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">;
24 //===----------------------------------------------------------------------===//
25 // ARM64-specific DAG Nodes.
28 // SDTBinaryArithWithFlagsOut - RES1, FLAGS = op LHS, RHS
29 def SDTBinaryArithWithFlagsOut : SDTypeProfile<2, 2,
32 SDTCisInt<0>, SDTCisVT<1, i32>]>;
34 // SDTBinaryArithWithFlagsIn - RES1, FLAGS = op LHS, RHS, FLAGS
35 def SDTBinaryArithWithFlagsIn : SDTypeProfile<1, 3,
41 // SDTBinaryArithWithFlagsInOut - RES1, FLAGS = op LHS, RHS, FLAGS
42 def SDTBinaryArithWithFlagsInOut : SDTypeProfile<2, 3,
49 def SDT_ARM64Brcond : SDTypeProfile<0, 3,
50 [SDTCisVT<0, OtherVT>, SDTCisVT<1, i32>,
52 def SDT_ARM64cbz : SDTypeProfile<0, 2, [SDTCisInt<0>, SDTCisVT<1, OtherVT>]>;
53 def SDT_ARM64tbz : SDTypeProfile<0, 3, [SDTCisVT<0, i64>, SDTCisVT<1, i64>,
54 SDTCisVT<2, OtherVT>]>;
57 def SDT_ARM64CSel : SDTypeProfile<1, 4,
62 def SDT_ARM64FCmp : SDTypeProfile<0, 2,
65 def SDT_ARM64Dup : SDTypeProfile<1, 1, [SDTCisVec<0>]>;
66 def SDT_ARM64DupLane : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisInt<2>]>;
67 def SDT_ARM64Zip : SDTypeProfile<1, 2, [SDTCisVec<0>,
70 def SDT_ARM64MOVIedit : SDTypeProfile<1, 1, [SDTCisInt<1>]>;
71 def SDT_ARM64MOVIshift : SDTypeProfile<1, 2, [SDTCisInt<1>, SDTCisInt<2>]>;
72 def SDT_ARM64vecimm : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0,1>,
73 SDTCisInt<2>, SDTCisInt<3>]>;
74 def SDT_ARM64UnaryVec: SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0,1>]>;
75 def SDT_ARM64ExtVec: SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0,1>,
76 SDTCisSameAs<0,2>, SDTCisInt<3>]>;
77 def SDT_ARM64vshift : SDTypeProfile<1, 2, [SDTCisSameAs<0,1>, SDTCisInt<2>]>;
79 def SDT_ARM64unvec : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0,1>]>;
80 def SDT_ARM64fcmpz : SDTypeProfile<1, 1, []>;
81 def SDT_ARM64fcmp : SDTypeProfile<1, 2, [SDTCisSameAs<1,2>]>;
82 def SDT_ARM64binvec : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisSameAs<0,1>,
84 def SDT_ARM64trivec : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0,1>,
87 def SDT_ARM64TCRET : SDTypeProfile<0, 1, [SDTCisPtrTy<0>]>;
88 def SDT_ARM64PREFETCH : SDTypeProfile<0, 2, [SDTCisVT<0, i32>, SDTCisPtrTy<1>]>;
90 def SDT_ARM64ITOF : SDTypeProfile<1, 1, [SDTCisFP<0>, SDTCisSameAs<0,1>]>;
92 def SDT_ARM64TLSDescCall : SDTypeProfile<0, -2, [SDTCisPtrTy<0>,
94 def SDT_ARM64WrapperLarge : SDTypeProfile<1, 4,
95 [SDTCisVT<0, i64>, SDTCisVT<1, i32>,
96 SDTCisSameAs<1, 2>, SDTCisSameAs<1, 3>,
101 def ARM64adrp : SDNode<"ARM64ISD::ADRP", SDTIntUnaryOp, []>;
102 def ARM64addlow : SDNode<"ARM64ISD::ADDlow", SDTIntBinOp, []>;
103 def ARM64LOADgot : SDNode<"ARM64ISD::LOADgot", SDTIntUnaryOp>;
104 def ARM64callseq_start : SDNode<"ISD::CALLSEQ_START",
105 SDCallSeqStart<[ SDTCisVT<0, i32> ]>,
106 [SDNPHasChain, SDNPOutGlue]>;
107 def ARM64callseq_end : SDNode<"ISD::CALLSEQ_END",
108 SDCallSeqEnd<[ SDTCisVT<0, i32>,
110 [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue]>;
111 def ARM64call : SDNode<"ARM64ISD::CALL",
112 SDTypeProfile<0, -1, [SDTCisPtrTy<0>]>,
113 [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue,
115 def ARM64brcond : SDNode<"ARM64ISD::BRCOND", SDT_ARM64Brcond,
117 def ARM64cbz : SDNode<"ARM64ISD::CBZ", SDT_ARM64cbz,
119 def ARM64cbnz : SDNode<"ARM64ISD::CBNZ", SDT_ARM64cbz,
121 def ARM64tbz : SDNode<"ARM64ISD::TBZ", SDT_ARM64tbz,
123 def ARM64tbnz : SDNode<"ARM64ISD::TBNZ", SDT_ARM64tbz,
127 def ARM64csel : SDNode<"ARM64ISD::CSEL", SDT_ARM64CSel>;
128 def ARM64csinv : SDNode<"ARM64ISD::CSINV", SDT_ARM64CSel>;
129 def ARM64csneg : SDNode<"ARM64ISD::CSNEG", SDT_ARM64CSel>;
130 def ARM64csinc : SDNode<"ARM64ISD::CSINC", SDT_ARM64CSel>;
131 def ARM64retflag : SDNode<"ARM64ISD::RET_FLAG", SDTNone,
132 [SDNPHasChain, SDNPOptInGlue, SDNPVariadic]>;
133 def ARM64adc : SDNode<"ARM64ISD::ADC", SDTBinaryArithWithFlagsIn >;
134 def ARM64sbc : SDNode<"ARM64ISD::SBC", SDTBinaryArithWithFlagsIn>;
135 def ARM64add_flag : SDNode<"ARM64ISD::ADDS", SDTBinaryArithWithFlagsOut,
137 def ARM64sub_flag : SDNode<"ARM64ISD::SUBS", SDTBinaryArithWithFlagsOut>;
138 def ARM64and_flag : SDNode<"ARM64ISD::ANDS", SDTBinaryArithWithFlagsOut,
140 def ARM64adc_flag : SDNode<"ARM64ISD::ADCS", SDTBinaryArithWithFlagsInOut>;
141 def ARM64sbc_flag : SDNode<"ARM64ISD::SBCS", SDTBinaryArithWithFlagsInOut>;
143 def ARM64threadpointer : SDNode<"ARM64ISD::THREAD_POINTER", SDTPtrLeaf>;
145 def ARM64fcmp : SDNode<"ARM64ISD::FCMP", SDT_ARM64FCmp>;
147 def ARM64fmax : SDNode<"ARM64ISD::FMAX", SDTFPBinOp>;
148 def ARM64fmin : SDNode<"ARM64ISD::FMIN", SDTFPBinOp>;
150 def ARM64dup : SDNode<"ARM64ISD::DUP", SDT_ARM64Dup>;
151 def ARM64duplane8 : SDNode<"ARM64ISD::DUPLANE8", SDT_ARM64DupLane>;
152 def ARM64duplane16 : SDNode<"ARM64ISD::DUPLANE16", SDT_ARM64DupLane>;
153 def ARM64duplane32 : SDNode<"ARM64ISD::DUPLANE32", SDT_ARM64DupLane>;
154 def ARM64duplane64 : SDNode<"ARM64ISD::DUPLANE64", SDT_ARM64DupLane>;
156 def ARM64zip1 : SDNode<"ARM64ISD::ZIP1", SDT_ARM64Zip>;
157 def ARM64zip2 : SDNode<"ARM64ISD::ZIP2", SDT_ARM64Zip>;
158 def ARM64uzp1 : SDNode<"ARM64ISD::UZP1", SDT_ARM64Zip>;
159 def ARM64uzp2 : SDNode<"ARM64ISD::UZP2", SDT_ARM64Zip>;
160 def ARM64trn1 : SDNode<"ARM64ISD::TRN1", SDT_ARM64Zip>;
161 def ARM64trn2 : SDNode<"ARM64ISD::TRN2", SDT_ARM64Zip>;
163 def ARM64movi_edit : SDNode<"ARM64ISD::MOVIedit", SDT_ARM64MOVIedit>;
164 def ARM64movi_shift : SDNode<"ARM64ISD::MOVIshift", SDT_ARM64MOVIshift>;
165 def ARM64movi_msl : SDNode<"ARM64ISD::MOVImsl", SDT_ARM64MOVIshift>;
166 def ARM64mvni_shift : SDNode<"ARM64ISD::MVNIshift", SDT_ARM64MOVIshift>;
167 def ARM64mvni_msl : SDNode<"ARM64ISD::MVNImsl", SDT_ARM64MOVIshift>;
168 def ARM64movi : SDNode<"ARM64ISD::MOVI", SDT_ARM64MOVIedit>;
169 def ARM64fmov : SDNode<"ARM64ISD::FMOV", SDT_ARM64MOVIedit>;
171 def ARM64rev16 : SDNode<"ARM64ISD::REV16", SDT_ARM64UnaryVec>;
172 def ARM64rev32 : SDNode<"ARM64ISD::REV32", SDT_ARM64UnaryVec>;
173 def ARM64rev64 : SDNode<"ARM64ISD::REV64", SDT_ARM64UnaryVec>;
174 def ARM64ext : SDNode<"ARM64ISD::EXT", SDT_ARM64ExtVec>;
176 def ARM64vashr : SDNode<"ARM64ISD::VASHR", SDT_ARM64vshift>;
177 def ARM64vlshr : SDNode<"ARM64ISD::VLSHR", SDT_ARM64vshift>;
178 def ARM64vshl : SDNode<"ARM64ISD::VSHL", SDT_ARM64vshift>;
179 def ARM64sqshli : SDNode<"ARM64ISD::SQSHL_I", SDT_ARM64vshift>;
180 def ARM64uqshli : SDNode<"ARM64ISD::UQSHL_I", SDT_ARM64vshift>;
181 def ARM64sqshlui : SDNode<"ARM64ISD::SQSHLU_I", SDT_ARM64vshift>;
182 def ARM64srshri : SDNode<"ARM64ISD::SRSHR_I", SDT_ARM64vshift>;
183 def ARM64urshri : SDNode<"ARM64ISD::URSHR_I", SDT_ARM64vshift>;
185 def ARM64not: SDNode<"ARM64ISD::NOT", SDT_ARM64unvec>;
186 def ARM64bit: SDNode<"ARM64ISD::BIT", SDT_ARM64trivec>;
187 def ARM64bsl: SDNode<"ARM64ISD::BSL", SDT_ARM64trivec>;
189 def ARM64cmeq: SDNode<"ARM64ISD::CMEQ", SDT_ARM64binvec>;
190 def ARM64cmge: SDNode<"ARM64ISD::CMGE", SDT_ARM64binvec>;
191 def ARM64cmgt: SDNode<"ARM64ISD::CMGT", SDT_ARM64binvec>;
192 def ARM64cmhi: SDNode<"ARM64ISD::CMHI", SDT_ARM64binvec>;
193 def ARM64cmhs: SDNode<"ARM64ISD::CMHS", SDT_ARM64binvec>;
195 def ARM64fcmeq: SDNode<"ARM64ISD::FCMEQ", SDT_ARM64fcmp>;
196 def ARM64fcmge: SDNode<"ARM64ISD::FCMGE", SDT_ARM64fcmp>;
197 def ARM64fcmgt: SDNode<"ARM64ISD::FCMGT", SDT_ARM64fcmp>;
199 def ARM64cmeqz: SDNode<"ARM64ISD::CMEQz", SDT_ARM64unvec>;
200 def ARM64cmgez: SDNode<"ARM64ISD::CMGEz", SDT_ARM64unvec>;
201 def ARM64cmgtz: SDNode<"ARM64ISD::CMGTz", SDT_ARM64unvec>;
202 def ARM64cmlez: SDNode<"ARM64ISD::CMLEz", SDT_ARM64unvec>;
203 def ARM64cmltz: SDNode<"ARM64ISD::CMLTz", SDT_ARM64unvec>;
204 def ARM64cmtst : PatFrag<(ops node:$LHS, node:$RHS),
205 (ARM64not (ARM64cmeqz (and node:$LHS, node:$RHS)))>;
207 def ARM64fcmeqz: SDNode<"ARM64ISD::FCMEQz", SDT_ARM64fcmpz>;
208 def ARM64fcmgez: SDNode<"ARM64ISD::FCMGEz", SDT_ARM64fcmpz>;
209 def ARM64fcmgtz: SDNode<"ARM64ISD::FCMGTz", SDT_ARM64fcmpz>;
210 def ARM64fcmlez: SDNode<"ARM64ISD::FCMLEz", SDT_ARM64fcmpz>;
211 def ARM64fcmltz: SDNode<"ARM64ISD::FCMLTz", SDT_ARM64fcmpz>;
213 def ARM64bici: SDNode<"ARM64ISD::BICi", SDT_ARM64vecimm>;
214 def ARM64orri: SDNode<"ARM64ISD::ORRi", SDT_ARM64vecimm>;
216 def ARM64neg : SDNode<"ARM64ISD::NEG", SDT_ARM64unvec>;
218 def ARM64tcret: SDNode<"ARM64ISD::TC_RETURN", SDT_ARM64TCRET,
219 [SDNPHasChain, SDNPOptInGlue, SDNPVariadic]>;
221 def ARM64Prefetch : SDNode<"ARM64ISD::PREFETCH", SDT_ARM64PREFETCH,
222 [SDNPHasChain, SDNPSideEffect]>;
224 def ARM64sitof: SDNode<"ARM64ISD::SITOF", SDT_ARM64ITOF>;
225 def ARM64uitof: SDNode<"ARM64ISD::UITOF", SDT_ARM64ITOF>;
227 def ARM64tlsdesc_call : SDNode<"ARM64ISD::TLSDESC_CALL", SDT_ARM64TLSDescCall,
228 [SDNPInGlue, SDNPOutGlue, SDNPHasChain,
231 def ARM64WrapperLarge : SDNode<"ARM64ISD::WrapperLarge", SDT_ARM64WrapperLarge>;
234 //===----------------------------------------------------------------------===//
236 //===----------------------------------------------------------------------===//
238 // ARM64 Instruction Predicate Definitions.
240 def HasZCZ : Predicate<"Subtarget->hasZeroCycleZeroing()">;
241 def NoZCZ : Predicate<"!Subtarget->hasZeroCycleZeroing()">;
242 def IsDarwin : Predicate<"Subtarget->isTargetDarwin()">;
243 def IsNotDarwin: Predicate<"!Subtarget->isTargetDarwin()">;
244 def ForCodeSize : Predicate<"ForCodeSize">;
245 def NotForCodeSize : Predicate<"!ForCodeSize">;
247 include "ARM64InstrFormats.td"
249 //===----------------------------------------------------------------------===//
251 //===----------------------------------------------------------------------===//
252 // Miscellaneous instructions.
253 //===----------------------------------------------------------------------===//
255 let Defs = [SP], Uses = [SP], hasSideEffects = 1, isCodeGenOnly = 1 in {
256 def ADJCALLSTACKDOWN : Pseudo<(outs), (ins i32imm:$amt),
257 [(ARM64callseq_start timm:$amt)]>;
258 def ADJCALLSTACKUP : Pseudo<(outs), (ins i32imm:$amt1, i32imm:$amt2),
259 [(ARM64callseq_end timm:$amt1, timm:$amt2)]>;
260 } // Defs = [SP], Uses = [SP], hasSideEffects = 1, isCodeGenOnly = 1
262 let isReMaterializable = 1, isCodeGenOnly = 1 in {
263 // FIXME: The following pseudo instructions are only needed because remat
264 // cannot handle multiple instructions. When that changes, they can be
265 // removed, along with the ARM64Wrapper node.
267 let AddedComplexity = 10 in
268 def LOADgot : Pseudo<(outs GPR64:$dst), (ins i64imm:$addr),
269 [(set GPR64:$dst, (ARM64LOADgot tglobaladdr:$addr))]>,
272 // The MOVaddr instruction should match only when the add is not folded
273 // into a load or store address.
275 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
276 [(set GPR64:$dst, (ARM64addlow (ARM64adrp tglobaladdr:$hi),
277 tglobaladdr:$low))]>,
278 Sched<[WriteAdrAdr]>;
280 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
281 [(set GPR64:$dst, (ARM64addlow (ARM64adrp tjumptable:$hi),
283 Sched<[WriteAdrAdr]>;
285 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
286 [(set GPR64:$dst, (ARM64addlow (ARM64adrp tconstpool:$hi),
288 Sched<[WriteAdrAdr]>;
290 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
291 [(set GPR64:$dst, (ARM64addlow (ARM64adrp tblockaddress:$hi),
292 tblockaddress:$low))]>,
293 Sched<[WriteAdrAdr]>;
295 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
296 [(set GPR64:$dst, (ARM64addlow (ARM64adrp tglobaltlsaddr:$hi),
297 tglobaltlsaddr:$low))]>,
298 Sched<[WriteAdrAdr]>;
300 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
301 [(set GPR64:$dst, (ARM64addlow (ARM64adrp texternalsym:$hi),
302 texternalsym:$low))]>,
303 Sched<[WriteAdrAdr]>;
305 } // isReMaterializable, isCodeGenOnly
307 def : Pat<(ARM64LOADgot tglobaltlsaddr:$addr),
308 (LOADgot tglobaltlsaddr:$addr)>;
310 def : Pat<(ARM64LOADgot texternalsym:$addr),
311 (LOADgot texternalsym:$addr)>;
313 def : Pat<(ARM64LOADgot tconstpool:$addr),
314 (LOADgot tconstpool:$addr)>;
316 //===----------------------------------------------------------------------===//
317 // System instructions.
318 //===----------------------------------------------------------------------===//
320 def HINT : HintI<"hint">;
321 def : InstAlias<"nop", (HINT 0b000)>;
322 def : InstAlias<"yield",(HINT 0b001)>;
323 def : InstAlias<"wfe", (HINT 0b010)>;
324 def : InstAlias<"wfi", (HINT 0b011)>;
325 def : InstAlias<"sev", (HINT 0b100)>;
326 def : InstAlias<"sevl", (HINT 0b101)>;
328 // As far as LLVM is concerned this writes to the system's exclusive monitors.
329 let mayLoad = 1, mayStore = 1 in
330 def CLREX : CRmSystemI<imm0_15, 0b010, "clrex">;
332 def DMB : CRmSystemI<barrier_op, 0b101, "dmb">;
333 def DSB : CRmSystemI<barrier_op, 0b100, "dsb">;
334 def ISB : CRmSystemI<barrier_op, 0b110, "isb">;
335 def : InstAlias<"clrex", (CLREX 0xf)>;
336 def : InstAlias<"isb", (ISB 0xf)>;
340 def MSRcpsr: MSRcpsrI;
342 // The thread pointer (on Linux, at least, where this has been implemented) is
344 def : Pat<(ARM64threadpointer), (MRS 0xde82)>;
346 // Generic system instructions
347 def SYSxt : SystemXtI<0, "sys">;
348 def SYSLxt : SystemLXtI<1, "sysl">;
350 def : InstAlias<"sys $op1, $Cn, $Cm, $op2",
351 (SYSxt imm0_7:$op1, sys_cr_op:$Cn,
352 sys_cr_op:$Cm, imm0_7:$op2, XZR)>;
354 //===----------------------------------------------------------------------===//
355 // Move immediate instructions.
356 //===----------------------------------------------------------------------===//
358 defm MOVK : InsertImmediate<0b11, "movk">;
359 defm MOVN : MoveImmediate<0b00, "movn">;
361 let PostEncoderMethod = "fixMOVZ" in
362 defm MOVZ : MoveImmediate<0b10, "movz">;
364 def : InstAlias<"movk $dst, $imm", (MOVKWi GPR32:$dst, imm0_65535:$imm, 0)>;
365 def : InstAlias<"movk $dst, $imm", (MOVKXi GPR64:$dst, imm0_65535:$imm, 0)>;
366 def : InstAlias<"movn $dst, $imm", (MOVNWi GPR32:$dst, imm0_65535:$imm, 0)>;
367 def : InstAlias<"movn $dst, $imm", (MOVNXi GPR64:$dst, imm0_65535:$imm, 0)>;
368 def : InstAlias<"movz $dst, $imm", (MOVZWi GPR32:$dst, imm0_65535:$imm, 0)>;
369 def : InstAlias<"movz $dst, $imm", (MOVZXi GPR64:$dst, imm0_65535:$imm, 0)>;
371 def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movz_symbol_g3:$sym, 48)>;
372 def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movz_symbol_g2:$sym, 32)>;
373 def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movz_symbol_g1:$sym, 16)>;
374 def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movz_symbol_g0:$sym, 0)>;
376 def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movz_symbol_g3:$sym, 48)>;
377 def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movz_symbol_g2:$sym, 32)>;
378 def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movz_symbol_g1:$sym, 16)>;
379 def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movz_symbol_g0:$sym, 0)>;
381 def : InstAlias<"movz $Rd, $sym", (MOVZWi GPR32:$Rd, movz_symbol_g3:$sym, 48)>;
382 def : InstAlias<"movz $Rd, $sym", (MOVZWi GPR32:$Rd, movz_symbol_g2:$sym, 32)>;
383 def : InstAlias<"movz $Rd, $sym", (MOVZWi GPR32:$Rd, movz_symbol_g1:$sym, 16)>;
384 def : InstAlias<"movz $Rd, $sym", (MOVZWi GPR32:$Rd, movz_symbol_g0:$sym, 0)>;
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<"movk $Rd, $sym", (MOVKWi GPR32:$Rd, movk_symbol_g2:$sym, 32)>;
391 def : InstAlias<"movk $Rd, $sym", (MOVKWi GPR32:$Rd, movk_symbol_g1:$sym, 16)>;
392 def : InstAlias<"movk $Rd, $sym", (MOVKWi GPR32:$Rd, movk_symbol_g0:$sym, 0)>;
394 let isReMaterializable = 1, isCodeGenOnly = 1, isMoveImm = 1,
395 isAsCheapAsAMove = 1 in {
396 // FIXME: The following pseudo instructions are only needed because remat
397 // cannot handle multiple instructions. When that changes, we can select
398 // directly to the real instructions and get rid of these pseudos.
401 : Pseudo<(outs GPR32:$dst), (ins i32imm:$src),
402 [(set GPR32:$dst, imm:$src)]>,
405 : Pseudo<(outs GPR64:$dst), (ins i64imm:$src),
406 [(set GPR64:$dst, imm:$src)]>,
408 } // isReMaterializable, isCodeGenOnly
410 // If possible, we want to use MOVi32imm even for 64-bit moves. This gives the
411 // eventual expansion code fewer bits to worry about getting right. Marshalling
412 // the types is a little tricky though:
413 def i64imm_32bit : ImmLeaf<i64, [{
414 return (Imm & 0xffffffffULL) == static_cast<uint64_t>(Imm);
417 def trunc_imm : SDNodeXForm<imm, [{
418 return CurDAG->getTargetConstant(N->getZExtValue(), MVT::i32);
421 def : Pat<(i64 i64imm_32bit:$src),
422 (SUBREG_TO_REG (i64 0), (MOVi32imm (trunc_imm imm:$src)), sub_32)>;
424 // Deal with the various forms of (ELF) large addressing with MOVZ/MOVK
426 def : Pat<(ARM64WrapperLarge tglobaladdr:$g3, tglobaladdr:$g2,
427 tglobaladdr:$g1, tglobaladdr:$g0),
428 (MOVKXi (MOVKXi (MOVKXi (MOVZXi tglobaladdr:$g3, 48),
429 tglobaladdr:$g2, 32),
430 tglobaladdr:$g1, 16),
431 tglobaladdr:$g0, 0)>;
433 def : Pat<(ARM64WrapperLarge tblockaddress:$g3, tblockaddress:$g2,
434 tblockaddress:$g1, tblockaddress:$g0),
435 (MOVKXi (MOVKXi (MOVKXi (MOVZXi tblockaddress:$g3, 48),
436 tblockaddress:$g2, 32),
437 tblockaddress:$g1, 16),
438 tblockaddress:$g0, 0)>;
440 def : Pat<(ARM64WrapperLarge tconstpool:$g3, tconstpool:$g2,
441 tconstpool:$g1, tconstpool:$g0),
442 (MOVKXi (MOVKXi (MOVKXi (MOVZXi tconstpool:$g3, 48),
447 def : Pat<(ARM64WrapperLarge tjumptable:$g3, tjumptable:$g2,
448 tjumptable:$g1, tjumptable:$g0),
449 (MOVKXi (MOVKXi (MOVKXi (MOVZXi tjumptable:$g3, 48),
455 //===----------------------------------------------------------------------===//
456 // Arithmetic instructions.
457 //===----------------------------------------------------------------------===//
459 // Add/subtract with carry.
460 defm ADC : AddSubCarry<0, "adc", "adcs", ARM64adc, ARM64adc_flag>;
461 defm SBC : AddSubCarry<1, "sbc", "sbcs", ARM64sbc, ARM64sbc_flag>;
463 def : InstAlias<"ngc $dst, $src", (SBCWr GPR32:$dst, WZR, GPR32:$src)>;
464 def : InstAlias<"ngc $dst, $src", (SBCXr GPR64:$dst, XZR, GPR64:$src)>;
465 def : InstAlias<"ngcs $dst, $src", (SBCSWr GPR32:$dst, WZR, GPR32:$src)>;
466 def : InstAlias<"ngcs $dst, $src", (SBCSXr GPR64:$dst, XZR, GPR64:$src)>;
469 defm ADD : AddSub<0, "add", add>;
470 defm SUB : AddSub<1, "sub">;
472 defm ADDS : AddSubS<0, "adds", ARM64add_flag>;
473 defm SUBS : AddSubS<1, "subs", ARM64sub_flag>;
475 // Use SUBS instead of SUB to enable CSE between SUBS and SUB.
476 def : Pat<(sub GPR32sp:$Rn, addsub_shifted_imm32:$imm),
477 (SUBSWri GPR32sp:$Rn, addsub_shifted_imm32:$imm)>;
478 def : Pat<(sub GPR64sp:$Rn, addsub_shifted_imm64:$imm),
479 (SUBSXri GPR64sp:$Rn, addsub_shifted_imm64:$imm)>;
480 def : Pat<(sub GPR32:$Rn, GPR32:$Rm),
481 (SUBSWrr GPR32:$Rn, GPR32:$Rm)>;
482 def : Pat<(sub GPR64:$Rn, GPR64:$Rm),
483 (SUBSXrr GPR64:$Rn, GPR64:$Rm)>;
484 def : Pat<(sub GPR32:$Rn, arith_shifted_reg32:$Rm),
485 (SUBSWrs GPR32:$Rn, arith_shifted_reg32:$Rm)>;
486 def : Pat<(sub GPR64:$Rn, arith_shifted_reg64:$Rm),
487 (SUBSXrs GPR64:$Rn, arith_shifted_reg64:$Rm)>;
488 def : Pat<(sub GPR32sp:$R2, arith_extended_reg32<i32>:$R3),
489 (SUBSWrx GPR32sp:$R2, arith_extended_reg32<i32>:$R3)>;
490 def : Pat<(sub GPR64sp:$R2, arith_extended_reg32to64<i64>:$R3),
491 (SUBSXrx GPR64sp:$R2, arith_extended_reg32to64<i64>:$R3)>;
493 // Because of the immediate format for add/sub-imm instructions, the
494 // expression (add x, -1) must be transformed to (SUB{W,X}ri x, 1).
495 // These patterns capture that transformation.
496 let AddedComplexity = 1 in {
497 def : Pat<(add GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
498 (SUBSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
499 def : Pat<(add GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
500 (SUBSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
501 def : Pat<(sub GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
502 (ADDWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
503 def : Pat<(sub GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
504 (ADDXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
507 def : InstAlias<"neg $dst, $src", (SUBWrs GPR32:$dst, WZR, GPR32:$src, 0)>;
508 def : InstAlias<"neg $dst, $src", (SUBXrs GPR64:$dst, XZR, GPR64:$src, 0)>;
509 def : InstAlias<"neg $dst, $src, $shift",
510 (SUBWrs GPR32:$dst, WZR, GPR32:$src, arith_shift:$shift)>;
511 def : InstAlias<"neg $dst, $src, $shift",
512 (SUBXrs GPR64:$dst, XZR, GPR64:$src, arith_shift:$shift)>;
514 // Because of the immediate format for add/sub-imm instructions, the
515 // expression (add x, -1) must be transformed to (SUB{W,X}ri x, 1).
516 // These patterns capture that transformation.
517 let AddedComplexity = 1 in {
518 def : Pat<(ARM64add_flag GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
519 (SUBSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
520 def : Pat<(ARM64add_flag GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
521 (SUBSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
522 def : Pat<(ARM64sub_flag GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
523 (ADDSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
524 def : Pat<(ARM64sub_flag GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
525 (ADDSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
528 def : InstAlias<"negs $dst, $src", (SUBSWrs GPR32:$dst, WZR, GPR32:$src, 0)>;
529 def : InstAlias<"negs $dst, $src", (SUBSXrs GPR64:$dst, XZR, GPR64:$src, 0)>;
530 def : InstAlias<"negs $dst, $src, $shift",
531 (SUBSWrs GPR32:$dst, WZR, GPR32:$src, arith_shift:$shift)>;
532 def : InstAlias<"negs $dst, $src, $shift",
533 (SUBSXrs GPR64:$dst, XZR, GPR64:$src, arith_shift:$shift)>;
535 // Unsigned/Signed divide
536 defm UDIV : Div<0, "udiv", udiv>;
537 defm SDIV : Div<1, "sdiv", sdiv>;
538 let isCodeGenOnly = 1 in {
539 defm UDIV_Int : Div<0, "udiv", int_arm64_udiv>;
540 defm SDIV_Int : Div<1, "sdiv", int_arm64_sdiv>;
544 defm ASRV : Shift<0b10, "asrv", sra>;
545 defm LSLV : Shift<0b00, "lslv", shl>;
546 defm LSRV : Shift<0b01, "lsrv", srl>;
547 defm RORV : Shift<0b11, "rorv", rotr>;
549 def : ShiftAlias<"asr", ASRVWr, GPR32>;
550 def : ShiftAlias<"asr", ASRVXr, GPR64>;
551 def : ShiftAlias<"lsl", LSLVWr, GPR32>;
552 def : ShiftAlias<"lsl", LSLVXr, GPR64>;
553 def : ShiftAlias<"lsr", LSRVWr, GPR32>;
554 def : ShiftAlias<"lsr", LSRVXr, GPR64>;
555 def : ShiftAlias<"ror", RORVWr, GPR32>;
556 def : ShiftAlias<"ror", RORVXr, GPR64>;
559 let AddedComplexity = 7 in {
560 defm MADD : MulAccum<0, "madd", add>;
561 defm MSUB : MulAccum<1, "msub", sub>;
563 def : Pat<(i32 (mul GPR32:$Rn, GPR32:$Rm)),
564 (MADDWrrr GPR32:$Rn, GPR32:$Rm, WZR)>;
565 def : Pat<(i64 (mul GPR64:$Rn, GPR64:$Rm)),
566 (MADDXrrr GPR64:$Rn, GPR64:$Rm, XZR)>;
568 def : Pat<(i32 (ineg (mul GPR32:$Rn, GPR32:$Rm))),
569 (MSUBWrrr GPR32:$Rn, GPR32:$Rm, WZR)>;
570 def : Pat<(i64 (ineg (mul GPR64:$Rn, GPR64:$Rm))),
571 (MSUBXrrr GPR64:$Rn, GPR64:$Rm, XZR)>;
572 } // AddedComplexity = 7
574 let AddedComplexity = 5 in {
575 def SMADDLrrr : WideMulAccum<0, 0b001, "smaddl", add, sext>;
576 def SMSUBLrrr : WideMulAccum<1, 0b001, "smsubl", sub, sext>;
577 def UMADDLrrr : WideMulAccum<0, 0b101, "umaddl", add, zext>;
578 def UMSUBLrrr : WideMulAccum<1, 0b101, "umsubl", sub, zext>;
580 def : Pat<(i64 (mul (sext GPR32:$Rn), (sext GPR32:$Rm))),
581 (SMADDLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
582 def : Pat<(i64 (mul (zext GPR32:$Rn), (zext GPR32:$Rm))),
583 (UMADDLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
585 def : Pat<(i64 (ineg (mul (sext GPR32:$Rn), (sext GPR32:$Rm)))),
586 (SMSUBLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
587 def : Pat<(i64 (ineg (mul (zext GPR32:$Rn), (zext GPR32:$Rm)))),
588 (UMSUBLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
589 } // AddedComplexity = 5
591 def : MulAccumWAlias<"mul", MADDWrrr>;
592 def : MulAccumXAlias<"mul", MADDXrrr>;
593 def : MulAccumWAlias<"mneg", MSUBWrrr>;
594 def : MulAccumXAlias<"mneg", MSUBXrrr>;
595 def : WideMulAccumAlias<"smull", SMADDLrrr>;
596 def : WideMulAccumAlias<"smnegl", SMSUBLrrr>;
597 def : WideMulAccumAlias<"umull", UMADDLrrr>;
598 def : WideMulAccumAlias<"umnegl", UMSUBLrrr>;
601 def SMULHrr : MulHi<0b010, "smulh", mulhs>;
602 def UMULHrr : MulHi<0b110, "umulh", mulhu>;
605 def CRC32Brr : BaseCRC32<0, 0b00, 0, GPR32, int_arm64_crc32b, "crc32b">;
606 def CRC32Hrr : BaseCRC32<0, 0b01, 0, GPR32, int_arm64_crc32h, "crc32h">;
607 def CRC32Wrr : BaseCRC32<0, 0b10, 0, GPR32, int_arm64_crc32w, "crc32w">;
608 def CRC32Xrr : BaseCRC32<1, 0b11, 0, GPR64, int_arm64_crc32x, "crc32x">;
610 def CRC32CBrr : BaseCRC32<0, 0b00, 1, GPR32, int_arm64_crc32cb, "crc32cb">;
611 def CRC32CHrr : BaseCRC32<0, 0b01, 1, GPR32, int_arm64_crc32ch, "crc32ch">;
612 def CRC32CWrr : BaseCRC32<0, 0b10, 1, GPR32, int_arm64_crc32cw, "crc32cw">;
613 def CRC32CXrr : BaseCRC32<1, 0b11, 1, GPR64, int_arm64_crc32cx, "crc32cx">;
616 //===----------------------------------------------------------------------===//
617 // Logical instructions.
618 //===----------------------------------------------------------------------===//
621 defm ANDS : LogicalImmS<0b11, "ands", ARM64and_flag>;
622 defm AND : LogicalImm<0b00, "and", and>;
623 defm EOR : LogicalImm<0b10, "eor", xor>;
624 defm ORR : LogicalImm<0b01, "orr", or>;
626 def : InstAlias<"mov $dst, $imm", (ORRWri GPR32sp:$dst, WZR,
627 logical_imm32:$imm)>;
628 def : InstAlias<"mov $dst, $imm", (ORRXri GPR64sp:$dst, XZR,
629 logical_imm64:$imm)>;
633 defm ANDS : LogicalRegS<0b11, 0, "ands", ARM64and_flag>;
634 defm BICS : LogicalRegS<0b11, 1, "bics",
635 BinOpFrag<(ARM64and_flag node:$LHS, (not node:$RHS))>>;
636 defm AND : LogicalReg<0b00, 0, "and", and>;
637 defm BIC : LogicalReg<0b00, 1, "bic",
638 BinOpFrag<(and node:$LHS, (not node:$RHS))>>;
639 defm EON : LogicalReg<0b10, 1, "eon",
640 BinOpFrag<(xor node:$LHS, (not node:$RHS))>>;
641 defm EOR : LogicalReg<0b10, 0, "eor", xor>;
642 defm ORN : LogicalReg<0b01, 1, "orn",
643 BinOpFrag<(or node:$LHS, (not node:$RHS))>>;
644 defm ORR : LogicalReg<0b01, 0, "orr", or>;
646 def : InstAlias<"tst $src1, $src2",
647 (ANDSWri WZR, GPR32:$src1, logical_imm32:$src2)>;
648 def : InstAlias<"tst $src1, $src2",
649 (ANDSXri XZR, GPR64:$src1, logical_imm64:$src2)>;
651 def : InstAlias<"tst $src1, $src2",
652 (ANDSWrs WZR, GPR32:$src1, GPR32:$src2, 0)>;
653 def : InstAlias<"tst $src1, $src2",
654 (ANDSXrs XZR, GPR64:$src1, GPR64:$src2, 0)>;
656 def : InstAlias<"tst $src1, $src2, $sh",
657 (ANDSWrs WZR, GPR32:$src1, GPR32:$src2, logical_shift:$sh)>;
658 def : InstAlias<"tst $src1, $src2, $sh",
659 (ANDSXrs XZR, GPR64:$src1, GPR64:$src2, logical_shift:$sh)>;
661 def : InstAlias<"mvn $Wd, $Wm",
662 (ORNWrs GPR32:$Wd, WZR, GPR32:$Wm, 0)>;
663 def : InstAlias<"mvn $Xd, $Xm",
664 (ORNXrs GPR64:$Xd, XZR, GPR64:$Xm, 0)>;
666 def : InstAlias<"mvn $Wd, $Wm, $sh",
667 (ORNWrs GPR32:$Wd, WZR, GPR32:$Wm, logical_shift:$sh)>;
668 def : InstAlias<"mvn $Xd, $Xm, $sh",
669 (ORNXrs GPR64:$Xd, XZR, GPR64:$Xm, logical_shift:$sh)>;
671 def : Pat<(not GPR32:$Wm), (ORNWrr WZR, GPR32:$Wm)>;
672 def : Pat<(not GPR64:$Xm), (ORNXrr XZR, GPR64:$Xm)>;
675 //===----------------------------------------------------------------------===//
676 // One operand data processing instructions.
677 //===----------------------------------------------------------------------===//
679 defm CLS : OneOperandData<0b101, "cls">;
680 defm CLZ : OneOperandData<0b100, "clz", ctlz>;
681 defm RBIT : OneOperandData<0b000, "rbit">;
682 def REV16Wr : OneWRegData<0b001, "rev16",
683 UnOpFrag<(rotr (bswap node:$LHS), (i64 16))>>;
684 def REV16Xr : OneXRegData<0b001, "rev16", null_frag>;
686 def : Pat<(cttz GPR32:$Rn),
687 (CLZWr (RBITWr GPR32:$Rn))>;
688 def : Pat<(cttz GPR64:$Rn),
689 (CLZXr (RBITXr GPR64:$Rn))>;
690 def : Pat<(ctlz (or (shl (xor (sra GPR32:$Rn, (i64 31)), GPR32:$Rn), (i64 1)),
693 def : Pat<(ctlz (or (shl (xor (sra GPR64:$Rn, (i64 63)), GPR64:$Rn), (i64 1)),
697 // Unlike the other one operand instructions, the instructions with the "rev"
698 // mnemonic do *not* just different in the size bit, but actually use different
699 // opcode bits for the different sizes.
700 def REVWr : OneWRegData<0b010, "rev", bswap>;
701 def REVXr : OneXRegData<0b011, "rev", bswap>;
702 def REV32Xr : OneXRegData<0b010, "rev32",
703 UnOpFrag<(rotr (bswap node:$LHS), (i64 32))>>;
705 // The bswap commutes with the rotr so we want a pattern for both possible
707 def : Pat<(bswap (rotr GPR32:$Rn, (i64 16))), (REV16Wr GPR32:$Rn)>;
708 def : Pat<(bswap (rotr GPR64:$Rn, (i64 32))), (REV32Xr GPR64:$Rn)>;
710 //===----------------------------------------------------------------------===//
711 // Bitfield immediate extraction instruction.
712 //===----------------------------------------------------------------------===//
713 let neverHasSideEffects = 1 in
714 defm EXTR : ExtractImm<"extr">;
715 def : InstAlias<"ror $dst, $src, $shift",
716 (EXTRWrri GPR32:$dst, GPR32:$src, GPR32:$src, imm0_31:$shift)>;
717 def : InstAlias<"ror $dst, $src, $shift",
718 (EXTRXrri GPR64:$dst, GPR64:$src, GPR64:$src, imm0_63:$shift)>;
720 def : Pat<(rotr GPR32:$Rn, (i64 imm0_31:$imm)),
721 (EXTRWrri GPR32:$Rn, GPR32:$Rn, imm0_31:$imm)>;
722 def : Pat<(rotr GPR64:$Rn, (i64 imm0_63:$imm)),
723 (EXTRXrri GPR64:$Rn, GPR64:$Rn, imm0_63:$imm)>;
725 //===----------------------------------------------------------------------===//
726 // Other bitfield immediate instructions.
727 //===----------------------------------------------------------------------===//
728 let neverHasSideEffects = 1 in {
729 defm BFM : BitfieldImmWith2RegArgs<0b01, "bfm">;
730 defm SBFM : BitfieldImm<0b00, "sbfm">;
731 defm UBFM : BitfieldImm<0b10, "ubfm">;
734 def i32shift_a : Operand<i64>, SDNodeXForm<imm, [{
735 uint64_t enc = (32 - N->getZExtValue()) & 0x1f;
736 return CurDAG->getTargetConstant(enc, MVT::i64);
739 def i32shift_b : Operand<i64>, SDNodeXForm<imm, [{
740 uint64_t enc = 31 - N->getZExtValue();
741 return CurDAG->getTargetConstant(enc, MVT::i64);
744 // min(7, 31 - shift_amt)
745 def i32shift_sext_i8 : Operand<i64>, SDNodeXForm<imm, [{
746 uint64_t enc = 31 - N->getZExtValue();
747 enc = enc > 7 ? 7 : enc;
748 return CurDAG->getTargetConstant(enc, MVT::i64);
751 // min(15, 31 - shift_amt)
752 def i32shift_sext_i16 : Operand<i64>, SDNodeXForm<imm, [{
753 uint64_t enc = 31 - N->getZExtValue();
754 enc = enc > 15 ? 15 : enc;
755 return CurDAG->getTargetConstant(enc, MVT::i64);
758 def i64shift_a : Operand<i64>, SDNodeXForm<imm, [{
759 uint64_t enc = (64 - N->getZExtValue()) & 0x3f;
760 return CurDAG->getTargetConstant(enc, MVT::i64);
763 def i64shift_b : Operand<i64>, SDNodeXForm<imm, [{
764 uint64_t enc = 63 - N->getZExtValue();
765 return CurDAG->getTargetConstant(enc, MVT::i64);
768 // min(7, 63 - shift_amt)
769 def i64shift_sext_i8 : Operand<i64>, SDNodeXForm<imm, [{
770 uint64_t enc = 63 - N->getZExtValue();
771 enc = enc > 7 ? 7 : enc;
772 return CurDAG->getTargetConstant(enc, MVT::i64);
775 // min(15, 63 - shift_amt)
776 def i64shift_sext_i16 : Operand<i64>, SDNodeXForm<imm, [{
777 uint64_t enc = 63 - N->getZExtValue();
778 enc = enc > 15 ? 15 : enc;
779 return CurDAG->getTargetConstant(enc, MVT::i64);
782 // min(31, 63 - shift_amt)
783 def i64shift_sext_i32 : Operand<i64>, SDNodeXForm<imm, [{
784 uint64_t enc = 63 - N->getZExtValue();
785 enc = enc > 31 ? 31 : enc;
786 return CurDAG->getTargetConstant(enc, MVT::i64);
789 def : Pat<(shl GPR32:$Rn, (i64 imm0_31:$imm)),
790 (UBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
791 (i64 (i32shift_b imm0_31:$imm)))>;
792 def : Pat<(shl GPR64:$Rn, (i64 imm0_63:$imm)),
793 (UBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
794 (i64 (i64shift_b imm0_63:$imm)))>;
796 let AddedComplexity = 10 in {
797 def : Pat<(sra GPR32:$Rn, (i64 imm0_31:$imm)),
798 (SBFMWri GPR32:$Rn, imm0_31:$imm, 31)>;
799 def : Pat<(sra GPR64:$Rn, (i64 imm0_63:$imm)),
800 (SBFMXri GPR64:$Rn, imm0_63:$imm, 63)>;
803 def : InstAlias<"asr $dst, $src, $shift",
804 (SBFMWri GPR32:$dst, GPR32:$src, imm0_31:$shift, 31)>;
805 def : InstAlias<"asr $dst, $src, $shift",
806 (SBFMXri GPR64:$dst, GPR64:$src, imm0_63:$shift, 63)>;
807 def : InstAlias<"sxtb $dst, $src", (SBFMWri GPR32:$dst, GPR32:$src, 0, 7)>;
808 def : InstAlias<"sxtb $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 7)>;
809 def : InstAlias<"sxth $dst, $src", (SBFMWri GPR32:$dst, GPR32:$src, 0, 15)>;
810 def : InstAlias<"sxth $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 15)>;
811 def : InstAlias<"sxtw $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 31)>;
813 def : Pat<(srl GPR32:$Rn, (i64 imm0_31:$imm)),
814 (UBFMWri GPR32:$Rn, imm0_31:$imm, 31)>;
815 def : Pat<(srl GPR64:$Rn, (i64 imm0_63:$imm)),
816 (UBFMXri GPR64:$Rn, imm0_63:$imm, 63)>;
818 def : InstAlias<"lsr $dst, $src, $shift",
819 (UBFMWri GPR32:$dst, GPR32:$src, imm0_31:$shift, 31)>;
820 def : InstAlias<"lsr $dst, $src, $shift",
821 (UBFMXri GPR64:$dst, GPR64:$src, imm0_63:$shift, 63)>;
822 def : InstAlias<"uxtb $dst, $src", (UBFMWri GPR32:$dst, GPR32:$src, 0, 7)>;
823 def : InstAlias<"uxtb $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 7)>;
824 def : InstAlias<"uxth $dst, $src", (UBFMWri GPR32:$dst, GPR32:$src, 0, 15)>;
825 def : InstAlias<"uxth $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 15)>;
826 def : InstAlias<"uxtw $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 31)>;
828 //===----------------------------------------------------------------------===//
829 // Conditionally set flags instructions.
830 //===----------------------------------------------------------------------===//
831 defm CCMN : CondSetFlagsImm<0, "ccmn">;
832 defm CCMP : CondSetFlagsImm<1, "ccmp">;
834 defm CCMN : CondSetFlagsReg<0, "ccmn">;
835 defm CCMP : CondSetFlagsReg<1, "ccmp">;
837 //===----------------------------------------------------------------------===//
838 // Conditional select instructions.
839 //===----------------------------------------------------------------------===//
840 defm CSEL : CondSelect<0, 0b00, "csel">;
842 def inc : PatFrag<(ops node:$in), (add node:$in, 1)>;
843 defm CSINC : CondSelectOp<0, 0b01, "csinc", inc>;
844 defm CSINV : CondSelectOp<1, 0b00, "csinv", not>;
845 defm CSNEG : CondSelectOp<1, 0b01, "csneg", ineg>;
847 def : Pat<(ARM64csinv GPR32:$tval, GPR32:$fval, (i32 imm:$cc), CPSR),
848 (CSINVWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
849 def : Pat<(ARM64csinv GPR64:$tval, GPR64:$fval, (i32 imm:$cc), CPSR),
850 (CSINVXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
851 def : Pat<(ARM64csneg GPR32:$tval, GPR32:$fval, (i32 imm:$cc), CPSR),
852 (CSNEGWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
853 def : Pat<(ARM64csneg GPR64:$tval, GPR64:$fval, (i32 imm:$cc), CPSR),
854 (CSNEGXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
855 def : Pat<(ARM64csinc GPR32:$tval, GPR32:$fval, (i32 imm:$cc), CPSR),
856 (CSINCWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
857 def : Pat<(ARM64csinc GPR64:$tval, GPR64:$fval, (i32 imm:$cc), CPSR),
858 (CSINCXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
860 def : Pat<(ARM64csel (i32 0), (i32 1), (i32 imm:$cc), CPSR),
861 (CSINCWr WZR, WZR, (i32 imm:$cc))>;
862 def : Pat<(ARM64csel (i64 0), (i64 1), (i32 imm:$cc), CPSR),
863 (CSINCXr XZR, XZR, (i32 imm:$cc))>;
864 def : Pat<(ARM64csel (i32 0), (i32 -1), (i32 imm:$cc), CPSR),
865 (CSINVWr WZR, WZR, (i32 imm:$cc))>;
866 def : Pat<(ARM64csel (i64 0), (i64 -1), (i32 imm:$cc), CPSR),
867 (CSINVXr XZR, XZR, (i32 imm:$cc))>;
869 // The inverse of the condition code from the alias instruction is what is used
870 // in the aliased instruction. The parser all ready inverts the condition code
871 // for these aliases.
872 // FIXME: Is this the correct way to handle these aliases?
873 def : InstAlias<"cset $dst, $cc", (CSINCWr GPR32:$dst, WZR, WZR, ccode:$cc)>;
874 def : InstAlias<"cset $dst, $cc", (CSINCXr GPR64:$dst, XZR, XZR, ccode:$cc)>;
876 def : InstAlias<"csetm $dst, $cc", (CSINVWr GPR32:$dst, WZR, WZR, ccode:$cc)>;
877 def : InstAlias<"csetm $dst, $cc", (CSINVXr GPR64:$dst, XZR, XZR, ccode:$cc)>;
879 def : InstAlias<"cinc $dst, $src, $cc",
880 (CSINCWr GPR32:$dst, GPR32:$src, GPR32:$src, ccode:$cc)>;
881 def : InstAlias<"cinc $dst, $src, $cc",
882 (CSINCXr GPR64:$dst, GPR64:$src, GPR64:$src, ccode:$cc)>;
884 def : InstAlias<"cinv $dst, $src, $cc",
885 (CSINVWr GPR32:$dst, GPR32:$src, GPR32:$src, ccode:$cc)>;
886 def : InstAlias<"cinv $dst, $src, $cc",
887 (CSINVXr GPR64:$dst, GPR64:$src, GPR64:$src, ccode:$cc)>;
889 def : InstAlias<"cneg $dst, $src, $cc",
890 (CSNEGWr GPR32:$dst, GPR32:$src, GPR32:$src, ccode:$cc)>;
891 def : InstAlias<"cneg $dst, $src, $cc",
892 (CSNEGXr GPR64:$dst, GPR64:$src, GPR64:$src, ccode:$cc)>;
894 //===----------------------------------------------------------------------===//
895 // PC-relative instructions.
896 //===----------------------------------------------------------------------===//
897 let isReMaterializable = 1 in {
898 let neverHasSideEffects = 1, mayStore = 0, mayLoad = 0 in {
899 def ADR : ADRI<0, "adr", adrlabel, []>;
900 } // neverHasSideEffects = 1
902 def ADRP : ADRI<1, "adrp", adrplabel,
903 [(set GPR64:$Xd, (ARM64adrp tglobaladdr:$label))]>;
904 } // isReMaterializable = 1
906 // page address of a constant pool entry, block address
907 def : Pat<(ARM64adrp tconstpool:$cp), (ADRP tconstpool:$cp)>;
908 def : Pat<(ARM64adrp tblockaddress:$cp), (ADRP tblockaddress:$cp)>;
910 //===----------------------------------------------------------------------===//
911 // Unconditional branch (register) instructions.
912 //===----------------------------------------------------------------------===//
914 let isReturn = 1, isTerminator = 1, isBarrier = 1 in {
915 def RET : BranchReg<0b0010, "ret", []>;
916 def DRPS : SpecialReturn<0b0101, "drps">;
917 def ERET : SpecialReturn<0b0100, "eret">;
918 } // isReturn = 1, isTerminator = 1, isBarrier = 1
920 // Default to the LR register.
921 def : InstAlias<"ret", (RET LR)>;
923 let isCall = 1, Defs = [LR], Uses = [SP] in {
924 def BLR : BranchReg<0b0001, "blr", [(ARM64call GPR64:$Rn)]>;
927 let isBranch = 1, isTerminator = 1, isBarrier = 1, isIndirectBranch = 1 in {
928 def BR : BranchReg<0b0000, "br", [(brind GPR64:$Rn)]>;
929 } // isBranch, isTerminator, isBarrier, isIndirectBranch
931 // Create a separate pseudo-instruction for codegen to use so that we don't
932 // flag lr as used in every function. It'll be restored before the RET by the
933 // epilogue if it's legitimately used.
934 def RET_ReallyLR : Pseudo<(outs), (ins), [(ARM64retflag)]> {
935 let isTerminator = 1;
940 // This is a directive-like pseudo-instruction. The purpose is to insert an
941 // R_AARCH64_TLSDESC_CALL relocation at the offset of the following instruction
942 // (which in the usual case is a BLR).
943 let hasSideEffects = 1 in
944 def TLSDESCCALL : Pseudo<(outs), (ins i64imm:$sym), []> {
945 let AsmString = ".tlsdesccall $sym";
948 // Pseudo-instruction representing a BLR with attached TLSDESC relocation. It
949 // gets expanded to two MCInsts during lowering.
950 let isCall = 1, Defs = [LR] in
952 : Pseudo<(outs), (ins GPR64:$dest, i64imm:$sym),
953 [(ARM64tlsdesc_call GPR64:$dest, tglobaltlsaddr:$sym)]>;
955 def : Pat<(ARM64tlsdesc_call GPR64:$dest, texternalsym:$sym),
956 (TLSDESC_BLR GPR64:$dest, texternalsym:$sym)>;
957 //===----------------------------------------------------------------------===//
958 // Conditional branch (immediate) instruction.
959 //===----------------------------------------------------------------------===//
960 def Bcc : BranchCond;
962 //===----------------------------------------------------------------------===//
963 // Compare-and-branch instructions.
964 //===----------------------------------------------------------------------===//
965 defm CBZ : CmpBranch<0, "cbz", ARM64cbz>;
966 defm CBNZ : CmpBranch<1, "cbnz", ARM64cbnz>;
968 //===----------------------------------------------------------------------===//
969 // Test-bit-and-branch instructions.
970 //===----------------------------------------------------------------------===//
971 def TBZ : TestBranch<0, "tbz", ARM64tbz>;
972 def TBNZ : TestBranch<1, "tbnz", ARM64tbnz>;
974 //===----------------------------------------------------------------------===//
975 // Unconditional branch (immediate) instructions.
976 //===----------------------------------------------------------------------===//
977 let isBranch = 1, isTerminator = 1, isBarrier = 1 in {
978 def B : BranchImm<0, "b", [(br bb:$addr)]>;
979 } // isBranch, isTerminator, isBarrier
981 let isCall = 1, Defs = [LR], Uses = [SP] in {
982 def BL : CallImm<1, "bl", [(ARM64call tglobaladdr:$addr)]>;
984 def : Pat<(ARM64call texternalsym:$func), (BL texternalsym:$func)>;
986 //===----------------------------------------------------------------------===//
987 // Exception generation instructions.
988 //===----------------------------------------------------------------------===//
989 def BRK : ExceptionGeneration<0b001, 0b00, "brk">;
990 def DCPS1 : ExceptionGeneration<0b101, 0b01, "dcps1">;
991 def DCPS2 : ExceptionGeneration<0b101, 0b10, "dcps2">;
992 def DCPS3 : ExceptionGeneration<0b101, 0b11, "dcps3">;
993 def HLT : ExceptionGeneration<0b010, 0b00, "hlt">;
994 def HVC : ExceptionGeneration<0b000, 0b10, "hvc">;
995 def SMC : ExceptionGeneration<0b000, 0b11, "smc">;
996 def SVC : ExceptionGeneration<0b000, 0b01, "svc">;
998 // DCPSn defaults to an immediate operand of zero if unspecified.
999 def : InstAlias<"dcps1", (DCPS1 0)>;
1000 def : InstAlias<"dcps2", (DCPS2 0)>;
1001 def : InstAlias<"dcps3", (DCPS3 0)>;
1003 //===----------------------------------------------------------------------===//
1004 // Load instructions.
1005 //===----------------------------------------------------------------------===//
1007 // Pair (indexed, offset)
1008 def LDPWi : LoadPairOffset<0b00, 0, GPR32, am_indexed32simm7, "ldp">;
1009 def LDPXi : LoadPairOffset<0b10, 0, GPR64, am_indexed64simm7, "ldp">;
1010 def LDPSi : LoadPairOffset<0b00, 1, FPR32, am_indexed32simm7, "ldp">;
1011 def LDPDi : LoadPairOffset<0b01, 1, FPR64, am_indexed64simm7, "ldp">;
1012 def LDPQi : LoadPairOffset<0b10, 1, FPR128, am_indexed128simm7, "ldp">;
1014 def LDPSWi : LoadPairOffset<0b01, 0, GPR64, am_indexed32simm7, "ldpsw">;
1016 // Pair (pre-indexed)
1017 def LDPWpre : LoadPairPreIdx<0b00, 0, GPR32, am_indexed32simm7_wb, "ldp">;
1018 def LDPXpre : LoadPairPreIdx<0b10, 0, GPR64, am_indexed64simm7_wb, "ldp">;
1019 def LDPSpre : LoadPairPreIdx<0b00, 1, FPR32, am_indexed32simm7_wb, "ldp">;
1020 def LDPDpre : LoadPairPreIdx<0b01, 1, FPR64, am_indexed64simm7_wb, "ldp">;
1021 def LDPQpre : LoadPairPreIdx<0b10, 1, FPR128, am_indexed128simm7_wb, "ldp">;
1023 def LDPSWpre : LoadPairPreIdx<0b01, 0, GPR64, am_indexed32simm7_wb, "ldpsw">;
1025 // Pair (post-indexed)
1026 def LDPWpost : LoadPairPostIdx<0b00, 0, GPR32, simm7s4, "ldp">;
1027 def LDPXpost : LoadPairPostIdx<0b10, 0, GPR64, simm7s8, "ldp">;
1028 def LDPSpost : LoadPairPostIdx<0b00, 1, FPR32, simm7s4, "ldp">;
1029 def LDPDpost : LoadPairPostIdx<0b01, 1, FPR64, simm7s8, "ldp">;
1030 def LDPQpost : LoadPairPostIdx<0b10, 1, FPR128, simm7s16, "ldp">;
1032 def LDPSWpost : LoadPairPostIdx<0b01, 0, GPR64, simm7s4, "ldpsw">;
1035 // Pair (no allocate)
1036 def LDNPWi : LoadPairNoAlloc<0b00, 0, GPR32, am_indexed32simm7, "ldnp">;
1037 def LDNPXi : LoadPairNoAlloc<0b10, 0, GPR64, am_indexed64simm7, "ldnp">;
1038 def LDNPSi : LoadPairNoAlloc<0b00, 1, FPR32, am_indexed32simm7, "ldnp">;
1039 def LDNPDi : LoadPairNoAlloc<0b01, 1, FPR64, am_indexed64simm7, "ldnp">;
1040 def LDNPQi : LoadPairNoAlloc<0b10, 1, FPR128, am_indexed128simm7, "ldnp">;
1043 // (register offset)
1046 let AddedComplexity = 10 in {
1048 def LDRBBro : Load8RO<0b00, 0, 0b01, GPR32, "ldrb",
1049 [(set GPR32:$Rt, (zextloadi8 ro_indexed8:$addr))]>;
1050 def LDRHHro : Load16RO<0b01, 0, 0b01, GPR32, "ldrh",
1051 [(set GPR32:$Rt, (zextloadi16 ro_indexed16:$addr))]>;
1052 def LDRWro : Load32RO<0b10, 0, 0b01, GPR32, "ldr",
1053 [(set GPR32:$Rt, (load ro_indexed32:$addr))]>;
1054 def LDRXro : Load64RO<0b11, 0, 0b01, GPR64, "ldr",
1055 [(set GPR64:$Rt, (load ro_indexed64:$addr))]>;
1058 def LDRBro : Load8RO<0b00, 1, 0b01, FPR8, "ldr",
1059 [(set FPR8:$Rt, (load ro_indexed8:$addr))]>;
1060 def LDRHro : Load16RO<0b01, 1, 0b01, FPR16, "ldr",
1061 [(set (f16 FPR16:$Rt), (load ro_indexed16:$addr))]>;
1062 def LDRSro : Load32RO<0b10, 1, 0b01, FPR32, "ldr",
1063 [(set (f32 FPR32:$Rt), (load ro_indexed32:$addr))]>;
1064 def LDRDro : Load64RO<0b11, 1, 0b01, FPR64, "ldr",
1065 [(set (f64 FPR64:$Rt), (load ro_indexed64:$addr))]>;
1066 def LDRQro : Load128RO<0b00, 1, 0b11, FPR128, "ldr", []> {
1070 // For regular load, we do not have any alignment requirement.
1071 // Thus, it is safe to directly map the vector loads with interesting
1072 // addressing modes.
1073 // FIXME: We could do the same for bitconvert to floating point vectors.
1074 def : Pat <(v8i8 (scalar_to_vector (i32 (extloadi8 ro_indexed8:$addr)))),
1075 (INSERT_SUBREG (v8i8 (IMPLICIT_DEF)),
1076 (LDRBro ro_indexed8:$addr), bsub)>;
1077 def : Pat <(v16i8 (scalar_to_vector (i32 (extloadi8 ro_indexed8:$addr)))),
1078 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
1079 (LDRBro ro_indexed8:$addr), bsub)>;
1080 def : Pat <(v4i16 (scalar_to_vector (i32 (extloadi16 ro_indexed16:$addr)))),
1081 (INSERT_SUBREG (v4i16 (IMPLICIT_DEF)),
1082 (LDRHro ro_indexed16:$addr), hsub)>;
1083 def : Pat <(v8i16 (scalar_to_vector (i32 (extloadi16 ro_indexed16:$addr)))),
1084 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
1085 (LDRHro ro_indexed16:$addr), hsub)>;
1086 def : Pat <(v2i32 (scalar_to_vector (i32 (load ro_indexed32:$addr)))),
1087 (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)),
1088 (LDRSro ro_indexed32:$addr), ssub)>;
1089 def : Pat <(v4i32 (scalar_to_vector (i32 (load ro_indexed32:$addr)))),
1090 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
1091 (LDRSro ro_indexed32:$addr), ssub)>;
1092 def : Pat <(v1i64 (scalar_to_vector (i64 (load ro_indexed64:$addr)))),
1093 (LDRDro ro_indexed64:$addr)>;
1094 def : Pat <(v2i64 (scalar_to_vector (i64 (load ro_indexed64:$addr)))),
1095 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
1096 (LDRDro ro_indexed64:$addr), dsub)>;
1098 // Match all load 64 bits width whose type is compatible with FPR64
1099 def : Pat<(v2f32 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1100 def : Pat<(v1f64 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1101 def : Pat<(v8i8 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1102 def : Pat<(v4i16 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1103 def : Pat<(v2i32 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1104 def : Pat<(v1i64 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1106 // Match all load 128 bits width whose type is compatible with FPR128
1107 def : Pat<(v4f32 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1108 def : Pat<(v2f64 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1109 def : Pat<(v16i8 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1110 def : Pat<(v8i16 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1111 def : Pat<(v4i32 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1112 def : Pat<(v2i64 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1113 def : Pat<(f128 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1115 // Load sign-extended half-word
1116 def LDRSHWro : Load16RO<0b01, 0, 0b11, GPR32, "ldrsh",
1117 [(set GPR32:$Rt, (sextloadi16 ro_indexed16:$addr))]>;
1118 def LDRSHXro : Load16RO<0b01, 0, 0b10, GPR64, "ldrsh",
1119 [(set GPR64:$Rt, (sextloadi16 ro_indexed16:$addr))]>;
1121 // Load sign-extended byte
1122 def LDRSBWro : Load8RO<0b00, 0, 0b11, GPR32, "ldrsb",
1123 [(set GPR32:$Rt, (sextloadi8 ro_indexed8:$addr))]>;
1124 def LDRSBXro : Load8RO<0b00, 0, 0b10, GPR64, "ldrsb",
1125 [(set GPR64:$Rt, (sextloadi8 ro_indexed8:$addr))]>;
1127 // Load sign-extended word
1128 def LDRSWro : Load32RO<0b10, 0, 0b10, GPR64, "ldrsw",
1129 [(set GPR64:$Rt, (sextloadi32 ro_indexed32:$addr))]>;
1132 def PRFMro : PrefetchRO<0b11, 0, 0b10, "prfm",
1133 [(ARM64Prefetch imm:$Rt, ro_indexed64:$addr)]>;
1136 def : Pat<(i64 (zextloadi8 ro_indexed8:$addr)),
1137 (SUBREG_TO_REG (i64 0), (LDRBBro ro_indexed8:$addr), sub_32)>;
1138 def : Pat<(i64 (zextloadi16 ro_indexed16:$addr)),
1139 (SUBREG_TO_REG (i64 0), (LDRHHro ro_indexed16:$addr), sub_32)>;
1140 def : Pat<(i64 (zextloadi32 ro_indexed32:$addr)),
1141 (SUBREG_TO_REG (i64 0), (LDRWro ro_indexed32:$addr), sub_32)>;
1143 // zextloadi1 -> zextloadi8
1144 def : Pat<(i32 (zextloadi1 ro_indexed8:$addr)), (LDRBBro ro_indexed8:$addr)>;
1145 def : Pat<(i64 (zextloadi1 ro_indexed8:$addr)),
1146 (SUBREG_TO_REG (i64 0), (LDRBBro ro_indexed8:$addr), sub_32)>;
1148 // extload -> zextload
1149 def : Pat<(i32 (extloadi16 ro_indexed16:$addr)), (LDRHHro ro_indexed16:$addr)>;
1150 def : Pat<(i32 (extloadi8 ro_indexed8:$addr)), (LDRBBro ro_indexed8:$addr)>;
1151 def : Pat<(i32 (extloadi1 ro_indexed8:$addr)), (LDRBBro ro_indexed8:$addr)>;
1152 def : Pat<(i64 (extloadi32 ro_indexed32:$addr)),
1153 (SUBREG_TO_REG (i64 0), (LDRWro ro_indexed32:$addr), sub_32)>;
1154 def : Pat<(i64 (extloadi16 ro_indexed16:$addr)),
1155 (SUBREG_TO_REG (i64 0), (LDRHHro ro_indexed16:$addr), sub_32)>;
1156 def : Pat<(i64 (extloadi8 ro_indexed8:$addr)),
1157 (SUBREG_TO_REG (i64 0), (LDRBBro ro_indexed8:$addr), sub_32)>;
1158 def : Pat<(i64 (extloadi1 ro_indexed8:$addr)),
1159 (SUBREG_TO_REG (i64 0), (LDRBBro ro_indexed8:$addr), sub_32)>;
1161 } // AddedComplexity = 10
1164 // (unsigned immediate)
1166 def LDRXui : LoadUI<0b11, 0, 0b01, GPR64, am_indexed64, "ldr",
1167 [(set GPR64:$Rt, (load am_indexed64:$addr))]>;
1168 def LDRWui : LoadUI<0b10, 0, 0b01, GPR32, am_indexed32, "ldr",
1169 [(set GPR32:$Rt, (load am_indexed32:$addr))]>;
1170 def LDRBui : LoadUI<0b00, 1, 0b01, FPR8, am_indexed8, "ldr",
1171 [(set FPR8:$Rt, (load am_indexed8:$addr))]>;
1172 def LDRHui : LoadUI<0b01, 1, 0b01, FPR16, am_indexed16, "ldr",
1173 [(set (f16 FPR16:$Rt), (load am_indexed16:$addr))]>;
1174 def LDRSui : LoadUI<0b10, 1, 0b01, FPR32, am_indexed32, "ldr",
1175 [(set (f32 FPR32:$Rt), (load am_indexed32:$addr))]>;
1176 def LDRDui : LoadUI<0b11, 1, 0b01, FPR64, am_indexed64, "ldr",
1177 [(set (f64 FPR64:$Rt), (load am_indexed64:$addr))]>;
1178 def LDRQui : LoadUI<0b00, 1, 0b11, FPR128, am_indexed128, "ldr",
1179 [(set (f128 FPR128:$Rt), (load am_indexed128:$addr))]>;
1181 // For regular load, we do not have any alignment requirement.
1182 // Thus, it is safe to directly map the vector loads with interesting
1183 // addressing modes.
1184 // FIXME: We could do the same for bitconvert to floating point vectors.
1185 def : Pat <(v8i8 (scalar_to_vector (i32 (extloadi8 am_indexed8:$addr)))),
1186 (INSERT_SUBREG (v8i8 (IMPLICIT_DEF)),
1187 (LDRBui am_indexed8:$addr), bsub)>;
1188 def : Pat <(v16i8 (scalar_to_vector (i32 (extloadi8 am_indexed8:$addr)))),
1189 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
1190 (LDRBui am_indexed8:$addr), bsub)>;
1191 def : Pat <(v4i16 (scalar_to_vector (i32 (extloadi16 am_indexed16:$addr)))),
1192 (INSERT_SUBREG (v4i16 (IMPLICIT_DEF)),
1193 (LDRHui am_indexed16:$addr), hsub)>;
1194 def : Pat <(v8i16 (scalar_to_vector (i32 (extloadi16 am_indexed16:$addr)))),
1195 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
1196 (LDRHui am_indexed16:$addr), hsub)>;
1197 def : Pat <(v2i32 (scalar_to_vector (i32 (load am_indexed32:$addr)))),
1198 (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)),
1199 (LDRSui am_indexed32:$addr), ssub)>;
1200 def : Pat <(v4i32 (scalar_to_vector (i32 (load am_indexed32:$addr)))),
1201 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
1202 (LDRSui am_indexed32:$addr), ssub)>;
1203 def : Pat <(v1i64 (scalar_to_vector (i64 (load am_indexed64:$addr)))),
1204 (LDRDui am_indexed64:$addr)>;
1205 def : Pat <(v2i64 (scalar_to_vector (i64 (load am_indexed64:$addr)))),
1206 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
1207 (LDRDui am_indexed64:$addr), dsub)>;
1209 // Match all load 64 bits width whose type is compatible with FPR64
1210 def : Pat<(v2f32 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1211 def : Pat<(v1f64 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1212 def : Pat<(v8i8 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1213 def : Pat<(v4i16 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1214 def : Pat<(v2i32 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1215 def : Pat<(v1i64 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1217 // Match all load 128 bits width whose type is compatible with FPR128
1218 def : Pat<(v4f32 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1219 def : Pat<(v2f64 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1220 def : Pat<(v16i8 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1221 def : Pat<(v8i16 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1222 def : Pat<(v4i32 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1223 def : Pat<(v2i64 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1224 def : Pat<(f128 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1226 def LDRHHui : LoadUI<0b01, 0, 0b01, GPR32, am_indexed16, "ldrh",
1227 [(set GPR32:$Rt, (zextloadi16 am_indexed16:$addr))]>;
1228 def LDRBBui : LoadUI<0b00, 0, 0b01, GPR32, am_indexed8, "ldrb",
1229 [(set GPR32:$Rt, (zextloadi8 am_indexed8:$addr))]>;
1231 def : Pat<(i64 (zextloadi8 am_indexed8:$addr)),
1232 (SUBREG_TO_REG (i64 0), (LDRBBui am_indexed8:$addr), sub_32)>;
1233 def : Pat<(i64 (zextloadi16 am_indexed16:$addr)),
1234 (SUBREG_TO_REG (i64 0), (LDRHHui am_indexed16:$addr), sub_32)>;
1236 // zextloadi1 -> zextloadi8
1237 def : Pat<(i32 (zextloadi1 am_indexed8:$addr)), (LDRBBui am_indexed8:$addr)>;
1238 def : Pat<(i64 (zextloadi1 am_indexed8:$addr)),
1239 (SUBREG_TO_REG (i64 0), (LDRBBui am_indexed8:$addr), sub_32)>;
1241 // extload -> zextload
1242 def : Pat<(i32 (extloadi16 am_indexed16:$addr)), (LDRHHui am_indexed16:$addr)>;
1243 def : Pat<(i32 (extloadi8 am_indexed8:$addr)), (LDRBBui am_indexed8:$addr)>;
1244 def : Pat<(i32 (extloadi1 am_indexed8:$addr)), (LDRBBui am_indexed8:$addr)>;
1245 def : Pat<(i64 (extloadi32 am_indexed32:$addr)),
1246 (SUBREG_TO_REG (i64 0), (LDRWui am_indexed32:$addr), sub_32)>;
1247 def : Pat<(i64 (extloadi16 am_indexed16:$addr)),
1248 (SUBREG_TO_REG (i64 0), (LDRHHui am_indexed16:$addr), sub_32)>;
1249 def : Pat<(i64 (extloadi8 am_indexed8:$addr)),
1250 (SUBREG_TO_REG (i64 0), (LDRBBui am_indexed8:$addr), sub_32)>;
1251 def : Pat<(i64 (extloadi1 am_indexed8:$addr)),
1252 (SUBREG_TO_REG (i64 0), (LDRBBui am_indexed8:$addr), sub_32)>;
1254 // load sign-extended half-word
1255 def LDRSHWui : LoadUI<0b01, 0, 0b11, GPR32, am_indexed16, "ldrsh",
1256 [(set GPR32:$Rt, (sextloadi16 am_indexed16:$addr))]>;
1257 def LDRSHXui : LoadUI<0b01, 0, 0b10, GPR64, am_indexed16, "ldrsh",
1258 [(set GPR64:$Rt, (sextloadi16 am_indexed16:$addr))]>;
1260 // load sign-extended byte
1261 def LDRSBWui : LoadUI<0b00, 0, 0b11, GPR32, am_indexed8, "ldrsb",
1262 [(set GPR32:$Rt, (sextloadi8 am_indexed8:$addr))]>;
1263 def LDRSBXui : LoadUI<0b00, 0, 0b10, GPR64, am_indexed8, "ldrsb",
1264 [(set GPR64:$Rt, (sextloadi8 am_indexed8:$addr))]>;
1266 // load sign-extended word
1267 def LDRSWui : LoadUI<0b10, 0, 0b10, GPR64, am_indexed32, "ldrsw",
1268 [(set GPR64:$Rt, (sextloadi32 am_indexed32:$addr))]>;
1270 // load zero-extended word
1271 def : Pat<(i64 (zextloadi32 am_indexed32:$addr)),
1272 (SUBREG_TO_REG (i64 0), (LDRWui am_indexed32:$addr), sub_32)>;
1275 def PRFMui : PrefetchUI<0b11, 0, 0b10, "prfm",
1276 [(ARM64Prefetch imm:$Rt, am_indexed64:$addr)]>;
1280 def LDRWl : LoadLiteral<0b00, 0, GPR32, "ldr">;
1281 def LDRXl : LoadLiteral<0b01, 0, GPR64, "ldr">;
1282 def LDRSl : LoadLiteral<0b00, 1, FPR32, "ldr">;
1283 def LDRDl : LoadLiteral<0b01, 1, FPR64, "ldr">;
1284 def LDRQl : LoadLiteral<0b10, 1, FPR128, "ldr">;
1286 // load sign-extended word
1287 def LDRSWl : LoadLiteral<0b10, 0, GPR64, "ldrsw">;
1290 def PRFMl : PrefetchLiteral<0b11, 0, "prfm", []>;
1291 // [(ARM64Prefetch imm:$Rt, tglobaladdr:$label)]>;
1294 // (unscaled immediate)
1295 def LDURXi : LoadUnscaled<0b11, 0, 0b01, GPR64, am_unscaled64, "ldur",
1296 [(set GPR64:$Rt, (load am_unscaled64:$addr))]>;
1297 def LDURWi : LoadUnscaled<0b10, 0, 0b01, GPR32, am_unscaled32, "ldur",
1298 [(set GPR32:$Rt, (load am_unscaled32:$addr))]>;
1299 def LDURBi : LoadUnscaled<0b00, 1, 0b01, FPR8, am_unscaled8, "ldur",
1300 [(set FPR8:$Rt, (load am_unscaled8:$addr))]>;
1301 def LDURHi : LoadUnscaled<0b01, 1, 0b01, FPR16, am_unscaled16, "ldur",
1302 [(set (f16 FPR16:$Rt), (load am_unscaled16:$addr))]>;
1303 def LDURSi : LoadUnscaled<0b10, 1, 0b01, FPR32, am_unscaled32, "ldur",
1304 [(set (f32 FPR32:$Rt), (load am_unscaled32:$addr))]>;
1305 def LDURDi : LoadUnscaled<0b11, 1, 0b01, FPR64, am_unscaled64, "ldur",
1306 [(set (f64 FPR64:$Rt), (load am_unscaled64:$addr))]>;
1307 def LDURQi : LoadUnscaled<0b00, 1, 0b11, FPR128, am_unscaled128, "ldur",
1308 [(set (v2f64 FPR128:$Rt), (load am_unscaled128:$addr))]>;
1311 : LoadUnscaled<0b01, 0, 0b01, GPR32, am_unscaled16, "ldurh",
1312 [(set GPR32:$Rt, (zextloadi16 am_unscaled16:$addr))]>;
1314 : LoadUnscaled<0b00, 0, 0b01, GPR32, am_unscaled8, "ldurb",
1315 [(set GPR32:$Rt, (zextloadi8 am_unscaled8:$addr))]>;
1317 // Match all load 64 bits width whose type is compatible with FPR64
1318 def : Pat<(v2f32 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1319 def : Pat<(v1f64 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1320 def : Pat<(v8i8 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1321 def : Pat<(v4i16 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1322 def : Pat<(v2i32 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1323 def : Pat<(v1i64 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1325 // Match all load 128 bits width whose type is compatible with FPR128
1326 def : Pat<(v4f32 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1327 def : Pat<(v2f64 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1328 def : Pat<(v16i8 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1329 def : Pat<(v8i16 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1330 def : Pat<(v4i32 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1331 def : Pat<(v2i64 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1332 def : Pat<(f128 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1335 def : Pat<(i32 (extloadi16 am_unscaled16:$addr)), (LDURHHi am_unscaled16:$addr)>;
1336 def : Pat<(i32 (extloadi8 am_unscaled8:$addr)), (LDURBBi am_unscaled8:$addr)>;
1337 def : Pat<(i32 (extloadi1 am_unscaled8:$addr)), (LDURBBi am_unscaled8:$addr)>;
1338 def : Pat<(i64 (extloadi32 am_unscaled32:$addr)),
1339 (SUBREG_TO_REG (i64 0), (LDURWi am_unscaled32:$addr), sub_32)>;
1340 def : Pat<(i64 (extloadi16 am_unscaled16:$addr)),
1341 (SUBREG_TO_REG (i64 0), (LDURHHi am_unscaled16:$addr), sub_32)>;
1342 def : Pat<(i64 (extloadi8 am_unscaled8:$addr)),
1343 (SUBREG_TO_REG (i64 0), (LDURBBi am_unscaled8:$addr), sub_32)>;
1344 def : Pat<(i64 (extloadi1 am_unscaled8:$addr)),
1345 (SUBREG_TO_REG (i64 0), (LDURBBi am_unscaled8:$addr), sub_32)>;
1347 def : Pat<(i32 (zextloadi16 am_unscaled16:$addr)),
1348 (LDURHHi am_unscaled16:$addr)>;
1349 def : Pat<(i32 (zextloadi8 am_unscaled8:$addr)),
1350 (LDURBBi am_unscaled8:$addr)>;
1351 def : Pat<(i32 (zextloadi1 am_unscaled8:$addr)),
1352 (LDURBBi am_unscaled8:$addr)>;
1353 def : Pat<(i64 (zextloadi32 am_unscaled32:$addr)),
1354 (SUBREG_TO_REG (i64 0), (LDURWi am_unscaled32:$addr), sub_32)>;
1355 def : Pat<(i64 (zextloadi16 am_unscaled16:$addr)),
1356 (SUBREG_TO_REG (i64 0), (LDURHHi am_unscaled16:$addr), sub_32)>;
1357 def : Pat<(i64 (zextloadi8 am_unscaled8:$addr)),
1358 (SUBREG_TO_REG (i64 0), (LDURBBi am_unscaled8:$addr), sub_32)>;
1359 def : Pat<(i64 (zextloadi1 am_unscaled8:$addr)),
1360 (SUBREG_TO_REG (i64 0), (LDURBBi am_unscaled8:$addr), sub_32)>;
1364 // LDR mnemonics fall back to LDUR for negative or unaligned offsets.
1366 // Define new assembler match classes as we want to only match these when
1367 // the don't otherwise match the scaled addressing mode for LDR/STR. Don't
1368 // associate a DiagnosticType either, as we want the diagnostic for the
1369 // canonical form (the scaled operand) to take precedence.
1370 def MemoryUnscaledFB8Operand : AsmOperandClass {
1371 let Name = "MemoryUnscaledFB8";
1372 let RenderMethod = "addMemoryUnscaledOperands";
1374 def MemoryUnscaledFB16Operand : AsmOperandClass {
1375 let Name = "MemoryUnscaledFB16";
1376 let RenderMethod = "addMemoryUnscaledOperands";
1378 def MemoryUnscaledFB32Operand : AsmOperandClass {
1379 let Name = "MemoryUnscaledFB32";
1380 let RenderMethod = "addMemoryUnscaledOperands";
1382 def MemoryUnscaledFB64Operand : AsmOperandClass {
1383 let Name = "MemoryUnscaledFB64";
1384 let RenderMethod = "addMemoryUnscaledOperands";
1386 def MemoryUnscaledFB128Operand : AsmOperandClass {
1387 let Name = "MemoryUnscaledFB128";
1388 let RenderMethod = "addMemoryUnscaledOperands";
1390 def am_unscaled_fb8 : Operand<i64> {
1391 let ParserMatchClass = MemoryUnscaledFB8Operand;
1392 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
1394 def am_unscaled_fb16 : Operand<i64> {
1395 let ParserMatchClass = MemoryUnscaledFB16Operand;
1396 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
1398 def am_unscaled_fb32 : Operand<i64> {
1399 let ParserMatchClass = MemoryUnscaledFB32Operand;
1400 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
1402 def am_unscaled_fb64 : Operand<i64> {
1403 let ParserMatchClass = MemoryUnscaledFB64Operand;
1404 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
1406 def am_unscaled_fb128 : Operand<i64> {
1407 let ParserMatchClass = MemoryUnscaledFB128Operand;
1408 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
1410 def : InstAlias<"ldr $Rt, $addr", (LDURXi GPR64:$Rt, am_unscaled_fb64:$addr)>;
1411 def : InstAlias<"ldr $Rt, $addr", (LDURWi GPR32:$Rt, am_unscaled_fb32:$addr)>;
1412 def : InstAlias<"ldr $Rt, $addr", (LDURBi FPR8:$Rt, am_unscaled_fb8:$addr)>;
1413 def : InstAlias<"ldr $Rt, $addr", (LDURHi FPR16:$Rt, am_unscaled_fb16:$addr)>;
1414 def : InstAlias<"ldr $Rt, $addr", (LDURSi FPR32:$Rt, am_unscaled_fb32:$addr)>;
1415 def : InstAlias<"ldr $Rt, $addr", (LDURDi FPR64:$Rt, am_unscaled_fb64:$addr)>;
1416 def : InstAlias<"ldr $Rt, $addr", (LDURQi FPR128:$Rt, am_unscaled_fb128:$addr)>;
1419 def : Pat<(i64 (zextloadi8 am_unscaled8:$addr)),
1420 (SUBREG_TO_REG (i64 0), (LDURBBi am_unscaled8:$addr), sub_32)>;
1421 def : Pat<(i64 (zextloadi16 am_unscaled16:$addr)),
1422 (SUBREG_TO_REG (i64 0), (LDURHHi am_unscaled16:$addr), sub_32)>;
1424 // load sign-extended half-word
1426 : LoadUnscaled<0b01, 0, 0b11, GPR32, am_unscaled16, "ldursh",
1427 [(set GPR32:$Rt, (sextloadi16 am_unscaled16:$addr))]>;
1429 : LoadUnscaled<0b01, 0, 0b10, GPR64, am_unscaled16, "ldursh",
1430 [(set GPR64:$Rt, (sextloadi16 am_unscaled16:$addr))]>;
1432 // load sign-extended byte
1434 : LoadUnscaled<0b00, 0, 0b11, GPR32, am_unscaled8, "ldursb",
1435 [(set GPR32:$Rt, (sextloadi8 am_unscaled8:$addr))]>;
1437 : LoadUnscaled<0b00, 0, 0b10, GPR64, am_unscaled8, "ldursb",
1438 [(set GPR64:$Rt, (sextloadi8 am_unscaled8:$addr))]>;
1440 // load sign-extended word
1442 : LoadUnscaled<0b10, 0, 0b10, GPR64, am_unscaled32, "ldursw",
1443 [(set GPR64:$Rt, (sextloadi32 am_unscaled32:$addr))]>;
1445 // zero and sign extending aliases from generic LDR* mnemonics to LDUR*.
1446 def : InstAlias<"ldrb $Rt, $addr", (LDURBBi GPR32:$Rt, am_unscaled_fb8:$addr)>;
1447 def : InstAlias<"ldrh $Rt, $addr", (LDURHHi GPR32:$Rt, am_unscaled_fb16:$addr)>;
1448 def : InstAlias<"ldrsb $Rt, $addr", (LDURSBWi GPR32:$Rt, am_unscaled_fb8:$addr)>;
1449 def : InstAlias<"ldrsb $Rt, $addr", (LDURSBXi GPR64:$Rt, am_unscaled_fb8:$addr)>;
1450 def : InstAlias<"ldrsh $Rt, $addr", (LDURSHWi GPR32:$Rt, am_unscaled_fb16:$addr)>;
1451 def : InstAlias<"ldrsh $Rt, $addr", (LDURSHXi GPR64:$Rt, am_unscaled_fb16:$addr)>;
1452 def : InstAlias<"ldrsw $Rt, $addr", (LDURSWi GPR64:$Rt, am_unscaled_fb32:$addr)>;
1455 def PRFUMi : PrefetchUnscaled<0b11, 0, 0b10, "prfum",
1456 [(ARM64Prefetch imm:$Rt, am_unscaled64:$addr)]>;
1459 // (unscaled immediate, unprivileged)
1460 def LDTRXi : LoadUnprivileged<0b11, 0, 0b01, GPR64, "ldtr">;
1461 def LDTRWi : LoadUnprivileged<0b10, 0, 0b01, GPR32, "ldtr">;
1463 def LDTRHi : LoadUnprivileged<0b01, 0, 0b01, GPR32, "ldtrh">;
1464 def LDTRBi : LoadUnprivileged<0b00, 0, 0b01, GPR32, "ldtrb">;
1466 // load sign-extended half-word
1467 def LDTRSHWi : LoadUnprivileged<0b01, 0, 0b11, GPR32, "ldtrsh">;
1468 def LDTRSHXi : LoadUnprivileged<0b01, 0, 0b10, GPR64, "ldtrsh">;
1470 // load sign-extended byte
1471 def LDTRSBWi : LoadUnprivileged<0b00, 0, 0b11, GPR32, "ldtrsb">;
1472 def LDTRSBXi : LoadUnprivileged<0b00, 0, 0b10, GPR64, "ldtrsb">;
1474 // load sign-extended word
1475 def LDTRSWi : LoadUnprivileged<0b10, 0, 0b10, GPR64, "ldtrsw">;
1478 // (immediate pre-indexed)
1479 def LDRWpre : LoadPreIdx<0b10, 0, 0b01, GPR32, "ldr">;
1480 def LDRXpre : LoadPreIdx<0b11, 0, 0b01, GPR64, "ldr">;
1481 def LDRBpre : LoadPreIdx<0b00, 1, 0b01, FPR8, "ldr">;
1482 def LDRHpre : LoadPreIdx<0b01, 1, 0b01, FPR16, "ldr">;
1483 def LDRSpre : LoadPreIdx<0b10, 1, 0b01, FPR32, "ldr">;
1484 def LDRDpre : LoadPreIdx<0b11, 1, 0b01, FPR64, "ldr">;
1485 def LDRQpre : LoadPreIdx<0b00, 1, 0b11, FPR128, "ldr">;
1487 // load sign-extended half-word
1488 def LDRSHWpre : LoadPreIdx<0b01, 0, 0b11, GPR32, "ldrsh">;
1489 def LDRSHXpre : LoadPreIdx<0b01, 0, 0b10, GPR64, "ldrsh">;
1491 // load sign-extended byte
1492 def LDRSBWpre : LoadPreIdx<0b00, 0, 0b11, GPR32, "ldrsb">;
1493 def LDRSBXpre : LoadPreIdx<0b00, 0, 0b10, GPR64, "ldrsb">;
1495 // load zero-extended byte
1496 def LDRBBpre : LoadPreIdx<0b00, 0, 0b01, GPR32, "ldrb">;
1497 def LDRHHpre : LoadPreIdx<0b01, 0, 0b01, GPR32, "ldrh">;
1499 // load sign-extended word
1500 def LDRSWpre : LoadPreIdx<0b10, 0, 0b10, GPR64, "ldrsw">;
1502 // ISel pseudos and patterns. See expanded comment on LoadPreIdxPseudo.
1503 def LDRDpre_isel : LoadPreIdxPseudo<FPR64>;
1504 def LDRSpre_isel : LoadPreIdxPseudo<FPR32>;
1505 def LDRXpre_isel : LoadPreIdxPseudo<GPR64>;
1506 def LDRWpre_isel : LoadPreIdxPseudo<GPR32>;
1507 def LDRHHpre_isel : LoadPreIdxPseudo<GPR32>;
1508 def LDRBBpre_isel : LoadPreIdxPseudo<GPR32>;
1510 def LDRSWpre_isel : LoadPreIdxPseudo<GPR64>;
1511 def LDRSHWpre_isel : LoadPreIdxPseudo<GPR32>;
1512 def LDRSHXpre_isel : LoadPreIdxPseudo<GPR64>;
1513 def LDRSBWpre_isel : LoadPreIdxPseudo<GPR32>;
1514 def LDRSBXpre_isel : LoadPreIdxPseudo<GPR64>;
1517 // (immediate post-indexed)
1518 def LDRWpost : LoadPostIdx<0b10, 0, 0b01, GPR32, "ldr">;
1519 def LDRXpost : LoadPostIdx<0b11, 0, 0b01, GPR64, "ldr">;
1520 def LDRBpost : LoadPostIdx<0b00, 1, 0b01, FPR8, "ldr">;
1521 def LDRHpost : LoadPostIdx<0b01, 1, 0b01, FPR16, "ldr">;
1522 def LDRSpost : LoadPostIdx<0b10, 1, 0b01, FPR32, "ldr">;
1523 def LDRDpost : LoadPostIdx<0b11, 1, 0b01, FPR64, "ldr">;
1524 def LDRQpost : LoadPostIdx<0b00, 1, 0b11, FPR128, "ldr">;
1526 // load sign-extended half-word
1527 def LDRSHWpost : LoadPostIdx<0b01, 0, 0b11, GPR32, "ldrsh">;
1528 def LDRSHXpost : LoadPostIdx<0b01, 0, 0b10, GPR64, "ldrsh">;
1530 // load sign-extended byte
1531 def LDRSBWpost : LoadPostIdx<0b00, 0, 0b11, GPR32, "ldrsb">;
1532 def LDRSBXpost : LoadPostIdx<0b00, 0, 0b10, GPR64, "ldrsb">;
1534 // load zero-extended byte
1535 def LDRBBpost : LoadPostIdx<0b00, 0, 0b01, GPR32, "ldrb">;
1536 def LDRHHpost : LoadPostIdx<0b01, 0, 0b01, GPR32, "ldrh">;
1538 // load sign-extended word
1539 def LDRSWpost : LoadPostIdx<0b10, 0, 0b10, GPR64, "ldrsw">;
1541 // ISel pseudos and patterns. See expanded comment on LoadPostIdxPseudo.
1542 def LDRDpost_isel : LoadPostIdxPseudo<FPR64>;
1543 def LDRSpost_isel : LoadPostIdxPseudo<FPR32>;
1544 def LDRXpost_isel : LoadPostIdxPseudo<GPR64>;
1545 def LDRWpost_isel : LoadPostIdxPseudo<GPR32>;
1546 def LDRHHpost_isel : LoadPostIdxPseudo<GPR32>;
1547 def LDRBBpost_isel : LoadPostIdxPseudo<GPR32>;
1549 def LDRSWpost_isel : LoadPostIdxPseudo<GPR64>;
1550 def LDRSHWpost_isel : LoadPostIdxPseudo<GPR32>;
1551 def LDRSHXpost_isel : LoadPostIdxPseudo<GPR64>;
1552 def LDRSBWpost_isel : LoadPostIdxPseudo<GPR32>;
1553 def LDRSBXpost_isel : LoadPostIdxPseudo<GPR64>;
1555 //===----------------------------------------------------------------------===//
1556 // Store instructions.
1557 //===----------------------------------------------------------------------===//
1559 // Pair (indexed, offset)
1560 // FIXME: Use dedicated range-checked addressing mode operand here.
1561 def STPWi : StorePairOffset<0b00, 0, GPR32, am_indexed32simm7, "stp">;
1562 def STPXi : StorePairOffset<0b10, 0, GPR64, am_indexed64simm7, "stp">;
1563 def STPSi : StorePairOffset<0b00, 1, FPR32, am_indexed32simm7, "stp">;
1564 def STPDi : StorePairOffset<0b01, 1, FPR64, am_indexed64simm7, "stp">;
1565 def STPQi : StorePairOffset<0b10, 1, FPR128, am_indexed128simm7, "stp">;
1567 // Pair (pre-indexed)
1568 def STPWpre : StorePairPreIdx<0b00, 0, GPR32, am_indexed32simm7_wb, "stp">;
1569 def STPXpre : StorePairPreIdx<0b10, 0, GPR64, am_indexed64simm7_wb, "stp">;
1570 def STPSpre : StorePairPreIdx<0b00, 1, FPR32, am_indexed32simm7_wb, "stp">;
1571 def STPDpre : StorePairPreIdx<0b01, 1, FPR64, am_indexed64simm7_wb, "stp">;
1572 def STPQpre : StorePairPreIdx<0b10, 1, FPR128, am_indexed128simm7_wb, "stp">;
1574 // Pair (pre-indexed)
1575 def STPWpost : StorePairPostIdx<0b00, 0, GPR32, simm7s4, "stp">;
1576 def STPXpost : StorePairPostIdx<0b10, 0, GPR64, simm7s8, "stp">;
1577 def STPSpost : StorePairPostIdx<0b00, 1, FPR32, simm7s4, "stp">;
1578 def STPDpost : StorePairPostIdx<0b01, 1, FPR64, simm7s8, "stp">;
1579 def STPQpost : StorePairPostIdx<0b10, 1, FPR128, simm7s16, "stp">;
1581 // Pair (no allocate)
1582 def STNPWi : StorePairNoAlloc<0b00, 0, GPR32, am_indexed32simm7, "stnp">;
1583 def STNPXi : StorePairNoAlloc<0b10, 0, GPR64, am_indexed64simm7, "stnp">;
1584 def STNPSi : StorePairNoAlloc<0b00, 1, FPR32, am_indexed32simm7, "stnp">;
1585 def STNPDi : StorePairNoAlloc<0b01, 1, FPR64, am_indexed64simm7, "stnp">;
1586 def STNPQi : StorePairNoAlloc<0b10, 1, FPR128, am_indexed128simm7, "stnp">;
1589 // (Register offset)
1591 let AddedComplexity = 10 in {
1594 def STRHHro : Store16RO<0b01, 0, 0b00, GPR32, "strh",
1595 [(truncstorei16 GPR32:$Rt, ro_indexed16:$addr)]>;
1596 def STRBBro : Store8RO<0b00, 0, 0b00, GPR32, "strb",
1597 [(truncstorei8 GPR32:$Rt, ro_indexed8:$addr)]>;
1598 def STRWro : Store32RO<0b10, 0, 0b00, GPR32, "str",
1599 [(store GPR32:$Rt, ro_indexed32:$addr)]>;
1600 def STRXro : Store64RO<0b11, 0, 0b00, GPR64, "str",
1601 [(store GPR64:$Rt, ro_indexed64:$addr)]>;
1604 def : Pat<(truncstorei8 GPR64:$Rt, ro_indexed8:$addr),
1605 (STRBBro (EXTRACT_SUBREG GPR64:$Rt, sub_32), ro_indexed8:$addr)>;
1606 def : Pat<(truncstorei16 GPR64:$Rt, ro_indexed16:$addr),
1607 (STRHHro (EXTRACT_SUBREG GPR64:$Rt, sub_32), ro_indexed16:$addr)>;
1608 def : Pat<(truncstorei32 GPR64:$Rt, ro_indexed32:$addr),
1609 (STRWro (EXTRACT_SUBREG GPR64:$Rt, sub_32), ro_indexed32:$addr)>;
1613 def STRBro : Store8RO<0b00, 1, 0b00, FPR8, "str",
1614 [(store FPR8:$Rt, ro_indexed8:$addr)]>;
1615 def STRHro : Store16RO<0b01, 1, 0b00, FPR16, "str",
1616 [(store (f16 FPR16:$Rt), ro_indexed16:$addr)]>;
1617 def STRSro : Store32RO<0b10, 1, 0b00, FPR32, "str",
1618 [(store (f32 FPR32:$Rt), ro_indexed32:$addr)]>;
1619 def STRDro : Store64RO<0b11, 1, 0b00, FPR64, "str",
1620 [(store (f64 FPR64:$Rt), ro_indexed64:$addr)]>;
1621 def STRQro : Store128RO<0b00, 1, 0b10, FPR128, "str", []> {
1625 // Match all store 64 bits width whose type is compatible with FPR64
1626 def : Pat<(store (v2f32 FPR64:$Rn), ro_indexed64:$addr),
1627 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1628 def : Pat<(store (v1f64 FPR64:$Rn), ro_indexed64:$addr),
1629 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1630 def : Pat<(store (v8i8 FPR64:$Rn), ro_indexed64:$addr),
1631 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1632 def : Pat<(store (v4i16 FPR64:$Rn), ro_indexed64:$addr),
1633 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1634 def : Pat<(store (v2i32 FPR64:$Rn), ro_indexed64:$addr),
1635 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1636 def : Pat<(store (v1i64 FPR64:$Rn), ro_indexed64:$addr),
1637 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1639 // Match all store 128 bits width whose type is compatible with FPR128
1640 def : Pat<(store (v4f32 FPR128:$Rn), ro_indexed128:$addr),
1641 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1642 def : Pat<(store (v2f64 FPR128:$Rn), ro_indexed128:$addr),
1643 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1644 def : Pat<(store (v16i8 FPR128:$Rn), ro_indexed128:$addr),
1645 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1646 def : Pat<(store (v8i16 FPR128:$Rn), ro_indexed128:$addr),
1647 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1648 def : Pat<(store (v4i32 FPR128:$Rn), ro_indexed128:$addr),
1649 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1650 def : Pat<(store (v2i64 FPR128:$Rn), ro_indexed128:$addr),
1651 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1652 def : Pat<(store (f128 FPR128:$Rn), ro_indexed128:$addr),
1653 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1656 // (unsigned immediate)
1657 def STRXui : StoreUI<0b11, 0, 0b00, GPR64, am_indexed64, "str",
1658 [(store GPR64:$Rt, am_indexed64:$addr)]>;
1659 def STRWui : StoreUI<0b10, 0, 0b00, GPR32, am_indexed32, "str",
1660 [(store GPR32:$Rt, am_indexed32:$addr)]>;
1661 def STRBui : StoreUI<0b00, 1, 0b00, FPR8, am_indexed8, "str",
1662 [(store FPR8:$Rt, am_indexed8:$addr)]>;
1663 def STRHui : StoreUI<0b01, 1, 0b00, FPR16, am_indexed16, "str",
1664 [(store (f16 FPR16:$Rt), am_indexed16:$addr)]>;
1665 def STRSui : StoreUI<0b10, 1, 0b00, FPR32, am_indexed32, "str",
1666 [(store (f32 FPR32:$Rt), am_indexed32:$addr)]>;
1667 def STRDui : StoreUI<0b11, 1, 0b00, FPR64, am_indexed64, "str",
1668 [(store (f64 FPR64:$Rt), am_indexed64:$addr)]>;
1669 def STRQui : StoreUI<0b00, 1, 0b10, FPR128, am_indexed128, "str", []> {
1673 // Match all store 64 bits width whose type is compatible with FPR64
1674 def : Pat<(store (v2f32 FPR64:$Rn), am_indexed64:$addr),
1675 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1676 def : Pat<(store (v1f64 FPR64:$Rn), am_indexed64:$addr),
1677 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1678 def : Pat<(store (v8i8 FPR64:$Rn), am_indexed64:$addr),
1679 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1680 def : Pat<(store (v4i16 FPR64:$Rn), am_indexed64:$addr),
1681 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1682 def : Pat<(store (v2i32 FPR64:$Rn), am_indexed64:$addr),
1683 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1684 def : Pat<(store (v1i64 FPR64:$Rn), am_indexed64:$addr),
1685 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1687 // Match all store 128 bits width whose type is compatible with FPR128
1688 def : Pat<(store (v4f32 FPR128:$Rn), am_indexed128:$addr),
1689 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1690 def : Pat<(store (v2f64 FPR128:$Rn), am_indexed128:$addr),
1691 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1692 def : Pat<(store (v16i8 FPR128:$Rn), am_indexed128:$addr),
1693 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1694 def : Pat<(store (v8i16 FPR128:$Rn), am_indexed128:$addr),
1695 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1696 def : Pat<(store (v4i32 FPR128:$Rn), am_indexed128:$addr),
1697 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1698 def : Pat<(store (v2i64 FPR128:$Rn), am_indexed128:$addr),
1699 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1700 def : Pat<(store (f128 FPR128:$Rn), am_indexed128:$addr),
1701 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1703 def STRHHui : StoreUI<0b01, 0, 0b00, GPR32, am_indexed16, "strh",
1704 [(truncstorei16 GPR32:$Rt, am_indexed16:$addr)]>;
1705 def STRBBui : StoreUI<0b00, 0, 0b00, GPR32, am_indexed8, "strb",
1706 [(truncstorei8 GPR32:$Rt, am_indexed8:$addr)]>;
1709 def : Pat<(truncstorei32 GPR64:$Rt, am_indexed32:$addr),
1710 (STRWui (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_indexed32:$addr)>;
1711 def : Pat<(truncstorei16 GPR64:$Rt, am_indexed16:$addr),
1712 (STRHHui (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_indexed16:$addr)>;
1713 def : Pat<(truncstorei8 GPR64:$Rt, am_indexed8:$addr),
1714 (STRBBui (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_indexed8:$addr)>;
1716 } // AddedComplexity = 10
1719 // (unscaled immediate)
1720 def STURXi : StoreUnscaled<0b11, 0, 0b00, GPR64, am_unscaled64, "stur",
1721 [(store GPR64:$Rt, am_unscaled64:$addr)]>;
1722 def STURWi : StoreUnscaled<0b10, 0, 0b00, GPR32, am_unscaled32, "stur",
1723 [(store GPR32:$Rt, am_unscaled32:$addr)]>;
1724 def STURBi : StoreUnscaled<0b00, 1, 0b00, FPR8, am_unscaled8, "stur",
1725 [(store FPR8:$Rt, am_unscaled8:$addr)]>;
1726 def STURHi : StoreUnscaled<0b01, 1, 0b00, FPR16, am_unscaled16, "stur",
1727 [(store (f16 FPR16:$Rt), am_unscaled16:$addr)]>;
1728 def STURSi : StoreUnscaled<0b10, 1, 0b00, FPR32, am_unscaled32, "stur",
1729 [(store (f32 FPR32:$Rt), am_unscaled32:$addr)]>;
1730 def STURDi : StoreUnscaled<0b11, 1, 0b00, FPR64, am_unscaled64, "stur",
1731 [(store (f64 FPR64:$Rt), am_unscaled64:$addr)]>;
1732 def STURQi : StoreUnscaled<0b00, 1, 0b10, FPR128, am_unscaled128, "stur",
1733 [(store (v2f64 FPR128:$Rt), am_unscaled128:$addr)]>;
1734 def STURHHi : StoreUnscaled<0b01, 0, 0b00, GPR32, am_unscaled16, "sturh",
1735 [(truncstorei16 GPR32:$Rt, am_unscaled16:$addr)]>;
1736 def STURBBi : StoreUnscaled<0b00, 0, 0b00, GPR32, am_unscaled8, "sturb",
1737 [(truncstorei8 GPR32:$Rt, am_unscaled8:$addr)]>;
1739 // Match all store 64 bits width whose type is compatible with FPR64
1740 def : Pat<(store (v2f32 FPR64:$Rn), am_unscaled64:$addr),
1741 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1742 def : Pat<(store (v1f64 FPR64:$Rn), am_unscaled64:$addr),
1743 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1744 def : Pat<(store (v8i8 FPR64:$Rn), am_unscaled64:$addr),
1745 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1746 def : Pat<(store (v4i16 FPR64:$Rn), am_unscaled64:$addr),
1747 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1748 def : Pat<(store (v2i32 FPR64:$Rn), am_unscaled64:$addr),
1749 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1750 def : Pat<(store (v1i64 FPR64:$Rn), am_unscaled64:$addr),
1751 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1753 // Match all store 128 bits width whose type is compatible with FPR128
1754 def : Pat<(store (v4f32 FPR128:$Rn), am_unscaled128:$addr),
1755 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1756 def : Pat<(store (v2f64 FPR128:$Rn), am_unscaled128:$addr),
1757 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1758 def : Pat<(store (v16i8 FPR128:$Rn), am_unscaled128:$addr),
1759 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1760 def : Pat<(store (v8i16 FPR128:$Rn), am_unscaled128:$addr),
1761 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1762 def : Pat<(store (v4i32 FPR128:$Rn), am_unscaled128:$addr),
1763 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1764 def : Pat<(store (v2i64 FPR128:$Rn), am_unscaled128:$addr),
1765 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1766 def : Pat<(store (f128 FPR128:$Rn), am_unscaled128:$addr),
1767 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1769 // unscaled i64 truncating stores
1770 def : Pat<(truncstorei32 GPR64:$Rt, am_unscaled32:$addr),
1771 (STURWi (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_unscaled32:$addr)>;
1772 def : Pat<(truncstorei16 GPR64:$Rt, am_unscaled16:$addr),
1773 (STURHHi (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_unscaled16:$addr)>;
1774 def : Pat<(truncstorei8 GPR64:$Rt, am_unscaled8:$addr),
1775 (STURBBi (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_unscaled8:$addr)>;
1778 // STR mnemonics fall back to STUR for negative or unaligned offsets.
1779 def : InstAlias<"str $Rt, $addr", (STURXi GPR64:$Rt, am_unscaled_fb64:$addr)>;
1780 def : InstAlias<"str $Rt, $addr", (STURWi GPR32:$Rt, am_unscaled_fb32:$addr)>;
1781 def : InstAlias<"str $Rt, $addr", (STURBi FPR8:$Rt, am_unscaled_fb8:$addr)>;
1782 def : InstAlias<"str $Rt, $addr", (STURHi FPR16:$Rt, am_unscaled_fb16:$addr)>;
1783 def : InstAlias<"str $Rt, $addr", (STURSi FPR32:$Rt, am_unscaled_fb32:$addr)>;
1784 def : InstAlias<"str $Rt, $addr", (STURDi FPR64:$Rt, am_unscaled_fb64:$addr)>;
1785 def : InstAlias<"str $Rt, $addr", (STURQi FPR128:$Rt, am_unscaled_fb128:$addr)>;
1787 def : InstAlias<"strb $Rt, $addr", (STURBBi GPR32:$Rt, am_unscaled_fb8:$addr)>;
1788 def : InstAlias<"strh $Rt, $addr", (STURHHi GPR32:$Rt, am_unscaled_fb16:$addr)>;
1791 // (unscaled immediate, unprivileged)
1792 def STTRWi : StoreUnprivileged<0b10, 0, 0b00, GPR32, "sttr">;
1793 def STTRXi : StoreUnprivileged<0b11, 0, 0b00, GPR64, "sttr">;
1795 def STTRHi : StoreUnprivileged<0b01, 0, 0b00, GPR32, "sttrh">;
1796 def STTRBi : StoreUnprivileged<0b00, 0, 0b00, GPR32, "sttrb">;
1799 // (immediate pre-indexed)
1800 def STRWpre : StorePreIdx<0b10, 0, 0b00, GPR32, "str">;
1801 def STRXpre : StorePreIdx<0b11, 0, 0b00, GPR64, "str">;
1802 def STRBpre : StorePreIdx<0b00, 1, 0b00, FPR8, "str">;
1803 def STRHpre : StorePreIdx<0b01, 1, 0b00, FPR16, "str">;
1804 def STRSpre : StorePreIdx<0b10, 1, 0b00, FPR32, "str">;
1805 def STRDpre : StorePreIdx<0b11, 1, 0b00, FPR64, "str">;
1806 def STRQpre : StorePreIdx<0b00, 1, 0b10, FPR128, "str">;
1808 def STRBBpre : StorePreIdx<0b00, 0, 0b00, GPR32, "strb">;
1809 def STRHHpre : StorePreIdx<0b01, 0, 0b00, GPR32, "strh">;
1811 // ISel pseudos and patterns. See expanded comment on StorePreIdxPseudo.
1812 defm STRDpre : StorePreIdxPseudo<FPR64, f64, pre_store>;
1813 defm STRSpre : StorePreIdxPseudo<FPR32, f32, pre_store>;
1814 defm STRXpre : StorePreIdxPseudo<GPR64, i64, pre_store>;
1815 defm STRWpre : StorePreIdxPseudo<GPR32, i32, pre_store>;
1816 defm STRHHpre : StorePreIdxPseudo<GPR32, i32, pre_truncsti16>;
1817 defm STRBBpre : StorePreIdxPseudo<GPR32, i32, pre_truncsti8>;
1819 def : Pat<(pre_truncsti32 GPR64:$Rt, am_noindex:$addr, simm9:$off),
1820 (STRWpre_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
1822 def : Pat<(pre_truncsti16 GPR64:$Rt, am_noindex:$addr, simm9:$off),
1823 (STRHHpre_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
1825 def : Pat<(pre_truncsti8 GPR64:$Rt, am_noindex:$addr, simm9:$off),
1826 (STRBBpre_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
1830 // (immediate post-indexed)
1831 def STRWpost : StorePostIdx<0b10, 0, 0b00, GPR32, "str">;
1832 def STRXpost : StorePostIdx<0b11, 0, 0b00, GPR64, "str">;
1833 def STRBpost : StorePostIdx<0b00, 1, 0b00, FPR8, "str">;
1834 def STRHpost : StorePostIdx<0b01, 1, 0b00, FPR16, "str">;
1835 def STRSpost : StorePostIdx<0b10, 1, 0b00, FPR32, "str">;
1836 def STRDpost : StorePostIdx<0b11, 1, 0b00, FPR64, "str">;
1837 def STRQpost : StorePostIdx<0b00, 1, 0b10, FPR128, "str">;
1839 def STRBBpost : StorePostIdx<0b00, 0, 0b00, GPR32, "strb">;
1840 def STRHHpost : StorePostIdx<0b01, 0, 0b00, GPR32, "strh">;
1842 // ISel pseudos and patterns. See expanded comment on StorePostIdxPseudo.
1843 defm STRDpost : StorePostIdxPseudo<FPR64, f64, post_store, STRDpost>;
1844 defm STRSpost : StorePostIdxPseudo<FPR32, f32, post_store, STRSpost>;
1845 defm STRXpost : StorePostIdxPseudo<GPR64, i64, post_store, STRXpost>;
1846 defm STRWpost : StorePostIdxPseudo<GPR32, i32, post_store, STRWpost>;
1847 defm STRHHpost : StorePostIdxPseudo<GPR32, i32, post_truncsti16, STRHHpost>;
1848 defm STRBBpost : StorePostIdxPseudo<GPR32, i32, post_truncsti8, STRBBpost>;
1850 def : Pat<(post_truncsti32 GPR64:$Rt, am_noindex:$addr, simm9:$off),
1851 (STRWpost_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
1853 def : Pat<(post_truncsti16 GPR64:$Rt, am_noindex:$addr, simm9:$off),
1854 (STRHHpost_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
1856 def : Pat<(post_truncsti8 GPR64:$Rt, am_noindex:$addr, simm9:$off),
1857 (STRBBpost_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
1861 //===----------------------------------------------------------------------===//
1862 // Load/store exclusive instructions.
1863 //===----------------------------------------------------------------------===//
1865 def LDARW : LoadAcquire <0b10, 1, 1, 0, 1, GPR32, "ldar">;
1866 def LDARX : LoadAcquire <0b11, 1, 1, 0, 1, GPR64, "ldar">;
1867 def LDARB : LoadAcquire <0b00, 1, 1, 0, 1, GPR32, "ldarb">;
1868 def LDARH : LoadAcquire <0b01, 1, 1, 0, 1, GPR32, "ldarh">;
1870 def LDAXRW : LoadExclusive <0b10, 0, 1, 0, 1, GPR32, "ldaxr">;
1871 def LDAXRX : LoadExclusive <0b11, 0, 1, 0, 1, GPR64, "ldaxr">;
1872 def LDAXRB : LoadExclusive <0b00, 0, 1, 0, 1, GPR32, "ldaxrb">;
1873 def LDAXRH : LoadExclusive <0b01, 0, 1, 0, 1, GPR32, "ldaxrh">;
1875 def LDXRW : LoadExclusive <0b10, 0, 1, 0, 0, GPR32, "ldxr">;
1876 def LDXRX : LoadExclusive <0b11, 0, 1, 0, 0, GPR64, "ldxr">;
1877 def LDXRB : LoadExclusive <0b00, 0, 1, 0, 0, GPR32, "ldxrb">;
1878 def LDXRH : LoadExclusive <0b01, 0, 1, 0, 0, GPR32, "ldxrh">;
1880 def STLRW : StoreRelease <0b10, 1, 0, 0, 1, GPR32, "stlr">;
1881 def STLRX : StoreRelease <0b11, 1, 0, 0, 1, GPR64, "stlr">;
1882 def STLRB : StoreRelease <0b00, 1, 0, 0, 1, GPR32, "stlrb">;
1883 def STLRH : StoreRelease <0b01, 1, 0, 0, 1, GPR32, "stlrh">;
1885 def STLXRW : StoreExclusive<0b10, 0, 0, 0, 1, GPR32, "stlxr">;
1886 def STLXRX : StoreExclusive<0b11, 0, 0, 0, 1, GPR64, "stlxr">;
1887 def STLXRB : StoreExclusive<0b00, 0, 0, 0, 1, GPR32, "stlxrb">;
1888 def STLXRH : StoreExclusive<0b01, 0, 0, 0, 1, GPR32, "stlxrh">;
1890 def STXRW : StoreExclusive<0b10, 0, 0, 0, 0, GPR32, "stxr">;
1891 def STXRX : StoreExclusive<0b11, 0, 0, 0, 0, GPR64, "stxr">;
1892 def STXRB : StoreExclusive<0b00, 0, 0, 0, 0, GPR32, "stxrb">;
1893 def STXRH : StoreExclusive<0b01, 0, 0, 0, 0, GPR32, "stxrh">;
1895 def LDAXPW : LoadExclusivePair<0b10, 0, 1, 1, 1, GPR32, "ldaxp">;
1896 def LDAXPX : LoadExclusivePair<0b11, 0, 1, 1, 1, GPR64, "ldaxp">;
1898 def LDXPW : LoadExclusivePair<0b10, 0, 1, 1, 0, GPR32, "ldxp">;
1899 def LDXPX : LoadExclusivePair<0b11, 0, 1, 1, 0, GPR64, "ldxp">;
1901 def STLXPW : StoreExclusivePair<0b10, 0, 0, 1, 1, GPR32, "stlxp">;
1902 def STLXPX : StoreExclusivePair<0b11, 0, 0, 1, 1, GPR64, "stlxp">;
1904 def STXPW : StoreExclusivePair<0b10, 0, 0, 1, 0, GPR32, "stxp">;
1905 def STXPX : StoreExclusivePair<0b11, 0, 0, 1, 0, GPR64, "stxp">;
1907 //===----------------------------------------------------------------------===//
1908 // Scaled floating point to integer conversion instructions.
1909 //===----------------------------------------------------------------------===//
1911 defm FCVTAS : FPToIntegerUnscaled<0b00, 0b100, "fcvtas", int_arm64_neon_fcvtas>;
1912 defm FCVTAU : FPToIntegerUnscaled<0b00, 0b101, "fcvtau", int_arm64_neon_fcvtau>;
1913 defm FCVTMS : FPToIntegerUnscaled<0b10, 0b000, "fcvtms", int_arm64_neon_fcvtms>;
1914 defm FCVTMU : FPToIntegerUnscaled<0b10, 0b001, "fcvtmu", int_arm64_neon_fcvtmu>;
1915 defm FCVTNS : FPToIntegerUnscaled<0b00, 0b000, "fcvtns", int_arm64_neon_fcvtns>;
1916 defm FCVTNU : FPToIntegerUnscaled<0b00, 0b001, "fcvtnu", int_arm64_neon_fcvtnu>;
1917 defm FCVTPS : FPToIntegerUnscaled<0b01, 0b000, "fcvtps", int_arm64_neon_fcvtps>;
1918 defm FCVTPU : FPToIntegerUnscaled<0b01, 0b001, "fcvtpu", int_arm64_neon_fcvtpu>;
1919 defm FCVTZS : FPToIntegerUnscaled<0b11, 0b000, "fcvtzs", fp_to_sint>;
1920 defm FCVTZU : FPToIntegerUnscaled<0b11, 0b001, "fcvtzu", fp_to_uint>;
1921 defm FCVTZS : FPToIntegerScaled<0b11, 0b000, "fcvtzs", fp_to_sint>;
1922 defm FCVTZU : FPToIntegerScaled<0b11, 0b001, "fcvtzu", fp_to_uint>;
1923 let isCodeGenOnly = 1 in {
1924 defm FCVTZS_Int : FPToIntegerUnscaled<0b11, 0b000, "fcvtzs", int_arm64_neon_fcvtzs>;
1925 defm FCVTZU_Int : FPToIntegerUnscaled<0b11, 0b001, "fcvtzu", int_arm64_neon_fcvtzu>;
1926 defm FCVTZS_Int : FPToIntegerScaled<0b11, 0b000, "fcvtzs", int_arm64_neon_fcvtzs>;
1927 defm FCVTZU_Int : FPToIntegerScaled<0b11, 0b001, "fcvtzu", int_arm64_neon_fcvtzu>;
1930 //===----------------------------------------------------------------------===//
1931 // Scaled integer to floating point conversion instructions.
1932 //===----------------------------------------------------------------------===//
1934 defm SCVTF : IntegerToFP<0, "scvtf", sint_to_fp>;
1935 defm UCVTF : IntegerToFP<1, "ucvtf", uint_to_fp>;
1937 //===----------------------------------------------------------------------===//
1938 // Unscaled integer to floating point conversion instruction.
1939 //===----------------------------------------------------------------------===//
1941 defm FMOV : UnscaledConversion<"fmov">;
1943 def : Pat<(f32 (fpimm0)), (FMOVWSr WZR)>, Requires<[NoZCZ]>;
1944 def : Pat<(f64 (fpimm0)), (FMOVXDr XZR)>, Requires<[NoZCZ]>;
1946 def : Pat<(v8i8 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1947 def : Pat<(v4i16 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1948 def : Pat<(v2i32 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1949 def : Pat<(v1i64 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1950 def : Pat<(v2f32 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1951 def : Pat<(v1f64 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1952 def : Pat<(v1i64 (scalar_to_vector GPR64:$Xn)),
1953 (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1954 def : Pat<(v1f64 (scalar_to_vector GPR64:$Xn)),
1955 (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1956 def : Pat<(v1f64 (scalar_to_vector (f64 FPR64:$Xn))), (v1f64 FPR64:$Xn)>;
1958 def : Pat<(i64 (bitconvert (v8i8 V64:$Vn))),
1959 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
1960 def : Pat<(i64 (bitconvert (v4i16 V64:$Vn))),
1961 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
1962 def : Pat<(i64 (bitconvert (v2i32 V64:$Vn))),
1963 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
1964 def : Pat<(i64 (bitconvert (v1i64 V64:$Vn))),
1965 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
1966 def : Pat<(i64 (bitconvert (v2f32 V64:$Vn))),
1967 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
1968 def : Pat<(i64 (bitconvert (v1f64 V64:$Vn))),
1969 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
1971 def : Pat<(f32 (bitconvert (i32 GPR32:$Xn))),
1972 (COPY_TO_REGCLASS GPR32:$Xn, FPR32)>;
1973 def : Pat<(i32 (bitconvert (f32 FPR32:$Xn))),
1974 (COPY_TO_REGCLASS FPR32:$Xn, GPR32)>;
1975 def : Pat<(f64 (bitconvert (i64 GPR64:$Xn))),
1976 (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1977 def : Pat<(i64 (bitconvert (f64 FPR64:$Xn))),
1978 (COPY_TO_REGCLASS FPR64:$Xn, GPR64)>;
1980 //===----------------------------------------------------------------------===//
1981 // Floating point conversion instruction.
1982 //===----------------------------------------------------------------------===//
1984 defm FCVT : FPConversion<"fcvt">;
1986 def : Pat<(f32_to_f16 FPR32:$Rn),
1987 (i32 (COPY_TO_REGCLASS
1988 (f32 (SUBREG_TO_REG (i32 0), (FCVTHSr FPR32:$Rn), hsub)),
1991 def FCVTSHpseudo : Pseudo<(outs FPR32:$Rd), (ins FPR32:$Rn),
1992 [(set (f32 FPR32:$Rd), (f16_to_f32 i32:$Rn))]>;
1994 //===----------------------------------------------------------------------===//
1995 // Floating point single operand instructions.
1996 //===----------------------------------------------------------------------===//
1998 defm FABS : SingleOperandFPData<0b0001, "fabs", fabs>;
1999 defm FMOV : SingleOperandFPData<0b0000, "fmov">;
2000 defm FNEG : SingleOperandFPData<0b0010, "fneg", fneg>;
2001 defm FRINTA : SingleOperandFPData<0b1100, "frinta", frnd>;
2002 defm FRINTI : SingleOperandFPData<0b1111, "frinti", fnearbyint>;
2003 defm FRINTM : SingleOperandFPData<0b1010, "frintm", ffloor>;
2004 defm FRINTN : SingleOperandFPData<0b1000, "frintn", int_arm64_neon_frintn>;
2005 defm FRINTP : SingleOperandFPData<0b1001, "frintp", fceil>;
2007 def : Pat<(v1f64 (int_arm64_neon_frintn (v1f64 FPR64:$Rn))),
2008 (FRINTNDr FPR64:$Rn)>;
2010 // FRINTX is inserted to set the flags as required by FENV_ACCESS ON behavior
2011 // in the C spec. Setting hasSideEffects ensures it is not DCE'd.
2012 // <rdar://problem/13715968>
2013 // TODO: We should really model the FPSR flags correctly. This is really ugly.
2014 let hasSideEffects = 1 in {
2015 defm FRINTX : SingleOperandFPData<0b1110, "frintx", frint>;
2018 defm FRINTZ : SingleOperandFPData<0b1011, "frintz", ftrunc>;
2020 let SchedRW = [WriteFDiv] in {
2021 defm FSQRT : SingleOperandFPData<0b0011, "fsqrt", fsqrt>;
2024 //===----------------------------------------------------------------------===//
2025 // Floating point two operand instructions.
2026 //===----------------------------------------------------------------------===//
2028 defm FADD : TwoOperandFPData<0b0010, "fadd", fadd>;
2029 let SchedRW = [WriteFDiv] in {
2030 defm FDIV : TwoOperandFPData<0b0001, "fdiv", fdiv>;
2032 defm FMAXNM : TwoOperandFPData<0b0110, "fmaxnm", int_arm64_neon_fmaxnm>;
2033 defm FMAX : TwoOperandFPData<0b0100, "fmax", ARM64fmax>;
2034 defm FMINNM : TwoOperandFPData<0b0111, "fminnm", int_arm64_neon_fminnm>;
2035 defm FMIN : TwoOperandFPData<0b0101, "fmin", ARM64fmin>;
2036 let SchedRW = [WriteFMul] in {
2037 defm FMUL : TwoOperandFPData<0b0000, "fmul", fmul>;
2038 defm FNMUL : TwoOperandFPDataNeg<0b1000, "fnmul", fmul>;
2040 defm FSUB : TwoOperandFPData<0b0011, "fsub", fsub>;
2042 def : Pat<(v1f64 (ARM64fmax (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2043 (FMAXDrr FPR64:$Rn, FPR64:$Rm)>;
2044 def : Pat<(v1f64 (ARM64fmin (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2045 (FMINDrr FPR64:$Rn, FPR64:$Rm)>;
2046 def : Pat<(v1f64 (int_arm64_neon_fmaxnm (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2047 (FMAXNMDrr FPR64:$Rn, FPR64:$Rm)>;
2048 def : Pat<(v1f64 (int_arm64_neon_fminnm (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2049 (FMINNMDrr FPR64:$Rn, FPR64:$Rm)>;
2051 //===----------------------------------------------------------------------===//
2052 // Floating point three operand instructions.
2053 //===----------------------------------------------------------------------===//
2055 defm FMADD : ThreeOperandFPData<0, 0, "fmadd", fma>;
2056 defm FMSUB : ThreeOperandFPData<0, 1, "fmsub",
2057 TriOpFrag<(fma node:$LHS, (fneg node:$MHS), node:$RHS)> >;
2058 defm FNMADD : ThreeOperandFPData<1, 0, "fnmadd",
2059 TriOpFrag<(fneg (fma node:$LHS, node:$MHS, node:$RHS))> >;
2060 defm FNMSUB : ThreeOperandFPData<1, 1, "fnmsub",
2061 TriOpFrag<(fma node:$LHS, node:$MHS, (fneg node:$RHS))> >;
2063 // The following def pats catch the case where the LHS of an FMA is negated.
2064 // The TriOpFrag above catches the case where the middle operand is negated.
2066 // N.b. FMSUB etc have the accumulator at the *end* of (outs), unlike
2067 // the NEON variant.
2068 def : Pat<(f32 (fma (fneg FPR32:$Rn), FPR32:$Rm, FPR32:$Ra)),
2069 (FMSUBSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
2071 def : Pat<(f64 (fma (fneg FPR64:$Rn), FPR64:$Rm, FPR64:$Ra)),
2072 (FMSUBDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
2074 // We handled -(a + b*c) for FNMADD above, now it's time for "(-a) + (-b)*c" and
2076 def : Pat<(f32 (fma (fneg FPR32:$Rn), FPR32:$Rm, (fneg FPR32:$Ra))),
2077 (FNMADDSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
2079 def : Pat<(f64 (fma (fneg FPR64:$Rn), FPR64:$Rm, (fneg FPR64:$Ra))),
2080 (FNMADDDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
2082 def : Pat<(f32 (fma FPR32:$Rn, (fneg FPR32:$Rm), (fneg FPR32:$Ra))),
2083 (FNMADDSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
2085 def : Pat<(f64 (fma FPR64:$Rn, (fneg FPR64:$Rm), (fneg FPR64:$Ra))),
2086 (FNMADDDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
2088 //===----------------------------------------------------------------------===//
2089 // Floating point comparison instructions.
2090 //===----------------------------------------------------------------------===//
2092 defm FCMPE : FPComparison<1, "fcmpe">;
2093 defm FCMP : FPComparison<0, "fcmp", ARM64fcmp>;
2095 //===----------------------------------------------------------------------===//
2096 // Floating point conditional comparison instructions.
2097 //===----------------------------------------------------------------------===//
2099 defm FCCMPE : FPCondComparison<1, "fccmpe">;
2100 defm FCCMP : FPCondComparison<0, "fccmp">;
2102 //===----------------------------------------------------------------------===//
2103 // Floating point conditional select instruction.
2104 //===----------------------------------------------------------------------===//
2106 defm FCSEL : FPCondSelect<"fcsel">;
2108 // CSEL instructions providing f128 types need to be handled by a
2109 // pseudo-instruction since the eventual code will need to introduce basic
2110 // blocks and control flow.
2111 def F128CSEL : Pseudo<(outs FPR128:$Rd),
2112 (ins FPR128:$Rn, FPR128:$Rm, ccode:$cond),
2113 [(set (f128 FPR128:$Rd),
2114 (ARM64csel FPR128:$Rn, FPR128:$Rm,
2115 (i32 imm:$cond), CPSR))]> {
2117 let usesCustomInserter = 1;
2121 //===----------------------------------------------------------------------===//
2122 // Floating point immediate move.
2123 //===----------------------------------------------------------------------===//
2125 let isReMaterializable = 1 in {
2126 defm FMOV : FPMoveImmediate<"fmov">;
2129 //===----------------------------------------------------------------------===//
2130 // Advanced SIMD two vector instructions.
2131 //===----------------------------------------------------------------------===//
2133 defm ABS : SIMDTwoVectorBHSD<0, 0b01011, "abs", int_arm64_neon_abs>;
2134 defm CLS : SIMDTwoVectorBHS<0, 0b00100, "cls", int_arm64_neon_cls>;
2135 defm CLZ : SIMDTwoVectorBHS<1, 0b00100, "clz", ctlz>;
2136 defm CMEQ : SIMDCmpTwoVector<0, 0b01001, "cmeq", ARM64cmeqz>;
2137 defm CMGE : SIMDCmpTwoVector<1, 0b01000, "cmge", ARM64cmgez>;
2138 defm CMGT : SIMDCmpTwoVector<0, 0b01000, "cmgt", ARM64cmgtz>;
2139 defm CMLE : SIMDCmpTwoVector<1, 0b01001, "cmle", ARM64cmlez>;
2140 defm CMLT : SIMDCmpTwoVector<0, 0b01010, "cmlt", ARM64cmltz>;
2141 defm CNT : SIMDTwoVectorB<0, 0b00, 0b00101, "cnt", ctpop>;
2142 defm FABS : SIMDTwoVectorFP<0, 1, 0b01111, "fabs", fabs>;
2144 defm FCMEQ : SIMDFPCmpTwoVector<0, 1, 0b01101, "fcmeq", ARM64fcmeqz>;
2145 defm FCMGE : SIMDFPCmpTwoVector<1, 1, 0b01100, "fcmge", ARM64fcmgez>;
2146 defm FCMGT : SIMDFPCmpTwoVector<0, 1, 0b01100, "fcmgt", ARM64fcmgtz>;
2147 defm FCMLE : SIMDFPCmpTwoVector<1, 1, 0b01101, "fcmle", ARM64fcmlez>;
2148 defm FCMLT : SIMDFPCmpTwoVector<0, 1, 0b01110, "fcmlt", ARM64fcmltz>;
2149 defm FCVTAS : SIMDTwoVectorFPToInt<0,0,0b11100, "fcvtas",int_arm64_neon_fcvtas>;
2150 defm FCVTAU : SIMDTwoVectorFPToInt<1,0,0b11100, "fcvtau",int_arm64_neon_fcvtau>;
2151 defm FCVTL : SIMDFPWidenTwoVector<0, 0, 0b10111, "fcvtl">;
2152 def : Pat<(v4f32 (int_arm64_neon_vcvthf2fp (v4i16 V64:$Rn))),
2153 (FCVTLv4i16 V64:$Rn)>;
2154 def : Pat<(v4f32 (int_arm64_neon_vcvthf2fp (extract_subvector (v8i16 V128:$Rn),
2156 (FCVTLv8i16 V128:$Rn)>;
2157 def : Pat<(v2f64 (fextend (v2f32 V64:$Rn))), (FCVTLv2i32 V64:$Rn)>;
2158 def : Pat<(v2f64 (fextend (v2f32 (extract_subvector (v4f32 V128:$Rn),
2160 (FCVTLv4i32 V128:$Rn)>;
2162 defm FCVTMS : SIMDTwoVectorFPToInt<0,0,0b11011, "fcvtms",int_arm64_neon_fcvtms>;
2163 defm FCVTMU : SIMDTwoVectorFPToInt<1,0,0b11011, "fcvtmu",int_arm64_neon_fcvtmu>;
2164 defm FCVTNS : SIMDTwoVectorFPToInt<0,0,0b11010, "fcvtns",int_arm64_neon_fcvtns>;
2165 defm FCVTNU : SIMDTwoVectorFPToInt<1,0,0b11010, "fcvtnu",int_arm64_neon_fcvtnu>;
2166 defm FCVTN : SIMDFPNarrowTwoVector<0, 0, 0b10110, "fcvtn">;
2167 def : Pat<(v4i16 (int_arm64_neon_vcvtfp2hf (v4f32 V128:$Rn))),
2168 (FCVTNv4i16 V128:$Rn)>;
2169 def : Pat<(concat_vectors V64:$Rd,
2170 (v4i16 (int_arm64_neon_vcvtfp2hf (v4f32 V128:$Rn)))),
2171 (FCVTNv8i16 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
2172 def : Pat<(v2f32 (fround (v2f64 V128:$Rn))), (FCVTNv2i32 V128:$Rn)>;
2173 def : Pat<(concat_vectors V64:$Rd, (v2f32 (fround (v2f64 V128:$Rn)))),
2174 (FCVTNv4i32 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
2175 defm FCVTPS : SIMDTwoVectorFPToInt<0,1,0b11010, "fcvtps",int_arm64_neon_fcvtps>;
2176 defm FCVTPU : SIMDTwoVectorFPToInt<1,1,0b11010, "fcvtpu",int_arm64_neon_fcvtpu>;
2177 defm FCVTXN : SIMDFPInexactCvtTwoVector<1, 0, 0b10110, "fcvtxn",
2178 int_arm64_neon_fcvtxn>;
2179 defm FCVTZS : SIMDTwoVectorFPToInt<0, 1, 0b11011, "fcvtzs", fp_to_sint>;
2180 defm FCVTZU : SIMDTwoVectorFPToInt<1, 1, 0b11011, "fcvtzu", fp_to_uint>;
2181 let isCodeGenOnly = 1 in {
2182 defm FCVTZS_Int : SIMDTwoVectorFPToInt<0, 1, 0b11011, "fcvtzs",
2183 int_arm64_neon_fcvtzs>;
2184 defm FCVTZU_Int : SIMDTwoVectorFPToInt<1, 1, 0b11011, "fcvtzu",
2185 int_arm64_neon_fcvtzu>;
2187 defm FNEG : SIMDTwoVectorFP<1, 1, 0b01111, "fneg", fneg>;
2188 defm FRECPE : SIMDTwoVectorFP<0, 1, 0b11101, "frecpe", int_arm64_neon_frecpe>;
2189 defm FRINTA : SIMDTwoVectorFP<1, 0, 0b11000, "frinta", frnd>;
2190 defm FRINTI : SIMDTwoVectorFP<1, 1, 0b11001, "frinti", fnearbyint>;
2191 defm FRINTM : SIMDTwoVectorFP<0, 0, 0b11001, "frintm", ffloor>;
2192 defm FRINTN : SIMDTwoVectorFP<0, 0, 0b11000, "frintn", int_arm64_neon_frintn>;
2193 defm FRINTP : SIMDTwoVectorFP<0, 1, 0b11000, "frintp", fceil>;
2194 defm FRINTX : SIMDTwoVectorFP<1, 0, 0b11001, "frintx", frint>;
2195 defm FRINTZ : SIMDTwoVectorFP<0, 1, 0b11001, "frintz", ftrunc>;
2196 defm FRSQRTE: SIMDTwoVectorFP<1, 1, 0b11101, "frsqrte", int_arm64_neon_frsqrte>;
2197 defm FSQRT : SIMDTwoVectorFP<1, 1, 0b11111, "fsqrt", fsqrt>;
2198 defm NEG : SIMDTwoVectorBHSD<1, 0b01011, "neg",
2199 UnOpFrag<(sub immAllZerosV, node:$LHS)> >;
2200 defm NOT : SIMDTwoVectorB<1, 0b00, 0b00101, "not", vnot>;
2201 // Aliases for MVN -> NOT.
2202 def : InstAlias<"mvn.8b $Vd, $Vn", (NOTv8i8 V64:$Vd, V64:$Vn)>;
2203 def : InstAlias<"mvn.16b $Vd, $Vn", (NOTv16i8 V128:$Vd, V128:$Vn)>;
2204 def : InstAlias<"mvn $Vd.8b, $Vn.8b", (NOTv8i8 V64:$Vd, V64:$Vn)>;
2205 def : InstAlias<"mvn $Vd.16b, $Vn.16b", (NOTv16i8 V128:$Vd, V128:$Vn)>;
2207 def : Pat<(ARM64neg (v8i8 V64:$Rn)), (NEGv8i8 V64:$Rn)>;
2208 def : Pat<(ARM64neg (v16i8 V128:$Rn)), (NEGv16i8 V128:$Rn)>;
2209 def : Pat<(ARM64neg (v4i16 V64:$Rn)), (NEGv4i16 V64:$Rn)>;
2210 def : Pat<(ARM64neg (v8i16 V128:$Rn)), (NEGv8i16 V128:$Rn)>;
2211 def : Pat<(ARM64neg (v2i32 V64:$Rn)), (NEGv2i32 V64:$Rn)>;
2212 def : Pat<(ARM64neg (v4i32 V128:$Rn)), (NEGv4i32 V128:$Rn)>;
2213 def : Pat<(ARM64neg (v2i64 V128:$Rn)), (NEGv2i64 V128:$Rn)>;
2215 def : Pat<(ARM64not (v8i8 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2216 def : Pat<(ARM64not (v16i8 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2217 def : Pat<(ARM64not (v4i16 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2218 def : Pat<(ARM64not (v8i16 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2219 def : Pat<(ARM64not (v2i32 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2220 def : Pat<(ARM64not (v1i64 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2221 def : Pat<(ARM64not (v4i32 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2222 def : Pat<(ARM64not (v2i64 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2224 def : Pat<(vnot (v4i16 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2225 def : Pat<(vnot (v8i16 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2226 def : Pat<(vnot (v2i32 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2227 def : Pat<(vnot (v4i32 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2228 def : Pat<(vnot (v2i64 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2230 defm RBIT : SIMDTwoVectorB<1, 0b01, 0b00101, "rbit", int_arm64_neon_rbit>;
2231 defm REV16 : SIMDTwoVectorB<0, 0b00, 0b00001, "rev16", ARM64rev16>;
2232 defm REV32 : SIMDTwoVectorBH<1, 0b00000, "rev32", ARM64rev32>;
2233 defm REV64 : SIMDTwoVectorBHS<0, 0b00000, "rev64", ARM64rev64>;
2234 defm SADALP : SIMDLongTwoVectorTied<0, 0b00110, "sadalp",
2235 BinOpFrag<(add node:$LHS, (int_arm64_neon_saddlp node:$RHS))> >;
2236 defm SADDLP : SIMDLongTwoVector<0, 0b00010, "saddlp", int_arm64_neon_saddlp>;
2237 defm SCVTF : SIMDTwoVectorIntToFP<0, 0, 0b11101, "scvtf", sint_to_fp>;
2238 defm SHLL : SIMDVectorLShiftLongBySizeBHS;
2239 defm SQABS : SIMDTwoVectorBHSD<0, 0b00111, "sqabs", int_arm64_neon_sqabs>;
2240 defm SQNEG : SIMDTwoVectorBHSD<1, 0b00111, "sqneg", int_arm64_neon_sqneg>;
2241 defm SQXTN : SIMDMixedTwoVector<0, 0b10100, "sqxtn", int_arm64_neon_sqxtn>;
2242 defm SQXTUN : SIMDMixedTwoVector<1, 0b10010, "sqxtun", int_arm64_neon_sqxtun>;
2243 defm SUQADD : SIMDTwoVectorBHSDTied<0, 0b00011, "suqadd",int_arm64_neon_suqadd>;
2244 defm UADALP : SIMDLongTwoVectorTied<1, 0b00110, "uadalp",
2245 BinOpFrag<(add node:$LHS, (int_arm64_neon_uaddlp node:$RHS))> >;
2246 defm UADDLP : SIMDLongTwoVector<1, 0b00010, "uaddlp",
2247 int_arm64_neon_uaddlp>;
2248 defm UCVTF : SIMDTwoVectorIntToFP<1, 0, 0b11101, "ucvtf", uint_to_fp>;
2249 defm UQXTN : SIMDMixedTwoVector<1, 0b10100, "uqxtn", int_arm64_neon_uqxtn>;
2250 defm URECPE : SIMDTwoVectorS<0, 1, 0b11100, "urecpe", int_arm64_neon_urecpe>;
2251 defm URSQRTE: SIMDTwoVectorS<1, 1, 0b11100, "ursqrte", int_arm64_neon_ursqrte>;
2252 defm USQADD : SIMDTwoVectorBHSDTied<1, 0b00011, "usqadd",int_arm64_neon_usqadd>;
2253 defm XTN : SIMDMixedTwoVector<0, 0b10010, "xtn", trunc>;
2255 def : Pat<(v2f32 (ARM64rev64 V64:$Rn)), (REV64v2i32 V64:$Rn)>;
2256 def : Pat<(v4f32 (ARM64rev64 V128:$Rn)), (REV64v4i32 V128:$Rn)>;
2258 // Patterns for vector long shift (by element width). These need to match all
2259 // three of zext, sext and anyext so it's easier to pull the patterns out of the
2261 multiclass SIMDVectorLShiftLongBySizeBHSPats<SDPatternOperator ext> {
2262 def : Pat<(ARM64vshl (v8i16 (ext (v8i8 V64:$Rn))), (i32 8)),
2263 (SHLLv8i8 V64:$Rn)>;
2264 def : Pat<(ARM64vshl (v8i16 (ext (extract_high_v16i8 V128:$Rn))), (i32 8)),
2265 (SHLLv16i8 V128:$Rn)>;
2266 def : Pat<(ARM64vshl (v4i32 (ext (v4i16 V64:$Rn))), (i32 16)),
2267 (SHLLv4i16 V64:$Rn)>;
2268 def : Pat<(ARM64vshl (v4i32 (ext (extract_high_v8i16 V128:$Rn))), (i32 16)),
2269 (SHLLv8i16 V128:$Rn)>;
2270 def : Pat<(ARM64vshl (v2i64 (ext (v2i32 V64:$Rn))), (i32 32)),
2271 (SHLLv2i32 V64:$Rn)>;
2272 def : Pat<(ARM64vshl (v2i64 (ext (extract_high_v4i32 V128:$Rn))), (i32 32)),
2273 (SHLLv4i32 V128:$Rn)>;
2276 defm : SIMDVectorLShiftLongBySizeBHSPats<anyext>;
2277 defm : SIMDVectorLShiftLongBySizeBHSPats<zext>;
2278 defm : SIMDVectorLShiftLongBySizeBHSPats<sext>;
2280 //===----------------------------------------------------------------------===//
2281 // Advanced SIMD three vector instructions.
2282 //===----------------------------------------------------------------------===//
2284 defm ADD : SIMDThreeSameVector<0, 0b10000, "add", add>;
2285 defm ADDP : SIMDThreeSameVector<0, 0b10111, "addp", int_arm64_neon_addp>;
2286 defm CMEQ : SIMDThreeSameVector<1, 0b10001, "cmeq", ARM64cmeq>;
2287 defm CMGE : SIMDThreeSameVector<0, 0b00111, "cmge", ARM64cmge>;
2288 defm CMGT : SIMDThreeSameVector<0, 0b00110, "cmgt", ARM64cmgt>;
2289 defm CMHI : SIMDThreeSameVector<1, 0b00110, "cmhi", ARM64cmhi>;
2290 defm CMHS : SIMDThreeSameVector<1, 0b00111, "cmhs", ARM64cmhs>;
2291 defm CMTST : SIMDThreeSameVector<0, 0b10001, "cmtst", ARM64cmtst>;
2292 defm FABD : SIMDThreeSameVectorFP<1,1,0b11010,"fabd", int_arm64_neon_fabd>;
2293 defm FACGE : SIMDThreeSameVectorFPCmp<1,0,0b11101,"facge",int_arm64_neon_facge>;
2294 defm FACGT : SIMDThreeSameVectorFPCmp<1,1,0b11101,"facgt",int_arm64_neon_facgt>;
2295 defm FADDP : SIMDThreeSameVectorFP<1,0,0b11010,"faddp",int_arm64_neon_addp>;
2296 defm FADD : SIMDThreeSameVectorFP<0,0,0b11010,"fadd", fadd>;
2297 defm FCMEQ : SIMDThreeSameVectorFPCmp<0, 0, 0b11100, "fcmeq", ARM64fcmeq>;
2298 defm FCMGE : SIMDThreeSameVectorFPCmp<1, 0, 0b11100, "fcmge", ARM64fcmge>;
2299 defm FCMGT : SIMDThreeSameVectorFPCmp<1, 1, 0b11100, "fcmgt", ARM64fcmgt>;
2300 defm FDIV : SIMDThreeSameVectorFP<1,0,0b11111,"fdiv", fdiv>;
2301 defm FMAXNMP : SIMDThreeSameVectorFP<1,0,0b11000,"fmaxnmp", int_arm64_neon_fmaxnmp>;
2302 defm FMAXNM : SIMDThreeSameVectorFP<0,0,0b11000,"fmaxnm", int_arm64_neon_fmaxnm>;
2303 defm FMAXP : SIMDThreeSameVectorFP<1,0,0b11110,"fmaxp", int_arm64_neon_fmaxp>;
2304 defm FMAX : SIMDThreeSameVectorFP<0,0,0b11110,"fmax", ARM64fmax>;
2305 defm FMINNMP : SIMDThreeSameVectorFP<1,1,0b11000,"fminnmp", int_arm64_neon_fminnmp>;
2306 defm FMINNM : SIMDThreeSameVectorFP<0,1,0b11000,"fminnm", int_arm64_neon_fminnm>;
2307 defm FMINP : SIMDThreeSameVectorFP<1,1,0b11110,"fminp", int_arm64_neon_fminp>;
2308 defm FMIN : SIMDThreeSameVectorFP<0,1,0b11110,"fmin", ARM64fmin>;
2310 // NOTE: The operands of the PatFrag are reordered on FMLA/FMLS because the
2311 // instruction expects the addend first, while the fma intrinsic puts it last.
2312 defm FMLA : SIMDThreeSameVectorFPTied<0, 0, 0b11001, "fmla",
2313 TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)> >;
2314 defm FMLS : SIMDThreeSameVectorFPTied<0, 1, 0b11001, "fmls",
2315 TriOpFrag<(fma node:$MHS, (fneg node:$RHS), node:$LHS)> >;
2317 // The following def pats catch the case where the LHS of an FMA is negated.
2318 // The TriOpFrag above catches the case where the middle operand is negated.
2319 def : Pat<(v2f32 (fma (fneg V64:$Rn), V64:$Rm, V64:$Rd)),
2320 (FMLSv2f32 V64:$Rd, V64:$Rn, V64:$Rm)>;
2322 def : Pat<(v4f32 (fma (fneg V128:$Rn), V128:$Rm, V128:$Rd)),
2323 (FMLSv4f32 V128:$Rd, V128:$Rn, V128:$Rm)>;
2325 def : Pat<(v2f64 (fma (fneg V128:$Rn), V128:$Rm, V128:$Rd)),
2326 (FMLSv2f64 V128:$Rd, V128:$Rn, V128:$Rm)>;
2328 defm FMULX : SIMDThreeSameVectorFP<0,0,0b11011,"fmulx", int_arm64_neon_fmulx>;
2329 defm FMUL : SIMDThreeSameVectorFP<1,0,0b11011,"fmul", fmul>;
2330 defm FRECPS : SIMDThreeSameVectorFP<0,0,0b11111,"frecps", int_arm64_neon_frecps>;
2331 defm FRSQRTS : SIMDThreeSameVectorFP<0,1,0b11111,"frsqrts", int_arm64_neon_frsqrts>;
2332 defm FSUB : SIMDThreeSameVectorFP<0,1,0b11010,"fsub", fsub>;
2333 defm MLA : SIMDThreeSameVectorBHSTied<0, 0b10010, "mla",
2334 TriOpFrag<(add node:$LHS, (mul node:$MHS, node:$RHS))> >;
2335 defm MLS : SIMDThreeSameVectorBHSTied<1, 0b10010, "mls",
2336 TriOpFrag<(sub node:$LHS, (mul node:$MHS, node:$RHS))> >;
2337 defm MUL : SIMDThreeSameVectorBHS<0, 0b10011, "mul", mul>;
2338 defm PMUL : SIMDThreeSameVectorB<1, 0b10011, "pmul", int_arm64_neon_pmul>;
2339 defm SABA : SIMDThreeSameVectorBHSTied<0, 0b01111, "saba",
2340 TriOpFrag<(add node:$LHS, (int_arm64_neon_sabd node:$MHS, node:$RHS))> >;
2341 defm SABD : SIMDThreeSameVectorBHS<0,0b01110,"sabd", int_arm64_neon_sabd>;
2342 defm SHADD : SIMDThreeSameVectorBHS<0,0b00000,"shadd", int_arm64_neon_shadd>;
2343 defm SHSUB : SIMDThreeSameVectorBHS<0,0b00100,"shsub", int_arm64_neon_shsub>;
2344 defm SMAXP : SIMDThreeSameVectorBHS<0,0b10100,"smaxp", int_arm64_neon_smaxp>;
2345 defm SMAX : SIMDThreeSameVectorBHS<0,0b01100,"smax", int_arm64_neon_smax>;
2346 defm SMINP : SIMDThreeSameVectorBHS<0,0b10101,"sminp", int_arm64_neon_sminp>;
2347 defm SMIN : SIMDThreeSameVectorBHS<0,0b01101,"smin", int_arm64_neon_smin>;
2348 defm SQADD : SIMDThreeSameVector<0,0b00001,"sqadd", int_arm64_neon_sqadd>;
2349 defm SQDMULH : SIMDThreeSameVectorHS<0,0b10110,"sqdmulh",int_arm64_neon_sqdmulh>;
2350 defm SQRDMULH : SIMDThreeSameVectorHS<1,0b10110,"sqrdmulh",int_arm64_neon_sqrdmulh>;
2351 defm SQRSHL : SIMDThreeSameVector<0,0b01011,"sqrshl", int_arm64_neon_sqrshl>;
2352 defm SQSHL : SIMDThreeSameVector<0,0b01001,"sqshl", int_arm64_neon_sqshl>;
2353 defm SQSUB : SIMDThreeSameVector<0,0b00101,"sqsub", int_arm64_neon_sqsub>;
2354 defm SRHADD : SIMDThreeSameVectorBHS<0,0b00010,"srhadd",int_arm64_neon_srhadd>;
2355 defm SRSHL : SIMDThreeSameVector<0,0b01010,"srshl", int_arm64_neon_srshl>;
2356 defm SSHL : SIMDThreeSameVector<0,0b01000,"sshl", int_arm64_neon_sshl>;
2357 defm SUB : SIMDThreeSameVector<1,0b10000,"sub", sub>;
2358 defm UABA : SIMDThreeSameVectorBHSTied<1, 0b01111, "uaba",
2359 TriOpFrag<(add node:$LHS, (int_arm64_neon_uabd node:$MHS, node:$RHS))> >;
2360 defm UABD : SIMDThreeSameVectorBHS<1,0b01110,"uabd", int_arm64_neon_uabd>;
2361 defm UHADD : SIMDThreeSameVectorBHS<1,0b00000,"uhadd", int_arm64_neon_uhadd>;
2362 defm UHSUB : SIMDThreeSameVectorBHS<1,0b00100,"uhsub", int_arm64_neon_uhsub>;
2363 defm UMAXP : SIMDThreeSameVectorBHS<1,0b10100,"umaxp", int_arm64_neon_umaxp>;
2364 defm UMAX : SIMDThreeSameVectorBHS<1,0b01100,"umax", int_arm64_neon_umax>;
2365 defm UMINP : SIMDThreeSameVectorBHS<1,0b10101,"uminp", int_arm64_neon_uminp>;
2366 defm UMIN : SIMDThreeSameVectorBHS<1,0b01101,"umin", int_arm64_neon_umin>;
2367 defm UQADD : SIMDThreeSameVector<1,0b00001,"uqadd", int_arm64_neon_uqadd>;
2368 defm UQRSHL : SIMDThreeSameVector<1,0b01011,"uqrshl", int_arm64_neon_uqrshl>;
2369 defm UQSHL : SIMDThreeSameVector<1,0b01001,"uqshl", int_arm64_neon_uqshl>;
2370 defm UQSUB : SIMDThreeSameVector<1,0b00101,"uqsub", int_arm64_neon_uqsub>;
2371 defm URHADD : SIMDThreeSameVectorBHS<1,0b00010,"urhadd", int_arm64_neon_urhadd>;
2372 defm URSHL : SIMDThreeSameVector<1,0b01010,"urshl", int_arm64_neon_urshl>;
2373 defm USHL : SIMDThreeSameVector<1,0b01000,"ushl", int_arm64_neon_ushl>;
2375 defm AND : SIMDLogicalThreeVector<0, 0b00, "and", and>;
2376 defm BIC : SIMDLogicalThreeVector<0, 0b01, "bic",
2377 BinOpFrag<(and node:$LHS, (vnot node:$RHS))> >;
2378 defm BIF : SIMDLogicalThreeVector<1, 0b11, "bif">;
2379 defm BIT : SIMDLogicalThreeVectorTied<1, 0b10, "bit", ARM64bit>;
2380 defm BSL : SIMDLogicalThreeVectorTied<1, 0b01, "bsl",
2381 TriOpFrag<(or (and node:$LHS, node:$MHS), (and (vnot node:$LHS), node:$RHS))>>;
2382 defm EOR : SIMDLogicalThreeVector<1, 0b00, "eor", xor>;
2383 defm ORN : SIMDLogicalThreeVector<0, 0b11, "orn",
2384 BinOpFrag<(or node:$LHS, (vnot node:$RHS))> >;
2385 defm ORR : SIMDLogicalThreeVector<0, 0b10, "orr", or>;
2387 def : Pat<(ARM64bsl (v8i8 V64:$Rd), V64:$Rn, V64:$Rm),
2388 (BSLv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
2389 def : Pat<(ARM64bsl (v4i16 V64:$Rd), V64:$Rn, V64:$Rm),
2390 (BSLv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
2391 def : Pat<(ARM64bsl (v2i32 V64:$Rd), V64:$Rn, V64:$Rm),
2392 (BSLv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
2393 def : Pat<(ARM64bsl (v1i64 V64:$Rd), V64:$Rn, V64:$Rm),
2394 (BSLv8i8 V64:$Rd, V64:$Rn, V64:$Rm)>;
2396 def : Pat<(ARM64bsl (v16i8 V128:$Rd), V128:$Rn, V128:$Rm),
2397 (BSLv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
2398 def : Pat<(ARM64bsl (v8i16 V128:$Rd), V128:$Rn, V128:$Rm),
2399 (BSLv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
2400 def : Pat<(ARM64bsl (v4i32 V128:$Rd), V128:$Rn, V128:$Rm),
2401 (BSLv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
2402 def : Pat<(ARM64bsl (v2i64 V128:$Rd), V128:$Rn, V128:$Rm),
2403 (BSLv16i8 V128:$Rd, V128:$Rn, V128:$Rm)>;
2405 // FIXME: the .16b and .8b variantes should be emitted by the
2406 // AsmWriter. TableGen's AsmWriter-generator doesn't deal with variant syntaxes
2407 // in aliases yet though.
2408 def : InstAlias<"mov{\t$dst.16b, $src.16b|.16b\t$dst, $src}",
2409 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2410 def : InstAlias<"{mov\t$dst.8h, $src.8h|mov.8h\t$dst, $src}",
2411 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2412 def : InstAlias<"{mov\t$dst.4s, $src.4s|mov.4s\t$dst, $src}",
2413 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2414 def : InstAlias<"{mov\t$dst.2d, $src.2d|mov.2d\t$dst, $src}",
2415 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2417 def : InstAlias<"{mov\t$dst.8b, $src.8b|mov.8b\t$dst, $src}",
2418 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2419 def : InstAlias<"{mov\t$dst.4h, $src.4h|mov.4h\t$dst, $src}",
2420 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2421 def : InstAlias<"{mov\t$dst.2s, $src.2s|mov.2s\t$dst, $src}",
2422 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2423 def : InstAlias<"{mov\t$dst.1d, $src.1d|mov.1d\t$dst, $src}",
2424 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2426 def : InstAlias<"{cmls\t$dst.8b, $src1.8b, $src2.8b" #
2427 "|cmls.8b\t$dst, $src1, $src2}",
2428 (CMHSv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2429 def : InstAlias<"{cmls\t$dst.16b, $src1.16b, $src2.16b" #
2430 "|cmls.16b\t$dst, $src1, $src2}",
2431 (CMHSv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2432 def : InstAlias<"{cmls\t$dst.4h, $src1.4h, $src2.4h" #
2433 "|cmls.4h\t$dst, $src1, $src2}",
2434 (CMHSv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2435 def : InstAlias<"{cmls\t$dst.8h, $src1.8h, $src2.8h" #
2436 "|cmls.8h\t$dst, $src1, $src2}",
2437 (CMHSv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2438 def : InstAlias<"{cmls\t$dst.2s, $src1.2s, $src2.2s" #
2439 "|cmls.2s\t$dst, $src1, $src2}",
2440 (CMHSv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2441 def : InstAlias<"{cmls\t$dst.4s, $src1.4s, $src2.4s" #
2442 "|cmls.4s\t$dst, $src1, $src2}",
2443 (CMHSv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2444 def : InstAlias<"{cmls\t$dst.2d, $src1.2d, $src2.2d" #
2445 "|cmls.2d\t$dst, $src1, $src2}",
2446 (CMHSv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2448 def : InstAlias<"{cmlo\t$dst.8b, $src1.8b, $src2.8b" #
2449 "|cmlo.8b\t$dst, $src1, $src2}",
2450 (CMHIv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2451 def : InstAlias<"{cmlo\t$dst.16b, $src1.16b, $src2.16b" #
2452 "|cmlo.16b\t$dst, $src1, $src2}",
2453 (CMHIv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2454 def : InstAlias<"{cmlo\t$dst.4h, $src1.4h, $src2.4h" #
2455 "|cmlo.4h\t$dst, $src1, $src2}",
2456 (CMHIv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2457 def : InstAlias<"{cmlo\t$dst.8h, $src1.8h, $src2.8h" #
2458 "|cmlo.8h\t$dst, $src1, $src2}",
2459 (CMHIv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2460 def : InstAlias<"{cmlo\t$dst.2s, $src1.2s, $src2.2s" #
2461 "|cmlo.2s\t$dst, $src1, $src2}",
2462 (CMHIv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2463 def : InstAlias<"{cmlo\t$dst.4s, $src1.4s, $src2.4s" #
2464 "|cmlo.4s\t$dst, $src1, $src2}",
2465 (CMHIv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2466 def : InstAlias<"{cmlo\t$dst.2d, $src1.2d, $src2.2d" #
2467 "|cmlo.2d\t$dst, $src1, $src2}",
2468 (CMHIv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2470 def : InstAlias<"{cmle\t$dst.8b, $src1.8b, $src2.8b" #
2471 "|cmle.8b\t$dst, $src1, $src2}",
2472 (CMGEv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2473 def : InstAlias<"{cmle\t$dst.16b, $src1.16b, $src2.16b" #
2474 "|cmle.16b\t$dst, $src1, $src2}",
2475 (CMGEv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2476 def : InstAlias<"{cmle\t$dst.4h, $src1.4h, $src2.4h" #
2477 "|cmle.4h\t$dst, $src1, $src2}",
2478 (CMGEv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2479 def : InstAlias<"{cmle\t$dst.8h, $src1.8h, $src2.8h" #
2480 "|cmle.8h\t$dst, $src1, $src2}",
2481 (CMGEv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2482 def : InstAlias<"{cmle\t$dst.2s, $src1.2s, $src2.2s" #
2483 "|cmle.2s\t$dst, $src1, $src2}",
2484 (CMGEv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2485 def : InstAlias<"{cmle\t$dst.4s, $src1.4s, $src2.4s" #
2486 "|cmle.4s\t$dst, $src1, $src2}",
2487 (CMGEv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2488 def : InstAlias<"{cmle\t$dst.2d, $src1.2d, $src2.2d" #
2489 "|cmle.2d\t$dst, $src1, $src2}",
2490 (CMGEv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2492 def : InstAlias<"{cmlt\t$dst.8b, $src1.8b, $src2.8b" #
2493 "|cmlt.8b\t$dst, $src1, $src2}",
2494 (CMGTv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2495 def : InstAlias<"{cmlt\t$dst.16b, $src1.16b, $src2.16b" #
2496 "|cmlt.16b\t$dst, $src1, $src2}",
2497 (CMGTv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2498 def : InstAlias<"{cmlt\t$dst.4h, $src1.4h, $src2.4h" #
2499 "|cmlt.4h\t$dst, $src1, $src2}",
2500 (CMGTv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2501 def : InstAlias<"{cmlt\t$dst.8h, $src1.8h, $src2.8h" #
2502 "|cmlt.8h\t$dst, $src1, $src2}",
2503 (CMGTv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2504 def : InstAlias<"{cmlt\t$dst.2s, $src1.2s, $src2.2s" #
2505 "|cmlt.2s\t$dst, $src1, $src2}",
2506 (CMGTv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2507 def : InstAlias<"{cmlt\t$dst.4s, $src1.4s, $src2.4s" #
2508 "|cmlt.4s\t$dst, $src1, $src2}",
2509 (CMGTv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2510 def : InstAlias<"{cmlt\t$dst.2d, $src1.2d, $src2.2d" #
2511 "|cmlt.2d\t$dst, $src1, $src2}",
2512 (CMGTv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2514 def : InstAlias<"{fcmle\t$dst.2s, $src1.2s, $src2.2s" #
2515 "|fcmle.2s\t$dst, $src1, $src2}",
2516 (FCMGEv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
2517 def : InstAlias<"{fcmle\t$dst.4s, $src1.4s, $src2.4s" #
2518 "|fcmle.4s\t$dst, $src1, $src2}",
2519 (FCMGEv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
2520 def : InstAlias<"{fcmle\t$dst.2d, $src1.2d, $src2.2d" #
2521 "|fcmle.2d\t$dst, $src1, $src2}",
2522 (FCMGEv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
2524 def : InstAlias<"{fcmlt\t$dst.2s, $src1.2s, $src2.2s" #
2525 "|fcmlt.2s\t$dst, $src1, $src2}",
2526 (FCMGTv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
2527 def : InstAlias<"{fcmlt\t$dst.4s, $src1.4s, $src2.4s" #
2528 "|fcmlt.4s\t$dst, $src1, $src2}",
2529 (FCMGTv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
2530 def : InstAlias<"{fcmlt\t$dst.2d, $src1.2d, $src2.2d" #
2531 "|fcmlt.2d\t$dst, $src1, $src2}",
2532 (FCMGTv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
2534 def : InstAlias<"{facle\t$dst.2s, $src1.2s, $src2.2s" #
2535 "|facle.2s\t$dst, $src1, $src2}",
2536 (FACGEv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
2537 def : InstAlias<"{facle\t$dst.4s, $src1.4s, $src2.4s" #
2538 "|facle.4s\t$dst, $src1, $src2}",
2539 (FACGEv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
2540 def : InstAlias<"{facle\t$dst.2d, $src1.2d, $src2.2d" #
2541 "|facle.2d\t$dst, $src1, $src2}",
2542 (FACGEv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
2544 def : InstAlias<"{faclt\t$dst.2s, $src1.2s, $src2.2s" #
2545 "|faclt.2s\t$dst, $src1, $src2}",
2546 (FACGTv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
2547 def : InstAlias<"{faclt\t$dst.4s, $src1.4s, $src2.4s" #
2548 "|faclt.4s\t$dst, $src1, $src2}",
2549 (FACGTv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
2550 def : InstAlias<"{faclt\t$dst.2d, $src1.2d, $src2.2d" #
2551 "|faclt.2d\t$dst, $src1, $src2}",
2552 (FACGTv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
2554 //===----------------------------------------------------------------------===//
2555 // Advanced SIMD three scalar instructions.
2556 //===----------------------------------------------------------------------===//
2558 defm ADD : SIMDThreeScalarD<0, 0b10000, "add", add>;
2559 defm CMEQ : SIMDThreeScalarD<1, 0b10001, "cmeq", ARM64cmeq>;
2560 defm CMGE : SIMDThreeScalarD<0, 0b00111, "cmge", ARM64cmge>;
2561 defm CMGT : SIMDThreeScalarD<0, 0b00110, "cmgt", ARM64cmgt>;
2562 defm CMHI : SIMDThreeScalarD<1, 0b00110, "cmhi", ARM64cmhi>;
2563 defm CMHS : SIMDThreeScalarD<1, 0b00111, "cmhs", ARM64cmhs>;
2564 defm CMTST : SIMDThreeScalarD<0, 0b10001, "cmtst", ARM64cmtst>;
2565 defm FABD : SIMDThreeScalarSD<1, 1, 0b11010, "fabd", int_arm64_sisd_fabd>;
2566 def : Pat<(v1f64 (int_arm64_neon_fabd (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2567 (FABD64 FPR64:$Rn, FPR64:$Rm)>;
2568 defm FACGE : SIMDThreeScalarFPCmp<1, 0, 0b11101, "facge",
2569 int_arm64_neon_facge>;
2570 defm FACGT : SIMDThreeScalarFPCmp<1, 1, 0b11101, "facgt",
2571 int_arm64_neon_facgt>;
2572 defm FCMEQ : SIMDThreeScalarFPCmp<0, 0, 0b11100, "fcmeq", ARM64fcmeq>;
2573 defm FCMGE : SIMDThreeScalarFPCmp<1, 0, 0b11100, "fcmge", ARM64fcmge>;
2574 defm FCMGT : SIMDThreeScalarFPCmp<1, 1, 0b11100, "fcmgt", ARM64fcmgt>;
2575 defm FMULX : SIMDThreeScalarSD<0, 0, 0b11011, "fmulx", int_arm64_neon_fmulx>;
2576 defm FRECPS : SIMDThreeScalarSD<0, 0, 0b11111, "frecps", int_arm64_neon_frecps>;
2577 defm FRSQRTS : SIMDThreeScalarSD<0, 1, 0b11111, "frsqrts", int_arm64_neon_frsqrts>;
2578 defm SQADD : SIMDThreeScalarBHSD<0, 0b00001, "sqadd", int_arm64_neon_sqadd>;
2579 defm SQDMULH : SIMDThreeScalarHS< 0, 0b10110, "sqdmulh", int_arm64_neon_sqdmulh>;
2580 defm SQRDMULH : SIMDThreeScalarHS< 1, 0b10110, "sqrdmulh", int_arm64_neon_sqrdmulh>;
2581 defm SQRSHL : SIMDThreeScalarBHSD<0, 0b01011, "sqrshl",int_arm64_neon_sqrshl>;
2582 defm SQSHL : SIMDThreeScalarBHSD<0, 0b01001, "sqshl", int_arm64_neon_sqshl>;
2583 defm SQSUB : SIMDThreeScalarBHSD<0, 0b00101, "sqsub", int_arm64_neon_sqsub>;
2584 defm SRSHL : SIMDThreeScalarD< 0, 0b01010, "srshl", int_arm64_neon_srshl>;
2585 defm SSHL : SIMDThreeScalarD< 0, 0b01000, "sshl", int_arm64_neon_sshl>;
2586 defm SUB : SIMDThreeScalarD< 1, 0b10000, "sub", sub>;
2587 defm UQADD : SIMDThreeScalarBHSD<1, 0b00001, "uqadd", int_arm64_neon_uqadd>;
2588 defm UQRSHL : SIMDThreeScalarBHSD<1, 0b01011, "uqrshl",int_arm64_neon_uqrshl>;
2589 defm UQSHL : SIMDThreeScalarBHSD<1, 0b01001, "uqshl", int_arm64_neon_uqshl>;
2590 defm UQSUB : SIMDThreeScalarBHSD<1, 0b00101, "uqsub", int_arm64_neon_uqsub>;
2591 defm URSHL : SIMDThreeScalarD< 1, 0b01010, "urshl", int_arm64_neon_urshl>;
2592 defm USHL : SIMDThreeScalarD< 1, 0b01000, "ushl", int_arm64_neon_ushl>;
2594 def : InstAlias<"cmls $dst, $src1, $src2",
2595 (CMHSv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2596 def : InstAlias<"cmle $dst, $src1, $src2",
2597 (CMGEv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2598 def : InstAlias<"cmlo $dst, $src1, $src2",
2599 (CMHIv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2600 def : InstAlias<"cmlt $dst, $src1, $src2",
2601 (CMGTv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2602 def : InstAlias<"fcmle $dst, $src1, $src2",
2603 (FCMGE32 FPR32:$dst, FPR32:$src2, FPR32:$src1)>;
2604 def : InstAlias<"fcmle $dst, $src1, $src2",
2605 (FCMGE64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2606 def : InstAlias<"fcmlt $dst, $src1, $src2",
2607 (FCMGT32 FPR32:$dst, FPR32:$src2, FPR32:$src1)>;
2608 def : InstAlias<"fcmlt $dst, $src1, $src2",
2609 (FCMGT64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2610 def : InstAlias<"facle $dst, $src1, $src2",
2611 (FACGE32 FPR32:$dst, FPR32:$src2, FPR32:$src1)>;
2612 def : InstAlias<"facle $dst, $src1, $src2",
2613 (FACGE64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2614 def : InstAlias<"faclt $dst, $src1, $src2",
2615 (FACGT32 FPR32:$dst, FPR32:$src2, FPR32:$src1)>;
2616 def : InstAlias<"faclt $dst, $src1, $src2",
2617 (FACGT64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2619 //===----------------------------------------------------------------------===//
2620 // Advanced SIMD three scalar instructions (mixed operands).
2621 //===----------------------------------------------------------------------===//
2622 defm SQDMULL : SIMDThreeScalarMixedHS<0, 0b11010, "sqdmull",
2623 int_arm64_neon_sqdmulls_scalar>;
2624 defm SQDMLAL : SIMDThreeScalarMixedTiedHS<0, 0b10010, "sqdmlal">;
2625 defm SQDMLSL : SIMDThreeScalarMixedTiedHS<0, 0b10110, "sqdmlsl">;
2627 def : Pat<(i64 (int_arm64_neon_sqadd (i64 FPR64:$Rd),
2628 (i64 (int_arm64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
2629 (i32 FPR32:$Rm))))),
2630 (SQDMLALi32 FPR64:$Rd, FPR32:$Rn, FPR32:$Rm)>;
2631 def : Pat<(i64 (int_arm64_neon_sqsub (i64 FPR64:$Rd),
2632 (i64 (int_arm64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
2633 (i32 FPR32:$Rm))))),
2634 (SQDMLSLi32 FPR64:$Rd, FPR32:$Rn, FPR32:$Rm)>;
2636 //===----------------------------------------------------------------------===//
2637 // Advanced SIMD two scalar instructions.
2638 //===----------------------------------------------------------------------===//
2640 defm ABS : SIMDTwoScalarD< 0, 0b01011, "abs", int_arm64_neon_abs>;
2641 defm CMEQ : SIMDCmpTwoScalarD< 0, 0b01001, "cmeq", ARM64cmeqz>;
2642 defm CMGE : SIMDCmpTwoScalarD< 1, 0b01000, "cmge", ARM64cmgez>;
2643 defm CMGT : SIMDCmpTwoScalarD< 0, 0b01000, "cmgt", ARM64cmgtz>;
2644 defm CMLE : SIMDCmpTwoScalarD< 1, 0b01001, "cmle", ARM64cmlez>;
2645 defm CMLT : SIMDCmpTwoScalarD< 0, 0b01010, "cmlt", ARM64cmltz>;
2646 defm FCMEQ : SIMDCmpTwoScalarSD<0, 1, 0b01101, "fcmeq", ARM64fcmeqz>;
2647 defm FCMGE : SIMDCmpTwoScalarSD<1, 1, 0b01100, "fcmge", ARM64fcmgez>;
2648 defm FCMGT : SIMDCmpTwoScalarSD<0, 1, 0b01100, "fcmgt", ARM64fcmgtz>;
2649 defm FCMLE : SIMDCmpTwoScalarSD<1, 1, 0b01101, "fcmle", ARM64fcmlez>;
2650 defm FCMLT : SIMDCmpTwoScalarSD<0, 1, 0b01110, "fcmlt", ARM64fcmltz>;
2651 defm FCVTAS : SIMDTwoScalarSD< 0, 0, 0b11100, "fcvtas">;
2652 defm FCVTAU : SIMDTwoScalarSD< 1, 0, 0b11100, "fcvtau">;
2653 defm FCVTMS : SIMDTwoScalarSD< 0, 0, 0b11011, "fcvtms">;
2654 defm FCVTMU : SIMDTwoScalarSD< 1, 0, 0b11011, "fcvtmu">;
2655 defm FCVTNS : SIMDTwoScalarSD< 0, 0, 0b11010, "fcvtns">;
2656 defm FCVTNU : SIMDTwoScalarSD< 1, 0, 0b11010, "fcvtnu">;
2657 defm FCVTPS : SIMDTwoScalarSD< 0, 1, 0b11010, "fcvtps">;
2658 defm FCVTPU : SIMDTwoScalarSD< 1, 1, 0b11010, "fcvtpu">;
2659 def FCVTXNv1i64 : SIMDInexactCvtTwoScalar<0b10110, "fcvtxn">;
2660 defm FCVTZS : SIMDTwoScalarSD< 0, 1, 0b11011, "fcvtzs">;
2661 defm FCVTZU : SIMDTwoScalarSD< 1, 1, 0b11011, "fcvtzu">;
2662 defm FRECPE : SIMDTwoScalarSD< 0, 1, 0b11101, "frecpe">;
2663 defm FRECPX : SIMDTwoScalarSD< 0, 1, 0b11111, "frecpx">;
2664 defm FRSQRTE : SIMDTwoScalarSD< 1, 1, 0b11101, "frsqrte">;
2665 defm NEG : SIMDTwoScalarD< 1, 0b01011, "neg",
2666 UnOpFrag<(sub immAllZerosV, node:$LHS)> >;
2667 defm SCVTF : SIMDTwoScalarCVTSD< 0, 0, 0b11101, "scvtf", ARM64sitof>;
2668 defm SQABS : SIMDTwoScalarBHSD< 0, 0b00111, "sqabs", int_arm64_neon_sqabs>;
2669 defm SQNEG : SIMDTwoScalarBHSD< 1, 0b00111, "sqneg", int_arm64_neon_sqneg>;
2670 defm SQXTN : SIMDTwoScalarMixedBHS< 0, 0b10100, "sqxtn", int_arm64_neon_scalar_sqxtn>;
2671 defm SQXTUN : SIMDTwoScalarMixedBHS< 1, 0b10010, "sqxtun", int_arm64_neon_scalar_sqxtun>;
2672 defm SUQADD : SIMDTwoScalarBHSDTied< 0, 0b00011, "suqadd",
2673 int_arm64_neon_suqadd>;
2674 defm UCVTF : SIMDTwoScalarCVTSD< 1, 0, 0b11101, "ucvtf", ARM64uitof>;
2675 defm UQXTN : SIMDTwoScalarMixedBHS<1, 0b10100, "uqxtn", int_arm64_neon_scalar_uqxtn>;
2676 defm USQADD : SIMDTwoScalarBHSDTied< 1, 0b00011, "usqadd",
2677 int_arm64_neon_usqadd>;
2679 def : Pat<(ARM64neg (v1i64 V64:$Rn)), (NEGv1i64 V64:$Rn)>;
2681 def : Pat<(v1i64 (int_arm64_neon_fcvtas (v1f64 FPR64:$Rn))),
2682 (FCVTASv1i64 FPR64:$Rn)>;
2683 def : Pat<(v1i64 (int_arm64_neon_fcvtau (v1f64 FPR64:$Rn))),
2684 (FCVTAUv1i64 FPR64:$Rn)>;
2685 def : Pat<(v1i64 (int_arm64_neon_fcvtms (v1f64 FPR64:$Rn))),
2686 (FCVTMSv1i64 FPR64:$Rn)>;
2687 def : Pat<(v1i64 (int_arm64_neon_fcvtmu (v1f64 FPR64:$Rn))),
2688 (FCVTMUv1i64 FPR64:$Rn)>;
2689 def : Pat<(v1i64 (int_arm64_neon_fcvtns (v1f64 FPR64:$Rn))),
2690 (FCVTNSv1i64 FPR64:$Rn)>;
2691 def : Pat<(v1i64 (int_arm64_neon_fcvtnu (v1f64 FPR64:$Rn))),
2692 (FCVTNUv1i64 FPR64:$Rn)>;
2693 def : Pat<(v1i64 (int_arm64_neon_fcvtps (v1f64 FPR64:$Rn))),
2694 (FCVTPSv1i64 FPR64:$Rn)>;
2695 def : Pat<(v1i64 (int_arm64_neon_fcvtpu (v1f64 FPR64:$Rn))),
2696 (FCVTPUv1i64 FPR64:$Rn)>;
2698 def : Pat<(f32 (int_arm64_neon_frecpe (f32 FPR32:$Rn))),
2699 (FRECPEv1i32 FPR32:$Rn)>;
2700 def : Pat<(f64 (int_arm64_neon_frecpe (f64 FPR64:$Rn))),
2701 (FRECPEv1i64 FPR64:$Rn)>;
2702 def : Pat<(v1f64 (int_arm64_neon_frecpe (v1f64 FPR64:$Rn))),
2703 (FRECPEv1i64 FPR64:$Rn)>;
2705 def : Pat<(f32 (int_arm64_neon_frecpx (f32 FPR32:$Rn))),
2706 (FRECPXv1i32 FPR32:$Rn)>;
2707 def : Pat<(f64 (int_arm64_neon_frecpx (f64 FPR64:$Rn))),
2708 (FRECPXv1i64 FPR64:$Rn)>;
2710 def : Pat<(f32 (int_arm64_neon_frsqrte (f32 FPR32:$Rn))),
2711 (FRSQRTEv1i32 FPR32:$Rn)>;
2712 def : Pat<(f64 (int_arm64_neon_frsqrte (f64 FPR64:$Rn))),
2713 (FRSQRTEv1i64 FPR64:$Rn)>;
2714 def : Pat<(v1f64 (int_arm64_neon_frsqrte (v1f64 FPR64:$Rn))),
2715 (FRSQRTEv1i64 FPR64:$Rn)>;
2717 // If an integer is about to be converted to a floating point value,
2718 // just load it on the floating point unit.
2719 // Here are the patterns for 8 and 16-bits to float.
2721 def : Pat <(f32 (uint_to_fp (i32 (zextloadi8 ro_indexed8:$addr)))),
2722 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2723 (LDRBro ro_indexed8:$addr), bsub))>;
2724 def : Pat <(f32 (uint_to_fp (i32 (zextloadi8 am_indexed8:$addr)))),
2725 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2726 (LDRBui am_indexed8:$addr), bsub))>;
2727 def : Pat <(f32 (uint_to_fp (i32 (zextloadi8 am_unscaled8:$addr)))),
2728 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2729 (LDURBi am_unscaled8:$addr), bsub))>;
2730 // 16-bits -> float.
2731 def : Pat <(f32 (uint_to_fp (i32 (zextloadi16 ro_indexed16:$addr)))),
2732 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2733 (LDRHro ro_indexed16:$addr), hsub))>;
2734 def : Pat <(f32 (uint_to_fp (i32 (zextloadi16 am_indexed16:$addr)))),
2735 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2736 (LDRHui am_indexed16:$addr), hsub))>;
2737 def : Pat <(f32 (uint_to_fp (i32 (zextloadi16 am_unscaled16:$addr)))),
2738 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2739 (LDURHi am_unscaled16:$addr), hsub))>;
2740 // 32-bits are handled in target specific dag combine:
2741 // performIntToFpCombine.
2742 // 64-bits integer to 32-bits floating point, not possible with
2743 // UCVTF on floating point registers (both source and destination
2744 // must have the same size).
2746 // Here are the patterns for 8, 16, 32, and 64-bits to double.
2747 // 8-bits -> double.
2748 def : Pat <(f64 (uint_to_fp (i32 (zextloadi8 ro_indexed8:$addr)))),
2749 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2750 (LDRBro ro_indexed8:$addr), bsub))>;
2751 def : Pat <(f64 (uint_to_fp (i32 (zextloadi8 am_indexed8:$addr)))),
2752 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2753 (LDRBui am_indexed8:$addr), bsub))>;
2754 def : Pat <(f64 (uint_to_fp (i32 (zextloadi8 am_unscaled8:$addr)))),
2755 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2756 (LDURBi am_unscaled8:$addr), bsub))>;
2757 // 16-bits -> double.
2758 def : Pat <(f64 (uint_to_fp (i32 (zextloadi16 ro_indexed16:$addr)))),
2759 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2760 (LDRHro ro_indexed16:$addr), hsub))>;
2761 def : Pat <(f64 (uint_to_fp (i32 (zextloadi16 am_indexed16:$addr)))),
2762 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2763 (LDRHui am_indexed16:$addr), hsub))>;
2764 def : Pat <(f64 (uint_to_fp (i32 (zextloadi16 am_unscaled16:$addr)))),
2765 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2766 (LDURHi am_unscaled16:$addr), hsub))>;
2767 // 32-bits -> double.
2768 def : Pat <(f64 (uint_to_fp (i32 (load ro_indexed32:$addr)))),
2769 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2770 (LDRSro ro_indexed32:$addr), ssub))>;
2771 def : Pat <(f64 (uint_to_fp (i32 (load am_indexed32:$addr)))),
2772 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2773 (LDRSui am_indexed32:$addr), ssub))>;
2774 def : Pat <(f64 (uint_to_fp (i32 (load am_unscaled32:$addr)))),
2775 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2776 (LDURSi am_unscaled32:$addr), ssub))>;
2777 // 64-bits -> double are handled in target specific dag combine:
2778 // performIntToFpCombine.
2780 //===----------------------------------------------------------------------===//
2781 // Advanced SIMD three different-sized vector instructions.
2782 //===----------------------------------------------------------------------===//
2784 defm ADDHN : SIMDNarrowThreeVectorBHS<0,0b0100,"addhn", int_arm64_neon_addhn>;
2785 defm SUBHN : SIMDNarrowThreeVectorBHS<0,0b0110,"subhn", int_arm64_neon_subhn>;
2786 defm RADDHN : SIMDNarrowThreeVectorBHS<1,0b0100,"raddhn",int_arm64_neon_raddhn>;
2787 defm RSUBHN : SIMDNarrowThreeVectorBHS<1,0b0110,"rsubhn",int_arm64_neon_rsubhn>;
2788 defm PMULL : SIMDDifferentThreeVectorBD<0,0b1110,"pmull",int_arm64_neon_pmull>;
2789 defm SABAL : SIMDLongThreeVectorTiedBHSabal<0,0b0101,"sabal",
2790 int_arm64_neon_sabd>;
2791 defm SABDL : SIMDLongThreeVectorBHSabdl<0, 0b0111, "sabdl",
2792 int_arm64_neon_sabd>;
2793 defm SADDL : SIMDLongThreeVectorBHS< 0, 0b0000, "saddl",
2794 BinOpFrag<(add (sext node:$LHS), (sext node:$RHS))>>;
2795 defm SADDW : SIMDWideThreeVectorBHS< 0, 0b0001, "saddw",
2796 BinOpFrag<(add node:$LHS, (sext node:$RHS))>>;
2797 defm SMLAL : SIMDLongThreeVectorTiedBHS<0, 0b1000, "smlal",
2798 TriOpFrag<(add node:$LHS, (int_arm64_neon_smull node:$MHS, node:$RHS))>>;
2799 defm SMLSL : SIMDLongThreeVectorTiedBHS<0, 0b1010, "smlsl",
2800 TriOpFrag<(sub node:$LHS, (int_arm64_neon_smull node:$MHS, node:$RHS))>>;
2801 defm SMULL : SIMDLongThreeVectorBHS<0, 0b1100, "smull", int_arm64_neon_smull>;
2802 defm SQDMLAL : SIMDLongThreeVectorSQDMLXTiedHS<0, 0b1001, "sqdmlal",
2803 int_arm64_neon_sqadd>;
2804 defm SQDMLSL : SIMDLongThreeVectorSQDMLXTiedHS<0, 0b1011, "sqdmlsl",
2805 int_arm64_neon_sqsub>;
2806 defm SQDMULL : SIMDLongThreeVectorHS<0, 0b1101, "sqdmull",
2807 int_arm64_neon_sqdmull>;
2808 defm SSUBL : SIMDLongThreeVectorBHS<0, 0b0010, "ssubl",
2809 BinOpFrag<(sub (sext node:$LHS), (sext node:$RHS))>>;
2810 defm SSUBW : SIMDWideThreeVectorBHS<0, 0b0011, "ssubw",
2811 BinOpFrag<(sub node:$LHS, (sext node:$RHS))>>;
2812 defm UABAL : SIMDLongThreeVectorTiedBHSabal<1, 0b0101, "uabal",
2813 int_arm64_neon_uabd>;
2814 defm UABDL : SIMDLongThreeVectorBHSabdl<1, 0b0111, "uabdl",
2815 int_arm64_neon_uabd>;
2816 defm UADDL : SIMDLongThreeVectorBHS<1, 0b0000, "uaddl",
2817 BinOpFrag<(add (zext node:$LHS), (zext node:$RHS))>>;
2818 defm UADDW : SIMDWideThreeVectorBHS<1, 0b0001, "uaddw",
2819 BinOpFrag<(add node:$LHS, (zext node:$RHS))>>;
2820 defm UMLAL : SIMDLongThreeVectorTiedBHS<1, 0b1000, "umlal",
2821 TriOpFrag<(add node:$LHS, (int_arm64_neon_umull node:$MHS, node:$RHS))>>;
2822 defm UMLSL : SIMDLongThreeVectorTiedBHS<1, 0b1010, "umlsl",
2823 TriOpFrag<(sub node:$LHS, (int_arm64_neon_umull node:$MHS, node:$RHS))>>;
2824 defm UMULL : SIMDLongThreeVectorBHS<1, 0b1100, "umull", int_arm64_neon_umull>;
2825 defm USUBL : SIMDLongThreeVectorBHS<1, 0b0010, "usubl",
2826 BinOpFrag<(sub (zext node:$LHS), (zext node:$RHS))>>;
2827 defm USUBW : SIMDWideThreeVectorBHS< 1, 0b0011, "usubw",
2828 BinOpFrag<(sub node:$LHS, (zext node:$RHS))>>;
2830 // Patterns for 64-bit pmull
2831 def : Pat<(int_arm64_neon_pmull64 V64:$Rn, V64:$Rm),
2832 (PMULLv1i64 V64:$Rn, V64:$Rm)>;
2833 def : Pat<(int_arm64_neon_pmull64 (vector_extract (v2i64 V128:$Rn), (i64 1)),
2834 (vector_extract (v2i64 V128:$Rm), (i64 1))),
2835 (PMULLv2i64 V128:$Rn, V128:$Rm)>;
2837 // CodeGen patterns for addhn and subhn instructions, which can actually be
2838 // written in LLVM IR without too much difficulty.
2841 def : Pat<(v8i8 (trunc (v8i16 (ARM64vlshr (add V128:$Rn, V128:$Rm), (i32 8))))),
2842 (ADDHNv8i16_v8i8 V128:$Rn, V128:$Rm)>;
2843 def : Pat<(v4i16 (trunc (v4i32 (ARM64vlshr (add V128:$Rn, V128:$Rm),
2845 (ADDHNv4i32_v4i16 V128:$Rn, V128:$Rm)>;
2846 def : Pat<(v2i32 (trunc (v2i64 (ARM64vlshr (add V128:$Rn, V128:$Rm),
2848 (ADDHNv2i64_v2i32 V128:$Rn, V128:$Rm)>;
2849 def : Pat<(concat_vectors (v8i8 V64:$Rd),
2850 (trunc (v8i16 (ARM64vlshr (add V128:$Rn, V128:$Rm),
2852 (ADDHNv8i16_v16i8 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
2853 V128:$Rn, V128:$Rm)>;
2854 def : Pat<(concat_vectors (v4i16 V64:$Rd),
2855 (trunc (v4i32 (ARM64vlshr (add V128:$Rn, V128:$Rm),
2857 (ADDHNv4i32_v8i16 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
2858 V128:$Rn, V128:$Rm)>;
2859 def : Pat<(concat_vectors (v2i32 V64:$Rd),
2860 (trunc (v2i64 (ARM64vlshr (add V128:$Rn, V128:$Rm),
2862 (ADDHNv2i64_v4i32 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
2863 V128:$Rn, V128:$Rm)>;
2866 def : Pat<(v8i8 (trunc (v8i16 (ARM64vlshr (sub V128:$Rn, V128:$Rm), (i32 8))))),
2867 (SUBHNv8i16_v8i8 V128:$Rn, V128:$Rm)>;
2868 def : Pat<(v4i16 (trunc (v4i32 (ARM64vlshr (sub V128:$Rn, V128:$Rm),
2870 (SUBHNv4i32_v4i16 V128:$Rn, V128:$Rm)>;
2871 def : Pat<(v2i32 (trunc (v2i64 (ARM64vlshr (sub V128:$Rn, V128:$Rm),
2873 (SUBHNv2i64_v2i32 V128:$Rn, V128:$Rm)>;
2874 def : Pat<(concat_vectors (v8i8 V64:$Rd),
2875 (trunc (v8i16 (ARM64vlshr (sub V128:$Rn, V128:$Rm),
2877 (SUBHNv8i16_v16i8 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
2878 V128:$Rn, V128:$Rm)>;
2879 def : Pat<(concat_vectors (v4i16 V64:$Rd),
2880 (trunc (v4i32 (ARM64vlshr (sub V128:$Rn, V128:$Rm),
2882 (SUBHNv4i32_v8i16 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
2883 V128:$Rn, V128:$Rm)>;
2884 def : Pat<(concat_vectors (v2i32 V64:$Rd),
2885 (trunc (v2i64 (ARM64vlshr (sub V128:$Rn, V128:$Rm),
2887 (SUBHNv2i64_v4i32 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
2888 V128:$Rn, V128:$Rm)>;
2890 //----------------------------------------------------------------------------
2891 // AdvSIMD bitwise extract from vector instruction.
2892 //----------------------------------------------------------------------------
2894 defm EXT : SIMDBitwiseExtract<"ext">;
2896 def : Pat<(v4i16 (ARM64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
2897 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
2898 def : Pat<(v8i16 (ARM64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
2899 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
2900 def : Pat<(v2i32 (ARM64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
2901 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
2902 def : Pat<(v2f32 (ARM64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
2903 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
2904 def : Pat<(v4i32 (ARM64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
2905 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
2906 def : Pat<(v4f32 (ARM64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
2907 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
2908 def : Pat<(v2i64 (ARM64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
2909 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
2910 def : Pat<(v2f64 (ARM64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
2911 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
2913 // We use EXT to handle extract_subvector to copy the upper 64-bits of a
2915 def : Pat<(v8i8 (extract_subvector V128:$Rn, (i64 8))),
2916 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
2917 def : Pat<(v4i16 (extract_subvector V128:$Rn, (i64 4))),
2918 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
2919 def : Pat<(v2i32 (extract_subvector V128:$Rn, (i64 2))),
2920 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
2921 def : Pat<(v1i64 (extract_subvector V128:$Rn, (i64 1))),
2922 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
2923 def : Pat<(v2f32 (extract_subvector V128:$Rn, (i64 2))),
2924 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
2925 def : Pat<(v1f64 (extract_subvector V128:$Rn, (i64 1))),
2926 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
2929 //----------------------------------------------------------------------------
2930 // AdvSIMD zip vector
2931 //----------------------------------------------------------------------------
2933 defm TRN1 : SIMDZipVector<0b010, "trn1", ARM64trn1>;
2934 defm TRN2 : SIMDZipVector<0b110, "trn2", ARM64trn2>;
2935 defm UZP1 : SIMDZipVector<0b001, "uzp1", ARM64uzp1>;
2936 defm UZP2 : SIMDZipVector<0b101, "uzp2", ARM64uzp2>;
2937 defm ZIP1 : SIMDZipVector<0b011, "zip1", ARM64zip1>;
2938 defm ZIP2 : SIMDZipVector<0b111, "zip2", ARM64zip2>;
2940 //----------------------------------------------------------------------------
2941 // AdvSIMD TBL/TBX instructions
2942 //----------------------------------------------------------------------------
2944 defm TBL : SIMDTableLookup< 0, "tbl">;
2945 defm TBX : SIMDTableLookupTied<1, "tbx">;
2947 def : Pat<(v8i8 (int_arm64_neon_tbl1 (v16i8 VecListOne128:$Rn), (v8i8 V64:$Ri))),
2948 (TBLv8i8One VecListOne128:$Rn, V64:$Ri)>;
2949 def : Pat<(v16i8 (int_arm64_neon_tbl1 (v16i8 V128:$Ri), (v16i8 V128:$Rn))),
2950 (TBLv16i8One V128:$Ri, V128:$Rn)>;
2952 def : Pat<(v8i8 (int_arm64_neon_tbx1 (v8i8 V64:$Rd),
2953 (v16i8 VecListOne128:$Rn), (v8i8 V64:$Ri))),
2954 (TBXv8i8One V64:$Rd, VecListOne128:$Rn, V64:$Ri)>;
2955 def : Pat<(v16i8 (int_arm64_neon_tbx1 (v16i8 V128:$Rd),
2956 (v16i8 V128:$Ri), (v16i8 V128:$Rn))),
2957 (TBXv16i8One V128:$Rd, V128:$Ri, V128:$Rn)>;
2960 //----------------------------------------------------------------------------
2961 // AdvSIMD scalar CPY instruction
2962 //----------------------------------------------------------------------------
2964 defm CPY : SIMDScalarCPY<"cpy">;
2966 //----------------------------------------------------------------------------
2967 // AdvSIMD scalar pairwise instructions
2968 //----------------------------------------------------------------------------
2970 defm ADDP : SIMDPairwiseScalarD<0, 0b11011, "addp">;
2971 defm FADDP : SIMDPairwiseScalarSD<1, 0, 0b01101, "faddp">;
2972 defm FMAXNMP : SIMDPairwiseScalarSD<1, 0, 0b01100, "fmaxnmp">;
2973 defm FMAXP : SIMDPairwiseScalarSD<1, 0, 0b01111, "fmaxp">;
2974 defm FMINNMP : SIMDPairwiseScalarSD<1, 1, 0b01100, "fminnmp">;
2975 defm FMINP : SIMDPairwiseScalarSD<1, 1, 0b01111, "fminp">;
2976 def : Pat<(i64 (int_arm64_neon_saddv (v2i64 V128:$Rn))),
2977 (ADDPv2i64p V128:$Rn)>;
2978 def : Pat<(i64 (int_arm64_neon_uaddv (v2i64 V128:$Rn))),
2979 (ADDPv2i64p V128:$Rn)>;
2980 def : Pat<(f32 (int_arm64_neon_faddv (v2f32 V64:$Rn))),
2981 (FADDPv2i32p V64:$Rn)>;
2982 def : Pat<(f32 (int_arm64_neon_faddv (v4f32 V128:$Rn))),
2983 (FADDPv2i32p (EXTRACT_SUBREG (FADDPv4f32 V128:$Rn, V128:$Rn), dsub))>;
2984 def : Pat<(f64 (int_arm64_neon_faddv (v2f64 V128:$Rn))),
2985 (FADDPv2i64p V128:$Rn)>;
2986 def : Pat<(f32 (int_arm64_neon_fmaxnmv (v2f32 V64:$Rn))),
2987 (FMAXNMPv2i32p V64:$Rn)>;
2988 def : Pat<(f64 (int_arm64_neon_fmaxnmv (v2f64 V128:$Rn))),
2989 (FMAXNMPv2i64p V128:$Rn)>;
2990 def : Pat<(f32 (int_arm64_neon_fmaxv (v2f32 V64:$Rn))),
2991 (FMAXPv2i32p V64:$Rn)>;
2992 def : Pat<(f64 (int_arm64_neon_fmaxv (v2f64 V128:$Rn))),
2993 (FMAXPv2i64p V128:$Rn)>;
2994 def : Pat<(f32 (int_arm64_neon_fminnmv (v2f32 V64:$Rn))),
2995 (FMINNMPv2i32p V64:$Rn)>;
2996 def : Pat<(f64 (int_arm64_neon_fminnmv (v2f64 V128:$Rn))),
2997 (FMINNMPv2i64p V128:$Rn)>;
2998 def : Pat<(f32 (int_arm64_neon_fminv (v2f32 V64:$Rn))),
2999 (FMINPv2i32p V64:$Rn)>;
3000 def : Pat<(f64 (int_arm64_neon_fminv (v2f64 V128:$Rn))),
3001 (FMINPv2i64p V128:$Rn)>;
3003 //----------------------------------------------------------------------------
3004 // AdvSIMD INS/DUP instructions
3005 //----------------------------------------------------------------------------
3007 def DUPv8i8gpr : SIMDDupFromMain<0, 0b00001, ".8b", v8i8, V64, GPR32>;
3008 def DUPv16i8gpr : SIMDDupFromMain<1, 0b00001, ".16b", v16i8, V128, GPR32>;
3009 def DUPv4i16gpr : SIMDDupFromMain<0, 0b00010, ".4h", v4i16, V64, GPR32>;
3010 def DUPv8i16gpr : SIMDDupFromMain<1, 0b00010, ".8h", v8i16, V128, GPR32>;
3011 def DUPv2i32gpr : SIMDDupFromMain<0, 0b00100, ".2s", v2i32, V64, GPR32>;
3012 def DUPv4i32gpr : SIMDDupFromMain<1, 0b00100, ".4s", v4i32, V128, GPR32>;
3013 def DUPv2i64gpr : SIMDDupFromMain<1, 0b01000, ".2d", v2i64, V128, GPR64>;
3015 def DUPv2i64lane : SIMDDup64FromElement;
3016 def DUPv2i32lane : SIMDDup32FromElement<0, ".2s", v2i32, V64>;
3017 def DUPv4i32lane : SIMDDup32FromElement<1, ".4s", v4i32, V128>;
3018 def DUPv4i16lane : SIMDDup16FromElement<0, ".4h", v4i16, V64>;
3019 def DUPv8i16lane : SIMDDup16FromElement<1, ".8h", v8i16, V128>;
3020 def DUPv8i8lane : SIMDDup8FromElement <0, ".8b", v8i8, V64>;
3021 def DUPv16i8lane : SIMDDup8FromElement <1, ".16b", v16i8, V128>;
3023 def : Pat<(v2f32 (ARM64dup (f32 FPR32:$Rn))),
3024 (v2f32 (DUPv2i32lane
3025 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rn, ssub),
3027 def : Pat<(v4f32 (ARM64dup (f32 FPR32:$Rn))),
3028 (v4f32 (DUPv4i32lane
3029 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rn, ssub),
3031 def : Pat<(v2f64 (ARM64dup (f64 FPR64:$Rn))),
3032 (v2f64 (DUPv2i64lane
3033 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$Rn, dsub),
3036 def : Pat<(v2f32 (ARM64duplane32 (v4f32 V128:$Rn), VectorIndexS:$imm)),
3037 (DUPv2i32lane V128:$Rn, VectorIndexS:$imm)>;
3038 def : Pat<(v4f32 (ARM64duplane32 (v4f32 V128:$Rn), VectorIndexS:$imm)),
3039 (DUPv4i32lane V128:$Rn, VectorIndexS:$imm)>;
3040 def : Pat<(v2f64 (ARM64duplane64 (v2f64 V128:$Rn), VectorIndexD:$imm)),
3041 (DUPv2i64lane V128:$Rn, VectorIndexD:$imm)>;
3043 // If there's an (ARM64dup (vector_extract ...) ...), we can use a duplane
3044 // instruction even if the types don't match: we just have to remap the lane
3045 // carefully. N.b. this trick only applies to truncations.
3046 def VecIndex_x2 : SDNodeXForm<imm, [{
3047 return CurDAG->getTargetConstant(2 * N->getZExtValue(), MVT::i64);
3049 def VecIndex_x4 : SDNodeXForm<imm, [{
3050 return CurDAG->getTargetConstant(4 * N->getZExtValue(), MVT::i64);
3052 def VecIndex_x8 : SDNodeXForm<imm, [{
3053 return CurDAG->getTargetConstant(8 * N->getZExtValue(), MVT::i64);
3056 multiclass DUPWithTruncPats<ValueType ResVT, ValueType Src64VT,
3057 ValueType Src128VT, ValueType ScalVT,
3058 Instruction DUP, SDNodeXForm IdxXFORM> {
3059 def : Pat<(ResVT (ARM64dup (ScalVT (vector_extract (Src128VT V128:$Rn),
3061 (DUP V128:$Rn, (IdxXFORM imm:$idx))>;
3063 def : Pat<(ResVT (ARM64dup (ScalVT (vector_extract (Src64VT V64:$Rn),
3065 (DUP (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), (IdxXFORM imm:$idx))>;
3068 defm : DUPWithTruncPats<v8i8, v4i16, v8i16, i32, DUPv8i8lane, VecIndex_x2>;
3069 defm : DUPWithTruncPats<v8i8, v2i32, v4i32, i32, DUPv8i8lane, VecIndex_x4>;
3070 defm : DUPWithTruncPats<v4i16, v2i32, v4i32, i32, DUPv4i16lane, VecIndex_x2>;
3072 defm : DUPWithTruncPats<v16i8, v4i16, v8i16, i32, DUPv16i8lane, VecIndex_x2>;
3073 defm : DUPWithTruncPats<v16i8, v2i32, v4i32, i32, DUPv16i8lane, VecIndex_x4>;
3074 defm : DUPWithTruncPats<v8i16, v2i32, v4i32, i32, DUPv8i16lane, VecIndex_x2>;
3076 multiclass DUPWithTrunci64Pats<ValueType ResVT, Instruction DUP,
3077 SDNodeXForm IdxXFORM> {
3078 def : Pat<(ResVT (ARM64dup (i32 (trunc (vector_extract (v2i64 V128:$Rn),
3080 (DUP V128:$Rn, (IdxXFORM imm:$idx))>;
3082 def : Pat<(ResVT (ARM64dup (i32 (trunc (vector_extract (v1i64 V64:$Rn),
3084 (DUP (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), (IdxXFORM imm:$idx))>;
3087 defm : DUPWithTrunci64Pats<v8i8, DUPv8i8lane, VecIndex_x8>;
3088 defm : DUPWithTrunci64Pats<v4i16, DUPv4i16lane, VecIndex_x4>;
3089 defm : DUPWithTrunci64Pats<v2i32, DUPv2i32lane, VecIndex_x2>;
3091 defm : DUPWithTrunci64Pats<v16i8, DUPv16i8lane, VecIndex_x8>;
3092 defm : DUPWithTrunci64Pats<v8i16, DUPv8i16lane, VecIndex_x4>;
3093 defm : DUPWithTrunci64Pats<v4i32, DUPv4i32lane, VecIndex_x2>;
3095 // SMOV and UMOV definitions, with some extra patterns for convenience
3099 def : Pat<(sext_inreg (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx), i8),
3100 (i32 (SMOVvi8to32 V128:$Rn, VectorIndexB:$idx))>;
3101 def : Pat<(sext_inreg (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx), i8),
3102 (i64 (SMOVvi8to64 V128:$Rn, VectorIndexB:$idx))>;
3103 def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
3104 (i32 (SMOVvi16to32 V128:$Rn, VectorIndexH:$idx))>;
3105 def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
3106 (i64 (SMOVvi16to64 V128:$Rn, VectorIndexH:$idx))>;
3107 def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
3108 (i32 (SMOVvi16to32 V128:$Rn, VectorIndexH:$idx))>;
3109 def : Pat<(sext (i32 (vector_extract (v4i32 V128:$Rn), VectorIndexS:$idx))),
3110 (i64 (SMOVvi32to64 V128:$Rn, VectorIndexS:$idx))>;
3112 // Extracting i8 or i16 elements will have the zero-extend transformed to
3113 // an 'and' mask by type legalization since neither i8 nor i16 are legal types
3114 // for ARM64. Match these patterns here since UMOV already zeroes out the high
3115 // bits of the destination register.
3116 def : Pat<(and (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx),
3118 (i32 (UMOVvi8 V128:$Rn, VectorIndexB:$idx))>;
3119 def : Pat<(and (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),
3121 (i32 (UMOVvi16 V128:$Rn, VectorIndexH:$idx))>;
3125 def : Pat<(v16i8 (scalar_to_vector GPR32:$Rn)),
3126 (SUBREG_TO_REG (i32 0),
3127 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3128 def : Pat<(v8i8 (scalar_to_vector GPR32:$Rn)),
3129 (SUBREG_TO_REG (i32 0),
3130 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3132 def : Pat<(v8i16 (scalar_to_vector GPR32:$Rn)),
3133 (SUBREG_TO_REG (i32 0),
3134 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3135 def : Pat<(v4i16 (scalar_to_vector GPR32:$Rn)),
3136 (SUBREG_TO_REG (i32 0),
3137 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3139 def : Pat<(v2i32 (scalar_to_vector (i32 FPR32:$Rn))),
3140 (v2i32 (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)),
3141 (i32 FPR32:$Rn), ssub))>;
3142 def : Pat<(v4i32 (scalar_to_vector (i32 FPR32:$Rn))),
3143 (v4i32 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
3144 (i32 FPR32:$Rn), ssub))>;
3145 def : Pat<(v2i64 (scalar_to_vector (i64 FPR64:$Rn))),
3146 (v2i64 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
3147 (i64 FPR64:$Rn), dsub))>;
3149 def : Pat<(v4f32 (scalar_to_vector (f32 FPR32:$Rn))),
3150 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rn, ssub)>;
3151 def : Pat<(v2f32 (scalar_to_vector (f32 FPR32:$Rn))),
3152 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), FPR32:$Rn, ssub)>;
3153 def : Pat<(v2f64 (scalar_to_vector (f64 FPR64:$Rn))),
3154 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$Rn, dsub)>;
3156 def : Pat<(v2f32 (vector_insert (v2f32 V64:$Rn),
3157 (f32 FPR32:$Rm), (i64 VectorIndexS:$imm))),
3160 (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), V64:$Rn, dsub)),
3162 (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rm, ssub)),
3165 def : Pat<(v4f32 (vector_insert (v4f32 V128:$Rn),
3166 (f32 FPR32:$Rm), (i64 VectorIndexS:$imm))),
3168 V128:$Rn, VectorIndexS:$imm,
3169 (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rm, ssub)),
3171 def : Pat<(v2f64 (vector_insert (v2f64 V128:$Rn),
3172 (f64 FPR64:$Rm), (i64 VectorIndexD:$imm))),
3174 V128:$Rn, VectorIndexD:$imm,
3175 (v2f64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$Rm, dsub)),
3178 // Copy an element at a constant index in one vector into a constant indexed
3179 // element of another.
3180 // FIXME refactor to a shared class/dev parameterized on vector type, vector
3181 // index type and INS extension
3182 def : Pat<(v16i8 (int_arm64_neon_vcopy_lane
3183 (v16i8 V128:$Vd), VectorIndexB:$idx, (v16i8 V128:$Vs),
3184 VectorIndexB:$idx2)),
3186 V128:$Vd, VectorIndexB:$idx, V128:$Vs, VectorIndexB:$idx2)
3188 def : Pat<(v8i16 (int_arm64_neon_vcopy_lane
3189 (v8i16 V128:$Vd), VectorIndexH:$idx, (v8i16 V128:$Vs),
3190 VectorIndexH:$idx2)),
3192 V128:$Vd, VectorIndexH:$idx, V128:$Vs, VectorIndexH:$idx2)
3194 def : Pat<(v4i32 (int_arm64_neon_vcopy_lane
3195 (v4i32 V128:$Vd), VectorIndexS:$idx, (v4i32 V128:$Vs),
3196 VectorIndexS:$idx2)),
3198 V128:$Vd, VectorIndexS:$idx, V128:$Vs, VectorIndexS:$idx2)
3200 def : Pat<(v2i64 (int_arm64_neon_vcopy_lane
3201 (v2i64 V128:$Vd), VectorIndexD:$idx, (v2i64 V128:$Vs),
3202 VectorIndexD:$idx2)),
3204 V128:$Vd, VectorIndexD:$idx, V128:$Vs, VectorIndexD:$idx2)
3207 multiclass Neon_INS_elt_pattern<ValueType VT128, ValueType VT64,
3208 ValueType VTScal, Instruction INS> {
3209 def : Pat<(VT128 (vector_insert V128:$src,
3210 (VTScal (vector_extract (VT128 V128:$Rn), imm:$Immn)),
3212 (INS V128:$src, imm:$Immd, V128:$Rn, imm:$Immn)>;
3214 def : Pat<(VT128 (vector_insert V128:$src,
3215 (VTScal (vector_extract (VT64 V64:$Rn), imm:$Immn)),
3217 (INS V128:$src, imm:$Immd,
3218 (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), imm:$Immn)>;
3220 def : Pat<(VT64 (vector_insert V64:$src,
3221 (VTScal (vector_extract (VT128 V128:$Rn), imm:$Immn)),
3223 (EXTRACT_SUBREG (INS (SUBREG_TO_REG (i64 0), V64:$src, dsub),
3224 imm:$Immd, V128:$Rn, imm:$Immn),
3227 def : Pat<(VT64 (vector_insert V64:$src,
3228 (VTScal (vector_extract (VT64 V64:$Rn), imm:$Immn)),
3231 (INS (SUBREG_TO_REG (i64 0), V64:$src, dsub), imm:$Immd,
3232 (SUBREG_TO_REG (i64 0), V64:$Rn, dsub), imm:$Immn),
3236 defm : Neon_INS_elt_pattern<v4f32, v2f32, f32, INSvi32lane>;
3237 defm : Neon_INS_elt_pattern<v2f64, v1f64, f64, INSvi64lane>;
3238 defm : Neon_INS_elt_pattern<v16i8, v8i8, i32, INSvi8lane>;
3239 defm : Neon_INS_elt_pattern<v8i16, v4i16, i32, INSvi16lane>;
3240 defm : Neon_INS_elt_pattern<v4i32, v2i32, i32, INSvi32lane>;
3241 defm : Neon_INS_elt_pattern<v2i64, v1i64, i64, INSvi32lane>;
3244 // Floating point vector extractions are codegen'd as either a sequence of
3245 // subregister extractions, possibly fed by an INS if the lane number is
3246 // anything other than zero.
3247 def : Pat<(vector_extract (v2f64 V128:$Rn), 0),
3248 (f64 (EXTRACT_SUBREG V128:$Rn, dsub))>;
3249 def : Pat<(vector_extract (v4f32 V128:$Rn), 0),
3250 (f32 (EXTRACT_SUBREG V128:$Rn, ssub))>;
3251 def : Pat<(vector_extract (v2f64 V128:$Rn), VectorIndexD:$idx),
3252 (f64 (EXTRACT_SUBREG
3253 (INSvi64lane (v2f64 (IMPLICIT_DEF)), 0,
3254 V128:$Rn, VectorIndexD:$idx),
3256 def : Pat<(vector_extract (v4f32 V128:$Rn), VectorIndexS:$idx),
3257 (f32 (EXTRACT_SUBREG
3258 (INSvi32lane (v4f32 (IMPLICIT_DEF)), 0,
3259 V128:$Rn, VectorIndexS:$idx),
3262 // All concat_vectors operations are canonicalised to act on i64 vectors for
3263 // ARM64. In the general case we need an instruction, which had just as well be
3265 class ConcatPat<ValueType DstTy, ValueType SrcTy>
3266 : Pat<(DstTy (concat_vectors (SrcTy V64:$Rd), V64:$Rn)),
3267 (INSvi64lane (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), 1,
3268 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rn, dsub), 0)>;
3270 def : ConcatPat<v2i64, v1i64>;
3271 def : ConcatPat<v2f64, v1f64>;
3272 def : ConcatPat<v4i32, v2i32>;
3273 def : ConcatPat<v4f32, v2f32>;
3274 def : ConcatPat<v8i16, v4i16>;
3275 def : ConcatPat<v16i8, v8i8>;
3277 // If the high lanes are undef, though, we can just ignore them:
3278 class ConcatUndefPat<ValueType DstTy, ValueType SrcTy>
3279 : Pat<(DstTy (concat_vectors (SrcTy V64:$Rn), undef)),
3280 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rn, dsub)>;
3282 def : ConcatUndefPat<v2i64, v1i64>;
3283 def : ConcatUndefPat<v2f64, v1f64>;
3284 def : ConcatUndefPat<v4i32, v2i32>;
3285 def : ConcatUndefPat<v4f32, v2f32>;
3286 def : ConcatUndefPat<v8i16, v4i16>;
3287 def : ConcatUndefPat<v16i8, v8i8>;
3289 //----------------------------------------------------------------------------
3290 // AdvSIMD across lanes instructions
3291 //----------------------------------------------------------------------------
3293 defm ADDV : SIMDAcrossLanesBHS<0, 0b11011, "addv">;
3294 defm SMAXV : SIMDAcrossLanesBHS<0, 0b01010, "smaxv">;
3295 defm SMINV : SIMDAcrossLanesBHS<0, 0b11010, "sminv">;
3296 defm UMAXV : SIMDAcrossLanesBHS<1, 0b01010, "umaxv">;
3297 defm UMINV : SIMDAcrossLanesBHS<1, 0b11010, "uminv">;
3298 defm SADDLV : SIMDAcrossLanesHSD<0, 0b00011, "saddlv">;
3299 defm UADDLV : SIMDAcrossLanesHSD<1, 0b00011, "uaddlv">;
3300 defm FMAXNMV : SIMDAcrossLanesS<0b01100, 0, "fmaxnmv", int_arm64_neon_fmaxnmv>;
3301 defm FMAXV : SIMDAcrossLanesS<0b01111, 0, "fmaxv", int_arm64_neon_fmaxv>;
3302 defm FMINNMV : SIMDAcrossLanesS<0b01100, 1, "fminnmv", int_arm64_neon_fminnmv>;
3303 defm FMINV : SIMDAcrossLanesS<0b01111, 1, "fminv", int_arm64_neon_fminv>;
3305 multiclass SIMDAcrossLanesSignedIntrinsic<string baseOpc, Intrinsic intOp> {
3306 // If there is a sign extension after this intrinsic, consume it as smov already
3308 def : Pat<(i32 (sext_inreg (i32 (intOp (v8i8 V64:$Rn))), i8)),
3310 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3311 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3313 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
3315 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3316 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3318 // If there is a sign extension after this intrinsic, consume it as smov already
3320 def : Pat<(i32 (sext_inreg (i32 (intOp (v16i8 V128:$Rn))), i8)),
3322 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3323 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3325 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
3327 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3328 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3330 // If there is a sign extension after this intrinsic, consume it as smov already
3332 def : Pat<(i32 (sext_inreg (i32 (intOp (v4i16 V64:$Rn))), i16)),
3334 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3335 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3337 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
3339 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3340 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3342 // If there is a sign extension after this intrinsic, consume it as smov already
3344 def : Pat<(i32 (sext_inreg (i32 (intOp (v8i16 V128:$Rn))), i16)),
3346 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3347 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3349 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
3351 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3352 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3355 def : Pat<(i32 (intOp (v4i32 V128:$Rn))),
3356 (i32 (EXTRACT_SUBREG
3357 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3358 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), ssub),
3362 multiclass SIMDAcrossLanesUnsignedIntrinsic<string baseOpc, Intrinsic intOp> {
3363 // If there is a masking operation keeping only what has been actually
3364 // generated, consume it.
3365 def : Pat<(i32 (and (i32 (intOp (v8i8 V64:$Rn))), maski8_or_more)),
3366 (i32 (EXTRACT_SUBREG
3367 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3368 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3370 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
3371 (i32 (EXTRACT_SUBREG
3372 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3373 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3375 // If there is a masking operation keeping only what has been actually
3376 // generated, consume it.
3377 def : Pat<(i32 (and (i32 (intOp (v16i8 V128:$Rn))), maski8_or_more)),
3378 (i32 (EXTRACT_SUBREG
3379 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3380 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3382 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
3383 (i32 (EXTRACT_SUBREG
3384 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3385 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3388 // If there is a masking operation keeping only what has been actually
3389 // generated, consume it.
3390 def : Pat<(i32 (and (i32 (intOp (v4i16 V64:$Rn))), maski16_or_more)),
3391 (i32 (EXTRACT_SUBREG
3392 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3393 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3395 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
3396 (i32 (EXTRACT_SUBREG
3397 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3398 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3400 // If there is a masking operation keeping only what has been actually
3401 // generated, consume it.
3402 def : Pat<(i32 (and (i32 (intOp (v8i16 V128:$Rn))), maski16_or_more)),
3403 (i32 (EXTRACT_SUBREG
3404 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3405 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3407 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
3408 (i32 (EXTRACT_SUBREG
3409 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3410 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3413 def : Pat<(i32 (intOp (v4i32 V128:$Rn))),
3414 (i32 (EXTRACT_SUBREG
3415 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3416 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), ssub),
3421 multiclass SIMDAcrossLanesSignedLongIntrinsic<string baseOpc, Intrinsic intOp> {
3422 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
3424 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3425 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), hsub),
3427 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
3429 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3430 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), hsub),
3433 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
3434 (i32 (EXTRACT_SUBREG
3435 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3436 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), ssub),
3438 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
3439 (i32 (EXTRACT_SUBREG
3440 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3441 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), ssub),
3444 def : Pat<(i64 (intOp (v4i32 V128:$Rn))),
3445 (i64 (EXTRACT_SUBREG
3446 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3447 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), dsub),
3451 multiclass SIMDAcrossLanesUnsignedLongIntrinsic<string baseOpc,
3453 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
3454 (i32 (EXTRACT_SUBREG
3455 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3456 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), hsub),
3458 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
3459 (i32 (EXTRACT_SUBREG
3460 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3461 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), hsub),
3464 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
3465 (i32 (EXTRACT_SUBREG
3466 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3467 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), ssub),
3469 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
3470 (i32 (EXTRACT_SUBREG
3471 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3472 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), ssub),
3475 def : Pat<(i64 (intOp (v4i32 V128:$Rn))),
3476 (i64 (EXTRACT_SUBREG
3477 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3478 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), dsub),
3482 defm : SIMDAcrossLanesSignedIntrinsic<"ADDV", int_arm64_neon_saddv>;
3483 // vaddv_[su]32 is special; -> ADDP Vd.2S,Vn.2S,Vm.2S; return Vd.s[0];Vn==Vm
3484 def : Pat<(i32 (int_arm64_neon_saddv (v2i32 V64:$Rn))),
3485 (EXTRACT_SUBREG (ADDPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3487 defm : SIMDAcrossLanesUnsignedIntrinsic<"ADDV", int_arm64_neon_uaddv>;
3488 // vaddv_[su]32 is special; -> ADDP Vd.2S,Vn.2S,Vm.2S; return Vd.s[0];Vn==Vm
3489 def : Pat<(i32 (int_arm64_neon_uaddv (v2i32 V64:$Rn))),
3490 (EXTRACT_SUBREG (ADDPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3492 defm : SIMDAcrossLanesSignedIntrinsic<"SMAXV", int_arm64_neon_smaxv>;
3493 def : Pat<(i32 (int_arm64_neon_smaxv (v2i32 V64:$Rn))),
3494 (EXTRACT_SUBREG (SMAXPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3496 defm : SIMDAcrossLanesSignedIntrinsic<"SMINV", int_arm64_neon_sminv>;
3497 def : Pat<(i32 (int_arm64_neon_sminv (v2i32 V64:$Rn))),
3498 (EXTRACT_SUBREG (SMINPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3500 defm : SIMDAcrossLanesUnsignedIntrinsic<"UMAXV", int_arm64_neon_umaxv>;
3501 def : Pat<(i32 (int_arm64_neon_umaxv (v2i32 V64:$Rn))),
3502 (EXTRACT_SUBREG (UMAXPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3504 defm : SIMDAcrossLanesUnsignedIntrinsic<"UMINV", int_arm64_neon_uminv>;
3505 def : Pat<(i32 (int_arm64_neon_uminv (v2i32 V64:$Rn))),
3506 (EXTRACT_SUBREG (UMINPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3508 defm : SIMDAcrossLanesSignedLongIntrinsic<"SADDLV", int_arm64_neon_saddlv>;
3509 defm : SIMDAcrossLanesUnsignedLongIntrinsic<"UADDLV", int_arm64_neon_uaddlv>;
3511 // The vaddlv_s32 intrinsic gets mapped to SADDLP.
3512 def : Pat<(i64 (int_arm64_neon_saddlv (v2i32 V64:$Rn))),
3513 (i64 (EXTRACT_SUBREG
3514 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3515 (SADDLPv2i32_v1i64 V64:$Rn), dsub),
3517 // The vaddlv_u32 intrinsic gets mapped to UADDLP.
3518 def : Pat<(i64 (int_arm64_neon_uaddlv (v2i32 V64:$Rn))),
3519 (i64 (EXTRACT_SUBREG
3520 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3521 (UADDLPv2i32_v1i64 V64:$Rn), dsub),
3524 //------------------------------------------------------------------------------
3525 // AdvSIMD modified immediate instructions
3526 //------------------------------------------------------------------------------
3529 defm BIC : SIMDModifiedImmVectorShiftTied<1, 0b11, 0b01, "bic", ARM64bici>;
3531 defm ORR : SIMDModifiedImmVectorShiftTied<0, 0b11, 0b01, "orr", ARM64orri>;
3535 def FMOVv2f64_ns : SIMDModifiedImmVectorNoShift<1, 1, 0b1111, V128, fpimm8,
3537 [(set (v2f64 V128:$Rd), (ARM64fmov imm0_255:$imm8))]>;
3538 def FMOVv2f32_ns : SIMDModifiedImmVectorNoShift<0, 0, 0b1111, V64, fpimm8,
3540 [(set (v2f32 V64:$Rd), (ARM64fmov imm0_255:$imm8))]>;
3541 def FMOVv4f32_ns : SIMDModifiedImmVectorNoShift<1, 0, 0b1111, V128, fpimm8,
3543 [(set (v4f32 V128:$Rd), (ARM64fmov imm0_255:$imm8))]>;
3547 // EDIT byte mask: scalar
3548 let isReMaterializable = 1, isAsCheapAsAMove = 1 in
3549 def MOVID : SIMDModifiedImmScalarNoShift<0, 1, 0b1110, "movi",
3550 [(set FPR64:$Rd, simdimmtype10:$imm8)]>;
3551 // The movi_edit node has the immediate value already encoded, so we use
3552 // a plain imm0_255 here.
3553 def : Pat<(f64 (ARM64movi_edit imm0_255:$shift)),
3554 (MOVID imm0_255:$shift)>;
3556 def : Pat<(v1i64 immAllZerosV), (MOVID (i32 0))>;
3557 def : Pat<(v2i32 immAllZerosV), (MOVID (i32 0))>;
3558 def : Pat<(v4i16 immAllZerosV), (MOVID (i32 0))>;
3559 def : Pat<(v8i8 immAllZerosV), (MOVID (i32 0))>;
3561 def : Pat<(v1i64 immAllOnesV), (MOVID (i32 255))>;
3562 def : Pat<(v2i32 immAllOnesV), (MOVID (i32 255))>;
3563 def : Pat<(v4i16 immAllOnesV), (MOVID (i32 255))>;
3564 def : Pat<(v8i8 immAllOnesV), (MOVID (i32 255))>;
3566 // EDIT byte mask: 2d
3568 // The movi_edit node has the immediate value already encoded, so we use
3569 // a plain imm0_255 in the pattern
3570 let isReMaterializable = 1, isAsCheapAsAMove = 1 in
3571 def MOVIv2d_ns : SIMDModifiedImmVectorNoShift<1, 1, 0b1110, V128,
3574 [(set (v2i64 V128:$Rd), (ARM64movi_edit imm0_255:$imm8))]>;
3577 // Use movi.2d to materialize 0.0 if the HW does zero-cycle zeroing.
3578 // Complexity is added to break a tie with a plain MOVI.
3579 let AddedComplexity = 1 in {
3580 def : Pat<(f32 fpimm0),
3581 (f32 (EXTRACT_SUBREG (v2i64 (MOVIv2d_ns (i32 0))), ssub))>,
3583 def : Pat<(f64 fpimm0),
3584 (f64 (EXTRACT_SUBREG (v2i64 (MOVIv2d_ns (i32 0))), dsub))>,
3588 def : Pat<(v2i64 immAllZerosV), (MOVIv2d_ns (i32 0))>;
3589 def : Pat<(v4i32 immAllZerosV), (MOVIv2d_ns (i32 0))>;
3590 def : Pat<(v8i16 immAllZerosV), (MOVIv2d_ns (i32 0))>;
3591 def : Pat<(v16i8 immAllZerosV), (MOVIv2d_ns (i32 0))>;
3593 def : Pat<(v2i64 immAllOnesV), (MOVIv2d_ns (i32 255))>;
3594 def : Pat<(v4i32 immAllOnesV), (MOVIv2d_ns (i32 255))>;
3595 def : Pat<(v8i16 immAllOnesV), (MOVIv2d_ns (i32 255))>;
3596 def : Pat<(v16i8 immAllOnesV), (MOVIv2d_ns (i32 255))>;
3598 def : Pat<(v2f64 (ARM64dup (f64 fpimm0))), (MOVIv2d_ns (i32 0))>;
3599 def : Pat<(v4f32 (ARM64dup (f32 fpimm0))), (MOVIv2d_ns (i32 0))>;
3601 // EDIT per word & halfword: 2s, 4h, 4s, & 8h
3602 defm MOVI : SIMDModifiedImmVectorShift<0, 0b10, 0b00, "movi">;
3603 def : Pat<(v2i32 (ARM64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
3604 (MOVIv2i32 imm0_255:$imm8, imm:$shift)>;
3605 def : Pat<(v4i32 (ARM64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
3606 (MOVIv4i32 imm0_255:$imm8, imm:$shift)>;
3607 def : Pat<(v4i16 (ARM64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
3608 (MOVIv4i16 imm0_255:$imm8, imm:$shift)>;
3609 def : Pat<(v8i16 (ARM64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
3610 (MOVIv8i16 imm0_255:$imm8, imm:$shift)>;
3612 // EDIT per word: 2s & 4s with MSL shifter
3613 def MOVIv2s_msl : SIMDModifiedImmMoveMSL<0, 0, {1,1,0,?}, V64, "movi", ".2s",
3614 [(set (v2i32 V64:$Rd),
3615 (ARM64movi_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
3616 def MOVIv4s_msl : SIMDModifiedImmMoveMSL<1, 0, {1,1,0,?}, V128, "movi", ".4s",
3617 [(set (v4i32 V128:$Rd),
3618 (ARM64movi_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
3620 // Per byte: 8b & 16b
3621 def MOVIv8b_ns : SIMDModifiedImmVectorNoShift<0, 0, 0b1110, V64, imm0_255,
3623 [(set (v8i8 V64:$Rd), (ARM64movi imm0_255:$imm8))]>;
3624 def MOVIv16b_ns : SIMDModifiedImmVectorNoShift<1, 0, 0b1110, V128, imm0_255,
3626 [(set (v16i8 V128:$Rd), (ARM64movi imm0_255:$imm8))]>;
3630 // EDIT per word & halfword: 2s, 4h, 4s, & 8h
3631 defm MVNI : SIMDModifiedImmVectorShift<1, 0b10, 0b00, "mvni">;
3632 def : Pat<(v2i32 (ARM64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
3633 (MVNIv2i32 imm0_255:$imm8, imm:$shift)>;
3634 def : Pat<(v4i32 (ARM64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
3635 (MVNIv4i32 imm0_255:$imm8, imm:$shift)>;
3636 def : Pat<(v4i16 (ARM64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
3637 (MVNIv4i16 imm0_255:$imm8, imm:$shift)>;
3638 def : Pat<(v8i16 (ARM64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
3639 (MVNIv8i16 imm0_255:$imm8, imm:$shift)>;
3641 // EDIT per word: 2s & 4s with MSL shifter
3642 def MVNIv2s_msl : SIMDModifiedImmMoveMSL<0, 1, {1,1,0,?}, V64, "mvni", ".2s",
3643 [(set (v2i32 V64:$Rd),
3644 (ARM64mvni_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
3645 def MVNIv4s_msl : SIMDModifiedImmMoveMSL<1, 1, {1,1,0,?}, V128, "mvni", ".4s",
3646 [(set (v4i32 V128:$Rd),
3647 (ARM64mvni_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
3649 //----------------------------------------------------------------------------
3650 // AdvSIMD indexed element
3651 //----------------------------------------------------------------------------
3653 let neverHasSideEffects = 1 in {
3654 defm FMLA : SIMDFPIndexedSDTied<0, 0b0001, "fmla">;
3655 defm FMLS : SIMDFPIndexedSDTied<0, 0b0101, "fmls">;
3658 // NOTE: Operands are reordered in the FMLA/FMLS PatFrags because the
3659 // instruction expects the addend first, while the intrinsic expects it last.
3661 // On the other hand, there are quite a few valid combinatorial options due to
3662 // the commutativity of multiplication and the fact that (-x) * y = x * (-y).
3663 defm : SIMDFPIndexedSDTiedPatterns<"FMLA",
3664 TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)>>;
3665 defm : SIMDFPIndexedSDTiedPatterns<"FMLA",
3666 TriOpFrag<(fma node:$MHS, node:$RHS, node:$LHS)>>;
3668 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
3669 TriOpFrag<(fma node:$MHS, (fneg node:$RHS), node:$LHS)> >;
3670 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
3671 TriOpFrag<(fma node:$RHS, (fneg node:$MHS), node:$LHS)> >;
3672 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
3673 TriOpFrag<(fma (fneg node:$RHS), node:$MHS, node:$LHS)> >;
3674 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
3675 TriOpFrag<(fma (fneg node:$MHS), node:$RHS, node:$LHS)> >;
3677 multiclass FMLSIndexedAfterNegPatterns<SDPatternOperator OpNode> {
3678 // 3 variants for the .2s version: DUPLANE from 128-bit, DUPLANE from 64-bit
3680 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
3681 (ARM64duplane32 (v4f32 (fneg V128:$Rm)),
3682 VectorIndexS:$idx))),
3683 (FMLSv2i32_indexed V64:$Rd, V64:$Rn, V128:$Rm, VectorIndexS:$idx)>;
3684 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
3685 (v2f32 (ARM64duplane32
3686 (v4f32 (insert_subvector undef,
3687 (v2f32 (fneg V64:$Rm)),
3689 VectorIndexS:$idx)))),
3690 (FMLSv2i32_indexed V64:$Rd, V64:$Rn,
3691 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub),
3692 VectorIndexS:$idx)>;
3693 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
3694 (ARM64dup (f32 (fneg FPR32Op:$Rm))))),
3695 (FMLSv2i32_indexed V64:$Rd, V64:$Rn,
3696 (SUBREG_TO_REG (i32 0), FPR32Op:$Rm, ssub), (i64 0))>;
3698 // 3 variants for the .4s version: DUPLANE from 128-bit, DUPLANE from 64-bit
3700 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
3701 (ARM64duplane32 (v4f32 (fneg V128:$Rm)),
3702 VectorIndexS:$idx))),
3703 (FMLSv4i32_indexed V128:$Rd, V128:$Rn, V128:$Rm,
3704 VectorIndexS:$idx)>;
3705 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
3706 (v4f32 (ARM64duplane32
3707 (v4f32 (insert_subvector undef,
3708 (v2f32 (fneg V64:$Rm)),
3710 VectorIndexS:$idx)))),
3711 (FMLSv4i32_indexed V128:$Rd, V128:$Rn,
3712 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub),
3713 VectorIndexS:$idx)>;
3714 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
3715 (ARM64dup (f32 (fneg FPR32Op:$Rm))))),
3716 (FMLSv4i32_indexed V128:$Rd, V128:$Rn,
3717 (SUBREG_TO_REG (i32 0), FPR32Op:$Rm, ssub), (i64 0))>;
3719 // 2 variants for the .2d version: DUPLANE from 128-bit, and DUP scalar
3720 // (DUPLANE from 64-bit would be trivial).
3721 def : Pat<(v2f64 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn),
3722 (ARM64duplane64 (v2f64 (fneg V128:$Rm)),
3723 VectorIndexD:$idx))),
3725 V128:$Rd, V128:$Rn, V128:$Rm, VectorIndexS:$idx)>;
3726 def : Pat<(v2f64 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn),
3727 (ARM64dup (f64 (fneg FPR64Op:$Rm))))),
3728 (FMLSv2i64_indexed V128:$Rd, V128:$Rn,
3729 (SUBREG_TO_REG (i32 0), FPR64Op:$Rm, dsub), (i64 0))>;
3731 // 2 variants for 32-bit scalar version: extract from .2s or from .4s
3732 def : Pat<(f32 (OpNode (f32 FPR32:$Rd), (f32 FPR32:$Rn),
3733 (vector_extract (v4f32 (fneg V128:$Rm)),
3734 VectorIndexS:$idx))),
3735 (FMLSv1i32_indexed FPR32:$Rd, FPR32:$Rn,
3736 V128:$Rm, VectorIndexS:$idx)>;
3737 def : Pat<(f32 (OpNode (f32 FPR32:$Rd), (f32 FPR32:$Rn),
3738 (vector_extract (v2f32 (fneg V64:$Rm)),
3739 VectorIndexS:$idx))),
3740 (FMLSv1i32_indexed FPR32:$Rd, FPR32:$Rn,
3741 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub), VectorIndexS:$idx)>;
3743 // 1 variant for 64-bit scalar version: extract from .1d or from .2d
3744 def : Pat<(f64 (OpNode (f64 FPR64:$Rd), (f64 FPR64:$Rn),
3745 (vector_extract (v2f64 (fneg V128:$Rm)),
3746 VectorIndexS:$idx))),
3747 (FMLSv1i64_indexed FPR64:$Rd, FPR64:$Rn,
3748 V128:$Rm, VectorIndexS:$idx)>;
3751 defm : FMLSIndexedAfterNegPatterns<
3752 TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)> >;
3753 defm : FMLSIndexedAfterNegPatterns<
3754 TriOpFrag<(fma node:$MHS, node:$RHS, node:$LHS)> >;
3756 defm FMULX : SIMDFPIndexedSD<1, 0b1001, "fmulx", int_arm64_neon_fmulx>;
3757 defm FMUL : SIMDFPIndexedSD<0, 0b1001, "fmul", fmul>;
3759 def : Pat<(v2f32 (fmul V64:$Rn, (ARM64dup (f32 FPR32:$Rm)))),
3760 (FMULv2i32_indexed V64:$Rn,
3761 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rm, ssub),
3763 def : Pat<(v4f32 (fmul V128:$Rn, (ARM64dup (f32 FPR32:$Rm)))),
3764 (FMULv4i32_indexed V128:$Rn,
3765 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rm, ssub),
3767 def : Pat<(v2f64 (fmul V128:$Rn, (ARM64dup (f64 FPR64:$Rm)))),
3768 (FMULv2i64_indexed V128:$Rn,
3769 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$Rm, dsub),
3772 defm SQDMULH : SIMDIndexedHS<0, 0b1100, "sqdmulh", int_arm64_neon_sqdmulh>;
3773 defm SQRDMULH : SIMDIndexedHS<0, 0b1101, "sqrdmulh", int_arm64_neon_sqrdmulh>;
3774 defm MLA : SIMDVectorIndexedHSTied<1, 0b0000, "mla",
3775 TriOpFrag<(add node:$LHS, (mul node:$MHS, node:$RHS))>>;
3776 defm MLS : SIMDVectorIndexedHSTied<1, 0b0100, "mls",
3777 TriOpFrag<(sub node:$LHS, (mul node:$MHS, node:$RHS))>>;
3778 defm MUL : SIMDVectorIndexedHS<0, 0b1000, "mul", mul>;
3779 defm SMLAL : SIMDVectorIndexedLongSDTied<0, 0b0010, "smlal",
3780 TriOpFrag<(add node:$LHS, (int_arm64_neon_smull node:$MHS, node:$RHS))>>;
3781 defm SMLSL : SIMDVectorIndexedLongSDTied<0, 0b0110, "smlsl",
3782 TriOpFrag<(sub node:$LHS, (int_arm64_neon_smull node:$MHS, node:$RHS))>>;
3783 defm SMULL : SIMDVectorIndexedLongSD<0, 0b1010, "smull",
3784 int_arm64_neon_smull>;
3785 defm SQDMLAL : SIMDIndexedLongSQDMLXSDTied<0, 0b0011, "sqdmlal",
3786 int_arm64_neon_sqadd>;
3787 defm SQDMLSL : SIMDIndexedLongSQDMLXSDTied<0, 0b0111, "sqdmlsl",
3788 int_arm64_neon_sqsub>;
3789 defm SQDMULL : SIMDIndexedLongSD<0, 0b1011, "sqdmull", int_arm64_neon_sqdmull>;
3790 defm UMLAL : SIMDVectorIndexedLongSDTied<1, 0b0010, "umlal",
3791 TriOpFrag<(add node:$LHS, (int_arm64_neon_umull node:$MHS, node:$RHS))>>;
3792 defm UMLSL : SIMDVectorIndexedLongSDTied<1, 0b0110, "umlsl",
3793 TriOpFrag<(sub node:$LHS, (int_arm64_neon_umull node:$MHS, node:$RHS))>>;
3794 defm UMULL : SIMDVectorIndexedLongSD<1, 0b1010, "umull",
3795 int_arm64_neon_umull>;
3797 // A scalar sqdmull with the second operand being a vector lane can be
3798 // handled directly with the indexed instruction encoding.
3799 def : Pat<(int_arm64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
3800 (vector_extract (v4i32 V128:$Vm),
3801 VectorIndexS:$idx)),
3802 (SQDMULLv1i64_indexed FPR32:$Rn, V128:$Vm, VectorIndexS:$idx)>;
3804 //----------------------------------------------------------------------------
3805 // AdvSIMD scalar shift instructions
3806 //----------------------------------------------------------------------------
3807 defm FCVTZS : SIMDScalarRShiftSD<0, 0b11111, "fcvtzs">;
3808 defm FCVTZU : SIMDScalarRShiftSD<1, 0b11111, "fcvtzu">;
3809 defm SCVTF : SIMDScalarRShiftSD<0, 0b11100, "scvtf">;
3810 defm UCVTF : SIMDScalarRShiftSD<1, 0b11100, "ucvtf">;
3811 // Codegen patterns for the above. We don't put these directly on the
3812 // instructions because TableGen's type inference can't handle the truth.
3813 // Having the same base pattern for fp <--> int totally freaks it out.
3814 def : Pat<(int_arm64_neon_vcvtfp2fxs FPR32:$Rn, vecshiftR32:$imm),
3815 (FCVTZSs FPR32:$Rn, vecshiftR32:$imm)>;
3816 def : Pat<(int_arm64_neon_vcvtfp2fxu FPR32:$Rn, vecshiftR32:$imm),
3817 (FCVTZUs FPR32:$Rn, vecshiftR32:$imm)>;
3818 def : Pat<(i64 (int_arm64_neon_vcvtfp2fxs (f64 FPR64:$Rn), vecshiftR64:$imm)),
3819 (FCVTZSd FPR64:$Rn, vecshiftR64:$imm)>;
3820 def : Pat<(i64 (int_arm64_neon_vcvtfp2fxu (f64 FPR64:$Rn), vecshiftR64:$imm)),
3821 (FCVTZUd FPR64:$Rn, vecshiftR64:$imm)>;
3822 def : Pat<(v1i64 (int_arm64_neon_vcvtfp2fxs (v1f64 FPR64:$Rn),
3824 (FCVTZSd FPR64:$Rn, vecshiftR64:$imm)>;
3825 def : Pat<(v1i64 (int_arm64_neon_vcvtfp2fxu (v1f64 FPR64:$Rn),
3827 (FCVTZUd FPR64:$Rn, vecshiftR64:$imm)>;
3828 def : Pat<(int_arm64_neon_vcvtfxs2fp FPR32:$Rn, vecshiftR32:$imm),
3829 (SCVTFs FPR32:$Rn, vecshiftR32:$imm)>;
3830 def : Pat<(int_arm64_neon_vcvtfxu2fp FPR32:$Rn, vecshiftR32:$imm),
3831 (UCVTFs FPR32:$Rn, vecshiftR32:$imm)>;
3832 def : Pat<(f64 (int_arm64_neon_vcvtfxs2fp (i64 FPR64:$Rn), vecshiftR64:$imm)),
3833 (SCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
3834 def : Pat<(f64 (int_arm64_neon_vcvtfxu2fp (i64 FPR64:$Rn), vecshiftR64:$imm)),
3835 (UCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
3836 def : Pat<(v1f64 (int_arm64_neon_vcvtfxs2fp (v1i64 FPR64:$Rn),
3838 (SCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
3839 def : Pat<(v1f64 (int_arm64_neon_vcvtfxu2fp (v1i64 FPR64:$Rn),
3841 (UCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
3843 defm SHL : SIMDScalarLShiftD< 0, 0b01010, "shl", ARM64vshl>;
3844 defm SLI : SIMDScalarLShiftDTied<1, 0b01010, "sli">;
3845 defm SQRSHRN : SIMDScalarRShiftBHS< 0, 0b10011, "sqrshrn",
3846 int_arm64_neon_sqrshrn>;
3847 defm SQRSHRUN : SIMDScalarRShiftBHS< 1, 0b10001, "sqrshrun",
3848 int_arm64_neon_sqrshrun>;
3849 defm SQSHLU : SIMDScalarLShiftBHSD<1, 0b01100, "sqshlu", ARM64sqshlui>;
3850 defm SQSHL : SIMDScalarLShiftBHSD<0, 0b01110, "sqshl", ARM64sqshli>;
3851 defm SQSHRN : SIMDScalarRShiftBHS< 0, 0b10010, "sqshrn",
3852 int_arm64_neon_sqshrn>;
3853 defm SQSHRUN : SIMDScalarRShiftBHS< 1, 0b10000, "sqshrun",
3854 int_arm64_neon_sqshrun>;
3855 defm SRI : SIMDScalarRShiftDTied< 1, 0b01000, "sri">;
3856 defm SRSHR : SIMDScalarRShiftD< 0, 0b00100, "srshr", ARM64srshri>;
3857 defm SRSRA : SIMDScalarRShiftDTied< 0, 0b00110, "srsra",
3858 TriOpFrag<(add node:$LHS,
3859 (ARM64srshri node:$MHS, node:$RHS))>>;
3860 defm SSHR : SIMDScalarRShiftD< 0, 0b00000, "sshr", ARM64vashr>;
3861 defm SSRA : SIMDScalarRShiftDTied< 0, 0b00010, "ssra",
3862 TriOpFrag<(add node:$LHS,
3863 (ARM64vashr node:$MHS, node:$RHS))>>;
3864 defm UQRSHRN : SIMDScalarRShiftBHS< 1, 0b10011, "uqrshrn",
3865 int_arm64_neon_uqrshrn>;
3866 defm UQSHL : SIMDScalarLShiftBHSD<1, 0b01110, "uqshl", ARM64uqshli>;
3867 defm UQSHRN : SIMDScalarRShiftBHS< 1, 0b10010, "uqshrn",
3868 int_arm64_neon_uqshrn>;
3869 defm URSHR : SIMDScalarRShiftD< 1, 0b00100, "urshr", ARM64urshri>;
3870 defm URSRA : SIMDScalarRShiftDTied< 1, 0b00110, "ursra",
3871 TriOpFrag<(add node:$LHS,
3872 (ARM64urshri node:$MHS, node:$RHS))>>;
3873 defm USHR : SIMDScalarRShiftD< 1, 0b00000, "ushr", ARM64vlshr>;
3874 defm USRA : SIMDScalarRShiftDTied< 1, 0b00010, "usra",
3875 TriOpFrag<(add node:$LHS,
3876 (ARM64vlshr node:$MHS, node:$RHS))>>;
3878 //----------------------------------------------------------------------------
3879 // AdvSIMD vector shift instructions
3880 //----------------------------------------------------------------------------
3881 defm FCVTZS:SIMDVectorRShiftSD<0, 0b11111, "fcvtzs", int_arm64_neon_vcvtfp2fxs>;
3882 defm FCVTZU:SIMDVectorRShiftSD<1, 0b11111, "fcvtzu", int_arm64_neon_vcvtfp2fxu>;
3883 defm SCVTF: SIMDVectorRShiftSDToFP<0, 0b11100, "scvtf",
3884 int_arm64_neon_vcvtfxs2fp>;
3885 defm RSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10001, "rshrn",
3886 int_arm64_neon_rshrn>;
3887 defm SHL : SIMDVectorLShiftBHSD<0, 0b01010, "shl", ARM64vshl>;
3888 defm SHRN : SIMDVectorRShiftNarrowBHS<0, 0b10000, "shrn",
3889 BinOpFrag<(trunc (ARM64vashr node:$LHS, node:$RHS))>>;
3890 defm SLI : SIMDVectorLShiftBHSDTied<1, 0b01010, "sli", int_arm64_neon_vsli>;
3891 def : Pat<(v1i64 (int_arm64_neon_vsli (v1i64 FPR64:$Rd), (v1i64 FPR64:$Rn),
3892 (i32 vecshiftL64:$imm))),
3893 (SLId FPR64:$Rd, FPR64:$Rn, vecshiftL64:$imm)>;
3894 defm SQRSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10011, "sqrshrn",
3895 int_arm64_neon_sqrshrn>;
3896 defm SQRSHRUN: SIMDVectorRShiftNarrowBHS<1, 0b10001, "sqrshrun",
3897 int_arm64_neon_sqrshrun>;
3898 defm SQSHLU : SIMDVectorLShiftBHSD<1, 0b01100, "sqshlu", ARM64sqshlui>;
3899 defm SQSHL : SIMDVectorLShiftBHSD<0, 0b01110, "sqshl", ARM64sqshli>;
3900 defm SQSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10010, "sqshrn",
3901 int_arm64_neon_sqshrn>;
3902 defm SQSHRUN : SIMDVectorRShiftNarrowBHS<1, 0b10000, "sqshrun",
3903 int_arm64_neon_sqshrun>;
3904 defm SRI : SIMDVectorRShiftBHSDTied<1, 0b01000, "sri", int_arm64_neon_vsri>;
3905 def : Pat<(v1i64 (int_arm64_neon_vsri (v1i64 FPR64:$Rd), (v1i64 FPR64:$Rn),
3906 (i32 vecshiftR64:$imm))),
3907 (SRId FPR64:$Rd, FPR64:$Rn, vecshiftR64:$imm)>;
3908 defm SRSHR : SIMDVectorRShiftBHSD<0, 0b00100, "srshr", ARM64srshri>;
3909 defm SRSRA : SIMDVectorRShiftBHSDTied<0, 0b00110, "srsra",
3910 TriOpFrag<(add node:$LHS,
3911 (ARM64srshri node:$MHS, node:$RHS))> >;
3912 defm SSHLL : SIMDVectorLShiftLongBHSD<0, 0b10100, "sshll",
3913 BinOpFrag<(ARM64vshl (sext node:$LHS), node:$RHS)>>;
3915 defm SSHR : SIMDVectorRShiftBHSD<0, 0b00000, "sshr", ARM64vashr>;
3916 defm SSRA : SIMDVectorRShiftBHSDTied<0, 0b00010, "ssra",
3917 TriOpFrag<(add node:$LHS, (ARM64vashr node:$MHS, node:$RHS))>>;
3918 defm UCVTF : SIMDVectorRShiftSDToFP<1, 0b11100, "ucvtf",
3919 int_arm64_neon_vcvtfxu2fp>;
3920 defm UQRSHRN : SIMDVectorRShiftNarrowBHS<1, 0b10011, "uqrshrn",
3921 int_arm64_neon_uqrshrn>;
3922 defm UQSHL : SIMDVectorLShiftBHSD<1, 0b01110, "uqshl", ARM64uqshli>;
3923 defm UQSHRN : SIMDVectorRShiftNarrowBHS<1, 0b10010, "uqshrn",
3924 int_arm64_neon_uqshrn>;
3925 defm URSHR : SIMDVectorRShiftBHSD<1, 0b00100, "urshr", ARM64urshri>;
3926 defm URSRA : SIMDVectorRShiftBHSDTied<1, 0b00110, "ursra",
3927 TriOpFrag<(add node:$LHS,
3928 (ARM64urshri node:$MHS, node:$RHS))> >;
3929 defm USHLL : SIMDVectorLShiftLongBHSD<1, 0b10100, "ushll",
3930 BinOpFrag<(ARM64vshl (zext node:$LHS), node:$RHS)>>;
3931 defm USHR : SIMDVectorRShiftBHSD<1, 0b00000, "ushr", ARM64vlshr>;
3932 defm USRA : SIMDVectorRShiftBHSDTied<1, 0b00010, "usra",
3933 TriOpFrag<(add node:$LHS, (ARM64vlshr node:$MHS, node:$RHS))> >;
3935 // SHRN patterns for when a logical right shift was used instead of arithmetic
3936 // (the immediate guarantees no sign bits actually end up in the result so it
3938 def : Pat<(v8i8 (trunc (ARM64vlshr (v8i16 V128:$Rn), vecshiftR16Narrow:$imm))),
3939 (SHRNv8i8_shift V128:$Rn, vecshiftR16Narrow:$imm)>;
3940 def : Pat<(v4i16 (trunc (ARM64vlshr (v4i32 V128:$Rn), vecshiftR32Narrow:$imm))),
3941 (SHRNv4i16_shift V128:$Rn, vecshiftR32Narrow:$imm)>;
3942 def : Pat<(v2i32 (trunc (ARM64vlshr (v2i64 V128:$Rn), vecshiftR64Narrow:$imm))),
3943 (SHRNv2i32_shift V128:$Rn, vecshiftR64Narrow:$imm)>;
3945 def : Pat<(v16i8 (concat_vectors (v8i8 V64:$Rd),
3946 (trunc (ARM64vlshr (v8i16 V128:$Rn),
3947 vecshiftR16Narrow:$imm)))),
3948 (SHRNv16i8_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
3949 V128:$Rn, vecshiftR16Narrow:$imm)>;
3950 def : Pat<(v8i16 (concat_vectors (v4i16 V64:$Rd),
3951 (trunc (ARM64vlshr (v4i32 V128:$Rn),
3952 vecshiftR32Narrow:$imm)))),
3953 (SHRNv8i16_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
3954 V128:$Rn, vecshiftR32Narrow:$imm)>;
3955 def : Pat<(v4i32 (concat_vectors (v2i32 V64:$Rd),
3956 (trunc (ARM64vlshr (v2i64 V128:$Rn),
3957 vecshiftR64Narrow:$imm)))),
3958 (SHRNv4i32_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
3959 V128:$Rn, vecshiftR32Narrow:$imm)>;
3961 // Vector sign and zero extensions are implemented with SSHLL and USSHLL.
3962 // Anyexts are implemented as zexts.
3963 def : Pat<(v8i16 (sext (v8i8 V64:$Rn))), (SSHLLv8i8_shift V64:$Rn, (i32 0))>;
3964 def : Pat<(v8i16 (zext (v8i8 V64:$Rn))), (USHLLv8i8_shift V64:$Rn, (i32 0))>;
3965 def : Pat<(v8i16 (anyext (v8i8 V64:$Rn))), (USHLLv8i8_shift V64:$Rn, (i32 0))>;
3966 def : Pat<(v4i32 (sext (v4i16 V64:$Rn))), (SSHLLv4i16_shift V64:$Rn, (i32 0))>;
3967 def : Pat<(v4i32 (zext (v4i16 V64:$Rn))), (USHLLv4i16_shift V64:$Rn, (i32 0))>;
3968 def : Pat<(v4i32 (anyext (v4i16 V64:$Rn))), (USHLLv4i16_shift V64:$Rn, (i32 0))>;
3969 def : Pat<(v2i64 (sext (v2i32 V64:$Rn))), (SSHLLv2i32_shift V64:$Rn, (i32 0))>;
3970 def : Pat<(v2i64 (zext (v2i32 V64:$Rn))), (USHLLv2i32_shift V64:$Rn, (i32 0))>;
3971 def : Pat<(v2i64 (anyext (v2i32 V64:$Rn))), (USHLLv2i32_shift V64:$Rn, (i32 0))>;
3972 // Also match an extend from the upper half of a 128 bit source register.
3973 def : Pat<(v8i16 (anyext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
3974 (USHLLv16i8_shift V128:$Rn, (i32 0))>;
3975 def : Pat<(v8i16 (zext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
3976 (USHLLv16i8_shift V128:$Rn, (i32 0))>;
3977 def : Pat<(v8i16 (sext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
3978 (SSHLLv16i8_shift V128:$Rn, (i32 0))>;
3979 def : Pat<(v4i32 (anyext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
3980 (USHLLv8i16_shift V128:$Rn, (i32 0))>;
3981 def : Pat<(v4i32 (zext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
3982 (USHLLv8i16_shift V128:$Rn, (i32 0))>;
3983 def : Pat<(v4i32 (sext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
3984 (SSHLLv8i16_shift V128:$Rn, (i32 0))>;
3985 def : Pat<(v2i64 (anyext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
3986 (USHLLv4i32_shift V128:$Rn, (i32 0))>;
3987 def : Pat<(v2i64 (zext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
3988 (USHLLv4i32_shift V128:$Rn, (i32 0))>;
3989 def : Pat<(v2i64 (sext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
3990 (SSHLLv4i32_shift V128:$Rn, (i32 0))>;
3992 // Vector shift sxtl aliases
3993 def : InstAlias<"sxtl.8h $dst, $src1",
3994 (SSHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
3995 def : InstAlias<"sxtl $dst.8h, $src1.8b",
3996 (SSHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
3997 def : InstAlias<"sxtl.4s $dst, $src1",
3998 (SSHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
3999 def : InstAlias<"sxtl $dst.4s, $src1.4h",
4000 (SSHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
4001 def : InstAlias<"sxtl.2d $dst, $src1",
4002 (SSHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
4003 def : InstAlias<"sxtl $dst.2d, $src1.2s",
4004 (SSHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
4006 // Vector shift sxtl2 aliases
4007 def : InstAlias<"sxtl2.8h $dst, $src1",
4008 (SSHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
4009 def : InstAlias<"sxtl2 $dst.8h, $src1.16b",
4010 (SSHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
4011 def : InstAlias<"sxtl2.4s $dst, $src1",
4012 (SSHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
4013 def : InstAlias<"sxtl2 $dst.4s, $src1.8h",
4014 (SSHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
4015 def : InstAlias<"sxtl2.2d $dst, $src1",
4016 (SSHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
4017 def : InstAlias<"sxtl2 $dst.2d, $src1.4s",
4018 (SSHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
4020 // Vector shift uxtl aliases
4021 def : InstAlias<"uxtl.8h $dst, $src1",
4022 (USHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
4023 def : InstAlias<"uxtl $dst.8h, $src1.8b",
4024 (USHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
4025 def : InstAlias<"uxtl.4s $dst, $src1",
4026 (USHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
4027 def : InstAlias<"uxtl $dst.4s, $src1.4h",
4028 (USHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
4029 def : InstAlias<"uxtl.2d $dst, $src1",
4030 (USHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
4031 def : InstAlias<"uxtl $dst.2d, $src1.2s",
4032 (USHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
4034 // Vector shift uxtl2 aliases
4035 def : InstAlias<"uxtl2.8h $dst, $src1",
4036 (USHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
4037 def : InstAlias<"uxtl2 $dst.8h, $src1.16b",
4038 (USHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
4039 def : InstAlias<"uxtl2.4s $dst, $src1",
4040 (USHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
4041 def : InstAlias<"uxtl2 $dst.4s, $src1.8h",
4042 (USHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
4043 def : InstAlias<"uxtl2.2d $dst, $src1",
4044 (USHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
4045 def : InstAlias<"uxtl2 $dst.2d, $src1.4s",
4046 (USHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
4048 // If an integer is about to be converted to a floating point value,
4049 // just load it on the floating point unit.
4050 // These patterns are more complex because floating point loads do not
4051 // support sign extension.
4052 // The sign extension has to be explicitly added and is only supported for
4053 // one step: byte-to-half, half-to-word, word-to-doubleword.
4054 // SCVTF GPR -> FPR is 9 cycles.
4055 // SCVTF FPR -> FPR is 4 cyclces.
4056 // (sign extension with lengthen) SXTL FPR -> FPR is 2 cycles.
4057 // Therefore, we can do 2 sign extensions and one SCVTF FPR -> FPR
4058 // and still being faster.
4059 // However, this is not good for code size.
4060 // 8-bits -> float. 2 sizes step-up.
4061 def : Pat <(f32 (sint_to_fp (i32 (sextloadi8 ro_indexed8:$addr)))),
4062 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
4067 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4068 (LDRBro ro_indexed8:$addr),
4073 ssub)))>, Requires<[NotForCodeSize]>;
4074 def : Pat <(f32 (sint_to_fp (i32 (sextloadi8 am_indexed8:$addr)))),
4075 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
4080 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4081 (LDRBui am_indexed8:$addr),
4086 ssub)))>, Requires<[NotForCodeSize]>;
4087 def : Pat <(f32 (sint_to_fp (i32 (sextloadi8 am_unscaled8:$addr)))),
4088 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
4093 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4094 (LDURBi am_unscaled8:$addr),
4099 ssub)))>, Requires<[NotForCodeSize]>;
4100 // 16-bits -> float. 1 size step-up.
4101 def : Pat <(f32 (sint_to_fp (i32 (sextloadi16 ro_indexed16:$addr)))),
4102 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
4104 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4105 (LDRHro ro_indexed16:$addr),
4108 ssub)))>, Requires<[NotForCodeSize]>;
4109 def : Pat <(f32 (sint_to_fp (i32 (sextloadi16 am_indexed16:$addr)))),
4110 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
4112 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4113 (LDRHui am_indexed16:$addr),
4116 ssub)))>, Requires<[NotForCodeSize]>;
4117 def : Pat <(f32 (sint_to_fp (i32 (sextloadi16 am_unscaled16:$addr)))),
4118 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
4120 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4121 (LDURHi am_unscaled16:$addr),
4124 ssub)))>, Requires<[NotForCodeSize]>;
4125 // 32-bits to 32-bits are handled in target specific dag combine:
4126 // performIntToFpCombine.
4127 // 64-bits integer to 32-bits floating point, not possible with
4128 // SCVTF on floating point registers (both source and destination
4129 // must have the same size).
4131 // Here are the patterns for 8, 16, 32, and 64-bits to double.
4132 // 8-bits -> double. 3 size step-up: give up.
4133 // 16-bits -> double. 2 size step.
4134 def : Pat <(f64 (sint_to_fp (i32 (sextloadi16 ro_indexed16:$addr)))),
4135 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4140 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4141 (LDRHro ro_indexed16:$addr),
4146 dsub)))>, Requires<[NotForCodeSize]>;
4147 def : Pat <(f64 (sint_to_fp (i32 (sextloadi16 am_indexed16:$addr)))),
4148 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4153 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4154 (LDRHui am_indexed16:$addr),
4159 dsub)))>, Requires<[NotForCodeSize]>;
4160 def : Pat <(f64 (sint_to_fp (i32 (sextloadi16 am_unscaled16:$addr)))),
4161 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4166 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4167 (LDURHi am_unscaled16:$addr),
4172 dsub)))>, Requires<[NotForCodeSize]>;
4173 // 32-bits -> double. 1 size step-up.
4174 def : Pat <(f64 (sint_to_fp (i32 (load ro_indexed32:$addr)))),
4175 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4177 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4178 (LDRSro ro_indexed32:$addr),
4181 dsub)))>, Requires<[NotForCodeSize]>;
4182 def : Pat <(f64 (sint_to_fp (i32 (load am_indexed32:$addr)))),
4183 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4185 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4186 (LDRSui am_indexed32:$addr),
4189 dsub)))>, Requires<[NotForCodeSize]>;
4190 def : Pat <(f64 (sint_to_fp (i32 (load am_unscaled32:$addr)))),
4191 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4193 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4194 (LDURSi am_unscaled32:$addr),
4197 dsub)))>, Requires<[NotForCodeSize]>;
4198 // 64-bits -> double are handled in target specific dag combine:
4199 // performIntToFpCombine.
4202 //----------------------------------------------------------------------------
4203 // AdvSIMD Load-Store Structure
4204 //----------------------------------------------------------------------------
4205 defm LD1 : SIMDLd1Multiple<"ld1">;
4206 defm LD2 : SIMDLd2Multiple<"ld2">;
4207 defm LD3 : SIMDLd3Multiple<"ld3">;
4208 defm LD4 : SIMDLd4Multiple<"ld4">;
4210 defm ST1 : SIMDSt1Multiple<"st1">;
4211 defm ST2 : SIMDSt2Multiple<"st2">;
4212 defm ST3 : SIMDSt3Multiple<"st3">;
4213 defm ST4 : SIMDSt4Multiple<"st4">;
4215 class Ld1Pat<ValueType ty, Instruction INST>
4216 : Pat<(ty (load am_simdnoindex:$vaddr)), (INST am_simdnoindex:$vaddr)>;
4218 def : Ld1Pat<v16i8, LD1Onev16b>;
4219 def : Ld1Pat<v8i16, LD1Onev8h>;
4220 def : Ld1Pat<v4i32, LD1Onev4s>;
4221 def : Ld1Pat<v2i64, LD1Onev2d>;
4222 def : Ld1Pat<v8i8, LD1Onev8b>;
4223 def : Ld1Pat<v4i16, LD1Onev4h>;
4224 def : Ld1Pat<v2i32, LD1Onev2s>;
4225 def : Ld1Pat<v1i64, LD1Onev1d>;
4227 class St1Pat<ValueType ty, Instruction INST>
4228 : Pat<(store ty:$Vt, am_simdnoindex:$vaddr),
4229 (INST ty:$Vt, am_simdnoindex:$vaddr)>;
4231 def : St1Pat<v16i8, ST1Onev16b>;
4232 def : St1Pat<v8i16, ST1Onev8h>;
4233 def : St1Pat<v4i32, ST1Onev4s>;
4234 def : St1Pat<v2i64, ST1Onev2d>;
4235 def : St1Pat<v8i8, ST1Onev8b>;
4236 def : St1Pat<v4i16, ST1Onev4h>;
4237 def : St1Pat<v2i32, ST1Onev2s>;
4238 def : St1Pat<v1i64, ST1Onev1d>;
4244 defm LD1R : SIMDLdR<0, 0b110, 0, "ld1r", "One", 1, 2, 4, 8>;
4245 defm LD2R : SIMDLdR<1, 0b110, 0, "ld2r", "Two", 2, 4, 8, 16>;
4246 defm LD3R : SIMDLdR<0, 0b111, 0, "ld3r", "Three", 3, 6, 12, 24>;
4247 defm LD4R : SIMDLdR<1, 0b111, 0, "ld4r", "Four", 4, 8, 16, 32>;
4248 let mayLoad = 1, neverHasSideEffects = 1 in {
4249 defm LD1 : SIMDLdSingleBTied<0, 0b000, "ld1", VecListOneb, GPR64pi1>;
4250 defm LD1 : SIMDLdSingleHTied<0, 0b010, 0, "ld1", VecListOneh, GPR64pi2>;
4251 defm LD1 : SIMDLdSingleSTied<0, 0b100, 0b00, "ld1", VecListOnes, GPR64pi4>;
4252 defm LD1 : SIMDLdSingleDTied<0, 0b100, 0b01, "ld1", VecListOned, GPR64pi8>;
4253 defm LD2 : SIMDLdSingleBTied<1, 0b000, "ld2", VecListTwob, GPR64pi2>;
4254 defm LD2 : SIMDLdSingleHTied<1, 0b010, 0, "ld2", VecListTwoh, GPR64pi4>;
4255 defm LD2 : SIMDLdSingleSTied<1, 0b100, 0b00, "ld2", VecListTwos, GPR64pi8>;
4256 defm LD2 : SIMDLdSingleDTied<1, 0b100, 0b01, "ld2", VecListTwod, GPR64pi16>;
4257 defm LD3 : SIMDLdSingleBTied<0, 0b001, "ld3", VecListThreeb, GPR64pi3>;
4258 defm LD3 : SIMDLdSingleHTied<0, 0b011, 0, "ld3", VecListThreeh, GPR64pi6>;
4259 defm LD3 : SIMDLdSingleSTied<0, 0b101, 0b00, "ld3", VecListThrees, GPR64pi12>;
4260 defm LD3 : SIMDLdSingleDTied<0, 0b101, 0b01, "ld3", VecListThreed, GPR64pi24>;
4261 defm LD4 : SIMDLdSingleBTied<1, 0b001, "ld4", VecListFourb, GPR64pi4>;
4262 defm LD4 : SIMDLdSingleHTied<1, 0b011, 0, "ld4", VecListFourh, GPR64pi8>;
4263 defm LD4 : SIMDLdSingleSTied<1, 0b101, 0b00, "ld4", VecListFours, GPR64pi16>;
4264 defm LD4 : SIMDLdSingleDTied<1, 0b101, 0b01, "ld4", VecListFourd, GPR64pi32>;
4267 def : Pat<(v8i8 (ARM64dup (i32 (extloadi8 am_simdnoindex:$vaddr)))),
4268 (LD1Rv8b am_simdnoindex:$vaddr)>;
4269 def : Pat<(v16i8 (ARM64dup (i32 (extloadi8 am_simdnoindex:$vaddr)))),
4270 (LD1Rv16b am_simdnoindex:$vaddr)>;
4271 def : Pat<(v4i16 (ARM64dup (i32 (extloadi16 am_simdnoindex:$vaddr)))),
4272 (LD1Rv4h am_simdnoindex:$vaddr)>;
4273 def : Pat<(v8i16 (ARM64dup (i32 (extloadi16 am_simdnoindex:$vaddr)))),
4274 (LD1Rv8h am_simdnoindex:$vaddr)>;
4275 def : Pat<(v2i32 (ARM64dup (i32 (load am_simdnoindex:$vaddr)))),
4276 (LD1Rv2s am_simdnoindex:$vaddr)>;
4277 def : Pat<(v4i32 (ARM64dup (i32 (load am_simdnoindex:$vaddr)))),
4278 (LD1Rv4s am_simdnoindex:$vaddr)>;
4279 def : Pat<(v2i64 (ARM64dup (i64 (load am_simdnoindex:$vaddr)))),
4280 (LD1Rv2d am_simdnoindex:$vaddr)>;
4281 def : Pat<(v1i64 (ARM64dup (i64 (load am_simdnoindex:$vaddr)))),
4282 (LD1Rv1d am_simdnoindex:$vaddr)>;
4283 // Grab the floating point version too
4284 def : Pat<(v2f32 (ARM64dup (f32 (load am_simdnoindex:$vaddr)))),
4285 (LD1Rv2s am_simdnoindex:$vaddr)>;
4286 def : Pat<(v4f32 (ARM64dup (f32 (load am_simdnoindex:$vaddr)))),
4287 (LD1Rv4s am_simdnoindex:$vaddr)>;
4288 def : Pat<(v2f64 (ARM64dup (f64 (load am_simdnoindex:$vaddr)))),
4289 (LD1Rv2d am_simdnoindex:$vaddr)>;
4290 def : Pat<(v1f64 (ARM64dup (f64 (load am_simdnoindex:$vaddr)))),
4291 (LD1Rv1d am_simdnoindex:$vaddr)>;
4293 class Ld1Lane128Pat<SDPatternOperator scalar_load, Operand VecIndex,
4294 ValueType VTy, ValueType STy, Instruction LD1>
4295 : Pat<(vector_insert (VTy VecListOne128:$Rd),
4296 (STy (scalar_load am_simdnoindex:$vaddr)), VecIndex:$idx),
4297 (LD1 VecListOne128:$Rd, VecIndex:$idx, am_simdnoindex:$vaddr)>;
4299 def : Ld1Lane128Pat<extloadi8, VectorIndexB, v16i8, i32, LD1i8>;
4300 def : Ld1Lane128Pat<extloadi16, VectorIndexH, v8i16, i32, LD1i16>;
4301 def : Ld1Lane128Pat<load, VectorIndexS, v4i32, i32, LD1i32>;
4302 def : Ld1Lane128Pat<load, VectorIndexS, v4f32, f32, LD1i32>;
4303 def : Ld1Lane128Pat<load, VectorIndexD, v2i64, i64, LD1i64>;
4304 def : Ld1Lane128Pat<load, VectorIndexD, v2f64, f64, LD1i64>;
4306 class Ld1Lane64Pat<SDPatternOperator scalar_load, Operand VecIndex,
4307 ValueType VTy, ValueType STy, Instruction LD1>
4308 : Pat<(vector_insert (VTy VecListOne64:$Rd),
4309 (STy (scalar_load am_simdnoindex:$vaddr)), VecIndex:$idx),
4311 (LD1 (SUBREG_TO_REG (i32 0), VecListOne64:$Rd, dsub),
4312 VecIndex:$idx, am_simdnoindex:$vaddr),
4315 def : Ld1Lane64Pat<extloadi8, VectorIndexB, v8i8, i32, LD1i8>;
4316 def : Ld1Lane64Pat<extloadi16, VectorIndexH, v4i16, i32, LD1i16>;
4317 def : Ld1Lane64Pat<load, VectorIndexS, v2i32, i32, LD1i32>;
4318 def : Ld1Lane64Pat<load, VectorIndexS, v2f32, f32, LD1i32>;
4321 defm LD1 : SIMDLdSt1SingleAliases<"ld1">;
4322 defm LD2 : SIMDLdSt2SingleAliases<"ld2">;
4323 defm LD3 : SIMDLdSt3SingleAliases<"ld3">;
4324 defm LD4 : SIMDLdSt4SingleAliases<"ld4">;
4327 defm ST1 : SIMDStSingleB<0, 0b000, "st1", VecListOneb, GPR64pi1>;
4328 defm ST1 : SIMDStSingleH<0, 0b010, 0, "st1", VecListOneh, GPR64pi2>;
4329 defm ST1 : SIMDStSingleS<0, 0b100, 0b00, "st1", VecListOnes, GPR64pi4>;
4330 defm ST1 : SIMDStSingleD<0, 0b100, 0b01, "st1", VecListOned, GPR64pi8>;
4332 let AddedComplexity = 8 in
4333 class St1Lane128Pat<SDPatternOperator scalar_store, Operand VecIndex,
4334 ValueType VTy, ValueType STy, Instruction ST1>
4336 (STy (vector_extract (VTy VecListOne128:$Vt), VecIndex:$idx)),
4337 am_simdnoindex:$vaddr),
4338 (ST1 VecListOne128:$Vt, VecIndex:$idx, am_simdnoindex:$vaddr)>;
4340 def : St1Lane128Pat<truncstorei8, VectorIndexB, v16i8, i32, ST1i8>;
4341 def : St1Lane128Pat<truncstorei16, VectorIndexH, v8i16, i32, ST1i16>;
4342 def : St1Lane128Pat<store, VectorIndexS, v4i32, i32, ST1i32>;
4343 def : St1Lane128Pat<store, VectorIndexS, v4f32, f32, ST1i32>;
4344 def : St1Lane128Pat<store, VectorIndexD, v2i64, i64, ST1i64>;
4345 def : St1Lane128Pat<store, VectorIndexD, v2f64, f64, ST1i64>;
4347 let AddedComplexity = 8 in
4348 class St1Lane64Pat<SDPatternOperator scalar_store, Operand VecIndex,
4349 ValueType VTy, ValueType STy, Instruction ST1>
4351 (STy (vector_extract (VTy VecListOne64:$Vt), VecIndex:$idx)),
4352 am_simdnoindex:$vaddr),
4353 (ST1 (SUBREG_TO_REG (i32 0), VecListOne64:$Vt, dsub),
4354 VecIndex:$idx, am_simdnoindex:$vaddr)>;
4356 def : St1Lane64Pat<truncstorei8, VectorIndexB, v8i8, i32, ST1i8>;
4357 def : St1Lane64Pat<truncstorei16, VectorIndexH, v4i16, i32, ST1i16>;
4358 def : St1Lane64Pat<store, VectorIndexS, v2i32, i32, ST1i32>;
4359 def : St1Lane64Pat<store, VectorIndexS, v2f32, f32, ST1i32>;
4361 let mayStore = 1, neverHasSideEffects = 1 in {
4362 defm ST2 : SIMDStSingleB<1, 0b000, "st2", VecListTwob, GPR64pi2>;
4363 defm ST2 : SIMDStSingleH<1, 0b010, 0, "st2", VecListTwoh, GPR64pi4>;
4364 defm ST2 : SIMDStSingleS<1, 0b100, 0b00, "st2", VecListTwos, GPR64pi8>;
4365 defm ST2 : SIMDStSingleD<1, 0b100, 0b01, "st2", VecListTwod, GPR64pi16>;
4366 defm ST3 : SIMDStSingleB<0, 0b001, "st3", VecListThreeb, GPR64pi3>;
4367 defm ST3 : SIMDStSingleH<0, 0b011, 0, "st3", VecListThreeh, GPR64pi6>;
4368 defm ST3 : SIMDStSingleS<0, 0b101, 0b00, "st3", VecListThrees, GPR64pi12>;
4369 defm ST3 : SIMDStSingleD<0, 0b101, 0b01, "st3", VecListThreed, GPR64pi24>;
4370 defm ST4 : SIMDStSingleB<1, 0b001, "st4", VecListFourb, GPR64pi4>;
4371 defm ST4 : SIMDStSingleH<1, 0b011, 0, "st4", VecListFourh, GPR64pi8>;
4372 defm ST4 : SIMDStSingleS<1, 0b101, 0b00, "st4", VecListFours, GPR64pi16>;
4373 defm ST4 : SIMDStSingleD<1, 0b101, 0b01, "st4", VecListFourd, GPR64pi32>;
4376 defm ST1 : SIMDLdSt1SingleAliases<"st1">;
4377 defm ST2 : SIMDLdSt2SingleAliases<"st2">;
4378 defm ST3 : SIMDLdSt3SingleAliases<"st3">;
4379 defm ST4 : SIMDLdSt4SingleAliases<"st4">;
4381 //----------------------------------------------------------------------------
4382 // Crypto extensions
4383 //----------------------------------------------------------------------------
4385 def AESErr : AESTiedInst<0b0100, "aese", int_arm64_crypto_aese>;
4386 def AESDrr : AESTiedInst<0b0101, "aesd", int_arm64_crypto_aesd>;
4387 def AESMCrr : AESInst< 0b0110, "aesmc", int_arm64_crypto_aesmc>;
4388 def AESIMCrr : AESInst< 0b0111, "aesimc", int_arm64_crypto_aesimc>;
4390 def SHA1Crrr : SHATiedInstQSV<0b000, "sha1c", int_arm64_crypto_sha1c>;
4391 def SHA1Prrr : SHATiedInstQSV<0b001, "sha1p", int_arm64_crypto_sha1p>;
4392 def SHA1Mrrr : SHATiedInstQSV<0b010, "sha1m", int_arm64_crypto_sha1m>;
4393 def SHA1SU0rrr : SHATiedInstVVV<0b011, "sha1su0", int_arm64_crypto_sha1su0>;
4394 def SHA256Hrrr : SHATiedInstQQV<0b100, "sha256h", int_arm64_crypto_sha256h>;
4395 def SHA256H2rrr : SHATiedInstQQV<0b101, "sha256h2",int_arm64_crypto_sha256h2>;
4396 def SHA256SU1rrr :SHATiedInstVVV<0b110, "sha256su1",int_arm64_crypto_sha256su1>;
4398 def SHA1Hrr : SHAInstSS< 0b0000, "sha1h", int_arm64_crypto_sha1h>;
4399 def SHA1SU1rr : SHATiedInstVV<0b0001, "sha1su1", int_arm64_crypto_sha1su1>;
4400 def SHA256SU0rr : SHATiedInstVV<0b0010, "sha256su0",int_arm64_crypto_sha256su0>;
4402 //----------------------------------------------------------------------------
4404 //----------------------------------------------------------------------------
4405 // FIXME: Like for X86, these should go in their own separate .td file.
4407 // Any instruction that defines a 32-bit result leaves the high half of the
4408 // register. Truncate can be lowered to EXTRACT_SUBREG. CopyFromReg may
4409 // be copying from a truncate. But any other 32-bit operation will zero-extend
4411 // FIXME: X86 also checks for CMOV here. Do we need something similar?
4412 def def32 : PatLeaf<(i32 GPR32:$src), [{
4413 return N->getOpcode() != ISD::TRUNCATE &&
4414 N->getOpcode() != TargetOpcode::EXTRACT_SUBREG &&
4415 N->getOpcode() != ISD::CopyFromReg;
4418 // In the case of a 32-bit def that is known to implicitly zero-extend,
4419 // we can use a SUBREG_TO_REG.
4420 def : Pat<(i64 (zext def32:$src)), (SUBREG_TO_REG (i64 0), GPR32:$src, sub_32)>;
4422 // For an anyext, we don't care what the high bits are, so we can perform an
4423 // INSERT_SUBREF into an IMPLICIT_DEF.
4424 def : Pat<(i64 (anyext GPR32:$src)),
4425 (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32)>;
4427 // When we need to explicitly zero-extend, we use an unsigned bitfield move
4428 // instruction (UBFM) on the enclosing super-reg.
4429 def : Pat<(i64 (zext GPR32:$src)),
4430 (UBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32), 0, 31)>;
4432 // To sign extend, we use a signed bitfield move instruction (SBFM) on the
4433 // containing super-reg.
4434 def : Pat<(i64 (sext GPR32:$src)),
4435 (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32), 0, 31)>;
4436 def : Pat<(i64 (sext_inreg GPR64:$src, i32)), (SBFMXri GPR64:$src, 0, 31)>;
4437 def : Pat<(i64 (sext_inreg GPR64:$src, i16)), (SBFMXri GPR64:$src, 0, 15)>;
4438 def : Pat<(i64 (sext_inreg GPR64:$src, i8)), (SBFMXri GPR64:$src, 0, 7)>;
4439 def : Pat<(i64 (sext_inreg GPR64:$src, i1)), (SBFMXri GPR64:$src, 0, 0)>;
4440 def : Pat<(i32 (sext_inreg GPR32:$src, i16)), (SBFMWri GPR32:$src, 0, 15)>;
4441 def : Pat<(i32 (sext_inreg GPR32:$src, i8)), (SBFMWri GPR32:$src, 0, 7)>;
4442 def : Pat<(i32 (sext_inreg GPR32:$src, i1)), (SBFMWri GPR32:$src, 0, 0)>;
4444 def : Pat<(shl (sext_inreg GPR32:$Rn, i8), (i64 imm0_31:$imm)),
4445 (SBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
4446 (i64 (i32shift_sext_i8 imm0_31:$imm)))>;
4447 def : Pat<(shl (sext_inreg GPR64:$Rn, i8), (i64 imm0_63:$imm)),
4448 (SBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
4449 (i64 (i64shift_sext_i8 imm0_63:$imm)))>;
4451 def : Pat<(shl (sext_inreg GPR32:$Rn, i16), (i64 imm0_31:$imm)),
4452 (SBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
4453 (i64 (i32shift_sext_i16 imm0_31:$imm)))>;
4454 def : Pat<(shl (sext_inreg GPR64:$Rn, i16), (i64 imm0_63:$imm)),
4455 (SBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
4456 (i64 (i64shift_sext_i16 imm0_63:$imm)))>;
4458 def : Pat<(shl (i64 (sext GPR32:$Rn)), (i64 imm0_63:$imm)),
4459 (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$Rn, sub_32),
4460 (i64 (i64shift_a imm0_63:$imm)),
4461 (i64 (i64shift_sext_i32 imm0_63:$imm)))>;
4463 // sra patterns have an AddedComplexity of 10, so make sure we have a higher
4464 // AddedComplexity for the following patterns since we want to match sext + sra
4465 // patterns before we attempt to match a single sra node.
4466 let AddedComplexity = 20 in {
4467 // We support all sext + sra combinations which preserve at least one bit of the
4468 // original value which is to be sign extended. E.g. we support shifts up to
4470 def : Pat<(sra (sext_inreg GPR32:$Rn, i8), (i64 imm0_7:$imm)),
4471 (SBFMWri GPR32:$Rn, (i64 imm0_7:$imm), 7)>;
4472 def : Pat<(sra (sext_inreg GPR64:$Rn, i8), (i64 imm0_7:$imm)),
4473 (SBFMXri GPR64:$Rn, (i64 imm0_7:$imm), 7)>;
4475 def : Pat<(sra (sext_inreg GPR32:$Rn, i16), (i64 imm0_15:$imm)),
4476 (SBFMWri GPR32:$Rn, (i64 imm0_15:$imm), 15)>;
4477 def : Pat<(sra (sext_inreg GPR64:$Rn, i16), (i64 imm0_15:$imm)),
4478 (SBFMXri GPR64:$Rn, (i64 imm0_15:$imm), 15)>;
4480 def : Pat<(sra (i64 (sext GPR32:$Rn)), (i64 imm0_31:$imm)),
4481 (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$Rn, sub_32),
4482 (i64 imm0_31:$imm), 31)>;
4483 } // AddedComplexity = 20
4485 // To truncate, we can simply extract from a subregister.
4486 def : Pat<(i32 (trunc GPR64sp:$src)),
4487 (i32 (EXTRACT_SUBREG GPR64sp:$src, sub_32))>;
4489 // __builtin_trap() uses the BRK instruction on ARM64.
4490 def : Pat<(trap), (BRK 1)>;
4492 // Conversions within AdvSIMD types in the same register size are free.
4494 def : Pat<(v1i64 (bitconvert (v2i32 FPR64:$src))), (v1i64 FPR64:$src)>;
4495 def : Pat<(v1i64 (bitconvert (v4i16 FPR64:$src))), (v1i64 FPR64:$src)>;
4496 def : Pat<(v1i64 (bitconvert (v8i8 FPR64:$src))), (v1i64 FPR64:$src)>;
4497 def : Pat<(v1i64 (bitconvert (f64 FPR64:$src))), (v1i64 FPR64:$src)>;
4498 def : Pat<(v1i64 (bitconvert (v2f32 FPR64:$src))), (v1i64 FPR64:$src)>;
4499 def : Pat<(v1i64 (bitconvert (v1f64 FPR64:$src))), (v1i64 FPR64:$src)>;
4501 def : Pat<(v2i32 (bitconvert (v1i64 FPR64:$src))), (v2i32 FPR64:$src)>;
4502 def : Pat<(v2i32 (bitconvert (v4i16 FPR64:$src))), (v2i32 FPR64:$src)>;
4503 def : Pat<(v2i32 (bitconvert (v8i8 FPR64:$src))), (v2i32 FPR64:$src)>;
4504 def : Pat<(v2i32 (bitconvert (f64 FPR64:$src))), (v2i32 FPR64:$src)>;
4505 def : Pat<(v2i32 (bitconvert (v2f32 FPR64:$src))), (v2i32 FPR64:$src)>;
4506 def : Pat<(v2i32 (bitconvert (v1f64 FPR64:$src))), (v2i32 FPR64:$src)>;
4508 def : Pat<(v4i16 (bitconvert (v1i64 FPR64:$src))), (v4i16 FPR64:$src)>;
4509 def : Pat<(v4i16 (bitconvert (v2i32 FPR64:$src))), (v4i16 FPR64:$src)>;
4510 def : Pat<(v4i16 (bitconvert (v8i8 FPR64:$src))), (v4i16 FPR64:$src)>;
4511 def : Pat<(v4i16 (bitconvert (f64 FPR64:$src))), (v4i16 FPR64:$src)>;
4512 def : Pat<(v4i16 (bitconvert (v2f32 FPR64:$src))), (v4i16 FPR64:$src)>;
4513 def : Pat<(v4i16 (bitconvert (v1f64 FPR64:$src))), (v4i16 FPR64:$src)>;
4515 def : Pat<(v8i8 (bitconvert (v1i64 FPR64:$src))), (v8i8 FPR64:$src)>;
4516 def : Pat<(v8i8 (bitconvert (v2i32 FPR64:$src))), (v8i8 FPR64:$src)>;
4517 def : Pat<(v8i8 (bitconvert (v4i16 FPR64:$src))), (v8i8 FPR64:$src)>;
4518 def : Pat<(v8i8 (bitconvert (f64 FPR64:$src))), (v8i8 FPR64:$src)>;
4519 def : Pat<(v8i8 (bitconvert (v2f32 FPR64:$src))), (v8i8 FPR64:$src)>;
4520 def : Pat<(v8i8 (bitconvert (v1f64 FPR64:$src))), (v8i8 FPR64:$src)>;
4522 def : Pat<(f64 (bitconvert (v1i64 FPR64:$src))), (f64 FPR64:$src)>;
4523 def : Pat<(f64 (bitconvert (v2i32 FPR64:$src))), (f64 FPR64:$src)>;
4524 def : Pat<(f64 (bitconvert (v4i16 FPR64:$src))), (f64 FPR64:$src)>;
4525 def : Pat<(f64 (bitconvert (v8i8 FPR64:$src))), (f64 FPR64:$src)>;
4526 def : Pat<(f64 (bitconvert (v2f32 FPR64:$src))), (f64 FPR64:$src)>;
4527 def : Pat<(f64 (bitconvert (v1f64 FPR64:$src))), (f64 FPR64:$src)>;
4529 def : Pat<(v1f64 (bitconvert (v1i64 FPR64:$src))), (v1f64 FPR64:$src)>;
4530 def : Pat<(v1f64 (bitconvert (v2i32 FPR64:$src))), (v1f64 FPR64:$src)>;
4531 def : Pat<(v1f64 (bitconvert (v4i16 FPR64:$src))), (v1f64 FPR64:$src)>;
4532 def : Pat<(v1f64 (bitconvert (v8i8 FPR64:$src))), (v1f64 FPR64:$src)>;
4533 def : Pat<(v1f64 (bitconvert (f64 FPR64:$src))), (v1f64 FPR64:$src)>;
4534 def : Pat<(v1f64 (bitconvert (v2f32 FPR64:$src))), (v1f64 FPR64:$src)>;
4536 def : Pat<(v2f32 (bitconvert (f64 FPR64:$src))), (v2f32 FPR64:$src)>;
4537 def : Pat<(v2f32 (bitconvert (v1i64 FPR64:$src))), (v2f32 FPR64:$src)>;
4538 def : Pat<(v2f32 (bitconvert (v2i32 FPR64:$src))), (v2f32 FPR64:$src)>;
4539 def : Pat<(v2f32 (bitconvert (v4i16 FPR64:$src))), (v2f32 FPR64:$src)>;
4540 def : Pat<(v2f32 (bitconvert (v8i8 FPR64:$src))), (v2f32 FPR64:$src)>;
4541 def : Pat<(v2f32 (bitconvert (v1f64 FPR64:$src))), (v2f32 FPR64:$src)>;
4544 def : Pat<(f128 (bitconvert (v2i64 FPR128:$src))), (f128 FPR128:$src)>;
4545 def : Pat<(f128 (bitconvert (v4i32 FPR128:$src))), (f128 FPR128:$src)>;
4546 def : Pat<(f128 (bitconvert (v8i16 FPR128:$src))), (f128 FPR128:$src)>;
4547 def : Pat<(f128 (bitconvert (v2f64 FPR128:$src))), (f128 FPR128:$src)>;
4548 def : Pat<(f128 (bitconvert (v4f32 FPR128:$src))), (f128 FPR128:$src)>;
4550 def : Pat<(v2f64 (bitconvert (f128 FPR128:$src))), (v2f64 FPR128:$src)>;
4551 def : Pat<(v2f64 (bitconvert (v4i32 FPR128:$src))), (v2f64 FPR128:$src)>;
4552 def : Pat<(v2f64 (bitconvert (v8i16 FPR128:$src))), (v2f64 FPR128:$src)>;
4553 def : Pat<(v2f64 (bitconvert (v16i8 FPR128:$src))), (v2f64 FPR128:$src)>;
4554 def : Pat<(v2f64 (bitconvert (v2i64 FPR128:$src))), (v2f64 FPR128:$src)>;
4555 def : Pat<(v2f64 (bitconvert (v4f32 FPR128:$src))), (v2f64 FPR128:$src)>;
4557 def : Pat<(v4f32 (bitconvert (f128 FPR128:$src))), (v4f32 FPR128:$src)>;
4558 def : Pat<(v4f32 (bitconvert (v4i32 FPR128:$src))), (v4f32 FPR128:$src)>;
4559 def : Pat<(v4f32 (bitconvert (v8i16 FPR128:$src))), (v4f32 FPR128:$src)>;
4560 def : Pat<(v4f32 (bitconvert (v16i8 FPR128:$src))), (v4f32 FPR128:$src)>;
4561 def : Pat<(v4f32 (bitconvert (v2i64 FPR128:$src))), (v4f32 FPR128:$src)>;
4562 def : Pat<(v4f32 (bitconvert (v2f64 FPR128:$src))), (v4f32 FPR128:$src)>;
4564 def : Pat<(v2i64 (bitconvert (f128 FPR128:$src))), (v2i64 FPR128:$src)>;
4565 def : Pat<(v2i64 (bitconvert (v4i32 FPR128:$src))), (v2i64 FPR128:$src)>;
4566 def : Pat<(v2i64 (bitconvert (v8i16 FPR128:$src))), (v2i64 FPR128:$src)>;
4567 def : Pat<(v2i64 (bitconvert (v16i8 FPR128:$src))), (v2i64 FPR128:$src)>;
4568 def : Pat<(v2i64 (bitconvert (v2f64 FPR128:$src))), (v2i64 FPR128:$src)>;
4569 def : Pat<(v2i64 (bitconvert (v4f32 FPR128:$src))), (v2i64 FPR128:$src)>;
4571 def : Pat<(v4i32 (bitconvert (f128 FPR128:$src))), (v4i32 FPR128:$src)>;
4572 def : Pat<(v4i32 (bitconvert (v2i64 FPR128:$src))), (v4i32 FPR128:$src)>;
4573 def : Pat<(v4i32 (bitconvert (v8i16 FPR128:$src))), (v4i32 FPR128:$src)>;
4574 def : Pat<(v4i32 (bitconvert (v16i8 FPR128:$src))), (v4i32 FPR128:$src)>;
4575 def : Pat<(v4i32 (bitconvert (v2f64 FPR128:$src))), (v4i32 FPR128:$src)>;
4576 def : Pat<(v4i32 (bitconvert (v4f32 FPR128:$src))), (v4i32 FPR128:$src)>;
4578 def : Pat<(v8i16 (bitconvert (f128 FPR128:$src))), (v8i16 FPR128:$src)>;
4579 def : Pat<(v8i16 (bitconvert (v2i64 FPR128:$src))), (v8i16 FPR128:$src)>;
4580 def : Pat<(v8i16 (bitconvert (v4i32 FPR128:$src))), (v8i16 FPR128:$src)>;
4581 def : Pat<(v8i16 (bitconvert (v16i8 FPR128:$src))), (v8i16 FPR128:$src)>;
4582 def : Pat<(v8i16 (bitconvert (v2f64 FPR128:$src))), (v8i16 FPR128:$src)>;
4583 def : Pat<(v8i16 (bitconvert (v4f32 FPR128:$src))), (v8i16 FPR128:$src)>;
4585 def : Pat<(v16i8 (bitconvert (f128 FPR128:$src))), (v16i8 FPR128:$src)>;
4586 def : Pat<(v16i8 (bitconvert (v2i64 FPR128:$src))), (v16i8 FPR128:$src)>;
4587 def : Pat<(v16i8 (bitconvert (v4i32 FPR128:$src))), (v16i8 FPR128:$src)>;
4588 def : Pat<(v16i8 (bitconvert (v8i16 FPR128:$src))), (v16i8 FPR128:$src)>;
4589 def : Pat<(v16i8 (bitconvert (v2f64 FPR128:$src))), (v16i8 FPR128:$src)>;
4590 def : Pat<(v16i8 (bitconvert (v4f32 FPR128:$src))), (v16i8 FPR128:$src)>;
4592 def : Pat<(v8i8 (extract_subvector (v16i8 FPR128:$Rn), (i64 1))),
4593 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
4594 def : Pat<(v4i16 (extract_subvector (v8i16 FPR128:$Rn), (i64 1))),
4595 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
4596 def : Pat<(v2i32 (extract_subvector (v4i32 FPR128:$Rn), (i64 1))),
4597 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
4598 def : Pat<(v1i64 (extract_subvector (v2i64 FPR128:$Rn), (i64 1))),
4599 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
4601 // A 64-bit subvector insert to the first 128-bit vector position
4602 // is a subregister copy that needs no instruction.
4603 def : Pat<(insert_subvector undef, (v1i64 FPR64:$src), (i32 0)),
4604 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
4605 def : Pat<(insert_subvector undef, (v1f64 FPR64:$src), (i32 0)),
4606 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
4607 def : Pat<(insert_subvector undef, (v2i32 FPR64:$src), (i32 0)),
4608 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
4609 def : Pat<(insert_subvector undef, (v2f32 FPR64:$src), (i32 0)),
4610 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
4611 def : Pat<(insert_subvector undef, (v4i16 FPR64:$src), (i32 0)),
4612 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
4613 def : Pat<(insert_subvector undef, (v8i8 FPR64:$src), (i32 0)),
4614 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
4616 // Use pair-wise add instructions when summing up the lanes for v2f64, v2i64
4618 def : Pat<(i64 (add (vector_extract (v2i64 FPR128:$Rn), (i64 0)),
4619 (vector_extract (v2i64 FPR128:$Rn), (i64 1)))),
4620 (i64 (ADDPv2i64p (v2i64 FPR128:$Rn)))>;
4621 def : Pat<(f64 (fadd (vector_extract (v2f64 FPR128:$Rn), (i64 0)),
4622 (vector_extract (v2f64 FPR128:$Rn), (i64 1)))),
4623 (f64 (FADDPv2i64p (v2f64 FPR128:$Rn)))>;
4624 // vector_extract on 64-bit vectors gets promoted to a 128 bit vector,
4625 // so we match on v4f32 here, not v2f32. This will also catch adding
4626 // the low two lanes of a true v4f32 vector.
4627 def : Pat<(fadd (vector_extract (v4f32 FPR128:$Rn), (i64 0)),
4628 (vector_extract (v4f32 FPR128:$Rn), (i64 1))),
4629 (f32 (FADDPv2i32p (EXTRACT_SUBREG FPR128:$Rn, dsub)))>;
4631 // Scalar 64-bit shifts in FPR64 registers.
4632 def : Pat<(i64 (int_arm64_neon_sshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
4633 (SSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
4634 def : Pat<(i64 (int_arm64_neon_ushl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
4635 (USHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
4636 def : Pat<(i64 (int_arm64_neon_srshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
4637 (SRSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
4638 def : Pat<(i64 (int_arm64_neon_urshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
4639 (URSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
4641 // Tail call return handling. These are all compiler pseudo-instructions,
4642 // so no encoding information or anything like that.
4643 let isCall = 1, isTerminator = 1, isReturn = 1, isBarrier = 1, Uses = [SP] in {
4644 def TCRETURNdi : Pseudo<(outs), (ins i64imm:$dst), []>;
4645 def TCRETURNri : Pseudo<(outs), (ins tcGPR64:$dst), []>;
4648 def : Pat<(ARM64tcret tcGPR64:$dst), (TCRETURNri tcGPR64:$dst)>;
4649 def : Pat<(ARM64tcret (i64 tglobaladdr:$dst)), (TCRETURNdi texternalsym:$dst)>;
4650 def : Pat<(ARM64tcret (i64 texternalsym:$dst)), (TCRETURNdi texternalsym:$dst)>;
4652 include "ARM64InstrAtomics.td"