1 //===- ARM64InstrInfo.td - Describe the ARM64 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 // ARM64 Instruction definitions.
12 //===----------------------------------------------------------------------===//
14 //===----------------------------------------------------------------------===//
15 // ARM Instruction Predicate Definitions.
17 def HasFPARMv8 : Predicate<"Subtarget->hasFPARMv8()">,
18 AssemblerPredicate<"FeatureFPARMv8", "fp-armv8">;
19 def HasNEON : Predicate<"Subtarget->hasNEON()">,
20 AssemblerPredicate<"FeatureNEON", "neon">;
21 def HasCrypto : Predicate<"Subtarget->hasCrypto()">,
22 AssemblerPredicate<"FeatureCrypto", "crypto">;
23 def HasCRC : Predicate<"Subtarget->hasCRC()">,
24 AssemblerPredicate<"FeatureCRC", "crc">;
25 def IsLE : Predicate<"Subtarget->isLittleEndian()">;
26 def IsBE : Predicate<"!Subtarget->isLittleEndian()">;
28 //===----------------------------------------------------------------------===//
29 // ARM64-specific DAG Nodes.
32 // SDTBinaryArithWithFlagsOut - RES1, FLAGS = op LHS, RHS
33 def SDTBinaryArithWithFlagsOut : SDTypeProfile<2, 2,
36 SDTCisInt<0>, SDTCisVT<1, i32>]>;
38 // SDTBinaryArithWithFlagsIn - RES1, FLAGS = op LHS, RHS, FLAGS
39 def SDTBinaryArithWithFlagsIn : SDTypeProfile<1, 3,
45 // SDTBinaryArithWithFlagsInOut - RES1, FLAGS = op LHS, RHS, FLAGS
46 def SDTBinaryArithWithFlagsInOut : SDTypeProfile<2, 3,
53 def SDT_ARM64Brcond : SDTypeProfile<0, 3,
54 [SDTCisVT<0, OtherVT>, SDTCisVT<1, i32>,
56 def SDT_ARM64cbz : SDTypeProfile<0, 2, [SDTCisInt<0>, SDTCisVT<1, OtherVT>]>;
57 def SDT_ARM64tbz : SDTypeProfile<0, 3, [SDTCisInt<0>, SDTCisInt<1>,
58 SDTCisVT<2, OtherVT>]>;
61 def SDT_ARM64CSel : SDTypeProfile<1, 4,
66 def SDT_ARM64FCmp : SDTypeProfile<0, 2,
69 def SDT_ARM64Dup : SDTypeProfile<1, 1, [SDTCisVec<0>]>;
70 def SDT_ARM64DupLane : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisInt<2>]>;
71 def SDT_ARM64Zip : SDTypeProfile<1, 2, [SDTCisVec<0>,
74 def SDT_ARM64MOVIedit : SDTypeProfile<1, 1, [SDTCisInt<1>]>;
75 def SDT_ARM64MOVIshift : SDTypeProfile<1, 2, [SDTCisInt<1>, SDTCisInt<2>]>;
76 def SDT_ARM64vecimm : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0,1>,
77 SDTCisInt<2>, SDTCisInt<3>]>;
78 def SDT_ARM64UnaryVec: SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0,1>]>;
79 def SDT_ARM64ExtVec: SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0,1>,
80 SDTCisSameAs<0,2>, SDTCisInt<3>]>;
81 def SDT_ARM64vshift : SDTypeProfile<1, 2, [SDTCisSameAs<0,1>, SDTCisInt<2>]>;
83 def SDT_ARM64unvec : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0,1>]>;
84 def SDT_ARM64fcmpz : SDTypeProfile<1, 1, []>;
85 def SDT_ARM64fcmp : SDTypeProfile<1, 2, [SDTCisSameAs<1,2>]>;
86 def SDT_ARM64binvec : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisSameAs<0,1>,
88 def SDT_ARM64trivec : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0,1>,
91 def SDT_ARM64TCRET : SDTypeProfile<0, 2, [SDTCisPtrTy<0>]>;
92 def SDT_ARM64PREFETCH : SDTypeProfile<0, 2, [SDTCisVT<0, i32>, SDTCisPtrTy<1>]>;
94 def SDT_ARM64ITOF : SDTypeProfile<1, 1, [SDTCisFP<0>, SDTCisSameAs<0,1>]>;
96 def SDT_ARM64TLSDescCall : SDTypeProfile<0, -2, [SDTCisPtrTy<0>,
98 def SDT_ARM64WrapperLarge : SDTypeProfile<1, 4,
99 [SDTCisVT<0, i64>, SDTCisVT<1, i32>,
100 SDTCisSameAs<1, 2>, SDTCisSameAs<1, 3>,
101 SDTCisSameAs<1, 4>]>;
105 def ARM64adrp : SDNode<"ARM64ISD::ADRP", SDTIntUnaryOp, []>;
106 def ARM64addlow : SDNode<"ARM64ISD::ADDlow", SDTIntBinOp, []>;
107 def ARM64LOADgot : SDNode<"ARM64ISD::LOADgot", SDTIntUnaryOp>;
108 def ARM64callseq_start : SDNode<"ISD::CALLSEQ_START",
109 SDCallSeqStart<[ SDTCisVT<0, i32> ]>,
110 [SDNPHasChain, SDNPOutGlue]>;
111 def ARM64callseq_end : SDNode<"ISD::CALLSEQ_END",
112 SDCallSeqEnd<[ SDTCisVT<0, i32>,
114 [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue]>;
115 def ARM64call : SDNode<"ARM64ISD::CALL",
116 SDTypeProfile<0, -1, [SDTCisPtrTy<0>]>,
117 [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue,
119 def ARM64brcond : SDNode<"ARM64ISD::BRCOND", SDT_ARM64Brcond,
121 def ARM64cbz : SDNode<"ARM64ISD::CBZ", SDT_ARM64cbz,
123 def ARM64cbnz : SDNode<"ARM64ISD::CBNZ", SDT_ARM64cbz,
125 def ARM64tbz : SDNode<"ARM64ISD::TBZ", SDT_ARM64tbz,
127 def ARM64tbnz : SDNode<"ARM64ISD::TBNZ", SDT_ARM64tbz,
131 def ARM64csel : SDNode<"ARM64ISD::CSEL", SDT_ARM64CSel>;
132 def ARM64csinv : SDNode<"ARM64ISD::CSINV", SDT_ARM64CSel>;
133 def ARM64csneg : SDNode<"ARM64ISD::CSNEG", SDT_ARM64CSel>;
134 def ARM64csinc : SDNode<"ARM64ISD::CSINC", SDT_ARM64CSel>;
135 def ARM64retflag : SDNode<"ARM64ISD::RET_FLAG", SDTNone,
136 [SDNPHasChain, SDNPOptInGlue, SDNPVariadic]>;
137 def ARM64adc : SDNode<"ARM64ISD::ADC", SDTBinaryArithWithFlagsIn >;
138 def ARM64sbc : SDNode<"ARM64ISD::SBC", SDTBinaryArithWithFlagsIn>;
139 def ARM64add_flag : SDNode<"ARM64ISD::ADDS", SDTBinaryArithWithFlagsOut,
141 def ARM64sub_flag : SDNode<"ARM64ISD::SUBS", SDTBinaryArithWithFlagsOut>;
142 def ARM64and_flag : SDNode<"ARM64ISD::ANDS", SDTBinaryArithWithFlagsOut,
144 def ARM64adc_flag : SDNode<"ARM64ISD::ADCS", SDTBinaryArithWithFlagsInOut>;
145 def ARM64sbc_flag : SDNode<"ARM64ISD::SBCS", SDTBinaryArithWithFlagsInOut>;
147 def ARM64threadpointer : SDNode<"ARM64ISD::THREAD_POINTER", SDTPtrLeaf>;
149 def ARM64fcmp : SDNode<"ARM64ISD::FCMP", SDT_ARM64FCmp>;
151 def ARM64fmax : SDNode<"ARM64ISD::FMAX", SDTFPBinOp>;
152 def ARM64fmin : SDNode<"ARM64ISD::FMIN", SDTFPBinOp>;
154 def ARM64dup : SDNode<"ARM64ISD::DUP", SDT_ARM64Dup>;
155 def ARM64duplane8 : SDNode<"ARM64ISD::DUPLANE8", SDT_ARM64DupLane>;
156 def ARM64duplane16 : SDNode<"ARM64ISD::DUPLANE16", SDT_ARM64DupLane>;
157 def ARM64duplane32 : SDNode<"ARM64ISD::DUPLANE32", SDT_ARM64DupLane>;
158 def ARM64duplane64 : SDNode<"ARM64ISD::DUPLANE64", SDT_ARM64DupLane>;
160 def ARM64zip1 : SDNode<"ARM64ISD::ZIP1", SDT_ARM64Zip>;
161 def ARM64zip2 : SDNode<"ARM64ISD::ZIP2", SDT_ARM64Zip>;
162 def ARM64uzp1 : SDNode<"ARM64ISD::UZP1", SDT_ARM64Zip>;
163 def ARM64uzp2 : SDNode<"ARM64ISD::UZP2", SDT_ARM64Zip>;
164 def ARM64trn1 : SDNode<"ARM64ISD::TRN1", SDT_ARM64Zip>;
165 def ARM64trn2 : SDNode<"ARM64ISD::TRN2", SDT_ARM64Zip>;
167 def ARM64movi_edit : SDNode<"ARM64ISD::MOVIedit", SDT_ARM64MOVIedit>;
168 def ARM64movi_shift : SDNode<"ARM64ISD::MOVIshift", SDT_ARM64MOVIshift>;
169 def ARM64movi_msl : SDNode<"ARM64ISD::MOVImsl", SDT_ARM64MOVIshift>;
170 def ARM64mvni_shift : SDNode<"ARM64ISD::MVNIshift", SDT_ARM64MOVIshift>;
171 def ARM64mvni_msl : SDNode<"ARM64ISD::MVNImsl", SDT_ARM64MOVIshift>;
172 def ARM64movi : SDNode<"ARM64ISD::MOVI", SDT_ARM64MOVIedit>;
173 def ARM64fmov : SDNode<"ARM64ISD::FMOV", SDT_ARM64MOVIedit>;
175 def ARM64rev16 : SDNode<"ARM64ISD::REV16", SDT_ARM64UnaryVec>;
176 def ARM64rev32 : SDNode<"ARM64ISD::REV32", SDT_ARM64UnaryVec>;
177 def ARM64rev64 : SDNode<"ARM64ISD::REV64", SDT_ARM64UnaryVec>;
178 def ARM64ext : SDNode<"ARM64ISD::EXT", SDT_ARM64ExtVec>;
180 def ARM64vashr : SDNode<"ARM64ISD::VASHR", SDT_ARM64vshift>;
181 def ARM64vlshr : SDNode<"ARM64ISD::VLSHR", SDT_ARM64vshift>;
182 def ARM64vshl : SDNode<"ARM64ISD::VSHL", SDT_ARM64vshift>;
183 def ARM64sqshli : SDNode<"ARM64ISD::SQSHL_I", SDT_ARM64vshift>;
184 def ARM64uqshli : SDNode<"ARM64ISD::UQSHL_I", SDT_ARM64vshift>;
185 def ARM64sqshlui : SDNode<"ARM64ISD::SQSHLU_I", SDT_ARM64vshift>;
186 def ARM64srshri : SDNode<"ARM64ISD::SRSHR_I", SDT_ARM64vshift>;
187 def ARM64urshri : SDNode<"ARM64ISD::URSHR_I", SDT_ARM64vshift>;
189 def ARM64not: SDNode<"ARM64ISD::NOT", SDT_ARM64unvec>;
190 def ARM64bit: SDNode<"ARM64ISD::BIT", SDT_ARM64trivec>;
191 def ARM64bsl: SDNode<"ARM64ISD::BSL", SDT_ARM64trivec>;
193 def ARM64cmeq: SDNode<"ARM64ISD::CMEQ", SDT_ARM64binvec>;
194 def ARM64cmge: SDNode<"ARM64ISD::CMGE", SDT_ARM64binvec>;
195 def ARM64cmgt: SDNode<"ARM64ISD::CMGT", SDT_ARM64binvec>;
196 def ARM64cmhi: SDNode<"ARM64ISD::CMHI", SDT_ARM64binvec>;
197 def ARM64cmhs: SDNode<"ARM64ISD::CMHS", SDT_ARM64binvec>;
199 def ARM64fcmeq: SDNode<"ARM64ISD::FCMEQ", SDT_ARM64fcmp>;
200 def ARM64fcmge: SDNode<"ARM64ISD::FCMGE", SDT_ARM64fcmp>;
201 def ARM64fcmgt: SDNode<"ARM64ISD::FCMGT", SDT_ARM64fcmp>;
203 def ARM64cmeqz: SDNode<"ARM64ISD::CMEQz", SDT_ARM64unvec>;
204 def ARM64cmgez: SDNode<"ARM64ISD::CMGEz", SDT_ARM64unvec>;
205 def ARM64cmgtz: SDNode<"ARM64ISD::CMGTz", SDT_ARM64unvec>;
206 def ARM64cmlez: SDNode<"ARM64ISD::CMLEz", SDT_ARM64unvec>;
207 def ARM64cmltz: SDNode<"ARM64ISD::CMLTz", SDT_ARM64unvec>;
208 def ARM64cmtst : PatFrag<(ops node:$LHS, node:$RHS),
209 (ARM64not (ARM64cmeqz (and node:$LHS, node:$RHS)))>;
211 def ARM64fcmeqz: SDNode<"ARM64ISD::FCMEQz", SDT_ARM64fcmpz>;
212 def ARM64fcmgez: SDNode<"ARM64ISD::FCMGEz", SDT_ARM64fcmpz>;
213 def ARM64fcmgtz: SDNode<"ARM64ISD::FCMGTz", SDT_ARM64fcmpz>;
214 def ARM64fcmlez: SDNode<"ARM64ISD::FCMLEz", SDT_ARM64fcmpz>;
215 def ARM64fcmltz: SDNode<"ARM64ISD::FCMLTz", SDT_ARM64fcmpz>;
217 def ARM64bici: SDNode<"ARM64ISD::BICi", SDT_ARM64vecimm>;
218 def ARM64orri: SDNode<"ARM64ISD::ORRi", SDT_ARM64vecimm>;
220 def ARM64neg : SDNode<"ARM64ISD::NEG", SDT_ARM64unvec>;
222 def ARM64tcret: SDNode<"ARM64ISD::TC_RETURN", SDT_ARM64TCRET,
223 [SDNPHasChain, SDNPOptInGlue, SDNPVariadic]>;
225 def ARM64Prefetch : SDNode<"ARM64ISD::PREFETCH", SDT_ARM64PREFETCH,
226 [SDNPHasChain, SDNPSideEffect]>;
228 def ARM64sitof: SDNode<"ARM64ISD::SITOF", SDT_ARM64ITOF>;
229 def ARM64uitof: SDNode<"ARM64ISD::UITOF", SDT_ARM64ITOF>;
231 def ARM64tlsdesc_call : SDNode<"ARM64ISD::TLSDESC_CALL", SDT_ARM64TLSDescCall,
232 [SDNPInGlue, SDNPOutGlue, SDNPHasChain,
235 def ARM64WrapperLarge : SDNode<"ARM64ISD::WrapperLarge", SDT_ARM64WrapperLarge>;
238 //===----------------------------------------------------------------------===//
240 //===----------------------------------------------------------------------===//
242 // ARM64 Instruction Predicate Definitions.
244 def HasZCZ : Predicate<"Subtarget->hasZeroCycleZeroing()">;
245 def NoZCZ : Predicate<"!Subtarget->hasZeroCycleZeroing()">;
246 def IsDarwin : Predicate<"Subtarget->isTargetDarwin()">;
247 def IsNotDarwin: Predicate<"!Subtarget->isTargetDarwin()">;
248 def ForCodeSize : Predicate<"ForCodeSize">;
249 def NotForCodeSize : Predicate<"!ForCodeSize">;
251 include "ARM64InstrFormats.td"
253 //===----------------------------------------------------------------------===//
255 //===----------------------------------------------------------------------===//
256 // Miscellaneous instructions.
257 //===----------------------------------------------------------------------===//
259 let Defs = [SP], Uses = [SP], hasSideEffects = 1, isCodeGenOnly = 1 in {
260 def ADJCALLSTACKDOWN : Pseudo<(outs), (ins i32imm:$amt),
261 [(ARM64callseq_start timm:$amt)]>;
262 def ADJCALLSTACKUP : Pseudo<(outs), (ins i32imm:$amt1, i32imm:$amt2),
263 [(ARM64callseq_end timm:$amt1, timm:$amt2)]>;
264 } // Defs = [SP], Uses = [SP], hasSideEffects = 1, isCodeGenOnly = 1
266 let isReMaterializable = 1, isCodeGenOnly = 1 in {
267 // FIXME: The following pseudo instructions are only needed because remat
268 // cannot handle multiple instructions. When that changes, they can be
269 // removed, along with the ARM64Wrapper node.
271 let AddedComplexity = 10 in
272 def LOADgot : Pseudo<(outs GPR64:$dst), (ins i64imm:$addr),
273 [(set GPR64:$dst, (ARM64LOADgot tglobaladdr:$addr))]>,
276 // The MOVaddr instruction should match only when the add is not folded
277 // into a load or store address.
279 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
280 [(set GPR64:$dst, (ARM64addlow (ARM64adrp tglobaladdr:$hi),
281 tglobaladdr:$low))]>,
282 Sched<[WriteAdrAdr]>;
284 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
285 [(set GPR64:$dst, (ARM64addlow (ARM64adrp tjumptable:$hi),
287 Sched<[WriteAdrAdr]>;
289 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
290 [(set GPR64:$dst, (ARM64addlow (ARM64adrp tconstpool:$hi),
292 Sched<[WriteAdrAdr]>;
294 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
295 [(set GPR64:$dst, (ARM64addlow (ARM64adrp tblockaddress:$hi),
296 tblockaddress:$low))]>,
297 Sched<[WriteAdrAdr]>;
299 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
300 [(set GPR64:$dst, (ARM64addlow (ARM64adrp tglobaltlsaddr:$hi),
301 tglobaltlsaddr:$low))]>,
302 Sched<[WriteAdrAdr]>;
304 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
305 [(set GPR64:$dst, (ARM64addlow (ARM64adrp texternalsym:$hi),
306 texternalsym:$low))]>,
307 Sched<[WriteAdrAdr]>;
309 } // isReMaterializable, isCodeGenOnly
311 def : Pat<(ARM64LOADgot tglobaltlsaddr:$addr),
312 (LOADgot tglobaltlsaddr:$addr)>;
314 def : Pat<(ARM64LOADgot texternalsym:$addr),
315 (LOADgot texternalsym:$addr)>;
317 def : Pat<(ARM64LOADgot tconstpool:$addr),
318 (LOADgot tconstpool:$addr)>;
320 //===----------------------------------------------------------------------===//
321 // System instructions.
322 //===----------------------------------------------------------------------===//
324 def HINT : HintI<"hint">;
325 def : InstAlias<"nop", (HINT 0b000)>;
326 def : InstAlias<"yield",(HINT 0b001)>;
327 def : InstAlias<"wfe", (HINT 0b010)>;
328 def : InstAlias<"wfi", (HINT 0b011)>;
329 def : InstAlias<"sev", (HINT 0b100)>;
330 def : InstAlias<"sevl", (HINT 0b101)>;
332 // As far as LLVM is concerned this writes to the system's exclusive monitors.
333 let mayLoad = 1, mayStore = 1 in
334 def CLREX : CRmSystemI<imm0_15, 0b010, "clrex">;
336 def DMB : CRmSystemI<barrier_op, 0b101, "dmb">;
337 def DSB : CRmSystemI<barrier_op, 0b100, "dsb">;
338 def ISB : CRmSystemI<barrier_op, 0b110, "isb">;
339 def : InstAlias<"clrex", (CLREX 0xf)>;
340 def : InstAlias<"isb", (ISB 0xf)>;
344 def MSRpstate: MSRpstateI;
346 // The thread pointer (on Linux, at least, where this has been implemented) is
348 def : Pat<(ARM64threadpointer), (MRS 0xde82)>;
350 // Generic system instructions
351 def SYSxt : SystemXtI<0, "sys">;
352 def SYSLxt : SystemLXtI<1, "sysl">;
354 def : InstAlias<"sys $op1, $Cn, $Cm, $op2",
355 (SYSxt imm0_7:$op1, sys_cr_op:$Cn,
356 sys_cr_op:$Cm, imm0_7:$op2, XZR)>;
358 //===----------------------------------------------------------------------===//
359 // Move immediate instructions.
360 //===----------------------------------------------------------------------===//
362 defm MOVK : InsertImmediate<0b11, "movk">;
363 defm MOVN : MoveImmediate<0b00, "movn">;
365 let PostEncoderMethod = "fixMOVZ" in
366 defm MOVZ : MoveImmediate<0b10, "movz">;
368 // First group of aliases covers an implicit "lsl #0".
369 def : InstAlias<"movk $dst, $imm", (MOVKWi GPR32:$dst, imm0_65535:$imm, 0)>;
370 def : InstAlias<"movk $dst, $imm", (MOVKXi GPR64:$dst, imm0_65535:$imm, 0)>;
371 def : InstAlias<"movn $dst, $imm", (MOVNWi GPR32:$dst, imm0_65535:$imm, 0)>;
372 def : InstAlias<"movn $dst, $imm", (MOVNXi GPR64:$dst, imm0_65535:$imm, 0)>;
373 def : InstAlias<"movz $dst, $imm", (MOVZWi GPR32:$dst, imm0_65535:$imm, 0)>;
374 def : InstAlias<"movz $dst, $imm", (MOVZXi GPR64:$dst, imm0_65535:$imm, 0)>;
376 // Next, we have various ELF relocations with the ":XYZ_g0:sym" syntax.
377 def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movz_symbol_g3:$sym, 48)>;
378 def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movz_symbol_g2:$sym, 32)>;
379 def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movz_symbol_g1:$sym, 16)>;
380 def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movz_symbol_g0:$sym, 0)>;
382 def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movz_symbol_g3:$sym, 48)>;
383 def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movz_symbol_g2:$sym, 32)>;
384 def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movz_symbol_g1:$sym, 16)>;
385 def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movz_symbol_g0:$sym, 0)>;
387 def : InstAlias<"movk $Rd, $sym", (MOVKXi GPR64:$Rd, movk_symbol_g3:$sym, 48)>;
388 def : InstAlias<"movk $Rd, $sym", (MOVKXi GPR64:$Rd, movk_symbol_g2:$sym, 32)>;
389 def : InstAlias<"movk $Rd, $sym", (MOVKXi GPR64:$Rd, movk_symbol_g1:$sym, 16)>;
390 def : InstAlias<"movk $Rd, $sym", (MOVKXi GPR64:$Rd, movk_symbol_g0:$sym, 0)>;
392 def : InstAlias<"movz $Rd, $sym", (MOVZWi GPR32:$Rd, movz_symbol_g1:$sym, 16)>;
393 def : InstAlias<"movz $Rd, $sym", (MOVZWi GPR32:$Rd, movz_symbol_g0:$sym, 0)>;
395 def : InstAlias<"movn $Rd, $sym", (MOVNWi GPR32:$Rd, movz_symbol_g1:$sym, 16)>;
396 def : InstAlias<"movn $Rd, $sym", (MOVNWi GPR32:$Rd, movz_symbol_g0:$sym, 0)>;
398 def : InstAlias<"movk $Rd, $sym", (MOVKWi GPR32:$Rd, movk_symbol_g1:$sym, 16)>;
399 def : InstAlias<"movk $Rd, $sym", (MOVKWi GPR32:$Rd, movk_symbol_g0:$sym, 0)>;
401 // Final group of aliases covers true "mov $Rd, $imm" cases.
402 multiclass movw_mov_alias<string basename,Instruction INST, RegisterClass GPR,
403 int width, int shift> {
404 def _asmoperand : AsmOperandClass {
405 let Name = basename # width # "_lsl" # shift # "MovAlias";
406 let PredicateMethod = "is" # basename # "MovAlias<" # width # ", "
408 let RenderMethod = "add" # basename # "MovAliasOperands<" # shift # ">";
411 def _movimm : Operand<i32> {
412 let ParserMatchClass = !cast<AsmOperandClass>(NAME # "_asmoperand");
415 def : InstAlias<"mov $Rd, $imm",
416 (INST GPR:$Rd, !cast<Operand>(NAME # "_movimm"):$imm, shift)>;
419 defm : movw_mov_alias<"MOVZ", MOVZWi, GPR32, 32, 0>;
420 defm : movw_mov_alias<"MOVZ", MOVZWi, GPR32, 32, 16>;
422 defm : movw_mov_alias<"MOVZ", MOVZXi, GPR64, 64, 0>;
423 defm : movw_mov_alias<"MOVZ", MOVZXi, GPR64, 64, 16>;
424 defm : movw_mov_alias<"MOVZ", MOVZXi, GPR64, 64, 32>;
425 defm : movw_mov_alias<"MOVZ", MOVZXi, GPR64, 64, 48>;
427 defm : movw_mov_alias<"MOVN", MOVNWi, GPR32, 32, 0>;
428 defm : movw_mov_alias<"MOVN", MOVNWi, GPR32, 32, 16>;
430 defm : movw_mov_alias<"MOVN", MOVNXi, GPR64, 64, 0>;
431 defm : movw_mov_alias<"MOVN", MOVNXi, GPR64, 64, 16>;
432 defm : movw_mov_alias<"MOVN", MOVNXi, GPR64, 64, 32>;
433 defm : movw_mov_alias<"MOVN", MOVNXi, GPR64, 64, 48>;
435 let isReMaterializable = 1, isCodeGenOnly = 1, isMoveImm = 1,
436 isAsCheapAsAMove = 1 in {
437 // FIXME: The following pseudo instructions are only needed because remat
438 // cannot handle multiple instructions. When that changes, we can select
439 // directly to the real instructions and get rid of these pseudos.
442 : Pseudo<(outs GPR32:$dst), (ins i32imm:$src),
443 [(set GPR32:$dst, imm:$src)]>,
446 : Pseudo<(outs GPR64:$dst), (ins i64imm:$src),
447 [(set GPR64:$dst, imm:$src)]>,
449 } // isReMaterializable, isCodeGenOnly
451 // If possible, we want to use MOVi32imm even for 64-bit moves. This gives the
452 // eventual expansion code fewer bits to worry about getting right. Marshalling
453 // the types is a little tricky though:
454 def i64imm_32bit : ImmLeaf<i64, [{
455 return (Imm & 0xffffffffULL) == static_cast<uint64_t>(Imm);
458 def trunc_imm : SDNodeXForm<imm, [{
459 return CurDAG->getTargetConstant(N->getZExtValue(), MVT::i32);
462 def : Pat<(i64 i64imm_32bit:$src),
463 (SUBREG_TO_REG (i64 0), (MOVi32imm (trunc_imm imm:$src)), sub_32)>;
465 // Deal with the various forms of (ELF) large addressing with MOVZ/MOVK
467 def : Pat<(ARM64WrapperLarge tglobaladdr:$g3, tglobaladdr:$g2,
468 tglobaladdr:$g1, tglobaladdr:$g0),
469 (MOVKXi (MOVKXi (MOVKXi (MOVZXi tglobaladdr:$g3, 48),
470 tglobaladdr:$g2, 32),
471 tglobaladdr:$g1, 16),
472 tglobaladdr:$g0, 0)>;
474 def : Pat<(ARM64WrapperLarge tblockaddress:$g3, tblockaddress:$g2,
475 tblockaddress:$g1, tblockaddress:$g0),
476 (MOVKXi (MOVKXi (MOVKXi (MOVZXi tblockaddress:$g3, 48),
477 tblockaddress:$g2, 32),
478 tblockaddress:$g1, 16),
479 tblockaddress:$g0, 0)>;
481 def : Pat<(ARM64WrapperLarge tconstpool:$g3, tconstpool:$g2,
482 tconstpool:$g1, tconstpool:$g0),
483 (MOVKXi (MOVKXi (MOVKXi (MOVZXi tconstpool:$g3, 48),
488 def : Pat<(ARM64WrapperLarge tjumptable:$g3, tjumptable:$g2,
489 tjumptable:$g1, tjumptable:$g0),
490 (MOVKXi (MOVKXi (MOVKXi (MOVZXi tjumptable:$g3, 48),
496 //===----------------------------------------------------------------------===//
497 // Arithmetic instructions.
498 //===----------------------------------------------------------------------===//
500 // Add/subtract with carry.
501 defm ADC : AddSubCarry<0, "adc", "adcs", ARM64adc, ARM64adc_flag>;
502 defm SBC : AddSubCarry<1, "sbc", "sbcs", ARM64sbc, ARM64sbc_flag>;
504 def : InstAlias<"ngc $dst, $src", (SBCWr GPR32:$dst, WZR, GPR32:$src)>;
505 def : InstAlias<"ngc $dst, $src", (SBCXr GPR64:$dst, XZR, GPR64:$src)>;
506 def : InstAlias<"ngcs $dst, $src", (SBCSWr GPR32:$dst, WZR, GPR32:$src)>;
507 def : InstAlias<"ngcs $dst, $src", (SBCSXr GPR64:$dst, XZR, GPR64:$src)>;
510 defm ADD : AddSub<0, "add", add>;
511 defm SUB : AddSub<1, "sub">;
513 def : InstAlias<"mov $dst, $src",
514 (ADDWri GPR32sponly:$dst, GPR32sp:$src, 0, 0)>;
515 def : InstAlias<"mov $dst, $src",
516 (ADDWri GPR32sp:$dst, GPR32sponly:$src, 0, 0)>;
517 def : InstAlias<"mov $dst, $src",
518 (ADDXri GPR64sponly:$dst, GPR64sp:$src, 0, 0)>;
519 def : InstAlias<"mov $dst, $src",
520 (ADDXri GPR64sp:$dst, GPR64sponly:$src, 0, 0)>;
522 defm ADDS : AddSubS<0, "adds", ARM64add_flag, "cmn">;
523 defm SUBS : AddSubS<1, "subs", ARM64sub_flag, "cmp">;
525 // Use SUBS instead of SUB to enable CSE between SUBS and SUB.
526 def : Pat<(sub GPR32sp:$Rn, addsub_shifted_imm32:$imm),
527 (SUBSWri GPR32sp:$Rn, addsub_shifted_imm32:$imm)>;
528 def : Pat<(sub GPR64sp:$Rn, addsub_shifted_imm64:$imm),
529 (SUBSXri GPR64sp:$Rn, addsub_shifted_imm64:$imm)>;
530 def : Pat<(sub GPR32:$Rn, GPR32:$Rm),
531 (SUBSWrr GPR32:$Rn, GPR32:$Rm)>;
532 def : Pat<(sub GPR64:$Rn, GPR64:$Rm),
533 (SUBSXrr GPR64:$Rn, GPR64:$Rm)>;
534 def : Pat<(sub GPR32:$Rn, arith_shifted_reg32:$Rm),
535 (SUBSWrs GPR32:$Rn, arith_shifted_reg32:$Rm)>;
536 def : Pat<(sub GPR64:$Rn, arith_shifted_reg64:$Rm),
537 (SUBSXrs GPR64:$Rn, arith_shifted_reg64:$Rm)>;
538 def : Pat<(sub GPR32sp:$R2, arith_extended_reg32<i32>:$R3),
539 (SUBSWrx GPR32sp:$R2, arith_extended_reg32<i32>:$R3)>;
540 def : Pat<(sub GPR64sp:$R2, arith_extended_reg32to64<i64>:$R3),
541 (SUBSXrx GPR64sp:$R2, arith_extended_reg32to64<i64>:$R3)>;
543 // Because of the immediate format for add/sub-imm instructions, the
544 // expression (add x, -1) must be transformed to (SUB{W,X}ri x, 1).
545 // These patterns capture that transformation.
546 let AddedComplexity = 1 in {
547 def : Pat<(add GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
548 (SUBSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
549 def : Pat<(add GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
550 (SUBSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
551 def : Pat<(sub GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
552 (ADDWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
553 def : Pat<(sub GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
554 (ADDXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
557 // Because of the immediate format for add/sub-imm instructions, the
558 // expression (add x, -1) must be transformed to (SUB{W,X}ri x, 1).
559 // These patterns capture that transformation.
560 let AddedComplexity = 1 in {
561 def : Pat<(ARM64add_flag GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
562 (SUBSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
563 def : Pat<(ARM64add_flag GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
564 (SUBSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
565 def : Pat<(ARM64sub_flag GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
566 (ADDSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
567 def : Pat<(ARM64sub_flag GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
568 (ADDSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
571 def : InstAlias<"neg $dst, $src", (SUBWrs GPR32:$dst, WZR, GPR32:$src, 0)>;
572 def : InstAlias<"neg $dst, $src", (SUBXrs GPR64:$dst, XZR, GPR64:$src, 0)>;
573 def : InstAlias<"neg $dst, $src$shift",
574 (SUBWrs GPR32:$dst, WZR, GPR32:$src, arith_shift32:$shift)>;
575 def : InstAlias<"neg $dst, $src$shift",
576 (SUBXrs GPR64:$dst, XZR, GPR64:$src, arith_shift64:$shift)>;
578 def : InstAlias<"negs $dst, $src", (SUBSWrs GPR32:$dst, WZR, GPR32:$src, 0)>;
579 def : InstAlias<"negs $dst, $src", (SUBSXrs GPR64:$dst, XZR, GPR64:$src, 0)>;
580 def : InstAlias<"negs $dst, $src$shift",
581 (SUBSWrs GPR32:$dst, WZR, GPR32:$src, arith_shift32:$shift)>;
582 def : InstAlias<"negs $dst, $src$shift",
583 (SUBSXrs GPR64:$dst, XZR, GPR64:$src, arith_shift64:$shift)>;
586 // Register/register aliases with no shift when SP is not used.
587 def : AddSubRegAlias<"add", ADDWrs, GPR32, GPR32, GPR32, 0>;
588 def : AddSubRegAlias<"add", ADDXrs, GPR64, GPR64, GPR64, 0>;
589 def : AddSubRegAlias<"sub", SUBWrs, GPR32, GPR32, GPR32, 0>;
590 def : AddSubRegAlias<"sub", SUBXrs, GPR64, GPR64, GPR64, 0>;
591 def : AddSubRegAlias<"adds", ADDSWrs, GPR32, GPR32, GPR32, 0>;
592 def : AddSubRegAlias<"adds", ADDSXrs, GPR64, GPR64, GPR64, 0>;
593 def : AddSubRegAlias<"subs", SUBSWrs, GPR32, GPR32, GPR32, 0>;
594 def : AddSubRegAlias<"subs", SUBSXrs, GPR64, GPR64, GPR64, 0>;
597 // Unsigned/Signed divide
598 defm UDIV : Div<0, "udiv", udiv>;
599 defm SDIV : Div<1, "sdiv", sdiv>;
600 let isCodeGenOnly = 1 in {
601 defm UDIV_Int : Div<0, "udiv", int_arm64_udiv>;
602 defm SDIV_Int : Div<1, "sdiv", int_arm64_sdiv>;
606 defm ASRV : Shift<0b10, "asr", sra>;
607 defm LSLV : Shift<0b00, "lsl", shl>;
608 defm LSRV : Shift<0b01, "lsr", srl>;
609 defm RORV : Shift<0b11, "ror", rotr>;
611 def : ShiftAlias<"asrv", ASRVWr, GPR32>;
612 def : ShiftAlias<"asrv", ASRVXr, GPR64>;
613 def : ShiftAlias<"lslv", LSLVWr, GPR32>;
614 def : ShiftAlias<"lslv", LSLVXr, GPR64>;
615 def : ShiftAlias<"lsrv", LSRVWr, GPR32>;
616 def : ShiftAlias<"lsrv", LSRVXr, GPR64>;
617 def : ShiftAlias<"rorv", RORVWr, GPR32>;
618 def : ShiftAlias<"rorv", RORVXr, GPR64>;
621 let AddedComplexity = 7 in {
622 defm MADD : MulAccum<0, "madd", add>;
623 defm MSUB : MulAccum<1, "msub", sub>;
625 def : Pat<(i32 (mul GPR32:$Rn, GPR32:$Rm)),
626 (MADDWrrr GPR32:$Rn, GPR32:$Rm, WZR)>;
627 def : Pat<(i64 (mul GPR64:$Rn, GPR64:$Rm)),
628 (MADDXrrr GPR64:$Rn, GPR64:$Rm, XZR)>;
630 def : Pat<(i32 (ineg (mul GPR32:$Rn, GPR32:$Rm))),
631 (MSUBWrrr GPR32:$Rn, GPR32:$Rm, WZR)>;
632 def : Pat<(i64 (ineg (mul GPR64:$Rn, GPR64:$Rm))),
633 (MSUBXrrr GPR64:$Rn, GPR64:$Rm, XZR)>;
634 } // AddedComplexity = 7
636 let AddedComplexity = 5 in {
637 def SMADDLrrr : WideMulAccum<0, 0b001, "smaddl", add, sext>;
638 def SMSUBLrrr : WideMulAccum<1, 0b001, "smsubl", sub, sext>;
639 def UMADDLrrr : WideMulAccum<0, 0b101, "umaddl", add, zext>;
640 def UMSUBLrrr : WideMulAccum<1, 0b101, "umsubl", sub, zext>;
642 def : Pat<(i64 (mul (sext GPR32:$Rn), (sext GPR32:$Rm))),
643 (SMADDLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
644 def : Pat<(i64 (mul (zext GPR32:$Rn), (zext GPR32:$Rm))),
645 (UMADDLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
647 def : Pat<(i64 (ineg (mul (sext GPR32:$Rn), (sext GPR32:$Rm)))),
648 (SMSUBLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
649 def : Pat<(i64 (ineg (mul (zext GPR32:$Rn), (zext GPR32:$Rm)))),
650 (UMSUBLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
651 } // AddedComplexity = 5
653 def : MulAccumWAlias<"mul", MADDWrrr>;
654 def : MulAccumXAlias<"mul", MADDXrrr>;
655 def : MulAccumWAlias<"mneg", MSUBWrrr>;
656 def : MulAccumXAlias<"mneg", MSUBXrrr>;
657 def : WideMulAccumAlias<"smull", SMADDLrrr>;
658 def : WideMulAccumAlias<"smnegl", SMSUBLrrr>;
659 def : WideMulAccumAlias<"umull", UMADDLrrr>;
660 def : WideMulAccumAlias<"umnegl", UMSUBLrrr>;
663 def SMULHrr : MulHi<0b010, "smulh", mulhs>;
664 def UMULHrr : MulHi<0b110, "umulh", mulhu>;
667 def CRC32Brr : BaseCRC32<0, 0b00, 0, GPR32, int_arm64_crc32b, "crc32b">;
668 def CRC32Hrr : BaseCRC32<0, 0b01, 0, GPR32, int_arm64_crc32h, "crc32h">;
669 def CRC32Wrr : BaseCRC32<0, 0b10, 0, GPR32, int_arm64_crc32w, "crc32w">;
670 def CRC32Xrr : BaseCRC32<1, 0b11, 0, GPR64, int_arm64_crc32x, "crc32x">;
672 def CRC32CBrr : BaseCRC32<0, 0b00, 1, GPR32, int_arm64_crc32cb, "crc32cb">;
673 def CRC32CHrr : BaseCRC32<0, 0b01, 1, GPR32, int_arm64_crc32ch, "crc32ch">;
674 def CRC32CWrr : BaseCRC32<0, 0b10, 1, GPR32, int_arm64_crc32cw, "crc32cw">;
675 def CRC32CXrr : BaseCRC32<1, 0b11, 1, GPR64, int_arm64_crc32cx, "crc32cx">;
678 //===----------------------------------------------------------------------===//
679 // Logical instructions.
680 //===----------------------------------------------------------------------===//
683 defm ANDS : LogicalImmS<0b11, "ands", ARM64and_flag>;
684 defm AND : LogicalImm<0b00, "and", and>;
685 defm EOR : LogicalImm<0b10, "eor", xor>;
686 defm ORR : LogicalImm<0b01, "orr", or>;
688 // FIXME: these aliases *are* canonical sometimes (when movz can't be
689 // used). Actually, it seems to be working right now, but putting logical_immXX
690 // here is a bit dodgy on the AsmParser side too.
691 def : InstAlias<"mov $dst, $imm", (ORRWri GPR32sp:$dst, WZR,
692 logical_imm32:$imm), 0>;
693 def : InstAlias<"mov $dst, $imm", (ORRXri GPR64sp:$dst, XZR,
694 logical_imm64:$imm), 0>;
698 defm ANDS : LogicalRegS<0b11, 0, "ands", ARM64and_flag>;
699 defm BICS : LogicalRegS<0b11, 1, "bics",
700 BinOpFrag<(ARM64and_flag node:$LHS, (not node:$RHS))>>;
701 defm AND : LogicalReg<0b00, 0, "and", and>;
702 defm BIC : LogicalReg<0b00, 1, "bic",
703 BinOpFrag<(and node:$LHS, (not node:$RHS))>>;
704 defm EON : LogicalReg<0b10, 1, "eon",
705 BinOpFrag<(xor node:$LHS, (not node:$RHS))>>;
706 defm EOR : LogicalReg<0b10, 0, "eor", xor>;
707 defm ORN : LogicalReg<0b01, 1, "orn",
708 BinOpFrag<(or node:$LHS, (not node:$RHS))>>;
709 defm ORR : LogicalReg<0b01, 0, "orr", or>;
711 // FIXME: these aliases are named so that they get considered by TableGen before
712 // the already instantiated anonymous_ABC ones. Some kind of explicit priority
713 // system would be better.
714 def AA_MOVWr : InstAlias<"mov $dst, $src",
715 (ORRWrs GPR32:$dst, WZR, GPR32:$src, 0)>;
716 def AA_MOVXr : InstAlias<"mov $dst, $src",
717 (ORRXrs GPR64:$dst, XZR, GPR64:$src, 0)>;
719 def AA_MVNWr : InstAlias<"mvn $Wd, $Wm",
720 (ORNWrs GPR32:$Wd, WZR, GPR32:$Wm, 0)>;
721 def AA_MVNXr : InstAlias<"mvn $Xd, $Xm",
722 (ORNXrs GPR64:$Xd, XZR, GPR64:$Xm, 0)>;
724 def AA_MVNWrs : InstAlias<"mvn $Wd, $Wm$sh",
725 (ORNWrs GPR32:$Wd, WZR, GPR32:$Wm, logical_shift32:$sh)>;
726 def AA_MVNXrs : InstAlias<"mvn $Xd, $Xm$sh",
727 (ORNXrs GPR64:$Xd, XZR, GPR64:$Xm, logical_shift64:$sh)>;
729 def AA_TSTWri : InstAlias<"tst $src1, $src2",
730 (ANDSWri WZR, GPR32:$src1, logical_imm32:$src2)>;
731 def AA_TSTXri : InstAlias<"tst $src1, $src2",
732 (ANDSXri XZR, GPR64:$src1, logical_imm64:$src2)>;
734 def AA_TSTWr: InstAlias<"tst $src1, $src2",
735 (ANDSWrs WZR, GPR32:$src1, GPR32:$src2, 0)>;
736 def AA_TSTXr: InstAlias<"tst $src1, $src2",
737 (ANDSXrs XZR, GPR64:$src1, GPR64:$src2, 0)>;
739 def AB_TSTWrs : InstAlias<"tst $src1, $src2$sh",
740 (ANDSWrs WZR, GPR32:$src1, GPR32:$src2, logical_shift32:$sh)>;
741 def AB_TSTXrs : InstAlias<"tst $src1, $src2$sh",
742 (ANDSXrs XZR, GPR64:$src1, GPR64:$src2, logical_shift64:$sh)>;
745 def : Pat<(not GPR32:$Wm), (ORNWrr WZR, GPR32:$Wm)>;
746 def : Pat<(not GPR64:$Xm), (ORNXrr XZR, GPR64:$Xm)>;
749 //===----------------------------------------------------------------------===//
750 // One operand data processing instructions.
751 //===----------------------------------------------------------------------===//
753 defm CLS : OneOperandData<0b101, "cls">;
754 defm CLZ : OneOperandData<0b100, "clz", ctlz>;
755 defm RBIT : OneOperandData<0b000, "rbit">;
756 def REV16Wr : OneWRegData<0b001, "rev16",
757 UnOpFrag<(rotr (bswap node:$LHS), (i64 16))>>;
758 def REV16Xr : OneXRegData<0b001, "rev16", null_frag>;
760 def : Pat<(cttz GPR32:$Rn),
761 (CLZWr (RBITWr GPR32:$Rn))>;
762 def : Pat<(cttz GPR64:$Rn),
763 (CLZXr (RBITXr GPR64:$Rn))>;
764 def : Pat<(ctlz (or (shl (xor (sra GPR32:$Rn, (i64 31)), GPR32:$Rn), (i64 1)),
767 def : Pat<(ctlz (or (shl (xor (sra GPR64:$Rn, (i64 63)), GPR64:$Rn), (i64 1)),
771 // Unlike the other one operand instructions, the instructions with the "rev"
772 // mnemonic do *not* just different in the size bit, but actually use different
773 // opcode bits for the different sizes.
774 def REVWr : OneWRegData<0b010, "rev", bswap>;
775 def REVXr : OneXRegData<0b011, "rev", bswap>;
776 def REV32Xr : OneXRegData<0b010, "rev32",
777 UnOpFrag<(rotr (bswap node:$LHS), (i64 32))>>;
779 // The bswap commutes with the rotr so we want a pattern for both possible
781 def : Pat<(bswap (rotr GPR32:$Rn, (i64 16))), (REV16Wr GPR32:$Rn)>;
782 def : Pat<(bswap (rotr GPR64:$Rn, (i64 32))), (REV32Xr GPR64:$Rn)>;
784 //===----------------------------------------------------------------------===//
785 // Bitfield immediate extraction instruction.
786 //===----------------------------------------------------------------------===//
787 let neverHasSideEffects = 1 in
788 defm EXTR : ExtractImm<"extr">;
789 def : InstAlias<"ror $dst, $src, $shift",
790 (EXTRWrri GPR32:$dst, GPR32:$src, GPR32:$src, imm0_31:$shift)>;
791 def : InstAlias<"ror $dst, $src, $shift",
792 (EXTRXrri GPR64:$dst, GPR64:$src, GPR64:$src, imm0_63:$shift)>;
794 def : Pat<(rotr GPR32:$Rn, (i64 imm0_31:$imm)),
795 (EXTRWrri GPR32:$Rn, GPR32:$Rn, imm0_31:$imm)>;
796 def : Pat<(rotr GPR64:$Rn, (i64 imm0_63:$imm)),
797 (EXTRXrri GPR64:$Rn, GPR64:$Rn, imm0_63:$imm)>;
799 //===----------------------------------------------------------------------===//
800 // Other bitfield immediate instructions.
801 //===----------------------------------------------------------------------===//
802 let neverHasSideEffects = 1 in {
803 defm BFM : BitfieldImmWith2RegArgs<0b01, "bfm">;
804 defm SBFM : BitfieldImm<0b00, "sbfm">;
805 defm UBFM : BitfieldImm<0b10, "ubfm">;
808 def i32shift_a : Operand<i64>, SDNodeXForm<imm, [{
809 uint64_t enc = (32 - N->getZExtValue()) & 0x1f;
810 return CurDAG->getTargetConstant(enc, MVT::i64);
813 def i32shift_b : Operand<i64>, SDNodeXForm<imm, [{
814 uint64_t enc = 31 - N->getZExtValue();
815 return CurDAG->getTargetConstant(enc, MVT::i64);
818 // min(7, 31 - shift_amt)
819 def i32shift_sext_i8 : Operand<i64>, SDNodeXForm<imm, [{
820 uint64_t enc = 31 - N->getZExtValue();
821 enc = enc > 7 ? 7 : enc;
822 return CurDAG->getTargetConstant(enc, MVT::i64);
825 // min(15, 31 - shift_amt)
826 def i32shift_sext_i16 : Operand<i64>, SDNodeXForm<imm, [{
827 uint64_t enc = 31 - N->getZExtValue();
828 enc = enc > 15 ? 15 : enc;
829 return CurDAG->getTargetConstant(enc, MVT::i64);
832 def i64shift_a : Operand<i64>, SDNodeXForm<imm, [{
833 uint64_t enc = (64 - N->getZExtValue()) & 0x3f;
834 return CurDAG->getTargetConstant(enc, MVT::i64);
837 def i64shift_b : Operand<i64>, SDNodeXForm<imm, [{
838 uint64_t enc = 63 - N->getZExtValue();
839 return CurDAG->getTargetConstant(enc, MVT::i64);
842 // min(7, 63 - shift_amt)
843 def i64shift_sext_i8 : Operand<i64>, SDNodeXForm<imm, [{
844 uint64_t enc = 63 - N->getZExtValue();
845 enc = enc > 7 ? 7 : enc;
846 return CurDAG->getTargetConstant(enc, MVT::i64);
849 // min(15, 63 - shift_amt)
850 def i64shift_sext_i16 : Operand<i64>, SDNodeXForm<imm, [{
851 uint64_t enc = 63 - N->getZExtValue();
852 enc = enc > 15 ? 15 : enc;
853 return CurDAG->getTargetConstant(enc, MVT::i64);
856 // min(31, 63 - shift_amt)
857 def i64shift_sext_i32 : Operand<i64>, SDNodeXForm<imm, [{
858 uint64_t enc = 63 - N->getZExtValue();
859 enc = enc > 31 ? 31 : enc;
860 return CurDAG->getTargetConstant(enc, MVT::i64);
863 def : Pat<(shl GPR32:$Rn, (i64 imm0_31:$imm)),
864 (UBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
865 (i64 (i32shift_b imm0_31:$imm)))>;
866 def : Pat<(shl GPR64:$Rn, (i64 imm0_63:$imm)),
867 (UBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
868 (i64 (i64shift_b imm0_63:$imm)))>;
870 let AddedComplexity = 10 in {
871 def : Pat<(sra GPR32:$Rn, (i64 imm0_31:$imm)),
872 (SBFMWri GPR32:$Rn, imm0_31:$imm, 31)>;
873 def : Pat<(sra GPR64:$Rn, (i64 imm0_63:$imm)),
874 (SBFMXri GPR64:$Rn, imm0_63:$imm, 63)>;
877 def : InstAlias<"asr $dst, $src, $shift",
878 (SBFMWri GPR32:$dst, GPR32:$src, imm0_31:$shift, 31)>;
879 def : InstAlias<"asr $dst, $src, $shift",
880 (SBFMXri GPR64:$dst, GPR64:$src, imm0_63:$shift, 63)>;
881 def : InstAlias<"sxtb $dst, $src", (SBFMWri GPR32:$dst, GPR32:$src, 0, 7)>;
882 def : InstAlias<"sxtb $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 7)>;
883 def : InstAlias<"sxth $dst, $src", (SBFMWri GPR32:$dst, GPR32:$src, 0, 15)>;
884 def : InstAlias<"sxth $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 15)>;
885 def : InstAlias<"sxtw $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 31)>;
887 def : Pat<(srl GPR32:$Rn, (i64 imm0_31:$imm)),
888 (UBFMWri GPR32:$Rn, imm0_31:$imm, 31)>;
889 def : Pat<(srl GPR64:$Rn, (i64 imm0_63:$imm)),
890 (UBFMXri GPR64:$Rn, imm0_63:$imm, 63)>;
892 def : InstAlias<"lsr $dst, $src, $shift",
893 (UBFMWri GPR32:$dst, GPR32:$src, imm0_31:$shift, 31)>;
894 def : InstAlias<"lsr $dst, $src, $shift",
895 (UBFMXri GPR64:$dst, GPR64:$src, imm0_63:$shift, 63)>;
896 def : InstAlias<"uxtb $dst, $src", (UBFMWri GPR32:$dst, GPR32:$src, 0, 7)>;
897 def : InstAlias<"uxtb $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 7)>;
898 def : InstAlias<"uxth $dst, $src", (UBFMWri GPR32:$dst, GPR32:$src, 0, 15)>;
899 def : InstAlias<"uxth $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 15)>;
900 def : InstAlias<"uxtw $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 31)>;
902 //===----------------------------------------------------------------------===//
903 // Conditionally set flags instructions.
904 //===----------------------------------------------------------------------===//
905 defm CCMN : CondSetFlagsImm<0, "ccmn">;
906 defm CCMP : CondSetFlagsImm<1, "ccmp">;
908 defm CCMN : CondSetFlagsReg<0, "ccmn">;
909 defm CCMP : CondSetFlagsReg<1, "ccmp">;
911 //===----------------------------------------------------------------------===//
912 // Conditional select instructions.
913 //===----------------------------------------------------------------------===//
914 defm CSEL : CondSelect<0, 0b00, "csel">;
916 def inc : PatFrag<(ops node:$in), (add node:$in, 1)>;
917 defm CSINC : CondSelectOp<0, 0b01, "csinc", inc>;
918 defm CSINV : CondSelectOp<1, 0b00, "csinv", not>;
919 defm CSNEG : CondSelectOp<1, 0b01, "csneg", ineg>;
921 def : Pat<(ARM64csinv GPR32:$tval, GPR32:$fval, (i32 imm:$cc), NZCV),
922 (CSINVWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
923 def : Pat<(ARM64csinv GPR64:$tval, GPR64:$fval, (i32 imm:$cc), NZCV),
924 (CSINVXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
925 def : Pat<(ARM64csneg GPR32:$tval, GPR32:$fval, (i32 imm:$cc), NZCV),
926 (CSNEGWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
927 def : Pat<(ARM64csneg GPR64:$tval, GPR64:$fval, (i32 imm:$cc), NZCV),
928 (CSNEGXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
929 def : Pat<(ARM64csinc GPR32:$tval, GPR32:$fval, (i32 imm:$cc), NZCV),
930 (CSINCWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
931 def : Pat<(ARM64csinc GPR64:$tval, GPR64:$fval, (i32 imm:$cc), NZCV),
932 (CSINCXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
934 def : Pat<(ARM64csel (i32 0), (i32 1), (i32 imm:$cc), NZCV),
935 (CSINCWr WZR, WZR, (i32 imm:$cc))>;
936 def : Pat<(ARM64csel (i64 0), (i64 1), (i32 imm:$cc), NZCV),
937 (CSINCXr XZR, XZR, (i32 imm:$cc))>;
938 def : Pat<(ARM64csel (i32 0), (i32 -1), (i32 imm:$cc), NZCV),
939 (CSINVWr WZR, WZR, (i32 imm:$cc))>;
940 def : Pat<(ARM64csel (i64 0), (i64 -1), (i32 imm:$cc), NZCV),
941 (CSINVXr XZR, XZR, (i32 imm:$cc))>;
943 // The inverse of the condition code from the alias instruction is what is used
944 // in the aliased instruction. The parser all ready inverts the condition code
945 // for these aliases.
946 def : InstAlias<"cset $dst, $cc",
947 (CSINCWr GPR32:$dst, WZR, WZR, inv_ccode:$cc)>;
948 def : InstAlias<"cset $dst, $cc",
949 (CSINCXr GPR64:$dst, XZR, XZR, inv_ccode:$cc)>;
951 def : InstAlias<"csetm $dst, $cc",
952 (CSINVWr GPR32:$dst, WZR, WZR, inv_ccode:$cc)>;
953 def : InstAlias<"csetm $dst, $cc",
954 (CSINVXr GPR64:$dst, XZR, XZR, inv_ccode:$cc)>;
956 def : InstAlias<"cinc $dst, $src, $cc",
957 (CSINCWr GPR32:$dst, GPR32:$src, GPR32:$src, inv_ccode:$cc)>;
958 def : InstAlias<"cinc $dst, $src, $cc",
959 (CSINCXr GPR64:$dst, GPR64:$src, GPR64:$src, inv_ccode:$cc)>;
961 def : InstAlias<"cinv $dst, $src, $cc",
962 (CSINVWr GPR32:$dst, GPR32:$src, GPR32:$src, inv_ccode:$cc)>;
963 def : InstAlias<"cinv $dst, $src, $cc",
964 (CSINVXr GPR64:$dst, GPR64:$src, GPR64:$src, inv_ccode:$cc)>;
966 def : InstAlias<"cneg $dst, $src, $cc",
967 (CSNEGWr GPR32:$dst, GPR32:$src, GPR32:$src, inv_ccode:$cc)>;
968 def : InstAlias<"cneg $dst, $src, $cc",
969 (CSNEGXr GPR64:$dst, GPR64:$src, GPR64:$src, inv_ccode:$cc)>;
971 //===----------------------------------------------------------------------===//
972 // PC-relative instructions.
973 //===----------------------------------------------------------------------===//
974 let isReMaterializable = 1 in {
975 let neverHasSideEffects = 1, mayStore = 0, mayLoad = 0 in {
976 def ADR : ADRI<0, "adr", adrlabel, []>;
977 } // neverHasSideEffects = 1
979 def ADRP : ADRI<1, "adrp", adrplabel,
980 [(set GPR64:$Xd, (ARM64adrp tglobaladdr:$label))]>;
981 } // isReMaterializable = 1
983 // page address of a constant pool entry, block address
984 def : Pat<(ARM64adrp tconstpool:$cp), (ADRP tconstpool:$cp)>;
985 def : Pat<(ARM64adrp tblockaddress:$cp), (ADRP tblockaddress:$cp)>;
987 //===----------------------------------------------------------------------===//
988 // Unconditional branch (register) instructions.
989 //===----------------------------------------------------------------------===//
991 let isReturn = 1, isTerminator = 1, isBarrier = 1 in {
992 def RET : BranchReg<0b0010, "ret", []>;
993 def DRPS : SpecialReturn<0b0101, "drps">;
994 def ERET : SpecialReturn<0b0100, "eret">;
995 } // isReturn = 1, isTerminator = 1, isBarrier = 1
997 // Default to the LR register.
998 def : InstAlias<"ret", (RET LR)>;
1000 let isCall = 1, Defs = [LR], Uses = [SP] in {
1001 def BLR : BranchReg<0b0001, "blr", [(ARM64call GPR64:$Rn)]>;
1004 let isBranch = 1, isTerminator = 1, isBarrier = 1, isIndirectBranch = 1 in {
1005 def BR : BranchReg<0b0000, "br", [(brind GPR64:$Rn)]>;
1006 } // isBranch, isTerminator, isBarrier, isIndirectBranch
1008 // Create a separate pseudo-instruction for codegen to use so that we don't
1009 // flag lr as used in every function. It'll be restored before the RET by the
1010 // epilogue if it's legitimately used.
1011 def RET_ReallyLR : Pseudo<(outs), (ins), [(ARM64retflag)]> {
1012 let isTerminator = 1;
1017 // This is a directive-like pseudo-instruction. The purpose is to insert an
1018 // R_AARCH64_TLSDESC_CALL relocation at the offset of the following instruction
1019 // (which in the usual case is a BLR).
1020 let hasSideEffects = 1 in
1021 def TLSDESCCALL : Pseudo<(outs), (ins i64imm:$sym), []> {
1022 let AsmString = ".tlsdesccall $sym";
1025 // Pseudo-instruction representing a BLR with attached TLSDESC relocation. It
1026 // gets expanded to two MCInsts during lowering.
1027 let isCall = 1, Defs = [LR] in
1029 : Pseudo<(outs), (ins GPR64:$dest, i64imm:$sym),
1030 [(ARM64tlsdesc_call GPR64:$dest, tglobaltlsaddr:$sym)]>;
1032 def : Pat<(ARM64tlsdesc_call GPR64:$dest, texternalsym:$sym),
1033 (TLSDESC_BLR GPR64:$dest, texternalsym:$sym)>;
1034 //===----------------------------------------------------------------------===//
1035 // Conditional branch (immediate) instruction.
1036 //===----------------------------------------------------------------------===//
1037 def Bcc : BranchCond;
1039 //===----------------------------------------------------------------------===//
1040 // Compare-and-branch instructions.
1041 //===----------------------------------------------------------------------===//
1042 defm CBZ : CmpBranch<0, "cbz", ARM64cbz>;
1043 defm CBNZ : CmpBranch<1, "cbnz", ARM64cbnz>;
1045 //===----------------------------------------------------------------------===//
1046 // Test-bit-and-branch instructions.
1047 //===----------------------------------------------------------------------===//
1048 defm TBZ : TestBranch<0, "tbz", ARM64tbz>;
1049 defm TBNZ : TestBranch<1, "tbnz", ARM64tbnz>;
1051 //===----------------------------------------------------------------------===//
1052 // Unconditional branch (immediate) instructions.
1053 //===----------------------------------------------------------------------===//
1054 let isBranch = 1, isTerminator = 1, isBarrier = 1 in {
1055 def B : BranchImm<0, "b", [(br bb:$addr)]>;
1056 } // isBranch, isTerminator, isBarrier
1058 let isCall = 1, Defs = [LR], Uses = [SP] in {
1059 def BL : CallImm<1, "bl", [(ARM64call tglobaladdr:$addr)]>;
1061 def : Pat<(ARM64call texternalsym:$func), (BL texternalsym:$func)>;
1063 //===----------------------------------------------------------------------===//
1064 // Exception generation instructions.
1065 //===----------------------------------------------------------------------===//
1066 def BRK : ExceptionGeneration<0b001, 0b00, "brk">;
1067 def DCPS1 : ExceptionGeneration<0b101, 0b01, "dcps1">;
1068 def DCPS2 : ExceptionGeneration<0b101, 0b10, "dcps2">;
1069 def DCPS3 : ExceptionGeneration<0b101, 0b11, "dcps3">;
1070 def HLT : ExceptionGeneration<0b010, 0b00, "hlt">;
1071 def HVC : ExceptionGeneration<0b000, 0b10, "hvc">;
1072 def SMC : ExceptionGeneration<0b000, 0b11, "smc">;
1073 def SVC : ExceptionGeneration<0b000, 0b01, "svc">;
1075 // DCPSn defaults to an immediate operand of zero if unspecified.
1076 def : InstAlias<"dcps1", (DCPS1 0)>;
1077 def : InstAlias<"dcps2", (DCPS2 0)>;
1078 def : InstAlias<"dcps3", (DCPS3 0)>;
1080 //===----------------------------------------------------------------------===//
1081 // Load instructions.
1082 //===----------------------------------------------------------------------===//
1084 // Pair (indexed, offset)
1085 def LDPWi : LoadPairOffset<0b00, 0, GPR32, am_indexed32simm7, "ldp">;
1086 def LDPXi : LoadPairOffset<0b10, 0, GPR64, am_indexed64simm7, "ldp">;
1087 def LDPSi : LoadPairOffset<0b00, 1, FPR32, am_indexed32simm7, "ldp">;
1088 def LDPDi : LoadPairOffset<0b01, 1, FPR64, am_indexed64simm7, "ldp">;
1089 def LDPQi : LoadPairOffset<0b10, 1, FPR128, am_indexed128simm7, "ldp">;
1091 def LDPSWi : LoadPairOffset<0b01, 0, GPR64, am_indexed32simm7, "ldpsw">;
1093 // Pair (pre-indexed)
1094 def LDPWpre : LoadPairPreIdx<0b00, 0, GPR32, am_indexed32simm7_wb, "ldp">;
1095 def LDPXpre : LoadPairPreIdx<0b10, 0, GPR64, am_indexed64simm7_wb, "ldp">;
1096 def LDPSpre : LoadPairPreIdx<0b00, 1, FPR32, am_indexed32simm7_wb, "ldp">;
1097 def LDPDpre : LoadPairPreIdx<0b01, 1, FPR64, am_indexed64simm7_wb, "ldp">;
1098 def LDPQpre : LoadPairPreIdx<0b10, 1, FPR128, am_indexed128simm7_wb, "ldp">;
1100 def LDPSWpre : LoadPairPreIdx<0b01, 0, GPR64, am_indexed32simm7_wb, "ldpsw">;
1102 // Pair (post-indexed)
1103 def LDPWpost : LoadPairPostIdx<0b00, 0, GPR32, simm7s4, "ldp">;
1104 def LDPXpost : LoadPairPostIdx<0b10, 0, GPR64, simm7s8, "ldp">;
1105 def LDPSpost : LoadPairPostIdx<0b00, 1, FPR32, simm7s4, "ldp">;
1106 def LDPDpost : LoadPairPostIdx<0b01, 1, FPR64, simm7s8, "ldp">;
1107 def LDPQpost : LoadPairPostIdx<0b10, 1, FPR128, simm7s16, "ldp">;
1109 def LDPSWpost : LoadPairPostIdx<0b01, 0, GPR64, simm7s4, "ldpsw">;
1112 // Pair (no allocate)
1113 def LDNPWi : LoadPairNoAlloc<0b00, 0, GPR32, am_indexed32simm7, "ldnp">;
1114 def LDNPXi : LoadPairNoAlloc<0b10, 0, GPR64, am_indexed64simm7, "ldnp">;
1115 def LDNPSi : LoadPairNoAlloc<0b00, 1, FPR32, am_indexed32simm7, "ldnp">;
1116 def LDNPDi : LoadPairNoAlloc<0b01, 1, FPR64, am_indexed64simm7, "ldnp">;
1117 def LDNPQi : LoadPairNoAlloc<0b10, 1, FPR128, am_indexed128simm7, "ldnp">;
1120 // (register offset)
1123 let AddedComplexity = 10 in {
1125 def LDRBBro : Load8RO<0b00, 0, 0b01, GPR32, "ldrb",
1126 [(set GPR32:$Rt, (zextloadi8 ro_indexed8:$addr))]>;
1127 def LDRHHro : Load16RO<0b01, 0, 0b01, GPR32, "ldrh",
1128 [(set GPR32:$Rt, (zextloadi16 ro_indexed16:$addr))]>;
1129 def LDRWro : Load32RO<0b10, 0, 0b01, GPR32, "ldr",
1130 [(set GPR32:$Rt, (load ro_indexed32:$addr))]>;
1131 def LDRXro : Load64RO<0b11, 0, 0b01, GPR64, "ldr",
1132 [(set GPR64:$Rt, (load ro_indexed64:$addr))]>;
1135 def LDRBro : Load8RO<0b00, 1, 0b01, FPR8, "ldr",
1136 [(set FPR8:$Rt, (load ro_indexed8:$addr))]>;
1137 def LDRHro : Load16RO<0b01, 1, 0b01, FPR16, "ldr",
1138 [(set (f16 FPR16:$Rt), (load ro_indexed16:$addr))]>;
1139 def LDRSro : Load32RO<0b10, 1, 0b01, FPR32, "ldr",
1140 [(set (f32 FPR32:$Rt), (load ro_indexed32:$addr))]>;
1141 def LDRDro : Load64RO<0b11, 1, 0b01, FPR64, "ldr",
1142 [(set (f64 FPR64:$Rt), (load ro_indexed64:$addr))]>;
1143 def LDRQro : Load128RO<0b00, 1, 0b11, FPR128, "ldr", []> {
1147 // For regular load, we do not have any alignment requirement.
1148 // Thus, it is safe to directly map the vector loads with interesting
1149 // addressing modes.
1150 // FIXME: We could do the same for bitconvert to floating point vectors.
1151 def : Pat <(v8i8 (scalar_to_vector (i32 (extloadi8 ro_indexed8:$addr)))),
1152 (INSERT_SUBREG (v8i8 (IMPLICIT_DEF)),
1153 (LDRBro ro_indexed8:$addr), bsub)>;
1154 def : Pat <(v16i8 (scalar_to_vector (i32 (extloadi8 ro_indexed8:$addr)))),
1155 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
1156 (LDRBro ro_indexed8:$addr), bsub)>;
1157 def : Pat <(v4i16 (scalar_to_vector (i32 (extloadi16 ro_indexed16:$addr)))),
1158 (INSERT_SUBREG (v4i16 (IMPLICIT_DEF)),
1159 (LDRHro ro_indexed16:$addr), hsub)>;
1160 def : Pat <(v8i16 (scalar_to_vector (i32 (extloadi16 ro_indexed16:$addr)))),
1161 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
1162 (LDRHro ro_indexed16:$addr), hsub)>;
1163 def : Pat <(v2i32 (scalar_to_vector (i32 (load ro_indexed32:$addr)))),
1164 (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)),
1165 (LDRSro ro_indexed32:$addr), ssub)>;
1166 def : Pat <(v4i32 (scalar_to_vector (i32 (load ro_indexed32:$addr)))),
1167 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
1168 (LDRSro ro_indexed32:$addr), ssub)>;
1169 def : Pat <(v1i64 (scalar_to_vector (i64 (load ro_indexed64:$addr)))),
1170 (LDRDro ro_indexed64:$addr)>;
1171 def : Pat <(v2i64 (scalar_to_vector (i64 (load ro_indexed64:$addr)))),
1172 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
1173 (LDRDro ro_indexed64:$addr), dsub)>;
1175 // Match all load 64 bits width whose type is compatible with FPR64
1176 let Predicates = [IsLE] in {
1177 // We must do vector loads with LD1 in big-endian.
1178 def : Pat<(v2f32 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1179 def : Pat<(v8i8 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1180 def : Pat<(v4i16 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1181 def : Pat<(v2i32 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1183 def : Pat<(v1f64 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1184 def : Pat<(v1i64 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1186 // Match all load 128 bits width whose type is compatible with FPR128
1187 let Predicates = [IsLE] in {
1188 // We must do vector loads with LD1 in big-endian.
1189 def : Pat<(v4f32 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1190 def : Pat<(v2f64 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1191 def : Pat<(v16i8 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1192 def : Pat<(v8i16 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1193 def : Pat<(v4i32 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1194 def : Pat<(v2i64 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1196 def : Pat<(f128 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1198 // Load sign-extended half-word
1199 def LDRSHWro : Load16RO<0b01, 0, 0b11, GPR32, "ldrsh",
1200 [(set GPR32:$Rt, (sextloadi16 ro_indexed16:$addr))]>;
1201 def LDRSHXro : Load16RO<0b01, 0, 0b10, GPR64, "ldrsh",
1202 [(set GPR64:$Rt, (sextloadi16 ro_indexed16:$addr))]>;
1204 // Load sign-extended byte
1205 def LDRSBWro : Load8RO<0b00, 0, 0b11, GPR32, "ldrsb",
1206 [(set GPR32:$Rt, (sextloadi8 ro_indexed8:$addr))]>;
1207 def LDRSBXro : Load8RO<0b00, 0, 0b10, GPR64, "ldrsb",
1208 [(set GPR64:$Rt, (sextloadi8 ro_indexed8:$addr))]>;
1210 // Load sign-extended word
1211 def LDRSWro : Load32RO<0b10, 0, 0b10, GPR64, "ldrsw",
1212 [(set GPR64:$Rt, (sextloadi32 ro_indexed32:$addr))]>;
1215 def PRFMro : PrefetchRO<0b11, 0, 0b10, "prfm",
1216 [(ARM64Prefetch imm:$Rt, ro_indexed64:$addr)]>;
1219 def : Pat<(i64 (zextloadi8 ro_indexed8:$addr)),
1220 (SUBREG_TO_REG (i64 0), (LDRBBro ro_indexed8:$addr), sub_32)>;
1221 def : Pat<(i64 (zextloadi16 ro_indexed16:$addr)),
1222 (SUBREG_TO_REG (i64 0), (LDRHHro ro_indexed16:$addr), sub_32)>;
1223 def : Pat<(i64 (zextloadi32 ro_indexed32:$addr)),
1224 (SUBREG_TO_REG (i64 0), (LDRWro ro_indexed32:$addr), sub_32)>;
1226 // zextloadi1 -> zextloadi8
1227 def : Pat<(i32 (zextloadi1 ro_indexed8:$addr)), (LDRBBro ro_indexed8:$addr)>;
1228 def : Pat<(i64 (zextloadi1 ro_indexed8:$addr)),
1229 (SUBREG_TO_REG (i64 0), (LDRBBro ro_indexed8:$addr), sub_32)>;
1231 // extload -> zextload
1232 def : Pat<(i32 (extloadi16 ro_indexed16:$addr)), (LDRHHro ro_indexed16:$addr)>;
1233 def : Pat<(i32 (extloadi8 ro_indexed8:$addr)), (LDRBBro ro_indexed8:$addr)>;
1234 def : Pat<(i32 (extloadi1 ro_indexed8:$addr)), (LDRBBro ro_indexed8:$addr)>;
1235 def : Pat<(i64 (extloadi32 ro_indexed32:$addr)),
1236 (SUBREG_TO_REG (i64 0), (LDRWro ro_indexed32:$addr), sub_32)>;
1237 def : Pat<(i64 (extloadi16 ro_indexed16:$addr)),
1238 (SUBREG_TO_REG (i64 0), (LDRHHro ro_indexed16:$addr), sub_32)>;
1239 def : Pat<(i64 (extloadi8 ro_indexed8:$addr)),
1240 (SUBREG_TO_REG (i64 0), (LDRBBro ro_indexed8:$addr), sub_32)>;
1241 def : Pat<(i64 (extloadi1 ro_indexed8:$addr)),
1242 (SUBREG_TO_REG (i64 0), (LDRBBro ro_indexed8:$addr), sub_32)>;
1244 } // AddedComplexity = 10
1247 // (unsigned immediate)
1249 def LDRXui : LoadUI<0b11, 0, 0b01, GPR64, am_indexed64, "ldr",
1250 [(set GPR64:$Rt, (load am_indexed64:$addr))]>;
1251 def LDRWui : LoadUI<0b10, 0, 0b01, GPR32, am_indexed32, "ldr",
1252 [(set GPR32:$Rt, (load am_indexed32:$addr))]>;
1253 def LDRBui : LoadUI<0b00, 1, 0b01, FPR8, am_indexed8, "ldr",
1254 [(set FPR8:$Rt, (load am_indexed8:$addr))]>;
1255 def LDRHui : LoadUI<0b01, 1, 0b01, FPR16, am_indexed16, "ldr",
1256 [(set (f16 FPR16:$Rt), (load am_indexed16:$addr))]>;
1257 def LDRSui : LoadUI<0b10, 1, 0b01, FPR32, am_indexed32, "ldr",
1258 [(set (f32 FPR32:$Rt), (load am_indexed32:$addr))]>;
1259 def LDRDui : LoadUI<0b11, 1, 0b01, FPR64, am_indexed64, "ldr",
1260 [(set (f64 FPR64:$Rt), (load am_indexed64:$addr))]>;
1261 def LDRQui : LoadUI<0b00, 1, 0b11, FPR128, am_indexed128, "ldr",
1262 [(set (f128 FPR128:$Rt), (load am_indexed128:$addr))]>;
1264 // For regular load, we do not have any alignment requirement.
1265 // Thus, it is safe to directly map the vector loads with interesting
1266 // addressing modes.
1267 // FIXME: We could do the same for bitconvert to floating point vectors.
1268 def : Pat <(v8i8 (scalar_to_vector (i32 (extloadi8 am_indexed8:$addr)))),
1269 (INSERT_SUBREG (v8i8 (IMPLICIT_DEF)),
1270 (LDRBui am_indexed8:$addr), bsub)>;
1271 def : Pat <(v16i8 (scalar_to_vector (i32 (extloadi8 am_indexed8:$addr)))),
1272 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
1273 (LDRBui am_indexed8:$addr), bsub)>;
1274 def : Pat <(v4i16 (scalar_to_vector (i32 (extloadi16 am_indexed16:$addr)))),
1275 (INSERT_SUBREG (v4i16 (IMPLICIT_DEF)),
1276 (LDRHui am_indexed16:$addr), hsub)>;
1277 def : Pat <(v8i16 (scalar_to_vector (i32 (extloadi16 am_indexed16:$addr)))),
1278 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
1279 (LDRHui am_indexed16:$addr), hsub)>;
1280 def : Pat <(v2i32 (scalar_to_vector (i32 (load am_indexed32:$addr)))),
1281 (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)),
1282 (LDRSui am_indexed32:$addr), ssub)>;
1283 def : Pat <(v4i32 (scalar_to_vector (i32 (load am_indexed32:$addr)))),
1284 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
1285 (LDRSui am_indexed32:$addr), ssub)>;
1286 def : Pat <(v1i64 (scalar_to_vector (i64 (load am_indexed64:$addr)))),
1287 (LDRDui am_indexed64:$addr)>;
1288 def : Pat <(v2i64 (scalar_to_vector (i64 (load am_indexed64:$addr)))),
1289 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
1290 (LDRDui am_indexed64:$addr), dsub)>;
1292 // Match all load 64 bits width whose type is compatible with FPR64
1293 let Predicates = [IsLE] in {
1294 // We must use LD1 to perform vector loads in big-endian.
1295 def : Pat<(v2f32 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1296 def : Pat<(v8i8 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1297 def : Pat<(v4i16 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1298 def : Pat<(v2i32 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1300 def : Pat<(v1f64 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1301 def : Pat<(v1i64 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1303 // Match all load 128 bits width whose type is compatible with FPR128
1304 let Predicates = [IsLE] in {
1305 // We must use LD1 to perform vector loads in big-endian.
1306 def : Pat<(v4f32 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1307 def : Pat<(v2f64 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1308 def : Pat<(v16i8 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1309 def : Pat<(v8i16 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1310 def : Pat<(v4i32 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1311 def : Pat<(v2i64 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1313 def : Pat<(f128 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1315 def LDRHHui : LoadUI<0b01, 0, 0b01, GPR32, am_indexed16, "ldrh",
1316 [(set GPR32:$Rt, (zextloadi16 am_indexed16:$addr))]>;
1317 def LDRBBui : LoadUI<0b00, 0, 0b01, GPR32, am_indexed8, "ldrb",
1318 [(set GPR32:$Rt, (zextloadi8 am_indexed8:$addr))]>;
1320 def : Pat<(i64 (zextloadi8 am_indexed8:$addr)),
1321 (SUBREG_TO_REG (i64 0), (LDRBBui am_indexed8:$addr), sub_32)>;
1322 def : Pat<(i64 (zextloadi16 am_indexed16:$addr)),
1323 (SUBREG_TO_REG (i64 0), (LDRHHui am_indexed16:$addr), sub_32)>;
1325 // zextloadi1 -> zextloadi8
1326 def : Pat<(i32 (zextloadi1 am_indexed8:$addr)), (LDRBBui am_indexed8:$addr)>;
1327 def : Pat<(i64 (zextloadi1 am_indexed8:$addr)),
1328 (SUBREG_TO_REG (i64 0), (LDRBBui am_indexed8:$addr), sub_32)>;
1330 // extload -> zextload
1331 def : Pat<(i32 (extloadi16 am_indexed16:$addr)), (LDRHHui am_indexed16:$addr)>;
1332 def : Pat<(i32 (extloadi8 am_indexed8:$addr)), (LDRBBui am_indexed8:$addr)>;
1333 def : Pat<(i32 (extloadi1 am_indexed8:$addr)), (LDRBBui am_indexed8:$addr)>;
1334 def : Pat<(i64 (extloadi32 am_indexed32:$addr)),
1335 (SUBREG_TO_REG (i64 0), (LDRWui am_indexed32:$addr), sub_32)>;
1336 def : Pat<(i64 (extloadi16 am_indexed16:$addr)),
1337 (SUBREG_TO_REG (i64 0), (LDRHHui am_indexed16:$addr), sub_32)>;
1338 def : Pat<(i64 (extloadi8 am_indexed8:$addr)),
1339 (SUBREG_TO_REG (i64 0), (LDRBBui am_indexed8:$addr), sub_32)>;
1340 def : Pat<(i64 (extloadi1 am_indexed8:$addr)),
1341 (SUBREG_TO_REG (i64 0), (LDRBBui am_indexed8:$addr), sub_32)>;
1343 // load sign-extended half-word
1344 def LDRSHWui : LoadUI<0b01, 0, 0b11, GPR32, am_indexed16, "ldrsh",
1345 [(set GPR32:$Rt, (sextloadi16 am_indexed16:$addr))]>;
1346 def LDRSHXui : LoadUI<0b01, 0, 0b10, GPR64, am_indexed16, "ldrsh",
1347 [(set GPR64:$Rt, (sextloadi16 am_indexed16:$addr))]>;
1349 // load sign-extended byte
1350 def LDRSBWui : LoadUI<0b00, 0, 0b11, GPR32, am_indexed8, "ldrsb",
1351 [(set GPR32:$Rt, (sextloadi8 am_indexed8:$addr))]>;
1352 def LDRSBXui : LoadUI<0b00, 0, 0b10, GPR64, am_indexed8, "ldrsb",
1353 [(set GPR64:$Rt, (sextloadi8 am_indexed8:$addr))]>;
1355 // load sign-extended word
1356 def LDRSWui : LoadUI<0b10, 0, 0b10, GPR64, am_indexed32, "ldrsw",
1357 [(set GPR64:$Rt, (sextloadi32 am_indexed32:$addr))]>;
1359 // load zero-extended word
1360 def : Pat<(i64 (zextloadi32 am_indexed32:$addr)),
1361 (SUBREG_TO_REG (i64 0), (LDRWui am_indexed32:$addr), sub_32)>;
1364 def PRFMui : PrefetchUI<0b11, 0, 0b10, "prfm",
1365 [(ARM64Prefetch imm:$Rt, am_indexed64:$addr)]>;
1369 def LDRWl : LoadLiteral<0b00, 0, GPR32, "ldr">;
1370 def LDRXl : LoadLiteral<0b01, 0, GPR64, "ldr">;
1371 def LDRSl : LoadLiteral<0b00, 1, FPR32, "ldr">;
1372 def LDRDl : LoadLiteral<0b01, 1, FPR64, "ldr">;
1373 def LDRQl : LoadLiteral<0b10, 1, FPR128, "ldr">;
1375 // load sign-extended word
1376 def LDRSWl : LoadLiteral<0b10, 0, GPR64, "ldrsw">;
1379 def PRFMl : PrefetchLiteral<0b11, 0, "prfm", []>;
1380 // [(ARM64Prefetch imm:$Rt, tglobaladdr:$label)]>;
1383 // (unscaled immediate)
1384 def LDURXi : LoadUnscaled<0b11, 0, 0b01, GPR64, am_unscaled64, "ldur",
1385 [(set GPR64:$Rt, (load am_unscaled64:$addr))]>;
1386 def LDURWi : LoadUnscaled<0b10, 0, 0b01, GPR32, am_unscaled32, "ldur",
1387 [(set GPR32:$Rt, (load am_unscaled32:$addr))]>;
1388 def LDURBi : LoadUnscaled<0b00, 1, 0b01, FPR8, am_unscaled8, "ldur",
1389 [(set FPR8:$Rt, (load am_unscaled8:$addr))]>;
1390 def LDURHi : LoadUnscaled<0b01, 1, 0b01, FPR16, am_unscaled16, "ldur",
1391 [(set (f16 FPR16:$Rt), (load am_unscaled16:$addr))]>;
1392 def LDURSi : LoadUnscaled<0b10, 1, 0b01, FPR32, am_unscaled32, "ldur",
1393 [(set (f32 FPR32:$Rt), (load am_unscaled32:$addr))]>;
1394 def LDURDi : LoadUnscaled<0b11, 1, 0b01, FPR64, am_unscaled64, "ldur",
1395 [(set (f64 FPR64:$Rt), (load am_unscaled64:$addr))]>;
1396 def LDURQi : LoadUnscaled<0b00, 1, 0b11, FPR128, am_unscaled128, "ldur",
1397 [(set (f128 FPR128:$Rt), (load am_unscaled128:$addr))]>;
1400 : LoadUnscaled<0b01, 0, 0b01, GPR32, am_unscaled16, "ldurh",
1401 [(set GPR32:$Rt, (zextloadi16 am_unscaled16:$addr))]>;
1403 : LoadUnscaled<0b00, 0, 0b01, GPR32, am_unscaled8, "ldurb",
1404 [(set GPR32:$Rt, (zextloadi8 am_unscaled8:$addr))]>;
1406 // Match all load 64 bits width whose type is compatible with FPR64
1407 let Predicates = [IsLE] in {
1408 def : Pat<(v2f32 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1409 def : Pat<(v8i8 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1410 def : Pat<(v4i16 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1411 def : Pat<(v2i32 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1413 def : Pat<(v1f64 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1414 def : Pat<(v1i64 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1416 // Match all load 128 bits width whose type is compatible with FPR128
1417 let Predicates = [IsLE] in {
1418 def : Pat<(v4f32 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1419 def : Pat<(v2f64 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1420 def : Pat<(v16i8 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1421 def : Pat<(v8i16 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1422 def : Pat<(v4i32 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1423 def : Pat<(v2i64 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1424 def : Pat<(v2f64 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1428 def : Pat<(i32 (extloadi16 am_unscaled16:$addr)), (LDURHHi am_unscaled16:$addr)>;
1429 def : Pat<(i32 (extloadi8 am_unscaled8:$addr)), (LDURBBi am_unscaled8:$addr)>;
1430 def : Pat<(i32 (extloadi1 am_unscaled8:$addr)), (LDURBBi am_unscaled8:$addr)>;
1431 def : Pat<(i64 (extloadi32 am_unscaled32:$addr)),
1432 (SUBREG_TO_REG (i64 0), (LDURWi am_unscaled32:$addr), sub_32)>;
1433 def : Pat<(i64 (extloadi16 am_unscaled16:$addr)),
1434 (SUBREG_TO_REG (i64 0), (LDURHHi am_unscaled16:$addr), sub_32)>;
1435 def : Pat<(i64 (extloadi8 am_unscaled8:$addr)),
1436 (SUBREG_TO_REG (i64 0), (LDURBBi am_unscaled8:$addr), sub_32)>;
1437 def : Pat<(i64 (extloadi1 am_unscaled8:$addr)),
1438 (SUBREG_TO_REG (i64 0), (LDURBBi am_unscaled8:$addr), sub_32)>;
1440 def : Pat<(i32 (zextloadi16 am_unscaled16:$addr)),
1441 (LDURHHi am_unscaled16:$addr)>;
1442 def : Pat<(i32 (zextloadi8 am_unscaled8:$addr)),
1443 (LDURBBi am_unscaled8:$addr)>;
1444 def : Pat<(i32 (zextloadi1 am_unscaled8:$addr)),
1445 (LDURBBi am_unscaled8:$addr)>;
1446 def : Pat<(i64 (zextloadi32 am_unscaled32:$addr)),
1447 (SUBREG_TO_REG (i64 0), (LDURWi am_unscaled32:$addr), sub_32)>;
1448 def : Pat<(i64 (zextloadi16 am_unscaled16:$addr)),
1449 (SUBREG_TO_REG (i64 0), (LDURHHi am_unscaled16:$addr), sub_32)>;
1450 def : Pat<(i64 (zextloadi8 am_unscaled8:$addr)),
1451 (SUBREG_TO_REG (i64 0), (LDURBBi am_unscaled8:$addr), sub_32)>;
1452 def : Pat<(i64 (zextloadi1 am_unscaled8:$addr)),
1453 (SUBREG_TO_REG (i64 0), (LDURBBi am_unscaled8:$addr), sub_32)>;
1457 // LDR mnemonics fall back to LDUR for negative or unaligned offsets.
1459 // Define new assembler match classes as we want to only match these when
1460 // the don't otherwise match the scaled addressing mode for LDR/STR. Don't
1461 // associate a DiagnosticType either, as we want the diagnostic for the
1462 // canonical form (the scaled operand) to take precedence.
1463 def MemoryUnscaledFB8Operand : AsmOperandClass {
1464 let Name = "MemoryUnscaledFB8";
1465 let RenderMethod = "addMemoryUnscaledOperands";
1467 def MemoryUnscaledFB16Operand : AsmOperandClass {
1468 let Name = "MemoryUnscaledFB16";
1469 let RenderMethod = "addMemoryUnscaledOperands";
1471 def MemoryUnscaledFB32Operand : AsmOperandClass {
1472 let Name = "MemoryUnscaledFB32";
1473 let RenderMethod = "addMemoryUnscaledOperands";
1475 def MemoryUnscaledFB64Operand : AsmOperandClass {
1476 let Name = "MemoryUnscaledFB64";
1477 let RenderMethod = "addMemoryUnscaledOperands";
1479 def MemoryUnscaledFB128Operand : AsmOperandClass {
1480 let Name = "MemoryUnscaledFB128";
1481 let RenderMethod = "addMemoryUnscaledOperands";
1483 def am_unscaled_fb8 : Operand<i64> {
1484 let ParserMatchClass = MemoryUnscaledFB8Operand;
1485 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
1487 def am_unscaled_fb16 : Operand<i64> {
1488 let ParserMatchClass = MemoryUnscaledFB16Operand;
1489 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
1491 def am_unscaled_fb32 : Operand<i64> {
1492 let ParserMatchClass = MemoryUnscaledFB32Operand;
1493 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
1495 def am_unscaled_fb64 : Operand<i64> {
1496 let ParserMatchClass = MemoryUnscaledFB64Operand;
1497 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
1499 def am_unscaled_fb128 : Operand<i64> {
1500 let ParserMatchClass = MemoryUnscaledFB128Operand;
1501 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
1503 def : InstAlias<"ldr $Rt, $addr", (LDURXi GPR64:$Rt, am_unscaled_fb64:$addr), 0>;
1504 def : InstAlias<"ldr $Rt, $addr", (LDURWi GPR32:$Rt, am_unscaled_fb32:$addr), 0>;
1505 def : InstAlias<"ldr $Rt, $addr", (LDURBi FPR8:$Rt, am_unscaled_fb8:$addr), 0>;
1506 def : InstAlias<"ldr $Rt, $addr", (LDURHi FPR16:$Rt, am_unscaled_fb16:$addr), 0>;
1507 def : InstAlias<"ldr $Rt, $addr", (LDURSi FPR32:$Rt, am_unscaled_fb32:$addr), 0>;
1508 def : InstAlias<"ldr $Rt, $addr", (LDURDi FPR64:$Rt, am_unscaled_fb64:$addr), 0>;
1509 def : InstAlias<"ldr $Rt, $addr", (LDURQi FPR128:$Rt, am_unscaled_fb128:$addr), 0>;
1512 def : Pat<(i64 (zextloadi8 am_unscaled8:$addr)),
1513 (SUBREG_TO_REG (i64 0), (LDURBBi am_unscaled8:$addr), sub_32)>;
1514 def : Pat<(i64 (zextloadi16 am_unscaled16:$addr)),
1515 (SUBREG_TO_REG (i64 0), (LDURHHi am_unscaled16:$addr), sub_32)>;
1517 // load sign-extended half-word
1519 : LoadUnscaled<0b01, 0, 0b11, GPR32, am_unscaled16, "ldursh",
1520 [(set GPR32:$Rt, (sextloadi16 am_unscaled16:$addr))]>;
1522 : LoadUnscaled<0b01, 0, 0b10, GPR64, am_unscaled16, "ldursh",
1523 [(set GPR64:$Rt, (sextloadi16 am_unscaled16:$addr))]>;
1525 // load sign-extended byte
1527 : LoadUnscaled<0b00, 0, 0b11, GPR32, am_unscaled8, "ldursb",
1528 [(set GPR32:$Rt, (sextloadi8 am_unscaled8:$addr))]>;
1530 : LoadUnscaled<0b00, 0, 0b10, GPR64, am_unscaled8, "ldursb",
1531 [(set GPR64:$Rt, (sextloadi8 am_unscaled8:$addr))]>;
1533 // load sign-extended word
1535 : LoadUnscaled<0b10, 0, 0b10, GPR64, am_unscaled32, "ldursw",
1536 [(set GPR64:$Rt, (sextloadi32 am_unscaled32:$addr))]>;
1538 // zero and sign extending aliases from generic LDR* mnemonics to LDUR*.
1539 def : InstAlias<"ldrb $Rt, $addr",
1540 (LDURBBi GPR32:$Rt, am_unscaled_fb8:$addr), 0>;
1541 def : InstAlias<"ldrh $Rt, $addr",
1542 (LDURHHi GPR32:$Rt, am_unscaled_fb16:$addr), 0>;
1543 def : InstAlias<"ldrsb $Rt, $addr",
1544 (LDURSBWi GPR32:$Rt, am_unscaled_fb8:$addr), 0>;
1545 def : InstAlias<"ldrsb $Rt, $addr",
1546 (LDURSBXi GPR64:$Rt, am_unscaled_fb8:$addr), 0>;
1547 def : InstAlias<"ldrsh $Rt, $addr",
1548 (LDURSHWi GPR32:$Rt, am_unscaled_fb16:$addr), 0>;
1549 def : InstAlias<"ldrsh $Rt, $addr",
1550 (LDURSHXi GPR64:$Rt, am_unscaled_fb16:$addr), 0>;
1551 def : InstAlias<"ldrsw $Rt, $addr",
1552 (LDURSWi GPR64:$Rt, am_unscaled_fb32:$addr), 0>;
1555 def PRFUMi : PrefetchUnscaled<0b11, 0, 0b10, "prfum",
1556 [(ARM64Prefetch imm:$Rt, am_unscaled64:$addr)]>;
1559 // (unscaled immediate, unprivileged)
1560 def LDTRXi : LoadUnprivileged<0b11, 0, 0b01, GPR64, "ldtr">;
1561 def LDTRWi : LoadUnprivileged<0b10, 0, 0b01, GPR32, "ldtr">;
1563 def LDTRHi : LoadUnprivileged<0b01, 0, 0b01, GPR32, "ldtrh">;
1564 def LDTRBi : LoadUnprivileged<0b00, 0, 0b01, GPR32, "ldtrb">;
1566 // load sign-extended half-word
1567 def LDTRSHWi : LoadUnprivileged<0b01, 0, 0b11, GPR32, "ldtrsh">;
1568 def LDTRSHXi : LoadUnprivileged<0b01, 0, 0b10, GPR64, "ldtrsh">;
1570 // load sign-extended byte
1571 def LDTRSBWi : LoadUnprivileged<0b00, 0, 0b11, GPR32, "ldtrsb">;
1572 def LDTRSBXi : LoadUnprivileged<0b00, 0, 0b10, GPR64, "ldtrsb">;
1574 // load sign-extended word
1575 def LDTRSWi : LoadUnprivileged<0b10, 0, 0b10, GPR64, "ldtrsw">;
1578 // (immediate pre-indexed)
1579 def LDRWpre : LoadPreIdx<0b10, 0, 0b01, GPR32, "ldr">;
1580 def LDRXpre : LoadPreIdx<0b11, 0, 0b01, GPR64, "ldr">;
1581 def LDRBpre : LoadPreIdx<0b00, 1, 0b01, FPR8, "ldr">;
1582 def LDRHpre : LoadPreIdx<0b01, 1, 0b01, FPR16, "ldr">;
1583 def LDRSpre : LoadPreIdx<0b10, 1, 0b01, FPR32, "ldr">;
1584 def LDRDpre : LoadPreIdx<0b11, 1, 0b01, FPR64, "ldr">;
1585 def LDRQpre : LoadPreIdx<0b00, 1, 0b11, FPR128, "ldr">;
1587 // load sign-extended half-word
1588 def LDRSHWpre : LoadPreIdx<0b01, 0, 0b11, GPR32, "ldrsh">;
1589 def LDRSHXpre : LoadPreIdx<0b01, 0, 0b10, GPR64, "ldrsh">;
1591 // load sign-extended byte
1592 def LDRSBWpre : LoadPreIdx<0b00, 0, 0b11, GPR32, "ldrsb">;
1593 def LDRSBXpre : LoadPreIdx<0b00, 0, 0b10, GPR64, "ldrsb">;
1595 // load zero-extended byte
1596 def LDRBBpre : LoadPreIdx<0b00, 0, 0b01, GPR32, "ldrb">;
1597 def LDRHHpre : LoadPreIdx<0b01, 0, 0b01, GPR32, "ldrh">;
1599 // load sign-extended word
1600 def LDRSWpre : LoadPreIdx<0b10, 0, 0b10, GPR64, "ldrsw">;
1602 // ISel pseudos and patterns. See expanded comment on LoadPreIdxPseudo.
1603 def LDRQpre_isel : LoadPreIdxPseudo<FPR128>;
1604 def LDRDpre_isel : LoadPreIdxPseudo<FPR64>;
1605 def LDRSpre_isel : LoadPreIdxPseudo<FPR32>;
1606 def LDRXpre_isel : LoadPreIdxPseudo<GPR64>;
1607 def LDRWpre_isel : LoadPreIdxPseudo<GPR32>;
1608 def LDRHHpre_isel : LoadPreIdxPseudo<GPR32>;
1609 def LDRBBpre_isel : LoadPreIdxPseudo<GPR32>;
1611 def LDRSWpre_isel : LoadPreIdxPseudo<GPR64>;
1612 def LDRSHWpre_isel : LoadPreIdxPseudo<GPR32>;
1613 def LDRSHXpre_isel : LoadPreIdxPseudo<GPR64>;
1614 def LDRSBWpre_isel : LoadPreIdxPseudo<GPR32>;
1615 def LDRSBXpre_isel : LoadPreIdxPseudo<GPR64>;
1618 // (immediate post-indexed)
1619 def LDRWpost : LoadPostIdx<0b10, 0, 0b01, GPR32, "ldr">;
1620 def LDRXpost : LoadPostIdx<0b11, 0, 0b01, GPR64, "ldr">;
1621 def LDRBpost : LoadPostIdx<0b00, 1, 0b01, FPR8, "ldr">;
1622 def LDRHpost : LoadPostIdx<0b01, 1, 0b01, FPR16, "ldr">;
1623 def LDRSpost : LoadPostIdx<0b10, 1, 0b01, FPR32, "ldr">;
1624 def LDRDpost : LoadPostIdx<0b11, 1, 0b01, FPR64, "ldr">;
1625 def LDRQpost : LoadPostIdx<0b00, 1, 0b11, FPR128, "ldr">;
1627 // load sign-extended half-word
1628 def LDRSHWpost : LoadPostIdx<0b01, 0, 0b11, GPR32, "ldrsh">;
1629 def LDRSHXpost : LoadPostIdx<0b01, 0, 0b10, GPR64, "ldrsh">;
1631 // load sign-extended byte
1632 def LDRSBWpost : LoadPostIdx<0b00, 0, 0b11, GPR32, "ldrsb">;
1633 def LDRSBXpost : LoadPostIdx<0b00, 0, 0b10, GPR64, "ldrsb">;
1635 // load zero-extended byte
1636 def LDRBBpost : LoadPostIdx<0b00, 0, 0b01, GPR32, "ldrb">;
1637 def LDRHHpost : LoadPostIdx<0b01, 0, 0b01, GPR32, "ldrh">;
1639 // load sign-extended word
1640 def LDRSWpost : LoadPostIdx<0b10, 0, 0b10, GPR64, "ldrsw">;
1642 // ISel pseudos and patterns. See expanded comment on LoadPostIdxPseudo.
1643 def LDRQpost_isel : LoadPostIdxPseudo<FPR128>;
1644 def LDRDpost_isel : LoadPostIdxPseudo<FPR64>;
1645 def LDRSpost_isel : LoadPostIdxPseudo<FPR32>;
1646 def LDRXpost_isel : LoadPostIdxPseudo<GPR64>;
1647 def LDRWpost_isel : LoadPostIdxPseudo<GPR32>;
1648 def LDRHHpost_isel : LoadPostIdxPseudo<GPR32>;
1649 def LDRBBpost_isel : LoadPostIdxPseudo<GPR32>;
1651 def LDRSWpost_isel : LoadPostIdxPseudo<GPR64>;
1652 def LDRSHWpost_isel : LoadPostIdxPseudo<GPR32>;
1653 def LDRSHXpost_isel : LoadPostIdxPseudo<GPR64>;
1654 def LDRSBWpost_isel : LoadPostIdxPseudo<GPR32>;
1655 def LDRSBXpost_isel : LoadPostIdxPseudo<GPR64>;
1657 //===----------------------------------------------------------------------===//
1658 // Store instructions.
1659 //===----------------------------------------------------------------------===//
1661 // Pair (indexed, offset)
1662 // FIXME: Use dedicated range-checked addressing mode operand here.
1663 def STPWi : StorePairOffset<0b00, 0, GPR32, am_indexed32simm7, "stp">;
1664 def STPXi : StorePairOffset<0b10, 0, GPR64, am_indexed64simm7, "stp">;
1665 def STPSi : StorePairOffset<0b00, 1, FPR32, am_indexed32simm7, "stp">;
1666 def STPDi : StorePairOffset<0b01, 1, FPR64, am_indexed64simm7, "stp">;
1667 def STPQi : StorePairOffset<0b10, 1, FPR128, am_indexed128simm7, "stp">;
1669 // Pair (pre-indexed)
1670 def STPWpre : StorePairPreIdx<0b00, 0, GPR32, am_indexed32simm7_wb, "stp">;
1671 def STPXpre : StorePairPreIdx<0b10, 0, GPR64, am_indexed64simm7_wb, "stp">;
1672 def STPSpre : StorePairPreIdx<0b00, 1, FPR32, am_indexed32simm7_wb, "stp">;
1673 def STPDpre : StorePairPreIdx<0b01, 1, FPR64, am_indexed64simm7_wb, "stp">;
1674 def STPQpre : StorePairPreIdx<0b10, 1, FPR128, am_indexed128simm7_wb, "stp">;
1676 // Pair (pre-indexed)
1677 def STPWpost : StorePairPostIdx<0b00, 0, GPR32, simm7s4, "stp">;
1678 def STPXpost : StorePairPostIdx<0b10, 0, GPR64, simm7s8, "stp">;
1679 def STPSpost : StorePairPostIdx<0b00, 1, FPR32, simm7s4, "stp">;
1680 def STPDpost : StorePairPostIdx<0b01, 1, FPR64, simm7s8, "stp">;
1681 def STPQpost : StorePairPostIdx<0b10, 1, FPR128, simm7s16, "stp">;
1683 // Pair (no allocate)
1684 def STNPWi : StorePairNoAlloc<0b00, 0, GPR32, am_indexed32simm7, "stnp">;
1685 def STNPXi : StorePairNoAlloc<0b10, 0, GPR64, am_indexed64simm7, "stnp">;
1686 def STNPSi : StorePairNoAlloc<0b00, 1, FPR32, am_indexed32simm7, "stnp">;
1687 def STNPDi : StorePairNoAlloc<0b01, 1, FPR64, am_indexed64simm7, "stnp">;
1688 def STNPQi : StorePairNoAlloc<0b10, 1, FPR128, am_indexed128simm7, "stnp">;
1691 // (Register offset)
1693 let AddedComplexity = 10 in {
1696 def STRHHro : Store16RO<0b01, 0, 0b00, GPR32, "strh",
1697 [(truncstorei16 GPR32:$Rt, ro_indexed16:$addr)]>;
1698 def STRBBro : Store8RO<0b00, 0, 0b00, GPR32, "strb",
1699 [(truncstorei8 GPR32:$Rt, ro_indexed8:$addr)]>;
1700 def STRWro : Store32RO<0b10, 0, 0b00, GPR32, "str",
1701 [(store GPR32:$Rt, ro_indexed32:$addr)]>;
1702 def STRXro : Store64RO<0b11, 0, 0b00, GPR64, "str",
1703 [(store GPR64:$Rt, ro_indexed64:$addr)]>;
1706 def : Pat<(truncstorei8 GPR64:$Rt, ro_indexed8:$addr),
1707 (STRBBro (EXTRACT_SUBREG GPR64:$Rt, sub_32), ro_indexed8:$addr)>;
1708 def : Pat<(truncstorei16 GPR64:$Rt, ro_indexed16:$addr),
1709 (STRHHro (EXTRACT_SUBREG GPR64:$Rt, sub_32), ro_indexed16:$addr)>;
1710 def : Pat<(truncstorei32 GPR64:$Rt, ro_indexed32:$addr),
1711 (STRWro (EXTRACT_SUBREG GPR64:$Rt, sub_32), ro_indexed32:$addr)>;
1715 def STRBro : Store8RO<0b00, 1, 0b00, FPR8, "str",
1716 [(store FPR8:$Rt, ro_indexed8:$addr)]>;
1717 def STRHro : Store16RO<0b01, 1, 0b00, FPR16, "str",
1718 [(store (f16 FPR16:$Rt), ro_indexed16:$addr)]>;
1719 def STRSro : Store32RO<0b10, 1, 0b00, FPR32, "str",
1720 [(store (f32 FPR32:$Rt), ro_indexed32:$addr)]>;
1721 def STRDro : Store64RO<0b11, 1, 0b00, FPR64, "str",
1722 [(store (f64 FPR64:$Rt), ro_indexed64:$addr)]>;
1723 def STRQro : Store128RO<0b00, 1, 0b10, FPR128, "str", []> {
1727 // Match all store 64 bits width whose type is compatible with FPR64
1728 let Predicates = [IsLE] in {
1729 // We must use ST1 to store vectors in big-endian.
1730 def : Pat<(store (v2f32 FPR64:$Rn), ro_indexed64:$addr),
1731 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1732 def : Pat<(store (v8i8 FPR64:$Rn), ro_indexed64:$addr),
1733 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1734 def : Pat<(store (v4i16 FPR64:$Rn), ro_indexed64:$addr),
1735 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1736 def : Pat<(store (v2i32 FPR64:$Rn), ro_indexed64:$addr),
1737 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1739 def : Pat<(store (v1f64 FPR64:$Rn), ro_indexed64:$addr),
1740 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1741 def : Pat<(store (v1i64 FPR64:$Rn), ro_indexed64:$addr),
1742 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1744 // Match all store 128 bits width whose type is compatible with FPR128
1745 let Predicates = [IsLE] in {
1746 // We must use ST1 to store vectors in big-endian.
1747 def : Pat<(store (v4f32 FPR128:$Rn), ro_indexed128:$addr),
1748 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1749 def : Pat<(store (v2f64 FPR128:$Rn), ro_indexed128:$addr),
1750 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1751 def : Pat<(store (v16i8 FPR128:$Rn), ro_indexed128:$addr),
1752 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1753 def : Pat<(store (v8i16 FPR128:$Rn), ro_indexed128:$addr),
1754 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1755 def : Pat<(store (v4i32 FPR128:$Rn), ro_indexed128:$addr),
1756 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1757 def : Pat<(store (v2i64 FPR128:$Rn), ro_indexed128:$addr),
1758 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1760 def : Pat<(store (f128 FPR128:$Rn), ro_indexed128:$addr),
1761 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1764 // (unsigned immediate)
1765 def STRXui : StoreUI<0b11, 0, 0b00, GPR64, am_indexed64, "str",
1766 [(store GPR64:$Rt, am_indexed64:$addr)]>;
1767 def STRWui : StoreUI<0b10, 0, 0b00, GPR32, am_indexed32, "str",
1768 [(store GPR32:$Rt, am_indexed32:$addr)]>;
1769 def STRBui : StoreUI<0b00, 1, 0b00, FPR8, am_indexed8, "str",
1770 [(store FPR8:$Rt, am_indexed8:$addr)]>;
1771 def STRHui : StoreUI<0b01, 1, 0b00, FPR16, am_indexed16, "str",
1772 [(store (f16 FPR16:$Rt), am_indexed16:$addr)]>;
1773 def STRSui : StoreUI<0b10, 1, 0b00, FPR32, am_indexed32, "str",
1774 [(store (f32 FPR32:$Rt), am_indexed32:$addr)]>;
1775 def STRDui : StoreUI<0b11, 1, 0b00, FPR64, am_indexed64, "str",
1776 [(store (f64 FPR64:$Rt), am_indexed64:$addr)]>;
1777 def STRQui : StoreUI<0b00, 1, 0b10, FPR128, am_indexed128, "str", []> {
1781 // Match all store 64 bits width whose type is compatible with FPR64
1782 let Predicates = [IsLE] in {
1783 // We must use ST1 to store vectors in big-endian.
1784 def : Pat<(store (v2f32 FPR64:$Rn), am_indexed64:$addr),
1785 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1786 def : Pat<(store (v8i8 FPR64:$Rn), am_indexed64:$addr),
1787 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1788 def : Pat<(store (v4i16 FPR64:$Rn), am_indexed64:$addr),
1789 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1790 def : Pat<(store (v2i32 FPR64:$Rn), am_indexed64:$addr),
1791 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1793 def : Pat<(store (v1f64 FPR64:$Rn), am_indexed64:$addr),
1794 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1795 def : Pat<(store (v1i64 FPR64:$Rn), am_indexed64:$addr),
1796 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1798 // Match all store 128 bits width whose type is compatible with FPR128
1799 let Predicates = [IsLE] in {
1800 // We must use ST1 to store vectors in big-endian.
1801 def : Pat<(store (v4f32 FPR128:$Rn), am_indexed128:$addr),
1802 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1803 def : Pat<(store (v2f64 FPR128:$Rn), am_indexed128:$addr),
1804 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1805 def : Pat<(store (v16i8 FPR128:$Rn), am_indexed128:$addr),
1806 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1807 def : Pat<(store (v8i16 FPR128:$Rn), am_indexed128:$addr),
1808 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1809 def : Pat<(store (v4i32 FPR128:$Rn), am_indexed128:$addr),
1810 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1811 def : Pat<(store (v2i64 FPR128:$Rn), am_indexed128:$addr),
1812 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1814 def : Pat<(store (f128 FPR128:$Rn), am_indexed128:$addr),
1815 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1817 def STRHHui : StoreUI<0b01, 0, 0b00, GPR32, am_indexed16, "strh",
1818 [(truncstorei16 GPR32:$Rt, am_indexed16:$addr)]>;
1819 def STRBBui : StoreUI<0b00, 0, 0b00, GPR32, am_indexed8, "strb",
1820 [(truncstorei8 GPR32:$Rt, am_indexed8:$addr)]>;
1823 def : Pat<(truncstorei32 GPR64:$Rt, am_indexed32:$addr),
1824 (STRWui (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_indexed32:$addr)>;
1825 def : Pat<(truncstorei16 GPR64:$Rt, am_indexed16:$addr),
1826 (STRHHui (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_indexed16:$addr)>;
1827 def : Pat<(truncstorei8 GPR64:$Rt, am_indexed8:$addr),
1828 (STRBBui (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_indexed8:$addr)>;
1830 } // AddedComplexity = 10
1833 // (unscaled immediate)
1834 def STURXi : StoreUnscaled<0b11, 0, 0b00, GPR64, am_unscaled64, "stur",
1835 [(store GPR64:$Rt, am_unscaled64:$addr)]>;
1836 def STURWi : StoreUnscaled<0b10, 0, 0b00, GPR32, am_unscaled32, "stur",
1837 [(store GPR32:$Rt, am_unscaled32:$addr)]>;
1838 def STURBi : StoreUnscaled<0b00, 1, 0b00, FPR8, am_unscaled8, "stur",
1839 [(store FPR8:$Rt, am_unscaled8:$addr)]>;
1840 def STURHi : StoreUnscaled<0b01, 1, 0b00, FPR16, am_unscaled16, "stur",
1841 [(store (f16 FPR16:$Rt), am_unscaled16:$addr)]>;
1842 def STURSi : StoreUnscaled<0b10, 1, 0b00, FPR32, am_unscaled32, "stur",
1843 [(store (f32 FPR32:$Rt), am_unscaled32:$addr)]>;
1844 def STURDi : StoreUnscaled<0b11, 1, 0b00, FPR64, am_unscaled64, "stur",
1845 [(store (f64 FPR64:$Rt), am_unscaled64:$addr)]>;
1846 def STURQi : StoreUnscaled<0b00, 1, 0b10, FPR128, am_unscaled128, "stur",
1847 [(store (f128 FPR128:$Rt), am_unscaled128:$addr)]>;
1848 def STURHHi : StoreUnscaled<0b01, 0, 0b00, GPR32, am_unscaled16, "sturh",
1849 [(truncstorei16 GPR32:$Rt, am_unscaled16:$addr)]>;
1850 def STURBBi : StoreUnscaled<0b00, 0, 0b00, GPR32, am_unscaled8, "sturb",
1851 [(truncstorei8 GPR32:$Rt, am_unscaled8:$addr)]>;
1853 // Match all store 64 bits width whose type is compatible with FPR64
1854 let Predicates = [IsLE] in {
1855 // We must use ST1 to store vectors in big-endian.
1856 def : Pat<(store (v2f32 FPR64:$Rn), am_unscaled64:$addr),
1857 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1858 def : Pat<(store (v8i8 FPR64:$Rn), am_unscaled64:$addr),
1859 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1860 def : Pat<(store (v4i16 FPR64:$Rn), am_unscaled64:$addr),
1861 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1862 def : Pat<(store (v2i32 FPR64:$Rn), am_unscaled64:$addr),
1863 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1865 def : Pat<(store (v1f64 FPR64:$Rn), am_unscaled64:$addr),
1866 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1867 def : Pat<(store (v1i64 FPR64:$Rn), am_unscaled64:$addr),
1868 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1870 // Match all store 128 bits width whose type is compatible with FPR128
1871 let Predicates = [IsLE] in {
1872 // We must use ST1 to store vectors in big-endian.
1873 def : Pat<(store (v4f32 FPR128:$Rn), am_unscaled128:$addr),
1874 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1875 def : Pat<(store (v2f64 FPR128:$Rn), am_unscaled128:$addr),
1876 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1877 def : Pat<(store (v16i8 FPR128:$Rn), am_unscaled128:$addr),
1878 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1879 def : Pat<(store (v8i16 FPR128:$Rn), am_unscaled128:$addr),
1880 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1881 def : Pat<(store (v4i32 FPR128:$Rn), am_unscaled128:$addr),
1882 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1883 def : Pat<(store (v2i64 FPR128:$Rn), am_unscaled128:$addr),
1884 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1885 def : Pat<(store (v2f64 FPR128:$Rn), am_unscaled128:$addr),
1886 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1889 // unscaled i64 truncating stores
1890 def : Pat<(truncstorei32 GPR64:$Rt, am_unscaled32:$addr),
1891 (STURWi (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_unscaled32:$addr)>;
1892 def : Pat<(truncstorei16 GPR64:$Rt, am_unscaled16:$addr),
1893 (STURHHi (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_unscaled16:$addr)>;
1894 def : Pat<(truncstorei8 GPR64:$Rt, am_unscaled8:$addr),
1895 (STURBBi (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_unscaled8:$addr)>;
1898 // STR mnemonics fall back to STUR for negative or unaligned offsets.
1899 def : InstAlias<"str $Rt, $addr",
1900 (STURXi GPR64:$Rt, am_unscaled_fb64:$addr), 0>;
1901 def : InstAlias<"str $Rt, $addr",
1902 (STURWi GPR32:$Rt, am_unscaled_fb32:$addr), 0>;
1903 def : InstAlias<"str $Rt, $addr",
1904 (STURBi FPR8:$Rt, am_unscaled_fb8:$addr), 0>;
1905 def : InstAlias<"str $Rt, $addr",
1906 (STURHi FPR16:$Rt, am_unscaled_fb16:$addr), 0>;
1907 def : InstAlias<"str $Rt, $addr",
1908 (STURSi FPR32:$Rt, am_unscaled_fb32:$addr), 0>;
1909 def : InstAlias<"str $Rt, $addr",
1910 (STURDi FPR64:$Rt, am_unscaled_fb64:$addr), 0>;
1911 def : InstAlias<"str $Rt, $addr",
1912 (STURQi FPR128:$Rt, am_unscaled_fb128:$addr), 0>;
1914 def : InstAlias<"strb $Rt, $addr",
1915 (STURBBi GPR32:$Rt, am_unscaled_fb8:$addr), 0>;
1916 def : InstAlias<"strh $Rt, $addr",
1917 (STURHHi GPR32:$Rt, am_unscaled_fb16:$addr), 0>;
1920 // (unscaled immediate, unprivileged)
1921 def STTRWi : StoreUnprivileged<0b10, 0, 0b00, GPR32, "sttr">;
1922 def STTRXi : StoreUnprivileged<0b11, 0, 0b00, GPR64, "sttr">;
1924 def STTRHi : StoreUnprivileged<0b01, 0, 0b00, GPR32, "sttrh">;
1925 def STTRBi : StoreUnprivileged<0b00, 0, 0b00, GPR32, "sttrb">;
1928 // (immediate pre-indexed)
1929 def STRWpre : StorePreIdx<0b10, 0, 0b00, GPR32, "str">;
1930 def STRXpre : StorePreIdx<0b11, 0, 0b00, GPR64, "str">;
1931 def STRBpre : StorePreIdx<0b00, 1, 0b00, FPR8, "str">;
1932 def STRHpre : StorePreIdx<0b01, 1, 0b00, FPR16, "str">;
1933 def STRSpre : StorePreIdx<0b10, 1, 0b00, FPR32, "str">;
1934 def STRDpre : StorePreIdx<0b11, 1, 0b00, FPR64, "str">;
1935 def STRQpre : StorePreIdx<0b00, 1, 0b10, FPR128, "str">;
1937 def STRBBpre : StorePreIdx<0b00, 0, 0b00, GPR32, "strb">;
1938 def STRHHpre : StorePreIdx<0b01, 0, 0b00, GPR32, "strh">;
1940 // ISel pseudos and patterns. See expanded comment on StorePreIdxPseudo.
1941 defm STRQpre : StorePreIdxPseudo<FPR128, f128, pre_store>;
1942 defm STRDpre : StorePreIdxPseudo<FPR64, f64, pre_store>;
1943 defm STRSpre : StorePreIdxPseudo<FPR32, f32, pre_store>;
1944 defm STRXpre : StorePreIdxPseudo<GPR64, i64, pre_store>;
1945 defm STRWpre : StorePreIdxPseudo<GPR32, i32, pre_store>;
1946 defm STRHHpre : StorePreIdxPseudo<GPR32, i32, pre_truncsti16>;
1947 defm STRBBpre : StorePreIdxPseudo<GPR32, i32, pre_truncsti8>;
1949 def : Pat<(pre_truncsti32 GPR64:$Rt, am_noindex:$addr, simm9:$off),
1950 (STRWpre_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
1952 def : Pat<(pre_truncsti16 GPR64:$Rt, am_noindex:$addr, simm9:$off),
1953 (STRHHpre_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
1955 def : Pat<(pre_truncsti8 GPR64:$Rt, am_noindex:$addr, simm9:$off),
1956 (STRBBpre_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
1959 def : Pat<(pre_store (v8i8 FPR64:$Rt), am_noindex:$addr, simm9:$off),
1960 (STRDpre_isel FPR64:$Rt, am_noindex:$addr, simm9:$off)>;
1961 def : Pat<(pre_store (v4i16 FPR64:$Rt), am_noindex:$addr, simm9:$off),
1962 (STRDpre_isel FPR64:$Rt, am_noindex:$addr, simm9:$off)>;
1963 def : Pat<(pre_store (v2i32 FPR64:$Rt), am_noindex:$addr, simm9:$off),
1964 (STRDpre_isel FPR64:$Rt, am_noindex:$addr, simm9:$off)>;
1965 def : Pat<(pre_store (v2f32 FPR64:$Rt), am_noindex:$addr, simm9:$off),
1966 (STRDpre_isel FPR64:$Rt, am_noindex:$addr, simm9:$off)>;
1967 def : Pat<(pre_store (v1i64 FPR64:$Rt), am_noindex:$addr, simm9:$off),
1968 (STRDpre_isel FPR64:$Rt, am_noindex:$addr, simm9:$off)>;
1969 def : Pat<(pre_store (v1f64 FPR64:$Rt), am_noindex:$addr, simm9:$off),
1970 (STRDpre_isel FPR64:$Rt, am_noindex:$addr, simm9:$off)>;
1972 def : Pat<(pre_store (v16i8 FPR128:$Rt), am_noindex:$addr, simm9:$off),
1973 (STRQpre_isel FPR128:$Rt, am_noindex:$addr, simm9:$off)>;
1974 def : Pat<(pre_store (v8i16 FPR128:$Rt), am_noindex:$addr, simm9:$off),
1975 (STRQpre_isel FPR128:$Rt, am_noindex:$addr, simm9:$off)>;
1976 def : Pat<(pre_store (v4i32 FPR128:$Rt), am_noindex:$addr, simm9:$off),
1977 (STRQpre_isel FPR128:$Rt, am_noindex:$addr, simm9:$off)>;
1978 def : Pat<(pre_store (v4f32 FPR128:$Rt), am_noindex:$addr, simm9:$off),
1979 (STRQpre_isel FPR128:$Rt, am_noindex:$addr, simm9:$off)>;
1980 def : Pat<(pre_store (v2i64 FPR128:$Rt), am_noindex:$addr, simm9:$off),
1981 (STRQpre_isel FPR128:$Rt, am_noindex:$addr, simm9:$off)>;
1982 def : Pat<(pre_store (v2f64 FPR128:$Rt), am_noindex:$addr, simm9:$off),
1983 (STRQpre_isel FPR128:$Rt, am_noindex:$addr, simm9:$off)>;
1986 // (immediate post-indexed)
1987 def STRWpost : StorePostIdx<0b10, 0, 0b00, GPR32, "str">;
1988 def STRXpost : StorePostIdx<0b11, 0, 0b00, GPR64, "str">;
1989 def STRBpost : StorePostIdx<0b00, 1, 0b00, FPR8, "str">;
1990 def STRHpost : StorePostIdx<0b01, 1, 0b00, FPR16, "str">;
1991 def STRSpost : StorePostIdx<0b10, 1, 0b00, FPR32, "str">;
1992 def STRDpost : StorePostIdx<0b11, 1, 0b00, FPR64, "str">;
1993 def STRQpost : StorePostIdx<0b00, 1, 0b10, FPR128, "str">;
1995 def STRBBpost : StorePostIdx<0b00, 0, 0b00, GPR32, "strb">;
1996 def STRHHpost : StorePostIdx<0b01, 0, 0b00, GPR32, "strh">;
1998 // ISel pseudos and patterns. See expanded comment on StorePostIdxPseudo.
1999 defm STRQpost : StorePostIdxPseudo<FPR128, f128, post_store, STRQpost>;
2000 defm STRDpost : StorePostIdxPseudo<FPR64, f64, post_store, STRDpost>;
2001 defm STRSpost : StorePostIdxPseudo<FPR32, f32, post_store, STRSpost>;
2002 defm STRXpost : StorePostIdxPseudo<GPR64, i64, post_store, STRXpost>;
2003 defm STRWpost : StorePostIdxPseudo<GPR32, i32, post_store, STRWpost>;
2004 defm STRHHpost : StorePostIdxPseudo<GPR32, i32, post_truncsti16, STRHHpost>;
2005 defm STRBBpost : StorePostIdxPseudo<GPR32, i32, post_truncsti8, STRBBpost>;
2007 def : Pat<(post_truncsti32 GPR64:$Rt, am_noindex:$addr, simm9:$off),
2008 (STRWpost_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
2010 def : Pat<(post_truncsti16 GPR64:$Rt, am_noindex:$addr, simm9:$off),
2011 (STRHHpost_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
2013 def : Pat<(post_truncsti8 GPR64:$Rt, am_noindex:$addr, simm9:$off),
2014 (STRBBpost_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
2017 def : Pat<(post_store (v8i8 FPR64:$Rt), am_noindex:$addr, simm9:$off),
2018 (STRDpost_isel FPR64:$Rt, am_noindex:$addr, simm9:$off)>;
2019 def : Pat<(post_store (v4i16 FPR64:$Rt), am_noindex:$addr, simm9:$off),
2020 (STRDpost_isel FPR64:$Rt, am_noindex:$addr, simm9:$off)>;
2021 def : Pat<(post_store (v2i32 FPR64:$Rt), am_noindex:$addr, simm9:$off),
2022 (STRDpost_isel FPR64:$Rt, am_noindex:$addr, simm9:$off)>;
2023 def : Pat<(post_store (v2f32 FPR64:$Rt), am_noindex:$addr, simm9:$off),
2024 (STRDpost_isel FPR64:$Rt, am_noindex:$addr, simm9:$off)>;
2025 def : Pat<(post_store (v1i64 FPR64:$Rt), am_noindex:$addr, simm9:$off),
2026 (STRDpost_isel FPR64:$Rt, am_noindex:$addr, simm9:$off)>;
2027 def : Pat<(post_store (v1f64 FPR64:$Rt), am_noindex:$addr, simm9:$off),
2028 (STRDpost_isel FPR64:$Rt, am_noindex:$addr, simm9:$off)>;
2030 def : Pat<(post_store (v16i8 FPR128:$Rt), am_noindex:$addr, simm9:$off),
2031 (STRQpost_isel FPR128:$Rt, am_noindex:$addr, simm9:$off)>;
2032 def : Pat<(post_store (v8i16 FPR128:$Rt), am_noindex:$addr, simm9:$off),
2033 (STRQpost_isel FPR128:$Rt, am_noindex:$addr, simm9:$off)>;
2034 def : Pat<(post_store (v4i32 FPR128:$Rt), am_noindex:$addr, simm9:$off),
2035 (STRQpost_isel FPR128:$Rt, am_noindex:$addr, simm9:$off)>;
2036 def : Pat<(post_store (v4f32 FPR128:$Rt), am_noindex:$addr, simm9:$off),
2037 (STRQpost_isel FPR128:$Rt, am_noindex:$addr, simm9:$off)>;
2038 def : Pat<(post_store (v2i64 FPR128:$Rt), am_noindex:$addr, simm9:$off),
2039 (STRQpost_isel FPR128:$Rt, am_noindex:$addr, simm9:$off)>;
2040 def : Pat<(post_store (v2f64 FPR128:$Rt), am_noindex:$addr, simm9:$off),
2041 (STRQpost_isel FPR128:$Rt, am_noindex:$addr, simm9:$off)>;
2043 //===----------------------------------------------------------------------===//
2044 // Load/store exclusive instructions.
2045 //===----------------------------------------------------------------------===//
2047 def LDARW : LoadAcquire <0b10, 1, 1, 0, 1, GPR32, "ldar">;
2048 def LDARX : LoadAcquire <0b11, 1, 1, 0, 1, GPR64, "ldar">;
2049 def LDARB : LoadAcquire <0b00, 1, 1, 0, 1, GPR32, "ldarb">;
2050 def LDARH : LoadAcquire <0b01, 1, 1, 0, 1, GPR32, "ldarh">;
2052 def LDAXRW : LoadExclusive <0b10, 0, 1, 0, 1, GPR32, "ldaxr">;
2053 def LDAXRX : LoadExclusive <0b11, 0, 1, 0, 1, GPR64, "ldaxr">;
2054 def LDAXRB : LoadExclusive <0b00, 0, 1, 0, 1, GPR32, "ldaxrb">;
2055 def LDAXRH : LoadExclusive <0b01, 0, 1, 0, 1, GPR32, "ldaxrh">;
2057 def LDXRW : LoadExclusive <0b10, 0, 1, 0, 0, GPR32, "ldxr">;
2058 def LDXRX : LoadExclusive <0b11, 0, 1, 0, 0, GPR64, "ldxr">;
2059 def LDXRB : LoadExclusive <0b00, 0, 1, 0, 0, GPR32, "ldxrb">;
2060 def LDXRH : LoadExclusive <0b01, 0, 1, 0, 0, GPR32, "ldxrh">;
2062 def STLRW : StoreRelease <0b10, 1, 0, 0, 1, GPR32, "stlr">;
2063 def STLRX : StoreRelease <0b11, 1, 0, 0, 1, GPR64, "stlr">;
2064 def STLRB : StoreRelease <0b00, 1, 0, 0, 1, GPR32, "stlrb">;
2065 def STLRH : StoreRelease <0b01, 1, 0, 0, 1, GPR32, "stlrh">;
2067 def STLXRW : StoreExclusive<0b10, 0, 0, 0, 1, GPR32, "stlxr">;
2068 def STLXRX : StoreExclusive<0b11, 0, 0, 0, 1, GPR64, "stlxr">;
2069 def STLXRB : StoreExclusive<0b00, 0, 0, 0, 1, GPR32, "stlxrb">;
2070 def STLXRH : StoreExclusive<0b01, 0, 0, 0, 1, GPR32, "stlxrh">;
2072 def STXRW : StoreExclusive<0b10, 0, 0, 0, 0, GPR32, "stxr">;
2073 def STXRX : StoreExclusive<0b11, 0, 0, 0, 0, GPR64, "stxr">;
2074 def STXRB : StoreExclusive<0b00, 0, 0, 0, 0, GPR32, "stxrb">;
2075 def STXRH : StoreExclusive<0b01, 0, 0, 0, 0, GPR32, "stxrh">;
2077 def LDAXPW : LoadExclusivePair<0b10, 0, 1, 1, 1, GPR32, "ldaxp">;
2078 def LDAXPX : LoadExclusivePair<0b11, 0, 1, 1, 1, GPR64, "ldaxp">;
2080 def LDXPW : LoadExclusivePair<0b10, 0, 1, 1, 0, GPR32, "ldxp">;
2081 def LDXPX : LoadExclusivePair<0b11, 0, 1, 1, 0, GPR64, "ldxp">;
2083 def STLXPW : StoreExclusivePair<0b10, 0, 0, 1, 1, GPR32, "stlxp">;
2084 def STLXPX : StoreExclusivePair<0b11, 0, 0, 1, 1, GPR64, "stlxp">;
2086 def STXPW : StoreExclusivePair<0b10, 0, 0, 1, 0, GPR32, "stxp">;
2087 def STXPX : StoreExclusivePair<0b11, 0, 0, 1, 0, GPR64, "stxp">;
2089 //===----------------------------------------------------------------------===//
2090 // Scaled floating point to integer conversion instructions.
2091 //===----------------------------------------------------------------------===//
2093 defm FCVTAS : FPToIntegerUnscaled<0b00, 0b100, "fcvtas", int_arm64_neon_fcvtas>;
2094 defm FCVTAU : FPToIntegerUnscaled<0b00, 0b101, "fcvtau", int_arm64_neon_fcvtau>;
2095 defm FCVTMS : FPToIntegerUnscaled<0b10, 0b000, "fcvtms", int_arm64_neon_fcvtms>;
2096 defm FCVTMU : FPToIntegerUnscaled<0b10, 0b001, "fcvtmu", int_arm64_neon_fcvtmu>;
2097 defm FCVTNS : FPToIntegerUnscaled<0b00, 0b000, "fcvtns", int_arm64_neon_fcvtns>;
2098 defm FCVTNU : FPToIntegerUnscaled<0b00, 0b001, "fcvtnu", int_arm64_neon_fcvtnu>;
2099 defm FCVTPS : FPToIntegerUnscaled<0b01, 0b000, "fcvtps", int_arm64_neon_fcvtps>;
2100 defm FCVTPU : FPToIntegerUnscaled<0b01, 0b001, "fcvtpu", int_arm64_neon_fcvtpu>;
2101 defm FCVTZS : FPToIntegerUnscaled<0b11, 0b000, "fcvtzs", fp_to_sint>;
2102 defm FCVTZU : FPToIntegerUnscaled<0b11, 0b001, "fcvtzu", fp_to_uint>;
2103 defm FCVTZS : FPToIntegerScaled<0b11, 0b000, "fcvtzs", fp_to_sint>;
2104 defm FCVTZU : FPToIntegerScaled<0b11, 0b001, "fcvtzu", fp_to_uint>;
2105 let isCodeGenOnly = 1 in {
2106 defm FCVTZS_Int : FPToIntegerUnscaled<0b11, 0b000, "fcvtzs", int_arm64_neon_fcvtzs>;
2107 defm FCVTZU_Int : FPToIntegerUnscaled<0b11, 0b001, "fcvtzu", int_arm64_neon_fcvtzu>;
2108 defm FCVTZS_Int : FPToIntegerScaled<0b11, 0b000, "fcvtzs", int_arm64_neon_fcvtzs>;
2109 defm FCVTZU_Int : FPToIntegerScaled<0b11, 0b001, "fcvtzu", int_arm64_neon_fcvtzu>;
2112 //===----------------------------------------------------------------------===//
2113 // Scaled integer to floating point conversion instructions.
2114 //===----------------------------------------------------------------------===//
2116 defm SCVTF : IntegerToFP<0, "scvtf", sint_to_fp>;
2117 defm UCVTF : IntegerToFP<1, "ucvtf", uint_to_fp>;
2119 //===----------------------------------------------------------------------===//
2120 // Unscaled integer to floating point conversion instruction.
2121 //===----------------------------------------------------------------------===//
2123 defm FMOV : UnscaledConversion<"fmov">;
2125 def : Pat<(f32 (fpimm0)), (FMOVWSr WZR)>, Requires<[NoZCZ]>;
2126 def : Pat<(f64 (fpimm0)), (FMOVXDr XZR)>, Requires<[NoZCZ]>;
2128 //===----------------------------------------------------------------------===//
2129 // Floating point conversion instruction.
2130 //===----------------------------------------------------------------------===//
2132 defm FCVT : FPConversion<"fcvt">;
2134 def : Pat<(f32_to_f16 FPR32:$Rn),
2135 (i32 (COPY_TO_REGCLASS
2136 (f32 (SUBREG_TO_REG (i32 0), (FCVTHSr FPR32:$Rn), hsub)),
2139 def FCVTSHpseudo : Pseudo<(outs FPR32:$Rd), (ins FPR32:$Rn),
2140 [(set (f32 FPR32:$Rd), (f16_to_f32 i32:$Rn))]>;
2142 //===----------------------------------------------------------------------===//
2143 // Floating point single operand instructions.
2144 //===----------------------------------------------------------------------===//
2146 defm FABS : SingleOperandFPData<0b0001, "fabs", fabs>;
2147 defm FMOV : SingleOperandFPData<0b0000, "fmov">;
2148 defm FNEG : SingleOperandFPData<0b0010, "fneg", fneg>;
2149 defm FRINTA : SingleOperandFPData<0b1100, "frinta", frnd>;
2150 defm FRINTI : SingleOperandFPData<0b1111, "frinti", fnearbyint>;
2151 defm FRINTM : SingleOperandFPData<0b1010, "frintm", ffloor>;
2152 defm FRINTN : SingleOperandFPData<0b1000, "frintn", int_arm64_neon_frintn>;
2153 defm FRINTP : SingleOperandFPData<0b1001, "frintp", fceil>;
2155 def : Pat<(v1f64 (int_arm64_neon_frintn (v1f64 FPR64:$Rn))),
2156 (FRINTNDr FPR64:$Rn)>;
2158 // FRINTX is inserted to set the flags as required by FENV_ACCESS ON behavior
2159 // in the C spec. Setting hasSideEffects ensures it is not DCE'd.
2160 // <rdar://problem/13715968>
2161 // TODO: We should really model the FPSR flags correctly. This is really ugly.
2162 let hasSideEffects = 1 in {
2163 defm FRINTX : SingleOperandFPData<0b1110, "frintx", frint>;
2166 defm FRINTZ : SingleOperandFPData<0b1011, "frintz", ftrunc>;
2168 let SchedRW = [WriteFDiv] in {
2169 defm FSQRT : SingleOperandFPData<0b0011, "fsqrt", fsqrt>;
2172 //===----------------------------------------------------------------------===//
2173 // Floating point two operand instructions.
2174 //===----------------------------------------------------------------------===//
2176 defm FADD : TwoOperandFPData<0b0010, "fadd", fadd>;
2177 let SchedRW = [WriteFDiv] in {
2178 defm FDIV : TwoOperandFPData<0b0001, "fdiv", fdiv>;
2180 defm FMAXNM : TwoOperandFPData<0b0110, "fmaxnm", int_arm64_neon_fmaxnm>;
2181 defm FMAX : TwoOperandFPData<0b0100, "fmax", ARM64fmax>;
2182 defm FMINNM : TwoOperandFPData<0b0111, "fminnm", int_arm64_neon_fminnm>;
2183 defm FMIN : TwoOperandFPData<0b0101, "fmin", ARM64fmin>;
2184 let SchedRW = [WriteFMul] in {
2185 defm FMUL : TwoOperandFPData<0b0000, "fmul", fmul>;
2186 defm FNMUL : TwoOperandFPDataNeg<0b1000, "fnmul", fmul>;
2188 defm FSUB : TwoOperandFPData<0b0011, "fsub", fsub>;
2190 def : Pat<(v1f64 (ARM64fmax (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2191 (FMAXDrr FPR64:$Rn, FPR64:$Rm)>;
2192 def : Pat<(v1f64 (ARM64fmin (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2193 (FMINDrr FPR64:$Rn, FPR64:$Rm)>;
2194 def : Pat<(v1f64 (int_arm64_neon_fmaxnm (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2195 (FMAXNMDrr FPR64:$Rn, FPR64:$Rm)>;
2196 def : Pat<(v1f64 (int_arm64_neon_fminnm (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2197 (FMINNMDrr FPR64:$Rn, FPR64:$Rm)>;
2199 //===----------------------------------------------------------------------===//
2200 // Floating point three operand instructions.
2201 //===----------------------------------------------------------------------===//
2203 defm FMADD : ThreeOperandFPData<0, 0, "fmadd", fma>;
2204 defm FMSUB : ThreeOperandFPData<0, 1, "fmsub",
2205 TriOpFrag<(fma node:$LHS, (fneg node:$MHS), node:$RHS)> >;
2206 defm FNMADD : ThreeOperandFPData<1, 0, "fnmadd",
2207 TriOpFrag<(fneg (fma node:$LHS, node:$MHS, node:$RHS))> >;
2208 defm FNMSUB : ThreeOperandFPData<1, 1, "fnmsub",
2209 TriOpFrag<(fma node:$LHS, node:$MHS, (fneg node:$RHS))> >;
2211 // The following def pats catch the case where the LHS of an FMA is negated.
2212 // The TriOpFrag above catches the case where the middle operand is negated.
2214 // N.b. FMSUB etc have the accumulator at the *end* of (outs), unlike
2215 // the NEON variant.
2216 def : Pat<(f32 (fma (fneg FPR32:$Rn), FPR32:$Rm, FPR32:$Ra)),
2217 (FMSUBSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
2219 def : Pat<(f64 (fma (fneg FPR64:$Rn), FPR64:$Rm, FPR64:$Ra)),
2220 (FMSUBDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
2222 // We handled -(a + b*c) for FNMADD above, now it's time for "(-a) + (-b)*c" and
2224 def : Pat<(f32 (fma (fneg FPR32:$Rn), FPR32:$Rm, (fneg FPR32:$Ra))),
2225 (FNMADDSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
2227 def : Pat<(f64 (fma (fneg FPR64:$Rn), FPR64:$Rm, (fneg FPR64:$Ra))),
2228 (FNMADDDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
2230 def : Pat<(f32 (fma FPR32:$Rn, (fneg FPR32:$Rm), (fneg FPR32:$Ra))),
2231 (FNMADDSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
2233 def : Pat<(f64 (fma FPR64:$Rn, (fneg FPR64:$Rm), (fneg FPR64:$Ra))),
2234 (FNMADDDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
2236 //===----------------------------------------------------------------------===//
2237 // Floating point comparison instructions.
2238 //===----------------------------------------------------------------------===//
2240 defm FCMPE : FPComparison<1, "fcmpe">;
2241 defm FCMP : FPComparison<0, "fcmp", ARM64fcmp>;
2243 //===----------------------------------------------------------------------===//
2244 // Floating point conditional comparison instructions.
2245 //===----------------------------------------------------------------------===//
2247 defm FCCMPE : FPCondComparison<1, "fccmpe">;
2248 defm FCCMP : FPCondComparison<0, "fccmp">;
2250 //===----------------------------------------------------------------------===//
2251 // Floating point conditional select instruction.
2252 //===----------------------------------------------------------------------===//
2254 defm FCSEL : FPCondSelect<"fcsel">;
2256 // CSEL instructions providing f128 types need to be handled by a
2257 // pseudo-instruction since the eventual code will need to introduce basic
2258 // blocks and control flow.
2259 def F128CSEL : Pseudo<(outs FPR128:$Rd),
2260 (ins FPR128:$Rn, FPR128:$Rm, ccode:$cond),
2261 [(set (f128 FPR128:$Rd),
2262 (ARM64csel FPR128:$Rn, FPR128:$Rm,
2263 (i32 imm:$cond), NZCV))]> {
2265 let usesCustomInserter = 1;
2269 //===----------------------------------------------------------------------===//
2270 // Floating point immediate move.
2271 //===----------------------------------------------------------------------===//
2273 let isReMaterializable = 1 in {
2274 defm FMOV : FPMoveImmediate<"fmov">;
2277 //===----------------------------------------------------------------------===//
2278 // Advanced SIMD two vector instructions.
2279 //===----------------------------------------------------------------------===//
2281 defm ABS : SIMDTwoVectorBHSD<0, 0b01011, "abs", int_arm64_neon_abs>;
2282 defm CLS : SIMDTwoVectorBHS<0, 0b00100, "cls", int_arm64_neon_cls>;
2283 defm CLZ : SIMDTwoVectorBHS<1, 0b00100, "clz", ctlz>;
2284 defm CMEQ : SIMDCmpTwoVector<0, 0b01001, "cmeq", ARM64cmeqz>;
2285 defm CMGE : SIMDCmpTwoVector<1, 0b01000, "cmge", ARM64cmgez>;
2286 defm CMGT : SIMDCmpTwoVector<0, 0b01000, "cmgt", ARM64cmgtz>;
2287 defm CMLE : SIMDCmpTwoVector<1, 0b01001, "cmle", ARM64cmlez>;
2288 defm CMLT : SIMDCmpTwoVector<0, 0b01010, "cmlt", ARM64cmltz>;
2289 defm CNT : SIMDTwoVectorB<0, 0b00, 0b00101, "cnt", ctpop>;
2290 defm FABS : SIMDTwoVectorFP<0, 1, 0b01111, "fabs", fabs>;
2292 defm FCMEQ : SIMDFPCmpTwoVector<0, 1, 0b01101, "fcmeq", ARM64fcmeqz>;
2293 defm FCMGE : SIMDFPCmpTwoVector<1, 1, 0b01100, "fcmge", ARM64fcmgez>;
2294 defm FCMGT : SIMDFPCmpTwoVector<0, 1, 0b01100, "fcmgt", ARM64fcmgtz>;
2295 defm FCMLE : SIMDFPCmpTwoVector<1, 1, 0b01101, "fcmle", ARM64fcmlez>;
2296 defm FCMLT : SIMDFPCmpTwoVector<0, 1, 0b01110, "fcmlt", ARM64fcmltz>;
2297 defm FCVTAS : SIMDTwoVectorFPToInt<0,0,0b11100, "fcvtas",int_arm64_neon_fcvtas>;
2298 defm FCVTAU : SIMDTwoVectorFPToInt<1,0,0b11100, "fcvtau",int_arm64_neon_fcvtau>;
2299 defm FCVTL : SIMDFPWidenTwoVector<0, 0, 0b10111, "fcvtl">;
2300 def : Pat<(v4f32 (int_arm64_neon_vcvthf2fp (v4i16 V64:$Rn))),
2301 (FCVTLv4i16 V64:$Rn)>;
2302 def : Pat<(v4f32 (int_arm64_neon_vcvthf2fp (extract_subvector (v8i16 V128:$Rn),
2304 (FCVTLv8i16 V128:$Rn)>;
2305 def : Pat<(v2f64 (fextend (v2f32 V64:$Rn))), (FCVTLv2i32 V64:$Rn)>;
2306 def : Pat<(v2f64 (fextend (v2f32 (extract_subvector (v4f32 V128:$Rn),
2308 (FCVTLv4i32 V128:$Rn)>;
2310 defm FCVTMS : SIMDTwoVectorFPToInt<0,0,0b11011, "fcvtms",int_arm64_neon_fcvtms>;
2311 defm FCVTMU : SIMDTwoVectorFPToInt<1,0,0b11011, "fcvtmu",int_arm64_neon_fcvtmu>;
2312 defm FCVTNS : SIMDTwoVectorFPToInt<0,0,0b11010, "fcvtns",int_arm64_neon_fcvtns>;
2313 defm FCVTNU : SIMDTwoVectorFPToInt<1,0,0b11010, "fcvtnu",int_arm64_neon_fcvtnu>;
2314 defm FCVTN : SIMDFPNarrowTwoVector<0, 0, 0b10110, "fcvtn">;
2315 def : Pat<(v4i16 (int_arm64_neon_vcvtfp2hf (v4f32 V128:$Rn))),
2316 (FCVTNv4i16 V128:$Rn)>;
2317 def : Pat<(concat_vectors V64:$Rd,
2318 (v4i16 (int_arm64_neon_vcvtfp2hf (v4f32 V128:$Rn)))),
2319 (FCVTNv8i16 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
2320 def : Pat<(v2f32 (fround (v2f64 V128:$Rn))), (FCVTNv2i32 V128:$Rn)>;
2321 def : Pat<(concat_vectors V64:$Rd, (v2f32 (fround (v2f64 V128:$Rn)))),
2322 (FCVTNv4i32 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
2323 defm FCVTPS : SIMDTwoVectorFPToInt<0,1,0b11010, "fcvtps",int_arm64_neon_fcvtps>;
2324 defm FCVTPU : SIMDTwoVectorFPToInt<1,1,0b11010, "fcvtpu",int_arm64_neon_fcvtpu>;
2325 defm FCVTXN : SIMDFPInexactCvtTwoVector<1, 0, 0b10110, "fcvtxn",
2326 int_arm64_neon_fcvtxn>;
2327 defm FCVTZS : SIMDTwoVectorFPToInt<0, 1, 0b11011, "fcvtzs", fp_to_sint>;
2328 defm FCVTZU : SIMDTwoVectorFPToInt<1, 1, 0b11011, "fcvtzu", fp_to_uint>;
2329 let isCodeGenOnly = 1 in {
2330 defm FCVTZS_Int : SIMDTwoVectorFPToInt<0, 1, 0b11011, "fcvtzs",
2331 int_arm64_neon_fcvtzs>;
2332 defm FCVTZU_Int : SIMDTwoVectorFPToInt<1, 1, 0b11011, "fcvtzu",
2333 int_arm64_neon_fcvtzu>;
2335 defm FNEG : SIMDTwoVectorFP<1, 1, 0b01111, "fneg", fneg>;
2336 defm FRECPE : SIMDTwoVectorFP<0, 1, 0b11101, "frecpe", int_arm64_neon_frecpe>;
2337 defm FRINTA : SIMDTwoVectorFP<1, 0, 0b11000, "frinta", frnd>;
2338 defm FRINTI : SIMDTwoVectorFP<1, 1, 0b11001, "frinti", fnearbyint>;
2339 defm FRINTM : SIMDTwoVectorFP<0, 0, 0b11001, "frintm", ffloor>;
2340 defm FRINTN : SIMDTwoVectorFP<0, 0, 0b11000, "frintn", int_arm64_neon_frintn>;
2341 defm FRINTP : SIMDTwoVectorFP<0, 1, 0b11000, "frintp", fceil>;
2342 defm FRINTX : SIMDTwoVectorFP<1, 0, 0b11001, "frintx", frint>;
2343 defm FRINTZ : SIMDTwoVectorFP<0, 1, 0b11001, "frintz", ftrunc>;
2344 defm FRSQRTE: SIMDTwoVectorFP<1, 1, 0b11101, "frsqrte", int_arm64_neon_frsqrte>;
2345 defm FSQRT : SIMDTwoVectorFP<1, 1, 0b11111, "fsqrt", fsqrt>;
2346 defm NEG : SIMDTwoVectorBHSD<1, 0b01011, "neg",
2347 UnOpFrag<(sub immAllZerosV, node:$LHS)> >;
2348 defm NOT : SIMDTwoVectorB<1, 0b00, 0b00101, "not", vnot>;
2349 // Aliases for MVN -> NOT.
2350 def : InstAlias<"mvn{ $Vd.8b, $Vn.8b|.8b $Vd, $Vn}",
2351 (NOTv8i8 V64:$Vd, V64:$Vn)>;
2352 def : InstAlias<"mvn{ $Vd.16b, $Vn.16b|.16b $Vd, $Vn}",
2353 (NOTv16i8 V128:$Vd, V128:$Vn)>;
2355 def : Pat<(ARM64neg (v8i8 V64:$Rn)), (NEGv8i8 V64:$Rn)>;
2356 def : Pat<(ARM64neg (v16i8 V128:$Rn)), (NEGv16i8 V128:$Rn)>;
2357 def : Pat<(ARM64neg (v4i16 V64:$Rn)), (NEGv4i16 V64:$Rn)>;
2358 def : Pat<(ARM64neg (v8i16 V128:$Rn)), (NEGv8i16 V128:$Rn)>;
2359 def : Pat<(ARM64neg (v2i32 V64:$Rn)), (NEGv2i32 V64:$Rn)>;
2360 def : Pat<(ARM64neg (v4i32 V128:$Rn)), (NEGv4i32 V128:$Rn)>;
2361 def : Pat<(ARM64neg (v2i64 V128:$Rn)), (NEGv2i64 V128:$Rn)>;
2363 def : Pat<(ARM64not (v8i8 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2364 def : Pat<(ARM64not (v16i8 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2365 def : Pat<(ARM64not (v4i16 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2366 def : Pat<(ARM64not (v8i16 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2367 def : Pat<(ARM64not (v2i32 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2368 def : Pat<(ARM64not (v1i64 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2369 def : Pat<(ARM64not (v4i32 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2370 def : Pat<(ARM64not (v2i64 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2372 def : Pat<(vnot (v4i16 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2373 def : Pat<(vnot (v8i16 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2374 def : Pat<(vnot (v2i32 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2375 def : Pat<(vnot (v4i32 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2376 def : Pat<(vnot (v2i64 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2378 defm RBIT : SIMDTwoVectorB<1, 0b01, 0b00101, "rbit", int_arm64_neon_rbit>;
2379 defm REV16 : SIMDTwoVectorB<0, 0b00, 0b00001, "rev16", ARM64rev16>;
2380 defm REV32 : SIMDTwoVectorBH<1, 0b00000, "rev32", ARM64rev32>;
2381 defm REV64 : SIMDTwoVectorBHS<0, 0b00000, "rev64", ARM64rev64>;
2382 defm SADALP : SIMDLongTwoVectorTied<0, 0b00110, "sadalp",
2383 BinOpFrag<(add node:$LHS, (int_arm64_neon_saddlp node:$RHS))> >;
2384 defm SADDLP : SIMDLongTwoVector<0, 0b00010, "saddlp", int_arm64_neon_saddlp>;
2385 defm SCVTF : SIMDTwoVectorIntToFP<0, 0, 0b11101, "scvtf", sint_to_fp>;
2386 defm SHLL : SIMDVectorLShiftLongBySizeBHS;
2387 defm SQABS : SIMDTwoVectorBHSD<0, 0b00111, "sqabs", int_arm64_neon_sqabs>;
2388 defm SQNEG : SIMDTwoVectorBHSD<1, 0b00111, "sqneg", int_arm64_neon_sqneg>;
2389 defm SQXTN : SIMDMixedTwoVector<0, 0b10100, "sqxtn", int_arm64_neon_sqxtn>;
2390 defm SQXTUN : SIMDMixedTwoVector<1, 0b10010, "sqxtun", int_arm64_neon_sqxtun>;
2391 defm SUQADD : SIMDTwoVectorBHSDTied<0, 0b00011, "suqadd",int_arm64_neon_suqadd>;
2392 defm UADALP : SIMDLongTwoVectorTied<1, 0b00110, "uadalp",
2393 BinOpFrag<(add node:$LHS, (int_arm64_neon_uaddlp node:$RHS))> >;
2394 defm UADDLP : SIMDLongTwoVector<1, 0b00010, "uaddlp",
2395 int_arm64_neon_uaddlp>;
2396 defm UCVTF : SIMDTwoVectorIntToFP<1, 0, 0b11101, "ucvtf", uint_to_fp>;
2397 defm UQXTN : SIMDMixedTwoVector<1, 0b10100, "uqxtn", int_arm64_neon_uqxtn>;
2398 defm URECPE : SIMDTwoVectorS<0, 1, 0b11100, "urecpe", int_arm64_neon_urecpe>;
2399 defm URSQRTE: SIMDTwoVectorS<1, 1, 0b11100, "ursqrte", int_arm64_neon_ursqrte>;
2400 defm USQADD : SIMDTwoVectorBHSDTied<1, 0b00011, "usqadd",int_arm64_neon_usqadd>;
2401 defm XTN : SIMDMixedTwoVector<0, 0b10010, "xtn", trunc>;
2403 def : Pat<(v2f32 (ARM64rev64 V64:$Rn)), (REV64v2i32 V64:$Rn)>;
2404 def : Pat<(v4f32 (ARM64rev64 V128:$Rn)), (REV64v4i32 V128:$Rn)>;
2406 // Patterns for vector long shift (by element width). These need to match all
2407 // three of zext, sext and anyext so it's easier to pull the patterns out of the
2409 multiclass SIMDVectorLShiftLongBySizeBHSPats<SDPatternOperator ext> {
2410 def : Pat<(ARM64vshl (v8i16 (ext (v8i8 V64:$Rn))), (i32 8)),
2411 (SHLLv8i8 V64:$Rn)>;
2412 def : Pat<(ARM64vshl (v8i16 (ext (extract_high_v16i8 V128:$Rn))), (i32 8)),
2413 (SHLLv16i8 V128:$Rn)>;
2414 def : Pat<(ARM64vshl (v4i32 (ext (v4i16 V64:$Rn))), (i32 16)),
2415 (SHLLv4i16 V64:$Rn)>;
2416 def : Pat<(ARM64vshl (v4i32 (ext (extract_high_v8i16 V128:$Rn))), (i32 16)),
2417 (SHLLv8i16 V128:$Rn)>;
2418 def : Pat<(ARM64vshl (v2i64 (ext (v2i32 V64:$Rn))), (i32 32)),
2419 (SHLLv2i32 V64:$Rn)>;
2420 def : Pat<(ARM64vshl (v2i64 (ext (extract_high_v4i32 V128:$Rn))), (i32 32)),
2421 (SHLLv4i32 V128:$Rn)>;
2424 defm : SIMDVectorLShiftLongBySizeBHSPats<anyext>;
2425 defm : SIMDVectorLShiftLongBySizeBHSPats<zext>;
2426 defm : SIMDVectorLShiftLongBySizeBHSPats<sext>;
2428 //===----------------------------------------------------------------------===//
2429 // Advanced SIMD three vector instructions.
2430 //===----------------------------------------------------------------------===//
2432 defm ADD : SIMDThreeSameVector<0, 0b10000, "add", add>;
2433 defm ADDP : SIMDThreeSameVector<0, 0b10111, "addp", int_arm64_neon_addp>;
2434 defm CMEQ : SIMDThreeSameVector<1, 0b10001, "cmeq", ARM64cmeq>;
2435 defm CMGE : SIMDThreeSameVector<0, 0b00111, "cmge", ARM64cmge>;
2436 defm CMGT : SIMDThreeSameVector<0, 0b00110, "cmgt", ARM64cmgt>;
2437 defm CMHI : SIMDThreeSameVector<1, 0b00110, "cmhi", ARM64cmhi>;
2438 defm CMHS : SIMDThreeSameVector<1, 0b00111, "cmhs", ARM64cmhs>;
2439 defm CMTST : SIMDThreeSameVector<0, 0b10001, "cmtst", ARM64cmtst>;
2440 defm FABD : SIMDThreeSameVectorFP<1,1,0b11010,"fabd", int_arm64_neon_fabd>;
2441 defm FACGE : SIMDThreeSameVectorFPCmp<1,0,0b11101,"facge",int_arm64_neon_facge>;
2442 defm FACGT : SIMDThreeSameVectorFPCmp<1,1,0b11101,"facgt",int_arm64_neon_facgt>;
2443 defm FADDP : SIMDThreeSameVectorFP<1,0,0b11010,"faddp",int_arm64_neon_addp>;
2444 defm FADD : SIMDThreeSameVectorFP<0,0,0b11010,"fadd", fadd>;
2445 defm FCMEQ : SIMDThreeSameVectorFPCmp<0, 0, 0b11100, "fcmeq", ARM64fcmeq>;
2446 defm FCMGE : SIMDThreeSameVectorFPCmp<1, 0, 0b11100, "fcmge", ARM64fcmge>;
2447 defm FCMGT : SIMDThreeSameVectorFPCmp<1, 1, 0b11100, "fcmgt", ARM64fcmgt>;
2448 defm FDIV : SIMDThreeSameVectorFP<1,0,0b11111,"fdiv", fdiv>;
2449 defm FMAXNMP : SIMDThreeSameVectorFP<1,0,0b11000,"fmaxnmp", int_arm64_neon_fmaxnmp>;
2450 defm FMAXNM : SIMDThreeSameVectorFP<0,0,0b11000,"fmaxnm", int_arm64_neon_fmaxnm>;
2451 defm FMAXP : SIMDThreeSameVectorFP<1,0,0b11110,"fmaxp", int_arm64_neon_fmaxp>;
2452 defm FMAX : SIMDThreeSameVectorFP<0,0,0b11110,"fmax", ARM64fmax>;
2453 defm FMINNMP : SIMDThreeSameVectorFP<1,1,0b11000,"fminnmp", int_arm64_neon_fminnmp>;
2454 defm FMINNM : SIMDThreeSameVectorFP<0,1,0b11000,"fminnm", int_arm64_neon_fminnm>;
2455 defm FMINP : SIMDThreeSameVectorFP<1,1,0b11110,"fminp", int_arm64_neon_fminp>;
2456 defm FMIN : SIMDThreeSameVectorFP<0,1,0b11110,"fmin", ARM64fmin>;
2458 // NOTE: The operands of the PatFrag are reordered on FMLA/FMLS because the
2459 // instruction expects the addend first, while the fma intrinsic puts it last.
2460 defm FMLA : SIMDThreeSameVectorFPTied<0, 0, 0b11001, "fmla",
2461 TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)> >;
2462 defm FMLS : SIMDThreeSameVectorFPTied<0, 1, 0b11001, "fmls",
2463 TriOpFrag<(fma node:$MHS, (fneg node:$RHS), node:$LHS)> >;
2465 // The following def pats catch the case where the LHS of an FMA is negated.
2466 // The TriOpFrag above catches the case where the middle operand is negated.
2467 def : Pat<(v2f32 (fma (fneg V64:$Rn), V64:$Rm, V64:$Rd)),
2468 (FMLSv2f32 V64:$Rd, V64:$Rn, V64:$Rm)>;
2470 def : Pat<(v4f32 (fma (fneg V128:$Rn), V128:$Rm, V128:$Rd)),
2471 (FMLSv4f32 V128:$Rd, V128:$Rn, V128:$Rm)>;
2473 def : Pat<(v2f64 (fma (fneg V128:$Rn), V128:$Rm, V128:$Rd)),
2474 (FMLSv2f64 V128:$Rd, V128:$Rn, V128:$Rm)>;
2476 defm FMULX : SIMDThreeSameVectorFP<0,0,0b11011,"fmulx", int_arm64_neon_fmulx>;
2477 defm FMUL : SIMDThreeSameVectorFP<1,0,0b11011,"fmul", fmul>;
2478 defm FRECPS : SIMDThreeSameVectorFP<0,0,0b11111,"frecps", int_arm64_neon_frecps>;
2479 defm FRSQRTS : SIMDThreeSameVectorFP<0,1,0b11111,"frsqrts", int_arm64_neon_frsqrts>;
2480 defm FSUB : SIMDThreeSameVectorFP<0,1,0b11010,"fsub", fsub>;
2481 defm MLA : SIMDThreeSameVectorBHSTied<0, 0b10010, "mla",
2482 TriOpFrag<(add node:$LHS, (mul node:$MHS, node:$RHS))> >;
2483 defm MLS : SIMDThreeSameVectorBHSTied<1, 0b10010, "mls",
2484 TriOpFrag<(sub node:$LHS, (mul node:$MHS, node:$RHS))> >;
2485 defm MUL : SIMDThreeSameVectorBHS<0, 0b10011, "mul", mul>;
2486 defm PMUL : SIMDThreeSameVectorB<1, 0b10011, "pmul", int_arm64_neon_pmul>;
2487 defm SABA : SIMDThreeSameVectorBHSTied<0, 0b01111, "saba",
2488 TriOpFrag<(add node:$LHS, (int_arm64_neon_sabd node:$MHS, node:$RHS))> >;
2489 defm SABD : SIMDThreeSameVectorBHS<0,0b01110,"sabd", int_arm64_neon_sabd>;
2490 defm SHADD : SIMDThreeSameVectorBHS<0,0b00000,"shadd", int_arm64_neon_shadd>;
2491 defm SHSUB : SIMDThreeSameVectorBHS<0,0b00100,"shsub", int_arm64_neon_shsub>;
2492 defm SMAXP : SIMDThreeSameVectorBHS<0,0b10100,"smaxp", int_arm64_neon_smaxp>;
2493 defm SMAX : SIMDThreeSameVectorBHS<0,0b01100,"smax", int_arm64_neon_smax>;
2494 defm SMINP : SIMDThreeSameVectorBHS<0,0b10101,"sminp", int_arm64_neon_sminp>;
2495 defm SMIN : SIMDThreeSameVectorBHS<0,0b01101,"smin", int_arm64_neon_smin>;
2496 defm SQADD : SIMDThreeSameVector<0,0b00001,"sqadd", int_arm64_neon_sqadd>;
2497 defm SQDMULH : SIMDThreeSameVectorHS<0,0b10110,"sqdmulh",int_arm64_neon_sqdmulh>;
2498 defm SQRDMULH : SIMDThreeSameVectorHS<1,0b10110,"sqrdmulh",int_arm64_neon_sqrdmulh>;
2499 defm SQRSHL : SIMDThreeSameVector<0,0b01011,"sqrshl", int_arm64_neon_sqrshl>;
2500 defm SQSHL : SIMDThreeSameVector<0,0b01001,"sqshl", int_arm64_neon_sqshl>;
2501 defm SQSUB : SIMDThreeSameVector<0,0b00101,"sqsub", int_arm64_neon_sqsub>;
2502 defm SRHADD : SIMDThreeSameVectorBHS<0,0b00010,"srhadd",int_arm64_neon_srhadd>;
2503 defm SRSHL : SIMDThreeSameVector<0,0b01010,"srshl", int_arm64_neon_srshl>;
2504 defm SSHL : SIMDThreeSameVector<0,0b01000,"sshl", int_arm64_neon_sshl>;
2505 defm SUB : SIMDThreeSameVector<1,0b10000,"sub", sub>;
2506 defm UABA : SIMDThreeSameVectorBHSTied<1, 0b01111, "uaba",
2507 TriOpFrag<(add node:$LHS, (int_arm64_neon_uabd node:$MHS, node:$RHS))> >;
2508 defm UABD : SIMDThreeSameVectorBHS<1,0b01110,"uabd", int_arm64_neon_uabd>;
2509 defm UHADD : SIMDThreeSameVectorBHS<1,0b00000,"uhadd", int_arm64_neon_uhadd>;
2510 defm UHSUB : SIMDThreeSameVectorBHS<1,0b00100,"uhsub", int_arm64_neon_uhsub>;
2511 defm UMAXP : SIMDThreeSameVectorBHS<1,0b10100,"umaxp", int_arm64_neon_umaxp>;
2512 defm UMAX : SIMDThreeSameVectorBHS<1,0b01100,"umax", int_arm64_neon_umax>;
2513 defm UMINP : SIMDThreeSameVectorBHS<1,0b10101,"uminp", int_arm64_neon_uminp>;
2514 defm UMIN : SIMDThreeSameVectorBHS<1,0b01101,"umin", int_arm64_neon_umin>;
2515 defm UQADD : SIMDThreeSameVector<1,0b00001,"uqadd", int_arm64_neon_uqadd>;
2516 defm UQRSHL : SIMDThreeSameVector<1,0b01011,"uqrshl", int_arm64_neon_uqrshl>;
2517 defm UQSHL : SIMDThreeSameVector<1,0b01001,"uqshl", int_arm64_neon_uqshl>;
2518 defm UQSUB : SIMDThreeSameVector<1,0b00101,"uqsub", int_arm64_neon_uqsub>;
2519 defm URHADD : SIMDThreeSameVectorBHS<1,0b00010,"urhadd", int_arm64_neon_urhadd>;
2520 defm URSHL : SIMDThreeSameVector<1,0b01010,"urshl", int_arm64_neon_urshl>;
2521 defm USHL : SIMDThreeSameVector<1,0b01000,"ushl", int_arm64_neon_ushl>;
2523 defm AND : SIMDLogicalThreeVector<0, 0b00, "and", and>;
2524 defm BIC : SIMDLogicalThreeVector<0, 0b01, "bic",
2525 BinOpFrag<(and node:$LHS, (vnot node:$RHS))> >;
2526 defm BIF : SIMDLogicalThreeVector<1, 0b11, "bif">;
2527 defm BIT : SIMDLogicalThreeVectorTied<1, 0b10, "bit", ARM64bit>;
2528 defm BSL : SIMDLogicalThreeVectorTied<1, 0b01, "bsl",
2529 TriOpFrag<(or (and node:$LHS, node:$MHS), (and (vnot node:$LHS), node:$RHS))>>;
2530 defm EOR : SIMDLogicalThreeVector<1, 0b00, "eor", xor>;
2531 defm ORN : SIMDLogicalThreeVector<0, 0b11, "orn",
2532 BinOpFrag<(or node:$LHS, (vnot node:$RHS))> >;
2533 defm ORR : SIMDLogicalThreeVector<0, 0b10, "orr", or>;
2535 def : Pat<(ARM64bsl (v8i8 V64:$Rd), V64:$Rn, V64:$Rm),
2536 (BSLv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
2537 def : Pat<(ARM64bsl (v4i16 V64:$Rd), V64:$Rn, V64:$Rm),
2538 (BSLv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
2539 def : Pat<(ARM64bsl (v2i32 V64:$Rd), V64:$Rn, V64:$Rm),
2540 (BSLv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
2541 def : Pat<(ARM64bsl (v1i64 V64:$Rd), V64:$Rn, V64:$Rm),
2542 (BSLv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
2544 def : Pat<(ARM64bsl (v16i8 V128:$Rd), V128:$Rn, V128:$Rm),
2545 (BSLv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
2546 def : Pat<(ARM64bsl (v8i16 V128:$Rd), V128:$Rn, V128:$Rm),
2547 (BSLv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
2548 def : Pat<(ARM64bsl (v4i32 V128:$Rd), V128:$Rn, V128:$Rm),
2549 (BSLv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
2550 def : Pat<(ARM64bsl (v2i64 V128:$Rd), V128:$Rn, V128:$Rm),
2551 (BSLv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
2553 def : InstAlias<"mov{\t$dst.16b, $src.16b|.16b\t$dst, $src}",
2554 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 1>;
2555 def : InstAlias<"mov{\t$dst.8h, $src.8h|.8h\t$dst, $src}",
2556 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2557 def : InstAlias<"mov{\t$dst.4s, $src.4s|.4s\t$dst, $src}",
2558 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2559 def : InstAlias<"mov{\t$dst.2d, $src.2d|.2d\t$dst, $src}",
2560 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2562 def : InstAlias<"mov{\t$dst.8b, $src.8b|.8b\t$dst, $src}",
2563 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 1>;
2564 def : InstAlias<"mov{\t$dst.4h, $src.4h|.4h\t$dst, $src}",
2565 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2566 def : InstAlias<"mov{\t$dst.2s, $src.2s|.2s\t$dst, $src}",
2567 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2568 def : InstAlias<"mov{\t$dst.1d, $src.1d|.1d\t$dst, $src}",
2569 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2571 def : InstAlias<"{cmls\t$dst.8b, $src1.8b, $src2.8b" #
2572 "|cmls.8b\t$dst, $src1, $src2}",
2573 (CMHSv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2574 def : InstAlias<"{cmls\t$dst.16b, $src1.16b, $src2.16b" #
2575 "|cmls.16b\t$dst, $src1, $src2}",
2576 (CMHSv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2577 def : InstAlias<"{cmls\t$dst.4h, $src1.4h, $src2.4h" #
2578 "|cmls.4h\t$dst, $src1, $src2}",
2579 (CMHSv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2580 def : InstAlias<"{cmls\t$dst.8h, $src1.8h, $src2.8h" #
2581 "|cmls.8h\t$dst, $src1, $src2}",
2582 (CMHSv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2583 def : InstAlias<"{cmls\t$dst.2s, $src1.2s, $src2.2s" #
2584 "|cmls.2s\t$dst, $src1, $src2}",
2585 (CMHSv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2586 def : InstAlias<"{cmls\t$dst.4s, $src1.4s, $src2.4s" #
2587 "|cmls.4s\t$dst, $src1, $src2}",
2588 (CMHSv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2589 def : InstAlias<"{cmls\t$dst.2d, $src1.2d, $src2.2d" #
2590 "|cmls.2d\t$dst, $src1, $src2}",
2591 (CMHSv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2593 def : InstAlias<"{cmlo\t$dst.8b, $src1.8b, $src2.8b" #
2594 "|cmlo.8b\t$dst, $src1, $src2}",
2595 (CMHIv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2596 def : InstAlias<"{cmlo\t$dst.16b, $src1.16b, $src2.16b" #
2597 "|cmlo.16b\t$dst, $src1, $src2}",
2598 (CMHIv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2599 def : InstAlias<"{cmlo\t$dst.4h, $src1.4h, $src2.4h" #
2600 "|cmlo.4h\t$dst, $src1, $src2}",
2601 (CMHIv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2602 def : InstAlias<"{cmlo\t$dst.8h, $src1.8h, $src2.8h" #
2603 "|cmlo.8h\t$dst, $src1, $src2}",
2604 (CMHIv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2605 def : InstAlias<"{cmlo\t$dst.2s, $src1.2s, $src2.2s" #
2606 "|cmlo.2s\t$dst, $src1, $src2}",
2607 (CMHIv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2608 def : InstAlias<"{cmlo\t$dst.4s, $src1.4s, $src2.4s" #
2609 "|cmlo.4s\t$dst, $src1, $src2}",
2610 (CMHIv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2611 def : InstAlias<"{cmlo\t$dst.2d, $src1.2d, $src2.2d" #
2612 "|cmlo.2d\t$dst, $src1, $src2}",
2613 (CMHIv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2615 def : InstAlias<"{cmle\t$dst.8b, $src1.8b, $src2.8b" #
2616 "|cmle.8b\t$dst, $src1, $src2}",
2617 (CMGEv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2618 def : InstAlias<"{cmle\t$dst.16b, $src1.16b, $src2.16b" #
2619 "|cmle.16b\t$dst, $src1, $src2}",
2620 (CMGEv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2621 def : InstAlias<"{cmle\t$dst.4h, $src1.4h, $src2.4h" #
2622 "|cmle.4h\t$dst, $src1, $src2}",
2623 (CMGEv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2624 def : InstAlias<"{cmle\t$dst.8h, $src1.8h, $src2.8h" #
2625 "|cmle.8h\t$dst, $src1, $src2}",
2626 (CMGEv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2627 def : InstAlias<"{cmle\t$dst.2s, $src1.2s, $src2.2s" #
2628 "|cmle.2s\t$dst, $src1, $src2}",
2629 (CMGEv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2630 def : InstAlias<"{cmle\t$dst.4s, $src1.4s, $src2.4s" #
2631 "|cmle.4s\t$dst, $src1, $src2}",
2632 (CMGEv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2633 def : InstAlias<"{cmle\t$dst.2d, $src1.2d, $src2.2d" #
2634 "|cmle.2d\t$dst, $src1, $src2}",
2635 (CMGEv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2637 def : InstAlias<"{cmlt\t$dst.8b, $src1.8b, $src2.8b" #
2638 "|cmlt.8b\t$dst, $src1, $src2}",
2639 (CMGTv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2640 def : InstAlias<"{cmlt\t$dst.16b, $src1.16b, $src2.16b" #
2641 "|cmlt.16b\t$dst, $src1, $src2}",
2642 (CMGTv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2643 def : InstAlias<"{cmlt\t$dst.4h, $src1.4h, $src2.4h" #
2644 "|cmlt.4h\t$dst, $src1, $src2}",
2645 (CMGTv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2646 def : InstAlias<"{cmlt\t$dst.8h, $src1.8h, $src2.8h" #
2647 "|cmlt.8h\t$dst, $src1, $src2}",
2648 (CMGTv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2649 def : InstAlias<"{cmlt\t$dst.2s, $src1.2s, $src2.2s" #
2650 "|cmlt.2s\t$dst, $src1, $src2}",
2651 (CMGTv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2652 def : InstAlias<"{cmlt\t$dst.4s, $src1.4s, $src2.4s" #
2653 "|cmlt.4s\t$dst, $src1, $src2}",
2654 (CMGTv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2655 def : InstAlias<"{cmlt\t$dst.2d, $src1.2d, $src2.2d" #
2656 "|cmlt.2d\t$dst, $src1, $src2}",
2657 (CMGTv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2659 def : InstAlias<"{fcmle\t$dst.2s, $src1.2s, $src2.2s" #
2660 "|fcmle.2s\t$dst, $src1, $src2}",
2661 (FCMGEv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
2662 def : InstAlias<"{fcmle\t$dst.4s, $src1.4s, $src2.4s" #
2663 "|fcmle.4s\t$dst, $src1, $src2}",
2664 (FCMGEv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
2665 def : InstAlias<"{fcmle\t$dst.2d, $src1.2d, $src2.2d" #
2666 "|fcmle.2d\t$dst, $src1, $src2}",
2667 (FCMGEv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
2669 def : InstAlias<"{fcmlt\t$dst.2s, $src1.2s, $src2.2s" #
2670 "|fcmlt.2s\t$dst, $src1, $src2}",
2671 (FCMGTv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
2672 def : InstAlias<"{fcmlt\t$dst.4s, $src1.4s, $src2.4s" #
2673 "|fcmlt.4s\t$dst, $src1, $src2}",
2674 (FCMGTv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
2675 def : InstAlias<"{fcmlt\t$dst.2d, $src1.2d, $src2.2d" #
2676 "|fcmlt.2d\t$dst, $src1, $src2}",
2677 (FCMGTv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
2679 def : InstAlias<"{facle\t$dst.2s, $src1.2s, $src2.2s" #
2680 "|facle.2s\t$dst, $src1, $src2}",
2681 (FACGEv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
2682 def : InstAlias<"{facle\t$dst.4s, $src1.4s, $src2.4s" #
2683 "|facle.4s\t$dst, $src1, $src2}",
2684 (FACGEv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
2685 def : InstAlias<"{facle\t$dst.2d, $src1.2d, $src2.2d" #
2686 "|facle.2d\t$dst, $src1, $src2}",
2687 (FACGEv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
2689 def : InstAlias<"{faclt\t$dst.2s, $src1.2s, $src2.2s" #
2690 "|faclt.2s\t$dst, $src1, $src2}",
2691 (FACGTv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
2692 def : InstAlias<"{faclt\t$dst.4s, $src1.4s, $src2.4s" #
2693 "|faclt.4s\t$dst, $src1, $src2}",
2694 (FACGTv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
2695 def : InstAlias<"{faclt\t$dst.2d, $src1.2d, $src2.2d" #
2696 "|faclt.2d\t$dst, $src1, $src2}",
2697 (FACGTv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
2699 //===----------------------------------------------------------------------===//
2700 // Advanced SIMD three scalar instructions.
2701 //===----------------------------------------------------------------------===//
2703 defm ADD : SIMDThreeScalarD<0, 0b10000, "add", add>;
2704 defm CMEQ : SIMDThreeScalarD<1, 0b10001, "cmeq", ARM64cmeq>;
2705 defm CMGE : SIMDThreeScalarD<0, 0b00111, "cmge", ARM64cmge>;
2706 defm CMGT : SIMDThreeScalarD<0, 0b00110, "cmgt", ARM64cmgt>;
2707 defm CMHI : SIMDThreeScalarD<1, 0b00110, "cmhi", ARM64cmhi>;
2708 defm CMHS : SIMDThreeScalarD<1, 0b00111, "cmhs", ARM64cmhs>;
2709 defm CMTST : SIMDThreeScalarD<0, 0b10001, "cmtst", ARM64cmtst>;
2710 defm FABD : SIMDThreeScalarSD<1, 1, 0b11010, "fabd", int_arm64_sisd_fabd>;
2711 def : Pat<(v1f64 (int_arm64_neon_fabd (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2712 (FABD64 FPR64:$Rn, FPR64:$Rm)>;
2713 defm FACGE : SIMDThreeScalarFPCmp<1, 0, 0b11101, "facge",
2714 int_arm64_neon_facge>;
2715 defm FACGT : SIMDThreeScalarFPCmp<1, 1, 0b11101, "facgt",
2716 int_arm64_neon_facgt>;
2717 defm FCMEQ : SIMDThreeScalarFPCmp<0, 0, 0b11100, "fcmeq", ARM64fcmeq>;
2718 defm FCMGE : SIMDThreeScalarFPCmp<1, 0, 0b11100, "fcmge", ARM64fcmge>;
2719 defm FCMGT : SIMDThreeScalarFPCmp<1, 1, 0b11100, "fcmgt", ARM64fcmgt>;
2720 defm FMULX : SIMDThreeScalarSD<0, 0, 0b11011, "fmulx", int_arm64_neon_fmulx>;
2721 defm FRECPS : SIMDThreeScalarSD<0, 0, 0b11111, "frecps", int_arm64_neon_frecps>;
2722 defm FRSQRTS : SIMDThreeScalarSD<0, 1, 0b11111, "frsqrts", int_arm64_neon_frsqrts>;
2723 defm SQADD : SIMDThreeScalarBHSD<0, 0b00001, "sqadd", int_arm64_neon_sqadd>;
2724 defm SQDMULH : SIMDThreeScalarHS< 0, 0b10110, "sqdmulh", int_arm64_neon_sqdmulh>;
2725 defm SQRDMULH : SIMDThreeScalarHS< 1, 0b10110, "sqrdmulh", int_arm64_neon_sqrdmulh>;
2726 defm SQRSHL : SIMDThreeScalarBHSD<0, 0b01011, "sqrshl",int_arm64_neon_sqrshl>;
2727 defm SQSHL : SIMDThreeScalarBHSD<0, 0b01001, "sqshl", int_arm64_neon_sqshl>;
2728 defm SQSUB : SIMDThreeScalarBHSD<0, 0b00101, "sqsub", int_arm64_neon_sqsub>;
2729 defm SRSHL : SIMDThreeScalarD< 0, 0b01010, "srshl", int_arm64_neon_srshl>;
2730 defm SSHL : SIMDThreeScalarD< 0, 0b01000, "sshl", int_arm64_neon_sshl>;
2731 defm SUB : SIMDThreeScalarD< 1, 0b10000, "sub", sub>;
2732 defm UQADD : SIMDThreeScalarBHSD<1, 0b00001, "uqadd", int_arm64_neon_uqadd>;
2733 defm UQRSHL : SIMDThreeScalarBHSD<1, 0b01011, "uqrshl",int_arm64_neon_uqrshl>;
2734 defm UQSHL : SIMDThreeScalarBHSD<1, 0b01001, "uqshl", int_arm64_neon_uqshl>;
2735 defm UQSUB : SIMDThreeScalarBHSD<1, 0b00101, "uqsub", int_arm64_neon_uqsub>;
2736 defm URSHL : SIMDThreeScalarD< 1, 0b01010, "urshl", int_arm64_neon_urshl>;
2737 defm USHL : SIMDThreeScalarD< 1, 0b01000, "ushl", int_arm64_neon_ushl>;
2739 def : InstAlias<"cmls $dst, $src1, $src2",
2740 (CMHSv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
2741 def : InstAlias<"cmle $dst, $src1, $src2",
2742 (CMGEv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
2743 def : InstAlias<"cmlo $dst, $src1, $src2",
2744 (CMHIv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
2745 def : InstAlias<"cmlt $dst, $src1, $src2",
2746 (CMGTv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
2747 def : InstAlias<"fcmle $dst, $src1, $src2",
2748 (FCMGE32 FPR32:$dst, FPR32:$src2, FPR32:$src1), 0>;
2749 def : InstAlias<"fcmle $dst, $src1, $src2",
2750 (FCMGE64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
2751 def : InstAlias<"fcmlt $dst, $src1, $src2",
2752 (FCMGT32 FPR32:$dst, FPR32:$src2, FPR32:$src1), 0>;
2753 def : InstAlias<"fcmlt $dst, $src1, $src2",
2754 (FCMGT64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
2755 def : InstAlias<"facle $dst, $src1, $src2",
2756 (FACGE32 FPR32:$dst, FPR32:$src2, FPR32:$src1), 0>;
2757 def : InstAlias<"facle $dst, $src1, $src2",
2758 (FACGE64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
2759 def : InstAlias<"faclt $dst, $src1, $src2",
2760 (FACGT32 FPR32:$dst, FPR32:$src2, FPR32:$src1), 0>;
2761 def : InstAlias<"faclt $dst, $src1, $src2",
2762 (FACGT64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
2764 //===----------------------------------------------------------------------===//
2765 // Advanced SIMD three scalar instructions (mixed operands).
2766 //===----------------------------------------------------------------------===//
2767 defm SQDMULL : SIMDThreeScalarMixedHS<0, 0b11010, "sqdmull",
2768 int_arm64_neon_sqdmulls_scalar>;
2769 defm SQDMLAL : SIMDThreeScalarMixedTiedHS<0, 0b10010, "sqdmlal">;
2770 defm SQDMLSL : SIMDThreeScalarMixedTiedHS<0, 0b10110, "sqdmlsl">;
2772 def : Pat<(i64 (int_arm64_neon_sqadd (i64 FPR64:$Rd),
2773 (i64 (int_arm64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
2774 (i32 FPR32:$Rm))))),
2775 (SQDMLALi32 FPR64:$Rd, FPR32:$Rn, FPR32:$Rm)>;
2776 def : Pat<(i64 (int_arm64_neon_sqsub (i64 FPR64:$Rd),
2777 (i64 (int_arm64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
2778 (i32 FPR32:$Rm))))),
2779 (SQDMLSLi32 FPR64:$Rd, FPR32:$Rn, FPR32:$Rm)>;
2781 //===----------------------------------------------------------------------===//
2782 // Advanced SIMD two scalar instructions.
2783 //===----------------------------------------------------------------------===//
2785 defm ABS : SIMDTwoScalarD< 0, 0b01011, "abs", int_arm64_neon_abs>;
2786 defm CMEQ : SIMDCmpTwoScalarD< 0, 0b01001, "cmeq", ARM64cmeqz>;
2787 defm CMGE : SIMDCmpTwoScalarD< 1, 0b01000, "cmge", ARM64cmgez>;
2788 defm CMGT : SIMDCmpTwoScalarD< 0, 0b01000, "cmgt", ARM64cmgtz>;
2789 defm CMLE : SIMDCmpTwoScalarD< 1, 0b01001, "cmle", ARM64cmlez>;
2790 defm CMLT : SIMDCmpTwoScalarD< 0, 0b01010, "cmlt", ARM64cmltz>;
2791 defm FCMEQ : SIMDCmpTwoScalarSD<0, 1, 0b01101, "fcmeq", ARM64fcmeqz>;
2792 defm FCMGE : SIMDCmpTwoScalarSD<1, 1, 0b01100, "fcmge", ARM64fcmgez>;
2793 defm FCMGT : SIMDCmpTwoScalarSD<0, 1, 0b01100, "fcmgt", ARM64fcmgtz>;
2794 defm FCMLE : SIMDCmpTwoScalarSD<1, 1, 0b01101, "fcmle", ARM64fcmlez>;
2795 defm FCMLT : SIMDCmpTwoScalarSD<0, 1, 0b01110, "fcmlt", ARM64fcmltz>;
2796 defm FCVTAS : SIMDTwoScalarSD< 0, 0, 0b11100, "fcvtas">;
2797 defm FCVTAU : SIMDTwoScalarSD< 1, 0, 0b11100, "fcvtau">;
2798 defm FCVTMS : SIMDTwoScalarSD< 0, 0, 0b11011, "fcvtms">;
2799 defm FCVTMU : SIMDTwoScalarSD< 1, 0, 0b11011, "fcvtmu">;
2800 defm FCVTNS : SIMDTwoScalarSD< 0, 0, 0b11010, "fcvtns">;
2801 defm FCVTNU : SIMDTwoScalarSD< 1, 0, 0b11010, "fcvtnu">;
2802 defm FCVTPS : SIMDTwoScalarSD< 0, 1, 0b11010, "fcvtps">;
2803 defm FCVTPU : SIMDTwoScalarSD< 1, 1, 0b11010, "fcvtpu">;
2804 def FCVTXNv1i64 : SIMDInexactCvtTwoScalar<0b10110, "fcvtxn">;
2805 defm FCVTZS : SIMDTwoScalarSD< 0, 1, 0b11011, "fcvtzs">;
2806 defm FCVTZU : SIMDTwoScalarSD< 1, 1, 0b11011, "fcvtzu">;
2807 defm FRECPE : SIMDTwoScalarSD< 0, 1, 0b11101, "frecpe">;
2808 defm FRECPX : SIMDTwoScalarSD< 0, 1, 0b11111, "frecpx">;
2809 defm FRSQRTE : SIMDTwoScalarSD< 1, 1, 0b11101, "frsqrte">;
2810 defm NEG : SIMDTwoScalarD< 1, 0b01011, "neg",
2811 UnOpFrag<(sub immAllZerosV, node:$LHS)> >;
2812 defm SCVTF : SIMDTwoScalarCVTSD< 0, 0, 0b11101, "scvtf", ARM64sitof>;
2813 defm SQABS : SIMDTwoScalarBHSD< 0, 0b00111, "sqabs", int_arm64_neon_sqabs>;
2814 defm SQNEG : SIMDTwoScalarBHSD< 1, 0b00111, "sqneg", int_arm64_neon_sqneg>;
2815 defm SQXTN : SIMDTwoScalarMixedBHS< 0, 0b10100, "sqxtn", int_arm64_neon_scalar_sqxtn>;
2816 defm SQXTUN : SIMDTwoScalarMixedBHS< 1, 0b10010, "sqxtun", int_arm64_neon_scalar_sqxtun>;
2817 defm SUQADD : SIMDTwoScalarBHSDTied< 0, 0b00011, "suqadd",
2818 int_arm64_neon_suqadd>;
2819 defm UCVTF : SIMDTwoScalarCVTSD< 1, 0, 0b11101, "ucvtf", ARM64uitof>;
2820 defm UQXTN : SIMDTwoScalarMixedBHS<1, 0b10100, "uqxtn", int_arm64_neon_scalar_uqxtn>;
2821 defm USQADD : SIMDTwoScalarBHSDTied< 1, 0b00011, "usqadd",
2822 int_arm64_neon_usqadd>;
2824 def : Pat<(ARM64neg (v1i64 V64:$Rn)), (NEGv1i64 V64:$Rn)>;
2826 def : Pat<(v1i64 (int_arm64_neon_fcvtas (v1f64 FPR64:$Rn))),
2827 (FCVTASv1i64 FPR64:$Rn)>;
2828 def : Pat<(v1i64 (int_arm64_neon_fcvtau (v1f64 FPR64:$Rn))),
2829 (FCVTAUv1i64 FPR64:$Rn)>;
2830 def : Pat<(v1i64 (int_arm64_neon_fcvtms (v1f64 FPR64:$Rn))),
2831 (FCVTMSv1i64 FPR64:$Rn)>;
2832 def : Pat<(v1i64 (int_arm64_neon_fcvtmu (v1f64 FPR64:$Rn))),
2833 (FCVTMUv1i64 FPR64:$Rn)>;
2834 def : Pat<(v1i64 (int_arm64_neon_fcvtns (v1f64 FPR64:$Rn))),
2835 (FCVTNSv1i64 FPR64:$Rn)>;
2836 def : Pat<(v1i64 (int_arm64_neon_fcvtnu (v1f64 FPR64:$Rn))),
2837 (FCVTNUv1i64 FPR64:$Rn)>;
2838 def : Pat<(v1i64 (int_arm64_neon_fcvtps (v1f64 FPR64:$Rn))),
2839 (FCVTPSv1i64 FPR64:$Rn)>;
2840 def : Pat<(v1i64 (int_arm64_neon_fcvtpu (v1f64 FPR64:$Rn))),
2841 (FCVTPUv1i64 FPR64:$Rn)>;
2843 def : Pat<(f32 (int_arm64_neon_frecpe (f32 FPR32:$Rn))),
2844 (FRECPEv1i32 FPR32:$Rn)>;
2845 def : Pat<(f64 (int_arm64_neon_frecpe (f64 FPR64:$Rn))),
2846 (FRECPEv1i64 FPR64:$Rn)>;
2847 def : Pat<(v1f64 (int_arm64_neon_frecpe (v1f64 FPR64:$Rn))),
2848 (FRECPEv1i64 FPR64:$Rn)>;
2850 def : Pat<(f32 (int_arm64_neon_frecpx (f32 FPR32:$Rn))),
2851 (FRECPXv1i32 FPR32:$Rn)>;
2852 def : Pat<(f64 (int_arm64_neon_frecpx (f64 FPR64:$Rn))),
2853 (FRECPXv1i64 FPR64:$Rn)>;
2855 def : Pat<(f32 (int_arm64_neon_frsqrte (f32 FPR32:$Rn))),
2856 (FRSQRTEv1i32 FPR32:$Rn)>;
2857 def : Pat<(f64 (int_arm64_neon_frsqrte (f64 FPR64:$Rn))),
2858 (FRSQRTEv1i64 FPR64:$Rn)>;
2859 def : Pat<(v1f64 (int_arm64_neon_frsqrte (v1f64 FPR64:$Rn))),
2860 (FRSQRTEv1i64 FPR64:$Rn)>;
2862 // If an integer is about to be converted to a floating point value,
2863 // just load it on the floating point unit.
2864 // Here are the patterns for 8 and 16-bits to float.
2866 def : Pat <(f32 (uint_to_fp (i32 (zextloadi8 ro_indexed8:$addr)))),
2867 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2868 (LDRBro ro_indexed8:$addr), bsub))>;
2869 def : Pat <(f32 (uint_to_fp (i32 (zextloadi8 am_indexed8:$addr)))),
2870 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2871 (LDRBui am_indexed8:$addr), bsub))>;
2872 def : Pat <(f32 (uint_to_fp (i32 (zextloadi8 am_unscaled8:$addr)))),
2873 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2874 (LDURBi am_unscaled8:$addr), bsub))>;
2875 // 16-bits -> float.
2876 def : Pat <(f32 (uint_to_fp (i32 (zextloadi16 ro_indexed16:$addr)))),
2877 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2878 (LDRHro ro_indexed16:$addr), hsub))>;
2879 def : Pat <(f32 (uint_to_fp (i32 (zextloadi16 am_indexed16:$addr)))),
2880 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2881 (LDRHui am_indexed16:$addr), hsub))>;
2882 def : Pat <(f32 (uint_to_fp (i32 (zextloadi16 am_unscaled16:$addr)))),
2883 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2884 (LDURHi am_unscaled16:$addr), hsub))>;
2885 // 32-bits are handled in target specific dag combine:
2886 // performIntToFpCombine.
2887 // 64-bits integer to 32-bits floating point, not possible with
2888 // UCVTF on floating point registers (both source and destination
2889 // must have the same size).
2891 // Here are the patterns for 8, 16, 32, and 64-bits to double.
2892 // 8-bits -> double.
2893 def : Pat <(f64 (uint_to_fp (i32 (zextloadi8 ro_indexed8:$addr)))),
2894 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2895 (LDRBro ro_indexed8:$addr), bsub))>;
2896 def : Pat <(f64 (uint_to_fp (i32 (zextloadi8 am_indexed8:$addr)))),
2897 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2898 (LDRBui am_indexed8:$addr), bsub))>;
2899 def : Pat <(f64 (uint_to_fp (i32 (zextloadi8 am_unscaled8:$addr)))),
2900 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2901 (LDURBi am_unscaled8:$addr), bsub))>;
2902 // 16-bits -> double.
2903 def : Pat <(f64 (uint_to_fp (i32 (zextloadi16 ro_indexed16:$addr)))),
2904 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2905 (LDRHro ro_indexed16:$addr), hsub))>;
2906 def : Pat <(f64 (uint_to_fp (i32 (zextloadi16 am_indexed16:$addr)))),
2907 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2908 (LDRHui am_indexed16:$addr), hsub))>;
2909 def : Pat <(f64 (uint_to_fp (i32 (zextloadi16 am_unscaled16:$addr)))),
2910 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2911 (LDURHi am_unscaled16:$addr), hsub))>;
2912 // 32-bits -> double.
2913 def : Pat <(f64 (uint_to_fp (i32 (load ro_indexed32:$addr)))),
2914 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2915 (LDRSro ro_indexed32:$addr), ssub))>;
2916 def : Pat <(f64 (uint_to_fp (i32 (load am_indexed32:$addr)))),
2917 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2918 (LDRSui am_indexed32:$addr), ssub))>;
2919 def : Pat <(f64 (uint_to_fp (i32 (load am_unscaled32:$addr)))),
2920 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2921 (LDURSi am_unscaled32:$addr), ssub))>;
2922 // 64-bits -> double are handled in target specific dag combine:
2923 // performIntToFpCombine.
2925 //===----------------------------------------------------------------------===//
2926 // Advanced SIMD three different-sized vector instructions.
2927 //===----------------------------------------------------------------------===//
2929 defm ADDHN : SIMDNarrowThreeVectorBHS<0,0b0100,"addhn", int_arm64_neon_addhn>;
2930 defm SUBHN : SIMDNarrowThreeVectorBHS<0,0b0110,"subhn", int_arm64_neon_subhn>;
2931 defm RADDHN : SIMDNarrowThreeVectorBHS<1,0b0100,"raddhn",int_arm64_neon_raddhn>;
2932 defm RSUBHN : SIMDNarrowThreeVectorBHS<1,0b0110,"rsubhn",int_arm64_neon_rsubhn>;
2933 defm PMULL : SIMDDifferentThreeVectorBD<0,0b1110,"pmull",int_arm64_neon_pmull>;
2934 defm SABAL : SIMDLongThreeVectorTiedBHSabal<0,0b0101,"sabal",
2935 int_arm64_neon_sabd>;
2936 defm SABDL : SIMDLongThreeVectorBHSabdl<0, 0b0111, "sabdl",
2937 int_arm64_neon_sabd>;
2938 defm SADDL : SIMDLongThreeVectorBHS< 0, 0b0000, "saddl",
2939 BinOpFrag<(add (sext node:$LHS), (sext node:$RHS))>>;
2940 defm SADDW : SIMDWideThreeVectorBHS< 0, 0b0001, "saddw",
2941 BinOpFrag<(add node:$LHS, (sext node:$RHS))>>;
2942 defm SMLAL : SIMDLongThreeVectorTiedBHS<0, 0b1000, "smlal",
2943 TriOpFrag<(add node:$LHS, (int_arm64_neon_smull node:$MHS, node:$RHS))>>;
2944 defm SMLSL : SIMDLongThreeVectorTiedBHS<0, 0b1010, "smlsl",
2945 TriOpFrag<(sub node:$LHS, (int_arm64_neon_smull node:$MHS, node:$RHS))>>;
2946 defm SMULL : SIMDLongThreeVectorBHS<0, 0b1100, "smull", int_arm64_neon_smull>;
2947 defm SQDMLAL : SIMDLongThreeVectorSQDMLXTiedHS<0, 0b1001, "sqdmlal",
2948 int_arm64_neon_sqadd>;
2949 defm SQDMLSL : SIMDLongThreeVectorSQDMLXTiedHS<0, 0b1011, "sqdmlsl",
2950 int_arm64_neon_sqsub>;
2951 defm SQDMULL : SIMDLongThreeVectorHS<0, 0b1101, "sqdmull",
2952 int_arm64_neon_sqdmull>;
2953 defm SSUBL : SIMDLongThreeVectorBHS<0, 0b0010, "ssubl",
2954 BinOpFrag<(sub (sext node:$LHS), (sext node:$RHS))>>;
2955 defm SSUBW : SIMDWideThreeVectorBHS<0, 0b0011, "ssubw",
2956 BinOpFrag<(sub node:$LHS, (sext node:$RHS))>>;
2957 defm UABAL : SIMDLongThreeVectorTiedBHSabal<1, 0b0101, "uabal",
2958 int_arm64_neon_uabd>;
2959 defm UABDL : SIMDLongThreeVectorBHSabdl<1, 0b0111, "uabdl",
2960 int_arm64_neon_uabd>;
2961 defm UADDL : SIMDLongThreeVectorBHS<1, 0b0000, "uaddl",
2962 BinOpFrag<(add (zext node:$LHS), (zext node:$RHS))>>;
2963 defm UADDW : SIMDWideThreeVectorBHS<1, 0b0001, "uaddw",
2964 BinOpFrag<(add node:$LHS, (zext node:$RHS))>>;
2965 defm UMLAL : SIMDLongThreeVectorTiedBHS<1, 0b1000, "umlal",
2966 TriOpFrag<(add node:$LHS, (int_arm64_neon_umull node:$MHS, node:$RHS))>>;
2967 defm UMLSL : SIMDLongThreeVectorTiedBHS<1, 0b1010, "umlsl",
2968 TriOpFrag<(sub node:$LHS, (int_arm64_neon_umull node:$MHS, node:$RHS))>>;
2969 defm UMULL : SIMDLongThreeVectorBHS<1, 0b1100, "umull", int_arm64_neon_umull>;
2970 defm USUBL : SIMDLongThreeVectorBHS<1, 0b0010, "usubl",
2971 BinOpFrag<(sub (zext node:$LHS), (zext node:$RHS))>>;
2972 defm USUBW : SIMDWideThreeVectorBHS< 1, 0b0011, "usubw",
2973 BinOpFrag<(sub node:$LHS, (zext node:$RHS))>>;
2975 // Patterns for 64-bit pmull
2976 def : Pat<(int_arm64_neon_pmull64 V64:$Rn, V64:$Rm),
2977 (PMULLv1i64 V64:$Rn, V64:$Rm)>;
2978 def : Pat<(int_arm64_neon_pmull64 (vector_extract (v2i64 V128:$Rn), (i64 1)),
2979 (vector_extract (v2i64 V128:$Rm), (i64 1))),
2980 (PMULLv2i64 V128:$Rn, V128:$Rm)>;
2982 // CodeGen patterns for addhn and subhn instructions, which can actually be
2983 // written in LLVM IR without too much difficulty.
2986 def : Pat<(v8i8 (trunc (v8i16 (ARM64vlshr (add V128:$Rn, V128:$Rm), (i32 8))))),
2987 (ADDHNv8i16_v8i8 V128:$Rn, V128:$Rm)>;
2988 def : Pat<(v4i16 (trunc (v4i32 (ARM64vlshr (add V128:$Rn, V128:$Rm),
2990 (ADDHNv4i32_v4i16 V128:$Rn, V128:$Rm)>;
2991 def : Pat<(v2i32 (trunc (v2i64 (ARM64vlshr (add V128:$Rn, V128:$Rm),
2993 (ADDHNv2i64_v2i32 V128:$Rn, V128:$Rm)>;
2994 def : Pat<(concat_vectors (v8i8 V64:$Rd),
2995 (trunc (v8i16 (ARM64vlshr (add V128:$Rn, V128:$Rm),
2997 (ADDHNv8i16_v16i8 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
2998 V128:$Rn, V128:$Rm)>;
2999 def : Pat<(concat_vectors (v4i16 V64:$Rd),
3000 (trunc (v4i32 (ARM64vlshr (add V128:$Rn, V128:$Rm),
3002 (ADDHNv4i32_v8i16 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3003 V128:$Rn, V128:$Rm)>;
3004 def : Pat<(concat_vectors (v2i32 V64:$Rd),
3005 (trunc (v2i64 (ARM64vlshr (add V128:$Rn, V128:$Rm),
3007 (ADDHNv2i64_v4i32 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3008 V128:$Rn, V128:$Rm)>;
3011 def : Pat<(v8i8 (trunc (v8i16 (ARM64vlshr (sub V128:$Rn, V128:$Rm), (i32 8))))),
3012 (SUBHNv8i16_v8i8 V128:$Rn, V128:$Rm)>;
3013 def : Pat<(v4i16 (trunc (v4i32 (ARM64vlshr (sub V128:$Rn, V128:$Rm),
3015 (SUBHNv4i32_v4i16 V128:$Rn, V128:$Rm)>;
3016 def : Pat<(v2i32 (trunc (v2i64 (ARM64vlshr (sub V128:$Rn, V128:$Rm),
3018 (SUBHNv2i64_v2i32 V128:$Rn, V128:$Rm)>;
3019 def : Pat<(concat_vectors (v8i8 V64:$Rd),
3020 (trunc (v8i16 (ARM64vlshr (sub V128:$Rn, V128:$Rm),
3022 (SUBHNv8i16_v16i8 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3023 V128:$Rn, V128:$Rm)>;
3024 def : Pat<(concat_vectors (v4i16 V64:$Rd),
3025 (trunc (v4i32 (ARM64vlshr (sub V128:$Rn, V128:$Rm),
3027 (SUBHNv4i32_v8i16 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3028 V128:$Rn, V128:$Rm)>;
3029 def : Pat<(concat_vectors (v2i32 V64:$Rd),
3030 (trunc (v2i64 (ARM64vlshr (sub V128:$Rn, V128:$Rm),
3032 (SUBHNv2i64_v4i32 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3033 V128:$Rn, V128:$Rm)>;
3035 //----------------------------------------------------------------------------
3036 // AdvSIMD bitwise extract from vector instruction.
3037 //----------------------------------------------------------------------------
3039 defm EXT : SIMDBitwiseExtract<"ext">;
3041 def : Pat<(v4i16 (ARM64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
3042 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
3043 def : Pat<(v8i16 (ARM64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3044 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3045 def : Pat<(v2i32 (ARM64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
3046 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
3047 def : Pat<(v2f32 (ARM64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
3048 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
3049 def : Pat<(v4i32 (ARM64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3050 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3051 def : Pat<(v4f32 (ARM64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3052 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3053 def : Pat<(v2i64 (ARM64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3054 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3055 def : Pat<(v2f64 (ARM64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3056 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3058 // We use EXT to handle extract_subvector to copy the upper 64-bits of a
3060 def : Pat<(v8i8 (extract_subvector V128:$Rn, (i64 8))),
3061 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3062 def : Pat<(v4i16 (extract_subvector V128:$Rn, (i64 4))),
3063 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3064 def : Pat<(v2i32 (extract_subvector V128:$Rn, (i64 2))),
3065 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3066 def : Pat<(v1i64 (extract_subvector V128:$Rn, (i64 1))),
3067 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3068 def : Pat<(v2f32 (extract_subvector V128:$Rn, (i64 2))),
3069 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3070 def : Pat<(v1f64 (extract_subvector V128:$Rn, (i64 1))),
3071 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3074 //----------------------------------------------------------------------------
3075 // AdvSIMD zip vector
3076 //----------------------------------------------------------------------------
3078 defm TRN1 : SIMDZipVector<0b010, "trn1", ARM64trn1>;
3079 defm TRN2 : SIMDZipVector<0b110, "trn2", ARM64trn2>;
3080 defm UZP1 : SIMDZipVector<0b001, "uzp1", ARM64uzp1>;
3081 defm UZP2 : SIMDZipVector<0b101, "uzp2", ARM64uzp2>;
3082 defm ZIP1 : SIMDZipVector<0b011, "zip1", ARM64zip1>;
3083 defm ZIP2 : SIMDZipVector<0b111, "zip2", ARM64zip2>;
3085 //----------------------------------------------------------------------------
3086 // AdvSIMD TBL/TBX instructions
3087 //----------------------------------------------------------------------------
3089 defm TBL : SIMDTableLookup< 0, "tbl">;
3090 defm TBX : SIMDTableLookupTied<1, "tbx">;
3092 def : Pat<(v8i8 (int_arm64_neon_tbl1 (v16i8 VecListOne128:$Rn), (v8i8 V64:$Ri))),
3093 (TBLv8i8One VecListOne128:$Rn, V64:$Ri)>;
3094 def : Pat<(v16i8 (int_arm64_neon_tbl1 (v16i8 V128:$Ri), (v16i8 V128:$Rn))),
3095 (TBLv16i8One V128:$Ri, V128:$Rn)>;
3097 def : Pat<(v8i8 (int_arm64_neon_tbx1 (v8i8 V64:$Rd),
3098 (v16i8 VecListOne128:$Rn), (v8i8 V64:$Ri))),
3099 (TBXv8i8One V64:$Rd, VecListOne128:$Rn, V64:$Ri)>;
3100 def : Pat<(v16i8 (int_arm64_neon_tbx1 (v16i8 V128:$Rd),
3101 (v16i8 V128:$Ri), (v16i8 V128:$Rn))),
3102 (TBXv16i8One V128:$Rd, V128:$Ri, V128:$Rn)>;
3105 //----------------------------------------------------------------------------
3106 // AdvSIMD scalar CPY instruction
3107 //----------------------------------------------------------------------------
3109 defm CPY : SIMDScalarCPY<"cpy">;
3111 //----------------------------------------------------------------------------
3112 // AdvSIMD scalar pairwise instructions
3113 //----------------------------------------------------------------------------
3115 defm ADDP : SIMDPairwiseScalarD<0, 0b11011, "addp">;
3116 defm FADDP : SIMDPairwiseScalarSD<1, 0, 0b01101, "faddp">;
3117 defm FMAXNMP : SIMDPairwiseScalarSD<1, 0, 0b01100, "fmaxnmp">;
3118 defm FMAXP : SIMDPairwiseScalarSD<1, 0, 0b01111, "fmaxp">;
3119 defm FMINNMP : SIMDPairwiseScalarSD<1, 1, 0b01100, "fminnmp">;
3120 defm FMINP : SIMDPairwiseScalarSD<1, 1, 0b01111, "fminp">;
3121 def : Pat<(i64 (int_arm64_neon_saddv (v2i64 V128:$Rn))),
3122 (ADDPv2i64p V128:$Rn)>;
3123 def : Pat<(i64 (int_arm64_neon_uaddv (v2i64 V128:$Rn))),
3124 (ADDPv2i64p V128:$Rn)>;
3125 def : Pat<(f32 (int_arm64_neon_faddv (v2f32 V64:$Rn))),
3126 (FADDPv2i32p V64:$Rn)>;
3127 def : Pat<(f32 (int_arm64_neon_faddv (v4f32 V128:$Rn))),
3128 (FADDPv2i32p (EXTRACT_SUBREG (FADDPv4f32 V128:$Rn, V128:$Rn), dsub))>;
3129 def : Pat<(f64 (int_arm64_neon_faddv (v2f64 V128:$Rn))),
3130 (FADDPv2i64p V128:$Rn)>;
3131 def : Pat<(f32 (int_arm64_neon_fmaxnmv (v2f32 V64:$Rn))),
3132 (FMAXNMPv2i32p V64:$Rn)>;
3133 def : Pat<(f64 (int_arm64_neon_fmaxnmv (v2f64 V128:$Rn))),
3134 (FMAXNMPv2i64p V128:$Rn)>;
3135 def : Pat<(f32 (int_arm64_neon_fmaxv (v2f32 V64:$Rn))),
3136 (FMAXPv2i32p V64:$Rn)>;
3137 def : Pat<(f64 (int_arm64_neon_fmaxv (v2f64 V128:$Rn))),
3138 (FMAXPv2i64p V128:$Rn)>;
3139 def : Pat<(f32 (int_arm64_neon_fminnmv (v2f32 V64:$Rn))),
3140 (FMINNMPv2i32p V64:$Rn)>;
3141 def : Pat<(f64 (int_arm64_neon_fminnmv (v2f64 V128:$Rn))),
3142 (FMINNMPv2i64p V128:$Rn)>;
3143 def : Pat<(f32 (int_arm64_neon_fminv (v2f32 V64:$Rn))),
3144 (FMINPv2i32p V64:$Rn)>;
3145 def : Pat<(f64 (int_arm64_neon_fminv (v2f64 V128:$Rn))),
3146 (FMINPv2i64p V128:$Rn)>;
3148 //----------------------------------------------------------------------------
3149 // AdvSIMD INS/DUP instructions
3150 //----------------------------------------------------------------------------
3152 def DUPv8i8gpr : SIMDDupFromMain<0, 0b00001, ".8b", v8i8, V64, GPR32>;
3153 def DUPv16i8gpr : SIMDDupFromMain<1, 0b00001, ".16b", v16i8, V128, GPR32>;
3154 def DUPv4i16gpr : SIMDDupFromMain<0, 0b00010, ".4h", v4i16, V64, GPR32>;
3155 def DUPv8i16gpr : SIMDDupFromMain<1, 0b00010, ".8h", v8i16, V128, GPR32>;
3156 def DUPv2i32gpr : SIMDDupFromMain<0, 0b00100, ".2s", v2i32, V64, GPR32>;
3157 def DUPv4i32gpr : SIMDDupFromMain<1, 0b00100, ".4s", v4i32, V128, GPR32>;
3158 def DUPv2i64gpr : SIMDDupFromMain<1, 0b01000, ".2d", v2i64, V128, GPR64>;
3160 def DUPv2i64lane : SIMDDup64FromElement;
3161 def DUPv2i32lane : SIMDDup32FromElement<0, ".2s", v2i32, V64>;
3162 def DUPv4i32lane : SIMDDup32FromElement<1, ".4s", v4i32, V128>;
3163 def DUPv4i16lane : SIMDDup16FromElement<0, ".4h", v4i16, V64>;
3164 def DUPv8i16lane : SIMDDup16FromElement<1, ".8h", v8i16, V128>;
3165 def DUPv8i8lane : SIMDDup8FromElement <0, ".8b", v8i8, V64>;
3166 def DUPv16i8lane : SIMDDup8FromElement <1, ".16b", v16i8, V128>;
3168 def : Pat<(v2f32 (ARM64dup (f32 FPR32:$Rn))),
3169 (v2f32 (DUPv2i32lane
3170 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rn, ssub),
3172 def : Pat<(v4f32 (ARM64dup (f32 FPR32:$Rn))),
3173 (v4f32 (DUPv4i32lane
3174 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rn, ssub),
3176 def : Pat<(v2f64 (ARM64dup (f64 FPR64:$Rn))),
3177 (v2f64 (DUPv2i64lane
3178 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$Rn, dsub),
3181 def : Pat<(v2f32 (ARM64duplane32 (v4f32 V128:$Rn), VectorIndexS:$imm)),
3182 (DUPv2i32lane V128:$Rn, VectorIndexS:$imm)>;
3183 def : Pat<(v4f32 (ARM64duplane32 (v4f32 V128:$Rn), VectorIndexS:$imm)),
3184 (DUPv4i32lane V128:$Rn, VectorIndexS:$imm)>;
3185 def : Pat<(v2f64 (ARM64duplane64 (v2f64 V128:$Rn), VectorIndexD:$imm)),
3186 (DUPv2i64lane V128:$Rn, VectorIndexD:$imm)>;
3188 // If there's an (ARM64dup (vector_extract ...) ...), we can use a duplane
3189 // instruction even if the types don't match: we just have to remap the lane
3190 // carefully. N.b. this trick only applies to truncations.
3191 def VecIndex_x2 : SDNodeXForm<imm, [{
3192 return CurDAG->getTargetConstant(2 * N->getZExtValue(), MVT::i64);
3194 def VecIndex_x4 : SDNodeXForm<imm, [{
3195 return CurDAG->getTargetConstant(4 * N->getZExtValue(), MVT::i64);
3197 def VecIndex_x8 : SDNodeXForm<imm, [{
3198 return CurDAG->getTargetConstant(8 * N->getZExtValue(), MVT::i64);
3201 multiclass DUPWithTruncPats<ValueType ResVT, ValueType Src64VT,
3202 ValueType Src128VT, ValueType ScalVT,
3203 Instruction DUP, SDNodeXForm IdxXFORM> {
3204 def : Pat<(ResVT (ARM64dup (ScalVT (vector_extract (Src128VT V128:$Rn),
3206 (DUP V128:$Rn, (IdxXFORM imm:$idx))>;
3208 def : Pat<(ResVT (ARM64dup (ScalVT (vector_extract (Src64VT V64:$Rn),
3210 (DUP (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), (IdxXFORM imm:$idx))>;
3213 defm : DUPWithTruncPats<v8i8, v4i16, v8i16, i32, DUPv8i8lane, VecIndex_x2>;
3214 defm : DUPWithTruncPats<v8i8, v2i32, v4i32, i32, DUPv8i8lane, VecIndex_x4>;
3215 defm : DUPWithTruncPats<v4i16, v2i32, v4i32, i32, DUPv4i16lane, VecIndex_x2>;
3217 defm : DUPWithTruncPats<v16i8, v4i16, v8i16, i32, DUPv16i8lane, VecIndex_x2>;
3218 defm : DUPWithTruncPats<v16i8, v2i32, v4i32, i32, DUPv16i8lane, VecIndex_x4>;
3219 defm : DUPWithTruncPats<v8i16, v2i32, v4i32, i32, DUPv8i16lane, VecIndex_x2>;
3221 multiclass DUPWithTrunci64Pats<ValueType ResVT, Instruction DUP,
3222 SDNodeXForm IdxXFORM> {
3223 def : Pat<(ResVT (ARM64dup (i32 (trunc (vector_extract (v2i64 V128:$Rn),
3225 (DUP V128:$Rn, (IdxXFORM imm:$idx))>;
3227 def : Pat<(ResVT (ARM64dup (i32 (trunc (vector_extract (v1i64 V64:$Rn),
3229 (DUP (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), (IdxXFORM imm:$idx))>;
3232 defm : DUPWithTrunci64Pats<v8i8, DUPv8i8lane, VecIndex_x8>;
3233 defm : DUPWithTrunci64Pats<v4i16, DUPv4i16lane, VecIndex_x4>;
3234 defm : DUPWithTrunci64Pats<v2i32, DUPv2i32lane, VecIndex_x2>;
3236 defm : DUPWithTrunci64Pats<v16i8, DUPv16i8lane, VecIndex_x8>;
3237 defm : DUPWithTrunci64Pats<v8i16, DUPv8i16lane, VecIndex_x4>;
3238 defm : DUPWithTrunci64Pats<v4i32, DUPv4i32lane, VecIndex_x2>;
3240 // SMOV and UMOV definitions, with some extra patterns for convenience
3244 def : Pat<(sext_inreg (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx), i8),
3245 (i32 (SMOVvi8to32 V128:$Rn, VectorIndexB:$idx))>;
3246 def : Pat<(sext_inreg (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx), i8),
3247 (i64 (SMOVvi8to64 V128:$Rn, VectorIndexB:$idx))>;
3248 def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
3249 (i32 (SMOVvi16to32 V128:$Rn, VectorIndexH:$idx))>;
3250 def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
3251 (i64 (SMOVvi16to64 V128:$Rn, VectorIndexH:$idx))>;
3252 def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
3253 (i32 (SMOVvi16to32 V128:$Rn, VectorIndexH:$idx))>;
3254 def : Pat<(sext (i32 (vector_extract (v4i32 V128:$Rn), VectorIndexS:$idx))),
3255 (i64 (SMOVvi32to64 V128:$Rn, VectorIndexS:$idx))>;
3257 // Extracting i8 or i16 elements will have the zero-extend transformed to
3258 // an 'and' mask by type legalization since neither i8 nor i16 are legal types
3259 // for ARM64. Match these patterns here since UMOV already zeroes out the high
3260 // bits of the destination register.
3261 def : Pat<(and (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx),
3263 (i32 (UMOVvi8 V128:$Rn, VectorIndexB:$idx))>;
3264 def : Pat<(and (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),
3266 (i32 (UMOVvi16 V128:$Rn, VectorIndexH:$idx))>;
3270 def : Pat<(v16i8 (scalar_to_vector GPR32:$Rn)),
3271 (SUBREG_TO_REG (i32 0),
3272 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3273 def : Pat<(v8i8 (scalar_to_vector GPR32:$Rn)),
3274 (SUBREG_TO_REG (i32 0),
3275 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3277 def : Pat<(v8i16 (scalar_to_vector GPR32:$Rn)),
3278 (SUBREG_TO_REG (i32 0),
3279 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3280 def : Pat<(v4i16 (scalar_to_vector GPR32:$Rn)),
3281 (SUBREG_TO_REG (i32 0),
3282 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3284 def : Pat<(v2i32 (scalar_to_vector (i32 FPR32:$Rn))),
3285 (v2i32 (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)),
3286 (i32 FPR32:$Rn), ssub))>;
3287 def : Pat<(v4i32 (scalar_to_vector (i32 FPR32:$Rn))),
3288 (v4i32 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
3289 (i32 FPR32:$Rn), ssub))>;
3290 def : Pat<(v2i64 (scalar_to_vector (i64 FPR64:$Rn))),
3291 (v2i64 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
3292 (i64 FPR64:$Rn), dsub))>;
3294 def : Pat<(v4f32 (scalar_to_vector (f32 FPR32:$Rn))),
3295 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rn, ssub)>;
3296 def : Pat<(v2f32 (scalar_to_vector (f32 FPR32:$Rn))),
3297 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), FPR32:$Rn, ssub)>;
3298 def : Pat<(v2f64 (scalar_to_vector (f64 FPR64:$Rn))),
3299 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$Rn, dsub)>;
3301 def : Pat<(v2f32 (vector_insert (v2f32 V64:$Rn),
3302 (f32 FPR32:$Rm), (i64 VectorIndexS:$imm))),
3305 (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), V64:$Rn, dsub)),
3307 (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rm, ssub)),
3310 def : Pat<(v4f32 (vector_insert (v4f32 V128:$Rn),
3311 (f32 FPR32:$Rm), (i64 VectorIndexS:$imm))),
3313 V128:$Rn, VectorIndexS:$imm,
3314 (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rm, ssub)),
3316 def : Pat<(v2f64 (vector_insert (v2f64 V128:$Rn),
3317 (f64 FPR64:$Rm), (i64 VectorIndexD:$imm))),
3319 V128:$Rn, VectorIndexD:$imm,
3320 (v2f64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$Rm, dsub)),
3323 // Copy an element at a constant index in one vector into a constant indexed
3324 // element of another.
3325 // FIXME refactor to a shared class/dev parameterized on vector type, vector
3326 // index type and INS extension
3327 def : Pat<(v16i8 (int_arm64_neon_vcopy_lane
3328 (v16i8 V128:$Vd), VectorIndexB:$idx, (v16i8 V128:$Vs),
3329 VectorIndexB:$idx2)),
3331 V128:$Vd, VectorIndexB:$idx, V128:$Vs, VectorIndexB:$idx2)
3333 def : Pat<(v8i16 (int_arm64_neon_vcopy_lane
3334 (v8i16 V128:$Vd), VectorIndexH:$idx, (v8i16 V128:$Vs),
3335 VectorIndexH:$idx2)),
3337 V128:$Vd, VectorIndexH:$idx, V128:$Vs, VectorIndexH:$idx2)
3339 def : Pat<(v4i32 (int_arm64_neon_vcopy_lane
3340 (v4i32 V128:$Vd), VectorIndexS:$idx, (v4i32 V128:$Vs),
3341 VectorIndexS:$idx2)),
3343 V128:$Vd, VectorIndexS:$idx, V128:$Vs, VectorIndexS:$idx2)
3345 def : Pat<(v2i64 (int_arm64_neon_vcopy_lane
3346 (v2i64 V128:$Vd), VectorIndexD:$idx, (v2i64 V128:$Vs),
3347 VectorIndexD:$idx2)),
3349 V128:$Vd, VectorIndexD:$idx, V128:$Vs, VectorIndexD:$idx2)
3352 multiclass Neon_INS_elt_pattern<ValueType VT128, ValueType VT64,
3353 ValueType VTScal, Instruction INS> {
3354 def : Pat<(VT128 (vector_insert V128:$src,
3355 (VTScal (vector_extract (VT128 V128:$Rn), imm:$Immn)),
3357 (INS V128:$src, imm:$Immd, V128:$Rn, imm:$Immn)>;
3359 def : Pat<(VT128 (vector_insert V128:$src,
3360 (VTScal (vector_extract (VT64 V64:$Rn), imm:$Immn)),
3362 (INS V128:$src, imm:$Immd,
3363 (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), imm:$Immn)>;
3365 def : Pat<(VT64 (vector_insert V64:$src,
3366 (VTScal (vector_extract (VT128 V128:$Rn), imm:$Immn)),
3368 (EXTRACT_SUBREG (INS (SUBREG_TO_REG (i64 0), V64:$src, dsub),
3369 imm:$Immd, V128:$Rn, imm:$Immn),
3372 def : Pat<(VT64 (vector_insert V64:$src,
3373 (VTScal (vector_extract (VT64 V64:$Rn), imm:$Immn)),
3376 (INS (SUBREG_TO_REG (i64 0), V64:$src, dsub), imm:$Immd,
3377 (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), imm:$Immn),
3381 defm : Neon_INS_elt_pattern<v4f32, v2f32, f32, INSvi32lane>;
3382 defm : Neon_INS_elt_pattern<v2f64, v1f64, f64, INSvi64lane>;
3383 defm : Neon_INS_elt_pattern<v16i8, v8i8, i32, INSvi8lane>;
3384 defm : Neon_INS_elt_pattern<v8i16, v4i16, i32, INSvi16lane>;
3385 defm : Neon_INS_elt_pattern<v4i32, v2i32, i32, INSvi32lane>;
3386 defm : Neon_INS_elt_pattern<v2i64, v1i64, i64, INSvi32lane>;
3389 // Floating point vector extractions are codegen'd as either a sequence of
3390 // subregister extractions, possibly fed by an INS if the lane number is
3391 // anything other than zero.
3392 def : Pat<(vector_extract (v2f64 V128:$Rn), 0),
3393 (f64 (EXTRACT_SUBREG V128:$Rn, dsub))>;
3394 def : Pat<(vector_extract (v4f32 V128:$Rn), 0),
3395 (f32 (EXTRACT_SUBREG V128:$Rn, ssub))>;
3396 def : Pat<(vector_extract (v2f64 V128:$Rn), VectorIndexD:$idx),
3397 (f64 (EXTRACT_SUBREG
3398 (INSvi64lane (v2f64 (IMPLICIT_DEF)), 0,
3399 V128:$Rn, VectorIndexD:$idx),
3401 def : Pat<(vector_extract (v4f32 V128:$Rn), VectorIndexS:$idx),
3402 (f32 (EXTRACT_SUBREG
3403 (INSvi32lane (v4f32 (IMPLICIT_DEF)), 0,
3404 V128:$Rn, VectorIndexS:$idx),
3407 // All concat_vectors operations are canonicalised to act on i64 vectors for
3408 // ARM64. In the general case we need an instruction, which had just as well be
3410 class ConcatPat<ValueType DstTy, ValueType SrcTy>
3411 : Pat<(DstTy (concat_vectors (SrcTy V64:$Rd), V64:$Rn)),
3412 (INSvi64lane (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), 1,
3413 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rn, dsub), 0)>;
3415 def : ConcatPat<v2i64, v1i64>;
3416 def : ConcatPat<v2f64, v1f64>;
3417 def : ConcatPat<v4i32, v2i32>;
3418 def : ConcatPat<v4f32, v2f32>;
3419 def : ConcatPat<v8i16, v4i16>;
3420 def : ConcatPat<v16i8, v8i8>;
3422 // If the high lanes are undef, though, we can just ignore them:
3423 class ConcatUndefPat<ValueType DstTy, ValueType SrcTy>
3424 : Pat<(DstTy (concat_vectors (SrcTy V64:$Rn), undef)),
3425 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rn, dsub)>;
3427 def : ConcatUndefPat<v2i64, v1i64>;
3428 def : ConcatUndefPat<v2f64, v1f64>;
3429 def : ConcatUndefPat<v4i32, v2i32>;
3430 def : ConcatUndefPat<v4f32, v2f32>;
3431 def : ConcatUndefPat<v8i16, v4i16>;
3432 def : ConcatUndefPat<v16i8, v8i8>;
3434 //----------------------------------------------------------------------------
3435 // AdvSIMD across lanes instructions
3436 //----------------------------------------------------------------------------
3438 defm ADDV : SIMDAcrossLanesBHS<0, 0b11011, "addv">;
3439 defm SMAXV : SIMDAcrossLanesBHS<0, 0b01010, "smaxv">;
3440 defm SMINV : SIMDAcrossLanesBHS<0, 0b11010, "sminv">;
3441 defm UMAXV : SIMDAcrossLanesBHS<1, 0b01010, "umaxv">;
3442 defm UMINV : SIMDAcrossLanesBHS<1, 0b11010, "uminv">;
3443 defm SADDLV : SIMDAcrossLanesHSD<0, 0b00011, "saddlv">;
3444 defm UADDLV : SIMDAcrossLanesHSD<1, 0b00011, "uaddlv">;
3445 defm FMAXNMV : SIMDAcrossLanesS<0b01100, 0, "fmaxnmv", int_arm64_neon_fmaxnmv>;
3446 defm FMAXV : SIMDAcrossLanesS<0b01111, 0, "fmaxv", int_arm64_neon_fmaxv>;
3447 defm FMINNMV : SIMDAcrossLanesS<0b01100, 1, "fminnmv", int_arm64_neon_fminnmv>;
3448 defm FMINV : SIMDAcrossLanesS<0b01111, 1, "fminv", int_arm64_neon_fminv>;
3450 multiclass SIMDAcrossLanesSignedIntrinsic<string baseOpc, Intrinsic intOp> {
3451 // If there is a sign extension after this intrinsic, consume it as smov already
3453 def : Pat<(i32 (sext_inreg (i32 (intOp (v8i8 V64:$Rn))), i8)),
3455 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3456 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3458 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
3460 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3461 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3463 // If there is a sign extension after this intrinsic, consume it as smov already
3465 def : Pat<(i32 (sext_inreg (i32 (intOp (v16i8 V128:$Rn))), i8)),
3467 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3468 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3470 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
3472 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3473 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3475 // If there is a sign extension after this intrinsic, consume it as smov already
3477 def : Pat<(i32 (sext_inreg (i32 (intOp (v4i16 V64:$Rn))), i16)),
3479 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3480 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3482 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
3484 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3485 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3487 // If there is a sign extension after this intrinsic, consume it as smov already
3489 def : Pat<(i32 (sext_inreg (i32 (intOp (v8i16 V128:$Rn))), i16)),
3491 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3492 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3494 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
3496 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3497 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3500 def : Pat<(i32 (intOp (v4i32 V128:$Rn))),
3501 (i32 (EXTRACT_SUBREG
3502 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3503 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), ssub),
3507 multiclass SIMDAcrossLanesUnsignedIntrinsic<string baseOpc, Intrinsic intOp> {
3508 // If there is a masking operation keeping only what has been actually
3509 // generated, consume it.
3510 def : Pat<(i32 (and (i32 (intOp (v8i8 V64:$Rn))), maski8_or_more)),
3511 (i32 (EXTRACT_SUBREG
3512 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3513 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3515 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
3516 (i32 (EXTRACT_SUBREG
3517 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3518 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3520 // If there is a masking operation keeping only what has been actually
3521 // generated, consume it.
3522 def : Pat<(i32 (and (i32 (intOp (v16i8 V128:$Rn))), maski8_or_more)),
3523 (i32 (EXTRACT_SUBREG
3524 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3525 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3527 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
3528 (i32 (EXTRACT_SUBREG
3529 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3530 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3533 // If there is a masking operation keeping only what has been actually
3534 // generated, consume it.
3535 def : Pat<(i32 (and (i32 (intOp (v4i16 V64:$Rn))), maski16_or_more)),
3536 (i32 (EXTRACT_SUBREG
3537 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3538 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3540 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
3541 (i32 (EXTRACT_SUBREG
3542 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3543 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3545 // If there is a masking operation keeping only what has been actually
3546 // generated, consume it.
3547 def : Pat<(i32 (and (i32 (intOp (v8i16 V128:$Rn))), maski16_or_more)),
3548 (i32 (EXTRACT_SUBREG
3549 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3550 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3552 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
3553 (i32 (EXTRACT_SUBREG
3554 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3555 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3558 def : Pat<(i32 (intOp (v4i32 V128:$Rn))),
3559 (i32 (EXTRACT_SUBREG
3560 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3561 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), ssub),
3566 multiclass SIMDAcrossLanesSignedLongIntrinsic<string baseOpc, Intrinsic intOp> {
3567 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
3569 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3570 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), hsub),
3572 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
3574 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3575 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), hsub),
3578 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
3579 (i32 (EXTRACT_SUBREG
3580 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3581 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), ssub),
3583 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
3584 (i32 (EXTRACT_SUBREG
3585 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3586 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), ssub),
3589 def : Pat<(i64 (intOp (v4i32 V128:$Rn))),
3590 (i64 (EXTRACT_SUBREG
3591 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3592 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), dsub),
3596 multiclass SIMDAcrossLanesUnsignedLongIntrinsic<string baseOpc,
3598 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
3599 (i32 (EXTRACT_SUBREG
3600 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3601 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), hsub),
3603 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
3604 (i32 (EXTRACT_SUBREG
3605 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3606 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), hsub),
3609 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
3610 (i32 (EXTRACT_SUBREG
3611 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3612 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), ssub),
3614 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
3615 (i32 (EXTRACT_SUBREG
3616 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3617 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), ssub),
3620 def : Pat<(i64 (intOp (v4i32 V128:$Rn))),
3621 (i64 (EXTRACT_SUBREG
3622 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3623 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), dsub),
3627 defm : SIMDAcrossLanesSignedIntrinsic<"ADDV", int_arm64_neon_saddv>;
3628 // vaddv_[su]32 is special; -> ADDP Vd.2S,Vn.2S,Vm.2S; return Vd.s[0];Vn==Vm
3629 def : Pat<(i32 (int_arm64_neon_saddv (v2i32 V64:$Rn))),
3630 (EXTRACT_SUBREG (ADDPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3632 defm : SIMDAcrossLanesUnsignedIntrinsic<"ADDV", int_arm64_neon_uaddv>;
3633 // vaddv_[su]32 is special; -> ADDP Vd.2S,Vn.2S,Vm.2S; return Vd.s[0];Vn==Vm
3634 def : Pat<(i32 (int_arm64_neon_uaddv (v2i32 V64:$Rn))),
3635 (EXTRACT_SUBREG (ADDPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3637 defm : SIMDAcrossLanesSignedIntrinsic<"SMAXV", int_arm64_neon_smaxv>;
3638 def : Pat<(i32 (int_arm64_neon_smaxv (v2i32 V64:$Rn))),
3639 (EXTRACT_SUBREG (SMAXPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3641 defm : SIMDAcrossLanesSignedIntrinsic<"SMINV", int_arm64_neon_sminv>;
3642 def : Pat<(i32 (int_arm64_neon_sminv (v2i32 V64:$Rn))),
3643 (EXTRACT_SUBREG (SMINPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3645 defm : SIMDAcrossLanesUnsignedIntrinsic<"UMAXV", int_arm64_neon_umaxv>;
3646 def : Pat<(i32 (int_arm64_neon_umaxv (v2i32 V64:$Rn))),
3647 (EXTRACT_SUBREG (UMAXPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3649 defm : SIMDAcrossLanesUnsignedIntrinsic<"UMINV", int_arm64_neon_uminv>;
3650 def : Pat<(i32 (int_arm64_neon_uminv (v2i32 V64:$Rn))),
3651 (EXTRACT_SUBREG (UMINPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3653 defm : SIMDAcrossLanesSignedLongIntrinsic<"SADDLV", int_arm64_neon_saddlv>;
3654 defm : SIMDAcrossLanesUnsignedLongIntrinsic<"UADDLV", int_arm64_neon_uaddlv>;
3656 // The vaddlv_s32 intrinsic gets mapped to SADDLP.
3657 def : Pat<(i64 (int_arm64_neon_saddlv (v2i32 V64:$Rn))),
3658 (i64 (EXTRACT_SUBREG
3659 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3660 (SADDLPv2i32_v1i64 V64:$Rn), dsub),
3662 // The vaddlv_u32 intrinsic gets mapped to UADDLP.
3663 def : Pat<(i64 (int_arm64_neon_uaddlv (v2i32 V64:$Rn))),
3664 (i64 (EXTRACT_SUBREG
3665 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3666 (UADDLPv2i32_v1i64 V64:$Rn), dsub),
3669 //------------------------------------------------------------------------------
3670 // AdvSIMD modified immediate instructions
3671 //------------------------------------------------------------------------------
3674 defm BIC : SIMDModifiedImmVectorShiftTied<1, 0b11, 0b01, "bic", ARM64bici>;
3676 defm ORR : SIMDModifiedImmVectorShiftTied<0, 0b11, 0b01, "orr", ARM64orri>;
3678 def : InstAlias<"bic $Vd.4h, $imm", (BICv4i16 V64:$Vd, imm0_255:$imm, 0)>;
3679 def : InstAlias<"bic $Vd.8h, $imm", (BICv8i16 V128:$Vd, imm0_255:$imm, 0)>;
3680 def : InstAlias<"bic $Vd.2s, $imm", (BICv2i32 V64:$Vd, imm0_255:$imm, 0)>;
3681 def : InstAlias<"bic $Vd.4s, $imm", (BICv4i32 V128:$Vd, imm0_255:$imm, 0)>;
3683 def : InstAlias<"bic.4h $Vd, $imm", (BICv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
3684 def : InstAlias<"bic.8h $Vd, $imm", (BICv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
3685 def : InstAlias<"bic.2s $Vd, $imm", (BICv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
3686 def : InstAlias<"bic.4s $Vd, $imm", (BICv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
3688 def : InstAlias<"orr $Vd.4h, $imm", (ORRv4i16 V64:$Vd, imm0_255:$imm, 0)>;
3689 def : InstAlias<"orr $Vd.8h, $imm", (ORRv8i16 V128:$Vd, imm0_255:$imm, 0)>;
3690 def : InstAlias<"orr $Vd.2s, $imm", (ORRv2i32 V64:$Vd, imm0_255:$imm, 0)>;
3691 def : InstAlias<"orr $Vd.4s, $imm", (ORRv4i32 V128:$Vd, imm0_255:$imm, 0)>;
3693 def : InstAlias<"orr.4h $Vd, $imm", (ORRv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
3694 def : InstAlias<"orr.8h $Vd, $imm", (ORRv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
3695 def : InstAlias<"orr.2s $Vd, $imm", (ORRv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
3696 def : InstAlias<"orr.4s $Vd, $imm", (ORRv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
3699 def FMOVv2f64_ns : SIMDModifiedImmVectorNoShift<1, 1, 0b1111, V128, fpimm8,
3701 [(set (v2f64 V128:$Rd), (ARM64fmov imm0_255:$imm8))]>;
3702 def FMOVv2f32_ns : SIMDModifiedImmVectorNoShift<0, 0, 0b1111, V64, fpimm8,
3704 [(set (v2f32 V64:$Rd), (ARM64fmov imm0_255:$imm8))]>;
3705 def FMOVv4f32_ns : SIMDModifiedImmVectorNoShift<1, 0, 0b1111, V128, fpimm8,
3707 [(set (v4f32 V128:$Rd), (ARM64fmov imm0_255:$imm8))]>;
3711 // EDIT byte mask: scalar
3712 let isReMaterializable = 1, isAsCheapAsAMove = 1 in
3713 def MOVID : SIMDModifiedImmScalarNoShift<0, 1, 0b1110, "movi",
3714 [(set FPR64:$Rd, simdimmtype10:$imm8)]>;
3715 // The movi_edit node has the immediate value already encoded, so we use
3716 // a plain imm0_255 here.
3717 def : Pat<(f64 (ARM64movi_edit imm0_255:$shift)),
3718 (MOVID imm0_255:$shift)>;
3720 def : Pat<(v1i64 immAllZerosV), (MOVID (i32 0))>;
3721 def : Pat<(v2i32 immAllZerosV), (MOVID (i32 0))>;
3722 def : Pat<(v4i16 immAllZerosV), (MOVID (i32 0))>;
3723 def : Pat<(v8i8 immAllZerosV), (MOVID (i32 0))>;
3725 def : Pat<(v1i64 immAllOnesV), (MOVID (i32 255))>;
3726 def : Pat<(v2i32 immAllOnesV), (MOVID (i32 255))>;
3727 def : Pat<(v4i16 immAllOnesV), (MOVID (i32 255))>;
3728 def : Pat<(v8i8 immAllOnesV), (MOVID (i32 255))>;
3730 // EDIT byte mask: 2d
3732 // The movi_edit node has the immediate value already encoded, so we use
3733 // a plain imm0_255 in the pattern
3734 let isReMaterializable = 1, isAsCheapAsAMove = 1 in
3735 def MOVIv2d_ns : SIMDModifiedImmVectorNoShift<1, 1, 0b1110, V128,
3738 [(set (v2i64 V128:$Rd), (ARM64movi_edit imm0_255:$imm8))]>;
3741 // Use movi.2d to materialize 0.0 if the HW does zero-cycle zeroing.
3742 // Complexity is added to break a tie with a plain MOVI.
3743 let AddedComplexity = 1 in {
3744 def : Pat<(f32 fpimm0),
3745 (f32 (EXTRACT_SUBREG (v2i64 (MOVIv2d_ns (i32 0))), ssub))>,
3747 def : Pat<(f64 fpimm0),
3748 (f64 (EXTRACT_SUBREG (v2i64 (MOVIv2d_ns (i32 0))), dsub))>,
3752 def : Pat<(v2i64 immAllZerosV), (MOVIv2d_ns (i32 0))>;
3753 def : Pat<(v4i32 immAllZerosV), (MOVIv2d_ns (i32 0))>;
3754 def : Pat<(v8i16 immAllZerosV), (MOVIv2d_ns (i32 0))>;
3755 def : Pat<(v16i8 immAllZerosV), (MOVIv2d_ns (i32 0))>;
3757 def : Pat<(v2i64 immAllOnesV), (MOVIv2d_ns (i32 255))>;
3758 def : Pat<(v4i32 immAllOnesV), (MOVIv2d_ns (i32 255))>;
3759 def : Pat<(v8i16 immAllOnesV), (MOVIv2d_ns (i32 255))>;
3760 def : Pat<(v16i8 immAllOnesV), (MOVIv2d_ns (i32 255))>;
3762 def : Pat<(v2f64 (ARM64dup (f64 fpimm0))), (MOVIv2d_ns (i32 0))>;
3763 def : Pat<(v4f32 (ARM64dup (f32 fpimm0))), (MOVIv2d_ns (i32 0))>;
3765 // EDIT per word & halfword: 2s, 4h, 4s, & 8h
3766 defm MOVI : SIMDModifiedImmVectorShift<0, 0b10, 0b00, "movi">;
3768 def : InstAlias<"movi $Vd.4h, $imm", (MOVIv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
3769 def : InstAlias<"movi $Vd.8h, $imm", (MOVIv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
3770 def : InstAlias<"movi $Vd.2s, $imm", (MOVIv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
3771 def : InstAlias<"movi $Vd.4s, $imm", (MOVIv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
3773 def : InstAlias<"movi.4h $Vd, $imm", (MOVIv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
3774 def : InstAlias<"movi.8h $Vd, $imm", (MOVIv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
3775 def : InstAlias<"movi.2s $Vd, $imm", (MOVIv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
3776 def : InstAlias<"movi.4s $Vd, $imm", (MOVIv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
3778 def : Pat<(v2i32 (ARM64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
3779 (MOVIv2i32 imm0_255:$imm8, imm:$shift)>;
3780 def : Pat<(v4i32 (ARM64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
3781 (MOVIv4i32 imm0_255:$imm8, imm:$shift)>;
3782 def : Pat<(v4i16 (ARM64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
3783 (MOVIv4i16 imm0_255:$imm8, imm:$shift)>;
3784 def : Pat<(v8i16 (ARM64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
3785 (MOVIv8i16 imm0_255:$imm8, imm:$shift)>;
3787 // EDIT per word: 2s & 4s with MSL shifter
3788 def MOVIv2s_msl : SIMDModifiedImmMoveMSL<0, 0, {1,1,0,?}, V64, "movi", ".2s",
3789 [(set (v2i32 V64:$Rd),
3790 (ARM64movi_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
3791 def MOVIv4s_msl : SIMDModifiedImmMoveMSL<1, 0, {1,1,0,?}, V128, "movi", ".4s",
3792 [(set (v4i32 V128:$Rd),
3793 (ARM64movi_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
3795 // Per byte: 8b & 16b
3796 def MOVIv8b_ns : SIMDModifiedImmVectorNoShift<0, 0, 0b1110, V64, imm0_255,
3798 [(set (v8i8 V64:$Rd), (ARM64movi imm0_255:$imm8))]>;
3799 def MOVIv16b_ns : SIMDModifiedImmVectorNoShift<1, 0, 0b1110, V128, imm0_255,
3801 [(set (v16i8 V128:$Rd), (ARM64movi imm0_255:$imm8))]>;
3805 // EDIT per word & halfword: 2s, 4h, 4s, & 8h
3806 defm MVNI : SIMDModifiedImmVectorShift<1, 0b10, 0b00, "mvni">;
3808 def : InstAlias<"mvni $Vd.4h, $imm", (MVNIv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
3809 def : InstAlias<"mvni $Vd.8h, $imm", (MVNIv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
3810 def : InstAlias<"mvni $Vd.2s, $imm", (MVNIv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
3811 def : InstAlias<"mvni $Vd.4s, $imm", (MVNIv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
3813 def : InstAlias<"mvni.4h $Vd, $imm", (MVNIv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
3814 def : InstAlias<"mvni.8h $Vd, $imm", (MVNIv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
3815 def : InstAlias<"mvni.2s $Vd, $imm", (MVNIv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
3816 def : InstAlias<"mvni.4s $Vd, $imm", (MVNIv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
3818 def : Pat<(v2i32 (ARM64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
3819 (MVNIv2i32 imm0_255:$imm8, imm:$shift)>;
3820 def : Pat<(v4i32 (ARM64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
3821 (MVNIv4i32 imm0_255:$imm8, imm:$shift)>;
3822 def : Pat<(v4i16 (ARM64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
3823 (MVNIv4i16 imm0_255:$imm8, imm:$shift)>;
3824 def : Pat<(v8i16 (ARM64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
3825 (MVNIv8i16 imm0_255:$imm8, imm:$shift)>;
3827 // EDIT per word: 2s & 4s with MSL shifter
3828 def MVNIv2s_msl : SIMDModifiedImmMoveMSL<0, 1, {1,1,0,?}, V64, "mvni", ".2s",
3829 [(set (v2i32 V64:$Rd),
3830 (ARM64mvni_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
3831 def MVNIv4s_msl : SIMDModifiedImmMoveMSL<1, 1, {1,1,0,?}, V128, "mvni", ".4s",
3832 [(set (v4i32 V128:$Rd),
3833 (ARM64mvni_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
3835 //----------------------------------------------------------------------------
3836 // AdvSIMD indexed element
3837 //----------------------------------------------------------------------------
3839 let neverHasSideEffects = 1 in {
3840 defm FMLA : SIMDFPIndexedSDTied<0, 0b0001, "fmla">;
3841 defm FMLS : SIMDFPIndexedSDTied<0, 0b0101, "fmls">;
3844 // NOTE: Operands are reordered in the FMLA/FMLS PatFrags because the
3845 // instruction expects the addend first, while the intrinsic expects it last.
3847 // On the other hand, there are quite a few valid combinatorial options due to
3848 // the commutativity of multiplication and the fact that (-x) * y = x * (-y).
3849 defm : SIMDFPIndexedSDTiedPatterns<"FMLA",
3850 TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)>>;
3851 defm : SIMDFPIndexedSDTiedPatterns<"FMLA",
3852 TriOpFrag<(fma node:$MHS, node:$RHS, node:$LHS)>>;
3854 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
3855 TriOpFrag<(fma node:$MHS, (fneg node:$RHS), node:$LHS)> >;
3856 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
3857 TriOpFrag<(fma node:$RHS, (fneg node:$MHS), node:$LHS)> >;
3858 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
3859 TriOpFrag<(fma (fneg node:$RHS), node:$MHS, node:$LHS)> >;
3860 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
3861 TriOpFrag<(fma (fneg node:$MHS), node:$RHS, node:$LHS)> >;
3863 multiclass FMLSIndexedAfterNegPatterns<SDPatternOperator OpNode> {
3864 // 3 variants for the .2s version: DUPLANE from 128-bit, DUPLANE from 64-bit
3866 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
3867 (ARM64duplane32 (v4f32 (fneg V128:$Rm)),
3868 VectorIndexS:$idx))),
3869 (FMLSv2i32_indexed V64:$Rd, V64:$Rn, V128:$Rm, VectorIndexS:$idx)>;
3870 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
3871 (v2f32 (ARM64duplane32
3872 (v4f32 (insert_subvector undef,
3873 (v2f32 (fneg V64:$Rm)),
3875 VectorIndexS:$idx)))),
3876 (FMLSv2i32_indexed V64:$Rd, V64:$Rn,
3877 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub),
3878 VectorIndexS:$idx)>;
3879 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
3880 (ARM64dup (f32 (fneg FPR32Op:$Rm))))),
3881 (FMLSv2i32_indexed V64:$Rd, V64:$Rn,
3882 (SUBREG_TO_REG (i32 0), FPR32Op:$Rm, ssub), (i64 0))>;
3884 // 3 variants for the .4s version: DUPLANE from 128-bit, DUPLANE from 64-bit
3886 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
3887 (ARM64duplane32 (v4f32 (fneg V128:$Rm)),
3888 VectorIndexS:$idx))),
3889 (FMLSv4i32_indexed V128:$Rd, V128:$Rn, V128:$Rm,
3890 VectorIndexS:$idx)>;
3891 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
3892 (v4f32 (ARM64duplane32
3893 (v4f32 (insert_subvector undef,
3894 (v2f32 (fneg V64:$Rm)),
3896 VectorIndexS:$idx)))),
3897 (FMLSv4i32_indexed V128:$Rd, V128:$Rn,
3898 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub),
3899 VectorIndexS:$idx)>;
3900 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
3901 (ARM64dup (f32 (fneg FPR32Op:$Rm))))),
3902 (FMLSv4i32_indexed V128:$Rd, V128:$Rn,
3903 (SUBREG_TO_REG (i32 0), FPR32Op:$Rm, ssub), (i64 0))>;
3905 // 2 variants for the .2d version: DUPLANE from 128-bit, and DUP scalar
3906 // (DUPLANE from 64-bit would be trivial).
3907 def : Pat<(v2f64 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn),
3908 (ARM64duplane64 (v2f64 (fneg V128:$Rm)),
3909 VectorIndexD:$idx))),
3911 V128:$Rd, V128:$Rn, V128:$Rm, VectorIndexS:$idx)>;
3912 def : Pat<(v2f64 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn),
3913 (ARM64dup (f64 (fneg FPR64Op:$Rm))))),
3914 (FMLSv2i64_indexed V128:$Rd, V128:$Rn,
3915 (SUBREG_TO_REG (i32 0), FPR64Op:$Rm, dsub), (i64 0))>;
3917 // 2 variants for 32-bit scalar version: extract from .2s or from .4s
3918 def : Pat<(f32 (OpNode (f32 FPR32:$Rd), (f32 FPR32:$Rn),
3919 (vector_extract (v4f32 (fneg V128:$Rm)),
3920 VectorIndexS:$idx))),
3921 (FMLSv1i32_indexed FPR32:$Rd, FPR32:$Rn,
3922 V128:$Rm, VectorIndexS:$idx)>;
3923 def : Pat<(f32 (OpNode (f32 FPR32:$Rd), (f32 FPR32:$Rn),
3924 (vector_extract (v2f32 (fneg V64:$Rm)),
3925 VectorIndexS:$idx))),
3926 (FMLSv1i32_indexed FPR32:$Rd, FPR32:$Rn,
3927 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub), VectorIndexS:$idx)>;
3929 // 1 variant for 64-bit scalar version: extract from .1d or from .2d
3930 def : Pat<(f64 (OpNode (f64 FPR64:$Rd), (f64 FPR64:$Rn),
3931 (vector_extract (v2f64 (fneg V128:$Rm)),
3932 VectorIndexS:$idx))),
3933 (FMLSv1i64_indexed FPR64:$Rd, FPR64:$Rn,
3934 V128:$Rm, VectorIndexS:$idx)>;
3937 defm : FMLSIndexedAfterNegPatterns<
3938 TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)> >;
3939 defm : FMLSIndexedAfterNegPatterns<
3940 TriOpFrag<(fma node:$MHS, node:$RHS, node:$LHS)> >;
3942 defm FMULX : SIMDFPIndexedSD<1, 0b1001, "fmulx", int_arm64_neon_fmulx>;
3943 defm FMUL : SIMDFPIndexedSD<0, 0b1001, "fmul", fmul>;
3945 def : Pat<(v2f32 (fmul V64:$Rn, (ARM64dup (f32 FPR32:$Rm)))),
3946 (FMULv2i32_indexed V64:$Rn,
3947 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rm, ssub),
3949 def : Pat<(v4f32 (fmul V128:$Rn, (ARM64dup (f32 FPR32:$Rm)))),
3950 (FMULv4i32_indexed V128:$Rn,
3951 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rm, ssub),
3953 def : Pat<(v2f64 (fmul V128:$Rn, (ARM64dup (f64 FPR64:$Rm)))),
3954 (FMULv2i64_indexed V128:$Rn,
3955 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$Rm, dsub),
3958 defm SQDMULH : SIMDIndexedHS<0, 0b1100, "sqdmulh", int_arm64_neon_sqdmulh>;
3959 defm SQRDMULH : SIMDIndexedHS<0, 0b1101, "sqrdmulh", int_arm64_neon_sqrdmulh>;
3960 defm MLA : SIMDVectorIndexedHSTied<1, 0b0000, "mla",
3961 TriOpFrag<(add node:$LHS, (mul node:$MHS, node:$RHS))>>;
3962 defm MLS : SIMDVectorIndexedHSTied<1, 0b0100, "mls",
3963 TriOpFrag<(sub node:$LHS, (mul node:$MHS, node:$RHS))>>;
3964 defm MUL : SIMDVectorIndexedHS<0, 0b1000, "mul", mul>;
3965 defm SMLAL : SIMDVectorIndexedLongSDTied<0, 0b0010, "smlal",
3966 TriOpFrag<(add node:$LHS, (int_arm64_neon_smull node:$MHS, node:$RHS))>>;
3967 defm SMLSL : SIMDVectorIndexedLongSDTied<0, 0b0110, "smlsl",
3968 TriOpFrag<(sub node:$LHS, (int_arm64_neon_smull node:$MHS, node:$RHS))>>;
3969 defm SMULL : SIMDVectorIndexedLongSD<0, 0b1010, "smull",
3970 int_arm64_neon_smull>;
3971 defm SQDMLAL : SIMDIndexedLongSQDMLXSDTied<0, 0b0011, "sqdmlal",
3972 int_arm64_neon_sqadd>;
3973 defm SQDMLSL : SIMDIndexedLongSQDMLXSDTied<0, 0b0111, "sqdmlsl",
3974 int_arm64_neon_sqsub>;
3975 defm SQDMULL : SIMDIndexedLongSD<0, 0b1011, "sqdmull", int_arm64_neon_sqdmull>;
3976 defm UMLAL : SIMDVectorIndexedLongSDTied<1, 0b0010, "umlal",
3977 TriOpFrag<(add node:$LHS, (int_arm64_neon_umull node:$MHS, node:$RHS))>>;
3978 defm UMLSL : SIMDVectorIndexedLongSDTied<1, 0b0110, "umlsl",
3979 TriOpFrag<(sub node:$LHS, (int_arm64_neon_umull node:$MHS, node:$RHS))>>;
3980 defm UMULL : SIMDVectorIndexedLongSD<1, 0b1010, "umull",
3981 int_arm64_neon_umull>;
3983 // A scalar sqdmull with the second operand being a vector lane can be
3984 // handled directly with the indexed instruction encoding.
3985 def : Pat<(int_arm64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
3986 (vector_extract (v4i32 V128:$Vm),
3987 VectorIndexS:$idx)),
3988 (SQDMULLv1i64_indexed FPR32:$Rn, V128:$Vm, VectorIndexS:$idx)>;
3990 //----------------------------------------------------------------------------
3991 // AdvSIMD scalar shift instructions
3992 //----------------------------------------------------------------------------
3993 defm FCVTZS : SIMDScalarRShiftSD<0, 0b11111, "fcvtzs">;
3994 defm FCVTZU : SIMDScalarRShiftSD<1, 0b11111, "fcvtzu">;
3995 defm SCVTF : SIMDScalarRShiftSD<0, 0b11100, "scvtf">;
3996 defm UCVTF : SIMDScalarRShiftSD<1, 0b11100, "ucvtf">;
3997 // Codegen patterns for the above. We don't put these directly on the
3998 // instructions because TableGen's type inference can't handle the truth.
3999 // Having the same base pattern for fp <--> int totally freaks it out.
4000 def : Pat<(int_arm64_neon_vcvtfp2fxs FPR32:$Rn, vecshiftR32:$imm),
4001 (FCVTZSs FPR32:$Rn, vecshiftR32:$imm)>;
4002 def : Pat<(int_arm64_neon_vcvtfp2fxu FPR32:$Rn, vecshiftR32:$imm),
4003 (FCVTZUs FPR32:$Rn, vecshiftR32:$imm)>;
4004 def : Pat<(i64 (int_arm64_neon_vcvtfp2fxs (f64 FPR64:$Rn), vecshiftR64:$imm)),
4005 (FCVTZSd FPR64:$Rn, vecshiftR64:$imm)>;
4006 def : Pat<(i64 (int_arm64_neon_vcvtfp2fxu (f64 FPR64:$Rn), vecshiftR64:$imm)),
4007 (FCVTZUd FPR64:$Rn, vecshiftR64:$imm)>;
4008 def : Pat<(v1i64 (int_arm64_neon_vcvtfp2fxs (v1f64 FPR64:$Rn),
4010 (FCVTZSd FPR64:$Rn, vecshiftR64:$imm)>;
4011 def : Pat<(v1i64 (int_arm64_neon_vcvtfp2fxu (v1f64 FPR64:$Rn),
4013 (FCVTZUd FPR64:$Rn, vecshiftR64:$imm)>;
4014 def : Pat<(int_arm64_neon_vcvtfxs2fp FPR32:$Rn, vecshiftR32:$imm),
4015 (SCVTFs FPR32:$Rn, vecshiftR32:$imm)>;
4016 def : Pat<(int_arm64_neon_vcvtfxu2fp FPR32:$Rn, vecshiftR32:$imm),
4017 (UCVTFs FPR32:$Rn, vecshiftR32:$imm)>;
4018 def : Pat<(f64 (int_arm64_neon_vcvtfxs2fp (i64 FPR64:$Rn), vecshiftR64:$imm)),
4019 (SCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
4020 def : Pat<(f64 (int_arm64_neon_vcvtfxu2fp (i64 FPR64:$Rn), vecshiftR64:$imm)),
4021 (UCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
4022 def : Pat<(v1f64 (int_arm64_neon_vcvtfxs2fp (v1i64 FPR64:$Rn),
4024 (SCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
4025 def : Pat<(v1f64 (int_arm64_neon_vcvtfxu2fp (v1i64 FPR64:$Rn),
4027 (UCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
4029 defm SHL : SIMDScalarLShiftD< 0, 0b01010, "shl", ARM64vshl>;
4030 defm SLI : SIMDScalarLShiftDTied<1, 0b01010, "sli">;
4031 defm SQRSHRN : SIMDScalarRShiftBHS< 0, 0b10011, "sqrshrn",
4032 int_arm64_neon_sqrshrn>;
4033 defm SQRSHRUN : SIMDScalarRShiftBHS< 1, 0b10001, "sqrshrun",
4034 int_arm64_neon_sqrshrun>;
4035 defm SQSHLU : SIMDScalarLShiftBHSD<1, 0b01100, "sqshlu", ARM64sqshlui>;
4036 defm SQSHL : SIMDScalarLShiftBHSD<0, 0b01110, "sqshl", ARM64sqshli>;
4037 defm SQSHRN : SIMDScalarRShiftBHS< 0, 0b10010, "sqshrn",
4038 int_arm64_neon_sqshrn>;
4039 defm SQSHRUN : SIMDScalarRShiftBHS< 1, 0b10000, "sqshrun",
4040 int_arm64_neon_sqshrun>;
4041 defm SRI : SIMDScalarRShiftDTied< 1, 0b01000, "sri">;
4042 defm SRSHR : SIMDScalarRShiftD< 0, 0b00100, "srshr", ARM64srshri>;
4043 defm SRSRA : SIMDScalarRShiftDTied< 0, 0b00110, "srsra",
4044 TriOpFrag<(add node:$LHS,
4045 (ARM64srshri node:$MHS, node:$RHS))>>;
4046 defm SSHR : SIMDScalarRShiftD< 0, 0b00000, "sshr", ARM64vashr>;
4047 defm SSRA : SIMDScalarRShiftDTied< 0, 0b00010, "ssra",
4048 TriOpFrag<(add node:$LHS,
4049 (ARM64vashr node:$MHS, node:$RHS))>>;
4050 defm UQRSHRN : SIMDScalarRShiftBHS< 1, 0b10011, "uqrshrn",
4051 int_arm64_neon_uqrshrn>;
4052 defm UQSHL : SIMDScalarLShiftBHSD<1, 0b01110, "uqshl", ARM64uqshli>;
4053 defm UQSHRN : SIMDScalarRShiftBHS< 1, 0b10010, "uqshrn",
4054 int_arm64_neon_uqshrn>;
4055 defm URSHR : SIMDScalarRShiftD< 1, 0b00100, "urshr", ARM64urshri>;
4056 defm URSRA : SIMDScalarRShiftDTied< 1, 0b00110, "ursra",
4057 TriOpFrag<(add node:$LHS,
4058 (ARM64urshri node:$MHS, node:$RHS))>>;
4059 defm USHR : SIMDScalarRShiftD< 1, 0b00000, "ushr", ARM64vlshr>;
4060 defm USRA : SIMDScalarRShiftDTied< 1, 0b00010, "usra",
4061 TriOpFrag<(add node:$LHS,
4062 (ARM64vlshr node:$MHS, node:$RHS))>>;
4064 //----------------------------------------------------------------------------
4065 // AdvSIMD vector shift instructions
4066 //----------------------------------------------------------------------------
4067 defm FCVTZS:SIMDVectorRShiftSD<0, 0b11111, "fcvtzs", int_arm64_neon_vcvtfp2fxs>;
4068 defm FCVTZU:SIMDVectorRShiftSD<1, 0b11111, "fcvtzu", int_arm64_neon_vcvtfp2fxu>;
4069 defm SCVTF: SIMDVectorRShiftSDToFP<0, 0b11100, "scvtf",
4070 int_arm64_neon_vcvtfxs2fp>;
4071 defm RSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10001, "rshrn",
4072 int_arm64_neon_rshrn>;
4073 defm SHL : SIMDVectorLShiftBHSD<0, 0b01010, "shl", ARM64vshl>;
4074 defm SHRN : SIMDVectorRShiftNarrowBHS<0, 0b10000, "shrn",
4075 BinOpFrag<(trunc (ARM64vashr node:$LHS, node:$RHS))>>;
4076 defm SLI : SIMDVectorLShiftBHSDTied<1, 0b01010, "sli", int_arm64_neon_vsli>;
4077 def : Pat<(v1i64 (int_arm64_neon_vsli (v1i64 FPR64:$Rd), (v1i64 FPR64:$Rn),
4078 (i32 vecshiftL64:$imm))),
4079 (SLId FPR64:$Rd, FPR64:$Rn, vecshiftL64:$imm)>;
4080 defm SQRSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10011, "sqrshrn",
4081 int_arm64_neon_sqrshrn>;
4082 defm SQRSHRUN: SIMDVectorRShiftNarrowBHS<1, 0b10001, "sqrshrun",
4083 int_arm64_neon_sqrshrun>;
4084 defm SQSHLU : SIMDVectorLShiftBHSD<1, 0b01100, "sqshlu", ARM64sqshlui>;
4085 defm SQSHL : SIMDVectorLShiftBHSD<0, 0b01110, "sqshl", ARM64sqshli>;
4086 defm SQSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10010, "sqshrn",
4087 int_arm64_neon_sqshrn>;
4088 defm SQSHRUN : SIMDVectorRShiftNarrowBHS<1, 0b10000, "sqshrun",
4089 int_arm64_neon_sqshrun>;
4090 defm SRI : SIMDVectorRShiftBHSDTied<1, 0b01000, "sri", int_arm64_neon_vsri>;
4091 def : Pat<(v1i64 (int_arm64_neon_vsri (v1i64 FPR64:$Rd), (v1i64 FPR64:$Rn),
4092 (i32 vecshiftR64:$imm))),
4093 (SRId FPR64:$Rd, FPR64:$Rn, vecshiftR64:$imm)>;
4094 defm SRSHR : SIMDVectorRShiftBHSD<0, 0b00100, "srshr", ARM64srshri>;
4095 defm SRSRA : SIMDVectorRShiftBHSDTied<0, 0b00110, "srsra",
4096 TriOpFrag<(add node:$LHS,
4097 (ARM64srshri node:$MHS, node:$RHS))> >;
4098 defm SSHLL : SIMDVectorLShiftLongBHSD<0, 0b10100, "sshll",
4099 BinOpFrag<(ARM64vshl (sext node:$LHS), node:$RHS)>>;
4101 defm SSHR : SIMDVectorRShiftBHSD<0, 0b00000, "sshr", ARM64vashr>;
4102 defm SSRA : SIMDVectorRShiftBHSDTied<0, 0b00010, "ssra",
4103 TriOpFrag<(add node:$LHS, (ARM64vashr node:$MHS, node:$RHS))>>;
4104 defm UCVTF : SIMDVectorRShiftSDToFP<1, 0b11100, "ucvtf",
4105 int_arm64_neon_vcvtfxu2fp>;
4106 defm UQRSHRN : SIMDVectorRShiftNarrowBHS<1, 0b10011, "uqrshrn",
4107 int_arm64_neon_uqrshrn>;
4108 defm UQSHL : SIMDVectorLShiftBHSD<1, 0b01110, "uqshl", ARM64uqshli>;
4109 defm UQSHRN : SIMDVectorRShiftNarrowBHS<1, 0b10010, "uqshrn",
4110 int_arm64_neon_uqshrn>;
4111 defm URSHR : SIMDVectorRShiftBHSD<1, 0b00100, "urshr", ARM64urshri>;
4112 defm URSRA : SIMDVectorRShiftBHSDTied<1, 0b00110, "ursra",
4113 TriOpFrag<(add node:$LHS,
4114 (ARM64urshri node:$MHS, node:$RHS))> >;
4115 defm USHLL : SIMDVectorLShiftLongBHSD<1, 0b10100, "ushll",
4116 BinOpFrag<(ARM64vshl (zext node:$LHS), node:$RHS)>>;
4117 defm USHR : SIMDVectorRShiftBHSD<1, 0b00000, "ushr", ARM64vlshr>;
4118 defm USRA : SIMDVectorRShiftBHSDTied<1, 0b00010, "usra",
4119 TriOpFrag<(add node:$LHS, (ARM64vlshr node:$MHS, node:$RHS))> >;
4121 // SHRN patterns for when a logical right shift was used instead of arithmetic
4122 // (the immediate guarantees no sign bits actually end up in the result so it
4124 def : Pat<(v8i8 (trunc (ARM64vlshr (v8i16 V128:$Rn), vecshiftR16Narrow:$imm))),
4125 (SHRNv8i8_shift V128:$Rn, vecshiftR16Narrow:$imm)>;
4126 def : Pat<(v4i16 (trunc (ARM64vlshr (v4i32 V128:$Rn), vecshiftR32Narrow:$imm))),
4127 (SHRNv4i16_shift V128:$Rn, vecshiftR32Narrow:$imm)>;
4128 def : Pat<(v2i32 (trunc (ARM64vlshr (v2i64 V128:$Rn), vecshiftR64Narrow:$imm))),
4129 (SHRNv2i32_shift V128:$Rn, vecshiftR64Narrow:$imm)>;
4131 def : Pat<(v16i8 (concat_vectors (v8i8 V64:$Rd),
4132 (trunc (ARM64vlshr (v8i16 V128:$Rn),
4133 vecshiftR16Narrow:$imm)))),
4134 (SHRNv16i8_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
4135 V128:$Rn, vecshiftR16Narrow:$imm)>;
4136 def : Pat<(v8i16 (concat_vectors (v4i16 V64:$Rd),
4137 (trunc (ARM64vlshr (v4i32 V128:$Rn),
4138 vecshiftR32Narrow:$imm)))),
4139 (SHRNv8i16_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
4140 V128:$Rn, vecshiftR32Narrow:$imm)>;
4141 def : Pat<(v4i32 (concat_vectors (v2i32 V64:$Rd),
4142 (trunc (ARM64vlshr (v2i64 V128:$Rn),
4143 vecshiftR64Narrow:$imm)))),
4144 (SHRNv4i32_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
4145 V128:$Rn, vecshiftR32Narrow:$imm)>;
4147 // Vector sign and zero extensions are implemented with SSHLL and USSHLL.
4148 // Anyexts are implemented as zexts.
4149 def : Pat<(v8i16 (sext (v8i8 V64:$Rn))), (SSHLLv8i8_shift V64:$Rn, (i32 0))>;
4150 def : Pat<(v8i16 (zext (v8i8 V64:$Rn))), (USHLLv8i8_shift V64:$Rn, (i32 0))>;
4151 def : Pat<(v8i16 (anyext (v8i8 V64:$Rn))), (USHLLv8i8_shift V64:$Rn, (i32 0))>;
4152 def : Pat<(v4i32 (sext (v4i16 V64:$Rn))), (SSHLLv4i16_shift V64:$Rn, (i32 0))>;
4153 def : Pat<(v4i32 (zext (v4i16 V64:$Rn))), (USHLLv4i16_shift V64:$Rn, (i32 0))>;
4154 def : Pat<(v4i32 (anyext (v4i16 V64:$Rn))), (USHLLv4i16_shift V64:$Rn, (i32 0))>;
4155 def : Pat<(v2i64 (sext (v2i32 V64:$Rn))), (SSHLLv2i32_shift V64:$Rn, (i32 0))>;
4156 def : Pat<(v2i64 (zext (v2i32 V64:$Rn))), (USHLLv2i32_shift V64:$Rn, (i32 0))>;
4157 def : Pat<(v2i64 (anyext (v2i32 V64:$Rn))), (USHLLv2i32_shift V64:$Rn, (i32 0))>;
4158 // Also match an extend from the upper half of a 128 bit source register.
4159 def : Pat<(v8i16 (anyext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
4160 (USHLLv16i8_shift V128:$Rn, (i32 0))>;
4161 def : Pat<(v8i16 (zext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
4162 (USHLLv16i8_shift V128:$Rn, (i32 0))>;
4163 def : Pat<(v8i16 (sext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
4164 (SSHLLv16i8_shift V128:$Rn, (i32 0))>;
4165 def : Pat<(v4i32 (anyext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
4166 (USHLLv8i16_shift V128:$Rn, (i32 0))>;
4167 def : Pat<(v4i32 (zext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
4168 (USHLLv8i16_shift V128:$Rn, (i32 0))>;
4169 def : Pat<(v4i32 (sext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
4170 (SSHLLv8i16_shift V128:$Rn, (i32 0))>;
4171 def : Pat<(v2i64 (anyext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
4172 (USHLLv4i32_shift V128:$Rn, (i32 0))>;
4173 def : Pat<(v2i64 (zext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
4174 (USHLLv4i32_shift V128:$Rn, (i32 0))>;
4175 def : Pat<(v2i64 (sext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
4176 (SSHLLv4i32_shift V128:$Rn, (i32 0))>;
4178 // Vector shift sxtl aliases
4179 def : InstAlias<"sxtl.8h $dst, $src1",
4180 (SSHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
4181 def : InstAlias<"sxtl $dst.8h, $src1.8b",
4182 (SSHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
4183 def : InstAlias<"sxtl.4s $dst, $src1",
4184 (SSHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
4185 def : InstAlias<"sxtl $dst.4s, $src1.4h",
4186 (SSHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
4187 def : InstAlias<"sxtl.2d $dst, $src1",
4188 (SSHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
4189 def : InstAlias<"sxtl $dst.2d, $src1.2s",
4190 (SSHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
4192 // Vector shift sxtl2 aliases
4193 def : InstAlias<"sxtl2.8h $dst, $src1",
4194 (SSHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
4195 def : InstAlias<"sxtl2 $dst.8h, $src1.16b",
4196 (SSHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
4197 def : InstAlias<"sxtl2.4s $dst, $src1",
4198 (SSHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
4199 def : InstAlias<"sxtl2 $dst.4s, $src1.8h",
4200 (SSHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
4201 def : InstAlias<"sxtl2.2d $dst, $src1",
4202 (SSHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
4203 def : InstAlias<"sxtl2 $dst.2d, $src1.4s",
4204 (SSHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
4206 // Vector shift uxtl aliases
4207 def : InstAlias<"uxtl.8h $dst, $src1",
4208 (USHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
4209 def : InstAlias<"uxtl $dst.8h, $src1.8b",
4210 (USHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
4211 def : InstAlias<"uxtl.4s $dst, $src1",
4212 (USHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
4213 def : InstAlias<"uxtl $dst.4s, $src1.4h",
4214 (USHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
4215 def : InstAlias<"uxtl.2d $dst, $src1",
4216 (USHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
4217 def : InstAlias<"uxtl $dst.2d, $src1.2s",
4218 (USHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
4220 // Vector shift uxtl2 aliases
4221 def : InstAlias<"uxtl2.8h $dst, $src1",
4222 (USHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
4223 def : InstAlias<"uxtl2 $dst.8h, $src1.16b",
4224 (USHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
4225 def : InstAlias<"uxtl2.4s $dst, $src1",
4226 (USHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
4227 def : InstAlias<"uxtl2 $dst.4s, $src1.8h",
4228 (USHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
4229 def : InstAlias<"uxtl2.2d $dst, $src1",
4230 (USHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
4231 def : InstAlias<"uxtl2 $dst.2d, $src1.4s",
4232 (USHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
4234 // If an integer is about to be converted to a floating point value,
4235 // just load it on the floating point unit.
4236 // These patterns are more complex because floating point loads do not
4237 // support sign extension.
4238 // The sign extension has to be explicitly added and is only supported for
4239 // one step: byte-to-half, half-to-word, word-to-doubleword.
4240 // SCVTF GPR -> FPR is 9 cycles.
4241 // SCVTF FPR -> FPR is 4 cyclces.
4242 // (sign extension with lengthen) SXTL FPR -> FPR is 2 cycles.
4243 // Therefore, we can do 2 sign extensions and one SCVTF FPR -> FPR
4244 // and still being faster.
4245 // However, this is not good for code size.
4246 // 8-bits -> float. 2 sizes step-up.
4247 def : Pat <(f32 (sint_to_fp (i32 (sextloadi8 ro_indexed8:$addr)))),
4248 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
4253 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4254 (LDRBro ro_indexed8:$addr),
4259 ssub)))>, Requires<[NotForCodeSize]>;
4260 def : Pat <(f32 (sint_to_fp (i32 (sextloadi8 am_indexed8:$addr)))),
4261 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
4266 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4267 (LDRBui am_indexed8:$addr),
4272 ssub)))>, Requires<[NotForCodeSize]>;
4273 def : Pat <(f32 (sint_to_fp (i32 (sextloadi8 am_unscaled8:$addr)))),
4274 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
4279 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4280 (LDURBi am_unscaled8:$addr),
4285 ssub)))>, Requires<[NotForCodeSize]>;
4286 // 16-bits -> float. 1 size step-up.
4287 def : Pat <(f32 (sint_to_fp (i32 (sextloadi16 ro_indexed16:$addr)))),
4288 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
4290 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4291 (LDRHro ro_indexed16:$addr),
4294 ssub)))>, Requires<[NotForCodeSize]>;
4295 def : Pat <(f32 (sint_to_fp (i32 (sextloadi16 am_indexed16:$addr)))),
4296 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
4298 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4299 (LDRHui am_indexed16:$addr),
4302 ssub)))>, Requires<[NotForCodeSize]>;
4303 def : Pat <(f32 (sint_to_fp (i32 (sextloadi16 am_unscaled16:$addr)))),
4304 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
4306 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4307 (LDURHi am_unscaled16:$addr),
4310 ssub)))>, Requires<[NotForCodeSize]>;
4311 // 32-bits to 32-bits are handled in target specific dag combine:
4312 // performIntToFpCombine.
4313 // 64-bits integer to 32-bits floating point, not possible with
4314 // SCVTF on floating point registers (both source and destination
4315 // must have the same size).
4317 // Here are the patterns for 8, 16, 32, and 64-bits to double.
4318 // 8-bits -> double. 3 size step-up: give up.
4319 // 16-bits -> double. 2 size step.
4320 def : Pat <(f64 (sint_to_fp (i32 (sextloadi16 ro_indexed16:$addr)))),
4321 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4326 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4327 (LDRHro ro_indexed16:$addr),
4332 dsub)))>, Requires<[NotForCodeSize]>;
4333 def : Pat <(f64 (sint_to_fp (i32 (sextloadi16 am_indexed16:$addr)))),
4334 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4339 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4340 (LDRHui am_indexed16:$addr),
4345 dsub)))>, Requires<[NotForCodeSize]>;
4346 def : Pat <(f64 (sint_to_fp (i32 (sextloadi16 am_unscaled16:$addr)))),
4347 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4352 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4353 (LDURHi am_unscaled16:$addr),
4358 dsub)))>, Requires<[NotForCodeSize]>;
4359 // 32-bits -> double. 1 size step-up.
4360 def : Pat <(f64 (sint_to_fp (i32 (load ro_indexed32:$addr)))),
4361 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4363 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4364 (LDRSro ro_indexed32:$addr),
4367 dsub)))>, Requires<[NotForCodeSize]>;
4368 def : Pat <(f64 (sint_to_fp (i32 (load am_indexed32:$addr)))),
4369 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4371 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4372 (LDRSui am_indexed32:$addr),
4375 dsub)))>, Requires<[NotForCodeSize]>;
4376 def : Pat <(f64 (sint_to_fp (i32 (load am_unscaled32:$addr)))),
4377 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4379 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4380 (LDURSi am_unscaled32:$addr),
4383 dsub)))>, Requires<[NotForCodeSize]>;
4384 // 64-bits -> double are handled in target specific dag combine:
4385 // performIntToFpCombine.
4388 //----------------------------------------------------------------------------
4389 // AdvSIMD Load-Store Structure
4390 //----------------------------------------------------------------------------
4391 defm LD1 : SIMDLd1Multiple<"ld1">;
4392 defm LD2 : SIMDLd2Multiple<"ld2">;
4393 defm LD3 : SIMDLd3Multiple<"ld3">;
4394 defm LD4 : SIMDLd4Multiple<"ld4">;
4396 defm ST1 : SIMDSt1Multiple<"st1">;
4397 defm ST2 : SIMDSt2Multiple<"st2">;
4398 defm ST3 : SIMDSt3Multiple<"st3">;
4399 defm ST4 : SIMDSt4Multiple<"st4">;
4401 class Ld1Pat<ValueType ty, Instruction INST>
4402 : Pat<(ty (load am_simdnoindex:$vaddr)), (INST am_simdnoindex:$vaddr)>;
4404 def : Ld1Pat<v16i8, LD1Onev16b>;
4405 def : Ld1Pat<v8i16, LD1Onev8h>;
4406 def : Ld1Pat<v4i32, LD1Onev4s>;
4407 def : Ld1Pat<v2i64, LD1Onev2d>;
4408 def : Ld1Pat<v8i8, LD1Onev8b>;
4409 def : Ld1Pat<v4i16, LD1Onev4h>;
4410 def : Ld1Pat<v2i32, LD1Onev2s>;
4411 def : Ld1Pat<v1i64, LD1Onev1d>;
4413 class St1Pat<ValueType ty, Instruction INST>
4414 : Pat<(store ty:$Vt, am_simdnoindex:$vaddr),
4415 (INST ty:$Vt, am_simdnoindex:$vaddr)>;
4417 def : St1Pat<v16i8, ST1Onev16b>;
4418 def : St1Pat<v8i16, ST1Onev8h>;
4419 def : St1Pat<v4i32, ST1Onev4s>;
4420 def : St1Pat<v2i64, ST1Onev2d>;
4421 def : St1Pat<v8i8, ST1Onev8b>;
4422 def : St1Pat<v4i16, ST1Onev4h>;
4423 def : St1Pat<v2i32, ST1Onev2s>;
4424 def : St1Pat<v1i64, ST1Onev1d>;
4430 defm LD1R : SIMDLdR<0, 0b110, 0, "ld1r", "One", 1, 2, 4, 8>;
4431 defm LD2R : SIMDLdR<1, 0b110, 0, "ld2r", "Two", 2, 4, 8, 16>;
4432 defm LD3R : SIMDLdR<0, 0b111, 0, "ld3r", "Three", 3, 6, 12, 24>;
4433 defm LD4R : SIMDLdR<1, 0b111, 0, "ld4r", "Four", 4, 8, 16, 32>;
4434 let mayLoad = 1, neverHasSideEffects = 1 in {
4435 defm LD1 : SIMDLdSingleBTied<0, 0b000, "ld1", VecListOneb, GPR64pi1>;
4436 defm LD1 : SIMDLdSingleHTied<0, 0b010, 0, "ld1", VecListOneh, GPR64pi2>;
4437 defm LD1 : SIMDLdSingleSTied<0, 0b100, 0b00, "ld1", VecListOnes, GPR64pi4>;
4438 defm LD1 : SIMDLdSingleDTied<0, 0b100, 0b01, "ld1", VecListOned, GPR64pi8>;
4439 defm LD2 : SIMDLdSingleBTied<1, 0b000, "ld2", VecListTwob, GPR64pi2>;
4440 defm LD2 : SIMDLdSingleHTied<1, 0b010, 0, "ld2", VecListTwoh, GPR64pi4>;
4441 defm LD2 : SIMDLdSingleSTied<1, 0b100, 0b00, "ld2", VecListTwos, GPR64pi8>;
4442 defm LD2 : SIMDLdSingleDTied<1, 0b100, 0b01, "ld2", VecListTwod, GPR64pi16>;
4443 defm LD3 : SIMDLdSingleBTied<0, 0b001, "ld3", VecListThreeb, GPR64pi3>;
4444 defm LD3 : SIMDLdSingleHTied<0, 0b011, 0, "ld3", VecListThreeh, GPR64pi6>;
4445 defm LD3 : SIMDLdSingleSTied<0, 0b101, 0b00, "ld3", VecListThrees, GPR64pi12>;
4446 defm LD3 : SIMDLdSingleDTied<0, 0b101, 0b01, "ld3", VecListThreed, GPR64pi24>;
4447 defm LD4 : SIMDLdSingleBTied<1, 0b001, "ld4", VecListFourb, GPR64pi4>;
4448 defm LD4 : SIMDLdSingleHTied<1, 0b011, 0, "ld4", VecListFourh, GPR64pi8>;
4449 defm LD4 : SIMDLdSingleSTied<1, 0b101, 0b00, "ld4", VecListFours, GPR64pi16>;
4450 defm LD4 : SIMDLdSingleDTied<1, 0b101, 0b01, "ld4", VecListFourd, GPR64pi32>;
4453 def : Pat<(v8i8 (ARM64dup (i32 (extloadi8 am_simdnoindex:$vaddr)))),
4454 (LD1Rv8b am_simdnoindex:$vaddr)>;
4455 def : Pat<(v16i8 (ARM64dup (i32 (extloadi8 am_simdnoindex:$vaddr)))),
4456 (LD1Rv16b am_simdnoindex:$vaddr)>;
4457 def : Pat<(v4i16 (ARM64dup (i32 (extloadi16 am_simdnoindex:$vaddr)))),
4458 (LD1Rv4h am_simdnoindex:$vaddr)>;
4459 def : Pat<(v8i16 (ARM64dup (i32 (extloadi16 am_simdnoindex:$vaddr)))),
4460 (LD1Rv8h am_simdnoindex:$vaddr)>;
4461 def : Pat<(v2i32 (ARM64dup (i32 (load am_simdnoindex:$vaddr)))),
4462 (LD1Rv2s am_simdnoindex:$vaddr)>;
4463 def : Pat<(v4i32 (ARM64dup (i32 (load am_simdnoindex:$vaddr)))),
4464 (LD1Rv4s am_simdnoindex:$vaddr)>;
4465 def : Pat<(v2i64 (ARM64dup (i64 (load am_simdnoindex:$vaddr)))),
4466 (LD1Rv2d am_simdnoindex:$vaddr)>;
4467 def : Pat<(v1i64 (ARM64dup (i64 (load am_simdnoindex:$vaddr)))),
4468 (LD1Rv1d am_simdnoindex:$vaddr)>;
4469 // Grab the floating point version too
4470 def : Pat<(v2f32 (ARM64dup (f32 (load am_simdnoindex:$vaddr)))),
4471 (LD1Rv2s am_simdnoindex:$vaddr)>;
4472 def : Pat<(v4f32 (ARM64dup (f32 (load am_simdnoindex:$vaddr)))),
4473 (LD1Rv4s am_simdnoindex:$vaddr)>;
4474 def : Pat<(v2f64 (ARM64dup (f64 (load am_simdnoindex:$vaddr)))),
4475 (LD1Rv2d am_simdnoindex:$vaddr)>;
4476 def : Pat<(v1f64 (ARM64dup (f64 (load am_simdnoindex:$vaddr)))),
4477 (LD1Rv1d am_simdnoindex:$vaddr)>;
4479 class Ld1Lane128Pat<SDPatternOperator scalar_load, Operand VecIndex,
4480 ValueType VTy, ValueType STy, Instruction LD1>
4481 : Pat<(vector_insert (VTy VecListOne128:$Rd),
4482 (STy (scalar_load am_simdnoindex:$vaddr)), VecIndex:$idx),
4483 (LD1 VecListOne128:$Rd, VecIndex:$idx, am_simdnoindex:$vaddr)>;
4485 def : Ld1Lane128Pat<extloadi8, VectorIndexB, v16i8, i32, LD1i8>;
4486 def : Ld1Lane128Pat<extloadi16, VectorIndexH, v8i16, i32, LD1i16>;
4487 def : Ld1Lane128Pat<load, VectorIndexS, v4i32, i32, LD1i32>;
4488 def : Ld1Lane128Pat<load, VectorIndexS, v4f32, f32, LD1i32>;
4489 def : Ld1Lane128Pat<load, VectorIndexD, v2i64, i64, LD1i64>;
4490 def : Ld1Lane128Pat<load, VectorIndexD, v2f64, f64, LD1i64>;
4492 class Ld1Lane64Pat<SDPatternOperator scalar_load, Operand VecIndex,
4493 ValueType VTy, ValueType STy, Instruction LD1>
4494 : Pat<(vector_insert (VTy VecListOne64:$Rd),
4495 (STy (scalar_load am_simdnoindex:$vaddr)), VecIndex:$idx),
4497 (LD1 (SUBREG_TO_REG (i32 0), VecListOne64:$Rd, dsub),
4498 VecIndex:$idx, am_simdnoindex:$vaddr),
4501 def : Ld1Lane64Pat<extloadi8, VectorIndexB, v8i8, i32, LD1i8>;
4502 def : Ld1Lane64Pat<extloadi16, VectorIndexH, v4i16, i32, LD1i16>;
4503 def : Ld1Lane64Pat<load, VectorIndexS, v2i32, i32, LD1i32>;
4504 def : Ld1Lane64Pat<load, VectorIndexS, v2f32, f32, LD1i32>;
4507 defm LD1 : SIMDLdSt1SingleAliases<"ld1">;
4508 defm LD2 : SIMDLdSt2SingleAliases<"ld2">;
4509 defm LD3 : SIMDLdSt3SingleAliases<"ld3">;
4510 defm LD4 : SIMDLdSt4SingleAliases<"ld4">;
4513 defm ST1 : SIMDStSingleB<0, 0b000, "st1", VecListOneb, GPR64pi1>;
4514 defm ST1 : SIMDStSingleH<0, 0b010, 0, "st1", VecListOneh, GPR64pi2>;
4515 defm ST1 : SIMDStSingleS<0, 0b100, 0b00, "st1", VecListOnes, GPR64pi4>;
4516 defm ST1 : SIMDStSingleD<0, 0b100, 0b01, "st1", VecListOned, GPR64pi8>;
4518 let AddedComplexity = 8 in
4519 class St1Lane128Pat<SDPatternOperator scalar_store, Operand VecIndex,
4520 ValueType VTy, ValueType STy, Instruction ST1>
4522 (STy (vector_extract (VTy VecListOne128:$Vt), VecIndex:$idx)),
4523 am_simdnoindex:$vaddr),
4524 (ST1 VecListOne128:$Vt, VecIndex:$idx, am_simdnoindex:$vaddr)>;
4526 def : St1Lane128Pat<truncstorei8, VectorIndexB, v16i8, i32, ST1i8>;
4527 def : St1Lane128Pat<truncstorei16, VectorIndexH, v8i16, i32, ST1i16>;
4528 def : St1Lane128Pat<store, VectorIndexS, v4i32, i32, ST1i32>;
4529 def : St1Lane128Pat<store, VectorIndexS, v4f32, f32, ST1i32>;
4530 def : St1Lane128Pat<store, VectorIndexD, v2i64, i64, ST1i64>;
4531 def : St1Lane128Pat<store, VectorIndexD, v2f64, f64, ST1i64>;
4533 let AddedComplexity = 8 in
4534 class St1Lane64Pat<SDPatternOperator scalar_store, Operand VecIndex,
4535 ValueType VTy, ValueType STy, Instruction ST1>
4537 (STy (vector_extract (VTy VecListOne64:$Vt), VecIndex:$idx)),
4538 am_simdnoindex:$vaddr),
4539 (ST1 (SUBREG_TO_REG (i32 0), VecListOne64:$Vt, dsub),
4540 VecIndex:$idx, am_simdnoindex:$vaddr)>;
4542 def : St1Lane64Pat<truncstorei8, VectorIndexB, v8i8, i32, ST1i8>;
4543 def : St1Lane64Pat<truncstorei16, VectorIndexH, v4i16, i32, ST1i16>;
4544 def : St1Lane64Pat<store, VectorIndexS, v2i32, i32, ST1i32>;
4545 def : St1Lane64Pat<store, VectorIndexS, v2f32, f32, ST1i32>;
4547 multiclass St1LanePost64Pat<SDPatternOperator scalar_store, Operand VecIndex,
4548 ValueType VTy, ValueType STy, Instruction ST1,
4550 def : Pat<(scalar_store
4551 (STy (vector_extract (VTy VecListOne64:$Vt), VecIndex:$idx)),
4552 am_simdnoindex:$vaddr, offset),
4553 (ST1 (SUBREG_TO_REG (i32 0), VecListOne64:$Vt, dsub),
4554 VecIndex:$idx, am_simdnoindex:$vaddr, XZR)>;
4556 def : Pat<(scalar_store
4557 (STy (vector_extract (VTy VecListOne64:$Vt), VecIndex:$idx)),
4558 am_simdnoindex:$vaddr, GPR64:$Rm),
4559 (ST1 (SUBREG_TO_REG (i32 0), VecListOne64:$Vt, dsub),
4560 VecIndex:$idx, am_simdnoindex:$vaddr, $Rm)>;
4563 defm : St1LanePost64Pat<post_truncsti8, VectorIndexB, v8i8, i32, ST1i8_POST, 1>;
4564 defm : St1LanePost64Pat<post_truncsti16, VectorIndexH, v4i16, i32, ST1i16_POST,
4566 defm : St1LanePost64Pat<post_store, VectorIndexS, v2i32, i32, ST1i32_POST, 4>;
4567 defm : St1LanePost64Pat<post_store, VectorIndexS, v2f32, f32, ST1i32_POST, 4>;
4568 defm : St1LanePost64Pat<post_store, VectorIndexD, v1i64, i64, ST1i64_POST, 8>;
4569 defm : St1LanePost64Pat<post_store, VectorIndexD, v1f64, f64, ST1i64_POST, 8>;
4571 multiclass St1LanePost128Pat<SDPatternOperator scalar_store, Operand VecIndex,
4572 ValueType VTy, ValueType STy, Instruction ST1,
4574 def : Pat<(scalar_store
4575 (STy (vector_extract (VTy VecListOne128:$Vt), VecIndex:$idx)),
4576 am_simdnoindex:$vaddr, offset),
4577 (ST1 VecListOne128:$Vt, VecIndex:$idx, am_simdnoindex:$vaddr, XZR)>;
4579 def : Pat<(scalar_store
4580 (STy (vector_extract (VTy VecListOne128:$Vt), VecIndex:$idx)),
4581 am_simdnoindex:$vaddr, GPR64:$Rm),
4582 (ST1 VecListOne128:$Vt, VecIndex:$idx, am_simdnoindex:$vaddr, $Rm)>;
4585 defm : St1LanePost128Pat<post_truncsti8, VectorIndexB, v16i8, i32, ST1i8_POST,
4587 defm : St1LanePost128Pat<post_truncsti16, VectorIndexH, v8i16, i32, ST1i16_POST,
4589 defm : St1LanePost128Pat<post_store, VectorIndexS, v4i32, i32, ST1i32_POST, 4>;
4590 defm : St1LanePost128Pat<post_store, VectorIndexS, v4f32, f32, ST1i32_POST, 4>;
4591 defm : St1LanePost128Pat<post_store, VectorIndexD, v2i64, i64, ST1i64_POST, 8>;
4592 defm : St1LanePost128Pat<post_store, VectorIndexD, v2f64, f64, ST1i64_POST, 8>;
4594 let mayStore = 1, neverHasSideEffects = 1 in {
4595 defm ST2 : SIMDStSingleB<1, 0b000, "st2", VecListTwob, GPR64pi2>;
4596 defm ST2 : SIMDStSingleH<1, 0b010, 0, "st2", VecListTwoh, GPR64pi4>;
4597 defm ST2 : SIMDStSingleS<1, 0b100, 0b00, "st2", VecListTwos, GPR64pi8>;
4598 defm ST2 : SIMDStSingleD<1, 0b100, 0b01, "st2", VecListTwod, GPR64pi16>;
4599 defm ST3 : SIMDStSingleB<0, 0b001, "st3", VecListThreeb, GPR64pi3>;
4600 defm ST3 : SIMDStSingleH<0, 0b011, 0, "st3", VecListThreeh, GPR64pi6>;
4601 defm ST3 : SIMDStSingleS<0, 0b101, 0b00, "st3", VecListThrees, GPR64pi12>;
4602 defm ST3 : SIMDStSingleD<0, 0b101, 0b01, "st3", VecListThreed, GPR64pi24>;
4603 defm ST4 : SIMDStSingleB<1, 0b001, "st4", VecListFourb, GPR64pi4>;
4604 defm ST4 : SIMDStSingleH<1, 0b011, 0, "st4", VecListFourh, GPR64pi8>;
4605 defm ST4 : SIMDStSingleS<1, 0b101, 0b00, "st4", VecListFours, GPR64pi16>;
4606 defm ST4 : SIMDStSingleD<1, 0b101, 0b01, "st4", VecListFourd, GPR64pi32>;
4609 defm ST1 : SIMDLdSt1SingleAliases<"st1">;
4610 defm ST2 : SIMDLdSt2SingleAliases<"st2">;
4611 defm ST3 : SIMDLdSt3SingleAliases<"st3">;
4612 defm ST4 : SIMDLdSt4SingleAliases<"st4">;
4614 //----------------------------------------------------------------------------
4615 // Crypto extensions
4616 //----------------------------------------------------------------------------
4618 def AESErr : AESTiedInst<0b0100, "aese", int_arm64_crypto_aese>;
4619 def AESDrr : AESTiedInst<0b0101, "aesd", int_arm64_crypto_aesd>;
4620 def AESMCrr : AESInst< 0b0110, "aesmc", int_arm64_crypto_aesmc>;
4621 def AESIMCrr : AESInst< 0b0111, "aesimc", int_arm64_crypto_aesimc>;
4623 def SHA1Crrr : SHATiedInstQSV<0b000, "sha1c", int_arm64_crypto_sha1c>;
4624 def SHA1Prrr : SHATiedInstQSV<0b001, "sha1p", int_arm64_crypto_sha1p>;
4625 def SHA1Mrrr : SHATiedInstQSV<0b010, "sha1m", int_arm64_crypto_sha1m>;
4626 def SHA1SU0rrr : SHATiedInstVVV<0b011, "sha1su0", int_arm64_crypto_sha1su0>;
4627 def SHA256Hrrr : SHATiedInstQQV<0b100, "sha256h", int_arm64_crypto_sha256h>;
4628 def SHA256H2rrr : SHATiedInstQQV<0b101, "sha256h2",int_arm64_crypto_sha256h2>;
4629 def SHA256SU1rrr :SHATiedInstVVV<0b110, "sha256su1",int_arm64_crypto_sha256su1>;
4631 def SHA1Hrr : SHAInstSS< 0b0000, "sha1h", int_arm64_crypto_sha1h>;
4632 def SHA1SU1rr : SHATiedInstVV<0b0001, "sha1su1", int_arm64_crypto_sha1su1>;
4633 def SHA256SU0rr : SHATiedInstVV<0b0010, "sha256su0",int_arm64_crypto_sha256su0>;
4635 //----------------------------------------------------------------------------
4637 //----------------------------------------------------------------------------
4638 // FIXME: Like for X86, these should go in their own separate .td file.
4640 // Any instruction that defines a 32-bit result leaves the high half of the
4641 // register. Truncate can be lowered to EXTRACT_SUBREG. CopyFromReg may
4642 // be copying from a truncate. But any other 32-bit operation will zero-extend
4644 // FIXME: X86 also checks for CMOV here. Do we need something similar?
4645 def def32 : PatLeaf<(i32 GPR32:$src), [{
4646 return N->getOpcode() != ISD::TRUNCATE &&
4647 N->getOpcode() != TargetOpcode::EXTRACT_SUBREG &&
4648 N->getOpcode() != ISD::CopyFromReg;
4651 // In the case of a 32-bit def that is known to implicitly zero-extend,
4652 // we can use a SUBREG_TO_REG.
4653 def : Pat<(i64 (zext def32:$src)), (SUBREG_TO_REG (i64 0), GPR32:$src, sub_32)>;
4655 // For an anyext, we don't care what the high bits are, so we can perform an
4656 // INSERT_SUBREF into an IMPLICIT_DEF.
4657 def : Pat<(i64 (anyext GPR32:$src)),
4658 (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32)>;
4660 // When we need to explicitly zero-extend, we use an unsigned bitfield move
4661 // instruction (UBFM) on the enclosing super-reg.
4662 def : Pat<(i64 (zext GPR32:$src)),
4663 (UBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32), 0, 31)>;
4665 // To sign extend, we use a signed bitfield move instruction (SBFM) on the
4666 // containing super-reg.
4667 def : Pat<(i64 (sext GPR32:$src)),
4668 (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32), 0, 31)>;
4669 def : Pat<(i64 (sext_inreg GPR64:$src, i32)), (SBFMXri GPR64:$src, 0, 31)>;
4670 def : Pat<(i64 (sext_inreg GPR64:$src, i16)), (SBFMXri GPR64:$src, 0, 15)>;
4671 def : Pat<(i64 (sext_inreg GPR64:$src, i8)), (SBFMXri GPR64:$src, 0, 7)>;
4672 def : Pat<(i64 (sext_inreg GPR64:$src, i1)), (SBFMXri GPR64:$src, 0, 0)>;
4673 def : Pat<(i32 (sext_inreg GPR32:$src, i16)), (SBFMWri GPR32:$src, 0, 15)>;
4674 def : Pat<(i32 (sext_inreg GPR32:$src, i8)), (SBFMWri GPR32:$src, 0, 7)>;
4675 def : Pat<(i32 (sext_inreg GPR32:$src, i1)), (SBFMWri GPR32:$src, 0, 0)>;
4677 def : Pat<(shl (sext_inreg GPR32:$Rn, i8), (i64 imm0_31:$imm)),
4678 (SBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
4679 (i64 (i32shift_sext_i8 imm0_31:$imm)))>;
4680 def : Pat<(shl (sext_inreg GPR64:$Rn, i8), (i64 imm0_63:$imm)),
4681 (SBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
4682 (i64 (i64shift_sext_i8 imm0_63:$imm)))>;
4684 def : Pat<(shl (sext_inreg GPR32:$Rn, i16), (i64 imm0_31:$imm)),
4685 (SBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
4686 (i64 (i32shift_sext_i16 imm0_31:$imm)))>;
4687 def : Pat<(shl (sext_inreg GPR64:$Rn, i16), (i64 imm0_63:$imm)),
4688 (SBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
4689 (i64 (i64shift_sext_i16 imm0_63:$imm)))>;
4691 def : Pat<(shl (i64 (sext GPR32:$Rn)), (i64 imm0_63:$imm)),
4692 (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$Rn, sub_32),
4693 (i64 (i64shift_a imm0_63:$imm)),
4694 (i64 (i64shift_sext_i32 imm0_63:$imm)))>;
4696 // sra patterns have an AddedComplexity of 10, so make sure we have a higher
4697 // AddedComplexity for the following patterns since we want to match sext + sra
4698 // patterns before we attempt to match a single sra node.
4699 let AddedComplexity = 20 in {
4700 // We support all sext + sra combinations which preserve at least one bit of the
4701 // original value which is to be sign extended. E.g. we support shifts up to
4703 def : Pat<(sra (sext_inreg GPR32:$Rn, i8), (i64 imm0_7:$imm)),
4704 (SBFMWri GPR32:$Rn, (i64 imm0_7:$imm), 7)>;
4705 def : Pat<(sra (sext_inreg GPR64:$Rn, i8), (i64 imm0_7:$imm)),
4706 (SBFMXri GPR64:$Rn, (i64 imm0_7:$imm), 7)>;
4708 def : Pat<(sra (sext_inreg GPR32:$Rn, i16), (i64 imm0_15:$imm)),
4709 (SBFMWri GPR32:$Rn, (i64 imm0_15:$imm), 15)>;
4710 def : Pat<(sra (sext_inreg GPR64:$Rn, i16), (i64 imm0_15:$imm)),
4711 (SBFMXri GPR64:$Rn, (i64 imm0_15:$imm), 15)>;
4713 def : Pat<(sra (i64 (sext GPR32:$Rn)), (i64 imm0_31:$imm)),
4714 (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$Rn, sub_32),
4715 (i64 imm0_31:$imm), 31)>;
4716 } // AddedComplexity = 20
4718 // To truncate, we can simply extract from a subregister.
4719 def : Pat<(i32 (trunc GPR64sp:$src)),
4720 (i32 (EXTRACT_SUBREG GPR64sp:$src, sub_32))>;
4722 // __builtin_trap() uses the BRK instruction on ARM64.
4723 def : Pat<(trap), (BRK 1)>;
4725 // Conversions within AdvSIMD types in the same register size are free.
4726 // But because we need a consistent lane ordering, in big endian many
4727 // conversions require one or more REV instructions.
4729 // Consider a simple memory load followed by a bitconvert then a store.
4731 // v1 = BITCAST v2i32 v0 to v4i16
4734 // In big endian mode every memory access has an implicit byte swap. LDR and
4735 // STR do a 64-bit byte swap, whereas LD1/ST1 do a byte swap per lane - that
4736 // is, they treat the vector as a sequence of elements to be byte-swapped.
4737 // The two pairs of instructions are fundamentally incompatible. We've decided
4738 // to use LD1/ST1 only to simplify compiler implementation.
4740 // LD1/ST1 perform the equivalent of a sequence of LDR/STR + REV. This makes
4741 // the original code sequence:
4743 // v1 = REV v2i32 (implicit)
4744 // v2 = BITCAST v2i32 v1 to v4i16
4745 // v3 = REV v4i16 v2 (implicit)
4748 // But this is now broken - the value stored is different to the value loaded
4749 // due to lane reordering. To fix this, on every BITCAST we must perform two
4752 // v1 = REV v2i32 (implicit)
4754 // v3 = BITCAST v2i32 v2 to v4i16
4756 // v5 = REV v4i16 v4 (implicit)
4759 // This means an extra two instructions, but actually in most cases the two REV
4760 // instructions can be combined into one. For example:
4761 // (REV64_2s (REV64_4h X)) === (REV32_4h X)
4763 // There is also no 128-bit REV instruction. This must be synthesized with an
4766 // Most bitconverts require some sort of conversion. The only exceptions are:
4767 // a) Identity conversions - vNfX <-> vNiX
4768 // b) Single-lane-to-scalar - v1fX <-> fX or v1iX <-> iX
4771 let Predicates = [IsLE] in {
4772 def : Pat<(v8i8 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
4773 def : Pat<(v4i16 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
4774 def : Pat<(v2i32 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
4775 def : Pat<(v2f32 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
4777 def : Pat<(i64 (bitconvert (v8i8 V64:$Vn))),
4778 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
4779 def : Pat<(i64 (bitconvert (v4i16 V64:$Vn))),
4780 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
4781 def : Pat<(i64 (bitconvert (v2i32 V64:$Vn))),
4782 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
4783 def : Pat<(i64 (bitconvert (v2f32 V64:$Vn))),
4784 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
4785 def : Pat<(i64 (bitconvert (v1f64 V64:$Vn))),
4786 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
4788 let Predicates = [IsBE] in {
4789 def : Pat<(v8i8 (bitconvert GPR64:$Xn)),
4790 (REV64v8i8 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
4791 def : Pat<(v4i16 (bitconvert GPR64:$Xn)),
4792 (REV64v4i16 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
4793 def : Pat<(v2i32 (bitconvert GPR64:$Xn)),
4794 (REV64v2i32 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
4795 def : Pat<(v2f32 (bitconvert GPR64:$Xn)),
4796 (REV64v2i32 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
4798 def : Pat<(i64 (bitconvert (v8i8 V64:$Vn))),
4799 (REV64v8i8 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
4800 def : Pat<(i64 (bitconvert (v4i16 V64:$Vn))),
4801 (REV64v4i16 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
4802 def : Pat<(i64 (bitconvert (v2i32 V64:$Vn))),
4803 (REV64v2i32 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
4804 def : Pat<(i64 (bitconvert (v2f32 V64:$Vn))),
4805 (REV64v2i32 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
4807 def : Pat<(v1i64 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
4808 def : Pat<(v1f64 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
4809 def : Pat<(i64 (bitconvert (v1i64 V64:$Vn))),
4810 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
4811 def : Pat<(v1i64 (scalar_to_vector GPR64:$Xn)),
4812 (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
4813 def : Pat<(v1f64 (scalar_to_vector GPR64:$Xn)),
4814 (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
4815 def : Pat<(v1f64 (scalar_to_vector (f64 FPR64:$Xn))), (v1f64 FPR64:$Xn)>;
4817 def : Pat<(f32 (bitconvert (i32 GPR32:$Xn))),
4818 (COPY_TO_REGCLASS GPR32:$Xn, FPR32)>;
4819 def : Pat<(i32 (bitconvert (f32 FPR32:$Xn))),
4820 (COPY_TO_REGCLASS FPR32:$Xn, GPR32)>;
4821 def : Pat<(f64 (bitconvert (i64 GPR64:$Xn))),
4822 (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
4823 def : Pat<(i64 (bitconvert (f64 FPR64:$Xn))),
4824 (COPY_TO_REGCLASS FPR64:$Xn, GPR64)>;
4825 def : Pat<(i64 (bitconvert (v1f64 V64:$Vn))),
4826 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
4828 let Predicates = [IsLE] in {
4829 def : Pat<(v1i64 (bitconvert (v2i32 FPR64:$src))), (v1i64 FPR64:$src)>;
4830 def : Pat<(v1i64 (bitconvert (v4i16 FPR64:$src))), (v1i64 FPR64:$src)>;
4831 def : Pat<(v1i64 (bitconvert (v8i8 FPR64:$src))), (v1i64 FPR64:$src)>;
4832 def : Pat<(v1i64 (bitconvert (v2f32 FPR64:$src))), (v1i64 FPR64:$src)>;
4834 let Predicates = [IsBE] in {
4835 def : Pat<(v1i64 (bitconvert (v2i32 FPR64:$src))),
4836 (v1i64 (REV64v2i32 FPR64:$src))>;
4837 def : Pat<(v1i64 (bitconvert (v4i16 FPR64:$src))),
4838 (v1i64 (REV64v4i16 FPR64:$src))>;
4839 def : Pat<(v1i64 (bitconvert (v8i8 FPR64:$src))),
4840 (v1i64 (REV64v8i8 FPR64:$src))>;
4841 def : Pat<(v1i64 (bitconvert (v2f32 FPR64:$src))),
4842 (v1i64 (REV64v2i32 FPR64:$src))>;
4844 def : Pat<(v1i64 (bitconvert (v1f64 FPR64:$src))), (v1i64 FPR64:$src)>;
4845 def : Pat<(v1i64 (bitconvert (f64 FPR64:$src))), (v1i64 FPR64:$src)>;
4847 let Predicates = [IsLE] in {
4848 def : Pat<(v2i32 (bitconvert (v1i64 FPR64:$src))), (v2i32 FPR64:$src)>;
4849 def : Pat<(v2i32 (bitconvert (v4i16 FPR64:$src))), (v2i32 FPR64:$src)>;
4850 def : Pat<(v2i32 (bitconvert (v8i8 FPR64:$src))), (v2i32 FPR64:$src)>;
4851 def : Pat<(v2i32 (bitconvert (f64 FPR64:$src))), (v2i32 FPR64:$src)>;
4852 def : Pat<(v2i32 (bitconvert (v1f64 FPR64:$src))), (v2i32 FPR64:$src)>;
4854 let Predicates = [IsBE] in {
4855 def : Pat<(v2i32 (bitconvert (v1i64 FPR64:$src))),
4856 (v2i32 (REV64v2i32 FPR64:$src))>;
4857 def : Pat<(v2i32 (bitconvert (v4i16 FPR64:$src))),
4858 (v2i32 (REV32v4i16 FPR64:$src))>;
4859 def : Pat<(v2i32 (bitconvert (v8i8 FPR64:$src))),
4860 (v2i32 (REV32v8i8 FPR64:$src))>;
4861 def : Pat<(v2i32 (bitconvert (f64 FPR64:$src))),
4862 (v2i32 (REV64v2i32 FPR64:$src))>;
4863 def : Pat<(v2i32 (bitconvert (v1f64 FPR64:$src))),
4864 (v2i32 (REV64v2i32 FPR64:$src))>;
4866 def : Pat<(v2i32 (bitconvert (v2f32 FPR64:$src))), (v2i32 FPR64:$src)>;
4868 let Predicates = [IsLE] in {
4869 def : Pat<(v4i16 (bitconvert (v1i64 FPR64:$src))), (v4i16 FPR64:$src)>;
4870 def : Pat<(v4i16 (bitconvert (v2i32 FPR64:$src))), (v4i16 FPR64:$src)>;
4871 def : Pat<(v4i16 (bitconvert (v8i8 FPR64:$src))), (v4i16 FPR64:$src)>;
4872 def : Pat<(v4i16 (bitconvert (f64 FPR64:$src))), (v4i16 FPR64:$src)>;
4873 def : Pat<(v4i16 (bitconvert (v2f32 FPR64:$src))), (v4i16 FPR64:$src)>;
4874 def : Pat<(v4i16 (bitconvert (v1f64 FPR64:$src))), (v4i16 FPR64:$src)>;
4876 let Predicates = [IsBE] in {
4877 def : Pat<(v4i16 (bitconvert (v1i64 FPR64:$src))),
4878 (v4i16 (REV64v4i16 FPR64:$src))>;
4879 def : Pat<(v4i16 (bitconvert (v2i32 FPR64:$src))),
4880 (v4i16 (REV32v4i16 FPR64:$src))>;
4881 def : Pat<(v4i16 (bitconvert (v8i8 FPR64:$src))),
4882 (v4i16 (REV16v8i8 FPR64:$src))>;
4883 def : Pat<(v4i16 (bitconvert (f64 FPR64:$src))),
4884 (v4i16 (REV64v4i16 FPR64:$src))>;
4885 def : Pat<(v4i16 (bitconvert (v2f32 FPR64:$src))),
4886 (v4i16 (REV32v4i16 FPR64:$src))>;
4887 def : Pat<(v4i16 (bitconvert (v1f64 FPR64:$src))),
4888 (v4i16 (REV64v4i16 FPR64:$src))>;
4891 let Predicates = [IsLE] in {
4892 def : Pat<(v8i8 (bitconvert (v1i64 FPR64:$src))), (v8i8 FPR64:$src)>;
4893 def : Pat<(v8i8 (bitconvert (v2i32 FPR64:$src))), (v8i8 FPR64:$src)>;
4894 def : Pat<(v8i8 (bitconvert (v4i16 FPR64:$src))), (v8i8 FPR64:$src)>;
4895 def : Pat<(v8i8 (bitconvert (f64 FPR64:$src))), (v8i8 FPR64:$src)>;
4896 def : Pat<(v8i8 (bitconvert (v2f32 FPR64:$src))), (v8i8 FPR64:$src)>;
4897 def : Pat<(v8i8 (bitconvert (v1f64 FPR64:$src))), (v8i8 FPR64:$src)>;
4899 let Predicates = [IsBE] in {
4900 def : Pat<(v8i8 (bitconvert (v1i64 FPR64:$src))),
4901 (v8i8 (REV64v8i8 FPR64:$src))>;
4902 def : Pat<(v8i8 (bitconvert (v2i32 FPR64:$src))),
4903 (v8i8 (REV32v8i8 FPR64:$src))>;
4904 def : Pat<(v8i8 (bitconvert (v4i16 FPR64:$src))),
4905 (v8i8 (REV16v8i8 FPR64:$src))>;
4906 def : Pat<(v8i8 (bitconvert (f64 FPR64:$src))),
4907 (v8i8 (REV64v8i8 FPR64:$src))>;
4908 def : Pat<(v8i8 (bitconvert (v2f32 FPR64:$src))),
4909 (v8i8 (REV32v8i8 FPR64:$src))>;
4910 def : Pat<(v8i8 (bitconvert (v1f64 FPR64:$src))),
4911 (v8i8 (REV64v8i8 FPR64:$src))>;
4914 let Predicates = [IsLE] in {
4915 def : Pat<(f64 (bitconvert (v2i32 FPR64:$src))), (f64 FPR64:$src)>;
4916 def : Pat<(f64 (bitconvert (v4i16 FPR64:$src))), (f64 FPR64:$src)>;
4917 def : Pat<(f64 (bitconvert (v2f32 FPR64:$src))), (f64 FPR64:$src)>;
4918 def : Pat<(f64 (bitconvert (v8i8 FPR64:$src))), (f64 FPR64:$src)>;
4920 let Predicates = [IsBE] in {
4921 def : Pat<(f64 (bitconvert (v2i32 FPR64:$src))),
4922 (f64 (REV64v2i32 FPR64:$src))>;
4923 def : Pat<(f64 (bitconvert (v4i16 FPR64:$src))),
4924 (f64 (REV64v4i16 FPR64:$src))>;
4925 def : Pat<(f64 (bitconvert (v2f32 FPR64:$src))),
4926 (f64 (REV64v2i32 FPR64:$src))>;
4927 def : Pat<(f64 (bitconvert (v8i8 FPR64:$src))),
4928 (f64 (REV64v8i8 FPR64:$src))>;
4930 def : Pat<(f64 (bitconvert (v1i64 FPR64:$src))), (f64 FPR64:$src)>;
4931 def : Pat<(f64 (bitconvert (v1f64 FPR64:$src))), (f64 FPR64:$src)>;
4933 let Predicates = [IsLE] in {
4934 def : Pat<(v1f64 (bitconvert (v2i32 FPR64:$src))), (v1f64 FPR64:$src)>;
4935 def : Pat<(v1f64 (bitconvert (v4i16 FPR64:$src))), (v1f64 FPR64:$src)>;
4936 def : Pat<(v1f64 (bitconvert (v8i8 FPR64:$src))), (v1f64 FPR64:$src)>;
4937 def : Pat<(v1f64 (bitconvert (v2f32 FPR64:$src))), (v1f64 FPR64:$src)>;
4939 let Predicates = [IsBE] in {
4940 def : Pat<(v1f64 (bitconvert (v2i32 FPR64:$src))),
4941 (v1f64 (REV64v2i32 FPR64:$src))>;
4942 def : Pat<(v1f64 (bitconvert (v4i16 FPR64:$src))),
4943 (v1f64 (REV64v4i16 FPR64:$src))>;
4944 def : Pat<(v1f64 (bitconvert (v8i8 FPR64:$src))),
4945 (v1f64 (REV64v8i8 FPR64:$src))>;
4946 def : Pat<(v1f64 (bitconvert (v2f32 FPR64:$src))),
4947 (v1f64 (REV64v2i32 FPR64:$src))>;
4949 def : Pat<(v1f64 (bitconvert (v1i64 FPR64:$src))), (v1f64 FPR64:$src)>;
4950 def : Pat<(v1f64 (bitconvert (f64 FPR64:$src))), (v1f64 FPR64:$src)>;
4952 let Predicates = [IsLE] in {
4953 def : Pat<(v2f32 (bitconvert (v1i64 FPR64:$src))), (v2f32 FPR64:$src)>;
4954 def : Pat<(v2f32 (bitconvert (v4i16 FPR64:$src))), (v2f32 FPR64:$src)>;
4955 def : Pat<(v2f32 (bitconvert (v8i8 FPR64:$src))), (v2f32 FPR64:$src)>;
4956 def : Pat<(v2f32 (bitconvert (v1f64 FPR64:$src))), (v2f32 FPR64:$src)>;
4957 def : Pat<(v2f32 (bitconvert (f64 FPR64:$src))), (v2f32 FPR64:$src)>;
4959 let Predicates = [IsBE] in {
4960 def : Pat<(v2f32 (bitconvert (v1i64 FPR64:$src))),
4961 (v2f32 (REV64v2i32 FPR64:$src))>;
4962 def : Pat<(v2f32 (bitconvert (v4i16 FPR64:$src))),
4963 (v2f32 (REV32v4i16 FPR64:$src))>;
4964 def : Pat<(v2f32 (bitconvert (v8i8 FPR64:$src))),
4965 (v2f32 (REV32v8i8 FPR64:$src))>;
4966 def : Pat<(v2f32 (bitconvert (v1f64 FPR64:$src))),
4967 (v2f32 (REV64v2i32 FPR64:$src))>;
4968 def : Pat<(v2f32 (bitconvert (f64 FPR64:$src))),
4969 (v2f32 (REV64v2i32 FPR64:$src))>;
4971 def : Pat<(v2f32 (bitconvert (v2i32 FPR64:$src))), (v2f32 FPR64:$src)>;
4973 let Predicates = [IsLE] in {
4974 def : Pat<(f128 (bitconvert (v2i64 FPR128:$src))), (f128 FPR128:$src)>;
4975 def : Pat<(f128 (bitconvert (v4i32 FPR128:$src))), (f128 FPR128:$src)>;
4976 def : Pat<(f128 (bitconvert (v8i16 FPR128:$src))), (f128 FPR128:$src)>;
4977 def : Pat<(f128 (bitconvert (v2f64 FPR128:$src))), (f128 FPR128:$src)>;
4978 def : Pat<(f128 (bitconvert (v4f32 FPR128:$src))), (f128 FPR128:$src)>;
4979 def : Pat<(f128 (bitconvert (v16i8 FPR128:$src))), (f128 FPR128:$src)>;
4981 let Predicates = [IsBE] in {
4982 def : Pat<(f128 (bitconvert (v2i64 FPR128:$src))),
4983 (f128 (EXTv16i8 FPR128:$src, FPR128:$src, (i32 8)))>;
4984 def : Pat<(f128 (bitconvert (v4i32 FPR128:$src))),
4985 (f128 (EXTv16i8 (REV64v4i32 FPR128:$src),
4986 (REV64v4i32 FPR128:$src), (i32 8)))>;
4987 def : Pat<(f128 (bitconvert (v8i16 FPR128:$src))),
4988 (f128 (EXTv16i8 (REV64v8i16 FPR128:$src),
4989 (REV64v8i16 FPR128:$src), (i32 8)))>;
4990 def : Pat<(f128 (bitconvert (v2f64 FPR128:$src))),
4991 (f128 (EXTv16i8 FPR128:$src, FPR128:$src, (i32 8)))>;
4992 def : Pat<(f128 (bitconvert (v4f32 FPR128:$src))),
4993 (f128 (EXTv16i8 (REV64v4i32 FPR128:$src),
4994 (REV64v4i32 FPR128:$src), (i32 8)))>;
4995 def : Pat<(f128 (bitconvert (v16i8 FPR128:$src))),
4996 (f128 (EXTv16i8 (REV64v16i8 FPR128:$src),
4997 (REV64v16i8 FPR128:$src), (i32 8)))>;
5000 let Predicates = [IsLE] in {
5001 def : Pat<(v2f64 (bitconvert (f128 FPR128:$src))), (v2f64 FPR128:$src)>;
5002 def : Pat<(v2f64 (bitconvert (v4i32 FPR128:$src))), (v2f64 FPR128:$src)>;
5003 def : Pat<(v2f64 (bitconvert (v8i16 FPR128:$src))), (v2f64 FPR128:$src)>;
5004 def : Pat<(v2f64 (bitconvert (v16i8 FPR128:$src))), (v2f64 FPR128:$src)>;
5005 def : Pat<(v2f64 (bitconvert (v4f32 FPR128:$src))), (v2f64 FPR128:$src)>;
5007 let Predicates = [IsBE] in {
5008 def : Pat<(v2f64 (bitconvert (f128 FPR128:$src))),
5009 (v2f64 (EXTv16i8 FPR128:$src,
5010 FPR128:$src, (i32 8)))>;
5011 def : Pat<(v2f64 (bitconvert (v4i32 FPR128:$src))),
5012 (v2f64 (REV64v4i32 FPR128:$src))>;
5013 def : Pat<(v2f64 (bitconvert (v8i16 FPR128:$src))),
5014 (v2f64 (REV64v8i16 FPR128:$src))>;
5015 def : Pat<(v2f64 (bitconvert (v16i8 FPR128:$src))),
5016 (v2f64 (REV64v16i8 FPR128:$src))>;
5017 def : Pat<(v2f64 (bitconvert (v4f32 FPR128:$src))),
5018 (v2f64 (REV64v4i32 FPR128:$src))>;
5020 def : Pat<(v2f64 (bitconvert (v2i64 FPR128:$src))), (v2f64 FPR128:$src)>;
5022 let Predicates = [IsLE] in {
5023 def : Pat<(v4f32 (bitconvert (f128 FPR128:$src))), (v4f32 FPR128:$src)>;
5024 def : Pat<(v4f32 (bitconvert (v8i16 FPR128:$src))), (v4f32 FPR128:$src)>;
5025 def : Pat<(v4f32 (bitconvert (v16i8 FPR128:$src))), (v4f32 FPR128:$src)>;
5026 def : Pat<(v4f32 (bitconvert (v2i64 FPR128:$src))), (v4f32 FPR128:$src)>;
5027 def : Pat<(v4f32 (bitconvert (v2f64 FPR128:$src))), (v4f32 FPR128:$src)>;
5029 let Predicates = [IsBE] in {
5030 def : Pat<(v4f32 (bitconvert (f128 FPR128:$src))),
5031 (v4f32 (EXTv16i8 (REV64v4i32 FPR128:$src),
5032 (REV64v4i32 FPR128:$src), (i32 8)))>;
5033 def : Pat<(v4f32 (bitconvert (v8i16 FPR128:$src))),
5034 (v4f32 (REV32v8i16 FPR128:$src))>;
5035 def : Pat<(v4f32 (bitconvert (v16i8 FPR128:$src))),
5036 (v4f32 (REV32v16i8 FPR128:$src))>;
5037 def : Pat<(v4f32 (bitconvert (v2i64 FPR128:$src))),
5038 (v4f32 (REV64v4i32 FPR128:$src))>;
5039 def : Pat<(v4f32 (bitconvert (v2f64 FPR128:$src))),
5040 (v4f32 (REV64v4i32 FPR128:$src))>;
5042 def : Pat<(v4f32 (bitconvert (v4i32 FPR128:$src))), (v4f32 FPR128:$src)>;
5044 let Predicates = [IsLE] in {
5045 def : Pat<(v2i64 (bitconvert (f128 FPR128:$src))), (v2i64 FPR128:$src)>;
5046 def : Pat<(v2i64 (bitconvert (v4i32 FPR128:$src))), (v2i64 FPR128:$src)>;
5047 def : Pat<(v2i64 (bitconvert (v8i16 FPR128:$src))), (v2i64 FPR128:$src)>;
5048 def : Pat<(v2i64 (bitconvert (v16i8 FPR128:$src))), (v2i64 FPR128:$src)>;
5049 def : Pat<(v2i64 (bitconvert (v4f32 FPR128:$src))), (v2i64 FPR128:$src)>;
5051 let Predicates = [IsBE] in {
5052 def : Pat<(v2i64 (bitconvert (f128 FPR128:$src))),
5053 (v2i64 (EXTv16i8 FPR128:$src,
5054 FPR128:$src, (i32 8)))>;
5055 def : Pat<(v2i64 (bitconvert (v4i32 FPR128:$src))),
5056 (v2i64 (REV64v4i32 FPR128:$src))>;
5057 def : Pat<(v2i64 (bitconvert (v8i16 FPR128:$src))),
5058 (v2i64 (REV64v8i16 FPR128:$src))>;
5059 def : Pat<(v2i64 (bitconvert (v16i8 FPR128:$src))),
5060 (v2i64 (REV64v16i8 FPR128:$src))>;
5061 def : Pat<(v2i64 (bitconvert (v4f32 FPR128:$src))),
5062 (v2i64 (REV64v4i32 FPR128:$src))>;
5064 def : Pat<(v2i64 (bitconvert (v2f64 FPR128:$src))), (v2i64 FPR128:$src)>;
5066 let Predicates = [IsLE] in {
5067 def : Pat<(v4i32 (bitconvert (f128 FPR128:$src))), (v4i32 FPR128:$src)>;
5068 def : Pat<(v4i32 (bitconvert (v2i64 FPR128:$src))), (v4i32 FPR128:$src)>;
5069 def : Pat<(v4i32 (bitconvert (v8i16 FPR128:$src))), (v4i32 FPR128:$src)>;
5070 def : Pat<(v4i32 (bitconvert (v16i8 FPR128:$src))), (v4i32 FPR128:$src)>;
5071 def : Pat<(v4i32 (bitconvert (v2f64 FPR128:$src))), (v4i32 FPR128:$src)>;
5073 let Predicates = [IsBE] in {
5074 def : Pat<(v4i32 (bitconvert (f128 FPR128:$src))),
5075 (v4i32 (EXTv16i8 (REV64v4i32 FPR128:$src),
5076 (REV64v4i32 FPR128:$src),
5078 def : Pat<(v4i32 (bitconvert (v2i64 FPR128:$src))),
5079 (v4i32 (REV64v4i32 FPR128:$src))>;
5080 def : Pat<(v4i32 (bitconvert (v8i16 FPR128:$src))),
5081 (v4i32 (REV32v8i16 FPR128:$src))>;
5082 def : Pat<(v4i32 (bitconvert (v16i8 FPR128:$src))),
5083 (v4i32 (REV32v16i8 FPR128:$src))>;
5084 def : Pat<(v4i32 (bitconvert (v2f64 FPR128:$src))),
5085 (v4i32 (REV64v4i32 FPR128:$src))>;
5087 def : Pat<(v4i32 (bitconvert (v4f32 FPR128:$src))), (v4i32 FPR128:$src)>;
5089 let Predicates = [IsLE] in {
5090 def : Pat<(v8i16 (bitconvert (f128 FPR128:$src))), (v8i16 FPR128:$src)>;
5091 def : Pat<(v8i16 (bitconvert (v2i64 FPR128:$src))), (v8i16 FPR128:$src)>;
5092 def : Pat<(v8i16 (bitconvert (v4i32 FPR128:$src))), (v8i16 FPR128:$src)>;
5093 def : Pat<(v8i16 (bitconvert (v16i8 FPR128:$src))), (v8i16 FPR128:$src)>;
5094 def : Pat<(v8i16 (bitconvert (v2f64 FPR128:$src))), (v8i16 FPR128:$src)>;
5095 def : Pat<(v8i16 (bitconvert (v4f32 FPR128:$src))), (v8i16 FPR128:$src)>;
5097 let Predicates = [IsBE] in {
5098 def : Pat<(v8i16 (bitconvert (f128 FPR128:$src))),
5099 (v8i16 (EXTv16i8 (REV64v8i16 FPR128:$src),
5100 (REV64v8i16 FPR128:$src),
5102 def : Pat<(v8i16 (bitconvert (v2i64 FPR128:$src))),
5103 (v8i16 (REV64v8i16 FPR128:$src))>;
5104 def : Pat<(v8i16 (bitconvert (v4i32 FPR128:$src))),
5105 (v8i16 (REV32v8i16 FPR128:$src))>;
5106 def : Pat<(v8i16 (bitconvert (v16i8 FPR128:$src))),
5107 (v8i16 (REV16v16i8 FPR128:$src))>;
5108 def : Pat<(v8i16 (bitconvert (v2f64 FPR128:$src))),
5109 (v8i16 (REV64v8i16 FPR128:$src))>;
5110 def : Pat<(v8i16 (bitconvert (v4f32 FPR128:$src))),
5111 (v8i16 (REV32v8i16 FPR128:$src))>;
5114 let Predicates = [IsLE] in {
5115 def : Pat<(v16i8 (bitconvert (f128 FPR128:$src))), (v16i8 FPR128:$src)>;
5116 def : Pat<(v16i8 (bitconvert (v2i64 FPR128:$src))), (v16i8 FPR128:$src)>;
5117 def : Pat<(v16i8 (bitconvert (v4i32 FPR128:$src))), (v16i8 FPR128:$src)>;
5118 def : Pat<(v16i8 (bitconvert (v8i16 FPR128:$src))), (v16i8 FPR128:$src)>;
5119 def : Pat<(v16i8 (bitconvert (v2f64 FPR128:$src))), (v16i8 FPR128:$src)>;
5120 def : Pat<(v16i8 (bitconvert (v4f32 FPR128:$src))), (v16i8 FPR128:$src)>;
5122 let Predicates = [IsBE] in {
5123 def : Pat<(v16i8 (bitconvert (f128 FPR128:$src))),
5124 (v16i8 (EXTv16i8 (REV64v16i8 FPR128:$src),
5125 (REV64v16i8 FPR128:$src),
5127 def : Pat<(v16i8 (bitconvert (v2i64 FPR128:$src))),
5128 (v16i8 (REV64v16i8 FPR128:$src))>;
5129 def : Pat<(v16i8 (bitconvert (v4i32 FPR128:$src))),
5130 (v16i8 (REV32v16i8 FPR128:$src))>;
5131 def : Pat<(v16i8 (bitconvert (v8i16 FPR128:$src))),
5132 (v16i8 (REV16v16i8 FPR128:$src))>;
5133 def : Pat<(v16i8 (bitconvert (v2f64 FPR128:$src))),
5134 (v16i8 (REV64v16i8 FPR128:$src))>;
5135 def : Pat<(v16i8 (bitconvert (v4f32 FPR128:$src))),
5136 (v16i8 (REV32v16i8 FPR128:$src))>;
5139 def : Pat<(v8i8 (extract_subvector (v16i8 FPR128:$Rn), (i64 1))),
5140 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
5141 def : Pat<(v4i16 (extract_subvector (v8i16 FPR128:$Rn), (i64 1))),
5142 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
5143 def : Pat<(v2i32 (extract_subvector (v4i32 FPR128:$Rn), (i64 1))),
5144 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
5145 def : Pat<(v1i64 (extract_subvector (v2i64 FPR128:$Rn), (i64 1))),
5146 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
5148 // A 64-bit subvector insert to the first 128-bit vector position
5149 // is a subregister copy that needs no instruction.
5150 def : Pat<(insert_subvector undef, (v1i64 FPR64:$src), (i32 0)),
5151 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5152 def : Pat<(insert_subvector undef, (v1f64 FPR64:$src), (i32 0)),
5153 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5154 def : Pat<(insert_subvector undef, (v2i32 FPR64:$src), (i32 0)),
5155 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5156 def : Pat<(insert_subvector undef, (v2f32 FPR64:$src), (i32 0)),
5157 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5158 def : Pat<(insert_subvector undef, (v4i16 FPR64:$src), (i32 0)),
5159 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5160 def : Pat<(insert_subvector undef, (v8i8 FPR64:$src), (i32 0)),
5161 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5163 // Use pair-wise add instructions when summing up the lanes for v2f64, v2i64
5165 def : Pat<(i64 (add (vector_extract (v2i64 FPR128:$Rn), (i64 0)),
5166 (vector_extract (v2i64 FPR128:$Rn), (i64 1)))),
5167 (i64 (ADDPv2i64p (v2i64 FPR128:$Rn)))>;
5168 def : Pat<(f64 (fadd (vector_extract (v2f64 FPR128:$Rn), (i64 0)),
5169 (vector_extract (v2f64 FPR128:$Rn), (i64 1)))),
5170 (f64 (FADDPv2i64p (v2f64 FPR128:$Rn)))>;
5171 // vector_extract on 64-bit vectors gets promoted to a 128 bit vector,
5172 // so we match on v4f32 here, not v2f32. This will also catch adding
5173 // the low two lanes of a true v4f32 vector.
5174 def : Pat<(fadd (vector_extract (v4f32 FPR128:$Rn), (i64 0)),
5175 (vector_extract (v4f32 FPR128:$Rn), (i64 1))),
5176 (f32 (FADDPv2i32p (EXTRACT_SUBREG FPR128:$Rn, dsub)))>;
5178 // Scalar 64-bit shifts in FPR64 registers.
5179 def : Pat<(i64 (int_arm64_neon_sshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
5180 (SSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
5181 def : Pat<(i64 (int_arm64_neon_ushl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
5182 (USHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
5183 def : Pat<(i64 (int_arm64_neon_srshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
5184 (SRSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
5185 def : Pat<(i64 (int_arm64_neon_urshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
5186 (URSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
5188 // Tail call return handling. These are all compiler pseudo-instructions,
5189 // so no encoding information or anything like that.
5190 let isCall = 1, isTerminator = 1, isReturn = 1, isBarrier = 1, Uses = [SP] in {
5191 def TCRETURNdi : Pseudo<(outs), (ins i64imm:$dst, i32imm:$FPDiff),[]>;
5192 def TCRETURNri : Pseudo<(outs), (ins tcGPR64:$dst, i32imm:$FPDiff), []>;
5195 def : Pat<(ARM64tcret tcGPR64:$dst, (i32 timm:$FPDiff)),
5196 (TCRETURNri tcGPR64:$dst, imm:$FPDiff)>;
5197 def : Pat<(ARM64tcret tglobaladdr:$dst, (i32 timm:$FPDiff)),
5198 (TCRETURNdi texternalsym:$dst, imm:$FPDiff)>;
5199 def : Pat<(ARM64tcret texternalsym:$dst, (i32 timm:$FPDiff)),
5200 (TCRETURNdi texternalsym:$dst, imm:$FPDiff)>;
5202 include "ARM64InstrAtomics.td"