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 SYSxt : SystemXtI<0, "sys">;
336 def SYSLxt : SystemLXtI<1, "sysl">;
338 def : InstAlias<"sys $op1, $Cn, $Cm, $op2",
339 (SYSxt imm0_7:$op1, sys_cr_op:$Cn,
340 sys_cr_op:$Cm, imm0_7:$op2, XZR)>;
342 //===----------------------------------------------------------------------===//
343 // Move immediate instructions.
344 //===----------------------------------------------------------------------===//
346 defm MOVK : InsertImmediate<0b11, "movk">;
347 defm MOVN : MoveImmediate<0b00, "movn">;
349 let PostEncoderMethod = "fixMOVZ" in
350 defm MOVZ : MoveImmediate<0b10, "movz">;
352 def : InstAlias<"movk $dst, $imm", (MOVKWi GPR32:$dst, imm0_65535:$imm, 0)>;
353 def : InstAlias<"movk $dst, $imm", (MOVKXi GPR64:$dst, imm0_65535:$imm, 0)>;
354 def : InstAlias<"movn $dst, $imm", (MOVNWi GPR32:$dst, imm0_65535:$imm, 0)>;
355 def : InstAlias<"movn $dst, $imm", (MOVNXi GPR64:$dst, imm0_65535:$imm, 0)>;
356 def : InstAlias<"movz $dst, $imm", (MOVZWi GPR32:$dst, imm0_65535:$imm, 0)>;
357 def : InstAlias<"movz $dst, $imm", (MOVZXi GPR64:$dst, imm0_65535:$imm, 0)>;
359 def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movz_symbol_g3:$sym, 48)>;
360 def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movz_symbol_g2:$sym, 32)>;
361 def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movz_symbol_g1:$sym, 16)>;
362 def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movz_symbol_g0:$sym, 0)>;
364 def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movz_symbol_g3:$sym, 48)>;
365 def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movz_symbol_g2:$sym, 32)>;
366 def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movz_symbol_g1:$sym, 16)>;
367 def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movz_symbol_g0:$sym, 0)>;
369 def : InstAlias<"movz $Rd, $sym", (MOVZWi GPR32:$Rd, movz_symbol_g3:$sym, 48)>;
370 def : InstAlias<"movz $Rd, $sym", (MOVZWi GPR32:$Rd, movz_symbol_g2:$sym, 32)>;
371 def : InstAlias<"movz $Rd, $sym", (MOVZWi GPR32:$Rd, movz_symbol_g1:$sym, 16)>;
372 def : InstAlias<"movz $Rd, $sym", (MOVZWi GPR32:$Rd, movz_symbol_g0:$sym, 0)>;
374 def : InstAlias<"movk $Rd, $sym", (MOVKXi GPR64:$Rd, movk_symbol_g2:$sym, 32)>;
375 def : InstAlias<"movk $Rd, $sym", (MOVKXi GPR64:$Rd, movk_symbol_g1:$sym, 16)>;
376 def : InstAlias<"movk $Rd, $sym", (MOVKXi GPR64:$Rd, movk_symbol_g0:$sym, 0)>;
378 def : InstAlias<"movk $Rd, $sym", (MOVKWi GPR32:$Rd, movk_symbol_g2:$sym, 32)>;
379 def : InstAlias<"movk $Rd, $sym", (MOVKWi GPR32:$Rd, movk_symbol_g1:$sym, 16)>;
380 def : InstAlias<"movk $Rd, $sym", (MOVKWi GPR32:$Rd, movk_symbol_g0:$sym, 0)>;
382 let isReMaterializable = 1, isCodeGenOnly = 1, isMoveImm = 1,
383 isAsCheapAsAMove = 1 in {
384 // FIXME: The following pseudo instructions are only needed because remat
385 // cannot handle multiple instructions. When that changes, we can select
386 // directly to the real instructions and get rid of these pseudos.
389 : Pseudo<(outs GPR32:$dst), (ins i32imm:$src),
390 [(set GPR32:$dst, imm:$src)]>,
393 : Pseudo<(outs GPR64:$dst), (ins i64imm:$src),
394 [(set GPR64:$dst, imm:$src)]>,
396 } // isReMaterializable, isCodeGenOnly
398 def : Pat<(ARM64WrapperLarge tglobaladdr:$g3, tglobaladdr:$g2,
399 tglobaladdr:$g1, tglobaladdr:$g0),
400 (MOVKXi (MOVKXi (MOVKXi (MOVZXi tglobaladdr:$g3, 48),
401 tglobaladdr:$g2, 32),
402 tglobaladdr:$g1, 16),
403 tglobaladdr:$g0, 0)>;
405 def : Pat<(ARM64WrapperLarge tblockaddress:$g3, tblockaddress:$g2,
406 tblockaddress:$g1, tblockaddress:$g0),
407 (MOVKXi (MOVKXi (MOVKXi (MOVZXi tblockaddress:$g3, 48),
408 tblockaddress:$g2, 32),
409 tblockaddress:$g1, 16),
410 tblockaddress:$g0, 0)>;
412 def : Pat<(ARM64WrapperLarge tconstpool:$g3, tconstpool:$g2,
413 tconstpool:$g1, tconstpool:$g0),
414 (MOVKXi (MOVKXi (MOVKXi (MOVZXi tconstpool:$g3, 48),
420 //===----------------------------------------------------------------------===//
421 // Arithmetic instructions.
422 //===----------------------------------------------------------------------===//
424 // Add/subtract with carry.
425 defm ADC : AddSubCarry<0, "adc", "adcs", ARM64adc, ARM64adc_flag>;
426 defm SBC : AddSubCarry<1, "sbc", "sbcs", ARM64sbc, ARM64sbc_flag>;
428 def : InstAlias<"ngc $dst, $src", (SBCWr GPR32:$dst, WZR, GPR32:$src)>;
429 def : InstAlias<"ngc $dst, $src", (SBCXr GPR64:$dst, XZR, GPR64:$src)>;
430 def : InstAlias<"ngcs $dst, $src", (SBCSWr GPR32:$dst, WZR, GPR32:$src)>;
431 def : InstAlias<"ngcs $dst, $src", (SBCSXr GPR64:$dst, XZR, GPR64:$src)>;
434 defm ADD : AddSub<0, "add", add>;
435 defm SUB : AddSub<1, "sub">;
437 defm ADDS : AddSubS<0, "adds", ARM64add_flag>;
438 defm SUBS : AddSubS<1, "subs", ARM64sub_flag>;
440 // Use SUBS instead of SUB to enable CSE between SUBS and SUB.
441 def : Pat<(sub GPR32sp:$Rn, addsub_shifted_imm32:$imm),
442 (SUBSWri GPR32sp:$Rn, addsub_shifted_imm32:$imm)>;
443 def : Pat<(sub GPR64sp:$Rn, addsub_shifted_imm64:$imm),
444 (SUBSXri GPR64sp:$Rn, addsub_shifted_imm64:$imm)>;
445 def : Pat<(sub GPR32:$Rn, GPR32:$Rm),
446 (SUBSWrr GPR32:$Rn, GPR32:$Rm)>;
447 def : Pat<(sub GPR64:$Rn, GPR64:$Rm),
448 (SUBSXrr GPR64:$Rn, GPR64:$Rm)>;
449 def : Pat<(sub GPR32:$Rn, arith_shifted_reg32:$Rm),
450 (SUBSWrs GPR32:$Rn, arith_shifted_reg32:$Rm)>;
451 def : Pat<(sub GPR64:$Rn, arith_shifted_reg64:$Rm),
452 (SUBSXrs GPR64:$Rn, arith_shifted_reg64:$Rm)>;
453 def : Pat<(sub GPR32sp:$R2, arith_extended_reg32<i32>:$R3),
454 (SUBSWrx GPR32sp:$R2, arith_extended_reg32<i32>:$R3)>;
455 def : Pat<(sub GPR64sp:$R2, arith_extended_reg32to64<i64>:$R3),
456 (SUBSXrx GPR64sp:$R2, arith_extended_reg32to64<i64>:$R3)>;
458 // Because of the immediate format for add/sub-imm instructions, the
459 // expression (add x, -1) must be transformed to (SUB{W,X}ri x, 1).
460 // These patterns capture that transformation.
461 let AddedComplexity = 1 in {
462 def : Pat<(add GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
463 (SUBSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
464 def : Pat<(add GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
465 (SUBSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
466 def : Pat<(sub GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
467 (ADDWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
468 def : Pat<(sub GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
469 (ADDXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
472 def : InstAlias<"neg $dst, $src", (SUBWrs GPR32:$dst, WZR, GPR32:$src, 0)>;
473 def : InstAlias<"neg $dst, $src", (SUBXrs GPR64:$dst, XZR, GPR64:$src, 0)>;
474 def : InstAlias<"neg $dst, $src, $shift",
475 (SUBWrs GPR32:$dst, WZR, GPR32:$src, arith_shift:$shift)>;
476 def : InstAlias<"neg $dst, $src, $shift",
477 (SUBXrs GPR64:$dst, XZR, GPR64:$src, arith_shift:$shift)>;
479 // Because of the immediate format for add/sub-imm instructions, the
480 // expression (add x, -1) must be transformed to (SUB{W,X}ri x, 1).
481 // These patterns capture that transformation.
482 let AddedComplexity = 1 in {
483 def : Pat<(ARM64add_flag GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
484 (SUBSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
485 def : Pat<(ARM64add_flag GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
486 (SUBSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
487 def : Pat<(ARM64sub_flag GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
488 (ADDSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
489 def : Pat<(ARM64sub_flag GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
490 (ADDSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
493 def : InstAlias<"negs $dst, $src", (SUBSWrs GPR32:$dst, WZR, GPR32:$src, 0)>;
494 def : InstAlias<"negs $dst, $src", (SUBSXrs GPR64:$dst, XZR, GPR64:$src, 0)>;
495 def : InstAlias<"negs $dst, $src, $shift",
496 (SUBSWrs GPR32:$dst, WZR, GPR32:$src, arith_shift:$shift)>;
497 def : InstAlias<"negs $dst, $src, $shift",
498 (SUBSXrs GPR64:$dst, XZR, GPR64:$src, arith_shift:$shift)>;
500 // Unsigned/Signed divide
501 defm UDIV : Div<0, "udiv", udiv>;
502 defm SDIV : Div<1, "sdiv", sdiv>;
503 let isCodeGenOnly = 1 in {
504 defm UDIV_Int : Div<0, "udiv", int_arm64_udiv>;
505 defm SDIV_Int : Div<1, "sdiv", int_arm64_sdiv>;
509 defm ASRV : Shift<0b10, "asrv", sra>;
510 defm LSLV : Shift<0b00, "lslv", shl>;
511 defm LSRV : Shift<0b01, "lsrv", srl>;
512 defm RORV : Shift<0b11, "rorv", rotr>;
514 def : ShiftAlias<"asr", ASRVWr, GPR32>;
515 def : ShiftAlias<"asr", ASRVXr, GPR64>;
516 def : ShiftAlias<"lsl", LSLVWr, GPR32>;
517 def : ShiftAlias<"lsl", LSLVXr, GPR64>;
518 def : ShiftAlias<"lsr", LSRVWr, GPR32>;
519 def : ShiftAlias<"lsr", LSRVXr, GPR64>;
520 def : ShiftAlias<"ror", RORVWr, GPR32>;
521 def : ShiftAlias<"ror", RORVXr, GPR64>;
524 let AddedComplexity = 7 in {
525 defm MADD : MulAccum<0, "madd", add>;
526 defm MSUB : MulAccum<1, "msub", sub>;
528 def : Pat<(i32 (mul GPR32:$Rn, GPR32:$Rm)),
529 (MADDWrrr GPR32:$Rn, GPR32:$Rm, WZR)>;
530 def : Pat<(i64 (mul GPR64:$Rn, GPR64:$Rm)),
531 (MADDXrrr GPR64:$Rn, GPR64:$Rm, XZR)>;
533 def : Pat<(i32 (ineg (mul GPR32:$Rn, GPR32:$Rm))),
534 (MSUBWrrr GPR32:$Rn, GPR32:$Rm, WZR)>;
535 def : Pat<(i64 (ineg (mul GPR64:$Rn, GPR64:$Rm))),
536 (MSUBXrrr GPR64:$Rn, GPR64:$Rm, XZR)>;
537 } // AddedComplexity = 7
539 let AddedComplexity = 5 in {
540 def SMADDLrrr : WideMulAccum<0, 0b001, "smaddl", add, sext>;
541 def SMSUBLrrr : WideMulAccum<1, 0b001, "smsubl", sub, sext>;
542 def UMADDLrrr : WideMulAccum<0, 0b101, "umaddl", add, zext>;
543 def UMSUBLrrr : WideMulAccum<1, 0b101, "umsubl", sub, zext>;
545 def : Pat<(i64 (mul (sext GPR32:$Rn), (sext GPR32:$Rm))),
546 (SMADDLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
547 def : Pat<(i64 (mul (zext GPR32:$Rn), (zext GPR32:$Rm))),
548 (UMADDLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
550 def : Pat<(i64 (ineg (mul (sext GPR32:$Rn), (sext GPR32:$Rm)))),
551 (SMSUBLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
552 def : Pat<(i64 (ineg (mul (zext GPR32:$Rn), (zext GPR32:$Rm)))),
553 (UMSUBLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
554 } // AddedComplexity = 5
556 def : MulAccumWAlias<"mul", MADDWrrr>;
557 def : MulAccumXAlias<"mul", MADDXrrr>;
558 def : MulAccumWAlias<"mneg", MSUBWrrr>;
559 def : MulAccumXAlias<"mneg", MSUBXrrr>;
560 def : WideMulAccumAlias<"smull", SMADDLrrr>;
561 def : WideMulAccumAlias<"smnegl", SMSUBLrrr>;
562 def : WideMulAccumAlias<"umull", UMADDLrrr>;
563 def : WideMulAccumAlias<"umnegl", UMSUBLrrr>;
566 def SMULHrr : MulHi<0b010, "smulh", mulhs>;
567 def UMULHrr : MulHi<0b110, "umulh", mulhu>;
570 def CRC32Brr : BaseCRC32<0, 0b00, 0, GPR32, int_arm64_crc32b, "crc32b">;
571 def CRC32Hrr : BaseCRC32<0, 0b01, 0, GPR32, int_arm64_crc32h, "crc32h">;
572 def CRC32Wrr : BaseCRC32<0, 0b10, 0, GPR32, int_arm64_crc32w, "crc32w">;
573 def CRC32Xrr : BaseCRC32<1, 0b11, 0, GPR64, int_arm64_crc32x, "crc32x">;
575 def CRC32CBrr : BaseCRC32<0, 0b00, 1, GPR32, int_arm64_crc32cb, "crc32cb">;
576 def CRC32CHrr : BaseCRC32<0, 0b01, 1, GPR32, int_arm64_crc32ch, "crc32ch">;
577 def CRC32CWrr : BaseCRC32<0, 0b10, 1, GPR32, int_arm64_crc32cw, "crc32cw">;
578 def CRC32CXrr : BaseCRC32<1, 0b11, 1, GPR64, int_arm64_crc32cx, "crc32cx">;
581 //===----------------------------------------------------------------------===//
582 // Logical instructions.
583 //===----------------------------------------------------------------------===//
586 defm ANDS : LogicalImmS<0b11, "ands", ARM64and_flag>;
587 defm AND : LogicalImm<0b00, "and", and>;
588 defm EOR : LogicalImm<0b10, "eor", xor>;
589 defm ORR : LogicalImm<0b01, "orr", or>;
591 def : InstAlias<"mov $dst, $imm", (ORRWri GPR32sp:$dst, WZR,
592 logical_imm32:$imm)>;
593 def : InstAlias<"mov $dst, $imm", (ORRXri GPR64sp:$dst, XZR,
594 logical_imm64:$imm)>;
598 defm ANDS : LogicalRegS<0b11, 0, "ands">;
599 defm BICS : LogicalRegS<0b11, 1, "bics">;
600 defm AND : LogicalReg<0b00, 0, "and", and>;
601 defm BIC : LogicalReg<0b00, 1, "bic",
602 BinOpFrag<(and node:$LHS, (not node:$RHS))>>;
603 defm EON : LogicalReg<0b10, 1, "eon",
604 BinOpFrag<(xor node:$LHS, (not node:$RHS))>>;
605 defm EOR : LogicalReg<0b10, 0, "eor", xor>;
606 defm ORN : LogicalReg<0b01, 1, "orn",
607 BinOpFrag<(or node:$LHS, (not node:$RHS))>>;
608 defm ORR : LogicalReg<0b01, 0, "orr", or>;
610 def : InstAlias<"tst $src1, $src2",
611 (ANDSWri WZR, GPR32:$src1, logical_imm32:$src2)>;
612 def : InstAlias<"tst $src1, $src2",
613 (ANDSXri XZR, GPR64:$src1, logical_imm64:$src2)>;
615 def : InstAlias<"tst $src1, $src2",
616 (ANDSWrs WZR, GPR32:$src1, GPR32:$src2, 0)>;
617 def : InstAlias<"tst $src1, $src2",
618 (ANDSXrs XZR, GPR64:$src1, GPR64:$src2, 0)>;
620 def : InstAlias<"tst $src1, $src2, $sh",
621 (ANDSWrs WZR, GPR32:$src1, GPR32:$src2, logical_shift:$sh)>;
622 def : InstAlias<"tst $src1, $src2, $sh",
623 (ANDSXrs XZR, GPR64:$src1, GPR64:$src2, logical_shift:$sh)>;
625 def : InstAlias<"mvn $Wd, $Wm",
626 (ORNWrs GPR32:$Wd, WZR, GPR32:$Wm, 0)>;
627 def : InstAlias<"mvn $Xd, $Xm",
628 (ORNXrs GPR64:$Xd, XZR, GPR64:$Xm, 0)>;
630 def : InstAlias<"mvn $Wd, $Wm, $sh",
631 (ORNWrs GPR32:$Wd, WZR, GPR32:$Wm, logical_shift:$sh)>;
632 def : InstAlias<"mvn $Xd, $Xm, $sh",
633 (ORNXrs GPR64:$Xd, XZR, GPR64:$Xm, logical_shift:$sh)>;
635 def : Pat<(not GPR32:$Wm), (ORNWrr WZR, GPR32:$Wm)>;
636 def : Pat<(not GPR64:$Xm), (ORNXrr XZR, GPR64:$Xm)>;
639 //===----------------------------------------------------------------------===//
640 // One operand data processing instructions.
641 //===----------------------------------------------------------------------===//
643 defm CLS : OneOperandData<0b101, "cls">;
644 defm CLZ : OneOperandData<0b100, "clz", ctlz>;
645 defm RBIT : OneOperandData<0b000, "rbit">;
646 def REV16Wr : OneWRegData<0b001, "rev16",
647 UnOpFrag<(rotr (bswap node:$LHS), (i64 16))>>;
648 def REV16Xr : OneXRegData<0b001, "rev16", null_frag>;
650 def : Pat<(cttz GPR32:$Rn),
651 (CLZWr (RBITWr GPR32:$Rn))>;
652 def : Pat<(cttz GPR64:$Rn),
653 (CLZXr (RBITXr GPR64:$Rn))>;
654 def : Pat<(ctlz (or (shl (xor (sra GPR32:$Rn, (i64 31)), GPR32:$Rn), (i64 1)),
657 def : Pat<(ctlz (or (shl (xor (sra GPR64:$Rn, (i64 63)), GPR64:$Rn), (i64 1)),
661 // Unlike the other one operand instructions, the instructions with the "rev"
662 // mnemonic do *not* just different in the size bit, but actually use different
663 // opcode bits for the different sizes.
664 def REVWr : OneWRegData<0b010, "rev", bswap>;
665 def REVXr : OneXRegData<0b011, "rev", bswap>;
666 def REV32Xr : OneXRegData<0b010, "rev32",
667 UnOpFrag<(rotr (bswap node:$LHS), (i64 32))>>;
669 // The bswap commutes with the rotr so we want a pattern for both possible
671 def : Pat<(bswap (rotr GPR32:$Rn, (i64 16))), (REV16Wr GPR32:$Rn)>;
672 def : Pat<(bswap (rotr GPR64:$Rn, (i64 32))), (REV32Xr GPR64:$Rn)>;
674 //===----------------------------------------------------------------------===//
675 // Bitfield immediate extraction instruction.
676 //===----------------------------------------------------------------------===//
677 let neverHasSideEffects = 1 in
678 defm EXTR : ExtractImm<"extr">;
679 def : InstAlias<"ror $dst, $src, $shift",
680 (EXTRWrri GPR32:$dst, GPR32:$src, GPR32:$src, imm0_31:$shift)>;
681 def : InstAlias<"ror $dst, $src, $shift",
682 (EXTRXrri GPR64:$dst, GPR64:$src, GPR64:$src, imm0_63:$shift)>;
684 def : Pat<(rotr GPR32:$Rn, (i64 imm0_31:$imm)),
685 (EXTRWrri GPR32:$Rn, GPR32:$Rn, imm0_31:$imm)>;
686 def : Pat<(rotr GPR64:$Rn, (i64 imm0_63:$imm)),
687 (EXTRXrri GPR64:$Rn, GPR64:$Rn, imm0_63:$imm)>;
689 //===----------------------------------------------------------------------===//
690 // Other bitfield immediate instructions.
691 //===----------------------------------------------------------------------===//
692 let neverHasSideEffects = 1 in {
693 defm BFM : BitfieldImmWith2RegArgs<0b01, "bfm">;
694 defm SBFM : BitfieldImm<0b00, "sbfm">;
695 defm UBFM : BitfieldImm<0b10, "ubfm">;
698 def i32shift_a : Operand<i64>, SDNodeXForm<imm, [{
699 uint64_t enc = (32 - N->getZExtValue()) & 0x1f;
700 return CurDAG->getTargetConstant(enc, MVT::i64);
703 def i32shift_b : Operand<i64>, SDNodeXForm<imm, [{
704 uint64_t enc = 31 - N->getZExtValue();
705 return CurDAG->getTargetConstant(enc, MVT::i64);
708 // min(7, 31 - shift_amt)
709 def i32shift_sext_i8 : Operand<i64>, SDNodeXForm<imm, [{
710 uint64_t enc = 31 - N->getZExtValue();
711 enc = enc > 7 ? 7 : enc;
712 return CurDAG->getTargetConstant(enc, MVT::i64);
715 // min(15, 31 - shift_amt)
716 def i32shift_sext_i16 : Operand<i64>, SDNodeXForm<imm, [{
717 uint64_t enc = 31 - N->getZExtValue();
718 enc = enc > 15 ? 15 : enc;
719 return CurDAG->getTargetConstant(enc, MVT::i64);
722 def i64shift_a : Operand<i64>, SDNodeXForm<imm, [{
723 uint64_t enc = (64 - N->getZExtValue()) & 0x3f;
724 return CurDAG->getTargetConstant(enc, MVT::i64);
727 def i64shift_b : Operand<i64>, SDNodeXForm<imm, [{
728 uint64_t enc = 63 - N->getZExtValue();
729 return CurDAG->getTargetConstant(enc, MVT::i64);
732 // min(7, 63 - shift_amt)
733 def i64shift_sext_i8 : Operand<i64>, SDNodeXForm<imm, [{
734 uint64_t enc = 63 - N->getZExtValue();
735 enc = enc > 7 ? 7 : enc;
736 return CurDAG->getTargetConstant(enc, MVT::i64);
739 // min(15, 63 - shift_amt)
740 def i64shift_sext_i16 : Operand<i64>, SDNodeXForm<imm, [{
741 uint64_t enc = 63 - N->getZExtValue();
742 enc = enc > 15 ? 15 : enc;
743 return CurDAG->getTargetConstant(enc, MVT::i64);
746 // min(31, 63 - shift_amt)
747 def i64shift_sext_i32 : Operand<i64>, SDNodeXForm<imm, [{
748 uint64_t enc = 63 - N->getZExtValue();
749 enc = enc > 31 ? 31 : enc;
750 return CurDAG->getTargetConstant(enc, MVT::i64);
753 def : Pat<(shl GPR32:$Rn, (i64 imm0_31:$imm)),
754 (UBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
755 (i64 (i32shift_b imm0_31:$imm)))>;
756 def : Pat<(shl GPR64:$Rn, (i64 imm0_63:$imm)),
757 (UBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
758 (i64 (i64shift_b imm0_63:$imm)))>;
760 let AddedComplexity = 10 in {
761 def : Pat<(sra GPR32:$Rn, (i64 imm0_31:$imm)),
762 (SBFMWri GPR32:$Rn, imm0_31:$imm, 31)>;
763 def : Pat<(sra GPR64:$Rn, (i64 imm0_63:$imm)),
764 (SBFMXri GPR64:$Rn, imm0_63:$imm, 63)>;
767 def : InstAlias<"asr $dst, $src, $shift",
768 (SBFMWri GPR32:$dst, GPR32:$src, imm0_31:$shift, 31)>;
769 def : InstAlias<"asr $dst, $src, $shift",
770 (SBFMXri GPR64:$dst, GPR64:$src, imm0_63:$shift, 63)>;
771 def : InstAlias<"sxtb $dst, $src", (SBFMWri GPR32:$dst, GPR32:$src, 0, 7)>;
772 def : InstAlias<"sxtb $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 7)>;
773 def : InstAlias<"sxth $dst, $src", (SBFMWri GPR32:$dst, GPR32:$src, 0, 15)>;
774 def : InstAlias<"sxth $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 15)>;
775 def : InstAlias<"sxtw $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 31)>;
777 def : Pat<(srl GPR32:$Rn, (i64 imm0_31:$imm)),
778 (UBFMWri GPR32:$Rn, imm0_31:$imm, 31)>;
779 def : Pat<(srl GPR64:$Rn, (i64 imm0_63:$imm)),
780 (UBFMXri GPR64:$Rn, imm0_63:$imm, 63)>;
782 def : InstAlias<"lsr $dst, $src, $shift",
783 (UBFMWri GPR32:$dst, GPR32:$src, imm0_31:$shift, 31)>;
784 def : InstAlias<"lsr $dst, $src, $shift",
785 (UBFMXri GPR64:$dst, GPR64:$src, imm0_63:$shift, 63)>;
786 def : InstAlias<"uxtb $dst, $src", (UBFMWri GPR32:$dst, GPR32:$src, 0, 7)>;
787 def : InstAlias<"uxtb $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 7)>;
788 def : InstAlias<"uxth $dst, $src", (UBFMWri GPR32:$dst, GPR32:$src, 0, 15)>;
789 def : InstAlias<"uxth $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 15)>;
790 def : InstAlias<"uxtw $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 31)>;
792 //===----------------------------------------------------------------------===//
793 // Conditionally set flags instructions.
794 //===----------------------------------------------------------------------===//
795 defm CCMN : CondSetFlagsImm<0, "ccmn">;
796 defm CCMP : CondSetFlagsImm<1, "ccmp">;
798 defm CCMN : CondSetFlagsReg<0, "ccmn">;
799 defm CCMP : CondSetFlagsReg<1, "ccmp">;
801 //===----------------------------------------------------------------------===//
802 // Conditional select instructions.
803 //===----------------------------------------------------------------------===//
804 defm CSEL : CondSelect<0, 0b00, "csel">;
806 def inc : PatFrag<(ops node:$in), (add node:$in, 1)>;
807 defm CSINC : CondSelectOp<0, 0b01, "csinc", inc>;
808 defm CSINV : CondSelectOp<1, 0b00, "csinv", not>;
809 defm CSNEG : CondSelectOp<1, 0b01, "csneg", ineg>;
811 def : Pat<(ARM64csinv GPR32:$tval, GPR32:$fval, (i32 imm:$cc), CPSR),
812 (CSINVWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
813 def : Pat<(ARM64csinv GPR64:$tval, GPR64:$fval, (i32 imm:$cc), CPSR),
814 (CSINVXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
815 def : Pat<(ARM64csneg GPR32:$tval, GPR32:$fval, (i32 imm:$cc), CPSR),
816 (CSNEGWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
817 def : Pat<(ARM64csneg GPR64:$tval, GPR64:$fval, (i32 imm:$cc), CPSR),
818 (CSNEGXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
819 def : Pat<(ARM64csinc GPR32:$tval, GPR32:$fval, (i32 imm:$cc), CPSR),
820 (CSINCWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
821 def : Pat<(ARM64csinc GPR64:$tval, GPR64:$fval, (i32 imm:$cc), CPSR),
822 (CSINCXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
824 def : Pat<(ARM64csel (i32 0), (i32 1), (i32 imm:$cc), CPSR),
825 (CSINCWr WZR, WZR, (i32 imm:$cc))>;
826 def : Pat<(ARM64csel (i64 0), (i64 1), (i32 imm:$cc), CPSR),
827 (CSINCXr XZR, XZR, (i32 imm:$cc))>;
828 def : Pat<(ARM64csel (i32 0), (i32 -1), (i32 imm:$cc), CPSR),
829 (CSINVWr WZR, WZR, (i32 imm:$cc))>;
830 def : Pat<(ARM64csel (i64 0), (i64 -1), (i32 imm:$cc), CPSR),
831 (CSINVXr XZR, XZR, (i32 imm:$cc))>;
833 // The inverse of the condition code from the alias instruction is what is used
834 // in the aliased instruction. The parser all ready inverts the condition code
835 // for these aliases.
836 // FIXME: Is this the correct way to handle these aliases?
837 def : InstAlias<"cset $dst, $cc", (CSINCWr GPR32:$dst, WZR, WZR, ccode:$cc)>;
838 def : InstAlias<"cset $dst, $cc", (CSINCXr GPR64:$dst, XZR, XZR, ccode:$cc)>;
840 def : InstAlias<"csetm $dst, $cc", (CSINVWr GPR32:$dst, WZR, WZR, ccode:$cc)>;
841 def : InstAlias<"csetm $dst, $cc", (CSINVXr GPR64:$dst, XZR, XZR, ccode:$cc)>;
843 def : InstAlias<"cinc $dst, $src, $cc",
844 (CSINCWr GPR32:$dst, GPR32:$src, GPR32:$src, ccode:$cc)>;
845 def : InstAlias<"cinc $dst, $src, $cc",
846 (CSINCXr GPR64:$dst, GPR64:$src, GPR64:$src, ccode:$cc)>;
848 def : InstAlias<"cinv $dst, $src, $cc",
849 (CSINVWr GPR32:$dst, GPR32:$src, GPR32:$src, ccode:$cc)>;
850 def : InstAlias<"cinv $dst, $src, $cc",
851 (CSINVXr GPR64:$dst, GPR64:$src, GPR64:$src, ccode:$cc)>;
853 def : InstAlias<"cneg $dst, $src, $cc",
854 (CSNEGWr GPR32:$dst, GPR32:$src, GPR32:$src, ccode:$cc)>;
855 def : InstAlias<"cneg $dst, $src, $cc",
856 (CSNEGXr GPR64:$dst, GPR64:$src, GPR64:$src, ccode:$cc)>;
858 //===----------------------------------------------------------------------===//
859 // PC-relative instructions.
860 //===----------------------------------------------------------------------===//
861 let isReMaterializable = 1 in {
862 let neverHasSideEffects = 1, mayStore = 0, mayLoad = 0 in {
863 def ADR : ADRI<0, "adr", adrlabel, []>;
864 } // neverHasSideEffects = 1
866 def ADRP : ADRI<1, "adrp", adrplabel,
867 [(set GPR64:$Xd, (ARM64adrp tglobaladdr:$label))]>;
868 } // isReMaterializable = 1
870 // page address of a constant pool entry, block address
871 def : Pat<(ARM64adrp tconstpool:$cp), (ADRP tconstpool:$cp)>;
872 def : Pat<(ARM64adrp tblockaddress:$cp), (ADRP tblockaddress:$cp)>;
874 //===----------------------------------------------------------------------===//
875 // Unconditional branch (register) instructions.
876 //===----------------------------------------------------------------------===//
878 let isReturn = 1, isTerminator = 1, isBarrier = 1 in {
879 def RET : BranchReg<0b0010, "ret", []>;
880 def DRPS : SpecialReturn<0b0101, "drps">;
881 def ERET : SpecialReturn<0b0100, "eret">;
882 } // isReturn = 1, isTerminator = 1, isBarrier = 1
884 // Default to the LR register.
885 def : InstAlias<"ret", (RET LR)>;
887 let isCall = 1, Defs = [LR], Uses = [SP] in {
888 def BLR : BranchReg<0b0001, "blr", [(ARM64call GPR64:$Rn)]>;
891 let isBranch = 1, isTerminator = 1, isBarrier = 1, isIndirectBranch = 1 in {
892 def BR : BranchReg<0b0000, "br", [(brind GPR64:$Rn)]>;
893 } // isBranch, isTerminator, isBarrier, isIndirectBranch
895 // Create a separate pseudo-instruction for codegen to use so that we don't
896 // flag lr as used in every function. It'll be restored before the RET by the
897 // epilogue if it's legitimately used.
898 def RET_ReallyLR : Pseudo<(outs), (ins), [(ARM64retflag)]> {
899 let isTerminator = 1;
904 // This is a directive-like pseudo-instruction. The purpose is to insert an
905 // R_AARCH64_TLSDESC_CALL relocation at the offset of the following instruction
906 // (which in the usual case is a BLR).
907 let hasSideEffects = 1 in
908 def TLSDESCCALL : Pseudo<(outs), (ins i64imm:$sym), []> {
909 let AsmString = ".tlsdesccall $sym";
912 // Pseudo-instruction representing a BLR with attached TLSDESC relocation. It
913 // gets expanded to two MCInsts during lowering.
914 let isCall = 1, Defs = [LR] in
916 : Pseudo<(outs), (ins GPR64:$dest, i64imm:$sym),
917 [(ARM64tlsdesc_call GPR64:$dest, tglobaltlsaddr:$sym)]>;
919 def : Pat<(ARM64tlsdesc_call GPR64:$dest, texternalsym:$sym),
920 (TLSDESC_BLR GPR64:$dest, texternalsym:$sym)>;
921 //===----------------------------------------------------------------------===//
922 // Conditional branch (immediate) instruction.
923 //===----------------------------------------------------------------------===//
924 def Bcc : BranchCond;
926 //===----------------------------------------------------------------------===//
927 // Compare-and-branch instructions.
928 //===----------------------------------------------------------------------===//
929 defm CBZ : CmpBranch<0, "cbz", ARM64cbz>;
930 defm CBNZ : CmpBranch<1, "cbnz", ARM64cbnz>;
932 //===----------------------------------------------------------------------===//
933 // Test-bit-and-branch instructions.
934 //===----------------------------------------------------------------------===//
935 def TBZ : TestBranch<0, "tbz", ARM64tbz>;
936 def TBNZ : TestBranch<1, "tbnz", ARM64tbnz>;
938 //===----------------------------------------------------------------------===//
939 // Unconditional branch (immediate) instructions.
940 //===----------------------------------------------------------------------===//
941 let isBranch = 1, isTerminator = 1, isBarrier = 1 in {
942 def B : BranchImm<0, "b", [(br bb:$addr)]>;
943 } // isBranch, isTerminator, isBarrier
945 let isCall = 1, Defs = [LR], Uses = [SP] in {
946 def BL : CallImm<1, "bl", [(ARM64call tglobaladdr:$addr)]>;
948 def : Pat<(ARM64call texternalsym:$func), (BL texternalsym:$func)>;
950 //===----------------------------------------------------------------------===//
951 // Exception generation instructions.
952 //===----------------------------------------------------------------------===//
953 def BRK : ExceptionGeneration<0b001, 0b00, "brk">;
954 def DCPS1 : ExceptionGeneration<0b101, 0b01, "dcps1">;
955 def DCPS2 : ExceptionGeneration<0b101, 0b10, "dcps2">;
956 def DCPS3 : ExceptionGeneration<0b101, 0b11, "dcps3">;
957 def HLT : ExceptionGeneration<0b010, 0b00, "hlt">;
958 def HVC : ExceptionGeneration<0b000, 0b10, "hvc">;
959 def SMC : ExceptionGeneration<0b000, 0b11, "smc">;
960 def SVC : ExceptionGeneration<0b000, 0b01, "svc">;
962 // DCPSn defaults to an immediate operand of zero if unspecified.
963 def : InstAlias<"dcps1", (DCPS1 0)>;
964 def : InstAlias<"dcps2", (DCPS2 0)>;
965 def : InstAlias<"dcps3", (DCPS3 0)>;
967 //===----------------------------------------------------------------------===//
968 // Load instructions.
969 //===----------------------------------------------------------------------===//
971 // Pair (indexed, offset)
972 def LDPWi : LoadPairOffset<0b00, 0, GPR32, am_indexed32simm7, "ldp">;
973 def LDPXi : LoadPairOffset<0b10, 0, GPR64, am_indexed64simm7, "ldp">;
974 def LDPSi : LoadPairOffset<0b00, 1, FPR32, am_indexed32simm7, "ldp">;
975 def LDPDi : LoadPairOffset<0b01, 1, FPR64, am_indexed64simm7, "ldp">;
976 def LDPQi : LoadPairOffset<0b10, 1, FPR128, am_indexed128simm7, "ldp">;
978 def LDPSWi : LoadPairOffset<0b01, 0, GPR64, am_indexed32simm7, "ldpsw">;
980 // Pair (pre-indexed)
981 def LDPWpre : LoadPairPreIdx<0b00, 0, GPR32, am_indexed32simm7_wb, "ldp">;
982 def LDPXpre : LoadPairPreIdx<0b10, 0, GPR64, am_indexed64simm7_wb, "ldp">;
983 def LDPSpre : LoadPairPreIdx<0b00, 1, FPR32, am_indexed32simm7_wb, "ldp">;
984 def LDPDpre : LoadPairPreIdx<0b01, 1, FPR64, am_indexed64simm7_wb, "ldp">;
985 def LDPQpre : LoadPairPreIdx<0b10, 1, FPR128, am_indexed128simm7_wb, "ldp">;
987 def LDPSWpre : LoadPairPreIdx<0b01, 0, GPR64, am_indexed32simm7_wb, "ldpsw">;
989 // Pair (post-indexed)
990 def LDPWpost : LoadPairPostIdx<0b00, 0, GPR32, simm7s4, "ldp">;
991 def LDPXpost : LoadPairPostIdx<0b10, 0, GPR64, simm7s8, "ldp">;
992 def LDPSpost : LoadPairPostIdx<0b00, 1, FPR32, simm7s4, "ldp">;
993 def LDPDpost : LoadPairPostIdx<0b01, 1, FPR64, simm7s8, "ldp">;
994 def LDPQpost : LoadPairPostIdx<0b10, 1, FPR128, simm7s16, "ldp">;
996 def LDPSWpost : LoadPairPostIdx<0b01, 0, GPR64, simm7s4, "ldpsw">;
999 // Pair (no allocate)
1000 def LDNPWi : LoadPairNoAlloc<0b00, 0, GPR32, am_indexed32simm7, "ldnp">;
1001 def LDNPXi : LoadPairNoAlloc<0b10, 0, GPR64, am_indexed64simm7, "ldnp">;
1002 def LDNPSi : LoadPairNoAlloc<0b00, 1, FPR32, am_indexed32simm7, "ldnp">;
1003 def LDNPDi : LoadPairNoAlloc<0b01, 1, FPR64, am_indexed64simm7, "ldnp">;
1004 def LDNPQi : LoadPairNoAlloc<0b10, 1, FPR128, am_indexed128simm7, "ldnp">;
1007 // (register offset)
1010 let AddedComplexity = 10 in {
1012 def LDRBBro : Load8RO<0b00, 0, 0b01, GPR32, "ldrb",
1013 [(set GPR32:$Rt, (zextloadi8 ro_indexed8:$addr))]>;
1014 def LDRHHro : Load16RO<0b01, 0, 0b01, GPR32, "ldrh",
1015 [(set GPR32:$Rt, (zextloadi16 ro_indexed16:$addr))]>;
1016 def LDRWro : Load32RO<0b10, 0, 0b01, GPR32, "ldr",
1017 [(set GPR32:$Rt, (load ro_indexed32:$addr))]>;
1018 def LDRXro : Load64RO<0b11, 0, 0b01, GPR64, "ldr",
1019 [(set GPR64:$Rt, (load ro_indexed64:$addr))]>;
1022 def LDRBro : Load8RO<0b00, 1, 0b01, FPR8, "ldr",
1023 [(set FPR8:$Rt, (load ro_indexed8:$addr))]>;
1024 def LDRHro : Load16RO<0b01, 1, 0b01, FPR16, "ldr",
1025 [(set (f16 FPR16:$Rt), (load ro_indexed16:$addr))]>;
1026 def LDRSro : Load32RO<0b10, 1, 0b01, FPR32, "ldr",
1027 [(set (f32 FPR32:$Rt), (load ro_indexed32:$addr))]>;
1028 def LDRDro : Load64RO<0b11, 1, 0b01, FPR64, "ldr",
1029 [(set (f64 FPR64:$Rt), (load ro_indexed64:$addr))]>;
1030 def LDRQro : Load128RO<0b00, 1, 0b11, FPR128, "ldr", []> {
1034 // For regular load, we do not have any alignment requirement.
1035 // Thus, it is safe to directly map the vector loads with interesting
1036 // addressing modes.
1037 // FIXME: We could do the same for bitconvert to floating point vectors.
1038 def : Pat <(v8i8 (scalar_to_vector (i32 (extloadi8 ro_indexed8:$addr)))),
1039 (INSERT_SUBREG (v8i8 (IMPLICIT_DEF)),
1040 (LDRBro ro_indexed8:$addr), bsub)>;
1041 def : Pat <(v16i8 (scalar_to_vector (i32 (extloadi8 ro_indexed8:$addr)))),
1042 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
1043 (LDRBro ro_indexed8:$addr), bsub)>;
1044 def : Pat <(v4i16 (scalar_to_vector (i32 (extloadi16 ro_indexed16:$addr)))),
1045 (INSERT_SUBREG (v4i16 (IMPLICIT_DEF)),
1046 (LDRHro ro_indexed16:$addr), hsub)>;
1047 def : Pat <(v8i16 (scalar_to_vector (i32 (extloadi16 ro_indexed16:$addr)))),
1048 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
1049 (LDRHro ro_indexed16:$addr), hsub)>;
1050 def : Pat <(v2i32 (scalar_to_vector (i32 (load ro_indexed32:$addr)))),
1051 (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)),
1052 (LDRSro ro_indexed32:$addr), ssub)>;
1053 def : Pat <(v4i32 (scalar_to_vector (i32 (load ro_indexed32:$addr)))),
1054 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
1055 (LDRSro ro_indexed32:$addr), ssub)>;
1056 def : Pat <(v1i64 (scalar_to_vector (i64 (load ro_indexed64:$addr)))),
1057 (LDRDro ro_indexed64:$addr)>;
1058 def : Pat <(v2i64 (scalar_to_vector (i64 (load ro_indexed64:$addr)))),
1059 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
1060 (LDRDro ro_indexed64:$addr), dsub)>;
1062 // Match all load 64 bits width whose type is compatible with FPR64
1063 def : Pat<(v2f32 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1064 def : Pat<(v1f64 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1065 def : Pat<(v8i8 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1066 def : Pat<(v4i16 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1067 def : Pat<(v2i32 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1068 def : Pat<(v1i64 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1070 // Match all load 128 bits width whose type is compatible with FPR128
1071 def : Pat<(v4f32 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1072 def : Pat<(v2f64 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1073 def : Pat<(v16i8 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1074 def : Pat<(v8i16 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1075 def : Pat<(v4i32 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1076 def : Pat<(v2i64 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1077 def : Pat<(f128 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1079 // Load sign-extended half-word
1080 def LDRSHWro : Load16RO<0b01, 0, 0b11, GPR32, "ldrsh",
1081 [(set GPR32:$Rt, (sextloadi16 ro_indexed16:$addr))]>;
1082 def LDRSHXro : Load16RO<0b01, 0, 0b10, GPR64, "ldrsh",
1083 [(set GPR64:$Rt, (sextloadi16 ro_indexed16:$addr))]>;
1085 // Load sign-extended byte
1086 def LDRSBWro : Load8RO<0b00, 0, 0b11, GPR32, "ldrsb",
1087 [(set GPR32:$Rt, (sextloadi8 ro_indexed8:$addr))]>;
1088 def LDRSBXro : Load8RO<0b00, 0, 0b10, GPR64, "ldrsb",
1089 [(set GPR64:$Rt, (sextloadi8 ro_indexed8:$addr))]>;
1091 // Load sign-extended word
1092 def LDRSWro : Load32RO<0b10, 0, 0b10, GPR64, "ldrsw",
1093 [(set GPR64:$Rt, (sextloadi32 ro_indexed32:$addr))]>;
1096 def PRFMro : PrefetchRO<0b11, 0, 0b10, "prfm",
1097 [(ARM64Prefetch imm:$Rt, ro_indexed64:$addr)]>;
1100 def : Pat<(i64 (zextloadi8 ro_indexed8:$addr)),
1101 (SUBREG_TO_REG (i64 0), (LDRBBro ro_indexed8:$addr), sub_32)>;
1102 def : Pat<(i64 (zextloadi16 ro_indexed16:$addr)),
1103 (SUBREG_TO_REG (i64 0), (LDRHHro ro_indexed16:$addr), sub_32)>;
1105 // zextloadi1 -> zextloadi8
1106 def : Pat<(i32 (zextloadi1 ro_indexed8:$addr)), (LDRBBro ro_indexed8:$addr)>;
1107 def : Pat<(i64 (zextloadi1 ro_indexed8:$addr)),
1108 (SUBREG_TO_REG (i64 0), (LDRBBro ro_indexed8:$addr), sub_32)>;
1110 // extload -> zextload
1111 def : Pat<(i32 (extloadi16 ro_indexed16:$addr)), (LDRHHro ro_indexed16:$addr)>;
1112 def : Pat<(i32 (extloadi8 ro_indexed8:$addr)), (LDRBBro ro_indexed8:$addr)>;
1113 def : Pat<(i32 (extloadi1 ro_indexed8:$addr)), (LDRBBro ro_indexed8:$addr)>;
1114 def : Pat<(i64 (extloadi32 ro_indexed32:$addr)),
1115 (SUBREG_TO_REG (i64 0), (LDRWro ro_indexed32:$addr), sub_32)>;
1116 def : Pat<(i64 (extloadi16 ro_indexed16:$addr)),
1117 (SUBREG_TO_REG (i64 0), (LDRHHro ro_indexed16:$addr), sub_32)>;
1118 def : Pat<(i64 (extloadi8 ro_indexed8:$addr)),
1119 (SUBREG_TO_REG (i64 0), (LDRBBro ro_indexed8:$addr), sub_32)>;
1120 def : Pat<(i64 (extloadi1 ro_indexed8:$addr)),
1121 (SUBREG_TO_REG (i64 0), (LDRBBro ro_indexed8:$addr), sub_32)>;
1123 } // AddedComplexity = 10
1126 // (unsigned immediate)
1128 def LDRXui : LoadUI<0b11, 0, 0b01, GPR64, am_indexed64, "ldr",
1129 [(set GPR64:$Rt, (load am_indexed64:$addr))]>;
1130 def LDRWui : LoadUI<0b10, 0, 0b01, GPR32, am_indexed32, "ldr",
1131 [(set GPR32:$Rt, (load am_indexed32:$addr))]>;
1132 def LDRBui : LoadUI<0b00, 1, 0b01, FPR8, am_indexed8, "ldr",
1133 [(set FPR8:$Rt, (load am_indexed8:$addr))]>;
1134 def LDRHui : LoadUI<0b01, 1, 0b01, FPR16, am_indexed16, "ldr",
1135 [(set (f16 FPR16:$Rt), (load am_indexed16:$addr))]>;
1136 def LDRSui : LoadUI<0b10, 1, 0b01, FPR32, am_indexed32, "ldr",
1137 [(set (f32 FPR32:$Rt), (load am_indexed32:$addr))]>;
1138 def LDRDui : LoadUI<0b11, 1, 0b01, FPR64, am_indexed64, "ldr",
1139 [(set (f64 FPR64:$Rt), (load am_indexed64:$addr))]>;
1140 def LDRQui : LoadUI<0b00, 1, 0b11, FPR128, am_indexed128, "ldr",
1141 [(set (f128 FPR128:$Rt), (load am_indexed128:$addr))]>;
1143 // For regular load, we do not have any alignment requirement.
1144 // Thus, it is safe to directly map the vector loads with interesting
1145 // addressing modes.
1146 // FIXME: We could do the same for bitconvert to floating point vectors.
1147 def : Pat <(v8i8 (scalar_to_vector (i32 (extloadi8 am_indexed8:$addr)))),
1148 (INSERT_SUBREG (v8i8 (IMPLICIT_DEF)),
1149 (LDRBui am_indexed8:$addr), bsub)>;
1150 def : Pat <(v16i8 (scalar_to_vector (i32 (extloadi8 am_indexed8:$addr)))),
1151 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
1152 (LDRBui am_indexed8:$addr), bsub)>;
1153 def : Pat <(v4i16 (scalar_to_vector (i32 (extloadi16 am_indexed16:$addr)))),
1154 (INSERT_SUBREG (v4i16 (IMPLICIT_DEF)),
1155 (LDRHui am_indexed16:$addr), hsub)>;
1156 def : Pat <(v8i16 (scalar_to_vector (i32 (extloadi16 am_indexed16:$addr)))),
1157 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
1158 (LDRHui am_indexed16:$addr), hsub)>;
1159 def : Pat <(v2i32 (scalar_to_vector (i32 (load am_indexed32:$addr)))),
1160 (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)),
1161 (LDRSui am_indexed32:$addr), ssub)>;
1162 def : Pat <(v4i32 (scalar_to_vector (i32 (load am_indexed32:$addr)))),
1163 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
1164 (LDRSui am_indexed32:$addr), ssub)>;
1165 def : Pat <(v1i64 (scalar_to_vector (i64 (load am_indexed64:$addr)))),
1166 (LDRDui am_indexed64:$addr)>;
1167 def : Pat <(v2i64 (scalar_to_vector (i64 (load am_indexed64:$addr)))),
1168 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
1169 (LDRDui am_indexed64:$addr), dsub)>;
1171 // Match all load 64 bits width whose type is compatible with FPR64
1172 def : Pat<(v2f32 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1173 def : Pat<(v1f64 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1174 def : Pat<(v8i8 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1175 def : Pat<(v4i16 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1176 def : Pat<(v2i32 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1177 def : Pat<(v1i64 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1179 // Match all load 128 bits width whose type is compatible with FPR128
1180 def : Pat<(v4f32 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1181 def : Pat<(v2f64 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1182 def : Pat<(v16i8 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1183 def : Pat<(v8i16 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1184 def : Pat<(v4i32 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1185 def : Pat<(v2i64 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1186 def : Pat<(f128 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1188 def LDRHHui : LoadUI<0b01, 0, 0b01, GPR32, am_indexed16, "ldrh",
1189 [(set GPR32:$Rt, (zextloadi16 am_indexed16:$addr))]>;
1190 def LDRBBui : LoadUI<0b00, 0, 0b01, GPR32, am_indexed8, "ldrb",
1191 [(set GPR32:$Rt, (zextloadi8 am_indexed8:$addr))]>;
1193 def : Pat<(i64 (zextloadi8 am_indexed8:$addr)),
1194 (SUBREG_TO_REG (i64 0), (LDRBBui am_indexed8:$addr), sub_32)>;
1195 def : Pat<(i64 (zextloadi16 am_indexed16:$addr)),
1196 (SUBREG_TO_REG (i64 0), (LDRHHui am_indexed16:$addr), sub_32)>;
1198 // zextloadi1 -> zextloadi8
1199 def : Pat<(i32 (zextloadi1 am_indexed8:$addr)), (LDRBBui am_indexed8:$addr)>;
1200 def : Pat<(i64 (zextloadi1 am_indexed8:$addr)),
1201 (SUBREG_TO_REG (i64 0), (LDRBBui am_indexed8:$addr), sub_32)>;
1203 // extload -> zextload
1204 def : Pat<(i32 (extloadi16 am_indexed16:$addr)), (LDRHHui am_indexed16:$addr)>;
1205 def : Pat<(i32 (extloadi8 am_indexed8:$addr)), (LDRBBui am_indexed8:$addr)>;
1206 def : Pat<(i32 (extloadi1 am_indexed8:$addr)), (LDRBBui am_indexed8:$addr)>;
1207 def : Pat<(i64 (extloadi32 am_indexed32:$addr)),
1208 (SUBREG_TO_REG (i64 0), (LDRWui am_indexed32:$addr), sub_32)>;
1209 def : Pat<(i64 (extloadi16 am_indexed16:$addr)),
1210 (SUBREG_TO_REG (i64 0), (LDRHHui am_indexed16:$addr), sub_32)>;
1211 def : Pat<(i64 (extloadi8 am_indexed8:$addr)),
1212 (SUBREG_TO_REG (i64 0), (LDRBBui am_indexed8:$addr), sub_32)>;
1213 def : Pat<(i64 (extloadi1 am_indexed8:$addr)),
1214 (SUBREG_TO_REG (i64 0), (LDRBBui am_indexed8:$addr), sub_32)>;
1216 // load sign-extended half-word
1217 def LDRSHWui : LoadUI<0b01, 0, 0b11, GPR32, am_indexed16, "ldrsh",
1218 [(set GPR32:$Rt, (sextloadi16 am_indexed16:$addr))]>;
1219 def LDRSHXui : LoadUI<0b01, 0, 0b10, GPR64, am_indexed16, "ldrsh",
1220 [(set GPR64:$Rt, (sextloadi16 am_indexed16:$addr))]>;
1222 // load sign-extended byte
1223 def LDRSBWui : LoadUI<0b00, 0, 0b11, GPR32, am_indexed8, "ldrsb",
1224 [(set GPR32:$Rt, (sextloadi8 am_indexed8:$addr))]>;
1225 def LDRSBXui : LoadUI<0b00, 0, 0b10, GPR64, am_indexed8, "ldrsb",
1226 [(set GPR64:$Rt, (sextloadi8 am_indexed8:$addr))]>;
1228 // load sign-extended word
1229 def LDRSWui : LoadUI<0b10, 0, 0b10, GPR64, am_indexed32, "ldrsw",
1230 [(set GPR64:$Rt, (sextloadi32 am_indexed32:$addr))]>;
1232 // load zero-extended word
1233 def : Pat<(i64 (zextloadi32 am_indexed32:$addr)),
1234 (SUBREG_TO_REG (i64 0), (LDRWui am_indexed32:$addr), sub_32)>;
1237 def PRFMui : PrefetchUI<0b11, 0, 0b10, "prfm",
1238 [(ARM64Prefetch imm:$Rt, am_indexed64:$addr)]>;
1242 def LDRWl : LoadLiteral<0b00, 0, GPR32, "ldr">;
1243 def LDRXl : LoadLiteral<0b01, 0, GPR64, "ldr">;
1244 def LDRSl : LoadLiteral<0b00, 1, FPR32, "ldr">;
1245 def LDRDl : LoadLiteral<0b01, 1, FPR64, "ldr">;
1246 def LDRQl : LoadLiteral<0b10, 1, FPR128, "ldr">;
1248 // load sign-extended word
1249 def LDRSWl : LoadLiteral<0b10, 0, GPR64, "ldrsw">;
1252 def PRFMl : PrefetchLiteral<0b11, 0, "prfm", []>;
1253 // [(ARM64Prefetch imm:$Rt, tglobaladdr:$label)]>;
1256 // (unscaled immediate)
1257 def LDURXi : LoadUnscaled<0b11, 0, 0b01, GPR64, am_unscaled64, "ldur",
1258 [(set GPR64:$Rt, (load am_unscaled64:$addr))]>;
1259 def LDURWi : LoadUnscaled<0b10, 0, 0b01, GPR32, am_unscaled32, "ldur",
1260 [(set GPR32:$Rt, (load am_unscaled32:$addr))]>;
1261 def LDURBi : LoadUnscaled<0b00, 1, 0b01, FPR8, am_unscaled8, "ldur",
1262 [(set FPR8:$Rt, (load am_unscaled8:$addr))]>;
1263 def LDURHi : LoadUnscaled<0b01, 1, 0b01, FPR16, am_unscaled16, "ldur",
1264 [(set (f16 FPR16:$Rt), (load am_unscaled16:$addr))]>;
1265 def LDURSi : LoadUnscaled<0b10, 1, 0b01, FPR32, am_unscaled32, "ldur",
1266 [(set (f32 FPR32:$Rt), (load am_unscaled32:$addr))]>;
1267 def LDURDi : LoadUnscaled<0b11, 1, 0b01, FPR64, am_unscaled64, "ldur",
1268 [(set (f64 FPR64:$Rt), (load am_unscaled64:$addr))]>;
1269 def LDURQi : LoadUnscaled<0b00, 1, 0b11, FPR128, am_unscaled128, "ldur",
1270 [(set (v2f64 FPR128:$Rt), (load am_unscaled128:$addr))]>;
1273 : LoadUnscaled<0b01, 0, 0b01, GPR32, am_unscaled16, "ldurh",
1274 [(set GPR32:$Rt, (zextloadi16 am_unscaled16:$addr))]>;
1276 : LoadUnscaled<0b00, 0, 0b01, GPR32, am_unscaled8, "ldurb",
1277 [(set GPR32:$Rt, (zextloadi8 am_unscaled8:$addr))]>;
1279 // Match all load 64 bits width whose type is compatible with FPR64
1280 def : Pat<(v2f32 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1281 def : Pat<(v1f64 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1282 def : Pat<(v8i8 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1283 def : Pat<(v4i16 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1284 def : Pat<(v2i32 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1285 def : Pat<(v1i64 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1287 // Match all load 128 bits width whose type is compatible with FPR128
1288 def : Pat<(v4f32 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1289 def : Pat<(v2f64 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1290 def : Pat<(v16i8 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1291 def : Pat<(v8i16 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1292 def : Pat<(v4i32 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1293 def : Pat<(v2i64 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1294 def : Pat<(f128 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1297 def : Pat<(i32 (extloadi16 am_unscaled16:$addr)), (LDURHHi am_unscaled16:$addr)>;
1298 def : Pat<(i32 (extloadi8 am_unscaled8:$addr)), (LDURBBi am_unscaled8:$addr)>;
1299 def : Pat<(i32 (extloadi1 am_unscaled8:$addr)), (LDURBBi am_unscaled8:$addr)>;
1300 def : Pat<(i64 (extloadi32 am_unscaled32:$addr)),
1301 (SUBREG_TO_REG (i64 0), (LDURWi am_unscaled32:$addr), sub_32)>;
1302 def : Pat<(i64 (extloadi16 am_unscaled16:$addr)),
1303 (SUBREG_TO_REG (i64 0), (LDURHHi am_unscaled16:$addr), sub_32)>;
1304 def : Pat<(i64 (extloadi8 am_unscaled8:$addr)),
1305 (SUBREG_TO_REG (i64 0), (LDURBBi am_unscaled8:$addr), sub_32)>;
1306 def : Pat<(i64 (extloadi1 am_unscaled8:$addr)),
1307 (SUBREG_TO_REG (i64 0), (LDURBBi am_unscaled8:$addr), sub_32)>;
1309 def : Pat<(i32 (zextloadi16 am_unscaled16:$addr)),
1310 (LDURHHi am_unscaled16:$addr)>;
1311 def : Pat<(i32 (zextloadi8 am_unscaled8:$addr)),
1312 (LDURBBi am_unscaled8:$addr)>;
1313 def : Pat<(i32 (zextloadi1 am_unscaled8:$addr)),
1314 (LDURBBi am_unscaled8:$addr)>;
1315 def : Pat<(i64 (zextloadi32 am_unscaled32:$addr)),
1316 (SUBREG_TO_REG (i64 0), (LDURWi am_unscaled32:$addr), sub_32)>;
1317 def : Pat<(i64 (zextloadi16 am_unscaled16:$addr)),
1318 (SUBREG_TO_REG (i64 0), (LDURHHi am_unscaled16:$addr), sub_32)>;
1319 def : Pat<(i64 (zextloadi8 am_unscaled8:$addr)),
1320 (SUBREG_TO_REG (i64 0), (LDURBBi am_unscaled8:$addr), sub_32)>;
1321 def : Pat<(i64 (zextloadi1 am_unscaled8:$addr)),
1322 (SUBREG_TO_REG (i64 0), (LDURBBi am_unscaled8:$addr), sub_32)>;
1326 // LDR mnemonics fall back to LDUR for negative or unaligned offsets.
1328 // Define new assembler match classes as we want to only match these when
1329 // the don't otherwise match the scaled addressing mode for LDR/STR. Don't
1330 // associate a DiagnosticType either, as we want the diagnostic for the
1331 // canonical form (the scaled operand) to take precedence.
1332 def MemoryUnscaledFB8Operand : AsmOperandClass {
1333 let Name = "MemoryUnscaledFB8";
1334 let RenderMethod = "addMemoryUnscaledOperands";
1336 def MemoryUnscaledFB16Operand : AsmOperandClass {
1337 let Name = "MemoryUnscaledFB16";
1338 let RenderMethod = "addMemoryUnscaledOperands";
1340 def MemoryUnscaledFB32Operand : AsmOperandClass {
1341 let Name = "MemoryUnscaledFB32";
1342 let RenderMethod = "addMemoryUnscaledOperands";
1344 def MemoryUnscaledFB64Operand : AsmOperandClass {
1345 let Name = "MemoryUnscaledFB64";
1346 let RenderMethod = "addMemoryUnscaledOperands";
1348 def MemoryUnscaledFB128Operand : AsmOperandClass {
1349 let Name = "MemoryUnscaledFB128";
1350 let RenderMethod = "addMemoryUnscaledOperands";
1352 def am_unscaled_fb8 : Operand<i64> {
1353 let ParserMatchClass = MemoryUnscaledFB8Operand;
1354 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
1356 def am_unscaled_fb16 : Operand<i64> {
1357 let ParserMatchClass = MemoryUnscaledFB16Operand;
1358 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
1360 def am_unscaled_fb32 : Operand<i64> {
1361 let ParserMatchClass = MemoryUnscaledFB32Operand;
1362 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
1364 def am_unscaled_fb64 : Operand<i64> {
1365 let ParserMatchClass = MemoryUnscaledFB64Operand;
1366 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
1368 def am_unscaled_fb128 : Operand<i64> {
1369 let ParserMatchClass = MemoryUnscaledFB128Operand;
1370 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
1372 def : InstAlias<"ldr $Rt, $addr", (LDURXi GPR64:$Rt, am_unscaled_fb64:$addr)>;
1373 def : InstAlias<"ldr $Rt, $addr", (LDURWi GPR32:$Rt, am_unscaled_fb32:$addr)>;
1374 def : InstAlias<"ldr $Rt, $addr", (LDURBi FPR8:$Rt, am_unscaled_fb8:$addr)>;
1375 def : InstAlias<"ldr $Rt, $addr", (LDURHi FPR16:$Rt, am_unscaled_fb16:$addr)>;
1376 def : InstAlias<"ldr $Rt, $addr", (LDURSi FPR32:$Rt, am_unscaled_fb32:$addr)>;
1377 def : InstAlias<"ldr $Rt, $addr", (LDURDi FPR64:$Rt, am_unscaled_fb64:$addr)>;
1378 def : InstAlias<"ldr $Rt, $addr", (LDURQi FPR128:$Rt, am_unscaled_fb128:$addr)>;
1381 def : Pat<(i64 (zextloadi8 am_unscaled8:$addr)),
1382 (SUBREG_TO_REG (i64 0), (LDURBBi am_unscaled8:$addr), sub_32)>;
1383 def : Pat<(i64 (zextloadi16 am_unscaled16:$addr)),
1384 (SUBREG_TO_REG (i64 0), (LDURHHi am_unscaled16:$addr), sub_32)>;
1386 // load sign-extended half-word
1388 : LoadUnscaled<0b01, 0, 0b11, GPR32, am_unscaled16, "ldursh",
1389 [(set GPR32:$Rt, (sextloadi16 am_unscaled16:$addr))]>;
1391 : LoadUnscaled<0b01, 0, 0b10, GPR64, am_unscaled16, "ldursh",
1392 [(set GPR64:$Rt, (sextloadi16 am_unscaled16:$addr))]>;
1394 // load sign-extended byte
1396 : LoadUnscaled<0b00, 0, 0b11, GPR32, am_unscaled8, "ldursb",
1397 [(set GPR32:$Rt, (sextloadi8 am_unscaled8:$addr))]>;
1399 : LoadUnscaled<0b00, 0, 0b10, GPR64, am_unscaled8, "ldursb",
1400 [(set GPR64:$Rt, (sextloadi8 am_unscaled8:$addr))]>;
1402 // load sign-extended word
1404 : LoadUnscaled<0b10, 0, 0b10, GPR64, am_unscaled32, "ldursw",
1405 [(set GPR64:$Rt, (sextloadi32 am_unscaled32:$addr))]>;
1407 // zero and sign extending aliases from generic LDR* mnemonics to LDUR*.
1408 def : InstAlias<"ldrb $Rt, $addr", (LDURBBi GPR32:$Rt, am_unscaled_fb8:$addr)>;
1409 def : InstAlias<"ldrh $Rt, $addr", (LDURHHi GPR32:$Rt, am_unscaled_fb16:$addr)>;
1410 def : InstAlias<"ldrsb $Rt, $addr", (LDURSBWi GPR32:$Rt, am_unscaled_fb8:$addr)>;
1411 def : InstAlias<"ldrsb $Rt, $addr", (LDURSBXi GPR64:$Rt, am_unscaled_fb8:$addr)>;
1412 def : InstAlias<"ldrsh $Rt, $addr", (LDURSHWi GPR32:$Rt, am_unscaled_fb16:$addr)>;
1413 def : InstAlias<"ldrsh $Rt, $addr", (LDURSHXi GPR64:$Rt, am_unscaled_fb16:$addr)>;
1414 def : InstAlias<"ldrsw $Rt, $addr", (LDURSWi GPR64:$Rt, am_unscaled_fb32:$addr)>;
1417 def PRFUMi : PrefetchUnscaled<0b11, 0, 0b10, "prfum",
1418 [(ARM64Prefetch imm:$Rt, am_unscaled64:$addr)]>;
1421 // (unscaled immediate, unprivileged)
1422 def LDTRXi : LoadUnprivileged<0b11, 0, 0b01, GPR64, "ldtr">;
1423 def LDTRWi : LoadUnprivileged<0b10, 0, 0b01, GPR32, "ldtr">;
1425 def LDTRHi : LoadUnprivileged<0b01, 0, 0b01, GPR32, "ldtrh">;
1426 def LDTRBi : LoadUnprivileged<0b00, 0, 0b01, GPR32, "ldtrb">;
1428 // load sign-extended half-word
1429 def LDTRSHWi : LoadUnprivileged<0b01, 0, 0b11, GPR32, "ldtrsh">;
1430 def LDTRSHXi : LoadUnprivileged<0b01, 0, 0b10, GPR64, "ldtrsh">;
1432 // load sign-extended byte
1433 def LDTRSBWi : LoadUnprivileged<0b00, 0, 0b11, GPR32, "ldtrsb">;
1434 def LDTRSBXi : LoadUnprivileged<0b00, 0, 0b10, GPR64, "ldtrsb">;
1436 // load sign-extended word
1437 def LDTRSWi : LoadUnprivileged<0b10, 0, 0b10, GPR64, "ldtrsw">;
1440 // (immediate pre-indexed)
1441 def LDRWpre : LoadPreIdx<0b10, 0, 0b01, GPR32, "ldr">;
1442 def LDRXpre : LoadPreIdx<0b11, 0, 0b01, GPR64, "ldr">;
1443 def LDRBpre : LoadPreIdx<0b00, 1, 0b01, FPR8, "ldr">;
1444 def LDRHpre : LoadPreIdx<0b01, 1, 0b01, FPR16, "ldr">;
1445 def LDRSpre : LoadPreIdx<0b10, 1, 0b01, FPR32, "ldr">;
1446 def LDRDpre : LoadPreIdx<0b11, 1, 0b01, FPR64, "ldr">;
1447 def LDRQpre : LoadPreIdx<0b00, 1, 0b11, FPR128, "ldr">;
1449 // load sign-extended half-word
1450 def LDRSHWpre : LoadPreIdx<0b01, 0, 0b11, GPR32, "ldrsh">;
1451 def LDRSHXpre : LoadPreIdx<0b01, 0, 0b10, GPR64, "ldrsh">;
1453 // load sign-extended byte
1454 def LDRSBWpre : LoadPreIdx<0b00, 0, 0b11, GPR32, "ldrsb">;
1455 def LDRSBXpre : LoadPreIdx<0b00, 0, 0b10, GPR64, "ldrsb">;
1457 // load zero-extended byte
1458 def LDRBBpre : LoadPreIdx<0b00, 0, 0b01, GPR32, "ldrb">;
1459 def LDRHHpre : LoadPreIdx<0b01, 0, 0b01, GPR32, "ldrh">;
1461 // load sign-extended word
1462 def LDRSWpre : LoadPreIdx<0b10, 0, 0b10, GPR64, "ldrsw">;
1464 // ISel pseudos and patterns. See expanded comment on LoadPreIdxPseudo.
1465 def LDRDpre_isel : LoadPreIdxPseudo<FPR64>;
1466 def LDRSpre_isel : LoadPreIdxPseudo<FPR32>;
1467 def LDRXpre_isel : LoadPreIdxPseudo<GPR64>;
1468 def LDRWpre_isel : LoadPreIdxPseudo<GPR32>;
1469 def LDRHHpre_isel : LoadPreIdxPseudo<GPR32>;
1470 def LDRBBpre_isel : LoadPreIdxPseudo<GPR32>;
1472 def LDRSWpre_isel : LoadPreIdxPseudo<GPR64>;
1473 def LDRSHWpre_isel : LoadPreIdxPseudo<GPR32>;
1474 def LDRSHXpre_isel : LoadPreIdxPseudo<GPR64>;
1475 def LDRSBWpre_isel : LoadPreIdxPseudo<GPR32>;
1476 def LDRSBXpre_isel : LoadPreIdxPseudo<GPR64>;
1479 // (immediate post-indexed)
1480 def LDRWpost : LoadPostIdx<0b10, 0, 0b01, GPR32, "ldr">;
1481 def LDRXpost : LoadPostIdx<0b11, 0, 0b01, GPR64, "ldr">;
1482 def LDRBpost : LoadPostIdx<0b00, 1, 0b01, FPR8, "ldr">;
1483 def LDRHpost : LoadPostIdx<0b01, 1, 0b01, FPR16, "ldr">;
1484 def LDRSpost : LoadPostIdx<0b10, 1, 0b01, FPR32, "ldr">;
1485 def LDRDpost : LoadPostIdx<0b11, 1, 0b01, FPR64, "ldr">;
1486 def LDRQpost : LoadPostIdx<0b00, 1, 0b11, FPR128, "ldr">;
1488 // load sign-extended half-word
1489 def LDRSHWpost : LoadPostIdx<0b01, 0, 0b11, GPR32, "ldrsh">;
1490 def LDRSHXpost : LoadPostIdx<0b01, 0, 0b10, GPR64, "ldrsh">;
1492 // load sign-extended byte
1493 def LDRSBWpost : LoadPostIdx<0b00, 0, 0b11, GPR32, "ldrsb">;
1494 def LDRSBXpost : LoadPostIdx<0b00, 0, 0b10, GPR64, "ldrsb">;
1496 // load zero-extended byte
1497 def LDRBBpost : LoadPostIdx<0b00, 0, 0b01, GPR32, "ldrb">;
1498 def LDRHHpost : LoadPostIdx<0b01, 0, 0b01, GPR32, "ldrh">;
1500 // load sign-extended word
1501 def LDRSWpost : LoadPostIdx<0b10, 0, 0b10, GPR64, "ldrsw">;
1503 // ISel pseudos and patterns. See expanded comment on LoadPostIdxPseudo.
1504 def LDRDpost_isel : LoadPostIdxPseudo<FPR64>;
1505 def LDRSpost_isel : LoadPostIdxPseudo<FPR32>;
1506 def LDRXpost_isel : LoadPostIdxPseudo<GPR64>;
1507 def LDRWpost_isel : LoadPostIdxPseudo<GPR32>;
1508 def LDRHHpost_isel : LoadPostIdxPseudo<GPR32>;
1509 def LDRBBpost_isel : LoadPostIdxPseudo<GPR32>;
1511 def LDRSWpost_isel : LoadPostIdxPseudo<GPR64>;
1512 def LDRSHWpost_isel : LoadPostIdxPseudo<GPR32>;
1513 def LDRSHXpost_isel : LoadPostIdxPseudo<GPR64>;
1514 def LDRSBWpost_isel : LoadPostIdxPseudo<GPR32>;
1515 def LDRSBXpost_isel : LoadPostIdxPseudo<GPR64>;
1517 //===----------------------------------------------------------------------===//
1518 // Store instructions.
1519 //===----------------------------------------------------------------------===//
1521 // Pair (indexed, offset)
1522 // FIXME: Use dedicated range-checked addressing mode operand here.
1523 def STPWi : StorePairOffset<0b00, 0, GPR32, am_indexed32simm7, "stp">;
1524 def STPXi : StorePairOffset<0b10, 0, GPR64, am_indexed64simm7, "stp">;
1525 def STPSi : StorePairOffset<0b00, 1, FPR32, am_indexed32simm7, "stp">;
1526 def STPDi : StorePairOffset<0b01, 1, FPR64, am_indexed64simm7, "stp">;
1527 def STPQi : StorePairOffset<0b10, 1, FPR128, am_indexed128simm7, "stp">;
1529 // Pair (pre-indexed)
1530 def STPWpre : StorePairPreIdx<0b00, 0, GPR32, am_indexed32simm7_wb, "stp">;
1531 def STPXpre : StorePairPreIdx<0b10, 0, GPR64, am_indexed64simm7_wb, "stp">;
1532 def STPSpre : StorePairPreIdx<0b00, 1, FPR32, am_indexed32simm7_wb, "stp">;
1533 def STPDpre : StorePairPreIdx<0b01, 1, FPR64, am_indexed64simm7_wb, "stp">;
1534 def STPQpre : StorePairPreIdx<0b10, 1, FPR128, am_indexed128simm7_wb, "stp">;
1536 // Pair (pre-indexed)
1537 def STPWpost : StorePairPostIdx<0b00, 0, GPR32, simm7s4, "stp">;
1538 def STPXpost : StorePairPostIdx<0b10, 0, GPR64, simm7s8, "stp">;
1539 def STPSpost : StorePairPostIdx<0b00, 1, FPR32, simm7s4, "stp">;
1540 def STPDpost : StorePairPostIdx<0b01, 1, FPR64, simm7s8, "stp">;
1541 def STPQpost : StorePairPostIdx<0b10, 1, FPR128, simm7s16, "stp">;
1543 // Pair (no allocate)
1544 def STNPWi : StorePairNoAlloc<0b00, 0, GPR32, am_indexed32simm7, "stnp">;
1545 def STNPXi : StorePairNoAlloc<0b10, 0, GPR64, am_indexed64simm7, "stnp">;
1546 def STNPSi : StorePairNoAlloc<0b00, 1, FPR32, am_indexed32simm7, "stnp">;
1547 def STNPDi : StorePairNoAlloc<0b01, 1, FPR64, am_indexed64simm7, "stnp">;
1548 def STNPQi : StorePairNoAlloc<0b10, 1, FPR128, am_indexed128simm7, "stnp">;
1551 // (Register offset)
1553 let AddedComplexity = 10 in {
1556 def STRHHro : Store16RO<0b01, 0, 0b00, GPR32, "strh",
1557 [(truncstorei16 GPR32:$Rt, ro_indexed16:$addr)]>;
1558 def STRBBro : Store8RO<0b00, 0, 0b00, GPR32, "strb",
1559 [(truncstorei8 GPR32:$Rt, ro_indexed8:$addr)]>;
1560 def STRWro : Store32RO<0b10, 0, 0b00, GPR32, "str",
1561 [(store GPR32:$Rt, ro_indexed32:$addr)]>;
1562 def STRXro : Store64RO<0b11, 0, 0b00, GPR64, "str",
1563 [(store GPR64:$Rt, ro_indexed64:$addr)]>;
1566 def : Pat<(truncstorei8 GPR64:$Rt, ro_indexed8:$addr),
1567 (STRBBro (EXTRACT_SUBREG GPR64:$Rt, sub_32), ro_indexed8:$addr)>;
1568 def : Pat<(truncstorei16 GPR64:$Rt, ro_indexed16:$addr),
1569 (STRHHro (EXTRACT_SUBREG GPR64:$Rt, sub_32), ro_indexed16:$addr)>;
1570 def : Pat<(truncstorei32 GPR64:$Rt, ro_indexed32:$addr),
1571 (STRWro (EXTRACT_SUBREG GPR64:$Rt, sub_32), ro_indexed32:$addr)>;
1575 def STRBro : Store8RO<0b00, 1, 0b00, FPR8, "str",
1576 [(store FPR8:$Rt, ro_indexed8:$addr)]>;
1577 def STRHro : Store16RO<0b01, 1, 0b00, FPR16, "str",
1578 [(store (f16 FPR16:$Rt), ro_indexed16:$addr)]>;
1579 def STRSro : Store32RO<0b10, 1, 0b00, FPR32, "str",
1580 [(store (f32 FPR32:$Rt), ro_indexed32:$addr)]>;
1581 def STRDro : Store64RO<0b11, 1, 0b00, FPR64, "str",
1582 [(store (f64 FPR64:$Rt), ro_indexed64:$addr)]>;
1583 def STRQro : Store128RO<0b00, 1, 0b10, FPR128, "str", []> {
1587 // Match all store 64 bits width whose type is compatible with FPR64
1588 def : Pat<(store (v2f32 FPR64:$Rn), ro_indexed64:$addr),
1589 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1590 def : Pat<(store (v1f64 FPR64:$Rn), ro_indexed64:$addr),
1591 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1592 def : Pat<(store (v8i8 FPR64:$Rn), ro_indexed64:$addr),
1593 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1594 def : Pat<(store (v4i16 FPR64:$Rn), ro_indexed64:$addr),
1595 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1596 def : Pat<(store (v2i32 FPR64:$Rn), ro_indexed64:$addr),
1597 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1598 def : Pat<(store (v1i64 FPR64:$Rn), ro_indexed64:$addr),
1599 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1601 // Match all store 128 bits width whose type is compatible with FPR128
1602 def : Pat<(store (v4f32 FPR128:$Rn), ro_indexed128:$addr),
1603 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1604 def : Pat<(store (v2f64 FPR128:$Rn), ro_indexed128:$addr),
1605 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1606 def : Pat<(store (v16i8 FPR128:$Rn), ro_indexed128:$addr),
1607 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1608 def : Pat<(store (v8i16 FPR128:$Rn), ro_indexed128:$addr),
1609 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1610 def : Pat<(store (v4i32 FPR128:$Rn), ro_indexed128:$addr),
1611 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1612 def : Pat<(store (v2i64 FPR128:$Rn), ro_indexed128:$addr),
1613 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1614 def : Pat<(store (f128 FPR128:$Rn), ro_indexed128:$addr),
1615 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1618 // (unsigned immediate)
1619 def STRXui : StoreUI<0b11, 0, 0b00, GPR64, am_indexed64, "str",
1620 [(store GPR64:$Rt, am_indexed64:$addr)]>;
1621 def STRWui : StoreUI<0b10, 0, 0b00, GPR32, am_indexed32, "str",
1622 [(store GPR32:$Rt, am_indexed32:$addr)]>;
1623 def STRBui : StoreUI<0b00, 1, 0b00, FPR8, am_indexed8, "str",
1624 [(store FPR8:$Rt, am_indexed8:$addr)]>;
1625 def STRHui : StoreUI<0b01, 1, 0b00, FPR16, am_indexed16, "str",
1626 [(store (f16 FPR16:$Rt), am_indexed16:$addr)]>;
1627 def STRSui : StoreUI<0b10, 1, 0b00, FPR32, am_indexed32, "str",
1628 [(store (f32 FPR32:$Rt), am_indexed32:$addr)]>;
1629 def STRDui : StoreUI<0b11, 1, 0b00, FPR64, am_indexed64, "str",
1630 [(store (f64 FPR64:$Rt), am_indexed64:$addr)]>;
1631 def STRQui : StoreUI<0b00, 1, 0b10, FPR128, am_indexed128, "str", []> {
1635 // Match all store 64 bits width whose type is compatible with FPR64
1636 def : Pat<(store (v2f32 FPR64:$Rn), am_indexed64:$addr),
1637 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1638 def : Pat<(store (v1f64 FPR64:$Rn), am_indexed64:$addr),
1639 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1640 def : Pat<(store (v8i8 FPR64:$Rn), am_indexed64:$addr),
1641 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1642 def : Pat<(store (v4i16 FPR64:$Rn), am_indexed64:$addr),
1643 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1644 def : Pat<(store (v2i32 FPR64:$Rn), am_indexed64:$addr),
1645 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1646 def : Pat<(store (v1i64 FPR64:$Rn), am_indexed64:$addr),
1647 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1649 // Match all store 128 bits width whose type is compatible with FPR128
1650 def : Pat<(store (v4f32 FPR128:$Rn), am_indexed128:$addr),
1651 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1652 def : Pat<(store (v2f64 FPR128:$Rn), am_indexed128:$addr),
1653 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1654 def : Pat<(store (v16i8 FPR128:$Rn), am_indexed128:$addr),
1655 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1656 def : Pat<(store (v8i16 FPR128:$Rn), am_indexed128:$addr),
1657 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1658 def : Pat<(store (v4i32 FPR128:$Rn), am_indexed128:$addr),
1659 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1660 def : Pat<(store (v2i64 FPR128:$Rn), am_indexed128:$addr),
1661 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1662 def : Pat<(store (f128 FPR128:$Rn), am_indexed128:$addr),
1663 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1665 def STRHHui : StoreUI<0b01, 0, 0b00, GPR32, am_indexed16, "strh",
1666 [(truncstorei16 GPR32:$Rt, am_indexed16:$addr)]>;
1667 def STRBBui : StoreUI<0b00, 0, 0b00, GPR32, am_indexed8, "strb",
1668 [(truncstorei8 GPR32:$Rt, am_indexed8:$addr)]>;
1671 def : Pat<(truncstorei32 GPR64:$Rt, am_indexed32:$addr),
1672 (STRWui (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_indexed32:$addr)>;
1673 def : Pat<(truncstorei16 GPR64:$Rt, am_indexed16:$addr),
1674 (STRHHui (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_indexed16:$addr)>;
1675 def : Pat<(truncstorei8 GPR64:$Rt, am_indexed8:$addr),
1676 (STRBBui (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_indexed8:$addr)>;
1678 } // AddedComplexity = 10
1681 // (unscaled immediate)
1682 def STURXi : StoreUnscaled<0b11, 0, 0b00, GPR64, am_unscaled64, "stur",
1683 [(store GPR64:$Rt, am_unscaled64:$addr)]>;
1684 def STURWi : StoreUnscaled<0b10, 0, 0b00, GPR32, am_unscaled32, "stur",
1685 [(store GPR32:$Rt, am_unscaled32:$addr)]>;
1686 def STURBi : StoreUnscaled<0b00, 1, 0b00, FPR8, am_unscaled8, "stur",
1687 [(store FPR8:$Rt, am_unscaled8:$addr)]>;
1688 def STURHi : StoreUnscaled<0b01, 1, 0b00, FPR16, am_unscaled16, "stur",
1689 [(store (f16 FPR16:$Rt), am_unscaled16:$addr)]>;
1690 def STURSi : StoreUnscaled<0b10, 1, 0b00, FPR32, am_unscaled32, "stur",
1691 [(store (f32 FPR32:$Rt), am_unscaled32:$addr)]>;
1692 def STURDi : StoreUnscaled<0b11, 1, 0b00, FPR64, am_unscaled64, "stur",
1693 [(store (f64 FPR64:$Rt), am_unscaled64:$addr)]>;
1694 def STURQi : StoreUnscaled<0b00, 1, 0b10, FPR128, am_unscaled128, "stur",
1695 [(store (v2f64 FPR128:$Rt), am_unscaled128:$addr)]>;
1696 def STURHHi : StoreUnscaled<0b01, 0, 0b00, GPR32, am_unscaled16, "sturh",
1697 [(truncstorei16 GPR32:$Rt, am_unscaled16:$addr)]>;
1698 def STURBBi : StoreUnscaled<0b00, 0, 0b00, GPR32, am_unscaled8, "sturb",
1699 [(truncstorei8 GPR32:$Rt, am_unscaled8:$addr)]>;
1701 // Match all store 64 bits width whose type is compatible with FPR64
1702 def : Pat<(store (v2f32 FPR64:$Rn), am_unscaled64:$addr),
1703 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1704 def : Pat<(store (v1f64 FPR64:$Rn), am_unscaled64:$addr),
1705 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1706 def : Pat<(store (v8i8 FPR64:$Rn), am_unscaled64:$addr),
1707 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1708 def : Pat<(store (v4i16 FPR64:$Rn), am_unscaled64:$addr),
1709 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1710 def : Pat<(store (v2i32 FPR64:$Rn), am_unscaled64:$addr),
1711 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1712 def : Pat<(store (v1i64 FPR64:$Rn), am_unscaled64:$addr),
1713 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1715 // Match all store 128 bits width whose type is compatible with FPR128
1716 def : Pat<(store (v4f32 FPR128:$Rn), am_unscaled128:$addr),
1717 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1718 def : Pat<(store (v2f64 FPR128:$Rn), am_unscaled128:$addr),
1719 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1720 def : Pat<(store (v16i8 FPR128:$Rn), am_unscaled128:$addr),
1721 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1722 def : Pat<(store (v8i16 FPR128:$Rn), am_unscaled128:$addr),
1723 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1724 def : Pat<(store (v4i32 FPR128:$Rn), am_unscaled128:$addr),
1725 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1726 def : Pat<(store (v2i64 FPR128:$Rn), am_unscaled128:$addr),
1727 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1728 def : Pat<(store (f128 FPR128:$Rn), am_unscaled128:$addr),
1729 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1731 // unscaled i64 truncating stores
1732 def : Pat<(truncstorei32 GPR64:$Rt, am_unscaled32:$addr),
1733 (STURWi (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_unscaled32:$addr)>;
1734 def : Pat<(truncstorei16 GPR64:$Rt, am_unscaled16:$addr),
1735 (STURHHi (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_unscaled16:$addr)>;
1736 def : Pat<(truncstorei8 GPR64:$Rt, am_unscaled8:$addr),
1737 (STURBBi (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_unscaled8:$addr)>;
1740 // STR mnemonics fall back to STUR for negative or unaligned offsets.
1741 def : InstAlias<"str $Rt, $addr", (STURXi GPR64:$Rt, am_unscaled_fb64:$addr)>;
1742 def : InstAlias<"str $Rt, $addr", (STURWi GPR32:$Rt, am_unscaled_fb32:$addr)>;
1743 def : InstAlias<"str $Rt, $addr", (STURBi FPR8:$Rt, am_unscaled_fb8:$addr)>;
1744 def : InstAlias<"str $Rt, $addr", (STURHi FPR16:$Rt, am_unscaled_fb16:$addr)>;
1745 def : InstAlias<"str $Rt, $addr", (STURSi FPR32:$Rt, am_unscaled_fb32:$addr)>;
1746 def : InstAlias<"str $Rt, $addr", (STURDi FPR64:$Rt, am_unscaled_fb64:$addr)>;
1747 def : InstAlias<"str $Rt, $addr", (STURQi FPR128:$Rt, am_unscaled_fb128:$addr)>;
1749 def : InstAlias<"strb $Rt, $addr", (STURBBi GPR32:$Rt, am_unscaled_fb8:$addr)>;
1750 def : InstAlias<"strh $Rt, $addr", (STURHHi GPR32:$Rt, am_unscaled_fb16:$addr)>;
1753 // (unscaled immediate, unprivileged)
1754 def STTRWi : StoreUnprivileged<0b10, 0, 0b00, GPR32, "sttr">;
1755 def STTRXi : StoreUnprivileged<0b11, 0, 0b00, GPR64, "sttr">;
1757 def STTRHi : StoreUnprivileged<0b01, 0, 0b00, GPR32, "sttrh">;
1758 def STTRBi : StoreUnprivileged<0b00, 0, 0b00, GPR32, "sttrb">;
1761 // (immediate pre-indexed)
1762 def STRWpre : StorePreIdx<0b10, 0, 0b00, GPR32, "str">;
1763 def STRXpre : StorePreIdx<0b11, 0, 0b00, GPR64, "str">;
1764 def STRBpre : StorePreIdx<0b00, 1, 0b00, FPR8, "str">;
1765 def STRHpre : StorePreIdx<0b01, 1, 0b00, FPR16, "str">;
1766 def STRSpre : StorePreIdx<0b10, 1, 0b00, FPR32, "str">;
1767 def STRDpre : StorePreIdx<0b11, 1, 0b00, FPR64, "str">;
1768 def STRQpre : StorePreIdx<0b00, 1, 0b10, FPR128, "str">;
1770 def STRBBpre : StorePreIdx<0b00, 0, 0b00, GPR32, "strb">;
1771 def STRHHpre : StorePreIdx<0b01, 0, 0b00, GPR32, "strh">;
1773 // ISel pseudos and patterns. See expanded comment on StorePreIdxPseudo.
1774 defm STRDpre : StorePreIdxPseudo<FPR64, f64, pre_store>;
1775 defm STRSpre : StorePreIdxPseudo<FPR32, f32, pre_store>;
1776 defm STRXpre : StorePreIdxPseudo<GPR64, i64, pre_store>;
1777 defm STRWpre : StorePreIdxPseudo<GPR32, i32, pre_store>;
1778 defm STRHHpre : StorePreIdxPseudo<GPR32, i32, pre_truncsti16>;
1779 defm STRBBpre : StorePreIdxPseudo<GPR32, i32, pre_truncsti8>;
1781 def : Pat<(pre_truncsti32 GPR64:$Rt, am_noindex:$addr, simm9:$off),
1782 (STRWpre_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
1784 def : Pat<(pre_truncsti16 GPR64:$Rt, am_noindex:$addr, simm9:$off),
1785 (STRHHpre_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
1787 def : Pat<(pre_truncsti8 GPR64:$Rt, am_noindex:$addr, simm9:$off),
1788 (STRBBpre_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
1792 // (immediate post-indexed)
1793 def STRWpost : StorePostIdx<0b10, 0, 0b00, GPR32, "str">;
1794 def STRXpost : StorePostIdx<0b11, 0, 0b00, GPR64, "str">;
1795 def STRBpost : StorePostIdx<0b00, 1, 0b00, FPR8, "str">;
1796 def STRHpost : StorePostIdx<0b01, 1, 0b00, FPR16, "str">;
1797 def STRSpost : StorePostIdx<0b10, 1, 0b00, FPR32, "str">;
1798 def STRDpost : StorePostIdx<0b11, 1, 0b00, FPR64, "str">;
1799 def STRQpost : StorePostIdx<0b00, 1, 0b10, FPR128, "str">;
1801 def STRBBpost : StorePostIdx<0b00, 0, 0b00, GPR32, "strb">;
1802 def STRHHpost : StorePostIdx<0b01, 0, 0b00, GPR32, "strh">;
1804 // ISel pseudos and patterns. See expanded comment on StorePostIdxPseudo.
1805 defm STRDpost : StorePostIdxPseudo<FPR64, f64, post_store, STRDpost>;
1806 defm STRSpost : StorePostIdxPseudo<FPR32, f32, post_store, STRSpost>;
1807 defm STRXpost : StorePostIdxPseudo<GPR64, i64, post_store, STRXpost>;
1808 defm STRWpost : StorePostIdxPseudo<GPR32, i32, post_store, STRWpost>;
1809 defm STRHHpost : StorePostIdxPseudo<GPR32, i32, post_truncsti16, STRHHpost>;
1810 defm STRBBpost : StorePostIdxPseudo<GPR32, i32, post_truncsti8, STRBBpost>;
1812 def : Pat<(post_truncsti32 GPR64:$Rt, am_noindex:$addr, simm9:$off),
1813 (STRWpost_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
1815 def : Pat<(post_truncsti16 GPR64:$Rt, am_noindex:$addr, simm9:$off),
1816 (STRHHpost_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
1818 def : Pat<(post_truncsti8 GPR64:$Rt, am_noindex:$addr, simm9:$off),
1819 (STRBBpost_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
1823 //===----------------------------------------------------------------------===//
1824 // Load/store exclusive instructions.
1825 //===----------------------------------------------------------------------===//
1827 def LDARW : LoadAcquire <0b10, 1, 1, 0, 1, GPR32, "ldar">;
1828 def LDARX : LoadAcquire <0b11, 1, 1, 0, 1, GPR64, "ldar">;
1829 def LDARB : LoadAcquire <0b00, 1, 1, 0, 1, GPR32, "ldarb">;
1830 def LDARH : LoadAcquire <0b01, 1, 1, 0, 1, GPR32, "ldarh">;
1832 def LDAXRW : LoadExclusive <0b10, 0, 1, 0, 1, GPR32, "ldaxr">;
1833 def LDAXRX : LoadExclusive <0b11, 0, 1, 0, 1, GPR64, "ldaxr">;
1834 def LDAXRB : LoadExclusive <0b00, 0, 1, 0, 1, GPR32, "ldaxrb">;
1835 def LDAXRH : LoadExclusive <0b01, 0, 1, 0, 1, GPR32, "ldaxrh">;
1837 def LDXRW : LoadExclusive <0b10, 0, 1, 0, 0, GPR32, "ldxr">;
1838 def LDXRX : LoadExclusive <0b11, 0, 1, 0, 0, GPR64, "ldxr">;
1839 def LDXRB : LoadExclusive <0b00, 0, 1, 0, 0, GPR32, "ldxrb">;
1840 def LDXRH : LoadExclusive <0b01, 0, 1, 0, 0, GPR32, "ldxrh">;
1842 def STLRW : StoreRelease <0b10, 1, 0, 0, 1, GPR32, "stlr">;
1843 def STLRX : StoreRelease <0b11, 1, 0, 0, 1, GPR64, "stlr">;
1844 def STLRB : StoreRelease <0b00, 1, 0, 0, 1, GPR32, "stlrb">;
1845 def STLRH : StoreRelease <0b01, 1, 0, 0, 1, GPR32, "stlrh">;
1847 def STLXRW : StoreExclusive<0b10, 0, 0, 0, 1, GPR32, "stlxr">;
1848 def STLXRX : StoreExclusive<0b11, 0, 0, 0, 1, GPR64, "stlxr">;
1849 def STLXRB : StoreExclusive<0b00, 0, 0, 0, 1, GPR32, "stlxrb">;
1850 def STLXRH : StoreExclusive<0b01, 0, 0, 0, 1, GPR32, "stlxrh">;
1852 def STXRW : StoreExclusive<0b10, 0, 0, 0, 0, GPR32, "stxr">;
1853 def STXRX : StoreExclusive<0b11, 0, 0, 0, 0, GPR64, "stxr">;
1854 def STXRB : StoreExclusive<0b00, 0, 0, 0, 0, GPR32, "stxrb">;
1855 def STXRH : StoreExclusive<0b01, 0, 0, 0, 0, GPR32, "stxrh">;
1857 def LDAXPW : LoadExclusivePair<0b10, 0, 1, 1, 1, GPR32, "ldaxp">;
1858 def LDAXPX : LoadExclusivePair<0b11, 0, 1, 1, 1, GPR64, "ldaxp">;
1860 def LDXPW : LoadExclusivePair<0b10, 0, 1, 1, 0, GPR32, "ldxp">;
1861 def LDXPX : LoadExclusivePair<0b11, 0, 1, 1, 0, GPR64, "ldxp">;
1863 def STLXPW : StoreExclusivePair<0b10, 0, 0, 1, 1, GPR32, "stlxp">;
1864 def STLXPX : StoreExclusivePair<0b11, 0, 0, 1, 1, GPR64, "stlxp">;
1866 def STXPW : StoreExclusivePair<0b10, 0, 0, 1, 0, GPR32, "stxp">;
1867 def STXPX : StoreExclusivePair<0b11, 0, 0, 1, 0, GPR64, "stxp">;
1869 //===----------------------------------------------------------------------===//
1870 // Scaled floating point to integer conversion instructions.
1871 //===----------------------------------------------------------------------===//
1873 defm FCVTAS : FPToIntegerUnscaled<0b00, 0b100, "fcvtas", int_arm64_neon_fcvtas>;
1874 defm FCVTAU : FPToIntegerUnscaled<0b00, 0b101, "fcvtau", int_arm64_neon_fcvtau>;
1875 defm FCVTMS : FPToIntegerUnscaled<0b10, 0b000, "fcvtms", int_arm64_neon_fcvtms>;
1876 defm FCVTMU : FPToIntegerUnscaled<0b10, 0b001, "fcvtmu", int_arm64_neon_fcvtmu>;
1877 defm FCVTNS : FPToIntegerUnscaled<0b00, 0b000, "fcvtns", int_arm64_neon_fcvtns>;
1878 defm FCVTNU : FPToIntegerUnscaled<0b00, 0b001, "fcvtnu", int_arm64_neon_fcvtnu>;
1879 defm FCVTPS : FPToIntegerUnscaled<0b01, 0b000, "fcvtps", int_arm64_neon_fcvtps>;
1880 defm FCVTPU : FPToIntegerUnscaled<0b01, 0b001, "fcvtpu", int_arm64_neon_fcvtpu>;
1881 defm FCVTZS : FPToIntegerUnscaled<0b11, 0b000, "fcvtzs", fp_to_sint>;
1882 defm FCVTZU : FPToIntegerUnscaled<0b11, 0b001, "fcvtzu", fp_to_uint>;
1883 defm FCVTZS : FPToIntegerScaled<0b11, 0b000, "fcvtzs", fp_to_sint>;
1884 defm FCVTZU : FPToIntegerScaled<0b11, 0b001, "fcvtzu", fp_to_uint>;
1885 let isCodeGenOnly = 1 in {
1886 defm FCVTZS_Int : FPToIntegerUnscaled<0b11, 0b000, "fcvtzs", int_arm64_neon_fcvtzs>;
1887 defm FCVTZU_Int : FPToIntegerUnscaled<0b11, 0b001, "fcvtzu", int_arm64_neon_fcvtzu>;
1888 defm FCVTZS_Int : FPToIntegerScaled<0b11, 0b000, "fcvtzs", int_arm64_neon_fcvtzs>;
1889 defm FCVTZU_Int : FPToIntegerScaled<0b11, 0b001, "fcvtzu", int_arm64_neon_fcvtzu>;
1892 //===----------------------------------------------------------------------===//
1893 // Scaled integer to floating point conversion instructions.
1894 //===----------------------------------------------------------------------===//
1896 defm SCVTF : IntegerToFP<0, "scvtf", sint_to_fp>;
1897 defm UCVTF : IntegerToFP<1, "ucvtf", uint_to_fp>;
1899 //===----------------------------------------------------------------------===//
1900 // Unscaled integer to floating point conversion instruction.
1901 //===----------------------------------------------------------------------===//
1903 defm FMOV : UnscaledConversion<"fmov">;
1905 def : Pat<(f32 (fpimm0)), (FMOVWSr WZR)>, Requires<[NoZCZ]>;
1906 def : Pat<(f64 (fpimm0)), (FMOVXDr XZR)>, Requires<[NoZCZ]>;
1908 def : Pat<(v8i8 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1909 def : Pat<(v4i16 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1910 def : Pat<(v2i32 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1911 def : Pat<(v1i64 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1912 def : Pat<(v2f32 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1913 def : Pat<(v1f64 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1914 def : Pat<(v1i64 (scalar_to_vector GPR64:$Xn)),
1915 (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1916 def : Pat<(v1f64 (scalar_to_vector GPR64:$Xn)),
1917 (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1918 def : Pat<(v1f64 (scalar_to_vector (f64 FPR64:$Xn))), (v1f64 FPR64:$Xn)>;
1920 def : Pat<(i64 (bitconvert (v8i8 V64:$Vn))),
1921 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
1922 def : Pat<(i64 (bitconvert (v4i16 V64:$Vn))),
1923 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
1924 def : Pat<(i64 (bitconvert (v2i32 V64:$Vn))),
1925 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
1926 def : Pat<(i64 (bitconvert (v1i64 V64:$Vn))),
1927 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
1928 def : Pat<(i64 (bitconvert (v2f32 V64:$Vn))),
1929 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
1930 def : Pat<(i64 (bitconvert (v1f64 V64:$Vn))),
1931 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
1933 def : Pat<(f32 (bitconvert (i32 GPR32:$Xn))),
1934 (COPY_TO_REGCLASS GPR32:$Xn, FPR32)>;
1935 def : Pat<(i32 (bitconvert (f32 FPR32:$Xn))),
1936 (COPY_TO_REGCLASS FPR32:$Xn, GPR32)>;
1937 def : Pat<(f64 (bitconvert (i64 GPR64:$Xn))),
1938 (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1939 def : Pat<(i64 (bitconvert (f64 FPR64:$Xn))),
1940 (COPY_TO_REGCLASS FPR64:$Xn, GPR64)>;
1942 //===----------------------------------------------------------------------===//
1943 // Floating point conversion instruction.
1944 //===----------------------------------------------------------------------===//
1946 defm FCVT : FPConversion<"fcvt">;
1948 def : Pat<(f32_to_f16 FPR32:$Rn),
1949 (i32 (COPY_TO_REGCLASS
1950 (f32 (SUBREG_TO_REG (i32 0), (FCVTHSr FPR32:$Rn), hsub)),
1953 def FCVTSHpseudo : Pseudo<(outs FPR32:$Rd), (ins FPR32:$Rn),
1954 [(set (f32 FPR32:$Rd), (f16_to_f32 i32:$Rn))]>;
1956 //===----------------------------------------------------------------------===//
1957 // Floating point single operand instructions.
1958 //===----------------------------------------------------------------------===//
1960 defm FABS : SingleOperandFPData<0b0001, "fabs", fabs>;
1961 defm FMOV : SingleOperandFPData<0b0000, "fmov">;
1962 defm FNEG : SingleOperandFPData<0b0010, "fneg", fneg>;
1963 defm FRINTA : SingleOperandFPData<0b1100, "frinta", frnd>;
1964 defm FRINTI : SingleOperandFPData<0b1111, "frinti", fnearbyint>;
1965 defm FRINTM : SingleOperandFPData<0b1010, "frintm", ffloor>;
1966 defm FRINTN : SingleOperandFPData<0b1000, "frintn", int_arm64_neon_frintn>;
1967 defm FRINTP : SingleOperandFPData<0b1001, "frintp", fceil>;
1969 def : Pat<(v1f64 (int_arm64_neon_frintn (v1f64 FPR64:$Rn))),
1970 (FRINTNDr FPR64:$Rn)>;
1972 // FRINTX is inserted to set the flags as required by FENV_ACCESS ON behavior
1973 // in the C spec. Setting hasSideEffects ensures it is not DCE'd.
1974 // <rdar://problem/13715968>
1975 // TODO: We should really model the FPSR flags correctly. This is really ugly.
1976 let hasSideEffects = 1 in {
1977 defm FRINTX : SingleOperandFPData<0b1110, "frintx", frint>;
1980 defm FRINTZ : SingleOperandFPData<0b1011, "frintz", ftrunc>;
1982 let SchedRW = [WriteFDiv] in {
1983 defm FSQRT : SingleOperandFPData<0b0011, "fsqrt", fsqrt>;
1986 //===----------------------------------------------------------------------===//
1987 // Floating point two operand instructions.
1988 //===----------------------------------------------------------------------===//
1990 defm FADD : TwoOperandFPData<0b0010, "fadd", fadd>;
1991 let SchedRW = [WriteFDiv] in {
1992 defm FDIV : TwoOperandFPData<0b0001, "fdiv", fdiv>;
1994 defm FMAXNM : TwoOperandFPData<0b0110, "fmaxnm", int_arm64_neon_fmaxnm>;
1995 defm FMAX : TwoOperandFPData<0b0100, "fmax", ARM64fmax>;
1996 defm FMINNM : TwoOperandFPData<0b0111, "fminnm", int_arm64_neon_fminnm>;
1997 defm FMIN : TwoOperandFPData<0b0101, "fmin", ARM64fmin>;
1998 let SchedRW = [WriteFMul] in {
1999 defm FMUL : TwoOperandFPData<0b0000, "fmul", fmul>;
2000 defm FNMUL : TwoOperandFPDataNeg<0b1000, "fnmul", fmul>;
2002 defm FSUB : TwoOperandFPData<0b0011, "fsub", fsub>;
2004 def : Pat<(v1f64 (ARM64fmax (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2005 (FMAXDrr FPR64:$Rn, FPR64:$Rm)>;
2006 def : Pat<(v1f64 (ARM64fmin (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2007 (FMINDrr FPR64:$Rn, FPR64:$Rm)>;
2008 def : Pat<(v1f64 (int_arm64_neon_fmaxnm (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2009 (FMAXNMDrr FPR64:$Rn, FPR64:$Rm)>;
2010 def : Pat<(v1f64 (int_arm64_neon_fminnm (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2011 (FMINNMDrr FPR64:$Rn, FPR64:$Rm)>;
2013 //===----------------------------------------------------------------------===//
2014 // Floating point three operand instructions.
2015 //===----------------------------------------------------------------------===//
2017 defm FMADD : ThreeOperandFPData<0, 0, "fmadd", fma>;
2018 defm FMSUB : ThreeOperandFPData<0, 1, "fmsub",
2019 TriOpFrag<(fma node:$LHS, (fneg node:$MHS), node:$RHS)> >;
2020 defm FNMADD : ThreeOperandFPData<1, 0, "fnmadd",
2021 TriOpFrag<(fneg (fma node:$LHS, node:$MHS, node:$RHS))> >;
2022 defm FNMSUB : ThreeOperandFPData<1, 1, "fnmsub",
2023 TriOpFrag<(fma node:$LHS, node:$MHS, (fneg node:$RHS))> >;
2025 // The following def pats catch the case where the LHS of an FMA is negated.
2026 // The TriOpFrag above catches the case where the middle operand is negated.
2028 // N.b. FMSUB etc have the accumulator at the *end* of (outs), unlike
2029 // the NEON variant.
2030 def : Pat<(f32 (fma (fneg FPR32:$Rn), FPR32:$Rm, FPR32:$Ra)),
2031 (FMSUBSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
2033 def : Pat<(f64 (fma (fneg FPR64:$Rn), FPR64:$Rm, FPR64:$Ra)),
2034 (FMSUBDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
2036 //===----------------------------------------------------------------------===//
2037 // Floating point comparison instructions.
2038 //===----------------------------------------------------------------------===//
2040 defm FCMPE : FPComparison<1, "fcmpe">;
2041 defm FCMP : FPComparison<0, "fcmp", ARM64fcmp>;
2043 //===----------------------------------------------------------------------===//
2044 // Floating point conditional comparison instructions.
2045 //===----------------------------------------------------------------------===//
2047 defm FCCMPE : FPCondComparison<1, "fccmpe">;
2048 defm FCCMP : FPCondComparison<0, "fccmp">;
2050 //===----------------------------------------------------------------------===//
2051 // Floating point conditional select instruction.
2052 //===----------------------------------------------------------------------===//
2054 defm FCSEL : FPCondSelect<"fcsel">;
2056 // CSEL instructions providing f128 types need to be handled by a
2057 // pseudo-instruction since the eventual code will need to introduce basic
2058 // blocks and control flow.
2059 def F128CSEL : Pseudo<(outs FPR128:$Rd),
2060 (ins FPR128:$Rn, FPR128:$Rm, ccode:$cond),
2061 [(set (f128 FPR128:$Rd),
2062 (ARM64csel FPR128:$Rn, FPR128:$Rm,
2063 (i32 imm:$cond), CPSR))]> {
2065 let usesCustomInserter = 1;
2069 //===----------------------------------------------------------------------===//
2070 // Floating point immediate move.
2071 //===----------------------------------------------------------------------===//
2073 let isReMaterializable = 1 in {
2074 defm FMOV : FPMoveImmediate<"fmov">;
2077 //===----------------------------------------------------------------------===//
2078 // Advanced SIMD two vector instructions.
2079 //===----------------------------------------------------------------------===//
2081 defm ABS : SIMDTwoVectorBHSD<0, 0b01011, "abs", int_arm64_neon_abs>;
2082 defm CLS : SIMDTwoVectorBHS<0, 0b00100, "cls", int_arm64_neon_cls>;
2083 defm CLZ : SIMDTwoVectorBHS<1, 0b00100, "clz", ctlz>;
2084 defm CMEQ : SIMDCmpTwoVector<0, 0b01001, "cmeq", ARM64cmeqz>;
2085 defm CMGE : SIMDCmpTwoVector<1, 0b01000, "cmge", ARM64cmgez>;
2086 defm CMGT : SIMDCmpTwoVector<0, 0b01000, "cmgt", ARM64cmgtz>;
2087 defm CMLE : SIMDCmpTwoVector<1, 0b01001, "cmle", ARM64cmlez>;
2088 defm CMLT : SIMDCmpTwoVector<0, 0b01010, "cmlt", ARM64cmltz>;
2089 defm CNT : SIMDTwoVectorB<0, 0b00, 0b00101, "cnt", ctpop>;
2090 defm FABS : SIMDTwoVectorFP<0, 1, 0b01111, "fabs", fabs>;
2092 defm FCMEQ : SIMDFPCmpTwoVector<0, 1, 0b01101, "fcmeq", ARM64fcmeqz>;
2093 defm FCMGE : SIMDFPCmpTwoVector<1, 1, 0b01100, "fcmge", ARM64fcmgez>;
2094 defm FCMGT : SIMDFPCmpTwoVector<0, 1, 0b01100, "fcmgt", ARM64fcmgtz>;
2095 defm FCMLE : SIMDFPCmpTwoVector<1, 1, 0b01101, "fcmle", ARM64fcmlez>;
2096 defm FCMLT : SIMDFPCmpTwoVector<0, 1, 0b01110, "fcmlt", ARM64fcmltz>;
2097 defm FCVTAS : SIMDTwoVectorFPToInt<0,0,0b11100, "fcvtas",int_arm64_neon_fcvtas>;
2098 defm FCVTAU : SIMDTwoVectorFPToInt<1,0,0b11100, "fcvtau",int_arm64_neon_fcvtau>;
2099 defm FCVTL : SIMDFPWidenTwoVector<0, 0, 0b10111, "fcvtl">;
2100 def : Pat<(v4f32 (int_arm64_neon_vcvthf2fp (v4i16 V64:$Rn))),
2101 (FCVTLv4i16 V64:$Rn)>;
2102 def : Pat<(v4f32 (int_arm64_neon_vcvthf2fp (extract_subvector (v8i16 V128:$Rn),
2104 (FCVTLv8i16 V128:$Rn)>;
2105 def : Pat<(v2f64 (fextend (v2f32 V64:$Rn))), (FCVTLv2i32 V64:$Rn)>;
2106 def : Pat<(v2f64 (fextend (v2f32 (extract_subvector (v4f32 V128:$Rn),
2108 (FCVTLv4i32 V128:$Rn)>;
2110 defm FCVTMS : SIMDTwoVectorFPToInt<0,0,0b11011, "fcvtms",int_arm64_neon_fcvtms>;
2111 defm FCVTMU : SIMDTwoVectorFPToInt<1,0,0b11011, "fcvtmu",int_arm64_neon_fcvtmu>;
2112 defm FCVTNS : SIMDTwoVectorFPToInt<0,0,0b11010, "fcvtns",int_arm64_neon_fcvtns>;
2113 defm FCVTNU : SIMDTwoVectorFPToInt<1,0,0b11010, "fcvtnu",int_arm64_neon_fcvtnu>;
2114 defm FCVTN : SIMDFPNarrowTwoVector<0, 0, 0b10110, "fcvtn">;
2115 def : Pat<(v4i16 (int_arm64_neon_vcvtfp2hf (v4f32 V128:$Rn))),
2116 (FCVTNv4i16 V128:$Rn)>;
2117 def : Pat<(concat_vectors V64:$Rd,
2118 (v4i16 (int_arm64_neon_vcvtfp2hf (v4f32 V128:$Rn)))),
2119 (FCVTNv8i16 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
2120 def : Pat<(v2f32 (fround (v2f64 V128:$Rn))), (FCVTNv2i32 V128:$Rn)>;
2121 def : Pat<(concat_vectors V64:$Rd, (v2f32 (fround (v2f64 V128:$Rn)))),
2122 (FCVTNv4i32 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
2123 defm FCVTPS : SIMDTwoVectorFPToInt<0,1,0b11010, "fcvtps",int_arm64_neon_fcvtps>;
2124 defm FCVTPU : SIMDTwoVectorFPToInt<1,1,0b11010, "fcvtpu",int_arm64_neon_fcvtpu>;
2125 defm FCVTXN : SIMDFPInexactCvtTwoVector<1, 0, 0b10110, "fcvtxn",
2126 int_arm64_neon_fcvtxn>;
2127 defm FCVTZS : SIMDTwoVectorFPToInt<0, 1, 0b11011, "fcvtzs", fp_to_sint>;
2128 defm FCVTZU : SIMDTwoVectorFPToInt<1, 1, 0b11011, "fcvtzu", fp_to_uint>;
2129 let isCodeGenOnly = 1 in {
2130 defm FCVTZS_Int : SIMDTwoVectorFPToInt<0, 1, 0b11011, "fcvtzs",
2131 int_arm64_neon_fcvtzs>;
2132 defm FCVTZU_Int : SIMDTwoVectorFPToInt<1, 1, 0b11011, "fcvtzu",
2133 int_arm64_neon_fcvtzu>;
2135 defm FNEG : SIMDTwoVectorFP<1, 1, 0b01111, "fneg", fneg>;
2136 defm FRECPE : SIMDTwoVectorFP<0, 1, 0b11101, "frecpe", int_arm64_neon_frecpe>;
2137 defm FRINTA : SIMDTwoVectorFP<1, 0, 0b11000, "frinta", frnd>;
2138 defm FRINTI : SIMDTwoVectorFP<1, 1, 0b11001, "frinti", fnearbyint>;
2139 defm FRINTM : SIMDTwoVectorFP<0, 0, 0b11001, "frintm", ffloor>;
2140 defm FRINTN : SIMDTwoVectorFP<0, 0, 0b11000, "frintn", int_arm64_neon_frintn>;
2141 defm FRINTP : SIMDTwoVectorFP<0, 1, 0b11000, "frintp", fceil>;
2142 defm FRINTX : SIMDTwoVectorFP<1, 0, 0b11001, "frintx", frint>;
2143 defm FRINTZ : SIMDTwoVectorFP<0, 1, 0b11001, "frintz", ftrunc>;
2144 defm FRSQRTE: SIMDTwoVectorFP<1, 1, 0b11101, "frsqrte", int_arm64_neon_frsqrte>;
2145 defm FSQRT : SIMDTwoVectorFP<1, 1, 0b11111, "fsqrt", fsqrt>;
2146 defm NEG : SIMDTwoVectorBHSD<1, 0b01011, "neg",
2147 UnOpFrag<(sub immAllZerosV, node:$LHS)> >;
2148 defm NOT : SIMDTwoVectorB<1, 0b00, 0b00101, "not", vnot>;
2149 // Aliases for MVN -> NOT.
2150 def : InstAlias<"mvn.8b $Vd, $Vn", (NOTv8i8 V64:$Vd, V64:$Vn)>;
2151 def : InstAlias<"mvn.16b $Vd, $Vn", (NOTv16i8 V128:$Vd, V128:$Vn)>;
2152 def : InstAlias<"mvn $Vd.8b, $Vn.8b", (NOTv8i8 V64:$Vd, V64:$Vn)>;
2153 def : InstAlias<"mvn $Vd.16b, $Vn.16b", (NOTv16i8 V128:$Vd, V128:$Vn)>;
2155 def : Pat<(ARM64neg (v8i8 V64:$Rn)), (NEGv8i8 V64:$Rn)>;
2156 def : Pat<(ARM64neg (v16i8 V128:$Rn)), (NEGv16i8 V128:$Rn)>;
2157 def : Pat<(ARM64neg (v4i16 V64:$Rn)), (NEGv4i16 V64:$Rn)>;
2158 def : Pat<(ARM64neg (v8i16 V128:$Rn)), (NEGv8i16 V128:$Rn)>;
2159 def : Pat<(ARM64neg (v2i32 V64:$Rn)), (NEGv2i32 V64:$Rn)>;
2160 def : Pat<(ARM64neg (v4i32 V128:$Rn)), (NEGv4i32 V128:$Rn)>;
2161 def : Pat<(ARM64neg (v2i64 V128:$Rn)), (NEGv2i64 V128:$Rn)>;
2163 def : Pat<(ARM64not (v8i8 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2164 def : Pat<(ARM64not (v16i8 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2165 def : Pat<(ARM64not (v4i16 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2166 def : Pat<(ARM64not (v8i16 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2167 def : Pat<(ARM64not (v2i32 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2168 def : Pat<(ARM64not (v1i64 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2169 def : Pat<(ARM64not (v4i32 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2170 def : Pat<(ARM64not (v2i64 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2172 def : Pat<(vnot (v4i16 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2173 def : Pat<(vnot (v8i16 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2174 def : Pat<(vnot (v2i32 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2175 def : Pat<(vnot (v4i32 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2176 def : Pat<(vnot (v2i64 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2178 defm RBIT : SIMDTwoVectorB<1, 0b01, 0b00101, "rbit", int_arm64_neon_rbit>;
2179 defm REV16 : SIMDTwoVectorB<0, 0b00, 0b00001, "rev16", ARM64rev16>;
2180 defm REV32 : SIMDTwoVectorBH<1, 0b00000, "rev32", ARM64rev32>;
2181 defm REV64 : SIMDTwoVectorBHS<0, 0b00000, "rev64", ARM64rev64>;
2182 defm SADALP : SIMDLongTwoVectorTied<0, 0b00110, "sadalp",
2183 BinOpFrag<(add node:$LHS, (int_arm64_neon_saddlp node:$RHS))> >;
2184 defm SADDLP : SIMDLongTwoVector<0, 0b00010, "saddlp", int_arm64_neon_saddlp>;
2185 defm SCVTF : SIMDTwoVectorIntToFP<0, 0, 0b11101, "scvtf", sint_to_fp>;
2186 defm SHLL : SIMDVectorLShiftLongBySizeBHS;
2187 defm SQABS : SIMDTwoVectorBHSD<0, 0b00111, "sqabs", int_arm64_neon_sqabs>;
2188 defm SQNEG : SIMDTwoVectorBHSD<1, 0b00111, "sqneg", int_arm64_neon_sqneg>;
2189 defm SQXTN : SIMDMixedTwoVector<0, 0b10100, "sqxtn", int_arm64_neon_sqxtn>;
2190 defm SQXTUN : SIMDMixedTwoVector<1, 0b10010, "sqxtun", int_arm64_neon_sqxtun>;
2191 defm SUQADD : SIMDTwoVectorBHSDTied<0, 0b00011, "suqadd",int_arm64_neon_suqadd>;
2192 defm UADALP : SIMDLongTwoVectorTied<1, 0b00110, "uadalp",
2193 BinOpFrag<(add node:$LHS, (int_arm64_neon_uaddlp node:$RHS))> >;
2194 defm UADDLP : SIMDLongTwoVector<1, 0b00010, "uaddlp",
2195 int_arm64_neon_uaddlp>;
2196 defm UCVTF : SIMDTwoVectorIntToFP<1, 0, 0b11101, "ucvtf", uint_to_fp>;
2197 defm UQXTN : SIMDMixedTwoVector<1, 0b10100, "uqxtn", int_arm64_neon_uqxtn>;
2198 defm URECPE : SIMDTwoVectorS<0, 1, 0b11100, "urecpe", int_arm64_neon_urecpe>;
2199 defm URSQRTE: SIMDTwoVectorS<1, 1, 0b11100, "ursqrte", int_arm64_neon_ursqrte>;
2200 defm USQADD : SIMDTwoVectorBHSDTied<1, 0b00011, "usqadd",int_arm64_neon_usqadd>;
2201 defm XTN : SIMDMixedTwoVector<0, 0b10010, "xtn", trunc>;
2203 def : Pat<(v2f32 (ARM64rev64 V64:$Rn)), (REV64v2i32 V64:$Rn)>;
2204 def : Pat<(v4f32 (ARM64rev64 V128:$Rn)), (REV64v4i32 V128:$Rn)>;
2206 // Patterns for vector long shift (by element width). These need to match all
2207 // three of zext, sext and anyext so it's easier to pull the patterns out of the
2209 multiclass SIMDVectorLShiftLongBySizeBHSPats<SDPatternOperator ext> {
2210 def : Pat<(ARM64vshl (v8i16 (ext (v8i8 V64:$Rn))), (i32 8)),
2211 (SHLLv8i8 V64:$Rn)>;
2212 def : Pat<(ARM64vshl (v8i16 (ext (extract_high_v16i8 V128:$Rn))), (i32 8)),
2213 (SHLLv16i8 V128:$Rn)>;
2214 def : Pat<(ARM64vshl (v4i32 (ext (v4i16 V64:$Rn))), (i32 16)),
2215 (SHLLv4i16 V64:$Rn)>;
2216 def : Pat<(ARM64vshl (v4i32 (ext (extract_high_v8i16 V128:$Rn))), (i32 16)),
2217 (SHLLv8i16 V128:$Rn)>;
2218 def : Pat<(ARM64vshl (v2i64 (ext (v2i32 V64:$Rn))), (i32 32)),
2219 (SHLLv2i32 V64:$Rn)>;
2220 def : Pat<(ARM64vshl (v2i64 (ext (extract_high_v4i32 V128:$Rn))), (i32 32)),
2221 (SHLLv4i32 V128:$Rn)>;
2224 defm : SIMDVectorLShiftLongBySizeBHSPats<anyext>;
2225 defm : SIMDVectorLShiftLongBySizeBHSPats<zext>;
2226 defm : SIMDVectorLShiftLongBySizeBHSPats<sext>;
2228 //===----------------------------------------------------------------------===//
2229 // Advanced SIMD three vector instructions.
2230 //===----------------------------------------------------------------------===//
2232 defm ADD : SIMDThreeSameVector<0, 0b10000, "add", add>;
2233 defm ADDP : SIMDThreeSameVector<0, 0b10111, "addp", int_arm64_neon_addp>;
2234 defm CMEQ : SIMDThreeSameVector<1, 0b10001, "cmeq", ARM64cmeq>;
2235 defm CMGE : SIMDThreeSameVector<0, 0b00111, "cmge", ARM64cmge>;
2236 defm CMGT : SIMDThreeSameVector<0, 0b00110, "cmgt", ARM64cmgt>;
2237 defm CMHI : SIMDThreeSameVector<1, 0b00110, "cmhi", ARM64cmhi>;
2238 defm CMHS : SIMDThreeSameVector<1, 0b00111, "cmhs", ARM64cmhs>;
2239 defm CMTST : SIMDThreeSameVector<0, 0b10001, "cmtst", ARM64cmtst>;
2240 defm FABD : SIMDThreeSameVectorFP<1,1,0b11010,"fabd", int_arm64_neon_fabd>;
2241 defm FACGE : SIMDThreeSameVectorFPCmp<1,0,0b11101,"facge",int_arm64_neon_facge>;
2242 defm FACGT : SIMDThreeSameVectorFPCmp<1,1,0b11101,"facgt",int_arm64_neon_facgt>;
2243 defm FADDP : SIMDThreeSameVectorFP<1,0,0b11010,"faddp",int_arm64_neon_addp>;
2244 defm FADD : SIMDThreeSameVectorFP<0,0,0b11010,"fadd", fadd>;
2245 defm FCMEQ : SIMDThreeSameVectorFPCmp<0, 0, 0b11100, "fcmeq", ARM64fcmeq>;
2246 defm FCMGE : SIMDThreeSameVectorFPCmp<1, 0, 0b11100, "fcmge", ARM64fcmge>;
2247 defm FCMGT : SIMDThreeSameVectorFPCmp<1, 1, 0b11100, "fcmgt", ARM64fcmgt>;
2248 defm FDIV : SIMDThreeSameVectorFP<1,0,0b11111,"fdiv", fdiv>;
2249 defm FMAXNMP : SIMDThreeSameVectorFP<1,0,0b11000,"fmaxnmp", int_arm64_neon_fmaxnmp>;
2250 defm FMAXNM : SIMDThreeSameVectorFP<0,0,0b11000,"fmaxnm", int_arm64_neon_fmaxnm>;
2251 defm FMAXP : SIMDThreeSameVectorFP<1,0,0b11110,"fmaxp", int_arm64_neon_fmaxp>;
2252 defm FMAX : SIMDThreeSameVectorFP<0,0,0b11110,"fmax", ARM64fmax>;
2253 defm FMINNMP : SIMDThreeSameVectorFP<1,1,0b11000,"fminnmp", int_arm64_neon_fminnmp>;
2254 defm FMINNM : SIMDThreeSameVectorFP<0,1,0b11000,"fminnm", int_arm64_neon_fminnm>;
2255 defm FMINP : SIMDThreeSameVectorFP<1,1,0b11110,"fminp", int_arm64_neon_fminp>;
2256 defm FMIN : SIMDThreeSameVectorFP<0,1,0b11110,"fmin", ARM64fmin>;
2258 // NOTE: The operands of the PatFrag are reordered on FMLA/FMLS because the
2259 // instruction expects the addend first, while the fma intrinsic puts it last.
2260 defm FMLA : SIMDThreeSameVectorFPTied<0, 0, 0b11001, "fmla",
2261 TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)> >;
2262 defm FMLS : SIMDThreeSameVectorFPTied<0, 1, 0b11001, "fmls",
2263 TriOpFrag<(fma node:$MHS, (fneg node:$RHS), node:$LHS)> >;
2265 // The following def pats catch the case where the LHS of an FMA is negated.
2266 // The TriOpFrag above catches the case where the middle operand is negated.
2267 def : Pat<(v2f32 (fma (fneg V64:$Rn), V64:$Rm, V64:$Rd)),
2268 (FMLSv2f32 V64:$Rd, V64:$Rn, V64:$Rm)>;
2270 def : Pat<(v4f32 (fma (fneg V128:$Rn), V128:$Rm, V128:$Rd)),
2271 (FMLSv4f32 V128:$Rd, V128:$Rn, V128:$Rm)>;
2273 def : Pat<(v2f64 (fma (fneg V128:$Rn), V128:$Rm, V128:$Rd)),
2274 (FMLSv2f64 V128:$Rd, V128:$Rn, V128:$Rm)>;
2276 defm FMULX : SIMDThreeSameVectorFP<0,0,0b11011,"fmulx", int_arm64_neon_fmulx>;
2277 defm FMUL : SIMDThreeSameVectorFP<1,0,0b11011,"fmul", fmul>;
2278 defm FRECPS : SIMDThreeSameVectorFP<0,0,0b11111,"frecps", int_arm64_neon_frecps>;
2279 defm FRSQRTS : SIMDThreeSameVectorFP<0,1,0b11111,"frsqrts", int_arm64_neon_frsqrts>;
2280 defm FSUB : SIMDThreeSameVectorFP<0,1,0b11010,"fsub", fsub>;
2281 defm MLA : SIMDThreeSameVectorBHSTied<0, 0b10010, "mla",
2282 TriOpFrag<(add node:$LHS, (mul node:$MHS, node:$RHS))> >;
2283 defm MLS : SIMDThreeSameVectorBHSTied<1, 0b10010, "mls",
2284 TriOpFrag<(sub node:$LHS, (mul node:$MHS, node:$RHS))> >;
2285 defm MUL : SIMDThreeSameVectorBHS<0, 0b10011, "mul", mul>;
2286 defm PMUL : SIMDThreeSameVectorB<1, 0b10011, "pmul", int_arm64_neon_pmul>;
2287 defm SABA : SIMDThreeSameVectorBHSTied<0, 0b01111, "saba",
2288 TriOpFrag<(add node:$LHS, (int_arm64_neon_sabd node:$MHS, node:$RHS))> >;
2289 defm SABD : SIMDThreeSameVectorBHS<0,0b01110,"sabd", int_arm64_neon_sabd>;
2290 defm SHADD : SIMDThreeSameVectorBHS<0,0b00000,"shadd", int_arm64_neon_shadd>;
2291 defm SHSUB : SIMDThreeSameVectorBHS<0,0b00100,"shsub", int_arm64_neon_shsub>;
2292 defm SMAXP : SIMDThreeSameVectorBHS<0,0b10100,"smaxp", int_arm64_neon_smaxp>;
2293 defm SMAX : SIMDThreeSameVectorBHS<0,0b01100,"smax", int_arm64_neon_smax>;
2294 defm SMINP : SIMDThreeSameVectorBHS<0,0b10101,"sminp", int_arm64_neon_sminp>;
2295 defm SMIN : SIMDThreeSameVectorBHS<0,0b01101,"smin", int_arm64_neon_smin>;
2296 defm SQADD : SIMDThreeSameVector<0,0b00001,"sqadd", int_arm64_neon_sqadd>;
2297 defm SQDMULH : SIMDThreeSameVectorHS<0,0b10110,"sqdmulh",int_arm64_neon_sqdmulh>;
2298 defm SQRDMULH : SIMDThreeSameVectorHS<1,0b10110,"sqrdmulh",int_arm64_neon_sqrdmulh>;
2299 defm SQRSHL : SIMDThreeSameVector<0,0b01011,"sqrshl", int_arm64_neon_sqrshl>;
2300 defm SQSHL : SIMDThreeSameVector<0,0b01001,"sqshl", int_arm64_neon_sqshl>;
2301 defm SQSUB : SIMDThreeSameVector<0,0b00101,"sqsub", int_arm64_neon_sqsub>;
2302 defm SRHADD : SIMDThreeSameVectorBHS<0,0b00010,"srhadd",int_arm64_neon_srhadd>;
2303 defm SRSHL : SIMDThreeSameVector<0,0b01010,"srshl", int_arm64_neon_srshl>;
2304 defm SSHL : SIMDThreeSameVector<0,0b01000,"sshl", int_arm64_neon_sshl>;
2305 defm SUB : SIMDThreeSameVector<1,0b10000,"sub", sub>;
2306 defm UABA : SIMDThreeSameVectorBHSTied<1, 0b01111, "uaba",
2307 TriOpFrag<(add node:$LHS, (int_arm64_neon_uabd node:$MHS, node:$RHS))> >;
2308 defm UABD : SIMDThreeSameVectorBHS<1,0b01110,"uabd", int_arm64_neon_uabd>;
2309 defm UHADD : SIMDThreeSameVectorBHS<1,0b00000,"uhadd", int_arm64_neon_uhadd>;
2310 defm UHSUB : SIMDThreeSameVectorBHS<1,0b00100,"uhsub", int_arm64_neon_uhsub>;
2311 defm UMAXP : SIMDThreeSameVectorBHS<1,0b10100,"umaxp", int_arm64_neon_umaxp>;
2312 defm UMAX : SIMDThreeSameVectorBHS<1,0b01100,"umax", int_arm64_neon_umax>;
2313 defm UMINP : SIMDThreeSameVectorBHS<1,0b10101,"uminp", int_arm64_neon_uminp>;
2314 defm UMIN : SIMDThreeSameVectorBHS<1,0b01101,"umin", int_arm64_neon_umin>;
2315 defm UQADD : SIMDThreeSameVector<1,0b00001,"uqadd", int_arm64_neon_uqadd>;
2316 defm UQRSHL : SIMDThreeSameVector<1,0b01011,"uqrshl", int_arm64_neon_uqrshl>;
2317 defm UQSHL : SIMDThreeSameVector<1,0b01001,"uqshl", int_arm64_neon_uqshl>;
2318 defm UQSUB : SIMDThreeSameVector<1,0b00101,"uqsub", int_arm64_neon_uqsub>;
2319 defm URHADD : SIMDThreeSameVectorBHS<1,0b00010,"urhadd", int_arm64_neon_urhadd>;
2320 defm URSHL : SIMDThreeSameVector<1,0b01010,"urshl", int_arm64_neon_urshl>;
2321 defm USHL : SIMDThreeSameVector<1,0b01000,"ushl", int_arm64_neon_ushl>;
2323 defm AND : SIMDLogicalThreeVector<0, 0b00, "and", and>;
2324 defm BIC : SIMDLogicalThreeVector<0, 0b01, "bic",
2325 BinOpFrag<(and node:$LHS, (vnot node:$RHS))> >;
2326 defm BIF : SIMDLogicalThreeVector<1, 0b11, "bif">;
2327 defm BIT : SIMDLogicalThreeVectorTied<1, 0b10, "bit", ARM64bit>;
2328 defm BSL : SIMDLogicalThreeVectorTied<1, 0b01, "bsl",
2329 TriOpFrag<(or (and node:$LHS, node:$MHS), (and (vnot node:$LHS), node:$RHS))>>;
2330 defm EOR : SIMDLogicalThreeVector<1, 0b00, "eor", xor>;
2331 defm ORN : SIMDLogicalThreeVector<0, 0b11, "orn",
2332 BinOpFrag<(or node:$LHS, (vnot node:$RHS))> >;
2333 defm ORR : SIMDLogicalThreeVector<0, 0b10, "orr", or>;
2335 // FIXME: the .16b and .8b variantes should be emitted by the
2336 // AsmWriter. TableGen's AsmWriter-generator doesn't deal with variant syntaxes
2337 // in aliases yet though.
2338 def : InstAlias<"mov{\t$dst.16b, $src.16b|.16b\t$dst, $src}",
2339 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2340 def : InstAlias<"{mov\t$dst.8h, $src.8h|mov.8h\t$dst, $src}",
2341 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2342 def : InstAlias<"{mov\t$dst.4s, $src.4s|mov.4s\t$dst, $src}",
2343 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2344 def : InstAlias<"{mov\t$dst.2d, $src.2d|mov.2d\t$dst, $src}",
2345 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2347 def : InstAlias<"{mov\t$dst.8b, $src.8b|mov.8b\t$dst, $src}",
2348 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2349 def : InstAlias<"{mov\t$dst.4h, $src.4h|mov.4h\t$dst, $src}",
2350 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2351 def : InstAlias<"{mov\t$dst.2s, $src.2s|mov.2s\t$dst, $src}",
2352 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2353 def : InstAlias<"{mov\t$dst.1d, $src.1d|mov.1d\t$dst, $src}",
2354 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2356 def : InstAlias<"{cmls\t$dst.8b, $src1.8b, $src2.8b" #
2357 "|cmls.8b\t$dst, $src1, $src2}",
2358 (CMHSv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2359 def : InstAlias<"{cmls\t$dst.16b, $src1.16b, $src2.16b" #
2360 "|cmls.16b\t$dst, $src1, $src2}",
2361 (CMHSv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2362 def : InstAlias<"{cmls\t$dst.4h, $src1.4h, $src2.4h" #
2363 "|cmls.4h\t$dst, $src1, $src2}",
2364 (CMHSv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2365 def : InstAlias<"{cmls\t$dst.8h, $src1.8h, $src2.8h" #
2366 "|cmls.8h\t$dst, $src1, $src2}",
2367 (CMHSv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2368 def : InstAlias<"{cmls\t$dst.2s, $src1.2s, $src2.2s" #
2369 "|cmls.2s\t$dst, $src1, $src2}",
2370 (CMHSv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2371 def : InstAlias<"{cmls\t$dst.4s, $src1.4s, $src2.4s" #
2372 "|cmls.4s\t$dst, $src1, $src2}",
2373 (CMHSv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2374 def : InstAlias<"{cmls\t$dst.2d, $src1.2d, $src2.2d" #
2375 "|cmls.2d\t$dst, $src1, $src2}",
2376 (CMHSv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2378 def : InstAlias<"{cmlo\t$dst.8b, $src1.8b, $src2.8b" #
2379 "|cmlo.8b\t$dst, $src1, $src2}",
2380 (CMHIv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2381 def : InstAlias<"{cmlo\t$dst.16b, $src1.16b, $src2.16b" #
2382 "|cmlo.16b\t$dst, $src1, $src2}",
2383 (CMHIv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2384 def : InstAlias<"{cmlo\t$dst.4h, $src1.4h, $src2.4h" #
2385 "|cmlo.4h\t$dst, $src1, $src2}",
2386 (CMHIv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2387 def : InstAlias<"{cmlo\t$dst.8h, $src1.8h, $src2.8h" #
2388 "|cmlo.8h\t$dst, $src1, $src2}",
2389 (CMHIv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2390 def : InstAlias<"{cmlo\t$dst.2s, $src1.2s, $src2.2s" #
2391 "|cmlo.2s\t$dst, $src1, $src2}",
2392 (CMHIv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2393 def : InstAlias<"{cmlo\t$dst.4s, $src1.4s, $src2.4s" #
2394 "|cmlo.4s\t$dst, $src1, $src2}",
2395 (CMHIv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2396 def : InstAlias<"{cmlo\t$dst.2d, $src1.2d, $src2.2d" #
2397 "|cmlo.2d\t$dst, $src1, $src2}",
2398 (CMHIv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2400 def : InstAlias<"{cmle\t$dst.8b, $src1.8b, $src2.8b" #
2401 "|cmle.8b\t$dst, $src1, $src2}",
2402 (CMGEv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2403 def : InstAlias<"{cmle\t$dst.16b, $src1.16b, $src2.16b" #
2404 "|cmle.16b\t$dst, $src1, $src2}",
2405 (CMGEv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2406 def : InstAlias<"{cmle\t$dst.4h, $src1.4h, $src2.4h" #
2407 "|cmle.4h\t$dst, $src1, $src2}",
2408 (CMGEv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2409 def : InstAlias<"{cmle\t$dst.8h, $src1.8h, $src2.8h" #
2410 "|cmle.8h\t$dst, $src1, $src2}",
2411 (CMGEv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2412 def : InstAlias<"{cmle\t$dst.2s, $src1.2s, $src2.2s" #
2413 "|cmle.2s\t$dst, $src1, $src2}",
2414 (CMGEv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2415 def : InstAlias<"{cmle\t$dst.4s, $src1.4s, $src2.4s" #
2416 "|cmle.4s\t$dst, $src1, $src2}",
2417 (CMGEv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2418 def : InstAlias<"{cmle\t$dst.2d, $src1.2d, $src2.2d" #
2419 "|cmle.2d\t$dst, $src1, $src2}",
2420 (CMGEv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2422 def : InstAlias<"{cmlt\t$dst.8b, $src1.8b, $src2.8b" #
2423 "|cmlt.8b\t$dst, $src1, $src2}",
2424 (CMGTv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2425 def : InstAlias<"{cmlt\t$dst.16b, $src1.16b, $src2.16b" #
2426 "|cmlt.16b\t$dst, $src1, $src2}",
2427 (CMGTv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2428 def : InstAlias<"{cmlt\t$dst.4h, $src1.4h, $src2.4h" #
2429 "|cmlt.4h\t$dst, $src1, $src2}",
2430 (CMGTv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2431 def : InstAlias<"{cmlt\t$dst.8h, $src1.8h, $src2.8h" #
2432 "|cmlt.8h\t$dst, $src1, $src2}",
2433 (CMGTv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2434 def : InstAlias<"{cmlt\t$dst.2s, $src1.2s, $src2.2s" #
2435 "|cmlt.2s\t$dst, $src1, $src2}",
2436 (CMGTv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2437 def : InstAlias<"{cmlt\t$dst.4s, $src1.4s, $src2.4s" #
2438 "|cmlt.4s\t$dst, $src1, $src2}",
2439 (CMGTv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2440 def : InstAlias<"{cmlt\t$dst.2d, $src1.2d, $src2.2d" #
2441 "|cmlt.2d\t$dst, $src1, $src2}",
2442 (CMGTv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2444 def : InstAlias<"{fcmle\t$dst.2s, $src1.2s, $src2.2s" #
2445 "|fcmle.2s\t$dst, $src1, $src2}",
2446 (FCMGEv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
2447 def : InstAlias<"{fcmle\t$dst.4s, $src1.4s, $src2.4s" #
2448 "|fcmle.4s\t$dst, $src1, $src2}",
2449 (FCMGEv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
2450 def : InstAlias<"{fcmle\t$dst.2d, $src1.2d, $src2.2d" #
2451 "|fcmle.2d\t$dst, $src1, $src2}",
2452 (FCMGEv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
2454 def : InstAlias<"{fcmlt\t$dst.2s, $src1.2s, $src2.2s" #
2455 "|fcmlt.2s\t$dst, $src1, $src2}",
2456 (FCMGTv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
2457 def : InstAlias<"{fcmlt\t$dst.4s, $src1.4s, $src2.4s" #
2458 "|fcmlt.4s\t$dst, $src1, $src2}",
2459 (FCMGTv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
2460 def : InstAlias<"{fcmlt\t$dst.2d, $src1.2d, $src2.2d" #
2461 "|fcmlt.2d\t$dst, $src1, $src2}",
2462 (FCMGTv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
2464 def : InstAlias<"{facle\t$dst.2s, $src1.2s, $src2.2s" #
2465 "|facle.2s\t$dst, $src1, $src2}",
2466 (FACGEv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
2467 def : InstAlias<"{facle\t$dst.4s, $src1.4s, $src2.4s" #
2468 "|facle.4s\t$dst, $src1, $src2}",
2469 (FACGEv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
2470 def : InstAlias<"{facle\t$dst.2d, $src1.2d, $src2.2d" #
2471 "|facle.2d\t$dst, $src1, $src2}",
2472 (FACGEv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
2474 def : InstAlias<"{faclt\t$dst.2s, $src1.2s, $src2.2s" #
2475 "|faclt.2s\t$dst, $src1, $src2}",
2476 (FACGTv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
2477 def : InstAlias<"{faclt\t$dst.4s, $src1.4s, $src2.4s" #
2478 "|faclt.4s\t$dst, $src1, $src2}",
2479 (FACGTv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
2480 def : InstAlias<"{faclt\t$dst.2d, $src1.2d, $src2.2d" #
2481 "|faclt.2d\t$dst, $src1, $src2}",
2482 (FACGTv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
2484 //===----------------------------------------------------------------------===//
2485 // Advanced SIMD three scalar instructions.
2486 //===----------------------------------------------------------------------===//
2488 defm ADD : SIMDThreeScalarD<0, 0b10000, "add", add>;
2489 defm CMEQ : SIMDThreeScalarD<1, 0b10001, "cmeq", ARM64cmeq>;
2490 defm CMGE : SIMDThreeScalarD<0, 0b00111, "cmge", ARM64cmge>;
2491 defm CMGT : SIMDThreeScalarD<0, 0b00110, "cmgt", ARM64cmgt>;
2492 defm CMHI : SIMDThreeScalarD<1, 0b00110, "cmhi", ARM64cmhi>;
2493 defm CMHS : SIMDThreeScalarD<1, 0b00111, "cmhs", ARM64cmhs>;
2494 defm CMTST : SIMDThreeScalarD<0, 0b10001, "cmtst", ARM64cmtst>;
2495 defm FABD : SIMDThreeScalarSD<1, 1, 0b11010, "fabd", int_arm64_sisd_fabd>;
2496 def : Pat<(v1f64 (int_arm64_neon_fabd (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2497 (FABD64 FPR64:$Rn, FPR64:$Rm)>;
2498 defm FACGE : SIMDThreeScalarFPCmp<1, 0, 0b11101, "facge",
2499 int_arm64_neon_facge>;
2500 defm FACGT : SIMDThreeScalarFPCmp<1, 1, 0b11101, "facgt",
2501 int_arm64_neon_facgt>;
2502 defm FCMEQ : SIMDThreeScalarFPCmp<0, 0, 0b11100, "fcmeq", ARM64fcmeq>;
2503 defm FCMGE : SIMDThreeScalarFPCmp<1, 0, 0b11100, "fcmge", ARM64fcmge>;
2504 defm FCMGT : SIMDThreeScalarFPCmp<1, 1, 0b11100, "fcmgt", ARM64fcmgt>;
2505 defm FMULX : SIMDThreeScalarSD<0, 0, 0b11011, "fmulx", int_arm64_neon_fmulx>;
2506 defm FRECPS : SIMDThreeScalarSD<0, 0, 0b11111, "frecps", int_arm64_neon_frecps>;
2507 defm FRSQRTS : SIMDThreeScalarSD<0, 1, 0b11111, "frsqrts", int_arm64_neon_frsqrts>;
2508 defm SQADD : SIMDThreeScalarBHSD<0, 0b00001, "sqadd", int_arm64_neon_sqadd>;
2509 defm SQDMULH : SIMDThreeScalarHS< 0, 0b10110, "sqdmulh", int_arm64_neon_sqdmulh>;
2510 defm SQRDMULH : SIMDThreeScalarHS< 1, 0b10110, "sqrdmulh", int_arm64_neon_sqrdmulh>;
2511 defm SQRSHL : SIMDThreeScalarBHSD<0, 0b01011, "sqrshl",int_arm64_neon_sqrshl>;
2512 defm SQSHL : SIMDThreeScalarBHSD<0, 0b01001, "sqshl", int_arm64_neon_sqshl>;
2513 defm SQSUB : SIMDThreeScalarBHSD<0, 0b00101, "sqsub", int_arm64_neon_sqsub>;
2514 defm SRSHL : SIMDThreeScalarD< 0, 0b01010, "srshl", int_arm64_neon_srshl>;
2515 defm SSHL : SIMDThreeScalarD< 0, 0b01000, "sshl", int_arm64_neon_sshl>;
2516 defm SUB : SIMDThreeScalarD< 1, 0b10000, "sub", sub>;
2517 defm UQADD : SIMDThreeScalarBHSD<1, 0b00001, "uqadd", int_arm64_neon_uqadd>;
2518 defm UQRSHL : SIMDThreeScalarBHSD<1, 0b01011, "uqrshl",int_arm64_neon_uqrshl>;
2519 defm UQSHL : SIMDThreeScalarBHSD<1, 0b01001, "uqshl", int_arm64_neon_uqshl>;
2520 defm UQSUB : SIMDThreeScalarBHSD<1, 0b00101, "uqsub", int_arm64_neon_uqsub>;
2521 defm URSHL : SIMDThreeScalarD< 1, 0b01010, "urshl", int_arm64_neon_urshl>;
2522 defm USHL : SIMDThreeScalarD< 1, 0b01000, "ushl", int_arm64_neon_ushl>;
2524 def : InstAlias<"cmls $dst, $src1, $src2",
2525 (CMHSv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2526 def : InstAlias<"cmle $dst, $src1, $src2",
2527 (CMGEv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2528 def : InstAlias<"cmlo $dst, $src1, $src2",
2529 (CMHIv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2530 def : InstAlias<"cmlt $dst, $src1, $src2",
2531 (CMGTv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2532 def : InstAlias<"fcmle $dst, $src1, $src2",
2533 (FCMGE32 FPR32:$dst, FPR32:$src2, FPR32:$src1)>;
2534 def : InstAlias<"fcmle $dst, $src1, $src2",
2535 (FCMGE64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2536 def : InstAlias<"fcmlt $dst, $src1, $src2",
2537 (FCMGT32 FPR32:$dst, FPR32:$src2, FPR32:$src1)>;
2538 def : InstAlias<"fcmlt $dst, $src1, $src2",
2539 (FCMGT64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2540 def : InstAlias<"facle $dst, $src1, $src2",
2541 (FACGE32 FPR32:$dst, FPR32:$src2, FPR32:$src1)>;
2542 def : InstAlias<"facle $dst, $src1, $src2",
2543 (FACGE64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2544 def : InstAlias<"faclt $dst, $src1, $src2",
2545 (FACGT32 FPR32:$dst, FPR32:$src2, FPR32:$src1)>;
2546 def : InstAlias<"faclt $dst, $src1, $src2",
2547 (FACGT64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2549 //===----------------------------------------------------------------------===//
2550 // Advanced SIMD three scalar instructions (mixed operands).
2551 //===----------------------------------------------------------------------===//
2552 defm SQDMULL : SIMDThreeScalarMixedHS<0, 0b11010, "sqdmull",
2553 int_arm64_neon_sqdmulls_scalar>;
2554 defm SQDMLAL : SIMDThreeScalarMixedTiedHS<0, 0b10010, "sqdmlal">;
2555 defm SQDMLSL : SIMDThreeScalarMixedTiedHS<0, 0b10110, "sqdmlsl">;
2557 def : Pat<(i64 (int_arm64_neon_sqadd (i64 FPR64:$Rd),
2558 (i64 (int_arm64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
2559 (i32 FPR32:$Rm))))),
2560 (SQDMLALi32 FPR64:$Rd, FPR32:$Rn, FPR32:$Rm)>;
2561 def : Pat<(i64 (int_arm64_neon_sqsub (i64 FPR64:$Rd),
2562 (i64 (int_arm64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
2563 (i32 FPR32:$Rm))))),
2564 (SQDMLSLi32 FPR64:$Rd, FPR32:$Rn, FPR32:$Rm)>;
2566 //===----------------------------------------------------------------------===//
2567 // Advanced SIMD two scalar instructions.
2568 //===----------------------------------------------------------------------===//
2570 defm ABS : SIMDTwoScalarD< 0, 0b01011, "abs", int_arm64_neon_abs>;
2571 defm CMEQ : SIMDCmpTwoScalarD< 0, 0b01001, "cmeq", ARM64cmeqz>;
2572 defm CMGE : SIMDCmpTwoScalarD< 1, 0b01000, "cmge", ARM64cmgez>;
2573 defm CMGT : SIMDCmpTwoScalarD< 0, 0b01000, "cmgt", ARM64cmgtz>;
2574 defm CMLE : SIMDCmpTwoScalarD< 1, 0b01001, "cmle", ARM64cmlez>;
2575 defm CMLT : SIMDCmpTwoScalarD< 0, 0b01010, "cmlt", ARM64cmltz>;
2576 defm FCMEQ : SIMDCmpTwoScalarSD<0, 1, 0b01101, "fcmeq", ARM64fcmeqz>;
2577 defm FCMGE : SIMDCmpTwoScalarSD<1, 1, 0b01100, "fcmge", ARM64fcmgez>;
2578 defm FCMGT : SIMDCmpTwoScalarSD<0, 1, 0b01100, "fcmgt", ARM64fcmgtz>;
2579 defm FCMLE : SIMDCmpTwoScalarSD<1, 1, 0b01101, "fcmle", ARM64fcmlez>;
2580 defm FCMLT : SIMDCmpTwoScalarSD<0, 1, 0b01110, "fcmlt", ARM64fcmltz>;
2581 defm FCVTAS : SIMDTwoScalarSD< 0, 0, 0b11100, "fcvtas">;
2582 defm FCVTAU : SIMDTwoScalarSD< 1, 0, 0b11100, "fcvtau">;
2583 defm FCVTMS : SIMDTwoScalarSD< 0, 0, 0b11011, "fcvtms">;
2584 defm FCVTMU : SIMDTwoScalarSD< 1, 0, 0b11011, "fcvtmu">;
2585 defm FCVTNS : SIMDTwoScalarSD< 0, 0, 0b11010, "fcvtns">;
2586 defm FCVTNU : SIMDTwoScalarSD< 1, 0, 0b11010, "fcvtnu">;
2587 defm FCVTPS : SIMDTwoScalarSD< 0, 1, 0b11010, "fcvtps">;
2588 defm FCVTPU : SIMDTwoScalarSD< 1, 1, 0b11010, "fcvtpu">;
2589 def FCVTXNv1i64 : SIMDInexactCvtTwoScalar<0b10110, "fcvtxn">;
2590 defm FCVTZS : SIMDTwoScalarSD< 0, 1, 0b11011, "fcvtzs">;
2591 defm FCVTZU : SIMDTwoScalarSD< 1, 1, 0b11011, "fcvtzu">;
2592 defm FRECPE : SIMDTwoScalarSD< 0, 1, 0b11101, "frecpe">;
2593 defm FRECPX : SIMDTwoScalarSD< 0, 1, 0b11111, "frecpx">;
2594 defm FRSQRTE : SIMDTwoScalarSD< 1, 1, 0b11101, "frsqrte">;
2595 defm NEG : SIMDTwoScalarD< 1, 0b01011, "neg",
2596 UnOpFrag<(sub immAllZerosV, node:$LHS)> >;
2597 defm SCVTF : SIMDTwoScalarCVTSD< 0, 0, 0b11101, "scvtf", ARM64sitof>;
2598 defm SQABS : SIMDTwoScalarBHSD< 0, 0b00111, "sqabs", int_arm64_neon_sqabs>;
2599 defm SQNEG : SIMDTwoScalarBHSD< 1, 0b00111, "sqneg", int_arm64_neon_sqneg>;
2600 defm SQXTN : SIMDTwoScalarMixedBHS< 0, 0b10100, "sqxtn", int_arm64_neon_scalar_sqxtn>;
2601 defm SQXTUN : SIMDTwoScalarMixedBHS< 1, 0b10010, "sqxtun", int_arm64_neon_scalar_sqxtun>;
2602 defm SUQADD : SIMDTwoScalarBHSDTied< 0, 0b00011, "suqadd",
2603 int_arm64_neon_suqadd>;
2604 defm UCVTF : SIMDTwoScalarCVTSD< 1, 0, 0b11101, "ucvtf", ARM64uitof>;
2605 defm UQXTN : SIMDTwoScalarMixedBHS<1, 0b10100, "uqxtn", int_arm64_neon_scalar_uqxtn>;
2606 defm USQADD : SIMDTwoScalarBHSDTied< 1, 0b00011, "usqadd",
2607 int_arm64_neon_usqadd>;
2609 def : Pat<(v1i64 (int_arm64_neon_fcvtas (v1f64 FPR64:$Rn))),
2610 (FCVTASv1i64 FPR64:$Rn)>;
2611 def : Pat<(v1i64 (int_arm64_neon_fcvtau (v1f64 FPR64:$Rn))),
2612 (FCVTAUv1i64 FPR64:$Rn)>;
2613 def : Pat<(v1i64 (int_arm64_neon_fcvtms (v1f64 FPR64:$Rn))),
2614 (FCVTMSv1i64 FPR64:$Rn)>;
2615 def : Pat<(v1i64 (int_arm64_neon_fcvtmu (v1f64 FPR64:$Rn))),
2616 (FCVTMUv1i64 FPR64:$Rn)>;
2617 def : Pat<(v1i64 (int_arm64_neon_fcvtns (v1f64 FPR64:$Rn))),
2618 (FCVTNSv1i64 FPR64:$Rn)>;
2619 def : Pat<(v1i64 (int_arm64_neon_fcvtnu (v1f64 FPR64:$Rn))),
2620 (FCVTNUv1i64 FPR64:$Rn)>;
2621 def : Pat<(v1i64 (int_arm64_neon_fcvtps (v1f64 FPR64:$Rn))),
2622 (FCVTPSv1i64 FPR64:$Rn)>;
2623 def : Pat<(v1i64 (int_arm64_neon_fcvtpu (v1f64 FPR64:$Rn))),
2624 (FCVTPUv1i64 FPR64:$Rn)>;
2626 def : Pat<(f32 (int_arm64_neon_frecpe (f32 FPR32:$Rn))),
2627 (FRECPEv1i32 FPR32:$Rn)>;
2628 def : Pat<(f64 (int_arm64_neon_frecpe (f64 FPR64:$Rn))),
2629 (FRECPEv1i64 FPR64:$Rn)>;
2630 def : Pat<(v1f64 (int_arm64_neon_frecpe (v1f64 FPR64:$Rn))),
2631 (FRECPEv1i64 FPR64:$Rn)>;
2633 def : Pat<(f32 (int_arm64_neon_frecpx (f32 FPR32:$Rn))),
2634 (FRECPXv1i32 FPR32:$Rn)>;
2635 def : Pat<(f64 (int_arm64_neon_frecpx (f64 FPR64:$Rn))),
2636 (FRECPXv1i64 FPR64:$Rn)>;
2638 def : Pat<(f32 (int_arm64_neon_frsqrte (f32 FPR32:$Rn))),
2639 (FRSQRTEv1i32 FPR32:$Rn)>;
2640 def : Pat<(f64 (int_arm64_neon_frsqrte (f64 FPR64:$Rn))),
2641 (FRSQRTEv1i64 FPR64:$Rn)>;
2642 def : Pat<(v1f64 (int_arm64_neon_frsqrte (v1f64 FPR64:$Rn))),
2643 (FRSQRTEv1i64 FPR64:$Rn)>;
2645 // If an integer is about to be converted to a floating point value,
2646 // just load it on the floating point unit.
2647 // Here are the patterns for 8 and 16-bits to float.
2649 def : Pat <(f32 (uint_to_fp (i32 (zextloadi8 ro_indexed8:$addr)))),
2650 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2651 (LDRBro ro_indexed8:$addr), bsub))>;
2652 def : Pat <(f32 (uint_to_fp (i32 (zextloadi8 am_indexed8:$addr)))),
2653 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2654 (LDRBui am_indexed8:$addr), bsub))>;
2655 def : Pat <(f32 (uint_to_fp (i32 (zextloadi8 am_unscaled8:$addr)))),
2656 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2657 (LDURBi am_unscaled8:$addr), bsub))>;
2658 // 16-bits -> float.
2659 def : Pat <(f32 (uint_to_fp (i32 (zextloadi16 ro_indexed16:$addr)))),
2660 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2661 (LDRHro ro_indexed16:$addr), hsub))>;
2662 def : Pat <(f32 (uint_to_fp (i32 (zextloadi16 am_indexed16:$addr)))),
2663 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2664 (LDRHui am_indexed16:$addr), hsub))>;
2665 def : Pat <(f32 (uint_to_fp (i32 (zextloadi16 am_unscaled16:$addr)))),
2666 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2667 (LDURHi am_unscaled16:$addr), hsub))>;
2668 // 32-bits are handled in target specific dag combine:
2669 // performIntToFpCombine.
2670 // 64-bits integer to 32-bits floating point, not possible with
2671 // UCVTF on floating point registers (both source and destination
2672 // must have the same size).
2674 // Here are the patterns for 8, 16, 32, and 64-bits to double.
2675 // 8-bits -> double.
2676 def : Pat <(f64 (uint_to_fp (i32 (zextloadi8 ro_indexed8:$addr)))),
2677 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2678 (LDRBro ro_indexed8:$addr), bsub))>;
2679 def : Pat <(f64 (uint_to_fp (i32 (zextloadi8 am_indexed8:$addr)))),
2680 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2681 (LDRBui am_indexed8:$addr), bsub))>;
2682 def : Pat <(f64 (uint_to_fp (i32 (zextloadi8 am_unscaled8:$addr)))),
2683 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2684 (LDURBi am_unscaled8:$addr), bsub))>;
2685 // 16-bits -> double.
2686 def : Pat <(f64 (uint_to_fp (i32 (zextloadi16 ro_indexed16:$addr)))),
2687 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2688 (LDRHro ro_indexed16:$addr), hsub))>;
2689 def : Pat <(f64 (uint_to_fp (i32 (zextloadi16 am_indexed16:$addr)))),
2690 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2691 (LDRHui am_indexed16:$addr), hsub))>;
2692 def : Pat <(f64 (uint_to_fp (i32 (zextloadi16 am_unscaled16:$addr)))),
2693 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2694 (LDURHi am_unscaled16:$addr), hsub))>;
2695 // 32-bits -> double.
2696 def : Pat <(f64 (uint_to_fp (i32 (load ro_indexed32:$addr)))),
2697 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2698 (LDRSro ro_indexed32:$addr), ssub))>;
2699 def : Pat <(f64 (uint_to_fp (i32 (load am_indexed32:$addr)))),
2700 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2701 (LDRSui am_indexed32:$addr), ssub))>;
2702 def : Pat <(f64 (uint_to_fp (i32 (load am_unscaled32:$addr)))),
2703 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2704 (LDURSi am_unscaled32:$addr), ssub))>;
2705 // 64-bits -> double are handled in target specific dag combine:
2706 // performIntToFpCombine.
2708 //===----------------------------------------------------------------------===//
2709 // Advanced SIMD three different-sized vector instructions.
2710 //===----------------------------------------------------------------------===//
2712 defm ADDHN : SIMDNarrowThreeVectorBHS<0,0b0100,"addhn", int_arm64_neon_addhn>;
2713 defm SUBHN : SIMDNarrowThreeVectorBHS<0,0b0110,"subhn", int_arm64_neon_subhn>;
2714 defm RADDHN : SIMDNarrowThreeVectorBHS<1,0b0100,"raddhn",int_arm64_neon_raddhn>;
2715 defm RSUBHN : SIMDNarrowThreeVectorBHS<1,0b0110,"rsubhn",int_arm64_neon_rsubhn>;
2716 defm PMULL : SIMDDifferentThreeVectorBD<0,0b1110,"pmull",int_arm64_neon_pmull>;
2717 defm SABAL : SIMDLongThreeVectorTiedBHSabal<0,0b0101,"sabal",
2718 int_arm64_neon_sabd>;
2719 defm SABDL : SIMDLongThreeVectorBHSabdl<0, 0b0111, "sabdl",
2720 int_arm64_neon_sabd>;
2721 defm SADDL : SIMDLongThreeVectorBHS< 0, 0b0000, "saddl",
2722 BinOpFrag<(add (sext node:$LHS), (sext node:$RHS))>>;
2723 defm SADDW : SIMDWideThreeVectorBHS< 0, 0b0001, "saddw",
2724 BinOpFrag<(add node:$LHS, (sext node:$RHS))>>;
2725 defm SMLAL : SIMDLongThreeVectorTiedBHS<0, 0b1000, "smlal",
2726 TriOpFrag<(add node:$LHS, (int_arm64_neon_smull node:$MHS, node:$RHS))>>;
2727 defm SMLSL : SIMDLongThreeVectorTiedBHS<0, 0b1010, "smlsl",
2728 TriOpFrag<(sub node:$LHS, (int_arm64_neon_smull node:$MHS, node:$RHS))>>;
2729 defm SMULL : SIMDLongThreeVectorBHS<0, 0b1100, "smull", int_arm64_neon_smull>;
2730 defm SQDMLAL : SIMDLongThreeVectorSQDMLXTiedHS<0, 0b1001, "sqdmlal",
2731 int_arm64_neon_sqadd>;
2732 defm SQDMLSL : SIMDLongThreeVectorSQDMLXTiedHS<0, 0b1011, "sqdmlsl",
2733 int_arm64_neon_sqsub>;
2734 defm SQDMULL : SIMDLongThreeVectorHS<0, 0b1101, "sqdmull",
2735 int_arm64_neon_sqdmull>;
2736 defm SSUBL : SIMDLongThreeVectorBHS<0, 0b0010, "ssubl",
2737 BinOpFrag<(sub (sext node:$LHS), (sext node:$RHS))>>;
2738 defm SSUBW : SIMDWideThreeVectorBHS<0, 0b0011, "ssubw",
2739 BinOpFrag<(sub node:$LHS, (sext node:$RHS))>>;
2740 defm UABAL : SIMDLongThreeVectorTiedBHSabal<1, 0b0101, "uabal",
2741 int_arm64_neon_uabd>;
2742 defm UABDL : SIMDLongThreeVectorBHSabdl<1, 0b0111, "uabdl",
2743 int_arm64_neon_uabd>;
2744 defm UADDL : SIMDLongThreeVectorBHS<1, 0b0000, "uaddl",
2745 BinOpFrag<(add (zext node:$LHS), (zext node:$RHS))>>;
2746 defm UADDW : SIMDWideThreeVectorBHS<1, 0b0001, "uaddw",
2747 BinOpFrag<(add node:$LHS, (zext node:$RHS))>>;
2748 defm UMLAL : SIMDLongThreeVectorTiedBHS<1, 0b1000, "umlal",
2749 TriOpFrag<(add node:$LHS, (int_arm64_neon_umull node:$MHS, node:$RHS))>>;
2750 defm UMLSL : SIMDLongThreeVectorTiedBHS<1, 0b1010, "umlsl",
2751 TriOpFrag<(sub node:$LHS, (int_arm64_neon_umull node:$MHS, node:$RHS))>>;
2752 defm UMULL : SIMDLongThreeVectorBHS<1, 0b1100, "umull", int_arm64_neon_umull>;
2753 defm USUBL : SIMDLongThreeVectorBHS<1, 0b0010, "usubl",
2754 BinOpFrag<(sub (zext node:$LHS), (zext node:$RHS))>>;
2755 defm USUBW : SIMDWideThreeVectorBHS< 1, 0b0011, "usubw",
2756 BinOpFrag<(sub node:$LHS, (zext node:$RHS))>>;
2758 // Patterns for 64-bit pmull
2759 def : Pat<(int_arm64_neon_pmull64 V64:$Rn, V64:$Rm),
2760 (PMULLv1i64 V64:$Rn, V64:$Rm)>;
2761 def : Pat<(int_arm64_neon_pmull64 (vector_extract (v2i64 V128:$Rn), (i64 1)),
2762 (vector_extract (v2i64 V128:$Rm), (i64 1))),
2763 (PMULLv2i64 V128:$Rn, V128:$Rm)>;
2765 // CodeGen patterns for addhn and subhn instructions, which can actually be
2766 // written in LLVM IR without too much difficulty.
2769 def : Pat<(v8i8 (trunc (v8i16 (ARM64vlshr (add V128:$Rn, V128:$Rm), (i32 8))))),
2770 (ADDHNv8i16_v8i8 V128:$Rn, V128:$Rm)>;
2771 def : Pat<(v4i16 (trunc (v4i32 (ARM64vlshr (add V128:$Rn, V128:$Rm),
2773 (ADDHNv4i32_v4i16 V128:$Rn, V128:$Rm)>;
2774 def : Pat<(v2i32 (trunc (v2i64 (ARM64vlshr (add V128:$Rn, V128:$Rm),
2776 (ADDHNv2i64_v2i32 V128:$Rn, V128:$Rm)>;
2777 def : Pat<(concat_vectors (v8i8 V64:$Rd),
2778 (trunc (v8i16 (ARM64vlshr (add V128:$Rn, V128:$Rm),
2780 (ADDHNv8i16_v16i8 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
2781 V128:$Rn, V128:$Rm)>;
2782 def : Pat<(concat_vectors (v4i16 V64:$Rd),
2783 (trunc (v4i32 (ARM64vlshr (add V128:$Rn, V128:$Rm),
2785 (ADDHNv4i32_v8i16 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
2786 V128:$Rn, V128:$Rm)>;
2787 def : Pat<(concat_vectors (v2i32 V64:$Rd),
2788 (trunc (v2i64 (ARM64vlshr (add V128:$Rn, V128:$Rm),
2790 (ADDHNv2i64_v4i32 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
2791 V128:$Rn, V128:$Rm)>;
2794 def : Pat<(v8i8 (trunc (v8i16 (ARM64vlshr (sub V128:$Rn, V128:$Rm), (i32 8))))),
2795 (SUBHNv8i16_v8i8 V128:$Rn, V128:$Rm)>;
2796 def : Pat<(v4i16 (trunc (v4i32 (ARM64vlshr (sub V128:$Rn, V128:$Rm),
2798 (SUBHNv4i32_v4i16 V128:$Rn, V128:$Rm)>;
2799 def : Pat<(v2i32 (trunc (v2i64 (ARM64vlshr (sub V128:$Rn, V128:$Rm),
2801 (SUBHNv2i64_v2i32 V128:$Rn, V128:$Rm)>;
2802 def : Pat<(concat_vectors (v8i8 V64:$Rd),
2803 (trunc (v8i16 (ARM64vlshr (sub V128:$Rn, V128:$Rm),
2805 (SUBHNv8i16_v16i8 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
2806 V128:$Rn, V128:$Rm)>;
2807 def : Pat<(concat_vectors (v4i16 V64:$Rd),
2808 (trunc (v4i32 (ARM64vlshr (sub V128:$Rn, V128:$Rm),
2810 (SUBHNv4i32_v8i16 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
2811 V128:$Rn, V128:$Rm)>;
2812 def : Pat<(concat_vectors (v2i32 V64:$Rd),
2813 (trunc (v2i64 (ARM64vlshr (sub V128:$Rn, V128:$Rm),
2815 (SUBHNv2i64_v4i32 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
2816 V128:$Rn, V128:$Rm)>;
2818 //----------------------------------------------------------------------------
2819 // AdvSIMD bitwise extract from vector instruction.
2820 //----------------------------------------------------------------------------
2822 defm EXT : SIMDBitwiseExtract<"ext">;
2824 def : Pat<(v4i16 (ARM64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
2825 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
2826 def : Pat<(v8i16 (ARM64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
2827 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
2828 def : Pat<(v2i32 (ARM64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
2829 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
2830 def : Pat<(v2f32 (ARM64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
2831 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
2832 def : Pat<(v4i32 (ARM64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
2833 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
2834 def : Pat<(v4f32 (ARM64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
2835 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
2836 def : Pat<(v2i64 (ARM64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
2837 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
2838 def : Pat<(v2f64 (ARM64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
2839 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
2841 // We use EXT to handle extract_subvector to copy the upper 64-bits of a
2843 def : Pat<(v8i8 (extract_subvector V128:$Rn, (i64 8))),
2844 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
2845 def : Pat<(v4i16 (extract_subvector V128:$Rn, (i64 4))),
2846 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
2847 def : Pat<(v2i32 (extract_subvector V128:$Rn, (i64 2))),
2848 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
2849 def : Pat<(v1i64 (extract_subvector V128:$Rn, (i64 1))),
2850 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
2851 def : Pat<(v2f32 (extract_subvector V128:$Rn, (i64 2))),
2852 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
2853 def : Pat<(v1f64 (extract_subvector V128:$Rn, (i64 1))),
2854 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
2857 //----------------------------------------------------------------------------
2858 // AdvSIMD zip vector
2859 //----------------------------------------------------------------------------
2861 defm TRN1 : SIMDZipVector<0b010, "trn1", ARM64trn1>;
2862 defm TRN2 : SIMDZipVector<0b110, "trn2", ARM64trn2>;
2863 defm UZP1 : SIMDZipVector<0b001, "uzp1", ARM64uzp1>;
2864 defm UZP2 : SIMDZipVector<0b101, "uzp2", ARM64uzp2>;
2865 defm ZIP1 : SIMDZipVector<0b011, "zip1", ARM64zip1>;
2866 defm ZIP2 : SIMDZipVector<0b111, "zip2", ARM64zip2>;
2868 //----------------------------------------------------------------------------
2869 // AdvSIMD TBL/TBX instructions
2870 //----------------------------------------------------------------------------
2872 defm TBL : SIMDTableLookup< 0, "tbl">;
2873 defm TBX : SIMDTableLookupTied<1, "tbx">;
2875 def : Pat<(v8i8 (int_arm64_neon_tbl1 (v16i8 VecListOne128:$Rn), (v8i8 V64:$Ri))),
2876 (TBLv8i8One VecListOne128:$Rn, V64:$Ri)>;
2877 def : Pat<(v16i8 (int_arm64_neon_tbl1 (v16i8 V128:$Ri), (v16i8 V128:$Rn))),
2878 (TBLv16i8One V128:$Ri, V128:$Rn)>;
2880 def : Pat<(v8i8 (int_arm64_neon_tbx1 (v8i8 V64:$Rd),
2881 (v16i8 VecListOne128:$Rn), (v8i8 V64:$Ri))),
2882 (TBXv8i8One V64:$Rd, VecListOne128:$Rn, V64:$Ri)>;
2883 def : Pat<(v16i8 (int_arm64_neon_tbx1 (v16i8 V128:$Rd),
2884 (v16i8 V128:$Ri), (v16i8 V128:$Rn))),
2885 (TBXv16i8One V128:$Rd, V128:$Ri, V128:$Rn)>;
2888 //----------------------------------------------------------------------------
2889 // AdvSIMD scalar CPY instruction
2890 //----------------------------------------------------------------------------
2892 defm CPY : SIMDScalarCPY<"cpy">;
2894 //----------------------------------------------------------------------------
2895 // AdvSIMD scalar pairwise instructions
2896 //----------------------------------------------------------------------------
2898 defm ADDP : SIMDPairwiseScalarD<0, 0b11011, "addp">;
2899 defm FADDP : SIMDPairwiseScalarSD<1, 0, 0b01101, "faddp">;
2900 defm FMAXNMP : SIMDPairwiseScalarSD<1, 0, 0b01100, "fmaxnmp">;
2901 defm FMAXP : SIMDPairwiseScalarSD<1, 0, 0b01111, "fmaxp">;
2902 defm FMINNMP : SIMDPairwiseScalarSD<1, 1, 0b01100, "fminnmp">;
2903 defm FMINP : SIMDPairwiseScalarSD<1, 1, 0b01111, "fminp">;
2904 def : Pat<(i64 (int_arm64_neon_saddv (v2i64 V128:$Rn))),
2905 (ADDPv2i64p V128:$Rn)>;
2906 def : Pat<(i64 (int_arm64_neon_uaddv (v2i64 V128:$Rn))),
2907 (ADDPv2i64p V128:$Rn)>;
2908 def : Pat<(f32 (int_arm64_neon_faddv (v2f32 V64:$Rn))),
2909 (FADDPv2i32p V64:$Rn)>;
2910 def : Pat<(f32 (int_arm64_neon_faddv (v4f32 V128:$Rn))),
2911 (FADDPv2i32p (EXTRACT_SUBREG (FADDPv4f32 V128:$Rn, V128:$Rn), dsub))>;
2912 def : Pat<(f64 (int_arm64_neon_faddv (v2f64 V128:$Rn))),
2913 (FADDPv2i64p V128:$Rn)>;
2914 def : Pat<(f32 (int_arm64_neon_fmaxnmv (v2f32 V64:$Rn))),
2915 (FMAXNMPv2i32p V64:$Rn)>;
2916 def : Pat<(f64 (int_arm64_neon_fmaxnmv (v2f64 V128:$Rn))),
2917 (FMAXNMPv2i64p V128:$Rn)>;
2918 def : Pat<(f32 (int_arm64_neon_fmaxv (v2f32 V64:$Rn))),
2919 (FMAXPv2i32p V64:$Rn)>;
2920 def : Pat<(f64 (int_arm64_neon_fmaxv (v2f64 V128:$Rn))),
2921 (FMAXPv2i64p V128:$Rn)>;
2922 def : Pat<(f32 (int_arm64_neon_fminnmv (v2f32 V64:$Rn))),
2923 (FMINNMPv2i32p V64:$Rn)>;
2924 def : Pat<(f64 (int_arm64_neon_fminnmv (v2f64 V128:$Rn))),
2925 (FMINNMPv2i64p V128:$Rn)>;
2926 def : Pat<(f32 (int_arm64_neon_fminv (v2f32 V64:$Rn))),
2927 (FMINPv2i32p V64:$Rn)>;
2928 def : Pat<(f64 (int_arm64_neon_fminv (v2f64 V128:$Rn))),
2929 (FMINPv2i64p V128:$Rn)>;
2931 //----------------------------------------------------------------------------
2932 // AdvSIMD INS/DUP instructions
2933 //----------------------------------------------------------------------------
2935 def DUPv8i8gpr : SIMDDupFromMain<0, 0b00001, ".8b", v8i8, V64, GPR32>;
2936 def DUPv16i8gpr : SIMDDupFromMain<1, 0b00001, ".16b", v16i8, V128, GPR32>;
2937 def DUPv4i16gpr : SIMDDupFromMain<0, 0b00010, ".4h", v4i16, V64, GPR32>;
2938 def DUPv8i16gpr : SIMDDupFromMain<1, 0b00010, ".8h", v8i16, V128, GPR32>;
2939 def DUPv2i32gpr : SIMDDupFromMain<0, 0b00100, ".2s", v2i32, V64, GPR32>;
2940 def DUPv4i32gpr : SIMDDupFromMain<1, 0b00100, ".4s", v4i32, V128, GPR32>;
2941 def DUPv2i64gpr : SIMDDupFromMain<1, 0b01000, ".2d", v2i64, V128, GPR64>;
2943 def DUPv2i64lane : SIMDDup64FromElement;
2944 def DUPv2i32lane : SIMDDup32FromElement<0, ".2s", v2i32, V64>;
2945 def DUPv4i32lane : SIMDDup32FromElement<1, ".4s", v4i32, V128>;
2946 def DUPv4i16lane : SIMDDup16FromElement<0, ".4h", v4i16, V64>;
2947 def DUPv8i16lane : SIMDDup16FromElement<1, ".8h", v8i16, V128>;
2948 def DUPv8i8lane : SIMDDup8FromElement <0, ".8b", v8i8, V64>;
2949 def DUPv16i8lane : SIMDDup8FromElement <1, ".16b", v16i8, V128>;
2951 def : Pat<(v2f32 (ARM64dup (f32 FPR32:$Rn))),
2952 (v2f32 (DUPv2i32lane
2953 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rn, ssub),
2955 def : Pat<(v4f32 (ARM64dup (f32 FPR32:$Rn))),
2956 (v4f32 (DUPv4i32lane
2957 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rn, ssub),
2959 def : Pat<(v2f64 (ARM64dup (f64 FPR64:$Rn))),
2960 (v2f64 (DUPv2i64lane
2961 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$Rn, dsub),
2964 def : Pat<(v2f32 (ARM64duplane32 (v4f32 V128:$Rn), VectorIndexS:$imm)),
2965 (DUPv2i32lane V128:$Rn, VectorIndexS:$imm)>;
2966 def : Pat<(v4f32 (ARM64duplane32 (v4f32 V128:$Rn), VectorIndexS:$imm)),
2967 (DUPv4i32lane V128:$Rn, VectorIndexS:$imm)>;
2968 def : Pat<(v2f64 (ARM64duplane64 (v2f64 V128:$Rn), VectorIndexD:$imm)),
2969 (DUPv2i64lane V128:$Rn, VectorIndexD:$imm)>;
2974 def : Pat<(sext_inreg (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx), i8),
2975 (i32 (SMOVvi8to32 V128:$Rn, VectorIndexB:$idx))>;
2976 def : Pat<(sext_inreg (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx), i8),
2977 (i64 (SMOVvi8to64 V128:$Rn, VectorIndexB:$idx))>;
2978 def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
2979 (i32 (SMOVvi16to32 V128:$Rn, VectorIndexH:$idx))>;
2980 def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
2981 (i64 (SMOVvi16to64 V128:$Rn, VectorIndexH:$idx))>;
2982 def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
2983 (i32 (SMOVvi16to32 V128:$Rn, VectorIndexH:$idx))>;
2984 def : Pat<(sext (i32 (vector_extract (v4i32 V128:$Rn), VectorIndexS:$idx))),
2985 (i64 (SMOVvi32to64 V128:$Rn, VectorIndexS:$idx))>;
2987 // Extracting i8 or i16 elements will have the zero-extend transformed to
2988 // an 'and' mask by type legalization since neither i8 nor i16 are legal types
2989 // for ARM64. Match these patterns here since UMOV already zeroes out the high
2990 // bits of the destination register.
2991 def : Pat<(and (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx),
2993 (i32 (UMOVvi8 V128:$Rn, VectorIndexB:$idx))>;
2994 def : Pat<(and (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),
2996 (i32 (UMOVvi16 V128:$Rn, VectorIndexH:$idx))>;
3000 def : Pat<(v16i8 (scalar_to_vector GPR32:$Rn)),
3001 (SUBREG_TO_REG (i32 0),
3002 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3003 def : Pat<(v8i8 (scalar_to_vector GPR32:$Rn)),
3004 (SUBREG_TO_REG (i32 0),
3005 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3007 def : Pat<(v8i16 (scalar_to_vector GPR32:$Rn)),
3008 (SUBREG_TO_REG (i32 0),
3009 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3010 def : Pat<(v4i16 (scalar_to_vector GPR32:$Rn)),
3011 (SUBREG_TO_REG (i32 0),
3012 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3014 def : Pat<(v2i32 (scalar_to_vector (i32 FPR32:$Rn))),
3015 (v2i32 (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)),
3016 (i32 FPR32:$Rn), ssub))>;
3017 def : Pat<(v4i32 (scalar_to_vector (i32 FPR32:$Rn))),
3018 (v4i32 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
3019 (i32 FPR32:$Rn), ssub))>;
3020 def : Pat<(v2i64 (scalar_to_vector (i64 FPR64:$Rn))),
3021 (v2i64 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
3022 (i64 FPR64:$Rn), dsub))>;
3024 def : Pat<(v4f32 (scalar_to_vector (f32 FPR32:$Rn))),
3025 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rn, ssub)>;
3026 def : Pat<(v2f32 (scalar_to_vector (f32 FPR32:$Rn))),
3027 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), FPR32:$Rn, ssub)>;
3028 def : Pat<(v2f64 (scalar_to_vector (f64 FPR64:$Rn))),
3029 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$Rn, dsub)>;
3031 def : Pat<(v2f32 (vector_insert (v2f32 V64:$Rn),
3032 (f32 FPR32:$Rm), (i64 VectorIndexS:$imm))),
3035 (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), V64:$Rn, dsub)),
3037 (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rm, ssub)),
3040 def : Pat<(v4f32 (vector_insert (v4f32 V128:$Rn),
3041 (f32 FPR32:$Rm), (i64 VectorIndexS:$imm))),
3043 V128:$Rn, VectorIndexS:$imm,
3044 (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rm, ssub)),
3046 def : Pat<(v2f64 (vector_insert (v2f64 V128:$Rn),
3047 (f64 FPR64:$Rm), (i64 VectorIndexD:$imm))),
3049 V128:$Rn, VectorIndexD:$imm,
3050 (v2f64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$Rm, dsub)),
3053 // Copy an element at a constant index in one vector into a constant indexed
3054 // element of another.
3055 // FIXME refactor to a shared class/dev parameterized on vector type, vector
3056 // index type and INS extension
3057 def : Pat<(v16i8 (int_arm64_neon_vcopy_lane
3058 (v16i8 V128:$Vd), VectorIndexB:$idx, (v16i8 V128:$Vs),
3059 VectorIndexB:$idx2)),
3061 V128:$Vd, VectorIndexB:$idx, V128:$Vs, VectorIndexB:$idx2)
3063 def : Pat<(v8i16 (int_arm64_neon_vcopy_lane
3064 (v8i16 V128:$Vd), VectorIndexH:$idx, (v8i16 V128:$Vs),
3065 VectorIndexH:$idx2)),
3067 V128:$Vd, VectorIndexH:$idx, V128:$Vs, VectorIndexH:$idx2)
3069 def : Pat<(v4i32 (int_arm64_neon_vcopy_lane
3070 (v4i32 V128:$Vd), VectorIndexS:$idx, (v4i32 V128:$Vs),
3071 VectorIndexS:$idx2)),
3073 V128:$Vd, VectorIndexS:$idx, V128:$Vs, VectorIndexS:$idx2)
3075 def : Pat<(v2i64 (int_arm64_neon_vcopy_lane
3076 (v2i64 V128:$Vd), VectorIndexD:$idx, (v2i64 V128:$Vs),
3077 VectorIndexD:$idx2)),
3079 V128:$Vd, VectorIndexD:$idx, V128:$Vs, VectorIndexD:$idx2)
3082 // Floating point vector extractions are codegen'd as either a sequence of
3083 // subregister extractions, possibly fed by an INS if the lane number is
3084 // anything other than zero.
3085 def : Pat<(vector_extract (v2f64 V128:$Rn), 0),
3086 (f64 (EXTRACT_SUBREG V128:$Rn, dsub))>;
3087 def : Pat<(vector_extract (v4f32 V128:$Rn), 0),
3088 (f32 (EXTRACT_SUBREG V128:$Rn, ssub))>;
3089 def : Pat<(vector_extract (v2f64 V128:$Rn), VectorIndexD:$idx),
3090 (f64 (EXTRACT_SUBREG
3091 (INSvi64lane (v2f64 (IMPLICIT_DEF)), 0,
3092 V128:$Rn, VectorIndexD:$idx),
3094 def : Pat<(vector_extract (v4f32 V128:$Rn), VectorIndexS:$idx),
3095 (f32 (EXTRACT_SUBREG
3096 (INSvi32lane (v4f32 (IMPLICIT_DEF)), 0,
3097 V128:$Rn, VectorIndexS:$idx),
3100 // All concat_vectors operations are canonicalised to act on i64 vectors for
3101 // ARM64. In the general case we need an instruction, which had just as well be
3103 class ConcatPat<ValueType DstTy, ValueType SrcTy>
3104 : Pat<(DstTy (concat_vectors (SrcTy V64:$Rd), V64:$Rn)),
3105 (INSvi64lane (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), 1,
3106 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rn, dsub), 0)>;
3108 def : ConcatPat<v2i64, v1i64>;
3109 def : ConcatPat<v2f64, v1f64>;
3110 def : ConcatPat<v4i32, v2i32>;
3111 def : ConcatPat<v4f32, v2f32>;
3112 def : ConcatPat<v8i16, v4i16>;
3113 def : ConcatPat<v16i8, v8i8>;
3115 // If the high lanes are undef, though, we can just ignore them:
3116 class ConcatUndefPat<ValueType DstTy, ValueType SrcTy>
3117 : Pat<(DstTy (concat_vectors (SrcTy V64:$Rn), undef)),
3118 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rn, dsub)>;
3120 def : ConcatUndefPat<v2i64, v1i64>;
3121 def : ConcatUndefPat<v2f64, v1f64>;
3122 def : ConcatUndefPat<v4i32, v2i32>;
3123 def : ConcatUndefPat<v4f32, v2f32>;
3124 def : ConcatUndefPat<v8i16, v4i16>;
3125 def : ConcatUndefPat<v16i8, v8i8>;
3127 //----------------------------------------------------------------------------
3128 // AdvSIMD across lanes instructions
3129 //----------------------------------------------------------------------------
3131 defm ADDV : SIMDAcrossLanesBHS<0, 0b11011, "addv">;
3132 defm SMAXV : SIMDAcrossLanesBHS<0, 0b01010, "smaxv">;
3133 defm SMINV : SIMDAcrossLanesBHS<0, 0b11010, "sminv">;
3134 defm UMAXV : SIMDAcrossLanesBHS<1, 0b01010, "umaxv">;
3135 defm UMINV : SIMDAcrossLanesBHS<1, 0b11010, "uminv">;
3136 defm SADDLV : SIMDAcrossLanesHSD<0, 0b00011, "saddlv">;
3137 defm UADDLV : SIMDAcrossLanesHSD<1, 0b00011, "uaddlv">;
3138 defm FMAXNMV : SIMDAcrossLanesS<0b01100, 0, "fmaxnmv", int_arm64_neon_fmaxnmv>;
3139 defm FMAXV : SIMDAcrossLanesS<0b01111, 0, "fmaxv", int_arm64_neon_fmaxv>;
3140 defm FMINNMV : SIMDAcrossLanesS<0b01100, 1, "fminnmv", int_arm64_neon_fminnmv>;
3141 defm FMINV : SIMDAcrossLanesS<0b01111, 1, "fminv", int_arm64_neon_fminv>;
3143 multiclass SIMDAcrossLanesSignedIntrinsic<string baseOpc, Intrinsic intOp> {
3144 // If there is a sign extension after this intrinsic, consume it as smov already
3146 def : Pat<(i32 (sext_inreg (i32 (intOp (v8i8 V64:$Rn))), i8)),
3148 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3149 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3151 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
3153 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3154 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3156 // If there is a sign extension after this intrinsic, consume it as smov already
3158 def : Pat<(i32 (sext_inreg (i32 (intOp (v16i8 V128:$Rn))), i8)),
3160 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3161 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3163 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
3165 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3166 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3168 // If there is a sign extension after this intrinsic, consume it as smov already
3170 def : Pat<(i32 (sext_inreg (i32 (intOp (v4i16 V64:$Rn))), i16)),
3172 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3173 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3175 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
3177 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3178 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3180 // If there is a sign extension after this intrinsic, consume it as smov already
3182 def : Pat<(i32 (sext_inreg (i32 (intOp (v8i16 V128:$Rn))), i16)),
3184 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3185 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3187 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
3189 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3190 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3193 def : Pat<(i32 (intOp (v4i32 V128:$Rn))),
3194 (i32 (EXTRACT_SUBREG
3195 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3196 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), ssub),
3200 multiclass SIMDAcrossLanesUnsignedIntrinsic<string baseOpc, Intrinsic intOp> {
3201 // If there is a masking operation keeping only what has been actually
3202 // generated, consume it.
3203 def : Pat<(i32 (and (i32 (intOp (v8i8 V64:$Rn))), maski8_or_more)),
3204 (i32 (EXTRACT_SUBREG
3205 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3206 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3208 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
3209 (i32 (EXTRACT_SUBREG
3210 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3211 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3213 // If there is a masking operation keeping only what has been actually
3214 // generated, consume it.
3215 def : Pat<(i32 (and (i32 (intOp (v16i8 V128:$Rn))), maski8_or_more)),
3216 (i32 (EXTRACT_SUBREG
3217 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3218 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3220 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
3221 (i32 (EXTRACT_SUBREG
3222 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3223 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3226 // If there is a masking operation keeping only what has been actually
3227 // generated, consume it.
3228 def : Pat<(i32 (and (i32 (intOp (v4i16 V64:$Rn))), maski16_or_more)),
3229 (i32 (EXTRACT_SUBREG
3230 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3231 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3233 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
3234 (i32 (EXTRACT_SUBREG
3235 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3236 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3238 // If there is a masking operation keeping only what has been actually
3239 // generated, consume it.
3240 def : Pat<(i32 (and (i32 (intOp (v8i16 V128:$Rn))), maski16_or_more)),
3241 (i32 (EXTRACT_SUBREG
3242 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3243 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3245 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
3246 (i32 (EXTRACT_SUBREG
3247 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3248 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3251 def : Pat<(i32 (intOp (v4i32 V128:$Rn))),
3252 (i32 (EXTRACT_SUBREG
3253 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3254 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), ssub),
3259 multiclass SIMDAcrossLanesSignedLongIntrinsic<string baseOpc, Intrinsic intOp> {
3260 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
3262 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3263 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), hsub),
3265 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
3267 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3268 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), hsub),
3271 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
3272 (i32 (EXTRACT_SUBREG
3273 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3274 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), ssub),
3276 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
3277 (i32 (EXTRACT_SUBREG
3278 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3279 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), ssub),
3282 def : Pat<(i64 (intOp (v4i32 V128:$Rn))),
3283 (i64 (EXTRACT_SUBREG
3284 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3285 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), dsub),
3289 multiclass SIMDAcrossLanesUnsignedLongIntrinsic<string baseOpc,
3291 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
3292 (i32 (EXTRACT_SUBREG
3293 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3294 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), hsub),
3296 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
3297 (i32 (EXTRACT_SUBREG
3298 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3299 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), hsub),
3302 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
3303 (i32 (EXTRACT_SUBREG
3304 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3305 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), ssub),
3307 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
3308 (i32 (EXTRACT_SUBREG
3309 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3310 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), ssub),
3313 def : Pat<(i64 (intOp (v4i32 V128:$Rn))),
3314 (i64 (EXTRACT_SUBREG
3315 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3316 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), dsub),
3320 defm : SIMDAcrossLanesSignedIntrinsic<"ADDV", int_arm64_neon_saddv>;
3321 // vaddv_[su]32 is special; -> ADDP Vd.2S,Vn.2S,Vm.2S; return Vd.s[0];Vn==Vm
3322 def : Pat<(i32 (int_arm64_neon_saddv (v2i32 V64:$Rn))),
3323 (EXTRACT_SUBREG (ADDPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3325 defm : SIMDAcrossLanesUnsignedIntrinsic<"ADDV", int_arm64_neon_uaddv>;
3326 // vaddv_[su]32 is special; -> ADDP Vd.2S,Vn.2S,Vm.2S; return Vd.s[0];Vn==Vm
3327 def : Pat<(i32 (int_arm64_neon_uaddv (v2i32 V64:$Rn))),
3328 (EXTRACT_SUBREG (ADDPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3330 defm : SIMDAcrossLanesSignedIntrinsic<"SMAXV", int_arm64_neon_smaxv>;
3331 def : Pat<(i32 (int_arm64_neon_smaxv (v2i32 V64:$Rn))),
3332 (EXTRACT_SUBREG (SMAXPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3334 defm : SIMDAcrossLanesSignedIntrinsic<"SMINV", int_arm64_neon_sminv>;
3335 def : Pat<(i32 (int_arm64_neon_sminv (v2i32 V64:$Rn))),
3336 (EXTRACT_SUBREG (SMINPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3338 defm : SIMDAcrossLanesUnsignedIntrinsic<"UMAXV", int_arm64_neon_umaxv>;
3339 def : Pat<(i32 (int_arm64_neon_umaxv (v2i32 V64:$Rn))),
3340 (EXTRACT_SUBREG (UMAXPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3342 defm : SIMDAcrossLanesUnsignedIntrinsic<"UMINV", int_arm64_neon_uminv>;
3343 def : Pat<(i32 (int_arm64_neon_uminv (v2i32 V64:$Rn))),
3344 (EXTRACT_SUBREG (UMINPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3346 defm : SIMDAcrossLanesSignedLongIntrinsic<"SADDLV", int_arm64_neon_saddlv>;
3347 defm : SIMDAcrossLanesUnsignedLongIntrinsic<"UADDLV", int_arm64_neon_uaddlv>;
3349 // The vaddlv_s32 intrinsic gets mapped to SADDLP.
3350 def : Pat<(i64 (int_arm64_neon_saddlv (v2i32 V64:$Rn))),
3351 (i64 (EXTRACT_SUBREG
3352 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3353 (SADDLPv2i32_v1i64 V64:$Rn), dsub),
3355 // The vaddlv_u32 intrinsic gets mapped to UADDLP.
3356 def : Pat<(i64 (int_arm64_neon_uaddlv (v2i32 V64:$Rn))),
3357 (i64 (EXTRACT_SUBREG
3358 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3359 (UADDLPv2i32_v1i64 V64:$Rn), dsub),
3362 //------------------------------------------------------------------------------
3363 // AdvSIMD modified immediate instructions
3364 //------------------------------------------------------------------------------
3367 defm BIC : SIMDModifiedImmVectorShiftTied<1, 0b11, 0b01, "bic", ARM64bici>;
3369 defm ORR : SIMDModifiedImmVectorShiftTied<0, 0b11, 0b01, "orr", ARM64orri>;
3373 def FMOVv2f64_ns : SIMDModifiedImmVectorNoShift<1, 1, 0b1111, V128, fpimm8,
3375 [(set (v2f64 V128:$Rd), (ARM64fmov imm0_255:$imm8))]>;
3376 def FMOVv2f32_ns : SIMDModifiedImmVectorNoShift<0, 0, 0b1111, V64, fpimm8,
3378 [(set (v2f32 V64:$Rd), (ARM64fmov imm0_255:$imm8))]>;
3379 def FMOVv4f32_ns : SIMDModifiedImmVectorNoShift<1, 0, 0b1111, V128, fpimm8,
3381 [(set (v4f32 V128:$Rd), (ARM64fmov imm0_255:$imm8))]>;
3385 // EDIT byte mask: scalar
3386 let isReMaterializable = 1, isAsCheapAsAMove = 1 in
3387 def MOVID : SIMDModifiedImmScalarNoShift<0, 1, 0b1110, "movi",
3388 [(set FPR64:$Rd, simdimmtype10:$imm8)]>;
3389 // The movi_edit node has the immediate value already encoded, so we use
3390 // a plain imm0_255 here.
3391 def : Pat<(f64 (ARM64movi_edit imm0_255:$shift)),
3392 (MOVID imm0_255:$shift)>;
3394 def : Pat<(v1i64 immAllZerosV), (MOVID (i32 0))>;
3395 def : Pat<(v2i32 immAllZerosV), (MOVID (i32 0))>;
3396 def : Pat<(v4i16 immAllZerosV), (MOVID (i32 0))>;
3397 def : Pat<(v8i8 immAllZerosV), (MOVID (i32 0))>;
3399 def : Pat<(v1i64 immAllOnesV), (MOVID (i32 255))>;
3400 def : Pat<(v2i32 immAllOnesV), (MOVID (i32 255))>;
3401 def : Pat<(v4i16 immAllOnesV), (MOVID (i32 255))>;
3402 def : Pat<(v8i8 immAllOnesV), (MOVID (i32 255))>;
3404 // EDIT byte mask: 2d
3406 // The movi_edit node has the immediate value already encoded, so we use
3407 // a plain imm0_255 in the pattern
3408 let isReMaterializable = 1, isAsCheapAsAMove = 1 in
3409 def MOVIv2d_ns : SIMDModifiedImmVectorNoShift<1, 1, 0b1110, V128,
3412 [(set (v2i64 V128:$Rd), (ARM64movi_edit imm0_255:$imm8))]>;
3415 // Use movi.2d to materialize 0.0 if the HW does zero-cycle zeroing.
3416 // Complexity is added to break a tie with a plain MOVI.
3417 let AddedComplexity = 1 in {
3418 def : Pat<(f32 fpimm0),
3419 (f32 (EXTRACT_SUBREG (v2i64 (MOVIv2d_ns (i32 0))), ssub))>,
3421 def : Pat<(f64 fpimm0),
3422 (f64 (EXTRACT_SUBREG (v2i64 (MOVIv2d_ns (i32 0))), dsub))>,
3426 def : Pat<(v2i64 immAllZerosV), (MOVIv2d_ns (i32 0))>;
3427 def : Pat<(v4i32 immAllZerosV), (MOVIv2d_ns (i32 0))>;
3428 def : Pat<(v8i16 immAllZerosV), (MOVIv2d_ns (i32 0))>;
3429 def : Pat<(v16i8 immAllZerosV), (MOVIv2d_ns (i32 0))>;
3431 def : Pat<(v2i64 immAllOnesV), (MOVIv2d_ns (i32 255))>;
3432 def : Pat<(v4i32 immAllOnesV), (MOVIv2d_ns (i32 255))>;
3433 def : Pat<(v8i16 immAllOnesV), (MOVIv2d_ns (i32 255))>;
3434 def : Pat<(v16i8 immAllOnesV), (MOVIv2d_ns (i32 255))>;
3436 // EDIT per word & halfword: 2s, 4h, 4s, & 8h
3437 defm MOVI : SIMDModifiedImmVectorShift<0, 0b10, 0b00, "movi">;
3438 def : Pat<(v2i32 (ARM64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
3439 (MOVIv2i32 imm0_255:$imm8, imm:$shift)>;
3440 def : Pat<(v4i32 (ARM64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
3441 (MOVIv4i32 imm0_255:$imm8, imm:$shift)>;
3442 def : Pat<(v4i16 (ARM64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
3443 (MOVIv4i16 imm0_255:$imm8, imm:$shift)>;
3444 def : Pat<(v8i16 (ARM64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
3445 (MOVIv8i16 imm0_255:$imm8, imm:$shift)>;
3447 // EDIT per word: 2s & 4s with MSL shifter
3448 def MOVIv2s_msl : SIMDModifiedImmMoveMSL<0, 0, {1,1,0,?}, V64, "movi", ".2s",
3449 [(set (v2i32 V64:$Rd),
3450 (ARM64movi_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
3451 def MOVIv4s_msl : SIMDModifiedImmMoveMSL<1, 0, {1,1,0,?}, V128, "movi", ".4s",
3452 [(set (v4i32 V128:$Rd),
3453 (ARM64movi_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
3455 // Per byte: 8b & 16b
3456 def MOVIv8b_ns : SIMDModifiedImmVectorNoShift<0, 0, 0b1110, V64, imm0_255,
3458 [(set (v8i8 V64:$Rd), (ARM64movi imm0_255:$imm8))]>;
3459 def MOVIv16b_ns : SIMDModifiedImmVectorNoShift<1, 0, 0b1110, V128, imm0_255,
3461 [(set (v16i8 V128:$Rd), (ARM64movi imm0_255:$imm8))]>;
3465 // EDIT per word & halfword: 2s, 4h, 4s, & 8h
3466 defm MVNI : SIMDModifiedImmVectorShift<1, 0b10, 0b00, "mvni">;
3467 def : Pat<(v2i32 (ARM64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
3468 (MVNIv2i32 imm0_255:$imm8, imm:$shift)>;
3469 def : Pat<(v4i32 (ARM64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
3470 (MVNIv4i32 imm0_255:$imm8, imm:$shift)>;
3471 def : Pat<(v4i16 (ARM64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
3472 (MVNIv4i16 imm0_255:$imm8, imm:$shift)>;
3473 def : Pat<(v8i16 (ARM64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
3474 (MVNIv8i16 imm0_255:$imm8, imm:$shift)>;
3476 // EDIT per word: 2s & 4s with MSL shifter
3477 def MVNIv2s_msl : SIMDModifiedImmMoveMSL<0, 1, {1,1,0,?}, V64, "mvni", ".2s",
3478 [(set (v2i32 V64:$Rd),
3479 (ARM64mvni_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
3480 def MVNIv4s_msl : SIMDModifiedImmMoveMSL<1, 1, {1,1,0,?}, V128, "mvni", ".4s",
3481 [(set (v4i32 V128:$Rd),
3482 (ARM64mvni_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
3484 //----------------------------------------------------------------------------
3485 // AdvSIMD indexed element
3486 //----------------------------------------------------------------------------
3488 let neverHasSideEffects = 1 in {
3489 defm FMLA : SIMDFPIndexedSDTied<0, 0b0001, "fmla">;
3490 defm FMLS : SIMDFPIndexedSDTied<0, 0b0101, "fmls">;
3493 // NOTE: Operands are reordered in the FMLA/FMLS PatFrags because the
3494 // instruction expects the addend first, while the intrinsic expects it last.
3496 // On the other hand, there are quite a few valid combinatorial options due to
3497 // the commutativity of multiplication and the fact that (-x) * y = x * (-y).
3498 defm : SIMDFPIndexedSDTiedPatterns<"FMLA",
3499 TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)>>;
3500 defm : SIMDFPIndexedSDTiedPatterns<"FMLA",
3501 TriOpFrag<(fma node:$MHS, node:$RHS, node:$LHS)>>;
3503 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
3504 TriOpFrag<(fma node:$MHS, (fneg node:$RHS), node:$LHS)> >;
3505 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
3506 TriOpFrag<(fma node:$RHS, (fneg node:$MHS), node:$LHS)> >;
3507 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
3508 TriOpFrag<(fma (fneg node:$RHS), node:$MHS, node:$LHS)> >;
3509 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
3510 TriOpFrag<(fma (fneg node:$MHS), node:$RHS, node:$LHS)> >;
3512 multiclass FMLSIndexedAfterNegPatterns<SDPatternOperator OpNode> {
3513 // 3 variants for the .2s version: DUPLANE from 128-bit, DUPLANE from 64-bit
3515 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
3516 (ARM64duplane32 (v4f32 (fneg V128:$Rm)),
3517 VectorIndexS:$idx))),
3518 (FMLSv2i32_indexed V64:$Rd, V64:$Rn, V128:$Rm, VectorIndexS:$idx)>;
3519 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
3520 (v2f32 (ARM64duplane32
3521 (v4f32 (insert_subvector undef,
3522 (v2f32 (fneg V64:$Rm)),
3524 VectorIndexS:$idx)))),
3525 (FMLSv2i32_indexed V64:$Rd, V64:$Rn,
3526 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub),
3527 VectorIndexS:$idx)>;
3528 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
3529 (ARM64dup (f32 (fneg FPR32Op:$Rm))))),
3530 (FMLSv2i32_indexed V64:$Rd, V64:$Rn,
3531 (SUBREG_TO_REG (i32 0), FPR32Op:$Rm, ssub), (i64 0))>;
3533 // 3 variants for the .4s version: DUPLANE from 128-bit, DUPLANE from 64-bit
3535 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
3536 (ARM64duplane32 (v4f32 (fneg V128:$Rm)),
3537 VectorIndexS:$idx))),
3538 (FMLSv4i32_indexed V128:$Rd, V128:$Rn, V128:$Rm,
3539 VectorIndexS:$idx)>;
3540 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
3541 (v4f32 (ARM64duplane32
3542 (v4f32 (insert_subvector undef,
3543 (v2f32 (fneg V64:$Rm)),
3545 VectorIndexS:$idx)))),
3546 (FMLSv4i32_indexed V128:$Rd, V128:$Rn,
3547 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub),
3548 VectorIndexS:$idx)>;
3549 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
3550 (ARM64dup (f32 (fneg FPR32Op:$Rm))))),
3551 (FMLSv4i32_indexed V128:$Rd, V128:$Rn,
3552 (SUBREG_TO_REG (i32 0), FPR32Op:$Rm, ssub), (i64 0))>;
3554 // 2 variants for the .2d version: DUPLANE from 128-bit, and DUP scalar
3555 // (DUPLANE from 64-bit would be trivial).
3556 def : Pat<(v2f64 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn),
3557 (ARM64duplane64 (v2f64 (fneg V128:$Rm)),
3558 VectorIndexD:$idx))),
3560 V128:$Rd, V128:$Rn, V128:$Rm, VectorIndexS:$idx)>;
3561 def : Pat<(v2f64 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn),
3562 (ARM64dup (f64 (fneg FPR64Op:$Rm))))),
3563 (FMLSv2i64_indexed V128:$Rd, V128:$Rn,
3564 (SUBREG_TO_REG (i32 0), FPR64Op:$Rm, dsub), (i64 0))>;
3566 // 2 variants for 32-bit scalar version: extract from .2s or from .4s
3567 def : Pat<(f32 (OpNode (f32 FPR32:$Rd), (f32 FPR32:$Rn),
3568 (vector_extract (v4f32 (fneg V128:$Rm)),
3569 VectorIndexS:$idx))),
3570 (FMLSv1i32_indexed FPR32:$Rd, FPR32:$Rn,
3571 V128:$Rm, VectorIndexS:$idx)>;
3572 def : Pat<(f32 (OpNode (f32 FPR32:$Rd), (f32 FPR32:$Rn),
3573 (vector_extract (v2f32 (fneg V64:$Rm)),
3574 VectorIndexS:$idx))),
3575 (FMLSv1i32_indexed FPR32:$Rd, FPR32:$Rn,
3576 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub), VectorIndexS:$idx)>;
3578 // 1 variant for 64-bit scalar version: extract from .1d or from .2d
3579 def : Pat<(f64 (OpNode (f64 FPR64:$Rd), (f64 FPR64:$Rn),
3580 (vector_extract (v2f64 (fneg V128:$Rm)),
3581 VectorIndexS:$idx))),
3582 (FMLSv1i64_indexed FPR64:$Rd, FPR64:$Rn,
3583 V128:$Rm, VectorIndexS:$idx)>;
3586 defm : FMLSIndexedAfterNegPatterns<
3587 TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)> >;
3588 defm : FMLSIndexedAfterNegPatterns<
3589 TriOpFrag<(fma node:$MHS, node:$RHS, node:$LHS)> >;
3591 defm FMULX : SIMDFPIndexedSD<1, 0b1001, "fmulx", int_arm64_neon_fmulx>;
3592 defm FMUL : SIMDFPIndexedSD<0, 0b1001, "fmul", fmul>;
3594 def : Pat<(v2f32 (fmul V64:$Rn, (ARM64dup (f32 FPR32:$Rm)))),
3595 (FMULv2i32_indexed V64:$Rn,
3596 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rm, ssub),
3598 def : Pat<(v4f32 (fmul V128:$Rn, (ARM64dup (f32 FPR32:$Rm)))),
3599 (FMULv4i32_indexed V128:$Rn,
3600 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rm, ssub),
3602 def : Pat<(v2f64 (fmul V128:$Rn, (ARM64dup (f64 FPR64:$Rm)))),
3603 (FMULv2i64_indexed V128:$Rn,
3604 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$Rm, dsub),
3607 defm SQDMULH : SIMDIndexedHS<0, 0b1100, "sqdmulh", int_arm64_neon_sqdmulh>;
3608 defm SQRDMULH : SIMDIndexedHS<0, 0b1101, "sqrdmulh", int_arm64_neon_sqrdmulh>;
3609 defm MLA : SIMDVectorIndexedHSTied<1, 0b0000, "mla",
3610 TriOpFrag<(add node:$LHS, (mul node:$MHS, node:$RHS))>>;
3611 defm MLS : SIMDVectorIndexedHSTied<1, 0b0100, "mls",
3612 TriOpFrag<(sub node:$LHS, (mul node:$MHS, node:$RHS))>>;
3613 defm MUL : SIMDVectorIndexedHS<0, 0b1000, "mul", mul>;
3614 defm SMLAL : SIMDVectorIndexedLongSDTied<0, 0b0010, "smlal",
3615 TriOpFrag<(add node:$LHS, (int_arm64_neon_smull node:$MHS, node:$RHS))>>;
3616 defm SMLSL : SIMDVectorIndexedLongSDTied<0, 0b0110, "smlsl",
3617 TriOpFrag<(sub node:$LHS, (int_arm64_neon_smull node:$MHS, node:$RHS))>>;
3618 defm SMULL : SIMDVectorIndexedLongSD<0, 0b1010, "smull",
3619 int_arm64_neon_smull>;
3620 defm SQDMLAL : SIMDIndexedLongSQDMLXSDTied<0, 0b0011, "sqdmlal",
3621 int_arm64_neon_sqadd>;
3622 defm SQDMLSL : SIMDIndexedLongSQDMLXSDTied<0, 0b0111, "sqdmlsl",
3623 int_arm64_neon_sqsub>;
3624 defm SQDMULL : SIMDIndexedLongSD<0, 0b1011, "sqdmull", int_arm64_neon_sqdmull>;
3625 defm UMLAL : SIMDVectorIndexedLongSDTied<1, 0b0010, "umlal",
3626 TriOpFrag<(add node:$LHS, (int_arm64_neon_umull node:$MHS, node:$RHS))>>;
3627 defm UMLSL : SIMDVectorIndexedLongSDTied<1, 0b0110, "umlsl",
3628 TriOpFrag<(sub node:$LHS, (int_arm64_neon_umull node:$MHS, node:$RHS))>>;
3629 defm UMULL : SIMDVectorIndexedLongSD<1, 0b1010, "umull",
3630 int_arm64_neon_umull>;
3632 // A scalar sqdmull with the second operand being a vector lane can be
3633 // handled directly with the indexed instruction encoding.
3634 def : Pat<(int_arm64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
3635 (vector_extract (v4i32 V128:$Vm),
3636 VectorIndexS:$idx)),
3637 (SQDMULLv1i64_indexed FPR32:$Rn, V128:$Vm, VectorIndexS:$idx)>;
3639 //----------------------------------------------------------------------------
3640 // AdvSIMD scalar shift instructions
3641 //----------------------------------------------------------------------------
3642 defm FCVTZS : SIMDScalarRShiftSD<0, 0b11111, "fcvtzs">;
3643 defm FCVTZU : SIMDScalarRShiftSD<1, 0b11111, "fcvtzu">;
3644 defm SCVTF : SIMDScalarRShiftSD<0, 0b11100, "scvtf">;
3645 defm UCVTF : SIMDScalarRShiftSD<1, 0b11100, "ucvtf">;
3646 // Codegen patterns for the above. We don't put these directly on the
3647 // instructions because TableGen's type inference can't handle the truth.
3648 // Having the same base pattern for fp <--> int totally freaks it out.
3649 def : Pat<(int_arm64_neon_vcvtfp2fxs FPR32:$Rn, vecshiftR32:$imm),
3650 (FCVTZSs FPR32:$Rn, vecshiftR32:$imm)>;
3651 def : Pat<(int_arm64_neon_vcvtfp2fxu FPR32:$Rn, vecshiftR32:$imm),
3652 (FCVTZUs FPR32:$Rn, vecshiftR32:$imm)>;
3653 def : Pat<(i64 (int_arm64_neon_vcvtfp2fxs (f64 FPR64:$Rn), vecshiftR64:$imm)),
3654 (FCVTZSd FPR64:$Rn, vecshiftR64:$imm)>;
3655 def : Pat<(i64 (int_arm64_neon_vcvtfp2fxu (f64 FPR64:$Rn), vecshiftR64:$imm)),
3656 (FCVTZUd FPR64:$Rn, vecshiftR64:$imm)>;
3657 def : Pat<(v1i64 (int_arm64_neon_vcvtfp2fxs (v1f64 FPR64:$Rn),
3659 (FCVTZSd FPR64:$Rn, vecshiftR64:$imm)>;
3660 def : Pat<(v1i64 (int_arm64_neon_vcvtfp2fxu (v1f64 FPR64:$Rn),
3662 (FCVTZUd FPR64:$Rn, vecshiftR64:$imm)>;
3663 def : Pat<(int_arm64_neon_vcvtfxs2fp FPR32:$Rn, vecshiftR32:$imm),
3664 (SCVTFs FPR32:$Rn, vecshiftR32:$imm)>;
3665 def : Pat<(int_arm64_neon_vcvtfxu2fp FPR32:$Rn, vecshiftR32:$imm),
3666 (UCVTFs FPR32:$Rn, vecshiftR32:$imm)>;
3667 def : Pat<(f64 (int_arm64_neon_vcvtfxs2fp (i64 FPR64:$Rn), vecshiftR64:$imm)),
3668 (SCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
3669 def : Pat<(f64 (int_arm64_neon_vcvtfxu2fp (i64 FPR64:$Rn), vecshiftR64:$imm)),
3670 (UCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
3671 def : Pat<(v1f64 (int_arm64_neon_vcvtfxs2fp (v1i64 FPR64:$Rn),
3673 (SCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
3674 def : Pat<(v1f64 (int_arm64_neon_vcvtfxu2fp (v1i64 FPR64:$Rn),
3676 (UCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
3678 defm SHL : SIMDScalarLShiftD< 0, 0b01010, "shl", ARM64vshl>;
3679 defm SLI : SIMDScalarLShiftDTied<1, 0b01010, "sli">;
3680 defm SQRSHRN : SIMDScalarRShiftBHS< 0, 0b10011, "sqrshrn",
3681 int_arm64_neon_sqrshrn>;
3682 defm SQRSHRUN : SIMDScalarRShiftBHS< 1, 0b10001, "sqrshrun",
3683 int_arm64_neon_sqrshrun>;
3684 defm SQSHLU : SIMDScalarLShiftBHSD<1, 0b01100, "sqshlu", ARM64sqshlui>;
3685 defm SQSHL : SIMDScalarLShiftBHSD<0, 0b01110, "sqshl", ARM64sqshli>;
3686 defm SQSHRN : SIMDScalarRShiftBHS< 0, 0b10010, "sqshrn",
3687 int_arm64_neon_sqshrn>;
3688 defm SQSHRUN : SIMDScalarRShiftBHS< 1, 0b10000, "sqshrun",
3689 int_arm64_neon_sqshrun>;
3690 defm SRI : SIMDScalarRShiftDTied< 1, 0b01000, "sri">;
3691 defm SRSHR : SIMDScalarRShiftD< 0, 0b00100, "srshr", ARM64srshri>;
3692 defm SRSRA : SIMDScalarRShiftDTied< 0, 0b00110, "srsra",
3693 TriOpFrag<(add node:$LHS,
3694 (ARM64srshri node:$MHS, node:$RHS))>>;
3695 defm SSHR : SIMDScalarRShiftD< 0, 0b00000, "sshr", ARM64vashr>;
3696 defm SSRA : SIMDScalarRShiftDTied< 0, 0b00010, "ssra",
3697 TriOpFrag<(add node:$LHS,
3698 (ARM64vashr node:$MHS, node:$RHS))>>;
3699 defm UQRSHRN : SIMDScalarRShiftBHS< 1, 0b10011, "uqrshrn",
3700 int_arm64_neon_uqrshrn>;
3701 defm UQSHL : SIMDScalarLShiftBHSD<1, 0b01110, "uqshl", ARM64uqshli>;
3702 defm UQSHRN : SIMDScalarRShiftBHS< 1, 0b10010, "uqshrn",
3703 int_arm64_neon_uqshrn>;
3704 defm URSHR : SIMDScalarRShiftD< 1, 0b00100, "urshr", ARM64urshri>;
3705 defm URSRA : SIMDScalarRShiftDTied< 1, 0b00110, "ursra",
3706 TriOpFrag<(add node:$LHS,
3707 (ARM64urshri node:$MHS, node:$RHS))>>;
3708 defm USHR : SIMDScalarRShiftD< 1, 0b00000, "ushr", ARM64vlshr>;
3709 defm USRA : SIMDScalarRShiftDTied< 1, 0b00010, "usra",
3710 TriOpFrag<(add node:$LHS,
3711 (ARM64vlshr node:$MHS, node:$RHS))>>;
3713 //----------------------------------------------------------------------------
3714 // AdvSIMD vector shift instructions
3715 //----------------------------------------------------------------------------
3716 defm FCVTZS:SIMDVectorRShiftSD<0, 0b11111, "fcvtzs", int_arm64_neon_vcvtfp2fxs>;
3717 defm FCVTZU:SIMDVectorRShiftSD<1, 0b11111, "fcvtzu", int_arm64_neon_vcvtfp2fxu>;
3718 defm SCVTF: SIMDVectorRShiftSDToFP<0, 0b11100, "scvtf",
3719 int_arm64_neon_vcvtfxs2fp>;
3720 defm RSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10001, "rshrn",
3721 int_arm64_neon_rshrn>;
3722 defm SHL : SIMDVectorLShiftBHSD<0, 0b01010, "shl", ARM64vshl>;
3723 defm SHRN : SIMDVectorRShiftNarrowBHS<0, 0b10000, "shrn",
3724 BinOpFrag<(trunc (ARM64vashr node:$LHS, node:$RHS))>>;
3725 defm SLI : SIMDVectorLShiftBHSDTied<1, 0b01010, "sli", int_arm64_neon_vsli>;
3726 def : Pat<(v1i64 (int_arm64_neon_vsli (v1i64 FPR64:$Rd), (v1i64 FPR64:$Rn),
3727 (i32 vecshiftL64:$imm))),
3728 (SLId FPR64:$Rd, FPR64:$Rn, vecshiftL64:$imm)>;
3729 defm SQRSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10011, "sqrshrn",
3730 int_arm64_neon_sqrshrn>;
3731 defm SQRSHRUN: SIMDVectorRShiftNarrowBHS<1, 0b10001, "sqrshrun",
3732 int_arm64_neon_sqrshrun>;
3733 defm SQSHLU : SIMDVectorLShiftBHSD<1, 0b01100, "sqshlu", ARM64sqshlui>;
3734 defm SQSHL : SIMDVectorLShiftBHSD<0, 0b01110, "sqshl", ARM64sqshli>;
3735 defm SQSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10010, "sqshrn",
3736 int_arm64_neon_sqshrn>;
3737 defm SQSHRUN : SIMDVectorRShiftNarrowBHS<1, 0b10000, "sqshrun",
3738 int_arm64_neon_sqshrun>;
3739 defm SRI : SIMDVectorRShiftBHSDTied<1, 0b01000, "sri", int_arm64_neon_vsri>;
3740 def : Pat<(v1i64 (int_arm64_neon_vsri (v1i64 FPR64:$Rd), (v1i64 FPR64:$Rn),
3741 (i32 vecshiftR64:$imm))),
3742 (SRId FPR64:$Rd, FPR64:$Rn, vecshiftR64:$imm)>;
3743 defm SRSHR : SIMDVectorRShiftBHSD<0, 0b00100, "srshr", ARM64srshri>;
3744 defm SRSRA : SIMDVectorRShiftBHSDTied<0, 0b00110, "srsra",
3745 TriOpFrag<(add node:$LHS,
3746 (ARM64srshri node:$MHS, node:$RHS))> >;
3747 defm SSHLL : SIMDVectorLShiftLongBHSD<0, 0b10100, "sshll",
3748 BinOpFrag<(ARM64vshl (sext node:$LHS), node:$RHS)>>;
3750 defm SSHR : SIMDVectorRShiftBHSD<0, 0b00000, "sshr", ARM64vashr>;
3751 defm SSRA : SIMDVectorRShiftBHSDTied<0, 0b00010, "ssra",
3752 TriOpFrag<(add node:$LHS, (ARM64vashr node:$MHS, node:$RHS))>>;
3753 defm UCVTF : SIMDVectorRShiftSDToFP<1, 0b11100, "ucvtf",
3754 int_arm64_neon_vcvtfxu2fp>;
3755 defm UQRSHRN : SIMDVectorRShiftNarrowBHS<1, 0b10011, "uqrshrn",
3756 int_arm64_neon_uqrshrn>;
3757 defm UQSHL : SIMDVectorLShiftBHSD<1, 0b01110, "uqshl", ARM64uqshli>;
3758 defm UQSHRN : SIMDVectorRShiftNarrowBHS<1, 0b10010, "uqshrn",
3759 int_arm64_neon_uqshrn>;
3760 defm URSHR : SIMDVectorRShiftBHSD<1, 0b00100, "urshr", ARM64urshri>;
3761 defm URSRA : SIMDVectorRShiftBHSDTied<1, 0b00110, "ursra",
3762 TriOpFrag<(add node:$LHS,
3763 (ARM64urshri node:$MHS, node:$RHS))> >;
3764 defm USHLL : SIMDVectorLShiftLongBHSD<1, 0b10100, "ushll",
3765 BinOpFrag<(ARM64vshl (zext node:$LHS), node:$RHS)>>;
3766 defm USHR : SIMDVectorRShiftBHSD<1, 0b00000, "ushr", ARM64vlshr>;
3767 defm USRA : SIMDVectorRShiftBHSDTied<1, 0b00010, "usra",
3768 TriOpFrag<(add node:$LHS, (ARM64vlshr node:$MHS, node:$RHS))> >;
3770 // SHRN patterns for when a logical right shift was used instead of arithmetic
3771 // (the immediate guarantees no sign bits actually end up in the result so it
3773 def : Pat<(v8i8 (trunc (ARM64vlshr (v8i16 V128:$Rn), vecshiftR16Narrow:$imm))),
3774 (SHRNv8i8_shift V128:$Rn, vecshiftR16Narrow:$imm)>;
3775 def : Pat<(v4i16 (trunc (ARM64vlshr (v4i32 V128:$Rn), vecshiftR32Narrow:$imm))),
3776 (SHRNv4i16_shift V128:$Rn, vecshiftR32Narrow:$imm)>;
3777 def : Pat<(v2i32 (trunc (ARM64vlshr (v2i64 V128:$Rn), vecshiftR64Narrow:$imm))),
3778 (SHRNv2i32_shift V128:$Rn, vecshiftR64Narrow:$imm)>;
3780 def : Pat<(v16i8 (concat_vectors (v8i8 V64:$Rd),
3781 (trunc (ARM64vlshr (v8i16 V128:$Rn),
3782 vecshiftR16Narrow:$imm)))),
3783 (SHRNv16i8_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
3784 V128:$Rn, vecshiftR16Narrow:$imm)>;
3785 def : Pat<(v8i16 (concat_vectors (v4i16 V64:$Rd),
3786 (trunc (ARM64vlshr (v4i32 V128:$Rn),
3787 vecshiftR32Narrow:$imm)))),
3788 (SHRNv8i16_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
3789 V128:$Rn, vecshiftR32Narrow:$imm)>;
3790 def : Pat<(v4i32 (concat_vectors (v2i32 V64:$Rd),
3791 (trunc (ARM64vlshr (v2i64 V128:$Rn),
3792 vecshiftR64Narrow:$imm)))),
3793 (SHRNv4i32_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
3794 V128:$Rn, vecshiftR32Narrow:$imm)>;
3796 // Vector sign and zero extensions are implemented with SSHLL and USSHLL.
3797 // Anyexts are implemented as zexts.
3798 def : Pat<(v8i16 (sext (v8i8 V64:$Rn))), (SSHLLv8i8_shift V64:$Rn, (i32 0))>;
3799 def : Pat<(v8i16 (zext (v8i8 V64:$Rn))), (USHLLv8i8_shift V64:$Rn, (i32 0))>;
3800 def : Pat<(v8i16 (anyext (v8i8 V64:$Rn))), (USHLLv8i8_shift V64:$Rn, (i32 0))>;
3801 def : Pat<(v4i32 (sext (v4i16 V64:$Rn))), (SSHLLv4i16_shift V64:$Rn, (i32 0))>;
3802 def : Pat<(v4i32 (zext (v4i16 V64:$Rn))), (USHLLv4i16_shift V64:$Rn, (i32 0))>;
3803 def : Pat<(v4i32 (anyext (v4i16 V64:$Rn))), (USHLLv4i16_shift V64:$Rn, (i32 0))>;
3804 def : Pat<(v2i64 (sext (v2i32 V64:$Rn))), (SSHLLv2i32_shift V64:$Rn, (i32 0))>;
3805 def : Pat<(v2i64 (zext (v2i32 V64:$Rn))), (USHLLv2i32_shift V64:$Rn, (i32 0))>;
3806 def : Pat<(v2i64 (anyext (v2i32 V64:$Rn))), (USHLLv2i32_shift V64:$Rn, (i32 0))>;
3807 // Also match an extend from the upper half of a 128 bit source register.
3808 def : Pat<(v8i16 (anyext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
3809 (USHLLv16i8_shift V128:$Rn, (i32 0))>;
3810 def : Pat<(v8i16 (zext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
3811 (USHLLv16i8_shift V128:$Rn, (i32 0))>;
3812 def : Pat<(v8i16 (sext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
3813 (SSHLLv16i8_shift V128:$Rn, (i32 0))>;
3814 def : Pat<(v4i32 (anyext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
3815 (USHLLv8i16_shift V128:$Rn, (i32 0))>;
3816 def : Pat<(v4i32 (zext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
3817 (USHLLv8i16_shift V128:$Rn, (i32 0))>;
3818 def : Pat<(v4i32 (sext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
3819 (SSHLLv8i16_shift V128:$Rn, (i32 0))>;
3820 def : Pat<(v2i64 (anyext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
3821 (USHLLv4i32_shift V128:$Rn, (i32 0))>;
3822 def : Pat<(v2i64 (zext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
3823 (USHLLv4i32_shift V128:$Rn, (i32 0))>;
3824 def : Pat<(v2i64 (sext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
3825 (SSHLLv4i32_shift V128:$Rn, (i32 0))>;
3827 // Vector shift sxtl aliases
3828 def : InstAlias<"sxtl.8h $dst, $src1",
3829 (SSHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
3830 def : InstAlias<"sxtl $dst.8h, $src1.8b",
3831 (SSHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
3832 def : InstAlias<"sxtl.4s $dst, $src1",
3833 (SSHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
3834 def : InstAlias<"sxtl $dst.4s, $src1.4h",
3835 (SSHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
3836 def : InstAlias<"sxtl.2d $dst, $src1",
3837 (SSHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
3838 def : InstAlias<"sxtl $dst.2d, $src1.2s",
3839 (SSHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
3841 // Vector shift sxtl2 aliases
3842 def : InstAlias<"sxtl2.8h $dst, $src1",
3843 (SSHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
3844 def : InstAlias<"sxtl2 $dst.8h, $src1.16b",
3845 (SSHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
3846 def : InstAlias<"sxtl2.4s $dst, $src1",
3847 (SSHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
3848 def : InstAlias<"sxtl2 $dst.4s, $src1.8h",
3849 (SSHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
3850 def : InstAlias<"sxtl2.2d $dst, $src1",
3851 (SSHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
3852 def : InstAlias<"sxtl2 $dst.2d, $src1.4s",
3853 (SSHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
3855 // Vector shift uxtl aliases
3856 def : InstAlias<"uxtl.8h $dst, $src1",
3857 (USHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
3858 def : InstAlias<"uxtl $dst.8h, $src1.8b",
3859 (USHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
3860 def : InstAlias<"uxtl.4s $dst, $src1",
3861 (USHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
3862 def : InstAlias<"uxtl $dst.4s, $src1.4h",
3863 (USHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
3864 def : InstAlias<"uxtl.2d $dst, $src1",
3865 (USHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
3866 def : InstAlias<"uxtl $dst.2d, $src1.2s",
3867 (USHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
3869 // Vector shift uxtl2 aliases
3870 def : InstAlias<"uxtl2.8h $dst, $src1",
3871 (USHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
3872 def : InstAlias<"uxtl2 $dst.8h, $src1.16b",
3873 (USHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
3874 def : InstAlias<"uxtl2.4s $dst, $src1",
3875 (USHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
3876 def : InstAlias<"uxtl2 $dst.4s, $src1.8h",
3877 (USHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
3878 def : InstAlias<"uxtl2.2d $dst, $src1",
3879 (USHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
3880 def : InstAlias<"uxtl2 $dst.2d, $src1.4s",
3881 (USHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
3883 // If an integer is about to be converted to a floating point value,
3884 // just load it on the floating point unit.
3885 // These patterns are more complex because floating point loads do not
3886 // support sign extension.
3887 // The sign extension has to be explicitly added and is only supported for
3888 // one step: byte-to-half, half-to-word, word-to-doubleword.
3889 // SCVTF GPR -> FPR is 9 cycles.
3890 // SCVTF FPR -> FPR is 4 cyclces.
3891 // (sign extension with lengthen) SXTL FPR -> FPR is 2 cycles.
3892 // Therefore, we can do 2 sign extensions and one SCVTF FPR -> FPR
3893 // and still being faster.
3894 // However, this is not good for code size.
3895 // 8-bits -> float. 2 sizes step-up.
3896 def : Pat <(f32 (sint_to_fp (i32 (sextloadi8 ro_indexed8:$addr)))),
3897 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
3902 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3903 (LDRBro ro_indexed8:$addr),
3908 ssub)))>, Requires<[NotForCodeSize]>;
3909 def : Pat <(f32 (sint_to_fp (i32 (sextloadi8 am_indexed8:$addr)))),
3910 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
3915 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3916 (LDRBui am_indexed8:$addr),
3921 ssub)))>, Requires<[NotForCodeSize]>;
3922 def : Pat <(f32 (sint_to_fp (i32 (sextloadi8 am_unscaled8:$addr)))),
3923 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
3928 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3929 (LDURBi am_unscaled8:$addr),
3934 ssub)))>, Requires<[NotForCodeSize]>;
3935 // 16-bits -> float. 1 size step-up.
3936 def : Pat <(f32 (sint_to_fp (i32 (sextloadi16 ro_indexed16:$addr)))),
3937 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
3939 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3940 (LDRHro ro_indexed16:$addr),
3943 ssub)))>, Requires<[NotForCodeSize]>;
3944 def : Pat <(f32 (sint_to_fp (i32 (sextloadi16 am_indexed16:$addr)))),
3945 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
3947 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3948 (LDRHui am_indexed16:$addr),
3951 ssub)))>, Requires<[NotForCodeSize]>;
3952 def : Pat <(f32 (sint_to_fp (i32 (sextloadi16 am_unscaled16:$addr)))),
3953 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
3955 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3956 (LDURHi am_unscaled16:$addr),
3959 ssub)))>, Requires<[NotForCodeSize]>;
3960 // 32-bits to 32-bits are handled in target specific dag combine:
3961 // performIntToFpCombine.
3962 // 64-bits integer to 32-bits floating point, not possible with
3963 // SCVTF on floating point registers (both source and destination
3964 // must have the same size).
3966 // Here are the patterns for 8, 16, 32, and 64-bits to double.
3967 // 8-bits -> double. 3 size step-up: give up.
3968 // 16-bits -> double. 2 size step.
3969 def : Pat <(f64 (sint_to_fp (i32 (sextloadi16 ro_indexed16:$addr)))),
3970 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
3975 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3976 (LDRHro ro_indexed16:$addr),
3981 dsub)))>, Requires<[NotForCodeSize]>;
3982 def : Pat <(f64 (sint_to_fp (i32 (sextloadi16 am_indexed16:$addr)))),
3983 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
3988 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3989 (LDRHui am_indexed16:$addr),
3994 dsub)))>, Requires<[NotForCodeSize]>;
3995 def : Pat <(f64 (sint_to_fp (i32 (sextloadi16 am_unscaled16:$addr)))),
3996 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4001 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4002 (LDURHi am_unscaled16:$addr),
4007 dsub)))>, Requires<[NotForCodeSize]>;
4008 // 32-bits -> double. 1 size step-up.
4009 def : Pat <(f64 (sint_to_fp (i32 (load ro_indexed32:$addr)))),
4010 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4012 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4013 (LDRSro ro_indexed32:$addr),
4016 dsub)))>, Requires<[NotForCodeSize]>;
4017 def : Pat <(f64 (sint_to_fp (i32 (load am_indexed32:$addr)))),
4018 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4020 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4021 (LDRSui am_indexed32:$addr),
4024 dsub)))>, Requires<[NotForCodeSize]>;
4025 def : Pat <(f64 (sint_to_fp (i32 (load am_unscaled32:$addr)))),
4026 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4028 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4029 (LDURSi am_unscaled32:$addr),
4032 dsub)))>, Requires<[NotForCodeSize]>;
4033 // 64-bits -> double are handled in target specific dag combine:
4034 // performIntToFpCombine.
4037 //----------------------------------------------------------------------------
4038 // AdvSIMD Load-Store Structure
4039 //----------------------------------------------------------------------------
4040 defm LD1 : SIMDLd1Multiple<"ld1">;
4041 defm LD2 : SIMDLd2Multiple<"ld2">;
4042 defm LD3 : SIMDLd3Multiple<"ld3">;
4043 defm LD4 : SIMDLd4Multiple<"ld4">;
4045 defm ST1 : SIMDSt1Multiple<"st1">;
4046 defm ST2 : SIMDSt2Multiple<"st2">;
4047 defm ST3 : SIMDSt3Multiple<"st3">;
4048 defm ST4 : SIMDSt4Multiple<"st4">;
4050 class Ld1Pat<ValueType ty, Instruction INST>
4051 : Pat<(ty (load am_simdnoindex:$vaddr)), (INST am_simdnoindex:$vaddr)>;
4053 def : Ld1Pat<v16i8, LD1Onev16b>;
4054 def : Ld1Pat<v8i16, LD1Onev8h>;
4055 def : Ld1Pat<v4i32, LD1Onev4s>;
4056 def : Ld1Pat<v2i64, LD1Onev2d>;
4057 def : Ld1Pat<v8i8, LD1Onev8b>;
4058 def : Ld1Pat<v4i16, LD1Onev4h>;
4059 def : Ld1Pat<v2i32, LD1Onev2s>;
4060 def : Ld1Pat<v1i64, LD1Onev1d>;
4062 class St1Pat<ValueType ty, Instruction INST>
4063 : Pat<(store ty:$Vt, am_simdnoindex:$vaddr),
4064 (INST ty:$Vt, am_simdnoindex:$vaddr)>;
4066 def : St1Pat<v16i8, ST1Onev16b>;
4067 def : St1Pat<v8i16, ST1Onev8h>;
4068 def : St1Pat<v4i32, ST1Onev4s>;
4069 def : St1Pat<v2i64, ST1Onev2d>;
4070 def : St1Pat<v8i8, ST1Onev8b>;
4071 def : St1Pat<v4i16, ST1Onev4h>;
4072 def : St1Pat<v2i32, ST1Onev2s>;
4073 def : St1Pat<v1i64, ST1Onev1d>;
4079 defm LD1R : SIMDLdR<0, 0b110, 0, "ld1r", "One", 1, 2, 4, 8>;
4080 defm LD2R : SIMDLdR<1, 0b110, 0, "ld2r", "Two", 2, 4, 8, 16>;
4081 defm LD3R : SIMDLdR<0, 0b111, 0, "ld3r", "Three", 3, 6, 12, 24>;
4082 defm LD4R : SIMDLdR<1, 0b111, 0, "ld4r", "Four", 4, 8, 16, 32>;
4083 let mayLoad = 1, neverHasSideEffects = 1 in {
4084 defm LD1 : SIMDLdSingleBTied<0, 0b000, "ld1", VecListOneb, GPR64pi1>;
4085 defm LD1 : SIMDLdSingleHTied<0, 0b010, 0, "ld1", VecListOneh, GPR64pi2>;
4086 defm LD1 : SIMDLdSingleSTied<0, 0b100, 0b00, "ld1", VecListOnes, GPR64pi4>;
4087 defm LD1 : SIMDLdSingleDTied<0, 0b100, 0b01, "ld1", VecListOned, GPR64pi8>;
4088 defm LD2 : SIMDLdSingleBTied<1, 0b000, "ld2", VecListTwob, GPR64pi2>;
4089 defm LD2 : SIMDLdSingleHTied<1, 0b010, 0, "ld2", VecListTwoh, GPR64pi4>;
4090 defm LD2 : SIMDLdSingleSTied<1, 0b100, 0b00, "ld2", VecListTwos, GPR64pi8>;
4091 defm LD2 : SIMDLdSingleDTied<1, 0b100, 0b01, "ld2", VecListTwod, GPR64pi16>;
4092 defm LD3 : SIMDLdSingleBTied<0, 0b001, "ld3", VecListThreeb, GPR64pi3>;
4093 defm LD3 : SIMDLdSingleHTied<0, 0b011, 0, "ld3", VecListThreeh, GPR64pi6>;
4094 defm LD3 : SIMDLdSingleSTied<0, 0b101, 0b00, "ld3", VecListThrees, GPR64pi12>;
4095 defm LD3 : SIMDLdSingleDTied<0, 0b101, 0b01, "ld3", VecListThreed, GPR64pi24>;
4096 defm LD4 : SIMDLdSingleBTied<1, 0b001, "ld4", VecListFourb, GPR64pi4>;
4097 defm LD4 : SIMDLdSingleHTied<1, 0b011, 0, "ld4", VecListFourh, GPR64pi8>;
4098 defm LD4 : SIMDLdSingleSTied<1, 0b101, 0b00, "ld4", VecListFours, GPR64pi16>;
4099 defm LD4 : SIMDLdSingleDTied<1, 0b101, 0b01, "ld4", VecListFourd, GPR64pi32>;
4102 def : Pat<(v8i8 (ARM64dup (i32 (extloadi8 am_simdnoindex:$vaddr)))),
4103 (LD1Rv8b am_simdnoindex:$vaddr)>;
4104 def : Pat<(v16i8 (ARM64dup (i32 (extloadi8 am_simdnoindex:$vaddr)))),
4105 (LD1Rv16b am_simdnoindex:$vaddr)>;
4106 def : Pat<(v4i16 (ARM64dup (i32 (extloadi16 am_simdnoindex:$vaddr)))),
4107 (LD1Rv4h am_simdnoindex:$vaddr)>;
4108 def : Pat<(v8i16 (ARM64dup (i32 (extloadi16 am_simdnoindex:$vaddr)))),
4109 (LD1Rv8h am_simdnoindex:$vaddr)>;
4110 def : Pat<(v2i32 (ARM64dup (i32 (load am_simdnoindex:$vaddr)))),
4111 (LD1Rv2s am_simdnoindex:$vaddr)>;
4112 def : Pat<(v4i32 (ARM64dup (i32 (load am_simdnoindex:$vaddr)))),
4113 (LD1Rv4s am_simdnoindex:$vaddr)>;
4114 def : Pat<(v2i64 (ARM64dup (i64 (load am_simdnoindex:$vaddr)))),
4115 (LD1Rv2d am_simdnoindex:$vaddr)>;
4116 def : Pat<(v1i64 (ARM64dup (i64 (load am_simdnoindex:$vaddr)))),
4117 (LD1Rv1d am_simdnoindex:$vaddr)>;
4118 // Grab the floating point version too
4119 def : Pat<(v2f32 (ARM64dup (f32 (load am_simdnoindex:$vaddr)))),
4120 (LD1Rv2s am_simdnoindex:$vaddr)>;
4121 def : Pat<(v4f32 (ARM64dup (f32 (load am_simdnoindex:$vaddr)))),
4122 (LD1Rv4s am_simdnoindex:$vaddr)>;
4123 def : Pat<(v2f64 (ARM64dup (f64 (load am_simdnoindex:$vaddr)))),
4124 (LD1Rv2d am_simdnoindex:$vaddr)>;
4125 def : Pat<(v1f64 (ARM64dup (f64 (load am_simdnoindex:$vaddr)))),
4126 (LD1Rv1d am_simdnoindex:$vaddr)>;
4128 class Ld1Lane128Pat<SDPatternOperator scalar_load, Operand VecIndex,
4129 ValueType VTy, ValueType STy, Instruction LD1>
4130 : Pat<(vector_insert (VTy VecListOne128:$Rd),
4131 (STy (scalar_load am_simdnoindex:$vaddr)), VecIndex:$idx),
4132 (LD1 VecListOne128:$Rd, VecIndex:$idx, am_simdnoindex:$vaddr)>;
4134 def : Ld1Lane128Pat<extloadi8, VectorIndexB, v16i8, i32, LD1i8>;
4135 def : Ld1Lane128Pat<extloadi16, VectorIndexH, v8i16, i32, LD1i16>;
4136 def : Ld1Lane128Pat<load, VectorIndexS, v4i32, i32, LD1i32>;
4137 def : Ld1Lane128Pat<load, VectorIndexS, v4f32, f32, LD1i32>;
4138 def : Ld1Lane128Pat<load, VectorIndexD, v2i64, i64, LD1i64>;
4139 def : Ld1Lane128Pat<load, VectorIndexD, v2f64, f64, LD1i64>;
4141 class Ld1Lane64Pat<SDPatternOperator scalar_load, Operand VecIndex,
4142 ValueType VTy, ValueType STy, Instruction LD1>
4143 : Pat<(vector_insert (VTy VecListOne64:$Rd),
4144 (STy (scalar_load am_simdnoindex:$vaddr)), VecIndex:$idx),
4146 (LD1 (SUBREG_TO_REG (i32 0), VecListOne64:$Rd, dsub),
4147 VecIndex:$idx, am_simdnoindex:$vaddr),
4150 def : Ld1Lane64Pat<extloadi8, VectorIndexB, v8i8, i32, LD1i8>;
4151 def : Ld1Lane64Pat<extloadi16, VectorIndexH, v4i16, i32, LD1i16>;
4152 def : Ld1Lane64Pat<load, VectorIndexS, v2i32, i32, LD1i32>;
4153 def : Ld1Lane64Pat<load, VectorIndexS, v2f32, f32, LD1i32>;
4156 defm LD1 : SIMDLdSt1SingleAliases<"ld1">;
4157 defm LD2 : SIMDLdSt2SingleAliases<"ld2">;
4158 defm LD3 : SIMDLdSt3SingleAliases<"ld3">;
4159 defm LD4 : SIMDLdSt4SingleAliases<"ld4">;
4162 defm ST1 : SIMDStSingleB<0, 0b000, "st1", VecListOneb, GPR64pi1>;
4163 defm ST1 : SIMDStSingleH<0, 0b010, 0, "st1", VecListOneh, GPR64pi2>;
4164 defm ST1 : SIMDStSingleS<0, 0b100, 0b00, "st1", VecListOnes, GPR64pi4>;
4165 defm ST1 : SIMDStSingleD<0, 0b100, 0b01, "st1", VecListOned, GPR64pi8>;
4167 let AddedComplexity = 8 in
4168 class St1Lane128Pat<SDPatternOperator scalar_store, Operand VecIndex,
4169 ValueType VTy, ValueType STy, Instruction ST1>
4171 (STy (vector_extract (VTy VecListOne128:$Vt), VecIndex:$idx)),
4172 am_simdnoindex:$vaddr),
4173 (ST1 VecListOne128:$Vt, VecIndex:$idx, am_simdnoindex:$vaddr)>;
4175 def : St1Lane128Pat<truncstorei8, VectorIndexB, v16i8, i32, ST1i8>;
4176 def : St1Lane128Pat<truncstorei16, VectorIndexH, v8i16, i32, ST1i16>;
4177 def : St1Lane128Pat<store, VectorIndexS, v4i32, i32, ST1i32>;
4178 def : St1Lane128Pat<store, VectorIndexS, v4f32, f32, ST1i32>;
4179 def : St1Lane128Pat<store, VectorIndexD, v2i64, i64, ST1i64>;
4180 def : St1Lane128Pat<store, VectorIndexD, v2f64, f64, ST1i64>;
4182 let AddedComplexity = 8 in
4183 class St1Lane64Pat<SDPatternOperator scalar_store, Operand VecIndex,
4184 ValueType VTy, ValueType STy, Instruction ST1>
4186 (STy (vector_extract (VTy VecListOne64:$Vt), VecIndex:$idx)),
4187 am_simdnoindex:$vaddr),
4188 (ST1 (SUBREG_TO_REG (i32 0), VecListOne64:$Vt, dsub),
4189 VecIndex:$idx, am_simdnoindex:$vaddr)>;
4191 def : St1Lane64Pat<truncstorei8, VectorIndexB, v8i8, i32, ST1i8>;
4192 def : St1Lane64Pat<truncstorei16, VectorIndexH, v4i16, i32, ST1i16>;
4193 def : St1Lane64Pat<store, VectorIndexS, v2i32, i32, ST1i32>;
4194 def : St1Lane64Pat<store, VectorIndexS, v2f32, f32, ST1i32>;
4196 let mayStore = 1, neverHasSideEffects = 1 in {
4197 defm ST2 : SIMDStSingleB<1, 0b000, "st2", VecListTwob, GPR64pi2>;
4198 defm ST2 : SIMDStSingleH<1, 0b010, 0, "st2", VecListTwoh, GPR64pi4>;
4199 defm ST2 : SIMDStSingleS<1, 0b100, 0b00, "st2", VecListTwos, GPR64pi8>;
4200 defm ST2 : SIMDStSingleD<1, 0b100, 0b01, "st2", VecListTwod, GPR64pi16>;
4201 defm ST3 : SIMDStSingleB<0, 0b001, "st3", VecListThreeb, GPR64pi3>;
4202 defm ST3 : SIMDStSingleH<0, 0b011, 0, "st3", VecListThreeh, GPR64pi6>;
4203 defm ST3 : SIMDStSingleS<0, 0b101, 0b00, "st3", VecListThrees, GPR64pi12>;
4204 defm ST3 : SIMDStSingleD<0, 0b101, 0b01, "st3", VecListThreed, GPR64pi24>;
4205 defm ST4 : SIMDStSingleB<1, 0b001, "st4", VecListFourb, GPR64pi4>;
4206 defm ST4 : SIMDStSingleH<1, 0b011, 0, "st4", VecListFourh, GPR64pi8>;
4207 defm ST4 : SIMDStSingleS<1, 0b101, 0b00, "st4", VecListFours, GPR64pi16>;
4208 defm ST4 : SIMDStSingleD<1, 0b101, 0b01, "st4", VecListFourd, GPR64pi32>;
4211 defm ST1 : SIMDLdSt1SingleAliases<"st1">;
4212 defm ST2 : SIMDLdSt2SingleAliases<"st2">;
4213 defm ST3 : SIMDLdSt3SingleAliases<"st3">;
4214 defm ST4 : SIMDLdSt4SingleAliases<"st4">;
4216 //----------------------------------------------------------------------------
4217 // Crypto extensions
4218 //----------------------------------------------------------------------------
4220 def AESErr : AESTiedInst<0b0100, "aese", int_arm64_crypto_aese>;
4221 def AESDrr : AESTiedInst<0b0101, "aesd", int_arm64_crypto_aesd>;
4222 def AESMCrr : AESInst< 0b0110, "aesmc", int_arm64_crypto_aesmc>;
4223 def AESIMCrr : AESInst< 0b0111, "aesimc", int_arm64_crypto_aesimc>;
4225 def SHA1Crrr : SHATiedInstQSV<0b000, "sha1c", int_arm64_crypto_sha1c>;
4226 def SHA1Prrr : SHATiedInstQSV<0b001, "sha1p", int_arm64_crypto_sha1p>;
4227 def SHA1Mrrr : SHATiedInstQSV<0b010, "sha1m", int_arm64_crypto_sha1m>;
4228 def SHA1SU0rrr : SHATiedInstVVV<0b011, "sha1su0", int_arm64_crypto_sha1su0>;
4229 def SHA256Hrrr : SHATiedInstQQV<0b100, "sha256h", int_arm64_crypto_sha256h>;
4230 def SHA256H2rrr : SHATiedInstQQV<0b101, "sha256h2",int_arm64_crypto_sha256h2>;
4231 def SHA256SU1rrr :SHATiedInstVVV<0b110, "sha256su1",int_arm64_crypto_sha256su1>;
4233 def SHA1Hrr : SHAInstSS< 0b0000, "sha1h", int_arm64_crypto_sha1h>;
4234 def SHA1SU1rr : SHATiedInstVV<0b0001, "sha1su1", int_arm64_crypto_sha1su1>;
4235 def SHA256SU0rr : SHATiedInstVV<0b0010, "sha256su0",int_arm64_crypto_sha256su0>;
4237 //----------------------------------------------------------------------------
4239 //----------------------------------------------------------------------------
4240 // FIXME: Like for X86, these should go in their own separate .td file.
4242 // Any instruction that defines a 32-bit result leaves the high half of the
4243 // register. Truncate can be lowered to EXTRACT_SUBREG. CopyFromReg may
4244 // be copying from a truncate. But any other 32-bit operation will zero-extend
4246 // FIXME: X86 also checks for CMOV here. Do we need something similar?
4247 def def32 : PatLeaf<(i32 GPR32:$src), [{
4248 return N->getOpcode() != ISD::TRUNCATE &&
4249 N->getOpcode() != TargetOpcode::EXTRACT_SUBREG &&
4250 N->getOpcode() != ISD::CopyFromReg;
4253 // In the case of a 32-bit def that is known to implicitly zero-extend,
4254 // we can use a SUBREG_TO_REG.
4255 def : Pat<(i64 (zext def32:$src)), (SUBREG_TO_REG (i64 0), GPR32:$src, sub_32)>;
4257 // For an anyext, we don't care what the high bits are, so we can perform an
4258 // INSERT_SUBREF into an IMPLICIT_DEF.
4259 def : Pat<(i64 (anyext GPR32:$src)),
4260 (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32)>;
4262 // When we need to explicitly zero-extend, we use an unsigned bitfield move
4263 // instruction (UBFM) on the enclosing super-reg.
4264 def : Pat<(i64 (zext GPR32:$src)),
4265 (UBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32), 0, 31)>;
4267 // To sign extend, we use a signed bitfield move instruction (SBFM) on the
4268 // containing super-reg.
4269 def : Pat<(i64 (sext GPR32:$src)),
4270 (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32), 0, 31)>;
4271 def : Pat<(i64 (sext_inreg GPR64:$src, i32)), (SBFMXri GPR64:$src, 0, 31)>;
4272 def : Pat<(i64 (sext_inreg GPR64:$src, i16)), (SBFMXri GPR64:$src, 0, 15)>;
4273 def : Pat<(i64 (sext_inreg GPR64:$src, i8)), (SBFMXri GPR64:$src, 0, 7)>;
4274 def : Pat<(i64 (sext_inreg GPR64:$src, i1)), (SBFMXri GPR64:$src, 0, 0)>;
4275 def : Pat<(i32 (sext_inreg GPR32:$src, i16)), (SBFMWri GPR32:$src, 0, 15)>;
4276 def : Pat<(i32 (sext_inreg GPR32:$src, i8)), (SBFMWri GPR32:$src, 0, 7)>;
4277 def : Pat<(i32 (sext_inreg GPR32:$src, i1)), (SBFMWri GPR32:$src, 0, 0)>;
4279 def : Pat<(shl (sext_inreg GPR32:$Rn, i8), (i64 imm0_31:$imm)),
4280 (SBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
4281 (i64 (i32shift_sext_i8 imm0_31:$imm)))>;
4282 def : Pat<(shl (sext_inreg GPR64:$Rn, i8), (i64 imm0_63:$imm)),
4283 (SBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
4284 (i64 (i64shift_sext_i8 imm0_63:$imm)))>;
4286 def : Pat<(shl (sext_inreg GPR32:$Rn, i16), (i64 imm0_31:$imm)),
4287 (SBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
4288 (i64 (i32shift_sext_i16 imm0_31:$imm)))>;
4289 def : Pat<(shl (sext_inreg GPR64:$Rn, i16), (i64 imm0_63:$imm)),
4290 (SBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
4291 (i64 (i64shift_sext_i16 imm0_63:$imm)))>;
4293 def : Pat<(shl (i64 (sext GPR32:$Rn)), (i64 imm0_63:$imm)),
4294 (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$Rn, sub_32),
4295 (i64 (i64shift_a imm0_63:$imm)),
4296 (i64 (i64shift_sext_i32 imm0_63:$imm)))>;
4298 // sra patterns have an AddedComplexity of 10, so make sure we have a higher
4299 // AddedComplexity for the following patterns since we want to match sext + sra
4300 // patterns before we attempt to match a single sra node.
4301 let AddedComplexity = 20 in {
4302 // We support all sext + sra combinations which preserve at least one bit of the
4303 // original value which is to be sign extended. E.g. we support shifts up to
4305 def : Pat<(sra (sext_inreg GPR32:$Rn, i8), (i64 imm0_7:$imm)),
4306 (SBFMWri GPR32:$Rn, (i64 imm0_7:$imm), 7)>;
4307 def : Pat<(sra (sext_inreg GPR64:$Rn, i8), (i64 imm0_7:$imm)),
4308 (SBFMXri GPR64:$Rn, (i64 imm0_7:$imm), 7)>;
4310 def : Pat<(sra (sext_inreg GPR32:$Rn, i16), (i64 imm0_15:$imm)),
4311 (SBFMWri GPR32:$Rn, (i64 imm0_15:$imm), 15)>;
4312 def : Pat<(sra (sext_inreg GPR64:$Rn, i16), (i64 imm0_15:$imm)),
4313 (SBFMXri GPR64:$Rn, (i64 imm0_15:$imm), 15)>;
4315 def : Pat<(sra (i64 (sext GPR32:$Rn)), (i64 imm0_31:$imm)),
4316 (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$Rn, sub_32),
4317 (i64 imm0_31:$imm), 31)>;
4318 } // AddedComplexity = 20
4320 // To truncate, we can simply extract from a subregister.
4321 def : Pat<(i32 (trunc GPR64sp:$src)),
4322 (i32 (EXTRACT_SUBREG GPR64sp:$src, sub_32))>;
4324 // __builtin_trap() uses the BRK instruction on ARM64.
4325 def : Pat<(trap), (BRK 1)>;
4327 // Conversions within AdvSIMD types in the same register size are free.
4329 def : Pat<(v1i64 (bitconvert (v2i32 FPR64:$src))), (v1i64 FPR64:$src)>;
4330 def : Pat<(v1i64 (bitconvert (v4i16 FPR64:$src))), (v1i64 FPR64:$src)>;
4331 def : Pat<(v1i64 (bitconvert (v8i8 FPR64:$src))), (v1i64 FPR64:$src)>;
4332 def : Pat<(v1i64 (bitconvert (f64 FPR64:$src))), (v1i64 FPR64:$src)>;
4333 def : Pat<(v1i64 (bitconvert (v2f32 FPR64:$src))), (v1i64 FPR64:$src)>;
4334 def : Pat<(v1i64 (bitconvert (v1f64 FPR64:$src))), (v1i64 FPR64:$src)>;
4336 def : Pat<(v2i32 (bitconvert (v1i64 FPR64:$src))), (v2i32 FPR64:$src)>;
4337 def : Pat<(v2i32 (bitconvert (v4i16 FPR64:$src))), (v2i32 FPR64:$src)>;
4338 def : Pat<(v2i32 (bitconvert (v8i8 FPR64:$src))), (v2i32 FPR64:$src)>;
4339 def : Pat<(v2i32 (bitconvert (f64 FPR64:$src))), (v2i32 FPR64:$src)>;
4340 def : Pat<(v2i32 (bitconvert (v2f32 FPR64:$src))), (v2i32 FPR64:$src)>;
4341 def : Pat<(v2i32 (bitconvert (v1f64 FPR64:$src))), (v2i32 FPR64:$src)>;
4343 def : Pat<(v4i16 (bitconvert (v1i64 FPR64:$src))), (v4i16 FPR64:$src)>;
4344 def : Pat<(v4i16 (bitconvert (v2i32 FPR64:$src))), (v4i16 FPR64:$src)>;
4345 def : Pat<(v4i16 (bitconvert (v8i8 FPR64:$src))), (v4i16 FPR64:$src)>;
4346 def : Pat<(v4i16 (bitconvert (f64 FPR64:$src))), (v4i16 FPR64:$src)>;
4347 def : Pat<(v4i16 (bitconvert (v2f32 FPR64:$src))), (v4i16 FPR64:$src)>;
4348 def : Pat<(v4i16 (bitconvert (v1f64 FPR64:$src))), (v4i16 FPR64:$src)>;
4350 def : Pat<(v8i8 (bitconvert (v1i64 FPR64:$src))), (v8i8 FPR64:$src)>;
4351 def : Pat<(v8i8 (bitconvert (v2i32 FPR64:$src))), (v8i8 FPR64:$src)>;
4352 def : Pat<(v8i8 (bitconvert (v4i16 FPR64:$src))), (v8i8 FPR64:$src)>;
4353 def : Pat<(v8i8 (bitconvert (f64 FPR64:$src))), (v8i8 FPR64:$src)>;
4354 def : Pat<(v8i8 (bitconvert (v2f32 FPR64:$src))), (v8i8 FPR64:$src)>;
4355 def : Pat<(v8i8 (bitconvert (v1f64 FPR64:$src))), (v8i8 FPR64:$src)>;
4357 def : Pat<(f64 (bitconvert (v1i64 FPR64:$src))), (f64 FPR64:$src)>;
4358 def : Pat<(f64 (bitconvert (v2i32 FPR64:$src))), (f64 FPR64:$src)>;
4359 def : Pat<(f64 (bitconvert (v4i16 FPR64:$src))), (f64 FPR64:$src)>;
4360 def : Pat<(f64 (bitconvert (v8i8 FPR64:$src))), (f64 FPR64:$src)>;
4361 def : Pat<(f64 (bitconvert (v2f32 FPR64:$src))), (f64 FPR64:$src)>;
4362 def : Pat<(f64 (bitconvert (v1f64 FPR64:$src))), (f64 FPR64:$src)>;
4364 def : Pat<(v1f64 (bitconvert (v1i64 FPR64:$src))), (v1f64 FPR64:$src)>;
4365 def : Pat<(v1f64 (bitconvert (v2i32 FPR64:$src))), (v1f64 FPR64:$src)>;
4366 def : Pat<(v1f64 (bitconvert (v4i16 FPR64:$src))), (v1f64 FPR64:$src)>;
4367 def : Pat<(v1f64 (bitconvert (v8i8 FPR64:$src))), (v1f64 FPR64:$src)>;
4368 def : Pat<(v1f64 (bitconvert (f64 FPR64:$src))), (v1f64 FPR64:$src)>;
4369 def : Pat<(v1f64 (bitconvert (v2f32 FPR64:$src))), (v1f64 FPR64:$src)>;
4371 def : Pat<(v2f32 (bitconvert (f64 FPR64:$src))), (v2f32 FPR64:$src)>;
4372 def : Pat<(v2f32 (bitconvert (v1i64 FPR64:$src))), (v2f32 FPR64:$src)>;
4373 def : Pat<(v2f32 (bitconvert (v2i32 FPR64:$src))), (v2f32 FPR64:$src)>;
4374 def : Pat<(v2f32 (bitconvert (v4i16 FPR64:$src))), (v2f32 FPR64:$src)>;
4375 def : Pat<(v2f32 (bitconvert (v8i8 FPR64:$src))), (v2f32 FPR64:$src)>;
4376 def : Pat<(v2f32 (bitconvert (v1f64 FPR64:$src))), (v2f32 FPR64:$src)>;
4379 def : Pat<(f128 (bitconvert (v2i64 FPR128:$src))), (f128 FPR128:$src)>;
4380 def : Pat<(f128 (bitconvert (v4i32 FPR128:$src))), (f128 FPR128:$src)>;
4381 def : Pat<(f128 (bitconvert (v8i16 FPR128:$src))), (f128 FPR128:$src)>;
4382 def : Pat<(f128 (bitconvert (v2f64 FPR128:$src))), (f128 FPR128:$src)>;
4383 def : Pat<(f128 (bitconvert (v4f32 FPR128:$src))), (f128 FPR128:$src)>;
4385 def : Pat<(v2f64 (bitconvert (f128 FPR128:$src))), (v2f64 FPR128:$src)>;
4386 def : Pat<(v2f64 (bitconvert (v4i32 FPR128:$src))), (v2f64 FPR128:$src)>;
4387 def : Pat<(v2f64 (bitconvert (v8i16 FPR128:$src))), (v2f64 FPR128:$src)>;
4388 def : Pat<(v2f64 (bitconvert (v16i8 FPR128:$src))), (v2f64 FPR128:$src)>;
4389 def : Pat<(v2f64 (bitconvert (v2i64 FPR128:$src))), (v2f64 FPR128:$src)>;
4390 def : Pat<(v2f64 (bitconvert (v4f32 FPR128:$src))), (v2f64 FPR128:$src)>;
4392 def : Pat<(v4f32 (bitconvert (f128 FPR128:$src))), (v4f32 FPR128:$src)>;
4393 def : Pat<(v4f32 (bitconvert (v4i32 FPR128:$src))), (v4f32 FPR128:$src)>;
4394 def : Pat<(v4f32 (bitconvert (v8i16 FPR128:$src))), (v4f32 FPR128:$src)>;
4395 def : Pat<(v4f32 (bitconvert (v16i8 FPR128:$src))), (v4f32 FPR128:$src)>;
4396 def : Pat<(v4f32 (bitconvert (v2i64 FPR128:$src))), (v4f32 FPR128:$src)>;
4397 def : Pat<(v4f32 (bitconvert (v2f64 FPR128:$src))), (v4f32 FPR128:$src)>;
4399 def : Pat<(v2i64 (bitconvert (f128 FPR128:$src))), (v2i64 FPR128:$src)>;
4400 def : Pat<(v2i64 (bitconvert (v4i32 FPR128:$src))), (v2i64 FPR128:$src)>;
4401 def : Pat<(v2i64 (bitconvert (v8i16 FPR128:$src))), (v2i64 FPR128:$src)>;
4402 def : Pat<(v2i64 (bitconvert (v16i8 FPR128:$src))), (v2i64 FPR128:$src)>;
4403 def : Pat<(v2i64 (bitconvert (v2f64 FPR128:$src))), (v2i64 FPR128:$src)>;
4404 def : Pat<(v2i64 (bitconvert (v4f32 FPR128:$src))), (v2i64 FPR128:$src)>;
4406 def : Pat<(v4i32 (bitconvert (f128 FPR128:$src))), (v4i32 FPR128:$src)>;
4407 def : Pat<(v4i32 (bitconvert (v2i64 FPR128:$src))), (v4i32 FPR128:$src)>;
4408 def : Pat<(v4i32 (bitconvert (v8i16 FPR128:$src))), (v4i32 FPR128:$src)>;
4409 def : Pat<(v4i32 (bitconvert (v16i8 FPR128:$src))), (v4i32 FPR128:$src)>;
4410 def : Pat<(v4i32 (bitconvert (v2f64 FPR128:$src))), (v4i32 FPR128:$src)>;
4411 def : Pat<(v4i32 (bitconvert (v4f32 FPR128:$src))), (v4i32 FPR128:$src)>;
4413 def : Pat<(v8i16 (bitconvert (f128 FPR128:$src))), (v8i16 FPR128:$src)>;
4414 def : Pat<(v8i16 (bitconvert (v2i64 FPR128:$src))), (v8i16 FPR128:$src)>;
4415 def : Pat<(v8i16 (bitconvert (v4i32 FPR128:$src))), (v8i16 FPR128:$src)>;
4416 def : Pat<(v8i16 (bitconvert (v16i8 FPR128:$src))), (v8i16 FPR128:$src)>;
4417 def : Pat<(v8i16 (bitconvert (v2f64 FPR128:$src))), (v8i16 FPR128:$src)>;
4418 def : Pat<(v8i16 (bitconvert (v4f32 FPR128:$src))), (v8i16 FPR128:$src)>;
4420 def : Pat<(v16i8 (bitconvert (f128 FPR128:$src))), (v16i8 FPR128:$src)>;
4421 def : Pat<(v16i8 (bitconvert (v2i64 FPR128:$src))), (v16i8 FPR128:$src)>;
4422 def : Pat<(v16i8 (bitconvert (v4i32 FPR128:$src))), (v16i8 FPR128:$src)>;
4423 def : Pat<(v16i8 (bitconvert (v8i16 FPR128:$src))), (v16i8 FPR128:$src)>;
4424 def : Pat<(v16i8 (bitconvert (v2f64 FPR128:$src))), (v16i8 FPR128:$src)>;
4425 def : Pat<(v16i8 (bitconvert (v4f32 FPR128:$src))), (v16i8 FPR128:$src)>;
4427 def : Pat<(v8i8 (extract_subvector (v16i8 FPR128:$Rn), (i64 1))),
4428 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
4429 def : Pat<(v4i16 (extract_subvector (v8i16 FPR128:$Rn), (i64 1))),
4430 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
4431 def : Pat<(v2i32 (extract_subvector (v4i32 FPR128:$Rn), (i64 1))),
4432 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
4433 def : Pat<(v1i64 (extract_subvector (v2i64 FPR128:$Rn), (i64 1))),
4434 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
4436 // A 64-bit subvector insert to the first 128-bit vector position
4437 // is a subregister copy that needs no instruction.
4438 def : Pat<(insert_subvector undef, (v1i64 FPR64:$src), (i32 0)),
4439 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
4440 def : Pat<(insert_subvector undef, (v1f64 FPR64:$src), (i32 0)),
4441 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
4442 def : Pat<(insert_subvector undef, (v2i32 FPR64:$src), (i32 0)),
4443 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
4444 def : Pat<(insert_subvector undef, (v2f32 FPR64:$src), (i32 0)),
4445 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
4446 def : Pat<(insert_subvector undef, (v4i16 FPR64:$src), (i32 0)),
4447 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
4448 def : Pat<(insert_subvector undef, (v8i8 FPR64:$src), (i32 0)),
4449 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
4451 // Use pair-wise add instructions when summing up the lanes for v2f64, v2i64
4453 def : Pat<(i64 (add (vector_extract (v2i64 FPR128:$Rn), (i64 0)),
4454 (vector_extract (v2i64 FPR128:$Rn), (i64 1)))),
4455 (i64 (ADDPv2i64p (v2i64 FPR128:$Rn)))>;
4456 def : Pat<(f64 (fadd (vector_extract (v2f64 FPR128:$Rn), (i64 0)),
4457 (vector_extract (v2f64 FPR128:$Rn), (i64 1)))),
4458 (f64 (FADDPv2i64p (v2f64 FPR128:$Rn)))>;
4459 // vector_extract on 64-bit vectors gets promoted to a 128 bit vector,
4460 // so we match on v4f32 here, not v2f32. This will also catch adding
4461 // the low two lanes of a true v4f32 vector.
4462 def : Pat<(fadd (vector_extract (v4f32 FPR128:$Rn), (i64 0)),
4463 (vector_extract (v4f32 FPR128:$Rn), (i64 1))),
4464 (f32 (FADDPv2i32p (EXTRACT_SUBREG FPR128:$Rn, dsub)))>;
4466 // Scalar 64-bit shifts in FPR64 registers.
4467 def : Pat<(i64 (int_arm64_neon_sshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
4468 (SSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
4469 def : Pat<(i64 (int_arm64_neon_ushl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
4470 (USHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
4471 def : Pat<(i64 (int_arm64_neon_srshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
4472 (SRSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
4473 def : Pat<(i64 (int_arm64_neon_urshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
4474 (URSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
4476 // Tail call return handling. These are all compiler pseudo-instructions,
4477 // so no encoding information or anything like that.
4478 let isCall = 1, isTerminator = 1, isReturn = 1, isBarrier = 1, Uses = [SP] in {
4479 def TCRETURNdi : Pseudo<(outs), (ins i64imm:$dst), []>;
4480 def TCRETURNri : Pseudo<(outs), (ins tcGPR64:$dst), []>;
4483 def : Pat<(ARM64tcret tcGPR64:$dst), (TCRETURNri tcGPR64:$dst)>;
4484 def : Pat<(ARM64tcret (i64 tglobaladdr:$dst)), (TCRETURNdi texternalsym:$dst)>;
4485 def : Pat<(ARM64tcret (i64 texternalsym:$dst)), (TCRETURNdi texternalsym:$dst)>;
4487 include "ARM64InstrAtomics.td"