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, [SDTCisVT<0, i64>, SDTCisVT<1, i64>,
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, 1, [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 def : InstAlias<"movk $dst, $imm", (MOVKWi GPR32:$dst, imm0_65535:$imm, 0)>;
369 def : InstAlias<"movk $dst, $imm", (MOVKXi GPR64:$dst, imm0_65535:$imm, 0)>;
370 def : InstAlias<"movn $dst, $imm", (MOVNWi GPR32:$dst, imm0_65535:$imm, 0)>;
371 def : InstAlias<"movn $dst, $imm", (MOVNXi GPR64:$dst, imm0_65535:$imm, 0)>;
372 def : InstAlias<"movz $dst, $imm", (MOVZWi GPR32:$dst, imm0_65535:$imm, 0)>;
373 def : InstAlias<"movz $dst, $imm", (MOVZXi GPR64:$dst, imm0_65535:$imm, 0)>;
375 def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movz_symbol_g3:$sym, 48)>;
376 def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movz_symbol_g2:$sym, 32)>;
377 def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movz_symbol_g1:$sym, 16)>;
378 def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movz_symbol_g0:$sym, 0)>;
380 def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movz_symbol_g3:$sym, 48)>;
381 def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movz_symbol_g2:$sym, 32)>;
382 def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movz_symbol_g1:$sym, 16)>;
383 def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movz_symbol_g0:$sym, 0)>;
385 def : InstAlias<"movk $Rd, $sym", (MOVKXi GPR64:$Rd, movk_symbol_g3:$sym, 48)>;
386 def : InstAlias<"movk $Rd, $sym", (MOVKXi GPR64:$Rd, movk_symbol_g2:$sym, 32)>;
387 def : InstAlias<"movk $Rd, $sym", (MOVKXi GPR64:$Rd, movk_symbol_g1:$sym, 16)>;
388 def : InstAlias<"movk $Rd, $sym", (MOVKXi GPR64:$Rd, movk_symbol_g0:$sym, 0)>;
390 def : InstAlias<"movz $Rd, $sym", (MOVZWi GPR32:$Rd, movz_symbol_g1:$sym, 16)>;
391 def : InstAlias<"movz $Rd, $sym", (MOVZWi GPR32:$Rd, movz_symbol_g0:$sym, 0)>;
393 def : InstAlias<"movn $Rd, $sym", (MOVNWi GPR32:$Rd, movz_symbol_g1:$sym, 16)>;
394 def : InstAlias<"movn $Rd, $sym", (MOVNWi GPR32:$Rd, movz_symbol_g0:$sym, 0)>;
396 def : InstAlias<"movk $Rd, $sym", (MOVKWi GPR32:$Rd, movk_symbol_g1:$sym, 16)>;
397 def : InstAlias<"movk $Rd, $sym", (MOVKWi GPR32:$Rd, movk_symbol_g0:$sym, 0)>;
399 let isReMaterializable = 1, isCodeGenOnly = 1, isMoveImm = 1,
400 isAsCheapAsAMove = 1 in {
401 // FIXME: The following pseudo instructions are only needed because remat
402 // cannot handle multiple instructions. When that changes, we can select
403 // directly to the real instructions and get rid of these pseudos.
406 : Pseudo<(outs GPR32:$dst), (ins i32imm:$src),
407 [(set GPR32:$dst, imm:$src)]>,
410 : Pseudo<(outs GPR64:$dst), (ins i64imm:$src),
411 [(set GPR64:$dst, imm:$src)]>,
413 } // isReMaterializable, isCodeGenOnly
415 // If possible, we want to use MOVi32imm even for 64-bit moves. This gives the
416 // eventual expansion code fewer bits to worry about getting right. Marshalling
417 // the types is a little tricky though:
418 def i64imm_32bit : ImmLeaf<i64, [{
419 return (Imm & 0xffffffffULL) == static_cast<uint64_t>(Imm);
422 def trunc_imm : SDNodeXForm<imm, [{
423 return CurDAG->getTargetConstant(N->getZExtValue(), MVT::i32);
426 def : Pat<(i64 i64imm_32bit:$src),
427 (SUBREG_TO_REG (i64 0), (MOVi32imm (trunc_imm imm:$src)), sub_32)>;
429 // Deal with the various forms of (ELF) large addressing with MOVZ/MOVK
431 def : Pat<(ARM64WrapperLarge tglobaladdr:$g3, tglobaladdr:$g2,
432 tglobaladdr:$g1, tglobaladdr:$g0),
433 (MOVKXi (MOVKXi (MOVKXi (MOVZXi tglobaladdr:$g3, 48),
434 tglobaladdr:$g2, 32),
435 tglobaladdr:$g1, 16),
436 tglobaladdr:$g0, 0)>;
438 def : Pat<(ARM64WrapperLarge tblockaddress:$g3, tblockaddress:$g2,
439 tblockaddress:$g1, tblockaddress:$g0),
440 (MOVKXi (MOVKXi (MOVKXi (MOVZXi tblockaddress:$g3, 48),
441 tblockaddress:$g2, 32),
442 tblockaddress:$g1, 16),
443 tblockaddress:$g0, 0)>;
445 def : Pat<(ARM64WrapperLarge tconstpool:$g3, tconstpool:$g2,
446 tconstpool:$g1, tconstpool:$g0),
447 (MOVKXi (MOVKXi (MOVKXi (MOVZXi tconstpool:$g3, 48),
452 def : Pat<(ARM64WrapperLarge tjumptable:$g3, tjumptable:$g2,
453 tjumptable:$g1, tjumptable:$g0),
454 (MOVKXi (MOVKXi (MOVKXi (MOVZXi tjumptable:$g3, 48),
460 //===----------------------------------------------------------------------===//
461 // Arithmetic instructions.
462 //===----------------------------------------------------------------------===//
464 // Add/subtract with carry.
465 defm ADC : AddSubCarry<0, "adc", "adcs", ARM64adc, ARM64adc_flag>;
466 defm SBC : AddSubCarry<1, "sbc", "sbcs", ARM64sbc, ARM64sbc_flag>;
468 def : InstAlias<"ngc $dst, $src", (SBCWr GPR32:$dst, WZR, GPR32:$src)>;
469 def : InstAlias<"ngc $dst, $src", (SBCXr GPR64:$dst, XZR, GPR64:$src)>;
470 def : InstAlias<"ngcs $dst, $src", (SBCSWr GPR32:$dst, WZR, GPR32:$src)>;
471 def : InstAlias<"ngcs $dst, $src", (SBCSXr GPR64:$dst, XZR, GPR64:$src)>;
474 defm ADD : AddSub<0, "add", add>;
475 defm SUB : AddSub<1, "sub">;
477 defm ADDS : AddSubS<0, "adds", ARM64add_flag>;
478 defm SUBS : AddSubS<1, "subs", ARM64sub_flag>;
480 // Use SUBS instead of SUB to enable CSE between SUBS and SUB.
481 def : Pat<(sub GPR32sp:$Rn, addsub_shifted_imm32:$imm),
482 (SUBSWri GPR32sp:$Rn, addsub_shifted_imm32:$imm)>;
483 def : Pat<(sub GPR64sp:$Rn, addsub_shifted_imm64:$imm),
484 (SUBSXri GPR64sp:$Rn, addsub_shifted_imm64:$imm)>;
485 def : Pat<(sub GPR32:$Rn, GPR32:$Rm),
486 (SUBSWrr GPR32:$Rn, GPR32:$Rm)>;
487 def : Pat<(sub GPR64:$Rn, GPR64:$Rm),
488 (SUBSXrr GPR64:$Rn, GPR64:$Rm)>;
489 def : Pat<(sub GPR32:$Rn, arith_shifted_reg32:$Rm),
490 (SUBSWrs GPR32:$Rn, arith_shifted_reg32:$Rm)>;
491 def : Pat<(sub GPR64:$Rn, arith_shifted_reg64:$Rm),
492 (SUBSXrs GPR64:$Rn, arith_shifted_reg64:$Rm)>;
493 def : Pat<(sub GPR32sp:$R2, arith_extended_reg32<i32>:$R3),
494 (SUBSWrx GPR32sp:$R2, arith_extended_reg32<i32>:$R3)>;
495 def : Pat<(sub GPR64sp:$R2, arith_extended_reg32to64<i64>:$R3),
496 (SUBSXrx GPR64sp:$R2, arith_extended_reg32to64<i64>:$R3)>;
498 // Because of the immediate format for add/sub-imm instructions, the
499 // expression (add x, -1) must be transformed to (SUB{W,X}ri x, 1).
500 // These patterns capture that transformation.
501 let AddedComplexity = 1 in {
502 def : Pat<(add GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
503 (SUBSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
504 def : Pat<(add GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
505 (SUBSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
506 def : Pat<(sub GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
507 (ADDWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
508 def : Pat<(sub GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
509 (ADDXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
512 def : InstAlias<"neg $dst, $src", (SUBWrs GPR32:$dst, WZR, GPR32:$src, 0)>;
513 def : InstAlias<"neg $dst, $src", (SUBXrs GPR64:$dst, XZR, GPR64:$src, 0)>;
514 def : InstAlias<"neg $dst, $src, $shift",
515 (SUBWrs GPR32:$dst, WZR, GPR32:$src, arith_shift:$shift)>;
516 def : InstAlias<"neg $dst, $src, $shift",
517 (SUBXrs GPR64:$dst, XZR, GPR64:$src, arith_shift:$shift)>;
519 // Because of the immediate format for add/sub-imm instructions, the
520 // expression (add x, -1) must be transformed to (SUB{W,X}ri x, 1).
521 // These patterns capture that transformation.
522 let AddedComplexity = 1 in {
523 def : Pat<(ARM64add_flag GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
524 (SUBSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
525 def : Pat<(ARM64add_flag GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
526 (SUBSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
527 def : Pat<(ARM64sub_flag GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
528 (ADDSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
529 def : Pat<(ARM64sub_flag GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
530 (ADDSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
533 def : InstAlias<"negs $dst, $src", (SUBSWrs GPR32:$dst, WZR, GPR32:$src, 0)>;
534 def : InstAlias<"negs $dst, $src", (SUBSXrs GPR64:$dst, XZR, GPR64:$src, 0)>;
535 def : InstAlias<"negs $dst, $src, $shift",
536 (SUBSWrs GPR32:$dst, WZR, GPR32:$src, arith_shift:$shift)>;
537 def : InstAlias<"negs $dst, $src, $shift",
538 (SUBSXrs GPR64:$dst, XZR, GPR64:$src, arith_shift:$shift)>;
540 // Unsigned/Signed divide
541 defm UDIV : Div<0, "udiv", udiv>;
542 defm SDIV : Div<1, "sdiv", sdiv>;
543 let isCodeGenOnly = 1 in {
544 defm UDIV_Int : Div<0, "udiv", int_arm64_udiv>;
545 defm SDIV_Int : Div<1, "sdiv", int_arm64_sdiv>;
549 defm ASRV : Shift<0b10, "asr", sra>;
550 defm LSLV : Shift<0b00, "lsl", shl>;
551 defm LSRV : Shift<0b01, "lsr", srl>;
552 defm RORV : Shift<0b11, "ror", rotr>;
554 def : ShiftAlias<"asrv", ASRVWr, GPR32>;
555 def : ShiftAlias<"asrv", ASRVXr, GPR64>;
556 def : ShiftAlias<"lslv", LSLVWr, GPR32>;
557 def : ShiftAlias<"lslv", LSLVXr, GPR64>;
558 def : ShiftAlias<"lsrv", LSRVWr, GPR32>;
559 def : ShiftAlias<"lsrv", LSRVXr, GPR64>;
560 def : ShiftAlias<"rorv", RORVWr, GPR32>;
561 def : ShiftAlias<"rorv", RORVXr, GPR64>;
564 let AddedComplexity = 7 in {
565 defm MADD : MulAccum<0, "madd", add>;
566 defm MSUB : MulAccum<1, "msub", sub>;
568 def : Pat<(i32 (mul GPR32:$Rn, GPR32:$Rm)),
569 (MADDWrrr GPR32:$Rn, GPR32:$Rm, WZR)>;
570 def : Pat<(i64 (mul GPR64:$Rn, GPR64:$Rm)),
571 (MADDXrrr GPR64:$Rn, GPR64:$Rm, XZR)>;
573 def : Pat<(i32 (ineg (mul GPR32:$Rn, GPR32:$Rm))),
574 (MSUBWrrr GPR32:$Rn, GPR32:$Rm, WZR)>;
575 def : Pat<(i64 (ineg (mul GPR64:$Rn, GPR64:$Rm))),
576 (MSUBXrrr GPR64:$Rn, GPR64:$Rm, XZR)>;
577 } // AddedComplexity = 7
579 let AddedComplexity = 5 in {
580 def SMADDLrrr : WideMulAccum<0, 0b001, "smaddl", add, sext>;
581 def SMSUBLrrr : WideMulAccum<1, 0b001, "smsubl", sub, sext>;
582 def UMADDLrrr : WideMulAccum<0, 0b101, "umaddl", add, zext>;
583 def UMSUBLrrr : WideMulAccum<1, 0b101, "umsubl", sub, zext>;
585 def : Pat<(i64 (mul (sext GPR32:$Rn), (sext GPR32:$Rm))),
586 (SMADDLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
587 def : Pat<(i64 (mul (zext GPR32:$Rn), (zext GPR32:$Rm))),
588 (UMADDLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
590 def : Pat<(i64 (ineg (mul (sext GPR32:$Rn), (sext GPR32:$Rm)))),
591 (SMSUBLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
592 def : Pat<(i64 (ineg (mul (zext GPR32:$Rn), (zext GPR32:$Rm)))),
593 (UMSUBLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
594 } // AddedComplexity = 5
596 def : MulAccumWAlias<"mul", MADDWrrr>;
597 def : MulAccumXAlias<"mul", MADDXrrr>;
598 def : MulAccumWAlias<"mneg", MSUBWrrr>;
599 def : MulAccumXAlias<"mneg", MSUBXrrr>;
600 def : WideMulAccumAlias<"smull", SMADDLrrr>;
601 def : WideMulAccumAlias<"smnegl", SMSUBLrrr>;
602 def : WideMulAccumAlias<"umull", UMADDLrrr>;
603 def : WideMulAccumAlias<"umnegl", UMSUBLrrr>;
606 def SMULHrr : MulHi<0b010, "smulh", mulhs>;
607 def UMULHrr : MulHi<0b110, "umulh", mulhu>;
610 def CRC32Brr : BaseCRC32<0, 0b00, 0, GPR32, int_arm64_crc32b, "crc32b">;
611 def CRC32Hrr : BaseCRC32<0, 0b01, 0, GPR32, int_arm64_crc32h, "crc32h">;
612 def CRC32Wrr : BaseCRC32<0, 0b10, 0, GPR32, int_arm64_crc32w, "crc32w">;
613 def CRC32Xrr : BaseCRC32<1, 0b11, 0, GPR64, int_arm64_crc32x, "crc32x">;
615 def CRC32CBrr : BaseCRC32<0, 0b00, 1, GPR32, int_arm64_crc32cb, "crc32cb">;
616 def CRC32CHrr : BaseCRC32<0, 0b01, 1, GPR32, int_arm64_crc32ch, "crc32ch">;
617 def CRC32CWrr : BaseCRC32<0, 0b10, 1, GPR32, int_arm64_crc32cw, "crc32cw">;
618 def CRC32CXrr : BaseCRC32<1, 0b11, 1, GPR64, int_arm64_crc32cx, "crc32cx">;
621 //===----------------------------------------------------------------------===//
622 // Logical instructions.
623 //===----------------------------------------------------------------------===//
626 defm ANDS : LogicalImmS<0b11, "ands", ARM64and_flag>;
627 defm AND : LogicalImm<0b00, "and", and>;
628 defm EOR : LogicalImm<0b10, "eor", xor>;
629 defm ORR : LogicalImm<0b01, "orr", or>;
631 def : InstAlias<"mov $dst, $imm", (ORRWri GPR32sp:$dst, WZR,
632 logical_imm32:$imm)>;
633 def : InstAlias<"mov $dst, $imm", (ORRXri GPR64sp:$dst, XZR,
634 logical_imm64:$imm)>;
638 defm ANDS : LogicalRegS<0b11, 0, "ands", ARM64and_flag>;
639 defm BICS : LogicalRegS<0b11, 1, "bics",
640 BinOpFrag<(ARM64and_flag node:$LHS, (not node:$RHS))>>;
641 defm AND : LogicalReg<0b00, 0, "and", and>;
642 defm BIC : LogicalReg<0b00, 1, "bic",
643 BinOpFrag<(and node:$LHS, (not node:$RHS))>>;
644 defm EON : LogicalReg<0b10, 1, "eon",
645 BinOpFrag<(xor node:$LHS, (not node:$RHS))>>;
646 defm EOR : LogicalReg<0b10, 0, "eor", xor>;
647 defm ORN : LogicalReg<0b01, 1, "orn",
648 BinOpFrag<(or node:$LHS, (not node:$RHS))>>;
649 defm ORR : LogicalReg<0b01, 0, "orr", or>;
651 def : InstAlias<"tst $src1, $src2",
652 (ANDSWri WZR, GPR32:$src1, logical_imm32:$src2)>;
653 def : InstAlias<"tst $src1, $src2",
654 (ANDSXri XZR, GPR64:$src1, logical_imm64:$src2)>;
656 def : InstAlias<"tst $src1, $src2",
657 (ANDSWrs WZR, GPR32:$src1, GPR32:$src2, 0)>;
658 def : InstAlias<"tst $src1, $src2",
659 (ANDSXrs XZR, GPR64:$src1, GPR64:$src2, 0)>;
661 def : InstAlias<"tst $src1, $src2, $sh",
662 (ANDSWrs WZR, GPR32:$src1, GPR32:$src2, logical_shift:$sh)>;
663 def : InstAlias<"tst $src1, $src2, $sh",
664 (ANDSXrs XZR, GPR64:$src1, GPR64:$src2, logical_shift:$sh)>;
666 def : InstAlias<"mvn $Wd, $Wm",
667 (ORNWrs GPR32:$Wd, WZR, GPR32:$Wm, 0)>;
668 def : InstAlias<"mvn $Xd, $Xm",
669 (ORNXrs GPR64:$Xd, XZR, GPR64:$Xm, 0)>;
671 def : InstAlias<"mvn $Wd, $Wm, $sh",
672 (ORNWrs GPR32:$Wd, WZR, GPR32:$Wm, logical_shift:$sh)>;
673 def : InstAlias<"mvn $Xd, $Xm, $sh",
674 (ORNXrs GPR64:$Xd, XZR, GPR64:$Xm, logical_shift:$sh)>;
676 def : Pat<(not GPR32:$Wm), (ORNWrr WZR, GPR32:$Wm)>;
677 def : Pat<(not GPR64:$Xm), (ORNXrr XZR, GPR64:$Xm)>;
680 //===----------------------------------------------------------------------===//
681 // One operand data processing instructions.
682 //===----------------------------------------------------------------------===//
684 defm CLS : OneOperandData<0b101, "cls">;
685 defm CLZ : OneOperandData<0b100, "clz", ctlz>;
686 defm RBIT : OneOperandData<0b000, "rbit">;
687 def REV16Wr : OneWRegData<0b001, "rev16",
688 UnOpFrag<(rotr (bswap node:$LHS), (i64 16))>>;
689 def REV16Xr : OneXRegData<0b001, "rev16", null_frag>;
691 def : Pat<(cttz GPR32:$Rn),
692 (CLZWr (RBITWr GPR32:$Rn))>;
693 def : Pat<(cttz GPR64:$Rn),
694 (CLZXr (RBITXr GPR64:$Rn))>;
695 def : Pat<(ctlz (or (shl (xor (sra GPR32:$Rn, (i64 31)), GPR32:$Rn), (i64 1)),
698 def : Pat<(ctlz (or (shl (xor (sra GPR64:$Rn, (i64 63)), GPR64:$Rn), (i64 1)),
702 // Unlike the other one operand instructions, the instructions with the "rev"
703 // mnemonic do *not* just different in the size bit, but actually use different
704 // opcode bits for the different sizes.
705 def REVWr : OneWRegData<0b010, "rev", bswap>;
706 def REVXr : OneXRegData<0b011, "rev", bswap>;
707 def REV32Xr : OneXRegData<0b010, "rev32",
708 UnOpFrag<(rotr (bswap node:$LHS), (i64 32))>>;
710 // The bswap commutes with the rotr so we want a pattern for both possible
712 def : Pat<(bswap (rotr GPR32:$Rn, (i64 16))), (REV16Wr GPR32:$Rn)>;
713 def : Pat<(bswap (rotr GPR64:$Rn, (i64 32))), (REV32Xr GPR64:$Rn)>;
715 //===----------------------------------------------------------------------===//
716 // Bitfield immediate extraction instruction.
717 //===----------------------------------------------------------------------===//
718 let neverHasSideEffects = 1 in
719 defm EXTR : ExtractImm<"extr">;
720 def : InstAlias<"ror $dst, $src, $shift",
721 (EXTRWrri GPR32:$dst, GPR32:$src, GPR32:$src, imm0_31:$shift)>;
722 def : InstAlias<"ror $dst, $src, $shift",
723 (EXTRXrri GPR64:$dst, GPR64:$src, GPR64:$src, imm0_63:$shift)>;
725 def : Pat<(rotr GPR32:$Rn, (i64 imm0_31:$imm)),
726 (EXTRWrri GPR32:$Rn, GPR32:$Rn, imm0_31:$imm)>;
727 def : Pat<(rotr GPR64:$Rn, (i64 imm0_63:$imm)),
728 (EXTRXrri GPR64:$Rn, GPR64:$Rn, imm0_63:$imm)>;
730 //===----------------------------------------------------------------------===//
731 // Other bitfield immediate instructions.
732 //===----------------------------------------------------------------------===//
733 let neverHasSideEffects = 1 in {
734 defm BFM : BitfieldImmWith2RegArgs<0b01, "bfm">;
735 defm SBFM : BitfieldImm<0b00, "sbfm">;
736 defm UBFM : BitfieldImm<0b10, "ubfm">;
739 def i32shift_a : Operand<i64>, SDNodeXForm<imm, [{
740 uint64_t enc = (32 - N->getZExtValue()) & 0x1f;
741 return CurDAG->getTargetConstant(enc, MVT::i64);
744 def i32shift_b : Operand<i64>, SDNodeXForm<imm, [{
745 uint64_t enc = 31 - N->getZExtValue();
746 return CurDAG->getTargetConstant(enc, MVT::i64);
749 // min(7, 31 - shift_amt)
750 def i32shift_sext_i8 : Operand<i64>, SDNodeXForm<imm, [{
751 uint64_t enc = 31 - N->getZExtValue();
752 enc = enc > 7 ? 7 : enc;
753 return CurDAG->getTargetConstant(enc, MVT::i64);
756 // min(15, 31 - shift_amt)
757 def i32shift_sext_i16 : Operand<i64>, SDNodeXForm<imm, [{
758 uint64_t enc = 31 - N->getZExtValue();
759 enc = enc > 15 ? 15 : enc;
760 return CurDAG->getTargetConstant(enc, MVT::i64);
763 def i64shift_a : Operand<i64>, SDNodeXForm<imm, [{
764 uint64_t enc = (64 - N->getZExtValue()) & 0x3f;
765 return CurDAG->getTargetConstant(enc, MVT::i64);
768 def i64shift_b : Operand<i64>, SDNodeXForm<imm, [{
769 uint64_t enc = 63 - N->getZExtValue();
770 return CurDAG->getTargetConstant(enc, MVT::i64);
773 // min(7, 63 - shift_amt)
774 def i64shift_sext_i8 : Operand<i64>, SDNodeXForm<imm, [{
775 uint64_t enc = 63 - N->getZExtValue();
776 enc = enc > 7 ? 7 : enc;
777 return CurDAG->getTargetConstant(enc, MVT::i64);
780 // min(15, 63 - shift_amt)
781 def i64shift_sext_i16 : Operand<i64>, SDNodeXForm<imm, [{
782 uint64_t enc = 63 - N->getZExtValue();
783 enc = enc > 15 ? 15 : enc;
784 return CurDAG->getTargetConstant(enc, MVT::i64);
787 // min(31, 63 - shift_amt)
788 def i64shift_sext_i32 : Operand<i64>, SDNodeXForm<imm, [{
789 uint64_t enc = 63 - N->getZExtValue();
790 enc = enc > 31 ? 31 : enc;
791 return CurDAG->getTargetConstant(enc, MVT::i64);
794 def : Pat<(shl GPR32:$Rn, (i64 imm0_31:$imm)),
795 (UBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
796 (i64 (i32shift_b imm0_31:$imm)))>;
797 def : Pat<(shl GPR64:$Rn, (i64 imm0_63:$imm)),
798 (UBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
799 (i64 (i64shift_b imm0_63:$imm)))>;
801 let AddedComplexity = 10 in {
802 def : Pat<(sra GPR32:$Rn, (i64 imm0_31:$imm)),
803 (SBFMWri GPR32:$Rn, imm0_31:$imm, 31)>;
804 def : Pat<(sra GPR64:$Rn, (i64 imm0_63:$imm)),
805 (SBFMXri GPR64:$Rn, imm0_63:$imm, 63)>;
808 def : InstAlias<"asr $dst, $src, $shift",
809 (SBFMWri GPR32:$dst, GPR32:$src, imm0_31:$shift, 31)>;
810 def : InstAlias<"asr $dst, $src, $shift",
811 (SBFMXri GPR64:$dst, GPR64:$src, imm0_63:$shift, 63)>;
812 def : InstAlias<"sxtb $dst, $src", (SBFMWri GPR32:$dst, GPR32:$src, 0, 7)>;
813 def : InstAlias<"sxtb $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 7)>;
814 def : InstAlias<"sxth $dst, $src", (SBFMWri GPR32:$dst, GPR32:$src, 0, 15)>;
815 def : InstAlias<"sxth $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 15)>;
816 def : InstAlias<"sxtw $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 31)>;
818 def : Pat<(srl GPR32:$Rn, (i64 imm0_31:$imm)),
819 (UBFMWri GPR32:$Rn, imm0_31:$imm, 31)>;
820 def : Pat<(srl GPR64:$Rn, (i64 imm0_63:$imm)),
821 (UBFMXri GPR64:$Rn, imm0_63:$imm, 63)>;
823 def : InstAlias<"lsr $dst, $src, $shift",
824 (UBFMWri GPR32:$dst, GPR32:$src, imm0_31:$shift, 31)>;
825 def : InstAlias<"lsr $dst, $src, $shift",
826 (UBFMXri GPR64:$dst, GPR64:$src, imm0_63:$shift, 63)>;
827 def : InstAlias<"uxtb $dst, $src", (UBFMWri GPR32:$dst, GPR32:$src, 0, 7)>;
828 def : InstAlias<"uxtb $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 7)>;
829 def : InstAlias<"uxth $dst, $src", (UBFMWri GPR32:$dst, GPR32:$src, 0, 15)>;
830 def : InstAlias<"uxth $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 15)>;
831 def : InstAlias<"uxtw $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 31)>;
833 //===----------------------------------------------------------------------===//
834 // Conditionally set flags instructions.
835 //===----------------------------------------------------------------------===//
836 defm CCMN : CondSetFlagsImm<0, "ccmn">;
837 defm CCMP : CondSetFlagsImm<1, "ccmp">;
839 defm CCMN : CondSetFlagsReg<0, "ccmn">;
840 defm CCMP : CondSetFlagsReg<1, "ccmp">;
842 //===----------------------------------------------------------------------===//
843 // Conditional select instructions.
844 //===----------------------------------------------------------------------===//
845 defm CSEL : CondSelect<0, 0b00, "csel">;
847 def inc : PatFrag<(ops node:$in), (add node:$in, 1)>;
848 defm CSINC : CondSelectOp<0, 0b01, "csinc", inc>;
849 defm CSINV : CondSelectOp<1, 0b00, "csinv", not>;
850 defm CSNEG : CondSelectOp<1, 0b01, "csneg", ineg>;
852 def : Pat<(ARM64csinv GPR32:$tval, GPR32:$fval, (i32 imm:$cc), NZCV),
853 (CSINVWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
854 def : Pat<(ARM64csinv GPR64:$tval, GPR64:$fval, (i32 imm:$cc), NZCV),
855 (CSINVXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
856 def : Pat<(ARM64csneg GPR32:$tval, GPR32:$fval, (i32 imm:$cc), NZCV),
857 (CSNEGWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
858 def : Pat<(ARM64csneg GPR64:$tval, GPR64:$fval, (i32 imm:$cc), NZCV),
859 (CSNEGXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
860 def : Pat<(ARM64csinc GPR32:$tval, GPR32:$fval, (i32 imm:$cc), NZCV),
861 (CSINCWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
862 def : Pat<(ARM64csinc GPR64:$tval, GPR64:$fval, (i32 imm:$cc), NZCV),
863 (CSINCXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
865 def : Pat<(ARM64csel (i32 0), (i32 1), (i32 imm:$cc), NZCV),
866 (CSINCWr WZR, WZR, (i32 imm:$cc))>;
867 def : Pat<(ARM64csel (i64 0), (i64 1), (i32 imm:$cc), NZCV),
868 (CSINCXr XZR, XZR, (i32 imm:$cc))>;
869 def : Pat<(ARM64csel (i32 0), (i32 -1), (i32 imm:$cc), NZCV),
870 (CSINVWr WZR, WZR, (i32 imm:$cc))>;
871 def : Pat<(ARM64csel (i64 0), (i64 -1), (i32 imm:$cc), NZCV),
872 (CSINVXr XZR, XZR, (i32 imm:$cc))>;
874 // The inverse of the condition code from the alias instruction is what is used
875 // in the aliased instruction. The parser all ready inverts the condition code
876 // for these aliases.
877 // FIXME: Is this the correct way to handle these aliases?
878 def : InstAlias<"cset $dst, $cc", (CSINCWr GPR32:$dst, WZR, WZR, ccode:$cc)>;
879 def : InstAlias<"cset $dst, $cc", (CSINCXr GPR64:$dst, XZR, XZR, ccode:$cc)>;
881 def : InstAlias<"csetm $dst, $cc", (CSINVWr GPR32:$dst, WZR, WZR, ccode:$cc)>;
882 def : InstAlias<"csetm $dst, $cc", (CSINVXr GPR64:$dst, XZR, XZR, ccode:$cc)>;
884 def : InstAlias<"cinc $dst, $src, $cc",
885 (CSINCWr GPR32:$dst, GPR32:$src, GPR32:$src, ccode:$cc)>;
886 def : InstAlias<"cinc $dst, $src, $cc",
887 (CSINCXr GPR64:$dst, GPR64:$src, GPR64:$src, ccode:$cc)>;
889 def : InstAlias<"cinv $dst, $src, $cc",
890 (CSINVWr GPR32:$dst, GPR32:$src, GPR32:$src, ccode:$cc)>;
891 def : InstAlias<"cinv $dst, $src, $cc",
892 (CSINVXr GPR64:$dst, GPR64:$src, GPR64:$src, ccode:$cc)>;
894 def : InstAlias<"cneg $dst, $src, $cc",
895 (CSNEGWr GPR32:$dst, GPR32:$src, GPR32:$src, ccode:$cc)>;
896 def : InstAlias<"cneg $dst, $src, $cc",
897 (CSNEGXr GPR64:$dst, GPR64:$src, GPR64:$src, ccode:$cc)>;
899 //===----------------------------------------------------------------------===//
900 // PC-relative instructions.
901 //===----------------------------------------------------------------------===//
902 let isReMaterializable = 1 in {
903 let neverHasSideEffects = 1, mayStore = 0, mayLoad = 0 in {
904 def ADR : ADRI<0, "adr", adrlabel, []>;
905 } // neverHasSideEffects = 1
907 def ADRP : ADRI<1, "adrp", adrplabel,
908 [(set GPR64:$Xd, (ARM64adrp tglobaladdr:$label))]>;
909 } // isReMaterializable = 1
911 // page address of a constant pool entry, block address
912 def : Pat<(ARM64adrp tconstpool:$cp), (ADRP tconstpool:$cp)>;
913 def : Pat<(ARM64adrp tblockaddress:$cp), (ADRP tblockaddress:$cp)>;
915 //===----------------------------------------------------------------------===//
916 // Unconditional branch (register) instructions.
917 //===----------------------------------------------------------------------===//
919 let isReturn = 1, isTerminator = 1, isBarrier = 1 in {
920 def RET : BranchReg<0b0010, "ret", []>;
921 def DRPS : SpecialReturn<0b0101, "drps">;
922 def ERET : SpecialReturn<0b0100, "eret">;
923 } // isReturn = 1, isTerminator = 1, isBarrier = 1
925 // Default to the LR register.
926 def : InstAlias<"ret", (RET LR)>;
928 let isCall = 1, Defs = [LR], Uses = [SP] in {
929 def BLR : BranchReg<0b0001, "blr", [(ARM64call GPR64:$Rn)]>;
932 let isBranch = 1, isTerminator = 1, isBarrier = 1, isIndirectBranch = 1 in {
933 def BR : BranchReg<0b0000, "br", [(brind GPR64:$Rn)]>;
934 } // isBranch, isTerminator, isBarrier, isIndirectBranch
936 // Create a separate pseudo-instruction for codegen to use so that we don't
937 // flag lr as used in every function. It'll be restored before the RET by the
938 // epilogue if it's legitimately used.
939 def RET_ReallyLR : Pseudo<(outs), (ins), [(ARM64retflag)]> {
940 let isTerminator = 1;
945 // This is a directive-like pseudo-instruction. The purpose is to insert an
946 // R_AARCH64_TLSDESC_CALL relocation at the offset of the following instruction
947 // (which in the usual case is a BLR).
948 let hasSideEffects = 1 in
949 def TLSDESCCALL : Pseudo<(outs), (ins i64imm:$sym), []> {
950 let AsmString = ".tlsdesccall $sym";
953 // Pseudo-instruction representing a BLR with attached TLSDESC relocation. It
954 // gets expanded to two MCInsts during lowering.
955 let isCall = 1, Defs = [LR] in
957 : Pseudo<(outs), (ins GPR64:$dest, i64imm:$sym),
958 [(ARM64tlsdesc_call GPR64:$dest, tglobaltlsaddr:$sym)]>;
960 def : Pat<(ARM64tlsdesc_call GPR64:$dest, texternalsym:$sym),
961 (TLSDESC_BLR GPR64:$dest, texternalsym:$sym)>;
962 //===----------------------------------------------------------------------===//
963 // Conditional branch (immediate) instruction.
964 //===----------------------------------------------------------------------===//
965 def Bcc : BranchCond;
967 //===----------------------------------------------------------------------===//
968 // Compare-and-branch instructions.
969 //===----------------------------------------------------------------------===//
970 defm CBZ : CmpBranch<0, "cbz", ARM64cbz>;
971 defm CBNZ : CmpBranch<1, "cbnz", ARM64cbnz>;
973 //===----------------------------------------------------------------------===//
974 // Test-bit-and-branch instructions.
975 //===----------------------------------------------------------------------===//
976 def TBZ : TestBranch<0, "tbz", ARM64tbz>;
977 def TBNZ : TestBranch<1, "tbnz", ARM64tbnz>;
979 //===----------------------------------------------------------------------===//
980 // Unconditional branch (immediate) instructions.
981 //===----------------------------------------------------------------------===//
982 let isBranch = 1, isTerminator = 1, isBarrier = 1 in {
983 def B : BranchImm<0, "b", [(br bb:$addr)]>;
984 } // isBranch, isTerminator, isBarrier
986 let isCall = 1, Defs = [LR], Uses = [SP] in {
987 def BL : CallImm<1, "bl", [(ARM64call tglobaladdr:$addr)]>;
989 def : Pat<(ARM64call texternalsym:$func), (BL texternalsym:$func)>;
991 //===----------------------------------------------------------------------===//
992 // Exception generation instructions.
993 //===----------------------------------------------------------------------===//
994 def BRK : ExceptionGeneration<0b001, 0b00, "brk">;
995 def DCPS1 : ExceptionGeneration<0b101, 0b01, "dcps1">;
996 def DCPS2 : ExceptionGeneration<0b101, 0b10, "dcps2">;
997 def DCPS3 : ExceptionGeneration<0b101, 0b11, "dcps3">;
998 def HLT : ExceptionGeneration<0b010, 0b00, "hlt">;
999 def HVC : ExceptionGeneration<0b000, 0b10, "hvc">;
1000 def SMC : ExceptionGeneration<0b000, 0b11, "smc">;
1001 def SVC : ExceptionGeneration<0b000, 0b01, "svc">;
1003 // DCPSn defaults to an immediate operand of zero if unspecified.
1004 def : InstAlias<"dcps1", (DCPS1 0)>;
1005 def : InstAlias<"dcps2", (DCPS2 0)>;
1006 def : InstAlias<"dcps3", (DCPS3 0)>;
1008 //===----------------------------------------------------------------------===//
1009 // Load instructions.
1010 //===----------------------------------------------------------------------===//
1012 // Pair (indexed, offset)
1013 def LDPWi : LoadPairOffset<0b00, 0, GPR32, am_indexed32simm7, "ldp">;
1014 def LDPXi : LoadPairOffset<0b10, 0, GPR64, am_indexed64simm7, "ldp">;
1015 def LDPSi : LoadPairOffset<0b00, 1, FPR32, am_indexed32simm7, "ldp">;
1016 def LDPDi : LoadPairOffset<0b01, 1, FPR64, am_indexed64simm7, "ldp">;
1017 def LDPQi : LoadPairOffset<0b10, 1, FPR128, am_indexed128simm7, "ldp">;
1019 def LDPSWi : LoadPairOffset<0b01, 0, GPR64, am_indexed32simm7, "ldpsw">;
1021 // Pair (pre-indexed)
1022 def LDPWpre : LoadPairPreIdx<0b00, 0, GPR32, am_indexed32simm7_wb, "ldp">;
1023 def LDPXpre : LoadPairPreIdx<0b10, 0, GPR64, am_indexed64simm7_wb, "ldp">;
1024 def LDPSpre : LoadPairPreIdx<0b00, 1, FPR32, am_indexed32simm7_wb, "ldp">;
1025 def LDPDpre : LoadPairPreIdx<0b01, 1, FPR64, am_indexed64simm7_wb, "ldp">;
1026 def LDPQpre : LoadPairPreIdx<0b10, 1, FPR128, am_indexed128simm7_wb, "ldp">;
1028 def LDPSWpre : LoadPairPreIdx<0b01, 0, GPR64, am_indexed32simm7_wb, "ldpsw">;
1030 // Pair (post-indexed)
1031 def LDPWpost : LoadPairPostIdx<0b00, 0, GPR32, simm7s4, "ldp">;
1032 def LDPXpost : LoadPairPostIdx<0b10, 0, GPR64, simm7s8, "ldp">;
1033 def LDPSpost : LoadPairPostIdx<0b00, 1, FPR32, simm7s4, "ldp">;
1034 def LDPDpost : LoadPairPostIdx<0b01, 1, FPR64, simm7s8, "ldp">;
1035 def LDPQpost : LoadPairPostIdx<0b10, 1, FPR128, simm7s16, "ldp">;
1037 def LDPSWpost : LoadPairPostIdx<0b01, 0, GPR64, simm7s4, "ldpsw">;
1040 // Pair (no allocate)
1041 def LDNPWi : LoadPairNoAlloc<0b00, 0, GPR32, am_indexed32simm7, "ldnp">;
1042 def LDNPXi : LoadPairNoAlloc<0b10, 0, GPR64, am_indexed64simm7, "ldnp">;
1043 def LDNPSi : LoadPairNoAlloc<0b00, 1, FPR32, am_indexed32simm7, "ldnp">;
1044 def LDNPDi : LoadPairNoAlloc<0b01, 1, FPR64, am_indexed64simm7, "ldnp">;
1045 def LDNPQi : LoadPairNoAlloc<0b10, 1, FPR128, am_indexed128simm7, "ldnp">;
1048 // (register offset)
1051 let AddedComplexity = 10 in {
1053 def LDRBBro : Load8RO<0b00, 0, 0b01, GPR32, "ldrb",
1054 [(set GPR32:$Rt, (zextloadi8 ro_indexed8:$addr))]>;
1055 def LDRHHro : Load16RO<0b01, 0, 0b01, GPR32, "ldrh",
1056 [(set GPR32:$Rt, (zextloadi16 ro_indexed16:$addr))]>;
1057 def LDRWro : Load32RO<0b10, 0, 0b01, GPR32, "ldr",
1058 [(set GPR32:$Rt, (load ro_indexed32:$addr))]>;
1059 def LDRXro : Load64RO<0b11, 0, 0b01, GPR64, "ldr",
1060 [(set GPR64:$Rt, (load ro_indexed64:$addr))]>;
1063 def LDRBro : Load8RO<0b00, 1, 0b01, FPR8, "ldr",
1064 [(set FPR8:$Rt, (load ro_indexed8:$addr))]>;
1065 def LDRHro : Load16RO<0b01, 1, 0b01, FPR16, "ldr",
1066 [(set (f16 FPR16:$Rt), (load ro_indexed16:$addr))]>;
1067 def LDRSro : Load32RO<0b10, 1, 0b01, FPR32, "ldr",
1068 [(set (f32 FPR32:$Rt), (load ro_indexed32:$addr))]>;
1069 def LDRDro : Load64RO<0b11, 1, 0b01, FPR64, "ldr",
1070 [(set (f64 FPR64:$Rt), (load ro_indexed64:$addr))]>;
1071 def LDRQro : Load128RO<0b00, 1, 0b11, FPR128, "ldr", []> {
1075 // For regular load, we do not have any alignment requirement.
1076 // Thus, it is safe to directly map the vector loads with interesting
1077 // addressing modes.
1078 // FIXME: We could do the same for bitconvert to floating point vectors.
1079 def : Pat <(v8i8 (scalar_to_vector (i32 (extloadi8 ro_indexed8:$addr)))),
1080 (INSERT_SUBREG (v8i8 (IMPLICIT_DEF)),
1081 (LDRBro ro_indexed8:$addr), bsub)>;
1082 def : Pat <(v16i8 (scalar_to_vector (i32 (extloadi8 ro_indexed8:$addr)))),
1083 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
1084 (LDRBro ro_indexed8:$addr), bsub)>;
1085 def : Pat <(v4i16 (scalar_to_vector (i32 (extloadi16 ro_indexed16:$addr)))),
1086 (INSERT_SUBREG (v4i16 (IMPLICIT_DEF)),
1087 (LDRHro ro_indexed16:$addr), hsub)>;
1088 def : Pat <(v8i16 (scalar_to_vector (i32 (extloadi16 ro_indexed16:$addr)))),
1089 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
1090 (LDRHro ro_indexed16:$addr), hsub)>;
1091 def : Pat <(v2i32 (scalar_to_vector (i32 (load ro_indexed32:$addr)))),
1092 (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)),
1093 (LDRSro ro_indexed32:$addr), ssub)>;
1094 def : Pat <(v4i32 (scalar_to_vector (i32 (load ro_indexed32:$addr)))),
1095 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
1096 (LDRSro ro_indexed32:$addr), ssub)>;
1097 def : Pat <(v1i64 (scalar_to_vector (i64 (load ro_indexed64:$addr)))),
1098 (LDRDro ro_indexed64:$addr)>;
1099 def : Pat <(v2i64 (scalar_to_vector (i64 (load ro_indexed64:$addr)))),
1100 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
1101 (LDRDro ro_indexed64:$addr), dsub)>;
1103 // Match all load 64 bits width whose type is compatible with FPR64
1104 let Predicates = [IsLE] in {
1105 // We must do vector loads with LD1 in big-endian.
1106 def : Pat<(v2f32 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1107 def : Pat<(v8i8 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1108 def : Pat<(v4i16 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1109 def : Pat<(v2i32 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1111 def : Pat<(v1f64 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1112 def : Pat<(v1i64 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1114 // Match all load 128 bits width whose type is compatible with FPR128
1115 let Predicates = [IsLE] in {
1116 // We must do vector loads with LD1 in big-endian.
1117 def : Pat<(v4f32 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1118 def : Pat<(v2f64 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1119 def : Pat<(v16i8 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1120 def : Pat<(v8i16 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1121 def : Pat<(v4i32 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1122 def : Pat<(v2i64 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1124 def : Pat<(f128 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1126 // Load sign-extended half-word
1127 def LDRSHWro : Load16RO<0b01, 0, 0b11, GPR32, "ldrsh",
1128 [(set GPR32:$Rt, (sextloadi16 ro_indexed16:$addr))]>;
1129 def LDRSHXro : Load16RO<0b01, 0, 0b10, GPR64, "ldrsh",
1130 [(set GPR64:$Rt, (sextloadi16 ro_indexed16:$addr))]>;
1132 // Load sign-extended byte
1133 def LDRSBWro : Load8RO<0b00, 0, 0b11, GPR32, "ldrsb",
1134 [(set GPR32:$Rt, (sextloadi8 ro_indexed8:$addr))]>;
1135 def LDRSBXro : Load8RO<0b00, 0, 0b10, GPR64, "ldrsb",
1136 [(set GPR64:$Rt, (sextloadi8 ro_indexed8:$addr))]>;
1138 // Load sign-extended word
1139 def LDRSWro : Load32RO<0b10, 0, 0b10, GPR64, "ldrsw",
1140 [(set GPR64:$Rt, (sextloadi32 ro_indexed32:$addr))]>;
1143 def PRFMro : PrefetchRO<0b11, 0, 0b10, "prfm",
1144 [(ARM64Prefetch imm:$Rt, ro_indexed64:$addr)]>;
1147 def : Pat<(i64 (zextloadi8 ro_indexed8:$addr)),
1148 (SUBREG_TO_REG (i64 0), (LDRBBro ro_indexed8:$addr), sub_32)>;
1149 def : Pat<(i64 (zextloadi16 ro_indexed16:$addr)),
1150 (SUBREG_TO_REG (i64 0), (LDRHHro ro_indexed16:$addr), sub_32)>;
1151 def : Pat<(i64 (zextloadi32 ro_indexed32:$addr)),
1152 (SUBREG_TO_REG (i64 0), (LDRWro ro_indexed32:$addr), sub_32)>;
1154 // zextloadi1 -> zextloadi8
1155 def : Pat<(i32 (zextloadi1 ro_indexed8:$addr)), (LDRBBro ro_indexed8:$addr)>;
1156 def : Pat<(i64 (zextloadi1 ro_indexed8:$addr)),
1157 (SUBREG_TO_REG (i64 0), (LDRBBro ro_indexed8:$addr), sub_32)>;
1159 // extload -> zextload
1160 def : Pat<(i32 (extloadi16 ro_indexed16:$addr)), (LDRHHro ro_indexed16:$addr)>;
1161 def : Pat<(i32 (extloadi8 ro_indexed8:$addr)), (LDRBBro ro_indexed8:$addr)>;
1162 def : Pat<(i32 (extloadi1 ro_indexed8:$addr)), (LDRBBro ro_indexed8:$addr)>;
1163 def : Pat<(i64 (extloadi32 ro_indexed32:$addr)),
1164 (SUBREG_TO_REG (i64 0), (LDRWro ro_indexed32:$addr), sub_32)>;
1165 def : Pat<(i64 (extloadi16 ro_indexed16:$addr)),
1166 (SUBREG_TO_REG (i64 0), (LDRHHro ro_indexed16:$addr), sub_32)>;
1167 def : Pat<(i64 (extloadi8 ro_indexed8:$addr)),
1168 (SUBREG_TO_REG (i64 0), (LDRBBro ro_indexed8:$addr), sub_32)>;
1169 def : Pat<(i64 (extloadi1 ro_indexed8:$addr)),
1170 (SUBREG_TO_REG (i64 0), (LDRBBro ro_indexed8:$addr), sub_32)>;
1172 } // AddedComplexity = 10
1175 // (unsigned immediate)
1177 def LDRXui : LoadUI<0b11, 0, 0b01, GPR64, am_indexed64, "ldr",
1178 [(set GPR64:$Rt, (load am_indexed64:$addr))]>;
1179 def LDRWui : LoadUI<0b10, 0, 0b01, GPR32, am_indexed32, "ldr",
1180 [(set GPR32:$Rt, (load am_indexed32:$addr))]>;
1181 def LDRBui : LoadUI<0b00, 1, 0b01, FPR8, am_indexed8, "ldr",
1182 [(set FPR8:$Rt, (load am_indexed8:$addr))]>;
1183 def LDRHui : LoadUI<0b01, 1, 0b01, FPR16, am_indexed16, "ldr",
1184 [(set (f16 FPR16:$Rt), (load am_indexed16:$addr))]>;
1185 def LDRSui : LoadUI<0b10, 1, 0b01, FPR32, am_indexed32, "ldr",
1186 [(set (f32 FPR32:$Rt), (load am_indexed32:$addr))]>;
1187 def LDRDui : LoadUI<0b11, 1, 0b01, FPR64, am_indexed64, "ldr",
1188 [(set (f64 FPR64:$Rt), (load am_indexed64:$addr))]>;
1189 def LDRQui : LoadUI<0b00, 1, 0b11, FPR128, am_indexed128, "ldr",
1190 [(set (f128 FPR128:$Rt), (load am_indexed128:$addr))]>;
1192 // For regular load, we do not have any alignment requirement.
1193 // Thus, it is safe to directly map the vector loads with interesting
1194 // addressing modes.
1195 // FIXME: We could do the same for bitconvert to floating point vectors.
1196 def : Pat <(v8i8 (scalar_to_vector (i32 (extloadi8 am_indexed8:$addr)))),
1197 (INSERT_SUBREG (v8i8 (IMPLICIT_DEF)),
1198 (LDRBui am_indexed8:$addr), bsub)>;
1199 def : Pat <(v16i8 (scalar_to_vector (i32 (extloadi8 am_indexed8:$addr)))),
1200 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
1201 (LDRBui am_indexed8:$addr), bsub)>;
1202 def : Pat <(v4i16 (scalar_to_vector (i32 (extloadi16 am_indexed16:$addr)))),
1203 (INSERT_SUBREG (v4i16 (IMPLICIT_DEF)),
1204 (LDRHui am_indexed16:$addr), hsub)>;
1205 def : Pat <(v8i16 (scalar_to_vector (i32 (extloadi16 am_indexed16:$addr)))),
1206 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
1207 (LDRHui am_indexed16:$addr), hsub)>;
1208 def : Pat <(v2i32 (scalar_to_vector (i32 (load am_indexed32:$addr)))),
1209 (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)),
1210 (LDRSui am_indexed32:$addr), ssub)>;
1211 def : Pat <(v4i32 (scalar_to_vector (i32 (load am_indexed32:$addr)))),
1212 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
1213 (LDRSui am_indexed32:$addr), ssub)>;
1214 def : Pat <(v1i64 (scalar_to_vector (i64 (load am_indexed64:$addr)))),
1215 (LDRDui am_indexed64:$addr)>;
1216 def : Pat <(v2i64 (scalar_to_vector (i64 (load am_indexed64:$addr)))),
1217 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
1218 (LDRDui am_indexed64:$addr), dsub)>;
1220 // Match all load 64 bits width whose type is compatible with FPR64
1221 let Predicates = [IsLE] in {
1222 // We must use LD1 to perform vector loads in big-endian.
1223 def : Pat<(v2f32 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1224 def : Pat<(v8i8 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1225 def : Pat<(v4i16 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1226 def : Pat<(v2i32 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1228 def : Pat<(v1f64 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1229 def : Pat<(v1i64 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1231 // Match all load 128 bits width whose type is compatible with FPR128
1232 let Predicates = [IsLE] in {
1233 // We must use LD1 to perform vector loads in big-endian.
1234 def : Pat<(v4f32 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1235 def : Pat<(v2f64 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1236 def : Pat<(v16i8 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1237 def : Pat<(v8i16 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1238 def : Pat<(v4i32 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1239 def : Pat<(v2i64 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1241 def : Pat<(f128 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1243 def LDRHHui : LoadUI<0b01, 0, 0b01, GPR32, am_indexed16, "ldrh",
1244 [(set GPR32:$Rt, (zextloadi16 am_indexed16:$addr))]>;
1245 def LDRBBui : LoadUI<0b00, 0, 0b01, GPR32, am_indexed8, "ldrb",
1246 [(set GPR32:$Rt, (zextloadi8 am_indexed8:$addr))]>;
1248 def : Pat<(i64 (zextloadi8 am_indexed8:$addr)),
1249 (SUBREG_TO_REG (i64 0), (LDRBBui am_indexed8:$addr), sub_32)>;
1250 def : Pat<(i64 (zextloadi16 am_indexed16:$addr)),
1251 (SUBREG_TO_REG (i64 0), (LDRHHui am_indexed16:$addr), sub_32)>;
1253 // zextloadi1 -> zextloadi8
1254 def : Pat<(i32 (zextloadi1 am_indexed8:$addr)), (LDRBBui am_indexed8:$addr)>;
1255 def : Pat<(i64 (zextloadi1 am_indexed8:$addr)),
1256 (SUBREG_TO_REG (i64 0), (LDRBBui am_indexed8:$addr), sub_32)>;
1258 // extload -> zextload
1259 def : Pat<(i32 (extloadi16 am_indexed16:$addr)), (LDRHHui am_indexed16:$addr)>;
1260 def : Pat<(i32 (extloadi8 am_indexed8:$addr)), (LDRBBui am_indexed8:$addr)>;
1261 def : Pat<(i32 (extloadi1 am_indexed8:$addr)), (LDRBBui am_indexed8:$addr)>;
1262 def : Pat<(i64 (extloadi32 am_indexed32:$addr)),
1263 (SUBREG_TO_REG (i64 0), (LDRWui am_indexed32:$addr), sub_32)>;
1264 def : Pat<(i64 (extloadi16 am_indexed16:$addr)),
1265 (SUBREG_TO_REG (i64 0), (LDRHHui am_indexed16:$addr), sub_32)>;
1266 def : Pat<(i64 (extloadi8 am_indexed8:$addr)),
1267 (SUBREG_TO_REG (i64 0), (LDRBBui am_indexed8:$addr), sub_32)>;
1268 def : Pat<(i64 (extloadi1 am_indexed8:$addr)),
1269 (SUBREG_TO_REG (i64 0), (LDRBBui am_indexed8:$addr), sub_32)>;
1271 // load sign-extended half-word
1272 def LDRSHWui : LoadUI<0b01, 0, 0b11, GPR32, am_indexed16, "ldrsh",
1273 [(set GPR32:$Rt, (sextloadi16 am_indexed16:$addr))]>;
1274 def LDRSHXui : LoadUI<0b01, 0, 0b10, GPR64, am_indexed16, "ldrsh",
1275 [(set GPR64:$Rt, (sextloadi16 am_indexed16:$addr))]>;
1277 // load sign-extended byte
1278 def LDRSBWui : LoadUI<0b00, 0, 0b11, GPR32, am_indexed8, "ldrsb",
1279 [(set GPR32:$Rt, (sextloadi8 am_indexed8:$addr))]>;
1280 def LDRSBXui : LoadUI<0b00, 0, 0b10, GPR64, am_indexed8, "ldrsb",
1281 [(set GPR64:$Rt, (sextloadi8 am_indexed8:$addr))]>;
1283 // load sign-extended word
1284 def LDRSWui : LoadUI<0b10, 0, 0b10, GPR64, am_indexed32, "ldrsw",
1285 [(set GPR64:$Rt, (sextloadi32 am_indexed32:$addr))]>;
1287 // load zero-extended word
1288 def : Pat<(i64 (zextloadi32 am_indexed32:$addr)),
1289 (SUBREG_TO_REG (i64 0), (LDRWui am_indexed32:$addr), sub_32)>;
1292 def PRFMui : PrefetchUI<0b11, 0, 0b10, "prfm",
1293 [(ARM64Prefetch imm:$Rt, am_indexed64:$addr)]>;
1297 def LDRWl : LoadLiteral<0b00, 0, GPR32, "ldr">;
1298 def LDRXl : LoadLiteral<0b01, 0, GPR64, "ldr">;
1299 def LDRSl : LoadLiteral<0b00, 1, FPR32, "ldr">;
1300 def LDRDl : LoadLiteral<0b01, 1, FPR64, "ldr">;
1301 def LDRQl : LoadLiteral<0b10, 1, FPR128, "ldr">;
1303 // load sign-extended word
1304 def LDRSWl : LoadLiteral<0b10, 0, GPR64, "ldrsw">;
1307 def PRFMl : PrefetchLiteral<0b11, 0, "prfm", []>;
1308 // [(ARM64Prefetch imm:$Rt, tglobaladdr:$label)]>;
1311 // (unscaled immediate)
1312 def LDURXi : LoadUnscaled<0b11, 0, 0b01, GPR64, am_unscaled64, "ldur",
1313 [(set GPR64:$Rt, (load am_unscaled64:$addr))]>;
1314 def LDURWi : LoadUnscaled<0b10, 0, 0b01, GPR32, am_unscaled32, "ldur",
1315 [(set GPR32:$Rt, (load am_unscaled32:$addr))]>;
1316 def LDURBi : LoadUnscaled<0b00, 1, 0b01, FPR8, am_unscaled8, "ldur",
1317 [(set FPR8:$Rt, (load am_unscaled8:$addr))]>;
1318 def LDURHi : LoadUnscaled<0b01, 1, 0b01, FPR16, am_unscaled16, "ldur",
1319 [(set (f16 FPR16:$Rt), (load am_unscaled16:$addr))]>;
1320 def LDURSi : LoadUnscaled<0b10, 1, 0b01, FPR32, am_unscaled32, "ldur",
1321 [(set (f32 FPR32:$Rt), (load am_unscaled32:$addr))]>;
1322 def LDURDi : LoadUnscaled<0b11, 1, 0b01, FPR64, am_unscaled64, "ldur",
1323 [(set (f64 FPR64:$Rt), (load am_unscaled64:$addr))]>;
1324 def LDURQi : LoadUnscaled<0b00, 1, 0b11, FPR128, am_unscaled128, "ldur",
1325 [(set (f128 FPR128:$Rt), (load am_unscaled128:$addr))]>;
1328 : LoadUnscaled<0b01, 0, 0b01, GPR32, am_unscaled16, "ldurh",
1329 [(set GPR32:$Rt, (zextloadi16 am_unscaled16:$addr))]>;
1331 : LoadUnscaled<0b00, 0, 0b01, GPR32, am_unscaled8, "ldurb",
1332 [(set GPR32:$Rt, (zextloadi8 am_unscaled8:$addr))]>;
1334 // Match all load 64 bits width whose type is compatible with FPR64
1335 let Predicates = [IsLE] in {
1336 def : Pat<(v2f32 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1337 def : Pat<(v8i8 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1338 def : Pat<(v4i16 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1339 def : Pat<(v2i32 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1341 def : Pat<(v1f64 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1342 def : Pat<(v1i64 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1344 // Match all load 128 bits width whose type is compatible with FPR128
1345 let Predicates = [IsLE] in {
1346 def : Pat<(v4f32 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1347 def : Pat<(v2f64 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1348 def : Pat<(v16i8 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1349 def : Pat<(v8i16 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1350 def : Pat<(v4i32 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1351 def : Pat<(v2i64 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1352 def : Pat<(v2f64 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1356 def : Pat<(i32 (extloadi16 am_unscaled16:$addr)), (LDURHHi am_unscaled16:$addr)>;
1357 def : Pat<(i32 (extloadi8 am_unscaled8:$addr)), (LDURBBi am_unscaled8:$addr)>;
1358 def : Pat<(i32 (extloadi1 am_unscaled8:$addr)), (LDURBBi am_unscaled8:$addr)>;
1359 def : Pat<(i64 (extloadi32 am_unscaled32:$addr)),
1360 (SUBREG_TO_REG (i64 0), (LDURWi am_unscaled32:$addr), sub_32)>;
1361 def : Pat<(i64 (extloadi16 am_unscaled16:$addr)),
1362 (SUBREG_TO_REG (i64 0), (LDURHHi am_unscaled16:$addr), sub_32)>;
1363 def : Pat<(i64 (extloadi8 am_unscaled8:$addr)),
1364 (SUBREG_TO_REG (i64 0), (LDURBBi am_unscaled8:$addr), sub_32)>;
1365 def : Pat<(i64 (extloadi1 am_unscaled8:$addr)),
1366 (SUBREG_TO_REG (i64 0), (LDURBBi am_unscaled8:$addr), sub_32)>;
1368 def : Pat<(i32 (zextloadi16 am_unscaled16:$addr)),
1369 (LDURHHi am_unscaled16:$addr)>;
1370 def : Pat<(i32 (zextloadi8 am_unscaled8:$addr)),
1371 (LDURBBi am_unscaled8:$addr)>;
1372 def : Pat<(i32 (zextloadi1 am_unscaled8:$addr)),
1373 (LDURBBi am_unscaled8:$addr)>;
1374 def : Pat<(i64 (zextloadi32 am_unscaled32:$addr)),
1375 (SUBREG_TO_REG (i64 0), (LDURWi am_unscaled32:$addr), sub_32)>;
1376 def : Pat<(i64 (zextloadi16 am_unscaled16:$addr)),
1377 (SUBREG_TO_REG (i64 0), (LDURHHi am_unscaled16:$addr), sub_32)>;
1378 def : Pat<(i64 (zextloadi8 am_unscaled8:$addr)),
1379 (SUBREG_TO_REG (i64 0), (LDURBBi am_unscaled8:$addr), sub_32)>;
1380 def : Pat<(i64 (zextloadi1 am_unscaled8:$addr)),
1381 (SUBREG_TO_REG (i64 0), (LDURBBi am_unscaled8:$addr), sub_32)>;
1385 // LDR mnemonics fall back to LDUR for negative or unaligned offsets.
1387 // Define new assembler match classes as we want to only match these when
1388 // the don't otherwise match the scaled addressing mode for LDR/STR. Don't
1389 // associate a DiagnosticType either, as we want the diagnostic for the
1390 // canonical form (the scaled operand) to take precedence.
1391 def MemoryUnscaledFB8Operand : AsmOperandClass {
1392 let Name = "MemoryUnscaledFB8";
1393 let RenderMethod = "addMemoryUnscaledOperands";
1395 def MemoryUnscaledFB16Operand : AsmOperandClass {
1396 let Name = "MemoryUnscaledFB16";
1397 let RenderMethod = "addMemoryUnscaledOperands";
1399 def MemoryUnscaledFB32Operand : AsmOperandClass {
1400 let Name = "MemoryUnscaledFB32";
1401 let RenderMethod = "addMemoryUnscaledOperands";
1403 def MemoryUnscaledFB64Operand : AsmOperandClass {
1404 let Name = "MemoryUnscaledFB64";
1405 let RenderMethod = "addMemoryUnscaledOperands";
1407 def MemoryUnscaledFB128Operand : AsmOperandClass {
1408 let Name = "MemoryUnscaledFB128";
1409 let RenderMethod = "addMemoryUnscaledOperands";
1411 def am_unscaled_fb8 : Operand<i64> {
1412 let ParserMatchClass = MemoryUnscaledFB8Operand;
1413 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
1415 def am_unscaled_fb16 : Operand<i64> {
1416 let ParserMatchClass = MemoryUnscaledFB16Operand;
1417 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
1419 def am_unscaled_fb32 : Operand<i64> {
1420 let ParserMatchClass = MemoryUnscaledFB32Operand;
1421 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
1423 def am_unscaled_fb64 : Operand<i64> {
1424 let ParserMatchClass = MemoryUnscaledFB64Operand;
1425 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
1427 def am_unscaled_fb128 : Operand<i64> {
1428 let ParserMatchClass = MemoryUnscaledFB128Operand;
1429 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
1431 def : InstAlias<"ldr $Rt, $addr", (LDURXi GPR64:$Rt, am_unscaled_fb64:$addr)>;
1432 def : InstAlias<"ldr $Rt, $addr", (LDURWi GPR32:$Rt, am_unscaled_fb32:$addr)>;
1433 def : InstAlias<"ldr $Rt, $addr", (LDURBi FPR8:$Rt, am_unscaled_fb8:$addr)>;
1434 def : InstAlias<"ldr $Rt, $addr", (LDURHi FPR16:$Rt, am_unscaled_fb16:$addr)>;
1435 def : InstAlias<"ldr $Rt, $addr", (LDURSi FPR32:$Rt, am_unscaled_fb32:$addr)>;
1436 def : InstAlias<"ldr $Rt, $addr", (LDURDi FPR64:$Rt, am_unscaled_fb64:$addr)>;
1437 def : InstAlias<"ldr $Rt, $addr", (LDURQi FPR128:$Rt, am_unscaled_fb128:$addr)>;
1440 def : Pat<(i64 (zextloadi8 am_unscaled8:$addr)),
1441 (SUBREG_TO_REG (i64 0), (LDURBBi am_unscaled8:$addr), sub_32)>;
1442 def : Pat<(i64 (zextloadi16 am_unscaled16:$addr)),
1443 (SUBREG_TO_REG (i64 0), (LDURHHi am_unscaled16:$addr), sub_32)>;
1445 // load sign-extended half-word
1447 : LoadUnscaled<0b01, 0, 0b11, GPR32, am_unscaled16, "ldursh",
1448 [(set GPR32:$Rt, (sextloadi16 am_unscaled16:$addr))]>;
1450 : LoadUnscaled<0b01, 0, 0b10, GPR64, am_unscaled16, "ldursh",
1451 [(set GPR64:$Rt, (sextloadi16 am_unscaled16:$addr))]>;
1453 // load sign-extended byte
1455 : LoadUnscaled<0b00, 0, 0b11, GPR32, am_unscaled8, "ldursb",
1456 [(set GPR32:$Rt, (sextloadi8 am_unscaled8:$addr))]>;
1458 : LoadUnscaled<0b00, 0, 0b10, GPR64, am_unscaled8, "ldursb",
1459 [(set GPR64:$Rt, (sextloadi8 am_unscaled8:$addr))]>;
1461 // load sign-extended word
1463 : LoadUnscaled<0b10, 0, 0b10, GPR64, am_unscaled32, "ldursw",
1464 [(set GPR64:$Rt, (sextloadi32 am_unscaled32:$addr))]>;
1466 // zero and sign extending aliases from generic LDR* mnemonics to LDUR*.
1467 def : InstAlias<"ldrb $Rt, $addr", (LDURBBi GPR32:$Rt, am_unscaled_fb8:$addr)>;
1468 def : InstAlias<"ldrh $Rt, $addr", (LDURHHi GPR32:$Rt, am_unscaled_fb16:$addr)>;
1469 def : InstAlias<"ldrsb $Rt, $addr", (LDURSBWi GPR32:$Rt, am_unscaled_fb8:$addr)>;
1470 def : InstAlias<"ldrsb $Rt, $addr", (LDURSBXi GPR64:$Rt, am_unscaled_fb8:$addr)>;
1471 def : InstAlias<"ldrsh $Rt, $addr", (LDURSHWi GPR32:$Rt, am_unscaled_fb16:$addr)>;
1472 def : InstAlias<"ldrsh $Rt, $addr", (LDURSHXi GPR64:$Rt, am_unscaled_fb16:$addr)>;
1473 def : InstAlias<"ldrsw $Rt, $addr", (LDURSWi GPR64:$Rt, am_unscaled_fb32:$addr)>;
1476 def PRFUMi : PrefetchUnscaled<0b11, 0, 0b10, "prfum",
1477 [(ARM64Prefetch imm:$Rt, am_unscaled64:$addr)]>;
1480 // (unscaled immediate, unprivileged)
1481 def LDTRXi : LoadUnprivileged<0b11, 0, 0b01, GPR64, "ldtr">;
1482 def LDTRWi : LoadUnprivileged<0b10, 0, 0b01, GPR32, "ldtr">;
1484 def LDTRHi : LoadUnprivileged<0b01, 0, 0b01, GPR32, "ldtrh">;
1485 def LDTRBi : LoadUnprivileged<0b00, 0, 0b01, GPR32, "ldtrb">;
1487 // load sign-extended half-word
1488 def LDTRSHWi : LoadUnprivileged<0b01, 0, 0b11, GPR32, "ldtrsh">;
1489 def LDTRSHXi : LoadUnprivileged<0b01, 0, 0b10, GPR64, "ldtrsh">;
1491 // load sign-extended byte
1492 def LDTRSBWi : LoadUnprivileged<0b00, 0, 0b11, GPR32, "ldtrsb">;
1493 def LDTRSBXi : LoadUnprivileged<0b00, 0, 0b10, GPR64, "ldtrsb">;
1495 // load sign-extended word
1496 def LDTRSWi : LoadUnprivileged<0b10, 0, 0b10, GPR64, "ldtrsw">;
1499 // (immediate pre-indexed)
1500 def LDRWpre : LoadPreIdx<0b10, 0, 0b01, GPR32, "ldr">;
1501 def LDRXpre : LoadPreIdx<0b11, 0, 0b01, GPR64, "ldr">;
1502 def LDRBpre : LoadPreIdx<0b00, 1, 0b01, FPR8, "ldr">;
1503 def LDRHpre : LoadPreIdx<0b01, 1, 0b01, FPR16, "ldr">;
1504 def LDRSpre : LoadPreIdx<0b10, 1, 0b01, FPR32, "ldr">;
1505 def LDRDpre : LoadPreIdx<0b11, 1, 0b01, FPR64, "ldr">;
1506 def LDRQpre : LoadPreIdx<0b00, 1, 0b11, FPR128, "ldr">;
1508 // load sign-extended half-word
1509 def LDRSHWpre : LoadPreIdx<0b01, 0, 0b11, GPR32, "ldrsh">;
1510 def LDRSHXpre : LoadPreIdx<0b01, 0, 0b10, GPR64, "ldrsh">;
1512 // load sign-extended byte
1513 def LDRSBWpre : LoadPreIdx<0b00, 0, 0b11, GPR32, "ldrsb">;
1514 def LDRSBXpre : LoadPreIdx<0b00, 0, 0b10, GPR64, "ldrsb">;
1516 // load zero-extended byte
1517 def LDRBBpre : LoadPreIdx<0b00, 0, 0b01, GPR32, "ldrb">;
1518 def LDRHHpre : LoadPreIdx<0b01, 0, 0b01, GPR32, "ldrh">;
1520 // load sign-extended word
1521 def LDRSWpre : LoadPreIdx<0b10, 0, 0b10, GPR64, "ldrsw">;
1523 // ISel pseudos and patterns. See expanded comment on LoadPreIdxPseudo.
1524 def LDRQpre_isel : LoadPreIdxPseudo<FPR128>;
1525 def LDRDpre_isel : LoadPreIdxPseudo<FPR64>;
1526 def LDRSpre_isel : LoadPreIdxPseudo<FPR32>;
1527 def LDRXpre_isel : LoadPreIdxPseudo<GPR64>;
1528 def LDRWpre_isel : LoadPreIdxPseudo<GPR32>;
1529 def LDRHHpre_isel : LoadPreIdxPseudo<GPR32>;
1530 def LDRBBpre_isel : LoadPreIdxPseudo<GPR32>;
1532 def LDRSWpre_isel : LoadPreIdxPseudo<GPR64>;
1533 def LDRSHWpre_isel : LoadPreIdxPseudo<GPR32>;
1534 def LDRSHXpre_isel : LoadPreIdxPseudo<GPR64>;
1535 def LDRSBWpre_isel : LoadPreIdxPseudo<GPR32>;
1536 def LDRSBXpre_isel : LoadPreIdxPseudo<GPR64>;
1539 // (immediate post-indexed)
1540 def LDRWpost : LoadPostIdx<0b10, 0, 0b01, GPR32, "ldr">;
1541 def LDRXpost : LoadPostIdx<0b11, 0, 0b01, GPR64, "ldr">;
1542 def LDRBpost : LoadPostIdx<0b00, 1, 0b01, FPR8, "ldr">;
1543 def LDRHpost : LoadPostIdx<0b01, 1, 0b01, FPR16, "ldr">;
1544 def LDRSpost : LoadPostIdx<0b10, 1, 0b01, FPR32, "ldr">;
1545 def LDRDpost : LoadPostIdx<0b11, 1, 0b01, FPR64, "ldr">;
1546 def LDRQpost : LoadPostIdx<0b00, 1, 0b11, FPR128, "ldr">;
1548 // load sign-extended half-word
1549 def LDRSHWpost : LoadPostIdx<0b01, 0, 0b11, GPR32, "ldrsh">;
1550 def LDRSHXpost : LoadPostIdx<0b01, 0, 0b10, GPR64, "ldrsh">;
1552 // load sign-extended byte
1553 def LDRSBWpost : LoadPostIdx<0b00, 0, 0b11, GPR32, "ldrsb">;
1554 def LDRSBXpost : LoadPostIdx<0b00, 0, 0b10, GPR64, "ldrsb">;
1556 // load zero-extended byte
1557 def LDRBBpost : LoadPostIdx<0b00, 0, 0b01, GPR32, "ldrb">;
1558 def LDRHHpost : LoadPostIdx<0b01, 0, 0b01, GPR32, "ldrh">;
1560 // load sign-extended word
1561 def LDRSWpost : LoadPostIdx<0b10, 0, 0b10, GPR64, "ldrsw">;
1563 // ISel pseudos and patterns. See expanded comment on LoadPostIdxPseudo.
1564 def LDRQpost_isel : LoadPostIdxPseudo<FPR128>;
1565 def LDRDpost_isel : LoadPostIdxPseudo<FPR64>;
1566 def LDRSpost_isel : LoadPostIdxPseudo<FPR32>;
1567 def LDRXpost_isel : LoadPostIdxPseudo<GPR64>;
1568 def LDRWpost_isel : LoadPostIdxPseudo<GPR32>;
1569 def LDRHHpost_isel : LoadPostIdxPseudo<GPR32>;
1570 def LDRBBpost_isel : LoadPostIdxPseudo<GPR32>;
1572 def LDRSWpost_isel : LoadPostIdxPseudo<GPR64>;
1573 def LDRSHWpost_isel : LoadPostIdxPseudo<GPR32>;
1574 def LDRSHXpost_isel : LoadPostIdxPseudo<GPR64>;
1575 def LDRSBWpost_isel : LoadPostIdxPseudo<GPR32>;
1576 def LDRSBXpost_isel : LoadPostIdxPseudo<GPR64>;
1578 //===----------------------------------------------------------------------===//
1579 // Store instructions.
1580 //===----------------------------------------------------------------------===//
1582 // Pair (indexed, offset)
1583 // FIXME: Use dedicated range-checked addressing mode operand here.
1584 def STPWi : StorePairOffset<0b00, 0, GPR32, am_indexed32simm7, "stp">;
1585 def STPXi : StorePairOffset<0b10, 0, GPR64, am_indexed64simm7, "stp">;
1586 def STPSi : StorePairOffset<0b00, 1, FPR32, am_indexed32simm7, "stp">;
1587 def STPDi : StorePairOffset<0b01, 1, FPR64, am_indexed64simm7, "stp">;
1588 def STPQi : StorePairOffset<0b10, 1, FPR128, am_indexed128simm7, "stp">;
1590 // Pair (pre-indexed)
1591 def STPWpre : StorePairPreIdx<0b00, 0, GPR32, am_indexed32simm7_wb, "stp">;
1592 def STPXpre : StorePairPreIdx<0b10, 0, GPR64, am_indexed64simm7_wb, "stp">;
1593 def STPSpre : StorePairPreIdx<0b00, 1, FPR32, am_indexed32simm7_wb, "stp">;
1594 def STPDpre : StorePairPreIdx<0b01, 1, FPR64, am_indexed64simm7_wb, "stp">;
1595 def STPQpre : StorePairPreIdx<0b10, 1, FPR128, am_indexed128simm7_wb, "stp">;
1597 // Pair (pre-indexed)
1598 def STPWpost : StorePairPostIdx<0b00, 0, GPR32, simm7s4, "stp">;
1599 def STPXpost : StorePairPostIdx<0b10, 0, GPR64, simm7s8, "stp">;
1600 def STPSpost : StorePairPostIdx<0b00, 1, FPR32, simm7s4, "stp">;
1601 def STPDpost : StorePairPostIdx<0b01, 1, FPR64, simm7s8, "stp">;
1602 def STPQpost : StorePairPostIdx<0b10, 1, FPR128, simm7s16, "stp">;
1604 // Pair (no allocate)
1605 def STNPWi : StorePairNoAlloc<0b00, 0, GPR32, am_indexed32simm7, "stnp">;
1606 def STNPXi : StorePairNoAlloc<0b10, 0, GPR64, am_indexed64simm7, "stnp">;
1607 def STNPSi : StorePairNoAlloc<0b00, 1, FPR32, am_indexed32simm7, "stnp">;
1608 def STNPDi : StorePairNoAlloc<0b01, 1, FPR64, am_indexed64simm7, "stnp">;
1609 def STNPQi : StorePairNoAlloc<0b10, 1, FPR128, am_indexed128simm7, "stnp">;
1612 // (Register offset)
1614 let AddedComplexity = 10 in {
1617 def STRHHro : Store16RO<0b01, 0, 0b00, GPR32, "strh",
1618 [(truncstorei16 GPR32:$Rt, ro_indexed16:$addr)]>;
1619 def STRBBro : Store8RO<0b00, 0, 0b00, GPR32, "strb",
1620 [(truncstorei8 GPR32:$Rt, ro_indexed8:$addr)]>;
1621 def STRWro : Store32RO<0b10, 0, 0b00, GPR32, "str",
1622 [(store GPR32:$Rt, ro_indexed32:$addr)]>;
1623 def STRXro : Store64RO<0b11, 0, 0b00, GPR64, "str",
1624 [(store GPR64:$Rt, ro_indexed64:$addr)]>;
1627 def : Pat<(truncstorei8 GPR64:$Rt, ro_indexed8:$addr),
1628 (STRBBro (EXTRACT_SUBREG GPR64:$Rt, sub_32), ro_indexed8:$addr)>;
1629 def : Pat<(truncstorei16 GPR64:$Rt, ro_indexed16:$addr),
1630 (STRHHro (EXTRACT_SUBREG GPR64:$Rt, sub_32), ro_indexed16:$addr)>;
1631 def : Pat<(truncstorei32 GPR64:$Rt, ro_indexed32:$addr),
1632 (STRWro (EXTRACT_SUBREG GPR64:$Rt, sub_32), ro_indexed32:$addr)>;
1636 def STRBro : Store8RO<0b00, 1, 0b00, FPR8, "str",
1637 [(store FPR8:$Rt, ro_indexed8:$addr)]>;
1638 def STRHro : Store16RO<0b01, 1, 0b00, FPR16, "str",
1639 [(store (f16 FPR16:$Rt), ro_indexed16:$addr)]>;
1640 def STRSro : Store32RO<0b10, 1, 0b00, FPR32, "str",
1641 [(store (f32 FPR32:$Rt), ro_indexed32:$addr)]>;
1642 def STRDro : Store64RO<0b11, 1, 0b00, FPR64, "str",
1643 [(store (f64 FPR64:$Rt), ro_indexed64:$addr)]>;
1644 def STRQro : Store128RO<0b00, 1, 0b10, FPR128, "str", []> {
1648 // Match all store 64 bits width whose type is compatible with FPR64
1649 let Predicates = [IsLE] in {
1650 // We must use ST1 to store vectors in big-endian.
1651 def : Pat<(store (v2f32 FPR64:$Rn), ro_indexed64:$addr),
1652 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1653 def : Pat<(store (v8i8 FPR64:$Rn), ro_indexed64:$addr),
1654 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1655 def : Pat<(store (v4i16 FPR64:$Rn), ro_indexed64:$addr),
1656 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1657 def : Pat<(store (v2i32 FPR64:$Rn), ro_indexed64:$addr),
1658 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1660 def : Pat<(store (v1f64 FPR64:$Rn), ro_indexed64:$addr),
1661 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1662 def : Pat<(store (v1i64 FPR64:$Rn), ro_indexed64:$addr),
1663 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1665 // Match all store 128 bits width whose type is compatible with FPR128
1666 let Predicates = [IsLE] in {
1667 // We must use ST1 to store vectors in big-endian.
1668 def : Pat<(store (v4f32 FPR128:$Rn), ro_indexed128:$addr),
1669 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1670 def : Pat<(store (v2f64 FPR128:$Rn), ro_indexed128:$addr),
1671 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1672 def : Pat<(store (v16i8 FPR128:$Rn), ro_indexed128:$addr),
1673 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1674 def : Pat<(store (v8i16 FPR128:$Rn), ro_indexed128:$addr),
1675 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1676 def : Pat<(store (v4i32 FPR128:$Rn), ro_indexed128:$addr),
1677 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1678 def : Pat<(store (v2i64 FPR128:$Rn), ro_indexed128:$addr),
1679 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1681 def : Pat<(store (f128 FPR128:$Rn), ro_indexed128:$addr),
1682 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1685 // (unsigned immediate)
1686 def STRXui : StoreUI<0b11, 0, 0b00, GPR64, am_indexed64, "str",
1687 [(store GPR64:$Rt, am_indexed64:$addr)]>;
1688 def STRWui : StoreUI<0b10, 0, 0b00, GPR32, am_indexed32, "str",
1689 [(store GPR32:$Rt, am_indexed32:$addr)]>;
1690 def STRBui : StoreUI<0b00, 1, 0b00, FPR8, am_indexed8, "str",
1691 [(store FPR8:$Rt, am_indexed8:$addr)]>;
1692 def STRHui : StoreUI<0b01, 1, 0b00, FPR16, am_indexed16, "str",
1693 [(store (f16 FPR16:$Rt), am_indexed16:$addr)]>;
1694 def STRSui : StoreUI<0b10, 1, 0b00, FPR32, am_indexed32, "str",
1695 [(store (f32 FPR32:$Rt), am_indexed32:$addr)]>;
1696 def STRDui : StoreUI<0b11, 1, 0b00, FPR64, am_indexed64, "str",
1697 [(store (f64 FPR64:$Rt), am_indexed64:$addr)]>;
1698 def STRQui : StoreUI<0b00, 1, 0b10, FPR128, am_indexed128, "str", []> {
1702 // Match all store 64 bits width whose type is compatible with FPR64
1703 let Predicates = [IsLE] in {
1704 // We must use ST1 to store vectors in big-endian.
1705 def : Pat<(store (v2f32 FPR64:$Rn), am_indexed64:$addr),
1706 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1707 def : Pat<(store (v8i8 FPR64:$Rn), am_indexed64:$addr),
1708 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1709 def : Pat<(store (v4i16 FPR64:$Rn), am_indexed64:$addr),
1710 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1711 def : Pat<(store (v2i32 FPR64:$Rn), am_indexed64:$addr),
1712 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1714 def : Pat<(store (v1f64 FPR64:$Rn), am_indexed64:$addr),
1715 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1716 def : Pat<(store (v1i64 FPR64:$Rn), am_indexed64:$addr),
1717 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1719 // Match all store 128 bits width whose type is compatible with FPR128
1720 let Predicates = [IsLE] in {
1721 // We must use ST1 to store vectors in big-endian.
1722 def : Pat<(store (v4f32 FPR128:$Rn), am_indexed128:$addr),
1723 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1724 def : Pat<(store (v2f64 FPR128:$Rn), am_indexed128:$addr),
1725 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1726 def : Pat<(store (v16i8 FPR128:$Rn), am_indexed128:$addr),
1727 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1728 def : Pat<(store (v8i16 FPR128:$Rn), am_indexed128:$addr),
1729 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1730 def : Pat<(store (v4i32 FPR128:$Rn), am_indexed128:$addr),
1731 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1732 def : Pat<(store (v2i64 FPR128:$Rn), am_indexed128:$addr),
1733 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1735 def : Pat<(store (f128 FPR128:$Rn), am_indexed128:$addr),
1736 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1738 def STRHHui : StoreUI<0b01, 0, 0b00, GPR32, am_indexed16, "strh",
1739 [(truncstorei16 GPR32:$Rt, am_indexed16:$addr)]>;
1740 def STRBBui : StoreUI<0b00, 0, 0b00, GPR32, am_indexed8, "strb",
1741 [(truncstorei8 GPR32:$Rt, am_indexed8:$addr)]>;
1744 def : Pat<(truncstorei32 GPR64:$Rt, am_indexed32:$addr),
1745 (STRWui (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_indexed32:$addr)>;
1746 def : Pat<(truncstorei16 GPR64:$Rt, am_indexed16:$addr),
1747 (STRHHui (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_indexed16:$addr)>;
1748 def : Pat<(truncstorei8 GPR64:$Rt, am_indexed8:$addr),
1749 (STRBBui (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_indexed8:$addr)>;
1751 } // AddedComplexity = 10
1754 // (unscaled immediate)
1755 def STURXi : StoreUnscaled<0b11, 0, 0b00, GPR64, am_unscaled64, "stur",
1756 [(store GPR64:$Rt, am_unscaled64:$addr)]>;
1757 def STURWi : StoreUnscaled<0b10, 0, 0b00, GPR32, am_unscaled32, "stur",
1758 [(store GPR32:$Rt, am_unscaled32:$addr)]>;
1759 def STURBi : StoreUnscaled<0b00, 1, 0b00, FPR8, am_unscaled8, "stur",
1760 [(store FPR8:$Rt, am_unscaled8:$addr)]>;
1761 def STURHi : StoreUnscaled<0b01, 1, 0b00, FPR16, am_unscaled16, "stur",
1762 [(store (f16 FPR16:$Rt), am_unscaled16:$addr)]>;
1763 def STURSi : StoreUnscaled<0b10, 1, 0b00, FPR32, am_unscaled32, "stur",
1764 [(store (f32 FPR32:$Rt), am_unscaled32:$addr)]>;
1765 def STURDi : StoreUnscaled<0b11, 1, 0b00, FPR64, am_unscaled64, "stur",
1766 [(store (f64 FPR64:$Rt), am_unscaled64:$addr)]>;
1767 def STURQi : StoreUnscaled<0b00, 1, 0b10, FPR128, am_unscaled128, "stur",
1768 [(store (f128 FPR128:$Rt), am_unscaled128:$addr)]>;
1769 def STURHHi : StoreUnscaled<0b01, 0, 0b00, GPR32, am_unscaled16, "sturh",
1770 [(truncstorei16 GPR32:$Rt, am_unscaled16:$addr)]>;
1771 def STURBBi : StoreUnscaled<0b00, 0, 0b00, GPR32, am_unscaled8, "sturb",
1772 [(truncstorei8 GPR32:$Rt, am_unscaled8:$addr)]>;
1774 // Match all store 64 bits width whose type is compatible with FPR64
1775 let Predicates = [IsLE] in {
1776 // We must use ST1 to store vectors in big-endian.
1777 def : Pat<(store (v2f32 FPR64:$Rn), am_unscaled64:$addr),
1778 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1779 def : Pat<(store (v8i8 FPR64:$Rn), am_unscaled64:$addr),
1780 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1781 def : Pat<(store (v4i16 FPR64:$Rn), am_unscaled64:$addr),
1782 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1783 def : Pat<(store (v2i32 FPR64:$Rn), am_unscaled64:$addr),
1784 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1786 def : Pat<(store (v1f64 FPR64:$Rn), am_unscaled64:$addr),
1787 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1788 def : Pat<(store (v1i64 FPR64:$Rn), am_unscaled64:$addr),
1789 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1791 // Match all store 128 bits width whose type is compatible with FPR128
1792 let Predicates = [IsLE] in {
1793 // We must use ST1 to store vectors in big-endian.
1794 def : Pat<(store (v4f32 FPR128:$Rn), am_unscaled128:$addr),
1795 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1796 def : Pat<(store (v2f64 FPR128:$Rn), am_unscaled128:$addr),
1797 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1798 def : Pat<(store (v16i8 FPR128:$Rn), am_unscaled128:$addr),
1799 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1800 def : Pat<(store (v8i16 FPR128:$Rn), am_unscaled128:$addr),
1801 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1802 def : Pat<(store (v4i32 FPR128:$Rn), am_unscaled128:$addr),
1803 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1804 def : Pat<(store (v2i64 FPR128:$Rn), am_unscaled128:$addr),
1805 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1806 def : Pat<(store (v2f64 FPR128:$Rn), am_unscaled128:$addr),
1807 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1810 // unscaled i64 truncating stores
1811 def : Pat<(truncstorei32 GPR64:$Rt, am_unscaled32:$addr),
1812 (STURWi (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_unscaled32:$addr)>;
1813 def : Pat<(truncstorei16 GPR64:$Rt, am_unscaled16:$addr),
1814 (STURHHi (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_unscaled16:$addr)>;
1815 def : Pat<(truncstorei8 GPR64:$Rt, am_unscaled8:$addr),
1816 (STURBBi (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_unscaled8:$addr)>;
1819 // STR mnemonics fall back to STUR for negative or unaligned offsets.
1820 def : InstAlias<"str $Rt, $addr", (STURXi GPR64:$Rt, am_unscaled_fb64:$addr)>;
1821 def : InstAlias<"str $Rt, $addr", (STURWi GPR32:$Rt, am_unscaled_fb32:$addr)>;
1822 def : InstAlias<"str $Rt, $addr", (STURBi FPR8:$Rt, am_unscaled_fb8:$addr)>;
1823 def : InstAlias<"str $Rt, $addr", (STURHi FPR16:$Rt, am_unscaled_fb16:$addr)>;
1824 def : InstAlias<"str $Rt, $addr", (STURSi FPR32:$Rt, am_unscaled_fb32:$addr)>;
1825 def : InstAlias<"str $Rt, $addr", (STURDi FPR64:$Rt, am_unscaled_fb64:$addr)>;
1826 def : InstAlias<"str $Rt, $addr", (STURQi FPR128:$Rt, am_unscaled_fb128:$addr)>;
1828 def : InstAlias<"strb $Rt, $addr", (STURBBi GPR32:$Rt, am_unscaled_fb8:$addr)>;
1829 def : InstAlias<"strh $Rt, $addr", (STURHHi GPR32:$Rt, am_unscaled_fb16:$addr)>;
1832 // (unscaled immediate, unprivileged)
1833 def STTRWi : StoreUnprivileged<0b10, 0, 0b00, GPR32, "sttr">;
1834 def STTRXi : StoreUnprivileged<0b11, 0, 0b00, GPR64, "sttr">;
1836 def STTRHi : StoreUnprivileged<0b01, 0, 0b00, GPR32, "sttrh">;
1837 def STTRBi : StoreUnprivileged<0b00, 0, 0b00, GPR32, "sttrb">;
1840 // (immediate pre-indexed)
1841 def STRWpre : StorePreIdx<0b10, 0, 0b00, GPR32, "str">;
1842 def STRXpre : StorePreIdx<0b11, 0, 0b00, GPR64, "str">;
1843 def STRBpre : StorePreIdx<0b00, 1, 0b00, FPR8, "str">;
1844 def STRHpre : StorePreIdx<0b01, 1, 0b00, FPR16, "str">;
1845 def STRSpre : StorePreIdx<0b10, 1, 0b00, FPR32, "str">;
1846 def STRDpre : StorePreIdx<0b11, 1, 0b00, FPR64, "str">;
1847 def STRQpre : StorePreIdx<0b00, 1, 0b10, FPR128, "str">;
1849 def STRBBpre : StorePreIdx<0b00, 0, 0b00, GPR32, "strb">;
1850 def STRHHpre : StorePreIdx<0b01, 0, 0b00, GPR32, "strh">;
1852 // ISel pseudos and patterns. See expanded comment on StorePreIdxPseudo.
1853 defm STRQpre : StorePreIdxPseudo<FPR128, f128, pre_store>;
1854 defm STRDpre : StorePreIdxPseudo<FPR64, f64, pre_store>;
1855 defm STRSpre : StorePreIdxPseudo<FPR32, f32, pre_store>;
1856 defm STRXpre : StorePreIdxPseudo<GPR64, i64, pre_store>;
1857 defm STRWpre : StorePreIdxPseudo<GPR32, i32, pre_store>;
1858 defm STRHHpre : StorePreIdxPseudo<GPR32, i32, pre_truncsti16>;
1859 defm STRBBpre : StorePreIdxPseudo<GPR32, i32, pre_truncsti8>;
1861 def : Pat<(pre_truncsti32 GPR64:$Rt, am_noindex:$addr, simm9:$off),
1862 (STRWpre_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
1864 def : Pat<(pre_truncsti16 GPR64:$Rt, am_noindex:$addr, simm9:$off),
1865 (STRHHpre_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
1867 def : Pat<(pre_truncsti8 GPR64:$Rt, am_noindex:$addr, simm9:$off),
1868 (STRBBpre_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
1871 def : Pat<(pre_store (v8i8 FPR64:$Rt), am_noindex:$addr, simm9:$off),
1872 (STRDpre_isel FPR64:$Rt, am_noindex:$addr, simm9:$off)>;
1873 def : Pat<(pre_store (v4i16 FPR64:$Rt), am_noindex:$addr, simm9:$off),
1874 (STRDpre_isel FPR64:$Rt, am_noindex:$addr, simm9:$off)>;
1875 def : Pat<(pre_store (v2i32 FPR64:$Rt), am_noindex:$addr, simm9:$off),
1876 (STRDpre_isel FPR64:$Rt, am_noindex:$addr, simm9:$off)>;
1877 def : Pat<(pre_store (v2f32 FPR64:$Rt), am_noindex:$addr, simm9:$off),
1878 (STRDpre_isel FPR64:$Rt, am_noindex:$addr, simm9:$off)>;
1879 def : Pat<(pre_store (v1i64 FPR64:$Rt), am_noindex:$addr, simm9:$off),
1880 (STRDpre_isel FPR64:$Rt, am_noindex:$addr, simm9:$off)>;
1881 def : Pat<(pre_store (v1f64 FPR64:$Rt), am_noindex:$addr, simm9:$off),
1882 (STRDpre_isel FPR64:$Rt, am_noindex:$addr, simm9:$off)>;
1884 def : Pat<(pre_store (v16i8 FPR128:$Rt), am_noindex:$addr, simm9:$off),
1885 (STRQpre_isel FPR128:$Rt, am_noindex:$addr, simm9:$off)>;
1886 def : Pat<(pre_store (v8i16 FPR128:$Rt), am_noindex:$addr, simm9:$off),
1887 (STRQpre_isel FPR128:$Rt, am_noindex:$addr, simm9:$off)>;
1888 def : Pat<(pre_store (v4i32 FPR128:$Rt), am_noindex:$addr, simm9:$off),
1889 (STRQpre_isel FPR128:$Rt, am_noindex:$addr, simm9:$off)>;
1890 def : Pat<(pre_store (v4f32 FPR128:$Rt), am_noindex:$addr, simm9:$off),
1891 (STRQpre_isel FPR128:$Rt, am_noindex:$addr, simm9:$off)>;
1892 def : Pat<(pre_store (v2i64 FPR128:$Rt), am_noindex:$addr, simm9:$off),
1893 (STRQpre_isel FPR128:$Rt, am_noindex:$addr, simm9:$off)>;
1894 def : Pat<(pre_store (v2f64 FPR128:$Rt), am_noindex:$addr, simm9:$off),
1895 (STRQpre_isel FPR128:$Rt, am_noindex:$addr, simm9:$off)>;
1898 // (immediate post-indexed)
1899 def STRWpost : StorePostIdx<0b10, 0, 0b00, GPR32, "str">;
1900 def STRXpost : StorePostIdx<0b11, 0, 0b00, GPR64, "str">;
1901 def STRBpost : StorePostIdx<0b00, 1, 0b00, FPR8, "str">;
1902 def STRHpost : StorePostIdx<0b01, 1, 0b00, FPR16, "str">;
1903 def STRSpost : StorePostIdx<0b10, 1, 0b00, FPR32, "str">;
1904 def STRDpost : StorePostIdx<0b11, 1, 0b00, FPR64, "str">;
1905 def STRQpost : StorePostIdx<0b00, 1, 0b10, FPR128, "str">;
1907 def STRBBpost : StorePostIdx<0b00, 0, 0b00, GPR32, "strb">;
1908 def STRHHpost : StorePostIdx<0b01, 0, 0b00, GPR32, "strh">;
1910 // ISel pseudos and patterns. See expanded comment on StorePostIdxPseudo.
1911 defm STRQpost : StorePostIdxPseudo<FPR128, f128, post_store, STRQpost>;
1912 defm STRDpost : StorePostIdxPseudo<FPR64, f64, post_store, STRDpost>;
1913 defm STRSpost : StorePostIdxPseudo<FPR32, f32, post_store, STRSpost>;
1914 defm STRXpost : StorePostIdxPseudo<GPR64, i64, post_store, STRXpost>;
1915 defm STRWpost : StorePostIdxPseudo<GPR32, i32, post_store, STRWpost>;
1916 defm STRHHpost : StorePostIdxPseudo<GPR32, i32, post_truncsti16, STRHHpost>;
1917 defm STRBBpost : StorePostIdxPseudo<GPR32, i32, post_truncsti8, STRBBpost>;
1919 def : Pat<(post_truncsti32 GPR64:$Rt, am_noindex:$addr, simm9:$off),
1920 (STRWpost_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
1922 def : Pat<(post_truncsti16 GPR64:$Rt, am_noindex:$addr, simm9:$off),
1923 (STRHHpost_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
1925 def : Pat<(post_truncsti8 GPR64:$Rt, am_noindex:$addr, simm9:$off),
1926 (STRBBpost_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
1929 def : Pat<(post_store (v8i8 FPR64:$Rt), am_noindex:$addr, simm9:$off),
1930 (STRDpost_isel FPR64:$Rt, am_noindex:$addr, simm9:$off)>;
1931 def : Pat<(post_store (v4i16 FPR64:$Rt), am_noindex:$addr, simm9:$off),
1932 (STRDpost_isel FPR64:$Rt, am_noindex:$addr, simm9:$off)>;
1933 def : Pat<(post_store (v2i32 FPR64:$Rt), am_noindex:$addr, simm9:$off),
1934 (STRDpost_isel FPR64:$Rt, am_noindex:$addr, simm9:$off)>;
1935 def : Pat<(post_store (v2f32 FPR64:$Rt), am_noindex:$addr, simm9:$off),
1936 (STRDpost_isel FPR64:$Rt, am_noindex:$addr, simm9:$off)>;
1937 def : Pat<(post_store (v1i64 FPR64:$Rt), am_noindex:$addr, simm9:$off),
1938 (STRDpost_isel FPR64:$Rt, am_noindex:$addr, simm9:$off)>;
1939 def : Pat<(post_store (v1f64 FPR64:$Rt), am_noindex:$addr, simm9:$off),
1940 (STRDpost_isel FPR64:$Rt, am_noindex:$addr, simm9:$off)>;
1942 def : Pat<(post_store (v16i8 FPR128:$Rt), am_noindex:$addr, simm9:$off),
1943 (STRQpost_isel FPR128:$Rt, am_noindex:$addr, simm9:$off)>;
1944 def : Pat<(post_store (v8i16 FPR128:$Rt), am_noindex:$addr, simm9:$off),
1945 (STRQpost_isel FPR128:$Rt, am_noindex:$addr, simm9:$off)>;
1946 def : Pat<(post_store (v4i32 FPR128:$Rt), am_noindex:$addr, simm9:$off),
1947 (STRQpost_isel FPR128:$Rt, am_noindex:$addr, simm9:$off)>;
1948 def : Pat<(post_store (v4f32 FPR128:$Rt), am_noindex:$addr, simm9:$off),
1949 (STRQpost_isel FPR128:$Rt, am_noindex:$addr, simm9:$off)>;
1950 def : Pat<(post_store (v2i64 FPR128:$Rt), am_noindex:$addr, simm9:$off),
1951 (STRQpost_isel FPR128:$Rt, am_noindex:$addr, simm9:$off)>;
1952 def : Pat<(post_store (v2f64 FPR128:$Rt), am_noindex:$addr, simm9:$off),
1953 (STRQpost_isel FPR128:$Rt, am_noindex:$addr, simm9:$off)>;
1955 //===----------------------------------------------------------------------===//
1956 // Load/store exclusive instructions.
1957 //===----------------------------------------------------------------------===//
1959 def LDARW : LoadAcquire <0b10, 1, 1, 0, 1, GPR32, "ldar">;
1960 def LDARX : LoadAcquire <0b11, 1, 1, 0, 1, GPR64, "ldar">;
1961 def LDARB : LoadAcquire <0b00, 1, 1, 0, 1, GPR32, "ldarb">;
1962 def LDARH : LoadAcquire <0b01, 1, 1, 0, 1, GPR32, "ldarh">;
1964 def LDAXRW : LoadExclusive <0b10, 0, 1, 0, 1, GPR32, "ldaxr">;
1965 def LDAXRX : LoadExclusive <0b11, 0, 1, 0, 1, GPR64, "ldaxr">;
1966 def LDAXRB : LoadExclusive <0b00, 0, 1, 0, 1, GPR32, "ldaxrb">;
1967 def LDAXRH : LoadExclusive <0b01, 0, 1, 0, 1, GPR32, "ldaxrh">;
1969 def LDXRW : LoadExclusive <0b10, 0, 1, 0, 0, GPR32, "ldxr">;
1970 def LDXRX : LoadExclusive <0b11, 0, 1, 0, 0, GPR64, "ldxr">;
1971 def LDXRB : LoadExclusive <0b00, 0, 1, 0, 0, GPR32, "ldxrb">;
1972 def LDXRH : LoadExclusive <0b01, 0, 1, 0, 0, GPR32, "ldxrh">;
1974 def STLRW : StoreRelease <0b10, 1, 0, 0, 1, GPR32, "stlr">;
1975 def STLRX : StoreRelease <0b11, 1, 0, 0, 1, GPR64, "stlr">;
1976 def STLRB : StoreRelease <0b00, 1, 0, 0, 1, GPR32, "stlrb">;
1977 def STLRH : StoreRelease <0b01, 1, 0, 0, 1, GPR32, "stlrh">;
1979 def STLXRW : StoreExclusive<0b10, 0, 0, 0, 1, GPR32, "stlxr">;
1980 def STLXRX : StoreExclusive<0b11, 0, 0, 0, 1, GPR64, "stlxr">;
1981 def STLXRB : StoreExclusive<0b00, 0, 0, 0, 1, GPR32, "stlxrb">;
1982 def STLXRH : StoreExclusive<0b01, 0, 0, 0, 1, GPR32, "stlxrh">;
1984 def STXRW : StoreExclusive<0b10, 0, 0, 0, 0, GPR32, "stxr">;
1985 def STXRX : StoreExclusive<0b11, 0, 0, 0, 0, GPR64, "stxr">;
1986 def STXRB : StoreExclusive<0b00, 0, 0, 0, 0, GPR32, "stxrb">;
1987 def STXRH : StoreExclusive<0b01, 0, 0, 0, 0, GPR32, "stxrh">;
1989 def LDAXPW : LoadExclusivePair<0b10, 0, 1, 1, 1, GPR32, "ldaxp">;
1990 def LDAXPX : LoadExclusivePair<0b11, 0, 1, 1, 1, GPR64, "ldaxp">;
1992 def LDXPW : LoadExclusivePair<0b10, 0, 1, 1, 0, GPR32, "ldxp">;
1993 def LDXPX : LoadExclusivePair<0b11, 0, 1, 1, 0, GPR64, "ldxp">;
1995 def STLXPW : StoreExclusivePair<0b10, 0, 0, 1, 1, GPR32, "stlxp">;
1996 def STLXPX : StoreExclusivePair<0b11, 0, 0, 1, 1, GPR64, "stlxp">;
1998 def STXPW : StoreExclusivePair<0b10, 0, 0, 1, 0, GPR32, "stxp">;
1999 def STXPX : StoreExclusivePair<0b11, 0, 0, 1, 0, GPR64, "stxp">;
2001 //===----------------------------------------------------------------------===//
2002 // Scaled floating point to integer conversion instructions.
2003 //===----------------------------------------------------------------------===//
2005 defm FCVTAS : FPToIntegerUnscaled<0b00, 0b100, "fcvtas", int_arm64_neon_fcvtas>;
2006 defm FCVTAU : FPToIntegerUnscaled<0b00, 0b101, "fcvtau", int_arm64_neon_fcvtau>;
2007 defm FCVTMS : FPToIntegerUnscaled<0b10, 0b000, "fcvtms", int_arm64_neon_fcvtms>;
2008 defm FCVTMU : FPToIntegerUnscaled<0b10, 0b001, "fcvtmu", int_arm64_neon_fcvtmu>;
2009 defm FCVTNS : FPToIntegerUnscaled<0b00, 0b000, "fcvtns", int_arm64_neon_fcvtns>;
2010 defm FCVTNU : FPToIntegerUnscaled<0b00, 0b001, "fcvtnu", int_arm64_neon_fcvtnu>;
2011 defm FCVTPS : FPToIntegerUnscaled<0b01, 0b000, "fcvtps", int_arm64_neon_fcvtps>;
2012 defm FCVTPU : FPToIntegerUnscaled<0b01, 0b001, "fcvtpu", int_arm64_neon_fcvtpu>;
2013 defm FCVTZS : FPToIntegerUnscaled<0b11, 0b000, "fcvtzs", fp_to_sint>;
2014 defm FCVTZU : FPToIntegerUnscaled<0b11, 0b001, "fcvtzu", fp_to_uint>;
2015 defm FCVTZS : FPToIntegerScaled<0b11, 0b000, "fcvtzs", fp_to_sint>;
2016 defm FCVTZU : FPToIntegerScaled<0b11, 0b001, "fcvtzu", fp_to_uint>;
2017 let isCodeGenOnly = 1 in {
2018 defm FCVTZS_Int : FPToIntegerUnscaled<0b11, 0b000, "fcvtzs", int_arm64_neon_fcvtzs>;
2019 defm FCVTZU_Int : FPToIntegerUnscaled<0b11, 0b001, "fcvtzu", int_arm64_neon_fcvtzu>;
2020 defm FCVTZS_Int : FPToIntegerScaled<0b11, 0b000, "fcvtzs", int_arm64_neon_fcvtzs>;
2021 defm FCVTZU_Int : FPToIntegerScaled<0b11, 0b001, "fcvtzu", int_arm64_neon_fcvtzu>;
2024 //===----------------------------------------------------------------------===//
2025 // Scaled integer to floating point conversion instructions.
2026 //===----------------------------------------------------------------------===//
2028 defm SCVTF : IntegerToFP<0, "scvtf", sint_to_fp>;
2029 defm UCVTF : IntegerToFP<1, "ucvtf", uint_to_fp>;
2031 //===----------------------------------------------------------------------===//
2032 // Unscaled integer to floating point conversion instruction.
2033 //===----------------------------------------------------------------------===//
2035 defm FMOV : UnscaledConversion<"fmov">;
2037 def : Pat<(f32 (fpimm0)), (FMOVWSr WZR)>, Requires<[NoZCZ]>;
2038 def : Pat<(f64 (fpimm0)), (FMOVXDr XZR)>, Requires<[NoZCZ]>;
2040 def : Pat<(v8i8 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
2041 def : Pat<(v4i16 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
2042 def : Pat<(v2i32 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
2043 def : Pat<(v1i64 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
2044 def : Pat<(v2f32 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
2045 def : Pat<(v1f64 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
2046 def : Pat<(v1i64 (scalar_to_vector GPR64:$Xn)),
2047 (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
2048 def : Pat<(v1f64 (scalar_to_vector GPR64:$Xn)),
2049 (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
2050 def : Pat<(v1f64 (scalar_to_vector (f64 FPR64:$Xn))), (v1f64 FPR64:$Xn)>;
2052 def : Pat<(i64 (bitconvert (v8i8 V64:$Vn))),
2053 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
2054 def : Pat<(i64 (bitconvert (v4i16 V64:$Vn))),
2055 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
2056 def : Pat<(i64 (bitconvert (v2i32 V64:$Vn))),
2057 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
2058 def : Pat<(i64 (bitconvert (v1i64 V64:$Vn))),
2059 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
2060 def : Pat<(i64 (bitconvert (v2f32 V64:$Vn))),
2061 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
2062 def : Pat<(i64 (bitconvert (v1f64 V64:$Vn))),
2063 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
2065 def : Pat<(f32 (bitconvert (i32 GPR32:$Xn))),
2066 (COPY_TO_REGCLASS GPR32:$Xn, FPR32)>;
2067 def : Pat<(i32 (bitconvert (f32 FPR32:$Xn))),
2068 (COPY_TO_REGCLASS FPR32:$Xn, GPR32)>;
2069 def : Pat<(f64 (bitconvert (i64 GPR64:$Xn))),
2070 (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
2071 def : Pat<(i64 (bitconvert (f64 FPR64:$Xn))),
2072 (COPY_TO_REGCLASS FPR64:$Xn, GPR64)>;
2074 //===----------------------------------------------------------------------===//
2075 // Floating point conversion instruction.
2076 //===----------------------------------------------------------------------===//
2078 defm FCVT : FPConversion<"fcvt">;
2080 def : Pat<(f32_to_f16 FPR32:$Rn),
2081 (i32 (COPY_TO_REGCLASS
2082 (f32 (SUBREG_TO_REG (i32 0), (FCVTHSr FPR32:$Rn), hsub)),
2085 def FCVTSHpseudo : Pseudo<(outs FPR32:$Rd), (ins FPR32:$Rn),
2086 [(set (f32 FPR32:$Rd), (f16_to_f32 i32:$Rn))]>;
2088 //===----------------------------------------------------------------------===//
2089 // Floating point single operand instructions.
2090 //===----------------------------------------------------------------------===//
2092 defm FABS : SingleOperandFPData<0b0001, "fabs", fabs>;
2093 defm FMOV : SingleOperandFPData<0b0000, "fmov">;
2094 defm FNEG : SingleOperandFPData<0b0010, "fneg", fneg>;
2095 defm FRINTA : SingleOperandFPData<0b1100, "frinta", frnd>;
2096 defm FRINTI : SingleOperandFPData<0b1111, "frinti", fnearbyint>;
2097 defm FRINTM : SingleOperandFPData<0b1010, "frintm", ffloor>;
2098 defm FRINTN : SingleOperandFPData<0b1000, "frintn", int_arm64_neon_frintn>;
2099 defm FRINTP : SingleOperandFPData<0b1001, "frintp", fceil>;
2101 def : Pat<(v1f64 (int_arm64_neon_frintn (v1f64 FPR64:$Rn))),
2102 (FRINTNDr FPR64:$Rn)>;
2104 // FRINTX is inserted to set the flags as required by FENV_ACCESS ON behavior
2105 // in the C spec. Setting hasSideEffects ensures it is not DCE'd.
2106 // <rdar://problem/13715968>
2107 // TODO: We should really model the FPSR flags correctly. This is really ugly.
2108 let hasSideEffects = 1 in {
2109 defm FRINTX : SingleOperandFPData<0b1110, "frintx", frint>;
2112 defm FRINTZ : SingleOperandFPData<0b1011, "frintz", ftrunc>;
2114 let SchedRW = [WriteFDiv] in {
2115 defm FSQRT : SingleOperandFPData<0b0011, "fsqrt", fsqrt>;
2118 //===----------------------------------------------------------------------===//
2119 // Floating point two operand instructions.
2120 //===----------------------------------------------------------------------===//
2122 defm FADD : TwoOperandFPData<0b0010, "fadd", fadd>;
2123 let SchedRW = [WriteFDiv] in {
2124 defm FDIV : TwoOperandFPData<0b0001, "fdiv", fdiv>;
2126 defm FMAXNM : TwoOperandFPData<0b0110, "fmaxnm", int_arm64_neon_fmaxnm>;
2127 defm FMAX : TwoOperandFPData<0b0100, "fmax", ARM64fmax>;
2128 defm FMINNM : TwoOperandFPData<0b0111, "fminnm", int_arm64_neon_fminnm>;
2129 defm FMIN : TwoOperandFPData<0b0101, "fmin", ARM64fmin>;
2130 let SchedRW = [WriteFMul] in {
2131 defm FMUL : TwoOperandFPData<0b0000, "fmul", fmul>;
2132 defm FNMUL : TwoOperandFPDataNeg<0b1000, "fnmul", fmul>;
2134 defm FSUB : TwoOperandFPData<0b0011, "fsub", fsub>;
2136 def : Pat<(v1f64 (ARM64fmax (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2137 (FMAXDrr FPR64:$Rn, FPR64:$Rm)>;
2138 def : Pat<(v1f64 (ARM64fmin (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2139 (FMINDrr FPR64:$Rn, FPR64:$Rm)>;
2140 def : Pat<(v1f64 (int_arm64_neon_fmaxnm (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2141 (FMAXNMDrr FPR64:$Rn, FPR64:$Rm)>;
2142 def : Pat<(v1f64 (int_arm64_neon_fminnm (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2143 (FMINNMDrr FPR64:$Rn, FPR64:$Rm)>;
2145 //===----------------------------------------------------------------------===//
2146 // Floating point three operand instructions.
2147 //===----------------------------------------------------------------------===//
2149 defm FMADD : ThreeOperandFPData<0, 0, "fmadd", fma>;
2150 defm FMSUB : ThreeOperandFPData<0, 1, "fmsub",
2151 TriOpFrag<(fma node:$LHS, (fneg node:$MHS), node:$RHS)> >;
2152 defm FNMADD : ThreeOperandFPData<1, 0, "fnmadd",
2153 TriOpFrag<(fneg (fma node:$LHS, node:$MHS, node:$RHS))> >;
2154 defm FNMSUB : ThreeOperandFPData<1, 1, "fnmsub",
2155 TriOpFrag<(fma node:$LHS, node:$MHS, (fneg node:$RHS))> >;
2157 // The following def pats catch the case where the LHS of an FMA is negated.
2158 // The TriOpFrag above catches the case where the middle operand is negated.
2160 // N.b. FMSUB etc have the accumulator at the *end* of (outs), unlike
2161 // the NEON variant.
2162 def : Pat<(f32 (fma (fneg FPR32:$Rn), FPR32:$Rm, FPR32:$Ra)),
2163 (FMSUBSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
2165 def : Pat<(f64 (fma (fneg FPR64:$Rn), FPR64:$Rm, FPR64:$Ra)),
2166 (FMSUBDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
2168 // We handled -(a + b*c) for FNMADD above, now it's time for "(-a) + (-b)*c" and
2170 def : Pat<(f32 (fma (fneg FPR32:$Rn), FPR32:$Rm, (fneg FPR32:$Ra))),
2171 (FNMADDSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
2173 def : Pat<(f64 (fma (fneg FPR64:$Rn), FPR64:$Rm, (fneg FPR64:$Ra))),
2174 (FNMADDDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
2176 def : Pat<(f32 (fma FPR32:$Rn, (fneg FPR32:$Rm), (fneg FPR32:$Ra))),
2177 (FNMADDSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
2179 def : Pat<(f64 (fma FPR64:$Rn, (fneg FPR64:$Rm), (fneg FPR64:$Ra))),
2180 (FNMADDDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
2182 //===----------------------------------------------------------------------===//
2183 // Floating point comparison instructions.
2184 //===----------------------------------------------------------------------===//
2186 defm FCMPE : FPComparison<1, "fcmpe">;
2187 defm FCMP : FPComparison<0, "fcmp", ARM64fcmp>;
2189 //===----------------------------------------------------------------------===//
2190 // Floating point conditional comparison instructions.
2191 //===----------------------------------------------------------------------===//
2193 defm FCCMPE : FPCondComparison<1, "fccmpe">;
2194 defm FCCMP : FPCondComparison<0, "fccmp">;
2196 //===----------------------------------------------------------------------===//
2197 // Floating point conditional select instruction.
2198 //===----------------------------------------------------------------------===//
2200 defm FCSEL : FPCondSelect<"fcsel">;
2202 // CSEL instructions providing f128 types need to be handled by a
2203 // pseudo-instruction since the eventual code will need to introduce basic
2204 // blocks and control flow.
2205 def F128CSEL : Pseudo<(outs FPR128:$Rd),
2206 (ins FPR128:$Rn, FPR128:$Rm, ccode:$cond),
2207 [(set (f128 FPR128:$Rd),
2208 (ARM64csel FPR128:$Rn, FPR128:$Rm,
2209 (i32 imm:$cond), NZCV))]> {
2211 let usesCustomInserter = 1;
2215 //===----------------------------------------------------------------------===//
2216 // Floating point immediate move.
2217 //===----------------------------------------------------------------------===//
2219 let isReMaterializable = 1 in {
2220 defm FMOV : FPMoveImmediate<"fmov">;
2223 //===----------------------------------------------------------------------===//
2224 // Advanced SIMD two vector instructions.
2225 //===----------------------------------------------------------------------===//
2227 defm ABS : SIMDTwoVectorBHSD<0, 0b01011, "abs", int_arm64_neon_abs>;
2228 defm CLS : SIMDTwoVectorBHS<0, 0b00100, "cls", int_arm64_neon_cls>;
2229 defm CLZ : SIMDTwoVectorBHS<1, 0b00100, "clz", ctlz>;
2230 defm CMEQ : SIMDCmpTwoVector<0, 0b01001, "cmeq", ARM64cmeqz>;
2231 defm CMGE : SIMDCmpTwoVector<1, 0b01000, "cmge", ARM64cmgez>;
2232 defm CMGT : SIMDCmpTwoVector<0, 0b01000, "cmgt", ARM64cmgtz>;
2233 defm CMLE : SIMDCmpTwoVector<1, 0b01001, "cmle", ARM64cmlez>;
2234 defm CMLT : SIMDCmpTwoVector<0, 0b01010, "cmlt", ARM64cmltz>;
2235 defm CNT : SIMDTwoVectorB<0, 0b00, 0b00101, "cnt", ctpop>;
2236 defm FABS : SIMDTwoVectorFP<0, 1, 0b01111, "fabs", fabs>;
2238 defm FCMEQ : SIMDFPCmpTwoVector<0, 1, 0b01101, "fcmeq", ARM64fcmeqz>;
2239 defm FCMGE : SIMDFPCmpTwoVector<1, 1, 0b01100, "fcmge", ARM64fcmgez>;
2240 defm FCMGT : SIMDFPCmpTwoVector<0, 1, 0b01100, "fcmgt", ARM64fcmgtz>;
2241 defm FCMLE : SIMDFPCmpTwoVector<1, 1, 0b01101, "fcmle", ARM64fcmlez>;
2242 defm FCMLT : SIMDFPCmpTwoVector<0, 1, 0b01110, "fcmlt", ARM64fcmltz>;
2243 defm FCVTAS : SIMDTwoVectorFPToInt<0,0,0b11100, "fcvtas",int_arm64_neon_fcvtas>;
2244 defm FCVTAU : SIMDTwoVectorFPToInt<1,0,0b11100, "fcvtau",int_arm64_neon_fcvtau>;
2245 defm FCVTL : SIMDFPWidenTwoVector<0, 0, 0b10111, "fcvtl">;
2246 def : Pat<(v4f32 (int_arm64_neon_vcvthf2fp (v4i16 V64:$Rn))),
2247 (FCVTLv4i16 V64:$Rn)>;
2248 def : Pat<(v4f32 (int_arm64_neon_vcvthf2fp (extract_subvector (v8i16 V128:$Rn),
2250 (FCVTLv8i16 V128:$Rn)>;
2251 def : Pat<(v2f64 (fextend (v2f32 V64:$Rn))), (FCVTLv2i32 V64:$Rn)>;
2252 def : Pat<(v2f64 (fextend (v2f32 (extract_subvector (v4f32 V128:$Rn),
2254 (FCVTLv4i32 V128:$Rn)>;
2256 defm FCVTMS : SIMDTwoVectorFPToInt<0,0,0b11011, "fcvtms",int_arm64_neon_fcvtms>;
2257 defm FCVTMU : SIMDTwoVectorFPToInt<1,0,0b11011, "fcvtmu",int_arm64_neon_fcvtmu>;
2258 defm FCVTNS : SIMDTwoVectorFPToInt<0,0,0b11010, "fcvtns",int_arm64_neon_fcvtns>;
2259 defm FCVTNU : SIMDTwoVectorFPToInt<1,0,0b11010, "fcvtnu",int_arm64_neon_fcvtnu>;
2260 defm FCVTN : SIMDFPNarrowTwoVector<0, 0, 0b10110, "fcvtn">;
2261 def : Pat<(v4i16 (int_arm64_neon_vcvtfp2hf (v4f32 V128:$Rn))),
2262 (FCVTNv4i16 V128:$Rn)>;
2263 def : Pat<(concat_vectors V64:$Rd,
2264 (v4i16 (int_arm64_neon_vcvtfp2hf (v4f32 V128:$Rn)))),
2265 (FCVTNv8i16 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
2266 def : Pat<(v2f32 (fround (v2f64 V128:$Rn))), (FCVTNv2i32 V128:$Rn)>;
2267 def : Pat<(concat_vectors V64:$Rd, (v2f32 (fround (v2f64 V128:$Rn)))),
2268 (FCVTNv4i32 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
2269 defm FCVTPS : SIMDTwoVectorFPToInt<0,1,0b11010, "fcvtps",int_arm64_neon_fcvtps>;
2270 defm FCVTPU : SIMDTwoVectorFPToInt<1,1,0b11010, "fcvtpu",int_arm64_neon_fcvtpu>;
2271 defm FCVTXN : SIMDFPInexactCvtTwoVector<1, 0, 0b10110, "fcvtxn",
2272 int_arm64_neon_fcvtxn>;
2273 defm FCVTZS : SIMDTwoVectorFPToInt<0, 1, 0b11011, "fcvtzs", fp_to_sint>;
2274 defm FCVTZU : SIMDTwoVectorFPToInt<1, 1, 0b11011, "fcvtzu", fp_to_uint>;
2275 let isCodeGenOnly = 1 in {
2276 defm FCVTZS_Int : SIMDTwoVectorFPToInt<0, 1, 0b11011, "fcvtzs",
2277 int_arm64_neon_fcvtzs>;
2278 defm FCVTZU_Int : SIMDTwoVectorFPToInt<1, 1, 0b11011, "fcvtzu",
2279 int_arm64_neon_fcvtzu>;
2281 defm FNEG : SIMDTwoVectorFP<1, 1, 0b01111, "fneg", fneg>;
2282 defm FRECPE : SIMDTwoVectorFP<0, 1, 0b11101, "frecpe", int_arm64_neon_frecpe>;
2283 defm FRINTA : SIMDTwoVectorFP<1, 0, 0b11000, "frinta", frnd>;
2284 defm FRINTI : SIMDTwoVectorFP<1, 1, 0b11001, "frinti", fnearbyint>;
2285 defm FRINTM : SIMDTwoVectorFP<0, 0, 0b11001, "frintm", ffloor>;
2286 defm FRINTN : SIMDTwoVectorFP<0, 0, 0b11000, "frintn", int_arm64_neon_frintn>;
2287 defm FRINTP : SIMDTwoVectorFP<0, 1, 0b11000, "frintp", fceil>;
2288 defm FRINTX : SIMDTwoVectorFP<1, 0, 0b11001, "frintx", frint>;
2289 defm FRINTZ : SIMDTwoVectorFP<0, 1, 0b11001, "frintz", ftrunc>;
2290 defm FRSQRTE: SIMDTwoVectorFP<1, 1, 0b11101, "frsqrte", int_arm64_neon_frsqrte>;
2291 defm FSQRT : SIMDTwoVectorFP<1, 1, 0b11111, "fsqrt", fsqrt>;
2292 defm NEG : SIMDTwoVectorBHSD<1, 0b01011, "neg",
2293 UnOpFrag<(sub immAllZerosV, node:$LHS)> >;
2294 defm NOT : SIMDTwoVectorB<1, 0b00, 0b00101, "not", vnot>;
2295 // Aliases for MVN -> NOT.
2296 def : InstAlias<"mvn.8b $Vd, $Vn", (NOTv8i8 V64:$Vd, V64:$Vn)>;
2297 def : InstAlias<"mvn.16b $Vd, $Vn", (NOTv16i8 V128:$Vd, V128:$Vn)>;
2298 def : InstAlias<"mvn $Vd.8b, $Vn.8b", (NOTv8i8 V64:$Vd, V64:$Vn)>;
2299 def : InstAlias<"mvn $Vd.16b, $Vn.16b", (NOTv16i8 V128:$Vd, V128:$Vn)>;
2301 def : Pat<(ARM64neg (v8i8 V64:$Rn)), (NEGv8i8 V64:$Rn)>;
2302 def : Pat<(ARM64neg (v16i8 V128:$Rn)), (NEGv16i8 V128:$Rn)>;
2303 def : Pat<(ARM64neg (v4i16 V64:$Rn)), (NEGv4i16 V64:$Rn)>;
2304 def : Pat<(ARM64neg (v8i16 V128:$Rn)), (NEGv8i16 V128:$Rn)>;
2305 def : Pat<(ARM64neg (v2i32 V64:$Rn)), (NEGv2i32 V64:$Rn)>;
2306 def : Pat<(ARM64neg (v4i32 V128:$Rn)), (NEGv4i32 V128:$Rn)>;
2307 def : Pat<(ARM64neg (v2i64 V128:$Rn)), (NEGv2i64 V128:$Rn)>;
2309 def : Pat<(ARM64not (v8i8 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2310 def : Pat<(ARM64not (v16i8 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2311 def : Pat<(ARM64not (v4i16 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2312 def : Pat<(ARM64not (v8i16 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2313 def : Pat<(ARM64not (v2i32 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2314 def : Pat<(ARM64not (v1i64 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2315 def : Pat<(ARM64not (v4i32 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2316 def : Pat<(ARM64not (v2i64 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2318 def : Pat<(vnot (v4i16 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2319 def : Pat<(vnot (v8i16 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2320 def : Pat<(vnot (v2i32 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2321 def : Pat<(vnot (v4i32 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2322 def : Pat<(vnot (v2i64 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2324 defm RBIT : SIMDTwoVectorB<1, 0b01, 0b00101, "rbit", int_arm64_neon_rbit>;
2325 defm REV16 : SIMDTwoVectorB<0, 0b00, 0b00001, "rev16", ARM64rev16>;
2326 defm REV32 : SIMDTwoVectorBH<1, 0b00000, "rev32", ARM64rev32>;
2327 defm REV64 : SIMDTwoVectorBHS<0, 0b00000, "rev64", ARM64rev64>;
2328 defm SADALP : SIMDLongTwoVectorTied<0, 0b00110, "sadalp",
2329 BinOpFrag<(add node:$LHS, (int_arm64_neon_saddlp node:$RHS))> >;
2330 defm SADDLP : SIMDLongTwoVector<0, 0b00010, "saddlp", int_arm64_neon_saddlp>;
2331 defm SCVTF : SIMDTwoVectorIntToFP<0, 0, 0b11101, "scvtf", sint_to_fp>;
2332 defm SHLL : SIMDVectorLShiftLongBySizeBHS;
2333 defm SQABS : SIMDTwoVectorBHSD<0, 0b00111, "sqabs", int_arm64_neon_sqabs>;
2334 defm SQNEG : SIMDTwoVectorBHSD<1, 0b00111, "sqneg", int_arm64_neon_sqneg>;
2335 defm SQXTN : SIMDMixedTwoVector<0, 0b10100, "sqxtn", int_arm64_neon_sqxtn>;
2336 defm SQXTUN : SIMDMixedTwoVector<1, 0b10010, "sqxtun", int_arm64_neon_sqxtun>;
2337 defm SUQADD : SIMDTwoVectorBHSDTied<0, 0b00011, "suqadd",int_arm64_neon_suqadd>;
2338 defm UADALP : SIMDLongTwoVectorTied<1, 0b00110, "uadalp",
2339 BinOpFrag<(add node:$LHS, (int_arm64_neon_uaddlp node:$RHS))> >;
2340 defm UADDLP : SIMDLongTwoVector<1, 0b00010, "uaddlp",
2341 int_arm64_neon_uaddlp>;
2342 defm UCVTF : SIMDTwoVectorIntToFP<1, 0, 0b11101, "ucvtf", uint_to_fp>;
2343 defm UQXTN : SIMDMixedTwoVector<1, 0b10100, "uqxtn", int_arm64_neon_uqxtn>;
2344 defm URECPE : SIMDTwoVectorS<0, 1, 0b11100, "urecpe", int_arm64_neon_urecpe>;
2345 defm URSQRTE: SIMDTwoVectorS<1, 1, 0b11100, "ursqrte", int_arm64_neon_ursqrte>;
2346 defm USQADD : SIMDTwoVectorBHSDTied<1, 0b00011, "usqadd",int_arm64_neon_usqadd>;
2347 defm XTN : SIMDMixedTwoVector<0, 0b10010, "xtn", trunc>;
2349 def : Pat<(v2f32 (ARM64rev64 V64:$Rn)), (REV64v2i32 V64:$Rn)>;
2350 def : Pat<(v4f32 (ARM64rev64 V128:$Rn)), (REV64v4i32 V128:$Rn)>;
2352 // Patterns for vector long shift (by element width). These need to match all
2353 // three of zext, sext and anyext so it's easier to pull the patterns out of the
2355 multiclass SIMDVectorLShiftLongBySizeBHSPats<SDPatternOperator ext> {
2356 def : Pat<(ARM64vshl (v8i16 (ext (v8i8 V64:$Rn))), (i32 8)),
2357 (SHLLv8i8 V64:$Rn)>;
2358 def : Pat<(ARM64vshl (v8i16 (ext (extract_high_v16i8 V128:$Rn))), (i32 8)),
2359 (SHLLv16i8 V128:$Rn)>;
2360 def : Pat<(ARM64vshl (v4i32 (ext (v4i16 V64:$Rn))), (i32 16)),
2361 (SHLLv4i16 V64:$Rn)>;
2362 def : Pat<(ARM64vshl (v4i32 (ext (extract_high_v8i16 V128:$Rn))), (i32 16)),
2363 (SHLLv8i16 V128:$Rn)>;
2364 def : Pat<(ARM64vshl (v2i64 (ext (v2i32 V64:$Rn))), (i32 32)),
2365 (SHLLv2i32 V64:$Rn)>;
2366 def : Pat<(ARM64vshl (v2i64 (ext (extract_high_v4i32 V128:$Rn))), (i32 32)),
2367 (SHLLv4i32 V128:$Rn)>;
2370 defm : SIMDVectorLShiftLongBySizeBHSPats<anyext>;
2371 defm : SIMDVectorLShiftLongBySizeBHSPats<zext>;
2372 defm : SIMDVectorLShiftLongBySizeBHSPats<sext>;
2374 //===----------------------------------------------------------------------===//
2375 // Advanced SIMD three vector instructions.
2376 //===----------------------------------------------------------------------===//
2378 defm ADD : SIMDThreeSameVector<0, 0b10000, "add", add>;
2379 defm ADDP : SIMDThreeSameVector<0, 0b10111, "addp", int_arm64_neon_addp>;
2380 defm CMEQ : SIMDThreeSameVector<1, 0b10001, "cmeq", ARM64cmeq>;
2381 defm CMGE : SIMDThreeSameVector<0, 0b00111, "cmge", ARM64cmge>;
2382 defm CMGT : SIMDThreeSameVector<0, 0b00110, "cmgt", ARM64cmgt>;
2383 defm CMHI : SIMDThreeSameVector<1, 0b00110, "cmhi", ARM64cmhi>;
2384 defm CMHS : SIMDThreeSameVector<1, 0b00111, "cmhs", ARM64cmhs>;
2385 defm CMTST : SIMDThreeSameVector<0, 0b10001, "cmtst", ARM64cmtst>;
2386 defm FABD : SIMDThreeSameVectorFP<1,1,0b11010,"fabd", int_arm64_neon_fabd>;
2387 defm FACGE : SIMDThreeSameVectorFPCmp<1,0,0b11101,"facge",int_arm64_neon_facge>;
2388 defm FACGT : SIMDThreeSameVectorFPCmp<1,1,0b11101,"facgt",int_arm64_neon_facgt>;
2389 defm FADDP : SIMDThreeSameVectorFP<1,0,0b11010,"faddp",int_arm64_neon_addp>;
2390 defm FADD : SIMDThreeSameVectorFP<0,0,0b11010,"fadd", fadd>;
2391 defm FCMEQ : SIMDThreeSameVectorFPCmp<0, 0, 0b11100, "fcmeq", ARM64fcmeq>;
2392 defm FCMGE : SIMDThreeSameVectorFPCmp<1, 0, 0b11100, "fcmge", ARM64fcmge>;
2393 defm FCMGT : SIMDThreeSameVectorFPCmp<1, 1, 0b11100, "fcmgt", ARM64fcmgt>;
2394 defm FDIV : SIMDThreeSameVectorFP<1,0,0b11111,"fdiv", fdiv>;
2395 defm FMAXNMP : SIMDThreeSameVectorFP<1,0,0b11000,"fmaxnmp", int_arm64_neon_fmaxnmp>;
2396 defm FMAXNM : SIMDThreeSameVectorFP<0,0,0b11000,"fmaxnm", int_arm64_neon_fmaxnm>;
2397 defm FMAXP : SIMDThreeSameVectorFP<1,0,0b11110,"fmaxp", int_arm64_neon_fmaxp>;
2398 defm FMAX : SIMDThreeSameVectorFP<0,0,0b11110,"fmax", ARM64fmax>;
2399 defm FMINNMP : SIMDThreeSameVectorFP<1,1,0b11000,"fminnmp", int_arm64_neon_fminnmp>;
2400 defm FMINNM : SIMDThreeSameVectorFP<0,1,0b11000,"fminnm", int_arm64_neon_fminnm>;
2401 defm FMINP : SIMDThreeSameVectorFP<1,1,0b11110,"fminp", int_arm64_neon_fminp>;
2402 defm FMIN : SIMDThreeSameVectorFP<0,1,0b11110,"fmin", ARM64fmin>;
2404 // NOTE: The operands of the PatFrag are reordered on FMLA/FMLS because the
2405 // instruction expects the addend first, while the fma intrinsic puts it last.
2406 defm FMLA : SIMDThreeSameVectorFPTied<0, 0, 0b11001, "fmla",
2407 TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)> >;
2408 defm FMLS : SIMDThreeSameVectorFPTied<0, 1, 0b11001, "fmls",
2409 TriOpFrag<(fma node:$MHS, (fneg node:$RHS), node:$LHS)> >;
2411 // The following def pats catch the case where the LHS of an FMA is negated.
2412 // The TriOpFrag above catches the case where the middle operand is negated.
2413 def : Pat<(v2f32 (fma (fneg V64:$Rn), V64:$Rm, V64:$Rd)),
2414 (FMLSv2f32 V64:$Rd, V64:$Rn, V64:$Rm)>;
2416 def : Pat<(v4f32 (fma (fneg V128:$Rn), V128:$Rm, V128:$Rd)),
2417 (FMLSv4f32 V128:$Rd, V128:$Rn, V128:$Rm)>;
2419 def : Pat<(v2f64 (fma (fneg V128:$Rn), V128:$Rm, V128:$Rd)),
2420 (FMLSv2f64 V128:$Rd, V128:$Rn, V128:$Rm)>;
2422 defm FMULX : SIMDThreeSameVectorFP<0,0,0b11011,"fmulx", int_arm64_neon_fmulx>;
2423 defm FMUL : SIMDThreeSameVectorFP<1,0,0b11011,"fmul", fmul>;
2424 defm FRECPS : SIMDThreeSameVectorFP<0,0,0b11111,"frecps", int_arm64_neon_frecps>;
2425 defm FRSQRTS : SIMDThreeSameVectorFP<0,1,0b11111,"frsqrts", int_arm64_neon_frsqrts>;
2426 defm FSUB : SIMDThreeSameVectorFP<0,1,0b11010,"fsub", fsub>;
2427 defm MLA : SIMDThreeSameVectorBHSTied<0, 0b10010, "mla",
2428 TriOpFrag<(add node:$LHS, (mul node:$MHS, node:$RHS))> >;
2429 defm MLS : SIMDThreeSameVectorBHSTied<1, 0b10010, "mls",
2430 TriOpFrag<(sub node:$LHS, (mul node:$MHS, node:$RHS))> >;
2431 defm MUL : SIMDThreeSameVectorBHS<0, 0b10011, "mul", mul>;
2432 defm PMUL : SIMDThreeSameVectorB<1, 0b10011, "pmul", int_arm64_neon_pmul>;
2433 defm SABA : SIMDThreeSameVectorBHSTied<0, 0b01111, "saba",
2434 TriOpFrag<(add node:$LHS, (int_arm64_neon_sabd node:$MHS, node:$RHS))> >;
2435 defm SABD : SIMDThreeSameVectorBHS<0,0b01110,"sabd", int_arm64_neon_sabd>;
2436 defm SHADD : SIMDThreeSameVectorBHS<0,0b00000,"shadd", int_arm64_neon_shadd>;
2437 defm SHSUB : SIMDThreeSameVectorBHS<0,0b00100,"shsub", int_arm64_neon_shsub>;
2438 defm SMAXP : SIMDThreeSameVectorBHS<0,0b10100,"smaxp", int_arm64_neon_smaxp>;
2439 defm SMAX : SIMDThreeSameVectorBHS<0,0b01100,"smax", int_arm64_neon_smax>;
2440 defm SMINP : SIMDThreeSameVectorBHS<0,0b10101,"sminp", int_arm64_neon_sminp>;
2441 defm SMIN : SIMDThreeSameVectorBHS<0,0b01101,"smin", int_arm64_neon_smin>;
2442 defm SQADD : SIMDThreeSameVector<0,0b00001,"sqadd", int_arm64_neon_sqadd>;
2443 defm SQDMULH : SIMDThreeSameVectorHS<0,0b10110,"sqdmulh",int_arm64_neon_sqdmulh>;
2444 defm SQRDMULH : SIMDThreeSameVectorHS<1,0b10110,"sqrdmulh",int_arm64_neon_sqrdmulh>;
2445 defm SQRSHL : SIMDThreeSameVector<0,0b01011,"sqrshl", int_arm64_neon_sqrshl>;
2446 defm SQSHL : SIMDThreeSameVector<0,0b01001,"sqshl", int_arm64_neon_sqshl>;
2447 defm SQSUB : SIMDThreeSameVector<0,0b00101,"sqsub", int_arm64_neon_sqsub>;
2448 defm SRHADD : SIMDThreeSameVectorBHS<0,0b00010,"srhadd",int_arm64_neon_srhadd>;
2449 defm SRSHL : SIMDThreeSameVector<0,0b01010,"srshl", int_arm64_neon_srshl>;
2450 defm SSHL : SIMDThreeSameVector<0,0b01000,"sshl", int_arm64_neon_sshl>;
2451 defm SUB : SIMDThreeSameVector<1,0b10000,"sub", sub>;
2452 defm UABA : SIMDThreeSameVectorBHSTied<1, 0b01111, "uaba",
2453 TriOpFrag<(add node:$LHS, (int_arm64_neon_uabd node:$MHS, node:$RHS))> >;
2454 defm UABD : SIMDThreeSameVectorBHS<1,0b01110,"uabd", int_arm64_neon_uabd>;
2455 defm UHADD : SIMDThreeSameVectorBHS<1,0b00000,"uhadd", int_arm64_neon_uhadd>;
2456 defm UHSUB : SIMDThreeSameVectorBHS<1,0b00100,"uhsub", int_arm64_neon_uhsub>;
2457 defm UMAXP : SIMDThreeSameVectorBHS<1,0b10100,"umaxp", int_arm64_neon_umaxp>;
2458 defm UMAX : SIMDThreeSameVectorBHS<1,0b01100,"umax", int_arm64_neon_umax>;
2459 defm UMINP : SIMDThreeSameVectorBHS<1,0b10101,"uminp", int_arm64_neon_uminp>;
2460 defm UMIN : SIMDThreeSameVectorBHS<1,0b01101,"umin", int_arm64_neon_umin>;
2461 defm UQADD : SIMDThreeSameVector<1,0b00001,"uqadd", int_arm64_neon_uqadd>;
2462 defm UQRSHL : SIMDThreeSameVector<1,0b01011,"uqrshl", int_arm64_neon_uqrshl>;
2463 defm UQSHL : SIMDThreeSameVector<1,0b01001,"uqshl", int_arm64_neon_uqshl>;
2464 defm UQSUB : SIMDThreeSameVector<1,0b00101,"uqsub", int_arm64_neon_uqsub>;
2465 defm URHADD : SIMDThreeSameVectorBHS<1,0b00010,"urhadd", int_arm64_neon_urhadd>;
2466 defm URSHL : SIMDThreeSameVector<1,0b01010,"urshl", int_arm64_neon_urshl>;
2467 defm USHL : SIMDThreeSameVector<1,0b01000,"ushl", int_arm64_neon_ushl>;
2469 defm AND : SIMDLogicalThreeVector<0, 0b00, "and", and>;
2470 defm BIC : SIMDLogicalThreeVector<0, 0b01, "bic",
2471 BinOpFrag<(and node:$LHS, (vnot node:$RHS))> >;
2472 defm BIF : SIMDLogicalThreeVector<1, 0b11, "bif">;
2473 defm BIT : SIMDLogicalThreeVectorTied<1, 0b10, "bit", ARM64bit>;
2474 defm BSL : SIMDLogicalThreeVectorTied<1, 0b01, "bsl",
2475 TriOpFrag<(or (and node:$LHS, node:$MHS), (and (vnot node:$LHS), node:$RHS))>>;
2476 defm EOR : SIMDLogicalThreeVector<1, 0b00, "eor", xor>;
2477 defm ORN : SIMDLogicalThreeVector<0, 0b11, "orn",
2478 BinOpFrag<(or node:$LHS, (vnot node:$RHS))> >;
2479 defm ORR : SIMDLogicalThreeVector<0, 0b10, "orr", or>;
2481 def : Pat<(ARM64bsl (v8i8 V64:$Rd), V64:$Rn, V64:$Rm),
2482 (BSLv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
2483 def : Pat<(ARM64bsl (v4i16 V64:$Rd), V64:$Rn, V64:$Rm),
2484 (BSLv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
2485 def : Pat<(ARM64bsl (v2i32 V64:$Rd), V64:$Rn, V64:$Rm),
2486 (BSLv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
2487 def : Pat<(ARM64bsl (v1i64 V64:$Rd), V64:$Rn, V64:$Rm),
2488 (BSLv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
2490 def : Pat<(ARM64bsl (v16i8 V128:$Rd), V128:$Rn, V128:$Rm),
2491 (BSLv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
2492 def : Pat<(ARM64bsl (v8i16 V128:$Rd), V128:$Rn, V128:$Rm),
2493 (BSLv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
2494 def : Pat<(ARM64bsl (v4i32 V128:$Rd), V128:$Rn, V128:$Rm),
2495 (BSLv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
2496 def : Pat<(ARM64bsl (v2i64 V128:$Rd), V128:$Rn, V128:$Rm),
2497 (BSLv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
2499 // FIXME: the .16b and .8b variantes should be emitted by the
2500 // AsmWriter. TableGen's AsmWriter-generator doesn't deal with variant syntaxes
2501 // in aliases yet though.
2502 def : InstAlias<"mov{\t$dst.16b, $src.16b|.16b\t$dst, $src}",
2503 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2504 def : InstAlias<"{mov\t$dst.8h, $src.8h|mov.8h\t$dst, $src}",
2505 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2506 def : InstAlias<"{mov\t$dst.4s, $src.4s|mov.4s\t$dst, $src}",
2507 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2508 def : InstAlias<"{mov\t$dst.2d, $src.2d|mov.2d\t$dst, $src}",
2509 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2511 def : InstAlias<"{mov\t$dst.8b, $src.8b|mov.8b\t$dst, $src}",
2512 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2513 def : InstAlias<"{mov\t$dst.4h, $src.4h|mov.4h\t$dst, $src}",
2514 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2515 def : InstAlias<"{mov\t$dst.2s, $src.2s|mov.2s\t$dst, $src}",
2516 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2517 def : InstAlias<"{mov\t$dst.1d, $src.1d|mov.1d\t$dst, $src}",
2518 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2520 def : InstAlias<"{cmls\t$dst.8b, $src1.8b, $src2.8b" #
2521 "|cmls.8b\t$dst, $src1, $src2}",
2522 (CMHSv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2523 def : InstAlias<"{cmls\t$dst.16b, $src1.16b, $src2.16b" #
2524 "|cmls.16b\t$dst, $src1, $src2}",
2525 (CMHSv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2526 def : InstAlias<"{cmls\t$dst.4h, $src1.4h, $src2.4h" #
2527 "|cmls.4h\t$dst, $src1, $src2}",
2528 (CMHSv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2529 def : InstAlias<"{cmls\t$dst.8h, $src1.8h, $src2.8h" #
2530 "|cmls.8h\t$dst, $src1, $src2}",
2531 (CMHSv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2532 def : InstAlias<"{cmls\t$dst.2s, $src1.2s, $src2.2s" #
2533 "|cmls.2s\t$dst, $src1, $src2}",
2534 (CMHSv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2535 def : InstAlias<"{cmls\t$dst.4s, $src1.4s, $src2.4s" #
2536 "|cmls.4s\t$dst, $src1, $src2}",
2537 (CMHSv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2538 def : InstAlias<"{cmls\t$dst.2d, $src1.2d, $src2.2d" #
2539 "|cmls.2d\t$dst, $src1, $src2}",
2540 (CMHSv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2542 def : InstAlias<"{cmlo\t$dst.8b, $src1.8b, $src2.8b" #
2543 "|cmlo.8b\t$dst, $src1, $src2}",
2544 (CMHIv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2545 def : InstAlias<"{cmlo\t$dst.16b, $src1.16b, $src2.16b" #
2546 "|cmlo.16b\t$dst, $src1, $src2}",
2547 (CMHIv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2548 def : InstAlias<"{cmlo\t$dst.4h, $src1.4h, $src2.4h" #
2549 "|cmlo.4h\t$dst, $src1, $src2}",
2550 (CMHIv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2551 def : InstAlias<"{cmlo\t$dst.8h, $src1.8h, $src2.8h" #
2552 "|cmlo.8h\t$dst, $src1, $src2}",
2553 (CMHIv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2554 def : InstAlias<"{cmlo\t$dst.2s, $src1.2s, $src2.2s" #
2555 "|cmlo.2s\t$dst, $src1, $src2}",
2556 (CMHIv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2557 def : InstAlias<"{cmlo\t$dst.4s, $src1.4s, $src2.4s" #
2558 "|cmlo.4s\t$dst, $src1, $src2}",
2559 (CMHIv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2560 def : InstAlias<"{cmlo\t$dst.2d, $src1.2d, $src2.2d" #
2561 "|cmlo.2d\t$dst, $src1, $src2}",
2562 (CMHIv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2564 def : InstAlias<"{cmle\t$dst.8b, $src1.8b, $src2.8b" #
2565 "|cmle.8b\t$dst, $src1, $src2}",
2566 (CMGEv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2567 def : InstAlias<"{cmle\t$dst.16b, $src1.16b, $src2.16b" #
2568 "|cmle.16b\t$dst, $src1, $src2}",
2569 (CMGEv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2570 def : InstAlias<"{cmle\t$dst.4h, $src1.4h, $src2.4h" #
2571 "|cmle.4h\t$dst, $src1, $src2}",
2572 (CMGEv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2573 def : InstAlias<"{cmle\t$dst.8h, $src1.8h, $src2.8h" #
2574 "|cmle.8h\t$dst, $src1, $src2}",
2575 (CMGEv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2576 def : InstAlias<"{cmle\t$dst.2s, $src1.2s, $src2.2s" #
2577 "|cmle.2s\t$dst, $src1, $src2}",
2578 (CMGEv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2579 def : InstAlias<"{cmle\t$dst.4s, $src1.4s, $src2.4s" #
2580 "|cmle.4s\t$dst, $src1, $src2}",
2581 (CMGEv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2582 def : InstAlias<"{cmle\t$dst.2d, $src1.2d, $src2.2d" #
2583 "|cmle.2d\t$dst, $src1, $src2}",
2584 (CMGEv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2586 def : InstAlias<"{cmlt\t$dst.8b, $src1.8b, $src2.8b" #
2587 "|cmlt.8b\t$dst, $src1, $src2}",
2588 (CMGTv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2589 def : InstAlias<"{cmlt\t$dst.16b, $src1.16b, $src2.16b" #
2590 "|cmlt.16b\t$dst, $src1, $src2}",
2591 (CMGTv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2592 def : InstAlias<"{cmlt\t$dst.4h, $src1.4h, $src2.4h" #
2593 "|cmlt.4h\t$dst, $src1, $src2}",
2594 (CMGTv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2595 def : InstAlias<"{cmlt\t$dst.8h, $src1.8h, $src2.8h" #
2596 "|cmlt.8h\t$dst, $src1, $src2}",
2597 (CMGTv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2598 def : InstAlias<"{cmlt\t$dst.2s, $src1.2s, $src2.2s" #
2599 "|cmlt.2s\t$dst, $src1, $src2}",
2600 (CMGTv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2601 def : InstAlias<"{cmlt\t$dst.4s, $src1.4s, $src2.4s" #
2602 "|cmlt.4s\t$dst, $src1, $src2}",
2603 (CMGTv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2604 def : InstAlias<"{cmlt\t$dst.2d, $src1.2d, $src2.2d" #
2605 "|cmlt.2d\t$dst, $src1, $src2}",
2606 (CMGTv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2608 def : InstAlias<"{fcmle\t$dst.2s, $src1.2s, $src2.2s" #
2609 "|fcmle.2s\t$dst, $src1, $src2}",
2610 (FCMGEv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
2611 def : InstAlias<"{fcmle\t$dst.4s, $src1.4s, $src2.4s" #
2612 "|fcmle.4s\t$dst, $src1, $src2}",
2613 (FCMGEv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
2614 def : InstAlias<"{fcmle\t$dst.2d, $src1.2d, $src2.2d" #
2615 "|fcmle.2d\t$dst, $src1, $src2}",
2616 (FCMGEv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
2618 def : InstAlias<"{fcmlt\t$dst.2s, $src1.2s, $src2.2s" #
2619 "|fcmlt.2s\t$dst, $src1, $src2}",
2620 (FCMGTv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
2621 def : InstAlias<"{fcmlt\t$dst.4s, $src1.4s, $src2.4s" #
2622 "|fcmlt.4s\t$dst, $src1, $src2}",
2623 (FCMGTv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
2624 def : InstAlias<"{fcmlt\t$dst.2d, $src1.2d, $src2.2d" #
2625 "|fcmlt.2d\t$dst, $src1, $src2}",
2626 (FCMGTv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
2628 def : InstAlias<"{facle\t$dst.2s, $src1.2s, $src2.2s" #
2629 "|facle.2s\t$dst, $src1, $src2}",
2630 (FACGEv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
2631 def : InstAlias<"{facle\t$dst.4s, $src1.4s, $src2.4s" #
2632 "|facle.4s\t$dst, $src1, $src2}",
2633 (FACGEv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
2634 def : InstAlias<"{facle\t$dst.2d, $src1.2d, $src2.2d" #
2635 "|facle.2d\t$dst, $src1, $src2}",
2636 (FACGEv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
2638 def : InstAlias<"{faclt\t$dst.2s, $src1.2s, $src2.2s" #
2639 "|faclt.2s\t$dst, $src1, $src2}",
2640 (FACGTv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
2641 def : InstAlias<"{faclt\t$dst.4s, $src1.4s, $src2.4s" #
2642 "|faclt.4s\t$dst, $src1, $src2}",
2643 (FACGTv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
2644 def : InstAlias<"{faclt\t$dst.2d, $src1.2d, $src2.2d" #
2645 "|faclt.2d\t$dst, $src1, $src2}",
2646 (FACGTv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
2648 //===----------------------------------------------------------------------===//
2649 // Advanced SIMD three scalar instructions.
2650 //===----------------------------------------------------------------------===//
2652 defm ADD : SIMDThreeScalarD<0, 0b10000, "add", add>;
2653 defm CMEQ : SIMDThreeScalarD<1, 0b10001, "cmeq", ARM64cmeq>;
2654 defm CMGE : SIMDThreeScalarD<0, 0b00111, "cmge", ARM64cmge>;
2655 defm CMGT : SIMDThreeScalarD<0, 0b00110, "cmgt", ARM64cmgt>;
2656 defm CMHI : SIMDThreeScalarD<1, 0b00110, "cmhi", ARM64cmhi>;
2657 defm CMHS : SIMDThreeScalarD<1, 0b00111, "cmhs", ARM64cmhs>;
2658 defm CMTST : SIMDThreeScalarD<0, 0b10001, "cmtst", ARM64cmtst>;
2659 defm FABD : SIMDThreeScalarSD<1, 1, 0b11010, "fabd", int_arm64_sisd_fabd>;
2660 def : Pat<(v1f64 (int_arm64_neon_fabd (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2661 (FABD64 FPR64:$Rn, FPR64:$Rm)>;
2662 defm FACGE : SIMDThreeScalarFPCmp<1, 0, 0b11101, "facge",
2663 int_arm64_neon_facge>;
2664 defm FACGT : SIMDThreeScalarFPCmp<1, 1, 0b11101, "facgt",
2665 int_arm64_neon_facgt>;
2666 defm FCMEQ : SIMDThreeScalarFPCmp<0, 0, 0b11100, "fcmeq", ARM64fcmeq>;
2667 defm FCMGE : SIMDThreeScalarFPCmp<1, 0, 0b11100, "fcmge", ARM64fcmge>;
2668 defm FCMGT : SIMDThreeScalarFPCmp<1, 1, 0b11100, "fcmgt", ARM64fcmgt>;
2669 defm FMULX : SIMDThreeScalarSD<0, 0, 0b11011, "fmulx", int_arm64_neon_fmulx>;
2670 defm FRECPS : SIMDThreeScalarSD<0, 0, 0b11111, "frecps", int_arm64_neon_frecps>;
2671 defm FRSQRTS : SIMDThreeScalarSD<0, 1, 0b11111, "frsqrts", int_arm64_neon_frsqrts>;
2672 defm SQADD : SIMDThreeScalarBHSD<0, 0b00001, "sqadd", int_arm64_neon_sqadd>;
2673 defm SQDMULH : SIMDThreeScalarHS< 0, 0b10110, "sqdmulh", int_arm64_neon_sqdmulh>;
2674 defm SQRDMULH : SIMDThreeScalarHS< 1, 0b10110, "sqrdmulh", int_arm64_neon_sqrdmulh>;
2675 defm SQRSHL : SIMDThreeScalarBHSD<0, 0b01011, "sqrshl",int_arm64_neon_sqrshl>;
2676 defm SQSHL : SIMDThreeScalarBHSD<0, 0b01001, "sqshl", int_arm64_neon_sqshl>;
2677 defm SQSUB : SIMDThreeScalarBHSD<0, 0b00101, "sqsub", int_arm64_neon_sqsub>;
2678 defm SRSHL : SIMDThreeScalarD< 0, 0b01010, "srshl", int_arm64_neon_srshl>;
2679 defm SSHL : SIMDThreeScalarD< 0, 0b01000, "sshl", int_arm64_neon_sshl>;
2680 defm SUB : SIMDThreeScalarD< 1, 0b10000, "sub", sub>;
2681 defm UQADD : SIMDThreeScalarBHSD<1, 0b00001, "uqadd", int_arm64_neon_uqadd>;
2682 defm UQRSHL : SIMDThreeScalarBHSD<1, 0b01011, "uqrshl",int_arm64_neon_uqrshl>;
2683 defm UQSHL : SIMDThreeScalarBHSD<1, 0b01001, "uqshl", int_arm64_neon_uqshl>;
2684 defm UQSUB : SIMDThreeScalarBHSD<1, 0b00101, "uqsub", int_arm64_neon_uqsub>;
2685 defm URSHL : SIMDThreeScalarD< 1, 0b01010, "urshl", int_arm64_neon_urshl>;
2686 defm USHL : SIMDThreeScalarD< 1, 0b01000, "ushl", int_arm64_neon_ushl>;
2688 def : InstAlias<"cmls $dst, $src1, $src2",
2689 (CMHSv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2690 def : InstAlias<"cmle $dst, $src1, $src2",
2691 (CMGEv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2692 def : InstAlias<"cmlo $dst, $src1, $src2",
2693 (CMHIv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2694 def : InstAlias<"cmlt $dst, $src1, $src2",
2695 (CMGTv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2696 def : InstAlias<"fcmle $dst, $src1, $src2",
2697 (FCMGE32 FPR32:$dst, FPR32:$src2, FPR32:$src1)>;
2698 def : InstAlias<"fcmle $dst, $src1, $src2",
2699 (FCMGE64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2700 def : InstAlias<"fcmlt $dst, $src1, $src2",
2701 (FCMGT32 FPR32:$dst, FPR32:$src2, FPR32:$src1)>;
2702 def : InstAlias<"fcmlt $dst, $src1, $src2",
2703 (FCMGT64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2704 def : InstAlias<"facle $dst, $src1, $src2",
2705 (FACGE32 FPR32:$dst, FPR32:$src2, FPR32:$src1)>;
2706 def : InstAlias<"facle $dst, $src1, $src2",
2707 (FACGE64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2708 def : InstAlias<"faclt $dst, $src1, $src2",
2709 (FACGT32 FPR32:$dst, FPR32:$src2, FPR32:$src1)>;
2710 def : InstAlias<"faclt $dst, $src1, $src2",
2711 (FACGT64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2713 //===----------------------------------------------------------------------===//
2714 // Advanced SIMD three scalar instructions (mixed operands).
2715 //===----------------------------------------------------------------------===//
2716 defm SQDMULL : SIMDThreeScalarMixedHS<0, 0b11010, "sqdmull",
2717 int_arm64_neon_sqdmulls_scalar>;
2718 defm SQDMLAL : SIMDThreeScalarMixedTiedHS<0, 0b10010, "sqdmlal">;
2719 defm SQDMLSL : SIMDThreeScalarMixedTiedHS<0, 0b10110, "sqdmlsl">;
2721 def : Pat<(i64 (int_arm64_neon_sqadd (i64 FPR64:$Rd),
2722 (i64 (int_arm64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
2723 (i32 FPR32:$Rm))))),
2724 (SQDMLALi32 FPR64:$Rd, FPR32:$Rn, FPR32:$Rm)>;
2725 def : Pat<(i64 (int_arm64_neon_sqsub (i64 FPR64:$Rd),
2726 (i64 (int_arm64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
2727 (i32 FPR32:$Rm))))),
2728 (SQDMLSLi32 FPR64:$Rd, FPR32:$Rn, FPR32:$Rm)>;
2730 //===----------------------------------------------------------------------===//
2731 // Advanced SIMD two scalar instructions.
2732 //===----------------------------------------------------------------------===//
2734 defm ABS : SIMDTwoScalarD< 0, 0b01011, "abs", int_arm64_neon_abs>;
2735 defm CMEQ : SIMDCmpTwoScalarD< 0, 0b01001, "cmeq", ARM64cmeqz>;
2736 defm CMGE : SIMDCmpTwoScalarD< 1, 0b01000, "cmge", ARM64cmgez>;
2737 defm CMGT : SIMDCmpTwoScalarD< 0, 0b01000, "cmgt", ARM64cmgtz>;
2738 defm CMLE : SIMDCmpTwoScalarD< 1, 0b01001, "cmle", ARM64cmlez>;
2739 defm CMLT : SIMDCmpTwoScalarD< 0, 0b01010, "cmlt", ARM64cmltz>;
2740 defm FCMEQ : SIMDCmpTwoScalarSD<0, 1, 0b01101, "fcmeq", ARM64fcmeqz>;
2741 defm FCMGE : SIMDCmpTwoScalarSD<1, 1, 0b01100, "fcmge", ARM64fcmgez>;
2742 defm FCMGT : SIMDCmpTwoScalarSD<0, 1, 0b01100, "fcmgt", ARM64fcmgtz>;
2743 defm FCMLE : SIMDCmpTwoScalarSD<1, 1, 0b01101, "fcmle", ARM64fcmlez>;
2744 defm FCMLT : SIMDCmpTwoScalarSD<0, 1, 0b01110, "fcmlt", ARM64fcmltz>;
2745 defm FCVTAS : SIMDTwoScalarSD< 0, 0, 0b11100, "fcvtas">;
2746 defm FCVTAU : SIMDTwoScalarSD< 1, 0, 0b11100, "fcvtau">;
2747 defm FCVTMS : SIMDTwoScalarSD< 0, 0, 0b11011, "fcvtms">;
2748 defm FCVTMU : SIMDTwoScalarSD< 1, 0, 0b11011, "fcvtmu">;
2749 defm FCVTNS : SIMDTwoScalarSD< 0, 0, 0b11010, "fcvtns">;
2750 defm FCVTNU : SIMDTwoScalarSD< 1, 0, 0b11010, "fcvtnu">;
2751 defm FCVTPS : SIMDTwoScalarSD< 0, 1, 0b11010, "fcvtps">;
2752 defm FCVTPU : SIMDTwoScalarSD< 1, 1, 0b11010, "fcvtpu">;
2753 def FCVTXNv1i64 : SIMDInexactCvtTwoScalar<0b10110, "fcvtxn">;
2754 defm FCVTZS : SIMDTwoScalarSD< 0, 1, 0b11011, "fcvtzs">;
2755 defm FCVTZU : SIMDTwoScalarSD< 1, 1, 0b11011, "fcvtzu">;
2756 defm FRECPE : SIMDTwoScalarSD< 0, 1, 0b11101, "frecpe">;
2757 defm FRECPX : SIMDTwoScalarSD< 0, 1, 0b11111, "frecpx">;
2758 defm FRSQRTE : SIMDTwoScalarSD< 1, 1, 0b11101, "frsqrte">;
2759 defm NEG : SIMDTwoScalarD< 1, 0b01011, "neg",
2760 UnOpFrag<(sub immAllZerosV, node:$LHS)> >;
2761 defm SCVTF : SIMDTwoScalarCVTSD< 0, 0, 0b11101, "scvtf", ARM64sitof>;
2762 defm SQABS : SIMDTwoScalarBHSD< 0, 0b00111, "sqabs", int_arm64_neon_sqabs>;
2763 defm SQNEG : SIMDTwoScalarBHSD< 1, 0b00111, "sqneg", int_arm64_neon_sqneg>;
2764 defm SQXTN : SIMDTwoScalarMixedBHS< 0, 0b10100, "sqxtn", int_arm64_neon_scalar_sqxtn>;
2765 defm SQXTUN : SIMDTwoScalarMixedBHS< 1, 0b10010, "sqxtun", int_arm64_neon_scalar_sqxtun>;
2766 defm SUQADD : SIMDTwoScalarBHSDTied< 0, 0b00011, "suqadd",
2767 int_arm64_neon_suqadd>;
2768 defm UCVTF : SIMDTwoScalarCVTSD< 1, 0, 0b11101, "ucvtf", ARM64uitof>;
2769 defm UQXTN : SIMDTwoScalarMixedBHS<1, 0b10100, "uqxtn", int_arm64_neon_scalar_uqxtn>;
2770 defm USQADD : SIMDTwoScalarBHSDTied< 1, 0b00011, "usqadd",
2771 int_arm64_neon_usqadd>;
2773 def : Pat<(ARM64neg (v1i64 V64:$Rn)), (NEGv1i64 V64:$Rn)>;
2775 def : Pat<(v1i64 (int_arm64_neon_fcvtas (v1f64 FPR64:$Rn))),
2776 (FCVTASv1i64 FPR64:$Rn)>;
2777 def : Pat<(v1i64 (int_arm64_neon_fcvtau (v1f64 FPR64:$Rn))),
2778 (FCVTAUv1i64 FPR64:$Rn)>;
2779 def : Pat<(v1i64 (int_arm64_neon_fcvtms (v1f64 FPR64:$Rn))),
2780 (FCVTMSv1i64 FPR64:$Rn)>;
2781 def : Pat<(v1i64 (int_arm64_neon_fcvtmu (v1f64 FPR64:$Rn))),
2782 (FCVTMUv1i64 FPR64:$Rn)>;
2783 def : Pat<(v1i64 (int_arm64_neon_fcvtns (v1f64 FPR64:$Rn))),
2784 (FCVTNSv1i64 FPR64:$Rn)>;
2785 def : Pat<(v1i64 (int_arm64_neon_fcvtnu (v1f64 FPR64:$Rn))),
2786 (FCVTNUv1i64 FPR64:$Rn)>;
2787 def : Pat<(v1i64 (int_arm64_neon_fcvtps (v1f64 FPR64:$Rn))),
2788 (FCVTPSv1i64 FPR64:$Rn)>;
2789 def : Pat<(v1i64 (int_arm64_neon_fcvtpu (v1f64 FPR64:$Rn))),
2790 (FCVTPUv1i64 FPR64:$Rn)>;
2792 def : Pat<(f32 (int_arm64_neon_frecpe (f32 FPR32:$Rn))),
2793 (FRECPEv1i32 FPR32:$Rn)>;
2794 def : Pat<(f64 (int_arm64_neon_frecpe (f64 FPR64:$Rn))),
2795 (FRECPEv1i64 FPR64:$Rn)>;
2796 def : Pat<(v1f64 (int_arm64_neon_frecpe (v1f64 FPR64:$Rn))),
2797 (FRECPEv1i64 FPR64:$Rn)>;
2799 def : Pat<(f32 (int_arm64_neon_frecpx (f32 FPR32:$Rn))),
2800 (FRECPXv1i32 FPR32:$Rn)>;
2801 def : Pat<(f64 (int_arm64_neon_frecpx (f64 FPR64:$Rn))),
2802 (FRECPXv1i64 FPR64:$Rn)>;
2804 def : Pat<(f32 (int_arm64_neon_frsqrte (f32 FPR32:$Rn))),
2805 (FRSQRTEv1i32 FPR32:$Rn)>;
2806 def : Pat<(f64 (int_arm64_neon_frsqrte (f64 FPR64:$Rn))),
2807 (FRSQRTEv1i64 FPR64:$Rn)>;
2808 def : Pat<(v1f64 (int_arm64_neon_frsqrte (v1f64 FPR64:$Rn))),
2809 (FRSQRTEv1i64 FPR64:$Rn)>;
2811 // If an integer is about to be converted to a floating point value,
2812 // just load it on the floating point unit.
2813 // Here are the patterns for 8 and 16-bits to float.
2815 def : Pat <(f32 (uint_to_fp (i32 (zextloadi8 ro_indexed8:$addr)))),
2816 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2817 (LDRBro ro_indexed8:$addr), bsub))>;
2818 def : Pat <(f32 (uint_to_fp (i32 (zextloadi8 am_indexed8:$addr)))),
2819 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2820 (LDRBui am_indexed8:$addr), bsub))>;
2821 def : Pat <(f32 (uint_to_fp (i32 (zextloadi8 am_unscaled8:$addr)))),
2822 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2823 (LDURBi am_unscaled8:$addr), bsub))>;
2824 // 16-bits -> float.
2825 def : Pat <(f32 (uint_to_fp (i32 (zextloadi16 ro_indexed16:$addr)))),
2826 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2827 (LDRHro ro_indexed16:$addr), hsub))>;
2828 def : Pat <(f32 (uint_to_fp (i32 (zextloadi16 am_indexed16:$addr)))),
2829 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2830 (LDRHui am_indexed16:$addr), hsub))>;
2831 def : Pat <(f32 (uint_to_fp (i32 (zextloadi16 am_unscaled16:$addr)))),
2832 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2833 (LDURHi am_unscaled16:$addr), hsub))>;
2834 // 32-bits are handled in target specific dag combine:
2835 // performIntToFpCombine.
2836 // 64-bits integer to 32-bits floating point, not possible with
2837 // UCVTF on floating point registers (both source and destination
2838 // must have the same size).
2840 // Here are the patterns for 8, 16, 32, and 64-bits to double.
2841 // 8-bits -> double.
2842 def : Pat <(f64 (uint_to_fp (i32 (zextloadi8 ro_indexed8:$addr)))),
2843 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2844 (LDRBro ro_indexed8:$addr), bsub))>;
2845 def : Pat <(f64 (uint_to_fp (i32 (zextloadi8 am_indexed8:$addr)))),
2846 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2847 (LDRBui am_indexed8:$addr), bsub))>;
2848 def : Pat <(f64 (uint_to_fp (i32 (zextloadi8 am_unscaled8:$addr)))),
2849 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2850 (LDURBi am_unscaled8:$addr), bsub))>;
2851 // 16-bits -> double.
2852 def : Pat <(f64 (uint_to_fp (i32 (zextloadi16 ro_indexed16:$addr)))),
2853 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2854 (LDRHro ro_indexed16:$addr), hsub))>;
2855 def : Pat <(f64 (uint_to_fp (i32 (zextloadi16 am_indexed16:$addr)))),
2856 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2857 (LDRHui am_indexed16:$addr), hsub))>;
2858 def : Pat <(f64 (uint_to_fp (i32 (zextloadi16 am_unscaled16:$addr)))),
2859 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2860 (LDURHi am_unscaled16:$addr), hsub))>;
2861 // 32-bits -> double.
2862 def : Pat <(f64 (uint_to_fp (i32 (load ro_indexed32:$addr)))),
2863 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2864 (LDRSro ro_indexed32:$addr), ssub))>;
2865 def : Pat <(f64 (uint_to_fp (i32 (load am_indexed32:$addr)))),
2866 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2867 (LDRSui am_indexed32:$addr), ssub))>;
2868 def : Pat <(f64 (uint_to_fp (i32 (load am_unscaled32:$addr)))),
2869 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2870 (LDURSi am_unscaled32:$addr), ssub))>;
2871 // 64-bits -> double are handled in target specific dag combine:
2872 // performIntToFpCombine.
2874 //===----------------------------------------------------------------------===//
2875 // Advanced SIMD three different-sized vector instructions.
2876 //===----------------------------------------------------------------------===//
2878 defm ADDHN : SIMDNarrowThreeVectorBHS<0,0b0100,"addhn", int_arm64_neon_addhn>;
2879 defm SUBHN : SIMDNarrowThreeVectorBHS<0,0b0110,"subhn", int_arm64_neon_subhn>;
2880 defm RADDHN : SIMDNarrowThreeVectorBHS<1,0b0100,"raddhn",int_arm64_neon_raddhn>;
2881 defm RSUBHN : SIMDNarrowThreeVectorBHS<1,0b0110,"rsubhn",int_arm64_neon_rsubhn>;
2882 defm PMULL : SIMDDifferentThreeVectorBD<0,0b1110,"pmull",int_arm64_neon_pmull>;
2883 defm SABAL : SIMDLongThreeVectorTiedBHSabal<0,0b0101,"sabal",
2884 int_arm64_neon_sabd>;
2885 defm SABDL : SIMDLongThreeVectorBHSabdl<0, 0b0111, "sabdl",
2886 int_arm64_neon_sabd>;
2887 defm SADDL : SIMDLongThreeVectorBHS< 0, 0b0000, "saddl",
2888 BinOpFrag<(add (sext node:$LHS), (sext node:$RHS))>>;
2889 defm SADDW : SIMDWideThreeVectorBHS< 0, 0b0001, "saddw",
2890 BinOpFrag<(add node:$LHS, (sext node:$RHS))>>;
2891 defm SMLAL : SIMDLongThreeVectorTiedBHS<0, 0b1000, "smlal",
2892 TriOpFrag<(add node:$LHS, (int_arm64_neon_smull node:$MHS, node:$RHS))>>;
2893 defm SMLSL : SIMDLongThreeVectorTiedBHS<0, 0b1010, "smlsl",
2894 TriOpFrag<(sub node:$LHS, (int_arm64_neon_smull node:$MHS, node:$RHS))>>;
2895 defm SMULL : SIMDLongThreeVectorBHS<0, 0b1100, "smull", int_arm64_neon_smull>;
2896 defm SQDMLAL : SIMDLongThreeVectorSQDMLXTiedHS<0, 0b1001, "sqdmlal",
2897 int_arm64_neon_sqadd>;
2898 defm SQDMLSL : SIMDLongThreeVectorSQDMLXTiedHS<0, 0b1011, "sqdmlsl",
2899 int_arm64_neon_sqsub>;
2900 defm SQDMULL : SIMDLongThreeVectorHS<0, 0b1101, "sqdmull",
2901 int_arm64_neon_sqdmull>;
2902 defm SSUBL : SIMDLongThreeVectorBHS<0, 0b0010, "ssubl",
2903 BinOpFrag<(sub (sext node:$LHS), (sext node:$RHS))>>;
2904 defm SSUBW : SIMDWideThreeVectorBHS<0, 0b0011, "ssubw",
2905 BinOpFrag<(sub node:$LHS, (sext node:$RHS))>>;
2906 defm UABAL : SIMDLongThreeVectorTiedBHSabal<1, 0b0101, "uabal",
2907 int_arm64_neon_uabd>;
2908 defm UABDL : SIMDLongThreeVectorBHSabdl<1, 0b0111, "uabdl",
2909 int_arm64_neon_uabd>;
2910 defm UADDL : SIMDLongThreeVectorBHS<1, 0b0000, "uaddl",
2911 BinOpFrag<(add (zext node:$LHS), (zext node:$RHS))>>;
2912 defm UADDW : SIMDWideThreeVectorBHS<1, 0b0001, "uaddw",
2913 BinOpFrag<(add node:$LHS, (zext node:$RHS))>>;
2914 defm UMLAL : SIMDLongThreeVectorTiedBHS<1, 0b1000, "umlal",
2915 TriOpFrag<(add node:$LHS, (int_arm64_neon_umull node:$MHS, node:$RHS))>>;
2916 defm UMLSL : SIMDLongThreeVectorTiedBHS<1, 0b1010, "umlsl",
2917 TriOpFrag<(sub node:$LHS, (int_arm64_neon_umull node:$MHS, node:$RHS))>>;
2918 defm UMULL : SIMDLongThreeVectorBHS<1, 0b1100, "umull", int_arm64_neon_umull>;
2919 defm USUBL : SIMDLongThreeVectorBHS<1, 0b0010, "usubl",
2920 BinOpFrag<(sub (zext node:$LHS), (zext node:$RHS))>>;
2921 defm USUBW : SIMDWideThreeVectorBHS< 1, 0b0011, "usubw",
2922 BinOpFrag<(sub node:$LHS, (zext node:$RHS))>>;
2924 // Patterns for 64-bit pmull
2925 def : Pat<(int_arm64_neon_pmull64 V64:$Rn, V64:$Rm),
2926 (PMULLv1i64 V64:$Rn, V64:$Rm)>;
2927 def : Pat<(int_arm64_neon_pmull64 (vector_extract (v2i64 V128:$Rn), (i64 1)),
2928 (vector_extract (v2i64 V128:$Rm), (i64 1))),
2929 (PMULLv2i64 V128:$Rn, V128:$Rm)>;
2931 // CodeGen patterns for addhn and subhn instructions, which can actually be
2932 // written in LLVM IR without too much difficulty.
2935 def : Pat<(v8i8 (trunc (v8i16 (ARM64vlshr (add V128:$Rn, V128:$Rm), (i32 8))))),
2936 (ADDHNv8i16_v8i8 V128:$Rn, V128:$Rm)>;
2937 def : Pat<(v4i16 (trunc (v4i32 (ARM64vlshr (add V128:$Rn, V128:$Rm),
2939 (ADDHNv4i32_v4i16 V128:$Rn, V128:$Rm)>;
2940 def : Pat<(v2i32 (trunc (v2i64 (ARM64vlshr (add V128:$Rn, V128:$Rm),
2942 (ADDHNv2i64_v2i32 V128:$Rn, V128:$Rm)>;
2943 def : Pat<(concat_vectors (v8i8 V64:$Rd),
2944 (trunc (v8i16 (ARM64vlshr (add V128:$Rn, V128:$Rm),
2946 (ADDHNv8i16_v16i8 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
2947 V128:$Rn, V128:$Rm)>;
2948 def : Pat<(concat_vectors (v4i16 V64:$Rd),
2949 (trunc (v4i32 (ARM64vlshr (add V128:$Rn, V128:$Rm),
2951 (ADDHNv4i32_v8i16 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
2952 V128:$Rn, V128:$Rm)>;
2953 def : Pat<(concat_vectors (v2i32 V64:$Rd),
2954 (trunc (v2i64 (ARM64vlshr (add V128:$Rn, V128:$Rm),
2956 (ADDHNv2i64_v4i32 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
2957 V128:$Rn, V128:$Rm)>;
2960 def : Pat<(v8i8 (trunc (v8i16 (ARM64vlshr (sub V128:$Rn, V128:$Rm), (i32 8))))),
2961 (SUBHNv8i16_v8i8 V128:$Rn, V128:$Rm)>;
2962 def : Pat<(v4i16 (trunc (v4i32 (ARM64vlshr (sub V128:$Rn, V128:$Rm),
2964 (SUBHNv4i32_v4i16 V128:$Rn, V128:$Rm)>;
2965 def : Pat<(v2i32 (trunc (v2i64 (ARM64vlshr (sub V128:$Rn, V128:$Rm),
2967 (SUBHNv2i64_v2i32 V128:$Rn, V128:$Rm)>;
2968 def : Pat<(concat_vectors (v8i8 V64:$Rd),
2969 (trunc (v8i16 (ARM64vlshr (sub V128:$Rn, V128:$Rm),
2971 (SUBHNv8i16_v16i8 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
2972 V128:$Rn, V128:$Rm)>;
2973 def : Pat<(concat_vectors (v4i16 V64:$Rd),
2974 (trunc (v4i32 (ARM64vlshr (sub V128:$Rn, V128:$Rm),
2976 (SUBHNv4i32_v8i16 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
2977 V128:$Rn, V128:$Rm)>;
2978 def : Pat<(concat_vectors (v2i32 V64:$Rd),
2979 (trunc (v2i64 (ARM64vlshr (sub V128:$Rn, V128:$Rm),
2981 (SUBHNv2i64_v4i32 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
2982 V128:$Rn, V128:$Rm)>;
2984 //----------------------------------------------------------------------------
2985 // AdvSIMD bitwise extract from vector instruction.
2986 //----------------------------------------------------------------------------
2988 defm EXT : SIMDBitwiseExtract<"ext">;
2990 def : Pat<(v4i16 (ARM64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
2991 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
2992 def : Pat<(v8i16 (ARM64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
2993 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
2994 def : Pat<(v2i32 (ARM64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
2995 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
2996 def : Pat<(v2f32 (ARM64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
2997 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
2998 def : Pat<(v4i32 (ARM64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
2999 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3000 def : Pat<(v4f32 (ARM64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3001 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3002 def : Pat<(v2i64 (ARM64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3003 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3004 def : Pat<(v2f64 (ARM64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
3005 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
3007 // We use EXT to handle extract_subvector to copy the upper 64-bits of a
3009 def : Pat<(v8i8 (extract_subvector V128:$Rn, (i64 8))),
3010 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3011 def : Pat<(v4i16 (extract_subvector V128:$Rn, (i64 4))),
3012 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3013 def : Pat<(v2i32 (extract_subvector V128:$Rn, (i64 2))),
3014 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3015 def : Pat<(v1i64 (extract_subvector V128:$Rn, (i64 1))),
3016 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3017 def : Pat<(v2f32 (extract_subvector V128:$Rn, (i64 2))),
3018 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3019 def : Pat<(v1f64 (extract_subvector V128:$Rn, (i64 1))),
3020 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
3023 //----------------------------------------------------------------------------
3024 // AdvSIMD zip vector
3025 //----------------------------------------------------------------------------
3027 defm TRN1 : SIMDZipVector<0b010, "trn1", ARM64trn1>;
3028 defm TRN2 : SIMDZipVector<0b110, "trn2", ARM64trn2>;
3029 defm UZP1 : SIMDZipVector<0b001, "uzp1", ARM64uzp1>;
3030 defm UZP2 : SIMDZipVector<0b101, "uzp2", ARM64uzp2>;
3031 defm ZIP1 : SIMDZipVector<0b011, "zip1", ARM64zip1>;
3032 defm ZIP2 : SIMDZipVector<0b111, "zip2", ARM64zip2>;
3034 //----------------------------------------------------------------------------
3035 // AdvSIMD TBL/TBX instructions
3036 //----------------------------------------------------------------------------
3038 defm TBL : SIMDTableLookup< 0, "tbl">;
3039 defm TBX : SIMDTableLookupTied<1, "tbx">;
3041 def : Pat<(v8i8 (int_arm64_neon_tbl1 (v16i8 VecListOne128:$Rn), (v8i8 V64:$Ri))),
3042 (TBLv8i8One VecListOne128:$Rn, V64:$Ri)>;
3043 def : Pat<(v16i8 (int_arm64_neon_tbl1 (v16i8 V128:$Ri), (v16i8 V128:$Rn))),
3044 (TBLv16i8One V128:$Ri, V128:$Rn)>;
3046 def : Pat<(v8i8 (int_arm64_neon_tbx1 (v8i8 V64:$Rd),
3047 (v16i8 VecListOne128:$Rn), (v8i8 V64:$Ri))),
3048 (TBXv8i8One V64:$Rd, VecListOne128:$Rn, V64:$Ri)>;
3049 def : Pat<(v16i8 (int_arm64_neon_tbx1 (v16i8 V128:$Rd),
3050 (v16i8 V128:$Ri), (v16i8 V128:$Rn))),
3051 (TBXv16i8One V128:$Rd, V128:$Ri, V128:$Rn)>;
3054 //----------------------------------------------------------------------------
3055 // AdvSIMD scalar CPY instruction
3056 //----------------------------------------------------------------------------
3058 defm CPY : SIMDScalarCPY<"cpy">;
3060 //----------------------------------------------------------------------------
3061 // AdvSIMD scalar pairwise instructions
3062 //----------------------------------------------------------------------------
3064 defm ADDP : SIMDPairwiseScalarD<0, 0b11011, "addp">;
3065 defm FADDP : SIMDPairwiseScalarSD<1, 0, 0b01101, "faddp">;
3066 defm FMAXNMP : SIMDPairwiseScalarSD<1, 0, 0b01100, "fmaxnmp">;
3067 defm FMAXP : SIMDPairwiseScalarSD<1, 0, 0b01111, "fmaxp">;
3068 defm FMINNMP : SIMDPairwiseScalarSD<1, 1, 0b01100, "fminnmp">;
3069 defm FMINP : SIMDPairwiseScalarSD<1, 1, 0b01111, "fminp">;
3070 def : Pat<(i64 (int_arm64_neon_saddv (v2i64 V128:$Rn))),
3071 (ADDPv2i64p V128:$Rn)>;
3072 def : Pat<(i64 (int_arm64_neon_uaddv (v2i64 V128:$Rn))),
3073 (ADDPv2i64p V128:$Rn)>;
3074 def : Pat<(f32 (int_arm64_neon_faddv (v2f32 V64:$Rn))),
3075 (FADDPv2i32p V64:$Rn)>;
3076 def : Pat<(f32 (int_arm64_neon_faddv (v4f32 V128:$Rn))),
3077 (FADDPv2i32p (EXTRACT_SUBREG (FADDPv4f32 V128:$Rn, V128:$Rn), dsub))>;
3078 def : Pat<(f64 (int_arm64_neon_faddv (v2f64 V128:$Rn))),
3079 (FADDPv2i64p V128:$Rn)>;
3080 def : Pat<(f32 (int_arm64_neon_fmaxnmv (v2f32 V64:$Rn))),
3081 (FMAXNMPv2i32p V64:$Rn)>;
3082 def : Pat<(f64 (int_arm64_neon_fmaxnmv (v2f64 V128:$Rn))),
3083 (FMAXNMPv2i64p V128:$Rn)>;
3084 def : Pat<(f32 (int_arm64_neon_fmaxv (v2f32 V64:$Rn))),
3085 (FMAXPv2i32p V64:$Rn)>;
3086 def : Pat<(f64 (int_arm64_neon_fmaxv (v2f64 V128:$Rn))),
3087 (FMAXPv2i64p V128:$Rn)>;
3088 def : Pat<(f32 (int_arm64_neon_fminnmv (v2f32 V64:$Rn))),
3089 (FMINNMPv2i32p V64:$Rn)>;
3090 def : Pat<(f64 (int_arm64_neon_fminnmv (v2f64 V128:$Rn))),
3091 (FMINNMPv2i64p V128:$Rn)>;
3092 def : Pat<(f32 (int_arm64_neon_fminv (v2f32 V64:$Rn))),
3093 (FMINPv2i32p V64:$Rn)>;
3094 def : Pat<(f64 (int_arm64_neon_fminv (v2f64 V128:$Rn))),
3095 (FMINPv2i64p V128:$Rn)>;
3097 //----------------------------------------------------------------------------
3098 // AdvSIMD INS/DUP instructions
3099 //----------------------------------------------------------------------------
3101 def DUPv8i8gpr : SIMDDupFromMain<0, 0b00001, ".8b", v8i8, V64, GPR32>;
3102 def DUPv16i8gpr : SIMDDupFromMain<1, 0b00001, ".16b", v16i8, V128, GPR32>;
3103 def DUPv4i16gpr : SIMDDupFromMain<0, 0b00010, ".4h", v4i16, V64, GPR32>;
3104 def DUPv8i16gpr : SIMDDupFromMain<1, 0b00010, ".8h", v8i16, V128, GPR32>;
3105 def DUPv2i32gpr : SIMDDupFromMain<0, 0b00100, ".2s", v2i32, V64, GPR32>;
3106 def DUPv4i32gpr : SIMDDupFromMain<1, 0b00100, ".4s", v4i32, V128, GPR32>;
3107 def DUPv2i64gpr : SIMDDupFromMain<1, 0b01000, ".2d", v2i64, V128, GPR64>;
3109 def DUPv2i64lane : SIMDDup64FromElement;
3110 def DUPv2i32lane : SIMDDup32FromElement<0, ".2s", v2i32, V64>;
3111 def DUPv4i32lane : SIMDDup32FromElement<1, ".4s", v4i32, V128>;
3112 def DUPv4i16lane : SIMDDup16FromElement<0, ".4h", v4i16, V64>;
3113 def DUPv8i16lane : SIMDDup16FromElement<1, ".8h", v8i16, V128>;
3114 def DUPv8i8lane : SIMDDup8FromElement <0, ".8b", v8i8, V64>;
3115 def DUPv16i8lane : SIMDDup8FromElement <1, ".16b", v16i8, V128>;
3117 def : Pat<(v2f32 (ARM64dup (f32 FPR32:$Rn))),
3118 (v2f32 (DUPv2i32lane
3119 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rn, ssub),
3121 def : Pat<(v4f32 (ARM64dup (f32 FPR32:$Rn))),
3122 (v4f32 (DUPv4i32lane
3123 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rn, ssub),
3125 def : Pat<(v2f64 (ARM64dup (f64 FPR64:$Rn))),
3126 (v2f64 (DUPv2i64lane
3127 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$Rn, dsub),
3130 def : Pat<(v2f32 (ARM64duplane32 (v4f32 V128:$Rn), VectorIndexS:$imm)),
3131 (DUPv2i32lane V128:$Rn, VectorIndexS:$imm)>;
3132 def : Pat<(v4f32 (ARM64duplane32 (v4f32 V128:$Rn), VectorIndexS:$imm)),
3133 (DUPv4i32lane V128:$Rn, VectorIndexS:$imm)>;
3134 def : Pat<(v2f64 (ARM64duplane64 (v2f64 V128:$Rn), VectorIndexD:$imm)),
3135 (DUPv2i64lane V128:$Rn, VectorIndexD:$imm)>;
3137 // If there's an (ARM64dup (vector_extract ...) ...), we can use a duplane
3138 // instruction even if the types don't match: we just have to remap the lane
3139 // carefully. N.b. this trick only applies to truncations.
3140 def VecIndex_x2 : SDNodeXForm<imm, [{
3141 return CurDAG->getTargetConstant(2 * N->getZExtValue(), MVT::i64);
3143 def VecIndex_x4 : SDNodeXForm<imm, [{
3144 return CurDAG->getTargetConstant(4 * N->getZExtValue(), MVT::i64);
3146 def VecIndex_x8 : SDNodeXForm<imm, [{
3147 return CurDAG->getTargetConstant(8 * N->getZExtValue(), MVT::i64);
3150 multiclass DUPWithTruncPats<ValueType ResVT, ValueType Src64VT,
3151 ValueType Src128VT, ValueType ScalVT,
3152 Instruction DUP, SDNodeXForm IdxXFORM> {
3153 def : Pat<(ResVT (ARM64dup (ScalVT (vector_extract (Src128VT V128:$Rn),
3155 (DUP V128:$Rn, (IdxXFORM imm:$idx))>;
3157 def : Pat<(ResVT (ARM64dup (ScalVT (vector_extract (Src64VT V64:$Rn),
3159 (DUP (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), (IdxXFORM imm:$idx))>;
3162 defm : DUPWithTruncPats<v8i8, v4i16, v8i16, i32, DUPv8i8lane, VecIndex_x2>;
3163 defm : DUPWithTruncPats<v8i8, v2i32, v4i32, i32, DUPv8i8lane, VecIndex_x4>;
3164 defm : DUPWithTruncPats<v4i16, v2i32, v4i32, i32, DUPv4i16lane, VecIndex_x2>;
3166 defm : DUPWithTruncPats<v16i8, v4i16, v8i16, i32, DUPv16i8lane, VecIndex_x2>;
3167 defm : DUPWithTruncPats<v16i8, v2i32, v4i32, i32, DUPv16i8lane, VecIndex_x4>;
3168 defm : DUPWithTruncPats<v8i16, v2i32, v4i32, i32, DUPv8i16lane, VecIndex_x2>;
3170 multiclass DUPWithTrunci64Pats<ValueType ResVT, Instruction DUP,
3171 SDNodeXForm IdxXFORM> {
3172 def : Pat<(ResVT (ARM64dup (i32 (trunc (vector_extract (v2i64 V128:$Rn),
3174 (DUP V128:$Rn, (IdxXFORM imm:$idx))>;
3176 def : Pat<(ResVT (ARM64dup (i32 (trunc (vector_extract (v1i64 V64:$Rn),
3178 (DUP (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), (IdxXFORM imm:$idx))>;
3181 defm : DUPWithTrunci64Pats<v8i8, DUPv8i8lane, VecIndex_x8>;
3182 defm : DUPWithTrunci64Pats<v4i16, DUPv4i16lane, VecIndex_x4>;
3183 defm : DUPWithTrunci64Pats<v2i32, DUPv2i32lane, VecIndex_x2>;
3185 defm : DUPWithTrunci64Pats<v16i8, DUPv16i8lane, VecIndex_x8>;
3186 defm : DUPWithTrunci64Pats<v8i16, DUPv8i16lane, VecIndex_x4>;
3187 defm : DUPWithTrunci64Pats<v4i32, DUPv4i32lane, VecIndex_x2>;
3189 // SMOV and UMOV definitions, with some extra patterns for convenience
3193 def : Pat<(sext_inreg (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx), i8),
3194 (i32 (SMOVvi8to32 V128:$Rn, VectorIndexB:$idx))>;
3195 def : Pat<(sext_inreg (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx), i8),
3196 (i64 (SMOVvi8to64 V128:$Rn, VectorIndexB:$idx))>;
3197 def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
3198 (i32 (SMOVvi16to32 V128:$Rn, VectorIndexH:$idx))>;
3199 def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
3200 (i64 (SMOVvi16to64 V128:$Rn, VectorIndexH:$idx))>;
3201 def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
3202 (i32 (SMOVvi16to32 V128:$Rn, VectorIndexH:$idx))>;
3203 def : Pat<(sext (i32 (vector_extract (v4i32 V128:$Rn), VectorIndexS:$idx))),
3204 (i64 (SMOVvi32to64 V128:$Rn, VectorIndexS:$idx))>;
3206 // Extracting i8 or i16 elements will have the zero-extend transformed to
3207 // an 'and' mask by type legalization since neither i8 nor i16 are legal types
3208 // for ARM64. Match these patterns here since UMOV already zeroes out the high
3209 // bits of the destination register.
3210 def : Pat<(and (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx),
3212 (i32 (UMOVvi8 V128:$Rn, VectorIndexB:$idx))>;
3213 def : Pat<(and (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),
3215 (i32 (UMOVvi16 V128:$Rn, VectorIndexH:$idx))>;
3219 def : Pat<(v16i8 (scalar_to_vector GPR32:$Rn)),
3220 (SUBREG_TO_REG (i32 0),
3221 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3222 def : Pat<(v8i8 (scalar_to_vector GPR32:$Rn)),
3223 (SUBREG_TO_REG (i32 0),
3224 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3226 def : Pat<(v8i16 (scalar_to_vector GPR32:$Rn)),
3227 (SUBREG_TO_REG (i32 0),
3228 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3229 def : Pat<(v4i16 (scalar_to_vector GPR32:$Rn)),
3230 (SUBREG_TO_REG (i32 0),
3231 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3233 def : Pat<(v2i32 (scalar_to_vector (i32 FPR32:$Rn))),
3234 (v2i32 (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)),
3235 (i32 FPR32:$Rn), ssub))>;
3236 def : Pat<(v4i32 (scalar_to_vector (i32 FPR32:$Rn))),
3237 (v4i32 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
3238 (i32 FPR32:$Rn), ssub))>;
3239 def : Pat<(v2i64 (scalar_to_vector (i64 FPR64:$Rn))),
3240 (v2i64 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
3241 (i64 FPR64:$Rn), dsub))>;
3243 def : Pat<(v4f32 (scalar_to_vector (f32 FPR32:$Rn))),
3244 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rn, ssub)>;
3245 def : Pat<(v2f32 (scalar_to_vector (f32 FPR32:$Rn))),
3246 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), FPR32:$Rn, ssub)>;
3247 def : Pat<(v2f64 (scalar_to_vector (f64 FPR64:$Rn))),
3248 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$Rn, dsub)>;
3250 def : Pat<(v2f32 (vector_insert (v2f32 V64:$Rn),
3251 (f32 FPR32:$Rm), (i64 VectorIndexS:$imm))),
3254 (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), V64:$Rn, dsub)),
3256 (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rm, ssub)),
3259 def : Pat<(v4f32 (vector_insert (v4f32 V128:$Rn),
3260 (f32 FPR32:$Rm), (i64 VectorIndexS:$imm))),
3262 V128:$Rn, VectorIndexS:$imm,
3263 (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rm, ssub)),
3265 def : Pat<(v2f64 (vector_insert (v2f64 V128:$Rn),
3266 (f64 FPR64:$Rm), (i64 VectorIndexD:$imm))),
3268 V128:$Rn, VectorIndexD:$imm,
3269 (v2f64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$Rm, dsub)),
3272 // Copy an element at a constant index in one vector into a constant indexed
3273 // element of another.
3274 // FIXME refactor to a shared class/dev parameterized on vector type, vector
3275 // index type and INS extension
3276 def : Pat<(v16i8 (int_arm64_neon_vcopy_lane
3277 (v16i8 V128:$Vd), VectorIndexB:$idx, (v16i8 V128:$Vs),
3278 VectorIndexB:$idx2)),
3280 V128:$Vd, VectorIndexB:$idx, V128:$Vs, VectorIndexB:$idx2)
3282 def : Pat<(v8i16 (int_arm64_neon_vcopy_lane
3283 (v8i16 V128:$Vd), VectorIndexH:$idx, (v8i16 V128:$Vs),
3284 VectorIndexH:$idx2)),
3286 V128:$Vd, VectorIndexH:$idx, V128:$Vs, VectorIndexH:$idx2)
3288 def : Pat<(v4i32 (int_arm64_neon_vcopy_lane
3289 (v4i32 V128:$Vd), VectorIndexS:$idx, (v4i32 V128:$Vs),
3290 VectorIndexS:$idx2)),
3292 V128:$Vd, VectorIndexS:$idx, V128:$Vs, VectorIndexS:$idx2)
3294 def : Pat<(v2i64 (int_arm64_neon_vcopy_lane
3295 (v2i64 V128:$Vd), VectorIndexD:$idx, (v2i64 V128:$Vs),
3296 VectorIndexD:$idx2)),
3298 V128:$Vd, VectorIndexD:$idx, V128:$Vs, VectorIndexD:$idx2)
3301 multiclass Neon_INS_elt_pattern<ValueType VT128, ValueType VT64,
3302 ValueType VTScal, Instruction INS> {
3303 def : Pat<(VT128 (vector_insert V128:$src,
3304 (VTScal (vector_extract (VT128 V128:$Rn), imm:$Immn)),
3306 (INS V128:$src, imm:$Immd, V128:$Rn, imm:$Immn)>;
3308 def : Pat<(VT128 (vector_insert V128:$src,
3309 (VTScal (vector_extract (VT64 V64:$Rn), imm:$Immn)),
3311 (INS V128:$src, imm:$Immd,
3312 (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), imm:$Immn)>;
3314 def : Pat<(VT64 (vector_insert V64:$src,
3315 (VTScal (vector_extract (VT128 V128:$Rn), imm:$Immn)),
3317 (EXTRACT_SUBREG (INS (SUBREG_TO_REG (i64 0), V64:$src, dsub),
3318 imm:$Immd, V128:$Rn, imm:$Immn),
3321 def : Pat<(VT64 (vector_insert V64:$src,
3322 (VTScal (vector_extract (VT64 V64:$Rn), imm:$Immn)),
3325 (INS (SUBREG_TO_REG (i64 0), V64:$src, dsub), imm:$Immd,
3326 (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), imm:$Immn),
3330 defm : Neon_INS_elt_pattern<v4f32, v2f32, f32, INSvi32lane>;
3331 defm : Neon_INS_elt_pattern<v2f64, v1f64, f64, INSvi64lane>;
3332 defm : Neon_INS_elt_pattern<v16i8, v8i8, i32, INSvi8lane>;
3333 defm : Neon_INS_elt_pattern<v8i16, v4i16, i32, INSvi16lane>;
3334 defm : Neon_INS_elt_pattern<v4i32, v2i32, i32, INSvi32lane>;
3335 defm : Neon_INS_elt_pattern<v2i64, v1i64, i64, INSvi32lane>;
3338 // Floating point vector extractions are codegen'd as either a sequence of
3339 // subregister extractions, possibly fed by an INS if the lane number is
3340 // anything other than zero.
3341 def : Pat<(vector_extract (v2f64 V128:$Rn), 0),
3342 (f64 (EXTRACT_SUBREG V128:$Rn, dsub))>;
3343 def : Pat<(vector_extract (v4f32 V128:$Rn), 0),
3344 (f32 (EXTRACT_SUBREG V128:$Rn, ssub))>;
3345 def : Pat<(vector_extract (v2f64 V128:$Rn), VectorIndexD:$idx),
3346 (f64 (EXTRACT_SUBREG
3347 (INSvi64lane (v2f64 (IMPLICIT_DEF)), 0,
3348 V128:$Rn, VectorIndexD:$idx),
3350 def : Pat<(vector_extract (v4f32 V128:$Rn), VectorIndexS:$idx),
3351 (f32 (EXTRACT_SUBREG
3352 (INSvi32lane (v4f32 (IMPLICIT_DEF)), 0,
3353 V128:$Rn, VectorIndexS:$idx),
3356 // All concat_vectors operations are canonicalised to act on i64 vectors for
3357 // ARM64. In the general case we need an instruction, which had just as well be
3359 class ConcatPat<ValueType DstTy, ValueType SrcTy>
3360 : Pat<(DstTy (concat_vectors (SrcTy V64:$Rd), V64:$Rn)),
3361 (INSvi64lane (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), 1,
3362 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rn, dsub), 0)>;
3364 def : ConcatPat<v2i64, v1i64>;
3365 def : ConcatPat<v2f64, v1f64>;
3366 def : ConcatPat<v4i32, v2i32>;
3367 def : ConcatPat<v4f32, v2f32>;
3368 def : ConcatPat<v8i16, v4i16>;
3369 def : ConcatPat<v16i8, v8i8>;
3371 // If the high lanes are undef, though, we can just ignore them:
3372 class ConcatUndefPat<ValueType DstTy, ValueType SrcTy>
3373 : Pat<(DstTy (concat_vectors (SrcTy V64:$Rn), undef)),
3374 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rn, dsub)>;
3376 def : ConcatUndefPat<v2i64, v1i64>;
3377 def : ConcatUndefPat<v2f64, v1f64>;
3378 def : ConcatUndefPat<v4i32, v2i32>;
3379 def : ConcatUndefPat<v4f32, v2f32>;
3380 def : ConcatUndefPat<v8i16, v4i16>;
3381 def : ConcatUndefPat<v16i8, v8i8>;
3383 //----------------------------------------------------------------------------
3384 // AdvSIMD across lanes instructions
3385 //----------------------------------------------------------------------------
3387 defm ADDV : SIMDAcrossLanesBHS<0, 0b11011, "addv">;
3388 defm SMAXV : SIMDAcrossLanesBHS<0, 0b01010, "smaxv">;
3389 defm SMINV : SIMDAcrossLanesBHS<0, 0b11010, "sminv">;
3390 defm UMAXV : SIMDAcrossLanesBHS<1, 0b01010, "umaxv">;
3391 defm UMINV : SIMDAcrossLanesBHS<1, 0b11010, "uminv">;
3392 defm SADDLV : SIMDAcrossLanesHSD<0, 0b00011, "saddlv">;
3393 defm UADDLV : SIMDAcrossLanesHSD<1, 0b00011, "uaddlv">;
3394 defm FMAXNMV : SIMDAcrossLanesS<0b01100, 0, "fmaxnmv", int_arm64_neon_fmaxnmv>;
3395 defm FMAXV : SIMDAcrossLanesS<0b01111, 0, "fmaxv", int_arm64_neon_fmaxv>;
3396 defm FMINNMV : SIMDAcrossLanesS<0b01100, 1, "fminnmv", int_arm64_neon_fminnmv>;
3397 defm FMINV : SIMDAcrossLanesS<0b01111, 1, "fminv", int_arm64_neon_fminv>;
3399 multiclass SIMDAcrossLanesSignedIntrinsic<string baseOpc, Intrinsic intOp> {
3400 // If there is a sign extension after this intrinsic, consume it as smov already
3402 def : Pat<(i32 (sext_inreg (i32 (intOp (v8i8 V64:$Rn))), i8)),
3404 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3405 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3407 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
3409 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3410 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3412 // If there is a sign extension after this intrinsic, consume it as smov already
3414 def : Pat<(i32 (sext_inreg (i32 (intOp (v16i8 V128:$Rn))), i8)),
3416 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3417 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3419 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
3421 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3422 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3424 // If there is a sign extension after this intrinsic, consume it as smov already
3426 def : Pat<(i32 (sext_inreg (i32 (intOp (v4i16 V64:$Rn))), i16)),
3428 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3429 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3431 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
3433 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3434 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3436 // If there is a sign extension after this intrinsic, consume it as smov already
3438 def : Pat<(i32 (sext_inreg (i32 (intOp (v8i16 V128:$Rn))), i16)),
3440 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3441 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3443 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
3445 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3446 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3449 def : Pat<(i32 (intOp (v4i32 V128:$Rn))),
3450 (i32 (EXTRACT_SUBREG
3451 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3452 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), ssub),
3456 multiclass SIMDAcrossLanesUnsignedIntrinsic<string baseOpc, Intrinsic intOp> {
3457 // If there is a masking operation keeping only what has been actually
3458 // generated, consume it.
3459 def : Pat<(i32 (and (i32 (intOp (v8i8 V64:$Rn))), maski8_or_more)),
3460 (i32 (EXTRACT_SUBREG
3461 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3462 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3464 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
3465 (i32 (EXTRACT_SUBREG
3466 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3467 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3469 // If there is a masking operation keeping only what has been actually
3470 // generated, consume it.
3471 def : Pat<(i32 (and (i32 (intOp (v16i8 V128:$Rn))), maski8_or_more)),
3472 (i32 (EXTRACT_SUBREG
3473 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3474 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3476 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
3477 (i32 (EXTRACT_SUBREG
3478 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3479 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3482 // If there is a masking operation keeping only what has been actually
3483 // generated, consume it.
3484 def : Pat<(i32 (and (i32 (intOp (v4i16 V64:$Rn))), maski16_or_more)),
3485 (i32 (EXTRACT_SUBREG
3486 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3487 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3489 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
3490 (i32 (EXTRACT_SUBREG
3491 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3492 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3494 // If there is a masking operation keeping only what has been actually
3495 // generated, consume it.
3496 def : Pat<(i32 (and (i32 (intOp (v8i16 V128:$Rn))), maski16_or_more)),
3497 (i32 (EXTRACT_SUBREG
3498 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3499 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3501 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
3502 (i32 (EXTRACT_SUBREG
3503 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3504 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3507 def : Pat<(i32 (intOp (v4i32 V128:$Rn))),
3508 (i32 (EXTRACT_SUBREG
3509 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3510 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), ssub),
3515 multiclass SIMDAcrossLanesSignedLongIntrinsic<string baseOpc, Intrinsic intOp> {
3516 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
3518 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3519 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), hsub),
3521 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
3523 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3524 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), hsub),
3527 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
3528 (i32 (EXTRACT_SUBREG
3529 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3530 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), ssub),
3532 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
3533 (i32 (EXTRACT_SUBREG
3534 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3535 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), ssub),
3538 def : Pat<(i64 (intOp (v4i32 V128:$Rn))),
3539 (i64 (EXTRACT_SUBREG
3540 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3541 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), dsub),
3545 multiclass SIMDAcrossLanesUnsignedLongIntrinsic<string baseOpc,
3547 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
3548 (i32 (EXTRACT_SUBREG
3549 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3550 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), hsub),
3552 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
3553 (i32 (EXTRACT_SUBREG
3554 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3555 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), hsub),
3558 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
3559 (i32 (EXTRACT_SUBREG
3560 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3561 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), ssub),
3563 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
3564 (i32 (EXTRACT_SUBREG
3565 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3566 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), ssub),
3569 def : Pat<(i64 (intOp (v4i32 V128:$Rn))),
3570 (i64 (EXTRACT_SUBREG
3571 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3572 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), dsub),
3576 defm : SIMDAcrossLanesSignedIntrinsic<"ADDV", int_arm64_neon_saddv>;
3577 // vaddv_[su]32 is special; -> ADDP Vd.2S,Vn.2S,Vm.2S; return Vd.s[0];Vn==Vm
3578 def : Pat<(i32 (int_arm64_neon_saddv (v2i32 V64:$Rn))),
3579 (EXTRACT_SUBREG (ADDPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3581 defm : SIMDAcrossLanesUnsignedIntrinsic<"ADDV", int_arm64_neon_uaddv>;
3582 // vaddv_[su]32 is special; -> ADDP Vd.2S,Vn.2S,Vm.2S; return Vd.s[0];Vn==Vm
3583 def : Pat<(i32 (int_arm64_neon_uaddv (v2i32 V64:$Rn))),
3584 (EXTRACT_SUBREG (ADDPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3586 defm : SIMDAcrossLanesSignedIntrinsic<"SMAXV", int_arm64_neon_smaxv>;
3587 def : Pat<(i32 (int_arm64_neon_smaxv (v2i32 V64:$Rn))),
3588 (EXTRACT_SUBREG (SMAXPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3590 defm : SIMDAcrossLanesSignedIntrinsic<"SMINV", int_arm64_neon_sminv>;
3591 def : Pat<(i32 (int_arm64_neon_sminv (v2i32 V64:$Rn))),
3592 (EXTRACT_SUBREG (SMINPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3594 defm : SIMDAcrossLanesUnsignedIntrinsic<"UMAXV", int_arm64_neon_umaxv>;
3595 def : Pat<(i32 (int_arm64_neon_umaxv (v2i32 V64:$Rn))),
3596 (EXTRACT_SUBREG (UMAXPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3598 defm : SIMDAcrossLanesUnsignedIntrinsic<"UMINV", int_arm64_neon_uminv>;
3599 def : Pat<(i32 (int_arm64_neon_uminv (v2i32 V64:$Rn))),
3600 (EXTRACT_SUBREG (UMINPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3602 defm : SIMDAcrossLanesSignedLongIntrinsic<"SADDLV", int_arm64_neon_saddlv>;
3603 defm : SIMDAcrossLanesUnsignedLongIntrinsic<"UADDLV", int_arm64_neon_uaddlv>;
3605 // The vaddlv_s32 intrinsic gets mapped to SADDLP.
3606 def : Pat<(i64 (int_arm64_neon_saddlv (v2i32 V64:$Rn))),
3607 (i64 (EXTRACT_SUBREG
3608 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3609 (SADDLPv2i32_v1i64 V64:$Rn), dsub),
3611 // The vaddlv_u32 intrinsic gets mapped to UADDLP.
3612 def : Pat<(i64 (int_arm64_neon_uaddlv (v2i32 V64:$Rn))),
3613 (i64 (EXTRACT_SUBREG
3614 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3615 (UADDLPv2i32_v1i64 V64:$Rn), dsub),
3618 //------------------------------------------------------------------------------
3619 // AdvSIMD modified immediate instructions
3620 //------------------------------------------------------------------------------
3623 defm BIC : SIMDModifiedImmVectorShiftTied<1, 0b11, 0b01, "bic", ARM64bici>;
3625 defm ORR : SIMDModifiedImmVectorShiftTied<0, 0b11, 0b01, "orr", ARM64orri>;
3629 def FMOVv2f64_ns : SIMDModifiedImmVectorNoShift<1, 1, 0b1111, V128, fpimm8,
3631 [(set (v2f64 V128:$Rd), (ARM64fmov imm0_255:$imm8))]>;
3632 def FMOVv2f32_ns : SIMDModifiedImmVectorNoShift<0, 0, 0b1111, V64, fpimm8,
3634 [(set (v2f32 V64:$Rd), (ARM64fmov imm0_255:$imm8))]>;
3635 def FMOVv4f32_ns : SIMDModifiedImmVectorNoShift<1, 0, 0b1111, V128, fpimm8,
3637 [(set (v4f32 V128:$Rd), (ARM64fmov imm0_255:$imm8))]>;
3641 // EDIT byte mask: scalar
3642 let isReMaterializable = 1, isAsCheapAsAMove = 1 in
3643 def MOVID : SIMDModifiedImmScalarNoShift<0, 1, 0b1110, "movi",
3644 [(set FPR64:$Rd, simdimmtype10:$imm8)]>;
3645 // The movi_edit node has the immediate value already encoded, so we use
3646 // a plain imm0_255 here.
3647 def : Pat<(f64 (ARM64movi_edit imm0_255:$shift)),
3648 (MOVID imm0_255:$shift)>;
3650 def : Pat<(v1i64 immAllZerosV), (MOVID (i32 0))>;
3651 def : Pat<(v2i32 immAllZerosV), (MOVID (i32 0))>;
3652 def : Pat<(v4i16 immAllZerosV), (MOVID (i32 0))>;
3653 def : Pat<(v8i8 immAllZerosV), (MOVID (i32 0))>;
3655 def : Pat<(v1i64 immAllOnesV), (MOVID (i32 255))>;
3656 def : Pat<(v2i32 immAllOnesV), (MOVID (i32 255))>;
3657 def : Pat<(v4i16 immAllOnesV), (MOVID (i32 255))>;
3658 def : Pat<(v8i8 immAllOnesV), (MOVID (i32 255))>;
3660 // EDIT byte mask: 2d
3662 // The movi_edit node has the immediate value already encoded, so we use
3663 // a plain imm0_255 in the pattern
3664 let isReMaterializable = 1, isAsCheapAsAMove = 1 in
3665 def MOVIv2d_ns : SIMDModifiedImmVectorNoShift<1, 1, 0b1110, V128,
3668 [(set (v2i64 V128:$Rd), (ARM64movi_edit imm0_255:$imm8))]>;
3671 // Use movi.2d to materialize 0.0 if the HW does zero-cycle zeroing.
3672 // Complexity is added to break a tie with a plain MOVI.
3673 let AddedComplexity = 1 in {
3674 def : Pat<(f32 fpimm0),
3675 (f32 (EXTRACT_SUBREG (v2i64 (MOVIv2d_ns (i32 0))), ssub))>,
3677 def : Pat<(f64 fpimm0),
3678 (f64 (EXTRACT_SUBREG (v2i64 (MOVIv2d_ns (i32 0))), dsub))>,
3682 def : Pat<(v2i64 immAllZerosV), (MOVIv2d_ns (i32 0))>;
3683 def : Pat<(v4i32 immAllZerosV), (MOVIv2d_ns (i32 0))>;
3684 def : Pat<(v8i16 immAllZerosV), (MOVIv2d_ns (i32 0))>;
3685 def : Pat<(v16i8 immAllZerosV), (MOVIv2d_ns (i32 0))>;
3687 def : Pat<(v2i64 immAllOnesV), (MOVIv2d_ns (i32 255))>;
3688 def : Pat<(v4i32 immAllOnesV), (MOVIv2d_ns (i32 255))>;
3689 def : Pat<(v8i16 immAllOnesV), (MOVIv2d_ns (i32 255))>;
3690 def : Pat<(v16i8 immAllOnesV), (MOVIv2d_ns (i32 255))>;
3692 def : Pat<(v2f64 (ARM64dup (f64 fpimm0))), (MOVIv2d_ns (i32 0))>;
3693 def : Pat<(v4f32 (ARM64dup (f32 fpimm0))), (MOVIv2d_ns (i32 0))>;
3695 // EDIT per word & halfword: 2s, 4h, 4s, & 8h
3696 defm MOVI : SIMDModifiedImmVectorShift<0, 0b10, 0b00, "movi">;
3697 def : Pat<(v2i32 (ARM64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
3698 (MOVIv2i32 imm0_255:$imm8, imm:$shift)>;
3699 def : Pat<(v4i32 (ARM64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
3700 (MOVIv4i32 imm0_255:$imm8, imm:$shift)>;
3701 def : Pat<(v4i16 (ARM64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
3702 (MOVIv4i16 imm0_255:$imm8, imm:$shift)>;
3703 def : Pat<(v8i16 (ARM64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
3704 (MOVIv8i16 imm0_255:$imm8, imm:$shift)>;
3706 // EDIT per word: 2s & 4s with MSL shifter
3707 def MOVIv2s_msl : SIMDModifiedImmMoveMSL<0, 0, {1,1,0,?}, V64, "movi", ".2s",
3708 [(set (v2i32 V64:$Rd),
3709 (ARM64movi_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
3710 def MOVIv4s_msl : SIMDModifiedImmMoveMSL<1, 0, {1,1,0,?}, V128, "movi", ".4s",
3711 [(set (v4i32 V128:$Rd),
3712 (ARM64movi_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
3714 // Per byte: 8b & 16b
3715 def MOVIv8b_ns : SIMDModifiedImmVectorNoShift<0, 0, 0b1110, V64, imm0_255,
3717 [(set (v8i8 V64:$Rd), (ARM64movi imm0_255:$imm8))]>;
3718 def MOVIv16b_ns : SIMDModifiedImmVectorNoShift<1, 0, 0b1110, V128, imm0_255,
3720 [(set (v16i8 V128:$Rd), (ARM64movi imm0_255:$imm8))]>;
3724 // EDIT per word & halfword: 2s, 4h, 4s, & 8h
3725 defm MVNI : SIMDModifiedImmVectorShift<1, 0b10, 0b00, "mvni">;
3726 def : Pat<(v2i32 (ARM64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
3727 (MVNIv2i32 imm0_255:$imm8, imm:$shift)>;
3728 def : Pat<(v4i32 (ARM64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
3729 (MVNIv4i32 imm0_255:$imm8, imm:$shift)>;
3730 def : Pat<(v4i16 (ARM64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
3731 (MVNIv4i16 imm0_255:$imm8, imm:$shift)>;
3732 def : Pat<(v8i16 (ARM64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
3733 (MVNIv8i16 imm0_255:$imm8, imm:$shift)>;
3735 // EDIT per word: 2s & 4s with MSL shifter
3736 def MVNIv2s_msl : SIMDModifiedImmMoveMSL<0, 1, {1,1,0,?}, V64, "mvni", ".2s",
3737 [(set (v2i32 V64:$Rd),
3738 (ARM64mvni_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
3739 def MVNIv4s_msl : SIMDModifiedImmMoveMSL<1, 1, {1,1,0,?}, V128, "mvni", ".4s",
3740 [(set (v4i32 V128:$Rd),
3741 (ARM64mvni_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
3743 //----------------------------------------------------------------------------
3744 // AdvSIMD indexed element
3745 //----------------------------------------------------------------------------
3747 let neverHasSideEffects = 1 in {
3748 defm FMLA : SIMDFPIndexedSDTied<0, 0b0001, "fmla">;
3749 defm FMLS : SIMDFPIndexedSDTied<0, 0b0101, "fmls">;
3752 // NOTE: Operands are reordered in the FMLA/FMLS PatFrags because the
3753 // instruction expects the addend first, while the intrinsic expects it last.
3755 // On the other hand, there are quite a few valid combinatorial options due to
3756 // the commutativity of multiplication and the fact that (-x) * y = x * (-y).
3757 defm : SIMDFPIndexedSDTiedPatterns<"FMLA",
3758 TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)>>;
3759 defm : SIMDFPIndexedSDTiedPatterns<"FMLA",
3760 TriOpFrag<(fma node:$MHS, node:$RHS, node:$LHS)>>;
3762 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
3763 TriOpFrag<(fma node:$MHS, (fneg node:$RHS), node:$LHS)> >;
3764 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
3765 TriOpFrag<(fma node:$RHS, (fneg node:$MHS), node:$LHS)> >;
3766 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
3767 TriOpFrag<(fma (fneg node:$RHS), node:$MHS, node:$LHS)> >;
3768 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
3769 TriOpFrag<(fma (fneg node:$MHS), node:$RHS, node:$LHS)> >;
3771 multiclass FMLSIndexedAfterNegPatterns<SDPatternOperator OpNode> {
3772 // 3 variants for the .2s version: DUPLANE from 128-bit, DUPLANE from 64-bit
3774 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
3775 (ARM64duplane32 (v4f32 (fneg V128:$Rm)),
3776 VectorIndexS:$idx))),
3777 (FMLSv2i32_indexed V64:$Rd, V64:$Rn, V128:$Rm, VectorIndexS:$idx)>;
3778 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
3779 (v2f32 (ARM64duplane32
3780 (v4f32 (insert_subvector undef,
3781 (v2f32 (fneg V64:$Rm)),
3783 VectorIndexS:$idx)))),
3784 (FMLSv2i32_indexed V64:$Rd, V64:$Rn,
3785 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub),
3786 VectorIndexS:$idx)>;
3787 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
3788 (ARM64dup (f32 (fneg FPR32Op:$Rm))))),
3789 (FMLSv2i32_indexed V64:$Rd, V64:$Rn,
3790 (SUBREG_TO_REG (i32 0), FPR32Op:$Rm, ssub), (i64 0))>;
3792 // 3 variants for the .4s version: DUPLANE from 128-bit, DUPLANE from 64-bit
3794 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
3795 (ARM64duplane32 (v4f32 (fneg V128:$Rm)),
3796 VectorIndexS:$idx))),
3797 (FMLSv4i32_indexed V128:$Rd, V128:$Rn, V128:$Rm,
3798 VectorIndexS:$idx)>;
3799 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
3800 (v4f32 (ARM64duplane32
3801 (v4f32 (insert_subvector undef,
3802 (v2f32 (fneg V64:$Rm)),
3804 VectorIndexS:$idx)))),
3805 (FMLSv4i32_indexed V128:$Rd, V128:$Rn,
3806 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub),
3807 VectorIndexS:$idx)>;
3808 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
3809 (ARM64dup (f32 (fneg FPR32Op:$Rm))))),
3810 (FMLSv4i32_indexed V128:$Rd, V128:$Rn,
3811 (SUBREG_TO_REG (i32 0), FPR32Op:$Rm, ssub), (i64 0))>;
3813 // 2 variants for the .2d version: DUPLANE from 128-bit, and DUP scalar
3814 // (DUPLANE from 64-bit would be trivial).
3815 def : Pat<(v2f64 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn),
3816 (ARM64duplane64 (v2f64 (fneg V128:$Rm)),
3817 VectorIndexD:$idx))),
3819 V128:$Rd, V128:$Rn, V128:$Rm, VectorIndexS:$idx)>;
3820 def : Pat<(v2f64 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn),
3821 (ARM64dup (f64 (fneg FPR64Op:$Rm))))),
3822 (FMLSv2i64_indexed V128:$Rd, V128:$Rn,
3823 (SUBREG_TO_REG (i32 0), FPR64Op:$Rm, dsub), (i64 0))>;
3825 // 2 variants for 32-bit scalar version: extract from .2s or from .4s
3826 def : Pat<(f32 (OpNode (f32 FPR32:$Rd), (f32 FPR32:$Rn),
3827 (vector_extract (v4f32 (fneg V128:$Rm)),
3828 VectorIndexS:$idx))),
3829 (FMLSv1i32_indexed FPR32:$Rd, FPR32:$Rn,
3830 V128:$Rm, VectorIndexS:$idx)>;
3831 def : Pat<(f32 (OpNode (f32 FPR32:$Rd), (f32 FPR32:$Rn),
3832 (vector_extract (v2f32 (fneg V64:$Rm)),
3833 VectorIndexS:$idx))),
3834 (FMLSv1i32_indexed FPR32:$Rd, FPR32:$Rn,
3835 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub), VectorIndexS:$idx)>;
3837 // 1 variant for 64-bit scalar version: extract from .1d or from .2d
3838 def : Pat<(f64 (OpNode (f64 FPR64:$Rd), (f64 FPR64:$Rn),
3839 (vector_extract (v2f64 (fneg V128:$Rm)),
3840 VectorIndexS:$idx))),
3841 (FMLSv1i64_indexed FPR64:$Rd, FPR64:$Rn,
3842 V128:$Rm, VectorIndexS:$idx)>;
3845 defm : FMLSIndexedAfterNegPatterns<
3846 TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)> >;
3847 defm : FMLSIndexedAfterNegPatterns<
3848 TriOpFrag<(fma node:$MHS, node:$RHS, node:$LHS)> >;
3850 defm FMULX : SIMDFPIndexedSD<1, 0b1001, "fmulx", int_arm64_neon_fmulx>;
3851 defm FMUL : SIMDFPIndexedSD<0, 0b1001, "fmul", fmul>;
3853 def : Pat<(v2f32 (fmul V64:$Rn, (ARM64dup (f32 FPR32:$Rm)))),
3854 (FMULv2i32_indexed V64:$Rn,
3855 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rm, ssub),
3857 def : Pat<(v4f32 (fmul V128:$Rn, (ARM64dup (f32 FPR32:$Rm)))),
3858 (FMULv4i32_indexed V128:$Rn,
3859 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rm, ssub),
3861 def : Pat<(v2f64 (fmul V128:$Rn, (ARM64dup (f64 FPR64:$Rm)))),
3862 (FMULv2i64_indexed V128:$Rn,
3863 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$Rm, dsub),
3866 defm SQDMULH : SIMDIndexedHS<0, 0b1100, "sqdmulh", int_arm64_neon_sqdmulh>;
3867 defm SQRDMULH : SIMDIndexedHS<0, 0b1101, "sqrdmulh", int_arm64_neon_sqrdmulh>;
3868 defm MLA : SIMDVectorIndexedHSTied<1, 0b0000, "mla",
3869 TriOpFrag<(add node:$LHS, (mul node:$MHS, node:$RHS))>>;
3870 defm MLS : SIMDVectorIndexedHSTied<1, 0b0100, "mls",
3871 TriOpFrag<(sub node:$LHS, (mul node:$MHS, node:$RHS))>>;
3872 defm MUL : SIMDVectorIndexedHS<0, 0b1000, "mul", mul>;
3873 defm SMLAL : SIMDVectorIndexedLongSDTied<0, 0b0010, "smlal",
3874 TriOpFrag<(add node:$LHS, (int_arm64_neon_smull node:$MHS, node:$RHS))>>;
3875 defm SMLSL : SIMDVectorIndexedLongSDTied<0, 0b0110, "smlsl",
3876 TriOpFrag<(sub node:$LHS, (int_arm64_neon_smull node:$MHS, node:$RHS))>>;
3877 defm SMULL : SIMDVectorIndexedLongSD<0, 0b1010, "smull",
3878 int_arm64_neon_smull>;
3879 defm SQDMLAL : SIMDIndexedLongSQDMLXSDTied<0, 0b0011, "sqdmlal",
3880 int_arm64_neon_sqadd>;
3881 defm SQDMLSL : SIMDIndexedLongSQDMLXSDTied<0, 0b0111, "sqdmlsl",
3882 int_arm64_neon_sqsub>;
3883 defm SQDMULL : SIMDIndexedLongSD<0, 0b1011, "sqdmull", int_arm64_neon_sqdmull>;
3884 defm UMLAL : SIMDVectorIndexedLongSDTied<1, 0b0010, "umlal",
3885 TriOpFrag<(add node:$LHS, (int_arm64_neon_umull node:$MHS, node:$RHS))>>;
3886 defm UMLSL : SIMDVectorIndexedLongSDTied<1, 0b0110, "umlsl",
3887 TriOpFrag<(sub node:$LHS, (int_arm64_neon_umull node:$MHS, node:$RHS))>>;
3888 defm UMULL : SIMDVectorIndexedLongSD<1, 0b1010, "umull",
3889 int_arm64_neon_umull>;
3891 // A scalar sqdmull with the second operand being a vector lane can be
3892 // handled directly with the indexed instruction encoding.
3893 def : Pat<(int_arm64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
3894 (vector_extract (v4i32 V128:$Vm),
3895 VectorIndexS:$idx)),
3896 (SQDMULLv1i64_indexed FPR32:$Rn, V128:$Vm, VectorIndexS:$idx)>;
3898 //----------------------------------------------------------------------------
3899 // AdvSIMD scalar shift instructions
3900 //----------------------------------------------------------------------------
3901 defm FCVTZS : SIMDScalarRShiftSD<0, 0b11111, "fcvtzs">;
3902 defm FCVTZU : SIMDScalarRShiftSD<1, 0b11111, "fcvtzu">;
3903 defm SCVTF : SIMDScalarRShiftSD<0, 0b11100, "scvtf">;
3904 defm UCVTF : SIMDScalarRShiftSD<1, 0b11100, "ucvtf">;
3905 // Codegen patterns for the above. We don't put these directly on the
3906 // instructions because TableGen's type inference can't handle the truth.
3907 // Having the same base pattern for fp <--> int totally freaks it out.
3908 def : Pat<(int_arm64_neon_vcvtfp2fxs FPR32:$Rn, vecshiftR32:$imm),
3909 (FCVTZSs FPR32:$Rn, vecshiftR32:$imm)>;
3910 def : Pat<(int_arm64_neon_vcvtfp2fxu FPR32:$Rn, vecshiftR32:$imm),
3911 (FCVTZUs FPR32:$Rn, vecshiftR32:$imm)>;
3912 def : Pat<(i64 (int_arm64_neon_vcvtfp2fxs (f64 FPR64:$Rn), vecshiftR64:$imm)),
3913 (FCVTZSd FPR64:$Rn, vecshiftR64:$imm)>;
3914 def : Pat<(i64 (int_arm64_neon_vcvtfp2fxu (f64 FPR64:$Rn), vecshiftR64:$imm)),
3915 (FCVTZUd FPR64:$Rn, vecshiftR64:$imm)>;
3916 def : Pat<(v1i64 (int_arm64_neon_vcvtfp2fxs (v1f64 FPR64:$Rn),
3918 (FCVTZSd FPR64:$Rn, vecshiftR64:$imm)>;
3919 def : Pat<(v1i64 (int_arm64_neon_vcvtfp2fxu (v1f64 FPR64:$Rn),
3921 (FCVTZUd FPR64:$Rn, vecshiftR64:$imm)>;
3922 def : Pat<(int_arm64_neon_vcvtfxs2fp FPR32:$Rn, vecshiftR32:$imm),
3923 (SCVTFs FPR32:$Rn, vecshiftR32:$imm)>;
3924 def : Pat<(int_arm64_neon_vcvtfxu2fp FPR32:$Rn, vecshiftR32:$imm),
3925 (UCVTFs FPR32:$Rn, vecshiftR32:$imm)>;
3926 def : Pat<(f64 (int_arm64_neon_vcvtfxs2fp (i64 FPR64:$Rn), vecshiftR64:$imm)),
3927 (SCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
3928 def : Pat<(f64 (int_arm64_neon_vcvtfxu2fp (i64 FPR64:$Rn), vecshiftR64:$imm)),
3929 (UCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
3930 def : Pat<(v1f64 (int_arm64_neon_vcvtfxs2fp (v1i64 FPR64:$Rn),
3932 (SCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
3933 def : Pat<(v1f64 (int_arm64_neon_vcvtfxu2fp (v1i64 FPR64:$Rn),
3935 (UCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
3937 defm SHL : SIMDScalarLShiftD< 0, 0b01010, "shl", ARM64vshl>;
3938 defm SLI : SIMDScalarLShiftDTied<1, 0b01010, "sli">;
3939 defm SQRSHRN : SIMDScalarRShiftBHS< 0, 0b10011, "sqrshrn",
3940 int_arm64_neon_sqrshrn>;
3941 defm SQRSHRUN : SIMDScalarRShiftBHS< 1, 0b10001, "sqrshrun",
3942 int_arm64_neon_sqrshrun>;
3943 defm SQSHLU : SIMDScalarLShiftBHSD<1, 0b01100, "sqshlu", ARM64sqshlui>;
3944 defm SQSHL : SIMDScalarLShiftBHSD<0, 0b01110, "sqshl", ARM64sqshli>;
3945 defm SQSHRN : SIMDScalarRShiftBHS< 0, 0b10010, "sqshrn",
3946 int_arm64_neon_sqshrn>;
3947 defm SQSHRUN : SIMDScalarRShiftBHS< 1, 0b10000, "sqshrun",
3948 int_arm64_neon_sqshrun>;
3949 defm SRI : SIMDScalarRShiftDTied< 1, 0b01000, "sri">;
3950 defm SRSHR : SIMDScalarRShiftD< 0, 0b00100, "srshr", ARM64srshri>;
3951 defm SRSRA : SIMDScalarRShiftDTied< 0, 0b00110, "srsra",
3952 TriOpFrag<(add node:$LHS,
3953 (ARM64srshri node:$MHS, node:$RHS))>>;
3954 defm SSHR : SIMDScalarRShiftD< 0, 0b00000, "sshr", ARM64vashr>;
3955 defm SSRA : SIMDScalarRShiftDTied< 0, 0b00010, "ssra",
3956 TriOpFrag<(add node:$LHS,
3957 (ARM64vashr node:$MHS, node:$RHS))>>;
3958 defm UQRSHRN : SIMDScalarRShiftBHS< 1, 0b10011, "uqrshrn",
3959 int_arm64_neon_uqrshrn>;
3960 defm UQSHL : SIMDScalarLShiftBHSD<1, 0b01110, "uqshl", ARM64uqshli>;
3961 defm UQSHRN : SIMDScalarRShiftBHS< 1, 0b10010, "uqshrn",
3962 int_arm64_neon_uqshrn>;
3963 defm URSHR : SIMDScalarRShiftD< 1, 0b00100, "urshr", ARM64urshri>;
3964 defm URSRA : SIMDScalarRShiftDTied< 1, 0b00110, "ursra",
3965 TriOpFrag<(add node:$LHS,
3966 (ARM64urshri node:$MHS, node:$RHS))>>;
3967 defm USHR : SIMDScalarRShiftD< 1, 0b00000, "ushr", ARM64vlshr>;
3968 defm USRA : SIMDScalarRShiftDTied< 1, 0b00010, "usra",
3969 TriOpFrag<(add node:$LHS,
3970 (ARM64vlshr node:$MHS, node:$RHS))>>;
3972 //----------------------------------------------------------------------------
3973 // AdvSIMD vector shift instructions
3974 //----------------------------------------------------------------------------
3975 defm FCVTZS:SIMDVectorRShiftSD<0, 0b11111, "fcvtzs", int_arm64_neon_vcvtfp2fxs>;
3976 defm FCVTZU:SIMDVectorRShiftSD<1, 0b11111, "fcvtzu", int_arm64_neon_vcvtfp2fxu>;
3977 defm SCVTF: SIMDVectorRShiftSDToFP<0, 0b11100, "scvtf",
3978 int_arm64_neon_vcvtfxs2fp>;
3979 defm RSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10001, "rshrn",
3980 int_arm64_neon_rshrn>;
3981 defm SHL : SIMDVectorLShiftBHSD<0, 0b01010, "shl", ARM64vshl>;
3982 defm SHRN : SIMDVectorRShiftNarrowBHS<0, 0b10000, "shrn",
3983 BinOpFrag<(trunc (ARM64vashr node:$LHS, node:$RHS))>>;
3984 defm SLI : SIMDVectorLShiftBHSDTied<1, 0b01010, "sli", int_arm64_neon_vsli>;
3985 def : Pat<(v1i64 (int_arm64_neon_vsli (v1i64 FPR64:$Rd), (v1i64 FPR64:$Rn),
3986 (i32 vecshiftL64:$imm))),
3987 (SLId FPR64:$Rd, FPR64:$Rn, vecshiftL64:$imm)>;
3988 defm SQRSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10011, "sqrshrn",
3989 int_arm64_neon_sqrshrn>;
3990 defm SQRSHRUN: SIMDVectorRShiftNarrowBHS<1, 0b10001, "sqrshrun",
3991 int_arm64_neon_sqrshrun>;
3992 defm SQSHLU : SIMDVectorLShiftBHSD<1, 0b01100, "sqshlu", ARM64sqshlui>;
3993 defm SQSHL : SIMDVectorLShiftBHSD<0, 0b01110, "sqshl", ARM64sqshli>;
3994 defm SQSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10010, "sqshrn",
3995 int_arm64_neon_sqshrn>;
3996 defm SQSHRUN : SIMDVectorRShiftNarrowBHS<1, 0b10000, "sqshrun",
3997 int_arm64_neon_sqshrun>;
3998 defm SRI : SIMDVectorRShiftBHSDTied<1, 0b01000, "sri", int_arm64_neon_vsri>;
3999 def : Pat<(v1i64 (int_arm64_neon_vsri (v1i64 FPR64:$Rd), (v1i64 FPR64:$Rn),
4000 (i32 vecshiftR64:$imm))),
4001 (SRId FPR64:$Rd, FPR64:$Rn, vecshiftR64:$imm)>;
4002 defm SRSHR : SIMDVectorRShiftBHSD<0, 0b00100, "srshr", ARM64srshri>;
4003 defm SRSRA : SIMDVectorRShiftBHSDTied<0, 0b00110, "srsra",
4004 TriOpFrag<(add node:$LHS,
4005 (ARM64srshri node:$MHS, node:$RHS))> >;
4006 defm SSHLL : SIMDVectorLShiftLongBHSD<0, 0b10100, "sshll",
4007 BinOpFrag<(ARM64vshl (sext node:$LHS), node:$RHS)>>;
4009 defm SSHR : SIMDVectorRShiftBHSD<0, 0b00000, "sshr", ARM64vashr>;
4010 defm SSRA : SIMDVectorRShiftBHSDTied<0, 0b00010, "ssra",
4011 TriOpFrag<(add node:$LHS, (ARM64vashr node:$MHS, node:$RHS))>>;
4012 defm UCVTF : SIMDVectorRShiftSDToFP<1, 0b11100, "ucvtf",
4013 int_arm64_neon_vcvtfxu2fp>;
4014 defm UQRSHRN : SIMDVectorRShiftNarrowBHS<1, 0b10011, "uqrshrn",
4015 int_arm64_neon_uqrshrn>;
4016 defm UQSHL : SIMDVectorLShiftBHSD<1, 0b01110, "uqshl", ARM64uqshli>;
4017 defm UQSHRN : SIMDVectorRShiftNarrowBHS<1, 0b10010, "uqshrn",
4018 int_arm64_neon_uqshrn>;
4019 defm URSHR : SIMDVectorRShiftBHSD<1, 0b00100, "urshr", ARM64urshri>;
4020 defm URSRA : SIMDVectorRShiftBHSDTied<1, 0b00110, "ursra",
4021 TriOpFrag<(add node:$LHS,
4022 (ARM64urshri node:$MHS, node:$RHS))> >;
4023 defm USHLL : SIMDVectorLShiftLongBHSD<1, 0b10100, "ushll",
4024 BinOpFrag<(ARM64vshl (zext node:$LHS), node:$RHS)>>;
4025 defm USHR : SIMDVectorRShiftBHSD<1, 0b00000, "ushr", ARM64vlshr>;
4026 defm USRA : SIMDVectorRShiftBHSDTied<1, 0b00010, "usra",
4027 TriOpFrag<(add node:$LHS, (ARM64vlshr node:$MHS, node:$RHS))> >;
4029 // SHRN patterns for when a logical right shift was used instead of arithmetic
4030 // (the immediate guarantees no sign bits actually end up in the result so it
4032 def : Pat<(v8i8 (trunc (ARM64vlshr (v8i16 V128:$Rn), vecshiftR16Narrow:$imm))),
4033 (SHRNv8i8_shift V128:$Rn, vecshiftR16Narrow:$imm)>;
4034 def : Pat<(v4i16 (trunc (ARM64vlshr (v4i32 V128:$Rn), vecshiftR32Narrow:$imm))),
4035 (SHRNv4i16_shift V128:$Rn, vecshiftR32Narrow:$imm)>;
4036 def : Pat<(v2i32 (trunc (ARM64vlshr (v2i64 V128:$Rn), vecshiftR64Narrow:$imm))),
4037 (SHRNv2i32_shift V128:$Rn, vecshiftR64Narrow:$imm)>;
4039 def : Pat<(v16i8 (concat_vectors (v8i8 V64:$Rd),
4040 (trunc (ARM64vlshr (v8i16 V128:$Rn),
4041 vecshiftR16Narrow:$imm)))),
4042 (SHRNv16i8_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
4043 V128:$Rn, vecshiftR16Narrow:$imm)>;
4044 def : Pat<(v8i16 (concat_vectors (v4i16 V64:$Rd),
4045 (trunc (ARM64vlshr (v4i32 V128:$Rn),
4046 vecshiftR32Narrow:$imm)))),
4047 (SHRNv8i16_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
4048 V128:$Rn, vecshiftR32Narrow:$imm)>;
4049 def : Pat<(v4i32 (concat_vectors (v2i32 V64:$Rd),
4050 (trunc (ARM64vlshr (v2i64 V128:$Rn),
4051 vecshiftR64Narrow:$imm)))),
4052 (SHRNv4i32_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
4053 V128:$Rn, vecshiftR32Narrow:$imm)>;
4055 // Vector sign and zero extensions are implemented with SSHLL and USSHLL.
4056 // Anyexts are implemented as zexts.
4057 def : Pat<(v8i16 (sext (v8i8 V64:$Rn))), (SSHLLv8i8_shift V64:$Rn, (i32 0))>;
4058 def : Pat<(v8i16 (zext (v8i8 V64:$Rn))), (USHLLv8i8_shift V64:$Rn, (i32 0))>;
4059 def : Pat<(v8i16 (anyext (v8i8 V64:$Rn))), (USHLLv8i8_shift V64:$Rn, (i32 0))>;
4060 def : Pat<(v4i32 (sext (v4i16 V64:$Rn))), (SSHLLv4i16_shift V64:$Rn, (i32 0))>;
4061 def : Pat<(v4i32 (zext (v4i16 V64:$Rn))), (USHLLv4i16_shift V64:$Rn, (i32 0))>;
4062 def : Pat<(v4i32 (anyext (v4i16 V64:$Rn))), (USHLLv4i16_shift V64:$Rn, (i32 0))>;
4063 def : Pat<(v2i64 (sext (v2i32 V64:$Rn))), (SSHLLv2i32_shift V64:$Rn, (i32 0))>;
4064 def : Pat<(v2i64 (zext (v2i32 V64:$Rn))), (USHLLv2i32_shift V64:$Rn, (i32 0))>;
4065 def : Pat<(v2i64 (anyext (v2i32 V64:$Rn))), (USHLLv2i32_shift V64:$Rn, (i32 0))>;
4066 // Also match an extend from the upper half of a 128 bit source register.
4067 def : Pat<(v8i16 (anyext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
4068 (USHLLv16i8_shift V128:$Rn, (i32 0))>;
4069 def : Pat<(v8i16 (zext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
4070 (USHLLv16i8_shift V128:$Rn, (i32 0))>;
4071 def : Pat<(v8i16 (sext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
4072 (SSHLLv16i8_shift V128:$Rn, (i32 0))>;
4073 def : Pat<(v4i32 (anyext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
4074 (USHLLv8i16_shift V128:$Rn, (i32 0))>;
4075 def : Pat<(v4i32 (zext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
4076 (USHLLv8i16_shift V128:$Rn, (i32 0))>;
4077 def : Pat<(v4i32 (sext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
4078 (SSHLLv8i16_shift V128:$Rn, (i32 0))>;
4079 def : Pat<(v2i64 (anyext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
4080 (USHLLv4i32_shift V128:$Rn, (i32 0))>;
4081 def : Pat<(v2i64 (zext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
4082 (USHLLv4i32_shift V128:$Rn, (i32 0))>;
4083 def : Pat<(v2i64 (sext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
4084 (SSHLLv4i32_shift V128:$Rn, (i32 0))>;
4086 // Vector shift sxtl aliases
4087 def : InstAlias<"sxtl.8h $dst, $src1",
4088 (SSHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
4089 def : InstAlias<"sxtl $dst.8h, $src1.8b",
4090 (SSHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
4091 def : InstAlias<"sxtl.4s $dst, $src1",
4092 (SSHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
4093 def : InstAlias<"sxtl $dst.4s, $src1.4h",
4094 (SSHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
4095 def : InstAlias<"sxtl.2d $dst, $src1",
4096 (SSHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
4097 def : InstAlias<"sxtl $dst.2d, $src1.2s",
4098 (SSHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
4100 // Vector shift sxtl2 aliases
4101 def : InstAlias<"sxtl2.8h $dst, $src1",
4102 (SSHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
4103 def : InstAlias<"sxtl2 $dst.8h, $src1.16b",
4104 (SSHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
4105 def : InstAlias<"sxtl2.4s $dst, $src1",
4106 (SSHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
4107 def : InstAlias<"sxtl2 $dst.4s, $src1.8h",
4108 (SSHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
4109 def : InstAlias<"sxtl2.2d $dst, $src1",
4110 (SSHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
4111 def : InstAlias<"sxtl2 $dst.2d, $src1.4s",
4112 (SSHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
4114 // Vector shift uxtl aliases
4115 def : InstAlias<"uxtl.8h $dst, $src1",
4116 (USHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
4117 def : InstAlias<"uxtl $dst.8h, $src1.8b",
4118 (USHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
4119 def : InstAlias<"uxtl.4s $dst, $src1",
4120 (USHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
4121 def : InstAlias<"uxtl $dst.4s, $src1.4h",
4122 (USHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
4123 def : InstAlias<"uxtl.2d $dst, $src1",
4124 (USHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
4125 def : InstAlias<"uxtl $dst.2d, $src1.2s",
4126 (USHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
4128 // Vector shift uxtl2 aliases
4129 def : InstAlias<"uxtl2.8h $dst, $src1",
4130 (USHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
4131 def : InstAlias<"uxtl2 $dst.8h, $src1.16b",
4132 (USHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
4133 def : InstAlias<"uxtl2.4s $dst, $src1",
4134 (USHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
4135 def : InstAlias<"uxtl2 $dst.4s, $src1.8h",
4136 (USHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
4137 def : InstAlias<"uxtl2.2d $dst, $src1",
4138 (USHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
4139 def : InstAlias<"uxtl2 $dst.2d, $src1.4s",
4140 (USHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
4142 // If an integer is about to be converted to a floating point value,
4143 // just load it on the floating point unit.
4144 // These patterns are more complex because floating point loads do not
4145 // support sign extension.
4146 // The sign extension has to be explicitly added and is only supported for
4147 // one step: byte-to-half, half-to-word, word-to-doubleword.
4148 // SCVTF GPR -> FPR is 9 cycles.
4149 // SCVTF FPR -> FPR is 4 cyclces.
4150 // (sign extension with lengthen) SXTL FPR -> FPR is 2 cycles.
4151 // Therefore, we can do 2 sign extensions and one SCVTF FPR -> FPR
4152 // and still being faster.
4153 // However, this is not good for code size.
4154 // 8-bits -> float. 2 sizes step-up.
4155 def : Pat <(f32 (sint_to_fp (i32 (sextloadi8 ro_indexed8:$addr)))),
4156 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
4161 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4162 (LDRBro ro_indexed8:$addr),
4167 ssub)))>, Requires<[NotForCodeSize]>;
4168 def : Pat <(f32 (sint_to_fp (i32 (sextloadi8 am_indexed8:$addr)))),
4169 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
4174 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4175 (LDRBui am_indexed8:$addr),
4180 ssub)))>, Requires<[NotForCodeSize]>;
4181 def : Pat <(f32 (sint_to_fp (i32 (sextloadi8 am_unscaled8:$addr)))),
4182 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
4187 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4188 (LDURBi am_unscaled8:$addr),
4193 ssub)))>, Requires<[NotForCodeSize]>;
4194 // 16-bits -> float. 1 size step-up.
4195 def : Pat <(f32 (sint_to_fp (i32 (sextloadi16 ro_indexed16:$addr)))),
4196 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
4198 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4199 (LDRHro ro_indexed16:$addr),
4202 ssub)))>, Requires<[NotForCodeSize]>;
4203 def : Pat <(f32 (sint_to_fp (i32 (sextloadi16 am_indexed16:$addr)))),
4204 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
4206 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4207 (LDRHui am_indexed16:$addr),
4210 ssub)))>, Requires<[NotForCodeSize]>;
4211 def : Pat <(f32 (sint_to_fp (i32 (sextloadi16 am_unscaled16:$addr)))),
4212 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
4214 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4215 (LDURHi am_unscaled16:$addr),
4218 ssub)))>, Requires<[NotForCodeSize]>;
4219 // 32-bits to 32-bits are handled in target specific dag combine:
4220 // performIntToFpCombine.
4221 // 64-bits integer to 32-bits floating point, not possible with
4222 // SCVTF on floating point registers (both source and destination
4223 // must have the same size).
4225 // Here are the patterns for 8, 16, 32, and 64-bits to double.
4226 // 8-bits -> double. 3 size step-up: give up.
4227 // 16-bits -> double. 2 size step.
4228 def : Pat <(f64 (sint_to_fp (i32 (sextloadi16 ro_indexed16:$addr)))),
4229 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4234 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4235 (LDRHro ro_indexed16:$addr),
4240 dsub)))>, Requires<[NotForCodeSize]>;
4241 def : Pat <(f64 (sint_to_fp (i32 (sextloadi16 am_indexed16:$addr)))),
4242 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4247 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4248 (LDRHui am_indexed16:$addr),
4253 dsub)))>, Requires<[NotForCodeSize]>;
4254 def : Pat <(f64 (sint_to_fp (i32 (sextloadi16 am_unscaled16:$addr)))),
4255 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4260 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4261 (LDURHi am_unscaled16:$addr),
4266 dsub)))>, Requires<[NotForCodeSize]>;
4267 // 32-bits -> double. 1 size step-up.
4268 def : Pat <(f64 (sint_to_fp (i32 (load ro_indexed32:$addr)))),
4269 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4271 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4272 (LDRSro ro_indexed32:$addr),
4275 dsub)))>, Requires<[NotForCodeSize]>;
4276 def : Pat <(f64 (sint_to_fp (i32 (load am_indexed32:$addr)))),
4277 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4279 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4280 (LDRSui am_indexed32:$addr),
4283 dsub)))>, Requires<[NotForCodeSize]>;
4284 def : Pat <(f64 (sint_to_fp (i32 (load am_unscaled32:$addr)))),
4285 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4287 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4288 (LDURSi am_unscaled32:$addr),
4291 dsub)))>, Requires<[NotForCodeSize]>;
4292 // 64-bits -> double are handled in target specific dag combine:
4293 // performIntToFpCombine.
4296 //----------------------------------------------------------------------------
4297 // AdvSIMD Load-Store Structure
4298 //----------------------------------------------------------------------------
4299 defm LD1 : SIMDLd1Multiple<"ld1">;
4300 defm LD2 : SIMDLd2Multiple<"ld2">;
4301 defm LD3 : SIMDLd3Multiple<"ld3">;
4302 defm LD4 : SIMDLd4Multiple<"ld4">;
4304 defm ST1 : SIMDSt1Multiple<"st1">;
4305 defm ST2 : SIMDSt2Multiple<"st2">;
4306 defm ST3 : SIMDSt3Multiple<"st3">;
4307 defm ST4 : SIMDSt4Multiple<"st4">;
4309 class Ld1Pat<ValueType ty, Instruction INST>
4310 : Pat<(ty (load am_simdnoindex:$vaddr)), (INST am_simdnoindex:$vaddr)>;
4312 def : Ld1Pat<v16i8, LD1Onev16b>;
4313 def : Ld1Pat<v8i16, LD1Onev8h>;
4314 def : Ld1Pat<v4i32, LD1Onev4s>;
4315 def : Ld1Pat<v2i64, LD1Onev2d>;
4316 def : Ld1Pat<v8i8, LD1Onev8b>;
4317 def : Ld1Pat<v4i16, LD1Onev4h>;
4318 def : Ld1Pat<v2i32, LD1Onev2s>;
4319 def : Ld1Pat<v1i64, LD1Onev1d>;
4321 class St1Pat<ValueType ty, Instruction INST>
4322 : Pat<(store ty:$Vt, am_simdnoindex:$vaddr),
4323 (INST ty:$Vt, am_simdnoindex:$vaddr)>;
4325 def : St1Pat<v16i8, ST1Onev16b>;
4326 def : St1Pat<v8i16, ST1Onev8h>;
4327 def : St1Pat<v4i32, ST1Onev4s>;
4328 def : St1Pat<v2i64, ST1Onev2d>;
4329 def : St1Pat<v8i8, ST1Onev8b>;
4330 def : St1Pat<v4i16, ST1Onev4h>;
4331 def : St1Pat<v2i32, ST1Onev2s>;
4332 def : St1Pat<v1i64, ST1Onev1d>;
4338 defm LD1R : SIMDLdR<0, 0b110, 0, "ld1r", "One", 1, 2, 4, 8>;
4339 defm LD2R : SIMDLdR<1, 0b110, 0, "ld2r", "Two", 2, 4, 8, 16>;
4340 defm LD3R : SIMDLdR<0, 0b111, 0, "ld3r", "Three", 3, 6, 12, 24>;
4341 defm LD4R : SIMDLdR<1, 0b111, 0, "ld4r", "Four", 4, 8, 16, 32>;
4342 let mayLoad = 1, neverHasSideEffects = 1 in {
4343 defm LD1 : SIMDLdSingleBTied<0, 0b000, "ld1", VecListOneb, GPR64pi1>;
4344 defm LD1 : SIMDLdSingleHTied<0, 0b010, 0, "ld1", VecListOneh, GPR64pi2>;
4345 defm LD1 : SIMDLdSingleSTied<0, 0b100, 0b00, "ld1", VecListOnes, GPR64pi4>;
4346 defm LD1 : SIMDLdSingleDTied<0, 0b100, 0b01, "ld1", VecListOned, GPR64pi8>;
4347 defm LD2 : SIMDLdSingleBTied<1, 0b000, "ld2", VecListTwob, GPR64pi2>;
4348 defm LD2 : SIMDLdSingleHTied<1, 0b010, 0, "ld2", VecListTwoh, GPR64pi4>;
4349 defm LD2 : SIMDLdSingleSTied<1, 0b100, 0b00, "ld2", VecListTwos, GPR64pi8>;
4350 defm LD2 : SIMDLdSingleDTied<1, 0b100, 0b01, "ld2", VecListTwod, GPR64pi16>;
4351 defm LD3 : SIMDLdSingleBTied<0, 0b001, "ld3", VecListThreeb, GPR64pi3>;
4352 defm LD3 : SIMDLdSingleHTied<0, 0b011, 0, "ld3", VecListThreeh, GPR64pi6>;
4353 defm LD3 : SIMDLdSingleSTied<0, 0b101, 0b00, "ld3", VecListThrees, GPR64pi12>;
4354 defm LD3 : SIMDLdSingleDTied<0, 0b101, 0b01, "ld3", VecListThreed, GPR64pi24>;
4355 defm LD4 : SIMDLdSingleBTied<1, 0b001, "ld4", VecListFourb, GPR64pi4>;
4356 defm LD4 : SIMDLdSingleHTied<1, 0b011, 0, "ld4", VecListFourh, GPR64pi8>;
4357 defm LD4 : SIMDLdSingleSTied<1, 0b101, 0b00, "ld4", VecListFours, GPR64pi16>;
4358 defm LD4 : SIMDLdSingleDTied<1, 0b101, 0b01, "ld4", VecListFourd, GPR64pi32>;
4361 def : Pat<(v8i8 (ARM64dup (i32 (extloadi8 am_simdnoindex:$vaddr)))),
4362 (LD1Rv8b am_simdnoindex:$vaddr)>;
4363 def : Pat<(v16i8 (ARM64dup (i32 (extloadi8 am_simdnoindex:$vaddr)))),
4364 (LD1Rv16b am_simdnoindex:$vaddr)>;
4365 def : Pat<(v4i16 (ARM64dup (i32 (extloadi16 am_simdnoindex:$vaddr)))),
4366 (LD1Rv4h am_simdnoindex:$vaddr)>;
4367 def : Pat<(v8i16 (ARM64dup (i32 (extloadi16 am_simdnoindex:$vaddr)))),
4368 (LD1Rv8h am_simdnoindex:$vaddr)>;
4369 def : Pat<(v2i32 (ARM64dup (i32 (load am_simdnoindex:$vaddr)))),
4370 (LD1Rv2s am_simdnoindex:$vaddr)>;
4371 def : Pat<(v4i32 (ARM64dup (i32 (load am_simdnoindex:$vaddr)))),
4372 (LD1Rv4s am_simdnoindex:$vaddr)>;
4373 def : Pat<(v2i64 (ARM64dup (i64 (load am_simdnoindex:$vaddr)))),
4374 (LD1Rv2d am_simdnoindex:$vaddr)>;
4375 def : Pat<(v1i64 (ARM64dup (i64 (load am_simdnoindex:$vaddr)))),
4376 (LD1Rv1d am_simdnoindex:$vaddr)>;
4377 // Grab the floating point version too
4378 def : Pat<(v2f32 (ARM64dup (f32 (load am_simdnoindex:$vaddr)))),
4379 (LD1Rv2s am_simdnoindex:$vaddr)>;
4380 def : Pat<(v4f32 (ARM64dup (f32 (load am_simdnoindex:$vaddr)))),
4381 (LD1Rv4s am_simdnoindex:$vaddr)>;
4382 def : Pat<(v2f64 (ARM64dup (f64 (load am_simdnoindex:$vaddr)))),
4383 (LD1Rv2d am_simdnoindex:$vaddr)>;
4384 def : Pat<(v1f64 (ARM64dup (f64 (load am_simdnoindex:$vaddr)))),
4385 (LD1Rv1d am_simdnoindex:$vaddr)>;
4387 class Ld1Lane128Pat<SDPatternOperator scalar_load, Operand VecIndex,
4388 ValueType VTy, ValueType STy, Instruction LD1>
4389 : Pat<(vector_insert (VTy VecListOne128:$Rd),
4390 (STy (scalar_load am_simdnoindex:$vaddr)), VecIndex:$idx),
4391 (LD1 VecListOne128:$Rd, VecIndex:$idx, am_simdnoindex:$vaddr)>;
4393 def : Ld1Lane128Pat<extloadi8, VectorIndexB, v16i8, i32, LD1i8>;
4394 def : Ld1Lane128Pat<extloadi16, VectorIndexH, v8i16, i32, LD1i16>;
4395 def : Ld1Lane128Pat<load, VectorIndexS, v4i32, i32, LD1i32>;
4396 def : Ld1Lane128Pat<load, VectorIndexS, v4f32, f32, LD1i32>;
4397 def : Ld1Lane128Pat<load, VectorIndexD, v2i64, i64, LD1i64>;
4398 def : Ld1Lane128Pat<load, VectorIndexD, v2f64, f64, LD1i64>;
4400 class Ld1Lane64Pat<SDPatternOperator scalar_load, Operand VecIndex,
4401 ValueType VTy, ValueType STy, Instruction LD1>
4402 : Pat<(vector_insert (VTy VecListOne64:$Rd),
4403 (STy (scalar_load am_simdnoindex:$vaddr)), VecIndex:$idx),
4405 (LD1 (SUBREG_TO_REG (i32 0), VecListOne64:$Rd, dsub),
4406 VecIndex:$idx, am_simdnoindex:$vaddr),
4409 def : Ld1Lane64Pat<extloadi8, VectorIndexB, v8i8, i32, LD1i8>;
4410 def : Ld1Lane64Pat<extloadi16, VectorIndexH, v4i16, i32, LD1i16>;
4411 def : Ld1Lane64Pat<load, VectorIndexS, v2i32, i32, LD1i32>;
4412 def : Ld1Lane64Pat<load, VectorIndexS, v2f32, f32, LD1i32>;
4415 defm LD1 : SIMDLdSt1SingleAliases<"ld1">;
4416 defm LD2 : SIMDLdSt2SingleAliases<"ld2">;
4417 defm LD3 : SIMDLdSt3SingleAliases<"ld3">;
4418 defm LD4 : SIMDLdSt4SingleAliases<"ld4">;
4421 defm ST1 : SIMDStSingleB<0, 0b000, "st1", VecListOneb, GPR64pi1>;
4422 defm ST1 : SIMDStSingleH<0, 0b010, 0, "st1", VecListOneh, GPR64pi2>;
4423 defm ST1 : SIMDStSingleS<0, 0b100, 0b00, "st1", VecListOnes, GPR64pi4>;
4424 defm ST1 : SIMDStSingleD<0, 0b100, 0b01, "st1", VecListOned, GPR64pi8>;
4426 let AddedComplexity = 8 in
4427 class St1Lane128Pat<SDPatternOperator scalar_store, Operand VecIndex,
4428 ValueType VTy, ValueType STy, Instruction ST1>
4430 (STy (vector_extract (VTy VecListOne128:$Vt), VecIndex:$idx)),
4431 am_simdnoindex:$vaddr),
4432 (ST1 VecListOne128:$Vt, VecIndex:$idx, am_simdnoindex:$vaddr)>;
4434 def : St1Lane128Pat<truncstorei8, VectorIndexB, v16i8, i32, ST1i8>;
4435 def : St1Lane128Pat<truncstorei16, VectorIndexH, v8i16, i32, ST1i16>;
4436 def : St1Lane128Pat<store, VectorIndexS, v4i32, i32, ST1i32>;
4437 def : St1Lane128Pat<store, VectorIndexS, v4f32, f32, ST1i32>;
4438 def : St1Lane128Pat<store, VectorIndexD, v2i64, i64, ST1i64>;
4439 def : St1Lane128Pat<store, VectorIndexD, v2f64, f64, ST1i64>;
4441 let AddedComplexity = 8 in
4442 class St1Lane64Pat<SDPatternOperator scalar_store, Operand VecIndex,
4443 ValueType VTy, ValueType STy, Instruction ST1>
4445 (STy (vector_extract (VTy VecListOne64:$Vt), VecIndex:$idx)),
4446 am_simdnoindex:$vaddr),
4447 (ST1 (SUBREG_TO_REG (i32 0), VecListOne64:$Vt, dsub),
4448 VecIndex:$idx, am_simdnoindex:$vaddr)>;
4450 def : St1Lane64Pat<truncstorei8, VectorIndexB, v8i8, i32, ST1i8>;
4451 def : St1Lane64Pat<truncstorei16, VectorIndexH, v4i16, i32, ST1i16>;
4452 def : St1Lane64Pat<store, VectorIndexS, v2i32, i32, ST1i32>;
4453 def : St1Lane64Pat<store, VectorIndexS, v2f32, f32, ST1i32>;
4455 multiclass St1LanePost64Pat<SDPatternOperator scalar_store, Operand VecIndex,
4456 ValueType VTy, ValueType STy, Instruction ST1,
4458 def : Pat<(scalar_store
4459 (STy (vector_extract (VTy VecListOne64:$Vt), VecIndex:$idx)),
4460 am_simdnoindex:$vaddr, offset),
4461 (ST1 (SUBREG_TO_REG (i32 0), VecListOne64:$Vt, dsub),
4462 VecIndex:$idx, am_simdnoindex:$vaddr, XZR)>;
4464 def : Pat<(scalar_store
4465 (STy (vector_extract (VTy VecListOne64:$Vt), VecIndex:$idx)),
4466 am_simdnoindex:$vaddr, GPR64:$Rm),
4467 (ST1 (SUBREG_TO_REG (i32 0), VecListOne64:$Vt, dsub),
4468 VecIndex:$idx, am_simdnoindex:$vaddr, $Rm)>;
4471 defm : St1LanePost64Pat<post_truncsti8, VectorIndexB, v8i8, i32, ST1i8_POST, 1>;
4472 defm : St1LanePost64Pat<post_truncsti16, VectorIndexH, v4i16, i32, ST1i16_POST,
4474 defm : St1LanePost64Pat<post_store, VectorIndexS, v2i32, i32, ST1i32_POST, 4>;
4475 defm : St1LanePost64Pat<post_store, VectorIndexS, v2f32, f32, ST1i32_POST, 4>;
4476 defm : St1LanePost64Pat<post_store, VectorIndexD, v1i64, i64, ST1i64_POST, 8>;
4477 defm : St1LanePost64Pat<post_store, VectorIndexD, v1f64, f64, ST1i64_POST, 8>;
4479 multiclass St1LanePost128Pat<SDPatternOperator scalar_store, Operand VecIndex,
4480 ValueType VTy, ValueType STy, Instruction ST1,
4482 def : Pat<(scalar_store
4483 (STy (vector_extract (VTy VecListOne128:$Vt), VecIndex:$idx)),
4484 am_simdnoindex:$vaddr, offset),
4485 (ST1 VecListOne128:$Vt, VecIndex:$idx, am_simdnoindex:$vaddr, XZR)>;
4487 def : Pat<(scalar_store
4488 (STy (vector_extract (VTy VecListOne128:$Vt), VecIndex:$idx)),
4489 am_simdnoindex:$vaddr, GPR64:$Rm),
4490 (ST1 VecListOne128:$Vt, VecIndex:$idx, am_simdnoindex:$vaddr, $Rm)>;
4493 defm : St1LanePost128Pat<post_truncsti8, VectorIndexB, v16i8, i32, ST1i8_POST,
4495 defm : St1LanePost128Pat<post_truncsti16, VectorIndexH, v8i16, i32, ST1i16_POST,
4497 defm : St1LanePost128Pat<post_store, VectorIndexS, v4i32, i32, ST1i32_POST, 4>;
4498 defm : St1LanePost128Pat<post_store, VectorIndexS, v4f32, f32, ST1i32_POST, 4>;
4499 defm : St1LanePost128Pat<post_store, VectorIndexD, v2i64, i64, ST1i64_POST, 8>;
4500 defm : St1LanePost128Pat<post_store, VectorIndexD, v2f64, f64, ST1i64_POST, 8>;
4502 let mayStore = 1, neverHasSideEffects = 1 in {
4503 defm ST2 : SIMDStSingleB<1, 0b000, "st2", VecListTwob, GPR64pi2>;
4504 defm ST2 : SIMDStSingleH<1, 0b010, 0, "st2", VecListTwoh, GPR64pi4>;
4505 defm ST2 : SIMDStSingleS<1, 0b100, 0b00, "st2", VecListTwos, GPR64pi8>;
4506 defm ST2 : SIMDStSingleD<1, 0b100, 0b01, "st2", VecListTwod, GPR64pi16>;
4507 defm ST3 : SIMDStSingleB<0, 0b001, "st3", VecListThreeb, GPR64pi3>;
4508 defm ST3 : SIMDStSingleH<0, 0b011, 0, "st3", VecListThreeh, GPR64pi6>;
4509 defm ST3 : SIMDStSingleS<0, 0b101, 0b00, "st3", VecListThrees, GPR64pi12>;
4510 defm ST3 : SIMDStSingleD<0, 0b101, 0b01, "st3", VecListThreed, GPR64pi24>;
4511 defm ST4 : SIMDStSingleB<1, 0b001, "st4", VecListFourb, GPR64pi4>;
4512 defm ST4 : SIMDStSingleH<1, 0b011, 0, "st4", VecListFourh, GPR64pi8>;
4513 defm ST4 : SIMDStSingleS<1, 0b101, 0b00, "st4", VecListFours, GPR64pi16>;
4514 defm ST4 : SIMDStSingleD<1, 0b101, 0b01, "st4", VecListFourd, GPR64pi32>;
4517 defm ST1 : SIMDLdSt1SingleAliases<"st1">;
4518 defm ST2 : SIMDLdSt2SingleAliases<"st2">;
4519 defm ST3 : SIMDLdSt3SingleAliases<"st3">;
4520 defm ST4 : SIMDLdSt4SingleAliases<"st4">;
4522 //----------------------------------------------------------------------------
4523 // Crypto extensions
4524 //----------------------------------------------------------------------------
4526 def AESErr : AESTiedInst<0b0100, "aese", int_arm64_crypto_aese>;
4527 def AESDrr : AESTiedInst<0b0101, "aesd", int_arm64_crypto_aesd>;
4528 def AESMCrr : AESInst< 0b0110, "aesmc", int_arm64_crypto_aesmc>;
4529 def AESIMCrr : AESInst< 0b0111, "aesimc", int_arm64_crypto_aesimc>;
4531 def SHA1Crrr : SHATiedInstQSV<0b000, "sha1c", int_arm64_crypto_sha1c>;
4532 def SHA1Prrr : SHATiedInstQSV<0b001, "sha1p", int_arm64_crypto_sha1p>;
4533 def SHA1Mrrr : SHATiedInstQSV<0b010, "sha1m", int_arm64_crypto_sha1m>;
4534 def SHA1SU0rrr : SHATiedInstVVV<0b011, "sha1su0", int_arm64_crypto_sha1su0>;
4535 def SHA256Hrrr : SHATiedInstQQV<0b100, "sha256h", int_arm64_crypto_sha256h>;
4536 def SHA256H2rrr : SHATiedInstQQV<0b101, "sha256h2",int_arm64_crypto_sha256h2>;
4537 def SHA256SU1rrr :SHATiedInstVVV<0b110, "sha256su1",int_arm64_crypto_sha256su1>;
4539 def SHA1Hrr : SHAInstSS< 0b0000, "sha1h", int_arm64_crypto_sha1h>;
4540 def SHA1SU1rr : SHATiedInstVV<0b0001, "sha1su1", int_arm64_crypto_sha1su1>;
4541 def SHA256SU0rr : SHATiedInstVV<0b0010, "sha256su0",int_arm64_crypto_sha256su0>;
4543 //----------------------------------------------------------------------------
4545 //----------------------------------------------------------------------------
4546 // FIXME: Like for X86, these should go in their own separate .td file.
4548 // Any instruction that defines a 32-bit result leaves the high half of the
4549 // register. Truncate can be lowered to EXTRACT_SUBREG. CopyFromReg may
4550 // be copying from a truncate. But any other 32-bit operation will zero-extend
4552 // FIXME: X86 also checks for CMOV here. Do we need something similar?
4553 def def32 : PatLeaf<(i32 GPR32:$src), [{
4554 return N->getOpcode() != ISD::TRUNCATE &&
4555 N->getOpcode() != TargetOpcode::EXTRACT_SUBREG &&
4556 N->getOpcode() != ISD::CopyFromReg;
4559 // In the case of a 32-bit def that is known to implicitly zero-extend,
4560 // we can use a SUBREG_TO_REG.
4561 def : Pat<(i64 (zext def32:$src)), (SUBREG_TO_REG (i64 0), GPR32:$src, sub_32)>;
4563 // For an anyext, we don't care what the high bits are, so we can perform an
4564 // INSERT_SUBREF into an IMPLICIT_DEF.
4565 def : Pat<(i64 (anyext GPR32:$src)),
4566 (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32)>;
4568 // When we need to explicitly zero-extend, we use an unsigned bitfield move
4569 // instruction (UBFM) on the enclosing super-reg.
4570 def : Pat<(i64 (zext GPR32:$src)),
4571 (UBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32), 0, 31)>;
4573 // To sign extend, we use a signed bitfield move instruction (SBFM) on the
4574 // containing super-reg.
4575 def : Pat<(i64 (sext GPR32:$src)),
4576 (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32), 0, 31)>;
4577 def : Pat<(i64 (sext_inreg GPR64:$src, i32)), (SBFMXri GPR64:$src, 0, 31)>;
4578 def : Pat<(i64 (sext_inreg GPR64:$src, i16)), (SBFMXri GPR64:$src, 0, 15)>;
4579 def : Pat<(i64 (sext_inreg GPR64:$src, i8)), (SBFMXri GPR64:$src, 0, 7)>;
4580 def : Pat<(i64 (sext_inreg GPR64:$src, i1)), (SBFMXri GPR64:$src, 0, 0)>;
4581 def : Pat<(i32 (sext_inreg GPR32:$src, i16)), (SBFMWri GPR32:$src, 0, 15)>;
4582 def : Pat<(i32 (sext_inreg GPR32:$src, i8)), (SBFMWri GPR32:$src, 0, 7)>;
4583 def : Pat<(i32 (sext_inreg GPR32:$src, i1)), (SBFMWri GPR32:$src, 0, 0)>;
4585 def : Pat<(shl (sext_inreg GPR32:$Rn, i8), (i64 imm0_31:$imm)),
4586 (SBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
4587 (i64 (i32shift_sext_i8 imm0_31:$imm)))>;
4588 def : Pat<(shl (sext_inreg GPR64:$Rn, i8), (i64 imm0_63:$imm)),
4589 (SBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
4590 (i64 (i64shift_sext_i8 imm0_63:$imm)))>;
4592 def : Pat<(shl (sext_inreg GPR32:$Rn, i16), (i64 imm0_31:$imm)),
4593 (SBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
4594 (i64 (i32shift_sext_i16 imm0_31:$imm)))>;
4595 def : Pat<(shl (sext_inreg GPR64:$Rn, i16), (i64 imm0_63:$imm)),
4596 (SBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
4597 (i64 (i64shift_sext_i16 imm0_63:$imm)))>;
4599 def : Pat<(shl (i64 (sext GPR32:$Rn)), (i64 imm0_63:$imm)),
4600 (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$Rn, sub_32),
4601 (i64 (i64shift_a imm0_63:$imm)),
4602 (i64 (i64shift_sext_i32 imm0_63:$imm)))>;
4604 // sra patterns have an AddedComplexity of 10, so make sure we have a higher
4605 // AddedComplexity for the following patterns since we want to match sext + sra
4606 // patterns before we attempt to match a single sra node.
4607 let AddedComplexity = 20 in {
4608 // We support all sext + sra combinations which preserve at least one bit of the
4609 // original value which is to be sign extended. E.g. we support shifts up to
4611 def : Pat<(sra (sext_inreg GPR32:$Rn, i8), (i64 imm0_7:$imm)),
4612 (SBFMWri GPR32:$Rn, (i64 imm0_7:$imm), 7)>;
4613 def : Pat<(sra (sext_inreg GPR64:$Rn, i8), (i64 imm0_7:$imm)),
4614 (SBFMXri GPR64:$Rn, (i64 imm0_7:$imm), 7)>;
4616 def : Pat<(sra (sext_inreg GPR32:$Rn, i16), (i64 imm0_15:$imm)),
4617 (SBFMWri GPR32:$Rn, (i64 imm0_15:$imm), 15)>;
4618 def : Pat<(sra (sext_inreg GPR64:$Rn, i16), (i64 imm0_15:$imm)),
4619 (SBFMXri GPR64:$Rn, (i64 imm0_15:$imm), 15)>;
4621 def : Pat<(sra (i64 (sext GPR32:$Rn)), (i64 imm0_31:$imm)),
4622 (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$Rn, sub_32),
4623 (i64 imm0_31:$imm), 31)>;
4624 } // AddedComplexity = 20
4626 // To truncate, we can simply extract from a subregister.
4627 def : Pat<(i32 (trunc GPR64sp:$src)),
4628 (i32 (EXTRACT_SUBREG GPR64sp:$src, sub_32))>;
4630 // __builtin_trap() uses the BRK instruction on ARM64.
4631 def : Pat<(trap), (BRK 1)>;
4633 // Conversions within AdvSIMD types in the same register size are free.
4635 def : Pat<(v1i64 (bitconvert (v2i32 FPR64:$src))), (v1i64 FPR64:$src)>;
4636 def : Pat<(v1i64 (bitconvert (v4i16 FPR64:$src))), (v1i64 FPR64:$src)>;
4637 def : Pat<(v1i64 (bitconvert (v8i8 FPR64:$src))), (v1i64 FPR64:$src)>;
4638 def : Pat<(v1i64 (bitconvert (f64 FPR64:$src))), (v1i64 FPR64:$src)>;
4639 def : Pat<(v1i64 (bitconvert (v2f32 FPR64:$src))), (v1i64 FPR64:$src)>;
4640 def : Pat<(v1i64 (bitconvert (v1f64 FPR64:$src))), (v1i64 FPR64:$src)>;
4642 def : Pat<(v2i32 (bitconvert (v1i64 FPR64:$src))), (v2i32 FPR64:$src)>;
4643 def : Pat<(v2i32 (bitconvert (v4i16 FPR64:$src))), (v2i32 FPR64:$src)>;
4644 def : Pat<(v2i32 (bitconvert (v8i8 FPR64:$src))), (v2i32 FPR64:$src)>;
4645 def : Pat<(v2i32 (bitconvert (f64 FPR64:$src))), (v2i32 FPR64:$src)>;
4646 def : Pat<(v2i32 (bitconvert (v2f32 FPR64:$src))), (v2i32 FPR64:$src)>;
4647 def : Pat<(v2i32 (bitconvert (v1f64 FPR64:$src))), (v2i32 FPR64:$src)>;
4649 def : Pat<(v4i16 (bitconvert (v1i64 FPR64:$src))), (v4i16 FPR64:$src)>;
4650 def : Pat<(v4i16 (bitconvert (v2i32 FPR64:$src))), (v4i16 FPR64:$src)>;
4651 def : Pat<(v4i16 (bitconvert (v8i8 FPR64:$src))), (v4i16 FPR64:$src)>;
4652 def : Pat<(v4i16 (bitconvert (f64 FPR64:$src))), (v4i16 FPR64:$src)>;
4653 def : Pat<(v4i16 (bitconvert (v2f32 FPR64:$src))), (v4i16 FPR64:$src)>;
4654 def : Pat<(v4i16 (bitconvert (v1f64 FPR64:$src))), (v4i16 FPR64:$src)>;
4656 def : Pat<(v8i8 (bitconvert (v1i64 FPR64:$src))), (v8i8 FPR64:$src)>;
4657 def : Pat<(v8i8 (bitconvert (v2i32 FPR64:$src))), (v8i8 FPR64:$src)>;
4658 def : Pat<(v8i8 (bitconvert (v4i16 FPR64:$src))), (v8i8 FPR64:$src)>;
4659 def : Pat<(v8i8 (bitconvert (f64 FPR64:$src))), (v8i8 FPR64:$src)>;
4660 def : Pat<(v8i8 (bitconvert (v2f32 FPR64:$src))), (v8i8 FPR64:$src)>;
4661 def : Pat<(v8i8 (bitconvert (v1f64 FPR64:$src))), (v8i8 FPR64:$src)>;
4663 def : Pat<(f64 (bitconvert (v1i64 FPR64:$src))), (f64 FPR64:$src)>;
4664 def : Pat<(f64 (bitconvert (v2i32 FPR64:$src))), (f64 FPR64:$src)>;
4665 def : Pat<(f64 (bitconvert (v4i16 FPR64:$src))), (f64 FPR64:$src)>;
4666 def : Pat<(f64 (bitconvert (v8i8 FPR64:$src))), (f64 FPR64:$src)>;
4667 def : Pat<(f64 (bitconvert (v2f32 FPR64:$src))), (f64 FPR64:$src)>;
4668 def : Pat<(f64 (bitconvert (v1f64 FPR64:$src))), (f64 FPR64:$src)>;
4670 def : Pat<(v1f64 (bitconvert (v1i64 FPR64:$src))), (v1f64 FPR64:$src)>;
4671 def : Pat<(v1f64 (bitconvert (v2i32 FPR64:$src))), (v1f64 FPR64:$src)>;
4672 def : Pat<(v1f64 (bitconvert (v4i16 FPR64:$src))), (v1f64 FPR64:$src)>;
4673 def : Pat<(v1f64 (bitconvert (v8i8 FPR64:$src))), (v1f64 FPR64:$src)>;
4674 def : Pat<(v1f64 (bitconvert (f64 FPR64:$src))), (v1f64 FPR64:$src)>;
4675 def : Pat<(v1f64 (bitconvert (v2f32 FPR64:$src))), (v1f64 FPR64:$src)>;
4677 def : Pat<(v2f32 (bitconvert (f64 FPR64:$src))), (v2f32 FPR64:$src)>;
4678 def : Pat<(v2f32 (bitconvert (v1i64 FPR64:$src))), (v2f32 FPR64:$src)>;
4679 def : Pat<(v2f32 (bitconvert (v2i32 FPR64:$src))), (v2f32 FPR64:$src)>;
4680 def : Pat<(v2f32 (bitconvert (v4i16 FPR64:$src))), (v2f32 FPR64:$src)>;
4681 def : Pat<(v2f32 (bitconvert (v8i8 FPR64:$src))), (v2f32 FPR64:$src)>;
4682 def : Pat<(v2f32 (bitconvert (v1f64 FPR64:$src))), (v2f32 FPR64:$src)>;
4685 def : Pat<(f128 (bitconvert (v2i64 FPR128:$src))), (f128 FPR128:$src)>;
4686 def : Pat<(f128 (bitconvert (v4i32 FPR128:$src))), (f128 FPR128:$src)>;
4687 def : Pat<(f128 (bitconvert (v8i16 FPR128:$src))), (f128 FPR128:$src)>;
4688 def : Pat<(f128 (bitconvert (v2f64 FPR128:$src))), (f128 FPR128:$src)>;
4689 def : Pat<(f128 (bitconvert (v4f32 FPR128:$src))), (f128 FPR128:$src)>;
4691 def : Pat<(v2f64 (bitconvert (f128 FPR128:$src))), (v2f64 FPR128:$src)>;
4692 def : Pat<(v2f64 (bitconvert (v4i32 FPR128:$src))), (v2f64 FPR128:$src)>;
4693 def : Pat<(v2f64 (bitconvert (v8i16 FPR128:$src))), (v2f64 FPR128:$src)>;
4694 def : Pat<(v2f64 (bitconvert (v16i8 FPR128:$src))), (v2f64 FPR128:$src)>;
4695 def : Pat<(v2f64 (bitconvert (v2i64 FPR128:$src))), (v2f64 FPR128:$src)>;
4696 def : Pat<(v2f64 (bitconvert (v4f32 FPR128:$src))), (v2f64 FPR128:$src)>;
4698 def : Pat<(v4f32 (bitconvert (f128 FPR128:$src))), (v4f32 FPR128:$src)>;
4699 def : Pat<(v4f32 (bitconvert (v4i32 FPR128:$src))), (v4f32 FPR128:$src)>;
4700 def : Pat<(v4f32 (bitconvert (v8i16 FPR128:$src))), (v4f32 FPR128:$src)>;
4701 def : Pat<(v4f32 (bitconvert (v16i8 FPR128:$src))), (v4f32 FPR128:$src)>;
4702 def : Pat<(v4f32 (bitconvert (v2i64 FPR128:$src))), (v4f32 FPR128:$src)>;
4703 def : Pat<(v4f32 (bitconvert (v2f64 FPR128:$src))), (v4f32 FPR128:$src)>;
4705 def : Pat<(v2i64 (bitconvert (f128 FPR128:$src))), (v2i64 FPR128:$src)>;
4706 def : Pat<(v2i64 (bitconvert (v4i32 FPR128:$src))), (v2i64 FPR128:$src)>;
4707 def : Pat<(v2i64 (bitconvert (v8i16 FPR128:$src))), (v2i64 FPR128:$src)>;
4708 def : Pat<(v2i64 (bitconvert (v16i8 FPR128:$src))), (v2i64 FPR128:$src)>;
4709 def : Pat<(v2i64 (bitconvert (v2f64 FPR128:$src))), (v2i64 FPR128:$src)>;
4710 def : Pat<(v2i64 (bitconvert (v4f32 FPR128:$src))), (v2i64 FPR128:$src)>;
4712 def : Pat<(v4i32 (bitconvert (f128 FPR128:$src))), (v4i32 FPR128:$src)>;
4713 def : Pat<(v4i32 (bitconvert (v2i64 FPR128:$src))), (v4i32 FPR128:$src)>;
4714 def : Pat<(v4i32 (bitconvert (v8i16 FPR128:$src))), (v4i32 FPR128:$src)>;
4715 def : Pat<(v4i32 (bitconvert (v16i8 FPR128:$src))), (v4i32 FPR128:$src)>;
4716 def : Pat<(v4i32 (bitconvert (v2f64 FPR128:$src))), (v4i32 FPR128:$src)>;
4717 def : Pat<(v4i32 (bitconvert (v4f32 FPR128:$src))), (v4i32 FPR128:$src)>;
4719 def : Pat<(v8i16 (bitconvert (f128 FPR128:$src))), (v8i16 FPR128:$src)>;
4720 def : Pat<(v8i16 (bitconvert (v2i64 FPR128:$src))), (v8i16 FPR128:$src)>;
4721 def : Pat<(v8i16 (bitconvert (v4i32 FPR128:$src))), (v8i16 FPR128:$src)>;
4722 def : Pat<(v8i16 (bitconvert (v16i8 FPR128:$src))), (v8i16 FPR128:$src)>;
4723 def : Pat<(v8i16 (bitconvert (v2f64 FPR128:$src))), (v8i16 FPR128:$src)>;
4724 def : Pat<(v8i16 (bitconvert (v4f32 FPR128:$src))), (v8i16 FPR128:$src)>;
4726 def : Pat<(v16i8 (bitconvert (f128 FPR128:$src))), (v16i8 FPR128:$src)>;
4727 def : Pat<(v16i8 (bitconvert (v2i64 FPR128:$src))), (v16i8 FPR128:$src)>;
4728 def : Pat<(v16i8 (bitconvert (v4i32 FPR128:$src))), (v16i8 FPR128:$src)>;
4729 def : Pat<(v16i8 (bitconvert (v8i16 FPR128:$src))), (v16i8 FPR128:$src)>;
4730 def : Pat<(v16i8 (bitconvert (v2f64 FPR128:$src))), (v16i8 FPR128:$src)>;
4731 def : Pat<(v16i8 (bitconvert (v4f32 FPR128:$src))), (v16i8 FPR128:$src)>;
4733 def : Pat<(v8i8 (extract_subvector (v16i8 FPR128:$Rn), (i64 1))),
4734 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
4735 def : Pat<(v4i16 (extract_subvector (v8i16 FPR128:$Rn), (i64 1))),
4736 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
4737 def : Pat<(v2i32 (extract_subvector (v4i32 FPR128:$Rn), (i64 1))),
4738 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
4739 def : Pat<(v1i64 (extract_subvector (v2i64 FPR128:$Rn), (i64 1))),
4740 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
4742 // A 64-bit subvector insert to the first 128-bit vector position
4743 // is a subregister copy that needs no instruction.
4744 def : Pat<(insert_subvector undef, (v1i64 FPR64:$src), (i32 0)),
4745 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
4746 def : Pat<(insert_subvector undef, (v1f64 FPR64:$src), (i32 0)),
4747 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
4748 def : Pat<(insert_subvector undef, (v2i32 FPR64:$src), (i32 0)),
4749 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
4750 def : Pat<(insert_subvector undef, (v2f32 FPR64:$src), (i32 0)),
4751 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
4752 def : Pat<(insert_subvector undef, (v4i16 FPR64:$src), (i32 0)),
4753 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
4754 def : Pat<(insert_subvector undef, (v8i8 FPR64:$src), (i32 0)),
4755 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
4757 // Use pair-wise add instructions when summing up the lanes for v2f64, v2i64
4759 def : Pat<(i64 (add (vector_extract (v2i64 FPR128:$Rn), (i64 0)),
4760 (vector_extract (v2i64 FPR128:$Rn), (i64 1)))),
4761 (i64 (ADDPv2i64p (v2i64 FPR128:$Rn)))>;
4762 def : Pat<(f64 (fadd (vector_extract (v2f64 FPR128:$Rn), (i64 0)),
4763 (vector_extract (v2f64 FPR128:$Rn), (i64 1)))),
4764 (f64 (FADDPv2i64p (v2f64 FPR128:$Rn)))>;
4765 // vector_extract on 64-bit vectors gets promoted to a 128 bit vector,
4766 // so we match on v4f32 here, not v2f32. This will also catch adding
4767 // the low two lanes of a true v4f32 vector.
4768 def : Pat<(fadd (vector_extract (v4f32 FPR128:$Rn), (i64 0)),
4769 (vector_extract (v4f32 FPR128:$Rn), (i64 1))),
4770 (f32 (FADDPv2i32p (EXTRACT_SUBREG FPR128:$Rn, dsub)))>;
4772 // Scalar 64-bit shifts in FPR64 registers.
4773 def : Pat<(i64 (int_arm64_neon_sshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
4774 (SSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
4775 def : Pat<(i64 (int_arm64_neon_ushl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
4776 (USHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
4777 def : Pat<(i64 (int_arm64_neon_srshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
4778 (SRSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
4779 def : Pat<(i64 (int_arm64_neon_urshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
4780 (URSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
4782 // Tail call return handling. These are all compiler pseudo-instructions,
4783 // so no encoding information or anything like that.
4784 let isCall = 1, isTerminator = 1, isReturn = 1, isBarrier = 1, Uses = [SP] in {
4785 def TCRETURNdi : Pseudo<(outs), (ins i64imm:$dst), []>;
4786 def TCRETURNri : Pseudo<(outs), (ins tcGPR64:$dst), []>;
4789 def : Pat<(ARM64tcret tcGPR64:$dst), (TCRETURNri tcGPR64:$dst)>;
4790 def : Pat<(ARM64tcret (i64 tglobaladdr:$dst)), (TCRETURNdi texternalsym:$dst)>;
4791 def : Pat<(ARM64tcret (i64 texternalsym:$dst)), (TCRETURNdi texternalsym:$dst)>;
4793 include "ARM64InstrAtomics.td"