1 //=- AArch64InstrInfo.td - Describe the AArch64 Instructions -*- tablegen -*-=//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // AArch64 Instruction definitions.
12 //===----------------------------------------------------------------------===//
14 //===----------------------------------------------------------------------===//
15 // ARM Instruction Predicate Definitions.
17 def HasV8_1a : Predicate<"Subtarget->hasV8_1aOps()">,
18 AssemblerPredicate<"HasV8_1aOps", "armv8.1a">;
19 def HasFPARMv8 : Predicate<"Subtarget->hasFPARMv8()">,
20 AssemblerPredicate<"FeatureFPARMv8", "fp-armv8">;
21 def HasNEON : Predicate<"Subtarget->hasNEON()">,
22 AssemblerPredicate<"FeatureNEON", "neon">;
23 def HasCrypto : Predicate<"Subtarget->hasCrypto()">,
24 AssemblerPredicate<"FeatureCrypto", "crypto">;
25 def HasCRC : Predicate<"Subtarget->hasCRC()">,
26 AssemblerPredicate<"FeatureCRC", "crc">;
27 def HasPerfMon : Predicate<"Subtarget->hasPerfMon()">;
28 def IsLE : Predicate<"Subtarget->isLittleEndian()">;
29 def IsBE : Predicate<"!Subtarget->isLittleEndian()">;
30 def IsCyclone : Predicate<"Subtarget->isCyclone()">;
32 //===----------------------------------------------------------------------===//
33 // AArch64-specific DAG Nodes.
36 // SDTBinaryArithWithFlagsOut - RES1, FLAGS = op LHS, RHS
37 def SDTBinaryArithWithFlagsOut : SDTypeProfile<2, 2,
40 SDTCisInt<0>, SDTCisVT<1, i32>]>;
42 // SDTBinaryArithWithFlagsIn - RES1, FLAGS = op LHS, RHS, FLAGS
43 def SDTBinaryArithWithFlagsIn : SDTypeProfile<1, 3,
49 // SDTBinaryArithWithFlagsInOut - RES1, FLAGS = op LHS, RHS, FLAGS
50 def SDTBinaryArithWithFlagsInOut : SDTypeProfile<2, 3,
57 def SDT_AArch64Brcond : SDTypeProfile<0, 3,
58 [SDTCisVT<0, OtherVT>, SDTCisVT<1, i32>,
60 def SDT_AArch64cbz : SDTypeProfile<0, 2, [SDTCisInt<0>, SDTCisVT<1, OtherVT>]>;
61 def SDT_AArch64tbz : SDTypeProfile<0, 3, [SDTCisInt<0>, SDTCisInt<1>,
62 SDTCisVT<2, OtherVT>]>;
65 def SDT_AArch64CSel : SDTypeProfile<1, 4,
70 def SDT_AArch64CCMP : SDTypeProfile<1, 5,
77 def SDT_AArch64FCCMP : SDTypeProfile<1, 5,
84 def SDT_AArch64FCmp : SDTypeProfile<0, 2,
87 def SDT_AArch64Dup : SDTypeProfile<1, 1, [SDTCisVec<0>]>;
88 def SDT_AArch64DupLane : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisInt<2>]>;
89 def SDT_AArch64Zip : SDTypeProfile<1, 2, [SDTCisVec<0>,
92 def SDT_AArch64MOVIedit : SDTypeProfile<1, 1, [SDTCisInt<1>]>;
93 def SDT_AArch64MOVIshift : SDTypeProfile<1, 2, [SDTCisInt<1>, SDTCisInt<2>]>;
94 def SDT_AArch64vecimm : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0,1>,
95 SDTCisInt<2>, SDTCisInt<3>]>;
96 def SDT_AArch64UnaryVec: SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0,1>]>;
97 def SDT_AArch64ExtVec: SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0,1>,
98 SDTCisSameAs<0,2>, SDTCisInt<3>]>;
99 def SDT_AArch64vshift : SDTypeProfile<1, 2, [SDTCisSameAs<0,1>, SDTCisInt<2>]>;
101 def SDT_AArch64unvec : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0,1>]>;
102 def SDT_AArch64fcmpz : SDTypeProfile<1, 1, []>;
103 def SDT_AArch64fcmp : SDTypeProfile<1, 2, [SDTCisSameAs<1,2>]>;
104 def SDT_AArch64binvec : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisSameAs<0,1>,
106 def SDT_AArch64trivec : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0,1>,
109 def SDT_AArch64TCRET : SDTypeProfile<0, 2, [SDTCisPtrTy<0>]>;
110 def SDT_AArch64PREFETCH : SDTypeProfile<0, 2, [SDTCisVT<0, i32>, SDTCisPtrTy<1>]>;
112 def SDT_AArch64ITOF : SDTypeProfile<1, 1, [SDTCisFP<0>, SDTCisSameAs<0,1>]>;
114 def SDT_AArch64TLSDescCall : SDTypeProfile<0, -2, [SDTCisPtrTy<0>,
117 // Generates the general dynamic sequences, i.e.
118 // adrp x0, :tlsdesc:var
119 // ldr x1, [x0, #:tlsdesc_lo12:var]
120 // add x0, x0, #:tlsdesc_lo12:var
124 // (the TPIDR_EL0 offset is put directly in X0, hence no "result" here)
125 // number of operands (the variable)
126 def SDT_AArch64TLSDescCallSeq : SDTypeProfile<0,1,
129 def SDT_AArch64WrapperLarge : SDTypeProfile<1, 4,
130 [SDTCisVT<0, i64>, SDTCisVT<1, i32>,
131 SDTCisSameAs<1, 2>, SDTCisSameAs<1, 3>,
132 SDTCisSameAs<1, 4>]>;
136 def AArch64adrp : SDNode<"AArch64ISD::ADRP", SDTIntUnaryOp, []>;
137 def AArch64addlow : SDNode<"AArch64ISD::ADDlow", SDTIntBinOp, []>;
138 def AArch64LOADgot : SDNode<"AArch64ISD::LOADgot", SDTIntUnaryOp>;
139 def AArch64callseq_start : SDNode<"ISD::CALLSEQ_START",
140 SDCallSeqStart<[ SDTCisVT<0, i32> ]>,
141 [SDNPHasChain, SDNPOutGlue]>;
142 def AArch64callseq_end : SDNode<"ISD::CALLSEQ_END",
143 SDCallSeqEnd<[ SDTCisVT<0, i32>,
145 [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue]>;
146 def AArch64call : SDNode<"AArch64ISD::CALL",
147 SDTypeProfile<0, -1, [SDTCisPtrTy<0>]>,
148 [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue,
150 def AArch64brcond : SDNode<"AArch64ISD::BRCOND", SDT_AArch64Brcond,
152 def AArch64cbz : SDNode<"AArch64ISD::CBZ", SDT_AArch64cbz,
154 def AArch64cbnz : SDNode<"AArch64ISD::CBNZ", SDT_AArch64cbz,
156 def AArch64tbz : SDNode<"AArch64ISD::TBZ", SDT_AArch64tbz,
158 def AArch64tbnz : SDNode<"AArch64ISD::TBNZ", SDT_AArch64tbz,
162 def AArch64csel : SDNode<"AArch64ISD::CSEL", SDT_AArch64CSel>;
163 def AArch64csinv : SDNode<"AArch64ISD::CSINV", SDT_AArch64CSel>;
164 def AArch64csneg : SDNode<"AArch64ISD::CSNEG", SDT_AArch64CSel>;
165 def AArch64csinc : SDNode<"AArch64ISD::CSINC", SDT_AArch64CSel>;
166 def AArch64retflag : SDNode<"AArch64ISD::RET_FLAG", SDTNone,
167 [SDNPHasChain, SDNPOptInGlue, SDNPVariadic]>;
168 def AArch64adc : SDNode<"AArch64ISD::ADC", SDTBinaryArithWithFlagsIn >;
169 def AArch64sbc : SDNode<"AArch64ISD::SBC", SDTBinaryArithWithFlagsIn>;
170 def AArch64add_flag : SDNode<"AArch64ISD::ADDS", SDTBinaryArithWithFlagsOut,
172 def AArch64sub_flag : SDNode<"AArch64ISD::SUBS", SDTBinaryArithWithFlagsOut>;
173 def AArch64and_flag : SDNode<"AArch64ISD::ANDS", SDTBinaryArithWithFlagsOut,
175 def AArch64adc_flag : SDNode<"AArch64ISD::ADCS", SDTBinaryArithWithFlagsInOut>;
176 def AArch64sbc_flag : SDNode<"AArch64ISD::SBCS", SDTBinaryArithWithFlagsInOut>;
178 def AArch64ccmp : SDNode<"AArch64ISD::CCMP", SDT_AArch64CCMP>;
179 def AArch64ccmn : SDNode<"AArch64ISD::CCMN", SDT_AArch64CCMP>;
180 def AArch64fccmp : SDNode<"AArch64ISD::FCCMP", SDT_AArch64FCCMP>;
182 def AArch64threadpointer : SDNode<"AArch64ISD::THREAD_POINTER", SDTPtrLeaf>;
184 def AArch64fcmp : SDNode<"AArch64ISD::FCMP", SDT_AArch64FCmp>;
186 def AArch64dup : SDNode<"AArch64ISD::DUP", SDT_AArch64Dup>;
187 def AArch64duplane8 : SDNode<"AArch64ISD::DUPLANE8", SDT_AArch64DupLane>;
188 def AArch64duplane16 : SDNode<"AArch64ISD::DUPLANE16", SDT_AArch64DupLane>;
189 def AArch64duplane32 : SDNode<"AArch64ISD::DUPLANE32", SDT_AArch64DupLane>;
190 def AArch64duplane64 : SDNode<"AArch64ISD::DUPLANE64", SDT_AArch64DupLane>;
192 def AArch64zip1 : SDNode<"AArch64ISD::ZIP1", SDT_AArch64Zip>;
193 def AArch64zip2 : SDNode<"AArch64ISD::ZIP2", SDT_AArch64Zip>;
194 def AArch64uzp1 : SDNode<"AArch64ISD::UZP1", SDT_AArch64Zip>;
195 def AArch64uzp2 : SDNode<"AArch64ISD::UZP2", SDT_AArch64Zip>;
196 def AArch64trn1 : SDNode<"AArch64ISD::TRN1", SDT_AArch64Zip>;
197 def AArch64trn2 : SDNode<"AArch64ISD::TRN2", SDT_AArch64Zip>;
199 def AArch64movi_edit : SDNode<"AArch64ISD::MOVIedit", SDT_AArch64MOVIedit>;
200 def AArch64movi_shift : SDNode<"AArch64ISD::MOVIshift", SDT_AArch64MOVIshift>;
201 def AArch64movi_msl : SDNode<"AArch64ISD::MOVImsl", SDT_AArch64MOVIshift>;
202 def AArch64mvni_shift : SDNode<"AArch64ISD::MVNIshift", SDT_AArch64MOVIshift>;
203 def AArch64mvni_msl : SDNode<"AArch64ISD::MVNImsl", SDT_AArch64MOVIshift>;
204 def AArch64movi : SDNode<"AArch64ISD::MOVI", SDT_AArch64MOVIedit>;
205 def AArch64fmov : SDNode<"AArch64ISD::FMOV", SDT_AArch64MOVIedit>;
207 def AArch64rev16 : SDNode<"AArch64ISD::REV16", SDT_AArch64UnaryVec>;
208 def AArch64rev32 : SDNode<"AArch64ISD::REV32", SDT_AArch64UnaryVec>;
209 def AArch64rev64 : SDNode<"AArch64ISD::REV64", SDT_AArch64UnaryVec>;
210 def AArch64ext : SDNode<"AArch64ISD::EXT", SDT_AArch64ExtVec>;
212 def AArch64vashr : SDNode<"AArch64ISD::VASHR", SDT_AArch64vshift>;
213 def AArch64vlshr : SDNode<"AArch64ISD::VLSHR", SDT_AArch64vshift>;
214 def AArch64vshl : SDNode<"AArch64ISD::VSHL", SDT_AArch64vshift>;
215 def AArch64sqshli : SDNode<"AArch64ISD::SQSHL_I", SDT_AArch64vshift>;
216 def AArch64uqshli : SDNode<"AArch64ISD::UQSHL_I", SDT_AArch64vshift>;
217 def AArch64sqshlui : SDNode<"AArch64ISD::SQSHLU_I", SDT_AArch64vshift>;
218 def AArch64srshri : SDNode<"AArch64ISD::SRSHR_I", SDT_AArch64vshift>;
219 def AArch64urshri : SDNode<"AArch64ISD::URSHR_I", SDT_AArch64vshift>;
221 def AArch64not: SDNode<"AArch64ISD::NOT", SDT_AArch64unvec>;
222 def AArch64bit: SDNode<"AArch64ISD::BIT", SDT_AArch64trivec>;
223 def AArch64bsl: SDNode<"AArch64ISD::BSL", SDT_AArch64trivec>;
225 def AArch64cmeq: SDNode<"AArch64ISD::CMEQ", SDT_AArch64binvec>;
226 def AArch64cmge: SDNode<"AArch64ISD::CMGE", SDT_AArch64binvec>;
227 def AArch64cmgt: SDNode<"AArch64ISD::CMGT", SDT_AArch64binvec>;
228 def AArch64cmhi: SDNode<"AArch64ISD::CMHI", SDT_AArch64binvec>;
229 def AArch64cmhs: SDNode<"AArch64ISD::CMHS", SDT_AArch64binvec>;
231 def AArch64fcmeq: SDNode<"AArch64ISD::FCMEQ", SDT_AArch64fcmp>;
232 def AArch64fcmge: SDNode<"AArch64ISD::FCMGE", SDT_AArch64fcmp>;
233 def AArch64fcmgt: SDNode<"AArch64ISD::FCMGT", SDT_AArch64fcmp>;
235 def AArch64cmeqz: SDNode<"AArch64ISD::CMEQz", SDT_AArch64unvec>;
236 def AArch64cmgez: SDNode<"AArch64ISD::CMGEz", SDT_AArch64unvec>;
237 def AArch64cmgtz: SDNode<"AArch64ISD::CMGTz", SDT_AArch64unvec>;
238 def AArch64cmlez: SDNode<"AArch64ISD::CMLEz", SDT_AArch64unvec>;
239 def AArch64cmltz: SDNode<"AArch64ISD::CMLTz", SDT_AArch64unvec>;
240 def AArch64cmtst : PatFrag<(ops node:$LHS, node:$RHS),
241 (AArch64not (AArch64cmeqz (and node:$LHS, node:$RHS)))>;
243 def AArch64fcmeqz: SDNode<"AArch64ISD::FCMEQz", SDT_AArch64fcmpz>;
244 def AArch64fcmgez: SDNode<"AArch64ISD::FCMGEz", SDT_AArch64fcmpz>;
245 def AArch64fcmgtz: SDNode<"AArch64ISD::FCMGTz", SDT_AArch64fcmpz>;
246 def AArch64fcmlez: SDNode<"AArch64ISD::FCMLEz", SDT_AArch64fcmpz>;
247 def AArch64fcmltz: SDNode<"AArch64ISD::FCMLTz", SDT_AArch64fcmpz>;
249 def AArch64bici: SDNode<"AArch64ISD::BICi", SDT_AArch64vecimm>;
250 def AArch64orri: SDNode<"AArch64ISD::ORRi", SDT_AArch64vecimm>;
252 def AArch64neg : SDNode<"AArch64ISD::NEG", SDT_AArch64unvec>;
254 def AArch64tcret: SDNode<"AArch64ISD::TC_RETURN", SDT_AArch64TCRET,
255 [SDNPHasChain, SDNPOptInGlue, SDNPVariadic]>;
257 def AArch64Prefetch : SDNode<"AArch64ISD::PREFETCH", SDT_AArch64PREFETCH,
258 [SDNPHasChain, SDNPSideEffect]>;
260 def AArch64sitof: SDNode<"AArch64ISD::SITOF", SDT_AArch64ITOF>;
261 def AArch64uitof: SDNode<"AArch64ISD::UITOF", SDT_AArch64ITOF>;
263 def AArch64tlsdesc_callseq : SDNode<"AArch64ISD::TLSDESC_CALLSEQ",
264 SDT_AArch64TLSDescCallSeq,
265 [SDNPInGlue, SDNPOutGlue, SDNPHasChain,
269 def AArch64WrapperLarge : SDNode<"AArch64ISD::WrapperLarge",
270 SDT_AArch64WrapperLarge>;
272 def AArch64NvCast : SDNode<"AArch64ISD::NVCAST", SDTUnaryOp>;
274 def SDT_AArch64mull : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
275 SDTCisSameAs<1, 2>]>;
276 def AArch64smull : SDNode<"AArch64ISD::SMULL", SDT_AArch64mull>;
277 def AArch64umull : SDNode<"AArch64ISD::UMULL", SDT_AArch64mull>;
279 def AArch64saddv : SDNode<"AArch64ISD::SADDV", SDT_AArch64UnaryVec>;
280 def AArch64uaddv : SDNode<"AArch64ISD::UADDV", SDT_AArch64UnaryVec>;
281 def AArch64sminv : SDNode<"AArch64ISD::SMINV", SDT_AArch64UnaryVec>;
282 def AArch64uminv : SDNode<"AArch64ISD::UMINV", SDT_AArch64UnaryVec>;
283 def AArch64smaxv : SDNode<"AArch64ISD::SMAXV", SDT_AArch64UnaryVec>;
284 def AArch64umaxv : SDNode<"AArch64ISD::UMAXV", SDT_AArch64UnaryVec>;
286 //===----------------------------------------------------------------------===//
288 //===----------------------------------------------------------------------===//
290 // AArch64 Instruction Predicate Definitions.
292 def HasZCZ : Predicate<"Subtarget->hasZeroCycleZeroing()">;
293 def NoZCZ : Predicate<"!Subtarget->hasZeroCycleZeroing()">;
294 def IsDarwin : Predicate<"Subtarget->isTargetDarwin()">;
295 def IsNotDarwin: Predicate<"!Subtarget->isTargetDarwin()">;
296 def ForCodeSize : Predicate<"ForCodeSize">;
297 def NotForCodeSize : Predicate<"!ForCodeSize">;
299 include "AArch64InstrFormats.td"
301 //===----------------------------------------------------------------------===//
303 //===----------------------------------------------------------------------===//
304 // Miscellaneous instructions.
305 //===----------------------------------------------------------------------===//
307 let Defs = [SP], Uses = [SP], hasSideEffects = 1, isCodeGenOnly = 1 in {
308 def ADJCALLSTACKDOWN : Pseudo<(outs), (ins i32imm:$amt),
309 [(AArch64callseq_start timm:$amt)]>;
310 def ADJCALLSTACKUP : Pseudo<(outs), (ins i32imm:$amt1, i32imm:$amt2),
311 [(AArch64callseq_end timm:$amt1, timm:$amt2)]>;
312 } // Defs = [SP], Uses = [SP], hasSideEffects = 1, isCodeGenOnly = 1
314 let isReMaterializable = 1, isCodeGenOnly = 1 in {
315 // FIXME: The following pseudo instructions are only needed because remat
316 // cannot handle multiple instructions. When that changes, they can be
317 // removed, along with the AArch64Wrapper node.
319 let AddedComplexity = 10 in
320 def LOADgot : Pseudo<(outs GPR64:$dst), (ins i64imm:$addr),
321 [(set GPR64:$dst, (AArch64LOADgot tglobaladdr:$addr))]>,
324 // The MOVaddr instruction should match only when the add is not folded
325 // into a load or store address.
327 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
328 [(set GPR64:$dst, (AArch64addlow (AArch64adrp tglobaladdr:$hi),
329 tglobaladdr:$low))]>,
330 Sched<[WriteAdrAdr]>;
332 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
333 [(set GPR64:$dst, (AArch64addlow (AArch64adrp tjumptable:$hi),
335 Sched<[WriteAdrAdr]>;
337 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
338 [(set GPR64:$dst, (AArch64addlow (AArch64adrp tconstpool:$hi),
340 Sched<[WriteAdrAdr]>;
342 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
343 [(set GPR64:$dst, (AArch64addlow (AArch64adrp tblockaddress:$hi),
344 tblockaddress:$low))]>,
345 Sched<[WriteAdrAdr]>;
347 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
348 [(set GPR64:$dst, (AArch64addlow (AArch64adrp tglobaltlsaddr:$hi),
349 tglobaltlsaddr:$low))]>,
350 Sched<[WriteAdrAdr]>;
352 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
353 [(set GPR64:$dst, (AArch64addlow (AArch64adrp texternalsym:$hi),
354 texternalsym:$low))]>,
355 Sched<[WriteAdrAdr]>;
357 } // isReMaterializable, isCodeGenOnly
359 def : Pat<(AArch64LOADgot tglobaltlsaddr:$addr),
360 (LOADgot tglobaltlsaddr:$addr)>;
362 def : Pat<(AArch64LOADgot texternalsym:$addr),
363 (LOADgot texternalsym:$addr)>;
365 def : Pat<(AArch64LOADgot tconstpool:$addr),
366 (LOADgot tconstpool:$addr)>;
368 //===----------------------------------------------------------------------===//
369 // System instructions.
370 //===----------------------------------------------------------------------===//
372 def HINT : HintI<"hint">;
373 def : InstAlias<"nop", (HINT 0b000)>;
374 def : InstAlias<"yield",(HINT 0b001)>;
375 def : InstAlias<"wfe", (HINT 0b010)>;
376 def : InstAlias<"wfi", (HINT 0b011)>;
377 def : InstAlias<"sev", (HINT 0b100)>;
378 def : InstAlias<"sevl", (HINT 0b101)>;
380 // As far as LLVM is concerned this writes to the system's exclusive monitors.
381 let mayLoad = 1, mayStore = 1 in
382 def CLREX : CRmSystemI<imm0_15, 0b010, "clrex">;
384 // NOTE: ideally, this would have mayStore = 0, mayLoad = 0, but we cannot
385 // model patterns with sufficiently fine granularity.
386 let mayLoad = ?, mayStore = ? in {
387 def DMB : CRmSystemI<barrier_op, 0b101, "dmb",
388 [(int_aarch64_dmb (i32 imm32_0_15:$CRm))]>;
390 def DSB : CRmSystemI<barrier_op, 0b100, "dsb",
391 [(int_aarch64_dsb (i32 imm32_0_15:$CRm))]>;
393 def ISB : CRmSystemI<barrier_op, 0b110, "isb",
394 [(int_aarch64_isb (i32 imm32_0_15:$CRm))]>;
397 def : InstAlias<"clrex", (CLREX 0xf)>;
398 def : InstAlias<"isb", (ISB 0xf)>;
402 def MSRpstate: MSRpstateI;
404 // The thread pointer (on Linux, at least, where this has been implemented) is
406 def : Pat<(AArch64threadpointer), (MRS 0xde82)>;
408 // The cycle counter PMC register is PMCCNTR_EL0.
409 let Predicates = [HasPerfMon] in
410 def : Pat<(readcyclecounter), (MRS 0xdce8)>;
412 // Generic system instructions
413 def SYSxt : SystemXtI<0, "sys">;
414 def SYSLxt : SystemLXtI<1, "sysl">;
416 def : InstAlias<"sys $op1, $Cn, $Cm, $op2",
417 (SYSxt imm0_7:$op1, sys_cr_op:$Cn,
418 sys_cr_op:$Cm, imm0_7:$op2, XZR)>;
420 //===----------------------------------------------------------------------===//
421 // Move immediate instructions.
422 //===----------------------------------------------------------------------===//
424 defm MOVK : InsertImmediate<0b11, "movk">;
425 defm MOVN : MoveImmediate<0b00, "movn">;
427 let PostEncoderMethod = "fixMOVZ" in
428 defm MOVZ : MoveImmediate<0b10, "movz">;
430 // First group of aliases covers an implicit "lsl #0".
431 def : InstAlias<"movk $dst, $imm", (MOVKWi GPR32:$dst, imm0_65535:$imm, 0)>;
432 def : InstAlias<"movk $dst, $imm", (MOVKXi GPR64:$dst, imm0_65535:$imm, 0)>;
433 def : InstAlias<"movn $dst, $imm", (MOVNWi GPR32:$dst, imm0_65535:$imm, 0)>;
434 def : InstAlias<"movn $dst, $imm", (MOVNXi GPR64:$dst, imm0_65535:$imm, 0)>;
435 def : InstAlias<"movz $dst, $imm", (MOVZWi GPR32:$dst, imm0_65535:$imm, 0)>;
436 def : InstAlias<"movz $dst, $imm", (MOVZXi GPR64:$dst, imm0_65535:$imm, 0)>;
438 // Next, we have various ELF relocations with the ":XYZ_g0:sym" syntax.
439 def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movz_symbol_g3:$sym, 48)>;
440 def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movz_symbol_g2:$sym, 32)>;
441 def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movz_symbol_g1:$sym, 16)>;
442 def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movz_symbol_g0:$sym, 0)>;
444 def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movz_symbol_g3:$sym, 48)>;
445 def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movz_symbol_g2:$sym, 32)>;
446 def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movz_symbol_g1:$sym, 16)>;
447 def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movz_symbol_g0:$sym, 0)>;
449 def : InstAlias<"movk $Rd, $sym", (MOVKXi GPR64:$Rd, movk_symbol_g3:$sym, 48)>;
450 def : InstAlias<"movk $Rd, $sym", (MOVKXi GPR64:$Rd, movk_symbol_g2:$sym, 32)>;
451 def : InstAlias<"movk $Rd, $sym", (MOVKXi GPR64:$Rd, movk_symbol_g1:$sym, 16)>;
452 def : InstAlias<"movk $Rd, $sym", (MOVKXi GPR64:$Rd, movk_symbol_g0:$sym, 0)>;
454 def : InstAlias<"movz $Rd, $sym", (MOVZWi GPR32:$Rd, movz_symbol_g1:$sym, 16)>;
455 def : InstAlias<"movz $Rd, $sym", (MOVZWi GPR32:$Rd, movz_symbol_g0:$sym, 0)>;
457 def : InstAlias<"movn $Rd, $sym", (MOVNWi GPR32:$Rd, movz_symbol_g1:$sym, 16)>;
458 def : InstAlias<"movn $Rd, $sym", (MOVNWi GPR32:$Rd, movz_symbol_g0:$sym, 0)>;
460 def : InstAlias<"movk $Rd, $sym", (MOVKWi GPR32:$Rd, movk_symbol_g1:$sym, 16)>;
461 def : InstAlias<"movk $Rd, $sym", (MOVKWi GPR32:$Rd, movk_symbol_g0:$sym, 0)>;
463 // Final group of aliases covers true "mov $Rd, $imm" cases.
464 multiclass movw_mov_alias<string basename,Instruction INST, RegisterClass GPR,
465 int width, int shift> {
466 def _asmoperand : AsmOperandClass {
467 let Name = basename # width # "_lsl" # shift # "MovAlias";
468 let PredicateMethod = "is" # basename # "MovAlias<" # width # ", "
470 let RenderMethod = "add" # basename # "MovAliasOperands<" # shift # ">";
473 def _movimm : Operand<i32> {
474 let ParserMatchClass = !cast<AsmOperandClass>(NAME # "_asmoperand");
477 def : InstAlias<"mov $Rd, $imm",
478 (INST GPR:$Rd, !cast<Operand>(NAME # "_movimm"):$imm, shift)>;
481 defm : movw_mov_alias<"MOVZ", MOVZWi, GPR32, 32, 0>;
482 defm : movw_mov_alias<"MOVZ", MOVZWi, GPR32, 32, 16>;
484 defm : movw_mov_alias<"MOVZ", MOVZXi, GPR64, 64, 0>;
485 defm : movw_mov_alias<"MOVZ", MOVZXi, GPR64, 64, 16>;
486 defm : movw_mov_alias<"MOVZ", MOVZXi, GPR64, 64, 32>;
487 defm : movw_mov_alias<"MOVZ", MOVZXi, GPR64, 64, 48>;
489 defm : movw_mov_alias<"MOVN", MOVNWi, GPR32, 32, 0>;
490 defm : movw_mov_alias<"MOVN", MOVNWi, GPR32, 32, 16>;
492 defm : movw_mov_alias<"MOVN", MOVNXi, GPR64, 64, 0>;
493 defm : movw_mov_alias<"MOVN", MOVNXi, GPR64, 64, 16>;
494 defm : movw_mov_alias<"MOVN", MOVNXi, GPR64, 64, 32>;
495 defm : movw_mov_alias<"MOVN", MOVNXi, GPR64, 64, 48>;
497 let isReMaterializable = 1, isCodeGenOnly = 1, isMoveImm = 1,
498 isAsCheapAsAMove = 1 in {
499 // FIXME: The following pseudo instructions are only needed because remat
500 // cannot handle multiple instructions. When that changes, we can select
501 // directly to the real instructions and get rid of these pseudos.
504 : Pseudo<(outs GPR32:$dst), (ins i32imm:$src),
505 [(set GPR32:$dst, imm:$src)]>,
508 : Pseudo<(outs GPR64:$dst), (ins i64imm:$src),
509 [(set GPR64:$dst, imm:$src)]>,
511 } // isReMaterializable, isCodeGenOnly
513 // If possible, we want to use MOVi32imm even for 64-bit moves. This gives the
514 // eventual expansion code fewer bits to worry about getting right. Marshalling
515 // the types is a little tricky though:
516 def i64imm_32bit : ImmLeaf<i64, [{
517 return (Imm & 0xffffffffULL) == static_cast<uint64_t>(Imm);
520 def trunc_imm : SDNodeXForm<imm, [{
521 return CurDAG->getTargetConstant(N->getZExtValue(), SDLoc(N), MVT::i32);
524 def : Pat<(i64 i64imm_32bit:$src),
525 (SUBREG_TO_REG (i64 0), (MOVi32imm (trunc_imm imm:$src)), sub_32)>;
527 // Materialize FP constants via MOVi32imm/MOVi64imm (MachO large code model).
528 def bitcast_fpimm_to_i32 : SDNodeXForm<fpimm, [{
529 return CurDAG->getTargetConstant(
530 N->getValueAPF().bitcastToAPInt().getZExtValue(), SDLoc(N), MVT::i32);
533 def bitcast_fpimm_to_i64 : SDNodeXForm<fpimm, [{
534 return CurDAG->getTargetConstant(
535 N->getValueAPF().bitcastToAPInt().getZExtValue(), SDLoc(N), MVT::i64);
539 def : Pat<(f32 fpimm:$in),
540 (COPY_TO_REGCLASS (MOVi32imm (bitcast_fpimm_to_i32 f32:$in)), FPR32)>;
541 def : Pat<(f64 fpimm:$in),
542 (COPY_TO_REGCLASS (MOVi64imm (bitcast_fpimm_to_i64 f64:$in)), FPR64)>;
545 // Deal with the various forms of (ELF) large addressing with MOVZ/MOVK
547 def : Pat<(AArch64WrapperLarge tglobaladdr:$g3, tglobaladdr:$g2,
548 tglobaladdr:$g1, tglobaladdr:$g0),
549 (MOVKXi (MOVKXi (MOVKXi (MOVZXi tglobaladdr:$g3, 48),
550 tglobaladdr:$g2, 32),
551 tglobaladdr:$g1, 16),
552 tglobaladdr:$g0, 0)>;
554 def : Pat<(AArch64WrapperLarge tblockaddress:$g3, tblockaddress:$g2,
555 tblockaddress:$g1, tblockaddress:$g0),
556 (MOVKXi (MOVKXi (MOVKXi (MOVZXi tblockaddress:$g3, 48),
557 tblockaddress:$g2, 32),
558 tblockaddress:$g1, 16),
559 tblockaddress:$g0, 0)>;
561 def : Pat<(AArch64WrapperLarge tconstpool:$g3, tconstpool:$g2,
562 tconstpool:$g1, tconstpool:$g0),
563 (MOVKXi (MOVKXi (MOVKXi (MOVZXi tconstpool:$g3, 48),
568 def : Pat<(AArch64WrapperLarge tjumptable:$g3, tjumptable:$g2,
569 tjumptable:$g1, tjumptable:$g0),
570 (MOVKXi (MOVKXi (MOVKXi (MOVZXi tjumptable:$g3, 48),
576 //===----------------------------------------------------------------------===//
577 // Arithmetic instructions.
578 //===----------------------------------------------------------------------===//
580 // Add/subtract with carry.
581 defm ADC : AddSubCarry<0, "adc", "adcs", AArch64adc, AArch64adc_flag>;
582 defm SBC : AddSubCarry<1, "sbc", "sbcs", AArch64sbc, AArch64sbc_flag>;
584 def : InstAlias<"ngc $dst, $src", (SBCWr GPR32:$dst, WZR, GPR32:$src)>;
585 def : InstAlias<"ngc $dst, $src", (SBCXr GPR64:$dst, XZR, GPR64:$src)>;
586 def : InstAlias<"ngcs $dst, $src", (SBCSWr GPR32:$dst, WZR, GPR32:$src)>;
587 def : InstAlias<"ngcs $dst, $src", (SBCSXr GPR64:$dst, XZR, GPR64:$src)>;
590 defm ADD : AddSub<0, "add", "sub", add>;
591 defm SUB : AddSub<1, "sub", "add">;
593 def : InstAlias<"mov $dst, $src",
594 (ADDWri GPR32sponly:$dst, GPR32sp:$src, 0, 0)>;
595 def : InstAlias<"mov $dst, $src",
596 (ADDWri GPR32sp:$dst, GPR32sponly:$src, 0, 0)>;
597 def : InstAlias<"mov $dst, $src",
598 (ADDXri GPR64sponly:$dst, GPR64sp:$src, 0, 0)>;
599 def : InstAlias<"mov $dst, $src",
600 (ADDXri GPR64sp:$dst, GPR64sponly:$src, 0, 0)>;
602 defm ADDS : AddSubS<0, "adds", AArch64add_flag, "cmn", "subs", "cmp">;
603 defm SUBS : AddSubS<1, "subs", AArch64sub_flag, "cmp", "adds", "cmn">;
605 // Use SUBS instead of SUB to enable CSE between SUBS and SUB.
606 def : Pat<(sub GPR32sp:$Rn, addsub_shifted_imm32:$imm),
607 (SUBSWri GPR32sp:$Rn, addsub_shifted_imm32:$imm)>;
608 def : Pat<(sub GPR64sp:$Rn, addsub_shifted_imm64:$imm),
609 (SUBSXri GPR64sp:$Rn, addsub_shifted_imm64:$imm)>;
610 def : Pat<(sub GPR32:$Rn, GPR32:$Rm),
611 (SUBSWrr GPR32:$Rn, GPR32:$Rm)>;
612 def : Pat<(sub GPR64:$Rn, GPR64:$Rm),
613 (SUBSXrr GPR64:$Rn, GPR64:$Rm)>;
614 def : Pat<(sub GPR32:$Rn, arith_shifted_reg32:$Rm),
615 (SUBSWrs GPR32:$Rn, arith_shifted_reg32:$Rm)>;
616 def : Pat<(sub GPR64:$Rn, arith_shifted_reg64:$Rm),
617 (SUBSXrs GPR64:$Rn, arith_shifted_reg64:$Rm)>;
618 let AddedComplexity = 1 in {
619 def : Pat<(sub GPR32sp:$R2, arith_extended_reg32<i32>:$R3),
620 (SUBSWrx GPR32sp:$R2, arith_extended_reg32<i32>:$R3)>;
621 def : Pat<(sub GPR64sp:$R2, arith_extended_reg32to64<i64>:$R3),
622 (SUBSXrx GPR64sp:$R2, arith_extended_reg32to64<i64>:$R3)>;
625 // Because of the immediate format for add/sub-imm instructions, the
626 // expression (add x, -1) must be transformed to (SUB{W,X}ri x, 1).
627 // These patterns capture that transformation.
628 let AddedComplexity = 1 in {
629 def : Pat<(add GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
630 (SUBSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
631 def : Pat<(add GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
632 (SUBSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
633 def : Pat<(sub GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
634 (ADDWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
635 def : Pat<(sub GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
636 (ADDXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
639 // Because of the immediate format for add/sub-imm instructions, the
640 // expression (add x, -1) must be transformed to (SUB{W,X}ri x, 1).
641 // These patterns capture that transformation.
642 let AddedComplexity = 1 in {
643 def : Pat<(AArch64add_flag GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
644 (SUBSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
645 def : Pat<(AArch64add_flag GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
646 (SUBSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
647 def : Pat<(AArch64sub_flag GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
648 (ADDSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
649 def : Pat<(AArch64sub_flag GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
650 (ADDSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
653 def : InstAlias<"neg $dst, $src", (SUBWrs GPR32:$dst, WZR, GPR32:$src, 0), 3>;
654 def : InstAlias<"neg $dst, $src", (SUBXrs GPR64:$dst, XZR, GPR64:$src, 0), 3>;
655 def : InstAlias<"neg $dst, $src$shift",
656 (SUBWrs GPR32:$dst, WZR, GPR32:$src, arith_shift32:$shift), 2>;
657 def : InstAlias<"neg $dst, $src$shift",
658 (SUBXrs GPR64:$dst, XZR, GPR64:$src, arith_shift64:$shift), 2>;
660 def : InstAlias<"negs $dst, $src", (SUBSWrs GPR32:$dst, WZR, GPR32:$src, 0), 3>;
661 def : InstAlias<"negs $dst, $src", (SUBSXrs GPR64:$dst, XZR, GPR64:$src, 0), 3>;
662 def : InstAlias<"negs $dst, $src$shift",
663 (SUBSWrs GPR32:$dst, WZR, GPR32:$src, arith_shift32:$shift), 2>;
664 def : InstAlias<"negs $dst, $src$shift",
665 (SUBSXrs GPR64:$dst, XZR, GPR64:$src, arith_shift64:$shift), 2>;
668 // Unsigned/Signed divide
669 defm UDIV : Div<0, "udiv", udiv>;
670 defm SDIV : Div<1, "sdiv", sdiv>;
671 let isCodeGenOnly = 1 in {
672 defm UDIV_Int : Div<0, "udiv", int_aarch64_udiv>;
673 defm SDIV_Int : Div<1, "sdiv", int_aarch64_sdiv>;
677 defm ASRV : Shift<0b10, "asr", sra>;
678 defm LSLV : Shift<0b00, "lsl", shl>;
679 defm LSRV : Shift<0b01, "lsr", srl>;
680 defm RORV : Shift<0b11, "ror", rotr>;
682 def : ShiftAlias<"asrv", ASRVWr, GPR32>;
683 def : ShiftAlias<"asrv", ASRVXr, GPR64>;
684 def : ShiftAlias<"lslv", LSLVWr, GPR32>;
685 def : ShiftAlias<"lslv", LSLVXr, GPR64>;
686 def : ShiftAlias<"lsrv", LSRVWr, GPR32>;
687 def : ShiftAlias<"lsrv", LSRVXr, GPR64>;
688 def : ShiftAlias<"rorv", RORVWr, GPR32>;
689 def : ShiftAlias<"rorv", RORVXr, GPR64>;
692 let AddedComplexity = 7 in {
693 defm MADD : MulAccum<0, "madd", add>;
694 defm MSUB : MulAccum<1, "msub", sub>;
696 def : Pat<(i32 (mul GPR32:$Rn, GPR32:$Rm)),
697 (MADDWrrr GPR32:$Rn, GPR32:$Rm, WZR)>;
698 def : Pat<(i64 (mul GPR64:$Rn, GPR64:$Rm)),
699 (MADDXrrr GPR64:$Rn, GPR64:$Rm, XZR)>;
701 def : Pat<(i32 (ineg (mul GPR32:$Rn, GPR32:$Rm))),
702 (MSUBWrrr GPR32:$Rn, GPR32:$Rm, WZR)>;
703 def : Pat<(i64 (ineg (mul GPR64:$Rn, GPR64:$Rm))),
704 (MSUBXrrr GPR64:$Rn, GPR64:$Rm, XZR)>;
705 def : Pat<(i32 (mul (ineg GPR32:$Rn), GPR32:$Rm)),
706 (MSUBWrrr GPR32:$Rn, GPR32:$Rm, WZR)>;
707 def : Pat<(i64 (mul (ineg GPR64:$Rn), GPR64:$Rm)),
708 (MSUBXrrr GPR64:$Rn, GPR64:$Rm, XZR)>;
709 } // AddedComplexity = 7
711 let AddedComplexity = 5 in {
712 def SMADDLrrr : WideMulAccum<0, 0b001, "smaddl", add, sext>;
713 def SMSUBLrrr : WideMulAccum<1, 0b001, "smsubl", sub, sext>;
714 def UMADDLrrr : WideMulAccum<0, 0b101, "umaddl", add, zext>;
715 def UMSUBLrrr : WideMulAccum<1, 0b101, "umsubl", sub, zext>;
717 def : Pat<(i64 (mul (sext GPR32:$Rn), (sext GPR32:$Rm))),
718 (SMADDLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
719 def : Pat<(i64 (mul (zext GPR32:$Rn), (zext GPR32:$Rm))),
720 (UMADDLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
722 def : Pat<(i64 (ineg (mul (sext GPR32:$Rn), (sext GPR32:$Rm)))),
723 (SMSUBLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
724 def : Pat<(i64 (ineg (mul (zext GPR32:$Rn), (zext GPR32:$Rm)))),
725 (UMSUBLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
726 } // AddedComplexity = 5
728 def : MulAccumWAlias<"mul", MADDWrrr>;
729 def : MulAccumXAlias<"mul", MADDXrrr>;
730 def : MulAccumWAlias<"mneg", MSUBWrrr>;
731 def : MulAccumXAlias<"mneg", MSUBXrrr>;
732 def : WideMulAccumAlias<"smull", SMADDLrrr>;
733 def : WideMulAccumAlias<"smnegl", SMSUBLrrr>;
734 def : WideMulAccumAlias<"umull", UMADDLrrr>;
735 def : WideMulAccumAlias<"umnegl", UMSUBLrrr>;
738 def SMULHrr : MulHi<0b010, "smulh", mulhs>;
739 def UMULHrr : MulHi<0b110, "umulh", mulhu>;
742 def CRC32Brr : BaseCRC32<0, 0b00, 0, GPR32, int_aarch64_crc32b, "crc32b">;
743 def CRC32Hrr : BaseCRC32<0, 0b01, 0, GPR32, int_aarch64_crc32h, "crc32h">;
744 def CRC32Wrr : BaseCRC32<0, 0b10, 0, GPR32, int_aarch64_crc32w, "crc32w">;
745 def CRC32Xrr : BaseCRC32<1, 0b11, 0, GPR64, int_aarch64_crc32x, "crc32x">;
747 def CRC32CBrr : BaseCRC32<0, 0b00, 1, GPR32, int_aarch64_crc32cb, "crc32cb">;
748 def CRC32CHrr : BaseCRC32<0, 0b01, 1, GPR32, int_aarch64_crc32ch, "crc32ch">;
749 def CRC32CWrr : BaseCRC32<0, 0b10, 1, GPR32, int_aarch64_crc32cw, "crc32cw">;
750 def CRC32CXrr : BaseCRC32<1, 0b11, 1, GPR64, int_aarch64_crc32cx, "crc32cx">;
753 defm CAS : CompareAndSwap<0, 0, "">;
754 defm CASA : CompareAndSwap<1, 0, "a">;
755 defm CASL : CompareAndSwap<0, 1, "l">;
756 defm CASAL : CompareAndSwap<1, 1, "al">;
759 defm CASP : CompareAndSwapPair<0, 0, "">;
760 defm CASPA : CompareAndSwapPair<1, 0, "a">;
761 defm CASPL : CompareAndSwapPair<0, 1, "l">;
762 defm CASPAL : CompareAndSwapPair<1, 1, "al">;
765 defm SWP : Swap<0, 0, "">;
766 defm SWPA : Swap<1, 0, "a">;
767 defm SWPL : Swap<0, 1, "l">;
768 defm SWPAL : Swap<1, 1, "al">;
770 // v8.1 atomic LD<OP>(register). Performs load and then ST<OP>(register)
771 defm LDADD : LDOPregister<0b000, "add", 0, 0, "">;
772 defm LDADDA : LDOPregister<0b000, "add", 1, 0, "a">;
773 defm LDADDL : LDOPregister<0b000, "add", 0, 1, "l">;
774 defm LDADDAL : LDOPregister<0b000, "add", 1, 1, "al">;
776 defm LDCLR : LDOPregister<0b001, "clr", 0, 0, "">;
777 defm LDCLRA : LDOPregister<0b001, "clr", 1, 0, "a">;
778 defm LDCLRL : LDOPregister<0b001, "clr", 0, 1, "l">;
779 defm LDCLRAL : LDOPregister<0b001, "clr", 1, 1, "al">;
781 defm LDEOR : LDOPregister<0b010, "eor", 0, 0, "">;
782 defm LDEORA : LDOPregister<0b010, "eor", 1, 0, "a">;
783 defm LDEORL : LDOPregister<0b010, "eor", 0, 1, "l">;
784 defm LDEORAL : LDOPregister<0b010, "eor", 1, 1, "al">;
786 defm LDSET : LDOPregister<0b011, "set", 0, 0, "">;
787 defm LDSETA : LDOPregister<0b011, "set", 1, 0, "a">;
788 defm LDSETL : LDOPregister<0b011, "set", 0, 1, "l">;
789 defm LDSETAL : LDOPregister<0b011, "set", 1, 1, "al">;
791 defm LDSMAX : LDOPregister<0b100, "smax", 0, 0, "">;
792 defm LDSMAXA : LDOPregister<0b100, "smax", 1, 0, "a">;
793 defm LDSMAXL : LDOPregister<0b100, "smax", 0, 1, "l">;
794 defm LDSMAXAL : LDOPregister<0b100, "smax", 1, 1, "al">;
796 defm LDSMIN : LDOPregister<0b101, "smin", 0, 0, "">;
797 defm LDSMINA : LDOPregister<0b101, "smin", 1, 0, "a">;
798 defm LDSMINL : LDOPregister<0b101, "smin", 0, 1, "l">;
799 defm LDSMINAL : LDOPregister<0b101, "smin", 1, 1, "al">;
801 defm LDUMAX : LDOPregister<0b110, "umax", 0, 0, "">;
802 defm LDUMAXA : LDOPregister<0b110, "umax", 1, 0, "a">;
803 defm LDUMAXL : LDOPregister<0b110, "umax", 0, 1, "l">;
804 defm LDUMAXAL : LDOPregister<0b110, "umax", 1, 1, "al">;
806 defm LDUMIN : LDOPregister<0b111, "umin", 0, 0, "">;
807 defm LDUMINA : LDOPregister<0b111, "umin", 1, 0, "a">;
808 defm LDUMINL : LDOPregister<0b111, "umin", 0, 1, "l">;
809 defm LDUMINAL : LDOPregister<0b111, "umin", 1, 1, "al">;
811 // v8.1 atomic ST<OP>(register) as aliases to "LD<OP>(register) when Rt=xZR"
812 defm : STOPregister<"stadd","LDADD">; // STADDx
813 defm : STOPregister<"stclr","LDCLR">; // STCLRx
814 defm : STOPregister<"steor","LDEOR">; // STEORx
815 defm : STOPregister<"stset","LDSET">; // STSETx
816 defm : STOPregister<"stsmax","LDSMAX">;// STSMAXx
817 defm : STOPregister<"stsmin","LDSMIN">;// STSMINx
818 defm : STOPregister<"stumax","LDUMAX">;// STUMAXx
819 defm : STOPregister<"stumin","LDUMIN">;// STUMINx
821 //===----------------------------------------------------------------------===//
822 // Logical instructions.
823 //===----------------------------------------------------------------------===//
826 defm ANDS : LogicalImmS<0b11, "ands", AArch64and_flag, "bics">;
827 defm AND : LogicalImm<0b00, "and", and, "bic">;
828 defm EOR : LogicalImm<0b10, "eor", xor, "eon">;
829 defm ORR : LogicalImm<0b01, "orr", or, "orn">;
831 // FIXME: these aliases *are* canonical sometimes (when movz can't be
832 // used). Actually, it seems to be working right now, but putting logical_immXX
833 // here is a bit dodgy on the AsmParser side too.
834 def : InstAlias<"mov $dst, $imm", (ORRWri GPR32sp:$dst, WZR,
835 logical_imm32:$imm), 0>;
836 def : InstAlias<"mov $dst, $imm", (ORRXri GPR64sp:$dst, XZR,
837 logical_imm64:$imm), 0>;
841 defm ANDS : LogicalRegS<0b11, 0, "ands", AArch64and_flag>;
842 defm BICS : LogicalRegS<0b11, 1, "bics",
843 BinOpFrag<(AArch64and_flag node:$LHS, (not node:$RHS))>>;
844 defm AND : LogicalReg<0b00, 0, "and", and>;
845 defm BIC : LogicalReg<0b00, 1, "bic",
846 BinOpFrag<(and node:$LHS, (not node:$RHS))>>;
847 defm EON : LogicalReg<0b10, 1, "eon",
848 BinOpFrag<(not (xor node:$LHS, node:$RHS))>>;
849 defm EOR : LogicalReg<0b10, 0, "eor", xor>;
850 defm ORN : LogicalReg<0b01, 1, "orn",
851 BinOpFrag<(or node:$LHS, (not node:$RHS))>>;
852 defm ORR : LogicalReg<0b01, 0, "orr", or>;
854 def : InstAlias<"mov $dst, $src", (ORRWrs GPR32:$dst, WZR, GPR32:$src, 0), 2>;
855 def : InstAlias<"mov $dst, $src", (ORRXrs GPR64:$dst, XZR, GPR64:$src, 0), 2>;
857 def : InstAlias<"mvn $Wd, $Wm", (ORNWrs GPR32:$Wd, WZR, GPR32:$Wm, 0), 3>;
858 def : InstAlias<"mvn $Xd, $Xm", (ORNXrs GPR64:$Xd, XZR, GPR64:$Xm, 0), 3>;
860 def : InstAlias<"mvn $Wd, $Wm$sh",
861 (ORNWrs GPR32:$Wd, WZR, GPR32:$Wm, logical_shift32:$sh), 2>;
862 def : InstAlias<"mvn $Xd, $Xm$sh",
863 (ORNXrs GPR64:$Xd, XZR, GPR64:$Xm, logical_shift64:$sh), 2>;
865 def : InstAlias<"tst $src1, $src2",
866 (ANDSWri WZR, GPR32:$src1, logical_imm32:$src2), 2>;
867 def : InstAlias<"tst $src1, $src2",
868 (ANDSXri XZR, GPR64:$src1, logical_imm64:$src2), 2>;
870 def : InstAlias<"tst $src1, $src2",
871 (ANDSWrs WZR, GPR32:$src1, GPR32:$src2, 0), 3>;
872 def : InstAlias<"tst $src1, $src2",
873 (ANDSXrs XZR, GPR64:$src1, GPR64:$src2, 0), 3>;
875 def : InstAlias<"tst $src1, $src2$sh",
876 (ANDSWrs WZR, GPR32:$src1, GPR32:$src2, logical_shift32:$sh), 2>;
877 def : InstAlias<"tst $src1, $src2$sh",
878 (ANDSXrs XZR, GPR64:$src1, GPR64:$src2, logical_shift64:$sh), 2>;
881 def : Pat<(not GPR32:$Wm), (ORNWrr WZR, GPR32:$Wm)>;
882 def : Pat<(not GPR64:$Xm), (ORNXrr XZR, GPR64:$Xm)>;
885 //===----------------------------------------------------------------------===//
886 // One operand data processing instructions.
887 //===----------------------------------------------------------------------===//
889 defm CLS : OneOperandData<0b101, "cls">;
890 defm CLZ : OneOperandData<0b100, "clz", ctlz>;
891 defm RBIT : OneOperandData<0b000, "rbit">;
893 def : Pat<(int_aarch64_rbit GPR32:$Rn), (RBITWr $Rn)>;
894 def : Pat<(int_aarch64_rbit GPR64:$Rn), (RBITXr $Rn)>;
896 def REV16Wr : OneWRegData<0b001, "rev16",
897 UnOpFrag<(rotr (bswap node:$LHS), (i64 16))>>;
898 def REV16Xr : OneXRegData<0b001, "rev16", null_frag>;
900 def : Pat<(cttz GPR32:$Rn),
901 (CLZWr (RBITWr GPR32:$Rn))>;
902 def : Pat<(cttz GPR64:$Rn),
903 (CLZXr (RBITXr GPR64:$Rn))>;
904 def : Pat<(ctlz (or (shl (xor (sra GPR32:$Rn, (i64 31)), GPR32:$Rn), (i64 1)),
907 def : Pat<(ctlz (or (shl (xor (sra GPR64:$Rn, (i64 63)), GPR64:$Rn), (i64 1)),
911 // Unlike the other one operand instructions, the instructions with the "rev"
912 // mnemonic do *not* just different in the size bit, but actually use different
913 // opcode bits for the different sizes.
914 def REVWr : OneWRegData<0b010, "rev", bswap>;
915 def REVXr : OneXRegData<0b011, "rev", bswap>;
916 def REV32Xr : OneXRegData<0b010, "rev32",
917 UnOpFrag<(rotr (bswap node:$LHS), (i64 32))>>;
919 def : InstAlias<"rev64 $Rd, $Rn", (REVXr GPR64:$Rd, GPR64:$Rn), 0>;
921 // The bswap commutes with the rotr so we want a pattern for both possible
923 def : Pat<(bswap (rotr GPR32:$Rn, (i64 16))), (REV16Wr GPR32:$Rn)>;
924 def : Pat<(bswap (rotr GPR64:$Rn, (i64 32))), (REV32Xr GPR64:$Rn)>;
926 //===----------------------------------------------------------------------===//
927 // Bitfield immediate extraction instruction.
928 //===----------------------------------------------------------------------===//
929 let hasSideEffects = 0 in
930 defm EXTR : ExtractImm<"extr">;
931 def : InstAlias<"ror $dst, $src, $shift",
932 (EXTRWrri GPR32:$dst, GPR32:$src, GPR32:$src, imm0_31:$shift)>;
933 def : InstAlias<"ror $dst, $src, $shift",
934 (EXTRXrri GPR64:$dst, GPR64:$src, GPR64:$src, imm0_63:$shift)>;
936 def : Pat<(rotr GPR32:$Rn, (i64 imm0_31:$imm)),
937 (EXTRWrri GPR32:$Rn, GPR32:$Rn, imm0_31:$imm)>;
938 def : Pat<(rotr GPR64:$Rn, (i64 imm0_63:$imm)),
939 (EXTRXrri GPR64:$Rn, GPR64:$Rn, imm0_63:$imm)>;
941 //===----------------------------------------------------------------------===//
942 // Other bitfield immediate instructions.
943 //===----------------------------------------------------------------------===//
944 let hasSideEffects = 0 in {
945 defm BFM : BitfieldImmWith2RegArgs<0b01, "bfm">;
946 defm SBFM : BitfieldImm<0b00, "sbfm">;
947 defm UBFM : BitfieldImm<0b10, "ubfm">;
950 def i32shift_a : Operand<i64>, SDNodeXForm<imm, [{
951 uint64_t enc = (32 - N->getZExtValue()) & 0x1f;
952 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
955 def i32shift_b : Operand<i64>, SDNodeXForm<imm, [{
956 uint64_t enc = 31 - N->getZExtValue();
957 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
960 // min(7, 31 - shift_amt)
961 def i32shift_sext_i8 : Operand<i64>, SDNodeXForm<imm, [{
962 uint64_t enc = 31 - N->getZExtValue();
963 enc = enc > 7 ? 7 : enc;
964 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
967 // min(15, 31 - shift_amt)
968 def i32shift_sext_i16 : Operand<i64>, SDNodeXForm<imm, [{
969 uint64_t enc = 31 - N->getZExtValue();
970 enc = enc > 15 ? 15 : enc;
971 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
974 def i64shift_a : Operand<i64>, SDNodeXForm<imm, [{
975 uint64_t enc = (64 - N->getZExtValue()) & 0x3f;
976 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
979 def i64shift_b : Operand<i64>, SDNodeXForm<imm, [{
980 uint64_t enc = 63 - N->getZExtValue();
981 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
984 // min(7, 63 - shift_amt)
985 def i64shift_sext_i8 : Operand<i64>, SDNodeXForm<imm, [{
986 uint64_t enc = 63 - N->getZExtValue();
987 enc = enc > 7 ? 7 : enc;
988 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
991 // min(15, 63 - shift_amt)
992 def i64shift_sext_i16 : Operand<i64>, SDNodeXForm<imm, [{
993 uint64_t enc = 63 - N->getZExtValue();
994 enc = enc > 15 ? 15 : enc;
995 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
998 // min(31, 63 - shift_amt)
999 def i64shift_sext_i32 : Operand<i64>, SDNodeXForm<imm, [{
1000 uint64_t enc = 63 - N->getZExtValue();
1001 enc = enc > 31 ? 31 : enc;
1002 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i64);
1005 def : Pat<(shl GPR32:$Rn, (i64 imm0_31:$imm)),
1006 (UBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
1007 (i64 (i32shift_b imm0_31:$imm)))>;
1008 def : Pat<(shl GPR64:$Rn, (i64 imm0_63:$imm)),
1009 (UBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
1010 (i64 (i64shift_b imm0_63:$imm)))>;
1012 let AddedComplexity = 10 in {
1013 def : Pat<(sra GPR32:$Rn, (i64 imm0_31:$imm)),
1014 (SBFMWri GPR32:$Rn, imm0_31:$imm, 31)>;
1015 def : Pat<(sra GPR64:$Rn, (i64 imm0_63:$imm)),
1016 (SBFMXri GPR64:$Rn, imm0_63:$imm, 63)>;
1019 def : InstAlias<"asr $dst, $src, $shift",
1020 (SBFMWri GPR32:$dst, GPR32:$src, imm0_31:$shift, 31)>;
1021 def : InstAlias<"asr $dst, $src, $shift",
1022 (SBFMXri GPR64:$dst, GPR64:$src, imm0_63:$shift, 63)>;
1023 def : InstAlias<"sxtb $dst, $src", (SBFMWri GPR32:$dst, GPR32:$src, 0, 7)>;
1024 def : InstAlias<"sxtb $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 7)>;
1025 def : InstAlias<"sxth $dst, $src", (SBFMWri GPR32:$dst, GPR32:$src, 0, 15)>;
1026 def : InstAlias<"sxth $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 15)>;
1027 def : InstAlias<"sxtw $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 31)>;
1029 def : Pat<(srl GPR32:$Rn, (i64 imm0_31:$imm)),
1030 (UBFMWri GPR32:$Rn, imm0_31:$imm, 31)>;
1031 def : Pat<(srl GPR64:$Rn, (i64 imm0_63:$imm)),
1032 (UBFMXri GPR64:$Rn, imm0_63:$imm, 63)>;
1034 def : InstAlias<"lsr $dst, $src, $shift",
1035 (UBFMWri GPR32:$dst, GPR32:$src, imm0_31:$shift, 31)>;
1036 def : InstAlias<"lsr $dst, $src, $shift",
1037 (UBFMXri GPR64:$dst, GPR64:$src, imm0_63:$shift, 63)>;
1038 def : InstAlias<"uxtb $dst, $src", (UBFMWri GPR32:$dst, GPR32:$src, 0, 7)>;
1039 def : InstAlias<"uxtb $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 7)>;
1040 def : InstAlias<"uxth $dst, $src", (UBFMWri GPR32:$dst, GPR32:$src, 0, 15)>;
1041 def : InstAlias<"uxth $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 15)>;
1042 def : InstAlias<"uxtw $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 31)>;
1044 //===----------------------------------------------------------------------===//
1045 // Conditional comparison instructions.
1046 //===----------------------------------------------------------------------===//
1047 defm CCMN : CondComparison<0, "ccmn", AArch64ccmn>;
1048 defm CCMP : CondComparison<1, "ccmp", AArch64ccmp>;
1050 //===----------------------------------------------------------------------===//
1051 // Conditional select instructions.
1052 //===----------------------------------------------------------------------===//
1053 defm CSEL : CondSelect<0, 0b00, "csel">;
1055 def inc : PatFrag<(ops node:$in), (add node:$in, 1)>;
1056 defm CSINC : CondSelectOp<0, 0b01, "csinc", inc>;
1057 defm CSINV : CondSelectOp<1, 0b00, "csinv", not>;
1058 defm CSNEG : CondSelectOp<1, 0b01, "csneg", ineg>;
1060 def : Pat<(AArch64csinv GPR32:$tval, GPR32:$fval, (i32 imm:$cc), NZCV),
1061 (CSINVWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
1062 def : Pat<(AArch64csinv GPR64:$tval, GPR64:$fval, (i32 imm:$cc), NZCV),
1063 (CSINVXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
1064 def : Pat<(AArch64csneg GPR32:$tval, GPR32:$fval, (i32 imm:$cc), NZCV),
1065 (CSNEGWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
1066 def : Pat<(AArch64csneg GPR64:$tval, GPR64:$fval, (i32 imm:$cc), NZCV),
1067 (CSNEGXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
1068 def : Pat<(AArch64csinc GPR32:$tval, GPR32:$fval, (i32 imm:$cc), NZCV),
1069 (CSINCWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
1070 def : Pat<(AArch64csinc GPR64:$tval, GPR64:$fval, (i32 imm:$cc), NZCV),
1071 (CSINCXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
1073 def : Pat<(AArch64csel (i32 0), (i32 1), (i32 imm:$cc), NZCV),
1074 (CSINCWr WZR, WZR, (i32 imm:$cc))>;
1075 def : Pat<(AArch64csel (i64 0), (i64 1), (i32 imm:$cc), NZCV),
1076 (CSINCXr XZR, XZR, (i32 imm:$cc))>;
1077 def : Pat<(AArch64csel (i32 0), (i32 -1), (i32 imm:$cc), NZCV),
1078 (CSINVWr WZR, WZR, (i32 imm:$cc))>;
1079 def : Pat<(AArch64csel (i64 0), (i64 -1), (i32 imm:$cc), NZCV),
1080 (CSINVXr XZR, XZR, (i32 imm:$cc))>;
1082 // The inverse of the condition code from the alias instruction is what is used
1083 // in the aliased instruction. The parser all ready inverts the condition code
1084 // for these aliases.
1085 def : InstAlias<"cset $dst, $cc",
1086 (CSINCWr GPR32:$dst, WZR, WZR, inv_ccode:$cc)>;
1087 def : InstAlias<"cset $dst, $cc",
1088 (CSINCXr GPR64:$dst, XZR, XZR, inv_ccode:$cc)>;
1090 def : InstAlias<"csetm $dst, $cc",
1091 (CSINVWr GPR32:$dst, WZR, WZR, inv_ccode:$cc)>;
1092 def : InstAlias<"csetm $dst, $cc",
1093 (CSINVXr GPR64:$dst, XZR, XZR, inv_ccode:$cc)>;
1095 def : InstAlias<"cinc $dst, $src, $cc",
1096 (CSINCWr GPR32:$dst, GPR32:$src, GPR32:$src, inv_ccode:$cc)>;
1097 def : InstAlias<"cinc $dst, $src, $cc",
1098 (CSINCXr GPR64:$dst, GPR64:$src, GPR64:$src, inv_ccode:$cc)>;
1100 def : InstAlias<"cinv $dst, $src, $cc",
1101 (CSINVWr GPR32:$dst, GPR32:$src, GPR32:$src, inv_ccode:$cc)>;
1102 def : InstAlias<"cinv $dst, $src, $cc",
1103 (CSINVXr GPR64:$dst, GPR64:$src, GPR64:$src, inv_ccode:$cc)>;
1105 def : InstAlias<"cneg $dst, $src, $cc",
1106 (CSNEGWr GPR32:$dst, GPR32:$src, GPR32:$src, inv_ccode:$cc)>;
1107 def : InstAlias<"cneg $dst, $src, $cc",
1108 (CSNEGXr GPR64:$dst, GPR64:$src, GPR64:$src, inv_ccode:$cc)>;
1110 //===----------------------------------------------------------------------===//
1111 // PC-relative instructions.
1112 //===----------------------------------------------------------------------===//
1113 let isReMaterializable = 1 in {
1114 let hasSideEffects = 0, mayStore = 0, mayLoad = 0 in {
1115 def ADR : ADRI<0, "adr", adrlabel, []>;
1116 } // hasSideEffects = 0
1118 def ADRP : ADRI<1, "adrp", adrplabel,
1119 [(set GPR64:$Xd, (AArch64adrp tglobaladdr:$label))]>;
1120 } // isReMaterializable = 1
1122 // page address of a constant pool entry, block address
1123 def : Pat<(AArch64adrp tconstpool:$cp), (ADRP tconstpool:$cp)>;
1124 def : Pat<(AArch64adrp tblockaddress:$cp), (ADRP tblockaddress:$cp)>;
1126 //===----------------------------------------------------------------------===//
1127 // Unconditional branch (register) instructions.
1128 //===----------------------------------------------------------------------===//
1130 let isReturn = 1, isTerminator = 1, isBarrier = 1 in {
1131 def RET : BranchReg<0b0010, "ret", []>;
1132 def DRPS : SpecialReturn<0b0101, "drps">;
1133 def ERET : SpecialReturn<0b0100, "eret">;
1134 } // isReturn = 1, isTerminator = 1, isBarrier = 1
1136 // Default to the LR register.
1137 def : InstAlias<"ret", (RET LR)>;
1139 let isCall = 1, Defs = [LR], Uses = [SP] in {
1140 def BLR : BranchReg<0b0001, "blr", [(AArch64call GPR64:$Rn)]>;
1143 let isBranch = 1, isTerminator = 1, isBarrier = 1, isIndirectBranch = 1 in {
1144 def BR : BranchReg<0b0000, "br", [(brind GPR64:$Rn)]>;
1145 } // isBranch, isTerminator, isBarrier, isIndirectBranch
1147 // Create a separate pseudo-instruction for codegen to use so that we don't
1148 // flag lr as used in every function. It'll be restored before the RET by the
1149 // epilogue if it's legitimately used.
1150 def RET_ReallyLR : Pseudo<(outs), (ins), [(AArch64retflag)]> {
1151 let isTerminator = 1;
1156 // This is a directive-like pseudo-instruction. The purpose is to insert an
1157 // R_AARCH64_TLSDESC_CALL relocation at the offset of the following instruction
1158 // (which in the usual case is a BLR).
1159 let hasSideEffects = 1 in
1160 def TLSDESCCALL : Pseudo<(outs), (ins i64imm:$sym), []> {
1161 let AsmString = ".tlsdesccall $sym";
1164 // FIXME: maybe the scratch register used shouldn't be fixed to X1?
1165 // FIXME: can "hasSideEffects be dropped?
1166 let isCall = 1, Defs = [LR, X0, X1], hasSideEffects = 1,
1167 isCodeGenOnly = 1 in
1169 : Pseudo<(outs), (ins i64imm:$sym),
1170 [(AArch64tlsdesc_callseq tglobaltlsaddr:$sym)]>;
1171 def : Pat<(AArch64tlsdesc_callseq texternalsym:$sym),
1172 (TLSDESC_CALLSEQ texternalsym:$sym)>;
1174 //===----------------------------------------------------------------------===//
1175 // Conditional branch (immediate) instruction.
1176 //===----------------------------------------------------------------------===//
1177 def Bcc : BranchCond;
1179 //===----------------------------------------------------------------------===//
1180 // Compare-and-branch instructions.
1181 //===----------------------------------------------------------------------===//
1182 defm CBZ : CmpBranch<0, "cbz", AArch64cbz>;
1183 defm CBNZ : CmpBranch<1, "cbnz", AArch64cbnz>;
1185 //===----------------------------------------------------------------------===//
1186 // Test-bit-and-branch instructions.
1187 //===----------------------------------------------------------------------===//
1188 defm TBZ : TestBranch<0, "tbz", AArch64tbz>;
1189 defm TBNZ : TestBranch<1, "tbnz", AArch64tbnz>;
1191 //===----------------------------------------------------------------------===//
1192 // Unconditional branch (immediate) instructions.
1193 //===----------------------------------------------------------------------===//
1194 let isBranch = 1, isTerminator = 1, isBarrier = 1 in {
1195 def B : BranchImm<0, "b", [(br bb:$addr)]>;
1196 } // isBranch, isTerminator, isBarrier
1198 let isCall = 1, Defs = [LR], Uses = [SP] in {
1199 def BL : CallImm<1, "bl", [(AArch64call tglobaladdr:$addr)]>;
1201 def : Pat<(AArch64call texternalsym:$func), (BL texternalsym:$func)>;
1203 //===----------------------------------------------------------------------===//
1204 // Exception generation instructions.
1205 //===----------------------------------------------------------------------===//
1206 def BRK : ExceptionGeneration<0b001, 0b00, "brk">;
1207 def DCPS1 : ExceptionGeneration<0b101, 0b01, "dcps1">;
1208 def DCPS2 : ExceptionGeneration<0b101, 0b10, "dcps2">;
1209 def DCPS3 : ExceptionGeneration<0b101, 0b11, "dcps3">;
1210 def HLT : ExceptionGeneration<0b010, 0b00, "hlt">;
1211 def HVC : ExceptionGeneration<0b000, 0b10, "hvc">;
1212 def SMC : ExceptionGeneration<0b000, 0b11, "smc">;
1213 def SVC : ExceptionGeneration<0b000, 0b01, "svc">;
1215 // DCPSn defaults to an immediate operand of zero if unspecified.
1216 def : InstAlias<"dcps1", (DCPS1 0)>;
1217 def : InstAlias<"dcps2", (DCPS2 0)>;
1218 def : InstAlias<"dcps3", (DCPS3 0)>;
1220 //===----------------------------------------------------------------------===//
1221 // Load instructions.
1222 //===----------------------------------------------------------------------===//
1224 // Pair (indexed, offset)
1225 defm LDPW : LoadPairOffset<0b00, 0, GPR32, simm7s4, "ldp">;
1226 defm LDPX : LoadPairOffset<0b10, 0, GPR64, simm7s8, "ldp">;
1227 defm LDPS : LoadPairOffset<0b00, 1, FPR32, simm7s4, "ldp">;
1228 defm LDPD : LoadPairOffset<0b01, 1, FPR64, simm7s8, "ldp">;
1229 defm LDPQ : LoadPairOffset<0b10, 1, FPR128, simm7s16, "ldp">;
1231 defm LDPSW : LoadPairOffset<0b01, 0, GPR64, simm7s4, "ldpsw">;
1233 // Pair (pre-indexed)
1234 def LDPWpre : LoadPairPreIdx<0b00, 0, GPR32, simm7s4, "ldp">;
1235 def LDPXpre : LoadPairPreIdx<0b10, 0, GPR64, simm7s8, "ldp">;
1236 def LDPSpre : LoadPairPreIdx<0b00, 1, FPR32, simm7s4, "ldp">;
1237 def LDPDpre : LoadPairPreIdx<0b01, 1, FPR64, simm7s8, "ldp">;
1238 def LDPQpre : LoadPairPreIdx<0b10, 1, FPR128, simm7s16, "ldp">;
1240 def LDPSWpre : LoadPairPreIdx<0b01, 0, GPR64, simm7s4, "ldpsw">;
1242 // Pair (post-indexed)
1243 def LDPWpost : LoadPairPostIdx<0b00, 0, GPR32, simm7s4, "ldp">;
1244 def LDPXpost : LoadPairPostIdx<0b10, 0, GPR64, simm7s8, "ldp">;
1245 def LDPSpost : LoadPairPostIdx<0b00, 1, FPR32, simm7s4, "ldp">;
1246 def LDPDpost : LoadPairPostIdx<0b01, 1, FPR64, simm7s8, "ldp">;
1247 def LDPQpost : LoadPairPostIdx<0b10, 1, FPR128, simm7s16, "ldp">;
1249 def LDPSWpost : LoadPairPostIdx<0b01, 0, GPR64, simm7s4, "ldpsw">;
1252 // Pair (no allocate)
1253 defm LDNPW : LoadPairNoAlloc<0b00, 0, GPR32, simm7s4, "ldnp">;
1254 defm LDNPX : LoadPairNoAlloc<0b10, 0, GPR64, simm7s8, "ldnp">;
1255 defm LDNPS : LoadPairNoAlloc<0b00, 1, FPR32, simm7s4, "ldnp">;
1256 defm LDNPD : LoadPairNoAlloc<0b01, 1, FPR64, simm7s8, "ldnp">;
1257 defm LDNPQ : LoadPairNoAlloc<0b10, 1, FPR128, simm7s16, "ldnp">;
1260 // (register offset)
1264 defm LDRBB : Load8RO<0b00, 0, 0b01, GPR32, "ldrb", i32, zextloadi8>;
1265 defm LDRHH : Load16RO<0b01, 0, 0b01, GPR32, "ldrh", i32, zextloadi16>;
1266 defm LDRW : Load32RO<0b10, 0, 0b01, GPR32, "ldr", i32, load>;
1267 defm LDRX : Load64RO<0b11, 0, 0b01, GPR64, "ldr", i64, load>;
1270 defm LDRB : Load8RO<0b00, 1, 0b01, FPR8, "ldr", untyped, load>;
1271 defm LDRH : Load16RO<0b01, 1, 0b01, FPR16, "ldr", f16, load>;
1272 defm LDRS : Load32RO<0b10, 1, 0b01, FPR32, "ldr", f32, load>;
1273 defm LDRD : Load64RO<0b11, 1, 0b01, FPR64, "ldr", f64, load>;
1274 defm LDRQ : Load128RO<0b00, 1, 0b11, FPR128, "ldr", f128, load>;
1276 // Load sign-extended half-word
1277 defm LDRSHW : Load16RO<0b01, 0, 0b11, GPR32, "ldrsh", i32, sextloadi16>;
1278 defm LDRSHX : Load16RO<0b01, 0, 0b10, GPR64, "ldrsh", i64, sextloadi16>;
1280 // Load sign-extended byte
1281 defm LDRSBW : Load8RO<0b00, 0, 0b11, GPR32, "ldrsb", i32, sextloadi8>;
1282 defm LDRSBX : Load8RO<0b00, 0, 0b10, GPR64, "ldrsb", i64, sextloadi8>;
1284 // Load sign-extended word
1285 defm LDRSW : Load32RO<0b10, 0, 0b10, GPR64, "ldrsw", i64, sextloadi32>;
1288 defm PRFM : PrefetchRO<0b11, 0, 0b10, "prfm">;
1290 // For regular load, we do not have any alignment requirement.
1291 // Thus, it is safe to directly map the vector loads with interesting
1292 // addressing modes.
1293 // FIXME: We could do the same for bitconvert to floating point vectors.
1294 multiclass ScalToVecROLoadPat<ROAddrMode ro, SDPatternOperator loadop,
1295 ValueType ScalTy, ValueType VecTy,
1296 Instruction LOADW, Instruction LOADX,
1298 def : Pat<(VecTy (scalar_to_vector (ScalTy
1299 (loadop (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$offset))))),
1300 (INSERT_SUBREG (VecTy (IMPLICIT_DEF)),
1301 (LOADW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$offset),
1304 def : Pat<(VecTy (scalar_to_vector (ScalTy
1305 (loadop (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$offset))))),
1306 (INSERT_SUBREG (VecTy (IMPLICIT_DEF)),
1307 (LOADX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$offset),
1311 let AddedComplexity = 10 in {
1312 defm : ScalToVecROLoadPat<ro8, extloadi8, i32, v8i8, LDRBroW, LDRBroX, bsub>;
1313 defm : ScalToVecROLoadPat<ro8, extloadi8, i32, v16i8, LDRBroW, LDRBroX, bsub>;
1315 defm : ScalToVecROLoadPat<ro16, extloadi16, i32, v4i16, LDRHroW, LDRHroX, hsub>;
1316 defm : ScalToVecROLoadPat<ro16, extloadi16, i32, v8i16, LDRHroW, LDRHroX, hsub>;
1318 defm : ScalToVecROLoadPat<ro16, load, i32, v4f16, LDRHroW, LDRHroX, hsub>;
1319 defm : ScalToVecROLoadPat<ro16, load, i32, v8f16, LDRHroW, LDRHroX, hsub>;
1321 defm : ScalToVecROLoadPat<ro32, load, i32, v2i32, LDRSroW, LDRSroX, ssub>;
1322 defm : ScalToVecROLoadPat<ro32, load, i32, v4i32, LDRSroW, LDRSroX, ssub>;
1324 defm : ScalToVecROLoadPat<ro32, load, f32, v2f32, LDRSroW, LDRSroX, ssub>;
1325 defm : ScalToVecROLoadPat<ro32, load, f32, v4f32, LDRSroW, LDRSroX, ssub>;
1327 defm : ScalToVecROLoadPat<ro64, load, i64, v2i64, LDRDroW, LDRDroX, dsub>;
1329 defm : ScalToVecROLoadPat<ro64, load, f64, v2f64, LDRDroW, LDRDroX, dsub>;
1332 def : Pat <(v1i64 (scalar_to_vector (i64
1333 (load (ro_Windexed64 GPR64sp:$Rn, GPR32:$Rm,
1334 ro_Wextend64:$extend))))),
1335 (LDRDroW GPR64sp:$Rn, GPR32:$Rm, ro_Wextend64:$extend)>;
1337 def : Pat <(v1i64 (scalar_to_vector (i64
1338 (load (ro_Xindexed64 GPR64sp:$Rn, GPR64:$Rm,
1339 ro_Xextend64:$extend))))),
1340 (LDRDroX GPR64sp:$Rn, GPR64:$Rm, ro_Xextend64:$extend)>;
1343 // Match all load 64 bits width whose type is compatible with FPR64
1344 multiclass VecROLoadPat<ROAddrMode ro, ValueType VecTy,
1345 Instruction LOADW, Instruction LOADX> {
1347 def : Pat<(VecTy (load (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend))),
1348 (LOADW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
1350 def : Pat<(VecTy (load (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend))),
1351 (LOADX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
1354 let AddedComplexity = 10 in {
1355 let Predicates = [IsLE] in {
1356 // We must do vector loads with LD1 in big-endian.
1357 defm : VecROLoadPat<ro64, v2i32, LDRDroW, LDRDroX>;
1358 defm : VecROLoadPat<ro64, v2f32, LDRDroW, LDRDroX>;
1359 defm : VecROLoadPat<ro64, v8i8, LDRDroW, LDRDroX>;
1360 defm : VecROLoadPat<ro64, v4i16, LDRDroW, LDRDroX>;
1361 defm : VecROLoadPat<ro64, v4f16, LDRDroW, LDRDroX>;
1364 defm : VecROLoadPat<ro64, v1i64, LDRDroW, LDRDroX>;
1365 defm : VecROLoadPat<ro64, v1f64, LDRDroW, LDRDroX>;
1367 // Match all load 128 bits width whose type is compatible with FPR128
1368 let Predicates = [IsLE] in {
1369 // We must do vector loads with LD1 in big-endian.
1370 defm : VecROLoadPat<ro128, v2i64, LDRQroW, LDRQroX>;
1371 defm : VecROLoadPat<ro128, v2f64, LDRQroW, LDRQroX>;
1372 defm : VecROLoadPat<ro128, v4i32, LDRQroW, LDRQroX>;
1373 defm : VecROLoadPat<ro128, v4f32, LDRQroW, LDRQroX>;
1374 defm : VecROLoadPat<ro128, v8i16, LDRQroW, LDRQroX>;
1375 defm : VecROLoadPat<ro128, v8f16, LDRQroW, LDRQroX>;
1376 defm : VecROLoadPat<ro128, v16i8, LDRQroW, LDRQroX>;
1378 } // AddedComplexity = 10
1381 multiclass ExtLoadTo64ROPat<ROAddrMode ro, SDPatternOperator loadop,
1382 Instruction INSTW, Instruction INSTX> {
1383 def : Pat<(i64 (loadop (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend))),
1384 (SUBREG_TO_REG (i64 0),
1385 (INSTW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend),
1388 def : Pat<(i64 (loadop (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend))),
1389 (SUBREG_TO_REG (i64 0),
1390 (INSTX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend),
1394 let AddedComplexity = 10 in {
1395 defm : ExtLoadTo64ROPat<ro8, zextloadi8, LDRBBroW, LDRBBroX>;
1396 defm : ExtLoadTo64ROPat<ro16, zextloadi16, LDRHHroW, LDRHHroX>;
1397 defm : ExtLoadTo64ROPat<ro32, zextloadi32, LDRWroW, LDRWroX>;
1399 // zextloadi1 -> zextloadi8
1400 defm : ExtLoadTo64ROPat<ro8, zextloadi1, LDRBBroW, LDRBBroX>;
1402 // extload -> zextload
1403 defm : ExtLoadTo64ROPat<ro8, extloadi8, LDRBBroW, LDRBBroX>;
1404 defm : ExtLoadTo64ROPat<ro16, extloadi16, LDRHHroW, LDRHHroX>;
1405 defm : ExtLoadTo64ROPat<ro32, extloadi32, LDRWroW, LDRWroX>;
1407 // extloadi1 -> zextloadi8
1408 defm : ExtLoadTo64ROPat<ro8, extloadi1, LDRBBroW, LDRBBroX>;
1413 multiclass ExtLoadTo32ROPat<ROAddrMode ro, SDPatternOperator loadop,
1414 Instruction INSTW, Instruction INSTX> {
1415 def : Pat<(i32 (loadop (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend))),
1416 (INSTW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
1418 def : Pat<(i32 (loadop (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend))),
1419 (INSTX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
1423 let AddedComplexity = 10 in {
1424 // extload -> zextload
1425 defm : ExtLoadTo32ROPat<ro8, extloadi8, LDRBBroW, LDRBBroX>;
1426 defm : ExtLoadTo32ROPat<ro16, extloadi16, LDRHHroW, LDRHHroX>;
1427 defm : ExtLoadTo32ROPat<ro32, extloadi32, LDRWroW, LDRWroX>;
1429 // zextloadi1 -> zextloadi8
1430 defm : ExtLoadTo32ROPat<ro8, zextloadi1, LDRBBroW, LDRBBroX>;
1434 // (unsigned immediate)
1436 defm LDRX : LoadUI<0b11, 0, 0b01, GPR64, uimm12s8, "ldr",
1438 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)))]>;
1439 defm LDRW : LoadUI<0b10, 0, 0b01, GPR32, uimm12s4, "ldr",
1441 (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset)))]>;
1442 defm LDRB : LoadUI<0b00, 1, 0b01, FPR8, uimm12s1, "ldr",
1444 (load (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset)))]>;
1445 defm LDRH : LoadUI<0b01, 1, 0b01, FPR16, uimm12s2, "ldr",
1446 [(set (f16 FPR16:$Rt),
1447 (load (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset)))]>;
1448 defm LDRS : LoadUI<0b10, 1, 0b01, FPR32, uimm12s4, "ldr",
1449 [(set (f32 FPR32:$Rt),
1450 (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset)))]>;
1451 defm LDRD : LoadUI<0b11, 1, 0b01, FPR64, uimm12s8, "ldr",
1452 [(set (f64 FPR64:$Rt),
1453 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)))]>;
1454 defm LDRQ : LoadUI<0b00, 1, 0b11, FPR128, uimm12s16, "ldr",
1455 [(set (f128 FPR128:$Rt),
1456 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)))]>;
1458 // For regular load, we do not have any alignment requirement.
1459 // Thus, it is safe to directly map the vector loads with interesting
1460 // addressing modes.
1461 // FIXME: We could do the same for bitconvert to floating point vectors.
1462 def : Pat <(v8i8 (scalar_to_vector (i32
1463 (extloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))))),
1464 (INSERT_SUBREG (v8i8 (IMPLICIT_DEF)),
1465 (LDRBui GPR64sp:$Rn, uimm12s1:$offset), bsub)>;
1466 def : Pat <(v16i8 (scalar_to_vector (i32
1467 (extloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))))),
1468 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
1469 (LDRBui GPR64sp:$Rn, uimm12s1:$offset), bsub)>;
1470 def : Pat <(v4i16 (scalar_to_vector (i32
1471 (extloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))))),
1472 (INSERT_SUBREG (v4i16 (IMPLICIT_DEF)),
1473 (LDRHui GPR64sp:$Rn, uimm12s2:$offset), hsub)>;
1474 def : Pat <(v8i16 (scalar_to_vector (i32
1475 (extloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))))),
1476 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
1477 (LDRHui GPR64sp:$Rn, uimm12s2:$offset), hsub)>;
1478 def : Pat <(v2i32 (scalar_to_vector (i32
1479 (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))))),
1480 (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)),
1481 (LDRSui GPR64sp:$Rn, uimm12s4:$offset), ssub)>;
1482 def : Pat <(v4i32 (scalar_to_vector (i32
1483 (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))))),
1484 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
1485 (LDRSui GPR64sp:$Rn, uimm12s4:$offset), ssub)>;
1486 def : Pat <(v1i64 (scalar_to_vector (i64
1487 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))))),
1488 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1489 def : Pat <(v2i64 (scalar_to_vector (i64
1490 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))))),
1491 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
1492 (LDRDui GPR64sp:$Rn, uimm12s8:$offset), dsub)>;
1494 // Match all load 64 bits width whose type is compatible with FPR64
1495 let Predicates = [IsLE] in {
1496 // We must use LD1 to perform vector loads in big-endian.
1497 def : Pat<(v2f32 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1498 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1499 def : Pat<(v8i8 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1500 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1501 def : Pat<(v4i16 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1502 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1503 def : Pat<(v2i32 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1504 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1505 def : Pat<(v4f16 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1506 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1508 def : Pat<(v1f64 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1509 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1510 def : Pat<(v1i64 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1511 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1513 // Match all load 128 bits width whose type is compatible with FPR128
1514 let Predicates = [IsLE] in {
1515 // We must use LD1 to perform vector loads in big-endian.
1516 def : Pat<(v4f32 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1517 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1518 def : Pat<(v2f64 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1519 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1520 def : Pat<(v16i8 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1521 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1522 def : Pat<(v8i16 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1523 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1524 def : Pat<(v4i32 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1525 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1526 def : Pat<(v2i64 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1527 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1528 def : Pat<(v8f16 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1529 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1531 def : Pat<(f128 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1532 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1534 defm LDRHH : LoadUI<0b01, 0, 0b01, GPR32, uimm12s2, "ldrh",
1536 (zextloadi16 (am_indexed16 GPR64sp:$Rn,
1537 uimm12s2:$offset)))]>;
1538 defm LDRBB : LoadUI<0b00, 0, 0b01, GPR32, uimm12s1, "ldrb",
1540 (zextloadi8 (am_indexed8 GPR64sp:$Rn,
1541 uimm12s1:$offset)))]>;
1543 def : Pat<(i64 (zextloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1544 (SUBREG_TO_REG (i64 0), (LDRBBui GPR64sp:$Rn, uimm12s1:$offset), sub_32)>;
1545 def : Pat<(i64 (zextloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))),
1546 (SUBREG_TO_REG (i64 0), (LDRHHui GPR64sp:$Rn, uimm12s2:$offset), sub_32)>;
1548 // zextloadi1 -> zextloadi8
1549 def : Pat<(i32 (zextloadi1 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1550 (LDRBBui GPR64sp:$Rn, uimm12s1:$offset)>;
1551 def : Pat<(i64 (zextloadi1 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1552 (SUBREG_TO_REG (i64 0), (LDRBBui GPR64sp:$Rn, uimm12s1:$offset), sub_32)>;
1554 // extload -> zextload
1555 def : Pat<(i32 (extloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))),
1556 (LDRHHui GPR64sp:$Rn, uimm12s2:$offset)>;
1557 def : Pat<(i32 (extloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1558 (LDRBBui GPR64sp:$Rn, uimm12s1:$offset)>;
1559 def : Pat<(i32 (extloadi1 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1560 (LDRBBui GPR64sp:$Rn, uimm12s1:$offset)>;
1561 def : Pat<(i64 (extloadi32 (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))),
1562 (SUBREG_TO_REG (i64 0), (LDRWui GPR64sp:$Rn, uimm12s4:$offset), sub_32)>;
1563 def : Pat<(i64 (extloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))),
1564 (SUBREG_TO_REG (i64 0), (LDRHHui GPR64sp:$Rn, uimm12s2:$offset), sub_32)>;
1565 def : Pat<(i64 (extloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1566 (SUBREG_TO_REG (i64 0), (LDRBBui GPR64sp:$Rn, uimm12s1:$offset), sub_32)>;
1567 def : Pat<(i64 (extloadi1 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1568 (SUBREG_TO_REG (i64 0), (LDRBBui GPR64sp:$Rn, uimm12s1:$offset), sub_32)>;
1570 // load sign-extended half-word
1571 defm LDRSHW : LoadUI<0b01, 0, 0b11, GPR32, uimm12s2, "ldrsh",
1573 (sextloadi16 (am_indexed16 GPR64sp:$Rn,
1574 uimm12s2:$offset)))]>;
1575 defm LDRSHX : LoadUI<0b01, 0, 0b10, GPR64, uimm12s2, "ldrsh",
1577 (sextloadi16 (am_indexed16 GPR64sp:$Rn,
1578 uimm12s2:$offset)))]>;
1580 // load sign-extended byte
1581 defm LDRSBW : LoadUI<0b00, 0, 0b11, GPR32, uimm12s1, "ldrsb",
1583 (sextloadi8 (am_indexed8 GPR64sp:$Rn,
1584 uimm12s1:$offset)))]>;
1585 defm LDRSBX : LoadUI<0b00, 0, 0b10, GPR64, uimm12s1, "ldrsb",
1587 (sextloadi8 (am_indexed8 GPR64sp:$Rn,
1588 uimm12s1:$offset)))]>;
1590 // load sign-extended word
1591 defm LDRSW : LoadUI<0b10, 0, 0b10, GPR64, uimm12s4, "ldrsw",
1593 (sextloadi32 (am_indexed32 GPR64sp:$Rn,
1594 uimm12s4:$offset)))]>;
1596 // load zero-extended word
1597 def : Pat<(i64 (zextloadi32 (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))),
1598 (SUBREG_TO_REG (i64 0), (LDRWui GPR64sp:$Rn, uimm12s4:$offset), sub_32)>;
1601 def PRFMui : PrefetchUI<0b11, 0, 0b10, "prfm",
1602 [(AArch64Prefetch imm:$Rt,
1603 (am_indexed64 GPR64sp:$Rn,
1604 uimm12s8:$offset))]>;
1606 def : InstAlias<"prfm $Rt, [$Rn]", (PRFMui prfop:$Rt, GPR64sp:$Rn, 0)>;
1610 def LDRWl : LoadLiteral<0b00, 0, GPR32, "ldr">;
1611 def LDRXl : LoadLiteral<0b01, 0, GPR64, "ldr">;
1612 def LDRSl : LoadLiteral<0b00, 1, FPR32, "ldr">;
1613 def LDRDl : LoadLiteral<0b01, 1, FPR64, "ldr">;
1614 def LDRQl : LoadLiteral<0b10, 1, FPR128, "ldr">;
1616 // load sign-extended word
1617 def LDRSWl : LoadLiteral<0b10, 0, GPR64, "ldrsw">;
1620 def PRFMl : PrefetchLiteral<0b11, 0, "prfm", []>;
1621 // [(AArch64Prefetch imm:$Rt, tglobaladdr:$label)]>;
1624 // (unscaled immediate)
1625 defm LDURX : LoadUnscaled<0b11, 0, 0b01, GPR64, "ldur",
1627 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset)))]>;
1628 defm LDURW : LoadUnscaled<0b10, 0, 0b01, GPR32, "ldur",
1630 (load (am_unscaled32 GPR64sp:$Rn, simm9:$offset)))]>;
1631 defm LDURB : LoadUnscaled<0b00, 1, 0b01, FPR8, "ldur",
1633 (load (am_unscaled8 GPR64sp:$Rn, simm9:$offset)))]>;
1634 defm LDURH : LoadUnscaled<0b01, 1, 0b01, FPR16, "ldur",
1636 (load (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
1637 defm LDURS : LoadUnscaled<0b10, 1, 0b01, FPR32, "ldur",
1638 [(set (f32 FPR32:$Rt),
1639 (load (am_unscaled32 GPR64sp:$Rn, simm9:$offset)))]>;
1640 defm LDURD : LoadUnscaled<0b11, 1, 0b01, FPR64, "ldur",
1641 [(set (f64 FPR64:$Rt),
1642 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset)))]>;
1643 defm LDURQ : LoadUnscaled<0b00, 1, 0b11, FPR128, "ldur",
1644 [(set (f128 FPR128:$Rt),
1645 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset)))]>;
1648 : LoadUnscaled<0b01, 0, 0b01, GPR32, "ldurh",
1650 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
1652 : LoadUnscaled<0b00, 0, 0b01, GPR32, "ldurb",
1654 (zextloadi8 (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
1656 // Match all load 64 bits width whose type is compatible with FPR64
1657 let Predicates = [IsLE] in {
1658 def : Pat<(v2f32 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1659 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1660 def : Pat<(v2i32 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1661 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1662 def : Pat<(v4i16 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1663 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1664 def : Pat<(v8i8 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1665 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1666 def : Pat<(v4f16 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1667 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1669 def : Pat<(v1f64 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1670 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1671 def : Pat<(v1i64 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1672 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1674 // Match all load 128 bits width whose type is compatible with FPR128
1675 let Predicates = [IsLE] in {
1676 def : Pat<(v2f64 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1677 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1678 def : Pat<(v2i64 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1679 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1680 def : Pat<(v4f32 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1681 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1682 def : Pat<(v4i32 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1683 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1684 def : Pat<(v8i16 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1685 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1686 def : Pat<(v16i8 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1687 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1688 def : Pat<(v8f16 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1689 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1693 def : Pat<(i32 (extloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
1694 (LDURHHi GPR64sp:$Rn, simm9:$offset)>;
1695 def : Pat<(i32 (extloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1696 (LDURBBi GPR64sp:$Rn, simm9:$offset)>;
1697 def : Pat<(i32 (extloadi1 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1698 (LDURBBi GPR64sp:$Rn, simm9:$offset)>;
1699 def : Pat<(i64 (extloadi32 (am_unscaled32 GPR64sp:$Rn, simm9:$offset))),
1700 (SUBREG_TO_REG (i64 0), (LDURWi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1701 def : Pat<(i64 (extloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
1702 (SUBREG_TO_REG (i64 0), (LDURHHi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1703 def : Pat<(i64 (extloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1704 (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1705 def : Pat<(i64 (extloadi1 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1706 (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1708 def : Pat<(i32 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
1709 (LDURHHi GPR64sp:$Rn, simm9:$offset)>;
1710 def : Pat<(i32 (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1711 (LDURBBi GPR64sp:$Rn, simm9:$offset)>;
1712 def : Pat<(i32 (zextloadi1 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1713 (LDURBBi GPR64sp:$Rn, simm9:$offset)>;
1714 def : Pat<(i64 (zextloadi32 (am_unscaled32 GPR64sp:$Rn, simm9:$offset))),
1715 (SUBREG_TO_REG (i64 0), (LDURWi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1716 def : Pat<(i64 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
1717 (SUBREG_TO_REG (i64 0), (LDURHHi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1718 def : Pat<(i64 (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1719 (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1720 def : Pat<(i64 (zextloadi1 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1721 (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1725 // LDR mnemonics fall back to LDUR for negative or unaligned offsets.
1727 // Define new assembler match classes as we want to only match these when
1728 // the don't otherwise match the scaled addressing mode for LDR/STR. Don't
1729 // associate a DiagnosticType either, as we want the diagnostic for the
1730 // canonical form (the scaled operand) to take precedence.
1731 class SImm9OffsetOperand<int Width> : AsmOperandClass {
1732 let Name = "SImm9OffsetFB" # Width;
1733 let PredicateMethod = "isSImm9OffsetFB<" # Width # ">";
1734 let RenderMethod = "addImmOperands";
1737 def SImm9OffsetFB8Operand : SImm9OffsetOperand<8>;
1738 def SImm9OffsetFB16Operand : SImm9OffsetOperand<16>;
1739 def SImm9OffsetFB32Operand : SImm9OffsetOperand<32>;
1740 def SImm9OffsetFB64Operand : SImm9OffsetOperand<64>;
1741 def SImm9OffsetFB128Operand : SImm9OffsetOperand<128>;
1743 def simm9_offset_fb8 : Operand<i64> {
1744 let ParserMatchClass = SImm9OffsetFB8Operand;
1746 def simm9_offset_fb16 : Operand<i64> {
1747 let ParserMatchClass = SImm9OffsetFB16Operand;
1749 def simm9_offset_fb32 : Operand<i64> {
1750 let ParserMatchClass = SImm9OffsetFB32Operand;
1752 def simm9_offset_fb64 : Operand<i64> {
1753 let ParserMatchClass = SImm9OffsetFB64Operand;
1755 def simm9_offset_fb128 : Operand<i64> {
1756 let ParserMatchClass = SImm9OffsetFB128Operand;
1759 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1760 (LDURXi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb64:$offset), 0>;
1761 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1762 (LDURWi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
1763 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1764 (LDURBi FPR8:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
1765 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1766 (LDURHi FPR16:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
1767 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1768 (LDURSi FPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
1769 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1770 (LDURDi FPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb64:$offset), 0>;
1771 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1772 (LDURQi FPR128:$Rt, GPR64sp:$Rn, simm9_offset_fb128:$offset), 0>;
1775 def : Pat<(i64 (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1776 (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1777 def : Pat<(i64 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
1778 (SUBREG_TO_REG (i64 0), (LDURHHi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1780 // load sign-extended half-word
1782 : LoadUnscaled<0b01, 0, 0b11, GPR32, "ldursh",
1784 (sextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
1786 : LoadUnscaled<0b01, 0, 0b10, GPR64, "ldursh",
1788 (sextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
1790 // load sign-extended byte
1792 : LoadUnscaled<0b00, 0, 0b11, GPR32, "ldursb",
1794 (sextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset)))]>;
1796 : LoadUnscaled<0b00, 0, 0b10, GPR64, "ldursb",
1798 (sextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset)))]>;
1800 // load sign-extended word
1802 : LoadUnscaled<0b10, 0, 0b10, GPR64, "ldursw",
1804 (sextloadi32 (am_unscaled32 GPR64sp:$Rn, simm9:$offset)))]>;
1806 // zero and sign extending aliases from generic LDR* mnemonics to LDUR*.
1807 def : InstAlias<"ldrb $Rt, [$Rn, $offset]",
1808 (LDURBBi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
1809 def : InstAlias<"ldrh $Rt, [$Rn, $offset]",
1810 (LDURHHi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
1811 def : InstAlias<"ldrsb $Rt, [$Rn, $offset]",
1812 (LDURSBWi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
1813 def : InstAlias<"ldrsb $Rt, [$Rn, $offset]",
1814 (LDURSBXi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
1815 def : InstAlias<"ldrsh $Rt, [$Rn, $offset]",
1816 (LDURSHWi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
1817 def : InstAlias<"ldrsh $Rt, [$Rn, $offset]",
1818 (LDURSHXi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
1819 def : InstAlias<"ldrsw $Rt, [$Rn, $offset]",
1820 (LDURSWi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
1823 defm PRFUM : PrefetchUnscaled<0b11, 0, 0b10, "prfum",
1824 [(AArch64Prefetch imm:$Rt,
1825 (am_unscaled64 GPR64sp:$Rn, simm9:$offset))]>;
1828 // (unscaled immediate, unprivileged)
1829 defm LDTRX : LoadUnprivileged<0b11, 0, 0b01, GPR64, "ldtr">;
1830 defm LDTRW : LoadUnprivileged<0b10, 0, 0b01, GPR32, "ldtr">;
1832 defm LDTRH : LoadUnprivileged<0b01, 0, 0b01, GPR32, "ldtrh">;
1833 defm LDTRB : LoadUnprivileged<0b00, 0, 0b01, GPR32, "ldtrb">;
1835 // load sign-extended half-word
1836 defm LDTRSHW : LoadUnprivileged<0b01, 0, 0b11, GPR32, "ldtrsh">;
1837 defm LDTRSHX : LoadUnprivileged<0b01, 0, 0b10, GPR64, "ldtrsh">;
1839 // load sign-extended byte
1840 defm LDTRSBW : LoadUnprivileged<0b00, 0, 0b11, GPR32, "ldtrsb">;
1841 defm LDTRSBX : LoadUnprivileged<0b00, 0, 0b10, GPR64, "ldtrsb">;
1843 // load sign-extended word
1844 defm LDTRSW : LoadUnprivileged<0b10, 0, 0b10, GPR64, "ldtrsw">;
1847 // (immediate pre-indexed)
1848 def LDRWpre : LoadPreIdx<0b10, 0, 0b01, GPR32, "ldr">;
1849 def LDRXpre : LoadPreIdx<0b11, 0, 0b01, GPR64, "ldr">;
1850 def LDRBpre : LoadPreIdx<0b00, 1, 0b01, FPR8, "ldr">;
1851 def LDRHpre : LoadPreIdx<0b01, 1, 0b01, FPR16, "ldr">;
1852 def LDRSpre : LoadPreIdx<0b10, 1, 0b01, FPR32, "ldr">;
1853 def LDRDpre : LoadPreIdx<0b11, 1, 0b01, FPR64, "ldr">;
1854 def LDRQpre : LoadPreIdx<0b00, 1, 0b11, FPR128, "ldr">;
1856 // load sign-extended half-word
1857 def LDRSHWpre : LoadPreIdx<0b01, 0, 0b11, GPR32, "ldrsh">;
1858 def LDRSHXpre : LoadPreIdx<0b01, 0, 0b10, GPR64, "ldrsh">;
1860 // load sign-extended byte
1861 def LDRSBWpre : LoadPreIdx<0b00, 0, 0b11, GPR32, "ldrsb">;
1862 def LDRSBXpre : LoadPreIdx<0b00, 0, 0b10, GPR64, "ldrsb">;
1864 // load zero-extended byte
1865 def LDRBBpre : LoadPreIdx<0b00, 0, 0b01, GPR32, "ldrb">;
1866 def LDRHHpre : LoadPreIdx<0b01, 0, 0b01, GPR32, "ldrh">;
1868 // load sign-extended word
1869 def LDRSWpre : LoadPreIdx<0b10, 0, 0b10, GPR64, "ldrsw">;
1872 // (immediate post-indexed)
1873 def LDRWpost : LoadPostIdx<0b10, 0, 0b01, GPR32, "ldr">;
1874 def LDRXpost : LoadPostIdx<0b11, 0, 0b01, GPR64, "ldr">;
1875 def LDRBpost : LoadPostIdx<0b00, 1, 0b01, FPR8, "ldr">;
1876 def LDRHpost : LoadPostIdx<0b01, 1, 0b01, FPR16, "ldr">;
1877 def LDRSpost : LoadPostIdx<0b10, 1, 0b01, FPR32, "ldr">;
1878 def LDRDpost : LoadPostIdx<0b11, 1, 0b01, FPR64, "ldr">;
1879 def LDRQpost : LoadPostIdx<0b00, 1, 0b11, FPR128, "ldr">;
1881 // load sign-extended half-word
1882 def LDRSHWpost : LoadPostIdx<0b01, 0, 0b11, GPR32, "ldrsh">;
1883 def LDRSHXpost : LoadPostIdx<0b01, 0, 0b10, GPR64, "ldrsh">;
1885 // load sign-extended byte
1886 def LDRSBWpost : LoadPostIdx<0b00, 0, 0b11, GPR32, "ldrsb">;
1887 def LDRSBXpost : LoadPostIdx<0b00, 0, 0b10, GPR64, "ldrsb">;
1889 // load zero-extended byte
1890 def LDRBBpost : LoadPostIdx<0b00, 0, 0b01, GPR32, "ldrb">;
1891 def LDRHHpost : LoadPostIdx<0b01, 0, 0b01, GPR32, "ldrh">;
1893 // load sign-extended word
1894 def LDRSWpost : LoadPostIdx<0b10, 0, 0b10, GPR64, "ldrsw">;
1896 //===----------------------------------------------------------------------===//
1897 // Store instructions.
1898 //===----------------------------------------------------------------------===//
1900 // Pair (indexed, offset)
1901 // FIXME: Use dedicated range-checked addressing mode operand here.
1902 defm STPW : StorePairOffset<0b00, 0, GPR32, simm7s4, "stp">;
1903 defm STPX : StorePairOffset<0b10, 0, GPR64, simm7s8, "stp">;
1904 defm STPS : StorePairOffset<0b00, 1, FPR32, simm7s4, "stp">;
1905 defm STPD : StorePairOffset<0b01, 1, FPR64, simm7s8, "stp">;
1906 defm STPQ : StorePairOffset<0b10, 1, FPR128, simm7s16, "stp">;
1908 // Pair (pre-indexed)
1909 def STPWpre : StorePairPreIdx<0b00, 0, GPR32, simm7s4, "stp">;
1910 def STPXpre : StorePairPreIdx<0b10, 0, GPR64, simm7s8, "stp">;
1911 def STPSpre : StorePairPreIdx<0b00, 1, FPR32, simm7s4, "stp">;
1912 def STPDpre : StorePairPreIdx<0b01, 1, FPR64, simm7s8, "stp">;
1913 def STPQpre : StorePairPreIdx<0b10, 1, FPR128, simm7s16, "stp">;
1915 // Pair (pre-indexed)
1916 def STPWpost : StorePairPostIdx<0b00, 0, GPR32, simm7s4, "stp">;
1917 def STPXpost : StorePairPostIdx<0b10, 0, GPR64, simm7s8, "stp">;
1918 def STPSpost : StorePairPostIdx<0b00, 1, FPR32, simm7s4, "stp">;
1919 def STPDpost : StorePairPostIdx<0b01, 1, FPR64, simm7s8, "stp">;
1920 def STPQpost : StorePairPostIdx<0b10, 1, FPR128, simm7s16, "stp">;
1922 // Pair (no allocate)
1923 defm STNPW : StorePairNoAlloc<0b00, 0, GPR32, simm7s4, "stnp">;
1924 defm STNPX : StorePairNoAlloc<0b10, 0, GPR64, simm7s8, "stnp">;
1925 defm STNPS : StorePairNoAlloc<0b00, 1, FPR32, simm7s4, "stnp">;
1926 defm STNPD : StorePairNoAlloc<0b01, 1, FPR64, simm7s8, "stnp">;
1927 defm STNPQ : StorePairNoAlloc<0b10, 1, FPR128, simm7s16, "stnp">;
1930 // (Register offset)
1933 defm STRBB : Store8RO< 0b00, 0, 0b00, GPR32, "strb", i32, truncstorei8>;
1934 defm STRHH : Store16RO<0b01, 0, 0b00, GPR32, "strh", i32, truncstorei16>;
1935 defm STRW : Store32RO<0b10, 0, 0b00, GPR32, "str", i32, store>;
1936 defm STRX : Store64RO<0b11, 0, 0b00, GPR64, "str", i64, store>;
1940 defm STRB : Store8RO< 0b00, 1, 0b00, FPR8, "str", untyped, store>;
1941 defm STRH : Store16RO<0b01, 1, 0b00, FPR16, "str", f16, store>;
1942 defm STRS : Store32RO<0b10, 1, 0b00, FPR32, "str", f32, store>;
1943 defm STRD : Store64RO<0b11, 1, 0b00, FPR64, "str", f64, store>;
1944 defm STRQ : Store128RO<0b00, 1, 0b10, FPR128, "str", f128, store>;
1946 multiclass TruncStoreFrom64ROPat<ROAddrMode ro, SDPatternOperator storeop,
1947 Instruction STRW, Instruction STRX> {
1949 def : Pat<(storeop GPR64:$Rt,
1950 (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)),
1951 (STRW (EXTRACT_SUBREG GPR64:$Rt, sub_32),
1952 GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
1954 def : Pat<(storeop GPR64:$Rt,
1955 (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)),
1956 (STRX (EXTRACT_SUBREG GPR64:$Rt, sub_32),
1957 GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
1960 let AddedComplexity = 10 in {
1962 defm : TruncStoreFrom64ROPat<ro8, truncstorei8, STRBBroW, STRBBroX>;
1963 defm : TruncStoreFrom64ROPat<ro16, truncstorei16, STRHHroW, STRHHroX>;
1964 defm : TruncStoreFrom64ROPat<ro32, truncstorei32, STRWroW, STRWroX>;
1967 multiclass VecROStorePat<ROAddrMode ro, ValueType VecTy, RegisterClass FPR,
1968 Instruction STRW, Instruction STRX> {
1969 def : Pat<(store (VecTy FPR:$Rt),
1970 (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)),
1971 (STRW FPR:$Rt, GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
1973 def : Pat<(store (VecTy FPR:$Rt),
1974 (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)),
1975 (STRX FPR:$Rt, GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
1978 let AddedComplexity = 10 in {
1979 // Match all store 64 bits width whose type is compatible with FPR64
1980 let Predicates = [IsLE] in {
1981 // We must use ST1 to store vectors in big-endian.
1982 defm : VecROStorePat<ro64, v2i32, FPR64, STRDroW, STRDroX>;
1983 defm : VecROStorePat<ro64, v2f32, FPR64, STRDroW, STRDroX>;
1984 defm : VecROStorePat<ro64, v4i16, FPR64, STRDroW, STRDroX>;
1985 defm : VecROStorePat<ro64, v8i8, FPR64, STRDroW, STRDroX>;
1986 defm : VecROStorePat<ro64, v4f16, FPR64, STRDroW, STRDroX>;
1989 defm : VecROStorePat<ro64, v1i64, FPR64, STRDroW, STRDroX>;
1990 defm : VecROStorePat<ro64, v1f64, FPR64, STRDroW, STRDroX>;
1992 // Match all store 128 bits width whose type is compatible with FPR128
1993 let Predicates = [IsLE] in {
1994 // We must use ST1 to store vectors in big-endian.
1995 defm : VecROStorePat<ro128, v2i64, FPR128, STRQroW, STRQroX>;
1996 defm : VecROStorePat<ro128, v2f64, FPR128, STRQroW, STRQroX>;
1997 defm : VecROStorePat<ro128, v4i32, FPR128, STRQroW, STRQroX>;
1998 defm : VecROStorePat<ro128, v4f32, FPR128, STRQroW, STRQroX>;
1999 defm : VecROStorePat<ro128, v8i16, FPR128, STRQroW, STRQroX>;
2000 defm : VecROStorePat<ro128, v16i8, FPR128, STRQroW, STRQroX>;
2001 defm : VecROStorePat<ro128, v8f16, FPR128, STRQroW, STRQroX>;
2003 } // AddedComplexity = 10
2005 // Match stores from lane 0 to the appropriate subreg's store.
2006 multiclass VecROStoreLane0Pat<ROAddrMode ro, SDPatternOperator storeop,
2007 ValueType VecTy, ValueType STy,
2008 SubRegIndex SubRegIdx,
2009 Instruction STRW, Instruction STRX> {
2011 def : Pat<(storeop (STy (vector_extract (VecTy VecListOne128:$Vt), 0)),
2012 (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)),
2013 (STRW (EXTRACT_SUBREG VecListOne128:$Vt, SubRegIdx),
2014 GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
2016 def : Pat<(storeop (STy (vector_extract (VecTy VecListOne128:$Vt), 0)),
2017 (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)),
2018 (STRX (EXTRACT_SUBREG VecListOne128:$Vt, SubRegIdx),
2019 GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
2022 let AddedComplexity = 19 in {
2023 defm : VecROStoreLane0Pat<ro16, truncstorei16, v8i16, i32, hsub, STRHroW, STRHroX>;
2024 defm : VecROStoreLane0Pat<ro16, store , v8i16, i16, hsub, STRHroW, STRHroX>;
2025 defm : VecROStoreLane0Pat<ro32, truncstorei32, v4i32, i32, ssub, STRSroW, STRSroX>;
2026 defm : VecROStoreLane0Pat<ro32, store , v4i32, i32, ssub, STRSroW, STRSroX>;
2027 defm : VecROStoreLane0Pat<ro32, store , v4f32, f32, ssub, STRSroW, STRSroX>;
2028 defm : VecROStoreLane0Pat<ro64, store , v2i64, i64, dsub, STRDroW, STRDroX>;
2029 defm : VecROStoreLane0Pat<ro64, store , v2f64, f64, dsub, STRDroW, STRDroX>;
2033 // (unsigned immediate)
2034 defm STRX : StoreUI<0b11, 0, 0b00, GPR64, uimm12s8, "str",
2036 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))]>;
2037 defm STRW : StoreUI<0b10, 0, 0b00, GPR32, uimm12s4, "str",
2039 (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))]>;
2040 defm STRB : StoreUI<0b00, 1, 0b00, FPR8, uimm12s1, "str",
2042 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))]>;
2043 defm STRH : StoreUI<0b01, 1, 0b00, FPR16, uimm12s2, "str",
2044 [(store (f16 FPR16:$Rt),
2045 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))]>;
2046 defm STRS : StoreUI<0b10, 1, 0b00, FPR32, uimm12s4, "str",
2047 [(store (f32 FPR32:$Rt),
2048 (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))]>;
2049 defm STRD : StoreUI<0b11, 1, 0b00, FPR64, uimm12s8, "str",
2050 [(store (f64 FPR64:$Rt),
2051 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))]>;
2052 defm STRQ : StoreUI<0b00, 1, 0b10, FPR128, uimm12s16, "str", []>;
2054 defm STRHH : StoreUI<0b01, 0, 0b00, GPR32, uimm12s2, "strh",
2055 [(truncstorei16 GPR32:$Rt,
2056 (am_indexed16 GPR64sp:$Rn,
2057 uimm12s2:$offset))]>;
2058 defm STRBB : StoreUI<0b00, 0, 0b00, GPR32, uimm12s1, "strb",
2059 [(truncstorei8 GPR32:$Rt,
2060 (am_indexed8 GPR64sp:$Rn,
2061 uimm12s1:$offset))]>;
2063 // Match all store 64 bits width whose type is compatible with FPR64
2064 let AddedComplexity = 10 in {
2065 let Predicates = [IsLE] in {
2066 // We must use ST1 to store vectors in big-endian.
2067 def : Pat<(store (v2f32 FPR64:$Rt),
2068 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
2069 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
2070 def : Pat<(store (v8i8 FPR64:$Rt),
2071 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
2072 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
2073 def : Pat<(store (v4i16 FPR64:$Rt),
2074 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
2075 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
2076 def : Pat<(store (v2i32 FPR64:$Rt),
2077 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
2078 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
2079 def : Pat<(store (v4f16 FPR64:$Rt),
2080 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
2081 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
2083 def : Pat<(store (v1f64 FPR64:$Rt),
2084 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
2085 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
2086 def : Pat<(store (v1i64 FPR64:$Rt),
2087 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
2088 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
2090 // Match all store 128 bits width whose type is compatible with FPR128
2091 let Predicates = [IsLE] in {
2092 // We must use ST1 to store vectors in big-endian.
2093 def : Pat<(store (v4f32 FPR128:$Rt),
2094 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
2095 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
2096 def : Pat<(store (v2f64 FPR128:$Rt),
2097 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
2098 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
2099 def : Pat<(store (v16i8 FPR128:$Rt),
2100 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
2101 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
2102 def : Pat<(store (v8i16 FPR128:$Rt),
2103 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
2104 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
2105 def : Pat<(store (v4i32 FPR128:$Rt),
2106 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
2107 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
2108 def : Pat<(store (v2i64 FPR128:$Rt),
2109 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
2110 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
2111 def : Pat<(store (v8f16 FPR128:$Rt),
2112 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
2113 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
2115 def : Pat<(store (f128 FPR128:$Rt),
2116 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
2117 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
2120 def : Pat<(truncstorei32 GPR64:$Rt,
2121 (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset)),
2122 (STRWui (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, uimm12s4:$offset)>;
2123 def : Pat<(truncstorei16 GPR64:$Rt,
2124 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset)),
2125 (STRHHui (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, uimm12s2:$offset)>;
2126 def : Pat<(truncstorei8 GPR64:$Rt, (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset)),
2127 (STRBBui (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, uimm12s1:$offset)>;
2129 } // AddedComplexity = 10
2132 // (unscaled immediate)
2133 defm STURX : StoreUnscaled<0b11, 0, 0b00, GPR64, "stur",
2135 (am_unscaled64 GPR64sp:$Rn, simm9:$offset))]>;
2136 defm STURW : StoreUnscaled<0b10, 0, 0b00, GPR32, "stur",
2138 (am_unscaled32 GPR64sp:$Rn, simm9:$offset))]>;
2139 defm STURB : StoreUnscaled<0b00, 1, 0b00, FPR8, "stur",
2141 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))]>;
2142 defm STURH : StoreUnscaled<0b01, 1, 0b00, FPR16, "stur",
2143 [(store (f16 FPR16:$Rt),
2144 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))]>;
2145 defm STURS : StoreUnscaled<0b10, 1, 0b00, FPR32, "stur",
2146 [(store (f32 FPR32:$Rt),
2147 (am_unscaled32 GPR64sp:$Rn, simm9:$offset))]>;
2148 defm STURD : StoreUnscaled<0b11, 1, 0b00, FPR64, "stur",
2149 [(store (f64 FPR64:$Rt),
2150 (am_unscaled64 GPR64sp:$Rn, simm9:$offset))]>;
2151 defm STURQ : StoreUnscaled<0b00, 1, 0b10, FPR128, "stur",
2152 [(store (f128 FPR128:$Rt),
2153 (am_unscaled128 GPR64sp:$Rn, simm9:$offset))]>;
2154 defm STURHH : StoreUnscaled<0b01, 0, 0b00, GPR32, "sturh",
2155 [(truncstorei16 GPR32:$Rt,
2156 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))]>;
2157 defm STURBB : StoreUnscaled<0b00, 0, 0b00, GPR32, "sturb",
2158 [(truncstorei8 GPR32:$Rt,
2159 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))]>;
2161 // Match all store 64 bits width whose type is compatible with FPR64
2162 let Predicates = [IsLE] in {
2163 // We must use ST1 to store vectors in big-endian.
2164 def : Pat<(store (v2f32 FPR64:$Rt),
2165 (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2166 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2167 def : Pat<(store (v8i8 FPR64:$Rt),
2168 (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2169 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2170 def : Pat<(store (v4i16 FPR64:$Rt),
2171 (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2172 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2173 def : Pat<(store (v2i32 FPR64:$Rt),
2174 (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2175 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2176 def : Pat<(store (v4f16 FPR64:$Rt),
2177 (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2178 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2180 def : Pat<(store (v1f64 FPR64:$Rt), (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2181 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2182 def : Pat<(store (v1i64 FPR64:$Rt), (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2183 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2185 // Match all store 128 bits width whose type is compatible with FPR128
2186 let Predicates = [IsLE] in {
2187 // We must use ST1 to store vectors in big-endian.
2188 def : Pat<(store (v4f32 FPR128:$Rt),
2189 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2190 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2191 def : Pat<(store (v2f64 FPR128:$Rt),
2192 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2193 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2194 def : Pat<(store (v16i8 FPR128:$Rt),
2195 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2196 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2197 def : Pat<(store (v8i16 FPR128:$Rt),
2198 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2199 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2200 def : Pat<(store (v4i32 FPR128:$Rt),
2201 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2202 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2203 def : Pat<(store (v2i64 FPR128:$Rt),
2204 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2205 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2206 def : Pat<(store (v2f64 FPR128:$Rt),
2207 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2208 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2209 def : Pat<(store (v8f16 FPR128:$Rt),
2210 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2211 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2214 // unscaled i64 truncating stores
2215 def : Pat<(truncstorei32 GPR64:$Rt, (am_unscaled32 GPR64sp:$Rn, simm9:$offset)),
2216 (STURWi (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, simm9:$offset)>;
2217 def : Pat<(truncstorei16 GPR64:$Rt, (am_unscaled16 GPR64sp:$Rn, simm9:$offset)),
2218 (STURHHi (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, simm9:$offset)>;
2219 def : Pat<(truncstorei8 GPR64:$Rt, (am_unscaled8 GPR64sp:$Rn, simm9:$offset)),
2220 (STURBBi (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, simm9:$offset)>;
2223 // STR mnemonics fall back to STUR for negative or unaligned offsets.
2224 def : InstAlias<"str $Rt, [$Rn, $offset]",
2225 (STURXi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb64:$offset), 0>;
2226 def : InstAlias<"str $Rt, [$Rn, $offset]",
2227 (STURWi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
2228 def : InstAlias<"str $Rt, [$Rn, $offset]",
2229 (STURBi FPR8:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
2230 def : InstAlias<"str $Rt, [$Rn, $offset]",
2231 (STURHi FPR16:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
2232 def : InstAlias<"str $Rt, [$Rn, $offset]",
2233 (STURSi FPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
2234 def : InstAlias<"str $Rt, [$Rn, $offset]",
2235 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb64:$offset), 0>;
2236 def : InstAlias<"str $Rt, [$Rn, $offset]",
2237 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9_offset_fb128:$offset), 0>;
2239 def : InstAlias<"strb $Rt, [$Rn, $offset]",
2240 (STURBBi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
2241 def : InstAlias<"strh $Rt, [$Rn, $offset]",
2242 (STURHHi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
2245 // (unscaled immediate, unprivileged)
2246 defm STTRW : StoreUnprivileged<0b10, 0, 0b00, GPR32, "sttr">;
2247 defm STTRX : StoreUnprivileged<0b11, 0, 0b00, GPR64, "sttr">;
2249 defm STTRH : StoreUnprivileged<0b01, 0, 0b00, GPR32, "sttrh">;
2250 defm STTRB : StoreUnprivileged<0b00, 0, 0b00, GPR32, "sttrb">;
2253 // (immediate pre-indexed)
2254 def STRWpre : StorePreIdx<0b10, 0, 0b00, GPR32, "str", pre_store, i32>;
2255 def STRXpre : StorePreIdx<0b11, 0, 0b00, GPR64, "str", pre_store, i64>;
2256 def STRBpre : StorePreIdx<0b00, 1, 0b00, FPR8, "str", pre_store, untyped>;
2257 def STRHpre : StorePreIdx<0b01, 1, 0b00, FPR16, "str", pre_store, f16>;
2258 def STRSpre : StorePreIdx<0b10, 1, 0b00, FPR32, "str", pre_store, f32>;
2259 def STRDpre : StorePreIdx<0b11, 1, 0b00, FPR64, "str", pre_store, f64>;
2260 def STRQpre : StorePreIdx<0b00, 1, 0b10, FPR128, "str", pre_store, f128>;
2262 def STRBBpre : StorePreIdx<0b00, 0, 0b00, GPR32, "strb", pre_truncsti8, i32>;
2263 def STRHHpre : StorePreIdx<0b01, 0, 0b00, GPR32, "strh", pre_truncsti16, i32>;
2266 def : Pat<(pre_truncsti32 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
2267 (STRWpre (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
2269 def : Pat<(pre_truncsti16 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
2270 (STRHHpre (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
2272 def : Pat<(pre_truncsti8 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
2273 (STRBBpre (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
2276 def : Pat<(pre_store (v8i8 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2277 (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2278 def : Pat<(pre_store (v4i16 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2279 (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2280 def : Pat<(pre_store (v2i32 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2281 (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2282 def : Pat<(pre_store (v2f32 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2283 (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2284 def : Pat<(pre_store (v1i64 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2285 (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2286 def : Pat<(pre_store (v1f64 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2287 (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2288 def : Pat<(pre_store (v4f16 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2289 (STRDpre FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2291 def : Pat<(pre_store (v16i8 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2292 (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2293 def : Pat<(pre_store (v8i16 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2294 (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2295 def : Pat<(pre_store (v4i32 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2296 (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2297 def : Pat<(pre_store (v4f32 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2298 (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2299 def : Pat<(pre_store (v2i64 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2300 (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2301 def : Pat<(pre_store (v2f64 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2302 (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2303 def : Pat<(pre_store (v8f16 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2304 (STRQpre FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2307 // (immediate post-indexed)
2308 def STRWpost : StorePostIdx<0b10, 0, 0b00, GPR32, "str", post_store, i32>;
2309 def STRXpost : StorePostIdx<0b11, 0, 0b00, GPR64, "str", post_store, i64>;
2310 def STRBpost : StorePostIdx<0b00, 1, 0b00, FPR8, "str", post_store, untyped>;
2311 def STRHpost : StorePostIdx<0b01, 1, 0b00, FPR16, "str", post_store, f16>;
2312 def STRSpost : StorePostIdx<0b10, 1, 0b00, FPR32, "str", post_store, f32>;
2313 def STRDpost : StorePostIdx<0b11, 1, 0b00, FPR64, "str", post_store, f64>;
2314 def STRQpost : StorePostIdx<0b00, 1, 0b10, FPR128, "str", post_store, f128>;
2316 def STRBBpost : StorePostIdx<0b00, 0, 0b00, GPR32, "strb", post_truncsti8, i32>;
2317 def STRHHpost : StorePostIdx<0b01, 0, 0b00, GPR32, "strh", post_truncsti16, i32>;
2320 def : Pat<(post_truncsti32 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
2321 (STRWpost (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
2323 def : Pat<(post_truncsti16 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
2324 (STRHHpost (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
2326 def : Pat<(post_truncsti8 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
2327 (STRBBpost (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
2330 def : Pat<(post_store (v8i8 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2331 (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2332 def : Pat<(post_store (v4i16 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2333 (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2334 def : Pat<(post_store (v2i32 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2335 (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2336 def : Pat<(post_store (v2f32 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2337 (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2338 def : Pat<(post_store (v1i64 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2339 (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2340 def : Pat<(post_store (v1f64 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2341 (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2342 def : Pat<(post_store (v4f16 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2343 (STRDpost FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2345 def : Pat<(post_store (v16i8 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2346 (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2347 def : Pat<(post_store (v8i16 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2348 (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2349 def : Pat<(post_store (v4i32 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2350 (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2351 def : Pat<(post_store (v4f32 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2352 (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2353 def : Pat<(post_store (v2i64 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2354 (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2355 def : Pat<(post_store (v2f64 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2356 (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2357 def : Pat<(post_store (v8f16 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2358 (STRQpost FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2360 //===----------------------------------------------------------------------===//
2361 // Load/store exclusive instructions.
2362 //===----------------------------------------------------------------------===//
2364 def LDARW : LoadAcquire <0b10, 1, 1, 0, 1, GPR32, "ldar">;
2365 def LDARX : LoadAcquire <0b11, 1, 1, 0, 1, GPR64, "ldar">;
2366 def LDARB : LoadAcquire <0b00, 1, 1, 0, 1, GPR32, "ldarb">;
2367 def LDARH : LoadAcquire <0b01, 1, 1, 0, 1, GPR32, "ldarh">;
2369 def LDAXRW : LoadExclusive <0b10, 0, 1, 0, 1, GPR32, "ldaxr">;
2370 def LDAXRX : LoadExclusive <0b11, 0, 1, 0, 1, GPR64, "ldaxr">;
2371 def LDAXRB : LoadExclusive <0b00, 0, 1, 0, 1, GPR32, "ldaxrb">;
2372 def LDAXRH : LoadExclusive <0b01, 0, 1, 0, 1, GPR32, "ldaxrh">;
2374 def LDXRW : LoadExclusive <0b10, 0, 1, 0, 0, GPR32, "ldxr">;
2375 def LDXRX : LoadExclusive <0b11, 0, 1, 0, 0, GPR64, "ldxr">;
2376 def LDXRB : LoadExclusive <0b00, 0, 1, 0, 0, GPR32, "ldxrb">;
2377 def LDXRH : LoadExclusive <0b01, 0, 1, 0, 0, GPR32, "ldxrh">;
2379 def STLRW : StoreRelease <0b10, 1, 0, 0, 1, GPR32, "stlr">;
2380 def STLRX : StoreRelease <0b11, 1, 0, 0, 1, GPR64, "stlr">;
2381 def STLRB : StoreRelease <0b00, 1, 0, 0, 1, GPR32, "stlrb">;
2382 def STLRH : StoreRelease <0b01, 1, 0, 0, 1, GPR32, "stlrh">;
2384 def STLXRW : StoreExclusive<0b10, 0, 0, 0, 1, GPR32, "stlxr">;
2385 def STLXRX : StoreExclusive<0b11, 0, 0, 0, 1, GPR64, "stlxr">;
2386 def STLXRB : StoreExclusive<0b00, 0, 0, 0, 1, GPR32, "stlxrb">;
2387 def STLXRH : StoreExclusive<0b01, 0, 0, 0, 1, GPR32, "stlxrh">;
2389 def STXRW : StoreExclusive<0b10, 0, 0, 0, 0, GPR32, "stxr">;
2390 def STXRX : StoreExclusive<0b11, 0, 0, 0, 0, GPR64, "stxr">;
2391 def STXRB : StoreExclusive<0b00, 0, 0, 0, 0, GPR32, "stxrb">;
2392 def STXRH : StoreExclusive<0b01, 0, 0, 0, 0, GPR32, "stxrh">;
2394 def LDAXPW : LoadExclusivePair<0b10, 0, 1, 1, 1, GPR32, "ldaxp">;
2395 def LDAXPX : LoadExclusivePair<0b11, 0, 1, 1, 1, GPR64, "ldaxp">;
2397 def LDXPW : LoadExclusivePair<0b10, 0, 1, 1, 0, GPR32, "ldxp">;
2398 def LDXPX : LoadExclusivePair<0b11, 0, 1, 1, 0, GPR64, "ldxp">;
2400 def STLXPW : StoreExclusivePair<0b10, 0, 0, 1, 1, GPR32, "stlxp">;
2401 def STLXPX : StoreExclusivePair<0b11, 0, 0, 1, 1, GPR64, "stlxp">;
2403 def STXPW : StoreExclusivePair<0b10, 0, 0, 1, 0, GPR32, "stxp">;
2404 def STXPX : StoreExclusivePair<0b11, 0, 0, 1, 0, GPR64, "stxp">;
2406 let Predicates = [HasV8_1a] in {
2407 // v8.1a "Limited Order Region" extension load-acquire instructions
2408 def LDLARW : LoadAcquire <0b10, 1, 1, 0, 0, GPR32, "ldlar">;
2409 def LDLARX : LoadAcquire <0b11, 1, 1, 0, 0, GPR64, "ldlar">;
2410 def LDLARB : LoadAcquire <0b00, 1, 1, 0, 0, GPR32, "ldlarb">;
2411 def LDLARH : LoadAcquire <0b01, 1, 1, 0, 0, GPR32, "ldlarh">;
2413 // v8.1a "Limited Order Region" extension store-release instructions
2414 def STLLRW : StoreRelease <0b10, 1, 0, 0, 0, GPR32, "stllr">;
2415 def STLLRX : StoreRelease <0b11, 1, 0, 0, 0, GPR64, "stllr">;
2416 def STLLRB : StoreRelease <0b00, 1, 0, 0, 0, GPR32, "stllrb">;
2417 def STLLRH : StoreRelease <0b01, 1, 0, 0, 0, GPR32, "stllrh">;
2420 //===----------------------------------------------------------------------===//
2421 // Scaled floating point to integer conversion instructions.
2422 //===----------------------------------------------------------------------===//
2424 defm FCVTAS : FPToIntegerUnscaled<0b00, 0b100, "fcvtas", int_aarch64_neon_fcvtas>;
2425 defm FCVTAU : FPToIntegerUnscaled<0b00, 0b101, "fcvtau", int_aarch64_neon_fcvtau>;
2426 defm FCVTMS : FPToIntegerUnscaled<0b10, 0b000, "fcvtms", int_aarch64_neon_fcvtms>;
2427 defm FCVTMU : FPToIntegerUnscaled<0b10, 0b001, "fcvtmu", int_aarch64_neon_fcvtmu>;
2428 defm FCVTNS : FPToIntegerUnscaled<0b00, 0b000, "fcvtns", int_aarch64_neon_fcvtns>;
2429 defm FCVTNU : FPToIntegerUnscaled<0b00, 0b001, "fcvtnu", int_aarch64_neon_fcvtnu>;
2430 defm FCVTPS : FPToIntegerUnscaled<0b01, 0b000, "fcvtps", int_aarch64_neon_fcvtps>;
2431 defm FCVTPU : FPToIntegerUnscaled<0b01, 0b001, "fcvtpu", int_aarch64_neon_fcvtpu>;
2432 defm FCVTZS : FPToIntegerUnscaled<0b11, 0b000, "fcvtzs", fp_to_sint>;
2433 defm FCVTZU : FPToIntegerUnscaled<0b11, 0b001, "fcvtzu", fp_to_uint>;
2434 defm FCVTZS : FPToIntegerScaled<0b11, 0b000, "fcvtzs", fp_to_sint>;
2435 defm FCVTZU : FPToIntegerScaled<0b11, 0b001, "fcvtzu", fp_to_uint>;
2436 let isCodeGenOnly = 1 in {
2437 defm FCVTZS_Int : FPToIntegerUnscaled<0b11, 0b000, "fcvtzs", int_aarch64_neon_fcvtzs>;
2438 defm FCVTZU_Int : FPToIntegerUnscaled<0b11, 0b001, "fcvtzu", int_aarch64_neon_fcvtzu>;
2439 defm FCVTZS_Int : FPToIntegerScaled<0b11, 0b000, "fcvtzs", int_aarch64_neon_fcvtzs>;
2440 defm FCVTZU_Int : FPToIntegerScaled<0b11, 0b001, "fcvtzu", int_aarch64_neon_fcvtzu>;
2443 multiclass FPToIntegerPats<SDNode to_int, SDNode round, string INST> {
2444 def : Pat<(i32 (to_int (round f32:$Rn))),
2445 (!cast<Instruction>(INST # UWSr) f32:$Rn)>;
2446 def : Pat<(i64 (to_int (round f32:$Rn))),
2447 (!cast<Instruction>(INST # UXSr) f32:$Rn)>;
2448 def : Pat<(i32 (to_int (round f64:$Rn))),
2449 (!cast<Instruction>(INST # UWDr) f64:$Rn)>;
2450 def : Pat<(i64 (to_int (round f64:$Rn))),
2451 (!cast<Instruction>(INST # UXDr) f64:$Rn)>;
2454 defm : FPToIntegerPats<fp_to_sint, fceil, "FCVTPS">;
2455 defm : FPToIntegerPats<fp_to_uint, fceil, "FCVTPU">;
2456 defm : FPToIntegerPats<fp_to_sint, ffloor, "FCVTMS">;
2457 defm : FPToIntegerPats<fp_to_uint, ffloor, "FCVTMU">;
2458 defm : FPToIntegerPats<fp_to_sint, ftrunc, "FCVTZS">;
2459 defm : FPToIntegerPats<fp_to_uint, ftrunc, "FCVTZU">;
2460 defm : FPToIntegerPats<fp_to_sint, frnd, "FCVTAS">;
2461 defm : FPToIntegerPats<fp_to_uint, frnd, "FCVTAU">;
2463 //===----------------------------------------------------------------------===//
2464 // Scaled integer to floating point conversion instructions.
2465 //===----------------------------------------------------------------------===//
2467 defm SCVTF : IntegerToFP<0, "scvtf", sint_to_fp>;
2468 defm UCVTF : IntegerToFP<1, "ucvtf", uint_to_fp>;
2470 //===----------------------------------------------------------------------===//
2471 // Unscaled integer to floating point conversion instruction.
2472 //===----------------------------------------------------------------------===//
2474 defm FMOV : UnscaledConversion<"fmov">;
2476 // Add pseudo ops for FMOV 0 so we can mark them as isReMaterializable
2477 let isReMaterializable = 1, isCodeGenOnly = 1 in {
2478 def FMOVS0 : Pseudo<(outs FPR32:$Rd), (ins), [(set f32:$Rd, (fpimm0))]>,
2479 PseudoInstExpansion<(FMOVWSr FPR32:$Rd, WZR)>,
2481 def FMOVD0 : Pseudo<(outs FPR64:$Rd), (ins), [(set f64:$Rd, (fpimm0))]>,
2482 PseudoInstExpansion<(FMOVXDr FPR64:$Rd, XZR)>,
2486 //===----------------------------------------------------------------------===//
2487 // Floating point conversion instruction.
2488 //===----------------------------------------------------------------------===//
2490 defm FCVT : FPConversion<"fcvt">;
2492 //===----------------------------------------------------------------------===//
2493 // Floating point single operand instructions.
2494 //===----------------------------------------------------------------------===//
2496 defm FABS : SingleOperandFPData<0b0001, "fabs", fabs>;
2497 defm FMOV : SingleOperandFPData<0b0000, "fmov">;
2498 defm FNEG : SingleOperandFPData<0b0010, "fneg", fneg>;
2499 defm FRINTA : SingleOperandFPData<0b1100, "frinta", frnd>;
2500 defm FRINTI : SingleOperandFPData<0b1111, "frinti", fnearbyint>;
2501 defm FRINTM : SingleOperandFPData<0b1010, "frintm", ffloor>;
2502 defm FRINTN : SingleOperandFPData<0b1000, "frintn", int_aarch64_neon_frintn>;
2503 defm FRINTP : SingleOperandFPData<0b1001, "frintp", fceil>;
2505 def : Pat<(v1f64 (int_aarch64_neon_frintn (v1f64 FPR64:$Rn))),
2506 (FRINTNDr FPR64:$Rn)>;
2508 defm FRINTX : SingleOperandFPData<0b1110, "frintx", frint>;
2509 defm FRINTZ : SingleOperandFPData<0b1011, "frintz", ftrunc>;
2511 let SchedRW = [WriteFDiv] in {
2512 defm FSQRT : SingleOperandFPData<0b0011, "fsqrt", fsqrt>;
2515 //===----------------------------------------------------------------------===//
2516 // Floating point two operand instructions.
2517 //===----------------------------------------------------------------------===//
2519 defm FADD : TwoOperandFPData<0b0010, "fadd", fadd>;
2520 let SchedRW = [WriteFDiv] in {
2521 defm FDIV : TwoOperandFPData<0b0001, "fdiv", fdiv>;
2523 defm FMAXNM : TwoOperandFPData<0b0110, "fmaxnm", fmaxnum>;
2524 defm FMAX : TwoOperandFPData<0b0100, "fmax", fmaxnan>;
2525 defm FMINNM : TwoOperandFPData<0b0111, "fminnm", fminnum>;
2526 defm FMIN : TwoOperandFPData<0b0101, "fmin", fminnan>;
2527 let SchedRW = [WriteFMul] in {
2528 defm FMUL : TwoOperandFPData<0b0000, "fmul", fmul>;
2529 defm FNMUL : TwoOperandFPDataNeg<0b1000, "fnmul", fmul>;
2531 defm FSUB : TwoOperandFPData<0b0011, "fsub", fsub>;
2533 def : Pat<(v1f64 (fmaxnan (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2534 (FMAXDrr FPR64:$Rn, FPR64:$Rm)>;
2535 def : Pat<(v1f64 (fminnan (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2536 (FMINDrr FPR64:$Rn, FPR64:$Rm)>;
2537 def : Pat<(v1f64 (fmaxnum (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2538 (FMAXNMDrr FPR64:$Rn, FPR64:$Rm)>;
2539 def : Pat<(v1f64 (fminnum (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2540 (FMINNMDrr FPR64:$Rn, FPR64:$Rm)>;
2542 //===----------------------------------------------------------------------===//
2543 // Floating point three operand instructions.
2544 //===----------------------------------------------------------------------===//
2546 defm FMADD : ThreeOperandFPData<0, 0, "fmadd", fma>;
2547 defm FMSUB : ThreeOperandFPData<0, 1, "fmsub",
2548 TriOpFrag<(fma node:$LHS, (fneg node:$MHS), node:$RHS)> >;
2549 defm FNMADD : ThreeOperandFPData<1, 0, "fnmadd",
2550 TriOpFrag<(fneg (fma node:$LHS, node:$MHS, node:$RHS))> >;
2551 defm FNMSUB : ThreeOperandFPData<1, 1, "fnmsub",
2552 TriOpFrag<(fma node:$LHS, node:$MHS, (fneg node:$RHS))> >;
2554 // The following def pats catch the case where the LHS of an FMA is negated.
2555 // The TriOpFrag above catches the case where the middle operand is negated.
2557 // N.b. FMSUB etc have the accumulator at the *end* of (outs), unlike
2558 // the NEON variant.
2559 def : Pat<(f32 (fma (fneg FPR32:$Rn), FPR32:$Rm, FPR32:$Ra)),
2560 (FMSUBSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
2562 def : Pat<(f64 (fma (fneg FPR64:$Rn), FPR64:$Rm, FPR64:$Ra)),
2563 (FMSUBDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
2565 // We handled -(a + b*c) for FNMADD above, now it's time for "(-a) + (-b)*c" and
2567 def : Pat<(f32 (fma (fneg FPR32:$Rn), FPR32:$Rm, (fneg FPR32:$Ra))),
2568 (FNMADDSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
2570 def : Pat<(f64 (fma (fneg FPR64:$Rn), FPR64:$Rm, (fneg FPR64:$Ra))),
2571 (FNMADDDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
2573 def : Pat<(f32 (fma FPR32:$Rn, (fneg FPR32:$Rm), (fneg FPR32:$Ra))),
2574 (FNMADDSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
2576 def : Pat<(f64 (fma FPR64:$Rn, (fneg FPR64:$Rm), (fneg FPR64:$Ra))),
2577 (FNMADDDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
2579 //===----------------------------------------------------------------------===//
2580 // Floating point comparison instructions.
2581 //===----------------------------------------------------------------------===//
2583 defm FCMPE : FPComparison<1, "fcmpe">;
2584 defm FCMP : FPComparison<0, "fcmp", AArch64fcmp>;
2586 //===----------------------------------------------------------------------===//
2587 // Floating point conditional comparison instructions.
2588 //===----------------------------------------------------------------------===//
2590 defm FCCMPE : FPCondComparison<1, "fccmpe">;
2591 defm FCCMP : FPCondComparison<0, "fccmp", AArch64fccmp>;
2593 //===----------------------------------------------------------------------===//
2594 // Floating point conditional select instruction.
2595 //===----------------------------------------------------------------------===//
2597 defm FCSEL : FPCondSelect<"fcsel">;
2599 // CSEL instructions providing f128 types need to be handled by a
2600 // pseudo-instruction since the eventual code will need to introduce basic
2601 // blocks and control flow.
2602 def F128CSEL : Pseudo<(outs FPR128:$Rd),
2603 (ins FPR128:$Rn, FPR128:$Rm, ccode:$cond),
2604 [(set (f128 FPR128:$Rd),
2605 (AArch64csel FPR128:$Rn, FPR128:$Rm,
2606 (i32 imm:$cond), NZCV))]> {
2608 let usesCustomInserter = 1;
2612 //===----------------------------------------------------------------------===//
2613 // Floating point immediate move.
2614 //===----------------------------------------------------------------------===//
2616 let isReMaterializable = 1 in {
2617 defm FMOV : FPMoveImmediate<"fmov">;
2620 //===----------------------------------------------------------------------===//
2621 // Advanced SIMD two vector instructions.
2622 //===----------------------------------------------------------------------===//
2624 defm ABS : SIMDTwoVectorBHSD<0, 0b01011, "abs", int_aarch64_neon_abs>;
2625 def : Pat<(xor (v8i8 (AArch64vashr V64:$src, (i32 7))),
2626 (v8i8 (add V64:$src, (AArch64vashr V64:$src, (i32 7))))),
2627 (ABSv8i8 V64:$src)>;
2628 def : Pat<(xor (v4i16 (AArch64vashr V64:$src, (i32 15))),
2629 (v4i16 (add V64:$src, (AArch64vashr V64:$src, (i32 15))))),
2630 (ABSv4i16 V64:$src)>;
2631 def : Pat<(xor (v2i32 (AArch64vashr V64:$src, (i32 31))),
2632 (v2i32 (add V64:$src, (AArch64vashr V64:$src, (i32 31))))),
2633 (ABSv2i32 V64:$src)>;
2634 def : Pat<(xor (v16i8 (AArch64vashr V128:$src, (i32 7))),
2635 (v16i8 (add V128:$src, (AArch64vashr V128:$src, (i32 7))))),
2636 (ABSv16i8 V128:$src)>;
2637 def : Pat<(xor (v8i16 (AArch64vashr V128:$src, (i32 15))),
2638 (v8i16 (add V128:$src, (AArch64vashr V128:$src, (i32 15))))),
2639 (ABSv8i16 V128:$src)>;
2640 def : Pat<(xor (v4i32 (AArch64vashr V128:$src, (i32 31))),
2641 (v4i32 (add V128:$src, (AArch64vashr V128:$src, (i32 31))))),
2642 (ABSv4i32 V128:$src)>;
2643 def : Pat<(xor (v2i64 (AArch64vashr V128:$src, (i32 63))),
2644 (v2i64 (add V128:$src, (AArch64vashr V128:$src, (i32 63))))),
2645 (ABSv2i64 V128:$src)>;
2647 defm CLS : SIMDTwoVectorBHS<0, 0b00100, "cls", int_aarch64_neon_cls>;
2648 defm CLZ : SIMDTwoVectorBHS<1, 0b00100, "clz", ctlz>;
2649 defm CMEQ : SIMDCmpTwoVector<0, 0b01001, "cmeq", AArch64cmeqz>;
2650 defm CMGE : SIMDCmpTwoVector<1, 0b01000, "cmge", AArch64cmgez>;
2651 defm CMGT : SIMDCmpTwoVector<0, 0b01000, "cmgt", AArch64cmgtz>;
2652 defm CMLE : SIMDCmpTwoVector<1, 0b01001, "cmle", AArch64cmlez>;
2653 defm CMLT : SIMDCmpTwoVector<0, 0b01010, "cmlt", AArch64cmltz>;
2654 defm CNT : SIMDTwoVectorB<0, 0b00, 0b00101, "cnt", ctpop>;
2655 defm FABS : SIMDTwoVectorFP<0, 1, 0b01111, "fabs", fabs>;
2657 defm FCMEQ : SIMDFPCmpTwoVector<0, 1, 0b01101, "fcmeq", AArch64fcmeqz>;
2658 defm FCMGE : SIMDFPCmpTwoVector<1, 1, 0b01100, "fcmge", AArch64fcmgez>;
2659 defm FCMGT : SIMDFPCmpTwoVector<0, 1, 0b01100, "fcmgt", AArch64fcmgtz>;
2660 defm FCMLE : SIMDFPCmpTwoVector<1, 1, 0b01101, "fcmle", AArch64fcmlez>;
2661 defm FCMLT : SIMDFPCmpTwoVector<0, 1, 0b01110, "fcmlt", AArch64fcmltz>;
2662 defm FCVTAS : SIMDTwoVectorFPToInt<0,0,0b11100, "fcvtas",int_aarch64_neon_fcvtas>;
2663 defm FCVTAU : SIMDTwoVectorFPToInt<1,0,0b11100, "fcvtau",int_aarch64_neon_fcvtau>;
2664 defm FCVTL : SIMDFPWidenTwoVector<0, 0, 0b10111, "fcvtl">;
2665 def : Pat<(v4f32 (int_aarch64_neon_vcvthf2fp (v4i16 V64:$Rn))),
2666 (FCVTLv4i16 V64:$Rn)>;
2667 def : Pat<(v4f32 (int_aarch64_neon_vcvthf2fp (extract_subvector (v8i16 V128:$Rn),
2669 (FCVTLv8i16 V128:$Rn)>;
2670 def : Pat<(v2f64 (fextend (v2f32 V64:$Rn))), (FCVTLv2i32 V64:$Rn)>;
2671 def : Pat<(v2f64 (fextend (v2f32 (extract_subvector (v4f32 V128:$Rn),
2673 (FCVTLv4i32 V128:$Rn)>;
2675 def : Pat<(v4f32 (fextend (v4f16 V64:$Rn))), (FCVTLv4i16 V64:$Rn)>;
2676 def : Pat<(v4f32 (fextend (v4f16 (extract_subvector (v8f16 V128:$Rn),
2678 (FCVTLv8i16 V128:$Rn)>;
2680 defm FCVTMS : SIMDTwoVectorFPToInt<0,0,0b11011, "fcvtms",int_aarch64_neon_fcvtms>;
2681 defm FCVTMU : SIMDTwoVectorFPToInt<1,0,0b11011, "fcvtmu",int_aarch64_neon_fcvtmu>;
2682 defm FCVTNS : SIMDTwoVectorFPToInt<0,0,0b11010, "fcvtns",int_aarch64_neon_fcvtns>;
2683 defm FCVTNU : SIMDTwoVectorFPToInt<1,0,0b11010, "fcvtnu",int_aarch64_neon_fcvtnu>;
2684 defm FCVTN : SIMDFPNarrowTwoVector<0, 0, 0b10110, "fcvtn">;
2685 def : Pat<(v4i16 (int_aarch64_neon_vcvtfp2hf (v4f32 V128:$Rn))),
2686 (FCVTNv4i16 V128:$Rn)>;
2687 def : Pat<(concat_vectors V64:$Rd,
2688 (v4i16 (int_aarch64_neon_vcvtfp2hf (v4f32 V128:$Rn)))),
2689 (FCVTNv8i16 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
2690 def : Pat<(v2f32 (fround (v2f64 V128:$Rn))), (FCVTNv2i32 V128:$Rn)>;
2691 def : Pat<(v4f16 (fround (v4f32 V128:$Rn))), (FCVTNv4i16 V128:$Rn)>;
2692 def : Pat<(concat_vectors V64:$Rd, (v2f32 (fround (v2f64 V128:$Rn)))),
2693 (FCVTNv4i32 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
2694 defm FCVTPS : SIMDTwoVectorFPToInt<0,1,0b11010, "fcvtps",int_aarch64_neon_fcvtps>;
2695 defm FCVTPU : SIMDTwoVectorFPToInt<1,1,0b11010, "fcvtpu",int_aarch64_neon_fcvtpu>;
2696 defm FCVTXN : SIMDFPInexactCvtTwoVector<1, 0, 0b10110, "fcvtxn",
2697 int_aarch64_neon_fcvtxn>;
2698 defm FCVTZS : SIMDTwoVectorFPToInt<0, 1, 0b11011, "fcvtzs", fp_to_sint>;
2699 defm FCVTZU : SIMDTwoVectorFPToInt<1, 1, 0b11011, "fcvtzu", fp_to_uint>;
2700 let isCodeGenOnly = 1 in {
2701 defm FCVTZS_Int : SIMDTwoVectorFPToInt<0, 1, 0b11011, "fcvtzs",
2702 int_aarch64_neon_fcvtzs>;
2703 defm FCVTZU_Int : SIMDTwoVectorFPToInt<1, 1, 0b11011, "fcvtzu",
2704 int_aarch64_neon_fcvtzu>;
2706 defm FNEG : SIMDTwoVectorFP<1, 1, 0b01111, "fneg", fneg>;
2707 defm FRECPE : SIMDTwoVectorFP<0, 1, 0b11101, "frecpe", int_aarch64_neon_frecpe>;
2708 defm FRINTA : SIMDTwoVectorFP<1, 0, 0b11000, "frinta", frnd>;
2709 defm FRINTI : SIMDTwoVectorFP<1, 1, 0b11001, "frinti", fnearbyint>;
2710 defm FRINTM : SIMDTwoVectorFP<0, 0, 0b11001, "frintm", ffloor>;
2711 defm FRINTN : SIMDTwoVectorFP<0, 0, 0b11000, "frintn", int_aarch64_neon_frintn>;
2712 defm FRINTP : SIMDTwoVectorFP<0, 1, 0b11000, "frintp", fceil>;
2713 defm FRINTX : SIMDTwoVectorFP<1, 0, 0b11001, "frintx", frint>;
2714 defm FRINTZ : SIMDTwoVectorFP<0, 1, 0b11001, "frintz", ftrunc>;
2715 defm FRSQRTE: SIMDTwoVectorFP<1, 1, 0b11101, "frsqrte", int_aarch64_neon_frsqrte>;
2716 defm FSQRT : SIMDTwoVectorFP<1, 1, 0b11111, "fsqrt", fsqrt>;
2717 defm NEG : SIMDTwoVectorBHSD<1, 0b01011, "neg",
2718 UnOpFrag<(sub immAllZerosV, node:$LHS)> >;
2719 defm NOT : SIMDTwoVectorB<1, 0b00, 0b00101, "not", vnot>;
2720 // Aliases for MVN -> NOT.
2721 def : InstAlias<"mvn{ $Vd.8b, $Vn.8b|.8b $Vd, $Vn}",
2722 (NOTv8i8 V64:$Vd, V64:$Vn)>;
2723 def : InstAlias<"mvn{ $Vd.16b, $Vn.16b|.16b $Vd, $Vn}",
2724 (NOTv16i8 V128:$Vd, V128:$Vn)>;
2726 def : Pat<(AArch64neg (v8i8 V64:$Rn)), (NEGv8i8 V64:$Rn)>;
2727 def : Pat<(AArch64neg (v16i8 V128:$Rn)), (NEGv16i8 V128:$Rn)>;
2728 def : Pat<(AArch64neg (v4i16 V64:$Rn)), (NEGv4i16 V64:$Rn)>;
2729 def : Pat<(AArch64neg (v8i16 V128:$Rn)), (NEGv8i16 V128:$Rn)>;
2730 def : Pat<(AArch64neg (v2i32 V64:$Rn)), (NEGv2i32 V64:$Rn)>;
2731 def : Pat<(AArch64neg (v4i32 V128:$Rn)), (NEGv4i32 V128:$Rn)>;
2732 def : Pat<(AArch64neg (v2i64 V128:$Rn)), (NEGv2i64 V128:$Rn)>;
2734 def : Pat<(AArch64not (v8i8 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2735 def : Pat<(AArch64not (v16i8 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2736 def : Pat<(AArch64not (v4i16 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2737 def : Pat<(AArch64not (v8i16 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2738 def : Pat<(AArch64not (v2i32 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2739 def : Pat<(AArch64not (v1i64 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2740 def : Pat<(AArch64not (v4i32 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2741 def : Pat<(AArch64not (v2i64 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2743 def : Pat<(vnot (v4i16 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2744 def : Pat<(vnot (v8i16 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2745 def : Pat<(vnot (v2i32 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2746 def : Pat<(vnot (v4i32 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2747 def : Pat<(vnot (v2i64 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2749 defm RBIT : SIMDTwoVectorB<1, 0b01, 0b00101, "rbit", int_aarch64_neon_rbit>;
2750 defm REV16 : SIMDTwoVectorB<0, 0b00, 0b00001, "rev16", AArch64rev16>;
2751 defm REV32 : SIMDTwoVectorBH<1, 0b00000, "rev32", AArch64rev32>;
2752 defm REV64 : SIMDTwoVectorBHS<0, 0b00000, "rev64", AArch64rev64>;
2753 defm SADALP : SIMDLongTwoVectorTied<0, 0b00110, "sadalp",
2754 BinOpFrag<(add node:$LHS, (int_aarch64_neon_saddlp node:$RHS))> >;
2755 defm SADDLP : SIMDLongTwoVector<0, 0b00010, "saddlp", int_aarch64_neon_saddlp>;
2756 defm SCVTF : SIMDTwoVectorIntToFP<0, 0, 0b11101, "scvtf", sint_to_fp>;
2757 defm SHLL : SIMDVectorLShiftLongBySizeBHS;
2758 defm SQABS : SIMDTwoVectorBHSD<0, 0b00111, "sqabs", int_aarch64_neon_sqabs>;
2759 defm SQNEG : SIMDTwoVectorBHSD<1, 0b00111, "sqneg", int_aarch64_neon_sqneg>;
2760 defm SQXTN : SIMDMixedTwoVector<0, 0b10100, "sqxtn", int_aarch64_neon_sqxtn>;
2761 defm SQXTUN : SIMDMixedTwoVector<1, 0b10010, "sqxtun", int_aarch64_neon_sqxtun>;
2762 defm SUQADD : SIMDTwoVectorBHSDTied<0, 0b00011, "suqadd",int_aarch64_neon_suqadd>;
2763 defm UADALP : SIMDLongTwoVectorTied<1, 0b00110, "uadalp",
2764 BinOpFrag<(add node:$LHS, (int_aarch64_neon_uaddlp node:$RHS))> >;
2765 defm UADDLP : SIMDLongTwoVector<1, 0b00010, "uaddlp",
2766 int_aarch64_neon_uaddlp>;
2767 defm UCVTF : SIMDTwoVectorIntToFP<1, 0, 0b11101, "ucvtf", uint_to_fp>;
2768 defm UQXTN : SIMDMixedTwoVector<1, 0b10100, "uqxtn", int_aarch64_neon_uqxtn>;
2769 defm URECPE : SIMDTwoVectorS<0, 1, 0b11100, "urecpe", int_aarch64_neon_urecpe>;
2770 defm URSQRTE: SIMDTwoVectorS<1, 1, 0b11100, "ursqrte", int_aarch64_neon_ursqrte>;
2771 defm USQADD : SIMDTwoVectorBHSDTied<1, 0b00011, "usqadd",int_aarch64_neon_usqadd>;
2772 defm XTN : SIMDMixedTwoVector<0, 0b10010, "xtn", trunc>;
2774 def : Pat<(v4f16 (AArch64rev32 V64:$Rn)), (REV32v4i16 V64:$Rn)>;
2775 def : Pat<(v4f16 (AArch64rev64 V64:$Rn)), (REV64v4i16 V64:$Rn)>;
2776 def : Pat<(v8f16 (AArch64rev32 V128:$Rn)), (REV32v8i16 V128:$Rn)>;
2777 def : Pat<(v8f16 (AArch64rev64 V128:$Rn)), (REV64v8i16 V128:$Rn)>;
2778 def : Pat<(v2f32 (AArch64rev64 V64:$Rn)), (REV64v2i32 V64:$Rn)>;
2779 def : Pat<(v4f32 (AArch64rev64 V128:$Rn)), (REV64v4i32 V128:$Rn)>;
2781 // Patterns for vector long shift (by element width). These need to match all
2782 // three of zext, sext and anyext so it's easier to pull the patterns out of the
2784 multiclass SIMDVectorLShiftLongBySizeBHSPats<SDPatternOperator ext> {
2785 def : Pat<(AArch64vshl (v8i16 (ext (v8i8 V64:$Rn))), (i32 8)),
2786 (SHLLv8i8 V64:$Rn)>;
2787 def : Pat<(AArch64vshl (v8i16 (ext (extract_high_v16i8 V128:$Rn))), (i32 8)),
2788 (SHLLv16i8 V128:$Rn)>;
2789 def : Pat<(AArch64vshl (v4i32 (ext (v4i16 V64:$Rn))), (i32 16)),
2790 (SHLLv4i16 V64:$Rn)>;
2791 def : Pat<(AArch64vshl (v4i32 (ext (extract_high_v8i16 V128:$Rn))), (i32 16)),
2792 (SHLLv8i16 V128:$Rn)>;
2793 def : Pat<(AArch64vshl (v2i64 (ext (v2i32 V64:$Rn))), (i32 32)),
2794 (SHLLv2i32 V64:$Rn)>;
2795 def : Pat<(AArch64vshl (v2i64 (ext (extract_high_v4i32 V128:$Rn))), (i32 32)),
2796 (SHLLv4i32 V128:$Rn)>;
2799 defm : SIMDVectorLShiftLongBySizeBHSPats<anyext>;
2800 defm : SIMDVectorLShiftLongBySizeBHSPats<zext>;
2801 defm : SIMDVectorLShiftLongBySizeBHSPats<sext>;
2803 //===----------------------------------------------------------------------===//
2804 // Advanced SIMD three vector instructions.
2805 //===----------------------------------------------------------------------===//
2807 defm ADD : SIMDThreeSameVector<0, 0b10000, "add", add>;
2808 defm ADDP : SIMDThreeSameVector<0, 0b10111, "addp", int_aarch64_neon_addp>;
2809 defm CMEQ : SIMDThreeSameVector<1, 0b10001, "cmeq", AArch64cmeq>;
2810 defm CMGE : SIMDThreeSameVector<0, 0b00111, "cmge", AArch64cmge>;
2811 defm CMGT : SIMDThreeSameVector<0, 0b00110, "cmgt", AArch64cmgt>;
2812 defm CMHI : SIMDThreeSameVector<1, 0b00110, "cmhi", AArch64cmhi>;
2813 defm CMHS : SIMDThreeSameVector<1, 0b00111, "cmhs", AArch64cmhs>;
2814 defm CMTST : SIMDThreeSameVector<0, 0b10001, "cmtst", AArch64cmtst>;
2815 defm FABD : SIMDThreeSameVectorFP<1,1,0b11010,"fabd", int_aarch64_neon_fabd>;
2816 defm FACGE : SIMDThreeSameVectorFPCmp<1,0,0b11101,"facge",int_aarch64_neon_facge>;
2817 defm FACGT : SIMDThreeSameVectorFPCmp<1,1,0b11101,"facgt",int_aarch64_neon_facgt>;
2818 defm FADDP : SIMDThreeSameVectorFP<1,0,0b11010,"faddp",int_aarch64_neon_addp>;
2819 defm FADD : SIMDThreeSameVectorFP<0,0,0b11010,"fadd", fadd>;
2820 defm FCMEQ : SIMDThreeSameVectorFPCmp<0, 0, 0b11100, "fcmeq", AArch64fcmeq>;
2821 defm FCMGE : SIMDThreeSameVectorFPCmp<1, 0, 0b11100, "fcmge", AArch64fcmge>;
2822 defm FCMGT : SIMDThreeSameVectorFPCmp<1, 1, 0b11100, "fcmgt", AArch64fcmgt>;
2823 defm FDIV : SIMDThreeSameVectorFP<1,0,0b11111,"fdiv", fdiv>;
2824 defm FMAXNMP : SIMDThreeSameVectorFP<1,0,0b11000,"fmaxnmp", int_aarch64_neon_fmaxnmp>;
2825 defm FMAXNM : SIMDThreeSameVectorFP<0,0,0b11000,"fmaxnm", fmaxnum>;
2826 defm FMAXP : SIMDThreeSameVectorFP<1,0,0b11110,"fmaxp", int_aarch64_neon_fmaxp>;
2827 defm FMAX : SIMDThreeSameVectorFP<0,0,0b11110,"fmax", fmaxnan>;
2828 defm FMINNMP : SIMDThreeSameVectorFP<1,1,0b11000,"fminnmp", int_aarch64_neon_fminnmp>;
2829 defm FMINNM : SIMDThreeSameVectorFP<0,1,0b11000,"fminnm", fminnum>;
2830 defm FMINP : SIMDThreeSameVectorFP<1,1,0b11110,"fminp", int_aarch64_neon_fminp>;
2831 defm FMIN : SIMDThreeSameVectorFP<0,1,0b11110,"fmin", fminnan>;
2833 // NOTE: The operands of the PatFrag are reordered on FMLA/FMLS because the
2834 // instruction expects the addend first, while the fma intrinsic puts it last.
2835 defm FMLA : SIMDThreeSameVectorFPTied<0, 0, 0b11001, "fmla",
2836 TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)> >;
2837 defm FMLS : SIMDThreeSameVectorFPTied<0, 1, 0b11001, "fmls",
2838 TriOpFrag<(fma node:$MHS, (fneg node:$RHS), node:$LHS)> >;
2840 // The following def pats catch the case where the LHS of an FMA is negated.
2841 // The TriOpFrag above catches the case where the middle operand is negated.
2842 def : Pat<(v2f32 (fma (fneg V64:$Rn), V64:$Rm, V64:$Rd)),
2843 (FMLSv2f32 V64:$Rd, V64:$Rn, V64:$Rm)>;
2845 def : Pat<(v4f32 (fma (fneg V128:$Rn), V128:$Rm, V128:$Rd)),
2846 (FMLSv4f32 V128:$Rd, V128:$Rn, V128:$Rm)>;
2848 def : Pat<(v2f64 (fma (fneg V128:$Rn), V128:$Rm, V128:$Rd)),
2849 (FMLSv2f64 V128:$Rd, V128:$Rn, V128:$Rm)>;
2851 defm FMULX : SIMDThreeSameVectorFP<0,0,0b11011,"fmulx", int_aarch64_neon_fmulx>;
2852 defm FMUL : SIMDThreeSameVectorFP<1,0,0b11011,"fmul", fmul>;
2853 defm FRECPS : SIMDThreeSameVectorFP<0,0,0b11111,"frecps", int_aarch64_neon_frecps>;
2854 defm FRSQRTS : SIMDThreeSameVectorFP<0,1,0b11111,"frsqrts", int_aarch64_neon_frsqrts>;
2855 defm FSUB : SIMDThreeSameVectorFP<0,1,0b11010,"fsub", fsub>;
2856 defm MLA : SIMDThreeSameVectorBHSTied<0, 0b10010, "mla",
2857 TriOpFrag<(add node:$LHS, (mul node:$MHS, node:$RHS))> >;
2858 defm MLS : SIMDThreeSameVectorBHSTied<1, 0b10010, "mls",
2859 TriOpFrag<(sub node:$LHS, (mul node:$MHS, node:$RHS))> >;
2860 defm MUL : SIMDThreeSameVectorBHS<0, 0b10011, "mul", mul>;
2861 defm PMUL : SIMDThreeSameVectorB<1, 0b10011, "pmul", int_aarch64_neon_pmul>;
2862 defm SABA : SIMDThreeSameVectorBHSTied<0, 0b01111, "saba",
2863 TriOpFrag<(add node:$LHS, (sabsdiff node:$MHS, node:$RHS))> >;
2864 defm SABD : SIMDThreeSameVectorBHS<0,0b01110,"sabd", sabsdiff>;
2865 defm SHADD : SIMDThreeSameVectorBHS<0,0b00000,"shadd", int_aarch64_neon_shadd>;
2866 defm SHSUB : SIMDThreeSameVectorBHS<0,0b00100,"shsub", int_aarch64_neon_shsub>;
2867 defm SMAXP : SIMDThreeSameVectorBHS<0,0b10100,"smaxp", int_aarch64_neon_smaxp>;
2868 defm SMAX : SIMDThreeSameVectorBHS<0,0b01100,"smax", smax>;
2869 defm SMINP : SIMDThreeSameVectorBHS<0,0b10101,"sminp", int_aarch64_neon_sminp>;
2870 defm SMIN : SIMDThreeSameVectorBHS<0,0b01101,"smin", smin>;
2871 defm SQADD : SIMDThreeSameVector<0,0b00001,"sqadd", int_aarch64_neon_sqadd>;
2872 defm SQDMULH : SIMDThreeSameVectorHS<0,0b10110,"sqdmulh",int_aarch64_neon_sqdmulh>;
2873 defm SQRDMULH : SIMDThreeSameVectorHS<1,0b10110,"sqrdmulh",int_aarch64_neon_sqrdmulh>;
2874 defm SQRSHL : SIMDThreeSameVector<0,0b01011,"sqrshl", int_aarch64_neon_sqrshl>;
2875 defm SQSHL : SIMDThreeSameVector<0,0b01001,"sqshl", int_aarch64_neon_sqshl>;
2876 defm SQSUB : SIMDThreeSameVector<0,0b00101,"sqsub", int_aarch64_neon_sqsub>;
2877 defm SRHADD : SIMDThreeSameVectorBHS<0,0b00010,"srhadd",int_aarch64_neon_srhadd>;
2878 defm SRSHL : SIMDThreeSameVector<0,0b01010,"srshl", int_aarch64_neon_srshl>;
2879 defm SSHL : SIMDThreeSameVector<0,0b01000,"sshl", int_aarch64_neon_sshl>;
2880 defm SUB : SIMDThreeSameVector<1,0b10000,"sub", sub>;
2881 defm UABA : SIMDThreeSameVectorBHSTied<1, 0b01111, "uaba",
2882 TriOpFrag<(add node:$LHS, (uabsdiff node:$MHS, node:$RHS))> >;
2883 defm UABD : SIMDThreeSameVectorBHS<1,0b01110,"uabd", uabsdiff>;
2884 defm UHADD : SIMDThreeSameVectorBHS<1,0b00000,"uhadd", int_aarch64_neon_uhadd>;
2885 defm UHSUB : SIMDThreeSameVectorBHS<1,0b00100,"uhsub", int_aarch64_neon_uhsub>;
2886 defm UMAXP : SIMDThreeSameVectorBHS<1,0b10100,"umaxp", int_aarch64_neon_umaxp>;
2887 defm UMAX : SIMDThreeSameVectorBHS<1,0b01100,"umax", umax>;
2888 defm UMINP : SIMDThreeSameVectorBHS<1,0b10101,"uminp", int_aarch64_neon_uminp>;
2889 defm UMIN : SIMDThreeSameVectorBHS<1,0b01101,"umin", umin>;
2890 defm UQADD : SIMDThreeSameVector<1,0b00001,"uqadd", int_aarch64_neon_uqadd>;
2891 defm UQRSHL : SIMDThreeSameVector<1,0b01011,"uqrshl", int_aarch64_neon_uqrshl>;
2892 defm UQSHL : SIMDThreeSameVector<1,0b01001,"uqshl", int_aarch64_neon_uqshl>;
2893 defm UQSUB : SIMDThreeSameVector<1,0b00101,"uqsub", int_aarch64_neon_uqsub>;
2894 defm URHADD : SIMDThreeSameVectorBHS<1,0b00010,"urhadd", int_aarch64_neon_urhadd>;
2895 defm URSHL : SIMDThreeSameVector<1,0b01010,"urshl", int_aarch64_neon_urshl>;
2896 defm USHL : SIMDThreeSameVector<1,0b01000,"ushl", int_aarch64_neon_ushl>;
2897 defm SQRDMLAH : SIMDThreeSameVectorSQRDMLxHTiedHS<1,0b10000,"sqrdmlah",
2898 int_aarch64_neon_sqadd>;
2899 defm SQRDMLSH : SIMDThreeSameVectorSQRDMLxHTiedHS<1,0b10001,"sqrdmlsh",
2900 int_aarch64_neon_sqsub>;
2902 defm AND : SIMDLogicalThreeVector<0, 0b00, "and", and>;
2903 defm BIC : SIMDLogicalThreeVector<0, 0b01, "bic",
2904 BinOpFrag<(and node:$LHS, (vnot node:$RHS))> >;
2905 defm BIF : SIMDLogicalThreeVector<1, 0b11, "bif">;
2906 defm BIT : SIMDLogicalThreeVectorTied<1, 0b10, "bit", AArch64bit>;
2907 defm BSL : SIMDLogicalThreeVectorTied<1, 0b01, "bsl",
2908 TriOpFrag<(or (and node:$LHS, node:$MHS), (and (vnot node:$LHS), node:$RHS))>>;
2909 defm EOR : SIMDLogicalThreeVector<1, 0b00, "eor", xor>;
2910 defm ORN : SIMDLogicalThreeVector<0, 0b11, "orn",
2911 BinOpFrag<(or node:$LHS, (vnot node:$RHS))> >;
2912 defm ORR : SIMDLogicalThreeVector<0, 0b10, "orr", or>;
2915 def : Pat<(AArch64bsl (v8i8 V64:$Rd), V64:$Rn, V64:$Rm),
2916 (BSLv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
2917 def : Pat<(AArch64bsl (v4i16 V64:$Rd), V64:$Rn, V64:$Rm),
2918 (BSLv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
2919 def : Pat<(AArch64bsl (v2i32 V64:$Rd), V64:$Rn, V64:$Rm),
2920 (BSLv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
2921 def : Pat<(AArch64bsl (v1i64 V64:$Rd), V64:$Rn, V64:$Rm),
2922 (BSLv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
2924 def : Pat<(AArch64bsl (v16i8 V128:$Rd), V128:$Rn, V128:$Rm),
2925 (BSLv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
2926 def : Pat<(AArch64bsl (v8i16 V128:$Rd), V128:$Rn, V128:$Rm),
2927 (BSLv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
2928 def : Pat<(AArch64bsl (v4i32 V128:$Rd), V128:$Rn, V128:$Rm),
2929 (BSLv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
2930 def : Pat<(AArch64bsl (v2i64 V128:$Rd), V128:$Rn, V128:$Rm),
2931 (BSLv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
2933 def : InstAlias<"mov{\t$dst.16b, $src.16b|.16b\t$dst, $src}",
2934 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 1>;
2935 def : InstAlias<"mov{\t$dst.8h, $src.8h|.8h\t$dst, $src}",
2936 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2937 def : InstAlias<"mov{\t$dst.4s, $src.4s|.4s\t$dst, $src}",
2938 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2939 def : InstAlias<"mov{\t$dst.2d, $src.2d|.2d\t$dst, $src}",
2940 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2942 def : InstAlias<"mov{\t$dst.8b, $src.8b|.8b\t$dst, $src}",
2943 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 1>;
2944 def : InstAlias<"mov{\t$dst.4h, $src.4h|.4h\t$dst, $src}",
2945 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2946 def : InstAlias<"mov{\t$dst.2s, $src.2s|.2s\t$dst, $src}",
2947 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2948 def : InstAlias<"mov{\t$dst.1d, $src.1d|.1d\t$dst, $src}",
2949 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2951 def : InstAlias<"{cmls\t$dst.8b, $src1.8b, $src2.8b" #
2952 "|cmls.8b\t$dst, $src1, $src2}",
2953 (CMHSv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2954 def : InstAlias<"{cmls\t$dst.16b, $src1.16b, $src2.16b" #
2955 "|cmls.16b\t$dst, $src1, $src2}",
2956 (CMHSv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2957 def : InstAlias<"{cmls\t$dst.4h, $src1.4h, $src2.4h" #
2958 "|cmls.4h\t$dst, $src1, $src2}",
2959 (CMHSv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2960 def : InstAlias<"{cmls\t$dst.8h, $src1.8h, $src2.8h" #
2961 "|cmls.8h\t$dst, $src1, $src2}",
2962 (CMHSv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2963 def : InstAlias<"{cmls\t$dst.2s, $src1.2s, $src2.2s" #
2964 "|cmls.2s\t$dst, $src1, $src2}",
2965 (CMHSv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2966 def : InstAlias<"{cmls\t$dst.4s, $src1.4s, $src2.4s" #
2967 "|cmls.4s\t$dst, $src1, $src2}",
2968 (CMHSv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2969 def : InstAlias<"{cmls\t$dst.2d, $src1.2d, $src2.2d" #
2970 "|cmls.2d\t$dst, $src1, $src2}",
2971 (CMHSv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2973 def : InstAlias<"{cmlo\t$dst.8b, $src1.8b, $src2.8b" #
2974 "|cmlo.8b\t$dst, $src1, $src2}",
2975 (CMHIv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2976 def : InstAlias<"{cmlo\t$dst.16b, $src1.16b, $src2.16b" #
2977 "|cmlo.16b\t$dst, $src1, $src2}",
2978 (CMHIv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2979 def : InstAlias<"{cmlo\t$dst.4h, $src1.4h, $src2.4h" #
2980 "|cmlo.4h\t$dst, $src1, $src2}",
2981 (CMHIv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2982 def : InstAlias<"{cmlo\t$dst.8h, $src1.8h, $src2.8h" #
2983 "|cmlo.8h\t$dst, $src1, $src2}",
2984 (CMHIv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2985 def : InstAlias<"{cmlo\t$dst.2s, $src1.2s, $src2.2s" #
2986 "|cmlo.2s\t$dst, $src1, $src2}",
2987 (CMHIv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2988 def : InstAlias<"{cmlo\t$dst.4s, $src1.4s, $src2.4s" #
2989 "|cmlo.4s\t$dst, $src1, $src2}",
2990 (CMHIv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2991 def : InstAlias<"{cmlo\t$dst.2d, $src1.2d, $src2.2d" #
2992 "|cmlo.2d\t$dst, $src1, $src2}",
2993 (CMHIv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2995 def : InstAlias<"{cmle\t$dst.8b, $src1.8b, $src2.8b" #
2996 "|cmle.8b\t$dst, $src1, $src2}",
2997 (CMGEv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2998 def : InstAlias<"{cmle\t$dst.16b, $src1.16b, $src2.16b" #
2999 "|cmle.16b\t$dst, $src1, $src2}",
3000 (CMGEv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
3001 def : InstAlias<"{cmle\t$dst.4h, $src1.4h, $src2.4h" #
3002 "|cmle.4h\t$dst, $src1, $src2}",
3003 (CMGEv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
3004 def : InstAlias<"{cmle\t$dst.8h, $src1.8h, $src2.8h" #
3005 "|cmle.8h\t$dst, $src1, $src2}",
3006 (CMGEv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
3007 def : InstAlias<"{cmle\t$dst.2s, $src1.2s, $src2.2s" #
3008 "|cmle.2s\t$dst, $src1, $src2}",
3009 (CMGEv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
3010 def : InstAlias<"{cmle\t$dst.4s, $src1.4s, $src2.4s" #
3011 "|cmle.4s\t$dst, $src1, $src2}",
3012 (CMGEv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
3013 def : InstAlias<"{cmle\t$dst.2d, $src1.2d, $src2.2d" #
3014 "|cmle.2d\t$dst, $src1, $src2}",
3015 (CMGEv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
3017 def : InstAlias<"{cmlt\t$dst.8b, $src1.8b, $src2.8b" #
3018 "|cmlt.8b\t$dst, $src1, $src2}",
3019 (CMGTv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
3020 def : InstAlias<"{cmlt\t$dst.16b, $src1.16b, $src2.16b" #
3021 "|cmlt.16b\t$dst, $src1, $src2}",
3022 (CMGTv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
3023 def : InstAlias<"{cmlt\t$dst.4h, $src1.4h, $src2.4h" #
3024 "|cmlt.4h\t$dst, $src1, $src2}",
3025 (CMGTv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
3026 def : InstAlias<"{cmlt\t$dst.8h, $src1.8h, $src2.8h" #
3027 "|cmlt.8h\t$dst, $src1, $src2}",
3028 (CMGTv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
3029 def : InstAlias<"{cmlt\t$dst.2s, $src1.2s, $src2.2s" #
3030 "|cmlt.2s\t$dst, $src1, $src2}",
3031 (CMGTv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
3032 def : InstAlias<"{cmlt\t$dst.4s, $src1.4s, $src2.4s" #
3033 "|cmlt.4s\t$dst, $src1, $src2}",
3034 (CMGTv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
3035 def : InstAlias<"{cmlt\t$dst.2d, $src1.2d, $src2.2d" #
3036 "|cmlt.2d\t$dst, $src1, $src2}",
3037 (CMGTv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
3039 def : InstAlias<"{fcmle\t$dst.2s, $src1.2s, $src2.2s" #
3040 "|fcmle.2s\t$dst, $src1, $src2}",
3041 (FCMGEv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
3042 def : InstAlias<"{fcmle\t$dst.4s, $src1.4s, $src2.4s" #
3043 "|fcmle.4s\t$dst, $src1, $src2}",
3044 (FCMGEv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
3045 def : InstAlias<"{fcmle\t$dst.2d, $src1.2d, $src2.2d" #
3046 "|fcmle.2d\t$dst, $src1, $src2}",
3047 (FCMGEv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
3049 def : InstAlias<"{fcmlt\t$dst.2s, $src1.2s, $src2.2s" #
3050 "|fcmlt.2s\t$dst, $src1, $src2}",
3051 (FCMGTv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
3052 def : InstAlias<"{fcmlt\t$dst.4s, $src1.4s, $src2.4s" #
3053 "|fcmlt.4s\t$dst, $src1, $src2}",
3054 (FCMGTv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
3055 def : InstAlias<"{fcmlt\t$dst.2d, $src1.2d, $src2.2d" #
3056 "|fcmlt.2d\t$dst, $src1, $src2}",
3057 (FCMGTv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
3059 def : InstAlias<"{facle\t$dst.2s, $src1.2s, $src2.2s" #
3060 "|facle.2s\t$dst, $src1, $src2}",
3061 (FACGEv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
3062 def : InstAlias<"{facle\t$dst.4s, $src1.4s, $src2.4s" #
3063 "|facle.4s\t$dst, $src1, $src2}",
3064 (FACGEv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
3065 def : InstAlias<"{facle\t$dst.2d, $src1.2d, $src2.2d" #
3066 "|facle.2d\t$dst, $src1, $src2}",
3067 (FACGEv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
3069 def : InstAlias<"{faclt\t$dst.2s, $src1.2s, $src2.2s" #
3070 "|faclt.2s\t$dst, $src1, $src2}",
3071 (FACGTv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
3072 def : InstAlias<"{faclt\t$dst.4s, $src1.4s, $src2.4s" #
3073 "|faclt.4s\t$dst, $src1, $src2}",
3074 (FACGTv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
3075 def : InstAlias<"{faclt\t$dst.2d, $src1.2d, $src2.2d" #
3076 "|faclt.2d\t$dst, $src1, $src2}",
3077 (FACGTv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
3079 //===----------------------------------------------------------------------===//
3080 // Advanced SIMD three scalar instructions.
3081 //===----------------------------------------------------------------------===//
3083 defm ADD : SIMDThreeScalarD<0, 0b10000, "add", add>;
3084 defm CMEQ : SIMDThreeScalarD<1, 0b10001, "cmeq", AArch64cmeq>;
3085 defm CMGE : SIMDThreeScalarD<0, 0b00111, "cmge", AArch64cmge>;
3086 defm CMGT : SIMDThreeScalarD<0, 0b00110, "cmgt", AArch64cmgt>;
3087 defm CMHI : SIMDThreeScalarD<1, 0b00110, "cmhi", AArch64cmhi>;
3088 defm CMHS : SIMDThreeScalarD<1, 0b00111, "cmhs", AArch64cmhs>;
3089 defm CMTST : SIMDThreeScalarD<0, 0b10001, "cmtst", AArch64cmtst>;
3090 defm FABD : SIMDThreeScalarSD<1, 1, 0b11010, "fabd", int_aarch64_sisd_fabd>;
3091 def : Pat<(v1f64 (int_aarch64_neon_fabd (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
3092 (FABD64 FPR64:$Rn, FPR64:$Rm)>;
3093 defm FACGE : SIMDThreeScalarFPCmp<1, 0, 0b11101, "facge",
3094 int_aarch64_neon_facge>;
3095 defm FACGT : SIMDThreeScalarFPCmp<1, 1, 0b11101, "facgt",
3096 int_aarch64_neon_facgt>;
3097 defm FCMEQ : SIMDThreeScalarFPCmp<0, 0, 0b11100, "fcmeq", AArch64fcmeq>;
3098 defm FCMGE : SIMDThreeScalarFPCmp<1, 0, 0b11100, "fcmge", AArch64fcmge>;
3099 defm FCMGT : SIMDThreeScalarFPCmp<1, 1, 0b11100, "fcmgt", AArch64fcmgt>;
3100 defm FMULX : SIMDThreeScalarSD<0, 0, 0b11011, "fmulx", int_aarch64_neon_fmulx>;
3101 defm FRECPS : SIMDThreeScalarSD<0, 0, 0b11111, "frecps", int_aarch64_neon_frecps>;
3102 defm FRSQRTS : SIMDThreeScalarSD<0, 1, 0b11111, "frsqrts", int_aarch64_neon_frsqrts>;
3103 defm SQADD : SIMDThreeScalarBHSD<0, 0b00001, "sqadd", int_aarch64_neon_sqadd>;
3104 defm SQDMULH : SIMDThreeScalarHS< 0, 0b10110, "sqdmulh", int_aarch64_neon_sqdmulh>;
3105 defm SQRDMULH : SIMDThreeScalarHS< 1, 0b10110, "sqrdmulh", int_aarch64_neon_sqrdmulh>;
3106 defm SQRSHL : SIMDThreeScalarBHSD<0, 0b01011, "sqrshl",int_aarch64_neon_sqrshl>;
3107 defm SQSHL : SIMDThreeScalarBHSD<0, 0b01001, "sqshl", int_aarch64_neon_sqshl>;
3108 defm SQSUB : SIMDThreeScalarBHSD<0, 0b00101, "sqsub", int_aarch64_neon_sqsub>;
3109 defm SRSHL : SIMDThreeScalarD< 0, 0b01010, "srshl", int_aarch64_neon_srshl>;
3110 defm SSHL : SIMDThreeScalarD< 0, 0b01000, "sshl", int_aarch64_neon_sshl>;
3111 defm SUB : SIMDThreeScalarD< 1, 0b10000, "sub", sub>;
3112 defm UQADD : SIMDThreeScalarBHSD<1, 0b00001, "uqadd", int_aarch64_neon_uqadd>;
3113 defm UQRSHL : SIMDThreeScalarBHSD<1, 0b01011, "uqrshl",int_aarch64_neon_uqrshl>;
3114 defm UQSHL : SIMDThreeScalarBHSD<1, 0b01001, "uqshl", int_aarch64_neon_uqshl>;
3115 defm UQSUB : SIMDThreeScalarBHSD<1, 0b00101, "uqsub", int_aarch64_neon_uqsub>;
3116 defm URSHL : SIMDThreeScalarD< 1, 0b01010, "urshl", int_aarch64_neon_urshl>;
3117 defm USHL : SIMDThreeScalarD< 1, 0b01000, "ushl", int_aarch64_neon_ushl>;
3118 let Predicates = [HasV8_1a] in {
3119 defm SQRDMLAH : SIMDThreeScalarHSTied<1, 0, 0b10000, "sqrdmlah">;
3120 defm SQRDMLSH : SIMDThreeScalarHSTied<1, 0, 0b10001, "sqrdmlsh">;
3121 def : Pat<(i32 (int_aarch64_neon_sqadd
3123 (i32 (int_aarch64_neon_sqrdmulh (i32 FPR32:$Rn),
3124 (i32 FPR32:$Rm))))),
3125 (SQRDMLAHv1i32 FPR32:$Rd, FPR32:$Rn, FPR32:$Rm)>;
3126 def : Pat<(i32 (int_aarch64_neon_sqsub
3128 (i32 (int_aarch64_neon_sqrdmulh (i32 FPR32:$Rn),
3129 (i32 FPR32:$Rm))))),
3130 (SQRDMLSHv1i32 FPR32:$Rd, FPR32:$Rn, FPR32:$Rm)>;
3133 def : InstAlias<"cmls $dst, $src1, $src2",
3134 (CMHSv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
3135 def : InstAlias<"cmle $dst, $src1, $src2",
3136 (CMGEv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
3137 def : InstAlias<"cmlo $dst, $src1, $src2",
3138 (CMHIv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
3139 def : InstAlias<"cmlt $dst, $src1, $src2",
3140 (CMGTv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
3141 def : InstAlias<"fcmle $dst, $src1, $src2",
3142 (FCMGE32 FPR32:$dst, FPR32:$src2, FPR32:$src1), 0>;
3143 def : InstAlias<"fcmle $dst, $src1, $src2",
3144 (FCMGE64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
3145 def : InstAlias<"fcmlt $dst, $src1, $src2",
3146 (FCMGT32 FPR32:$dst, FPR32:$src2, FPR32:$src1), 0>;
3147 def : InstAlias<"fcmlt $dst, $src1, $src2",
3148 (FCMGT64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
3149 def : InstAlias<"facle $dst, $src1, $src2",
3150 (FACGE32 FPR32:$dst, FPR32:$src2, FPR32:$src1), 0>;
3151 def : InstAlias<"facle $dst, $src1, $src2",
3152 (FACGE64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
3153 def : InstAlias<"faclt $dst, $src1, $src2",
3154 (FACGT32 FPR32:$dst, FPR32:$src2, FPR32:$src1), 0>;
3155 def : InstAlias<"faclt $dst, $src1, $src2",
3156 (FACGT64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
3158 //===----------------------------------------------------------------------===//
3159 // Advanced SIMD three scalar instructions (mixed operands).
3160 //===----------------------------------------------------------------------===//
3161 defm SQDMULL : SIMDThreeScalarMixedHS<0, 0b11010, "sqdmull",
3162 int_aarch64_neon_sqdmulls_scalar>;
3163 defm SQDMLAL : SIMDThreeScalarMixedTiedHS<0, 0b10010, "sqdmlal">;
3164 defm SQDMLSL : SIMDThreeScalarMixedTiedHS<0, 0b10110, "sqdmlsl">;
3166 def : Pat<(i64 (int_aarch64_neon_sqadd (i64 FPR64:$Rd),
3167 (i64 (int_aarch64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
3168 (i32 FPR32:$Rm))))),
3169 (SQDMLALi32 FPR64:$Rd, FPR32:$Rn, FPR32:$Rm)>;
3170 def : Pat<(i64 (int_aarch64_neon_sqsub (i64 FPR64:$Rd),
3171 (i64 (int_aarch64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
3172 (i32 FPR32:$Rm))))),
3173 (SQDMLSLi32 FPR64:$Rd, FPR32:$Rn, FPR32:$Rm)>;
3175 //===----------------------------------------------------------------------===//
3176 // Advanced SIMD two scalar instructions.
3177 //===----------------------------------------------------------------------===//
3179 defm ABS : SIMDTwoScalarD< 0, 0b01011, "abs", int_aarch64_neon_abs>;
3180 defm CMEQ : SIMDCmpTwoScalarD< 0, 0b01001, "cmeq", AArch64cmeqz>;
3181 defm CMGE : SIMDCmpTwoScalarD< 1, 0b01000, "cmge", AArch64cmgez>;
3182 defm CMGT : SIMDCmpTwoScalarD< 0, 0b01000, "cmgt", AArch64cmgtz>;
3183 defm CMLE : SIMDCmpTwoScalarD< 1, 0b01001, "cmle", AArch64cmlez>;
3184 defm CMLT : SIMDCmpTwoScalarD< 0, 0b01010, "cmlt", AArch64cmltz>;
3185 defm FCMEQ : SIMDFPCmpTwoScalar<0, 1, 0b01101, "fcmeq", AArch64fcmeqz>;
3186 defm FCMGE : SIMDFPCmpTwoScalar<1, 1, 0b01100, "fcmge", AArch64fcmgez>;
3187 defm FCMGT : SIMDFPCmpTwoScalar<0, 1, 0b01100, "fcmgt", AArch64fcmgtz>;
3188 defm FCMLE : SIMDFPCmpTwoScalar<1, 1, 0b01101, "fcmle", AArch64fcmlez>;
3189 defm FCMLT : SIMDFPCmpTwoScalar<0, 1, 0b01110, "fcmlt", AArch64fcmltz>;
3190 defm FCVTAS : SIMDFPTwoScalar< 0, 0, 0b11100, "fcvtas">;
3191 defm FCVTAU : SIMDFPTwoScalar< 1, 0, 0b11100, "fcvtau">;
3192 defm FCVTMS : SIMDFPTwoScalar< 0, 0, 0b11011, "fcvtms">;
3193 defm FCVTMU : SIMDFPTwoScalar< 1, 0, 0b11011, "fcvtmu">;
3194 defm FCVTNS : SIMDFPTwoScalar< 0, 0, 0b11010, "fcvtns">;
3195 defm FCVTNU : SIMDFPTwoScalar< 1, 0, 0b11010, "fcvtnu">;
3196 defm FCVTPS : SIMDFPTwoScalar< 0, 1, 0b11010, "fcvtps">;
3197 defm FCVTPU : SIMDFPTwoScalar< 1, 1, 0b11010, "fcvtpu">;
3198 def FCVTXNv1i64 : SIMDInexactCvtTwoScalar<0b10110, "fcvtxn">;
3199 defm FCVTZS : SIMDFPTwoScalar< 0, 1, 0b11011, "fcvtzs">;
3200 defm FCVTZU : SIMDFPTwoScalar< 1, 1, 0b11011, "fcvtzu">;
3201 defm FRECPE : SIMDFPTwoScalar< 0, 1, 0b11101, "frecpe">;
3202 defm FRECPX : SIMDFPTwoScalar< 0, 1, 0b11111, "frecpx">;
3203 defm FRSQRTE : SIMDFPTwoScalar< 1, 1, 0b11101, "frsqrte">;
3204 defm NEG : SIMDTwoScalarD< 1, 0b01011, "neg",
3205 UnOpFrag<(sub immAllZerosV, node:$LHS)> >;
3206 defm SCVTF : SIMDTwoScalarCVTSD< 0, 0, 0b11101, "scvtf", AArch64sitof>;
3207 defm SQABS : SIMDTwoScalarBHSD< 0, 0b00111, "sqabs", int_aarch64_neon_sqabs>;
3208 defm SQNEG : SIMDTwoScalarBHSD< 1, 0b00111, "sqneg", int_aarch64_neon_sqneg>;
3209 defm SQXTN : SIMDTwoScalarMixedBHS< 0, 0b10100, "sqxtn", int_aarch64_neon_scalar_sqxtn>;
3210 defm SQXTUN : SIMDTwoScalarMixedBHS< 1, 0b10010, "sqxtun", int_aarch64_neon_scalar_sqxtun>;
3211 defm SUQADD : SIMDTwoScalarBHSDTied< 0, 0b00011, "suqadd",
3212 int_aarch64_neon_suqadd>;
3213 defm UCVTF : SIMDTwoScalarCVTSD< 1, 0, 0b11101, "ucvtf", AArch64uitof>;
3214 defm UQXTN : SIMDTwoScalarMixedBHS<1, 0b10100, "uqxtn", int_aarch64_neon_scalar_uqxtn>;
3215 defm USQADD : SIMDTwoScalarBHSDTied< 1, 0b00011, "usqadd",
3216 int_aarch64_neon_usqadd>;
3218 def : Pat<(AArch64neg (v1i64 V64:$Rn)), (NEGv1i64 V64:$Rn)>;
3220 def : Pat<(v1i64 (int_aarch64_neon_fcvtas (v1f64 FPR64:$Rn))),
3221 (FCVTASv1i64 FPR64:$Rn)>;
3222 def : Pat<(v1i64 (int_aarch64_neon_fcvtau (v1f64 FPR64:$Rn))),
3223 (FCVTAUv1i64 FPR64:$Rn)>;
3224 def : Pat<(v1i64 (int_aarch64_neon_fcvtms (v1f64 FPR64:$Rn))),
3225 (FCVTMSv1i64 FPR64:$Rn)>;
3226 def : Pat<(v1i64 (int_aarch64_neon_fcvtmu (v1f64 FPR64:$Rn))),
3227 (FCVTMUv1i64 FPR64:$Rn)>;
3228 def : Pat<(v1i64 (int_aarch64_neon_fcvtns (v1f64 FPR64:$Rn))),
3229 (FCVTNSv1i64 FPR64:$Rn)>;
3230 def : Pat<(v1i64 (int_aarch64_neon_fcvtnu (v1f64 FPR64:$Rn))),
3231 (FCVTNUv1i64 FPR64:$Rn)>;
3232 def : Pat<(v1i64 (int_aarch64_neon_fcvtps (v1f64 FPR64:$Rn))),
3233 (FCVTPSv1i64 FPR64:$Rn)>;
3234 def : Pat<(v1i64 (int_aarch64_neon_fcvtpu (v1f64 FPR64:$Rn))),
3235 (FCVTPUv1i64 FPR64:$Rn)>;
3237 def : Pat<(f32 (int_aarch64_neon_frecpe (f32 FPR32:$Rn))),
3238 (FRECPEv1i32 FPR32:$Rn)>;
3239 def : Pat<(f64 (int_aarch64_neon_frecpe (f64 FPR64:$Rn))),
3240 (FRECPEv1i64 FPR64:$Rn)>;
3241 def : Pat<(v1f64 (int_aarch64_neon_frecpe (v1f64 FPR64:$Rn))),
3242 (FRECPEv1i64 FPR64:$Rn)>;
3244 def : Pat<(f32 (int_aarch64_neon_frecpx (f32 FPR32:$Rn))),
3245 (FRECPXv1i32 FPR32:$Rn)>;
3246 def : Pat<(f64 (int_aarch64_neon_frecpx (f64 FPR64:$Rn))),
3247 (FRECPXv1i64 FPR64:$Rn)>;
3249 def : Pat<(f32 (int_aarch64_neon_frsqrte (f32 FPR32:$Rn))),
3250 (FRSQRTEv1i32 FPR32:$Rn)>;
3251 def : Pat<(f64 (int_aarch64_neon_frsqrte (f64 FPR64:$Rn))),
3252 (FRSQRTEv1i64 FPR64:$Rn)>;
3253 def : Pat<(v1f64 (int_aarch64_neon_frsqrte (v1f64 FPR64:$Rn))),
3254 (FRSQRTEv1i64 FPR64:$Rn)>;
3256 // If an integer is about to be converted to a floating point value,
3257 // just load it on the floating point unit.
3258 // Here are the patterns for 8 and 16-bits to float.
3260 multiclass UIntToFPROLoadPat<ValueType DstTy, ValueType SrcTy,
3261 SDPatternOperator loadop, Instruction UCVTF,
3262 ROAddrMode ro, Instruction LDRW, Instruction LDRX,
3264 def : Pat<(DstTy (uint_to_fp (SrcTy
3265 (loadop (ro.Wpat GPR64sp:$Rn, GPR32:$Rm,
3266 ro.Wext:$extend))))),
3267 (UCVTF (INSERT_SUBREG (DstTy (IMPLICIT_DEF)),
3268 (LDRW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend),
3271 def : Pat<(DstTy (uint_to_fp (SrcTy
3272 (loadop (ro.Xpat GPR64sp:$Rn, GPR64:$Rm,
3273 ro.Wext:$extend))))),
3274 (UCVTF (INSERT_SUBREG (DstTy (IMPLICIT_DEF)),
3275 (LDRX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend),
3279 defm : UIntToFPROLoadPat<f32, i32, zextloadi8,
3280 UCVTFv1i32, ro8, LDRBroW, LDRBroX, bsub>;
3281 def : Pat <(f32 (uint_to_fp (i32
3282 (zextloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))))),
3283 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
3284 (LDRBui GPR64sp:$Rn, uimm12s1:$offset), bsub))>;
3285 def : Pat <(f32 (uint_to_fp (i32
3286 (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))))),
3287 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
3288 (LDURBi GPR64sp:$Rn, simm9:$offset), bsub))>;
3289 // 16-bits -> float.
3290 defm : UIntToFPROLoadPat<f32, i32, zextloadi16,
3291 UCVTFv1i32, ro16, LDRHroW, LDRHroX, hsub>;
3292 def : Pat <(f32 (uint_to_fp (i32
3293 (zextloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))))),
3294 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
3295 (LDRHui GPR64sp:$Rn, uimm12s2:$offset), hsub))>;
3296 def : Pat <(f32 (uint_to_fp (i32
3297 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))))),
3298 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
3299 (LDURHi GPR64sp:$Rn, simm9:$offset), hsub))>;
3300 // 32-bits are handled in target specific dag combine:
3301 // performIntToFpCombine.
3302 // 64-bits integer to 32-bits floating point, not possible with
3303 // UCVTF on floating point registers (both source and destination
3304 // must have the same size).
3306 // Here are the patterns for 8, 16, 32, and 64-bits to double.
3307 // 8-bits -> double.
3308 defm : UIntToFPROLoadPat<f64, i32, zextloadi8,
3309 UCVTFv1i64, ro8, LDRBroW, LDRBroX, bsub>;
3310 def : Pat <(f64 (uint_to_fp (i32
3311 (zextloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))))),
3312 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3313 (LDRBui GPR64sp:$Rn, uimm12s1:$offset), bsub))>;
3314 def : Pat <(f64 (uint_to_fp (i32
3315 (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))))),
3316 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3317 (LDURBi GPR64sp:$Rn, simm9:$offset), bsub))>;
3318 // 16-bits -> double.
3319 defm : UIntToFPROLoadPat<f64, i32, zextloadi16,
3320 UCVTFv1i64, ro16, LDRHroW, LDRHroX, hsub>;
3321 def : Pat <(f64 (uint_to_fp (i32
3322 (zextloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))))),
3323 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3324 (LDRHui GPR64sp:$Rn, uimm12s2:$offset), hsub))>;
3325 def : Pat <(f64 (uint_to_fp (i32
3326 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))))),
3327 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3328 (LDURHi GPR64sp:$Rn, simm9:$offset), hsub))>;
3329 // 32-bits -> double.
3330 defm : UIntToFPROLoadPat<f64, i32, load,
3331 UCVTFv1i64, ro32, LDRSroW, LDRSroX, ssub>;
3332 def : Pat <(f64 (uint_to_fp (i32
3333 (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))))),
3334 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3335 (LDRSui GPR64sp:$Rn, uimm12s4:$offset), ssub))>;
3336 def : Pat <(f64 (uint_to_fp (i32
3337 (load (am_unscaled32 GPR64sp:$Rn, simm9:$offset))))),
3338 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3339 (LDURSi GPR64sp:$Rn, simm9:$offset), ssub))>;
3340 // 64-bits -> double are handled in target specific dag combine:
3341 // performIntToFpCombine.
3343 //===----------------------------------------------------------------------===//
3344 // Advanced SIMD three different-sized vector instructions.
3345 //===----------------------------------------------------------------------===//
3347 defm ADDHN : SIMDNarrowThreeVectorBHS<0,0b0100,"addhn", int_aarch64_neon_addhn>;
3348 defm SUBHN : SIMDNarrowThreeVectorBHS<0,0b0110,"subhn", int_aarch64_neon_subhn>;
3349 defm RADDHN : SIMDNarrowThreeVectorBHS<1,0b0100,"raddhn",int_aarch64_neon_raddhn>;
3350 defm RSUBHN : SIMDNarrowThreeVectorBHS<1,0b0110,"rsubhn",int_aarch64_neon_rsubhn>;
3351 defm PMULL : SIMDDifferentThreeVectorBD<0,0b1110,"pmull",int_aarch64_neon_pmull>;
3352 defm SABAL : SIMDLongThreeVectorTiedBHSabal<0,0b0101,"sabal",
3354 defm SABDL : SIMDLongThreeVectorBHSabdl<0, 0b0111, "sabdl",
3356 defm SADDL : SIMDLongThreeVectorBHS< 0, 0b0000, "saddl",
3357 BinOpFrag<(add (sext node:$LHS), (sext node:$RHS))>>;
3358 defm SADDW : SIMDWideThreeVectorBHS< 0, 0b0001, "saddw",
3359 BinOpFrag<(add node:$LHS, (sext node:$RHS))>>;
3360 defm SMLAL : SIMDLongThreeVectorTiedBHS<0, 0b1000, "smlal",
3361 TriOpFrag<(add node:$LHS, (int_aarch64_neon_smull node:$MHS, node:$RHS))>>;
3362 defm SMLSL : SIMDLongThreeVectorTiedBHS<0, 0b1010, "smlsl",
3363 TriOpFrag<(sub node:$LHS, (int_aarch64_neon_smull node:$MHS, node:$RHS))>>;
3364 defm SMULL : SIMDLongThreeVectorBHS<0, 0b1100, "smull", int_aarch64_neon_smull>;
3365 defm SQDMLAL : SIMDLongThreeVectorSQDMLXTiedHS<0, 0b1001, "sqdmlal",
3366 int_aarch64_neon_sqadd>;
3367 defm SQDMLSL : SIMDLongThreeVectorSQDMLXTiedHS<0, 0b1011, "sqdmlsl",
3368 int_aarch64_neon_sqsub>;
3369 defm SQDMULL : SIMDLongThreeVectorHS<0, 0b1101, "sqdmull",
3370 int_aarch64_neon_sqdmull>;
3371 defm SSUBL : SIMDLongThreeVectorBHS<0, 0b0010, "ssubl",
3372 BinOpFrag<(sub (sext node:$LHS), (sext node:$RHS))>>;
3373 defm SSUBW : SIMDWideThreeVectorBHS<0, 0b0011, "ssubw",
3374 BinOpFrag<(sub node:$LHS, (sext node:$RHS))>>;
3375 defm UABAL : SIMDLongThreeVectorTiedBHSabal<1, 0b0101, "uabal",
3377 defm UABDL : SIMDLongThreeVectorBHSabdl<1, 0b0111, "uabdl",
3379 defm UADDL : SIMDLongThreeVectorBHS<1, 0b0000, "uaddl",
3380 BinOpFrag<(add (zext node:$LHS), (zext node:$RHS))>>;
3381 defm UADDW : SIMDWideThreeVectorBHS<1, 0b0001, "uaddw",
3382 BinOpFrag<(add node:$LHS, (zext node:$RHS))>>;
3383 defm UMLAL : SIMDLongThreeVectorTiedBHS<1, 0b1000, "umlal",
3384 TriOpFrag<(add node:$LHS, (int_aarch64_neon_umull node:$MHS, node:$RHS))>>;
3385 defm UMLSL : SIMDLongThreeVectorTiedBHS<1, 0b1010, "umlsl",
3386 TriOpFrag<(sub node:$LHS, (int_aarch64_neon_umull node:$MHS, node:$RHS))>>;
3387 defm UMULL : SIMDLongThreeVectorBHS<1, 0b1100, "umull", int_aarch64_neon_umull>;
3388 defm USUBL : SIMDLongThreeVectorBHS<1, 0b0010, "usubl",
3389 BinOpFrag<(sub (zext node:$LHS), (zext node:$RHS))>>;
3390 defm USUBW : SIMDWideThreeVectorBHS< 1, 0b0011, "usubw",
3391 BinOpFrag<(sub node:$LHS, (zext node:$RHS))>>;
3393 // Additional patterns for SMULL and UMULL
3394 multiclass Neon_mul_widen_patterns<SDPatternOperator opnode,
3395 Instruction INST8B, Instruction INST4H, Instruction INST2S> {
3396 def : Pat<(v8i16 (opnode (v8i8 V64:$Rn), (v8i8 V64:$Rm))),
3397 (INST8B V64:$Rn, V64:$Rm)>;
3398 def : Pat<(v4i32 (opnode (v4i16 V64:$Rn), (v4i16 V64:$Rm))),
3399 (INST4H V64:$Rn, V64:$Rm)>;
3400 def : Pat<(v2i64 (opnode (v2i32 V64:$Rn), (v2i32 V64:$Rm))),
3401 (INST2S V64:$Rn, V64:$Rm)>;
3404 defm : Neon_mul_widen_patterns<AArch64smull, SMULLv8i8_v8i16,
3405 SMULLv4i16_v4i32, SMULLv2i32_v2i64>;
3406 defm : Neon_mul_widen_patterns<AArch64umull, UMULLv8i8_v8i16,
3407 UMULLv4i16_v4i32, UMULLv2i32_v2i64>;
3409 // Additional patterns for SMLAL/SMLSL and UMLAL/UMLSL
3410 multiclass Neon_mulacc_widen_patterns<SDPatternOperator opnode,
3411 Instruction INST8B, Instruction INST4H, Instruction INST2S> {
3412 def : Pat<(v8i16 (opnode (v8i16 V128:$Rd), (v8i8 V64:$Rn), (v8i8 V64:$Rm))),
3413 (INST8B V128:$Rd, V64:$Rn, V64:$Rm)>;
3414 def : Pat<(v4i32 (opnode (v4i32 V128:$Rd), (v4i16 V64:$Rn), (v4i16 V64:$Rm))),
3415 (INST4H V128:$Rd, V64:$Rn, V64:$Rm)>;
3416 def : Pat<(v2i64 (opnode (v2i64 V128:$Rd), (v2i32 V64:$Rn), (v2i32 V64:$Rm))),
3417 (INST2S V128:$Rd, V64:$Rn, V64:$Rm)>;
3420 defm : Neon_mulacc_widen_patterns<
3421 TriOpFrag<(add node:$LHS, (AArch64smull node:$MHS, node:$RHS))>,
3422 SMLALv8i8_v8i16, SMLALv4i16_v4i32, SMLALv2i32_v2i64>;
3423 defm : Neon_mulacc_widen_patterns<
3424 TriOpFrag<(add node:$LHS, (AArch64umull node:$MHS, node:$RHS))>,
3425 UMLALv8i8_v8i16, UMLALv4i16_v4i32, UMLALv2i32_v2i64>;
3426 defm : Neon_mulacc_widen_patterns<
3427 TriOpFrag<(sub node:$LHS, (AArch64smull node:$MHS, node:$RHS))>,
3428 SMLSLv8i8_v8i16, SMLSLv4i16_v4i32, SMLSLv2i32_v2i64>;
3429 defm : Neon_mulacc_widen_patterns<
3430 TriOpFrag<(sub node:$LHS, (AArch64umull node:$MHS, node:$RHS))>,
3431 UMLSLv8i8_v8i16, UMLSLv4i16_v4i32, UMLSLv2i32_v2i64>;
3433 // Patterns for 64-bit pmull
3434 def : Pat<(int_aarch64_neon_pmull64 V64:$Rn, V64:$Rm),
3435 (PMULLv1i64 V64:$Rn, V64:$Rm)>;
3436 def : Pat<(int_aarch64_neon_pmull64 (vector_extract (v2i64 V128:$Rn), (i64 1)),
3437 (vector_extract (v2i64 V128:$Rm), (i64 1))),
3438 (PMULLv2i64 V128:$Rn, V128:$Rm)>;
3440 // CodeGen patterns for addhn and subhn instructions, which can actually be
3441 // written in LLVM IR without too much difficulty.
3444 def : Pat<(v8i8 (trunc (v8i16 (AArch64vlshr (add V128:$Rn, V128:$Rm), (i32 8))))),
3445 (ADDHNv8i16_v8i8 V128:$Rn, V128:$Rm)>;
3446 def : Pat<(v4i16 (trunc (v4i32 (AArch64vlshr (add V128:$Rn, V128:$Rm),
3448 (ADDHNv4i32_v4i16 V128:$Rn, V128:$Rm)>;
3449 def : Pat<(v2i32 (trunc (v2i64 (AArch64vlshr (add V128:$Rn, V128:$Rm),
3451 (ADDHNv2i64_v2i32 V128:$Rn, V128:$Rm)>;
3452 def : Pat<(concat_vectors (v8i8 V64:$Rd),
3453 (trunc (v8i16 (AArch64vlshr (add V128:$Rn, V128:$Rm),
3455 (ADDHNv8i16_v16i8 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3456 V128:$Rn, V128:$Rm)>;
3457 def : Pat<(concat_vectors (v4i16 V64:$Rd),
3458 (trunc (v4i32 (AArch64vlshr (add V128:$Rn, V128:$Rm),
3460 (ADDHNv4i32_v8i16 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3461 V128:$Rn, V128:$Rm)>;
3462 def : Pat<(concat_vectors (v2i32 V64:$Rd),
3463 (trunc (v2i64 (AArch64vlshr (add V128:$Rn, V128:$Rm),
3465 (ADDHNv2i64_v4i32 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3466 V128:$Rn, V128:$Rm)>;
3469 def : Pat<(v8i8 (trunc (v8i16 (AArch64vlshr (sub V128:$Rn, V128:$Rm), (i32 8))))),
3470 (SUBHNv8i16_v8i8 V128:$Rn, V128:$Rm)>;
3471 def : Pat<(v4i16 (trunc (v4i32 (AArch64vlshr (sub V128:$Rn, V128:$Rm),
3473 (SUBHNv4i32_v4i16 V128:$Rn, V128:$Rm)>;
3474 def : Pat<(v2i32 (trunc (v2i64 (AArch64vlshr (sub V128:$Rn, V128:$Rm),
3476 (SUBHNv2i64_v2i32 V128:$Rn, V128:$Rm)>;
3477 def : Pat<(concat_vectors (v8i8 V64:$Rd),
3478 (trunc (v8i16 (AArch64vlshr (sub V128:$Rn, V128:$Rm),
3480 (SUBHNv8i16_v16i8 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3481 V128:$Rn, V128:$Rm)>;
3482 def : Pat<(concat_vectors (v4i16 V64:$Rd),
3483 (trunc (v4i32 (AArch64vlshr (sub V128:$Rn, V128:$Rm),
3485 (SUBHNv4i32_v8i16 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3486 V128:$Rn, V128:$Rm)>;
3487 def : Pat<(concat_vectors (v2i32 V64:$Rd),
3488 (trunc (v2i64 (AArch64vlshr (sub V128:$Rn, V128:$Rm),
3490 (SUBHNv2i64_v4i32 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3491 V128:$Rn, V128:$Rm)>;
3493 //----------------------------------------------------------------------------
3494 // AdvSIMD bitwise extract from vector instruction.
3495 //----------------------------------------------------------------------------
3497 defm EXT : SIMDBitwiseExtract<"ext">;
3499 def : Pat<(v4i16 (AArch64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
3500 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
3501 def : Pat<(v8i16 (AArch64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3502 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3503 def : Pat<(v2i32 (AArch64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
3504 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
3505 def : Pat<(v2f32 (AArch64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
3506 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
3507 def : Pat<(v4i32 (AArch64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3508 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3509 def : Pat<(v4f32 (AArch64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3510 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3511 def : Pat<(v2i64 (AArch64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3512 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3513 def : Pat<(v2f64 (AArch64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3514 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3515 def : Pat<(v4f16 (AArch64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
3516 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
3517 def : Pat<(v8f16 (AArch64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3518 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3520 // We use EXT to handle extract_subvector to copy the upper 64-bits of a
3522 def : Pat<(v8i8 (extract_subvector V128:$Rn, (i64 8))),
3523 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3524 def : Pat<(v4i16 (extract_subvector V128:$Rn, (i64 4))),
3525 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3526 def : Pat<(v2i32 (extract_subvector V128:$Rn, (i64 2))),
3527 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3528 def : Pat<(v1i64 (extract_subvector V128:$Rn, (i64 1))),
3529 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3530 def : Pat<(v4f16 (extract_subvector V128:$Rn, (i64 4))),
3531 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3532 def : Pat<(v2f32 (extract_subvector V128:$Rn, (i64 2))),
3533 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3534 def : Pat<(v1f64 (extract_subvector V128:$Rn, (i64 1))),
3535 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3538 //----------------------------------------------------------------------------
3539 // AdvSIMD zip vector
3540 //----------------------------------------------------------------------------
3542 defm TRN1 : SIMDZipVector<0b010, "trn1", AArch64trn1>;
3543 defm TRN2 : SIMDZipVector<0b110, "trn2", AArch64trn2>;
3544 defm UZP1 : SIMDZipVector<0b001, "uzp1", AArch64uzp1>;
3545 defm UZP2 : SIMDZipVector<0b101, "uzp2", AArch64uzp2>;
3546 defm ZIP1 : SIMDZipVector<0b011, "zip1", AArch64zip1>;
3547 defm ZIP2 : SIMDZipVector<0b111, "zip2", AArch64zip2>;
3549 //----------------------------------------------------------------------------
3550 // AdvSIMD TBL/TBX instructions
3551 //----------------------------------------------------------------------------
3553 defm TBL : SIMDTableLookup< 0, "tbl">;
3554 defm TBX : SIMDTableLookupTied<1, "tbx">;
3556 def : Pat<(v8i8 (int_aarch64_neon_tbl1 (v16i8 VecListOne128:$Rn), (v8i8 V64:$Ri))),
3557 (TBLv8i8One VecListOne128:$Rn, V64:$Ri)>;
3558 def : Pat<(v16i8 (int_aarch64_neon_tbl1 (v16i8 V128:$Ri), (v16i8 V128:$Rn))),
3559 (TBLv16i8One V128:$Ri, V128:$Rn)>;
3561 def : Pat<(v8i8 (int_aarch64_neon_tbx1 (v8i8 V64:$Rd),
3562 (v16i8 VecListOne128:$Rn), (v8i8 V64:$Ri))),
3563 (TBXv8i8One V64:$Rd, VecListOne128:$Rn, V64:$Ri)>;
3564 def : Pat<(v16i8 (int_aarch64_neon_tbx1 (v16i8 V128:$Rd),
3565 (v16i8 V128:$Ri), (v16i8 V128:$Rn))),
3566 (TBXv16i8One V128:$Rd, V128:$Ri, V128:$Rn)>;
3569 //----------------------------------------------------------------------------
3570 // AdvSIMD scalar CPY instruction
3571 //----------------------------------------------------------------------------
3573 defm CPY : SIMDScalarCPY<"cpy">;
3575 //----------------------------------------------------------------------------
3576 // AdvSIMD scalar pairwise instructions
3577 //----------------------------------------------------------------------------
3579 defm ADDP : SIMDPairwiseScalarD<0, 0b11011, "addp">;
3580 defm FADDP : SIMDFPPairwiseScalar<1, 0, 0b01101, "faddp">;
3581 defm FMAXNMP : SIMDFPPairwiseScalar<1, 0, 0b01100, "fmaxnmp">;
3582 defm FMAXP : SIMDFPPairwiseScalar<1, 0, 0b01111, "fmaxp">;
3583 defm FMINNMP : SIMDFPPairwiseScalar<1, 1, 0b01100, "fminnmp">;
3584 defm FMINP : SIMDFPPairwiseScalar<1, 1, 0b01111, "fminp">;
3585 def : Pat<(v2i64 (AArch64saddv V128:$Rn)),
3586 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), (ADDPv2i64p V128:$Rn), dsub)>;
3587 def : Pat<(v2i64 (AArch64uaddv V128:$Rn)),
3588 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), (ADDPv2i64p V128:$Rn), dsub)>;
3589 def : Pat<(f32 (int_aarch64_neon_faddv (v2f32 V64:$Rn))),
3590 (FADDPv2i32p V64:$Rn)>;
3591 def : Pat<(f32 (int_aarch64_neon_faddv (v4f32 V128:$Rn))),
3592 (FADDPv2i32p (EXTRACT_SUBREG (FADDPv4f32 V128:$Rn, V128:$Rn), dsub))>;
3593 def : Pat<(f64 (int_aarch64_neon_faddv (v2f64 V128:$Rn))),
3594 (FADDPv2i64p V128:$Rn)>;
3595 def : Pat<(f32 (int_aarch64_neon_fmaxnmv (v2f32 V64:$Rn))),
3596 (FMAXNMPv2i32p V64:$Rn)>;
3597 def : Pat<(f64 (int_aarch64_neon_fmaxnmv (v2f64 V128:$Rn))),
3598 (FMAXNMPv2i64p V128:$Rn)>;
3599 def : Pat<(f32 (int_aarch64_neon_fmaxv (v2f32 V64:$Rn))),
3600 (FMAXPv2i32p V64:$Rn)>;
3601 def : Pat<(f64 (int_aarch64_neon_fmaxv (v2f64 V128:$Rn))),
3602 (FMAXPv2i64p V128:$Rn)>;
3603 def : Pat<(f32 (int_aarch64_neon_fminnmv (v2f32 V64:$Rn))),
3604 (FMINNMPv2i32p V64:$Rn)>;
3605 def : Pat<(f64 (int_aarch64_neon_fminnmv (v2f64 V128:$Rn))),
3606 (FMINNMPv2i64p V128:$Rn)>;
3607 def : Pat<(f32 (int_aarch64_neon_fminv (v2f32 V64:$Rn))),
3608 (FMINPv2i32p V64:$Rn)>;
3609 def : Pat<(f64 (int_aarch64_neon_fminv (v2f64 V128:$Rn))),
3610 (FMINPv2i64p V128:$Rn)>;
3612 //----------------------------------------------------------------------------
3613 // AdvSIMD INS/DUP instructions
3614 //----------------------------------------------------------------------------
3616 def DUPv8i8gpr : SIMDDupFromMain<0, {?,?,?,?,1}, ".8b", v8i8, V64, GPR32>;
3617 def DUPv16i8gpr : SIMDDupFromMain<1, {?,?,?,?,1}, ".16b", v16i8, V128, GPR32>;
3618 def DUPv4i16gpr : SIMDDupFromMain<0, {?,?,?,1,0}, ".4h", v4i16, V64, GPR32>;
3619 def DUPv8i16gpr : SIMDDupFromMain<1, {?,?,?,1,0}, ".8h", v8i16, V128, GPR32>;
3620 def DUPv2i32gpr : SIMDDupFromMain<0, {?,?,1,0,0}, ".2s", v2i32, V64, GPR32>;
3621 def DUPv4i32gpr : SIMDDupFromMain<1, {?,?,1,0,0}, ".4s", v4i32, V128, GPR32>;
3622 def DUPv2i64gpr : SIMDDupFromMain<1, {?,1,0,0,0}, ".2d", v2i64, V128, GPR64>;
3624 def DUPv2i64lane : SIMDDup64FromElement;
3625 def DUPv2i32lane : SIMDDup32FromElement<0, ".2s", v2i32, V64>;
3626 def DUPv4i32lane : SIMDDup32FromElement<1, ".4s", v4i32, V128>;
3627 def DUPv4i16lane : SIMDDup16FromElement<0, ".4h", v4i16, V64>;
3628 def DUPv8i16lane : SIMDDup16FromElement<1, ".8h", v8i16, V128>;
3629 def DUPv8i8lane : SIMDDup8FromElement <0, ".8b", v8i8, V64>;
3630 def DUPv16i8lane : SIMDDup8FromElement <1, ".16b", v16i8, V128>;
3632 def : Pat<(v2f32 (AArch64dup (f32 FPR32:$Rn))),
3633 (v2f32 (DUPv2i32lane
3634 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rn, ssub),
3636 def : Pat<(v4f32 (AArch64dup (f32 FPR32:$Rn))),
3637 (v4f32 (DUPv4i32lane
3638 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rn, ssub),
3640 def : Pat<(v2f64 (AArch64dup (f64 FPR64:$Rn))),
3641 (v2f64 (DUPv2i64lane
3642 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$Rn, dsub),
3644 def : Pat<(v4f16 (AArch64dup (f16 FPR16:$Rn))),
3645 (v4f16 (DUPv4i16lane
3646 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), FPR16:$Rn, hsub),
3648 def : Pat<(v8f16 (AArch64dup (f16 FPR16:$Rn))),
3649 (v8f16 (DUPv8i16lane
3650 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), FPR16:$Rn, hsub),
3653 def : Pat<(v4f16 (AArch64duplane16 (v8f16 V128:$Rn), VectorIndexH:$imm)),
3654 (DUPv4i16lane V128:$Rn, VectorIndexH:$imm)>;
3655 def : Pat<(v8f16 (AArch64duplane16 (v8f16 V128:$Rn), VectorIndexH:$imm)),
3656 (DUPv8i16lane V128:$Rn, VectorIndexH:$imm)>;
3658 def : Pat<(v2f32 (AArch64duplane32 (v4f32 V128:$Rn), VectorIndexS:$imm)),
3659 (DUPv2i32lane V128:$Rn, VectorIndexS:$imm)>;
3660 def : Pat<(v4f32 (AArch64duplane32 (v4f32 V128:$Rn), VectorIndexS:$imm)),
3661 (DUPv4i32lane V128:$Rn, VectorIndexS:$imm)>;
3662 def : Pat<(v2f64 (AArch64duplane64 (v2f64 V128:$Rn), VectorIndexD:$imm)),
3663 (DUPv2i64lane V128:$Rn, VectorIndexD:$imm)>;
3665 // If there's an (AArch64dup (vector_extract ...) ...), we can use a duplane
3666 // instruction even if the types don't match: we just have to remap the lane
3667 // carefully. N.b. this trick only applies to truncations.
3668 def VecIndex_x2 : SDNodeXForm<imm, [{
3669 return CurDAG->getTargetConstant(2 * N->getZExtValue(), SDLoc(N), MVT::i64);
3671 def VecIndex_x4 : SDNodeXForm<imm, [{
3672 return CurDAG->getTargetConstant(4 * N->getZExtValue(), SDLoc(N), MVT::i64);
3674 def VecIndex_x8 : SDNodeXForm<imm, [{
3675 return CurDAG->getTargetConstant(8 * N->getZExtValue(), SDLoc(N), MVT::i64);
3678 multiclass DUPWithTruncPats<ValueType ResVT, ValueType Src64VT,
3679 ValueType Src128VT, ValueType ScalVT,
3680 Instruction DUP, SDNodeXForm IdxXFORM> {
3681 def : Pat<(ResVT (AArch64dup (ScalVT (vector_extract (Src128VT V128:$Rn),
3683 (DUP V128:$Rn, (IdxXFORM imm:$idx))>;
3685 def : Pat<(ResVT (AArch64dup (ScalVT (vector_extract (Src64VT V64:$Rn),
3687 (DUP (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), (IdxXFORM imm:$idx))>;
3690 defm : DUPWithTruncPats<v8i8, v4i16, v8i16, i32, DUPv8i8lane, VecIndex_x2>;
3691 defm : DUPWithTruncPats<v8i8, v2i32, v4i32, i32, DUPv8i8lane, VecIndex_x4>;
3692 defm : DUPWithTruncPats<v4i16, v2i32, v4i32, i32, DUPv4i16lane, VecIndex_x2>;
3694 defm : DUPWithTruncPats<v16i8, v4i16, v8i16, i32, DUPv16i8lane, VecIndex_x2>;
3695 defm : DUPWithTruncPats<v16i8, v2i32, v4i32, i32, DUPv16i8lane, VecIndex_x4>;
3696 defm : DUPWithTruncPats<v8i16, v2i32, v4i32, i32, DUPv8i16lane, VecIndex_x2>;
3698 multiclass DUPWithTrunci64Pats<ValueType ResVT, Instruction DUP,
3699 SDNodeXForm IdxXFORM> {
3700 def : Pat<(ResVT (AArch64dup (i32 (trunc (vector_extract (v2i64 V128:$Rn),
3702 (DUP V128:$Rn, (IdxXFORM imm:$idx))>;
3704 def : Pat<(ResVT (AArch64dup (i32 (trunc (vector_extract (v1i64 V64:$Rn),
3706 (DUP (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), (IdxXFORM imm:$idx))>;
3709 defm : DUPWithTrunci64Pats<v8i8, DUPv8i8lane, VecIndex_x8>;
3710 defm : DUPWithTrunci64Pats<v4i16, DUPv4i16lane, VecIndex_x4>;
3711 defm : DUPWithTrunci64Pats<v2i32, DUPv2i32lane, VecIndex_x2>;
3713 defm : DUPWithTrunci64Pats<v16i8, DUPv16i8lane, VecIndex_x8>;
3714 defm : DUPWithTrunci64Pats<v8i16, DUPv8i16lane, VecIndex_x4>;
3715 defm : DUPWithTrunci64Pats<v4i32, DUPv4i32lane, VecIndex_x2>;
3717 // SMOV and UMOV definitions, with some extra patterns for convenience
3721 def : Pat<(sext_inreg (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx), i8),
3722 (i32 (SMOVvi8to32 V128:$Rn, VectorIndexB:$idx))>;
3723 def : Pat<(sext_inreg (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx), i8),
3724 (i64 (SMOVvi8to64 V128:$Rn, VectorIndexB:$idx))>;
3725 def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
3726 (i32 (SMOVvi16to32 V128:$Rn, VectorIndexH:$idx))>;
3727 def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
3728 (i64 (SMOVvi16to64 V128:$Rn, VectorIndexH:$idx))>;
3729 def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
3730 (i32 (SMOVvi16to32 V128:$Rn, VectorIndexH:$idx))>;
3731 def : Pat<(sext (i32 (vector_extract (v4i32 V128:$Rn), VectorIndexS:$idx))),
3732 (i64 (SMOVvi32to64 V128:$Rn, VectorIndexS:$idx))>;
3734 // Extracting i8 or i16 elements will have the zero-extend transformed to
3735 // an 'and' mask by type legalization since neither i8 nor i16 are legal types
3736 // for AArch64. Match these patterns here since UMOV already zeroes out the high
3737 // bits of the destination register.
3738 def : Pat<(and (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx),
3740 (i32 (UMOVvi8 V128:$Rn, VectorIndexB:$idx))>;
3741 def : Pat<(and (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),
3743 (i32 (UMOVvi16 V128:$Rn, VectorIndexH:$idx))>;
3747 def : Pat<(v16i8 (scalar_to_vector GPR32:$Rn)),
3748 (SUBREG_TO_REG (i32 0),
3749 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3750 def : Pat<(v8i8 (scalar_to_vector GPR32:$Rn)),
3751 (SUBREG_TO_REG (i32 0),
3752 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3754 def : Pat<(v8i16 (scalar_to_vector GPR32:$Rn)),
3755 (SUBREG_TO_REG (i32 0),
3756 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3757 def : Pat<(v4i16 (scalar_to_vector GPR32:$Rn)),
3758 (SUBREG_TO_REG (i32 0),
3759 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3761 def : Pat<(v2i32 (scalar_to_vector (i32 FPR32:$Rn))),
3762 (v2i32 (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)),
3763 (i32 FPR32:$Rn), ssub))>;
3764 def : Pat<(v4i32 (scalar_to_vector (i32 FPR32:$Rn))),
3765 (v4i32 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
3766 (i32 FPR32:$Rn), ssub))>;
3767 def : Pat<(v2i64 (scalar_to_vector (i64 FPR64:$Rn))),
3768 (v2i64 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
3769 (i64 FPR64:$Rn), dsub))>;
3771 def : Pat<(v4f32 (scalar_to_vector (f32 FPR32:$Rn))),
3772 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rn, ssub)>;
3773 def : Pat<(v2f32 (scalar_to_vector (f32 FPR32:$Rn))),
3774 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), FPR32:$Rn, ssub)>;
3775 def : Pat<(v2f64 (scalar_to_vector (f64 FPR64:$Rn))),
3776 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$Rn, dsub)>;
3778 def : Pat<(v4f16 (vector_insert (v4f16 V64:$Rn),
3779 (f16 FPR16:$Rm), (i64 VectorIndexS:$imm))),
3782 (v8f16 (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), V64:$Rn, dsub)),
3784 (v8f16 (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), FPR16:$Rm, hsub)),
3788 def : Pat<(v8f16 (vector_insert (v8f16 V128:$Rn),
3789 (f16 FPR16:$Rm), (i64 VectorIndexH:$imm))),
3791 V128:$Rn, VectorIndexH:$imm,
3792 (v8f16 (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), FPR16:$Rm, hsub)),
3795 def : Pat<(v2f32 (vector_insert (v2f32 V64:$Rn),
3796 (f32 FPR32:$Rm), (i64 VectorIndexS:$imm))),
3799 (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), V64:$Rn, dsub)),
3801 (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rm, ssub)),
3804 def : Pat<(v4f32 (vector_insert (v4f32 V128:$Rn),
3805 (f32 FPR32:$Rm), (i64 VectorIndexS:$imm))),
3807 V128:$Rn, VectorIndexS:$imm,
3808 (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rm, ssub)),
3810 def : Pat<(v2f64 (vector_insert (v2f64 V128:$Rn),
3811 (f64 FPR64:$Rm), (i64 VectorIndexD:$imm))),
3813 V128:$Rn, VectorIndexD:$imm,
3814 (v2f64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$Rm, dsub)),
3817 // Copy an element at a constant index in one vector into a constant indexed
3818 // element of another.
3819 // FIXME refactor to a shared class/dev parameterized on vector type, vector
3820 // index type and INS extension
3821 def : Pat<(v16i8 (int_aarch64_neon_vcopy_lane
3822 (v16i8 V128:$Vd), VectorIndexB:$idx, (v16i8 V128:$Vs),
3823 VectorIndexB:$idx2)),
3825 V128:$Vd, VectorIndexB:$idx, V128:$Vs, VectorIndexB:$idx2)
3827 def : Pat<(v8i16 (int_aarch64_neon_vcopy_lane
3828 (v8i16 V128:$Vd), VectorIndexH:$idx, (v8i16 V128:$Vs),
3829 VectorIndexH:$idx2)),
3831 V128:$Vd, VectorIndexH:$idx, V128:$Vs, VectorIndexH:$idx2)
3833 def : Pat<(v4i32 (int_aarch64_neon_vcopy_lane
3834 (v4i32 V128:$Vd), VectorIndexS:$idx, (v4i32 V128:$Vs),
3835 VectorIndexS:$idx2)),
3837 V128:$Vd, VectorIndexS:$idx, V128:$Vs, VectorIndexS:$idx2)
3839 def : Pat<(v2i64 (int_aarch64_neon_vcopy_lane
3840 (v2i64 V128:$Vd), VectorIndexD:$idx, (v2i64 V128:$Vs),
3841 VectorIndexD:$idx2)),
3843 V128:$Vd, VectorIndexD:$idx, V128:$Vs, VectorIndexD:$idx2)
3846 multiclass Neon_INS_elt_pattern<ValueType VT128, ValueType VT64,
3847 ValueType VTScal, Instruction INS> {
3848 def : Pat<(VT128 (vector_insert V128:$src,
3849 (VTScal (vector_extract (VT128 V128:$Rn), imm:$Immn)),
3851 (INS V128:$src, imm:$Immd, V128:$Rn, imm:$Immn)>;
3853 def : Pat<(VT128 (vector_insert V128:$src,
3854 (VTScal (vector_extract (VT64 V64:$Rn), imm:$Immn)),
3856 (INS V128:$src, imm:$Immd,
3857 (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), imm:$Immn)>;
3859 def : Pat<(VT64 (vector_insert V64:$src,
3860 (VTScal (vector_extract (VT128 V128:$Rn), imm:$Immn)),
3862 (EXTRACT_SUBREG (INS (SUBREG_TO_REG (i64 0), V64:$src, dsub),
3863 imm:$Immd, V128:$Rn, imm:$Immn),
3866 def : Pat<(VT64 (vector_insert V64:$src,
3867 (VTScal (vector_extract (VT64 V64:$Rn), imm:$Immn)),
3870 (INS (SUBREG_TO_REG (i64 0), V64:$src, dsub), imm:$Immd,
3871 (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), imm:$Immn),
3875 defm : Neon_INS_elt_pattern<v8f16, v4f16, f16, INSvi16lane>;
3876 defm : Neon_INS_elt_pattern<v4f32, v2f32, f32, INSvi32lane>;
3877 defm : Neon_INS_elt_pattern<v2f64, v1f64, f64, INSvi64lane>;
3880 // Floating point vector extractions are codegen'd as either a sequence of
3881 // subregister extractions, or a MOV (aka CPY here, alias for DUP) if
3882 // the lane number is anything other than zero.
3883 def : Pat<(vector_extract (v2f64 V128:$Rn), 0),
3884 (f64 (EXTRACT_SUBREG V128:$Rn, dsub))>;
3885 def : Pat<(vector_extract (v4f32 V128:$Rn), 0),
3886 (f32 (EXTRACT_SUBREG V128:$Rn, ssub))>;
3887 def : Pat<(vector_extract (v8f16 V128:$Rn), 0),
3888 (f16 (EXTRACT_SUBREG V128:$Rn, hsub))>;
3890 def : Pat<(vector_extract (v2f64 V128:$Rn), VectorIndexD:$idx),
3891 (f64 (CPYi64 V128:$Rn, VectorIndexD:$idx))>;
3892 def : Pat<(vector_extract (v4f32 V128:$Rn), VectorIndexS:$idx),
3893 (f32 (CPYi32 V128:$Rn, VectorIndexS:$idx))>;
3894 def : Pat<(vector_extract (v8f16 V128:$Rn), VectorIndexH:$idx),
3895 (f16 (CPYi16 V128:$Rn, VectorIndexH:$idx))>;
3897 // All concat_vectors operations are canonicalised to act on i64 vectors for
3898 // AArch64. In the general case we need an instruction, which had just as well be
3900 class ConcatPat<ValueType DstTy, ValueType SrcTy>
3901 : Pat<(DstTy (concat_vectors (SrcTy V64:$Rd), V64:$Rn)),
3902 (INSvi64lane (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), 1,
3903 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rn, dsub), 0)>;
3905 def : ConcatPat<v2i64, v1i64>;
3906 def : ConcatPat<v2f64, v1f64>;
3907 def : ConcatPat<v4i32, v2i32>;
3908 def : ConcatPat<v4f32, v2f32>;
3909 def : ConcatPat<v8i16, v4i16>;
3910 def : ConcatPat<v8f16, v4f16>;
3911 def : ConcatPat<v16i8, v8i8>;
3913 // If the high lanes are undef, though, we can just ignore them:
3914 class ConcatUndefPat<ValueType DstTy, ValueType SrcTy>
3915 : Pat<(DstTy (concat_vectors (SrcTy V64:$Rn), undef)),
3916 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rn, dsub)>;
3918 def : ConcatUndefPat<v2i64, v1i64>;
3919 def : ConcatUndefPat<v2f64, v1f64>;
3920 def : ConcatUndefPat<v4i32, v2i32>;
3921 def : ConcatUndefPat<v4f32, v2f32>;
3922 def : ConcatUndefPat<v8i16, v4i16>;
3923 def : ConcatUndefPat<v16i8, v8i8>;
3925 //----------------------------------------------------------------------------
3926 // AdvSIMD across lanes instructions
3927 //----------------------------------------------------------------------------
3929 defm ADDV : SIMDAcrossLanesBHS<0, 0b11011, "addv">;
3930 defm SMAXV : SIMDAcrossLanesBHS<0, 0b01010, "smaxv">;
3931 defm SMINV : SIMDAcrossLanesBHS<0, 0b11010, "sminv">;
3932 defm UMAXV : SIMDAcrossLanesBHS<1, 0b01010, "umaxv">;
3933 defm UMINV : SIMDAcrossLanesBHS<1, 0b11010, "uminv">;
3934 defm SADDLV : SIMDAcrossLanesHSD<0, 0b00011, "saddlv">;
3935 defm UADDLV : SIMDAcrossLanesHSD<1, 0b00011, "uaddlv">;
3936 defm FMAXNMV : SIMDAcrossLanesS<0b01100, 0, "fmaxnmv", int_aarch64_neon_fmaxnmv>;
3937 defm FMAXV : SIMDAcrossLanesS<0b01111, 0, "fmaxv", int_aarch64_neon_fmaxv>;
3938 defm FMINNMV : SIMDAcrossLanesS<0b01100, 1, "fminnmv", int_aarch64_neon_fminnmv>;
3939 defm FMINV : SIMDAcrossLanesS<0b01111, 1, "fminv", int_aarch64_neon_fminv>;
3941 // Patterns for across-vector intrinsics, that have a node equivalent, that
3942 // returns a vector (with only the low lane defined) instead of a scalar.
3943 // In effect, opNode is the same as (scalar_to_vector (IntNode)).
3944 multiclass SIMDAcrossLanesIntrinsic<string baseOpc,
3945 SDPatternOperator opNode> {
3946 // If a lane instruction caught the vector_extract around opNode, we can
3947 // directly match the latter to the instruction.
3948 def : Pat<(v8i8 (opNode V64:$Rn)),
3949 (INSERT_SUBREG (v8i8 (IMPLICIT_DEF)),
3950 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub)>;
3951 def : Pat<(v16i8 (opNode V128:$Rn)),
3952 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3953 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub)>;
3954 def : Pat<(v4i16 (opNode V64:$Rn)),
3955 (INSERT_SUBREG (v4i16 (IMPLICIT_DEF)),
3956 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub)>;
3957 def : Pat<(v8i16 (opNode V128:$Rn)),
3958 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
3959 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub)>;
3960 def : Pat<(v4i32 (opNode V128:$Rn)),
3961 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
3962 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), ssub)>;
3965 // If none did, fallback to the explicit patterns, consuming the vector_extract.
3966 def : Pat<(i32 (vector_extract (insert_subvector undef, (v8i8 (opNode V64:$Rn)),
3967 (i32 0)), (i64 0))),
3968 (EXTRACT_SUBREG (INSERT_SUBREG (v8i8 (IMPLICIT_DEF)),
3969 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn),
3971 def : Pat<(i32 (vector_extract (v16i8 (opNode V128:$Rn)), (i64 0))),
3972 (EXTRACT_SUBREG (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3973 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn),
3975 def : Pat<(i32 (vector_extract (insert_subvector undef,
3976 (v4i16 (opNode V64:$Rn)), (i32 0)), (i64 0))),
3977 (EXTRACT_SUBREG (INSERT_SUBREG (v4i16 (IMPLICIT_DEF)),
3978 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn),
3980 def : Pat<(i32 (vector_extract (v8i16 (opNode V128:$Rn)), (i64 0))),
3981 (EXTRACT_SUBREG (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
3982 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn),
3984 def : Pat<(i32 (vector_extract (v4i32 (opNode V128:$Rn)), (i64 0))),
3985 (EXTRACT_SUBREG (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
3986 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn),
3991 multiclass SIMDAcrossLanesSignedIntrinsic<string baseOpc,
3992 SDPatternOperator opNode>
3993 : SIMDAcrossLanesIntrinsic<baseOpc, opNode> {
3994 // If there is a sign extension after this intrinsic, consume it as smov already
3996 def : Pat<(i32 (sext_inreg (i32 (vector_extract (insert_subvector undef,
3997 (opNode (v8i8 V64:$Rn)), (i32 0)), (i64 0))), i8)),
3999 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4000 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
4002 def : Pat<(i32 (sext_inreg (i32 (vector_extract
4003 (opNode (v16i8 V128:$Rn)), (i64 0))), i8)),
4005 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4006 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
4008 def : Pat<(i32 (sext_inreg (i32 (vector_extract (insert_subvector undef,
4009 (opNode (v4i16 V64:$Rn)), (i32 0)), (i64 0))), i16)),
4011 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4012 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
4014 def : Pat<(i32 (sext_inreg (i32 (vector_extract
4015 (opNode (v8i16 V128:$Rn)), (i64 0))), i16)),
4017 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4018 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
4022 multiclass SIMDAcrossLanesUnsignedIntrinsic<string baseOpc,
4023 SDPatternOperator opNode>
4024 : SIMDAcrossLanesIntrinsic<baseOpc, opNode> {
4025 // If there is a masking operation keeping only what has been actually
4026 // generated, consume it.
4027 def : Pat<(i32 (and (i32 (vector_extract (insert_subvector undef,
4028 (opNode (v8i8 V64:$Rn)), (i32 0)), (i64 0))), maski8_or_more)),
4029 (i32 (EXTRACT_SUBREG
4030 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4031 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
4033 def : Pat<(i32 (and (i32 (vector_extract (opNode (v16i8 V128:$Rn)), (i64 0))),
4035 (i32 (EXTRACT_SUBREG
4036 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4037 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
4039 def : Pat<(i32 (and (i32 (vector_extract (insert_subvector undef,
4040 (opNode (v4i16 V64:$Rn)), (i32 0)), (i64 0))), maski16_or_more)),
4041 (i32 (EXTRACT_SUBREG
4042 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4043 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
4045 def : Pat<(i32 (and (i32 (vector_extract (opNode (v8i16 V128:$Rn)), (i64 0))),
4047 (i32 (EXTRACT_SUBREG
4048 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4049 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
4053 defm : SIMDAcrossLanesSignedIntrinsic<"ADDV", AArch64saddv>;
4054 // vaddv_[su]32 is special; -> ADDP Vd.2S,Vn.2S,Vm.2S; return Vd.s[0];Vn==Vm
4055 def : Pat<(v2i32 (AArch64saddv (v2i32 V64:$Rn))),
4056 (ADDPv2i32 V64:$Rn, V64:$Rn)>;
4058 defm : SIMDAcrossLanesUnsignedIntrinsic<"ADDV", AArch64uaddv>;
4059 // vaddv_[su]32 is special; -> ADDP Vd.2S,Vn.2S,Vm.2S; return Vd.s[0];Vn==Vm
4060 def : Pat<(v2i32 (AArch64uaddv (v2i32 V64:$Rn))),
4061 (ADDPv2i32 V64:$Rn, V64:$Rn)>;
4063 defm : SIMDAcrossLanesSignedIntrinsic<"SMAXV", AArch64smaxv>;
4064 def : Pat<(v2i32 (AArch64smaxv (v2i32 V64:$Rn))),
4065 (SMAXPv2i32 V64:$Rn, V64:$Rn)>;
4067 defm : SIMDAcrossLanesSignedIntrinsic<"SMINV", AArch64sminv>;
4068 def : Pat<(v2i32 (AArch64sminv (v2i32 V64:$Rn))),
4069 (SMINPv2i32 V64:$Rn, V64:$Rn)>;
4071 defm : SIMDAcrossLanesUnsignedIntrinsic<"UMAXV", AArch64umaxv>;
4072 def : Pat<(v2i32 (AArch64umaxv (v2i32 V64:$Rn))),
4073 (UMAXPv2i32 V64:$Rn, V64:$Rn)>;
4075 defm : SIMDAcrossLanesUnsignedIntrinsic<"UMINV", AArch64uminv>;
4076 def : Pat<(v2i32 (AArch64uminv (v2i32 V64:$Rn))),
4077 (UMINPv2i32 V64:$Rn, V64:$Rn)>;
4079 multiclass SIMDAcrossLanesSignedLongIntrinsic<string baseOpc, Intrinsic intOp> {
4080 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
4082 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4083 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), hsub),
4085 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
4087 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4088 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), hsub),
4091 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
4092 (i32 (EXTRACT_SUBREG
4093 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4094 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), ssub),
4096 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
4097 (i32 (EXTRACT_SUBREG
4098 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4099 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), ssub),
4102 def : Pat<(i64 (intOp (v4i32 V128:$Rn))),
4103 (i64 (EXTRACT_SUBREG
4104 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4105 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), dsub),
4109 multiclass SIMDAcrossLanesUnsignedLongIntrinsic<string baseOpc,
4111 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
4112 (i32 (EXTRACT_SUBREG
4113 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4114 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), hsub),
4116 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
4117 (i32 (EXTRACT_SUBREG
4118 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4119 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), hsub),
4122 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
4123 (i32 (EXTRACT_SUBREG
4124 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4125 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), ssub),
4127 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
4128 (i32 (EXTRACT_SUBREG
4129 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4130 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), ssub),
4133 def : Pat<(i64 (intOp (v4i32 V128:$Rn))),
4134 (i64 (EXTRACT_SUBREG
4135 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4136 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), dsub),
4140 defm : SIMDAcrossLanesSignedLongIntrinsic<"SADDLV", int_aarch64_neon_saddlv>;
4141 defm : SIMDAcrossLanesUnsignedLongIntrinsic<"UADDLV", int_aarch64_neon_uaddlv>;
4143 // The vaddlv_s32 intrinsic gets mapped to SADDLP.
4144 def : Pat<(i64 (int_aarch64_neon_saddlv (v2i32 V64:$Rn))),
4145 (i64 (EXTRACT_SUBREG
4146 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4147 (SADDLPv2i32_v1i64 V64:$Rn), dsub),
4149 // The vaddlv_u32 intrinsic gets mapped to UADDLP.
4150 def : Pat<(i64 (int_aarch64_neon_uaddlv (v2i32 V64:$Rn))),
4151 (i64 (EXTRACT_SUBREG
4152 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4153 (UADDLPv2i32_v1i64 V64:$Rn), dsub),
4156 //------------------------------------------------------------------------------
4157 // AdvSIMD modified immediate instructions
4158 //------------------------------------------------------------------------------
4161 defm BIC : SIMDModifiedImmVectorShiftTied<1, 0b11, 0b01, "bic", AArch64bici>;
4163 defm ORR : SIMDModifiedImmVectorShiftTied<0, 0b11, 0b01, "orr", AArch64orri>;
4165 def : InstAlias<"bic $Vd.4h, $imm", (BICv4i16 V64:$Vd, imm0_255:$imm, 0)>;
4166 def : InstAlias<"bic $Vd.8h, $imm", (BICv8i16 V128:$Vd, imm0_255:$imm, 0)>;
4167 def : InstAlias<"bic $Vd.2s, $imm", (BICv2i32 V64:$Vd, imm0_255:$imm, 0)>;
4168 def : InstAlias<"bic $Vd.4s, $imm", (BICv4i32 V128:$Vd, imm0_255:$imm, 0)>;
4170 def : InstAlias<"bic.4h $Vd, $imm", (BICv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
4171 def : InstAlias<"bic.8h $Vd, $imm", (BICv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
4172 def : InstAlias<"bic.2s $Vd, $imm", (BICv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
4173 def : InstAlias<"bic.4s $Vd, $imm", (BICv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
4175 def : InstAlias<"orr $Vd.4h, $imm", (ORRv4i16 V64:$Vd, imm0_255:$imm, 0)>;
4176 def : InstAlias<"orr $Vd.8h, $imm", (ORRv8i16 V128:$Vd, imm0_255:$imm, 0)>;
4177 def : InstAlias<"orr $Vd.2s, $imm", (ORRv2i32 V64:$Vd, imm0_255:$imm, 0)>;
4178 def : InstAlias<"orr $Vd.4s, $imm", (ORRv4i32 V128:$Vd, imm0_255:$imm, 0)>;
4180 def : InstAlias<"orr.4h $Vd, $imm", (ORRv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
4181 def : InstAlias<"orr.8h $Vd, $imm", (ORRv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
4182 def : InstAlias<"orr.2s $Vd, $imm", (ORRv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
4183 def : InstAlias<"orr.4s $Vd, $imm", (ORRv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
4186 def FMOVv2f64_ns : SIMDModifiedImmVectorNoShift<1, 1, 0b1111, V128, fpimm8,
4188 [(set (v2f64 V128:$Rd), (AArch64fmov imm0_255:$imm8))]>;
4189 def FMOVv2f32_ns : SIMDModifiedImmVectorNoShift<0, 0, 0b1111, V64, fpimm8,
4191 [(set (v2f32 V64:$Rd), (AArch64fmov imm0_255:$imm8))]>;
4192 def FMOVv4f32_ns : SIMDModifiedImmVectorNoShift<1, 0, 0b1111, V128, fpimm8,
4194 [(set (v4f32 V128:$Rd), (AArch64fmov imm0_255:$imm8))]>;
4198 // EDIT byte mask: scalar
4199 let isReMaterializable = 1, isAsCheapAsAMove = 1 in
4200 def MOVID : SIMDModifiedImmScalarNoShift<0, 1, 0b1110, "movi",
4201 [(set FPR64:$Rd, simdimmtype10:$imm8)]>;
4202 // The movi_edit node has the immediate value already encoded, so we use
4203 // a plain imm0_255 here.
4204 def : Pat<(f64 (AArch64movi_edit imm0_255:$shift)),
4205 (MOVID imm0_255:$shift)>;
4207 def : Pat<(v1i64 immAllZerosV), (MOVID (i32 0))>;
4208 def : Pat<(v2i32 immAllZerosV), (MOVID (i32 0))>;
4209 def : Pat<(v4i16 immAllZerosV), (MOVID (i32 0))>;
4210 def : Pat<(v8i8 immAllZerosV), (MOVID (i32 0))>;
4212 def : Pat<(v1i64 immAllOnesV), (MOVID (i32 255))>;
4213 def : Pat<(v2i32 immAllOnesV), (MOVID (i32 255))>;
4214 def : Pat<(v4i16 immAllOnesV), (MOVID (i32 255))>;
4215 def : Pat<(v8i8 immAllOnesV), (MOVID (i32 255))>;
4217 // EDIT byte mask: 2d
4219 // The movi_edit node has the immediate value already encoded, so we use
4220 // a plain imm0_255 in the pattern
4221 let isReMaterializable = 1, isAsCheapAsAMove = 1 in
4222 def MOVIv2d_ns : SIMDModifiedImmVectorNoShift<1, 1, 0b1110, V128,
4225 [(set (v2i64 V128:$Rd), (AArch64movi_edit imm0_255:$imm8))]>;
4228 // Use movi.2d to materialize 0.0 if the HW does zero-cycle zeroing.
4229 // Complexity is added to break a tie with a plain MOVI.
4230 let AddedComplexity = 1 in {
4231 def : Pat<(f32 fpimm0),
4232 (f32 (EXTRACT_SUBREG (v2i64 (MOVIv2d_ns (i32 0))), ssub))>,
4234 def : Pat<(f64 fpimm0),
4235 (f64 (EXTRACT_SUBREG (v2i64 (MOVIv2d_ns (i32 0))), dsub))>,
4239 def : Pat<(v2i64 immAllZerosV), (MOVIv2d_ns (i32 0))>;
4240 def : Pat<(v4i32 immAllZerosV), (MOVIv2d_ns (i32 0))>;
4241 def : Pat<(v8i16 immAllZerosV), (MOVIv2d_ns (i32 0))>;
4242 def : Pat<(v16i8 immAllZerosV), (MOVIv2d_ns (i32 0))>;
4244 def : Pat<(v2i64 immAllOnesV), (MOVIv2d_ns (i32 255))>;
4245 def : Pat<(v4i32 immAllOnesV), (MOVIv2d_ns (i32 255))>;
4246 def : Pat<(v8i16 immAllOnesV), (MOVIv2d_ns (i32 255))>;
4247 def : Pat<(v16i8 immAllOnesV), (MOVIv2d_ns (i32 255))>;
4249 def : Pat<(v2f64 (AArch64dup (f64 fpimm0))), (MOVIv2d_ns (i32 0))>;
4250 def : Pat<(v4f32 (AArch64dup (f32 fpimm0))), (MOVIv2d_ns (i32 0))>;
4252 // EDIT per word & halfword: 2s, 4h, 4s, & 8h
4253 defm MOVI : SIMDModifiedImmVectorShift<0, 0b10, 0b00, "movi">;
4255 def : InstAlias<"movi $Vd.4h, $imm", (MOVIv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
4256 def : InstAlias<"movi $Vd.8h, $imm", (MOVIv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
4257 def : InstAlias<"movi $Vd.2s, $imm", (MOVIv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
4258 def : InstAlias<"movi $Vd.4s, $imm", (MOVIv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
4260 def : InstAlias<"movi.4h $Vd, $imm", (MOVIv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
4261 def : InstAlias<"movi.8h $Vd, $imm", (MOVIv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
4262 def : InstAlias<"movi.2s $Vd, $imm", (MOVIv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
4263 def : InstAlias<"movi.4s $Vd, $imm", (MOVIv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
4265 def : Pat<(v2i32 (AArch64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
4266 (MOVIv2i32 imm0_255:$imm8, imm:$shift)>;
4267 def : Pat<(v4i32 (AArch64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
4268 (MOVIv4i32 imm0_255:$imm8, imm:$shift)>;
4269 def : Pat<(v4i16 (AArch64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
4270 (MOVIv4i16 imm0_255:$imm8, imm:$shift)>;
4271 def : Pat<(v8i16 (AArch64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
4272 (MOVIv8i16 imm0_255:$imm8, imm:$shift)>;
4274 // EDIT per word: 2s & 4s with MSL shifter
4275 def MOVIv2s_msl : SIMDModifiedImmMoveMSL<0, 0, {1,1,0,?}, V64, "movi", ".2s",
4276 [(set (v2i32 V64:$Rd),
4277 (AArch64movi_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
4278 def MOVIv4s_msl : SIMDModifiedImmMoveMSL<1, 0, {1,1,0,?}, V128, "movi", ".4s",
4279 [(set (v4i32 V128:$Rd),
4280 (AArch64movi_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
4282 // Per byte: 8b & 16b
4283 def MOVIv8b_ns : SIMDModifiedImmVectorNoShift<0, 0, 0b1110, V64, imm0_255,
4285 [(set (v8i8 V64:$Rd), (AArch64movi imm0_255:$imm8))]>;
4286 def MOVIv16b_ns : SIMDModifiedImmVectorNoShift<1, 0, 0b1110, V128, imm0_255,
4288 [(set (v16i8 V128:$Rd), (AArch64movi imm0_255:$imm8))]>;
4292 // EDIT per word & halfword: 2s, 4h, 4s, & 8h
4293 defm MVNI : SIMDModifiedImmVectorShift<1, 0b10, 0b00, "mvni">;
4295 def : InstAlias<"mvni $Vd.4h, $imm", (MVNIv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
4296 def : InstAlias<"mvni $Vd.8h, $imm", (MVNIv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
4297 def : InstAlias<"mvni $Vd.2s, $imm", (MVNIv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
4298 def : InstAlias<"mvni $Vd.4s, $imm", (MVNIv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
4300 def : InstAlias<"mvni.4h $Vd, $imm", (MVNIv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
4301 def : InstAlias<"mvni.8h $Vd, $imm", (MVNIv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
4302 def : InstAlias<"mvni.2s $Vd, $imm", (MVNIv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
4303 def : InstAlias<"mvni.4s $Vd, $imm", (MVNIv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
4305 def : Pat<(v2i32 (AArch64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
4306 (MVNIv2i32 imm0_255:$imm8, imm:$shift)>;
4307 def : Pat<(v4i32 (AArch64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
4308 (MVNIv4i32 imm0_255:$imm8, imm:$shift)>;
4309 def : Pat<(v4i16 (AArch64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
4310 (MVNIv4i16 imm0_255:$imm8, imm:$shift)>;
4311 def : Pat<(v8i16 (AArch64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
4312 (MVNIv8i16 imm0_255:$imm8, imm:$shift)>;
4314 // EDIT per word: 2s & 4s with MSL shifter
4315 def MVNIv2s_msl : SIMDModifiedImmMoveMSL<0, 1, {1,1,0,?}, V64, "mvni", ".2s",
4316 [(set (v2i32 V64:$Rd),
4317 (AArch64mvni_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
4318 def MVNIv4s_msl : SIMDModifiedImmMoveMSL<1, 1, {1,1,0,?}, V128, "mvni", ".4s",
4319 [(set (v4i32 V128:$Rd),
4320 (AArch64mvni_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
4322 //----------------------------------------------------------------------------
4323 // AdvSIMD indexed element
4324 //----------------------------------------------------------------------------
4326 let hasSideEffects = 0 in {
4327 defm FMLA : SIMDFPIndexedTied<0, 0b0001, "fmla">;
4328 defm FMLS : SIMDFPIndexedTied<0, 0b0101, "fmls">;
4331 // NOTE: Operands are reordered in the FMLA/FMLS PatFrags because the
4332 // instruction expects the addend first, while the intrinsic expects it last.
4334 // On the other hand, there are quite a few valid combinatorial options due to
4335 // the commutativity of multiplication and the fact that (-x) * y = x * (-y).
4336 defm : SIMDFPIndexedTiedPatterns<"FMLA",
4337 TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)>>;
4338 defm : SIMDFPIndexedTiedPatterns<"FMLA",
4339 TriOpFrag<(fma node:$MHS, node:$RHS, node:$LHS)>>;
4341 defm : SIMDFPIndexedTiedPatterns<"FMLS",
4342 TriOpFrag<(fma node:$MHS, (fneg node:$RHS), node:$LHS)> >;
4343 defm : SIMDFPIndexedTiedPatterns<"FMLS",
4344 TriOpFrag<(fma node:$RHS, (fneg node:$MHS), node:$LHS)> >;
4345 defm : SIMDFPIndexedTiedPatterns<"FMLS",
4346 TriOpFrag<(fma (fneg node:$RHS), node:$MHS, node:$LHS)> >;
4347 defm : SIMDFPIndexedTiedPatterns<"FMLS",
4348 TriOpFrag<(fma (fneg node:$MHS), node:$RHS, node:$LHS)> >;
4350 multiclass FMLSIndexedAfterNegPatterns<SDPatternOperator OpNode> {
4351 // 3 variants for the .2s version: DUPLANE from 128-bit, DUPLANE from 64-bit
4353 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
4354 (AArch64duplane32 (v4f32 (fneg V128:$Rm)),
4355 VectorIndexS:$idx))),
4356 (FMLSv2i32_indexed V64:$Rd, V64:$Rn, V128:$Rm, VectorIndexS:$idx)>;
4357 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
4358 (v2f32 (AArch64duplane32
4359 (v4f32 (insert_subvector undef,
4360 (v2f32 (fneg V64:$Rm)),
4362 VectorIndexS:$idx)))),
4363 (FMLSv2i32_indexed V64:$Rd, V64:$Rn,
4364 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub),
4365 VectorIndexS:$idx)>;
4366 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
4367 (AArch64dup (f32 (fneg FPR32Op:$Rm))))),
4368 (FMLSv2i32_indexed V64:$Rd, V64:$Rn,
4369 (SUBREG_TO_REG (i32 0), FPR32Op:$Rm, ssub), (i64 0))>;
4371 // 3 variants for the .4s version: DUPLANE from 128-bit, DUPLANE from 64-bit
4373 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
4374 (AArch64duplane32 (v4f32 (fneg V128:$Rm)),
4375 VectorIndexS:$idx))),
4376 (FMLSv4i32_indexed V128:$Rd, V128:$Rn, V128:$Rm,
4377 VectorIndexS:$idx)>;
4378 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
4379 (v4f32 (AArch64duplane32
4380 (v4f32 (insert_subvector undef,
4381 (v2f32 (fneg V64:$Rm)),
4383 VectorIndexS:$idx)))),
4384 (FMLSv4i32_indexed V128:$Rd, V128:$Rn,
4385 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub),
4386 VectorIndexS:$idx)>;
4387 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
4388 (AArch64dup (f32 (fneg FPR32Op:$Rm))))),
4389 (FMLSv4i32_indexed V128:$Rd, V128:$Rn,
4390 (SUBREG_TO_REG (i32 0), FPR32Op:$Rm, ssub), (i64 0))>;
4392 // 2 variants for the .2d version: DUPLANE from 128-bit, and DUP scalar
4393 // (DUPLANE from 64-bit would be trivial).
4394 def : Pat<(v2f64 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn),
4395 (AArch64duplane64 (v2f64 (fneg V128:$Rm)),
4396 VectorIndexD:$idx))),
4398 V128:$Rd, V128:$Rn, V128:$Rm, VectorIndexS:$idx)>;
4399 def : Pat<(v2f64 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn),
4400 (AArch64dup (f64 (fneg FPR64Op:$Rm))))),
4401 (FMLSv2i64_indexed V128:$Rd, V128:$Rn,
4402 (SUBREG_TO_REG (i32 0), FPR64Op:$Rm, dsub), (i64 0))>;
4404 // 2 variants for 32-bit scalar version: extract from .2s or from .4s
4405 def : Pat<(f32 (OpNode (f32 FPR32:$Rd), (f32 FPR32:$Rn),
4406 (vector_extract (v4f32 (fneg V128:$Rm)),
4407 VectorIndexS:$idx))),
4408 (FMLSv1i32_indexed FPR32:$Rd, FPR32:$Rn,
4409 V128:$Rm, VectorIndexS:$idx)>;
4410 def : Pat<(f32 (OpNode (f32 FPR32:$Rd), (f32 FPR32:$Rn),
4411 (vector_extract (v4f32 (insert_subvector undef,
4412 (v2f32 (fneg V64:$Rm)),
4414 VectorIndexS:$idx))),
4415 (FMLSv1i32_indexed FPR32:$Rd, FPR32:$Rn,
4416 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub), VectorIndexS:$idx)>;
4418 // 1 variant for 64-bit scalar version: extract from .1d or from .2d
4419 def : Pat<(f64 (OpNode (f64 FPR64:$Rd), (f64 FPR64:$Rn),
4420 (vector_extract (v2f64 (fneg V128:$Rm)),
4421 VectorIndexS:$idx))),
4422 (FMLSv1i64_indexed FPR64:$Rd, FPR64:$Rn,
4423 V128:$Rm, VectorIndexS:$idx)>;
4426 defm : FMLSIndexedAfterNegPatterns<
4427 TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)> >;
4428 defm : FMLSIndexedAfterNegPatterns<
4429 TriOpFrag<(fma node:$MHS, node:$RHS, node:$LHS)> >;
4431 defm FMULX : SIMDFPIndexed<1, 0b1001, "fmulx", int_aarch64_neon_fmulx>;
4432 defm FMUL : SIMDFPIndexed<0, 0b1001, "fmul", fmul>;
4434 def : Pat<(v2f32 (fmul V64:$Rn, (AArch64dup (f32 FPR32:$Rm)))),
4435 (FMULv2i32_indexed V64:$Rn,
4436 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rm, ssub),
4438 def : Pat<(v4f32 (fmul V128:$Rn, (AArch64dup (f32 FPR32:$Rm)))),
4439 (FMULv4i32_indexed V128:$Rn,
4440 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rm, ssub),
4442 def : Pat<(v2f64 (fmul V128:$Rn, (AArch64dup (f64 FPR64:$Rm)))),
4443 (FMULv2i64_indexed V128:$Rn,
4444 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$Rm, dsub),
4447 defm SQDMULH : SIMDIndexedHS<0, 0b1100, "sqdmulh", int_aarch64_neon_sqdmulh>;
4448 defm SQRDMULH : SIMDIndexedHS<0, 0b1101, "sqrdmulh", int_aarch64_neon_sqrdmulh>;
4449 defm MLA : SIMDVectorIndexedHSTied<1, 0b0000, "mla",
4450 TriOpFrag<(add node:$LHS, (mul node:$MHS, node:$RHS))>>;
4451 defm MLS : SIMDVectorIndexedHSTied<1, 0b0100, "mls",
4452 TriOpFrag<(sub node:$LHS, (mul node:$MHS, node:$RHS))>>;
4453 defm MUL : SIMDVectorIndexedHS<0, 0b1000, "mul", mul>;
4454 defm SMLAL : SIMDVectorIndexedLongSDTied<0, 0b0010, "smlal",
4455 TriOpFrag<(add node:$LHS, (int_aarch64_neon_smull node:$MHS, node:$RHS))>>;
4456 defm SMLSL : SIMDVectorIndexedLongSDTied<0, 0b0110, "smlsl",
4457 TriOpFrag<(sub node:$LHS, (int_aarch64_neon_smull node:$MHS, node:$RHS))>>;
4458 defm SMULL : SIMDVectorIndexedLongSD<0, 0b1010, "smull",
4459 int_aarch64_neon_smull>;
4460 defm SQDMLAL : SIMDIndexedLongSQDMLXSDTied<0, 0b0011, "sqdmlal",
4461 int_aarch64_neon_sqadd>;
4462 defm SQDMLSL : SIMDIndexedLongSQDMLXSDTied<0, 0b0111, "sqdmlsl",
4463 int_aarch64_neon_sqsub>;
4464 defm SQRDMLAH : SIMDIndexedSQRDMLxHSDTied<1, 0b1101, "sqrdmlah",
4465 int_aarch64_neon_sqadd>;
4466 defm SQRDMLSH : SIMDIndexedSQRDMLxHSDTied<1, 0b1111, "sqrdmlsh",
4467 int_aarch64_neon_sqsub>;
4468 defm SQDMULL : SIMDIndexedLongSD<0, 0b1011, "sqdmull", int_aarch64_neon_sqdmull>;
4469 defm UMLAL : SIMDVectorIndexedLongSDTied<1, 0b0010, "umlal",
4470 TriOpFrag<(add node:$LHS, (int_aarch64_neon_umull node:$MHS, node:$RHS))>>;
4471 defm UMLSL : SIMDVectorIndexedLongSDTied<1, 0b0110, "umlsl",
4472 TriOpFrag<(sub node:$LHS, (int_aarch64_neon_umull node:$MHS, node:$RHS))>>;
4473 defm UMULL : SIMDVectorIndexedLongSD<1, 0b1010, "umull",
4474 int_aarch64_neon_umull>;
4476 // A scalar sqdmull with the second operand being a vector lane can be
4477 // handled directly with the indexed instruction encoding.
4478 def : Pat<(int_aarch64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
4479 (vector_extract (v4i32 V128:$Vm),
4480 VectorIndexS:$idx)),
4481 (SQDMULLv1i64_indexed FPR32:$Rn, V128:$Vm, VectorIndexS:$idx)>;
4483 //----------------------------------------------------------------------------
4484 // AdvSIMD scalar shift instructions
4485 //----------------------------------------------------------------------------
4486 defm FCVTZS : SIMDScalarRShiftSD<0, 0b11111, "fcvtzs">;
4487 defm FCVTZU : SIMDScalarRShiftSD<1, 0b11111, "fcvtzu">;
4488 defm SCVTF : SIMDScalarRShiftSD<0, 0b11100, "scvtf">;
4489 defm UCVTF : SIMDScalarRShiftSD<1, 0b11100, "ucvtf">;
4490 // Codegen patterns for the above. We don't put these directly on the
4491 // instructions because TableGen's type inference can't handle the truth.
4492 // Having the same base pattern for fp <--> int totally freaks it out.
4493 def : Pat<(int_aarch64_neon_vcvtfp2fxs FPR32:$Rn, vecshiftR32:$imm),
4494 (FCVTZSs FPR32:$Rn, vecshiftR32:$imm)>;
4495 def : Pat<(int_aarch64_neon_vcvtfp2fxu FPR32:$Rn, vecshiftR32:$imm),
4496 (FCVTZUs FPR32:$Rn, vecshiftR32:$imm)>;
4497 def : Pat<(i64 (int_aarch64_neon_vcvtfp2fxs (f64 FPR64:$Rn), vecshiftR64:$imm)),
4498 (FCVTZSd FPR64:$Rn, vecshiftR64:$imm)>;
4499 def : Pat<(i64 (int_aarch64_neon_vcvtfp2fxu (f64 FPR64:$Rn), vecshiftR64:$imm)),
4500 (FCVTZUd FPR64:$Rn, vecshiftR64:$imm)>;
4501 def : Pat<(v1i64 (int_aarch64_neon_vcvtfp2fxs (v1f64 FPR64:$Rn),
4503 (FCVTZSd FPR64:$Rn, vecshiftR64:$imm)>;
4504 def : Pat<(v1i64 (int_aarch64_neon_vcvtfp2fxu (v1f64 FPR64:$Rn),
4506 (FCVTZUd FPR64:$Rn, vecshiftR64:$imm)>;
4507 def : Pat<(int_aarch64_neon_vcvtfxs2fp FPR32:$Rn, vecshiftR32:$imm),
4508 (SCVTFs FPR32:$Rn, vecshiftR32:$imm)>;
4509 def : Pat<(int_aarch64_neon_vcvtfxu2fp FPR32:$Rn, vecshiftR32:$imm),
4510 (UCVTFs FPR32:$Rn, vecshiftR32:$imm)>;
4511 def : Pat<(f64 (int_aarch64_neon_vcvtfxs2fp (i64 FPR64:$Rn), vecshiftR64:$imm)),
4512 (SCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
4513 def : Pat<(f64 (int_aarch64_neon_vcvtfxu2fp (i64 FPR64:$Rn), vecshiftR64:$imm)),
4514 (UCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
4515 def : Pat<(v1f64 (int_aarch64_neon_vcvtfxs2fp (v1i64 FPR64:$Rn),
4517 (SCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
4518 def : Pat<(v1f64 (int_aarch64_neon_vcvtfxu2fp (v1i64 FPR64:$Rn),
4520 (UCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
4522 defm SHL : SIMDScalarLShiftD< 0, 0b01010, "shl", AArch64vshl>;
4523 defm SLI : SIMDScalarLShiftDTied<1, 0b01010, "sli">;
4524 defm SQRSHRN : SIMDScalarRShiftBHS< 0, 0b10011, "sqrshrn",
4525 int_aarch64_neon_sqrshrn>;
4526 defm SQRSHRUN : SIMDScalarRShiftBHS< 1, 0b10001, "sqrshrun",
4527 int_aarch64_neon_sqrshrun>;
4528 defm SQSHLU : SIMDScalarLShiftBHSD<1, 0b01100, "sqshlu", AArch64sqshlui>;
4529 defm SQSHL : SIMDScalarLShiftBHSD<0, 0b01110, "sqshl", AArch64sqshli>;
4530 defm SQSHRN : SIMDScalarRShiftBHS< 0, 0b10010, "sqshrn",
4531 int_aarch64_neon_sqshrn>;
4532 defm SQSHRUN : SIMDScalarRShiftBHS< 1, 0b10000, "sqshrun",
4533 int_aarch64_neon_sqshrun>;
4534 defm SRI : SIMDScalarRShiftDTied< 1, 0b01000, "sri">;
4535 defm SRSHR : SIMDScalarRShiftD< 0, 0b00100, "srshr", AArch64srshri>;
4536 defm SRSRA : SIMDScalarRShiftDTied< 0, 0b00110, "srsra",
4537 TriOpFrag<(add node:$LHS,
4538 (AArch64srshri node:$MHS, node:$RHS))>>;
4539 defm SSHR : SIMDScalarRShiftD< 0, 0b00000, "sshr", AArch64vashr>;
4540 defm SSRA : SIMDScalarRShiftDTied< 0, 0b00010, "ssra",
4541 TriOpFrag<(add node:$LHS,
4542 (AArch64vashr node:$MHS, node:$RHS))>>;
4543 defm UQRSHRN : SIMDScalarRShiftBHS< 1, 0b10011, "uqrshrn",
4544 int_aarch64_neon_uqrshrn>;
4545 defm UQSHL : SIMDScalarLShiftBHSD<1, 0b01110, "uqshl", AArch64uqshli>;
4546 defm UQSHRN : SIMDScalarRShiftBHS< 1, 0b10010, "uqshrn",
4547 int_aarch64_neon_uqshrn>;
4548 defm URSHR : SIMDScalarRShiftD< 1, 0b00100, "urshr", AArch64urshri>;
4549 defm URSRA : SIMDScalarRShiftDTied< 1, 0b00110, "ursra",
4550 TriOpFrag<(add node:$LHS,
4551 (AArch64urshri node:$MHS, node:$RHS))>>;
4552 defm USHR : SIMDScalarRShiftD< 1, 0b00000, "ushr", AArch64vlshr>;
4553 defm USRA : SIMDScalarRShiftDTied< 1, 0b00010, "usra",
4554 TriOpFrag<(add node:$LHS,
4555 (AArch64vlshr node:$MHS, node:$RHS))>>;
4557 //----------------------------------------------------------------------------
4558 // AdvSIMD vector shift instructions
4559 //----------------------------------------------------------------------------
4560 defm FCVTZS:SIMDVectorRShiftSD<0, 0b11111, "fcvtzs", int_aarch64_neon_vcvtfp2fxs>;
4561 defm FCVTZU:SIMDVectorRShiftSD<1, 0b11111, "fcvtzu", int_aarch64_neon_vcvtfp2fxu>;
4562 defm SCVTF: SIMDVectorRShiftSDToFP<0, 0b11100, "scvtf",
4563 int_aarch64_neon_vcvtfxs2fp>;
4564 defm RSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10001, "rshrn",
4565 int_aarch64_neon_rshrn>;
4566 defm SHL : SIMDVectorLShiftBHSD<0, 0b01010, "shl", AArch64vshl>;
4567 defm SHRN : SIMDVectorRShiftNarrowBHS<0, 0b10000, "shrn",
4568 BinOpFrag<(trunc (AArch64vashr node:$LHS, node:$RHS))>>;
4569 defm SLI : SIMDVectorLShiftBHSDTied<1, 0b01010, "sli", int_aarch64_neon_vsli>;
4570 def : Pat<(v1i64 (int_aarch64_neon_vsli (v1i64 FPR64:$Rd), (v1i64 FPR64:$Rn),
4571 (i32 vecshiftL64:$imm))),
4572 (SLId FPR64:$Rd, FPR64:$Rn, vecshiftL64:$imm)>;
4573 defm SQRSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10011, "sqrshrn",
4574 int_aarch64_neon_sqrshrn>;
4575 defm SQRSHRUN: SIMDVectorRShiftNarrowBHS<1, 0b10001, "sqrshrun",
4576 int_aarch64_neon_sqrshrun>;
4577 defm SQSHLU : SIMDVectorLShiftBHSD<1, 0b01100, "sqshlu", AArch64sqshlui>;
4578 defm SQSHL : SIMDVectorLShiftBHSD<0, 0b01110, "sqshl", AArch64sqshli>;
4579 defm SQSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10010, "sqshrn",
4580 int_aarch64_neon_sqshrn>;
4581 defm SQSHRUN : SIMDVectorRShiftNarrowBHS<1, 0b10000, "sqshrun",
4582 int_aarch64_neon_sqshrun>;
4583 defm SRI : SIMDVectorRShiftBHSDTied<1, 0b01000, "sri", int_aarch64_neon_vsri>;
4584 def : Pat<(v1i64 (int_aarch64_neon_vsri (v1i64 FPR64:$Rd), (v1i64 FPR64:$Rn),
4585 (i32 vecshiftR64:$imm))),
4586 (SRId FPR64:$Rd, FPR64:$Rn, vecshiftR64:$imm)>;
4587 defm SRSHR : SIMDVectorRShiftBHSD<0, 0b00100, "srshr", AArch64srshri>;
4588 defm SRSRA : SIMDVectorRShiftBHSDTied<0, 0b00110, "srsra",
4589 TriOpFrag<(add node:$LHS,
4590 (AArch64srshri node:$MHS, node:$RHS))> >;
4591 defm SSHLL : SIMDVectorLShiftLongBHSD<0, 0b10100, "sshll",
4592 BinOpFrag<(AArch64vshl (sext node:$LHS), node:$RHS)>>;
4594 defm SSHR : SIMDVectorRShiftBHSD<0, 0b00000, "sshr", AArch64vashr>;
4595 defm SSRA : SIMDVectorRShiftBHSDTied<0, 0b00010, "ssra",
4596 TriOpFrag<(add node:$LHS, (AArch64vashr node:$MHS, node:$RHS))>>;
4597 defm UCVTF : SIMDVectorRShiftSDToFP<1, 0b11100, "ucvtf",
4598 int_aarch64_neon_vcvtfxu2fp>;
4599 defm UQRSHRN : SIMDVectorRShiftNarrowBHS<1, 0b10011, "uqrshrn",
4600 int_aarch64_neon_uqrshrn>;
4601 defm UQSHL : SIMDVectorLShiftBHSD<1, 0b01110, "uqshl", AArch64uqshli>;
4602 defm UQSHRN : SIMDVectorRShiftNarrowBHS<1, 0b10010, "uqshrn",
4603 int_aarch64_neon_uqshrn>;
4604 defm URSHR : SIMDVectorRShiftBHSD<1, 0b00100, "urshr", AArch64urshri>;
4605 defm URSRA : SIMDVectorRShiftBHSDTied<1, 0b00110, "ursra",
4606 TriOpFrag<(add node:$LHS,
4607 (AArch64urshri node:$MHS, node:$RHS))> >;
4608 defm USHLL : SIMDVectorLShiftLongBHSD<1, 0b10100, "ushll",
4609 BinOpFrag<(AArch64vshl (zext node:$LHS), node:$RHS)>>;
4610 defm USHR : SIMDVectorRShiftBHSD<1, 0b00000, "ushr", AArch64vlshr>;
4611 defm USRA : SIMDVectorRShiftBHSDTied<1, 0b00010, "usra",
4612 TriOpFrag<(add node:$LHS, (AArch64vlshr node:$MHS, node:$RHS))> >;
4614 // SHRN patterns for when a logical right shift was used instead of arithmetic
4615 // (the immediate guarantees no sign bits actually end up in the result so it
4617 def : Pat<(v8i8 (trunc (AArch64vlshr (v8i16 V128:$Rn), vecshiftR16Narrow:$imm))),
4618 (SHRNv8i8_shift V128:$Rn, vecshiftR16Narrow:$imm)>;
4619 def : Pat<(v4i16 (trunc (AArch64vlshr (v4i32 V128:$Rn), vecshiftR32Narrow:$imm))),
4620 (SHRNv4i16_shift V128:$Rn, vecshiftR32Narrow:$imm)>;
4621 def : Pat<(v2i32 (trunc (AArch64vlshr (v2i64 V128:$Rn), vecshiftR64Narrow:$imm))),
4622 (SHRNv2i32_shift V128:$Rn, vecshiftR64Narrow:$imm)>;
4624 def : Pat<(v16i8 (concat_vectors (v8i8 V64:$Rd),
4625 (trunc (AArch64vlshr (v8i16 V128:$Rn),
4626 vecshiftR16Narrow:$imm)))),
4627 (SHRNv16i8_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
4628 V128:$Rn, vecshiftR16Narrow:$imm)>;
4629 def : Pat<(v8i16 (concat_vectors (v4i16 V64:$Rd),
4630 (trunc (AArch64vlshr (v4i32 V128:$Rn),
4631 vecshiftR32Narrow:$imm)))),
4632 (SHRNv8i16_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
4633 V128:$Rn, vecshiftR32Narrow:$imm)>;
4634 def : Pat<(v4i32 (concat_vectors (v2i32 V64:$Rd),
4635 (trunc (AArch64vlshr (v2i64 V128:$Rn),
4636 vecshiftR64Narrow:$imm)))),
4637 (SHRNv4i32_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
4638 V128:$Rn, vecshiftR32Narrow:$imm)>;
4640 // Vector sign and zero extensions are implemented with SSHLL and USSHLL.
4641 // Anyexts are implemented as zexts.
4642 def : Pat<(v8i16 (sext (v8i8 V64:$Rn))), (SSHLLv8i8_shift V64:$Rn, (i32 0))>;
4643 def : Pat<(v8i16 (zext (v8i8 V64:$Rn))), (USHLLv8i8_shift V64:$Rn, (i32 0))>;
4644 def : Pat<(v8i16 (anyext (v8i8 V64:$Rn))), (USHLLv8i8_shift V64:$Rn, (i32 0))>;
4645 def : Pat<(v4i32 (sext (v4i16 V64:$Rn))), (SSHLLv4i16_shift V64:$Rn, (i32 0))>;
4646 def : Pat<(v4i32 (zext (v4i16 V64:$Rn))), (USHLLv4i16_shift V64:$Rn, (i32 0))>;
4647 def : Pat<(v4i32 (anyext (v4i16 V64:$Rn))), (USHLLv4i16_shift V64:$Rn, (i32 0))>;
4648 def : Pat<(v2i64 (sext (v2i32 V64:$Rn))), (SSHLLv2i32_shift V64:$Rn, (i32 0))>;
4649 def : Pat<(v2i64 (zext (v2i32 V64:$Rn))), (USHLLv2i32_shift V64:$Rn, (i32 0))>;
4650 def : Pat<(v2i64 (anyext (v2i32 V64:$Rn))), (USHLLv2i32_shift V64:$Rn, (i32 0))>;
4651 // Also match an extend from the upper half of a 128 bit source register.
4652 def : Pat<(v8i16 (anyext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
4653 (USHLLv16i8_shift V128:$Rn, (i32 0))>;
4654 def : Pat<(v8i16 (zext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
4655 (USHLLv16i8_shift V128:$Rn, (i32 0))>;
4656 def : Pat<(v8i16 (sext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
4657 (SSHLLv16i8_shift V128:$Rn, (i32 0))>;
4658 def : Pat<(v4i32 (anyext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
4659 (USHLLv8i16_shift V128:$Rn, (i32 0))>;
4660 def : Pat<(v4i32 (zext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
4661 (USHLLv8i16_shift V128:$Rn, (i32 0))>;
4662 def : Pat<(v4i32 (sext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
4663 (SSHLLv8i16_shift V128:$Rn, (i32 0))>;
4664 def : Pat<(v2i64 (anyext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
4665 (USHLLv4i32_shift V128:$Rn, (i32 0))>;
4666 def : Pat<(v2i64 (zext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
4667 (USHLLv4i32_shift V128:$Rn, (i32 0))>;
4668 def : Pat<(v2i64 (sext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
4669 (SSHLLv4i32_shift V128:$Rn, (i32 0))>;
4671 // Vector shift sxtl aliases
4672 def : InstAlias<"sxtl.8h $dst, $src1",
4673 (SSHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
4674 def : InstAlias<"sxtl $dst.8h, $src1.8b",
4675 (SSHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
4676 def : InstAlias<"sxtl.4s $dst, $src1",
4677 (SSHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
4678 def : InstAlias<"sxtl $dst.4s, $src1.4h",
4679 (SSHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
4680 def : InstAlias<"sxtl.2d $dst, $src1",
4681 (SSHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
4682 def : InstAlias<"sxtl $dst.2d, $src1.2s",
4683 (SSHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
4685 // Vector shift sxtl2 aliases
4686 def : InstAlias<"sxtl2.8h $dst, $src1",
4687 (SSHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
4688 def : InstAlias<"sxtl2 $dst.8h, $src1.16b",
4689 (SSHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
4690 def : InstAlias<"sxtl2.4s $dst, $src1",
4691 (SSHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
4692 def : InstAlias<"sxtl2 $dst.4s, $src1.8h",
4693 (SSHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
4694 def : InstAlias<"sxtl2.2d $dst, $src1",
4695 (SSHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
4696 def : InstAlias<"sxtl2 $dst.2d, $src1.4s",
4697 (SSHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
4699 // Vector shift uxtl aliases
4700 def : InstAlias<"uxtl.8h $dst, $src1",
4701 (USHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
4702 def : InstAlias<"uxtl $dst.8h, $src1.8b",
4703 (USHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
4704 def : InstAlias<"uxtl.4s $dst, $src1",
4705 (USHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
4706 def : InstAlias<"uxtl $dst.4s, $src1.4h",
4707 (USHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
4708 def : InstAlias<"uxtl.2d $dst, $src1",
4709 (USHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
4710 def : InstAlias<"uxtl $dst.2d, $src1.2s",
4711 (USHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
4713 // Vector shift uxtl2 aliases
4714 def : InstAlias<"uxtl2.8h $dst, $src1",
4715 (USHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
4716 def : InstAlias<"uxtl2 $dst.8h, $src1.16b",
4717 (USHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
4718 def : InstAlias<"uxtl2.4s $dst, $src1",
4719 (USHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
4720 def : InstAlias<"uxtl2 $dst.4s, $src1.8h",
4721 (USHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
4722 def : InstAlias<"uxtl2.2d $dst, $src1",
4723 (USHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
4724 def : InstAlias<"uxtl2 $dst.2d, $src1.4s",
4725 (USHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
4727 // If an integer is about to be converted to a floating point value,
4728 // just load it on the floating point unit.
4729 // These patterns are more complex because floating point loads do not
4730 // support sign extension.
4731 // The sign extension has to be explicitly added and is only supported for
4732 // one step: byte-to-half, half-to-word, word-to-doubleword.
4733 // SCVTF GPR -> FPR is 9 cycles.
4734 // SCVTF FPR -> FPR is 4 cyclces.
4735 // (sign extension with lengthen) SXTL FPR -> FPR is 2 cycles.
4736 // Therefore, we can do 2 sign extensions and one SCVTF FPR -> FPR
4737 // and still being faster.
4738 // However, this is not good for code size.
4739 // 8-bits -> float. 2 sizes step-up.
4740 class SExtLoadi8CVTf32Pat<dag addrmode, dag INST>
4741 : Pat<(f32 (sint_to_fp (i32 (sextloadi8 addrmode)))),
4742 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
4747 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4753 ssub)))>, Requires<[NotForCodeSize, IsCyclone]>;
4755 def : SExtLoadi8CVTf32Pat<(ro8.Wpat GPR64sp:$Rn, GPR32:$Rm, ro8.Wext:$ext),
4756 (LDRBroW GPR64sp:$Rn, GPR32:$Rm, ro8.Wext:$ext)>;
4757 def : SExtLoadi8CVTf32Pat<(ro8.Xpat GPR64sp:$Rn, GPR64:$Rm, ro8.Xext:$ext),
4758 (LDRBroX GPR64sp:$Rn, GPR64:$Rm, ro8.Xext:$ext)>;
4759 def : SExtLoadi8CVTf32Pat<(am_indexed8 GPR64sp:$Rn, uimm12s1:$offset),
4760 (LDRBui GPR64sp:$Rn, uimm12s1:$offset)>;
4761 def : SExtLoadi8CVTf32Pat<(am_unscaled8 GPR64sp:$Rn, simm9:$offset),
4762 (LDURBi GPR64sp:$Rn, simm9:$offset)>;
4764 // 16-bits -> float. 1 size step-up.
4765 class SExtLoadi16CVTf32Pat<dag addrmode, dag INST>
4766 : Pat<(f32 (sint_to_fp (i32 (sextloadi16 addrmode)))),
4767 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
4769 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4773 ssub)))>, Requires<[NotForCodeSize]>;
4775 def : SExtLoadi16CVTf32Pat<(ro16.Wpat GPR64sp:$Rn, GPR32:$Rm, ro16.Wext:$ext),
4776 (LDRHroW GPR64sp:$Rn, GPR32:$Rm, ro16.Wext:$ext)>;
4777 def : SExtLoadi16CVTf32Pat<(ro16.Xpat GPR64sp:$Rn, GPR64:$Rm, ro16.Xext:$ext),
4778 (LDRHroX GPR64sp:$Rn, GPR64:$Rm, ro16.Xext:$ext)>;
4779 def : SExtLoadi16CVTf32Pat<(am_indexed16 GPR64sp:$Rn, uimm12s2:$offset),
4780 (LDRHui GPR64sp:$Rn, uimm12s2:$offset)>;
4781 def : SExtLoadi16CVTf32Pat<(am_unscaled16 GPR64sp:$Rn, simm9:$offset),
4782 (LDURHi GPR64sp:$Rn, simm9:$offset)>;
4784 // 32-bits to 32-bits are handled in target specific dag combine:
4785 // performIntToFpCombine.
4786 // 64-bits integer to 32-bits floating point, not possible with
4787 // SCVTF on floating point registers (both source and destination
4788 // must have the same size).
4790 // Here are the patterns for 8, 16, 32, and 64-bits to double.
4791 // 8-bits -> double. 3 size step-up: give up.
4792 // 16-bits -> double. 2 size step.
4793 class SExtLoadi16CVTf64Pat<dag addrmode, dag INST>
4794 : Pat <(f64 (sint_to_fp (i32 (sextloadi16 addrmode)))),
4795 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4800 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4806 dsub)))>, Requires<[NotForCodeSize, IsCyclone]>;
4808 def : SExtLoadi16CVTf64Pat<(ro16.Wpat GPR64sp:$Rn, GPR32:$Rm, ro16.Wext:$ext),
4809 (LDRHroW GPR64sp:$Rn, GPR32:$Rm, ro16.Wext:$ext)>;
4810 def : SExtLoadi16CVTf64Pat<(ro16.Xpat GPR64sp:$Rn, GPR64:$Rm, ro16.Xext:$ext),
4811 (LDRHroX GPR64sp:$Rn, GPR64:$Rm, ro16.Xext:$ext)>;
4812 def : SExtLoadi16CVTf64Pat<(am_indexed16 GPR64sp:$Rn, uimm12s2:$offset),
4813 (LDRHui GPR64sp:$Rn, uimm12s2:$offset)>;
4814 def : SExtLoadi16CVTf64Pat<(am_unscaled16 GPR64sp:$Rn, simm9:$offset),
4815 (LDURHi GPR64sp:$Rn, simm9:$offset)>;
4816 // 32-bits -> double. 1 size step-up.
4817 class SExtLoadi32CVTf64Pat<dag addrmode, dag INST>
4818 : Pat <(f64 (sint_to_fp (i32 (load addrmode)))),
4819 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4821 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4825 dsub)))>, Requires<[NotForCodeSize]>;
4827 def : SExtLoadi32CVTf64Pat<(ro32.Wpat GPR64sp:$Rn, GPR32:$Rm, ro32.Wext:$ext),
4828 (LDRSroW GPR64sp:$Rn, GPR32:$Rm, ro32.Wext:$ext)>;
4829 def : SExtLoadi32CVTf64Pat<(ro32.Xpat GPR64sp:$Rn, GPR64:$Rm, ro32.Xext:$ext),
4830 (LDRSroX GPR64sp:$Rn, GPR64:$Rm, ro32.Xext:$ext)>;
4831 def : SExtLoadi32CVTf64Pat<(am_indexed32 GPR64sp:$Rn, uimm12s4:$offset),
4832 (LDRSui GPR64sp:$Rn, uimm12s4:$offset)>;
4833 def : SExtLoadi32CVTf64Pat<(am_unscaled32 GPR64sp:$Rn, simm9:$offset),
4834 (LDURSi GPR64sp:$Rn, simm9:$offset)>;
4836 // 64-bits -> double are handled in target specific dag combine:
4837 // performIntToFpCombine.
4840 //----------------------------------------------------------------------------
4841 // AdvSIMD Load-Store Structure
4842 //----------------------------------------------------------------------------
4843 defm LD1 : SIMDLd1Multiple<"ld1">;
4844 defm LD2 : SIMDLd2Multiple<"ld2">;
4845 defm LD3 : SIMDLd3Multiple<"ld3">;
4846 defm LD4 : SIMDLd4Multiple<"ld4">;
4848 defm ST1 : SIMDSt1Multiple<"st1">;
4849 defm ST2 : SIMDSt2Multiple<"st2">;
4850 defm ST3 : SIMDSt3Multiple<"st3">;
4851 defm ST4 : SIMDSt4Multiple<"st4">;
4853 class Ld1Pat<ValueType ty, Instruction INST>
4854 : Pat<(ty (load GPR64sp:$Rn)), (INST GPR64sp:$Rn)>;
4856 def : Ld1Pat<v16i8, LD1Onev16b>;
4857 def : Ld1Pat<v8i16, LD1Onev8h>;
4858 def : Ld1Pat<v4i32, LD1Onev4s>;
4859 def : Ld1Pat<v2i64, LD1Onev2d>;
4860 def : Ld1Pat<v8i8, LD1Onev8b>;
4861 def : Ld1Pat<v4i16, LD1Onev4h>;
4862 def : Ld1Pat<v2i32, LD1Onev2s>;
4863 def : Ld1Pat<v1i64, LD1Onev1d>;
4865 class St1Pat<ValueType ty, Instruction INST>
4866 : Pat<(store ty:$Vt, GPR64sp:$Rn),
4867 (INST ty:$Vt, GPR64sp:$Rn)>;
4869 def : St1Pat<v16i8, ST1Onev16b>;
4870 def : St1Pat<v8i16, ST1Onev8h>;
4871 def : St1Pat<v4i32, ST1Onev4s>;
4872 def : St1Pat<v2i64, ST1Onev2d>;
4873 def : St1Pat<v8i8, ST1Onev8b>;
4874 def : St1Pat<v4i16, ST1Onev4h>;
4875 def : St1Pat<v2i32, ST1Onev2s>;
4876 def : St1Pat<v1i64, ST1Onev1d>;
4882 defm LD1R : SIMDLdR<0, 0b110, 0, "ld1r", "One", 1, 2, 4, 8>;
4883 defm LD2R : SIMDLdR<1, 0b110, 0, "ld2r", "Two", 2, 4, 8, 16>;
4884 defm LD3R : SIMDLdR<0, 0b111, 0, "ld3r", "Three", 3, 6, 12, 24>;
4885 defm LD4R : SIMDLdR<1, 0b111, 0, "ld4r", "Four", 4, 8, 16, 32>;
4886 let mayLoad = 1, hasSideEffects = 0 in {
4887 defm LD1 : SIMDLdSingleBTied<0, 0b000, "ld1", VecListOneb, GPR64pi1>;
4888 defm LD1 : SIMDLdSingleHTied<0, 0b010, 0, "ld1", VecListOneh, GPR64pi2>;
4889 defm LD1 : SIMDLdSingleSTied<0, 0b100, 0b00, "ld1", VecListOnes, GPR64pi4>;
4890 defm LD1 : SIMDLdSingleDTied<0, 0b100, 0b01, "ld1", VecListOned, GPR64pi8>;
4891 defm LD2 : SIMDLdSingleBTied<1, 0b000, "ld2", VecListTwob, GPR64pi2>;
4892 defm LD2 : SIMDLdSingleHTied<1, 0b010, 0, "ld2", VecListTwoh, GPR64pi4>;
4893 defm LD2 : SIMDLdSingleSTied<1, 0b100, 0b00, "ld2", VecListTwos, GPR64pi8>;
4894 defm LD2 : SIMDLdSingleDTied<1, 0b100, 0b01, "ld2", VecListTwod, GPR64pi16>;
4895 defm LD3 : SIMDLdSingleBTied<0, 0b001, "ld3", VecListThreeb, GPR64pi3>;
4896 defm LD3 : SIMDLdSingleHTied<0, 0b011, 0, "ld3", VecListThreeh, GPR64pi6>;
4897 defm LD3 : SIMDLdSingleSTied<0, 0b101, 0b00, "ld3", VecListThrees, GPR64pi12>;
4898 defm LD3 : SIMDLdSingleDTied<0, 0b101, 0b01, "ld3", VecListThreed, GPR64pi24>;
4899 defm LD4 : SIMDLdSingleBTied<1, 0b001, "ld4", VecListFourb, GPR64pi4>;
4900 defm LD4 : SIMDLdSingleHTied<1, 0b011, 0, "ld4", VecListFourh, GPR64pi8>;
4901 defm LD4 : SIMDLdSingleSTied<1, 0b101, 0b00, "ld4", VecListFours, GPR64pi16>;
4902 defm LD4 : SIMDLdSingleDTied<1, 0b101, 0b01, "ld4", VecListFourd, GPR64pi32>;
4905 def : Pat<(v8i8 (AArch64dup (i32 (extloadi8 GPR64sp:$Rn)))),
4906 (LD1Rv8b GPR64sp:$Rn)>;
4907 def : Pat<(v16i8 (AArch64dup (i32 (extloadi8 GPR64sp:$Rn)))),
4908 (LD1Rv16b GPR64sp:$Rn)>;
4909 def : Pat<(v4i16 (AArch64dup (i32 (extloadi16 GPR64sp:$Rn)))),
4910 (LD1Rv4h GPR64sp:$Rn)>;
4911 def : Pat<(v8i16 (AArch64dup (i32 (extloadi16 GPR64sp:$Rn)))),
4912 (LD1Rv8h GPR64sp:$Rn)>;
4913 def : Pat<(v2i32 (AArch64dup (i32 (load GPR64sp:$Rn)))),
4914 (LD1Rv2s GPR64sp:$Rn)>;
4915 def : Pat<(v4i32 (AArch64dup (i32 (load GPR64sp:$Rn)))),
4916 (LD1Rv4s GPR64sp:$Rn)>;
4917 def : Pat<(v2i64 (AArch64dup (i64 (load GPR64sp:$Rn)))),
4918 (LD1Rv2d GPR64sp:$Rn)>;
4919 def : Pat<(v1i64 (AArch64dup (i64 (load GPR64sp:$Rn)))),
4920 (LD1Rv1d GPR64sp:$Rn)>;
4921 // Grab the floating point version too
4922 def : Pat<(v2f32 (AArch64dup (f32 (load GPR64sp:$Rn)))),
4923 (LD1Rv2s GPR64sp:$Rn)>;
4924 def : Pat<(v4f32 (AArch64dup (f32 (load GPR64sp:$Rn)))),
4925 (LD1Rv4s GPR64sp:$Rn)>;
4926 def : Pat<(v2f64 (AArch64dup (f64 (load GPR64sp:$Rn)))),
4927 (LD1Rv2d GPR64sp:$Rn)>;
4928 def : Pat<(v1f64 (AArch64dup (f64 (load GPR64sp:$Rn)))),
4929 (LD1Rv1d GPR64sp:$Rn)>;
4930 def : Pat<(v4f16 (AArch64dup (f16 (load GPR64sp:$Rn)))),
4931 (LD1Rv4h GPR64sp:$Rn)>;
4932 def : Pat<(v8f16 (AArch64dup (f16 (load GPR64sp:$Rn)))),
4933 (LD1Rv8h GPR64sp:$Rn)>;
4935 class Ld1Lane128Pat<SDPatternOperator scalar_load, Operand VecIndex,
4936 ValueType VTy, ValueType STy, Instruction LD1>
4937 : Pat<(vector_insert (VTy VecListOne128:$Rd),
4938 (STy (scalar_load GPR64sp:$Rn)), VecIndex:$idx),
4939 (LD1 VecListOne128:$Rd, VecIndex:$idx, GPR64sp:$Rn)>;
4941 def : Ld1Lane128Pat<extloadi8, VectorIndexB, v16i8, i32, LD1i8>;
4942 def : Ld1Lane128Pat<extloadi16, VectorIndexH, v8i16, i32, LD1i16>;
4943 def : Ld1Lane128Pat<load, VectorIndexS, v4i32, i32, LD1i32>;
4944 def : Ld1Lane128Pat<load, VectorIndexS, v4f32, f32, LD1i32>;
4945 def : Ld1Lane128Pat<load, VectorIndexD, v2i64, i64, LD1i64>;
4946 def : Ld1Lane128Pat<load, VectorIndexD, v2f64, f64, LD1i64>;
4947 def : Ld1Lane128Pat<load, VectorIndexH, v8f16, f16, LD1i16>;
4949 class Ld1Lane64Pat<SDPatternOperator scalar_load, Operand VecIndex,
4950 ValueType VTy, ValueType STy, Instruction LD1>
4951 : Pat<(vector_insert (VTy VecListOne64:$Rd),
4952 (STy (scalar_load GPR64sp:$Rn)), VecIndex:$idx),
4954 (LD1 (SUBREG_TO_REG (i32 0), VecListOne64:$Rd, dsub),
4955 VecIndex:$idx, GPR64sp:$Rn),
4958 def : Ld1Lane64Pat<extloadi8, VectorIndexB, v8i8, i32, LD1i8>;
4959 def : Ld1Lane64Pat<extloadi16, VectorIndexH, v4i16, i32, LD1i16>;
4960 def : Ld1Lane64Pat<load, VectorIndexS, v2i32, i32, LD1i32>;
4961 def : Ld1Lane64Pat<load, VectorIndexS, v2f32, f32, LD1i32>;
4962 def : Ld1Lane64Pat<load, VectorIndexH, v4f16, f16, LD1i16>;
4965 defm LD1 : SIMDLdSt1SingleAliases<"ld1">;
4966 defm LD2 : SIMDLdSt2SingleAliases<"ld2">;
4967 defm LD3 : SIMDLdSt3SingleAliases<"ld3">;
4968 defm LD4 : SIMDLdSt4SingleAliases<"ld4">;
4971 defm ST1 : SIMDStSingleB<0, 0b000, "st1", VecListOneb, GPR64pi1>;
4972 defm ST1 : SIMDStSingleH<0, 0b010, 0, "st1", VecListOneh, GPR64pi2>;
4973 defm ST1 : SIMDStSingleS<0, 0b100, 0b00, "st1", VecListOnes, GPR64pi4>;
4974 defm ST1 : SIMDStSingleD<0, 0b100, 0b01, "st1", VecListOned, GPR64pi8>;
4976 let AddedComplexity = 19 in
4977 class St1Lane128Pat<SDPatternOperator scalar_store, Operand VecIndex,
4978 ValueType VTy, ValueType STy, Instruction ST1>
4980 (STy (vector_extract (VTy VecListOne128:$Vt), VecIndex:$idx)),
4982 (ST1 VecListOne128:$Vt, VecIndex:$idx, GPR64sp:$Rn)>;
4984 def : St1Lane128Pat<truncstorei8, VectorIndexB, v16i8, i32, ST1i8>;
4985 def : St1Lane128Pat<truncstorei16, VectorIndexH, v8i16, i32, ST1i16>;
4986 def : St1Lane128Pat<store, VectorIndexS, v4i32, i32, ST1i32>;
4987 def : St1Lane128Pat<store, VectorIndexS, v4f32, f32, ST1i32>;
4988 def : St1Lane128Pat<store, VectorIndexD, v2i64, i64, ST1i64>;
4989 def : St1Lane128Pat<store, VectorIndexD, v2f64, f64, ST1i64>;
4990 def : St1Lane128Pat<store, VectorIndexH, v8f16, f16, ST1i16>;
4992 let AddedComplexity = 19 in
4993 class St1Lane64Pat<SDPatternOperator scalar_store, Operand VecIndex,
4994 ValueType VTy, ValueType STy, Instruction ST1>
4996 (STy (vector_extract (VTy VecListOne64:$Vt), VecIndex:$idx)),
4998 (ST1 (SUBREG_TO_REG (i32 0), VecListOne64:$Vt, dsub),
4999 VecIndex:$idx, GPR64sp:$Rn)>;
5001 def : St1Lane64Pat<truncstorei8, VectorIndexB, v8i8, i32, ST1i8>;
5002 def : St1Lane64Pat<truncstorei16, VectorIndexH, v4i16, i32, ST1i16>;
5003 def : St1Lane64Pat<store, VectorIndexS, v2i32, i32, ST1i32>;
5004 def : St1Lane64Pat<store, VectorIndexS, v2f32, f32, ST1i32>;
5005 def : St1Lane64Pat<store, VectorIndexH, v4f16, f16, ST1i16>;
5007 multiclass St1LanePost64Pat<SDPatternOperator scalar_store, Operand VecIndex,
5008 ValueType VTy, ValueType STy, Instruction ST1,
5010 def : Pat<(scalar_store
5011 (STy (vector_extract (VTy VecListOne64:$Vt), VecIndex:$idx)),
5012 GPR64sp:$Rn, offset),
5013 (ST1 (SUBREG_TO_REG (i32 0), VecListOne64:$Vt, dsub),
5014 VecIndex:$idx, GPR64sp:$Rn, XZR)>;
5016 def : Pat<(scalar_store
5017 (STy (vector_extract (VTy VecListOne64:$Vt), VecIndex:$idx)),
5018 GPR64sp:$Rn, GPR64:$Rm),
5019 (ST1 (SUBREG_TO_REG (i32 0), VecListOne64:$Vt, dsub),
5020 VecIndex:$idx, GPR64sp:$Rn, $Rm)>;
5023 defm : St1LanePost64Pat<post_truncsti8, VectorIndexB, v8i8, i32, ST1i8_POST, 1>;
5024 defm : St1LanePost64Pat<post_truncsti16, VectorIndexH, v4i16, i32, ST1i16_POST,
5026 defm : St1LanePost64Pat<post_store, VectorIndexS, v2i32, i32, ST1i32_POST, 4>;
5027 defm : St1LanePost64Pat<post_store, VectorIndexS, v2f32, f32, ST1i32_POST, 4>;
5028 defm : St1LanePost64Pat<post_store, VectorIndexD, v1i64, i64, ST1i64_POST, 8>;
5029 defm : St1LanePost64Pat<post_store, VectorIndexD, v1f64, f64, ST1i64_POST, 8>;
5030 defm : St1LanePost64Pat<post_store, VectorIndexH, v4f16, f16, ST1i16_POST, 2>;
5032 multiclass St1LanePost128Pat<SDPatternOperator scalar_store, Operand VecIndex,
5033 ValueType VTy, ValueType STy, Instruction ST1,
5035 def : Pat<(scalar_store
5036 (STy (vector_extract (VTy VecListOne128:$Vt), VecIndex:$idx)),
5037 GPR64sp:$Rn, offset),
5038 (ST1 VecListOne128:$Vt, VecIndex:$idx, GPR64sp:$Rn, XZR)>;
5040 def : Pat<(scalar_store
5041 (STy (vector_extract (VTy VecListOne128:$Vt), VecIndex:$idx)),
5042 GPR64sp:$Rn, GPR64:$Rm),
5043 (ST1 VecListOne128:$Vt, VecIndex:$idx, GPR64sp:$Rn, $Rm)>;
5046 defm : St1LanePost128Pat<post_truncsti8, VectorIndexB, v16i8, i32, ST1i8_POST,
5048 defm : St1LanePost128Pat<post_truncsti16, VectorIndexH, v8i16, i32, ST1i16_POST,
5050 defm : St1LanePost128Pat<post_store, VectorIndexS, v4i32, i32, ST1i32_POST, 4>;
5051 defm : St1LanePost128Pat<post_store, VectorIndexS, v4f32, f32, ST1i32_POST, 4>;
5052 defm : St1LanePost128Pat<post_store, VectorIndexD, v2i64, i64, ST1i64_POST, 8>;
5053 defm : St1LanePost128Pat<post_store, VectorIndexD, v2f64, f64, ST1i64_POST, 8>;
5054 defm : St1LanePost128Pat<post_store, VectorIndexH, v8f16, f16, ST1i16_POST, 2>;
5056 let mayStore = 1, hasSideEffects = 0 in {
5057 defm ST2 : SIMDStSingleB<1, 0b000, "st2", VecListTwob, GPR64pi2>;
5058 defm ST2 : SIMDStSingleH<1, 0b010, 0, "st2", VecListTwoh, GPR64pi4>;
5059 defm ST2 : SIMDStSingleS<1, 0b100, 0b00, "st2", VecListTwos, GPR64pi8>;
5060 defm ST2 : SIMDStSingleD<1, 0b100, 0b01, "st2", VecListTwod, GPR64pi16>;
5061 defm ST3 : SIMDStSingleB<0, 0b001, "st3", VecListThreeb, GPR64pi3>;
5062 defm ST3 : SIMDStSingleH<0, 0b011, 0, "st3", VecListThreeh, GPR64pi6>;
5063 defm ST3 : SIMDStSingleS<0, 0b101, 0b00, "st3", VecListThrees, GPR64pi12>;
5064 defm ST3 : SIMDStSingleD<0, 0b101, 0b01, "st3", VecListThreed, GPR64pi24>;
5065 defm ST4 : SIMDStSingleB<1, 0b001, "st4", VecListFourb, GPR64pi4>;
5066 defm ST4 : SIMDStSingleH<1, 0b011, 0, "st4", VecListFourh, GPR64pi8>;
5067 defm ST4 : SIMDStSingleS<1, 0b101, 0b00, "st4", VecListFours, GPR64pi16>;
5068 defm ST4 : SIMDStSingleD<1, 0b101, 0b01, "st4", VecListFourd, GPR64pi32>;
5071 defm ST1 : SIMDLdSt1SingleAliases<"st1">;
5072 defm ST2 : SIMDLdSt2SingleAliases<"st2">;
5073 defm ST3 : SIMDLdSt3SingleAliases<"st3">;
5074 defm ST4 : SIMDLdSt4SingleAliases<"st4">;
5076 //----------------------------------------------------------------------------
5077 // Crypto extensions
5078 //----------------------------------------------------------------------------
5080 def AESErr : AESTiedInst<0b0100, "aese", int_aarch64_crypto_aese>;
5081 def AESDrr : AESTiedInst<0b0101, "aesd", int_aarch64_crypto_aesd>;
5082 def AESMCrr : AESInst< 0b0110, "aesmc", int_aarch64_crypto_aesmc>;
5083 def AESIMCrr : AESInst< 0b0111, "aesimc", int_aarch64_crypto_aesimc>;
5085 def SHA1Crrr : SHATiedInstQSV<0b000, "sha1c", int_aarch64_crypto_sha1c>;
5086 def SHA1Prrr : SHATiedInstQSV<0b001, "sha1p", int_aarch64_crypto_sha1p>;
5087 def SHA1Mrrr : SHATiedInstQSV<0b010, "sha1m", int_aarch64_crypto_sha1m>;
5088 def SHA1SU0rrr : SHATiedInstVVV<0b011, "sha1su0", int_aarch64_crypto_sha1su0>;
5089 def SHA256Hrrr : SHATiedInstQQV<0b100, "sha256h", int_aarch64_crypto_sha256h>;
5090 def SHA256H2rrr : SHATiedInstQQV<0b101, "sha256h2",int_aarch64_crypto_sha256h2>;
5091 def SHA256SU1rrr :SHATiedInstVVV<0b110, "sha256su1",int_aarch64_crypto_sha256su1>;
5093 def SHA1Hrr : SHAInstSS< 0b0000, "sha1h", int_aarch64_crypto_sha1h>;
5094 def SHA1SU1rr : SHATiedInstVV<0b0001, "sha1su1", int_aarch64_crypto_sha1su1>;
5095 def SHA256SU0rr : SHATiedInstVV<0b0010, "sha256su0",int_aarch64_crypto_sha256su0>;
5097 //----------------------------------------------------------------------------
5099 //----------------------------------------------------------------------------
5100 // FIXME: Like for X86, these should go in their own separate .td file.
5102 // Any instruction that defines a 32-bit result leaves the high half of the
5103 // register. Truncate can be lowered to EXTRACT_SUBREG. CopyFromReg may
5104 // be copying from a truncate. But any other 32-bit operation will zero-extend
5106 // FIXME: X86 also checks for CMOV here. Do we need something similar?
5107 def def32 : PatLeaf<(i32 GPR32:$src), [{
5108 return N->getOpcode() != ISD::TRUNCATE &&
5109 N->getOpcode() != TargetOpcode::EXTRACT_SUBREG &&
5110 N->getOpcode() != ISD::CopyFromReg;
5113 // In the case of a 32-bit def that is known to implicitly zero-extend,
5114 // we can use a SUBREG_TO_REG.
5115 def : Pat<(i64 (zext def32:$src)), (SUBREG_TO_REG (i64 0), GPR32:$src, sub_32)>;
5117 // For an anyext, we don't care what the high bits are, so we can perform an
5118 // INSERT_SUBREF into an IMPLICIT_DEF.
5119 def : Pat<(i64 (anyext GPR32:$src)),
5120 (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32)>;
5122 // When we need to explicitly zero-extend, we use a 32-bit MOV instruction and
5123 // then assert the extension has happened.
5124 def : Pat<(i64 (zext GPR32:$src)),
5125 (SUBREG_TO_REG (i32 0), (ORRWrs WZR, GPR32:$src, 0), sub_32)>;
5127 // To sign extend, we use a signed bitfield move instruction (SBFM) on the
5128 // containing super-reg.
5129 def : Pat<(i64 (sext GPR32:$src)),
5130 (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32), 0, 31)>;
5131 def : Pat<(i64 (sext_inreg GPR64:$src, i32)), (SBFMXri GPR64:$src, 0, 31)>;
5132 def : Pat<(i64 (sext_inreg GPR64:$src, i16)), (SBFMXri GPR64:$src, 0, 15)>;
5133 def : Pat<(i64 (sext_inreg GPR64:$src, i8)), (SBFMXri GPR64:$src, 0, 7)>;
5134 def : Pat<(i64 (sext_inreg GPR64:$src, i1)), (SBFMXri GPR64:$src, 0, 0)>;
5135 def : Pat<(i32 (sext_inreg GPR32:$src, i16)), (SBFMWri GPR32:$src, 0, 15)>;
5136 def : Pat<(i32 (sext_inreg GPR32:$src, i8)), (SBFMWri GPR32:$src, 0, 7)>;
5137 def : Pat<(i32 (sext_inreg GPR32:$src, i1)), (SBFMWri GPR32:$src, 0, 0)>;
5139 def : Pat<(shl (sext_inreg GPR32:$Rn, i8), (i64 imm0_31:$imm)),
5140 (SBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
5141 (i64 (i32shift_sext_i8 imm0_31:$imm)))>;
5142 def : Pat<(shl (sext_inreg GPR64:$Rn, i8), (i64 imm0_63:$imm)),
5143 (SBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
5144 (i64 (i64shift_sext_i8 imm0_63:$imm)))>;
5146 def : Pat<(shl (sext_inreg GPR32:$Rn, i16), (i64 imm0_31:$imm)),
5147 (SBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
5148 (i64 (i32shift_sext_i16 imm0_31:$imm)))>;
5149 def : Pat<(shl (sext_inreg GPR64:$Rn, i16), (i64 imm0_63:$imm)),
5150 (SBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
5151 (i64 (i64shift_sext_i16 imm0_63:$imm)))>;
5153 def : Pat<(shl (i64 (sext GPR32:$Rn)), (i64 imm0_63:$imm)),
5154 (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$Rn, sub_32),
5155 (i64 (i64shift_a imm0_63:$imm)),
5156 (i64 (i64shift_sext_i32 imm0_63:$imm)))>;
5158 // sra patterns have an AddedComplexity of 10, so make sure we have a higher
5159 // AddedComplexity for the following patterns since we want to match sext + sra
5160 // patterns before we attempt to match a single sra node.
5161 let AddedComplexity = 20 in {
5162 // We support all sext + sra combinations which preserve at least one bit of the
5163 // original value which is to be sign extended. E.g. we support shifts up to
5165 def : Pat<(sra (sext_inreg GPR32:$Rn, i8), (i64 imm0_7:$imm)),
5166 (SBFMWri GPR32:$Rn, (i64 imm0_7:$imm), 7)>;
5167 def : Pat<(sra (sext_inreg GPR64:$Rn, i8), (i64 imm0_7:$imm)),
5168 (SBFMXri GPR64:$Rn, (i64 imm0_7:$imm), 7)>;
5170 def : Pat<(sra (sext_inreg GPR32:$Rn, i16), (i64 imm0_15:$imm)),
5171 (SBFMWri GPR32:$Rn, (i64 imm0_15:$imm), 15)>;
5172 def : Pat<(sra (sext_inreg GPR64:$Rn, i16), (i64 imm0_15:$imm)),
5173 (SBFMXri GPR64:$Rn, (i64 imm0_15:$imm), 15)>;
5175 def : Pat<(sra (i64 (sext GPR32:$Rn)), (i64 imm0_31:$imm)),
5176 (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$Rn, sub_32),
5177 (i64 imm0_31:$imm), 31)>;
5178 } // AddedComplexity = 20
5180 // To truncate, we can simply extract from a subregister.
5181 def : Pat<(i32 (trunc GPR64sp:$src)),
5182 (i32 (EXTRACT_SUBREG GPR64sp:$src, sub_32))>;
5184 // __builtin_trap() uses the BRK instruction on AArch64.
5185 def : Pat<(trap), (BRK 1)>;
5187 // Conversions within AdvSIMD types in the same register size are free.
5188 // But because we need a consistent lane ordering, in big endian many
5189 // conversions require one or more REV instructions.
5191 // Consider a simple memory load followed by a bitconvert then a store.
5193 // v1 = BITCAST v2i32 v0 to v4i16
5196 // In big endian mode every memory access has an implicit byte swap. LDR and
5197 // STR do a 64-bit byte swap, whereas LD1/ST1 do a byte swap per lane - that
5198 // is, they treat the vector as a sequence of elements to be byte-swapped.
5199 // The two pairs of instructions are fundamentally incompatible. We've decided
5200 // to use LD1/ST1 only to simplify compiler implementation.
5202 // LD1/ST1 perform the equivalent of a sequence of LDR/STR + REV. This makes
5203 // the original code sequence:
5205 // v1 = REV v2i32 (implicit)
5206 // v2 = BITCAST v2i32 v1 to v4i16
5207 // v3 = REV v4i16 v2 (implicit)
5210 // But this is now broken - the value stored is different to the value loaded
5211 // due to lane reordering. To fix this, on every BITCAST we must perform two
5214 // v1 = REV v2i32 (implicit)
5216 // v3 = BITCAST v2i32 v2 to v4i16
5218 // v5 = REV v4i16 v4 (implicit)
5221 // This means an extra two instructions, but actually in most cases the two REV
5222 // instructions can be combined into one. For example:
5223 // (REV64_2s (REV64_4h X)) === (REV32_4h X)
5225 // There is also no 128-bit REV instruction. This must be synthesized with an
5228 // Most bitconverts require some sort of conversion. The only exceptions are:
5229 // a) Identity conversions - vNfX <-> vNiX
5230 // b) Single-lane-to-scalar - v1fX <-> fX or v1iX <-> iX
5233 // Natural vector casts (64 bit)
5234 def : Pat<(v8i8 (AArch64NvCast (v2i32 FPR64:$src))), (v8i8 FPR64:$src)>;
5235 def : Pat<(v4i16 (AArch64NvCast (v2i32 FPR64:$src))), (v4i16 FPR64:$src)>;
5236 def : Pat<(v4f16 (AArch64NvCast (v2i32 FPR64:$src))), (v4f16 FPR64:$src)>;
5237 def : Pat<(v2i32 (AArch64NvCast (v2i32 FPR64:$src))), (v2i32 FPR64:$src)>;
5238 def : Pat<(v2f32 (AArch64NvCast (v2i32 FPR64:$src))), (v2f32 FPR64:$src)>;
5239 def : Pat<(v1i64 (AArch64NvCast (v2i32 FPR64:$src))), (v1i64 FPR64:$src)>;
5241 def : Pat<(v8i8 (AArch64NvCast (v4i16 FPR64:$src))), (v8i8 FPR64:$src)>;
5242 def : Pat<(v4i16 (AArch64NvCast (v4i16 FPR64:$src))), (v4i16 FPR64:$src)>;
5243 def : Pat<(v4f16 (AArch64NvCast (v4i16 FPR64:$src))), (v4f16 FPR64:$src)>;
5244 def : Pat<(v2i32 (AArch64NvCast (v4i16 FPR64:$src))), (v2i32 FPR64:$src)>;
5245 def : Pat<(v1i64 (AArch64NvCast (v4i16 FPR64:$src))), (v1i64 FPR64:$src)>;
5247 def : Pat<(v8i8 (AArch64NvCast (v8i8 FPR64:$src))), (v8i8 FPR64:$src)>;
5248 def : Pat<(v4i16 (AArch64NvCast (v8i8 FPR64:$src))), (v4i16 FPR64:$src)>;
5249 def : Pat<(v4f16 (AArch64NvCast (v8i8 FPR64:$src))), (v4f16 FPR64:$src)>;
5250 def : Pat<(v2i32 (AArch64NvCast (v8i8 FPR64:$src))), (v2i32 FPR64:$src)>;
5251 def : Pat<(v1i64 (AArch64NvCast (v8i8 FPR64:$src))), (v1i64 FPR64:$src)>;
5253 def : Pat<(v8i8 (AArch64NvCast (f64 FPR64:$src))), (v8i8 FPR64:$src)>;
5254 def : Pat<(v4i16 (AArch64NvCast (f64 FPR64:$src))), (v4i16 FPR64:$src)>;
5255 def : Pat<(v4f16 (AArch64NvCast (f64 FPR64:$src))), (v4f16 FPR64:$src)>;
5256 def : Pat<(v2i32 (AArch64NvCast (f64 FPR64:$src))), (v2i32 FPR64:$src)>;
5257 def : Pat<(v2f32 (AArch64NvCast (f64 FPR64:$src))), (v2f32 FPR64:$src)>;
5258 def : Pat<(v1i64 (AArch64NvCast (f64 FPR64:$src))), (v1i64 FPR64:$src)>;
5259 def : Pat<(v1f64 (AArch64NvCast (f64 FPR64:$src))), (v1f64 FPR64:$src)>;
5261 def : Pat<(v8i8 (AArch64NvCast (v2f32 FPR64:$src))), (v8i8 FPR64:$src)>;
5262 def : Pat<(v4i16 (AArch64NvCast (v2f32 FPR64:$src))), (v4i16 FPR64:$src)>;
5263 def : Pat<(v2i32 (AArch64NvCast (v2f32 FPR64:$src))), (v2i32 FPR64:$src)>;
5264 def : Pat<(v2f32 (AArch64NvCast (v2f32 FPR64:$src))), (v2f32 FPR64:$src)>;
5265 def : Pat<(v1i64 (AArch64NvCast (v2f32 FPR64:$src))), (v1i64 FPR64:$src)>;
5267 // Natural vector casts (128 bit)
5268 def : Pat<(v16i8 (AArch64NvCast (v4i32 FPR128:$src))), (v16i8 FPR128:$src)>;
5269 def : Pat<(v8i16 (AArch64NvCast (v4i32 FPR128:$src))), (v8i16 FPR128:$src)>;
5270 def : Pat<(v8f16 (AArch64NvCast (v4i32 FPR128:$src))), (v8f16 FPR128:$src)>;
5271 def : Pat<(v4i32 (AArch64NvCast (v4i32 FPR128:$src))), (v4i32 FPR128:$src)>;
5272 def : Pat<(v4f32 (AArch64NvCast (v4i32 FPR128:$src))), (v4f32 FPR128:$src)>;
5273 def : Pat<(v2i64 (AArch64NvCast (v4i32 FPR128:$src))), (v2i64 FPR128:$src)>;
5274 def : Pat<(v2f64 (AArch64NvCast (v4i32 FPR128:$src))), (v2f64 FPR128:$src)>;
5276 def : Pat<(v16i8 (AArch64NvCast (v8i16 FPR128:$src))), (v16i8 FPR128:$src)>;
5277 def : Pat<(v8i16 (AArch64NvCast (v8i16 FPR128:$src))), (v8i16 FPR128:$src)>;
5278 def : Pat<(v8f16 (AArch64NvCast (v8i16 FPR128:$src))), (v8f16 FPR128:$src)>;
5279 def : Pat<(v4i32 (AArch64NvCast (v8i16 FPR128:$src))), (v4i32 FPR128:$src)>;
5280 def : Pat<(v2i64 (AArch64NvCast (v8i16 FPR128:$src))), (v2i64 FPR128:$src)>;
5281 def : Pat<(v4f32 (AArch64NvCast (v8i16 FPR128:$src))), (v4f32 FPR128:$src)>;
5282 def : Pat<(v2f64 (AArch64NvCast (v8i16 FPR128:$src))), (v2f64 FPR128:$src)>;
5284 def : Pat<(v16i8 (AArch64NvCast (v16i8 FPR128:$src))), (v16i8 FPR128:$src)>;
5285 def : Pat<(v8i16 (AArch64NvCast (v16i8 FPR128:$src))), (v8i16 FPR128:$src)>;
5286 def : Pat<(v8f16 (AArch64NvCast (v16i8 FPR128:$src))), (v8f16 FPR128:$src)>;
5287 def : Pat<(v4i32 (AArch64NvCast (v16i8 FPR128:$src))), (v4i32 FPR128:$src)>;
5288 def : Pat<(v2i64 (AArch64NvCast (v16i8 FPR128:$src))), (v2i64 FPR128:$src)>;
5289 def : Pat<(v4f32 (AArch64NvCast (v16i8 FPR128:$src))), (v4f32 FPR128:$src)>;
5290 def : Pat<(v2f64 (AArch64NvCast (v16i8 FPR128:$src))), (v2f64 FPR128:$src)>;
5292 def : Pat<(v16i8 (AArch64NvCast (v2i64 FPR128:$src))), (v16i8 FPR128:$src)>;
5293 def : Pat<(v8i16 (AArch64NvCast (v2i64 FPR128:$src))), (v8i16 FPR128:$src)>;
5294 def : Pat<(v8f16 (AArch64NvCast (v2i64 FPR128:$src))), (v8f16 FPR128:$src)>;
5295 def : Pat<(v4i32 (AArch64NvCast (v2i64 FPR128:$src))), (v4i32 FPR128:$src)>;
5296 def : Pat<(v2i64 (AArch64NvCast (v2i64 FPR128:$src))), (v2i64 FPR128:$src)>;
5297 def : Pat<(v4f32 (AArch64NvCast (v2i64 FPR128:$src))), (v4f32 FPR128:$src)>;
5298 def : Pat<(v2f64 (AArch64NvCast (v2i64 FPR128:$src))), (v2f64 FPR128:$src)>;
5300 def : Pat<(v16i8 (AArch64NvCast (v4f32 FPR128:$src))), (v16i8 FPR128:$src)>;
5301 def : Pat<(v8i16 (AArch64NvCast (v4f32 FPR128:$src))), (v8i16 FPR128:$src)>;
5302 def : Pat<(v4i32 (AArch64NvCast (v4f32 FPR128:$src))), (v4i32 FPR128:$src)>;
5303 def : Pat<(v4f32 (AArch64NvCast (v4f32 FPR128:$src))), (v4f32 FPR128:$src)>;
5304 def : Pat<(v2i64 (AArch64NvCast (v4f32 FPR128:$src))), (v2i64 FPR128:$src)>;
5305 def : Pat<(v8f16 (AArch64NvCast (v4f32 FPR128:$src))), (v8f16 FPR128:$src)>;
5306 def : Pat<(v2f64 (AArch64NvCast (v4f32 FPR128:$src))), (v2f64 FPR128:$src)>;
5308 def : Pat<(v16i8 (AArch64NvCast (v2f64 FPR128:$src))), (v16i8 FPR128:$src)>;
5309 def : Pat<(v8i16 (AArch64NvCast (v2f64 FPR128:$src))), (v8i16 FPR128:$src)>;
5310 def : Pat<(v4i32 (AArch64NvCast (v2f64 FPR128:$src))), (v4i32 FPR128:$src)>;
5311 def : Pat<(v2i64 (AArch64NvCast (v2f64 FPR128:$src))), (v2i64 FPR128:$src)>;
5312 def : Pat<(v2f64 (AArch64NvCast (v2f64 FPR128:$src))), (v2f64 FPR128:$src)>;
5313 def : Pat<(v8f16 (AArch64NvCast (v2f64 FPR128:$src))), (v8f16 FPR128:$src)>;
5314 def : Pat<(v4f32 (AArch64NvCast (v2f64 FPR128:$src))), (v4f32 FPR128:$src)>;
5316 let Predicates = [IsLE] in {
5317 def : Pat<(v8i8 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5318 def : Pat<(v4i16 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5319 def : Pat<(v2i32 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5320 def : Pat<(v4f16 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5321 def : Pat<(v2f32 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5323 def : Pat<(i64 (bitconvert (v8i8 V64:$Vn))),
5324 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5325 def : Pat<(i64 (bitconvert (v4i16 V64:$Vn))),
5326 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5327 def : Pat<(i64 (bitconvert (v2i32 V64:$Vn))),
5328 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5329 def : Pat<(i64 (bitconvert (v4f16 V64:$Vn))),
5330 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5331 def : Pat<(i64 (bitconvert (v2f32 V64:$Vn))),
5332 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5333 def : Pat<(i64 (bitconvert (v1f64 V64:$Vn))),
5334 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5336 let Predicates = [IsBE] in {
5337 def : Pat<(v8i8 (bitconvert GPR64:$Xn)),
5338 (REV64v8i8 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
5339 def : Pat<(v4i16 (bitconvert GPR64:$Xn)),
5340 (REV64v4i16 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
5341 def : Pat<(v2i32 (bitconvert GPR64:$Xn)),
5342 (REV64v2i32 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
5343 def : Pat<(v4f16 (bitconvert GPR64:$Xn)),
5344 (REV64v4i16 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
5345 def : Pat<(v2f32 (bitconvert GPR64:$Xn)),
5346 (REV64v2i32 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
5348 def : Pat<(i64 (bitconvert (v8i8 V64:$Vn))),
5349 (REV64v8i8 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
5350 def : Pat<(i64 (bitconvert (v4i16 V64:$Vn))),
5351 (REV64v4i16 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
5352 def : Pat<(i64 (bitconvert (v2i32 V64:$Vn))),
5353 (REV64v2i32 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
5354 def : Pat<(i64 (bitconvert (v4f16 V64:$Vn))),
5355 (REV64v4i16 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
5356 def : Pat<(i64 (bitconvert (v2f32 V64:$Vn))),
5357 (REV64v2i32 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
5359 def : Pat<(v1i64 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5360 def : Pat<(v1f64 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5361 def : Pat<(i64 (bitconvert (v1i64 V64:$Vn))),
5362 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5363 def : Pat<(v1i64 (scalar_to_vector GPR64:$Xn)),
5364 (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5365 def : Pat<(v1f64 (scalar_to_vector GPR64:$Xn)),
5366 (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5367 def : Pat<(v1f64 (scalar_to_vector (f64 FPR64:$Xn))), (v1f64 FPR64:$Xn)>;
5369 def : Pat<(f32 (bitconvert (i32 GPR32:$Xn))),
5370 (COPY_TO_REGCLASS GPR32:$Xn, FPR32)>;
5371 def : Pat<(i32 (bitconvert (f32 FPR32:$Xn))),
5372 (COPY_TO_REGCLASS FPR32:$Xn, GPR32)>;
5373 def : Pat<(f64 (bitconvert (i64 GPR64:$Xn))),
5374 (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
5375 def : Pat<(i64 (bitconvert (f64 FPR64:$Xn))),
5376 (COPY_TO_REGCLASS FPR64:$Xn, GPR64)>;
5377 def : Pat<(i64 (bitconvert (v1f64 V64:$Vn))),
5378 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
5380 let Predicates = [IsLE] in {
5381 def : Pat<(v1i64 (bitconvert (v2i32 FPR64:$src))), (v1i64 FPR64:$src)>;
5382 def : Pat<(v1i64 (bitconvert (v4i16 FPR64:$src))), (v1i64 FPR64:$src)>;
5383 def : Pat<(v1i64 (bitconvert (v8i8 FPR64:$src))), (v1i64 FPR64:$src)>;
5384 def : Pat<(v1i64 (bitconvert (v4f16 FPR64:$src))), (v1i64 FPR64:$src)>;
5385 def : Pat<(v1i64 (bitconvert (v2f32 FPR64:$src))), (v1i64 FPR64:$src)>;
5387 let Predicates = [IsBE] in {
5388 def : Pat<(v1i64 (bitconvert (v2i32 FPR64:$src))),
5389 (v1i64 (REV64v2i32 FPR64:$src))>;
5390 def : Pat<(v1i64 (bitconvert (v4i16 FPR64:$src))),
5391 (v1i64 (REV64v4i16 FPR64:$src))>;
5392 def : Pat<(v1i64 (bitconvert (v8i8 FPR64:$src))),
5393 (v1i64 (REV64v8i8 FPR64:$src))>;
5394 def : Pat<(v1i64 (bitconvert (v4f16 FPR64:$src))),
5395 (v1i64 (REV64v4i16 FPR64:$src))>;
5396 def : Pat<(v1i64 (bitconvert (v2f32 FPR64:$src))),
5397 (v1i64 (REV64v2i32 FPR64:$src))>;
5399 def : Pat<(v1i64 (bitconvert (v1f64 FPR64:$src))), (v1i64 FPR64:$src)>;
5400 def : Pat<(v1i64 (bitconvert (f64 FPR64:$src))), (v1i64 FPR64:$src)>;
5402 let Predicates = [IsLE] in {
5403 def : Pat<(v2i32 (bitconvert (v1i64 FPR64:$src))), (v2i32 FPR64:$src)>;
5404 def : Pat<(v2i32 (bitconvert (v4i16 FPR64:$src))), (v2i32 FPR64:$src)>;
5405 def : Pat<(v2i32 (bitconvert (v8i8 FPR64:$src))), (v2i32 FPR64:$src)>;
5406 def : Pat<(v2i32 (bitconvert (f64 FPR64:$src))), (v2i32 FPR64:$src)>;
5407 def : Pat<(v2i32 (bitconvert (v1f64 FPR64:$src))), (v2i32 FPR64:$src)>;
5408 def : Pat<(v2i32 (bitconvert (v4f16 FPR64:$src))), (v2i32 FPR64:$src)>;
5410 let Predicates = [IsBE] in {
5411 def : Pat<(v2i32 (bitconvert (v1i64 FPR64:$src))),
5412 (v2i32 (REV64v2i32 FPR64:$src))>;
5413 def : Pat<(v2i32 (bitconvert (v4i16 FPR64:$src))),
5414 (v2i32 (REV32v4i16 FPR64:$src))>;
5415 def : Pat<(v2i32 (bitconvert (v8i8 FPR64:$src))),
5416 (v2i32 (REV32v8i8 FPR64:$src))>;
5417 def : Pat<(v2i32 (bitconvert (f64 FPR64:$src))),
5418 (v2i32 (REV64v2i32 FPR64:$src))>;
5419 def : Pat<(v2i32 (bitconvert (v1f64 FPR64:$src))),
5420 (v2i32 (REV64v2i32 FPR64:$src))>;
5421 def : Pat<(v2i32 (bitconvert (v4f16 FPR64:$src))),
5422 (v2i32 (REV64v4i16 FPR64:$src))>;
5424 def : Pat<(v2i32 (bitconvert (v2f32 FPR64:$src))), (v2i32 FPR64:$src)>;
5426 let Predicates = [IsLE] in {
5427 def : Pat<(v4i16 (bitconvert (v1i64 FPR64:$src))), (v4i16 FPR64:$src)>;
5428 def : Pat<(v4i16 (bitconvert (v2i32 FPR64:$src))), (v4i16 FPR64:$src)>;
5429 def : Pat<(v4i16 (bitconvert (v8i8 FPR64:$src))), (v4i16 FPR64:$src)>;
5430 def : Pat<(v4i16 (bitconvert (f64 FPR64:$src))), (v4i16 FPR64:$src)>;
5431 def : Pat<(v4i16 (bitconvert (v4f16 FPR64:$src))), (v4i16 FPR64:$src)>;
5432 def : Pat<(v4i16 (bitconvert (v2f32 FPR64:$src))), (v4i16 FPR64:$src)>;
5433 def : Pat<(v4i16 (bitconvert (v1f64 FPR64:$src))), (v4i16 FPR64:$src)>;
5435 let Predicates = [IsBE] in {
5436 def : Pat<(v4i16 (bitconvert (v1i64 FPR64:$src))),
5437 (v4i16 (REV64v4i16 FPR64:$src))>;
5438 def : Pat<(v4i16 (bitconvert (v2i32 FPR64:$src))),
5439 (v4i16 (REV32v4i16 FPR64:$src))>;
5440 def : Pat<(v4i16 (bitconvert (v8i8 FPR64:$src))),
5441 (v4i16 (REV16v8i8 FPR64:$src))>;
5442 def : Pat<(v4i16 (bitconvert (f64 FPR64:$src))),
5443 (v4i16 (REV64v4i16 FPR64:$src))>;
5444 def : Pat<(v4i16 (bitconvert (v4f16 FPR64:$src))),
5445 (v4i16 (REV32v4i16 FPR64:$src))>;
5446 def : Pat<(v4i16 (bitconvert (v2f32 FPR64:$src))),
5447 (v4i16 (REV32v4i16 FPR64:$src))>;
5448 def : Pat<(v4i16 (bitconvert (v1f64 FPR64:$src))),
5449 (v4i16 (REV64v4i16 FPR64:$src))>;
5452 let Predicates = [IsLE] in {
5453 def : Pat<(v4f16 (bitconvert (v1i64 FPR64:$src))), (v4f16 FPR64:$src)>;
5454 def : Pat<(v4f16 (bitconvert (v2i32 FPR64:$src))), (v4f16 FPR64:$src)>;
5455 def : Pat<(v4f16 (bitconvert (v4i16 FPR64:$src))), (v4f16 FPR64:$src)>;
5456 def : Pat<(v4f16 (bitconvert (v8i8 FPR64:$src))), (v4f16 FPR64:$src)>;
5457 def : Pat<(v4f16 (bitconvert (f64 FPR64:$src))), (v4f16 FPR64:$src)>;
5458 def : Pat<(v4f16 (bitconvert (v2f32 FPR64:$src))), (v4f16 FPR64:$src)>;
5459 def : Pat<(v4f16 (bitconvert (v1f64 FPR64:$src))), (v4f16 FPR64:$src)>;
5461 let Predicates = [IsBE] in {
5462 def : Pat<(v4f16 (bitconvert (v1i64 FPR64:$src))),
5463 (v4f16 (REV64v4i16 FPR64:$src))>;
5464 def : Pat<(v4f16 (bitconvert (v2i32 FPR64:$src))),
5465 (v4f16 (REV64v4i16 FPR64:$src))>;
5466 def : Pat<(v4f16 (bitconvert (v4i16 FPR64:$src))),
5467 (v4f16 (REV64v4i16 FPR64:$src))>;
5468 def : Pat<(v4f16 (bitconvert (v8i8 FPR64:$src))),
5469 (v4f16 (REV16v8i8 FPR64:$src))>;
5470 def : Pat<(v4f16 (bitconvert (f64 FPR64:$src))),
5471 (v4f16 (REV64v4i16 FPR64:$src))>;
5472 def : Pat<(v4f16 (bitconvert (v2f32 FPR64:$src))),
5473 (v4f16 (REV64v4i16 FPR64:$src))>;
5474 def : Pat<(v4f16 (bitconvert (v1f64 FPR64:$src))),
5475 (v4f16 (REV64v4i16 FPR64:$src))>;
5480 let Predicates = [IsLE] in {
5481 def : Pat<(v8i8 (bitconvert (v1i64 FPR64:$src))), (v8i8 FPR64:$src)>;
5482 def : Pat<(v8i8 (bitconvert (v2i32 FPR64:$src))), (v8i8 FPR64:$src)>;
5483 def : Pat<(v8i8 (bitconvert (v4i16 FPR64:$src))), (v8i8 FPR64:$src)>;
5484 def : Pat<(v8i8 (bitconvert (f64 FPR64:$src))), (v8i8 FPR64:$src)>;
5485 def : Pat<(v8i8 (bitconvert (v2f32 FPR64:$src))), (v8i8 FPR64:$src)>;
5486 def : Pat<(v8i8 (bitconvert (v1f64 FPR64:$src))), (v8i8 FPR64:$src)>;
5487 def : Pat<(v8i8 (bitconvert (v4f16 FPR64:$src))), (v8i8 FPR64:$src)>;
5489 let Predicates = [IsBE] in {
5490 def : Pat<(v8i8 (bitconvert (v1i64 FPR64:$src))),
5491 (v8i8 (REV64v8i8 FPR64:$src))>;
5492 def : Pat<(v8i8 (bitconvert (v2i32 FPR64:$src))),
5493 (v8i8 (REV32v8i8 FPR64:$src))>;
5494 def : Pat<(v8i8 (bitconvert (v4i16 FPR64:$src))),
5495 (v8i8 (REV16v8i8 FPR64:$src))>;
5496 def : Pat<(v8i8 (bitconvert (f64 FPR64:$src))),
5497 (v8i8 (REV64v8i8 FPR64:$src))>;
5498 def : Pat<(v8i8 (bitconvert (v2f32 FPR64:$src))),
5499 (v8i8 (REV32v8i8 FPR64:$src))>;
5500 def : Pat<(v8i8 (bitconvert (v1f64 FPR64:$src))),
5501 (v8i8 (REV64v8i8 FPR64:$src))>;
5502 def : Pat<(v8i8 (bitconvert (v4f16 FPR64:$src))),
5503 (v8i8 (REV16v8i8 FPR64:$src))>;
5506 let Predicates = [IsLE] in {
5507 def : Pat<(f64 (bitconvert (v2i32 FPR64:$src))), (f64 FPR64:$src)>;
5508 def : Pat<(f64 (bitconvert (v4i16 FPR64:$src))), (f64 FPR64:$src)>;
5509 def : Pat<(f64 (bitconvert (v2f32 FPR64:$src))), (f64 FPR64:$src)>;
5510 def : Pat<(f64 (bitconvert (v8i8 FPR64:$src))), (f64 FPR64:$src)>;
5511 def : Pat<(f64 (bitconvert (v4f16 FPR64:$src))), (f64 FPR64:$src)>;
5513 let Predicates = [IsBE] in {
5514 def : Pat<(f64 (bitconvert (v2i32 FPR64:$src))),
5515 (f64 (REV64v2i32 FPR64:$src))>;
5516 def : Pat<(f64 (bitconvert (v4i16 FPR64:$src))),
5517 (f64 (REV64v4i16 FPR64:$src))>;
5518 def : Pat<(f64 (bitconvert (v2f32 FPR64:$src))),
5519 (f64 (REV64v2i32 FPR64:$src))>;
5520 def : Pat<(f64 (bitconvert (v8i8 FPR64:$src))),
5521 (f64 (REV64v8i8 FPR64:$src))>;
5522 def : Pat<(f64 (bitconvert (v4f16 FPR64:$src))),
5523 (f64 (REV64v4i16 FPR64:$src))>;
5525 def : Pat<(f64 (bitconvert (v1i64 FPR64:$src))), (f64 FPR64:$src)>;
5526 def : Pat<(f64 (bitconvert (v1f64 FPR64:$src))), (f64 FPR64:$src)>;
5528 let Predicates = [IsLE] in {
5529 def : Pat<(v1f64 (bitconvert (v2i32 FPR64:$src))), (v1f64 FPR64:$src)>;
5530 def : Pat<(v1f64 (bitconvert (v4i16 FPR64:$src))), (v1f64 FPR64:$src)>;
5531 def : Pat<(v1f64 (bitconvert (v8i8 FPR64:$src))), (v1f64 FPR64:$src)>;
5532 def : Pat<(v1f64 (bitconvert (v2f32 FPR64:$src))), (v1f64 FPR64:$src)>;
5533 def : Pat<(v1f64 (bitconvert (v4f16 FPR64:$src))), (v1f64 FPR64:$src)>;
5535 let Predicates = [IsBE] in {
5536 def : Pat<(v1f64 (bitconvert (v2i32 FPR64:$src))),
5537 (v1f64 (REV64v2i32 FPR64:$src))>;
5538 def : Pat<(v1f64 (bitconvert (v4i16 FPR64:$src))),
5539 (v1f64 (REV64v4i16 FPR64:$src))>;
5540 def : Pat<(v1f64 (bitconvert (v8i8 FPR64:$src))),
5541 (v1f64 (REV64v8i8 FPR64:$src))>;
5542 def : Pat<(v1f64 (bitconvert (v2f32 FPR64:$src))),
5543 (v1f64 (REV64v2i32 FPR64:$src))>;
5544 def : Pat<(v1f64 (bitconvert (v4f16 FPR64:$src))),
5545 (v1f64 (REV64v4i16 FPR64:$src))>;
5547 def : Pat<(v1f64 (bitconvert (v1i64 FPR64:$src))), (v1f64 FPR64:$src)>;
5548 def : Pat<(v1f64 (bitconvert (f64 FPR64:$src))), (v1f64 FPR64:$src)>;
5550 let Predicates = [IsLE] in {
5551 def : Pat<(v2f32 (bitconvert (v1i64 FPR64:$src))), (v2f32 FPR64:$src)>;
5552 def : Pat<(v2f32 (bitconvert (v4i16 FPR64:$src))), (v2f32 FPR64:$src)>;
5553 def : Pat<(v2f32 (bitconvert (v8i8 FPR64:$src))), (v2f32 FPR64:$src)>;
5554 def : Pat<(v2f32 (bitconvert (v1f64 FPR64:$src))), (v2f32 FPR64:$src)>;
5555 def : Pat<(v2f32 (bitconvert (f64 FPR64:$src))), (v2f32 FPR64:$src)>;
5556 def : Pat<(v2f32 (bitconvert (v4f16 FPR64:$src))), (v2f32 FPR64:$src)>;
5558 let Predicates = [IsBE] in {
5559 def : Pat<(v2f32 (bitconvert (v1i64 FPR64:$src))),
5560 (v2f32 (REV64v2i32 FPR64:$src))>;
5561 def : Pat<(v2f32 (bitconvert (v4i16 FPR64:$src))),
5562 (v2f32 (REV32v4i16 FPR64:$src))>;
5563 def : Pat<(v2f32 (bitconvert (v8i8 FPR64:$src))),
5564 (v2f32 (REV32v8i8 FPR64:$src))>;
5565 def : Pat<(v2f32 (bitconvert (v1f64 FPR64:$src))),
5566 (v2f32 (REV64v2i32 FPR64:$src))>;
5567 def : Pat<(v2f32 (bitconvert (f64 FPR64:$src))),
5568 (v2f32 (REV64v2i32 FPR64:$src))>;
5569 def : Pat<(v2f32 (bitconvert (v4f16 FPR64:$src))),
5570 (v2f32 (REV64v4i16 FPR64:$src))>;
5572 def : Pat<(v2f32 (bitconvert (v2i32 FPR64:$src))), (v2f32 FPR64:$src)>;
5574 let Predicates = [IsLE] in {
5575 def : Pat<(f128 (bitconvert (v2i64 FPR128:$src))), (f128 FPR128:$src)>;
5576 def : Pat<(f128 (bitconvert (v4i32 FPR128:$src))), (f128 FPR128:$src)>;
5577 def : Pat<(f128 (bitconvert (v8i16 FPR128:$src))), (f128 FPR128:$src)>;
5578 def : Pat<(f128 (bitconvert (v2f64 FPR128:$src))), (f128 FPR128:$src)>;
5579 def : Pat<(f128 (bitconvert (v4f32 FPR128:$src))), (f128 FPR128:$src)>;
5580 def : Pat<(f128 (bitconvert (v8f16 FPR128:$src))), (f128 FPR128:$src)>;
5581 def : Pat<(f128 (bitconvert (v16i8 FPR128:$src))), (f128 FPR128:$src)>;
5583 let Predicates = [IsBE] in {
5584 def : Pat<(f128 (bitconvert (v2i64 FPR128:$src))),
5585 (f128 (EXTv16i8 FPR128:$src, FPR128:$src, (i32 8)))>;
5586 def : Pat<(f128 (bitconvert (v4i32 FPR128:$src))),
5587 (f128 (EXTv16i8 (REV64v4i32 FPR128:$src),
5588 (REV64v4i32 FPR128:$src), (i32 8)))>;
5589 def : Pat<(f128 (bitconvert (v8i16 FPR128:$src))),
5590 (f128 (EXTv16i8 (REV64v8i16 FPR128:$src),
5591 (REV64v8i16 FPR128:$src), (i32 8)))>;
5592 def : Pat<(f128 (bitconvert (v8f16 FPR128:$src))),
5593 (f128 (EXTv16i8 (REV64v8i16 FPR128:$src),
5594 (REV64v8i16 FPR128:$src), (i32 8)))>;
5595 def : Pat<(f128 (bitconvert (v2f64 FPR128:$src))),
5596 (f128 (EXTv16i8 FPR128:$src, FPR128:$src, (i32 8)))>;
5597 def : Pat<(f128 (bitconvert (v4f32 FPR128:$src))),
5598 (f128 (EXTv16i8 (REV64v4i32 FPR128:$src),
5599 (REV64v4i32 FPR128:$src), (i32 8)))>;
5600 def : Pat<(f128 (bitconvert (v16i8 FPR128:$src))),
5601 (f128 (EXTv16i8 (REV64v16i8 FPR128:$src),
5602 (REV64v16i8 FPR128:$src), (i32 8)))>;
5605 let Predicates = [IsLE] in {
5606 def : Pat<(v2f64 (bitconvert (f128 FPR128:$src))), (v2f64 FPR128:$src)>;
5607 def : Pat<(v2f64 (bitconvert (v4i32 FPR128:$src))), (v2f64 FPR128:$src)>;
5608 def : Pat<(v2f64 (bitconvert (v8i16 FPR128:$src))), (v2f64 FPR128:$src)>;
5609 def : Pat<(v2f64 (bitconvert (v8f16 FPR128:$src))), (v2f64 FPR128:$src)>;
5610 def : Pat<(v2f64 (bitconvert (v16i8 FPR128:$src))), (v2f64 FPR128:$src)>;
5611 def : Pat<(v2f64 (bitconvert (v4f32 FPR128:$src))), (v2f64 FPR128:$src)>;
5613 let Predicates = [IsBE] in {
5614 def : Pat<(v2f64 (bitconvert (f128 FPR128:$src))),
5615 (v2f64 (EXTv16i8 FPR128:$src,
5616 FPR128:$src, (i32 8)))>;
5617 def : Pat<(v2f64 (bitconvert (v4i32 FPR128:$src))),
5618 (v2f64 (REV64v4i32 FPR128:$src))>;
5619 def : Pat<(v2f64 (bitconvert (v8i16 FPR128:$src))),
5620 (v2f64 (REV64v8i16 FPR128:$src))>;
5621 def : Pat<(v2f64 (bitconvert (v8f16 FPR128:$src))),
5622 (v2f64 (REV64v8i16 FPR128:$src))>;
5623 def : Pat<(v2f64 (bitconvert (v16i8 FPR128:$src))),
5624 (v2f64 (REV64v16i8 FPR128:$src))>;
5625 def : Pat<(v2f64 (bitconvert (v4f32 FPR128:$src))),
5626 (v2f64 (REV64v4i32 FPR128:$src))>;
5628 def : Pat<(v2f64 (bitconvert (v2i64 FPR128:$src))), (v2f64 FPR128:$src)>;
5630 let Predicates = [IsLE] in {
5631 def : Pat<(v4f32 (bitconvert (f128 FPR128:$src))), (v4f32 FPR128:$src)>;
5632 def : Pat<(v4f32 (bitconvert (v8i16 FPR128:$src))), (v4f32 FPR128:$src)>;
5633 def : Pat<(v4f32 (bitconvert (v8f16 FPR128:$src))), (v4f32 FPR128:$src)>;
5634 def : Pat<(v4f32 (bitconvert (v16i8 FPR128:$src))), (v4f32 FPR128:$src)>;
5635 def : Pat<(v4f32 (bitconvert (v2i64 FPR128:$src))), (v4f32 FPR128:$src)>;
5636 def : Pat<(v4f32 (bitconvert (v2f64 FPR128:$src))), (v4f32 FPR128:$src)>;
5638 let Predicates = [IsBE] in {
5639 def : Pat<(v4f32 (bitconvert (f128 FPR128:$src))),
5640 (v4f32 (EXTv16i8 (REV64v4i32 FPR128:$src),
5641 (REV64v4i32 FPR128:$src), (i32 8)))>;
5642 def : Pat<(v4f32 (bitconvert (v8i16 FPR128:$src))),
5643 (v4f32 (REV32v8i16 FPR128:$src))>;
5644 def : Pat<(v4f32 (bitconvert (v8f16 FPR128:$src))),
5645 (v4f32 (REV32v8i16 FPR128:$src))>;
5646 def : Pat<(v4f32 (bitconvert (v16i8 FPR128:$src))),
5647 (v4f32 (REV32v16i8 FPR128:$src))>;
5648 def : Pat<(v4f32 (bitconvert (v2i64 FPR128:$src))),
5649 (v4f32 (REV64v4i32 FPR128:$src))>;
5650 def : Pat<(v4f32 (bitconvert (v2f64 FPR128:$src))),
5651 (v4f32 (REV64v4i32 FPR128:$src))>;
5653 def : Pat<(v4f32 (bitconvert (v4i32 FPR128:$src))), (v4f32 FPR128:$src)>;
5655 let Predicates = [IsLE] in {
5656 def : Pat<(v2i64 (bitconvert (f128 FPR128:$src))), (v2i64 FPR128:$src)>;
5657 def : Pat<(v2i64 (bitconvert (v4i32 FPR128:$src))), (v2i64 FPR128:$src)>;
5658 def : Pat<(v2i64 (bitconvert (v8i16 FPR128:$src))), (v2i64 FPR128:$src)>;
5659 def : Pat<(v2i64 (bitconvert (v16i8 FPR128:$src))), (v2i64 FPR128:$src)>;
5660 def : Pat<(v2i64 (bitconvert (v4f32 FPR128:$src))), (v2i64 FPR128:$src)>;
5661 def : Pat<(v2i64 (bitconvert (v8f16 FPR128:$src))), (v2i64 FPR128:$src)>;
5663 let Predicates = [IsBE] in {
5664 def : Pat<(v2i64 (bitconvert (f128 FPR128:$src))),
5665 (v2i64 (EXTv16i8 FPR128:$src,
5666 FPR128:$src, (i32 8)))>;
5667 def : Pat<(v2i64 (bitconvert (v4i32 FPR128:$src))),
5668 (v2i64 (REV64v4i32 FPR128:$src))>;
5669 def : Pat<(v2i64 (bitconvert (v8i16 FPR128:$src))),
5670 (v2i64 (REV64v8i16 FPR128:$src))>;
5671 def : Pat<(v2i64 (bitconvert (v16i8 FPR128:$src))),
5672 (v2i64 (REV64v16i8 FPR128:$src))>;
5673 def : Pat<(v2i64 (bitconvert (v4f32 FPR128:$src))),
5674 (v2i64 (REV64v4i32 FPR128:$src))>;
5675 def : Pat<(v2i64 (bitconvert (v8f16 FPR128:$src))),
5676 (v2i64 (REV64v8i16 FPR128:$src))>;
5678 def : Pat<(v2i64 (bitconvert (v2f64 FPR128:$src))), (v2i64 FPR128:$src)>;
5680 let Predicates = [IsLE] in {
5681 def : Pat<(v4i32 (bitconvert (f128 FPR128:$src))), (v4i32 FPR128:$src)>;
5682 def : Pat<(v4i32 (bitconvert (v2i64 FPR128:$src))), (v4i32 FPR128:$src)>;
5683 def : Pat<(v4i32 (bitconvert (v8i16 FPR128:$src))), (v4i32 FPR128:$src)>;
5684 def : Pat<(v4i32 (bitconvert (v16i8 FPR128:$src))), (v4i32 FPR128:$src)>;
5685 def : Pat<(v4i32 (bitconvert (v2f64 FPR128:$src))), (v4i32 FPR128:$src)>;
5686 def : Pat<(v4i32 (bitconvert (v8f16 FPR128:$src))), (v4i32 FPR128:$src)>;
5688 let Predicates = [IsBE] in {
5689 def : Pat<(v4i32 (bitconvert (f128 FPR128:$src))),
5690 (v4i32 (EXTv16i8 (REV64v4i32 FPR128:$src),
5691 (REV64v4i32 FPR128:$src),
5693 def : Pat<(v4i32 (bitconvert (v2i64 FPR128:$src))),
5694 (v4i32 (REV64v4i32 FPR128:$src))>;
5695 def : Pat<(v4i32 (bitconvert (v8i16 FPR128:$src))),
5696 (v4i32 (REV32v8i16 FPR128:$src))>;
5697 def : Pat<(v4i32 (bitconvert (v16i8 FPR128:$src))),
5698 (v4i32 (REV32v16i8 FPR128:$src))>;
5699 def : Pat<(v4i32 (bitconvert (v2f64 FPR128:$src))),
5700 (v4i32 (REV64v4i32 FPR128:$src))>;
5701 def : Pat<(v4i32 (bitconvert (v8f16 FPR128:$src))),
5702 (v4i32 (REV32v8i16 FPR128:$src))>;
5704 def : Pat<(v4i32 (bitconvert (v4f32 FPR128:$src))), (v4i32 FPR128:$src)>;
5706 let Predicates = [IsLE] in {
5707 def : Pat<(v8i16 (bitconvert (f128 FPR128:$src))), (v8i16 FPR128:$src)>;
5708 def : Pat<(v8i16 (bitconvert (v2i64 FPR128:$src))), (v8i16 FPR128:$src)>;
5709 def : Pat<(v8i16 (bitconvert (v4i32 FPR128:$src))), (v8i16 FPR128:$src)>;
5710 def : Pat<(v8i16 (bitconvert (v16i8 FPR128:$src))), (v8i16 FPR128:$src)>;
5711 def : Pat<(v8i16 (bitconvert (v2f64 FPR128:$src))), (v8i16 FPR128:$src)>;
5712 def : Pat<(v8i16 (bitconvert (v4f32 FPR128:$src))), (v8i16 FPR128:$src)>;
5713 def : Pat<(v8i16 (bitconvert (v8f16 FPR128:$src))), (v8i16 FPR128:$src)>;
5715 let Predicates = [IsBE] in {
5716 def : Pat<(v8i16 (bitconvert (f128 FPR128:$src))),
5717 (v8i16 (EXTv16i8 (REV64v8i16 FPR128:$src),
5718 (REV64v8i16 FPR128:$src),
5720 def : Pat<(v8i16 (bitconvert (v2i64 FPR128:$src))),
5721 (v8i16 (REV64v8i16 FPR128:$src))>;
5722 def : Pat<(v8i16 (bitconvert (v4i32 FPR128:$src))),
5723 (v8i16 (REV32v8i16 FPR128:$src))>;
5724 def : Pat<(v8i16 (bitconvert (v16i8 FPR128:$src))),
5725 (v8i16 (REV16v16i8 FPR128:$src))>;
5726 def : Pat<(v8i16 (bitconvert (v2f64 FPR128:$src))),
5727 (v8i16 (REV64v8i16 FPR128:$src))>;
5728 def : Pat<(v8i16 (bitconvert (v4f32 FPR128:$src))),
5729 (v8i16 (REV32v8i16 FPR128:$src))>;
5730 def : Pat<(v8i16 (bitconvert (v8f16 FPR128:$src))),
5731 (v8i16 (REV32v8i16 FPR128:$src))>;
5734 let Predicates = [IsLE] in {
5735 def : Pat<(v8f16 (bitconvert (f128 FPR128:$src))), (v8f16 FPR128:$src)>;
5736 def : Pat<(v8f16 (bitconvert (v2i64 FPR128:$src))), (v8f16 FPR128:$src)>;
5737 def : Pat<(v8f16 (bitconvert (v4i32 FPR128:$src))), (v8f16 FPR128:$src)>;
5738 def : Pat<(v8f16 (bitconvert (v8i16 FPR128:$src))), (v8f16 FPR128:$src)>;
5739 def : Pat<(v8f16 (bitconvert (v16i8 FPR128:$src))), (v8f16 FPR128:$src)>;
5740 def : Pat<(v8f16 (bitconvert (v2f64 FPR128:$src))), (v8f16 FPR128:$src)>;
5741 def : Pat<(v8f16 (bitconvert (v4f32 FPR128:$src))), (v8f16 FPR128:$src)>;
5743 let Predicates = [IsBE] in {
5744 def : Pat<(v8f16 (bitconvert (f128 FPR128:$src))),
5745 (v8f16 (EXTv16i8 (REV64v8i16 FPR128:$src),
5746 (REV64v8i16 FPR128:$src),
5748 def : Pat<(v8f16 (bitconvert (v2i64 FPR128:$src))),
5749 (v8f16 (REV64v8i16 FPR128:$src))>;
5750 def : Pat<(v8f16 (bitconvert (v4i32 FPR128:$src))),
5751 (v8f16 (REV32v8i16 FPR128:$src))>;
5752 def : Pat<(v8f16 (bitconvert (v8i16 FPR128:$src))),
5753 (v8f16 (REV64v8i16 FPR128:$src))>;
5754 def : Pat<(v8f16 (bitconvert (v16i8 FPR128:$src))),
5755 (v8f16 (REV16v16i8 FPR128:$src))>;
5756 def : Pat<(v8f16 (bitconvert (v2f64 FPR128:$src))),
5757 (v8f16 (REV64v8i16 FPR128:$src))>;
5758 def : Pat<(v8f16 (bitconvert (v4f32 FPR128:$src))),
5759 (v8f16 (REV32v8i16 FPR128:$src))>;
5762 let Predicates = [IsLE] in {
5763 def : Pat<(v16i8 (bitconvert (f128 FPR128:$src))), (v16i8 FPR128:$src)>;
5764 def : Pat<(v16i8 (bitconvert (v2i64 FPR128:$src))), (v16i8 FPR128:$src)>;
5765 def : Pat<(v16i8 (bitconvert (v4i32 FPR128:$src))), (v16i8 FPR128:$src)>;
5766 def : Pat<(v16i8 (bitconvert (v8i16 FPR128:$src))), (v16i8 FPR128:$src)>;
5767 def : Pat<(v16i8 (bitconvert (v2f64 FPR128:$src))), (v16i8 FPR128:$src)>;
5768 def : Pat<(v16i8 (bitconvert (v4f32 FPR128:$src))), (v16i8 FPR128:$src)>;
5769 def : Pat<(v16i8 (bitconvert (v8f16 FPR128:$src))), (v16i8 FPR128:$src)>;
5771 let Predicates = [IsBE] in {
5772 def : Pat<(v16i8 (bitconvert (f128 FPR128:$src))),
5773 (v16i8 (EXTv16i8 (REV64v16i8 FPR128:$src),
5774 (REV64v16i8 FPR128:$src),
5776 def : Pat<(v16i8 (bitconvert (v2i64 FPR128:$src))),
5777 (v16i8 (REV64v16i8 FPR128:$src))>;
5778 def : Pat<(v16i8 (bitconvert (v4i32 FPR128:$src))),
5779 (v16i8 (REV32v16i8 FPR128:$src))>;
5780 def : Pat<(v16i8 (bitconvert (v8i16 FPR128:$src))),
5781 (v16i8 (REV16v16i8 FPR128:$src))>;
5782 def : Pat<(v16i8 (bitconvert (v2f64 FPR128:$src))),
5783 (v16i8 (REV64v16i8 FPR128:$src))>;
5784 def : Pat<(v16i8 (bitconvert (v4f32 FPR128:$src))),
5785 (v16i8 (REV32v16i8 FPR128:$src))>;
5786 def : Pat<(v16i8 (bitconvert (v8f16 FPR128:$src))),
5787 (v16i8 (REV16v16i8 FPR128:$src))>;
5790 def : Pat<(v8i8 (extract_subvector (v16i8 FPR128:$Rn), (i64 1))),
5791 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
5792 def : Pat<(v4i16 (extract_subvector (v8i16 FPR128:$Rn), (i64 1))),
5793 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
5794 def : Pat<(v2i32 (extract_subvector (v4i32 FPR128:$Rn), (i64 1))),
5795 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
5796 def : Pat<(v1i64 (extract_subvector (v2i64 FPR128:$Rn), (i64 1))),
5797 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
5799 // A 64-bit subvector insert to the first 128-bit vector position
5800 // is a subregister copy that needs no instruction.
5801 def : Pat<(insert_subvector undef, (v1i64 FPR64:$src), (i32 0)),
5802 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5803 def : Pat<(insert_subvector undef, (v1f64 FPR64:$src), (i32 0)),
5804 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5805 def : Pat<(insert_subvector undef, (v2i32 FPR64:$src), (i32 0)),
5806 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5807 def : Pat<(insert_subvector undef, (v2f32 FPR64:$src), (i32 0)),
5808 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5809 def : Pat<(insert_subvector undef, (v4i16 FPR64:$src), (i32 0)),
5810 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5811 def : Pat<(insert_subvector undef, (v4f16 FPR64:$src), (i32 0)),
5812 (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5813 def : Pat<(insert_subvector undef, (v8i8 FPR64:$src), (i32 0)),
5814 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5816 // Use pair-wise add instructions when summing up the lanes for v2f64, v2i64
5818 def : Pat<(i64 (add (vector_extract (v2i64 FPR128:$Rn), (i64 0)),
5819 (vector_extract (v2i64 FPR128:$Rn), (i64 1)))),
5820 (i64 (ADDPv2i64p (v2i64 FPR128:$Rn)))>;
5821 def : Pat<(f64 (fadd (vector_extract (v2f64 FPR128:$Rn), (i64 0)),
5822 (vector_extract (v2f64 FPR128:$Rn), (i64 1)))),
5823 (f64 (FADDPv2i64p (v2f64 FPR128:$Rn)))>;
5824 // vector_extract on 64-bit vectors gets promoted to a 128 bit vector,
5825 // so we match on v4f32 here, not v2f32. This will also catch adding
5826 // the low two lanes of a true v4f32 vector.
5827 def : Pat<(fadd (vector_extract (v4f32 FPR128:$Rn), (i64 0)),
5828 (vector_extract (v4f32 FPR128:$Rn), (i64 1))),
5829 (f32 (FADDPv2i32p (EXTRACT_SUBREG FPR128:$Rn, dsub)))>;
5831 // Scalar 64-bit shifts in FPR64 registers.
5832 def : Pat<(i64 (int_aarch64_neon_sshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
5833 (SSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
5834 def : Pat<(i64 (int_aarch64_neon_ushl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
5835 (USHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
5836 def : Pat<(i64 (int_aarch64_neon_srshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
5837 (SRSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
5838 def : Pat<(i64 (int_aarch64_neon_urshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
5839 (URSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
5841 // Patterns for nontemporal/no-allocate stores.
5842 // We have to resort to tricks to turn a single-input store into a store pair,
5843 // because there is no single-input nontemporal store, only STNP.
5844 let Predicates = [IsLE] in {
5845 let AddedComplexity = 15 in {
5846 class NTStore128Pat<ValueType VT> :
5847 Pat<(nontemporalstore (VT FPR128:$Rt),
5848 (am_indexed7s64 GPR64sp:$Rn, simm7s8:$offset)),
5849 (STNPDi (EXTRACT_SUBREG FPR128:$Rt, dsub),
5850 (CPYi64 FPR128:$Rt, (i64 1)),
5851 GPR64sp:$Rn, simm7s8:$offset)>;
5853 def : NTStore128Pat<v2i64>;
5854 def : NTStore128Pat<v4i32>;
5855 def : NTStore128Pat<v8i16>;
5856 def : NTStore128Pat<v16i8>;
5858 class NTStore64Pat<ValueType VT> :
5859 Pat<(nontemporalstore (VT FPR64:$Rt),
5860 (am_indexed7s32 GPR64sp:$Rn, simm7s4:$offset)),
5861 (STNPSi (EXTRACT_SUBREG FPR64:$Rt, ssub),
5862 (CPYi32 (SUBREG_TO_REG (i64 0), FPR64:$Rt, dsub), (i64 1)),
5863 GPR64sp:$Rn, simm7s4:$offset)>;
5865 // FIXME: Shouldn't v1f64 loads/stores be promoted to v1i64?
5866 def : NTStore64Pat<v1f64>;
5867 def : NTStore64Pat<v1i64>;
5868 def : NTStore64Pat<v2i32>;
5869 def : NTStore64Pat<v4i16>;
5870 def : NTStore64Pat<v8i8>;
5872 def : Pat<(nontemporalstore GPR64:$Rt,
5873 (am_indexed7s32 GPR64sp:$Rn, simm7s4:$offset)),
5874 (STNPWi (EXTRACT_SUBREG GPR64:$Rt, sub_32),
5875 (EXTRACT_SUBREG (UBFMXri GPR64:$Rt, 0, 31), sub_32),
5876 GPR64sp:$Rn, simm7s4:$offset)>;
5877 } // AddedComplexity=10
5878 } // Predicates = [IsLE]
5880 // Tail call return handling. These are all compiler pseudo-instructions,
5881 // so no encoding information or anything like that.
5882 let isCall = 1, isTerminator = 1, isReturn = 1, isBarrier = 1, Uses = [SP] in {
5883 def TCRETURNdi : Pseudo<(outs), (ins i64imm:$dst, i32imm:$FPDiff),[]>;
5884 def TCRETURNri : Pseudo<(outs), (ins tcGPR64:$dst, i32imm:$FPDiff), []>;
5887 def : Pat<(AArch64tcret tcGPR64:$dst, (i32 timm:$FPDiff)),
5888 (TCRETURNri tcGPR64:$dst, imm:$FPDiff)>;
5889 def : Pat<(AArch64tcret tglobaladdr:$dst, (i32 timm:$FPDiff)),
5890 (TCRETURNdi texternalsym:$dst, imm:$FPDiff)>;
5891 def : Pat<(AArch64tcret texternalsym:$dst, (i32 timm:$FPDiff)),
5892 (TCRETURNdi texternalsym:$dst, imm:$FPDiff)>;
5894 include "AArch64InstrAtomics.td"