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), 3>;
572 def : InstAlias<"neg $dst, $src", (SUBXrs GPR64:$dst, XZR, GPR64:$src, 0), 3>;
573 def : InstAlias<"neg $dst, $src$shift",
574 (SUBWrs GPR32:$dst, WZR, GPR32:$src, arith_shift32:$shift), 2>;
575 def : InstAlias<"neg $dst, $src$shift",
576 (SUBXrs GPR64:$dst, XZR, GPR64:$src, arith_shift64:$shift), 2>;
578 def : InstAlias<"negs $dst, $src", (SUBSWrs GPR32:$dst, WZR, GPR32:$src, 0), 3>;
579 def : InstAlias<"negs $dst, $src", (SUBSXrs GPR64:$dst, XZR, GPR64:$src, 0), 3>;
580 def : InstAlias<"negs $dst, $src$shift",
581 (SUBSWrs GPR32:$dst, WZR, GPR32:$src, arith_shift32:$shift), 2>;
582 def : InstAlias<"negs $dst, $src$shift",
583 (SUBSXrs GPR64:$dst, XZR, GPR64:$src, arith_shift64:$shift), 2>;
586 // Unsigned/Signed divide
587 defm UDIV : Div<0, "udiv", udiv>;
588 defm SDIV : Div<1, "sdiv", sdiv>;
589 let isCodeGenOnly = 1 in {
590 defm UDIV_Int : Div<0, "udiv", int_arm64_udiv>;
591 defm SDIV_Int : Div<1, "sdiv", int_arm64_sdiv>;
595 defm ASRV : Shift<0b10, "asr", sra>;
596 defm LSLV : Shift<0b00, "lsl", shl>;
597 defm LSRV : Shift<0b01, "lsr", srl>;
598 defm RORV : Shift<0b11, "ror", rotr>;
600 def : ShiftAlias<"asrv", ASRVWr, GPR32>;
601 def : ShiftAlias<"asrv", ASRVXr, GPR64>;
602 def : ShiftAlias<"lslv", LSLVWr, GPR32>;
603 def : ShiftAlias<"lslv", LSLVXr, GPR64>;
604 def : ShiftAlias<"lsrv", LSRVWr, GPR32>;
605 def : ShiftAlias<"lsrv", LSRVXr, GPR64>;
606 def : ShiftAlias<"rorv", RORVWr, GPR32>;
607 def : ShiftAlias<"rorv", RORVXr, GPR64>;
610 let AddedComplexity = 7 in {
611 defm MADD : MulAccum<0, "madd", add>;
612 defm MSUB : MulAccum<1, "msub", sub>;
614 def : Pat<(i32 (mul GPR32:$Rn, GPR32:$Rm)),
615 (MADDWrrr GPR32:$Rn, GPR32:$Rm, WZR)>;
616 def : Pat<(i64 (mul GPR64:$Rn, GPR64:$Rm)),
617 (MADDXrrr GPR64:$Rn, GPR64:$Rm, XZR)>;
619 def : Pat<(i32 (ineg (mul GPR32:$Rn, GPR32:$Rm))),
620 (MSUBWrrr GPR32:$Rn, GPR32:$Rm, WZR)>;
621 def : Pat<(i64 (ineg (mul GPR64:$Rn, GPR64:$Rm))),
622 (MSUBXrrr GPR64:$Rn, GPR64:$Rm, XZR)>;
623 } // AddedComplexity = 7
625 let AddedComplexity = 5 in {
626 def SMADDLrrr : WideMulAccum<0, 0b001, "smaddl", add, sext>;
627 def SMSUBLrrr : WideMulAccum<1, 0b001, "smsubl", sub, sext>;
628 def UMADDLrrr : WideMulAccum<0, 0b101, "umaddl", add, zext>;
629 def UMSUBLrrr : WideMulAccum<1, 0b101, "umsubl", sub, zext>;
631 def : Pat<(i64 (mul (sext GPR32:$Rn), (sext GPR32:$Rm))),
632 (SMADDLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
633 def : Pat<(i64 (mul (zext GPR32:$Rn), (zext GPR32:$Rm))),
634 (UMADDLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
636 def : Pat<(i64 (ineg (mul (sext GPR32:$Rn), (sext GPR32:$Rm)))),
637 (SMSUBLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
638 def : Pat<(i64 (ineg (mul (zext GPR32:$Rn), (zext GPR32:$Rm)))),
639 (UMSUBLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
640 } // AddedComplexity = 5
642 def : MulAccumWAlias<"mul", MADDWrrr>;
643 def : MulAccumXAlias<"mul", MADDXrrr>;
644 def : MulAccumWAlias<"mneg", MSUBWrrr>;
645 def : MulAccumXAlias<"mneg", MSUBXrrr>;
646 def : WideMulAccumAlias<"smull", SMADDLrrr>;
647 def : WideMulAccumAlias<"smnegl", SMSUBLrrr>;
648 def : WideMulAccumAlias<"umull", UMADDLrrr>;
649 def : WideMulAccumAlias<"umnegl", UMSUBLrrr>;
652 def SMULHrr : MulHi<0b010, "smulh", mulhs>;
653 def UMULHrr : MulHi<0b110, "umulh", mulhu>;
656 def CRC32Brr : BaseCRC32<0, 0b00, 0, GPR32, int_arm64_crc32b, "crc32b">;
657 def CRC32Hrr : BaseCRC32<0, 0b01, 0, GPR32, int_arm64_crc32h, "crc32h">;
658 def CRC32Wrr : BaseCRC32<0, 0b10, 0, GPR32, int_arm64_crc32w, "crc32w">;
659 def CRC32Xrr : BaseCRC32<1, 0b11, 0, GPR64, int_arm64_crc32x, "crc32x">;
661 def CRC32CBrr : BaseCRC32<0, 0b00, 1, GPR32, int_arm64_crc32cb, "crc32cb">;
662 def CRC32CHrr : BaseCRC32<0, 0b01, 1, GPR32, int_arm64_crc32ch, "crc32ch">;
663 def CRC32CWrr : BaseCRC32<0, 0b10, 1, GPR32, int_arm64_crc32cw, "crc32cw">;
664 def CRC32CXrr : BaseCRC32<1, 0b11, 1, GPR64, int_arm64_crc32cx, "crc32cx">;
667 //===----------------------------------------------------------------------===//
668 // Logical instructions.
669 //===----------------------------------------------------------------------===//
672 defm ANDS : LogicalImmS<0b11, "ands", ARM64and_flag>;
673 defm AND : LogicalImm<0b00, "and", and>;
674 defm EOR : LogicalImm<0b10, "eor", xor>;
675 defm ORR : LogicalImm<0b01, "orr", or>;
677 // FIXME: these aliases *are* canonical sometimes (when movz can't be
678 // used). Actually, it seems to be working right now, but putting logical_immXX
679 // here is a bit dodgy on the AsmParser side too.
680 def : InstAlias<"mov $dst, $imm", (ORRWri GPR32sp:$dst, WZR,
681 logical_imm32:$imm), 0>;
682 def : InstAlias<"mov $dst, $imm", (ORRXri GPR64sp:$dst, XZR,
683 logical_imm64:$imm), 0>;
687 defm ANDS : LogicalRegS<0b11, 0, "ands", ARM64and_flag>;
688 defm BICS : LogicalRegS<0b11, 1, "bics",
689 BinOpFrag<(ARM64and_flag node:$LHS, (not node:$RHS))>>;
690 defm AND : LogicalReg<0b00, 0, "and", and>;
691 defm BIC : LogicalReg<0b00, 1, "bic",
692 BinOpFrag<(and node:$LHS, (not node:$RHS))>>;
693 defm EON : LogicalReg<0b10, 1, "eon",
694 BinOpFrag<(xor node:$LHS, (not node:$RHS))>>;
695 defm EOR : LogicalReg<0b10, 0, "eor", xor>;
696 defm ORN : LogicalReg<0b01, 1, "orn",
697 BinOpFrag<(or node:$LHS, (not node:$RHS))>>;
698 defm ORR : LogicalReg<0b01, 0, "orr", or>;
700 def : InstAlias<"mov $dst, $src", (ORRWrs GPR32:$dst, WZR, GPR32:$src, 0), 2>;
701 def : InstAlias<"mov $dst, $src", (ORRXrs GPR64:$dst, XZR, GPR64:$src, 0), 2>;
703 def : InstAlias<"mvn $Wd, $Wm", (ORNWrs GPR32:$Wd, WZR, GPR32:$Wm, 0), 3>;
704 def : InstAlias<"mvn $Xd, $Xm", (ORNXrs GPR64:$Xd, XZR, GPR64:$Xm, 0), 3>;
706 def : InstAlias<"mvn $Wd, $Wm$sh",
707 (ORNWrs GPR32:$Wd, WZR, GPR32:$Wm, logical_shift32:$sh), 2>;
708 def : InstAlias<"mvn $Xd, $Xm$sh",
709 (ORNXrs GPR64:$Xd, XZR, GPR64:$Xm, logical_shift64:$sh), 2>;
711 def : InstAlias<"tst $src1, $src2",
712 (ANDSWri WZR, GPR32:$src1, logical_imm32:$src2), 2>;
713 def : InstAlias<"tst $src1, $src2",
714 (ANDSXri XZR, GPR64:$src1, logical_imm64:$src2), 2>;
716 def : InstAlias<"tst $src1, $src2",
717 (ANDSWrs WZR, GPR32:$src1, GPR32:$src2, 0), 3>;
718 def : InstAlias<"tst $src1, $src2",
719 (ANDSXrs XZR, GPR64:$src1, GPR64:$src2, 0), 3>;
721 def : InstAlias<"tst $src1, $src2$sh",
722 (ANDSWrs WZR, GPR32:$src1, GPR32:$src2, logical_shift32:$sh), 2>;
723 def : InstAlias<"tst $src1, $src2$sh",
724 (ANDSXrs XZR, GPR64:$src1, GPR64:$src2, logical_shift64:$sh), 2>;
727 def : Pat<(not GPR32:$Wm), (ORNWrr WZR, GPR32:$Wm)>;
728 def : Pat<(not GPR64:$Xm), (ORNXrr XZR, GPR64:$Xm)>;
731 //===----------------------------------------------------------------------===//
732 // One operand data processing instructions.
733 //===----------------------------------------------------------------------===//
735 defm CLS : OneOperandData<0b101, "cls">;
736 defm CLZ : OneOperandData<0b100, "clz", ctlz>;
737 defm RBIT : OneOperandData<0b000, "rbit">;
738 def REV16Wr : OneWRegData<0b001, "rev16",
739 UnOpFrag<(rotr (bswap node:$LHS), (i64 16))>>;
740 def REV16Xr : OneXRegData<0b001, "rev16", null_frag>;
742 def : Pat<(cttz GPR32:$Rn),
743 (CLZWr (RBITWr GPR32:$Rn))>;
744 def : Pat<(cttz GPR64:$Rn),
745 (CLZXr (RBITXr GPR64:$Rn))>;
746 def : Pat<(ctlz (or (shl (xor (sra GPR32:$Rn, (i64 31)), GPR32:$Rn), (i64 1)),
749 def : Pat<(ctlz (or (shl (xor (sra GPR64:$Rn, (i64 63)), GPR64:$Rn), (i64 1)),
753 // Unlike the other one operand instructions, the instructions with the "rev"
754 // mnemonic do *not* just different in the size bit, but actually use different
755 // opcode bits for the different sizes.
756 def REVWr : OneWRegData<0b010, "rev", bswap>;
757 def REVXr : OneXRegData<0b011, "rev", bswap>;
758 def REV32Xr : OneXRegData<0b010, "rev32",
759 UnOpFrag<(rotr (bswap node:$LHS), (i64 32))>>;
761 // The bswap commutes with the rotr so we want a pattern for both possible
763 def : Pat<(bswap (rotr GPR32:$Rn, (i64 16))), (REV16Wr GPR32:$Rn)>;
764 def : Pat<(bswap (rotr GPR64:$Rn, (i64 32))), (REV32Xr GPR64:$Rn)>;
766 //===----------------------------------------------------------------------===//
767 // Bitfield immediate extraction instruction.
768 //===----------------------------------------------------------------------===//
769 let neverHasSideEffects = 1 in
770 defm EXTR : ExtractImm<"extr">;
771 def : InstAlias<"ror $dst, $src, $shift",
772 (EXTRWrri GPR32:$dst, GPR32:$src, GPR32:$src, imm0_31:$shift)>;
773 def : InstAlias<"ror $dst, $src, $shift",
774 (EXTRXrri GPR64:$dst, GPR64:$src, GPR64:$src, imm0_63:$shift)>;
776 def : Pat<(rotr GPR32:$Rn, (i64 imm0_31:$imm)),
777 (EXTRWrri GPR32:$Rn, GPR32:$Rn, imm0_31:$imm)>;
778 def : Pat<(rotr GPR64:$Rn, (i64 imm0_63:$imm)),
779 (EXTRXrri GPR64:$Rn, GPR64:$Rn, imm0_63:$imm)>;
781 //===----------------------------------------------------------------------===//
782 // Other bitfield immediate instructions.
783 //===----------------------------------------------------------------------===//
784 let neverHasSideEffects = 1 in {
785 defm BFM : BitfieldImmWith2RegArgs<0b01, "bfm">;
786 defm SBFM : BitfieldImm<0b00, "sbfm">;
787 defm UBFM : BitfieldImm<0b10, "ubfm">;
790 def i32shift_a : Operand<i64>, SDNodeXForm<imm, [{
791 uint64_t enc = (32 - N->getZExtValue()) & 0x1f;
792 return CurDAG->getTargetConstant(enc, MVT::i64);
795 def i32shift_b : Operand<i64>, SDNodeXForm<imm, [{
796 uint64_t enc = 31 - N->getZExtValue();
797 return CurDAG->getTargetConstant(enc, MVT::i64);
800 // min(7, 31 - shift_amt)
801 def i32shift_sext_i8 : Operand<i64>, SDNodeXForm<imm, [{
802 uint64_t enc = 31 - N->getZExtValue();
803 enc = enc > 7 ? 7 : enc;
804 return CurDAG->getTargetConstant(enc, MVT::i64);
807 // min(15, 31 - shift_amt)
808 def i32shift_sext_i16 : Operand<i64>, SDNodeXForm<imm, [{
809 uint64_t enc = 31 - N->getZExtValue();
810 enc = enc > 15 ? 15 : enc;
811 return CurDAG->getTargetConstant(enc, MVT::i64);
814 def i64shift_a : Operand<i64>, SDNodeXForm<imm, [{
815 uint64_t enc = (64 - N->getZExtValue()) & 0x3f;
816 return CurDAG->getTargetConstant(enc, MVT::i64);
819 def i64shift_b : Operand<i64>, SDNodeXForm<imm, [{
820 uint64_t enc = 63 - N->getZExtValue();
821 return CurDAG->getTargetConstant(enc, MVT::i64);
824 // min(7, 63 - shift_amt)
825 def i64shift_sext_i8 : Operand<i64>, SDNodeXForm<imm, [{
826 uint64_t enc = 63 - N->getZExtValue();
827 enc = enc > 7 ? 7 : enc;
828 return CurDAG->getTargetConstant(enc, MVT::i64);
831 // min(15, 63 - shift_amt)
832 def i64shift_sext_i16 : Operand<i64>, SDNodeXForm<imm, [{
833 uint64_t enc = 63 - N->getZExtValue();
834 enc = enc > 15 ? 15 : enc;
835 return CurDAG->getTargetConstant(enc, MVT::i64);
838 // min(31, 63 - shift_amt)
839 def i64shift_sext_i32 : Operand<i64>, SDNodeXForm<imm, [{
840 uint64_t enc = 63 - N->getZExtValue();
841 enc = enc > 31 ? 31 : enc;
842 return CurDAG->getTargetConstant(enc, MVT::i64);
845 def : Pat<(shl GPR32:$Rn, (i64 imm0_31:$imm)),
846 (UBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
847 (i64 (i32shift_b imm0_31:$imm)))>;
848 def : Pat<(shl GPR64:$Rn, (i64 imm0_63:$imm)),
849 (UBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
850 (i64 (i64shift_b imm0_63:$imm)))>;
852 let AddedComplexity = 10 in {
853 def : Pat<(sra GPR32:$Rn, (i64 imm0_31:$imm)),
854 (SBFMWri GPR32:$Rn, imm0_31:$imm, 31)>;
855 def : Pat<(sra GPR64:$Rn, (i64 imm0_63:$imm)),
856 (SBFMXri GPR64:$Rn, imm0_63:$imm, 63)>;
859 def : InstAlias<"asr $dst, $src, $shift",
860 (SBFMWri GPR32:$dst, GPR32:$src, imm0_31:$shift, 31)>;
861 def : InstAlias<"asr $dst, $src, $shift",
862 (SBFMXri GPR64:$dst, GPR64:$src, imm0_63:$shift, 63)>;
863 def : InstAlias<"sxtb $dst, $src", (SBFMWri GPR32:$dst, GPR32:$src, 0, 7)>;
864 def : InstAlias<"sxtb $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 7)>;
865 def : InstAlias<"sxth $dst, $src", (SBFMWri GPR32:$dst, GPR32:$src, 0, 15)>;
866 def : InstAlias<"sxth $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 15)>;
867 def : InstAlias<"sxtw $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 31)>;
869 def : Pat<(srl GPR32:$Rn, (i64 imm0_31:$imm)),
870 (UBFMWri GPR32:$Rn, imm0_31:$imm, 31)>;
871 def : Pat<(srl GPR64:$Rn, (i64 imm0_63:$imm)),
872 (UBFMXri GPR64:$Rn, imm0_63:$imm, 63)>;
874 def : InstAlias<"lsr $dst, $src, $shift",
875 (UBFMWri GPR32:$dst, GPR32:$src, imm0_31:$shift, 31)>;
876 def : InstAlias<"lsr $dst, $src, $shift",
877 (UBFMXri GPR64:$dst, GPR64:$src, imm0_63:$shift, 63)>;
878 def : InstAlias<"uxtb $dst, $src", (UBFMWri GPR32:$dst, GPR32:$src, 0, 7)>;
879 def : InstAlias<"uxtb $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 7)>;
880 def : InstAlias<"uxth $dst, $src", (UBFMWri GPR32:$dst, GPR32:$src, 0, 15)>;
881 def : InstAlias<"uxth $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 15)>;
882 def : InstAlias<"uxtw $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 31)>;
884 //===----------------------------------------------------------------------===//
885 // Conditionally set flags instructions.
886 //===----------------------------------------------------------------------===//
887 defm CCMN : CondSetFlagsImm<0, "ccmn">;
888 defm CCMP : CondSetFlagsImm<1, "ccmp">;
890 defm CCMN : CondSetFlagsReg<0, "ccmn">;
891 defm CCMP : CondSetFlagsReg<1, "ccmp">;
893 //===----------------------------------------------------------------------===//
894 // Conditional select instructions.
895 //===----------------------------------------------------------------------===//
896 defm CSEL : CondSelect<0, 0b00, "csel">;
898 def inc : PatFrag<(ops node:$in), (add node:$in, 1)>;
899 defm CSINC : CondSelectOp<0, 0b01, "csinc", inc>;
900 defm CSINV : CondSelectOp<1, 0b00, "csinv", not>;
901 defm CSNEG : CondSelectOp<1, 0b01, "csneg", ineg>;
903 def : Pat<(ARM64csinv GPR32:$tval, GPR32:$fval, (i32 imm:$cc), NZCV),
904 (CSINVWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
905 def : Pat<(ARM64csinv GPR64:$tval, GPR64:$fval, (i32 imm:$cc), NZCV),
906 (CSINVXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
907 def : Pat<(ARM64csneg GPR32:$tval, GPR32:$fval, (i32 imm:$cc), NZCV),
908 (CSNEGWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
909 def : Pat<(ARM64csneg GPR64:$tval, GPR64:$fval, (i32 imm:$cc), NZCV),
910 (CSNEGXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
911 def : Pat<(ARM64csinc GPR32:$tval, GPR32:$fval, (i32 imm:$cc), NZCV),
912 (CSINCWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
913 def : Pat<(ARM64csinc GPR64:$tval, GPR64:$fval, (i32 imm:$cc), NZCV),
914 (CSINCXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
916 def : Pat<(ARM64csel (i32 0), (i32 1), (i32 imm:$cc), NZCV),
917 (CSINCWr WZR, WZR, (i32 imm:$cc))>;
918 def : Pat<(ARM64csel (i64 0), (i64 1), (i32 imm:$cc), NZCV),
919 (CSINCXr XZR, XZR, (i32 imm:$cc))>;
920 def : Pat<(ARM64csel (i32 0), (i32 -1), (i32 imm:$cc), NZCV),
921 (CSINVWr WZR, WZR, (i32 imm:$cc))>;
922 def : Pat<(ARM64csel (i64 0), (i64 -1), (i32 imm:$cc), NZCV),
923 (CSINVXr XZR, XZR, (i32 imm:$cc))>;
925 // The inverse of the condition code from the alias instruction is what is used
926 // in the aliased instruction. The parser all ready inverts the condition code
927 // for these aliases.
928 def : InstAlias<"cset $dst, $cc",
929 (CSINCWr GPR32:$dst, WZR, WZR, inv_ccode:$cc)>;
930 def : InstAlias<"cset $dst, $cc",
931 (CSINCXr GPR64:$dst, XZR, XZR, inv_ccode:$cc)>;
933 def : InstAlias<"csetm $dst, $cc",
934 (CSINVWr GPR32:$dst, WZR, WZR, inv_ccode:$cc)>;
935 def : InstAlias<"csetm $dst, $cc",
936 (CSINVXr GPR64:$dst, XZR, XZR, inv_ccode:$cc)>;
938 def : InstAlias<"cinc $dst, $src, $cc",
939 (CSINCWr GPR32:$dst, GPR32:$src, GPR32:$src, inv_ccode:$cc)>;
940 def : InstAlias<"cinc $dst, $src, $cc",
941 (CSINCXr GPR64:$dst, GPR64:$src, GPR64:$src, inv_ccode:$cc)>;
943 def : InstAlias<"cinv $dst, $src, $cc",
944 (CSINVWr GPR32:$dst, GPR32:$src, GPR32:$src, inv_ccode:$cc)>;
945 def : InstAlias<"cinv $dst, $src, $cc",
946 (CSINVXr GPR64:$dst, GPR64:$src, GPR64:$src, inv_ccode:$cc)>;
948 def : InstAlias<"cneg $dst, $src, $cc",
949 (CSNEGWr GPR32:$dst, GPR32:$src, GPR32:$src, inv_ccode:$cc)>;
950 def : InstAlias<"cneg $dst, $src, $cc",
951 (CSNEGXr GPR64:$dst, GPR64:$src, GPR64:$src, inv_ccode:$cc)>;
953 //===----------------------------------------------------------------------===//
954 // PC-relative instructions.
955 //===----------------------------------------------------------------------===//
956 let isReMaterializable = 1 in {
957 let neverHasSideEffects = 1, mayStore = 0, mayLoad = 0 in {
958 def ADR : ADRI<0, "adr", adrlabel, []>;
959 } // neverHasSideEffects = 1
961 def ADRP : ADRI<1, "adrp", adrplabel,
962 [(set GPR64:$Xd, (ARM64adrp tglobaladdr:$label))]>;
963 } // isReMaterializable = 1
965 // page address of a constant pool entry, block address
966 def : Pat<(ARM64adrp tconstpool:$cp), (ADRP tconstpool:$cp)>;
967 def : Pat<(ARM64adrp tblockaddress:$cp), (ADRP tblockaddress:$cp)>;
969 //===----------------------------------------------------------------------===//
970 // Unconditional branch (register) instructions.
971 //===----------------------------------------------------------------------===//
973 let isReturn = 1, isTerminator = 1, isBarrier = 1 in {
974 def RET : BranchReg<0b0010, "ret", []>;
975 def DRPS : SpecialReturn<0b0101, "drps">;
976 def ERET : SpecialReturn<0b0100, "eret">;
977 } // isReturn = 1, isTerminator = 1, isBarrier = 1
979 // Default to the LR register.
980 def : InstAlias<"ret", (RET LR)>;
982 let isCall = 1, Defs = [LR], Uses = [SP] in {
983 def BLR : BranchReg<0b0001, "blr", [(ARM64call GPR64:$Rn)]>;
986 let isBranch = 1, isTerminator = 1, isBarrier = 1, isIndirectBranch = 1 in {
987 def BR : BranchReg<0b0000, "br", [(brind GPR64:$Rn)]>;
988 } // isBranch, isTerminator, isBarrier, isIndirectBranch
990 // Create a separate pseudo-instruction for codegen to use so that we don't
991 // flag lr as used in every function. It'll be restored before the RET by the
992 // epilogue if it's legitimately used.
993 def RET_ReallyLR : Pseudo<(outs), (ins), [(ARM64retflag)]> {
994 let isTerminator = 1;
999 // This is a directive-like pseudo-instruction. The purpose is to insert an
1000 // R_AARCH64_TLSDESC_CALL relocation at the offset of the following instruction
1001 // (which in the usual case is a BLR).
1002 let hasSideEffects = 1 in
1003 def TLSDESCCALL : Pseudo<(outs), (ins i64imm:$sym), []> {
1004 let AsmString = ".tlsdesccall $sym";
1007 // Pseudo-instruction representing a BLR with attached TLSDESC relocation. It
1008 // gets expanded to two MCInsts during lowering.
1009 let isCall = 1, Defs = [LR] in
1011 : Pseudo<(outs), (ins GPR64:$dest, i64imm:$sym),
1012 [(ARM64tlsdesc_call GPR64:$dest, tglobaltlsaddr:$sym)]>;
1014 def : Pat<(ARM64tlsdesc_call GPR64:$dest, texternalsym:$sym),
1015 (TLSDESC_BLR GPR64:$dest, texternalsym:$sym)>;
1016 //===----------------------------------------------------------------------===//
1017 // Conditional branch (immediate) instruction.
1018 //===----------------------------------------------------------------------===//
1019 def Bcc : BranchCond;
1021 //===----------------------------------------------------------------------===//
1022 // Compare-and-branch instructions.
1023 //===----------------------------------------------------------------------===//
1024 defm CBZ : CmpBranch<0, "cbz", ARM64cbz>;
1025 defm CBNZ : CmpBranch<1, "cbnz", ARM64cbnz>;
1027 //===----------------------------------------------------------------------===//
1028 // Test-bit-and-branch instructions.
1029 //===----------------------------------------------------------------------===//
1030 defm TBZ : TestBranch<0, "tbz", ARM64tbz>;
1031 defm TBNZ : TestBranch<1, "tbnz", ARM64tbnz>;
1033 //===----------------------------------------------------------------------===//
1034 // Unconditional branch (immediate) instructions.
1035 //===----------------------------------------------------------------------===//
1036 let isBranch = 1, isTerminator = 1, isBarrier = 1 in {
1037 def B : BranchImm<0, "b", [(br bb:$addr)]>;
1038 } // isBranch, isTerminator, isBarrier
1040 let isCall = 1, Defs = [LR], Uses = [SP] in {
1041 def BL : CallImm<1, "bl", [(ARM64call tglobaladdr:$addr)]>;
1043 def : Pat<(ARM64call texternalsym:$func), (BL texternalsym:$func)>;
1045 //===----------------------------------------------------------------------===//
1046 // Exception generation instructions.
1047 //===----------------------------------------------------------------------===//
1048 def BRK : ExceptionGeneration<0b001, 0b00, "brk">;
1049 def DCPS1 : ExceptionGeneration<0b101, 0b01, "dcps1">;
1050 def DCPS2 : ExceptionGeneration<0b101, 0b10, "dcps2">;
1051 def DCPS3 : ExceptionGeneration<0b101, 0b11, "dcps3">;
1052 def HLT : ExceptionGeneration<0b010, 0b00, "hlt">;
1053 def HVC : ExceptionGeneration<0b000, 0b10, "hvc">;
1054 def SMC : ExceptionGeneration<0b000, 0b11, "smc">;
1055 def SVC : ExceptionGeneration<0b000, 0b01, "svc">;
1057 // DCPSn defaults to an immediate operand of zero if unspecified.
1058 def : InstAlias<"dcps1", (DCPS1 0)>;
1059 def : InstAlias<"dcps2", (DCPS2 0)>;
1060 def : InstAlias<"dcps3", (DCPS3 0)>;
1062 //===----------------------------------------------------------------------===//
1063 // Load instructions.
1064 //===----------------------------------------------------------------------===//
1066 // Pair (indexed, offset)
1067 def LDPWi : LoadPairOffset<0b00, 0, GPR32, am_indexed32simm7, "ldp">;
1068 def LDPXi : LoadPairOffset<0b10, 0, GPR64, am_indexed64simm7, "ldp">;
1069 def LDPSi : LoadPairOffset<0b00, 1, FPR32, am_indexed32simm7, "ldp">;
1070 def LDPDi : LoadPairOffset<0b01, 1, FPR64, am_indexed64simm7, "ldp">;
1071 def LDPQi : LoadPairOffset<0b10, 1, FPR128, am_indexed128simm7, "ldp">;
1073 def LDPSWi : LoadPairOffset<0b01, 0, GPR64, am_indexed32simm7, "ldpsw">;
1075 // Pair (pre-indexed)
1076 def LDPWpre : LoadPairPreIdx<0b00, 0, GPR32, am_indexed32simm7_wb, "ldp">;
1077 def LDPXpre : LoadPairPreIdx<0b10, 0, GPR64, am_indexed64simm7_wb, "ldp">;
1078 def LDPSpre : LoadPairPreIdx<0b00, 1, FPR32, am_indexed32simm7_wb, "ldp">;
1079 def LDPDpre : LoadPairPreIdx<0b01, 1, FPR64, am_indexed64simm7_wb, "ldp">;
1080 def LDPQpre : LoadPairPreIdx<0b10, 1, FPR128, am_indexed128simm7_wb, "ldp">;
1082 def LDPSWpre : LoadPairPreIdx<0b01, 0, GPR64, am_indexed32simm7_wb, "ldpsw">;
1084 // Pair (post-indexed)
1085 def LDPWpost : LoadPairPostIdx<0b00, 0, GPR32, simm7s4, "ldp">;
1086 def LDPXpost : LoadPairPostIdx<0b10, 0, GPR64, simm7s8, "ldp">;
1087 def LDPSpost : LoadPairPostIdx<0b00, 1, FPR32, simm7s4, "ldp">;
1088 def LDPDpost : LoadPairPostIdx<0b01, 1, FPR64, simm7s8, "ldp">;
1089 def LDPQpost : LoadPairPostIdx<0b10, 1, FPR128, simm7s16, "ldp">;
1091 def LDPSWpost : LoadPairPostIdx<0b01, 0, GPR64, simm7s4, "ldpsw">;
1094 // Pair (no allocate)
1095 def LDNPWi : LoadPairNoAlloc<0b00, 0, GPR32, am_indexed32simm7, "ldnp">;
1096 def LDNPXi : LoadPairNoAlloc<0b10, 0, GPR64, am_indexed64simm7, "ldnp">;
1097 def LDNPSi : LoadPairNoAlloc<0b00, 1, FPR32, am_indexed32simm7, "ldnp">;
1098 def LDNPDi : LoadPairNoAlloc<0b01, 1, FPR64, am_indexed64simm7, "ldnp">;
1099 def LDNPQi : LoadPairNoAlloc<0b10, 1, FPR128, am_indexed128simm7, "ldnp">;
1102 // (register offset)
1105 let AddedComplexity = 10 in {
1107 def LDRBBro : Load8RO<0b00, 0, 0b01, GPR32, "ldrb",
1108 [(set GPR32:$Rt, (zextloadi8 ro_indexed8:$addr))]>;
1109 def LDRHHro : Load16RO<0b01, 0, 0b01, GPR32, "ldrh",
1110 [(set GPR32:$Rt, (zextloadi16 ro_indexed16:$addr))]>;
1111 def LDRWro : Load32RO<0b10, 0, 0b01, GPR32, "ldr",
1112 [(set GPR32:$Rt, (load ro_indexed32:$addr))]>;
1113 def LDRXro : Load64RO<0b11, 0, 0b01, GPR64, "ldr",
1114 [(set GPR64:$Rt, (load ro_indexed64:$addr))]>;
1117 def LDRBro : Load8RO<0b00, 1, 0b01, FPR8, "ldr",
1118 [(set FPR8:$Rt, (load ro_indexed8:$addr))]>;
1119 def LDRHro : Load16RO<0b01, 1, 0b01, FPR16, "ldr",
1120 [(set (f16 FPR16:$Rt), (load ro_indexed16:$addr))]>;
1121 def LDRSro : Load32RO<0b10, 1, 0b01, FPR32, "ldr",
1122 [(set (f32 FPR32:$Rt), (load ro_indexed32:$addr))]>;
1123 def LDRDro : Load64RO<0b11, 1, 0b01, FPR64, "ldr",
1124 [(set (f64 FPR64:$Rt), (load ro_indexed64:$addr))]>;
1125 def LDRQro : Load128RO<0b00, 1, 0b11, FPR128, "ldr", []> {
1129 // For regular load, we do not have any alignment requirement.
1130 // Thus, it is safe to directly map the vector loads with interesting
1131 // addressing modes.
1132 // FIXME: We could do the same for bitconvert to floating point vectors.
1133 def : Pat <(v8i8 (scalar_to_vector (i32 (extloadi8 ro_indexed8:$addr)))),
1134 (INSERT_SUBREG (v8i8 (IMPLICIT_DEF)),
1135 (LDRBro ro_indexed8:$addr), bsub)>;
1136 def : Pat <(v16i8 (scalar_to_vector (i32 (extloadi8 ro_indexed8:$addr)))),
1137 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
1138 (LDRBro ro_indexed8:$addr), bsub)>;
1139 def : Pat <(v4i16 (scalar_to_vector (i32 (extloadi16 ro_indexed16:$addr)))),
1140 (INSERT_SUBREG (v4i16 (IMPLICIT_DEF)),
1141 (LDRHro ro_indexed16:$addr), hsub)>;
1142 def : Pat <(v8i16 (scalar_to_vector (i32 (extloadi16 ro_indexed16:$addr)))),
1143 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
1144 (LDRHro ro_indexed16:$addr), hsub)>;
1145 def : Pat <(v2i32 (scalar_to_vector (i32 (load ro_indexed32:$addr)))),
1146 (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)),
1147 (LDRSro ro_indexed32:$addr), ssub)>;
1148 def : Pat <(v4i32 (scalar_to_vector (i32 (load ro_indexed32:$addr)))),
1149 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
1150 (LDRSro ro_indexed32:$addr), ssub)>;
1151 def : Pat <(v1i64 (scalar_to_vector (i64 (load ro_indexed64:$addr)))),
1152 (LDRDro ro_indexed64:$addr)>;
1153 def : Pat <(v2i64 (scalar_to_vector (i64 (load ro_indexed64:$addr)))),
1154 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
1155 (LDRDro ro_indexed64:$addr), dsub)>;
1157 // Match all load 64 bits width whose type is compatible with FPR64
1158 let Predicates = [IsLE] in {
1159 // We must do vector loads with LD1 in big-endian.
1160 def : Pat<(v2f32 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1161 def : Pat<(v8i8 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1162 def : Pat<(v4i16 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1163 def : Pat<(v2i32 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1165 def : Pat<(v1f64 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1166 def : Pat<(v1i64 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1168 // Match all load 128 bits width whose type is compatible with FPR128
1169 let Predicates = [IsLE] in {
1170 // We must do vector loads with LD1 in big-endian.
1171 def : Pat<(v4f32 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1172 def : Pat<(v2f64 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1173 def : Pat<(v16i8 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1174 def : Pat<(v8i16 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1175 def : Pat<(v4i32 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1176 def : Pat<(v2i64 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1178 def : Pat<(f128 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1180 // Load sign-extended half-word
1181 def LDRSHWro : Load16RO<0b01, 0, 0b11, GPR32, "ldrsh",
1182 [(set GPR32:$Rt, (sextloadi16 ro_indexed16:$addr))]>;
1183 def LDRSHXro : Load16RO<0b01, 0, 0b10, GPR64, "ldrsh",
1184 [(set GPR64:$Rt, (sextloadi16 ro_indexed16:$addr))]>;
1186 // Load sign-extended byte
1187 def LDRSBWro : Load8RO<0b00, 0, 0b11, GPR32, "ldrsb",
1188 [(set GPR32:$Rt, (sextloadi8 ro_indexed8:$addr))]>;
1189 def LDRSBXro : Load8RO<0b00, 0, 0b10, GPR64, "ldrsb",
1190 [(set GPR64:$Rt, (sextloadi8 ro_indexed8:$addr))]>;
1192 // Load sign-extended word
1193 def LDRSWro : Load32RO<0b10, 0, 0b10, GPR64, "ldrsw",
1194 [(set GPR64:$Rt, (sextloadi32 ro_indexed32:$addr))]>;
1197 def PRFMro : PrefetchRO<0b11, 0, 0b10, "prfm",
1198 [(ARM64Prefetch imm:$Rt, ro_indexed64:$addr)]>;
1201 def : Pat<(i64 (zextloadi8 ro_indexed8:$addr)),
1202 (SUBREG_TO_REG (i64 0), (LDRBBro ro_indexed8:$addr), sub_32)>;
1203 def : Pat<(i64 (zextloadi16 ro_indexed16:$addr)),
1204 (SUBREG_TO_REG (i64 0), (LDRHHro ro_indexed16:$addr), sub_32)>;
1205 def : Pat<(i64 (zextloadi32 ro_indexed32:$addr)),
1206 (SUBREG_TO_REG (i64 0), (LDRWro ro_indexed32:$addr), sub_32)>;
1208 // zextloadi1 -> zextloadi8
1209 def : Pat<(i32 (zextloadi1 ro_indexed8:$addr)), (LDRBBro ro_indexed8:$addr)>;
1210 def : Pat<(i64 (zextloadi1 ro_indexed8:$addr)),
1211 (SUBREG_TO_REG (i64 0), (LDRBBro ro_indexed8:$addr), sub_32)>;
1213 // extload -> zextload
1214 def : Pat<(i32 (extloadi16 ro_indexed16:$addr)), (LDRHHro ro_indexed16:$addr)>;
1215 def : Pat<(i32 (extloadi8 ro_indexed8:$addr)), (LDRBBro ro_indexed8:$addr)>;
1216 def : Pat<(i32 (extloadi1 ro_indexed8:$addr)), (LDRBBro ro_indexed8:$addr)>;
1217 def : Pat<(i64 (extloadi32 ro_indexed32:$addr)),
1218 (SUBREG_TO_REG (i64 0), (LDRWro ro_indexed32:$addr), sub_32)>;
1219 def : Pat<(i64 (extloadi16 ro_indexed16:$addr)),
1220 (SUBREG_TO_REG (i64 0), (LDRHHro ro_indexed16:$addr), sub_32)>;
1221 def : Pat<(i64 (extloadi8 ro_indexed8:$addr)),
1222 (SUBREG_TO_REG (i64 0), (LDRBBro ro_indexed8:$addr), sub_32)>;
1223 def : Pat<(i64 (extloadi1 ro_indexed8:$addr)),
1224 (SUBREG_TO_REG (i64 0), (LDRBBro ro_indexed8:$addr), sub_32)>;
1226 } // AddedComplexity = 10
1229 // (unsigned immediate)
1231 def LDRXui : LoadUI<0b11, 0, 0b01, GPR64, am_indexed64, "ldr",
1232 [(set GPR64:$Rt, (load am_indexed64:$addr))]>;
1233 def LDRWui : LoadUI<0b10, 0, 0b01, GPR32, am_indexed32, "ldr",
1234 [(set GPR32:$Rt, (load am_indexed32:$addr))]>;
1235 def LDRBui : LoadUI<0b00, 1, 0b01, FPR8, am_indexed8, "ldr",
1236 [(set FPR8:$Rt, (load am_indexed8:$addr))]>;
1237 def LDRHui : LoadUI<0b01, 1, 0b01, FPR16, am_indexed16, "ldr",
1238 [(set (f16 FPR16:$Rt), (load am_indexed16:$addr))]>;
1239 def LDRSui : LoadUI<0b10, 1, 0b01, FPR32, am_indexed32, "ldr",
1240 [(set (f32 FPR32:$Rt), (load am_indexed32:$addr))]>;
1241 def LDRDui : LoadUI<0b11, 1, 0b01, FPR64, am_indexed64, "ldr",
1242 [(set (f64 FPR64:$Rt), (load am_indexed64:$addr))]>;
1243 def LDRQui : LoadUI<0b00, 1, 0b11, FPR128, am_indexed128, "ldr",
1244 [(set (f128 FPR128:$Rt), (load am_indexed128:$addr))]>;
1246 // For regular load, we do not have any alignment requirement.
1247 // Thus, it is safe to directly map the vector loads with interesting
1248 // addressing modes.
1249 // FIXME: We could do the same for bitconvert to floating point vectors.
1250 def : Pat <(v8i8 (scalar_to_vector (i32 (extloadi8 am_indexed8:$addr)))),
1251 (INSERT_SUBREG (v8i8 (IMPLICIT_DEF)),
1252 (LDRBui am_indexed8:$addr), bsub)>;
1253 def : Pat <(v16i8 (scalar_to_vector (i32 (extloadi8 am_indexed8:$addr)))),
1254 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
1255 (LDRBui am_indexed8:$addr), bsub)>;
1256 def : Pat <(v4i16 (scalar_to_vector (i32 (extloadi16 am_indexed16:$addr)))),
1257 (INSERT_SUBREG (v4i16 (IMPLICIT_DEF)),
1258 (LDRHui am_indexed16:$addr), hsub)>;
1259 def : Pat <(v8i16 (scalar_to_vector (i32 (extloadi16 am_indexed16:$addr)))),
1260 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
1261 (LDRHui am_indexed16:$addr), hsub)>;
1262 def : Pat <(v2i32 (scalar_to_vector (i32 (load am_indexed32:$addr)))),
1263 (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)),
1264 (LDRSui am_indexed32:$addr), ssub)>;
1265 def : Pat <(v4i32 (scalar_to_vector (i32 (load am_indexed32:$addr)))),
1266 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
1267 (LDRSui am_indexed32:$addr), ssub)>;
1268 def : Pat <(v1i64 (scalar_to_vector (i64 (load am_indexed64:$addr)))),
1269 (LDRDui am_indexed64:$addr)>;
1270 def : Pat <(v2i64 (scalar_to_vector (i64 (load am_indexed64:$addr)))),
1271 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
1272 (LDRDui am_indexed64:$addr), dsub)>;
1274 // Match all load 64 bits width whose type is compatible with FPR64
1275 let Predicates = [IsLE] in {
1276 // We must use LD1 to perform vector loads in big-endian.
1277 def : Pat<(v2f32 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1278 def : Pat<(v8i8 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1279 def : Pat<(v4i16 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1280 def : Pat<(v2i32 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1282 def : Pat<(v1f64 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1283 def : Pat<(v1i64 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1285 // Match all load 128 bits width whose type is compatible with FPR128
1286 let Predicates = [IsLE] in {
1287 // We must use LD1 to perform vector loads in big-endian.
1288 def : Pat<(v4f32 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1289 def : Pat<(v2f64 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1290 def : Pat<(v16i8 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1291 def : Pat<(v8i16 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1292 def : Pat<(v4i32 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1293 def : Pat<(v2i64 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1295 def : Pat<(f128 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1297 def LDRHHui : LoadUI<0b01, 0, 0b01, GPR32, am_indexed16, "ldrh",
1298 [(set GPR32:$Rt, (zextloadi16 am_indexed16:$addr))]>;
1299 def LDRBBui : LoadUI<0b00, 0, 0b01, GPR32, am_indexed8, "ldrb",
1300 [(set GPR32:$Rt, (zextloadi8 am_indexed8:$addr))]>;
1302 def : Pat<(i64 (zextloadi8 am_indexed8:$addr)),
1303 (SUBREG_TO_REG (i64 0), (LDRBBui am_indexed8:$addr), sub_32)>;
1304 def : Pat<(i64 (zextloadi16 am_indexed16:$addr)),
1305 (SUBREG_TO_REG (i64 0), (LDRHHui am_indexed16:$addr), sub_32)>;
1307 // zextloadi1 -> zextloadi8
1308 def : Pat<(i32 (zextloadi1 am_indexed8:$addr)), (LDRBBui am_indexed8:$addr)>;
1309 def : Pat<(i64 (zextloadi1 am_indexed8:$addr)),
1310 (SUBREG_TO_REG (i64 0), (LDRBBui am_indexed8:$addr), sub_32)>;
1312 // extload -> zextload
1313 def : Pat<(i32 (extloadi16 am_indexed16:$addr)), (LDRHHui am_indexed16:$addr)>;
1314 def : Pat<(i32 (extloadi8 am_indexed8:$addr)), (LDRBBui am_indexed8:$addr)>;
1315 def : Pat<(i32 (extloadi1 am_indexed8:$addr)), (LDRBBui am_indexed8:$addr)>;
1316 def : Pat<(i64 (extloadi32 am_indexed32:$addr)),
1317 (SUBREG_TO_REG (i64 0), (LDRWui am_indexed32:$addr), sub_32)>;
1318 def : Pat<(i64 (extloadi16 am_indexed16:$addr)),
1319 (SUBREG_TO_REG (i64 0), (LDRHHui am_indexed16:$addr), sub_32)>;
1320 def : Pat<(i64 (extloadi8 am_indexed8:$addr)),
1321 (SUBREG_TO_REG (i64 0), (LDRBBui am_indexed8:$addr), sub_32)>;
1322 def : Pat<(i64 (extloadi1 am_indexed8:$addr)),
1323 (SUBREG_TO_REG (i64 0), (LDRBBui am_indexed8:$addr), sub_32)>;
1325 // load sign-extended half-word
1326 def LDRSHWui : LoadUI<0b01, 0, 0b11, GPR32, am_indexed16, "ldrsh",
1327 [(set GPR32:$Rt, (sextloadi16 am_indexed16:$addr))]>;
1328 def LDRSHXui : LoadUI<0b01, 0, 0b10, GPR64, am_indexed16, "ldrsh",
1329 [(set GPR64:$Rt, (sextloadi16 am_indexed16:$addr))]>;
1331 // load sign-extended byte
1332 def LDRSBWui : LoadUI<0b00, 0, 0b11, GPR32, am_indexed8, "ldrsb",
1333 [(set GPR32:$Rt, (sextloadi8 am_indexed8:$addr))]>;
1334 def LDRSBXui : LoadUI<0b00, 0, 0b10, GPR64, am_indexed8, "ldrsb",
1335 [(set GPR64:$Rt, (sextloadi8 am_indexed8:$addr))]>;
1337 // load sign-extended word
1338 def LDRSWui : LoadUI<0b10, 0, 0b10, GPR64, am_indexed32, "ldrsw",
1339 [(set GPR64:$Rt, (sextloadi32 am_indexed32:$addr))]>;
1341 // load zero-extended word
1342 def : Pat<(i64 (zextloadi32 am_indexed32:$addr)),
1343 (SUBREG_TO_REG (i64 0), (LDRWui am_indexed32:$addr), sub_32)>;
1346 def PRFMui : PrefetchUI<0b11, 0, 0b10, "prfm",
1347 [(ARM64Prefetch imm:$Rt, am_indexed64:$addr)]>;
1351 def LDRWl : LoadLiteral<0b00, 0, GPR32, "ldr">;
1352 def LDRXl : LoadLiteral<0b01, 0, GPR64, "ldr">;
1353 def LDRSl : LoadLiteral<0b00, 1, FPR32, "ldr">;
1354 def LDRDl : LoadLiteral<0b01, 1, FPR64, "ldr">;
1355 def LDRQl : LoadLiteral<0b10, 1, FPR128, "ldr">;
1357 // load sign-extended word
1358 def LDRSWl : LoadLiteral<0b10, 0, GPR64, "ldrsw">;
1361 def PRFMl : PrefetchLiteral<0b11, 0, "prfm", []>;
1362 // [(ARM64Prefetch imm:$Rt, tglobaladdr:$label)]>;
1365 // (unscaled immediate)
1366 def LDURXi : LoadUnscaled<0b11, 0, 0b01, GPR64, am_unscaled64, "ldur",
1367 [(set GPR64:$Rt, (load am_unscaled64:$addr))]>;
1368 def LDURWi : LoadUnscaled<0b10, 0, 0b01, GPR32, am_unscaled32, "ldur",
1369 [(set GPR32:$Rt, (load am_unscaled32:$addr))]>;
1370 def LDURBi : LoadUnscaled<0b00, 1, 0b01, FPR8, am_unscaled8, "ldur",
1371 [(set FPR8:$Rt, (load am_unscaled8:$addr))]>;
1372 def LDURHi : LoadUnscaled<0b01, 1, 0b01, FPR16, am_unscaled16, "ldur",
1373 [(set (f16 FPR16:$Rt), (load am_unscaled16:$addr))]>;
1374 def LDURSi : LoadUnscaled<0b10, 1, 0b01, FPR32, am_unscaled32, "ldur",
1375 [(set (f32 FPR32:$Rt), (load am_unscaled32:$addr))]>;
1376 def LDURDi : LoadUnscaled<0b11, 1, 0b01, FPR64, am_unscaled64, "ldur",
1377 [(set (f64 FPR64:$Rt), (load am_unscaled64:$addr))]>;
1378 def LDURQi : LoadUnscaled<0b00, 1, 0b11, FPR128, am_unscaled128, "ldur",
1379 [(set (f128 FPR128:$Rt), (load am_unscaled128:$addr))]>;
1382 : LoadUnscaled<0b01, 0, 0b01, GPR32, am_unscaled16, "ldurh",
1383 [(set GPR32:$Rt, (zextloadi16 am_unscaled16:$addr))]>;
1385 : LoadUnscaled<0b00, 0, 0b01, GPR32, am_unscaled8, "ldurb",
1386 [(set GPR32:$Rt, (zextloadi8 am_unscaled8:$addr))]>;
1388 // Match all load 64 bits width whose type is compatible with FPR64
1389 let Predicates = [IsLE] in {
1390 def : Pat<(v2f32 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1391 def : Pat<(v8i8 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1392 def : Pat<(v4i16 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1393 def : Pat<(v2i32 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1395 def : Pat<(v1f64 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1396 def : Pat<(v1i64 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1398 // Match all load 128 bits width whose type is compatible with FPR128
1399 let Predicates = [IsLE] in {
1400 def : Pat<(v4f32 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1401 def : Pat<(v2f64 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1402 def : Pat<(v16i8 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1403 def : Pat<(v8i16 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1404 def : Pat<(v4i32 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1405 def : Pat<(v2i64 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1406 def : Pat<(v2f64 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1410 def : Pat<(i32 (extloadi16 am_unscaled16:$addr)), (LDURHHi am_unscaled16:$addr)>;
1411 def : Pat<(i32 (extloadi8 am_unscaled8:$addr)), (LDURBBi am_unscaled8:$addr)>;
1412 def : Pat<(i32 (extloadi1 am_unscaled8:$addr)), (LDURBBi am_unscaled8:$addr)>;
1413 def : Pat<(i64 (extloadi32 am_unscaled32:$addr)),
1414 (SUBREG_TO_REG (i64 0), (LDURWi am_unscaled32:$addr), sub_32)>;
1415 def : Pat<(i64 (extloadi16 am_unscaled16:$addr)),
1416 (SUBREG_TO_REG (i64 0), (LDURHHi am_unscaled16:$addr), sub_32)>;
1417 def : Pat<(i64 (extloadi8 am_unscaled8:$addr)),
1418 (SUBREG_TO_REG (i64 0), (LDURBBi am_unscaled8:$addr), sub_32)>;
1419 def : Pat<(i64 (extloadi1 am_unscaled8:$addr)),
1420 (SUBREG_TO_REG (i64 0), (LDURBBi am_unscaled8:$addr), sub_32)>;
1422 def : Pat<(i32 (zextloadi16 am_unscaled16:$addr)),
1423 (LDURHHi am_unscaled16:$addr)>;
1424 def : Pat<(i32 (zextloadi8 am_unscaled8:$addr)),
1425 (LDURBBi am_unscaled8:$addr)>;
1426 def : Pat<(i32 (zextloadi1 am_unscaled8:$addr)),
1427 (LDURBBi am_unscaled8:$addr)>;
1428 def : Pat<(i64 (zextloadi32 am_unscaled32:$addr)),
1429 (SUBREG_TO_REG (i64 0), (LDURWi am_unscaled32:$addr), sub_32)>;
1430 def : Pat<(i64 (zextloadi16 am_unscaled16:$addr)),
1431 (SUBREG_TO_REG (i64 0), (LDURHHi am_unscaled16:$addr), sub_32)>;
1432 def : Pat<(i64 (zextloadi8 am_unscaled8:$addr)),
1433 (SUBREG_TO_REG (i64 0), (LDURBBi am_unscaled8:$addr), sub_32)>;
1434 def : Pat<(i64 (zextloadi1 am_unscaled8:$addr)),
1435 (SUBREG_TO_REG (i64 0), (LDURBBi am_unscaled8:$addr), sub_32)>;
1439 // LDR mnemonics fall back to LDUR for negative or unaligned offsets.
1441 // Define new assembler match classes as we want to only match these when
1442 // the don't otherwise match the scaled addressing mode for LDR/STR. Don't
1443 // associate a DiagnosticType either, as we want the diagnostic for the
1444 // canonical form (the scaled operand) to take precedence.
1445 def MemoryUnscaledFB8Operand : AsmOperandClass {
1446 let Name = "MemoryUnscaledFB8";
1447 let RenderMethod = "addMemoryUnscaledOperands";
1449 def MemoryUnscaledFB16Operand : AsmOperandClass {
1450 let Name = "MemoryUnscaledFB16";
1451 let RenderMethod = "addMemoryUnscaledOperands";
1453 def MemoryUnscaledFB32Operand : AsmOperandClass {
1454 let Name = "MemoryUnscaledFB32";
1455 let RenderMethod = "addMemoryUnscaledOperands";
1457 def MemoryUnscaledFB64Operand : AsmOperandClass {
1458 let Name = "MemoryUnscaledFB64";
1459 let RenderMethod = "addMemoryUnscaledOperands";
1461 def MemoryUnscaledFB128Operand : AsmOperandClass {
1462 let Name = "MemoryUnscaledFB128";
1463 let RenderMethod = "addMemoryUnscaledOperands";
1465 def am_unscaled_fb8 : Operand<i64> {
1466 let ParserMatchClass = MemoryUnscaledFB8Operand;
1467 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
1469 def am_unscaled_fb16 : Operand<i64> {
1470 let ParserMatchClass = MemoryUnscaledFB16Operand;
1471 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
1473 def am_unscaled_fb32 : Operand<i64> {
1474 let ParserMatchClass = MemoryUnscaledFB32Operand;
1475 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
1477 def am_unscaled_fb64 : Operand<i64> {
1478 let ParserMatchClass = MemoryUnscaledFB64Operand;
1479 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
1481 def am_unscaled_fb128 : Operand<i64> {
1482 let ParserMatchClass = MemoryUnscaledFB128Operand;
1483 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
1485 def : InstAlias<"ldr $Rt, $addr", (LDURXi GPR64:$Rt, am_unscaled_fb64:$addr), 0>;
1486 def : InstAlias<"ldr $Rt, $addr", (LDURWi GPR32:$Rt, am_unscaled_fb32:$addr), 0>;
1487 def : InstAlias<"ldr $Rt, $addr", (LDURBi FPR8:$Rt, am_unscaled_fb8:$addr), 0>;
1488 def : InstAlias<"ldr $Rt, $addr", (LDURHi FPR16:$Rt, am_unscaled_fb16:$addr), 0>;
1489 def : InstAlias<"ldr $Rt, $addr", (LDURSi FPR32:$Rt, am_unscaled_fb32:$addr), 0>;
1490 def : InstAlias<"ldr $Rt, $addr", (LDURDi FPR64:$Rt, am_unscaled_fb64:$addr), 0>;
1491 def : InstAlias<"ldr $Rt, $addr", (LDURQi FPR128:$Rt, am_unscaled_fb128:$addr), 0>;
1494 def : Pat<(i64 (zextloadi8 am_unscaled8:$addr)),
1495 (SUBREG_TO_REG (i64 0), (LDURBBi am_unscaled8:$addr), sub_32)>;
1496 def : Pat<(i64 (zextloadi16 am_unscaled16:$addr)),
1497 (SUBREG_TO_REG (i64 0), (LDURHHi am_unscaled16:$addr), sub_32)>;
1499 // load sign-extended half-word
1501 : LoadUnscaled<0b01, 0, 0b11, GPR32, am_unscaled16, "ldursh",
1502 [(set GPR32:$Rt, (sextloadi16 am_unscaled16:$addr))]>;
1504 : LoadUnscaled<0b01, 0, 0b10, GPR64, am_unscaled16, "ldursh",
1505 [(set GPR64:$Rt, (sextloadi16 am_unscaled16:$addr))]>;
1507 // load sign-extended byte
1509 : LoadUnscaled<0b00, 0, 0b11, GPR32, am_unscaled8, "ldursb",
1510 [(set GPR32:$Rt, (sextloadi8 am_unscaled8:$addr))]>;
1512 : LoadUnscaled<0b00, 0, 0b10, GPR64, am_unscaled8, "ldursb",
1513 [(set GPR64:$Rt, (sextloadi8 am_unscaled8:$addr))]>;
1515 // load sign-extended word
1517 : LoadUnscaled<0b10, 0, 0b10, GPR64, am_unscaled32, "ldursw",
1518 [(set GPR64:$Rt, (sextloadi32 am_unscaled32:$addr))]>;
1520 // zero and sign extending aliases from generic LDR* mnemonics to LDUR*.
1521 def : InstAlias<"ldrb $Rt, $addr",
1522 (LDURBBi GPR32:$Rt, am_unscaled_fb8:$addr), 0>;
1523 def : InstAlias<"ldrh $Rt, $addr",
1524 (LDURHHi GPR32:$Rt, am_unscaled_fb16:$addr), 0>;
1525 def : InstAlias<"ldrsb $Rt, $addr",
1526 (LDURSBWi GPR32:$Rt, am_unscaled_fb8:$addr), 0>;
1527 def : InstAlias<"ldrsb $Rt, $addr",
1528 (LDURSBXi GPR64:$Rt, am_unscaled_fb8:$addr), 0>;
1529 def : InstAlias<"ldrsh $Rt, $addr",
1530 (LDURSHWi GPR32:$Rt, am_unscaled_fb16:$addr), 0>;
1531 def : InstAlias<"ldrsh $Rt, $addr",
1532 (LDURSHXi GPR64:$Rt, am_unscaled_fb16:$addr), 0>;
1533 def : InstAlias<"ldrsw $Rt, $addr",
1534 (LDURSWi GPR64:$Rt, am_unscaled_fb32:$addr), 0>;
1537 def PRFUMi : PrefetchUnscaled<0b11, 0, 0b10, "prfum",
1538 [(ARM64Prefetch imm:$Rt, am_unscaled64:$addr)]>;
1541 // (unscaled immediate, unprivileged)
1542 def LDTRXi : LoadUnprivileged<0b11, 0, 0b01, GPR64, "ldtr">;
1543 def LDTRWi : LoadUnprivileged<0b10, 0, 0b01, GPR32, "ldtr">;
1545 def LDTRHi : LoadUnprivileged<0b01, 0, 0b01, GPR32, "ldtrh">;
1546 def LDTRBi : LoadUnprivileged<0b00, 0, 0b01, GPR32, "ldtrb">;
1548 // load sign-extended half-word
1549 def LDTRSHWi : LoadUnprivileged<0b01, 0, 0b11, GPR32, "ldtrsh">;
1550 def LDTRSHXi : LoadUnprivileged<0b01, 0, 0b10, GPR64, "ldtrsh">;
1552 // load sign-extended byte
1553 def LDTRSBWi : LoadUnprivileged<0b00, 0, 0b11, GPR32, "ldtrsb">;
1554 def LDTRSBXi : LoadUnprivileged<0b00, 0, 0b10, GPR64, "ldtrsb">;
1556 // load sign-extended word
1557 def LDTRSWi : LoadUnprivileged<0b10, 0, 0b10, GPR64, "ldtrsw">;
1560 // (immediate pre-indexed)
1561 def LDRWpre : LoadPreIdx<0b10, 0, 0b01, GPR32, "ldr">;
1562 def LDRXpre : LoadPreIdx<0b11, 0, 0b01, GPR64, "ldr">;
1563 def LDRBpre : LoadPreIdx<0b00, 1, 0b01, FPR8, "ldr">;
1564 def LDRHpre : LoadPreIdx<0b01, 1, 0b01, FPR16, "ldr">;
1565 def LDRSpre : LoadPreIdx<0b10, 1, 0b01, FPR32, "ldr">;
1566 def LDRDpre : LoadPreIdx<0b11, 1, 0b01, FPR64, "ldr">;
1567 def LDRQpre : LoadPreIdx<0b00, 1, 0b11, FPR128, "ldr">;
1569 // load sign-extended half-word
1570 def LDRSHWpre : LoadPreIdx<0b01, 0, 0b11, GPR32, "ldrsh">;
1571 def LDRSHXpre : LoadPreIdx<0b01, 0, 0b10, GPR64, "ldrsh">;
1573 // load sign-extended byte
1574 def LDRSBWpre : LoadPreIdx<0b00, 0, 0b11, GPR32, "ldrsb">;
1575 def LDRSBXpre : LoadPreIdx<0b00, 0, 0b10, GPR64, "ldrsb">;
1577 // load zero-extended byte
1578 def LDRBBpre : LoadPreIdx<0b00, 0, 0b01, GPR32, "ldrb">;
1579 def LDRHHpre : LoadPreIdx<0b01, 0, 0b01, GPR32, "ldrh">;
1581 // load sign-extended word
1582 def LDRSWpre : LoadPreIdx<0b10, 0, 0b10, GPR64, "ldrsw">;
1584 // ISel pseudos and patterns. See expanded comment on LoadPreIdxPseudo.
1585 def LDRQpre_isel : LoadPreIdxPseudo<FPR128>;
1586 def LDRDpre_isel : LoadPreIdxPseudo<FPR64>;
1587 def LDRSpre_isel : LoadPreIdxPseudo<FPR32>;
1588 def LDRXpre_isel : LoadPreIdxPseudo<GPR64>;
1589 def LDRWpre_isel : LoadPreIdxPseudo<GPR32>;
1590 def LDRHHpre_isel : LoadPreIdxPseudo<GPR32>;
1591 def LDRBBpre_isel : LoadPreIdxPseudo<GPR32>;
1593 def LDRSWpre_isel : LoadPreIdxPseudo<GPR64>;
1594 def LDRSHWpre_isel : LoadPreIdxPseudo<GPR32>;
1595 def LDRSHXpre_isel : LoadPreIdxPseudo<GPR64>;
1596 def LDRSBWpre_isel : LoadPreIdxPseudo<GPR32>;
1597 def LDRSBXpre_isel : LoadPreIdxPseudo<GPR64>;
1600 // (immediate post-indexed)
1601 def LDRWpost : LoadPostIdx<0b10, 0, 0b01, GPR32, "ldr">;
1602 def LDRXpost : LoadPostIdx<0b11, 0, 0b01, GPR64, "ldr">;
1603 def LDRBpost : LoadPostIdx<0b00, 1, 0b01, FPR8, "ldr">;
1604 def LDRHpost : LoadPostIdx<0b01, 1, 0b01, FPR16, "ldr">;
1605 def LDRSpost : LoadPostIdx<0b10, 1, 0b01, FPR32, "ldr">;
1606 def LDRDpost : LoadPostIdx<0b11, 1, 0b01, FPR64, "ldr">;
1607 def LDRQpost : LoadPostIdx<0b00, 1, 0b11, FPR128, "ldr">;
1609 // load sign-extended half-word
1610 def LDRSHWpost : LoadPostIdx<0b01, 0, 0b11, GPR32, "ldrsh">;
1611 def LDRSHXpost : LoadPostIdx<0b01, 0, 0b10, GPR64, "ldrsh">;
1613 // load sign-extended byte
1614 def LDRSBWpost : LoadPostIdx<0b00, 0, 0b11, GPR32, "ldrsb">;
1615 def LDRSBXpost : LoadPostIdx<0b00, 0, 0b10, GPR64, "ldrsb">;
1617 // load zero-extended byte
1618 def LDRBBpost : LoadPostIdx<0b00, 0, 0b01, GPR32, "ldrb">;
1619 def LDRHHpost : LoadPostIdx<0b01, 0, 0b01, GPR32, "ldrh">;
1621 // load sign-extended word
1622 def LDRSWpost : LoadPostIdx<0b10, 0, 0b10, GPR64, "ldrsw">;
1624 // ISel pseudos and patterns. See expanded comment on LoadPostIdxPseudo.
1625 def LDRQpost_isel : LoadPostIdxPseudo<FPR128>;
1626 def LDRDpost_isel : LoadPostIdxPseudo<FPR64>;
1627 def LDRSpost_isel : LoadPostIdxPseudo<FPR32>;
1628 def LDRXpost_isel : LoadPostIdxPseudo<GPR64>;
1629 def LDRWpost_isel : LoadPostIdxPseudo<GPR32>;
1630 def LDRHHpost_isel : LoadPostIdxPseudo<GPR32>;
1631 def LDRBBpost_isel : LoadPostIdxPseudo<GPR32>;
1633 def LDRSWpost_isel : LoadPostIdxPseudo<GPR64>;
1634 def LDRSHWpost_isel : LoadPostIdxPseudo<GPR32>;
1635 def LDRSHXpost_isel : LoadPostIdxPseudo<GPR64>;
1636 def LDRSBWpost_isel : LoadPostIdxPseudo<GPR32>;
1637 def LDRSBXpost_isel : LoadPostIdxPseudo<GPR64>;
1639 //===----------------------------------------------------------------------===//
1640 // Store instructions.
1641 //===----------------------------------------------------------------------===//
1643 // Pair (indexed, offset)
1644 // FIXME: Use dedicated range-checked addressing mode operand here.
1645 def STPWi : StorePairOffset<0b00, 0, GPR32, am_indexed32simm7, "stp">;
1646 def STPXi : StorePairOffset<0b10, 0, GPR64, am_indexed64simm7, "stp">;
1647 def STPSi : StorePairOffset<0b00, 1, FPR32, am_indexed32simm7, "stp">;
1648 def STPDi : StorePairOffset<0b01, 1, FPR64, am_indexed64simm7, "stp">;
1649 def STPQi : StorePairOffset<0b10, 1, FPR128, am_indexed128simm7, "stp">;
1651 // Pair (pre-indexed)
1652 def STPWpre : StorePairPreIdx<0b00, 0, GPR32, am_indexed32simm7_wb, "stp">;
1653 def STPXpre : StorePairPreIdx<0b10, 0, GPR64, am_indexed64simm7_wb, "stp">;
1654 def STPSpre : StorePairPreIdx<0b00, 1, FPR32, am_indexed32simm7_wb, "stp">;
1655 def STPDpre : StorePairPreIdx<0b01, 1, FPR64, am_indexed64simm7_wb, "stp">;
1656 def STPQpre : StorePairPreIdx<0b10, 1, FPR128, am_indexed128simm7_wb, "stp">;
1658 // Pair (pre-indexed)
1659 def STPWpost : StorePairPostIdx<0b00, 0, GPR32, simm7s4, "stp">;
1660 def STPXpost : StorePairPostIdx<0b10, 0, GPR64, simm7s8, "stp">;
1661 def STPSpost : StorePairPostIdx<0b00, 1, FPR32, simm7s4, "stp">;
1662 def STPDpost : StorePairPostIdx<0b01, 1, FPR64, simm7s8, "stp">;
1663 def STPQpost : StorePairPostIdx<0b10, 1, FPR128, simm7s16, "stp">;
1665 // Pair (no allocate)
1666 def STNPWi : StorePairNoAlloc<0b00, 0, GPR32, am_indexed32simm7, "stnp">;
1667 def STNPXi : StorePairNoAlloc<0b10, 0, GPR64, am_indexed64simm7, "stnp">;
1668 def STNPSi : StorePairNoAlloc<0b00, 1, FPR32, am_indexed32simm7, "stnp">;
1669 def STNPDi : StorePairNoAlloc<0b01, 1, FPR64, am_indexed64simm7, "stnp">;
1670 def STNPQi : StorePairNoAlloc<0b10, 1, FPR128, am_indexed128simm7, "stnp">;
1673 // (Register offset)
1675 let AddedComplexity = 10 in {
1678 def STRHHro : Store16RO<0b01, 0, 0b00, GPR32, "strh",
1679 [(truncstorei16 GPR32:$Rt, ro_indexed16:$addr)]>;
1680 def STRBBro : Store8RO<0b00, 0, 0b00, GPR32, "strb",
1681 [(truncstorei8 GPR32:$Rt, ro_indexed8:$addr)]>;
1682 def STRWro : Store32RO<0b10, 0, 0b00, GPR32, "str",
1683 [(store GPR32:$Rt, ro_indexed32:$addr)]>;
1684 def STRXro : Store64RO<0b11, 0, 0b00, GPR64, "str",
1685 [(store GPR64:$Rt, ro_indexed64:$addr)]>;
1688 def : Pat<(truncstorei8 GPR64:$Rt, ro_indexed8:$addr),
1689 (STRBBro (EXTRACT_SUBREG GPR64:$Rt, sub_32), ro_indexed8:$addr)>;
1690 def : Pat<(truncstorei16 GPR64:$Rt, ro_indexed16:$addr),
1691 (STRHHro (EXTRACT_SUBREG GPR64:$Rt, sub_32), ro_indexed16:$addr)>;
1692 def : Pat<(truncstorei32 GPR64:$Rt, ro_indexed32:$addr),
1693 (STRWro (EXTRACT_SUBREG GPR64:$Rt, sub_32), ro_indexed32:$addr)>;
1697 def STRBro : Store8RO<0b00, 1, 0b00, FPR8, "str",
1698 [(store FPR8:$Rt, ro_indexed8:$addr)]>;
1699 def STRHro : Store16RO<0b01, 1, 0b00, FPR16, "str",
1700 [(store (f16 FPR16:$Rt), ro_indexed16:$addr)]>;
1701 def STRSro : Store32RO<0b10, 1, 0b00, FPR32, "str",
1702 [(store (f32 FPR32:$Rt), ro_indexed32:$addr)]>;
1703 def STRDro : Store64RO<0b11, 1, 0b00, FPR64, "str",
1704 [(store (f64 FPR64:$Rt), ro_indexed64:$addr)]>;
1705 def STRQro : Store128RO<0b00, 1, 0b10, FPR128, "str", []> {
1709 // Match all store 64 bits width whose type is compatible with FPR64
1710 let Predicates = [IsLE] in {
1711 // We must use ST1 to store vectors in big-endian.
1712 def : Pat<(store (v2f32 FPR64:$Rn), ro_indexed64:$addr),
1713 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1714 def : Pat<(store (v8i8 FPR64:$Rn), ro_indexed64:$addr),
1715 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1716 def : Pat<(store (v4i16 FPR64:$Rn), ro_indexed64:$addr),
1717 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1718 def : Pat<(store (v2i32 FPR64:$Rn), ro_indexed64:$addr),
1719 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1721 def : Pat<(store (v1f64 FPR64:$Rn), ro_indexed64:$addr),
1722 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1723 def : Pat<(store (v1i64 FPR64:$Rn), ro_indexed64:$addr),
1724 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1726 // Match all store 128 bits width whose type is compatible with FPR128
1727 let Predicates = [IsLE] in {
1728 // We must use ST1 to store vectors in big-endian.
1729 def : Pat<(store (v4f32 FPR128:$Rn), ro_indexed128:$addr),
1730 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1731 def : Pat<(store (v2f64 FPR128:$Rn), ro_indexed128:$addr),
1732 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1733 def : Pat<(store (v16i8 FPR128:$Rn), ro_indexed128:$addr),
1734 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1735 def : Pat<(store (v8i16 FPR128:$Rn), ro_indexed128:$addr),
1736 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1737 def : Pat<(store (v4i32 FPR128:$Rn), ro_indexed128:$addr),
1738 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1739 def : Pat<(store (v2i64 FPR128:$Rn), ro_indexed128:$addr),
1740 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1742 def : Pat<(store (f128 FPR128:$Rn), ro_indexed128:$addr),
1743 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1746 // (unsigned immediate)
1747 def STRXui : StoreUI<0b11, 0, 0b00, GPR64, am_indexed64, "str",
1748 [(store GPR64:$Rt, am_indexed64:$addr)]>;
1749 def STRWui : StoreUI<0b10, 0, 0b00, GPR32, am_indexed32, "str",
1750 [(store GPR32:$Rt, am_indexed32:$addr)]>;
1751 def STRBui : StoreUI<0b00, 1, 0b00, FPR8, am_indexed8, "str",
1752 [(store FPR8:$Rt, am_indexed8:$addr)]>;
1753 def STRHui : StoreUI<0b01, 1, 0b00, FPR16, am_indexed16, "str",
1754 [(store (f16 FPR16:$Rt), am_indexed16:$addr)]>;
1755 def STRSui : StoreUI<0b10, 1, 0b00, FPR32, am_indexed32, "str",
1756 [(store (f32 FPR32:$Rt), am_indexed32:$addr)]>;
1757 def STRDui : StoreUI<0b11, 1, 0b00, FPR64, am_indexed64, "str",
1758 [(store (f64 FPR64:$Rt), am_indexed64:$addr)]>;
1759 def STRQui : StoreUI<0b00, 1, 0b10, FPR128, am_indexed128, "str", []> {
1763 // Match all store 64 bits width whose type is compatible with FPR64
1764 let Predicates = [IsLE] in {
1765 // We must use ST1 to store vectors in big-endian.
1766 def : Pat<(store (v2f32 FPR64:$Rn), am_indexed64:$addr),
1767 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1768 def : Pat<(store (v8i8 FPR64:$Rn), am_indexed64:$addr),
1769 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1770 def : Pat<(store (v4i16 FPR64:$Rn), am_indexed64:$addr),
1771 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1772 def : Pat<(store (v2i32 FPR64:$Rn), am_indexed64:$addr),
1773 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1775 def : Pat<(store (v1f64 FPR64:$Rn), am_indexed64:$addr),
1776 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1777 def : Pat<(store (v1i64 FPR64:$Rn), am_indexed64:$addr),
1778 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1780 // Match all store 128 bits width whose type is compatible with FPR128
1781 let Predicates = [IsLE] in {
1782 // We must use ST1 to store vectors in big-endian.
1783 def : Pat<(store (v4f32 FPR128:$Rn), am_indexed128:$addr),
1784 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1785 def : Pat<(store (v2f64 FPR128:$Rn), am_indexed128:$addr),
1786 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1787 def : Pat<(store (v16i8 FPR128:$Rn), am_indexed128:$addr),
1788 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1789 def : Pat<(store (v8i16 FPR128:$Rn), am_indexed128:$addr),
1790 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1791 def : Pat<(store (v4i32 FPR128:$Rn), am_indexed128:$addr),
1792 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1793 def : Pat<(store (v2i64 FPR128:$Rn), am_indexed128:$addr),
1794 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1796 def : Pat<(store (f128 FPR128:$Rn), am_indexed128:$addr),
1797 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1799 def STRHHui : StoreUI<0b01, 0, 0b00, GPR32, am_indexed16, "strh",
1800 [(truncstorei16 GPR32:$Rt, am_indexed16:$addr)]>;
1801 def STRBBui : StoreUI<0b00, 0, 0b00, GPR32, am_indexed8, "strb",
1802 [(truncstorei8 GPR32:$Rt, am_indexed8:$addr)]>;
1805 def : Pat<(truncstorei32 GPR64:$Rt, am_indexed32:$addr),
1806 (STRWui (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_indexed32:$addr)>;
1807 def : Pat<(truncstorei16 GPR64:$Rt, am_indexed16:$addr),
1808 (STRHHui (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_indexed16:$addr)>;
1809 def : Pat<(truncstorei8 GPR64:$Rt, am_indexed8:$addr),
1810 (STRBBui (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_indexed8:$addr)>;
1812 } // AddedComplexity = 10
1815 // (unscaled immediate)
1816 def STURXi : StoreUnscaled<0b11, 0, 0b00, GPR64, am_unscaled64, "stur",
1817 [(store GPR64:$Rt, am_unscaled64:$addr)]>;
1818 def STURWi : StoreUnscaled<0b10, 0, 0b00, GPR32, am_unscaled32, "stur",
1819 [(store GPR32:$Rt, am_unscaled32:$addr)]>;
1820 def STURBi : StoreUnscaled<0b00, 1, 0b00, FPR8, am_unscaled8, "stur",
1821 [(store FPR8:$Rt, am_unscaled8:$addr)]>;
1822 def STURHi : StoreUnscaled<0b01, 1, 0b00, FPR16, am_unscaled16, "stur",
1823 [(store (f16 FPR16:$Rt), am_unscaled16:$addr)]>;
1824 def STURSi : StoreUnscaled<0b10, 1, 0b00, FPR32, am_unscaled32, "stur",
1825 [(store (f32 FPR32:$Rt), am_unscaled32:$addr)]>;
1826 def STURDi : StoreUnscaled<0b11, 1, 0b00, FPR64, am_unscaled64, "stur",
1827 [(store (f64 FPR64:$Rt), am_unscaled64:$addr)]>;
1828 def STURQi : StoreUnscaled<0b00, 1, 0b10, FPR128, am_unscaled128, "stur",
1829 [(store (f128 FPR128:$Rt), am_unscaled128:$addr)]>;
1830 def STURHHi : StoreUnscaled<0b01, 0, 0b00, GPR32, am_unscaled16, "sturh",
1831 [(truncstorei16 GPR32:$Rt, am_unscaled16:$addr)]>;
1832 def STURBBi : StoreUnscaled<0b00, 0, 0b00, GPR32, am_unscaled8, "sturb",
1833 [(truncstorei8 GPR32:$Rt, am_unscaled8:$addr)]>;
1835 // Match all store 64 bits width whose type is compatible with FPR64
1836 let Predicates = [IsLE] in {
1837 // We must use ST1 to store vectors in big-endian.
1838 def : Pat<(store (v2f32 FPR64:$Rn), am_unscaled64:$addr),
1839 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1840 def : Pat<(store (v8i8 FPR64:$Rn), am_unscaled64:$addr),
1841 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1842 def : Pat<(store (v4i16 FPR64:$Rn), am_unscaled64:$addr),
1843 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1844 def : Pat<(store (v2i32 FPR64:$Rn), am_unscaled64:$addr),
1845 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1847 def : Pat<(store (v1f64 FPR64:$Rn), am_unscaled64:$addr),
1848 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1849 def : Pat<(store (v1i64 FPR64:$Rn), am_unscaled64:$addr),
1850 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1852 // Match all store 128 bits width whose type is compatible with FPR128
1853 let Predicates = [IsLE] in {
1854 // We must use ST1 to store vectors in big-endian.
1855 def : Pat<(store (v4f32 FPR128:$Rn), am_unscaled128:$addr),
1856 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1857 def : Pat<(store (v2f64 FPR128:$Rn), am_unscaled128:$addr),
1858 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1859 def : Pat<(store (v16i8 FPR128:$Rn), am_unscaled128:$addr),
1860 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1861 def : Pat<(store (v8i16 FPR128:$Rn), am_unscaled128:$addr),
1862 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1863 def : Pat<(store (v4i32 FPR128:$Rn), am_unscaled128:$addr),
1864 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1865 def : Pat<(store (v2i64 FPR128:$Rn), am_unscaled128:$addr),
1866 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1867 def : Pat<(store (v2f64 FPR128:$Rn), am_unscaled128:$addr),
1868 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1871 // unscaled i64 truncating stores
1872 def : Pat<(truncstorei32 GPR64:$Rt, am_unscaled32:$addr),
1873 (STURWi (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_unscaled32:$addr)>;
1874 def : Pat<(truncstorei16 GPR64:$Rt, am_unscaled16:$addr),
1875 (STURHHi (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_unscaled16:$addr)>;
1876 def : Pat<(truncstorei8 GPR64:$Rt, am_unscaled8:$addr),
1877 (STURBBi (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_unscaled8:$addr)>;
1880 // STR mnemonics fall back to STUR for negative or unaligned offsets.
1881 def : InstAlias<"str $Rt, $addr",
1882 (STURXi GPR64:$Rt, am_unscaled_fb64:$addr), 0>;
1883 def : InstAlias<"str $Rt, $addr",
1884 (STURWi GPR32:$Rt, am_unscaled_fb32:$addr), 0>;
1885 def : InstAlias<"str $Rt, $addr",
1886 (STURBi FPR8:$Rt, am_unscaled_fb8:$addr), 0>;
1887 def : InstAlias<"str $Rt, $addr",
1888 (STURHi FPR16:$Rt, am_unscaled_fb16:$addr), 0>;
1889 def : InstAlias<"str $Rt, $addr",
1890 (STURSi FPR32:$Rt, am_unscaled_fb32:$addr), 0>;
1891 def : InstAlias<"str $Rt, $addr",
1892 (STURDi FPR64:$Rt, am_unscaled_fb64:$addr), 0>;
1893 def : InstAlias<"str $Rt, $addr",
1894 (STURQi FPR128:$Rt, am_unscaled_fb128:$addr), 0>;
1896 def : InstAlias<"strb $Rt, $addr",
1897 (STURBBi GPR32:$Rt, am_unscaled_fb8:$addr), 0>;
1898 def : InstAlias<"strh $Rt, $addr",
1899 (STURHHi GPR32:$Rt, am_unscaled_fb16:$addr), 0>;
1902 // (unscaled immediate, unprivileged)
1903 def STTRWi : StoreUnprivileged<0b10, 0, 0b00, GPR32, "sttr">;
1904 def STTRXi : StoreUnprivileged<0b11, 0, 0b00, GPR64, "sttr">;
1906 def STTRHi : StoreUnprivileged<0b01, 0, 0b00, GPR32, "sttrh">;
1907 def STTRBi : StoreUnprivileged<0b00, 0, 0b00, GPR32, "sttrb">;
1910 // (immediate pre-indexed)
1911 def STRWpre : StorePreIdx<0b10, 0, 0b00, GPR32, "str">;
1912 def STRXpre : StorePreIdx<0b11, 0, 0b00, GPR64, "str">;
1913 def STRBpre : StorePreIdx<0b00, 1, 0b00, FPR8, "str">;
1914 def STRHpre : StorePreIdx<0b01, 1, 0b00, FPR16, "str">;
1915 def STRSpre : StorePreIdx<0b10, 1, 0b00, FPR32, "str">;
1916 def STRDpre : StorePreIdx<0b11, 1, 0b00, FPR64, "str">;
1917 def STRQpre : StorePreIdx<0b00, 1, 0b10, FPR128, "str">;
1919 def STRBBpre : StorePreIdx<0b00, 0, 0b00, GPR32, "strb">;
1920 def STRHHpre : StorePreIdx<0b01, 0, 0b00, GPR32, "strh">;
1922 // ISel pseudos and patterns. See expanded comment on StorePreIdxPseudo.
1923 defm STRQpre : StorePreIdxPseudo<FPR128, f128, pre_store>;
1924 defm STRDpre : StorePreIdxPseudo<FPR64, f64, pre_store>;
1925 defm STRSpre : StorePreIdxPseudo<FPR32, f32, pre_store>;
1926 defm STRXpre : StorePreIdxPseudo<GPR64, i64, pre_store>;
1927 defm STRWpre : StorePreIdxPseudo<GPR32, i32, pre_store>;
1928 defm STRHHpre : StorePreIdxPseudo<GPR32, i32, pre_truncsti16>;
1929 defm STRBBpre : StorePreIdxPseudo<GPR32, i32, pre_truncsti8>;
1931 def : Pat<(pre_truncsti32 GPR64:$Rt, am_noindex:$addr, simm9:$off),
1932 (STRWpre_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
1934 def : Pat<(pre_truncsti16 GPR64:$Rt, am_noindex:$addr, simm9:$off),
1935 (STRHHpre_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
1937 def : Pat<(pre_truncsti8 GPR64:$Rt, am_noindex:$addr, simm9:$off),
1938 (STRBBpre_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
1941 def : Pat<(pre_store (v8i8 FPR64:$Rt), am_noindex:$addr, simm9:$off),
1942 (STRDpre_isel FPR64:$Rt, am_noindex:$addr, simm9:$off)>;
1943 def : Pat<(pre_store (v4i16 FPR64:$Rt), am_noindex:$addr, simm9:$off),
1944 (STRDpre_isel FPR64:$Rt, am_noindex:$addr, simm9:$off)>;
1945 def : Pat<(pre_store (v2i32 FPR64:$Rt), am_noindex:$addr, simm9:$off),
1946 (STRDpre_isel FPR64:$Rt, am_noindex:$addr, simm9:$off)>;
1947 def : Pat<(pre_store (v2f32 FPR64:$Rt), am_noindex:$addr, simm9:$off),
1948 (STRDpre_isel FPR64:$Rt, am_noindex:$addr, simm9:$off)>;
1949 def : Pat<(pre_store (v1i64 FPR64:$Rt), am_noindex:$addr, simm9:$off),
1950 (STRDpre_isel FPR64:$Rt, am_noindex:$addr, simm9:$off)>;
1951 def : Pat<(pre_store (v1f64 FPR64:$Rt), am_noindex:$addr, simm9:$off),
1952 (STRDpre_isel FPR64:$Rt, am_noindex:$addr, simm9:$off)>;
1954 def : Pat<(pre_store (v16i8 FPR128:$Rt), am_noindex:$addr, simm9:$off),
1955 (STRQpre_isel FPR128:$Rt, am_noindex:$addr, simm9:$off)>;
1956 def : Pat<(pre_store (v8i16 FPR128:$Rt), am_noindex:$addr, simm9:$off),
1957 (STRQpre_isel FPR128:$Rt, am_noindex:$addr, simm9:$off)>;
1958 def : Pat<(pre_store (v4i32 FPR128:$Rt), am_noindex:$addr, simm9:$off),
1959 (STRQpre_isel FPR128:$Rt, am_noindex:$addr, simm9:$off)>;
1960 def : Pat<(pre_store (v4f32 FPR128:$Rt), am_noindex:$addr, simm9:$off),
1961 (STRQpre_isel FPR128:$Rt, am_noindex:$addr, simm9:$off)>;
1962 def : Pat<(pre_store (v2i64 FPR128:$Rt), am_noindex:$addr, simm9:$off),
1963 (STRQpre_isel FPR128:$Rt, am_noindex:$addr, simm9:$off)>;
1964 def : Pat<(pre_store (v2f64 FPR128:$Rt), am_noindex:$addr, simm9:$off),
1965 (STRQpre_isel FPR128:$Rt, am_noindex:$addr, simm9:$off)>;
1968 // (immediate post-indexed)
1969 def STRWpost : StorePostIdx<0b10, 0, 0b00, GPR32, "str">;
1970 def STRXpost : StorePostIdx<0b11, 0, 0b00, GPR64, "str">;
1971 def STRBpost : StorePostIdx<0b00, 1, 0b00, FPR8, "str">;
1972 def STRHpost : StorePostIdx<0b01, 1, 0b00, FPR16, "str">;
1973 def STRSpost : StorePostIdx<0b10, 1, 0b00, FPR32, "str">;
1974 def STRDpost : StorePostIdx<0b11, 1, 0b00, FPR64, "str">;
1975 def STRQpost : StorePostIdx<0b00, 1, 0b10, FPR128, "str">;
1977 def STRBBpost : StorePostIdx<0b00, 0, 0b00, GPR32, "strb">;
1978 def STRHHpost : StorePostIdx<0b01, 0, 0b00, GPR32, "strh">;
1980 // ISel pseudos and patterns. See expanded comment on StorePostIdxPseudo.
1981 defm STRQpost : StorePostIdxPseudo<FPR128, f128, post_store, STRQpost>;
1982 defm STRDpost : StorePostIdxPseudo<FPR64, f64, post_store, STRDpost>;
1983 defm STRSpost : StorePostIdxPseudo<FPR32, f32, post_store, STRSpost>;
1984 defm STRXpost : StorePostIdxPseudo<GPR64, i64, post_store, STRXpost>;
1985 defm STRWpost : StorePostIdxPseudo<GPR32, i32, post_store, STRWpost>;
1986 defm STRHHpost : StorePostIdxPseudo<GPR32, i32, post_truncsti16, STRHHpost>;
1987 defm STRBBpost : StorePostIdxPseudo<GPR32, i32, post_truncsti8, STRBBpost>;
1989 def : Pat<(post_truncsti32 GPR64:$Rt, am_noindex:$addr, simm9:$off),
1990 (STRWpost_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
1992 def : Pat<(post_truncsti16 GPR64:$Rt, am_noindex:$addr, simm9:$off),
1993 (STRHHpost_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
1995 def : Pat<(post_truncsti8 GPR64:$Rt, am_noindex:$addr, simm9:$off),
1996 (STRBBpost_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
1999 def : Pat<(post_store (v8i8 FPR64:$Rt), am_noindex:$addr, simm9:$off),
2000 (STRDpost_isel FPR64:$Rt, am_noindex:$addr, simm9:$off)>;
2001 def : Pat<(post_store (v4i16 FPR64:$Rt), am_noindex:$addr, simm9:$off),
2002 (STRDpost_isel FPR64:$Rt, am_noindex:$addr, simm9:$off)>;
2003 def : Pat<(post_store (v2i32 FPR64:$Rt), am_noindex:$addr, simm9:$off),
2004 (STRDpost_isel FPR64:$Rt, am_noindex:$addr, simm9:$off)>;
2005 def : Pat<(post_store (v2f32 FPR64:$Rt), am_noindex:$addr, simm9:$off),
2006 (STRDpost_isel FPR64:$Rt, am_noindex:$addr, simm9:$off)>;
2007 def : Pat<(post_store (v1i64 FPR64:$Rt), am_noindex:$addr, simm9:$off),
2008 (STRDpost_isel FPR64:$Rt, am_noindex:$addr, simm9:$off)>;
2009 def : Pat<(post_store (v1f64 FPR64:$Rt), am_noindex:$addr, simm9:$off),
2010 (STRDpost_isel FPR64:$Rt, am_noindex:$addr, simm9:$off)>;
2012 def : Pat<(post_store (v16i8 FPR128:$Rt), am_noindex:$addr, simm9:$off),
2013 (STRQpost_isel FPR128:$Rt, am_noindex:$addr, simm9:$off)>;
2014 def : Pat<(post_store (v8i16 FPR128:$Rt), am_noindex:$addr, simm9:$off),
2015 (STRQpost_isel FPR128:$Rt, am_noindex:$addr, simm9:$off)>;
2016 def : Pat<(post_store (v4i32 FPR128:$Rt), am_noindex:$addr, simm9:$off),
2017 (STRQpost_isel FPR128:$Rt, am_noindex:$addr, simm9:$off)>;
2018 def : Pat<(post_store (v4f32 FPR128:$Rt), am_noindex:$addr, simm9:$off),
2019 (STRQpost_isel FPR128:$Rt, am_noindex:$addr, simm9:$off)>;
2020 def : Pat<(post_store (v2i64 FPR128:$Rt), am_noindex:$addr, simm9:$off),
2021 (STRQpost_isel FPR128:$Rt, am_noindex:$addr, simm9:$off)>;
2022 def : Pat<(post_store (v2f64 FPR128:$Rt), am_noindex:$addr, simm9:$off),
2023 (STRQpost_isel FPR128:$Rt, am_noindex:$addr, simm9:$off)>;
2025 //===----------------------------------------------------------------------===//
2026 // Load/store exclusive instructions.
2027 //===----------------------------------------------------------------------===//
2029 def LDARW : LoadAcquire <0b10, 1, 1, 0, 1, GPR32, "ldar">;
2030 def LDARX : LoadAcquire <0b11, 1, 1, 0, 1, GPR64, "ldar">;
2031 def LDARB : LoadAcquire <0b00, 1, 1, 0, 1, GPR32, "ldarb">;
2032 def LDARH : LoadAcquire <0b01, 1, 1, 0, 1, GPR32, "ldarh">;
2034 def LDAXRW : LoadExclusive <0b10, 0, 1, 0, 1, GPR32, "ldaxr">;
2035 def LDAXRX : LoadExclusive <0b11, 0, 1, 0, 1, GPR64, "ldaxr">;
2036 def LDAXRB : LoadExclusive <0b00, 0, 1, 0, 1, GPR32, "ldaxrb">;
2037 def LDAXRH : LoadExclusive <0b01, 0, 1, 0, 1, GPR32, "ldaxrh">;
2039 def LDXRW : LoadExclusive <0b10, 0, 1, 0, 0, GPR32, "ldxr">;
2040 def LDXRX : LoadExclusive <0b11, 0, 1, 0, 0, GPR64, "ldxr">;
2041 def LDXRB : LoadExclusive <0b00, 0, 1, 0, 0, GPR32, "ldxrb">;
2042 def LDXRH : LoadExclusive <0b01, 0, 1, 0, 0, GPR32, "ldxrh">;
2044 def STLRW : StoreRelease <0b10, 1, 0, 0, 1, GPR32, "stlr">;
2045 def STLRX : StoreRelease <0b11, 1, 0, 0, 1, GPR64, "stlr">;
2046 def STLRB : StoreRelease <0b00, 1, 0, 0, 1, GPR32, "stlrb">;
2047 def STLRH : StoreRelease <0b01, 1, 0, 0, 1, GPR32, "stlrh">;
2049 def STLXRW : StoreExclusive<0b10, 0, 0, 0, 1, GPR32, "stlxr">;
2050 def STLXRX : StoreExclusive<0b11, 0, 0, 0, 1, GPR64, "stlxr">;
2051 def STLXRB : StoreExclusive<0b00, 0, 0, 0, 1, GPR32, "stlxrb">;
2052 def STLXRH : StoreExclusive<0b01, 0, 0, 0, 1, GPR32, "stlxrh">;
2054 def STXRW : StoreExclusive<0b10, 0, 0, 0, 0, GPR32, "stxr">;
2055 def STXRX : StoreExclusive<0b11, 0, 0, 0, 0, GPR64, "stxr">;
2056 def STXRB : StoreExclusive<0b00, 0, 0, 0, 0, GPR32, "stxrb">;
2057 def STXRH : StoreExclusive<0b01, 0, 0, 0, 0, GPR32, "stxrh">;
2059 def LDAXPW : LoadExclusivePair<0b10, 0, 1, 1, 1, GPR32, "ldaxp">;
2060 def LDAXPX : LoadExclusivePair<0b11, 0, 1, 1, 1, GPR64, "ldaxp">;
2062 def LDXPW : LoadExclusivePair<0b10, 0, 1, 1, 0, GPR32, "ldxp">;
2063 def LDXPX : LoadExclusivePair<0b11, 0, 1, 1, 0, GPR64, "ldxp">;
2065 def STLXPW : StoreExclusivePair<0b10, 0, 0, 1, 1, GPR32, "stlxp">;
2066 def STLXPX : StoreExclusivePair<0b11, 0, 0, 1, 1, GPR64, "stlxp">;
2068 def STXPW : StoreExclusivePair<0b10, 0, 0, 1, 0, GPR32, "stxp">;
2069 def STXPX : StoreExclusivePair<0b11, 0, 0, 1, 0, GPR64, "stxp">;
2071 //===----------------------------------------------------------------------===//
2072 // Scaled floating point to integer conversion instructions.
2073 //===----------------------------------------------------------------------===//
2075 defm FCVTAS : FPToIntegerUnscaled<0b00, 0b100, "fcvtas", int_arm64_neon_fcvtas>;
2076 defm FCVTAU : FPToIntegerUnscaled<0b00, 0b101, "fcvtau", int_arm64_neon_fcvtau>;
2077 defm FCVTMS : FPToIntegerUnscaled<0b10, 0b000, "fcvtms", int_arm64_neon_fcvtms>;
2078 defm FCVTMU : FPToIntegerUnscaled<0b10, 0b001, "fcvtmu", int_arm64_neon_fcvtmu>;
2079 defm FCVTNS : FPToIntegerUnscaled<0b00, 0b000, "fcvtns", int_arm64_neon_fcvtns>;
2080 defm FCVTNU : FPToIntegerUnscaled<0b00, 0b001, "fcvtnu", int_arm64_neon_fcvtnu>;
2081 defm FCVTPS : FPToIntegerUnscaled<0b01, 0b000, "fcvtps", int_arm64_neon_fcvtps>;
2082 defm FCVTPU : FPToIntegerUnscaled<0b01, 0b001, "fcvtpu", int_arm64_neon_fcvtpu>;
2083 defm FCVTZS : FPToIntegerUnscaled<0b11, 0b000, "fcvtzs", fp_to_sint>;
2084 defm FCVTZU : FPToIntegerUnscaled<0b11, 0b001, "fcvtzu", fp_to_uint>;
2085 defm FCVTZS : FPToIntegerScaled<0b11, 0b000, "fcvtzs", fp_to_sint>;
2086 defm FCVTZU : FPToIntegerScaled<0b11, 0b001, "fcvtzu", fp_to_uint>;
2087 let isCodeGenOnly = 1 in {
2088 defm FCVTZS_Int : FPToIntegerUnscaled<0b11, 0b000, "fcvtzs", int_arm64_neon_fcvtzs>;
2089 defm FCVTZU_Int : FPToIntegerUnscaled<0b11, 0b001, "fcvtzu", int_arm64_neon_fcvtzu>;
2090 defm FCVTZS_Int : FPToIntegerScaled<0b11, 0b000, "fcvtzs", int_arm64_neon_fcvtzs>;
2091 defm FCVTZU_Int : FPToIntegerScaled<0b11, 0b001, "fcvtzu", int_arm64_neon_fcvtzu>;
2094 //===----------------------------------------------------------------------===//
2095 // Scaled integer to floating point conversion instructions.
2096 //===----------------------------------------------------------------------===//
2098 defm SCVTF : IntegerToFP<0, "scvtf", sint_to_fp>;
2099 defm UCVTF : IntegerToFP<1, "ucvtf", uint_to_fp>;
2101 //===----------------------------------------------------------------------===//
2102 // Unscaled integer to floating point conversion instruction.
2103 //===----------------------------------------------------------------------===//
2105 defm FMOV : UnscaledConversion<"fmov">;
2107 def : Pat<(f32 (fpimm0)), (FMOVWSr WZR)>, Requires<[NoZCZ]>;
2108 def : Pat<(f64 (fpimm0)), (FMOVXDr XZR)>, Requires<[NoZCZ]>;
2110 //===----------------------------------------------------------------------===//
2111 // Floating point conversion instruction.
2112 //===----------------------------------------------------------------------===//
2114 defm FCVT : FPConversion<"fcvt">;
2116 def : Pat<(f32_to_f16 FPR32:$Rn),
2117 (i32 (COPY_TO_REGCLASS
2118 (f32 (SUBREG_TO_REG (i32 0), (FCVTHSr FPR32:$Rn), hsub)),
2121 def FCVTSHpseudo : Pseudo<(outs FPR32:$Rd), (ins FPR32:$Rn),
2122 [(set (f32 FPR32:$Rd), (f16_to_f32 i32:$Rn))]>;
2124 //===----------------------------------------------------------------------===//
2125 // Floating point single operand instructions.
2126 //===----------------------------------------------------------------------===//
2128 defm FABS : SingleOperandFPData<0b0001, "fabs", fabs>;
2129 defm FMOV : SingleOperandFPData<0b0000, "fmov">;
2130 defm FNEG : SingleOperandFPData<0b0010, "fneg", fneg>;
2131 defm FRINTA : SingleOperandFPData<0b1100, "frinta", frnd>;
2132 defm FRINTI : SingleOperandFPData<0b1111, "frinti", fnearbyint>;
2133 defm FRINTM : SingleOperandFPData<0b1010, "frintm", ffloor>;
2134 defm FRINTN : SingleOperandFPData<0b1000, "frintn", int_arm64_neon_frintn>;
2135 defm FRINTP : SingleOperandFPData<0b1001, "frintp", fceil>;
2137 def : Pat<(v1f64 (int_arm64_neon_frintn (v1f64 FPR64:$Rn))),
2138 (FRINTNDr FPR64:$Rn)>;
2140 // FRINTX is inserted to set the flags as required by FENV_ACCESS ON behavior
2141 // in the C spec. Setting hasSideEffects ensures it is not DCE'd.
2142 // <rdar://problem/13715968>
2143 // TODO: We should really model the FPSR flags correctly. This is really ugly.
2144 let hasSideEffects = 1 in {
2145 defm FRINTX : SingleOperandFPData<0b1110, "frintx", frint>;
2148 defm FRINTZ : SingleOperandFPData<0b1011, "frintz", ftrunc>;
2150 let SchedRW = [WriteFDiv] in {
2151 defm FSQRT : SingleOperandFPData<0b0011, "fsqrt", fsqrt>;
2154 //===----------------------------------------------------------------------===//
2155 // Floating point two operand instructions.
2156 //===----------------------------------------------------------------------===//
2158 defm FADD : TwoOperandFPData<0b0010, "fadd", fadd>;
2159 let SchedRW = [WriteFDiv] in {
2160 defm FDIV : TwoOperandFPData<0b0001, "fdiv", fdiv>;
2162 defm FMAXNM : TwoOperandFPData<0b0110, "fmaxnm", int_arm64_neon_fmaxnm>;
2163 defm FMAX : TwoOperandFPData<0b0100, "fmax", ARM64fmax>;
2164 defm FMINNM : TwoOperandFPData<0b0111, "fminnm", int_arm64_neon_fminnm>;
2165 defm FMIN : TwoOperandFPData<0b0101, "fmin", ARM64fmin>;
2166 let SchedRW = [WriteFMul] in {
2167 defm FMUL : TwoOperandFPData<0b0000, "fmul", fmul>;
2168 defm FNMUL : TwoOperandFPDataNeg<0b1000, "fnmul", fmul>;
2170 defm FSUB : TwoOperandFPData<0b0011, "fsub", fsub>;
2172 def : Pat<(v1f64 (ARM64fmax (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2173 (FMAXDrr FPR64:$Rn, FPR64:$Rm)>;
2174 def : Pat<(v1f64 (ARM64fmin (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2175 (FMINDrr FPR64:$Rn, FPR64:$Rm)>;
2176 def : Pat<(v1f64 (int_arm64_neon_fmaxnm (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2177 (FMAXNMDrr FPR64:$Rn, FPR64:$Rm)>;
2178 def : Pat<(v1f64 (int_arm64_neon_fminnm (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2179 (FMINNMDrr FPR64:$Rn, FPR64:$Rm)>;
2181 //===----------------------------------------------------------------------===//
2182 // Floating point three operand instructions.
2183 //===----------------------------------------------------------------------===//
2185 defm FMADD : ThreeOperandFPData<0, 0, "fmadd", fma>;
2186 defm FMSUB : ThreeOperandFPData<0, 1, "fmsub",
2187 TriOpFrag<(fma node:$LHS, (fneg node:$MHS), node:$RHS)> >;
2188 defm FNMADD : ThreeOperandFPData<1, 0, "fnmadd",
2189 TriOpFrag<(fneg (fma node:$LHS, node:$MHS, node:$RHS))> >;
2190 defm FNMSUB : ThreeOperandFPData<1, 1, "fnmsub",
2191 TriOpFrag<(fma node:$LHS, node:$MHS, (fneg node:$RHS))> >;
2193 // The following def pats catch the case where the LHS of an FMA is negated.
2194 // The TriOpFrag above catches the case where the middle operand is negated.
2196 // N.b. FMSUB etc have the accumulator at the *end* of (outs), unlike
2197 // the NEON variant.
2198 def : Pat<(f32 (fma (fneg FPR32:$Rn), FPR32:$Rm, FPR32:$Ra)),
2199 (FMSUBSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
2201 def : Pat<(f64 (fma (fneg FPR64:$Rn), FPR64:$Rm, FPR64:$Ra)),
2202 (FMSUBDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
2204 // We handled -(a + b*c) for FNMADD above, now it's time for "(-a) + (-b)*c" and
2206 def : Pat<(f32 (fma (fneg FPR32:$Rn), FPR32:$Rm, (fneg FPR32:$Ra))),
2207 (FNMADDSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
2209 def : Pat<(f64 (fma (fneg FPR64:$Rn), FPR64:$Rm, (fneg FPR64:$Ra))),
2210 (FNMADDDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
2212 def : Pat<(f32 (fma FPR32:$Rn, (fneg FPR32:$Rm), (fneg FPR32:$Ra))),
2213 (FNMADDSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
2215 def : Pat<(f64 (fma FPR64:$Rn, (fneg FPR64:$Rm), (fneg FPR64:$Ra))),
2216 (FNMADDDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
2218 //===----------------------------------------------------------------------===//
2219 // Floating point comparison instructions.
2220 //===----------------------------------------------------------------------===//
2222 defm FCMPE : FPComparison<1, "fcmpe">;
2223 defm FCMP : FPComparison<0, "fcmp", ARM64fcmp>;
2225 //===----------------------------------------------------------------------===//
2226 // Floating point conditional comparison instructions.
2227 //===----------------------------------------------------------------------===//
2229 defm FCCMPE : FPCondComparison<1, "fccmpe">;
2230 defm FCCMP : FPCondComparison<0, "fccmp">;
2232 //===----------------------------------------------------------------------===//
2233 // Floating point conditional select instruction.
2234 //===----------------------------------------------------------------------===//
2236 defm FCSEL : FPCondSelect<"fcsel">;
2238 // CSEL instructions providing f128 types need to be handled by a
2239 // pseudo-instruction since the eventual code will need to introduce basic
2240 // blocks and control flow.
2241 def F128CSEL : Pseudo<(outs FPR128:$Rd),
2242 (ins FPR128:$Rn, FPR128:$Rm, ccode:$cond),
2243 [(set (f128 FPR128:$Rd),
2244 (ARM64csel FPR128:$Rn, FPR128:$Rm,
2245 (i32 imm:$cond), NZCV))]> {
2247 let usesCustomInserter = 1;
2251 //===----------------------------------------------------------------------===//
2252 // Floating point immediate move.
2253 //===----------------------------------------------------------------------===//
2255 let isReMaterializable = 1 in {
2256 defm FMOV : FPMoveImmediate<"fmov">;
2259 //===----------------------------------------------------------------------===//
2260 // Advanced SIMD two vector instructions.
2261 //===----------------------------------------------------------------------===//
2263 defm ABS : SIMDTwoVectorBHSD<0, 0b01011, "abs", int_arm64_neon_abs>;
2264 defm CLS : SIMDTwoVectorBHS<0, 0b00100, "cls", int_arm64_neon_cls>;
2265 defm CLZ : SIMDTwoVectorBHS<1, 0b00100, "clz", ctlz>;
2266 defm CMEQ : SIMDCmpTwoVector<0, 0b01001, "cmeq", ARM64cmeqz>;
2267 defm CMGE : SIMDCmpTwoVector<1, 0b01000, "cmge", ARM64cmgez>;
2268 defm CMGT : SIMDCmpTwoVector<0, 0b01000, "cmgt", ARM64cmgtz>;
2269 defm CMLE : SIMDCmpTwoVector<1, 0b01001, "cmle", ARM64cmlez>;
2270 defm CMLT : SIMDCmpTwoVector<0, 0b01010, "cmlt", ARM64cmltz>;
2271 defm CNT : SIMDTwoVectorB<0, 0b00, 0b00101, "cnt", ctpop>;
2272 defm FABS : SIMDTwoVectorFP<0, 1, 0b01111, "fabs", fabs>;
2274 defm FCMEQ : SIMDFPCmpTwoVector<0, 1, 0b01101, "fcmeq", ARM64fcmeqz>;
2275 defm FCMGE : SIMDFPCmpTwoVector<1, 1, 0b01100, "fcmge", ARM64fcmgez>;
2276 defm FCMGT : SIMDFPCmpTwoVector<0, 1, 0b01100, "fcmgt", ARM64fcmgtz>;
2277 defm FCMLE : SIMDFPCmpTwoVector<1, 1, 0b01101, "fcmle", ARM64fcmlez>;
2278 defm FCMLT : SIMDFPCmpTwoVector<0, 1, 0b01110, "fcmlt", ARM64fcmltz>;
2279 defm FCVTAS : SIMDTwoVectorFPToInt<0,0,0b11100, "fcvtas",int_arm64_neon_fcvtas>;
2280 defm FCVTAU : SIMDTwoVectorFPToInt<1,0,0b11100, "fcvtau",int_arm64_neon_fcvtau>;
2281 defm FCVTL : SIMDFPWidenTwoVector<0, 0, 0b10111, "fcvtl">;
2282 def : Pat<(v4f32 (int_arm64_neon_vcvthf2fp (v4i16 V64:$Rn))),
2283 (FCVTLv4i16 V64:$Rn)>;
2284 def : Pat<(v4f32 (int_arm64_neon_vcvthf2fp (extract_subvector (v8i16 V128:$Rn),
2286 (FCVTLv8i16 V128:$Rn)>;
2287 def : Pat<(v2f64 (fextend (v2f32 V64:$Rn))), (FCVTLv2i32 V64:$Rn)>;
2288 def : Pat<(v2f64 (fextend (v2f32 (extract_subvector (v4f32 V128:$Rn),
2290 (FCVTLv4i32 V128:$Rn)>;
2292 defm FCVTMS : SIMDTwoVectorFPToInt<0,0,0b11011, "fcvtms",int_arm64_neon_fcvtms>;
2293 defm FCVTMU : SIMDTwoVectorFPToInt<1,0,0b11011, "fcvtmu",int_arm64_neon_fcvtmu>;
2294 defm FCVTNS : SIMDTwoVectorFPToInt<0,0,0b11010, "fcvtns",int_arm64_neon_fcvtns>;
2295 defm FCVTNU : SIMDTwoVectorFPToInt<1,0,0b11010, "fcvtnu",int_arm64_neon_fcvtnu>;
2296 defm FCVTN : SIMDFPNarrowTwoVector<0, 0, 0b10110, "fcvtn">;
2297 def : Pat<(v4i16 (int_arm64_neon_vcvtfp2hf (v4f32 V128:$Rn))),
2298 (FCVTNv4i16 V128:$Rn)>;
2299 def : Pat<(concat_vectors V64:$Rd,
2300 (v4i16 (int_arm64_neon_vcvtfp2hf (v4f32 V128:$Rn)))),
2301 (FCVTNv8i16 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
2302 def : Pat<(v2f32 (fround (v2f64 V128:$Rn))), (FCVTNv2i32 V128:$Rn)>;
2303 def : Pat<(concat_vectors V64:$Rd, (v2f32 (fround (v2f64 V128:$Rn)))),
2304 (FCVTNv4i32 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
2305 defm FCVTPS : SIMDTwoVectorFPToInt<0,1,0b11010, "fcvtps",int_arm64_neon_fcvtps>;
2306 defm FCVTPU : SIMDTwoVectorFPToInt<1,1,0b11010, "fcvtpu",int_arm64_neon_fcvtpu>;
2307 defm FCVTXN : SIMDFPInexactCvtTwoVector<1, 0, 0b10110, "fcvtxn",
2308 int_arm64_neon_fcvtxn>;
2309 defm FCVTZS : SIMDTwoVectorFPToInt<0, 1, 0b11011, "fcvtzs", fp_to_sint>;
2310 defm FCVTZU : SIMDTwoVectorFPToInt<1, 1, 0b11011, "fcvtzu", fp_to_uint>;
2311 let isCodeGenOnly = 1 in {
2312 defm FCVTZS_Int : SIMDTwoVectorFPToInt<0, 1, 0b11011, "fcvtzs",
2313 int_arm64_neon_fcvtzs>;
2314 defm FCVTZU_Int : SIMDTwoVectorFPToInt<1, 1, 0b11011, "fcvtzu",
2315 int_arm64_neon_fcvtzu>;
2317 defm FNEG : SIMDTwoVectorFP<1, 1, 0b01111, "fneg", fneg>;
2318 defm FRECPE : SIMDTwoVectorFP<0, 1, 0b11101, "frecpe", int_arm64_neon_frecpe>;
2319 defm FRINTA : SIMDTwoVectorFP<1, 0, 0b11000, "frinta", frnd>;
2320 defm FRINTI : SIMDTwoVectorFP<1, 1, 0b11001, "frinti", fnearbyint>;
2321 defm FRINTM : SIMDTwoVectorFP<0, 0, 0b11001, "frintm", ffloor>;
2322 defm FRINTN : SIMDTwoVectorFP<0, 0, 0b11000, "frintn", int_arm64_neon_frintn>;
2323 defm FRINTP : SIMDTwoVectorFP<0, 1, 0b11000, "frintp", fceil>;
2324 defm FRINTX : SIMDTwoVectorFP<1, 0, 0b11001, "frintx", frint>;
2325 defm FRINTZ : SIMDTwoVectorFP<0, 1, 0b11001, "frintz", ftrunc>;
2326 defm FRSQRTE: SIMDTwoVectorFP<1, 1, 0b11101, "frsqrte", int_arm64_neon_frsqrte>;
2327 defm FSQRT : SIMDTwoVectorFP<1, 1, 0b11111, "fsqrt", fsqrt>;
2328 defm NEG : SIMDTwoVectorBHSD<1, 0b01011, "neg",
2329 UnOpFrag<(sub immAllZerosV, node:$LHS)> >;
2330 defm NOT : SIMDTwoVectorB<1, 0b00, 0b00101, "not", vnot>;
2331 // Aliases for MVN -> NOT.
2332 def : InstAlias<"mvn{ $Vd.8b, $Vn.8b|.8b $Vd, $Vn}",
2333 (NOTv8i8 V64:$Vd, V64:$Vn)>;
2334 def : InstAlias<"mvn{ $Vd.16b, $Vn.16b|.16b $Vd, $Vn}",
2335 (NOTv16i8 V128:$Vd, V128:$Vn)>;
2337 def : Pat<(ARM64neg (v8i8 V64:$Rn)), (NEGv8i8 V64:$Rn)>;
2338 def : Pat<(ARM64neg (v16i8 V128:$Rn)), (NEGv16i8 V128:$Rn)>;
2339 def : Pat<(ARM64neg (v4i16 V64:$Rn)), (NEGv4i16 V64:$Rn)>;
2340 def : Pat<(ARM64neg (v8i16 V128:$Rn)), (NEGv8i16 V128:$Rn)>;
2341 def : Pat<(ARM64neg (v2i32 V64:$Rn)), (NEGv2i32 V64:$Rn)>;
2342 def : Pat<(ARM64neg (v4i32 V128:$Rn)), (NEGv4i32 V128:$Rn)>;
2343 def : Pat<(ARM64neg (v2i64 V128:$Rn)), (NEGv2i64 V128:$Rn)>;
2345 def : Pat<(ARM64not (v8i8 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2346 def : Pat<(ARM64not (v16i8 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2347 def : Pat<(ARM64not (v4i16 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2348 def : Pat<(ARM64not (v8i16 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2349 def : Pat<(ARM64not (v2i32 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2350 def : Pat<(ARM64not (v1i64 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2351 def : Pat<(ARM64not (v4i32 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2352 def : Pat<(ARM64not (v2i64 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2354 def : Pat<(vnot (v4i16 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2355 def : Pat<(vnot (v8i16 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2356 def : Pat<(vnot (v2i32 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2357 def : Pat<(vnot (v4i32 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2358 def : Pat<(vnot (v2i64 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2360 defm RBIT : SIMDTwoVectorB<1, 0b01, 0b00101, "rbit", int_arm64_neon_rbit>;
2361 defm REV16 : SIMDTwoVectorB<0, 0b00, 0b00001, "rev16", ARM64rev16>;
2362 defm REV32 : SIMDTwoVectorBH<1, 0b00000, "rev32", ARM64rev32>;
2363 defm REV64 : SIMDTwoVectorBHS<0, 0b00000, "rev64", ARM64rev64>;
2364 defm SADALP : SIMDLongTwoVectorTied<0, 0b00110, "sadalp",
2365 BinOpFrag<(add node:$LHS, (int_arm64_neon_saddlp node:$RHS))> >;
2366 defm SADDLP : SIMDLongTwoVector<0, 0b00010, "saddlp", int_arm64_neon_saddlp>;
2367 defm SCVTF : SIMDTwoVectorIntToFP<0, 0, 0b11101, "scvtf", sint_to_fp>;
2368 defm SHLL : SIMDVectorLShiftLongBySizeBHS;
2369 defm SQABS : SIMDTwoVectorBHSD<0, 0b00111, "sqabs", int_arm64_neon_sqabs>;
2370 defm SQNEG : SIMDTwoVectorBHSD<1, 0b00111, "sqneg", int_arm64_neon_sqneg>;
2371 defm SQXTN : SIMDMixedTwoVector<0, 0b10100, "sqxtn", int_arm64_neon_sqxtn>;
2372 defm SQXTUN : SIMDMixedTwoVector<1, 0b10010, "sqxtun", int_arm64_neon_sqxtun>;
2373 defm SUQADD : SIMDTwoVectorBHSDTied<0, 0b00011, "suqadd",int_arm64_neon_suqadd>;
2374 defm UADALP : SIMDLongTwoVectorTied<1, 0b00110, "uadalp",
2375 BinOpFrag<(add node:$LHS, (int_arm64_neon_uaddlp node:$RHS))> >;
2376 defm UADDLP : SIMDLongTwoVector<1, 0b00010, "uaddlp",
2377 int_arm64_neon_uaddlp>;
2378 defm UCVTF : SIMDTwoVectorIntToFP<1, 0, 0b11101, "ucvtf", uint_to_fp>;
2379 defm UQXTN : SIMDMixedTwoVector<1, 0b10100, "uqxtn", int_arm64_neon_uqxtn>;
2380 defm URECPE : SIMDTwoVectorS<0, 1, 0b11100, "urecpe", int_arm64_neon_urecpe>;
2381 defm URSQRTE: SIMDTwoVectorS<1, 1, 0b11100, "ursqrte", int_arm64_neon_ursqrte>;
2382 defm USQADD : SIMDTwoVectorBHSDTied<1, 0b00011, "usqadd",int_arm64_neon_usqadd>;
2383 defm XTN : SIMDMixedTwoVector<0, 0b10010, "xtn", trunc>;
2385 def : Pat<(v2f32 (ARM64rev64 V64:$Rn)), (REV64v2i32 V64:$Rn)>;
2386 def : Pat<(v4f32 (ARM64rev64 V128:$Rn)), (REV64v4i32 V128:$Rn)>;
2388 // Patterns for vector long shift (by element width). These need to match all
2389 // three of zext, sext and anyext so it's easier to pull the patterns out of the
2391 multiclass SIMDVectorLShiftLongBySizeBHSPats<SDPatternOperator ext> {
2392 def : Pat<(ARM64vshl (v8i16 (ext (v8i8 V64:$Rn))), (i32 8)),
2393 (SHLLv8i8 V64:$Rn)>;
2394 def : Pat<(ARM64vshl (v8i16 (ext (extract_high_v16i8 V128:$Rn))), (i32 8)),
2395 (SHLLv16i8 V128:$Rn)>;
2396 def : Pat<(ARM64vshl (v4i32 (ext (v4i16 V64:$Rn))), (i32 16)),
2397 (SHLLv4i16 V64:$Rn)>;
2398 def : Pat<(ARM64vshl (v4i32 (ext (extract_high_v8i16 V128:$Rn))), (i32 16)),
2399 (SHLLv8i16 V128:$Rn)>;
2400 def : Pat<(ARM64vshl (v2i64 (ext (v2i32 V64:$Rn))), (i32 32)),
2401 (SHLLv2i32 V64:$Rn)>;
2402 def : Pat<(ARM64vshl (v2i64 (ext (extract_high_v4i32 V128:$Rn))), (i32 32)),
2403 (SHLLv4i32 V128:$Rn)>;
2406 defm : SIMDVectorLShiftLongBySizeBHSPats<anyext>;
2407 defm : SIMDVectorLShiftLongBySizeBHSPats<zext>;
2408 defm : SIMDVectorLShiftLongBySizeBHSPats<sext>;
2410 //===----------------------------------------------------------------------===//
2411 // Advanced SIMD three vector instructions.
2412 //===----------------------------------------------------------------------===//
2414 defm ADD : SIMDThreeSameVector<0, 0b10000, "add", add>;
2415 defm ADDP : SIMDThreeSameVector<0, 0b10111, "addp", int_arm64_neon_addp>;
2416 defm CMEQ : SIMDThreeSameVector<1, 0b10001, "cmeq", ARM64cmeq>;
2417 defm CMGE : SIMDThreeSameVector<0, 0b00111, "cmge", ARM64cmge>;
2418 defm CMGT : SIMDThreeSameVector<0, 0b00110, "cmgt", ARM64cmgt>;
2419 defm CMHI : SIMDThreeSameVector<1, 0b00110, "cmhi", ARM64cmhi>;
2420 defm CMHS : SIMDThreeSameVector<1, 0b00111, "cmhs", ARM64cmhs>;
2421 defm CMTST : SIMDThreeSameVector<0, 0b10001, "cmtst", ARM64cmtst>;
2422 defm FABD : SIMDThreeSameVectorFP<1,1,0b11010,"fabd", int_arm64_neon_fabd>;
2423 defm FACGE : SIMDThreeSameVectorFPCmp<1,0,0b11101,"facge",int_arm64_neon_facge>;
2424 defm FACGT : SIMDThreeSameVectorFPCmp<1,1,0b11101,"facgt",int_arm64_neon_facgt>;
2425 defm FADDP : SIMDThreeSameVectorFP<1,0,0b11010,"faddp",int_arm64_neon_addp>;
2426 defm FADD : SIMDThreeSameVectorFP<0,0,0b11010,"fadd", fadd>;
2427 defm FCMEQ : SIMDThreeSameVectorFPCmp<0, 0, 0b11100, "fcmeq", ARM64fcmeq>;
2428 defm FCMGE : SIMDThreeSameVectorFPCmp<1, 0, 0b11100, "fcmge", ARM64fcmge>;
2429 defm FCMGT : SIMDThreeSameVectorFPCmp<1, 1, 0b11100, "fcmgt", ARM64fcmgt>;
2430 defm FDIV : SIMDThreeSameVectorFP<1,0,0b11111,"fdiv", fdiv>;
2431 defm FMAXNMP : SIMDThreeSameVectorFP<1,0,0b11000,"fmaxnmp", int_arm64_neon_fmaxnmp>;
2432 defm FMAXNM : SIMDThreeSameVectorFP<0,0,0b11000,"fmaxnm", int_arm64_neon_fmaxnm>;
2433 defm FMAXP : SIMDThreeSameVectorFP<1,0,0b11110,"fmaxp", int_arm64_neon_fmaxp>;
2434 defm FMAX : SIMDThreeSameVectorFP<0,0,0b11110,"fmax", ARM64fmax>;
2435 defm FMINNMP : SIMDThreeSameVectorFP<1,1,0b11000,"fminnmp", int_arm64_neon_fminnmp>;
2436 defm FMINNM : SIMDThreeSameVectorFP<0,1,0b11000,"fminnm", int_arm64_neon_fminnm>;
2437 defm FMINP : SIMDThreeSameVectorFP<1,1,0b11110,"fminp", int_arm64_neon_fminp>;
2438 defm FMIN : SIMDThreeSameVectorFP<0,1,0b11110,"fmin", ARM64fmin>;
2440 // NOTE: The operands of the PatFrag are reordered on FMLA/FMLS because the
2441 // instruction expects the addend first, while the fma intrinsic puts it last.
2442 defm FMLA : SIMDThreeSameVectorFPTied<0, 0, 0b11001, "fmla",
2443 TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)> >;
2444 defm FMLS : SIMDThreeSameVectorFPTied<0, 1, 0b11001, "fmls",
2445 TriOpFrag<(fma node:$MHS, (fneg node:$RHS), node:$LHS)> >;
2447 // The following def pats catch the case where the LHS of an FMA is negated.
2448 // The TriOpFrag above catches the case where the middle operand is negated.
2449 def : Pat<(v2f32 (fma (fneg V64:$Rn), V64:$Rm, V64:$Rd)),
2450 (FMLSv2f32 V64:$Rd, V64:$Rn, V64:$Rm)>;
2452 def : Pat<(v4f32 (fma (fneg V128:$Rn), V128:$Rm, V128:$Rd)),
2453 (FMLSv4f32 V128:$Rd, V128:$Rn, V128:$Rm)>;
2455 def : Pat<(v2f64 (fma (fneg V128:$Rn), V128:$Rm, V128:$Rd)),
2456 (FMLSv2f64 V128:$Rd, V128:$Rn, V128:$Rm)>;
2458 defm FMULX : SIMDThreeSameVectorFP<0,0,0b11011,"fmulx", int_arm64_neon_fmulx>;
2459 defm FMUL : SIMDThreeSameVectorFP<1,0,0b11011,"fmul", fmul>;
2460 defm FRECPS : SIMDThreeSameVectorFP<0,0,0b11111,"frecps", int_arm64_neon_frecps>;
2461 defm FRSQRTS : SIMDThreeSameVectorFP<0,1,0b11111,"frsqrts", int_arm64_neon_frsqrts>;
2462 defm FSUB : SIMDThreeSameVectorFP<0,1,0b11010,"fsub", fsub>;
2463 defm MLA : SIMDThreeSameVectorBHSTied<0, 0b10010, "mla",
2464 TriOpFrag<(add node:$LHS, (mul node:$MHS, node:$RHS))> >;
2465 defm MLS : SIMDThreeSameVectorBHSTied<1, 0b10010, "mls",
2466 TriOpFrag<(sub node:$LHS, (mul node:$MHS, node:$RHS))> >;
2467 defm MUL : SIMDThreeSameVectorBHS<0, 0b10011, "mul", mul>;
2468 defm PMUL : SIMDThreeSameVectorB<1, 0b10011, "pmul", int_arm64_neon_pmul>;
2469 defm SABA : SIMDThreeSameVectorBHSTied<0, 0b01111, "saba",
2470 TriOpFrag<(add node:$LHS, (int_arm64_neon_sabd node:$MHS, node:$RHS))> >;
2471 defm SABD : SIMDThreeSameVectorBHS<0,0b01110,"sabd", int_arm64_neon_sabd>;
2472 defm SHADD : SIMDThreeSameVectorBHS<0,0b00000,"shadd", int_arm64_neon_shadd>;
2473 defm SHSUB : SIMDThreeSameVectorBHS<0,0b00100,"shsub", int_arm64_neon_shsub>;
2474 defm SMAXP : SIMDThreeSameVectorBHS<0,0b10100,"smaxp", int_arm64_neon_smaxp>;
2475 defm SMAX : SIMDThreeSameVectorBHS<0,0b01100,"smax", int_arm64_neon_smax>;
2476 defm SMINP : SIMDThreeSameVectorBHS<0,0b10101,"sminp", int_arm64_neon_sminp>;
2477 defm SMIN : SIMDThreeSameVectorBHS<0,0b01101,"smin", int_arm64_neon_smin>;
2478 defm SQADD : SIMDThreeSameVector<0,0b00001,"sqadd", int_arm64_neon_sqadd>;
2479 defm SQDMULH : SIMDThreeSameVectorHS<0,0b10110,"sqdmulh",int_arm64_neon_sqdmulh>;
2480 defm SQRDMULH : SIMDThreeSameVectorHS<1,0b10110,"sqrdmulh",int_arm64_neon_sqrdmulh>;
2481 defm SQRSHL : SIMDThreeSameVector<0,0b01011,"sqrshl", int_arm64_neon_sqrshl>;
2482 defm SQSHL : SIMDThreeSameVector<0,0b01001,"sqshl", int_arm64_neon_sqshl>;
2483 defm SQSUB : SIMDThreeSameVector<0,0b00101,"sqsub", int_arm64_neon_sqsub>;
2484 defm SRHADD : SIMDThreeSameVectorBHS<0,0b00010,"srhadd",int_arm64_neon_srhadd>;
2485 defm SRSHL : SIMDThreeSameVector<0,0b01010,"srshl", int_arm64_neon_srshl>;
2486 defm SSHL : SIMDThreeSameVector<0,0b01000,"sshl", int_arm64_neon_sshl>;
2487 defm SUB : SIMDThreeSameVector<1,0b10000,"sub", sub>;
2488 defm UABA : SIMDThreeSameVectorBHSTied<1, 0b01111, "uaba",
2489 TriOpFrag<(add node:$LHS, (int_arm64_neon_uabd node:$MHS, node:$RHS))> >;
2490 defm UABD : SIMDThreeSameVectorBHS<1,0b01110,"uabd", int_arm64_neon_uabd>;
2491 defm UHADD : SIMDThreeSameVectorBHS<1,0b00000,"uhadd", int_arm64_neon_uhadd>;
2492 defm UHSUB : SIMDThreeSameVectorBHS<1,0b00100,"uhsub", int_arm64_neon_uhsub>;
2493 defm UMAXP : SIMDThreeSameVectorBHS<1,0b10100,"umaxp", int_arm64_neon_umaxp>;
2494 defm UMAX : SIMDThreeSameVectorBHS<1,0b01100,"umax", int_arm64_neon_umax>;
2495 defm UMINP : SIMDThreeSameVectorBHS<1,0b10101,"uminp", int_arm64_neon_uminp>;
2496 defm UMIN : SIMDThreeSameVectorBHS<1,0b01101,"umin", int_arm64_neon_umin>;
2497 defm UQADD : SIMDThreeSameVector<1,0b00001,"uqadd", int_arm64_neon_uqadd>;
2498 defm UQRSHL : SIMDThreeSameVector<1,0b01011,"uqrshl", int_arm64_neon_uqrshl>;
2499 defm UQSHL : SIMDThreeSameVector<1,0b01001,"uqshl", int_arm64_neon_uqshl>;
2500 defm UQSUB : SIMDThreeSameVector<1,0b00101,"uqsub", int_arm64_neon_uqsub>;
2501 defm URHADD : SIMDThreeSameVectorBHS<1,0b00010,"urhadd", int_arm64_neon_urhadd>;
2502 defm URSHL : SIMDThreeSameVector<1,0b01010,"urshl", int_arm64_neon_urshl>;
2503 defm USHL : SIMDThreeSameVector<1,0b01000,"ushl", int_arm64_neon_ushl>;
2505 defm AND : SIMDLogicalThreeVector<0, 0b00, "and", and>;
2506 defm BIC : SIMDLogicalThreeVector<0, 0b01, "bic",
2507 BinOpFrag<(and node:$LHS, (vnot node:$RHS))> >;
2508 defm BIF : SIMDLogicalThreeVector<1, 0b11, "bif">;
2509 defm BIT : SIMDLogicalThreeVectorTied<1, 0b10, "bit", ARM64bit>;
2510 defm BSL : SIMDLogicalThreeVectorTied<1, 0b01, "bsl",
2511 TriOpFrag<(or (and node:$LHS, node:$MHS), (and (vnot node:$LHS), node:$RHS))>>;
2512 defm EOR : SIMDLogicalThreeVector<1, 0b00, "eor", xor>;
2513 defm ORN : SIMDLogicalThreeVector<0, 0b11, "orn",
2514 BinOpFrag<(or node:$LHS, (vnot node:$RHS))> >;
2515 defm ORR : SIMDLogicalThreeVector<0, 0b10, "orr", or>;
2517 def : Pat<(ARM64bsl (v8i8 V64:$Rd), V64:$Rn, V64:$Rm),
2518 (BSLv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
2519 def : Pat<(ARM64bsl (v4i16 V64:$Rd), V64:$Rn, V64:$Rm),
2520 (BSLv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
2521 def : Pat<(ARM64bsl (v2i32 V64:$Rd), V64:$Rn, V64:$Rm),
2522 (BSLv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
2523 def : Pat<(ARM64bsl (v1i64 V64:$Rd), V64:$Rn, V64:$Rm),
2524 (BSLv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
2526 def : Pat<(ARM64bsl (v16i8 V128:$Rd), V128:$Rn, V128:$Rm),
2527 (BSLv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
2528 def : Pat<(ARM64bsl (v8i16 V128:$Rd), V128:$Rn, V128:$Rm),
2529 (BSLv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
2530 def : Pat<(ARM64bsl (v4i32 V128:$Rd), V128:$Rn, V128:$Rm),
2531 (BSLv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
2532 def : Pat<(ARM64bsl (v2i64 V128:$Rd), V128:$Rn, V128:$Rm),
2533 (BSLv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
2535 def : InstAlias<"mov{\t$dst.16b, $src.16b|.16b\t$dst, $src}",
2536 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 1>;
2537 def : InstAlias<"mov{\t$dst.8h, $src.8h|.8h\t$dst, $src}",
2538 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2539 def : InstAlias<"mov{\t$dst.4s, $src.4s|.4s\t$dst, $src}",
2540 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2541 def : InstAlias<"mov{\t$dst.2d, $src.2d|.2d\t$dst, $src}",
2542 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2544 def : InstAlias<"mov{\t$dst.8b, $src.8b|.8b\t$dst, $src}",
2545 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 1>;
2546 def : InstAlias<"mov{\t$dst.4h, $src.4h|.4h\t$dst, $src}",
2547 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2548 def : InstAlias<"mov{\t$dst.2s, $src.2s|.2s\t$dst, $src}",
2549 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2550 def : InstAlias<"mov{\t$dst.1d, $src.1d|.1d\t$dst, $src}",
2551 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2553 def : InstAlias<"{cmls\t$dst.8b, $src1.8b, $src2.8b" #
2554 "|cmls.8b\t$dst, $src1, $src2}",
2555 (CMHSv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2556 def : InstAlias<"{cmls\t$dst.16b, $src1.16b, $src2.16b" #
2557 "|cmls.16b\t$dst, $src1, $src2}",
2558 (CMHSv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2559 def : InstAlias<"{cmls\t$dst.4h, $src1.4h, $src2.4h" #
2560 "|cmls.4h\t$dst, $src1, $src2}",
2561 (CMHSv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2562 def : InstAlias<"{cmls\t$dst.8h, $src1.8h, $src2.8h" #
2563 "|cmls.8h\t$dst, $src1, $src2}",
2564 (CMHSv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2565 def : InstAlias<"{cmls\t$dst.2s, $src1.2s, $src2.2s" #
2566 "|cmls.2s\t$dst, $src1, $src2}",
2567 (CMHSv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2568 def : InstAlias<"{cmls\t$dst.4s, $src1.4s, $src2.4s" #
2569 "|cmls.4s\t$dst, $src1, $src2}",
2570 (CMHSv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2571 def : InstAlias<"{cmls\t$dst.2d, $src1.2d, $src2.2d" #
2572 "|cmls.2d\t$dst, $src1, $src2}",
2573 (CMHSv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2575 def : InstAlias<"{cmlo\t$dst.8b, $src1.8b, $src2.8b" #
2576 "|cmlo.8b\t$dst, $src1, $src2}",
2577 (CMHIv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2578 def : InstAlias<"{cmlo\t$dst.16b, $src1.16b, $src2.16b" #
2579 "|cmlo.16b\t$dst, $src1, $src2}",
2580 (CMHIv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2581 def : InstAlias<"{cmlo\t$dst.4h, $src1.4h, $src2.4h" #
2582 "|cmlo.4h\t$dst, $src1, $src2}",
2583 (CMHIv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2584 def : InstAlias<"{cmlo\t$dst.8h, $src1.8h, $src2.8h" #
2585 "|cmlo.8h\t$dst, $src1, $src2}",
2586 (CMHIv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2587 def : InstAlias<"{cmlo\t$dst.2s, $src1.2s, $src2.2s" #
2588 "|cmlo.2s\t$dst, $src1, $src2}",
2589 (CMHIv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2590 def : InstAlias<"{cmlo\t$dst.4s, $src1.4s, $src2.4s" #
2591 "|cmlo.4s\t$dst, $src1, $src2}",
2592 (CMHIv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2593 def : InstAlias<"{cmlo\t$dst.2d, $src1.2d, $src2.2d" #
2594 "|cmlo.2d\t$dst, $src1, $src2}",
2595 (CMHIv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2597 def : InstAlias<"{cmle\t$dst.8b, $src1.8b, $src2.8b" #
2598 "|cmle.8b\t$dst, $src1, $src2}",
2599 (CMGEv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2600 def : InstAlias<"{cmle\t$dst.16b, $src1.16b, $src2.16b" #
2601 "|cmle.16b\t$dst, $src1, $src2}",
2602 (CMGEv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2603 def : InstAlias<"{cmle\t$dst.4h, $src1.4h, $src2.4h" #
2604 "|cmle.4h\t$dst, $src1, $src2}",
2605 (CMGEv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2606 def : InstAlias<"{cmle\t$dst.8h, $src1.8h, $src2.8h" #
2607 "|cmle.8h\t$dst, $src1, $src2}",
2608 (CMGEv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2609 def : InstAlias<"{cmle\t$dst.2s, $src1.2s, $src2.2s" #
2610 "|cmle.2s\t$dst, $src1, $src2}",
2611 (CMGEv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2612 def : InstAlias<"{cmle\t$dst.4s, $src1.4s, $src2.4s" #
2613 "|cmle.4s\t$dst, $src1, $src2}",
2614 (CMGEv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2615 def : InstAlias<"{cmle\t$dst.2d, $src1.2d, $src2.2d" #
2616 "|cmle.2d\t$dst, $src1, $src2}",
2617 (CMGEv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2619 def : InstAlias<"{cmlt\t$dst.8b, $src1.8b, $src2.8b" #
2620 "|cmlt.8b\t$dst, $src1, $src2}",
2621 (CMGTv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2622 def : InstAlias<"{cmlt\t$dst.16b, $src1.16b, $src2.16b" #
2623 "|cmlt.16b\t$dst, $src1, $src2}",
2624 (CMGTv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2625 def : InstAlias<"{cmlt\t$dst.4h, $src1.4h, $src2.4h" #
2626 "|cmlt.4h\t$dst, $src1, $src2}",
2627 (CMGTv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2628 def : InstAlias<"{cmlt\t$dst.8h, $src1.8h, $src2.8h" #
2629 "|cmlt.8h\t$dst, $src1, $src2}",
2630 (CMGTv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2631 def : InstAlias<"{cmlt\t$dst.2s, $src1.2s, $src2.2s" #
2632 "|cmlt.2s\t$dst, $src1, $src2}",
2633 (CMGTv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2634 def : InstAlias<"{cmlt\t$dst.4s, $src1.4s, $src2.4s" #
2635 "|cmlt.4s\t$dst, $src1, $src2}",
2636 (CMGTv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2637 def : InstAlias<"{cmlt\t$dst.2d, $src1.2d, $src2.2d" #
2638 "|cmlt.2d\t$dst, $src1, $src2}",
2639 (CMGTv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2641 def : InstAlias<"{fcmle\t$dst.2s, $src1.2s, $src2.2s" #
2642 "|fcmle.2s\t$dst, $src1, $src2}",
2643 (FCMGEv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
2644 def : InstAlias<"{fcmle\t$dst.4s, $src1.4s, $src2.4s" #
2645 "|fcmle.4s\t$dst, $src1, $src2}",
2646 (FCMGEv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
2647 def : InstAlias<"{fcmle\t$dst.2d, $src1.2d, $src2.2d" #
2648 "|fcmle.2d\t$dst, $src1, $src2}",
2649 (FCMGEv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
2651 def : InstAlias<"{fcmlt\t$dst.2s, $src1.2s, $src2.2s" #
2652 "|fcmlt.2s\t$dst, $src1, $src2}",
2653 (FCMGTv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
2654 def : InstAlias<"{fcmlt\t$dst.4s, $src1.4s, $src2.4s" #
2655 "|fcmlt.4s\t$dst, $src1, $src2}",
2656 (FCMGTv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
2657 def : InstAlias<"{fcmlt\t$dst.2d, $src1.2d, $src2.2d" #
2658 "|fcmlt.2d\t$dst, $src1, $src2}",
2659 (FCMGTv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
2661 def : InstAlias<"{facle\t$dst.2s, $src1.2s, $src2.2s" #
2662 "|facle.2s\t$dst, $src1, $src2}",
2663 (FACGEv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
2664 def : InstAlias<"{facle\t$dst.4s, $src1.4s, $src2.4s" #
2665 "|facle.4s\t$dst, $src1, $src2}",
2666 (FACGEv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
2667 def : InstAlias<"{facle\t$dst.2d, $src1.2d, $src2.2d" #
2668 "|facle.2d\t$dst, $src1, $src2}",
2669 (FACGEv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
2671 def : InstAlias<"{faclt\t$dst.2s, $src1.2s, $src2.2s" #
2672 "|faclt.2s\t$dst, $src1, $src2}",
2673 (FACGTv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
2674 def : InstAlias<"{faclt\t$dst.4s, $src1.4s, $src2.4s" #
2675 "|faclt.4s\t$dst, $src1, $src2}",
2676 (FACGTv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
2677 def : InstAlias<"{faclt\t$dst.2d, $src1.2d, $src2.2d" #
2678 "|faclt.2d\t$dst, $src1, $src2}",
2679 (FACGTv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
2681 //===----------------------------------------------------------------------===//
2682 // Advanced SIMD three scalar instructions.
2683 //===----------------------------------------------------------------------===//
2685 defm ADD : SIMDThreeScalarD<0, 0b10000, "add", add>;
2686 defm CMEQ : SIMDThreeScalarD<1, 0b10001, "cmeq", ARM64cmeq>;
2687 defm CMGE : SIMDThreeScalarD<0, 0b00111, "cmge", ARM64cmge>;
2688 defm CMGT : SIMDThreeScalarD<0, 0b00110, "cmgt", ARM64cmgt>;
2689 defm CMHI : SIMDThreeScalarD<1, 0b00110, "cmhi", ARM64cmhi>;
2690 defm CMHS : SIMDThreeScalarD<1, 0b00111, "cmhs", ARM64cmhs>;
2691 defm CMTST : SIMDThreeScalarD<0, 0b10001, "cmtst", ARM64cmtst>;
2692 defm FABD : SIMDThreeScalarSD<1, 1, 0b11010, "fabd", int_arm64_sisd_fabd>;
2693 def : Pat<(v1f64 (int_arm64_neon_fabd (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2694 (FABD64 FPR64:$Rn, FPR64:$Rm)>;
2695 defm FACGE : SIMDThreeScalarFPCmp<1, 0, 0b11101, "facge",
2696 int_arm64_neon_facge>;
2697 defm FACGT : SIMDThreeScalarFPCmp<1, 1, 0b11101, "facgt",
2698 int_arm64_neon_facgt>;
2699 defm FCMEQ : SIMDThreeScalarFPCmp<0, 0, 0b11100, "fcmeq", ARM64fcmeq>;
2700 defm FCMGE : SIMDThreeScalarFPCmp<1, 0, 0b11100, "fcmge", ARM64fcmge>;
2701 defm FCMGT : SIMDThreeScalarFPCmp<1, 1, 0b11100, "fcmgt", ARM64fcmgt>;
2702 defm FMULX : SIMDThreeScalarSD<0, 0, 0b11011, "fmulx", int_arm64_neon_fmulx>;
2703 defm FRECPS : SIMDThreeScalarSD<0, 0, 0b11111, "frecps", int_arm64_neon_frecps>;
2704 defm FRSQRTS : SIMDThreeScalarSD<0, 1, 0b11111, "frsqrts", int_arm64_neon_frsqrts>;
2705 defm SQADD : SIMDThreeScalarBHSD<0, 0b00001, "sqadd", int_arm64_neon_sqadd>;
2706 defm SQDMULH : SIMDThreeScalarHS< 0, 0b10110, "sqdmulh", int_arm64_neon_sqdmulh>;
2707 defm SQRDMULH : SIMDThreeScalarHS< 1, 0b10110, "sqrdmulh", int_arm64_neon_sqrdmulh>;
2708 defm SQRSHL : SIMDThreeScalarBHSD<0, 0b01011, "sqrshl",int_arm64_neon_sqrshl>;
2709 defm SQSHL : SIMDThreeScalarBHSD<0, 0b01001, "sqshl", int_arm64_neon_sqshl>;
2710 defm SQSUB : SIMDThreeScalarBHSD<0, 0b00101, "sqsub", int_arm64_neon_sqsub>;
2711 defm SRSHL : SIMDThreeScalarD< 0, 0b01010, "srshl", int_arm64_neon_srshl>;
2712 defm SSHL : SIMDThreeScalarD< 0, 0b01000, "sshl", int_arm64_neon_sshl>;
2713 defm SUB : SIMDThreeScalarD< 1, 0b10000, "sub", sub>;
2714 defm UQADD : SIMDThreeScalarBHSD<1, 0b00001, "uqadd", int_arm64_neon_uqadd>;
2715 defm UQRSHL : SIMDThreeScalarBHSD<1, 0b01011, "uqrshl",int_arm64_neon_uqrshl>;
2716 defm UQSHL : SIMDThreeScalarBHSD<1, 0b01001, "uqshl", int_arm64_neon_uqshl>;
2717 defm UQSUB : SIMDThreeScalarBHSD<1, 0b00101, "uqsub", int_arm64_neon_uqsub>;
2718 defm URSHL : SIMDThreeScalarD< 1, 0b01010, "urshl", int_arm64_neon_urshl>;
2719 defm USHL : SIMDThreeScalarD< 1, 0b01000, "ushl", int_arm64_neon_ushl>;
2721 def : InstAlias<"cmls $dst, $src1, $src2",
2722 (CMHSv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
2723 def : InstAlias<"cmle $dst, $src1, $src2",
2724 (CMGEv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
2725 def : InstAlias<"cmlo $dst, $src1, $src2",
2726 (CMHIv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
2727 def : InstAlias<"cmlt $dst, $src1, $src2",
2728 (CMGTv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
2729 def : InstAlias<"fcmle $dst, $src1, $src2",
2730 (FCMGE32 FPR32:$dst, FPR32:$src2, FPR32:$src1), 0>;
2731 def : InstAlias<"fcmle $dst, $src1, $src2",
2732 (FCMGE64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
2733 def : InstAlias<"fcmlt $dst, $src1, $src2",
2734 (FCMGT32 FPR32:$dst, FPR32:$src2, FPR32:$src1), 0>;
2735 def : InstAlias<"fcmlt $dst, $src1, $src2",
2736 (FCMGT64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
2737 def : InstAlias<"facle $dst, $src1, $src2",
2738 (FACGE32 FPR32:$dst, FPR32:$src2, FPR32:$src1), 0>;
2739 def : InstAlias<"facle $dst, $src1, $src2",
2740 (FACGE64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
2741 def : InstAlias<"faclt $dst, $src1, $src2",
2742 (FACGT32 FPR32:$dst, FPR32:$src2, FPR32:$src1), 0>;
2743 def : InstAlias<"faclt $dst, $src1, $src2",
2744 (FACGT64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
2746 //===----------------------------------------------------------------------===//
2747 // Advanced SIMD three scalar instructions (mixed operands).
2748 //===----------------------------------------------------------------------===//
2749 defm SQDMULL : SIMDThreeScalarMixedHS<0, 0b11010, "sqdmull",
2750 int_arm64_neon_sqdmulls_scalar>;
2751 defm SQDMLAL : SIMDThreeScalarMixedTiedHS<0, 0b10010, "sqdmlal">;
2752 defm SQDMLSL : SIMDThreeScalarMixedTiedHS<0, 0b10110, "sqdmlsl">;
2754 def : Pat<(i64 (int_arm64_neon_sqadd (i64 FPR64:$Rd),
2755 (i64 (int_arm64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
2756 (i32 FPR32:$Rm))))),
2757 (SQDMLALi32 FPR64:$Rd, FPR32:$Rn, FPR32:$Rm)>;
2758 def : Pat<(i64 (int_arm64_neon_sqsub (i64 FPR64:$Rd),
2759 (i64 (int_arm64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
2760 (i32 FPR32:$Rm))))),
2761 (SQDMLSLi32 FPR64:$Rd, FPR32:$Rn, FPR32:$Rm)>;
2763 //===----------------------------------------------------------------------===//
2764 // Advanced SIMD two scalar instructions.
2765 //===----------------------------------------------------------------------===//
2767 defm ABS : SIMDTwoScalarD< 0, 0b01011, "abs", int_arm64_neon_abs>;
2768 defm CMEQ : SIMDCmpTwoScalarD< 0, 0b01001, "cmeq", ARM64cmeqz>;
2769 defm CMGE : SIMDCmpTwoScalarD< 1, 0b01000, "cmge", ARM64cmgez>;
2770 defm CMGT : SIMDCmpTwoScalarD< 0, 0b01000, "cmgt", ARM64cmgtz>;
2771 defm CMLE : SIMDCmpTwoScalarD< 1, 0b01001, "cmle", ARM64cmlez>;
2772 defm CMLT : SIMDCmpTwoScalarD< 0, 0b01010, "cmlt", ARM64cmltz>;
2773 defm FCMEQ : SIMDCmpTwoScalarSD<0, 1, 0b01101, "fcmeq", ARM64fcmeqz>;
2774 defm FCMGE : SIMDCmpTwoScalarSD<1, 1, 0b01100, "fcmge", ARM64fcmgez>;
2775 defm FCMGT : SIMDCmpTwoScalarSD<0, 1, 0b01100, "fcmgt", ARM64fcmgtz>;
2776 defm FCMLE : SIMDCmpTwoScalarSD<1, 1, 0b01101, "fcmle", ARM64fcmlez>;
2777 defm FCMLT : SIMDCmpTwoScalarSD<0, 1, 0b01110, "fcmlt", ARM64fcmltz>;
2778 defm FCVTAS : SIMDTwoScalarSD< 0, 0, 0b11100, "fcvtas">;
2779 defm FCVTAU : SIMDTwoScalarSD< 1, 0, 0b11100, "fcvtau">;
2780 defm FCVTMS : SIMDTwoScalarSD< 0, 0, 0b11011, "fcvtms">;
2781 defm FCVTMU : SIMDTwoScalarSD< 1, 0, 0b11011, "fcvtmu">;
2782 defm FCVTNS : SIMDTwoScalarSD< 0, 0, 0b11010, "fcvtns">;
2783 defm FCVTNU : SIMDTwoScalarSD< 1, 0, 0b11010, "fcvtnu">;
2784 defm FCVTPS : SIMDTwoScalarSD< 0, 1, 0b11010, "fcvtps">;
2785 defm FCVTPU : SIMDTwoScalarSD< 1, 1, 0b11010, "fcvtpu">;
2786 def FCVTXNv1i64 : SIMDInexactCvtTwoScalar<0b10110, "fcvtxn">;
2787 defm FCVTZS : SIMDTwoScalarSD< 0, 1, 0b11011, "fcvtzs">;
2788 defm FCVTZU : SIMDTwoScalarSD< 1, 1, 0b11011, "fcvtzu">;
2789 defm FRECPE : SIMDTwoScalarSD< 0, 1, 0b11101, "frecpe">;
2790 defm FRECPX : SIMDTwoScalarSD< 0, 1, 0b11111, "frecpx">;
2791 defm FRSQRTE : SIMDTwoScalarSD< 1, 1, 0b11101, "frsqrte">;
2792 defm NEG : SIMDTwoScalarD< 1, 0b01011, "neg",
2793 UnOpFrag<(sub immAllZerosV, node:$LHS)> >;
2794 defm SCVTF : SIMDTwoScalarCVTSD< 0, 0, 0b11101, "scvtf", ARM64sitof>;
2795 defm SQABS : SIMDTwoScalarBHSD< 0, 0b00111, "sqabs", int_arm64_neon_sqabs>;
2796 defm SQNEG : SIMDTwoScalarBHSD< 1, 0b00111, "sqneg", int_arm64_neon_sqneg>;
2797 defm SQXTN : SIMDTwoScalarMixedBHS< 0, 0b10100, "sqxtn", int_arm64_neon_scalar_sqxtn>;
2798 defm SQXTUN : SIMDTwoScalarMixedBHS< 1, 0b10010, "sqxtun", int_arm64_neon_scalar_sqxtun>;
2799 defm SUQADD : SIMDTwoScalarBHSDTied< 0, 0b00011, "suqadd",
2800 int_arm64_neon_suqadd>;
2801 defm UCVTF : SIMDTwoScalarCVTSD< 1, 0, 0b11101, "ucvtf", ARM64uitof>;
2802 defm UQXTN : SIMDTwoScalarMixedBHS<1, 0b10100, "uqxtn", int_arm64_neon_scalar_uqxtn>;
2803 defm USQADD : SIMDTwoScalarBHSDTied< 1, 0b00011, "usqadd",
2804 int_arm64_neon_usqadd>;
2806 def : Pat<(ARM64neg (v1i64 V64:$Rn)), (NEGv1i64 V64:$Rn)>;
2808 def : Pat<(v1i64 (int_arm64_neon_fcvtas (v1f64 FPR64:$Rn))),
2809 (FCVTASv1i64 FPR64:$Rn)>;
2810 def : Pat<(v1i64 (int_arm64_neon_fcvtau (v1f64 FPR64:$Rn))),
2811 (FCVTAUv1i64 FPR64:$Rn)>;
2812 def : Pat<(v1i64 (int_arm64_neon_fcvtms (v1f64 FPR64:$Rn))),
2813 (FCVTMSv1i64 FPR64:$Rn)>;
2814 def : Pat<(v1i64 (int_arm64_neon_fcvtmu (v1f64 FPR64:$Rn))),
2815 (FCVTMUv1i64 FPR64:$Rn)>;
2816 def : Pat<(v1i64 (int_arm64_neon_fcvtns (v1f64 FPR64:$Rn))),
2817 (FCVTNSv1i64 FPR64:$Rn)>;
2818 def : Pat<(v1i64 (int_arm64_neon_fcvtnu (v1f64 FPR64:$Rn))),
2819 (FCVTNUv1i64 FPR64:$Rn)>;
2820 def : Pat<(v1i64 (int_arm64_neon_fcvtps (v1f64 FPR64:$Rn))),
2821 (FCVTPSv1i64 FPR64:$Rn)>;
2822 def : Pat<(v1i64 (int_arm64_neon_fcvtpu (v1f64 FPR64:$Rn))),
2823 (FCVTPUv1i64 FPR64:$Rn)>;
2825 def : Pat<(f32 (int_arm64_neon_frecpe (f32 FPR32:$Rn))),
2826 (FRECPEv1i32 FPR32:$Rn)>;
2827 def : Pat<(f64 (int_arm64_neon_frecpe (f64 FPR64:$Rn))),
2828 (FRECPEv1i64 FPR64:$Rn)>;
2829 def : Pat<(v1f64 (int_arm64_neon_frecpe (v1f64 FPR64:$Rn))),
2830 (FRECPEv1i64 FPR64:$Rn)>;
2832 def : Pat<(f32 (int_arm64_neon_frecpx (f32 FPR32:$Rn))),
2833 (FRECPXv1i32 FPR32:$Rn)>;
2834 def : Pat<(f64 (int_arm64_neon_frecpx (f64 FPR64:$Rn))),
2835 (FRECPXv1i64 FPR64:$Rn)>;
2837 def : Pat<(f32 (int_arm64_neon_frsqrte (f32 FPR32:$Rn))),
2838 (FRSQRTEv1i32 FPR32:$Rn)>;
2839 def : Pat<(f64 (int_arm64_neon_frsqrte (f64 FPR64:$Rn))),
2840 (FRSQRTEv1i64 FPR64:$Rn)>;
2841 def : Pat<(v1f64 (int_arm64_neon_frsqrte (v1f64 FPR64:$Rn))),
2842 (FRSQRTEv1i64 FPR64:$Rn)>;
2844 // If an integer is about to be converted to a floating point value,
2845 // just load it on the floating point unit.
2846 // Here are the patterns for 8 and 16-bits to float.
2848 def : Pat <(f32 (uint_to_fp (i32 (zextloadi8 ro_indexed8:$addr)))),
2849 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2850 (LDRBro ro_indexed8:$addr), bsub))>;
2851 def : Pat <(f32 (uint_to_fp (i32 (zextloadi8 am_indexed8:$addr)))),
2852 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2853 (LDRBui am_indexed8:$addr), bsub))>;
2854 def : Pat <(f32 (uint_to_fp (i32 (zextloadi8 am_unscaled8:$addr)))),
2855 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2856 (LDURBi am_unscaled8:$addr), bsub))>;
2857 // 16-bits -> float.
2858 def : Pat <(f32 (uint_to_fp (i32 (zextloadi16 ro_indexed16:$addr)))),
2859 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2860 (LDRHro ro_indexed16:$addr), hsub))>;
2861 def : Pat <(f32 (uint_to_fp (i32 (zextloadi16 am_indexed16:$addr)))),
2862 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2863 (LDRHui am_indexed16:$addr), hsub))>;
2864 def : Pat <(f32 (uint_to_fp (i32 (zextloadi16 am_unscaled16:$addr)))),
2865 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2866 (LDURHi am_unscaled16:$addr), hsub))>;
2867 // 32-bits are handled in target specific dag combine:
2868 // performIntToFpCombine.
2869 // 64-bits integer to 32-bits floating point, not possible with
2870 // UCVTF on floating point registers (both source and destination
2871 // must have the same size).
2873 // Here are the patterns for 8, 16, 32, and 64-bits to double.
2874 // 8-bits -> double.
2875 def : Pat <(f64 (uint_to_fp (i32 (zextloadi8 ro_indexed8:$addr)))),
2876 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2877 (LDRBro ro_indexed8:$addr), bsub))>;
2878 def : Pat <(f64 (uint_to_fp (i32 (zextloadi8 am_indexed8:$addr)))),
2879 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2880 (LDRBui am_indexed8:$addr), bsub))>;
2881 def : Pat <(f64 (uint_to_fp (i32 (zextloadi8 am_unscaled8:$addr)))),
2882 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2883 (LDURBi am_unscaled8:$addr), bsub))>;
2884 // 16-bits -> double.
2885 def : Pat <(f64 (uint_to_fp (i32 (zextloadi16 ro_indexed16:$addr)))),
2886 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2887 (LDRHro ro_indexed16:$addr), hsub))>;
2888 def : Pat <(f64 (uint_to_fp (i32 (zextloadi16 am_indexed16:$addr)))),
2889 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2890 (LDRHui am_indexed16:$addr), hsub))>;
2891 def : Pat <(f64 (uint_to_fp (i32 (zextloadi16 am_unscaled16:$addr)))),
2892 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2893 (LDURHi am_unscaled16:$addr), hsub))>;
2894 // 32-bits -> double.
2895 def : Pat <(f64 (uint_to_fp (i32 (load ro_indexed32:$addr)))),
2896 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2897 (LDRSro ro_indexed32:$addr), ssub))>;
2898 def : Pat <(f64 (uint_to_fp (i32 (load am_indexed32:$addr)))),
2899 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2900 (LDRSui am_indexed32:$addr), ssub))>;
2901 def : Pat <(f64 (uint_to_fp (i32 (load am_unscaled32:$addr)))),
2902 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2903 (LDURSi am_unscaled32:$addr), ssub))>;
2904 // 64-bits -> double are handled in target specific dag combine:
2905 // performIntToFpCombine.
2907 //===----------------------------------------------------------------------===//
2908 // Advanced SIMD three different-sized vector instructions.
2909 //===----------------------------------------------------------------------===//
2911 defm ADDHN : SIMDNarrowThreeVectorBHS<0,0b0100,"addhn", int_arm64_neon_addhn>;
2912 defm SUBHN : SIMDNarrowThreeVectorBHS<0,0b0110,"subhn", int_arm64_neon_subhn>;
2913 defm RADDHN : SIMDNarrowThreeVectorBHS<1,0b0100,"raddhn",int_arm64_neon_raddhn>;
2914 defm RSUBHN : SIMDNarrowThreeVectorBHS<1,0b0110,"rsubhn",int_arm64_neon_rsubhn>;
2915 defm PMULL : SIMDDifferentThreeVectorBD<0,0b1110,"pmull",int_arm64_neon_pmull>;
2916 defm SABAL : SIMDLongThreeVectorTiedBHSabal<0,0b0101,"sabal",
2917 int_arm64_neon_sabd>;
2918 defm SABDL : SIMDLongThreeVectorBHSabdl<0, 0b0111, "sabdl",
2919 int_arm64_neon_sabd>;
2920 defm SADDL : SIMDLongThreeVectorBHS< 0, 0b0000, "saddl",
2921 BinOpFrag<(add (sext node:$LHS), (sext node:$RHS))>>;
2922 defm SADDW : SIMDWideThreeVectorBHS< 0, 0b0001, "saddw",
2923 BinOpFrag<(add node:$LHS, (sext node:$RHS))>>;
2924 defm SMLAL : SIMDLongThreeVectorTiedBHS<0, 0b1000, "smlal",
2925 TriOpFrag<(add node:$LHS, (int_arm64_neon_smull node:$MHS, node:$RHS))>>;
2926 defm SMLSL : SIMDLongThreeVectorTiedBHS<0, 0b1010, "smlsl",
2927 TriOpFrag<(sub node:$LHS, (int_arm64_neon_smull node:$MHS, node:$RHS))>>;
2928 defm SMULL : SIMDLongThreeVectorBHS<0, 0b1100, "smull", int_arm64_neon_smull>;
2929 defm SQDMLAL : SIMDLongThreeVectorSQDMLXTiedHS<0, 0b1001, "sqdmlal",
2930 int_arm64_neon_sqadd>;
2931 defm SQDMLSL : SIMDLongThreeVectorSQDMLXTiedHS<0, 0b1011, "sqdmlsl",
2932 int_arm64_neon_sqsub>;
2933 defm SQDMULL : SIMDLongThreeVectorHS<0, 0b1101, "sqdmull",
2934 int_arm64_neon_sqdmull>;
2935 defm SSUBL : SIMDLongThreeVectorBHS<0, 0b0010, "ssubl",
2936 BinOpFrag<(sub (sext node:$LHS), (sext node:$RHS))>>;
2937 defm SSUBW : SIMDWideThreeVectorBHS<0, 0b0011, "ssubw",
2938 BinOpFrag<(sub node:$LHS, (sext node:$RHS))>>;
2939 defm UABAL : SIMDLongThreeVectorTiedBHSabal<1, 0b0101, "uabal",
2940 int_arm64_neon_uabd>;
2941 defm UABDL : SIMDLongThreeVectorBHSabdl<1, 0b0111, "uabdl",
2942 int_arm64_neon_uabd>;
2943 defm UADDL : SIMDLongThreeVectorBHS<1, 0b0000, "uaddl",
2944 BinOpFrag<(add (zext node:$LHS), (zext node:$RHS))>>;
2945 defm UADDW : SIMDWideThreeVectorBHS<1, 0b0001, "uaddw",
2946 BinOpFrag<(add node:$LHS, (zext node:$RHS))>>;
2947 defm UMLAL : SIMDLongThreeVectorTiedBHS<1, 0b1000, "umlal",
2948 TriOpFrag<(add node:$LHS, (int_arm64_neon_umull node:$MHS, node:$RHS))>>;
2949 defm UMLSL : SIMDLongThreeVectorTiedBHS<1, 0b1010, "umlsl",
2950 TriOpFrag<(sub node:$LHS, (int_arm64_neon_umull node:$MHS, node:$RHS))>>;
2951 defm UMULL : SIMDLongThreeVectorBHS<1, 0b1100, "umull", int_arm64_neon_umull>;
2952 defm USUBL : SIMDLongThreeVectorBHS<1, 0b0010, "usubl",
2953 BinOpFrag<(sub (zext node:$LHS), (zext node:$RHS))>>;
2954 defm USUBW : SIMDWideThreeVectorBHS< 1, 0b0011, "usubw",
2955 BinOpFrag<(sub node:$LHS, (zext node:$RHS))>>;
2957 // Patterns for 64-bit pmull
2958 def : Pat<(int_arm64_neon_pmull64 V64:$Rn, V64:$Rm),
2959 (PMULLv1i64 V64:$Rn, V64:$Rm)>;
2960 def : Pat<(int_arm64_neon_pmull64 (vector_extract (v2i64 V128:$Rn), (i64 1)),
2961 (vector_extract (v2i64 V128:$Rm), (i64 1))),
2962 (PMULLv2i64 V128:$Rn, V128:$Rm)>;
2964 // CodeGen patterns for addhn and subhn instructions, which can actually be
2965 // written in LLVM IR without too much difficulty.
2968 def : Pat<(v8i8 (trunc (v8i16 (ARM64vlshr (add V128:$Rn, V128:$Rm), (i32 8))))),
2969 (ADDHNv8i16_v8i8 V128:$Rn, V128:$Rm)>;
2970 def : Pat<(v4i16 (trunc (v4i32 (ARM64vlshr (add V128:$Rn, V128:$Rm),
2972 (ADDHNv4i32_v4i16 V128:$Rn, V128:$Rm)>;
2973 def : Pat<(v2i32 (trunc (v2i64 (ARM64vlshr (add V128:$Rn, V128:$Rm),
2975 (ADDHNv2i64_v2i32 V128:$Rn, V128:$Rm)>;
2976 def : Pat<(concat_vectors (v8i8 V64:$Rd),
2977 (trunc (v8i16 (ARM64vlshr (add V128:$Rn, V128:$Rm),
2979 (ADDHNv8i16_v16i8 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
2980 V128:$Rn, V128:$Rm)>;
2981 def : Pat<(concat_vectors (v4i16 V64:$Rd),
2982 (trunc (v4i32 (ARM64vlshr (add V128:$Rn, V128:$Rm),
2984 (ADDHNv4i32_v8i16 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
2985 V128:$Rn, V128:$Rm)>;
2986 def : Pat<(concat_vectors (v2i32 V64:$Rd),
2987 (trunc (v2i64 (ARM64vlshr (add V128:$Rn, V128:$Rm),
2989 (ADDHNv2i64_v4i32 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
2990 V128:$Rn, V128:$Rm)>;
2993 def : Pat<(v8i8 (trunc (v8i16 (ARM64vlshr (sub V128:$Rn, V128:$Rm), (i32 8))))),
2994 (SUBHNv8i16_v8i8 V128:$Rn, V128:$Rm)>;
2995 def : Pat<(v4i16 (trunc (v4i32 (ARM64vlshr (sub V128:$Rn, V128:$Rm),
2997 (SUBHNv4i32_v4i16 V128:$Rn, V128:$Rm)>;
2998 def : Pat<(v2i32 (trunc (v2i64 (ARM64vlshr (sub V128:$Rn, V128:$Rm),
3000 (SUBHNv2i64_v2i32 V128:$Rn, V128:$Rm)>;
3001 def : Pat<(concat_vectors (v8i8 V64:$Rd),
3002 (trunc (v8i16 (ARM64vlshr (sub V128:$Rn, V128:$Rm),
3004 (SUBHNv8i16_v16i8 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3005 V128:$Rn, V128:$Rm)>;
3006 def : Pat<(concat_vectors (v4i16 V64:$Rd),
3007 (trunc (v4i32 (ARM64vlshr (sub V128:$Rn, V128:$Rm),
3009 (SUBHNv4i32_v8i16 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3010 V128:$Rn, V128:$Rm)>;
3011 def : Pat<(concat_vectors (v2i32 V64:$Rd),
3012 (trunc (v2i64 (ARM64vlshr (sub V128:$Rn, V128:$Rm),
3014 (SUBHNv2i64_v4i32 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3015 V128:$Rn, V128:$Rm)>;
3017 //----------------------------------------------------------------------------
3018 // AdvSIMD bitwise extract from vector instruction.
3019 //----------------------------------------------------------------------------
3021 defm EXT : SIMDBitwiseExtract<"ext">;
3023 def : Pat<(v4i16 (ARM64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
3024 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
3025 def : Pat<(v8i16 (ARM64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3026 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3027 def : Pat<(v2i32 (ARM64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
3028 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
3029 def : Pat<(v2f32 (ARM64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
3030 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
3031 def : Pat<(v4i32 (ARM64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3032 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3033 def : Pat<(v4f32 (ARM64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3034 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3035 def : Pat<(v2i64 (ARM64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3036 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3037 def : Pat<(v2f64 (ARM64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3038 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3040 // We use EXT to handle extract_subvector to copy the upper 64-bits of a
3042 def : Pat<(v8i8 (extract_subvector V128:$Rn, (i64 8))),
3043 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3044 def : Pat<(v4i16 (extract_subvector V128:$Rn, (i64 4))),
3045 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3046 def : Pat<(v2i32 (extract_subvector V128:$Rn, (i64 2))),
3047 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3048 def : Pat<(v1i64 (extract_subvector V128:$Rn, (i64 1))),
3049 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3050 def : Pat<(v2f32 (extract_subvector V128:$Rn, (i64 2))),
3051 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3052 def : Pat<(v1f64 (extract_subvector V128:$Rn, (i64 1))),
3053 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3056 //----------------------------------------------------------------------------
3057 // AdvSIMD zip vector
3058 //----------------------------------------------------------------------------
3060 defm TRN1 : SIMDZipVector<0b010, "trn1", ARM64trn1>;
3061 defm TRN2 : SIMDZipVector<0b110, "trn2", ARM64trn2>;
3062 defm UZP1 : SIMDZipVector<0b001, "uzp1", ARM64uzp1>;
3063 defm UZP2 : SIMDZipVector<0b101, "uzp2", ARM64uzp2>;
3064 defm ZIP1 : SIMDZipVector<0b011, "zip1", ARM64zip1>;
3065 defm ZIP2 : SIMDZipVector<0b111, "zip2", ARM64zip2>;
3067 //----------------------------------------------------------------------------
3068 // AdvSIMD TBL/TBX instructions
3069 //----------------------------------------------------------------------------
3071 defm TBL : SIMDTableLookup< 0, "tbl">;
3072 defm TBX : SIMDTableLookupTied<1, "tbx">;
3074 def : Pat<(v8i8 (int_arm64_neon_tbl1 (v16i8 VecListOne128:$Rn), (v8i8 V64:$Ri))),
3075 (TBLv8i8One VecListOne128:$Rn, V64:$Ri)>;
3076 def : Pat<(v16i8 (int_arm64_neon_tbl1 (v16i8 V128:$Ri), (v16i8 V128:$Rn))),
3077 (TBLv16i8One V128:$Ri, V128:$Rn)>;
3079 def : Pat<(v8i8 (int_arm64_neon_tbx1 (v8i8 V64:$Rd),
3080 (v16i8 VecListOne128:$Rn), (v8i8 V64:$Ri))),
3081 (TBXv8i8One V64:$Rd, VecListOne128:$Rn, V64:$Ri)>;
3082 def : Pat<(v16i8 (int_arm64_neon_tbx1 (v16i8 V128:$Rd),
3083 (v16i8 V128:$Ri), (v16i8 V128:$Rn))),
3084 (TBXv16i8One V128:$Rd, V128:$Ri, V128:$Rn)>;
3087 //----------------------------------------------------------------------------
3088 // AdvSIMD scalar CPY instruction
3089 //----------------------------------------------------------------------------
3091 defm CPY : SIMDScalarCPY<"cpy">;
3093 //----------------------------------------------------------------------------
3094 // AdvSIMD scalar pairwise instructions
3095 //----------------------------------------------------------------------------
3097 defm ADDP : SIMDPairwiseScalarD<0, 0b11011, "addp">;
3098 defm FADDP : SIMDPairwiseScalarSD<1, 0, 0b01101, "faddp">;
3099 defm FMAXNMP : SIMDPairwiseScalarSD<1, 0, 0b01100, "fmaxnmp">;
3100 defm FMAXP : SIMDPairwiseScalarSD<1, 0, 0b01111, "fmaxp">;
3101 defm FMINNMP : SIMDPairwiseScalarSD<1, 1, 0b01100, "fminnmp">;
3102 defm FMINP : SIMDPairwiseScalarSD<1, 1, 0b01111, "fminp">;
3103 def : Pat<(i64 (int_arm64_neon_saddv (v2i64 V128:$Rn))),
3104 (ADDPv2i64p V128:$Rn)>;
3105 def : Pat<(i64 (int_arm64_neon_uaddv (v2i64 V128:$Rn))),
3106 (ADDPv2i64p V128:$Rn)>;
3107 def : Pat<(f32 (int_arm64_neon_faddv (v2f32 V64:$Rn))),
3108 (FADDPv2i32p V64:$Rn)>;
3109 def : Pat<(f32 (int_arm64_neon_faddv (v4f32 V128:$Rn))),
3110 (FADDPv2i32p (EXTRACT_SUBREG (FADDPv4f32 V128:$Rn, V128:$Rn), dsub))>;
3111 def : Pat<(f64 (int_arm64_neon_faddv (v2f64 V128:$Rn))),
3112 (FADDPv2i64p V128:$Rn)>;
3113 def : Pat<(f32 (int_arm64_neon_fmaxnmv (v2f32 V64:$Rn))),
3114 (FMAXNMPv2i32p V64:$Rn)>;
3115 def : Pat<(f64 (int_arm64_neon_fmaxnmv (v2f64 V128:$Rn))),
3116 (FMAXNMPv2i64p V128:$Rn)>;
3117 def : Pat<(f32 (int_arm64_neon_fmaxv (v2f32 V64:$Rn))),
3118 (FMAXPv2i32p V64:$Rn)>;
3119 def : Pat<(f64 (int_arm64_neon_fmaxv (v2f64 V128:$Rn))),
3120 (FMAXPv2i64p V128:$Rn)>;
3121 def : Pat<(f32 (int_arm64_neon_fminnmv (v2f32 V64:$Rn))),
3122 (FMINNMPv2i32p V64:$Rn)>;
3123 def : Pat<(f64 (int_arm64_neon_fminnmv (v2f64 V128:$Rn))),
3124 (FMINNMPv2i64p V128:$Rn)>;
3125 def : Pat<(f32 (int_arm64_neon_fminv (v2f32 V64:$Rn))),
3126 (FMINPv2i32p V64:$Rn)>;
3127 def : Pat<(f64 (int_arm64_neon_fminv (v2f64 V128:$Rn))),
3128 (FMINPv2i64p V128:$Rn)>;
3130 //----------------------------------------------------------------------------
3131 // AdvSIMD INS/DUP instructions
3132 //----------------------------------------------------------------------------
3134 def DUPv8i8gpr : SIMDDupFromMain<0, 0b00001, ".8b", v8i8, V64, GPR32>;
3135 def DUPv16i8gpr : SIMDDupFromMain<1, 0b00001, ".16b", v16i8, V128, GPR32>;
3136 def DUPv4i16gpr : SIMDDupFromMain<0, 0b00010, ".4h", v4i16, V64, GPR32>;
3137 def DUPv8i16gpr : SIMDDupFromMain<1, 0b00010, ".8h", v8i16, V128, GPR32>;
3138 def DUPv2i32gpr : SIMDDupFromMain<0, 0b00100, ".2s", v2i32, V64, GPR32>;
3139 def DUPv4i32gpr : SIMDDupFromMain<1, 0b00100, ".4s", v4i32, V128, GPR32>;
3140 def DUPv2i64gpr : SIMDDupFromMain<1, 0b01000, ".2d", v2i64, V128, GPR64>;
3142 def DUPv2i64lane : SIMDDup64FromElement;
3143 def DUPv2i32lane : SIMDDup32FromElement<0, ".2s", v2i32, V64>;
3144 def DUPv4i32lane : SIMDDup32FromElement<1, ".4s", v4i32, V128>;
3145 def DUPv4i16lane : SIMDDup16FromElement<0, ".4h", v4i16, V64>;
3146 def DUPv8i16lane : SIMDDup16FromElement<1, ".8h", v8i16, V128>;
3147 def DUPv8i8lane : SIMDDup8FromElement <0, ".8b", v8i8, V64>;
3148 def DUPv16i8lane : SIMDDup8FromElement <1, ".16b", v16i8, V128>;
3150 def : Pat<(v2f32 (ARM64dup (f32 FPR32:$Rn))),
3151 (v2f32 (DUPv2i32lane
3152 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rn, ssub),
3154 def : Pat<(v4f32 (ARM64dup (f32 FPR32:$Rn))),
3155 (v4f32 (DUPv4i32lane
3156 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rn, ssub),
3158 def : Pat<(v2f64 (ARM64dup (f64 FPR64:$Rn))),
3159 (v2f64 (DUPv2i64lane
3160 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$Rn, dsub),
3163 def : Pat<(v2f32 (ARM64duplane32 (v4f32 V128:$Rn), VectorIndexS:$imm)),
3164 (DUPv2i32lane V128:$Rn, VectorIndexS:$imm)>;
3165 def : Pat<(v4f32 (ARM64duplane32 (v4f32 V128:$Rn), VectorIndexS:$imm)),
3166 (DUPv4i32lane V128:$Rn, VectorIndexS:$imm)>;
3167 def : Pat<(v2f64 (ARM64duplane64 (v2f64 V128:$Rn), VectorIndexD:$imm)),
3168 (DUPv2i64lane V128:$Rn, VectorIndexD:$imm)>;
3170 // If there's an (ARM64dup (vector_extract ...) ...), we can use a duplane
3171 // instruction even if the types don't match: we just have to remap the lane
3172 // carefully. N.b. this trick only applies to truncations.
3173 def VecIndex_x2 : SDNodeXForm<imm, [{
3174 return CurDAG->getTargetConstant(2 * N->getZExtValue(), MVT::i64);
3176 def VecIndex_x4 : SDNodeXForm<imm, [{
3177 return CurDAG->getTargetConstant(4 * N->getZExtValue(), MVT::i64);
3179 def VecIndex_x8 : SDNodeXForm<imm, [{
3180 return CurDAG->getTargetConstant(8 * N->getZExtValue(), MVT::i64);
3183 multiclass DUPWithTruncPats<ValueType ResVT, ValueType Src64VT,
3184 ValueType Src128VT, ValueType ScalVT,
3185 Instruction DUP, SDNodeXForm IdxXFORM> {
3186 def : Pat<(ResVT (ARM64dup (ScalVT (vector_extract (Src128VT V128:$Rn),
3188 (DUP V128:$Rn, (IdxXFORM imm:$idx))>;
3190 def : Pat<(ResVT (ARM64dup (ScalVT (vector_extract (Src64VT V64:$Rn),
3192 (DUP (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), (IdxXFORM imm:$idx))>;
3195 defm : DUPWithTruncPats<v8i8, v4i16, v8i16, i32, DUPv8i8lane, VecIndex_x2>;
3196 defm : DUPWithTruncPats<v8i8, v2i32, v4i32, i32, DUPv8i8lane, VecIndex_x4>;
3197 defm : DUPWithTruncPats<v4i16, v2i32, v4i32, i32, DUPv4i16lane, VecIndex_x2>;
3199 defm : DUPWithTruncPats<v16i8, v4i16, v8i16, i32, DUPv16i8lane, VecIndex_x2>;
3200 defm : DUPWithTruncPats<v16i8, v2i32, v4i32, i32, DUPv16i8lane, VecIndex_x4>;
3201 defm : DUPWithTruncPats<v8i16, v2i32, v4i32, i32, DUPv8i16lane, VecIndex_x2>;
3203 multiclass DUPWithTrunci64Pats<ValueType ResVT, Instruction DUP,
3204 SDNodeXForm IdxXFORM> {
3205 def : Pat<(ResVT (ARM64dup (i32 (trunc (vector_extract (v2i64 V128:$Rn),
3207 (DUP V128:$Rn, (IdxXFORM imm:$idx))>;
3209 def : Pat<(ResVT (ARM64dup (i32 (trunc (vector_extract (v1i64 V64:$Rn),
3211 (DUP (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), (IdxXFORM imm:$idx))>;
3214 defm : DUPWithTrunci64Pats<v8i8, DUPv8i8lane, VecIndex_x8>;
3215 defm : DUPWithTrunci64Pats<v4i16, DUPv4i16lane, VecIndex_x4>;
3216 defm : DUPWithTrunci64Pats<v2i32, DUPv2i32lane, VecIndex_x2>;
3218 defm : DUPWithTrunci64Pats<v16i8, DUPv16i8lane, VecIndex_x8>;
3219 defm : DUPWithTrunci64Pats<v8i16, DUPv8i16lane, VecIndex_x4>;
3220 defm : DUPWithTrunci64Pats<v4i32, DUPv4i32lane, VecIndex_x2>;
3222 // SMOV and UMOV definitions, with some extra patterns for convenience
3226 def : Pat<(sext_inreg (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx), i8),
3227 (i32 (SMOVvi8to32 V128:$Rn, VectorIndexB:$idx))>;
3228 def : Pat<(sext_inreg (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx), i8),
3229 (i64 (SMOVvi8to64 V128:$Rn, VectorIndexB:$idx))>;
3230 def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
3231 (i32 (SMOVvi16to32 V128:$Rn, VectorIndexH:$idx))>;
3232 def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
3233 (i64 (SMOVvi16to64 V128:$Rn, VectorIndexH:$idx))>;
3234 def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
3235 (i32 (SMOVvi16to32 V128:$Rn, VectorIndexH:$idx))>;
3236 def : Pat<(sext (i32 (vector_extract (v4i32 V128:$Rn), VectorIndexS:$idx))),
3237 (i64 (SMOVvi32to64 V128:$Rn, VectorIndexS:$idx))>;
3239 // Extracting i8 or i16 elements will have the zero-extend transformed to
3240 // an 'and' mask by type legalization since neither i8 nor i16 are legal types
3241 // for ARM64. Match these patterns here since UMOV already zeroes out the high
3242 // bits of the destination register.
3243 def : Pat<(and (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx),
3245 (i32 (UMOVvi8 V128:$Rn, VectorIndexB:$idx))>;
3246 def : Pat<(and (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),
3248 (i32 (UMOVvi16 V128:$Rn, VectorIndexH:$idx))>;
3252 def : Pat<(v16i8 (scalar_to_vector GPR32:$Rn)),
3253 (SUBREG_TO_REG (i32 0),
3254 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3255 def : Pat<(v8i8 (scalar_to_vector GPR32:$Rn)),
3256 (SUBREG_TO_REG (i32 0),
3257 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3259 def : Pat<(v8i16 (scalar_to_vector GPR32:$Rn)),
3260 (SUBREG_TO_REG (i32 0),
3261 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3262 def : Pat<(v4i16 (scalar_to_vector GPR32:$Rn)),
3263 (SUBREG_TO_REG (i32 0),
3264 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3266 def : Pat<(v2i32 (scalar_to_vector (i32 FPR32:$Rn))),
3267 (v2i32 (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)),
3268 (i32 FPR32:$Rn), ssub))>;
3269 def : Pat<(v4i32 (scalar_to_vector (i32 FPR32:$Rn))),
3270 (v4i32 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
3271 (i32 FPR32:$Rn), ssub))>;
3272 def : Pat<(v2i64 (scalar_to_vector (i64 FPR64:$Rn))),
3273 (v2i64 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
3274 (i64 FPR64:$Rn), dsub))>;
3276 def : Pat<(v4f32 (scalar_to_vector (f32 FPR32:$Rn))),
3277 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rn, ssub)>;
3278 def : Pat<(v2f32 (scalar_to_vector (f32 FPR32:$Rn))),
3279 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), FPR32:$Rn, ssub)>;
3280 def : Pat<(v2f64 (scalar_to_vector (f64 FPR64:$Rn))),
3281 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$Rn, dsub)>;
3283 def : Pat<(v2f32 (vector_insert (v2f32 V64:$Rn),
3284 (f32 FPR32:$Rm), (i64 VectorIndexS:$imm))),
3287 (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), V64:$Rn, dsub)),
3289 (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rm, ssub)),
3292 def : Pat<(v4f32 (vector_insert (v4f32 V128:$Rn),
3293 (f32 FPR32:$Rm), (i64 VectorIndexS:$imm))),
3295 V128:$Rn, VectorIndexS:$imm,
3296 (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rm, ssub)),
3298 def : Pat<(v2f64 (vector_insert (v2f64 V128:$Rn),
3299 (f64 FPR64:$Rm), (i64 VectorIndexD:$imm))),
3301 V128:$Rn, VectorIndexD:$imm,
3302 (v2f64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$Rm, dsub)),
3305 // Copy an element at a constant index in one vector into a constant indexed
3306 // element of another.
3307 // FIXME refactor to a shared class/dev parameterized on vector type, vector
3308 // index type and INS extension
3309 def : Pat<(v16i8 (int_arm64_neon_vcopy_lane
3310 (v16i8 V128:$Vd), VectorIndexB:$idx, (v16i8 V128:$Vs),
3311 VectorIndexB:$idx2)),
3313 V128:$Vd, VectorIndexB:$idx, V128:$Vs, VectorIndexB:$idx2)
3315 def : Pat<(v8i16 (int_arm64_neon_vcopy_lane
3316 (v8i16 V128:$Vd), VectorIndexH:$idx, (v8i16 V128:$Vs),
3317 VectorIndexH:$idx2)),
3319 V128:$Vd, VectorIndexH:$idx, V128:$Vs, VectorIndexH:$idx2)
3321 def : Pat<(v4i32 (int_arm64_neon_vcopy_lane
3322 (v4i32 V128:$Vd), VectorIndexS:$idx, (v4i32 V128:$Vs),
3323 VectorIndexS:$idx2)),
3325 V128:$Vd, VectorIndexS:$idx, V128:$Vs, VectorIndexS:$idx2)
3327 def : Pat<(v2i64 (int_arm64_neon_vcopy_lane
3328 (v2i64 V128:$Vd), VectorIndexD:$idx, (v2i64 V128:$Vs),
3329 VectorIndexD:$idx2)),
3331 V128:$Vd, VectorIndexD:$idx, V128:$Vs, VectorIndexD:$idx2)
3334 multiclass Neon_INS_elt_pattern<ValueType VT128, ValueType VT64,
3335 ValueType VTScal, Instruction INS> {
3336 def : Pat<(VT128 (vector_insert V128:$src,
3337 (VTScal (vector_extract (VT128 V128:$Rn), imm:$Immn)),
3339 (INS V128:$src, imm:$Immd, V128:$Rn, imm:$Immn)>;
3341 def : Pat<(VT128 (vector_insert V128:$src,
3342 (VTScal (vector_extract (VT64 V64:$Rn), imm:$Immn)),
3344 (INS V128:$src, imm:$Immd,
3345 (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), imm:$Immn)>;
3347 def : Pat<(VT64 (vector_insert V64:$src,
3348 (VTScal (vector_extract (VT128 V128:$Rn), imm:$Immn)),
3350 (EXTRACT_SUBREG (INS (SUBREG_TO_REG (i64 0), V64:$src, dsub),
3351 imm:$Immd, V128:$Rn, imm:$Immn),
3354 def : Pat<(VT64 (vector_insert V64:$src,
3355 (VTScal (vector_extract (VT64 V64:$Rn), imm:$Immn)),
3358 (INS (SUBREG_TO_REG (i64 0), V64:$src, dsub), imm:$Immd,
3359 (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), imm:$Immn),
3363 defm : Neon_INS_elt_pattern<v4f32, v2f32, f32, INSvi32lane>;
3364 defm : Neon_INS_elt_pattern<v2f64, v1f64, f64, INSvi64lane>;
3365 defm : Neon_INS_elt_pattern<v16i8, v8i8, i32, INSvi8lane>;
3366 defm : Neon_INS_elt_pattern<v8i16, v4i16, i32, INSvi16lane>;
3367 defm : Neon_INS_elt_pattern<v4i32, v2i32, i32, INSvi32lane>;
3368 defm : Neon_INS_elt_pattern<v2i64, v1i64, i64, INSvi32lane>;
3371 // Floating point vector extractions are codegen'd as either a sequence of
3372 // subregister extractions, possibly fed by an INS if the lane number is
3373 // anything other than zero.
3374 def : Pat<(vector_extract (v2f64 V128:$Rn), 0),
3375 (f64 (EXTRACT_SUBREG V128:$Rn, dsub))>;
3376 def : Pat<(vector_extract (v4f32 V128:$Rn), 0),
3377 (f32 (EXTRACT_SUBREG V128:$Rn, ssub))>;
3378 def : Pat<(vector_extract (v2f64 V128:$Rn), VectorIndexD:$idx),
3379 (f64 (EXTRACT_SUBREG
3380 (INSvi64lane (v2f64 (IMPLICIT_DEF)), 0,
3381 V128:$Rn, VectorIndexD:$idx),
3383 def : Pat<(vector_extract (v4f32 V128:$Rn), VectorIndexS:$idx),
3384 (f32 (EXTRACT_SUBREG
3385 (INSvi32lane (v4f32 (IMPLICIT_DEF)), 0,
3386 V128:$Rn, VectorIndexS:$idx),
3389 // All concat_vectors operations are canonicalised to act on i64 vectors for
3390 // ARM64. In the general case we need an instruction, which had just as well be
3392 class ConcatPat<ValueType DstTy, ValueType SrcTy>
3393 : Pat<(DstTy (concat_vectors (SrcTy V64:$Rd), V64:$Rn)),
3394 (INSvi64lane (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), 1,
3395 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rn, dsub), 0)>;
3397 def : ConcatPat<v2i64, v1i64>;
3398 def : ConcatPat<v2f64, v1f64>;
3399 def : ConcatPat<v4i32, v2i32>;
3400 def : ConcatPat<v4f32, v2f32>;
3401 def : ConcatPat<v8i16, v4i16>;
3402 def : ConcatPat<v16i8, v8i8>;
3404 // If the high lanes are undef, though, we can just ignore them:
3405 class ConcatUndefPat<ValueType DstTy, ValueType SrcTy>
3406 : Pat<(DstTy (concat_vectors (SrcTy V64:$Rn), undef)),
3407 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rn, dsub)>;
3409 def : ConcatUndefPat<v2i64, v1i64>;
3410 def : ConcatUndefPat<v2f64, v1f64>;
3411 def : ConcatUndefPat<v4i32, v2i32>;
3412 def : ConcatUndefPat<v4f32, v2f32>;
3413 def : ConcatUndefPat<v8i16, v4i16>;
3414 def : ConcatUndefPat<v16i8, v8i8>;
3416 //----------------------------------------------------------------------------
3417 // AdvSIMD across lanes instructions
3418 //----------------------------------------------------------------------------
3420 defm ADDV : SIMDAcrossLanesBHS<0, 0b11011, "addv">;
3421 defm SMAXV : SIMDAcrossLanesBHS<0, 0b01010, "smaxv">;
3422 defm SMINV : SIMDAcrossLanesBHS<0, 0b11010, "sminv">;
3423 defm UMAXV : SIMDAcrossLanesBHS<1, 0b01010, "umaxv">;
3424 defm UMINV : SIMDAcrossLanesBHS<1, 0b11010, "uminv">;
3425 defm SADDLV : SIMDAcrossLanesHSD<0, 0b00011, "saddlv">;
3426 defm UADDLV : SIMDAcrossLanesHSD<1, 0b00011, "uaddlv">;
3427 defm FMAXNMV : SIMDAcrossLanesS<0b01100, 0, "fmaxnmv", int_arm64_neon_fmaxnmv>;
3428 defm FMAXV : SIMDAcrossLanesS<0b01111, 0, "fmaxv", int_arm64_neon_fmaxv>;
3429 defm FMINNMV : SIMDAcrossLanesS<0b01100, 1, "fminnmv", int_arm64_neon_fminnmv>;
3430 defm FMINV : SIMDAcrossLanesS<0b01111, 1, "fminv", int_arm64_neon_fminv>;
3432 multiclass SIMDAcrossLanesSignedIntrinsic<string baseOpc, Intrinsic intOp> {
3433 // If there is a sign extension after this intrinsic, consume it as smov already
3435 def : Pat<(i32 (sext_inreg (i32 (intOp (v8i8 V64:$Rn))), i8)),
3437 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3438 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3440 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
3442 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3443 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3445 // If there is a sign extension after this intrinsic, consume it as smov already
3447 def : Pat<(i32 (sext_inreg (i32 (intOp (v16i8 V128:$Rn))), i8)),
3449 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3450 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3452 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
3454 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3455 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3457 // If there is a sign extension after this intrinsic, consume it as smov already
3459 def : Pat<(i32 (sext_inreg (i32 (intOp (v4i16 V64:$Rn))), i16)),
3461 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3462 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3464 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
3466 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3467 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3469 // If there is a sign extension after this intrinsic, consume it as smov already
3471 def : Pat<(i32 (sext_inreg (i32 (intOp (v8i16 V128:$Rn))), i16)),
3473 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3474 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3476 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
3478 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3479 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3482 def : Pat<(i32 (intOp (v4i32 V128:$Rn))),
3483 (i32 (EXTRACT_SUBREG
3484 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3485 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), ssub),
3489 multiclass SIMDAcrossLanesUnsignedIntrinsic<string baseOpc, Intrinsic intOp> {
3490 // If there is a masking operation keeping only what has been actually
3491 // generated, consume it.
3492 def : Pat<(i32 (and (i32 (intOp (v8i8 V64:$Rn))), maski8_or_more)),
3493 (i32 (EXTRACT_SUBREG
3494 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3495 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3497 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
3498 (i32 (EXTRACT_SUBREG
3499 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3500 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3502 // If there is a masking operation keeping only what has been actually
3503 // generated, consume it.
3504 def : Pat<(i32 (and (i32 (intOp (v16i8 V128:$Rn))), maski8_or_more)),
3505 (i32 (EXTRACT_SUBREG
3506 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3507 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3509 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
3510 (i32 (EXTRACT_SUBREG
3511 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3512 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3515 // If there is a masking operation keeping only what has been actually
3516 // generated, consume it.
3517 def : Pat<(i32 (and (i32 (intOp (v4i16 V64:$Rn))), maski16_or_more)),
3518 (i32 (EXTRACT_SUBREG
3519 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3520 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3522 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
3523 (i32 (EXTRACT_SUBREG
3524 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3525 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3527 // If there is a masking operation keeping only what has been actually
3528 // generated, consume it.
3529 def : Pat<(i32 (and (i32 (intOp (v8i16 V128:$Rn))), maski16_or_more)),
3530 (i32 (EXTRACT_SUBREG
3531 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3532 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3534 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
3535 (i32 (EXTRACT_SUBREG
3536 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3537 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3540 def : Pat<(i32 (intOp (v4i32 V128:$Rn))),
3541 (i32 (EXTRACT_SUBREG
3542 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3543 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), ssub),
3548 multiclass SIMDAcrossLanesSignedLongIntrinsic<string baseOpc, Intrinsic intOp> {
3549 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
3551 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3552 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), hsub),
3554 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
3556 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3557 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), hsub),
3560 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
3561 (i32 (EXTRACT_SUBREG
3562 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3563 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), ssub),
3565 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
3566 (i32 (EXTRACT_SUBREG
3567 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3568 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), ssub),
3571 def : Pat<(i64 (intOp (v4i32 V128:$Rn))),
3572 (i64 (EXTRACT_SUBREG
3573 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3574 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), dsub),
3578 multiclass SIMDAcrossLanesUnsignedLongIntrinsic<string baseOpc,
3580 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
3581 (i32 (EXTRACT_SUBREG
3582 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3583 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), hsub),
3585 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
3586 (i32 (EXTRACT_SUBREG
3587 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3588 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), hsub),
3591 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
3592 (i32 (EXTRACT_SUBREG
3593 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3594 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), ssub),
3596 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
3597 (i32 (EXTRACT_SUBREG
3598 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3599 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), ssub),
3602 def : Pat<(i64 (intOp (v4i32 V128:$Rn))),
3603 (i64 (EXTRACT_SUBREG
3604 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3605 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), dsub),
3609 defm : SIMDAcrossLanesSignedIntrinsic<"ADDV", int_arm64_neon_saddv>;
3610 // vaddv_[su]32 is special; -> ADDP Vd.2S,Vn.2S,Vm.2S; return Vd.s[0];Vn==Vm
3611 def : Pat<(i32 (int_arm64_neon_saddv (v2i32 V64:$Rn))),
3612 (EXTRACT_SUBREG (ADDPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3614 defm : SIMDAcrossLanesUnsignedIntrinsic<"ADDV", int_arm64_neon_uaddv>;
3615 // vaddv_[su]32 is special; -> ADDP Vd.2S,Vn.2S,Vm.2S; return Vd.s[0];Vn==Vm
3616 def : Pat<(i32 (int_arm64_neon_uaddv (v2i32 V64:$Rn))),
3617 (EXTRACT_SUBREG (ADDPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3619 defm : SIMDAcrossLanesSignedIntrinsic<"SMAXV", int_arm64_neon_smaxv>;
3620 def : Pat<(i32 (int_arm64_neon_smaxv (v2i32 V64:$Rn))),
3621 (EXTRACT_SUBREG (SMAXPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3623 defm : SIMDAcrossLanesSignedIntrinsic<"SMINV", int_arm64_neon_sminv>;
3624 def : Pat<(i32 (int_arm64_neon_sminv (v2i32 V64:$Rn))),
3625 (EXTRACT_SUBREG (SMINPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3627 defm : SIMDAcrossLanesUnsignedIntrinsic<"UMAXV", int_arm64_neon_umaxv>;
3628 def : Pat<(i32 (int_arm64_neon_umaxv (v2i32 V64:$Rn))),
3629 (EXTRACT_SUBREG (UMAXPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3631 defm : SIMDAcrossLanesUnsignedIntrinsic<"UMINV", int_arm64_neon_uminv>;
3632 def : Pat<(i32 (int_arm64_neon_uminv (v2i32 V64:$Rn))),
3633 (EXTRACT_SUBREG (UMINPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3635 defm : SIMDAcrossLanesSignedLongIntrinsic<"SADDLV", int_arm64_neon_saddlv>;
3636 defm : SIMDAcrossLanesUnsignedLongIntrinsic<"UADDLV", int_arm64_neon_uaddlv>;
3638 // The vaddlv_s32 intrinsic gets mapped to SADDLP.
3639 def : Pat<(i64 (int_arm64_neon_saddlv (v2i32 V64:$Rn))),
3640 (i64 (EXTRACT_SUBREG
3641 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3642 (SADDLPv2i32_v1i64 V64:$Rn), dsub),
3644 // The vaddlv_u32 intrinsic gets mapped to UADDLP.
3645 def : Pat<(i64 (int_arm64_neon_uaddlv (v2i32 V64:$Rn))),
3646 (i64 (EXTRACT_SUBREG
3647 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3648 (UADDLPv2i32_v1i64 V64:$Rn), dsub),
3651 //------------------------------------------------------------------------------
3652 // AdvSIMD modified immediate instructions
3653 //------------------------------------------------------------------------------
3656 defm BIC : SIMDModifiedImmVectorShiftTied<1, 0b11, 0b01, "bic", ARM64bici>;
3658 defm ORR : SIMDModifiedImmVectorShiftTied<0, 0b11, 0b01, "orr", ARM64orri>;
3660 def : InstAlias<"bic $Vd.4h, $imm", (BICv4i16 V64:$Vd, imm0_255:$imm, 0)>;
3661 def : InstAlias<"bic $Vd.8h, $imm", (BICv8i16 V128:$Vd, imm0_255:$imm, 0)>;
3662 def : InstAlias<"bic $Vd.2s, $imm", (BICv2i32 V64:$Vd, imm0_255:$imm, 0)>;
3663 def : InstAlias<"bic $Vd.4s, $imm", (BICv4i32 V128:$Vd, imm0_255:$imm, 0)>;
3665 def : InstAlias<"bic.4h $Vd, $imm", (BICv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
3666 def : InstAlias<"bic.8h $Vd, $imm", (BICv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
3667 def : InstAlias<"bic.2s $Vd, $imm", (BICv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
3668 def : InstAlias<"bic.4s $Vd, $imm", (BICv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
3670 def : InstAlias<"orr $Vd.4h, $imm", (ORRv4i16 V64:$Vd, imm0_255:$imm, 0)>;
3671 def : InstAlias<"orr $Vd.8h, $imm", (ORRv8i16 V128:$Vd, imm0_255:$imm, 0)>;
3672 def : InstAlias<"orr $Vd.2s, $imm", (ORRv2i32 V64:$Vd, imm0_255:$imm, 0)>;
3673 def : InstAlias<"orr $Vd.4s, $imm", (ORRv4i32 V128:$Vd, imm0_255:$imm, 0)>;
3675 def : InstAlias<"orr.4h $Vd, $imm", (ORRv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
3676 def : InstAlias<"orr.8h $Vd, $imm", (ORRv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
3677 def : InstAlias<"orr.2s $Vd, $imm", (ORRv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
3678 def : InstAlias<"orr.4s $Vd, $imm", (ORRv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
3681 def FMOVv2f64_ns : SIMDModifiedImmVectorNoShift<1, 1, 0b1111, V128, fpimm8,
3683 [(set (v2f64 V128:$Rd), (ARM64fmov imm0_255:$imm8))]>;
3684 def FMOVv2f32_ns : SIMDModifiedImmVectorNoShift<0, 0, 0b1111, V64, fpimm8,
3686 [(set (v2f32 V64:$Rd), (ARM64fmov imm0_255:$imm8))]>;
3687 def FMOVv4f32_ns : SIMDModifiedImmVectorNoShift<1, 0, 0b1111, V128, fpimm8,
3689 [(set (v4f32 V128:$Rd), (ARM64fmov imm0_255:$imm8))]>;
3693 // EDIT byte mask: scalar
3694 let isReMaterializable = 1, isAsCheapAsAMove = 1 in
3695 def MOVID : SIMDModifiedImmScalarNoShift<0, 1, 0b1110, "movi",
3696 [(set FPR64:$Rd, simdimmtype10:$imm8)]>;
3697 // The movi_edit node has the immediate value already encoded, so we use
3698 // a plain imm0_255 here.
3699 def : Pat<(f64 (ARM64movi_edit imm0_255:$shift)),
3700 (MOVID imm0_255:$shift)>;
3702 def : Pat<(v1i64 immAllZerosV), (MOVID (i32 0))>;
3703 def : Pat<(v2i32 immAllZerosV), (MOVID (i32 0))>;
3704 def : Pat<(v4i16 immAllZerosV), (MOVID (i32 0))>;
3705 def : Pat<(v8i8 immAllZerosV), (MOVID (i32 0))>;
3707 def : Pat<(v1i64 immAllOnesV), (MOVID (i32 255))>;
3708 def : Pat<(v2i32 immAllOnesV), (MOVID (i32 255))>;
3709 def : Pat<(v4i16 immAllOnesV), (MOVID (i32 255))>;
3710 def : Pat<(v8i8 immAllOnesV), (MOVID (i32 255))>;
3712 // EDIT byte mask: 2d
3714 // The movi_edit node has the immediate value already encoded, so we use
3715 // a plain imm0_255 in the pattern
3716 let isReMaterializable = 1, isAsCheapAsAMove = 1 in
3717 def MOVIv2d_ns : SIMDModifiedImmVectorNoShift<1, 1, 0b1110, V128,
3720 [(set (v2i64 V128:$Rd), (ARM64movi_edit imm0_255:$imm8))]>;
3723 // Use movi.2d to materialize 0.0 if the HW does zero-cycle zeroing.
3724 // Complexity is added to break a tie with a plain MOVI.
3725 let AddedComplexity = 1 in {
3726 def : Pat<(f32 fpimm0),
3727 (f32 (EXTRACT_SUBREG (v2i64 (MOVIv2d_ns (i32 0))), ssub))>,
3729 def : Pat<(f64 fpimm0),
3730 (f64 (EXTRACT_SUBREG (v2i64 (MOVIv2d_ns (i32 0))), dsub))>,
3734 def : Pat<(v2i64 immAllZerosV), (MOVIv2d_ns (i32 0))>;
3735 def : Pat<(v4i32 immAllZerosV), (MOVIv2d_ns (i32 0))>;
3736 def : Pat<(v8i16 immAllZerosV), (MOVIv2d_ns (i32 0))>;
3737 def : Pat<(v16i8 immAllZerosV), (MOVIv2d_ns (i32 0))>;
3739 def : Pat<(v2i64 immAllOnesV), (MOVIv2d_ns (i32 255))>;
3740 def : Pat<(v4i32 immAllOnesV), (MOVIv2d_ns (i32 255))>;
3741 def : Pat<(v8i16 immAllOnesV), (MOVIv2d_ns (i32 255))>;
3742 def : Pat<(v16i8 immAllOnesV), (MOVIv2d_ns (i32 255))>;
3744 def : Pat<(v2f64 (ARM64dup (f64 fpimm0))), (MOVIv2d_ns (i32 0))>;
3745 def : Pat<(v4f32 (ARM64dup (f32 fpimm0))), (MOVIv2d_ns (i32 0))>;
3747 // EDIT per word & halfword: 2s, 4h, 4s, & 8h
3748 defm MOVI : SIMDModifiedImmVectorShift<0, 0b10, 0b00, "movi">;
3750 def : InstAlias<"movi $Vd.4h, $imm", (MOVIv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
3751 def : InstAlias<"movi $Vd.8h, $imm", (MOVIv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
3752 def : InstAlias<"movi $Vd.2s, $imm", (MOVIv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
3753 def : InstAlias<"movi $Vd.4s, $imm", (MOVIv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
3755 def : InstAlias<"movi.4h $Vd, $imm", (MOVIv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
3756 def : InstAlias<"movi.8h $Vd, $imm", (MOVIv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
3757 def : InstAlias<"movi.2s $Vd, $imm", (MOVIv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
3758 def : InstAlias<"movi.4s $Vd, $imm", (MOVIv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
3760 def : Pat<(v2i32 (ARM64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
3761 (MOVIv2i32 imm0_255:$imm8, imm:$shift)>;
3762 def : Pat<(v4i32 (ARM64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
3763 (MOVIv4i32 imm0_255:$imm8, imm:$shift)>;
3764 def : Pat<(v4i16 (ARM64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
3765 (MOVIv4i16 imm0_255:$imm8, imm:$shift)>;
3766 def : Pat<(v8i16 (ARM64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
3767 (MOVIv8i16 imm0_255:$imm8, imm:$shift)>;
3769 // EDIT per word: 2s & 4s with MSL shifter
3770 def MOVIv2s_msl : SIMDModifiedImmMoveMSL<0, 0, {1,1,0,?}, V64, "movi", ".2s",
3771 [(set (v2i32 V64:$Rd),
3772 (ARM64movi_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
3773 def MOVIv4s_msl : SIMDModifiedImmMoveMSL<1, 0, {1,1,0,?}, V128, "movi", ".4s",
3774 [(set (v4i32 V128:$Rd),
3775 (ARM64movi_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
3777 // Per byte: 8b & 16b
3778 def MOVIv8b_ns : SIMDModifiedImmVectorNoShift<0, 0, 0b1110, V64, imm0_255,
3780 [(set (v8i8 V64:$Rd), (ARM64movi imm0_255:$imm8))]>;
3781 def MOVIv16b_ns : SIMDModifiedImmVectorNoShift<1, 0, 0b1110, V128, imm0_255,
3783 [(set (v16i8 V128:$Rd), (ARM64movi imm0_255:$imm8))]>;
3787 // EDIT per word & halfword: 2s, 4h, 4s, & 8h
3788 defm MVNI : SIMDModifiedImmVectorShift<1, 0b10, 0b00, "mvni">;
3790 def : InstAlias<"mvni $Vd.4h, $imm", (MVNIv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
3791 def : InstAlias<"mvni $Vd.8h, $imm", (MVNIv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
3792 def : InstAlias<"mvni $Vd.2s, $imm", (MVNIv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
3793 def : InstAlias<"mvni $Vd.4s, $imm", (MVNIv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
3795 def : InstAlias<"mvni.4h $Vd, $imm", (MVNIv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
3796 def : InstAlias<"mvni.8h $Vd, $imm", (MVNIv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
3797 def : InstAlias<"mvni.2s $Vd, $imm", (MVNIv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
3798 def : InstAlias<"mvni.4s $Vd, $imm", (MVNIv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
3800 def : Pat<(v2i32 (ARM64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
3801 (MVNIv2i32 imm0_255:$imm8, imm:$shift)>;
3802 def : Pat<(v4i32 (ARM64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
3803 (MVNIv4i32 imm0_255:$imm8, imm:$shift)>;
3804 def : Pat<(v4i16 (ARM64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
3805 (MVNIv4i16 imm0_255:$imm8, imm:$shift)>;
3806 def : Pat<(v8i16 (ARM64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
3807 (MVNIv8i16 imm0_255:$imm8, imm:$shift)>;
3809 // EDIT per word: 2s & 4s with MSL shifter
3810 def MVNIv2s_msl : SIMDModifiedImmMoveMSL<0, 1, {1,1,0,?}, V64, "mvni", ".2s",
3811 [(set (v2i32 V64:$Rd),
3812 (ARM64mvni_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
3813 def MVNIv4s_msl : SIMDModifiedImmMoveMSL<1, 1, {1,1,0,?}, V128, "mvni", ".4s",
3814 [(set (v4i32 V128:$Rd),
3815 (ARM64mvni_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
3817 //----------------------------------------------------------------------------
3818 // AdvSIMD indexed element
3819 //----------------------------------------------------------------------------
3821 let neverHasSideEffects = 1 in {
3822 defm FMLA : SIMDFPIndexedSDTied<0, 0b0001, "fmla">;
3823 defm FMLS : SIMDFPIndexedSDTied<0, 0b0101, "fmls">;
3826 // NOTE: Operands are reordered in the FMLA/FMLS PatFrags because the
3827 // instruction expects the addend first, while the intrinsic expects it last.
3829 // On the other hand, there are quite a few valid combinatorial options due to
3830 // the commutativity of multiplication and the fact that (-x) * y = x * (-y).
3831 defm : SIMDFPIndexedSDTiedPatterns<"FMLA",
3832 TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)>>;
3833 defm : SIMDFPIndexedSDTiedPatterns<"FMLA",
3834 TriOpFrag<(fma node:$MHS, node:$RHS, node:$LHS)>>;
3836 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
3837 TriOpFrag<(fma node:$MHS, (fneg node:$RHS), node:$LHS)> >;
3838 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
3839 TriOpFrag<(fma node:$RHS, (fneg node:$MHS), node:$LHS)> >;
3840 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
3841 TriOpFrag<(fma (fneg node:$RHS), node:$MHS, node:$LHS)> >;
3842 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
3843 TriOpFrag<(fma (fneg node:$MHS), node:$RHS, node:$LHS)> >;
3845 multiclass FMLSIndexedAfterNegPatterns<SDPatternOperator OpNode> {
3846 // 3 variants for the .2s version: DUPLANE from 128-bit, DUPLANE from 64-bit
3848 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
3849 (ARM64duplane32 (v4f32 (fneg V128:$Rm)),
3850 VectorIndexS:$idx))),
3851 (FMLSv2i32_indexed V64:$Rd, V64:$Rn, V128:$Rm, VectorIndexS:$idx)>;
3852 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
3853 (v2f32 (ARM64duplane32
3854 (v4f32 (insert_subvector undef,
3855 (v2f32 (fneg V64:$Rm)),
3857 VectorIndexS:$idx)))),
3858 (FMLSv2i32_indexed V64:$Rd, V64:$Rn,
3859 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub),
3860 VectorIndexS:$idx)>;
3861 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
3862 (ARM64dup (f32 (fneg FPR32Op:$Rm))))),
3863 (FMLSv2i32_indexed V64:$Rd, V64:$Rn,
3864 (SUBREG_TO_REG (i32 0), FPR32Op:$Rm, ssub), (i64 0))>;
3866 // 3 variants for the .4s version: DUPLANE from 128-bit, DUPLANE from 64-bit
3868 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
3869 (ARM64duplane32 (v4f32 (fneg V128:$Rm)),
3870 VectorIndexS:$idx))),
3871 (FMLSv4i32_indexed V128:$Rd, V128:$Rn, V128:$Rm,
3872 VectorIndexS:$idx)>;
3873 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
3874 (v4f32 (ARM64duplane32
3875 (v4f32 (insert_subvector undef,
3876 (v2f32 (fneg V64:$Rm)),
3878 VectorIndexS:$idx)))),
3879 (FMLSv4i32_indexed V128:$Rd, V128:$Rn,
3880 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub),
3881 VectorIndexS:$idx)>;
3882 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
3883 (ARM64dup (f32 (fneg FPR32Op:$Rm))))),
3884 (FMLSv4i32_indexed V128:$Rd, V128:$Rn,
3885 (SUBREG_TO_REG (i32 0), FPR32Op:$Rm, ssub), (i64 0))>;
3887 // 2 variants for the .2d version: DUPLANE from 128-bit, and DUP scalar
3888 // (DUPLANE from 64-bit would be trivial).
3889 def : Pat<(v2f64 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn),
3890 (ARM64duplane64 (v2f64 (fneg V128:$Rm)),
3891 VectorIndexD:$idx))),
3893 V128:$Rd, V128:$Rn, V128:$Rm, VectorIndexS:$idx)>;
3894 def : Pat<(v2f64 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn),
3895 (ARM64dup (f64 (fneg FPR64Op:$Rm))))),
3896 (FMLSv2i64_indexed V128:$Rd, V128:$Rn,
3897 (SUBREG_TO_REG (i32 0), FPR64Op:$Rm, dsub), (i64 0))>;
3899 // 2 variants for 32-bit scalar version: extract from .2s or from .4s
3900 def : Pat<(f32 (OpNode (f32 FPR32:$Rd), (f32 FPR32:$Rn),
3901 (vector_extract (v4f32 (fneg V128:$Rm)),
3902 VectorIndexS:$idx))),
3903 (FMLSv1i32_indexed FPR32:$Rd, FPR32:$Rn,
3904 V128:$Rm, VectorIndexS:$idx)>;
3905 def : Pat<(f32 (OpNode (f32 FPR32:$Rd), (f32 FPR32:$Rn),
3906 (vector_extract (v2f32 (fneg V64:$Rm)),
3907 VectorIndexS:$idx))),
3908 (FMLSv1i32_indexed FPR32:$Rd, FPR32:$Rn,
3909 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub), VectorIndexS:$idx)>;
3911 // 1 variant for 64-bit scalar version: extract from .1d or from .2d
3912 def : Pat<(f64 (OpNode (f64 FPR64:$Rd), (f64 FPR64:$Rn),
3913 (vector_extract (v2f64 (fneg V128:$Rm)),
3914 VectorIndexS:$idx))),
3915 (FMLSv1i64_indexed FPR64:$Rd, FPR64:$Rn,
3916 V128:$Rm, VectorIndexS:$idx)>;
3919 defm : FMLSIndexedAfterNegPatterns<
3920 TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)> >;
3921 defm : FMLSIndexedAfterNegPatterns<
3922 TriOpFrag<(fma node:$MHS, node:$RHS, node:$LHS)> >;
3924 defm FMULX : SIMDFPIndexedSD<1, 0b1001, "fmulx", int_arm64_neon_fmulx>;
3925 defm FMUL : SIMDFPIndexedSD<0, 0b1001, "fmul", fmul>;
3927 def : Pat<(v2f32 (fmul V64:$Rn, (ARM64dup (f32 FPR32:$Rm)))),
3928 (FMULv2i32_indexed V64:$Rn,
3929 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rm, ssub),
3931 def : Pat<(v4f32 (fmul V128:$Rn, (ARM64dup (f32 FPR32:$Rm)))),
3932 (FMULv4i32_indexed V128:$Rn,
3933 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rm, ssub),
3935 def : Pat<(v2f64 (fmul V128:$Rn, (ARM64dup (f64 FPR64:$Rm)))),
3936 (FMULv2i64_indexed V128:$Rn,
3937 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$Rm, dsub),
3940 defm SQDMULH : SIMDIndexedHS<0, 0b1100, "sqdmulh", int_arm64_neon_sqdmulh>;
3941 defm SQRDMULH : SIMDIndexedHS<0, 0b1101, "sqrdmulh", int_arm64_neon_sqrdmulh>;
3942 defm MLA : SIMDVectorIndexedHSTied<1, 0b0000, "mla",
3943 TriOpFrag<(add node:$LHS, (mul node:$MHS, node:$RHS))>>;
3944 defm MLS : SIMDVectorIndexedHSTied<1, 0b0100, "mls",
3945 TriOpFrag<(sub node:$LHS, (mul node:$MHS, node:$RHS))>>;
3946 defm MUL : SIMDVectorIndexedHS<0, 0b1000, "mul", mul>;
3947 defm SMLAL : SIMDVectorIndexedLongSDTied<0, 0b0010, "smlal",
3948 TriOpFrag<(add node:$LHS, (int_arm64_neon_smull node:$MHS, node:$RHS))>>;
3949 defm SMLSL : SIMDVectorIndexedLongSDTied<0, 0b0110, "smlsl",
3950 TriOpFrag<(sub node:$LHS, (int_arm64_neon_smull node:$MHS, node:$RHS))>>;
3951 defm SMULL : SIMDVectorIndexedLongSD<0, 0b1010, "smull",
3952 int_arm64_neon_smull>;
3953 defm SQDMLAL : SIMDIndexedLongSQDMLXSDTied<0, 0b0011, "sqdmlal",
3954 int_arm64_neon_sqadd>;
3955 defm SQDMLSL : SIMDIndexedLongSQDMLXSDTied<0, 0b0111, "sqdmlsl",
3956 int_arm64_neon_sqsub>;
3957 defm SQDMULL : SIMDIndexedLongSD<0, 0b1011, "sqdmull", int_arm64_neon_sqdmull>;
3958 defm UMLAL : SIMDVectorIndexedLongSDTied<1, 0b0010, "umlal",
3959 TriOpFrag<(add node:$LHS, (int_arm64_neon_umull node:$MHS, node:$RHS))>>;
3960 defm UMLSL : SIMDVectorIndexedLongSDTied<1, 0b0110, "umlsl",
3961 TriOpFrag<(sub node:$LHS, (int_arm64_neon_umull node:$MHS, node:$RHS))>>;
3962 defm UMULL : SIMDVectorIndexedLongSD<1, 0b1010, "umull",
3963 int_arm64_neon_umull>;
3965 // A scalar sqdmull with the second operand being a vector lane can be
3966 // handled directly with the indexed instruction encoding.
3967 def : Pat<(int_arm64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
3968 (vector_extract (v4i32 V128:$Vm),
3969 VectorIndexS:$idx)),
3970 (SQDMULLv1i64_indexed FPR32:$Rn, V128:$Vm, VectorIndexS:$idx)>;
3972 //----------------------------------------------------------------------------
3973 // AdvSIMD scalar shift instructions
3974 //----------------------------------------------------------------------------
3975 defm FCVTZS : SIMDScalarRShiftSD<0, 0b11111, "fcvtzs">;
3976 defm FCVTZU : SIMDScalarRShiftSD<1, 0b11111, "fcvtzu">;
3977 defm SCVTF : SIMDScalarRShiftSD<0, 0b11100, "scvtf">;
3978 defm UCVTF : SIMDScalarRShiftSD<1, 0b11100, "ucvtf">;
3979 // Codegen patterns for the above. We don't put these directly on the
3980 // instructions because TableGen's type inference can't handle the truth.
3981 // Having the same base pattern for fp <--> int totally freaks it out.
3982 def : Pat<(int_arm64_neon_vcvtfp2fxs FPR32:$Rn, vecshiftR32:$imm),
3983 (FCVTZSs FPR32:$Rn, vecshiftR32:$imm)>;
3984 def : Pat<(int_arm64_neon_vcvtfp2fxu FPR32:$Rn, vecshiftR32:$imm),
3985 (FCVTZUs FPR32:$Rn, vecshiftR32:$imm)>;
3986 def : Pat<(i64 (int_arm64_neon_vcvtfp2fxs (f64 FPR64:$Rn), vecshiftR64:$imm)),
3987 (FCVTZSd FPR64:$Rn, vecshiftR64:$imm)>;
3988 def : Pat<(i64 (int_arm64_neon_vcvtfp2fxu (f64 FPR64:$Rn), vecshiftR64:$imm)),
3989 (FCVTZUd FPR64:$Rn, vecshiftR64:$imm)>;
3990 def : Pat<(v1i64 (int_arm64_neon_vcvtfp2fxs (v1f64 FPR64:$Rn),
3992 (FCVTZSd FPR64:$Rn, vecshiftR64:$imm)>;
3993 def : Pat<(v1i64 (int_arm64_neon_vcvtfp2fxu (v1f64 FPR64:$Rn),
3995 (FCVTZUd FPR64:$Rn, vecshiftR64:$imm)>;
3996 def : Pat<(int_arm64_neon_vcvtfxs2fp FPR32:$Rn, vecshiftR32:$imm),
3997 (SCVTFs FPR32:$Rn, vecshiftR32:$imm)>;
3998 def : Pat<(int_arm64_neon_vcvtfxu2fp FPR32:$Rn, vecshiftR32:$imm),
3999 (UCVTFs FPR32:$Rn, vecshiftR32:$imm)>;
4000 def : Pat<(f64 (int_arm64_neon_vcvtfxs2fp (i64 FPR64:$Rn), vecshiftR64:$imm)),
4001 (SCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
4002 def : Pat<(f64 (int_arm64_neon_vcvtfxu2fp (i64 FPR64:$Rn), vecshiftR64:$imm)),
4003 (UCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
4004 def : Pat<(v1f64 (int_arm64_neon_vcvtfxs2fp (v1i64 FPR64:$Rn),
4006 (SCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
4007 def : Pat<(v1f64 (int_arm64_neon_vcvtfxu2fp (v1i64 FPR64:$Rn),
4009 (UCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
4011 defm SHL : SIMDScalarLShiftD< 0, 0b01010, "shl", ARM64vshl>;
4012 defm SLI : SIMDScalarLShiftDTied<1, 0b01010, "sli">;
4013 defm SQRSHRN : SIMDScalarRShiftBHS< 0, 0b10011, "sqrshrn",
4014 int_arm64_neon_sqrshrn>;
4015 defm SQRSHRUN : SIMDScalarRShiftBHS< 1, 0b10001, "sqrshrun",
4016 int_arm64_neon_sqrshrun>;
4017 defm SQSHLU : SIMDScalarLShiftBHSD<1, 0b01100, "sqshlu", ARM64sqshlui>;
4018 defm SQSHL : SIMDScalarLShiftBHSD<0, 0b01110, "sqshl", ARM64sqshli>;
4019 defm SQSHRN : SIMDScalarRShiftBHS< 0, 0b10010, "sqshrn",
4020 int_arm64_neon_sqshrn>;
4021 defm SQSHRUN : SIMDScalarRShiftBHS< 1, 0b10000, "sqshrun",
4022 int_arm64_neon_sqshrun>;
4023 defm SRI : SIMDScalarRShiftDTied< 1, 0b01000, "sri">;
4024 defm SRSHR : SIMDScalarRShiftD< 0, 0b00100, "srshr", ARM64srshri>;
4025 defm SRSRA : SIMDScalarRShiftDTied< 0, 0b00110, "srsra",
4026 TriOpFrag<(add node:$LHS,
4027 (ARM64srshri node:$MHS, node:$RHS))>>;
4028 defm SSHR : SIMDScalarRShiftD< 0, 0b00000, "sshr", ARM64vashr>;
4029 defm SSRA : SIMDScalarRShiftDTied< 0, 0b00010, "ssra",
4030 TriOpFrag<(add node:$LHS,
4031 (ARM64vashr node:$MHS, node:$RHS))>>;
4032 defm UQRSHRN : SIMDScalarRShiftBHS< 1, 0b10011, "uqrshrn",
4033 int_arm64_neon_uqrshrn>;
4034 defm UQSHL : SIMDScalarLShiftBHSD<1, 0b01110, "uqshl", ARM64uqshli>;
4035 defm UQSHRN : SIMDScalarRShiftBHS< 1, 0b10010, "uqshrn",
4036 int_arm64_neon_uqshrn>;
4037 defm URSHR : SIMDScalarRShiftD< 1, 0b00100, "urshr", ARM64urshri>;
4038 defm URSRA : SIMDScalarRShiftDTied< 1, 0b00110, "ursra",
4039 TriOpFrag<(add node:$LHS,
4040 (ARM64urshri node:$MHS, node:$RHS))>>;
4041 defm USHR : SIMDScalarRShiftD< 1, 0b00000, "ushr", ARM64vlshr>;
4042 defm USRA : SIMDScalarRShiftDTied< 1, 0b00010, "usra",
4043 TriOpFrag<(add node:$LHS,
4044 (ARM64vlshr node:$MHS, node:$RHS))>>;
4046 //----------------------------------------------------------------------------
4047 // AdvSIMD vector shift instructions
4048 //----------------------------------------------------------------------------
4049 defm FCVTZS:SIMDVectorRShiftSD<0, 0b11111, "fcvtzs", int_arm64_neon_vcvtfp2fxs>;
4050 defm FCVTZU:SIMDVectorRShiftSD<1, 0b11111, "fcvtzu", int_arm64_neon_vcvtfp2fxu>;
4051 defm SCVTF: SIMDVectorRShiftSDToFP<0, 0b11100, "scvtf",
4052 int_arm64_neon_vcvtfxs2fp>;
4053 defm RSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10001, "rshrn",
4054 int_arm64_neon_rshrn>;
4055 defm SHL : SIMDVectorLShiftBHSD<0, 0b01010, "shl", ARM64vshl>;
4056 defm SHRN : SIMDVectorRShiftNarrowBHS<0, 0b10000, "shrn",
4057 BinOpFrag<(trunc (ARM64vashr node:$LHS, node:$RHS))>>;
4058 defm SLI : SIMDVectorLShiftBHSDTied<1, 0b01010, "sli", int_arm64_neon_vsli>;
4059 def : Pat<(v1i64 (int_arm64_neon_vsli (v1i64 FPR64:$Rd), (v1i64 FPR64:$Rn),
4060 (i32 vecshiftL64:$imm))),
4061 (SLId FPR64:$Rd, FPR64:$Rn, vecshiftL64:$imm)>;
4062 defm SQRSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10011, "sqrshrn",
4063 int_arm64_neon_sqrshrn>;
4064 defm SQRSHRUN: SIMDVectorRShiftNarrowBHS<1, 0b10001, "sqrshrun",
4065 int_arm64_neon_sqrshrun>;
4066 defm SQSHLU : SIMDVectorLShiftBHSD<1, 0b01100, "sqshlu", ARM64sqshlui>;
4067 defm SQSHL : SIMDVectorLShiftBHSD<0, 0b01110, "sqshl", ARM64sqshli>;
4068 defm SQSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10010, "sqshrn",
4069 int_arm64_neon_sqshrn>;
4070 defm SQSHRUN : SIMDVectorRShiftNarrowBHS<1, 0b10000, "sqshrun",
4071 int_arm64_neon_sqshrun>;
4072 defm SRI : SIMDVectorRShiftBHSDTied<1, 0b01000, "sri", int_arm64_neon_vsri>;
4073 def : Pat<(v1i64 (int_arm64_neon_vsri (v1i64 FPR64:$Rd), (v1i64 FPR64:$Rn),
4074 (i32 vecshiftR64:$imm))),
4075 (SRId FPR64:$Rd, FPR64:$Rn, vecshiftR64:$imm)>;
4076 defm SRSHR : SIMDVectorRShiftBHSD<0, 0b00100, "srshr", ARM64srshri>;
4077 defm SRSRA : SIMDVectorRShiftBHSDTied<0, 0b00110, "srsra",
4078 TriOpFrag<(add node:$LHS,
4079 (ARM64srshri node:$MHS, node:$RHS))> >;
4080 defm SSHLL : SIMDVectorLShiftLongBHSD<0, 0b10100, "sshll",
4081 BinOpFrag<(ARM64vshl (sext node:$LHS), node:$RHS)>>;
4083 defm SSHR : SIMDVectorRShiftBHSD<0, 0b00000, "sshr", ARM64vashr>;
4084 defm SSRA : SIMDVectorRShiftBHSDTied<0, 0b00010, "ssra",
4085 TriOpFrag<(add node:$LHS, (ARM64vashr node:$MHS, node:$RHS))>>;
4086 defm UCVTF : SIMDVectorRShiftSDToFP<1, 0b11100, "ucvtf",
4087 int_arm64_neon_vcvtfxu2fp>;
4088 defm UQRSHRN : SIMDVectorRShiftNarrowBHS<1, 0b10011, "uqrshrn",
4089 int_arm64_neon_uqrshrn>;
4090 defm UQSHL : SIMDVectorLShiftBHSD<1, 0b01110, "uqshl", ARM64uqshli>;
4091 defm UQSHRN : SIMDVectorRShiftNarrowBHS<1, 0b10010, "uqshrn",
4092 int_arm64_neon_uqshrn>;
4093 defm URSHR : SIMDVectorRShiftBHSD<1, 0b00100, "urshr", ARM64urshri>;
4094 defm URSRA : SIMDVectorRShiftBHSDTied<1, 0b00110, "ursra",
4095 TriOpFrag<(add node:$LHS,
4096 (ARM64urshri node:$MHS, node:$RHS))> >;
4097 defm USHLL : SIMDVectorLShiftLongBHSD<1, 0b10100, "ushll",
4098 BinOpFrag<(ARM64vshl (zext node:$LHS), node:$RHS)>>;
4099 defm USHR : SIMDVectorRShiftBHSD<1, 0b00000, "ushr", ARM64vlshr>;
4100 defm USRA : SIMDVectorRShiftBHSDTied<1, 0b00010, "usra",
4101 TriOpFrag<(add node:$LHS, (ARM64vlshr node:$MHS, node:$RHS))> >;
4103 // SHRN patterns for when a logical right shift was used instead of arithmetic
4104 // (the immediate guarantees no sign bits actually end up in the result so it
4106 def : Pat<(v8i8 (trunc (ARM64vlshr (v8i16 V128:$Rn), vecshiftR16Narrow:$imm))),
4107 (SHRNv8i8_shift V128:$Rn, vecshiftR16Narrow:$imm)>;
4108 def : Pat<(v4i16 (trunc (ARM64vlshr (v4i32 V128:$Rn), vecshiftR32Narrow:$imm))),
4109 (SHRNv4i16_shift V128:$Rn, vecshiftR32Narrow:$imm)>;
4110 def : Pat<(v2i32 (trunc (ARM64vlshr (v2i64 V128:$Rn), vecshiftR64Narrow:$imm))),
4111 (SHRNv2i32_shift V128:$Rn, vecshiftR64Narrow:$imm)>;
4113 def : Pat<(v16i8 (concat_vectors (v8i8 V64:$Rd),
4114 (trunc (ARM64vlshr (v8i16 V128:$Rn),
4115 vecshiftR16Narrow:$imm)))),
4116 (SHRNv16i8_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
4117 V128:$Rn, vecshiftR16Narrow:$imm)>;
4118 def : Pat<(v8i16 (concat_vectors (v4i16 V64:$Rd),
4119 (trunc (ARM64vlshr (v4i32 V128:$Rn),
4120 vecshiftR32Narrow:$imm)))),
4121 (SHRNv8i16_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
4122 V128:$Rn, vecshiftR32Narrow:$imm)>;
4123 def : Pat<(v4i32 (concat_vectors (v2i32 V64:$Rd),
4124 (trunc (ARM64vlshr (v2i64 V128:$Rn),
4125 vecshiftR64Narrow:$imm)))),
4126 (SHRNv4i32_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
4127 V128:$Rn, vecshiftR32Narrow:$imm)>;
4129 // Vector sign and zero extensions are implemented with SSHLL and USSHLL.
4130 // Anyexts are implemented as zexts.
4131 def : Pat<(v8i16 (sext (v8i8 V64:$Rn))), (SSHLLv8i8_shift V64:$Rn, (i32 0))>;
4132 def : Pat<(v8i16 (zext (v8i8 V64:$Rn))), (USHLLv8i8_shift V64:$Rn, (i32 0))>;
4133 def : Pat<(v8i16 (anyext (v8i8 V64:$Rn))), (USHLLv8i8_shift V64:$Rn, (i32 0))>;
4134 def : Pat<(v4i32 (sext (v4i16 V64:$Rn))), (SSHLLv4i16_shift V64:$Rn, (i32 0))>;
4135 def : Pat<(v4i32 (zext (v4i16 V64:$Rn))), (USHLLv4i16_shift V64:$Rn, (i32 0))>;
4136 def : Pat<(v4i32 (anyext (v4i16 V64:$Rn))), (USHLLv4i16_shift V64:$Rn, (i32 0))>;
4137 def : Pat<(v2i64 (sext (v2i32 V64:$Rn))), (SSHLLv2i32_shift V64:$Rn, (i32 0))>;
4138 def : Pat<(v2i64 (zext (v2i32 V64:$Rn))), (USHLLv2i32_shift V64:$Rn, (i32 0))>;
4139 def : Pat<(v2i64 (anyext (v2i32 V64:$Rn))), (USHLLv2i32_shift V64:$Rn, (i32 0))>;
4140 // Also match an extend from the upper half of a 128 bit source register.
4141 def : Pat<(v8i16 (anyext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
4142 (USHLLv16i8_shift V128:$Rn, (i32 0))>;
4143 def : Pat<(v8i16 (zext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
4144 (USHLLv16i8_shift V128:$Rn, (i32 0))>;
4145 def : Pat<(v8i16 (sext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
4146 (SSHLLv16i8_shift V128:$Rn, (i32 0))>;
4147 def : Pat<(v4i32 (anyext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
4148 (USHLLv8i16_shift V128:$Rn, (i32 0))>;
4149 def : Pat<(v4i32 (zext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
4150 (USHLLv8i16_shift V128:$Rn, (i32 0))>;
4151 def : Pat<(v4i32 (sext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
4152 (SSHLLv8i16_shift V128:$Rn, (i32 0))>;
4153 def : Pat<(v2i64 (anyext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
4154 (USHLLv4i32_shift V128:$Rn, (i32 0))>;
4155 def : Pat<(v2i64 (zext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
4156 (USHLLv4i32_shift V128:$Rn, (i32 0))>;
4157 def : Pat<(v2i64 (sext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
4158 (SSHLLv4i32_shift V128:$Rn, (i32 0))>;
4160 // Vector shift sxtl aliases
4161 def : InstAlias<"sxtl.8h $dst, $src1",
4162 (SSHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
4163 def : InstAlias<"sxtl $dst.8h, $src1.8b",
4164 (SSHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
4165 def : InstAlias<"sxtl.4s $dst, $src1",
4166 (SSHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
4167 def : InstAlias<"sxtl $dst.4s, $src1.4h",
4168 (SSHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
4169 def : InstAlias<"sxtl.2d $dst, $src1",
4170 (SSHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
4171 def : InstAlias<"sxtl $dst.2d, $src1.2s",
4172 (SSHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
4174 // Vector shift sxtl2 aliases
4175 def : InstAlias<"sxtl2.8h $dst, $src1",
4176 (SSHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
4177 def : InstAlias<"sxtl2 $dst.8h, $src1.16b",
4178 (SSHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
4179 def : InstAlias<"sxtl2.4s $dst, $src1",
4180 (SSHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
4181 def : InstAlias<"sxtl2 $dst.4s, $src1.8h",
4182 (SSHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
4183 def : InstAlias<"sxtl2.2d $dst, $src1",
4184 (SSHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
4185 def : InstAlias<"sxtl2 $dst.2d, $src1.4s",
4186 (SSHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
4188 // Vector shift uxtl aliases
4189 def : InstAlias<"uxtl.8h $dst, $src1",
4190 (USHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
4191 def : InstAlias<"uxtl $dst.8h, $src1.8b",
4192 (USHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
4193 def : InstAlias<"uxtl.4s $dst, $src1",
4194 (USHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
4195 def : InstAlias<"uxtl $dst.4s, $src1.4h",
4196 (USHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
4197 def : InstAlias<"uxtl.2d $dst, $src1",
4198 (USHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
4199 def : InstAlias<"uxtl $dst.2d, $src1.2s",
4200 (USHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
4202 // Vector shift uxtl2 aliases
4203 def : InstAlias<"uxtl2.8h $dst, $src1",
4204 (USHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
4205 def : InstAlias<"uxtl2 $dst.8h, $src1.16b",
4206 (USHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
4207 def : InstAlias<"uxtl2.4s $dst, $src1",
4208 (USHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
4209 def : InstAlias<"uxtl2 $dst.4s, $src1.8h",
4210 (USHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
4211 def : InstAlias<"uxtl2.2d $dst, $src1",
4212 (USHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
4213 def : InstAlias<"uxtl2 $dst.2d, $src1.4s",
4214 (USHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
4216 // If an integer is about to be converted to a floating point value,
4217 // just load it on the floating point unit.
4218 // These patterns are more complex because floating point loads do not
4219 // support sign extension.
4220 // The sign extension has to be explicitly added and is only supported for
4221 // one step: byte-to-half, half-to-word, word-to-doubleword.
4222 // SCVTF GPR -> FPR is 9 cycles.
4223 // SCVTF FPR -> FPR is 4 cyclces.
4224 // (sign extension with lengthen) SXTL FPR -> FPR is 2 cycles.
4225 // Therefore, we can do 2 sign extensions and one SCVTF FPR -> FPR
4226 // and still being faster.
4227 // However, this is not good for code size.
4228 // 8-bits -> float. 2 sizes step-up.
4229 def : Pat <(f32 (sint_to_fp (i32 (sextloadi8 ro_indexed8:$addr)))),
4230 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
4235 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4236 (LDRBro ro_indexed8:$addr),
4241 ssub)))>, Requires<[NotForCodeSize]>;
4242 def : Pat <(f32 (sint_to_fp (i32 (sextloadi8 am_indexed8:$addr)))),
4243 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
4248 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4249 (LDRBui am_indexed8:$addr),
4254 ssub)))>, Requires<[NotForCodeSize]>;
4255 def : Pat <(f32 (sint_to_fp (i32 (sextloadi8 am_unscaled8:$addr)))),
4256 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
4261 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4262 (LDURBi am_unscaled8:$addr),
4267 ssub)))>, Requires<[NotForCodeSize]>;
4268 // 16-bits -> float. 1 size step-up.
4269 def : Pat <(f32 (sint_to_fp (i32 (sextloadi16 ro_indexed16:$addr)))),
4270 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
4272 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4273 (LDRHro ro_indexed16:$addr),
4276 ssub)))>, Requires<[NotForCodeSize]>;
4277 def : Pat <(f32 (sint_to_fp (i32 (sextloadi16 am_indexed16:$addr)))),
4278 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
4280 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4281 (LDRHui am_indexed16:$addr),
4284 ssub)))>, Requires<[NotForCodeSize]>;
4285 def : Pat <(f32 (sint_to_fp (i32 (sextloadi16 am_unscaled16:$addr)))),
4286 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
4288 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4289 (LDURHi am_unscaled16:$addr),
4292 ssub)))>, Requires<[NotForCodeSize]>;
4293 // 32-bits to 32-bits are handled in target specific dag combine:
4294 // performIntToFpCombine.
4295 // 64-bits integer to 32-bits floating point, not possible with
4296 // SCVTF on floating point registers (both source and destination
4297 // must have the same size).
4299 // Here are the patterns for 8, 16, 32, and 64-bits to double.
4300 // 8-bits -> double. 3 size step-up: give up.
4301 // 16-bits -> double. 2 size step.
4302 def : Pat <(f64 (sint_to_fp (i32 (sextloadi16 ro_indexed16:$addr)))),
4303 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4308 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4309 (LDRHro ro_indexed16:$addr),
4314 dsub)))>, Requires<[NotForCodeSize]>;
4315 def : Pat <(f64 (sint_to_fp (i32 (sextloadi16 am_indexed16:$addr)))),
4316 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4321 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4322 (LDRHui am_indexed16:$addr),
4327 dsub)))>, Requires<[NotForCodeSize]>;
4328 def : Pat <(f64 (sint_to_fp (i32 (sextloadi16 am_unscaled16:$addr)))),
4329 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4334 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4335 (LDURHi am_unscaled16:$addr),
4340 dsub)))>, Requires<[NotForCodeSize]>;
4341 // 32-bits -> double. 1 size step-up.
4342 def : Pat <(f64 (sint_to_fp (i32 (load ro_indexed32:$addr)))),
4343 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4345 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4346 (LDRSro ro_indexed32:$addr),
4349 dsub)))>, Requires<[NotForCodeSize]>;
4350 def : Pat <(f64 (sint_to_fp (i32 (load am_indexed32:$addr)))),
4351 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4353 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4354 (LDRSui am_indexed32:$addr),
4357 dsub)))>, Requires<[NotForCodeSize]>;
4358 def : Pat <(f64 (sint_to_fp (i32 (load am_unscaled32:$addr)))),
4359 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4361 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4362 (LDURSi am_unscaled32:$addr),
4365 dsub)))>, Requires<[NotForCodeSize]>;
4366 // 64-bits -> double are handled in target specific dag combine:
4367 // performIntToFpCombine.
4370 //----------------------------------------------------------------------------
4371 // AdvSIMD Load-Store Structure
4372 //----------------------------------------------------------------------------
4373 defm LD1 : SIMDLd1Multiple<"ld1">;
4374 defm LD2 : SIMDLd2Multiple<"ld2">;
4375 defm LD3 : SIMDLd3Multiple<"ld3">;
4376 defm LD4 : SIMDLd4Multiple<"ld4">;
4378 defm ST1 : SIMDSt1Multiple<"st1">;
4379 defm ST2 : SIMDSt2Multiple<"st2">;
4380 defm ST3 : SIMDSt3Multiple<"st3">;
4381 defm ST4 : SIMDSt4Multiple<"st4">;
4383 class Ld1Pat<ValueType ty, Instruction INST>
4384 : Pat<(ty (load am_simdnoindex:$vaddr)), (INST am_simdnoindex:$vaddr)>;
4386 def : Ld1Pat<v16i8, LD1Onev16b>;
4387 def : Ld1Pat<v8i16, LD1Onev8h>;
4388 def : Ld1Pat<v4i32, LD1Onev4s>;
4389 def : Ld1Pat<v2i64, LD1Onev2d>;
4390 def : Ld1Pat<v8i8, LD1Onev8b>;
4391 def : Ld1Pat<v4i16, LD1Onev4h>;
4392 def : Ld1Pat<v2i32, LD1Onev2s>;
4393 def : Ld1Pat<v1i64, LD1Onev1d>;
4395 class St1Pat<ValueType ty, Instruction INST>
4396 : Pat<(store ty:$Vt, am_simdnoindex:$vaddr),
4397 (INST ty:$Vt, am_simdnoindex:$vaddr)>;
4399 def : St1Pat<v16i8, ST1Onev16b>;
4400 def : St1Pat<v8i16, ST1Onev8h>;
4401 def : St1Pat<v4i32, ST1Onev4s>;
4402 def : St1Pat<v2i64, ST1Onev2d>;
4403 def : St1Pat<v8i8, ST1Onev8b>;
4404 def : St1Pat<v4i16, ST1Onev4h>;
4405 def : St1Pat<v2i32, ST1Onev2s>;
4406 def : St1Pat<v1i64, ST1Onev1d>;
4412 defm LD1R : SIMDLdR<0, 0b110, 0, "ld1r", "One", 1, 2, 4, 8>;
4413 defm LD2R : SIMDLdR<1, 0b110, 0, "ld2r", "Two", 2, 4, 8, 16>;
4414 defm LD3R : SIMDLdR<0, 0b111, 0, "ld3r", "Three", 3, 6, 12, 24>;
4415 defm LD4R : SIMDLdR<1, 0b111, 0, "ld4r", "Four", 4, 8, 16, 32>;
4416 let mayLoad = 1, neverHasSideEffects = 1 in {
4417 defm LD1 : SIMDLdSingleBTied<0, 0b000, "ld1", VecListOneb, GPR64pi1>;
4418 defm LD1 : SIMDLdSingleHTied<0, 0b010, 0, "ld1", VecListOneh, GPR64pi2>;
4419 defm LD1 : SIMDLdSingleSTied<0, 0b100, 0b00, "ld1", VecListOnes, GPR64pi4>;
4420 defm LD1 : SIMDLdSingleDTied<0, 0b100, 0b01, "ld1", VecListOned, GPR64pi8>;
4421 defm LD2 : SIMDLdSingleBTied<1, 0b000, "ld2", VecListTwob, GPR64pi2>;
4422 defm LD2 : SIMDLdSingleHTied<1, 0b010, 0, "ld2", VecListTwoh, GPR64pi4>;
4423 defm LD2 : SIMDLdSingleSTied<1, 0b100, 0b00, "ld2", VecListTwos, GPR64pi8>;
4424 defm LD2 : SIMDLdSingleDTied<1, 0b100, 0b01, "ld2", VecListTwod, GPR64pi16>;
4425 defm LD3 : SIMDLdSingleBTied<0, 0b001, "ld3", VecListThreeb, GPR64pi3>;
4426 defm LD3 : SIMDLdSingleHTied<0, 0b011, 0, "ld3", VecListThreeh, GPR64pi6>;
4427 defm LD3 : SIMDLdSingleSTied<0, 0b101, 0b00, "ld3", VecListThrees, GPR64pi12>;
4428 defm LD3 : SIMDLdSingleDTied<0, 0b101, 0b01, "ld3", VecListThreed, GPR64pi24>;
4429 defm LD4 : SIMDLdSingleBTied<1, 0b001, "ld4", VecListFourb, GPR64pi4>;
4430 defm LD4 : SIMDLdSingleHTied<1, 0b011, 0, "ld4", VecListFourh, GPR64pi8>;
4431 defm LD4 : SIMDLdSingleSTied<1, 0b101, 0b00, "ld4", VecListFours, GPR64pi16>;
4432 defm LD4 : SIMDLdSingleDTied<1, 0b101, 0b01, "ld4", VecListFourd, GPR64pi32>;
4435 def : Pat<(v8i8 (ARM64dup (i32 (extloadi8 am_simdnoindex:$vaddr)))),
4436 (LD1Rv8b am_simdnoindex:$vaddr)>;
4437 def : Pat<(v16i8 (ARM64dup (i32 (extloadi8 am_simdnoindex:$vaddr)))),
4438 (LD1Rv16b am_simdnoindex:$vaddr)>;
4439 def : Pat<(v4i16 (ARM64dup (i32 (extloadi16 am_simdnoindex:$vaddr)))),
4440 (LD1Rv4h am_simdnoindex:$vaddr)>;
4441 def : Pat<(v8i16 (ARM64dup (i32 (extloadi16 am_simdnoindex:$vaddr)))),
4442 (LD1Rv8h am_simdnoindex:$vaddr)>;
4443 def : Pat<(v2i32 (ARM64dup (i32 (load am_simdnoindex:$vaddr)))),
4444 (LD1Rv2s am_simdnoindex:$vaddr)>;
4445 def : Pat<(v4i32 (ARM64dup (i32 (load am_simdnoindex:$vaddr)))),
4446 (LD1Rv4s am_simdnoindex:$vaddr)>;
4447 def : Pat<(v2i64 (ARM64dup (i64 (load am_simdnoindex:$vaddr)))),
4448 (LD1Rv2d am_simdnoindex:$vaddr)>;
4449 def : Pat<(v1i64 (ARM64dup (i64 (load am_simdnoindex:$vaddr)))),
4450 (LD1Rv1d am_simdnoindex:$vaddr)>;
4451 // Grab the floating point version too
4452 def : Pat<(v2f32 (ARM64dup (f32 (load am_simdnoindex:$vaddr)))),
4453 (LD1Rv2s am_simdnoindex:$vaddr)>;
4454 def : Pat<(v4f32 (ARM64dup (f32 (load am_simdnoindex:$vaddr)))),
4455 (LD1Rv4s am_simdnoindex:$vaddr)>;
4456 def : Pat<(v2f64 (ARM64dup (f64 (load am_simdnoindex:$vaddr)))),
4457 (LD1Rv2d am_simdnoindex:$vaddr)>;
4458 def : Pat<(v1f64 (ARM64dup (f64 (load am_simdnoindex:$vaddr)))),
4459 (LD1Rv1d am_simdnoindex:$vaddr)>;
4461 class Ld1Lane128Pat<SDPatternOperator scalar_load, Operand VecIndex,
4462 ValueType VTy, ValueType STy, Instruction LD1>
4463 : Pat<(vector_insert (VTy VecListOne128:$Rd),
4464 (STy (scalar_load am_simdnoindex:$vaddr)), VecIndex:$idx),
4465 (LD1 VecListOne128:$Rd, VecIndex:$idx, am_simdnoindex:$vaddr)>;
4467 def : Ld1Lane128Pat<extloadi8, VectorIndexB, v16i8, i32, LD1i8>;
4468 def : Ld1Lane128Pat<extloadi16, VectorIndexH, v8i16, i32, LD1i16>;
4469 def : Ld1Lane128Pat<load, VectorIndexS, v4i32, i32, LD1i32>;
4470 def : Ld1Lane128Pat<load, VectorIndexS, v4f32, f32, LD1i32>;
4471 def : Ld1Lane128Pat<load, VectorIndexD, v2i64, i64, LD1i64>;
4472 def : Ld1Lane128Pat<load, VectorIndexD, v2f64, f64, LD1i64>;
4474 class Ld1Lane64Pat<SDPatternOperator scalar_load, Operand VecIndex,
4475 ValueType VTy, ValueType STy, Instruction LD1>
4476 : Pat<(vector_insert (VTy VecListOne64:$Rd),
4477 (STy (scalar_load am_simdnoindex:$vaddr)), VecIndex:$idx),
4479 (LD1 (SUBREG_TO_REG (i32 0), VecListOne64:$Rd, dsub),
4480 VecIndex:$idx, am_simdnoindex:$vaddr),
4483 def : Ld1Lane64Pat<extloadi8, VectorIndexB, v8i8, i32, LD1i8>;
4484 def : Ld1Lane64Pat<extloadi16, VectorIndexH, v4i16, i32, LD1i16>;
4485 def : Ld1Lane64Pat<load, VectorIndexS, v2i32, i32, LD1i32>;
4486 def : Ld1Lane64Pat<load, VectorIndexS, v2f32, f32, LD1i32>;
4489 defm LD1 : SIMDLdSt1SingleAliases<"ld1">;
4490 defm LD2 : SIMDLdSt2SingleAliases<"ld2">;
4491 defm LD3 : SIMDLdSt3SingleAliases<"ld3">;
4492 defm LD4 : SIMDLdSt4SingleAliases<"ld4">;
4495 defm ST1 : SIMDStSingleB<0, 0b000, "st1", VecListOneb, GPR64pi1>;
4496 defm ST1 : SIMDStSingleH<0, 0b010, 0, "st1", VecListOneh, GPR64pi2>;
4497 defm ST1 : SIMDStSingleS<0, 0b100, 0b00, "st1", VecListOnes, GPR64pi4>;
4498 defm ST1 : SIMDStSingleD<0, 0b100, 0b01, "st1", VecListOned, GPR64pi8>;
4500 let AddedComplexity = 8 in
4501 class St1Lane128Pat<SDPatternOperator scalar_store, Operand VecIndex,
4502 ValueType VTy, ValueType STy, Instruction ST1>
4504 (STy (vector_extract (VTy VecListOne128:$Vt), VecIndex:$idx)),
4505 am_simdnoindex:$vaddr),
4506 (ST1 VecListOne128:$Vt, VecIndex:$idx, am_simdnoindex:$vaddr)>;
4508 def : St1Lane128Pat<truncstorei8, VectorIndexB, v16i8, i32, ST1i8>;
4509 def : St1Lane128Pat<truncstorei16, VectorIndexH, v8i16, i32, ST1i16>;
4510 def : St1Lane128Pat<store, VectorIndexS, v4i32, i32, ST1i32>;
4511 def : St1Lane128Pat<store, VectorIndexS, v4f32, f32, ST1i32>;
4512 def : St1Lane128Pat<store, VectorIndexD, v2i64, i64, ST1i64>;
4513 def : St1Lane128Pat<store, VectorIndexD, v2f64, f64, ST1i64>;
4515 let AddedComplexity = 8 in
4516 class St1Lane64Pat<SDPatternOperator scalar_store, Operand VecIndex,
4517 ValueType VTy, ValueType STy, Instruction ST1>
4519 (STy (vector_extract (VTy VecListOne64:$Vt), VecIndex:$idx)),
4520 am_simdnoindex:$vaddr),
4521 (ST1 (SUBREG_TO_REG (i32 0), VecListOne64:$Vt, dsub),
4522 VecIndex:$idx, am_simdnoindex:$vaddr)>;
4524 def : St1Lane64Pat<truncstorei8, VectorIndexB, v8i8, i32, ST1i8>;
4525 def : St1Lane64Pat<truncstorei16, VectorIndexH, v4i16, i32, ST1i16>;
4526 def : St1Lane64Pat<store, VectorIndexS, v2i32, i32, ST1i32>;
4527 def : St1Lane64Pat<store, VectorIndexS, v2f32, f32, ST1i32>;
4529 multiclass St1LanePost64Pat<SDPatternOperator scalar_store, Operand VecIndex,
4530 ValueType VTy, ValueType STy, Instruction ST1,
4532 def : Pat<(scalar_store
4533 (STy (vector_extract (VTy VecListOne64:$Vt), VecIndex:$idx)),
4534 am_simdnoindex:$vaddr, offset),
4535 (ST1 (SUBREG_TO_REG (i32 0), VecListOne64:$Vt, dsub),
4536 VecIndex:$idx, am_simdnoindex:$vaddr, XZR)>;
4538 def : Pat<(scalar_store
4539 (STy (vector_extract (VTy VecListOne64:$Vt), VecIndex:$idx)),
4540 am_simdnoindex:$vaddr, GPR64:$Rm),
4541 (ST1 (SUBREG_TO_REG (i32 0), VecListOne64:$Vt, dsub),
4542 VecIndex:$idx, am_simdnoindex:$vaddr, $Rm)>;
4545 defm : St1LanePost64Pat<post_truncsti8, VectorIndexB, v8i8, i32, ST1i8_POST, 1>;
4546 defm : St1LanePost64Pat<post_truncsti16, VectorIndexH, v4i16, i32, ST1i16_POST,
4548 defm : St1LanePost64Pat<post_store, VectorIndexS, v2i32, i32, ST1i32_POST, 4>;
4549 defm : St1LanePost64Pat<post_store, VectorIndexS, v2f32, f32, ST1i32_POST, 4>;
4550 defm : St1LanePost64Pat<post_store, VectorIndexD, v1i64, i64, ST1i64_POST, 8>;
4551 defm : St1LanePost64Pat<post_store, VectorIndexD, v1f64, f64, ST1i64_POST, 8>;
4553 multiclass St1LanePost128Pat<SDPatternOperator scalar_store, Operand VecIndex,
4554 ValueType VTy, ValueType STy, Instruction ST1,
4556 def : Pat<(scalar_store
4557 (STy (vector_extract (VTy VecListOne128:$Vt), VecIndex:$idx)),
4558 am_simdnoindex:$vaddr, offset),
4559 (ST1 VecListOne128:$Vt, VecIndex:$idx, am_simdnoindex:$vaddr, XZR)>;
4561 def : Pat<(scalar_store
4562 (STy (vector_extract (VTy VecListOne128:$Vt), VecIndex:$idx)),
4563 am_simdnoindex:$vaddr, GPR64:$Rm),
4564 (ST1 VecListOne128:$Vt, VecIndex:$idx, am_simdnoindex:$vaddr, $Rm)>;
4567 defm : St1LanePost128Pat<post_truncsti8, VectorIndexB, v16i8, i32, ST1i8_POST,
4569 defm : St1LanePost128Pat<post_truncsti16, VectorIndexH, v8i16, i32, ST1i16_POST,
4571 defm : St1LanePost128Pat<post_store, VectorIndexS, v4i32, i32, ST1i32_POST, 4>;
4572 defm : St1LanePost128Pat<post_store, VectorIndexS, v4f32, f32, ST1i32_POST, 4>;
4573 defm : St1LanePost128Pat<post_store, VectorIndexD, v2i64, i64, ST1i64_POST, 8>;
4574 defm : St1LanePost128Pat<post_store, VectorIndexD, v2f64, f64, ST1i64_POST, 8>;
4576 let mayStore = 1, neverHasSideEffects = 1 in {
4577 defm ST2 : SIMDStSingleB<1, 0b000, "st2", VecListTwob, GPR64pi2>;
4578 defm ST2 : SIMDStSingleH<1, 0b010, 0, "st2", VecListTwoh, GPR64pi4>;
4579 defm ST2 : SIMDStSingleS<1, 0b100, 0b00, "st2", VecListTwos, GPR64pi8>;
4580 defm ST2 : SIMDStSingleD<1, 0b100, 0b01, "st2", VecListTwod, GPR64pi16>;
4581 defm ST3 : SIMDStSingleB<0, 0b001, "st3", VecListThreeb, GPR64pi3>;
4582 defm ST3 : SIMDStSingleH<0, 0b011, 0, "st3", VecListThreeh, GPR64pi6>;
4583 defm ST3 : SIMDStSingleS<0, 0b101, 0b00, "st3", VecListThrees, GPR64pi12>;
4584 defm ST3 : SIMDStSingleD<0, 0b101, 0b01, "st3", VecListThreed, GPR64pi24>;
4585 defm ST4 : SIMDStSingleB<1, 0b001, "st4", VecListFourb, GPR64pi4>;
4586 defm ST4 : SIMDStSingleH<1, 0b011, 0, "st4", VecListFourh, GPR64pi8>;
4587 defm ST4 : SIMDStSingleS<1, 0b101, 0b00, "st4", VecListFours, GPR64pi16>;
4588 defm ST4 : SIMDStSingleD<1, 0b101, 0b01, "st4", VecListFourd, GPR64pi32>;
4591 defm ST1 : SIMDLdSt1SingleAliases<"st1">;
4592 defm ST2 : SIMDLdSt2SingleAliases<"st2">;
4593 defm ST3 : SIMDLdSt3SingleAliases<"st3">;
4594 defm ST4 : SIMDLdSt4SingleAliases<"st4">;
4596 //----------------------------------------------------------------------------
4597 // Crypto extensions
4598 //----------------------------------------------------------------------------
4600 def AESErr : AESTiedInst<0b0100, "aese", int_arm64_crypto_aese>;
4601 def AESDrr : AESTiedInst<0b0101, "aesd", int_arm64_crypto_aesd>;
4602 def AESMCrr : AESInst< 0b0110, "aesmc", int_arm64_crypto_aesmc>;
4603 def AESIMCrr : AESInst< 0b0111, "aesimc", int_arm64_crypto_aesimc>;
4605 def SHA1Crrr : SHATiedInstQSV<0b000, "sha1c", int_arm64_crypto_sha1c>;
4606 def SHA1Prrr : SHATiedInstQSV<0b001, "sha1p", int_arm64_crypto_sha1p>;
4607 def SHA1Mrrr : SHATiedInstQSV<0b010, "sha1m", int_arm64_crypto_sha1m>;
4608 def SHA1SU0rrr : SHATiedInstVVV<0b011, "sha1su0", int_arm64_crypto_sha1su0>;
4609 def SHA256Hrrr : SHATiedInstQQV<0b100, "sha256h", int_arm64_crypto_sha256h>;
4610 def SHA256H2rrr : SHATiedInstQQV<0b101, "sha256h2",int_arm64_crypto_sha256h2>;
4611 def SHA256SU1rrr :SHATiedInstVVV<0b110, "sha256su1",int_arm64_crypto_sha256su1>;
4613 def SHA1Hrr : SHAInstSS< 0b0000, "sha1h", int_arm64_crypto_sha1h>;
4614 def SHA1SU1rr : SHATiedInstVV<0b0001, "sha1su1", int_arm64_crypto_sha1su1>;
4615 def SHA256SU0rr : SHATiedInstVV<0b0010, "sha256su0",int_arm64_crypto_sha256su0>;
4617 //----------------------------------------------------------------------------
4619 //----------------------------------------------------------------------------
4620 // FIXME: Like for X86, these should go in their own separate .td file.
4622 // Any instruction that defines a 32-bit result leaves the high half of the
4623 // register. Truncate can be lowered to EXTRACT_SUBREG. CopyFromReg may
4624 // be copying from a truncate. But any other 32-bit operation will zero-extend
4626 // FIXME: X86 also checks for CMOV here. Do we need something similar?
4627 def def32 : PatLeaf<(i32 GPR32:$src), [{
4628 return N->getOpcode() != ISD::TRUNCATE &&
4629 N->getOpcode() != TargetOpcode::EXTRACT_SUBREG &&
4630 N->getOpcode() != ISD::CopyFromReg;
4633 // In the case of a 32-bit def that is known to implicitly zero-extend,
4634 // we can use a SUBREG_TO_REG.
4635 def : Pat<(i64 (zext def32:$src)), (SUBREG_TO_REG (i64 0), GPR32:$src, sub_32)>;
4637 // For an anyext, we don't care what the high bits are, so we can perform an
4638 // INSERT_SUBREF into an IMPLICIT_DEF.
4639 def : Pat<(i64 (anyext GPR32:$src)),
4640 (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32)>;
4642 // When we need to explicitly zero-extend, we use an unsigned bitfield move
4643 // instruction (UBFM) on the enclosing super-reg.
4644 def : Pat<(i64 (zext GPR32:$src)),
4645 (UBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32), 0, 31)>;
4647 // To sign extend, we use a signed bitfield move instruction (SBFM) on the
4648 // containing super-reg.
4649 def : Pat<(i64 (sext GPR32:$src)),
4650 (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32), 0, 31)>;
4651 def : Pat<(i64 (sext_inreg GPR64:$src, i32)), (SBFMXri GPR64:$src, 0, 31)>;
4652 def : Pat<(i64 (sext_inreg GPR64:$src, i16)), (SBFMXri GPR64:$src, 0, 15)>;
4653 def : Pat<(i64 (sext_inreg GPR64:$src, i8)), (SBFMXri GPR64:$src, 0, 7)>;
4654 def : Pat<(i64 (sext_inreg GPR64:$src, i1)), (SBFMXri GPR64:$src, 0, 0)>;
4655 def : Pat<(i32 (sext_inreg GPR32:$src, i16)), (SBFMWri GPR32:$src, 0, 15)>;
4656 def : Pat<(i32 (sext_inreg GPR32:$src, i8)), (SBFMWri GPR32:$src, 0, 7)>;
4657 def : Pat<(i32 (sext_inreg GPR32:$src, i1)), (SBFMWri GPR32:$src, 0, 0)>;
4659 def : Pat<(shl (sext_inreg GPR32:$Rn, i8), (i64 imm0_31:$imm)),
4660 (SBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
4661 (i64 (i32shift_sext_i8 imm0_31:$imm)))>;
4662 def : Pat<(shl (sext_inreg GPR64:$Rn, i8), (i64 imm0_63:$imm)),
4663 (SBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
4664 (i64 (i64shift_sext_i8 imm0_63:$imm)))>;
4666 def : Pat<(shl (sext_inreg GPR32:$Rn, i16), (i64 imm0_31:$imm)),
4667 (SBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
4668 (i64 (i32shift_sext_i16 imm0_31:$imm)))>;
4669 def : Pat<(shl (sext_inreg GPR64:$Rn, i16), (i64 imm0_63:$imm)),
4670 (SBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
4671 (i64 (i64shift_sext_i16 imm0_63:$imm)))>;
4673 def : Pat<(shl (i64 (sext GPR32:$Rn)), (i64 imm0_63:$imm)),
4674 (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$Rn, sub_32),
4675 (i64 (i64shift_a imm0_63:$imm)),
4676 (i64 (i64shift_sext_i32 imm0_63:$imm)))>;
4678 // sra patterns have an AddedComplexity of 10, so make sure we have a higher
4679 // AddedComplexity for the following patterns since we want to match sext + sra
4680 // patterns before we attempt to match a single sra node.
4681 let AddedComplexity = 20 in {
4682 // We support all sext + sra combinations which preserve at least one bit of the
4683 // original value which is to be sign extended. E.g. we support shifts up to
4685 def : Pat<(sra (sext_inreg GPR32:$Rn, i8), (i64 imm0_7:$imm)),
4686 (SBFMWri GPR32:$Rn, (i64 imm0_7:$imm), 7)>;
4687 def : Pat<(sra (sext_inreg GPR64:$Rn, i8), (i64 imm0_7:$imm)),
4688 (SBFMXri GPR64:$Rn, (i64 imm0_7:$imm), 7)>;
4690 def : Pat<(sra (sext_inreg GPR32:$Rn, i16), (i64 imm0_15:$imm)),
4691 (SBFMWri GPR32:$Rn, (i64 imm0_15:$imm), 15)>;
4692 def : Pat<(sra (sext_inreg GPR64:$Rn, i16), (i64 imm0_15:$imm)),
4693 (SBFMXri GPR64:$Rn, (i64 imm0_15:$imm), 15)>;
4695 def : Pat<(sra (i64 (sext GPR32:$Rn)), (i64 imm0_31:$imm)),
4696 (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$Rn, sub_32),
4697 (i64 imm0_31:$imm), 31)>;
4698 } // AddedComplexity = 20
4700 // To truncate, we can simply extract from a subregister.
4701 def : Pat<(i32 (trunc GPR64sp:$src)),
4702 (i32 (EXTRACT_SUBREG GPR64sp:$src, sub_32))>;
4704 // __builtin_trap() uses the BRK instruction on ARM64.
4705 def : Pat<(trap), (BRK 1)>;
4707 // Conversions within AdvSIMD types in the same register size are free.
4708 // But because we need a consistent lane ordering, in big endian many
4709 // conversions require one or more REV instructions.
4711 // Consider a simple memory load followed by a bitconvert then a store.
4713 // v1 = BITCAST v2i32 v0 to v4i16
4716 // In big endian mode every memory access has an implicit byte swap. LDR and
4717 // STR do a 64-bit byte swap, whereas LD1/ST1 do a byte swap per lane - that
4718 // is, they treat the vector as a sequence of elements to be byte-swapped.
4719 // The two pairs of instructions are fundamentally incompatible. We've decided
4720 // to use LD1/ST1 only to simplify compiler implementation.
4722 // LD1/ST1 perform the equivalent of a sequence of LDR/STR + REV. This makes
4723 // the original code sequence:
4725 // v1 = REV v2i32 (implicit)
4726 // v2 = BITCAST v2i32 v1 to v4i16
4727 // v3 = REV v4i16 v2 (implicit)
4730 // But this is now broken - the value stored is different to the value loaded
4731 // due to lane reordering. To fix this, on every BITCAST we must perform two
4734 // v1 = REV v2i32 (implicit)
4736 // v3 = BITCAST v2i32 v2 to v4i16
4738 // v5 = REV v4i16 v4 (implicit)
4741 // This means an extra two instructions, but actually in most cases the two REV
4742 // instructions can be combined into one. For example:
4743 // (REV64_2s (REV64_4h X)) === (REV32_4h X)
4745 // There is also no 128-bit REV instruction. This must be synthesized with an
4748 // Most bitconverts require some sort of conversion. The only exceptions are:
4749 // a) Identity conversions - vNfX <-> vNiX
4750 // b) Single-lane-to-scalar - v1fX <-> fX or v1iX <-> iX
4753 let Predicates = [IsLE] in {
4754 def : Pat<(v8i8 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
4755 def : Pat<(v4i16 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
4756 def : Pat<(v2i32 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
4757 def : Pat<(v2f32 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
4759 def : Pat<(i64 (bitconvert (v8i8 V64:$Vn))),
4760 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
4761 def : Pat<(i64 (bitconvert (v4i16 V64:$Vn))),
4762 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
4763 def : Pat<(i64 (bitconvert (v2i32 V64:$Vn))),
4764 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
4765 def : Pat<(i64 (bitconvert (v2f32 V64:$Vn))),
4766 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
4767 def : Pat<(i64 (bitconvert (v1f64 V64:$Vn))),
4768 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
4770 let Predicates = [IsBE] in {
4771 def : Pat<(v8i8 (bitconvert GPR64:$Xn)),
4772 (REV64v8i8 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
4773 def : Pat<(v4i16 (bitconvert GPR64:$Xn)),
4774 (REV64v4i16 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
4775 def : Pat<(v2i32 (bitconvert GPR64:$Xn)),
4776 (REV64v2i32 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
4777 def : Pat<(v2f32 (bitconvert GPR64:$Xn)),
4778 (REV64v2i32 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
4780 def : Pat<(i64 (bitconvert (v8i8 V64:$Vn))),
4781 (REV64v8i8 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
4782 def : Pat<(i64 (bitconvert (v4i16 V64:$Vn))),
4783 (REV64v4i16 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
4784 def : Pat<(i64 (bitconvert (v2i32 V64:$Vn))),
4785 (REV64v2i32 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
4786 def : Pat<(i64 (bitconvert (v2f32 V64:$Vn))),
4787 (REV64v2i32 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
4789 def : Pat<(v1i64 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
4790 def : Pat<(v1f64 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
4791 def : Pat<(i64 (bitconvert (v1i64 V64:$Vn))),
4792 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
4793 def : Pat<(v1i64 (scalar_to_vector GPR64:$Xn)),
4794 (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
4795 def : Pat<(v1f64 (scalar_to_vector GPR64:$Xn)),
4796 (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
4797 def : Pat<(v1f64 (scalar_to_vector (f64 FPR64:$Xn))), (v1f64 FPR64:$Xn)>;
4799 def : Pat<(f32 (bitconvert (i32 GPR32:$Xn))),
4800 (COPY_TO_REGCLASS GPR32:$Xn, FPR32)>;
4801 def : Pat<(i32 (bitconvert (f32 FPR32:$Xn))),
4802 (COPY_TO_REGCLASS FPR32:$Xn, GPR32)>;
4803 def : Pat<(f64 (bitconvert (i64 GPR64:$Xn))),
4804 (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
4805 def : Pat<(i64 (bitconvert (f64 FPR64:$Xn))),
4806 (COPY_TO_REGCLASS FPR64:$Xn, GPR64)>;
4807 def : Pat<(i64 (bitconvert (v1f64 V64:$Vn))),
4808 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
4810 let Predicates = [IsLE] in {
4811 def : Pat<(v1i64 (bitconvert (v2i32 FPR64:$src))), (v1i64 FPR64:$src)>;
4812 def : Pat<(v1i64 (bitconvert (v4i16 FPR64:$src))), (v1i64 FPR64:$src)>;
4813 def : Pat<(v1i64 (bitconvert (v8i8 FPR64:$src))), (v1i64 FPR64:$src)>;
4814 def : Pat<(v1i64 (bitconvert (v2f32 FPR64:$src))), (v1i64 FPR64:$src)>;
4816 let Predicates = [IsBE] in {
4817 def : Pat<(v1i64 (bitconvert (v2i32 FPR64:$src))),
4818 (v1i64 (REV64v2i32 FPR64:$src))>;
4819 def : Pat<(v1i64 (bitconvert (v4i16 FPR64:$src))),
4820 (v1i64 (REV64v4i16 FPR64:$src))>;
4821 def : Pat<(v1i64 (bitconvert (v8i8 FPR64:$src))),
4822 (v1i64 (REV64v8i8 FPR64:$src))>;
4823 def : Pat<(v1i64 (bitconvert (v2f32 FPR64:$src))),
4824 (v1i64 (REV64v2i32 FPR64:$src))>;
4826 def : Pat<(v1i64 (bitconvert (v1f64 FPR64:$src))), (v1i64 FPR64:$src)>;
4827 def : Pat<(v1i64 (bitconvert (f64 FPR64:$src))), (v1i64 FPR64:$src)>;
4829 let Predicates = [IsLE] in {
4830 def : Pat<(v2i32 (bitconvert (v1i64 FPR64:$src))), (v2i32 FPR64:$src)>;
4831 def : Pat<(v2i32 (bitconvert (v4i16 FPR64:$src))), (v2i32 FPR64:$src)>;
4832 def : Pat<(v2i32 (bitconvert (v8i8 FPR64:$src))), (v2i32 FPR64:$src)>;
4833 def : Pat<(v2i32 (bitconvert (f64 FPR64:$src))), (v2i32 FPR64:$src)>;
4834 def : Pat<(v2i32 (bitconvert (v1f64 FPR64:$src))), (v2i32 FPR64:$src)>;
4836 let Predicates = [IsBE] in {
4837 def : Pat<(v2i32 (bitconvert (v1i64 FPR64:$src))),
4838 (v2i32 (REV64v2i32 FPR64:$src))>;
4839 def : Pat<(v2i32 (bitconvert (v4i16 FPR64:$src))),
4840 (v2i32 (REV32v4i16 FPR64:$src))>;
4841 def : Pat<(v2i32 (bitconvert (v8i8 FPR64:$src))),
4842 (v2i32 (REV32v8i8 FPR64:$src))>;
4843 def : Pat<(v2i32 (bitconvert (f64 FPR64:$src))),
4844 (v2i32 (REV64v2i32 FPR64:$src))>;
4845 def : Pat<(v2i32 (bitconvert (v1f64 FPR64:$src))),
4846 (v2i32 (REV64v2i32 FPR64:$src))>;
4848 def : Pat<(v2i32 (bitconvert (v2f32 FPR64:$src))), (v2i32 FPR64:$src)>;
4850 let Predicates = [IsLE] in {
4851 def : Pat<(v4i16 (bitconvert (v1i64 FPR64:$src))), (v4i16 FPR64:$src)>;
4852 def : Pat<(v4i16 (bitconvert (v2i32 FPR64:$src))), (v4i16 FPR64:$src)>;
4853 def : Pat<(v4i16 (bitconvert (v8i8 FPR64:$src))), (v4i16 FPR64:$src)>;
4854 def : Pat<(v4i16 (bitconvert (f64 FPR64:$src))), (v4i16 FPR64:$src)>;
4855 def : Pat<(v4i16 (bitconvert (v2f32 FPR64:$src))), (v4i16 FPR64:$src)>;
4856 def : Pat<(v4i16 (bitconvert (v1f64 FPR64:$src))), (v4i16 FPR64:$src)>;
4858 let Predicates = [IsBE] in {
4859 def : Pat<(v4i16 (bitconvert (v1i64 FPR64:$src))),
4860 (v4i16 (REV64v4i16 FPR64:$src))>;
4861 def : Pat<(v4i16 (bitconvert (v2i32 FPR64:$src))),
4862 (v4i16 (REV32v4i16 FPR64:$src))>;
4863 def : Pat<(v4i16 (bitconvert (v8i8 FPR64:$src))),
4864 (v4i16 (REV16v8i8 FPR64:$src))>;
4865 def : Pat<(v4i16 (bitconvert (f64 FPR64:$src))),
4866 (v4i16 (REV64v4i16 FPR64:$src))>;
4867 def : Pat<(v4i16 (bitconvert (v2f32 FPR64:$src))),
4868 (v4i16 (REV32v4i16 FPR64:$src))>;
4869 def : Pat<(v4i16 (bitconvert (v1f64 FPR64:$src))),
4870 (v4i16 (REV64v4i16 FPR64:$src))>;
4873 let Predicates = [IsLE] in {
4874 def : Pat<(v8i8 (bitconvert (v1i64 FPR64:$src))), (v8i8 FPR64:$src)>;
4875 def : Pat<(v8i8 (bitconvert (v2i32 FPR64:$src))), (v8i8 FPR64:$src)>;
4876 def : Pat<(v8i8 (bitconvert (v4i16 FPR64:$src))), (v8i8 FPR64:$src)>;
4877 def : Pat<(v8i8 (bitconvert (f64 FPR64:$src))), (v8i8 FPR64:$src)>;
4878 def : Pat<(v8i8 (bitconvert (v2f32 FPR64:$src))), (v8i8 FPR64:$src)>;
4879 def : Pat<(v8i8 (bitconvert (v1f64 FPR64:$src))), (v8i8 FPR64:$src)>;
4881 let Predicates = [IsBE] in {
4882 def : Pat<(v8i8 (bitconvert (v1i64 FPR64:$src))),
4883 (v8i8 (REV64v8i8 FPR64:$src))>;
4884 def : Pat<(v8i8 (bitconvert (v2i32 FPR64:$src))),
4885 (v8i8 (REV32v8i8 FPR64:$src))>;
4886 def : Pat<(v8i8 (bitconvert (v4i16 FPR64:$src))),
4887 (v8i8 (REV16v8i8 FPR64:$src))>;
4888 def : Pat<(v8i8 (bitconvert (f64 FPR64:$src))),
4889 (v8i8 (REV64v8i8 FPR64:$src))>;
4890 def : Pat<(v8i8 (bitconvert (v2f32 FPR64:$src))),
4891 (v8i8 (REV32v8i8 FPR64:$src))>;
4892 def : Pat<(v8i8 (bitconvert (v1f64 FPR64:$src))),
4893 (v8i8 (REV64v8i8 FPR64:$src))>;
4896 let Predicates = [IsLE] in {
4897 def : Pat<(f64 (bitconvert (v2i32 FPR64:$src))), (f64 FPR64:$src)>;
4898 def : Pat<(f64 (bitconvert (v4i16 FPR64:$src))), (f64 FPR64:$src)>;
4899 def : Pat<(f64 (bitconvert (v2f32 FPR64:$src))), (f64 FPR64:$src)>;
4900 def : Pat<(f64 (bitconvert (v8i8 FPR64:$src))), (f64 FPR64:$src)>;
4902 let Predicates = [IsBE] in {
4903 def : Pat<(f64 (bitconvert (v2i32 FPR64:$src))),
4904 (f64 (REV64v2i32 FPR64:$src))>;
4905 def : Pat<(f64 (bitconvert (v4i16 FPR64:$src))),
4906 (f64 (REV64v4i16 FPR64:$src))>;
4907 def : Pat<(f64 (bitconvert (v2f32 FPR64:$src))),
4908 (f64 (REV64v2i32 FPR64:$src))>;
4909 def : Pat<(f64 (bitconvert (v8i8 FPR64:$src))),
4910 (f64 (REV64v8i8 FPR64:$src))>;
4912 def : Pat<(f64 (bitconvert (v1i64 FPR64:$src))), (f64 FPR64:$src)>;
4913 def : Pat<(f64 (bitconvert (v1f64 FPR64:$src))), (f64 FPR64:$src)>;
4915 let Predicates = [IsLE] in {
4916 def : Pat<(v1f64 (bitconvert (v2i32 FPR64:$src))), (v1f64 FPR64:$src)>;
4917 def : Pat<(v1f64 (bitconvert (v4i16 FPR64:$src))), (v1f64 FPR64:$src)>;
4918 def : Pat<(v1f64 (bitconvert (v8i8 FPR64:$src))), (v1f64 FPR64:$src)>;
4919 def : Pat<(v1f64 (bitconvert (v2f32 FPR64:$src))), (v1f64 FPR64:$src)>;
4921 let Predicates = [IsBE] in {
4922 def : Pat<(v1f64 (bitconvert (v2i32 FPR64:$src))),
4923 (v1f64 (REV64v2i32 FPR64:$src))>;
4924 def : Pat<(v1f64 (bitconvert (v4i16 FPR64:$src))),
4925 (v1f64 (REV64v4i16 FPR64:$src))>;
4926 def : Pat<(v1f64 (bitconvert (v8i8 FPR64:$src))),
4927 (v1f64 (REV64v8i8 FPR64:$src))>;
4928 def : Pat<(v1f64 (bitconvert (v2f32 FPR64:$src))),
4929 (v1f64 (REV64v2i32 FPR64:$src))>;
4931 def : Pat<(v1f64 (bitconvert (v1i64 FPR64:$src))), (v1f64 FPR64:$src)>;
4932 def : Pat<(v1f64 (bitconvert (f64 FPR64:$src))), (v1f64 FPR64:$src)>;
4934 let Predicates = [IsLE] in {
4935 def : Pat<(v2f32 (bitconvert (v1i64 FPR64:$src))), (v2f32 FPR64:$src)>;
4936 def : Pat<(v2f32 (bitconvert (v4i16 FPR64:$src))), (v2f32 FPR64:$src)>;
4937 def : Pat<(v2f32 (bitconvert (v8i8 FPR64:$src))), (v2f32 FPR64:$src)>;
4938 def : Pat<(v2f32 (bitconvert (v1f64 FPR64:$src))), (v2f32 FPR64:$src)>;
4939 def : Pat<(v2f32 (bitconvert (f64 FPR64:$src))), (v2f32 FPR64:$src)>;
4941 let Predicates = [IsBE] in {
4942 def : Pat<(v2f32 (bitconvert (v1i64 FPR64:$src))),
4943 (v2f32 (REV64v2i32 FPR64:$src))>;
4944 def : Pat<(v2f32 (bitconvert (v4i16 FPR64:$src))),
4945 (v2f32 (REV32v4i16 FPR64:$src))>;
4946 def : Pat<(v2f32 (bitconvert (v8i8 FPR64:$src))),
4947 (v2f32 (REV32v8i8 FPR64:$src))>;
4948 def : Pat<(v2f32 (bitconvert (v1f64 FPR64:$src))),
4949 (v2f32 (REV64v2i32 FPR64:$src))>;
4950 def : Pat<(v2f32 (bitconvert (f64 FPR64:$src))),
4951 (v2f32 (REV64v2i32 FPR64:$src))>;
4953 def : Pat<(v2f32 (bitconvert (v2i32 FPR64:$src))), (v2f32 FPR64:$src)>;
4955 let Predicates = [IsLE] in {
4956 def : Pat<(f128 (bitconvert (v2i64 FPR128:$src))), (f128 FPR128:$src)>;
4957 def : Pat<(f128 (bitconvert (v4i32 FPR128:$src))), (f128 FPR128:$src)>;
4958 def : Pat<(f128 (bitconvert (v8i16 FPR128:$src))), (f128 FPR128:$src)>;
4959 def : Pat<(f128 (bitconvert (v2f64 FPR128:$src))), (f128 FPR128:$src)>;
4960 def : Pat<(f128 (bitconvert (v4f32 FPR128:$src))), (f128 FPR128:$src)>;
4961 def : Pat<(f128 (bitconvert (v16i8 FPR128:$src))), (f128 FPR128:$src)>;
4963 let Predicates = [IsBE] in {
4964 def : Pat<(f128 (bitconvert (v2i64 FPR128:$src))),
4965 (f128 (EXTv16i8 FPR128:$src, FPR128:$src, (i32 8)))>;
4966 def : Pat<(f128 (bitconvert (v4i32 FPR128:$src))),
4967 (f128 (EXTv16i8 (REV64v4i32 FPR128:$src),
4968 (REV64v4i32 FPR128:$src), (i32 8)))>;
4969 def : Pat<(f128 (bitconvert (v8i16 FPR128:$src))),
4970 (f128 (EXTv16i8 (REV64v8i16 FPR128:$src),
4971 (REV64v8i16 FPR128:$src), (i32 8)))>;
4972 def : Pat<(f128 (bitconvert (v2f64 FPR128:$src))),
4973 (f128 (EXTv16i8 FPR128:$src, FPR128:$src, (i32 8)))>;
4974 def : Pat<(f128 (bitconvert (v4f32 FPR128:$src))),
4975 (f128 (EXTv16i8 (REV64v4i32 FPR128:$src),
4976 (REV64v4i32 FPR128:$src), (i32 8)))>;
4977 def : Pat<(f128 (bitconvert (v16i8 FPR128:$src))),
4978 (f128 (EXTv16i8 (REV64v16i8 FPR128:$src),
4979 (REV64v16i8 FPR128:$src), (i32 8)))>;
4982 let Predicates = [IsLE] in {
4983 def : Pat<(v2f64 (bitconvert (f128 FPR128:$src))), (v2f64 FPR128:$src)>;
4984 def : Pat<(v2f64 (bitconvert (v4i32 FPR128:$src))), (v2f64 FPR128:$src)>;
4985 def : Pat<(v2f64 (bitconvert (v8i16 FPR128:$src))), (v2f64 FPR128:$src)>;
4986 def : Pat<(v2f64 (bitconvert (v16i8 FPR128:$src))), (v2f64 FPR128:$src)>;
4987 def : Pat<(v2f64 (bitconvert (v4f32 FPR128:$src))), (v2f64 FPR128:$src)>;
4989 let Predicates = [IsBE] in {
4990 def : Pat<(v2f64 (bitconvert (f128 FPR128:$src))),
4991 (v2f64 (EXTv16i8 FPR128:$src,
4992 FPR128:$src, (i32 8)))>;
4993 def : Pat<(v2f64 (bitconvert (v4i32 FPR128:$src))),
4994 (v2f64 (REV64v4i32 FPR128:$src))>;
4995 def : Pat<(v2f64 (bitconvert (v8i16 FPR128:$src))),
4996 (v2f64 (REV64v8i16 FPR128:$src))>;
4997 def : Pat<(v2f64 (bitconvert (v16i8 FPR128:$src))),
4998 (v2f64 (REV64v16i8 FPR128:$src))>;
4999 def : Pat<(v2f64 (bitconvert (v4f32 FPR128:$src))),
5000 (v2f64 (REV64v4i32 FPR128:$src))>;
5002 def : Pat<(v2f64 (bitconvert (v2i64 FPR128:$src))), (v2f64 FPR128:$src)>;
5004 let Predicates = [IsLE] in {
5005 def : Pat<(v4f32 (bitconvert (f128 FPR128:$src))), (v4f32 FPR128:$src)>;
5006 def : Pat<(v4f32 (bitconvert (v8i16 FPR128:$src))), (v4f32 FPR128:$src)>;
5007 def : Pat<(v4f32 (bitconvert (v16i8 FPR128:$src))), (v4f32 FPR128:$src)>;
5008 def : Pat<(v4f32 (bitconvert (v2i64 FPR128:$src))), (v4f32 FPR128:$src)>;
5009 def : Pat<(v4f32 (bitconvert (v2f64 FPR128:$src))), (v4f32 FPR128:$src)>;
5011 let Predicates = [IsBE] in {
5012 def : Pat<(v4f32 (bitconvert (f128 FPR128:$src))),
5013 (v4f32 (EXTv16i8 (REV64v4i32 FPR128:$src),
5014 (REV64v4i32 FPR128:$src), (i32 8)))>;
5015 def : Pat<(v4f32 (bitconvert (v8i16 FPR128:$src))),
5016 (v4f32 (REV32v8i16 FPR128:$src))>;
5017 def : Pat<(v4f32 (bitconvert (v16i8 FPR128:$src))),
5018 (v4f32 (REV32v16i8 FPR128:$src))>;
5019 def : Pat<(v4f32 (bitconvert (v2i64 FPR128:$src))),
5020 (v4f32 (REV64v4i32 FPR128:$src))>;
5021 def : Pat<(v4f32 (bitconvert (v2f64 FPR128:$src))),
5022 (v4f32 (REV64v4i32 FPR128:$src))>;
5024 def : Pat<(v4f32 (bitconvert (v4i32 FPR128:$src))), (v4f32 FPR128:$src)>;
5026 let Predicates = [IsLE] in {
5027 def : Pat<(v2i64 (bitconvert (f128 FPR128:$src))), (v2i64 FPR128:$src)>;
5028 def : Pat<(v2i64 (bitconvert (v4i32 FPR128:$src))), (v2i64 FPR128:$src)>;
5029 def : Pat<(v2i64 (bitconvert (v8i16 FPR128:$src))), (v2i64 FPR128:$src)>;
5030 def : Pat<(v2i64 (bitconvert (v16i8 FPR128:$src))), (v2i64 FPR128:$src)>;
5031 def : Pat<(v2i64 (bitconvert (v4f32 FPR128:$src))), (v2i64 FPR128:$src)>;
5033 let Predicates = [IsBE] in {
5034 def : Pat<(v2i64 (bitconvert (f128 FPR128:$src))),
5035 (v2i64 (EXTv16i8 FPR128:$src,
5036 FPR128:$src, (i32 8)))>;
5037 def : Pat<(v2i64 (bitconvert (v4i32 FPR128:$src))),
5038 (v2i64 (REV64v4i32 FPR128:$src))>;
5039 def : Pat<(v2i64 (bitconvert (v8i16 FPR128:$src))),
5040 (v2i64 (REV64v8i16 FPR128:$src))>;
5041 def : Pat<(v2i64 (bitconvert (v16i8 FPR128:$src))),
5042 (v2i64 (REV64v16i8 FPR128:$src))>;
5043 def : Pat<(v2i64 (bitconvert (v4f32 FPR128:$src))),
5044 (v2i64 (REV64v4i32 FPR128:$src))>;
5046 def : Pat<(v2i64 (bitconvert (v2f64 FPR128:$src))), (v2i64 FPR128:$src)>;
5048 let Predicates = [IsLE] in {
5049 def : Pat<(v4i32 (bitconvert (f128 FPR128:$src))), (v4i32 FPR128:$src)>;
5050 def : Pat<(v4i32 (bitconvert (v2i64 FPR128:$src))), (v4i32 FPR128:$src)>;
5051 def : Pat<(v4i32 (bitconvert (v8i16 FPR128:$src))), (v4i32 FPR128:$src)>;
5052 def : Pat<(v4i32 (bitconvert (v16i8 FPR128:$src))), (v4i32 FPR128:$src)>;
5053 def : Pat<(v4i32 (bitconvert (v2f64 FPR128:$src))), (v4i32 FPR128:$src)>;
5055 let Predicates = [IsBE] in {
5056 def : Pat<(v4i32 (bitconvert (f128 FPR128:$src))),
5057 (v4i32 (EXTv16i8 (REV64v4i32 FPR128:$src),
5058 (REV64v4i32 FPR128:$src),
5060 def : Pat<(v4i32 (bitconvert (v2i64 FPR128:$src))),
5061 (v4i32 (REV64v4i32 FPR128:$src))>;
5062 def : Pat<(v4i32 (bitconvert (v8i16 FPR128:$src))),
5063 (v4i32 (REV32v8i16 FPR128:$src))>;
5064 def : Pat<(v4i32 (bitconvert (v16i8 FPR128:$src))),
5065 (v4i32 (REV32v16i8 FPR128:$src))>;
5066 def : Pat<(v4i32 (bitconvert (v2f64 FPR128:$src))),
5067 (v4i32 (REV64v4i32 FPR128:$src))>;
5069 def : Pat<(v4i32 (bitconvert (v4f32 FPR128:$src))), (v4i32 FPR128:$src)>;
5071 let Predicates = [IsLE] in {
5072 def : Pat<(v8i16 (bitconvert (f128 FPR128:$src))), (v8i16 FPR128:$src)>;
5073 def : Pat<(v8i16 (bitconvert (v2i64 FPR128:$src))), (v8i16 FPR128:$src)>;
5074 def : Pat<(v8i16 (bitconvert (v4i32 FPR128:$src))), (v8i16 FPR128:$src)>;
5075 def : Pat<(v8i16 (bitconvert (v16i8 FPR128:$src))), (v8i16 FPR128:$src)>;
5076 def : Pat<(v8i16 (bitconvert (v2f64 FPR128:$src))), (v8i16 FPR128:$src)>;
5077 def : Pat<(v8i16 (bitconvert (v4f32 FPR128:$src))), (v8i16 FPR128:$src)>;
5079 let Predicates = [IsBE] in {
5080 def : Pat<(v8i16 (bitconvert (f128 FPR128:$src))),
5081 (v8i16 (EXTv16i8 (REV64v8i16 FPR128:$src),
5082 (REV64v8i16 FPR128:$src),
5084 def : Pat<(v8i16 (bitconvert (v2i64 FPR128:$src))),
5085 (v8i16 (REV64v8i16 FPR128:$src))>;
5086 def : Pat<(v8i16 (bitconvert (v4i32 FPR128:$src))),
5087 (v8i16 (REV32v8i16 FPR128:$src))>;
5088 def : Pat<(v8i16 (bitconvert (v16i8 FPR128:$src))),
5089 (v8i16 (REV16v16i8 FPR128:$src))>;
5090 def : Pat<(v8i16 (bitconvert (v2f64 FPR128:$src))),
5091 (v8i16 (REV64v8i16 FPR128:$src))>;
5092 def : Pat<(v8i16 (bitconvert (v4f32 FPR128:$src))),
5093 (v8i16 (REV32v8i16 FPR128:$src))>;
5096 let Predicates = [IsLE] in {
5097 def : Pat<(v16i8 (bitconvert (f128 FPR128:$src))), (v16i8 FPR128:$src)>;
5098 def : Pat<(v16i8 (bitconvert (v2i64 FPR128:$src))), (v16i8 FPR128:$src)>;
5099 def : Pat<(v16i8 (bitconvert (v4i32 FPR128:$src))), (v16i8 FPR128:$src)>;
5100 def : Pat<(v16i8 (bitconvert (v8i16 FPR128:$src))), (v16i8 FPR128:$src)>;
5101 def : Pat<(v16i8 (bitconvert (v2f64 FPR128:$src))), (v16i8 FPR128:$src)>;
5102 def : Pat<(v16i8 (bitconvert (v4f32 FPR128:$src))), (v16i8 FPR128:$src)>;
5104 let Predicates = [IsBE] in {
5105 def : Pat<(v16i8 (bitconvert (f128 FPR128:$src))),
5106 (v16i8 (EXTv16i8 (REV64v16i8 FPR128:$src),
5107 (REV64v16i8 FPR128:$src),
5109 def : Pat<(v16i8 (bitconvert (v2i64 FPR128:$src))),
5110 (v16i8 (REV64v16i8 FPR128:$src))>;
5111 def : Pat<(v16i8 (bitconvert (v4i32 FPR128:$src))),
5112 (v16i8 (REV32v16i8 FPR128:$src))>;
5113 def : Pat<(v16i8 (bitconvert (v8i16 FPR128:$src))),
5114 (v16i8 (REV16v16i8 FPR128:$src))>;
5115 def : Pat<(v16i8 (bitconvert (v2f64 FPR128:$src))),
5116 (v16i8 (REV64v16i8 FPR128:$src))>;
5117 def : Pat<(v16i8 (bitconvert (v4f32 FPR128:$src))),
5118 (v16i8 (REV32v16i8 FPR128:$src))>;
5121 def : Pat<(v8i8 (extract_subvector (v16i8 FPR128:$Rn), (i64 1))),
5122 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
5123 def : Pat<(v4i16 (extract_subvector (v8i16 FPR128:$Rn), (i64 1))),
5124 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
5125 def : Pat<(v2i32 (extract_subvector (v4i32 FPR128:$Rn), (i64 1))),
5126 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
5127 def : Pat<(v1i64 (extract_subvector (v2i64 FPR128:$Rn), (i64 1))),
5128 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
5130 // A 64-bit subvector insert to the first 128-bit vector position
5131 // is a subregister copy that needs no instruction.
5132 def : Pat<(insert_subvector undef, (v1i64 FPR64:$src), (i32 0)),
5133 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5134 def : Pat<(insert_subvector undef, (v1f64 FPR64:$src), (i32 0)),
5135 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5136 def : Pat<(insert_subvector undef, (v2i32 FPR64:$src), (i32 0)),
5137 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5138 def : Pat<(insert_subvector undef, (v2f32 FPR64:$src), (i32 0)),
5139 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5140 def : Pat<(insert_subvector undef, (v4i16 FPR64:$src), (i32 0)),
5141 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5142 def : Pat<(insert_subvector undef, (v8i8 FPR64:$src), (i32 0)),
5143 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5145 // Use pair-wise add instructions when summing up the lanes for v2f64, v2i64
5147 def : Pat<(i64 (add (vector_extract (v2i64 FPR128:$Rn), (i64 0)),
5148 (vector_extract (v2i64 FPR128:$Rn), (i64 1)))),
5149 (i64 (ADDPv2i64p (v2i64 FPR128:$Rn)))>;
5150 def : Pat<(f64 (fadd (vector_extract (v2f64 FPR128:$Rn), (i64 0)),
5151 (vector_extract (v2f64 FPR128:$Rn), (i64 1)))),
5152 (f64 (FADDPv2i64p (v2f64 FPR128:$Rn)))>;
5153 // vector_extract on 64-bit vectors gets promoted to a 128 bit vector,
5154 // so we match on v4f32 here, not v2f32. This will also catch adding
5155 // the low two lanes of a true v4f32 vector.
5156 def : Pat<(fadd (vector_extract (v4f32 FPR128:$Rn), (i64 0)),
5157 (vector_extract (v4f32 FPR128:$Rn), (i64 1))),
5158 (f32 (FADDPv2i32p (EXTRACT_SUBREG FPR128:$Rn, dsub)))>;
5160 // Scalar 64-bit shifts in FPR64 registers.
5161 def : Pat<(i64 (int_arm64_neon_sshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
5162 (SSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
5163 def : Pat<(i64 (int_arm64_neon_ushl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
5164 (USHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
5165 def : Pat<(i64 (int_arm64_neon_srshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
5166 (SRSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
5167 def : Pat<(i64 (int_arm64_neon_urshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
5168 (URSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
5170 // Tail call return handling. These are all compiler pseudo-instructions,
5171 // so no encoding information or anything like that.
5172 let isCall = 1, isTerminator = 1, isReturn = 1, isBarrier = 1, Uses = [SP] in {
5173 def TCRETURNdi : Pseudo<(outs), (ins i64imm:$dst, i32imm:$FPDiff),[]>;
5174 def TCRETURNri : Pseudo<(outs), (ins tcGPR64:$dst, i32imm:$FPDiff), []>;
5177 def : Pat<(ARM64tcret tcGPR64:$dst, (i32 timm:$FPDiff)),
5178 (TCRETURNri tcGPR64:$dst, imm:$FPDiff)>;
5179 def : Pat<(ARM64tcret tglobaladdr:$dst, (i32 timm:$FPDiff)),
5180 (TCRETURNdi texternalsym:$dst, imm:$FPDiff)>;
5181 def : Pat<(ARM64tcret texternalsym:$dst, (i32 timm:$FPDiff)),
5182 (TCRETURNdi texternalsym:$dst, imm:$FPDiff)>;
5184 include "ARM64InstrAtomics.td"