst, XZR, XZR, inv_ccode:$cc)>;
978 def : InstAlias<"csetm $dst, $cc",
979 (CSINVWr GPR32:$dst, WZR, WZR, inv_ccode:$cc)>;
980 def : InstAlias<"csetm $dst, $cc",
981 (CSINVXr GPR64:$dst, XZR, XZR, inv_ccode:$cc)>;
983 def : InstAlias<"cinc $dst, $src, $cc",
984 (CSINCWr GPR32:$dst, GPR32:$src, GPR32:$src, inv_ccode:$cc)>;
985 def : InstAlias<"cinc $dst, $src, $cc",
986 (CSINCXr GPR64:$dst, GPR64:$src, GPR64:$src, inv_ccode:$cc)>;
988 def : InstAlias<"cinv $dst, $src, $cc",
989 (CSINVWr GPR32:$dst, GPR32:$src, GPR32:$src, inv_ccode:$cc)>;
990 def : InstAlias<"cinv $dst, $src, $cc",
991 (CSINVXr GPR64:$dst, GPR64:$src, GPR64:$src, inv_ccode:$cc)>;
993 def : InstAlias<"cneg $dst, $src, $cc",
994 (CSNEGWr GPR32:$dst, GPR32:$src, GPR32:$src, inv_ccode:$cc)>;
995 def : InstAlias<"cneg $dst, $src, $cc",
996 (CSNEGXr GPR64:$dst, GPR64:$src, GPR64:$src, inv_ccode:$cc)>;
998 //===----------------------------------------------------------------------===//
999 // PC-relative instructions.
1000 //===----------------------------------------------------------------------===//
1001 let isReMaterializable = 1 in {
1002 let hasSideEffects = 0, mayStore = 0, mayLoad = 0 in {
1003 def ADR : ADRI<0, "adr", adrlabel, []>;
1004 } // hasSideEffects = 0
1006 def ADRP : ADRI<1, "adrp", adrplabel,
1007 [(set GPR64:$Xd, (AArch64adrp tglobaladdr:$label))]>;
1008 } // isReMaterializable = 1
1010 // page address of a constant pool entry, block address
1011 def : Pat<(AArch64adrp tconstpool:$cp), (ADRP tconstpool:$cp)>;
1012 def : Pat<(AArch64adrp tblockaddress:$cp), (ADRP tblockaddress:$cp)>;
1014 //===----------------------------------------------------------------------===//
1015 // Unconditional branch (register) instructions.
1016 //===----------------------------------------------------------------------===//
1018 let isReturn = 1, isTerminator = 1, isBarrier = 1 in {
1019 def RET : BranchReg<0b0010, "ret", []>;
1020 def DRPS : SpecialReturn<0b0101, "drps">;
1021 def ERET : SpecialReturn<0b0100, "eret">;
1022 } // isReturn = 1, isTerminator = 1, isBarrier = 1
1024 // Default to the LR register.
1025 def : InstAlias<"ret", (RET LR)>;
1027 let isCall = 1, Defs = [LR], Uses = [SP] in {
1028 def BLR : BranchReg<0b0001, "blr", [(AArch64call GPR64:$Rn)]>;
1031 let isBranch = 1, isTerminator = 1, isBarrier = 1, isIndirectBranch = 1 in {
1032 def BR : BranchReg<0b0000, "br", [(brind GPR64:$Rn)]>;
1033 } // isBranch, isTerminator, isBarrier, isIndirectBranch
1035 // Create a separate pseudo-instruction for codegen to use so that we don't
1036 // flag lr as used in every function. It'll be restored before the RET by the
1037 // epilogue if it's legitimately used.
1038 def RET_ReallyLR : Pseudo<(outs), (ins), [(AArch64retflag)]> {
1039 let isTerminator = 1;
1044 // This is a directive-like pseudo-instruction. The purpose is to insert an
1045 // R_AARCH64_TLSDESC_CALL relocation at the offset of the following instruction
1046 // (which in the usual case is a BLR).
1047 let hasSideEffects = 1 in
1048 def TLSDESCCALL : Pseudo<(outs), (ins i64imm:$sym), []> {
1049 let AsmString = ".tlsdesccall $sym";
1052 // Pseudo-instruction representing a BLR with attached TLSDESC relocation. It
1053 // gets expanded to two MCInsts during lowering.
1054 let isCall = 1, Defs = [LR] in
1056 : Pseudo<(outs), (ins GPR64:$dest, i64imm:$sym),
1057 [(AArch64tlsdesc_call GPR64:$dest, tglobaltlsaddr:$sym)]>;
1059 def : Pat<(AArch64tlsdesc_call GPR64:$dest, texternalsym:$sym),
1060 (TLSDESC_BLR GPR64:$dest, texternalsym:$sym)>;
1061 //===----------------------------------------------------------------------===//
1062 // Conditional branch (immediate) instruction.
1063 //===----------------------------------------------------------------------===//
1064 def Bcc : BranchCond;
1066 //===----------------------------------------------------------------------===//
1067 // Compare-and-branch instructions.
1068 //===----------------------------------------------------------------------===//
1069 defm CBZ : CmpBranch<0, "cbz", AArch64cbz>;
1070 defm CBNZ : CmpBranch<1, "cbnz", AArch64cbnz>;
1072 //===----------------------------------------------------------------------===//
1073 // Test-bit-and-branch instructions.
1074 //===----------------------------------------------------------------------===//
1075 defm TBZ : TestBranch<0, "tbz", AArch64tbz>;
1076 defm TBNZ : TestBranch<1, "tbnz", AArch64tbnz>;
1078 //===----------------------------------------------------------------------===//
1079 // Unconditional branch (immediate) instructions.
1080 //===----------------------------------------------------------------------===//
1081 let isBranch = 1, isTerminator = 1, isBarrier = 1 in {
1082 def B : BranchImm<0, "b", [(br bb:$addr)]>;
1083 } // isBranch, isTerminator, isBarrier
1085 let isCall = 1, Defs = [LR], Uses = [SP] in {
1086 def BL : CallImm<1, "bl", [(AArch64call tglobaladdr:$addr)]>;
1088 def : Pat<(AArch64call texternalsym:$func), (BL texternalsym:$func)>;
1090 //===----------------------------------------------------------------------===//
1091 // Exception generation instructions.
1092 //===----------------------------------------------------------------------===//
1093 def BRK : ExceptionGeneration<0b001, 0b00, "brk">;
1094 def DCPS1 : ExceptionGeneration<0b101, 0b01, "dcps1">;
1095 def DCPS2 : ExceptionGeneration<0b101, 0b10, "dcps2">;
1096 def DCPS3 : ExceptionGeneration<0b101, 0b11, "dcps3">;
1097 def HLT : ExceptionGeneration<0b010, 0b00, "hlt">;
1098 def HVC : ExceptionGeneration<0b000, 0b10, "hvc">;
1099 def SMC : ExceptionGeneration<0b000, 0b11, "smc">;
1100 def SVC : ExceptionGeneration<0b000, 0b01, "svc">;
1102 // DCPSn defaults to an immediate operand of zero if unspecified.
1103 def : InstAlias<"dcps1", (DCPS1 0)>;
1104 def : InstAlias<"dcps2", (DCPS2 0)>;
1105 def : InstAlias<"dcps3", (DCPS3 0)>;
1107 //===----------------------------------------------------------------------===//
1108 // Load instructions.
1109 //===----------------------------------------------------------------------===//
1111 // Pair (indexed, offset)
1112 defm LDPW : LoadPairOffset<0b00, 0, GPR32, simm7s4, "ldp">;
1113 defm LDPX : LoadPairOffset<0b10, 0, GPR64, simm7s8, "ldp">;
1114 defm LDPS : LoadPairOffset<0b00, 1, FPR32, simm7s4, "ldp">;
1115 defm LDPD : LoadPairOffset<0b01, 1, FPR64, simm7s8, "ldp">;
1116 defm LDPQ : LoadPairOffset<0b10, 1, FPR128, simm7s16, "ldp">;
1118 defm LDPSW : LoadPairOffset<0b01, 0, GPR64, simm7s4, "ldpsw">;
1120 // Pair (pre-indexed)
1121 def LDPWpre : LoadPairPreIdx<0b00, 0, GPR32, simm7s4, "ldp">;
1122 def LDPXpre : LoadPairPreIdx<0b10, 0, GPR64, simm7s8, "ldp">;
1123 def LDPSpre : LoadPairPreIdx<0b00, 1, FPR32, simm7s4, "ldp">;
1124 def LDPDpre : LoadPairPreIdx<0b01, 1, FPR64, simm7s8, "ldp">;
1125 def LDPQpre : LoadPairPreIdx<0b10, 1, FPR128, simm7s16, "ldp">;
1127 def LDPSWpre : LoadPairPreIdx<0b01, 0, GPR64, simm7s4, "ldpsw">;
1129 // Pair (post-indexed)
1130 def LDPWpost : LoadPairPostIdx<0b00, 0, GPR32, simm7s4, "ldp">;
1131 def LDPXpost : LoadPairPostIdx<0b10, 0, GPR64, simm7s8, "ldp">;
1132 def LDPSpost : LoadPairPostIdx<0b00, 1, FPR32, simm7s4, "ldp">;
1133 def LDPDpost : LoadPairPostIdx<0b01, 1, FPR64, simm7s8, "ldp">;
1134 def LDPQpost : LoadPairPostIdx<0b10, 1, FPR128, simm7s16, "ldp">;
1136 def LDPSWpost : LoadPairPostIdx<0b01, 0, GPR64, simm7s4, "ldpsw">;
1139 // Pair (no allocate)
1140 defm LDNPW : LoadPairNoAlloc<0b00, 0, GPR32, simm7s4, "ldnp">;
1141 defm LDNPX : LoadPairNoAlloc<0b10, 0, GPR64, simm7s8, "ldnp">;
1142 defm LDNPS : LoadPairNoAlloc<0b00, 1, FPR32, simm7s4, "ldnp">;
1143 defm LDNPD : LoadPairNoAlloc<0b01, 1, FPR64, simm7s8, "ldnp">;
1144 defm LDNPQ : LoadPairNoAlloc<0b10, 1, FPR128, simm7s16, "ldnp">;
1147 // (register offset)
1151 defm LDRBB : Load8RO<0b00, 0, 0b01, GPR32, "ldrb", i32, zextloadi8>;
1152 defm LDRHH : Load16RO<0b01, 0, 0b01, GPR32, "ldrh", i32, zextloadi16>;
1153 defm LDRW : Load32RO<0b10, 0, 0b01, GPR32, "ldr", i32, load>;
1154 defm LDRX : Load64RO<0b11, 0, 0b01, GPR64, "ldr", i64, load>;
1157 defm LDRB : Load8RO<0b00, 1, 0b01, FPR8, "ldr", untyped, load>;
1158 defm LDRH : Load16RO<0b01, 1, 0b01, FPR16, "ldr", f16, load>;
1159 defm LDRS : Load32RO<0b10, 1, 0b01, FPR32, "ldr", f32, load>;
1160 defm LDRD : Load64RO<0b11, 1, 0b01, FPR64, "ldr", f64, load>;
1161 defm LDRQ : Load128RO<0b00, 1, 0b11, FPR128, "ldr", f128, load>;
1163 // Load sign-extended half-word
1164 defm LDRSHW : Load16RO<0b01, 0, 0b11, GPR32, "ldrsh", i32, sextloadi16>;
1165 defm LDRSHX : Load16RO<0b01, 0, 0b10, GPR64, "ldrsh", i64, sextloadi16>;
1167 // Load sign-extended byte
1168 defm LDRSBW : Load8RO<0b00, 0, 0b11, GPR32, "ldrsb", i32, sextloadi8>;
1169 defm LDRSBX : Load8RO<0b00, 0, 0b10, GPR64, "ldrsb", i64, sextloadi8>;
1171 // Load sign-extended word
1172 defm LDRSW : Load32RO<0b10, 0, 0b10, GPR64, "ldrsw", i64, sextloadi32>;
1175 defm PRFM : PrefetchRO<0b11, 0, 0b10, "prfm">;
1177 // For regular load, we do not have any alignment requirement.
1178 // Thus, it is safe to directly map the vector loads with interesting
1179 // addressing modes.
1180 // FIXME: We could do the same for bitconvert to floating point vectors.
1181 multiclass ScalToVecROLoadPat<ROAddrMode ro, SDPatternOperator loadop,
1182 ValueType ScalTy, ValueType VecTy,
1183 Instruction LOADW, Instruction LOADX,
1185 def : Pat<(VecTy (scalar_to_vector (ScalTy
1186 (loadop (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$offset))))),
1187 (INSERT_SUBREG (VecTy (IMPLICIT_DEF)),
1188 (LOADW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$offset),
1191 def : Pat<(VecTy (scalar_to_vector (ScalTy
1192 (loadop (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$offset))))),
1193 (INSERT_SUBREG (VecTy (IMPLICIT_DEF)),
1194 (LOADX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$offset),
1198 let AddedComplexity = 10 in {
1199 defm : ScalToVecROLoadPat<ro8, extloadi8, i32, v8i8, LDRBroW, LDRBroX, bsub>;
1200 defm : ScalToVecROLoadPat<ro8, extloadi8, i32, v16i8, LDRBroW, LDRBroX, bsub>;
1202 defm : ScalToVecROLoadPat<ro16, extloadi16, i32, v4i16, LDRHroW, LDRHroX, hsub>;
1203 defm : ScalToVecROLoadPat<ro16, extloadi16, i32, v8i16, LDRHroW, LDRHroX, hsub>;
1205 defm : ScalToVecROLoadPat<ro16, load, i32, v4f16, LDRHroW, LDRHroX, hsub>;
1206 defm : ScalToVecROLoadPat<ro16, load, i32, v8f16, LDRHroW, LDRHroX, hsub>;
1208 defm : ScalToVecROLoadPat<ro32, load, i32, v2i32, LDRSroW, LDRSroX, ssub>;
1209 defm : ScalToVecROLoadPat<ro32, load, i32, v4i32, LDRSroW, LDRSroX, ssub>;
1211 defm : ScalToVecROLoadPat<ro32, load, f32, v2f32, LDRSroW, LDRSroX, ssub>;
1212 defm : ScalToVecROLoadPat<ro32, load, f32, v4f32, LDRSroW, LDRSroX, ssub>;
1214 defm : ScalToVecROLoadPat<ro64, load, i64, v2i64, LDRDroW, LDRDroX, dsub>;
1216 defm : ScalToVecROLoadPat<ro64, load, f64, v2f64, LDRDroW, LDRDroX, dsub>;
1219 def : Pat <(v1i64 (scalar_to_vector (i64
1220 (load (ro_Windexed64 GPR64sp:$Rn, GPR32:$Rm,
1221 ro_Wextend64:$extend))))),
1222 (LDRDroW GPR64sp:$Rn, GPR32:$Rm, ro_Wextend64:$extend)>;
1224 def : Pat <(v1i64 (scalar_to_vector (i64
1225 (load (ro_Xindexed64 GPR64sp:$Rn, GPR64:$Rm,
1226 ro_Xextend64:$extend))))),
1227 (LDRDroX GPR64sp:$Rn, GPR64:$Rm, ro_Xextend64:$extend)>;
1230 // Match all load 64 bits width whose type is compatible with FPR64
1231 multiclass VecROLoadPat<ROAddrMode ro, ValueType VecTy,
1232 Instruction LOADW, Instruction LOADX> {
1234 def : Pat<(VecTy (load (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend))),
1235 (LOADW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
1237 def : Pat<(VecTy (load (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend))),
1238 (LOADX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
1241 let AddedComplexity = 10 in {
1242 let Predicates = [IsLE] in {
1243 // We must do vector loads with LD1 in big-endian.
1244 defm : VecROLoadPat<ro64, v2i32, LDRDroW, LDRDroX>;
1245 defm : VecROLoadPat<ro64, v2f32, LDRDroW, LDRDroX>;
1246 defm : VecROLoadPat<ro64, v8i8, LDRDroW, LDRDroX>;
1247 defm : VecROLoadPat<ro64, v4i16, LDRDroW, LDRDroX>;
1248 defm : VecROLoadPat<ro64, v4f16, LDRDroW, LDRDroX>;
1251 defm : VecROLoadPat<ro64, v1i64, LDRDroW, LDRDroX>;
1252 defm : VecROLoadPat<ro64, v1f64, LDRDroW, LDRDroX>;
1254 // Match all load 128 bits width whose type is compatible with FPR128
1255 let Predicates = [IsLE] in {
1256 // We must do vector loads with LD1 in big-endian.
1257 defm : VecROLoadPat<ro128, v2i64, LDRQroW, LDRQroX>;
1258 defm : VecROLoadPat<ro128, v2f64, LDRQroW, LDRQroX>;
1259 defm : VecROLoadPat<ro128, v4i32, LDRQroW, LDRQroX>;
1260 defm : VecROLoadPat<ro128, v4f32, LDRQroW, LDRQroX>;
1261 defm : VecROLoadPat<ro128, v8i16, LDRQroW, LDRQroX>;
1262 defm : VecROLoadPat<ro128, v8f16, LDRQroW, LDRQroX>;
1263 defm : VecROLoadPat<ro128, v16i8, LDRQroW, LDRQroX>;
1265 } // AddedComplexity = 10
1268 multiclass ExtLoadTo64ROPat<ROAddrMode ro, SDPatternOperator loadop,
1269 Instruction INSTW, Instruction INSTX> {
1270 def : Pat<(i64 (loadop (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend))),
1271 (SUBREG_TO_REG (i64 0),
1272 (INSTW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend),
1275 def : Pat<(i64 (loadop (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend))),
1276 (SUBREG_TO_REG (i64 0),
1277 (INSTX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend),
1281 let AddedComplexity = 10 in {
1282 defm : ExtLoadTo64ROPat<ro8, zextloadi8, LDRBBroW, LDRBBroX>;
1283 defm : ExtLoadTo64ROPat<ro16, zextloadi16, LDRHHroW, LDRHHroX>;
1284 defm : ExtLoadTo64ROPat<ro32, zextloadi32, LDRWroW, LDRWroX>;
1286 // zextloadi1 -> zextloadi8
1287 defm : ExtLoadTo64ROPat<ro8, zextloadi1, LDRBBroW, LDRBBroX>;
1289 // extload -> zextload
1290 defm : ExtLoadTo64ROPat<ro8, extloadi8, LDRBBroW, LDRBBroX>;
1291 defm : ExtLoadTo64ROPat<ro16, extloadi16, LDRHHroW, LDRHHroX>;
1292 defm : ExtLoadTo64ROPat<ro32, extloadi32, LDRWroW, LDRWroX>;
1294 // extloadi1 -> zextloadi8
1295 defm : ExtLoadTo64ROPat<ro8, extloadi1, LDRBBroW, LDRBBroX>;
1300 multiclass ExtLoadTo32ROPat<ROAddrMode ro, SDPatternOperator loadop,
1301 Instruction INSTW, Instruction INSTX> {
1302 def : Pat<(i32 (loadop (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend))),
1303 (INSTW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
1305 def : Pat<(i32 (loadop (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend))),
1306 (INSTX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
1310 let AddedComplexity = 10 in {
1311 // extload -> zextload
1312 defm : ExtLoadTo32ROPat<ro8, extloadi8, LDRBBroW, LDRBBroX>;
1313 defm : ExtLoadTo32ROPat<ro16, extloadi16, LDRHHroW, LDRHHroX>;
1314 defm : ExtLoadTo32ROPat<ro32, extloadi32, LDRWroW, LDRWroX>;
1316 // zextloadi1 -> zextloadi8
1317 defm : ExtLoadTo32ROPat<ro8, zextloadi1, LDRBBroW, LDRBBroX>;
1321 // (unsigned immediate)
1323 defm LDRX : LoadUI<0b11, 0, 0b01, GPR64, uimm12s8, "ldr",
1325 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)))]>;
1326 defm LDRW : LoadUI<0b10, 0, 0b01, GPR32, uimm12s4, "ldr",
1328 (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset)))]>;
1329 defm LDRB : LoadUI<0b00, 1, 0b01, FPR8, uimm12s1, "ldr",
1331 (load (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset)))]>;
1332 defm LDRH : LoadUI<0b01, 1, 0b01, FPR16, uimm12s2, "ldr",
1333 [(set (f16 FPR16:$Rt),
1334 (load (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset)))]>;
1335 defm LDRS : LoadUI<0b10, 1, 0b01, FPR32, uimm12s4, "ldr",
1336 [(set (f32 FPR32:$Rt),
1337 (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset)))]>;
1338 defm LDRD : LoadUI<0b11, 1, 0b01, FPR64, uimm12s8, "ldr",
1339 [(set (f64 FPR64:$Rt),
1340 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)))]>;
1341 defm LDRQ : LoadUI<0b00, 1, 0b11, FPR128, uimm12s16, "ldr",
1342 [(set (f128 FPR128:$Rt),
1343 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)))]>;
1345 // For regular load, we do not have any alignment requirement.
1346 // Thus, it is safe to directly map the vector loads with interesting
1347 // addressing modes.
1348 // FIXME: We could do the same for bitconvert to floating point vectors.
1349 def : Pat <(v8i8 (scalar_to_vector (i32
1350 (extloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))))),
1351 (INSERT_SUBREG (v8i8 (IMPLICIT_DEF)),
1352 (LDRBui GPR64sp:$Rn, uimm12s1:$offset), bsub)>;
1353 def : Pat <(v16i8 (scalar_to_vector (i32
1354 (extloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))))),
1355 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
1356 (LDRBui GPR64sp:$Rn, uimm12s1:$offset), bsub)>;
1357 def : Pat <(v4i16 (scalar_to_vector (i32
1358 (extloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))))),
1359 (INSERT_SUBREG (v4i16 (IMPLICIT_DEF)),
1360 (LDRHui GPR64sp:$Rn, uimm12s2:$offset), hsub)>;
1361 def : Pat <(v8i16 (scalar_to_vector (i32
1362 (extloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))))),
1363 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
1364 (LDRHui GPR64sp:$Rn, uimm12s2:$offset), hsub)>;
1365 def : Pat <(v2i32 (scalar_to_vector (i32
1366 (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))))),
1367 (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)),
1368 (LDRSui GPR64sp:$Rn, uimm12s4:$offset), ssub)>;
1369 def : Pat <(v4i32 (scalar_to_vector (i32
1370 (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))))),
1371 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
1372 (LDRSui GPR64sp:$Rn, uimm12s4:$offset), ssub)>;
1373 def : Pat <(v1i64 (scalar_to_vector (i64
1374 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))))),
1375 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1376 def : Pat <(v2i64 (scalar_to_vector (i64
1377 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))))),
1378 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
1379 (LDRDui GPR64sp:$Rn, uimm12s8:$offset), dsub)>;
1381 // Match all load 64 bits width whose type is compatible with FPR64
1382 let Predicates = [IsLE] in {
1383 // We must use LD1 to perform vector loads in big-endian.
1384 def : Pat<(v2f32 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1385 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1386 def : Pat<(v8i8 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1387 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1388 def : Pat<(v4i16 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1389 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1390 def : Pat<(v2i32 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1391 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1392 def : Pat<(v4f16 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1393 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1395 def : Pat<(v1f64 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1396 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1397 def : Pat<(v1i64 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1398 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1400 // Match all load 128 bits width whose type is compatible with FPR128
1401 let Predicates = [IsLE] in {
1402 // We must use LD1 to perform vector loads in big-endian.
1403 def : Pat<(v4f32 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1404 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1405 def : Pat<(v2f64 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1406 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1407 def : Pat<(v16i8 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1408 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1409 def : Pat<(v8i16 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1410 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1411 def : Pat<(v4i32 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1412 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1413 def : Pat<(v2i64 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1414 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1415 def : Pat<(v8f16 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1416 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1418 def : Pat<(f128 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1419 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1421 defm LDRHH : LoadUI<0b01, 0, 0b01, GPR32, uimm12s2, "ldrh",
1423 (zextloadi16 (am_indexed16 GPR64sp:$Rn,
1424 uimm12s2:$offset)))]>;
1425 defm LDRBB : LoadUI<0b00, 0, 0b01, GPR32, uimm12s1, "ldrb",
1427 (zextloadi8 (am_indexed8 GPR64sp:$Rn,
1428 uimm12s1:$offset)))]>;
1430 def : Pat<(i64 (zextloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1431 (SUBREG_TO_REG (i64 0), (LDRBBui GPR64sp:$Rn, uimm12s1:$offset), sub_32)>;
1432 def : Pat<(i64 (zextloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))),
1433 (SUBREG_TO_REG (i64 0), (LDRHHui GPR64sp:$Rn, uimm12s2:$offset), sub_32)>;
1435 // zextloadi1 -> zextloadi8
1436 def : Pat<(i32 (zextloadi1 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1437 (LDRBBui GPR64sp:$Rn, uimm12s1:$offset)>;
1438 def : Pat<(i64 (zextloadi1 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1439 (SUBREG_TO_REG (i64 0), (LDRBBui GPR64sp:$Rn, uimm12s1:$offset), sub_32)>;
1441 // extload -> zextload
1442 def : Pat<(i32 (extloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))),
1443 (LDRHHui GPR64sp:$Rn, uimm12s2:$offset)>;
1444 def : Pat<(i32 (extloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1445 (LDRBBui GPR64sp:$Rn, uimm12s1:$offset)>;
1446 def : Pat<(i32 (extloadi1 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1447 (LDRBBui GPR64sp:$Rn, uimm12s1:$offset)>;
1448 def : Pat<(i64 (extloadi32 (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))),
1449 (SUBREG_TO_REG (i64 0), (LDRWui GPR64sp:$Rn, uimm12s4:$offset), sub_32)>;
1450 def : Pat<(i64 (extloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))),
1451 (SUBREG_TO_REG (i64 0), (LDRHHui GPR64sp:$Rn, uimm12s2:$offset), sub_32)>;
1452 def : Pat<(i64 (extloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1453 (SUBREG_TO_REG (i64 0), (LDRBBui GPR64sp:$Rn, uimm12s1:$offset), sub_32)>;
1454 def : Pat<(i64 (extloadi1 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1455 (SUBREG_TO_REG (i64 0), (LDRBBui GPR64sp:$Rn, uimm12s1:$offset), sub_32)>;
1457 // load sign-extended half-word
1458 defm LDRSHW : LoadUI<0b01, 0, 0b11, GPR32, uimm12s2, "ldrsh",
1460 (sextloadi16 (am_indexed16 GPR64sp:$Rn,
1461 uimm12s2:$offset)))]>;
1462 defm LDRSHX : LoadUI<0b01, 0, 0b10, GPR64, uimm12s2, "ldrsh",
1464 (sextloadi16 (am_indexed16 GPR64sp:$Rn,
1465 uimm12s2:$offset)))]>;
1467 // load sign-extended byte
1468 defm LDRSBW : LoadUI<0b00, 0, 0b11, GPR32, uimm12s1, "ldrsb",
1470 (sextloadi8 (am_indexed8 GPR64sp:$Rn,
1471 uimm12s1:$offset)))]>;
1472 defm LDRSBX : LoadUI<0b00, 0, 0b10, GPR64, uimm12s1, "ldrsb",
1474 (sextloadi8 (am_indexed8 GPR64sp:$Rn,
1475 uimm12s1:$offset)))]>;
1477 // load sign-extended word
1478 defm LDRSW : LoadUI<0b10, 0, 0b10, GPR64, uimm12s4, "ldrsw",
1480 (sextloadi32 (am_indexed32 GPR64sp:$Rn,
1481 uimm12s4:$offset)))]>;
1483 // load zero-extended word
1484 def : Pat<(i64 (zextloadi32 (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))),
1485 (SUBREG_TO_REG (i64 0), (LDRWui GPR64sp:$Rn, uimm12s4:$offset), sub_32)>;
1488 def PRFMui : PrefetchUI<0b11, 0, 0b10, "prfm",
1489 [(AArch64Prefetch imm:$Rt,
1490 (am_indexed64 GPR64sp:$Rn,
1491 uimm12s8:$offset))]>;
1493 def : InstAlias<"prfm $Rt, [$Rn]", (PRFMui prfop:$Rt, GPR64sp:$Rn, 0)>;
1497 def LDRWl : LoadLiteral<0b00, 0, GPR32, "ldr">;
1498 def LDRXl : LoadLiteral<0b01, 0, GPR64, "ldr">;
1499 def LDRSl : LoadLiteral<0b00, 1, FPR32, "ldr">;
1500 def LDRDl : LoadLiteral<0b01, 1, FPR64, "ldr">;
1501 def LDRQl : LoadLiteral<0b10, 1, FPR128, "ldr">;
1503 // load sign-extended word
1504 def LDRSWl : LoadLiteral<0b10, 0, GPR64, "ldrsw">;
1507 def PRFMl : PrefetchLiteral<0b11, 0, "prfm", []>;
1508 // [(AArch64Prefetch imm:$Rt, tglobaladdr:$label)]>;
1511 // (unscaled immediate)
1512 defm LDURX : LoadUnscaled<0b11, 0, 0b01, GPR64, "ldur",
1514 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset)))]>;
1515 defm LDURW : LoadUnscaled<0b10, 0, 0b01, GPR32, "ldur",
1517 (load (am_unscaled32 GPR64sp:$Rn, simm9:$offset)))]>;
1518 defm LDURB : LoadUnscaled<0b00, 1, 0b01, FPR8, "ldur",
1520 (load (am_unscaled8 GPR64sp:$Rn, simm9:$offset)))]>;
1521 defm LDURH : LoadUnscaled<0b01, 1, 0b01, FPR16, "ldur",
1523 (load (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
1524 defm LDURS : LoadUnscaled<0b10, 1, 0b01, FPR32, "ldur",
1525 [(set (f32 FPR32:$Rt),
1526 (load (am_unscaled32 GPR64sp:$Rn, simm9:$offset)))]>;
1527 defm LDURD : LoadUnscaled<0b11, 1, 0b01, FPR64, "ldur",
1528 [(set (f64 FPR64:$Rt),
1529 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset)))]>;
1530 defm LDURQ : LoadUnscaled<0b00, 1, 0b11, FPR128, "ldur",
1531 [(set (f128 FPR128:$Rt),
1532 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset)))]>;
1535 : LoadUnscaled<0b01, 0, 0b01, GPR32, "ldurh",
1537 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
1539 : LoadUnscaled<0b00, 0, 0b01, GPR32, "ldurb",
1541 (zextloadi8 (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
1543 // Match all load 64 bits width whose type is compatible with FPR64
1544 let Predicates = [IsLE] in {
1545 def : Pat<(v2f32 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1546 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1547 def : Pat<(v2i32 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1548 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1549 def : Pat<(v4i16 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1550 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1551 def : Pat<(v8i8 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1552 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1553 def : Pat<(v4f16 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1554 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1556 def : Pat<(v1f64 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1557 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1558 def : Pat<(v1i64 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1559 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1561 // Match all load 128 bits width whose type is compatible with FPR128
1562 let Predicates = [IsLE] in {
1563 def : Pat<(v2f64 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1564 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1565 def : Pat<(v2i64 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1566 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1567 def : Pat<(v4f32 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1568 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1569 def : Pat<(v4i32 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1570 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1571 def : Pat<(v8i16 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1572 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1573 def : Pat<(v16i8 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1574 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1575 def : Pat<(v8f16 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1576 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1580 def : Pat<(i32 (extloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
1581 (LDURHHi GPR64sp:$Rn, simm9:$offset)>;
1582 def : Pat<(i32 (extloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1583 (LDURBBi GPR64sp:$Rn, simm9:$offset)>;
1584 def : Pat<(i32 (extloadi1 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1585 (LDURBBi GPR64sp:$Rn, simm9:$offset)>;
1586 def : Pat<(i64 (extloadi32 (am_unscaled32 GPR64sp:$Rn, simm9:$offset))),
1587 (SUBREG_TO_REG (i64 0), (LDURWi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1588 def : Pat<(i64 (extloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
1589 (SUBREG_TO_REG (i64 0), (LDURHHi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1590 def : Pat<(i64 (extloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1591 (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1592 def : Pat<(i64 (extloadi1 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1593 (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1595 def : Pat<(i32 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
1596 (LDURHHi GPR64sp:$Rn, simm9:$offset)>;
1597 def : Pat<(i32 (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1598 (LDURBBi GPR64sp:$Rn, simm9:$offset)>;
1599 def : Pat<(i32 (zextloadi1 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1600 (LDURBBi GPR64sp:$Rn, simm9:$offset)>;
1601 def : Pat<(i64 (zextloadi32 (am_unscaled32 GPR64sp:$Rn, simm9:$offset))),
1602 (SUBREG_TO_REG (i64 0), (LDURWi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1603 def : Pat<(i64 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
1604 (SUBREG_TO_REG (i64 0), (LDURHHi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1605 def : Pat<(i64 (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1606 (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1607 def : Pat<(i64 (zextloadi1 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1608 (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1612 // LDR mnemonics fall back to LDUR for negative or unaligned offsets.
1614 // Define new assembler match classes as we want to only match these when
1615 // the don't otherwise match the scaled addressing mode for LDR/STR. Don't
1616 // associate a DiagnosticType either, as we want the diagnostic for the
1617 // canonical form (the scaled operand) to take precedence.
1618 class SImm9OffsetOperand<int Width> : AsmOperandClass {
1619 let Name = "SImm9OffsetFB" # Width;
1620 let PredicateMethod = "isSImm9OffsetFB<" # Width # ">";
1621 let RenderMethod = "addImmOperands";
1624 def SImm9OffsetFB8Operand : SImm9OffsetOperand<8>;
1625 def SImm9OffsetFB16Operand : SImm9OffsetOperand<16>;
1626 def SImm9OffsetFB32Operand : SImm9OffsetOperand<32>;
1627 def SImm9OffsetFB64Operand : SImm9OffsetOperand<64>;
1628 def SImm9OffsetFB128Operand : SImm9OffsetOperand<128>;
1630 def simm9_offset_fb8 : Operand<i64> {
1631 let ParserMatchClass = SImm9OffsetFB8Operand;
1633 def simm9_offset_fb16 : Operand<i64> {
1634 let ParserMatchClass = SImm9OffsetFB16Operand;
1636 def simm9_offset_fb32 : Operand<i64> {
1637 let ParserMatchClass = SImm9OffsetFB32Operand;
1639 def simm9_offset_fb64 : Operand<i64> {
1640 let ParserMatchClass = SImm9OffsetFB64Operand;
1642 def simm9_offset_fb128 : Operand<i64> {
1643 let ParserMatchClass = SImm9OffsetFB128Operand;
1646 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1647 (LDURXi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb64:$offset), 0>;
1648 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1649 (LDURWi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
1650 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1651 (LDURBi FPR8:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
1652 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1653 (LDURHi FPR16:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
1654 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1655 (LDURSi FPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
1656 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1657 (LDURDi FPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb64:$offset), 0>;
1658 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1659 (LDURQi FPR128:$Rt, GPR64sp:$Rn, simm9_offset_fb128:$offset), 0>;
1662 def : Pat<(i64 (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1663 (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1664 def : Pat<(i64 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
1665 (SUBREG_TO_REG (i64 0), (LDURHHi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1667 // load sign-extended half-word
1669 : LoadUnscaled<0b01, 0, 0b11, GPR32, "ldursh",
1671 (sextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
1673 : LoadUnscaled<0b01, 0, 0b10, GPR64, "ldursh",
1675 (sextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
1677 // load sign-extended byte
1679 : LoadUnscaled<0b00, 0, 0b11, GPR32, "ldursb",
1681 (sextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset)))]>;
1683 : LoadUnscaled<0b00, 0, 0b10, GPR64, "ldursb",
1685 (sextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset)))]>;
1687 // load sign-extended word
1689 : LoadUnscaled<0b10, 0, 0b10, GPR64, "ldursw",
1691 (sextloadi32 (am_unscaled32 GPR64sp:$Rn, simm9:$offset)))]>;
1693 // zero and sign extending aliases from generic LDR* mnemonics to LDUR*.
1694 def : InstAlias<"ldrb $Rt, [$Rn, $offset]",
1695 (LDURBBi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
1696 def : InstAlias<"ldrh $Rt, [$Rn, $offset]",
1697 (LDURHHi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
1698 def : InstAlias<"ldrsb $Rt, [$Rn, $offset]",
1699 (LDURSBWi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
1700 def : InstAlias<"ldrsb $Rt, [$Rn, $offset]",
1701 (LDURSBXi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
1702 def : InstAlias<"ldrsh $Rt, [$Rn, $offset]",
1703 (LDURSHWi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
1704 def : InstAlias<"ldrsh $Rt, [$Rn, $offset]",
1705 (LDURSHXi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
1706 def : InstAlias<"ldrsw $Rt, [$Rn, $offset]",
1707 (LDURSWi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
1710 defm PRFUM : PrefetchUnscaled<0b11, 0, 0b10, "prfum",
1711 [(AArch64Prefetch imm:$Rt,
1712 (am_unscaled64 GPR64sp:$Rn, simm9:$offset))]>;
1715 // (unscaled immediate, unprivileged)
1716 defm LDTRX : LoadUnprivileged<0b11, 0, 0b01, GPR64, "ldtr">;
1717 defm LDTRW : LoadUnprivileged<0b10, 0, 0b01, GPR32, "ldtr">;
1719 defm LDTRH : LoadUnprivileged<0b01, 0, 0b01, GPR32, "ldtrh">;
1720 defm LDTRB : LoadUnprivileged<0b00, 0, 0b01, GPR32, "ldtrb">;
1722 // load sign-extended half-word
1723 defm LDTRSHW : LoadUnprivileged<0b01, 0, 0b11, GPR32, "ldtrsh">;
1724 defm LDTRSHX : LoadUnprivileged<0b01, 0, 0b10, GPR64, "ldtrsh">;
1726 // load sign-extended byte
1727 defm LDTRSBW : LoadUnprivileged<0b00, 0, 0b11, GPR32, "ldtrsb">;
1728 defm LDTRSBX : LoadUnprivileged<0b00, 0, 0b10, GPR64, "ldtrsb">;
1730 // load sign-extended word
1731 defm LDTRSW : LoadUnprivileged<0b10, 0, 0b10, GPR64, "ldtrsw">;
1734 // (immediate pre-indexed)
1735 def LDRWpre : LoadPreIdx<0b10, 0, 0b01, GPR32, "ldr">;
1736 def LDRXpre : LoadPreIdx<0b11, 0, 0b01, GPR64, "ldr">;
1737 def LDRBpre : LoadPreIdx<0b00, 1, 0b01, FPR8, "ldr">;
1738 def LDRHpre : LoadPreIdx<0b01, 1, 0b01, FPR16, "ldr">;
1739 def LDRSpre : LoadPreIdx<0b10, 1, 0b01, FPR32, "ldr">;
1740 def LDRDpre : LoadPreIdx<0b11, 1, 0b01, FPR64, "ldr">;
1741 def LDRQpre : LoadPreIdx<0b00, 1, 0b11, FPR128, "ldr">;
1743 // load sign-extended half-word
1744 def LDRSHWpre : LoadPreIdx<0b01, 0, 0b11, GPR32, "ldrsh">;
1745 def LDRSHXpre : LoadPreIdx<0b01, 0, 0b10, GPR64, "ldrsh">;
1747 // load sign-extended byte
1748 def LDRSBWpre : LoadPreIdx<0b00, 0, 0b11, GPR32, "ldrsb">;
1749 def LDRSBXpre : LoadPreIdx<0b00, 0, 0b10, GPR64, "ldrsb">;
1751 // load zero-extended byte
1752 def LDRBBpre : LoadPreIdx<0b00, 0, 0b01, GPR32, "ldrb">;
1753 def LDRHHpre : LoadPreIdx<0b01, 0, 0b01, GPR32, "ldrh">;
1755 // load sign-extended word
1756 def LDRSWpre : LoadPreIdx<0b10, 0, 0b10, GPR64, "ldrsw">;
1759 // (immediate post-indexed)
1760 def LDRWpost : LoadPostIdx<0b10, 0, 0b01, GPR32, "ldr">;
1761 def LDRXpost : LoadPostIdx<0b11, 0, 0b01, GPR64, "ldr">;
1762 def LDRBpost : LoadPostIdx<0b00, 1, 0b01, FPR8, "ldr">;
1763 def LDRHpost : LoadPostIdx<0b01, 1, 0b01, FPR16, "ldr">;
1764 def LDRSpost : LoadPostIdx<0b10, 1, 0b01, FPR32, "ldr">;
1765 def LDRDpost : LoadPostIdx<0b11, 1, 0b01, FPR64, "ldr">;
1766 def LDRQpost : LoadPostIdx<0b00, 1, 0b11, FPR128, "ldr">;
1768 // load sign-extended half-word
1769 def LDRSHWpost : LoadPostIdx<0b01, 0, 0b11, GPR32, "ldrsh">;
1770 def LDRSHXpost : LoadPostIdx<0b01, 0, 0b10, GPR64, "ldrsh">;
1772 // load sign-extended byte
1773 def LDRSBWpost : LoadPostIdx<0b00, 0, 0b11, GPR32, "ldrsb">;
1774 def LDRSBXpost : LoadPostIdx<0b00, 0, 0b10, GPR64, "ldrsb">;
1776 // load zero-extended byte
1777 def LDRBBpost : LoadPostIdx<0b00, 0, 0b01, GPR32, "ldrb">;
1778 def LDRHHpost : LoadPostIdx<0b01, 0, 0b01, GPR32, "ldrh">;
1780 // load sign-extended word
1781 def LDRSWpost : LoadPostIdx<0b10, 0, 0b10, GPR64, "ldrsw">;
1783 //===----------------------------------------------------------------------===//
1784 // Store instructions.
1785 //===----------------------------------------------------------------------===//
1787 // Pair (indexed, offset)
1788 // FIXME: Use dedicated range-checked addressing mode operand here.
1789 defm STPW : StorePairOffset<0b00, 0, GPR32, simm7s4, "stp">;
1790 defm STPX : StorePairOffset<0b10, 0, GPR64, simm7s8, "stp">;
1791 defm STPS : StorePairOffset<0b00, 1, FPR32, simm7s4, "stp">;
1792 defm STPD : StorePairOffset<0b01, 1, FPR64, simm7s8, "stp">;
1793 defm STPQ : StorePairOffset<0b10, 1, FPR128, simm7s16, "stp">;
1795 // Pair (pre-indexed)
1796 def STPWpre : StorePairPreIdx<0b00, 0, GPR32, simm7s4, "stp">;
1797 def STPXpre : StorePairPreIdx<0b10, 0, GPR64, simm7s8, "stp">;
1798 def STPSpre : StorePairPreIdx<0b00, 1, FPR32, simm7s4, "stp">;
1799 def STPDpre : StorePairPreIdx<0b01, 1, FPR64, simm7s8, "stp">;
1800 def STPQpre : StorePairPreIdx<0b10, 1, FPR128, simm7s16, "stp">;
1802 // Pair (pre-indexed)
1803 def STPWpost : StorePairPostIdx<0b00, 0, GPR32, simm7s4, "stp">;
1804 def STPXpost : StorePairPostIdx<0b10, 0, GPR64, simm7s8, "stp">;
1805 def STPSpost : StorePairPostIdx<0b00, 1, FPR32, simm7s4, "stp">;
1806 def STPDpost : StorePairPostIdx<0b01, 1, FPR64, simm7s8, "stp">;
1807 def STPQpost : StorePairPostIdx<0b10, 1, FPR128, simm7s16, "stp">;
1809 // Pair (no allocate)
1810 defm STNPW : StorePairNoAlloc<0b00, 0, GPR32, simm7s4, "stnp">;
1811 defm STNPX : StorePairNoAlloc<0b10, 0, GPR64, simm7s8, "stnp">;
1812 defm STNPS : StorePairNoAlloc<0b00, 1, FPR32, simm7s4, "stnp">;
1813 defm STNPD : StorePairNoAlloc<0b01, 1, FPR64, simm7s8, "stnp">;
1814 defm STNPQ : StorePairNoAlloc<0b10, 1, FPR128, simm7s16, "stnp">;
1817 // (Register offset)
1820 defm STRBB : Store8RO< 0b00, 0, 0b00, GPR32, "strb", i32, truncstorei8>;
1821 defm STRHH : Store16RO<0b01, 0, 0b00, GPR32, "strh", i32, truncstorei16>;
1822 defm STRW : Store32RO<0b10, 0, 0b00, GPR32, "str", i32, store>;
1823 defm STRX : Store64RO<0b11, 0, 0b00, GPR64, "str", i64, store>;
1827 defm STRB : Store8RO< 0b00, 1, 0b00, FPR8, "str", untyped, store>;
1828 defm STRH : Store16RO<0b01, 1, 0b00, FPR16, "str", f16, store>;
1829 defm STRS : Store32RO<0b10, 1, 0b00, FPR32, "str", f32, store>;
1830 defm STRD : Store64RO<0b11, 1, 0b00, FPR64, "str", f64, store>;
1831 defm STRQ : Store128RO<0b00, 1, 0b10, FPR128, "str", f128, store>;
1833 multiclass TruncStoreFrom64ROPat<ROAddrMode ro, SDPatternOperator storeop,
1834 Instruction STRW, Instruction STRX> {
1836 def : Pat<(storeop GPR64:$Rt,
1837 (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)),
1838 (STRW (EXTRACT_SUBREG GPR64:$Rt, sub_32),
1839 GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
1841 def : Pat<(storeop GPR64:$Rt,
1842 (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)),
1843 (STRX (EXTRACT_SUBREG GPR64:$Rt, sub_32),
1844 GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
1847 let AddedComplexity = 10 in {
1849 defm : TruncStoreFrom64ROPat<ro8, truncstorei8, STRBBroW, STRBBroX>;
1850 defm : TruncStoreFrom64ROPat<ro16, truncstorei16, STRHHroW, STRHHroX>;
1851 defm : TruncStoreFrom64ROPat<ro32, truncstorei32, STRWroW, STRWroX>;
1854 multiclass VecROStorePat<ROAddrMode ro, ValueType VecTy, RegisterClass FPR,
1855 Instruction STRW, Instruction STRX> {
1856 def : Pat<(store (VecTy FPR:$Rt),
1857 (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)),
1858 (STRW FPR:$Rt, GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
1860 def : Pat<(store (VecTy FPR:$Rt),
1861 (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)),
1862 (STRX FPR:$Rt, GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
1865 let AddedComplexity = 10 in {
1866 // Match all store 64 bits width whose type is compatible with FPR64
1867 let Predicates = [IsLE] in {
1868 // We must use ST1 to store vectors in big-endian.
1869 defm : VecROStorePat<ro64, v2i32, FPR64, STRDroW, STRDroX>;
1870 defm : VecROStorePat<ro64, v2f32, FPR64, STRDroW, STRDroX>;
1871 defm : VecROStorePat<ro64, v4i16, FPR64, STRDroW, STRDroX>;
1872 defm : VecROStorePat<ro64, v8i8, FPR64, STRDroW, STRDroX>;
1873 defm : VecROStorePat<ro64, v4f16, FPR64, STRDroW, STRDroX>;
1876 defm : VecROStorePat<ro64, v1i64, FPR64, STRDroW, STRDroX>;
1877 defm : VecROStorePat<ro64, v1f64, FPR64, STRDroW, STRDroX>;
1879 // Match all store 128 bits width whose type is compatible with FPR128
1880 let Predicates = [IsLE] in {
1881 // We must use ST1 to store vectors in big-endian.
1882 defm : VecROStorePat<ro128, v2i64, FPR128, STRQroW, STRQroX>;
1883 defm : VecROStorePat<ro128, v2f64, FPR128, STRQroW, STRQroX>;
1884 defm : VecROStorePat<ro128, v4i32, FPR128, STRQroW, STRQroX>;
1885 defm : VecROStorePat<ro128, v4f32, FPR128, STRQroW, STRQroX>;
1886 defm : VecROStorePat<ro128, v8i16, FPR128, STRQroW, STRQroX>;
1887 defm : VecROStorePat<ro128, v16i8, FPR128, STRQroW, STRQroX>;
1888 defm : VecROStorePat<ro128, v8f16, FPR128, STRQroW, STRQroX>;
1890 } // AddedComplexity = 10
1892 // Match stores from lane 0 to the appropriate subreg's store.
1893 multiclass VecROStoreLane0Pat<ROAddrMode ro, SDPatternOperator storeop,
1894 ValueType VecTy, ValueType STy,
1895 SubRegIndex SubRegIdx,
1896 Instruction STRW, Instruction STRX> {
1898 def : Pat<(storeop (STy (vector_extract (VecTy VecListOne128:$Vt), 0)),
1899 (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)),
1900 (STRW (EXTRACT_SUBREG VecListOne128:$Vt, SubRegIdx),
1901 GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
1903 def : Pat<(storeop (STy (vector_extract (VecTy VecListOne128:$Vt), 0)),
1904 (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)),
1905 (STRX (EXTRACT_SUBREG VecListOne128:$Vt, SubRegIdx),
1906 GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
1909 let AddedComplexity = 19 in {
1910 defm : VecROStoreLane0Pat<ro16, truncstorei16, v8i16, i32, hsub, STRHroW, STRHroX>;
1911 defm : VecROStoreLane0Pat<ro16, store , v8i16, i16, hsub, STRHroW, STRHroX>;
1912 defm : VecROStoreLane0Pat<ro32, truncstorei32, v4i32, i32, ssub, STRSroW, STRSroX>;
1913 defm : VecROStoreLane0Pat<ro32, store , v4i32, i32, ssub, STRSroW, STRSroX>;
1914 defm : VecROStoreLane0Pat<ro32, store , v4f32, f32, ssub, STRSroW, STRSroX>;
1915 defm : VecROStoreLane0Pat<ro64, store , v2i64, i64, dsub, STRDroW, STRDroX>;
1916 defm : VecROStoreLane0Pat<ro64, store , v2f64, f64, dsub, STRDroW, STRDroX>;
1920 // (unsigned immediate)
1921 defm STRX : StoreUI<0b11, 0, 0b00, GPR64, uimm12s8, "str",
1923 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))]>;
1924 defm STRW : StoreUI<0b10, 0, 0b00, GPR32, uimm12s4, "str",
1926 (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))]>;
1927 defm STRB : StoreUI<0b00, 1, 0b00, FPR8, uimm12s1, "str",
1929 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))]>;
1930 defm STRH : StoreUI<0b01, 1, 0b00, FPR16, uimm12s2, "str",
1931 [(store (f16 FPR16:$Rt),
1932 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))]>;
1933 defm STRS : StoreUI<0b10, 1, 0b00, FPR32, uimm12s4, "str",
1934 [(store (f32 FPR32:$Rt),
1935 (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))]>;
1936 defm STRD : StoreUI<0b11, 1, 0b00, FPR64, uimm12s8, "str",
1937 [(store (f64 FPR64:$Rt),
1938 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))]>;
1939 defm STRQ : StoreUI<0b00, 1, 0b10, FPR128, uimm12s16, "str", []>;
1941 defm STRHH : StoreUI<0b01, 0, 0b00, GPR32, uimm12s2, "strh",
1942 [(truncstorei16 GPR32:$Rt,
1943 (am_indexed16 GPR64sp:$Rn,
1944 uimm12s2:$offset))]>;
1945 defm STRBB : StoreUI<0b00, 0, 0b00, GPR32, uimm12s1, "strb",
1946 [(truncstorei8 GPR32:$Rt,
1947 (am_indexed8 GPR64sp:$Rn,
1948 uimm12s1:$offset))]>;
1950 // Match all store 64 bits width whose type is compatible with FPR64
1951 let AddedComplexity = 10 in {
1952 let Predicates = [IsLE] in {
1953 // We must use ST1 to store vectors in big-endian.
1954 def : Pat<(store (v2f32 FPR64:$Rt),
1955 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
1956 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
1957 def : Pat<(store (v8i8 FPR64:$Rt),
1958 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
1959 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
1960 def : Pat<(store (v4i16 FPR64:$Rt),
1961 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
1962 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
1963 def : Pat<(store (v2i32 FPR64:$Rt),
1964 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
1965 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
1966 def : Pat<(store (v4f16 FPR64:$Rt),
1967 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
1968 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
1970 def : Pat<(store (v1f64 FPR64:$Rt),
1971 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
1972 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
1973 def : Pat<(store (v1i64 FPR64:$Rt),
1974 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
1975 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
1977 // Match all store 128 bits width whose type is compatible with FPR128
1978 let Predicates = [IsLE] in {
1979 // We must use ST1 to store vectors in big-endian.
1980 def : Pat<(store (v4f32 FPR128:$Rt),
1981 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
1982 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
1983 def : Pat<(store (v2f64 FPR128:$Rt),
1984 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
1985 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
1986 def : Pat<(store (v16i8 FPR128:$Rt),
1987 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
1988 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
1989 def : Pat<(store (v8i16 FPR128:$Rt),
1990 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
1991 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
1992 def : Pat<(store (v4i32 FPR128:$Rt),
1993 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
1994 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
1995 def : Pat<(store (v2i64 FPR128:$Rt),
1996 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
1997 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
1998 def : Pat<(store (v8f16 FPR128:$Rt),
1999 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
2000 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
2002 def : Pat<(store (f128 FPR128:$Rt),
2003 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
2004 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
2007 def : Pat<(truncstorei32 GPR64:$Rt,
2008 (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset)),
2009 (STRWui (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, uimm12s4:$offset)>;
2010 def : Pat<(truncstorei16 GPR64:$Rt,
2011 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset)),
2012 (STRHHui (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, uimm12s2:$offset)>;
2013 def : Pat<(truncstorei8 GPR64:$Rt, (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset)),
2014 (STRBBui (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, uimm12s1:$offset)>;
2016 } // AddedComplexity = 10
2019 // (unscaled immediate)
2020 defm STURX : StoreUnscaled<0b11, 0, 0b00, GPR64, "stur",
2022 (am_unscaled64 GPR64sp:$Rn, simm9:$offset))]>;
2023 defm STURW : StoreUnscaled<0b10, 0, 0b00, GPR32, "stur",
2025 (am_unscaled32 GPR64sp:$Rn, simm9:$offset))]>;
2026 defm STURB : StoreUnscaled<0b00, 1, 0b00, FPR8, "stur",
2028 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))]>;
2029 defm STURH : StoreUnscaled<0b01, 1, 0b00, FPR16, "stur",
2030 [(store (f16 FPR16:$Rt),
2031 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))]>;
2032 defm STURS : StoreUnscaled<0b10, 1, 0b00, FPR32, "stur",
2033 [(store (f32 FPR32:$Rt),
2034 (am_unscaled32 GPR64sp:$Rn, simm9:$offset))]>;
2035 defm STURD : StoreUnscaled<0b11, 1, 0b00, FPR64, "stur",
2036 [(store (f64 FPR64:$Rt),
2037 (am_unscaled64 GPR64sp:$Rn, simm9:$offset))]>;
2038 defm STURQ : StoreUnscaled<0b00, 1, 0b10, FPR128, "stur",
2039 [(store (f128 FPR128:$Rt),
2040 (am_unscaled128 GPR64sp:$Rn, simm9:$offset))]>;
2041 defm STURHH : StoreUnscaled<0b01, 0, 0b00, GPR32, "sturh",
2042 [(truncstorei16 GPR32:$Rt,
2043 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))]>;
2044 defm STURBB : StoreUnscaled<0b00, 0, 0b00, GPR32, "sturb",
2045 [(truncstorei8 GPR32:$Rt,
2046 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))]>;
2048 // Match all store 64 bits width whose type is compatible with FPR64
2049 let Predicates = [IsLE] in {
2050 // We must use ST1 to store vectors in big-endian.
2051 def : Pat<(store (v2f32 FPR64:$Rt),
2052 (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2053 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2054 def : Pat<(store (v8i8 FPR64:$Rt),
2055 (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2056 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2057 def : Pat<(store (v4i16 FPR64:$Rt),
2058 (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2059 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2060 def : Pat<(store (v2i32 FPR64:$Rt),
2061 (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2062 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2063 def : Pat<(store (v4f16 FPR64:$Rt),
2064 (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2065 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2067 def : Pat<(store (v1f64 FPR64:$Rt), (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2068 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2069 def : Pat<(store (v1i64 FPR64:$Rt), (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2070 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2072 // Match all store 128 bits width whose type is compatible with FPR128
2073 let Predicates = [IsLE] in {
2074 // We must use ST1 to store vectors in big-endian.
2075 def : Pat<(store (v4f32 FPR128:$Rt),
2076 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2077 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2078 def : Pat<(store (v2f64 FPR128:$Rt),
2079 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2080 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2081 def : Pat<(store (v16i8 FPR128:$Rt),
2082 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2083 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2084 def : Pat<(store (v8i16 FPR128:$Rt),
2085 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2086 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2087 def : Pat<(store (v4i32 FPR128:$Rt),
2088 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2089 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2090 def : Pat<(store (v2i64 FPR128:$Rt),
2091 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2092 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2093 def : Pat<(store (v2f64 FPR128:$Rt),
2094 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2095 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2096 def : Pat<(store (v8f16 FPR128:$Rt),
2097 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2098 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2101 // unscaled i64 truncating stores
2102 def : Pat<(truncstorei32 GPR64:$Rt, (am_unscaled32 GPR64sp:$Rn, simm9:$offset)),
2103 (STURWi (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, simm9:$offset)>;
2104 def : Pat<(truncstorei16 GPR64:$Rt, (am_unscaled16 GPR64sp:$Rn, simm9:$offset)),
2105 (STURHHi (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, simm9:$offset)>;
2106 def : Pat<(truncstorei8 GPR64:$Rt, (am_unscaled8 GPR64sp:$Rn, simm9:$offset)),
2107 (STURBBi (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, simm9:$offset)>;
2110 // STR mnemonics fall back to STUR for negative or unaligned offsets.
2111 def : InstAlias<"str $Rt, [$Rn, $offset]",
2112 (STURXi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb64:$offset), 0>;
2113 def : InstAlias<"str $Rt, [$Rn, $offset]",
2114 (STURWi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
2115 def : InstAlias<"str $Rt, [$Rn, $offset]",
2116 (STURBi FPR8:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
2117 def : InstAlias<"str $Rt, [$Rn, $offset]",
2118 (STURHi FPR16:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
2119 def : InstAlias<"str $Rt, [$Rn, $offset]",
2120 (STURSi FPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
2121 def : InstAlias<"str $Rt, [$Rn, $offset]",
2122 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb64:$offset), 0>;
2123 def : InstAlias<"str $Rt, [$Rn, $offset]",
2124 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9_offset_fb128:$offset), 0>;
2126 def : InstAlias<"strb $Rt, [$Rn, $offset]",
2127 (STURBBi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
2128 def : InstAlias<"strh $Rt, [$Rn, $offset]",
2129 (STURHHi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
2132 // (unscaled immediate, unprivileged)
2133 defm STTRW : StoreUnprivileged<0b10, 0, 0b00, GPR32, "sttr">;
2134 defm STTRX : StoreUnprivileged<0b11, 0, 0b00, GPR64, "sttr">;
2136 defm STTRH : StoreUnprivileged<0b01, 0, 0b00, GPR32, "sttrh">;
2137 defm STTRB : StoreUnprivileged<0b00, 0, 0b00, GPR32, "sttrb">;
2140 // (immediate pre-indexed)
2141 def STRWpre : StorePreIdx<0b10, 0, 0b00, GPR32, "str", pre_store, i32>;
2142 def STRXpre : StorePreIdx<0b11, 0, 0b00, GPR64, "str", pre_store, i64>;
2143 def STRBpre : StorePreIdx<0b00, 1, 0b00, FPR8, "str", pre_store, untyped>;
2144 def STRHpre : StorePreIdx<0b01, 1, 0b00, FPR16, "str", pre_store, f16>;
2145 def STRSpre : StorePreIdx<0b10, 1, 0b00, FPR32, "str", pre_store, f32>;
2146 def STRDpre : StorePreIdx<0b11, 1, 0b00, FPR64, "str", pre_store, f64>;
2147 def STRQpre : StorePreIdx<0b00, 1, 0b10, FPR128, "str", pre_store, f128>;
2149 def STRBBpre : StorePreIdx<0b00, 0, 0b00, GPR32, "strb", pre_truncsti8, i32>;
2150 def STRHHpre : StorePreIdx<0b01, 0, 0b00, GPR32, "strh", pre_truncsti16, i32>;
2153 def : Pat<(pre_truncsti32 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
2154 (STRWpre (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
2156 def : Pat<(pre_truncsti16 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
2157 (STRHHpre (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
2159 def : Pat<(pre_truncsti8 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
2160 (STRBBpre (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
2163 def : Pat<(pre_store (v8i8 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2164 (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2165 def : Pat<(pre_store (v4i16 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2166 (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2167 def : Pat<(pre_store (v2i32 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2168 (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2169 def : Pat<(pre_store (v2f32 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2170 (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2171 def : Pat<(pre_store (v1i64 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2172 (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2173 def : Pat<(pre_store (v1f64 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2174 (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2175 def : Pat<(pre_store (v4f16 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2176 (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2178 def : Pat<(pre_store (v16i8 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2179 (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2180 def : Pat<(pre_store (v8i16 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2181 (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2182 def : Pat<(pre_store (v4i32 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2183 (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2184 def : Pat<(pre_store (v4f32 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2185 (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2186 def : Pat<(pre_store (v2i64 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2187 (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2188 def : Pat<(pre_store (v2f64 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2189 (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2190 def : Pat<(pre_store (v8f16 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2191 (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2194 // (immediate post-indexed)
2195 def STRWpost : StorePostIdx<0b10, 0, 0b00, GPR32, "str", post_store, i32>;
2196 def STRXpost : StorePostIdx<0b11, 0, 0b00, GPR64, "str", post_store, i64>;
2197 def STRBpost : StorePostIdx<0b00, 1, 0b00, FPR8, "str", post_store, untyped>;
2198 def STRHpost : StorePostIdx<0b01, 1, 0b00, FPR16, "str", post_store, f16>;
2199 def STRSpost : StorePostIdx<0b10, 1, 0b00, FPR32, "str", post_store, f32>;
2200 def STRDpost : StorePostIdx<0b11, 1, 0b00, FPR64, "str", post_store, f64>;
2201 def STRQpost : StorePostIdx<0b00, 1, 0b10, FPR128, "str", post_store, f128>;
2203 def STRBBpost : StorePostIdx<0b00, 0, 0b00, GPR32, "strb", post_truncsti8, i32>;
2204 def STRHHpost : StorePostIdx<0b01, 0, 0b00, GPR32, "strh", post_truncsti16, i32>;
2207 def : Pat<(post_truncsti32 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
2208 (STRWpost (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
2210 def : Pat<(post_truncsti16 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
2211 (STRHHpost (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
2213 def : Pat<(post_truncsti8 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
2214 (STRBBpost (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
2217 def : Pat<(post_store (v8i8 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2218 (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2219 def : Pat<(post_store (v4i16 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2220 (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2221 def : Pat<(post_store (v2i32 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2222 (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2223 def : Pat<(post_store (v2f32 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2224 (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2225 def : Pat<(post_store (v1i64 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2226 (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2227 def : Pat<(post_store (v1f64 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2228 (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2229 def : Pat<(post_store (v4f16 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2230 (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2232 def : Pat<(post_store (v16i8 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2233 (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2234 def : Pat<(post_store (v8i16 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2235 (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2236 def : Pat<(post_store (v4i32 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2237 (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2238 def : Pat<(post_store (v4f32 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2239 (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2240 def : Pat<(post_store (v2i64 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2241 (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2242 def : Pat<(post_store (v2f64 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2243 (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2244 def : Pat<(post_store (v8f16 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2245 (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2247 //===----------------------------------------------------------------------===//
2248 // Load/store exclusive instructions.
2249 //===----------------------------------------------------------------------===//
2251 def LDARW : LoadAcquire <0b10, 1, 1, 0, 1, GPR32, "ldar">;
2252 def LDARX : LoadAcquire <0b11, 1, 1, 0, 1, GPR64, "ldar">;
2253 def LDARB : LoadAcquire <0b00, 1, 1, 0, 1, GPR32, "ldarb">;
2254 def LDARH : LoadAcquire <0b01, 1, 1, 0, 1, GPR32, "ldarh">;
2256 def LDAXRW : LoadExclusive <0b10, 0, 1, 0, 1, GPR32, "ldaxr">;
2257 def LDAXRX : LoadExclusive <0b11, 0, 1, 0, 1, GPR64, "ldaxr">;
2258 def LDAXRB : LoadExclusive <0b00, 0, 1, 0, 1, GPR32, "ldaxrb">;
2259 def LDAXRH : LoadExclusive <0b01, 0, 1, 0, 1, GPR32, "ldaxrh">;
2261 def LDXRW : LoadExclusive <0b10, 0, 1, 0, 0, GPR32, "ldxr">;
2262 def LDXRX : LoadExclusive <0b11, 0, 1, 0, 0, GPR64, "ldxr">;
2263 def LDXRB : LoadExclusive <0b00, 0, 1, 0, 0, GPR32, "ldxrb">;
2264 def LDXRH : LoadExclusive <0b01, 0, 1, 0, 0, GPR32, "ldxrh">;
2266 def STLRW : StoreRelease <0b10, 1, 0, 0, 1, GPR32, "stlr">;
2267 def STLRX : StoreRelease <0b11, 1, 0, 0, 1, GPR64, "stlr">;
2268 def STLRB : StoreRelease <0b00, 1, 0, 0, 1, GPR32, "stlrb">;
2269 def STLRH : StoreRelease <0b01, 1, 0, 0, 1, GPR32, "stlrh">;
2271 def STLXRW : StoreExclusive<0b10, 0, 0, 0, 1, GPR32, "stlxr">;
2272 def STLXRX : StoreExclusive<0b11, 0, 0, 0, 1, GPR64, "stlxr">;
2273 def STLXRB : StoreExclusive<0b00, 0, 0, 0, 1, GPR32, "stlxrb">;
2274 def STLXRH : StoreExclusive<0b01, 0, 0, 0, 1, GPR32, "stlxrh">;
2276 def STXRW : StoreExclusive<0b10, 0, 0, 0, 0, GPR32, "stxr">;
2277 def STXRX : StoreExclusive<0b11, 0, 0, 0, 0, GPR64, "stxr">;
2278 def STXRB : StoreExclusive<0b00, 0, 0, 0, 0, GPR32, "stxrb">;
2279 def STXRH : StoreExclusive<0b01, 0, 0, 0, 0, GPR32, "stxrh">;
2281 def LDAXPW : LoadExclusivePair<0b10, 0, 1, 1, 1, GPR32, "ldaxp">;
2282 def LDAXPX : LoadExclusivePair<0b11, 0, 1, 1, 1, GPR64, "ldaxp">;
2284 def LDXPW : LoadExclusivePair<0b10, 0, 1, 1, 0, GPR32, "ldxp">;
2285 def LDXPX : LoadExclusivePair<0b11, 0, 1, 1, 0, GPR64, "ldxp">;
2287 def STLXPW : StoreExclusivePair<0b10, 0, 0, 1, 1, GPR32, "stlxp">;
2288 def STLXPX : StoreExclusivePair<0b11, 0, 0, 1, 1, GPR64, "stlxp">;
2290 def STXPW : StoreExclusivePair<0b10, 0, 0, 1, 0, GPR32, "stxp">;
2291 def STXPX : StoreExclusivePair<0b11, 0, 0, 1, 0, GPR64, "stxp">;
2293 //===----------------------------------------------------------------------===//
2294 // Scaled floating point to integer conversion instructions.
2295 //===----------------------------------------------------------------------===//
2297 defm FCVTAS : FPToIntegerUnscaled<0b00, 0b100, "fcvtas", int_aarch64_neon_fcvtas>;
2298 defm FCVTAU : FPToIntegerUnscaled<0b00, 0b101, "fcvtau", int_aarch64_neon_fcvtau>;
2299 defm FCVTMS : FPToIntegerUnscaled<0b10, 0b000, "fcvtms", int_aarch64_neon_fcvtms>;
2300 defm FCVTMU : FPToIntegerUnscaled<0b10, 0b001, "fcvtmu", int_aarch64_neon_fcvtmu>;
2301 defm FCVTNS : FPToIntegerUnscaled<0b00, 0b000, "fcvtns", int_aarch64_neon_fcvtns>;
2302 defm FCVTNU : FPToIntegerUnscaled<0b00, 0b001, "fcvtnu", int_aarch64_neon_fcvtnu>;
2303 defm FCVTPS : FPToIntegerUnscaled<0b01, 0b000, "fcvtps", int_aarch64_neon_fcvtps>;
2304 defm FCVTPU : FPToIntegerUnscaled<0b01, 0b001, "fcvtpu", int_aarch64_neon_fcvtpu>;
2305 defm FCVTZS : FPToIntegerUnscaled<0b11, 0b000, "fcvtzs", fp_to_sint>;
2306 defm FCVTZU : FPToIntegerUnscaled<0b11, 0b001, "fcvtzu", fp_to_uint>;
2307 defm FCVTZS : FPToIntegerScaled<0b11, 0b000, "fcvtzs", fp_to_sint>;
2308 defm FCVTZU : FPToIntegerScaled<0b11, 0b001, "fcvtzu", fp_to_uint>;
2309 let isCodeGenOnly = 1 in {
2310 defm FCVTZS_Int : FPToIntegerUnscaled<0b11, 0b000, "fcvtzs", int_aarch64_neon_fcvtzs>;
2311 defm FCVTZU_Int : FPToIntegerUnscaled<0b11, 0b001, "fcvtzu", int_aarch64_neon_fcvtzu>;
2312 defm FCVTZS_Int : FPToIntegerScaled<0b11, 0b000, "fcvtzs", int_aarch64_neon_fcvtzs>;
2313 defm FCVTZU_Int : FPToIntegerScaled<0b11, 0b001, "fcvtzu", int_aarch64_neon_fcvtzu>;
2316 //===----------------------------------------------------------------------===//
2317 // Scaled integer to floating point conversion instructions.
2318 //===----------------------------------------------------------------------===//
2320 defm SCVTF : IntegerToFP<0, "scvtf", sint_to_fp>;
2321 defm UCVTF : IntegerToFP<1, "ucvtf", uint_to_fp>;
2323 //===----------------------------------------------------------------------===//
2324 // Unscaled integer to floating point conversion instruction.
2325 //===----------------------------------------------------------------------===//
2327 defm FMOV : UnscaledConversion<"fmov">;
2329 def : Pat<(f32 (fpimm0)), (FMOVWSr WZR)>, Requires<[NoZCZ]>;
2330 def : Pat<(f64 (fpimm0)), (FMOVXDr XZR)>, Requires<[NoZCZ]>;
2332 //===----------------------------------------------------------------------===//
2333 // Floating point conversion instruction.
2334 //===----------------------------------------------------------------------===//
2336 defm FCVT : FPConversion<"fcvt">;
2338 //===----------------------------------------------------------------------===//
2339 // Floating point single operand instructions.
2340 //===----------------------------------------------------------------------===//
2342 defm FABS : SingleOperandFPData<0b0001, "fabs", fabs>;
2343 defm FMOV : SingleOperandFPData<0b0000, "fmov">;
2344 defm FNEG : SingleOperandFPData<0b0010, "fneg", fneg>;
2345 defm FRINTA : SingleOperandFPData<0b1100, "frinta", frnd>;
2346 defm FRINTI : SingleOperandFPData<0b1111, "frinti", fnearbyint>;
2347 defm FRINTM : SingleOperandFPData<0b1010, "frintm", ffloor>;
2348 defm FRINTN : SingleOperandFPData<0b1000, "frintn", int_aarch64_neon_frintn>;
2349 defm FRINTP : SingleOperandFPData<0b1001, "frintp", fceil>;
2351 def : Pat<(v1f64 (int_aarch64_neon_frintn (v1f64 FPR64:$Rn))),
2352 (FRINTNDr FPR64:$Rn)>;
2354 // FRINTX is inserted to set the flags as required by FENV_ACCESS ON behavior
2355 // in the C spec. Setting hasSideEffects ensures it is not DCE'd.
2356 // <rdar://problem/13715968>
2357 // TODO: We should really model the FPSR flags correctly. This is really ugly.
2358 let hasSideEffects = 1 in {
2359 defm FRINTX : SingleOperandFPData<0b1110, "frintx", frint>;
2362 defm FRINTZ : SingleOperandFPData<0b1011, "frintz", ftrunc>;
2364 let SchedRW = [WriteFDiv] in {
2365 defm FSQRT : SingleOperandFPData<0b0011, "fsqrt", fsqrt>;
2368 //===----------------------------------------------------------------------===//
2369 // Floating point two operand instructions.
2370 //===----------------------------------------------------------------------===//
2372 defm FADD : TwoOperandFPData<0b0010, "fadd", fadd>;
2373 let SchedRW = [WriteFDiv] in {
2374 defm FDIV : TwoOperandFPData<0b0001, "fdiv", fdiv>;
2376 defm FMAXNM : TwoOperandFPData<0b0110, "fmaxnm", int_aarch64_neon_fmaxnm>;
2377 defm FMAX : TwoOperandFPData<0b0100, "fmax", AArch64fmax>;
2378 defm FMINNM : TwoOperandFPData<0b0111, "fminnm", int_aarch64_neon_fminnm>;
2379 defm FMIN : TwoOperandFPData<0b0101, "fmin", AArch64fmin>;
2380 let SchedRW = [WriteFMul] in {
2381 defm FMUL : TwoOperandFPData<0b0000, "fmul", fmul>;
2382 defm FNMUL : TwoOperandFPDataNeg<0b1000, "fnmul", fmul>;
2384 defm FSUB : TwoOperandFPData<0b0011, "fsub", fsub>;
2386 def : Pat<(v1f64 (AArch64fmax (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2387 (FMAXDrr FPR64:$Rn, FPR64:$Rm)>;
2388 def : Pat<(v1f64 (AArch64fmin (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2389 (FMINDrr FPR64:$Rn, FPR64:$Rm)>;
2390 def : Pat<(v1f64 (int_aarch64_neon_fmaxnm (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2391 (FMAXNMDrr FPR64:$Rn, FPR64:$Rm)>;
2392 def : Pat<(v1f64 (int_aarch64_neon_fminnm (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2393 (FMINNMDrr FPR64:$Rn, FPR64:$Rm)>;
2395 //===----------------------------------------------------------------------===//
2396 // Floating point three operand instructions.
2397 //===----------------------------------------------------------------------===//
2399 defm FMADD : ThreeOperandFPData<0, 0, "fmadd", fma>;
2400 defm FMSUB : ThreeOperandFPData<0, 1, "fmsub",
2401 TriOpFrag<(fma node:$LHS, (fneg node:$MHS), node:$RHS)> >;
2402 defm FNMADD : ThreeOperandFPData<1, 0, "fnmadd",
2403 TriOpFrag<(fneg (fma node:$LHS, node:$MHS, node:$RHS))> >;
2404 defm FNMSUB : ThreeOperandFPData<1, 1, "fnmsub",
2405 TriOpFrag<(fma node:$LHS, node:$MHS, (fneg node:$RHS))> >;
2407 // The following def pats catch the case where the LHS of an FMA is negated.
2408 // The TriOpFrag above catches the case where the middle operand is negated.
2410 // N.b. FMSUB etc have the accumulator at the *end* of (outs), unlike
2411 // the NEON variant.
2412 def : Pat<(f32 (fma (fneg FPR32:$Rn), FPR32:$Rm, FPR32:$Ra)),
2413 (FMSUBSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
2415 def : Pat<(f64 (fma (fneg FPR64:$Rn), FPR64:$Rm, FPR64:$Ra)),
2416 (FMSUBDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
2418 // We handled -(a + b*c) for FNMADD above, now it's time for "(-a) + (-b)*c" and
2420 def : Pat<(f32 (fma (fneg FPR32:$Rn), FPR32:$Rm, (fneg FPR32:$Ra))),
2421 (FNMADDSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
2423 def : Pat<(f64 (fma (fneg FPR64:$Rn), FPR64:$Rm, (fneg FPR64:$Ra))),
2424 (FNMADDDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
2426 def : Pat<(f32 (fma FPR32:$Rn, (fneg FPR32:$Rm), (fneg FPR32:$Ra))),
2427 (FNMADDSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
2429 def : Pat<(f64 (fma FPR64:$Rn, (fneg FPR64:$Rm), (fneg FPR64:$Ra))),
2430 (FNMADDDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
2432 //===----------------------------------------------------------------------===//
2433 // Floating point comparison instructions.
2434 //===----------------------------------------------------------------------===//
2436 defm FCMPE : FPComparison<1, "fcmpe">;
2437 defm FCMP : FPComparison<0, "fcmp", AArch64fcmp>;
2439 //===----------------------------------------------------------------------===//
2440 // Floating point conditional comparison instructions.
2441 //===----------------------------------------------------------------------===//
2443 defm FCCMPE : FPCondComparison<1, "fccmpe">;
2444 defm FCCMP : FPCondComparison<0, "fccmp">;
2446 //===----------------------------------------------------------------------===//
2447 // Floating point conditional select instruction.
2448 //===----------------------------------------------------------------------===//
2450 defm FCSEL : FPCondSelect<"fcsel">;
2452 // CSEL instructions providing f128 types need to be handled by a
2453 // pseudo-instruction since the eventual code will need to introduce basic
2454 // blocks and control flow.
2455 def F128CSEL : Pseudo<(outs FPR128:$Rd),
2456 (ins FPR128:$Rn, FPR128:$Rm, ccode:$cond),
2457 [(set (f128 FPR128:$Rd),
2458 (AArch64csel FPR128:$Rn, FPR128:$Rm,
2459 (i32 imm:$cond), NZCV))]> {
2461 let usesCustomInserter = 1;
2465 //===----------------------------------------------------------------------===//
2466 // Floating point immediate move.
2467 //===----------------------------------------------------------------------===//
2469 let isReMaterializable = 1 in {
2470 defm FMOV : FPMoveImmediate<"fmov">;
2473 //===----------------------------------------------------------------------===//
2474 // Advanced SIMD two vector instructions.
2475 //===----------------------------------------------------------------------===//
2477 defm ABS : SIMDTwoVectorBHSD<0, 0b01011, "abs", int_aarch64_neon_abs>;
2478 def : Pat<(xor (v8i8 (AArch64vashr V64:$src, (i32 7))),
2479 (v8i8 (add V64:$src, (AArch64vashr V64:$src, (i32 7))))),
2480 (ABSv8i8 V64:$src)>;
2481 def : Pat<(xor (v4i16 (AArch64vashr V64:$src, (i32 15))),
2482 (v4i16 (add V64:$src, (AArch64vashr V64:$src, (i32 15))))),
2483 (ABSv4i16 V64:$src)>;
2484 def : Pat<(xor (v2i32 (AArch64vashr V64:$src, (i32 31))),
2485 (v2i32 (add V64:$src, (AArch64vashr V64:$src, (i32 31))))),
2486 (ABSv2i32 V64:$src)>;
2487 def : Pat<(xor (v16i8 (AArch64vashr V128:$src, (i32 7))),
2488 (v16i8 (add V128:$src, (AArch64vashr V128:$src, (i32 7))))),
2489 (ABSv16i8 V128:$src)>;
2490 def : Pat<(xor (v8i16 (AArch64vashr V128:$src, (i32 15))),
2491 (v8i16 (add V128:$src, (AArch64vashr V128:$src, (i32 15))))),
2492 (ABSv8i16 V128:$src)>;
2493 def : Pat<(xor (v4i32 (AArch64vashr V128:$src, (i32 31))),
2494 (v4i32 (add V128:$src, (AArch64vashr V128:$src, (i32 31))))),
2495 (ABSv4i32 V128:$src)>;
2496 def : Pat<(xor (v2i64 (AArch64vashr V128:$src, (i32 63))),
2497 (v2i64 (add V128:$src, (AArch64vashr V128:$src, (i32 63))))),
2498 (ABSv2i64 V128:$src)>;
2500 defm CLS : SIMDTwoVectorBHS<0, 0b00100, "cls", int_aarch64_neon_cls>;
2501 defm CLZ : SIMDTwoVectorBHS<1, 0b00100, "clz", ctlz>;
2502 defm CMEQ : SIMDCmpTwoVector<0, 0b01001, "cmeq", AArch64cmeqz>;
2503 defm CMGE : SIMDCmpTwoVector<1, 0b01000, "cmge", AArch64cmgez>;
2504 defm CMGT : SIMDCmpTwoVector<0, 0b01000, "cmgt", AArch64cmgtz>;
2505 defm CMLE : SIMDCmpTwoVector<1, 0b01001, "cmle", AArch64cmlez>;
2506 defm CMLT : SIMDCmpTwoVector<0, 0b01010, "cmlt", AArch64cmltz>;
2507 defm CNT : SIMDTwoVectorB<0, 0b00, 0b00101, "cnt", ctpop>;
2508 defm FABS : SIMDTwoVectorFP<0, 1, 0b01111, "fabs", fabs>;
2510 defm FCMEQ : SIMDFPCmpTwoVector<0, 1, 0b01101, "fcmeq", AArch64fcmeqz>;
2511 defm FCMGE : SIMDFPCmpTwoVector<1, 1, 0b01100, "fcmge", AArch64fcmgez>;
2512 defm FCMGT : SIMDFPCmpTwoVector<0, 1, 0b01100, "fcmgt", AArch64fcmgtz>;
2513 defm FCMLE : SIMDFPCmpTwoVector<1, 1, 0b01101, "fcmle", AArch64fcmlez>;
2514 defm FCMLT : SIMDFPCmpTwoVector<0, 1, 0b01110, "fcmlt", AArch64fcmltz>;
2515 defm FCVTAS : SIMDTwoVectorFPToInt<0,0,0b11100, "fcvtas",int_aarch64_neon_fcvtas>;
2516 defm FCVTAU : SIMDTwoVectorFPToInt<1,0,0b11100, "fcvtau",int_aarch64_neon_fcvtau>;
2517 defm FCVTL : SIMDFPWidenTwoVector<0, 0, 0b10111, "fcvtl">;
2518 def : Pat<(v4f32 (int_aarch64_neon_vcvthf2fp (v4i16 V64:$Rn))),
2519 (FCVTLv4i16 V64:$Rn)>;
2520 def : Pat<(v4f32 (int_aarch64_neon_vcvthf2fp (extract_subvector (v8i16 V128:$Rn),
2522 (FCVTLv8i16 V128:$Rn)>;
2523 def : Pat<(v2f64 (fextend (v2f32 V64:$Rn))), (FCVTLv2i32 V64:$Rn)>;
2524 def : Pat<(v2f64 (fextend (v2f32 (extract_subvector (v4f32 V128:$Rn),
2526 (FCVTLv4i32 V128:$Rn)>;
2528 def : Pat<(v4f32 (fextend (v4f16 V64:$Rn))), (FCVTLv4i16 V64:$Rn)>;
2529 def : Pat<(v4f32 (fextend (v4f16 (extract_subvector (v8f16 V128:$Rn),
2531 (FCVTLv8i16 V128:$Rn)>;
2533 defm FCVTMS : SIMDTwoVectorFPToInt<0,0,0b11011, "fcvtms",int_aarch64_neon_fcvtms>;
2534 defm FCVTMU : SIMDTwoVectorFPToInt<1,0,0b11011, "fcvtmu",int_aarch64_neon_fcvtmu>;
2535 defm FCVTNS : SIMDTwoVectorFPToInt<0,0,0b11010, "fcvtns",int_aarch64_neon_fcvtns>;
2536 defm FCVTNU : SIMDTwoVectorFPToInt<1,0,0b11010, "fcvtnu",int_aarch64_neon_fcvtnu>;
2537 defm FCVTN : SIMDFPNarrowTwoVector<0, 0, 0b10110, "fcvtn">;
2538 def : Pat<(v4i16 (int_aarch64_neon_vcvtfp2hf (v4f32 V128:$Rn))),
2539 (FCVTNv4i16 V128:$Rn)>;
2540 def : Pat<(concat_vectors V64:$Rd,
2541 (v4i16 (int_aarch64_neon_vcvtfp2hf (v4f32 V128:$Rn)))),
2542 (FCVTNv8i16 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
2543 def : Pat<(v2f32 (fround (v2f64 V128:$Rn))), (FCVTNv2i32 V128:$Rn)>;
2544 def : Pat<(v4f16 (fround (v4f32 V128:$Rn))), (FCVTNv4i16 V128:$Rn)>;
2545 def : Pat<(concat_vectors V64:$Rd, (v2f32 (fround (v2f64 V128:$Rn)))),
2546 (FCVTNv4i32 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
2547 defm FCVTPS : SIMDTwoVectorFPToInt<0,1,0b11010, "fcvtps",int_aarch64_neon_fcvtps>;
2548 defm FCVTPU : SIMDTwoVectorFPToInt<1,1,0b11010, "fcvtpu",int_aarch64_neon_fcvtpu>;
2549 defm FCVTXN : SIMDFPInexactCvtTwoVector<1, 0, 0b10110, "fcvtxn",
2550 int_aarch64_neon_fcvtxn>;
2551 defm FCVTZS : SIMDTwoVectorFPToInt<0, 1, 0b11011, "fcvtzs", fp_to_sint>;
2552 defm FCVTZU : SIMDTwoVectorFPToInt<1, 1, 0b11011, "fcvtzu", fp_to_uint>;
2553 let isCodeGenOnly = 1 in {
2554 defm FCVTZS_Int : SIMDTwoVectorFPToInt<0, 1, 0b11011, "fcvtzs",
2555 int_aarch64_neon_fcvtzs>;
2556 defm FCVTZU_Int : SIMDTwoVectorFPToInt<1, 1, 0b11011, "fcvtzu",
2557 int_aarch64_neon_fcvtzu>;
2559 defm FNEG : SIMDTwoVectorFP<1, 1, 0b01111, "fneg", fneg>;
2560 defm FRECPE : SIMDTwoVectorFP<0, 1, 0b11101, "frecpe", int_aarch64_neon_frecpe>;
2561 defm FRINTA : SIMDTwoVectorFP<1, 0, 0b11000, "frinta", frnd>;
2562 defm FRINTI : SIMDTwoVectorFP<1, 1, 0b11001, "frinti", fnearbyint>;
2563 defm FRINTM : SIMDTwoVectorFP<0, 0, 0b11001, "frintm", ffloor>;
2564 defm FRINTN : SIMDTwoVectorFP<0, 0, 0b11000, "frintn", int_aarch64_neon_frintn>;
2565 defm FRINTP : SIMDTwoVectorFP<0, 1, 0b11000, "frintp", fceil>;
2566 defm FRINTX : SIMDTwoVectorFP<1, 0, 0b11001, "frintx", frint>;
2567 defm FRINTZ : SIMDTwoVectorFP<0, 1, 0b11001, "frintz", ftrunc>;
2568 defm FRSQRTE: SIMDTwoVectorFP<1, 1, 0b11101, "frsqrte", int_aarch64_neon_frsqrte>;
2569 defm FSQRT : SIMDTwoVectorFP<1, 1, 0b11111, "fsqrt", fsqrt>;
2570 defm NEG : SIMDTwoVectorBHSD<1, 0b01011, "neg",
2571 UnOpFrag<(sub immAllZerosV, node:$LHS)> >;
2572 defm NOT : SIMDTwoVectorB<1, 0b00, 0b00101, "not", vnot>;
2573 // Aliases for MVN -> NOT.
2574 def : InstAlias<"mvn{ $Vd.8b, $Vn.8b|.8b $Vd, $Vn}",
2575 (NOTv8i8 V64:$Vd, V64:$Vn)>;
2576 def : InstAlias<"mvn{ $Vd.16b, $Vn.16b|.16b $Vd, $Vn}",
2577 (NOTv16i8 V128:$Vd, V128:$Vn)>;
2579 def : Pat<(AArch64neg (v8i8 V64:$Rn)), (NEGv8i8 V64:$Rn)>;
2580 def : Pat<(AArch64neg (v16i8 V128:$Rn)), (NEGv16i8 V128:$Rn)>;
2581 def : Pat<(AArch64neg (v4i16 V64:$Rn)), (NEGv4i16 V64:$Rn)>;
2582 def : Pat<(AArch64neg (v8i16 V128:$Rn)), (NEGv8i16 V128:$Rn)>;
2583 def : Pat<(AArch64neg (v2i32 V64:$Rn)), (NEGv2i32 V64:$Rn)>;
2584 def : Pat<(AArch64neg (v4i32 V128:$Rn)), (NEGv4i32 V128:$Rn)>;
2585 def : Pat<(AArch64neg (v2i64 V128:$Rn)), (NEGv2i64 V128:$Rn)>;
2587 def : Pat<(AArch64not (v8i8 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2588 def : Pat<(AArch64not (v16i8 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2589 def : Pat<(AArch64not (v4i16 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2590 def : Pat<(AArch64not (v8i16 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2591 def : Pat<(AArch64not (v2i32 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2592 def : Pat<(AArch64not (v1i64 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2593 def : Pat<(AArch64not (v4i32 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2594 def : Pat<(AArch64not (v2i64 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2596 def : Pat<(vnot (v4i16 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2597 def : Pat<(vnot (v8i16 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2598 def : Pat<(vnot (v2i32 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2599 def : Pat<(vnot (v4i32 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2600 def : Pat<(vnot (v2i64 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2602 defm RBIT : SIMDTwoVectorB<1, 0b01, 0b00101, "rbit", int_aarch64_neon_rbit>;
2603 defm REV16 : SIMDTwoVectorB<0, 0b00, 0b00001, "rev16", AArch64rev16>;
2604 defm REV32 : SIMDTwoVectorBH<1, 0b00000, "rev32", AArch64rev32>;
2605 defm REV64 : SIMDTwoVectorBHS<0, 0b00000, "rev64", AArch64rev64>;
2606 defm SADALP : SIMDLongTwoVectorTied<0, 0b00110, "sadalp",
2607 BinOpFrag<(add node:$LHS, (int_aarch64_neon_saddlp node:$RHS))> >;
2608 defm SADDLP : SIMDLongTwoVector<0, 0b00010, "saddlp", int_aarch64_neon_saddlp>;
2609 defm SCVTF : SIMDTwoVectorIntToFP<0, 0, 0b11101, "scvtf", sint_to_fp>;
2610 defm SHLL : SIMDVectorLShiftLongBySizeBHS;
2611 defm SQABS : SIMDTwoVectorBHSD<0, 0b00111, "sqabs", int_aarch64_neon_sqabs>;
2612 defm SQNEG : SIMDTwoVectorBHSD<1, 0b00111, "sqneg", int_aarch64_neon_sqneg>;
2613 defm SQXTN : SIMDMixedTwoVector<0, 0b10100, "sqxtn", int_aarch64_neon_sqxtn>;
2614 defm SQXTUN : SIMDMixedTwoVector<1, 0b10010, "sqxtun", int_aarch64_neon_sqxtun>;
2615 defm SUQADD : SIMDTwoVectorBHSDTied<0, 0b00011, "suqadd",int_aarch64_neon_suqadd>;
2616 defm UADALP : SIMDLongTwoVectorTied<1, 0b00110, "uadalp",
2617 BinOpFrag<(add node:$LHS, (int_aarch64_neon_uaddlp node:$RHS))> >;
2618 defm UADDLP : SIMDLongTwoVector<1, 0b00010, "uaddlp",
2619 int_aarch64_neon_uaddlp>;
2620 defm UCVTF : SIMDTwoVectorIntToFP<1, 0, 0b11101, "ucvtf", uint_to_fp>;
2621 defm UQXTN : SIMDMixedTwoVector<1, 0b10100, "uqxtn", int_aarch64_neon_uqxtn>;
2622 defm URECPE : SIMDTwoVectorS<0, 1, 0b11100, "urecpe", int_aarch64_neon_urecpe>;
2623 defm URSQRTE: SIMDTwoVectorS<1, 1, 0b11100, "ursqrte", int_aarch64_neon_ursqrte>;
2624 defm USQADD : SIMDTwoVectorBHSDTied<1, 0b00011, "usqadd",int_aarch64_neon_usqadd>;
2625 defm XTN : SIMDMixedTwoVector<0, 0b10010, "xtn", trunc>;
2627 def : Pat<(v4f16 (AArch64rev32 V64:$Rn)), (REV32v4i16 V64:$Rn)>;
2628 def : Pat<(v4f16 (AArch64rev64 V64:$Rn)), (REV64v4i16 V64:$Rn)>;
2629 def : Pat<(v8f16 (AArch64rev32 V128:$Rn)), (REV32v8i16 V128:$Rn)>;
2630 def : Pat<(v8f16 (AArch64rev64 V128:$Rn)), (REV64v8i16 V128:$Rn)>;
2631 def : Pat<(v2f32 (AArch64rev64 V64:$Rn)), (REV64v2i32 V64:$Rn)>;
2632 def : Pat<(v4f32 (AArch64rev64 V128:$Rn)), (REV64v4i32 V128:$Rn)>;
2634 // Patterns for vector long shift (by element width). These need to match all
2635 // three of zext, sext and anyext so it's easier to pull the patterns out of the
2637 multiclass SIMDVectorLShiftLongBySizeBHSPats<SDPatternOperator ext> {
2638 def : Pat<(AArch64vshl (v8i16 (ext (v8i8 V64:$Rn))), (i32 8)),
2639 (SHLLv8i8 V64:$Rn)>;
2640 def : Pat<(AArch64vshl (v8i16 (ext (extract_high_v16i8 V128:$Rn))), (i32 8)),
2641 (SHLLv16i8 V128:$Rn)>;
2642 def : Pat<(AArch64vshl (v4i32 (ext (v4i16 V64:$Rn))), (i32 16)),
2643 (SHLLv4i16 V64:$Rn)>;
2644 def : Pat<(AArch64vshl (v4i32 (ext (extract_high_v8i16 V128:$Rn))), (i32 16)),
2645 (SHLLv8i16 V128:$Rn)>;
2646 def : Pat<(AArch64vshl (v2i64 (ext (v2i32 V64:$Rn))), (i32 32)),
2647 (SHLLv2i32 V64:$Rn)>;
2648 def : Pat<(AArch64vshl (v2i64 (ext (extract_high_v4i32 V128:$Rn))), (i32 32)),
2649 (SHLLv4i32 V128:$Rn)>;
2652 defm : SIMDVectorLShiftLongBySizeBHSPats<anyext>;
2653 defm : SIMDVectorLShiftLongBySizeBHSPats<zext>;
2654 defm : SIMDVectorLShiftLongBySizeBHSPats<sext>;
2656 //===----------------------------------------------------------------------===//
2657 // Advanced SIMD three vector instructions.
2658 //===----------------------------------------------------------------------===//
2660 defm ADD : SIMDThreeSameVector<0, 0b10000, "add", add>;
2661 defm ADDP : SIMDThreeSameVector<0, 0b10111, "addp", int_aarch64_neon_addp>;
2662 defm CMEQ : SIMDThreeSameVector<1, 0b10001, "cmeq", AArch64cmeq>;
2663 defm CMGE : SIMDThreeSameVector<0, 0b00111, "cmge", AArch64cmge>;
2664 defm CMGT : SIMDThreeSameVector<0, 0b00110, "cmgt", AArch64cmgt>;
2665 defm CMHI : SIMDThreeSameVector<1, 0b00110, "cmhi", AArch64cmhi>;
2666 defm CMHS : SIMDThreeSameVector<1, 0b00111, "cmhs", AArch64cmhs>;
2667 defm CMTST : SIMDThreeSameVector<0, 0b10001, "cmtst", AArch64cmtst>;
2668 defm FABD : SIMDThreeSameVectorFP<1,1,0b11010,"fabd", int_aarch64_neon_fabd>;
2669 defm FACGE : SIMDThreeSameVectorFPCmp<1,0,0b11101,"facge",int_aarch64_neon_facge>;
2670 defm FACGT : SIMDThreeSameVectorFPCmp<1,1,0b11101,"facgt",int_aarch64_neon_facgt>;
2671 defm FADDP : SIMDThreeSameVectorFP<1,0,0b11010,"faddp",int_aarch64_neon_addp>;
2672 defm FADD : SIMDThreeSameVectorFP<0,0,0b11010,"fadd", fadd>;
2673 defm FCMEQ : SIMDThreeSameVectorFPCmp<0, 0, 0b11100, "fcmeq", AArch64fcmeq>;
2674 defm FCMGE : SIMDThreeSameVectorFPCmp<1, 0, 0b11100, "fcmge", AArch64fcmge>;
2675 defm FCMGT : SIMDThreeSameVectorFPCmp<1, 1, 0b11100, "fcmgt", AArch64fcmgt>;
2676 defm FDIV : SIMDThreeSameVectorFP<1,0,0b11111,"fdiv", fdiv>;
2677 defm FMAXNMP : SIMDThreeSameVectorFP<1,0,0b11000,"fmaxnmp", int_aarch64_neon_fmaxnmp>;
2678 defm FMAXNM : SIMDThreeSameVectorFP<0,0,0b11000,"fmaxnm", int_aarch64_neon_fmaxnm>;
2679 defm FMAXP : SIMDThreeSameVectorFP<1,0,0b11110,"fmaxp", int_aarch64_neon_fmaxp>;
2680 defm FMAX : SIMDThreeSameVectorFP<0,0,0b11110,"fmax", AArch64fmax>;
2681 defm FMINNMP : SIMDThreeSameVectorFP<1,1,0b11000,"fminnmp", int_aarch64_neon_fminnmp>;
2682 defm FMINNM : SIMDThreeSameVectorFP<0,1,0b11000,"fminnm", int_aarch64_neon_fminnm>;
2683 defm FMINP : SIMDThreeSameVectorFP<1,1,0b11110,"fminp", int_aarch64_neon_fminp>;
2684 defm FMIN : SIMDThreeSameVectorFP<0,1,0b11110,"fmin", AArch64fmin>;
2686 // NOTE: The operands of the PatFrag are reordered on FMLA/FMLS because the
2687 // instruction expects the addend first, while the fma intrinsic puts it last.
2688 defm FMLA : SIMDThreeSameVectorFPTied<0, 0, 0b11001, "fmla",
2689 TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)> >;
2690 defm FMLS : SIMDThreeSameVectorFPTied<0, 1, 0b11001, "fmls",
2691 TriOpFrag<(fma node:$MHS, (fneg node:$RHS), node:$LHS)> >;
2693 // The following def pats catch the case where the LHS of an FMA is negated.
2694 // The TriOpFrag above catches the case where the middle operand is negated.
2695 def : Pat<(v2f32 (fma (fneg V64:$Rn), V64:$Rm, V64:$Rd)),
2696 (FMLSv2f32 V64:$Rd, V64:$Rn, V64:$Rm)>;
2698 def : Pat<(v4f32 (fma (fneg V128:$Rn), V128:$Rm, V128:$Rd)),
2699 (FMLSv4f32 V128:$Rd, V128:$Rn, V128:$Rm)>;
2701 def : Pat<(v2f64 (fma (fneg V128:$Rn), V128:$Rm, V128:$Rd)),
2702 (FMLSv2f64 V128:$Rd, V128:$Rn, V128:$Rm)>;
2704 defm FMULX : SIMDThreeSameVectorFP<0,0,0b11011,"fmulx", int_aarch64_neon_fmulx>;
2705 defm FMUL : SIMDThreeSameVectorFP<1,0,0b11011,"fmul", fmul>;
2706 defm FRECPS : SIMDThreeSameVectorFP<0,0,0b11111,"frecps", int_aarch64_neon_frecps>;
2707 defm FRSQRTS : SIMDThreeSameVectorFP<0,1,0b11111,"frsqrts", int_aarch64_neon_frsqrts>;
2708 defm FSUB : SIMDThreeSameVectorFP<0,1,0b11010,"fsub", fsub>;
2709 defm MLA : SIMDThreeSameVectorBHSTied<0, 0b10010, "mla",
2710 TriOpFrag<(add node:$LHS, (mul node:$MHS, node:$RHS))> >;
2711 defm MLS : SIMDThreeSameVectorBHSTied<1, 0b10010, "mls",
2712 TriOpFrag<(sub node:$LHS, (mul node:$MHS, node:$RHS))> >;
2713 defm MUL : SIMDThreeSameVectorBHS<0, 0b10011, "mul", mul>;
2714 defm PMUL : SIMDThreeSameVectorB<1, 0b10011, "pmul", int_aarch64_neon_pmul>;
2715 defm SABA : SIMDThreeSameVectorBHSTied<0, 0b01111, "saba",
2716 TriOpFrag<(add node:$LHS, (int_aarch64_neon_sabd node:$MHS, node:$RHS))> >;
2717 defm SABD : SIMDThreeSameVectorBHS<0,0b01110,"sabd", int_aarch64_neon_sabd>;
2718 defm SHADD : SIMDThreeSameVectorBHS<0,0b00000,"shadd", int_aarch64_neon_shadd>;
2719 defm SHSUB : SIMDThreeSameVectorBHS<0,0b00100,"shsub", int_aarch64_neon_shsub>;
2720 defm SMAXP : SIMDThreeSameVectorBHS<0,0b10100,"smaxp", int_aarch64_neon_smaxp>;
2721 defm SMAX : SIMDThreeSameVectorBHS<0,0b01100,"smax", int_aarch64_neon_smax>;
2722 defm SMINP : SIMDThreeSameVectorBHS<0,0b10101,"sminp", int_aarch64_neon_sminp>;
2723 defm SMIN : SIMDThreeSameVectorBHS<0,0b01101,"smin", int_aarch64_neon_smin>;
2724 defm SQADD : SIMDThreeSameVector<0,0b00001,"sqadd", int_aarch64_neon_sqadd>;
2725 defm SQDMULH : SIMDThreeSameVectorHS<0,0b10110,"sqdmulh",int_aarch64_neon_sqdmulh>;
2726 defm SQRDMULH : SIMDThreeSameVectorHS<1,0b10110,"sqrdmulh",int_aarch64_neon_sqrdmulh>;
2727 defm SQRSHL : SIMDThreeSameVector<0,0b01011,"sqrshl", int_aarch64_neon_sqrshl>;
2728 defm SQSHL : SIMDThreeSameVector<0,0b01001,"sqshl", int_aarch64_neon_sqshl>;
2729 defm SQSUB : SIMDThreeSameVector<0,0b00101,"sqsub", int_aarch64_neon_sqsub>;
2730 defm SRHADD : SIMDThreeSameVectorBHS<0,0b00010,"srhadd",int_aarch64_neon_srhadd>;
2731 defm SRSHL : SIMDThreeSameVector<0,0b01010,"srshl", int_aarch64_neon_srshl>;
2732 defm SSHL : SIMDThreeSameVector<0,0b01000,"sshl", int_aarch64_neon_sshl>;
2733 defm SUB : SIMDThreeSameVector<1,0b10000,"sub", sub>;
2734 defm UABA : SIMDThreeSameVectorBHSTied<1, 0b01111, "uaba",
2735 TriOpFrag<(add node:$LHS, (int_aarch64_neon_uabd node:$MHS, node:$RHS))> >;
2736 defm UABD : SIMDThreeSameVectorBHS<1,0b01110,"uabd", int_aarch64_neon_uabd>;
2737 defm UHADD : SIMDThreeSameVectorBHS<1,0b00000,"uhadd", int_aarch64_neon_uhadd>;
2738 defm UHSUB : SIMDThreeSameVectorBHS<1,0b00100,"uhsub", int_aarch64_neon_uhsub>;
2739 defm UMAXP : SIMDThreeSameVectorBHS<1,0b10100,"umaxp", int_aarch64_neon_umaxp>;
2740 defm UMAX : SIMDThreeSameVectorBHS<1,0b01100,"umax", int_aarch64_neon_umax>;
2741 defm UMINP : SIMDThreeSameVectorBHS<1,0b10101,"uminp", int_aarch64_neon_uminp>;
2742 defm UMIN : SIMDThreeSameVectorBHS<1,0b01101,"umin", int_aarch64_neon_umin>;
2743 defm UQADD : SIMDThreeSameVector<1,0b00001,"uqadd", int_aarch64_neon_uqadd>;
2744 defm UQRSHL : SIMDThreeSameVector<1,0b01011,"uqrshl", int_aarch64_neon_uqrshl>;
2745 defm UQSHL : SIMDThreeSameVector<1,0b01001,"uqshl", int_aarch64_neon_uqshl>;
2746 defm UQSUB : SIMDThreeSameVector<1,0b00101,"uqsub", int_aarch64_neon_uqsub>;
2747 defm URHADD : SIMDThreeSameVectorBHS<1,0b00010,"urhadd", int_aarch64_neon_urhadd>;
2748 defm URSHL : SIMDThreeSameVector<1,0b01010,"urshl", int_aarch64_neon_urshl>;
2749 defm USHL : SIMDThreeSameVector<1,0b01000,"ushl", int_aarch64_neon_ushl>;
2751 defm AND : SIMDLogicalThreeVector<0, 0b00, "and", and>;
2752 defm BIC : SIMDLogicalThreeVector<0, 0b01, "bic",
2753 BinOpFrag<(and node:$LHS, (vnot node:$RHS))> >;
2754 defm BIF : SIMDLogicalThreeVector<1, 0b11, "bif">;
2755 defm BIT : SIMDLogicalThreeVectorTied<1, 0b10, "bit", AArch64bit>;
2756 defm BSL : SIMDLogicalThreeVectorTied<1, 0b01, "bsl",
2757 TriOpFrag<(or (and node:$LHS, node:$MHS), (and (vnot node:$LHS), node:$RHS))>>;
2758 defm EOR : SIMDLogicalThreeVector<1, 0b00, "eor", xor>;
2759 defm ORN : SIMDLogicalThreeVector<0, 0b11, "orn",
2760 BinOpFrag<(or node:$LHS, (vnot node:$RHS))> >;
2761 defm ORR : SIMDLogicalThreeVector<0, 0b10, "orr", or>;
2763 // SABD Vd.<T>, Vn.<T>, Vm.<T> Subtracts the elements of Vm from the corresponding
2764 // elements of Vn, and places the absolute values of the results in the elements of Vd.
2765 def : Pat<(xor (v8i8 (AArch64vashr (v8i8(sub V64:$Rn, V64:$Rm)), (i32 7))),
2766 (v8i8 (add (v8i8(sub V64:$Rn, V64:$Rm)),
2767 (AArch64vashr (v8i8(sub V64:$Rn, V64:$Rm)), (i32 7))))),
2768 (SABDv8i8 V64:$Rn, V64:$Rm)>;
2769 def : Pat<(xor (v4i16 (AArch64vashr (v4i16(sub V64:$Rn, V64:$Rm)), (i32 15))),
2770 (v4i16 (add (v4i16(sub V64:$Rn, V64:$Rm)),
2771 (AArch64vashr (v4i16(sub V64:$Rn, V64:$Rm)), (i32 15))))),
2772 (SABDv4i16 V64:$Rn, V64:$Rm)>;
2773 def : Pat<(xor (v2i32 (AArch64vashr (v2i32(sub V64:$Rn, V64:$Rm)), (i32 31))),
2774 (v2i32 (add (v2i32(sub V64:$Rn, V64:$Rm)),
2775 (AArch64vashr (v2i32(sub V64:$Rn, V64:$Rm)), (i32 31))))),
2776 (SABDv2i32 V64:$Rn, V64:$Rm)>;
2777 def : Pat<(xor (v16i8 (AArch64vashr (v16i8(sub V128:$Rn, V128:$Rm)), (i32 7))),
2778 (v16i8 (add (v16i8(sub V128:$Rn, V128:$Rm)),
2779 (AArch64vashr (v16i8(sub V128:$Rn, V128:$Rm)), (i32 7))))),
2780 (SABDv16i8 V128:$Rn, V128:$Rm)>;
2781 def : Pat<(xor (v8i16 (AArch64vashr (v8i16(sub V128:$Rn, V128:$Rm)), (i32 15))),
2782 (v8i16 (add (v8i16(sub V128:$Rn, V128:$Rm)),
2783 (AArch64vashr (v8i16(sub V128:$Rn, V128:$Rm)), (i32 15))))),
2784 (SABDv8i16 V128:$Rn, V128:$Rm)>;
2785 def : Pat<(xor (v4i32 (AArch64vashr (v4i32(sub V128:$Rn, V128:$Rm)), (i32 31))),
2786 (v4i32 (add (v4i32(sub V128:$Rn, V128:$Rm)),
2787 (AArch64vashr (v4i32(sub V128:$Rn, V128:$Rm)), (i32 31))))),
2788 (SABDv4i32 V128:$Rn, V128:$Rm)>;
2790 def : Pat<(v2f32 (fabs (fsub V64:$Rn, V64:$Rm))),
2791 (FABDv2f32 V64:$Rn, V64:$Rm)>;
2792 def : Pat<(v4f32 (fabs (fsub V128:$Rn, V128:$Rm))),
2793 (FABDv4f32 V128:$Rn, V128:$Rm)>;
2794 def : Pat<(v2f64 (fabs (fsub V128:$Rn, V128:$Rm))),
2795 (FABDv2f64 V128:$Rn, V128:$Rm)>;
2797 def : Pat<(AArch64bsl (v8i8 V64:$Rd), V64:$Rn, V64:$Rm),
2798 (BSLv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
2799 def : Pat<(AArch64bsl (v4i16 V64:$Rd), V64:$Rn, V64:$Rm),
2800 (BSLv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
2801 def : Pat<(AArch64bsl (v2i32 V64:$Rd), V64:$Rn, V64:$Rm),
2802 (BSLv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
2803 def : Pat<(AArch64bsl (v1i64 V64:$Rd), V64:$Rn, V64:$Rm),
2804 (BSLv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
2806 def : Pat<(AArch64bsl (v16i8 V128:$Rd), V128:$Rn, V128:$Rm),
2807 (BSLv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
2808 def : Pat<(AArch64bsl (v8i16 V128:$Rd), V128:$Rn, V128:$Rm),
2809 (BSLv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
2810 def : Pat<(AArch64bsl (v4i32 V128:$Rd), V128:$Rn, V128:$Rm),
2811 (BSLv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
2812 def : Pat<(AArch64bsl (v2i64 V128:$Rd), V128:$Rn, V128:$Rm),
2813 (BSLv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
2815 def : InstAlias<"mov{\t$dst.16b, $src.16b|.16b\t$dst, $src}",
2816 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 1>;
2817 def : InstAlias<"mov{\t$dst.8h, $src.8h|.8h\t$dst, $src}",
2818 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2819 def : InstAlias<"mov{\t$dst.4s, $src.4s|.4s\t$dst, $src}",
2820 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2821 def : InstAlias<"mov{\t$dst.2d, $src.2d|.2d\t$dst, $src}",
2822 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2824 def : InstAlias<"mov{\t$dst.8b, $src.8b|.8b\t$dst, $src}",
2825 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 1>;
2826 def : InstAlias<"mov{\t$dst.4h, $src.4h|.4h\t$dst, $src}",
2827 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2828 def : InstAlias<"mov{\t$dst.2s, $src.2s|.2s\t$dst, $src}",
2829 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2830 def : InstAlias<"mov{\t$dst.1d, $src.1d|.1d\t$dst, $src}",
2831 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2833 def : InstAlias<"{cmls\t$dst.8b, $src1.8b, $src2.8b" #
2834 "|cmls.8b\t$dst, $src1, $src2}",
2835 (CMHSv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2836 def : InstAlias<"{cmls\t$dst.16b, $src1.16b, $src2.16b" #
2837 "|cmls.16b\t$dst, $src1, $src2}",
2838 (CMHSv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2839 def : InstAlias<"{cmls\t$dst.4h, $src1.4h, $src2.4h" #
2840 "|cmls.4h\t$dst, $src1, $src2}",
2841 (CMHSv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2842 def : InstAlias<"{cmls\t$dst.8h, $src1.8h, $src2.8h" #
2843 "|cmls.8h\t$dst, $src1, $src2}",
2844 (CMHSv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2845 def : InstAlias<"{cmls\t$dst.2s, $src1.2s, $src2.2s" #
2846 "|cmls.2s\t$dst, $src1, $src2}",
2847 (CMHSv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2848 def : InstAlias<"{cmls\t$dst.4s, $src1.4s, $src2.4s" #
2849 "|cmls.4s\t$dst, $src1, $src2}",
2850 (CMHSv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2851 def : InstAlias<"{cmls\t$dst.2d, $src1.2d, $src2.2d" #
2852 "|cmls.2d\t$dst, $src1, $src2}",
2853 (CMHSv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2855 def : InstAlias<"{cmlo\t$dst.8b, $src1.8b, $src2.8b" #
2856 "|cmlo.8b\t$dst, $src1, $src2}",
2857 (CMHIv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2858 def : InstAlias<"{cmlo\t$dst.16b, $src1.16b, $src2.16b" #
2859 "|cmlo.16b\t$dst, $src1, $src2}",
2860 (CMHIv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2861 def : InstAlias<"{cmlo\t$dst.4h, $src1.4h, $src2.4h" #
2862 "|cmlo.4h\t$dst, $src1, $src2}",
2863 (CMHIv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2864 def : InstAlias<"{cmlo\t$dst.8h, $src1.8h, $src2.8h" #
2865 "|cmlo.8h\t$dst, $src1, $src2}",
2866 (CMHIv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2867 def : InstAlias<"{cmlo\t$dst.2s, $src1.2s, $src2.2s" #
2868 "|cmlo.2s\t$dst, $src1, $src2}",
2869 (CMHIv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2870 def : InstAlias<"{cmlo\t$dst.4s, $src1.4s, $src2.4s" #
2871 "|cmlo.4s\t$dst, $src1, $src2}",
2872 (CMHIv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2873 def : InstAlias<"{cmlo\t$dst.2d, $src1.2d, $src2.2d" #
2874 "|cmlo.2d\t$dst, $src1, $src2}",
2875 (CMHIv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2877 def : InstAlias<"{cmle\t$dst.8b, $src1.8b, $src2.8b" #
2878 "|cmle.8b\t$dst, $src1, $src2}",
2879 (CMGEv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2880 def : InstAlias<"{cmle\t$dst.16b, $src1.16b, $src2.16b" #
2881 "|cmle.16b\t$dst, $src1, $src2}",
2882 (CMGEv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2883 def : InstAlias<"{cmle\t$dst.4h, $src1.4h, $src2.4h" #
2884 "|cmle.4h\t$dst, $src1, $src2}",
2885 (CMGEv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2886 def : InstAlias<"{cmle\t$dst.8h, $src1.8h, $src2.8h" #
2887 "|cmle.8h\t$dst, $src1, $src2}",
2888 (CMGEv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2889 def : InstAlias<"{cmle\t$dst.2s, $src1.2s, $src2.2s" #
2890 "|cmle.2s\t$dst, $src1, $src2}",
2891 (CMGEv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2892 def : InstAlias<"{cmle\t$dst.4s, $src1.4s, $src2.4s" #
2893 "|cmle.4s\t$dst, $src1, $src2}",
2894 (CMGEv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2895 def : InstAlias<"{cmle\t$dst.2d, $src1.2d, $src2.2d" #
2896 "|cmle.2d\t$dst, $src1, $src2}",
2897 (CMGEv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2899 def : InstAlias<"{cmlt\t$dst.8b, $src1.8b, $src2.8b" #
2900 "|cmlt.8b\t$dst, $src1, $src2}",
2901 (CMGTv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2902 def : InstAlias<"{cmlt\t$dst.16b, $src1.16b, $src2.16b" #
2903 "|cmlt.16b\t$dst, $src1, $src2}",
2904 (CMGTv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2905 def : InstAlias<"{cmlt\t$dst.4h, $src1.4h, $src2.4h" #
2906 "|cmlt.4h\t$dst, $src1, $src2}",
2907 (CMGTv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2908 def : InstAlias<"{cmlt\t$dst.8h, $src1.8h, $src2.8h" #
2909 "|cmlt.8h\t$dst, $src1, $src2}",
2910 (CMGTv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2911 def : InstAlias<"{cmlt\t$dst.2s, $src1.2s, $src2.2s" #
2912 "|cmlt.2s\t$dst, $src1, $src2}",
2913 (CMGTv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2914 def : InstAlias<"{cmlt\t$dst.4s, $src1.4s, $src2.4s" #
2915 "|cmlt.4s\t$dst, $src1, $src2}",
2916 (CMGTv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2917 def : InstAlias<"{cmlt\t$dst.2d, $src1.2d, $src2.2d" #
2918 "|cmlt.2d\t$dst, $src1, $src2}",
2919 (CMGTv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2921 def : InstAlias<"{fcmle\t$dst.2s, $src1.2s, $src2.2s" #
2922 "|fcmle.2s\t$dst, $src1, $src2}",
2923 (FCMGEv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
2924 def : InstAlias<"{fcmle\t$dst.4s, $src1.4s, $src2.4s" #
2925 "|fcmle.4s\t$dst, $src1, $src2}",
2926 (FCMGEv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
2927 def : InstAlias<"{fcmle\t$dst.2d, $src1.2d, $src2.2d" #
2928 "|fcmle.2d\t$dst, $src1, $src2}",
2929 (FCMGEv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
2931 def : InstAlias<"{fcmlt\t$dst.2s, $src1.2s, $src2.2s" #
2932 "|fcmlt.2s\t$dst, $src1, $src2}",
2933 (FCMGTv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
2934 def : InstAlias<"{fcmlt\t$dst.4s, $src1.4s, $src2.4s" #
2935 "|fcmlt.4s\t$dst, $src1, $src2}",
2936 (FCMGTv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
2937 def : InstAlias<"{fcmlt\t$dst.2d, $src1.2d, $src2.2d" #
2938 "|fcmlt.2d\t$dst, $src1, $src2}",
2939 (FCMGTv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
2941 def : InstAlias<"{facle\t$dst.2s, $src1.2s, $src2.2s" #
2942 "|facle.2s\t$dst, $src1, $src2}",
2943 (FACGEv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
2944 def : InstAlias<"{facle\t$dst.4s, $src1.4s, $src2.4s" #
2945 "|facle.4s\t$dst, $src1, $src2}",
2946 (FACGEv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
2947 def : InstAlias<"{facle\t$dst.2d, $src1.2d, $src2.2d" #
2948 "|facle.2d\t$dst, $src1, $src2}",
2949 (FACGEv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
2951 def : InstAlias<"{faclt\t$dst.2s, $src1.2s, $src2.2s" #
2952 "|faclt.2s\t$dst, $src1, $src2}",
2953 (FACGTv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
2954 def : InstAlias<"{faclt\t$dst.4s, $src1.4s, $src2.4s" #
2955 "|faclt.4s\t$dst, $src1, $src2}",
2956 (FACGTv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
2957 def : InstAlias<"{faclt\t$dst.2d, $src1.2d, $src2.2d" #
2958 "|faclt.2d\t$dst, $src1, $src2}",
2959 (FACGTv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
2961 //===----------------------------------------------------------------------===//
2962 // Advanced SIMD three scalar instructions.
2963 //===----------------------------------------------------------------------===//
2965 defm ADD : SIMDThreeScalarD<0, 0b10000, "add", add>;
2966 defm CMEQ : SIMDThreeScalarD<1, 0b10001, "cmeq", AArch64cmeq>;
2967 defm CMGE : SIMDThreeScalarD<0, 0b00111, "cmge", AArch64cmge>;
2968 defm CMGT : SIMDThreeScalarD<0, 0b00110, "cmgt", AArch64cmgt>;
2969 defm CMHI : SIMDThreeScalarD<1, 0b00110, "cmhi", AArch64cmhi>;
2970 defm CMHS : SIMDThreeScalarD<1, 0b00111, "cmhs", AArch64cmhs>;
2971 defm CMTST : SIMDThreeScalarD<0, 0b10001, "cmtst", AArch64cmtst>;
2972 defm FABD : SIMDThreeScalarSD<1, 1, 0b11010, "fabd", int_aarch64_sisd_fabd>;
2973 def : Pat<(v1f64 (int_aarch64_neon_fabd (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2974 (FABD64 FPR64:$Rn, FPR64:$Rm)>;
2975 defm FACGE : SIMDThreeScalarFPCmp<1, 0, 0b11101, "facge",
2976 int_aarch64_neon_facge>;
2977 defm FACGT : SIMDThreeScalarFPCmp<1, 1, 0b11101, "facgt",
2978 int_aarch64_neon_facgt>;
2979 defm FCMEQ : SIMDThreeScalarFPCmp<0, 0, 0b11100, "fcmeq", AArch64fcmeq>;
2980 defm FCMGE : SIMDThreeScalarFPCmp<1, 0, 0b11100, "fcmge", AArch64fcmge>;
2981 defm FCMGT : SIMDThreeScalarFPCmp<1, 1, 0b11100, "fcmgt", AArch64fcmgt>;
2982 defm FMULX : SIMDThreeScalarSD<0, 0, 0b11011, "fmulx", int_aarch64_neon_fmulx>;
2983 defm FRECPS : SIMDThreeScalarSD<0, 0, 0b11111, "frecps", int_aarch64_neon_frecps>;
2984 defm FRSQRTS : SIMDThreeScalarSD<0, 1, 0b11111, "frsqrts", int_aarch64_neon_frsqrts>;
2985 defm SQADD : SIMDThreeScalarBHSD<0, 0b00001, "sqadd", int_aarch64_neon_sqadd>;
2986 defm SQDMULH : SIMDThreeScalarHS< 0, 0b10110, "sqdmulh", int_aarch64_neon_sqdmulh>;
2987 defm SQRDMULH : SIMDThreeScalarHS< 1, 0b10110, "sqrdmulh", int_aarch64_neon_sqrdmulh>;
2988 defm SQRSHL : SIMDThreeScalarBHSD<0, 0b01011, "sqrshl",int_aarch64_neon_sqrshl>;
2989 defm SQSHL : SIMDThreeScalarBHSD<0, 0b01001, "sqshl", int_aarch64_neon_sqshl>;
2990 defm SQSUB : SIMDThreeScalarBHSD<0, 0b00101, "sqsub", int_aarch64_neon_sqsub>;
2991 defm SRSHL : SIMDThreeScalarD< 0, 0b01010, "srshl", int_aarch64_neon_srshl>;
2992 defm SSHL : SIMDThreeScalarD< 0, 0b01000, "sshl", int_aarch64_neon_sshl>;
2993 defm SUB : SIMDThreeScalarD< 1, 0b10000, "sub", sub>;
2994 defm UQADD : SIMDThreeScalarBHSD<1, 0b00001, "uqadd", int_aarch64_neon_uqadd>;
2995 defm UQRSHL : SIMDThreeScalarBHSD<1, 0b01011, "uqrshl",int_aarch64_neon_uqrshl>;
2996 defm UQSHL : SIMDThreeScalarBHSD<1, 0b01001, "uqshl", int_aarch64_neon_uqshl>;
2997 defm UQSUB : SIMDThreeScalarBHSD<1, 0b00101, "uqsub", int_aarch64_neon_uqsub>;
2998 defm URSHL : SIMDThreeScalarD< 1, 0b01010, "urshl", int_aarch64_neon_urshl>;
2999 defm USHL : SIMDThreeScalarD< 1, 0b01000, "ushl", int_aarch64_neon_ushl>;
3001 def : InstAlias<"cmls $dst, $src1, $src2",
3002 (CMHSv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
3003 def : InstAlias<"cmle $dst, $src1, $src2",
3004 (CMGEv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
3005 def : InstAlias<"cmlo $dst, $src1, $src2",
3006 (CMHIv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
3007 def : InstAlias<"cmlt $dst, $src1, $src2",
3008 (CMGTv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
3009 def : InstAlias<"fcmle $dst, $src1, $src2",
3010 (FCMGE32 FPR32:$dst, FPR32:$src2, FPR32:$src1), 0>;
3011 def : InstAlias<"fcmle $dst, $src1, $src2",
3012 (FCMGE64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
3013 def : InstAlias<"fcmlt $dst, $src1, $src2",
3014 (FCMGT32 FPR32:$dst, FPR32:$src2, FPR32:$src1), 0>;
3015 def : InstAlias<"fcmlt $dst, $src1, $src2",
3016 (FCMGT64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
3017 def : InstAlias<"facle $dst, $src1, $src2",
3018 (FACGE32 FPR32:$dst, FPR32:$src2, FPR32:$src1), 0>;
3019 def : InstAlias<"facle $dst, $src1, $src2",
3020 (FACGE64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
3021 def : InstAlias<"faclt $dst, $src1, $src2",
3022 (FACGT32 FPR32:$dst, FPR32:$src2, FPR32:$src1), 0>;
3023 def : InstAlias<"faclt $dst, $src1, $src2",
3024 (FACGT64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
3026 //===----------------------------------------------------------------------===//
3027 // Advanced SIMD three scalar instructions (mixed operands).
3028 //===----------------------------------------------------------------------===//
3029 defm SQDMULL : SIMDThreeScalarMixedHS<0, 0b11010, "sqdmull",
3030 int_aarch64_neon_sqdmulls_scalar>;
3031 defm SQDMLAL : SIMDThreeScalarMixedTiedHS<0, 0b10010, "sqdmlal">;
3032 defm SQDMLSL : SIMDThreeScalarMixedTiedHS<0, 0b10110, "sqdmlsl">;
3034 def : Pat<(i64 (int_aarch64_neon_sqadd (i64 FPR64:$Rd),
3035 (i64 (int_aarch64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
3036 (i32 FPR32:$Rm))))),
3037 (SQDMLALi32 FPR64:$Rd, FPR32:$Rn, FPR32:$Rm)>;
3038 def : Pat<(i64 (int_aarch64_neon_sqsub (i64 FPR64:$Rd),
3039 (i64 (int_aarch64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
3040 (i32 FPR32:$Rm))))),
3041 (SQDMLSLi32 FPR64:$Rd, FPR32:$Rn, FPR32:$Rm)>;
3043 //===----------------------------------------------------------------------===//
3044 // Advanced SIMD two scalar instructions.
3045 //===----------------------------------------------------------------------===//
3047 defm ABS : SIMDTwoScalarD< 0, 0b01011, "abs", int_aarch64_neon_abs>;
3048 defm CMEQ : SIMDCmpTwoScalarD< 0, 0b01001, "cmeq", AArch64cmeqz>;
3049 defm CMGE : SIMDCmpTwoScalarD< 1, 0b01000, "cmge", AArch64cmgez>;
3050 defm CMGT : SIMDCmpTwoScalarD< 0, 0b01000, "cmgt", AArch64cmgtz>;
3051 defm CMLE : SIMDCmpTwoScalarD< 1, 0b01001, "cmle", AArch64cmlez>;
3052 defm CMLT : SIMDCmpTwoScalarD< 0, 0b01010, "cmlt", AArch64cmltz>;
3053 defm FCMEQ : SIMDCmpTwoScalarSD<0, 1, 0b01101, "fcmeq", AArch64fcmeqz>;
3054 defm FCMGE : SIMDCmpTwoScalarSD<1, 1, 0b01100, "fcmge", AArch64fcmgez>;
3055 defm FCMGT : SIMDCmpTwoScalarSD<0, 1, 0b01100, "fcmgt", AArch64fcmgtz>;
3056 defm FCMLE : SIMDCmpTwoScalarSD<1, 1, 0b01101, "fcmle", AArch64fcmlez>;
3057 defm FCMLT : SIMDCmpTwoScalarSD<0, 1, 0b01110, "fcmlt", AArch64fcmltz>;
3058 defm FCVTAS : SIMDTwoScalarSD< 0, 0, 0b11100, "fcvtas">;
3059 defm FCVTAU : SIMDTwoScalarSD< 1, 0, 0b11100, "fcvtau">;
3060 defm FCVTMS : SIMDTwoScalarSD< 0, 0, 0b11011, "fcvtms">;
3061 defm FCVTMU : SIMDTwoScalarSD< 1, 0, 0b11011, "fcvtmu">;
3062 defm FCVTNS : SIMDTwoScalarSD< 0, 0, 0b11010, "fcvtns">;
3063 defm FCVTNU : SIMDTwoScalarSD< 1, 0, 0b11010, "fcvtnu">;
3064 defm FCVTPS : SIMDTwoScalarSD< 0, 1, 0b11010, "fcvtps">;
3065 defm FCVTPU : SIMDTwoScalarSD< 1, 1, 0b11010, "fcvtpu">;
3066 def FCVTXNv1i64 : SIMDInexactCvtTwoScalar<0b10110, "fcvtxn">;
3067 defm FCVTZS : SIMDTwoScalarSD< 0, 1, 0b11011, "fcvtzs">;
3068 defm FCVTZU : SIMDTwoScalarSD< 1, 1, 0b11011, "fcvtzu">;
3069 defm FRECPE : SIMDTwoScalarSD< 0, 1, 0b11101, "frecpe">;
3070 defm FRECPX : SIMDTwoScalarSD< 0, 1, 0b11111, "frecpx">;
3071 defm FRSQRTE : SIMDTwoScalarSD< 1, 1, 0b11101, "frsqrte">;
3072 defm NEG : SIMDTwoScalarD< 1, 0b01011, "neg",
3073 UnOpFrag<(sub immAllZerosV, node:$LHS)> >;
3074 defm SCVTF : SIMDTwoScalarCVTSD< 0, 0, 0b11101, "scvtf", AArch64sitof>;
3075 defm SQABS : SIMDTwoScalarBHSD< 0, 0b00111, "sqabs", int_aarch64_neon_sqabs>;
3076 defm SQNEG : SIMDTwoScalarBHSD< 1, 0b00111, "sqneg", int_aarch64_neon_sqneg>;
3077 defm SQXTN : SIMDTwoScalarMixedBHS< 0, 0b10100, "sqxtn", int_aarch64_neon_scalar_sqxtn>;
3078 defm SQXTUN : SIMDTwoScalarMixedBHS< 1, 0b10010, "sqxtun", int_aarch64_neon_scalar_sqxtun>;
3079 defm SUQADD : SIMDTwoScalarBHSDTied< 0, 0b00011, "suqadd",
3080 int_aarch64_neon_suqadd>;
3081 defm UCVTF : SIMDTwoScalarCVTSD< 1, 0, 0b11101, "ucvtf", AArch64uitof>;
3082 defm UQXTN : SIMDTwoScalarMixedBHS<1, 0b10100, "uqxtn", int_aarch64_neon_scalar_uqxtn>;
3083 defm USQADD : SIMDTwoScalarBHSDTied< 1, 0b00011, "usqadd",
3084 int_aarch64_neon_usqadd>;
3086 def : Pat<(f32 (fabs (fsub FPR32:$Rn, FPR32:$Rm))),
3087 (FABD32 FPR32:$Rn, FPR32:$Rm)>;
3088 def : Pat<(f64 (fabs (fsub FPR64:$Rn, FPR64:$Rm))),
3089 (FABD64 FPR64:$Rn, FPR64:$Rm)>;
3091 def : Pat<(AArch64neg (v1i64 V64:$Rn)), (NEGv1i64 V64:$Rn)>;
3093 def : Pat<(v1i64 (int_aarch64_neon_fcvtas (v1f64 FPR64:$Rn))),
3094 (FCVTASv1i64 FPR64:$Rn)>;
3095 def : Pat<(v1i64 (int_aarch64_neon_fcvtau (v1f64 FPR64:$Rn))),
3096 (FCVTAUv1i64 FPR64:$Rn)>;
3097 def : Pat<(v1i64 (int_aarch64_neon_fcvtms (v1f64 FPR64:$Rn))),
3098 (FCVTMSv1i64 FPR64:$Rn)>;
3099 def : Pat<(v1i64 (int_aarch64_neon_fcvtmu (v1f64 FPR64:$Rn))),
3100 (FCVTMUv1i64 FPR64:$Rn)>;
3101 def : Pat<(v1i64 (int_aarch64_neon_fcvtns (v1f64 FPR64:$Rn))),
3102 (FCVTNSv1i64 FPR64:$Rn)>;
3103 def : Pat<(v1i64 (int_aarch64_neon_fcvtnu (v1f64 FPR64:$Rn))),
3104 (FCVTNUv1i64 FPR64:$Rn)>;
3105 def : Pat<(v1i64 (int_aarch64_neon_fcvtps (v1f64 FPR64:$Rn))),
3106 (FCVTPSv1i64 FPR64:$Rn)>;
3107 def : Pat<(v1i64 (int_aarch64_neon_fcvtpu (v1f64 FPR64:$Rn))),
3108 (FCVTPUv1i64 FPR64:$Rn)>;
3110 def : Pat<(f32 (int_aarch64_neon_frecpe (f32 FPR32:$Rn))),
3111 (FRECPEv1i32 FPR32:$Rn)>;
3112 def : Pat<(f64 (int_aarch64_neon_frecpe (f64 FPR64:$Rn))),
3113 (FRECPEv1i64 FPR64:$Rn)>;
3114 def : Pat<(v1f64 (int_aarch64_neon_frecpe (v1f64 FPR64:$Rn))),
3115 (FRECPEv1i64 FPR64:$Rn)>;
3117 def : Pat<(f32 (int_aarch64_neon_frecpx (f32 FPR32:$Rn))),
3118 (FRECPXv1i32 FPR32:$Rn)>;
3119 def : Pat<(f64 (int_aarch64_neon_frecpx (f64 FPR64:$Rn))),
3120 (FRECPXv1i64 FPR64:$Rn)>;
3122 def : Pat<(f32 (int_aarch64_neon_frsqrte (f32 FPR32:$Rn))),
3123 (FRSQRTEv1i32 FPR32:$Rn)>;
3124 def : Pat<(f64 (int_aarch64_neon_frsqrte (f64 FPR64:$Rn))),
3125 (FRSQRTEv1i64 FPR64:$Rn)>;
3126 def : Pat<(v1f64 (int_aarch64_neon_frsqrte (v1f64 FPR64:$Rn))),
3127 (FRSQRTEv1i64 FPR64:$Rn)>;
3129 // If an integer is about to be converted to a floating point value,
3130 // just load it on the floating point unit.
3131 // Here are the patterns for 8 and 16-bits to float.
3133 multiclass UIntToFPROLoadPat<ValueType DstTy, ValueType SrcTy,
3134 SDPatternOperator loadop, Instruction UCVTF,
3135 ROAddrMode ro, Instruction LDRW, Instruction LDRX,
3137 def : Pat<(DstTy (uint_to_fp (SrcTy
3138 (loadop (ro.Wpat GPR64sp:$Rn, GPR32:$Rm,
3139 ro.Wext:$extend))))),
3140 (UCVTF (INSERT_SUBREG (DstTy (IMPLICIT_DEF)),
3141 (LDRW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend),
3144 def : Pat<(DstTy (uint_to_fp (SrcTy
3145 (loadop (ro.Xpat GPR64sp:$Rn, GPR64:$Rm,
3146 ro.Wext:$extend))))),
3147 (UCVTF (INSERT_SUBREG (DstTy (IMPLICIT_DEF)),
3148 (LDRX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend),
3152 defm : UIntToFPROLoadPat<f32, i32, zextloadi8,
3153 UCVTFv1i32, ro8, LDRBroW, LDRBroX, bsub>;
3154 def : Pat <(f32 (uint_to_fp (i32
3155 (zextloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))))),
3156 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
3157 (LDRBui GPR64sp:$Rn, uimm12s1:$offset), bsub))>;
3158 def : Pat <(f32 (uint_to_fp (i32
3159 (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))))),
3160 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
3161 (LDURBi GPR64sp:$Rn, simm9:$offset), bsub))>;
3162 // 16-bits -> float.
3163 defm : UIntToFPROLoadPat<f32, i32, zextloadi16,
3164 UCVTFv1i32, ro16, LDRHroW, LDRHroX, hsub>;
3165 def : Pat <(f32 (uint_to_fp (i32
3166 (zextloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))))),
3167 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
3168 (LDRHui GPR64sp:$Rn, uimm12s2:$offset), hsub))>;
3169 def : Pat <(f32 (uint_to_fp (i32
3170 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))))),
3171 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
3172 (LDURHi GPR64sp:$Rn, simm9:$offset), hsub))>;
3173 // 32-bits are handled in target specific dag combine:
3174 // performIntToFpCombine.
3175 // 64-bits integer to 32-bits floating point, not possible with
3176 // UCVTF on floating point registers (both source and destination
3177 // must have the same size).
3179 // Here are the patterns for 8, 16, 32, and 64-bits to double.
3180 // 8-bits -> double.
3181 defm : UIntToFPROLoadPat<f64, i32, zextloadi8,
3182 UCVTFv1i64, ro8, LDRBroW, LDRBroX, bsub>;
3183 def : Pat <(f64 (uint_to_fp (i32
3184 (zextloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))))),
3185 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3186 (LDRBui GPR64sp:$Rn, uimm12s1:$offset), bsub))>;
3187 def : Pat <(f64 (uint_to_fp (i32
3188 (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))))),
3189 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3190 (LDURBi GPR64sp:$Rn, simm9:$offset), bsub))>;
3191 // 16-bits -> double.
3192 defm : UIntToFPROLoadPat<f64, i32, zextloadi16,
3193 UCVTFv1i64, ro16, LDRHroW, LDRHroX, hsub>;
3194 def : Pat <(f64 (uint_to_fp (i32
3195 (zextloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))))),
3196 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3197 (LDRHui GPR64sp:$Rn, uimm12s2:$offset), hsub))>;
3198 def : Pat <(f64 (uint_to_fp (i32
3199 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))))),
3200 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3201 (LDURHi GPR64sp:$Rn, simm9:$offset), hsub))>;
3202 // 32-bits -> double.
3203 defm : UIntToFPROLoadPat<f64, i32, load,
3204 UCVTFv1i64, ro32, LDRSroW, LDRSroX, ssub>;
3205 def : Pat <(f64 (uint_to_fp (i32
3206 (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))))),
3207 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3208 (LDRSui GPR64sp:$Rn, uimm12s4:$offset), ssub))>;
3209 def : Pat <(f64 (uint_to_fp (i32
3210 (load (am_unscaled32 GPR64sp:$Rn, simm9:$offset))))),
3211 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3212 (LDURSi GPR64sp:$Rn, simm9:$offset), ssub))>;
3213 // 64-bits -> double are handled in target specific dag combine:
3214 // performIntToFpCombine.
3216 //===----------------------------------------------------------------------===//
3217 // Advanced SIMD three different-sized vector instructions.
3218 //===----------------------------------------------------------------------===//
3220 defm ADDHN : SIMDNarrowThreeVectorBHS<0,0b0100,"addhn", int_aarch64_neon_addhn>;
3221 defm SUBHN : SIMDNarrowThreeVectorBHS<0,0b0110,"subhn", int_aarch64_neon_subhn>;
3222 defm RADDHN : SIMDNarrowThreeVectorBHS<1,0b0100,"raddhn",int_aarch64_neon_raddhn>;
3223 defm RSUBHN : SIMDNarrowThreeVectorBHS<1,0b0110,"rsubhn",int_aarch64_neon_rsubhn>;
3224 defm PMULL : SIMDDifferentThreeVectorBD<0,0b1110,"pmull",int_aarch64_neon_pmull>;
3225 defm SABAL : SIMDLongThreeVectorTiedBHSabal<0,0b0101,"sabal",
3226 int_aarch64_neon_sabd>;
3227 defm SABDL : SIMDLongThreeVectorBHSabdl<0, 0b0111, "sabdl",
3228 int_aarch64_neon_sabd>;
3229 defm SADDL : SIMDLongThreeVectorBHS< 0, 0b0000, "saddl",
3230 BinOpFrag<(add (sext node:$LHS), (sext node:$RHS))>>;
3231 defm SADDW : SIMDWideThreeVectorBHS< 0, 0b0001, "saddw",
3232 BinOpFrag<(add node:$LHS, (sext node:$RHS))>>;
3233 defm SMLAL : SIMDLongThreeVectorTiedBHS<0, 0b1000, "smlal",
3234 TriOpFrag<(add node:$LHS, (int_aarch64_neon_smull node:$MHS, node:$RHS))>>;
3235 defm SMLSL : SIMDLongThreeVectorTiedBHS<0, 0b1010, "smlsl",
3236 TriOpFrag<(sub node:$LHS, (int_aarch64_neon_smull node:$MHS, node:$RHS))>>;
3237 defm SMULL : SIMDLongThreeVectorBHS<0, 0b1100, "smull", int_aarch64_neon_smull>;
3238 defm SQDMLAL : SIMDLongThreeVectorSQDMLXTiedHS<0, 0b1001, "sqdmlal",
3239 int_aarch64_neon_sqadd>;
3240 defm SQDMLSL : SIMDLongThreeVectorSQDMLXTiedHS<0, 0b1011, "sqdmlsl",
3241 int_aarch64_neon_sqsub>;
3242 defm SQDMULL : SIMDLongThreeVectorHS<0, 0b1101, "sqdmull",
3243 int_aarch64_neon_sqdmull>;
3244 defm SSUBL : SIMDLongThreeVectorBHS<0, 0b0010, "ssubl",
3245 BinOpFrag<(sub (sext node:$LHS), (sext node:$RHS))>>;
3246 defm SSUBW : SIMDWideThreeVectorBHS<0, 0b0011, "ssubw",
3247 BinOpFrag<(sub node:$LHS, (sext node:$RHS))>>;
3248 defm UABAL : SIMDLongThreeVectorTiedBHSabal<1, 0b0101, "uabal",
3249 int_aarch64_neon_uabd>;
3250 defm UABDL : SIMDLongThreeVectorBHSabdl<1, 0b0111, "uabdl",
3251 int_aarch64_neon_uabd>;
3252 defm UADDL : SIMDLongThreeVectorBHS<1, 0b0000, "uaddl",
3253 BinOpFrag<(add (zext node:$LHS), (zext node:$RHS))>>;
3254 defm UADDW : SIMDWideThreeVectorBHS<1, 0b0001, "uaddw",
3255 BinOpFrag<(add node:$LHS, (zext node:$RHS))>>;
3256 defm UMLAL : SIMDLongThreeVectorTiedBHS<1, 0b1000, "umlal",
3257 TriOpFrag<(add node:$LHS, (int_aarch64_neon_umull node:$MHS, node:$RHS))>>;
3258 defm UMLSL : SIMDLongThreeVectorTiedBHS<1, 0b1010, "umlsl",
3259 TriOpFrag<(sub node:$LHS, (int_aarch64_neon_umull node:$MHS, node:$RHS))>>;
3260 defm UMULL : SIMDLongThreeVectorBHS<1, 0b1100, "umull", int_aarch64_neon_umull>;
3261 defm USUBL : SIMDLongThreeVectorBHS<1, 0b0010, "usubl",
3262 BinOpFrag<(sub (zext node:$LHS), (zext node:$RHS))>>;
3263 defm USUBW : SIMDWideThreeVectorBHS< 1, 0b0011, "usubw",
3264 BinOpFrag<(sub node:$LHS, (zext node:$RHS))>>;
3266 // Additional patterns for SMULL and UMULL
3267 multiclass Neon_mul_widen_patterns<SDPatternOperator opnode,
3268 Instruction INST8B, Instruction INST4H, Instruction INST2S> {
3269 def : Pat<(v8i16 (opnode (v8i8 V64:$Rn), (v8i8 V64:$Rm))),
3270 (INST8B V64:$Rn, V64:$Rm)>;
3271 def : Pat<(v4i32 (opnode (v4i16 V64:$Rn), (v4i16 V64:$Rm))),
3272 (INST4H V64:$Rn, V64:$Rm)>;
3273 def : Pat<(v2i64 (opnode (v2i32 V64:$Rn), (v2i32 V64:$Rm))),
3274 (INST2S V64:$Rn, V64:$Rm)>;
3277 defm : Neon_mul_widen_patterns<AArch64smull, SMULLv8i8_v8i16,
3278 SMULLv4i16_v4i32, SMULLv2i32_v2i64>;
3279 defm : Neon_mul_widen_patterns<AArch64umull, UMULLv8i8_v8i16,
3280 UMULLv4i16_v4i32, UMULLv2i32_v2i64>;
3282 // Additional patterns for SMLAL/SMLSL and UMLAL/UMLSL
3283 multiclass Neon_mulacc_widen_patterns<SDPatternOperator opnode,
3284 Instruction INST8B, Instruction INST4H, Instruction INST2S> {
3285 def : Pat<(v8i16 (opnode (v8i16 V128:$Rd), (v8i8 V64:$Rn), (v8i8 V64:$Rm))),
3286 (INST8B V128:$Rd, V64:$Rn, V64:$Rm)>;
3287 def : Pat<(v4i32 (opnode (v4i32 V128:$Rd), (v4i16 V64:$Rn), (v4i16 V64:$Rm))),
3288 (INST4H V128:$Rd, V64:$Rn, V64:$Rm)>;
3289 def : Pat<(v2i64 (opnode (v2i64 V128:$Rd), (v2i32 V64:$Rn), (v2i32 V64:$Rm))),
3290 (INST2S V128:$Rd, V64:$Rn, V64:$Rm)>;
3293 defm : Neon_mulacc_widen_patterns<
3294 TriOpFrag<(add node:$LHS, (AArch64smull node:$MHS, node:$RHS))>,
3295 SMLALv8i8_v8i16, SMLALv4i16_v4i32, SMLALv2i32_v2i64>;
3296 defm : Neon_mulacc_widen_patterns<
3297 TriOpFrag<(add node:$LHS, (AArch64umull node:$MHS, node:$RHS))>,
3298 UMLALv8i8_v8i16, UMLALv4i16_v4i32, UMLALv2i32_v2i64>;
3299 defm : Neon_mulacc_widen_patterns<
3300 TriOpFrag<(sub node:$LHS, (AArch64smull node:$MHS, node:$RHS))>,
3301 SMLSLv8i8_v8i16, SMLSLv4i16_v4i32, SMLSLv2i32_v2i64>;
3302 defm : Neon_mulacc_widen_patterns<
3303 TriOpFrag<(sub node:$LHS, (AArch64umull node:$MHS, node:$RHS))>,
3304 UMLSLv8i8_v8i16, UMLSLv4i16_v4i32, UMLSLv2i32_v2i64>;
3306 // Patterns for 64-bit pmull
3307 def : Pat<(int_aarch64_neon_pmull64 V64:$Rn, V64:$Rm),
3308 (PMULLv1i64 V64:$Rn, V64:$Rm)>;
3309 def : Pat<(int_aarch64_neon_pmull64 (vector_extract (v2i64 V128:$Rn), (i64 1)),
3310 (vector_extract (v2i64 V128:$Rm), (i64 1))),
3311 (PMULLv2i64 V128:$Rn, V128:$Rm)>;
3313 // CodeGen patterns for addhn and subhn instructions, which can actually be
3314 // written in LLVM IR without too much difficulty.
3317 def : Pat<(v8i8 (trunc (v8i16 (AArch64vlshr (add V128:$Rn, V128:$Rm), (i32 8))))),
3318 (ADDHNv8i16_v8i8 V128:$Rn, V128:$Rm)>;
3319 def : Pat<(v4i16 (trunc (v4i32 (AArch64vlshr (add V128:$Rn, V128:$Rm),
3321 (ADDHNv4i32_v4i16 V128:$Rn, V128:$Rm)>;
3322 def : Pat<(v2i32 (trunc (v2i64 (AArch64vlshr (add V128:$Rn, V128:$Rm),
3324 (ADDHNv2i64_v2i32 V128:$Rn, V128:$Rm)>;
3325 def : Pat<(concat_vectors (v8i8 V64:$Rd),
3326 (trunc (v8i16 (AArch64vlshr (add V128:$Rn, V128:$Rm),
3328 (ADDHNv8i16_v16i8 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3329 V128:$Rn, V128:$Rm)>;
3330 def : Pat<(concat_vectors (v4i16 V64:$Rd),
3331 (trunc (v4i32 (AArch64vlshr (add V128:$Rn, V128:$Rm),
3333 (ADDHNv4i32_v8i16 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3334 V128:$Rn, V128:$Rm)>;
3335 def : Pat<(concat_vectors (v2i32 V64:$Rd),
3336 (trunc (v2i64 (AArch64vlshr (add V128:$Rn, V128:$Rm),
3338 (ADDHNv2i64_v4i32 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3339 V128:$Rn, V128:$Rm)>;
3342 def : Pat<(v8i8 (trunc (v8i16 (AArch64vlshr (sub V128:$Rn, V128:$Rm), (i32 8))))),
3343 (SUBHNv8i16_v8i8 V128:$Rn, V128:$Rm)>;
3344 def : Pat<(v4i16 (trunc (v4i32 (AArch64vlshr (sub V128:$Rn, V128:$Rm),
3346 (SUBHNv4i32_v4i16 V128:$Rn, V128:$Rm)>;
3347 def : Pat<(v2i32 (trunc (v2i64 (AArch64vlshr (sub V128:$Rn, V128:$Rm),
3349 (SUBHNv2i64_v2i32 V128:$Rn, V128:$Rm)>;
3350 def : Pat<(concat_vectors (v8i8 V64:$Rd),
3351 (trunc (v8i16 (AArch64vlshr (sub V128:$Rn, V128:$Rm),
3353 (SUBHNv8i16_v16i8 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3354 V128:$Rn, V128:$Rm)>;
3355 def : Pat<(concat_vectors (v4i16 V64:$Rd),
3356 (trunc (v4i32 (AArch64vlshr (sub V128:$Rn, V128:$Rm),
3358 (SUBHNv4i32_v8i16 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3359 V128:$Rn, V128:$Rm)>;
3360 def : Pat<(concat_vectors (v2i32 V64:$Rd),
3361 (trunc (v2i64 (AArch64vlshr (sub V128:$Rn, V128:$Rm),
3363 (SUBHNv2i64_v4i32 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3364 V128:$Rn, V128:$Rm)>;
3366 //----------------------------------------------------------------------------
3367 // AdvSIMD bitwise extract from vector instruction.
3368 //----------------------------------------------------------------------------
3370 defm EXT : SIMDBitwiseExtract<"ext">;
3372 def : Pat<(v4i16 (AArch64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
3373 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
3374 def : Pat<(v8i16 (AArch64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3375 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3376 def : Pat<(v2i32 (AArch64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
3377 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
3378 def : Pat<(v2f32 (AArch64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
3379 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
3380 def : Pat<(v4i32 (AArch64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3381 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3382 def : Pat<(v4f32 (AArch64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3383 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3384 def : Pat<(v2i64 (AArch64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3385 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3386 def : Pat<(v2f64 (AArch64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3387 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3388 def : Pat<(v4f16 (AArch64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
3389 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
3390 def : Pat<(v8f16 (AArch64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3391 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3393 // We use EXT to handle extract_subvector to copy the upper 64-bits of a
3395 def : Pat<(v8i8 (extract_subvector V128:$Rn, (i64 8))),
3396 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3397 def : Pat<(v4i16 (extract_subvector V128:$Rn, (i64 4))),
3398 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3399 def : Pat<(v2i32 (extract_subvector V128:$Rn, (i64 2))),
3400 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3401 def : Pat<(v1i64 (extract_subvector V128:$Rn, (i64 1))),
3402 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3403 def : Pat<(v4f16 (extract_subvector V128:$Rn, (i64 4))),
3404 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3405 def : Pat<(v2f32 (extract_subvector V128:$Rn, (i64 2))),
3406 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3407 def : Pat<(v1f64 (extract_subvector V128:$Rn, (i64 1))),
3408 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3411 //----------------------------------------------------------------------------
3412 // AdvSIMD zip vector
3413 //----------------------------------------------------------------------------
3415 defm TRN1 : SIMDZipVector<0b010, "trn1", AArch64trn1>;
3416 defm TRN2 : SIMDZipVector<0b110, "trn2", AArch64trn2>;
3417 defm UZP1 : SIMDZipVector<0b001, "uzp1", AArch64uzp1>;
3418 defm UZP2 : SIMDZipVector<0b101, "uzp2", AArch64uzp2>;
3419 defm ZIP1 : SIMDZipVector<0b011, "zip1", AArch64zip1>;
3420 defm ZIP2 : SIMDZipVector<0b111, "zip2", AArch64zip2>;
3422 //----------------------------------------------------------------------------
3423 // AdvSIMD TBL/TBX instructions
3424 //----------------------------------------------------------------------------
3426 defm TBL : SIMDTableLookup< 0, "tbl">;
3427 defm TBX : SIMDTableLookupTied<1, "tbx">;
3429 def : Pat<(v8i8 (int_aarch64_neon_tbl1 (v16i8 VecListOne128:$Rn), (v8i8 V64:$Ri))),
3430 (TBLv8i8One VecListOne128:$Rn, V64:$Ri)>;
3431 def : Pat<(v16i8 (int_aarch64_neon_tbl1 (v16i8 V128:$Ri), (v16i8 V128:$Rn))),
3432 (TBLv16i8One V128:$Ri, V128:$Rn)>;
3434 def : Pat<(v8i8 (int_aarch64_neon_tbx1 (v8i8 V64:$Rd),
3435 (v16i8 VecListOne128:$Rn), (v8i8 V64:$Ri))),
3436 (TBXv8i8One V64:$Rd, VecListOne128:$Rn, V64:$Ri)>;
3437 def : Pat<(v16i8 (int_aarch64_neon_tbx1 (v16i8 V128:$Rd),
3438 (v16i8 V128:$Ri), (v16i8 V128:$Rn))),
3439 (TBXv16i8One V128:$Rd, V128:$Ri, V128:$Rn)>;
3442 //----------------------------------------------------------------------------
3443 // AdvSIMD scalar CPY instruction
3444 //----------------------------------------------------------------------------
3446 defm CPY : SIMDScalarCPY<"cpy">;
3448 //----------------------------------------------------------------------------
3449 // AdvSIMD scalar pairwise instructions
3450 //----------------------------------------------------------------------------
3452 defm ADDP : SIMDPairwiseScalarD<0, 0b11011, "addp">;
3453 defm FADDP : SIMDPairwiseScalarSD<1, 0, 0b01101, "faddp">;
3454 defm FMAXNMP : SIMDPairwiseScalarSD<1, 0, 0b01100, "fmaxnmp">;
3455 defm FMAXP : SIMDPairwiseScalarSD<1, 0, 0b01111, "fmaxp">;
3456 defm FMINNMP : SIMDPairwiseScalarSD<1, 1, 0b01100, "fminnmp">;
3457 defm FMINP : SIMDPairwiseScalarSD<1, 1, 0b01111, "fminp">;
3458 def : Pat<(i64 (int_aarch64_neon_saddv (v2i64 V128:$Rn))),
3459 (ADDPv2i64p V128:$Rn)>;
3460 def : Pat<(i64 (int_aarch64_neon_uaddv (v2i64 V128:$Rn))),
3461 (ADDPv2i64p V128:$Rn)>;
3462 def : Pat<(f32 (int_aarch64_neon_faddv (v2f32 V64:$Rn))),
3463 (FADDPv2i32p V64:$Rn)>;
3464 def : Pat<(f32 (int_aarch64_neon_faddv (v4f32 V128:$Rn))),
3465 (FADDPv2i32p (EXTRACT_SUBREG (FADDPv4f32 V128:$Rn, V128:$Rn), dsub))>;
3466 def : Pat<(f64 (int_aarch64_neon_faddv (v2f64 V128:$Rn))),
3467 (FADDPv2i64p V128:$Rn)>;
3468 def : Pat<(f32 (int_aarch64_neon_fmaxnmv (v2f32 V64:$Rn))),
3469 (FMAXNMPv2i32p V64:$Rn)>;
3470 def : Pat<(f64 (int_aarch64_neon_fmaxnmv (v2f64 V128:$Rn))),
3471 (FMAXNMPv2i64p V128:$Rn)>;
3472 def : Pat<(f32 (int_aarch64_neon_fmaxv (v2f32 V64:$Rn))),
3473 (FMAXPv2i32p V64:$Rn)>;
3474 def : Pat<(f64 (int_aarch64_neon_fmaxv (v2f64 V128:$Rn))),
3475 (FMAXPv2i64p V128:$Rn)>;
3476 def : Pat<(f32 (int_aarch64_neon_fminnmv (v2f32 V64:$Rn))),
3477 (FMINNMPv2i32p V64:$Rn)>;
3478 def : Pat<(f64 (int_aarch64_neon_fminnmv (v2f64 V128:$Rn))),
3479 (FMINNMPv2i64p V128:$Rn)>;
3480 def : Pat<(f32 (int_aarch64_neon_fminv (v2f32 V64:$Rn))),
3481 (FMINPv2i32p V64:$Rn)>;
3482 def : Pat<(f64 (int_aarch64_neon_fminv (v2f64 V128:$Rn))),
3483 (FMINPv2i64p V128:$Rn)>;
3485 //----------------------------------------------------------------------------
3486 // AdvSIMD INS/DUP instructions
3487 //----------------------------------------------------------------------------
3489 def DUPv8i8gpr : SIMDDupFromMain<0, 0b00001, ".8b", v8i8, V64, GPR32>;
3490 def DUPv16i8gpr : SIMDDupFromMain<1, 0b00001, ".16b", v16i8, V128, GPR32>;
3491 def DUPv4i16gpr : SIMDDupFromMain<0, 0b00010, ".4h", v4i16, V64, GPR32>;
3492 def DUPv8i16gpr : SIMDDupFromMain<1, 0b00010, ".8h", v8i16, V128, GPR32>;
3493 def DUPv2i32gpr : SIMDDupFromMain<0, 0b00100, ".2s", v2i32, V64, GPR32>;
3494 def DUPv4i32gpr : SIMDDupFromMain<1, 0b00100, ".4s", v4i32, V128, GPR32>;
3495 def DUPv2i64gpr : SIMDDupFromMain<1, 0b01000, ".2d", v2i64, V128, GPR64>;
3497 def DUPv2i64lane : SIMDDup64FromElement;
3498 def DUPv2i32lane : SIMDDup32FromElement<0, ".2s", v2i32, V64>;
3499 def DUPv4i32lane : SIMDDup32FromElement<1, ".4s", v4i32, V128>;
3500 def DUPv4i16lane : SIMDDup16FromElement<0, ".4h", v4i16, V64>;
3501 def DUPv8i16lane : SIMDDup16FromElement<1, ".8h", v8i16, V128>;
3502 def DUPv8i8lane : SIMDDup8FromElement <0, ".8b", v8i8, V64>;
3503 def DUPv16i8lane : SIMDDup8FromElement <1, ".16b", v16i8, V128>;
3505 def : Pat<(v2f32 (AArch64dup (f32 FPR32:$Rn))),
3506 (v2f32 (DUPv2i32lane
3507 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rn, ssub),
3509 def : Pat<(v4f32 (AArch64dup (f32 FPR32:$Rn))),
3510 (v4f32 (DUPv4i32lane
3511 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rn, ssub),
3513 def : Pat<(v2f64 (AArch64dup (f64 FPR64:$Rn))),
3514 (v2f64 (DUPv2i64lane
3515 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$Rn, dsub),
3517 def : Pat<(v4f16 (AArch64dup (f16 FPR16:$Rn))),
3518 (v4f16 (DUPv4i16lane
3519 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), FPR16:$Rn, hsub),
3521 def : Pat<(v8f16 (AArch64dup (f16 FPR16:$Rn))),
3522 (v8f16 (DUPv8i16lane
3523 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), FPR16:$Rn, hsub),
3526 def : Pat<(v4f16 (AArch64duplane16 (v8f16 V128:$Rn), VectorIndexH:$imm)),
3527 (DUPv4i16lane V128:$Rn, VectorIndexH:$imm)>;
3528 def : Pat<(v8f16 (AArch64duplane16 (v8f16 V128:$Rn), VectorIndexH:$imm)),
3529 (DUPv8i16lane V128:$Rn, VectorIndexH:$imm)>;
3531 def : Pat<(v2f32 (AArch64duplane32 (v4f32 V128:$Rn), VectorIndexS:$imm)),
3532 (DUPv2i32lane V128:$Rn, VectorIndexS:$imm)>;
3533 def : Pat<(v4f32 (AArch64duplane32 (v4f32 V128:$Rn), VectorIndexS:$imm)),
3534 (DUPv4i32lane V128:$Rn, VectorIndexS:$imm)>;
3535 def : Pat<(v2f64 (AArch64duplane64 (v2f64 V128:$Rn), VectorIndexD:$imm)),
3536 (DUPv2i64lane V128:$Rn, VectorIndexD:$imm)>;
3538 // If there's an (AArch64dup (vector_extract ...) ...), we can use a duplane
3539 // instruction even if the types don't match: we just have to remap the lane
3540 // carefully. N.b. this trick only applies to truncations.
3541 def VecIndex_x2 : SDNodeXForm<imm, [{
3542 return CurDAG->getTargetConstant(2 * N->getZExtValue(), MVT::i64);
3544 def VecIndex_x4 : SDNodeXForm<imm, [{
3545 return CurDAG->getTargetConstant(4 * N->getZExtValue(), MVT::i64);
3547 def VecIndex_x8 : SDNodeXForm<imm, [{
3548 return CurDAG->getTargetConstant(8 * N->getZExtValue(), MVT::i64);
3551 multiclass DUPWithTruncPats<ValueType ResVT, ValueType Src64VT,
3552 ValueType Src128VT, ValueType ScalVT,
3553 Instruction DUP, SDNodeXForm IdxXFORM> {
3554 def : Pat<(ResVT (AArch64dup (ScalVT (vector_extract (Src128VT V128:$Rn),
3556 (DUP V128:$Rn, (IdxXFORM imm:$idx))>;
3558 def : Pat<(ResVT (AArch64dup (ScalVT (vector_extract (Src64VT V64:$Rn),
3560 (DUP (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), (IdxXFORM imm:$idx))>;
3563 defm : DUPWithTruncPats<v8i8, v4i16, v8i16, i32, DUPv8i8lane, VecIndex_x2>;
3564 defm : DUPWithTruncPats<v8i8, v2i32, v4i32, i32, DUPv8i8lane, VecIndex_x4>;
3565 defm : DUPWithTruncPats<v4i16, v2i32, v4i32, i32, DUPv4i16lane, VecIndex_x2>;
3567 defm : DUPWithTruncPats<v16i8, v4i16, v8i16, i32, DUPv16i8lane, VecIndex_x2>;
3568 defm : DUPWithTruncPats<v16i8, v2i32, v4i32, i32, DUPv16i8lane, VecIndex_x4>;
3569 defm : DUPWithTruncPats<v8i16, v2i32, v4i32, i32, DUPv8i16lane, VecIndex_x2>;
3571 multiclass DUPWithTrunci64Pats<ValueType ResVT, Instruction DUP,
3572 SDNodeXForm IdxXFORM> {
3573 def : Pat<(ResVT (AArch64dup (i32 (trunc (vector_extract (v2i64 V128:$Rn),
3575 (DUP V128:$Rn, (IdxXFORM imm:$idx))>;
3577 def : Pat<(ResVT (AArch64dup (i32 (trunc (vector_extract (v1i64 V64:$Rn),
3579 (DUP (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), (IdxXFORM imm:$idx))>;
3582 defm : DUPWithTrunci64Pats<v8i8, DUPv8i8lane, VecIndex_x8>;
3583 defm : DUPWithTrunci64Pats<v4i16, DUPv4i16lane, VecIndex_x4>;
3584 defm : DUPWithTrunci64Pats<v2i32, DUPv2i32lane, VecIndex_x2>;
3586 defm : DUPWithTrunci64Pats<v16i8, DUPv16i8lane, VecIndex_x8>;
3587 defm : DUPWithTrunci64Pats<v8i16, DUPv8i16lane, VecIndex_x4>;
3588 defm : DUPWithTrunci64Pats<v4i32, DUPv4i32lane, VecIndex_x2>;
3590 // SMOV and UMOV definitions, with some extra patterns for convenience
3594 def : Pat<(sext_inreg (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx), i8),
3595 (i32 (SMOVvi8to32 V128:$Rn, VectorIndexB:$idx))>;
3596 def : Pat<(sext_inreg (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx), i8),
3597 (i64 (SMOVvi8to64 V128:$Rn, VectorIndexB:$idx))>;
3598 def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
3599 (i32 (SMOVvi16to32 V128:$Rn, VectorIndexH:$idx))>;
3600 def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
3601 (i64 (SMOVvi16to64 V128:$Rn, VectorIndexH:$idx))>;
3602 def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
3603 (i32 (SMOVvi16to32 V128:$Rn, VectorIndexH:$idx))>;
3604 def : Pat<(sext (i32 (vector_extract (v4i32 V128:$Rn), VectorIndexS:$idx))),
3605 (i64 (SMOVvi32to64 V128:$Rn, VectorIndexS:$idx))>;
3607 // Extracting i8 or i16 elements will have the zero-extend transformed to
3608 // an 'and' mask by type legalization since neither i8 nor i16 are legal types
3609 // for AArch64. Match these patterns here since UMOV already zeroes out the high
3610 // bits of the destination register.
3611 def : Pat<(and (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx),
3613 (i32 (UMOVvi8 V128:$Rn, VectorIndexB:$idx))>;
3614 def : Pat<(and (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),
3616 (i32 (UMOVvi16 V128:$Rn, VectorIndexH:$idx))>;
3620 def : Pat<(v16i8 (scalar_to_vector GPR32:$Rn)),
3621 (SUBREG_TO_REG (i32 0),
3622 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3623 def : Pat<(v8i8 (scalar_to_vector GPR32:$Rn)),
3624 (SUBREG_TO_REG (i32 0),
3625 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3627 def : Pat<(v8i16 (scalar_to_vector GPR32:$Rn)),
3628 (SUBREG_TO_REG (i32 0),
3629 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3630 def : Pat<(v4i16 (scalar_to_vector GPR32:$Rn)),
3631 (SUBREG_TO_REG (i32 0),
3632 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3634 def : Pat<(v2i32 (scalar_to_vector (i32 FPR32:$Rn))),
3635 (v2i32 (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)),
3636 (i32 FPR32:$Rn), ssub))>;
3637 def : Pat<(v4i32 (scalar_to_vector (i32 FPR32:$Rn))),
3638 (v4i32 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
3639 (i32 FPR32:$Rn), ssub))>;
3640 def : Pat<(v2i64 (scalar_to_vector (i64 FPR64:$Rn))),
3641 (v2i64 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
3642 (i64 FPR64:$Rn), dsub))>;
3644 def : Pat<(v4f32 (scalar_to_vector (f32 FPR32:$Rn))),
3645 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rn, ssub)>;
3646 def : Pat<(v2f32 (scalar_to_vector (f32 FPR32:$Rn))),
3647 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), FPR32:$Rn, ssub)>;
3648 def : Pat<(v2f64 (scalar_to_vector (f64 FPR64:$Rn))),
3649 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$Rn, dsub)>;
3651 def : Pat<(v4f16 (vector_insert (v4f16 V64:$Rn),
3652 (f16 FPR16:$Rm), (i64 VectorIndexS:$imm))),
3655 (v8f16 (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), V64:$Rn, dsub)),
3657 (v8f16 (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), FPR16:$Rm, hsub)),
3661 def : Pat<(v8f16 (vector_insert (v8f16 V128:$Rn),
3662 (f16 FPR16:$Rm), (i64 VectorIndexH:$imm))),
3664 V128:$Rn, VectorIndexH:$imm,
3665 (v8f16 (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), FPR16:$Rm, hsub)),
3668 def : Pat<(v2f32 (vector_insert (v2f32 V64:$Rn),
3669 (f32 FPR32:$Rm), (i64 VectorIndexS:$imm))),
3672 (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), V64:$Rn, dsub)),
3674 (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rm, ssub)),
3677 def : Pat<(v4f32 (vector_insert (v4f32 V128:$Rn),
3678 (f32 FPR32:$Rm), (i64 VectorIndexS:$imm))),
3680 V128:$Rn, VectorIndexS:$imm,
3681 (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rm, ssub)),
3683 def : Pat<(v2f64 (vector_insert (v2f64 V128:$Rn),
3684 (f64 FPR64:$Rm), (i64 VectorIndexD:$imm))),
3686 V128:$Rn, VectorIndexD:$imm,
3687 (v2f64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$Rm, dsub)),
3690 // Copy an element at a constant index in one vector into a constant indexed
3691 // element of another.
3692 // FIXME refactor to a shared class/dev parameterized on vector type, vector
3693 // index type and INS extension
3694 def : Pat<(v16i8 (int_aarch64_neon_vcopy_lane
3695 (v16i8 V128:$Vd), VectorIndexB:$idx, (v16i8 V128:$Vs),
3696 VectorIndexB:$idx2)),
3698 V128:$Vd, VectorIndexB:$idx, V128:$Vs, VectorIndexB:$idx2)
3700 def : Pat<(v8i16 (int_aarch64_neon_vcopy_lane
3701 (v8i16 V128:$Vd), VectorIndexH:$idx, (v8i16 V128:$Vs),
3702 VectorIndexH:$idx2)),
3704 V128:$Vd, VectorIndexH:$idx, V128:$Vs, VectorIndexH:$idx2)
3706 def : Pat<(v4i32 (int_aarch64_neon_vcopy_lane
3707 (v4i32 V128:$Vd), VectorIndexS:$idx, (v4i32 V128:$Vs),
3708 VectorIndexS:$idx2)),
3710 V128:$Vd, VectorIndexS:$idx, V128:$Vs, VectorIndexS:$idx2)
3712 def : Pat<(v2i64 (int_aarch64_neon_vcopy_lane
3713 (v2i64 V128:$Vd), VectorIndexD:$idx, (v2i64 V128:$Vs),
3714 VectorIndexD:$idx2)),
3716 V128:$Vd, VectorIndexD:$idx, V128:$Vs, VectorIndexD:$idx2)
3719 multiclass Neon_INS_elt_pattern<ValueType VT128, ValueType VT64,
3720 ValueType VTScal, Instruction INS> {
3721 def : Pat<(VT128 (vector_insert V128:$src,
3722 (VTScal (vector_extract (VT128 V128:$Rn), imm:$Immn)),
3724 (INS V128:$src, imm:$Immd, V128:$Rn, imm:$Immn)>;
3726 def : Pat<(VT128 (vector_insert V128:$src,
3727 (VTScal (vector_extract (VT64 V64:$Rn), imm:$Immn)),
3729 (INS V128:$src, imm:$Immd,
3730 (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), imm:$Immn)>;
3732 def : Pat<(VT64 (vector_insert V64:$src,
3733 (VTScal (vector_extract (VT128 V128:$Rn), imm:$Immn)),
3735 (EXTRACT_SUBREG (INS (SUBREG_TO_REG (i64 0), V64:$src, dsub),
3736 imm:$Immd, V128:$Rn, imm:$Immn),
3739 def : Pat<(VT64 (vector_insert V64:$src,
3740 (VTScal (vector_extract (VT64 V64:$Rn), imm:$Immn)),
3743 (INS (SUBREG_TO_REG (i64 0), V64:$src, dsub), imm:$Immd,
3744 (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), imm:$Immn),
3748 defm : Neon_INS_elt_pattern<v8f16, v4f16, f16, INSvi16lane>;
3749 defm : Neon_INS_elt_pattern<v4f32, v2f32, f32, INSvi32lane>;
3750 defm : Neon_INS_elt_pattern<v2f64, v1f64, f64, INSvi64lane>;
3751 defm : Neon_INS_elt_pattern<v16i8, v8i8, i32, INSvi8lane>;
3752 defm : Neon_INS_elt_pattern<v8i16, v4i16, i32, INSvi16lane>;
3753 defm : Neon_INS_elt_pattern<v4i32, v2i32, i32, INSvi32lane>;
3754 defm : Neon_INS_elt_pattern<v2i64, v1i64, i64, INSvi32lane>;
3757 // Floating point vector extractions are codegen'd as either a sequence of
3758 // subregister extractions, possibly fed by an INS if the lane number is
3759 // anything other than zero.
3760 def : Pat<(vector_extract (v2f64 V128:$Rn), 0),
3761 (f64 (EXTRACT_SUBREG V128:$Rn, dsub))>;
3762 def : Pat<(vector_extract (v4f32 V128:$Rn), 0),
3763 (f32 (EXTRACT_SUBREG V128:$Rn, ssub))>;
3764 def : Pat<(vector_extract (v8f16 V128:$Rn), 0),
3765 (f16 (EXTRACT_SUBREG V128:$Rn, hsub))>;
3766 def : Pat<(vector_extract (v2f64 V128:$Rn), VectorIndexD:$idx),
3767 (f64 (EXTRACT_SUBREG
3768 (INSvi64lane (v2f64 (IMPLICIT_DEF)), 0,
3769 V128:$Rn, VectorIndexD:$idx),
3771 def : Pat<(vector_extract (v4f32 V128:$Rn), VectorIndexS:$idx),
3772 (f32 (EXTRACT_SUBREG
3773 (INSvi32lane (v4f32 (IMPLICIT_DEF)), 0,
3774 V128:$Rn, VectorIndexS:$idx),
3776 def : Pat<(vector_extract (v8f16 V128:$Rn), VectorIndexH:$idx),
3777 (f16 (EXTRACT_SUBREG
3778 (INSvi16lane (v8f16 (IMPLICIT_DEF)), 0,
3779 V128:$Rn, VectorIndexH:$idx),
3782 // All concat_vectors operations are canonicalised to act on i64 vectors for
3783 // AArch64. In the general case we need an instruction, which had just as well be
3785 class ConcatPat<ValueType DstTy, ValueType SrcTy>
3786 : Pat<(DstTy (concat_vectors (SrcTy V64:$Rd), V64:$Rn)),
3787 (INSvi64lane (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), 1,
3788 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rn, dsub), 0)>;
3790 def : ConcatPat<v2i64, v1i64>;
3791 def : ConcatPat<v2f64, v1f64>;
3792 def : ConcatPat<v4i32, v2i32>;
3793 def : ConcatPat<v4f32, v2f32>;
3794 def : ConcatPat<v8i16, v4i16>;
3795 def : ConcatPat<v8f16, v4f16>;
3796 def : ConcatPat<v16i8, v8i8>;
3798 // If the high lanes are undef, though, we can just ignore them:
3799 class ConcatUndefPat<ValueType DstTy, ValueType SrcTy>
3800 : Pat<(DstTy (concat_vectors (SrcTy V64:$Rn), undef)),
3801 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rn, dsub)>;
3803 def : ConcatUndefPat<v2i64, v1i64>;
3804 def : ConcatUndefPat<v2f64, v1f64>;
3805 def : ConcatUndefPat<v4i32, v2i32>;
3806 def : ConcatUndefPat<v4f32, v2f32>;
3807 def : ConcatUndefPat<v8i16, v4i16>;
3808 def : ConcatUndefPat<v16i8, v8i8>;
3810 //----------------------------------------------------------------------------
3811 // AdvSIMD across lanes instructions
3812 //----------------------------------------------------------------------------
3814 defm ADDV : SIMDAcrossLanesBHS<0, 0b11011, "addv">;
3815 defm SMAXV : SIMDAcrossLanesBHS<0, 0b01010, "smaxv">;
3816 defm SMINV : SIMDAcrossLanesBHS<0, 0b11010, "sminv">;
3817 defm UMAXV : SIMDAcrossLanesBHS<1, 0b01010, "umaxv">;
3818 defm UMINV : SIMDAcrossLanesBHS<1, 0b11010, "uminv">;
3819 defm SADDLV : SIMDAcrossLanesHSD<0, 0b00011, "saddlv">;
3820 defm UADDLV : SIMDAcrossLanesHSD<1, 0b00011, "uaddlv">;
3821 defm FMAXNMV : SIMDAcrossLanesS<0b01100, 0, "fmaxnmv", int_aarch64_neon_fmaxnmv>;
3822 defm FMAXV : SIMDAcrossLanesS<0b01111, 0, "fmaxv", int_aarch64_neon_fmaxv>;
3823 defm FMINNMV : SIMDAcrossLanesS<0b01100, 1, "fminnmv", int_aarch64_neon_fminnmv>;
3824 defm FMINV : SIMDAcrossLanesS<0b01111, 1, "fminv", int_aarch64_neon_fminv>;
3826 multiclass SIMDAcrossLanesSignedIntrinsic<string baseOpc, Intrinsic intOp> {
3827 // If there is a sign extension after this intrinsic, consume it as smov already
3829 def : Pat<(i32 (sext_inreg (i32 (intOp (v8i8 V64:$Rn))), i8)),
3831 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3832 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3834 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
3836 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3837 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3839 // If there is a sign extension after this intrinsic, consume it as smov already
3841 def : Pat<(i32 (sext_inreg (i32 (intOp (v16i8 V128:$Rn))), i8)),
3843 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3844 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3846 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
3848 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3849 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3851 // If there is a sign extension after this intrinsic, consume it as smov already
3853 def : Pat<(i32 (sext_inreg (i32 (intOp (v4i16 V64:$Rn))), i16)),
3855 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3856 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3858 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
3860 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3861 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3863 // If there is a sign extension after this intrinsic, consume it as smov already
3865 def : Pat<(i32 (sext_inreg (i32 (intOp (v8i16 V128:$Rn))), i16)),
3867 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3868 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3870 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
3872 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3873 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3876 def : Pat<(i32 (intOp (v4i32 V128:$Rn))),
3877 (i32 (EXTRACT_SUBREG
3878 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3879 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), ssub),
3883 multiclass SIMDAcrossLanesUnsignedIntrinsic<string baseOpc, Intrinsic intOp> {
3884 // If there is a masking operation keeping only what has been actually
3885 // generated, consume it.
3886 def : Pat<(i32 (and (i32 (intOp (v8i8 V64:$Rn))), maski8_or_more)),
3887 (i32 (EXTRACT_SUBREG
3888 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3889 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3891 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
3892 (i32 (EXTRACT_SUBREG
3893 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3894 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3896 // If there is a masking operation keeping only what has been actually
3897 // generated, consume it.
3898 def : Pat<(i32 (and (i32 (intOp (v16i8 V128:$Rn))), maski8_or_more)),
3899 (i32 (EXTRACT_SUBREG
3900 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3901 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3903 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
3904 (i32 (EXTRACT_SUBREG
3905 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3906 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3909 // If there is a masking operation keeping only what has been actually
3910 // generated, consume it.
3911 def : Pat<(i32 (and (i32 (intOp (v4i16 V64:$Rn))), maski16_or_more)),
3912 (i32 (EXTRACT_SUBREG
3913 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3914 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3916 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
3917 (i32 (EXTRACT_SUBREG
3918 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3919 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3921 // If there is a masking operation keeping only what has been actually
3922 // generated, consume it.
3923 def : Pat<(i32 (and (i32 (intOp (v8i16 V128:$Rn))), maski16_or_more)),
3924 (i32 (EXTRACT_SUBREG
3925 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3926 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3928 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
3929 (i32 (EXTRACT_SUBREG
3930 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3931 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3934 def : Pat<(i32 (intOp (v4i32 V128:$Rn))),
3935 (i32 (EXTRACT_SUBREG
3936 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3937 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), ssub),
3942 multiclass SIMDAcrossLanesSignedLongIntrinsic<string baseOpc, Intrinsic intOp> {
3943 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
3945 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3946 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), hsub),
3948 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
3950 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3951 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), hsub),
3954 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
3955 (i32 (EXTRACT_SUBREG
3956 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3957 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), ssub),
3959 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
3960 (i32 (EXTRACT_SUBREG
3961 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3962 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), ssub),
3965 def : Pat<(i64 (intOp (v4i32 V128:$Rn))),
3966 (i64 (EXTRACT_SUBREG
3967 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3968 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), dsub),
3972 multiclass SIMDAcrossLanesUnsignedLongIntrinsic<string baseOpc,
3974 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
3975 (i32 (EXTRACT_SUBREG
3976 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3977 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), hsub),
3979 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
3980 (i32 (EXTRACT_SUBREG
3981 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3982 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), hsub),
3985 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
3986 (i32 (EXTRACT_SUBREG
3987 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3988 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), ssub),
3990 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
3991 (i32 (EXTRACT_SUBREG
3992 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3993 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), ssub),
3996 def : Pat<(i64 (intOp (v4i32 V128:$Rn))),
3997 (i64 (EXTRACT_SUBREG
3998 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3999 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), dsub),
4003 defm : SIMDAcrossLanesSignedIntrinsic<"ADDV", int_aarch64_neon_saddv>;
4004 // vaddv_[su]32 is special; -> ADDP Vd.2S,Vn.2S,Vm.2S; return Vd.s[0];Vn==Vm
4005 def : Pat<(i32 (int_aarch64_neon_saddv (v2i32 V64:$Rn))),
4006 (EXTRACT_SUBREG (ADDPv2i32 V64:$Rn, V64:$Rn), ssub)>;
4008 defm : SIMDAcrossLanesUnsignedIntrinsic<"ADDV", int_aarch64_neon_uaddv>;
4009 // vaddv_[su]32 is special; -> ADDP Vd.2S,Vn.2S,Vm.2S; return Vd.s[0];Vn==Vm
4010 def : Pat<(i32 (int_aarch64_neon_uaddv (v2i32 V64:$Rn))),
4011 (EXTRACT_SUBREG (ADDPv2i32 V64:$Rn, V64:$Rn), ssub)>;
4013 defm : SIMDAcrossLanesSignedIntrinsic<"SMAXV", int_aarch64_neon_smaxv>;
4014 def : Pat<(i32 (int_aarch64_neon_smaxv (v2i32 V64:$Rn))),
4015 (EXTRACT_SUBREG (SMAXPv2i32 V64:$Rn, V64:$Rn), ssub)>;
4017 defm : SIMDAcrossLanesSignedIntrinsic<"SMINV", int_aarch64_neon_sminv>;
4018 def : Pat<(i32 (int_aarch64_neon_sminv (v2i32 V64:$Rn))),
4019 (EXTRACT_SUBREG (SMINPv2i32 V64:$Rn, V64:$Rn), ssub)>;
4021 defm : SIMDAcrossLanesUnsignedIntrinsic<"UMAXV", int_aarch64_neon_umaxv>;
4022 def : Pat<(i32 (int_aarch64_neon_umaxv (v2i32 V64:$Rn))),
4023 (EXTRACT_SUBREG (UMAXPv2i32 V64:$Rn, V64:$Rn), ssub)>;
4025 defm : SIMDAcrossLanesUnsignedIntrinsic<"UMINV", int_aarch64_neon_uminv>;
4026 def : Pat<(i32 (int_aarch64_neon_uminv (v2i32 V64:$Rn))),
4027 (EXTRACT_SUBREG (UMINPv2i32 V64:$Rn, V64:$Rn), ssub)>;
4029 defm : SIMDAcrossLanesSignedLongIntrinsic<"SADDLV", int_aarch64_neon_saddlv>;
4030 defm : SIMDAcrossLanesUnsignedLongIntrinsic<"UADDLV", int_aarch64_neon_uaddlv>;
4032 // The vaddlv_s32 intrinsic gets mapped to SADDLP.
4033 def : Pat<(i64 (int_aarch64_neon_saddlv (v2i32 V64:$Rn))),
4034 (i64 (EXTRACT_SUBREG
4035 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4036 (SADDLPv2i32_v1i64 V64:$Rn), dsub),
4038 // The vaddlv_u32 intrinsic gets mapped to UADDLP.
4039 def : Pat<(i64 (int_aarch64_neon_uaddlv (v2i32 V64:$Rn))),
4040 (i64 (EXTRACT_SUBREG
4041 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4042 (UADDLPv2i32_v1i64 V64:$Rn), dsub),
4045 //------------------------------------------------------------------------------
4046 // AdvSIMD modified immediate instructions
4047 //------------------------------------------------------------------------------
4050 defm BIC : SIMDModifiedImmVectorShiftTied<1, 0b11, 0b01, "bic", AArch64bici>;
4052 defm ORR : SIMDModifiedImmVectorShiftTied<0, 0b11, 0b01, "orr", AArch64orri>;
4054 def : InstAlias<"bic $Vd.4h, $imm", (BICv4i16 V64:$Vd, imm0_255:$imm, 0)>;
4055 def : InstAlias<"bic $Vd.8h, $imm", (BICv8i16 V128:$Vd, imm0_255:$imm, 0)>;
4056 def : InstAlias<"bic $Vd.2s, $imm", (BICv2i32 V64:$Vd, imm0_255:$imm, 0)>;
4057 def : InstAlias<"bic $Vd.4s, $imm", (BICv4i32 V128:$Vd, imm0_255:$imm, 0)>;
4059 def : InstAlias<"bic.4h $Vd, $imm", (BICv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
4060 def : InstAlias<"bic.8h $Vd, $imm", (BICv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
4061 def : InstAlias<"bic.2s $Vd, $imm", (BICv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
4062 def : InstAlias<"bic.4s $Vd, $imm", (BICv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
4064 def : InstAlias<"orr $Vd.4h, $imm", (ORRv4i16 V64:$Vd, imm0_255:$imm, 0)>;
4065 def : InstAlias<"orr $Vd.8h, $imm", (ORRv8i16 V128:$Vd, imm0_255:$imm, 0)>;
4066 def : InstAlias<"orr $Vd.2s, $imm", (ORRv2i32 V64:$Vd, imm0_255:$imm, 0)>;
4067 def : InstAlias<"orr $Vd.4s, $imm", (ORRv4i32 V128:$Vd, imm0_255:$imm, 0)>;
4069 def : InstAlias<"orr.4h $Vd, $imm", (ORRv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
4070 def : InstAlias<"orr.8h $Vd, $imm", (ORRv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
4071 def : InstAlias<"orr.2s $Vd, $imm", (ORRv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
4072 def : InstAlias<"orr.4s $Vd, $imm", (ORRv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
4075 def FMOVv2f64_ns : SIMDModifiedImmVectorNoShift<1, 1, 0b1111, V128, fpimm8,
4077 [(set (v2f64 V128:$Rd), (AArch64fmov imm0_255:$imm8))]>;
4078 def FMOVv2f32_ns : SIMDModifiedImmVectorNoShift<0, 0, 0b1111, V64, fpimm8,
4080 [(set (v2f32 V64:$Rd), (AArch64fmov imm0_255:$imm8))]>;
4081 def FMOVv4f32_ns : SIMDModifiedImmVectorNoShift<1, 0, 0b1111, V128, fpimm8,
4083 [(set (v4f32 V128:$Rd), (AArch64fmov imm0_255:$imm8))]>;
4087 // EDIT byte mask: scalar
4088 let isReMaterializable = 1, isAsCheapAsAMove = 1 in
4089 def MOVID : SIMDModifiedImmScalarNoShift<0, 1, 0b1110, "movi",
4090 [(set FPR64:$Rd, simdimmtype10:$imm8)]>;
4091 // The movi_edit node has the immediate value already encoded, so we use
4092 // a plain imm0_255 here.
4093 def : Pat<(f64 (AArch64movi_edit imm0_255:$shift)),
4094 (MOVID imm0_255:$shift)>;
4096 def : Pat<(v1i64 immAllZerosV), (MOVID (i32 0))>;
4097 def : Pat<(v2i32 immAllZerosV), (MOVID (i32 0))>;
4098 def : Pat<(v4i16 immAllZerosV), (MOVID (i32 0))>;
4099 def : Pat<(v8i8 immAllZerosV), (MOVID (i32 0))>;
4101 def : Pat<(v1i64 immAllOnesV), (MOVID (i32 255))>;
4102 def : Pat<(v2i32 immAllOnesV), (MOVID (i32 255))>;
4103 def : Pat<(v4i16 immAllOnesV), (MOVID (i32 255))>;
4104 def : Pat<(v8i8 immAllOnesV), (MOVID (i32 255))>;
4106 // EDIT byte mask: 2d
4108 // The movi_edit node has the immediate value already encoded, so we use
4109 // a plain imm0_255 in the pattern
4110 let isReMaterializable = 1, isAsCheapAsAMove = 1 in
4111 def MOVIv2d_ns : SIMDModifiedImmVectorNoShift<1, 1, 0b1110, V128,
4114 [(set (v2i64 V128:$Rd), (AArch64movi_edit imm0_255:$imm8))]>;
4117 // Use movi.2d to materialize 0.0 if the HW does zero-cycle zeroing.
4118 // Complexity is added to break a tie with a plain MOVI.
4119 let AddedComplexity = 1 in {
4120 def : Pat<(f32 fpimm0),
4121 (f32 (EXTRACT_SUBREG (v2i64 (MOVIv2d_ns (i32 0))), ssub))>,
4123 def : Pat<(f64 fpimm0),
4124 (f64 (EXTRACT_SUBREG (v2i64 (MOVIv2d_ns (i32 0))), dsub))>,
4128 def : Pat<(v2i64 immAllZerosV), (MOVIv2d_ns (i32 0))>;
4129 def : Pat<(v4i32 immAllZerosV), (MOVIv2d_ns (i32 0))>;
4130 def : Pat<(v8i16 immAllZerosV), (MOVIv2d_ns (i32 0))>;
4131 def : Pat<(v16i8 immAllZerosV), (MOVIv2d_ns (i32 0))>;
4133 def : Pat<(v2i64 immAllOnesV), (MOVIv2d_ns (i32 255))>;
4134 def : Pat<(v4i32 immAllOnesV), (MOVIv2d_ns (i32 255))>;
4135 def : Pat<(v8i16 immAllOnesV), (MOVIv2d_ns (i32 255))>;
4136 def : Pat<(v16i8 immAllOnesV), (MOVIv2d_ns (i32 255))>;
4138 def : Pat<(v2f64 (AArch64dup (f64 fpimm0))), (MOVIv2d_ns (i32 0))>;
4139 def : Pat<(v4f32 (AArch64dup (f32 fpimm0))), (MOVIv2d_ns (i32 0))>;
4141 // EDIT per word & halfword: 2s, 4h, 4s, & 8h
4142 defm MOVI : SIMDModifiedImmVectorShift<0, 0b10, 0b00, "movi">;
4144 def : InstAlias<"movi $Vd.4h, $imm", (MOVIv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
4145 def : InstAlias<"movi $Vd.8h, $imm", (MOVIv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
4146 def : InstAlias<"movi $Vd.2s, $imm", (MOVIv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
4147 def : InstAlias<"movi $Vd.4s, $imm", (MOVIv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
4149 def : InstAlias<"movi.4h $Vd, $imm", (MOVIv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
4150 def : InstAlias<"movi.8h $Vd, $imm", (MOVIv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
4151 def : InstAlias<"movi.2s $Vd, $imm", (MOVIv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
4152 def : InstAlias<"movi.4s $Vd, $imm", (MOVIv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
4154 def : Pat<(v2i32 (AArch64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
4155 (MOVIv2i32 imm0_255:$imm8, imm:$shift)>;
4156 def : Pat<(v4i32 (AArch64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
4157 (MOVIv4i32 imm0_255:$imm8, imm:$shift)>;
4158 def : Pat<(v4i16 (AArch64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
4159 (MOVIv4i16 imm0_255:$imm8, imm:$shift)>;
4160 def : Pat<(v8i16 (AArch64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
4161 (MOVIv8i16 imm0_255:$imm8, imm:$shift)>;
4163 // EDIT per word: 2s & 4s with MSL shifter
4164 def MOVIv2s_msl : SIMDModifiedImmMoveMSL<0, 0, {1,1,0,?}, V64, "movi", ".2s",
4165 [(set (v2i32 V64:$Rd),
4166 (AArch64movi_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
4167 def MOVIv4s_msl : SIMDModifiedImmMoveMSL<1, 0, {1,1,0,?}, V128, "movi", ".4s",
4168 [(set (v4i32 V128:$Rd),
4169 (AArch64movi_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
4171 // Per byte: 8b & 16b
4172 def MOVIv8b_ns : SIMDModifiedImmVectorNoShift<0, 0, 0b1110, V64, imm0_255,
4174 [(set (v8i8 V64:$Rd), (AArch64movi imm0_255:$imm8))]>;
4175 def MOVIv16b_ns : SIMDModifiedImmVectorNoShift<1, 0, 0b1110, V128, imm0_255,
4177 [(set (v16i8 V128:$Rd), (AArch64movi imm0_255:$imm8))]>;
4181 // EDIT per word & halfword: 2s, 4h, 4s, & 8h
4182 defm MVNI : SIMDModifiedImmVectorShift<1, 0b10, 0b00, "mvni">;
4184 def : InstAlias<"mvni $Vd.4h, $imm", (MVNIv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
4185 def : InstAlias<"mvni $Vd.8h, $imm", (MVNIv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
4186 def : InstAlias<"mvni $Vd.2s, $imm", (MVNIv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
4187 def : InstAlias<"mvni $Vd.4s, $imm", (MVNIv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
4189 def : InstAlias<"mvni.4h $Vd, $imm", (MVNIv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
4190 def : InstAlias<"mvni.8h $Vd, $imm", (MVNIv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
4191 def : InstAlias<"mvni.2s $Vd, $imm", (MVNIv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
4192 def : InstAlias<"mvni.4s $Vd, $imm", (MVNIv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
4194 def : Pat<(v2i32 (AArch64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
4195 (MVNIv2i32 imm0_255:$imm8, imm:$shift)>;
4196 def : Pat<(v4i32 (AArch64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
4197 (MVNIv4i32 imm0_255:$imm8, imm:$shift)>;
4198 def : Pat<(v4i16 (AArch64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
4199 (MVNIv4i16 imm0_255:$imm8, imm:$shift)>;
4200 def : Pat<(v8i16 (AArch64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
4201 (MVNIv8i16 imm0_255:$imm8, imm:$shift)>;
4203 // EDIT per word: 2s & 4s with MSL shifter
4204 def MVNIv2s_msl : SIMDModifiedImmMoveMSL<0, 1, {1,1,0,?}, V64, "mvni", ".2s",
4205 [(set (v2i32 V64:$Rd),
4206 (AArch64mvni_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
4207 def MVNIv4s_msl : SIMDModifiedImmMoveMSL<1, 1, {1,1,0,?}, V128, "mvni", ".4s",
4208 [(set (v4i32 V128:$Rd),
4209 (AArch64mvni_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
4211 //----------------------------------------------------------------------------
4212 // AdvSIMD indexed element
4213 //----------------------------------------------------------------------------
4215 let hasSideEffects = 0 in {
4216 defm FMLA : SIMDFPIndexedSDTied<0, 0b0001, "fmla">;
4217 defm FMLS : SIMDFPIndexedSDTied<0, 0b0101, "fmls">;
4220 // NOTE: Operands are reordered in the FMLA/FMLS PatFrags because the
4221 // instruction expects the addend first, while the intrinsic expects it last.
4223 // On the other hand, there are quite a few valid combinatorial options due to
4224 // the commutativity of multiplication and the fact that (-x) * y = x * (-y).
4225 defm : SIMDFPIndexedSDTiedPatterns<"FMLA",
4226 TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)>>;
4227 defm : SIMDFPIndexedSDTiedPatterns<"FMLA",
4228 TriOpFrag<(fma node:$MHS, node:$RHS, node:$LHS)>>;
4230 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
4231 TriOpFrag<(fma node:$MHS, (fneg node:$RHS), node:$LHS)> >;
4232 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
4233 TriOpFrag<(fma node:$RHS, (fneg node:$MHS), node:$LHS)> >;
4234 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
4235 TriOpFrag<(fma (fneg node:$RHS), node:$MHS, node:$LHS)> >;
4236 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
4237 TriOpFrag<(fma (fneg node:$MHS), node:$RHS, node:$LHS)> >;
4239 multiclass FMLSIndexedAfterNegPatterns<SDPatternOperator OpNode> {
4240 // 3 variants for the .2s version: DUPLANE from 128-bit, DUPLANE from 64-bit
4242 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
4243 (AArch64duplane32 (v4f32 (fneg V128:$Rm)),
4244 VectorIndexS:$idx))),
4245 (FMLSv2i32_indexed V64:$Rd, V64:$Rn, V128:$Rm, VectorIndexS:$idx)>;
4246 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
4247 (v2f32 (AArch64duplane32
4248 (v4f32 (insert_subvector undef,
4249 (v2f32 (fneg V64:$Rm)),
4251 VectorIndexS:$idx)))),
4252 (FMLSv2i32_indexed V64:$Rd, V64:$Rn,
4253 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub),
4254 VectorIndexS:$idx)>;
4255 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
4256 (AArch64dup (f32 (fneg FPR32Op:$Rm))))),
4257 (FMLSv2i32_indexed V64:$Rd, V64:$Rn,
4258 (SUBREG_TO_REG (i32 0), FPR32Op:$Rm, ssub), (i64 0))>;
4260 // 3 variants for the .4s version: DUPLANE from 128-bit, DUPLANE from 64-bit
4262 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
4263 (AArch64duplane32 (v4f32 (fneg V128:$Rm)),
4264 VectorIndexS:$idx))),
4265 (FMLSv4i32_indexed V128:$Rd, V128:$Rn, V128:$Rm,
4266 VectorIndexS:$idx)>;
4267 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
4268 (v4f32 (AArch64duplane32
4269 (v4f32 (insert_subvector undef,
4270 (v2f32 (fneg V64:$Rm)),
4272 VectorIndexS:$idx)))),
4273 (FMLSv4i32_indexed V128:$Rd, V128:$Rn,
4274 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub),
4275 VectorIndexS:$idx)>;
4276 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
4277 (AArch64dup (f32 (fneg FPR32Op:$Rm))))),
4278 (FMLSv4i32_indexed V128:$Rd, V128:$Rn,
4279 (SUBREG_TO_REG (i32 0), FPR32Op:$Rm, ssub), (i64 0))>;
4281 // 2 variants for the .2d version: DUPLANE from 128-bit, and DUP scalar
4282 // (DUPLANE from 64-bit would be trivial).
4283 def : Pat<(v2f64 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn),
4284 (AArch64duplane64 (v2f64 (fneg V128:$Rm)),
4285 VectorIndexD:$idx))),
4287 V128:$Rd, V128:$Rn, V128:$Rm, VectorIndexS:$idx)>;
4288 def : Pat<(v2f64 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn),
4289 (AArch64dup (f64 (fneg FPR64Op:$Rm))))),
4290 (FMLSv2i64_indexed V128:$Rd, V128:$Rn,
4291 (SUBREG_TO_REG (i32 0), FPR64Op:$Rm, dsub), (i64 0))>;
4293 // 2 variants for 32-bit scalar version: extract from .2s or from .4s
4294 def : Pat<(f32 (OpNode (f32 FPR32:$Rd), (f32 FPR32:$Rn),
4295 (vector_extract (v4f32 (fneg V128:$Rm)),
4296 VectorIndexS:$idx))),
4297 (FMLSv1i32_indexed FPR32:$Rd, FPR32:$Rn,
4298 V128:$Rm, VectorIndexS:$idx)>;
4299 def : Pat<(f32 (OpNode (f32 FPR32:$Rd), (f32 FPR32:$Rn),
4300 (vector_extract (v2f32 (fneg V64:$Rm)),
4301 VectorIndexS:$idx))),
4302 (FMLSv1i32_indexed FPR32:$Rd, FPR32:$Rn,
4303 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub), VectorIndexS:$idx)>;
4305 // 1 variant for 64-bit scalar version: extract from .1d or from .2d
4306 def : Pat<(f64 (OpNode (f64 FPR64:$Rd), (f64 FPR64:$Rn),
4307 (vector_extract (v2f64 (fneg V128:$Rm)),
4308 VectorIndexS:$idx))),
4309 (FMLSv1i64_indexed FPR64:$Rd, FPR64:$Rn,
4310 V128:$Rm, VectorIndexS:$idx)>;
4313 defm : FMLSIndexedAfterNegPatterns<
4314 TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)> >;
4315 defm : FMLSIndexedAfterNegPatterns<
4316 TriOpFrag<(fma node:$MHS, node:$RHS, node:$LHS)> >;
4318 defm FMULX : SIMDFPIndexedSD<1, 0b1001, "fmulx", int_aarch64_neon_fmulx>;
4319 defm FMUL : SIMDFPIndexedSD<0, 0b1001, "fmul", fmul>;
4321 def : Pat<(v2f32 (fmul V64:$Rn, (AArch64dup (f32 FPR32:$Rm)))),
4322 (FMULv2i32_indexed V64:$Rn,
4323 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rm, ssub),
4325 def : Pat<(v4f32 (fmul V128:$Rn, (AArch64dup (f32 FPR32:$Rm)))),
4326 (FMULv4i32_indexed V128:$Rn,
4327 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rm, ssub),
4329 def : Pat<(v2f64 (fmul V128:$Rn, (AArch64dup (f64 FPR64:$Rm)))),
4330 (FMULv2i64_indexed V128:$Rn,
4331 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$Rm, dsub),
4334 defm SQDMULH : SIMDIndexedHS<0, 0b1100, "sqdmulh", int_aarch64_neon_sqdmulh>;
4335 defm SQRDMULH : SIMDIndexedHS<0, 0b1101, "sqrdmulh", int_aarch64_neon_sqrdmulh>;
4336 defm MLA : SIMDVectorIndexedHSTied<1, 0b0000, "mla",
4337 TriOpFrag<(add node:$LHS, (mul node:$MHS, node:$RHS))>>;
4338 defm MLS : SIMDVectorIndexedHSTied<1, 0b0100, "mls",
4339 TriOpFrag<(sub node:$LHS, (mul node:$MHS, node:$RHS))>>;
4340 defm MUL : SIMDVectorIndexedHS<0, 0b1000, "mul", mul>;
4341 defm SMLAL : SIMDVectorIndexedLongSDTied<0, 0b0010, "smlal",
4342 TriOpFrag<(add node:$LHS, (int_aarch64_neon_smull node:$MHS, node:$RHS))>>;
4343 defm SMLSL : SIMDVectorIndexedLongSDTied<0, 0b0110, "smlsl",
4344 TriOpFrag<(sub node:$LHS, (int_aarch64_neon_smull node:$MHS, node:$RHS))>>;
4345 defm SMULL : SIMDVectorIndexedLongSD<0, 0b1010, "smull",
4346 int_aarch64_neon_smull>;
4347 defm SQDMLAL : SIMDIndexedLongSQDMLXSDTied<0, 0b0011, "sqdmlal",
4348 int_aarch64_neon_sqadd>;
4349 defm SQDMLSL : SIMDIndexedLongSQDMLXSDTied<0, 0b0111, "sqdmlsl",
4350 int_aarch64_neon_sqsub>;
4351 defm SQDMULL : SIMDIndexedLongSD<0, 0b1011, "sqdmull", int_aarch64_neon_sqdmull>;
4352 defm UMLAL : SIMDVectorIndexedLongSDTied<1, 0b0010, "umlal",
4353 TriOpFrag<(add node:$LHS, (int_aarch64_neon_umull node:$MHS, node:$RHS))>>;
4354 defm UMLSL : SIMDVectorIndexedLongSDTied<1, 0b0110, "umlsl",
4355 TriOpFrag<(sub node:$LHS, (int_aarch64_neon_umull node:$MHS, node:$RHS))>>;
4356 defm UMULL : SIMDVectorIndexedLongSD<1, 0b1010, "umull",
4357 int_aarch64_neon_umull>;
4359 // A scalar sqdmull with the second operand being a vector lane can be
4360 // handled directly with the indexed instruction encoding.
4361 def : Pat<(int_aarch64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
4362 (vector_extract (v4i32 V128:$Vm),
4363 VectorIndexS:$idx)),
4364 (SQDMULLv1i64_indexed FPR32:$Rn, V128:$Vm, VectorIndexS:$idx)>;
4366 //----------------------------------------------------------------------------
4367 // AdvSIMD scalar shift instructions
4368 //----------------------------------------------------------------------------
4369 defm FCVTZS : SIMDScalarRShiftSD<0, 0b11111, "fcvtzs">;
4370 defm FCVTZU : SIMDScalarRShiftSD<1, 0b11111, "fcvtzu">;
4371 defm SCVTF : SIMDScalarRShiftSD<0, 0b11100, "scvtf">;
4372 defm UCVTF : SIMDScalarRShiftSD<1, 0b11100, "ucvtf">;
4373 // Codegen patterns for the above. We don't put these directly on the
4374 // instructions because TableGen's type inference can't handle the truth.
4375 // Having the same base pattern for fp <--> int totally freaks it out.
4376 def : Pat<(int_aarch64_neon_vcvtfp2fxs FPR32:$Rn, vecshiftR32:$imm),
4377 (FCVTZSs FPR32:$Rn, vecshiftR32:$imm)>;
4378 def : Pat<(int_aarch64_neon_vcvtfp2fxu FPR32:$Rn, vecshiftR32:$imm),
4379 (FCVTZUs FPR32:$Rn, vecshiftR32:$imm)>;
4380 def : Pat<(i64 (int_aarch64_neon_vcvtfp2fxs (f64 FPR64:$Rn), vecshiftR64:$imm)),
4381 (FCVTZSd FPR64:$Rn, vecshiftR64:$imm)>;
4382 def : Pat<(i64 (int_aarch64_neon_vcvtfp2fxu (f64 FPR64:$Rn), vecshiftR64:$imm)),
4383 (FCVTZUd FPR64:$Rn, vecshiftR64:$imm)>;
4384 def : Pat<(v1i64 (int_aarch64_neon_vcvtfp2fxs (v1f64 FPR64:$Rn),
4386 (FCVTZSd FPR64:$Rn, vecshiftR64:$imm)>;
4387 def : Pat<(v1i64 (int_aarch64_neon_vcvtfp2fxu (v1f64 FPR64:$Rn),
4389 (FCVTZUd FPR64:$Rn, vecshiftR64:$imm)>;
4390 def : Pat<(int_aarch64_neon_vcvtfxs2fp FPR32:$Rn, vecshiftR32:$imm),
4391 (SCVTFs FPR32:$Rn, vecshiftR32:$imm)>;
4392 def : Pat<(int_aarch64_neon_vcvtfxu2fp FPR32:$Rn, vecshiftR32:$imm),
4393 (UCVTFs FPR32:$Rn, vecshiftR32:$imm)>;
4394 def : Pat<(f64 (int_aarch64_neon_vcvtfxs2fp (i64 FPR64:$Rn), vecshiftR64:$imm)),
4395 (SCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
4396 def : Pat<(f64 (int_aarch64_neon_vcvtfxu2fp (i64 FPR64:$Rn), vecshiftR64:$imm)),
4397 (UCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
4398 def : Pat<(v1f64 (int_aarch64_neon_vcvtfxs2fp (v1i64 FPR64:$Rn),
4400 (SCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
4401 def : Pat<(v1f64 (int_aarch64_neon_vcvtfxu2fp (v1i64 FPR64:$Rn),
4403 (UCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
4405 defm SHL : SIMDScalarLShiftD< 0, 0b01010, "shl", AArch64vshl>;
4406 defm SLI : SIMDScalarLShiftDTied<1, 0b01010, "sli">;
4407 defm SQRSHRN : SIMDScalarRShiftBHS< 0, 0b10011, "sqrshrn",
4408 int_aarch64_neon_sqrshrn>;
4409 defm SQRSHRUN : SIMDScalarRShiftBHS< 1, 0b10001, "sqrshrun",
4410 int_aarch64_neon_sqrshrun>;
4411 defm SQSHLU : SIMDScalarLShiftBHSD<1, 0b01100, "sqshlu", AArch64sqshlui>;
4412 defm SQSHL : SIMDScalarLShiftBHSD<0, 0b01110, "sqshl", AArch64sqshli>;
4413 defm SQSHRN : SIMDScalarRShiftBHS< 0, 0b10010, "sqshrn",
4414 int_aarch64_neon_sqshrn>;
4415 defm SQSHRUN : SIMDScalarRShiftBHS< 1, 0b10000, "sqshrun",
4416 int_aarch64_neon_sqshrun>;
4417 defm SRI : SIMDScalarRShiftDTied< 1, 0b01000, "sri">;
4418 defm SRSHR : SIMDScalarRShiftD< 0, 0b00100, "srshr", AArch64srshri>;
4419 defm SRSRA : SIMDScalarRShiftDTied< 0, 0b00110, "srsra",
4420 TriOpFrag<(add node:$LHS,
4421 (AArch64srshri node:$MHS, node:$RHS))>>;
4422 defm SSHR : SIMDScalarRShiftD< 0, 0b00000, "sshr", AArch64vashr>;
4423 defm SSRA : SIMDScalarRShiftDTied< 0, 0b00010, "ssra",
4424 TriOpFrag<(add node:$LHS,
4425 (AArch64vashr node:$MHS, node:$RHS))>>;
4426 defm UQRSHRN : SIMDScalarRShiftBHS< 1, 0b10011, "uqrshrn",
4427 int_aarch64_neon_uqrshrn>;
4428 defm UQSHL : SIMDScalarLShiftBHSD<1, 0b01110, "uqshl", AArch64uqshli>;
4429 defm UQSHRN : SIMDScalarRShiftBHS< 1, 0b10010, "uqshrn",
4430 int_aarch64_neon_uqshrn>;
4431 defm URSHR : SIMDScalarRShiftD< 1, 0b00100, "urshr", AArch64urshri>;
4432 defm URSRA : SIMDScalarRShiftDTied< 1, 0b00110, "ursra",
4433 TriOpFrag<(add node:$LHS,
4434 (AArch64urshri node:$MHS, node:$RHS))>>;
4435 defm USHR : SIMDScalarRShiftD< 1, 0b00000, "ushr", AArch64vlshr>;
4436 defm USRA : SIMDScalarRShiftDTied< 1, 0b00010, "usra",
4437 TriOpFrag<(add node:$LHS,
4438 (AArch64vlshr node:$MHS, node:$RHS))>>;
4440 //----------------------------------------------------------------------------
4441 // AdvSIMD vector shift instructions
4442 //----------------------------------------------------------------------------
4443 defm FCVTZS:SIMDVectorRShiftSD<0, 0b11111, "fcvtzs", int_aarch64_neon_vcvtfp2fxs>;
4444 defm FCVTZU:SIMDVectorRShiftSD<1, 0b11111, "fcvtzu", int_aarch64_neon_vcvtfp2fxu>;
4445 defm SCVTF: SIMDVectorRShiftSDToFP<0, 0b11100, "scvtf",
4446 int_aarch64_neon_vcvtfxs2fp>;
4447 defm RSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10001, "rshrn",
4448 int_aarch64_neon_rshrn>;
4449 defm SHL : SIMDVectorLShiftBHSD<0, 0b01010, "shl", AArch64vshl>;
4450 defm SHRN : SIMDVectorRShiftNarrowBHS<0, 0b10000, "shrn",
4451 BinOpFrag<(trunc (AArch64vashr node:$LHS, node:$RHS))>>;
4452 defm SLI : SIMDVectorLShiftBHSDTied<1, 0b01010, "sli", int_aarch64_neon_vsli>;
4453 def : Pat<(v1i64 (int_aarch64_neon_vsli (v1i64 FPR64:$Rd), (v1i64 FPR64:$Rn),
4454 (i32 vecshiftL64:$imm))),
4455 (SLId FPR64:$Rd, FPR64:$Rn, vecshiftL64:$imm)>;
4456 defm SQRSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10011, "sqrshrn",
4457 int_aarch64_neon_sqrshrn>;
4458 defm SQRSHRUN: SIMDVectorRShiftNarrowBHS<1, 0b10001, "sqrshrun",
4459 int_aarch64_neon_sqrshrun>;
4460 defm SQSHLU : SIMDVectorLShiftBHSD<1, 0b01100, "sqshlu", AArch64sqshlui>;
4461 defm SQSHL : SIMDVectorLShiftBHSD<0, 0b01110, "sqshl", AArch64sqshli>;
4462 defm SQSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10010, "sqshrn",
4463 int_aarch64_neon_sqshrn>;
4464 defm SQSHRUN : SIMDVectorRShiftNarrowBHS<1, 0b10000, "sqshrun",
4465 int_aarch64_neon_sqshrun>;
4466 defm SRI : SIMDVectorRShiftBHSDTied<1, 0b01000, "sri", int_aarch64_neon_vsri>;
4467 def : Pat<(v1i64 (int_aarch64_neon_vsri (v1i64 FPR64:$Rd), (v1i64 FPR64:$Rn),
4468 (i32 vecshiftR64:$imm))),
4469 (SRId FPR64:$Rd, FPR64:$Rn, vecshiftR64:$imm)>;
4470 defm SRSHR : SIMDVectorRShiftBHSD<0, 0b00100, "srshr", AArch64srshri>;
4471 defm SRSRA : SIMDVectorRShiftBHSDTied<0, 0b00110, "srsra",
4472 TriOpFrag<(add node:$LHS,
4473 (AArch64srshri node:$MHS, node:$RHS))> >;
4474 defm SSHLL : SIMDVectorLShiftLongBHSD<0, 0b10100, "sshll",
4475 BinOpFrag<(AArch64vshl (sext node:$LHS), node:$RHS)>>;
4477 defm SSHR : SIMDVectorRShiftBHSD<0, 0b00000, "sshr", AArch64vashr>;
4478 defm SSRA : SIMDVectorRShiftBHSDTied<0, 0b00010, "ssra",
4479 TriOpFrag<(add node:$LHS, (AArch64vashr node:$MHS, node:$RHS))>>;
4480 defm UCVTF : SIMDVectorRShiftSDToFP<1, 0b11100, "ucvtf",
4481 int_aarch64_neon_vcvtfxu2fp>;
4482 defm UQRSHRN : SIMDVectorRShiftNarrowBHS<1, 0b10011, "uqrshrn",
4483 int_aarch64_neon_uqrshrn>;
4484 defm UQSHL : SIMDVectorLShiftBHSD<1, 0b01110, "uqshl", AArch64uqshli>;
4485 defm UQSHRN : SIMDVectorRShiftNarrowBHS<1, 0b10010, "uqshrn",
4486 int_aarch64_neon_uqshrn>;
4487 defm URSHR : SIMDVectorRShiftBHSD<1, 0b00100, "urshr", AArch64urshri>;
4488 defm URSRA : SIMDVectorRShiftBHSDTied<1, 0b00110, "ursra",
4489 TriOpFrag<(add node:$LHS,
4490 (AArch64urshri node:$MHS, node:$RHS))> >;
4491 defm USHLL : SIMDVectorLShiftLongBHSD<1, 0b10100, "ushll",
4492 BinOpFrag<(AArch64vshl (zext node:$LHS), node:$RHS)>>;
4493 defm USHR : SIMDVectorRShiftBHSD<1, 0b00000, "ushr", AArch64vlshr>;
4494 defm USRA : SIMDVectorRShiftBHSDTied<1, 0b00010, "usra",
4495 TriOpFrag<(add node:$LHS, (AArch64vlshr node:$MHS, node:$RHS))> >;
4497 // SHRN patterns for when a logical right shift was used instead of arithmetic
4498 // (the immediate guarantees no sign bits actually end up in the result so it
4500 def : Pat<(v8i8 (trunc (AArch64vlshr (v8i16 V128:$Rn), vecshiftR16Narrow:$imm))),
4501 (SHRNv8i8_shift V128:$Rn, vecshiftR16Narrow:$imm)>;
4502 def : Pat<(v4i16 (trunc (AArch64vlshr (v4i32 V128:$Rn), vecshiftR32Narrow:$imm))),
4503 (SHRNv4i16_shift V128:$Rn, vecshiftR32Narrow:$imm)>;
4504 def : Pat<(v2i32 (trunc (AArch64vlshr (v2i64 V128:$Rn), vecshiftR64Narrow:$imm))),
4505 (SHRNv2i32_shift V128:$Rn, vecshiftR64Narrow:$imm)>;
4507 def : Pat<(v16i8 (concat_vectors (v8i8 V64:$Rd),
4508 (trunc (AArch64vlshr (v8i16 V128:$Rn),
4509 vecshiftR16Narrow:$imm)))),
4510 (SHRNv16i8_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
4511 V128:$Rn, vecshiftR16Narrow:$imm)>;
4512 def : Pat<(v8i16 (concat_vectors (v4i16 V64:$Rd),
4513 (trunc (AArch64vlshr (v4i32 V128:$Rn),
4514 vecshiftR32Narrow:$imm)))),
4515 (SHRNv8i16_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
4516 V128:$Rn, vecshiftR32Narrow:$imm)>;
4517 def : Pat<(v4i32 (concat_vectors (v2i32 V64:$Rd),
4518 (trunc (AArch64vlshr (v2i64 V128:$Rn),
4519 vecshiftR64Narrow:$imm)))),
4520 (SHRNv4i32_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
4521 V128:$Rn, vecshiftR32Narrow:$imm)>;
4523 // Vector sign and zero extensions are implemented with SSHLL and USSHLL.
4524 // Anyexts are implemented as zexts.
4525 def : Pat<(v8i16 (sext (v8i8 V64:$Rn))), (SSHLLv8i8_shift V64:$Rn, (i32 0))>;
4526 def : Pat<(v8i16 (zext (v8i8 V64:$Rn))), (USHLLv8i8_shift V64:$Rn, (i32 0))>;
4527 def : Pat<(v8i16 (anyext (v8i8 V64:$Rn))), (USHLLv8i8_shift V64:$Rn, (i32 0))>;
4528 def : Pat<(v4i32 (sext (v4i16 V64:$Rn))), (SSHLLv4i16_shift V64:$Rn, (i32 0))>;
4529 def : Pat<(v4i32 (zext (v4i16 V64:$Rn))), (USHLLv4i16_shift V64:$Rn, (i32 0))>;
4530 def : Pat<(v4i32 (anyext (v4i16 V64:$Rn))), (USHLLv4i16_shift V64:$Rn, (i32 0))>;
4531 def : Pat<(v2i64 (sext (v2i32 V64:$Rn))), (SSHLLv2i32_shift V64:$Rn, (i32 0))>;
4532 def : Pat<(v2i64 (zext (v2i32 V64:$Rn))), (USHLLv2i32_shift V64:$Rn, (i32 0))>;
4533 def : Pat<(v2i64 (anyext (v2i32 V64:$Rn))), (USHLLv2i32_shift V64:$Rn, (i32 0))>;
4534 // Also match an extend from the upper half of a 128 bit source register.
4535 def : Pat<(v8i16 (anyext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
4536 (USHLLv16i8_shift V128:$Rn, (i32 0))>;
4537 def : Pat<(v8i16 (zext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
4538 (USHLLv16i8_shift V128:$Rn, (i32 0))>;
4539 def : Pat<(v8i16 (sext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
4540 (SSHLLv16i8_shift V128:$Rn, (i32 0))>;
4541 def : Pat<(v4i32 (anyext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
4542 (USHLLv8i16_shift V128:$Rn, (i32 0))>;
4543 def : Pat<(v4i32 (zext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
4544 (USHLLv8i16_shift V128:$Rn, (i32 0))>;
4545 def : Pat<(v4i32 (sext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
4546 (SSHLLv8i16_shift V128:$Rn, (i32 0))>;
4547 def : Pat<(v2i64 (anyext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
4548 (USHLLv4i32_shift V128:$Rn, (i32 0))>;
4549 def : Pat<(v2i64 (zext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
4550 (USHLLv4i32_shift V128:$Rn, (i32 0))>;
4551 def : Pat<(v2i64 (sext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
4552 (SSHLLv4i32_shift V128:$Rn, (i32 0))>;
4554 // Vector shift sxtl aliases
4555 def : InstAlias<"sxtl.8h $dst, $src1",
4556 (SSHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
4557 def : InstAlias<"sxtl $dst.8h, $src1.8b",
4558 (SSHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
4559 def : InstAlias<"sxtl.4s $dst, $src1",
4560 (SSHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
4561 def : InstAlias<"sxtl $dst.4s, $src1.4h",
4562 (SSHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
4563 def : InstAlias<"sxtl.2d $dst, $src1",
4564 (SSHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
4565 def : InstAlias<"sxtl $dst.2d, $src1.2s",
4566 (SSHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
4568 // Vector shift sxtl2 aliases
4569 def : InstAlias<"sxtl2.8h $dst, $src1",
4570 (SSHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
4571 def : InstAlias<"sxtl2 $dst.8h, $src1.16b",
4572 (SSHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
4573 def : InstAlias<"sxtl2.4s $dst, $src1",
4574 (SSHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
4575 def : InstAlias<"sxtl2 $dst.4s, $src1.8h",
4576 (SSHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
4577 def : InstAlias<"sxtl2.2d $dst, $src1",
4578 (SSHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
4579 def : InstAlias<"sxtl2 $dst.2d, $src1.4s",
4580 (SSHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
4582 // Vector shift uxtl aliases
4583 def : InstAlias<"uxtl.8h $dst, $src1",
4584 (USHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
4585 def : InstAlias<"uxtl $dst.8h, $src1.8b",
4586 (USHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
4587 def : InstAlias<"uxtl.4s $dst, $src1",
4588 (USHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
4589 def : InstAlias<"uxtl $dst.4s, $src1.4h",
4590 (USHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
4591 def : InstAlias<"uxtl.2d $dst, $src1",
4592 (USHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
4593 def : InstAlias<"uxtl $dst.2d, $src1.2s",
4594 (USHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
4596 // Vector shift uxtl2 aliases
4597 def : InstAlias<"uxtl2.8h $dst, $src1",
4598 (USHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
4599 def : InstAlias<"uxtl2 $dst.8h, $src1.16b",
4600 (USHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
4601 def : InstAlias<"uxtl2.4s $dst, $src1",
4602 (USHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
4603 def : InstAlias<"uxtl2 $dst.4s, $src1.8h",
4604 (USHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
4605 def : InstAlias<"uxtl2.2d $dst, $src1",
4606 (USHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
4607 def : InstAlias<"uxtl2 $dst.2d, $src1.4s",
4608 (USHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
4610 // If an integer is about to be converted to a floating point value,
4611 // just load it on the floating point unit.
4612 // These patterns are more complex because floating point loads do not
4613 // support sign extension.
4614 // The sign extension has to be explicitly added and is only supported for
4615 // one step: byte-to-half, half-to-word, word-to-doubleword.
4616 // SCVTF GPR -> FPR is 9 cycles.
4617 // SCVTF FPR -> FPR is 4 cyclces.
4618 // (sign extension with lengthen) SXTL FPR -> FPR is 2 cycles.
4619 // Therefore, we can do 2 sign extensions and one SCVTF FPR -> FPR
4620 // and still being faster.
4621 // However, this is not good for code size.
4622 // 8-bits -> float. 2 sizes step-up.
4623 class SExtLoadi8CVTf32Pat<dag addrmode, dag INST>
4624 : Pat<(f32 (sint_to_fp (i32 (sextloadi8 addrmode)))),
4625 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
4630 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4636 ssub)))>, Requires<[NotForCodeSize, IsCyclone]>;
4638 def : SExtLoadi8CVTf32Pat<(ro8.Wpat GPR64sp:$Rn, GPR32:$Rm, ro8.Wext:$ext),
4639 (LDRBroW GPR64sp:$Rn, GPR32:$Rm, ro8.Wext:$ext)>;
4640 def : SExtLoadi8CVTf32Pat<(ro8.Xpat GPR64sp:$Rn, GPR64:$Rm, ro8.Xext:$ext),
4641 (LDRBroX GPR64sp:$Rn, GPR64:$Rm, ro8.Xext:$ext)>;
4642 def : SExtLoadi8CVTf32Pat<(am_indexed8 GPR64sp:$Rn, uimm12s1:$offset),
4643 (LDRBui GPR64sp:$Rn, uimm12s1:$offset)>;
4644 def : SExtLoadi8CVTf32Pat<(am_unscaled8 GPR64sp:$Rn, simm9:$offset),
4645 (LDURBi GPR64sp:$Rn, simm9:$offset)>;
4647 // 16-bits -> float. 1 size step-up.
4648 class SExtLoadi16CVTf32Pat<dag addrmode, dag INST>
4649 : Pat<(f32 (sint_to_fp (i32 (sextloadi16 addrmode)))),
4650 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
4652 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4656 ssub)))>, Requires<[NotForCodeSize]>;
4658 def : SExtLoadi16CVTf32Pat<(ro16.Wpat GPR64sp:$Rn, GPR32:$Rm, ro16.Wext:$ext),
4659 (LDRHroW GPR64sp:$Rn, GPR32:$Rm, ro16.Wext:$ext)>;
4660 def : SExtLoadi16CVTf32Pat<(ro16.Xpat GPR64sp:$Rn, GPR64:$Rm, ro16.Xext:$ext),
4661 (LDRHroX GPR64sp:$Rn, GPR64:$Rm, ro16.Xext:$ext)>;
4662 def : SExtLoadi16CVTf32Pat<(am_indexed16 GPR64sp:$Rn, uimm12s2:$offset),
4663 (LDRHui GPR64sp:$Rn, uimm12s2:$offset)>;
4664 def : SExtLoadi16CVTf32Pat<(am_unscaled16 GPR64sp:$Rn, simm9:$offset),
4665 (LDURHi GPR64sp:$Rn, simm9:$offset)>;
4667 // 32-bits to 32-bits are handled in target specific dag combine:
4668 // performIntToFpCombine.
4669 // 64-bits integer to 32-bits floating point, not possible with
4670 // SCVTF on floating point registers (both source and destination
4671 // must have the same size).
4673 // Here are the patterns for 8, 16, 32, and 64-bits to double.
4674 // 8-bits -> double. 3 size step-up: give up.
4675 // 16-bits -> double. 2 size step.
4676 class SExtLoadi16CVTf64Pat<dag addrmode, dag INST>
4677 : Pat <(f64 (sint_to_fp (i32 (sextloadi16 addrmode)))),
4678 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4683 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4689 dsub)))>, Requires<[NotForCodeSize, IsCyclone]>;
4691 def : SExtLoadi16CVTf64Pat<(ro16.Wpat GPR64sp:$Rn, GPR32:$Rm, ro16.Wext:$ext),
4692 (LDRHroW GPR64sp:$Rn, GPR32:$Rm, ro16.Wext:$ext)>;
4693 def : SExtLoadi16CVTf64Pat<(ro16.Xpat GPR64sp:$Rn, GPR64:$Rm, ro16.Xext:$ext),
4694 (LDRHroX GPR64sp:$Rn, GPR64:$Rm, ro16.Xext:$ext)>;
4695 def : SExtLoadi16CVTf64Pat<(am_indexed16 GPR64sp:$Rn, uimm12s2:$offset),
4696 (LDRHui GPR64sp:$Rn, uimm12s2:$offset)>;
4697 def : SExtLoadi16CVTf64Pat<(am_unscaled16 GPR64sp:$Rn, simm9:$offset),
4698 (LDURHi GPR64sp:$Rn, simm9:$offset)>;
4699 // 32-bits -> double. 1 size step-up.
4700 class SExtLoadi32CVTf64Pat<dag addrmode, dag INST>
4701 : Pat <(f64 (sint_to_fp (i32 (load addrmode)))),
4702 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4704 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4708 dsub)))>, Requires<[NotForCodeSize]>;
4710 def : SExtLoadi32CVTf64Pat<(ro32.Wpat GPR64sp:$Rn, GPR32:$Rm, ro32.Wext:$ext),
4711 (LDRSroW GPR64sp:$Rn, GPR32:$Rm, ro32.Wext:$ext)>;
4712 def : SExtLoadi32CVTf64Pat<(ro32.Xpat GPR64sp:$Rn, GPR64:$Rm, ro32.Xext:$ext),
4713 (LDRSroX GPR64sp:$Rn, GPR64:$Rm, ro32.Xext:$ext)>;
4714 def : SExtLoadi32CVTf64Pat<(am_indexed32 GPR64sp:$Rn, uimm12s4:$offset),
4715 (LDRSui GPR64sp:$Rn, uimm12s4:$offset)>;
4716 def : SExtLoadi32CVTf64Pat<(am_unscaled32 GPR64sp:$Rn, simm9:$offset),
4717 (LDURSi GPR64sp:$Rn, simm9:$offset)>;
4719 // 64-bits -> double are handled in target specific dag combine:
4720 // performIntToFpCombine.
4723 //----------------------------------------------------------------------------
4724 // AdvSIMD Load-Store Structure
4725 //----------------------------------------------------------------------------
4726 defm LD1 : SIMDLd1Multiple<"ld1">;
4727 defm LD2 : SIMDLd2Multiple<"ld2">;
4728 defm LD3 : SIMDLd3Multiple<"ld3">;
4729 defm LD4 : SIMDLd4Multiple<"ld4">;
4731 defm ST1 : SIMDSt1Multiple<"st1">;
4732 defm ST2 : SIMDSt2Multiple<"st2">;
4733 defm ST3 : SIMDSt3Multiple<"st3">;
4734 defm ST4 : SIMDSt4Multiple<"st4">;
4736 class Ld1Pat<ValueType ty, Instruction INST>
4737 : Pat<(ty (load GPR64sp:$Rn)), (INST GPR64sp:$Rn)>;
4739 def : Ld1Pat<v16i8, LD1Onev16b>;
4740 def : Ld1Pat<v8i16, LD1Onev8h>;
4741 def : Ld1Pat<v4i32, LD1Onev4s>;
4742 def : Ld1Pat<v2i64, LD1Onev2d>;
4743 def : Ld1Pat<v8i8, LD1Onev8b>;
4744 def : Ld1Pat<v4i16, LD1Onev4h>;
4745 def : Ld1Pat<v2i32, LD1Onev2s>;
4746 def : Ld1Pat<v1i64, LD1Onev1d>;
4748 class St1Pat<ValueType ty, Instruction INST>
4749 : Pat<(store ty:$Vt, GPR64sp:$Rn),
4750 (INST ty:$Vt, GPR64sp:$Rn)>;
4752 def : St1Pat<v16i8, ST1Onev16b>;
4753 def : St1Pat<v8i16, ST1Onev8h>;
4754 def : St1Pat<v4i32, ST1Onev4s>;
4755 def : St1Pat<v2i64, ST1Onev2d>;
4756 def : St1Pat<v8i8, ST1Onev8b>;
4757 def : St1Pat<v4i16, ST1Onev4h>;
4758 def : St1Pat<v2i32, ST1Onev2s>;
4759 def : St1Pat<v1i64, ST1Onev1d>;
4765 defm LD1R : SIMDLdR<0, 0b110, 0, "ld1r", "One", 1, 2, 4, 8>;
4766 defm LD2R : SIMDLdR<1, 0b110, 0, "ld2r", "Two", 2, 4, 8, 16>;
4767 defm LD3R : SIMDLdR<0, 0b111, 0, "ld3r", "Three", 3, 6, 12, 24>;
4768 defm LD4R : SIMDLdR<1, 0b111, 0, "ld4r", "Four", 4, 8, 16, 32>;
4769 let mayLoad = 1, hasSideEffects = 0 in {
4770 defm LD1 : SIMDLdSingleBTied<0, 0b000, "ld1", VecListOneb, GPR64pi1>;
4771 defm LD1 : SIMDLdSingleHTied<0, 0b010, 0, "ld1", VecListOneh, GPR64pi2>;
4772 defm LD1 : SIMDLdSingleSTied<0, 0b100, 0b00, "ld1", VecListOnes, GPR64pi4>;
4773 defm LD1 : SIMDLdSingleDTied<0, 0b100, 0b01, "ld1", VecListOned, GPR64pi8>;
4774 defm LD2 : SIMDLdSingleBTied<1, 0b000, "ld2", VecListTwob, GPR64pi2>;
4775 defm LD2 : SIMDLdSingleHTied<1, 0b010, 0, "ld2", VecListTwoh, GPR64pi4>;
4776 defm LD2 : SIMDLdSingleSTied<1, 0b100, 0b00, "ld2", VecListTwos, GPR64pi8>;
4777 defm LD2 : SIMDLdSingleDTied<1, 0b100, 0b01, "ld2", VecListTwod, GPR64pi16>;
4778 defm LD3 : SIMDLdSingleBTied<0, 0b001, "ld3", VecListThreeb, GPR64pi3>;
4779 defm LD3 : SIMDLdSingleHTied<0, 0b011, 0, "ld3", VecListThreeh, GPR64pi6>;
4780 defm LD3 : SIMDLdSingleSTied<0, 0b101, 0b00, "ld3", VecListThrees, GPR64pi12>;
4781 defm LD3 : SIMDLdSingleDTied<0, 0b101, 0b01, "ld3", VecListThreed, GPR64pi24>;
4782 defm LD4 : SIMDLdSingleBTied<1, 0b001, "ld4", VecListFourb, GPR64pi4>;
4783 defm LD4 : SIMDLdSingleHTied<1, 0b011, 0, "ld4", VecListFourh, GPR64pi8>;
4784 defm LD4 : SIMDLdSingleSTied<1, 0b101, 0b00, "ld4", VecListFours, GPR64pi16>;
4785 defm LD4 : SIMDLdSingleDTied<1, 0b101, 0b01, "ld4", VecListFourd, GPR64pi32>;
4788 def : Pat<(v8i8 (AArch64dup (i32 (extloadi8 GPR64sp:$Rn)))),
4789 (LD1Rv8b GPR64sp:$Rn)>;
4790 def : Pat<(v16i8 (AArch64dup (i32 (extloadi8 GPR64sp:$Rn)))),
4791 (LD1Rv16b GPR64sp:$Rn)>;
4792 def : Pat<(v4i16 (AArch64dup (i32 (extloadi16 GPR64sp:$Rn)))),
4793 (LD1Rv4h GPR64sp:$Rn)>;
4794 def : Pat<(v8i16 (AArch64dup (i32 (extloadi16 GPR64sp:$Rn)))),
4795 (LD1Rv8h GPR64sp:$Rn)>;
4796 def : Pat<(v2i32 (AArch64dup (i32 (load GPR64sp:$Rn)))),
4797 (LD1Rv2s GPR64sp:$Rn)>;
4798 def : Pat<(v4i32 (AArch64dup (i32 (load GPR64sp:$Rn)))),
4799 (LD1Rv4s GPR64sp:$Rn)>;
4800 def : Pat<(v2i64 (AArch64dup (i64 (load GPR64sp:$Rn)))),
4801 (LD1Rv2d GPR64sp:$Rn)>;
4802 def : Pat<(v1i64 (AArch64dup (i64 (load GPR64sp:$Rn)))),
4803 (LD1Rv1d GPR64sp:$Rn)>;
4804 // Grab the floating point version too
4805 def : Pat<(v2f32 (AArch64dup (f32 (load GPR64sp:$Rn)))),
4806 (LD1Rv2s GPR64sp:$Rn)>;
4807 def : Pat<(v4f32 (AArch64dup (f32 (load GPR64sp:$Rn)))),
4808 (LD1Rv4s GPR64sp:$Rn)>;
4809 def : Pat<(v2f64 (AArch64dup (f64 (load GPR64sp:$Rn)))),
4810 (LD1Rv2d GPR64sp:$Rn)>;
4811 def : Pat<(v1f64 (AArch64dup (f64 (load GPR64sp:$Rn)))),
4812 (LD1Rv1d GPR64sp:$Rn)>;
4813 def : Pat<(v4f16 (AArch64dup (f16 (load GPR64sp:$Rn)))),
4814 (LD1Rv4h GPR64sp:$Rn)>;
4815 def : Pat<(v8f16 (AArch64dup (f16 (load GPR64sp:$Rn)))),
4816 (LD1Rv8h GPR64sp:$Rn)>;
4818 class Ld1Lane128Pat<SDPatternOperator scalar_load, Operand VecIndex,
4819 ValueType VTy, ValueType STy, Instruction LD1>
4820 : Pat<(vector_insert (VTy VecListOne128:$Rd),
4821 (STy (scalar_load GPR64sp:$Rn)), VecIndex:$idx),
4822 (LD1 VecListOne128:$Rd, VecIndex:$idx, GPR64sp:$Rn)>;
4824 def : Ld1Lane128Pat<extloadi8, VectorIndexB, v16i8, i32, LD1i8>;
4825 def : Ld1Lane128Pat<extloadi16, VectorIndexH, v8i16, i32, LD1i16>;
4826 def : Ld1Lane128Pat<load, VectorIndexS, v4i32, i32, LD1i32>;
4827 def : Ld1Lane128Pat<load, VectorIndexS, v4f32, f32, LD1i32>;
4828 def : Ld1Lane128Pat<load, VectorIndexD, v2i64, i64, LD1i64>;
4829 def : Ld1Lane128Pat<load, VectorIndexD, v2f64, f64, LD1i64>;
4830 def : Ld1Lane128Pat<load, VectorIndexH, v8f16, f16, LD1i16>;
4832 class Ld1Lane64Pat<SDPatternOperator scalar_load, Operand VecIndex,
4833 ValueType VTy, ValueType STy, Instruction LD1>
4834 : Pat<(vector_insert (VTy VecListOne64:$Rd),
4835 (STy (scalar_load GPR64sp:$Rn)), VecIndex:$idx),
4837 (LD1 (SUBREG_TO_REG (i32 0), VecListOne64:$Rd, dsub),
4838 VecIndex:$idx, GPR64sp:$Rn),
4841 def : Ld1Lane64Pat<extloadi8, VectorIndexB, v8i8, i32, LD1i8>;
4842 def : Ld1Lane64Pat<extloadi16, VectorIndexH, v4i16, i32, LD1i16>;
4843 def : Ld1Lane64Pat<load, VectorIndexS, v2i32, i32, LD1i32>;
4844 def : Ld1Lane64Pat<load, VectorIndexS, v2f32, f32, LD1i32>;
4845 def : Ld1Lane64Pat<load, VectorIndexH, v4f16, f16, LD1i16>;
4848 defm LD1 : SIMDLdSt1SingleAliases<"ld1">;
4849 defm LD2 : SIMDLdSt2SingleAliases<"ld2">;
4850 defm LD3 : SIMDLdSt3SingleAliases<"ld3">;
4851 defm LD4 : SIMDLdSt4SingleAliases<"ld4">;
4854 defm ST1 : SIMDStSingleB<0, 0b000, "st1", VecListOneb, GPR64pi1>;
4855 defm ST1 : SIMDStSingleH<0, 0b010, 0, "st1", VecListOneh, GPR64pi2>;
4856 defm ST1 : SIMDStSingleS<0, 0b100, 0b00, "st1", VecListOnes, GPR64pi4>;
4857 defm ST1 : SIMDStSingleD<0, 0b100, 0b01, "st1", VecListOned, GPR64pi8>;
4859 let AddedComplexity = 19 in
4860 class St1Lane128Pat<SDPatternOperator scalar_store, Operand VecIndex,
4861 ValueType VTy, ValueType STy, Instruction ST1>
4863 (STy (vector_extract (VTy VecListOne128:$Vt), VecIndex:$idx)),
4865 (ST1 VecListOne128:$Vt, VecIndex:$idx, GPR64sp:$Rn)>;
4867 def : St1Lane128Pat<truncstorei8, VectorIndexB, v16i8, i32, ST1i8>;
4868 def : St1Lane128Pat<truncstorei16, VectorIndexH, v8i16, i32, ST1i16>;
4869 def : St1Lane128Pat<store, VectorIndexS, v4i32, i32, ST1i32>;
4870 def : St1Lane128Pat<store, VectorIndexS, v4f32, f32, ST1i32>;
4871 def : St1Lane128Pat<store, VectorIndexD, v2i64, i64, ST1i64>;
4872 def : St1Lane128Pat<store, VectorIndexD, v2f64, f64, ST1i64>;
4873 def : St1Lane128Pat<store, VectorIndexH, v8f16, f16, ST1i16>;
4875 let AddedComplexity = 19 in
4876 class St1Lane64Pat<SDPatternOperator scalar_store, Operand VecIndex,
4877 ValueType VTy, ValueType STy, Instruction ST1>
4879 (STy (vector_extract (VTy VecListOne64:$Vt), VecIndex:$idx)),
4881 (ST1 (SUBREG_TO_REG (i32 0), VecListOne64:$Vt, dsub),
4882 VecIndex:$idx, GPR64sp:$Rn)>;
4884 def : St1Lane64Pat<truncstorei8, VectorIndexB, v8i8, i32, ST1i8>;
4885 def : St1Lane64Pat<truncstorei16, VectorIndexH, v4i16, i32, ST1i16>;
4886 def : St1Lane64Pat<store, VectorIndexS, v2i32, i32, ST1i32>;
4887 def : St1Lane64Pat<store, VectorIndexS, v2f32, f32, ST1i32>;
4888 def : St1Lane64Pat<store, VectorIndexH, v4f16, f16, ST1i16>;
4890 multiclass St1LanePost64Pat<SDPatternOperator scalar_store, Operand VecIndex,
4891 ValueType VTy, ValueType STy, Instruction ST1,
4893 def : Pat<(scalar_store
4894 (STy (vector_extract (VTy VecListOne64:$Vt), VecIndex:$idx)),
4895 GPR64sp:$Rn, offset),
4896 (ST1 (SUBREG_TO_REG (i32 0), VecListOne64:$Vt, dsub),
4897 VecIndex:$idx, GPR64sp:$Rn, XZR)>;
4899 def : Pat<(scalar_store
4900 (STy (vector_extract (VTy VecListOne64:$Vt), VecIndex:$idx)),
4901 GPR64sp:$Rn, GPR64:$Rm),
4902 (ST1 (SUBREG_TO_REG (i32 0), VecListOne64:$Vt, dsub),
4903 VecIndex:$idx, GPR64sp:$Rn, $Rm)>;
4906 defm : St1LanePost64Pat<post_truncsti8, VectorIndexB, v8i8, i32, ST1i8_POST, 1>;
4907 defm : St1LanePost64Pat<post_truncsti16, VectorIndexH, v4i16, i32, ST1i16_POST,
4909 defm : St1LanePost64Pat<post_store, VectorIndexS, v2i32, i32, ST1i32_POST, 4>;
4910 defm : St1LanePost64Pat<post_store, VectorIndexS, v2f32, f32, ST1i32_POST, 4>;
4911 defm : St1LanePost64Pat<post_store, VectorIndexD, v1i64, i64, ST1i64_POST, 8>;
4912 defm : St1LanePost64Pat<post_store, VectorIndexD, v1f64, f64, ST1i64_POST, 8>;
4913 defm : St1LanePost64Pat<post_store, VectorIndexH, v4f16, f16, ST1i16_POST, 2>;
4915 multiclass St1LanePost128Pat<SDPatternOperator scalar_store, Operand VecIndex,
4916 ValueType VTy, ValueType STy, Instruction ST1,
4918 def : Pat<(scalar_store
4919 (STy (vector_extract (VTy VecListOne128:$Vt), VecIndex:$idx)),
4920 GPR64sp:$Rn, offset),
4921 (ST1 VecListOne128:$Vt, VecIndex:$idx, GPR64sp:$Rn, XZR)>;
4923 def : Pat<(scalar_store
4924 (STy (vector_extract (VTy VecListOne128:$Vt), VecIndex:$idx)),
4925 GPR64sp:$Rn, GPR64:$Rm),
4926 (ST1 VecListOne128:$Vt, VecIndex:$idx, GPR64sp:$Rn, $Rm)>;
4929 defm : St1LanePost128Pat<post_truncsti8, VectorIndexB, v16i8, i32, ST1i8_POST,
4931 defm : St1LanePost128Pat<post_truncsti16, VectorIndexH, v8i16, i32, ST1i16_POST,
4933 defm : St1LanePost128Pat<post_store, VectorIndexS, v4i32, i32, ST1i32_POST, 4>;
4934 defm : St1LanePost128Pat<post_store, VectorIndexS, v4f32, f32, ST1i32_POST, 4>;
4935 defm : St1LanePost128Pat<post_store, VectorIndexD, v2i64, i64, ST1i64_POST, 8>;
4936 defm : St1LanePost128Pat<post_store, VectorIndexD, v2f64, f64, ST1i64_POST, 8>;
4937 defm : St1LanePost128Pat<post_store, VectorIndexH, v8f16, f16, ST1i16_POST, 2>;
4939 let mayStore = 1, hasSideEffects = 0 in {
4940 defm ST2 : SIMDStSingleB<1, 0b000, "st2", VecListTwob, GPR64pi2>;
4941 defm ST2 : SIMDStSingleH<1, 0b010, 0, "st2", VecListTwoh, GPR64pi4>;
4942 defm ST2 : SIMDStSingleS<1, 0b100, 0b00, "st2", VecListTwos, GPR64pi8>;
4943 defm ST2 : SIMDStSingleD<1, 0b100, 0b01, "st2", VecListTwod, GPR64pi16>;
4944 defm ST3 : SIMDStSingleB<0, 0b001, "st3", VecListThreeb, GPR64pi3>;
4945 defm ST3 : SIMDStSingleH<0, 0b011, 0, "st3", VecListThreeh, GPR64pi6>;
4946 defm ST3 : SIMDStSingleS<0, 0b101, 0b00, "st3", VecListThrees, GPR64pi12>;
4947 defm ST3 : SIMDStSingleD<0, 0b101, 0b01, "st3", VecListThreed, GPR64pi24>;
4948 defm ST4 : SIMDStSingleB<1, 0b001, "st4", VecListFourb, GPR64pi4>;
4949 defm ST4 : SIMDStSingleH<1, 0b011, 0, "st4", VecListFourh, GPR64pi8>;
4950 defm ST4 : SIMDStSingleS<1, 0b101, 0b00, "st4", VecListFours, GPR64pi16>;
4951 defm ST4 : SIMDStSingleD<1, 0b101, 0b01, "st4", VecListFourd, GPR64pi32>;
4954 defm ST1 : SIMDLdSt1SingleAliases<"st1">;
4955 defm ST2 : SIMDLdSt2SingleAliases<"st2">;
4956 defm ST3 : SIMDLdSt3SingleAliases<"st3">;
4957 defm ST4 : SIMDLdSt4SingleAliases<"st4">;
4959 //----------------------------------------------------------------------------
4960 // Crypto extensions
4961 //----------------------------------------------------------------------------
4963 def AESErr : AESTiedInst<0b0100, "aese", int_aarch64_crypto_aese>;
4964 def AESDrr : AESTiedInst<0b0101, "aesd", int_aarch64_crypto_aesd>;
4965 def AESMCrr : AESInst< 0b0110, "aesmc", int_aarch64_crypto_aesmc>;
4966 def AESIMCrr : AESInst< 0b0111, "aesimc", int_aarch64_crypto_aesimc>;
4968 def SHA1Crrr : SHATiedInstQSV<0b000, "sha1c", int_aarch64_crypto_sha1c>;
4969 def SHA1Prrr : SHATiedInstQSV<0b001, "sha1p", int_aarch64_crypto_sha1p>;
4970 def SHA1Mrrr : SHATiedInstQSV<0b010, "sha1m", int_aarch64_crypto_sha1m>;
4971 def SHA1SU0rrr : SHATiedInstVVV<0b011, "sha1su0", int_aarch64_crypto_sha1su0>;
4972 def SHA256Hrrr : SHATiedInstQQV<0b100, "sha256h", int_aarch64_crypto_sha256h>;
4973 def SHA256H2rrr : SHATiedInstQQV<0b101, "sha256h2",int_aarch64_crypto_sha256h2>;
4974 def SHA256SU1rrr :SHATiedInstVVV<0b110, "sha256su1",int_aarch64_crypto_sha256su1>;
4976 def SHA1Hrr : SHAInstSS< 0b0000, "sha1h", int_aarch64_crypto_sha1h>;
4977 def SHA1SU1rr : SHATiedInstVV<0b0001, "sha1su1", int_aarch64_crypto_sha1su1>;
4978 def SHA256SU0rr : SHATiedInstVV<0b0010, "sha256su0",int_aarch64_crypto_sha256su0>;
4980 //----------------------------------------------------------------------------
4982 //----------------------------------------------------------------------------
4983 // FIXME: Like for X86, these should go in their own separate .td file.
4985 // Any instruction that defines a 32-bit result leaves the high half of the
4986 // register. Truncate can be lowered to EXTRACT_SUBREG. CopyFromReg may
4987 // be copying from a truncate. But any other 32-bit operation will zero-extend
4989 // FIXME: X86 also checks for CMOV here. Do we need something similar?
4990 def def32 : PatLeaf<(i32 GPR32:$src), [{
4991 return N->getOpcode() != ISD::TRUNCATE &&
4992 N->getOpcode() != TargetOpcode::EXTRACT_SUBREG &&
4993 N->getOpcode() != ISD::CopyFromReg;
4996 // In the case of a 32-bit def that is known to implicitly zero-extend,
4997 // we can use a SUBREG_TO_REG.
4998 def : Pat<(i64 (zext def32:$src)), (SUBREG_TO_REG (i64 0), GPR32:$src, sub_32)>;
5000 // For an anyext, we don't care what the high bits are, so we can perform an
5001 // INSERT_SUBREF into an IMPLICIT_DEF.
5002 def : Pat<(i64 (anyext GPR32:$src)),
5003 (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32)>;
5005 // When we need to explicitly zero-extend, we use an unsigned bitfield move
5006 // instruction (UBFM) on the enclosing super-reg.
5007 def : Pat<(i64 (zext GPR32:$src)),
5008 (UBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32), 0, 31)>;
5010 // To sign extend, we use a signed bitfield move instruction (SBFM) on the
5011 // containing super-reg.
5012 def : Pat<(i64 (sext GPR32:$src)),
5013 (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32), 0, 31)>;
5014 def : Pat<(i64 (sext_inreg GPR64:$src, i32)), (SBFMXri GPR64:$src, 0, 31)>;
5015 def : Pat<(i64 (sext_inreg GPR64:$src, i16)), (SBFMXri GPR64:$src, 0, 15)>;
5016 def : Pat<(i64 (sext_inreg GPR64:$src, i8)), (SBFMXri GPR64:$src, 0, 7)>;
5017 def : Pat<(i64 (sext_inreg GPR64:$src, i1)), (SBFMXri GPR64:$src, 0, 0)>;
5018 def : Pat<(i32 (sext_inreg GPR32:$src, i16)), (SBFMWri GPR32:$src, 0, 15)>;
5019 def : Pat<(i32 (sext_inreg GPR32:$src, i8)), (SBFMWri GPR32:$src, 0, 7)>;
5020 def : Pat<(i32 (sext_inreg GPR32:$src, i1)), (SBFMWri GPR32:$src, 0, 0)>;
5022 def : Pat<(shl (sext_inreg GPR32:$Rn, i8), (i64 imm0_31:$imm)),
5023 (SBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
5024 (i64 (i32shift_sext_i8 imm0_31:$imm)))>;
5025 def : Pat<(shl (sext_inreg GPR64:$Rn, i8), (i64 imm0_63:$imm)),
5026 (SBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
5027 (i64 (i64shift_sext_i8 imm0_63:$imm)))>;
5029 def : Pat<(shl (sext_inreg GPR32:$Rn, i16), (i64 imm0_31:$imm)),
5030 (SBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
5031 (i64 (i32shift_sext_i16 imm0_31:$imm)))>;
5032 def : Pat<(shl (sext_inreg GPR64:$Rn, i16), (i64 imm0_63:$imm)),
5033 (SBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
5034 (i64 (i64shift_sext_i16 imm0_63:$imm)))>;
5036 def : Pat<(shl (i64 (sext GPR32:$Rn)), (i64 imm0_63:$imm)),
5037 (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$Rn, sub_32),
5038 (i64 (i64shift_a imm0_63:$imm)),
5039 (i64 (i64shift_sext_i32 imm0_63:$imm)))>;
5041 // sra patterns have an AddedComplexity of 10, so make sure we have a higher
5042 // AddedComplexity for the following patterns since we want to match sext + sra
5043 // patterns before we attempt to match a single sra node.
5044 let AddedComplexity = 20 in {
5045 // We support all sext + sra combinations which preserve at least one bit of the
5046 // original value which is to be sign extended. E.g. we support shifts up to
5048 def : Pat<(sra (sext_inreg GPR32:$Rn, i8), (i64 imm0_7:$imm)),
5049 (SBFMWri GPR32:$Rn, (i64 imm0_7:$imm), 7)>;
5050 def : Pat<(sra (sext_inreg GPR64:$Rn, i8), (i64 imm0_7:$imm)),
5051 (SBFMXri GPR64:$Rn, (i64 imm0_7:$imm), 7)>;
5053 def : Pat<(sra (sext_inreg GPR32:$Rn, i16), (i64 imm0_15:$imm)),
5054 (SBFMWri GPR32:$Rn, (i64 imm0_15:$imm), 15)>;
5055 def : Pat<(sra (sext_inreg GPR64:$Rn, i16), (i64 imm0_15:$imm)),
5056 (SBFMXri GPR64:$Rn, (i64 imm0_15:$imm), 15)>;
5058 def : Pat<(sra (i64 (sext GPR32:$Rn)), (i64 imm0_31:$imm)),
5059 (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$Rn, sub_32),
5060 (i64 imm0_31:$imm), 31)>;
5061 } // AddedComplexity = 20
5063 // To truncate, we can simply extract from a subregister.
5064 def : Pat<(i32 (trunc GPR64sp:$src)),
5065 (i32 (EXTRACT_SUBREG GPR64sp:$src, sub_32))>;
5067 // __builtin_trap() uses the BRK instruction on AArch64.
5068 def : Pat<(trap), (BRK 1)>;
5070 // Conversions within AdvSIMD types in the same register size are free.
5071 // But because we need a consistent lane ordering, in big endian many
5072 // conversions require one or more REV instructions.
5074 // Consider a simple memory load followed by a bitconvert then a store.
5076 // v1 = BITCAST v2i32 v0 to v4i16
5079 // In big endian mode every memory access has an implicit byte swap. LDR and
5080 // STR do a 64-bit byte swap, whereas LD1/ST1 do a byte swap per lane - that
5081 // is, they treat the vector as a sequence of elements to be byte-swapped.
5082 // The two pairs of instructions are fundamentally incompatible. We've decided
5083 // to use LD1/ST1 only to simplify compiler implementation.
5085 // LD1/ST1 perform the equivalent of a sequence of LDR/STR + REV. This makes
5086 // the original code sequence:
5088 // v1 = REV v2i32 (implicit)
5089 // v2 = BITCAST v2i32 v1 to v4i16
5090 // v3 = REV v4i16 v2 (implicit)
5093 // But this is now broken - the value stored is different to the value loaded
5094 // due to lane reordering. To fix this, on every BITCAST we must perform two
5097 // v1 = REV v2i32 (implicit)
5099 // v3 = BITCAST v2i32 v2 to v4i16
5101 // v5 = REV v4i16 v4 (implicit)
5104 // This means an extra two instructions, but actually in most cases the two REV
5105 // instructions can be combined into one. For example:
5106 // (REV64_2s (REV64_4h X)) === (REV32_4h X)
5108 // There is also no 128-bit REV instruction. This must be synthesized with an
5111 // Most bitconverts require some sort of conversion. The only exceptions are:
5112 // a) Identity conversions - vNfX <-> vNiX
5113 // b) Single-lane-to-scalar - v1fX <-> fX or v1iX <-> iX
5116 // Natural vector casts (64 bit)
5117 def : Pat<(v8i8 (AArch64NvCast (v2i32 FPR64:$src))), (v8i8 FPR64:$src)>;
5118 def : Pat<(v4i16 (AArch64NvCast (v2i32 FPR64:$src))), (v4i16 FPR64:$src)>;
5119 def : Pat<(v2i32 (AArch64NvCast (v2i32 FPR64:$src))), (v2i32 FPR64:$src)>;
5120 def : Pat<(v2f32 (AArch64NvCast (v2i32 FPR64:$src))), (v2f32 FPR64:$src)>;
5121 def : Pat<(v1i64 (AArch64NvCast (v2i32 FPR64:$src))), (v1i64 FPR64:$src)>;
5123 def : Pat<(v8i8 (AArch64NvCast (v4i16 FPR64:$src))), (v8i8 FPR64:$src)>;
5124 def : Pat<(v4i16 (AArch64NvCast (v4i16 FPR64:$src))), (v4i16 FPR64:$src)>;
5125 def : Pat<(v2i32 (AArch64NvCast (v4i16 FPR64:$src))), (v2i32 FPR64:$src)>;
5126 def : Pat<(v1i64 (AArch64NvCast (v4i16 FPR64:$src))), (v1i64 FPR64:$src)>;
5128 def : Pat<(v8i8 (AArch64NvCast (v8i8 FPR64:$src))), (v8i8 FPR64:$src)>;
5129 def : Pat<(v4i16 (AArch64NvCast (v8i8 FPR64:$src))), (v4i16 FPR64:$src)>;
5130 def : Pat<(v2i32 (AArch64NvCast (v8i8 FPR64:$src))), (v2i32 FPR64:$src)>;
5131 def : Pat<(v1i64 (AArch64NvCast (v8i8 FPR64:$src))), (v1i64 FPR64:$src)>;
5133 def : Pat<(v8i8 (AArch64NvCast (f64 FPR64:$src))), (v8i8 FPR64:$src)>;
5134 def : Pat<(v4i16 (AArch64NvCast (f64 FPR64:$src))), (v4i16 FPR64:$src)>;
5135 def : Pat<(v2i32 (AArch64NvCast (f64 FPR64:$src))), (v2i32 FPR64:$src)>;
5136 def : Pat<(v2f32 (AArch64NvCast (f64 FPR64:$src))), (v2f32 FPR64:$src)>;
5137 def : Pat<(v1i64 (AArch64NvCast (f64 FPR64:$src))), (v1i64 FPR64:$src)>;
5138 def : Pat<(v1f64 (AArch64NvCast (f64 FPR64:$src))), (v1f64 FPR64:$src)>;
5140 def : Pat<(v8i8 (AArch64NvCast (v2f32 FPR64:$src))), (v8i8 FPR64:$src)>;
5141 def : Pat<(v4i16 (AArch64NvCast (v2f32 FPR64:$src))), (v4i16 FPR64:$src)>;
5142 def : Pat<(v2i32 (AArch64NvCast (v2f32 FPR64:$src))), (v2i32 FPR64:$src)>;
5143 def : Pat<(v2f32 (AArch64NvCast (v2f32 FPR64:$src))), (v2f32 FPR64:$src)>;
5144 def : Pat<(v1i64 (AArch64NvCast (v2f32 FPR64:$src))), (v1i64 FPR64:$src)>;
5146 // Natural vector casts (128 bit)
5147 def : Pat<(v16i8 (AArch64NvCast (v4i32 FPR128:$src))), (v16i8 FPR128:$src)>;
5148 def : Pat<(v8i16 (AArch64NvCast (v4i32 FPR128:$src))), (v8i16 FPR128:$src)>;
5149 def : Pat<(v4i32 (AArch64NvCast (v4i32 FPR128:$src))), (v4i32 FPR128:$src)>;
5150 def : Pat<(v4f32 (AArch64NvCast (v4i32 FPR128:$src))), (v4f32 FPR128:$src)>;
5151 def : Pat<(v2i64 (AArch64NvCast (v4i32 FPR128:$src))), (v2i64 FPR128:$src)>;
5153 def : Pat<(v16i8 (AArch64NvCast (v8i16 FPR128:$src))), (v16i8 FPR128:$src)>;
5154 def : Pat<(v8i16 (AArch64NvCast (v8i16 FPR128:$src))), (v8i16 FPR128:$src)>;
5155 def : Pat<(v4i32 (AArch64NvCast (v8i16 FPR128:$src))), (v4i32 FPR128:$src)>;
5156 def : Pat<(v2i64 (AArch64NvCast (v8i16 FPR128:$src))), (v2i64 FPR128:$src)>;
5158 def : Pat<(v16i8 (AArch64NvCast (v16i8 FPR128:$src))), (v16i8 FPR128:$src)>;
5159 def : Pat<(v8i16 (AArch64NvCast (v16i8 FPR128:$src))), (v8i16 FPR128:$src)>;
5160 def : Pat<(v4i32 (AArch64NvCast (v16i8 FPR128:$src))), (v4i32 FPR128:$src)>;
5161 def : Pat<(v2i64 (AArch64NvCast (v16i8 FPR128:$src))), (v2i64 FPR128:$src)>;
5163 def : Pat<(v16i8 (AArch64NvCast (v2i64 FPR128:$src))), (v16i8 FPR128:$src)>;
5164 def : Pat<(v8i16 (AArch64NvCast (v2i64 FPR128:$src))), (v8i16 FPR128:$src)>;
5165 def : Pat<(v4i32 (AArch64NvCast (v2i64 FPR128:$src))), (v4i32 FPR128:$src)>;
5166 def : Pat<(v2i64 (AArch64NvCast (v2i64 FPR128:$src))), (v2i64 FPR128:$src)>;
5167 def : Pat<(v4f32 (AArch64NvCast (v2i64 FPR128:$src))), (v4f32 FPR128:$src)>;
5168 def : Pat<(v2f64 (AArch64NvCast (v2i64 FPR128:$src))), (v2f64 FPR128:$src)>;
5170 def : Pat<(v16i8 (AArch64NvCast (v4f32 FPR128:$src))), (v16i8 FPR128:$src)>;
5171 def : Pat<(v8i16 (AArch64NvCast (v4f32 FPR128:$src))), (v8i16 FPR128:$src)>;
5172 def : Pat<(v4i32 (AArch64NvCast (v4f32 FPR128:$src))), (v4i32 FPR128:$src)>;
5173 def : Pat<(v4f32 (AArch64NvCast (v4f32 FPR128:$src))), (v4f32 FPR128:$src)>;
5174 def : Pat<(v2i64 (AArch64NvCast (v4f32 FPR128:$src))), (v2i64 FPR128:$src)>;
5176 def : Pat<(v16i8 (AArch64NvCast (v2f64 FPR128:$src))), (v16i8 FPR128:$src)>;
5177 def : Pat<(v8i16 (AArch64NvCast (v2f64 FPR128:$src))), (v8i16 FPR128:$src)>;
5178 def : Pat<(v4i32 (AArch64NvCast (v2f64 FPR128:$src))), (v4i32 FPR128:$src)>;
5179 def : Pat<(v2i64 (AArch64NvCast (v2f64 FPR128:$src))), (v2i64 FPR128:$src)>;
5180 def : Pat<(v2f64 (AArch64NvCast (v2f64 FPR128:$src))), (v2f64 FPR128:$src)>;
5182 let Predicates = [IsLE] in {
5183 def : Pat<(v8i8 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5184 def : Pat<(v4i16 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5185 def : Pat<(v2i32 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5186 def : Pat<(v4f16 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5187 def : Pat<(v2f32 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5189 def : Pat<(i64 (bitconvert (v8i8 V64:$Vn))),
5190 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5191 def : Pat<(i64 (bitconvert (v4i16 V64:$Vn))),
5192 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5193 def : Pat<(i64 (bitconvert (v2i32 V64:$Vn))),
5194 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5195 def : Pat<(i64 (bitconvert (v4f16 V64:$Vn))),
5196 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5197 def : Pat<(i64 (bitconvert (v2f32 V64:$Vn))),
5198 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5199 def : Pat<(i64 (bitconvert (v1f64 V64:$Vn))),
5200 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5202 let Predicates = [IsBE] in {
5203 def : Pat<(v8i8 (bitconvert GPR64:$Xn)),
5204 (REV64v8i8 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
5205 def : Pat<(v4i16 (bitconvert GPR64:$Xn)),
5206 (REV64v4i16 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
5207 def : Pat<(v2i32 (bitconvert GPR64:$Xn)),
5208 (REV64v2i32 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
5209 def : Pat<(v4f16 (bitconvert GPR64:$Xn)),
5210 (REV64v4i16 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
5211 def : Pat<(v2f32 (bitconvert GPR64:$Xn)),
5212 (REV64v2i32 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
5214 def : Pat<(i64 (bitconvert (v8i8 V64:$Vn))),
5215 (REV64v8i8 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
5216 def : Pat<(i64 (bitconvert (v4i16 V64:$Vn))),
5217 (REV64v4i16 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
5218 def : Pat<(i64 (bitconvert (v2i32 V64:$Vn))),
5219 (REV64v2i32 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
5220 def : Pat<(i64 (bitconvert (v4f16 V64:$Vn))),
5221 (REV64v4i16 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
5222 def : Pat<(i64 (bitconvert (v2f32 V64:$Vn))),
5223 (REV64v2i32 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
5225 def : Pat<(v1i64 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5226 def : Pat<(v1f64 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5227 def : Pat<(i64 (bitconvert (v1i64 V64:$Vn))),
5228 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5229 def : Pat<(v1i64 (scalar_to_vector GPR64:$Xn)),
5230 (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5231 def : Pat<(v1f64 (scalar_to_vector GPR64:$Xn)),
5232 (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5233 def : Pat<(v1f64 (scalar_to_vector (f64 FPR64:$Xn))), (v1f64 FPR64:$Xn)>;
5235 def : Pat<(f32 (bitconvert (i32 GPR32:$Xn))),
5236 (COPY_TO_REGCLASS GPR32:$Xn, FPR32)>;
5237 def : Pat<(i32 (bitconvert (f32 FPR32:$Xn))),
5238 (COPY_TO_REGCLASS FPR32:$Xn, GPR32)>;
5239 def : Pat<(f64 (bitconvert (i64 GPR64:$Xn))),
5240 (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5241 def : Pat<(i64 (bitconvert (f64 FPR64:$Xn))),
5242 (COPY_TO_REGCLASS FPR64:$Xn, GPR64)>;
5243 def : Pat<(i64 (bitconvert (v1f64 V64:$Vn))),
5244 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5246 let Predicates = [IsLE] in {
5247 def : Pat<(v1i64 (bitconvert (v2i32 FPR64:$src))), (v1i64 FPR64:$src)>;
5248 def : Pat<(v1i64 (bitconvert (v4i16 FPR64:$src))), (v1i64 FPR64:$src)>;
5249 def : Pat<(v1i64 (bitconvert (v8i8 FPR64:$src))), (v1i64 FPR64:$src)>;
5250 def : Pat<(v1i64 (bitconvert (v4f16 FPR64:$src))), (v1i64 FPR64:$src)>;
5251 def : Pat<(v1i64 (bitconvert (v2f32 FPR64:$src))), (v1i64 FPR64:$src)>;
5253 let Predicates = [IsBE] in {
5254 def : Pat<(v1i64 (bitconvert (v2i32 FPR64:$src))),
5255 (v1i64 (REV64v2i32 FPR64:$src))>;
5256 def : Pat<(v1i64 (bitconvert (v4i16 FPR64:$src))),
5257 (v1i64 (REV64v4i16 FPR64:$src))>;
5258 def : Pat<(v1i64 (bitconvert (v8i8 FPR64:$src))),
5259 (v1i64 (REV64v8i8 FPR64:$src))>;
5260 def : Pat<(v1i64 (bitconvert (v4f16 FPR64:$src))),
5261 (v1i64 (REV64v4i16 FPR64:$src))>;
5262 def : Pat<(v1i64 (bitconvert (v2f32 FPR64:$src))),
5263 (v1i64 (REV64v2i32 FPR64:$src))>;
5265 def : Pat<(v1i64 (bitconvert (v1f64 FPR64:$src))), (v1i64 FPR64:$src)>;
5266 def : Pat<(v1i64 (bitconvert (f64 FPR64:$src))), (v1i64 FPR64:$src)>;
5268 let Predicates = [IsLE] in {
5269 def : Pat<(v2i32 (bitconvert (v1i64 FPR64:$src))), (v2i32 FPR64:$src)>;
5270 def : Pat<(v2i32 (bitconvert (v4i16 FPR64:$src))), (v2i32 FPR64:$src)>;
5271 def : Pat<(v2i32 (bitconvert (v8i8 FPR64:$src))), (v2i32 FPR64:$src)>;
5272 def : Pat<(v2i32 (bitconvert (f64 FPR64:$src))), (v2i32 FPR64:$src)>;
5273 def : Pat<(v2i32 (bitconvert (v1f64 FPR64:$src))), (v2i32 FPR64:$src)>;
5274 def : Pat<(v2i32 (bitconvert (v4f16 FPR64:$src))), (v2i32 FPR64:$src)>;
5276 let Predicates = [IsBE] in {
5277 def : Pat<(v2i32 (bitconvert (v1i64 FPR64:$src))),
5278 (v2i32 (REV64v2i32 FPR64:$src))>;
5279 def : Pat<(v2i32 (bitconvert (v4i16 FPR64:$src))),
5280 (v2i32 (REV32v4i16 FPR64:$src))>;
5281 def : Pat<(v2i32 (bitconvert (v8i8 FPR64:$src))),
5282 (v2i32 (REV32v8i8 FPR64:$src))>;
5283 def : Pat<(v2i32 (bitconvert (f64 FPR64:$src))),
5284 (v2i32 (REV64v2i32 FPR64:$src))>;
5285 def : Pat<(v2i32 (bitconvert (v1f64 FPR64:$src))),
5286 (v2i32 (REV64v2i32 FPR64:$src))>;
5287 def : Pat<(v2i32 (bitconvert (v4f16 FPR64:$src))),
5288 (v2i32 (REV64v4i16 FPR64:$src))>;
5290 def : Pat<(v2i32 (bitconvert (v2f32 FPR64:$src))), (v2i32 FPR64:$src)>;
5292 let Predicates = [IsLE] in {
5293 def : Pat<(v4i16 (bitconvert (v1i64 FPR64:$src))), (v4i16 FPR64:$src)>;
5294 def : Pat<(v4i16 (bitconvert (v2i32 FPR64:$src))), (v4i16 FPR64:$src)>;
5295 def : Pat<(v4i16 (bitconvert (v8i8 FPR64:$src))), (v4i16 FPR64:$src)>;
5296 def : Pat<(v4i16 (bitconvert (f64 FPR64:$src))), (v4i16 FPR64:$src)>;
5297 def : Pat<(v4i16 (bitconvert (v4f16 FPR64:$src))), (v4i16 FPR64:$src)>;
5298 def : Pat<(v4i16 (bitconvert (v2f32 FPR64:$src))), (v4i16 FPR64:$src)>;
5299 def : Pat<(v4i16 (bitconvert (v1f64 FPR64:$src))), (v4i16 FPR64:$src)>;
5301 let Predicates = [IsBE] in {
5302 def : Pat<(v4i16 (bitconvert (v1i64 FPR64:$src))),
5303 (v4i16 (REV64v4i16 FPR64:$src))>;
5304 def : Pat<(v4i16 (bitconvert (v2i32 FPR64:$src))),
5305 (v4i16 (REV32v4i16 FPR64:$src))>;
5306 def : Pat<(v4i16 (bitconvert (v8i8 FPR64:$src))),
5307 (v4i16 (REV16v8i8 FPR64:$src))>;
5308 def : Pat<(v4i16 (bitconvert (f64 FPR64:$src))),
5309 (v4i16 (REV64v4i16 FPR64:$src))>;
5310 def : Pat<(v4i16 (bitconvert (v4f16 FPR64:$src))),
5311 (v4i16 (REV32v4i16 FPR64:$src))>;
5312 def : Pat<(v4i16 (bitconvert (v2f32 FPR64:$src))),
5313 (v4i16 (REV32v4i16 FPR64:$src))>;
5314 def : Pat<(v4i16 (bitconvert (v1f64 FPR64:$src))),
5315 (v4i16 (REV64v4i16 FPR64:$src))>;
5318 let Predicates = [IsLE] in {
5319 def : Pat<(v4f16 (bitconvert (v1i64 FPR64:$src))), (v4f16 FPR64:$src)>;
5320 def : Pat<(v4f16 (bitconvert (v2i32 FPR64:$src))), (v4f16 FPR64:$src)>;
5321 def : Pat<(v4f16 (bitconvert (v4i16 FPR64:$src))), (v4f16 FPR64:$src)>;
5322 def : Pat<(v4f16 (bitconvert (v8i8 FPR64:$src))), (v4f16 FPR64:$src)>;
5323 def : Pat<(v4f16 (bitconvert (f64 FPR64:$src))), (v4f16 FPR64:$src)>;
5324 def : Pat<(v4f16 (bitconvert (v2f32 FPR64:$src))), (v4f16 FPR64:$src)>;
5325 def : Pat<(v4f16 (bitconvert (v1f64 FPR64:$src))), (v4f16 FPR64:$src)>;
5327 let Predicates = [IsBE] in {
5328 def : Pat<(v4f16 (bitconvert (v1i64 FPR64:$src))),
5329 (v4f16 (REV64v4i16 FPR64:$src))>;
5330 def : Pat<(v4f16 (bitconvert (v2i32 FPR64:$src))),
5331 (v4f16 (REV64v4i16 FPR64:$src))>;
5332 def : Pat<(v4f16 (bitconvert (v4i16 FPR64:$src))),
5333 (v4f16 (REV64v4i16 FPR64:$src))>;
5334 def : Pat<(v4f16 (bitconvert (v8i8 FPR64:$src))),
5335 (v4f16 (REV16v8i8 FPR64:$src))>;
5336 def : Pat<(v4f16 (bitconvert (f64 FPR64:$src))),
5337 (v4f16 (REV64v4i16 FPR64:$src))>;
5338 def : Pat<(v4f16 (bitconvert (v2f32 FPR64:$src))),
5339 (v4f16 (REV64v4i16 FPR64:$src))>;
5340 def : Pat<(v4f16 (bitconvert (v1f64 FPR64:$src))),
5341 (v4f16 (REV64v4i16 FPR64:$src))>;
5346 let Predicates = [IsLE] in {
5347 def : Pat<(v8i8 (bitconvert (v1i64 FPR64:$src))), (v8i8 FPR64:$src)>;
5348 def : Pat<(v8i8 (bitconvert (v2i32 FPR64:$src))), (v8i8 FPR64:$src)>;
5349 def : Pat<(v8i8 (bitconvert (v4i16 FPR64:$src))), (v8i8 FPR64:$src)>;
5350 def : Pat<(v8i8 (bitconvert (f64 FPR64:$src))), (v8i8 FPR64:$src)>;
5351 def : Pat<(v8i8 (bitconvert (v2f32 FPR64:$src))), (v8i8 FPR64:$src)>;
5352 def : Pat<(v8i8 (bitconvert (v1f64 FPR64:$src))), (v8i8 FPR64:$src)>;
5353 def : Pat<(v8i8 (bitconvert (v4f16 FPR64:$src))), (v8i8 FPR64:$src)>;
5355 let Predicates = [IsBE] in {
5356 def : Pat<(v8i8 (bitconvert (v1i64 FPR64:$src))),
5357 (v8i8 (REV64v8i8 FPR64:$src))>;
5358 def : Pat<(v8i8 (bitconvert (v2i32 FPR64:$src))),
5359 (v8i8 (REV32v8i8 FPR64:$src))>;
5360 def : Pat<(v8i8 (bitconvert (v4i16 FPR64:$src))),
5361 (v8i8 (REV16v8i8 FPR64:$src))>;
5362 def : Pat<(v8i8 (bitconvert (f64 FPR64:$src))),
5363 (v8i8 (REV64v8i8 FPR64:$src))>;
5364 def : Pat<(v8i8 (bitconvert (v2f32 FPR64:$src))),
5365 (v8i8 (REV32v8i8 FPR64:$src))>;
5366 def : Pat<(v8i8 (bitconvert (v1f64 FPR64:$src))),
5367 (v8i8 (REV64v8i8 FPR64:$src))>;
5368 def : Pat<(v8i8 (bitconvert (v4f16 FPR64:$src))),
5369 (v8i8 (REV16v8i8 FPR64:$src))>;
5372 let Predicates = [IsLE] in {
5373 def : Pat<(f64 (bitconvert (v2i32 FPR64:$src))), (f64 FPR64:$src)>;
5374 def : Pat<(f64 (bitconvert (v4i16 FPR64:$src))), (f64 FPR64:$src)>;
5375 def : Pat<(f64 (bitconvert (v2f32 FPR64:$src))), (f64 FPR64:$src)>;
5376 def : Pat<(f64 (bitconvert (v8i8 FPR64:$src))), (f64 FPR64:$src)>;
5377 def : Pat<(f64 (bitconvert (v4f16 FPR64:$src))), (f64 FPR64:$src)>;
5379 let Predicates = [IsBE] in {
5380 def : Pat<(f64 (bitconvert (v2i32 FPR64:$src))),
5381 (f64 (REV64v2i32 FPR64:$src))>;
5382 def : Pat<(f64 (bitconvert (v4i16 FPR64:$src))),
5383 (f64 (REV64v4i16 FPR64:$src))>;
5384 def : Pat<(f64 (bitconvert (v2f32 FPR64:$src))),
5385 (f64 (REV64v2i32 FPR64:$src))>;
5386 def : Pat<(f64 (bitconvert (v8i8 FPR64:$src))),
5387 (f64 (REV64v8i8 FPR64:$src))>;
5388 def : Pat<(f64 (bitconvert (v4f16 FPR64:$src))),
5389 (f64 (REV64v4i16 FPR64:$src))>;
5391 def : Pat<(f64 (bitconvert (v1i64 FPR64:$src))), (f64 FPR64:$src)>;
5392 def : Pat<(f64 (bitconvert (v1f64 FPR64:$src))), (f64 FPR64:$src)>;
5394 let Predicates = [IsLE] in {
5395 def : Pat<(v1f64 (bitconvert (v2i32 FPR64:$src))), (v1f64 FPR64:$src)>;
5396 def : Pat<(v1f64 (bitconvert (v4i16 FPR64:$src))), (v1f64 FPR64:$src)>;
5397 def : Pat<(v1f64 (bitconvert (v8i8 FPR64:$src))), (v1f64 FPR64:$src)>;
5398 def : Pat<(v1f64 (bitconvert (v2f32 FPR64:$src))), (v1f64 FPR64:$src)>;
5399 def : Pat<(v1f64 (bitconvert (v4f16 FPR64:$src))), (v1f64 FPR64:$src)>;
5401 let Predicates = [IsBE] in {
5402 def : Pat<(v1f64 (bitconvert (v2i32 FPR64:$src))),
5403 (v1f64 (REV64v2i32 FPR64:$src))>;
5404 def : Pat<(v1f64 (bitconvert (v4i16 FPR64:$src))),
5405 (v1f64 (REV64v4i16 FPR64:$src))>;
5406 def : Pat<(v1f64 (bitconvert (v8i8 FPR64:$src))),
5407 (v1f64 (REV64v8i8 FPR64:$src))>;
5408 def : Pat<(v1f64 (bitconvert (v2f32 FPR64:$src))),
5409 (v1f64 (REV64v2i32 FPR64:$src))>;
5410 def : Pat<(v1f64 (bitconvert (v4f16 FPR64:$src))),
5411 (v1f64 (REV64v4i16 FPR64:$src))>;
5413 def : Pat<(v1f64 (bitconvert (v1i64 FPR64:$src))), (v1f64 FPR64:$src)>;
5414 def : Pat<(v1f64 (bitconvert (f64 FPR64:$src))), (v1f64 FPR64:$src)>;
5416 let Predicates = [IsLE] in {
5417 def : Pat<(v2f32 (bitconvert (v1i64 FPR64:$src))), (v2f32 FPR64:$src)>;
5418 def : Pat<(v2f32 (bitconvert (v4i16 FPR64:$src))), (v2f32 FPR64:$src)>;
5419 def : Pat<(v2f32 (bitconvert (v8i8 FPR64:$src))), (v2f32 FPR64:$src)>;
5420 def : Pat<(v2f32 (bitconvert (v1f64 FPR64:$src))), (v2f32 FPR64:$src)>;
5421 def : Pat<(v2f32 (bitconvert (f64 FPR64:$src))), (v2f32 FPR64:$src)>;
5422 def : Pat<(v2f32 (bitconvert (v4f16 FPR64:$src))), (v2f32 FPR64:$src)>;
5424 let Predicates = [IsBE] in {
5425 def : Pat<(v2f32 (bitconvert (v1i64 FPR64:$src))),
5426 (v2f32 (REV64v2i32 FPR64:$src))>;
5427 def : Pat<(v2f32 (bitconvert (v4i16 FPR64:$src))),
5428 (v2f32 (REV32v4i16 FPR64:$src))>;
5429 def : Pat<(v2f32 (bitconvert (v8i8 FPR64:$src))),
5430 (v2f32 (REV32v8i8 FPR64:$src))>;
5431 def : Pat<(v2f32 (bitconvert (v1f64 FPR64:$src))),
5432 (v2f32 (REV64v2i32 FPR64:$src))>;
5433 def : Pat<(v2f32 (bitconvert (f64 FPR64:$src))),
5434 (v2f32 (REV64v2i32 FPR64:$src))>;
5435 def : Pat<(v2f32 (bitconvert (v4f16 FPR64:$src))),
5436 (v2f32 (REV64v4i16 FPR64:$src))>;
5438 def : Pat<(v2f32 (bitconvert (v2i32 FPR64:$src))), (v2f32 FPR64:$src)>;
5440 let Predicates = [IsLE] in {
5441 def : Pat<(f128 (bitconvert (v2i64 FPR128:$src))), (f128 FPR128:$src)>;
5442 def : Pat<(f128 (bitconvert (v4i32 FPR128:$src))), (f128 FPR128:$src)>;
5443 def : Pat<(f128 (bitconvert (v8i16 FPR128:$src))), (f128 FPR128:$src)>;
5444 def : Pat<(f128 (bitconvert (v2f64 FPR128:$src))), (f128 FPR128:$src)>;
5445 def : Pat<(f128 (bitconvert (v4f32 FPR128:$src))), (f128 FPR128:$src)>;
5446 def : Pat<(f128 (bitconvert (v8f16 FPR128:$src))), (f128 FPR128:$src)>;
5447 def : Pat<(f128 (bitconvert (v16i8 FPR128:$src))), (f128 FPR128:$src)>;
5449 let Predicates = [IsBE] in {
5450 def : Pat<(f128 (bitconvert (v2i64 FPR128:$src))),
5451 (f128 (EXTv16i8 FPR128:$src, FPR128:$src, (i32 8)))>;
5452 def : Pat<(f128 (bitconvert (v4i32 FPR128:$src))),
5453 (f128 (EXTv16i8 (REV64v4i32 FPR128:$src),
5454 (REV64v4i32 FPR128:$src), (i32 8)))>;
5455 def : Pat<(f128 (bitconvert (v8i16 FPR128:$src))),
5456 (f128 (EXTv16i8 (REV64v8i16 FPR128:$src),
5457 (REV64v8i16 FPR128:$src), (i32 8)))>;
5458 def : Pat<(f128 (bitconvert (v8f16 FPR128:$src))),
5459 (f128 (EXTv16i8 (REV64v8i16 FPR128:$src),
5460 (REV64v8i16 FPR128:$src), (i32 8)))>;
5461 def : Pat<(f128 (bitconvert (v2f64 FPR128:$src))),
5462 (f128 (EXTv16i8 FPR128:$src, FPR128:$src, (i32 8)))>;
5463 def : Pat<(f128 (bitconvert (v4f32 FPR128:$src))),
5464 (f128 (EXTv16i8 (REV64v4i32 FPR128:$src),
5465 (REV64v4i32 FPR128:$src), (i32 8)))>;
5466 def : Pat<(f128 (bitconvert (v16i8 FPR128:$src))),
5467 (f128 (EXTv16i8 (REV64v16i8 FPR128:$src),
5468 (REV64v16i8 FPR128:$src), (i32 8)))>;
5471 let Predicates = [IsLE] in {
5472 def : Pat<(v2f64 (bitconvert (f128 FPR128:$src))), (v2f64 FPR128:$src)>;
5473 def : Pat<(v2f64 (bitconvert (v4i32 FPR128:$src))), (v2f64 FPR128:$src)>;
5474 def : Pat<(v2f64 (bitconvert (v8i16 FPR128:$src))), (v2f64 FPR128:$src)>;
5475 def : Pat<(v2f64 (bitconvert (v8f16 FPR128:$src))), (v2f64 FPR128:$src)>;
5476 def : Pat<(v2f64 (bitconvert (v16i8 FPR128:$src))), (v2f64 FPR128:$src)>;
5477 def : Pat<(v2f64 (bitconvert (v4f32 FPR128:$src))), (v2f64 FPR128:$src)>;
5479 let Predicates = [IsBE] in {
5480 def : Pat<(v2f64 (bitconvert (f128 FPR128:$src))),
5481 (v2f64 (EXTv16i8 FPR128:$src,
5482 FPR128:$src, (i32 8)))>;
5483 def : Pat<(v2f64 (bitconvert (v4i32 FPR128:$src))),
5484 (v2f64 (REV64v4i32 FPR128:$src))>;
5485 def : Pat<(v2f64 (bitconvert (v8i16 FPR128:$src))),
5486 (v2f64 (REV64v8i16 FPR128:$src))>;
5487 def : Pat<(v2f64 (bitconvert (v8f16 FPR128:$src))),
5488 (v2f64 (REV64v8i16 FPR128:$src))>;
5489 def : Pat<(v2f64 (bitconvert (v16i8 FPR128:$src))),
5490 (v2f64 (REV64v16i8 FPR128:$src))>;
5491 def : Pat<(v2f64 (bitconvert (v4f32 FPR128:$src))),
5492 (v2f64 (REV64v4i32 FPR128:$src))>;
5494 def : Pat<(v2f64 (bitconvert (v2i64 FPR128:$src))), (v2f64 FPR128:$src)>;
5496 let Predicates = [IsLE] in {
5497 def : Pat<(v4f32 (bitconvert (f128 FPR128:$src))), (v4f32 FPR128:$src)>;
5498 def : Pat<(v4f32 (bitconvert (v8i16 FPR128:$src))), (v4f32 FPR128:$src)>;
5499 def : Pat<(v4f32 (bitconvert (v8f16 FPR128:$src))), (v4f32 FPR128:$src)>;
5500 def : Pat<(v4f32 (bitconvert (v16i8 FPR128:$src))), (v4f32 FPR128:$src)>;
5501 def : Pat<(v4f32 (bitconvert (v2i64 FPR128:$src))), (v4f32 FPR128:$src)>;
5502 def : Pat<(v4f32 (bitconvert (v2f64 FPR128:$src))), (v4f32 FPR128:$src)>;
5504 let Predicates = [IsBE] in {
5505 def : Pat<(v4f32 (bitconvert (f128 FPR128:$src))),
5506 (v4f32 (EXTv16i8 (REV64v4i32 FPR128:$src),
5507 (REV64v4i32 FPR128:$src), (i32 8)))>;
5508 def : Pat<(v4f32 (bitconvert (v8i16 FPR128:$src))),
5509 (v4f32 (REV32v8i16 FPR128:$src))>;
5510 def : Pat<(v4f32 (bitconvert (v8f16 FPR128:$src))),
5511 (v4f32 (REV32v8i16 FPR128:$src))>;
5512 def : Pat<(v4f32 (bitconvert (v16i8 FPR128:$src))),
5513 (v4f32 (REV32v16i8 FPR128:$src))>;
5514 def : Pat<(v4f32 (bitconvert (v2i64 FPR128:$src))),
5515 (v4f32 (REV64v4i32 FPR128:$src))>;
5516 def : Pat<(v4f32 (bitconvert (v2f64 FPR128:$src))),
5517 (v4f32 (REV64v4i32 FPR128:$src))>;
5519 def : Pat<(v4f32 (bitconvert (v4i32 FPR128:$src))), (v4f32 FPR128:$src)>;
5521 let Predicates = [IsLE] in {
5522 def : Pat<(v2i64 (bitconvert (f128 FPR128:$src))), (v2i64 FPR128:$src)>;
5523 def : Pat<(v2i64 (bitconvert (v4i32 FPR128:$src))), (v2i64 FPR128:$src)>;
5524 def : Pat<(v2i64 (bitconvert (v8i16 FPR128:$src))), (v2i64 FPR128:$src)>;
5525 def : Pat<(v2i64 (bitconvert (v16i8 FPR128:$src))), (v2i64 FPR128:$src)>;
5526 def : Pat<(v2i64 (bitconvert (v4f32 FPR128:$src))), (v2i64 FPR128:$src)>;
5527 def : Pat<(v2i64 (bitconvert (v8f16 FPR128:$src))), (v2i64 FPR128:$src)>;
5529 let Predicates = [IsBE] in {
5530 def : Pat<(v2i64 (bitconvert (f128 FPR128:$src))),
5531 (v2i64 (EXTv16i8 FPR128:$src,
5532 FPR128:$src, (i32 8)))>;
5533 def : Pat<(v2i64 (bitconvert (v4i32 FPR128:$src))),
5534 (v2i64 (REV64v4i32 FPR128:$src))>;
5535 def : Pat<(v2i64 (bitconvert (v8i16 FPR128:$src))),
5536 (v2i64 (REV64v8i16 FPR128:$src))>;
5537 def : Pat<(v2i64 (bitconvert (v16i8 FPR128:$src))),
5538 (v2i64 (REV64v16i8 FPR128:$src))>;
5539 def : Pat<(v2i64 (bitconvert (v4f32 FPR128:$src))),
5540 (v2i64 (REV64v4i32 FPR128:$src))>;
5541 def : Pat<(v2i64 (bitconvert (v8f16 FPR128:$src))),
5542 (v2i64 (REV64v8i16 FPR128:$src))>;
5544 def : Pat<(v2i64 (bitconvert (v2f64 FPR128:$src))), (v2i64 FPR128:$src)>;
5546 let Predicates = [IsLE] in {
5547 def : Pat<(v4i32 (bitconvert (f128 FPR128:$src))), (v4i32 FPR128:$src)>;
5548 def : Pat<(v4i32 (bitconvert (v2i64 FPR128:$src))), (v4i32 FPR128:$src)>;
5549 def : Pat<(v4i32 (bitconvert (v8i16 FPR128:$src))), (v4i32 FPR128:$src)>;
5550 def : Pat<(v4i32 (bitconvert (v16i8 FPR128:$src))), (v4i32 FPR128:$src)>;
5551 def : Pat<(v4i32 (bitconvert (v2f64 FPR128:$src))), (v4i32 FPR128:$src)>;
5552 def : Pat<(v4i32 (bitconvert (v8f16 FPR128:$src))), (v4i32 FPR128:$src)>;
5554 let Predicates = [IsBE] in {
5555 def : Pat<(v4i32 (bitconvert (f128 FPR128:$src))),
5556 (v4i32 (EXTv16i8 (REV64v4i32 FPR128:$src),
5557 (REV64v4i32 FPR128:$src),
5559 def : Pat<(v4i32 (bitconvert (v2i64 FPR128:$src))),
5560 (v4i32 (REV64v4i32 FPR128:$src))>;
5561 def : Pat<(v4i32 (bitconvert (v8i16 FPR128:$src))),
5562 (v4i32 (REV32v8i16 FPR128:$src))>;
5563 def : Pat<(v4i32 (bitconvert (v16i8 FPR128:$src))),
5564 (v4i32 (REV32v16i8 FPR128:$src))>;
5565 def : Pat<(v4i32 (bitconvert (v2f64 FPR128:$src))),
5566 (v4i32 (REV64v4i32 FPR128:$src))>;
5567 def : Pat<(v4i32 (bitconvert (v8f16 FPR128:$src))),
5568 (v4i32 (REV32v8i16 FPR128:$src))>;
5570 def : Pat<(v4i32 (bitconvert (v4f32 FPR128:$src))), (v4i32 FPR128:$src)>;
5572 let Predicates = [IsLE] in {
5573 def : Pat<(v8i16 (bitconvert (f128 FPR128:$src))), (v8i16 FPR128:$src)>;
5574 def : Pat<(v8i16 (bitconvert (v2i64 FPR128:$src))), (v8i16 FPR128:$src)>;
5575 def : Pat<(v8i16 (bitconvert (v4i32 FPR128:$src))), (v8i16 FPR128:$src)>;
5576 def : Pat<(v8i16 (bitconvert (v16i8 FPR128:$src))), (v8i16 FPR128:$src)>;
5577 def : Pat<(v8i16 (bitconvert (v2f64 FPR128:$src))), (v8i16 FPR128:$src)>;
5578 def : Pat<(v8i16 (bitconvert (v4f32 FPR128:$src))), (v8i16 FPR128:$src)>;
5579 def : Pat<(v8i16 (bitconvert (v8f16 FPR128:$src))), (v8i16 FPR128:$src)>;
5581 let Predicates = [IsBE] in {
5582 def : Pat<(v8i16 (bitconvert (f128 FPR128:$src))),
5583 (v8i16 (EXTv16i8 (REV64v8i16 FPR128:$src),
5584 (REV64v8i16 FPR128:$src),
5586 def : Pat<(v8i16 (bitconvert (v2i64 FPR128:$src))),
5587 (v8i16 (REV64v8i16 FPR128:$src))>;
5588 def : Pat<(v8i16 (bitconvert (v4i32 FPR128:$src))),
5589 (v8i16 (REV32v8i16 FPR128:$src))>;
5590 def : Pat<(v8i16 (bitconvert (v16i8 FPR128:$src))),
5591 (v8i16 (REV16v16i8 FPR128:$src))>;
5592 def : Pat<(v8i16 (bitconvert (v2f64 FPR128:$src))),
5593 (v8i16 (REV64v8i16 FPR128:$src))>;
5594 def : Pat<(v8i16 (bitconvert (v4f32 FPR128:$src))),
5595 (v8i16 (REV32v8i16 FPR128:$src))>;
5596 def : Pat<(v8i16 (bitconvert (v8f16 FPR128:$src))),
5597 (v8i16 (REV32v8i16 FPR128:$src))>;
5600 let Predicates = [IsLE] in {
5601 def : Pat<(v8f16 (bitconvert (f128 FPR128:$src))), (v8f16 FPR128:$src)>;
5602 def : Pat<(v8f16 (bitconvert (v2i64 FPR128:$src))), (v8f16 FPR128:$src)>;
5603 def : Pat<(v8f16 (bitconvert (v4i32 FPR128:$src))), (v8f16 FPR128:$src)>;
5604 def : Pat<(v8f16 (bitconvert (v8i16 FPR128:$src))), (v8f16 FPR128:$src)>;
5605 def : Pat<(v8f16 (bitconvert (v16i8 FPR128:$src))), (v8f16 FPR128:$src)>;
5606 def : Pat<(v8f16 (bitconvert (v2f64 FPR128:$src))), (v8f16 FPR128:$src)>;
5607 def : Pat<(v8f16 (bitconvert (v4f32 FPR128:$src))), (v8f16 FPR128:$src)>;
5609 let Predicates = [IsBE] in {
5610 def : Pat<(v8f16 (bitconvert (f128 FPR128:$src))),
5611 (v8f16 (EXTv16i8 (REV64v8i16 FPR128:$src),
5612 (REV64v8i16 FPR128:$src),
5614 def : Pat<(v8f16 (bitconvert (v2i64 FPR128:$src))),
5615 (v8f16 (REV64v8i16 FPR128:$src))>;
5616 def : Pat<(v8f16 (bitconvert (v4i32 FPR128:$src))),
5617 (v8f16 (REV32v8i16 FPR128:$src))>;
5618 def : Pat<(v8f16 (bitconvert (v8i16 FPR128:$src))),
5619 (v8f16 (REV64v8i16 FPR128:$src))>;
5620 def : Pat<(v8f16 (bitconvert (v16i8 FPR128:$src))),
5621 (v8f16 (REV16v16i8 FPR128:$src))>;
5622 def : Pat<(v8f16 (bitconvert (v2f64 FPR128:$src))),
5623 (v8f16 (REV64v8i16 FPR128:$src))>;
5624 def : Pat<(v8f16 (bitconvert (v4f32 FPR128:$src))),
5625 (v8f16 (REV32v8i16 FPR128:$src))>;
5628 let Predicates = [IsLE] in {
5629 def : Pat<(v16i8 (bitconvert (f128 FPR128:$src))), (v16i8 FPR128:$src)>;
5630 def : Pat<(v16i8 (bitconvert (v2i64 FPR128:$src))), (v16i8 FPR128:$src)>;
5631 def : Pat<(v16i8 (bitconvert (v4i32 FPR128:$src))), (v16i8 FPR128:$src)>;
5632 def : Pat<(v16i8 (bitconvert (v8i16 FPR128:$src))), (v16i8 FPR128:$src)>;
5633 def : Pat<(v16i8 (bitconvert (v2f64 FPR128:$src))), (v16i8 FPR128:$src)>;
5634 def : Pat<(v16i8 (bitconvert (v4f32 FPR128:$src))), (v16i8 FPR128:$src)>;
5635 def : Pat<(v16i8 (bitconvert (v8f16 FPR128:$src))), (v16i8 FPR128:$src)>;
5637 let Predicates = [IsBE] in {
5638 def : Pat<(v16i8 (bitconvert (f128 FPR128:$src))),
5639 (v16i8 (EXTv16i8 (REV64v16i8 FPR128:$src),
5640 (REV64v16i8 FPR128:$src),
5642 def : Pat<(v16i8 (bitconvert (v2i64 FPR128:$src))),
5643 (v16i8 (REV64v16i8 FPR128:$src))>;
5644 def : Pat<(v16i8 (bitconvert (v4i32 FPR128:$src))),
5645 (v16i8 (REV32v16i8 FPR128:$src))>;
5646 def : Pat<(v16i8 (bitconvert (v8i16 FPR128:$src))),
5647 (v16i8 (REV16v16i8 FPR128:$src))>;
5648 def : Pat<(v16i8 (bitconvert (v2f64 FPR128:$src))),
5649 (v16i8 (REV64v16i8 FPR128:$src))>;
5650 def : Pat<(v16i8 (bitconvert (v4f32 FPR128:$src))),
5651 (v16i8 (REV32v16i8 FPR128:$src))>;
5652 def : Pat<(v16i8 (bitconvert (v8f16 FPR128:$src))),
5653 (v16i8 (REV16v16i8 FPR128:$src))>;
5656 def : Pat<(v8i8 (extract_subvector (v16i8 FPR128:$Rn), (i64 1))),
5657 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
5658 def : Pat<(v4i16 (extract_subvector (v8i16 FPR128:$Rn), (i64 1))),
5659 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
5660 def : Pat<(v2i32 (extract_subvector (v4i32 FPR128:$Rn), (i64 1))),
5661 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
5662 def : Pat<(v1i64 (extract_subvector (v2i64 FPR128:$Rn), (i64 1))),
5663 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
5665 // A 64-bit subvector insert to the first 128-bit vector position
5666 // is a subregister copy that needs no instruction.
5667 def : Pat<(insert_subvector undef, (v1i64 FPR64:$src), (i32 0)),
5668 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5669 def : Pat<(insert_subvector undef, (v1f64 FPR64:$src), (i32 0)),
5670 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5671 def : Pat<(insert_subvector undef, (v2i32 FPR64:$src), (i32 0)),
5672 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5673 def : Pat<(insert_subvector undef, (v2f32 FPR64:$src), (i32 0)),
5674 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5675 def : Pat<(insert_subvector undef, (v4i16 FPR64:$src), (i32 0)),
5676 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5677 def : Pat<(insert_subvector undef, (v4f16 FPR64:$src), (i32 0)),
5678 (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5679 def : Pat<(insert_subvector undef, (v8i8 FPR64:$src), (i32 0)),
5680 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5682 // Use pair-wise add instructions when summing up the lanes for v2f64, v2i64
5684 def : Pat<(i64 (add (vector_extract (v2i64 FPR128:$Rn), (i64 0)),
5685 (vector_extract (v2i64 FPR128:$Rn), (i64 1)))),
5686 (i64 (ADDPv2i64p (v2i64 FPR128:$Rn)))>;
5687 def : Pat<(f64 (fadd (vector_extract (v2f64 FPR128:$Rn), (i64 0)),
5688 (vector_extract (v2f64 FPR128:$Rn), (i64 1)))),
5689 (f64 (FADDPv2i64p (v2f64 FPR128:$Rn)))>;
5690 // vector_extract on 64-bit vectors gets promoted to a 128 bit vector,
5691 // so we match on v4f32 here, not v2f32. This will also catch adding
5692 // the low two lanes of a true v4f32 vector.
5693 def : Pat<(fadd (vector_extract (v4f32 FPR128:$Rn), (i64 0)),
5694 (vector_extract (v4f32 FPR128:$Rn), (i64 1))),
5695 (f32 (FADDPv2i32p (EXTRACT_SUBREG FPR128:$Rn, dsub)))>;
5697 // Scalar 64-bit shifts in FPR64 registers.
5698 def : Pat<(i64 (int_aarch64_neon_sshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
5699 (SSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
5700 def : Pat<(i64 (int_aarch64_neon_ushl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
5701 (USHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
5702 def : Pat<(i64 (int_aarch64_neon_srshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
5703 (SRSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
5704 def : Pat<(i64 (int_aarch64_neon_urshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
5705 (URSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
5707 // Tail call return handling. These are all compiler pseudo-instructions,
5708 // so no encoding information or anything like that.
5709 let isCall = 1, isTerminator = 1, isReturn = 1, isBarrier = 1, Uses = [SP] in {
5710 def TCRETURNdi : Pseudo<(outs), (ins i64imm:$dst, i32imm:$FPDiff),[]>;
5711 def TCRETURNri : Pseudo<(outs), (ins tcGPR64:$dst, i32imm:$FPDiff), []>;
5714 def : Pat<(AArch64tcret tcGPR64:$dst, (i32 timm:$FPDiff)),
5715 (TCRETURNri tcGPR64:$dst, imm:$FPDiff)>;
5716 def : Pat<(AArch64tcret tglobaladdr:$dst, (i32 timm:$FPDiff)),
5717 (TCRETURNdi texternalsym:$dst, imm:$FPDiff)>;
5718 def : Pat<(AArch64tcret texternalsym:$dst, (i32 timm:$FPDiff)),
5719 (TCRETURNdi texternalsym:$dst, imm:$FPDiff)>;
5721 include "AArch64InstrAtomics.td"