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 // ARM64-specific DAG Nodes.
18 // SDTBinaryArithWithFlagsOut - RES1, FLAGS = op LHS, RHS
19 def SDTBinaryArithWithFlagsOut : SDTypeProfile<2, 2,
22 SDTCisInt<0>, SDTCisVT<1, i32>]>;
24 // SDTBinaryArithWithFlagsIn - RES1, FLAGS = op LHS, RHS, FLAGS
25 def SDTBinaryArithWithFlagsIn : SDTypeProfile<1, 3,
31 // SDTBinaryArithWithFlagsInOut - RES1, FLAGS = op LHS, RHS, FLAGS
32 def SDTBinaryArithWithFlagsInOut : SDTypeProfile<2, 3,
39 def SDT_ARM64Brcond : SDTypeProfile<0, 3,
40 [SDTCisVT<0, OtherVT>, SDTCisVT<1, i32>,
42 def SDT_ARM64cbz : SDTypeProfile<0, 2, [SDTCisInt<0>, SDTCisVT<1, OtherVT>]>;
43 def SDT_ARM64tbz : SDTypeProfile<0, 3, [SDTCisVT<0, i64>, SDTCisVT<1, i64>,
44 SDTCisVT<2, OtherVT>]>;
47 def SDT_ARM64CSel : SDTypeProfile<1, 4,
52 def SDT_ARM64FCmp : SDTypeProfile<0, 2,
55 def SDT_ARM64Dup : SDTypeProfile<1, 1, [SDTCisVec<0>]>;
56 def SDT_ARM64DupLane : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisInt<2>]>;
57 def SDT_ARM64Zip : SDTypeProfile<1, 2, [SDTCisVec<0>,
60 def SDT_ARM64MOVIedit : SDTypeProfile<1, 1, [SDTCisInt<1>]>;
61 def SDT_ARM64MOVIshift : SDTypeProfile<1, 2, [SDTCisInt<1>, SDTCisInt<2>]>;
62 def SDT_ARM64vecimm : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0,1>,
63 SDTCisInt<2>, SDTCisInt<3>]>;
64 def SDT_ARM64UnaryVec: SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0,1>]>;
65 def SDT_ARM64ExtVec: SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0,1>,
66 SDTCisSameAs<0,2>, SDTCisInt<3>]>;
67 def SDT_ARM64vshift : SDTypeProfile<1, 2, [SDTCisSameAs<0,1>, SDTCisInt<2>]>;
69 def SDT_ARM64unvec : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0,1>]>;
70 def SDT_ARM64fcmpz : SDTypeProfile<1, 1, []>;
71 def SDT_ARM64fcmp : SDTypeProfile<1, 2, [SDTCisSameAs<1,2>]>;
72 def SDT_ARM64binvec : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisSameAs<0,1>,
74 def SDT_ARM64trivec : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0,1>,
77 def SDT_ARM64TCRET : SDTypeProfile<0, 1, [SDTCisPtrTy<0>]>;
78 def SDT_ARM64PREFETCH : SDTypeProfile<0, 2, [SDTCisVT<0, i32>, SDTCisPtrTy<1>]>;
80 def SDT_ARM64ITOF : SDTypeProfile<1, 1, [SDTCisFP<0>, SDTCisSameAs<0,1>]>;
82 def SDT_ARM64TLSDescCall : SDTypeProfile<0, -2, [SDTCisPtrTy<0>,
84 def SDT_ARM64WrapperLarge : SDTypeProfile<1, 4,
85 [SDTCisVT<0, i64>, SDTCisVT<1, i32>,
86 SDTCisSameAs<1, 2>, SDTCisSameAs<1, 3>,
91 def ARM64adrp : SDNode<"ARM64ISD::ADRP", SDTIntUnaryOp, []>;
92 def ARM64addlow : SDNode<"ARM64ISD::ADDlow", SDTIntBinOp, []>;
93 def ARM64LOADgot : SDNode<"ARM64ISD::LOADgot", SDTIntUnaryOp>;
94 def ARM64callseq_start : SDNode<"ISD::CALLSEQ_START",
95 SDCallSeqStart<[ SDTCisVT<0, i32> ]>,
96 [SDNPHasChain, SDNPOutGlue]>;
97 def ARM64callseq_end : SDNode<"ISD::CALLSEQ_END",
98 SDCallSeqEnd<[ SDTCisVT<0, i32>,
100 [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue]>;
101 def ARM64call : SDNode<"ARM64ISD::CALL",
102 SDTypeProfile<0, -1, [SDTCisPtrTy<0>]>,
103 [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue,
105 def ARM64brcond : SDNode<"ARM64ISD::BRCOND", SDT_ARM64Brcond,
107 def ARM64cbz : SDNode<"ARM64ISD::CBZ", SDT_ARM64cbz,
109 def ARM64cbnz : SDNode<"ARM64ISD::CBNZ", SDT_ARM64cbz,
111 def ARM64tbz : SDNode<"ARM64ISD::TBZ", SDT_ARM64tbz,
113 def ARM64tbnz : SDNode<"ARM64ISD::TBNZ", SDT_ARM64tbz,
117 def ARM64csel : SDNode<"ARM64ISD::CSEL", SDT_ARM64CSel>;
118 def ARM64csinv : SDNode<"ARM64ISD::CSINV", SDT_ARM64CSel>;
119 def ARM64csneg : SDNode<"ARM64ISD::CSNEG", SDT_ARM64CSel>;
120 def ARM64csinc : SDNode<"ARM64ISD::CSINC", SDT_ARM64CSel>;
121 def ARM64retflag : SDNode<"ARM64ISD::RET_FLAG", SDTNone,
122 [SDNPHasChain, SDNPOptInGlue, SDNPVariadic]>;
123 def ARM64adc : SDNode<"ARM64ISD::ADC", SDTBinaryArithWithFlagsIn >;
124 def ARM64sbc : SDNode<"ARM64ISD::SBC", SDTBinaryArithWithFlagsIn>;
125 def ARM64add_flag : SDNode<"ARM64ISD::ADDS", SDTBinaryArithWithFlagsOut,
127 def ARM64sub_flag : SDNode<"ARM64ISD::SUBS", SDTBinaryArithWithFlagsOut>;
128 def ARM64and_flag : SDNode<"ARM64ISD::ANDS", SDTBinaryArithWithFlagsOut>;
129 def ARM64adc_flag : SDNode<"ARM64ISD::ADCS", SDTBinaryArithWithFlagsInOut>;
130 def ARM64sbc_flag : SDNode<"ARM64ISD::SBCS", SDTBinaryArithWithFlagsInOut>;
132 def ARM64threadpointer : SDNode<"ARM64ISD::THREAD_POINTER", SDTPtrLeaf>;
134 def ARM64fcmp : SDNode<"ARM64ISD::FCMP", SDT_ARM64FCmp>;
136 def ARM64fmax : SDNode<"ARM64ISD::FMAX", SDTFPBinOp>;
137 def ARM64fmin : SDNode<"ARM64ISD::FMIN", SDTFPBinOp>;
139 def ARM64dup : SDNode<"ARM64ISD::DUP", SDT_ARM64Dup>;
140 def ARM64duplane8 : SDNode<"ARM64ISD::DUPLANE8", SDT_ARM64DupLane>;
141 def ARM64duplane16 : SDNode<"ARM64ISD::DUPLANE16", SDT_ARM64DupLane>;
142 def ARM64duplane32 : SDNode<"ARM64ISD::DUPLANE32", SDT_ARM64DupLane>;
143 def ARM64duplane64 : SDNode<"ARM64ISD::DUPLANE64", SDT_ARM64DupLane>;
145 def ARM64zip1 : SDNode<"ARM64ISD::ZIP1", SDT_ARM64Zip>;
146 def ARM64zip2 : SDNode<"ARM64ISD::ZIP2", SDT_ARM64Zip>;
147 def ARM64uzp1 : SDNode<"ARM64ISD::UZP1", SDT_ARM64Zip>;
148 def ARM64uzp2 : SDNode<"ARM64ISD::UZP2", SDT_ARM64Zip>;
149 def ARM64trn1 : SDNode<"ARM64ISD::TRN1", SDT_ARM64Zip>;
150 def ARM64trn2 : SDNode<"ARM64ISD::TRN2", SDT_ARM64Zip>;
152 def ARM64movi_edit : SDNode<"ARM64ISD::MOVIedit", SDT_ARM64MOVIedit>;
153 def ARM64movi_shift : SDNode<"ARM64ISD::MOVIshift", SDT_ARM64MOVIshift>;
154 def ARM64movi_msl : SDNode<"ARM64ISD::MOVImsl", SDT_ARM64MOVIshift>;
155 def ARM64mvni_shift : SDNode<"ARM64ISD::MVNIshift", SDT_ARM64MOVIshift>;
156 def ARM64mvni_msl : SDNode<"ARM64ISD::MVNImsl", SDT_ARM64MOVIshift>;
157 def ARM64movi : SDNode<"ARM64ISD::MOVI", SDT_ARM64MOVIedit>;
158 def ARM64fmov : SDNode<"ARM64ISD::FMOV", SDT_ARM64MOVIedit>;
160 def ARM64rev16 : SDNode<"ARM64ISD::REV16", SDT_ARM64UnaryVec>;
161 def ARM64rev32 : SDNode<"ARM64ISD::REV32", SDT_ARM64UnaryVec>;
162 def ARM64rev64 : SDNode<"ARM64ISD::REV64", SDT_ARM64UnaryVec>;
163 def ARM64ext : SDNode<"ARM64ISD::EXT", SDT_ARM64ExtVec>;
165 def ARM64vashr : SDNode<"ARM64ISD::VASHR", SDT_ARM64vshift>;
166 def ARM64vlshr : SDNode<"ARM64ISD::VLSHR", SDT_ARM64vshift>;
167 def ARM64vshl : SDNode<"ARM64ISD::VSHL", SDT_ARM64vshift>;
168 def ARM64sqshli : SDNode<"ARM64ISD::SQSHL_I", SDT_ARM64vshift>;
169 def ARM64uqshli : SDNode<"ARM64ISD::UQSHL_I", SDT_ARM64vshift>;
170 def ARM64sqshlui : SDNode<"ARM64ISD::SQSHLU_I", SDT_ARM64vshift>;
171 def ARM64srshri : SDNode<"ARM64ISD::SRSHR_I", SDT_ARM64vshift>;
172 def ARM64urshri : SDNode<"ARM64ISD::URSHR_I", SDT_ARM64vshift>;
174 def ARM64not: SDNode<"ARM64ISD::NOT", SDT_ARM64unvec>;
175 def ARM64bit: SDNode<"ARM64ISD::BIT", SDT_ARM64trivec>;
177 def ARM64cmeq: SDNode<"ARM64ISD::CMEQ", SDT_ARM64binvec>;
178 def ARM64cmge: SDNode<"ARM64ISD::CMGE", SDT_ARM64binvec>;
179 def ARM64cmgt: SDNode<"ARM64ISD::CMGT", SDT_ARM64binvec>;
180 def ARM64cmhi: SDNode<"ARM64ISD::CMHI", SDT_ARM64binvec>;
181 def ARM64cmhs: SDNode<"ARM64ISD::CMHS", SDT_ARM64binvec>;
183 def ARM64fcmeq: SDNode<"ARM64ISD::FCMEQ", SDT_ARM64fcmp>;
184 def ARM64fcmge: SDNode<"ARM64ISD::FCMGE", SDT_ARM64fcmp>;
185 def ARM64fcmgt: SDNode<"ARM64ISD::FCMGT", SDT_ARM64fcmp>;
187 def ARM64cmeqz: SDNode<"ARM64ISD::CMEQz", SDT_ARM64unvec>;
188 def ARM64cmgez: SDNode<"ARM64ISD::CMGEz", SDT_ARM64unvec>;
189 def ARM64cmgtz: SDNode<"ARM64ISD::CMGTz", SDT_ARM64unvec>;
190 def ARM64cmlez: SDNode<"ARM64ISD::CMLEz", SDT_ARM64unvec>;
191 def ARM64cmltz: SDNode<"ARM64ISD::CMLTz", SDT_ARM64unvec>;
192 def ARM64cmtst : PatFrag<(ops node:$LHS, node:$RHS),
193 (ARM64not (ARM64cmeqz (and node:$LHS, node:$RHS)))>;
195 def ARM64fcmeqz: SDNode<"ARM64ISD::FCMEQz", SDT_ARM64fcmpz>;
196 def ARM64fcmgez: SDNode<"ARM64ISD::FCMGEz", SDT_ARM64fcmpz>;
197 def ARM64fcmgtz: SDNode<"ARM64ISD::FCMGTz", SDT_ARM64fcmpz>;
198 def ARM64fcmlez: SDNode<"ARM64ISD::FCMLEz", SDT_ARM64fcmpz>;
199 def ARM64fcmltz: SDNode<"ARM64ISD::FCMLTz", SDT_ARM64fcmpz>;
201 def ARM64bici: SDNode<"ARM64ISD::BICi", SDT_ARM64vecimm>;
202 def ARM64orri: SDNode<"ARM64ISD::ORRi", SDT_ARM64vecimm>;
204 def ARM64neg : SDNode<"ARM64ISD::NEG", SDT_ARM64unvec>;
206 def ARM64tcret: SDNode<"ARM64ISD::TC_RETURN", SDT_ARM64TCRET,
207 [SDNPHasChain, SDNPOptInGlue, SDNPVariadic]>;
209 def ARM64Prefetch : SDNode<"ARM64ISD::PREFETCH", SDT_ARM64PREFETCH,
210 [SDNPHasChain, SDNPSideEffect]>;
212 def ARM64sitof: SDNode<"ARM64ISD::SITOF", SDT_ARM64ITOF>;
213 def ARM64uitof: SDNode<"ARM64ISD::UITOF", SDT_ARM64ITOF>;
215 def ARM64tlsdesc_call : SDNode<"ARM64ISD::TLSDESC_CALL", SDT_ARM64TLSDescCall,
216 [SDNPInGlue, SDNPOutGlue, SDNPHasChain,
219 def ARM64WrapperLarge : SDNode<"ARM64ISD::WrapperLarge", SDT_ARM64WrapperLarge>;
222 //===----------------------------------------------------------------------===//
224 //===----------------------------------------------------------------------===//
226 // ARM64 Instruction Predicate Definitions.
228 def HasZCZ : Predicate<"Subtarget->hasZeroCycleZeroing()">;
229 def NoZCZ : Predicate<"!Subtarget->hasZeroCycleZeroing()">;
230 def IsDarwin : Predicate<"Subtarget->isTargetDarwin()">;
231 def IsNotDarwin: Predicate<"!Subtarget->isTargetDarwin()">;
232 def ForCodeSize : Predicate<"ForCodeSize">;
233 def NotForCodeSize : Predicate<"!ForCodeSize">;
235 include "ARM64InstrFormats.td"
237 //===----------------------------------------------------------------------===//
239 //===----------------------------------------------------------------------===//
240 // Miscellaneous instructions.
241 //===----------------------------------------------------------------------===//
243 let Defs = [SP], Uses = [SP], hasSideEffects = 1, isCodeGenOnly = 1 in {
244 def ADJCALLSTACKDOWN : Pseudo<(outs), (ins i32imm:$amt),
245 [(ARM64callseq_start timm:$amt)]>;
246 def ADJCALLSTACKUP : Pseudo<(outs), (ins i32imm:$amt1, i32imm:$amt2),
247 [(ARM64callseq_end timm:$amt1, timm:$amt2)]>;
248 } // Defs = [SP], Uses = [SP], hasSideEffects = 1, isCodeGenOnly = 1
250 let isReMaterializable = 1, isCodeGenOnly = 1 in {
251 // FIXME: The following pseudo instructions are only needed because remat
252 // cannot handle multiple instructions. When that changes, they can be
253 // removed, along with the ARM64Wrapper node.
255 let AddedComplexity = 10 in
256 def LOADgot : Pseudo<(outs GPR64:$dst), (ins i64imm:$addr),
257 [(set GPR64:$dst, (ARM64LOADgot tglobaladdr:$addr))]>,
260 // The MOVaddr instruction should match only when the add is not folded
261 // into a load or store address.
263 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
264 [(set GPR64:$dst, (ARM64addlow (ARM64adrp tglobaladdr:$hi),
265 tglobaladdr:$low))]>,
266 Sched<[WriteAdrAdr]>;
268 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
269 [(set GPR64:$dst, (ARM64addlow (ARM64adrp tjumptable:$hi),
271 Sched<[WriteAdrAdr]>;
273 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
274 [(set GPR64:$dst, (ARM64addlow (ARM64adrp tconstpool:$hi),
276 Sched<[WriteAdrAdr]>;
278 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
279 [(set GPR64:$dst, (ARM64addlow (ARM64adrp tblockaddress:$hi),
280 tblockaddress:$low))]>,
281 Sched<[WriteAdrAdr]>;
283 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
284 [(set GPR64:$dst, (ARM64addlow (ARM64adrp tglobaltlsaddr:$hi),
285 tglobaltlsaddr:$low))]>,
286 Sched<[WriteAdrAdr]>;
288 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
289 [(set GPR64:$dst, (ARM64addlow (ARM64adrp texternalsym:$hi),
290 texternalsym:$low))]>,
291 Sched<[WriteAdrAdr]>;
293 } // isReMaterializable, isCodeGenOnly
295 def : Pat<(ARM64LOADgot tglobaltlsaddr:$addr),
296 (LOADgot tglobaltlsaddr:$addr)>;
298 def : Pat<(ARM64LOADgot texternalsym:$addr),
299 (LOADgot texternalsym:$addr)>;
301 def : Pat<(ARM64LOADgot tconstpool:$addr),
302 (LOADgot tconstpool:$addr)>;
304 //===----------------------------------------------------------------------===//
305 // System instructions.
306 //===----------------------------------------------------------------------===//
308 def HINT : HintI<"hint">;
309 def : InstAlias<"nop", (HINT 0b000)>;
310 def : InstAlias<"yield",(HINT 0b001)>;
311 def : InstAlias<"wfe", (HINT 0b010)>;
312 def : InstAlias<"wfi", (HINT 0b011)>;
313 def : InstAlias<"sev", (HINT 0b100)>;
314 def : InstAlias<"sevl", (HINT 0b101)>;
316 // As far as LLVM is concerned this writes to the system's exclusive monitors.
317 let mayLoad = 1, mayStore = 1 in
318 def CLREX : CRmSystemI<imm0_15, 0b010, "clrex">;
320 def DMB : CRmSystemI<barrier_op, 0b101, "dmb">;
321 def DSB : CRmSystemI<barrier_op, 0b100, "dsb">;
322 def ISB : CRmSystemI<barrier_op, 0b110, "isb">;
323 def : InstAlias<"clrex", (CLREX 0xf)>;
324 def : InstAlias<"isb", (ISB 0xf)>;
328 def MSRcpsr: MSRcpsrI;
330 // The thread pointer (on Linux, at least, where this has been implemented) is
332 def : Pat<(ARM64threadpointer), (MRS 0xde82)>;
334 // Generic system instructions
335 def SYS : SystemI<0, "sys">;
336 def SYSxt : SystemXtI<0, "sys">;
337 def SYSLxt : SystemLXtI<1, "sysl">;
339 //===----------------------------------------------------------------------===//
340 // Move immediate instructions.
341 //===----------------------------------------------------------------------===//
343 defm MOVK : InsertImmediate<0b11, "movk">;
344 defm MOVN : MoveImmediate<0b00, "movn">;
346 let PostEncoderMethod = "fixMOVZ" in
347 defm MOVZ : MoveImmediate<0b10, "movz">;
349 def : InstAlias<"movk $dst, $imm", (MOVKWi GPR32:$dst, imm0_65535:$imm, 0)>;
350 def : InstAlias<"movk $dst, $imm", (MOVKXi GPR64:$dst, imm0_65535:$imm, 0)>;
351 def : InstAlias<"movn $dst, $imm", (MOVNWi GPR32:$dst, imm0_65535:$imm, 0)>;
352 def : InstAlias<"movn $dst, $imm", (MOVNXi GPR64:$dst, imm0_65535:$imm, 0)>;
353 def : InstAlias<"movz $dst, $imm", (MOVZWi GPR32:$dst, imm0_65535:$imm, 0)>;
354 def : InstAlias<"movz $dst, $imm", (MOVZXi GPR64:$dst, imm0_65535:$imm, 0)>;
356 def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movz_symbol_g3:$sym, 48)>;
357 def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movz_symbol_g2:$sym, 32)>;
358 def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movz_symbol_g1:$sym, 16)>;
359 def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movz_symbol_g0:$sym, 0)>;
361 def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movz_symbol_g3:$sym, 48)>;
362 def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movz_symbol_g2:$sym, 32)>;
363 def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movz_symbol_g1:$sym, 16)>;
364 def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movz_symbol_g0:$sym, 0)>;
366 def : InstAlias<"movz $Rd, $sym", (MOVZWi GPR32:$Rd, movz_symbol_g3:$sym, 48)>;
367 def : InstAlias<"movz $Rd, $sym", (MOVZWi GPR32:$Rd, movz_symbol_g2:$sym, 32)>;
368 def : InstAlias<"movz $Rd, $sym", (MOVZWi GPR32:$Rd, movz_symbol_g1:$sym, 16)>;
369 def : InstAlias<"movz $Rd, $sym", (MOVZWi GPR32:$Rd, movz_symbol_g0:$sym, 0)>;
371 def : InstAlias<"movk $Rd, $sym", (MOVKXi GPR64:$Rd, movk_symbol_g2:$sym, 32)>;
372 def : InstAlias<"movk $Rd, $sym", (MOVKXi GPR64:$Rd, movk_symbol_g1:$sym, 16)>;
373 def : InstAlias<"movk $Rd, $sym", (MOVKXi GPR64:$Rd, movk_symbol_g0:$sym, 0)>;
375 def : InstAlias<"movk $Rd, $sym", (MOVKWi GPR32:$Rd, movk_symbol_g2:$sym, 32)>;
376 def : InstAlias<"movk $Rd, $sym", (MOVKWi GPR32:$Rd, movk_symbol_g1:$sym, 16)>;
377 def : InstAlias<"movk $Rd, $sym", (MOVKWi GPR32:$Rd, movk_symbol_g0:$sym, 0)>;
379 let isReMaterializable = 1, isCodeGenOnly = 1, isMoveImm = 1,
380 isAsCheapAsAMove = 1 in {
381 // FIXME: The following pseudo instructions are only needed because remat
382 // cannot handle multiple instructions. When that changes, we can select
383 // directly to the real instructions and get rid of these pseudos.
386 : Pseudo<(outs GPR32:$dst), (ins i32imm:$src),
387 [(set GPR32:$dst, imm:$src)]>,
390 : Pseudo<(outs GPR64:$dst), (ins i64imm:$src),
391 [(set GPR64:$dst, imm:$src)]>,
393 } // isReMaterializable, isCodeGenOnly
395 def : Pat<(ARM64WrapperLarge tglobaladdr:$g3, tglobaladdr:$g2,
396 tglobaladdr:$g1, tglobaladdr:$g0),
397 (MOVKXi (MOVKXi (MOVKXi (MOVZXi tglobaladdr:$g3, 48),
398 tglobaladdr:$g2, 32),
399 tglobaladdr:$g1, 16),
400 tglobaladdr:$g0, 0)>;
402 def : Pat<(ARM64WrapperLarge tblockaddress:$g3, tblockaddress:$g2,
403 tblockaddress:$g1, tblockaddress:$g0),
404 (MOVKXi (MOVKXi (MOVKXi (MOVZXi tblockaddress:$g3, 48),
405 tblockaddress:$g2, 32),
406 tblockaddress:$g1, 16),
407 tblockaddress:$g0, 0)>;
409 def : Pat<(ARM64WrapperLarge tconstpool:$g3, tconstpool:$g2,
410 tconstpool:$g1, tconstpool:$g0),
411 (MOVKXi (MOVKXi (MOVKXi (MOVZXi tconstpool:$g3, 48),
417 //===----------------------------------------------------------------------===//
418 // Arithmetic instructions.
419 //===----------------------------------------------------------------------===//
421 // Add/subtract with carry.
422 defm ADC : AddSubCarry<0, "adc", "adcs", ARM64adc, ARM64adc_flag>;
423 defm SBC : AddSubCarry<1, "sbc", "sbcs", ARM64sbc, ARM64sbc_flag>;
425 def : InstAlias<"ngc $dst, $src", (SBCWr GPR32:$dst, WZR, GPR32:$src)>;
426 def : InstAlias<"ngc $dst, $src", (SBCXr GPR64:$dst, XZR, GPR64:$src)>;
427 def : InstAlias<"ngcs $dst, $src", (SBCSWr GPR32:$dst, WZR, GPR32:$src)>;
428 def : InstAlias<"ngcs $dst, $src", (SBCSXr GPR64:$dst, XZR, GPR64:$src)>;
431 defm ADD : AddSub<0, "add", add>;
432 defm SUB : AddSub<1, "sub">;
434 defm ADDS : AddSubS<0, "adds", ARM64add_flag>;
435 defm SUBS : AddSubS<1, "subs", ARM64sub_flag>;
437 // Use SUBS instead of SUB to enable CSE between SUBS and SUB.
438 def : Pat<(sub GPR32sp:$Rn, addsub_shifted_imm32:$imm),
439 (SUBSWri GPR32sp:$Rn, addsub_shifted_imm32:$imm)>;
440 def : Pat<(sub GPR64sp:$Rn, addsub_shifted_imm64:$imm),
441 (SUBSXri GPR64sp:$Rn, addsub_shifted_imm64:$imm)>;
442 def : Pat<(sub GPR32:$Rn, GPR32:$Rm),
443 (SUBSWrr GPR32:$Rn, GPR32:$Rm)>;
444 def : Pat<(sub GPR64:$Rn, GPR64:$Rm),
445 (SUBSXrr GPR64:$Rn, GPR64:$Rm)>;
446 def : Pat<(sub GPR32:$Rn, arith_shifted_reg32:$Rm),
447 (SUBSWrs GPR32:$Rn, arith_shifted_reg32:$Rm)>;
448 def : Pat<(sub GPR64:$Rn, arith_shifted_reg64:$Rm),
449 (SUBSXrs GPR64:$Rn, arith_shifted_reg64:$Rm)>;
450 def : Pat<(sub GPR32sp:$R2, arith_extended_reg32<i32>:$R3),
451 (SUBSWrx GPR32sp:$R2, arith_extended_reg32<i32>:$R3)>;
452 def : Pat<(sub GPR64sp:$R2, arith_extended_reg32to64<i64>:$R3),
453 (SUBSXrx GPR64sp:$R2, arith_extended_reg32to64<i64>:$R3)>;
455 // Because of the immediate format for add/sub-imm instructions, the
456 // expression (add x, -1) must be transformed to (SUB{W,X}ri x, 1).
457 // These patterns capture that transformation.
458 let AddedComplexity = 1 in {
459 def : Pat<(add GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
460 (SUBSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
461 def : Pat<(add GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
462 (SUBSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
463 def : Pat<(sub GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
464 (ADDWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
465 def : Pat<(sub GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
466 (ADDXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
469 def : InstAlias<"neg $dst, $src", (SUBWrs GPR32:$dst, WZR, GPR32:$src, 0)>;
470 def : InstAlias<"neg $dst, $src", (SUBXrs GPR64:$dst, XZR, GPR64:$src, 0)>;
471 def : InstAlias<"neg $dst, $src, $shift",
472 (SUBWrs GPR32:$dst, WZR, GPR32:$src, arith_shift:$shift)>;
473 def : InstAlias<"neg $dst, $src, $shift",
474 (SUBXrs GPR64:$dst, XZR, GPR64:$src, arith_shift:$shift)>;
476 // Because of the immediate format for add/sub-imm instructions, the
477 // expression (add x, -1) must be transformed to (SUB{W,X}ri x, 1).
478 // These patterns capture that transformation.
479 let AddedComplexity = 1 in {
480 def : Pat<(ARM64add_flag GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
481 (SUBSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
482 def : Pat<(ARM64add_flag GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
483 (SUBSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
484 def : Pat<(ARM64sub_flag GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
485 (ADDSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
486 def : Pat<(ARM64sub_flag GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
487 (ADDSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
490 def : InstAlias<"negs $dst, $src", (SUBSWrs GPR32:$dst, WZR, GPR32:$src, 0)>;
491 def : InstAlias<"negs $dst, $src", (SUBSXrs GPR64:$dst, XZR, GPR64:$src, 0)>;
492 def : InstAlias<"negs $dst, $src, $shift",
493 (SUBSWrs GPR32:$dst, WZR, GPR32:$src, arith_shift:$shift)>;
494 def : InstAlias<"negs $dst, $src, $shift",
495 (SUBSXrs GPR64:$dst, XZR, GPR64:$src, arith_shift:$shift)>;
497 // Unsigned/Signed divide
498 defm UDIV : Div<0, "udiv", udiv>;
499 defm SDIV : Div<1, "sdiv", sdiv>;
500 let isCodeGenOnly = 1 in {
501 defm UDIV_Int : Div<0, "udiv", int_arm64_udiv>;
502 defm SDIV_Int : Div<1, "sdiv", int_arm64_sdiv>;
506 defm ASRV : Shift<0b10, "asrv", sra>;
507 defm LSLV : Shift<0b00, "lslv", shl>;
508 defm LSRV : Shift<0b01, "lsrv", srl>;
509 defm RORV : Shift<0b11, "rorv", rotr>;
511 def : ShiftAlias<"asr", ASRVWr, GPR32>;
512 def : ShiftAlias<"asr", ASRVXr, GPR64>;
513 def : ShiftAlias<"lsl", LSLVWr, GPR32>;
514 def : ShiftAlias<"lsl", LSLVXr, GPR64>;
515 def : ShiftAlias<"lsr", LSRVWr, GPR32>;
516 def : ShiftAlias<"lsr", LSRVXr, GPR64>;
517 def : ShiftAlias<"ror", RORVWr, GPR32>;
518 def : ShiftAlias<"ror", RORVXr, GPR64>;
521 let AddedComplexity = 7 in {
522 defm MADD : MulAccum<0, "madd", add>;
523 defm MSUB : MulAccum<1, "msub", sub>;
525 def : Pat<(i32 (mul GPR32:$Rn, GPR32:$Rm)),
526 (MADDWrrr GPR32:$Rn, GPR32:$Rm, WZR)>;
527 def : Pat<(i64 (mul GPR64:$Rn, GPR64:$Rm)),
528 (MADDXrrr GPR64:$Rn, GPR64:$Rm, XZR)>;
530 def : Pat<(i32 (ineg (mul GPR32:$Rn, GPR32:$Rm))),
531 (MSUBWrrr GPR32:$Rn, GPR32:$Rm, WZR)>;
532 def : Pat<(i64 (ineg (mul GPR64:$Rn, GPR64:$Rm))),
533 (MSUBXrrr GPR64:$Rn, GPR64:$Rm, XZR)>;
534 } // AddedComplexity = 7
536 let AddedComplexity = 5 in {
537 def SMADDLrrr : WideMulAccum<0, 0b001, "smaddl", add, sext>;
538 def SMSUBLrrr : WideMulAccum<1, 0b001, "smsubl", sub, sext>;
539 def UMADDLrrr : WideMulAccum<0, 0b101, "umaddl", add, zext>;
540 def UMSUBLrrr : WideMulAccum<1, 0b101, "umsubl", sub, zext>;
542 def : Pat<(i64 (mul (sext GPR32:$Rn), (sext GPR32:$Rm))),
543 (SMADDLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
544 def : Pat<(i64 (mul (zext GPR32:$Rn), (zext GPR32:$Rm))),
545 (UMADDLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
547 def : Pat<(i64 (ineg (mul (sext GPR32:$Rn), (sext GPR32:$Rm)))),
548 (SMSUBLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
549 def : Pat<(i64 (ineg (mul (zext GPR32:$Rn), (zext GPR32:$Rm)))),
550 (UMSUBLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
551 } // AddedComplexity = 5
553 def : MulAccumWAlias<"mul", MADDWrrr>;
554 def : MulAccumXAlias<"mul", MADDXrrr>;
555 def : MulAccumWAlias<"mneg", MSUBWrrr>;
556 def : MulAccumXAlias<"mneg", MSUBXrrr>;
557 def : WideMulAccumAlias<"smull", SMADDLrrr>;
558 def : WideMulAccumAlias<"smnegl", SMSUBLrrr>;
559 def : WideMulAccumAlias<"umull", UMADDLrrr>;
560 def : WideMulAccumAlias<"umnegl", UMSUBLrrr>;
563 def SMULHrr : MulHi<0b010, "smulh", mulhs>;
564 def UMULHrr : MulHi<0b110, "umulh", mulhu>;
567 def CRC32Brr : BaseCRC32<0, 0b00, 0, GPR32, int_arm64_crc32b, "crc32b">;
568 def CRC32Hrr : BaseCRC32<0, 0b01, 0, GPR32, int_arm64_crc32h, "crc32h">;
569 def CRC32Wrr : BaseCRC32<0, 0b10, 0, GPR32, int_arm64_crc32w, "crc32w">;
570 def CRC32Xrr : BaseCRC32<1, 0b11, 0, GPR64, int_arm64_crc32x, "crc32x">;
572 def CRC32CBrr : BaseCRC32<0, 0b00, 1, GPR32, int_arm64_crc32cb, "crc32cb">;
573 def CRC32CHrr : BaseCRC32<0, 0b01, 1, GPR32, int_arm64_crc32ch, "crc32ch">;
574 def CRC32CWrr : BaseCRC32<0, 0b10, 1, GPR32, int_arm64_crc32cw, "crc32cw">;
575 def CRC32CXrr : BaseCRC32<1, 0b11, 1, GPR64, int_arm64_crc32cx, "crc32cx">;
578 //===----------------------------------------------------------------------===//
579 // Logical instructions.
580 //===----------------------------------------------------------------------===//
583 defm ANDS : LogicalImmS<0b11, "ands", ARM64and_flag>;
584 defm AND : LogicalImm<0b00, "and", and>;
585 defm EOR : LogicalImm<0b10, "eor", xor>;
586 defm ORR : LogicalImm<0b01, "orr", or>;
588 def : InstAlias<"mov $dst, $imm", (ORRWri GPR32sp:$dst, WZR,
589 logical_imm32:$imm)>;
590 def : InstAlias<"mov $dst, $imm", (ORRXri GPR64sp:$dst, XZR,
591 logical_imm64:$imm)>;
595 defm ANDS : LogicalRegS<0b11, 0, "ands">;
596 defm BICS : LogicalRegS<0b11, 1, "bics">;
597 defm AND : LogicalReg<0b00, 0, "and", and>;
598 defm BIC : LogicalReg<0b00, 1, "bic",
599 BinOpFrag<(and node:$LHS, (not node:$RHS))>>;
600 defm EON : LogicalReg<0b10, 1, "eon",
601 BinOpFrag<(xor node:$LHS, (not node:$RHS))>>;
602 defm EOR : LogicalReg<0b10, 0, "eor", xor>;
603 defm ORN : LogicalReg<0b01, 1, "orn",
604 BinOpFrag<(or node:$LHS, (not node:$RHS))>>;
605 defm ORR : LogicalReg<0b01, 0, "orr", or>;
607 def : InstAlias<"tst $src1, $src2",
608 (ANDSWri WZR, GPR32:$src1, logical_imm32:$src2)>;
609 def : InstAlias<"tst $src1, $src2",
610 (ANDSXri XZR, GPR64:$src1, logical_imm64:$src2)>;
612 def : InstAlias<"tst $src1, $src2",
613 (ANDSWrs WZR, GPR32:$src1, GPR32:$src2, 0)>;
614 def : InstAlias<"tst $src1, $src2",
615 (ANDSXrs XZR, GPR64:$src1, GPR64:$src2, 0)>;
617 def : InstAlias<"tst $src1, $src2, $sh",
618 (ANDSWrs WZR, GPR32:$src1, GPR32:$src2, logical_shift:$sh)>;
619 def : InstAlias<"tst $src1, $src2, $sh",
620 (ANDSXrs XZR, GPR64:$src1, GPR64:$src2, logical_shift:$sh)>;
622 def : InstAlias<"mvn $Wd, $Wm",
623 (ORNWrs GPR32:$Wd, WZR, GPR32:$Wm, 0)>;
624 def : InstAlias<"mvn $Xd, $Xm",
625 (ORNXrs GPR64:$Xd, XZR, GPR64:$Xm, 0)>;
627 def : InstAlias<"mvn $Wd, $Wm, $sh",
628 (ORNWrs GPR32:$Wd, WZR, GPR32:$Wm, logical_shift:$sh)>;
629 def : InstAlias<"mvn $Xd, $Xm, $sh",
630 (ORNXrs GPR64:$Xd, XZR, GPR64:$Xm, logical_shift:$sh)>;
632 def : Pat<(not GPR32:$Wm), (ORNWrr WZR, GPR32:$Wm)>;
633 def : Pat<(not GPR64:$Xm), (ORNXrr XZR, GPR64:$Xm)>;
636 //===----------------------------------------------------------------------===//
637 // One operand data processing instructions.
638 //===----------------------------------------------------------------------===//
640 defm CLS : OneOperandData<0b101, "cls">;
641 defm CLZ : OneOperandData<0b100, "clz", ctlz>;
642 defm RBIT : OneOperandData<0b000, "rbit">;
643 def REV16Wr : OneWRegData<0b001, "rev16",
644 UnOpFrag<(rotr (bswap node:$LHS), (i64 16))>>;
645 def REV16Xr : OneXRegData<0b001, "rev16",
646 UnOpFrag<(rotr (bswap node:$LHS), (i64 16))>>;
648 def : Pat<(cttz GPR32:$Rn),
649 (CLZWr (RBITWr GPR32:$Rn))>;
650 def : Pat<(cttz GPR64:$Rn),
651 (CLZXr (RBITXr GPR64:$Rn))>;
653 // Unlike the other one operand instructions, the instructions with the "rev"
654 // mnemonic do *not* just different in the size bit, but actually use different
655 // opcode bits for the different sizes.
656 def REVWr : OneWRegData<0b010, "rev", bswap>;
657 def REVXr : OneXRegData<0b011, "rev", bswap>;
658 def REV32Xr : OneXRegData<0b010, "rev32",
659 UnOpFrag<(rotr (bswap node:$LHS), (i64 32))>>;
661 //===----------------------------------------------------------------------===//
662 // Bitfield immediate extraction instruction.
663 //===----------------------------------------------------------------------===//
664 let neverHasSideEffects = 1 in
665 defm EXTR : ExtractImm<"extr">;
666 def : InstAlias<"ror $dst, $src, $shift",
667 (EXTRWrri GPR32:$dst, GPR32:$src, GPR32:$src, imm0_31:$shift)>;
668 def : InstAlias<"ror $dst, $src, $shift",
669 (EXTRXrri GPR64:$dst, GPR64:$src, GPR64:$src, imm0_63:$shift)>;
671 def : Pat<(rotr GPR32:$Rn, (i64 imm0_31:$imm)),
672 (EXTRWrri GPR32:$Rn, GPR32:$Rn, imm0_31:$imm)>;
673 def : Pat<(rotr GPR64:$Rn, (i64 imm0_63:$imm)),
674 (EXTRXrri GPR64:$Rn, GPR64:$Rn, imm0_63:$imm)>;
676 //===----------------------------------------------------------------------===//
677 // Other bitfield immediate instructions.
678 //===----------------------------------------------------------------------===//
679 let neverHasSideEffects = 1 in {
680 defm BFM : BitfieldImmWith2RegArgs<0b01, "bfm">;
681 defm SBFM : BitfieldImm<0b00, "sbfm">;
682 defm UBFM : BitfieldImm<0b10, "ubfm">;
685 def i32shift_a : Operand<i64>, SDNodeXForm<imm, [{
686 uint64_t enc = (32 - N->getZExtValue()) & 0x1f;
687 return CurDAG->getTargetConstant(enc, MVT::i64);
690 def i32shift_b : Operand<i64>, SDNodeXForm<imm, [{
691 uint64_t enc = 31 - N->getZExtValue();
692 return CurDAG->getTargetConstant(enc, MVT::i64);
695 // min(7, 31 - shift_amt)
696 def i32shift_sext_i8 : Operand<i64>, SDNodeXForm<imm, [{
697 uint64_t enc = 31 - N->getZExtValue();
698 enc = enc > 7 ? 7 : enc;
699 return CurDAG->getTargetConstant(enc, MVT::i64);
702 // min(15, 31 - shift_amt)
703 def i32shift_sext_i16 : Operand<i64>, SDNodeXForm<imm, [{
704 uint64_t enc = 31 - N->getZExtValue();
705 enc = enc > 15 ? 15 : enc;
706 return CurDAG->getTargetConstant(enc, MVT::i64);
709 def i64shift_a : Operand<i64>, SDNodeXForm<imm, [{
710 uint64_t enc = (64 - N->getZExtValue()) & 0x3f;
711 return CurDAG->getTargetConstant(enc, MVT::i64);
714 def i64shift_b : Operand<i64>, SDNodeXForm<imm, [{
715 uint64_t enc = 63 - N->getZExtValue();
716 return CurDAG->getTargetConstant(enc, MVT::i64);
719 // min(7, 63 - shift_amt)
720 def i64shift_sext_i8 : Operand<i64>, SDNodeXForm<imm, [{
721 uint64_t enc = 63 - N->getZExtValue();
722 enc = enc > 7 ? 7 : enc;
723 return CurDAG->getTargetConstant(enc, MVT::i64);
726 // min(15, 63 - shift_amt)
727 def i64shift_sext_i16 : Operand<i64>, SDNodeXForm<imm, [{
728 uint64_t enc = 63 - N->getZExtValue();
729 enc = enc > 15 ? 15 : enc;
730 return CurDAG->getTargetConstant(enc, MVT::i64);
733 // min(31, 63 - shift_amt)
734 def i64shift_sext_i32 : Operand<i64>, SDNodeXForm<imm, [{
735 uint64_t enc = 63 - N->getZExtValue();
736 enc = enc > 31 ? 31 : enc;
737 return CurDAG->getTargetConstant(enc, MVT::i64);
740 def : Pat<(shl GPR32:$Rn, (i64 imm0_31:$imm)),
741 (UBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
742 (i64 (i32shift_b imm0_31:$imm)))>;
743 def : Pat<(shl GPR64:$Rn, (i64 imm0_63:$imm)),
744 (UBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
745 (i64 (i64shift_b imm0_63:$imm)))>;
747 let AddedComplexity = 10 in {
748 def : Pat<(sra GPR32:$Rn, (i64 imm0_31:$imm)),
749 (SBFMWri GPR32:$Rn, imm0_31:$imm, 31)>;
750 def : Pat<(sra GPR64:$Rn, (i64 imm0_63:$imm)),
751 (SBFMXri GPR64:$Rn, imm0_63:$imm, 63)>;
754 def : InstAlias<"asr $dst, $src, $shift",
755 (SBFMWri GPR32:$dst, GPR32:$src, imm0_31:$shift, 31)>;
756 def : InstAlias<"asr $dst, $src, $shift",
757 (SBFMXri GPR64:$dst, GPR64:$src, imm0_63:$shift, 63)>;
758 def : InstAlias<"sxtb $dst, $src", (SBFMWri GPR32:$dst, GPR32:$src, 0, 7)>;
759 def : InstAlias<"sxtb $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 7)>;
760 def : InstAlias<"sxth $dst, $src", (SBFMWri GPR32:$dst, GPR32:$src, 0, 15)>;
761 def : InstAlias<"sxth $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 15)>;
762 def : InstAlias<"sxtw $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 31)>;
764 def : Pat<(srl GPR32:$Rn, (i64 imm0_31:$imm)),
765 (UBFMWri GPR32:$Rn, imm0_31:$imm, 31)>;
766 def : Pat<(srl GPR64:$Rn, (i64 imm0_63:$imm)),
767 (UBFMXri GPR64:$Rn, imm0_63:$imm, 63)>;
769 def : InstAlias<"lsr $dst, $src, $shift",
770 (UBFMWri GPR32:$dst, GPR32:$src, imm0_31:$shift, 31)>;
771 def : InstAlias<"lsr $dst, $src, $shift",
772 (UBFMXri GPR64:$dst, GPR64:$src, imm0_63:$shift, 63)>;
773 def : InstAlias<"uxtb $dst, $src", (UBFMWri GPR32:$dst, GPR32:$src, 0, 7)>;
774 def : InstAlias<"uxtb $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 7)>;
775 def : InstAlias<"uxth $dst, $src", (UBFMWri GPR32:$dst, GPR32:$src, 0, 15)>;
776 def : InstAlias<"uxth $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 15)>;
777 def : InstAlias<"uxtw $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 31)>;
779 //===----------------------------------------------------------------------===//
780 // Conditionally set flags instructions.
781 //===----------------------------------------------------------------------===//
782 defm CCMN : CondSetFlagsImm<0, "ccmn">;
783 defm CCMP : CondSetFlagsImm<1, "ccmp">;
785 defm CCMN : CondSetFlagsReg<0, "ccmn">;
786 defm CCMP : CondSetFlagsReg<1, "ccmp">;
788 //===----------------------------------------------------------------------===//
789 // Conditional select instructions.
790 //===----------------------------------------------------------------------===//
791 defm CSEL : CondSelect<0, 0b00, "csel">;
793 def inc : PatFrag<(ops node:$in), (add node:$in, 1)>;
794 defm CSINC : CondSelectOp<0, 0b01, "csinc", inc>;
795 defm CSINV : CondSelectOp<1, 0b00, "csinv", not>;
796 defm CSNEG : CondSelectOp<1, 0b01, "csneg", ineg>;
798 def : Pat<(ARM64csinv GPR32:$tval, GPR32:$fval, (i32 imm:$cc), CPSR),
799 (CSINVWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
800 def : Pat<(ARM64csinv GPR64:$tval, GPR64:$fval, (i32 imm:$cc), CPSR),
801 (CSINVXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
802 def : Pat<(ARM64csneg GPR32:$tval, GPR32:$fval, (i32 imm:$cc), CPSR),
803 (CSNEGWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
804 def : Pat<(ARM64csneg GPR64:$tval, GPR64:$fval, (i32 imm:$cc), CPSR),
805 (CSNEGXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
806 def : Pat<(ARM64csinc GPR32:$tval, GPR32:$fval, (i32 imm:$cc), CPSR),
807 (CSINCWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
808 def : Pat<(ARM64csinc GPR64:$tval, GPR64:$fval, (i32 imm:$cc), CPSR),
809 (CSINCXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
811 def : Pat<(ARM64csel (i32 0), (i32 1), (i32 imm:$cc), CPSR),
812 (CSINCWr WZR, WZR, (i32 imm:$cc))>;
813 def : Pat<(ARM64csel (i64 0), (i64 1), (i32 imm:$cc), CPSR),
814 (CSINCXr XZR, XZR, (i32 imm:$cc))>;
815 def : Pat<(ARM64csel (i32 0), (i32 -1), (i32 imm:$cc), CPSR),
816 (CSINVWr WZR, WZR, (i32 imm:$cc))>;
817 def : Pat<(ARM64csel (i64 0), (i64 -1), (i32 imm:$cc), CPSR),
818 (CSINVXr XZR, XZR, (i32 imm:$cc))>;
820 // The inverse of the condition code from the alias instruction is what is used
821 // in the aliased instruction. The parser all ready inverts the condition code
822 // for these aliases.
823 // FIXME: Is this the correct way to handle these aliases?
824 def : InstAlias<"cset $dst, $cc", (CSINCWr GPR32:$dst, WZR, WZR, ccode:$cc)>;
825 def : InstAlias<"cset $dst, $cc", (CSINCXr GPR64:$dst, XZR, XZR, ccode:$cc)>;
827 def : InstAlias<"csetm $dst, $cc", (CSINVWr GPR32:$dst, WZR, WZR, ccode:$cc)>;
828 def : InstAlias<"csetm $dst, $cc", (CSINVXr GPR64:$dst, XZR, XZR, ccode:$cc)>;
830 def : InstAlias<"cinc $dst, $src, $cc",
831 (CSINCWr GPR32:$dst, GPR32:$src, GPR32:$src, ccode:$cc)>;
832 def : InstAlias<"cinc $dst, $src, $cc",
833 (CSINCXr GPR64:$dst, GPR64:$src, GPR64:$src, ccode:$cc)>;
835 def : InstAlias<"cinv $dst, $src, $cc",
836 (CSINVWr GPR32:$dst, GPR32:$src, GPR32:$src, ccode:$cc)>;
837 def : InstAlias<"cinv $dst, $src, $cc",
838 (CSINVXr GPR64:$dst, GPR64:$src, GPR64:$src, ccode:$cc)>;
840 def : InstAlias<"cneg $dst, $src, $cc",
841 (CSNEGWr GPR32:$dst, GPR32:$src, GPR32:$src, ccode:$cc)>;
842 def : InstAlias<"cneg $dst, $src, $cc",
843 (CSNEGXr GPR64:$dst, GPR64:$src, GPR64:$src, ccode:$cc)>;
845 //===----------------------------------------------------------------------===//
846 // PC-relative instructions.
847 //===----------------------------------------------------------------------===//
848 let isReMaterializable = 1 in {
849 let neverHasSideEffects = 1, mayStore = 0, mayLoad = 0 in {
850 def ADR : ADRI<0, "adr", adrlabel, []>;
851 } // neverHasSideEffects = 1
853 def ADRP : ADRI<1, "adrp", adrplabel,
854 [(set GPR64:$Xd, (ARM64adrp tglobaladdr:$label))]>;
855 } // isReMaterializable = 1
857 // page address of a constant pool entry, block address
858 def : Pat<(ARM64adrp tconstpool:$cp), (ADRP tconstpool:$cp)>;
859 def : Pat<(ARM64adrp tblockaddress:$cp), (ADRP tblockaddress:$cp)>;
861 //===----------------------------------------------------------------------===//
862 // Unconditional branch (register) instructions.
863 //===----------------------------------------------------------------------===//
865 let isReturn = 1, isTerminator = 1, isBarrier = 1 in {
866 def RET : BranchReg<0b0010, "ret", []>;
867 def DRPS : SpecialReturn<0b0101, "drps">;
868 def ERET : SpecialReturn<0b0100, "eret">;
869 } // isReturn = 1, isTerminator = 1, isBarrier = 1
871 // Default to the LR register.
872 def : InstAlias<"ret", (RET LR)>;
874 let isCall = 1, Defs = [LR], Uses = [SP] in {
875 def BLR : BranchReg<0b0001, "blr", [(ARM64call GPR64:$Rn)]>;
878 let isBranch = 1, isTerminator = 1, isBarrier = 1, isIndirectBranch = 1 in {
879 def BR : BranchReg<0b0000, "br", [(brind GPR64:$Rn)]>;
880 } // isBranch, isTerminator, isBarrier, isIndirectBranch
882 // Create a separate pseudo-instruction for codegen to use so that we don't
883 // flag lr as used in every function. It'll be restored before the RET by the
884 // epilogue if it's legitimately used.
885 def RET_ReallyLR : Pseudo<(outs), (ins), [(ARM64retflag)]> {
886 let isTerminator = 1;
891 // This is a directive-like pseudo-instruction. The purpose is to insert an
892 // R_AARCH64_TLSDESC_CALL relocation at the offset of the following instruction
893 // (which in the usual case is a BLR).
894 let hasSideEffects = 1 in
895 def TLSDESCCALL : Pseudo<(outs), (ins i64imm:$sym), []> {
896 let AsmString = ".tlsdesccall $sym";
899 // Pseudo-instruction representing a BLR with attached TLSDESC relocation. It
900 // gets expanded to two MCInsts during lowering.
901 let isCall = 1, Defs = [LR] in
903 : Pseudo<(outs), (ins GPR64:$dest, i64imm:$sym),
904 [(ARM64tlsdesc_call GPR64:$dest, tglobaltlsaddr:$sym)]>;
906 def : Pat<(ARM64tlsdesc_call GPR64:$dest, texternalsym:$sym),
907 (TLSDESC_BLR GPR64:$dest, texternalsym:$sym)>;
908 //===----------------------------------------------------------------------===//
909 // Conditional branch (immediate) instruction.
910 //===----------------------------------------------------------------------===//
911 def Bcc : BranchCond;
913 //===----------------------------------------------------------------------===//
914 // Compare-and-branch instructions.
915 //===----------------------------------------------------------------------===//
916 defm CBZ : CmpBranch<0, "cbz", ARM64cbz>;
917 defm CBNZ : CmpBranch<1, "cbnz", ARM64cbnz>;
919 //===----------------------------------------------------------------------===//
920 // Test-bit-and-branch instructions.
921 //===----------------------------------------------------------------------===//
922 def TBZ : TestBranch<0, "tbz", ARM64tbz>;
923 def TBNZ : TestBranch<1, "tbnz", ARM64tbnz>;
925 //===----------------------------------------------------------------------===//
926 // Unconditional branch (immediate) instructions.
927 //===----------------------------------------------------------------------===//
928 let isBranch = 1, isTerminator = 1, isBarrier = 1 in {
929 def B : BranchImm<0, "b", [(br bb:$addr)]>;
930 } // isBranch, isTerminator, isBarrier
932 let isCall = 1, Defs = [LR], Uses = [SP] in {
933 def BL : CallImm<1, "bl", [(ARM64call tglobaladdr:$addr)]>;
935 def : Pat<(ARM64call texternalsym:$func), (BL texternalsym:$func)>;
937 //===----------------------------------------------------------------------===//
938 // Exception generation instructions.
939 //===----------------------------------------------------------------------===//
940 def BRK : ExceptionGeneration<0b001, 0b00, "brk">;
941 def DCPS1 : ExceptionGeneration<0b101, 0b01, "dcps1">;
942 def DCPS2 : ExceptionGeneration<0b101, 0b10, "dcps2">;
943 def DCPS3 : ExceptionGeneration<0b101, 0b11, "dcps3">;
944 def HLT : ExceptionGeneration<0b010, 0b00, "hlt">;
945 def HVC : ExceptionGeneration<0b000, 0b10, "hvc">;
946 def SMC : ExceptionGeneration<0b000, 0b11, "smc">;
947 def SVC : ExceptionGeneration<0b000, 0b01, "svc">;
949 // DCPSn defaults to an immediate operand of zero if unspecified.
950 def : InstAlias<"dcps1", (DCPS1 0)>;
951 def : InstAlias<"dcps2", (DCPS2 0)>;
952 def : InstAlias<"dcps3", (DCPS3 0)>;
954 //===----------------------------------------------------------------------===//
955 // Load instructions.
956 //===----------------------------------------------------------------------===//
958 // Pair (indexed, offset)
959 def LDPWi : LoadPairOffset<0b00, 0, GPR32, am_indexed32simm7, "ldp">;
960 def LDPXi : LoadPairOffset<0b10, 0, GPR64, am_indexed64simm7, "ldp">;
961 def LDPSi : LoadPairOffset<0b00, 1, FPR32, am_indexed32simm7, "ldp">;
962 def LDPDi : LoadPairOffset<0b01, 1, FPR64, am_indexed64simm7, "ldp">;
963 def LDPQi : LoadPairOffset<0b10, 1, FPR128, am_indexed128simm7, "ldp">;
965 def LDPSWi : LoadPairOffset<0b01, 0, GPR64, am_indexed32simm7, "ldpsw">;
967 // Pair (pre-indexed)
968 def LDPWpre : LoadPairPreIdx<0b00, 0, GPR32, am_indexed32simm7, "ldp">;
969 def LDPXpre : LoadPairPreIdx<0b10, 0, GPR64, am_indexed64simm7, "ldp">;
970 def LDPSpre : LoadPairPreIdx<0b00, 1, FPR32, am_indexed32simm7, "ldp">;
971 def LDPDpre : LoadPairPreIdx<0b01, 1, FPR64, am_indexed64simm7, "ldp">;
972 def LDPQpre : LoadPairPreIdx<0b10, 1, FPR128, am_indexed128simm7, "ldp">;
974 def LDPSWpre : LoadPairPreIdx<0b01, 0, GPR64, am_indexed32simm7, "ldpsw">;
976 // Pair (post-indexed)
977 def LDPWpost : LoadPairPostIdx<0b00, 0, GPR32, simm7s4, "ldp">;
978 def LDPXpost : LoadPairPostIdx<0b10, 0, GPR64, simm7s8, "ldp">;
979 def LDPSpost : LoadPairPostIdx<0b00, 1, FPR32, simm7s4, "ldp">;
980 def LDPDpost : LoadPairPostIdx<0b01, 1, FPR64, simm7s8, "ldp">;
981 def LDPQpost : LoadPairPostIdx<0b10, 1, FPR128, simm7s16, "ldp">;
983 def LDPSWpost : LoadPairPostIdx<0b01, 0, GPR64, simm7s4, "ldpsw">;
986 // Pair (no allocate)
987 def LDNPWi : LoadPairNoAlloc<0b00, 0, GPR32, am_indexed32simm7, "ldnp">;
988 def LDNPXi : LoadPairNoAlloc<0b10, 0, GPR64, am_indexed64simm7, "ldnp">;
989 def LDNPSi : LoadPairNoAlloc<0b00, 1, FPR32, am_indexed32simm7, "ldnp">;
990 def LDNPDi : LoadPairNoAlloc<0b01, 1, FPR64, am_indexed64simm7, "ldnp">;
991 def LDNPQi : LoadPairNoAlloc<0b10, 1, FPR128, am_indexed128simm7, "ldnp">;
997 let AddedComplexity = 10 in {
999 def LDRBBro : Load8RO<0b00, 0, 0b01, GPR32, "ldrb",
1000 [(set GPR32:$Rt, (zextloadi8 ro_indexed8:$addr))]>;
1001 def LDRHHro : Load16RO<0b01, 0, 0b01, GPR32, "ldrh",
1002 [(set GPR32:$Rt, (zextloadi16 ro_indexed16:$addr))]>;
1003 def LDRWro : Load32RO<0b10, 0, 0b01, GPR32, "ldr",
1004 [(set GPR32:$Rt, (load ro_indexed32:$addr))]>;
1005 def LDRXro : Load64RO<0b11, 0, 0b01, GPR64, "ldr",
1006 [(set GPR64:$Rt, (load ro_indexed64:$addr))]>;
1009 def LDRBro : Load8RO<0b00, 1, 0b01, FPR8, "ldr",
1010 [(set FPR8:$Rt, (load ro_indexed8:$addr))]>;
1011 def LDRHro : Load16RO<0b01, 1, 0b01, FPR16, "ldr",
1012 [(set FPR16:$Rt, (load ro_indexed16:$addr))]>;
1013 def LDRSro : Load32RO<0b10, 1, 0b01, FPR32, "ldr",
1014 [(set (f32 FPR32:$Rt), (load ro_indexed32:$addr))]>;
1015 def LDRDro : Load64RO<0b11, 1, 0b01, FPR64, "ldr",
1016 [(set (f64 FPR64:$Rt), (load ro_indexed64:$addr))]>;
1017 def LDRQro : Load128RO<0b00, 1, 0b11, FPR128, "ldr", []> {
1021 // For regular load, we do not have any alignment requirement.
1022 // Thus, it is safe to directly map the vector loads with interesting
1023 // addressing modes.
1024 // FIXME: We could do the same for bitconvert to floating point vectors.
1025 def : Pat <(v8i8 (scalar_to_vector (i32 (extloadi8 ro_indexed8:$addr)))),
1026 (INSERT_SUBREG (v8i8 (IMPLICIT_DEF)),
1027 (LDRBro ro_indexed8:$addr), bsub)>;
1028 def : Pat <(v16i8 (scalar_to_vector (i32 (extloadi8 ro_indexed8:$addr)))),
1029 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
1030 (LDRBro ro_indexed8:$addr), bsub)>;
1031 def : Pat <(v4i16 (scalar_to_vector (i32 (extloadi16 ro_indexed16:$addr)))),
1032 (INSERT_SUBREG (v4i16 (IMPLICIT_DEF)),
1033 (LDRHro ro_indexed16:$addr), hsub)>;
1034 def : Pat <(v8i16 (scalar_to_vector (i32 (extloadi16 ro_indexed16:$addr)))),
1035 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
1036 (LDRHro ro_indexed16:$addr), hsub)>;
1037 def : Pat <(v2i32 (scalar_to_vector (i32 (load ro_indexed32:$addr)))),
1038 (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)),
1039 (LDRSro ro_indexed32:$addr), ssub)>;
1040 def : Pat <(v4i32 (scalar_to_vector (i32 (load ro_indexed32:$addr)))),
1041 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
1042 (LDRSro ro_indexed32:$addr), ssub)>;
1043 def : Pat <(v1i64 (scalar_to_vector (i64 (load ro_indexed64:$addr)))),
1044 (LDRDro ro_indexed64:$addr)>;
1045 def : Pat <(v2i64 (scalar_to_vector (i64 (load ro_indexed64:$addr)))),
1046 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
1047 (LDRDro ro_indexed64:$addr), dsub)>;
1049 // Match all load 64 bits width whose type is compatible with FPR64
1050 def : Pat<(v2f32 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1051 def : Pat<(v1f64 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1052 def : Pat<(v8i8 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1053 def : Pat<(v4i16 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1054 def : Pat<(v2i32 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1055 def : Pat<(v1i64 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1057 // Match all load 128 bits width whose type is compatible with FPR128
1058 def : Pat<(v4f32 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1059 def : Pat<(v2f64 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1060 def : Pat<(v16i8 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1061 def : Pat<(v8i16 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1062 def : Pat<(v4i32 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1063 def : Pat<(v2i64 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1064 def : Pat<(f128 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1066 // Load sign-extended half-word
1067 def LDRSHWro : Load16RO<0b01, 0, 0b11, GPR32, "ldrsh",
1068 [(set GPR32:$Rt, (sextloadi16 ro_indexed16:$addr))]>;
1069 def LDRSHXro : Load16RO<0b01, 0, 0b10, GPR64, "ldrsh",
1070 [(set GPR64:$Rt, (sextloadi16 ro_indexed16:$addr))]>;
1072 // Load sign-extended byte
1073 def LDRSBWro : Load8RO<0b00, 0, 0b11, GPR32, "ldrsb",
1074 [(set GPR32:$Rt, (sextloadi8 ro_indexed8:$addr))]>;
1075 def LDRSBXro : Load8RO<0b00, 0, 0b10, GPR64, "ldrsb",
1076 [(set GPR64:$Rt, (sextloadi8 ro_indexed8:$addr))]>;
1078 // Load sign-extended word
1079 def LDRSWro : Load32RO<0b10, 0, 0b10, GPR64, "ldrsw",
1080 [(set GPR64:$Rt, (sextloadi32 ro_indexed32:$addr))]>;
1083 def PRFMro : PrefetchRO<0b11, 0, 0b10, "prfm",
1084 [(ARM64Prefetch imm:$Rt, ro_indexed64:$addr)]>;
1087 def : Pat<(i64 (zextloadi8 ro_indexed8:$addr)),
1088 (SUBREG_TO_REG (i64 0), (LDRBBro ro_indexed8:$addr), sub_32)>;
1089 def : Pat<(i64 (zextloadi16 ro_indexed16:$addr)),
1090 (SUBREG_TO_REG (i64 0), (LDRHHro ro_indexed16:$addr), sub_32)>;
1092 // zextloadi1 -> zextloadi8
1093 def : Pat<(i32 (zextloadi1 ro_indexed8:$addr)), (LDRBBro ro_indexed8:$addr)>;
1094 def : Pat<(i64 (zextloadi1 ro_indexed8:$addr)),
1095 (SUBREG_TO_REG (i64 0), (LDRBBro ro_indexed8:$addr), sub_32)>;
1097 // extload -> zextload
1098 def : Pat<(i32 (extloadi16 ro_indexed16:$addr)), (LDRHHro ro_indexed16:$addr)>;
1099 def : Pat<(i32 (extloadi8 ro_indexed8:$addr)), (LDRBBro ro_indexed8:$addr)>;
1100 def : Pat<(i32 (extloadi1 ro_indexed8:$addr)), (LDRBBro ro_indexed8:$addr)>;
1101 def : Pat<(i64 (extloadi32 ro_indexed32:$addr)),
1102 (SUBREG_TO_REG (i64 0), (LDRWro ro_indexed32:$addr), sub_32)>;
1103 def : Pat<(i64 (extloadi16 ro_indexed16:$addr)),
1104 (SUBREG_TO_REG (i64 0), (LDRHHro ro_indexed16:$addr), sub_32)>;
1105 def : Pat<(i64 (extloadi8 ro_indexed8:$addr)),
1106 (SUBREG_TO_REG (i64 0), (LDRBBro ro_indexed8:$addr), sub_32)>;
1107 def : Pat<(i64 (extloadi1 ro_indexed8:$addr)),
1108 (SUBREG_TO_REG (i64 0), (LDRBBro ro_indexed8:$addr), sub_32)>;
1110 } // AddedComplexity = 10
1113 // (unsigned immediate)
1115 def LDRXui : LoadUI<0b11, 0, 0b01, GPR64, am_indexed64, "ldr",
1116 [(set GPR64:$Rt, (load am_indexed64:$addr))]>;
1117 def LDRWui : LoadUI<0b10, 0, 0b01, GPR32, am_indexed32, "ldr",
1118 [(set GPR32:$Rt, (load am_indexed32:$addr))]>;
1119 def LDRBui : LoadUI<0b00, 1, 0b01, FPR8, am_indexed8, "ldr",
1120 [(set FPR8:$Rt, (load am_indexed8:$addr))]>;
1121 def LDRHui : LoadUI<0b01, 1, 0b01, FPR16, am_indexed16, "ldr",
1122 [(set FPR16:$Rt, (load am_indexed16:$addr))]>;
1123 def LDRSui : LoadUI<0b10, 1, 0b01, FPR32, am_indexed32, "ldr",
1124 [(set (f32 FPR32:$Rt), (load am_indexed32:$addr))]>;
1125 def LDRDui : LoadUI<0b11, 1, 0b01, FPR64, am_indexed64, "ldr",
1126 [(set (f64 FPR64:$Rt), (load am_indexed64:$addr))]>;
1127 def LDRQui : LoadUI<0b00, 1, 0b11, FPR128, am_indexed128, "ldr",
1128 [(set (f128 FPR128:$Rt), (load am_indexed128:$addr))]>;
1130 // For regular load, we do not have any alignment requirement.
1131 // Thus, it is safe to directly map the vector loads with interesting
1132 // addressing modes.
1133 // FIXME: We could do the same for bitconvert to floating point vectors.
1134 def : Pat <(v8i8 (scalar_to_vector (i32 (extloadi8 am_indexed8:$addr)))),
1135 (INSERT_SUBREG (v8i8 (IMPLICIT_DEF)),
1136 (LDRBui am_indexed8:$addr), bsub)>;
1137 def : Pat <(v16i8 (scalar_to_vector (i32 (extloadi8 am_indexed8:$addr)))),
1138 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
1139 (LDRBui am_indexed8:$addr), bsub)>;
1140 def : Pat <(v4i16 (scalar_to_vector (i32 (extloadi16 am_indexed16:$addr)))),
1141 (INSERT_SUBREG (v4i16 (IMPLICIT_DEF)),
1142 (LDRHui am_indexed16:$addr), hsub)>;
1143 def : Pat <(v8i16 (scalar_to_vector (i32 (extloadi16 am_indexed16:$addr)))),
1144 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
1145 (LDRHui am_indexed16:$addr), hsub)>;
1146 def : Pat <(v2i32 (scalar_to_vector (i32 (load am_indexed32:$addr)))),
1147 (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)),
1148 (LDRSui am_indexed32:$addr), ssub)>;
1149 def : Pat <(v4i32 (scalar_to_vector (i32 (load am_indexed32:$addr)))),
1150 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
1151 (LDRSui am_indexed32:$addr), ssub)>;
1152 def : Pat <(v1i64 (scalar_to_vector (i64 (load am_indexed64:$addr)))),
1153 (LDRDui am_indexed64:$addr)>;
1154 def : Pat <(v2i64 (scalar_to_vector (i64 (load am_indexed64:$addr)))),
1155 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
1156 (LDRDui am_indexed64:$addr), dsub)>;
1158 // Match all load 64 bits width whose type is compatible with FPR64
1159 def : Pat<(v2f32 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1160 def : Pat<(v1f64 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1161 def : Pat<(v8i8 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1162 def : Pat<(v4i16 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1163 def : Pat<(v2i32 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1164 def : Pat<(v1i64 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1166 // Match all load 128 bits width whose type is compatible with FPR128
1167 def : Pat<(v4f32 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1168 def : Pat<(v2f64 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1169 def : Pat<(v16i8 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1170 def : Pat<(v8i16 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1171 def : Pat<(v4i32 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1172 def : Pat<(v2i64 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1173 def : Pat<(f128 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1175 def LDRHHui : LoadUI<0b01, 0, 0b01, GPR32, am_indexed16, "ldrh",
1176 [(set GPR32:$Rt, (zextloadi16 am_indexed16:$addr))]>;
1177 def LDRBBui : LoadUI<0b00, 0, 0b01, GPR32, am_indexed8, "ldrb",
1178 [(set GPR32:$Rt, (zextloadi8 am_indexed8:$addr))]>;
1180 def : Pat<(i64 (zextloadi8 am_indexed8:$addr)),
1181 (SUBREG_TO_REG (i64 0), (LDRBBui am_indexed8:$addr), sub_32)>;
1182 def : Pat<(i64 (zextloadi16 am_indexed16:$addr)),
1183 (SUBREG_TO_REG (i64 0), (LDRHHui am_indexed16:$addr), sub_32)>;
1185 // zextloadi1 -> zextloadi8
1186 def : Pat<(i32 (zextloadi1 am_indexed8:$addr)), (LDRBBui am_indexed8:$addr)>;
1187 def : Pat<(i64 (zextloadi1 am_indexed8:$addr)),
1188 (SUBREG_TO_REG (i64 0), (LDRBBui am_indexed8:$addr), sub_32)>;
1190 // extload -> zextload
1191 def : Pat<(i32 (extloadi16 am_indexed16:$addr)), (LDRHHui am_indexed16:$addr)>;
1192 def : Pat<(i32 (extloadi8 am_indexed8:$addr)), (LDRBBui am_indexed8:$addr)>;
1193 def : Pat<(i32 (extloadi1 am_indexed8:$addr)), (LDRBBui am_indexed8:$addr)>;
1194 def : Pat<(i64 (extloadi32 am_indexed32:$addr)),
1195 (SUBREG_TO_REG (i64 0), (LDRWui am_indexed32:$addr), sub_32)>;
1196 def : Pat<(i64 (extloadi16 am_indexed16:$addr)),
1197 (SUBREG_TO_REG (i64 0), (LDRHHui am_indexed16:$addr), sub_32)>;
1198 def : Pat<(i64 (extloadi8 am_indexed8:$addr)),
1199 (SUBREG_TO_REG (i64 0), (LDRBBui am_indexed8:$addr), sub_32)>;
1200 def : Pat<(i64 (extloadi1 am_indexed8:$addr)),
1201 (SUBREG_TO_REG (i64 0), (LDRBBui am_indexed8:$addr), sub_32)>;
1203 // load sign-extended half-word
1204 def LDRSHWui : LoadUI<0b01, 0, 0b11, GPR32, am_indexed16, "ldrsh",
1205 [(set GPR32:$Rt, (sextloadi16 am_indexed16:$addr))]>;
1206 def LDRSHXui : LoadUI<0b01, 0, 0b10, GPR64, am_indexed16, "ldrsh",
1207 [(set GPR64:$Rt, (sextloadi16 am_indexed16:$addr))]>;
1209 // load sign-extended byte
1210 def LDRSBWui : LoadUI<0b00, 0, 0b11, GPR32, am_indexed8, "ldrsb",
1211 [(set GPR32:$Rt, (sextloadi8 am_indexed8:$addr))]>;
1212 def LDRSBXui : LoadUI<0b00, 0, 0b10, GPR64, am_indexed8, "ldrsb",
1213 [(set GPR64:$Rt, (sextloadi8 am_indexed8:$addr))]>;
1215 // load sign-extended word
1216 def LDRSWui : LoadUI<0b10, 0, 0b10, GPR64, am_indexed32, "ldrsw",
1217 [(set GPR64:$Rt, (sextloadi32 am_indexed32:$addr))]>;
1219 // load zero-extended word
1220 def : Pat<(i64 (zextloadi32 am_indexed32:$addr)),
1221 (SUBREG_TO_REG (i64 0), (LDRWui am_indexed32:$addr), sub_32)>;
1224 def PRFMui : PrefetchUI<0b11, 0, 0b10, "prfm",
1225 [(ARM64Prefetch imm:$Rt, am_indexed64:$addr)]>;
1229 def LDRWl : LoadLiteral<0b00, 0, GPR32, "ldr">;
1230 def LDRXl : LoadLiteral<0b01, 0, GPR64, "ldr">;
1231 def LDRSl : LoadLiteral<0b00, 1, FPR32, "ldr">;
1232 def LDRDl : LoadLiteral<0b01, 1, FPR64, "ldr">;
1233 def LDRQl : LoadLiteral<0b10, 1, FPR128, "ldr">;
1235 // load sign-extended word
1236 def LDRSWl : LoadLiteral<0b10, 0, GPR64, "ldrsw">;
1239 def PRFMl : PrefetchLiteral<0b11, 0, "prfm", []>;
1240 // [(ARM64Prefetch imm:$Rt, tglobaladdr:$label)]>;
1243 // (unscaled immediate)
1244 def LDURXi : LoadUnscaled<0b11, 0, 0b01, GPR64, am_unscaled64, "ldur",
1245 [(set GPR64:$Rt, (load am_unscaled64:$addr))]>;
1246 def LDURWi : LoadUnscaled<0b10, 0, 0b01, GPR32, am_unscaled32, "ldur",
1247 [(set GPR32:$Rt, (load am_unscaled32:$addr))]>;
1248 def LDURBi : LoadUnscaled<0b00, 1, 0b01, FPR8, am_unscaled8, "ldur",
1249 [(set FPR8:$Rt, (load am_unscaled8:$addr))]>;
1250 def LDURHi : LoadUnscaled<0b01, 1, 0b01, FPR16, am_unscaled16, "ldur",
1251 [(set FPR16:$Rt, (load am_unscaled16:$addr))]>;
1252 def LDURSi : LoadUnscaled<0b10, 1, 0b01, FPR32, am_unscaled32, "ldur",
1253 [(set (f32 FPR32:$Rt), (load am_unscaled32:$addr))]>;
1254 def LDURDi : LoadUnscaled<0b11, 1, 0b01, FPR64, am_unscaled64, "ldur",
1255 [(set (f64 FPR64:$Rt), (load am_unscaled64:$addr))]>;
1256 def LDURQi : LoadUnscaled<0b00, 1, 0b11, FPR128, am_unscaled128, "ldur",
1257 [(set (v2f64 FPR128:$Rt), (load am_unscaled128:$addr))]>;
1260 : LoadUnscaled<0b01, 0, 0b01, GPR32, am_unscaled16, "ldurh",
1261 [(set GPR32:$Rt, (zextloadi16 am_unscaled16:$addr))]>;
1263 : LoadUnscaled<0b00, 0, 0b01, GPR32, am_unscaled8, "ldurb",
1264 [(set GPR32:$Rt, (zextloadi8 am_unscaled8:$addr))]>;
1266 // Match all load 64 bits width whose type is compatible with FPR64
1267 def : Pat<(v2f32 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1268 def : Pat<(v1f64 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1269 def : Pat<(v8i8 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1270 def : Pat<(v4i16 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1271 def : Pat<(v2i32 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1272 def : Pat<(v1i64 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1274 // Match all load 128 bits width whose type is compatible with FPR128
1275 def : Pat<(v4f32 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1276 def : Pat<(v2f64 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1277 def : Pat<(v16i8 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1278 def : Pat<(v8i16 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1279 def : Pat<(v4i32 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1280 def : Pat<(v2i64 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1281 def : Pat<(f128 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1284 def : Pat<(i32 (extloadi16 am_unscaled16:$addr)), (LDURHHi am_unscaled16:$addr)>;
1285 def : Pat<(i32 (extloadi8 am_unscaled8:$addr)), (LDURBBi am_unscaled8:$addr)>;
1286 def : Pat<(i32 (extloadi1 am_unscaled8:$addr)), (LDURBBi am_unscaled8:$addr)>;
1287 def : Pat<(i64 (extloadi32 am_unscaled32:$addr)),
1288 (SUBREG_TO_REG (i64 0), (LDURWi am_unscaled32:$addr), sub_32)>;
1289 def : Pat<(i64 (extloadi16 am_unscaled16:$addr)),
1290 (SUBREG_TO_REG (i64 0), (LDURHHi am_unscaled16:$addr), sub_32)>;
1291 def : Pat<(i64 (extloadi8 am_unscaled8:$addr)),
1292 (SUBREG_TO_REG (i64 0), (LDURBBi am_unscaled8:$addr), sub_32)>;
1293 def : Pat<(i64 (extloadi1 am_unscaled8:$addr)),
1294 (SUBREG_TO_REG (i64 0), (LDURBBi am_unscaled8:$addr), sub_32)>;
1296 def : Pat<(i32 (zextloadi16 am_unscaled16:$addr)),
1297 (LDURHHi am_unscaled16:$addr)>;
1298 def : Pat<(i32 (zextloadi8 am_unscaled8:$addr)),
1299 (LDURBBi am_unscaled8:$addr)>;
1300 def : Pat<(i32 (zextloadi1 am_unscaled8:$addr)),
1301 (LDURBBi am_unscaled8:$addr)>;
1302 def : Pat<(i64 (zextloadi32 am_unscaled32:$addr)),
1303 (SUBREG_TO_REG (i64 0), (LDURWi am_unscaled32:$addr), sub_32)>;
1304 def : Pat<(i64 (zextloadi16 am_unscaled16:$addr)),
1305 (SUBREG_TO_REG (i64 0), (LDURHHi am_unscaled16:$addr), sub_32)>;
1306 def : Pat<(i64 (zextloadi8 am_unscaled8:$addr)),
1307 (SUBREG_TO_REG (i64 0), (LDURBBi am_unscaled8:$addr), sub_32)>;
1308 def : Pat<(i64 (zextloadi1 am_unscaled8:$addr)),
1309 (SUBREG_TO_REG (i64 0), (LDURBBi am_unscaled8:$addr), sub_32)>;
1313 // LDR mnemonics fall back to LDUR for negative or unaligned offsets.
1315 // Define new assembler match classes as we want to only match these when
1316 // the don't otherwise match the scaled addressing mode for LDR/STR. Don't
1317 // associate a DiagnosticType either, as we want the diagnostic for the
1318 // canonical form (the scaled operand) to take precedence.
1319 def MemoryUnscaledFB8Operand : AsmOperandClass {
1320 let Name = "MemoryUnscaledFB8";
1321 let RenderMethod = "addMemoryUnscaledOperands";
1323 def MemoryUnscaledFB16Operand : AsmOperandClass {
1324 let Name = "MemoryUnscaledFB16";
1325 let RenderMethod = "addMemoryUnscaledOperands";
1327 def MemoryUnscaledFB32Operand : AsmOperandClass {
1328 let Name = "MemoryUnscaledFB32";
1329 let RenderMethod = "addMemoryUnscaledOperands";
1331 def MemoryUnscaledFB64Operand : AsmOperandClass {
1332 let Name = "MemoryUnscaledFB64";
1333 let RenderMethod = "addMemoryUnscaledOperands";
1335 def MemoryUnscaledFB128Operand : AsmOperandClass {
1336 let Name = "MemoryUnscaledFB128";
1337 let RenderMethod = "addMemoryUnscaledOperands";
1339 def am_unscaled_fb8 : Operand<i64> {
1340 let ParserMatchClass = MemoryUnscaledFB8Operand;
1341 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
1343 def am_unscaled_fb16 : Operand<i64> {
1344 let ParserMatchClass = MemoryUnscaledFB16Operand;
1345 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
1347 def am_unscaled_fb32 : Operand<i64> {
1348 let ParserMatchClass = MemoryUnscaledFB32Operand;
1349 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
1351 def am_unscaled_fb64 : Operand<i64> {
1352 let ParserMatchClass = MemoryUnscaledFB64Operand;
1353 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
1355 def am_unscaled_fb128 : Operand<i64> {
1356 let ParserMatchClass = MemoryUnscaledFB128Operand;
1357 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
1359 def : InstAlias<"ldr $Rt, $addr", (LDURXi GPR64:$Rt, am_unscaled_fb64:$addr)>;
1360 def : InstAlias<"ldr $Rt, $addr", (LDURWi GPR32:$Rt, am_unscaled_fb32:$addr)>;
1361 def : InstAlias<"ldr $Rt, $addr", (LDURBi FPR8:$Rt, am_unscaled_fb8:$addr)>;
1362 def : InstAlias<"ldr $Rt, $addr", (LDURHi FPR16:$Rt, am_unscaled_fb16:$addr)>;
1363 def : InstAlias<"ldr $Rt, $addr", (LDURSi FPR32:$Rt, am_unscaled_fb32:$addr)>;
1364 def : InstAlias<"ldr $Rt, $addr", (LDURDi FPR64:$Rt, am_unscaled_fb64:$addr)>;
1365 def : InstAlias<"ldr $Rt, $addr", (LDURQi FPR128:$Rt, am_unscaled_fb128:$addr)>;
1368 def : Pat<(i64 (zextloadi8 am_unscaled8:$addr)),
1369 (SUBREG_TO_REG (i64 0), (LDURBBi am_unscaled8:$addr), sub_32)>;
1370 def : Pat<(i64 (zextloadi16 am_unscaled16:$addr)),
1371 (SUBREG_TO_REG (i64 0), (LDURHHi am_unscaled16:$addr), sub_32)>;
1373 // load sign-extended half-word
1375 : LoadUnscaled<0b01, 0, 0b11, GPR32, am_unscaled16, "ldursh",
1376 [(set GPR32:$Rt, (sextloadi16 am_unscaled16:$addr))]>;
1378 : LoadUnscaled<0b01, 0, 0b10, GPR64, am_unscaled16, "ldursh",
1379 [(set GPR64:$Rt, (sextloadi16 am_unscaled16:$addr))]>;
1381 // load sign-extended byte
1383 : LoadUnscaled<0b00, 0, 0b11, GPR32, am_unscaled8, "ldursb",
1384 [(set GPR32:$Rt, (sextloadi8 am_unscaled8:$addr))]>;
1386 : LoadUnscaled<0b00, 0, 0b10, GPR64, am_unscaled8, "ldursb",
1387 [(set GPR64:$Rt, (sextloadi8 am_unscaled8:$addr))]>;
1389 // load sign-extended word
1391 : LoadUnscaled<0b10, 0, 0b10, GPR64, am_unscaled32, "ldursw",
1392 [(set GPR64:$Rt, (sextloadi32 am_unscaled32:$addr))]>;
1394 // zero and sign extending aliases from generic LDR* mnemonics to LDUR*.
1395 def : InstAlias<"ldrb $Rt, $addr", (LDURBBi GPR32:$Rt, am_unscaled_fb8:$addr)>;
1396 def : InstAlias<"ldrh $Rt, $addr", (LDURHHi GPR32:$Rt, am_unscaled_fb16:$addr)>;
1397 def : InstAlias<"ldrsb $Rt, $addr", (LDURSBWi GPR32:$Rt, am_unscaled_fb8:$addr)>;
1398 def : InstAlias<"ldrsb $Rt, $addr", (LDURSBXi GPR64:$Rt, am_unscaled_fb8:$addr)>;
1399 def : InstAlias<"ldrsh $Rt, $addr", (LDURSHWi GPR32:$Rt, am_unscaled_fb16:$addr)>;
1400 def : InstAlias<"ldrsh $Rt, $addr", (LDURSHXi GPR64:$Rt, am_unscaled_fb16:$addr)>;
1401 def : InstAlias<"ldrsw $Rt, $addr", (LDURSWi GPR64:$Rt, am_unscaled_fb32:$addr)>;
1404 def PRFUMi : PrefetchUnscaled<0b11, 0, 0b10, "prfum",
1405 [(ARM64Prefetch imm:$Rt, am_unscaled64:$addr)]>;
1408 // (unscaled immediate, unprivileged)
1409 def LDTRXi : LoadUnprivileged<0b11, 0, 0b01, GPR64, "ldtr">;
1410 def LDTRWi : LoadUnprivileged<0b10, 0, 0b01, GPR32, "ldtr">;
1412 def LDTRHi : LoadUnprivileged<0b01, 0, 0b01, GPR32, "ldtrh">;
1413 def LDTRBi : LoadUnprivileged<0b00, 0, 0b01, GPR32, "ldtrb">;
1415 // load sign-extended half-word
1416 def LDTRSHWi : LoadUnprivileged<0b01, 0, 0b11, GPR32, "ldtrsh">;
1417 def LDTRSHXi : LoadUnprivileged<0b01, 0, 0b10, GPR64, "ldtrsh">;
1419 // load sign-extended byte
1420 def LDTRSBWi : LoadUnprivileged<0b00, 0, 0b11, GPR32, "ldtrsb">;
1421 def LDTRSBXi : LoadUnprivileged<0b00, 0, 0b10, GPR64, "ldtrsb">;
1423 // load sign-extended word
1424 def LDTRSWi : LoadUnprivileged<0b10, 0, 0b10, GPR64, "ldtrsw">;
1427 // (immediate pre-indexed)
1428 def LDRWpre : LoadPreIdx<0b10, 0, 0b01, GPR32, "ldr">;
1429 def LDRXpre : LoadPreIdx<0b11, 0, 0b01, GPR64, "ldr">;
1430 def LDRBpre : LoadPreIdx<0b00, 1, 0b01, FPR8, "ldr">;
1431 def LDRHpre : LoadPreIdx<0b01, 1, 0b01, FPR16, "ldr">;
1432 def LDRSpre : LoadPreIdx<0b10, 1, 0b01, FPR32, "ldr">;
1433 def LDRDpre : LoadPreIdx<0b11, 1, 0b01, FPR64, "ldr">;
1434 def LDRQpre : LoadPreIdx<0b00, 1, 0b11, FPR128, "ldr">;
1436 // load sign-extended half-word
1437 def LDRSHWpre : LoadPreIdx<0b01, 0, 0b11, GPR32, "ldrsh">;
1438 def LDRSHXpre : LoadPreIdx<0b01, 0, 0b10, GPR64, "ldrsh">;
1440 // load sign-extended byte
1441 def LDRSBWpre : LoadPreIdx<0b00, 0, 0b11, GPR32, "ldrsb">;
1442 def LDRSBXpre : LoadPreIdx<0b00, 0, 0b10, GPR64, "ldrsb">;
1444 // load zero-extended byte
1445 def LDRBBpre : LoadPreIdx<0b00, 0, 0b01, GPR32, "ldrb">;
1446 def LDRHHpre : LoadPreIdx<0b01, 0, 0b01, GPR32, "ldrh">;
1448 // load sign-extended word
1449 def LDRSWpre : LoadPreIdx<0b10, 0, 0b10, GPR64, "ldrsw">;
1451 // ISel pseudos and patterns. See expanded comment on LoadPreIdxPseudo.
1452 def LDRDpre_isel : LoadPreIdxPseudo<FPR64>;
1453 def LDRSpre_isel : LoadPreIdxPseudo<FPR32>;
1454 def LDRXpre_isel : LoadPreIdxPseudo<GPR64>;
1455 def LDRWpre_isel : LoadPreIdxPseudo<GPR32>;
1456 def LDRHHpre_isel : LoadPreIdxPseudo<GPR32>;
1457 def LDRBBpre_isel : LoadPreIdxPseudo<GPR32>;
1459 def LDRSWpre_isel : LoadPreIdxPseudo<GPR64>;
1460 def LDRSHWpre_isel : LoadPreIdxPseudo<GPR32>;
1461 def LDRSHXpre_isel : LoadPreIdxPseudo<GPR64>;
1462 def LDRSBWpre_isel : LoadPreIdxPseudo<GPR32>;
1463 def LDRSBXpre_isel : LoadPreIdxPseudo<GPR64>;
1466 // (immediate post-indexed)
1467 def LDRWpost : LoadPostIdx<0b10, 0, 0b01, GPR32, "ldr">;
1468 def LDRXpost : LoadPostIdx<0b11, 0, 0b01, GPR64, "ldr">;
1469 def LDRBpost : LoadPostIdx<0b00, 1, 0b01, FPR8, "ldr">;
1470 def LDRHpost : LoadPostIdx<0b01, 1, 0b01, FPR16, "ldr">;
1471 def LDRSpost : LoadPostIdx<0b10, 1, 0b01, FPR32, "ldr">;
1472 def LDRDpost : LoadPostIdx<0b11, 1, 0b01, FPR64, "ldr">;
1473 def LDRQpost : LoadPostIdx<0b00, 1, 0b11, FPR128, "ldr">;
1475 // load sign-extended half-word
1476 def LDRSHWpost : LoadPostIdx<0b01, 0, 0b11, GPR32, "ldrsh">;
1477 def LDRSHXpost : LoadPostIdx<0b01, 0, 0b10, GPR64, "ldrsh">;
1479 // load sign-extended byte
1480 def LDRSBWpost : LoadPostIdx<0b00, 0, 0b11, GPR32, "ldrsb">;
1481 def LDRSBXpost : LoadPostIdx<0b00, 0, 0b10, GPR64, "ldrsb">;
1483 // load zero-extended byte
1484 def LDRBBpost : LoadPostIdx<0b00, 0, 0b01, GPR32, "ldrb">;
1485 def LDRHHpost : LoadPostIdx<0b01, 0, 0b01, GPR32, "ldrh">;
1487 // load sign-extended word
1488 def LDRSWpost : LoadPostIdx<0b10, 0, 0b10, GPR64, "ldrsw">;
1490 // ISel pseudos and patterns. See expanded comment on LoadPostIdxPseudo.
1491 def LDRDpost_isel : LoadPostIdxPseudo<FPR64>;
1492 def LDRSpost_isel : LoadPostIdxPseudo<FPR32>;
1493 def LDRXpost_isel : LoadPostIdxPseudo<GPR64>;
1494 def LDRWpost_isel : LoadPostIdxPseudo<GPR32>;
1495 def LDRHHpost_isel : LoadPostIdxPseudo<GPR32>;
1496 def LDRBBpost_isel : LoadPostIdxPseudo<GPR32>;
1498 def LDRSWpost_isel : LoadPostIdxPseudo<GPR64>;
1499 def LDRSHWpost_isel : LoadPostIdxPseudo<GPR32>;
1500 def LDRSHXpost_isel : LoadPostIdxPseudo<GPR64>;
1501 def LDRSBWpost_isel : LoadPostIdxPseudo<GPR32>;
1502 def LDRSBXpost_isel : LoadPostIdxPseudo<GPR64>;
1504 //===----------------------------------------------------------------------===//
1505 // Store instructions.
1506 //===----------------------------------------------------------------------===//
1508 // Pair (indexed, offset)
1509 // FIXME: Use dedicated range-checked addressing mode operand here.
1510 def STPWi : StorePairOffset<0b00, 0, GPR32, am_indexed32simm7, "stp">;
1511 def STPXi : StorePairOffset<0b10, 0, GPR64, am_indexed64simm7, "stp">;
1512 def STPSi : StorePairOffset<0b00, 1, FPR32, am_indexed32simm7, "stp">;
1513 def STPDi : StorePairOffset<0b01, 1, FPR64, am_indexed64simm7, "stp">;
1514 def STPQi : StorePairOffset<0b10, 1, FPR128, am_indexed128simm7, "stp">;
1516 // Pair (pre-indexed)
1517 def STPWpre : StorePairPreIdx<0b00, 0, GPR32, am_indexed32simm7, "stp">;
1518 def STPXpre : StorePairPreIdx<0b10, 0, GPR64, am_indexed64simm7, "stp">;
1519 def STPSpre : StorePairPreIdx<0b00, 1, FPR32, am_indexed32simm7, "stp">;
1520 def STPDpre : StorePairPreIdx<0b01, 1, FPR64, am_indexed64simm7, "stp">;
1521 def STPQpre : StorePairPreIdx<0b10, 1, FPR128, am_indexed128simm7, "stp">;
1523 // Pair (pre-indexed)
1524 def STPWpost : StorePairPostIdx<0b00, 0, GPR32, simm7s4, "stp">;
1525 def STPXpost : StorePairPostIdx<0b10, 0, GPR64, simm7s8, "stp">;
1526 def STPSpost : StorePairPostIdx<0b00, 1, FPR32, simm7s4, "stp">;
1527 def STPDpost : StorePairPostIdx<0b01, 1, FPR64, simm7s8, "stp">;
1528 def STPQpost : StorePairPostIdx<0b10, 1, FPR128, simm7s16, "stp">;
1530 // Pair (no allocate)
1531 def STNPWi : StorePairNoAlloc<0b00, 0, GPR32, am_indexed32simm7, "stnp">;
1532 def STNPXi : StorePairNoAlloc<0b10, 0, GPR64, am_indexed64simm7, "stnp">;
1533 def STNPSi : StorePairNoAlloc<0b00, 1, FPR32, am_indexed32simm7, "stnp">;
1534 def STNPDi : StorePairNoAlloc<0b01, 1, FPR64, am_indexed64simm7, "stnp">;
1535 def STNPQi : StorePairNoAlloc<0b10, 1, FPR128, am_indexed128simm7, "stnp">;
1538 // (Register offset)
1540 let AddedComplexity = 10 in {
1543 def STRHHro : Store16RO<0b01, 0, 0b00, GPR32, "strh",
1544 [(truncstorei16 GPR32:$Rt, ro_indexed16:$addr)]>;
1545 def STRBBro : Store8RO<0b00, 0, 0b00, GPR32, "strb",
1546 [(truncstorei8 GPR32:$Rt, ro_indexed8:$addr)]>;
1547 def STRWro : Store32RO<0b10, 0, 0b00, GPR32, "str",
1548 [(store GPR32:$Rt, ro_indexed32:$addr)]>;
1549 def STRXro : Store64RO<0b11, 0, 0b00, GPR64, "str",
1550 [(store GPR64:$Rt, ro_indexed64:$addr)]>;
1553 def : Pat<(truncstorei8 GPR64:$Rt, ro_indexed8:$addr),
1554 (STRBBro (EXTRACT_SUBREG GPR64:$Rt, sub_32), ro_indexed8:$addr)>;
1555 def : Pat<(truncstorei16 GPR64:$Rt, ro_indexed16:$addr),
1556 (STRHHro (EXTRACT_SUBREG GPR64:$Rt, sub_32), ro_indexed16:$addr)>;
1557 def : Pat<(truncstorei32 GPR64:$Rt, ro_indexed32:$addr),
1558 (STRWro (EXTRACT_SUBREG GPR64:$Rt, sub_32), ro_indexed32:$addr)>;
1562 def STRBro : Store8RO<0b00, 1, 0b00, FPR8, "str",
1563 [(store FPR8:$Rt, ro_indexed8:$addr)]>;
1564 def STRHro : Store16RO<0b01, 1, 0b00, FPR16, "str",
1565 [(store FPR16:$Rt, ro_indexed16:$addr)]>;
1566 def STRSro : Store32RO<0b10, 1, 0b00, FPR32, "str",
1567 [(store (f32 FPR32:$Rt), ro_indexed32:$addr)]>;
1568 def STRDro : Store64RO<0b11, 1, 0b00, FPR64, "str",
1569 [(store (f64 FPR64:$Rt), ro_indexed64:$addr)]>;
1570 def STRQro : Store128RO<0b00, 1, 0b10, FPR128, "str", []> {
1574 // Match all store 64 bits width whose type is compatible with FPR64
1575 def : Pat<(store (v2f32 FPR64:$Rn), ro_indexed64:$addr),
1576 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1577 def : Pat<(store (v1f64 FPR64:$Rn), ro_indexed64:$addr),
1578 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1579 def : Pat<(store (v8i8 FPR64:$Rn), ro_indexed64:$addr),
1580 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1581 def : Pat<(store (v4i16 FPR64:$Rn), ro_indexed64:$addr),
1582 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1583 def : Pat<(store (v2i32 FPR64:$Rn), ro_indexed64:$addr),
1584 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1585 def : Pat<(store (v1i64 FPR64:$Rn), ro_indexed64:$addr),
1586 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1588 // Match all store 128 bits width whose type is compatible with FPR128
1589 def : Pat<(store (v4f32 FPR128:$Rn), ro_indexed128:$addr),
1590 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1591 def : Pat<(store (v2f64 FPR128:$Rn), ro_indexed128:$addr),
1592 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1593 def : Pat<(store (v16i8 FPR128:$Rn), ro_indexed128:$addr),
1594 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1595 def : Pat<(store (v8i16 FPR128:$Rn), ro_indexed128:$addr),
1596 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1597 def : Pat<(store (v4i32 FPR128:$Rn), ro_indexed128:$addr),
1598 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1599 def : Pat<(store (v2i64 FPR128:$Rn), ro_indexed128:$addr),
1600 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1601 def : Pat<(store (f128 FPR128:$Rn), ro_indexed128:$addr),
1602 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1605 // (unsigned immediate)
1606 def STRXui : StoreUI<0b11, 0, 0b00, GPR64, am_indexed64, "str",
1607 [(store GPR64:$Rt, am_indexed64:$addr)]>;
1608 def STRWui : StoreUI<0b10, 0, 0b00, GPR32, am_indexed32, "str",
1609 [(store GPR32:$Rt, am_indexed32:$addr)]>;
1610 def STRBui : StoreUI<0b00, 1, 0b00, FPR8, am_indexed8, "str",
1611 [(store FPR8:$Rt, am_indexed8:$addr)]>;
1612 def STRHui : StoreUI<0b01, 1, 0b00, FPR16, am_indexed16, "str",
1613 [(store FPR16:$Rt, am_indexed16:$addr)]>;
1614 def STRSui : StoreUI<0b10, 1, 0b00, FPR32, am_indexed32, "str",
1615 [(store (f32 FPR32:$Rt), am_indexed32:$addr)]>;
1616 def STRDui : StoreUI<0b11, 1, 0b00, FPR64, am_indexed64, "str",
1617 [(store (f64 FPR64:$Rt), am_indexed64:$addr)]>;
1618 def STRQui : StoreUI<0b00, 1, 0b10, FPR128, am_indexed128, "str", []> {
1622 // Match all store 64 bits width whose type is compatible with FPR64
1623 def : Pat<(store (v2f32 FPR64:$Rn), am_indexed64:$addr),
1624 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1625 def : Pat<(store (v1f64 FPR64:$Rn), am_indexed64:$addr),
1626 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1627 def : Pat<(store (v8i8 FPR64:$Rn), am_indexed64:$addr),
1628 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1629 def : Pat<(store (v4i16 FPR64:$Rn), am_indexed64:$addr),
1630 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1631 def : Pat<(store (v2i32 FPR64:$Rn), am_indexed64:$addr),
1632 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1633 def : Pat<(store (v1i64 FPR64:$Rn), am_indexed64:$addr),
1634 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1636 // Match all store 128 bits width whose type is compatible with FPR128
1637 def : Pat<(store (v4f32 FPR128:$Rn), am_indexed128:$addr),
1638 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1639 def : Pat<(store (v2f64 FPR128:$Rn), am_indexed128:$addr),
1640 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1641 def : Pat<(store (v16i8 FPR128:$Rn), am_indexed128:$addr),
1642 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1643 def : Pat<(store (v8i16 FPR128:$Rn), am_indexed128:$addr),
1644 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1645 def : Pat<(store (v4i32 FPR128:$Rn), am_indexed128:$addr),
1646 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1647 def : Pat<(store (v2i64 FPR128:$Rn), am_indexed128:$addr),
1648 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1649 def : Pat<(store (f128 FPR128:$Rn), am_indexed128:$addr),
1650 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1652 def STRHHui : StoreUI<0b01, 0, 0b00, GPR32, am_indexed16, "strh",
1653 [(truncstorei16 GPR32:$Rt, am_indexed16:$addr)]>;
1654 def STRBBui : StoreUI<0b00, 0, 0b00, GPR32, am_indexed8, "strb",
1655 [(truncstorei8 GPR32:$Rt, am_indexed8:$addr)]>;
1658 def : Pat<(truncstorei32 GPR64:$Rt, am_indexed32:$addr),
1659 (STRWui (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_indexed32:$addr)>;
1660 def : Pat<(truncstorei16 GPR64:$Rt, am_indexed16:$addr),
1661 (STRHHui (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_indexed16:$addr)>;
1662 def : Pat<(truncstorei8 GPR64:$Rt, am_indexed8:$addr),
1663 (STRBBui (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_indexed8:$addr)>;
1665 } // AddedComplexity = 10
1668 // (unscaled immediate)
1669 def STURXi : StoreUnscaled<0b11, 0, 0b00, GPR64, am_unscaled64, "stur",
1670 [(store GPR64:$Rt, am_unscaled64:$addr)]>;
1671 def STURWi : StoreUnscaled<0b10, 0, 0b00, GPR32, am_unscaled32, "stur",
1672 [(store GPR32:$Rt, am_unscaled32:$addr)]>;
1673 def STURBi : StoreUnscaled<0b00, 1, 0b00, FPR8, am_unscaled8, "stur",
1674 [(store FPR8:$Rt, am_unscaled8:$addr)]>;
1675 def STURHi : StoreUnscaled<0b01, 1, 0b00, FPR16, am_unscaled16, "stur",
1676 [(store FPR16:$Rt, am_unscaled16:$addr)]>;
1677 def STURSi : StoreUnscaled<0b10, 1, 0b00, FPR32, am_unscaled32, "stur",
1678 [(store (f32 FPR32:$Rt), am_unscaled32:$addr)]>;
1679 def STURDi : StoreUnscaled<0b11, 1, 0b00, FPR64, am_unscaled64, "stur",
1680 [(store (f64 FPR64:$Rt), am_unscaled64:$addr)]>;
1681 def STURQi : StoreUnscaled<0b00, 1, 0b10, FPR128, am_unscaled128, "stur",
1682 [(store (v2f64 FPR128:$Rt), am_unscaled128:$addr)]>;
1683 def STURHHi : StoreUnscaled<0b01, 0, 0b00, GPR32, am_unscaled16, "sturh",
1684 [(truncstorei16 GPR32:$Rt, am_unscaled16:$addr)]>;
1685 def STURBBi : StoreUnscaled<0b00, 0, 0b00, GPR32, am_unscaled8, "sturb",
1686 [(truncstorei8 GPR32:$Rt, am_unscaled8:$addr)]>;
1688 // Match all store 64 bits width whose type is compatible with FPR64
1689 def : Pat<(store (v2f32 FPR64:$Rn), am_unscaled64:$addr),
1690 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1691 def : Pat<(store (v1f64 FPR64:$Rn), am_unscaled64:$addr),
1692 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1693 def : Pat<(store (v8i8 FPR64:$Rn), am_unscaled64:$addr),
1694 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1695 def : Pat<(store (v4i16 FPR64:$Rn), am_unscaled64:$addr),
1696 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1697 def : Pat<(store (v2i32 FPR64:$Rn), am_unscaled64:$addr),
1698 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1699 def : Pat<(store (v1i64 FPR64:$Rn), am_unscaled64:$addr),
1700 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1702 // Match all store 128 bits width whose type is compatible with FPR128
1703 def : Pat<(store (v4f32 FPR128:$Rn), am_unscaled128:$addr),
1704 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1705 def : Pat<(store (v2f64 FPR128:$Rn), am_unscaled128:$addr),
1706 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1707 def : Pat<(store (v16i8 FPR128:$Rn), am_unscaled128:$addr),
1708 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1709 def : Pat<(store (v8i16 FPR128:$Rn), am_unscaled128:$addr),
1710 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1711 def : Pat<(store (v4i32 FPR128:$Rn), am_unscaled128:$addr),
1712 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1713 def : Pat<(store (v2i64 FPR128:$Rn), am_unscaled128:$addr),
1714 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1715 def : Pat<(store (f128 FPR128:$Rn), am_unscaled128:$addr),
1716 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1718 // unscaled i64 truncating stores
1719 def : Pat<(truncstorei32 GPR64:$Rt, am_unscaled32:$addr),
1720 (STURWi (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_unscaled32:$addr)>;
1721 def : Pat<(truncstorei16 GPR64:$Rt, am_unscaled16:$addr),
1722 (STURHHi (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_unscaled16:$addr)>;
1723 def : Pat<(truncstorei8 GPR64:$Rt, am_unscaled8:$addr),
1724 (STURBBi (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_unscaled8:$addr)>;
1727 // STR mnemonics fall back to STUR for negative or unaligned offsets.
1728 def : InstAlias<"str $Rt, $addr", (STURXi GPR64:$Rt, am_unscaled_fb64:$addr)>;
1729 def : InstAlias<"str $Rt, $addr", (STURWi GPR32:$Rt, am_unscaled_fb32:$addr)>;
1730 def : InstAlias<"str $Rt, $addr", (STURBi FPR8:$Rt, am_unscaled_fb8:$addr)>;
1731 def : InstAlias<"str $Rt, $addr", (STURHi FPR16:$Rt, am_unscaled_fb16:$addr)>;
1732 def : InstAlias<"str $Rt, $addr", (STURSi FPR32:$Rt, am_unscaled_fb32:$addr)>;
1733 def : InstAlias<"str $Rt, $addr", (STURDi FPR64:$Rt, am_unscaled_fb64:$addr)>;
1734 def : InstAlias<"str $Rt, $addr", (STURQi FPR128:$Rt, am_unscaled_fb128:$addr)>;
1736 def : InstAlias<"strb $Rt, $addr", (STURBBi GPR32:$Rt, am_unscaled_fb8:$addr)>;
1737 def : InstAlias<"strh $Rt, $addr", (STURHHi GPR32:$Rt, am_unscaled_fb16:$addr)>;
1740 // (unscaled immediate, unprivileged)
1741 def STTRWi : StoreUnprivileged<0b10, 0, 0b00, GPR32, "sttr">;
1742 def STTRXi : StoreUnprivileged<0b11, 0, 0b00, GPR64, "sttr">;
1744 def STTRHi : StoreUnprivileged<0b01, 0, 0b00, GPR32, "sttrh">;
1745 def STTRBi : StoreUnprivileged<0b00, 0, 0b00, GPR32, "sttrb">;
1748 // (immediate pre-indexed)
1749 def STRWpre : StorePreIdx<0b10, 0, 0b00, GPR32, "str">;
1750 def STRXpre : StorePreIdx<0b11, 0, 0b00, GPR64, "str">;
1751 def STRBpre : StorePreIdx<0b00, 1, 0b00, FPR8, "str">;
1752 def STRHpre : StorePreIdx<0b01, 1, 0b00, FPR16, "str">;
1753 def STRSpre : StorePreIdx<0b10, 1, 0b00, FPR32, "str">;
1754 def STRDpre : StorePreIdx<0b11, 1, 0b00, FPR64, "str">;
1755 def STRQpre : StorePreIdx<0b00, 1, 0b10, FPR128, "str">;
1757 def STRBBpre : StorePreIdx<0b00, 0, 0b00, GPR32, "strb">;
1758 def STRHHpre : StorePreIdx<0b01, 0, 0b00, GPR32, "strh">;
1760 // ISel pseudos and patterns. See expanded comment on StorePreIdxPseudo.
1761 defm STRDpre : StorePreIdxPseudo<FPR64, f64, pre_store>;
1762 defm STRSpre : StorePreIdxPseudo<FPR32, f32, pre_store>;
1763 defm STRXpre : StorePreIdxPseudo<GPR64, i64, pre_store>;
1764 defm STRWpre : StorePreIdxPseudo<GPR32, i32, pre_store>;
1765 defm STRHHpre : StorePreIdxPseudo<GPR32, i32, pre_truncsti16>;
1766 defm STRBBpre : StorePreIdxPseudo<GPR32, i32, pre_truncsti8>;
1768 def : Pat<(pre_truncsti32 GPR64:$Rt, am_noindex:$addr, simm9:$off),
1769 (STRWpre_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
1771 def : Pat<(pre_truncsti16 GPR64:$Rt, am_noindex:$addr, simm9:$off),
1772 (STRHHpre_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
1774 def : Pat<(pre_truncsti8 GPR64:$Rt, am_noindex:$addr, simm9:$off),
1775 (STRBBpre_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
1779 // (immediate post-indexed)
1780 def STRWpost : StorePostIdx<0b10, 0, 0b00, GPR32, "str">;
1781 def STRXpost : StorePostIdx<0b11, 0, 0b00, GPR64, "str">;
1782 def STRBpost : StorePostIdx<0b00, 1, 0b00, FPR8, "str">;
1783 def STRHpost : StorePostIdx<0b01, 1, 0b00, FPR16, "str">;
1784 def STRSpost : StorePostIdx<0b10, 1, 0b00, FPR32, "str">;
1785 def STRDpost : StorePostIdx<0b11, 1, 0b00, FPR64, "str">;
1786 def STRQpost : StorePostIdx<0b00, 1, 0b10, FPR128, "str">;
1788 def STRBBpost : StorePostIdx<0b00, 0, 0b00, GPR32, "strb">;
1789 def STRHHpost : StorePostIdx<0b01, 0, 0b00, GPR32, "strh">;
1791 // ISel pseudos and patterns. See expanded comment on StorePostIdxPseudo.
1792 defm STRDpost : StorePostIdxPseudo<FPR64, f64, post_store, STRDpost>;
1793 defm STRSpost : StorePostIdxPseudo<FPR32, f32, post_store, STRSpost>;
1794 defm STRXpost : StorePostIdxPseudo<GPR64, i64, post_store, STRXpost>;
1795 defm STRWpost : StorePostIdxPseudo<GPR32, i32, post_store, STRWpost>;
1796 defm STRHHpost : StorePostIdxPseudo<GPR32, i32, post_truncsti16, STRHHpost>;
1797 defm STRBBpost : StorePostIdxPseudo<GPR32, i32, post_truncsti8, STRBBpost>;
1799 def : Pat<(post_truncsti32 GPR64:$Rt, am_noindex:$addr, simm9:$off),
1800 (STRWpost_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
1802 def : Pat<(post_truncsti16 GPR64:$Rt, am_noindex:$addr, simm9:$off),
1803 (STRHHpost_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
1805 def : Pat<(post_truncsti8 GPR64:$Rt, am_noindex:$addr, simm9:$off),
1806 (STRBBpost_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
1810 //===----------------------------------------------------------------------===//
1811 // Load/store exclusive instructions.
1812 //===----------------------------------------------------------------------===//
1814 def LDARW : LoadAcquire <0b10, 1, 1, 0, 1, GPR32, "ldar">;
1815 def LDARX : LoadAcquire <0b11, 1, 1, 0, 1, GPR64, "ldar">;
1816 def LDARB : LoadAcquire <0b00, 1, 1, 0, 1, GPR32, "ldarb">;
1817 def LDARH : LoadAcquire <0b01, 1, 1, 0, 1, GPR32, "ldarh">;
1819 def LDAXRW : LoadExclusive <0b10, 0, 1, 0, 1, GPR32, "ldaxr">;
1820 def LDAXRX : LoadExclusive <0b11, 0, 1, 0, 1, GPR64, "ldaxr">;
1821 def LDAXRB : LoadExclusive <0b00, 0, 1, 0, 1, GPR32, "ldaxrb">;
1822 def LDAXRH : LoadExclusive <0b01, 0, 1, 0, 1, GPR32, "ldaxrh">;
1824 def LDXRW : LoadExclusive <0b10, 0, 1, 0, 0, GPR32, "ldxr">;
1825 def LDXRX : LoadExclusive <0b11, 0, 1, 0, 0, GPR64, "ldxr">;
1826 def LDXRB : LoadExclusive <0b00, 0, 1, 0, 0, GPR32, "ldxrb">;
1827 def LDXRH : LoadExclusive <0b01, 0, 1, 0, 0, GPR32, "ldxrh">;
1829 def STLRW : StoreRelease <0b10, 1, 0, 0, 1, GPR32, "stlr">;
1830 def STLRX : StoreRelease <0b11, 1, 0, 0, 1, GPR64, "stlr">;
1831 def STLRB : StoreRelease <0b00, 1, 0, 0, 1, GPR32, "stlrb">;
1832 def STLRH : StoreRelease <0b01, 1, 0, 0, 1, GPR32, "stlrh">;
1834 def STLXRW : StoreExclusive<0b10, 0, 0, 0, 1, GPR32, "stlxr">;
1835 def STLXRX : StoreExclusive<0b11, 0, 0, 0, 1, GPR64, "stlxr">;
1836 def STLXRB : StoreExclusive<0b00, 0, 0, 0, 1, GPR32, "stlxrb">;
1837 def STLXRH : StoreExclusive<0b01, 0, 0, 0, 1, GPR32, "stlxrh">;
1839 def STXRW : StoreExclusive<0b10, 0, 0, 0, 0, GPR32, "stxr">;
1840 def STXRX : StoreExclusive<0b11, 0, 0, 0, 0, GPR64, "stxr">;
1841 def STXRB : StoreExclusive<0b00, 0, 0, 0, 0, GPR32, "stxrb">;
1842 def STXRH : StoreExclusive<0b01, 0, 0, 0, 0, GPR32, "stxrh">;
1844 def LDAXPW : LoadExclusivePair<0b10, 0, 1, 1, 1, GPR32, "ldaxp">;
1845 def LDAXPX : LoadExclusivePair<0b11, 0, 1, 1, 1, GPR64, "ldaxp">;
1847 def LDXPW : LoadExclusivePair<0b10, 0, 1, 1, 0, GPR32, "ldxp">;
1848 def LDXPX : LoadExclusivePair<0b11, 0, 1, 1, 0, GPR64, "ldxp">;
1850 def STLXPW : StoreExclusivePair<0b10, 0, 0, 1, 1, GPR32, "stlxp">;
1851 def STLXPX : StoreExclusivePair<0b11, 0, 0, 1, 1, GPR64, "stlxp">;
1853 def STXPW : StoreExclusivePair<0b10, 0, 0, 1, 0, GPR32, "stxp">;
1854 def STXPX : StoreExclusivePair<0b11, 0, 0, 1, 0, GPR64, "stxp">;
1856 //===----------------------------------------------------------------------===//
1857 // Scaled floating point to integer conversion instructions.
1858 //===----------------------------------------------------------------------===//
1860 defm FCVTAS : FPToIntegerUnscaled<0b00, 0b100, "fcvtas", int_arm64_neon_fcvtas>;
1861 defm FCVTAU : FPToIntegerUnscaled<0b00, 0b101, "fcvtau", int_arm64_neon_fcvtau>;
1862 defm FCVTMS : FPToIntegerUnscaled<0b10, 0b000, "fcvtms", int_arm64_neon_fcvtms>;
1863 defm FCVTMU : FPToIntegerUnscaled<0b10, 0b001, "fcvtmu", int_arm64_neon_fcvtmu>;
1864 defm FCVTNS : FPToIntegerUnscaled<0b00, 0b000, "fcvtns", int_arm64_neon_fcvtns>;
1865 defm FCVTNU : FPToIntegerUnscaled<0b00, 0b001, "fcvtnu", int_arm64_neon_fcvtnu>;
1866 defm FCVTPS : FPToIntegerUnscaled<0b01, 0b000, "fcvtps", int_arm64_neon_fcvtps>;
1867 defm FCVTPU : FPToIntegerUnscaled<0b01, 0b001, "fcvtpu", int_arm64_neon_fcvtpu>;
1868 defm FCVTZS : FPToIntegerUnscaled<0b11, 0b000, "fcvtzs", fp_to_sint>;
1869 defm FCVTZU : FPToIntegerUnscaled<0b11, 0b001, "fcvtzu", fp_to_uint>;
1870 defm FCVTZS : FPToIntegerScaled<0b11, 0b000, "fcvtzs", fp_to_sint>;
1871 defm FCVTZU : FPToIntegerScaled<0b11, 0b001, "fcvtzu", fp_to_uint>;
1872 let isCodeGenOnly = 1 in {
1873 defm FCVTZS_Int : FPToIntegerUnscaled<0b11, 0b000, "fcvtzs", int_arm64_neon_fcvtzs>;
1874 defm FCVTZU_Int : FPToIntegerUnscaled<0b11, 0b001, "fcvtzu", int_arm64_neon_fcvtzu>;
1875 defm FCVTZS_Int : FPToIntegerScaled<0b11, 0b000, "fcvtzs", int_arm64_neon_fcvtzs>;
1876 defm FCVTZU_Int : FPToIntegerScaled<0b11, 0b001, "fcvtzu", int_arm64_neon_fcvtzu>;
1879 //===----------------------------------------------------------------------===//
1880 // Scaled integer to floating point conversion instructions.
1881 //===----------------------------------------------------------------------===//
1883 defm SCVTF : IntegerToFP<0, "scvtf", sint_to_fp>;
1884 defm UCVTF : IntegerToFP<1, "ucvtf", uint_to_fp>;
1886 //===----------------------------------------------------------------------===//
1887 // Unscaled integer to floating point conversion instruction.
1888 //===----------------------------------------------------------------------===//
1890 defm FMOV : UnscaledConversion<"fmov">;
1892 def : Pat<(f32 (fpimm0)), (FMOVWSr WZR)>, Requires<[NoZCZ]>;
1893 def : Pat<(f64 (fpimm0)), (FMOVXDr XZR)>, Requires<[NoZCZ]>;
1895 def : Pat<(v8i8 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1896 def : Pat<(v4i16 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1897 def : Pat<(v2i32 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1898 def : Pat<(v1i64 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1899 def : Pat<(v2f32 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1900 def : Pat<(v1f64 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1901 def : Pat<(v1i64 (scalar_to_vector GPR64:$Xn)),
1902 (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1903 def : Pat<(v1f64 (scalar_to_vector GPR64:$Xn)),
1904 (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1905 def : Pat<(v1f64 (scalar_to_vector (f64 FPR64:$Xn))), (v1f64 FPR64:$Xn)>;
1907 def : Pat<(i64 (bitconvert (v8i8 V64:$Vn))),
1908 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
1909 def : Pat<(i64 (bitconvert (v4i16 V64:$Vn))),
1910 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
1911 def : Pat<(i64 (bitconvert (v2i32 V64:$Vn))),
1912 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
1913 def : Pat<(i64 (bitconvert (v1i64 V64:$Vn))),
1914 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
1915 def : Pat<(i64 (bitconvert (v2f32 V64:$Vn))),
1916 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
1917 def : Pat<(i64 (bitconvert (v1f64 V64:$Vn))),
1918 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
1920 def : Pat<(f32 (bitconvert (i32 GPR32:$Xn))),
1921 (COPY_TO_REGCLASS GPR32:$Xn, FPR32)>;
1922 def : Pat<(i32 (bitconvert (f32 FPR32:$Xn))),
1923 (COPY_TO_REGCLASS FPR32:$Xn, GPR32)>;
1924 def : Pat<(f64 (bitconvert (i64 GPR64:$Xn))),
1925 (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1926 def : Pat<(i64 (bitconvert (f64 FPR64:$Xn))),
1927 (COPY_TO_REGCLASS FPR64:$Xn, GPR64)>;
1929 //===----------------------------------------------------------------------===//
1930 // Floating point conversion instruction.
1931 //===----------------------------------------------------------------------===//
1933 defm FCVT : FPConversion<"fcvt">;
1935 def : Pat<(f32_to_f16 FPR32:$Rn),
1936 (i32 (COPY_TO_REGCLASS
1937 (f32 (SUBREG_TO_REG (i32 0), (FCVTHSr FPR32:$Rn), hsub)),
1940 def FCVTSHpseudo : Pseudo<(outs FPR32:$Rd), (ins FPR32:$Rn),
1941 [(set (f32 FPR32:$Rd), (f16_to_f32 i32:$Rn))]>;
1943 //===----------------------------------------------------------------------===//
1944 // Floating point single operand instructions.
1945 //===----------------------------------------------------------------------===//
1947 defm FABS : SingleOperandFPData<0b0001, "fabs", fabs>;
1948 defm FMOV : SingleOperandFPData<0b0000, "fmov">;
1949 defm FNEG : SingleOperandFPData<0b0010, "fneg", fneg>;
1950 defm FRINTA : SingleOperandFPData<0b1100, "frinta", frnd>;
1951 defm FRINTI : SingleOperandFPData<0b1111, "frinti", fnearbyint>;
1952 defm FRINTM : SingleOperandFPData<0b1010, "frintm", ffloor>;
1953 defm FRINTN : SingleOperandFPData<0b1000, "frintn", int_arm64_neon_frintn>;
1954 defm FRINTP : SingleOperandFPData<0b1001, "frintp", fceil>;
1956 def : Pat<(v1f64 (int_arm64_neon_frintn (v1f64 FPR64:$Rn))),
1957 (FRINTNDr FPR64:$Rn)>;
1959 // FRINTX is inserted to set the flags as required by FENV_ACCESS ON behavior
1960 // in the C spec. Setting hasSideEffects ensures it is not DCE'd.
1961 // <rdar://problem/13715968>
1962 // TODO: We should really model the FPSR flags correctly. This is really ugly.
1963 let hasSideEffects = 1 in {
1964 defm FRINTX : SingleOperandFPData<0b1110, "frintx", frint>;
1967 defm FRINTZ : SingleOperandFPData<0b1011, "frintz", ftrunc>;
1969 let SchedRW = [WriteFDiv] in {
1970 defm FSQRT : SingleOperandFPData<0b0011, "fsqrt", fsqrt>;
1973 //===----------------------------------------------------------------------===//
1974 // Floating point two operand instructions.
1975 //===----------------------------------------------------------------------===//
1977 defm FADD : TwoOperandFPData<0b0010, "fadd", fadd>;
1978 let SchedRW = [WriteFDiv] in {
1979 defm FDIV : TwoOperandFPData<0b0001, "fdiv", fdiv>;
1981 defm FMAXNM : TwoOperandFPData<0b0110, "fmaxnm", int_arm64_neon_fmaxnm>;
1982 defm FMAX : TwoOperandFPData<0b0100, "fmax", ARM64fmax>;
1983 defm FMINNM : TwoOperandFPData<0b0111, "fminnm", int_arm64_neon_fminnm>;
1984 defm FMIN : TwoOperandFPData<0b0101, "fmin", ARM64fmin>;
1985 let SchedRW = [WriteFMul] in {
1986 defm FMUL : TwoOperandFPData<0b0000, "fmul", fmul>;
1987 defm FNMUL : TwoOperandFPDataNeg<0b1000, "fnmul", fmul>;
1989 defm FSUB : TwoOperandFPData<0b0011, "fsub", fsub>;
1991 def : Pat<(v1f64 (ARM64fmax (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
1992 (FMAXDrr FPR64:$Rn, FPR64:$Rm)>;
1993 def : Pat<(v1f64 (ARM64fmin (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
1994 (FMINDrr FPR64:$Rn, FPR64:$Rm)>;
1995 def : Pat<(v1f64 (int_arm64_neon_fmaxnm (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
1996 (FMAXNMDrr FPR64:$Rn, FPR64:$Rm)>;
1997 def : Pat<(v1f64 (int_arm64_neon_fminnm (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
1998 (FMINNMDrr FPR64:$Rn, FPR64:$Rm)>;
2000 //===----------------------------------------------------------------------===//
2001 // Floating point three operand instructions.
2002 //===----------------------------------------------------------------------===//
2004 defm FMADD : ThreeOperandFPData<0, 0, "fmadd", fma>;
2005 defm FMSUB : ThreeOperandFPData<0, 1, "fmsub",
2006 TriOpFrag<(fma node:$LHS, (fneg node:$MHS), node:$RHS)> >;
2007 defm FNMADD : ThreeOperandFPData<1, 0, "fnmadd",
2008 TriOpFrag<(fneg (fma node:$LHS, node:$MHS, node:$RHS))> >;
2009 defm FNMSUB : ThreeOperandFPData<1, 1, "fnmsub",
2010 TriOpFrag<(fma node:$LHS, node:$MHS, (fneg node:$RHS))> >;
2012 // The following def pats catch the case where the LHS of an FMA is negated.
2013 // The TriOpFrag above catches the case where the middle operand is negated.
2015 // N.b. FMSUB etc have the accumulator at the *end* of (outs), unlike
2016 // the NEON variant.
2017 def : Pat<(f32 (fma (fneg FPR32:$Rn), FPR32:$Rm, FPR32:$Ra)),
2018 (FMSUBSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
2020 def : Pat<(f64 (fma (fneg FPR64:$Rn), FPR64:$Rm, FPR64:$Ra)),
2021 (FMSUBDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
2023 //===----------------------------------------------------------------------===//
2024 // Floating point comparison instructions.
2025 //===----------------------------------------------------------------------===//
2027 defm FCMPE : FPComparison<1, "fcmpe">;
2028 defm FCMP : FPComparison<0, "fcmp", ARM64fcmp>;
2030 //===----------------------------------------------------------------------===//
2031 // Floating point conditional comparison instructions.
2032 //===----------------------------------------------------------------------===//
2034 defm FCCMPE : FPCondComparison<1, "fccmpe">;
2035 defm FCCMP : FPCondComparison<0, "fccmp">;
2037 //===----------------------------------------------------------------------===//
2038 // Floating point conditional select instruction.
2039 //===----------------------------------------------------------------------===//
2041 defm FCSEL : FPCondSelect<"fcsel">;
2043 // CSEL instructions providing f128 types need to be handled by a
2044 // pseudo-instruction since the eventual code will need to introduce basic
2045 // blocks and control flow.
2046 def F128CSEL : Pseudo<(outs FPR128:$Rd),
2047 (ins FPR128:$Rn, FPR128:$Rm, ccode:$cond),
2048 [(set (f128 FPR128:$Rd),
2049 (ARM64csel FPR128:$Rn, FPR128:$Rm,
2050 (i32 imm:$cond), CPSR))]> {
2052 let usesCustomInserter = 1;
2056 //===----------------------------------------------------------------------===//
2057 // Floating point immediate move.
2058 //===----------------------------------------------------------------------===//
2060 let isReMaterializable = 1 in {
2061 defm FMOV : FPMoveImmediate<"fmov">;
2064 //===----------------------------------------------------------------------===//
2065 // Advanced SIMD two vector instructions.
2066 //===----------------------------------------------------------------------===//
2068 defm ABS : SIMDTwoVectorBHSD<0, 0b01011, "abs", int_arm64_neon_abs>;
2069 defm CLS : SIMDTwoVectorBHS<0, 0b00100, "cls", int_arm64_neon_cls>;
2070 defm CLZ : SIMDTwoVectorBHS<1, 0b00100, "clz", ctlz>;
2071 defm CMEQ : SIMDCmpTwoVector<0, 0b01001, "cmeq", ARM64cmeqz>;
2072 defm CMGE : SIMDCmpTwoVector<1, 0b01000, "cmge", ARM64cmgez>;
2073 defm CMGT : SIMDCmpTwoVector<0, 0b01000, "cmgt", ARM64cmgtz>;
2074 defm CMLE : SIMDCmpTwoVector<1, 0b01001, "cmle", ARM64cmlez>;
2075 defm CMLT : SIMDCmpTwoVector<0, 0b01010, "cmlt", ARM64cmltz>;
2076 defm CNT : SIMDTwoVectorB<0, 0b00, 0b00101, "cnt", ctpop>;
2077 defm FABS : SIMDTwoVectorFP<0, 1, 0b01111, "fabs", fabs>;
2079 defm FCMEQ : SIMDFPCmpTwoVector<0, 1, 0b01101, "fcmeq", ARM64fcmeqz>;
2080 defm FCMGE : SIMDFPCmpTwoVector<1, 1, 0b01100, "fcmge", ARM64fcmgez>;
2081 defm FCMGT : SIMDFPCmpTwoVector<0, 1, 0b01100, "fcmgt", ARM64fcmgtz>;
2082 defm FCMLE : SIMDFPCmpTwoVector<1, 1, 0b01101, "fcmle", ARM64fcmlez>;
2083 defm FCMLT : SIMDFPCmpTwoVector<0, 1, 0b01110, "fcmlt", ARM64fcmltz>;
2084 defm FCVTAS : SIMDTwoVectorFPToInt<0,0,0b11100, "fcvtas",int_arm64_neon_fcvtas>;
2085 defm FCVTAU : SIMDTwoVectorFPToInt<1,0,0b11100, "fcvtau",int_arm64_neon_fcvtau>;
2086 defm FCVTL : SIMDFPWidenTwoVector<0, 0, 0b10111, "fcvtl">;
2087 def : Pat<(v4f32 (int_arm64_neon_vcvthf2fp (v4i16 V64:$Rn))),
2088 (FCVTLv4i16 V64:$Rn)>;
2089 def : Pat<(v4f32 (int_arm64_neon_vcvthf2fp (extract_subvector (v8i16 V128:$Rn),
2091 (FCVTLv8i16 V128:$Rn)>;
2092 def : Pat<(v2f64 (fextend (v2f32 V64:$Rn))), (FCVTLv2i32 V64:$Rn)>;
2093 def : Pat<(v2f64 (fextend (v2f32 (extract_subvector (v4f32 V128:$Rn),
2095 (FCVTLv4i32 V128:$Rn)>;
2097 defm FCVTMS : SIMDTwoVectorFPToInt<0,0,0b11011, "fcvtms",int_arm64_neon_fcvtms>;
2098 defm FCVTMU : SIMDTwoVectorFPToInt<1,0,0b11011, "fcvtmu",int_arm64_neon_fcvtmu>;
2099 defm FCVTNS : SIMDTwoVectorFPToInt<0,0,0b11010, "fcvtns",int_arm64_neon_fcvtns>;
2100 defm FCVTNU : SIMDTwoVectorFPToInt<1,0,0b11010, "fcvtnu",int_arm64_neon_fcvtnu>;
2101 defm FCVTN : SIMDFPNarrowTwoVector<0, 0, 0b10110, "fcvtn">;
2102 def : Pat<(v4i16 (int_arm64_neon_vcvtfp2hf (v4f32 V128:$Rn))),
2103 (FCVTNv4i16 V128:$Rn)>;
2104 def : Pat<(concat_vectors V64:$Rd,
2105 (v4i16 (int_arm64_neon_vcvtfp2hf (v4f32 V128:$Rn)))),
2106 (FCVTNv8i16 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
2107 def : Pat<(v2f32 (fround (v2f64 V128:$Rn))), (FCVTNv2i32 V128:$Rn)>;
2108 def : Pat<(concat_vectors V64:$Rd, (v2f32 (fround (v2f64 V128:$Rn)))),
2109 (FCVTNv4i32 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
2110 defm FCVTPS : SIMDTwoVectorFPToInt<0,1,0b11010, "fcvtps",int_arm64_neon_fcvtps>;
2111 defm FCVTPU : SIMDTwoVectorFPToInt<1,1,0b11010, "fcvtpu",int_arm64_neon_fcvtpu>;
2112 defm FCVTXN : SIMDFPInexactCvtTwoVector<1, 0, 0b10110, "fcvtxn",
2113 int_arm64_neon_fcvtxn>;
2114 defm FCVTZS : SIMDTwoVectorFPToInt<0, 1, 0b11011, "fcvtzs", fp_to_sint>;
2115 defm FCVTZU : SIMDTwoVectorFPToInt<1, 1, 0b11011, "fcvtzu", fp_to_uint>;
2116 let isCodeGenOnly = 1 in {
2117 defm FCVTZS_Int : SIMDTwoVectorFPToInt<0, 1, 0b11011, "fcvtzs",
2118 int_arm64_neon_fcvtzs>;
2119 defm FCVTZU_Int : SIMDTwoVectorFPToInt<1, 1, 0b11011, "fcvtzu",
2120 int_arm64_neon_fcvtzu>;
2122 defm FNEG : SIMDTwoVectorFP<1, 1, 0b01111, "fneg", fneg>;
2123 defm FRECPE : SIMDTwoVectorFP<0, 1, 0b11101, "frecpe", int_arm64_neon_frecpe>;
2124 defm FRINTA : SIMDTwoVectorFP<1, 0, 0b11000, "frinta", frnd>;
2125 defm FRINTI : SIMDTwoVectorFP<1, 1, 0b11001, "frinti", fnearbyint>;
2126 defm FRINTM : SIMDTwoVectorFP<0, 0, 0b11001, "frintm", ffloor>;
2127 defm FRINTN : SIMDTwoVectorFP<0, 0, 0b11000, "frintn", int_arm64_neon_frintn>;
2128 defm FRINTP : SIMDTwoVectorFP<0, 1, 0b11000, "frintp", fceil>;
2129 defm FRINTX : SIMDTwoVectorFP<1, 0, 0b11001, "frintx", frint>;
2130 defm FRINTZ : SIMDTwoVectorFP<0, 1, 0b11001, "frintz", ftrunc>;
2131 defm FRSQRTE: SIMDTwoVectorFP<1, 1, 0b11101, "frsqrte", int_arm64_neon_frsqrte>;
2132 defm FSQRT : SIMDTwoVectorFP<1, 1, 0b11111, "fsqrt", fsqrt>;
2133 defm NEG : SIMDTwoVectorBHSD<1, 0b01011, "neg",
2134 UnOpFrag<(sub immAllZerosV, node:$LHS)> >;
2135 defm NOT : SIMDTwoVectorB<1, 0b00, 0b00101, "not", vnot>;
2136 // Aliases for MVN -> NOT.
2137 def : InstAlias<"mvn.8b $Vd, $Vn", (NOTv8i8 V64:$Vd, V64:$Vn)>;
2138 def : InstAlias<"mvn.16b $Vd, $Vn", (NOTv16i8 V128:$Vd, V128:$Vn)>;
2139 def : InstAlias<"mvn $Vd.8b, $Vn.8b", (NOTv8i8 V64:$Vd, V64:$Vn)>;
2140 def : InstAlias<"mvn $Vd.16b, $Vn.16b", (NOTv16i8 V128:$Vd, V128:$Vn)>;
2142 def : Pat<(ARM64neg (v8i8 V64:$Rn)), (NEGv8i8 V64:$Rn)>;
2143 def : Pat<(ARM64neg (v16i8 V128:$Rn)), (NEGv16i8 V128:$Rn)>;
2144 def : Pat<(ARM64neg (v4i16 V64:$Rn)), (NEGv4i16 V64:$Rn)>;
2145 def : Pat<(ARM64neg (v8i16 V128:$Rn)), (NEGv8i16 V128:$Rn)>;
2146 def : Pat<(ARM64neg (v2i32 V64:$Rn)), (NEGv2i32 V64:$Rn)>;
2147 def : Pat<(ARM64neg (v4i32 V128:$Rn)), (NEGv4i32 V128:$Rn)>;
2148 def : Pat<(ARM64neg (v2i64 V128:$Rn)), (NEGv2i64 V128:$Rn)>;
2150 def : Pat<(ARM64not (v8i8 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2151 def : Pat<(ARM64not (v16i8 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2152 def : Pat<(ARM64not (v4i16 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2153 def : Pat<(ARM64not (v8i16 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2154 def : Pat<(ARM64not (v2i32 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2155 def : Pat<(ARM64not (v1i64 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2156 def : Pat<(ARM64not (v4i32 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2157 def : Pat<(ARM64not (v2i64 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2159 def : Pat<(vnot (v4i16 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2160 def : Pat<(vnot (v8i16 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2161 def : Pat<(vnot (v2i32 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2162 def : Pat<(vnot (v4i32 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2163 def : Pat<(vnot (v2i64 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2165 defm RBIT : SIMDTwoVectorB<1, 0b01, 0b00101, "rbit", int_arm64_neon_rbit>;
2166 defm REV16 : SIMDTwoVectorB<0, 0b00, 0b00001, "rev16", ARM64rev16>;
2167 defm REV32 : SIMDTwoVectorBH<1, 0b00000, "rev32", ARM64rev32>;
2168 defm REV64 : SIMDTwoVectorBHS<0, 0b00000, "rev64", ARM64rev64>;
2169 defm SADALP : SIMDLongTwoVectorTied<0, 0b00110, "sadalp",
2170 BinOpFrag<(add node:$LHS, (int_arm64_neon_saddlp node:$RHS))> >;
2171 defm SADDLP : SIMDLongTwoVector<0, 0b00010, "saddlp", int_arm64_neon_saddlp>;
2172 defm SCVTF : SIMDTwoVectorIntToFP<0, 0, 0b11101, "scvtf", sint_to_fp>;
2173 defm SHLL : SIMDVectorLShiftLongBySizeBHS;
2174 defm SQABS : SIMDTwoVectorBHSD<0, 0b00111, "sqabs", int_arm64_neon_sqabs>;
2175 defm SQNEG : SIMDTwoVectorBHSD<1, 0b00111, "sqneg", int_arm64_neon_sqneg>;
2176 defm SQXTN : SIMDMixedTwoVector<0, 0b10100, "sqxtn", int_arm64_neon_sqxtn>;
2177 defm SQXTUN : SIMDMixedTwoVector<1, 0b10010, "sqxtun", int_arm64_neon_sqxtun>;
2178 defm SUQADD : SIMDTwoVectorBHSDTied<0, 0b00011, "suqadd",int_arm64_neon_suqadd>;
2179 defm UADALP : SIMDLongTwoVectorTied<1, 0b00110, "uadalp",
2180 BinOpFrag<(add node:$LHS, (int_arm64_neon_uaddlp node:$RHS))> >;
2181 defm UADDLP : SIMDLongTwoVector<1, 0b00010, "uaddlp",
2182 int_arm64_neon_uaddlp>;
2183 defm UCVTF : SIMDTwoVectorIntToFP<1, 0, 0b11101, "ucvtf", uint_to_fp>;
2184 defm UQXTN : SIMDMixedTwoVector<1, 0b10100, "uqxtn", int_arm64_neon_uqxtn>;
2185 defm URECPE : SIMDTwoVectorS<0, 1, 0b11100, "urecpe", int_arm64_neon_urecpe>;
2186 defm URSQRTE: SIMDTwoVectorS<1, 1, 0b11100, "ursqrte", int_arm64_neon_ursqrte>;
2187 defm USQADD : SIMDTwoVectorBHSDTied<1, 0b00011, "usqadd",int_arm64_neon_usqadd>;
2188 defm XTN : SIMDMixedTwoVector<0, 0b10010, "xtn", trunc>;
2190 def : Pat<(v2f32 (ARM64rev64 V64:$Rn)), (REV64v2i32 V64:$Rn)>;
2191 def : Pat<(v4f32 (ARM64rev64 V128:$Rn)), (REV64v4i32 V128:$Rn)>;
2193 // Patterns for vector long shift (by element width). These need to match all
2194 // three of zext, sext and anyext so it's easier to pull the patterns out of the
2196 multiclass SIMDVectorLShiftLongBySizeBHSPats<SDPatternOperator ext> {
2197 def : Pat<(ARM64vshl (v8i16 (ext (v8i8 V64:$Rn))), (i32 8)),
2198 (SHLLv8i8 V64:$Rn)>;
2199 def : Pat<(ARM64vshl (v8i16 (ext (extract_high_v16i8 V128:$Rn))), (i32 8)),
2200 (SHLLv16i8 V128:$Rn)>;
2201 def : Pat<(ARM64vshl (v4i32 (ext (v4i16 V64:$Rn))), (i32 16)),
2202 (SHLLv4i16 V64:$Rn)>;
2203 def : Pat<(ARM64vshl (v4i32 (ext (extract_high_v8i16 V128:$Rn))), (i32 16)),
2204 (SHLLv8i16 V128:$Rn)>;
2205 def : Pat<(ARM64vshl (v2i64 (ext (v2i32 V64:$Rn))), (i32 32)),
2206 (SHLLv2i32 V64:$Rn)>;
2207 def : Pat<(ARM64vshl (v2i64 (ext (extract_high_v4i32 V128:$Rn))), (i32 32)),
2208 (SHLLv4i32 V128:$Rn)>;
2211 defm : SIMDVectorLShiftLongBySizeBHSPats<anyext>;
2212 defm : SIMDVectorLShiftLongBySizeBHSPats<zext>;
2213 defm : SIMDVectorLShiftLongBySizeBHSPats<sext>;
2215 //===----------------------------------------------------------------------===//
2216 // Advanced SIMD three vector instructions.
2217 //===----------------------------------------------------------------------===//
2219 defm ADD : SIMDThreeSameVector<0, 0b10000, "add", add>;
2220 defm ADDP : SIMDThreeSameVector<0, 0b10111, "addp", int_arm64_neon_addp>;
2221 defm CMEQ : SIMDThreeSameVector<1, 0b10001, "cmeq", ARM64cmeq>;
2222 defm CMGE : SIMDThreeSameVector<0, 0b00111, "cmge", ARM64cmge>;
2223 defm CMGT : SIMDThreeSameVector<0, 0b00110, "cmgt", ARM64cmgt>;
2224 defm CMHI : SIMDThreeSameVector<1, 0b00110, "cmhi", ARM64cmhi>;
2225 defm CMHS : SIMDThreeSameVector<1, 0b00111, "cmhs", ARM64cmhs>;
2226 defm CMTST : SIMDThreeSameVector<0, 0b10001, "cmtst", ARM64cmtst>;
2227 defm FABD : SIMDThreeSameVectorFP<1,1,0b11010,"fabd", int_arm64_neon_fabd>;
2228 defm FACGE : SIMDThreeSameVectorFPCmp<1,0,0b11101,"facge",int_arm64_neon_facge>;
2229 defm FACGT : SIMDThreeSameVectorFPCmp<1,1,0b11101,"facgt",int_arm64_neon_facgt>;
2230 defm FADDP : SIMDThreeSameVectorFP<1,0,0b11010,"faddp",int_arm64_neon_addp>;
2231 defm FADD : SIMDThreeSameVectorFP<0,0,0b11010,"fadd", fadd>;
2232 defm FCMEQ : SIMDThreeSameVectorFPCmp<0, 0, 0b11100, "fcmeq", ARM64fcmeq>;
2233 defm FCMGE : SIMDThreeSameVectorFPCmp<1, 0, 0b11100, "fcmge", ARM64fcmge>;
2234 defm FCMGT : SIMDThreeSameVectorFPCmp<1, 1, 0b11100, "fcmgt", ARM64fcmgt>;
2235 defm FDIV : SIMDThreeSameVectorFP<1,0,0b11111,"fdiv", fdiv>;
2236 defm FMAXNMP : SIMDThreeSameVectorFP<1,0,0b11000,"fmaxnmp", int_arm64_neon_fmaxnmp>;
2237 defm FMAXNM : SIMDThreeSameVectorFP<0,0,0b11000,"fmaxnm", int_arm64_neon_fmaxnm>;
2238 defm FMAXP : SIMDThreeSameVectorFP<1,0,0b11110,"fmaxp", int_arm64_neon_fmaxp>;
2239 defm FMAX : SIMDThreeSameVectorFP<0,0,0b11110,"fmax", ARM64fmax>;
2240 defm FMINNMP : SIMDThreeSameVectorFP<1,1,0b11000,"fminnmp", int_arm64_neon_fminnmp>;
2241 defm FMINNM : SIMDThreeSameVectorFP<0,1,0b11000,"fminnm", int_arm64_neon_fminnm>;
2242 defm FMINP : SIMDThreeSameVectorFP<1,1,0b11110,"fminp", int_arm64_neon_fminp>;
2243 defm FMIN : SIMDThreeSameVectorFP<0,1,0b11110,"fmin", ARM64fmin>;
2245 // NOTE: The operands of the PatFrag are reordered on FMLA/FMLS because the
2246 // instruction expects the addend first, while the fma intrinsic puts it last.
2247 defm FMLA : SIMDThreeSameVectorFPTied<0, 0, 0b11001, "fmla",
2248 TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)> >;
2249 defm FMLS : SIMDThreeSameVectorFPTied<0, 1, 0b11001, "fmls",
2250 TriOpFrag<(fma node:$MHS, (fneg node:$RHS), node:$LHS)> >;
2252 // The following def pats catch the case where the LHS of an FMA is negated.
2253 // The TriOpFrag above catches the case where the middle operand is negated.
2254 def : Pat<(v2f32 (fma (fneg V64:$Rn), V64:$Rm, V64:$Rd)),
2255 (FMLSv2f32 V64:$Rd, V64:$Rn, V64:$Rm)>;
2257 def : Pat<(v4f32 (fma (fneg V128:$Rn), V128:$Rm, V128:$Rd)),
2258 (FMLSv4f32 V128:$Rd, V128:$Rn, V128:$Rm)>;
2260 def : Pat<(v2f64 (fma (fneg V128:$Rn), V128:$Rm, V128:$Rd)),
2261 (FMLSv2f64 V128:$Rd, V128:$Rn, V128:$Rm)>;
2263 defm FMULX : SIMDThreeSameVectorFP<0,0,0b11011,"fmulx", int_arm64_neon_fmulx>;
2264 defm FMUL : SIMDThreeSameVectorFP<1,0,0b11011,"fmul", fmul>;
2265 defm FRECPS : SIMDThreeSameVectorFP<0,0,0b11111,"frecps", int_arm64_neon_frecps>;
2266 defm FRSQRTS : SIMDThreeSameVectorFP<0,1,0b11111,"frsqrts", int_arm64_neon_frsqrts>;
2267 defm FSUB : SIMDThreeSameVectorFP<0,1,0b11010,"fsub", fsub>;
2268 defm MLA : SIMDThreeSameVectorBHSTied<0, 0b10010, "mla",
2269 TriOpFrag<(add node:$LHS, (mul node:$MHS, node:$RHS))> >;
2270 defm MLS : SIMDThreeSameVectorBHSTied<1, 0b10010, "mls",
2271 TriOpFrag<(sub node:$LHS, (mul node:$MHS, node:$RHS))> >;
2272 defm MUL : SIMDThreeSameVectorBHS<0, 0b10011, "mul", mul>;
2273 defm PMUL : SIMDThreeSameVectorB<1, 0b10011, "pmul", int_arm64_neon_pmul>;
2274 defm SABA : SIMDThreeSameVectorBHSTied<0, 0b01111, "saba",
2275 TriOpFrag<(add node:$LHS, (int_arm64_neon_sabd node:$MHS, node:$RHS))> >;
2276 defm SABD : SIMDThreeSameVectorBHS<0,0b01110,"sabd", int_arm64_neon_sabd>;
2277 defm SHADD : SIMDThreeSameVectorBHS<0,0b00000,"shadd", int_arm64_neon_shadd>;
2278 defm SHSUB : SIMDThreeSameVectorBHS<0,0b00100,"shsub", int_arm64_neon_shsub>;
2279 defm SMAXP : SIMDThreeSameVectorBHS<0,0b10100,"smaxp", int_arm64_neon_smaxp>;
2280 defm SMAX : SIMDThreeSameVectorBHS<0,0b01100,"smax", int_arm64_neon_smax>;
2281 defm SMINP : SIMDThreeSameVectorBHS<0,0b10101,"sminp", int_arm64_neon_sminp>;
2282 defm SMIN : SIMDThreeSameVectorBHS<0,0b01101,"smin", int_arm64_neon_smin>;
2283 defm SQADD : SIMDThreeSameVector<0,0b00001,"sqadd", int_arm64_neon_sqadd>;
2284 defm SQDMULH : SIMDThreeSameVectorHS<0,0b10110,"sqdmulh",int_arm64_neon_sqdmulh>;
2285 defm SQRDMULH : SIMDThreeSameVectorHS<1,0b10110,"sqrdmulh",int_arm64_neon_sqrdmulh>;
2286 defm SQRSHL : SIMDThreeSameVector<0,0b01011,"sqrshl", int_arm64_neon_sqrshl>;
2287 defm SQSHL : SIMDThreeSameVector<0,0b01001,"sqshl", int_arm64_neon_sqshl>;
2288 defm SQSUB : SIMDThreeSameVector<0,0b00101,"sqsub", int_arm64_neon_sqsub>;
2289 defm SRHADD : SIMDThreeSameVectorBHS<0,0b00010,"srhadd",int_arm64_neon_srhadd>;
2290 defm SRSHL : SIMDThreeSameVector<0,0b01010,"srshl", int_arm64_neon_srshl>;
2291 defm SSHL : SIMDThreeSameVector<0,0b01000,"sshl", int_arm64_neon_sshl>;
2292 defm SUB : SIMDThreeSameVector<1,0b10000,"sub", sub>;
2293 defm UABA : SIMDThreeSameVectorBHSTied<1, 0b01111, "uaba",
2294 TriOpFrag<(add node:$LHS, (int_arm64_neon_uabd node:$MHS, node:$RHS))> >;
2295 defm UABD : SIMDThreeSameVectorBHS<1,0b01110,"uabd", int_arm64_neon_uabd>;
2296 defm UHADD : SIMDThreeSameVectorBHS<1,0b00000,"uhadd", int_arm64_neon_uhadd>;
2297 defm UHSUB : SIMDThreeSameVectorBHS<1,0b00100,"uhsub", int_arm64_neon_uhsub>;
2298 defm UMAXP : SIMDThreeSameVectorBHS<1,0b10100,"umaxp", int_arm64_neon_umaxp>;
2299 defm UMAX : SIMDThreeSameVectorBHS<1,0b01100,"umax", int_arm64_neon_umax>;
2300 defm UMINP : SIMDThreeSameVectorBHS<1,0b10101,"uminp", int_arm64_neon_uminp>;
2301 defm UMIN : SIMDThreeSameVectorBHS<1,0b01101,"umin", int_arm64_neon_umin>;
2302 defm UQADD : SIMDThreeSameVector<1,0b00001,"uqadd", int_arm64_neon_uqadd>;
2303 defm UQRSHL : SIMDThreeSameVector<1,0b01011,"uqrshl", int_arm64_neon_uqrshl>;
2304 defm UQSHL : SIMDThreeSameVector<1,0b01001,"uqshl", int_arm64_neon_uqshl>;
2305 defm UQSUB : SIMDThreeSameVector<1,0b00101,"uqsub", int_arm64_neon_uqsub>;
2306 defm URHADD : SIMDThreeSameVectorBHS<1,0b00010,"urhadd", int_arm64_neon_urhadd>;
2307 defm URSHL : SIMDThreeSameVector<1,0b01010,"urshl", int_arm64_neon_urshl>;
2308 defm USHL : SIMDThreeSameVector<1,0b01000,"ushl", int_arm64_neon_ushl>;
2310 defm AND : SIMDLogicalThreeVector<0, 0b00, "and", and>;
2311 defm BIC : SIMDLogicalThreeVector<0, 0b01, "bic",
2312 BinOpFrag<(and node:$LHS, (vnot node:$RHS))> >;
2313 defm BIF : SIMDLogicalThreeVector<1, 0b11, "bif">;
2314 defm BIT : SIMDLogicalThreeVectorTied<1, 0b10, "bit", ARM64bit>;
2315 defm BSL : SIMDLogicalThreeVectorTied<1, 0b01, "bsl",
2316 TriOpFrag<(or (and node:$LHS, node:$MHS), (and (vnot node:$LHS), node:$RHS))>>;
2317 defm EOR : SIMDLogicalThreeVector<1, 0b00, "eor", xor>;
2318 defm ORN : SIMDLogicalThreeVector<0, 0b11, "orn",
2319 BinOpFrag<(or node:$LHS, (vnot node:$RHS))> >;
2320 defm ORR : SIMDLogicalThreeVector<0, 0b10, "orr", or>;
2322 // FIXME: the .16b and .8b variantes should be emitted by the
2323 // AsmWriter. TableGen's AsmWriter-generator doesn't deal with variant syntaxes
2324 // in aliases yet though.
2325 def : InstAlias<"mov{\t$dst.16b, $src.16b|.16b\t$dst, $src}",
2326 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2327 def : InstAlias<"{mov\t$dst.8h, $src.8h|mov.8h\t$dst, $src}",
2328 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2329 def : InstAlias<"{mov\t$dst.4s, $src.4s|mov.4s\t$dst, $src}",
2330 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2331 def : InstAlias<"{mov\t$dst.2d, $src.2d|mov.2d\t$dst, $src}",
2332 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2334 def : InstAlias<"{mov\t$dst.8b, $src.8b|mov.8b\t$dst, $src}",
2335 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2336 def : InstAlias<"{mov\t$dst.4h, $src.4h|mov.4h\t$dst, $src}",
2337 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2338 def : InstAlias<"{mov\t$dst.2s, $src.2s|mov.2s\t$dst, $src}",
2339 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2340 def : InstAlias<"{mov\t$dst.1d, $src.1d|mov.1d\t$dst, $src}",
2341 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2343 def : InstAlias<"{cmls\t$dst.8b, $src1.8b, $src2.8b" #
2344 "|cmls.8b\t$dst, $src1, $src2}",
2345 (CMHSv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2346 def : InstAlias<"{cmls\t$dst.16b, $src1.16b, $src2.16b" #
2347 "|cmls.16b\t$dst, $src1, $src2}",
2348 (CMHSv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2349 def : InstAlias<"{cmls\t$dst.4h, $src1.4h, $src2.4h" #
2350 "|cmls.4h\t$dst, $src1, $src2}",
2351 (CMHSv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2352 def : InstAlias<"{cmls\t$dst.8h, $src1.8h, $src2.8h" #
2353 "|cmls.8h\t$dst, $src1, $src2}",
2354 (CMHSv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2355 def : InstAlias<"{cmls\t$dst.2s, $src1.2s, $src2.2s" #
2356 "|cmls.2s\t$dst, $src1, $src2}",
2357 (CMHSv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2358 def : InstAlias<"{cmls\t$dst.4s, $src1.4s, $src2.4s" #
2359 "|cmls.4s\t$dst, $src1, $src2}",
2360 (CMHSv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2361 def : InstAlias<"{cmls\t$dst.2d, $src1.2d, $src2.2d" #
2362 "|cmls.2d\t$dst, $src1, $src2}",
2363 (CMHSv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2365 def : InstAlias<"{cmlo\t$dst.8b, $src1.8b, $src2.8b" #
2366 "|cmlo.8b\t$dst, $src1, $src2}",
2367 (CMHIv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2368 def : InstAlias<"{cmlo\t$dst.16b, $src1.16b, $src2.16b" #
2369 "|cmlo.16b\t$dst, $src1, $src2}",
2370 (CMHIv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2371 def : InstAlias<"{cmlo\t$dst.4h, $src1.4h, $src2.4h" #
2372 "|cmlo.4h\t$dst, $src1, $src2}",
2373 (CMHIv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2374 def : InstAlias<"{cmlo\t$dst.8h, $src1.8h, $src2.8h" #
2375 "|cmlo.8h\t$dst, $src1, $src2}",
2376 (CMHIv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2377 def : InstAlias<"{cmlo\t$dst.2s, $src1.2s, $src2.2s" #
2378 "|cmlo.2s\t$dst, $src1, $src2}",
2379 (CMHIv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2380 def : InstAlias<"{cmlo\t$dst.4s, $src1.4s, $src2.4s" #
2381 "|cmlo.4s\t$dst, $src1, $src2}",
2382 (CMHIv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2383 def : InstAlias<"{cmlo\t$dst.2d, $src1.2d, $src2.2d" #
2384 "|cmlo.2d\t$dst, $src1, $src2}",
2385 (CMHIv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2387 def : InstAlias<"{cmle\t$dst.8b, $src1.8b, $src2.8b" #
2388 "|cmle.8b\t$dst, $src1, $src2}",
2389 (CMGEv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2390 def : InstAlias<"{cmle\t$dst.16b, $src1.16b, $src2.16b" #
2391 "|cmle.16b\t$dst, $src1, $src2}",
2392 (CMGEv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2393 def : InstAlias<"{cmle\t$dst.4h, $src1.4h, $src2.4h" #
2394 "|cmle.4h\t$dst, $src1, $src2}",
2395 (CMGEv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2396 def : InstAlias<"{cmle\t$dst.8h, $src1.8h, $src2.8h" #
2397 "|cmle.8h\t$dst, $src1, $src2}",
2398 (CMGEv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2399 def : InstAlias<"{cmle\t$dst.2s, $src1.2s, $src2.2s" #
2400 "|cmle.2s\t$dst, $src1, $src2}",
2401 (CMGEv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2402 def : InstAlias<"{cmle\t$dst.4s, $src1.4s, $src2.4s" #
2403 "|cmle.4s\t$dst, $src1, $src2}",
2404 (CMGEv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2405 def : InstAlias<"{cmle\t$dst.2d, $src1.2d, $src2.2d" #
2406 "|cmle.2d\t$dst, $src1, $src2}",
2407 (CMGEv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2409 def : InstAlias<"{cmlt\t$dst.8b, $src1.8b, $src2.8b" #
2410 "|cmlt.8b\t$dst, $src1, $src2}",
2411 (CMGTv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2412 def : InstAlias<"{cmlt\t$dst.16b, $src1.16b, $src2.16b" #
2413 "|cmlt.16b\t$dst, $src1, $src2}",
2414 (CMGTv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2415 def : InstAlias<"{cmlt\t$dst.4h, $src1.4h, $src2.4h" #
2416 "|cmlt.4h\t$dst, $src1, $src2}",
2417 (CMGTv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2418 def : InstAlias<"{cmlt\t$dst.8h, $src1.8h, $src2.8h" #
2419 "|cmlt.8h\t$dst, $src1, $src2}",
2420 (CMGTv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2421 def : InstAlias<"{cmlt\t$dst.2s, $src1.2s, $src2.2s" #
2422 "|cmlt.2s\t$dst, $src1, $src2}",
2423 (CMGTv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2424 def : InstAlias<"{cmlt\t$dst.4s, $src1.4s, $src2.4s" #
2425 "|cmlt.4s\t$dst, $src1, $src2}",
2426 (CMGTv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2427 def : InstAlias<"{cmlt\t$dst.2d, $src1.2d, $src2.2d" #
2428 "|cmlt.2d\t$dst, $src1, $src2}",
2429 (CMGTv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2431 def : InstAlias<"{fcmle\t$dst.2s, $src1.2s, $src2.2s" #
2432 "|fcmle.2s\t$dst, $src1, $src2}",
2433 (FCMGEv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
2434 def : InstAlias<"{fcmle\t$dst.4s, $src1.4s, $src2.4s" #
2435 "|fcmle.4s\t$dst, $src1, $src2}",
2436 (FCMGEv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
2437 def : InstAlias<"{fcmle\t$dst.2d, $src1.2d, $src2.2d" #
2438 "|fcmle.2d\t$dst, $src1, $src2}",
2439 (FCMGEv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
2441 def : InstAlias<"{fcmlt\t$dst.2s, $src1.2s, $src2.2s" #
2442 "|fcmlt.2s\t$dst, $src1, $src2}",
2443 (FCMGTv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
2444 def : InstAlias<"{fcmlt\t$dst.4s, $src1.4s, $src2.4s" #
2445 "|fcmlt.4s\t$dst, $src1, $src2}",
2446 (FCMGTv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
2447 def : InstAlias<"{fcmlt\t$dst.2d, $src1.2d, $src2.2d" #
2448 "|fcmlt.2d\t$dst, $src1, $src2}",
2449 (FCMGTv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
2451 def : InstAlias<"{facle\t$dst.2s, $src1.2s, $src2.2s" #
2452 "|facle.2s\t$dst, $src1, $src2}",
2453 (FACGEv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
2454 def : InstAlias<"{facle\t$dst.4s, $src1.4s, $src2.4s" #
2455 "|facle.4s\t$dst, $src1, $src2}",
2456 (FACGEv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
2457 def : InstAlias<"{facle\t$dst.2d, $src1.2d, $src2.2d" #
2458 "|facle.2d\t$dst, $src1, $src2}",
2459 (FACGEv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
2461 def : InstAlias<"{faclt\t$dst.2s, $src1.2s, $src2.2s" #
2462 "|faclt.2s\t$dst, $src1, $src2}",
2463 (FACGTv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
2464 def : InstAlias<"{faclt\t$dst.4s, $src1.4s, $src2.4s" #
2465 "|faclt.4s\t$dst, $src1, $src2}",
2466 (FACGTv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
2467 def : InstAlias<"{faclt\t$dst.2d, $src1.2d, $src2.2d" #
2468 "|faclt.2d\t$dst, $src1, $src2}",
2469 (FACGTv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
2471 //===----------------------------------------------------------------------===//
2472 // Advanced SIMD three scalar instructions.
2473 //===----------------------------------------------------------------------===//
2475 defm ADD : SIMDThreeScalarD<0, 0b10000, "add", add>;
2476 defm CMEQ : SIMDThreeScalarD<1, 0b10001, "cmeq", ARM64cmeq>;
2477 defm CMGE : SIMDThreeScalarD<0, 0b00111, "cmge", ARM64cmge>;
2478 defm CMGT : SIMDThreeScalarD<0, 0b00110, "cmgt", ARM64cmgt>;
2479 defm CMHI : SIMDThreeScalarD<1, 0b00110, "cmhi", ARM64cmhi>;
2480 defm CMHS : SIMDThreeScalarD<1, 0b00111, "cmhs", ARM64cmhs>;
2481 defm CMTST : SIMDThreeScalarD<0, 0b10001, "cmtst", ARM64cmtst>;
2482 defm FABD : SIMDThreeScalarSD<1, 1, 0b11010, "fabd", int_arm64_sisd_fabd>;
2483 def : Pat<(v1f64 (int_arm64_neon_fabd (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2484 (FABD64 FPR64:$Rn, FPR64:$Rm)>;
2485 defm FACGE : SIMDThreeScalarFPCmp<1, 0, 0b11101, "facge",
2486 int_arm64_neon_facge>;
2487 defm FACGT : SIMDThreeScalarFPCmp<1, 1, 0b11101, "facgt",
2488 int_arm64_neon_facgt>;
2489 defm FCMEQ : SIMDThreeScalarFPCmp<0, 0, 0b11100, "fcmeq", ARM64fcmeq>;
2490 defm FCMGE : SIMDThreeScalarFPCmp<1, 0, 0b11100, "fcmge", ARM64fcmge>;
2491 defm FCMGT : SIMDThreeScalarFPCmp<1, 1, 0b11100, "fcmgt", ARM64fcmgt>;
2492 defm FMULX : SIMDThreeScalarSD<0, 0, 0b11011, "fmulx", int_arm64_neon_fmulx>;
2493 defm FRECPS : SIMDThreeScalarSD<0, 0, 0b11111, "frecps", int_arm64_neon_frecps>;
2494 defm FRSQRTS : SIMDThreeScalarSD<0, 1, 0b11111, "frsqrts", int_arm64_neon_frsqrts>;
2495 defm SQADD : SIMDThreeScalarBHSD<0, 0b00001, "sqadd", int_arm64_neon_sqadd>;
2496 defm SQDMULH : SIMDThreeScalarHS< 0, 0b10110, "sqdmulh", int_arm64_neon_sqdmulh>;
2497 defm SQRDMULH : SIMDThreeScalarHS< 1, 0b10110, "sqrdmulh", int_arm64_neon_sqrdmulh>;
2498 defm SQRSHL : SIMDThreeScalarBHSD<0, 0b01011, "sqrshl",int_arm64_neon_sqrshl>;
2499 defm SQSHL : SIMDThreeScalarBHSD<0, 0b01001, "sqshl", int_arm64_neon_sqshl>;
2500 defm SQSUB : SIMDThreeScalarBHSD<0, 0b00101, "sqsub", int_arm64_neon_sqsub>;
2501 defm SRSHL : SIMDThreeScalarD< 0, 0b01010, "srshl", int_arm64_neon_srshl>;
2502 defm SSHL : SIMDThreeScalarD< 0, 0b01000, "sshl", int_arm64_neon_sshl>;
2503 defm SUB : SIMDThreeScalarD< 1, 0b10000, "sub", sub>;
2504 defm UQADD : SIMDThreeScalarBHSD<1, 0b00001, "uqadd", int_arm64_neon_uqadd>;
2505 defm UQRSHL : SIMDThreeScalarBHSD<1, 0b01011, "uqrshl",int_arm64_neon_uqrshl>;
2506 defm UQSHL : SIMDThreeScalarBHSD<1, 0b01001, "uqshl", int_arm64_neon_uqshl>;
2507 defm UQSUB : SIMDThreeScalarBHSD<1, 0b00101, "uqsub", int_arm64_neon_uqsub>;
2508 defm URSHL : SIMDThreeScalarD< 1, 0b01010, "urshl", int_arm64_neon_urshl>;
2509 defm USHL : SIMDThreeScalarD< 1, 0b01000, "ushl", int_arm64_neon_ushl>;
2511 def : InstAlias<"cmls $dst, $src1, $src2",
2512 (CMHSv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2513 def : InstAlias<"cmle $dst, $src1, $src2",
2514 (CMGEv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2515 def : InstAlias<"cmlo $dst, $src1, $src2",
2516 (CMHIv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2517 def : InstAlias<"cmlt $dst, $src1, $src2",
2518 (CMGTv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2519 def : InstAlias<"fcmle $dst, $src1, $src2",
2520 (FCMGE32 FPR32:$dst, FPR32:$src2, FPR32:$src1)>;
2521 def : InstAlias<"fcmle $dst, $src1, $src2",
2522 (FCMGE64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2523 def : InstAlias<"fcmlt $dst, $src1, $src2",
2524 (FCMGT32 FPR32:$dst, FPR32:$src2, FPR32:$src1)>;
2525 def : InstAlias<"fcmlt $dst, $src1, $src2",
2526 (FCMGT64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2527 def : InstAlias<"facle $dst, $src1, $src2",
2528 (FACGE32 FPR32:$dst, FPR32:$src2, FPR32:$src1)>;
2529 def : InstAlias<"facle $dst, $src1, $src2",
2530 (FACGE64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2531 def : InstAlias<"faclt $dst, $src1, $src2",
2532 (FACGT32 FPR32:$dst, FPR32:$src2, FPR32:$src1)>;
2533 def : InstAlias<"faclt $dst, $src1, $src2",
2534 (FACGT64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2536 //===----------------------------------------------------------------------===//
2537 // Advanced SIMD three scalar instructions (mixed operands).
2538 //===----------------------------------------------------------------------===//
2539 defm SQDMULL : SIMDThreeScalarMixedHS<0, 0b11010, "sqdmull",
2540 int_arm64_neon_sqdmulls_scalar>;
2541 defm SQDMLAL : SIMDThreeScalarMixedTiedHS<0, 0b10010, "sqdmlal">;
2542 defm SQDMLSL : SIMDThreeScalarMixedTiedHS<0, 0b10110, "sqdmlsl">;
2544 def : Pat<(i64 (int_arm64_neon_sqadd (i64 FPR64:$Rd),
2545 (i64 (int_arm64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
2546 (i32 FPR32:$Rm))))),
2547 (SQDMLALi32 FPR64:$Rd, FPR32:$Rn, FPR32:$Rm)>;
2548 def : Pat<(i64 (int_arm64_neon_sqsub (i64 FPR64:$Rd),
2549 (i64 (int_arm64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
2550 (i32 FPR32:$Rm))))),
2551 (SQDMLSLi32 FPR64:$Rd, FPR32:$Rn, FPR32:$Rm)>;
2553 //===----------------------------------------------------------------------===//
2554 // Advanced SIMD two scalar instructions.
2555 //===----------------------------------------------------------------------===//
2557 defm ABS : SIMDTwoScalarD< 0, 0b01011, "abs", int_arm64_neon_abs>;
2558 defm CMEQ : SIMDCmpTwoScalarD< 0, 0b01001, "cmeq", ARM64cmeqz>;
2559 defm CMGE : SIMDCmpTwoScalarD< 1, 0b01000, "cmge", ARM64cmgez>;
2560 defm CMGT : SIMDCmpTwoScalarD< 0, 0b01000, "cmgt", ARM64cmgtz>;
2561 defm CMLE : SIMDCmpTwoScalarD< 1, 0b01001, "cmle", ARM64cmlez>;
2562 defm CMLT : SIMDCmpTwoScalarD< 0, 0b01010, "cmlt", ARM64cmltz>;
2563 defm FCMEQ : SIMDCmpTwoScalarSD<0, 1, 0b01101, "fcmeq", ARM64fcmeqz>;
2564 defm FCMGE : SIMDCmpTwoScalarSD<1, 1, 0b01100, "fcmge", ARM64fcmgez>;
2565 defm FCMGT : SIMDCmpTwoScalarSD<0, 1, 0b01100, "fcmgt", ARM64fcmgtz>;
2566 defm FCMLE : SIMDCmpTwoScalarSD<1, 1, 0b01101, "fcmle", ARM64fcmlez>;
2567 defm FCMLT : SIMDCmpTwoScalarSD<0, 1, 0b01110, "fcmlt", ARM64fcmltz>;
2568 defm FCVTAS : SIMDTwoScalarSD< 0, 0, 0b11100, "fcvtas">;
2569 defm FCVTAU : SIMDTwoScalarSD< 1, 0, 0b11100, "fcvtau">;
2570 defm FCVTMS : SIMDTwoScalarSD< 0, 0, 0b11011, "fcvtms">;
2571 defm FCVTMU : SIMDTwoScalarSD< 1, 0, 0b11011, "fcvtmu">;
2572 defm FCVTNS : SIMDTwoScalarSD< 0, 0, 0b11010, "fcvtns">;
2573 defm FCVTNU : SIMDTwoScalarSD< 1, 0, 0b11010, "fcvtnu">;
2574 defm FCVTPS : SIMDTwoScalarSD< 0, 1, 0b11010, "fcvtps">;
2575 defm FCVTPU : SIMDTwoScalarSD< 1, 1, 0b11010, "fcvtpu">;
2576 def FCVTXNv1i64 : SIMDInexactCvtTwoScalar<0b10110, "fcvtxn">;
2577 defm FCVTZS : SIMDTwoScalarSD< 0, 1, 0b11011, "fcvtzs">;
2578 defm FCVTZU : SIMDTwoScalarSD< 1, 1, 0b11011, "fcvtzu">;
2579 defm FRECPE : SIMDTwoScalarSD< 0, 1, 0b11101, "frecpe">;
2580 defm FRECPX : SIMDTwoScalarSD< 0, 1, 0b11111, "frecpx">;
2581 defm FRSQRTE : SIMDTwoScalarSD< 1, 1, 0b11101, "frsqrte">;
2582 defm NEG : SIMDTwoScalarD< 1, 0b01011, "neg",
2583 UnOpFrag<(sub immAllZerosV, node:$LHS)> >;
2584 defm SCVTF : SIMDTwoScalarCVTSD< 0, 0, 0b11101, "scvtf", ARM64sitof>;
2585 defm SQABS : SIMDTwoScalarBHSD< 0, 0b00111, "sqabs", int_arm64_neon_sqabs>;
2586 defm SQNEG : SIMDTwoScalarBHSD< 1, 0b00111, "sqneg", int_arm64_neon_sqneg>;
2587 defm SQXTN : SIMDTwoScalarMixedBHS< 0, 0b10100, "sqxtn", int_arm64_neon_scalar_sqxtn>;
2588 defm SQXTUN : SIMDTwoScalarMixedBHS< 1, 0b10010, "sqxtun", int_arm64_neon_scalar_sqxtun>;
2589 defm SUQADD : SIMDTwoScalarBHSDTied< 0, 0b00011, "suqadd",
2590 int_arm64_neon_suqadd>;
2591 defm UCVTF : SIMDTwoScalarCVTSD< 1, 0, 0b11101, "ucvtf", ARM64uitof>;
2592 defm UQXTN : SIMDTwoScalarMixedBHS<1, 0b10100, "uqxtn", int_arm64_neon_scalar_uqxtn>;
2593 defm USQADD : SIMDTwoScalarBHSDTied< 1, 0b00011, "usqadd",
2594 int_arm64_neon_usqadd>;
2596 def : Pat<(v1i64 (int_arm64_neon_fcvtas (v1f64 FPR64:$Rn))),
2597 (FCVTASv1i64 FPR64:$Rn)>;
2598 def : Pat<(v1i64 (int_arm64_neon_fcvtau (v1f64 FPR64:$Rn))),
2599 (FCVTAUv1i64 FPR64:$Rn)>;
2600 def : Pat<(v1i64 (int_arm64_neon_fcvtms (v1f64 FPR64:$Rn))),
2601 (FCVTMSv1i64 FPR64:$Rn)>;
2602 def : Pat<(v1i64 (int_arm64_neon_fcvtmu (v1f64 FPR64:$Rn))),
2603 (FCVTMUv1i64 FPR64:$Rn)>;
2604 def : Pat<(v1i64 (int_arm64_neon_fcvtns (v1f64 FPR64:$Rn))),
2605 (FCVTNSv1i64 FPR64:$Rn)>;
2606 def : Pat<(v1i64 (int_arm64_neon_fcvtnu (v1f64 FPR64:$Rn))),
2607 (FCVTNUv1i64 FPR64:$Rn)>;
2608 def : Pat<(v1i64 (int_arm64_neon_fcvtps (v1f64 FPR64:$Rn))),
2609 (FCVTPSv1i64 FPR64:$Rn)>;
2610 def : Pat<(v1i64 (int_arm64_neon_fcvtpu (v1f64 FPR64:$Rn))),
2611 (FCVTPUv1i64 FPR64:$Rn)>;
2613 def : Pat<(f32 (int_arm64_neon_frecpe (f32 FPR32:$Rn))),
2614 (FRECPEv1i32 FPR32:$Rn)>;
2615 def : Pat<(f64 (int_arm64_neon_frecpe (f64 FPR64:$Rn))),
2616 (FRECPEv1i64 FPR64:$Rn)>;
2617 def : Pat<(v1f64 (int_arm64_neon_frecpe (v1f64 FPR64:$Rn))),
2618 (FRECPEv1i64 FPR64:$Rn)>;
2620 def : Pat<(f32 (int_arm64_neon_frecpx (f32 FPR32:$Rn))),
2621 (FRECPXv1i32 FPR32:$Rn)>;
2622 def : Pat<(f64 (int_arm64_neon_frecpx (f64 FPR64:$Rn))),
2623 (FRECPXv1i64 FPR64:$Rn)>;
2625 def : Pat<(f32 (int_arm64_neon_frsqrte (f32 FPR32:$Rn))),
2626 (FRSQRTEv1i32 FPR32:$Rn)>;
2627 def : Pat<(f64 (int_arm64_neon_frsqrte (f64 FPR64:$Rn))),
2628 (FRSQRTEv1i64 FPR64:$Rn)>;
2629 def : Pat<(v1f64 (int_arm64_neon_frsqrte (v1f64 FPR64:$Rn))),
2630 (FRSQRTEv1i64 FPR64:$Rn)>;
2632 // If an integer is about to be converted to a floating point value,
2633 // just load it on the floating point unit.
2634 // Here are the patterns for 8 and 16-bits to float.
2636 def : Pat <(f32 (uint_to_fp (i32 (zextloadi8 ro_indexed8:$addr)))),
2637 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2638 (LDRBro ro_indexed8:$addr), bsub))>;
2639 def : Pat <(f32 (uint_to_fp (i32 (zextloadi8 am_indexed8:$addr)))),
2640 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2641 (LDRBui am_indexed8:$addr), bsub))>;
2642 def : Pat <(f32 (uint_to_fp (i32 (zextloadi8 am_unscaled8:$addr)))),
2643 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2644 (LDURBi am_unscaled8:$addr), bsub))>;
2645 // 16-bits -> float.
2646 def : Pat <(f32 (uint_to_fp (i32 (zextloadi16 ro_indexed16:$addr)))),
2647 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2648 (LDRHro ro_indexed16:$addr), hsub))>;
2649 def : Pat <(f32 (uint_to_fp (i32 (zextloadi16 am_indexed16:$addr)))),
2650 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2651 (LDRHui am_indexed16:$addr), hsub))>;
2652 def : Pat <(f32 (uint_to_fp (i32 (zextloadi16 am_unscaled16:$addr)))),
2653 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2654 (LDURHi am_unscaled16:$addr), hsub))>;
2655 // 32-bits are handled in target specific dag combine:
2656 // performIntToFpCombine.
2657 // 64-bits integer to 32-bits floating point, not possible with
2658 // UCVTF on floating point registers (both source and destination
2659 // must have the same size).
2661 // Here are the patterns for 8, 16, 32, and 64-bits to double.
2662 // 8-bits -> double.
2663 def : Pat <(f64 (uint_to_fp (i32 (zextloadi8 ro_indexed8:$addr)))),
2664 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2665 (LDRBro ro_indexed8:$addr), bsub))>;
2666 def : Pat <(f64 (uint_to_fp (i32 (zextloadi8 am_indexed8:$addr)))),
2667 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2668 (LDRBui am_indexed8:$addr), bsub))>;
2669 def : Pat <(f64 (uint_to_fp (i32 (zextloadi8 am_unscaled8:$addr)))),
2670 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2671 (LDURBi am_unscaled8:$addr), bsub))>;
2672 // 16-bits -> double.
2673 def : Pat <(f64 (uint_to_fp (i32 (zextloadi16 ro_indexed16:$addr)))),
2674 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2675 (LDRHro ro_indexed16:$addr), hsub))>;
2676 def : Pat <(f64 (uint_to_fp (i32 (zextloadi16 am_indexed16:$addr)))),
2677 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2678 (LDRHui am_indexed16:$addr), hsub))>;
2679 def : Pat <(f64 (uint_to_fp (i32 (zextloadi16 am_unscaled16:$addr)))),
2680 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2681 (LDURHi am_unscaled16:$addr), hsub))>;
2682 // 32-bits -> double.
2683 def : Pat <(f64 (uint_to_fp (i32 (load ro_indexed32:$addr)))),
2684 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2685 (LDRSro ro_indexed32:$addr), ssub))>;
2686 def : Pat <(f64 (uint_to_fp (i32 (load am_indexed32:$addr)))),
2687 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2688 (LDRSui am_indexed32:$addr), ssub))>;
2689 def : Pat <(f64 (uint_to_fp (i32 (load am_unscaled32:$addr)))),
2690 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2691 (LDURSi am_unscaled32:$addr), ssub))>;
2692 // 64-bits -> double are handled in target specific dag combine:
2693 // performIntToFpCombine.
2695 //===----------------------------------------------------------------------===//
2696 // Advanced SIMD three different-sized vector instructions.
2697 //===----------------------------------------------------------------------===//
2699 defm ADDHN : SIMDNarrowThreeVectorBHS<0,0b0100,"addhn", int_arm64_neon_addhn>;
2700 defm SUBHN : SIMDNarrowThreeVectorBHS<0,0b0110,"subhn", int_arm64_neon_subhn>;
2701 defm RADDHN : SIMDNarrowThreeVectorBHS<1,0b0100,"raddhn",int_arm64_neon_raddhn>;
2702 defm RSUBHN : SIMDNarrowThreeVectorBHS<1,0b0110,"rsubhn",int_arm64_neon_rsubhn>;
2703 defm PMULL : SIMDDifferentThreeVectorBD<0,0b1110,"pmull",int_arm64_neon_pmull>;
2704 defm SABAL : SIMDLongThreeVectorTiedBHSabal<0,0b0101,"sabal",
2705 int_arm64_neon_sabd>;
2706 defm SABDL : SIMDLongThreeVectorBHSabdl<0, 0b0111, "sabdl",
2707 int_arm64_neon_sabd>;
2708 defm SADDL : SIMDLongThreeVectorBHS< 0, 0b0000, "saddl",
2709 BinOpFrag<(add (sext node:$LHS), (sext node:$RHS))>>;
2710 defm SADDW : SIMDWideThreeVectorBHS< 0, 0b0001, "saddw",
2711 BinOpFrag<(add node:$LHS, (sext node:$RHS))>>;
2712 defm SMLAL : SIMDLongThreeVectorTiedBHS<0, 0b1000, "smlal",
2713 TriOpFrag<(add node:$LHS, (int_arm64_neon_smull node:$MHS, node:$RHS))>>;
2714 defm SMLSL : SIMDLongThreeVectorTiedBHS<0, 0b1010, "smlsl",
2715 TriOpFrag<(sub node:$LHS, (int_arm64_neon_smull node:$MHS, node:$RHS))>>;
2716 defm SMULL : SIMDLongThreeVectorBHS<0, 0b1100, "smull", int_arm64_neon_smull>;
2717 defm SQDMLAL : SIMDLongThreeVectorSQDMLXTiedHS<0, 0b1001, "sqdmlal",
2718 int_arm64_neon_sqadd>;
2719 defm SQDMLSL : SIMDLongThreeVectorSQDMLXTiedHS<0, 0b1011, "sqdmlsl",
2720 int_arm64_neon_sqsub>;
2721 defm SQDMULL : SIMDLongThreeVectorHS<0, 0b1101, "sqdmull",
2722 int_arm64_neon_sqdmull>;
2723 defm SSUBL : SIMDLongThreeVectorBHS<0, 0b0010, "ssubl",
2724 BinOpFrag<(sub (sext node:$LHS), (sext node:$RHS))>>;
2725 defm SSUBW : SIMDWideThreeVectorBHS<0, 0b0011, "ssubw",
2726 BinOpFrag<(sub node:$LHS, (sext node:$RHS))>>;
2727 defm UABAL : SIMDLongThreeVectorTiedBHSabal<1, 0b0101, "uabal",
2728 int_arm64_neon_uabd>;
2729 defm UABDL : SIMDLongThreeVectorBHSabdl<1, 0b0111, "uabdl",
2730 int_arm64_neon_uabd>;
2731 defm UADDL : SIMDLongThreeVectorBHS<1, 0b0000, "uaddl",
2732 BinOpFrag<(add (zext node:$LHS), (zext node:$RHS))>>;
2733 defm UADDW : SIMDWideThreeVectorBHS<1, 0b0001, "uaddw",
2734 BinOpFrag<(add node:$LHS, (zext node:$RHS))>>;
2735 defm UMLAL : SIMDLongThreeVectorTiedBHS<1, 0b1000, "umlal",
2736 TriOpFrag<(add node:$LHS, (int_arm64_neon_umull node:$MHS, node:$RHS))>>;
2737 defm UMLSL : SIMDLongThreeVectorTiedBHS<1, 0b1010, "umlsl",
2738 TriOpFrag<(sub node:$LHS, (int_arm64_neon_umull node:$MHS, node:$RHS))>>;
2739 defm UMULL : SIMDLongThreeVectorBHS<1, 0b1100, "umull", int_arm64_neon_umull>;
2740 defm USUBL : SIMDLongThreeVectorBHS<1, 0b0010, "usubl",
2741 BinOpFrag<(sub (zext node:$LHS), (zext node:$RHS))>>;
2742 defm USUBW : SIMDWideThreeVectorBHS< 1, 0b0011, "usubw",
2743 BinOpFrag<(sub node:$LHS, (zext node:$RHS))>>;
2745 // Patterns for 64-bit pmull
2746 def : Pat<(int_arm64_neon_pmull64 V64:$Rn, V64:$Rm),
2747 (PMULLv1i64 V64:$Rn, V64:$Rm)>;
2748 def : Pat<(int_arm64_neon_pmull64 (vector_extract (v2i64 V128:$Rn), (i64 1)),
2749 (vector_extract (v2i64 V128:$Rm), (i64 1))),
2750 (PMULLv2i64 V128:$Rn, V128:$Rm)>;
2752 // CodeGen patterns for addhn and subhn instructions, which can actually be
2753 // written in LLVM IR without too much difficulty.
2756 def : Pat<(v8i8 (trunc (v8i16 (ARM64vlshr (add V128:$Rn, V128:$Rm), (i32 8))))),
2757 (ADDHNv8i16_v8i8 V128:$Rn, V128:$Rm)>;
2758 def : Pat<(v4i16 (trunc (v4i32 (ARM64vlshr (add V128:$Rn, V128:$Rm),
2760 (ADDHNv4i32_v4i16 V128:$Rn, V128:$Rm)>;
2761 def : Pat<(v2i32 (trunc (v2i64 (ARM64vlshr (add V128:$Rn, V128:$Rm),
2763 (ADDHNv2i64_v2i32 V128:$Rn, V128:$Rm)>;
2764 def : Pat<(concat_vectors (v8i8 V64:$Rd),
2765 (trunc (v8i16 (ARM64vlshr (add V128:$Rn, V128:$Rm),
2767 (ADDHNv8i16_v16i8 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
2768 V128:$Rn, V128:$Rm)>;
2769 def : Pat<(concat_vectors (v4i16 V64:$Rd),
2770 (trunc (v4i32 (ARM64vlshr (add V128:$Rn, V128:$Rm),
2772 (ADDHNv4i32_v8i16 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
2773 V128:$Rn, V128:$Rm)>;
2774 def : Pat<(concat_vectors (v2i32 V64:$Rd),
2775 (trunc (v2i64 (ARM64vlshr (add V128:$Rn, V128:$Rm),
2777 (ADDHNv2i64_v4i32 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
2778 V128:$Rn, V128:$Rm)>;
2781 def : Pat<(v8i8 (trunc (v8i16 (ARM64vlshr (sub V128:$Rn, V128:$Rm), (i32 8))))),
2782 (SUBHNv8i16_v8i8 V128:$Rn, V128:$Rm)>;
2783 def : Pat<(v4i16 (trunc (v4i32 (ARM64vlshr (sub V128:$Rn, V128:$Rm),
2785 (SUBHNv4i32_v4i16 V128:$Rn, V128:$Rm)>;
2786 def : Pat<(v2i32 (trunc (v2i64 (ARM64vlshr (sub V128:$Rn, V128:$Rm),
2788 (SUBHNv2i64_v2i32 V128:$Rn, V128:$Rm)>;
2789 def : Pat<(concat_vectors (v8i8 V64:$Rd),
2790 (trunc (v8i16 (ARM64vlshr (sub V128:$Rn, V128:$Rm),
2792 (SUBHNv8i16_v16i8 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
2793 V128:$Rn, V128:$Rm)>;
2794 def : Pat<(concat_vectors (v4i16 V64:$Rd),
2795 (trunc (v4i32 (ARM64vlshr (sub V128:$Rn, V128:$Rm),
2797 (SUBHNv4i32_v8i16 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
2798 V128:$Rn, V128:$Rm)>;
2799 def : Pat<(concat_vectors (v2i32 V64:$Rd),
2800 (trunc (v2i64 (ARM64vlshr (sub V128:$Rn, V128:$Rm),
2802 (SUBHNv2i64_v4i32 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
2803 V128:$Rn, V128:$Rm)>;
2805 //----------------------------------------------------------------------------
2806 // AdvSIMD bitwise extract from vector instruction.
2807 //----------------------------------------------------------------------------
2809 defm EXT : SIMDBitwiseExtract<"ext">;
2811 def : Pat<(v4i16 (ARM64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
2812 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
2813 def : Pat<(v8i16 (ARM64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
2814 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
2815 def : Pat<(v2i32 (ARM64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
2816 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
2817 def : Pat<(v2f32 (ARM64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
2818 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
2819 def : Pat<(v4i32 (ARM64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
2820 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
2821 def : Pat<(v4f32 (ARM64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
2822 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
2823 def : Pat<(v2i64 (ARM64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
2824 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
2825 def : Pat<(v2f64 (ARM64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
2826 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
2828 // We use EXT to handle extract_subvector to copy the upper 64-bits of a
2830 def : Pat<(v8i8 (extract_subvector V128:$Rn, (i64 8))),
2831 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
2832 def : Pat<(v4i16 (extract_subvector V128:$Rn, (i64 4))),
2833 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
2834 def : Pat<(v2i32 (extract_subvector V128:$Rn, (i64 2))),
2835 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
2836 def : Pat<(v1i64 (extract_subvector V128:$Rn, (i64 1))),
2837 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
2838 def : Pat<(v2f32 (extract_subvector V128:$Rn, (i64 2))),
2839 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
2840 def : Pat<(v1f64 (extract_subvector V128:$Rn, (i64 1))),
2841 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
2844 //----------------------------------------------------------------------------
2845 // AdvSIMD zip vector
2846 //----------------------------------------------------------------------------
2848 defm TRN1 : SIMDZipVector<0b010, "trn1", ARM64trn1>;
2849 defm TRN2 : SIMDZipVector<0b110, "trn2", ARM64trn2>;
2850 defm UZP1 : SIMDZipVector<0b001, "uzp1", ARM64uzp1>;
2851 defm UZP2 : SIMDZipVector<0b101, "uzp2", ARM64uzp2>;
2852 defm ZIP1 : SIMDZipVector<0b011, "zip1", ARM64zip1>;
2853 defm ZIP2 : SIMDZipVector<0b111, "zip2", ARM64zip2>;
2855 //----------------------------------------------------------------------------
2856 // AdvSIMD TBL/TBX instructions
2857 //----------------------------------------------------------------------------
2859 defm TBL : SIMDTableLookup< 0, "tbl">;
2860 defm TBX : SIMDTableLookupTied<1, "tbx">;
2862 def : Pat<(v8i8 (int_arm64_neon_tbl1 (v16i8 VecListOne128:$Rn), (v8i8 V64:$Ri))),
2863 (TBLv8i8One VecListOne128:$Rn, V64:$Ri)>;
2864 def : Pat<(v16i8 (int_arm64_neon_tbl1 (v16i8 V128:$Ri), (v16i8 V128:$Rn))),
2865 (TBLv16i8One V128:$Ri, V128:$Rn)>;
2867 def : Pat<(v8i8 (int_arm64_neon_tbx1 (v8i8 V64:$Rd),
2868 (v16i8 VecListOne128:$Rn), (v8i8 V64:$Ri))),
2869 (TBXv8i8One V64:$Rd, VecListOne128:$Rn, V64:$Ri)>;
2870 def : Pat<(v16i8 (int_arm64_neon_tbx1 (v16i8 V128:$Rd),
2871 (v16i8 V128:$Ri), (v16i8 V128:$Rn))),
2872 (TBXv16i8One V128:$Rd, V128:$Ri, V128:$Rn)>;
2875 //----------------------------------------------------------------------------
2876 // AdvSIMD scalar CPY instruction
2877 //----------------------------------------------------------------------------
2879 defm CPY : SIMDScalarCPY<"cpy">;
2881 //----------------------------------------------------------------------------
2882 // AdvSIMD scalar pairwise instructions
2883 //----------------------------------------------------------------------------
2885 defm ADDP : SIMDPairwiseScalarD<0, 0b11011, "addp">;
2886 defm FADDP : SIMDPairwiseScalarSD<1, 0, 0b01101, "faddp">;
2887 defm FMAXNMP : SIMDPairwiseScalarSD<1, 0, 0b01100, "fmaxnmp">;
2888 defm FMAXP : SIMDPairwiseScalarSD<1, 0, 0b01111, "fmaxp">;
2889 defm FMINNMP : SIMDPairwiseScalarSD<1, 1, 0b01100, "fminnmp">;
2890 defm FMINP : SIMDPairwiseScalarSD<1, 1, 0b01111, "fminp">;
2891 def : Pat<(i64 (int_arm64_neon_saddv (v2i64 V128:$Rn))),
2892 (ADDPv2i64p V128:$Rn)>;
2893 def : Pat<(i64 (int_arm64_neon_uaddv (v2i64 V128:$Rn))),
2894 (ADDPv2i64p V128:$Rn)>;
2895 def : Pat<(f32 (int_arm64_neon_faddv (v2f32 V64:$Rn))),
2896 (FADDPv2i32p V64:$Rn)>;
2897 def : Pat<(f32 (int_arm64_neon_faddv (v4f32 V128:$Rn))),
2898 (FADDPv2i32p (EXTRACT_SUBREG (FADDPv4f32 V128:$Rn, V128:$Rn), dsub))>;
2899 def : Pat<(f64 (int_arm64_neon_faddv (v2f64 V128:$Rn))),
2900 (FADDPv2i64p V128:$Rn)>;
2901 def : Pat<(f32 (int_arm64_neon_fmaxnmv (v2f32 V64:$Rn))),
2902 (FMAXNMPv2i32p V64:$Rn)>;
2903 def : Pat<(f64 (int_arm64_neon_fmaxnmv (v2f64 V128:$Rn))),
2904 (FMAXNMPv2i64p V128:$Rn)>;
2905 def : Pat<(f32 (int_arm64_neon_fmaxv (v2f32 V64:$Rn))),
2906 (FMAXPv2i32p V64:$Rn)>;
2907 def : Pat<(f64 (int_arm64_neon_fmaxv (v2f64 V128:$Rn))),
2908 (FMAXPv2i64p V128:$Rn)>;
2909 def : Pat<(f32 (int_arm64_neon_fminnmv (v2f32 V64:$Rn))),
2910 (FMINNMPv2i32p V64:$Rn)>;
2911 def : Pat<(f64 (int_arm64_neon_fminnmv (v2f64 V128:$Rn))),
2912 (FMINNMPv2i64p V128:$Rn)>;
2913 def : Pat<(f32 (int_arm64_neon_fminv (v2f32 V64:$Rn))),
2914 (FMINPv2i32p V64:$Rn)>;
2915 def : Pat<(f64 (int_arm64_neon_fminv (v2f64 V128:$Rn))),
2916 (FMINPv2i64p V128:$Rn)>;
2918 //----------------------------------------------------------------------------
2919 // AdvSIMD INS/DUP instructions
2920 //----------------------------------------------------------------------------
2922 def DUPv8i8gpr : SIMDDupFromMain<0, 0b00001, ".8b", v8i8, V64, GPR32>;
2923 def DUPv16i8gpr : SIMDDupFromMain<1, 0b00001, ".16b", v16i8, V128, GPR32>;
2924 def DUPv4i16gpr : SIMDDupFromMain<0, 0b00010, ".4h", v4i16, V64, GPR32>;
2925 def DUPv8i16gpr : SIMDDupFromMain<1, 0b00010, ".8h", v8i16, V128, GPR32>;
2926 def DUPv2i32gpr : SIMDDupFromMain<0, 0b00100, ".2s", v2i32, V64, GPR32>;
2927 def DUPv4i32gpr : SIMDDupFromMain<1, 0b00100, ".4s", v4i32, V128, GPR32>;
2928 def DUPv2i64gpr : SIMDDupFromMain<1, 0b01000, ".2d", v2i64, V128, GPR64>;
2930 def DUPv2i64lane : SIMDDup64FromElement;
2931 def DUPv2i32lane : SIMDDup32FromElement<0, ".2s", v2i32, V64>;
2932 def DUPv4i32lane : SIMDDup32FromElement<1, ".4s", v4i32, V128>;
2933 def DUPv4i16lane : SIMDDup16FromElement<0, ".4h", v4i16, V64>;
2934 def DUPv8i16lane : SIMDDup16FromElement<1, ".8h", v8i16, V128>;
2935 def DUPv8i8lane : SIMDDup8FromElement <0, ".8b", v8i8, V64>;
2936 def DUPv16i8lane : SIMDDup8FromElement <1, ".16b", v16i8, V128>;
2938 def : Pat<(v2f32 (ARM64dup (f32 FPR32:$Rn))),
2939 (v2f32 (DUPv2i32lane
2940 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rn, ssub),
2942 def : Pat<(v4f32 (ARM64dup (f32 FPR32:$Rn))),
2943 (v4f32 (DUPv4i32lane
2944 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rn, ssub),
2946 def : Pat<(v2f64 (ARM64dup (f64 FPR64:$Rn))),
2947 (v2f64 (DUPv2i64lane
2948 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$Rn, dsub),
2951 def : Pat<(v2f32 (ARM64duplane32 (v4f32 V128:$Rn), VectorIndexS:$imm)),
2952 (DUPv2i32lane V128:$Rn, VectorIndexS:$imm)>;
2953 def : Pat<(v4f32 (ARM64duplane32 (v4f32 V128:$Rn), VectorIndexS:$imm)),
2954 (DUPv4i32lane V128:$Rn, VectorIndexS:$imm)>;
2955 def : Pat<(v2f64 (ARM64duplane64 (v2f64 V128:$Rn), VectorIndexD:$imm)),
2956 (DUPv2i64lane V128:$Rn, VectorIndexD:$imm)>;
2961 def : Pat<(sext_inreg (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx), i8),
2962 (i32 (SMOVvi8to32 V128:$Rn, VectorIndexB:$idx))>;
2963 def : Pat<(sext_inreg (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx), i8),
2964 (i64 (SMOVvi8to64 V128:$Rn, VectorIndexB:$idx))>;
2965 def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
2966 (i32 (SMOVvi16to32 V128:$Rn, VectorIndexH:$idx))>;
2967 def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
2968 (i64 (SMOVvi16to64 V128:$Rn, VectorIndexH:$idx))>;
2969 def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
2970 (i32 (SMOVvi16to32 V128:$Rn, VectorIndexH:$idx))>;
2971 def : Pat<(sext (i32 (vector_extract (v4i32 V128:$Rn), VectorIndexS:$idx))),
2972 (i64 (SMOVvi32to64 V128:$Rn, VectorIndexS:$idx))>;
2974 // Extracting i8 or i16 elements will have the zero-extend transformed to
2975 // an 'and' mask by type legalization since neither i8 nor i16 are legal types
2976 // for ARM64. Match these patterns here since UMOV already zeroes out the high
2977 // bits of the destination register.
2978 def : Pat<(and (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx),
2980 (i32 (UMOVvi8 V128:$Rn, VectorIndexB:$idx))>;
2981 def : Pat<(and (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),
2983 (i32 (UMOVvi16 V128:$Rn, VectorIndexH:$idx))>;
2987 def : Pat<(v16i8 (scalar_to_vector GPR32:$Rn)),
2988 (SUBREG_TO_REG (i32 0),
2989 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
2990 def : Pat<(v8i8 (scalar_to_vector GPR32:$Rn)),
2991 (SUBREG_TO_REG (i32 0),
2992 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
2994 def : Pat<(v8i16 (scalar_to_vector GPR32:$Rn)),
2995 (SUBREG_TO_REG (i32 0),
2996 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
2997 def : Pat<(v4i16 (scalar_to_vector GPR32:$Rn)),
2998 (SUBREG_TO_REG (i32 0),
2999 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3001 def : Pat<(v2i32 (scalar_to_vector (i32 FPR32:$Rn))),
3002 (v2i32 (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)),
3003 (i32 FPR32:$Rn), ssub))>;
3004 def : Pat<(v4i32 (scalar_to_vector (i32 FPR32:$Rn))),
3005 (v4i32 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
3006 (i32 FPR32:$Rn), ssub))>;
3007 def : Pat<(v2i64 (scalar_to_vector (i64 FPR64:$Rn))),
3008 (v2i64 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
3009 (i64 FPR64:$Rn), dsub))>;
3011 def : Pat<(v4f32 (scalar_to_vector (f32 FPR32:$Rn))),
3012 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rn, ssub)>;
3013 def : Pat<(v2f32 (scalar_to_vector (f32 FPR32:$Rn))),
3014 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), FPR32:$Rn, ssub)>;
3015 def : Pat<(v2f64 (scalar_to_vector (f64 FPR64:$Rn))),
3016 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$Rn, dsub)>;
3018 def : Pat<(v2f32 (vector_insert (v2f32 V64:$Rn),
3019 (f32 FPR32:$Rm), (i64 VectorIndexS:$imm))),
3022 (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), V64:$Rn, dsub)),
3024 (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rm, ssub)),
3027 def : Pat<(v4f32 (vector_insert (v4f32 V128:$Rn),
3028 (f32 FPR32:$Rm), (i64 VectorIndexS:$imm))),
3030 V128:$Rn, VectorIndexS:$imm,
3031 (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rm, ssub)),
3033 def : Pat<(v2f64 (vector_insert (v2f64 V128:$Rn),
3034 (f64 FPR64:$Rm), (i64 VectorIndexD:$imm))),
3036 V128:$Rn, VectorIndexD:$imm,
3037 (v2f64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$Rm, dsub)),
3040 // Copy an element at a constant index in one vector into a constant indexed
3041 // element of another.
3042 // FIXME refactor to a shared class/dev parameterized on vector type, vector
3043 // index type and INS extension
3044 def : Pat<(v16i8 (int_arm64_neon_vcopy_lane
3045 (v16i8 V128:$Vd), VectorIndexB:$idx, (v16i8 V128:$Vs),
3046 VectorIndexB:$idx2)),
3048 V128:$Vd, VectorIndexB:$idx, V128:$Vs, VectorIndexB:$idx2)
3050 def : Pat<(v8i16 (int_arm64_neon_vcopy_lane
3051 (v8i16 V128:$Vd), VectorIndexH:$idx, (v8i16 V128:$Vs),
3052 VectorIndexH:$idx2)),
3054 V128:$Vd, VectorIndexH:$idx, V128:$Vs, VectorIndexH:$idx2)
3056 def : Pat<(v4i32 (int_arm64_neon_vcopy_lane
3057 (v4i32 V128:$Vd), VectorIndexS:$idx, (v4i32 V128:$Vs),
3058 VectorIndexS:$idx2)),
3060 V128:$Vd, VectorIndexS:$idx, V128:$Vs, VectorIndexS:$idx2)
3062 def : Pat<(v2i64 (int_arm64_neon_vcopy_lane
3063 (v2i64 V128:$Vd), VectorIndexD:$idx, (v2i64 V128:$Vs),
3064 VectorIndexD:$idx2)),
3066 V128:$Vd, VectorIndexD:$idx, V128:$Vs, VectorIndexD:$idx2)
3069 // Floating point vector extractions are codegen'd as either a sequence of
3070 // subregister extractions, possibly fed by an INS if the lane number is
3071 // anything other than zero.
3072 def : Pat<(vector_extract (v2f64 V128:$Rn), 0),
3073 (f64 (EXTRACT_SUBREG V128:$Rn, dsub))>;
3074 def : Pat<(vector_extract (v4f32 V128:$Rn), 0),
3075 (f32 (EXTRACT_SUBREG V128:$Rn, ssub))>;
3076 def : Pat<(vector_extract (v2f64 V128:$Rn), VectorIndexD:$idx),
3077 (f64 (EXTRACT_SUBREG
3078 (INSvi64lane (v2f64 (IMPLICIT_DEF)), 0,
3079 V128:$Rn, VectorIndexD:$idx),
3081 def : Pat<(vector_extract (v4f32 V128:$Rn), VectorIndexS:$idx),
3082 (f32 (EXTRACT_SUBREG
3083 (INSvi32lane (v4f32 (IMPLICIT_DEF)), 0,
3084 V128:$Rn, VectorIndexS:$idx),
3087 // All concat_vectors operations are canonicalised to act on i64 vectors for
3088 // ARM64. In the general case we need an instruction, which had just as well be
3090 class ConcatPat<ValueType DstTy, ValueType SrcTy>
3091 : Pat<(DstTy (concat_vectors (SrcTy V64:$Rd), V64:$Rn)),
3092 (INSvi64lane (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), 1,
3093 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rn, dsub), 0)>;
3095 def : ConcatPat<v2i64, v1i64>;
3096 def : ConcatPat<v2f64, v1f64>;
3097 def : ConcatPat<v4i32, v2i32>;
3098 def : ConcatPat<v4f32, v2f32>;
3099 def : ConcatPat<v8i16, v4i16>;
3100 def : ConcatPat<v16i8, v8i8>;
3102 // If the high lanes are undef, though, we can just ignore them:
3103 class ConcatUndefPat<ValueType DstTy, ValueType SrcTy>
3104 : Pat<(DstTy (concat_vectors (SrcTy V64:$Rn), undef)),
3105 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rn, dsub)>;
3107 def : ConcatUndefPat<v2i64, v1i64>;
3108 def : ConcatUndefPat<v2f64, v1f64>;
3109 def : ConcatUndefPat<v4i32, v2i32>;
3110 def : ConcatUndefPat<v4f32, v2f32>;
3111 def : ConcatUndefPat<v8i16, v4i16>;
3112 def : ConcatUndefPat<v16i8, v8i8>;
3114 //----------------------------------------------------------------------------
3115 // AdvSIMD across lanes instructions
3116 //----------------------------------------------------------------------------
3118 defm ADDV : SIMDAcrossLanesBHS<0, 0b11011, "addv">;
3119 defm SMAXV : SIMDAcrossLanesBHS<0, 0b01010, "smaxv">;
3120 defm SMINV : SIMDAcrossLanesBHS<0, 0b11010, "sminv">;
3121 defm UMAXV : SIMDAcrossLanesBHS<1, 0b01010, "umaxv">;
3122 defm UMINV : SIMDAcrossLanesBHS<1, 0b11010, "uminv">;
3123 defm SADDLV : SIMDAcrossLanesHSD<0, 0b00011, "saddlv">;
3124 defm UADDLV : SIMDAcrossLanesHSD<1, 0b00011, "uaddlv">;
3125 defm FMAXNMV : SIMDAcrossLanesS<0b01100, 0, "fmaxnmv", int_arm64_neon_fmaxnmv>;
3126 defm FMAXV : SIMDAcrossLanesS<0b01111, 0, "fmaxv", int_arm64_neon_fmaxv>;
3127 defm FMINNMV : SIMDAcrossLanesS<0b01100, 1, "fminnmv", int_arm64_neon_fminnmv>;
3128 defm FMINV : SIMDAcrossLanesS<0b01111, 1, "fminv", int_arm64_neon_fminv>;
3130 multiclass SIMDAcrossLanesSignedIntrinsic<string baseOpc, Intrinsic intOp> {
3131 // If there is a sign extension after this intrinsic, consume it as smov already
3133 def : Pat<(i32 (sext_inreg (i32 (intOp (v8i8 V64:$Rn))), i8)),
3135 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3136 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3138 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
3140 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3141 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3143 // If there is a sign extension after this intrinsic, consume it as smov already
3145 def : Pat<(i32 (sext_inreg (i32 (intOp (v16i8 V128:$Rn))), i8)),
3147 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3148 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3150 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
3152 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3153 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3155 // If there is a sign extension after this intrinsic, consume it as smov already
3157 def : Pat<(i32 (sext_inreg (i32 (intOp (v4i16 V64:$Rn))), i16)),
3159 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3160 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3162 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
3164 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3165 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3167 // If there is a sign extension after this intrinsic, consume it as smov already
3169 def : Pat<(i32 (sext_inreg (i32 (intOp (v8i16 V128:$Rn))), i16)),
3171 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3172 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3174 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
3176 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3177 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3180 def : Pat<(i32 (intOp (v4i32 V128:$Rn))),
3181 (i32 (EXTRACT_SUBREG
3182 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3183 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), ssub),
3187 multiclass SIMDAcrossLanesUnsignedIntrinsic<string baseOpc, Intrinsic intOp> {
3188 // If there is a masking operation keeping only what has been actually
3189 // generated, consume it.
3190 def : Pat<(i32 (and (i32 (intOp (v8i8 V64:$Rn))), maski8_or_more)),
3191 (i32 (EXTRACT_SUBREG
3192 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3193 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3195 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
3196 (i32 (EXTRACT_SUBREG
3197 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3198 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3200 // If there is a masking operation keeping only what has been actually
3201 // generated, consume it.
3202 def : Pat<(i32 (and (i32 (intOp (v16i8 V128:$Rn))), maski8_or_more)),
3203 (i32 (EXTRACT_SUBREG
3204 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3205 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3207 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
3208 (i32 (EXTRACT_SUBREG
3209 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3210 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3213 // If there is a masking operation keeping only what has been actually
3214 // generated, consume it.
3215 def : Pat<(i32 (and (i32 (intOp (v4i16 V64:$Rn))), maski16_or_more)),
3216 (i32 (EXTRACT_SUBREG
3217 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3218 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3220 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
3221 (i32 (EXTRACT_SUBREG
3222 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3223 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3225 // If there is a masking operation keeping only what has been actually
3226 // generated, consume it.
3227 def : Pat<(i32 (and (i32 (intOp (v8i16 V128:$Rn))), maski16_or_more)),
3228 (i32 (EXTRACT_SUBREG
3229 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3230 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3232 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
3233 (i32 (EXTRACT_SUBREG
3234 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3235 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3238 def : Pat<(i32 (intOp (v4i32 V128:$Rn))),
3239 (i32 (EXTRACT_SUBREG
3240 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3241 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), ssub),
3246 multiclass SIMDAcrossLanesSignedLongIntrinsic<string baseOpc, Intrinsic intOp> {
3247 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
3249 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3250 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), hsub),
3252 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
3254 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3255 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), hsub),
3258 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
3259 (i32 (EXTRACT_SUBREG
3260 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3261 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), ssub),
3263 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
3264 (i32 (EXTRACT_SUBREG
3265 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3266 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), ssub),
3269 def : Pat<(i64 (intOp (v4i32 V128:$Rn))),
3270 (i64 (EXTRACT_SUBREG
3271 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3272 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), dsub),
3276 multiclass SIMDAcrossLanesUnsignedLongIntrinsic<string baseOpc,
3278 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
3279 (i32 (EXTRACT_SUBREG
3280 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3281 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), hsub),
3283 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
3284 (i32 (EXTRACT_SUBREG
3285 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3286 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), hsub),
3289 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
3290 (i32 (EXTRACT_SUBREG
3291 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3292 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), ssub),
3294 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
3295 (i32 (EXTRACT_SUBREG
3296 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3297 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), ssub),
3300 def : Pat<(i64 (intOp (v4i32 V128:$Rn))),
3301 (i64 (EXTRACT_SUBREG
3302 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3303 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), dsub),
3307 defm : SIMDAcrossLanesSignedIntrinsic<"ADDV", int_arm64_neon_saddv>;
3308 // vaddv_[su]32 is special; -> ADDP Vd.2S,Vn.2S,Vm.2S; return Vd.s[0];Vn==Vm
3309 def : Pat<(i32 (int_arm64_neon_saddv (v2i32 V64:$Rn))),
3310 (EXTRACT_SUBREG (ADDPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3312 defm : SIMDAcrossLanesUnsignedIntrinsic<"ADDV", int_arm64_neon_uaddv>;
3313 // vaddv_[su]32 is special; -> ADDP Vd.2S,Vn.2S,Vm.2S; return Vd.s[0];Vn==Vm
3314 def : Pat<(i32 (int_arm64_neon_uaddv (v2i32 V64:$Rn))),
3315 (EXTRACT_SUBREG (ADDPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3317 defm : SIMDAcrossLanesSignedIntrinsic<"SMAXV", int_arm64_neon_smaxv>;
3318 def : Pat<(i32 (int_arm64_neon_smaxv (v2i32 V64:$Rn))),
3319 (EXTRACT_SUBREG (SMAXPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3321 defm : SIMDAcrossLanesSignedIntrinsic<"SMINV", int_arm64_neon_sminv>;
3322 def : Pat<(i32 (int_arm64_neon_sminv (v2i32 V64:$Rn))),
3323 (EXTRACT_SUBREG (SMINPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3325 defm : SIMDAcrossLanesUnsignedIntrinsic<"UMAXV", int_arm64_neon_umaxv>;
3326 def : Pat<(i32 (int_arm64_neon_umaxv (v2i32 V64:$Rn))),
3327 (EXTRACT_SUBREG (UMAXPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3329 defm : SIMDAcrossLanesUnsignedIntrinsic<"UMINV", int_arm64_neon_uminv>;
3330 def : Pat<(i32 (int_arm64_neon_uminv (v2i32 V64:$Rn))),
3331 (EXTRACT_SUBREG (UMINPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3333 defm : SIMDAcrossLanesSignedLongIntrinsic<"SADDLV", int_arm64_neon_saddlv>;
3334 defm : SIMDAcrossLanesUnsignedLongIntrinsic<"UADDLV", int_arm64_neon_uaddlv>;
3336 // The vaddlv_s32 intrinsic gets mapped to SADDLP.
3337 def : Pat<(i64 (int_arm64_neon_saddlv (v2i32 V64:$Rn))),
3338 (i64 (EXTRACT_SUBREG
3339 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3340 (SADDLPv2i32_v1i64 V64:$Rn), dsub),
3342 // The vaddlv_u32 intrinsic gets mapped to UADDLP.
3343 def : Pat<(i64 (int_arm64_neon_uaddlv (v2i32 V64:$Rn))),
3344 (i64 (EXTRACT_SUBREG
3345 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3346 (UADDLPv2i32_v1i64 V64:$Rn), dsub),
3349 //------------------------------------------------------------------------------
3350 // AdvSIMD modified immediate instructions
3351 //------------------------------------------------------------------------------
3354 defm BIC : SIMDModifiedImmVectorShiftTied<1, 0b11, 0b01, "bic", ARM64bici>;
3356 defm ORR : SIMDModifiedImmVectorShiftTied<0, 0b11, 0b01, "orr", ARM64orri>;
3360 def FMOVv2f64_ns : SIMDModifiedImmVectorNoShift<1, 1, 0b1111, V128, fpimm8,
3362 [(set (v2f64 V128:$Rd), (ARM64fmov imm0_255:$imm8))]>;
3363 def FMOVv2f32_ns : SIMDModifiedImmVectorNoShift<0, 0, 0b1111, V64, fpimm8,
3365 [(set (v2f32 V64:$Rd), (ARM64fmov imm0_255:$imm8))]>;
3366 def FMOVv4f32_ns : SIMDModifiedImmVectorNoShift<1, 0, 0b1111, V128, fpimm8,
3368 [(set (v4f32 V128:$Rd), (ARM64fmov imm0_255:$imm8))]>;
3372 // EDIT byte mask: scalar
3373 let isReMaterializable = 1, isAsCheapAsAMove = 1 in
3374 def MOVID : SIMDModifiedImmScalarNoShift<0, 1, 0b1110, "movi",
3375 [(set FPR64:$Rd, simdimmtype10:$imm8)]>;
3376 // The movi_edit node has the immediate value already encoded, so we use
3377 // a plain imm0_255 here.
3378 def : Pat<(f64 (ARM64movi_edit imm0_255:$shift)),
3379 (MOVID imm0_255:$shift)>;
3381 def : Pat<(v1i64 immAllZerosV), (MOVID (i32 0))>;
3382 def : Pat<(v2i32 immAllZerosV), (MOVID (i32 0))>;
3383 def : Pat<(v4i16 immAllZerosV), (MOVID (i32 0))>;
3384 def : Pat<(v8i8 immAllZerosV), (MOVID (i32 0))>;
3386 def : Pat<(v1i64 immAllOnesV), (MOVID (i32 255))>;
3387 def : Pat<(v2i32 immAllOnesV), (MOVID (i32 255))>;
3388 def : Pat<(v4i16 immAllOnesV), (MOVID (i32 255))>;
3389 def : Pat<(v8i8 immAllOnesV), (MOVID (i32 255))>;
3391 // EDIT byte mask: 2d
3393 // The movi_edit node has the immediate value already encoded, so we use
3394 // a plain imm0_255 in the pattern
3395 let isReMaterializable = 1, isAsCheapAsAMove = 1 in
3396 def MOVIv2d_ns : SIMDModifiedImmVectorNoShift<1, 1, 0b1110, V128,
3399 [(set (v2i64 V128:$Rd), (ARM64movi_edit imm0_255:$imm8))]>;
3402 // Use movi.2d to materialize 0.0 if the HW does zero-cycle zeroing.
3403 // Complexity is added to break a tie with a plain MOVI.
3404 let AddedComplexity = 1 in {
3405 def : Pat<(f32 fpimm0),
3406 (f32 (EXTRACT_SUBREG (v2i64 (MOVIv2d_ns (i32 0))), ssub))>,
3408 def : Pat<(f64 fpimm0),
3409 (f64 (EXTRACT_SUBREG (v2i64 (MOVIv2d_ns (i32 0))), dsub))>,
3413 def : Pat<(v2i64 immAllZerosV), (MOVIv2d_ns (i32 0))>;
3414 def : Pat<(v4i32 immAllZerosV), (MOVIv2d_ns (i32 0))>;
3415 def : Pat<(v8i16 immAllZerosV), (MOVIv2d_ns (i32 0))>;
3416 def : Pat<(v16i8 immAllZerosV), (MOVIv2d_ns (i32 0))>;
3418 def : Pat<(v2i64 immAllOnesV), (MOVIv2d_ns (i32 255))>;
3419 def : Pat<(v4i32 immAllOnesV), (MOVIv2d_ns (i32 255))>;
3420 def : Pat<(v8i16 immAllOnesV), (MOVIv2d_ns (i32 255))>;
3421 def : Pat<(v16i8 immAllOnesV), (MOVIv2d_ns (i32 255))>;
3423 // EDIT per word & halfword: 2s, 4h, 4s, & 8h
3424 defm MOVI : SIMDModifiedImmVectorShift<0, 0b10, 0b00, "movi">;
3425 def : Pat<(v2i32 (ARM64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
3426 (MOVIv2i32 imm0_255:$imm8, imm:$shift)>;
3427 def : Pat<(v4i32 (ARM64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
3428 (MOVIv4i32 imm0_255:$imm8, imm:$shift)>;
3429 def : Pat<(v4i16 (ARM64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
3430 (MOVIv4i16 imm0_255:$imm8, imm:$shift)>;
3431 def : Pat<(v8i16 (ARM64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
3432 (MOVIv8i16 imm0_255:$imm8, imm:$shift)>;
3434 // EDIT per word: 2s & 4s with MSL shifter
3435 def MOVIv2s_msl : SIMDModifiedImmMoveMSL<0, 0, {1,1,0,?}, V64, "movi", ".2s",
3436 [(set (v2i32 V64:$Rd),
3437 (ARM64movi_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
3438 def MOVIv4s_msl : SIMDModifiedImmMoveMSL<1, 0, {1,1,0,?}, V128, "movi", ".4s",
3439 [(set (v4i32 V128:$Rd),
3440 (ARM64movi_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
3442 // Per byte: 8b & 16b
3443 def MOVIv8b_ns : SIMDModifiedImmVectorNoShift<0, 0, 0b1110, V64, imm0_255,
3445 [(set (v8i8 V64:$Rd), (ARM64movi imm0_255:$imm8))]>;
3446 def MOVIv16b_ns : SIMDModifiedImmVectorNoShift<1, 0, 0b1110, V128, imm0_255,
3448 [(set (v16i8 V128:$Rd), (ARM64movi imm0_255:$imm8))]>;
3452 // EDIT per word & halfword: 2s, 4h, 4s, & 8h
3453 defm MVNI : SIMDModifiedImmVectorShift<1, 0b10, 0b00, "mvni">;
3454 def : Pat<(v2i32 (ARM64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
3455 (MVNIv2i32 imm0_255:$imm8, imm:$shift)>;
3456 def : Pat<(v4i32 (ARM64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
3457 (MVNIv4i32 imm0_255:$imm8, imm:$shift)>;
3458 def : Pat<(v4i16 (ARM64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
3459 (MVNIv4i16 imm0_255:$imm8, imm:$shift)>;
3460 def : Pat<(v8i16 (ARM64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
3461 (MVNIv8i16 imm0_255:$imm8, imm:$shift)>;
3463 // EDIT per word: 2s & 4s with MSL shifter
3464 def MVNIv2s_msl : SIMDModifiedImmMoveMSL<0, 1, {1,1,0,?}, V64, "mvni", ".2s",
3465 [(set (v2i32 V64:$Rd),
3466 (ARM64mvni_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
3467 def MVNIv4s_msl : SIMDModifiedImmMoveMSL<1, 1, {1,1,0,?}, V128, "mvni", ".4s",
3468 [(set (v4i32 V128:$Rd),
3469 (ARM64mvni_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
3471 //----------------------------------------------------------------------------
3472 // AdvSIMD indexed element
3473 //----------------------------------------------------------------------------
3475 let neverHasSideEffects = 1 in {
3476 defm FMLA : SIMDFPIndexedSDTied<0, 0b0001, "fmla">;
3477 defm FMLS : SIMDFPIndexedSDTied<0, 0b0101, "fmls">;
3480 // NOTE: Operands are reordered in the FMLA/FMLS PatFrags because the
3481 // instruction expects the addend first, while the intrinsic expects it last.
3483 // On the other hand, there are quite a few valid combinatorial options due to
3484 // the commutativity of multiplication and the fact that (-x) * y = x * (-y).
3485 defm : SIMDFPIndexedSDTiedPatterns<"FMLA",
3486 TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)>>;
3487 defm : SIMDFPIndexedSDTiedPatterns<"FMLA",
3488 TriOpFrag<(fma node:$MHS, node:$RHS, node:$LHS)>>;
3490 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
3491 TriOpFrag<(fma node:$MHS, (fneg node:$RHS), node:$LHS)> >;
3492 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
3493 TriOpFrag<(fma node:$RHS, (fneg node:$MHS), node:$LHS)> >;
3494 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
3495 TriOpFrag<(fma (fneg node:$RHS), node:$MHS, node:$LHS)> >;
3496 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
3497 TriOpFrag<(fma (fneg node:$MHS), node:$RHS, node:$LHS)> >;
3499 multiclass FMLSIndexedAfterNegPatterns<SDPatternOperator OpNode> {
3500 // 3 variants for the .2s version: DUPLANE from 128-bit, DUPLANE from 64-bit
3502 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
3503 (ARM64duplane32 (v4f32 (fneg V128:$Rm)),
3504 VectorIndexS:$idx))),
3505 (FMLSv2i32_indexed V64:$Rd, V64:$Rn, V128:$Rm, VectorIndexS:$idx)>;
3506 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
3507 (v2f32 (ARM64duplane32
3508 (v4f32 (insert_subvector undef,
3509 (v2f32 (fneg V64:$Rm)),
3511 VectorIndexS:$idx)))),
3512 (FMLSv2i32_indexed V64:$Rd, V64:$Rn,
3513 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub),
3514 VectorIndexS:$idx)>;
3515 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
3516 (ARM64dup (f32 (fneg FPR32Op:$Rm))))),
3517 (FMLSv2i32_indexed V64:$Rd, V64:$Rn,
3518 (SUBREG_TO_REG (i32 0), FPR32Op:$Rm, ssub), (i64 0))>;
3520 // 3 variants for the .4s version: DUPLANE from 128-bit, DUPLANE from 64-bit
3522 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
3523 (ARM64duplane32 (v4f32 (fneg V128:$Rm)),
3524 VectorIndexS:$idx))),
3525 (FMLSv4i32_indexed V128:$Rd, V128:$Rn, V128:$Rm,
3526 VectorIndexS:$idx)>;
3527 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
3528 (v4f32 (ARM64duplane32
3529 (v4f32 (insert_subvector undef,
3530 (v2f32 (fneg V64:$Rm)),
3532 VectorIndexS:$idx)))),
3533 (FMLSv4i32_indexed V128:$Rd, V128:$Rn,
3534 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub),
3535 VectorIndexS:$idx)>;
3536 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
3537 (ARM64dup (f32 (fneg FPR32Op:$Rm))))),
3538 (FMLSv4i32_indexed V128:$Rd, V128:$Rn,
3539 (SUBREG_TO_REG (i32 0), FPR32Op:$Rm, ssub), (i64 0))>;
3541 // 2 variants for the .2d version: DUPLANE from 128-bit, and DUP scalar
3542 // (DUPLANE from 64-bit would be trivial).
3543 def : Pat<(v2f64 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn),
3544 (ARM64duplane64 (v2f64 (fneg V128:$Rm)),
3545 VectorIndexD:$idx))),
3547 V128:$Rd, V128:$Rn, V128:$Rm, VectorIndexS:$idx)>;
3548 def : Pat<(v2f64 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn),
3549 (ARM64dup (f64 (fneg FPR64Op:$Rm))))),
3550 (FMLSv2i64_indexed V128:$Rd, V128:$Rn,
3551 (SUBREG_TO_REG (i32 0), FPR64Op:$Rm, dsub), (i64 0))>;
3553 // 2 variants for 32-bit scalar version: extract from .2s or from .4s
3554 def : Pat<(f32 (OpNode (f32 FPR32:$Rd), (f32 FPR32:$Rn),
3555 (vector_extract (v4f32 (fneg V128:$Rm)),
3556 VectorIndexS:$idx))),
3557 (FMLSv1i32_indexed FPR32:$Rd, FPR32:$Rn,
3558 V128:$Rm, VectorIndexS:$idx)>;
3559 def : Pat<(f32 (OpNode (f32 FPR32:$Rd), (f32 FPR32:$Rn),
3560 (vector_extract (v2f32 (fneg V64:$Rm)),
3561 VectorIndexS:$idx))),
3562 (FMLSv1i32_indexed FPR32:$Rd, FPR32:$Rn,
3563 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub), VectorIndexS:$idx)>;
3565 // 1 variant for 64-bit scalar version: extract from .1d or from .2d
3566 def : Pat<(f64 (OpNode (f64 FPR64:$Rd), (f64 FPR64:$Rn),
3567 (vector_extract (v2f64 (fneg V128:$Rm)),
3568 VectorIndexS:$idx))),
3569 (FMLSv1i64_indexed FPR64:$Rd, FPR64:$Rn,
3570 V128:$Rm, VectorIndexS:$idx)>;
3573 defm : FMLSIndexedAfterNegPatterns<
3574 TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)> >;
3575 defm : FMLSIndexedAfterNegPatterns<
3576 TriOpFrag<(fma node:$MHS, node:$RHS, node:$LHS)> >;
3578 defm FMULX : SIMDFPIndexedSD<1, 0b1001, "fmulx", int_arm64_neon_fmulx>;
3579 defm FMUL : SIMDFPIndexedSD<0, 0b1001, "fmul", fmul>;
3581 def : Pat<(v2f32 (fmul V64:$Rn, (ARM64dup (f32 FPR32:$Rm)))),
3582 (FMULv2i32_indexed V64:$Rn,
3583 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rm, ssub),
3585 def : Pat<(v4f32 (fmul V128:$Rn, (ARM64dup (f32 FPR32:$Rm)))),
3586 (FMULv4i32_indexed V128:$Rn,
3587 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rm, ssub),
3589 def : Pat<(v2f64 (fmul V128:$Rn, (ARM64dup (f64 FPR64:$Rm)))),
3590 (FMULv2i64_indexed V128:$Rn,
3591 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$Rm, dsub),
3594 defm SQDMULH : SIMDIndexedHS<0, 0b1100, "sqdmulh", int_arm64_neon_sqdmulh>;
3595 defm SQRDMULH : SIMDIndexedHS<0, 0b1101, "sqrdmulh", int_arm64_neon_sqrdmulh>;
3596 defm MLA : SIMDVectorIndexedHSTied<1, 0b0000, "mla",
3597 TriOpFrag<(add node:$LHS, (mul node:$MHS, node:$RHS))>>;
3598 defm MLS : SIMDVectorIndexedHSTied<1, 0b0100, "mls",
3599 TriOpFrag<(sub node:$LHS, (mul node:$MHS, node:$RHS))>>;
3600 defm MUL : SIMDVectorIndexedHS<0, 0b1000, "mul", mul>;
3601 defm SMLAL : SIMDVectorIndexedLongSDTied<0, 0b0010, "smlal",
3602 TriOpFrag<(add node:$LHS, (int_arm64_neon_smull node:$MHS, node:$RHS))>>;
3603 defm SMLSL : SIMDVectorIndexedLongSDTied<0, 0b0110, "smlsl",
3604 TriOpFrag<(sub node:$LHS, (int_arm64_neon_smull node:$MHS, node:$RHS))>>;
3605 defm SMULL : SIMDVectorIndexedLongSD<0, 0b1010, "smull",
3606 int_arm64_neon_smull>;
3607 defm SQDMLAL : SIMDIndexedLongSQDMLXSDTied<0, 0b0011, "sqdmlal",
3608 int_arm64_neon_sqadd>;
3609 defm SQDMLSL : SIMDIndexedLongSQDMLXSDTied<0, 0b0111, "sqdmlsl",
3610 int_arm64_neon_sqsub>;
3611 defm SQDMULL : SIMDIndexedLongSD<0, 0b1011, "sqdmull", int_arm64_neon_sqdmull>;
3612 defm UMLAL : SIMDVectorIndexedLongSDTied<1, 0b0010, "umlal",
3613 TriOpFrag<(add node:$LHS, (int_arm64_neon_umull node:$MHS, node:$RHS))>>;
3614 defm UMLSL : SIMDVectorIndexedLongSDTied<1, 0b0110, "umlsl",
3615 TriOpFrag<(sub node:$LHS, (int_arm64_neon_umull node:$MHS, node:$RHS))>>;
3616 defm UMULL : SIMDVectorIndexedLongSD<1, 0b1010, "umull",
3617 int_arm64_neon_umull>;
3619 // A scalar sqdmull with the second operand being a vector lane can be
3620 // handled directly with the indexed instruction encoding.
3621 def : Pat<(int_arm64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
3622 (vector_extract (v4i32 V128:$Vm),
3623 VectorIndexS:$idx)),
3624 (SQDMULLv1i64_indexed FPR32:$Rn, V128:$Vm, VectorIndexS:$idx)>;
3626 //----------------------------------------------------------------------------
3627 // AdvSIMD scalar shift instructions
3628 //----------------------------------------------------------------------------
3629 defm FCVTZS : SIMDScalarRShiftSD<0, 0b11111, "fcvtzs">;
3630 defm FCVTZU : SIMDScalarRShiftSD<1, 0b11111, "fcvtzu">;
3631 defm SCVTF : SIMDScalarRShiftSD<0, 0b11100, "scvtf">;
3632 defm UCVTF : SIMDScalarRShiftSD<1, 0b11100, "ucvtf">;
3633 // Codegen patterns for the above. We don't put these directly on the
3634 // instructions because TableGen's type inference can't handle the truth.
3635 // Having the same base pattern for fp <--> int totally freaks it out.
3636 def : Pat<(int_arm64_neon_vcvtfp2fxs FPR32:$Rn, vecshiftR32:$imm),
3637 (FCVTZSs FPR32:$Rn, vecshiftR32:$imm)>;
3638 def : Pat<(int_arm64_neon_vcvtfp2fxu FPR32:$Rn, vecshiftR32:$imm),
3639 (FCVTZUs FPR32:$Rn, vecshiftR32:$imm)>;
3640 def : Pat<(i64 (int_arm64_neon_vcvtfp2fxs (f64 FPR64:$Rn), vecshiftR64:$imm)),
3641 (FCVTZSd FPR64:$Rn, vecshiftR64:$imm)>;
3642 def : Pat<(i64 (int_arm64_neon_vcvtfp2fxu (f64 FPR64:$Rn), vecshiftR64:$imm)),
3643 (FCVTZUd FPR64:$Rn, vecshiftR64:$imm)>;
3644 def : Pat<(v1i64 (int_arm64_neon_vcvtfp2fxs (v1f64 FPR64:$Rn),
3646 (FCVTZSd FPR64:$Rn, vecshiftR64:$imm)>;
3647 def : Pat<(v1i64 (int_arm64_neon_vcvtfp2fxu (v1f64 FPR64:$Rn),
3649 (FCVTZUd FPR64:$Rn, vecshiftR64:$imm)>;
3650 def : Pat<(int_arm64_neon_vcvtfxs2fp FPR32:$Rn, vecshiftR32:$imm),
3651 (SCVTFs FPR32:$Rn, vecshiftR32:$imm)>;
3652 def : Pat<(int_arm64_neon_vcvtfxu2fp FPR32:$Rn, vecshiftR32:$imm),
3653 (UCVTFs FPR32:$Rn, vecshiftR32:$imm)>;
3654 def : Pat<(f64 (int_arm64_neon_vcvtfxs2fp (i64 FPR64:$Rn), vecshiftR64:$imm)),
3655 (SCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
3656 def : Pat<(f64 (int_arm64_neon_vcvtfxu2fp (i64 FPR64:$Rn), vecshiftR64:$imm)),
3657 (UCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
3658 def : Pat<(v1f64 (int_arm64_neon_vcvtfxs2fp (v1i64 FPR64:$Rn),
3660 (SCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
3661 def : Pat<(v1f64 (int_arm64_neon_vcvtfxu2fp (v1i64 FPR64:$Rn),
3663 (UCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
3665 defm SHL : SIMDScalarLShiftD< 0, 0b01010, "shl", ARM64vshl>;
3666 defm SLI : SIMDScalarLShiftDTied<1, 0b01010, "sli">;
3667 defm SQRSHRN : SIMDScalarRShiftBHS< 0, 0b10011, "sqrshrn",
3668 int_arm64_neon_sqrshrn>;
3669 defm SQRSHRUN : SIMDScalarRShiftBHS< 1, 0b10001, "sqrshrun",
3670 int_arm64_neon_sqrshrun>;
3671 defm SQSHLU : SIMDScalarLShiftBHSD<1, 0b01100, "sqshlu", ARM64sqshlui>;
3672 defm SQSHL : SIMDScalarLShiftBHSD<0, 0b01110, "sqshl", ARM64sqshli>;
3673 defm SQSHRN : SIMDScalarRShiftBHS< 0, 0b10010, "sqshrn",
3674 int_arm64_neon_sqshrn>;
3675 defm SQSHRUN : SIMDScalarRShiftBHS< 1, 0b10000, "sqshrun",
3676 int_arm64_neon_sqshrun>;
3677 defm SRI : SIMDScalarRShiftDTied< 1, 0b01000, "sri">;
3678 defm SRSHR : SIMDScalarRShiftD< 0, 0b00100, "srshr", ARM64srshri>;
3679 defm SRSRA : SIMDScalarRShiftDTied< 0, 0b00110, "srsra",
3680 TriOpFrag<(add node:$LHS,
3681 (ARM64srshri node:$MHS, node:$RHS))>>;
3682 defm SSHR : SIMDScalarRShiftD< 0, 0b00000, "sshr", ARM64vashr>;
3683 defm SSRA : SIMDScalarRShiftDTied< 0, 0b00010, "ssra",
3684 TriOpFrag<(add node:$LHS,
3685 (ARM64vashr node:$MHS, node:$RHS))>>;
3686 defm UQRSHRN : SIMDScalarRShiftBHS< 1, 0b10011, "uqrshrn",
3687 int_arm64_neon_uqrshrn>;
3688 defm UQSHL : SIMDScalarLShiftBHSD<1, 0b01110, "uqshl", ARM64uqshli>;
3689 defm UQSHRN : SIMDScalarRShiftBHS< 1, 0b10010, "uqshrn",
3690 int_arm64_neon_uqshrn>;
3691 defm URSHR : SIMDScalarRShiftD< 1, 0b00100, "urshr", ARM64urshri>;
3692 defm URSRA : SIMDScalarRShiftDTied< 1, 0b00110, "ursra",
3693 TriOpFrag<(add node:$LHS,
3694 (ARM64urshri node:$MHS, node:$RHS))>>;
3695 defm USHR : SIMDScalarRShiftD< 1, 0b00000, "ushr", ARM64vlshr>;
3696 defm USRA : SIMDScalarRShiftDTied< 1, 0b00010, "usra",
3697 TriOpFrag<(add node:$LHS,
3698 (ARM64vlshr node:$MHS, node:$RHS))>>;
3700 //----------------------------------------------------------------------------
3701 // AdvSIMD vector shift instructions
3702 //----------------------------------------------------------------------------
3703 defm FCVTZS:SIMDVectorRShiftSD<0, 0b11111, "fcvtzs", int_arm64_neon_vcvtfp2fxs>;
3704 defm FCVTZU:SIMDVectorRShiftSD<1, 0b11111, "fcvtzu", int_arm64_neon_vcvtfp2fxu>;
3705 defm SCVTF: SIMDVectorRShiftSDToFP<0, 0b11100, "scvtf",
3706 int_arm64_neon_vcvtfxs2fp>;
3707 defm RSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10001, "rshrn",
3708 int_arm64_neon_rshrn>;
3709 defm SHL : SIMDVectorLShiftBHSD<0, 0b01010, "shl", ARM64vshl>;
3710 defm SHRN : SIMDVectorRShiftNarrowBHS<0, 0b10000, "shrn",
3711 BinOpFrag<(trunc (ARM64vashr node:$LHS, node:$RHS))>>;
3712 defm SLI : SIMDVectorLShiftBHSDTied<1, 0b01010, "sli", int_arm64_neon_vsli>;
3713 def : Pat<(v1i64 (int_arm64_neon_vsli (v1i64 FPR64:$Rd), (v1i64 FPR64:$Rn),
3714 (i32 vecshiftL64:$imm))),
3715 (SLId FPR64:$Rd, FPR64:$Rn, vecshiftL64:$imm)>;
3716 defm SQRSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10011, "sqrshrn",
3717 int_arm64_neon_sqrshrn>;
3718 defm SQRSHRUN: SIMDVectorRShiftNarrowBHS<1, 0b10001, "sqrshrun",
3719 int_arm64_neon_sqrshrun>;
3720 defm SQSHLU : SIMDVectorLShiftBHSD<1, 0b01100, "sqshlu", ARM64sqshlui>;
3721 defm SQSHL : SIMDVectorLShiftBHSD<0, 0b01110, "sqshl", ARM64sqshli>;
3722 defm SQSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10010, "sqshrn",
3723 int_arm64_neon_sqshrn>;
3724 defm SQSHRUN : SIMDVectorRShiftNarrowBHS<1, 0b10000, "sqshrun",
3725 int_arm64_neon_sqshrun>;
3726 defm SRI : SIMDVectorRShiftBHSDTied<1, 0b01000, "sri", int_arm64_neon_vsri>;
3727 def : Pat<(v1i64 (int_arm64_neon_vsri (v1i64 FPR64:$Rd), (v1i64 FPR64:$Rn),
3728 (i32 vecshiftR64:$imm))),
3729 (SRId FPR64:$Rd, FPR64:$Rn, vecshiftR64:$imm)>;
3730 defm SRSHR : SIMDVectorRShiftBHSD<0, 0b00100, "srshr", ARM64srshri>;
3731 defm SRSRA : SIMDVectorRShiftBHSDTied<0, 0b00110, "srsra",
3732 TriOpFrag<(add node:$LHS,
3733 (ARM64srshri node:$MHS, node:$RHS))> >;
3734 defm SSHLL : SIMDVectorLShiftLongBHSD<0, 0b10100, "sshll",
3735 BinOpFrag<(ARM64vshl (sext node:$LHS), node:$RHS)>>;
3737 defm SSHR : SIMDVectorRShiftBHSD<0, 0b00000, "sshr", ARM64vashr>;
3738 defm SSRA : SIMDVectorRShiftBHSDTied<0, 0b00010, "ssra",
3739 TriOpFrag<(add node:$LHS, (ARM64vashr node:$MHS, node:$RHS))>>;
3740 defm UCVTF : SIMDVectorRShiftSDToFP<1, 0b11100, "ucvtf",
3741 int_arm64_neon_vcvtfxu2fp>;
3742 defm UQRSHRN : SIMDVectorRShiftNarrowBHS<1, 0b10011, "uqrshrn",
3743 int_arm64_neon_uqrshrn>;
3744 defm UQSHL : SIMDVectorLShiftBHSD<1, 0b01110, "uqshl", ARM64uqshli>;
3745 defm UQSHRN : SIMDVectorRShiftNarrowBHS<1, 0b10010, "uqshrn",
3746 int_arm64_neon_uqshrn>;
3747 defm URSHR : SIMDVectorRShiftBHSD<1, 0b00100, "urshr", ARM64urshri>;
3748 defm URSRA : SIMDVectorRShiftBHSDTied<1, 0b00110, "ursra",
3749 TriOpFrag<(add node:$LHS,
3750 (ARM64urshri node:$MHS, node:$RHS))> >;
3751 defm USHLL : SIMDVectorLShiftLongBHSD<1, 0b10100, "ushll",
3752 BinOpFrag<(ARM64vshl (zext node:$LHS), node:$RHS)>>;
3753 defm USHR : SIMDVectorRShiftBHSD<1, 0b00000, "ushr", ARM64vlshr>;
3754 defm USRA : SIMDVectorRShiftBHSDTied<1, 0b00010, "usra",
3755 TriOpFrag<(add node:$LHS, (ARM64vlshr node:$MHS, node:$RHS))> >;
3757 // SHRN patterns for when a logical right shift was used instead of arithmetic
3758 // (the immediate guarantees no sign bits actually end up in the result so it
3760 def : Pat<(v8i8 (trunc (ARM64vlshr (v8i16 V128:$Rn), vecshiftR16Narrow:$imm))),
3761 (SHRNv8i8_shift V128:$Rn, vecshiftR16Narrow:$imm)>;
3762 def : Pat<(v4i16 (trunc (ARM64vlshr (v4i32 V128:$Rn), vecshiftR32Narrow:$imm))),
3763 (SHRNv4i16_shift V128:$Rn, vecshiftR32Narrow:$imm)>;
3764 def : Pat<(v2i32 (trunc (ARM64vlshr (v2i64 V128:$Rn), vecshiftR64Narrow:$imm))),
3765 (SHRNv2i32_shift V128:$Rn, vecshiftR64Narrow:$imm)>;
3767 def : Pat<(v16i8 (concat_vectors (v8i8 V64:$Rd),
3768 (trunc (ARM64vlshr (v8i16 V128:$Rn),
3769 vecshiftR16Narrow:$imm)))),
3770 (SHRNv16i8_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
3771 V128:$Rn, vecshiftR16Narrow:$imm)>;
3772 def : Pat<(v8i16 (concat_vectors (v4i16 V64:$Rd),
3773 (trunc (ARM64vlshr (v4i32 V128:$Rn),
3774 vecshiftR32Narrow:$imm)))),
3775 (SHRNv8i16_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
3776 V128:$Rn, vecshiftR32Narrow:$imm)>;
3777 def : Pat<(v4i32 (concat_vectors (v2i32 V64:$Rd),
3778 (trunc (ARM64vlshr (v2i64 V128:$Rn),
3779 vecshiftR64Narrow:$imm)))),
3780 (SHRNv4i32_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
3781 V128:$Rn, vecshiftR32Narrow:$imm)>;
3783 // Vector sign and zero extensions are implemented with SSHLL and USSHLL.
3784 // Anyexts are implemented as zexts.
3785 def : Pat<(v8i16 (sext (v8i8 V64:$Rn))), (SSHLLv8i8_shift V64:$Rn, (i32 0))>;
3786 def : Pat<(v8i16 (zext (v8i8 V64:$Rn))), (USHLLv8i8_shift V64:$Rn, (i32 0))>;
3787 def : Pat<(v8i16 (anyext (v8i8 V64:$Rn))), (USHLLv8i8_shift V64:$Rn, (i32 0))>;
3788 def : Pat<(v4i32 (sext (v4i16 V64:$Rn))), (SSHLLv4i16_shift V64:$Rn, (i32 0))>;
3789 def : Pat<(v4i32 (zext (v4i16 V64:$Rn))), (USHLLv4i16_shift V64:$Rn, (i32 0))>;
3790 def : Pat<(v4i32 (anyext (v4i16 V64:$Rn))), (USHLLv4i16_shift V64:$Rn, (i32 0))>;
3791 def : Pat<(v2i64 (sext (v2i32 V64:$Rn))), (SSHLLv2i32_shift V64:$Rn, (i32 0))>;
3792 def : Pat<(v2i64 (zext (v2i32 V64:$Rn))), (USHLLv2i32_shift V64:$Rn, (i32 0))>;
3793 def : Pat<(v2i64 (anyext (v2i32 V64:$Rn))), (USHLLv2i32_shift V64:$Rn, (i32 0))>;
3794 // Also match an extend from the upper half of a 128 bit source register.
3795 def : Pat<(v8i16 (anyext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
3796 (USHLLv16i8_shift V128:$Rn, (i32 0))>;
3797 def : Pat<(v8i16 (zext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
3798 (USHLLv16i8_shift V128:$Rn, (i32 0))>;
3799 def : Pat<(v8i16 (sext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
3800 (SSHLLv16i8_shift V128:$Rn, (i32 0))>;
3801 def : Pat<(v4i32 (anyext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
3802 (USHLLv8i16_shift V128:$Rn, (i32 0))>;
3803 def : Pat<(v4i32 (zext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
3804 (USHLLv8i16_shift V128:$Rn, (i32 0))>;
3805 def : Pat<(v4i32 (sext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
3806 (SSHLLv8i16_shift V128:$Rn, (i32 0))>;
3807 def : Pat<(v2i64 (anyext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
3808 (USHLLv4i32_shift V128:$Rn, (i32 0))>;
3809 def : Pat<(v2i64 (zext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
3810 (USHLLv4i32_shift V128:$Rn, (i32 0))>;
3811 def : Pat<(v2i64 (sext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
3812 (SSHLLv4i32_shift V128:$Rn, (i32 0))>;
3814 // Vector shift sxtl aliases
3815 def : InstAlias<"sxtl.8h $dst, $src1",
3816 (SSHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
3817 def : InstAlias<"sxtl $dst.8h, $src1.8b",
3818 (SSHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
3819 def : InstAlias<"sxtl.4s $dst, $src1",
3820 (SSHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
3821 def : InstAlias<"sxtl $dst.4s, $src1.4h",
3822 (SSHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
3823 def : InstAlias<"sxtl.2d $dst, $src1",
3824 (SSHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
3825 def : InstAlias<"sxtl $dst.2d, $src1.2s",
3826 (SSHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
3828 // Vector shift sxtl2 aliases
3829 def : InstAlias<"sxtl2.8h $dst, $src1",
3830 (SSHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
3831 def : InstAlias<"sxtl2 $dst.8h, $src1.16b",
3832 (SSHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
3833 def : InstAlias<"sxtl2.4s $dst, $src1",
3834 (SSHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
3835 def : InstAlias<"sxtl2 $dst.4s, $src1.8h",
3836 (SSHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
3837 def : InstAlias<"sxtl2.2d $dst, $src1",
3838 (SSHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
3839 def : InstAlias<"sxtl2 $dst.2d, $src1.4s",
3840 (SSHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
3842 // Vector shift uxtl aliases
3843 def : InstAlias<"uxtl.8h $dst, $src1",
3844 (USHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
3845 def : InstAlias<"uxtl $dst.8h, $src1.8b",
3846 (USHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
3847 def : InstAlias<"uxtl.4s $dst, $src1",
3848 (USHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
3849 def : InstAlias<"uxtl $dst.4s, $src1.4h",
3850 (USHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
3851 def : InstAlias<"uxtl.2d $dst, $src1",
3852 (USHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
3853 def : InstAlias<"uxtl $dst.2d, $src1.2s",
3854 (USHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
3856 // Vector shift uxtl2 aliases
3857 def : InstAlias<"uxtl2.8h $dst, $src1",
3858 (USHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
3859 def : InstAlias<"uxtl2 $dst.8h, $src1.16b",
3860 (USHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
3861 def : InstAlias<"uxtl2.4s $dst, $src1",
3862 (USHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
3863 def : InstAlias<"uxtl2 $dst.4s, $src1.8h",
3864 (USHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
3865 def : InstAlias<"uxtl2.2d $dst, $src1",
3866 (USHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
3867 def : InstAlias<"uxtl2 $dst.2d, $src1.4s",
3868 (USHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
3870 // If an integer is about to be converted to a floating point value,
3871 // just load it on the floating point unit.
3872 // These patterns are more complex because floating point loads do not
3873 // support sign extension.
3874 // The sign extension has to be explicitly added and is only supported for
3875 // one step: byte-to-half, half-to-word, word-to-doubleword.
3876 // SCVTF GPR -> FPR is 9 cycles.
3877 // SCVTF FPR -> FPR is 4 cyclces.
3878 // (sign extension with lengthen) SXTL FPR -> FPR is 2 cycles.
3879 // Therefore, we can do 2 sign extensions and one SCVTF FPR -> FPR
3880 // and still being faster.
3881 // However, this is not good for code size.
3882 // 8-bits -> float. 2 sizes step-up.
3883 def : Pat <(f32 (sint_to_fp (i32 (sextloadi8 ro_indexed8:$addr)))),
3884 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
3889 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3890 (LDRBro ro_indexed8:$addr),
3895 ssub)))>, Requires<[NotForCodeSize]>;
3896 def : Pat <(f32 (sint_to_fp (i32 (sextloadi8 am_indexed8:$addr)))),
3897 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
3902 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3903 (LDRBui am_indexed8:$addr),
3908 ssub)))>, Requires<[NotForCodeSize]>;
3909 def : Pat <(f32 (sint_to_fp (i32 (sextloadi8 am_unscaled8:$addr)))),
3910 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
3915 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3916 (LDURBi am_unscaled8:$addr),
3921 ssub)))>, Requires<[NotForCodeSize]>;
3922 // 16-bits -> float. 1 size step-up.
3923 def : Pat <(f32 (sint_to_fp (i32 (sextloadi16 ro_indexed16:$addr)))),
3924 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
3926 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3927 (LDRHro ro_indexed16:$addr),
3930 ssub)))>, Requires<[NotForCodeSize]>;
3931 def : Pat <(f32 (sint_to_fp (i32 (sextloadi16 am_indexed16:$addr)))),
3932 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
3934 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3935 (LDRHui am_indexed16:$addr),
3938 ssub)))>, Requires<[NotForCodeSize]>;
3939 def : Pat <(f32 (sint_to_fp (i32 (sextloadi16 am_unscaled16:$addr)))),
3940 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
3942 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3943 (LDURHi am_unscaled16:$addr),
3946 ssub)))>, Requires<[NotForCodeSize]>;
3947 // 32-bits to 32-bits are handled in target specific dag combine:
3948 // performIntToFpCombine.
3949 // 64-bits integer to 32-bits floating point, not possible with
3950 // SCVTF on floating point registers (both source and destination
3951 // must have the same size).
3953 // Here are the patterns for 8, 16, 32, and 64-bits to double.
3954 // 8-bits -> double. 3 size step-up: give up.
3955 // 16-bits -> double. 2 size step.
3956 def : Pat <(f64 (sint_to_fp (i32 (sextloadi16 ro_indexed16:$addr)))),
3957 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
3962 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3963 (LDRHro ro_indexed16:$addr),
3968 dsub)))>, Requires<[NotForCodeSize]>;
3969 def : Pat <(f64 (sint_to_fp (i32 (sextloadi16 am_indexed16:$addr)))),
3970 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
3975 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3976 (LDRHui am_indexed16:$addr),
3981 dsub)))>, Requires<[NotForCodeSize]>;
3982 def : Pat <(f64 (sint_to_fp (i32 (sextloadi16 am_unscaled16:$addr)))),
3983 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
3988 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3989 (LDURHi am_unscaled16:$addr),
3994 dsub)))>, Requires<[NotForCodeSize]>;
3995 // 32-bits -> double. 1 size step-up.
3996 def : Pat <(f64 (sint_to_fp (i32 (load ro_indexed32:$addr)))),
3997 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
3999 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4000 (LDRSro ro_indexed32:$addr),
4003 dsub)))>, Requires<[NotForCodeSize]>;
4004 def : Pat <(f64 (sint_to_fp (i32 (load am_indexed32:$addr)))),
4005 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4007 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4008 (LDRSui am_indexed32:$addr),
4011 dsub)))>, Requires<[NotForCodeSize]>;
4012 def : Pat <(f64 (sint_to_fp (i32 (load am_unscaled32:$addr)))),
4013 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4015 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4016 (LDURSi am_unscaled32:$addr),
4019 dsub)))>, Requires<[NotForCodeSize]>;
4020 // 64-bits -> double are handled in target specific dag combine:
4021 // performIntToFpCombine.
4024 //----------------------------------------------------------------------------
4025 // AdvSIMD Load-Store Structure
4026 //----------------------------------------------------------------------------
4027 defm LD1 : SIMDLd1Multiple<"ld1">;
4028 defm LD2 : SIMDLd2Multiple<"ld2">;
4029 defm LD3 : SIMDLd3Multiple<"ld3">;
4030 defm LD4 : SIMDLd4Multiple<"ld4">;
4032 defm ST1 : SIMDSt1Multiple<"st1">;
4033 defm ST2 : SIMDSt2Multiple<"st2">;
4034 defm ST3 : SIMDSt3Multiple<"st3">;
4035 defm ST4 : SIMDSt4Multiple<"st4">;
4037 class Ld1Pat<ValueType ty, Instruction INST>
4038 : Pat<(ty (load am_simdnoindex:$vaddr)), (INST am_simdnoindex:$vaddr)>;
4040 def : Ld1Pat<v16i8, LD1Onev16b>;
4041 def : Ld1Pat<v8i16, LD1Onev8h>;
4042 def : Ld1Pat<v4i32, LD1Onev4s>;
4043 def : Ld1Pat<v2i64, LD1Onev2d>;
4044 def : Ld1Pat<v8i8, LD1Onev8b>;
4045 def : Ld1Pat<v4i16, LD1Onev4h>;
4046 def : Ld1Pat<v2i32, LD1Onev2s>;
4047 def : Ld1Pat<v1i64, LD1Onev1d>;
4049 class St1Pat<ValueType ty, Instruction INST>
4050 : Pat<(store ty:$Vt, am_simdnoindex:$vaddr),
4051 (INST ty:$Vt, am_simdnoindex:$vaddr)>;
4053 def : St1Pat<v16i8, ST1Onev16b>;
4054 def : St1Pat<v8i16, ST1Onev8h>;
4055 def : St1Pat<v4i32, ST1Onev4s>;
4056 def : St1Pat<v2i64, ST1Onev2d>;
4057 def : St1Pat<v8i8, ST1Onev8b>;
4058 def : St1Pat<v4i16, ST1Onev4h>;
4059 def : St1Pat<v2i32, ST1Onev2s>;
4060 def : St1Pat<v1i64, ST1Onev1d>;
4066 defm LD1R : SIMDLdR<0, 0b110, 0, "ld1r", "One", 1, 2, 4, 8>;
4067 defm LD2R : SIMDLdR<1, 0b110, 0, "ld2r", "Two", 2, 4, 8, 16>;
4068 defm LD3R : SIMDLdR<0, 0b111, 0, "ld3r", "Three", 3, 6, 12, 24>;
4069 defm LD4R : SIMDLdR<1, 0b111, 0, "ld4r", "Four", 4, 8, 16, 32>;
4070 let mayLoad = 1, neverHasSideEffects = 1 in {
4071 defm LD1 : SIMDLdSingleBTied<0, 0b000, "ld1", VecListOneb, GPR64pi1>;
4072 defm LD1 : SIMDLdSingleHTied<0, 0b010, 0, "ld1", VecListOneh, GPR64pi2>;
4073 defm LD1 : SIMDLdSingleSTied<0, 0b100, 0b00, "ld1", VecListOnes, GPR64pi4>;
4074 defm LD1 : SIMDLdSingleDTied<0, 0b100, 0b01, "ld1", VecListOned, GPR64pi8>;
4075 defm LD2 : SIMDLdSingleBTied<1, 0b000, "ld2", VecListTwob, GPR64pi2>;
4076 defm LD2 : SIMDLdSingleHTied<1, 0b010, 0, "ld2", VecListTwoh, GPR64pi4>;
4077 defm LD2 : SIMDLdSingleSTied<1, 0b100, 0b00, "ld2", VecListTwos, GPR64pi8>;
4078 defm LD2 : SIMDLdSingleDTied<1, 0b100, 0b01, "ld2", VecListTwod, GPR64pi16>;
4079 defm LD3 : SIMDLdSingleBTied<0, 0b001, "ld3", VecListThreeb, GPR64pi3>;
4080 defm LD3 : SIMDLdSingleHTied<0, 0b011, 0, "ld3", VecListThreeh, GPR64pi6>;
4081 defm LD3 : SIMDLdSingleSTied<0, 0b101, 0b00, "ld3", VecListThrees, GPR64pi12>;
4082 defm LD3 : SIMDLdSingleDTied<0, 0b101, 0b01, "ld3", VecListThreed, GPR64pi24>;
4083 defm LD4 : SIMDLdSingleBTied<1, 0b001, "ld4", VecListFourb, GPR64pi4>;
4084 defm LD4 : SIMDLdSingleHTied<1, 0b011, 0, "ld4", VecListFourh, GPR64pi8>;
4085 defm LD4 : SIMDLdSingleSTied<1, 0b101, 0b00, "ld4", VecListFours, GPR64pi16>;
4086 defm LD4 : SIMDLdSingleDTied<1, 0b101, 0b01, "ld4", VecListFourd, GPR64pi32>;
4089 def : Pat<(v8i8 (ARM64dup (i32 (extloadi8 am_simdnoindex:$vaddr)))),
4090 (LD1Rv8b am_simdnoindex:$vaddr)>;
4091 def : Pat<(v16i8 (ARM64dup (i32 (extloadi8 am_simdnoindex:$vaddr)))),
4092 (LD1Rv16b am_simdnoindex:$vaddr)>;
4093 def : Pat<(v4i16 (ARM64dup (i32 (extloadi16 am_simdnoindex:$vaddr)))),
4094 (LD1Rv4h am_simdnoindex:$vaddr)>;
4095 def : Pat<(v8i16 (ARM64dup (i32 (extloadi16 am_simdnoindex:$vaddr)))),
4096 (LD1Rv8h am_simdnoindex:$vaddr)>;
4097 def : Pat<(v2i32 (ARM64dup (i32 (load am_simdnoindex:$vaddr)))),
4098 (LD1Rv2s am_simdnoindex:$vaddr)>;
4099 def : Pat<(v4i32 (ARM64dup (i32 (load am_simdnoindex:$vaddr)))),
4100 (LD1Rv4s am_simdnoindex:$vaddr)>;
4101 def : Pat<(v2i64 (ARM64dup (i64 (load am_simdnoindex:$vaddr)))),
4102 (LD1Rv2d am_simdnoindex:$vaddr)>;
4103 def : Pat<(v1i64 (ARM64dup (i64 (load am_simdnoindex:$vaddr)))),
4104 (LD1Rv1d am_simdnoindex:$vaddr)>;
4105 // Grab the floating point version too
4106 def : Pat<(v2f32 (ARM64dup (f32 (load am_simdnoindex:$vaddr)))),
4107 (LD1Rv2s am_simdnoindex:$vaddr)>;
4108 def : Pat<(v4f32 (ARM64dup (f32 (load am_simdnoindex:$vaddr)))),
4109 (LD1Rv4s am_simdnoindex:$vaddr)>;
4110 def : Pat<(v2f64 (ARM64dup (f64 (load am_simdnoindex:$vaddr)))),
4111 (LD1Rv2d am_simdnoindex:$vaddr)>;
4112 def : Pat<(v1f64 (ARM64dup (f64 (load am_simdnoindex:$vaddr)))),
4113 (LD1Rv1d am_simdnoindex:$vaddr)>;
4115 class Ld1Lane128Pat<SDPatternOperator scalar_load, Operand VecIndex,
4116 ValueType VTy, ValueType STy, Instruction LD1>
4117 : Pat<(vector_insert (VTy VecListOne128:$Rd),
4118 (STy (scalar_load am_simdnoindex:$vaddr)), VecIndex:$idx),
4119 (LD1 VecListOne128:$Rd, VecIndex:$idx, am_simdnoindex:$vaddr)>;
4121 def : Ld1Lane128Pat<extloadi8, VectorIndexB, v16i8, i32, LD1i8>;
4122 def : Ld1Lane128Pat<extloadi16, VectorIndexH, v8i16, i32, LD1i16>;
4123 def : Ld1Lane128Pat<load, VectorIndexS, v4i32, i32, LD1i32>;
4124 def : Ld1Lane128Pat<load, VectorIndexS, v4f32, f32, LD1i32>;
4125 def : Ld1Lane128Pat<load, VectorIndexD, v2i64, i64, LD1i64>;
4126 def : Ld1Lane128Pat<load, VectorIndexD, v2f64, f64, LD1i64>;
4128 class Ld1Lane64Pat<SDPatternOperator scalar_load, Operand VecIndex,
4129 ValueType VTy, ValueType STy, Instruction LD1>
4130 : Pat<(vector_insert (VTy VecListOne64:$Rd),
4131 (STy (scalar_load am_simdnoindex:$vaddr)), VecIndex:$idx),
4133 (LD1 (SUBREG_TO_REG (i32 0), VecListOne64:$Rd, dsub),
4134 VecIndex:$idx, am_simdnoindex:$vaddr),
4137 def : Ld1Lane64Pat<extloadi8, VectorIndexB, v8i8, i32, LD1i8>;
4138 def : Ld1Lane64Pat<extloadi16, VectorIndexH, v4i16, i32, LD1i16>;
4139 def : Ld1Lane64Pat<load, VectorIndexS, v2i32, i32, LD1i32>;
4140 def : Ld1Lane64Pat<load, VectorIndexS, v2f32, f32, LD1i32>;
4143 defm LD1 : SIMDLdSt1SingleAliases<"ld1">;
4144 defm LD2 : SIMDLdSt2SingleAliases<"ld2">;
4145 defm LD3 : SIMDLdSt3SingleAliases<"ld3">;
4146 defm LD4 : SIMDLdSt4SingleAliases<"ld4">;
4149 defm ST1 : SIMDStSingleB<0, 0b000, "st1", VecListOneb, GPR64pi1>;
4150 defm ST1 : SIMDStSingleH<0, 0b010, 0, "st1", VecListOneh, GPR64pi2>;
4151 defm ST1 : SIMDStSingleS<0, 0b100, 0b00, "st1", VecListOnes, GPR64pi4>;
4152 defm ST1 : SIMDStSingleD<0, 0b100, 0b01, "st1", VecListOned, GPR64pi8>;
4154 let AddedComplexity = 8 in
4155 class St1Lane128Pat<SDPatternOperator scalar_store, Operand VecIndex,
4156 ValueType VTy, ValueType STy, Instruction ST1>
4158 (STy (vector_extract (VTy VecListOne128:$Vt), VecIndex:$idx)),
4159 am_simdnoindex:$vaddr),
4160 (ST1 VecListOne128:$Vt, VecIndex:$idx, am_simdnoindex:$vaddr)>;
4162 def : St1Lane128Pat<truncstorei8, VectorIndexB, v16i8, i32, ST1i8>;
4163 def : St1Lane128Pat<truncstorei16, VectorIndexH, v8i16, i32, ST1i16>;
4164 def : St1Lane128Pat<store, VectorIndexS, v4i32, i32, ST1i32>;
4165 def : St1Lane128Pat<store, VectorIndexS, v4f32, f32, ST1i32>;
4166 def : St1Lane128Pat<store, VectorIndexD, v2i64, i64, ST1i64>;
4167 def : St1Lane128Pat<store, VectorIndexD, v2f64, f64, ST1i64>;
4169 let AddedComplexity = 8 in
4170 class St1Lane64Pat<SDPatternOperator scalar_store, Operand VecIndex,
4171 ValueType VTy, ValueType STy, Instruction ST1>
4173 (STy (vector_extract (VTy VecListOne64:$Vt), VecIndex:$idx)),
4174 am_simdnoindex:$vaddr),
4175 (ST1 (SUBREG_TO_REG (i32 0), VecListOne64:$Vt, dsub),
4176 VecIndex:$idx, am_simdnoindex:$vaddr)>;
4178 def : St1Lane64Pat<truncstorei8, VectorIndexB, v8i8, i32, ST1i8>;
4179 def : St1Lane64Pat<truncstorei16, VectorIndexH, v4i16, i32, ST1i16>;
4180 def : St1Lane64Pat<store, VectorIndexS, v2i32, i32, ST1i32>;
4181 def : St1Lane64Pat<store, VectorIndexS, v2f32, f32, ST1i32>;
4183 let mayStore = 1, neverHasSideEffects = 1 in {
4184 defm ST2 : SIMDStSingleB<1, 0b000, "st2", VecListTwob, GPR64pi2>;
4185 defm ST2 : SIMDStSingleH<1, 0b010, 0, "st2", VecListTwoh, GPR64pi4>;
4186 defm ST2 : SIMDStSingleS<1, 0b100, 0b00, "st2", VecListTwos, GPR64pi8>;
4187 defm ST2 : SIMDStSingleD<1, 0b100, 0b01, "st2", VecListTwod, GPR64pi16>;
4188 defm ST3 : SIMDStSingleB<0, 0b001, "st3", VecListThreeb, GPR64pi3>;
4189 defm ST3 : SIMDStSingleH<0, 0b011, 0, "st3", VecListThreeh, GPR64pi6>;
4190 defm ST3 : SIMDStSingleS<0, 0b101, 0b00, "st3", VecListThrees, GPR64pi12>;
4191 defm ST3 : SIMDStSingleD<0, 0b101, 0b01, "st3", VecListThreed, GPR64pi24>;
4192 defm ST4 : SIMDStSingleB<1, 0b001, "st4", VecListFourb, GPR64pi4>;
4193 defm ST4 : SIMDStSingleH<1, 0b011, 0, "st4", VecListFourh, GPR64pi8>;
4194 defm ST4 : SIMDStSingleS<1, 0b101, 0b00, "st4", VecListFours, GPR64pi16>;
4195 defm ST4 : SIMDStSingleD<1, 0b101, 0b01, "st4", VecListFourd, GPR64pi32>;
4198 defm ST1 : SIMDLdSt1SingleAliases<"st1">;
4199 defm ST2 : SIMDLdSt2SingleAliases<"st2">;
4200 defm ST3 : SIMDLdSt3SingleAliases<"st3">;
4201 defm ST4 : SIMDLdSt4SingleAliases<"st4">;
4203 //----------------------------------------------------------------------------
4204 // Crypto extensions
4205 //----------------------------------------------------------------------------
4207 def AESErr : AESTiedInst<0b0100, "aese", int_arm64_crypto_aese>;
4208 def AESDrr : AESTiedInst<0b0101, "aesd", int_arm64_crypto_aesd>;
4209 def AESMCrr : AESInst< 0b0110, "aesmc", int_arm64_crypto_aesmc>;
4210 def AESIMCrr : AESInst< 0b0111, "aesimc", int_arm64_crypto_aesimc>;
4212 def SHA1Crrr : SHATiedInstQSV<0b000, "sha1c", int_arm64_crypto_sha1c>;
4213 def SHA1Prrr : SHATiedInstQSV<0b001, "sha1p", int_arm64_crypto_sha1p>;
4214 def SHA1Mrrr : SHATiedInstQSV<0b010, "sha1m", int_arm64_crypto_sha1m>;
4215 def SHA1SU0rrr : SHATiedInstVVV<0b011, "sha1su0", int_arm64_crypto_sha1su0>;
4216 def SHA256Hrrr : SHATiedInstQQV<0b100, "sha256h", int_arm64_crypto_sha256h>;
4217 def SHA256H2rrr : SHATiedInstQQV<0b101, "sha256h2",int_arm64_crypto_sha256h2>;
4218 def SHA256SU1rrr :SHATiedInstVVV<0b110, "sha256su1",int_arm64_crypto_sha256su1>;
4220 def SHA1Hrr : SHAInstSS< 0b0000, "sha1h", int_arm64_crypto_sha1h>;
4221 def SHA1SU1rr : SHATiedInstVV<0b0001, "sha1su1", int_arm64_crypto_sha1su1>;
4222 def SHA256SU0rr : SHATiedInstVV<0b0010, "sha256su0",int_arm64_crypto_sha256su0>;
4224 //----------------------------------------------------------------------------
4226 //----------------------------------------------------------------------------
4227 // FIXME: Like for X86, these should go in their own separate .td file.
4229 // Any instruction that defines a 32-bit result leaves the high half of the
4230 // register. Truncate can be lowered to EXTRACT_SUBREG. CopyFromReg may
4231 // be copying from a truncate. But any other 32-bit operation will zero-extend
4233 // FIXME: X86 also checks for CMOV here. Do we need something similar?
4234 def def32 : PatLeaf<(i32 GPR32:$src), [{
4235 return N->getOpcode() != ISD::TRUNCATE &&
4236 N->getOpcode() != TargetOpcode::EXTRACT_SUBREG &&
4237 N->getOpcode() != ISD::CopyFromReg;
4240 // In the case of a 32-bit def that is known to implicitly zero-extend,
4241 // we can use a SUBREG_TO_REG.
4242 def : Pat<(i64 (zext def32:$src)), (SUBREG_TO_REG (i64 0), GPR32:$src, sub_32)>;
4244 // For an anyext, we don't care what the high bits are, so we can perform an
4245 // INSERT_SUBREF into an IMPLICIT_DEF.
4246 def : Pat<(i64 (anyext GPR32:$src)),
4247 (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32)>;
4249 // When we need to explicitly zero-extend, we use an unsigned bitfield move
4250 // instruction (UBFM) on the enclosing super-reg.
4251 def : Pat<(i64 (zext GPR32:$src)),
4252 (UBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32), 0, 31)>;
4254 // To sign extend, we use a signed bitfield move instruction (SBFM) on the
4255 // containing super-reg.
4256 def : Pat<(i64 (sext GPR32:$src)),
4257 (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32), 0, 31)>;
4258 def : Pat<(i64 (sext_inreg GPR64:$src, i32)), (SBFMXri GPR64:$src, 0, 31)>;
4259 def : Pat<(i64 (sext_inreg GPR64:$src, i16)), (SBFMXri GPR64:$src, 0, 15)>;
4260 def : Pat<(i64 (sext_inreg GPR64:$src, i8)), (SBFMXri GPR64:$src, 0, 7)>;
4261 def : Pat<(i64 (sext_inreg GPR64:$src, i1)), (SBFMXri GPR64:$src, 0, 0)>;
4262 def : Pat<(i32 (sext_inreg GPR32:$src, i16)), (SBFMWri GPR32:$src, 0, 15)>;
4263 def : Pat<(i32 (sext_inreg GPR32:$src, i8)), (SBFMWri GPR32:$src, 0, 7)>;
4264 def : Pat<(i32 (sext_inreg GPR32:$src, i1)), (SBFMWri GPR32:$src, 0, 0)>;
4266 def : Pat<(shl (sext_inreg GPR32:$Rn, i8), (i64 imm0_31:$imm)),
4267 (SBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
4268 (i64 (i32shift_sext_i8 imm0_31:$imm)))>;
4269 def : Pat<(shl (sext_inreg GPR64:$Rn, i8), (i64 imm0_63:$imm)),
4270 (SBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
4271 (i64 (i64shift_sext_i8 imm0_63:$imm)))>;
4273 def : Pat<(shl (sext_inreg GPR32:$Rn, i16), (i64 imm0_31:$imm)),
4274 (SBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
4275 (i64 (i32shift_sext_i16 imm0_31:$imm)))>;
4276 def : Pat<(shl (sext_inreg GPR64:$Rn, i16), (i64 imm0_63:$imm)),
4277 (SBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
4278 (i64 (i64shift_sext_i16 imm0_63:$imm)))>;
4280 def : Pat<(shl (i64 (sext GPR32:$Rn)), (i64 imm0_63:$imm)),
4281 (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$Rn, sub_32),
4282 (i64 (i64shift_a imm0_63:$imm)),
4283 (i64 (i64shift_sext_i32 imm0_63:$imm)))>;
4285 // sra patterns have an AddedComplexity of 10, so make sure we have a higher
4286 // AddedComplexity for the following patterns since we want to match sext + sra
4287 // patterns before we attempt to match a single sra node.
4288 let AddedComplexity = 20 in {
4289 // We support all sext + sra combinations which preserve at least one bit of the
4290 // original value which is to be sign extended. E.g. we support shifts up to
4292 def : Pat<(sra (sext_inreg GPR32:$Rn, i8), (i64 imm0_7:$imm)),
4293 (SBFMWri GPR32:$Rn, (i64 imm0_7:$imm), 7)>;
4294 def : Pat<(sra (sext_inreg GPR64:$Rn, i8), (i64 imm0_7:$imm)),
4295 (SBFMXri GPR64:$Rn, (i64 imm0_7:$imm), 7)>;
4297 def : Pat<(sra (sext_inreg GPR32:$Rn, i16), (i64 imm0_15:$imm)),
4298 (SBFMWri GPR32:$Rn, (i64 imm0_15:$imm), 15)>;
4299 def : Pat<(sra (sext_inreg GPR64:$Rn, i16), (i64 imm0_15:$imm)),
4300 (SBFMXri GPR64:$Rn, (i64 imm0_15:$imm), 15)>;
4302 def : Pat<(sra (i64 (sext GPR32:$Rn)), (i64 imm0_31:$imm)),
4303 (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$Rn, sub_32),
4304 (i64 imm0_31:$imm), 31)>;
4305 } // AddedComplexity = 20
4307 // To truncate, we can simply extract from a subregister.
4308 def : Pat<(i32 (trunc GPR64sp:$src)),
4309 (i32 (EXTRACT_SUBREG GPR64sp:$src, sub_32))>;
4311 // __builtin_trap() uses the BRK instruction on ARM64.
4312 def : Pat<(trap), (BRK 1)>;
4314 // Conversions within AdvSIMD types in the same register size are free.
4316 def : Pat<(v1i64 (bitconvert (v2i32 FPR64:$src))), (v1i64 FPR64:$src)>;
4317 def : Pat<(v1i64 (bitconvert (v4i16 FPR64:$src))), (v1i64 FPR64:$src)>;
4318 def : Pat<(v1i64 (bitconvert (v8i8 FPR64:$src))), (v1i64 FPR64:$src)>;
4319 def : Pat<(v1i64 (bitconvert (f64 FPR64:$src))), (v1i64 FPR64:$src)>;
4320 def : Pat<(v1i64 (bitconvert (v2f32 FPR64:$src))), (v1i64 FPR64:$src)>;
4321 def : Pat<(v1i64 (bitconvert (v1f64 FPR64:$src))), (v1i64 FPR64:$src)>;
4323 def : Pat<(v2i32 (bitconvert (v1i64 FPR64:$src))), (v2i32 FPR64:$src)>;
4324 def : Pat<(v2i32 (bitconvert (v4i16 FPR64:$src))), (v2i32 FPR64:$src)>;
4325 def : Pat<(v2i32 (bitconvert (v8i8 FPR64:$src))), (v2i32 FPR64:$src)>;
4326 def : Pat<(v2i32 (bitconvert (f64 FPR64:$src))), (v2i32 FPR64:$src)>;
4327 def : Pat<(v2i32 (bitconvert (v2f32 FPR64:$src))), (v2i32 FPR64:$src)>;
4328 def : Pat<(v2i32 (bitconvert (v1f64 FPR64:$src))), (v2i32 FPR64:$src)>;
4330 def : Pat<(v4i16 (bitconvert (v1i64 FPR64:$src))), (v4i16 FPR64:$src)>;
4331 def : Pat<(v4i16 (bitconvert (v2i32 FPR64:$src))), (v4i16 FPR64:$src)>;
4332 def : Pat<(v4i16 (bitconvert (v8i8 FPR64:$src))), (v4i16 FPR64:$src)>;
4333 def : Pat<(v4i16 (bitconvert (f64 FPR64:$src))), (v4i16 FPR64:$src)>;
4334 def : Pat<(v4i16 (bitconvert (v2f32 FPR64:$src))), (v4i16 FPR64:$src)>;
4335 def : Pat<(v4i16 (bitconvert (v1f64 FPR64:$src))), (v4i16 FPR64:$src)>;
4337 def : Pat<(v8i8 (bitconvert (v1i64 FPR64:$src))), (v8i8 FPR64:$src)>;
4338 def : Pat<(v8i8 (bitconvert (v2i32 FPR64:$src))), (v8i8 FPR64:$src)>;
4339 def : Pat<(v8i8 (bitconvert (v4i16 FPR64:$src))), (v8i8 FPR64:$src)>;
4340 def : Pat<(v8i8 (bitconvert (f64 FPR64:$src))), (v8i8 FPR64:$src)>;
4341 def : Pat<(v8i8 (bitconvert (v2f32 FPR64:$src))), (v8i8 FPR64:$src)>;
4342 def : Pat<(v8i8 (bitconvert (v1f64 FPR64:$src))), (v8i8 FPR64:$src)>;
4344 def : Pat<(f64 (bitconvert (v1i64 FPR64:$src))), (f64 FPR64:$src)>;
4345 def : Pat<(f64 (bitconvert (v2i32 FPR64:$src))), (f64 FPR64:$src)>;
4346 def : Pat<(f64 (bitconvert (v4i16 FPR64:$src))), (f64 FPR64:$src)>;
4347 def : Pat<(f64 (bitconvert (v8i8 FPR64:$src))), (f64 FPR64:$src)>;
4348 def : Pat<(f64 (bitconvert (v2f32 FPR64:$src))), (f64 FPR64:$src)>;
4349 def : Pat<(f64 (bitconvert (v1f64 FPR64:$src))), (f64 FPR64:$src)>;
4351 def : Pat<(v1f64 (bitconvert (v1i64 FPR64:$src))), (v1f64 FPR64:$src)>;
4352 def : Pat<(v1f64 (bitconvert (v2i32 FPR64:$src))), (v1f64 FPR64:$src)>;
4353 def : Pat<(v1f64 (bitconvert (v4i16 FPR64:$src))), (v1f64 FPR64:$src)>;
4354 def : Pat<(v1f64 (bitconvert (v8i8 FPR64:$src))), (v1f64 FPR64:$src)>;
4355 def : Pat<(v1f64 (bitconvert (f64 FPR64:$src))), (v1f64 FPR64:$src)>;
4356 def : Pat<(v1f64 (bitconvert (v2f32 FPR64:$src))), (v1f64 FPR64:$src)>;
4358 def : Pat<(v2f32 (bitconvert (f64 FPR64:$src))), (v2f32 FPR64:$src)>;
4359 def : Pat<(v2f32 (bitconvert (v1i64 FPR64:$src))), (v2f32 FPR64:$src)>;
4360 def : Pat<(v2f32 (bitconvert (v2i32 FPR64:$src))), (v2f32 FPR64:$src)>;
4361 def : Pat<(v2f32 (bitconvert (v4i16 FPR64:$src))), (v2f32 FPR64:$src)>;
4362 def : Pat<(v2f32 (bitconvert (v8i8 FPR64:$src))), (v2f32 FPR64:$src)>;
4363 def : Pat<(v2f32 (bitconvert (v1f64 FPR64:$src))), (v2f32 FPR64:$src)>;
4366 def : Pat<(f128 (bitconvert (v2i64 FPR128:$src))), (f128 FPR128:$src)>;
4367 def : Pat<(f128 (bitconvert (v4i32 FPR128:$src))), (f128 FPR128:$src)>;
4368 def : Pat<(f128 (bitconvert (v8i16 FPR128:$src))), (f128 FPR128:$src)>;
4369 def : Pat<(f128 (bitconvert (v2f64 FPR128:$src))), (f128 FPR128:$src)>;
4370 def : Pat<(f128 (bitconvert (v4f32 FPR128:$src))), (f128 FPR128:$src)>;
4372 def : Pat<(v2f64 (bitconvert (f128 FPR128:$src))), (v2f64 FPR128:$src)>;
4373 def : Pat<(v2f64 (bitconvert (v4i32 FPR128:$src))), (v2f64 FPR128:$src)>;
4374 def : Pat<(v2f64 (bitconvert (v8i16 FPR128:$src))), (v2f64 FPR128:$src)>;
4375 def : Pat<(v2f64 (bitconvert (v16i8 FPR128:$src))), (v2f64 FPR128:$src)>;
4376 def : Pat<(v2f64 (bitconvert (v2i64 FPR128:$src))), (v2f64 FPR128:$src)>;
4377 def : Pat<(v2f64 (bitconvert (v4f32 FPR128:$src))), (v2f64 FPR128:$src)>;
4379 def : Pat<(v4f32 (bitconvert (f128 FPR128:$src))), (v4f32 FPR128:$src)>;
4380 def : Pat<(v4f32 (bitconvert (v4i32 FPR128:$src))), (v4f32 FPR128:$src)>;
4381 def : Pat<(v4f32 (bitconvert (v8i16 FPR128:$src))), (v4f32 FPR128:$src)>;
4382 def : Pat<(v4f32 (bitconvert (v16i8 FPR128:$src))), (v4f32 FPR128:$src)>;
4383 def : Pat<(v4f32 (bitconvert (v2i64 FPR128:$src))), (v4f32 FPR128:$src)>;
4384 def : Pat<(v4f32 (bitconvert (v2f64 FPR128:$src))), (v4f32 FPR128:$src)>;
4386 def : Pat<(v2i64 (bitconvert (f128 FPR128:$src))), (v2i64 FPR128:$src)>;
4387 def : Pat<(v2i64 (bitconvert (v4i32 FPR128:$src))), (v2i64 FPR128:$src)>;
4388 def : Pat<(v2i64 (bitconvert (v8i16 FPR128:$src))), (v2i64 FPR128:$src)>;
4389 def : Pat<(v2i64 (bitconvert (v16i8 FPR128:$src))), (v2i64 FPR128:$src)>;
4390 def : Pat<(v2i64 (bitconvert (v2f64 FPR128:$src))), (v2i64 FPR128:$src)>;
4391 def : Pat<(v2i64 (bitconvert (v4f32 FPR128:$src))), (v2i64 FPR128:$src)>;
4393 def : Pat<(v4i32 (bitconvert (f128 FPR128:$src))), (v4i32 FPR128:$src)>;
4394 def : Pat<(v4i32 (bitconvert (v2i64 FPR128:$src))), (v4i32 FPR128:$src)>;
4395 def : Pat<(v4i32 (bitconvert (v8i16 FPR128:$src))), (v4i32 FPR128:$src)>;
4396 def : Pat<(v4i32 (bitconvert (v16i8 FPR128:$src))), (v4i32 FPR128:$src)>;
4397 def : Pat<(v4i32 (bitconvert (v2f64 FPR128:$src))), (v4i32 FPR128:$src)>;
4398 def : Pat<(v4i32 (bitconvert (v4f32 FPR128:$src))), (v4i32 FPR128:$src)>;
4400 def : Pat<(v8i16 (bitconvert (f128 FPR128:$src))), (v8i16 FPR128:$src)>;
4401 def : Pat<(v8i16 (bitconvert (v2i64 FPR128:$src))), (v8i16 FPR128:$src)>;
4402 def : Pat<(v8i16 (bitconvert (v4i32 FPR128:$src))), (v8i16 FPR128:$src)>;
4403 def : Pat<(v8i16 (bitconvert (v16i8 FPR128:$src))), (v8i16 FPR128:$src)>;
4404 def : Pat<(v8i16 (bitconvert (v2f64 FPR128:$src))), (v8i16 FPR128:$src)>;
4405 def : Pat<(v8i16 (bitconvert (v4f32 FPR128:$src))), (v8i16 FPR128:$src)>;
4407 def : Pat<(v16i8 (bitconvert (f128 FPR128:$src))), (v16i8 FPR128:$src)>;
4408 def : Pat<(v16i8 (bitconvert (v2i64 FPR128:$src))), (v16i8 FPR128:$src)>;
4409 def : Pat<(v16i8 (bitconvert (v4i32 FPR128:$src))), (v16i8 FPR128:$src)>;
4410 def : Pat<(v16i8 (bitconvert (v8i16 FPR128:$src))), (v16i8 FPR128:$src)>;
4411 def : Pat<(v16i8 (bitconvert (v2f64 FPR128:$src))), (v16i8 FPR128:$src)>;
4412 def : Pat<(v16i8 (bitconvert (v4f32 FPR128:$src))), (v16i8 FPR128:$src)>;
4414 def : Pat<(v8i8 (extract_subvector (v16i8 FPR128:$Rn), (i64 1))),
4415 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
4416 def : Pat<(v4i16 (extract_subvector (v8i16 FPR128:$Rn), (i64 1))),
4417 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
4418 def : Pat<(v2i32 (extract_subvector (v4i32 FPR128:$Rn), (i64 1))),
4419 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
4420 def : Pat<(v1i64 (extract_subvector (v2i64 FPR128:$Rn), (i64 1))),
4421 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
4423 // A 64-bit subvector insert to the first 128-bit vector position
4424 // is a subregister copy that needs no instruction.
4425 def : Pat<(insert_subvector undef, (v1i64 FPR64:$src), (i32 0)),
4426 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
4427 def : Pat<(insert_subvector undef, (v1f64 FPR64:$src), (i32 0)),
4428 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
4429 def : Pat<(insert_subvector undef, (v2i32 FPR64:$src), (i32 0)),
4430 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
4431 def : Pat<(insert_subvector undef, (v2f32 FPR64:$src), (i32 0)),
4432 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
4433 def : Pat<(insert_subvector undef, (v4i16 FPR64:$src), (i32 0)),
4434 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
4435 def : Pat<(insert_subvector undef, (v8i8 FPR64:$src), (i32 0)),
4436 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
4438 // Use pair-wise add instructions when summing up the lanes for v2f64, v2i64
4440 def : Pat<(i64 (add (vector_extract (v2i64 FPR128:$Rn), (i64 0)),
4441 (vector_extract (v2i64 FPR128:$Rn), (i64 1)))),
4442 (i64 (ADDPv2i64p (v2i64 FPR128:$Rn)))>;
4443 def : Pat<(f64 (fadd (vector_extract (v2f64 FPR128:$Rn), (i64 0)),
4444 (vector_extract (v2f64 FPR128:$Rn), (i64 1)))),
4445 (f64 (FADDPv2i64p (v2f64 FPR128:$Rn)))>;
4446 // vector_extract on 64-bit vectors gets promoted to a 128 bit vector,
4447 // so we match on v4f32 here, not v2f32. This will also catch adding
4448 // the low two lanes of a true v4f32 vector.
4449 def : Pat<(fadd (vector_extract (v4f32 FPR128:$Rn), (i64 0)),
4450 (vector_extract (v4f32 FPR128:$Rn), (i64 1))),
4451 (f32 (FADDPv2i32p (EXTRACT_SUBREG FPR128:$Rn, dsub)))>;
4453 // Scalar 64-bit shifts in FPR64 registers.
4454 def : Pat<(i64 (int_arm64_neon_sshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
4455 (SSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
4456 def : Pat<(i64 (int_arm64_neon_ushl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
4457 (USHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
4458 def : Pat<(i64 (int_arm64_neon_srshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
4459 (SRSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
4460 def : Pat<(i64 (int_arm64_neon_urshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
4461 (URSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
4463 // Tail call return handling. These are all compiler pseudo-instructions,
4464 // so no encoding information or anything like that.
4465 let isCall = 1, isTerminator = 1, isReturn = 1, isBarrier = 1, Uses = [SP] in {
4466 def TCRETURNdi : Pseudo<(outs), (ins i64imm:$dst), []>;
4467 def TCRETURNri : Pseudo<(outs), (ins tcGPR64:$dst), []>;
4470 def : Pat<(ARM64tcret tcGPR64:$dst), (TCRETURNri tcGPR64:$dst)>;
4471 def : Pat<(ARM64tcret (i64 tglobaladdr:$dst)), (TCRETURNdi texternalsym:$dst)>;
4472 def : Pat<(ARM64tcret (i64 texternalsym:$dst)), (TCRETURNdi texternalsym:$dst)>;
4474 include "ARM64InstrAtomics.td"