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),
419 def : Pat<(ARM64WrapperLarge tjumptable:$g3, tjumptable:$g2,
420 tjumptable:$g1, tjumptable:$g0),
421 (MOVKXi (MOVKXi (MOVKXi (MOVZXi tjumptable:$g3, 48),
427 //===----------------------------------------------------------------------===//
428 // Arithmetic instructions.
429 //===----------------------------------------------------------------------===//
431 // Add/subtract with carry.
432 defm ADC : AddSubCarry<0, "adc", "adcs", ARM64adc, ARM64adc_flag>;
433 defm SBC : AddSubCarry<1, "sbc", "sbcs", ARM64sbc, ARM64sbc_flag>;
435 def : InstAlias<"ngc $dst, $src", (SBCWr GPR32:$dst, WZR, GPR32:$src)>;
436 def : InstAlias<"ngc $dst, $src", (SBCXr GPR64:$dst, XZR, GPR64:$src)>;
437 def : InstAlias<"ngcs $dst, $src", (SBCSWr GPR32:$dst, WZR, GPR32:$src)>;
438 def : InstAlias<"ngcs $dst, $src", (SBCSXr GPR64:$dst, XZR, GPR64:$src)>;
441 defm ADD : AddSub<0, "add", add>;
442 defm SUB : AddSub<1, "sub">;
444 defm ADDS : AddSubS<0, "adds", ARM64add_flag>;
445 defm SUBS : AddSubS<1, "subs", ARM64sub_flag>;
447 // Use SUBS instead of SUB to enable CSE between SUBS and SUB.
448 def : Pat<(sub GPR32sp:$Rn, addsub_shifted_imm32:$imm),
449 (SUBSWri GPR32sp:$Rn, addsub_shifted_imm32:$imm)>;
450 def : Pat<(sub GPR64sp:$Rn, addsub_shifted_imm64:$imm),
451 (SUBSXri GPR64sp:$Rn, addsub_shifted_imm64:$imm)>;
452 def : Pat<(sub GPR32:$Rn, GPR32:$Rm),
453 (SUBSWrr GPR32:$Rn, GPR32:$Rm)>;
454 def : Pat<(sub GPR64:$Rn, GPR64:$Rm),
455 (SUBSXrr GPR64:$Rn, GPR64:$Rm)>;
456 def : Pat<(sub GPR32:$Rn, arith_shifted_reg32:$Rm),
457 (SUBSWrs GPR32:$Rn, arith_shifted_reg32:$Rm)>;
458 def : Pat<(sub GPR64:$Rn, arith_shifted_reg64:$Rm),
459 (SUBSXrs GPR64:$Rn, arith_shifted_reg64:$Rm)>;
460 def : Pat<(sub GPR32sp:$R2, arith_extended_reg32<i32>:$R3),
461 (SUBSWrx GPR32sp:$R2, arith_extended_reg32<i32>:$R3)>;
462 def : Pat<(sub GPR64sp:$R2, arith_extended_reg32to64<i64>:$R3),
463 (SUBSXrx GPR64sp:$R2, arith_extended_reg32to64<i64>:$R3)>;
465 // Because of the immediate format for add/sub-imm instructions, the
466 // expression (add x, -1) must be transformed to (SUB{W,X}ri x, 1).
467 // These patterns capture that transformation.
468 let AddedComplexity = 1 in {
469 def : Pat<(add GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
470 (SUBSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
471 def : Pat<(add GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
472 (SUBSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
473 def : Pat<(sub GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
474 (ADDWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
475 def : Pat<(sub GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
476 (ADDXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
479 def : InstAlias<"neg $dst, $src", (SUBWrs GPR32:$dst, WZR, GPR32:$src, 0)>;
480 def : InstAlias<"neg $dst, $src", (SUBXrs GPR64:$dst, XZR, GPR64:$src, 0)>;
481 def : InstAlias<"neg $dst, $src, $shift",
482 (SUBWrs GPR32:$dst, WZR, GPR32:$src, arith_shift:$shift)>;
483 def : InstAlias<"neg $dst, $src, $shift",
484 (SUBXrs GPR64:$dst, XZR, GPR64:$src, arith_shift:$shift)>;
486 // Because of the immediate format for add/sub-imm instructions, the
487 // expression (add x, -1) must be transformed to (SUB{W,X}ri x, 1).
488 // These patterns capture that transformation.
489 let AddedComplexity = 1 in {
490 def : Pat<(ARM64add_flag GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
491 (SUBSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
492 def : Pat<(ARM64add_flag GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
493 (SUBSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
494 def : Pat<(ARM64sub_flag GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
495 (ADDSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
496 def : Pat<(ARM64sub_flag GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
497 (ADDSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
500 def : InstAlias<"negs $dst, $src", (SUBSWrs GPR32:$dst, WZR, GPR32:$src, 0)>;
501 def : InstAlias<"negs $dst, $src", (SUBSXrs GPR64:$dst, XZR, GPR64:$src, 0)>;
502 def : InstAlias<"negs $dst, $src, $shift",
503 (SUBSWrs GPR32:$dst, WZR, GPR32:$src, arith_shift:$shift)>;
504 def : InstAlias<"negs $dst, $src, $shift",
505 (SUBSXrs GPR64:$dst, XZR, GPR64:$src, arith_shift:$shift)>;
507 // Unsigned/Signed divide
508 defm UDIV : Div<0, "udiv", udiv>;
509 defm SDIV : Div<1, "sdiv", sdiv>;
510 let isCodeGenOnly = 1 in {
511 defm UDIV_Int : Div<0, "udiv", int_arm64_udiv>;
512 defm SDIV_Int : Div<1, "sdiv", int_arm64_sdiv>;
516 defm ASRV : Shift<0b10, "asrv", sra>;
517 defm LSLV : Shift<0b00, "lslv", shl>;
518 defm LSRV : Shift<0b01, "lsrv", srl>;
519 defm RORV : Shift<0b11, "rorv", rotr>;
521 def : ShiftAlias<"asr", ASRVWr, GPR32>;
522 def : ShiftAlias<"asr", ASRVXr, GPR64>;
523 def : ShiftAlias<"lsl", LSLVWr, GPR32>;
524 def : ShiftAlias<"lsl", LSLVXr, GPR64>;
525 def : ShiftAlias<"lsr", LSRVWr, GPR32>;
526 def : ShiftAlias<"lsr", LSRVXr, GPR64>;
527 def : ShiftAlias<"ror", RORVWr, GPR32>;
528 def : ShiftAlias<"ror", RORVXr, GPR64>;
531 let AddedComplexity = 7 in {
532 defm MADD : MulAccum<0, "madd", add>;
533 defm MSUB : MulAccum<1, "msub", sub>;
535 def : Pat<(i32 (mul GPR32:$Rn, GPR32:$Rm)),
536 (MADDWrrr GPR32:$Rn, GPR32:$Rm, WZR)>;
537 def : Pat<(i64 (mul GPR64:$Rn, GPR64:$Rm)),
538 (MADDXrrr GPR64:$Rn, GPR64:$Rm, XZR)>;
540 def : Pat<(i32 (ineg (mul GPR32:$Rn, GPR32:$Rm))),
541 (MSUBWrrr GPR32:$Rn, GPR32:$Rm, WZR)>;
542 def : Pat<(i64 (ineg (mul GPR64:$Rn, GPR64:$Rm))),
543 (MSUBXrrr GPR64:$Rn, GPR64:$Rm, XZR)>;
544 } // AddedComplexity = 7
546 let AddedComplexity = 5 in {
547 def SMADDLrrr : WideMulAccum<0, 0b001, "smaddl", add, sext>;
548 def SMSUBLrrr : WideMulAccum<1, 0b001, "smsubl", sub, sext>;
549 def UMADDLrrr : WideMulAccum<0, 0b101, "umaddl", add, zext>;
550 def UMSUBLrrr : WideMulAccum<1, 0b101, "umsubl", sub, zext>;
552 def : Pat<(i64 (mul (sext GPR32:$Rn), (sext GPR32:$Rm))),
553 (SMADDLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
554 def : Pat<(i64 (mul (zext GPR32:$Rn), (zext GPR32:$Rm))),
555 (UMADDLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
557 def : Pat<(i64 (ineg (mul (sext GPR32:$Rn), (sext GPR32:$Rm)))),
558 (SMSUBLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
559 def : Pat<(i64 (ineg (mul (zext GPR32:$Rn), (zext GPR32:$Rm)))),
560 (UMSUBLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
561 } // AddedComplexity = 5
563 def : MulAccumWAlias<"mul", MADDWrrr>;
564 def : MulAccumXAlias<"mul", MADDXrrr>;
565 def : MulAccumWAlias<"mneg", MSUBWrrr>;
566 def : MulAccumXAlias<"mneg", MSUBXrrr>;
567 def : WideMulAccumAlias<"smull", SMADDLrrr>;
568 def : WideMulAccumAlias<"smnegl", SMSUBLrrr>;
569 def : WideMulAccumAlias<"umull", UMADDLrrr>;
570 def : WideMulAccumAlias<"umnegl", UMSUBLrrr>;
573 def SMULHrr : MulHi<0b010, "smulh", mulhs>;
574 def UMULHrr : MulHi<0b110, "umulh", mulhu>;
577 def CRC32Brr : BaseCRC32<0, 0b00, 0, GPR32, int_arm64_crc32b, "crc32b">;
578 def CRC32Hrr : BaseCRC32<0, 0b01, 0, GPR32, int_arm64_crc32h, "crc32h">;
579 def CRC32Wrr : BaseCRC32<0, 0b10, 0, GPR32, int_arm64_crc32w, "crc32w">;
580 def CRC32Xrr : BaseCRC32<1, 0b11, 0, GPR64, int_arm64_crc32x, "crc32x">;
582 def CRC32CBrr : BaseCRC32<0, 0b00, 1, GPR32, int_arm64_crc32cb, "crc32cb">;
583 def CRC32CHrr : BaseCRC32<0, 0b01, 1, GPR32, int_arm64_crc32ch, "crc32ch">;
584 def CRC32CWrr : BaseCRC32<0, 0b10, 1, GPR32, int_arm64_crc32cw, "crc32cw">;
585 def CRC32CXrr : BaseCRC32<1, 0b11, 1, GPR64, int_arm64_crc32cx, "crc32cx">;
588 //===----------------------------------------------------------------------===//
589 // Logical instructions.
590 //===----------------------------------------------------------------------===//
593 defm ANDS : LogicalImmS<0b11, "ands", ARM64and_flag>;
594 defm AND : LogicalImm<0b00, "and", and>;
595 defm EOR : LogicalImm<0b10, "eor", xor>;
596 defm ORR : LogicalImm<0b01, "orr", or>;
598 def : InstAlias<"mov $dst, $imm", (ORRWri GPR32sp:$dst, WZR,
599 logical_imm32:$imm)>;
600 def : InstAlias<"mov $dst, $imm", (ORRXri GPR64sp:$dst, XZR,
601 logical_imm64:$imm)>;
605 defm ANDS : LogicalRegS<0b11, 0, "ands">;
606 defm BICS : LogicalRegS<0b11, 1, "bics">;
607 defm AND : LogicalReg<0b00, 0, "and", and>;
608 defm BIC : LogicalReg<0b00, 1, "bic",
609 BinOpFrag<(and node:$LHS, (not node:$RHS))>>;
610 defm EON : LogicalReg<0b10, 1, "eon",
611 BinOpFrag<(xor node:$LHS, (not node:$RHS))>>;
612 defm EOR : LogicalReg<0b10, 0, "eor", xor>;
613 defm ORN : LogicalReg<0b01, 1, "orn",
614 BinOpFrag<(or node:$LHS, (not node:$RHS))>>;
615 defm ORR : LogicalReg<0b01, 0, "orr", or>;
617 def : InstAlias<"tst $src1, $src2",
618 (ANDSWri WZR, GPR32:$src1, logical_imm32:$src2)>;
619 def : InstAlias<"tst $src1, $src2",
620 (ANDSXri XZR, GPR64:$src1, logical_imm64:$src2)>;
622 def : InstAlias<"tst $src1, $src2",
623 (ANDSWrs WZR, GPR32:$src1, GPR32:$src2, 0)>;
624 def : InstAlias<"tst $src1, $src2",
625 (ANDSXrs XZR, GPR64:$src1, GPR64:$src2, 0)>;
627 def : InstAlias<"tst $src1, $src2, $sh",
628 (ANDSWrs WZR, GPR32:$src1, GPR32:$src2, logical_shift:$sh)>;
629 def : InstAlias<"tst $src1, $src2, $sh",
630 (ANDSXrs XZR, GPR64:$src1, GPR64:$src2, logical_shift:$sh)>;
632 def : InstAlias<"mvn $Wd, $Wm",
633 (ORNWrs GPR32:$Wd, WZR, GPR32:$Wm, 0)>;
634 def : InstAlias<"mvn $Xd, $Xm",
635 (ORNXrs GPR64:$Xd, XZR, GPR64:$Xm, 0)>;
637 def : InstAlias<"mvn $Wd, $Wm, $sh",
638 (ORNWrs GPR32:$Wd, WZR, GPR32:$Wm, logical_shift:$sh)>;
639 def : InstAlias<"mvn $Xd, $Xm, $sh",
640 (ORNXrs GPR64:$Xd, XZR, GPR64:$Xm, logical_shift:$sh)>;
642 def : Pat<(not GPR32:$Wm), (ORNWrr WZR, GPR32:$Wm)>;
643 def : Pat<(not GPR64:$Xm), (ORNXrr XZR, GPR64:$Xm)>;
646 //===----------------------------------------------------------------------===//
647 // One operand data processing instructions.
648 //===----------------------------------------------------------------------===//
650 defm CLS : OneOperandData<0b101, "cls">;
651 defm CLZ : OneOperandData<0b100, "clz", ctlz>;
652 defm RBIT : OneOperandData<0b000, "rbit">;
653 def REV16Wr : OneWRegData<0b001, "rev16",
654 UnOpFrag<(rotr (bswap node:$LHS), (i64 16))>>;
655 def REV16Xr : OneXRegData<0b001, "rev16", null_frag>;
657 def : Pat<(cttz GPR32:$Rn),
658 (CLZWr (RBITWr GPR32:$Rn))>;
659 def : Pat<(cttz GPR64:$Rn),
660 (CLZXr (RBITXr GPR64:$Rn))>;
661 def : Pat<(ctlz (or (shl (xor (sra GPR32:$Rn, (i64 31)), GPR32:$Rn), (i64 1)),
664 def : Pat<(ctlz (or (shl (xor (sra GPR64:$Rn, (i64 63)), GPR64:$Rn), (i64 1)),
668 // Unlike the other one operand instructions, the instructions with the "rev"
669 // mnemonic do *not* just different in the size bit, but actually use different
670 // opcode bits for the different sizes.
671 def REVWr : OneWRegData<0b010, "rev", bswap>;
672 def REVXr : OneXRegData<0b011, "rev", bswap>;
673 def REV32Xr : OneXRegData<0b010, "rev32",
674 UnOpFrag<(rotr (bswap node:$LHS), (i64 32))>>;
676 // The bswap commutes with the rotr so we want a pattern for both possible
678 def : Pat<(bswap (rotr GPR32:$Rn, (i64 16))), (REV16Wr GPR32:$Rn)>;
679 def : Pat<(bswap (rotr GPR64:$Rn, (i64 32))), (REV32Xr GPR64:$Rn)>;
681 //===----------------------------------------------------------------------===//
682 // Bitfield immediate extraction instruction.
683 //===----------------------------------------------------------------------===//
684 let neverHasSideEffects = 1 in
685 defm EXTR : ExtractImm<"extr">;
686 def : InstAlias<"ror $dst, $src, $shift",
687 (EXTRWrri GPR32:$dst, GPR32:$src, GPR32:$src, imm0_31:$shift)>;
688 def : InstAlias<"ror $dst, $src, $shift",
689 (EXTRXrri GPR64:$dst, GPR64:$src, GPR64:$src, imm0_63:$shift)>;
691 def : Pat<(rotr GPR32:$Rn, (i64 imm0_31:$imm)),
692 (EXTRWrri GPR32:$Rn, GPR32:$Rn, imm0_31:$imm)>;
693 def : Pat<(rotr GPR64:$Rn, (i64 imm0_63:$imm)),
694 (EXTRXrri GPR64:$Rn, GPR64:$Rn, imm0_63:$imm)>;
696 //===----------------------------------------------------------------------===//
697 // Other bitfield immediate instructions.
698 //===----------------------------------------------------------------------===//
699 let neverHasSideEffects = 1 in {
700 defm BFM : BitfieldImmWith2RegArgs<0b01, "bfm">;
701 defm SBFM : BitfieldImm<0b00, "sbfm">;
702 defm UBFM : BitfieldImm<0b10, "ubfm">;
705 def i32shift_a : Operand<i64>, SDNodeXForm<imm, [{
706 uint64_t enc = (32 - N->getZExtValue()) & 0x1f;
707 return CurDAG->getTargetConstant(enc, MVT::i64);
710 def i32shift_b : Operand<i64>, SDNodeXForm<imm, [{
711 uint64_t enc = 31 - N->getZExtValue();
712 return CurDAG->getTargetConstant(enc, MVT::i64);
715 // min(7, 31 - shift_amt)
716 def i32shift_sext_i8 : Operand<i64>, SDNodeXForm<imm, [{
717 uint64_t enc = 31 - N->getZExtValue();
718 enc = enc > 7 ? 7 : enc;
719 return CurDAG->getTargetConstant(enc, MVT::i64);
722 // min(15, 31 - shift_amt)
723 def i32shift_sext_i16 : Operand<i64>, SDNodeXForm<imm, [{
724 uint64_t enc = 31 - N->getZExtValue();
725 enc = enc > 15 ? 15 : enc;
726 return CurDAG->getTargetConstant(enc, MVT::i64);
729 def i64shift_a : Operand<i64>, SDNodeXForm<imm, [{
730 uint64_t enc = (64 - N->getZExtValue()) & 0x3f;
731 return CurDAG->getTargetConstant(enc, MVT::i64);
734 def i64shift_b : Operand<i64>, SDNodeXForm<imm, [{
735 uint64_t enc = 63 - N->getZExtValue();
736 return CurDAG->getTargetConstant(enc, MVT::i64);
739 // min(7, 63 - shift_amt)
740 def i64shift_sext_i8 : Operand<i64>, SDNodeXForm<imm, [{
741 uint64_t enc = 63 - N->getZExtValue();
742 enc = enc > 7 ? 7 : enc;
743 return CurDAG->getTargetConstant(enc, MVT::i64);
746 // min(15, 63 - shift_amt)
747 def i64shift_sext_i16 : Operand<i64>, SDNodeXForm<imm, [{
748 uint64_t enc = 63 - N->getZExtValue();
749 enc = enc > 15 ? 15 : enc;
750 return CurDAG->getTargetConstant(enc, MVT::i64);
753 // min(31, 63 - shift_amt)
754 def i64shift_sext_i32 : Operand<i64>, SDNodeXForm<imm, [{
755 uint64_t enc = 63 - N->getZExtValue();
756 enc = enc > 31 ? 31 : enc;
757 return CurDAG->getTargetConstant(enc, MVT::i64);
760 def : Pat<(shl GPR32:$Rn, (i64 imm0_31:$imm)),
761 (UBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
762 (i64 (i32shift_b imm0_31:$imm)))>;
763 def : Pat<(shl GPR64:$Rn, (i64 imm0_63:$imm)),
764 (UBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
765 (i64 (i64shift_b imm0_63:$imm)))>;
767 let AddedComplexity = 10 in {
768 def : Pat<(sra GPR32:$Rn, (i64 imm0_31:$imm)),
769 (SBFMWri GPR32:$Rn, imm0_31:$imm, 31)>;
770 def : Pat<(sra GPR64:$Rn, (i64 imm0_63:$imm)),
771 (SBFMXri GPR64:$Rn, imm0_63:$imm, 63)>;
774 def : InstAlias<"asr $dst, $src, $shift",
775 (SBFMWri GPR32:$dst, GPR32:$src, imm0_31:$shift, 31)>;
776 def : InstAlias<"asr $dst, $src, $shift",
777 (SBFMXri GPR64:$dst, GPR64:$src, imm0_63:$shift, 63)>;
778 def : InstAlias<"sxtb $dst, $src", (SBFMWri GPR32:$dst, GPR32:$src, 0, 7)>;
779 def : InstAlias<"sxtb $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 7)>;
780 def : InstAlias<"sxth $dst, $src", (SBFMWri GPR32:$dst, GPR32:$src, 0, 15)>;
781 def : InstAlias<"sxth $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 15)>;
782 def : InstAlias<"sxtw $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 31)>;
784 def : Pat<(srl GPR32:$Rn, (i64 imm0_31:$imm)),
785 (UBFMWri GPR32:$Rn, imm0_31:$imm, 31)>;
786 def : Pat<(srl GPR64:$Rn, (i64 imm0_63:$imm)),
787 (UBFMXri GPR64:$Rn, imm0_63:$imm, 63)>;
789 def : InstAlias<"lsr $dst, $src, $shift",
790 (UBFMWri GPR32:$dst, GPR32:$src, imm0_31:$shift, 31)>;
791 def : InstAlias<"lsr $dst, $src, $shift",
792 (UBFMXri GPR64:$dst, GPR64:$src, imm0_63:$shift, 63)>;
793 def : InstAlias<"uxtb $dst, $src", (UBFMWri GPR32:$dst, GPR32:$src, 0, 7)>;
794 def : InstAlias<"uxtb $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 7)>;
795 def : InstAlias<"uxth $dst, $src", (UBFMWri GPR32:$dst, GPR32:$src, 0, 15)>;
796 def : InstAlias<"uxth $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 15)>;
797 def : InstAlias<"uxtw $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 31)>;
799 //===----------------------------------------------------------------------===//
800 // Conditionally set flags instructions.
801 //===----------------------------------------------------------------------===//
802 defm CCMN : CondSetFlagsImm<0, "ccmn">;
803 defm CCMP : CondSetFlagsImm<1, "ccmp">;
805 defm CCMN : CondSetFlagsReg<0, "ccmn">;
806 defm CCMP : CondSetFlagsReg<1, "ccmp">;
808 //===----------------------------------------------------------------------===//
809 // Conditional select instructions.
810 //===----------------------------------------------------------------------===//
811 defm CSEL : CondSelect<0, 0b00, "csel">;
813 def inc : PatFrag<(ops node:$in), (add node:$in, 1)>;
814 defm CSINC : CondSelectOp<0, 0b01, "csinc", inc>;
815 defm CSINV : CondSelectOp<1, 0b00, "csinv", not>;
816 defm CSNEG : CondSelectOp<1, 0b01, "csneg", ineg>;
818 def : Pat<(ARM64csinv GPR32:$tval, GPR32:$fval, (i32 imm:$cc), CPSR),
819 (CSINVWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
820 def : Pat<(ARM64csinv GPR64:$tval, GPR64:$fval, (i32 imm:$cc), CPSR),
821 (CSINVXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
822 def : Pat<(ARM64csneg GPR32:$tval, GPR32:$fval, (i32 imm:$cc), CPSR),
823 (CSNEGWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
824 def : Pat<(ARM64csneg GPR64:$tval, GPR64:$fval, (i32 imm:$cc), CPSR),
825 (CSNEGXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
826 def : Pat<(ARM64csinc GPR32:$tval, GPR32:$fval, (i32 imm:$cc), CPSR),
827 (CSINCWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
828 def : Pat<(ARM64csinc GPR64:$tval, GPR64:$fval, (i32 imm:$cc), CPSR),
829 (CSINCXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
831 def : Pat<(ARM64csel (i32 0), (i32 1), (i32 imm:$cc), CPSR),
832 (CSINCWr WZR, WZR, (i32 imm:$cc))>;
833 def : Pat<(ARM64csel (i64 0), (i64 1), (i32 imm:$cc), CPSR),
834 (CSINCXr XZR, XZR, (i32 imm:$cc))>;
835 def : Pat<(ARM64csel (i32 0), (i32 -1), (i32 imm:$cc), CPSR),
836 (CSINVWr WZR, WZR, (i32 imm:$cc))>;
837 def : Pat<(ARM64csel (i64 0), (i64 -1), (i32 imm:$cc), CPSR),
838 (CSINVXr XZR, XZR, (i32 imm:$cc))>;
840 // The inverse of the condition code from the alias instruction is what is used
841 // in the aliased instruction. The parser all ready inverts the condition code
842 // for these aliases.
843 // FIXME: Is this the correct way to handle these aliases?
844 def : InstAlias<"cset $dst, $cc", (CSINCWr GPR32:$dst, WZR, WZR, ccode:$cc)>;
845 def : InstAlias<"cset $dst, $cc", (CSINCXr GPR64:$dst, XZR, XZR, ccode:$cc)>;
847 def : InstAlias<"csetm $dst, $cc", (CSINVWr GPR32:$dst, WZR, WZR, ccode:$cc)>;
848 def : InstAlias<"csetm $dst, $cc", (CSINVXr GPR64:$dst, XZR, XZR, ccode:$cc)>;
850 def : InstAlias<"cinc $dst, $src, $cc",
851 (CSINCWr GPR32:$dst, GPR32:$src, GPR32:$src, ccode:$cc)>;
852 def : InstAlias<"cinc $dst, $src, $cc",
853 (CSINCXr GPR64:$dst, GPR64:$src, GPR64:$src, ccode:$cc)>;
855 def : InstAlias<"cinv $dst, $src, $cc",
856 (CSINVWr GPR32:$dst, GPR32:$src, GPR32:$src, ccode:$cc)>;
857 def : InstAlias<"cinv $dst, $src, $cc",
858 (CSINVXr GPR64:$dst, GPR64:$src, GPR64:$src, ccode:$cc)>;
860 def : InstAlias<"cneg $dst, $src, $cc",
861 (CSNEGWr GPR32:$dst, GPR32:$src, GPR32:$src, ccode:$cc)>;
862 def : InstAlias<"cneg $dst, $src, $cc",
863 (CSNEGXr GPR64:$dst, GPR64:$src, GPR64:$src, ccode:$cc)>;
865 //===----------------------------------------------------------------------===//
866 // PC-relative instructions.
867 //===----------------------------------------------------------------------===//
868 let isReMaterializable = 1 in {
869 let neverHasSideEffects = 1, mayStore = 0, mayLoad = 0 in {
870 def ADR : ADRI<0, "adr", adrlabel, []>;
871 } // neverHasSideEffects = 1
873 def ADRP : ADRI<1, "adrp", adrplabel,
874 [(set GPR64:$Xd, (ARM64adrp tglobaladdr:$label))]>;
875 } // isReMaterializable = 1
877 // page address of a constant pool entry, block address
878 def : Pat<(ARM64adrp tconstpool:$cp), (ADRP tconstpool:$cp)>;
879 def : Pat<(ARM64adrp tblockaddress:$cp), (ADRP tblockaddress:$cp)>;
881 //===----------------------------------------------------------------------===//
882 // Unconditional branch (register) instructions.
883 //===----------------------------------------------------------------------===//
885 let isReturn = 1, isTerminator = 1, isBarrier = 1 in {
886 def RET : BranchReg<0b0010, "ret", []>;
887 def DRPS : SpecialReturn<0b0101, "drps">;
888 def ERET : SpecialReturn<0b0100, "eret">;
889 } // isReturn = 1, isTerminator = 1, isBarrier = 1
891 // Default to the LR register.
892 def : InstAlias<"ret", (RET LR)>;
894 let isCall = 1, Defs = [LR], Uses = [SP] in {
895 def BLR : BranchReg<0b0001, "blr", [(ARM64call GPR64:$Rn)]>;
898 let isBranch = 1, isTerminator = 1, isBarrier = 1, isIndirectBranch = 1 in {
899 def BR : BranchReg<0b0000, "br", [(brind GPR64:$Rn)]>;
900 } // isBranch, isTerminator, isBarrier, isIndirectBranch
902 // Create a separate pseudo-instruction for codegen to use so that we don't
903 // flag lr as used in every function. It'll be restored before the RET by the
904 // epilogue if it's legitimately used.
905 def RET_ReallyLR : Pseudo<(outs), (ins), [(ARM64retflag)]> {
906 let isTerminator = 1;
911 // This is a directive-like pseudo-instruction. The purpose is to insert an
912 // R_AARCH64_TLSDESC_CALL relocation at the offset of the following instruction
913 // (which in the usual case is a BLR).
914 let hasSideEffects = 1 in
915 def TLSDESCCALL : Pseudo<(outs), (ins i64imm:$sym), []> {
916 let AsmString = ".tlsdesccall $sym";
919 // Pseudo-instruction representing a BLR with attached TLSDESC relocation. It
920 // gets expanded to two MCInsts during lowering.
921 let isCall = 1, Defs = [LR] in
923 : Pseudo<(outs), (ins GPR64:$dest, i64imm:$sym),
924 [(ARM64tlsdesc_call GPR64:$dest, tglobaltlsaddr:$sym)]>;
926 def : Pat<(ARM64tlsdesc_call GPR64:$dest, texternalsym:$sym),
927 (TLSDESC_BLR GPR64:$dest, texternalsym:$sym)>;
928 //===----------------------------------------------------------------------===//
929 // Conditional branch (immediate) instruction.
930 //===----------------------------------------------------------------------===//
931 def Bcc : BranchCond;
933 //===----------------------------------------------------------------------===//
934 // Compare-and-branch instructions.
935 //===----------------------------------------------------------------------===//
936 defm CBZ : CmpBranch<0, "cbz", ARM64cbz>;
937 defm CBNZ : CmpBranch<1, "cbnz", ARM64cbnz>;
939 //===----------------------------------------------------------------------===//
940 // Test-bit-and-branch instructions.
941 //===----------------------------------------------------------------------===//
942 def TBZ : TestBranch<0, "tbz", ARM64tbz>;
943 def TBNZ : TestBranch<1, "tbnz", ARM64tbnz>;
945 //===----------------------------------------------------------------------===//
946 // Unconditional branch (immediate) instructions.
947 //===----------------------------------------------------------------------===//
948 let isBranch = 1, isTerminator = 1, isBarrier = 1 in {
949 def B : BranchImm<0, "b", [(br bb:$addr)]>;
950 } // isBranch, isTerminator, isBarrier
952 let isCall = 1, Defs = [LR], Uses = [SP] in {
953 def BL : CallImm<1, "bl", [(ARM64call tglobaladdr:$addr)]>;
955 def : Pat<(ARM64call texternalsym:$func), (BL texternalsym:$func)>;
957 //===----------------------------------------------------------------------===//
958 // Exception generation instructions.
959 //===----------------------------------------------------------------------===//
960 def BRK : ExceptionGeneration<0b001, 0b00, "brk">;
961 def DCPS1 : ExceptionGeneration<0b101, 0b01, "dcps1">;
962 def DCPS2 : ExceptionGeneration<0b101, 0b10, "dcps2">;
963 def DCPS3 : ExceptionGeneration<0b101, 0b11, "dcps3">;
964 def HLT : ExceptionGeneration<0b010, 0b00, "hlt">;
965 def HVC : ExceptionGeneration<0b000, 0b10, "hvc">;
966 def SMC : ExceptionGeneration<0b000, 0b11, "smc">;
967 def SVC : ExceptionGeneration<0b000, 0b01, "svc">;
969 // DCPSn defaults to an immediate operand of zero if unspecified.
970 def : InstAlias<"dcps1", (DCPS1 0)>;
971 def : InstAlias<"dcps2", (DCPS2 0)>;
972 def : InstAlias<"dcps3", (DCPS3 0)>;
974 //===----------------------------------------------------------------------===//
975 // Load instructions.
976 //===----------------------------------------------------------------------===//
978 // Pair (indexed, offset)
979 def LDPWi : LoadPairOffset<0b00, 0, GPR32, am_indexed32simm7, "ldp">;
980 def LDPXi : LoadPairOffset<0b10, 0, GPR64, am_indexed64simm7, "ldp">;
981 def LDPSi : LoadPairOffset<0b00, 1, FPR32, am_indexed32simm7, "ldp">;
982 def LDPDi : LoadPairOffset<0b01, 1, FPR64, am_indexed64simm7, "ldp">;
983 def LDPQi : LoadPairOffset<0b10, 1, FPR128, am_indexed128simm7, "ldp">;
985 def LDPSWi : LoadPairOffset<0b01, 0, GPR64, am_indexed32simm7, "ldpsw">;
987 // Pair (pre-indexed)
988 def LDPWpre : LoadPairPreIdx<0b00, 0, GPR32, am_indexed32simm7_wb, "ldp">;
989 def LDPXpre : LoadPairPreIdx<0b10, 0, GPR64, am_indexed64simm7_wb, "ldp">;
990 def LDPSpre : LoadPairPreIdx<0b00, 1, FPR32, am_indexed32simm7_wb, "ldp">;
991 def LDPDpre : LoadPairPreIdx<0b01, 1, FPR64, am_indexed64simm7_wb, "ldp">;
992 def LDPQpre : LoadPairPreIdx<0b10, 1, FPR128, am_indexed128simm7_wb, "ldp">;
994 def LDPSWpre : LoadPairPreIdx<0b01, 0, GPR64, am_indexed32simm7_wb, "ldpsw">;
996 // Pair (post-indexed)
997 def LDPWpost : LoadPairPostIdx<0b00, 0, GPR32, simm7s4, "ldp">;
998 def LDPXpost : LoadPairPostIdx<0b10, 0, GPR64, simm7s8, "ldp">;
999 def LDPSpost : LoadPairPostIdx<0b00, 1, FPR32, simm7s4, "ldp">;
1000 def LDPDpost : LoadPairPostIdx<0b01, 1, FPR64, simm7s8, "ldp">;
1001 def LDPQpost : LoadPairPostIdx<0b10, 1, FPR128, simm7s16, "ldp">;
1003 def LDPSWpost : LoadPairPostIdx<0b01, 0, GPR64, simm7s4, "ldpsw">;
1006 // Pair (no allocate)
1007 def LDNPWi : LoadPairNoAlloc<0b00, 0, GPR32, am_indexed32simm7, "ldnp">;
1008 def LDNPXi : LoadPairNoAlloc<0b10, 0, GPR64, am_indexed64simm7, "ldnp">;
1009 def LDNPSi : LoadPairNoAlloc<0b00, 1, FPR32, am_indexed32simm7, "ldnp">;
1010 def LDNPDi : LoadPairNoAlloc<0b01, 1, FPR64, am_indexed64simm7, "ldnp">;
1011 def LDNPQi : LoadPairNoAlloc<0b10, 1, FPR128, am_indexed128simm7, "ldnp">;
1014 // (register offset)
1017 let AddedComplexity = 10 in {
1019 def LDRBBro : Load8RO<0b00, 0, 0b01, GPR32, "ldrb",
1020 [(set GPR32:$Rt, (zextloadi8 ro_indexed8:$addr))]>;
1021 def LDRHHro : Load16RO<0b01, 0, 0b01, GPR32, "ldrh",
1022 [(set GPR32:$Rt, (zextloadi16 ro_indexed16:$addr))]>;
1023 def LDRWro : Load32RO<0b10, 0, 0b01, GPR32, "ldr",
1024 [(set GPR32:$Rt, (load ro_indexed32:$addr))]>;
1025 def LDRXro : Load64RO<0b11, 0, 0b01, GPR64, "ldr",
1026 [(set GPR64:$Rt, (load ro_indexed64:$addr))]>;
1029 def LDRBro : Load8RO<0b00, 1, 0b01, FPR8, "ldr",
1030 [(set FPR8:$Rt, (load ro_indexed8:$addr))]>;
1031 def LDRHro : Load16RO<0b01, 1, 0b01, FPR16, "ldr",
1032 [(set (f16 FPR16:$Rt), (load ro_indexed16:$addr))]>;
1033 def LDRSro : Load32RO<0b10, 1, 0b01, FPR32, "ldr",
1034 [(set (f32 FPR32:$Rt), (load ro_indexed32:$addr))]>;
1035 def LDRDro : Load64RO<0b11, 1, 0b01, FPR64, "ldr",
1036 [(set (f64 FPR64:$Rt), (load ro_indexed64:$addr))]>;
1037 def LDRQro : Load128RO<0b00, 1, 0b11, FPR128, "ldr", []> {
1041 // For regular load, we do not have any alignment requirement.
1042 // Thus, it is safe to directly map the vector loads with interesting
1043 // addressing modes.
1044 // FIXME: We could do the same for bitconvert to floating point vectors.
1045 def : Pat <(v8i8 (scalar_to_vector (i32 (extloadi8 ro_indexed8:$addr)))),
1046 (INSERT_SUBREG (v8i8 (IMPLICIT_DEF)),
1047 (LDRBro ro_indexed8:$addr), bsub)>;
1048 def : Pat <(v16i8 (scalar_to_vector (i32 (extloadi8 ro_indexed8:$addr)))),
1049 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
1050 (LDRBro ro_indexed8:$addr), bsub)>;
1051 def : Pat <(v4i16 (scalar_to_vector (i32 (extloadi16 ro_indexed16:$addr)))),
1052 (INSERT_SUBREG (v4i16 (IMPLICIT_DEF)),
1053 (LDRHro ro_indexed16:$addr), hsub)>;
1054 def : Pat <(v8i16 (scalar_to_vector (i32 (extloadi16 ro_indexed16:$addr)))),
1055 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
1056 (LDRHro ro_indexed16:$addr), hsub)>;
1057 def : Pat <(v2i32 (scalar_to_vector (i32 (load ro_indexed32:$addr)))),
1058 (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)),
1059 (LDRSro ro_indexed32:$addr), ssub)>;
1060 def : Pat <(v4i32 (scalar_to_vector (i32 (load ro_indexed32:$addr)))),
1061 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
1062 (LDRSro ro_indexed32:$addr), ssub)>;
1063 def : Pat <(v1i64 (scalar_to_vector (i64 (load ro_indexed64:$addr)))),
1064 (LDRDro ro_indexed64:$addr)>;
1065 def : Pat <(v2i64 (scalar_to_vector (i64 (load ro_indexed64:$addr)))),
1066 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
1067 (LDRDro ro_indexed64:$addr), dsub)>;
1069 // Match all load 64 bits width whose type is compatible with FPR64
1070 def : Pat<(v2f32 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1071 def : Pat<(v1f64 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1072 def : Pat<(v8i8 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1073 def : Pat<(v4i16 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1074 def : Pat<(v2i32 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1075 def : Pat<(v1i64 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1077 // Match all load 128 bits width whose type is compatible with FPR128
1078 def : Pat<(v4f32 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1079 def : Pat<(v2f64 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1080 def : Pat<(v16i8 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1081 def : Pat<(v8i16 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1082 def : Pat<(v4i32 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1083 def : Pat<(v2i64 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1084 def : Pat<(f128 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1086 // Load sign-extended half-word
1087 def LDRSHWro : Load16RO<0b01, 0, 0b11, GPR32, "ldrsh",
1088 [(set GPR32:$Rt, (sextloadi16 ro_indexed16:$addr))]>;
1089 def LDRSHXro : Load16RO<0b01, 0, 0b10, GPR64, "ldrsh",
1090 [(set GPR64:$Rt, (sextloadi16 ro_indexed16:$addr))]>;
1092 // Load sign-extended byte
1093 def LDRSBWro : Load8RO<0b00, 0, 0b11, GPR32, "ldrsb",
1094 [(set GPR32:$Rt, (sextloadi8 ro_indexed8:$addr))]>;
1095 def LDRSBXro : Load8RO<0b00, 0, 0b10, GPR64, "ldrsb",
1096 [(set GPR64:$Rt, (sextloadi8 ro_indexed8:$addr))]>;
1098 // Load sign-extended word
1099 def LDRSWro : Load32RO<0b10, 0, 0b10, GPR64, "ldrsw",
1100 [(set GPR64:$Rt, (sextloadi32 ro_indexed32:$addr))]>;
1103 def PRFMro : PrefetchRO<0b11, 0, 0b10, "prfm",
1104 [(ARM64Prefetch imm:$Rt, ro_indexed64:$addr)]>;
1107 def : Pat<(i64 (zextloadi8 ro_indexed8:$addr)),
1108 (SUBREG_TO_REG (i64 0), (LDRBBro ro_indexed8:$addr), sub_32)>;
1109 def : Pat<(i64 (zextloadi16 ro_indexed16:$addr)),
1110 (SUBREG_TO_REG (i64 0), (LDRHHro ro_indexed16:$addr), sub_32)>;
1111 def : Pat<(i64 (zextloadi32 ro_indexed32:$addr)),
1112 (SUBREG_TO_REG (i64 0), (LDRWro ro_indexed32:$addr), sub_32)>;
1114 // zextloadi1 -> zextloadi8
1115 def : Pat<(i32 (zextloadi1 ro_indexed8:$addr)), (LDRBBro ro_indexed8:$addr)>;
1116 def : Pat<(i64 (zextloadi1 ro_indexed8:$addr)),
1117 (SUBREG_TO_REG (i64 0), (LDRBBro ro_indexed8:$addr), sub_32)>;
1119 // extload -> zextload
1120 def : Pat<(i32 (extloadi16 ro_indexed16:$addr)), (LDRHHro ro_indexed16:$addr)>;
1121 def : Pat<(i32 (extloadi8 ro_indexed8:$addr)), (LDRBBro ro_indexed8:$addr)>;
1122 def : Pat<(i32 (extloadi1 ro_indexed8:$addr)), (LDRBBro ro_indexed8:$addr)>;
1123 def : Pat<(i64 (extloadi32 ro_indexed32:$addr)),
1124 (SUBREG_TO_REG (i64 0), (LDRWro ro_indexed32:$addr), sub_32)>;
1125 def : Pat<(i64 (extloadi16 ro_indexed16:$addr)),
1126 (SUBREG_TO_REG (i64 0), (LDRHHro ro_indexed16:$addr), sub_32)>;
1127 def : Pat<(i64 (extloadi8 ro_indexed8:$addr)),
1128 (SUBREG_TO_REG (i64 0), (LDRBBro ro_indexed8:$addr), sub_32)>;
1129 def : Pat<(i64 (extloadi1 ro_indexed8:$addr)),
1130 (SUBREG_TO_REG (i64 0), (LDRBBro ro_indexed8:$addr), sub_32)>;
1132 } // AddedComplexity = 10
1135 // (unsigned immediate)
1137 def LDRXui : LoadUI<0b11, 0, 0b01, GPR64, am_indexed64, "ldr",
1138 [(set GPR64:$Rt, (load am_indexed64:$addr))]>;
1139 def LDRWui : LoadUI<0b10, 0, 0b01, GPR32, am_indexed32, "ldr",
1140 [(set GPR32:$Rt, (load am_indexed32:$addr))]>;
1141 def LDRBui : LoadUI<0b00, 1, 0b01, FPR8, am_indexed8, "ldr",
1142 [(set FPR8:$Rt, (load am_indexed8:$addr))]>;
1143 def LDRHui : LoadUI<0b01, 1, 0b01, FPR16, am_indexed16, "ldr",
1144 [(set (f16 FPR16:$Rt), (load am_indexed16:$addr))]>;
1145 def LDRSui : LoadUI<0b10, 1, 0b01, FPR32, am_indexed32, "ldr",
1146 [(set (f32 FPR32:$Rt), (load am_indexed32:$addr))]>;
1147 def LDRDui : LoadUI<0b11, 1, 0b01, FPR64, am_indexed64, "ldr",
1148 [(set (f64 FPR64:$Rt), (load am_indexed64:$addr))]>;
1149 def LDRQui : LoadUI<0b00, 1, 0b11, FPR128, am_indexed128, "ldr",
1150 [(set (f128 FPR128:$Rt), (load am_indexed128:$addr))]>;
1152 // For regular load, we do not have any alignment requirement.
1153 // Thus, it is safe to directly map the vector loads with interesting
1154 // addressing modes.
1155 // FIXME: We could do the same for bitconvert to floating point vectors.
1156 def : Pat <(v8i8 (scalar_to_vector (i32 (extloadi8 am_indexed8:$addr)))),
1157 (INSERT_SUBREG (v8i8 (IMPLICIT_DEF)),
1158 (LDRBui am_indexed8:$addr), bsub)>;
1159 def : Pat <(v16i8 (scalar_to_vector (i32 (extloadi8 am_indexed8:$addr)))),
1160 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
1161 (LDRBui am_indexed8:$addr), bsub)>;
1162 def : Pat <(v4i16 (scalar_to_vector (i32 (extloadi16 am_indexed16:$addr)))),
1163 (INSERT_SUBREG (v4i16 (IMPLICIT_DEF)),
1164 (LDRHui am_indexed16:$addr), hsub)>;
1165 def : Pat <(v8i16 (scalar_to_vector (i32 (extloadi16 am_indexed16:$addr)))),
1166 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
1167 (LDRHui am_indexed16:$addr), hsub)>;
1168 def : Pat <(v2i32 (scalar_to_vector (i32 (load am_indexed32:$addr)))),
1169 (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)),
1170 (LDRSui am_indexed32:$addr), ssub)>;
1171 def : Pat <(v4i32 (scalar_to_vector (i32 (load am_indexed32:$addr)))),
1172 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
1173 (LDRSui am_indexed32:$addr), ssub)>;
1174 def : Pat <(v1i64 (scalar_to_vector (i64 (load am_indexed64:$addr)))),
1175 (LDRDui am_indexed64:$addr)>;
1176 def : Pat <(v2i64 (scalar_to_vector (i64 (load am_indexed64:$addr)))),
1177 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
1178 (LDRDui am_indexed64:$addr), dsub)>;
1180 // Match all load 64 bits width whose type is compatible with FPR64
1181 def : Pat<(v2f32 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1182 def : Pat<(v1f64 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1183 def : Pat<(v8i8 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1184 def : Pat<(v4i16 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1185 def : Pat<(v2i32 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1186 def : Pat<(v1i64 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1188 // Match all load 128 bits width whose type is compatible with FPR128
1189 def : Pat<(v4f32 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1190 def : Pat<(v2f64 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1191 def : Pat<(v16i8 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1192 def : Pat<(v8i16 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1193 def : Pat<(v4i32 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1194 def : Pat<(v2i64 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1195 def : Pat<(f128 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1197 def LDRHHui : LoadUI<0b01, 0, 0b01, GPR32, am_indexed16, "ldrh",
1198 [(set GPR32:$Rt, (zextloadi16 am_indexed16:$addr))]>;
1199 def LDRBBui : LoadUI<0b00, 0, 0b01, GPR32, am_indexed8, "ldrb",
1200 [(set GPR32:$Rt, (zextloadi8 am_indexed8:$addr))]>;
1202 def : Pat<(i64 (zextloadi8 am_indexed8:$addr)),
1203 (SUBREG_TO_REG (i64 0), (LDRBBui am_indexed8:$addr), sub_32)>;
1204 def : Pat<(i64 (zextloadi16 am_indexed16:$addr)),
1205 (SUBREG_TO_REG (i64 0), (LDRHHui am_indexed16:$addr), sub_32)>;
1207 // zextloadi1 -> zextloadi8
1208 def : Pat<(i32 (zextloadi1 am_indexed8:$addr)), (LDRBBui am_indexed8:$addr)>;
1209 def : Pat<(i64 (zextloadi1 am_indexed8:$addr)),
1210 (SUBREG_TO_REG (i64 0), (LDRBBui am_indexed8:$addr), sub_32)>;
1212 // extload -> zextload
1213 def : Pat<(i32 (extloadi16 am_indexed16:$addr)), (LDRHHui am_indexed16:$addr)>;
1214 def : Pat<(i32 (extloadi8 am_indexed8:$addr)), (LDRBBui am_indexed8:$addr)>;
1215 def : Pat<(i32 (extloadi1 am_indexed8:$addr)), (LDRBBui am_indexed8:$addr)>;
1216 def : Pat<(i64 (extloadi32 am_indexed32:$addr)),
1217 (SUBREG_TO_REG (i64 0), (LDRWui am_indexed32:$addr), sub_32)>;
1218 def : Pat<(i64 (extloadi16 am_indexed16:$addr)),
1219 (SUBREG_TO_REG (i64 0), (LDRHHui am_indexed16:$addr), sub_32)>;
1220 def : Pat<(i64 (extloadi8 am_indexed8:$addr)),
1221 (SUBREG_TO_REG (i64 0), (LDRBBui am_indexed8:$addr), sub_32)>;
1222 def : Pat<(i64 (extloadi1 am_indexed8:$addr)),
1223 (SUBREG_TO_REG (i64 0), (LDRBBui am_indexed8:$addr), sub_32)>;
1225 // load sign-extended half-word
1226 def LDRSHWui : LoadUI<0b01, 0, 0b11, GPR32, am_indexed16, "ldrsh",
1227 [(set GPR32:$Rt, (sextloadi16 am_indexed16:$addr))]>;
1228 def LDRSHXui : LoadUI<0b01, 0, 0b10, GPR64, am_indexed16, "ldrsh",
1229 [(set GPR64:$Rt, (sextloadi16 am_indexed16:$addr))]>;
1231 // load sign-extended byte
1232 def LDRSBWui : LoadUI<0b00, 0, 0b11, GPR32, am_indexed8, "ldrsb",
1233 [(set GPR32:$Rt, (sextloadi8 am_indexed8:$addr))]>;
1234 def LDRSBXui : LoadUI<0b00, 0, 0b10, GPR64, am_indexed8, "ldrsb",
1235 [(set GPR64:$Rt, (sextloadi8 am_indexed8:$addr))]>;
1237 // load sign-extended word
1238 def LDRSWui : LoadUI<0b10, 0, 0b10, GPR64, am_indexed32, "ldrsw",
1239 [(set GPR64:$Rt, (sextloadi32 am_indexed32:$addr))]>;
1241 // load zero-extended word
1242 def : Pat<(i64 (zextloadi32 am_indexed32:$addr)),
1243 (SUBREG_TO_REG (i64 0), (LDRWui am_indexed32:$addr), sub_32)>;
1246 def PRFMui : PrefetchUI<0b11, 0, 0b10, "prfm",
1247 [(ARM64Prefetch imm:$Rt, am_indexed64:$addr)]>;
1251 def LDRWl : LoadLiteral<0b00, 0, GPR32, "ldr">;
1252 def LDRXl : LoadLiteral<0b01, 0, GPR64, "ldr">;
1253 def LDRSl : LoadLiteral<0b00, 1, FPR32, "ldr">;
1254 def LDRDl : LoadLiteral<0b01, 1, FPR64, "ldr">;
1255 def LDRQl : LoadLiteral<0b10, 1, FPR128, "ldr">;
1257 // load sign-extended word
1258 def LDRSWl : LoadLiteral<0b10, 0, GPR64, "ldrsw">;
1261 def PRFMl : PrefetchLiteral<0b11, 0, "prfm", []>;
1262 // [(ARM64Prefetch imm:$Rt, tglobaladdr:$label)]>;
1265 // (unscaled immediate)
1266 def LDURXi : LoadUnscaled<0b11, 0, 0b01, GPR64, am_unscaled64, "ldur",
1267 [(set GPR64:$Rt, (load am_unscaled64:$addr))]>;
1268 def LDURWi : LoadUnscaled<0b10, 0, 0b01, GPR32, am_unscaled32, "ldur",
1269 [(set GPR32:$Rt, (load am_unscaled32:$addr))]>;
1270 def LDURBi : LoadUnscaled<0b00, 1, 0b01, FPR8, am_unscaled8, "ldur",
1271 [(set FPR8:$Rt, (load am_unscaled8:$addr))]>;
1272 def LDURHi : LoadUnscaled<0b01, 1, 0b01, FPR16, am_unscaled16, "ldur",
1273 [(set (f16 FPR16:$Rt), (load am_unscaled16:$addr))]>;
1274 def LDURSi : LoadUnscaled<0b10, 1, 0b01, FPR32, am_unscaled32, "ldur",
1275 [(set (f32 FPR32:$Rt), (load am_unscaled32:$addr))]>;
1276 def LDURDi : LoadUnscaled<0b11, 1, 0b01, FPR64, am_unscaled64, "ldur",
1277 [(set (f64 FPR64:$Rt), (load am_unscaled64:$addr))]>;
1278 def LDURQi : LoadUnscaled<0b00, 1, 0b11, FPR128, am_unscaled128, "ldur",
1279 [(set (v2f64 FPR128:$Rt), (load am_unscaled128:$addr))]>;
1282 : LoadUnscaled<0b01, 0, 0b01, GPR32, am_unscaled16, "ldurh",
1283 [(set GPR32:$Rt, (zextloadi16 am_unscaled16:$addr))]>;
1285 : LoadUnscaled<0b00, 0, 0b01, GPR32, am_unscaled8, "ldurb",
1286 [(set GPR32:$Rt, (zextloadi8 am_unscaled8:$addr))]>;
1288 // Match all load 64 bits width whose type is compatible with FPR64
1289 def : Pat<(v2f32 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1290 def : Pat<(v1f64 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1291 def : Pat<(v8i8 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1292 def : Pat<(v4i16 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1293 def : Pat<(v2i32 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1294 def : Pat<(v1i64 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1296 // Match all load 128 bits width whose type is compatible with FPR128
1297 def : Pat<(v4f32 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1298 def : Pat<(v2f64 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1299 def : Pat<(v16i8 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1300 def : Pat<(v8i16 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1301 def : Pat<(v4i32 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1302 def : Pat<(v2i64 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1303 def : Pat<(f128 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1306 def : Pat<(i32 (extloadi16 am_unscaled16:$addr)), (LDURHHi am_unscaled16:$addr)>;
1307 def : Pat<(i32 (extloadi8 am_unscaled8:$addr)), (LDURBBi am_unscaled8:$addr)>;
1308 def : Pat<(i32 (extloadi1 am_unscaled8:$addr)), (LDURBBi am_unscaled8:$addr)>;
1309 def : Pat<(i64 (extloadi32 am_unscaled32:$addr)),
1310 (SUBREG_TO_REG (i64 0), (LDURWi am_unscaled32:$addr), sub_32)>;
1311 def : Pat<(i64 (extloadi16 am_unscaled16:$addr)),
1312 (SUBREG_TO_REG (i64 0), (LDURHHi am_unscaled16:$addr), sub_32)>;
1313 def : Pat<(i64 (extloadi8 am_unscaled8:$addr)),
1314 (SUBREG_TO_REG (i64 0), (LDURBBi am_unscaled8:$addr), sub_32)>;
1315 def : Pat<(i64 (extloadi1 am_unscaled8:$addr)),
1316 (SUBREG_TO_REG (i64 0), (LDURBBi am_unscaled8:$addr), sub_32)>;
1318 def : Pat<(i32 (zextloadi16 am_unscaled16:$addr)),
1319 (LDURHHi am_unscaled16:$addr)>;
1320 def : Pat<(i32 (zextloadi8 am_unscaled8:$addr)),
1321 (LDURBBi am_unscaled8:$addr)>;
1322 def : Pat<(i32 (zextloadi1 am_unscaled8:$addr)),
1323 (LDURBBi am_unscaled8:$addr)>;
1324 def : Pat<(i64 (zextloadi32 am_unscaled32:$addr)),
1325 (SUBREG_TO_REG (i64 0), (LDURWi am_unscaled32:$addr), sub_32)>;
1326 def : Pat<(i64 (zextloadi16 am_unscaled16:$addr)),
1327 (SUBREG_TO_REG (i64 0), (LDURHHi am_unscaled16:$addr), sub_32)>;
1328 def : Pat<(i64 (zextloadi8 am_unscaled8:$addr)),
1329 (SUBREG_TO_REG (i64 0), (LDURBBi am_unscaled8:$addr), sub_32)>;
1330 def : Pat<(i64 (zextloadi1 am_unscaled8:$addr)),
1331 (SUBREG_TO_REG (i64 0), (LDURBBi am_unscaled8:$addr), sub_32)>;
1335 // LDR mnemonics fall back to LDUR for negative or unaligned offsets.
1337 // Define new assembler match classes as we want to only match these when
1338 // the don't otherwise match the scaled addressing mode for LDR/STR. Don't
1339 // associate a DiagnosticType either, as we want the diagnostic for the
1340 // canonical form (the scaled operand) to take precedence.
1341 def MemoryUnscaledFB8Operand : AsmOperandClass {
1342 let Name = "MemoryUnscaledFB8";
1343 let RenderMethod = "addMemoryUnscaledOperands";
1345 def MemoryUnscaledFB16Operand : AsmOperandClass {
1346 let Name = "MemoryUnscaledFB16";
1347 let RenderMethod = "addMemoryUnscaledOperands";
1349 def MemoryUnscaledFB32Operand : AsmOperandClass {
1350 let Name = "MemoryUnscaledFB32";
1351 let RenderMethod = "addMemoryUnscaledOperands";
1353 def MemoryUnscaledFB64Operand : AsmOperandClass {
1354 let Name = "MemoryUnscaledFB64";
1355 let RenderMethod = "addMemoryUnscaledOperands";
1357 def MemoryUnscaledFB128Operand : AsmOperandClass {
1358 let Name = "MemoryUnscaledFB128";
1359 let RenderMethod = "addMemoryUnscaledOperands";
1361 def am_unscaled_fb8 : Operand<i64> {
1362 let ParserMatchClass = MemoryUnscaledFB8Operand;
1363 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
1365 def am_unscaled_fb16 : Operand<i64> {
1366 let ParserMatchClass = MemoryUnscaledFB16Operand;
1367 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
1369 def am_unscaled_fb32 : Operand<i64> {
1370 let ParserMatchClass = MemoryUnscaledFB32Operand;
1371 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
1373 def am_unscaled_fb64 : Operand<i64> {
1374 let ParserMatchClass = MemoryUnscaledFB64Operand;
1375 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
1377 def am_unscaled_fb128 : Operand<i64> {
1378 let ParserMatchClass = MemoryUnscaledFB128Operand;
1379 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
1381 def : InstAlias<"ldr $Rt, $addr", (LDURXi GPR64:$Rt, am_unscaled_fb64:$addr)>;
1382 def : InstAlias<"ldr $Rt, $addr", (LDURWi GPR32:$Rt, am_unscaled_fb32:$addr)>;
1383 def : InstAlias<"ldr $Rt, $addr", (LDURBi FPR8:$Rt, am_unscaled_fb8:$addr)>;
1384 def : InstAlias<"ldr $Rt, $addr", (LDURHi FPR16:$Rt, am_unscaled_fb16:$addr)>;
1385 def : InstAlias<"ldr $Rt, $addr", (LDURSi FPR32:$Rt, am_unscaled_fb32:$addr)>;
1386 def : InstAlias<"ldr $Rt, $addr", (LDURDi FPR64:$Rt, am_unscaled_fb64:$addr)>;
1387 def : InstAlias<"ldr $Rt, $addr", (LDURQi FPR128:$Rt, am_unscaled_fb128:$addr)>;
1390 def : Pat<(i64 (zextloadi8 am_unscaled8:$addr)),
1391 (SUBREG_TO_REG (i64 0), (LDURBBi am_unscaled8:$addr), sub_32)>;
1392 def : Pat<(i64 (zextloadi16 am_unscaled16:$addr)),
1393 (SUBREG_TO_REG (i64 0), (LDURHHi am_unscaled16:$addr), sub_32)>;
1395 // load sign-extended half-word
1397 : LoadUnscaled<0b01, 0, 0b11, GPR32, am_unscaled16, "ldursh",
1398 [(set GPR32:$Rt, (sextloadi16 am_unscaled16:$addr))]>;
1400 : LoadUnscaled<0b01, 0, 0b10, GPR64, am_unscaled16, "ldursh",
1401 [(set GPR64:$Rt, (sextloadi16 am_unscaled16:$addr))]>;
1403 // load sign-extended byte
1405 : LoadUnscaled<0b00, 0, 0b11, GPR32, am_unscaled8, "ldursb",
1406 [(set GPR32:$Rt, (sextloadi8 am_unscaled8:$addr))]>;
1408 : LoadUnscaled<0b00, 0, 0b10, GPR64, am_unscaled8, "ldursb",
1409 [(set GPR64:$Rt, (sextloadi8 am_unscaled8:$addr))]>;
1411 // load sign-extended word
1413 : LoadUnscaled<0b10, 0, 0b10, GPR64, am_unscaled32, "ldursw",
1414 [(set GPR64:$Rt, (sextloadi32 am_unscaled32:$addr))]>;
1416 // zero and sign extending aliases from generic LDR* mnemonics to LDUR*.
1417 def : InstAlias<"ldrb $Rt, $addr", (LDURBBi GPR32:$Rt, am_unscaled_fb8:$addr)>;
1418 def : InstAlias<"ldrh $Rt, $addr", (LDURHHi GPR32:$Rt, am_unscaled_fb16:$addr)>;
1419 def : InstAlias<"ldrsb $Rt, $addr", (LDURSBWi GPR32:$Rt, am_unscaled_fb8:$addr)>;
1420 def : InstAlias<"ldrsb $Rt, $addr", (LDURSBXi GPR64:$Rt, am_unscaled_fb8:$addr)>;
1421 def : InstAlias<"ldrsh $Rt, $addr", (LDURSHWi GPR32:$Rt, am_unscaled_fb16:$addr)>;
1422 def : InstAlias<"ldrsh $Rt, $addr", (LDURSHXi GPR64:$Rt, am_unscaled_fb16:$addr)>;
1423 def : InstAlias<"ldrsw $Rt, $addr", (LDURSWi GPR64:$Rt, am_unscaled_fb32:$addr)>;
1426 def PRFUMi : PrefetchUnscaled<0b11, 0, 0b10, "prfum",
1427 [(ARM64Prefetch imm:$Rt, am_unscaled64:$addr)]>;
1430 // (unscaled immediate, unprivileged)
1431 def LDTRXi : LoadUnprivileged<0b11, 0, 0b01, GPR64, "ldtr">;
1432 def LDTRWi : LoadUnprivileged<0b10, 0, 0b01, GPR32, "ldtr">;
1434 def LDTRHi : LoadUnprivileged<0b01, 0, 0b01, GPR32, "ldtrh">;
1435 def LDTRBi : LoadUnprivileged<0b00, 0, 0b01, GPR32, "ldtrb">;
1437 // load sign-extended half-word
1438 def LDTRSHWi : LoadUnprivileged<0b01, 0, 0b11, GPR32, "ldtrsh">;
1439 def LDTRSHXi : LoadUnprivileged<0b01, 0, 0b10, GPR64, "ldtrsh">;
1441 // load sign-extended byte
1442 def LDTRSBWi : LoadUnprivileged<0b00, 0, 0b11, GPR32, "ldtrsb">;
1443 def LDTRSBXi : LoadUnprivileged<0b00, 0, 0b10, GPR64, "ldtrsb">;
1445 // load sign-extended word
1446 def LDTRSWi : LoadUnprivileged<0b10, 0, 0b10, GPR64, "ldtrsw">;
1449 // (immediate pre-indexed)
1450 def LDRWpre : LoadPreIdx<0b10, 0, 0b01, GPR32, "ldr">;
1451 def LDRXpre : LoadPreIdx<0b11, 0, 0b01, GPR64, "ldr">;
1452 def LDRBpre : LoadPreIdx<0b00, 1, 0b01, FPR8, "ldr">;
1453 def LDRHpre : LoadPreIdx<0b01, 1, 0b01, FPR16, "ldr">;
1454 def LDRSpre : LoadPreIdx<0b10, 1, 0b01, FPR32, "ldr">;
1455 def LDRDpre : LoadPreIdx<0b11, 1, 0b01, FPR64, "ldr">;
1456 def LDRQpre : LoadPreIdx<0b00, 1, 0b11, FPR128, "ldr">;
1458 // load sign-extended half-word
1459 def LDRSHWpre : LoadPreIdx<0b01, 0, 0b11, GPR32, "ldrsh">;
1460 def LDRSHXpre : LoadPreIdx<0b01, 0, 0b10, GPR64, "ldrsh">;
1462 // load sign-extended byte
1463 def LDRSBWpre : LoadPreIdx<0b00, 0, 0b11, GPR32, "ldrsb">;
1464 def LDRSBXpre : LoadPreIdx<0b00, 0, 0b10, GPR64, "ldrsb">;
1466 // load zero-extended byte
1467 def LDRBBpre : LoadPreIdx<0b00, 0, 0b01, GPR32, "ldrb">;
1468 def LDRHHpre : LoadPreIdx<0b01, 0, 0b01, GPR32, "ldrh">;
1470 // load sign-extended word
1471 def LDRSWpre : LoadPreIdx<0b10, 0, 0b10, GPR64, "ldrsw">;
1473 // ISel pseudos and patterns. See expanded comment on LoadPreIdxPseudo.
1474 def LDRDpre_isel : LoadPreIdxPseudo<FPR64>;
1475 def LDRSpre_isel : LoadPreIdxPseudo<FPR32>;
1476 def LDRXpre_isel : LoadPreIdxPseudo<GPR64>;
1477 def LDRWpre_isel : LoadPreIdxPseudo<GPR32>;
1478 def LDRHHpre_isel : LoadPreIdxPseudo<GPR32>;
1479 def LDRBBpre_isel : LoadPreIdxPseudo<GPR32>;
1481 def LDRSWpre_isel : LoadPreIdxPseudo<GPR64>;
1482 def LDRSHWpre_isel : LoadPreIdxPseudo<GPR32>;
1483 def LDRSHXpre_isel : LoadPreIdxPseudo<GPR64>;
1484 def LDRSBWpre_isel : LoadPreIdxPseudo<GPR32>;
1485 def LDRSBXpre_isel : LoadPreIdxPseudo<GPR64>;
1488 // (immediate post-indexed)
1489 def LDRWpost : LoadPostIdx<0b10, 0, 0b01, GPR32, "ldr">;
1490 def LDRXpost : LoadPostIdx<0b11, 0, 0b01, GPR64, "ldr">;
1491 def LDRBpost : LoadPostIdx<0b00, 1, 0b01, FPR8, "ldr">;
1492 def LDRHpost : LoadPostIdx<0b01, 1, 0b01, FPR16, "ldr">;
1493 def LDRSpost : LoadPostIdx<0b10, 1, 0b01, FPR32, "ldr">;
1494 def LDRDpost : LoadPostIdx<0b11, 1, 0b01, FPR64, "ldr">;
1495 def LDRQpost : LoadPostIdx<0b00, 1, 0b11, FPR128, "ldr">;
1497 // load sign-extended half-word
1498 def LDRSHWpost : LoadPostIdx<0b01, 0, 0b11, GPR32, "ldrsh">;
1499 def LDRSHXpost : LoadPostIdx<0b01, 0, 0b10, GPR64, "ldrsh">;
1501 // load sign-extended byte
1502 def LDRSBWpost : LoadPostIdx<0b00, 0, 0b11, GPR32, "ldrsb">;
1503 def LDRSBXpost : LoadPostIdx<0b00, 0, 0b10, GPR64, "ldrsb">;
1505 // load zero-extended byte
1506 def LDRBBpost : LoadPostIdx<0b00, 0, 0b01, GPR32, "ldrb">;
1507 def LDRHHpost : LoadPostIdx<0b01, 0, 0b01, GPR32, "ldrh">;
1509 // load sign-extended word
1510 def LDRSWpost : LoadPostIdx<0b10, 0, 0b10, GPR64, "ldrsw">;
1512 // ISel pseudos and patterns. See expanded comment on LoadPostIdxPseudo.
1513 def LDRDpost_isel : LoadPostIdxPseudo<FPR64>;
1514 def LDRSpost_isel : LoadPostIdxPseudo<FPR32>;
1515 def LDRXpost_isel : LoadPostIdxPseudo<GPR64>;
1516 def LDRWpost_isel : LoadPostIdxPseudo<GPR32>;
1517 def LDRHHpost_isel : LoadPostIdxPseudo<GPR32>;
1518 def LDRBBpost_isel : LoadPostIdxPseudo<GPR32>;
1520 def LDRSWpost_isel : LoadPostIdxPseudo<GPR64>;
1521 def LDRSHWpost_isel : LoadPostIdxPseudo<GPR32>;
1522 def LDRSHXpost_isel : LoadPostIdxPseudo<GPR64>;
1523 def LDRSBWpost_isel : LoadPostIdxPseudo<GPR32>;
1524 def LDRSBXpost_isel : LoadPostIdxPseudo<GPR64>;
1526 //===----------------------------------------------------------------------===//
1527 // Store instructions.
1528 //===----------------------------------------------------------------------===//
1530 // Pair (indexed, offset)
1531 // FIXME: Use dedicated range-checked addressing mode operand here.
1532 def STPWi : StorePairOffset<0b00, 0, GPR32, am_indexed32simm7, "stp">;
1533 def STPXi : StorePairOffset<0b10, 0, GPR64, am_indexed64simm7, "stp">;
1534 def STPSi : StorePairOffset<0b00, 1, FPR32, am_indexed32simm7, "stp">;
1535 def STPDi : StorePairOffset<0b01, 1, FPR64, am_indexed64simm7, "stp">;
1536 def STPQi : StorePairOffset<0b10, 1, FPR128, am_indexed128simm7, "stp">;
1538 // Pair (pre-indexed)
1539 def STPWpre : StorePairPreIdx<0b00, 0, GPR32, am_indexed32simm7_wb, "stp">;
1540 def STPXpre : StorePairPreIdx<0b10, 0, GPR64, am_indexed64simm7_wb, "stp">;
1541 def STPSpre : StorePairPreIdx<0b00, 1, FPR32, am_indexed32simm7_wb, "stp">;
1542 def STPDpre : StorePairPreIdx<0b01, 1, FPR64, am_indexed64simm7_wb, "stp">;
1543 def STPQpre : StorePairPreIdx<0b10, 1, FPR128, am_indexed128simm7_wb, "stp">;
1545 // Pair (pre-indexed)
1546 def STPWpost : StorePairPostIdx<0b00, 0, GPR32, simm7s4, "stp">;
1547 def STPXpost : StorePairPostIdx<0b10, 0, GPR64, simm7s8, "stp">;
1548 def STPSpost : StorePairPostIdx<0b00, 1, FPR32, simm7s4, "stp">;
1549 def STPDpost : StorePairPostIdx<0b01, 1, FPR64, simm7s8, "stp">;
1550 def STPQpost : StorePairPostIdx<0b10, 1, FPR128, simm7s16, "stp">;
1552 // Pair (no allocate)
1553 def STNPWi : StorePairNoAlloc<0b00, 0, GPR32, am_indexed32simm7, "stnp">;
1554 def STNPXi : StorePairNoAlloc<0b10, 0, GPR64, am_indexed64simm7, "stnp">;
1555 def STNPSi : StorePairNoAlloc<0b00, 1, FPR32, am_indexed32simm7, "stnp">;
1556 def STNPDi : StorePairNoAlloc<0b01, 1, FPR64, am_indexed64simm7, "stnp">;
1557 def STNPQi : StorePairNoAlloc<0b10, 1, FPR128, am_indexed128simm7, "stnp">;
1560 // (Register offset)
1562 let AddedComplexity = 10 in {
1565 def STRHHro : Store16RO<0b01, 0, 0b00, GPR32, "strh",
1566 [(truncstorei16 GPR32:$Rt, ro_indexed16:$addr)]>;
1567 def STRBBro : Store8RO<0b00, 0, 0b00, GPR32, "strb",
1568 [(truncstorei8 GPR32:$Rt, ro_indexed8:$addr)]>;
1569 def STRWro : Store32RO<0b10, 0, 0b00, GPR32, "str",
1570 [(store GPR32:$Rt, ro_indexed32:$addr)]>;
1571 def STRXro : Store64RO<0b11, 0, 0b00, GPR64, "str",
1572 [(store GPR64:$Rt, ro_indexed64:$addr)]>;
1575 def : Pat<(truncstorei8 GPR64:$Rt, ro_indexed8:$addr),
1576 (STRBBro (EXTRACT_SUBREG GPR64:$Rt, sub_32), ro_indexed8:$addr)>;
1577 def : Pat<(truncstorei16 GPR64:$Rt, ro_indexed16:$addr),
1578 (STRHHro (EXTRACT_SUBREG GPR64:$Rt, sub_32), ro_indexed16:$addr)>;
1579 def : Pat<(truncstorei32 GPR64:$Rt, ro_indexed32:$addr),
1580 (STRWro (EXTRACT_SUBREG GPR64:$Rt, sub_32), ro_indexed32:$addr)>;
1584 def STRBro : Store8RO<0b00, 1, 0b00, FPR8, "str",
1585 [(store FPR8:$Rt, ro_indexed8:$addr)]>;
1586 def STRHro : Store16RO<0b01, 1, 0b00, FPR16, "str",
1587 [(store (f16 FPR16:$Rt), ro_indexed16:$addr)]>;
1588 def STRSro : Store32RO<0b10, 1, 0b00, FPR32, "str",
1589 [(store (f32 FPR32:$Rt), ro_indexed32:$addr)]>;
1590 def STRDro : Store64RO<0b11, 1, 0b00, FPR64, "str",
1591 [(store (f64 FPR64:$Rt), ro_indexed64:$addr)]>;
1592 def STRQro : Store128RO<0b00, 1, 0b10, FPR128, "str", []> {
1596 // Match all store 64 bits width whose type is compatible with FPR64
1597 def : Pat<(store (v2f32 FPR64:$Rn), ro_indexed64:$addr),
1598 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1599 def : Pat<(store (v1f64 FPR64:$Rn), ro_indexed64:$addr),
1600 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1601 def : Pat<(store (v8i8 FPR64:$Rn), ro_indexed64:$addr),
1602 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1603 def : Pat<(store (v4i16 FPR64:$Rn), ro_indexed64:$addr),
1604 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1605 def : Pat<(store (v2i32 FPR64:$Rn), ro_indexed64:$addr),
1606 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1607 def : Pat<(store (v1i64 FPR64:$Rn), ro_indexed64:$addr),
1608 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1610 // Match all store 128 bits width whose type is compatible with FPR128
1611 def : Pat<(store (v4f32 FPR128:$Rn), ro_indexed128:$addr),
1612 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1613 def : Pat<(store (v2f64 FPR128:$Rn), ro_indexed128:$addr),
1614 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1615 def : Pat<(store (v16i8 FPR128:$Rn), ro_indexed128:$addr),
1616 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1617 def : Pat<(store (v8i16 FPR128:$Rn), ro_indexed128:$addr),
1618 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1619 def : Pat<(store (v4i32 FPR128:$Rn), ro_indexed128:$addr),
1620 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1621 def : Pat<(store (v2i64 FPR128:$Rn), ro_indexed128:$addr),
1622 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1623 def : Pat<(store (f128 FPR128:$Rn), ro_indexed128:$addr),
1624 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1627 // (unsigned immediate)
1628 def STRXui : StoreUI<0b11, 0, 0b00, GPR64, am_indexed64, "str",
1629 [(store GPR64:$Rt, am_indexed64:$addr)]>;
1630 def STRWui : StoreUI<0b10, 0, 0b00, GPR32, am_indexed32, "str",
1631 [(store GPR32:$Rt, am_indexed32:$addr)]>;
1632 def STRBui : StoreUI<0b00, 1, 0b00, FPR8, am_indexed8, "str",
1633 [(store FPR8:$Rt, am_indexed8:$addr)]>;
1634 def STRHui : StoreUI<0b01, 1, 0b00, FPR16, am_indexed16, "str",
1635 [(store (f16 FPR16:$Rt), am_indexed16:$addr)]>;
1636 def STRSui : StoreUI<0b10, 1, 0b00, FPR32, am_indexed32, "str",
1637 [(store (f32 FPR32:$Rt), am_indexed32:$addr)]>;
1638 def STRDui : StoreUI<0b11, 1, 0b00, FPR64, am_indexed64, "str",
1639 [(store (f64 FPR64:$Rt), am_indexed64:$addr)]>;
1640 def STRQui : StoreUI<0b00, 1, 0b10, FPR128, am_indexed128, "str", []> {
1644 // Match all store 64 bits width whose type is compatible with FPR64
1645 def : Pat<(store (v2f32 FPR64:$Rn), am_indexed64:$addr),
1646 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1647 def : Pat<(store (v1f64 FPR64:$Rn), am_indexed64:$addr),
1648 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1649 def : Pat<(store (v8i8 FPR64:$Rn), am_indexed64:$addr),
1650 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1651 def : Pat<(store (v4i16 FPR64:$Rn), am_indexed64:$addr),
1652 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1653 def : Pat<(store (v2i32 FPR64:$Rn), am_indexed64:$addr),
1654 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1655 def : Pat<(store (v1i64 FPR64:$Rn), am_indexed64:$addr),
1656 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1658 // Match all store 128 bits width whose type is compatible with FPR128
1659 def : Pat<(store (v4f32 FPR128:$Rn), am_indexed128:$addr),
1660 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1661 def : Pat<(store (v2f64 FPR128:$Rn), am_indexed128:$addr),
1662 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1663 def : Pat<(store (v16i8 FPR128:$Rn), am_indexed128:$addr),
1664 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1665 def : Pat<(store (v8i16 FPR128:$Rn), am_indexed128:$addr),
1666 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1667 def : Pat<(store (v4i32 FPR128:$Rn), am_indexed128:$addr),
1668 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1669 def : Pat<(store (v2i64 FPR128:$Rn), am_indexed128:$addr),
1670 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1671 def : Pat<(store (f128 FPR128:$Rn), am_indexed128:$addr),
1672 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1674 def STRHHui : StoreUI<0b01, 0, 0b00, GPR32, am_indexed16, "strh",
1675 [(truncstorei16 GPR32:$Rt, am_indexed16:$addr)]>;
1676 def STRBBui : StoreUI<0b00, 0, 0b00, GPR32, am_indexed8, "strb",
1677 [(truncstorei8 GPR32:$Rt, am_indexed8:$addr)]>;
1680 def : Pat<(truncstorei32 GPR64:$Rt, am_indexed32:$addr),
1681 (STRWui (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_indexed32:$addr)>;
1682 def : Pat<(truncstorei16 GPR64:$Rt, am_indexed16:$addr),
1683 (STRHHui (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_indexed16:$addr)>;
1684 def : Pat<(truncstorei8 GPR64:$Rt, am_indexed8:$addr),
1685 (STRBBui (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_indexed8:$addr)>;
1687 } // AddedComplexity = 10
1690 // (unscaled immediate)
1691 def STURXi : StoreUnscaled<0b11, 0, 0b00, GPR64, am_unscaled64, "stur",
1692 [(store GPR64:$Rt, am_unscaled64:$addr)]>;
1693 def STURWi : StoreUnscaled<0b10, 0, 0b00, GPR32, am_unscaled32, "stur",
1694 [(store GPR32:$Rt, am_unscaled32:$addr)]>;
1695 def STURBi : StoreUnscaled<0b00, 1, 0b00, FPR8, am_unscaled8, "stur",
1696 [(store FPR8:$Rt, am_unscaled8:$addr)]>;
1697 def STURHi : StoreUnscaled<0b01, 1, 0b00, FPR16, am_unscaled16, "stur",
1698 [(store (f16 FPR16:$Rt), am_unscaled16:$addr)]>;
1699 def STURSi : StoreUnscaled<0b10, 1, 0b00, FPR32, am_unscaled32, "stur",
1700 [(store (f32 FPR32:$Rt), am_unscaled32:$addr)]>;
1701 def STURDi : StoreUnscaled<0b11, 1, 0b00, FPR64, am_unscaled64, "stur",
1702 [(store (f64 FPR64:$Rt), am_unscaled64:$addr)]>;
1703 def STURQi : StoreUnscaled<0b00, 1, 0b10, FPR128, am_unscaled128, "stur",
1704 [(store (v2f64 FPR128:$Rt), am_unscaled128:$addr)]>;
1705 def STURHHi : StoreUnscaled<0b01, 0, 0b00, GPR32, am_unscaled16, "sturh",
1706 [(truncstorei16 GPR32:$Rt, am_unscaled16:$addr)]>;
1707 def STURBBi : StoreUnscaled<0b00, 0, 0b00, GPR32, am_unscaled8, "sturb",
1708 [(truncstorei8 GPR32:$Rt, am_unscaled8:$addr)]>;
1710 // Match all store 64 bits width whose type is compatible with FPR64
1711 def : Pat<(store (v2f32 FPR64:$Rn), am_unscaled64:$addr),
1712 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1713 def : Pat<(store (v1f64 FPR64:$Rn), am_unscaled64:$addr),
1714 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1715 def : Pat<(store (v8i8 FPR64:$Rn), am_unscaled64:$addr),
1716 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1717 def : Pat<(store (v4i16 FPR64:$Rn), am_unscaled64:$addr),
1718 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1719 def : Pat<(store (v2i32 FPR64:$Rn), am_unscaled64:$addr),
1720 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1721 def : Pat<(store (v1i64 FPR64:$Rn), am_unscaled64:$addr),
1722 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1724 // Match all store 128 bits width whose type is compatible with FPR128
1725 def : Pat<(store (v4f32 FPR128:$Rn), am_unscaled128:$addr),
1726 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1727 def : Pat<(store (v2f64 FPR128:$Rn), am_unscaled128:$addr),
1728 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1729 def : Pat<(store (v16i8 FPR128:$Rn), am_unscaled128:$addr),
1730 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1731 def : Pat<(store (v8i16 FPR128:$Rn), am_unscaled128:$addr),
1732 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1733 def : Pat<(store (v4i32 FPR128:$Rn), am_unscaled128:$addr),
1734 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1735 def : Pat<(store (v2i64 FPR128:$Rn), am_unscaled128:$addr),
1736 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1737 def : Pat<(store (f128 FPR128:$Rn), am_unscaled128:$addr),
1738 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1740 // unscaled i64 truncating stores
1741 def : Pat<(truncstorei32 GPR64:$Rt, am_unscaled32:$addr),
1742 (STURWi (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_unscaled32:$addr)>;
1743 def : Pat<(truncstorei16 GPR64:$Rt, am_unscaled16:$addr),
1744 (STURHHi (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_unscaled16:$addr)>;
1745 def : Pat<(truncstorei8 GPR64:$Rt, am_unscaled8:$addr),
1746 (STURBBi (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_unscaled8:$addr)>;
1749 // STR mnemonics fall back to STUR for negative or unaligned offsets.
1750 def : InstAlias<"str $Rt, $addr", (STURXi GPR64:$Rt, am_unscaled_fb64:$addr)>;
1751 def : InstAlias<"str $Rt, $addr", (STURWi GPR32:$Rt, am_unscaled_fb32:$addr)>;
1752 def : InstAlias<"str $Rt, $addr", (STURBi FPR8:$Rt, am_unscaled_fb8:$addr)>;
1753 def : InstAlias<"str $Rt, $addr", (STURHi FPR16:$Rt, am_unscaled_fb16:$addr)>;
1754 def : InstAlias<"str $Rt, $addr", (STURSi FPR32:$Rt, am_unscaled_fb32:$addr)>;
1755 def : InstAlias<"str $Rt, $addr", (STURDi FPR64:$Rt, am_unscaled_fb64:$addr)>;
1756 def : InstAlias<"str $Rt, $addr", (STURQi FPR128:$Rt, am_unscaled_fb128:$addr)>;
1758 def : InstAlias<"strb $Rt, $addr", (STURBBi GPR32:$Rt, am_unscaled_fb8:$addr)>;
1759 def : InstAlias<"strh $Rt, $addr", (STURHHi GPR32:$Rt, am_unscaled_fb16:$addr)>;
1762 // (unscaled immediate, unprivileged)
1763 def STTRWi : StoreUnprivileged<0b10, 0, 0b00, GPR32, "sttr">;
1764 def STTRXi : StoreUnprivileged<0b11, 0, 0b00, GPR64, "sttr">;
1766 def STTRHi : StoreUnprivileged<0b01, 0, 0b00, GPR32, "sttrh">;
1767 def STTRBi : StoreUnprivileged<0b00, 0, 0b00, GPR32, "sttrb">;
1770 // (immediate pre-indexed)
1771 def STRWpre : StorePreIdx<0b10, 0, 0b00, GPR32, "str">;
1772 def STRXpre : StorePreIdx<0b11, 0, 0b00, GPR64, "str">;
1773 def STRBpre : StorePreIdx<0b00, 1, 0b00, FPR8, "str">;
1774 def STRHpre : StorePreIdx<0b01, 1, 0b00, FPR16, "str">;
1775 def STRSpre : StorePreIdx<0b10, 1, 0b00, FPR32, "str">;
1776 def STRDpre : StorePreIdx<0b11, 1, 0b00, FPR64, "str">;
1777 def STRQpre : StorePreIdx<0b00, 1, 0b10, FPR128, "str">;
1779 def STRBBpre : StorePreIdx<0b00, 0, 0b00, GPR32, "strb">;
1780 def STRHHpre : StorePreIdx<0b01, 0, 0b00, GPR32, "strh">;
1782 // ISel pseudos and patterns. See expanded comment on StorePreIdxPseudo.
1783 defm STRDpre : StorePreIdxPseudo<FPR64, f64, pre_store>;
1784 defm STRSpre : StorePreIdxPseudo<FPR32, f32, pre_store>;
1785 defm STRXpre : StorePreIdxPseudo<GPR64, i64, pre_store>;
1786 defm STRWpre : StorePreIdxPseudo<GPR32, i32, pre_store>;
1787 defm STRHHpre : StorePreIdxPseudo<GPR32, i32, pre_truncsti16>;
1788 defm STRBBpre : StorePreIdxPseudo<GPR32, i32, pre_truncsti8>;
1790 def : Pat<(pre_truncsti32 GPR64:$Rt, am_noindex:$addr, simm9:$off),
1791 (STRWpre_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
1793 def : Pat<(pre_truncsti16 GPR64:$Rt, am_noindex:$addr, simm9:$off),
1794 (STRHHpre_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
1796 def : Pat<(pre_truncsti8 GPR64:$Rt, am_noindex:$addr, simm9:$off),
1797 (STRBBpre_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
1801 // (immediate post-indexed)
1802 def STRWpost : StorePostIdx<0b10, 0, 0b00, GPR32, "str">;
1803 def STRXpost : StorePostIdx<0b11, 0, 0b00, GPR64, "str">;
1804 def STRBpost : StorePostIdx<0b00, 1, 0b00, FPR8, "str">;
1805 def STRHpost : StorePostIdx<0b01, 1, 0b00, FPR16, "str">;
1806 def STRSpost : StorePostIdx<0b10, 1, 0b00, FPR32, "str">;
1807 def STRDpost : StorePostIdx<0b11, 1, 0b00, FPR64, "str">;
1808 def STRQpost : StorePostIdx<0b00, 1, 0b10, FPR128, "str">;
1810 def STRBBpost : StorePostIdx<0b00, 0, 0b00, GPR32, "strb">;
1811 def STRHHpost : StorePostIdx<0b01, 0, 0b00, GPR32, "strh">;
1813 // ISel pseudos and patterns. See expanded comment on StorePostIdxPseudo.
1814 defm STRDpost : StorePostIdxPseudo<FPR64, f64, post_store, STRDpost>;
1815 defm STRSpost : StorePostIdxPseudo<FPR32, f32, post_store, STRSpost>;
1816 defm STRXpost : StorePostIdxPseudo<GPR64, i64, post_store, STRXpost>;
1817 defm STRWpost : StorePostIdxPseudo<GPR32, i32, post_store, STRWpost>;
1818 defm STRHHpost : StorePostIdxPseudo<GPR32, i32, post_truncsti16, STRHHpost>;
1819 defm STRBBpost : StorePostIdxPseudo<GPR32, i32, post_truncsti8, STRBBpost>;
1821 def : Pat<(post_truncsti32 GPR64:$Rt, am_noindex:$addr, simm9:$off),
1822 (STRWpost_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
1824 def : Pat<(post_truncsti16 GPR64:$Rt, am_noindex:$addr, simm9:$off),
1825 (STRHHpost_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
1827 def : Pat<(post_truncsti8 GPR64:$Rt, am_noindex:$addr, simm9:$off),
1828 (STRBBpost_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
1832 //===----------------------------------------------------------------------===//
1833 // Load/store exclusive instructions.
1834 //===----------------------------------------------------------------------===//
1836 def LDARW : LoadAcquire <0b10, 1, 1, 0, 1, GPR32, "ldar">;
1837 def LDARX : LoadAcquire <0b11, 1, 1, 0, 1, GPR64, "ldar">;
1838 def LDARB : LoadAcquire <0b00, 1, 1, 0, 1, GPR32, "ldarb">;
1839 def LDARH : LoadAcquire <0b01, 1, 1, 0, 1, GPR32, "ldarh">;
1841 def LDAXRW : LoadExclusive <0b10, 0, 1, 0, 1, GPR32, "ldaxr">;
1842 def LDAXRX : LoadExclusive <0b11, 0, 1, 0, 1, GPR64, "ldaxr">;
1843 def LDAXRB : LoadExclusive <0b00, 0, 1, 0, 1, GPR32, "ldaxrb">;
1844 def LDAXRH : LoadExclusive <0b01, 0, 1, 0, 1, GPR32, "ldaxrh">;
1846 def LDXRW : LoadExclusive <0b10, 0, 1, 0, 0, GPR32, "ldxr">;
1847 def LDXRX : LoadExclusive <0b11, 0, 1, 0, 0, GPR64, "ldxr">;
1848 def LDXRB : LoadExclusive <0b00, 0, 1, 0, 0, GPR32, "ldxrb">;
1849 def LDXRH : LoadExclusive <0b01, 0, 1, 0, 0, GPR32, "ldxrh">;
1851 def STLRW : StoreRelease <0b10, 1, 0, 0, 1, GPR32, "stlr">;
1852 def STLRX : StoreRelease <0b11, 1, 0, 0, 1, GPR64, "stlr">;
1853 def STLRB : StoreRelease <0b00, 1, 0, 0, 1, GPR32, "stlrb">;
1854 def STLRH : StoreRelease <0b01, 1, 0, 0, 1, GPR32, "stlrh">;
1856 def STLXRW : StoreExclusive<0b10, 0, 0, 0, 1, GPR32, "stlxr">;
1857 def STLXRX : StoreExclusive<0b11, 0, 0, 0, 1, GPR64, "stlxr">;
1858 def STLXRB : StoreExclusive<0b00, 0, 0, 0, 1, GPR32, "stlxrb">;
1859 def STLXRH : StoreExclusive<0b01, 0, 0, 0, 1, GPR32, "stlxrh">;
1861 def STXRW : StoreExclusive<0b10, 0, 0, 0, 0, GPR32, "stxr">;
1862 def STXRX : StoreExclusive<0b11, 0, 0, 0, 0, GPR64, "stxr">;
1863 def STXRB : StoreExclusive<0b00, 0, 0, 0, 0, GPR32, "stxrb">;
1864 def STXRH : StoreExclusive<0b01, 0, 0, 0, 0, GPR32, "stxrh">;
1866 def LDAXPW : LoadExclusivePair<0b10, 0, 1, 1, 1, GPR32, "ldaxp">;
1867 def LDAXPX : LoadExclusivePair<0b11, 0, 1, 1, 1, GPR64, "ldaxp">;
1869 def LDXPW : LoadExclusivePair<0b10, 0, 1, 1, 0, GPR32, "ldxp">;
1870 def LDXPX : LoadExclusivePair<0b11, 0, 1, 1, 0, GPR64, "ldxp">;
1872 def STLXPW : StoreExclusivePair<0b10, 0, 0, 1, 1, GPR32, "stlxp">;
1873 def STLXPX : StoreExclusivePair<0b11, 0, 0, 1, 1, GPR64, "stlxp">;
1875 def STXPW : StoreExclusivePair<0b10, 0, 0, 1, 0, GPR32, "stxp">;
1876 def STXPX : StoreExclusivePair<0b11, 0, 0, 1, 0, GPR64, "stxp">;
1878 //===----------------------------------------------------------------------===//
1879 // Scaled floating point to integer conversion instructions.
1880 //===----------------------------------------------------------------------===//
1882 defm FCVTAS : FPToIntegerUnscaled<0b00, 0b100, "fcvtas", int_arm64_neon_fcvtas>;
1883 defm FCVTAU : FPToIntegerUnscaled<0b00, 0b101, "fcvtau", int_arm64_neon_fcvtau>;
1884 defm FCVTMS : FPToIntegerUnscaled<0b10, 0b000, "fcvtms", int_arm64_neon_fcvtms>;
1885 defm FCVTMU : FPToIntegerUnscaled<0b10, 0b001, "fcvtmu", int_arm64_neon_fcvtmu>;
1886 defm FCVTNS : FPToIntegerUnscaled<0b00, 0b000, "fcvtns", int_arm64_neon_fcvtns>;
1887 defm FCVTNU : FPToIntegerUnscaled<0b00, 0b001, "fcvtnu", int_arm64_neon_fcvtnu>;
1888 defm FCVTPS : FPToIntegerUnscaled<0b01, 0b000, "fcvtps", int_arm64_neon_fcvtps>;
1889 defm FCVTPU : FPToIntegerUnscaled<0b01, 0b001, "fcvtpu", int_arm64_neon_fcvtpu>;
1890 defm FCVTZS : FPToIntegerUnscaled<0b11, 0b000, "fcvtzs", fp_to_sint>;
1891 defm FCVTZU : FPToIntegerUnscaled<0b11, 0b001, "fcvtzu", fp_to_uint>;
1892 defm FCVTZS : FPToIntegerScaled<0b11, 0b000, "fcvtzs", fp_to_sint>;
1893 defm FCVTZU : FPToIntegerScaled<0b11, 0b001, "fcvtzu", fp_to_uint>;
1894 let isCodeGenOnly = 1 in {
1895 defm FCVTZS_Int : FPToIntegerUnscaled<0b11, 0b000, "fcvtzs", int_arm64_neon_fcvtzs>;
1896 defm FCVTZU_Int : FPToIntegerUnscaled<0b11, 0b001, "fcvtzu", int_arm64_neon_fcvtzu>;
1897 defm FCVTZS_Int : FPToIntegerScaled<0b11, 0b000, "fcvtzs", int_arm64_neon_fcvtzs>;
1898 defm FCVTZU_Int : FPToIntegerScaled<0b11, 0b001, "fcvtzu", int_arm64_neon_fcvtzu>;
1901 //===----------------------------------------------------------------------===//
1902 // Scaled integer to floating point conversion instructions.
1903 //===----------------------------------------------------------------------===//
1905 defm SCVTF : IntegerToFP<0, "scvtf", sint_to_fp>;
1906 defm UCVTF : IntegerToFP<1, "ucvtf", uint_to_fp>;
1908 //===----------------------------------------------------------------------===//
1909 // Unscaled integer to floating point conversion instruction.
1910 //===----------------------------------------------------------------------===//
1912 defm FMOV : UnscaledConversion<"fmov">;
1914 def : Pat<(f32 (fpimm0)), (FMOVWSr WZR)>, Requires<[NoZCZ]>;
1915 def : Pat<(f64 (fpimm0)), (FMOVXDr XZR)>, Requires<[NoZCZ]>;
1917 def : Pat<(v8i8 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1918 def : Pat<(v4i16 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1919 def : Pat<(v2i32 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1920 def : Pat<(v1i64 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1921 def : Pat<(v2f32 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1922 def : Pat<(v1f64 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1923 def : Pat<(v1i64 (scalar_to_vector GPR64:$Xn)),
1924 (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1925 def : Pat<(v1f64 (scalar_to_vector GPR64:$Xn)),
1926 (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1927 def : Pat<(v1f64 (scalar_to_vector (f64 FPR64:$Xn))), (v1f64 FPR64:$Xn)>;
1929 def : Pat<(i64 (bitconvert (v8i8 V64:$Vn))),
1930 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
1931 def : Pat<(i64 (bitconvert (v4i16 V64:$Vn))),
1932 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
1933 def : Pat<(i64 (bitconvert (v2i32 V64:$Vn))),
1934 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
1935 def : Pat<(i64 (bitconvert (v1i64 V64:$Vn))),
1936 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
1937 def : Pat<(i64 (bitconvert (v2f32 V64:$Vn))),
1938 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
1939 def : Pat<(i64 (bitconvert (v1f64 V64:$Vn))),
1940 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
1942 def : Pat<(f32 (bitconvert (i32 GPR32:$Xn))),
1943 (COPY_TO_REGCLASS GPR32:$Xn, FPR32)>;
1944 def : Pat<(i32 (bitconvert (f32 FPR32:$Xn))),
1945 (COPY_TO_REGCLASS FPR32:$Xn, GPR32)>;
1946 def : Pat<(f64 (bitconvert (i64 GPR64:$Xn))),
1947 (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1948 def : Pat<(i64 (bitconvert (f64 FPR64:$Xn))),
1949 (COPY_TO_REGCLASS FPR64:$Xn, GPR64)>;
1951 //===----------------------------------------------------------------------===//
1952 // Floating point conversion instruction.
1953 //===----------------------------------------------------------------------===//
1955 defm FCVT : FPConversion<"fcvt">;
1957 def : Pat<(f32_to_f16 FPR32:$Rn),
1958 (i32 (COPY_TO_REGCLASS
1959 (f32 (SUBREG_TO_REG (i32 0), (FCVTHSr FPR32:$Rn), hsub)),
1962 def FCVTSHpseudo : Pseudo<(outs FPR32:$Rd), (ins FPR32:$Rn),
1963 [(set (f32 FPR32:$Rd), (f16_to_f32 i32:$Rn))]>;
1965 //===----------------------------------------------------------------------===//
1966 // Floating point single operand instructions.
1967 //===----------------------------------------------------------------------===//
1969 defm FABS : SingleOperandFPData<0b0001, "fabs", fabs>;
1970 defm FMOV : SingleOperandFPData<0b0000, "fmov">;
1971 defm FNEG : SingleOperandFPData<0b0010, "fneg", fneg>;
1972 defm FRINTA : SingleOperandFPData<0b1100, "frinta", frnd>;
1973 defm FRINTI : SingleOperandFPData<0b1111, "frinti", fnearbyint>;
1974 defm FRINTM : SingleOperandFPData<0b1010, "frintm", ffloor>;
1975 defm FRINTN : SingleOperandFPData<0b1000, "frintn", int_arm64_neon_frintn>;
1976 defm FRINTP : SingleOperandFPData<0b1001, "frintp", fceil>;
1978 def : Pat<(v1f64 (int_arm64_neon_frintn (v1f64 FPR64:$Rn))),
1979 (FRINTNDr FPR64:$Rn)>;
1981 // FRINTX is inserted to set the flags as required by FENV_ACCESS ON behavior
1982 // in the C spec. Setting hasSideEffects ensures it is not DCE'd.
1983 // <rdar://problem/13715968>
1984 // TODO: We should really model the FPSR flags correctly. This is really ugly.
1985 let hasSideEffects = 1 in {
1986 defm FRINTX : SingleOperandFPData<0b1110, "frintx", frint>;
1989 defm FRINTZ : SingleOperandFPData<0b1011, "frintz", ftrunc>;
1991 let SchedRW = [WriteFDiv] in {
1992 defm FSQRT : SingleOperandFPData<0b0011, "fsqrt", fsqrt>;
1995 //===----------------------------------------------------------------------===//
1996 // Floating point two operand instructions.
1997 //===----------------------------------------------------------------------===//
1999 defm FADD : TwoOperandFPData<0b0010, "fadd", fadd>;
2000 let SchedRW = [WriteFDiv] in {
2001 defm FDIV : TwoOperandFPData<0b0001, "fdiv", fdiv>;
2003 defm FMAXNM : TwoOperandFPData<0b0110, "fmaxnm", int_arm64_neon_fmaxnm>;
2004 defm FMAX : TwoOperandFPData<0b0100, "fmax", ARM64fmax>;
2005 defm FMINNM : TwoOperandFPData<0b0111, "fminnm", int_arm64_neon_fminnm>;
2006 defm FMIN : TwoOperandFPData<0b0101, "fmin", ARM64fmin>;
2007 let SchedRW = [WriteFMul] in {
2008 defm FMUL : TwoOperandFPData<0b0000, "fmul", fmul>;
2009 defm FNMUL : TwoOperandFPDataNeg<0b1000, "fnmul", fmul>;
2011 defm FSUB : TwoOperandFPData<0b0011, "fsub", fsub>;
2013 def : Pat<(v1f64 (ARM64fmax (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2014 (FMAXDrr FPR64:$Rn, FPR64:$Rm)>;
2015 def : Pat<(v1f64 (ARM64fmin (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2016 (FMINDrr FPR64:$Rn, FPR64:$Rm)>;
2017 def : Pat<(v1f64 (int_arm64_neon_fmaxnm (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2018 (FMAXNMDrr FPR64:$Rn, FPR64:$Rm)>;
2019 def : Pat<(v1f64 (int_arm64_neon_fminnm (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2020 (FMINNMDrr FPR64:$Rn, FPR64:$Rm)>;
2022 //===----------------------------------------------------------------------===//
2023 // Floating point three operand instructions.
2024 //===----------------------------------------------------------------------===//
2026 defm FMADD : ThreeOperandFPData<0, 0, "fmadd", fma>;
2027 defm FMSUB : ThreeOperandFPData<0, 1, "fmsub",
2028 TriOpFrag<(fma node:$LHS, (fneg node:$MHS), node:$RHS)> >;
2029 defm FNMADD : ThreeOperandFPData<1, 0, "fnmadd",
2030 TriOpFrag<(fneg (fma node:$LHS, node:$MHS, node:$RHS))> >;
2031 defm FNMSUB : ThreeOperandFPData<1, 1, "fnmsub",
2032 TriOpFrag<(fma node:$LHS, node:$MHS, (fneg node:$RHS))> >;
2034 // The following def pats catch the case where the LHS of an FMA is negated.
2035 // The TriOpFrag above catches the case where the middle operand is negated.
2037 // N.b. FMSUB etc have the accumulator at the *end* of (outs), unlike
2038 // the NEON variant.
2039 def : Pat<(f32 (fma (fneg FPR32:$Rn), FPR32:$Rm, FPR32:$Ra)),
2040 (FMSUBSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
2042 def : Pat<(f64 (fma (fneg FPR64:$Rn), FPR64:$Rm, FPR64:$Ra)),
2043 (FMSUBDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
2045 // We handled -(a + b*c) for FNMADD above, now it's time for "(-a) + (-b)*c" and
2047 def : Pat<(f32 (fma (fneg FPR32:$Rn), FPR32:$Rm, (fneg FPR32:$Ra))),
2048 (FNMADDSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
2050 def : Pat<(f64 (fma (fneg FPR64:$Rn), FPR64:$Rm, (fneg FPR64:$Ra))),
2051 (FNMADDDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
2053 def : Pat<(f32 (fma FPR32:$Rn, (fneg FPR32:$Rm), (fneg FPR32:$Ra))),
2054 (FNMADDSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
2056 def : Pat<(f64 (fma FPR64:$Rn, (fneg FPR64:$Rm), (fneg FPR64:$Ra))),
2057 (FNMADDDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
2059 //===----------------------------------------------------------------------===//
2060 // Floating point comparison instructions.
2061 //===----------------------------------------------------------------------===//
2063 defm FCMPE : FPComparison<1, "fcmpe">;
2064 defm FCMP : FPComparison<0, "fcmp", ARM64fcmp>;
2066 //===----------------------------------------------------------------------===//
2067 // Floating point conditional comparison instructions.
2068 //===----------------------------------------------------------------------===//
2070 defm FCCMPE : FPCondComparison<1, "fccmpe">;
2071 defm FCCMP : FPCondComparison<0, "fccmp">;
2073 //===----------------------------------------------------------------------===//
2074 // Floating point conditional select instruction.
2075 //===----------------------------------------------------------------------===//
2077 defm FCSEL : FPCondSelect<"fcsel">;
2079 // CSEL instructions providing f128 types need to be handled by a
2080 // pseudo-instruction since the eventual code will need to introduce basic
2081 // blocks and control flow.
2082 def F128CSEL : Pseudo<(outs FPR128:$Rd),
2083 (ins FPR128:$Rn, FPR128:$Rm, ccode:$cond),
2084 [(set (f128 FPR128:$Rd),
2085 (ARM64csel FPR128:$Rn, FPR128:$Rm,
2086 (i32 imm:$cond), CPSR))]> {
2088 let usesCustomInserter = 1;
2092 //===----------------------------------------------------------------------===//
2093 // Floating point immediate move.
2094 //===----------------------------------------------------------------------===//
2096 let isReMaterializable = 1 in {
2097 defm FMOV : FPMoveImmediate<"fmov">;
2100 //===----------------------------------------------------------------------===//
2101 // Advanced SIMD two vector instructions.
2102 //===----------------------------------------------------------------------===//
2104 defm ABS : SIMDTwoVectorBHSD<0, 0b01011, "abs", int_arm64_neon_abs>;
2105 defm CLS : SIMDTwoVectorBHS<0, 0b00100, "cls", int_arm64_neon_cls>;
2106 defm CLZ : SIMDTwoVectorBHS<1, 0b00100, "clz", ctlz>;
2107 defm CMEQ : SIMDCmpTwoVector<0, 0b01001, "cmeq", ARM64cmeqz>;
2108 defm CMGE : SIMDCmpTwoVector<1, 0b01000, "cmge", ARM64cmgez>;
2109 defm CMGT : SIMDCmpTwoVector<0, 0b01000, "cmgt", ARM64cmgtz>;
2110 defm CMLE : SIMDCmpTwoVector<1, 0b01001, "cmle", ARM64cmlez>;
2111 defm CMLT : SIMDCmpTwoVector<0, 0b01010, "cmlt", ARM64cmltz>;
2112 defm CNT : SIMDTwoVectorB<0, 0b00, 0b00101, "cnt", ctpop>;
2113 defm FABS : SIMDTwoVectorFP<0, 1, 0b01111, "fabs", fabs>;
2115 defm FCMEQ : SIMDFPCmpTwoVector<0, 1, 0b01101, "fcmeq", ARM64fcmeqz>;
2116 defm FCMGE : SIMDFPCmpTwoVector<1, 1, 0b01100, "fcmge", ARM64fcmgez>;
2117 defm FCMGT : SIMDFPCmpTwoVector<0, 1, 0b01100, "fcmgt", ARM64fcmgtz>;
2118 defm FCMLE : SIMDFPCmpTwoVector<1, 1, 0b01101, "fcmle", ARM64fcmlez>;
2119 defm FCMLT : SIMDFPCmpTwoVector<0, 1, 0b01110, "fcmlt", ARM64fcmltz>;
2120 defm FCVTAS : SIMDTwoVectorFPToInt<0,0,0b11100, "fcvtas",int_arm64_neon_fcvtas>;
2121 defm FCVTAU : SIMDTwoVectorFPToInt<1,0,0b11100, "fcvtau",int_arm64_neon_fcvtau>;
2122 defm FCVTL : SIMDFPWidenTwoVector<0, 0, 0b10111, "fcvtl">;
2123 def : Pat<(v4f32 (int_arm64_neon_vcvthf2fp (v4i16 V64:$Rn))),
2124 (FCVTLv4i16 V64:$Rn)>;
2125 def : Pat<(v4f32 (int_arm64_neon_vcvthf2fp (extract_subvector (v8i16 V128:$Rn),
2127 (FCVTLv8i16 V128:$Rn)>;
2128 def : Pat<(v2f64 (fextend (v2f32 V64:$Rn))), (FCVTLv2i32 V64:$Rn)>;
2129 def : Pat<(v2f64 (fextend (v2f32 (extract_subvector (v4f32 V128:$Rn),
2131 (FCVTLv4i32 V128:$Rn)>;
2133 defm FCVTMS : SIMDTwoVectorFPToInt<0,0,0b11011, "fcvtms",int_arm64_neon_fcvtms>;
2134 defm FCVTMU : SIMDTwoVectorFPToInt<1,0,0b11011, "fcvtmu",int_arm64_neon_fcvtmu>;
2135 defm FCVTNS : SIMDTwoVectorFPToInt<0,0,0b11010, "fcvtns",int_arm64_neon_fcvtns>;
2136 defm FCVTNU : SIMDTwoVectorFPToInt<1,0,0b11010, "fcvtnu",int_arm64_neon_fcvtnu>;
2137 defm FCVTN : SIMDFPNarrowTwoVector<0, 0, 0b10110, "fcvtn">;
2138 def : Pat<(v4i16 (int_arm64_neon_vcvtfp2hf (v4f32 V128:$Rn))),
2139 (FCVTNv4i16 V128:$Rn)>;
2140 def : Pat<(concat_vectors V64:$Rd,
2141 (v4i16 (int_arm64_neon_vcvtfp2hf (v4f32 V128:$Rn)))),
2142 (FCVTNv8i16 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
2143 def : Pat<(v2f32 (fround (v2f64 V128:$Rn))), (FCVTNv2i32 V128:$Rn)>;
2144 def : Pat<(concat_vectors V64:$Rd, (v2f32 (fround (v2f64 V128:$Rn)))),
2145 (FCVTNv4i32 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
2146 defm FCVTPS : SIMDTwoVectorFPToInt<0,1,0b11010, "fcvtps",int_arm64_neon_fcvtps>;
2147 defm FCVTPU : SIMDTwoVectorFPToInt<1,1,0b11010, "fcvtpu",int_arm64_neon_fcvtpu>;
2148 defm FCVTXN : SIMDFPInexactCvtTwoVector<1, 0, 0b10110, "fcvtxn",
2149 int_arm64_neon_fcvtxn>;
2150 defm FCVTZS : SIMDTwoVectorFPToInt<0, 1, 0b11011, "fcvtzs", fp_to_sint>;
2151 defm FCVTZU : SIMDTwoVectorFPToInt<1, 1, 0b11011, "fcvtzu", fp_to_uint>;
2152 let isCodeGenOnly = 1 in {
2153 defm FCVTZS_Int : SIMDTwoVectorFPToInt<0, 1, 0b11011, "fcvtzs",
2154 int_arm64_neon_fcvtzs>;
2155 defm FCVTZU_Int : SIMDTwoVectorFPToInt<1, 1, 0b11011, "fcvtzu",
2156 int_arm64_neon_fcvtzu>;
2158 defm FNEG : SIMDTwoVectorFP<1, 1, 0b01111, "fneg", fneg>;
2159 defm FRECPE : SIMDTwoVectorFP<0, 1, 0b11101, "frecpe", int_arm64_neon_frecpe>;
2160 defm FRINTA : SIMDTwoVectorFP<1, 0, 0b11000, "frinta", frnd>;
2161 defm FRINTI : SIMDTwoVectorFP<1, 1, 0b11001, "frinti", fnearbyint>;
2162 defm FRINTM : SIMDTwoVectorFP<0, 0, 0b11001, "frintm", ffloor>;
2163 defm FRINTN : SIMDTwoVectorFP<0, 0, 0b11000, "frintn", int_arm64_neon_frintn>;
2164 defm FRINTP : SIMDTwoVectorFP<0, 1, 0b11000, "frintp", fceil>;
2165 defm FRINTX : SIMDTwoVectorFP<1, 0, 0b11001, "frintx", frint>;
2166 defm FRINTZ : SIMDTwoVectorFP<0, 1, 0b11001, "frintz", ftrunc>;
2167 defm FRSQRTE: SIMDTwoVectorFP<1, 1, 0b11101, "frsqrte", int_arm64_neon_frsqrte>;
2168 defm FSQRT : SIMDTwoVectorFP<1, 1, 0b11111, "fsqrt", fsqrt>;
2169 defm NEG : SIMDTwoVectorBHSD<1, 0b01011, "neg",
2170 UnOpFrag<(sub immAllZerosV, node:$LHS)> >;
2171 defm NOT : SIMDTwoVectorB<1, 0b00, 0b00101, "not", vnot>;
2172 // Aliases for MVN -> NOT.
2173 def : InstAlias<"mvn.8b $Vd, $Vn", (NOTv8i8 V64:$Vd, V64:$Vn)>;
2174 def : InstAlias<"mvn.16b $Vd, $Vn", (NOTv16i8 V128:$Vd, V128:$Vn)>;
2175 def : InstAlias<"mvn $Vd.8b, $Vn.8b", (NOTv8i8 V64:$Vd, V64:$Vn)>;
2176 def : InstAlias<"mvn $Vd.16b, $Vn.16b", (NOTv16i8 V128:$Vd, V128:$Vn)>;
2178 def : Pat<(ARM64neg (v8i8 V64:$Rn)), (NEGv8i8 V64:$Rn)>;
2179 def : Pat<(ARM64neg (v16i8 V128:$Rn)), (NEGv16i8 V128:$Rn)>;
2180 def : Pat<(ARM64neg (v4i16 V64:$Rn)), (NEGv4i16 V64:$Rn)>;
2181 def : Pat<(ARM64neg (v8i16 V128:$Rn)), (NEGv8i16 V128:$Rn)>;
2182 def : Pat<(ARM64neg (v2i32 V64:$Rn)), (NEGv2i32 V64:$Rn)>;
2183 def : Pat<(ARM64neg (v4i32 V128:$Rn)), (NEGv4i32 V128:$Rn)>;
2184 def : Pat<(ARM64neg (v2i64 V128:$Rn)), (NEGv2i64 V128:$Rn)>;
2186 def : Pat<(ARM64not (v8i8 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2187 def : Pat<(ARM64not (v16i8 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2188 def : Pat<(ARM64not (v4i16 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2189 def : Pat<(ARM64not (v8i16 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2190 def : Pat<(ARM64not (v2i32 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2191 def : Pat<(ARM64not (v1i64 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2192 def : Pat<(ARM64not (v4i32 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2193 def : Pat<(ARM64not (v2i64 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2195 def : Pat<(vnot (v4i16 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2196 def : Pat<(vnot (v8i16 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2197 def : Pat<(vnot (v2i32 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2198 def : Pat<(vnot (v4i32 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2199 def : Pat<(vnot (v2i64 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2201 defm RBIT : SIMDTwoVectorB<1, 0b01, 0b00101, "rbit", int_arm64_neon_rbit>;
2202 defm REV16 : SIMDTwoVectorB<0, 0b00, 0b00001, "rev16", ARM64rev16>;
2203 defm REV32 : SIMDTwoVectorBH<1, 0b00000, "rev32", ARM64rev32>;
2204 defm REV64 : SIMDTwoVectorBHS<0, 0b00000, "rev64", ARM64rev64>;
2205 defm SADALP : SIMDLongTwoVectorTied<0, 0b00110, "sadalp",
2206 BinOpFrag<(add node:$LHS, (int_arm64_neon_saddlp node:$RHS))> >;
2207 defm SADDLP : SIMDLongTwoVector<0, 0b00010, "saddlp", int_arm64_neon_saddlp>;
2208 defm SCVTF : SIMDTwoVectorIntToFP<0, 0, 0b11101, "scvtf", sint_to_fp>;
2209 defm SHLL : SIMDVectorLShiftLongBySizeBHS;
2210 defm SQABS : SIMDTwoVectorBHSD<0, 0b00111, "sqabs", int_arm64_neon_sqabs>;
2211 defm SQNEG : SIMDTwoVectorBHSD<1, 0b00111, "sqneg", int_arm64_neon_sqneg>;
2212 defm SQXTN : SIMDMixedTwoVector<0, 0b10100, "sqxtn", int_arm64_neon_sqxtn>;
2213 defm SQXTUN : SIMDMixedTwoVector<1, 0b10010, "sqxtun", int_arm64_neon_sqxtun>;
2214 defm SUQADD : SIMDTwoVectorBHSDTied<0, 0b00011, "suqadd",int_arm64_neon_suqadd>;
2215 defm UADALP : SIMDLongTwoVectorTied<1, 0b00110, "uadalp",
2216 BinOpFrag<(add node:$LHS, (int_arm64_neon_uaddlp node:$RHS))> >;
2217 defm UADDLP : SIMDLongTwoVector<1, 0b00010, "uaddlp",
2218 int_arm64_neon_uaddlp>;
2219 defm UCVTF : SIMDTwoVectorIntToFP<1, 0, 0b11101, "ucvtf", uint_to_fp>;
2220 defm UQXTN : SIMDMixedTwoVector<1, 0b10100, "uqxtn", int_arm64_neon_uqxtn>;
2221 defm URECPE : SIMDTwoVectorS<0, 1, 0b11100, "urecpe", int_arm64_neon_urecpe>;
2222 defm URSQRTE: SIMDTwoVectorS<1, 1, 0b11100, "ursqrte", int_arm64_neon_ursqrte>;
2223 defm USQADD : SIMDTwoVectorBHSDTied<1, 0b00011, "usqadd",int_arm64_neon_usqadd>;
2224 defm XTN : SIMDMixedTwoVector<0, 0b10010, "xtn", trunc>;
2226 def : Pat<(v2f32 (ARM64rev64 V64:$Rn)), (REV64v2i32 V64:$Rn)>;
2227 def : Pat<(v4f32 (ARM64rev64 V128:$Rn)), (REV64v4i32 V128:$Rn)>;
2229 // Patterns for vector long shift (by element width). These need to match all
2230 // three of zext, sext and anyext so it's easier to pull the patterns out of the
2232 multiclass SIMDVectorLShiftLongBySizeBHSPats<SDPatternOperator ext> {
2233 def : Pat<(ARM64vshl (v8i16 (ext (v8i8 V64:$Rn))), (i32 8)),
2234 (SHLLv8i8 V64:$Rn)>;
2235 def : Pat<(ARM64vshl (v8i16 (ext (extract_high_v16i8 V128:$Rn))), (i32 8)),
2236 (SHLLv16i8 V128:$Rn)>;
2237 def : Pat<(ARM64vshl (v4i32 (ext (v4i16 V64:$Rn))), (i32 16)),
2238 (SHLLv4i16 V64:$Rn)>;
2239 def : Pat<(ARM64vshl (v4i32 (ext (extract_high_v8i16 V128:$Rn))), (i32 16)),
2240 (SHLLv8i16 V128:$Rn)>;
2241 def : Pat<(ARM64vshl (v2i64 (ext (v2i32 V64:$Rn))), (i32 32)),
2242 (SHLLv2i32 V64:$Rn)>;
2243 def : Pat<(ARM64vshl (v2i64 (ext (extract_high_v4i32 V128:$Rn))), (i32 32)),
2244 (SHLLv4i32 V128:$Rn)>;
2247 defm : SIMDVectorLShiftLongBySizeBHSPats<anyext>;
2248 defm : SIMDVectorLShiftLongBySizeBHSPats<zext>;
2249 defm : SIMDVectorLShiftLongBySizeBHSPats<sext>;
2251 //===----------------------------------------------------------------------===//
2252 // Advanced SIMD three vector instructions.
2253 //===----------------------------------------------------------------------===//
2255 defm ADD : SIMDThreeSameVector<0, 0b10000, "add", add>;
2256 defm ADDP : SIMDThreeSameVector<0, 0b10111, "addp", int_arm64_neon_addp>;
2257 defm CMEQ : SIMDThreeSameVector<1, 0b10001, "cmeq", ARM64cmeq>;
2258 defm CMGE : SIMDThreeSameVector<0, 0b00111, "cmge", ARM64cmge>;
2259 defm CMGT : SIMDThreeSameVector<0, 0b00110, "cmgt", ARM64cmgt>;
2260 defm CMHI : SIMDThreeSameVector<1, 0b00110, "cmhi", ARM64cmhi>;
2261 defm CMHS : SIMDThreeSameVector<1, 0b00111, "cmhs", ARM64cmhs>;
2262 defm CMTST : SIMDThreeSameVector<0, 0b10001, "cmtst", ARM64cmtst>;
2263 defm FABD : SIMDThreeSameVectorFP<1,1,0b11010,"fabd", int_arm64_neon_fabd>;
2264 defm FACGE : SIMDThreeSameVectorFPCmp<1,0,0b11101,"facge",int_arm64_neon_facge>;
2265 defm FACGT : SIMDThreeSameVectorFPCmp<1,1,0b11101,"facgt",int_arm64_neon_facgt>;
2266 defm FADDP : SIMDThreeSameVectorFP<1,0,0b11010,"faddp",int_arm64_neon_addp>;
2267 defm FADD : SIMDThreeSameVectorFP<0,0,0b11010,"fadd", fadd>;
2268 defm FCMEQ : SIMDThreeSameVectorFPCmp<0, 0, 0b11100, "fcmeq", ARM64fcmeq>;
2269 defm FCMGE : SIMDThreeSameVectorFPCmp<1, 0, 0b11100, "fcmge", ARM64fcmge>;
2270 defm FCMGT : SIMDThreeSameVectorFPCmp<1, 1, 0b11100, "fcmgt", ARM64fcmgt>;
2271 defm FDIV : SIMDThreeSameVectorFP<1,0,0b11111,"fdiv", fdiv>;
2272 defm FMAXNMP : SIMDThreeSameVectorFP<1,0,0b11000,"fmaxnmp", int_arm64_neon_fmaxnmp>;
2273 defm FMAXNM : SIMDThreeSameVectorFP<0,0,0b11000,"fmaxnm", int_arm64_neon_fmaxnm>;
2274 defm FMAXP : SIMDThreeSameVectorFP<1,0,0b11110,"fmaxp", int_arm64_neon_fmaxp>;
2275 defm FMAX : SIMDThreeSameVectorFP<0,0,0b11110,"fmax", ARM64fmax>;
2276 defm FMINNMP : SIMDThreeSameVectorFP<1,1,0b11000,"fminnmp", int_arm64_neon_fminnmp>;
2277 defm FMINNM : SIMDThreeSameVectorFP<0,1,0b11000,"fminnm", int_arm64_neon_fminnm>;
2278 defm FMINP : SIMDThreeSameVectorFP<1,1,0b11110,"fminp", int_arm64_neon_fminp>;
2279 defm FMIN : SIMDThreeSameVectorFP<0,1,0b11110,"fmin", ARM64fmin>;
2281 // NOTE: The operands of the PatFrag are reordered on FMLA/FMLS because the
2282 // instruction expects the addend first, while the fma intrinsic puts it last.
2283 defm FMLA : SIMDThreeSameVectorFPTied<0, 0, 0b11001, "fmla",
2284 TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)> >;
2285 defm FMLS : SIMDThreeSameVectorFPTied<0, 1, 0b11001, "fmls",
2286 TriOpFrag<(fma node:$MHS, (fneg node:$RHS), node:$LHS)> >;
2288 // The following def pats catch the case where the LHS of an FMA is negated.
2289 // The TriOpFrag above catches the case where the middle operand is negated.
2290 def : Pat<(v2f32 (fma (fneg V64:$Rn), V64:$Rm, V64:$Rd)),
2291 (FMLSv2f32 V64:$Rd, V64:$Rn, V64:$Rm)>;
2293 def : Pat<(v4f32 (fma (fneg V128:$Rn), V128:$Rm, V128:$Rd)),
2294 (FMLSv4f32 V128:$Rd, V128:$Rn, V128:$Rm)>;
2296 def : Pat<(v2f64 (fma (fneg V128:$Rn), V128:$Rm, V128:$Rd)),
2297 (FMLSv2f64 V128:$Rd, V128:$Rn, V128:$Rm)>;
2299 defm FMULX : SIMDThreeSameVectorFP<0,0,0b11011,"fmulx", int_arm64_neon_fmulx>;
2300 defm FMUL : SIMDThreeSameVectorFP<1,0,0b11011,"fmul", fmul>;
2301 defm FRECPS : SIMDThreeSameVectorFP<0,0,0b11111,"frecps", int_arm64_neon_frecps>;
2302 defm FRSQRTS : SIMDThreeSameVectorFP<0,1,0b11111,"frsqrts", int_arm64_neon_frsqrts>;
2303 defm FSUB : SIMDThreeSameVectorFP<0,1,0b11010,"fsub", fsub>;
2304 defm MLA : SIMDThreeSameVectorBHSTied<0, 0b10010, "mla",
2305 TriOpFrag<(add node:$LHS, (mul node:$MHS, node:$RHS))> >;
2306 defm MLS : SIMDThreeSameVectorBHSTied<1, 0b10010, "mls",
2307 TriOpFrag<(sub node:$LHS, (mul node:$MHS, node:$RHS))> >;
2308 defm MUL : SIMDThreeSameVectorBHS<0, 0b10011, "mul", mul>;
2309 defm PMUL : SIMDThreeSameVectorB<1, 0b10011, "pmul", int_arm64_neon_pmul>;
2310 defm SABA : SIMDThreeSameVectorBHSTied<0, 0b01111, "saba",
2311 TriOpFrag<(add node:$LHS, (int_arm64_neon_sabd node:$MHS, node:$RHS))> >;
2312 defm SABD : SIMDThreeSameVectorBHS<0,0b01110,"sabd", int_arm64_neon_sabd>;
2313 defm SHADD : SIMDThreeSameVectorBHS<0,0b00000,"shadd", int_arm64_neon_shadd>;
2314 defm SHSUB : SIMDThreeSameVectorBHS<0,0b00100,"shsub", int_arm64_neon_shsub>;
2315 defm SMAXP : SIMDThreeSameVectorBHS<0,0b10100,"smaxp", int_arm64_neon_smaxp>;
2316 defm SMAX : SIMDThreeSameVectorBHS<0,0b01100,"smax", int_arm64_neon_smax>;
2317 defm SMINP : SIMDThreeSameVectorBHS<0,0b10101,"sminp", int_arm64_neon_sminp>;
2318 defm SMIN : SIMDThreeSameVectorBHS<0,0b01101,"smin", int_arm64_neon_smin>;
2319 defm SQADD : SIMDThreeSameVector<0,0b00001,"sqadd", int_arm64_neon_sqadd>;
2320 defm SQDMULH : SIMDThreeSameVectorHS<0,0b10110,"sqdmulh",int_arm64_neon_sqdmulh>;
2321 defm SQRDMULH : SIMDThreeSameVectorHS<1,0b10110,"sqrdmulh",int_arm64_neon_sqrdmulh>;
2322 defm SQRSHL : SIMDThreeSameVector<0,0b01011,"sqrshl", int_arm64_neon_sqrshl>;
2323 defm SQSHL : SIMDThreeSameVector<0,0b01001,"sqshl", int_arm64_neon_sqshl>;
2324 defm SQSUB : SIMDThreeSameVector<0,0b00101,"sqsub", int_arm64_neon_sqsub>;
2325 defm SRHADD : SIMDThreeSameVectorBHS<0,0b00010,"srhadd",int_arm64_neon_srhadd>;
2326 defm SRSHL : SIMDThreeSameVector<0,0b01010,"srshl", int_arm64_neon_srshl>;
2327 defm SSHL : SIMDThreeSameVector<0,0b01000,"sshl", int_arm64_neon_sshl>;
2328 defm SUB : SIMDThreeSameVector<1,0b10000,"sub", sub>;
2329 defm UABA : SIMDThreeSameVectorBHSTied<1, 0b01111, "uaba",
2330 TriOpFrag<(add node:$LHS, (int_arm64_neon_uabd node:$MHS, node:$RHS))> >;
2331 defm UABD : SIMDThreeSameVectorBHS<1,0b01110,"uabd", int_arm64_neon_uabd>;
2332 defm UHADD : SIMDThreeSameVectorBHS<1,0b00000,"uhadd", int_arm64_neon_uhadd>;
2333 defm UHSUB : SIMDThreeSameVectorBHS<1,0b00100,"uhsub", int_arm64_neon_uhsub>;
2334 defm UMAXP : SIMDThreeSameVectorBHS<1,0b10100,"umaxp", int_arm64_neon_umaxp>;
2335 defm UMAX : SIMDThreeSameVectorBHS<1,0b01100,"umax", int_arm64_neon_umax>;
2336 defm UMINP : SIMDThreeSameVectorBHS<1,0b10101,"uminp", int_arm64_neon_uminp>;
2337 defm UMIN : SIMDThreeSameVectorBHS<1,0b01101,"umin", int_arm64_neon_umin>;
2338 defm UQADD : SIMDThreeSameVector<1,0b00001,"uqadd", int_arm64_neon_uqadd>;
2339 defm UQRSHL : SIMDThreeSameVector<1,0b01011,"uqrshl", int_arm64_neon_uqrshl>;
2340 defm UQSHL : SIMDThreeSameVector<1,0b01001,"uqshl", int_arm64_neon_uqshl>;
2341 defm UQSUB : SIMDThreeSameVector<1,0b00101,"uqsub", int_arm64_neon_uqsub>;
2342 defm URHADD : SIMDThreeSameVectorBHS<1,0b00010,"urhadd", int_arm64_neon_urhadd>;
2343 defm URSHL : SIMDThreeSameVector<1,0b01010,"urshl", int_arm64_neon_urshl>;
2344 defm USHL : SIMDThreeSameVector<1,0b01000,"ushl", int_arm64_neon_ushl>;
2346 defm AND : SIMDLogicalThreeVector<0, 0b00, "and", and>;
2347 defm BIC : SIMDLogicalThreeVector<0, 0b01, "bic",
2348 BinOpFrag<(and node:$LHS, (vnot node:$RHS))> >;
2349 defm BIF : SIMDLogicalThreeVector<1, 0b11, "bif">;
2350 defm BIT : SIMDLogicalThreeVectorTied<1, 0b10, "bit", ARM64bit>;
2351 defm BSL : SIMDLogicalThreeVectorTied<1, 0b01, "bsl",
2352 TriOpFrag<(or (and node:$LHS, node:$MHS), (and (vnot node:$LHS), node:$RHS))>>;
2353 defm EOR : SIMDLogicalThreeVector<1, 0b00, "eor", xor>;
2354 defm ORN : SIMDLogicalThreeVector<0, 0b11, "orn",
2355 BinOpFrag<(or node:$LHS, (vnot node:$RHS))> >;
2356 defm ORR : SIMDLogicalThreeVector<0, 0b10, "orr", or>;
2358 // FIXME: the .16b and .8b variantes should be emitted by the
2359 // AsmWriter. TableGen's AsmWriter-generator doesn't deal with variant syntaxes
2360 // in aliases yet though.
2361 def : InstAlias<"mov{\t$dst.16b, $src.16b|.16b\t$dst, $src}",
2362 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2363 def : InstAlias<"{mov\t$dst.8h, $src.8h|mov.8h\t$dst, $src}",
2364 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2365 def : InstAlias<"{mov\t$dst.4s, $src.4s|mov.4s\t$dst, $src}",
2366 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2367 def : InstAlias<"{mov\t$dst.2d, $src.2d|mov.2d\t$dst, $src}",
2368 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2370 def : InstAlias<"{mov\t$dst.8b, $src.8b|mov.8b\t$dst, $src}",
2371 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2372 def : InstAlias<"{mov\t$dst.4h, $src.4h|mov.4h\t$dst, $src}",
2373 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2374 def : InstAlias<"{mov\t$dst.2s, $src.2s|mov.2s\t$dst, $src}",
2375 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2376 def : InstAlias<"{mov\t$dst.1d, $src.1d|mov.1d\t$dst, $src}",
2377 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2379 def : InstAlias<"{cmls\t$dst.8b, $src1.8b, $src2.8b" #
2380 "|cmls.8b\t$dst, $src1, $src2}",
2381 (CMHSv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2382 def : InstAlias<"{cmls\t$dst.16b, $src1.16b, $src2.16b" #
2383 "|cmls.16b\t$dst, $src1, $src2}",
2384 (CMHSv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2385 def : InstAlias<"{cmls\t$dst.4h, $src1.4h, $src2.4h" #
2386 "|cmls.4h\t$dst, $src1, $src2}",
2387 (CMHSv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2388 def : InstAlias<"{cmls\t$dst.8h, $src1.8h, $src2.8h" #
2389 "|cmls.8h\t$dst, $src1, $src2}",
2390 (CMHSv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2391 def : InstAlias<"{cmls\t$dst.2s, $src1.2s, $src2.2s" #
2392 "|cmls.2s\t$dst, $src1, $src2}",
2393 (CMHSv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2394 def : InstAlias<"{cmls\t$dst.4s, $src1.4s, $src2.4s" #
2395 "|cmls.4s\t$dst, $src1, $src2}",
2396 (CMHSv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2397 def : InstAlias<"{cmls\t$dst.2d, $src1.2d, $src2.2d" #
2398 "|cmls.2d\t$dst, $src1, $src2}",
2399 (CMHSv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2401 def : InstAlias<"{cmlo\t$dst.8b, $src1.8b, $src2.8b" #
2402 "|cmlo.8b\t$dst, $src1, $src2}",
2403 (CMHIv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2404 def : InstAlias<"{cmlo\t$dst.16b, $src1.16b, $src2.16b" #
2405 "|cmlo.16b\t$dst, $src1, $src2}",
2406 (CMHIv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2407 def : InstAlias<"{cmlo\t$dst.4h, $src1.4h, $src2.4h" #
2408 "|cmlo.4h\t$dst, $src1, $src2}",
2409 (CMHIv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2410 def : InstAlias<"{cmlo\t$dst.8h, $src1.8h, $src2.8h" #
2411 "|cmlo.8h\t$dst, $src1, $src2}",
2412 (CMHIv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2413 def : InstAlias<"{cmlo\t$dst.2s, $src1.2s, $src2.2s" #
2414 "|cmlo.2s\t$dst, $src1, $src2}",
2415 (CMHIv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2416 def : InstAlias<"{cmlo\t$dst.4s, $src1.4s, $src2.4s" #
2417 "|cmlo.4s\t$dst, $src1, $src2}",
2418 (CMHIv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2419 def : InstAlias<"{cmlo\t$dst.2d, $src1.2d, $src2.2d" #
2420 "|cmlo.2d\t$dst, $src1, $src2}",
2421 (CMHIv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2423 def : InstAlias<"{cmle\t$dst.8b, $src1.8b, $src2.8b" #
2424 "|cmle.8b\t$dst, $src1, $src2}",
2425 (CMGEv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2426 def : InstAlias<"{cmle\t$dst.16b, $src1.16b, $src2.16b" #
2427 "|cmle.16b\t$dst, $src1, $src2}",
2428 (CMGEv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2429 def : InstAlias<"{cmle\t$dst.4h, $src1.4h, $src2.4h" #
2430 "|cmle.4h\t$dst, $src1, $src2}",
2431 (CMGEv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2432 def : InstAlias<"{cmle\t$dst.8h, $src1.8h, $src2.8h" #
2433 "|cmle.8h\t$dst, $src1, $src2}",
2434 (CMGEv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2435 def : InstAlias<"{cmle\t$dst.2s, $src1.2s, $src2.2s" #
2436 "|cmle.2s\t$dst, $src1, $src2}",
2437 (CMGEv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2438 def : InstAlias<"{cmle\t$dst.4s, $src1.4s, $src2.4s" #
2439 "|cmle.4s\t$dst, $src1, $src2}",
2440 (CMGEv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2441 def : InstAlias<"{cmle\t$dst.2d, $src1.2d, $src2.2d" #
2442 "|cmle.2d\t$dst, $src1, $src2}",
2443 (CMGEv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2445 def : InstAlias<"{cmlt\t$dst.8b, $src1.8b, $src2.8b" #
2446 "|cmlt.8b\t$dst, $src1, $src2}",
2447 (CMGTv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2448 def : InstAlias<"{cmlt\t$dst.16b, $src1.16b, $src2.16b" #
2449 "|cmlt.16b\t$dst, $src1, $src2}",
2450 (CMGTv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2451 def : InstAlias<"{cmlt\t$dst.4h, $src1.4h, $src2.4h" #
2452 "|cmlt.4h\t$dst, $src1, $src2}",
2453 (CMGTv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2454 def : InstAlias<"{cmlt\t$dst.8h, $src1.8h, $src2.8h" #
2455 "|cmlt.8h\t$dst, $src1, $src2}",
2456 (CMGTv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2457 def : InstAlias<"{cmlt\t$dst.2s, $src1.2s, $src2.2s" #
2458 "|cmlt.2s\t$dst, $src1, $src2}",
2459 (CMGTv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2460 def : InstAlias<"{cmlt\t$dst.4s, $src1.4s, $src2.4s" #
2461 "|cmlt.4s\t$dst, $src1, $src2}",
2462 (CMGTv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2463 def : InstAlias<"{cmlt\t$dst.2d, $src1.2d, $src2.2d" #
2464 "|cmlt.2d\t$dst, $src1, $src2}",
2465 (CMGTv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2467 def : InstAlias<"{fcmle\t$dst.2s, $src1.2s, $src2.2s" #
2468 "|fcmle.2s\t$dst, $src1, $src2}",
2469 (FCMGEv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
2470 def : InstAlias<"{fcmle\t$dst.4s, $src1.4s, $src2.4s" #
2471 "|fcmle.4s\t$dst, $src1, $src2}",
2472 (FCMGEv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
2473 def : InstAlias<"{fcmle\t$dst.2d, $src1.2d, $src2.2d" #
2474 "|fcmle.2d\t$dst, $src1, $src2}",
2475 (FCMGEv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
2477 def : InstAlias<"{fcmlt\t$dst.2s, $src1.2s, $src2.2s" #
2478 "|fcmlt.2s\t$dst, $src1, $src2}",
2479 (FCMGTv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
2480 def : InstAlias<"{fcmlt\t$dst.4s, $src1.4s, $src2.4s" #
2481 "|fcmlt.4s\t$dst, $src1, $src2}",
2482 (FCMGTv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
2483 def : InstAlias<"{fcmlt\t$dst.2d, $src1.2d, $src2.2d" #
2484 "|fcmlt.2d\t$dst, $src1, $src2}",
2485 (FCMGTv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
2487 def : InstAlias<"{facle\t$dst.2s, $src1.2s, $src2.2s" #
2488 "|facle.2s\t$dst, $src1, $src2}",
2489 (FACGEv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
2490 def : InstAlias<"{facle\t$dst.4s, $src1.4s, $src2.4s" #
2491 "|facle.4s\t$dst, $src1, $src2}",
2492 (FACGEv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
2493 def : InstAlias<"{facle\t$dst.2d, $src1.2d, $src2.2d" #
2494 "|facle.2d\t$dst, $src1, $src2}",
2495 (FACGEv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
2497 def : InstAlias<"{faclt\t$dst.2s, $src1.2s, $src2.2s" #
2498 "|faclt.2s\t$dst, $src1, $src2}",
2499 (FACGTv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
2500 def : InstAlias<"{faclt\t$dst.4s, $src1.4s, $src2.4s" #
2501 "|faclt.4s\t$dst, $src1, $src2}",
2502 (FACGTv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
2503 def : InstAlias<"{faclt\t$dst.2d, $src1.2d, $src2.2d" #
2504 "|faclt.2d\t$dst, $src1, $src2}",
2505 (FACGTv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
2507 //===----------------------------------------------------------------------===//
2508 // Advanced SIMD three scalar instructions.
2509 //===----------------------------------------------------------------------===//
2511 defm ADD : SIMDThreeScalarD<0, 0b10000, "add", add>;
2512 defm CMEQ : SIMDThreeScalarD<1, 0b10001, "cmeq", ARM64cmeq>;
2513 defm CMGE : SIMDThreeScalarD<0, 0b00111, "cmge", ARM64cmge>;
2514 defm CMGT : SIMDThreeScalarD<0, 0b00110, "cmgt", ARM64cmgt>;
2515 defm CMHI : SIMDThreeScalarD<1, 0b00110, "cmhi", ARM64cmhi>;
2516 defm CMHS : SIMDThreeScalarD<1, 0b00111, "cmhs", ARM64cmhs>;
2517 defm CMTST : SIMDThreeScalarD<0, 0b10001, "cmtst", ARM64cmtst>;
2518 defm FABD : SIMDThreeScalarSD<1, 1, 0b11010, "fabd", int_arm64_sisd_fabd>;
2519 def : Pat<(v1f64 (int_arm64_neon_fabd (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2520 (FABD64 FPR64:$Rn, FPR64:$Rm)>;
2521 defm FACGE : SIMDThreeScalarFPCmp<1, 0, 0b11101, "facge",
2522 int_arm64_neon_facge>;
2523 defm FACGT : SIMDThreeScalarFPCmp<1, 1, 0b11101, "facgt",
2524 int_arm64_neon_facgt>;
2525 defm FCMEQ : SIMDThreeScalarFPCmp<0, 0, 0b11100, "fcmeq", ARM64fcmeq>;
2526 defm FCMGE : SIMDThreeScalarFPCmp<1, 0, 0b11100, "fcmge", ARM64fcmge>;
2527 defm FCMGT : SIMDThreeScalarFPCmp<1, 1, 0b11100, "fcmgt", ARM64fcmgt>;
2528 defm FMULX : SIMDThreeScalarSD<0, 0, 0b11011, "fmulx", int_arm64_neon_fmulx>;
2529 defm FRECPS : SIMDThreeScalarSD<0, 0, 0b11111, "frecps", int_arm64_neon_frecps>;
2530 defm FRSQRTS : SIMDThreeScalarSD<0, 1, 0b11111, "frsqrts", int_arm64_neon_frsqrts>;
2531 defm SQADD : SIMDThreeScalarBHSD<0, 0b00001, "sqadd", int_arm64_neon_sqadd>;
2532 defm SQDMULH : SIMDThreeScalarHS< 0, 0b10110, "sqdmulh", int_arm64_neon_sqdmulh>;
2533 defm SQRDMULH : SIMDThreeScalarHS< 1, 0b10110, "sqrdmulh", int_arm64_neon_sqrdmulh>;
2534 defm SQRSHL : SIMDThreeScalarBHSD<0, 0b01011, "sqrshl",int_arm64_neon_sqrshl>;
2535 defm SQSHL : SIMDThreeScalarBHSD<0, 0b01001, "sqshl", int_arm64_neon_sqshl>;
2536 defm SQSUB : SIMDThreeScalarBHSD<0, 0b00101, "sqsub", int_arm64_neon_sqsub>;
2537 defm SRSHL : SIMDThreeScalarD< 0, 0b01010, "srshl", int_arm64_neon_srshl>;
2538 defm SSHL : SIMDThreeScalarD< 0, 0b01000, "sshl", int_arm64_neon_sshl>;
2539 defm SUB : SIMDThreeScalarD< 1, 0b10000, "sub", sub>;
2540 defm UQADD : SIMDThreeScalarBHSD<1, 0b00001, "uqadd", int_arm64_neon_uqadd>;
2541 defm UQRSHL : SIMDThreeScalarBHSD<1, 0b01011, "uqrshl",int_arm64_neon_uqrshl>;
2542 defm UQSHL : SIMDThreeScalarBHSD<1, 0b01001, "uqshl", int_arm64_neon_uqshl>;
2543 defm UQSUB : SIMDThreeScalarBHSD<1, 0b00101, "uqsub", int_arm64_neon_uqsub>;
2544 defm URSHL : SIMDThreeScalarD< 1, 0b01010, "urshl", int_arm64_neon_urshl>;
2545 defm USHL : SIMDThreeScalarD< 1, 0b01000, "ushl", int_arm64_neon_ushl>;
2547 def : InstAlias<"cmls $dst, $src1, $src2",
2548 (CMHSv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2549 def : InstAlias<"cmle $dst, $src1, $src2",
2550 (CMGEv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2551 def : InstAlias<"cmlo $dst, $src1, $src2",
2552 (CMHIv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2553 def : InstAlias<"cmlt $dst, $src1, $src2",
2554 (CMGTv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2555 def : InstAlias<"fcmle $dst, $src1, $src2",
2556 (FCMGE32 FPR32:$dst, FPR32:$src2, FPR32:$src1)>;
2557 def : InstAlias<"fcmle $dst, $src1, $src2",
2558 (FCMGE64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2559 def : InstAlias<"fcmlt $dst, $src1, $src2",
2560 (FCMGT32 FPR32:$dst, FPR32:$src2, FPR32:$src1)>;
2561 def : InstAlias<"fcmlt $dst, $src1, $src2",
2562 (FCMGT64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2563 def : InstAlias<"facle $dst, $src1, $src2",
2564 (FACGE32 FPR32:$dst, FPR32:$src2, FPR32:$src1)>;
2565 def : InstAlias<"facle $dst, $src1, $src2",
2566 (FACGE64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2567 def : InstAlias<"faclt $dst, $src1, $src2",
2568 (FACGT32 FPR32:$dst, FPR32:$src2, FPR32:$src1)>;
2569 def : InstAlias<"faclt $dst, $src1, $src2",
2570 (FACGT64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2572 //===----------------------------------------------------------------------===//
2573 // Advanced SIMD three scalar instructions (mixed operands).
2574 //===----------------------------------------------------------------------===//
2575 defm SQDMULL : SIMDThreeScalarMixedHS<0, 0b11010, "sqdmull",
2576 int_arm64_neon_sqdmulls_scalar>;
2577 defm SQDMLAL : SIMDThreeScalarMixedTiedHS<0, 0b10010, "sqdmlal">;
2578 defm SQDMLSL : SIMDThreeScalarMixedTiedHS<0, 0b10110, "sqdmlsl">;
2580 def : Pat<(i64 (int_arm64_neon_sqadd (i64 FPR64:$Rd),
2581 (i64 (int_arm64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
2582 (i32 FPR32:$Rm))))),
2583 (SQDMLALi32 FPR64:$Rd, FPR32:$Rn, FPR32:$Rm)>;
2584 def : Pat<(i64 (int_arm64_neon_sqsub (i64 FPR64:$Rd),
2585 (i64 (int_arm64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
2586 (i32 FPR32:$Rm))))),
2587 (SQDMLSLi32 FPR64:$Rd, FPR32:$Rn, FPR32:$Rm)>;
2589 //===----------------------------------------------------------------------===//
2590 // Advanced SIMD two scalar instructions.
2591 //===----------------------------------------------------------------------===//
2593 defm ABS : SIMDTwoScalarD< 0, 0b01011, "abs", int_arm64_neon_abs>;
2594 defm CMEQ : SIMDCmpTwoScalarD< 0, 0b01001, "cmeq", ARM64cmeqz>;
2595 defm CMGE : SIMDCmpTwoScalarD< 1, 0b01000, "cmge", ARM64cmgez>;
2596 defm CMGT : SIMDCmpTwoScalarD< 0, 0b01000, "cmgt", ARM64cmgtz>;
2597 defm CMLE : SIMDCmpTwoScalarD< 1, 0b01001, "cmle", ARM64cmlez>;
2598 defm CMLT : SIMDCmpTwoScalarD< 0, 0b01010, "cmlt", ARM64cmltz>;
2599 defm FCMEQ : SIMDCmpTwoScalarSD<0, 1, 0b01101, "fcmeq", ARM64fcmeqz>;
2600 defm FCMGE : SIMDCmpTwoScalarSD<1, 1, 0b01100, "fcmge", ARM64fcmgez>;
2601 defm FCMGT : SIMDCmpTwoScalarSD<0, 1, 0b01100, "fcmgt", ARM64fcmgtz>;
2602 defm FCMLE : SIMDCmpTwoScalarSD<1, 1, 0b01101, "fcmle", ARM64fcmlez>;
2603 defm FCMLT : SIMDCmpTwoScalarSD<0, 1, 0b01110, "fcmlt", ARM64fcmltz>;
2604 defm FCVTAS : SIMDTwoScalarSD< 0, 0, 0b11100, "fcvtas">;
2605 defm FCVTAU : SIMDTwoScalarSD< 1, 0, 0b11100, "fcvtau">;
2606 defm FCVTMS : SIMDTwoScalarSD< 0, 0, 0b11011, "fcvtms">;
2607 defm FCVTMU : SIMDTwoScalarSD< 1, 0, 0b11011, "fcvtmu">;
2608 defm FCVTNS : SIMDTwoScalarSD< 0, 0, 0b11010, "fcvtns">;
2609 defm FCVTNU : SIMDTwoScalarSD< 1, 0, 0b11010, "fcvtnu">;
2610 defm FCVTPS : SIMDTwoScalarSD< 0, 1, 0b11010, "fcvtps">;
2611 defm FCVTPU : SIMDTwoScalarSD< 1, 1, 0b11010, "fcvtpu">;
2612 def FCVTXNv1i64 : SIMDInexactCvtTwoScalar<0b10110, "fcvtxn">;
2613 defm FCVTZS : SIMDTwoScalarSD< 0, 1, 0b11011, "fcvtzs">;
2614 defm FCVTZU : SIMDTwoScalarSD< 1, 1, 0b11011, "fcvtzu">;
2615 defm FRECPE : SIMDTwoScalarSD< 0, 1, 0b11101, "frecpe">;
2616 defm FRECPX : SIMDTwoScalarSD< 0, 1, 0b11111, "frecpx">;
2617 defm FRSQRTE : SIMDTwoScalarSD< 1, 1, 0b11101, "frsqrte">;
2618 defm NEG : SIMDTwoScalarD< 1, 0b01011, "neg",
2619 UnOpFrag<(sub immAllZerosV, node:$LHS)> >;
2620 defm SCVTF : SIMDTwoScalarCVTSD< 0, 0, 0b11101, "scvtf", ARM64sitof>;
2621 defm SQABS : SIMDTwoScalarBHSD< 0, 0b00111, "sqabs", int_arm64_neon_sqabs>;
2622 defm SQNEG : SIMDTwoScalarBHSD< 1, 0b00111, "sqneg", int_arm64_neon_sqneg>;
2623 defm SQXTN : SIMDTwoScalarMixedBHS< 0, 0b10100, "sqxtn", int_arm64_neon_scalar_sqxtn>;
2624 defm SQXTUN : SIMDTwoScalarMixedBHS< 1, 0b10010, "sqxtun", int_arm64_neon_scalar_sqxtun>;
2625 defm SUQADD : SIMDTwoScalarBHSDTied< 0, 0b00011, "suqadd",
2626 int_arm64_neon_suqadd>;
2627 defm UCVTF : SIMDTwoScalarCVTSD< 1, 0, 0b11101, "ucvtf", ARM64uitof>;
2628 defm UQXTN : SIMDTwoScalarMixedBHS<1, 0b10100, "uqxtn", int_arm64_neon_scalar_uqxtn>;
2629 defm USQADD : SIMDTwoScalarBHSDTied< 1, 0b00011, "usqadd",
2630 int_arm64_neon_usqadd>;
2632 def : Pat<(v1i64 (int_arm64_neon_fcvtas (v1f64 FPR64:$Rn))),
2633 (FCVTASv1i64 FPR64:$Rn)>;
2634 def : Pat<(v1i64 (int_arm64_neon_fcvtau (v1f64 FPR64:$Rn))),
2635 (FCVTAUv1i64 FPR64:$Rn)>;
2636 def : Pat<(v1i64 (int_arm64_neon_fcvtms (v1f64 FPR64:$Rn))),
2637 (FCVTMSv1i64 FPR64:$Rn)>;
2638 def : Pat<(v1i64 (int_arm64_neon_fcvtmu (v1f64 FPR64:$Rn))),
2639 (FCVTMUv1i64 FPR64:$Rn)>;
2640 def : Pat<(v1i64 (int_arm64_neon_fcvtns (v1f64 FPR64:$Rn))),
2641 (FCVTNSv1i64 FPR64:$Rn)>;
2642 def : Pat<(v1i64 (int_arm64_neon_fcvtnu (v1f64 FPR64:$Rn))),
2643 (FCVTNUv1i64 FPR64:$Rn)>;
2644 def : Pat<(v1i64 (int_arm64_neon_fcvtps (v1f64 FPR64:$Rn))),
2645 (FCVTPSv1i64 FPR64:$Rn)>;
2646 def : Pat<(v1i64 (int_arm64_neon_fcvtpu (v1f64 FPR64:$Rn))),
2647 (FCVTPUv1i64 FPR64:$Rn)>;
2649 def : Pat<(f32 (int_arm64_neon_frecpe (f32 FPR32:$Rn))),
2650 (FRECPEv1i32 FPR32:$Rn)>;
2651 def : Pat<(f64 (int_arm64_neon_frecpe (f64 FPR64:$Rn))),
2652 (FRECPEv1i64 FPR64:$Rn)>;
2653 def : Pat<(v1f64 (int_arm64_neon_frecpe (v1f64 FPR64:$Rn))),
2654 (FRECPEv1i64 FPR64:$Rn)>;
2656 def : Pat<(f32 (int_arm64_neon_frecpx (f32 FPR32:$Rn))),
2657 (FRECPXv1i32 FPR32:$Rn)>;
2658 def : Pat<(f64 (int_arm64_neon_frecpx (f64 FPR64:$Rn))),
2659 (FRECPXv1i64 FPR64:$Rn)>;
2661 def : Pat<(f32 (int_arm64_neon_frsqrte (f32 FPR32:$Rn))),
2662 (FRSQRTEv1i32 FPR32:$Rn)>;
2663 def : Pat<(f64 (int_arm64_neon_frsqrte (f64 FPR64:$Rn))),
2664 (FRSQRTEv1i64 FPR64:$Rn)>;
2665 def : Pat<(v1f64 (int_arm64_neon_frsqrte (v1f64 FPR64:$Rn))),
2666 (FRSQRTEv1i64 FPR64:$Rn)>;
2668 // If an integer is about to be converted to a floating point value,
2669 // just load it on the floating point unit.
2670 // Here are the patterns for 8 and 16-bits to float.
2672 def : Pat <(f32 (uint_to_fp (i32 (zextloadi8 ro_indexed8:$addr)))),
2673 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2674 (LDRBro ro_indexed8:$addr), bsub))>;
2675 def : Pat <(f32 (uint_to_fp (i32 (zextloadi8 am_indexed8:$addr)))),
2676 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2677 (LDRBui am_indexed8:$addr), bsub))>;
2678 def : Pat <(f32 (uint_to_fp (i32 (zextloadi8 am_unscaled8:$addr)))),
2679 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2680 (LDURBi am_unscaled8:$addr), bsub))>;
2681 // 16-bits -> float.
2682 def : Pat <(f32 (uint_to_fp (i32 (zextloadi16 ro_indexed16:$addr)))),
2683 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2684 (LDRHro ro_indexed16:$addr), hsub))>;
2685 def : Pat <(f32 (uint_to_fp (i32 (zextloadi16 am_indexed16:$addr)))),
2686 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2687 (LDRHui am_indexed16:$addr), hsub))>;
2688 def : Pat <(f32 (uint_to_fp (i32 (zextloadi16 am_unscaled16:$addr)))),
2689 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2690 (LDURHi am_unscaled16:$addr), hsub))>;
2691 // 32-bits are handled in target specific dag combine:
2692 // performIntToFpCombine.
2693 // 64-bits integer to 32-bits floating point, not possible with
2694 // UCVTF on floating point registers (both source and destination
2695 // must have the same size).
2697 // Here are the patterns for 8, 16, 32, and 64-bits to double.
2698 // 8-bits -> double.
2699 def : Pat <(f64 (uint_to_fp (i32 (zextloadi8 ro_indexed8:$addr)))),
2700 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2701 (LDRBro ro_indexed8:$addr), bsub))>;
2702 def : Pat <(f64 (uint_to_fp (i32 (zextloadi8 am_indexed8:$addr)))),
2703 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2704 (LDRBui am_indexed8:$addr), bsub))>;
2705 def : Pat <(f64 (uint_to_fp (i32 (zextloadi8 am_unscaled8:$addr)))),
2706 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2707 (LDURBi am_unscaled8:$addr), bsub))>;
2708 // 16-bits -> double.
2709 def : Pat <(f64 (uint_to_fp (i32 (zextloadi16 ro_indexed16:$addr)))),
2710 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2711 (LDRHro ro_indexed16:$addr), hsub))>;
2712 def : Pat <(f64 (uint_to_fp (i32 (zextloadi16 am_indexed16:$addr)))),
2713 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2714 (LDRHui am_indexed16:$addr), hsub))>;
2715 def : Pat <(f64 (uint_to_fp (i32 (zextloadi16 am_unscaled16:$addr)))),
2716 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2717 (LDURHi am_unscaled16:$addr), hsub))>;
2718 // 32-bits -> double.
2719 def : Pat <(f64 (uint_to_fp (i32 (load ro_indexed32:$addr)))),
2720 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2721 (LDRSro ro_indexed32:$addr), ssub))>;
2722 def : Pat <(f64 (uint_to_fp (i32 (load am_indexed32:$addr)))),
2723 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2724 (LDRSui am_indexed32:$addr), ssub))>;
2725 def : Pat <(f64 (uint_to_fp (i32 (load am_unscaled32:$addr)))),
2726 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2727 (LDURSi am_unscaled32:$addr), ssub))>;
2728 // 64-bits -> double are handled in target specific dag combine:
2729 // performIntToFpCombine.
2731 //===----------------------------------------------------------------------===//
2732 // Advanced SIMD three different-sized vector instructions.
2733 //===----------------------------------------------------------------------===//
2735 defm ADDHN : SIMDNarrowThreeVectorBHS<0,0b0100,"addhn", int_arm64_neon_addhn>;
2736 defm SUBHN : SIMDNarrowThreeVectorBHS<0,0b0110,"subhn", int_arm64_neon_subhn>;
2737 defm RADDHN : SIMDNarrowThreeVectorBHS<1,0b0100,"raddhn",int_arm64_neon_raddhn>;
2738 defm RSUBHN : SIMDNarrowThreeVectorBHS<1,0b0110,"rsubhn",int_arm64_neon_rsubhn>;
2739 defm PMULL : SIMDDifferentThreeVectorBD<0,0b1110,"pmull",int_arm64_neon_pmull>;
2740 defm SABAL : SIMDLongThreeVectorTiedBHSabal<0,0b0101,"sabal",
2741 int_arm64_neon_sabd>;
2742 defm SABDL : SIMDLongThreeVectorBHSabdl<0, 0b0111, "sabdl",
2743 int_arm64_neon_sabd>;
2744 defm SADDL : SIMDLongThreeVectorBHS< 0, 0b0000, "saddl",
2745 BinOpFrag<(add (sext node:$LHS), (sext node:$RHS))>>;
2746 defm SADDW : SIMDWideThreeVectorBHS< 0, 0b0001, "saddw",
2747 BinOpFrag<(add node:$LHS, (sext node:$RHS))>>;
2748 defm SMLAL : SIMDLongThreeVectorTiedBHS<0, 0b1000, "smlal",
2749 TriOpFrag<(add node:$LHS, (int_arm64_neon_smull node:$MHS, node:$RHS))>>;
2750 defm SMLSL : SIMDLongThreeVectorTiedBHS<0, 0b1010, "smlsl",
2751 TriOpFrag<(sub node:$LHS, (int_arm64_neon_smull node:$MHS, node:$RHS))>>;
2752 defm SMULL : SIMDLongThreeVectorBHS<0, 0b1100, "smull", int_arm64_neon_smull>;
2753 defm SQDMLAL : SIMDLongThreeVectorSQDMLXTiedHS<0, 0b1001, "sqdmlal",
2754 int_arm64_neon_sqadd>;
2755 defm SQDMLSL : SIMDLongThreeVectorSQDMLXTiedHS<0, 0b1011, "sqdmlsl",
2756 int_arm64_neon_sqsub>;
2757 defm SQDMULL : SIMDLongThreeVectorHS<0, 0b1101, "sqdmull",
2758 int_arm64_neon_sqdmull>;
2759 defm SSUBL : SIMDLongThreeVectorBHS<0, 0b0010, "ssubl",
2760 BinOpFrag<(sub (sext node:$LHS), (sext node:$RHS))>>;
2761 defm SSUBW : SIMDWideThreeVectorBHS<0, 0b0011, "ssubw",
2762 BinOpFrag<(sub node:$LHS, (sext node:$RHS))>>;
2763 defm UABAL : SIMDLongThreeVectorTiedBHSabal<1, 0b0101, "uabal",
2764 int_arm64_neon_uabd>;
2765 defm UABDL : SIMDLongThreeVectorBHSabdl<1, 0b0111, "uabdl",
2766 int_arm64_neon_uabd>;
2767 defm UADDL : SIMDLongThreeVectorBHS<1, 0b0000, "uaddl",
2768 BinOpFrag<(add (zext node:$LHS), (zext node:$RHS))>>;
2769 defm UADDW : SIMDWideThreeVectorBHS<1, 0b0001, "uaddw",
2770 BinOpFrag<(add node:$LHS, (zext node:$RHS))>>;
2771 defm UMLAL : SIMDLongThreeVectorTiedBHS<1, 0b1000, "umlal",
2772 TriOpFrag<(add node:$LHS, (int_arm64_neon_umull node:$MHS, node:$RHS))>>;
2773 defm UMLSL : SIMDLongThreeVectorTiedBHS<1, 0b1010, "umlsl",
2774 TriOpFrag<(sub node:$LHS, (int_arm64_neon_umull node:$MHS, node:$RHS))>>;
2775 defm UMULL : SIMDLongThreeVectorBHS<1, 0b1100, "umull", int_arm64_neon_umull>;
2776 defm USUBL : SIMDLongThreeVectorBHS<1, 0b0010, "usubl",
2777 BinOpFrag<(sub (zext node:$LHS), (zext node:$RHS))>>;
2778 defm USUBW : SIMDWideThreeVectorBHS< 1, 0b0011, "usubw",
2779 BinOpFrag<(sub node:$LHS, (zext node:$RHS))>>;
2781 // Patterns for 64-bit pmull
2782 def : Pat<(int_arm64_neon_pmull64 V64:$Rn, V64:$Rm),
2783 (PMULLv1i64 V64:$Rn, V64:$Rm)>;
2784 def : Pat<(int_arm64_neon_pmull64 (vector_extract (v2i64 V128:$Rn), (i64 1)),
2785 (vector_extract (v2i64 V128:$Rm), (i64 1))),
2786 (PMULLv2i64 V128:$Rn, V128:$Rm)>;
2788 // CodeGen patterns for addhn and subhn instructions, which can actually be
2789 // written in LLVM IR without too much difficulty.
2792 def : Pat<(v8i8 (trunc (v8i16 (ARM64vlshr (add V128:$Rn, V128:$Rm), (i32 8))))),
2793 (ADDHNv8i16_v8i8 V128:$Rn, V128:$Rm)>;
2794 def : Pat<(v4i16 (trunc (v4i32 (ARM64vlshr (add V128:$Rn, V128:$Rm),
2796 (ADDHNv4i32_v4i16 V128:$Rn, V128:$Rm)>;
2797 def : Pat<(v2i32 (trunc (v2i64 (ARM64vlshr (add V128:$Rn, V128:$Rm),
2799 (ADDHNv2i64_v2i32 V128:$Rn, V128:$Rm)>;
2800 def : Pat<(concat_vectors (v8i8 V64:$Rd),
2801 (trunc (v8i16 (ARM64vlshr (add V128:$Rn, V128:$Rm),
2803 (ADDHNv8i16_v16i8 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
2804 V128:$Rn, V128:$Rm)>;
2805 def : Pat<(concat_vectors (v4i16 V64:$Rd),
2806 (trunc (v4i32 (ARM64vlshr (add V128:$Rn, V128:$Rm),
2808 (ADDHNv4i32_v8i16 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
2809 V128:$Rn, V128:$Rm)>;
2810 def : Pat<(concat_vectors (v2i32 V64:$Rd),
2811 (trunc (v2i64 (ARM64vlshr (add V128:$Rn, V128:$Rm),
2813 (ADDHNv2i64_v4i32 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
2814 V128:$Rn, V128:$Rm)>;
2817 def : Pat<(v8i8 (trunc (v8i16 (ARM64vlshr (sub V128:$Rn, V128:$Rm), (i32 8))))),
2818 (SUBHNv8i16_v8i8 V128:$Rn, V128:$Rm)>;
2819 def : Pat<(v4i16 (trunc (v4i32 (ARM64vlshr (sub V128:$Rn, V128:$Rm),
2821 (SUBHNv4i32_v4i16 V128:$Rn, V128:$Rm)>;
2822 def : Pat<(v2i32 (trunc (v2i64 (ARM64vlshr (sub V128:$Rn, V128:$Rm),
2824 (SUBHNv2i64_v2i32 V128:$Rn, V128:$Rm)>;
2825 def : Pat<(concat_vectors (v8i8 V64:$Rd),
2826 (trunc (v8i16 (ARM64vlshr (sub V128:$Rn, V128:$Rm),
2828 (SUBHNv8i16_v16i8 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
2829 V128:$Rn, V128:$Rm)>;
2830 def : Pat<(concat_vectors (v4i16 V64:$Rd),
2831 (trunc (v4i32 (ARM64vlshr (sub V128:$Rn, V128:$Rm),
2833 (SUBHNv4i32_v8i16 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
2834 V128:$Rn, V128:$Rm)>;
2835 def : Pat<(concat_vectors (v2i32 V64:$Rd),
2836 (trunc (v2i64 (ARM64vlshr (sub V128:$Rn, V128:$Rm),
2838 (SUBHNv2i64_v4i32 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
2839 V128:$Rn, V128:$Rm)>;
2841 //----------------------------------------------------------------------------
2842 // AdvSIMD bitwise extract from vector instruction.
2843 //----------------------------------------------------------------------------
2845 defm EXT : SIMDBitwiseExtract<"ext">;
2847 def : Pat<(v4i16 (ARM64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
2848 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
2849 def : Pat<(v8i16 (ARM64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
2850 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
2851 def : Pat<(v2i32 (ARM64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
2852 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
2853 def : Pat<(v2f32 (ARM64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
2854 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
2855 def : Pat<(v4i32 (ARM64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
2856 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
2857 def : Pat<(v4f32 (ARM64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
2858 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
2859 def : Pat<(v2i64 (ARM64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
2860 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
2861 def : Pat<(v2f64 (ARM64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
2862 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
2864 // We use EXT to handle extract_subvector to copy the upper 64-bits of a
2866 def : Pat<(v8i8 (extract_subvector V128:$Rn, (i64 8))),
2867 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
2868 def : Pat<(v4i16 (extract_subvector V128:$Rn, (i64 4))),
2869 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
2870 def : Pat<(v2i32 (extract_subvector V128:$Rn, (i64 2))),
2871 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
2872 def : Pat<(v1i64 (extract_subvector V128:$Rn, (i64 1))),
2873 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
2874 def : Pat<(v2f32 (extract_subvector V128:$Rn, (i64 2))),
2875 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
2876 def : Pat<(v1f64 (extract_subvector V128:$Rn, (i64 1))),
2877 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
2880 //----------------------------------------------------------------------------
2881 // AdvSIMD zip vector
2882 //----------------------------------------------------------------------------
2884 defm TRN1 : SIMDZipVector<0b010, "trn1", ARM64trn1>;
2885 defm TRN2 : SIMDZipVector<0b110, "trn2", ARM64trn2>;
2886 defm UZP1 : SIMDZipVector<0b001, "uzp1", ARM64uzp1>;
2887 defm UZP2 : SIMDZipVector<0b101, "uzp2", ARM64uzp2>;
2888 defm ZIP1 : SIMDZipVector<0b011, "zip1", ARM64zip1>;
2889 defm ZIP2 : SIMDZipVector<0b111, "zip2", ARM64zip2>;
2891 //----------------------------------------------------------------------------
2892 // AdvSIMD TBL/TBX instructions
2893 //----------------------------------------------------------------------------
2895 defm TBL : SIMDTableLookup< 0, "tbl">;
2896 defm TBX : SIMDTableLookupTied<1, "tbx">;
2898 def : Pat<(v8i8 (int_arm64_neon_tbl1 (v16i8 VecListOne128:$Rn), (v8i8 V64:$Ri))),
2899 (TBLv8i8One VecListOne128:$Rn, V64:$Ri)>;
2900 def : Pat<(v16i8 (int_arm64_neon_tbl1 (v16i8 V128:$Ri), (v16i8 V128:$Rn))),
2901 (TBLv16i8One V128:$Ri, V128:$Rn)>;
2903 def : Pat<(v8i8 (int_arm64_neon_tbx1 (v8i8 V64:$Rd),
2904 (v16i8 VecListOne128:$Rn), (v8i8 V64:$Ri))),
2905 (TBXv8i8One V64:$Rd, VecListOne128:$Rn, V64:$Ri)>;
2906 def : Pat<(v16i8 (int_arm64_neon_tbx1 (v16i8 V128:$Rd),
2907 (v16i8 V128:$Ri), (v16i8 V128:$Rn))),
2908 (TBXv16i8One V128:$Rd, V128:$Ri, V128:$Rn)>;
2911 //----------------------------------------------------------------------------
2912 // AdvSIMD scalar CPY instruction
2913 //----------------------------------------------------------------------------
2915 defm CPY : SIMDScalarCPY<"cpy">;
2917 //----------------------------------------------------------------------------
2918 // AdvSIMD scalar pairwise instructions
2919 //----------------------------------------------------------------------------
2921 defm ADDP : SIMDPairwiseScalarD<0, 0b11011, "addp">;
2922 defm FADDP : SIMDPairwiseScalarSD<1, 0, 0b01101, "faddp">;
2923 defm FMAXNMP : SIMDPairwiseScalarSD<1, 0, 0b01100, "fmaxnmp">;
2924 defm FMAXP : SIMDPairwiseScalarSD<1, 0, 0b01111, "fmaxp">;
2925 defm FMINNMP : SIMDPairwiseScalarSD<1, 1, 0b01100, "fminnmp">;
2926 defm FMINP : SIMDPairwiseScalarSD<1, 1, 0b01111, "fminp">;
2927 def : Pat<(i64 (int_arm64_neon_saddv (v2i64 V128:$Rn))),
2928 (ADDPv2i64p V128:$Rn)>;
2929 def : Pat<(i64 (int_arm64_neon_uaddv (v2i64 V128:$Rn))),
2930 (ADDPv2i64p V128:$Rn)>;
2931 def : Pat<(f32 (int_arm64_neon_faddv (v2f32 V64:$Rn))),
2932 (FADDPv2i32p V64:$Rn)>;
2933 def : Pat<(f32 (int_arm64_neon_faddv (v4f32 V128:$Rn))),
2934 (FADDPv2i32p (EXTRACT_SUBREG (FADDPv4f32 V128:$Rn, V128:$Rn), dsub))>;
2935 def : Pat<(f64 (int_arm64_neon_faddv (v2f64 V128:$Rn))),
2936 (FADDPv2i64p V128:$Rn)>;
2937 def : Pat<(f32 (int_arm64_neon_fmaxnmv (v2f32 V64:$Rn))),
2938 (FMAXNMPv2i32p V64:$Rn)>;
2939 def : Pat<(f64 (int_arm64_neon_fmaxnmv (v2f64 V128:$Rn))),
2940 (FMAXNMPv2i64p V128:$Rn)>;
2941 def : Pat<(f32 (int_arm64_neon_fmaxv (v2f32 V64:$Rn))),
2942 (FMAXPv2i32p V64:$Rn)>;
2943 def : Pat<(f64 (int_arm64_neon_fmaxv (v2f64 V128:$Rn))),
2944 (FMAXPv2i64p V128:$Rn)>;
2945 def : Pat<(f32 (int_arm64_neon_fminnmv (v2f32 V64:$Rn))),
2946 (FMINNMPv2i32p V64:$Rn)>;
2947 def : Pat<(f64 (int_arm64_neon_fminnmv (v2f64 V128:$Rn))),
2948 (FMINNMPv2i64p V128:$Rn)>;
2949 def : Pat<(f32 (int_arm64_neon_fminv (v2f32 V64:$Rn))),
2950 (FMINPv2i32p V64:$Rn)>;
2951 def : Pat<(f64 (int_arm64_neon_fminv (v2f64 V128:$Rn))),
2952 (FMINPv2i64p V128:$Rn)>;
2954 //----------------------------------------------------------------------------
2955 // AdvSIMD INS/DUP instructions
2956 //----------------------------------------------------------------------------
2958 def DUPv8i8gpr : SIMDDupFromMain<0, 0b00001, ".8b", v8i8, V64, GPR32>;
2959 def DUPv16i8gpr : SIMDDupFromMain<1, 0b00001, ".16b", v16i8, V128, GPR32>;
2960 def DUPv4i16gpr : SIMDDupFromMain<0, 0b00010, ".4h", v4i16, V64, GPR32>;
2961 def DUPv8i16gpr : SIMDDupFromMain<1, 0b00010, ".8h", v8i16, V128, GPR32>;
2962 def DUPv2i32gpr : SIMDDupFromMain<0, 0b00100, ".2s", v2i32, V64, GPR32>;
2963 def DUPv4i32gpr : SIMDDupFromMain<1, 0b00100, ".4s", v4i32, V128, GPR32>;
2964 def DUPv2i64gpr : SIMDDupFromMain<1, 0b01000, ".2d", v2i64, V128, GPR64>;
2966 def DUPv2i64lane : SIMDDup64FromElement;
2967 def DUPv2i32lane : SIMDDup32FromElement<0, ".2s", v2i32, V64>;
2968 def DUPv4i32lane : SIMDDup32FromElement<1, ".4s", v4i32, V128>;
2969 def DUPv4i16lane : SIMDDup16FromElement<0, ".4h", v4i16, V64>;
2970 def DUPv8i16lane : SIMDDup16FromElement<1, ".8h", v8i16, V128>;
2971 def DUPv8i8lane : SIMDDup8FromElement <0, ".8b", v8i8, V64>;
2972 def DUPv16i8lane : SIMDDup8FromElement <1, ".16b", v16i8, V128>;
2974 def : Pat<(v2f32 (ARM64dup (f32 FPR32:$Rn))),
2975 (v2f32 (DUPv2i32lane
2976 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rn, ssub),
2978 def : Pat<(v4f32 (ARM64dup (f32 FPR32:$Rn))),
2979 (v4f32 (DUPv4i32lane
2980 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rn, ssub),
2982 def : Pat<(v2f64 (ARM64dup (f64 FPR64:$Rn))),
2983 (v2f64 (DUPv2i64lane
2984 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$Rn, dsub),
2987 def : Pat<(v2f32 (ARM64duplane32 (v4f32 V128:$Rn), VectorIndexS:$imm)),
2988 (DUPv2i32lane V128:$Rn, VectorIndexS:$imm)>;
2989 def : Pat<(v4f32 (ARM64duplane32 (v4f32 V128:$Rn), VectorIndexS:$imm)),
2990 (DUPv4i32lane V128:$Rn, VectorIndexS:$imm)>;
2991 def : Pat<(v2f64 (ARM64duplane64 (v2f64 V128:$Rn), VectorIndexD:$imm)),
2992 (DUPv2i64lane V128:$Rn, VectorIndexD:$imm)>;
2997 def : Pat<(sext_inreg (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx), i8),
2998 (i32 (SMOVvi8to32 V128:$Rn, VectorIndexB:$idx))>;
2999 def : Pat<(sext_inreg (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx), i8),
3000 (i64 (SMOVvi8to64 V128:$Rn, VectorIndexB:$idx))>;
3001 def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
3002 (i32 (SMOVvi16to32 V128:$Rn, VectorIndexH:$idx))>;
3003 def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
3004 (i64 (SMOVvi16to64 V128:$Rn, VectorIndexH:$idx))>;
3005 def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
3006 (i32 (SMOVvi16to32 V128:$Rn, VectorIndexH:$idx))>;
3007 def : Pat<(sext (i32 (vector_extract (v4i32 V128:$Rn), VectorIndexS:$idx))),
3008 (i64 (SMOVvi32to64 V128:$Rn, VectorIndexS:$idx))>;
3010 // Extracting i8 or i16 elements will have the zero-extend transformed to
3011 // an 'and' mask by type legalization since neither i8 nor i16 are legal types
3012 // for ARM64. Match these patterns here since UMOV already zeroes out the high
3013 // bits of the destination register.
3014 def : Pat<(and (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx),
3016 (i32 (UMOVvi8 V128:$Rn, VectorIndexB:$idx))>;
3017 def : Pat<(and (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),
3019 (i32 (UMOVvi16 V128:$Rn, VectorIndexH:$idx))>;
3023 def : Pat<(v16i8 (scalar_to_vector GPR32:$Rn)),
3024 (SUBREG_TO_REG (i32 0),
3025 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3026 def : Pat<(v8i8 (scalar_to_vector GPR32:$Rn)),
3027 (SUBREG_TO_REG (i32 0),
3028 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3030 def : Pat<(v8i16 (scalar_to_vector GPR32:$Rn)),
3031 (SUBREG_TO_REG (i32 0),
3032 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3033 def : Pat<(v4i16 (scalar_to_vector GPR32:$Rn)),
3034 (SUBREG_TO_REG (i32 0),
3035 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3037 def : Pat<(v2i32 (scalar_to_vector (i32 FPR32:$Rn))),
3038 (v2i32 (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)),
3039 (i32 FPR32:$Rn), ssub))>;
3040 def : Pat<(v4i32 (scalar_to_vector (i32 FPR32:$Rn))),
3041 (v4i32 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
3042 (i32 FPR32:$Rn), ssub))>;
3043 def : Pat<(v2i64 (scalar_to_vector (i64 FPR64:$Rn))),
3044 (v2i64 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
3045 (i64 FPR64:$Rn), dsub))>;
3047 def : Pat<(v4f32 (scalar_to_vector (f32 FPR32:$Rn))),
3048 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rn, ssub)>;
3049 def : Pat<(v2f32 (scalar_to_vector (f32 FPR32:$Rn))),
3050 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), FPR32:$Rn, ssub)>;
3051 def : Pat<(v2f64 (scalar_to_vector (f64 FPR64:$Rn))),
3052 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$Rn, dsub)>;
3054 def : Pat<(v2f32 (vector_insert (v2f32 V64:$Rn),
3055 (f32 FPR32:$Rm), (i64 VectorIndexS:$imm))),
3058 (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), V64:$Rn, dsub)),
3060 (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rm, ssub)),
3063 def : Pat<(v4f32 (vector_insert (v4f32 V128:$Rn),
3064 (f32 FPR32:$Rm), (i64 VectorIndexS:$imm))),
3066 V128:$Rn, VectorIndexS:$imm,
3067 (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rm, ssub)),
3069 def : Pat<(v2f64 (vector_insert (v2f64 V128:$Rn),
3070 (f64 FPR64:$Rm), (i64 VectorIndexD:$imm))),
3072 V128:$Rn, VectorIndexD:$imm,
3073 (v2f64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$Rm, dsub)),
3076 // Copy an element at a constant index in one vector into a constant indexed
3077 // element of another.
3078 // FIXME refactor to a shared class/dev parameterized on vector type, vector
3079 // index type and INS extension
3080 def : Pat<(v16i8 (int_arm64_neon_vcopy_lane
3081 (v16i8 V128:$Vd), VectorIndexB:$idx, (v16i8 V128:$Vs),
3082 VectorIndexB:$idx2)),
3084 V128:$Vd, VectorIndexB:$idx, V128:$Vs, VectorIndexB:$idx2)
3086 def : Pat<(v8i16 (int_arm64_neon_vcopy_lane
3087 (v8i16 V128:$Vd), VectorIndexH:$idx, (v8i16 V128:$Vs),
3088 VectorIndexH:$idx2)),
3090 V128:$Vd, VectorIndexH:$idx, V128:$Vs, VectorIndexH:$idx2)
3092 def : Pat<(v4i32 (int_arm64_neon_vcopy_lane
3093 (v4i32 V128:$Vd), VectorIndexS:$idx, (v4i32 V128:$Vs),
3094 VectorIndexS:$idx2)),
3096 V128:$Vd, VectorIndexS:$idx, V128:$Vs, VectorIndexS:$idx2)
3098 def : Pat<(v2i64 (int_arm64_neon_vcopy_lane
3099 (v2i64 V128:$Vd), VectorIndexD:$idx, (v2i64 V128:$Vs),
3100 VectorIndexD:$idx2)),
3102 V128:$Vd, VectorIndexD:$idx, V128:$Vs, VectorIndexD:$idx2)
3105 // Floating point vector extractions are codegen'd as either a sequence of
3106 // subregister extractions, possibly fed by an INS if the lane number is
3107 // anything other than zero.
3108 def : Pat<(vector_extract (v2f64 V128:$Rn), 0),
3109 (f64 (EXTRACT_SUBREG V128:$Rn, dsub))>;
3110 def : Pat<(vector_extract (v4f32 V128:$Rn), 0),
3111 (f32 (EXTRACT_SUBREG V128:$Rn, ssub))>;
3112 def : Pat<(vector_extract (v2f64 V128:$Rn), VectorIndexD:$idx),
3113 (f64 (EXTRACT_SUBREG
3114 (INSvi64lane (v2f64 (IMPLICIT_DEF)), 0,
3115 V128:$Rn, VectorIndexD:$idx),
3117 def : Pat<(vector_extract (v4f32 V128:$Rn), VectorIndexS:$idx),
3118 (f32 (EXTRACT_SUBREG
3119 (INSvi32lane (v4f32 (IMPLICIT_DEF)), 0,
3120 V128:$Rn, VectorIndexS:$idx),
3123 // All concat_vectors operations are canonicalised to act on i64 vectors for
3124 // ARM64. In the general case we need an instruction, which had just as well be
3126 class ConcatPat<ValueType DstTy, ValueType SrcTy>
3127 : Pat<(DstTy (concat_vectors (SrcTy V64:$Rd), V64:$Rn)),
3128 (INSvi64lane (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), 1,
3129 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rn, dsub), 0)>;
3131 def : ConcatPat<v2i64, v1i64>;
3132 def : ConcatPat<v2f64, v1f64>;
3133 def : ConcatPat<v4i32, v2i32>;
3134 def : ConcatPat<v4f32, v2f32>;
3135 def : ConcatPat<v8i16, v4i16>;
3136 def : ConcatPat<v16i8, v8i8>;
3138 // If the high lanes are undef, though, we can just ignore them:
3139 class ConcatUndefPat<ValueType DstTy, ValueType SrcTy>
3140 : Pat<(DstTy (concat_vectors (SrcTy V64:$Rn), undef)),
3141 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rn, dsub)>;
3143 def : ConcatUndefPat<v2i64, v1i64>;
3144 def : ConcatUndefPat<v2f64, v1f64>;
3145 def : ConcatUndefPat<v4i32, v2i32>;
3146 def : ConcatUndefPat<v4f32, v2f32>;
3147 def : ConcatUndefPat<v8i16, v4i16>;
3148 def : ConcatUndefPat<v16i8, v8i8>;
3150 //----------------------------------------------------------------------------
3151 // AdvSIMD across lanes instructions
3152 //----------------------------------------------------------------------------
3154 defm ADDV : SIMDAcrossLanesBHS<0, 0b11011, "addv">;
3155 defm SMAXV : SIMDAcrossLanesBHS<0, 0b01010, "smaxv">;
3156 defm SMINV : SIMDAcrossLanesBHS<0, 0b11010, "sminv">;
3157 defm UMAXV : SIMDAcrossLanesBHS<1, 0b01010, "umaxv">;
3158 defm UMINV : SIMDAcrossLanesBHS<1, 0b11010, "uminv">;
3159 defm SADDLV : SIMDAcrossLanesHSD<0, 0b00011, "saddlv">;
3160 defm UADDLV : SIMDAcrossLanesHSD<1, 0b00011, "uaddlv">;
3161 defm FMAXNMV : SIMDAcrossLanesS<0b01100, 0, "fmaxnmv", int_arm64_neon_fmaxnmv>;
3162 defm FMAXV : SIMDAcrossLanesS<0b01111, 0, "fmaxv", int_arm64_neon_fmaxv>;
3163 defm FMINNMV : SIMDAcrossLanesS<0b01100, 1, "fminnmv", int_arm64_neon_fminnmv>;
3164 defm FMINV : SIMDAcrossLanesS<0b01111, 1, "fminv", int_arm64_neon_fminv>;
3166 multiclass SIMDAcrossLanesSignedIntrinsic<string baseOpc, Intrinsic intOp> {
3167 // If there is a sign extension after this intrinsic, consume it as smov already
3169 def : Pat<(i32 (sext_inreg (i32 (intOp (v8i8 V64:$Rn))), i8)),
3171 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3172 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3174 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
3176 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3177 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3179 // If there is a sign extension after this intrinsic, consume it as smov already
3181 def : Pat<(i32 (sext_inreg (i32 (intOp (v16i8 V128:$Rn))), i8)),
3183 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3184 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3186 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
3188 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3189 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3191 // If there is a sign extension after this intrinsic, consume it as smov already
3193 def : Pat<(i32 (sext_inreg (i32 (intOp (v4i16 V64:$Rn))), i16)),
3195 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3196 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3198 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
3200 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3201 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3203 // If there is a sign extension after this intrinsic, consume it as smov already
3205 def : Pat<(i32 (sext_inreg (i32 (intOp (v8i16 V128:$Rn))), i16)),
3207 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3208 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3210 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
3212 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3213 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3216 def : Pat<(i32 (intOp (v4i32 V128:$Rn))),
3217 (i32 (EXTRACT_SUBREG
3218 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3219 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), ssub),
3223 multiclass SIMDAcrossLanesUnsignedIntrinsic<string baseOpc, Intrinsic intOp> {
3224 // If there is a masking operation keeping only what has been actually
3225 // generated, consume it.
3226 def : Pat<(i32 (and (i32 (intOp (v8i8 V64:$Rn))), maski8_or_more)),
3227 (i32 (EXTRACT_SUBREG
3228 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3229 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3231 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
3232 (i32 (EXTRACT_SUBREG
3233 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3234 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3236 // If there is a masking operation keeping only what has been actually
3237 // generated, consume it.
3238 def : Pat<(i32 (and (i32 (intOp (v16i8 V128:$Rn))), maski8_or_more)),
3239 (i32 (EXTRACT_SUBREG
3240 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3241 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3243 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
3244 (i32 (EXTRACT_SUBREG
3245 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3246 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3249 // If there is a masking operation keeping only what has been actually
3250 // generated, consume it.
3251 def : Pat<(i32 (and (i32 (intOp (v4i16 V64:$Rn))), maski16_or_more)),
3252 (i32 (EXTRACT_SUBREG
3253 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3254 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3256 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
3257 (i32 (EXTRACT_SUBREG
3258 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3259 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3261 // If there is a masking operation keeping only what has been actually
3262 // generated, consume it.
3263 def : Pat<(i32 (and (i32 (intOp (v8i16 V128:$Rn))), maski16_or_more)),
3264 (i32 (EXTRACT_SUBREG
3265 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3266 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3268 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
3269 (i32 (EXTRACT_SUBREG
3270 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3271 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3274 def : Pat<(i32 (intOp (v4i32 V128:$Rn))),
3275 (i32 (EXTRACT_SUBREG
3276 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3277 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), ssub),
3282 multiclass SIMDAcrossLanesSignedLongIntrinsic<string baseOpc, Intrinsic intOp> {
3283 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
3285 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3286 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), hsub),
3288 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
3290 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3291 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), hsub),
3294 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
3295 (i32 (EXTRACT_SUBREG
3296 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3297 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), ssub),
3299 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
3300 (i32 (EXTRACT_SUBREG
3301 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3302 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), ssub),
3305 def : Pat<(i64 (intOp (v4i32 V128:$Rn))),
3306 (i64 (EXTRACT_SUBREG
3307 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3308 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), dsub),
3312 multiclass SIMDAcrossLanesUnsignedLongIntrinsic<string baseOpc,
3314 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
3315 (i32 (EXTRACT_SUBREG
3316 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3317 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), hsub),
3319 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
3320 (i32 (EXTRACT_SUBREG
3321 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3322 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), hsub),
3325 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
3326 (i32 (EXTRACT_SUBREG
3327 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3328 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), ssub),
3330 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
3331 (i32 (EXTRACT_SUBREG
3332 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3333 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), ssub),
3336 def : Pat<(i64 (intOp (v4i32 V128:$Rn))),
3337 (i64 (EXTRACT_SUBREG
3338 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3339 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), dsub),
3343 defm : SIMDAcrossLanesSignedIntrinsic<"ADDV", int_arm64_neon_saddv>;
3344 // vaddv_[su]32 is special; -> ADDP Vd.2S,Vn.2S,Vm.2S; return Vd.s[0];Vn==Vm
3345 def : Pat<(i32 (int_arm64_neon_saddv (v2i32 V64:$Rn))),
3346 (EXTRACT_SUBREG (ADDPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3348 defm : SIMDAcrossLanesUnsignedIntrinsic<"ADDV", int_arm64_neon_uaddv>;
3349 // vaddv_[su]32 is special; -> ADDP Vd.2S,Vn.2S,Vm.2S; return Vd.s[0];Vn==Vm
3350 def : Pat<(i32 (int_arm64_neon_uaddv (v2i32 V64:$Rn))),
3351 (EXTRACT_SUBREG (ADDPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3353 defm : SIMDAcrossLanesSignedIntrinsic<"SMAXV", int_arm64_neon_smaxv>;
3354 def : Pat<(i32 (int_arm64_neon_smaxv (v2i32 V64:$Rn))),
3355 (EXTRACT_SUBREG (SMAXPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3357 defm : SIMDAcrossLanesSignedIntrinsic<"SMINV", int_arm64_neon_sminv>;
3358 def : Pat<(i32 (int_arm64_neon_sminv (v2i32 V64:$Rn))),
3359 (EXTRACT_SUBREG (SMINPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3361 defm : SIMDAcrossLanesUnsignedIntrinsic<"UMAXV", int_arm64_neon_umaxv>;
3362 def : Pat<(i32 (int_arm64_neon_umaxv (v2i32 V64:$Rn))),
3363 (EXTRACT_SUBREG (UMAXPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3365 defm : SIMDAcrossLanesUnsignedIntrinsic<"UMINV", int_arm64_neon_uminv>;
3366 def : Pat<(i32 (int_arm64_neon_uminv (v2i32 V64:$Rn))),
3367 (EXTRACT_SUBREG (UMINPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3369 defm : SIMDAcrossLanesSignedLongIntrinsic<"SADDLV", int_arm64_neon_saddlv>;
3370 defm : SIMDAcrossLanesUnsignedLongIntrinsic<"UADDLV", int_arm64_neon_uaddlv>;
3372 // The vaddlv_s32 intrinsic gets mapped to SADDLP.
3373 def : Pat<(i64 (int_arm64_neon_saddlv (v2i32 V64:$Rn))),
3374 (i64 (EXTRACT_SUBREG
3375 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3376 (SADDLPv2i32_v1i64 V64:$Rn), dsub),
3378 // The vaddlv_u32 intrinsic gets mapped to UADDLP.
3379 def : Pat<(i64 (int_arm64_neon_uaddlv (v2i32 V64:$Rn))),
3380 (i64 (EXTRACT_SUBREG
3381 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3382 (UADDLPv2i32_v1i64 V64:$Rn), dsub),
3385 //------------------------------------------------------------------------------
3386 // AdvSIMD modified immediate instructions
3387 //------------------------------------------------------------------------------
3390 defm BIC : SIMDModifiedImmVectorShiftTied<1, 0b11, 0b01, "bic", ARM64bici>;
3392 defm ORR : SIMDModifiedImmVectorShiftTied<0, 0b11, 0b01, "orr", ARM64orri>;
3396 def FMOVv2f64_ns : SIMDModifiedImmVectorNoShift<1, 1, 0b1111, V128, fpimm8,
3398 [(set (v2f64 V128:$Rd), (ARM64fmov imm0_255:$imm8))]>;
3399 def FMOVv2f32_ns : SIMDModifiedImmVectorNoShift<0, 0, 0b1111, V64, fpimm8,
3401 [(set (v2f32 V64:$Rd), (ARM64fmov imm0_255:$imm8))]>;
3402 def FMOVv4f32_ns : SIMDModifiedImmVectorNoShift<1, 0, 0b1111, V128, fpimm8,
3404 [(set (v4f32 V128:$Rd), (ARM64fmov imm0_255:$imm8))]>;
3408 // EDIT byte mask: scalar
3409 let isReMaterializable = 1, isAsCheapAsAMove = 1 in
3410 def MOVID : SIMDModifiedImmScalarNoShift<0, 1, 0b1110, "movi",
3411 [(set FPR64:$Rd, simdimmtype10:$imm8)]>;
3412 // The movi_edit node has the immediate value already encoded, so we use
3413 // a plain imm0_255 here.
3414 def : Pat<(f64 (ARM64movi_edit imm0_255:$shift)),
3415 (MOVID imm0_255:$shift)>;
3417 def : Pat<(v1i64 immAllZerosV), (MOVID (i32 0))>;
3418 def : Pat<(v2i32 immAllZerosV), (MOVID (i32 0))>;
3419 def : Pat<(v4i16 immAllZerosV), (MOVID (i32 0))>;
3420 def : Pat<(v8i8 immAllZerosV), (MOVID (i32 0))>;
3422 def : Pat<(v1i64 immAllOnesV), (MOVID (i32 255))>;
3423 def : Pat<(v2i32 immAllOnesV), (MOVID (i32 255))>;
3424 def : Pat<(v4i16 immAllOnesV), (MOVID (i32 255))>;
3425 def : Pat<(v8i8 immAllOnesV), (MOVID (i32 255))>;
3427 // EDIT byte mask: 2d
3429 // The movi_edit node has the immediate value already encoded, so we use
3430 // a plain imm0_255 in the pattern
3431 let isReMaterializable = 1, isAsCheapAsAMove = 1 in
3432 def MOVIv2d_ns : SIMDModifiedImmVectorNoShift<1, 1, 0b1110, V128,
3435 [(set (v2i64 V128:$Rd), (ARM64movi_edit imm0_255:$imm8))]>;
3438 // Use movi.2d to materialize 0.0 if the HW does zero-cycle zeroing.
3439 // Complexity is added to break a tie with a plain MOVI.
3440 let AddedComplexity = 1 in {
3441 def : Pat<(f32 fpimm0),
3442 (f32 (EXTRACT_SUBREG (v2i64 (MOVIv2d_ns (i32 0))), ssub))>,
3444 def : Pat<(f64 fpimm0),
3445 (f64 (EXTRACT_SUBREG (v2i64 (MOVIv2d_ns (i32 0))), dsub))>,
3449 def : Pat<(v2i64 immAllZerosV), (MOVIv2d_ns (i32 0))>;
3450 def : Pat<(v4i32 immAllZerosV), (MOVIv2d_ns (i32 0))>;
3451 def : Pat<(v8i16 immAllZerosV), (MOVIv2d_ns (i32 0))>;
3452 def : Pat<(v16i8 immAllZerosV), (MOVIv2d_ns (i32 0))>;
3454 def : Pat<(v2i64 immAllOnesV), (MOVIv2d_ns (i32 255))>;
3455 def : Pat<(v4i32 immAllOnesV), (MOVIv2d_ns (i32 255))>;
3456 def : Pat<(v8i16 immAllOnesV), (MOVIv2d_ns (i32 255))>;
3457 def : Pat<(v16i8 immAllOnesV), (MOVIv2d_ns (i32 255))>;
3459 // EDIT per word & halfword: 2s, 4h, 4s, & 8h
3460 defm MOVI : SIMDModifiedImmVectorShift<0, 0b10, 0b00, "movi">;
3461 def : Pat<(v2i32 (ARM64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
3462 (MOVIv2i32 imm0_255:$imm8, imm:$shift)>;
3463 def : Pat<(v4i32 (ARM64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
3464 (MOVIv4i32 imm0_255:$imm8, imm:$shift)>;
3465 def : Pat<(v4i16 (ARM64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
3466 (MOVIv4i16 imm0_255:$imm8, imm:$shift)>;
3467 def : Pat<(v8i16 (ARM64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
3468 (MOVIv8i16 imm0_255:$imm8, imm:$shift)>;
3470 // EDIT per word: 2s & 4s with MSL shifter
3471 def MOVIv2s_msl : SIMDModifiedImmMoveMSL<0, 0, {1,1,0,?}, V64, "movi", ".2s",
3472 [(set (v2i32 V64:$Rd),
3473 (ARM64movi_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
3474 def MOVIv4s_msl : SIMDModifiedImmMoveMSL<1, 0, {1,1,0,?}, V128, "movi", ".4s",
3475 [(set (v4i32 V128:$Rd),
3476 (ARM64movi_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
3478 // Per byte: 8b & 16b
3479 def MOVIv8b_ns : SIMDModifiedImmVectorNoShift<0, 0, 0b1110, V64, imm0_255,
3481 [(set (v8i8 V64:$Rd), (ARM64movi imm0_255:$imm8))]>;
3482 def MOVIv16b_ns : SIMDModifiedImmVectorNoShift<1, 0, 0b1110, V128, imm0_255,
3484 [(set (v16i8 V128:$Rd), (ARM64movi imm0_255:$imm8))]>;
3488 // EDIT per word & halfword: 2s, 4h, 4s, & 8h
3489 defm MVNI : SIMDModifiedImmVectorShift<1, 0b10, 0b00, "mvni">;
3490 def : Pat<(v2i32 (ARM64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
3491 (MVNIv2i32 imm0_255:$imm8, imm:$shift)>;
3492 def : Pat<(v4i32 (ARM64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
3493 (MVNIv4i32 imm0_255:$imm8, imm:$shift)>;
3494 def : Pat<(v4i16 (ARM64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
3495 (MVNIv4i16 imm0_255:$imm8, imm:$shift)>;
3496 def : Pat<(v8i16 (ARM64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
3497 (MVNIv8i16 imm0_255:$imm8, imm:$shift)>;
3499 // EDIT per word: 2s & 4s with MSL shifter
3500 def MVNIv2s_msl : SIMDModifiedImmMoveMSL<0, 1, {1,1,0,?}, V64, "mvni", ".2s",
3501 [(set (v2i32 V64:$Rd),
3502 (ARM64mvni_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
3503 def MVNIv4s_msl : SIMDModifiedImmMoveMSL<1, 1, {1,1,0,?}, V128, "mvni", ".4s",
3504 [(set (v4i32 V128:$Rd),
3505 (ARM64mvni_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
3507 //----------------------------------------------------------------------------
3508 // AdvSIMD indexed element
3509 //----------------------------------------------------------------------------
3511 let neverHasSideEffects = 1 in {
3512 defm FMLA : SIMDFPIndexedSDTied<0, 0b0001, "fmla">;
3513 defm FMLS : SIMDFPIndexedSDTied<0, 0b0101, "fmls">;
3516 // NOTE: Operands are reordered in the FMLA/FMLS PatFrags because the
3517 // instruction expects the addend first, while the intrinsic expects it last.
3519 // On the other hand, there are quite a few valid combinatorial options due to
3520 // the commutativity of multiplication and the fact that (-x) * y = x * (-y).
3521 defm : SIMDFPIndexedSDTiedPatterns<"FMLA",
3522 TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)>>;
3523 defm : SIMDFPIndexedSDTiedPatterns<"FMLA",
3524 TriOpFrag<(fma node:$MHS, node:$RHS, node:$LHS)>>;
3526 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
3527 TriOpFrag<(fma node:$MHS, (fneg node:$RHS), node:$LHS)> >;
3528 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
3529 TriOpFrag<(fma node:$RHS, (fneg node:$MHS), node:$LHS)> >;
3530 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
3531 TriOpFrag<(fma (fneg node:$RHS), node:$MHS, node:$LHS)> >;
3532 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
3533 TriOpFrag<(fma (fneg node:$MHS), node:$RHS, node:$LHS)> >;
3535 multiclass FMLSIndexedAfterNegPatterns<SDPatternOperator OpNode> {
3536 // 3 variants for the .2s version: DUPLANE from 128-bit, DUPLANE from 64-bit
3538 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
3539 (ARM64duplane32 (v4f32 (fneg V128:$Rm)),
3540 VectorIndexS:$idx))),
3541 (FMLSv2i32_indexed V64:$Rd, V64:$Rn, V128:$Rm, VectorIndexS:$idx)>;
3542 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
3543 (v2f32 (ARM64duplane32
3544 (v4f32 (insert_subvector undef,
3545 (v2f32 (fneg V64:$Rm)),
3547 VectorIndexS:$idx)))),
3548 (FMLSv2i32_indexed V64:$Rd, V64:$Rn,
3549 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub),
3550 VectorIndexS:$idx)>;
3551 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
3552 (ARM64dup (f32 (fneg FPR32Op:$Rm))))),
3553 (FMLSv2i32_indexed V64:$Rd, V64:$Rn,
3554 (SUBREG_TO_REG (i32 0), FPR32Op:$Rm, ssub), (i64 0))>;
3556 // 3 variants for the .4s version: DUPLANE from 128-bit, DUPLANE from 64-bit
3558 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
3559 (ARM64duplane32 (v4f32 (fneg V128:$Rm)),
3560 VectorIndexS:$idx))),
3561 (FMLSv4i32_indexed V128:$Rd, V128:$Rn, V128:$Rm,
3562 VectorIndexS:$idx)>;
3563 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
3564 (v4f32 (ARM64duplane32
3565 (v4f32 (insert_subvector undef,
3566 (v2f32 (fneg V64:$Rm)),
3568 VectorIndexS:$idx)))),
3569 (FMLSv4i32_indexed V128:$Rd, V128:$Rn,
3570 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub),
3571 VectorIndexS:$idx)>;
3572 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
3573 (ARM64dup (f32 (fneg FPR32Op:$Rm))))),
3574 (FMLSv4i32_indexed V128:$Rd, V128:$Rn,
3575 (SUBREG_TO_REG (i32 0), FPR32Op:$Rm, ssub), (i64 0))>;
3577 // 2 variants for the .2d version: DUPLANE from 128-bit, and DUP scalar
3578 // (DUPLANE from 64-bit would be trivial).
3579 def : Pat<(v2f64 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn),
3580 (ARM64duplane64 (v2f64 (fneg V128:$Rm)),
3581 VectorIndexD:$idx))),
3583 V128:$Rd, V128:$Rn, V128:$Rm, VectorIndexS:$idx)>;
3584 def : Pat<(v2f64 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn),
3585 (ARM64dup (f64 (fneg FPR64Op:$Rm))))),
3586 (FMLSv2i64_indexed V128:$Rd, V128:$Rn,
3587 (SUBREG_TO_REG (i32 0), FPR64Op:$Rm, dsub), (i64 0))>;
3589 // 2 variants for 32-bit scalar version: extract from .2s or from .4s
3590 def : Pat<(f32 (OpNode (f32 FPR32:$Rd), (f32 FPR32:$Rn),
3591 (vector_extract (v4f32 (fneg V128:$Rm)),
3592 VectorIndexS:$idx))),
3593 (FMLSv1i32_indexed FPR32:$Rd, FPR32:$Rn,
3594 V128:$Rm, VectorIndexS:$idx)>;
3595 def : Pat<(f32 (OpNode (f32 FPR32:$Rd), (f32 FPR32:$Rn),
3596 (vector_extract (v2f32 (fneg V64:$Rm)),
3597 VectorIndexS:$idx))),
3598 (FMLSv1i32_indexed FPR32:$Rd, FPR32:$Rn,
3599 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub), VectorIndexS:$idx)>;
3601 // 1 variant for 64-bit scalar version: extract from .1d or from .2d
3602 def : Pat<(f64 (OpNode (f64 FPR64:$Rd), (f64 FPR64:$Rn),
3603 (vector_extract (v2f64 (fneg V128:$Rm)),
3604 VectorIndexS:$idx))),
3605 (FMLSv1i64_indexed FPR64:$Rd, FPR64:$Rn,
3606 V128:$Rm, VectorIndexS:$idx)>;
3609 defm : FMLSIndexedAfterNegPatterns<
3610 TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)> >;
3611 defm : FMLSIndexedAfterNegPatterns<
3612 TriOpFrag<(fma node:$MHS, node:$RHS, node:$LHS)> >;
3614 defm FMULX : SIMDFPIndexedSD<1, 0b1001, "fmulx", int_arm64_neon_fmulx>;
3615 defm FMUL : SIMDFPIndexedSD<0, 0b1001, "fmul", fmul>;
3617 def : Pat<(v2f32 (fmul V64:$Rn, (ARM64dup (f32 FPR32:$Rm)))),
3618 (FMULv2i32_indexed V64:$Rn,
3619 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rm, ssub),
3621 def : Pat<(v4f32 (fmul V128:$Rn, (ARM64dup (f32 FPR32:$Rm)))),
3622 (FMULv4i32_indexed V128:$Rn,
3623 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rm, ssub),
3625 def : Pat<(v2f64 (fmul V128:$Rn, (ARM64dup (f64 FPR64:$Rm)))),
3626 (FMULv2i64_indexed V128:$Rn,
3627 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$Rm, dsub),
3630 defm SQDMULH : SIMDIndexedHS<0, 0b1100, "sqdmulh", int_arm64_neon_sqdmulh>;
3631 defm SQRDMULH : SIMDIndexedHS<0, 0b1101, "sqrdmulh", int_arm64_neon_sqrdmulh>;
3632 defm MLA : SIMDVectorIndexedHSTied<1, 0b0000, "mla",
3633 TriOpFrag<(add node:$LHS, (mul node:$MHS, node:$RHS))>>;
3634 defm MLS : SIMDVectorIndexedHSTied<1, 0b0100, "mls",
3635 TriOpFrag<(sub node:$LHS, (mul node:$MHS, node:$RHS))>>;
3636 defm MUL : SIMDVectorIndexedHS<0, 0b1000, "mul", mul>;
3637 defm SMLAL : SIMDVectorIndexedLongSDTied<0, 0b0010, "smlal",
3638 TriOpFrag<(add node:$LHS, (int_arm64_neon_smull node:$MHS, node:$RHS))>>;
3639 defm SMLSL : SIMDVectorIndexedLongSDTied<0, 0b0110, "smlsl",
3640 TriOpFrag<(sub node:$LHS, (int_arm64_neon_smull node:$MHS, node:$RHS))>>;
3641 defm SMULL : SIMDVectorIndexedLongSD<0, 0b1010, "smull",
3642 int_arm64_neon_smull>;
3643 defm SQDMLAL : SIMDIndexedLongSQDMLXSDTied<0, 0b0011, "sqdmlal",
3644 int_arm64_neon_sqadd>;
3645 defm SQDMLSL : SIMDIndexedLongSQDMLXSDTied<0, 0b0111, "sqdmlsl",
3646 int_arm64_neon_sqsub>;
3647 defm SQDMULL : SIMDIndexedLongSD<0, 0b1011, "sqdmull", int_arm64_neon_sqdmull>;
3648 defm UMLAL : SIMDVectorIndexedLongSDTied<1, 0b0010, "umlal",
3649 TriOpFrag<(add node:$LHS, (int_arm64_neon_umull node:$MHS, node:$RHS))>>;
3650 defm UMLSL : SIMDVectorIndexedLongSDTied<1, 0b0110, "umlsl",
3651 TriOpFrag<(sub node:$LHS, (int_arm64_neon_umull node:$MHS, node:$RHS))>>;
3652 defm UMULL : SIMDVectorIndexedLongSD<1, 0b1010, "umull",
3653 int_arm64_neon_umull>;
3655 // A scalar sqdmull with the second operand being a vector lane can be
3656 // handled directly with the indexed instruction encoding.
3657 def : Pat<(int_arm64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
3658 (vector_extract (v4i32 V128:$Vm),
3659 VectorIndexS:$idx)),
3660 (SQDMULLv1i64_indexed FPR32:$Rn, V128:$Vm, VectorIndexS:$idx)>;
3662 //----------------------------------------------------------------------------
3663 // AdvSIMD scalar shift instructions
3664 //----------------------------------------------------------------------------
3665 defm FCVTZS : SIMDScalarRShiftSD<0, 0b11111, "fcvtzs">;
3666 defm FCVTZU : SIMDScalarRShiftSD<1, 0b11111, "fcvtzu">;
3667 defm SCVTF : SIMDScalarRShiftSD<0, 0b11100, "scvtf">;
3668 defm UCVTF : SIMDScalarRShiftSD<1, 0b11100, "ucvtf">;
3669 // Codegen patterns for the above. We don't put these directly on the
3670 // instructions because TableGen's type inference can't handle the truth.
3671 // Having the same base pattern for fp <--> int totally freaks it out.
3672 def : Pat<(int_arm64_neon_vcvtfp2fxs FPR32:$Rn, vecshiftR32:$imm),
3673 (FCVTZSs FPR32:$Rn, vecshiftR32:$imm)>;
3674 def : Pat<(int_arm64_neon_vcvtfp2fxu FPR32:$Rn, vecshiftR32:$imm),
3675 (FCVTZUs FPR32:$Rn, vecshiftR32:$imm)>;
3676 def : Pat<(i64 (int_arm64_neon_vcvtfp2fxs (f64 FPR64:$Rn), vecshiftR64:$imm)),
3677 (FCVTZSd FPR64:$Rn, vecshiftR64:$imm)>;
3678 def : Pat<(i64 (int_arm64_neon_vcvtfp2fxu (f64 FPR64:$Rn), vecshiftR64:$imm)),
3679 (FCVTZUd FPR64:$Rn, vecshiftR64:$imm)>;
3680 def : Pat<(v1i64 (int_arm64_neon_vcvtfp2fxs (v1f64 FPR64:$Rn),
3682 (FCVTZSd FPR64:$Rn, vecshiftR64:$imm)>;
3683 def : Pat<(v1i64 (int_arm64_neon_vcvtfp2fxu (v1f64 FPR64:$Rn),
3685 (FCVTZUd FPR64:$Rn, vecshiftR64:$imm)>;
3686 def : Pat<(int_arm64_neon_vcvtfxs2fp FPR32:$Rn, vecshiftR32:$imm),
3687 (SCVTFs FPR32:$Rn, vecshiftR32:$imm)>;
3688 def : Pat<(int_arm64_neon_vcvtfxu2fp FPR32:$Rn, vecshiftR32:$imm),
3689 (UCVTFs FPR32:$Rn, vecshiftR32:$imm)>;
3690 def : Pat<(f64 (int_arm64_neon_vcvtfxs2fp (i64 FPR64:$Rn), vecshiftR64:$imm)),
3691 (SCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
3692 def : Pat<(f64 (int_arm64_neon_vcvtfxu2fp (i64 FPR64:$Rn), vecshiftR64:$imm)),
3693 (UCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
3694 def : Pat<(v1f64 (int_arm64_neon_vcvtfxs2fp (v1i64 FPR64:$Rn),
3696 (SCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
3697 def : Pat<(v1f64 (int_arm64_neon_vcvtfxu2fp (v1i64 FPR64:$Rn),
3699 (UCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
3701 defm SHL : SIMDScalarLShiftD< 0, 0b01010, "shl", ARM64vshl>;
3702 defm SLI : SIMDScalarLShiftDTied<1, 0b01010, "sli">;
3703 defm SQRSHRN : SIMDScalarRShiftBHS< 0, 0b10011, "sqrshrn",
3704 int_arm64_neon_sqrshrn>;
3705 defm SQRSHRUN : SIMDScalarRShiftBHS< 1, 0b10001, "sqrshrun",
3706 int_arm64_neon_sqrshrun>;
3707 defm SQSHLU : SIMDScalarLShiftBHSD<1, 0b01100, "sqshlu", ARM64sqshlui>;
3708 defm SQSHL : SIMDScalarLShiftBHSD<0, 0b01110, "sqshl", ARM64sqshli>;
3709 defm SQSHRN : SIMDScalarRShiftBHS< 0, 0b10010, "sqshrn",
3710 int_arm64_neon_sqshrn>;
3711 defm SQSHRUN : SIMDScalarRShiftBHS< 1, 0b10000, "sqshrun",
3712 int_arm64_neon_sqshrun>;
3713 defm SRI : SIMDScalarRShiftDTied< 1, 0b01000, "sri">;
3714 defm SRSHR : SIMDScalarRShiftD< 0, 0b00100, "srshr", ARM64srshri>;
3715 defm SRSRA : SIMDScalarRShiftDTied< 0, 0b00110, "srsra",
3716 TriOpFrag<(add node:$LHS,
3717 (ARM64srshri node:$MHS, node:$RHS))>>;
3718 defm SSHR : SIMDScalarRShiftD< 0, 0b00000, "sshr", ARM64vashr>;
3719 defm SSRA : SIMDScalarRShiftDTied< 0, 0b00010, "ssra",
3720 TriOpFrag<(add node:$LHS,
3721 (ARM64vashr node:$MHS, node:$RHS))>>;
3722 defm UQRSHRN : SIMDScalarRShiftBHS< 1, 0b10011, "uqrshrn",
3723 int_arm64_neon_uqrshrn>;
3724 defm UQSHL : SIMDScalarLShiftBHSD<1, 0b01110, "uqshl", ARM64uqshli>;
3725 defm UQSHRN : SIMDScalarRShiftBHS< 1, 0b10010, "uqshrn",
3726 int_arm64_neon_uqshrn>;
3727 defm URSHR : SIMDScalarRShiftD< 1, 0b00100, "urshr", ARM64urshri>;
3728 defm URSRA : SIMDScalarRShiftDTied< 1, 0b00110, "ursra",
3729 TriOpFrag<(add node:$LHS,
3730 (ARM64urshri node:$MHS, node:$RHS))>>;
3731 defm USHR : SIMDScalarRShiftD< 1, 0b00000, "ushr", ARM64vlshr>;
3732 defm USRA : SIMDScalarRShiftDTied< 1, 0b00010, "usra",
3733 TriOpFrag<(add node:$LHS,
3734 (ARM64vlshr node:$MHS, node:$RHS))>>;
3736 //----------------------------------------------------------------------------
3737 // AdvSIMD vector shift instructions
3738 //----------------------------------------------------------------------------
3739 defm FCVTZS:SIMDVectorRShiftSD<0, 0b11111, "fcvtzs", int_arm64_neon_vcvtfp2fxs>;
3740 defm FCVTZU:SIMDVectorRShiftSD<1, 0b11111, "fcvtzu", int_arm64_neon_vcvtfp2fxu>;
3741 defm SCVTF: SIMDVectorRShiftSDToFP<0, 0b11100, "scvtf",
3742 int_arm64_neon_vcvtfxs2fp>;
3743 defm RSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10001, "rshrn",
3744 int_arm64_neon_rshrn>;
3745 defm SHL : SIMDVectorLShiftBHSD<0, 0b01010, "shl", ARM64vshl>;
3746 defm SHRN : SIMDVectorRShiftNarrowBHS<0, 0b10000, "shrn",
3747 BinOpFrag<(trunc (ARM64vashr node:$LHS, node:$RHS))>>;
3748 defm SLI : SIMDVectorLShiftBHSDTied<1, 0b01010, "sli", int_arm64_neon_vsli>;
3749 def : Pat<(v1i64 (int_arm64_neon_vsli (v1i64 FPR64:$Rd), (v1i64 FPR64:$Rn),
3750 (i32 vecshiftL64:$imm))),
3751 (SLId FPR64:$Rd, FPR64:$Rn, vecshiftL64:$imm)>;
3752 defm SQRSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10011, "sqrshrn",
3753 int_arm64_neon_sqrshrn>;
3754 defm SQRSHRUN: SIMDVectorRShiftNarrowBHS<1, 0b10001, "sqrshrun",
3755 int_arm64_neon_sqrshrun>;
3756 defm SQSHLU : SIMDVectorLShiftBHSD<1, 0b01100, "sqshlu", ARM64sqshlui>;
3757 defm SQSHL : SIMDVectorLShiftBHSD<0, 0b01110, "sqshl", ARM64sqshli>;
3758 defm SQSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10010, "sqshrn",
3759 int_arm64_neon_sqshrn>;
3760 defm SQSHRUN : SIMDVectorRShiftNarrowBHS<1, 0b10000, "sqshrun",
3761 int_arm64_neon_sqshrun>;
3762 defm SRI : SIMDVectorRShiftBHSDTied<1, 0b01000, "sri", int_arm64_neon_vsri>;
3763 def : Pat<(v1i64 (int_arm64_neon_vsri (v1i64 FPR64:$Rd), (v1i64 FPR64:$Rn),
3764 (i32 vecshiftR64:$imm))),
3765 (SRId FPR64:$Rd, FPR64:$Rn, vecshiftR64:$imm)>;
3766 defm SRSHR : SIMDVectorRShiftBHSD<0, 0b00100, "srshr", ARM64srshri>;
3767 defm SRSRA : SIMDVectorRShiftBHSDTied<0, 0b00110, "srsra",
3768 TriOpFrag<(add node:$LHS,
3769 (ARM64srshri node:$MHS, node:$RHS))> >;
3770 defm SSHLL : SIMDVectorLShiftLongBHSD<0, 0b10100, "sshll",
3771 BinOpFrag<(ARM64vshl (sext node:$LHS), node:$RHS)>>;
3773 defm SSHR : SIMDVectorRShiftBHSD<0, 0b00000, "sshr", ARM64vashr>;
3774 defm SSRA : SIMDVectorRShiftBHSDTied<0, 0b00010, "ssra",
3775 TriOpFrag<(add node:$LHS, (ARM64vashr node:$MHS, node:$RHS))>>;
3776 defm UCVTF : SIMDVectorRShiftSDToFP<1, 0b11100, "ucvtf",
3777 int_arm64_neon_vcvtfxu2fp>;
3778 defm UQRSHRN : SIMDVectorRShiftNarrowBHS<1, 0b10011, "uqrshrn",
3779 int_arm64_neon_uqrshrn>;
3780 defm UQSHL : SIMDVectorLShiftBHSD<1, 0b01110, "uqshl", ARM64uqshli>;
3781 defm UQSHRN : SIMDVectorRShiftNarrowBHS<1, 0b10010, "uqshrn",
3782 int_arm64_neon_uqshrn>;
3783 defm URSHR : SIMDVectorRShiftBHSD<1, 0b00100, "urshr", ARM64urshri>;
3784 defm URSRA : SIMDVectorRShiftBHSDTied<1, 0b00110, "ursra",
3785 TriOpFrag<(add node:$LHS,
3786 (ARM64urshri node:$MHS, node:$RHS))> >;
3787 defm USHLL : SIMDVectorLShiftLongBHSD<1, 0b10100, "ushll",
3788 BinOpFrag<(ARM64vshl (zext node:$LHS), node:$RHS)>>;
3789 defm USHR : SIMDVectorRShiftBHSD<1, 0b00000, "ushr", ARM64vlshr>;
3790 defm USRA : SIMDVectorRShiftBHSDTied<1, 0b00010, "usra",
3791 TriOpFrag<(add node:$LHS, (ARM64vlshr node:$MHS, node:$RHS))> >;
3793 // SHRN patterns for when a logical right shift was used instead of arithmetic
3794 // (the immediate guarantees no sign bits actually end up in the result so it
3796 def : Pat<(v8i8 (trunc (ARM64vlshr (v8i16 V128:$Rn), vecshiftR16Narrow:$imm))),
3797 (SHRNv8i8_shift V128:$Rn, vecshiftR16Narrow:$imm)>;
3798 def : Pat<(v4i16 (trunc (ARM64vlshr (v4i32 V128:$Rn), vecshiftR32Narrow:$imm))),
3799 (SHRNv4i16_shift V128:$Rn, vecshiftR32Narrow:$imm)>;
3800 def : Pat<(v2i32 (trunc (ARM64vlshr (v2i64 V128:$Rn), vecshiftR64Narrow:$imm))),
3801 (SHRNv2i32_shift V128:$Rn, vecshiftR64Narrow:$imm)>;
3803 def : Pat<(v16i8 (concat_vectors (v8i8 V64:$Rd),
3804 (trunc (ARM64vlshr (v8i16 V128:$Rn),
3805 vecshiftR16Narrow:$imm)))),
3806 (SHRNv16i8_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
3807 V128:$Rn, vecshiftR16Narrow:$imm)>;
3808 def : Pat<(v8i16 (concat_vectors (v4i16 V64:$Rd),
3809 (trunc (ARM64vlshr (v4i32 V128:$Rn),
3810 vecshiftR32Narrow:$imm)))),
3811 (SHRNv8i16_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
3812 V128:$Rn, vecshiftR32Narrow:$imm)>;
3813 def : Pat<(v4i32 (concat_vectors (v2i32 V64:$Rd),
3814 (trunc (ARM64vlshr (v2i64 V128:$Rn),
3815 vecshiftR64Narrow:$imm)))),
3816 (SHRNv4i32_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
3817 V128:$Rn, vecshiftR32Narrow:$imm)>;
3819 // Vector sign and zero extensions are implemented with SSHLL and USSHLL.
3820 // Anyexts are implemented as zexts.
3821 def : Pat<(v8i16 (sext (v8i8 V64:$Rn))), (SSHLLv8i8_shift V64:$Rn, (i32 0))>;
3822 def : Pat<(v8i16 (zext (v8i8 V64:$Rn))), (USHLLv8i8_shift V64:$Rn, (i32 0))>;
3823 def : Pat<(v8i16 (anyext (v8i8 V64:$Rn))), (USHLLv8i8_shift V64:$Rn, (i32 0))>;
3824 def : Pat<(v4i32 (sext (v4i16 V64:$Rn))), (SSHLLv4i16_shift V64:$Rn, (i32 0))>;
3825 def : Pat<(v4i32 (zext (v4i16 V64:$Rn))), (USHLLv4i16_shift V64:$Rn, (i32 0))>;
3826 def : Pat<(v4i32 (anyext (v4i16 V64:$Rn))), (USHLLv4i16_shift V64:$Rn, (i32 0))>;
3827 def : Pat<(v2i64 (sext (v2i32 V64:$Rn))), (SSHLLv2i32_shift V64:$Rn, (i32 0))>;
3828 def : Pat<(v2i64 (zext (v2i32 V64:$Rn))), (USHLLv2i32_shift V64:$Rn, (i32 0))>;
3829 def : Pat<(v2i64 (anyext (v2i32 V64:$Rn))), (USHLLv2i32_shift V64:$Rn, (i32 0))>;
3830 // Also match an extend from the upper half of a 128 bit source register.
3831 def : Pat<(v8i16 (anyext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
3832 (USHLLv16i8_shift V128:$Rn, (i32 0))>;
3833 def : Pat<(v8i16 (zext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
3834 (USHLLv16i8_shift V128:$Rn, (i32 0))>;
3835 def : Pat<(v8i16 (sext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
3836 (SSHLLv16i8_shift V128:$Rn, (i32 0))>;
3837 def : Pat<(v4i32 (anyext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
3838 (USHLLv8i16_shift V128:$Rn, (i32 0))>;
3839 def : Pat<(v4i32 (zext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
3840 (USHLLv8i16_shift V128:$Rn, (i32 0))>;
3841 def : Pat<(v4i32 (sext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
3842 (SSHLLv8i16_shift V128:$Rn, (i32 0))>;
3843 def : Pat<(v2i64 (anyext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
3844 (USHLLv4i32_shift V128:$Rn, (i32 0))>;
3845 def : Pat<(v2i64 (zext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
3846 (USHLLv4i32_shift V128:$Rn, (i32 0))>;
3847 def : Pat<(v2i64 (sext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
3848 (SSHLLv4i32_shift V128:$Rn, (i32 0))>;
3850 // Vector shift sxtl aliases
3851 def : InstAlias<"sxtl.8h $dst, $src1",
3852 (SSHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
3853 def : InstAlias<"sxtl $dst.8h, $src1.8b",
3854 (SSHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
3855 def : InstAlias<"sxtl.4s $dst, $src1",
3856 (SSHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
3857 def : InstAlias<"sxtl $dst.4s, $src1.4h",
3858 (SSHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
3859 def : InstAlias<"sxtl.2d $dst, $src1",
3860 (SSHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
3861 def : InstAlias<"sxtl $dst.2d, $src1.2s",
3862 (SSHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
3864 // Vector shift sxtl2 aliases
3865 def : InstAlias<"sxtl2.8h $dst, $src1",
3866 (SSHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
3867 def : InstAlias<"sxtl2 $dst.8h, $src1.16b",
3868 (SSHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
3869 def : InstAlias<"sxtl2.4s $dst, $src1",
3870 (SSHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
3871 def : InstAlias<"sxtl2 $dst.4s, $src1.8h",
3872 (SSHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
3873 def : InstAlias<"sxtl2.2d $dst, $src1",
3874 (SSHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
3875 def : InstAlias<"sxtl2 $dst.2d, $src1.4s",
3876 (SSHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
3878 // Vector shift uxtl aliases
3879 def : InstAlias<"uxtl.8h $dst, $src1",
3880 (USHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
3881 def : InstAlias<"uxtl $dst.8h, $src1.8b",
3882 (USHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
3883 def : InstAlias<"uxtl.4s $dst, $src1",
3884 (USHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
3885 def : InstAlias<"uxtl $dst.4s, $src1.4h",
3886 (USHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
3887 def : InstAlias<"uxtl.2d $dst, $src1",
3888 (USHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
3889 def : InstAlias<"uxtl $dst.2d, $src1.2s",
3890 (USHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
3892 // Vector shift uxtl2 aliases
3893 def : InstAlias<"uxtl2.8h $dst, $src1",
3894 (USHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
3895 def : InstAlias<"uxtl2 $dst.8h, $src1.16b",
3896 (USHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
3897 def : InstAlias<"uxtl2.4s $dst, $src1",
3898 (USHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
3899 def : InstAlias<"uxtl2 $dst.4s, $src1.8h",
3900 (USHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
3901 def : InstAlias<"uxtl2.2d $dst, $src1",
3902 (USHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
3903 def : InstAlias<"uxtl2 $dst.2d, $src1.4s",
3904 (USHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
3906 // If an integer is about to be converted to a floating point value,
3907 // just load it on the floating point unit.
3908 // These patterns are more complex because floating point loads do not
3909 // support sign extension.
3910 // The sign extension has to be explicitly added and is only supported for
3911 // one step: byte-to-half, half-to-word, word-to-doubleword.
3912 // SCVTF GPR -> FPR is 9 cycles.
3913 // SCVTF FPR -> FPR is 4 cyclces.
3914 // (sign extension with lengthen) SXTL FPR -> FPR is 2 cycles.
3915 // Therefore, we can do 2 sign extensions and one SCVTF FPR -> FPR
3916 // and still being faster.
3917 // However, this is not good for code size.
3918 // 8-bits -> float. 2 sizes step-up.
3919 def : Pat <(f32 (sint_to_fp (i32 (sextloadi8 ro_indexed8:$addr)))),
3920 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
3925 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3926 (LDRBro ro_indexed8:$addr),
3931 ssub)))>, Requires<[NotForCodeSize]>;
3932 def : Pat <(f32 (sint_to_fp (i32 (sextloadi8 am_indexed8:$addr)))),
3933 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
3938 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3939 (LDRBui am_indexed8:$addr),
3944 ssub)))>, Requires<[NotForCodeSize]>;
3945 def : Pat <(f32 (sint_to_fp (i32 (sextloadi8 am_unscaled8:$addr)))),
3946 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
3951 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3952 (LDURBi am_unscaled8:$addr),
3957 ssub)))>, Requires<[NotForCodeSize]>;
3958 // 16-bits -> float. 1 size step-up.
3959 def : Pat <(f32 (sint_to_fp (i32 (sextloadi16 ro_indexed16:$addr)))),
3960 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
3962 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3963 (LDRHro ro_indexed16:$addr),
3966 ssub)))>, Requires<[NotForCodeSize]>;
3967 def : Pat <(f32 (sint_to_fp (i32 (sextloadi16 am_indexed16:$addr)))),
3968 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
3970 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3971 (LDRHui am_indexed16:$addr),
3974 ssub)))>, Requires<[NotForCodeSize]>;
3975 def : Pat <(f32 (sint_to_fp (i32 (sextloadi16 am_unscaled16:$addr)))),
3976 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
3978 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3979 (LDURHi am_unscaled16:$addr),
3982 ssub)))>, Requires<[NotForCodeSize]>;
3983 // 32-bits to 32-bits are handled in target specific dag combine:
3984 // performIntToFpCombine.
3985 // 64-bits integer to 32-bits floating point, not possible with
3986 // SCVTF on floating point registers (both source and destination
3987 // must have the same size).
3989 // Here are the patterns for 8, 16, 32, and 64-bits to double.
3990 // 8-bits -> double. 3 size step-up: give up.
3991 // 16-bits -> double. 2 size step.
3992 def : Pat <(f64 (sint_to_fp (i32 (sextloadi16 ro_indexed16:$addr)))),
3993 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
3998 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3999 (LDRHro ro_indexed16:$addr),
4004 dsub)))>, Requires<[NotForCodeSize]>;
4005 def : Pat <(f64 (sint_to_fp (i32 (sextloadi16 am_indexed16:$addr)))),
4006 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4011 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4012 (LDRHui am_indexed16:$addr),
4017 dsub)))>, Requires<[NotForCodeSize]>;
4018 def : Pat <(f64 (sint_to_fp (i32 (sextloadi16 am_unscaled16:$addr)))),
4019 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4024 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4025 (LDURHi am_unscaled16:$addr),
4030 dsub)))>, Requires<[NotForCodeSize]>;
4031 // 32-bits -> double. 1 size step-up.
4032 def : Pat <(f64 (sint_to_fp (i32 (load ro_indexed32:$addr)))),
4033 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4035 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4036 (LDRSro ro_indexed32:$addr),
4039 dsub)))>, Requires<[NotForCodeSize]>;
4040 def : Pat <(f64 (sint_to_fp (i32 (load am_indexed32:$addr)))),
4041 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4043 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4044 (LDRSui am_indexed32:$addr),
4047 dsub)))>, Requires<[NotForCodeSize]>;
4048 def : Pat <(f64 (sint_to_fp (i32 (load am_unscaled32:$addr)))),
4049 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4051 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4052 (LDURSi am_unscaled32:$addr),
4055 dsub)))>, Requires<[NotForCodeSize]>;
4056 // 64-bits -> double are handled in target specific dag combine:
4057 // performIntToFpCombine.
4060 //----------------------------------------------------------------------------
4061 // AdvSIMD Load-Store Structure
4062 //----------------------------------------------------------------------------
4063 defm LD1 : SIMDLd1Multiple<"ld1">;
4064 defm LD2 : SIMDLd2Multiple<"ld2">;
4065 defm LD3 : SIMDLd3Multiple<"ld3">;
4066 defm LD4 : SIMDLd4Multiple<"ld4">;
4068 defm ST1 : SIMDSt1Multiple<"st1">;
4069 defm ST2 : SIMDSt2Multiple<"st2">;
4070 defm ST3 : SIMDSt3Multiple<"st3">;
4071 defm ST4 : SIMDSt4Multiple<"st4">;
4073 class Ld1Pat<ValueType ty, Instruction INST>
4074 : Pat<(ty (load am_simdnoindex:$vaddr)), (INST am_simdnoindex:$vaddr)>;
4076 def : Ld1Pat<v16i8, LD1Onev16b>;
4077 def : Ld1Pat<v8i16, LD1Onev8h>;
4078 def : Ld1Pat<v4i32, LD1Onev4s>;
4079 def : Ld1Pat<v2i64, LD1Onev2d>;
4080 def : Ld1Pat<v8i8, LD1Onev8b>;
4081 def : Ld1Pat<v4i16, LD1Onev4h>;
4082 def : Ld1Pat<v2i32, LD1Onev2s>;
4083 def : Ld1Pat<v1i64, LD1Onev1d>;
4085 class St1Pat<ValueType ty, Instruction INST>
4086 : Pat<(store ty:$Vt, am_simdnoindex:$vaddr),
4087 (INST ty:$Vt, am_simdnoindex:$vaddr)>;
4089 def : St1Pat<v16i8, ST1Onev16b>;
4090 def : St1Pat<v8i16, ST1Onev8h>;
4091 def : St1Pat<v4i32, ST1Onev4s>;
4092 def : St1Pat<v2i64, ST1Onev2d>;
4093 def : St1Pat<v8i8, ST1Onev8b>;
4094 def : St1Pat<v4i16, ST1Onev4h>;
4095 def : St1Pat<v2i32, ST1Onev2s>;
4096 def : St1Pat<v1i64, ST1Onev1d>;
4102 defm LD1R : SIMDLdR<0, 0b110, 0, "ld1r", "One", 1, 2, 4, 8>;
4103 defm LD2R : SIMDLdR<1, 0b110, 0, "ld2r", "Two", 2, 4, 8, 16>;
4104 defm LD3R : SIMDLdR<0, 0b111, 0, "ld3r", "Three", 3, 6, 12, 24>;
4105 defm LD4R : SIMDLdR<1, 0b111, 0, "ld4r", "Four", 4, 8, 16, 32>;
4106 let mayLoad = 1, neverHasSideEffects = 1 in {
4107 defm LD1 : SIMDLdSingleBTied<0, 0b000, "ld1", VecListOneb, GPR64pi1>;
4108 defm LD1 : SIMDLdSingleHTied<0, 0b010, 0, "ld1", VecListOneh, GPR64pi2>;
4109 defm LD1 : SIMDLdSingleSTied<0, 0b100, 0b00, "ld1", VecListOnes, GPR64pi4>;
4110 defm LD1 : SIMDLdSingleDTied<0, 0b100, 0b01, "ld1", VecListOned, GPR64pi8>;
4111 defm LD2 : SIMDLdSingleBTied<1, 0b000, "ld2", VecListTwob, GPR64pi2>;
4112 defm LD2 : SIMDLdSingleHTied<1, 0b010, 0, "ld2", VecListTwoh, GPR64pi4>;
4113 defm LD2 : SIMDLdSingleSTied<1, 0b100, 0b00, "ld2", VecListTwos, GPR64pi8>;
4114 defm LD2 : SIMDLdSingleDTied<1, 0b100, 0b01, "ld2", VecListTwod, GPR64pi16>;
4115 defm LD3 : SIMDLdSingleBTied<0, 0b001, "ld3", VecListThreeb, GPR64pi3>;
4116 defm LD3 : SIMDLdSingleHTied<0, 0b011, 0, "ld3", VecListThreeh, GPR64pi6>;
4117 defm LD3 : SIMDLdSingleSTied<0, 0b101, 0b00, "ld3", VecListThrees, GPR64pi12>;
4118 defm LD3 : SIMDLdSingleDTied<0, 0b101, 0b01, "ld3", VecListThreed, GPR64pi24>;
4119 defm LD4 : SIMDLdSingleBTied<1, 0b001, "ld4", VecListFourb, GPR64pi4>;
4120 defm LD4 : SIMDLdSingleHTied<1, 0b011, 0, "ld4", VecListFourh, GPR64pi8>;
4121 defm LD4 : SIMDLdSingleSTied<1, 0b101, 0b00, "ld4", VecListFours, GPR64pi16>;
4122 defm LD4 : SIMDLdSingleDTied<1, 0b101, 0b01, "ld4", VecListFourd, GPR64pi32>;
4125 def : Pat<(v8i8 (ARM64dup (i32 (extloadi8 am_simdnoindex:$vaddr)))),
4126 (LD1Rv8b am_simdnoindex:$vaddr)>;
4127 def : Pat<(v16i8 (ARM64dup (i32 (extloadi8 am_simdnoindex:$vaddr)))),
4128 (LD1Rv16b am_simdnoindex:$vaddr)>;
4129 def : Pat<(v4i16 (ARM64dup (i32 (extloadi16 am_simdnoindex:$vaddr)))),
4130 (LD1Rv4h am_simdnoindex:$vaddr)>;
4131 def : Pat<(v8i16 (ARM64dup (i32 (extloadi16 am_simdnoindex:$vaddr)))),
4132 (LD1Rv8h am_simdnoindex:$vaddr)>;
4133 def : Pat<(v2i32 (ARM64dup (i32 (load am_simdnoindex:$vaddr)))),
4134 (LD1Rv2s am_simdnoindex:$vaddr)>;
4135 def : Pat<(v4i32 (ARM64dup (i32 (load am_simdnoindex:$vaddr)))),
4136 (LD1Rv4s am_simdnoindex:$vaddr)>;
4137 def : Pat<(v2i64 (ARM64dup (i64 (load am_simdnoindex:$vaddr)))),
4138 (LD1Rv2d am_simdnoindex:$vaddr)>;
4139 def : Pat<(v1i64 (ARM64dup (i64 (load am_simdnoindex:$vaddr)))),
4140 (LD1Rv1d am_simdnoindex:$vaddr)>;
4141 // Grab the floating point version too
4142 def : Pat<(v2f32 (ARM64dup (f32 (load am_simdnoindex:$vaddr)))),
4143 (LD1Rv2s am_simdnoindex:$vaddr)>;
4144 def : Pat<(v4f32 (ARM64dup (f32 (load am_simdnoindex:$vaddr)))),
4145 (LD1Rv4s am_simdnoindex:$vaddr)>;
4146 def : Pat<(v2f64 (ARM64dup (f64 (load am_simdnoindex:$vaddr)))),
4147 (LD1Rv2d am_simdnoindex:$vaddr)>;
4148 def : Pat<(v1f64 (ARM64dup (f64 (load am_simdnoindex:$vaddr)))),
4149 (LD1Rv1d am_simdnoindex:$vaddr)>;
4151 class Ld1Lane128Pat<SDPatternOperator scalar_load, Operand VecIndex,
4152 ValueType VTy, ValueType STy, Instruction LD1>
4153 : Pat<(vector_insert (VTy VecListOne128:$Rd),
4154 (STy (scalar_load am_simdnoindex:$vaddr)), VecIndex:$idx),
4155 (LD1 VecListOne128:$Rd, VecIndex:$idx, am_simdnoindex:$vaddr)>;
4157 def : Ld1Lane128Pat<extloadi8, VectorIndexB, v16i8, i32, LD1i8>;
4158 def : Ld1Lane128Pat<extloadi16, VectorIndexH, v8i16, i32, LD1i16>;
4159 def : Ld1Lane128Pat<load, VectorIndexS, v4i32, i32, LD1i32>;
4160 def : Ld1Lane128Pat<load, VectorIndexS, v4f32, f32, LD1i32>;
4161 def : Ld1Lane128Pat<load, VectorIndexD, v2i64, i64, LD1i64>;
4162 def : Ld1Lane128Pat<load, VectorIndexD, v2f64, f64, LD1i64>;
4164 class Ld1Lane64Pat<SDPatternOperator scalar_load, Operand VecIndex,
4165 ValueType VTy, ValueType STy, Instruction LD1>
4166 : Pat<(vector_insert (VTy VecListOne64:$Rd),
4167 (STy (scalar_load am_simdnoindex:$vaddr)), VecIndex:$idx),
4169 (LD1 (SUBREG_TO_REG (i32 0), VecListOne64:$Rd, dsub),
4170 VecIndex:$idx, am_simdnoindex:$vaddr),
4173 def : Ld1Lane64Pat<extloadi8, VectorIndexB, v8i8, i32, LD1i8>;
4174 def : Ld1Lane64Pat<extloadi16, VectorIndexH, v4i16, i32, LD1i16>;
4175 def : Ld1Lane64Pat<load, VectorIndexS, v2i32, i32, LD1i32>;
4176 def : Ld1Lane64Pat<load, VectorIndexS, v2f32, f32, LD1i32>;
4179 defm LD1 : SIMDLdSt1SingleAliases<"ld1">;
4180 defm LD2 : SIMDLdSt2SingleAliases<"ld2">;
4181 defm LD3 : SIMDLdSt3SingleAliases<"ld3">;
4182 defm LD4 : SIMDLdSt4SingleAliases<"ld4">;
4185 defm ST1 : SIMDStSingleB<0, 0b000, "st1", VecListOneb, GPR64pi1>;
4186 defm ST1 : SIMDStSingleH<0, 0b010, 0, "st1", VecListOneh, GPR64pi2>;
4187 defm ST1 : SIMDStSingleS<0, 0b100, 0b00, "st1", VecListOnes, GPR64pi4>;
4188 defm ST1 : SIMDStSingleD<0, 0b100, 0b01, "st1", VecListOned, GPR64pi8>;
4190 let AddedComplexity = 8 in
4191 class St1Lane128Pat<SDPatternOperator scalar_store, Operand VecIndex,
4192 ValueType VTy, ValueType STy, Instruction ST1>
4194 (STy (vector_extract (VTy VecListOne128:$Vt), VecIndex:$idx)),
4195 am_simdnoindex:$vaddr),
4196 (ST1 VecListOne128:$Vt, VecIndex:$idx, am_simdnoindex:$vaddr)>;
4198 def : St1Lane128Pat<truncstorei8, VectorIndexB, v16i8, i32, ST1i8>;
4199 def : St1Lane128Pat<truncstorei16, VectorIndexH, v8i16, i32, ST1i16>;
4200 def : St1Lane128Pat<store, VectorIndexS, v4i32, i32, ST1i32>;
4201 def : St1Lane128Pat<store, VectorIndexS, v4f32, f32, ST1i32>;
4202 def : St1Lane128Pat<store, VectorIndexD, v2i64, i64, ST1i64>;
4203 def : St1Lane128Pat<store, VectorIndexD, v2f64, f64, ST1i64>;
4205 let AddedComplexity = 8 in
4206 class St1Lane64Pat<SDPatternOperator scalar_store, Operand VecIndex,
4207 ValueType VTy, ValueType STy, Instruction ST1>
4209 (STy (vector_extract (VTy VecListOne64:$Vt), VecIndex:$idx)),
4210 am_simdnoindex:$vaddr),
4211 (ST1 (SUBREG_TO_REG (i32 0), VecListOne64:$Vt, dsub),
4212 VecIndex:$idx, am_simdnoindex:$vaddr)>;
4214 def : St1Lane64Pat<truncstorei8, VectorIndexB, v8i8, i32, ST1i8>;
4215 def : St1Lane64Pat<truncstorei16, VectorIndexH, v4i16, i32, ST1i16>;
4216 def : St1Lane64Pat<store, VectorIndexS, v2i32, i32, ST1i32>;
4217 def : St1Lane64Pat<store, VectorIndexS, v2f32, f32, ST1i32>;
4219 let mayStore = 1, neverHasSideEffects = 1 in {
4220 defm ST2 : SIMDStSingleB<1, 0b000, "st2", VecListTwob, GPR64pi2>;
4221 defm ST2 : SIMDStSingleH<1, 0b010, 0, "st2", VecListTwoh, GPR64pi4>;
4222 defm ST2 : SIMDStSingleS<1, 0b100, 0b00, "st2", VecListTwos, GPR64pi8>;
4223 defm ST2 : SIMDStSingleD<1, 0b100, 0b01, "st2", VecListTwod, GPR64pi16>;
4224 defm ST3 : SIMDStSingleB<0, 0b001, "st3", VecListThreeb, GPR64pi3>;
4225 defm ST3 : SIMDStSingleH<0, 0b011, 0, "st3", VecListThreeh, GPR64pi6>;
4226 defm ST3 : SIMDStSingleS<0, 0b101, 0b00, "st3", VecListThrees, GPR64pi12>;
4227 defm ST3 : SIMDStSingleD<0, 0b101, 0b01, "st3", VecListThreed, GPR64pi24>;
4228 defm ST4 : SIMDStSingleB<1, 0b001, "st4", VecListFourb, GPR64pi4>;
4229 defm ST4 : SIMDStSingleH<1, 0b011, 0, "st4", VecListFourh, GPR64pi8>;
4230 defm ST4 : SIMDStSingleS<1, 0b101, 0b00, "st4", VecListFours, GPR64pi16>;
4231 defm ST4 : SIMDStSingleD<1, 0b101, 0b01, "st4", VecListFourd, GPR64pi32>;
4234 defm ST1 : SIMDLdSt1SingleAliases<"st1">;
4235 defm ST2 : SIMDLdSt2SingleAliases<"st2">;
4236 defm ST3 : SIMDLdSt3SingleAliases<"st3">;
4237 defm ST4 : SIMDLdSt4SingleAliases<"st4">;
4239 //----------------------------------------------------------------------------
4240 // Crypto extensions
4241 //----------------------------------------------------------------------------
4243 def AESErr : AESTiedInst<0b0100, "aese", int_arm64_crypto_aese>;
4244 def AESDrr : AESTiedInst<0b0101, "aesd", int_arm64_crypto_aesd>;
4245 def AESMCrr : AESInst< 0b0110, "aesmc", int_arm64_crypto_aesmc>;
4246 def AESIMCrr : AESInst< 0b0111, "aesimc", int_arm64_crypto_aesimc>;
4248 def SHA1Crrr : SHATiedInstQSV<0b000, "sha1c", int_arm64_crypto_sha1c>;
4249 def SHA1Prrr : SHATiedInstQSV<0b001, "sha1p", int_arm64_crypto_sha1p>;
4250 def SHA1Mrrr : SHATiedInstQSV<0b010, "sha1m", int_arm64_crypto_sha1m>;
4251 def SHA1SU0rrr : SHATiedInstVVV<0b011, "sha1su0", int_arm64_crypto_sha1su0>;
4252 def SHA256Hrrr : SHATiedInstQQV<0b100, "sha256h", int_arm64_crypto_sha256h>;
4253 def SHA256H2rrr : SHATiedInstQQV<0b101, "sha256h2",int_arm64_crypto_sha256h2>;
4254 def SHA256SU1rrr :SHATiedInstVVV<0b110, "sha256su1",int_arm64_crypto_sha256su1>;
4256 def SHA1Hrr : SHAInstSS< 0b0000, "sha1h", int_arm64_crypto_sha1h>;
4257 def SHA1SU1rr : SHATiedInstVV<0b0001, "sha1su1", int_arm64_crypto_sha1su1>;
4258 def SHA256SU0rr : SHATiedInstVV<0b0010, "sha256su0",int_arm64_crypto_sha256su0>;
4260 //----------------------------------------------------------------------------
4262 //----------------------------------------------------------------------------
4263 // FIXME: Like for X86, these should go in their own separate .td file.
4265 // Any instruction that defines a 32-bit result leaves the high half of the
4266 // register. Truncate can be lowered to EXTRACT_SUBREG. CopyFromReg may
4267 // be copying from a truncate. But any other 32-bit operation will zero-extend
4269 // FIXME: X86 also checks for CMOV here. Do we need something similar?
4270 def def32 : PatLeaf<(i32 GPR32:$src), [{
4271 return N->getOpcode() != ISD::TRUNCATE &&
4272 N->getOpcode() != TargetOpcode::EXTRACT_SUBREG &&
4273 N->getOpcode() != ISD::CopyFromReg;
4276 // In the case of a 32-bit def that is known to implicitly zero-extend,
4277 // we can use a SUBREG_TO_REG.
4278 def : Pat<(i64 (zext def32:$src)), (SUBREG_TO_REG (i64 0), GPR32:$src, sub_32)>;
4280 // For an anyext, we don't care what the high bits are, so we can perform an
4281 // INSERT_SUBREF into an IMPLICIT_DEF.
4282 def : Pat<(i64 (anyext GPR32:$src)),
4283 (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32)>;
4285 // When we need to explicitly zero-extend, we use an unsigned bitfield move
4286 // instruction (UBFM) on the enclosing super-reg.
4287 def : Pat<(i64 (zext GPR32:$src)),
4288 (UBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32), 0, 31)>;
4290 // To sign extend, we use a signed bitfield move instruction (SBFM) on the
4291 // containing super-reg.
4292 def : Pat<(i64 (sext GPR32:$src)),
4293 (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32), 0, 31)>;
4294 def : Pat<(i64 (sext_inreg GPR64:$src, i32)), (SBFMXri GPR64:$src, 0, 31)>;
4295 def : Pat<(i64 (sext_inreg GPR64:$src, i16)), (SBFMXri GPR64:$src, 0, 15)>;
4296 def : Pat<(i64 (sext_inreg GPR64:$src, i8)), (SBFMXri GPR64:$src, 0, 7)>;
4297 def : Pat<(i64 (sext_inreg GPR64:$src, i1)), (SBFMXri GPR64:$src, 0, 0)>;
4298 def : Pat<(i32 (sext_inreg GPR32:$src, i16)), (SBFMWri GPR32:$src, 0, 15)>;
4299 def : Pat<(i32 (sext_inreg GPR32:$src, i8)), (SBFMWri GPR32:$src, 0, 7)>;
4300 def : Pat<(i32 (sext_inreg GPR32:$src, i1)), (SBFMWri GPR32:$src, 0, 0)>;
4302 def : Pat<(shl (sext_inreg GPR32:$Rn, i8), (i64 imm0_31:$imm)),
4303 (SBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
4304 (i64 (i32shift_sext_i8 imm0_31:$imm)))>;
4305 def : Pat<(shl (sext_inreg GPR64:$Rn, i8), (i64 imm0_63:$imm)),
4306 (SBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
4307 (i64 (i64shift_sext_i8 imm0_63:$imm)))>;
4309 def : Pat<(shl (sext_inreg GPR32:$Rn, i16), (i64 imm0_31:$imm)),
4310 (SBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
4311 (i64 (i32shift_sext_i16 imm0_31:$imm)))>;
4312 def : Pat<(shl (sext_inreg GPR64:$Rn, i16), (i64 imm0_63:$imm)),
4313 (SBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
4314 (i64 (i64shift_sext_i16 imm0_63:$imm)))>;
4316 def : Pat<(shl (i64 (sext GPR32:$Rn)), (i64 imm0_63:$imm)),
4317 (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$Rn, sub_32),
4318 (i64 (i64shift_a imm0_63:$imm)),
4319 (i64 (i64shift_sext_i32 imm0_63:$imm)))>;
4321 // sra patterns have an AddedComplexity of 10, so make sure we have a higher
4322 // AddedComplexity for the following patterns since we want to match sext + sra
4323 // patterns before we attempt to match a single sra node.
4324 let AddedComplexity = 20 in {
4325 // We support all sext + sra combinations which preserve at least one bit of the
4326 // original value which is to be sign extended. E.g. we support shifts up to
4328 def : Pat<(sra (sext_inreg GPR32:$Rn, i8), (i64 imm0_7:$imm)),
4329 (SBFMWri GPR32:$Rn, (i64 imm0_7:$imm), 7)>;
4330 def : Pat<(sra (sext_inreg GPR64:$Rn, i8), (i64 imm0_7:$imm)),
4331 (SBFMXri GPR64:$Rn, (i64 imm0_7:$imm), 7)>;
4333 def : Pat<(sra (sext_inreg GPR32:$Rn, i16), (i64 imm0_15:$imm)),
4334 (SBFMWri GPR32:$Rn, (i64 imm0_15:$imm), 15)>;
4335 def : Pat<(sra (sext_inreg GPR64:$Rn, i16), (i64 imm0_15:$imm)),
4336 (SBFMXri GPR64:$Rn, (i64 imm0_15:$imm), 15)>;
4338 def : Pat<(sra (i64 (sext GPR32:$Rn)), (i64 imm0_31:$imm)),
4339 (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$Rn, sub_32),
4340 (i64 imm0_31:$imm), 31)>;
4341 } // AddedComplexity = 20
4343 // To truncate, we can simply extract from a subregister.
4344 def : Pat<(i32 (trunc GPR64sp:$src)),
4345 (i32 (EXTRACT_SUBREG GPR64sp:$src, sub_32))>;
4347 // __builtin_trap() uses the BRK instruction on ARM64.
4348 def : Pat<(trap), (BRK 1)>;
4350 // Conversions within AdvSIMD types in the same register size are free.
4352 def : Pat<(v1i64 (bitconvert (v2i32 FPR64:$src))), (v1i64 FPR64:$src)>;
4353 def : Pat<(v1i64 (bitconvert (v4i16 FPR64:$src))), (v1i64 FPR64:$src)>;
4354 def : Pat<(v1i64 (bitconvert (v8i8 FPR64:$src))), (v1i64 FPR64:$src)>;
4355 def : Pat<(v1i64 (bitconvert (f64 FPR64:$src))), (v1i64 FPR64:$src)>;
4356 def : Pat<(v1i64 (bitconvert (v2f32 FPR64:$src))), (v1i64 FPR64:$src)>;
4357 def : Pat<(v1i64 (bitconvert (v1f64 FPR64:$src))), (v1i64 FPR64:$src)>;
4359 def : Pat<(v2i32 (bitconvert (v1i64 FPR64:$src))), (v2i32 FPR64:$src)>;
4360 def : Pat<(v2i32 (bitconvert (v4i16 FPR64:$src))), (v2i32 FPR64:$src)>;
4361 def : Pat<(v2i32 (bitconvert (v8i8 FPR64:$src))), (v2i32 FPR64:$src)>;
4362 def : Pat<(v2i32 (bitconvert (f64 FPR64:$src))), (v2i32 FPR64:$src)>;
4363 def : Pat<(v2i32 (bitconvert (v2f32 FPR64:$src))), (v2i32 FPR64:$src)>;
4364 def : Pat<(v2i32 (bitconvert (v1f64 FPR64:$src))), (v2i32 FPR64:$src)>;
4366 def : Pat<(v4i16 (bitconvert (v1i64 FPR64:$src))), (v4i16 FPR64:$src)>;
4367 def : Pat<(v4i16 (bitconvert (v2i32 FPR64:$src))), (v4i16 FPR64:$src)>;
4368 def : Pat<(v4i16 (bitconvert (v8i8 FPR64:$src))), (v4i16 FPR64:$src)>;
4369 def : Pat<(v4i16 (bitconvert (f64 FPR64:$src))), (v4i16 FPR64:$src)>;
4370 def : Pat<(v4i16 (bitconvert (v2f32 FPR64:$src))), (v4i16 FPR64:$src)>;
4371 def : Pat<(v4i16 (bitconvert (v1f64 FPR64:$src))), (v4i16 FPR64:$src)>;
4373 def : Pat<(v8i8 (bitconvert (v1i64 FPR64:$src))), (v8i8 FPR64:$src)>;
4374 def : Pat<(v8i8 (bitconvert (v2i32 FPR64:$src))), (v8i8 FPR64:$src)>;
4375 def : Pat<(v8i8 (bitconvert (v4i16 FPR64:$src))), (v8i8 FPR64:$src)>;
4376 def : Pat<(v8i8 (bitconvert (f64 FPR64:$src))), (v8i8 FPR64:$src)>;
4377 def : Pat<(v8i8 (bitconvert (v2f32 FPR64:$src))), (v8i8 FPR64:$src)>;
4378 def : Pat<(v8i8 (bitconvert (v1f64 FPR64:$src))), (v8i8 FPR64:$src)>;
4380 def : Pat<(f64 (bitconvert (v1i64 FPR64:$src))), (f64 FPR64:$src)>;
4381 def : Pat<(f64 (bitconvert (v2i32 FPR64:$src))), (f64 FPR64:$src)>;
4382 def : Pat<(f64 (bitconvert (v4i16 FPR64:$src))), (f64 FPR64:$src)>;
4383 def : Pat<(f64 (bitconvert (v8i8 FPR64:$src))), (f64 FPR64:$src)>;
4384 def : Pat<(f64 (bitconvert (v2f32 FPR64:$src))), (f64 FPR64:$src)>;
4385 def : Pat<(f64 (bitconvert (v1f64 FPR64:$src))), (f64 FPR64:$src)>;
4387 def : Pat<(v1f64 (bitconvert (v1i64 FPR64:$src))), (v1f64 FPR64:$src)>;
4388 def : Pat<(v1f64 (bitconvert (v2i32 FPR64:$src))), (v1f64 FPR64:$src)>;
4389 def : Pat<(v1f64 (bitconvert (v4i16 FPR64:$src))), (v1f64 FPR64:$src)>;
4390 def : Pat<(v1f64 (bitconvert (v8i8 FPR64:$src))), (v1f64 FPR64:$src)>;
4391 def : Pat<(v1f64 (bitconvert (f64 FPR64:$src))), (v1f64 FPR64:$src)>;
4392 def : Pat<(v1f64 (bitconvert (v2f32 FPR64:$src))), (v1f64 FPR64:$src)>;
4394 def : Pat<(v2f32 (bitconvert (f64 FPR64:$src))), (v2f32 FPR64:$src)>;
4395 def : Pat<(v2f32 (bitconvert (v1i64 FPR64:$src))), (v2f32 FPR64:$src)>;
4396 def : Pat<(v2f32 (bitconvert (v2i32 FPR64:$src))), (v2f32 FPR64:$src)>;
4397 def : Pat<(v2f32 (bitconvert (v4i16 FPR64:$src))), (v2f32 FPR64:$src)>;
4398 def : Pat<(v2f32 (bitconvert (v8i8 FPR64:$src))), (v2f32 FPR64:$src)>;
4399 def : Pat<(v2f32 (bitconvert (v1f64 FPR64:$src))), (v2f32 FPR64:$src)>;
4402 def : Pat<(f128 (bitconvert (v2i64 FPR128:$src))), (f128 FPR128:$src)>;
4403 def : Pat<(f128 (bitconvert (v4i32 FPR128:$src))), (f128 FPR128:$src)>;
4404 def : Pat<(f128 (bitconvert (v8i16 FPR128:$src))), (f128 FPR128:$src)>;
4405 def : Pat<(f128 (bitconvert (v2f64 FPR128:$src))), (f128 FPR128:$src)>;
4406 def : Pat<(f128 (bitconvert (v4f32 FPR128:$src))), (f128 FPR128:$src)>;
4408 def : Pat<(v2f64 (bitconvert (f128 FPR128:$src))), (v2f64 FPR128:$src)>;
4409 def : Pat<(v2f64 (bitconvert (v4i32 FPR128:$src))), (v2f64 FPR128:$src)>;
4410 def : Pat<(v2f64 (bitconvert (v8i16 FPR128:$src))), (v2f64 FPR128:$src)>;
4411 def : Pat<(v2f64 (bitconvert (v16i8 FPR128:$src))), (v2f64 FPR128:$src)>;
4412 def : Pat<(v2f64 (bitconvert (v2i64 FPR128:$src))), (v2f64 FPR128:$src)>;
4413 def : Pat<(v2f64 (bitconvert (v4f32 FPR128:$src))), (v2f64 FPR128:$src)>;
4415 def : Pat<(v4f32 (bitconvert (f128 FPR128:$src))), (v4f32 FPR128:$src)>;
4416 def : Pat<(v4f32 (bitconvert (v4i32 FPR128:$src))), (v4f32 FPR128:$src)>;
4417 def : Pat<(v4f32 (bitconvert (v8i16 FPR128:$src))), (v4f32 FPR128:$src)>;
4418 def : Pat<(v4f32 (bitconvert (v16i8 FPR128:$src))), (v4f32 FPR128:$src)>;
4419 def : Pat<(v4f32 (bitconvert (v2i64 FPR128:$src))), (v4f32 FPR128:$src)>;
4420 def : Pat<(v4f32 (bitconvert (v2f64 FPR128:$src))), (v4f32 FPR128:$src)>;
4422 def : Pat<(v2i64 (bitconvert (f128 FPR128:$src))), (v2i64 FPR128:$src)>;
4423 def : Pat<(v2i64 (bitconvert (v4i32 FPR128:$src))), (v2i64 FPR128:$src)>;
4424 def : Pat<(v2i64 (bitconvert (v8i16 FPR128:$src))), (v2i64 FPR128:$src)>;
4425 def : Pat<(v2i64 (bitconvert (v16i8 FPR128:$src))), (v2i64 FPR128:$src)>;
4426 def : Pat<(v2i64 (bitconvert (v2f64 FPR128:$src))), (v2i64 FPR128:$src)>;
4427 def : Pat<(v2i64 (bitconvert (v4f32 FPR128:$src))), (v2i64 FPR128:$src)>;
4429 def : Pat<(v4i32 (bitconvert (f128 FPR128:$src))), (v4i32 FPR128:$src)>;
4430 def : Pat<(v4i32 (bitconvert (v2i64 FPR128:$src))), (v4i32 FPR128:$src)>;
4431 def : Pat<(v4i32 (bitconvert (v8i16 FPR128:$src))), (v4i32 FPR128:$src)>;
4432 def : Pat<(v4i32 (bitconvert (v16i8 FPR128:$src))), (v4i32 FPR128:$src)>;
4433 def : Pat<(v4i32 (bitconvert (v2f64 FPR128:$src))), (v4i32 FPR128:$src)>;
4434 def : Pat<(v4i32 (bitconvert (v4f32 FPR128:$src))), (v4i32 FPR128:$src)>;
4436 def : Pat<(v8i16 (bitconvert (f128 FPR128:$src))), (v8i16 FPR128:$src)>;
4437 def : Pat<(v8i16 (bitconvert (v2i64 FPR128:$src))), (v8i16 FPR128:$src)>;
4438 def : Pat<(v8i16 (bitconvert (v4i32 FPR128:$src))), (v8i16 FPR128:$src)>;
4439 def : Pat<(v8i16 (bitconvert (v16i8 FPR128:$src))), (v8i16 FPR128:$src)>;
4440 def : Pat<(v8i16 (bitconvert (v2f64 FPR128:$src))), (v8i16 FPR128:$src)>;
4441 def : Pat<(v8i16 (bitconvert (v4f32 FPR128:$src))), (v8i16 FPR128:$src)>;
4443 def : Pat<(v16i8 (bitconvert (f128 FPR128:$src))), (v16i8 FPR128:$src)>;
4444 def : Pat<(v16i8 (bitconvert (v2i64 FPR128:$src))), (v16i8 FPR128:$src)>;
4445 def : Pat<(v16i8 (bitconvert (v4i32 FPR128:$src))), (v16i8 FPR128:$src)>;
4446 def : Pat<(v16i8 (bitconvert (v8i16 FPR128:$src))), (v16i8 FPR128:$src)>;
4447 def : Pat<(v16i8 (bitconvert (v2f64 FPR128:$src))), (v16i8 FPR128:$src)>;
4448 def : Pat<(v16i8 (bitconvert (v4f32 FPR128:$src))), (v16i8 FPR128:$src)>;
4450 def : Pat<(v8i8 (extract_subvector (v16i8 FPR128:$Rn), (i64 1))),
4451 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
4452 def : Pat<(v4i16 (extract_subvector (v8i16 FPR128:$Rn), (i64 1))),
4453 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
4454 def : Pat<(v2i32 (extract_subvector (v4i32 FPR128:$Rn), (i64 1))),
4455 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
4456 def : Pat<(v1i64 (extract_subvector (v2i64 FPR128:$Rn), (i64 1))),
4457 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
4459 // A 64-bit subvector insert to the first 128-bit vector position
4460 // is a subregister copy that needs no instruction.
4461 def : Pat<(insert_subvector undef, (v1i64 FPR64:$src), (i32 0)),
4462 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
4463 def : Pat<(insert_subvector undef, (v1f64 FPR64:$src), (i32 0)),
4464 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
4465 def : Pat<(insert_subvector undef, (v2i32 FPR64:$src), (i32 0)),
4466 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
4467 def : Pat<(insert_subvector undef, (v2f32 FPR64:$src), (i32 0)),
4468 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
4469 def : Pat<(insert_subvector undef, (v4i16 FPR64:$src), (i32 0)),
4470 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
4471 def : Pat<(insert_subvector undef, (v8i8 FPR64:$src), (i32 0)),
4472 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
4474 // Use pair-wise add instructions when summing up the lanes for v2f64, v2i64
4476 def : Pat<(i64 (add (vector_extract (v2i64 FPR128:$Rn), (i64 0)),
4477 (vector_extract (v2i64 FPR128:$Rn), (i64 1)))),
4478 (i64 (ADDPv2i64p (v2i64 FPR128:$Rn)))>;
4479 def : Pat<(f64 (fadd (vector_extract (v2f64 FPR128:$Rn), (i64 0)),
4480 (vector_extract (v2f64 FPR128:$Rn), (i64 1)))),
4481 (f64 (FADDPv2i64p (v2f64 FPR128:$Rn)))>;
4482 // vector_extract on 64-bit vectors gets promoted to a 128 bit vector,
4483 // so we match on v4f32 here, not v2f32. This will also catch adding
4484 // the low two lanes of a true v4f32 vector.
4485 def : Pat<(fadd (vector_extract (v4f32 FPR128:$Rn), (i64 0)),
4486 (vector_extract (v4f32 FPR128:$Rn), (i64 1))),
4487 (f32 (FADDPv2i32p (EXTRACT_SUBREG FPR128:$Rn, dsub)))>;
4489 // Scalar 64-bit shifts in FPR64 registers.
4490 def : Pat<(i64 (int_arm64_neon_sshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
4491 (SSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
4492 def : Pat<(i64 (int_arm64_neon_ushl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
4493 (USHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
4494 def : Pat<(i64 (int_arm64_neon_srshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
4495 (SRSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
4496 def : Pat<(i64 (int_arm64_neon_urshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
4497 (URSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
4499 // Tail call return handling. These are all compiler pseudo-instructions,
4500 // so no encoding information or anything like that.
4501 let isCall = 1, isTerminator = 1, isReturn = 1, isBarrier = 1, Uses = [SP] in {
4502 def TCRETURNdi : Pseudo<(outs), (ins i64imm:$dst), []>;
4503 def TCRETURNri : Pseudo<(outs), (ins tcGPR64:$dst), []>;
4506 def : Pat<(ARM64tcret tcGPR64:$dst), (TCRETURNri tcGPR64:$dst)>;
4507 def : Pat<(ARM64tcret (i64 tglobaladdr:$dst)), (TCRETURNdi texternalsym:$dst)>;
4508 def : Pat<(ARM64tcret (i64 texternalsym:$dst)), (TCRETURNdi texternalsym:$dst)>;
4510 include "ARM64InstrAtomics.td"