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 defm LDPW : LoadPairOffset<0b00, 0, GPR32, simm7s4, "ldp">;
1068 defm LDPX : LoadPairOffset<0b10, 0, GPR64, simm7s8, "ldp">;
1069 defm LDPS : LoadPairOffset<0b00, 1, FPR32, simm7s4, "ldp">;
1070 defm LDPD : LoadPairOffset<0b01, 1, FPR64, simm7s8, "ldp">;
1071 defm LDPQ : LoadPairOffset<0b10, 1, FPR128, simm7s16, "ldp">;
1073 defm LDPSW : LoadPairOffset<0b01, 0, GPR64, simm7s4, "ldpsw">;
1075 // Pair (pre-indexed)
1076 def LDPWpre : LoadPairPreIdx<0b00, 0, GPR32, simm7s4, "ldp">;
1077 def LDPXpre : LoadPairPreIdx<0b10, 0, GPR64, simm7s8, "ldp">;
1078 def LDPSpre : LoadPairPreIdx<0b00, 1, FPR32, simm7s4, "ldp">;
1079 def LDPDpre : LoadPairPreIdx<0b01, 1, FPR64, simm7s8, "ldp">;
1080 def LDPQpre : LoadPairPreIdx<0b10, 1, FPR128, simm7s16, "ldp">;
1082 def LDPSWpre : LoadPairPreIdx<0b01, 0, GPR64, simm7s4, "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 defm LDNPW : LoadPairNoAlloc<0b00, 0, GPR32, simm7s4, "ldnp">;
1096 defm LDNPX : LoadPairNoAlloc<0b10, 0, GPR64, simm7s8, "ldnp">;
1097 defm LDNPS : LoadPairNoAlloc<0b00, 1, FPR32, simm7s4, "ldnp">;
1098 defm LDNPD : LoadPairNoAlloc<0b01, 1, FPR64, simm7s8, "ldnp">;
1099 defm LDNPQ : LoadPairNoAlloc<0b10, 1, FPR128, simm7s16, "ldnp">;
1102 // (register offset)
1106 defm LDRBB : Load8RO<0b00, 0, 0b01, GPR32, "ldrb", i32, zextloadi8>;
1107 defm LDRHH : Load16RO<0b01, 0, 0b01, GPR32, "ldrh", i32, zextloadi16>;
1108 defm LDRW : Load32RO<0b10, 0, 0b01, GPR32, "ldr", i32, load>;
1109 defm LDRX : Load64RO<0b11, 0, 0b01, GPR64, "ldr", i64, load>;
1112 defm LDRB : Load8RO<0b00, 1, 0b01, FPR8, "ldr", untyped, load>;
1113 defm LDRH : Load16RO<0b01, 1, 0b01, FPR16, "ldr", f16, load>;
1114 defm LDRS : Load32RO<0b10, 1, 0b01, FPR32, "ldr", f32, load>;
1115 defm LDRD : Load64RO<0b11, 1, 0b01, FPR64, "ldr", f64, load>;
1116 defm LDRQ : Load128RO<0b00, 1, 0b11, FPR128, "ldr", f128, load>;
1118 // Load sign-extended half-word
1119 defm LDRSHW : Load16RO<0b01, 0, 0b11, GPR32, "ldrsh", i32, sextloadi16>;
1120 defm LDRSHX : Load16RO<0b01, 0, 0b10, GPR64, "ldrsh", i64, sextloadi16>;
1122 // Load sign-extended byte
1123 defm LDRSBW : Load8RO<0b00, 0, 0b11, GPR32, "ldrsb", i32, sextloadi8>;
1124 defm LDRSBX : Load8RO<0b00, 0, 0b10, GPR64, "ldrsb", i64, sextloadi8>;
1126 // Load sign-extended word
1127 defm LDRSW : Load32RO<0b10, 0, 0b10, GPR64, "ldrsw", i64, sextloadi32>;
1130 defm PRFM : PrefetchRO<0b11, 0, 0b10, "prfm">;
1132 // For regular load, we do not have any alignment requirement.
1133 // Thus, it is safe to directly map the vector loads with interesting
1134 // addressing modes.
1135 // FIXME: We could do the same for bitconvert to floating point vectors.
1136 multiclass ScalToVecROLoadPat<ROAddrMode ro, SDPatternOperator loadop,
1137 ValueType ScalTy, ValueType VecTy,
1138 Instruction LOADW, Instruction LOADX,
1140 def : Pat<(VecTy (scalar_to_vector (ScalTy
1141 (loadop (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$offset))))),
1142 (INSERT_SUBREG (VecTy (IMPLICIT_DEF)),
1143 (LOADW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$offset),
1146 def : Pat<(VecTy (scalar_to_vector (ScalTy
1147 (loadop (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$offset))))),
1148 (INSERT_SUBREG (VecTy (IMPLICIT_DEF)),
1149 (LOADX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$offset),
1153 let AddedComplexity = 10 in {
1154 defm : ScalToVecROLoadPat<ro8, extloadi8, i32, v8i8, LDRBroW, LDRBroX, bsub>;
1155 defm : ScalToVecROLoadPat<ro8, extloadi8, i32, v16i8, LDRBroW, LDRBroX, bsub>;
1157 defm : ScalToVecROLoadPat<ro16, extloadi16, i32, v4i16, LDRHroW, LDRHroX, hsub>;
1158 defm : ScalToVecROLoadPat<ro16, extloadi16, i32, v8i16, LDRHroW, LDRHroX, hsub>;
1160 defm : ScalToVecROLoadPat<ro32, load, i32, v2i32, LDRSroW, LDRSroX, ssub>;
1161 defm : ScalToVecROLoadPat<ro32, load, i32, v4i32, LDRSroW, LDRSroX, ssub>;
1163 defm : ScalToVecROLoadPat<ro32, load, f32, v2f32, LDRSroW, LDRSroX, ssub>;
1164 defm : ScalToVecROLoadPat<ro32, load, f32, v4f32, LDRSroW, LDRSroX, ssub>;
1166 defm : ScalToVecROLoadPat<ro64, load, i64, v2i64, LDRDroW, LDRDroX, dsub>;
1168 defm : ScalToVecROLoadPat<ro64, load, f64, v2f64, LDRDroW, LDRDroX, dsub>;
1171 def : Pat <(v1i64 (scalar_to_vector (i64
1172 (load (ro_Windexed64 GPR64sp:$Rn, GPR32:$Rm,
1173 ro_Wextend64:$extend))))),
1174 (LDRDroW GPR64sp:$Rn, GPR32:$Rm, ro_Wextend64:$extend)>;
1176 def : Pat <(v1i64 (scalar_to_vector (i64
1177 (load (ro_Xindexed64 GPR64sp:$Rn, GPR64:$Rm,
1178 ro_Xextend64:$extend))))),
1179 (LDRDroX GPR64sp:$Rn, GPR64:$Rm, ro_Xextend64:$extend)>;
1182 // Match all load 64 bits width whose type is compatible with FPR64
1183 multiclass VecROLoadPat<ROAddrMode ro, ValueType VecTy,
1184 Instruction LOADW, Instruction LOADX> {
1186 def : Pat<(VecTy (load (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend))),
1187 (LOADW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
1189 def : Pat<(VecTy (load (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend))),
1190 (LOADX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
1193 let AddedComplexity = 10 in {
1194 let Predicates = [IsLE] in {
1195 // We must do vector loads with LD1 in big-endian.
1196 defm : VecROLoadPat<ro64, v2i32, LDRDroW, LDRDroX>;
1197 defm : VecROLoadPat<ro64, v2f32, LDRDroW, LDRDroX>;
1198 defm : VecROLoadPat<ro64, v8i8, LDRDroW, LDRDroX>;
1199 defm : VecROLoadPat<ro64, v4i16, LDRDroW, LDRDroX>;
1202 defm : VecROLoadPat<ro64, v1i64, LDRDroW, LDRDroX>;
1203 defm : VecROLoadPat<ro64, v1f64, LDRDroW, LDRDroX>;
1205 // Match all load 128 bits width whose type is compatible with FPR128
1206 let Predicates = [IsLE] in {
1207 // We must do vector loads with LD1 in big-endian.
1208 defm : VecROLoadPat<ro128, v2i64, LDRQroW, LDRQroX>;
1209 defm : VecROLoadPat<ro128, v2f64, LDRQroW, LDRQroX>;
1210 defm : VecROLoadPat<ro128, v4i32, LDRQroW, LDRQroX>;
1211 defm : VecROLoadPat<ro128, v4f32, LDRQroW, LDRQroX>;
1212 defm : VecROLoadPat<ro128, v8i16, LDRQroW, LDRQroX>;
1213 defm : VecROLoadPat<ro128, v16i8, LDRQroW, LDRQroX>;
1215 } // AddedComplexity = 10
1218 multiclass ExtLoadTo64ROPat<ROAddrMode ro, SDPatternOperator loadop,
1219 Instruction INSTW, Instruction INSTX> {
1220 def : Pat<(i64 (loadop (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend))),
1221 (SUBREG_TO_REG (i64 0),
1222 (INSTW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend),
1225 def : Pat<(i64 (loadop (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend))),
1226 (SUBREG_TO_REG (i64 0),
1227 (INSTX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend),
1231 let AddedComplexity = 10 in {
1232 defm : ExtLoadTo64ROPat<ro8, zextloadi8, LDRBBroW, LDRBBroX>;
1233 defm : ExtLoadTo64ROPat<ro16, zextloadi16, LDRHHroW, LDRHHroX>;
1234 defm : ExtLoadTo64ROPat<ro32, zextloadi32, LDRWroW, LDRWroX>;
1236 // zextloadi1 -> zextloadi8
1237 defm : ExtLoadTo64ROPat<ro8, zextloadi1, LDRBBroW, LDRBBroX>;
1239 // extload -> zextload
1240 defm : ExtLoadTo64ROPat<ro8, extloadi8, LDRBBroW, LDRBBroX>;
1241 defm : ExtLoadTo64ROPat<ro16, extloadi16, LDRHHroW, LDRHHroX>;
1242 defm : ExtLoadTo64ROPat<ro32, extloadi32, LDRWroW, LDRWroX>;
1244 // extloadi1 -> zextloadi8
1245 defm : ExtLoadTo64ROPat<ro8, extloadi1, LDRBBroW, LDRBBroX>;
1250 multiclass ExtLoadTo32ROPat<ROAddrMode ro, SDPatternOperator loadop,
1251 Instruction INSTW, Instruction INSTX> {
1252 def : Pat<(i32 (loadop (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend))),
1253 (INSTW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
1255 def : Pat<(i32 (loadop (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend))),
1256 (INSTX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
1260 let AddedComplexity = 10 in {
1261 // extload -> zextload
1262 defm : ExtLoadTo32ROPat<ro8, extloadi8, LDRBBroW, LDRBBroX>;
1263 defm : ExtLoadTo32ROPat<ro16, extloadi16, LDRHHroW, LDRHHroX>;
1264 defm : ExtLoadTo32ROPat<ro32, extloadi32, LDRWroW, LDRWroX>;
1266 // zextloadi1 -> zextloadi8
1267 defm : ExtLoadTo32ROPat<ro8, zextloadi1, LDRBBroW, LDRBBroX>;
1271 // (unsigned immediate)
1273 defm LDRX : LoadUI<0b11, 0, 0b01, GPR64, uimm12s8, "ldr",
1275 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)))]>;
1276 defm LDRW : LoadUI<0b10, 0, 0b01, GPR32, uimm12s4, "ldr",
1278 (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset)))]>;
1279 defm LDRB : LoadUI<0b00, 1, 0b01, FPR8, uimm12s1, "ldr",
1281 (load (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset)))]>;
1282 defm LDRH : LoadUI<0b01, 1, 0b01, FPR16, uimm12s2, "ldr",
1283 [(set (f16 FPR16:$Rt),
1284 (load (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset)))]>;
1285 defm LDRS : LoadUI<0b10, 1, 0b01, FPR32, uimm12s4, "ldr",
1286 [(set (f32 FPR32:$Rt),
1287 (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset)))]>;
1288 defm LDRD : LoadUI<0b11, 1, 0b01, FPR64, uimm12s8, "ldr",
1289 [(set (f64 FPR64:$Rt),
1290 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)))]>;
1291 defm LDRQ : LoadUI<0b00, 1, 0b11, FPR128, uimm12s16, "ldr",
1292 [(set (f128 FPR128:$Rt),
1293 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)))]>;
1295 // For regular load, we do not have any alignment requirement.
1296 // Thus, it is safe to directly map the vector loads with interesting
1297 // addressing modes.
1298 // FIXME: We could do the same for bitconvert to floating point vectors.
1299 def : Pat <(v8i8 (scalar_to_vector (i32
1300 (extloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))))),
1301 (INSERT_SUBREG (v8i8 (IMPLICIT_DEF)),
1302 (LDRBui GPR64sp:$Rn, uimm12s1:$offset), bsub)>;
1303 def : Pat <(v16i8 (scalar_to_vector (i32
1304 (extloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))))),
1305 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
1306 (LDRBui GPR64sp:$Rn, uimm12s1:$offset), bsub)>;
1307 def : Pat <(v4i16 (scalar_to_vector (i32
1308 (extloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))))),
1309 (INSERT_SUBREG (v4i16 (IMPLICIT_DEF)),
1310 (LDRHui GPR64sp:$Rn, uimm12s2:$offset), hsub)>;
1311 def : Pat <(v8i16 (scalar_to_vector (i32
1312 (extloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))))),
1313 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
1314 (LDRHui GPR64sp:$Rn, uimm12s2:$offset), hsub)>;
1315 def : Pat <(v2i32 (scalar_to_vector (i32
1316 (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))))),
1317 (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)),
1318 (LDRSui GPR64sp:$Rn, uimm12s4:$offset), ssub)>;
1319 def : Pat <(v4i32 (scalar_to_vector (i32
1320 (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))))),
1321 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
1322 (LDRSui GPR64sp:$Rn, uimm12s4:$offset), ssub)>;
1323 def : Pat <(v1i64 (scalar_to_vector (i64
1324 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))))),
1325 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1326 def : Pat <(v2i64 (scalar_to_vector (i64
1327 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))))),
1328 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
1329 (LDRDui GPR64sp:$Rn, uimm12s8:$offset), dsub)>;
1331 // Match all load 64 bits width whose type is compatible with FPR64
1332 let Predicates = [IsLE] in {
1333 // We must use LD1 to perform vector loads in big-endian.
1334 def : Pat<(v2f32 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1335 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1336 def : Pat<(v8i8 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1337 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1338 def : Pat<(v4i16 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1339 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1340 def : Pat<(v2i32 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1341 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1343 def : Pat<(v1f64 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1344 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1345 def : Pat<(v1i64 (load (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))),
1346 (LDRDui GPR64sp:$Rn, uimm12s8:$offset)>;
1348 // Match all load 128 bits width whose type is compatible with FPR128
1349 let Predicates = [IsLE] in {
1350 // We must use LD1 to perform vector loads in big-endian.
1351 def : Pat<(v4f32 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1352 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1353 def : Pat<(v2f64 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1354 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1355 def : Pat<(v16i8 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1356 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1357 def : Pat<(v8i16 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1358 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1359 def : Pat<(v4i32 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1360 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1361 def : Pat<(v2i64 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1362 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1364 def : Pat<(f128 (load (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset))),
1365 (LDRQui GPR64sp:$Rn, uimm12s16:$offset)>;
1367 defm LDRHH : LoadUI<0b01, 0, 0b01, GPR32, uimm12s2, "ldrh",
1369 (zextloadi16 (am_indexed16 GPR64sp:$Rn,
1370 uimm12s2:$offset)))]>;
1371 defm LDRBB : LoadUI<0b00, 0, 0b01, GPR32, uimm12s1, "ldrb",
1373 (zextloadi8 (am_indexed8 GPR64sp:$Rn,
1374 uimm12s1:$offset)))]>;
1376 def : Pat<(i64 (zextloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1377 (SUBREG_TO_REG (i64 0), (LDRBBui GPR64sp:$Rn, uimm12s1:$offset), sub_32)>;
1378 def : Pat<(i64 (zextloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))),
1379 (SUBREG_TO_REG (i64 0), (LDRHHui GPR64sp:$Rn, uimm12s2:$offset), sub_32)>;
1381 // zextloadi1 -> zextloadi8
1382 def : Pat<(i32 (zextloadi1 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1383 (LDRBBui GPR64sp:$Rn, uimm12s1:$offset)>;
1384 def : Pat<(i64 (zextloadi1 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1385 (SUBREG_TO_REG (i64 0), (LDRBBui GPR64sp:$Rn, uimm12s1:$offset), sub_32)>;
1387 // extload -> zextload
1388 def : Pat<(i32 (extloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))),
1389 (LDRHHui GPR64sp:$Rn, uimm12s2:$offset)>;
1390 def : Pat<(i32 (extloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1391 (LDRBBui GPR64sp:$Rn, uimm12s1:$offset)>;
1392 def : Pat<(i32 (extloadi1 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1393 (LDRBBui GPR64sp:$Rn, uimm12s1:$offset)>;
1394 def : Pat<(i64 (extloadi32 (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))),
1395 (SUBREG_TO_REG (i64 0), (LDRWui GPR64sp:$Rn, uimm12s4:$offset), sub_32)>;
1396 def : Pat<(i64 (extloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))),
1397 (SUBREG_TO_REG (i64 0), (LDRHHui GPR64sp:$Rn, uimm12s2:$offset), sub_32)>;
1398 def : Pat<(i64 (extloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1399 (SUBREG_TO_REG (i64 0), (LDRBBui GPR64sp:$Rn, uimm12s1:$offset), sub_32)>;
1400 def : Pat<(i64 (extloadi1 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))),
1401 (SUBREG_TO_REG (i64 0), (LDRBBui GPR64sp:$Rn, uimm12s1:$offset), sub_32)>;
1403 // load sign-extended half-word
1404 defm LDRSHW : LoadUI<0b01, 0, 0b11, GPR32, uimm12s2, "ldrsh",
1406 (sextloadi16 (am_indexed16 GPR64sp:$Rn,
1407 uimm12s2:$offset)))]>;
1408 defm LDRSHX : LoadUI<0b01, 0, 0b10, GPR64, uimm12s2, "ldrsh",
1410 (sextloadi16 (am_indexed16 GPR64sp:$Rn,
1411 uimm12s2:$offset)))]>;
1413 // load sign-extended byte
1414 defm LDRSBW : LoadUI<0b00, 0, 0b11, GPR32, uimm12s1, "ldrsb",
1416 (sextloadi8 (am_indexed8 GPR64sp:$Rn,
1417 uimm12s1:$offset)))]>;
1418 defm LDRSBX : LoadUI<0b00, 0, 0b10, GPR64, uimm12s1, "ldrsb",
1420 (sextloadi8 (am_indexed8 GPR64sp:$Rn,
1421 uimm12s1:$offset)))]>;
1423 // load sign-extended word
1424 defm LDRSW : LoadUI<0b10, 0, 0b10, GPR64, uimm12s4, "ldrsw",
1426 (sextloadi32 (am_indexed32 GPR64sp:$Rn,
1427 uimm12s4:$offset)))]>;
1429 // load zero-extended word
1430 def : Pat<(i64 (zextloadi32 (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))),
1431 (SUBREG_TO_REG (i64 0), (LDRWui GPR64sp:$Rn, uimm12s4:$offset), sub_32)>;
1434 def PRFMui : PrefetchUI<0b11, 0, 0b10, "prfm",
1435 [(ARM64Prefetch imm:$Rt,
1436 (am_indexed64 GPR64sp:$Rn,
1437 uimm12s8:$offset))]>;
1439 def : InstAlias<"prfm $Rt, [$Rn]", (PRFMui prfop:$Rt, GPR64sp:$Rn, 0)>;
1443 def LDRWl : LoadLiteral<0b00, 0, GPR32, "ldr">;
1444 def LDRXl : LoadLiteral<0b01, 0, GPR64, "ldr">;
1445 def LDRSl : LoadLiteral<0b00, 1, FPR32, "ldr">;
1446 def LDRDl : LoadLiteral<0b01, 1, FPR64, "ldr">;
1447 def LDRQl : LoadLiteral<0b10, 1, FPR128, "ldr">;
1449 // load sign-extended word
1450 def LDRSWl : LoadLiteral<0b10, 0, GPR64, "ldrsw">;
1453 def PRFMl : PrefetchLiteral<0b11, 0, "prfm", []>;
1454 // [(ARM64Prefetch imm:$Rt, tglobaladdr:$label)]>;
1457 // (unscaled immediate)
1458 defm LDURX : LoadUnscaled<0b11, 0, 0b01, GPR64, "ldur",
1460 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset)))]>;
1461 defm LDURW : LoadUnscaled<0b10, 0, 0b01, GPR32, "ldur",
1463 (load (am_unscaled32 GPR64sp:$Rn, simm9:$offset)))]>;
1464 defm LDURB : LoadUnscaled<0b00, 1, 0b01, FPR8, "ldur",
1466 (load (am_unscaled8 GPR64sp:$Rn, simm9:$offset)))]>;
1467 defm LDURH : LoadUnscaled<0b01, 1, 0b01, FPR16, "ldur",
1469 (load (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
1470 defm LDURS : LoadUnscaled<0b10, 1, 0b01, FPR32, "ldur",
1471 [(set (f32 FPR32:$Rt),
1472 (load (am_unscaled32 GPR64sp:$Rn, simm9:$offset)))]>;
1473 defm LDURD : LoadUnscaled<0b11, 1, 0b01, FPR64, "ldur",
1474 [(set (f64 FPR64:$Rt),
1475 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset)))]>;
1476 defm LDURQ : LoadUnscaled<0b00, 1, 0b11, FPR128, "ldur",
1477 [(set (f128 FPR128:$Rt),
1478 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset)))]>;
1481 : LoadUnscaled<0b01, 0, 0b01, GPR32, "ldurh",
1483 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
1485 : LoadUnscaled<0b00, 0, 0b01, GPR32, "ldurb",
1487 (zextloadi8 (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
1489 // Match all load 64 bits width whose type is compatible with FPR64
1490 let Predicates = [IsLE] in {
1491 def : Pat<(v2f32 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1492 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1493 def : Pat<(v2i32 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1494 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1495 def : Pat<(v4i16 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1496 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1497 def : Pat<(v8i8 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1498 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1500 def : Pat<(v1f64 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1501 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1502 def : Pat<(v1i64 (load (am_unscaled64 GPR64sp:$Rn, simm9:$offset))),
1503 (LDURDi GPR64sp:$Rn, simm9:$offset)>;
1505 // Match all load 128 bits width whose type is compatible with FPR128
1506 let Predicates = [IsLE] in {
1507 def : Pat<(v2f64 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1508 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1509 def : Pat<(v2i64 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1510 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1511 def : Pat<(v4f32 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1512 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1513 def : Pat<(v4i32 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1514 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1515 def : Pat<(v8i16 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1516 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1517 def : Pat<(v16i8 (load (am_unscaled128 GPR64sp:$Rn, simm9:$offset))),
1518 (LDURQi GPR64sp:$Rn, simm9:$offset)>;
1522 def : Pat<(i32 (extloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
1523 (LDURHHi GPR64sp:$Rn, simm9:$offset)>;
1524 def : Pat<(i32 (extloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1525 (LDURBBi GPR64sp:$Rn, simm9:$offset)>;
1526 def : Pat<(i32 (extloadi1 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1527 (LDURBBi GPR64sp:$Rn, simm9:$offset)>;
1528 def : Pat<(i64 (extloadi32 (am_unscaled32 GPR64sp:$Rn, simm9:$offset))),
1529 (SUBREG_TO_REG (i64 0), (LDURWi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1530 def : Pat<(i64 (extloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
1531 (SUBREG_TO_REG (i64 0), (LDURHHi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1532 def : Pat<(i64 (extloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1533 (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1534 def : Pat<(i64 (extloadi1 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1535 (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1537 def : Pat<(i32 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
1538 (LDURHHi GPR64sp:$Rn, simm9:$offset)>;
1539 def : Pat<(i32 (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1540 (LDURBBi GPR64sp:$Rn, simm9:$offset)>;
1541 def : Pat<(i32 (zextloadi1 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1542 (LDURBBi GPR64sp:$Rn, simm9:$offset)>;
1543 def : Pat<(i64 (zextloadi32 (am_unscaled32 GPR64sp:$Rn, simm9:$offset))),
1544 (SUBREG_TO_REG (i64 0), (LDURWi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1545 def : Pat<(i64 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
1546 (SUBREG_TO_REG (i64 0), (LDURHHi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1547 def : Pat<(i64 (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1548 (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1549 def : Pat<(i64 (zextloadi1 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1550 (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1554 // LDR mnemonics fall back to LDUR for negative or unaligned offsets.
1556 // Define new assembler match classes as we want to only match these when
1557 // the don't otherwise match the scaled addressing mode for LDR/STR. Don't
1558 // associate a DiagnosticType either, as we want the diagnostic for the
1559 // canonical form (the scaled operand) to take precedence.
1560 class SImm9OffsetOperand<int Width> : AsmOperandClass {
1561 let Name = "SImm9OffsetFB" # Width;
1562 let PredicateMethod = "isSImm9OffsetFB<" # Width # ">";
1563 let RenderMethod = "addImmOperands";
1566 def SImm9OffsetFB8Operand : SImm9OffsetOperand<8>;
1567 def SImm9OffsetFB16Operand : SImm9OffsetOperand<16>;
1568 def SImm9OffsetFB32Operand : SImm9OffsetOperand<32>;
1569 def SImm9OffsetFB64Operand : SImm9OffsetOperand<64>;
1570 def SImm9OffsetFB128Operand : SImm9OffsetOperand<128>;
1572 def simm9_offset_fb8 : Operand<i64> {
1573 let ParserMatchClass = SImm9OffsetFB8Operand;
1575 def simm9_offset_fb16 : Operand<i64> {
1576 let ParserMatchClass = SImm9OffsetFB16Operand;
1578 def simm9_offset_fb32 : Operand<i64> {
1579 let ParserMatchClass = SImm9OffsetFB32Operand;
1581 def simm9_offset_fb64 : Operand<i64> {
1582 let ParserMatchClass = SImm9OffsetFB64Operand;
1584 def simm9_offset_fb128 : Operand<i64> {
1585 let ParserMatchClass = SImm9OffsetFB128Operand;
1588 // FIXME: these don't work
1589 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1590 (LDURXi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb64:$offset), 0>;
1591 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1592 (LDURWi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
1593 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1594 (LDURBi FPR8:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
1595 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1596 (LDURHi FPR16:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
1597 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1598 (LDURSi FPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
1599 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1600 (LDURDi FPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb64:$offset), 0>;
1601 def : InstAlias<"ldr $Rt, [$Rn, $offset]",
1602 (LDURQi FPR128:$Rt, GPR64sp:$Rn, simm9_offset_fb128:$offset), 0>;
1605 def : Pat<(i64 (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))),
1606 (SUBREG_TO_REG (i64 0), (LDURBBi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1607 def : Pat<(i64 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))),
1608 (SUBREG_TO_REG (i64 0), (LDURHHi GPR64sp:$Rn, simm9:$offset), sub_32)>;
1610 // load sign-extended half-word
1612 : LoadUnscaled<0b01, 0, 0b11, GPR32, "ldursh",
1614 (sextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
1616 : LoadUnscaled<0b01, 0, 0b10, GPR64, "ldursh",
1618 (sextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset)))]>;
1620 // load sign-extended byte
1622 : LoadUnscaled<0b00, 0, 0b11, GPR32, "ldursb",
1624 (sextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset)))]>;
1626 : LoadUnscaled<0b00, 0, 0b10, GPR64, "ldursb",
1628 (sextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset)))]>;
1630 // load sign-extended word
1632 : LoadUnscaled<0b10, 0, 0b10, GPR64, "ldursw",
1634 (sextloadi32 (am_unscaled32 GPR64sp:$Rn, simm9:$offset)))]>;
1636 // zero and sign extending aliases from generic LDR* mnemonics to LDUR*.
1637 // FIXME: these don't work now
1638 def : InstAlias<"ldrb $Rt, [$Rn, $offset]",
1639 (LDURBBi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
1640 def : InstAlias<"ldrh $Rt, [$Rn, $offset]",
1641 (LDURHHi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
1642 def : InstAlias<"ldrsb $Rt, [$Rn, $offset]",
1643 (LDURSBWi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
1644 def : InstAlias<"ldrsb $Rt, [$Rn, $offset]",
1645 (LDURSBXi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
1646 def : InstAlias<"ldrsh $Rt, [$Rn, $offset]",
1647 (LDURSHWi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
1648 def : InstAlias<"ldrsh $Rt, [$Rn, $offset]",
1649 (LDURSHXi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
1650 def : InstAlias<"ldrsw $Rt, [$Rn, $offset]",
1651 (LDURSWi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
1654 defm PRFUM : PrefetchUnscaled<0b11, 0, 0b10, "prfum",
1655 [(ARM64Prefetch imm:$Rt,
1656 (am_unscaled64 GPR64sp:$Rn, simm9:$offset))]>;
1659 // (unscaled immediate, unprivileged)
1660 defm LDTRX : LoadUnprivileged<0b11, 0, 0b01, GPR64, "ldtr">;
1661 defm LDTRW : LoadUnprivileged<0b10, 0, 0b01, GPR32, "ldtr">;
1663 defm LDTRH : LoadUnprivileged<0b01, 0, 0b01, GPR32, "ldtrh">;
1664 defm LDTRB : LoadUnprivileged<0b00, 0, 0b01, GPR32, "ldtrb">;
1666 // load sign-extended half-word
1667 defm LDTRSHW : LoadUnprivileged<0b01, 0, 0b11, GPR32, "ldtrsh">;
1668 defm LDTRSHX : LoadUnprivileged<0b01, 0, 0b10, GPR64, "ldtrsh">;
1670 // load sign-extended byte
1671 defm LDTRSBW : LoadUnprivileged<0b00, 0, 0b11, GPR32, "ldtrsb">;
1672 defm LDTRSBX : LoadUnprivileged<0b00, 0, 0b10, GPR64, "ldtrsb">;
1674 // load sign-extended word
1675 defm LDTRSW : LoadUnprivileged<0b10, 0, 0b10, GPR64, "ldtrsw">;
1678 // (immediate pre-indexed)
1679 def LDRWpre : LoadPreIdx<0b10, 0, 0b01, GPR32, "ldr">;
1680 def LDRXpre : LoadPreIdx<0b11, 0, 0b01, GPR64, "ldr">;
1681 def LDRBpre : LoadPreIdx<0b00, 1, 0b01, FPR8, "ldr">;
1682 def LDRHpre : LoadPreIdx<0b01, 1, 0b01, FPR16, "ldr">;
1683 def LDRSpre : LoadPreIdx<0b10, 1, 0b01, FPR32, "ldr">;
1684 def LDRDpre : LoadPreIdx<0b11, 1, 0b01, FPR64, "ldr">;
1685 def LDRQpre : LoadPreIdx<0b00, 1, 0b11, FPR128, "ldr">;
1687 // load sign-extended half-word
1688 def LDRSHWpre : LoadPreIdx<0b01, 0, 0b11, GPR32, "ldrsh">;
1689 def LDRSHXpre : LoadPreIdx<0b01, 0, 0b10, GPR64, "ldrsh">;
1691 // load sign-extended byte
1692 def LDRSBWpre : LoadPreIdx<0b00, 0, 0b11, GPR32, "ldrsb">;
1693 def LDRSBXpre : LoadPreIdx<0b00, 0, 0b10, GPR64, "ldrsb">;
1695 // load zero-extended byte
1696 def LDRBBpre : LoadPreIdx<0b00, 0, 0b01, GPR32, "ldrb">;
1697 def LDRHHpre : LoadPreIdx<0b01, 0, 0b01, GPR32, "ldrh">;
1699 // load sign-extended word
1700 def LDRSWpre : LoadPreIdx<0b10, 0, 0b10, GPR64, "ldrsw">;
1702 // ISel pseudos and patterns. See expanded comment on LoadPreIdxPseudo.
1703 def LDRQpre_isel : LoadPreIdxPseudo<FPR128>;
1704 def LDRDpre_isel : LoadPreIdxPseudo<FPR64>;
1705 def LDRSpre_isel : LoadPreIdxPseudo<FPR32>;
1706 def LDRXpre_isel : LoadPreIdxPseudo<GPR64>;
1707 def LDRWpre_isel : LoadPreIdxPseudo<GPR32>;
1708 def LDRHHpre_isel : LoadPreIdxPseudo<GPR32>;
1709 def LDRBBpre_isel : LoadPreIdxPseudo<GPR32>;
1711 def LDRSWpre_isel : LoadPreIdxPseudo<GPR64>;
1712 def LDRSHWpre_isel : LoadPreIdxPseudo<GPR32>;
1713 def LDRSHXpre_isel : LoadPreIdxPseudo<GPR64>;
1714 def LDRSBWpre_isel : LoadPreIdxPseudo<GPR32>;
1715 def LDRSBXpre_isel : LoadPreIdxPseudo<GPR64>;
1718 // (immediate post-indexed)
1719 def LDRWpost : LoadPostIdx<0b10, 0, 0b01, GPR32, "ldr">;
1720 def LDRXpost : LoadPostIdx<0b11, 0, 0b01, GPR64, "ldr">;
1721 def LDRBpost : LoadPostIdx<0b00, 1, 0b01, FPR8, "ldr">;
1722 def LDRHpost : LoadPostIdx<0b01, 1, 0b01, FPR16, "ldr">;
1723 def LDRSpost : LoadPostIdx<0b10, 1, 0b01, FPR32, "ldr">;
1724 def LDRDpost : LoadPostIdx<0b11, 1, 0b01, FPR64, "ldr">;
1725 def LDRQpost : LoadPostIdx<0b00, 1, 0b11, FPR128, "ldr">;
1727 // load sign-extended half-word
1728 def LDRSHWpost : LoadPostIdx<0b01, 0, 0b11, GPR32, "ldrsh">;
1729 def LDRSHXpost : LoadPostIdx<0b01, 0, 0b10, GPR64, "ldrsh">;
1731 // load sign-extended byte
1732 def LDRSBWpost : LoadPostIdx<0b00, 0, 0b11, GPR32, "ldrsb">;
1733 def LDRSBXpost : LoadPostIdx<0b00, 0, 0b10, GPR64, "ldrsb">;
1735 // load zero-extended byte
1736 def LDRBBpost : LoadPostIdx<0b00, 0, 0b01, GPR32, "ldrb">;
1737 def LDRHHpost : LoadPostIdx<0b01, 0, 0b01, GPR32, "ldrh">;
1739 // load sign-extended word
1740 def LDRSWpost : LoadPostIdx<0b10, 0, 0b10, GPR64, "ldrsw">;
1742 // ISel pseudos and patterns. See expanded comment on LoadPostIdxPseudo.
1743 def LDRQpost_isel : LoadPostIdxPseudo<FPR128>;
1744 def LDRDpost_isel : LoadPostIdxPseudo<FPR64>;
1745 def LDRSpost_isel : LoadPostIdxPseudo<FPR32>;
1746 def LDRXpost_isel : LoadPostIdxPseudo<GPR64>;
1747 def LDRWpost_isel : LoadPostIdxPseudo<GPR32>;
1748 def LDRHHpost_isel : LoadPostIdxPseudo<GPR32>;
1749 def LDRBBpost_isel : LoadPostIdxPseudo<GPR32>;
1751 def LDRSWpost_isel : LoadPostIdxPseudo<GPR64>;
1752 def LDRSHWpost_isel : LoadPostIdxPseudo<GPR32>;
1753 def LDRSHXpost_isel : LoadPostIdxPseudo<GPR64>;
1754 def LDRSBWpost_isel : LoadPostIdxPseudo<GPR32>;
1755 def LDRSBXpost_isel : LoadPostIdxPseudo<GPR64>;
1757 //===----------------------------------------------------------------------===//
1758 // Store instructions.
1759 //===----------------------------------------------------------------------===//
1761 // Pair (indexed, offset)
1762 // FIXME: Use dedicated range-checked addressing mode operand here.
1763 defm STPW : StorePairOffset<0b00, 0, GPR32, simm7s4, "stp">;
1764 defm STPX : StorePairOffset<0b10, 0, GPR64, simm7s8, "stp">;
1765 defm STPS : StorePairOffset<0b00, 1, FPR32, simm7s4, "stp">;
1766 defm STPD : StorePairOffset<0b01, 1, FPR64, simm7s8, "stp">;
1767 defm STPQ : StorePairOffset<0b10, 1, FPR128, simm7s16, "stp">;
1769 // Pair (pre-indexed)
1770 def STPWpre : StorePairPreIdx<0b00, 0, GPR32, simm7s4, "stp">;
1771 def STPXpre : StorePairPreIdx<0b10, 0, GPR64, simm7s8, "stp">;
1772 def STPSpre : StorePairPreIdx<0b00, 1, FPR32, simm7s4, "stp">;
1773 def STPDpre : StorePairPreIdx<0b01, 1, FPR64, simm7s8, "stp">;
1774 def STPQpre : StorePairPreIdx<0b10, 1, FPR128, simm7s16, "stp">;
1776 // Pair (pre-indexed)
1777 def STPWpost : StorePairPostIdx<0b00, 0, GPR32, simm7s4, "stp">;
1778 def STPXpost : StorePairPostIdx<0b10, 0, GPR64, simm7s8, "stp">;
1779 def STPSpost : StorePairPostIdx<0b00, 1, FPR32, simm7s4, "stp">;
1780 def STPDpost : StorePairPostIdx<0b01, 1, FPR64, simm7s8, "stp">;
1781 def STPQpost : StorePairPostIdx<0b10, 1, FPR128, simm7s16, "stp">;
1783 // Pair (no allocate)
1784 defm STNPW : StorePairNoAlloc<0b00, 0, GPR32, simm7s4, "stnp">;
1785 defm STNPX : StorePairNoAlloc<0b10, 0, GPR64, simm7s8, "stnp">;
1786 defm STNPS : StorePairNoAlloc<0b00, 1, FPR32, simm7s4, "stnp">;
1787 defm STNPD : StorePairNoAlloc<0b01, 1, FPR64, simm7s8, "stnp">;
1788 defm STNPQ : StorePairNoAlloc<0b10, 1, FPR128, simm7s16, "stnp">;
1791 // (Register offset)
1794 defm STRBB : Store8RO< 0b00, 0, 0b00, GPR32, "strb", i32, truncstorei8>;
1795 defm STRHH : Store16RO<0b01, 0, 0b00, GPR32, "strh", i32, truncstorei16>;
1796 defm STRW : Store32RO<0b10, 0, 0b00, GPR32, "str", i32, store>;
1797 defm STRX : Store64RO<0b11, 0, 0b00, GPR64, "str", i64, store>;
1801 defm STRB : Store8RO< 0b00, 1, 0b00, FPR8, "str", untyped, store>;
1802 defm STRH : Store16RO<0b01, 1, 0b00, FPR16, "str", f16, store>;
1803 defm STRS : Store32RO<0b10, 1, 0b00, FPR32, "str", f32, store>;
1804 defm STRD : Store64RO<0b11, 1, 0b00, FPR64, "str", f64, store>;
1805 defm STRQ : Store128RO<0b00, 1, 0b10, FPR128, "str", f128, store>;
1807 multiclass TruncStoreFrom64ROPat<ROAddrMode ro, SDPatternOperator storeop,
1808 Instruction STRW, Instruction STRX> {
1810 def : Pat<(storeop GPR64:$Rt,
1811 (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)),
1812 (STRW (EXTRACT_SUBREG GPR64:$Rt, sub_32),
1813 GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
1815 def : Pat<(storeop GPR64:$Rt,
1816 (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)),
1817 (STRX (EXTRACT_SUBREG GPR64:$Rt, sub_32),
1818 GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
1821 let AddedComplexity = 10 in {
1823 defm : TruncStoreFrom64ROPat<ro8, truncstorei8, STRBBroW, STRBBroX>;
1824 defm : TruncStoreFrom64ROPat<ro16, truncstorei16, STRHHroW, STRHHroX>;
1825 defm : TruncStoreFrom64ROPat<ro32, truncstorei32, STRWroW, STRWroX>;
1828 multiclass VecROStorePat<ROAddrMode ro, ValueType VecTy, RegisterClass FPR,
1829 Instruction STRW, Instruction STRX> {
1830 def : Pat<(store (VecTy FPR:$Rt),
1831 (ro.Wpat GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)),
1832 (STRW FPR:$Rt, GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend)>;
1834 def : Pat<(store (VecTy FPR:$Rt),
1835 (ro.Xpat GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)),
1836 (STRX FPR:$Rt, GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend)>;
1839 let AddedComplexity = 10 in {
1840 // Match all store 64 bits width whose type is compatible with FPR64
1841 let Predicates = [IsLE] in {
1842 // We must use ST1 to store vectors in big-endian.
1843 defm : VecROStorePat<ro64, v2i32, FPR64, STRDroW, STRDroX>;
1844 defm : VecROStorePat<ro64, v2f32, FPR64, STRDroW, STRDroX>;
1845 defm : VecROStorePat<ro64, v4i16, FPR64, STRDroW, STRDroX>;
1846 defm : VecROStorePat<ro64, v8i8, FPR64, STRDroW, STRDroX>;
1849 defm : VecROStorePat<ro64, v1i64, FPR64, STRDroW, STRDroX>;
1850 defm : VecROStorePat<ro64, v1f64, FPR64, STRDroW, STRDroX>;
1852 // Match all store 128 bits width whose type is compatible with FPR128
1853 let Predicates = [IsLE] in {
1854 // We must use ST1 to store vectors in big-endian.
1855 defm : VecROStorePat<ro128, v2i64, FPR128, STRQroW, STRQroX>;
1856 defm : VecROStorePat<ro128, v2f64, FPR128, STRQroW, STRQroX>;
1857 defm : VecROStorePat<ro128, v4i32, FPR128, STRQroW, STRQroX>;
1858 defm : VecROStorePat<ro128, v4f32, FPR128, STRQroW, STRQroX>;
1859 defm : VecROStorePat<ro128, v8i16, FPR128, STRQroW, STRQroX>;
1860 defm : VecROStorePat<ro128, v16i8, FPR128, STRQroW, STRQroX>;
1862 } // AddedComplexity = 10
1865 // (unsigned immediate)
1866 defm STRX : StoreUI<0b11, 0, 0b00, GPR64, uimm12s8, "str",
1868 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))]>;
1869 defm STRW : StoreUI<0b10, 0, 0b00, GPR32, uimm12s4, "str",
1871 (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))]>;
1872 defm STRB : StoreUI<0b00, 1, 0b00, FPR8, uimm12s1, "str",
1874 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))]>;
1875 defm STRH : StoreUI<0b01, 1, 0b00, FPR16, uimm12s2, "str",
1876 [(store (f16 FPR16:$Rt),
1877 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))]>;
1878 defm STRS : StoreUI<0b10, 1, 0b00, FPR32, uimm12s4, "str",
1879 [(store (f32 FPR32:$Rt),
1880 (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))]>;
1881 defm STRD : StoreUI<0b11, 1, 0b00, FPR64, uimm12s8, "str",
1882 [(store (f64 FPR64:$Rt),
1883 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset))]>;
1884 defm STRQ : StoreUI<0b00, 1, 0b10, FPR128, uimm12s16, "str", []>;
1886 defm STRHH : StoreUI<0b01, 0, 0b00, GPR32, uimm12s2, "strh",
1887 [(truncstorei16 GPR32:$Rt,
1888 (am_indexed16 GPR64sp:$Rn,
1889 uimm12s2:$offset))]>;
1890 defm STRBB : StoreUI<0b00, 0, 0b00, GPR32, uimm12s1, "strb",
1891 [(truncstorei8 GPR32:$Rt,
1892 (am_indexed8 GPR64sp:$Rn,
1893 uimm12s1:$offset))]>;
1895 // Match all store 64 bits width whose type is compatible with FPR64
1896 let AddedComplexity = 10 in {
1897 let Predicates = [IsLE] in {
1898 // We must use ST1 to store vectors in big-endian.
1899 def : Pat<(store (v2f32 FPR64:$Rt),
1900 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
1901 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
1902 def : Pat<(store (v8i8 FPR64:$Rt),
1903 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
1904 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
1905 def : Pat<(store (v4i16 FPR64:$Rt),
1906 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
1907 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
1908 def : Pat<(store (v2i32 FPR64:$Rt),
1909 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
1910 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
1912 def : Pat<(store (v1f64 FPR64:$Rt),
1913 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
1914 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
1915 def : Pat<(store (v1i64 FPR64:$Rt),
1916 (am_indexed64 GPR64sp:$Rn, uimm12s8:$offset)),
1917 (STRDui FPR64:$Rt, GPR64sp:$Rn, uimm12s8:$offset)>;
1919 // Match all store 128 bits width whose type is compatible with FPR128
1920 let Predicates = [IsLE] in {
1921 // We must use ST1 to store vectors in big-endian.
1922 def : Pat<(store (v4f32 FPR128:$Rt),
1923 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
1924 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
1925 def : Pat<(store (v2f64 FPR128:$Rt),
1926 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
1927 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
1928 def : Pat<(store (v16i8 FPR128:$Rt),
1929 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
1930 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
1931 def : Pat<(store (v8i16 FPR128:$Rt),
1932 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
1933 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
1934 def : Pat<(store (v4i32 FPR128:$Rt),
1935 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
1936 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
1937 def : Pat<(store (v2i64 FPR128:$Rt),
1938 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
1939 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
1941 def : Pat<(store (f128 FPR128:$Rt),
1942 (am_indexed128 GPR64sp:$Rn, uimm12s16:$offset)),
1943 (STRQui FPR128:$Rt, GPR64sp:$Rn, uimm12s16:$offset)>;
1946 def : Pat<(truncstorei32 GPR64:$Rt,
1947 (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset)),
1948 (STRWui (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, uimm12s4:$offset)>;
1949 def : Pat<(truncstorei16 GPR64:$Rt,
1950 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset)),
1951 (STRHHui (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, uimm12s2:$offset)>;
1952 def : Pat<(truncstorei8 GPR64:$Rt, (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset)),
1953 (STRBBui (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, uimm12s1:$offset)>;
1955 } // AddedComplexity = 10
1958 // (unscaled immediate)
1959 defm STURX : StoreUnscaled<0b11, 0, 0b00, GPR64, "stur",
1961 (am_unscaled64 GPR64sp:$Rn, simm9:$offset))]>;
1962 defm STURW : StoreUnscaled<0b10, 0, 0b00, GPR32, "stur",
1964 (am_unscaled32 GPR64sp:$Rn, simm9:$offset))]>;
1965 defm STURB : StoreUnscaled<0b00, 1, 0b00, FPR8, "stur",
1967 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))]>;
1968 defm STURH : StoreUnscaled<0b01, 1, 0b00, FPR16, "stur",
1969 [(store (f16 FPR16:$Rt),
1970 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))]>;
1971 defm STURS : StoreUnscaled<0b10, 1, 0b00, FPR32, "stur",
1972 [(store (f32 FPR32:$Rt),
1973 (am_unscaled32 GPR64sp:$Rn, simm9:$offset))]>;
1974 defm STURD : StoreUnscaled<0b11, 1, 0b00, FPR64, "stur",
1975 [(store (f64 FPR64:$Rt),
1976 (am_unscaled64 GPR64sp:$Rn, simm9:$offset))]>;
1977 defm STURQ : StoreUnscaled<0b00, 1, 0b10, FPR128, "stur",
1978 [(store (f128 FPR128:$Rt),
1979 (am_unscaled128 GPR64sp:$Rn, simm9:$offset))]>;
1980 defm STURHH : StoreUnscaled<0b01, 0, 0b00, GPR32, "sturh",
1981 [(truncstorei16 GPR32:$Rt,
1982 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))]>;
1983 defm STURBB : StoreUnscaled<0b00, 0, 0b00, GPR32, "sturb",
1984 [(truncstorei8 GPR32:$Rt,
1985 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))]>;
1987 // Match all store 64 bits width whose type is compatible with FPR64
1988 let Predicates = [IsLE] in {
1989 // We must use ST1 to store vectors in big-endian.
1990 def : Pat<(store (v2f32 FPR64:$Rt),
1991 (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
1992 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
1993 def : Pat<(store (v8i8 FPR64:$Rt),
1994 (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
1995 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
1996 def : Pat<(store (v4i16 FPR64:$Rt),
1997 (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
1998 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
1999 def : Pat<(store (v2i32 FPR64:$Rt),
2000 (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2001 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2003 def : Pat<(store (v1f64 FPR64:$Rt), (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2004 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2005 def : Pat<(store (v1i64 FPR64:$Rt), (am_unscaled64 GPR64sp:$Rn, simm9:$offset)),
2006 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2008 // Match all store 128 bits width whose type is compatible with FPR128
2009 let Predicates = [IsLE] in {
2010 // We must use ST1 to store vectors in big-endian.
2011 def : Pat<(store (v4f32 FPR128:$Rt),
2012 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2013 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2014 def : Pat<(store (v2f64 FPR128:$Rt),
2015 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2016 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2017 def : Pat<(store (v16i8 FPR128:$Rt),
2018 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2019 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2020 def : Pat<(store (v8i16 FPR128:$Rt),
2021 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2022 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2023 def : Pat<(store (v4i32 FPR128:$Rt),
2024 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2025 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2026 def : Pat<(store (v2i64 FPR128:$Rt),
2027 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2028 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2029 def : Pat<(store (v2f64 FPR128:$Rt),
2030 (am_unscaled128 GPR64sp:$Rn, simm9:$offset)),
2031 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9:$offset)>;
2034 // unscaled i64 truncating stores
2035 def : Pat<(truncstorei32 GPR64:$Rt, (am_unscaled32 GPR64sp:$Rn, simm9:$offset)),
2036 (STURWi (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, simm9:$offset)>;
2037 def : Pat<(truncstorei16 GPR64:$Rt, (am_unscaled16 GPR64sp:$Rn, simm9:$offset)),
2038 (STURHHi (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, simm9:$offset)>;
2039 def : Pat<(truncstorei8 GPR64:$Rt, (am_unscaled8 GPR64sp:$Rn, simm9:$offset)),
2040 (STURBBi (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$Rn, simm9:$offset)>;
2043 // STR mnemonics fall back to STUR for negative or unaligned offsets.
2044 // FIXME: these don't work now.
2045 def : InstAlias<"str $Rt, [$Rn, $offset]",
2046 (STURXi GPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb64:$offset), 0>;
2047 def : InstAlias<"str $Rt, [$Rn, $offset]",
2048 (STURWi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
2049 def : InstAlias<"str $Rt, [$Rn, $offset]",
2050 (STURBi FPR8:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
2051 def : InstAlias<"str $Rt, [$Rn, $offset]",
2052 (STURHi FPR16:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
2053 def : InstAlias<"str $Rt, [$Rn, $offset]",
2054 (STURSi FPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb32:$offset), 0>;
2055 def : InstAlias<"str $Rt, [$Rn, $offset]",
2056 (STURDi FPR64:$Rt, GPR64sp:$Rn, simm9_offset_fb64:$offset), 0>;
2057 def : InstAlias<"str $Rt, [$Rn, $offset]",
2058 (STURQi FPR128:$Rt, GPR64sp:$Rn, simm9_offset_fb128:$offset), 0>;
2060 def : InstAlias<"strb $Rt, [$Rn, $offset]",
2061 (STURBBi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb8:$offset), 0>;
2062 def : InstAlias<"strh $Rt, [$Rn, $offset]",
2063 (STURHHi GPR32:$Rt, GPR64sp:$Rn, simm9_offset_fb16:$offset), 0>;
2066 // (unscaled immediate, unprivileged)
2067 defm STTRW : StoreUnprivileged<0b10, 0, 0b00, GPR32, "sttr">;
2068 defm STTRX : StoreUnprivileged<0b11, 0, 0b00, GPR64, "sttr">;
2070 defm STTRH : StoreUnprivileged<0b01, 0, 0b00, GPR32, "sttrh">;
2071 defm STTRB : StoreUnprivileged<0b00, 0, 0b00, GPR32, "sttrb">;
2074 // (immediate pre-indexed)
2075 def STRWpre : StorePreIdx<0b10, 0, 0b00, GPR32, "str">;
2076 def STRXpre : StorePreIdx<0b11, 0, 0b00, GPR64, "str">;
2077 def STRBpre : StorePreIdx<0b00, 1, 0b00, FPR8, "str">;
2078 def STRHpre : StorePreIdx<0b01, 1, 0b00, FPR16, "str">;
2079 def STRSpre : StorePreIdx<0b10, 1, 0b00, FPR32, "str">;
2080 def STRDpre : StorePreIdx<0b11, 1, 0b00, FPR64, "str">;
2081 def STRQpre : StorePreIdx<0b00, 1, 0b10, FPR128, "str">;
2083 def STRBBpre : StorePreIdx<0b00, 0, 0b00, GPR32, "strb">;
2084 def STRHHpre : StorePreIdx<0b01, 0, 0b00, GPR32, "strh">;
2086 // ISel pseudos and patterns. See expanded comment on StorePreIdxPseudo.
2087 defm STRQpre : StorePreIdxPseudo<FPR128, f128, pre_store>;
2088 defm STRDpre : StorePreIdxPseudo<FPR64, f64, pre_store>;
2089 defm STRSpre : StorePreIdxPseudo<FPR32, f32, pre_store>;
2090 defm STRXpre : StorePreIdxPseudo<GPR64, i64, pre_store>;
2091 defm STRWpre : StorePreIdxPseudo<GPR32, i32, pre_store>;
2092 defm STRHHpre : StorePreIdxPseudo<GPR32, i32, pre_truncsti16>;
2093 defm STRBBpre : StorePreIdxPseudo<GPR32, i32, pre_truncsti8>;
2095 def : Pat<(pre_truncsti32 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
2096 (STRWpre_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
2098 def : Pat<(pre_truncsti16 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
2099 (STRHHpre_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
2101 def : Pat<(pre_truncsti8 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
2102 (STRBBpre_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
2105 def : Pat<(pre_store (v8i8 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2106 (STRDpre_isel FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2107 def : Pat<(pre_store (v4i16 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2108 (STRDpre_isel FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2109 def : Pat<(pre_store (v2i32 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2110 (STRDpre_isel FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2111 def : Pat<(pre_store (v2f32 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2112 (STRDpre_isel FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2113 def : Pat<(pre_store (v1i64 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2114 (STRDpre_isel FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2115 def : Pat<(pre_store (v1f64 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2116 (STRDpre_isel FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2118 def : Pat<(pre_store (v16i8 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2119 (STRQpre_isel FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2120 def : Pat<(pre_store (v8i16 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2121 (STRQpre_isel FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2122 def : Pat<(pre_store (v4i32 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2123 (STRQpre_isel FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2124 def : Pat<(pre_store (v4f32 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2125 (STRQpre_isel FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2126 def : Pat<(pre_store (v2i64 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2127 (STRQpre_isel FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2128 def : Pat<(pre_store (v2f64 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2129 (STRQpre_isel FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2132 // (immediate post-indexed)
2133 def STRWpost : StorePostIdx<0b10, 0, 0b00, GPR32, "str">;
2134 def STRXpost : StorePostIdx<0b11, 0, 0b00, GPR64, "str">;
2135 def STRBpost : StorePostIdx<0b00, 1, 0b00, FPR8, "str">;
2136 def STRHpost : StorePostIdx<0b01, 1, 0b00, FPR16, "str">;
2137 def STRSpost : StorePostIdx<0b10, 1, 0b00, FPR32, "str">;
2138 def STRDpost : StorePostIdx<0b11, 1, 0b00, FPR64, "str">;
2139 def STRQpost : StorePostIdx<0b00, 1, 0b10, FPR128, "str">;
2141 def STRBBpost : StorePostIdx<0b00, 0, 0b00, GPR32, "strb">;
2142 def STRHHpost : StorePostIdx<0b01, 0, 0b00, GPR32, "strh">;
2144 // ISel pseudos and patterns. See expanded comment on StorePostIdxPseudo.
2145 defm STRQpost : StorePostIdxPseudo<FPR128, f128, post_store, STRQpost>;
2146 defm STRDpost : StorePostIdxPseudo<FPR64, f64, post_store, STRDpost>;
2147 defm STRSpost : StorePostIdxPseudo<FPR32, f32, post_store, STRSpost>;
2148 defm STRXpost : StorePostIdxPseudo<GPR64, i64, post_store, STRXpost>;
2149 defm STRWpost : StorePostIdxPseudo<GPR32, i32, post_store, STRWpost>;
2150 defm STRHHpost : StorePostIdxPseudo<GPR32, i32, post_truncsti16, STRHHpost>;
2151 defm STRBBpost : StorePostIdxPseudo<GPR32, i32, post_truncsti8, STRBBpost>;
2153 def : Pat<(post_truncsti32 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
2154 (STRWpost_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
2156 def : Pat<(post_truncsti16 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
2157 (STRHHpost_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
2159 def : Pat<(post_truncsti8 GPR64:$Rt, GPR64sp:$addr, simm9:$off),
2160 (STRBBpost_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), GPR64sp:$addr,
2163 def : Pat<(post_store (v8i8 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2164 (STRDpost_isel FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2165 def : Pat<(post_store (v4i16 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2166 (STRDpost_isel FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2167 def : Pat<(post_store (v2i32 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2168 (STRDpost_isel FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2169 def : Pat<(post_store (v2f32 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2170 (STRDpost_isel FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2171 def : Pat<(post_store (v1i64 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2172 (STRDpost_isel FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2173 def : Pat<(post_store (v1f64 FPR64:$Rt), GPR64sp:$addr, simm9:$off),
2174 (STRDpost_isel FPR64:$Rt, GPR64sp:$addr, simm9:$off)>;
2176 def : Pat<(post_store (v16i8 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2177 (STRQpost_isel FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2178 def : Pat<(post_store (v8i16 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2179 (STRQpost_isel FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2180 def : Pat<(post_store (v4i32 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2181 (STRQpost_isel FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2182 def : Pat<(post_store (v4f32 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2183 (STRQpost_isel FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2184 def : Pat<(post_store (v2i64 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2185 (STRQpost_isel FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2186 def : Pat<(post_store (v2f64 FPR128:$Rt), GPR64sp:$addr, simm9:$off),
2187 (STRQpost_isel FPR128:$Rt, GPR64sp:$addr, simm9:$off)>;
2189 //===----------------------------------------------------------------------===//
2190 // Load/store exclusive instructions.
2191 //===----------------------------------------------------------------------===//
2193 def LDARW : LoadAcquire <0b10, 1, 1, 0, 1, GPR32, "ldar">;
2194 def LDARX : LoadAcquire <0b11, 1, 1, 0, 1, GPR64, "ldar">;
2195 def LDARB : LoadAcquire <0b00, 1, 1, 0, 1, GPR32, "ldarb">;
2196 def LDARH : LoadAcquire <0b01, 1, 1, 0, 1, GPR32, "ldarh">;
2198 def LDAXRW : LoadExclusive <0b10, 0, 1, 0, 1, GPR32, "ldaxr">;
2199 def LDAXRX : LoadExclusive <0b11, 0, 1, 0, 1, GPR64, "ldaxr">;
2200 def LDAXRB : LoadExclusive <0b00, 0, 1, 0, 1, GPR32, "ldaxrb">;
2201 def LDAXRH : LoadExclusive <0b01, 0, 1, 0, 1, GPR32, "ldaxrh">;
2203 def LDXRW : LoadExclusive <0b10, 0, 1, 0, 0, GPR32, "ldxr">;
2204 def LDXRX : LoadExclusive <0b11, 0, 1, 0, 0, GPR64, "ldxr">;
2205 def LDXRB : LoadExclusive <0b00, 0, 1, 0, 0, GPR32, "ldxrb">;
2206 def LDXRH : LoadExclusive <0b01, 0, 1, 0, 0, GPR32, "ldxrh">;
2208 def STLRW : StoreRelease <0b10, 1, 0, 0, 1, GPR32, "stlr">;
2209 def STLRX : StoreRelease <0b11, 1, 0, 0, 1, GPR64, "stlr">;
2210 def STLRB : StoreRelease <0b00, 1, 0, 0, 1, GPR32, "stlrb">;
2211 def STLRH : StoreRelease <0b01, 1, 0, 0, 1, GPR32, "stlrh">;
2213 def STLXRW : StoreExclusive<0b10, 0, 0, 0, 1, GPR32, "stlxr">;
2214 def STLXRX : StoreExclusive<0b11, 0, 0, 0, 1, GPR64, "stlxr">;
2215 def STLXRB : StoreExclusive<0b00, 0, 0, 0, 1, GPR32, "stlxrb">;
2216 def STLXRH : StoreExclusive<0b01, 0, 0, 0, 1, GPR32, "stlxrh">;
2218 def STXRW : StoreExclusive<0b10, 0, 0, 0, 0, GPR32, "stxr">;
2219 def STXRX : StoreExclusive<0b11, 0, 0, 0, 0, GPR64, "stxr">;
2220 def STXRB : StoreExclusive<0b00, 0, 0, 0, 0, GPR32, "stxrb">;
2221 def STXRH : StoreExclusive<0b01, 0, 0, 0, 0, GPR32, "stxrh">;
2223 def LDAXPW : LoadExclusivePair<0b10, 0, 1, 1, 1, GPR32, "ldaxp">;
2224 def LDAXPX : LoadExclusivePair<0b11, 0, 1, 1, 1, GPR64, "ldaxp">;
2226 def LDXPW : LoadExclusivePair<0b10, 0, 1, 1, 0, GPR32, "ldxp">;
2227 def LDXPX : LoadExclusivePair<0b11, 0, 1, 1, 0, GPR64, "ldxp">;
2229 def STLXPW : StoreExclusivePair<0b10, 0, 0, 1, 1, GPR32, "stlxp">;
2230 def STLXPX : StoreExclusivePair<0b11, 0, 0, 1, 1, GPR64, "stlxp">;
2232 def STXPW : StoreExclusivePair<0b10, 0, 0, 1, 0, GPR32, "stxp">;
2233 def STXPX : StoreExclusivePair<0b11, 0, 0, 1, 0, GPR64, "stxp">;
2235 //===----------------------------------------------------------------------===//
2236 // Scaled floating point to integer conversion instructions.
2237 //===----------------------------------------------------------------------===//
2239 defm FCVTAS : FPToIntegerUnscaled<0b00, 0b100, "fcvtas", int_arm64_neon_fcvtas>;
2240 defm FCVTAU : FPToIntegerUnscaled<0b00, 0b101, "fcvtau", int_arm64_neon_fcvtau>;
2241 defm FCVTMS : FPToIntegerUnscaled<0b10, 0b000, "fcvtms", int_arm64_neon_fcvtms>;
2242 defm FCVTMU : FPToIntegerUnscaled<0b10, 0b001, "fcvtmu", int_arm64_neon_fcvtmu>;
2243 defm FCVTNS : FPToIntegerUnscaled<0b00, 0b000, "fcvtns", int_arm64_neon_fcvtns>;
2244 defm FCVTNU : FPToIntegerUnscaled<0b00, 0b001, "fcvtnu", int_arm64_neon_fcvtnu>;
2245 defm FCVTPS : FPToIntegerUnscaled<0b01, 0b000, "fcvtps", int_arm64_neon_fcvtps>;
2246 defm FCVTPU : FPToIntegerUnscaled<0b01, 0b001, "fcvtpu", int_arm64_neon_fcvtpu>;
2247 defm FCVTZS : FPToIntegerUnscaled<0b11, 0b000, "fcvtzs", fp_to_sint>;
2248 defm FCVTZU : FPToIntegerUnscaled<0b11, 0b001, "fcvtzu", fp_to_uint>;
2249 defm FCVTZS : FPToIntegerScaled<0b11, 0b000, "fcvtzs", fp_to_sint>;
2250 defm FCVTZU : FPToIntegerScaled<0b11, 0b001, "fcvtzu", fp_to_uint>;
2251 let isCodeGenOnly = 1 in {
2252 defm FCVTZS_Int : FPToIntegerUnscaled<0b11, 0b000, "fcvtzs", int_arm64_neon_fcvtzs>;
2253 defm FCVTZU_Int : FPToIntegerUnscaled<0b11, 0b001, "fcvtzu", int_arm64_neon_fcvtzu>;
2254 defm FCVTZS_Int : FPToIntegerScaled<0b11, 0b000, "fcvtzs", int_arm64_neon_fcvtzs>;
2255 defm FCVTZU_Int : FPToIntegerScaled<0b11, 0b001, "fcvtzu", int_arm64_neon_fcvtzu>;
2258 //===----------------------------------------------------------------------===//
2259 // Scaled integer to floating point conversion instructions.
2260 //===----------------------------------------------------------------------===//
2262 defm SCVTF : IntegerToFP<0, "scvtf", sint_to_fp>;
2263 defm UCVTF : IntegerToFP<1, "ucvtf", uint_to_fp>;
2265 //===----------------------------------------------------------------------===//
2266 // Unscaled integer to floating point conversion instruction.
2267 //===----------------------------------------------------------------------===//
2269 defm FMOV : UnscaledConversion<"fmov">;
2271 def : Pat<(f32 (fpimm0)), (FMOVWSr WZR)>, Requires<[NoZCZ]>;
2272 def : Pat<(f64 (fpimm0)), (FMOVXDr XZR)>, Requires<[NoZCZ]>;
2274 //===----------------------------------------------------------------------===//
2275 // Floating point conversion instruction.
2276 //===----------------------------------------------------------------------===//
2278 defm FCVT : FPConversion<"fcvt">;
2280 def : Pat<(f32_to_f16 FPR32:$Rn),
2281 (i32 (COPY_TO_REGCLASS
2282 (f32 (SUBREG_TO_REG (i32 0), (FCVTHSr FPR32:$Rn), hsub)),
2285 def FCVTSHpseudo : Pseudo<(outs FPR32:$Rd), (ins FPR32:$Rn),
2286 [(set (f32 FPR32:$Rd), (f16_to_f32 i32:$Rn))]>;
2288 //===----------------------------------------------------------------------===//
2289 // Floating point single operand instructions.
2290 //===----------------------------------------------------------------------===//
2292 defm FABS : SingleOperandFPData<0b0001, "fabs", fabs>;
2293 defm FMOV : SingleOperandFPData<0b0000, "fmov">;
2294 defm FNEG : SingleOperandFPData<0b0010, "fneg", fneg>;
2295 defm FRINTA : SingleOperandFPData<0b1100, "frinta", frnd>;
2296 defm FRINTI : SingleOperandFPData<0b1111, "frinti", fnearbyint>;
2297 defm FRINTM : SingleOperandFPData<0b1010, "frintm", ffloor>;
2298 defm FRINTN : SingleOperandFPData<0b1000, "frintn", int_arm64_neon_frintn>;
2299 defm FRINTP : SingleOperandFPData<0b1001, "frintp", fceil>;
2301 def : Pat<(v1f64 (int_arm64_neon_frintn (v1f64 FPR64:$Rn))),
2302 (FRINTNDr FPR64:$Rn)>;
2304 // FRINTX is inserted to set the flags as required by FENV_ACCESS ON behavior
2305 // in the C spec. Setting hasSideEffects ensures it is not DCE'd.
2306 // <rdar://problem/13715968>
2307 // TODO: We should really model the FPSR flags correctly. This is really ugly.
2308 let hasSideEffects = 1 in {
2309 defm FRINTX : SingleOperandFPData<0b1110, "frintx", frint>;
2312 defm FRINTZ : SingleOperandFPData<0b1011, "frintz", ftrunc>;
2314 let SchedRW = [WriteFDiv] in {
2315 defm FSQRT : SingleOperandFPData<0b0011, "fsqrt", fsqrt>;
2318 //===----------------------------------------------------------------------===//
2319 // Floating point two operand instructions.
2320 //===----------------------------------------------------------------------===//
2322 defm FADD : TwoOperandFPData<0b0010, "fadd", fadd>;
2323 let SchedRW = [WriteFDiv] in {
2324 defm FDIV : TwoOperandFPData<0b0001, "fdiv", fdiv>;
2326 defm FMAXNM : TwoOperandFPData<0b0110, "fmaxnm", int_arm64_neon_fmaxnm>;
2327 defm FMAX : TwoOperandFPData<0b0100, "fmax", ARM64fmax>;
2328 defm FMINNM : TwoOperandFPData<0b0111, "fminnm", int_arm64_neon_fminnm>;
2329 defm FMIN : TwoOperandFPData<0b0101, "fmin", ARM64fmin>;
2330 let SchedRW = [WriteFMul] in {
2331 defm FMUL : TwoOperandFPData<0b0000, "fmul", fmul>;
2332 defm FNMUL : TwoOperandFPDataNeg<0b1000, "fnmul", fmul>;
2334 defm FSUB : TwoOperandFPData<0b0011, "fsub", fsub>;
2336 def : Pat<(v1f64 (ARM64fmax (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2337 (FMAXDrr FPR64:$Rn, FPR64:$Rm)>;
2338 def : Pat<(v1f64 (ARM64fmin (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2339 (FMINDrr FPR64:$Rn, FPR64:$Rm)>;
2340 def : Pat<(v1f64 (int_arm64_neon_fmaxnm (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2341 (FMAXNMDrr FPR64:$Rn, FPR64:$Rm)>;
2342 def : Pat<(v1f64 (int_arm64_neon_fminnm (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2343 (FMINNMDrr FPR64:$Rn, FPR64:$Rm)>;
2345 //===----------------------------------------------------------------------===//
2346 // Floating point three operand instructions.
2347 //===----------------------------------------------------------------------===//
2349 defm FMADD : ThreeOperandFPData<0, 0, "fmadd", fma>;
2350 defm FMSUB : ThreeOperandFPData<0, 1, "fmsub",
2351 TriOpFrag<(fma node:$LHS, (fneg node:$MHS), node:$RHS)> >;
2352 defm FNMADD : ThreeOperandFPData<1, 0, "fnmadd",
2353 TriOpFrag<(fneg (fma node:$LHS, node:$MHS, node:$RHS))> >;
2354 defm FNMSUB : ThreeOperandFPData<1, 1, "fnmsub",
2355 TriOpFrag<(fma node:$LHS, node:$MHS, (fneg node:$RHS))> >;
2357 // The following def pats catch the case where the LHS of an FMA is negated.
2358 // The TriOpFrag above catches the case where the middle operand is negated.
2360 // N.b. FMSUB etc have the accumulator at the *end* of (outs), unlike
2361 // the NEON variant.
2362 def : Pat<(f32 (fma (fneg FPR32:$Rn), FPR32:$Rm, FPR32:$Ra)),
2363 (FMSUBSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
2365 def : Pat<(f64 (fma (fneg FPR64:$Rn), FPR64:$Rm, FPR64:$Ra)),
2366 (FMSUBDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
2368 // We handled -(a + b*c) for FNMADD above, now it's time for "(-a) + (-b)*c" and
2370 def : Pat<(f32 (fma (fneg FPR32:$Rn), FPR32:$Rm, (fneg FPR32:$Ra))),
2371 (FNMADDSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
2373 def : Pat<(f64 (fma (fneg FPR64:$Rn), FPR64:$Rm, (fneg FPR64:$Ra))),
2374 (FNMADDDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
2376 def : Pat<(f32 (fma FPR32:$Rn, (fneg FPR32:$Rm), (fneg FPR32:$Ra))),
2377 (FNMADDSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
2379 def : Pat<(f64 (fma FPR64:$Rn, (fneg FPR64:$Rm), (fneg FPR64:$Ra))),
2380 (FNMADDDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
2382 //===----------------------------------------------------------------------===//
2383 // Floating point comparison instructions.
2384 //===----------------------------------------------------------------------===//
2386 defm FCMPE : FPComparison<1, "fcmpe">;
2387 defm FCMP : FPComparison<0, "fcmp", ARM64fcmp>;
2389 //===----------------------------------------------------------------------===//
2390 // Floating point conditional comparison instructions.
2391 //===----------------------------------------------------------------------===//
2393 defm FCCMPE : FPCondComparison<1, "fccmpe">;
2394 defm FCCMP : FPCondComparison<0, "fccmp">;
2396 //===----------------------------------------------------------------------===//
2397 // Floating point conditional select instruction.
2398 //===----------------------------------------------------------------------===//
2400 defm FCSEL : FPCondSelect<"fcsel">;
2402 // CSEL instructions providing f128 types need to be handled by a
2403 // pseudo-instruction since the eventual code will need to introduce basic
2404 // blocks and control flow.
2405 def F128CSEL : Pseudo<(outs FPR128:$Rd),
2406 (ins FPR128:$Rn, FPR128:$Rm, ccode:$cond),
2407 [(set (f128 FPR128:$Rd),
2408 (ARM64csel FPR128:$Rn, FPR128:$Rm,
2409 (i32 imm:$cond), NZCV))]> {
2411 let usesCustomInserter = 1;
2415 //===----------------------------------------------------------------------===//
2416 // Floating point immediate move.
2417 //===----------------------------------------------------------------------===//
2419 let isReMaterializable = 1 in {
2420 defm FMOV : FPMoveImmediate<"fmov">;
2423 //===----------------------------------------------------------------------===//
2424 // Advanced SIMD two vector instructions.
2425 //===----------------------------------------------------------------------===//
2427 defm ABS : SIMDTwoVectorBHSD<0, 0b01011, "abs", int_arm64_neon_abs>;
2428 defm CLS : SIMDTwoVectorBHS<0, 0b00100, "cls", int_arm64_neon_cls>;
2429 defm CLZ : SIMDTwoVectorBHS<1, 0b00100, "clz", ctlz>;
2430 defm CMEQ : SIMDCmpTwoVector<0, 0b01001, "cmeq", ARM64cmeqz>;
2431 defm CMGE : SIMDCmpTwoVector<1, 0b01000, "cmge", ARM64cmgez>;
2432 defm CMGT : SIMDCmpTwoVector<0, 0b01000, "cmgt", ARM64cmgtz>;
2433 defm CMLE : SIMDCmpTwoVector<1, 0b01001, "cmle", ARM64cmlez>;
2434 defm CMLT : SIMDCmpTwoVector<0, 0b01010, "cmlt", ARM64cmltz>;
2435 defm CNT : SIMDTwoVectorB<0, 0b00, 0b00101, "cnt", ctpop>;
2436 defm FABS : SIMDTwoVectorFP<0, 1, 0b01111, "fabs", fabs>;
2438 defm FCMEQ : SIMDFPCmpTwoVector<0, 1, 0b01101, "fcmeq", ARM64fcmeqz>;
2439 defm FCMGE : SIMDFPCmpTwoVector<1, 1, 0b01100, "fcmge", ARM64fcmgez>;
2440 defm FCMGT : SIMDFPCmpTwoVector<0, 1, 0b01100, "fcmgt", ARM64fcmgtz>;
2441 defm FCMLE : SIMDFPCmpTwoVector<1, 1, 0b01101, "fcmle", ARM64fcmlez>;
2442 defm FCMLT : SIMDFPCmpTwoVector<0, 1, 0b01110, "fcmlt", ARM64fcmltz>;
2443 defm FCVTAS : SIMDTwoVectorFPToInt<0,0,0b11100, "fcvtas",int_arm64_neon_fcvtas>;
2444 defm FCVTAU : SIMDTwoVectorFPToInt<1,0,0b11100, "fcvtau",int_arm64_neon_fcvtau>;
2445 defm FCVTL : SIMDFPWidenTwoVector<0, 0, 0b10111, "fcvtl">;
2446 def : Pat<(v4f32 (int_arm64_neon_vcvthf2fp (v4i16 V64:$Rn))),
2447 (FCVTLv4i16 V64:$Rn)>;
2448 def : Pat<(v4f32 (int_arm64_neon_vcvthf2fp (extract_subvector (v8i16 V128:$Rn),
2450 (FCVTLv8i16 V128:$Rn)>;
2451 def : Pat<(v2f64 (fextend (v2f32 V64:$Rn))), (FCVTLv2i32 V64:$Rn)>;
2452 def : Pat<(v2f64 (fextend (v2f32 (extract_subvector (v4f32 V128:$Rn),
2454 (FCVTLv4i32 V128:$Rn)>;
2456 defm FCVTMS : SIMDTwoVectorFPToInt<0,0,0b11011, "fcvtms",int_arm64_neon_fcvtms>;
2457 defm FCVTMU : SIMDTwoVectorFPToInt<1,0,0b11011, "fcvtmu",int_arm64_neon_fcvtmu>;
2458 defm FCVTNS : SIMDTwoVectorFPToInt<0,0,0b11010, "fcvtns",int_arm64_neon_fcvtns>;
2459 defm FCVTNU : SIMDTwoVectorFPToInt<1,0,0b11010, "fcvtnu",int_arm64_neon_fcvtnu>;
2460 defm FCVTN : SIMDFPNarrowTwoVector<0, 0, 0b10110, "fcvtn">;
2461 def : Pat<(v4i16 (int_arm64_neon_vcvtfp2hf (v4f32 V128:$Rn))),
2462 (FCVTNv4i16 V128:$Rn)>;
2463 def : Pat<(concat_vectors V64:$Rd,
2464 (v4i16 (int_arm64_neon_vcvtfp2hf (v4f32 V128:$Rn)))),
2465 (FCVTNv8i16 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
2466 def : Pat<(v2f32 (fround (v2f64 V128:$Rn))), (FCVTNv2i32 V128:$Rn)>;
2467 def : Pat<(concat_vectors V64:$Rd, (v2f32 (fround (v2f64 V128:$Rn)))),
2468 (FCVTNv4i32 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
2469 defm FCVTPS : SIMDTwoVectorFPToInt<0,1,0b11010, "fcvtps",int_arm64_neon_fcvtps>;
2470 defm FCVTPU : SIMDTwoVectorFPToInt<1,1,0b11010, "fcvtpu",int_arm64_neon_fcvtpu>;
2471 defm FCVTXN : SIMDFPInexactCvtTwoVector<1, 0, 0b10110, "fcvtxn",
2472 int_arm64_neon_fcvtxn>;
2473 defm FCVTZS : SIMDTwoVectorFPToInt<0, 1, 0b11011, "fcvtzs", fp_to_sint>;
2474 defm FCVTZU : SIMDTwoVectorFPToInt<1, 1, 0b11011, "fcvtzu", fp_to_uint>;
2475 let isCodeGenOnly = 1 in {
2476 defm FCVTZS_Int : SIMDTwoVectorFPToInt<0, 1, 0b11011, "fcvtzs",
2477 int_arm64_neon_fcvtzs>;
2478 defm FCVTZU_Int : SIMDTwoVectorFPToInt<1, 1, 0b11011, "fcvtzu",
2479 int_arm64_neon_fcvtzu>;
2481 defm FNEG : SIMDTwoVectorFP<1, 1, 0b01111, "fneg", fneg>;
2482 defm FRECPE : SIMDTwoVectorFP<0, 1, 0b11101, "frecpe", int_arm64_neon_frecpe>;
2483 defm FRINTA : SIMDTwoVectorFP<1, 0, 0b11000, "frinta", frnd>;
2484 defm FRINTI : SIMDTwoVectorFP<1, 1, 0b11001, "frinti", fnearbyint>;
2485 defm FRINTM : SIMDTwoVectorFP<0, 0, 0b11001, "frintm", ffloor>;
2486 defm FRINTN : SIMDTwoVectorFP<0, 0, 0b11000, "frintn", int_arm64_neon_frintn>;
2487 defm FRINTP : SIMDTwoVectorFP<0, 1, 0b11000, "frintp", fceil>;
2488 defm FRINTX : SIMDTwoVectorFP<1, 0, 0b11001, "frintx", frint>;
2489 defm FRINTZ : SIMDTwoVectorFP<0, 1, 0b11001, "frintz", ftrunc>;
2490 defm FRSQRTE: SIMDTwoVectorFP<1, 1, 0b11101, "frsqrte", int_arm64_neon_frsqrte>;
2491 defm FSQRT : SIMDTwoVectorFP<1, 1, 0b11111, "fsqrt", fsqrt>;
2492 defm NEG : SIMDTwoVectorBHSD<1, 0b01011, "neg",
2493 UnOpFrag<(sub immAllZerosV, node:$LHS)> >;
2494 defm NOT : SIMDTwoVectorB<1, 0b00, 0b00101, "not", vnot>;
2495 // Aliases for MVN -> NOT.
2496 def : InstAlias<"mvn{ $Vd.8b, $Vn.8b|.8b $Vd, $Vn}",
2497 (NOTv8i8 V64:$Vd, V64:$Vn)>;
2498 def : InstAlias<"mvn{ $Vd.16b, $Vn.16b|.16b $Vd, $Vn}",
2499 (NOTv16i8 V128:$Vd, V128:$Vn)>;
2501 def : Pat<(ARM64neg (v8i8 V64:$Rn)), (NEGv8i8 V64:$Rn)>;
2502 def : Pat<(ARM64neg (v16i8 V128:$Rn)), (NEGv16i8 V128:$Rn)>;
2503 def : Pat<(ARM64neg (v4i16 V64:$Rn)), (NEGv4i16 V64:$Rn)>;
2504 def : Pat<(ARM64neg (v8i16 V128:$Rn)), (NEGv8i16 V128:$Rn)>;
2505 def : Pat<(ARM64neg (v2i32 V64:$Rn)), (NEGv2i32 V64:$Rn)>;
2506 def : Pat<(ARM64neg (v4i32 V128:$Rn)), (NEGv4i32 V128:$Rn)>;
2507 def : Pat<(ARM64neg (v2i64 V128:$Rn)), (NEGv2i64 V128:$Rn)>;
2509 def : Pat<(ARM64not (v8i8 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2510 def : Pat<(ARM64not (v16i8 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2511 def : Pat<(ARM64not (v4i16 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2512 def : Pat<(ARM64not (v8i16 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2513 def : Pat<(ARM64not (v2i32 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2514 def : Pat<(ARM64not (v1i64 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2515 def : Pat<(ARM64not (v4i32 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2516 def : Pat<(ARM64not (v2i64 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2518 def : Pat<(vnot (v4i16 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2519 def : Pat<(vnot (v8i16 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2520 def : Pat<(vnot (v2i32 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2521 def : Pat<(vnot (v4i32 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2522 def : Pat<(vnot (v2i64 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2524 defm RBIT : SIMDTwoVectorB<1, 0b01, 0b00101, "rbit", int_arm64_neon_rbit>;
2525 defm REV16 : SIMDTwoVectorB<0, 0b00, 0b00001, "rev16", ARM64rev16>;
2526 defm REV32 : SIMDTwoVectorBH<1, 0b00000, "rev32", ARM64rev32>;
2527 defm REV64 : SIMDTwoVectorBHS<0, 0b00000, "rev64", ARM64rev64>;
2528 defm SADALP : SIMDLongTwoVectorTied<0, 0b00110, "sadalp",
2529 BinOpFrag<(add node:$LHS, (int_arm64_neon_saddlp node:$RHS))> >;
2530 defm SADDLP : SIMDLongTwoVector<0, 0b00010, "saddlp", int_arm64_neon_saddlp>;
2531 defm SCVTF : SIMDTwoVectorIntToFP<0, 0, 0b11101, "scvtf", sint_to_fp>;
2532 defm SHLL : SIMDVectorLShiftLongBySizeBHS;
2533 defm SQABS : SIMDTwoVectorBHSD<0, 0b00111, "sqabs", int_arm64_neon_sqabs>;
2534 defm SQNEG : SIMDTwoVectorBHSD<1, 0b00111, "sqneg", int_arm64_neon_sqneg>;
2535 defm SQXTN : SIMDMixedTwoVector<0, 0b10100, "sqxtn", int_arm64_neon_sqxtn>;
2536 defm SQXTUN : SIMDMixedTwoVector<1, 0b10010, "sqxtun", int_arm64_neon_sqxtun>;
2537 defm SUQADD : SIMDTwoVectorBHSDTied<0, 0b00011, "suqadd",int_arm64_neon_suqadd>;
2538 defm UADALP : SIMDLongTwoVectorTied<1, 0b00110, "uadalp",
2539 BinOpFrag<(add node:$LHS, (int_arm64_neon_uaddlp node:$RHS))> >;
2540 defm UADDLP : SIMDLongTwoVector<1, 0b00010, "uaddlp",
2541 int_arm64_neon_uaddlp>;
2542 defm UCVTF : SIMDTwoVectorIntToFP<1, 0, 0b11101, "ucvtf", uint_to_fp>;
2543 defm UQXTN : SIMDMixedTwoVector<1, 0b10100, "uqxtn", int_arm64_neon_uqxtn>;
2544 defm URECPE : SIMDTwoVectorS<0, 1, 0b11100, "urecpe", int_arm64_neon_urecpe>;
2545 defm URSQRTE: SIMDTwoVectorS<1, 1, 0b11100, "ursqrte", int_arm64_neon_ursqrte>;
2546 defm USQADD : SIMDTwoVectorBHSDTied<1, 0b00011, "usqadd",int_arm64_neon_usqadd>;
2547 defm XTN : SIMDMixedTwoVector<0, 0b10010, "xtn", trunc>;
2549 def : Pat<(v2f32 (ARM64rev64 V64:$Rn)), (REV64v2i32 V64:$Rn)>;
2550 def : Pat<(v4f32 (ARM64rev64 V128:$Rn)), (REV64v4i32 V128:$Rn)>;
2552 // Patterns for vector long shift (by element width). These need to match all
2553 // three of zext, sext and anyext so it's easier to pull the patterns out of the
2555 multiclass SIMDVectorLShiftLongBySizeBHSPats<SDPatternOperator ext> {
2556 def : Pat<(ARM64vshl (v8i16 (ext (v8i8 V64:$Rn))), (i32 8)),
2557 (SHLLv8i8 V64:$Rn)>;
2558 def : Pat<(ARM64vshl (v8i16 (ext (extract_high_v16i8 V128:$Rn))), (i32 8)),
2559 (SHLLv16i8 V128:$Rn)>;
2560 def : Pat<(ARM64vshl (v4i32 (ext (v4i16 V64:$Rn))), (i32 16)),
2561 (SHLLv4i16 V64:$Rn)>;
2562 def : Pat<(ARM64vshl (v4i32 (ext (extract_high_v8i16 V128:$Rn))), (i32 16)),
2563 (SHLLv8i16 V128:$Rn)>;
2564 def : Pat<(ARM64vshl (v2i64 (ext (v2i32 V64:$Rn))), (i32 32)),
2565 (SHLLv2i32 V64:$Rn)>;
2566 def : Pat<(ARM64vshl (v2i64 (ext (extract_high_v4i32 V128:$Rn))), (i32 32)),
2567 (SHLLv4i32 V128:$Rn)>;
2570 defm : SIMDVectorLShiftLongBySizeBHSPats<anyext>;
2571 defm : SIMDVectorLShiftLongBySizeBHSPats<zext>;
2572 defm : SIMDVectorLShiftLongBySizeBHSPats<sext>;
2574 //===----------------------------------------------------------------------===//
2575 // Advanced SIMD three vector instructions.
2576 //===----------------------------------------------------------------------===//
2578 defm ADD : SIMDThreeSameVector<0, 0b10000, "add", add>;
2579 defm ADDP : SIMDThreeSameVector<0, 0b10111, "addp", int_arm64_neon_addp>;
2580 defm CMEQ : SIMDThreeSameVector<1, 0b10001, "cmeq", ARM64cmeq>;
2581 defm CMGE : SIMDThreeSameVector<0, 0b00111, "cmge", ARM64cmge>;
2582 defm CMGT : SIMDThreeSameVector<0, 0b00110, "cmgt", ARM64cmgt>;
2583 defm CMHI : SIMDThreeSameVector<1, 0b00110, "cmhi", ARM64cmhi>;
2584 defm CMHS : SIMDThreeSameVector<1, 0b00111, "cmhs", ARM64cmhs>;
2585 defm CMTST : SIMDThreeSameVector<0, 0b10001, "cmtst", ARM64cmtst>;
2586 defm FABD : SIMDThreeSameVectorFP<1,1,0b11010,"fabd", int_arm64_neon_fabd>;
2587 defm FACGE : SIMDThreeSameVectorFPCmp<1,0,0b11101,"facge",int_arm64_neon_facge>;
2588 defm FACGT : SIMDThreeSameVectorFPCmp<1,1,0b11101,"facgt",int_arm64_neon_facgt>;
2589 defm FADDP : SIMDThreeSameVectorFP<1,0,0b11010,"faddp",int_arm64_neon_addp>;
2590 defm FADD : SIMDThreeSameVectorFP<0,0,0b11010,"fadd", fadd>;
2591 defm FCMEQ : SIMDThreeSameVectorFPCmp<0, 0, 0b11100, "fcmeq", ARM64fcmeq>;
2592 defm FCMGE : SIMDThreeSameVectorFPCmp<1, 0, 0b11100, "fcmge", ARM64fcmge>;
2593 defm FCMGT : SIMDThreeSameVectorFPCmp<1, 1, 0b11100, "fcmgt", ARM64fcmgt>;
2594 defm FDIV : SIMDThreeSameVectorFP<1,0,0b11111,"fdiv", fdiv>;
2595 defm FMAXNMP : SIMDThreeSameVectorFP<1,0,0b11000,"fmaxnmp", int_arm64_neon_fmaxnmp>;
2596 defm FMAXNM : SIMDThreeSameVectorFP<0,0,0b11000,"fmaxnm", int_arm64_neon_fmaxnm>;
2597 defm FMAXP : SIMDThreeSameVectorFP<1,0,0b11110,"fmaxp", int_arm64_neon_fmaxp>;
2598 defm FMAX : SIMDThreeSameVectorFP<0,0,0b11110,"fmax", ARM64fmax>;
2599 defm FMINNMP : SIMDThreeSameVectorFP<1,1,0b11000,"fminnmp", int_arm64_neon_fminnmp>;
2600 defm FMINNM : SIMDThreeSameVectorFP<0,1,0b11000,"fminnm", int_arm64_neon_fminnm>;
2601 defm FMINP : SIMDThreeSameVectorFP<1,1,0b11110,"fminp", int_arm64_neon_fminp>;
2602 defm FMIN : SIMDThreeSameVectorFP<0,1,0b11110,"fmin", ARM64fmin>;
2604 // NOTE: The operands of the PatFrag are reordered on FMLA/FMLS because the
2605 // instruction expects the addend first, while the fma intrinsic puts it last.
2606 defm FMLA : SIMDThreeSameVectorFPTied<0, 0, 0b11001, "fmla",
2607 TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)> >;
2608 defm FMLS : SIMDThreeSameVectorFPTied<0, 1, 0b11001, "fmls",
2609 TriOpFrag<(fma node:$MHS, (fneg node:$RHS), node:$LHS)> >;
2611 // The following def pats catch the case where the LHS of an FMA is negated.
2612 // The TriOpFrag above catches the case where the middle operand is negated.
2613 def : Pat<(v2f32 (fma (fneg V64:$Rn), V64:$Rm, V64:$Rd)),
2614 (FMLSv2f32 V64:$Rd, V64:$Rn, V64:$Rm)>;
2616 def : Pat<(v4f32 (fma (fneg V128:$Rn), V128:$Rm, V128:$Rd)),
2617 (FMLSv4f32 V128:$Rd, V128:$Rn, V128:$Rm)>;
2619 def : Pat<(v2f64 (fma (fneg V128:$Rn), V128:$Rm, V128:$Rd)),
2620 (FMLSv2f64 V128:$Rd, V128:$Rn, V128:$Rm)>;
2622 defm FMULX : SIMDThreeSameVectorFP<0,0,0b11011,"fmulx", int_arm64_neon_fmulx>;
2623 defm FMUL : SIMDThreeSameVectorFP<1,0,0b11011,"fmul", fmul>;
2624 defm FRECPS : SIMDThreeSameVectorFP<0,0,0b11111,"frecps", int_arm64_neon_frecps>;
2625 defm FRSQRTS : SIMDThreeSameVectorFP<0,1,0b11111,"frsqrts", int_arm64_neon_frsqrts>;
2626 defm FSUB : SIMDThreeSameVectorFP<0,1,0b11010,"fsub", fsub>;
2627 defm MLA : SIMDThreeSameVectorBHSTied<0, 0b10010, "mla",
2628 TriOpFrag<(add node:$LHS, (mul node:$MHS, node:$RHS))> >;
2629 defm MLS : SIMDThreeSameVectorBHSTied<1, 0b10010, "mls",
2630 TriOpFrag<(sub node:$LHS, (mul node:$MHS, node:$RHS))> >;
2631 defm MUL : SIMDThreeSameVectorBHS<0, 0b10011, "mul", mul>;
2632 defm PMUL : SIMDThreeSameVectorB<1, 0b10011, "pmul", int_arm64_neon_pmul>;
2633 defm SABA : SIMDThreeSameVectorBHSTied<0, 0b01111, "saba",
2634 TriOpFrag<(add node:$LHS, (int_arm64_neon_sabd node:$MHS, node:$RHS))> >;
2635 defm SABD : SIMDThreeSameVectorBHS<0,0b01110,"sabd", int_arm64_neon_sabd>;
2636 defm SHADD : SIMDThreeSameVectorBHS<0,0b00000,"shadd", int_arm64_neon_shadd>;
2637 defm SHSUB : SIMDThreeSameVectorBHS<0,0b00100,"shsub", int_arm64_neon_shsub>;
2638 defm SMAXP : SIMDThreeSameVectorBHS<0,0b10100,"smaxp", int_arm64_neon_smaxp>;
2639 defm SMAX : SIMDThreeSameVectorBHS<0,0b01100,"smax", int_arm64_neon_smax>;
2640 defm SMINP : SIMDThreeSameVectorBHS<0,0b10101,"sminp", int_arm64_neon_sminp>;
2641 defm SMIN : SIMDThreeSameVectorBHS<0,0b01101,"smin", int_arm64_neon_smin>;
2642 defm SQADD : SIMDThreeSameVector<0,0b00001,"sqadd", int_arm64_neon_sqadd>;
2643 defm SQDMULH : SIMDThreeSameVectorHS<0,0b10110,"sqdmulh",int_arm64_neon_sqdmulh>;
2644 defm SQRDMULH : SIMDThreeSameVectorHS<1,0b10110,"sqrdmulh",int_arm64_neon_sqrdmulh>;
2645 defm SQRSHL : SIMDThreeSameVector<0,0b01011,"sqrshl", int_arm64_neon_sqrshl>;
2646 defm SQSHL : SIMDThreeSameVector<0,0b01001,"sqshl", int_arm64_neon_sqshl>;
2647 defm SQSUB : SIMDThreeSameVector<0,0b00101,"sqsub", int_arm64_neon_sqsub>;
2648 defm SRHADD : SIMDThreeSameVectorBHS<0,0b00010,"srhadd",int_arm64_neon_srhadd>;
2649 defm SRSHL : SIMDThreeSameVector<0,0b01010,"srshl", int_arm64_neon_srshl>;
2650 defm SSHL : SIMDThreeSameVector<0,0b01000,"sshl", int_arm64_neon_sshl>;
2651 defm SUB : SIMDThreeSameVector<1,0b10000,"sub", sub>;
2652 defm UABA : SIMDThreeSameVectorBHSTied<1, 0b01111, "uaba",
2653 TriOpFrag<(add node:$LHS, (int_arm64_neon_uabd node:$MHS, node:$RHS))> >;
2654 defm UABD : SIMDThreeSameVectorBHS<1,0b01110,"uabd", int_arm64_neon_uabd>;
2655 defm UHADD : SIMDThreeSameVectorBHS<1,0b00000,"uhadd", int_arm64_neon_uhadd>;
2656 defm UHSUB : SIMDThreeSameVectorBHS<1,0b00100,"uhsub", int_arm64_neon_uhsub>;
2657 defm UMAXP : SIMDThreeSameVectorBHS<1,0b10100,"umaxp", int_arm64_neon_umaxp>;
2658 defm UMAX : SIMDThreeSameVectorBHS<1,0b01100,"umax", int_arm64_neon_umax>;
2659 defm UMINP : SIMDThreeSameVectorBHS<1,0b10101,"uminp", int_arm64_neon_uminp>;
2660 defm UMIN : SIMDThreeSameVectorBHS<1,0b01101,"umin", int_arm64_neon_umin>;
2661 defm UQADD : SIMDThreeSameVector<1,0b00001,"uqadd", int_arm64_neon_uqadd>;
2662 defm UQRSHL : SIMDThreeSameVector<1,0b01011,"uqrshl", int_arm64_neon_uqrshl>;
2663 defm UQSHL : SIMDThreeSameVector<1,0b01001,"uqshl", int_arm64_neon_uqshl>;
2664 defm UQSUB : SIMDThreeSameVector<1,0b00101,"uqsub", int_arm64_neon_uqsub>;
2665 defm URHADD : SIMDThreeSameVectorBHS<1,0b00010,"urhadd", int_arm64_neon_urhadd>;
2666 defm URSHL : SIMDThreeSameVector<1,0b01010,"urshl", int_arm64_neon_urshl>;
2667 defm USHL : SIMDThreeSameVector<1,0b01000,"ushl", int_arm64_neon_ushl>;
2669 defm AND : SIMDLogicalThreeVector<0, 0b00, "and", and>;
2670 defm BIC : SIMDLogicalThreeVector<0, 0b01, "bic",
2671 BinOpFrag<(and node:$LHS, (vnot node:$RHS))> >;
2672 defm BIF : SIMDLogicalThreeVector<1, 0b11, "bif">;
2673 defm BIT : SIMDLogicalThreeVectorTied<1, 0b10, "bit", ARM64bit>;
2674 defm BSL : SIMDLogicalThreeVectorTied<1, 0b01, "bsl",
2675 TriOpFrag<(or (and node:$LHS, node:$MHS), (and (vnot node:$LHS), node:$RHS))>>;
2676 defm EOR : SIMDLogicalThreeVector<1, 0b00, "eor", xor>;
2677 defm ORN : SIMDLogicalThreeVector<0, 0b11, "orn",
2678 BinOpFrag<(or node:$LHS, (vnot node:$RHS))> >;
2679 defm ORR : SIMDLogicalThreeVector<0, 0b10, "orr", or>;
2681 def : Pat<(ARM64bsl (v8i8 V64:$Rd), V64:$Rn, V64:$Rm),
2682 (BSLv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
2683 def : Pat<(ARM64bsl (v4i16 V64:$Rd), V64:$Rn, V64:$Rm),
2684 (BSLv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
2685 def : Pat<(ARM64bsl (v2i32 V64:$Rd), V64:$Rn, V64:$Rm),
2686 (BSLv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
2687 def : Pat<(ARM64bsl (v1i64 V64:$Rd), V64:$Rn, V64:$Rm),
2688 (BSLv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
2690 def : Pat<(ARM64bsl (v16i8 V128:$Rd), V128:$Rn, V128:$Rm),
2691 (BSLv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
2692 def : Pat<(ARM64bsl (v8i16 V128:$Rd), V128:$Rn, V128:$Rm),
2693 (BSLv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
2694 def : Pat<(ARM64bsl (v4i32 V128:$Rd), V128:$Rn, V128:$Rm),
2695 (BSLv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
2696 def : Pat<(ARM64bsl (v2i64 V128:$Rd), V128:$Rn, V128:$Rm),
2697 (BSLv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
2699 def : InstAlias<"mov{\t$dst.16b, $src.16b|.16b\t$dst, $src}",
2700 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 1>;
2701 def : InstAlias<"mov{\t$dst.8h, $src.8h|.8h\t$dst, $src}",
2702 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2703 def : InstAlias<"mov{\t$dst.4s, $src.4s|.4s\t$dst, $src}",
2704 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2705 def : InstAlias<"mov{\t$dst.2d, $src.2d|.2d\t$dst, $src}",
2706 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2708 def : InstAlias<"mov{\t$dst.8b, $src.8b|.8b\t$dst, $src}",
2709 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 1>;
2710 def : InstAlias<"mov{\t$dst.4h, $src.4h|.4h\t$dst, $src}",
2711 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2712 def : InstAlias<"mov{\t$dst.2s, $src.2s|.2s\t$dst, $src}",
2713 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2714 def : InstAlias<"mov{\t$dst.1d, $src.1d|.1d\t$dst, $src}",
2715 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2717 def : InstAlias<"{cmls\t$dst.8b, $src1.8b, $src2.8b" #
2718 "|cmls.8b\t$dst, $src1, $src2}",
2719 (CMHSv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2720 def : InstAlias<"{cmls\t$dst.16b, $src1.16b, $src2.16b" #
2721 "|cmls.16b\t$dst, $src1, $src2}",
2722 (CMHSv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2723 def : InstAlias<"{cmls\t$dst.4h, $src1.4h, $src2.4h" #
2724 "|cmls.4h\t$dst, $src1, $src2}",
2725 (CMHSv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2726 def : InstAlias<"{cmls\t$dst.8h, $src1.8h, $src2.8h" #
2727 "|cmls.8h\t$dst, $src1, $src2}",
2728 (CMHSv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2729 def : InstAlias<"{cmls\t$dst.2s, $src1.2s, $src2.2s" #
2730 "|cmls.2s\t$dst, $src1, $src2}",
2731 (CMHSv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2732 def : InstAlias<"{cmls\t$dst.4s, $src1.4s, $src2.4s" #
2733 "|cmls.4s\t$dst, $src1, $src2}",
2734 (CMHSv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2735 def : InstAlias<"{cmls\t$dst.2d, $src1.2d, $src2.2d" #
2736 "|cmls.2d\t$dst, $src1, $src2}",
2737 (CMHSv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2739 def : InstAlias<"{cmlo\t$dst.8b, $src1.8b, $src2.8b" #
2740 "|cmlo.8b\t$dst, $src1, $src2}",
2741 (CMHIv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2742 def : InstAlias<"{cmlo\t$dst.16b, $src1.16b, $src2.16b" #
2743 "|cmlo.16b\t$dst, $src1, $src2}",
2744 (CMHIv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2745 def : InstAlias<"{cmlo\t$dst.4h, $src1.4h, $src2.4h" #
2746 "|cmlo.4h\t$dst, $src1, $src2}",
2747 (CMHIv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2748 def : InstAlias<"{cmlo\t$dst.8h, $src1.8h, $src2.8h" #
2749 "|cmlo.8h\t$dst, $src1, $src2}",
2750 (CMHIv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2751 def : InstAlias<"{cmlo\t$dst.2s, $src1.2s, $src2.2s" #
2752 "|cmlo.2s\t$dst, $src1, $src2}",
2753 (CMHIv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2754 def : InstAlias<"{cmlo\t$dst.4s, $src1.4s, $src2.4s" #
2755 "|cmlo.4s\t$dst, $src1, $src2}",
2756 (CMHIv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2757 def : InstAlias<"{cmlo\t$dst.2d, $src1.2d, $src2.2d" #
2758 "|cmlo.2d\t$dst, $src1, $src2}",
2759 (CMHIv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2761 def : InstAlias<"{cmle\t$dst.8b, $src1.8b, $src2.8b" #
2762 "|cmle.8b\t$dst, $src1, $src2}",
2763 (CMGEv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2764 def : InstAlias<"{cmle\t$dst.16b, $src1.16b, $src2.16b" #
2765 "|cmle.16b\t$dst, $src1, $src2}",
2766 (CMGEv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2767 def : InstAlias<"{cmle\t$dst.4h, $src1.4h, $src2.4h" #
2768 "|cmle.4h\t$dst, $src1, $src2}",
2769 (CMGEv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2770 def : InstAlias<"{cmle\t$dst.8h, $src1.8h, $src2.8h" #
2771 "|cmle.8h\t$dst, $src1, $src2}",
2772 (CMGEv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2773 def : InstAlias<"{cmle\t$dst.2s, $src1.2s, $src2.2s" #
2774 "|cmle.2s\t$dst, $src1, $src2}",
2775 (CMGEv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2776 def : InstAlias<"{cmle\t$dst.4s, $src1.4s, $src2.4s" #
2777 "|cmle.4s\t$dst, $src1, $src2}",
2778 (CMGEv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2779 def : InstAlias<"{cmle\t$dst.2d, $src1.2d, $src2.2d" #
2780 "|cmle.2d\t$dst, $src1, $src2}",
2781 (CMGEv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2783 def : InstAlias<"{cmlt\t$dst.8b, $src1.8b, $src2.8b" #
2784 "|cmlt.8b\t$dst, $src1, $src2}",
2785 (CMGTv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2786 def : InstAlias<"{cmlt\t$dst.16b, $src1.16b, $src2.16b" #
2787 "|cmlt.16b\t$dst, $src1, $src2}",
2788 (CMGTv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2789 def : InstAlias<"{cmlt\t$dst.4h, $src1.4h, $src2.4h" #
2790 "|cmlt.4h\t$dst, $src1, $src2}",
2791 (CMGTv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2792 def : InstAlias<"{cmlt\t$dst.8h, $src1.8h, $src2.8h" #
2793 "|cmlt.8h\t$dst, $src1, $src2}",
2794 (CMGTv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2795 def : InstAlias<"{cmlt\t$dst.2s, $src1.2s, $src2.2s" #
2796 "|cmlt.2s\t$dst, $src1, $src2}",
2797 (CMGTv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2798 def : InstAlias<"{cmlt\t$dst.4s, $src1.4s, $src2.4s" #
2799 "|cmlt.4s\t$dst, $src1, $src2}",
2800 (CMGTv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2801 def : InstAlias<"{cmlt\t$dst.2d, $src1.2d, $src2.2d" #
2802 "|cmlt.2d\t$dst, $src1, $src2}",
2803 (CMGTv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2805 def : InstAlias<"{fcmle\t$dst.2s, $src1.2s, $src2.2s" #
2806 "|fcmle.2s\t$dst, $src1, $src2}",
2807 (FCMGEv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
2808 def : InstAlias<"{fcmle\t$dst.4s, $src1.4s, $src2.4s" #
2809 "|fcmle.4s\t$dst, $src1, $src2}",
2810 (FCMGEv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
2811 def : InstAlias<"{fcmle\t$dst.2d, $src1.2d, $src2.2d" #
2812 "|fcmle.2d\t$dst, $src1, $src2}",
2813 (FCMGEv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
2815 def : InstAlias<"{fcmlt\t$dst.2s, $src1.2s, $src2.2s" #
2816 "|fcmlt.2s\t$dst, $src1, $src2}",
2817 (FCMGTv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
2818 def : InstAlias<"{fcmlt\t$dst.4s, $src1.4s, $src2.4s" #
2819 "|fcmlt.4s\t$dst, $src1, $src2}",
2820 (FCMGTv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
2821 def : InstAlias<"{fcmlt\t$dst.2d, $src1.2d, $src2.2d" #
2822 "|fcmlt.2d\t$dst, $src1, $src2}",
2823 (FCMGTv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
2825 def : InstAlias<"{facle\t$dst.2s, $src1.2s, $src2.2s" #
2826 "|facle.2s\t$dst, $src1, $src2}",
2827 (FACGEv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
2828 def : InstAlias<"{facle\t$dst.4s, $src1.4s, $src2.4s" #
2829 "|facle.4s\t$dst, $src1, $src2}",
2830 (FACGEv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
2831 def : InstAlias<"{facle\t$dst.2d, $src1.2d, $src2.2d" #
2832 "|facle.2d\t$dst, $src1, $src2}",
2833 (FACGEv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
2835 def : InstAlias<"{faclt\t$dst.2s, $src1.2s, $src2.2s" #
2836 "|faclt.2s\t$dst, $src1, $src2}",
2837 (FACGTv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
2838 def : InstAlias<"{faclt\t$dst.4s, $src1.4s, $src2.4s" #
2839 "|faclt.4s\t$dst, $src1, $src2}",
2840 (FACGTv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
2841 def : InstAlias<"{faclt\t$dst.2d, $src1.2d, $src2.2d" #
2842 "|faclt.2d\t$dst, $src1, $src2}",
2843 (FACGTv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
2845 //===----------------------------------------------------------------------===//
2846 // Advanced SIMD three scalar instructions.
2847 //===----------------------------------------------------------------------===//
2849 defm ADD : SIMDThreeScalarD<0, 0b10000, "add", add>;
2850 defm CMEQ : SIMDThreeScalarD<1, 0b10001, "cmeq", ARM64cmeq>;
2851 defm CMGE : SIMDThreeScalarD<0, 0b00111, "cmge", ARM64cmge>;
2852 defm CMGT : SIMDThreeScalarD<0, 0b00110, "cmgt", ARM64cmgt>;
2853 defm CMHI : SIMDThreeScalarD<1, 0b00110, "cmhi", ARM64cmhi>;
2854 defm CMHS : SIMDThreeScalarD<1, 0b00111, "cmhs", ARM64cmhs>;
2855 defm CMTST : SIMDThreeScalarD<0, 0b10001, "cmtst", ARM64cmtst>;
2856 defm FABD : SIMDThreeScalarSD<1, 1, 0b11010, "fabd", int_arm64_sisd_fabd>;
2857 def : Pat<(v1f64 (int_arm64_neon_fabd (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2858 (FABD64 FPR64:$Rn, FPR64:$Rm)>;
2859 defm FACGE : SIMDThreeScalarFPCmp<1, 0, 0b11101, "facge",
2860 int_arm64_neon_facge>;
2861 defm FACGT : SIMDThreeScalarFPCmp<1, 1, 0b11101, "facgt",
2862 int_arm64_neon_facgt>;
2863 defm FCMEQ : SIMDThreeScalarFPCmp<0, 0, 0b11100, "fcmeq", ARM64fcmeq>;
2864 defm FCMGE : SIMDThreeScalarFPCmp<1, 0, 0b11100, "fcmge", ARM64fcmge>;
2865 defm FCMGT : SIMDThreeScalarFPCmp<1, 1, 0b11100, "fcmgt", ARM64fcmgt>;
2866 defm FMULX : SIMDThreeScalarSD<0, 0, 0b11011, "fmulx", int_arm64_neon_fmulx>;
2867 defm FRECPS : SIMDThreeScalarSD<0, 0, 0b11111, "frecps", int_arm64_neon_frecps>;
2868 defm FRSQRTS : SIMDThreeScalarSD<0, 1, 0b11111, "frsqrts", int_arm64_neon_frsqrts>;
2869 defm SQADD : SIMDThreeScalarBHSD<0, 0b00001, "sqadd", int_arm64_neon_sqadd>;
2870 defm SQDMULH : SIMDThreeScalarHS< 0, 0b10110, "sqdmulh", int_arm64_neon_sqdmulh>;
2871 defm SQRDMULH : SIMDThreeScalarHS< 1, 0b10110, "sqrdmulh", int_arm64_neon_sqrdmulh>;
2872 defm SQRSHL : SIMDThreeScalarBHSD<0, 0b01011, "sqrshl",int_arm64_neon_sqrshl>;
2873 defm SQSHL : SIMDThreeScalarBHSD<0, 0b01001, "sqshl", int_arm64_neon_sqshl>;
2874 defm SQSUB : SIMDThreeScalarBHSD<0, 0b00101, "sqsub", int_arm64_neon_sqsub>;
2875 defm SRSHL : SIMDThreeScalarD< 0, 0b01010, "srshl", int_arm64_neon_srshl>;
2876 defm SSHL : SIMDThreeScalarD< 0, 0b01000, "sshl", int_arm64_neon_sshl>;
2877 defm SUB : SIMDThreeScalarD< 1, 0b10000, "sub", sub>;
2878 defm UQADD : SIMDThreeScalarBHSD<1, 0b00001, "uqadd", int_arm64_neon_uqadd>;
2879 defm UQRSHL : SIMDThreeScalarBHSD<1, 0b01011, "uqrshl",int_arm64_neon_uqrshl>;
2880 defm UQSHL : SIMDThreeScalarBHSD<1, 0b01001, "uqshl", int_arm64_neon_uqshl>;
2881 defm UQSUB : SIMDThreeScalarBHSD<1, 0b00101, "uqsub", int_arm64_neon_uqsub>;
2882 defm URSHL : SIMDThreeScalarD< 1, 0b01010, "urshl", int_arm64_neon_urshl>;
2883 defm USHL : SIMDThreeScalarD< 1, 0b01000, "ushl", int_arm64_neon_ushl>;
2885 def : InstAlias<"cmls $dst, $src1, $src2",
2886 (CMHSv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
2887 def : InstAlias<"cmle $dst, $src1, $src2",
2888 (CMGEv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
2889 def : InstAlias<"cmlo $dst, $src1, $src2",
2890 (CMHIv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
2891 def : InstAlias<"cmlt $dst, $src1, $src2",
2892 (CMGTv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
2893 def : InstAlias<"fcmle $dst, $src1, $src2",
2894 (FCMGE32 FPR32:$dst, FPR32:$src2, FPR32:$src1), 0>;
2895 def : InstAlias<"fcmle $dst, $src1, $src2",
2896 (FCMGE64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
2897 def : InstAlias<"fcmlt $dst, $src1, $src2",
2898 (FCMGT32 FPR32:$dst, FPR32:$src2, FPR32:$src1), 0>;
2899 def : InstAlias<"fcmlt $dst, $src1, $src2",
2900 (FCMGT64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
2901 def : InstAlias<"facle $dst, $src1, $src2",
2902 (FACGE32 FPR32:$dst, FPR32:$src2, FPR32:$src1), 0>;
2903 def : InstAlias<"facle $dst, $src1, $src2",
2904 (FACGE64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
2905 def : InstAlias<"faclt $dst, $src1, $src2",
2906 (FACGT32 FPR32:$dst, FPR32:$src2, FPR32:$src1), 0>;
2907 def : InstAlias<"faclt $dst, $src1, $src2",
2908 (FACGT64 FPR64:$dst, FPR64:$src2, FPR64:$src1), 0>;
2910 //===----------------------------------------------------------------------===//
2911 // Advanced SIMD three scalar instructions (mixed operands).
2912 //===----------------------------------------------------------------------===//
2913 defm SQDMULL : SIMDThreeScalarMixedHS<0, 0b11010, "sqdmull",
2914 int_arm64_neon_sqdmulls_scalar>;
2915 defm SQDMLAL : SIMDThreeScalarMixedTiedHS<0, 0b10010, "sqdmlal">;
2916 defm SQDMLSL : SIMDThreeScalarMixedTiedHS<0, 0b10110, "sqdmlsl">;
2918 def : Pat<(i64 (int_arm64_neon_sqadd (i64 FPR64:$Rd),
2919 (i64 (int_arm64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
2920 (i32 FPR32:$Rm))))),
2921 (SQDMLALi32 FPR64:$Rd, FPR32:$Rn, FPR32:$Rm)>;
2922 def : Pat<(i64 (int_arm64_neon_sqsub (i64 FPR64:$Rd),
2923 (i64 (int_arm64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
2924 (i32 FPR32:$Rm))))),
2925 (SQDMLSLi32 FPR64:$Rd, FPR32:$Rn, FPR32:$Rm)>;
2927 //===----------------------------------------------------------------------===//
2928 // Advanced SIMD two scalar instructions.
2929 //===----------------------------------------------------------------------===//
2931 defm ABS : SIMDTwoScalarD< 0, 0b01011, "abs", int_arm64_neon_abs>;
2932 defm CMEQ : SIMDCmpTwoScalarD< 0, 0b01001, "cmeq", ARM64cmeqz>;
2933 defm CMGE : SIMDCmpTwoScalarD< 1, 0b01000, "cmge", ARM64cmgez>;
2934 defm CMGT : SIMDCmpTwoScalarD< 0, 0b01000, "cmgt", ARM64cmgtz>;
2935 defm CMLE : SIMDCmpTwoScalarD< 1, 0b01001, "cmle", ARM64cmlez>;
2936 defm CMLT : SIMDCmpTwoScalarD< 0, 0b01010, "cmlt", ARM64cmltz>;
2937 defm FCMEQ : SIMDCmpTwoScalarSD<0, 1, 0b01101, "fcmeq", ARM64fcmeqz>;
2938 defm FCMGE : SIMDCmpTwoScalarSD<1, 1, 0b01100, "fcmge", ARM64fcmgez>;
2939 defm FCMGT : SIMDCmpTwoScalarSD<0, 1, 0b01100, "fcmgt", ARM64fcmgtz>;
2940 defm FCMLE : SIMDCmpTwoScalarSD<1, 1, 0b01101, "fcmle", ARM64fcmlez>;
2941 defm FCMLT : SIMDCmpTwoScalarSD<0, 1, 0b01110, "fcmlt", ARM64fcmltz>;
2942 defm FCVTAS : SIMDTwoScalarSD< 0, 0, 0b11100, "fcvtas">;
2943 defm FCVTAU : SIMDTwoScalarSD< 1, 0, 0b11100, "fcvtau">;
2944 defm FCVTMS : SIMDTwoScalarSD< 0, 0, 0b11011, "fcvtms">;
2945 defm FCVTMU : SIMDTwoScalarSD< 1, 0, 0b11011, "fcvtmu">;
2946 defm FCVTNS : SIMDTwoScalarSD< 0, 0, 0b11010, "fcvtns">;
2947 defm FCVTNU : SIMDTwoScalarSD< 1, 0, 0b11010, "fcvtnu">;
2948 defm FCVTPS : SIMDTwoScalarSD< 0, 1, 0b11010, "fcvtps">;
2949 defm FCVTPU : SIMDTwoScalarSD< 1, 1, 0b11010, "fcvtpu">;
2950 def FCVTXNv1i64 : SIMDInexactCvtTwoScalar<0b10110, "fcvtxn">;
2951 defm FCVTZS : SIMDTwoScalarSD< 0, 1, 0b11011, "fcvtzs">;
2952 defm FCVTZU : SIMDTwoScalarSD< 1, 1, 0b11011, "fcvtzu">;
2953 defm FRECPE : SIMDTwoScalarSD< 0, 1, 0b11101, "frecpe">;
2954 defm FRECPX : SIMDTwoScalarSD< 0, 1, 0b11111, "frecpx">;
2955 defm FRSQRTE : SIMDTwoScalarSD< 1, 1, 0b11101, "frsqrte">;
2956 defm NEG : SIMDTwoScalarD< 1, 0b01011, "neg",
2957 UnOpFrag<(sub immAllZerosV, node:$LHS)> >;
2958 defm SCVTF : SIMDTwoScalarCVTSD< 0, 0, 0b11101, "scvtf", ARM64sitof>;
2959 defm SQABS : SIMDTwoScalarBHSD< 0, 0b00111, "sqabs", int_arm64_neon_sqabs>;
2960 defm SQNEG : SIMDTwoScalarBHSD< 1, 0b00111, "sqneg", int_arm64_neon_sqneg>;
2961 defm SQXTN : SIMDTwoScalarMixedBHS< 0, 0b10100, "sqxtn", int_arm64_neon_scalar_sqxtn>;
2962 defm SQXTUN : SIMDTwoScalarMixedBHS< 1, 0b10010, "sqxtun", int_arm64_neon_scalar_sqxtun>;
2963 defm SUQADD : SIMDTwoScalarBHSDTied< 0, 0b00011, "suqadd",
2964 int_arm64_neon_suqadd>;
2965 defm UCVTF : SIMDTwoScalarCVTSD< 1, 0, 0b11101, "ucvtf", ARM64uitof>;
2966 defm UQXTN : SIMDTwoScalarMixedBHS<1, 0b10100, "uqxtn", int_arm64_neon_scalar_uqxtn>;
2967 defm USQADD : SIMDTwoScalarBHSDTied< 1, 0b00011, "usqadd",
2968 int_arm64_neon_usqadd>;
2970 def : Pat<(ARM64neg (v1i64 V64:$Rn)), (NEGv1i64 V64:$Rn)>;
2972 def : Pat<(v1i64 (int_arm64_neon_fcvtas (v1f64 FPR64:$Rn))),
2973 (FCVTASv1i64 FPR64:$Rn)>;
2974 def : Pat<(v1i64 (int_arm64_neon_fcvtau (v1f64 FPR64:$Rn))),
2975 (FCVTAUv1i64 FPR64:$Rn)>;
2976 def : Pat<(v1i64 (int_arm64_neon_fcvtms (v1f64 FPR64:$Rn))),
2977 (FCVTMSv1i64 FPR64:$Rn)>;
2978 def : Pat<(v1i64 (int_arm64_neon_fcvtmu (v1f64 FPR64:$Rn))),
2979 (FCVTMUv1i64 FPR64:$Rn)>;
2980 def : Pat<(v1i64 (int_arm64_neon_fcvtns (v1f64 FPR64:$Rn))),
2981 (FCVTNSv1i64 FPR64:$Rn)>;
2982 def : Pat<(v1i64 (int_arm64_neon_fcvtnu (v1f64 FPR64:$Rn))),
2983 (FCVTNUv1i64 FPR64:$Rn)>;
2984 def : Pat<(v1i64 (int_arm64_neon_fcvtps (v1f64 FPR64:$Rn))),
2985 (FCVTPSv1i64 FPR64:$Rn)>;
2986 def : Pat<(v1i64 (int_arm64_neon_fcvtpu (v1f64 FPR64:$Rn))),
2987 (FCVTPUv1i64 FPR64:$Rn)>;
2989 def : Pat<(f32 (int_arm64_neon_frecpe (f32 FPR32:$Rn))),
2990 (FRECPEv1i32 FPR32:$Rn)>;
2991 def : Pat<(f64 (int_arm64_neon_frecpe (f64 FPR64:$Rn))),
2992 (FRECPEv1i64 FPR64:$Rn)>;
2993 def : Pat<(v1f64 (int_arm64_neon_frecpe (v1f64 FPR64:$Rn))),
2994 (FRECPEv1i64 FPR64:$Rn)>;
2996 def : Pat<(f32 (int_arm64_neon_frecpx (f32 FPR32:$Rn))),
2997 (FRECPXv1i32 FPR32:$Rn)>;
2998 def : Pat<(f64 (int_arm64_neon_frecpx (f64 FPR64:$Rn))),
2999 (FRECPXv1i64 FPR64:$Rn)>;
3001 def : Pat<(f32 (int_arm64_neon_frsqrte (f32 FPR32:$Rn))),
3002 (FRSQRTEv1i32 FPR32:$Rn)>;
3003 def : Pat<(f64 (int_arm64_neon_frsqrte (f64 FPR64:$Rn))),
3004 (FRSQRTEv1i64 FPR64:$Rn)>;
3005 def : Pat<(v1f64 (int_arm64_neon_frsqrte (v1f64 FPR64:$Rn))),
3006 (FRSQRTEv1i64 FPR64:$Rn)>;
3008 // If an integer is about to be converted to a floating point value,
3009 // just load it on the floating point unit.
3010 // Here are the patterns for 8 and 16-bits to float.
3012 multiclass UIntToFPROLoadPat<ValueType DstTy, ValueType SrcTy,
3013 SDPatternOperator loadop, Instruction UCVTF,
3014 ROAddrMode ro, Instruction LDRW, Instruction LDRX,
3016 def : Pat<(DstTy (uint_to_fp (SrcTy
3017 (loadop (ro.Wpat GPR64sp:$Rn, GPR32:$Rm,
3018 ro.Wext:$extend))))),
3019 (UCVTF (INSERT_SUBREG (DstTy (IMPLICIT_DEF)),
3020 (LDRW GPR64sp:$Rn, GPR32:$Rm, ro.Wext:$extend),
3023 def : Pat<(DstTy (uint_to_fp (SrcTy
3024 (loadop (ro.Xpat GPR64sp:$Rn, GPR64:$Rm,
3025 ro.Wext:$extend))))),
3026 (UCVTF (INSERT_SUBREG (DstTy (IMPLICIT_DEF)),
3027 (LDRX GPR64sp:$Rn, GPR64:$Rm, ro.Xext:$extend),
3031 defm : UIntToFPROLoadPat<f32, i32, zextloadi8,
3032 UCVTFv1i32, ro8, LDRBroW, LDRBroX, bsub>;
3033 def : Pat <(f32 (uint_to_fp (i32
3034 (zextloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))))),
3035 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
3036 (LDRBui GPR64sp:$Rn, uimm12s1:$offset), bsub))>;
3037 def : Pat <(f32 (uint_to_fp (i32
3038 (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))))),
3039 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
3040 (LDURBi GPR64sp:$Rn, simm9:$offset), bsub))>;
3041 // 16-bits -> float.
3042 defm : UIntToFPROLoadPat<f32, i32, zextloadi16,
3043 UCVTFv1i32, ro16, LDRHroW, LDRHroX, hsub>;
3044 def : Pat <(f32 (uint_to_fp (i32
3045 (zextloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))))),
3046 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
3047 (LDRHui GPR64sp:$Rn, uimm12s2:$offset), hsub))>;
3048 def : Pat <(f32 (uint_to_fp (i32
3049 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))))),
3050 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
3051 (LDURHi GPR64sp:$Rn, simm9:$offset), hsub))>;
3052 // 32-bits are handled in target specific dag combine:
3053 // performIntToFpCombine.
3054 // 64-bits integer to 32-bits floating point, not possible with
3055 // UCVTF on floating point registers (both source and destination
3056 // must have the same size).
3058 // Here are the patterns for 8, 16, 32, and 64-bits to double.
3059 // 8-bits -> double.
3060 defm : UIntToFPROLoadPat<f64, i32, zextloadi8,
3061 UCVTFv1i64, ro8, LDRBroW, LDRBroX, bsub>;
3062 def : Pat <(f64 (uint_to_fp (i32
3063 (zextloadi8 (am_indexed8 GPR64sp:$Rn, uimm12s1:$offset))))),
3064 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3065 (LDRBui GPR64sp:$Rn, uimm12s1:$offset), bsub))>;
3066 def : Pat <(f64 (uint_to_fp (i32
3067 (zextloadi8 (am_unscaled8 GPR64sp:$Rn, simm9:$offset))))),
3068 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3069 (LDURBi GPR64sp:$Rn, simm9:$offset), bsub))>;
3070 // 16-bits -> double.
3071 defm : UIntToFPROLoadPat<f64, i32, zextloadi16,
3072 UCVTFv1i64, ro16, LDRHroW, LDRHroX, hsub>;
3073 def : Pat <(f64 (uint_to_fp (i32
3074 (zextloadi16 (am_indexed16 GPR64sp:$Rn, uimm12s2:$offset))))),
3075 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3076 (LDRHui GPR64sp:$Rn, uimm12s2:$offset), hsub))>;
3077 def : Pat <(f64 (uint_to_fp (i32
3078 (zextloadi16 (am_unscaled16 GPR64sp:$Rn, simm9:$offset))))),
3079 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3080 (LDURHi GPR64sp:$Rn, simm9:$offset), hsub))>;
3081 // 32-bits -> double.
3082 defm : UIntToFPROLoadPat<f64, i32, load,
3083 UCVTFv1i64, ro32, LDRSroW, LDRSroX, ssub>;
3084 def : Pat <(f64 (uint_to_fp (i32
3085 (load (am_indexed32 GPR64sp:$Rn, uimm12s4:$offset))))),
3086 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3087 (LDRSui GPR64sp:$Rn, uimm12s4:$offset), ssub))>;
3088 def : Pat <(f64 (uint_to_fp (i32
3089 (load (am_unscaled32 GPR64sp:$Rn, simm9:$offset))))),
3090 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3091 (LDURSi GPR64sp:$Rn, simm9:$offset), ssub))>;
3092 // 64-bits -> double are handled in target specific dag combine:
3093 // performIntToFpCombine.
3095 //===----------------------------------------------------------------------===//
3096 // Advanced SIMD three different-sized vector instructions.
3097 //===----------------------------------------------------------------------===//
3099 defm ADDHN : SIMDNarrowThreeVectorBHS<0,0b0100,"addhn", int_arm64_neon_addhn>;
3100 defm SUBHN : SIMDNarrowThreeVectorBHS<0,0b0110,"subhn", int_arm64_neon_subhn>;
3101 defm RADDHN : SIMDNarrowThreeVectorBHS<1,0b0100,"raddhn",int_arm64_neon_raddhn>;
3102 defm RSUBHN : SIMDNarrowThreeVectorBHS<1,0b0110,"rsubhn",int_arm64_neon_rsubhn>;
3103 defm PMULL : SIMDDifferentThreeVectorBD<0,0b1110,"pmull",int_arm64_neon_pmull>;
3104 defm SABAL : SIMDLongThreeVectorTiedBHSabal<0,0b0101,"sabal",
3105 int_arm64_neon_sabd>;
3106 defm SABDL : SIMDLongThreeVectorBHSabdl<0, 0b0111, "sabdl",
3107 int_arm64_neon_sabd>;
3108 defm SADDL : SIMDLongThreeVectorBHS< 0, 0b0000, "saddl",
3109 BinOpFrag<(add (sext node:$LHS), (sext node:$RHS))>>;
3110 defm SADDW : SIMDWideThreeVectorBHS< 0, 0b0001, "saddw",
3111 BinOpFrag<(add node:$LHS, (sext node:$RHS))>>;
3112 defm SMLAL : SIMDLongThreeVectorTiedBHS<0, 0b1000, "smlal",
3113 TriOpFrag<(add node:$LHS, (int_arm64_neon_smull node:$MHS, node:$RHS))>>;
3114 defm SMLSL : SIMDLongThreeVectorTiedBHS<0, 0b1010, "smlsl",
3115 TriOpFrag<(sub node:$LHS, (int_arm64_neon_smull node:$MHS, node:$RHS))>>;
3116 defm SMULL : SIMDLongThreeVectorBHS<0, 0b1100, "smull", int_arm64_neon_smull>;
3117 defm SQDMLAL : SIMDLongThreeVectorSQDMLXTiedHS<0, 0b1001, "sqdmlal",
3118 int_arm64_neon_sqadd>;
3119 defm SQDMLSL : SIMDLongThreeVectorSQDMLXTiedHS<0, 0b1011, "sqdmlsl",
3120 int_arm64_neon_sqsub>;
3121 defm SQDMULL : SIMDLongThreeVectorHS<0, 0b1101, "sqdmull",
3122 int_arm64_neon_sqdmull>;
3123 defm SSUBL : SIMDLongThreeVectorBHS<0, 0b0010, "ssubl",
3124 BinOpFrag<(sub (sext node:$LHS), (sext node:$RHS))>>;
3125 defm SSUBW : SIMDWideThreeVectorBHS<0, 0b0011, "ssubw",
3126 BinOpFrag<(sub node:$LHS, (sext node:$RHS))>>;
3127 defm UABAL : SIMDLongThreeVectorTiedBHSabal<1, 0b0101, "uabal",
3128 int_arm64_neon_uabd>;
3129 defm UABDL : SIMDLongThreeVectorBHSabdl<1, 0b0111, "uabdl",
3130 int_arm64_neon_uabd>;
3131 defm UADDL : SIMDLongThreeVectorBHS<1, 0b0000, "uaddl",
3132 BinOpFrag<(add (zext node:$LHS), (zext node:$RHS))>>;
3133 defm UADDW : SIMDWideThreeVectorBHS<1, 0b0001, "uaddw",
3134 BinOpFrag<(add node:$LHS, (zext node:$RHS))>>;
3135 defm UMLAL : SIMDLongThreeVectorTiedBHS<1, 0b1000, "umlal",
3136 TriOpFrag<(add node:$LHS, (int_arm64_neon_umull node:$MHS, node:$RHS))>>;
3137 defm UMLSL : SIMDLongThreeVectorTiedBHS<1, 0b1010, "umlsl",
3138 TriOpFrag<(sub node:$LHS, (int_arm64_neon_umull node:$MHS, node:$RHS))>>;
3139 defm UMULL : SIMDLongThreeVectorBHS<1, 0b1100, "umull", int_arm64_neon_umull>;
3140 defm USUBL : SIMDLongThreeVectorBHS<1, 0b0010, "usubl",
3141 BinOpFrag<(sub (zext node:$LHS), (zext node:$RHS))>>;
3142 defm USUBW : SIMDWideThreeVectorBHS< 1, 0b0011, "usubw",
3143 BinOpFrag<(sub node:$LHS, (zext node:$RHS))>>;
3145 // Patterns for 64-bit pmull
3146 def : Pat<(int_arm64_neon_pmull64 V64:$Rn, V64:$Rm),
3147 (PMULLv1i64 V64:$Rn, V64:$Rm)>;
3148 def : Pat<(int_arm64_neon_pmull64 (vector_extract (v2i64 V128:$Rn), (i64 1)),
3149 (vector_extract (v2i64 V128:$Rm), (i64 1))),
3150 (PMULLv2i64 V128:$Rn, V128:$Rm)>;
3152 // CodeGen patterns for addhn and subhn instructions, which can actually be
3153 // written in LLVM IR without too much difficulty.
3156 def : Pat<(v8i8 (trunc (v8i16 (ARM64vlshr (add V128:$Rn, V128:$Rm), (i32 8))))),
3157 (ADDHNv8i16_v8i8 V128:$Rn, V128:$Rm)>;
3158 def : Pat<(v4i16 (trunc (v4i32 (ARM64vlshr (add V128:$Rn, V128:$Rm),
3160 (ADDHNv4i32_v4i16 V128:$Rn, V128:$Rm)>;
3161 def : Pat<(v2i32 (trunc (v2i64 (ARM64vlshr (add V128:$Rn, V128:$Rm),
3163 (ADDHNv2i64_v2i32 V128:$Rn, V128:$Rm)>;
3164 def : Pat<(concat_vectors (v8i8 V64:$Rd),
3165 (trunc (v8i16 (ARM64vlshr (add V128:$Rn, V128:$Rm),
3167 (ADDHNv8i16_v16i8 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3168 V128:$Rn, V128:$Rm)>;
3169 def : Pat<(concat_vectors (v4i16 V64:$Rd),
3170 (trunc (v4i32 (ARM64vlshr (add V128:$Rn, V128:$Rm),
3172 (ADDHNv4i32_v8i16 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3173 V128:$Rn, V128:$Rm)>;
3174 def : Pat<(concat_vectors (v2i32 V64:$Rd),
3175 (trunc (v2i64 (ARM64vlshr (add V128:$Rn, V128:$Rm),
3177 (ADDHNv2i64_v4i32 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3178 V128:$Rn, V128:$Rm)>;
3181 def : Pat<(v8i8 (trunc (v8i16 (ARM64vlshr (sub V128:$Rn, V128:$Rm), (i32 8))))),
3182 (SUBHNv8i16_v8i8 V128:$Rn, V128:$Rm)>;
3183 def : Pat<(v4i16 (trunc (v4i32 (ARM64vlshr (sub V128:$Rn, V128:$Rm),
3185 (SUBHNv4i32_v4i16 V128:$Rn, V128:$Rm)>;
3186 def : Pat<(v2i32 (trunc (v2i64 (ARM64vlshr (sub V128:$Rn, V128:$Rm),
3188 (SUBHNv2i64_v2i32 V128:$Rn, V128:$Rm)>;
3189 def : Pat<(concat_vectors (v8i8 V64:$Rd),
3190 (trunc (v8i16 (ARM64vlshr (sub V128:$Rn, V128:$Rm),
3192 (SUBHNv8i16_v16i8 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3193 V128:$Rn, V128:$Rm)>;
3194 def : Pat<(concat_vectors (v4i16 V64:$Rd),
3195 (trunc (v4i32 (ARM64vlshr (sub V128:$Rn, V128:$Rm),
3197 (SUBHNv4i32_v8i16 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3198 V128:$Rn, V128:$Rm)>;
3199 def : Pat<(concat_vectors (v2i32 V64:$Rd),
3200 (trunc (v2i64 (ARM64vlshr (sub V128:$Rn, V128:$Rm),
3202 (SUBHNv2i64_v4i32 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
3203 V128:$Rn, V128:$Rm)>;
3205 //----------------------------------------------------------------------------
3206 // AdvSIMD bitwise extract from vector instruction.
3207 //----------------------------------------------------------------------------
3209 defm EXT : SIMDBitwiseExtract<"ext">;
3211 def : Pat<(v4i16 (ARM64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
3212 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
3213 def : Pat<(v8i16 (ARM64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3214 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3215 def : Pat<(v2i32 (ARM64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
3216 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
3217 def : Pat<(v2f32 (ARM64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
3218 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
3219 def : Pat<(v4i32 (ARM64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3220 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3221 def : Pat<(v4f32 (ARM64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3222 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3223 def : Pat<(v2i64 (ARM64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3224 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3225 def : Pat<(v2f64 (ARM64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3226 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3228 // We use EXT to handle extract_subvector to copy the upper 64-bits of a
3230 def : Pat<(v8i8 (extract_subvector V128:$Rn, (i64 8))),
3231 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3232 def : Pat<(v4i16 (extract_subvector V128:$Rn, (i64 4))),
3233 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3234 def : Pat<(v2i32 (extract_subvector V128:$Rn, (i64 2))),
3235 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3236 def : Pat<(v1i64 (extract_subvector V128:$Rn, (i64 1))),
3237 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3238 def : Pat<(v2f32 (extract_subvector V128:$Rn, (i64 2))),
3239 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3240 def : Pat<(v1f64 (extract_subvector V128:$Rn, (i64 1))),
3241 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3244 //----------------------------------------------------------------------------
3245 // AdvSIMD zip vector
3246 //----------------------------------------------------------------------------
3248 defm TRN1 : SIMDZipVector<0b010, "trn1", ARM64trn1>;
3249 defm TRN2 : SIMDZipVector<0b110, "trn2", ARM64trn2>;
3250 defm UZP1 : SIMDZipVector<0b001, "uzp1", ARM64uzp1>;
3251 defm UZP2 : SIMDZipVector<0b101, "uzp2", ARM64uzp2>;
3252 defm ZIP1 : SIMDZipVector<0b011, "zip1", ARM64zip1>;
3253 defm ZIP2 : SIMDZipVector<0b111, "zip2", ARM64zip2>;
3255 //----------------------------------------------------------------------------
3256 // AdvSIMD TBL/TBX instructions
3257 //----------------------------------------------------------------------------
3259 defm TBL : SIMDTableLookup< 0, "tbl">;
3260 defm TBX : SIMDTableLookupTied<1, "tbx">;
3262 def : Pat<(v8i8 (int_arm64_neon_tbl1 (v16i8 VecListOne128:$Rn), (v8i8 V64:$Ri))),
3263 (TBLv8i8One VecListOne128:$Rn, V64:$Ri)>;
3264 def : Pat<(v16i8 (int_arm64_neon_tbl1 (v16i8 V128:$Ri), (v16i8 V128:$Rn))),
3265 (TBLv16i8One V128:$Ri, V128:$Rn)>;
3267 def : Pat<(v8i8 (int_arm64_neon_tbx1 (v8i8 V64:$Rd),
3268 (v16i8 VecListOne128:$Rn), (v8i8 V64:$Ri))),
3269 (TBXv8i8One V64:$Rd, VecListOne128:$Rn, V64:$Ri)>;
3270 def : Pat<(v16i8 (int_arm64_neon_tbx1 (v16i8 V128:$Rd),
3271 (v16i8 V128:$Ri), (v16i8 V128:$Rn))),
3272 (TBXv16i8One V128:$Rd, V128:$Ri, V128:$Rn)>;
3275 //----------------------------------------------------------------------------
3276 // AdvSIMD scalar CPY instruction
3277 //----------------------------------------------------------------------------
3279 defm CPY : SIMDScalarCPY<"cpy">;
3281 //----------------------------------------------------------------------------
3282 // AdvSIMD scalar pairwise instructions
3283 //----------------------------------------------------------------------------
3285 defm ADDP : SIMDPairwiseScalarD<0, 0b11011, "addp">;
3286 defm FADDP : SIMDPairwiseScalarSD<1, 0, 0b01101, "faddp">;
3287 defm FMAXNMP : SIMDPairwiseScalarSD<1, 0, 0b01100, "fmaxnmp">;
3288 defm FMAXP : SIMDPairwiseScalarSD<1, 0, 0b01111, "fmaxp">;
3289 defm FMINNMP : SIMDPairwiseScalarSD<1, 1, 0b01100, "fminnmp">;
3290 defm FMINP : SIMDPairwiseScalarSD<1, 1, 0b01111, "fminp">;
3291 def : Pat<(i64 (int_arm64_neon_saddv (v2i64 V128:$Rn))),
3292 (ADDPv2i64p V128:$Rn)>;
3293 def : Pat<(i64 (int_arm64_neon_uaddv (v2i64 V128:$Rn))),
3294 (ADDPv2i64p V128:$Rn)>;
3295 def : Pat<(f32 (int_arm64_neon_faddv (v2f32 V64:$Rn))),
3296 (FADDPv2i32p V64:$Rn)>;
3297 def : Pat<(f32 (int_arm64_neon_faddv (v4f32 V128:$Rn))),
3298 (FADDPv2i32p (EXTRACT_SUBREG (FADDPv4f32 V128:$Rn, V128:$Rn), dsub))>;
3299 def : Pat<(f64 (int_arm64_neon_faddv (v2f64 V128:$Rn))),
3300 (FADDPv2i64p V128:$Rn)>;
3301 def : Pat<(f32 (int_arm64_neon_fmaxnmv (v2f32 V64:$Rn))),
3302 (FMAXNMPv2i32p V64:$Rn)>;
3303 def : Pat<(f64 (int_arm64_neon_fmaxnmv (v2f64 V128:$Rn))),
3304 (FMAXNMPv2i64p V128:$Rn)>;
3305 def : Pat<(f32 (int_arm64_neon_fmaxv (v2f32 V64:$Rn))),
3306 (FMAXPv2i32p V64:$Rn)>;
3307 def : Pat<(f64 (int_arm64_neon_fmaxv (v2f64 V128:$Rn))),
3308 (FMAXPv2i64p V128:$Rn)>;
3309 def : Pat<(f32 (int_arm64_neon_fminnmv (v2f32 V64:$Rn))),
3310 (FMINNMPv2i32p V64:$Rn)>;
3311 def : Pat<(f64 (int_arm64_neon_fminnmv (v2f64 V128:$Rn))),
3312 (FMINNMPv2i64p V128:$Rn)>;
3313 def : Pat<(f32 (int_arm64_neon_fminv (v2f32 V64:$Rn))),
3314 (FMINPv2i32p V64:$Rn)>;
3315 def : Pat<(f64 (int_arm64_neon_fminv (v2f64 V128:$Rn))),
3316 (FMINPv2i64p V128:$Rn)>;
3318 //----------------------------------------------------------------------------
3319 // AdvSIMD INS/DUP instructions
3320 //----------------------------------------------------------------------------
3322 def DUPv8i8gpr : SIMDDupFromMain<0, 0b00001, ".8b", v8i8, V64, GPR32>;
3323 def DUPv16i8gpr : SIMDDupFromMain<1, 0b00001, ".16b", v16i8, V128, GPR32>;
3324 def DUPv4i16gpr : SIMDDupFromMain<0, 0b00010, ".4h", v4i16, V64, GPR32>;
3325 def DUPv8i16gpr : SIMDDupFromMain<1, 0b00010, ".8h", v8i16, V128, GPR32>;
3326 def DUPv2i32gpr : SIMDDupFromMain<0, 0b00100, ".2s", v2i32, V64, GPR32>;
3327 def DUPv4i32gpr : SIMDDupFromMain<1, 0b00100, ".4s", v4i32, V128, GPR32>;
3328 def DUPv2i64gpr : SIMDDupFromMain<1, 0b01000, ".2d", v2i64, V128, GPR64>;
3330 def DUPv2i64lane : SIMDDup64FromElement;
3331 def DUPv2i32lane : SIMDDup32FromElement<0, ".2s", v2i32, V64>;
3332 def DUPv4i32lane : SIMDDup32FromElement<1, ".4s", v4i32, V128>;
3333 def DUPv4i16lane : SIMDDup16FromElement<0, ".4h", v4i16, V64>;
3334 def DUPv8i16lane : SIMDDup16FromElement<1, ".8h", v8i16, V128>;
3335 def DUPv8i8lane : SIMDDup8FromElement <0, ".8b", v8i8, V64>;
3336 def DUPv16i8lane : SIMDDup8FromElement <1, ".16b", v16i8, V128>;
3338 def : Pat<(v2f32 (ARM64dup (f32 FPR32:$Rn))),
3339 (v2f32 (DUPv2i32lane
3340 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rn, ssub),
3342 def : Pat<(v4f32 (ARM64dup (f32 FPR32:$Rn))),
3343 (v4f32 (DUPv4i32lane
3344 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rn, ssub),
3346 def : Pat<(v2f64 (ARM64dup (f64 FPR64:$Rn))),
3347 (v2f64 (DUPv2i64lane
3348 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$Rn, dsub),
3351 def : Pat<(v2f32 (ARM64duplane32 (v4f32 V128:$Rn), VectorIndexS:$imm)),
3352 (DUPv2i32lane V128:$Rn, VectorIndexS:$imm)>;
3353 def : Pat<(v4f32 (ARM64duplane32 (v4f32 V128:$Rn), VectorIndexS:$imm)),
3354 (DUPv4i32lane V128:$Rn, VectorIndexS:$imm)>;
3355 def : Pat<(v2f64 (ARM64duplane64 (v2f64 V128:$Rn), VectorIndexD:$imm)),
3356 (DUPv2i64lane V128:$Rn, VectorIndexD:$imm)>;
3358 // If there's an (ARM64dup (vector_extract ...) ...), we can use a duplane
3359 // instruction even if the types don't match: we just have to remap the lane
3360 // carefully. N.b. this trick only applies to truncations.
3361 def VecIndex_x2 : SDNodeXForm<imm, [{
3362 return CurDAG->getTargetConstant(2 * N->getZExtValue(), MVT::i64);
3364 def VecIndex_x4 : SDNodeXForm<imm, [{
3365 return CurDAG->getTargetConstant(4 * N->getZExtValue(), MVT::i64);
3367 def VecIndex_x8 : SDNodeXForm<imm, [{
3368 return CurDAG->getTargetConstant(8 * N->getZExtValue(), MVT::i64);
3371 multiclass DUPWithTruncPats<ValueType ResVT, ValueType Src64VT,
3372 ValueType Src128VT, ValueType ScalVT,
3373 Instruction DUP, SDNodeXForm IdxXFORM> {
3374 def : Pat<(ResVT (ARM64dup (ScalVT (vector_extract (Src128VT V128:$Rn),
3376 (DUP V128:$Rn, (IdxXFORM imm:$idx))>;
3378 def : Pat<(ResVT (ARM64dup (ScalVT (vector_extract (Src64VT V64:$Rn),
3380 (DUP (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), (IdxXFORM imm:$idx))>;
3383 defm : DUPWithTruncPats<v8i8, v4i16, v8i16, i32, DUPv8i8lane, VecIndex_x2>;
3384 defm : DUPWithTruncPats<v8i8, v2i32, v4i32, i32, DUPv8i8lane, VecIndex_x4>;
3385 defm : DUPWithTruncPats<v4i16, v2i32, v4i32, i32, DUPv4i16lane, VecIndex_x2>;
3387 defm : DUPWithTruncPats<v16i8, v4i16, v8i16, i32, DUPv16i8lane, VecIndex_x2>;
3388 defm : DUPWithTruncPats<v16i8, v2i32, v4i32, i32, DUPv16i8lane, VecIndex_x4>;
3389 defm : DUPWithTruncPats<v8i16, v2i32, v4i32, i32, DUPv8i16lane, VecIndex_x2>;
3391 multiclass DUPWithTrunci64Pats<ValueType ResVT, Instruction DUP,
3392 SDNodeXForm IdxXFORM> {
3393 def : Pat<(ResVT (ARM64dup (i32 (trunc (vector_extract (v2i64 V128:$Rn),
3395 (DUP V128:$Rn, (IdxXFORM imm:$idx))>;
3397 def : Pat<(ResVT (ARM64dup (i32 (trunc (vector_extract (v1i64 V64:$Rn),
3399 (DUP (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), (IdxXFORM imm:$idx))>;
3402 defm : DUPWithTrunci64Pats<v8i8, DUPv8i8lane, VecIndex_x8>;
3403 defm : DUPWithTrunci64Pats<v4i16, DUPv4i16lane, VecIndex_x4>;
3404 defm : DUPWithTrunci64Pats<v2i32, DUPv2i32lane, VecIndex_x2>;
3406 defm : DUPWithTrunci64Pats<v16i8, DUPv16i8lane, VecIndex_x8>;
3407 defm : DUPWithTrunci64Pats<v8i16, DUPv8i16lane, VecIndex_x4>;
3408 defm : DUPWithTrunci64Pats<v4i32, DUPv4i32lane, VecIndex_x2>;
3410 // SMOV and UMOV definitions, with some extra patterns for convenience
3414 def : Pat<(sext_inreg (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx), i8),
3415 (i32 (SMOVvi8to32 V128:$Rn, VectorIndexB:$idx))>;
3416 def : Pat<(sext_inreg (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx), i8),
3417 (i64 (SMOVvi8to64 V128:$Rn, VectorIndexB:$idx))>;
3418 def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
3419 (i32 (SMOVvi16to32 V128:$Rn, VectorIndexH:$idx))>;
3420 def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
3421 (i64 (SMOVvi16to64 V128:$Rn, VectorIndexH:$idx))>;
3422 def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
3423 (i32 (SMOVvi16to32 V128:$Rn, VectorIndexH:$idx))>;
3424 def : Pat<(sext (i32 (vector_extract (v4i32 V128:$Rn), VectorIndexS:$idx))),
3425 (i64 (SMOVvi32to64 V128:$Rn, VectorIndexS:$idx))>;
3427 // Extracting i8 or i16 elements will have the zero-extend transformed to
3428 // an 'and' mask by type legalization since neither i8 nor i16 are legal types
3429 // for ARM64. Match these patterns here since UMOV already zeroes out the high
3430 // bits of the destination register.
3431 def : Pat<(and (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx),
3433 (i32 (UMOVvi8 V128:$Rn, VectorIndexB:$idx))>;
3434 def : Pat<(and (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),
3436 (i32 (UMOVvi16 V128:$Rn, VectorIndexH:$idx))>;
3440 def : Pat<(v16i8 (scalar_to_vector GPR32:$Rn)),
3441 (SUBREG_TO_REG (i32 0),
3442 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3443 def : Pat<(v8i8 (scalar_to_vector GPR32:$Rn)),
3444 (SUBREG_TO_REG (i32 0),
3445 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3447 def : Pat<(v8i16 (scalar_to_vector GPR32:$Rn)),
3448 (SUBREG_TO_REG (i32 0),
3449 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3450 def : Pat<(v4i16 (scalar_to_vector GPR32:$Rn)),
3451 (SUBREG_TO_REG (i32 0),
3452 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3454 def : Pat<(v2i32 (scalar_to_vector (i32 FPR32:$Rn))),
3455 (v2i32 (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)),
3456 (i32 FPR32:$Rn), ssub))>;
3457 def : Pat<(v4i32 (scalar_to_vector (i32 FPR32:$Rn))),
3458 (v4i32 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
3459 (i32 FPR32:$Rn), ssub))>;
3460 def : Pat<(v2i64 (scalar_to_vector (i64 FPR64:$Rn))),
3461 (v2i64 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
3462 (i64 FPR64:$Rn), dsub))>;
3464 def : Pat<(v4f32 (scalar_to_vector (f32 FPR32:$Rn))),
3465 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rn, ssub)>;
3466 def : Pat<(v2f32 (scalar_to_vector (f32 FPR32:$Rn))),
3467 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), FPR32:$Rn, ssub)>;
3468 def : Pat<(v2f64 (scalar_to_vector (f64 FPR64:$Rn))),
3469 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$Rn, dsub)>;
3471 def : Pat<(v2f32 (vector_insert (v2f32 V64:$Rn),
3472 (f32 FPR32:$Rm), (i64 VectorIndexS:$imm))),
3475 (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), V64:$Rn, dsub)),
3477 (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rm, ssub)),
3480 def : Pat<(v4f32 (vector_insert (v4f32 V128:$Rn),
3481 (f32 FPR32:$Rm), (i64 VectorIndexS:$imm))),
3483 V128:$Rn, VectorIndexS:$imm,
3484 (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rm, ssub)),
3486 def : Pat<(v2f64 (vector_insert (v2f64 V128:$Rn),
3487 (f64 FPR64:$Rm), (i64 VectorIndexD:$imm))),
3489 V128:$Rn, VectorIndexD:$imm,
3490 (v2f64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$Rm, dsub)),
3493 // Copy an element at a constant index in one vector into a constant indexed
3494 // element of another.
3495 // FIXME refactor to a shared class/dev parameterized on vector type, vector
3496 // index type and INS extension
3497 def : Pat<(v16i8 (int_arm64_neon_vcopy_lane
3498 (v16i8 V128:$Vd), VectorIndexB:$idx, (v16i8 V128:$Vs),
3499 VectorIndexB:$idx2)),
3501 V128:$Vd, VectorIndexB:$idx, V128:$Vs, VectorIndexB:$idx2)
3503 def : Pat<(v8i16 (int_arm64_neon_vcopy_lane
3504 (v8i16 V128:$Vd), VectorIndexH:$idx, (v8i16 V128:$Vs),
3505 VectorIndexH:$idx2)),
3507 V128:$Vd, VectorIndexH:$idx, V128:$Vs, VectorIndexH:$idx2)
3509 def : Pat<(v4i32 (int_arm64_neon_vcopy_lane
3510 (v4i32 V128:$Vd), VectorIndexS:$idx, (v4i32 V128:$Vs),
3511 VectorIndexS:$idx2)),
3513 V128:$Vd, VectorIndexS:$idx, V128:$Vs, VectorIndexS:$idx2)
3515 def : Pat<(v2i64 (int_arm64_neon_vcopy_lane
3516 (v2i64 V128:$Vd), VectorIndexD:$idx, (v2i64 V128:$Vs),
3517 VectorIndexD:$idx2)),
3519 V128:$Vd, VectorIndexD:$idx, V128:$Vs, VectorIndexD:$idx2)
3522 multiclass Neon_INS_elt_pattern<ValueType VT128, ValueType VT64,
3523 ValueType VTScal, Instruction INS> {
3524 def : Pat<(VT128 (vector_insert V128:$src,
3525 (VTScal (vector_extract (VT128 V128:$Rn), imm:$Immn)),
3527 (INS V128:$src, imm:$Immd, V128:$Rn, imm:$Immn)>;
3529 def : Pat<(VT128 (vector_insert V128:$src,
3530 (VTScal (vector_extract (VT64 V64:$Rn), imm:$Immn)),
3532 (INS V128:$src, imm:$Immd,
3533 (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), imm:$Immn)>;
3535 def : Pat<(VT64 (vector_insert V64:$src,
3536 (VTScal (vector_extract (VT128 V128:$Rn), imm:$Immn)),
3538 (EXTRACT_SUBREG (INS (SUBREG_TO_REG (i64 0), V64:$src, dsub),
3539 imm:$Immd, V128:$Rn, imm:$Immn),
3542 def : Pat<(VT64 (vector_insert V64:$src,
3543 (VTScal (vector_extract (VT64 V64:$Rn), imm:$Immn)),
3546 (INS (SUBREG_TO_REG (i64 0), V64:$src, dsub), imm:$Immd,
3547 (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), imm:$Immn),
3551 defm : Neon_INS_elt_pattern<v4f32, v2f32, f32, INSvi32lane>;
3552 defm : Neon_INS_elt_pattern<v2f64, v1f64, f64, INSvi64lane>;
3553 defm : Neon_INS_elt_pattern<v16i8, v8i8, i32, INSvi8lane>;
3554 defm : Neon_INS_elt_pattern<v8i16, v4i16, i32, INSvi16lane>;
3555 defm : Neon_INS_elt_pattern<v4i32, v2i32, i32, INSvi32lane>;
3556 defm : Neon_INS_elt_pattern<v2i64, v1i64, i64, INSvi32lane>;
3559 // Floating point vector extractions are codegen'd as either a sequence of
3560 // subregister extractions, possibly fed by an INS if the lane number is
3561 // anything other than zero.
3562 def : Pat<(vector_extract (v2f64 V128:$Rn), 0),
3563 (f64 (EXTRACT_SUBREG V128:$Rn, dsub))>;
3564 def : Pat<(vector_extract (v4f32 V128:$Rn), 0),
3565 (f32 (EXTRACT_SUBREG V128:$Rn, ssub))>;
3566 def : Pat<(vector_extract (v2f64 V128:$Rn), VectorIndexD:$idx),
3567 (f64 (EXTRACT_SUBREG
3568 (INSvi64lane (v2f64 (IMPLICIT_DEF)), 0,
3569 V128:$Rn, VectorIndexD:$idx),
3571 def : Pat<(vector_extract (v4f32 V128:$Rn), VectorIndexS:$idx),
3572 (f32 (EXTRACT_SUBREG
3573 (INSvi32lane (v4f32 (IMPLICIT_DEF)), 0,
3574 V128:$Rn, VectorIndexS:$idx),
3577 // All concat_vectors operations are canonicalised to act on i64 vectors for
3578 // ARM64. In the general case we need an instruction, which had just as well be
3580 class ConcatPat<ValueType DstTy, ValueType SrcTy>
3581 : Pat<(DstTy (concat_vectors (SrcTy V64:$Rd), V64:$Rn)),
3582 (INSvi64lane (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), 1,
3583 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rn, dsub), 0)>;
3585 def : ConcatPat<v2i64, v1i64>;
3586 def : ConcatPat<v2f64, v1f64>;
3587 def : ConcatPat<v4i32, v2i32>;
3588 def : ConcatPat<v4f32, v2f32>;
3589 def : ConcatPat<v8i16, v4i16>;
3590 def : ConcatPat<v16i8, v8i8>;
3592 // If the high lanes are undef, though, we can just ignore them:
3593 class ConcatUndefPat<ValueType DstTy, ValueType SrcTy>
3594 : Pat<(DstTy (concat_vectors (SrcTy V64:$Rn), undef)),
3595 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rn, dsub)>;
3597 def : ConcatUndefPat<v2i64, v1i64>;
3598 def : ConcatUndefPat<v2f64, v1f64>;
3599 def : ConcatUndefPat<v4i32, v2i32>;
3600 def : ConcatUndefPat<v4f32, v2f32>;
3601 def : ConcatUndefPat<v8i16, v4i16>;
3602 def : ConcatUndefPat<v16i8, v8i8>;
3604 //----------------------------------------------------------------------------
3605 // AdvSIMD across lanes instructions
3606 //----------------------------------------------------------------------------
3608 defm ADDV : SIMDAcrossLanesBHS<0, 0b11011, "addv">;
3609 defm SMAXV : SIMDAcrossLanesBHS<0, 0b01010, "smaxv">;
3610 defm SMINV : SIMDAcrossLanesBHS<0, 0b11010, "sminv">;
3611 defm UMAXV : SIMDAcrossLanesBHS<1, 0b01010, "umaxv">;
3612 defm UMINV : SIMDAcrossLanesBHS<1, 0b11010, "uminv">;
3613 defm SADDLV : SIMDAcrossLanesHSD<0, 0b00011, "saddlv">;
3614 defm UADDLV : SIMDAcrossLanesHSD<1, 0b00011, "uaddlv">;
3615 defm FMAXNMV : SIMDAcrossLanesS<0b01100, 0, "fmaxnmv", int_arm64_neon_fmaxnmv>;
3616 defm FMAXV : SIMDAcrossLanesS<0b01111, 0, "fmaxv", int_arm64_neon_fmaxv>;
3617 defm FMINNMV : SIMDAcrossLanesS<0b01100, 1, "fminnmv", int_arm64_neon_fminnmv>;
3618 defm FMINV : SIMDAcrossLanesS<0b01111, 1, "fminv", int_arm64_neon_fminv>;
3620 multiclass SIMDAcrossLanesSignedIntrinsic<string baseOpc, Intrinsic intOp> {
3621 // If there is a sign extension after this intrinsic, consume it as smov already
3623 def : Pat<(i32 (sext_inreg (i32 (intOp (v8i8 V64:$Rn))), i8)),
3625 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3626 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3628 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
3630 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3631 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3633 // If there is a sign extension after this intrinsic, consume it as smov already
3635 def : Pat<(i32 (sext_inreg (i32 (intOp (v16i8 V128:$Rn))), i8)),
3637 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3638 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3640 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
3642 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3643 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3645 // If there is a sign extension after this intrinsic, consume it as smov already
3647 def : Pat<(i32 (sext_inreg (i32 (intOp (v4i16 V64:$Rn))), i16)),
3649 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3650 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3652 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
3654 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3655 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3657 // If there is a sign extension after this intrinsic, consume it as smov already
3659 def : Pat<(i32 (sext_inreg (i32 (intOp (v8i16 V128:$Rn))), i16)),
3661 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3662 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3664 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
3666 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3667 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3670 def : Pat<(i32 (intOp (v4i32 V128:$Rn))),
3671 (i32 (EXTRACT_SUBREG
3672 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3673 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), ssub),
3677 multiclass SIMDAcrossLanesUnsignedIntrinsic<string baseOpc, Intrinsic intOp> {
3678 // If there is a masking operation keeping only what has been actually
3679 // generated, consume it.
3680 def : Pat<(i32 (and (i32 (intOp (v8i8 V64:$Rn))), maski8_or_more)),
3681 (i32 (EXTRACT_SUBREG
3682 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3683 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3685 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
3686 (i32 (EXTRACT_SUBREG
3687 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3688 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3690 // If there is a masking operation keeping only what has been actually
3691 // generated, consume it.
3692 def : Pat<(i32 (and (i32 (intOp (v16i8 V128:$Rn))), maski8_or_more)),
3693 (i32 (EXTRACT_SUBREG
3694 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3695 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3697 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
3698 (i32 (EXTRACT_SUBREG
3699 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3700 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3703 // If there is a masking operation keeping only what has been actually
3704 // generated, consume it.
3705 def : Pat<(i32 (and (i32 (intOp (v4i16 V64:$Rn))), maski16_or_more)),
3706 (i32 (EXTRACT_SUBREG
3707 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3708 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3710 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
3711 (i32 (EXTRACT_SUBREG
3712 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3713 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3715 // If there is a masking operation keeping only what has been actually
3716 // generated, consume it.
3717 def : Pat<(i32 (and (i32 (intOp (v8i16 V128:$Rn))), maski16_or_more)),
3718 (i32 (EXTRACT_SUBREG
3719 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3720 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3722 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
3723 (i32 (EXTRACT_SUBREG
3724 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3725 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3728 def : Pat<(i32 (intOp (v4i32 V128:$Rn))),
3729 (i32 (EXTRACT_SUBREG
3730 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3731 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), ssub),
3736 multiclass SIMDAcrossLanesSignedLongIntrinsic<string baseOpc, Intrinsic intOp> {
3737 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
3739 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3740 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), hsub),
3742 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
3744 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3745 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), hsub),
3748 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
3749 (i32 (EXTRACT_SUBREG
3750 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3751 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), ssub),
3753 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
3754 (i32 (EXTRACT_SUBREG
3755 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3756 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), ssub),
3759 def : Pat<(i64 (intOp (v4i32 V128:$Rn))),
3760 (i64 (EXTRACT_SUBREG
3761 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3762 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), dsub),
3766 multiclass SIMDAcrossLanesUnsignedLongIntrinsic<string baseOpc,
3768 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
3769 (i32 (EXTRACT_SUBREG
3770 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3771 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), hsub),
3773 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
3774 (i32 (EXTRACT_SUBREG
3775 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3776 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), hsub),
3779 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
3780 (i32 (EXTRACT_SUBREG
3781 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3782 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), ssub),
3784 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
3785 (i32 (EXTRACT_SUBREG
3786 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3787 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), ssub),
3790 def : Pat<(i64 (intOp (v4i32 V128:$Rn))),
3791 (i64 (EXTRACT_SUBREG
3792 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3793 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), dsub),
3797 defm : SIMDAcrossLanesSignedIntrinsic<"ADDV", int_arm64_neon_saddv>;
3798 // vaddv_[su]32 is special; -> ADDP Vd.2S,Vn.2S,Vm.2S; return Vd.s[0];Vn==Vm
3799 def : Pat<(i32 (int_arm64_neon_saddv (v2i32 V64:$Rn))),
3800 (EXTRACT_SUBREG (ADDPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3802 defm : SIMDAcrossLanesUnsignedIntrinsic<"ADDV", int_arm64_neon_uaddv>;
3803 // vaddv_[su]32 is special; -> ADDP Vd.2S,Vn.2S,Vm.2S; return Vd.s[0];Vn==Vm
3804 def : Pat<(i32 (int_arm64_neon_uaddv (v2i32 V64:$Rn))),
3805 (EXTRACT_SUBREG (ADDPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3807 defm : SIMDAcrossLanesSignedIntrinsic<"SMAXV", int_arm64_neon_smaxv>;
3808 def : Pat<(i32 (int_arm64_neon_smaxv (v2i32 V64:$Rn))),
3809 (EXTRACT_SUBREG (SMAXPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3811 defm : SIMDAcrossLanesSignedIntrinsic<"SMINV", int_arm64_neon_sminv>;
3812 def : Pat<(i32 (int_arm64_neon_sminv (v2i32 V64:$Rn))),
3813 (EXTRACT_SUBREG (SMINPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3815 defm : SIMDAcrossLanesUnsignedIntrinsic<"UMAXV", int_arm64_neon_umaxv>;
3816 def : Pat<(i32 (int_arm64_neon_umaxv (v2i32 V64:$Rn))),
3817 (EXTRACT_SUBREG (UMAXPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3819 defm : SIMDAcrossLanesUnsignedIntrinsic<"UMINV", int_arm64_neon_uminv>;
3820 def : Pat<(i32 (int_arm64_neon_uminv (v2i32 V64:$Rn))),
3821 (EXTRACT_SUBREG (UMINPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3823 defm : SIMDAcrossLanesSignedLongIntrinsic<"SADDLV", int_arm64_neon_saddlv>;
3824 defm : SIMDAcrossLanesUnsignedLongIntrinsic<"UADDLV", int_arm64_neon_uaddlv>;
3826 // The vaddlv_s32 intrinsic gets mapped to SADDLP.
3827 def : Pat<(i64 (int_arm64_neon_saddlv (v2i32 V64:$Rn))),
3828 (i64 (EXTRACT_SUBREG
3829 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3830 (SADDLPv2i32_v1i64 V64:$Rn), dsub),
3832 // The vaddlv_u32 intrinsic gets mapped to UADDLP.
3833 def : Pat<(i64 (int_arm64_neon_uaddlv (v2i32 V64:$Rn))),
3834 (i64 (EXTRACT_SUBREG
3835 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3836 (UADDLPv2i32_v1i64 V64:$Rn), dsub),
3839 //------------------------------------------------------------------------------
3840 // AdvSIMD modified immediate instructions
3841 //------------------------------------------------------------------------------
3844 defm BIC : SIMDModifiedImmVectorShiftTied<1, 0b11, 0b01, "bic", ARM64bici>;
3846 defm ORR : SIMDModifiedImmVectorShiftTied<0, 0b11, 0b01, "orr", ARM64orri>;
3848 def : InstAlias<"bic $Vd.4h, $imm", (BICv4i16 V64:$Vd, imm0_255:$imm, 0)>;
3849 def : InstAlias<"bic $Vd.8h, $imm", (BICv8i16 V128:$Vd, imm0_255:$imm, 0)>;
3850 def : InstAlias<"bic $Vd.2s, $imm", (BICv2i32 V64:$Vd, imm0_255:$imm, 0)>;
3851 def : InstAlias<"bic $Vd.4s, $imm", (BICv4i32 V128:$Vd, imm0_255:$imm, 0)>;
3853 def : InstAlias<"bic.4h $Vd, $imm", (BICv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
3854 def : InstAlias<"bic.8h $Vd, $imm", (BICv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
3855 def : InstAlias<"bic.2s $Vd, $imm", (BICv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
3856 def : InstAlias<"bic.4s $Vd, $imm", (BICv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
3858 def : InstAlias<"orr $Vd.4h, $imm", (ORRv4i16 V64:$Vd, imm0_255:$imm, 0)>;
3859 def : InstAlias<"orr $Vd.8h, $imm", (ORRv8i16 V128:$Vd, imm0_255:$imm, 0)>;
3860 def : InstAlias<"orr $Vd.2s, $imm", (ORRv2i32 V64:$Vd, imm0_255:$imm, 0)>;
3861 def : InstAlias<"orr $Vd.4s, $imm", (ORRv4i32 V128:$Vd, imm0_255:$imm, 0)>;
3863 def : InstAlias<"orr.4h $Vd, $imm", (ORRv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
3864 def : InstAlias<"orr.8h $Vd, $imm", (ORRv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
3865 def : InstAlias<"orr.2s $Vd, $imm", (ORRv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
3866 def : InstAlias<"orr.4s $Vd, $imm", (ORRv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
3869 def FMOVv2f64_ns : SIMDModifiedImmVectorNoShift<1, 1, 0b1111, V128, fpimm8,
3871 [(set (v2f64 V128:$Rd), (ARM64fmov imm0_255:$imm8))]>;
3872 def FMOVv2f32_ns : SIMDModifiedImmVectorNoShift<0, 0, 0b1111, V64, fpimm8,
3874 [(set (v2f32 V64:$Rd), (ARM64fmov imm0_255:$imm8))]>;
3875 def FMOVv4f32_ns : SIMDModifiedImmVectorNoShift<1, 0, 0b1111, V128, fpimm8,
3877 [(set (v4f32 V128:$Rd), (ARM64fmov imm0_255:$imm8))]>;
3881 // EDIT byte mask: scalar
3882 let isReMaterializable = 1, isAsCheapAsAMove = 1 in
3883 def MOVID : SIMDModifiedImmScalarNoShift<0, 1, 0b1110, "movi",
3884 [(set FPR64:$Rd, simdimmtype10:$imm8)]>;
3885 // The movi_edit node has the immediate value already encoded, so we use
3886 // a plain imm0_255 here.
3887 def : Pat<(f64 (ARM64movi_edit imm0_255:$shift)),
3888 (MOVID imm0_255:$shift)>;
3890 def : Pat<(v1i64 immAllZerosV), (MOVID (i32 0))>;
3891 def : Pat<(v2i32 immAllZerosV), (MOVID (i32 0))>;
3892 def : Pat<(v4i16 immAllZerosV), (MOVID (i32 0))>;
3893 def : Pat<(v8i8 immAllZerosV), (MOVID (i32 0))>;
3895 def : Pat<(v1i64 immAllOnesV), (MOVID (i32 255))>;
3896 def : Pat<(v2i32 immAllOnesV), (MOVID (i32 255))>;
3897 def : Pat<(v4i16 immAllOnesV), (MOVID (i32 255))>;
3898 def : Pat<(v8i8 immAllOnesV), (MOVID (i32 255))>;
3900 // EDIT byte mask: 2d
3902 // The movi_edit node has the immediate value already encoded, so we use
3903 // a plain imm0_255 in the pattern
3904 let isReMaterializable = 1, isAsCheapAsAMove = 1 in
3905 def MOVIv2d_ns : SIMDModifiedImmVectorNoShift<1, 1, 0b1110, V128,
3908 [(set (v2i64 V128:$Rd), (ARM64movi_edit imm0_255:$imm8))]>;
3911 // Use movi.2d to materialize 0.0 if the HW does zero-cycle zeroing.
3912 // Complexity is added to break a tie with a plain MOVI.
3913 let AddedComplexity = 1 in {
3914 def : Pat<(f32 fpimm0),
3915 (f32 (EXTRACT_SUBREG (v2i64 (MOVIv2d_ns (i32 0))), ssub))>,
3917 def : Pat<(f64 fpimm0),
3918 (f64 (EXTRACT_SUBREG (v2i64 (MOVIv2d_ns (i32 0))), dsub))>,
3922 def : Pat<(v2i64 immAllZerosV), (MOVIv2d_ns (i32 0))>;
3923 def : Pat<(v4i32 immAllZerosV), (MOVIv2d_ns (i32 0))>;
3924 def : Pat<(v8i16 immAllZerosV), (MOVIv2d_ns (i32 0))>;
3925 def : Pat<(v16i8 immAllZerosV), (MOVIv2d_ns (i32 0))>;
3927 def : Pat<(v2i64 immAllOnesV), (MOVIv2d_ns (i32 255))>;
3928 def : Pat<(v4i32 immAllOnesV), (MOVIv2d_ns (i32 255))>;
3929 def : Pat<(v8i16 immAllOnesV), (MOVIv2d_ns (i32 255))>;
3930 def : Pat<(v16i8 immAllOnesV), (MOVIv2d_ns (i32 255))>;
3932 def : Pat<(v2f64 (ARM64dup (f64 fpimm0))), (MOVIv2d_ns (i32 0))>;
3933 def : Pat<(v4f32 (ARM64dup (f32 fpimm0))), (MOVIv2d_ns (i32 0))>;
3935 // EDIT per word & halfword: 2s, 4h, 4s, & 8h
3936 defm MOVI : SIMDModifiedImmVectorShift<0, 0b10, 0b00, "movi">;
3938 def : InstAlias<"movi $Vd.4h, $imm", (MOVIv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
3939 def : InstAlias<"movi $Vd.8h, $imm", (MOVIv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
3940 def : InstAlias<"movi $Vd.2s, $imm", (MOVIv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
3941 def : InstAlias<"movi $Vd.4s, $imm", (MOVIv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
3943 def : InstAlias<"movi.4h $Vd, $imm", (MOVIv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
3944 def : InstAlias<"movi.8h $Vd, $imm", (MOVIv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
3945 def : InstAlias<"movi.2s $Vd, $imm", (MOVIv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
3946 def : InstAlias<"movi.4s $Vd, $imm", (MOVIv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
3948 def : Pat<(v2i32 (ARM64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
3949 (MOVIv2i32 imm0_255:$imm8, imm:$shift)>;
3950 def : Pat<(v4i32 (ARM64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
3951 (MOVIv4i32 imm0_255:$imm8, imm:$shift)>;
3952 def : Pat<(v4i16 (ARM64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
3953 (MOVIv4i16 imm0_255:$imm8, imm:$shift)>;
3954 def : Pat<(v8i16 (ARM64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
3955 (MOVIv8i16 imm0_255:$imm8, imm:$shift)>;
3957 // EDIT per word: 2s & 4s with MSL shifter
3958 def MOVIv2s_msl : SIMDModifiedImmMoveMSL<0, 0, {1,1,0,?}, V64, "movi", ".2s",
3959 [(set (v2i32 V64:$Rd),
3960 (ARM64movi_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
3961 def MOVIv4s_msl : SIMDModifiedImmMoveMSL<1, 0, {1,1,0,?}, V128, "movi", ".4s",
3962 [(set (v4i32 V128:$Rd),
3963 (ARM64movi_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
3965 // Per byte: 8b & 16b
3966 def MOVIv8b_ns : SIMDModifiedImmVectorNoShift<0, 0, 0b1110, V64, imm0_255,
3968 [(set (v8i8 V64:$Rd), (ARM64movi imm0_255:$imm8))]>;
3969 def MOVIv16b_ns : SIMDModifiedImmVectorNoShift<1, 0, 0b1110, V128, imm0_255,
3971 [(set (v16i8 V128:$Rd), (ARM64movi imm0_255:$imm8))]>;
3975 // EDIT per word & halfword: 2s, 4h, 4s, & 8h
3976 defm MVNI : SIMDModifiedImmVectorShift<1, 0b10, 0b00, "mvni">;
3978 def : InstAlias<"mvni $Vd.4h, $imm", (MVNIv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
3979 def : InstAlias<"mvni $Vd.8h, $imm", (MVNIv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
3980 def : InstAlias<"mvni $Vd.2s, $imm", (MVNIv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
3981 def : InstAlias<"mvni $Vd.4s, $imm", (MVNIv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
3983 def : InstAlias<"mvni.4h $Vd, $imm", (MVNIv4i16 V64:$Vd, imm0_255:$imm, 0), 0>;
3984 def : InstAlias<"mvni.8h $Vd, $imm", (MVNIv8i16 V128:$Vd, imm0_255:$imm, 0), 0>;
3985 def : InstAlias<"mvni.2s $Vd, $imm", (MVNIv2i32 V64:$Vd, imm0_255:$imm, 0), 0>;
3986 def : InstAlias<"mvni.4s $Vd, $imm", (MVNIv4i32 V128:$Vd, imm0_255:$imm, 0), 0>;
3988 def : Pat<(v2i32 (ARM64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
3989 (MVNIv2i32 imm0_255:$imm8, imm:$shift)>;
3990 def : Pat<(v4i32 (ARM64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
3991 (MVNIv4i32 imm0_255:$imm8, imm:$shift)>;
3992 def : Pat<(v4i16 (ARM64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
3993 (MVNIv4i16 imm0_255:$imm8, imm:$shift)>;
3994 def : Pat<(v8i16 (ARM64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
3995 (MVNIv8i16 imm0_255:$imm8, imm:$shift)>;
3997 // EDIT per word: 2s & 4s with MSL shifter
3998 def MVNIv2s_msl : SIMDModifiedImmMoveMSL<0, 1, {1,1,0,?}, V64, "mvni", ".2s",
3999 [(set (v2i32 V64:$Rd),
4000 (ARM64mvni_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
4001 def MVNIv4s_msl : SIMDModifiedImmMoveMSL<1, 1, {1,1,0,?}, V128, "mvni", ".4s",
4002 [(set (v4i32 V128:$Rd),
4003 (ARM64mvni_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
4005 //----------------------------------------------------------------------------
4006 // AdvSIMD indexed element
4007 //----------------------------------------------------------------------------
4009 let neverHasSideEffects = 1 in {
4010 defm FMLA : SIMDFPIndexedSDTied<0, 0b0001, "fmla">;
4011 defm FMLS : SIMDFPIndexedSDTied<0, 0b0101, "fmls">;
4014 // NOTE: Operands are reordered in the FMLA/FMLS PatFrags because the
4015 // instruction expects the addend first, while the intrinsic expects it last.
4017 // On the other hand, there are quite a few valid combinatorial options due to
4018 // the commutativity of multiplication and the fact that (-x) * y = x * (-y).
4019 defm : SIMDFPIndexedSDTiedPatterns<"FMLA",
4020 TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)>>;
4021 defm : SIMDFPIndexedSDTiedPatterns<"FMLA",
4022 TriOpFrag<(fma node:$MHS, node:$RHS, node:$LHS)>>;
4024 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
4025 TriOpFrag<(fma node:$MHS, (fneg node:$RHS), node:$LHS)> >;
4026 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
4027 TriOpFrag<(fma node:$RHS, (fneg node:$MHS), node:$LHS)> >;
4028 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
4029 TriOpFrag<(fma (fneg node:$RHS), node:$MHS, node:$LHS)> >;
4030 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
4031 TriOpFrag<(fma (fneg node:$MHS), node:$RHS, node:$LHS)> >;
4033 multiclass FMLSIndexedAfterNegPatterns<SDPatternOperator OpNode> {
4034 // 3 variants for the .2s version: DUPLANE from 128-bit, DUPLANE from 64-bit
4036 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
4037 (ARM64duplane32 (v4f32 (fneg V128:$Rm)),
4038 VectorIndexS:$idx))),
4039 (FMLSv2i32_indexed V64:$Rd, V64:$Rn, V128:$Rm, VectorIndexS:$idx)>;
4040 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
4041 (v2f32 (ARM64duplane32
4042 (v4f32 (insert_subvector undef,
4043 (v2f32 (fneg V64:$Rm)),
4045 VectorIndexS:$idx)))),
4046 (FMLSv2i32_indexed V64:$Rd, V64:$Rn,
4047 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub),
4048 VectorIndexS:$idx)>;
4049 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
4050 (ARM64dup (f32 (fneg FPR32Op:$Rm))))),
4051 (FMLSv2i32_indexed V64:$Rd, V64:$Rn,
4052 (SUBREG_TO_REG (i32 0), FPR32Op:$Rm, ssub), (i64 0))>;
4054 // 3 variants for the .4s version: DUPLANE from 128-bit, DUPLANE from 64-bit
4056 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
4057 (ARM64duplane32 (v4f32 (fneg V128:$Rm)),
4058 VectorIndexS:$idx))),
4059 (FMLSv4i32_indexed V128:$Rd, V128:$Rn, V128:$Rm,
4060 VectorIndexS:$idx)>;
4061 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
4062 (v4f32 (ARM64duplane32
4063 (v4f32 (insert_subvector undef,
4064 (v2f32 (fneg V64:$Rm)),
4066 VectorIndexS:$idx)))),
4067 (FMLSv4i32_indexed V128:$Rd, V128:$Rn,
4068 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub),
4069 VectorIndexS:$idx)>;
4070 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
4071 (ARM64dup (f32 (fneg FPR32Op:$Rm))))),
4072 (FMLSv4i32_indexed V128:$Rd, V128:$Rn,
4073 (SUBREG_TO_REG (i32 0), FPR32Op:$Rm, ssub), (i64 0))>;
4075 // 2 variants for the .2d version: DUPLANE from 128-bit, and DUP scalar
4076 // (DUPLANE from 64-bit would be trivial).
4077 def : Pat<(v2f64 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn),
4078 (ARM64duplane64 (v2f64 (fneg V128:$Rm)),
4079 VectorIndexD:$idx))),
4081 V128:$Rd, V128:$Rn, V128:$Rm, VectorIndexS:$idx)>;
4082 def : Pat<(v2f64 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn),
4083 (ARM64dup (f64 (fneg FPR64Op:$Rm))))),
4084 (FMLSv2i64_indexed V128:$Rd, V128:$Rn,
4085 (SUBREG_TO_REG (i32 0), FPR64Op:$Rm, dsub), (i64 0))>;
4087 // 2 variants for 32-bit scalar version: extract from .2s or from .4s
4088 def : Pat<(f32 (OpNode (f32 FPR32:$Rd), (f32 FPR32:$Rn),
4089 (vector_extract (v4f32 (fneg V128:$Rm)),
4090 VectorIndexS:$idx))),
4091 (FMLSv1i32_indexed FPR32:$Rd, FPR32:$Rn,
4092 V128:$Rm, VectorIndexS:$idx)>;
4093 def : Pat<(f32 (OpNode (f32 FPR32:$Rd), (f32 FPR32:$Rn),
4094 (vector_extract (v2f32 (fneg V64:$Rm)),
4095 VectorIndexS:$idx))),
4096 (FMLSv1i32_indexed FPR32:$Rd, FPR32:$Rn,
4097 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub), VectorIndexS:$idx)>;
4099 // 1 variant for 64-bit scalar version: extract from .1d or from .2d
4100 def : Pat<(f64 (OpNode (f64 FPR64:$Rd), (f64 FPR64:$Rn),
4101 (vector_extract (v2f64 (fneg V128:$Rm)),
4102 VectorIndexS:$idx))),
4103 (FMLSv1i64_indexed FPR64:$Rd, FPR64:$Rn,
4104 V128:$Rm, VectorIndexS:$idx)>;
4107 defm : FMLSIndexedAfterNegPatterns<
4108 TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)> >;
4109 defm : FMLSIndexedAfterNegPatterns<
4110 TriOpFrag<(fma node:$MHS, node:$RHS, node:$LHS)> >;
4112 defm FMULX : SIMDFPIndexedSD<1, 0b1001, "fmulx", int_arm64_neon_fmulx>;
4113 defm FMUL : SIMDFPIndexedSD<0, 0b1001, "fmul", fmul>;
4115 def : Pat<(v2f32 (fmul V64:$Rn, (ARM64dup (f32 FPR32:$Rm)))),
4116 (FMULv2i32_indexed V64:$Rn,
4117 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rm, ssub),
4119 def : Pat<(v4f32 (fmul V128:$Rn, (ARM64dup (f32 FPR32:$Rm)))),
4120 (FMULv4i32_indexed V128:$Rn,
4121 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rm, ssub),
4123 def : Pat<(v2f64 (fmul V128:$Rn, (ARM64dup (f64 FPR64:$Rm)))),
4124 (FMULv2i64_indexed V128:$Rn,
4125 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$Rm, dsub),
4128 defm SQDMULH : SIMDIndexedHS<0, 0b1100, "sqdmulh", int_arm64_neon_sqdmulh>;
4129 defm SQRDMULH : SIMDIndexedHS<0, 0b1101, "sqrdmulh", int_arm64_neon_sqrdmulh>;
4130 defm MLA : SIMDVectorIndexedHSTied<1, 0b0000, "mla",
4131 TriOpFrag<(add node:$LHS, (mul node:$MHS, node:$RHS))>>;
4132 defm MLS : SIMDVectorIndexedHSTied<1, 0b0100, "mls",
4133 TriOpFrag<(sub node:$LHS, (mul node:$MHS, node:$RHS))>>;
4134 defm MUL : SIMDVectorIndexedHS<0, 0b1000, "mul", mul>;
4135 defm SMLAL : SIMDVectorIndexedLongSDTied<0, 0b0010, "smlal",
4136 TriOpFrag<(add node:$LHS, (int_arm64_neon_smull node:$MHS, node:$RHS))>>;
4137 defm SMLSL : SIMDVectorIndexedLongSDTied<0, 0b0110, "smlsl",
4138 TriOpFrag<(sub node:$LHS, (int_arm64_neon_smull node:$MHS, node:$RHS))>>;
4139 defm SMULL : SIMDVectorIndexedLongSD<0, 0b1010, "smull",
4140 int_arm64_neon_smull>;
4141 defm SQDMLAL : SIMDIndexedLongSQDMLXSDTied<0, 0b0011, "sqdmlal",
4142 int_arm64_neon_sqadd>;
4143 defm SQDMLSL : SIMDIndexedLongSQDMLXSDTied<0, 0b0111, "sqdmlsl",
4144 int_arm64_neon_sqsub>;
4145 defm SQDMULL : SIMDIndexedLongSD<0, 0b1011, "sqdmull", int_arm64_neon_sqdmull>;
4146 defm UMLAL : SIMDVectorIndexedLongSDTied<1, 0b0010, "umlal",
4147 TriOpFrag<(add node:$LHS, (int_arm64_neon_umull node:$MHS, node:$RHS))>>;
4148 defm UMLSL : SIMDVectorIndexedLongSDTied<1, 0b0110, "umlsl",
4149 TriOpFrag<(sub node:$LHS, (int_arm64_neon_umull node:$MHS, node:$RHS))>>;
4150 defm UMULL : SIMDVectorIndexedLongSD<1, 0b1010, "umull",
4151 int_arm64_neon_umull>;
4153 // A scalar sqdmull with the second operand being a vector lane can be
4154 // handled directly with the indexed instruction encoding.
4155 def : Pat<(int_arm64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
4156 (vector_extract (v4i32 V128:$Vm),
4157 VectorIndexS:$idx)),
4158 (SQDMULLv1i64_indexed FPR32:$Rn, V128:$Vm, VectorIndexS:$idx)>;
4160 //----------------------------------------------------------------------------
4161 // AdvSIMD scalar shift instructions
4162 //----------------------------------------------------------------------------
4163 defm FCVTZS : SIMDScalarRShiftSD<0, 0b11111, "fcvtzs">;
4164 defm FCVTZU : SIMDScalarRShiftSD<1, 0b11111, "fcvtzu">;
4165 defm SCVTF : SIMDScalarRShiftSD<0, 0b11100, "scvtf">;
4166 defm UCVTF : SIMDScalarRShiftSD<1, 0b11100, "ucvtf">;
4167 // Codegen patterns for the above. We don't put these directly on the
4168 // instructions because TableGen's type inference can't handle the truth.
4169 // Having the same base pattern for fp <--> int totally freaks it out.
4170 def : Pat<(int_arm64_neon_vcvtfp2fxs FPR32:$Rn, vecshiftR32:$imm),
4171 (FCVTZSs FPR32:$Rn, vecshiftR32:$imm)>;
4172 def : Pat<(int_arm64_neon_vcvtfp2fxu FPR32:$Rn, vecshiftR32:$imm),
4173 (FCVTZUs FPR32:$Rn, vecshiftR32:$imm)>;
4174 def : Pat<(i64 (int_arm64_neon_vcvtfp2fxs (f64 FPR64:$Rn), vecshiftR64:$imm)),
4175 (FCVTZSd FPR64:$Rn, vecshiftR64:$imm)>;
4176 def : Pat<(i64 (int_arm64_neon_vcvtfp2fxu (f64 FPR64:$Rn), vecshiftR64:$imm)),
4177 (FCVTZUd FPR64:$Rn, vecshiftR64:$imm)>;
4178 def : Pat<(v1i64 (int_arm64_neon_vcvtfp2fxs (v1f64 FPR64:$Rn),
4180 (FCVTZSd FPR64:$Rn, vecshiftR64:$imm)>;
4181 def : Pat<(v1i64 (int_arm64_neon_vcvtfp2fxu (v1f64 FPR64:$Rn),
4183 (FCVTZUd FPR64:$Rn, vecshiftR64:$imm)>;
4184 def : Pat<(int_arm64_neon_vcvtfxs2fp FPR32:$Rn, vecshiftR32:$imm),
4185 (SCVTFs FPR32:$Rn, vecshiftR32:$imm)>;
4186 def : Pat<(int_arm64_neon_vcvtfxu2fp FPR32:$Rn, vecshiftR32:$imm),
4187 (UCVTFs FPR32:$Rn, vecshiftR32:$imm)>;
4188 def : Pat<(f64 (int_arm64_neon_vcvtfxs2fp (i64 FPR64:$Rn), vecshiftR64:$imm)),
4189 (SCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
4190 def : Pat<(f64 (int_arm64_neon_vcvtfxu2fp (i64 FPR64:$Rn), vecshiftR64:$imm)),
4191 (UCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
4192 def : Pat<(v1f64 (int_arm64_neon_vcvtfxs2fp (v1i64 FPR64:$Rn),
4194 (SCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
4195 def : Pat<(v1f64 (int_arm64_neon_vcvtfxu2fp (v1i64 FPR64:$Rn),
4197 (UCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
4199 defm SHL : SIMDScalarLShiftD< 0, 0b01010, "shl", ARM64vshl>;
4200 defm SLI : SIMDScalarLShiftDTied<1, 0b01010, "sli">;
4201 defm SQRSHRN : SIMDScalarRShiftBHS< 0, 0b10011, "sqrshrn",
4202 int_arm64_neon_sqrshrn>;
4203 defm SQRSHRUN : SIMDScalarRShiftBHS< 1, 0b10001, "sqrshrun",
4204 int_arm64_neon_sqrshrun>;
4205 defm SQSHLU : SIMDScalarLShiftBHSD<1, 0b01100, "sqshlu", ARM64sqshlui>;
4206 defm SQSHL : SIMDScalarLShiftBHSD<0, 0b01110, "sqshl", ARM64sqshli>;
4207 defm SQSHRN : SIMDScalarRShiftBHS< 0, 0b10010, "sqshrn",
4208 int_arm64_neon_sqshrn>;
4209 defm SQSHRUN : SIMDScalarRShiftBHS< 1, 0b10000, "sqshrun",
4210 int_arm64_neon_sqshrun>;
4211 defm SRI : SIMDScalarRShiftDTied< 1, 0b01000, "sri">;
4212 defm SRSHR : SIMDScalarRShiftD< 0, 0b00100, "srshr", ARM64srshri>;
4213 defm SRSRA : SIMDScalarRShiftDTied< 0, 0b00110, "srsra",
4214 TriOpFrag<(add node:$LHS,
4215 (ARM64srshri node:$MHS, node:$RHS))>>;
4216 defm SSHR : SIMDScalarRShiftD< 0, 0b00000, "sshr", ARM64vashr>;
4217 defm SSRA : SIMDScalarRShiftDTied< 0, 0b00010, "ssra",
4218 TriOpFrag<(add node:$LHS,
4219 (ARM64vashr node:$MHS, node:$RHS))>>;
4220 defm UQRSHRN : SIMDScalarRShiftBHS< 1, 0b10011, "uqrshrn",
4221 int_arm64_neon_uqrshrn>;
4222 defm UQSHL : SIMDScalarLShiftBHSD<1, 0b01110, "uqshl", ARM64uqshli>;
4223 defm UQSHRN : SIMDScalarRShiftBHS< 1, 0b10010, "uqshrn",
4224 int_arm64_neon_uqshrn>;
4225 defm URSHR : SIMDScalarRShiftD< 1, 0b00100, "urshr", ARM64urshri>;
4226 defm URSRA : SIMDScalarRShiftDTied< 1, 0b00110, "ursra",
4227 TriOpFrag<(add node:$LHS,
4228 (ARM64urshri node:$MHS, node:$RHS))>>;
4229 defm USHR : SIMDScalarRShiftD< 1, 0b00000, "ushr", ARM64vlshr>;
4230 defm USRA : SIMDScalarRShiftDTied< 1, 0b00010, "usra",
4231 TriOpFrag<(add node:$LHS,
4232 (ARM64vlshr node:$MHS, node:$RHS))>>;
4234 //----------------------------------------------------------------------------
4235 // AdvSIMD vector shift instructions
4236 //----------------------------------------------------------------------------
4237 defm FCVTZS:SIMDVectorRShiftSD<0, 0b11111, "fcvtzs", int_arm64_neon_vcvtfp2fxs>;
4238 defm FCVTZU:SIMDVectorRShiftSD<1, 0b11111, "fcvtzu", int_arm64_neon_vcvtfp2fxu>;
4239 defm SCVTF: SIMDVectorRShiftSDToFP<0, 0b11100, "scvtf",
4240 int_arm64_neon_vcvtfxs2fp>;
4241 defm RSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10001, "rshrn",
4242 int_arm64_neon_rshrn>;
4243 defm SHL : SIMDVectorLShiftBHSD<0, 0b01010, "shl", ARM64vshl>;
4244 defm SHRN : SIMDVectorRShiftNarrowBHS<0, 0b10000, "shrn",
4245 BinOpFrag<(trunc (ARM64vashr node:$LHS, node:$RHS))>>;
4246 defm SLI : SIMDVectorLShiftBHSDTied<1, 0b01010, "sli", int_arm64_neon_vsli>;
4247 def : Pat<(v1i64 (int_arm64_neon_vsli (v1i64 FPR64:$Rd), (v1i64 FPR64:$Rn),
4248 (i32 vecshiftL64:$imm))),
4249 (SLId FPR64:$Rd, FPR64:$Rn, vecshiftL64:$imm)>;
4250 defm SQRSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10011, "sqrshrn",
4251 int_arm64_neon_sqrshrn>;
4252 defm SQRSHRUN: SIMDVectorRShiftNarrowBHS<1, 0b10001, "sqrshrun",
4253 int_arm64_neon_sqrshrun>;
4254 defm SQSHLU : SIMDVectorLShiftBHSD<1, 0b01100, "sqshlu", ARM64sqshlui>;
4255 defm SQSHL : SIMDVectorLShiftBHSD<0, 0b01110, "sqshl", ARM64sqshli>;
4256 defm SQSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10010, "sqshrn",
4257 int_arm64_neon_sqshrn>;
4258 defm SQSHRUN : SIMDVectorRShiftNarrowBHS<1, 0b10000, "sqshrun",
4259 int_arm64_neon_sqshrun>;
4260 defm SRI : SIMDVectorRShiftBHSDTied<1, 0b01000, "sri", int_arm64_neon_vsri>;
4261 def : Pat<(v1i64 (int_arm64_neon_vsri (v1i64 FPR64:$Rd), (v1i64 FPR64:$Rn),
4262 (i32 vecshiftR64:$imm))),
4263 (SRId FPR64:$Rd, FPR64:$Rn, vecshiftR64:$imm)>;
4264 defm SRSHR : SIMDVectorRShiftBHSD<0, 0b00100, "srshr", ARM64srshri>;
4265 defm SRSRA : SIMDVectorRShiftBHSDTied<0, 0b00110, "srsra",
4266 TriOpFrag<(add node:$LHS,
4267 (ARM64srshri node:$MHS, node:$RHS))> >;
4268 defm SSHLL : SIMDVectorLShiftLongBHSD<0, 0b10100, "sshll",
4269 BinOpFrag<(ARM64vshl (sext node:$LHS), node:$RHS)>>;
4271 defm SSHR : SIMDVectorRShiftBHSD<0, 0b00000, "sshr", ARM64vashr>;
4272 defm SSRA : SIMDVectorRShiftBHSDTied<0, 0b00010, "ssra",
4273 TriOpFrag<(add node:$LHS, (ARM64vashr node:$MHS, node:$RHS))>>;
4274 defm UCVTF : SIMDVectorRShiftSDToFP<1, 0b11100, "ucvtf",
4275 int_arm64_neon_vcvtfxu2fp>;
4276 defm UQRSHRN : SIMDVectorRShiftNarrowBHS<1, 0b10011, "uqrshrn",
4277 int_arm64_neon_uqrshrn>;
4278 defm UQSHL : SIMDVectorLShiftBHSD<1, 0b01110, "uqshl", ARM64uqshli>;
4279 defm UQSHRN : SIMDVectorRShiftNarrowBHS<1, 0b10010, "uqshrn",
4280 int_arm64_neon_uqshrn>;
4281 defm URSHR : SIMDVectorRShiftBHSD<1, 0b00100, "urshr", ARM64urshri>;
4282 defm URSRA : SIMDVectorRShiftBHSDTied<1, 0b00110, "ursra",
4283 TriOpFrag<(add node:$LHS,
4284 (ARM64urshri node:$MHS, node:$RHS))> >;
4285 defm USHLL : SIMDVectorLShiftLongBHSD<1, 0b10100, "ushll",
4286 BinOpFrag<(ARM64vshl (zext node:$LHS), node:$RHS)>>;
4287 defm USHR : SIMDVectorRShiftBHSD<1, 0b00000, "ushr", ARM64vlshr>;
4288 defm USRA : SIMDVectorRShiftBHSDTied<1, 0b00010, "usra",
4289 TriOpFrag<(add node:$LHS, (ARM64vlshr node:$MHS, node:$RHS))> >;
4291 // SHRN patterns for when a logical right shift was used instead of arithmetic
4292 // (the immediate guarantees no sign bits actually end up in the result so it
4294 def : Pat<(v8i8 (trunc (ARM64vlshr (v8i16 V128:$Rn), vecshiftR16Narrow:$imm))),
4295 (SHRNv8i8_shift V128:$Rn, vecshiftR16Narrow:$imm)>;
4296 def : Pat<(v4i16 (trunc (ARM64vlshr (v4i32 V128:$Rn), vecshiftR32Narrow:$imm))),
4297 (SHRNv4i16_shift V128:$Rn, vecshiftR32Narrow:$imm)>;
4298 def : Pat<(v2i32 (trunc (ARM64vlshr (v2i64 V128:$Rn), vecshiftR64Narrow:$imm))),
4299 (SHRNv2i32_shift V128:$Rn, vecshiftR64Narrow:$imm)>;
4301 def : Pat<(v16i8 (concat_vectors (v8i8 V64:$Rd),
4302 (trunc (ARM64vlshr (v8i16 V128:$Rn),
4303 vecshiftR16Narrow:$imm)))),
4304 (SHRNv16i8_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
4305 V128:$Rn, vecshiftR16Narrow:$imm)>;
4306 def : Pat<(v8i16 (concat_vectors (v4i16 V64:$Rd),
4307 (trunc (ARM64vlshr (v4i32 V128:$Rn),
4308 vecshiftR32Narrow:$imm)))),
4309 (SHRNv8i16_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
4310 V128:$Rn, vecshiftR32Narrow:$imm)>;
4311 def : Pat<(v4i32 (concat_vectors (v2i32 V64:$Rd),
4312 (trunc (ARM64vlshr (v2i64 V128:$Rn),
4313 vecshiftR64Narrow:$imm)))),
4314 (SHRNv4i32_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
4315 V128:$Rn, vecshiftR32Narrow:$imm)>;
4317 // Vector sign and zero extensions are implemented with SSHLL and USSHLL.
4318 // Anyexts are implemented as zexts.
4319 def : Pat<(v8i16 (sext (v8i8 V64:$Rn))), (SSHLLv8i8_shift V64:$Rn, (i32 0))>;
4320 def : Pat<(v8i16 (zext (v8i8 V64:$Rn))), (USHLLv8i8_shift V64:$Rn, (i32 0))>;
4321 def : Pat<(v8i16 (anyext (v8i8 V64:$Rn))), (USHLLv8i8_shift V64:$Rn, (i32 0))>;
4322 def : Pat<(v4i32 (sext (v4i16 V64:$Rn))), (SSHLLv4i16_shift V64:$Rn, (i32 0))>;
4323 def : Pat<(v4i32 (zext (v4i16 V64:$Rn))), (USHLLv4i16_shift V64:$Rn, (i32 0))>;
4324 def : Pat<(v4i32 (anyext (v4i16 V64:$Rn))), (USHLLv4i16_shift V64:$Rn, (i32 0))>;
4325 def : Pat<(v2i64 (sext (v2i32 V64:$Rn))), (SSHLLv2i32_shift V64:$Rn, (i32 0))>;
4326 def : Pat<(v2i64 (zext (v2i32 V64:$Rn))), (USHLLv2i32_shift V64:$Rn, (i32 0))>;
4327 def : Pat<(v2i64 (anyext (v2i32 V64:$Rn))), (USHLLv2i32_shift V64:$Rn, (i32 0))>;
4328 // Also match an extend from the upper half of a 128 bit source register.
4329 def : Pat<(v8i16 (anyext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
4330 (USHLLv16i8_shift V128:$Rn, (i32 0))>;
4331 def : Pat<(v8i16 (zext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
4332 (USHLLv16i8_shift V128:$Rn, (i32 0))>;
4333 def : Pat<(v8i16 (sext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
4334 (SSHLLv16i8_shift V128:$Rn, (i32 0))>;
4335 def : Pat<(v4i32 (anyext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
4336 (USHLLv8i16_shift V128:$Rn, (i32 0))>;
4337 def : Pat<(v4i32 (zext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
4338 (USHLLv8i16_shift V128:$Rn, (i32 0))>;
4339 def : Pat<(v4i32 (sext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
4340 (SSHLLv8i16_shift V128:$Rn, (i32 0))>;
4341 def : Pat<(v2i64 (anyext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
4342 (USHLLv4i32_shift V128:$Rn, (i32 0))>;
4343 def : Pat<(v2i64 (zext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
4344 (USHLLv4i32_shift V128:$Rn, (i32 0))>;
4345 def : Pat<(v2i64 (sext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
4346 (SSHLLv4i32_shift V128:$Rn, (i32 0))>;
4348 // Vector shift sxtl aliases
4349 def : InstAlias<"sxtl.8h $dst, $src1",
4350 (SSHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
4351 def : InstAlias<"sxtl $dst.8h, $src1.8b",
4352 (SSHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
4353 def : InstAlias<"sxtl.4s $dst, $src1",
4354 (SSHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
4355 def : InstAlias<"sxtl $dst.4s, $src1.4h",
4356 (SSHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
4357 def : InstAlias<"sxtl.2d $dst, $src1",
4358 (SSHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
4359 def : InstAlias<"sxtl $dst.2d, $src1.2s",
4360 (SSHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
4362 // Vector shift sxtl2 aliases
4363 def : InstAlias<"sxtl2.8h $dst, $src1",
4364 (SSHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
4365 def : InstAlias<"sxtl2 $dst.8h, $src1.16b",
4366 (SSHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
4367 def : InstAlias<"sxtl2.4s $dst, $src1",
4368 (SSHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
4369 def : InstAlias<"sxtl2 $dst.4s, $src1.8h",
4370 (SSHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
4371 def : InstAlias<"sxtl2.2d $dst, $src1",
4372 (SSHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
4373 def : InstAlias<"sxtl2 $dst.2d, $src1.4s",
4374 (SSHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
4376 // Vector shift uxtl aliases
4377 def : InstAlias<"uxtl.8h $dst, $src1",
4378 (USHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
4379 def : InstAlias<"uxtl $dst.8h, $src1.8b",
4380 (USHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
4381 def : InstAlias<"uxtl.4s $dst, $src1",
4382 (USHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
4383 def : InstAlias<"uxtl $dst.4s, $src1.4h",
4384 (USHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
4385 def : InstAlias<"uxtl.2d $dst, $src1",
4386 (USHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
4387 def : InstAlias<"uxtl $dst.2d, $src1.2s",
4388 (USHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
4390 // Vector shift uxtl2 aliases
4391 def : InstAlias<"uxtl2.8h $dst, $src1",
4392 (USHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
4393 def : InstAlias<"uxtl2 $dst.8h, $src1.16b",
4394 (USHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
4395 def : InstAlias<"uxtl2.4s $dst, $src1",
4396 (USHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
4397 def : InstAlias<"uxtl2 $dst.4s, $src1.8h",
4398 (USHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
4399 def : InstAlias<"uxtl2.2d $dst, $src1",
4400 (USHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
4401 def : InstAlias<"uxtl2 $dst.2d, $src1.4s",
4402 (USHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
4404 // If an integer is about to be converted to a floating point value,
4405 // just load it on the floating point unit.
4406 // These patterns are more complex because floating point loads do not
4407 // support sign extension.
4408 // The sign extension has to be explicitly added and is only supported for
4409 // one step: byte-to-half, half-to-word, word-to-doubleword.
4410 // SCVTF GPR -> FPR is 9 cycles.
4411 // SCVTF FPR -> FPR is 4 cyclces.
4412 // (sign extension with lengthen) SXTL FPR -> FPR is 2 cycles.
4413 // Therefore, we can do 2 sign extensions and one SCVTF FPR -> FPR
4414 // and still being faster.
4415 // However, this is not good for code size.
4416 // 8-bits -> float. 2 sizes step-up.
4417 class SExtLoadi8CVTf32Pat<dag addrmode, dag INST>
4418 : Pat<(f32 (sint_to_fp (i32 (sextloadi8 addrmode)))),
4419 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
4424 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4430 ssub)))>, Requires<[NotForCodeSize]>;
4432 def : SExtLoadi8CVTf32Pat<(ro8.Wpat GPR64sp:$Rn, GPR32:$Rm, ro8.Wext:$ext),
4433 (LDRBroW GPR64sp:$Rn, GPR32:$Rm, ro8.Wext:$ext)>;
4434 def : SExtLoadi8CVTf32Pat<(ro8.Xpat GPR64sp:$Rn, GPR64:$Rm, ro8.Xext:$ext),
4435 (LDRBroX GPR64sp:$Rn, GPR64:$Rm, ro8.Xext:$ext)>;
4436 def : SExtLoadi8CVTf32Pat<(am_indexed8 GPR64sp:$Rn, uimm12s1:$offset),
4437 (LDRBui GPR64sp:$Rn, uimm12s1:$offset)>;
4438 def : SExtLoadi8CVTf32Pat<(am_unscaled8 GPR64sp:$Rn, simm9:$offset),
4439 (LDURBi GPR64sp:$Rn, simm9:$offset)>;
4441 // 16-bits -> float. 1 size step-up.
4442 class SExtLoadi16CVTf32Pat<dag addrmode, dag INST>
4443 : Pat<(f32 (sint_to_fp (i32 (sextloadi16 addrmode)))),
4444 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
4446 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4450 ssub)))>, Requires<[NotForCodeSize]>;
4452 def : SExtLoadi16CVTf32Pat<(ro16.Wpat GPR64sp:$Rn, GPR32:$Rm, ro16.Wext:$ext),
4453 (LDRHroW GPR64sp:$Rn, GPR32:$Rm, ro16.Wext:$ext)>;
4454 def : SExtLoadi16CVTf32Pat<(ro16.Xpat GPR64sp:$Rn, GPR64:$Rm, ro16.Xext:$ext),
4455 (LDRHroX GPR64sp:$Rn, GPR64:$Rm, ro16.Xext:$ext)>;
4456 def : SExtLoadi16CVTf32Pat<(am_indexed16 GPR64sp:$Rn, uimm12s2:$offset),
4457 (LDRHui GPR64sp:$Rn, uimm12s2:$offset)>;
4458 def : SExtLoadi16CVTf32Pat<(am_unscaled16 GPR64sp:$Rn, simm9:$offset),
4459 (LDURHi GPR64sp:$Rn, simm9:$offset)>;
4461 // 32-bits to 32-bits are handled in target specific dag combine:
4462 // performIntToFpCombine.
4463 // 64-bits integer to 32-bits floating point, not possible with
4464 // SCVTF on floating point registers (both source and destination
4465 // must have the same size).
4467 // Here are the patterns for 8, 16, 32, and 64-bits to double.
4468 // 8-bits -> double. 3 size step-up: give up.
4469 // 16-bits -> double. 2 size step.
4470 class SExtLoadi16CVTf64Pat<dag addrmode, dag INST>
4471 : Pat <(f64 (sint_to_fp (i32 (sextloadi16 addrmode)))),
4472 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4477 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4483 dsub)))>, Requires<[NotForCodeSize]>;
4485 def : SExtLoadi16CVTf64Pat<(ro16.Wpat GPR64sp:$Rn, GPR32:$Rm, ro16.Wext:$ext),
4486 (LDRHroW GPR64sp:$Rn, GPR32:$Rm, ro16.Wext:$ext)>;
4487 def : SExtLoadi16CVTf64Pat<(ro16.Xpat GPR64sp:$Rn, GPR64:$Rm, ro16.Xext:$ext),
4488 (LDRHroX GPR64sp:$Rn, GPR64:$Rm, ro16.Xext:$ext)>;
4489 def : SExtLoadi16CVTf64Pat<(am_indexed16 GPR64sp:$Rn, uimm12s2:$offset),
4490 (LDRHui GPR64sp:$Rn, uimm12s2:$offset)>;
4491 def : SExtLoadi16CVTf64Pat<(am_unscaled16 GPR64sp:$Rn, simm9:$offset),
4492 (LDURHi GPR64sp:$Rn, simm9:$offset)>;
4493 // 32-bits -> double. 1 size step-up.
4494 class SExtLoadi32CVTf64Pat<dag addrmode, dag INST>
4495 : Pat <(f64 (sint_to_fp (i32 (load addrmode)))),
4496 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4498 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4502 dsub)))>, Requires<[NotForCodeSize]>;
4504 def : SExtLoadi32CVTf64Pat<(ro32.Wpat GPR64sp:$Rn, GPR32:$Rm, ro32.Wext:$ext),
4505 (LDRSroW GPR64sp:$Rn, GPR32:$Rm, ro32.Wext:$ext)>;
4506 def : SExtLoadi32CVTf64Pat<(ro32.Xpat GPR64sp:$Rn, GPR64:$Rm, ro32.Xext:$ext),
4507 (LDRSroX GPR64sp:$Rn, GPR64:$Rm, ro32.Xext:$ext)>;
4508 def : SExtLoadi32CVTf64Pat<(am_indexed32 GPR64sp:$Rn, uimm12s4:$offset),
4509 (LDRSui GPR64sp:$Rn, uimm12s4:$offset)>;
4510 def : SExtLoadi32CVTf64Pat<(am_unscaled32 GPR64sp:$Rn, simm9:$offset),
4511 (LDURSi GPR64sp:$Rn, simm9:$offset)>;
4513 // 64-bits -> double are handled in target specific dag combine:
4514 // performIntToFpCombine.
4517 //----------------------------------------------------------------------------
4518 // AdvSIMD Load-Store Structure
4519 //----------------------------------------------------------------------------
4520 defm LD1 : SIMDLd1Multiple<"ld1">;
4521 defm LD2 : SIMDLd2Multiple<"ld2">;
4522 defm LD3 : SIMDLd3Multiple<"ld3">;
4523 defm LD4 : SIMDLd4Multiple<"ld4">;
4525 defm ST1 : SIMDSt1Multiple<"st1">;
4526 defm ST2 : SIMDSt2Multiple<"st2">;
4527 defm ST3 : SIMDSt3Multiple<"st3">;
4528 defm ST4 : SIMDSt4Multiple<"st4">;
4530 class Ld1Pat<ValueType ty, Instruction INST>
4531 : Pat<(ty (load GPR64sp:$Rn)), (INST GPR64sp:$Rn)>;
4533 def : Ld1Pat<v16i8, LD1Onev16b>;
4534 def : Ld1Pat<v8i16, LD1Onev8h>;
4535 def : Ld1Pat<v4i32, LD1Onev4s>;
4536 def : Ld1Pat<v2i64, LD1Onev2d>;
4537 def : Ld1Pat<v8i8, LD1Onev8b>;
4538 def : Ld1Pat<v4i16, LD1Onev4h>;
4539 def : Ld1Pat<v2i32, LD1Onev2s>;
4540 def : Ld1Pat<v1i64, LD1Onev1d>;
4542 class St1Pat<ValueType ty, Instruction INST>
4543 : Pat<(store ty:$Vt, GPR64sp:$Rn),
4544 (INST ty:$Vt, GPR64sp:$Rn)>;
4546 def : St1Pat<v16i8, ST1Onev16b>;
4547 def : St1Pat<v8i16, ST1Onev8h>;
4548 def : St1Pat<v4i32, ST1Onev4s>;
4549 def : St1Pat<v2i64, ST1Onev2d>;
4550 def : St1Pat<v8i8, ST1Onev8b>;
4551 def : St1Pat<v4i16, ST1Onev4h>;
4552 def : St1Pat<v2i32, ST1Onev2s>;
4553 def : St1Pat<v1i64, ST1Onev1d>;
4559 defm LD1R : SIMDLdR<0, 0b110, 0, "ld1r", "One", 1, 2, 4, 8>;
4560 defm LD2R : SIMDLdR<1, 0b110, 0, "ld2r", "Two", 2, 4, 8, 16>;
4561 defm LD3R : SIMDLdR<0, 0b111, 0, "ld3r", "Three", 3, 6, 12, 24>;
4562 defm LD4R : SIMDLdR<1, 0b111, 0, "ld4r", "Four", 4, 8, 16, 32>;
4563 let mayLoad = 1, neverHasSideEffects = 1 in {
4564 defm LD1 : SIMDLdSingleBTied<0, 0b000, "ld1", VecListOneb, GPR64pi1>;
4565 defm LD1 : SIMDLdSingleHTied<0, 0b010, 0, "ld1", VecListOneh, GPR64pi2>;
4566 defm LD1 : SIMDLdSingleSTied<0, 0b100, 0b00, "ld1", VecListOnes, GPR64pi4>;
4567 defm LD1 : SIMDLdSingleDTied<0, 0b100, 0b01, "ld1", VecListOned, GPR64pi8>;
4568 defm LD2 : SIMDLdSingleBTied<1, 0b000, "ld2", VecListTwob, GPR64pi2>;
4569 defm LD2 : SIMDLdSingleHTied<1, 0b010, 0, "ld2", VecListTwoh, GPR64pi4>;
4570 defm LD2 : SIMDLdSingleSTied<1, 0b100, 0b00, "ld2", VecListTwos, GPR64pi8>;
4571 defm LD2 : SIMDLdSingleDTied<1, 0b100, 0b01, "ld2", VecListTwod, GPR64pi16>;
4572 defm LD3 : SIMDLdSingleBTied<0, 0b001, "ld3", VecListThreeb, GPR64pi3>;
4573 defm LD3 : SIMDLdSingleHTied<0, 0b011, 0, "ld3", VecListThreeh, GPR64pi6>;
4574 defm LD3 : SIMDLdSingleSTied<0, 0b101, 0b00, "ld3", VecListThrees, GPR64pi12>;
4575 defm LD3 : SIMDLdSingleDTied<0, 0b101, 0b01, "ld3", VecListThreed, GPR64pi24>;
4576 defm LD4 : SIMDLdSingleBTied<1, 0b001, "ld4", VecListFourb, GPR64pi4>;
4577 defm LD4 : SIMDLdSingleHTied<1, 0b011, 0, "ld4", VecListFourh, GPR64pi8>;
4578 defm LD4 : SIMDLdSingleSTied<1, 0b101, 0b00, "ld4", VecListFours, GPR64pi16>;
4579 defm LD4 : SIMDLdSingleDTied<1, 0b101, 0b01, "ld4", VecListFourd, GPR64pi32>;
4582 def : Pat<(v8i8 (ARM64dup (i32 (extloadi8 GPR64sp:$Rn)))),
4583 (LD1Rv8b GPR64sp:$Rn)>;
4584 def : Pat<(v16i8 (ARM64dup (i32 (extloadi8 GPR64sp:$Rn)))),
4585 (LD1Rv16b GPR64sp:$Rn)>;
4586 def : Pat<(v4i16 (ARM64dup (i32 (extloadi16 GPR64sp:$Rn)))),
4587 (LD1Rv4h GPR64sp:$Rn)>;
4588 def : Pat<(v8i16 (ARM64dup (i32 (extloadi16 GPR64sp:$Rn)))),
4589 (LD1Rv8h GPR64sp:$Rn)>;
4590 def : Pat<(v2i32 (ARM64dup (i32 (load GPR64sp:$Rn)))),
4591 (LD1Rv2s GPR64sp:$Rn)>;
4592 def : Pat<(v4i32 (ARM64dup (i32 (load GPR64sp:$Rn)))),
4593 (LD1Rv4s GPR64sp:$Rn)>;
4594 def : Pat<(v2i64 (ARM64dup (i64 (load GPR64sp:$Rn)))),
4595 (LD1Rv2d GPR64sp:$Rn)>;
4596 def : Pat<(v1i64 (ARM64dup (i64 (load GPR64sp:$Rn)))),
4597 (LD1Rv1d GPR64sp:$Rn)>;
4598 // Grab the floating point version too
4599 def : Pat<(v2f32 (ARM64dup (f32 (load GPR64sp:$Rn)))),
4600 (LD1Rv2s GPR64sp:$Rn)>;
4601 def : Pat<(v4f32 (ARM64dup (f32 (load GPR64sp:$Rn)))),
4602 (LD1Rv4s GPR64sp:$Rn)>;
4603 def : Pat<(v2f64 (ARM64dup (f64 (load GPR64sp:$Rn)))),
4604 (LD1Rv2d GPR64sp:$Rn)>;
4605 def : Pat<(v1f64 (ARM64dup (f64 (load GPR64sp:$Rn)))),
4606 (LD1Rv1d GPR64sp:$Rn)>;
4608 class Ld1Lane128Pat<SDPatternOperator scalar_load, Operand VecIndex,
4609 ValueType VTy, ValueType STy, Instruction LD1>
4610 : Pat<(vector_insert (VTy VecListOne128:$Rd),
4611 (STy (scalar_load GPR64sp:$Rn)), VecIndex:$idx),
4612 (LD1 VecListOne128:$Rd, VecIndex:$idx, GPR64sp:$Rn)>;
4614 def : Ld1Lane128Pat<extloadi8, VectorIndexB, v16i8, i32, LD1i8>;
4615 def : Ld1Lane128Pat<extloadi16, VectorIndexH, v8i16, i32, LD1i16>;
4616 def : Ld1Lane128Pat<load, VectorIndexS, v4i32, i32, LD1i32>;
4617 def : Ld1Lane128Pat<load, VectorIndexS, v4f32, f32, LD1i32>;
4618 def : Ld1Lane128Pat<load, VectorIndexD, v2i64, i64, LD1i64>;
4619 def : Ld1Lane128Pat<load, VectorIndexD, v2f64, f64, LD1i64>;
4621 class Ld1Lane64Pat<SDPatternOperator scalar_load, Operand VecIndex,
4622 ValueType VTy, ValueType STy, Instruction LD1>
4623 : Pat<(vector_insert (VTy VecListOne64:$Rd),
4624 (STy (scalar_load GPR64sp:$Rn)), VecIndex:$idx),
4626 (LD1 (SUBREG_TO_REG (i32 0), VecListOne64:$Rd, dsub),
4627 VecIndex:$idx, GPR64sp:$Rn),
4630 def : Ld1Lane64Pat<extloadi8, VectorIndexB, v8i8, i32, LD1i8>;
4631 def : Ld1Lane64Pat<extloadi16, VectorIndexH, v4i16, i32, LD1i16>;
4632 def : Ld1Lane64Pat<load, VectorIndexS, v2i32, i32, LD1i32>;
4633 def : Ld1Lane64Pat<load, VectorIndexS, v2f32, f32, LD1i32>;
4636 defm LD1 : SIMDLdSt1SingleAliases<"ld1">;
4637 defm LD2 : SIMDLdSt2SingleAliases<"ld2">;
4638 defm LD3 : SIMDLdSt3SingleAliases<"ld3">;
4639 defm LD4 : SIMDLdSt4SingleAliases<"ld4">;
4642 defm ST1 : SIMDStSingleB<0, 0b000, "st1", VecListOneb, GPR64pi1>;
4643 defm ST1 : SIMDStSingleH<0, 0b010, 0, "st1", VecListOneh, GPR64pi2>;
4644 defm ST1 : SIMDStSingleS<0, 0b100, 0b00, "st1", VecListOnes, GPR64pi4>;
4645 defm ST1 : SIMDStSingleD<0, 0b100, 0b01, "st1", VecListOned, GPR64pi8>;
4647 let AddedComplexity = 15 in
4648 class St1Lane128Pat<SDPatternOperator scalar_store, Operand VecIndex,
4649 ValueType VTy, ValueType STy, Instruction ST1>
4651 (STy (vector_extract (VTy VecListOne128:$Vt), VecIndex:$idx)),
4653 (ST1 VecListOne128:$Vt, VecIndex:$idx, GPR64sp:$Rn)>;
4655 def : St1Lane128Pat<truncstorei8, VectorIndexB, v16i8, i32, ST1i8>;
4656 def : St1Lane128Pat<truncstorei16, VectorIndexH, v8i16, i32, ST1i16>;
4657 def : St1Lane128Pat<store, VectorIndexS, v4i32, i32, ST1i32>;
4658 def : St1Lane128Pat<store, VectorIndexS, v4f32, f32, ST1i32>;
4659 def : St1Lane128Pat<store, VectorIndexD, v2i64, i64, ST1i64>;
4660 def : St1Lane128Pat<store, VectorIndexD, v2f64, f64, ST1i64>;
4662 let AddedComplexity = 15 in
4663 class St1Lane64Pat<SDPatternOperator scalar_store, Operand VecIndex,
4664 ValueType VTy, ValueType STy, Instruction ST1>
4666 (STy (vector_extract (VTy VecListOne64:$Vt), VecIndex:$idx)),
4668 (ST1 (SUBREG_TO_REG (i32 0), VecListOne64:$Vt, dsub),
4669 VecIndex:$idx, GPR64sp:$Rn)>;
4671 def : St1Lane64Pat<truncstorei8, VectorIndexB, v8i8, i32, ST1i8>;
4672 def : St1Lane64Pat<truncstorei16, VectorIndexH, v4i16, i32, ST1i16>;
4673 def : St1Lane64Pat<store, VectorIndexS, v2i32, i32, ST1i32>;
4674 def : St1Lane64Pat<store, VectorIndexS, v2f32, f32, ST1i32>;
4676 multiclass St1LanePost64Pat<SDPatternOperator scalar_store, Operand VecIndex,
4677 ValueType VTy, ValueType STy, Instruction ST1,
4679 def : Pat<(scalar_store
4680 (STy (vector_extract (VTy VecListOne64:$Vt), VecIndex:$idx)),
4681 GPR64sp:$Rn, offset),
4682 (ST1 (SUBREG_TO_REG (i32 0), VecListOne64:$Vt, dsub),
4683 VecIndex:$idx, GPR64sp:$Rn, XZR)>;
4685 def : Pat<(scalar_store
4686 (STy (vector_extract (VTy VecListOne64:$Vt), VecIndex:$idx)),
4687 GPR64sp:$Rn, GPR64:$Rm),
4688 (ST1 (SUBREG_TO_REG (i32 0), VecListOne64:$Vt, dsub),
4689 VecIndex:$idx, GPR64sp:$Rn, $Rm)>;
4692 defm : St1LanePost64Pat<post_truncsti8, VectorIndexB, v8i8, i32, ST1i8_POST, 1>;
4693 defm : St1LanePost64Pat<post_truncsti16, VectorIndexH, v4i16, i32, ST1i16_POST,
4695 defm : St1LanePost64Pat<post_store, VectorIndexS, v2i32, i32, ST1i32_POST, 4>;
4696 defm : St1LanePost64Pat<post_store, VectorIndexS, v2f32, f32, ST1i32_POST, 4>;
4697 defm : St1LanePost64Pat<post_store, VectorIndexD, v1i64, i64, ST1i64_POST, 8>;
4698 defm : St1LanePost64Pat<post_store, VectorIndexD, v1f64, f64, ST1i64_POST, 8>;
4700 multiclass St1LanePost128Pat<SDPatternOperator scalar_store, Operand VecIndex,
4701 ValueType VTy, ValueType STy, Instruction ST1,
4703 def : Pat<(scalar_store
4704 (STy (vector_extract (VTy VecListOne128:$Vt), VecIndex:$idx)),
4705 GPR64sp:$Rn, offset),
4706 (ST1 VecListOne128:$Vt, VecIndex:$idx, GPR64sp:$Rn, XZR)>;
4708 def : Pat<(scalar_store
4709 (STy (vector_extract (VTy VecListOne128:$Vt), VecIndex:$idx)),
4710 GPR64sp:$Rn, GPR64:$Rm),
4711 (ST1 VecListOne128:$Vt, VecIndex:$idx, GPR64sp:$Rn, $Rm)>;
4714 defm : St1LanePost128Pat<post_truncsti8, VectorIndexB, v16i8, i32, ST1i8_POST,
4716 defm : St1LanePost128Pat<post_truncsti16, VectorIndexH, v8i16, i32, ST1i16_POST,
4718 defm : St1LanePost128Pat<post_store, VectorIndexS, v4i32, i32, ST1i32_POST, 4>;
4719 defm : St1LanePost128Pat<post_store, VectorIndexS, v4f32, f32, ST1i32_POST, 4>;
4720 defm : St1LanePost128Pat<post_store, VectorIndexD, v2i64, i64, ST1i64_POST, 8>;
4721 defm : St1LanePost128Pat<post_store, VectorIndexD, v2f64, f64, ST1i64_POST, 8>;
4723 let mayStore = 1, neverHasSideEffects = 1 in {
4724 defm ST2 : SIMDStSingleB<1, 0b000, "st2", VecListTwob, GPR64pi2>;
4725 defm ST2 : SIMDStSingleH<1, 0b010, 0, "st2", VecListTwoh, GPR64pi4>;
4726 defm ST2 : SIMDStSingleS<1, 0b100, 0b00, "st2", VecListTwos, GPR64pi8>;
4727 defm ST2 : SIMDStSingleD<1, 0b100, 0b01, "st2", VecListTwod, GPR64pi16>;
4728 defm ST3 : SIMDStSingleB<0, 0b001, "st3", VecListThreeb, GPR64pi3>;
4729 defm ST3 : SIMDStSingleH<0, 0b011, 0, "st3", VecListThreeh, GPR64pi6>;
4730 defm ST3 : SIMDStSingleS<0, 0b101, 0b00, "st3", VecListThrees, GPR64pi12>;
4731 defm ST3 : SIMDStSingleD<0, 0b101, 0b01, "st3", VecListThreed, GPR64pi24>;
4732 defm ST4 : SIMDStSingleB<1, 0b001, "st4", VecListFourb, GPR64pi4>;
4733 defm ST4 : SIMDStSingleH<1, 0b011, 0, "st4", VecListFourh, GPR64pi8>;
4734 defm ST4 : SIMDStSingleS<1, 0b101, 0b00, "st4", VecListFours, GPR64pi16>;
4735 defm ST4 : SIMDStSingleD<1, 0b101, 0b01, "st4", VecListFourd, GPR64pi32>;
4738 defm ST1 : SIMDLdSt1SingleAliases<"st1">;
4739 defm ST2 : SIMDLdSt2SingleAliases<"st2">;
4740 defm ST3 : SIMDLdSt3SingleAliases<"st3">;
4741 defm ST4 : SIMDLdSt4SingleAliases<"st4">;
4743 //----------------------------------------------------------------------------
4744 // Crypto extensions
4745 //----------------------------------------------------------------------------
4747 def AESErr : AESTiedInst<0b0100, "aese", int_arm64_crypto_aese>;
4748 def AESDrr : AESTiedInst<0b0101, "aesd", int_arm64_crypto_aesd>;
4749 def AESMCrr : AESInst< 0b0110, "aesmc", int_arm64_crypto_aesmc>;
4750 def AESIMCrr : AESInst< 0b0111, "aesimc", int_arm64_crypto_aesimc>;
4752 def SHA1Crrr : SHATiedInstQSV<0b000, "sha1c", int_arm64_crypto_sha1c>;
4753 def SHA1Prrr : SHATiedInstQSV<0b001, "sha1p", int_arm64_crypto_sha1p>;
4754 def SHA1Mrrr : SHATiedInstQSV<0b010, "sha1m", int_arm64_crypto_sha1m>;
4755 def SHA1SU0rrr : SHATiedInstVVV<0b011, "sha1su0", int_arm64_crypto_sha1su0>;
4756 def SHA256Hrrr : SHATiedInstQQV<0b100, "sha256h", int_arm64_crypto_sha256h>;
4757 def SHA256H2rrr : SHATiedInstQQV<0b101, "sha256h2",int_arm64_crypto_sha256h2>;
4758 def SHA256SU1rrr :SHATiedInstVVV<0b110, "sha256su1",int_arm64_crypto_sha256su1>;
4760 def SHA1Hrr : SHAInstSS< 0b0000, "sha1h", int_arm64_crypto_sha1h>;
4761 def SHA1SU1rr : SHATiedInstVV<0b0001, "sha1su1", int_arm64_crypto_sha1su1>;
4762 def SHA256SU0rr : SHATiedInstVV<0b0010, "sha256su0",int_arm64_crypto_sha256su0>;
4764 //----------------------------------------------------------------------------
4766 //----------------------------------------------------------------------------
4767 // FIXME: Like for X86, these should go in their own separate .td file.
4769 // Any instruction that defines a 32-bit result leaves the high half of the
4770 // register. Truncate can be lowered to EXTRACT_SUBREG. CopyFromReg may
4771 // be copying from a truncate. But any other 32-bit operation will zero-extend
4773 // FIXME: X86 also checks for CMOV here. Do we need something similar?
4774 def def32 : PatLeaf<(i32 GPR32:$src), [{
4775 return N->getOpcode() != ISD::TRUNCATE &&
4776 N->getOpcode() != TargetOpcode::EXTRACT_SUBREG &&
4777 N->getOpcode() != ISD::CopyFromReg;
4780 // In the case of a 32-bit def that is known to implicitly zero-extend,
4781 // we can use a SUBREG_TO_REG.
4782 def : Pat<(i64 (zext def32:$src)), (SUBREG_TO_REG (i64 0), GPR32:$src, sub_32)>;
4784 // For an anyext, we don't care what the high bits are, so we can perform an
4785 // INSERT_SUBREF into an IMPLICIT_DEF.
4786 def : Pat<(i64 (anyext GPR32:$src)),
4787 (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32)>;
4789 // When we need to explicitly zero-extend, we use an unsigned bitfield move
4790 // instruction (UBFM) on the enclosing super-reg.
4791 def : Pat<(i64 (zext GPR32:$src)),
4792 (UBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32), 0, 31)>;
4794 // To sign extend, we use a signed bitfield move instruction (SBFM) on the
4795 // containing super-reg.
4796 def : Pat<(i64 (sext GPR32:$src)),
4797 (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32), 0, 31)>;
4798 def : Pat<(i64 (sext_inreg GPR64:$src, i32)), (SBFMXri GPR64:$src, 0, 31)>;
4799 def : Pat<(i64 (sext_inreg GPR64:$src, i16)), (SBFMXri GPR64:$src, 0, 15)>;
4800 def : Pat<(i64 (sext_inreg GPR64:$src, i8)), (SBFMXri GPR64:$src, 0, 7)>;
4801 def : Pat<(i64 (sext_inreg GPR64:$src, i1)), (SBFMXri GPR64:$src, 0, 0)>;
4802 def : Pat<(i32 (sext_inreg GPR32:$src, i16)), (SBFMWri GPR32:$src, 0, 15)>;
4803 def : Pat<(i32 (sext_inreg GPR32:$src, i8)), (SBFMWri GPR32:$src, 0, 7)>;
4804 def : Pat<(i32 (sext_inreg GPR32:$src, i1)), (SBFMWri GPR32:$src, 0, 0)>;
4806 def : Pat<(shl (sext_inreg GPR32:$Rn, i8), (i64 imm0_31:$imm)),
4807 (SBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
4808 (i64 (i32shift_sext_i8 imm0_31:$imm)))>;
4809 def : Pat<(shl (sext_inreg GPR64:$Rn, i8), (i64 imm0_63:$imm)),
4810 (SBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
4811 (i64 (i64shift_sext_i8 imm0_63:$imm)))>;
4813 def : Pat<(shl (sext_inreg GPR32:$Rn, i16), (i64 imm0_31:$imm)),
4814 (SBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
4815 (i64 (i32shift_sext_i16 imm0_31:$imm)))>;
4816 def : Pat<(shl (sext_inreg GPR64:$Rn, i16), (i64 imm0_63:$imm)),
4817 (SBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
4818 (i64 (i64shift_sext_i16 imm0_63:$imm)))>;
4820 def : Pat<(shl (i64 (sext GPR32:$Rn)), (i64 imm0_63:$imm)),
4821 (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$Rn, sub_32),
4822 (i64 (i64shift_a imm0_63:$imm)),
4823 (i64 (i64shift_sext_i32 imm0_63:$imm)))>;
4825 // sra patterns have an AddedComplexity of 10, so make sure we have a higher
4826 // AddedComplexity for the following patterns since we want to match sext + sra
4827 // patterns before we attempt to match a single sra node.
4828 let AddedComplexity = 20 in {
4829 // We support all sext + sra combinations which preserve at least one bit of the
4830 // original value which is to be sign extended. E.g. we support shifts up to
4832 def : Pat<(sra (sext_inreg GPR32:$Rn, i8), (i64 imm0_7:$imm)),
4833 (SBFMWri GPR32:$Rn, (i64 imm0_7:$imm), 7)>;
4834 def : Pat<(sra (sext_inreg GPR64:$Rn, i8), (i64 imm0_7:$imm)),
4835 (SBFMXri GPR64:$Rn, (i64 imm0_7:$imm), 7)>;
4837 def : Pat<(sra (sext_inreg GPR32:$Rn, i16), (i64 imm0_15:$imm)),
4838 (SBFMWri GPR32:$Rn, (i64 imm0_15:$imm), 15)>;
4839 def : Pat<(sra (sext_inreg GPR64:$Rn, i16), (i64 imm0_15:$imm)),
4840 (SBFMXri GPR64:$Rn, (i64 imm0_15:$imm), 15)>;
4842 def : Pat<(sra (i64 (sext GPR32:$Rn)), (i64 imm0_31:$imm)),
4843 (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$Rn, sub_32),
4844 (i64 imm0_31:$imm), 31)>;
4845 } // AddedComplexity = 20
4847 // To truncate, we can simply extract from a subregister.
4848 def : Pat<(i32 (trunc GPR64sp:$src)),
4849 (i32 (EXTRACT_SUBREG GPR64sp:$src, sub_32))>;
4851 // __builtin_trap() uses the BRK instruction on ARM64.
4852 def : Pat<(trap), (BRK 1)>;
4854 // Conversions within AdvSIMD types in the same register size are free.
4855 // But because we need a consistent lane ordering, in big endian many
4856 // conversions require one or more REV instructions.
4858 // Consider a simple memory load followed by a bitconvert then a store.
4860 // v1 = BITCAST v2i32 v0 to v4i16
4863 // In big endian mode every memory access has an implicit byte swap. LDR and
4864 // STR do a 64-bit byte swap, whereas LD1/ST1 do a byte swap per lane - that
4865 // is, they treat the vector as a sequence of elements to be byte-swapped.
4866 // The two pairs of instructions are fundamentally incompatible. We've decided
4867 // to use LD1/ST1 only to simplify compiler implementation.
4869 // LD1/ST1 perform the equivalent of a sequence of LDR/STR + REV. This makes
4870 // the original code sequence:
4872 // v1 = REV v2i32 (implicit)
4873 // v2 = BITCAST v2i32 v1 to v4i16
4874 // v3 = REV v4i16 v2 (implicit)
4877 // But this is now broken - the value stored is different to the value loaded
4878 // due to lane reordering. To fix this, on every BITCAST we must perform two
4881 // v1 = REV v2i32 (implicit)
4883 // v3 = BITCAST v2i32 v2 to v4i16
4885 // v5 = REV v4i16 v4 (implicit)
4888 // This means an extra two instructions, but actually in most cases the two REV
4889 // instructions can be combined into one. For example:
4890 // (REV64_2s (REV64_4h X)) === (REV32_4h X)
4892 // There is also no 128-bit REV instruction. This must be synthesized with an
4895 // Most bitconverts require some sort of conversion. The only exceptions are:
4896 // a) Identity conversions - vNfX <-> vNiX
4897 // b) Single-lane-to-scalar - v1fX <-> fX or v1iX <-> iX
4900 let Predicates = [IsLE] in {
4901 def : Pat<(v8i8 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
4902 def : Pat<(v4i16 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
4903 def : Pat<(v2i32 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
4904 def : Pat<(v2f32 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
4906 def : Pat<(i64 (bitconvert (v8i8 V64:$Vn))),
4907 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
4908 def : Pat<(i64 (bitconvert (v4i16 V64:$Vn))),
4909 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
4910 def : Pat<(i64 (bitconvert (v2i32 V64:$Vn))),
4911 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
4912 def : Pat<(i64 (bitconvert (v2f32 V64:$Vn))),
4913 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
4914 def : Pat<(i64 (bitconvert (v1f64 V64:$Vn))),
4915 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
4917 let Predicates = [IsBE] in {
4918 def : Pat<(v8i8 (bitconvert GPR64:$Xn)),
4919 (REV64v8i8 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
4920 def : Pat<(v4i16 (bitconvert GPR64:$Xn)),
4921 (REV64v4i16 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
4922 def : Pat<(v2i32 (bitconvert GPR64:$Xn)),
4923 (REV64v2i32 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
4924 def : Pat<(v2f32 (bitconvert GPR64:$Xn)),
4925 (REV64v2i32 (COPY_TO_REGCLASS GPR64:$Xn, FPR64))>;
4927 def : Pat<(i64 (bitconvert (v8i8 V64:$Vn))),
4928 (REV64v8i8 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
4929 def : Pat<(i64 (bitconvert (v4i16 V64:$Vn))),
4930 (REV64v4i16 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
4931 def : Pat<(i64 (bitconvert (v2i32 V64:$Vn))),
4932 (REV64v2i32 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
4933 def : Pat<(i64 (bitconvert (v2f32 V64:$Vn))),
4934 (REV64v2i32 (COPY_TO_REGCLASS V64:$Vn, GPR64))>;
4936 def : Pat<(v1i64 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
4937 def : Pat<(v1f64 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
4938 def : Pat<(i64 (bitconvert (v1i64 V64:$Vn))),
4939 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
4940 def : Pat<(v1i64 (scalar_to_vector GPR64:$Xn)),
4941 (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
4942 def : Pat<(v1f64 (scalar_to_vector GPR64:$Xn)),
4943 (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
4944 def : Pat<(v1f64 (scalar_to_vector (f64 FPR64:$Xn))), (v1f64 FPR64:$Xn)>;
4946 def : Pat<(f32 (bitconvert (i32 GPR32:$Xn))),
4947 (COPY_TO_REGCLASS GPR32:$Xn, FPR32)>;
4948 def : Pat<(i32 (bitconvert (f32 FPR32:$Xn))),
4949 (COPY_TO_REGCLASS FPR32:$Xn, GPR32)>;
4950 def : Pat<(f64 (bitconvert (i64 GPR64:$Xn))),
4951 (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
4952 def : Pat<(i64 (bitconvert (f64 FPR64:$Xn))),
4953 (COPY_TO_REGCLASS FPR64:$Xn, GPR64)>;
4954 def : Pat<(i64 (bitconvert (v1f64 V64:$Vn))),
4955 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
4957 let Predicates = [IsLE] in {
4958 def : Pat<(v1i64 (bitconvert (v2i32 FPR64:$src))), (v1i64 FPR64:$src)>;
4959 def : Pat<(v1i64 (bitconvert (v4i16 FPR64:$src))), (v1i64 FPR64:$src)>;
4960 def : Pat<(v1i64 (bitconvert (v8i8 FPR64:$src))), (v1i64 FPR64:$src)>;
4961 def : Pat<(v1i64 (bitconvert (v2f32 FPR64:$src))), (v1i64 FPR64:$src)>;
4963 let Predicates = [IsBE] in {
4964 def : Pat<(v1i64 (bitconvert (v2i32 FPR64:$src))),
4965 (v1i64 (REV64v2i32 FPR64:$src))>;
4966 def : Pat<(v1i64 (bitconvert (v4i16 FPR64:$src))),
4967 (v1i64 (REV64v4i16 FPR64:$src))>;
4968 def : Pat<(v1i64 (bitconvert (v8i8 FPR64:$src))),
4969 (v1i64 (REV64v8i8 FPR64:$src))>;
4970 def : Pat<(v1i64 (bitconvert (v2f32 FPR64:$src))),
4971 (v1i64 (REV64v2i32 FPR64:$src))>;
4973 def : Pat<(v1i64 (bitconvert (v1f64 FPR64:$src))), (v1i64 FPR64:$src)>;
4974 def : Pat<(v1i64 (bitconvert (f64 FPR64:$src))), (v1i64 FPR64:$src)>;
4976 let Predicates = [IsLE] in {
4977 def : Pat<(v2i32 (bitconvert (v1i64 FPR64:$src))), (v2i32 FPR64:$src)>;
4978 def : Pat<(v2i32 (bitconvert (v4i16 FPR64:$src))), (v2i32 FPR64:$src)>;
4979 def : Pat<(v2i32 (bitconvert (v8i8 FPR64:$src))), (v2i32 FPR64:$src)>;
4980 def : Pat<(v2i32 (bitconvert (f64 FPR64:$src))), (v2i32 FPR64:$src)>;
4981 def : Pat<(v2i32 (bitconvert (v1f64 FPR64:$src))), (v2i32 FPR64:$src)>;
4983 let Predicates = [IsBE] in {
4984 def : Pat<(v2i32 (bitconvert (v1i64 FPR64:$src))),
4985 (v2i32 (REV64v2i32 FPR64:$src))>;
4986 def : Pat<(v2i32 (bitconvert (v4i16 FPR64:$src))),
4987 (v2i32 (REV32v4i16 FPR64:$src))>;
4988 def : Pat<(v2i32 (bitconvert (v8i8 FPR64:$src))),
4989 (v2i32 (REV32v8i8 FPR64:$src))>;
4990 def : Pat<(v2i32 (bitconvert (f64 FPR64:$src))),
4991 (v2i32 (REV64v2i32 FPR64:$src))>;
4992 def : Pat<(v2i32 (bitconvert (v1f64 FPR64:$src))),
4993 (v2i32 (REV64v2i32 FPR64:$src))>;
4995 def : Pat<(v2i32 (bitconvert (v2f32 FPR64:$src))), (v2i32 FPR64:$src)>;
4997 let Predicates = [IsLE] in {
4998 def : Pat<(v4i16 (bitconvert (v1i64 FPR64:$src))), (v4i16 FPR64:$src)>;
4999 def : Pat<(v4i16 (bitconvert (v2i32 FPR64:$src))), (v4i16 FPR64:$src)>;
5000 def : Pat<(v4i16 (bitconvert (v8i8 FPR64:$src))), (v4i16 FPR64:$src)>;
5001 def : Pat<(v4i16 (bitconvert (f64 FPR64:$src))), (v4i16 FPR64:$src)>;
5002 def : Pat<(v4i16 (bitconvert (v2f32 FPR64:$src))), (v4i16 FPR64:$src)>;
5003 def : Pat<(v4i16 (bitconvert (v1f64 FPR64:$src))), (v4i16 FPR64:$src)>;
5005 let Predicates = [IsBE] in {
5006 def : Pat<(v4i16 (bitconvert (v1i64 FPR64:$src))),
5007 (v4i16 (REV64v4i16 FPR64:$src))>;
5008 def : Pat<(v4i16 (bitconvert (v2i32 FPR64:$src))),
5009 (v4i16 (REV32v4i16 FPR64:$src))>;
5010 def : Pat<(v4i16 (bitconvert (v8i8 FPR64:$src))),
5011 (v4i16 (REV16v8i8 FPR64:$src))>;
5012 def : Pat<(v4i16 (bitconvert (f64 FPR64:$src))),
5013 (v4i16 (REV64v4i16 FPR64:$src))>;
5014 def : Pat<(v4i16 (bitconvert (v2f32 FPR64:$src))),
5015 (v4i16 (REV32v4i16 FPR64:$src))>;
5016 def : Pat<(v4i16 (bitconvert (v1f64 FPR64:$src))),
5017 (v4i16 (REV64v4i16 FPR64:$src))>;
5020 let Predicates = [IsLE] in {
5021 def : Pat<(v8i8 (bitconvert (v1i64 FPR64:$src))), (v8i8 FPR64:$src)>;
5022 def : Pat<(v8i8 (bitconvert (v2i32 FPR64:$src))), (v8i8 FPR64:$src)>;
5023 def : Pat<(v8i8 (bitconvert (v4i16 FPR64:$src))), (v8i8 FPR64:$src)>;
5024 def : Pat<(v8i8 (bitconvert (f64 FPR64:$src))), (v8i8 FPR64:$src)>;
5025 def : Pat<(v8i8 (bitconvert (v2f32 FPR64:$src))), (v8i8 FPR64:$src)>;
5026 def : Pat<(v8i8 (bitconvert (v1f64 FPR64:$src))), (v8i8 FPR64:$src)>;
5028 let Predicates = [IsBE] in {
5029 def : Pat<(v8i8 (bitconvert (v1i64 FPR64:$src))),
5030 (v8i8 (REV64v8i8 FPR64:$src))>;
5031 def : Pat<(v8i8 (bitconvert (v2i32 FPR64:$src))),
5032 (v8i8 (REV32v8i8 FPR64:$src))>;
5033 def : Pat<(v8i8 (bitconvert (v4i16 FPR64:$src))),
5034 (v8i8 (REV16v8i8 FPR64:$src))>;
5035 def : Pat<(v8i8 (bitconvert (f64 FPR64:$src))),
5036 (v8i8 (REV64v8i8 FPR64:$src))>;
5037 def : Pat<(v8i8 (bitconvert (v2f32 FPR64:$src))),
5038 (v8i8 (REV32v8i8 FPR64:$src))>;
5039 def : Pat<(v8i8 (bitconvert (v1f64 FPR64:$src))),
5040 (v8i8 (REV64v8i8 FPR64:$src))>;
5043 let Predicates = [IsLE] in {
5044 def : Pat<(f64 (bitconvert (v2i32 FPR64:$src))), (f64 FPR64:$src)>;
5045 def : Pat<(f64 (bitconvert (v4i16 FPR64:$src))), (f64 FPR64:$src)>;
5046 def : Pat<(f64 (bitconvert (v2f32 FPR64:$src))), (f64 FPR64:$src)>;
5047 def : Pat<(f64 (bitconvert (v8i8 FPR64:$src))), (f64 FPR64:$src)>;
5049 let Predicates = [IsBE] in {
5050 def : Pat<(f64 (bitconvert (v2i32 FPR64:$src))),
5051 (f64 (REV64v2i32 FPR64:$src))>;
5052 def : Pat<(f64 (bitconvert (v4i16 FPR64:$src))),
5053 (f64 (REV64v4i16 FPR64:$src))>;
5054 def : Pat<(f64 (bitconvert (v2f32 FPR64:$src))),
5055 (f64 (REV64v2i32 FPR64:$src))>;
5056 def : Pat<(f64 (bitconvert (v8i8 FPR64:$src))),
5057 (f64 (REV64v8i8 FPR64:$src))>;
5059 def : Pat<(f64 (bitconvert (v1i64 FPR64:$src))), (f64 FPR64:$src)>;
5060 def : Pat<(f64 (bitconvert (v1f64 FPR64:$src))), (f64 FPR64:$src)>;
5062 let Predicates = [IsLE] in {
5063 def : Pat<(v1f64 (bitconvert (v2i32 FPR64:$src))), (v1f64 FPR64:$src)>;
5064 def : Pat<(v1f64 (bitconvert (v4i16 FPR64:$src))), (v1f64 FPR64:$src)>;
5065 def : Pat<(v1f64 (bitconvert (v8i8 FPR64:$src))), (v1f64 FPR64:$src)>;
5066 def : Pat<(v1f64 (bitconvert (v2f32 FPR64:$src))), (v1f64 FPR64:$src)>;
5068 let Predicates = [IsBE] in {
5069 def : Pat<(v1f64 (bitconvert (v2i32 FPR64:$src))),
5070 (v1f64 (REV64v2i32 FPR64:$src))>;
5071 def : Pat<(v1f64 (bitconvert (v4i16 FPR64:$src))),
5072 (v1f64 (REV64v4i16 FPR64:$src))>;
5073 def : Pat<(v1f64 (bitconvert (v8i8 FPR64:$src))),
5074 (v1f64 (REV64v8i8 FPR64:$src))>;
5075 def : Pat<(v1f64 (bitconvert (v2f32 FPR64:$src))),
5076 (v1f64 (REV64v2i32 FPR64:$src))>;
5078 def : Pat<(v1f64 (bitconvert (v1i64 FPR64:$src))), (v1f64 FPR64:$src)>;
5079 def : Pat<(v1f64 (bitconvert (f64 FPR64:$src))), (v1f64 FPR64:$src)>;
5081 let Predicates = [IsLE] in {
5082 def : Pat<(v2f32 (bitconvert (v1i64 FPR64:$src))), (v2f32 FPR64:$src)>;
5083 def : Pat<(v2f32 (bitconvert (v4i16 FPR64:$src))), (v2f32 FPR64:$src)>;
5084 def : Pat<(v2f32 (bitconvert (v8i8 FPR64:$src))), (v2f32 FPR64:$src)>;
5085 def : Pat<(v2f32 (bitconvert (v1f64 FPR64:$src))), (v2f32 FPR64:$src)>;
5086 def : Pat<(v2f32 (bitconvert (f64 FPR64:$src))), (v2f32 FPR64:$src)>;
5088 let Predicates = [IsBE] in {
5089 def : Pat<(v2f32 (bitconvert (v1i64 FPR64:$src))),
5090 (v2f32 (REV64v2i32 FPR64:$src))>;
5091 def : Pat<(v2f32 (bitconvert (v4i16 FPR64:$src))),
5092 (v2f32 (REV32v4i16 FPR64:$src))>;
5093 def : Pat<(v2f32 (bitconvert (v8i8 FPR64:$src))),
5094 (v2f32 (REV32v8i8 FPR64:$src))>;
5095 def : Pat<(v2f32 (bitconvert (v1f64 FPR64:$src))),
5096 (v2f32 (REV64v2i32 FPR64:$src))>;
5097 def : Pat<(v2f32 (bitconvert (f64 FPR64:$src))),
5098 (v2f32 (REV64v2i32 FPR64:$src))>;
5100 def : Pat<(v2f32 (bitconvert (v2i32 FPR64:$src))), (v2f32 FPR64:$src)>;
5102 let Predicates = [IsLE] in {
5103 def : Pat<(f128 (bitconvert (v2i64 FPR128:$src))), (f128 FPR128:$src)>;
5104 def : Pat<(f128 (bitconvert (v4i32 FPR128:$src))), (f128 FPR128:$src)>;
5105 def : Pat<(f128 (bitconvert (v8i16 FPR128:$src))), (f128 FPR128:$src)>;
5106 def : Pat<(f128 (bitconvert (v2f64 FPR128:$src))), (f128 FPR128:$src)>;
5107 def : Pat<(f128 (bitconvert (v4f32 FPR128:$src))), (f128 FPR128:$src)>;
5108 def : Pat<(f128 (bitconvert (v16i8 FPR128:$src))), (f128 FPR128:$src)>;
5110 let Predicates = [IsBE] in {
5111 def : Pat<(f128 (bitconvert (v2i64 FPR128:$src))),
5112 (f128 (EXTv16i8 FPR128:$src, FPR128:$src, (i32 8)))>;
5113 def : Pat<(f128 (bitconvert (v4i32 FPR128:$src))),
5114 (f128 (EXTv16i8 (REV64v4i32 FPR128:$src),
5115 (REV64v4i32 FPR128:$src), (i32 8)))>;
5116 def : Pat<(f128 (bitconvert (v8i16 FPR128:$src))),
5117 (f128 (EXTv16i8 (REV64v8i16 FPR128:$src),
5118 (REV64v8i16 FPR128:$src), (i32 8)))>;
5119 def : Pat<(f128 (bitconvert (v2f64 FPR128:$src))),
5120 (f128 (EXTv16i8 FPR128:$src, FPR128:$src, (i32 8)))>;
5121 def : Pat<(f128 (bitconvert (v4f32 FPR128:$src))),
5122 (f128 (EXTv16i8 (REV64v4i32 FPR128:$src),
5123 (REV64v4i32 FPR128:$src), (i32 8)))>;
5124 def : Pat<(f128 (bitconvert (v16i8 FPR128:$src))),
5125 (f128 (EXTv16i8 (REV64v16i8 FPR128:$src),
5126 (REV64v16i8 FPR128:$src), (i32 8)))>;
5129 let Predicates = [IsLE] in {
5130 def : Pat<(v2f64 (bitconvert (f128 FPR128:$src))), (v2f64 FPR128:$src)>;
5131 def : Pat<(v2f64 (bitconvert (v4i32 FPR128:$src))), (v2f64 FPR128:$src)>;
5132 def : Pat<(v2f64 (bitconvert (v8i16 FPR128:$src))), (v2f64 FPR128:$src)>;
5133 def : Pat<(v2f64 (bitconvert (v16i8 FPR128:$src))), (v2f64 FPR128:$src)>;
5134 def : Pat<(v2f64 (bitconvert (v4f32 FPR128:$src))), (v2f64 FPR128:$src)>;
5136 let Predicates = [IsBE] in {
5137 def : Pat<(v2f64 (bitconvert (f128 FPR128:$src))),
5138 (v2f64 (EXTv16i8 FPR128:$src,
5139 FPR128:$src, (i32 8)))>;
5140 def : Pat<(v2f64 (bitconvert (v4i32 FPR128:$src))),
5141 (v2f64 (REV64v4i32 FPR128:$src))>;
5142 def : Pat<(v2f64 (bitconvert (v8i16 FPR128:$src))),
5143 (v2f64 (REV64v8i16 FPR128:$src))>;
5144 def : Pat<(v2f64 (bitconvert (v16i8 FPR128:$src))),
5145 (v2f64 (REV64v16i8 FPR128:$src))>;
5146 def : Pat<(v2f64 (bitconvert (v4f32 FPR128:$src))),
5147 (v2f64 (REV64v4i32 FPR128:$src))>;
5149 def : Pat<(v2f64 (bitconvert (v2i64 FPR128:$src))), (v2f64 FPR128:$src)>;
5151 let Predicates = [IsLE] in {
5152 def : Pat<(v4f32 (bitconvert (f128 FPR128:$src))), (v4f32 FPR128:$src)>;
5153 def : Pat<(v4f32 (bitconvert (v8i16 FPR128:$src))), (v4f32 FPR128:$src)>;
5154 def : Pat<(v4f32 (bitconvert (v16i8 FPR128:$src))), (v4f32 FPR128:$src)>;
5155 def : Pat<(v4f32 (bitconvert (v2i64 FPR128:$src))), (v4f32 FPR128:$src)>;
5156 def : Pat<(v4f32 (bitconvert (v2f64 FPR128:$src))), (v4f32 FPR128:$src)>;
5158 let Predicates = [IsBE] in {
5159 def : Pat<(v4f32 (bitconvert (f128 FPR128:$src))),
5160 (v4f32 (EXTv16i8 (REV64v4i32 FPR128:$src),
5161 (REV64v4i32 FPR128:$src), (i32 8)))>;
5162 def : Pat<(v4f32 (bitconvert (v8i16 FPR128:$src))),
5163 (v4f32 (REV32v8i16 FPR128:$src))>;
5164 def : Pat<(v4f32 (bitconvert (v16i8 FPR128:$src))),
5165 (v4f32 (REV32v16i8 FPR128:$src))>;
5166 def : Pat<(v4f32 (bitconvert (v2i64 FPR128:$src))),
5167 (v4f32 (REV64v4i32 FPR128:$src))>;
5168 def : Pat<(v4f32 (bitconvert (v2f64 FPR128:$src))),
5169 (v4f32 (REV64v4i32 FPR128:$src))>;
5171 def : Pat<(v4f32 (bitconvert (v4i32 FPR128:$src))), (v4f32 FPR128:$src)>;
5173 let Predicates = [IsLE] in {
5174 def : Pat<(v2i64 (bitconvert (f128 FPR128:$src))), (v2i64 FPR128:$src)>;
5175 def : Pat<(v2i64 (bitconvert (v4i32 FPR128:$src))), (v2i64 FPR128:$src)>;
5176 def : Pat<(v2i64 (bitconvert (v8i16 FPR128:$src))), (v2i64 FPR128:$src)>;
5177 def : Pat<(v2i64 (bitconvert (v16i8 FPR128:$src))), (v2i64 FPR128:$src)>;
5178 def : Pat<(v2i64 (bitconvert (v4f32 FPR128:$src))), (v2i64 FPR128:$src)>;
5180 let Predicates = [IsBE] in {
5181 def : Pat<(v2i64 (bitconvert (f128 FPR128:$src))),
5182 (v2i64 (EXTv16i8 FPR128:$src,
5183 FPR128:$src, (i32 8)))>;
5184 def : Pat<(v2i64 (bitconvert (v4i32 FPR128:$src))),
5185 (v2i64 (REV64v4i32 FPR128:$src))>;
5186 def : Pat<(v2i64 (bitconvert (v8i16 FPR128:$src))),
5187 (v2i64 (REV64v8i16 FPR128:$src))>;
5188 def : Pat<(v2i64 (bitconvert (v16i8 FPR128:$src))),
5189 (v2i64 (REV64v16i8 FPR128:$src))>;
5190 def : Pat<(v2i64 (bitconvert (v4f32 FPR128:$src))),
5191 (v2i64 (REV64v4i32 FPR128:$src))>;
5193 def : Pat<(v2i64 (bitconvert (v2f64 FPR128:$src))), (v2i64 FPR128:$src)>;
5195 let Predicates = [IsLE] in {
5196 def : Pat<(v4i32 (bitconvert (f128 FPR128:$src))), (v4i32 FPR128:$src)>;
5197 def : Pat<(v4i32 (bitconvert (v2i64 FPR128:$src))), (v4i32 FPR128:$src)>;
5198 def : Pat<(v4i32 (bitconvert (v8i16 FPR128:$src))), (v4i32 FPR128:$src)>;
5199 def : Pat<(v4i32 (bitconvert (v16i8 FPR128:$src))), (v4i32 FPR128:$src)>;
5200 def : Pat<(v4i32 (bitconvert (v2f64 FPR128:$src))), (v4i32 FPR128:$src)>;
5202 let Predicates = [IsBE] in {
5203 def : Pat<(v4i32 (bitconvert (f128 FPR128:$src))),
5204 (v4i32 (EXTv16i8 (REV64v4i32 FPR128:$src),
5205 (REV64v4i32 FPR128:$src),
5207 def : Pat<(v4i32 (bitconvert (v2i64 FPR128:$src))),
5208 (v4i32 (REV64v4i32 FPR128:$src))>;
5209 def : Pat<(v4i32 (bitconvert (v8i16 FPR128:$src))),
5210 (v4i32 (REV32v8i16 FPR128:$src))>;
5211 def : Pat<(v4i32 (bitconvert (v16i8 FPR128:$src))),
5212 (v4i32 (REV32v16i8 FPR128:$src))>;
5213 def : Pat<(v4i32 (bitconvert (v2f64 FPR128:$src))),
5214 (v4i32 (REV64v4i32 FPR128:$src))>;
5216 def : Pat<(v4i32 (bitconvert (v4f32 FPR128:$src))), (v4i32 FPR128:$src)>;
5218 let Predicates = [IsLE] in {
5219 def : Pat<(v8i16 (bitconvert (f128 FPR128:$src))), (v8i16 FPR128:$src)>;
5220 def : Pat<(v8i16 (bitconvert (v2i64 FPR128:$src))), (v8i16 FPR128:$src)>;
5221 def : Pat<(v8i16 (bitconvert (v4i32 FPR128:$src))), (v8i16 FPR128:$src)>;
5222 def : Pat<(v8i16 (bitconvert (v16i8 FPR128:$src))), (v8i16 FPR128:$src)>;
5223 def : Pat<(v8i16 (bitconvert (v2f64 FPR128:$src))), (v8i16 FPR128:$src)>;
5224 def : Pat<(v8i16 (bitconvert (v4f32 FPR128:$src))), (v8i16 FPR128:$src)>;
5226 let Predicates = [IsBE] in {
5227 def : Pat<(v8i16 (bitconvert (f128 FPR128:$src))),
5228 (v8i16 (EXTv16i8 (REV64v8i16 FPR128:$src),
5229 (REV64v8i16 FPR128:$src),
5231 def : Pat<(v8i16 (bitconvert (v2i64 FPR128:$src))),
5232 (v8i16 (REV64v8i16 FPR128:$src))>;
5233 def : Pat<(v8i16 (bitconvert (v4i32 FPR128:$src))),
5234 (v8i16 (REV32v8i16 FPR128:$src))>;
5235 def : Pat<(v8i16 (bitconvert (v16i8 FPR128:$src))),
5236 (v8i16 (REV16v16i8 FPR128:$src))>;
5237 def : Pat<(v8i16 (bitconvert (v2f64 FPR128:$src))),
5238 (v8i16 (REV64v8i16 FPR128:$src))>;
5239 def : Pat<(v8i16 (bitconvert (v4f32 FPR128:$src))),
5240 (v8i16 (REV32v8i16 FPR128:$src))>;
5243 let Predicates = [IsLE] in {
5244 def : Pat<(v16i8 (bitconvert (f128 FPR128:$src))), (v16i8 FPR128:$src)>;
5245 def : Pat<(v16i8 (bitconvert (v2i64 FPR128:$src))), (v16i8 FPR128:$src)>;
5246 def : Pat<(v16i8 (bitconvert (v4i32 FPR128:$src))), (v16i8 FPR128:$src)>;
5247 def : Pat<(v16i8 (bitconvert (v8i16 FPR128:$src))), (v16i8 FPR128:$src)>;
5248 def : Pat<(v16i8 (bitconvert (v2f64 FPR128:$src))), (v16i8 FPR128:$src)>;
5249 def : Pat<(v16i8 (bitconvert (v4f32 FPR128:$src))), (v16i8 FPR128:$src)>;
5251 let Predicates = [IsBE] in {
5252 def : Pat<(v16i8 (bitconvert (f128 FPR128:$src))),
5253 (v16i8 (EXTv16i8 (REV64v16i8 FPR128:$src),
5254 (REV64v16i8 FPR128:$src),
5256 def : Pat<(v16i8 (bitconvert (v2i64 FPR128:$src))),
5257 (v16i8 (REV64v16i8 FPR128:$src))>;
5258 def : Pat<(v16i8 (bitconvert (v4i32 FPR128:$src))),
5259 (v16i8 (REV32v16i8 FPR128:$src))>;
5260 def : Pat<(v16i8 (bitconvert (v8i16 FPR128:$src))),
5261 (v16i8 (REV16v16i8 FPR128:$src))>;
5262 def : Pat<(v16i8 (bitconvert (v2f64 FPR128:$src))),
5263 (v16i8 (REV64v16i8 FPR128:$src))>;
5264 def : Pat<(v16i8 (bitconvert (v4f32 FPR128:$src))),
5265 (v16i8 (REV32v16i8 FPR128:$src))>;
5268 def : Pat<(v8i8 (extract_subvector (v16i8 FPR128:$Rn), (i64 1))),
5269 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
5270 def : Pat<(v4i16 (extract_subvector (v8i16 FPR128:$Rn), (i64 1))),
5271 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
5272 def : Pat<(v2i32 (extract_subvector (v4i32 FPR128:$Rn), (i64 1))),
5273 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
5274 def : Pat<(v1i64 (extract_subvector (v2i64 FPR128:$Rn), (i64 1))),
5275 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
5277 // A 64-bit subvector insert to the first 128-bit vector position
5278 // is a subregister copy that needs no instruction.
5279 def : Pat<(insert_subvector undef, (v1i64 FPR64:$src), (i32 0)),
5280 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5281 def : Pat<(insert_subvector undef, (v1f64 FPR64:$src), (i32 0)),
5282 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5283 def : Pat<(insert_subvector undef, (v2i32 FPR64:$src), (i32 0)),
5284 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5285 def : Pat<(insert_subvector undef, (v2f32 FPR64:$src), (i32 0)),
5286 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5287 def : Pat<(insert_subvector undef, (v4i16 FPR64:$src), (i32 0)),
5288 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5289 def : Pat<(insert_subvector undef, (v8i8 FPR64:$src), (i32 0)),
5290 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
5292 // Use pair-wise add instructions when summing up the lanes for v2f64, v2i64
5294 def : Pat<(i64 (add (vector_extract (v2i64 FPR128:$Rn), (i64 0)),
5295 (vector_extract (v2i64 FPR128:$Rn), (i64 1)))),
5296 (i64 (ADDPv2i64p (v2i64 FPR128:$Rn)))>;
5297 def : Pat<(f64 (fadd (vector_extract (v2f64 FPR128:$Rn), (i64 0)),
5298 (vector_extract (v2f64 FPR128:$Rn), (i64 1)))),
5299 (f64 (FADDPv2i64p (v2f64 FPR128:$Rn)))>;
5300 // vector_extract on 64-bit vectors gets promoted to a 128 bit vector,
5301 // so we match on v4f32 here, not v2f32. This will also catch adding
5302 // the low two lanes of a true v4f32 vector.
5303 def : Pat<(fadd (vector_extract (v4f32 FPR128:$Rn), (i64 0)),
5304 (vector_extract (v4f32 FPR128:$Rn), (i64 1))),
5305 (f32 (FADDPv2i32p (EXTRACT_SUBREG FPR128:$Rn, dsub)))>;
5307 // Scalar 64-bit shifts in FPR64 registers.
5308 def : Pat<(i64 (int_arm64_neon_sshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
5309 (SSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
5310 def : Pat<(i64 (int_arm64_neon_ushl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
5311 (USHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
5312 def : Pat<(i64 (int_arm64_neon_srshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
5313 (SRSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
5314 def : Pat<(i64 (int_arm64_neon_urshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
5315 (URSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
5317 // Tail call return handling. These are all compiler pseudo-instructions,
5318 // so no encoding information or anything like that.
5319 let isCall = 1, isTerminator = 1, isReturn = 1, isBarrier = 1, Uses = [SP] in {
5320 def TCRETURNdi : Pseudo<(outs), (ins i64imm:$dst, i32imm:$FPDiff),[]>;
5321 def TCRETURNri : Pseudo<(outs), (ins tcGPR64:$dst, i32imm:$FPDiff), []>;
5324 def : Pat<(ARM64tcret tcGPR64:$dst, (i32 timm:$FPDiff)),
5325 (TCRETURNri tcGPR64:$dst, imm:$FPDiff)>;
5326 def : Pat<(ARM64tcret tglobaladdr:$dst, (i32 timm:$FPDiff)),
5327 (TCRETURNdi texternalsym:$dst, imm:$FPDiff)>;
5328 def : Pat<(ARM64tcret texternalsym:$dst, (i32 timm:$FPDiff)),
5329 (TCRETURNdi texternalsym:$dst, imm:$FPDiff)>;
5331 include "ARM64InstrAtomics.td"