1 //===- ARM64InstrInfo.td - Describe the ARM64 Instructions -*- tablegen -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // ARM64 Instruction definitions.
12 //===----------------------------------------------------------------------===//
14 //===----------------------------------------------------------------------===//
15 // ARM64-specific DAG Nodes.
18 // SDTBinaryArithWithFlagsOut - RES1, FLAGS = op LHS, RHS
19 def SDTBinaryArithWithFlagsOut : SDTypeProfile<2, 2,
22 SDTCisInt<0>, SDTCisVT<1, i32>]>;
24 // SDTBinaryArithWithFlagsIn - RES1, FLAGS = op LHS, RHS, FLAGS
25 def SDTBinaryArithWithFlagsIn : SDTypeProfile<1, 3,
31 // SDTBinaryArithWithFlagsInOut - RES1, FLAGS = op LHS, RHS, FLAGS
32 def SDTBinaryArithWithFlagsInOut : SDTypeProfile<2, 3,
39 def SDT_ARM64Brcond : SDTypeProfile<0, 3,
40 [SDTCisVT<0, OtherVT>, SDTCisVT<1, i32>,
42 def SDT_ARM64cbz : SDTypeProfile<0, 2, [SDTCisInt<0>, SDTCisVT<1, OtherVT>]>;
43 def SDT_ARM64tbz : SDTypeProfile<0, 3, [SDTCisVT<0, i64>, SDTCisVT<1, i64>,
44 SDTCisVT<2, OtherVT>]>;
47 def SDT_ARM64CSel : SDTypeProfile<1, 4,
52 def SDT_ARM64FCmp : SDTypeProfile<0, 2,
55 def SDT_ARM64Dup : SDTypeProfile<1, 1, [SDTCisVec<0>]>;
56 def SDT_ARM64DupLane : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisInt<2>]>;
57 def SDT_ARM64Zip : SDTypeProfile<1, 2, [SDTCisVec<0>,
60 def SDT_ARM64MOVIedit : SDTypeProfile<1, 1, [SDTCisInt<1>]>;
61 def SDT_ARM64MOVIshift : SDTypeProfile<1, 2, [SDTCisInt<1>, SDTCisInt<2>]>;
62 def SDT_ARM64vecimm : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0,1>,
63 SDTCisInt<2>, SDTCisInt<3>]>;
64 def SDT_ARM64UnaryVec: SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0,1>]>;
65 def SDT_ARM64ExtVec: SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0,1>,
66 SDTCisSameAs<0,2>, SDTCisInt<3>]>;
67 def SDT_ARM64vshift : SDTypeProfile<1, 2, [SDTCisSameAs<0,1>, SDTCisInt<2>]>;
69 def SDT_ARM64unvec : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0,1>]>;
70 def SDT_ARM64fcmpz : SDTypeProfile<1, 1, []>;
71 def SDT_ARM64fcmp : SDTypeProfile<1, 2, [SDTCisSameAs<1,2>]>;
72 def SDT_ARM64binvec : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisSameAs<0,1>,
74 def SDT_ARM64trivec : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0,1>,
77 def SDT_ARM64TCRET : SDTypeProfile<0, 1, [SDTCisPtrTy<0>]>;
78 def SDT_ARM64PREFETCH : SDTypeProfile<0, 2, [SDTCisVT<0, i32>, SDTCisPtrTy<1>]>;
80 def SDT_ARM64ITOF : SDTypeProfile<1, 1, [SDTCisFP<0>, SDTCisSameAs<0,1>]>;
82 def SDT_ARM64TLSDescCall : SDTypeProfile<0, -2, [SDTCisPtrTy<0>,
84 def SDT_ARM64WrapperLarge : SDTypeProfile<1, 4,
85 [SDTCisVT<0, i64>, SDTCisVT<1, i32>,
86 SDTCisSameAs<1, 2>, SDTCisSameAs<1, 3>,
91 def ARM64adrp : SDNode<"ARM64ISD::ADRP", SDTIntUnaryOp, []>;
92 def ARM64addlow : SDNode<"ARM64ISD::ADDlow", SDTIntBinOp, []>;
93 def ARM64LOADgot : SDNode<"ARM64ISD::LOADgot", SDTIntUnaryOp>;
94 def ARM64callseq_start : SDNode<"ISD::CALLSEQ_START",
95 SDCallSeqStart<[ SDTCisVT<0, i32> ]>,
96 [SDNPHasChain, SDNPOutGlue]>;
97 def ARM64callseq_end : SDNode<"ISD::CALLSEQ_END",
98 SDCallSeqEnd<[ SDTCisVT<0, i32>,
100 [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue]>;
101 def ARM64call : SDNode<"ARM64ISD::CALL",
102 SDTypeProfile<0, -1, [SDTCisPtrTy<0>]>,
103 [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue,
105 def ARM64brcond : SDNode<"ARM64ISD::BRCOND", SDT_ARM64Brcond,
107 def ARM64cbz : SDNode<"ARM64ISD::CBZ", SDT_ARM64cbz,
109 def ARM64cbnz : SDNode<"ARM64ISD::CBNZ", SDT_ARM64cbz,
111 def ARM64tbz : SDNode<"ARM64ISD::TBZ", SDT_ARM64tbz,
113 def ARM64tbnz : SDNode<"ARM64ISD::TBNZ", SDT_ARM64tbz,
117 def ARM64csel : SDNode<"ARM64ISD::CSEL", SDT_ARM64CSel>;
118 def ARM64csinv : SDNode<"ARM64ISD::CSINV", SDT_ARM64CSel>;
119 def ARM64csneg : SDNode<"ARM64ISD::CSNEG", SDT_ARM64CSel>;
120 def ARM64csinc : SDNode<"ARM64ISD::CSINC", SDT_ARM64CSel>;
121 def ARM64retflag : SDNode<"ARM64ISD::RET_FLAG", SDTNone,
122 [SDNPHasChain, SDNPOptInGlue, SDNPVariadic]>;
123 def ARM64adc : SDNode<"ARM64ISD::ADC", SDTBinaryArithWithFlagsIn >;
124 def ARM64sbc : SDNode<"ARM64ISD::SBC", SDTBinaryArithWithFlagsIn>;
125 def ARM64add_flag : SDNode<"ARM64ISD::ADDS", SDTBinaryArithWithFlagsOut,
127 def ARM64sub_flag : SDNode<"ARM64ISD::SUBS", SDTBinaryArithWithFlagsOut>;
128 def ARM64and_flag : SDNode<"ARM64ISD::ANDS", SDTBinaryArithWithFlagsOut>;
129 def ARM64adc_flag : SDNode<"ARM64ISD::ADCS", SDTBinaryArithWithFlagsInOut>;
130 def ARM64sbc_flag : SDNode<"ARM64ISD::SBCS", SDTBinaryArithWithFlagsInOut>;
132 def ARM64threadpointer : SDNode<"ARM64ISD::THREAD_POINTER", SDTPtrLeaf>;
134 def ARM64fcmp : SDNode<"ARM64ISD::FCMP", SDT_ARM64FCmp>;
136 def ARM64fmax : SDNode<"ARM64ISD::FMAX", SDTFPBinOp>;
137 def ARM64fmin : SDNode<"ARM64ISD::FMIN", SDTFPBinOp>;
139 def ARM64dup : SDNode<"ARM64ISD::DUP", SDT_ARM64Dup>;
140 def ARM64duplane8 : SDNode<"ARM64ISD::DUPLANE8", SDT_ARM64DupLane>;
141 def ARM64duplane16 : SDNode<"ARM64ISD::DUPLANE16", SDT_ARM64DupLane>;
142 def ARM64duplane32 : SDNode<"ARM64ISD::DUPLANE32", SDT_ARM64DupLane>;
143 def ARM64duplane64 : SDNode<"ARM64ISD::DUPLANE64", SDT_ARM64DupLane>;
145 def ARM64zip1 : SDNode<"ARM64ISD::ZIP1", SDT_ARM64Zip>;
146 def ARM64zip2 : SDNode<"ARM64ISD::ZIP2", SDT_ARM64Zip>;
147 def ARM64uzp1 : SDNode<"ARM64ISD::UZP1", SDT_ARM64Zip>;
148 def ARM64uzp2 : SDNode<"ARM64ISD::UZP2", SDT_ARM64Zip>;
149 def ARM64trn1 : SDNode<"ARM64ISD::TRN1", SDT_ARM64Zip>;
150 def ARM64trn2 : SDNode<"ARM64ISD::TRN2", SDT_ARM64Zip>;
152 def ARM64movi_edit : SDNode<"ARM64ISD::MOVIedit", SDT_ARM64MOVIedit>;
153 def ARM64movi_shift : SDNode<"ARM64ISD::MOVIshift", SDT_ARM64MOVIshift>;
154 def ARM64movi_msl : SDNode<"ARM64ISD::MOVImsl", SDT_ARM64MOVIshift>;
155 def ARM64mvni_shift : SDNode<"ARM64ISD::MVNIshift", SDT_ARM64MOVIshift>;
156 def ARM64mvni_msl : SDNode<"ARM64ISD::MVNImsl", SDT_ARM64MOVIshift>;
157 def ARM64movi : SDNode<"ARM64ISD::MOVI", SDT_ARM64MOVIedit>;
158 def ARM64fmov : SDNode<"ARM64ISD::FMOV", SDT_ARM64MOVIedit>;
160 def ARM64rev16 : SDNode<"ARM64ISD::REV16", SDT_ARM64UnaryVec>;
161 def ARM64rev32 : SDNode<"ARM64ISD::REV32", SDT_ARM64UnaryVec>;
162 def ARM64rev64 : SDNode<"ARM64ISD::REV64", SDT_ARM64UnaryVec>;
163 def ARM64ext : SDNode<"ARM64ISD::EXT", SDT_ARM64ExtVec>;
165 def ARM64vashr : SDNode<"ARM64ISD::VASHR", SDT_ARM64vshift>;
166 def ARM64vlshr : SDNode<"ARM64ISD::VLSHR", SDT_ARM64vshift>;
167 def ARM64vshl : SDNode<"ARM64ISD::VSHL", SDT_ARM64vshift>;
168 def ARM64sqshli : SDNode<"ARM64ISD::SQSHL_I", SDT_ARM64vshift>;
169 def ARM64uqshli : SDNode<"ARM64ISD::UQSHL_I", SDT_ARM64vshift>;
170 def ARM64sqshlui : SDNode<"ARM64ISD::SQSHLU_I", SDT_ARM64vshift>;
171 def ARM64srshri : SDNode<"ARM64ISD::SRSHR_I", SDT_ARM64vshift>;
172 def ARM64urshri : SDNode<"ARM64ISD::URSHR_I", SDT_ARM64vshift>;
174 def ARM64not: SDNode<"ARM64ISD::NOT", SDT_ARM64unvec>;
175 def ARM64bit: SDNode<"ARM64ISD::BIT", SDT_ARM64trivec>;
177 def ARM64cmeq: SDNode<"ARM64ISD::CMEQ", SDT_ARM64binvec>;
178 def ARM64cmge: SDNode<"ARM64ISD::CMGE", SDT_ARM64binvec>;
179 def ARM64cmgt: SDNode<"ARM64ISD::CMGT", SDT_ARM64binvec>;
180 def ARM64cmhi: SDNode<"ARM64ISD::CMHI", SDT_ARM64binvec>;
181 def ARM64cmhs: SDNode<"ARM64ISD::CMHS", SDT_ARM64binvec>;
183 def ARM64fcmeq: SDNode<"ARM64ISD::FCMEQ", SDT_ARM64fcmp>;
184 def ARM64fcmge: SDNode<"ARM64ISD::FCMGE", SDT_ARM64fcmp>;
185 def ARM64fcmgt: SDNode<"ARM64ISD::FCMGT", SDT_ARM64fcmp>;
187 def ARM64cmeqz: SDNode<"ARM64ISD::CMEQz", SDT_ARM64unvec>;
188 def ARM64cmgez: SDNode<"ARM64ISD::CMGEz", SDT_ARM64unvec>;
189 def ARM64cmgtz: SDNode<"ARM64ISD::CMGTz", SDT_ARM64unvec>;
190 def ARM64cmlez: SDNode<"ARM64ISD::CMLEz", SDT_ARM64unvec>;
191 def ARM64cmltz: SDNode<"ARM64ISD::CMLTz", SDT_ARM64unvec>;
192 def ARM64cmtst : PatFrag<(ops node:$LHS, node:$RHS),
193 (ARM64not (ARM64cmeqz (and node:$LHS, node:$RHS)))>;
195 def ARM64fcmeqz: SDNode<"ARM64ISD::FCMEQz", SDT_ARM64fcmpz>;
196 def ARM64fcmgez: SDNode<"ARM64ISD::FCMGEz", SDT_ARM64fcmpz>;
197 def ARM64fcmgtz: SDNode<"ARM64ISD::FCMGTz", SDT_ARM64fcmpz>;
198 def ARM64fcmlez: SDNode<"ARM64ISD::FCMLEz", SDT_ARM64fcmpz>;
199 def ARM64fcmltz: SDNode<"ARM64ISD::FCMLTz", SDT_ARM64fcmpz>;
201 def ARM64bici: SDNode<"ARM64ISD::BICi", SDT_ARM64vecimm>;
202 def ARM64orri: SDNode<"ARM64ISD::ORRi", SDT_ARM64vecimm>;
204 def ARM64neg : SDNode<"ARM64ISD::NEG", SDT_ARM64unvec>;
206 def ARM64tcret: SDNode<"ARM64ISD::TC_RETURN", SDT_ARM64TCRET,
207 [SDNPHasChain, SDNPOptInGlue, SDNPVariadic]>;
209 def ARM64Prefetch : SDNode<"ARM64ISD::PREFETCH", SDT_ARM64PREFETCH,
210 [SDNPHasChain, SDNPSideEffect]>;
212 def ARM64sitof: SDNode<"ARM64ISD::SITOF", SDT_ARM64ITOF>;
213 def ARM64uitof: SDNode<"ARM64ISD::UITOF", SDT_ARM64ITOF>;
215 def ARM64tlsdesc_call : SDNode<"ARM64ISD::TLSDESC_CALL", SDT_ARM64TLSDescCall,
216 [SDNPInGlue, SDNPOutGlue, SDNPHasChain,
219 def ARM64WrapperLarge : SDNode<"ARM64ISD::WrapperLarge", SDT_ARM64WrapperLarge>;
222 //===----------------------------------------------------------------------===//
224 //===----------------------------------------------------------------------===//
226 // ARM64 Instruction Predicate Definitions.
228 def HasZCZ : Predicate<"Subtarget->hasZeroCycleZeroing()">;
229 def NoZCZ : Predicate<"!Subtarget->hasZeroCycleZeroing()">;
230 def IsDarwin : Predicate<"Subtarget->isTargetDarwin()">;
231 def IsNotDarwin: Predicate<"!Subtarget->isTargetDarwin()">;
232 def ForCodeSize : Predicate<"ForCodeSize">;
233 def NotForCodeSize : Predicate<"!ForCodeSize">;
235 include "ARM64InstrFormats.td"
237 //===----------------------------------------------------------------------===//
239 //===----------------------------------------------------------------------===//
240 // Miscellaneous instructions.
241 //===----------------------------------------------------------------------===//
243 let Defs = [SP], Uses = [SP], hasSideEffects = 1, isCodeGenOnly = 1 in {
244 def ADJCALLSTACKDOWN : Pseudo<(outs), (ins i32imm:$amt),
245 [(ARM64callseq_start timm:$amt)]>;
246 def ADJCALLSTACKUP : Pseudo<(outs), (ins i32imm:$amt1, i32imm:$amt2),
247 [(ARM64callseq_end timm:$amt1, timm:$amt2)]>;
248 } // Defs = [SP], Uses = [SP], hasSideEffects = 1, isCodeGenOnly = 1
250 let isReMaterializable = 1, isCodeGenOnly = 1 in {
251 // FIXME: The following pseudo instructions are only needed because remat
252 // cannot handle multiple instructions. When that changes, they can be
253 // removed, along with the ARM64Wrapper node.
255 let AddedComplexity = 10 in
256 def LOADgot : Pseudo<(outs GPR64:$dst), (ins i64imm:$addr),
257 [(set GPR64:$dst, (ARM64LOADgot tglobaladdr:$addr))]>,
260 // The MOVaddr instruction should match only when the add is not folded
261 // into a load or store address.
263 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
264 [(set GPR64:$dst, (ARM64addlow (ARM64adrp tglobaladdr:$hi),
265 tglobaladdr:$low))]>,
266 Sched<[WriteAdrAdr]>;
268 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
269 [(set GPR64:$dst, (ARM64addlow (ARM64adrp tjumptable:$hi),
271 Sched<[WriteAdrAdr]>;
273 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
274 [(set GPR64:$dst, (ARM64addlow (ARM64adrp tconstpool:$hi),
276 Sched<[WriteAdrAdr]>;
278 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
279 [(set GPR64:$dst, (ARM64addlow (ARM64adrp tblockaddress:$hi),
280 tblockaddress:$low))]>,
281 Sched<[WriteAdrAdr]>;
283 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
284 [(set GPR64:$dst, (ARM64addlow (ARM64adrp tglobaltlsaddr:$hi),
285 tglobaltlsaddr:$low))]>,
286 Sched<[WriteAdrAdr]>;
288 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
289 [(set GPR64:$dst, (ARM64addlow (ARM64adrp texternalsym:$hi),
290 texternalsym:$low))]>,
291 Sched<[WriteAdrAdr]>;
293 } // isReMaterializable, isCodeGenOnly
295 def : Pat<(ARM64LOADgot tglobaltlsaddr:$addr),
296 (LOADgot tglobaltlsaddr:$addr)>;
298 def : Pat<(ARM64LOADgot texternalsym:$addr),
299 (LOADgot texternalsym:$addr)>;
301 def : Pat<(ARM64LOADgot tconstpool:$addr),
302 (LOADgot tconstpool:$addr)>;
304 //===----------------------------------------------------------------------===//
305 // System instructions.
306 //===----------------------------------------------------------------------===//
308 def HINT : HintI<"hint">;
309 def : InstAlias<"nop", (HINT 0b000)>;
310 def : InstAlias<"yield",(HINT 0b001)>;
311 def : InstAlias<"wfe", (HINT 0b010)>;
312 def : InstAlias<"wfi", (HINT 0b011)>;
313 def : InstAlias<"sev", (HINT 0b100)>;
314 def : InstAlias<"sevl", (HINT 0b101)>;
316 // As far as LLVM is concerned this writes to the system's exclusive monitors.
317 let mayLoad = 1, mayStore = 1 in
318 def CLREX : CRmSystemI<imm0_15, 0b010, "clrex">;
320 def DMB : CRmSystemI<barrier_op, 0b101, "dmb">;
321 def DSB : CRmSystemI<barrier_op, 0b100, "dsb">;
322 def ISB : CRmSystemI<barrier_op, 0b110, "isb">;
323 def : InstAlias<"clrex", (CLREX 0xf)>;
324 def : InstAlias<"isb", (ISB 0xf)>;
328 def MSRcpsr: MSRcpsrI;
330 // The thread pointer (on Linux, at least, where this has been implemented) is
332 def : Pat<(ARM64threadpointer), (MRS 0xde82)>;
334 // Generic system instructions
335 def SYSxt : SystemXtI<0, "sys">;
336 def SYSLxt : SystemLXtI<1, "sysl">;
338 def : InstAlias<"sys $op1, $Cn, $Cm, $op2",
339 (SYSxt imm0_7:$op1, sys_cr_op:$Cn,
340 sys_cr_op:$Cm, imm0_7:$op2, XZR)>;
342 //===----------------------------------------------------------------------===//
343 // Move immediate instructions.
344 //===----------------------------------------------------------------------===//
346 defm MOVK : InsertImmediate<0b11, "movk">;
347 defm MOVN : MoveImmediate<0b00, "movn">;
349 let PostEncoderMethod = "fixMOVZ" in
350 defm MOVZ : MoveImmediate<0b10, "movz">;
352 def : InstAlias<"movk $dst, $imm", (MOVKWi GPR32:$dst, imm0_65535:$imm, 0)>;
353 def : InstAlias<"movk $dst, $imm", (MOVKXi GPR64:$dst, imm0_65535:$imm, 0)>;
354 def : InstAlias<"movn $dst, $imm", (MOVNWi GPR32:$dst, imm0_65535:$imm, 0)>;
355 def : InstAlias<"movn $dst, $imm", (MOVNXi GPR64:$dst, imm0_65535:$imm, 0)>;
356 def : InstAlias<"movz $dst, $imm", (MOVZWi GPR32:$dst, imm0_65535:$imm, 0)>;
357 def : InstAlias<"movz $dst, $imm", (MOVZXi GPR64:$dst, imm0_65535:$imm, 0)>;
359 def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movz_symbol_g3:$sym, 48)>;
360 def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movz_symbol_g2:$sym, 32)>;
361 def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movz_symbol_g1:$sym, 16)>;
362 def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movz_symbol_g0:$sym, 0)>;
364 def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movz_symbol_g3:$sym, 48)>;
365 def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movz_symbol_g2:$sym, 32)>;
366 def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movz_symbol_g1:$sym, 16)>;
367 def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movz_symbol_g0:$sym, 0)>;
369 def : InstAlias<"movz $Rd, $sym", (MOVZWi GPR32:$Rd, movz_symbol_g3:$sym, 48)>;
370 def : InstAlias<"movz $Rd, $sym", (MOVZWi GPR32:$Rd, movz_symbol_g2:$sym, 32)>;
371 def : InstAlias<"movz $Rd, $sym", (MOVZWi GPR32:$Rd, movz_symbol_g1:$sym, 16)>;
372 def : InstAlias<"movz $Rd, $sym", (MOVZWi GPR32:$Rd, movz_symbol_g0:$sym, 0)>;
374 def : InstAlias<"movk $Rd, $sym", (MOVKXi GPR64:$Rd, movk_symbol_g2:$sym, 32)>;
375 def : InstAlias<"movk $Rd, $sym", (MOVKXi GPR64:$Rd, movk_symbol_g1:$sym, 16)>;
376 def : InstAlias<"movk $Rd, $sym", (MOVKXi GPR64:$Rd, movk_symbol_g0:$sym, 0)>;
378 def : InstAlias<"movk $Rd, $sym", (MOVKWi GPR32:$Rd, movk_symbol_g2:$sym, 32)>;
379 def : InstAlias<"movk $Rd, $sym", (MOVKWi GPR32:$Rd, movk_symbol_g1:$sym, 16)>;
380 def : InstAlias<"movk $Rd, $sym", (MOVKWi GPR32:$Rd, movk_symbol_g0:$sym, 0)>;
382 let isReMaterializable = 1, isCodeGenOnly = 1, isMoveImm = 1,
383 isAsCheapAsAMove = 1 in {
384 // FIXME: The following pseudo instructions are only needed because remat
385 // cannot handle multiple instructions. When that changes, we can select
386 // directly to the real instructions and get rid of these pseudos.
389 : Pseudo<(outs GPR32:$dst), (ins i32imm:$src),
390 [(set GPR32:$dst, imm:$src)]>,
393 : Pseudo<(outs GPR64:$dst), (ins i64imm:$src),
394 [(set GPR64:$dst, imm:$src)]>,
396 } // isReMaterializable, isCodeGenOnly
398 def : Pat<(ARM64WrapperLarge tglobaladdr:$g3, tglobaladdr:$g2,
399 tglobaladdr:$g1, tglobaladdr:$g0),
400 (MOVKXi (MOVKXi (MOVKXi (MOVZXi tglobaladdr:$g3, 48),
401 tglobaladdr:$g2, 32),
402 tglobaladdr:$g1, 16),
403 tglobaladdr:$g0, 0)>;
405 def : Pat<(ARM64WrapperLarge tblockaddress:$g3, tblockaddress:$g2,
406 tblockaddress:$g1, tblockaddress:$g0),
407 (MOVKXi (MOVKXi (MOVKXi (MOVZXi tblockaddress:$g3, 48),
408 tblockaddress:$g2, 32),
409 tblockaddress:$g1, 16),
410 tblockaddress:$g0, 0)>;
412 def : Pat<(ARM64WrapperLarge tconstpool:$g3, tconstpool:$g2,
413 tconstpool:$g1, tconstpool:$g0),
414 (MOVKXi (MOVKXi (MOVKXi (MOVZXi tconstpool:$g3, 48),
420 //===----------------------------------------------------------------------===//
421 // Arithmetic instructions.
422 //===----------------------------------------------------------------------===//
424 // Add/subtract with carry.
425 defm ADC : AddSubCarry<0, "adc", "adcs", ARM64adc, ARM64adc_flag>;
426 defm SBC : AddSubCarry<1, "sbc", "sbcs", ARM64sbc, ARM64sbc_flag>;
428 def : InstAlias<"ngc $dst, $src", (SBCWr GPR32:$dst, WZR, GPR32:$src)>;
429 def : InstAlias<"ngc $dst, $src", (SBCXr GPR64:$dst, XZR, GPR64:$src)>;
430 def : InstAlias<"ngcs $dst, $src", (SBCSWr GPR32:$dst, WZR, GPR32:$src)>;
431 def : InstAlias<"ngcs $dst, $src", (SBCSXr GPR64:$dst, XZR, GPR64:$src)>;
434 defm ADD : AddSub<0, "add", add>;
435 defm SUB : AddSub<1, "sub">;
437 defm ADDS : AddSubS<0, "adds", ARM64add_flag>;
438 defm SUBS : AddSubS<1, "subs", ARM64sub_flag>;
440 // Use SUBS instead of SUB to enable CSE between SUBS and SUB.
441 def : Pat<(sub GPR32sp:$Rn, addsub_shifted_imm32:$imm),
442 (SUBSWri GPR32sp:$Rn, addsub_shifted_imm32:$imm)>;
443 def : Pat<(sub GPR64sp:$Rn, addsub_shifted_imm64:$imm),
444 (SUBSXri GPR64sp:$Rn, addsub_shifted_imm64:$imm)>;
445 def : Pat<(sub GPR32:$Rn, GPR32:$Rm),
446 (SUBSWrr GPR32:$Rn, GPR32:$Rm)>;
447 def : Pat<(sub GPR64:$Rn, GPR64:$Rm),
448 (SUBSXrr GPR64:$Rn, GPR64:$Rm)>;
449 def : Pat<(sub GPR32:$Rn, arith_shifted_reg32:$Rm),
450 (SUBSWrs GPR32:$Rn, arith_shifted_reg32:$Rm)>;
451 def : Pat<(sub GPR64:$Rn, arith_shifted_reg64:$Rm),
452 (SUBSXrs GPR64:$Rn, arith_shifted_reg64:$Rm)>;
453 def : Pat<(sub GPR32sp:$R2, arith_extended_reg32<i32>:$R3),
454 (SUBSWrx GPR32sp:$R2, arith_extended_reg32<i32>:$R3)>;
455 def : Pat<(sub GPR64sp:$R2, arith_extended_reg32to64<i64>:$R3),
456 (SUBSXrx GPR64sp:$R2, arith_extended_reg32to64<i64>:$R3)>;
458 // Because of the immediate format for add/sub-imm instructions, the
459 // expression (add x, -1) must be transformed to (SUB{W,X}ri x, 1).
460 // These patterns capture that transformation.
461 let AddedComplexity = 1 in {
462 def : Pat<(add GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
463 (SUBSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
464 def : Pat<(add GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
465 (SUBSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
466 def : Pat<(sub GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
467 (ADDWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
468 def : Pat<(sub GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
469 (ADDXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
472 def : InstAlias<"neg $dst, $src", (SUBWrs GPR32:$dst, WZR, GPR32:$src, 0)>;
473 def : InstAlias<"neg $dst, $src", (SUBXrs GPR64:$dst, XZR, GPR64:$src, 0)>;
474 def : InstAlias<"neg $dst, $src, $shift",
475 (SUBWrs GPR32:$dst, WZR, GPR32:$src, arith_shift:$shift)>;
476 def : InstAlias<"neg $dst, $src, $shift",
477 (SUBXrs GPR64:$dst, XZR, GPR64:$src, arith_shift:$shift)>;
479 // Because of the immediate format for add/sub-imm instructions, the
480 // expression (add x, -1) must be transformed to (SUB{W,X}ri x, 1).
481 // These patterns capture that transformation.
482 let AddedComplexity = 1 in {
483 def : Pat<(ARM64add_flag GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
484 (SUBSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
485 def : Pat<(ARM64add_flag GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
486 (SUBSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
487 def : Pat<(ARM64sub_flag GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
488 (ADDSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
489 def : Pat<(ARM64sub_flag GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
490 (ADDSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
493 def : InstAlias<"negs $dst, $src", (SUBSWrs GPR32:$dst, WZR, GPR32:$src, 0)>;
494 def : InstAlias<"negs $dst, $src", (SUBSXrs GPR64:$dst, XZR, GPR64:$src, 0)>;
495 def : InstAlias<"negs $dst, $src, $shift",
496 (SUBSWrs GPR32:$dst, WZR, GPR32:$src, arith_shift:$shift)>;
497 def : InstAlias<"negs $dst, $src, $shift",
498 (SUBSXrs GPR64:$dst, XZR, GPR64:$src, arith_shift:$shift)>;
500 // Unsigned/Signed divide
501 defm UDIV : Div<0, "udiv", udiv>;
502 defm SDIV : Div<1, "sdiv", sdiv>;
503 let isCodeGenOnly = 1 in {
504 defm UDIV_Int : Div<0, "udiv", int_arm64_udiv>;
505 defm SDIV_Int : Div<1, "sdiv", int_arm64_sdiv>;
509 defm ASRV : Shift<0b10, "asrv", sra>;
510 defm LSLV : Shift<0b00, "lslv", shl>;
511 defm LSRV : Shift<0b01, "lsrv", srl>;
512 defm RORV : Shift<0b11, "rorv", rotr>;
514 def : ShiftAlias<"asr", ASRVWr, GPR32>;
515 def : ShiftAlias<"asr", ASRVXr, GPR64>;
516 def : ShiftAlias<"lsl", LSLVWr, GPR32>;
517 def : ShiftAlias<"lsl", LSLVXr, GPR64>;
518 def : ShiftAlias<"lsr", LSRVWr, GPR32>;
519 def : ShiftAlias<"lsr", LSRVXr, GPR64>;
520 def : ShiftAlias<"ror", RORVWr, GPR32>;
521 def : ShiftAlias<"ror", RORVXr, GPR64>;
524 let AddedComplexity = 7 in {
525 defm MADD : MulAccum<0, "madd", add>;
526 defm MSUB : MulAccum<1, "msub", sub>;
528 def : Pat<(i32 (mul GPR32:$Rn, GPR32:$Rm)),
529 (MADDWrrr GPR32:$Rn, GPR32:$Rm, WZR)>;
530 def : Pat<(i64 (mul GPR64:$Rn, GPR64:$Rm)),
531 (MADDXrrr GPR64:$Rn, GPR64:$Rm, XZR)>;
533 def : Pat<(i32 (ineg (mul GPR32:$Rn, GPR32:$Rm))),
534 (MSUBWrrr GPR32:$Rn, GPR32:$Rm, WZR)>;
535 def : Pat<(i64 (ineg (mul GPR64:$Rn, GPR64:$Rm))),
536 (MSUBXrrr GPR64:$Rn, GPR64:$Rm, XZR)>;
537 } // AddedComplexity = 7
539 let AddedComplexity = 5 in {
540 def SMADDLrrr : WideMulAccum<0, 0b001, "smaddl", add, sext>;
541 def SMSUBLrrr : WideMulAccum<1, 0b001, "smsubl", sub, sext>;
542 def UMADDLrrr : WideMulAccum<0, 0b101, "umaddl", add, zext>;
543 def UMSUBLrrr : WideMulAccum<1, 0b101, "umsubl", sub, zext>;
545 def : Pat<(i64 (mul (sext GPR32:$Rn), (sext GPR32:$Rm))),
546 (SMADDLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
547 def : Pat<(i64 (mul (zext GPR32:$Rn), (zext GPR32:$Rm))),
548 (UMADDLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
550 def : Pat<(i64 (ineg (mul (sext GPR32:$Rn), (sext GPR32:$Rm)))),
551 (SMSUBLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
552 def : Pat<(i64 (ineg (mul (zext GPR32:$Rn), (zext GPR32:$Rm)))),
553 (UMSUBLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
554 } // AddedComplexity = 5
556 def : MulAccumWAlias<"mul", MADDWrrr>;
557 def : MulAccumXAlias<"mul", MADDXrrr>;
558 def : MulAccumWAlias<"mneg", MSUBWrrr>;
559 def : MulAccumXAlias<"mneg", MSUBXrrr>;
560 def : WideMulAccumAlias<"smull", SMADDLrrr>;
561 def : WideMulAccumAlias<"smnegl", SMSUBLrrr>;
562 def : WideMulAccumAlias<"umull", UMADDLrrr>;
563 def : WideMulAccumAlias<"umnegl", UMSUBLrrr>;
566 def SMULHrr : MulHi<0b010, "smulh", mulhs>;
567 def UMULHrr : MulHi<0b110, "umulh", mulhu>;
570 def CRC32Brr : BaseCRC32<0, 0b00, 0, GPR32, int_arm64_crc32b, "crc32b">;
571 def CRC32Hrr : BaseCRC32<0, 0b01, 0, GPR32, int_arm64_crc32h, "crc32h">;
572 def CRC32Wrr : BaseCRC32<0, 0b10, 0, GPR32, int_arm64_crc32w, "crc32w">;
573 def CRC32Xrr : BaseCRC32<1, 0b11, 0, GPR64, int_arm64_crc32x, "crc32x">;
575 def CRC32CBrr : BaseCRC32<0, 0b00, 1, GPR32, int_arm64_crc32cb, "crc32cb">;
576 def CRC32CHrr : BaseCRC32<0, 0b01, 1, GPR32, int_arm64_crc32ch, "crc32ch">;
577 def CRC32CWrr : BaseCRC32<0, 0b10, 1, GPR32, int_arm64_crc32cw, "crc32cw">;
578 def CRC32CXrr : BaseCRC32<1, 0b11, 1, GPR64, int_arm64_crc32cx, "crc32cx">;
581 //===----------------------------------------------------------------------===//
582 // Logical instructions.
583 //===----------------------------------------------------------------------===//
586 defm ANDS : LogicalImmS<0b11, "ands", ARM64and_flag>;
587 defm AND : LogicalImm<0b00, "and", and>;
588 defm EOR : LogicalImm<0b10, "eor", xor>;
589 defm ORR : LogicalImm<0b01, "orr", or>;
591 def : InstAlias<"mov $dst, $imm", (ORRWri GPR32sp:$dst, WZR,
592 logical_imm32:$imm)>;
593 def : InstAlias<"mov $dst, $imm", (ORRXri GPR64sp:$dst, XZR,
594 logical_imm64:$imm)>;
598 defm ANDS : LogicalRegS<0b11, 0, "ands">;
599 defm BICS : LogicalRegS<0b11, 1, "bics">;
600 defm AND : LogicalReg<0b00, 0, "and", and>;
601 defm BIC : LogicalReg<0b00, 1, "bic",
602 BinOpFrag<(and node:$LHS, (not node:$RHS))>>;
603 defm EON : LogicalReg<0b10, 1, "eon",
604 BinOpFrag<(xor node:$LHS, (not node:$RHS))>>;
605 defm EOR : LogicalReg<0b10, 0, "eor", xor>;
606 defm ORN : LogicalReg<0b01, 1, "orn",
607 BinOpFrag<(or node:$LHS, (not node:$RHS))>>;
608 defm ORR : LogicalReg<0b01, 0, "orr", or>;
610 def : InstAlias<"tst $src1, $src2",
611 (ANDSWri WZR, GPR32:$src1, logical_imm32:$src2)>;
612 def : InstAlias<"tst $src1, $src2",
613 (ANDSXri XZR, GPR64:$src1, logical_imm64:$src2)>;
615 def : InstAlias<"tst $src1, $src2",
616 (ANDSWrs WZR, GPR32:$src1, GPR32:$src2, 0)>;
617 def : InstAlias<"tst $src1, $src2",
618 (ANDSXrs XZR, GPR64:$src1, GPR64:$src2, 0)>;
620 def : InstAlias<"tst $src1, $src2, $sh",
621 (ANDSWrs WZR, GPR32:$src1, GPR32:$src2, logical_shift:$sh)>;
622 def : InstAlias<"tst $src1, $src2, $sh",
623 (ANDSXrs XZR, GPR64:$src1, GPR64:$src2, logical_shift:$sh)>;
625 def : InstAlias<"mvn $Wd, $Wm",
626 (ORNWrs GPR32:$Wd, WZR, GPR32:$Wm, 0)>;
627 def : InstAlias<"mvn $Xd, $Xm",
628 (ORNXrs GPR64:$Xd, XZR, GPR64:$Xm, 0)>;
630 def : InstAlias<"mvn $Wd, $Wm, $sh",
631 (ORNWrs GPR32:$Wd, WZR, GPR32:$Wm, logical_shift:$sh)>;
632 def : InstAlias<"mvn $Xd, $Xm, $sh",
633 (ORNXrs GPR64:$Xd, XZR, GPR64:$Xm, logical_shift:$sh)>;
635 def : Pat<(not GPR32:$Wm), (ORNWrr WZR, GPR32:$Wm)>;
636 def : Pat<(not GPR64:$Xm), (ORNXrr XZR, GPR64:$Xm)>;
639 //===----------------------------------------------------------------------===//
640 // One operand data processing instructions.
641 //===----------------------------------------------------------------------===//
643 defm CLS : OneOperandData<0b101, "cls">;
644 defm CLZ : OneOperandData<0b100, "clz", ctlz>;
645 defm RBIT : OneOperandData<0b000, "rbit">;
646 def REV16Wr : OneWRegData<0b001, "rev16",
647 UnOpFrag<(rotr (bswap node:$LHS), (i64 16))>>;
648 def REV16Xr : OneXRegData<0b001, "rev16",
649 UnOpFrag<(rotr (bswap node:$LHS), (i64 16))>>;
651 def : Pat<(cttz GPR32:$Rn),
652 (CLZWr (RBITWr GPR32:$Rn))>;
653 def : Pat<(cttz GPR64:$Rn),
654 (CLZXr (RBITXr GPR64:$Rn))>;
656 // Unlike the other one operand instructions, the instructions with the "rev"
657 // mnemonic do *not* just different in the size bit, but actually use different
658 // opcode bits for the different sizes.
659 def REVWr : OneWRegData<0b010, "rev", bswap>;
660 def REVXr : OneXRegData<0b011, "rev", bswap>;
661 def REV32Xr : OneXRegData<0b010, "rev32",
662 UnOpFrag<(rotr (bswap node:$LHS), (i64 32))>>;
664 //===----------------------------------------------------------------------===//
665 // Bitfield immediate extraction instruction.
666 //===----------------------------------------------------------------------===//
667 let neverHasSideEffects = 1 in
668 defm EXTR : ExtractImm<"extr">;
669 def : InstAlias<"ror $dst, $src, $shift",
670 (EXTRWrri GPR32:$dst, GPR32:$src, GPR32:$src, imm0_31:$shift)>;
671 def : InstAlias<"ror $dst, $src, $shift",
672 (EXTRXrri GPR64:$dst, GPR64:$src, GPR64:$src, imm0_63:$shift)>;
674 def : Pat<(rotr GPR32:$Rn, (i64 imm0_31:$imm)),
675 (EXTRWrri GPR32:$Rn, GPR32:$Rn, imm0_31:$imm)>;
676 def : Pat<(rotr GPR64:$Rn, (i64 imm0_63:$imm)),
677 (EXTRXrri GPR64:$Rn, GPR64:$Rn, imm0_63:$imm)>;
679 //===----------------------------------------------------------------------===//
680 // Other bitfield immediate instructions.
681 //===----------------------------------------------------------------------===//
682 let neverHasSideEffects = 1 in {
683 defm BFM : BitfieldImmWith2RegArgs<0b01, "bfm">;
684 defm SBFM : BitfieldImm<0b00, "sbfm">;
685 defm UBFM : BitfieldImm<0b10, "ubfm">;
688 def i32shift_a : Operand<i64>, SDNodeXForm<imm, [{
689 uint64_t enc = (32 - N->getZExtValue()) & 0x1f;
690 return CurDAG->getTargetConstant(enc, MVT::i64);
693 def i32shift_b : Operand<i64>, SDNodeXForm<imm, [{
694 uint64_t enc = 31 - N->getZExtValue();
695 return CurDAG->getTargetConstant(enc, MVT::i64);
698 // min(7, 31 - shift_amt)
699 def i32shift_sext_i8 : Operand<i64>, SDNodeXForm<imm, [{
700 uint64_t enc = 31 - N->getZExtValue();
701 enc = enc > 7 ? 7 : enc;
702 return CurDAG->getTargetConstant(enc, MVT::i64);
705 // min(15, 31 - shift_amt)
706 def i32shift_sext_i16 : Operand<i64>, SDNodeXForm<imm, [{
707 uint64_t enc = 31 - N->getZExtValue();
708 enc = enc > 15 ? 15 : enc;
709 return CurDAG->getTargetConstant(enc, MVT::i64);
712 def i64shift_a : Operand<i64>, SDNodeXForm<imm, [{
713 uint64_t enc = (64 - N->getZExtValue()) & 0x3f;
714 return CurDAG->getTargetConstant(enc, MVT::i64);
717 def i64shift_b : Operand<i64>, SDNodeXForm<imm, [{
718 uint64_t enc = 63 - N->getZExtValue();
719 return CurDAG->getTargetConstant(enc, MVT::i64);
722 // min(7, 63 - shift_amt)
723 def i64shift_sext_i8 : Operand<i64>, SDNodeXForm<imm, [{
724 uint64_t enc = 63 - N->getZExtValue();
725 enc = enc > 7 ? 7 : enc;
726 return CurDAG->getTargetConstant(enc, MVT::i64);
729 // min(15, 63 - shift_amt)
730 def i64shift_sext_i16 : Operand<i64>, SDNodeXForm<imm, [{
731 uint64_t enc = 63 - N->getZExtValue();
732 enc = enc > 15 ? 15 : enc;
733 return CurDAG->getTargetConstant(enc, MVT::i64);
736 // min(31, 63 - shift_amt)
737 def i64shift_sext_i32 : Operand<i64>, SDNodeXForm<imm, [{
738 uint64_t enc = 63 - N->getZExtValue();
739 enc = enc > 31 ? 31 : enc;
740 return CurDAG->getTargetConstant(enc, MVT::i64);
743 def : Pat<(shl GPR32:$Rn, (i64 imm0_31:$imm)),
744 (UBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
745 (i64 (i32shift_b imm0_31:$imm)))>;
746 def : Pat<(shl GPR64:$Rn, (i64 imm0_63:$imm)),
747 (UBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
748 (i64 (i64shift_b imm0_63:$imm)))>;
750 let AddedComplexity = 10 in {
751 def : Pat<(sra GPR32:$Rn, (i64 imm0_31:$imm)),
752 (SBFMWri GPR32:$Rn, imm0_31:$imm, 31)>;
753 def : Pat<(sra GPR64:$Rn, (i64 imm0_63:$imm)),
754 (SBFMXri GPR64:$Rn, imm0_63:$imm, 63)>;
757 def : InstAlias<"asr $dst, $src, $shift",
758 (SBFMWri GPR32:$dst, GPR32:$src, imm0_31:$shift, 31)>;
759 def : InstAlias<"asr $dst, $src, $shift",
760 (SBFMXri GPR64:$dst, GPR64:$src, imm0_63:$shift, 63)>;
761 def : InstAlias<"sxtb $dst, $src", (SBFMWri GPR32:$dst, GPR32:$src, 0, 7)>;
762 def : InstAlias<"sxtb $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 7)>;
763 def : InstAlias<"sxth $dst, $src", (SBFMWri GPR32:$dst, GPR32:$src, 0, 15)>;
764 def : InstAlias<"sxth $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 15)>;
765 def : InstAlias<"sxtw $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 31)>;
767 def : Pat<(srl GPR32:$Rn, (i64 imm0_31:$imm)),
768 (UBFMWri GPR32:$Rn, imm0_31:$imm, 31)>;
769 def : Pat<(srl GPR64:$Rn, (i64 imm0_63:$imm)),
770 (UBFMXri GPR64:$Rn, imm0_63:$imm, 63)>;
772 def : InstAlias<"lsr $dst, $src, $shift",
773 (UBFMWri GPR32:$dst, GPR32:$src, imm0_31:$shift, 31)>;
774 def : InstAlias<"lsr $dst, $src, $shift",
775 (UBFMXri GPR64:$dst, GPR64:$src, imm0_63:$shift, 63)>;
776 def : InstAlias<"uxtb $dst, $src", (UBFMWri GPR32:$dst, GPR32:$src, 0, 7)>;
777 def : InstAlias<"uxtb $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 7)>;
778 def : InstAlias<"uxth $dst, $src", (UBFMWri GPR32:$dst, GPR32:$src, 0, 15)>;
779 def : InstAlias<"uxth $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 15)>;
780 def : InstAlias<"uxtw $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 31)>;
782 //===----------------------------------------------------------------------===//
783 // Conditionally set flags instructions.
784 //===----------------------------------------------------------------------===//
785 defm CCMN : CondSetFlagsImm<0, "ccmn">;
786 defm CCMP : CondSetFlagsImm<1, "ccmp">;
788 defm CCMN : CondSetFlagsReg<0, "ccmn">;
789 defm CCMP : CondSetFlagsReg<1, "ccmp">;
791 //===----------------------------------------------------------------------===//
792 // Conditional select instructions.
793 //===----------------------------------------------------------------------===//
794 defm CSEL : CondSelect<0, 0b00, "csel">;
796 def inc : PatFrag<(ops node:$in), (add node:$in, 1)>;
797 defm CSINC : CondSelectOp<0, 0b01, "csinc", inc>;
798 defm CSINV : CondSelectOp<1, 0b00, "csinv", not>;
799 defm CSNEG : CondSelectOp<1, 0b01, "csneg", ineg>;
801 def : Pat<(ARM64csinv GPR32:$tval, GPR32:$fval, (i32 imm:$cc), CPSR),
802 (CSINVWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
803 def : Pat<(ARM64csinv GPR64:$tval, GPR64:$fval, (i32 imm:$cc), CPSR),
804 (CSINVXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
805 def : Pat<(ARM64csneg GPR32:$tval, GPR32:$fval, (i32 imm:$cc), CPSR),
806 (CSNEGWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
807 def : Pat<(ARM64csneg GPR64:$tval, GPR64:$fval, (i32 imm:$cc), CPSR),
808 (CSNEGXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
809 def : Pat<(ARM64csinc GPR32:$tval, GPR32:$fval, (i32 imm:$cc), CPSR),
810 (CSINCWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
811 def : Pat<(ARM64csinc GPR64:$tval, GPR64:$fval, (i32 imm:$cc), CPSR),
812 (CSINCXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
814 def : Pat<(ARM64csel (i32 0), (i32 1), (i32 imm:$cc), CPSR),
815 (CSINCWr WZR, WZR, (i32 imm:$cc))>;
816 def : Pat<(ARM64csel (i64 0), (i64 1), (i32 imm:$cc), CPSR),
817 (CSINCXr XZR, XZR, (i32 imm:$cc))>;
818 def : Pat<(ARM64csel (i32 0), (i32 -1), (i32 imm:$cc), CPSR),
819 (CSINVWr WZR, WZR, (i32 imm:$cc))>;
820 def : Pat<(ARM64csel (i64 0), (i64 -1), (i32 imm:$cc), CPSR),
821 (CSINVXr XZR, XZR, (i32 imm:$cc))>;
823 // The inverse of the condition code from the alias instruction is what is used
824 // in the aliased instruction. The parser all ready inverts the condition code
825 // for these aliases.
826 // FIXME: Is this the correct way to handle these aliases?
827 def : InstAlias<"cset $dst, $cc", (CSINCWr GPR32:$dst, WZR, WZR, ccode:$cc)>;
828 def : InstAlias<"cset $dst, $cc", (CSINCXr GPR64:$dst, XZR, XZR, ccode:$cc)>;
830 def : InstAlias<"csetm $dst, $cc", (CSINVWr GPR32:$dst, WZR, WZR, ccode:$cc)>;
831 def : InstAlias<"csetm $dst, $cc", (CSINVXr GPR64:$dst, XZR, XZR, ccode:$cc)>;
833 def : InstAlias<"cinc $dst, $src, $cc",
834 (CSINCWr GPR32:$dst, GPR32:$src, GPR32:$src, ccode:$cc)>;
835 def : InstAlias<"cinc $dst, $src, $cc",
836 (CSINCXr GPR64:$dst, GPR64:$src, GPR64:$src, ccode:$cc)>;
838 def : InstAlias<"cinv $dst, $src, $cc",
839 (CSINVWr GPR32:$dst, GPR32:$src, GPR32:$src, ccode:$cc)>;
840 def : InstAlias<"cinv $dst, $src, $cc",
841 (CSINVXr GPR64:$dst, GPR64:$src, GPR64:$src, ccode:$cc)>;
843 def : InstAlias<"cneg $dst, $src, $cc",
844 (CSNEGWr GPR32:$dst, GPR32:$src, GPR32:$src, ccode:$cc)>;
845 def : InstAlias<"cneg $dst, $src, $cc",
846 (CSNEGXr GPR64:$dst, GPR64:$src, GPR64:$src, ccode:$cc)>;
848 //===----------------------------------------------------------------------===//
849 // PC-relative instructions.
850 //===----------------------------------------------------------------------===//
851 let isReMaterializable = 1 in {
852 let neverHasSideEffects = 1, mayStore = 0, mayLoad = 0 in {
853 def ADR : ADRI<0, "adr", adrlabel, []>;
854 } // neverHasSideEffects = 1
856 def ADRP : ADRI<1, "adrp", adrplabel,
857 [(set GPR64:$Xd, (ARM64adrp tglobaladdr:$label))]>;
858 } // isReMaterializable = 1
860 // page address of a constant pool entry, block address
861 def : Pat<(ARM64adrp tconstpool:$cp), (ADRP tconstpool:$cp)>;
862 def : Pat<(ARM64adrp tblockaddress:$cp), (ADRP tblockaddress:$cp)>;
864 //===----------------------------------------------------------------------===//
865 // Unconditional branch (register) instructions.
866 //===----------------------------------------------------------------------===//
868 let isReturn = 1, isTerminator = 1, isBarrier = 1 in {
869 def RET : BranchReg<0b0010, "ret", []>;
870 def DRPS : SpecialReturn<0b0101, "drps">;
871 def ERET : SpecialReturn<0b0100, "eret">;
872 } // isReturn = 1, isTerminator = 1, isBarrier = 1
874 // Default to the LR register.
875 def : InstAlias<"ret", (RET LR)>;
877 let isCall = 1, Defs = [LR], Uses = [SP] in {
878 def BLR : BranchReg<0b0001, "blr", [(ARM64call GPR64:$Rn)]>;
881 let isBranch = 1, isTerminator = 1, isBarrier = 1, isIndirectBranch = 1 in {
882 def BR : BranchReg<0b0000, "br", [(brind GPR64:$Rn)]>;
883 } // isBranch, isTerminator, isBarrier, isIndirectBranch
885 // Create a separate pseudo-instruction for codegen to use so that we don't
886 // flag lr as used in every function. It'll be restored before the RET by the
887 // epilogue if it's legitimately used.
888 def RET_ReallyLR : Pseudo<(outs), (ins), [(ARM64retflag)]> {
889 let isTerminator = 1;
894 // This is a directive-like pseudo-instruction. The purpose is to insert an
895 // R_AARCH64_TLSDESC_CALL relocation at the offset of the following instruction
896 // (which in the usual case is a BLR).
897 let hasSideEffects = 1 in
898 def TLSDESCCALL : Pseudo<(outs), (ins i64imm:$sym), []> {
899 let AsmString = ".tlsdesccall $sym";
902 // Pseudo-instruction representing a BLR with attached TLSDESC relocation. It
903 // gets expanded to two MCInsts during lowering.
904 let isCall = 1, Defs = [LR] in
906 : Pseudo<(outs), (ins GPR64:$dest, i64imm:$sym),
907 [(ARM64tlsdesc_call GPR64:$dest, tglobaltlsaddr:$sym)]>;
909 def : Pat<(ARM64tlsdesc_call GPR64:$dest, texternalsym:$sym),
910 (TLSDESC_BLR GPR64:$dest, texternalsym:$sym)>;
911 //===----------------------------------------------------------------------===//
912 // Conditional branch (immediate) instruction.
913 //===----------------------------------------------------------------------===//
914 def Bcc : BranchCond;
916 //===----------------------------------------------------------------------===//
917 // Compare-and-branch instructions.
918 //===----------------------------------------------------------------------===//
919 defm CBZ : CmpBranch<0, "cbz", ARM64cbz>;
920 defm CBNZ : CmpBranch<1, "cbnz", ARM64cbnz>;
922 //===----------------------------------------------------------------------===//
923 // Test-bit-and-branch instructions.
924 //===----------------------------------------------------------------------===//
925 def TBZ : TestBranch<0, "tbz", ARM64tbz>;
926 def TBNZ : TestBranch<1, "tbnz", ARM64tbnz>;
928 //===----------------------------------------------------------------------===//
929 // Unconditional branch (immediate) instructions.
930 //===----------------------------------------------------------------------===//
931 let isBranch = 1, isTerminator = 1, isBarrier = 1 in {
932 def B : BranchImm<0, "b", [(br bb:$addr)]>;
933 } // isBranch, isTerminator, isBarrier
935 let isCall = 1, Defs = [LR], Uses = [SP] in {
936 def BL : CallImm<1, "bl", [(ARM64call tglobaladdr:$addr)]>;
938 def : Pat<(ARM64call texternalsym:$func), (BL texternalsym:$func)>;
940 //===----------------------------------------------------------------------===//
941 // Exception generation instructions.
942 //===----------------------------------------------------------------------===//
943 def BRK : ExceptionGeneration<0b001, 0b00, "brk">;
944 def DCPS1 : ExceptionGeneration<0b101, 0b01, "dcps1">;
945 def DCPS2 : ExceptionGeneration<0b101, 0b10, "dcps2">;
946 def DCPS3 : ExceptionGeneration<0b101, 0b11, "dcps3">;
947 def HLT : ExceptionGeneration<0b010, 0b00, "hlt">;
948 def HVC : ExceptionGeneration<0b000, 0b10, "hvc">;
949 def SMC : ExceptionGeneration<0b000, 0b11, "smc">;
950 def SVC : ExceptionGeneration<0b000, 0b01, "svc">;
952 // DCPSn defaults to an immediate operand of zero if unspecified.
953 def : InstAlias<"dcps1", (DCPS1 0)>;
954 def : InstAlias<"dcps2", (DCPS2 0)>;
955 def : InstAlias<"dcps3", (DCPS3 0)>;
957 //===----------------------------------------------------------------------===//
958 // Load instructions.
959 //===----------------------------------------------------------------------===//
961 // Pair (indexed, offset)
962 def LDPWi : LoadPairOffset<0b00, 0, GPR32, am_indexed32simm7, "ldp">;
963 def LDPXi : LoadPairOffset<0b10, 0, GPR64, am_indexed64simm7, "ldp">;
964 def LDPSi : LoadPairOffset<0b00, 1, FPR32, am_indexed32simm7, "ldp">;
965 def LDPDi : LoadPairOffset<0b01, 1, FPR64, am_indexed64simm7, "ldp">;
966 def LDPQi : LoadPairOffset<0b10, 1, FPR128, am_indexed128simm7, "ldp">;
968 def LDPSWi : LoadPairOffset<0b01, 0, GPR64, am_indexed32simm7, "ldpsw">;
970 // Pair (pre-indexed)
971 def LDPWpre : LoadPairPreIdx<0b00, 0, GPR32, am_indexed32simm7_wb, "ldp">;
972 def LDPXpre : LoadPairPreIdx<0b10, 0, GPR64, am_indexed64simm7_wb, "ldp">;
973 def LDPSpre : LoadPairPreIdx<0b00, 1, FPR32, am_indexed32simm7_wb, "ldp">;
974 def LDPDpre : LoadPairPreIdx<0b01, 1, FPR64, am_indexed64simm7_wb, "ldp">;
975 def LDPQpre : LoadPairPreIdx<0b10, 1, FPR128, am_indexed128simm7_wb, "ldp">;
977 def LDPSWpre : LoadPairPreIdx<0b01, 0, GPR64, am_indexed32simm7_wb, "ldpsw">;
979 // Pair (post-indexed)
980 def LDPWpost : LoadPairPostIdx<0b00, 0, GPR32, simm7s4, "ldp">;
981 def LDPXpost : LoadPairPostIdx<0b10, 0, GPR64, simm7s8, "ldp">;
982 def LDPSpost : LoadPairPostIdx<0b00, 1, FPR32, simm7s4, "ldp">;
983 def LDPDpost : LoadPairPostIdx<0b01, 1, FPR64, simm7s8, "ldp">;
984 def LDPQpost : LoadPairPostIdx<0b10, 1, FPR128, simm7s16, "ldp">;
986 def LDPSWpost : LoadPairPostIdx<0b01, 0, GPR64, simm7s4, "ldpsw">;
989 // Pair (no allocate)
990 def LDNPWi : LoadPairNoAlloc<0b00, 0, GPR32, am_indexed32simm7, "ldnp">;
991 def LDNPXi : LoadPairNoAlloc<0b10, 0, GPR64, am_indexed64simm7, "ldnp">;
992 def LDNPSi : LoadPairNoAlloc<0b00, 1, FPR32, am_indexed32simm7, "ldnp">;
993 def LDNPDi : LoadPairNoAlloc<0b01, 1, FPR64, am_indexed64simm7, "ldnp">;
994 def LDNPQi : LoadPairNoAlloc<0b10, 1, FPR128, am_indexed128simm7, "ldnp">;
1000 let AddedComplexity = 10 in {
1002 def LDRBBro : Load8RO<0b00, 0, 0b01, GPR32, "ldrb",
1003 [(set GPR32:$Rt, (zextloadi8 ro_indexed8:$addr))]>;
1004 def LDRHHro : Load16RO<0b01, 0, 0b01, GPR32, "ldrh",
1005 [(set GPR32:$Rt, (zextloadi16 ro_indexed16:$addr))]>;
1006 def LDRWro : Load32RO<0b10, 0, 0b01, GPR32, "ldr",
1007 [(set GPR32:$Rt, (load ro_indexed32:$addr))]>;
1008 def LDRXro : Load64RO<0b11, 0, 0b01, GPR64, "ldr",
1009 [(set GPR64:$Rt, (load ro_indexed64:$addr))]>;
1012 def LDRBro : Load8RO<0b00, 1, 0b01, FPR8, "ldr",
1013 [(set FPR8:$Rt, (load ro_indexed8:$addr))]>;
1014 def LDRHro : Load16RO<0b01, 1, 0b01, FPR16, "ldr",
1015 [(set FPR16:$Rt, (load ro_indexed16:$addr))]>;
1016 def LDRSro : Load32RO<0b10, 1, 0b01, FPR32, "ldr",
1017 [(set (f32 FPR32:$Rt), (load ro_indexed32:$addr))]>;
1018 def LDRDro : Load64RO<0b11, 1, 0b01, FPR64, "ldr",
1019 [(set (f64 FPR64:$Rt), (load ro_indexed64:$addr))]>;
1020 def LDRQro : Load128RO<0b00, 1, 0b11, FPR128, "ldr", []> {
1024 // For regular load, we do not have any alignment requirement.
1025 // Thus, it is safe to directly map the vector loads with interesting
1026 // addressing modes.
1027 // FIXME: We could do the same for bitconvert to floating point vectors.
1028 def : Pat <(v8i8 (scalar_to_vector (i32 (extloadi8 ro_indexed8:$addr)))),
1029 (INSERT_SUBREG (v8i8 (IMPLICIT_DEF)),
1030 (LDRBro ro_indexed8:$addr), bsub)>;
1031 def : Pat <(v16i8 (scalar_to_vector (i32 (extloadi8 ro_indexed8:$addr)))),
1032 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
1033 (LDRBro ro_indexed8:$addr), bsub)>;
1034 def : Pat <(v4i16 (scalar_to_vector (i32 (extloadi16 ro_indexed16:$addr)))),
1035 (INSERT_SUBREG (v4i16 (IMPLICIT_DEF)),
1036 (LDRHro ro_indexed16:$addr), hsub)>;
1037 def : Pat <(v8i16 (scalar_to_vector (i32 (extloadi16 ro_indexed16:$addr)))),
1038 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
1039 (LDRHro ro_indexed16:$addr), hsub)>;
1040 def : Pat <(v2i32 (scalar_to_vector (i32 (load ro_indexed32:$addr)))),
1041 (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)),
1042 (LDRSro ro_indexed32:$addr), ssub)>;
1043 def : Pat <(v4i32 (scalar_to_vector (i32 (load ro_indexed32:$addr)))),
1044 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
1045 (LDRSro ro_indexed32:$addr), ssub)>;
1046 def : Pat <(v1i64 (scalar_to_vector (i64 (load ro_indexed64:$addr)))),
1047 (LDRDro ro_indexed64:$addr)>;
1048 def : Pat <(v2i64 (scalar_to_vector (i64 (load ro_indexed64:$addr)))),
1049 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
1050 (LDRDro ro_indexed64:$addr), dsub)>;
1052 // Match all load 64 bits width whose type is compatible with FPR64
1053 def : Pat<(v2f32 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1054 def : Pat<(v1f64 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1055 def : Pat<(v8i8 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1056 def : Pat<(v4i16 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1057 def : Pat<(v2i32 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1058 def : Pat<(v1i64 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1060 // Match all load 128 bits width whose type is compatible with FPR128
1061 def : Pat<(v4f32 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1062 def : Pat<(v2f64 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1063 def : Pat<(v16i8 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1064 def : Pat<(v8i16 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1065 def : Pat<(v4i32 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1066 def : Pat<(v2i64 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1067 def : Pat<(f128 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1069 // Load sign-extended half-word
1070 def LDRSHWro : Load16RO<0b01, 0, 0b11, GPR32, "ldrsh",
1071 [(set GPR32:$Rt, (sextloadi16 ro_indexed16:$addr))]>;
1072 def LDRSHXro : Load16RO<0b01, 0, 0b10, GPR64, "ldrsh",
1073 [(set GPR64:$Rt, (sextloadi16 ro_indexed16:$addr))]>;
1075 // Load sign-extended byte
1076 def LDRSBWro : Load8RO<0b00, 0, 0b11, GPR32, "ldrsb",
1077 [(set GPR32:$Rt, (sextloadi8 ro_indexed8:$addr))]>;
1078 def LDRSBXro : Load8RO<0b00, 0, 0b10, GPR64, "ldrsb",
1079 [(set GPR64:$Rt, (sextloadi8 ro_indexed8:$addr))]>;
1081 // Load sign-extended word
1082 def LDRSWro : Load32RO<0b10, 0, 0b10, GPR64, "ldrsw",
1083 [(set GPR64:$Rt, (sextloadi32 ro_indexed32:$addr))]>;
1086 def PRFMro : PrefetchRO<0b11, 0, 0b10, "prfm",
1087 [(ARM64Prefetch imm:$Rt, ro_indexed64:$addr)]>;
1090 def : Pat<(i64 (zextloadi8 ro_indexed8:$addr)),
1091 (SUBREG_TO_REG (i64 0), (LDRBBro ro_indexed8:$addr), sub_32)>;
1092 def : Pat<(i64 (zextloadi16 ro_indexed16:$addr)),
1093 (SUBREG_TO_REG (i64 0), (LDRHHro ro_indexed16:$addr), sub_32)>;
1095 // zextloadi1 -> zextloadi8
1096 def : Pat<(i32 (zextloadi1 ro_indexed8:$addr)), (LDRBBro ro_indexed8:$addr)>;
1097 def : Pat<(i64 (zextloadi1 ro_indexed8:$addr)),
1098 (SUBREG_TO_REG (i64 0), (LDRBBro ro_indexed8:$addr), sub_32)>;
1100 // extload -> zextload
1101 def : Pat<(i32 (extloadi16 ro_indexed16:$addr)), (LDRHHro ro_indexed16:$addr)>;
1102 def : Pat<(i32 (extloadi8 ro_indexed8:$addr)), (LDRBBro ro_indexed8:$addr)>;
1103 def : Pat<(i32 (extloadi1 ro_indexed8:$addr)), (LDRBBro ro_indexed8:$addr)>;
1104 def : Pat<(i64 (extloadi32 ro_indexed32:$addr)),
1105 (SUBREG_TO_REG (i64 0), (LDRWro ro_indexed32:$addr), sub_32)>;
1106 def : Pat<(i64 (extloadi16 ro_indexed16:$addr)),
1107 (SUBREG_TO_REG (i64 0), (LDRHHro ro_indexed16:$addr), sub_32)>;
1108 def : Pat<(i64 (extloadi8 ro_indexed8:$addr)),
1109 (SUBREG_TO_REG (i64 0), (LDRBBro ro_indexed8:$addr), sub_32)>;
1110 def : Pat<(i64 (extloadi1 ro_indexed8:$addr)),
1111 (SUBREG_TO_REG (i64 0), (LDRBBro ro_indexed8:$addr), sub_32)>;
1113 } // AddedComplexity = 10
1116 // (unsigned immediate)
1118 def LDRXui : LoadUI<0b11, 0, 0b01, GPR64, am_indexed64, "ldr",
1119 [(set GPR64:$Rt, (load am_indexed64:$addr))]>;
1120 def LDRWui : LoadUI<0b10, 0, 0b01, GPR32, am_indexed32, "ldr",
1121 [(set GPR32:$Rt, (load am_indexed32:$addr))]>;
1122 def LDRBui : LoadUI<0b00, 1, 0b01, FPR8, am_indexed8, "ldr",
1123 [(set FPR8:$Rt, (load am_indexed8:$addr))]>;
1124 def LDRHui : LoadUI<0b01, 1, 0b01, FPR16, am_indexed16, "ldr",
1125 [(set FPR16:$Rt, (load am_indexed16:$addr))]>;
1126 def LDRSui : LoadUI<0b10, 1, 0b01, FPR32, am_indexed32, "ldr",
1127 [(set (f32 FPR32:$Rt), (load am_indexed32:$addr))]>;
1128 def LDRDui : LoadUI<0b11, 1, 0b01, FPR64, am_indexed64, "ldr",
1129 [(set (f64 FPR64:$Rt), (load am_indexed64:$addr))]>;
1130 def LDRQui : LoadUI<0b00, 1, 0b11, FPR128, am_indexed128, "ldr",
1131 [(set (f128 FPR128:$Rt), (load am_indexed128:$addr))]>;
1133 // For regular load, we do not have any alignment requirement.
1134 // Thus, it is safe to directly map the vector loads with interesting
1135 // addressing modes.
1136 // FIXME: We could do the same for bitconvert to floating point vectors.
1137 def : Pat <(v8i8 (scalar_to_vector (i32 (extloadi8 am_indexed8:$addr)))),
1138 (INSERT_SUBREG (v8i8 (IMPLICIT_DEF)),
1139 (LDRBui am_indexed8:$addr), bsub)>;
1140 def : Pat <(v16i8 (scalar_to_vector (i32 (extloadi8 am_indexed8:$addr)))),
1141 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
1142 (LDRBui am_indexed8:$addr), bsub)>;
1143 def : Pat <(v4i16 (scalar_to_vector (i32 (extloadi16 am_indexed16:$addr)))),
1144 (INSERT_SUBREG (v4i16 (IMPLICIT_DEF)),
1145 (LDRHui am_indexed16:$addr), hsub)>;
1146 def : Pat <(v8i16 (scalar_to_vector (i32 (extloadi16 am_indexed16:$addr)))),
1147 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
1148 (LDRHui am_indexed16:$addr), hsub)>;
1149 def : Pat <(v2i32 (scalar_to_vector (i32 (load am_indexed32:$addr)))),
1150 (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)),
1151 (LDRSui am_indexed32:$addr), ssub)>;
1152 def : Pat <(v4i32 (scalar_to_vector (i32 (load am_indexed32:$addr)))),
1153 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
1154 (LDRSui am_indexed32:$addr), ssub)>;
1155 def : Pat <(v1i64 (scalar_to_vector (i64 (load am_indexed64:$addr)))),
1156 (LDRDui am_indexed64:$addr)>;
1157 def : Pat <(v2i64 (scalar_to_vector (i64 (load am_indexed64:$addr)))),
1158 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
1159 (LDRDui am_indexed64:$addr), dsub)>;
1161 // Match all load 64 bits width whose type is compatible with FPR64
1162 def : Pat<(v2f32 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1163 def : Pat<(v1f64 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1164 def : Pat<(v8i8 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1165 def : Pat<(v4i16 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1166 def : Pat<(v2i32 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1167 def : Pat<(v1i64 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1169 // Match all load 128 bits width whose type is compatible with FPR128
1170 def : Pat<(v4f32 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1171 def : Pat<(v2f64 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1172 def : Pat<(v16i8 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1173 def : Pat<(v8i16 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1174 def : Pat<(v4i32 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1175 def : Pat<(v2i64 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1176 def : Pat<(f128 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1178 def LDRHHui : LoadUI<0b01, 0, 0b01, GPR32, am_indexed16, "ldrh",
1179 [(set GPR32:$Rt, (zextloadi16 am_indexed16:$addr))]>;
1180 def LDRBBui : LoadUI<0b00, 0, 0b01, GPR32, am_indexed8, "ldrb",
1181 [(set GPR32:$Rt, (zextloadi8 am_indexed8:$addr))]>;
1183 def : Pat<(i64 (zextloadi8 am_indexed8:$addr)),
1184 (SUBREG_TO_REG (i64 0), (LDRBBui am_indexed8:$addr), sub_32)>;
1185 def : Pat<(i64 (zextloadi16 am_indexed16:$addr)),
1186 (SUBREG_TO_REG (i64 0), (LDRHHui am_indexed16:$addr), sub_32)>;
1188 // zextloadi1 -> zextloadi8
1189 def : Pat<(i32 (zextloadi1 am_indexed8:$addr)), (LDRBBui am_indexed8:$addr)>;
1190 def : Pat<(i64 (zextloadi1 am_indexed8:$addr)),
1191 (SUBREG_TO_REG (i64 0), (LDRBBui am_indexed8:$addr), sub_32)>;
1193 // extload -> zextload
1194 def : Pat<(i32 (extloadi16 am_indexed16:$addr)), (LDRHHui am_indexed16:$addr)>;
1195 def : Pat<(i32 (extloadi8 am_indexed8:$addr)), (LDRBBui am_indexed8:$addr)>;
1196 def : Pat<(i32 (extloadi1 am_indexed8:$addr)), (LDRBBui am_indexed8:$addr)>;
1197 def : Pat<(i64 (extloadi32 am_indexed32:$addr)),
1198 (SUBREG_TO_REG (i64 0), (LDRWui am_indexed32:$addr), sub_32)>;
1199 def : Pat<(i64 (extloadi16 am_indexed16:$addr)),
1200 (SUBREG_TO_REG (i64 0), (LDRHHui am_indexed16:$addr), sub_32)>;
1201 def : Pat<(i64 (extloadi8 am_indexed8:$addr)),
1202 (SUBREG_TO_REG (i64 0), (LDRBBui am_indexed8:$addr), sub_32)>;
1203 def : Pat<(i64 (extloadi1 am_indexed8:$addr)),
1204 (SUBREG_TO_REG (i64 0), (LDRBBui am_indexed8:$addr), sub_32)>;
1206 // load sign-extended half-word
1207 def LDRSHWui : LoadUI<0b01, 0, 0b11, GPR32, am_indexed16, "ldrsh",
1208 [(set GPR32:$Rt, (sextloadi16 am_indexed16:$addr))]>;
1209 def LDRSHXui : LoadUI<0b01, 0, 0b10, GPR64, am_indexed16, "ldrsh",
1210 [(set GPR64:$Rt, (sextloadi16 am_indexed16:$addr))]>;
1212 // load sign-extended byte
1213 def LDRSBWui : LoadUI<0b00, 0, 0b11, GPR32, am_indexed8, "ldrsb",
1214 [(set GPR32:$Rt, (sextloadi8 am_indexed8:$addr))]>;
1215 def LDRSBXui : LoadUI<0b00, 0, 0b10, GPR64, am_indexed8, "ldrsb",
1216 [(set GPR64:$Rt, (sextloadi8 am_indexed8:$addr))]>;
1218 // load sign-extended word
1219 def LDRSWui : LoadUI<0b10, 0, 0b10, GPR64, am_indexed32, "ldrsw",
1220 [(set GPR64:$Rt, (sextloadi32 am_indexed32:$addr))]>;
1222 // load zero-extended word
1223 def : Pat<(i64 (zextloadi32 am_indexed32:$addr)),
1224 (SUBREG_TO_REG (i64 0), (LDRWui am_indexed32:$addr), sub_32)>;
1227 def PRFMui : PrefetchUI<0b11, 0, 0b10, "prfm",
1228 [(ARM64Prefetch imm:$Rt, am_indexed64:$addr)]>;
1232 def LDRWl : LoadLiteral<0b00, 0, GPR32, "ldr">;
1233 def LDRXl : LoadLiteral<0b01, 0, GPR64, "ldr">;
1234 def LDRSl : LoadLiteral<0b00, 1, FPR32, "ldr">;
1235 def LDRDl : LoadLiteral<0b01, 1, FPR64, "ldr">;
1236 def LDRQl : LoadLiteral<0b10, 1, FPR128, "ldr">;
1238 // load sign-extended word
1239 def LDRSWl : LoadLiteral<0b10, 0, GPR64, "ldrsw">;
1242 def PRFMl : PrefetchLiteral<0b11, 0, "prfm", []>;
1243 // [(ARM64Prefetch imm:$Rt, tglobaladdr:$label)]>;
1246 // (unscaled immediate)
1247 def LDURXi : LoadUnscaled<0b11, 0, 0b01, GPR64, am_unscaled64, "ldur",
1248 [(set GPR64:$Rt, (load am_unscaled64:$addr))]>;
1249 def LDURWi : LoadUnscaled<0b10, 0, 0b01, GPR32, am_unscaled32, "ldur",
1250 [(set GPR32:$Rt, (load am_unscaled32:$addr))]>;
1251 def LDURBi : LoadUnscaled<0b00, 1, 0b01, FPR8, am_unscaled8, "ldur",
1252 [(set FPR8:$Rt, (load am_unscaled8:$addr))]>;
1253 def LDURHi : LoadUnscaled<0b01, 1, 0b01, FPR16, am_unscaled16, "ldur",
1254 [(set FPR16:$Rt, (load am_unscaled16:$addr))]>;
1255 def LDURSi : LoadUnscaled<0b10, 1, 0b01, FPR32, am_unscaled32, "ldur",
1256 [(set (f32 FPR32:$Rt), (load am_unscaled32:$addr))]>;
1257 def LDURDi : LoadUnscaled<0b11, 1, 0b01, FPR64, am_unscaled64, "ldur",
1258 [(set (f64 FPR64:$Rt), (load am_unscaled64:$addr))]>;
1259 def LDURQi : LoadUnscaled<0b00, 1, 0b11, FPR128, am_unscaled128, "ldur",
1260 [(set (v2f64 FPR128:$Rt), (load am_unscaled128:$addr))]>;
1263 : LoadUnscaled<0b01, 0, 0b01, GPR32, am_unscaled16, "ldurh",
1264 [(set GPR32:$Rt, (zextloadi16 am_unscaled16:$addr))]>;
1266 : LoadUnscaled<0b00, 0, 0b01, GPR32, am_unscaled8, "ldurb",
1267 [(set GPR32:$Rt, (zextloadi8 am_unscaled8:$addr))]>;
1269 // Match all load 64 bits width whose type is compatible with FPR64
1270 def : Pat<(v2f32 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1271 def : Pat<(v1f64 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1272 def : Pat<(v8i8 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1273 def : Pat<(v4i16 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1274 def : Pat<(v2i32 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1275 def : Pat<(v1i64 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1277 // Match all load 128 bits width whose type is compatible with FPR128
1278 def : Pat<(v4f32 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1279 def : Pat<(v2f64 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1280 def : Pat<(v16i8 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1281 def : Pat<(v8i16 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1282 def : Pat<(v4i32 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1283 def : Pat<(v2i64 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1284 def : Pat<(f128 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1287 def : Pat<(i32 (extloadi16 am_unscaled16:$addr)), (LDURHHi am_unscaled16:$addr)>;
1288 def : Pat<(i32 (extloadi8 am_unscaled8:$addr)), (LDURBBi am_unscaled8:$addr)>;
1289 def : Pat<(i32 (extloadi1 am_unscaled8:$addr)), (LDURBBi am_unscaled8:$addr)>;
1290 def : Pat<(i64 (extloadi32 am_unscaled32:$addr)),
1291 (SUBREG_TO_REG (i64 0), (LDURWi am_unscaled32:$addr), sub_32)>;
1292 def : Pat<(i64 (extloadi16 am_unscaled16:$addr)),
1293 (SUBREG_TO_REG (i64 0), (LDURHHi am_unscaled16:$addr), sub_32)>;
1294 def : Pat<(i64 (extloadi8 am_unscaled8:$addr)),
1295 (SUBREG_TO_REG (i64 0), (LDURBBi am_unscaled8:$addr), sub_32)>;
1296 def : Pat<(i64 (extloadi1 am_unscaled8:$addr)),
1297 (SUBREG_TO_REG (i64 0), (LDURBBi am_unscaled8:$addr), sub_32)>;
1299 def : Pat<(i32 (zextloadi16 am_unscaled16:$addr)),
1300 (LDURHHi am_unscaled16:$addr)>;
1301 def : Pat<(i32 (zextloadi8 am_unscaled8:$addr)),
1302 (LDURBBi am_unscaled8:$addr)>;
1303 def : Pat<(i32 (zextloadi1 am_unscaled8:$addr)),
1304 (LDURBBi am_unscaled8:$addr)>;
1305 def : Pat<(i64 (zextloadi32 am_unscaled32:$addr)),
1306 (SUBREG_TO_REG (i64 0), (LDURWi am_unscaled32:$addr), sub_32)>;
1307 def : Pat<(i64 (zextloadi16 am_unscaled16:$addr)),
1308 (SUBREG_TO_REG (i64 0), (LDURHHi am_unscaled16:$addr), sub_32)>;
1309 def : Pat<(i64 (zextloadi8 am_unscaled8:$addr)),
1310 (SUBREG_TO_REG (i64 0), (LDURBBi am_unscaled8:$addr), sub_32)>;
1311 def : Pat<(i64 (zextloadi1 am_unscaled8:$addr)),
1312 (SUBREG_TO_REG (i64 0), (LDURBBi am_unscaled8:$addr), sub_32)>;
1316 // LDR mnemonics fall back to LDUR for negative or unaligned offsets.
1318 // Define new assembler match classes as we want to only match these when
1319 // the don't otherwise match the scaled addressing mode for LDR/STR. Don't
1320 // associate a DiagnosticType either, as we want the diagnostic for the
1321 // canonical form (the scaled operand) to take precedence.
1322 def MemoryUnscaledFB8Operand : AsmOperandClass {
1323 let Name = "MemoryUnscaledFB8";
1324 let RenderMethod = "addMemoryUnscaledOperands";
1326 def MemoryUnscaledFB16Operand : AsmOperandClass {
1327 let Name = "MemoryUnscaledFB16";
1328 let RenderMethod = "addMemoryUnscaledOperands";
1330 def MemoryUnscaledFB32Operand : AsmOperandClass {
1331 let Name = "MemoryUnscaledFB32";
1332 let RenderMethod = "addMemoryUnscaledOperands";
1334 def MemoryUnscaledFB64Operand : AsmOperandClass {
1335 let Name = "MemoryUnscaledFB64";
1336 let RenderMethod = "addMemoryUnscaledOperands";
1338 def MemoryUnscaledFB128Operand : AsmOperandClass {
1339 let Name = "MemoryUnscaledFB128";
1340 let RenderMethod = "addMemoryUnscaledOperands";
1342 def am_unscaled_fb8 : Operand<i64> {
1343 let ParserMatchClass = MemoryUnscaledFB8Operand;
1344 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
1346 def am_unscaled_fb16 : Operand<i64> {
1347 let ParserMatchClass = MemoryUnscaledFB16Operand;
1348 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
1350 def am_unscaled_fb32 : Operand<i64> {
1351 let ParserMatchClass = MemoryUnscaledFB32Operand;
1352 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
1354 def am_unscaled_fb64 : Operand<i64> {
1355 let ParserMatchClass = MemoryUnscaledFB64Operand;
1356 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
1358 def am_unscaled_fb128 : Operand<i64> {
1359 let ParserMatchClass = MemoryUnscaledFB128Operand;
1360 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
1362 def : InstAlias<"ldr $Rt, $addr", (LDURXi GPR64:$Rt, am_unscaled_fb64:$addr)>;
1363 def : InstAlias<"ldr $Rt, $addr", (LDURWi GPR32:$Rt, am_unscaled_fb32:$addr)>;
1364 def : InstAlias<"ldr $Rt, $addr", (LDURBi FPR8:$Rt, am_unscaled_fb8:$addr)>;
1365 def : InstAlias<"ldr $Rt, $addr", (LDURHi FPR16:$Rt, am_unscaled_fb16:$addr)>;
1366 def : InstAlias<"ldr $Rt, $addr", (LDURSi FPR32:$Rt, am_unscaled_fb32:$addr)>;
1367 def : InstAlias<"ldr $Rt, $addr", (LDURDi FPR64:$Rt, am_unscaled_fb64:$addr)>;
1368 def : InstAlias<"ldr $Rt, $addr", (LDURQi FPR128:$Rt, am_unscaled_fb128:$addr)>;
1371 def : Pat<(i64 (zextloadi8 am_unscaled8:$addr)),
1372 (SUBREG_TO_REG (i64 0), (LDURBBi am_unscaled8:$addr), sub_32)>;
1373 def : Pat<(i64 (zextloadi16 am_unscaled16:$addr)),
1374 (SUBREG_TO_REG (i64 0), (LDURHHi am_unscaled16:$addr), sub_32)>;
1376 // load sign-extended half-word
1378 : LoadUnscaled<0b01, 0, 0b11, GPR32, am_unscaled16, "ldursh",
1379 [(set GPR32:$Rt, (sextloadi16 am_unscaled16:$addr))]>;
1381 : LoadUnscaled<0b01, 0, 0b10, GPR64, am_unscaled16, "ldursh",
1382 [(set GPR64:$Rt, (sextloadi16 am_unscaled16:$addr))]>;
1384 // load sign-extended byte
1386 : LoadUnscaled<0b00, 0, 0b11, GPR32, am_unscaled8, "ldursb",
1387 [(set GPR32:$Rt, (sextloadi8 am_unscaled8:$addr))]>;
1389 : LoadUnscaled<0b00, 0, 0b10, GPR64, am_unscaled8, "ldursb",
1390 [(set GPR64:$Rt, (sextloadi8 am_unscaled8:$addr))]>;
1392 // load sign-extended word
1394 : LoadUnscaled<0b10, 0, 0b10, GPR64, am_unscaled32, "ldursw",
1395 [(set GPR64:$Rt, (sextloadi32 am_unscaled32:$addr))]>;
1397 // zero and sign extending aliases from generic LDR* mnemonics to LDUR*.
1398 def : InstAlias<"ldrb $Rt, $addr", (LDURBBi GPR32:$Rt, am_unscaled_fb8:$addr)>;
1399 def : InstAlias<"ldrh $Rt, $addr", (LDURHHi GPR32:$Rt, am_unscaled_fb16:$addr)>;
1400 def : InstAlias<"ldrsb $Rt, $addr", (LDURSBWi GPR32:$Rt, am_unscaled_fb8:$addr)>;
1401 def : InstAlias<"ldrsb $Rt, $addr", (LDURSBXi GPR64:$Rt, am_unscaled_fb8:$addr)>;
1402 def : InstAlias<"ldrsh $Rt, $addr", (LDURSHWi GPR32:$Rt, am_unscaled_fb16:$addr)>;
1403 def : InstAlias<"ldrsh $Rt, $addr", (LDURSHXi GPR64:$Rt, am_unscaled_fb16:$addr)>;
1404 def : InstAlias<"ldrsw $Rt, $addr", (LDURSWi GPR64:$Rt, am_unscaled_fb32:$addr)>;
1407 def PRFUMi : PrefetchUnscaled<0b11, 0, 0b10, "prfum",
1408 [(ARM64Prefetch imm:$Rt, am_unscaled64:$addr)]>;
1411 // (unscaled immediate, unprivileged)
1412 def LDTRXi : LoadUnprivileged<0b11, 0, 0b01, GPR64, "ldtr">;
1413 def LDTRWi : LoadUnprivileged<0b10, 0, 0b01, GPR32, "ldtr">;
1415 def LDTRHi : LoadUnprivileged<0b01, 0, 0b01, GPR32, "ldtrh">;
1416 def LDTRBi : LoadUnprivileged<0b00, 0, 0b01, GPR32, "ldtrb">;
1418 // load sign-extended half-word
1419 def LDTRSHWi : LoadUnprivileged<0b01, 0, 0b11, GPR32, "ldtrsh">;
1420 def LDTRSHXi : LoadUnprivileged<0b01, 0, 0b10, GPR64, "ldtrsh">;
1422 // load sign-extended byte
1423 def LDTRSBWi : LoadUnprivileged<0b00, 0, 0b11, GPR32, "ldtrsb">;
1424 def LDTRSBXi : LoadUnprivileged<0b00, 0, 0b10, GPR64, "ldtrsb">;
1426 // load sign-extended word
1427 def LDTRSWi : LoadUnprivileged<0b10, 0, 0b10, GPR64, "ldtrsw">;
1430 // (immediate pre-indexed)
1431 def LDRWpre : LoadPreIdx<0b10, 0, 0b01, GPR32, "ldr">;
1432 def LDRXpre : LoadPreIdx<0b11, 0, 0b01, GPR64, "ldr">;
1433 def LDRBpre : LoadPreIdx<0b00, 1, 0b01, FPR8, "ldr">;
1434 def LDRHpre : LoadPreIdx<0b01, 1, 0b01, FPR16, "ldr">;
1435 def LDRSpre : LoadPreIdx<0b10, 1, 0b01, FPR32, "ldr">;
1436 def LDRDpre : LoadPreIdx<0b11, 1, 0b01, FPR64, "ldr">;
1437 def LDRQpre : LoadPreIdx<0b00, 1, 0b11, FPR128, "ldr">;
1439 // load sign-extended half-word
1440 def LDRSHWpre : LoadPreIdx<0b01, 0, 0b11, GPR32, "ldrsh">;
1441 def LDRSHXpre : LoadPreIdx<0b01, 0, 0b10, GPR64, "ldrsh">;
1443 // load sign-extended byte
1444 def LDRSBWpre : LoadPreIdx<0b00, 0, 0b11, GPR32, "ldrsb">;
1445 def LDRSBXpre : LoadPreIdx<0b00, 0, 0b10, GPR64, "ldrsb">;
1447 // load zero-extended byte
1448 def LDRBBpre : LoadPreIdx<0b00, 0, 0b01, GPR32, "ldrb">;
1449 def LDRHHpre : LoadPreIdx<0b01, 0, 0b01, GPR32, "ldrh">;
1451 // load sign-extended word
1452 def LDRSWpre : LoadPreIdx<0b10, 0, 0b10, GPR64, "ldrsw">;
1454 // ISel pseudos and patterns. See expanded comment on LoadPreIdxPseudo.
1455 def LDRDpre_isel : LoadPreIdxPseudo<FPR64>;
1456 def LDRSpre_isel : LoadPreIdxPseudo<FPR32>;
1457 def LDRXpre_isel : LoadPreIdxPseudo<GPR64>;
1458 def LDRWpre_isel : LoadPreIdxPseudo<GPR32>;
1459 def LDRHHpre_isel : LoadPreIdxPseudo<GPR32>;
1460 def LDRBBpre_isel : LoadPreIdxPseudo<GPR32>;
1462 def LDRSWpre_isel : LoadPreIdxPseudo<GPR64>;
1463 def LDRSHWpre_isel : LoadPreIdxPseudo<GPR32>;
1464 def LDRSHXpre_isel : LoadPreIdxPseudo<GPR64>;
1465 def LDRSBWpre_isel : LoadPreIdxPseudo<GPR32>;
1466 def LDRSBXpre_isel : LoadPreIdxPseudo<GPR64>;
1469 // (immediate post-indexed)
1470 def LDRWpost : LoadPostIdx<0b10, 0, 0b01, GPR32, "ldr">;
1471 def LDRXpost : LoadPostIdx<0b11, 0, 0b01, GPR64, "ldr">;
1472 def LDRBpost : LoadPostIdx<0b00, 1, 0b01, FPR8, "ldr">;
1473 def LDRHpost : LoadPostIdx<0b01, 1, 0b01, FPR16, "ldr">;
1474 def LDRSpost : LoadPostIdx<0b10, 1, 0b01, FPR32, "ldr">;
1475 def LDRDpost : LoadPostIdx<0b11, 1, 0b01, FPR64, "ldr">;
1476 def LDRQpost : LoadPostIdx<0b00, 1, 0b11, FPR128, "ldr">;
1478 // load sign-extended half-word
1479 def LDRSHWpost : LoadPostIdx<0b01, 0, 0b11, GPR32, "ldrsh">;
1480 def LDRSHXpost : LoadPostIdx<0b01, 0, 0b10, GPR64, "ldrsh">;
1482 // load sign-extended byte
1483 def LDRSBWpost : LoadPostIdx<0b00, 0, 0b11, GPR32, "ldrsb">;
1484 def LDRSBXpost : LoadPostIdx<0b00, 0, 0b10, GPR64, "ldrsb">;
1486 // load zero-extended byte
1487 def LDRBBpost : LoadPostIdx<0b00, 0, 0b01, GPR32, "ldrb">;
1488 def LDRHHpost : LoadPostIdx<0b01, 0, 0b01, GPR32, "ldrh">;
1490 // load sign-extended word
1491 def LDRSWpost : LoadPostIdx<0b10, 0, 0b10, GPR64, "ldrsw">;
1493 // ISel pseudos and patterns. See expanded comment on LoadPostIdxPseudo.
1494 def LDRDpost_isel : LoadPostIdxPseudo<FPR64>;
1495 def LDRSpost_isel : LoadPostIdxPseudo<FPR32>;
1496 def LDRXpost_isel : LoadPostIdxPseudo<GPR64>;
1497 def LDRWpost_isel : LoadPostIdxPseudo<GPR32>;
1498 def LDRHHpost_isel : LoadPostIdxPseudo<GPR32>;
1499 def LDRBBpost_isel : LoadPostIdxPseudo<GPR32>;
1501 def LDRSWpost_isel : LoadPostIdxPseudo<GPR64>;
1502 def LDRSHWpost_isel : LoadPostIdxPseudo<GPR32>;
1503 def LDRSHXpost_isel : LoadPostIdxPseudo<GPR64>;
1504 def LDRSBWpost_isel : LoadPostIdxPseudo<GPR32>;
1505 def LDRSBXpost_isel : LoadPostIdxPseudo<GPR64>;
1507 //===----------------------------------------------------------------------===//
1508 // Store instructions.
1509 //===----------------------------------------------------------------------===//
1511 // Pair (indexed, offset)
1512 // FIXME: Use dedicated range-checked addressing mode operand here.
1513 def STPWi : StorePairOffset<0b00, 0, GPR32, am_indexed32simm7, "stp">;
1514 def STPXi : StorePairOffset<0b10, 0, GPR64, am_indexed64simm7, "stp">;
1515 def STPSi : StorePairOffset<0b00, 1, FPR32, am_indexed32simm7, "stp">;
1516 def STPDi : StorePairOffset<0b01, 1, FPR64, am_indexed64simm7, "stp">;
1517 def STPQi : StorePairOffset<0b10, 1, FPR128, am_indexed128simm7, "stp">;
1519 // Pair (pre-indexed)
1520 def STPWpre : StorePairPreIdx<0b00, 0, GPR32, am_indexed32simm7_wb, "stp">;
1521 def STPXpre : StorePairPreIdx<0b10, 0, GPR64, am_indexed64simm7_wb, "stp">;
1522 def STPSpre : StorePairPreIdx<0b00, 1, FPR32, am_indexed32simm7_wb, "stp">;
1523 def STPDpre : StorePairPreIdx<0b01, 1, FPR64, am_indexed64simm7_wb, "stp">;
1524 def STPQpre : StorePairPreIdx<0b10, 1, FPR128, am_indexed128simm7_wb, "stp">;
1526 // Pair (pre-indexed)
1527 def STPWpost : StorePairPostIdx<0b00, 0, GPR32, simm7s4, "stp">;
1528 def STPXpost : StorePairPostIdx<0b10, 0, GPR64, simm7s8, "stp">;
1529 def STPSpost : StorePairPostIdx<0b00, 1, FPR32, simm7s4, "stp">;
1530 def STPDpost : StorePairPostIdx<0b01, 1, FPR64, simm7s8, "stp">;
1531 def STPQpost : StorePairPostIdx<0b10, 1, FPR128, simm7s16, "stp">;
1533 // Pair (no allocate)
1534 def STNPWi : StorePairNoAlloc<0b00, 0, GPR32, am_indexed32simm7, "stnp">;
1535 def STNPXi : StorePairNoAlloc<0b10, 0, GPR64, am_indexed64simm7, "stnp">;
1536 def STNPSi : StorePairNoAlloc<0b00, 1, FPR32, am_indexed32simm7, "stnp">;
1537 def STNPDi : StorePairNoAlloc<0b01, 1, FPR64, am_indexed64simm7, "stnp">;
1538 def STNPQi : StorePairNoAlloc<0b10, 1, FPR128, am_indexed128simm7, "stnp">;
1541 // (Register offset)
1543 let AddedComplexity = 10 in {
1546 def STRHHro : Store16RO<0b01, 0, 0b00, GPR32, "strh",
1547 [(truncstorei16 GPR32:$Rt, ro_indexed16:$addr)]>;
1548 def STRBBro : Store8RO<0b00, 0, 0b00, GPR32, "strb",
1549 [(truncstorei8 GPR32:$Rt, ro_indexed8:$addr)]>;
1550 def STRWro : Store32RO<0b10, 0, 0b00, GPR32, "str",
1551 [(store GPR32:$Rt, ro_indexed32:$addr)]>;
1552 def STRXro : Store64RO<0b11, 0, 0b00, GPR64, "str",
1553 [(store GPR64:$Rt, ro_indexed64:$addr)]>;
1556 def : Pat<(truncstorei8 GPR64:$Rt, ro_indexed8:$addr),
1557 (STRBBro (EXTRACT_SUBREG GPR64:$Rt, sub_32), ro_indexed8:$addr)>;
1558 def : Pat<(truncstorei16 GPR64:$Rt, ro_indexed16:$addr),
1559 (STRHHro (EXTRACT_SUBREG GPR64:$Rt, sub_32), ro_indexed16:$addr)>;
1560 def : Pat<(truncstorei32 GPR64:$Rt, ro_indexed32:$addr),
1561 (STRWro (EXTRACT_SUBREG GPR64:$Rt, sub_32), ro_indexed32:$addr)>;
1565 def STRBro : Store8RO<0b00, 1, 0b00, FPR8, "str",
1566 [(store FPR8:$Rt, ro_indexed8:$addr)]>;
1567 def STRHro : Store16RO<0b01, 1, 0b00, FPR16, "str",
1568 [(store FPR16:$Rt, ro_indexed16:$addr)]>;
1569 def STRSro : Store32RO<0b10, 1, 0b00, FPR32, "str",
1570 [(store (f32 FPR32:$Rt), ro_indexed32:$addr)]>;
1571 def STRDro : Store64RO<0b11, 1, 0b00, FPR64, "str",
1572 [(store (f64 FPR64:$Rt), ro_indexed64:$addr)]>;
1573 def STRQro : Store128RO<0b00, 1, 0b10, FPR128, "str", []> {
1577 // Match all store 64 bits width whose type is compatible with FPR64
1578 def : Pat<(store (v2f32 FPR64:$Rn), ro_indexed64:$addr),
1579 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1580 def : Pat<(store (v1f64 FPR64:$Rn), ro_indexed64:$addr),
1581 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1582 def : Pat<(store (v8i8 FPR64:$Rn), ro_indexed64:$addr),
1583 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1584 def : Pat<(store (v4i16 FPR64:$Rn), ro_indexed64:$addr),
1585 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1586 def : Pat<(store (v2i32 FPR64:$Rn), ro_indexed64:$addr),
1587 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1588 def : Pat<(store (v1i64 FPR64:$Rn), ro_indexed64:$addr),
1589 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1591 // Match all store 128 bits width whose type is compatible with FPR128
1592 def : Pat<(store (v4f32 FPR128:$Rn), ro_indexed128:$addr),
1593 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1594 def : Pat<(store (v2f64 FPR128:$Rn), ro_indexed128:$addr),
1595 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1596 def : Pat<(store (v16i8 FPR128:$Rn), ro_indexed128:$addr),
1597 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1598 def : Pat<(store (v8i16 FPR128:$Rn), ro_indexed128:$addr),
1599 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1600 def : Pat<(store (v4i32 FPR128:$Rn), ro_indexed128:$addr),
1601 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1602 def : Pat<(store (v2i64 FPR128:$Rn), ro_indexed128:$addr),
1603 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1604 def : Pat<(store (f128 FPR128:$Rn), ro_indexed128:$addr),
1605 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1608 // (unsigned immediate)
1609 def STRXui : StoreUI<0b11, 0, 0b00, GPR64, am_indexed64, "str",
1610 [(store GPR64:$Rt, am_indexed64:$addr)]>;
1611 def STRWui : StoreUI<0b10, 0, 0b00, GPR32, am_indexed32, "str",
1612 [(store GPR32:$Rt, am_indexed32:$addr)]>;
1613 def STRBui : StoreUI<0b00, 1, 0b00, FPR8, am_indexed8, "str",
1614 [(store FPR8:$Rt, am_indexed8:$addr)]>;
1615 def STRHui : StoreUI<0b01, 1, 0b00, FPR16, am_indexed16, "str",
1616 [(store FPR16:$Rt, am_indexed16:$addr)]>;
1617 def STRSui : StoreUI<0b10, 1, 0b00, FPR32, am_indexed32, "str",
1618 [(store (f32 FPR32:$Rt), am_indexed32:$addr)]>;
1619 def STRDui : StoreUI<0b11, 1, 0b00, FPR64, am_indexed64, "str",
1620 [(store (f64 FPR64:$Rt), am_indexed64:$addr)]>;
1621 def STRQui : StoreUI<0b00, 1, 0b10, FPR128, am_indexed128, "str", []> {
1625 // Match all store 64 bits width whose type is compatible with FPR64
1626 def : Pat<(store (v2f32 FPR64:$Rn), am_indexed64:$addr),
1627 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1628 def : Pat<(store (v1f64 FPR64:$Rn), am_indexed64:$addr),
1629 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1630 def : Pat<(store (v8i8 FPR64:$Rn), am_indexed64:$addr),
1631 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1632 def : Pat<(store (v4i16 FPR64:$Rn), am_indexed64:$addr),
1633 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1634 def : Pat<(store (v2i32 FPR64:$Rn), am_indexed64:$addr),
1635 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1636 def : Pat<(store (v1i64 FPR64:$Rn), am_indexed64:$addr),
1637 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1639 // Match all store 128 bits width whose type is compatible with FPR128
1640 def : Pat<(store (v4f32 FPR128:$Rn), am_indexed128:$addr),
1641 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1642 def : Pat<(store (v2f64 FPR128:$Rn), am_indexed128:$addr),
1643 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1644 def : Pat<(store (v16i8 FPR128:$Rn), am_indexed128:$addr),
1645 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1646 def : Pat<(store (v8i16 FPR128:$Rn), am_indexed128:$addr),
1647 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1648 def : Pat<(store (v4i32 FPR128:$Rn), am_indexed128:$addr),
1649 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1650 def : Pat<(store (v2i64 FPR128:$Rn), am_indexed128:$addr),
1651 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1652 def : Pat<(store (f128 FPR128:$Rn), am_indexed128:$addr),
1653 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1655 def STRHHui : StoreUI<0b01, 0, 0b00, GPR32, am_indexed16, "strh",
1656 [(truncstorei16 GPR32:$Rt, am_indexed16:$addr)]>;
1657 def STRBBui : StoreUI<0b00, 0, 0b00, GPR32, am_indexed8, "strb",
1658 [(truncstorei8 GPR32:$Rt, am_indexed8:$addr)]>;
1661 def : Pat<(truncstorei32 GPR64:$Rt, am_indexed32:$addr),
1662 (STRWui (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_indexed32:$addr)>;
1663 def : Pat<(truncstorei16 GPR64:$Rt, am_indexed16:$addr),
1664 (STRHHui (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_indexed16:$addr)>;
1665 def : Pat<(truncstorei8 GPR64:$Rt, am_indexed8:$addr),
1666 (STRBBui (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_indexed8:$addr)>;
1668 } // AddedComplexity = 10
1671 // (unscaled immediate)
1672 def STURXi : StoreUnscaled<0b11, 0, 0b00, GPR64, am_unscaled64, "stur",
1673 [(store GPR64:$Rt, am_unscaled64:$addr)]>;
1674 def STURWi : StoreUnscaled<0b10, 0, 0b00, GPR32, am_unscaled32, "stur",
1675 [(store GPR32:$Rt, am_unscaled32:$addr)]>;
1676 def STURBi : StoreUnscaled<0b00, 1, 0b00, FPR8, am_unscaled8, "stur",
1677 [(store FPR8:$Rt, am_unscaled8:$addr)]>;
1678 def STURHi : StoreUnscaled<0b01, 1, 0b00, FPR16, am_unscaled16, "stur",
1679 [(store FPR16:$Rt, am_unscaled16:$addr)]>;
1680 def STURSi : StoreUnscaled<0b10, 1, 0b00, FPR32, am_unscaled32, "stur",
1681 [(store (f32 FPR32:$Rt), am_unscaled32:$addr)]>;
1682 def STURDi : StoreUnscaled<0b11, 1, 0b00, FPR64, am_unscaled64, "stur",
1683 [(store (f64 FPR64:$Rt), am_unscaled64:$addr)]>;
1684 def STURQi : StoreUnscaled<0b00, 1, 0b10, FPR128, am_unscaled128, "stur",
1685 [(store (v2f64 FPR128:$Rt), am_unscaled128:$addr)]>;
1686 def STURHHi : StoreUnscaled<0b01, 0, 0b00, GPR32, am_unscaled16, "sturh",
1687 [(truncstorei16 GPR32:$Rt, am_unscaled16:$addr)]>;
1688 def STURBBi : StoreUnscaled<0b00, 0, 0b00, GPR32, am_unscaled8, "sturb",
1689 [(truncstorei8 GPR32:$Rt, am_unscaled8:$addr)]>;
1691 // Match all store 64 bits width whose type is compatible with FPR64
1692 def : Pat<(store (v2f32 FPR64:$Rn), am_unscaled64:$addr),
1693 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1694 def : Pat<(store (v1f64 FPR64:$Rn), am_unscaled64:$addr),
1695 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1696 def : Pat<(store (v8i8 FPR64:$Rn), am_unscaled64:$addr),
1697 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1698 def : Pat<(store (v4i16 FPR64:$Rn), am_unscaled64:$addr),
1699 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1700 def : Pat<(store (v2i32 FPR64:$Rn), am_unscaled64:$addr),
1701 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1702 def : Pat<(store (v1i64 FPR64:$Rn), am_unscaled64:$addr),
1703 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1705 // Match all store 128 bits width whose type is compatible with FPR128
1706 def : Pat<(store (v4f32 FPR128:$Rn), am_unscaled128:$addr),
1707 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1708 def : Pat<(store (v2f64 FPR128:$Rn), am_unscaled128:$addr),
1709 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1710 def : Pat<(store (v16i8 FPR128:$Rn), am_unscaled128:$addr),
1711 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1712 def : Pat<(store (v8i16 FPR128:$Rn), am_unscaled128:$addr),
1713 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1714 def : Pat<(store (v4i32 FPR128:$Rn), am_unscaled128:$addr),
1715 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1716 def : Pat<(store (v2i64 FPR128:$Rn), am_unscaled128:$addr),
1717 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1718 def : Pat<(store (f128 FPR128:$Rn), am_unscaled128:$addr),
1719 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1721 // unscaled i64 truncating stores
1722 def : Pat<(truncstorei32 GPR64:$Rt, am_unscaled32:$addr),
1723 (STURWi (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_unscaled32:$addr)>;
1724 def : Pat<(truncstorei16 GPR64:$Rt, am_unscaled16:$addr),
1725 (STURHHi (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_unscaled16:$addr)>;
1726 def : Pat<(truncstorei8 GPR64:$Rt, am_unscaled8:$addr),
1727 (STURBBi (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_unscaled8:$addr)>;
1730 // STR mnemonics fall back to STUR for negative or unaligned offsets.
1731 def : InstAlias<"str $Rt, $addr", (STURXi GPR64:$Rt, am_unscaled_fb64:$addr)>;
1732 def : InstAlias<"str $Rt, $addr", (STURWi GPR32:$Rt, am_unscaled_fb32:$addr)>;
1733 def : InstAlias<"str $Rt, $addr", (STURBi FPR8:$Rt, am_unscaled_fb8:$addr)>;
1734 def : InstAlias<"str $Rt, $addr", (STURHi FPR16:$Rt, am_unscaled_fb16:$addr)>;
1735 def : InstAlias<"str $Rt, $addr", (STURSi FPR32:$Rt, am_unscaled_fb32:$addr)>;
1736 def : InstAlias<"str $Rt, $addr", (STURDi FPR64:$Rt, am_unscaled_fb64:$addr)>;
1737 def : InstAlias<"str $Rt, $addr", (STURQi FPR128:$Rt, am_unscaled_fb128:$addr)>;
1739 def : InstAlias<"strb $Rt, $addr", (STURBBi GPR32:$Rt, am_unscaled_fb8:$addr)>;
1740 def : InstAlias<"strh $Rt, $addr", (STURHHi GPR32:$Rt, am_unscaled_fb16:$addr)>;
1743 // (unscaled immediate, unprivileged)
1744 def STTRWi : StoreUnprivileged<0b10, 0, 0b00, GPR32, "sttr">;
1745 def STTRXi : StoreUnprivileged<0b11, 0, 0b00, GPR64, "sttr">;
1747 def STTRHi : StoreUnprivileged<0b01, 0, 0b00, GPR32, "sttrh">;
1748 def STTRBi : StoreUnprivileged<0b00, 0, 0b00, GPR32, "sttrb">;
1751 // (immediate pre-indexed)
1752 def STRWpre : StorePreIdx<0b10, 0, 0b00, GPR32, "str">;
1753 def STRXpre : StorePreIdx<0b11, 0, 0b00, GPR64, "str">;
1754 def STRBpre : StorePreIdx<0b00, 1, 0b00, FPR8, "str">;
1755 def STRHpre : StorePreIdx<0b01, 1, 0b00, FPR16, "str">;
1756 def STRSpre : StorePreIdx<0b10, 1, 0b00, FPR32, "str">;
1757 def STRDpre : StorePreIdx<0b11, 1, 0b00, FPR64, "str">;
1758 def STRQpre : StorePreIdx<0b00, 1, 0b10, FPR128, "str">;
1760 def STRBBpre : StorePreIdx<0b00, 0, 0b00, GPR32, "strb">;
1761 def STRHHpre : StorePreIdx<0b01, 0, 0b00, GPR32, "strh">;
1763 // ISel pseudos and patterns. See expanded comment on StorePreIdxPseudo.
1764 defm STRDpre : StorePreIdxPseudo<FPR64, f64, pre_store>;
1765 defm STRSpre : StorePreIdxPseudo<FPR32, f32, pre_store>;
1766 defm STRXpre : StorePreIdxPseudo<GPR64, i64, pre_store>;
1767 defm STRWpre : StorePreIdxPseudo<GPR32, i32, pre_store>;
1768 defm STRHHpre : StorePreIdxPseudo<GPR32, i32, pre_truncsti16>;
1769 defm STRBBpre : StorePreIdxPseudo<GPR32, i32, pre_truncsti8>;
1771 def : Pat<(pre_truncsti32 GPR64:$Rt, am_noindex:$addr, simm9:$off),
1772 (STRWpre_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
1774 def : Pat<(pre_truncsti16 GPR64:$Rt, am_noindex:$addr, simm9:$off),
1775 (STRHHpre_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
1777 def : Pat<(pre_truncsti8 GPR64:$Rt, am_noindex:$addr, simm9:$off),
1778 (STRBBpre_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
1782 // (immediate post-indexed)
1783 def STRWpost : StorePostIdx<0b10, 0, 0b00, GPR32, "str">;
1784 def STRXpost : StorePostIdx<0b11, 0, 0b00, GPR64, "str">;
1785 def STRBpost : StorePostIdx<0b00, 1, 0b00, FPR8, "str">;
1786 def STRHpost : StorePostIdx<0b01, 1, 0b00, FPR16, "str">;
1787 def STRSpost : StorePostIdx<0b10, 1, 0b00, FPR32, "str">;
1788 def STRDpost : StorePostIdx<0b11, 1, 0b00, FPR64, "str">;
1789 def STRQpost : StorePostIdx<0b00, 1, 0b10, FPR128, "str">;
1791 def STRBBpost : StorePostIdx<0b00, 0, 0b00, GPR32, "strb">;
1792 def STRHHpost : StorePostIdx<0b01, 0, 0b00, GPR32, "strh">;
1794 // ISel pseudos and patterns. See expanded comment on StorePostIdxPseudo.
1795 defm STRDpost : StorePostIdxPseudo<FPR64, f64, post_store, STRDpost>;
1796 defm STRSpost : StorePostIdxPseudo<FPR32, f32, post_store, STRSpost>;
1797 defm STRXpost : StorePostIdxPseudo<GPR64, i64, post_store, STRXpost>;
1798 defm STRWpost : StorePostIdxPseudo<GPR32, i32, post_store, STRWpost>;
1799 defm STRHHpost : StorePostIdxPseudo<GPR32, i32, post_truncsti16, STRHHpost>;
1800 defm STRBBpost : StorePostIdxPseudo<GPR32, i32, post_truncsti8, STRBBpost>;
1802 def : Pat<(post_truncsti32 GPR64:$Rt, am_noindex:$addr, simm9:$off),
1803 (STRWpost_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
1805 def : Pat<(post_truncsti16 GPR64:$Rt, am_noindex:$addr, simm9:$off),
1806 (STRHHpost_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
1808 def : Pat<(post_truncsti8 GPR64:$Rt, am_noindex:$addr, simm9:$off),
1809 (STRBBpost_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
1813 //===----------------------------------------------------------------------===//
1814 // Load/store exclusive instructions.
1815 //===----------------------------------------------------------------------===//
1817 def LDARW : LoadAcquire <0b10, 1, 1, 0, 1, GPR32, "ldar">;
1818 def LDARX : LoadAcquire <0b11, 1, 1, 0, 1, GPR64, "ldar">;
1819 def LDARB : LoadAcquire <0b00, 1, 1, 0, 1, GPR32, "ldarb">;
1820 def LDARH : LoadAcquire <0b01, 1, 1, 0, 1, GPR32, "ldarh">;
1822 def LDAXRW : LoadExclusive <0b10, 0, 1, 0, 1, GPR32, "ldaxr">;
1823 def LDAXRX : LoadExclusive <0b11, 0, 1, 0, 1, GPR64, "ldaxr">;
1824 def LDAXRB : LoadExclusive <0b00, 0, 1, 0, 1, GPR32, "ldaxrb">;
1825 def LDAXRH : LoadExclusive <0b01, 0, 1, 0, 1, GPR32, "ldaxrh">;
1827 def LDXRW : LoadExclusive <0b10, 0, 1, 0, 0, GPR32, "ldxr">;
1828 def LDXRX : LoadExclusive <0b11, 0, 1, 0, 0, GPR64, "ldxr">;
1829 def LDXRB : LoadExclusive <0b00, 0, 1, 0, 0, GPR32, "ldxrb">;
1830 def LDXRH : LoadExclusive <0b01, 0, 1, 0, 0, GPR32, "ldxrh">;
1832 def STLRW : StoreRelease <0b10, 1, 0, 0, 1, GPR32, "stlr">;
1833 def STLRX : StoreRelease <0b11, 1, 0, 0, 1, GPR64, "stlr">;
1834 def STLRB : StoreRelease <0b00, 1, 0, 0, 1, GPR32, "stlrb">;
1835 def STLRH : StoreRelease <0b01, 1, 0, 0, 1, GPR32, "stlrh">;
1837 def STLXRW : StoreExclusive<0b10, 0, 0, 0, 1, GPR32, "stlxr">;
1838 def STLXRX : StoreExclusive<0b11, 0, 0, 0, 1, GPR64, "stlxr">;
1839 def STLXRB : StoreExclusive<0b00, 0, 0, 0, 1, GPR32, "stlxrb">;
1840 def STLXRH : StoreExclusive<0b01, 0, 0, 0, 1, GPR32, "stlxrh">;
1842 def STXRW : StoreExclusive<0b10, 0, 0, 0, 0, GPR32, "stxr">;
1843 def STXRX : StoreExclusive<0b11, 0, 0, 0, 0, GPR64, "stxr">;
1844 def STXRB : StoreExclusive<0b00, 0, 0, 0, 0, GPR32, "stxrb">;
1845 def STXRH : StoreExclusive<0b01, 0, 0, 0, 0, GPR32, "stxrh">;
1847 def LDAXPW : LoadExclusivePair<0b10, 0, 1, 1, 1, GPR32, "ldaxp">;
1848 def LDAXPX : LoadExclusivePair<0b11, 0, 1, 1, 1, GPR64, "ldaxp">;
1850 def LDXPW : LoadExclusivePair<0b10, 0, 1, 1, 0, GPR32, "ldxp">;
1851 def LDXPX : LoadExclusivePair<0b11, 0, 1, 1, 0, GPR64, "ldxp">;
1853 def STLXPW : StoreExclusivePair<0b10, 0, 0, 1, 1, GPR32, "stlxp">;
1854 def STLXPX : StoreExclusivePair<0b11, 0, 0, 1, 1, GPR64, "stlxp">;
1856 def STXPW : StoreExclusivePair<0b10, 0, 0, 1, 0, GPR32, "stxp">;
1857 def STXPX : StoreExclusivePair<0b11, 0, 0, 1, 0, GPR64, "stxp">;
1859 //===----------------------------------------------------------------------===//
1860 // Scaled floating point to integer conversion instructions.
1861 //===----------------------------------------------------------------------===//
1863 defm FCVTAS : FPToIntegerUnscaled<0b00, 0b100, "fcvtas", int_arm64_neon_fcvtas>;
1864 defm FCVTAU : FPToIntegerUnscaled<0b00, 0b101, "fcvtau", int_arm64_neon_fcvtau>;
1865 defm FCVTMS : FPToIntegerUnscaled<0b10, 0b000, "fcvtms", int_arm64_neon_fcvtms>;
1866 defm FCVTMU : FPToIntegerUnscaled<0b10, 0b001, "fcvtmu", int_arm64_neon_fcvtmu>;
1867 defm FCVTNS : FPToIntegerUnscaled<0b00, 0b000, "fcvtns", int_arm64_neon_fcvtns>;
1868 defm FCVTNU : FPToIntegerUnscaled<0b00, 0b001, "fcvtnu", int_arm64_neon_fcvtnu>;
1869 defm FCVTPS : FPToIntegerUnscaled<0b01, 0b000, "fcvtps", int_arm64_neon_fcvtps>;
1870 defm FCVTPU : FPToIntegerUnscaled<0b01, 0b001, "fcvtpu", int_arm64_neon_fcvtpu>;
1871 defm FCVTZS : FPToIntegerUnscaled<0b11, 0b000, "fcvtzs", fp_to_sint>;
1872 defm FCVTZU : FPToIntegerUnscaled<0b11, 0b001, "fcvtzu", fp_to_uint>;
1873 defm FCVTZS : FPToIntegerScaled<0b11, 0b000, "fcvtzs", fp_to_sint>;
1874 defm FCVTZU : FPToIntegerScaled<0b11, 0b001, "fcvtzu", fp_to_uint>;
1875 let isCodeGenOnly = 1 in {
1876 defm FCVTZS_Int : FPToIntegerUnscaled<0b11, 0b000, "fcvtzs", int_arm64_neon_fcvtzs>;
1877 defm FCVTZU_Int : FPToIntegerUnscaled<0b11, 0b001, "fcvtzu", int_arm64_neon_fcvtzu>;
1878 defm FCVTZS_Int : FPToIntegerScaled<0b11, 0b000, "fcvtzs", int_arm64_neon_fcvtzs>;
1879 defm FCVTZU_Int : FPToIntegerScaled<0b11, 0b001, "fcvtzu", int_arm64_neon_fcvtzu>;
1882 //===----------------------------------------------------------------------===//
1883 // Scaled integer to floating point conversion instructions.
1884 //===----------------------------------------------------------------------===//
1886 defm SCVTF : IntegerToFP<0, "scvtf", sint_to_fp>;
1887 defm UCVTF : IntegerToFP<1, "ucvtf", uint_to_fp>;
1889 //===----------------------------------------------------------------------===//
1890 // Unscaled integer to floating point conversion instruction.
1891 //===----------------------------------------------------------------------===//
1893 defm FMOV : UnscaledConversion<"fmov">;
1895 def : Pat<(f32 (fpimm0)), (FMOVWSr WZR)>, Requires<[NoZCZ]>;
1896 def : Pat<(f64 (fpimm0)), (FMOVXDr XZR)>, Requires<[NoZCZ]>;
1898 def : Pat<(v8i8 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1899 def : Pat<(v4i16 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1900 def : Pat<(v2i32 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1901 def : Pat<(v1i64 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1902 def : Pat<(v2f32 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1903 def : Pat<(v1f64 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1904 def : Pat<(v1i64 (scalar_to_vector GPR64:$Xn)),
1905 (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1906 def : Pat<(v1f64 (scalar_to_vector GPR64:$Xn)),
1907 (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1908 def : Pat<(v1f64 (scalar_to_vector (f64 FPR64:$Xn))), (v1f64 FPR64:$Xn)>;
1910 def : Pat<(i64 (bitconvert (v8i8 V64:$Vn))),
1911 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
1912 def : Pat<(i64 (bitconvert (v4i16 V64:$Vn))),
1913 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
1914 def : Pat<(i64 (bitconvert (v2i32 V64:$Vn))),
1915 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
1916 def : Pat<(i64 (bitconvert (v1i64 V64:$Vn))),
1917 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
1918 def : Pat<(i64 (bitconvert (v2f32 V64:$Vn))),
1919 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
1920 def : Pat<(i64 (bitconvert (v1f64 V64:$Vn))),
1921 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
1923 def : Pat<(f32 (bitconvert (i32 GPR32:$Xn))),
1924 (COPY_TO_REGCLASS GPR32:$Xn, FPR32)>;
1925 def : Pat<(i32 (bitconvert (f32 FPR32:$Xn))),
1926 (COPY_TO_REGCLASS FPR32:$Xn, GPR32)>;
1927 def : Pat<(f64 (bitconvert (i64 GPR64:$Xn))),
1928 (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1929 def : Pat<(i64 (bitconvert (f64 FPR64:$Xn))),
1930 (COPY_TO_REGCLASS FPR64:$Xn, GPR64)>;
1932 //===----------------------------------------------------------------------===//
1933 // Floating point conversion instruction.
1934 //===----------------------------------------------------------------------===//
1936 defm FCVT : FPConversion<"fcvt">;
1938 def : Pat<(f32_to_f16 FPR32:$Rn),
1939 (i32 (COPY_TO_REGCLASS
1940 (f32 (SUBREG_TO_REG (i32 0), (FCVTHSr FPR32:$Rn), hsub)),
1943 def FCVTSHpseudo : Pseudo<(outs FPR32:$Rd), (ins FPR32:$Rn),
1944 [(set (f32 FPR32:$Rd), (f16_to_f32 i32:$Rn))]>;
1946 //===----------------------------------------------------------------------===//
1947 // Floating point single operand instructions.
1948 //===----------------------------------------------------------------------===//
1950 defm FABS : SingleOperandFPData<0b0001, "fabs", fabs>;
1951 defm FMOV : SingleOperandFPData<0b0000, "fmov">;
1952 defm FNEG : SingleOperandFPData<0b0010, "fneg", fneg>;
1953 defm FRINTA : SingleOperandFPData<0b1100, "frinta", frnd>;
1954 defm FRINTI : SingleOperandFPData<0b1111, "frinti", fnearbyint>;
1955 defm FRINTM : SingleOperandFPData<0b1010, "frintm", ffloor>;
1956 defm FRINTN : SingleOperandFPData<0b1000, "frintn", int_arm64_neon_frintn>;
1957 defm FRINTP : SingleOperandFPData<0b1001, "frintp", fceil>;
1959 def : Pat<(v1f64 (int_arm64_neon_frintn (v1f64 FPR64:$Rn))),
1960 (FRINTNDr FPR64:$Rn)>;
1962 // FRINTX is inserted to set the flags as required by FENV_ACCESS ON behavior
1963 // in the C spec. Setting hasSideEffects ensures it is not DCE'd.
1964 // <rdar://problem/13715968>
1965 // TODO: We should really model the FPSR flags correctly. This is really ugly.
1966 let hasSideEffects = 1 in {
1967 defm FRINTX : SingleOperandFPData<0b1110, "frintx", frint>;
1970 defm FRINTZ : SingleOperandFPData<0b1011, "frintz", ftrunc>;
1972 let SchedRW = [WriteFDiv] in {
1973 defm FSQRT : SingleOperandFPData<0b0011, "fsqrt", fsqrt>;
1976 //===----------------------------------------------------------------------===//
1977 // Floating point two operand instructions.
1978 //===----------------------------------------------------------------------===//
1980 defm FADD : TwoOperandFPData<0b0010, "fadd", fadd>;
1981 let SchedRW = [WriteFDiv] in {
1982 defm FDIV : TwoOperandFPData<0b0001, "fdiv", fdiv>;
1984 defm FMAXNM : TwoOperandFPData<0b0110, "fmaxnm", int_arm64_neon_fmaxnm>;
1985 defm FMAX : TwoOperandFPData<0b0100, "fmax", ARM64fmax>;
1986 defm FMINNM : TwoOperandFPData<0b0111, "fminnm", int_arm64_neon_fminnm>;
1987 defm FMIN : TwoOperandFPData<0b0101, "fmin", ARM64fmin>;
1988 let SchedRW = [WriteFMul] in {
1989 defm FMUL : TwoOperandFPData<0b0000, "fmul", fmul>;
1990 defm FNMUL : TwoOperandFPDataNeg<0b1000, "fnmul", fmul>;
1992 defm FSUB : TwoOperandFPData<0b0011, "fsub", fsub>;
1994 def : Pat<(v1f64 (ARM64fmax (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
1995 (FMAXDrr FPR64:$Rn, FPR64:$Rm)>;
1996 def : Pat<(v1f64 (ARM64fmin (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
1997 (FMINDrr FPR64:$Rn, FPR64:$Rm)>;
1998 def : Pat<(v1f64 (int_arm64_neon_fmaxnm (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
1999 (FMAXNMDrr FPR64:$Rn, FPR64:$Rm)>;
2000 def : Pat<(v1f64 (int_arm64_neon_fminnm (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2001 (FMINNMDrr FPR64:$Rn, FPR64:$Rm)>;
2003 //===----------------------------------------------------------------------===//
2004 // Floating point three operand instructions.
2005 //===----------------------------------------------------------------------===//
2007 defm FMADD : ThreeOperandFPData<0, 0, "fmadd", fma>;
2008 defm FMSUB : ThreeOperandFPData<0, 1, "fmsub",
2009 TriOpFrag<(fma node:$LHS, (fneg node:$MHS), node:$RHS)> >;
2010 defm FNMADD : ThreeOperandFPData<1, 0, "fnmadd",
2011 TriOpFrag<(fneg (fma node:$LHS, node:$MHS, node:$RHS))> >;
2012 defm FNMSUB : ThreeOperandFPData<1, 1, "fnmsub",
2013 TriOpFrag<(fma node:$LHS, node:$MHS, (fneg node:$RHS))> >;
2015 // The following def pats catch the case where the LHS of an FMA is negated.
2016 // The TriOpFrag above catches the case where the middle operand is negated.
2018 // N.b. FMSUB etc have the accumulator at the *end* of (outs), unlike
2019 // the NEON variant.
2020 def : Pat<(f32 (fma (fneg FPR32:$Rn), FPR32:$Rm, FPR32:$Ra)),
2021 (FMSUBSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
2023 def : Pat<(f64 (fma (fneg FPR64:$Rn), FPR64:$Rm, FPR64:$Ra)),
2024 (FMSUBDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
2026 //===----------------------------------------------------------------------===//
2027 // Floating point comparison instructions.
2028 //===----------------------------------------------------------------------===//
2030 defm FCMPE : FPComparison<1, "fcmpe">;
2031 defm FCMP : FPComparison<0, "fcmp", ARM64fcmp>;
2033 //===----------------------------------------------------------------------===//
2034 // Floating point conditional comparison instructions.
2035 //===----------------------------------------------------------------------===//
2037 defm FCCMPE : FPCondComparison<1, "fccmpe">;
2038 defm FCCMP : FPCondComparison<0, "fccmp">;
2040 //===----------------------------------------------------------------------===//
2041 // Floating point conditional select instruction.
2042 //===----------------------------------------------------------------------===//
2044 defm FCSEL : FPCondSelect<"fcsel">;
2046 // CSEL instructions providing f128 types need to be handled by a
2047 // pseudo-instruction since the eventual code will need to introduce basic
2048 // blocks and control flow.
2049 def F128CSEL : Pseudo<(outs FPR128:$Rd),
2050 (ins FPR128:$Rn, FPR128:$Rm, ccode:$cond),
2051 [(set (f128 FPR128:$Rd),
2052 (ARM64csel FPR128:$Rn, FPR128:$Rm,
2053 (i32 imm:$cond), CPSR))]> {
2055 let usesCustomInserter = 1;
2059 //===----------------------------------------------------------------------===//
2060 // Floating point immediate move.
2061 //===----------------------------------------------------------------------===//
2063 let isReMaterializable = 1 in {
2064 defm FMOV : FPMoveImmediate<"fmov">;
2067 //===----------------------------------------------------------------------===//
2068 // Advanced SIMD two vector instructions.
2069 //===----------------------------------------------------------------------===//
2071 defm ABS : SIMDTwoVectorBHSD<0, 0b01011, "abs", int_arm64_neon_abs>;
2072 defm CLS : SIMDTwoVectorBHS<0, 0b00100, "cls", int_arm64_neon_cls>;
2073 defm CLZ : SIMDTwoVectorBHS<1, 0b00100, "clz", ctlz>;
2074 defm CMEQ : SIMDCmpTwoVector<0, 0b01001, "cmeq", ARM64cmeqz>;
2075 defm CMGE : SIMDCmpTwoVector<1, 0b01000, "cmge", ARM64cmgez>;
2076 defm CMGT : SIMDCmpTwoVector<0, 0b01000, "cmgt", ARM64cmgtz>;
2077 defm CMLE : SIMDCmpTwoVector<1, 0b01001, "cmle", ARM64cmlez>;
2078 defm CMLT : SIMDCmpTwoVector<0, 0b01010, "cmlt", ARM64cmltz>;
2079 defm CNT : SIMDTwoVectorB<0, 0b00, 0b00101, "cnt", ctpop>;
2080 defm FABS : SIMDTwoVectorFP<0, 1, 0b01111, "fabs", fabs>;
2082 defm FCMEQ : SIMDFPCmpTwoVector<0, 1, 0b01101, "fcmeq", ARM64fcmeqz>;
2083 defm FCMGE : SIMDFPCmpTwoVector<1, 1, 0b01100, "fcmge", ARM64fcmgez>;
2084 defm FCMGT : SIMDFPCmpTwoVector<0, 1, 0b01100, "fcmgt", ARM64fcmgtz>;
2085 defm FCMLE : SIMDFPCmpTwoVector<1, 1, 0b01101, "fcmle", ARM64fcmlez>;
2086 defm FCMLT : SIMDFPCmpTwoVector<0, 1, 0b01110, "fcmlt", ARM64fcmltz>;
2087 defm FCVTAS : SIMDTwoVectorFPToInt<0,0,0b11100, "fcvtas",int_arm64_neon_fcvtas>;
2088 defm FCVTAU : SIMDTwoVectorFPToInt<1,0,0b11100, "fcvtau",int_arm64_neon_fcvtau>;
2089 defm FCVTL : SIMDFPWidenTwoVector<0, 0, 0b10111, "fcvtl">;
2090 def : Pat<(v4f32 (int_arm64_neon_vcvthf2fp (v4i16 V64:$Rn))),
2091 (FCVTLv4i16 V64:$Rn)>;
2092 def : Pat<(v4f32 (int_arm64_neon_vcvthf2fp (extract_subvector (v8i16 V128:$Rn),
2094 (FCVTLv8i16 V128:$Rn)>;
2095 def : Pat<(v2f64 (fextend (v2f32 V64:$Rn))), (FCVTLv2i32 V64:$Rn)>;
2096 def : Pat<(v2f64 (fextend (v2f32 (extract_subvector (v4f32 V128:$Rn),
2098 (FCVTLv4i32 V128:$Rn)>;
2100 defm FCVTMS : SIMDTwoVectorFPToInt<0,0,0b11011, "fcvtms",int_arm64_neon_fcvtms>;
2101 defm FCVTMU : SIMDTwoVectorFPToInt<1,0,0b11011, "fcvtmu",int_arm64_neon_fcvtmu>;
2102 defm FCVTNS : SIMDTwoVectorFPToInt<0,0,0b11010, "fcvtns",int_arm64_neon_fcvtns>;
2103 defm FCVTNU : SIMDTwoVectorFPToInt<1,0,0b11010, "fcvtnu",int_arm64_neon_fcvtnu>;
2104 defm FCVTN : SIMDFPNarrowTwoVector<0, 0, 0b10110, "fcvtn">;
2105 def : Pat<(v4i16 (int_arm64_neon_vcvtfp2hf (v4f32 V128:$Rn))),
2106 (FCVTNv4i16 V128:$Rn)>;
2107 def : Pat<(concat_vectors V64:$Rd,
2108 (v4i16 (int_arm64_neon_vcvtfp2hf (v4f32 V128:$Rn)))),
2109 (FCVTNv8i16 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
2110 def : Pat<(v2f32 (fround (v2f64 V128:$Rn))), (FCVTNv2i32 V128:$Rn)>;
2111 def : Pat<(concat_vectors V64:$Rd, (v2f32 (fround (v2f64 V128:$Rn)))),
2112 (FCVTNv4i32 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
2113 defm FCVTPS : SIMDTwoVectorFPToInt<0,1,0b11010, "fcvtps",int_arm64_neon_fcvtps>;
2114 defm FCVTPU : SIMDTwoVectorFPToInt<1,1,0b11010, "fcvtpu",int_arm64_neon_fcvtpu>;
2115 defm FCVTXN : SIMDFPInexactCvtTwoVector<1, 0, 0b10110, "fcvtxn",
2116 int_arm64_neon_fcvtxn>;
2117 defm FCVTZS : SIMDTwoVectorFPToInt<0, 1, 0b11011, "fcvtzs", fp_to_sint>;
2118 defm FCVTZU : SIMDTwoVectorFPToInt<1, 1, 0b11011, "fcvtzu", fp_to_uint>;
2119 let isCodeGenOnly = 1 in {
2120 defm FCVTZS_Int : SIMDTwoVectorFPToInt<0, 1, 0b11011, "fcvtzs",
2121 int_arm64_neon_fcvtzs>;
2122 defm FCVTZU_Int : SIMDTwoVectorFPToInt<1, 1, 0b11011, "fcvtzu",
2123 int_arm64_neon_fcvtzu>;
2125 defm FNEG : SIMDTwoVectorFP<1, 1, 0b01111, "fneg", fneg>;
2126 defm FRECPE : SIMDTwoVectorFP<0, 1, 0b11101, "frecpe", int_arm64_neon_frecpe>;
2127 defm FRINTA : SIMDTwoVectorFP<1, 0, 0b11000, "frinta", frnd>;
2128 defm FRINTI : SIMDTwoVectorFP<1, 1, 0b11001, "frinti", fnearbyint>;
2129 defm FRINTM : SIMDTwoVectorFP<0, 0, 0b11001, "frintm", ffloor>;
2130 defm FRINTN : SIMDTwoVectorFP<0, 0, 0b11000, "frintn", int_arm64_neon_frintn>;
2131 defm FRINTP : SIMDTwoVectorFP<0, 1, 0b11000, "frintp", fceil>;
2132 defm FRINTX : SIMDTwoVectorFP<1, 0, 0b11001, "frintx", frint>;
2133 defm FRINTZ : SIMDTwoVectorFP<0, 1, 0b11001, "frintz", ftrunc>;
2134 defm FRSQRTE: SIMDTwoVectorFP<1, 1, 0b11101, "frsqrte", int_arm64_neon_frsqrte>;
2135 defm FSQRT : SIMDTwoVectorFP<1, 1, 0b11111, "fsqrt", fsqrt>;
2136 defm NEG : SIMDTwoVectorBHSD<1, 0b01011, "neg",
2137 UnOpFrag<(sub immAllZerosV, node:$LHS)> >;
2138 defm NOT : SIMDTwoVectorB<1, 0b00, 0b00101, "not", vnot>;
2139 // Aliases for MVN -> NOT.
2140 def : InstAlias<"mvn.8b $Vd, $Vn", (NOTv8i8 V64:$Vd, V64:$Vn)>;
2141 def : InstAlias<"mvn.16b $Vd, $Vn", (NOTv16i8 V128:$Vd, V128:$Vn)>;
2142 def : InstAlias<"mvn $Vd.8b, $Vn.8b", (NOTv8i8 V64:$Vd, V64:$Vn)>;
2143 def : InstAlias<"mvn $Vd.16b, $Vn.16b", (NOTv16i8 V128:$Vd, V128:$Vn)>;
2145 def : Pat<(ARM64neg (v8i8 V64:$Rn)), (NEGv8i8 V64:$Rn)>;
2146 def : Pat<(ARM64neg (v16i8 V128:$Rn)), (NEGv16i8 V128:$Rn)>;
2147 def : Pat<(ARM64neg (v4i16 V64:$Rn)), (NEGv4i16 V64:$Rn)>;
2148 def : Pat<(ARM64neg (v8i16 V128:$Rn)), (NEGv8i16 V128:$Rn)>;
2149 def : Pat<(ARM64neg (v2i32 V64:$Rn)), (NEGv2i32 V64:$Rn)>;
2150 def : Pat<(ARM64neg (v4i32 V128:$Rn)), (NEGv4i32 V128:$Rn)>;
2151 def : Pat<(ARM64neg (v2i64 V128:$Rn)), (NEGv2i64 V128:$Rn)>;
2153 def : Pat<(ARM64not (v8i8 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2154 def : Pat<(ARM64not (v16i8 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2155 def : Pat<(ARM64not (v4i16 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2156 def : Pat<(ARM64not (v8i16 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2157 def : Pat<(ARM64not (v2i32 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2158 def : Pat<(ARM64not (v1i64 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2159 def : Pat<(ARM64not (v4i32 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2160 def : Pat<(ARM64not (v2i64 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2162 def : Pat<(vnot (v4i16 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2163 def : Pat<(vnot (v8i16 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2164 def : Pat<(vnot (v2i32 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2165 def : Pat<(vnot (v4i32 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2166 def : Pat<(vnot (v2i64 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2168 defm RBIT : SIMDTwoVectorB<1, 0b01, 0b00101, "rbit", int_arm64_neon_rbit>;
2169 defm REV16 : SIMDTwoVectorB<0, 0b00, 0b00001, "rev16", ARM64rev16>;
2170 defm REV32 : SIMDTwoVectorBH<1, 0b00000, "rev32", ARM64rev32>;
2171 defm REV64 : SIMDTwoVectorBHS<0, 0b00000, "rev64", ARM64rev64>;
2172 defm SADALP : SIMDLongTwoVectorTied<0, 0b00110, "sadalp",
2173 BinOpFrag<(add node:$LHS, (int_arm64_neon_saddlp node:$RHS))> >;
2174 defm SADDLP : SIMDLongTwoVector<0, 0b00010, "saddlp", int_arm64_neon_saddlp>;
2175 defm SCVTF : SIMDTwoVectorIntToFP<0, 0, 0b11101, "scvtf", sint_to_fp>;
2176 defm SHLL : SIMDVectorLShiftLongBySizeBHS;
2177 defm SQABS : SIMDTwoVectorBHSD<0, 0b00111, "sqabs", int_arm64_neon_sqabs>;
2178 defm SQNEG : SIMDTwoVectorBHSD<1, 0b00111, "sqneg", int_arm64_neon_sqneg>;
2179 defm SQXTN : SIMDMixedTwoVector<0, 0b10100, "sqxtn", int_arm64_neon_sqxtn>;
2180 defm SQXTUN : SIMDMixedTwoVector<1, 0b10010, "sqxtun", int_arm64_neon_sqxtun>;
2181 defm SUQADD : SIMDTwoVectorBHSDTied<0, 0b00011, "suqadd",int_arm64_neon_suqadd>;
2182 defm UADALP : SIMDLongTwoVectorTied<1, 0b00110, "uadalp",
2183 BinOpFrag<(add node:$LHS, (int_arm64_neon_uaddlp node:$RHS))> >;
2184 defm UADDLP : SIMDLongTwoVector<1, 0b00010, "uaddlp",
2185 int_arm64_neon_uaddlp>;
2186 defm UCVTF : SIMDTwoVectorIntToFP<1, 0, 0b11101, "ucvtf", uint_to_fp>;
2187 defm UQXTN : SIMDMixedTwoVector<1, 0b10100, "uqxtn", int_arm64_neon_uqxtn>;
2188 defm URECPE : SIMDTwoVectorS<0, 1, 0b11100, "urecpe", int_arm64_neon_urecpe>;
2189 defm URSQRTE: SIMDTwoVectorS<1, 1, 0b11100, "ursqrte", int_arm64_neon_ursqrte>;
2190 defm USQADD : SIMDTwoVectorBHSDTied<1, 0b00011, "usqadd",int_arm64_neon_usqadd>;
2191 defm XTN : SIMDMixedTwoVector<0, 0b10010, "xtn", trunc>;
2193 def : Pat<(v2f32 (ARM64rev64 V64:$Rn)), (REV64v2i32 V64:$Rn)>;
2194 def : Pat<(v4f32 (ARM64rev64 V128:$Rn)), (REV64v4i32 V128:$Rn)>;
2196 // Patterns for vector long shift (by element width). These need to match all
2197 // three of zext, sext and anyext so it's easier to pull the patterns out of the
2199 multiclass SIMDVectorLShiftLongBySizeBHSPats<SDPatternOperator ext> {
2200 def : Pat<(ARM64vshl (v8i16 (ext (v8i8 V64:$Rn))), (i32 8)),
2201 (SHLLv8i8 V64:$Rn)>;
2202 def : Pat<(ARM64vshl (v8i16 (ext (extract_high_v16i8 V128:$Rn))), (i32 8)),
2203 (SHLLv16i8 V128:$Rn)>;
2204 def : Pat<(ARM64vshl (v4i32 (ext (v4i16 V64:$Rn))), (i32 16)),
2205 (SHLLv4i16 V64:$Rn)>;
2206 def : Pat<(ARM64vshl (v4i32 (ext (extract_high_v8i16 V128:$Rn))), (i32 16)),
2207 (SHLLv8i16 V128:$Rn)>;
2208 def : Pat<(ARM64vshl (v2i64 (ext (v2i32 V64:$Rn))), (i32 32)),
2209 (SHLLv2i32 V64:$Rn)>;
2210 def : Pat<(ARM64vshl (v2i64 (ext (extract_high_v4i32 V128:$Rn))), (i32 32)),
2211 (SHLLv4i32 V128:$Rn)>;
2214 defm : SIMDVectorLShiftLongBySizeBHSPats<anyext>;
2215 defm : SIMDVectorLShiftLongBySizeBHSPats<zext>;
2216 defm : SIMDVectorLShiftLongBySizeBHSPats<sext>;
2218 //===----------------------------------------------------------------------===//
2219 // Advanced SIMD three vector instructions.
2220 //===----------------------------------------------------------------------===//
2222 defm ADD : SIMDThreeSameVector<0, 0b10000, "add", add>;
2223 defm ADDP : SIMDThreeSameVector<0, 0b10111, "addp", int_arm64_neon_addp>;
2224 defm CMEQ : SIMDThreeSameVector<1, 0b10001, "cmeq", ARM64cmeq>;
2225 defm CMGE : SIMDThreeSameVector<0, 0b00111, "cmge", ARM64cmge>;
2226 defm CMGT : SIMDThreeSameVector<0, 0b00110, "cmgt", ARM64cmgt>;
2227 defm CMHI : SIMDThreeSameVector<1, 0b00110, "cmhi", ARM64cmhi>;
2228 defm CMHS : SIMDThreeSameVector<1, 0b00111, "cmhs", ARM64cmhs>;
2229 defm CMTST : SIMDThreeSameVector<0, 0b10001, "cmtst", ARM64cmtst>;
2230 defm FABD : SIMDThreeSameVectorFP<1,1,0b11010,"fabd", int_arm64_neon_fabd>;
2231 defm FACGE : SIMDThreeSameVectorFPCmp<1,0,0b11101,"facge",int_arm64_neon_facge>;
2232 defm FACGT : SIMDThreeSameVectorFPCmp<1,1,0b11101,"facgt",int_arm64_neon_facgt>;
2233 defm FADDP : SIMDThreeSameVectorFP<1,0,0b11010,"faddp",int_arm64_neon_addp>;
2234 defm FADD : SIMDThreeSameVectorFP<0,0,0b11010,"fadd", fadd>;
2235 defm FCMEQ : SIMDThreeSameVectorFPCmp<0, 0, 0b11100, "fcmeq", ARM64fcmeq>;
2236 defm FCMGE : SIMDThreeSameVectorFPCmp<1, 0, 0b11100, "fcmge", ARM64fcmge>;
2237 defm FCMGT : SIMDThreeSameVectorFPCmp<1, 1, 0b11100, "fcmgt", ARM64fcmgt>;
2238 defm FDIV : SIMDThreeSameVectorFP<1,0,0b11111,"fdiv", fdiv>;
2239 defm FMAXNMP : SIMDThreeSameVectorFP<1,0,0b11000,"fmaxnmp", int_arm64_neon_fmaxnmp>;
2240 defm FMAXNM : SIMDThreeSameVectorFP<0,0,0b11000,"fmaxnm", int_arm64_neon_fmaxnm>;
2241 defm FMAXP : SIMDThreeSameVectorFP<1,0,0b11110,"fmaxp", int_arm64_neon_fmaxp>;
2242 defm FMAX : SIMDThreeSameVectorFP<0,0,0b11110,"fmax", ARM64fmax>;
2243 defm FMINNMP : SIMDThreeSameVectorFP<1,1,0b11000,"fminnmp", int_arm64_neon_fminnmp>;
2244 defm FMINNM : SIMDThreeSameVectorFP<0,1,0b11000,"fminnm", int_arm64_neon_fminnm>;
2245 defm FMINP : SIMDThreeSameVectorFP<1,1,0b11110,"fminp", int_arm64_neon_fminp>;
2246 defm FMIN : SIMDThreeSameVectorFP<0,1,0b11110,"fmin", ARM64fmin>;
2248 // NOTE: The operands of the PatFrag are reordered on FMLA/FMLS because the
2249 // instruction expects the addend first, while the fma intrinsic puts it last.
2250 defm FMLA : SIMDThreeSameVectorFPTied<0, 0, 0b11001, "fmla",
2251 TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)> >;
2252 defm FMLS : SIMDThreeSameVectorFPTied<0, 1, 0b11001, "fmls",
2253 TriOpFrag<(fma node:$MHS, (fneg node:$RHS), node:$LHS)> >;
2255 // The following def pats catch the case where the LHS of an FMA is negated.
2256 // The TriOpFrag above catches the case where the middle operand is negated.
2257 def : Pat<(v2f32 (fma (fneg V64:$Rn), V64:$Rm, V64:$Rd)),
2258 (FMLSv2f32 V64:$Rd, V64:$Rn, V64:$Rm)>;
2260 def : Pat<(v4f32 (fma (fneg V128:$Rn), V128:$Rm, V128:$Rd)),
2261 (FMLSv4f32 V128:$Rd, V128:$Rn, V128:$Rm)>;
2263 def : Pat<(v2f64 (fma (fneg V128:$Rn), V128:$Rm, V128:$Rd)),
2264 (FMLSv2f64 V128:$Rd, V128:$Rn, V128:$Rm)>;
2266 defm FMULX : SIMDThreeSameVectorFP<0,0,0b11011,"fmulx", int_arm64_neon_fmulx>;
2267 defm FMUL : SIMDThreeSameVectorFP<1,0,0b11011,"fmul", fmul>;
2268 defm FRECPS : SIMDThreeSameVectorFP<0,0,0b11111,"frecps", int_arm64_neon_frecps>;
2269 defm FRSQRTS : SIMDThreeSameVectorFP<0,1,0b11111,"frsqrts", int_arm64_neon_frsqrts>;
2270 defm FSUB : SIMDThreeSameVectorFP<0,1,0b11010,"fsub", fsub>;
2271 defm MLA : SIMDThreeSameVectorBHSTied<0, 0b10010, "mla",
2272 TriOpFrag<(add node:$LHS, (mul node:$MHS, node:$RHS))> >;
2273 defm MLS : SIMDThreeSameVectorBHSTied<1, 0b10010, "mls",
2274 TriOpFrag<(sub node:$LHS, (mul node:$MHS, node:$RHS))> >;
2275 defm MUL : SIMDThreeSameVectorBHS<0, 0b10011, "mul", mul>;
2276 defm PMUL : SIMDThreeSameVectorB<1, 0b10011, "pmul", int_arm64_neon_pmul>;
2277 defm SABA : SIMDThreeSameVectorBHSTied<0, 0b01111, "saba",
2278 TriOpFrag<(add node:$LHS, (int_arm64_neon_sabd node:$MHS, node:$RHS))> >;
2279 defm SABD : SIMDThreeSameVectorBHS<0,0b01110,"sabd", int_arm64_neon_sabd>;
2280 defm SHADD : SIMDThreeSameVectorBHS<0,0b00000,"shadd", int_arm64_neon_shadd>;
2281 defm SHSUB : SIMDThreeSameVectorBHS<0,0b00100,"shsub", int_arm64_neon_shsub>;
2282 defm SMAXP : SIMDThreeSameVectorBHS<0,0b10100,"smaxp", int_arm64_neon_smaxp>;
2283 defm SMAX : SIMDThreeSameVectorBHS<0,0b01100,"smax", int_arm64_neon_smax>;
2284 defm SMINP : SIMDThreeSameVectorBHS<0,0b10101,"sminp", int_arm64_neon_sminp>;
2285 defm SMIN : SIMDThreeSameVectorBHS<0,0b01101,"smin", int_arm64_neon_smin>;
2286 defm SQADD : SIMDThreeSameVector<0,0b00001,"sqadd", int_arm64_neon_sqadd>;
2287 defm SQDMULH : SIMDThreeSameVectorHS<0,0b10110,"sqdmulh",int_arm64_neon_sqdmulh>;
2288 defm SQRDMULH : SIMDThreeSameVectorHS<1,0b10110,"sqrdmulh",int_arm64_neon_sqrdmulh>;
2289 defm SQRSHL : SIMDThreeSameVector<0,0b01011,"sqrshl", int_arm64_neon_sqrshl>;
2290 defm SQSHL : SIMDThreeSameVector<0,0b01001,"sqshl", int_arm64_neon_sqshl>;
2291 defm SQSUB : SIMDThreeSameVector<0,0b00101,"sqsub", int_arm64_neon_sqsub>;
2292 defm SRHADD : SIMDThreeSameVectorBHS<0,0b00010,"srhadd",int_arm64_neon_srhadd>;
2293 defm SRSHL : SIMDThreeSameVector<0,0b01010,"srshl", int_arm64_neon_srshl>;
2294 defm SSHL : SIMDThreeSameVector<0,0b01000,"sshl", int_arm64_neon_sshl>;
2295 defm SUB : SIMDThreeSameVector<1,0b10000,"sub", sub>;
2296 defm UABA : SIMDThreeSameVectorBHSTied<1, 0b01111, "uaba",
2297 TriOpFrag<(add node:$LHS, (int_arm64_neon_uabd node:$MHS, node:$RHS))> >;
2298 defm UABD : SIMDThreeSameVectorBHS<1,0b01110,"uabd", int_arm64_neon_uabd>;
2299 defm UHADD : SIMDThreeSameVectorBHS<1,0b00000,"uhadd", int_arm64_neon_uhadd>;
2300 defm UHSUB : SIMDThreeSameVectorBHS<1,0b00100,"uhsub", int_arm64_neon_uhsub>;
2301 defm UMAXP : SIMDThreeSameVectorBHS<1,0b10100,"umaxp", int_arm64_neon_umaxp>;
2302 defm UMAX : SIMDThreeSameVectorBHS<1,0b01100,"umax", int_arm64_neon_umax>;
2303 defm UMINP : SIMDThreeSameVectorBHS<1,0b10101,"uminp", int_arm64_neon_uminp>;
2304 defm UMIN : SIMDThreeSameVectorBHS<1,0b01101,"umin", int_arm64_neon_umin>;
2305 defm UQADD : SIMDThreeSameVector<1,0b00001,"uqadd", int_arm64_neon_uqadd>;
2306 defm UQRSHL : SIMDThreeSameVector<1,0b01011,"uqrshl", int_arm64_neon_uqrshl>;
2307 defm UQSHL : SIMDThreeSameVector<1,0b01001,"uqshl", int_arm64_neon_uqshl>;
2308 defm UQSUB : SIMDThreeSameVector<1,0b00101,"uqsub", int_arm64_neon_uqsub>;
2309 defm URHADD : SIMDThreeSameVectorBHS<1,0b00010,"urhadd", int_arm64_neon_urhadd>;
2310 defm URSHL : SIMDThreeSameVector<1,0b01010,"urshl", int_arm64_neon_urshl>;
2311 defm USHL : SIMDThreeSameVector<1,0b01000,"ushl", int_arm64_neon_ushl>;
2313 defm AND : SIMDLogicalThreeVector<0, 0b00, "and", and>;
2314 defm BIC : SIMDLogicalThreeVector<0, 0b01, "bic",
2315 BinOpFrag<(and node:$LHS, (vnot node:$RHS))> >;
2316 defm BIF : SIMDLogicalThreeVector<1, 0b11, "bif">;
2317 defm BIT : SIMDLogicalThreeVectorTied<1, 0b10, "bit", ARM64bit>;
2318 defm BSL : SIMDLogicalThreeVectorTied<1, 0b01, "bsl",
2319 TriOpFrag<(or (and node:$LHS, node:$MHS), (and (vnot node:$LHS), node:$RHS))>>;
2320 defm EOR : SIMDLogicalThreeVector<1, 0b00, "eor", xor>;
2321 defm ORN : SIMDLogicalThreeVector<0, 0b11, "orn",
2322 BinOpFrag<(or node:$LHS, (vnot node:$RHS))> >;
2323 defm ORR : SIMDLogicalThreeVector<0, 0b10, "orr", or>;
2325 // FIXME: the .16b and .8b variantes should be emitted by the
2326 // AsmWriter. TableGen's AsmWriter-generator doesn't deal with variant syntaxes
2327 // in aliases yet though.
2328 def : InstAlias<"mov{\t$dst.16b, $src.16b|.16b\t$dst, $src}",
2329 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2330 def : InstAlias<"{mov\t$dst.8h, $src.8h|mov.8h\t$dst, $src}",
2331 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2332 def : InstAlias<"{mov\t$dst.4s, $src.4s|mov.4s\t$dst, $src}",
2333 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2334 def : InstAlias<"{mov\t$dst.2d, $src.2d|mov.2d\t$dst, $src}",
2335 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2337 def : InstAlias<"{mov\t$dst.8b, $src.8b|mov.8b\t$dst, $src}",
2338 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2339 def : InstAlias<"{mov\t$dst.4h, $src.4h|mov.4h\t$dst, $src}",
2340 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2341 def : InstAlias<"{mov\t$dst.2s, $src.2s|mov.2s\t$dst, $src}",
2342 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2343 def : InstAlias<"{mov\t$dst.1d, $src.1d|mov.1d\t$dst, $src}",
2344 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2346 def : InstAlias<"{cmls\t$dst.8b, $src1.8b, $src2.8b" #
2347 "|cmls.8b\t$dst, $src1, $src2}",
2348 (CMHSv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2349 def : InstAlias<"{cmls\t$dst.16b, $src1.16b, $src2.16b" #
2350 "|cmls.16b\t$dst, $src1, $src2}",
2351 (CMHSv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2352 def : InstAlias<"{cmls\t$dst.4h, $src1.4h, $src2.4h" #
2353 "|cmls.4h\t$dst, $src1, $src2}",
2354 (CMHSv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2355 def : InstAlias<"{cmls\t$dst.8h, $src1.8h, $src2.8h" #
2356 "|cmls.8h\t$dst, $src1, $src2}",
2357 (CMHSv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2358 def : InstAlias<"{cmls\t$dst.2s, $src1.2s, $src2.2s" #
2359 "|cmls.2s\t$dst, $src1, $src2}",
2360 (CMHSv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2361 def : InstAlias<"{cmls\t$dst.4s, $src1.4s, $src2.4s" #
2362 "|cmls.4s\t$dst, $src1, $src2}",
2363 (CMHSv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2364 def : InstAlias<"{cmls\t$dst.2d, $src1.2d, $src2.2d" #
2365 "|cmls.2d\t$dst, $src1, $src2}",
2366 (CMHSv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2368 def : InstAlias<"{cmlo\t$dst.8b, $src1.8b, $src2.8b" #
2369 "|cmlo.8b\t$dst, $src1, $src2}",
2370 (CMHIv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2371 def : InstAlias<"{cmlo\t$dst.16b, $src1.16b, $src2.16b" #
2372 "|cmlo.16b\t$dst, $src1, $src2}",
2373 (CMHIv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2374 def : InstAlias<"{cmlo\t$dst.4h, $src1.4h, $src2.4h" #
2375 "|cmlo.4h\t$dst, $src1, $src2}",
2376 (CMHIv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2377 def : InstAlias<"{cmlo\t$dst.8h, $src1.8h, $src2.8h" #
2378 "|cmlo.8h\t$dst, $src1, $src2}",
2379 (CMHIv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2380 def : InstAlias<"{cmlo\t$dst.2s, $src1.2s, $src2.2s" #
2381 "|cmlo.2s\t$dst, $src1, $src2}",
2382 (CMHIv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2383 def : InstAlias<"{cmlo\t$dst.4s, $src1.4s, $src2.4s" #
2384 "|cmlo.4s\t$dst, $src1, $src2}",
2385 (CMHIv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2386 def : InstAlias<"{cmlo\t$dst.2d, $src1.2d, $src2.2d" #
2387 "|cmlo.2d\t$dst, $src1, $src2}",
2388 (CMHIv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2390 def : InstAlias<"{cmle\t$dst.8b, $src1.8b, $src2.8b" #
2391 "|cmle.8b\t$dst, $src1, $src2}",
2392 (CMGEv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2393 def : InstAlias<"{cmle\t$dst.16b, $src1.16b, $src2.16b" #
2394 "|cmle.16b\t$dst, $src1, $src2}",
2395 (CMGEv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2396 def : InstAlias<"{cmle\t$dst.4h, $src1.4h, $src2.4h" #
2397 "|cmle.4h\t$dst, $src1, $src2}",
2398 (CMGEv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2399 def : InstAlias<"{cmle\t$dst.8h, $src1.8h, $src2.8h" #
2400 "|cmle.8h\t$dst, $src1, $src2}",
2401 (CMGEv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2402 def : InstAlias<"{cmle\t$dst.2s, $src1.2s, $src2.2s" #
2403 "|cmle.2s\t$dst, $src1, $src2}",
2404 (CMGEv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2405 def : InstAlias<"{cmle\t$dst.4s, $src1.4s, $src2.4s" #
2406 "|cmle.4s\t$dst, $src1, $src2}",
2407 (CMGEv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2408 def : InstAlias<"{cmle\t$dst.2d, $src1.2d, $src2.2d" #
2409 "|cmle.2d\t$dst, $src1, $src2}",
2410 (CMGEv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2412 def : InstAlias<"{cmlt\t$dst.8b, $src1.8b, $src2.8b" #
2413 "|cmlt.8b\t$dst, $src1, $src2}",
2414 (CMGTv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2415 def : InstAlias<"{cmlt\t$dst.16b, $src1.16b, $src2.16b" #
2416 "|cmlt.16b\t$dst, $src1, $src2}",
2417 (CMGTv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2418 def : InstAlias<"{cmlt\t$dst.4h, $src1.4h, $src2.4h" #
2419 "|cmlt.4h\t$dst, $src1, $src2}",
2420 (CMGTv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2421 def : InstAlias<"{cmlt\t$dst.8h, $src1.8h, $src2.8h" #
2422 "|cmlt.8h\t$dst, $src1, $src2}",
2423 (CMGTv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2424 def : InstAlias<"{cmlt\t$dst.2s, $src1.2s, $src2.2s" #
2425 "|cmlt.2s\t$dst, $src1, $src2}",
2426 (CMGTv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2427 def : InstAlias<"{cmlt\t$dst.4s, $src1.4s, $src2.4s" #
2428 "|cmlt.4s\t$dst, $src1, $src2}",
2429 (CMGTv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2430 def : InstAlias<"{cmlt\t$dst.2d, $src1.2d, $src2.2d" #
2431 "|cmlt.2d\t$dst, $src1, $src2}",
2432 (CMGTv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2434 def : InstAlias<"{fcmle\t$dst.2s, $src1.2s, $src2.2s" #
2435 "|fcmle.2s\t$dst, $src1, $src2}",
2436 (FCMGEv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
2437 def : InstAlias<"{fcmle\t$dst.4s, $src1.4s, $src2.4s" #
2438 "|fcmle.4s\t$dst, $src1, $src2}",
2439 (FCMGEv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
2440 def : InstAlias<"{fcmle\t$dst.2d, $src1.2d, $src2.2d" #
2441 "|fcmle.2d\t$dst, $src1, $src2}",
2442 (FCMGEv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
2444 def : InstAlias<"{fcmlt\t$dst.2s, $src1.2s, $src2.2s" #
2445 "|fcmlt.2s\t$dst, $src1, $src2}",
2446 (FCMGTv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
2447 def : InstAlias<"{fcmlt\t$dst.4s, $src1.4s, $src2.4s" #
2448 "|fcmlt.4s\t$dst, $src1, $src2}",
2449 (FCMGTv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
2450 def : InstAlias<"{fcmlt\t$dst.2d, $src1.2d, $src2.2d" #
2451 "|fcmlt.2d\t$dst, $src1, $src2}",
2452 (FCMGTv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
2454 def : InstAlias<"{facle\t$dst.2s, $src1.2s, $src2.2s" #
2455 "|facle.2s\t$dst, $src1, $src2}",
2456 (FACGEv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
2457 def : InstAlias<"{facle\t$dst.4s, $src1.4s, $src2.4s" #
2458 "|facle.4s\t$dst, $src1, $src2}",
2459 (FACGEv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
2460 def : InstAlias<"{facle\t$dst.2d, $src1.2d, $src2.2d" #
2461 "|facle.2d\t$dst, $src1, $src2}",
2462 (FACGEv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
2464 def : InstAlias<"{faclt\t$dst.2s, $src1.2s, $src2.2s" #
2465 "|faclt.2s\t$dst, $src1, $src2}",
2466 (FACGTv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
2467 def : InstAlias<"{faclt\t$dst.4s, $src1.4s, $src2.4s" #
2468 "|faclt.4s\t$dst, $src1, $src2}",
2469 (FACGTv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
2470 def : InstAlias<"{faclt\t$dst.2d, $src1.2d, $src2.2d" #
2471 "|faclt.2d\t$dst, $src1, $src2}",
2472 (FACGTv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
2474 //===----------------------------------------------------------------------===//
2475 // Advanced SIMD three scalar instructions.
2476 //===----------------------------------------------------------------------===//
2478 defm ADD : SIMDThreeScalarD<0, 0b10000, "add", add>;
2479 defm CMEQ : SIMDThreeScalarD<1, 0b10001, "cmeq", ARM64cmeq>;
2480 defm CMGE : SIMDThreeScalarD<0, 0b00111, "cmge", ARM64cmge>;
2481 defm CMGT : SIMDThreeScalarD<0, 0b00110, "cmgt", ARM64cmgt>;
2482 defm CMHI : SIMDThreeScalarD<1, 0b00110, "cmhi", ARM64cmhi>;
2483 defm CMHS : SIMDThreeScalarD<1, 0b00111, "cmhs", ARM64cmhs>;
2484 defm CMTST : SIMDThreeScalarD<0, 0b10001, "cmtst", ARM64cmtst>;
2485 defm FABD : SIMDThreeScalarSD<1, 1, 0b11010, "fabd", int_arm64_sisd_fabd>;
2486 def : Pat<(v1f64 (int_arm64_neon_fabd (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2487 (FABD64 FPR64:$Rn, FPR64:$Rm)>;
2488 defm FACGE : SIMDThreeScalarFPCmp<1, 0, 0b11101, "facge",
2489 int_arm64_neon_facge>;
2490 defm FACGT : SIMDThreeScalarFPCmp<1, 1, 0b11101, "facgt",
2491 int_arm64_neon_facgt>;
2492 defm FCMEQ : SIMDThreeScalarFPCmp<0, 0, 0b11100, "fcmeq", ARM64fcmeq>;
2493 defm FCMGE : SIMDThreeScalarFPCmp<1, 0, 0b11100, "fcmge", ARM64fcmge>;
2494 defm FCMGT : SIMDThreeScalarFPCmp<1, 1, 0b11100, "fcmgt", ARM64fcmgt>;
2495 defm FMULX : SIMDThreeScalarSD<0, 0, 0b11011, "fmulx", int_arm64_neon_fmulx>;
2496 defm FRECPS : SIMDThreeScalarSD<0, 0, 0b11111, "frecps", int_arm64_neon_frecps>;
2497 defm FRSQRTS : SIMDThreeScalarSD<0, 1, 0b11111, "frsqrts", int_arm64_neon_frsqrts>;
2498 defm SQADD : SIMDThreeScalarBHSD<0, 0b00001, "sqadd", int_arm64_neon_sqadd>;
2499 defm SQDMULH : SIMDThreeScalarHS< 0, 0b10110, "sqdmulh", int_arm64_neon_sqdmulh>;
2500 defm SQRDMULH : SIMDThreeScalarHS< 1, 0b10110, "sqrdmulh", int_arm64_neon_sqrdmulh>;
2501 defm SQRSHL : SIMDThreeScalarBHSD<0, 0b01011, "sqrshl",int_arm64_neon_sqrshl>;
2502 defm SQSHL : SIMDThreeScalarBHSD<0, 0b01001, "sqshl", int_arm64_neon_sqshl>;
2503 defm SQSUB : SIMDThreeScalarBHSD<0, 0b00101, "sqsub", int_arm64_neon_sqsub>;
2504 defm SRSHL : SIMDThreeScalarD< 0, 0b01010, "srshl", int_arm64_neon_srshl>;
2505 defm SSHL : SIMDThreeScalarD< 0, 0b01000, "sshl", int_arm64_neon_sshl>;
2506 defm SUB : SIMDThreeScalarD< 1, 0b10000, "sub", sub>;
2507 defm UQADD : SIMDThreeScalarBHSD<1, 0b00001, "uqadd", int_arm64_neon_uqadd>;
2508 defm UQRSHL : SIMDThreeScalarBHSD<1, 0b01011, "uqrshl",int_arm64_neon_uqrshl>;
2509 defm UQSHL : SIMDThreeScalarBHSD<1, 0b01001, "uqshl", int_arm64_neon_uqshl>;
2510 defm UQSUB : SIMDThreeScalarBHSD<1, 0b00101, "uqsub", int_arm64_neon_uqsub>;
2511 defm URSHL : SIMDThreeScalarD< 1, 0b01010, "urshl", int_arm64_neon_urshl>;
2512 defm USHL : SIMDThreeScalarD< 1, 0b01000, "ushl", int_arm64_neon_ushl>;
2514 def : InstAlias<"cmls $dst, $src1, $src2",
2515 (CMHSv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2516 def : InstAlias<"cmle $dst, $src1, $src2",
2517 (CMGEv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2518 def : InstAlias<"cmlo $dst, $src1, $src2",
2519 (CMHIv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2520 def : InstAlias<"cmlt $dst, $src1, $src2",
2521 (CMGTv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2522 def : InstAlias<"fcmle $dst, $src1, $src2",
2523 (FCMGE32 FPR32:$dst, FPR32:$src2, FPR32:$src1)>;
2524 def : InstAlias<"fcmle $dst, $src1, $src2",
2525 (FCMGE64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2526 def : InstAlias<"fcmlt $dst, $src1, $src2",
2527 (FCMGT32 FPR32:$dst, FPR32:$src2, FPR32:$src1)>;
2528 def : InstAlias<"fcmlt $dst, $src1, $src2",
2529 (FCMGT64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2530 def : InstAlias<"facle $dst, $src1, $src2",
2531 (FACGE32 FPR32:$dst, FPR32:$src2, FPR32:$src1)>;
2532 def : InstAlias<"facle $dst, $src1, $src2",
2533 (FACGE64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2534 def : InstAlias<"faclt $dst, $src1, $src2",
2535 (FACGT32 FPR32:$dst, FPR32:$src2, FPR32:$src1)>;
2536 def : InstAlias<"faclt $dst, $src1, $src2",
2537 (FACGT64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2539 //===----------------------------------------------------------------------===//
2540 // Advanced SIMD three scalar instructions (mixed operands).
2541 //===----------------------------------------------------------------------===//
2542 defm SQDMULL : SIMDThreeScalarMixedHS<0, 0b11010, "sqdmull",
2543 int_arm64_neon_sqdmulls_scalar>;
2544 defm SQDMLAL : SIMDThreeScalarMixedTiedHS<0, 0b10010, "sqdmlal">;
2545 defm SQDMLSL : SIMDThreeScalarMixedTiedHS<0, 0b10110, "sqdmlsl">;
2547 def : Pat<(i64 (int_arm64_neon_sqadd (i64 FPR64:$Rd),
2548 (i64 (int_arm64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
2549 (i32 FPR32:$Rm))))),
2550 (SQDMLALi32 FPR64:$Rd, FPR32:$Rn, FPR32:$Rm)>;
2551 def : Pat<(i64 (int_arm64_neon_sqsub (i64 FPR64:$Rd),
2552 (i64 (int_arm64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
2553 (i32 FPR32:$Rm))))),
2554 (SQDMLSLi32 FPR64:$Rd, FPR32:$Rn, FPR32:$Rm)>;
2556 //===----------------------------------------------------------------------===//
2557 // Advanced SIMD two scalar instructions.
2558 //===----------------------------------------------------------------------===//
2560 defm ABS : SIMDTwoScalarD< 0, 0b01011, "abs", int_arm64_neon_abs>;
2561 defm CMEQ : SIMDCmpTwoScalarD< 0, 0b01001, "cmeq", ARM64cmeqz>;
2562 defm CMGE : SIMDCmpTwoScalarD< 1, 0b01000, "cmge", ARM64cmgez>;
2563 defm CMGT : SIMDCmpTwoScalarD< 0, 0b01000, "cmgt", ARM64cmgtz>;
2564 defm CMLE : SIMDCmpTwoScalarD< 1, 0b01001, "cmle", ARM64cmlez>;
2565 defm CMLT : SIMDCmpTwoScalarD< 0, 0b01010, "cmlt", ARM64cmltz>;
2566 defm FCMEQ : SIMDCmpTwoScalarSD<0, 1, 0b01101, "fcmeq", ARM64fcmeqz>;
2567 defm FCMGE : SIMDCmpTwoScalarSD<1, 1, 0b01100, "fcmge", ARM64fcmgez>;
2568 defm FCMGT : SIMDCmpTwoScalarSD<0, 1, 0b01100, "fcmgt", ARM64fcmgtz>;
2569 defm FCMLE : SIMDCmpTwoScalarSD<1, 1, 0b01101, "fcmle", ARM64fcmlez>;
2570 defm FCMLT : SIMDCmpTwoScalarSD<0, 1, 0b01110, "fcmlt", ARM64fcmltz>;
2571 defm FCVTAS : SIMDTwoScalarSD< 0, 0, 0b11100, "fcvtas">;
2572 defm FCVTAU : SIMDTwoScalarSD< 1, 0, 0b11100, "fcvtau">;
2573 defm FCVTMS : SIMDTwoScalarSD< 0, 0, 0b11011, "fcvtms">;
2574 defm FCVTMU : SIMDTwoScalarSD< 1, 0, 0b11011, "fcvtmu">;
2575 defm FCVTNS : SIMDTwoScalarSD< 0, 0, 0b11010, "fcvtns">;
2576 defm FCVTNU : SIMDTwoScalarSD< 1, 0, 0b11010, "fcvtnu">;
2577 defm FCVTPS : SIMDTwoScalarSD< 0, 1, 0b11010, "fcvtps">;
2578 defm FCVTPU : SIMDTwoScalarSD< 1, 1, 0b11010, "fcvtpu">;
2579 def FCVTXNv1i64 : SIMDInexactCvtTwoScalar<0b10110, "fcvtxn">;
2580 defm FCVTZS : SIMDTwoScalarSD< 0, 1, 0b11011, "fcvtzs">;
2581 defm FCVTZU : SIMDTwoScalarSD< 1, 1, 0b11011, "fcvtzu">;
2582 defm FRECPE : SIMDTwoScalarSD< 0, 1, 0b11101, "frecpe">;
2583 defm FRECPX : SIMDTwoScalarSD< 0, 1, 0b11111, "frecpx">;
2584 defm FRSQRTE : SIMDTwoScalarSD< 1, 1, 0b11101, "frsqrte">;
2585 defm NEG : SIMDTwoScalarD< 1, 0b01011, "neg",
2586 UnOpFrag<(sub immAllZerosV, node:$LHS)> >;
2587 defm SCVTF : SIMDTwoScalarCVTSD< 0, 0, 0b11101, "scvtf", ARM64sitof>;
2588 defm SQABS : SIMDTwoScalarBHSD< 0, 0b00111, "sqabs", int_arm64_neon_sqabs>;
2589 defm SQNEG : SIMDTwoScalarBHSD< 1, 0b00111, "sqneg", int_arm64_neon_sqneg>;
2590 defm SQXTN : SIMDTwoScalarMixedBHS< 0, 0b10100, "sqxtn", int_arm64_neon_scalar_sqxtn>;
2591 defm SQXTUN : SIMDTwoScalarMixedBHS< 1, 0b10010, "sqxtun", int_arm64_neon_scalar_sqxtun>;
2592 defm SUQADD : SIMDTwoScalarBHSDTied< 0, 0b00011, "suqadd",
2593 int_arm64_neon_suqadd>;
2594 defm UCVTF : SIMDTwoScalarCVTSD< 1, 0, 0b11101, "ucvtf", ARM64uitof>;
2595 defm UQXTN : SIMDTwoScalarMixedBHS<1, 0b10100, "uqxtn", int_arm64_neon_scalar_uqxtn>;
2596 defm USQADD : SIMDTwoScalarBHSDTied< 1, 0b00011, "usqadd",
2597 int_arm64_neon_usqadd>;
2599 def : Pat<(v1i64 (int_arm64_neon_fcvtas (v1f64 FPR64:$Rn))),
2600 (FCVTASv1i64 FPR64:$Rn)>;
2601 def : Pat<(v1i64 (int_arm64_neon_fcvtau (v1f64 FPR64:$Rn))),
2602 (FCVTAUv1i64 FPR64:$Rn)>;
2603 def : Pat<(v1i64 (int_arm64_neon_fcvtms (v1f64 FPR64:$Rn))),
2604 (FCVTMSv1i64 FPR64:$Rn)>;
2605 def : Pat<(v1i64 (int_arm64_neon_fcvtmu (v1f64 FPR64:$Rn))),
2606 (FCVTMUv1i64 FPR64:$Rn)>;
2607 def : Pat<(v1i64 (int_arm64_neon_fcvtns (v1f64 FPR64:$Rn))),
2608 (FCVTNSv1i64 FPR64:$Rn)>;
2609 def : Pat<(v1i64 (int_arm64_neon_fcvtnu (v1f64 FPR64:$Rn))),
2610 (FCVTNUv1i64 FPR64:$Rn)>;
2611 def : Pat<(v1i64 (int_arm64_neon_fcvtps (v1f64 FPR64:$Rn))),
2612 (FCVTPSv1i64 FPR64:$Rn)>;
2613 def : Pat<(v1i64 (int_arm64_neon_fcvtpu (v1f64 FPR64:$Rn))),
2614 (FCVTPUv1i64 FPR64:$Rn)>;
2616 def : Pat<(f32 (int_arm64_neon_frecpe (f32 FPR32:$Rn))),
2617 (FRECPEv1i32 FPR32:$Rn)>;
2618 def : Pat<(f64 (int_arm64_neon_frecpe (f64 FPR64:$Rn))),
2619 (FRECPEv1i64 FPR64:$Rn)>;
2620 def : Pat<(v1f64 (int_arm64_neon_frecpe (v1f64 FPR64:$Rn))),
2621 (FRECPEv1i64 FPR64:$Rn)>;
2623 def : Pat<(f32 (int_arm64_neon_frecpx (f32 FPR32:$Rn))),
2624 (FRECPXv1i32 FPR32:$Rn)>;
2625 def : Pat<(f64 (int_arm64_neon_frecpx (f64 FPR64:$Rn))),
2626 (FRECPXv1i64 FPR64:$Rn)>;
2628 def : Pat<(f32 (int_arm64_neon_frsqrte (f32 FPR32:$Rn))),
2629 (FRSQRTEv1i32 FPR32:$Rn)>;
2630 def : Pat<(f64 (int_arm64_neon_frsqrte (f64 FPR64:$Rn))),
2631 (FRSQRTEv1i64 FPR64:$Rn)>;
2632 def : Pat<(v1f64 (int_arm64_neon_frsqrte (v1f64 FPR64:$Rn))),
2633 (FRSQRTEv1i64 FPR64:$Rn)>;
2635 // If an integer is about to be converted to a floating point value,
2636 // just load it on the floating point unit.
2637 // Here are the patterns for 8 and 16-bits to float.
2639 def : Pat <(f32 (uint_to_fp (i32 (zextloadi8 ro_indexed8:$addr)))),
2640 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2641 (LDRBro ro_indexed8:$addr), bsub))>;
2642 def : Pat <(f32 (uint_to_fp (i32 (zextloadi8 am_indexed8:$addr)))),
2643 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2644 (LDRBui am_indexed8:$addr), bsub))>;
2645 def : Pat <(f32 (uint_to_fp (i32 (zextloadi8 am_unscaled8:$addr)))),
2646 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2647 (LDURBi am_unscaled8:$addr), bsub))>;
2648 // 16-bits -> float.
2649 def : Pat <(f32 (uint_to_fp (i32 (zextloadi16 ro_indexed16:$addr)))),
2650 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2651 (LDRHro ro_indexed16:$addr), hsub))>;
2652 def : Pat <(f32 (uint_to_fp (i32 (zextloadi16 am_indexed16:$addr)))),
2653 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2654 (LDRHui am_indexed16:$addr), hsub))>;
2655 def : Pat <(f32 (uint_to_fp (i32 (zextloadi16 am_unscaled16:$addr)))),
2656 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2657 (LDURHi am_unscaled16:$addr), hsub))>;
2658 // 32-bits are handled in target specific dag combine:
2659 // performIntToFpCombine.
2660 // 64-bits integer to 32-bits floating point, not possible with
2661 // UCVTF on floating point registers (both source and destination
2662 // must have the same size).
2664 // Here are the patterns for 8, 16, 32, and 64-bits to double.
2665 // 8-bits -> double.
2666 def : Pat <(f64 (uint_to_fp (i32 (zextloadi8 ro_indexed8:$addr)))),
2667 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2668 (LDRBro ro_indexed8:$addr), bsub))>;
2669 def : Pat <(f64 (uint_to_fp (i32 (zextloadi8 am_indexed8:$addr)))),
2670 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2671 (LDRBui am_indexed8:$addr), bsub))>;
2672 def : Pat <(f64 (uint_to_fp (i32 (zextloadi8 am_unscaled8:$addr)))),
2673 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2674 (LDURBi am_unscaled8:$addr), bsub))>;
2675 // 16-bits -> double.
2676 def : Pat <(f64 (uint_to_fp (i32 (zextloadi16 ro_indexed16:$addr)))),
2677 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2678 (LDRHro ro_indexed16:$addr), hsub))>;
2679 def : Pat <(f64 (uint_to_fp (i32 (zextloadi16 am_indexed16:$addr)))),
2680 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2681 (LDRHui am_indexed16:$addr), hsub))>;
2682 def : Pat <(f64 (uint_to_fp (i32 (zextloadi16 am_unscaled16:$addr)))),
2683 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2684 (LDURHi am_unscaled16:$addr), hsub))>;
2685 // 32-bits -> double.
2686 def : Pat <(f64 (uint_to_fp (i32 (load ro_indexed32:$addr)))),
2687 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2688 (LDRSro ro_indexed32:$addr), ssub))>;
2689 def : Pat <(f64 (uint_to_fp (i32 (load am_indexed32:$addr)))),
2690 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2691 (LDRSui am_indexed32:$addr), ssub))>;
2692 def : Pat <(f64 (uint_to_fp (i32 (load am_unscaled32:$addr)))),
2693 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2694 (LDURSi am_unscaled32:$addr), ssub))>;
2695 // 64-bits -> double are handled in target specific dag combine:
2696 // performIntToFpCombine.
2698 //===----------------------------------------------------------------------===//
2699 // Advanced SIMD three different-sized vector instructions.
2700 //===----------------------------------------------------------------------===//
2702 defm ADDHN : SIMDNarrowThreeVectorBHS<0,0b0100,"addhn", int_arm64_neon_addhn>;
2703 defm SUBHN : SIMDNarrowThreeVectorBHS<0,0b0110,"subhn", int_arm64_neon_subhn>;
2704 defm RADDHN : SIMDNarrowThreeVectorBHS<1,0b0100,"raddhn",int_arm64_neon_raddhn>;
2705 defm RSUBHN : SIMDNarrowThreeVectorBHS<1,0b0110,"rsubhn",int_arm64_neon_rsubhn>;
2706 defm PMULL : SIMDDifferentThreeVectorBD<0,0b1110,"pmull",int_arm64_neon_pmull>;
2707 defm SABAL : SIMDLongThreeVectorTiedBHSabal<0,0b0101,"sabal",
2708 int_arm64_neon_sabd>;
2709 defm SABDL : SIMDLongThreeVectorBHSabdl<0, 0b0111, "sabdl",
2710 int_arm64_neon_sabd>;
2711 defm SADDL : SIMDLongThreeVectorBHS< 0, 0b0000, "saddl",
2712 BinOpFrag<(add (sext node:$LHS), (sext node:$RHS))>>;
2713 defm SADDW : SIMDWideThreeVectorBHS< 0, 0b0001, "saddw",
2714 BinOpFrag<(add node:$LHS, (sext node:$RHS))>>;
2715 defm SMLAL : SIMDLongThreeVectorTiedBHS<0, 0b1000, "smlal",
2716 TriOpFrag<(add node:$LHS, (int_arm64_neon_smull node:$MHS, node:$RHS))>>;
2717 defm SMLSL : SIMDLongThreeVectorTiedBHS<0, 0b1010, "smlsl",
2718 TriOpFrag<(sub node:$LHS, (int_arm64_neon_smull node:$MHS, node:$RHS))>>;
2719 defm SMULL : SIMDLongThreeVectorBHS<0, 0b1100, "smull", int_arm64_neon_smull>;
2720 defm SQDMLAL : SIMDLongThreeVectorSQDMLXTiedHS<0, 0b1001, "sqdmlal",
2721 int_arm64_neon_sqadd>;
2722 defm SQDMLSL : SIMDLongThreeVectorSQDMLXTiedHS<0, 0b1011, "sqdmlsl",
2723 int_arm64_neon_sqsub>;
2724 defm SQDMULL : SIMDLongThreeVectorHS<0, 0b1101, "sqdmull",
2725 int_arm64_neon_sqdmull>;
2726 defm SSUBL : SIMDLongThreeVectorBHS<0, 0b0010, "ssubl",
2727 BinOpFrag<(sub (sext node:$LHS), (sext node:$RHS))>>;
2728 defm SSUBW : SIMDWideThreeVectorBHS<0, 0b0011, "ssubw",
2729 BinOpFrag<(sub node:$LHS, (sext node:$RHS))>>;
2730 defm UABAL : SIMDLongThreeVectorTiedBHSabal<1, 0b0101, "uabal",
2731 int_arm64_neon_uabd>;
2732 defm UABDL : SIMDLongThreeVectorBHSabdl<1, 0b0111, "uabdl",
2733 int_arm64_neon_uabd>;
2734 defm UADDL : SIMDLongThreeVectorBHS<1, 0b0000, "uaddl",
2735 BinOpFrag<(add (zext node:$LHS), (zext node:$RHS))>>;
2736 defm UADDW : SIMDWideThreeVectorBHS<1, 0b0001, "uaddw",
2737 BinOpFrag<(add node:$LHS, (zext node:$RHS))>>;
2738 defm UMLAL : SIMDLongThreeVectorTiedBHS<1, 0b1000, "umlal",
2739 TriOpFrag<(add node:$LHS, (int_arm64_neon_umull node:$MHS, node:$RHS))>>;
2740 defm UMLSL : SIMDLongThreeVectorTiedBHS<1, 0b1010, "umlsl",
2741 TriOpFrag<(sub node:$LHS, (int_arm64_neon_umull node:$MHS, node:$RHS))>>;
2742 defm UMULL : SIMDLongThreeVectorBHS<1, 0b1100, "umull", int_arm64_neon_umull>;
2743 defm USUBL : SIMDLongThreeVectorBHS<1, 0b0010, "usubl",
2744 BinOpFrag<(sub (zext node:$LHS), (zext node:$RHS))>>;
2745 defm USUBW : SIMDWideThreeVectorBHS< 1, 0b0011, "usubw",
2746 BinOpFrag<(sub node:$LHS, (zext node:$RHS))>>;
2748 // Patterns for 64-bit pmull
2749 def : Pat<(int_arm64_neon_pmull64 V64:$Rn, V64:$Rm),
2750 (PMULLv1i64 V64:$Rn, V64:$Rm)>;
2751 def : Pat<(int_arm64_neon_pmull64 (vector_extract (v2i64 V128:$Rn), (i64 1)),
2752 (vector_extract (v2i64 V128:$Rm), (i64 1))),
2753 (PMULLv2i64 V128:$Rn, V128:$Rm)>;
2755 // CodeGen patterns for addhn and subhn instructions, which can actually be
2756 // written in LLVM IR without too much difficulty.
2759 def : Pat<(v8i8 (trunc (v8i16 (ARM64vlshr (add V128:$Rn, V128:$Rm), (i32 8))))),
2760 (ADDHNv8i16_v8i8 V128:$Rn, V128:$Rm)>;
2761 def : Pat<(v4i16 (trunc (v4i32 (ARM64vlshr (add V128:$Rn, V128:$Rm),
2763 (ADDHNv4i32_v4i16 V128:$Rn, V128:$Rm)>;
2764 def : Pat<(v2i32 (trunc (v2i64 (ARM64vlshr (add V128:$Rn, V128:$Rm),
2766 (ADDHNv2i64_v2i32 V128:$Rn, V128:$Rm)>;
2767 def : Pat<(concat_vectors (v8i8 V64:$Rd),
2768 (trunc (v8i16 (ARM64vlshr (add V128:$Rn, V128:$Rm),
2770 (ADDHNv8i16_v16i8 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
2771 V128:$Rn, V128:$Rm)>;
2772 def : Pat<(concat_vectors (v4i16 V64:$Rd),
2773 (trunc (v4i32 (ARM64vlshr (add V128:$Rn, V128:$Rm),
2775 (ADDHNv4i32_v8i16 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
2776 V128:$Rn, V128:$Rm)>;
2777 def : Pat<(concat_vectors (v2i32 V64:$Rd),
2778 (trunc (v2i64 (ARM64vlshr (add V128:$Rn, V128:$Rm),
2780 (ADDHNv2i64_v4i32 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
2781 V128:$Rn, V128:$Rm)>;
2784 def : Pat<(v8i8 (trunc (v8i16 (ARM64vlshr (sub V128:$Rn, V128:$Rm), (i32 8))))),
2785 (SUBHNv8i16_v8i8 V128:$Rn, V128:$Rm)>;
2786 def : Pat<(v4i16 (trunc (v4i32 (ARM64vlshr (sub V128:$Rn, V128:$Rm),
2788 (SUBHNv4i32_v4i16 V128:$Rn, V128:$Rm)>;
2789 def : Pat<(v2i32 (trunc (v2i64 (ARM64vlshr (sub V128:$Rn, V128:$Rm),
2791 (SUBHNv2i64_v2i32 V128:$Rn, V128:$Rm)>;
2792 def : Pat<(concat_vectors (v8i8 V64:$Rd),
2793 (trunc (v8i16 (ARM64vlshr (sub V128:$Rn, V128:$Rm),
2795 (SUBHNv8i16_v16i8 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
2796 V128:$Rn, V128:$Rm)>;
2797 def : Pat<(concat_vectors (v4i16 V64:$Rd),
2798 (trunc (v4i32 (ARM64vlshr (sub V128:$Rn, V128:$Rm),
2800 (SUBHNv4i32_v8i16 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
2801 V128:$Rn, V128:$Rm)>;
2802 def : Pat<(concat_vectors (v2i32 V64:$Rd),
2803 (trunc (v2i64 (ARM64vlshr (sub V128:$Rn, V128:$Rm),
2805 (SUBHNv2i64_v4i32 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
2806 V128:$Rn, V128:$Rm)>;
2808 //----------------------------------------------------------------------------
2809 // AdvSIMD bitwise extract from vector instruction.
2810 //----------------------------------------------------------------------------
2812 defm EXT : SIMDBitwiseExtract<"ext">;
2814 def : Pat<(v4i16 (ARM64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
2815 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
2816 def : Pat<(v8i16 (ARM64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
2817 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
2818 def : Pat<(v2i32 (ARM64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
2819 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
2820 def : Pat<(v2f32 (ARM64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
2821 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
2822 def : Pat<(v4i32 (ARM64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
2823 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
2824 def : Pat<(v4f32 (ARM64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
2825 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
2826 def : Pat<(v2i64 (ARM64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
2827 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
2828 def : Pat<(v2f64 (ARM64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
2829 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
2831 // We use EXT to handle extract_subvector to copy the upper 64-bits of a
2833 def : Pat<(v8i8 (extract_subvector V128:$Rn, (i64 8))),
2834 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
2835 def : Pat<(v4i16 (extract_subvector V128:$Rn, (i64 4))),
2836 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
2837 def : Pat<(v2i32 (extract_subvector V128:$Rn, (i64 2))),
2838 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
2839 def : Pat<(v1i64 (extract_subvector V128:$Rn, (i64 1))),
2840 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
2841 def : Pat<(v2f32 (extract_subvector V128:$Rn, (i64 2))),
2842 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
2843 def : Pat<(v1f64 (extract_subvector V128:$Rn, (i64 1))),
2844 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
2847 //----------------------------------------------------------------------------
2848 // AdvSIMD zip vector
2849 //----------------------------------------------------------------------------
2851 defm TRN1 : SIMDZipVector<0b010, "trn1", ARM64trn1>;
2852 defm TRN2 : SIMDZipVector<0b110, "trn2", ARM64trn2>;
2853 defm UZP1 : SIMDZipVector<0b001, "uzp1", ARM64uzp1>;
2854 defm UZP2 : SIMDZipVector<0b101, "uzp2", ARM64uzp2>;
2855 defm ZIP1 : SIMDZipVector<0b011, "zip1", ARM64zip1>;
2856 defm ZIP2 : SIMDZipVector<0b111, "zip2", ARM64zip2>;
2858 //----------------------------------------------------------------------------
2859 // AdvSIMD TBL/TBX instructions
2860 //----------------------------------------------------------------------------
2862 defm TBL : SIMDTableLookup< 0, "tbl">;
2863 defm TBX : SIMDTableLookupTied<1, "tbx">;
2865 def : Pat<(v8i8 (int_arm64_neon_tbl1 (v16i8 VecListOne128:$Rn), (v8i8 V64:$Ri))),
2866 (TBLv8i8One VecListOne128:$Rn, V64:$Ri)>;
2867 def : Pat<(v16i8 (int_arm64_neon_tbl1 (v16i8 V128:$Ri), (v16i8 V128:$Rn))),
2868 (TBLv16i8One V128:$Ri, V128:$Rn)>;
2870 def : Pat<(v8i8 (int_arm64_neon_tbx1 (v8i8 V64:$Rd),
2871 (v16i8 VecListOne128:$Rn), (v8i8 V64:$Ri))),
2872 (TBXv8i8One V64:$Rd, VecListOne128:$Rn, V64:$Ri)>;
2873 def : Pat<(v16i8 (int_arm64_neon_tbx1 (v16i8 V128:$Rd),
2874 (v16i8 V128:$Ri), (v16i8 V128:$Rn))),
2875 (TBXv16i8One V128:$Rd, V128:$Ri, V128:$Rn)>;
2878 //----------------------------------------------------------------------------
2879 // AdvSIMD scalar CPY instruction
2880 //----------------------------------------------------------------------------
2882 defm CPY : SIMDScalarCPY<"cpy">;
2884 //----------------------------------------------------------------------------
2885 // AdvSIMD scalar pairwise instructions
2886 //----------------------------------------------------------------------------
2888 defm ADDP : SIMDPairwiseScalarD<0, 0b11011, "addp">;
2889 defm FADDP : SIMDPairwiseScalarSD<1, 0, 0b01101, "faddp">;
2890 defm FMAXNMP : SIMDPairwiseScalarSD<1, 0, 0b01100, "fmaxnmp">;
2891 defm FMAXP : SIMDPairwiseScalarSD<1, 0, 0b01111, "fmaxp">;
2892 defm FMINNMP : SIMDPairwiseScalarSD<1, 1, 0b01100, "fminnmp">;
2893 defm FMINP : SIMDPairwiseScalarSD<1, 1, 0b01111, "fminp">;
2894 def : Pat<(i64 (int_arm64_neon_saddv (v2i64 V128:$Rn))),
2895 (ADDPv2i64p V128:$Rn)>;
2896 def : Pat<(i64 (int_arm64_neon_uaddv (v2i64 V128:$Rn))),
2897 (ADDPv2i64p V128:$Rn)>;
2898 def : Pat<(f32 (int_arm64_neon_faddv (v2f32 V64:$Rn))),
2899 (FADDPv2i32p V64:$Rn)>;
2900 def : Pat<(f32 (int_arm64_neon_faddv (v4f32 V128:$Rn))),
2901 (FADDPv2i32p (EXTRACT_SUBREG (FADDPv4f32 V128:$Rn, V128:$Rn), dsub))>;
2902 def : Pat<(f64 (int_arm64_neon_faddv (v2f64 V128:$Rn))),
2903 (FADDPv2i64p V128:$Rn)>;
2904 def : Pat<(f32 (int_arm64_neon_fmaxnmv (v2f32 V64:$Rn))),
2905 (FMAXNMPv2i32p V64:$Rn)>;
2906 def : Pat<(f64 (int_arm64_neon_fmaxnmv (v2f64 V128:$Rn))),
2907 (FMAXNMPv2i64p V128:$Rn)>;
2908 def : Pat<(f32 (int_arm64_neon_fmaxv (v2f32 V64:$Rn))),
2909 (FMAXPv2i32p V64:$Rn)>;
2910 def : Pat<(f64 (int_arm64_neon_fmaxv (v2f64 V128:$Rn))),
2911 (FMAXPv2i64p V128:$Rn)>;
2912 def : Pat<(f32 (int_arm64_neon_fminnmv (v2f32 V64:$Rn))),
2913 (FMINNMPv2i32p V64:$Rn)>;
2914 def : Pat<(f64 (int_arm64_neon_fminnmv (v2f64 V128:$Rn))),
2915 (FMINNMPv2i64p V128:$Rn)>;
2916 def : Pat<(f32 (int_arm64_neon_fminv (v2f32 V64:$Rn))),
2917 (FMINPv2i32p V64:$Rn)>;
2918 def : Pat<(f64 (int_arm64_neon_fminv (v2f64 V128:$Rn))),
2919 (FMINPv2i64p V128:$Rn)>;
2921 //----------------------------------------------------------------------------
2922 // AdvSIMD INS/DUP instructions
2923 //----------------------------------------------------------------------------
2925 def DUPv8i8gpr : SIMDDupFromMain<0, 0b00001, ".8b", v8i8, V64, GPR32>;
2926 def DUPv16i8gpr : SIMDDupFromMain<1, 0b00001, ".16b", v16i8, V128, GPR32>;
2927 def DUPv4i16gpr : SIMDDupFromMain<0, 0b00010, ".4h", v4i16, V64, GPR32>;
2928 def DUPv8i16gpr : SIMDDupFromMain<1, 0b00010, ".8h", v8i16, V128, GPR32>;
2929 def DUPv2i32gpr : SIMDDupFromMain<0, 0b00100, ".2s", v2i32, V64, GPR32>;
2930 def DUPv4i32gpr : SIMDDupFromMain<1, 0b00100, ".4s", v4i32, V128, GPR32>;
2931 def DUPv2i64gpr : SIMDDupFromMain<1, 0b01000, ".2d", v2i64, V128, GPR64>;
2933 def DUPv2i64lane : SIMDDup64FromElement;
2934 def DUPv2i32lane : SIMDDup32FromElement<0, ".2s", v2i32, V64>;
2935 def DUPv4i32lane : SIMDDup32FromElement<1, ".4s", v4i32, V128>;
2936 def DUPv4i16lane : SIMDDup16FromElement<0, ".4h", v4i16, V64>;
2937 def DUPv8i16lane : SIMDDup16FromElement<1, ".8h", v8i16, V128>;
2938 def DUPv8i8lane : SIMDDup8FromElement <0, ".8b", v8i8, V64>;
2939 def DUPv16i8lane : SIMDDup8FromElement <1, ".16b", v16i8, V128>;
2941 def : Pat<(v2f32 (ARM64dup (f32 FPR32:$Rn))),
2942 (v2f32 (DUPv2i32lane
2943 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rn, ssub),
2945 def : Pat<(v4f32 (ARM64dup (f32 FPR32:$Rn))),
2946 (v4f32 (DUPv4i32lane
2947 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rn, ssub),
2949 def : Pat<(v2f64 (ARM64dup (f64 FPR64:$Rn))),
2950 (v2f64 (DUPv2i64lane
2951 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$Rn, dsub),
2954 def : Pat<(v2f32 (ARM64duplane32 (v4f32 V128:$Rn), VectorIndexS:$imm)),
2955 (DUPv2i32lane V128:$Rn, VectorIndexS:$imm)>;
2956 def : Pat<(v4f32 (ARM64duplane32 (v4f32 V128:$Rn), VectorIndexS:$imm)),
2957 (DUPv4i32lane V128:$Rn, VectorIndexS:$imm)>;
2958 def : Pat<(v2f64 (ARM64duplane64 (v2f64 V128:$Rn), VectorIndexD:$imm)),
2959 (DUPv2i64lane V128:$Rn, VectorIndexD:$imm)>;
2964 def : Pat<(sext_inreg (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx), i8),
2965 (i32 (SMOVvi8to32 V128:$Rn, VectorIndexB:$idx))>;
2966 def : Pat<(sext_inreg (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx), i8),
2967 (i64 (SMOVvi8to64 V128:$Rn, VectorIndexB:$idx))>;
2968 def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
2969 (i32 (SMOVvi16to32 V128:$Rn, VectorIndexH:$idx))>;
2970 def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
2971 (i64 (SMOVvi16to64 V128:$Rn, VectorIndexH:$idx))>;
2972 def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
2973 (i32 (SMOVvi16to32 V128:$Rn, VectorIndexH:$idx))>;
2974 def : Pat<(sext (i32 (vector_extract (v4i32 V128:$Rn), VectorIndexS:$idx))),
2975 (i64 (SMOVvi32to64 V128:$Rn, VectorIndexS:$idx))>;
2977 // Extracting i8 or i16 elements will have the zero-extend transformed to
2978 // an 'and' mask by type legalization since neither i8 nor i16 are legal types
2979 // for ARM64. Match these patterns here since UMOV already zeroes out the high
2980 // bits of the destination register.
2981 def : Pat<(and (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx),
2983 (i32 (UMOVvi8 V128:$Rn, VectorIndexB:$idx))>;
2984 def : Pat<(and (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),
2986 (i32 (UMOVvi16 V128:$Rn, VectorIndexH:$idx))>;
2990 def : Pat<(v16i8 (scalar_to_vector GPR32:$Rn)),
2991 (SUBREG_TO_REG (i32 0),
2992 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
2993 def : Pat<(v8i8 (scalar_to_vector GPR32:$Rn)),
2994 (SUBREG_TO_REG (i32 0),
2995 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
2997 def : Pat<(v8i16 (scalar_to_vector GPR32:$Rn)),
2998 (SUBREG_TO_REG (i32 0),
2999 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3000 def : Pat<(v4i16 (scalar_to_vector GPR32:$Rn)),
3001 (SUBREG_TO_REG (i32 0),
3002 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3004 def : Pat<(v2i32 (scalar_to_vector (i32 FPR32:$Rn))),
3005 (v2i32 (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)),
3006 (i32 FPR32:$Rn), ssub))>;
3007 def : Pat<(v4i32 (scalar_to_vector (i32 FPR32:$Rn))),
3008 (v4i32 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
3009 (i32 FPR32:$Rn), ssub))>;
3010 def : Pat<(v2i64 (scalar_to_vector (i64 FPR64:$Rn))),
3011 (v2i64 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
3012 (i64 FPR64:$Rn), dsub))>;
3014 def : Pat<(v4f32 (scalar_to_vector (f32 FPR32:$Rn))),
3015 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rn, ssub)>;
3016 def : Pat<(v2f32 (scalar_to_vector (f32 FPR32:$Rn))),
3017 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), FPR32:$Rn, ssub)>;
3018 def : Pat<(v2f64 (scalar_to_vector (f64 FPR64:$Rn))),
3019 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$Rn, dsub)>;
3021 def : Pat<(v2f32 (vector_insert (v2f32 V64:$Rn),
3022 (f32 FPR32:$Rm), (i64 VectorIndexS:$imm))),
3025 (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), V64:$Rn, dsub)),
3027 (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rm, ssub)),
3030 def : Pat<(v4f32 (vector_insert (v4f32 V128:$Rn),
3031 (f32 FPR32:$Rm), (i64 VectorIndexS:$imm))),
3033 V128:$Rn, VectorIndexS:$imm,
3034 (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rm, ssub)),
3036 def : Pat<(v2f64 (vector_insert (v2f64 V128:$Rn),
3037 (f64 FPR64:$Rm), (i64 VectorIndexD:$imm))),
3039 V128:$Rn, VectorIndexD:$imm,
3040 (v2f64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$Rm, dsub)),
3043 // Copy an element at a constant index in one vector into a constant indexed
3044 // element of another.
3045 // FIXME refactor to a shared class/dev parameterized on vector type, vector
3046 // index type and INS extension
3047 def : Pat<(v16i8 (int_arm64_neon_vcopy_lane
3048 (v16i8 V128:$Vd), VectorIndexB:$idx, (v16i8 V128:$Vs),
3049 VectorIndexB:$idx2)),
3051 V128:$Vd, VectorIndexB:$idx, V128:$Vs, VectorIndexB:$idx2)
3053 def : Pat<(v8i16 (int_arm64_neon_vcopy_lane
3054 (v8i16 V128:$Vd), VectorIndexH:$idx, (v8i16 V128:$Vs),
3055 VectorIndexH:$idx2)),
3057 V128:$Vd, VectorIndexH:$idx, V128:$Vs, VectorIndexH:$idx2)
3059 def : Pat<(v4i32 (int_arm64_neon_vcopy_lane
3060 (v4i32 V128:$Vd), VectorIndexS:$idx, (v4i32 V128:$Vs),
3061 VectorIndexS:$idx2)),
3063 V128:$Vd, VectorIndexS:$idx, V128:$Vs, VectorIndexS:$idx2)
3065 def : Pat<(v2i64 (int_arm64_neon_vcopy_lane
3066 (v2i64 V128:$Vd), VectorIndexD:$idx, (v2i64 V128:$Vs),
3067 VectorIndexD:$idx2)),
3069 V128:$Vd, VectorIndexD:$idx, V128:$Vs, VectorIndexD:$idx2)
3072 // Floating point vector extractions are codegen'd as either a sequence of
3073 // subregister extractions, possibly fed by an INS if the lane number is
3074 // anything other than zero.
3075 def : Pat<(vector_extract (v2f64 V128:$Rn), 0),
3076 (f64 (EXTRACT_SUBREG V128:$Rn, dsub))>;
3077 def : Pat<(vector_extract (v4f32 V128:$Rn), 0),
3078 (f32 (EXTRACT_SUBREG V128:$Rn, ssub))>;
3079 def : Pat<(vector_extract (v2f64 V128:$Rn), VectorIndexD:$idx),
3080 (f64 (EXTRACT_SUBREG
3081 (INSvi64lane (v2f64 (IMPLICIT_DEF)), 0,
3082 V128:$Rn, VectorIndexD:$idx),
3084 def : Pat<(vector_extract (v4f32 V128:$Rn), VectorIndexS:$idx),
3085 (f32 (EXTRACT_SUBREG
3086 (INSvi32lane (v4f32 (IMPLICIT_DEF)), 0,
3087 V128:$Rn, VectorIndexS:$idx),
3090 // All concat_vectors operations are canonicalised to act on i64 vectors for
3091 // ARM64. In the general case we need an instruction, which had just as well be
3093 class ConcatPat<ValueType DstTy, ValueType SrcTy>
3094 : Pat<(DstTy (concat_vectors (SrcTy V64:$Rd), V64:$Rn)),
3095 (INSvi64lane (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), 1,
3096 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rn, dsub), 0)>;
3098 def : ConcatPat<v2i64, v1i64>;
3099 def : ConcatPat<v2f64, v1f64>;
3100 def : ConcatPat<v4i32, v2i32>;
3101 def : ConcatPat<v4f32, v2f32>;
3102 def : ConcatPat<v8i16, v4i16>;
3103 def : ConcatPat<v16i8, v8i8>;
3105 // If the high lanes are undef, though, we can just ignore them:
3106 class ConcatUndefPat<ValueType DstTy, ValueType SrcTy>
3107 : Pat<(DstTy (concat_vectors (SrcTy V64:$Rn), undef)),
3108 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rn, dsub)>;
3110 def : ConcatUndefPat<v2i64, v1i64>;
3111 def : ConcatUndefPat<v2f64, v1f64>;
3112 def : ConcatUndefPat<v4i32, v2i32>;
3113 def : ConcatUndefPat<v4f32, v2f32>;
3114 def : ConcatUndefPat<v8i16, v4i16>;
3115 def : ConcatUndefPat<v16i8, v8i8>;
3117 //----------------------------------------------------------------------------
3118 // AdvSIMD across lanes instructions
3119 //----------------------------------------------------------------------------
3121 defm ADDV : SIMDAcrossLanesBHS<0, 0b11011, "addv">;
3122 defm SMAXV : SIMDAcrossLanesBHS<0, 0b01010, "smaxv">;
3123 defm SMINV : SIMDAcrossLanesBHS<0, 0b11010, "sminv">;
3124 defm UMAXV : SIMDAcrossLanesBHS<1, 0b01010, "umaxv">;
3125 defm UMINV : SIMDAcrossLanesBHS<1, 0b11010, "uminv">;
3126 defm SADDLV : SIMDAcrossLanesHSD<0, 0b00011, "saddlv">;
3127 defm UADDLV : SIMDAcrossLanesHSD<1, 0b00011, "uaddlv">;
3128 defm FMAXNMV : SIMDAcrossLanesS<0b01100, 0, "fmaxnmv", int_arm64_neon_fmaxnmv>;
3129 defm FMAXV : SIMDAcrossLanesS<0b01111, 0, "fmaxv", int_arm64_neon_fmaxv>;
3130 defm FMINNMV : SIMDAcrossLanesS<0b01100, 1, "fminnmv", int_arm64_neon_fminnmv>;
3131 defm FMINV : SIMDAcrossLanesS<0b01111, 1, "fminv", int_arm64_neon_fminv>;
3133 multiclass SIMDAcrossLanesSignedIntrinsic<string baseOpc, Intrinsic intOp> {
3134 // If there is a sign extension after this intrinsic, consume it as smov already
3136 def : Pat<(i32 (sext_inreg (i32 (intOp (v8i8 V64:$Rn))), i8)),
3138 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3139 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3141 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
3143 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3144 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3146 // If there is a sign extension after this intrinsic, consume it as smov already
3148 def : Pat<(i32 (sext_inreg (i32 (intOp (v16i8 V128:$Rn))), i8)),
3150 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3151 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3153 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
3155 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3156 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3158 // If there is a sign extension after this intrinsic, consume it as smov already
3160 def : Pat<(i32 (sext_inreg (i32 (intOp (v4i16 V64:$Rn))), i16)),
3162 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3163 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3165 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
3167 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3168 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3170 // If there is a sign extension after this intrinsic, consume it as smov already
3172 def : Pat<(i32 (sext_inreg (i32 (intOp (v8i16 V128:$Rn))), i16)),
3174 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3175 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3177 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
3179 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3180 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3183 def : Pat<(i32 (intOp (v4i32 V128:$Rn))),
3184 (i32 (EXTRACT_SUBREG
3185 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3186 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), ssub),
3190 multiclass SIMDAcrossLanesUnsignedIntrinsic<string baseOpc, Intrinsic intOp> {
3191 // If there is a masking operation keeping only what has been actually
3192 // generated, consume it.
3193 def : Pat<(i32 (and (i32 (intOp (v8i8 V64:$Rn))), maski8_or_more)),
3194 (i32 (EXTRACT_SUBREG
3195 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3196 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3198 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
3199 (i32 (EXTRACT_SUBREG
3200 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3201 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3203 // If there is a masking operation keeping only what has been actually
3204 // generated, consume it.
3205 def : Pat<(i32 (and (i32 (intOp (v16i8 V128:$Rn))), maski8_or_more)),
3206 (i32 (EXTRACT_SUBREG
3207 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3208 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3210 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
3211 (i32 (EXTRACT_SUBREG
3212 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3213 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3216 // If there is a masking operation keeping only what has been actually
3217 // generated, consume it.
3218 def : Pat<(i32 (and (i32 (intOp (v4i16 V64:$Rn))), maski16_or_more)),
3219 (i32 (EXTRACT_SUBREG
3220 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3221 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3223 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
3224 (i32 (EXTRACT_SUBREG
3225 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3226 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3228 // If there is a masking operation keeping only what has been actually
3229 // generated, consume it.
3230 def : Pat<(i32 (and (i32 (intOp (v8i16 V128:$Rn))), maski16_or_more)),
3231 (i32 (EXTRACT_SUBREG
3232 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3233 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3235 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
3236 (i32 (EXTRACT_SUBREG
3237 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3238 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3241 def : Pat<(i32 (intOp (v4i32 V128:$Rn))),
3242 (i32 (EXTRACT_SUBREG
3243 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3244 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), ssub),
3249 multiclass SIMDAcrossLanesSignedLongIntrinsic<string baseOpc, Intrinsic intOp> {
3250 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
3252 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3253 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), hsub),
3255 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
3257 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3258 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), hsub),
3261 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
3262 (i32 (EXTRACT_SUBREG
3263 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3264 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), ssub),
3266 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
3267 (i32 (EXTRACT_SUBREG
3268 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3269 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), ssub),
3272 def : Pat<(i64 (intOp (v4i32 V128:$Rn))),
3273 (i64 (EXTRACT_SUBREG
3274 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3275 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), dsub),
3279 multiclass SIMDAcrossLanesUnsignedLongIntrinsic<string baseOpc,
3281 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
3282 (i32 (EXTRACT_SUBREG
3283 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3284 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), hsub),
3286 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
3287 (i32 (EXTRACT_SUBREG
3288 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3289 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), hsub),
3292 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
3293 (i32 (EXTRACT_SUBREG
3294 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3295 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), ssub),
3297 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
3298 (i32 (EXTRACT_SUBREG
3299 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3300 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), ssub),
3303 def : Pat<(i64 (intOp (v4i32 V128:$Rn))),
3304 (i64 (EXTRACT_SUBREG
3305 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3306 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), dsub),
3310 defm : SIMDAcrossLanesSignedIntrinsic<"ADDV", int_arm64_neon_saddv>;
3311 // vaddv_[su]32 is special; -> ADDP Vd.2S,Vn.2S,Vm.2S; return Vd.s[0];Vn==Vm
3312 def : Pat<(i32 (int_arm64_neon_saddv (v2i32 V64:$Rn))),
3313 (EXTRACT_SUBREG (ADDPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3315 defm : SIMDAcrossLanesUnsignedIntrinsic<"ADDV", int_arm64_neon_uaddv>;
3316 // vaddv_[su]32 is special; -> ADDP Vd.2S,Vn.2S,Vm.2S; return Vd.s[0];Vn==Vm
3317 def : Pat<(i32 (int_arm64_neon_uaddv (v2i32 V64:$Rn))),
3318 (EXTRACT_SUBREG (ADDPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3320 defm : SIMDAcrossLanesSignedIntrinsic<"SMAXV", int_arm64_neon_smaxv>;
3321 def : Pat<(i32 (int_arm64_neon_smaxv (v2i32 V64:$Rn))),
3322 (EXTRACT_SUBREG (SMAXPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3324 defm : SIMDAcrossLanesSignedIntrinsic<"SMINV", int_arm64_neon_sminv>;
3325 def : Pat<(i32 (int_arm64_neon_sminv (v2i32 V64:$Rn))),
3326 (EXTRACT_SUBREG (SMINPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3328 defm : SIMDAcrossLanesUnsignedIntrinsic<"UMAXV", int_arm64_neon_umaxv>;
3329 def : Pat<(i32 (int_arm64_neon_umaxv (v2i32 V64:$Rn))),
3330 (EXTRACT_SUBREG (UMAXPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3332 defm : SIMDAcrossLanesUnsignedIntrinsic<"UMINV", int_arm64_neon_uminv>;
3333 def : Pat<(i32 (int_arm64_neon_uminv (v2i32 V64:$Rn))),
3334 (EXTRACT_SUBREG (UMINPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3336 defm : SIMDAcrossLanesSignedLongIntrinsic<"SADDLV", int_arm64_neon_saddlv>;
3337 defm : SIMDAcrossLanesUnsignedLongIntrinsic<"UADDLV", int_arm64_neon_uaddlv>;
3339 // The vaddlv_s32 intrinsic gets mapped to SADDLP.
3340 def : Pat<(i64 (int_arm64_neon_saddlv (v2i32 V64:$Rn))),
3341 (i64 (EXTRACT_SUBREG
3342 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3343 (SADDLPv2i32_v1i64 V64:$Rn), dsub),
3345 // The vaddlv_u32 intrinsic gets mapped to UADDLP.
3346 def : Pat<(i64 (int_arm64_neon_uaddlv (v2i32 V64:$Rn))),
3347 (i64 (EXTRACT_SUBREG
3348 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3349 (UADDLPv2i32_v1i64 V64:$Rn), dsub),
3352 //------------------------------------------------------------------------------
3353 // AdvSIMD modified immediate instructions
3354 //------------------------------------------------------------------------------
3357 defm BIC : SIMDModifiedImmVectorShiftTied<1, 0b11, 0b01, "bic", ARM64bici>;
3359 defm ORR : SIMDModifiedImmVectorShiftTied<0, 0b11, 0b01, "orr", ARM64orri>;
3363 def FMOVv2f64_ns : SIMDModifiedImmVectorNoShift<1, 1, 0b1111, V128, fpimm8,
3365 [(set (v2f64 V128:$Rd), (ARM64fmov imm0_255:$imm8))]>;
3366 def FMOVv2f32_ns : SIMDModifiedImmVectorNoShift<0, 0, 0b1111, V64, fpimm8,
3368 [(set (v2f32 V64:$Rd), (ARM64fmov imm0_255:$imm8))]>;
3369 def FMOVv4f32_ns : SIMDModifiedImmVectorNoShift<1, 0, 0b1111, V128, fpimm8,
3371 [(set (v4f32 V128:$Rd), (ARM64fmov imm0_255:$imm8))]>;
3375 // EDIT byte mask: scalar
3376 let isReMaterializable = 1, isAsCheapAsAMove = 1 in
3377 def MOVID : SIMDModifiedImmScalarNoShift<0, 1, 0b1110, "movi",
3378 [(set FPR64:$Rd, simdimmtype10:$imm8)]>;
3379 // The movi_edit node has the immediate value already encoded, so we use
3380 // a plain imm0_255 here.
3381 def : Pat<(f64 (ARM64movi_edit imm0_255:$shift)),
3382 (MOVID imm0_255:$shift)>;
3384 def : Pat<(v1i64 immAllZerosV), (MOVID (i32 0))>;
3385 def : Pat<(v2i32 immAllZerosV), (MOVID (i32 0))>;
3386 def : Pat<(v4i16 immAllZerosV), (MOVID (i32 0))>;
3387 def : Pat<(v8i8 immAllZerosV), (MOVID (i32 0))>;
3389 def : Pat<(v1i64 immAllOnesV), (MOVID (i32 255))>;
3390 def : Pat<(v2i32 immAllOnesV), (MOVID (i32 255))>;
3391 def : Pat<(v4i16 immAllOnesV), (MOVID (i32 255))>;
3392 def : Pat<(v8i8 immAllOnesV), (MOVID (i32 255))>;
3394 // EDIT byte mask: 2d
3396 // The movi_edit node has the immediate value already encoded, so we use
3397 // a plain imm0_255 in the pattern
3398 let isReMaterializable = 1, isAsCheapAsAMove = 1 in
3399 def MOVIv2d_ns : SIMDModifiedImmVectorNoShift<1, 1, 0b1110, V128,
3402 [(set (v2i64 V128:$Rd), (ARM64movi_edit imm0_255:$imm8))]>;
3405 // Use movi.2d to materialize 0.0 if the HW does zero-cycle zeroing.
3406 // Complexity is added to break a tie with a plain MOVI.
3407 let AddedComplexity = 1 in {
3408 def : Pat<(f32 fpimm0),
3409 (f32 (EXTRACT_SUBREG (v2i64 (MOVIv2d_ns (i32 0))), ssub))>,
3411 def : Pat<(f64 fpimm0),
3412 (f64 (EXTRACT_SUBREG (v2i64 (MOVIv2d_ns (i32 0))), dsub))>,
3416 def : Pat<(v2i64 immAllZerosV), (MOVIv2d_ns (i32 0))>;
3417 def : Pat<(v4i32 immAllZerosV), (MOVIv2d_ns (i32 0))>;
3418 def : Pat<(v8i16 immAllZerosV), (MOVIv2d_ns (i32 0))>;
3419 def : Pat<(v16i8 immAllZerosV), (MOVIv2d_ns (i32 0))>;
3421 def : Pat<(v2i64 immAllOnesV), (MOVIv2d_ns (i32 255))>;
3422 def : Pat<(v4i32 immAllOnesV), (MOVIv2d_ns (i32 255))>;
3423 def : Pat<(v8i16 immAllOnesV), (MOVIv2d_ns (i32 255))>;
3424 def : Pat<(v16i8 immAllOnesV), (MOVIv2d_ns (i32 255))>;
3426 // EDIT per word & halfword: 2s, 4h, 4s, & 8h
3427 defm MOVI : SIMDModifiedImmVectorShift<0, 0b10, 0b00, "movi">;
3428 def : Pat<(v2i32 (ARM64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
3429 (MOVIv2i32 imm0_255:$imm8, imm:$shift)>;
3430 def : Pat<(v4i32 (ARM64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
3431 (MOVIv4i32 imm0_255:$imm8, imm:$shift)>;
3432 def : Pat<(v4i16 (ARM64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
3433 (MOVIv4i16 imm0_255:$imm8, imm:$shift)>;
3434 def : Pat<(v8i16 (ARM64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
3435 (MOVIv8i16 imm0_255:$imm8, imm:$shift)>;
3437 // EDIT per word: 2s & 4s with MSL shifter
3438 def MOVIv2s_msl : SIMDModifiedImmMoveMSL<0, 0, {1,1,0,?}, V64, "movi", ".2s",
3439 [(set (v2i32 V64:$Rd),
3440 (ARM64movi_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
3441 def MOVIv4s_msl : SIMDModifiedImmMoveMSL<1, 0, {1,1,0,?}, V128, "movi", ".4s",
3442 [(set (v4i32 V128:$Rd),
3443 (ARM64movi_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
3445 // Per byte: 8b & 16b
3446 def MOVIv8b_ns : SIMDModifiedImmVectorNoShift<0, 0, 0b1110, V64, imm0_255,
3448 [(set (v8i8 V64:$Rd), (ARM64movi imm0_255:$imm8))]>;
3449 def MOVIv16b_ns : SIMDModifiedImmVectorNoShift<1, 0, 0b1110, V128, imm0_255,
3451 [(set (v16i8 V128:$Rd), (ARM64movi imm0_255:$imm8))]>;
3455 // EDIT per word & halfword: 2s, 4h, 4s, & 8h
3456 defm MVNI : SIMDModifiedImmVectorShift<1, 0b10, 0b00, "mvni">;
3457 def : Pat<(v2i32 (ARM64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
3458 (MVNIv2i32 imm0_255:$imm8, imm:$shift)>;
3459 def : Pat<(v4i32 (ARM64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
3460 (MVNIv4i32 imm0_255:$imm8, imm:$shift)>;
3461 def : Pat<(v4i16 (ARM64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
3462 (MVNIv4i16 imm0_255:$imm8, imm:$shift)>;
3463 def : Pat<(v8i16 (ARM64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
3464 (MVNIv8i16 imm0_255:$imm8, imm:$shift)>;
3466 // EDIT per word: 2s & 4s with MSL shifter
3467 def MVNIv2s_msl : SIMDModifiedImmMoveMSL<0, 1, {1,1,0,?}, V64, "mvni", ".2s",
3468 [(set (v2i32 V64:$Rd),
3469 (ARM64mvni_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
3470 def MVNIv4s_msl : SIMDModifiedImmMoveMSL<1, 1, {1,1,0,?}, V128, "mvni", ".4s",
3471 [(set (v4i32 V128:$Rd),
3472 (ARM64mvni_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
3474 //----------------------------------------------------------------------------
3475 // AdvSIMD indexed element
3476 //----------------------------------------------------------------------------
3478 let neverHasSideEffects = 1 in {
3479 defm FMLA : SIMDFPIndexedSDTied<0, 0b0001, "fmla">;
3480 defm FMLS : SIMDFPIndexedSDTied<0, 0b0101, "fmls">;
3483 // NOTE: Operands are reordered in the FMLA/FMLS PatFrags because the
3484 // instruction expects the addend first, while the intrinsic expects it last.
3486 // On the other hand, there are quite a few valid combinatorial options due to
3487 // the commutativity of multiplication and the fact that (-x) * y = x * (-y).
3488 defm : SIMDFPIndexedSDTiedPatterns<"FMLA",
3489 TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)>>;
3490 defm : SIMDFPIndexedSDTiedPatterns<"FMLA",
3491 TriOpFrag<(fma node:$MHS, node:$RHS, node:$LHS)>>;
3493 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
3494 TriOpFrag<(fma node:$MHS, (fneg node:$RHS), node:$LHS)> >;
3495 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
3496 TriOpFrag<(fma node:$RHS, (fneg node:$MHS), node:$LHS)> >;
3497 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
3498 TriOpFrag<(fma (fneg node:$RHS), node:$MHS, node:$LHS)> >;
3499 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
3500 TriOpFrag<(fma (fneg node:$MHS), node:$RHS, node:$LHS)> >;
3502 multiclass FMLSIndexedAfterNegPatterns<SDPatternOperator OpNode> {
3503 // 3 variants for the .2s version: DUPLANE from 128-bit, DUPLANE from 64-bit
3505 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
3506 (ARM64duplane32 (v4f32 (fneg V128:$Rm)),
3507 VectorIndexS:$idx))),
3508 (FMLSv2i32_indexed V64:$Rd, V64:$Rn, V128:$Rm, VectorIndexS:$idx)>;
3509 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
3510 (v2f32 (ARM64duplane32
3511 (v4f32 (insert_subvector undef,
3512 (v2f32 (fneg V64:$Rm)),
3514 VectorIndexS:$idx)))),
3515 (FMLSv2i32_indexed V64:$Rd, V64:$Rn,
3516 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub),
3517 VectorIndexS:$idx)>;
3518 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
3519 (ARM64dup (f32 (fneg FPR32Op:$Rm))))),
3520 (FMLSv2i32_indexed V64:$Rd, V64:$Rn,
3521 (SUBREG_TO_REG (i32 0), FPR32Op:$Rm, ssub), (i64 0))>;
3523 // 3 variants for the .4s version: DUPLANE from 128-bit, DUPLANE from 64-bit
3525 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
3526 (ARM64duplane32 (v4f32 (fneg V128:$Rm)),
3527 VectorIndexS:$idx))),
3528 (FMLSv4i32_indexed V128:$Rd, V128:$Rn, V128:$Rm,
3529 VectorIndexS:$idx)>;
3530 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
3531 (v4f32 (ARM64duplane32
3532 (v4f32 (insert_subvector undef,
3533 (v2f32 (fneg V64:$Rm)),
3535 VectorIndexS:$idx)))),
3536 (FMLSv4i32_indexed V128:$Rd, V128:$Rn,
3537 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub),
3538 VectorIndexS:$idx)>;
3539 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
3540 (ARM64dup (f32 (fneg FPR32Op:$Rm))))),
3541 (FMLSv4i32_indexed V128:$Rd, V128:$Rn,
3542 (SUBREG_TO_REG (i32 0), FPR32Op:$Rm, ssub), (i64 0))>;
3544 // 2 variants for the .2d version: DUPLANE from 128-bit, and DUP scalar
3545 // (DUPLANE from 64-bit would be trivial).
3546 def : Pat<(v2f64 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn),
3547 (ARM64duplane64 (v2f64 (fneg V128:$Rm)),
3548 VectorIndexD:$idx))),
3550 V128:$Rd, V128:$Rn, V128:$Rm, VectorIndexS:$idx)>;
3551 def : Pat<(v2f64 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn),
3552 (ARM64dup (f64 (fneg FPR64Op:$Rm))))),
3553 (FMLSv2i64_indexed V128:$Rd, V128:$Rn,
3554 (SUBREG_TO_REG (i32 0), FPR64Op:$Rm, dsub), (i64 0))>;
3556 // 2 variants for 32-bit scalar version: extract from .2s or from .4s
3557 def : Pat<(f32 (OpNode (f32 FPR32:$Rd), (f32 FPR32:$Rn),
3558 (vector_extract (v4f32 (fneg V128:$Rm)),
3559 VectorIndexS:$idx))),
3560 (FMLSv1i32_indexed FPR32:$Rd, FPR32:$Rn,
3561 V128:$Rm, VectorIndexS:$idx)>;
3562 def : Pat<(f32 (OpNode (f32 FPR32:$Rd), (f32 FPR32:$Rn),
3563 (vector_extract (v2f32 (fneg V64:$Rm)),
3564 VectorIndexS:$idx))),
3565 (FMLSv1i32_indexed FPR32:$Rd, FPR32:$Rn,
3566 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub), VectorIndexS:$idx)>;
3568 // 1 variant for 64-bit scalar version: extract from .1d or from .2d
3569 def : Pat<(f64 (OpNode (f64 FPR64:$Rd), (f64 FPR64:$Rn),
3570 (vector_extract (v2f64 (fneg V128:$Rm)),
3571 VectorIndexS:$idx))),
3572 (FMLSv1i64_indexed FPR64:$Rd, FPR64:$Rn,
3573 V128:$Rm, VectorIndexS:$idx)>;
3576 defm : FMLSIndexedAfterNegPatterns<
3577 TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)> >;
3578 defm : FMLSIndexedAfterNegPatterns<
3579 TriOpFrag<(fma node:$MHS, node:$RHS, node:$LHS)> >;
3581 defm FMULX : SIMDFPIndexedSD<1, 0b1001, "fmulx", int_arm64_neon_fmulx>;
3582 defm FMUL : SIMDFPIndexedSD<0, 0b1001, "fmul", fmul>;
3584 def : Pat<(v2f32 (fmul V64:$Rn, (ARM64dup (f32 FPR32:$Rm)))),
3585 (FMULv2i32_indexed V64:$Rn,
3586 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rm, ssub),
3588 def : Pat<(v4f32 (fmul V128:$Rn, (ARM64dup (f32 FPR32:$Rm)))),
3589 (FMULv4i32_indexed V128:$Rn,
3590 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rm, ssub),
3592 def : Pat<(v2f64 (fmul V128:$Rn, (ARM64dup (f64 FPR64:$Rm)))),
3593 (FMULv2i64_indexed V128:$Rn,
3594 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$Rm, dsub),
3597 defm SQDMULH : SIMDIndexedHS<0, 0b1100, "sqdmulh", int_arm64_neon_sqdmulh>;
3598 defm SQRDMULH : SIMDIndexedHS<0, 0b1101, "sqrdmulh", int_arm64_neon_sqrdmulh>;
3599 defm MLA : SIMDVectorIndexedHSTied<1, 0b0000, "mla",
3600 TriOpFrag<(add node:$LHS, (mul node:$MHS, node:$RHS))>>;
3601 defm MLS : SIMDVectorIndexedHSTied<1, 0b0100, "mls",
3602 TriOpFrag<(sub node:$LHS, (mul node:$MHS, node:$RHS))>>;
3603 defm MUL : SIMDVectorIndexedHS<0, 0b1000, "mul", mul>;
3604 defm SMLAL : SIMDVectorIndexedLongSDTied<0, 0b0010, "smlal",
3605 TriOpFrag<(add node:$LHS, (int_arm64_neon_smull node:$MHS, node:$RHS))>>;
3606 defm SMLSL : SIMDVectorIndexedLongSDTied<0, 0b0110, "smlsl",
3607 TriOpFrag<(sub node:$LHS, (int_arm64_neon_smull node:$MHS, node:$RHS))>>;
3608 defm SMULL : SIMDVectorIndexedLongSD<0, 0b1010, "smull",
3609 int_arm64_neon_smull>;
3610 defm SQDMLAL : SIMDIndexedLongSQDMLXSDTied<0, 0b0011, "sqdmlal",
3611 int_arm64_neon_sqadd>;
3612 defm SQDMLSL : SIMDIndexedLongSQDMLXSDTied<0, 0b0111, "sqdmlsl",
3613 int_arm64_neon_sqsub>;
3614 defm SQDMULL : SIMDIndexedLongSD<0, 0b1011, "sqdmull", int_arm64_neon_sqdmull>;
3615 defm UMLAL : SIMDVectorIndexedLongSDTied<1, 0b0010, "umlal",
3616 TriOpFrag<(add node:$LHS, (int_arm64_neon_umull node:$MHS, node:$RHS))>>;
3617 defm UMLSL : SIMDVectorIndexedLongSDTied<1, 0b0110, "umlsl",
3618 TriOpFrag<(sub node:$LHS, (int_arm64_neon_umull node:$MHS, node:$RHS))>>;
3619 defm UMULL : SIMDVectorIndexedLongSD<1, 0b1010, "umull",
3620 int_arm64_neon_umull>;
3622 // A scalar sqdmull with the second operand being a vector lane can be
3623 // handled directly with the indexed instruction encoding.
3624 def : Pat<(int_arm64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
3625 (vector_extract (v4i32 V128:$Vm),
3626 VectorIndexS:$idx)),
3627 (SQDMULLv1i64_indexed FPR32:$Rn, V128:$Vm, VectorIndexS:$idx)>;
3629 //----------------------------------------------------------------------------
3630 // AdvSIMD scalar shift instructions
3631 //----------------------------------------------------------------------------
3632 defm FCVTZS : SIMDScalarRShiftSD<0, 0b11111, "fcvtzs">;
3633 defm FCVTZU : SIMDScalarRShiftSD<1, 0b11111, "fcvtzu">;
3634 defm SCVTF : SIMDScalarRShiftSD<0, 0b11100, "scvtf">;
3635 defm UCVTF : SIMDScalarRShiftSD<1, 0b11100, "ucvtf">;
3636 // Codegen patterns for the above. We don't put these directly on the
3637 // instructions because TableGen's type inference can't handle the truth.
3638 // Having the same base pattern for fp <--> int totally freaks it out.
3639 def : Pat<(int_arm64_neon_vcvtfp2fxs FPR32:$Rn, vecshiftR32:$imm),
3640 (FCVTZSs FPR32:$Rn, vecshiftR32:$imm)>;
3641 def : Pat<(int_arm64_neon_vcvtfp2fxu FPR32:$Rn, vecshiftR32:$imm),
3642 (FCVTZUs FPR32:$Rn, vecshiftR32:$imm)>;
3643 def : Pat<(i64 (int_arm64_neon_vcvtfp2fxs (f64 FPR64:$Rn), vecshiftR64:$imm)),
3644 (FCVTZSd FPR64:$Rn, vecshiftR64:$imm)>;
3645 def : Pat<(i64 (int_arm64_neon_vcvtfp2fxu (f64 FPR64:$Rn), vecshiftR64:$imm)),
3646 (FCVTZUd FPR64:$Rn, vecshiftR64:$imm)>;
3647 def : Pat<(v1i64 (int_arm64_neon_vcvtfp2fxs (v1f64 FPR64:$Rn),
3649 (FCVTZSd FPR64:$Rn, vecshiftR64:$imm)>;
3650 def : Pat<(v1i64 (int_arm64_neon_vcvtfp2fxu (v1f64 FPR64:$Rn),
3652 (FCVTZUd FPR64:$Rn, vecshiftR64:$imm)>;
3653 def : Pat<(int_arm64_neon_vcvtfxs2fp FPR32:$Rn, vecshiftR32:$imm),
3654 (SCVTFs FPR32:$Rn, vecshiftR32:$imm)>;
3655 def : Pat<(int_arm64_neon_vcvtfxu2fp FPR32:$Rn, vecshiftR32:$imm),
3656 (UCVTFs FPR32:$Rn, vecshiftR32:$imm)>;
3657 def : Pat<(f64 (int_arm64_neon_vcvtfxs2fp (i64 FPR64:$Rn), vecshiftR64:$imm)),
3658 (SCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
3659 def : Pat<(f64 (int_arm64_neon_vcvtfxu2fp (i64 FPR64:$Rn), vecshiftR64:$imm)),
3660 (UCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
3661 def : Pat<(v1f64 (int_arm64_neon_vcvtfxs2fp (v1i64 FPR64:$Rn),
3663 (SCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
3664 def : Pat<(v1f64 (int_arm64_neon_vcvtfxu2fp (v1i64 FPR64:$Rn),
3666 (UCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
3668 defm SHL : SIMDScalarLShiftD< 0, 0b01010, "shl", ARM64vshl>;
3669 defm SLI : SIMDScalarLShiftDTied<1, 0b01010, "sli">;
3670 defm SQRSHRN : SIMDScalarRShiftBHS< 0, 0b10011, "sqrshrn",
3671 int_arm64_neon_sqrshrn>;
3672 defm SQRSHRUN : SIMDScalarRShiftBHS< 1, 0b10001, "sqrshrun",
3673 int_arm64_neon_sqrshrun>;
3674 defm SQSHLU : SIMDScalarLShiftBHSD<1, 0b01100, "sqshlu", ARM64sqshlui>;
3675 defm SQSHL : SIMDScalarLShiftBHSD<0, 0b01110, "sqshl", ARM64sqshli>;
3676 defm SQSHRN : SIMDScalarRShiftBHS< 0, 0b10010, "sqshrn",
3677 int_arm64_neon_sqshrn>;
3678 defm SQSHRUN : SIMDScalarRShiftBHS< 1, 0b10000, "sqshrun",
3679 int_arm64_neon_sqshrun>;
3680 defm SRI : SIMDScalarRShiftDTied< 1, 0b01000, "sri">;
3681 defm SRSHR : SIMDScalarRShiftD< 0, 0b00100, "srshr", ARM64srshri>;
3682 defm SRSRA : SIMDScalarRShiftDTied< 0, 0b00110, "srsra",
3683 TriOpFrag<(add node:$LHS,
3684 (ARM64srshri node:$MHS, node:$RHS))>>;
3685 defm SSHR : SIMDScalarRShiftD< 0, 0b00000, "sshr", ARM64vashr>;
3686 defm SSRA : SIMDScalarRShiftDTied< 0, 0b00010, "ssra",
3687 TriOpFrag<(add node:$LHS,
3688 (ARM64vashr node:$MHS, node:$RHS))>>;
3689 defm UQRSHRN : SIMDScalarRShiftBHS< 1, 0b10011, "uqrshrn",
3690 int_arm64_neon_uqrshrn>;
3691 defm UQSHL : SIMDScalarLShiftBHSD<1, 0b01110, "uqshl", ARM64uqshli>;
3692 defm UQSHRN : SIMDScalarRShiftBHS< 1, 0b10010, "uqshrn",
3693 int_arm64_neon_uqshrn>;
3694 defm URSHR : SIMDScalarRShiftD< 1, 0b00100, "urshr", ARM64urshri>;
3695 defm URSRA : SIMDScalarRShiftDTied< 1, 0b00110, "ursra",
3696 TriOpFrag<(add node:$LHS,
3697 (ARM64urshri node:$MHS, node:$RHS))>>;
3698 defm USHR : SIMDScalarRShiftD< 1, 0b00000, "ushr", ARM64vlshr>;
3699 defm USRA : SIMDScalarRShiftDTied< 1, 0b00010, "usra",
3700 TriOpFrag<(add node:$LHS,
3701 (ARM64vlshr node:$MHS, node:$RHS))>>;
3703 //----------------------------------------------------------------------------
3704 // AdvSIMD vector shift instructions
3705 //----------------------------------------------------------------------------
3706 defm FCVTZS:SIMDVectorRShiftSD<0, 0b11111, "fcvtzs", int_arm64_neon_vcvtfp2fxs>;
3707 defm FCVTZU:SIMDVectorRShiftSD<1, 0b11111, "fcvtzu", int_arm64_neon_vcvtfp2fxu>;
3708 defm SCVTF: SIMDVectorRShiftSDToFP<0, 0b11100, "scvtf",
3709 int_arm64_neon_vcvtfxs2fp>;
3710 defm RSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10001, "rshrn",
3711 int_arm64_neon_rshrn>;
3712 defm SHL : SIMDVectorLShiftBHSD<0, 0b01010, "shl", ARM64vshl>;
3713 defm SHRN : SIMDVectorRShiftNarrowBHS<0, 0b10000, "shrn",
3714 BinOpFrag<(trunc (ARM64vashr node:$LHS, node:$RHS))>>;
3715 defm SLI : SIMDVectorLShiftBHSDTied<1, 0b01010, "sli", int_arm64_neon_vsli>;
3716 def : Pat<(v1i64 (int_arm64_neon_vsli (v1i64 FPR64:$Rd), (v1i64 FPR64:$Rn),
3717 (i32 vecshiftL64:$imm))),
3718 (SLId FPR64:$Rd, FPR64:$Rn, vecshiftL64:$imm)>;
3719 defm SQRSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10011, "sqrshrn",
3720 int_arm64_neon_sqrshrn>;
3721 defm SQRSHRUN: SIMDVectorRShiftNarrowBHS<1, 0b10001, "sqrshrun",
3722 int_arm64_neon_sqrshrun>;
3723 defm SQSHLU : SIMDVectorLShiftBHSD<1, 0b01100, "sqshlu", ARM64sqshlui>;
3724 defm SQSHL : SIMDVectorLShiftBHSD<0, 0b01110, "sqshl", ARM64sqshli>;
3725 defm SQSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10010, "sqshrn",
3726 int_arm64_neon_sqshrn>;
3727 defm SQSHRUN : SIMDVectorRShiftNarrowBHS<1, 0b10000, "sqshrun",
3728 int_arm64_neon_sqshrun>;
3729 defm SRI : SIMDVectorRShiftBHSDTied<1, 0b01000, "sri", int_arm64_neon_vsri>;
3730 def : Pat<(v1i64 (int_arm64_neon_vsri (v1i64 FPR64:$Rd), (v1i64 FPR64:$Rn),
3731 (i32 vecshiftR64:$imm))),
3732 (SRId FPR64:$Rd, FPR64:$Rn, vecshiftR64:$imm)>;
3733 defm SRSHR : SIMDVectorRShiftBHSD<0, 0b00100, "srshr", ARM64srshri>;
3734 defm SRSRA : SIMDVectorRShiftBHSDTied<0, 0b00110, "srsra",
3735 TriOpFrag<(add node:$LHS,
3736 (ARM64srshri node:$MHS, node:$RHS))> >;
3737 defm SSHLL : SIMDVectorLShiftLongBHSD<0, 0b10100, "sshll",
3738 BinOpFrag<(ARM64vshl (sext node:$LHS), node:$RHS)>>;
3740 defm SSHR : SIMDVectorRShiftBHSD<0, 0b00000, "sshr", ARM64vashr>;
3741 defm SSRA : SIMDVectorRShiftBHSDTied<0, 0b00010, "ssra",
3742 TriOpFrag<(add node:$LHS, (ARM64vashr node:$MHS, node:$RHS))>>;
3743 defm UCVTF : SIMDVectorRShiftSDToFP<1, 0b11100, "ucvtf",
3744 int_arm64_neon_vcvtfxu2fp>;
3745 defm UQRSHRN : SIMDVectorRShiftNarrowBHS<1, 0b10011, "uqrshrn",
3746 int_arm64_neon_uqrshrn>;
3747 defm UQSHL : SIMDVectorLShiftBHSD<1, 0b01110, "uqshl", ARM64uqshli>;
3748 defm UQSHRN : SIMDVectorRShiftNarrowBHS<1, 0b10010, "uqshrn",
3749 int_arm64_neon_uqshrn>;
3750 defm URSHR : SIMDVectorRShiftBHSD<1, 0b00100, "urshr", ARM64urshri>;
3751 defm URSRA : SIMDVectorRShiftBHSDTied<1, 0b00110, "ursra",
3752 TriOpFrag<(add node:$LHS,
3753 (ARM64urshri node:$MHS, node:$RHS))> >;
3754 defm USHLL : SIMDVectorLShiftLongBHSD<1, 0b10100, "ushll",
3755 BinOpFrag<(ARM64vshl (zext node:$LHS), node:$RHS)>>;
3756 defm USHR : SIMDVectorRShiftBHSD<1, 0b00000, "ushr", ARM64vlshr>;
3757 defm USRA : SIMDVectorRShiftBHSDTied<1, 0b00010, "usra",
3758 TriOpFrag<(add node:$LHS, (ARM64vlshr node:$MHS, node:$RHS))> >;
3760 // SHRN patterns for when a logical right shift was used instead of arithmetic
3761 // (the immediate guarantees no sign bits actually end up in the result so it
3763 def : Pat<(v8i8 (trunc (ARM64vlshr (v8i16 V128:$Rn), vecshiftR16Narrow:$imm))),
3764 (SHRNv8i8_shift V128:$Rn, vecshiftR16Narrow:$imm)>;
3765 def : Pat<(v4i16 (trunc (ARM64vlshr (v4i32 V128:$Rn), vecshiftR32Narrow:$imm))),
3766 (SHRNv4i16_shift V128:$Rn, vecshiftR32Narrow:$imm)>;
3767 def : Pat<(v2i32 (trunc (ARM64vlshr (v2i64 V128:$Rn), vecshiftR64Narrow:$imm))),
3768 (SHRNv2i32_shift V128:$Rn, vecshiftR64Narrow:$imm)>;
3770 def : Pat<(v16i8 (concat_vectors (v8i8 V64:$Rd),
3771 (trunc (ARM64vlshr (v8i16 V128:$Rn),
3772 vecshiftR16Narrow:$imm)))),
3773 (SHRNv16i8_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
3774 V128:$Rn, vecshiftR16Narrow:$imm)>;
3775 def : Pat<(v8i16 (concat_vectors (v4i16 V64:$Rd),
3776 (trunc (ARM64vlshr (v4i32 V128:$Rn),
3777 vecshiftR32Narrow:$imm)))),
3778 (SHRNv8i16_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
3779 V128:$Rn, vecshiftR32Narrow:$imm)>;
3780 def : Pat<(v4i32 (concat_vectors (v2i32 V64:$Rd),
3781 (trunc (ARM64vlshr (v2i64 V128:$Rn),
3782 vecshiftR64Narrow:$imm)))),
3783 (SHRNv4i32_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
3784 V128:$Rn, vecshiftR32Narrow:$imm)>;
3786 // Vector sign and zero extensions are implemented with SSHLL and USSHLL.
3787 // Anyexts are implemented as zexts.
3788 def : Pat<(v8i16 (sext (v8i8 V64:$Rn))), (SSHLLv8i8_shift V64:$Rn, (i32 0))>;
3789 def : Pat<(v8i16 (zext (v8i8 V64:$Rn))), (USHLLv8i8_shift V64:$Rn, (i32 0))>;
3790 def : Pat<(v8i16 (anyext (v8i8 V64:$Rn))), (USHLLv8i8_shift V64:$Rn, (i32 0))>;
3791 def : Pat<(v4i32 (sext (v4i16 V64:$Rn))), (SSHLLv4i16_shift V64:$Rn, (i32 0))>;
3792 def : Pat<(v4i32 (zext (v4i16 V64:$Rn))), (USHLLv4i16_shift V64:$Rn, (i32 0))>;
3793 def : Pat<(v4i32 (anyext (v4i16 V64:$Rn))), (USHLLv4i16_shift V64:$Rn, (i32 0))>;
3794 def : Pat<(v2i64 (sext (v2i32 V64:$Rn))), (SSHLLv2i32_shift V64:$Rn, (i32 0))>;
3795 def : Pat<(v2i64 (zext (v2i32 V64:$Rn))), (USHLLv2i32_shift V64:$Rn, (i32 0))>;
3796 def : Pat<(v2i64 (anyext (v2i32 V64:$Rn))), (USHLLv2i32_shift V64:$Rn, (i32 0))>;
3797 // Also match an extend from the upper half of a 128 bit source register.
3798 def : Pat<(v8i16 (anyext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
3799 (USHLLv16i8_shift V128:$Rn, (i32 0))>;
3800 def : Pat<(v8i16 (zext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
3801 (USHLLv16i8_shift V128:$Rn, (i32 0))>;
3802 def : Pat<(v8i16 (sext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
3803 (SSHLLv16i8_shift V128:$Rn, (i32 0))>;
3804 def : Pat<(v4i32 (anyext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
3805 (USHLLv8i16_shift V128:$Rn, (i32 0))>;
3806 def : Pat<(v4i32 (zext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
3807 (USHLLv8i16_shift V128:$Rn, (i32 0))>;
3808 def : Pat<(v4i32 (sext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
3809 (SSHLLv8i16_shift V128:$Rn, (i32 0))>;
3810 def : Pat<(v2i64 (anyext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
3811 (USHLLv4i32_shift V128:$Rn, (i32 0))>;
3812 def : Pat<(v2i64 (zext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
3813 (USHLLv4i32_shift V128:$Rn, (i32 0))>;
3814 def : Pat<(v2i64 (sext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
3815 (SSHLLv4i32_shift V128:$Rn, (i32 0))>;
3817 // Vector shift sxtl aliases
3818 def : InstAlias<"sxtl.8h $dst, $src1",
3819 (SSHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
3820 def : InstAlias<"sxtl $dst.8h, $src1.8b",
3821 (SSHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
3822 def : InstAlias<"sxtl.4s $dst, $src1",
3823 (SSHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
3824 def : InstAlias<"sxtl $dst.4s, $src1.4h",
3825 (SSHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
3826 def : InstAlias<"sxtl.2d $dst, $src1",
3827 (SSHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
3828 def : InstAlias<"sxtl $dst.2d, $src1.2s",
3829 (SSHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
3831 // Vector shift sxtl2 aliases
3832 def : InstAlias<"sxtl2.8h $dst, $src1",
3833 (SSHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
3834 def : InstAlias<"sxtl2 $dst.8h, $src1.16b",
3835 (SSHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
3836 def : InstAlias<"sxtl2.4s $dst, $src1",
3837 (SSHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
3838 def : InstAlias<"sxtl2 $dst.4s, $src1.8h",
3839 (SSHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
3840 def : InstAlias<"sxtl2.2d $dst, $src1",
3841 (SSHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
3842 def : InstAlias<"sxtl2 $dst.2d, $src1.4s",
3843 (SSHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
3845 // Vector shift uxtl aliases
3846 def : InstAlias<"uxtl.8h $dst, $src1",
3847 (USHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
3848 def : InstAlias<"uxtl $dst.8h, $src1.8b",
3849 (USHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
3850 def : InstAlias<"uxtl.4s $dst, $src1",
3851 (USHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
3852 def : InstAlias<"uxtl $dst.4s, $src1.4h",
3853 (USHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
3854 def : InstAlias<"uxtl.2d $dst, $src1",
3855 (USHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
3856 def : InstAlias<"uxtl $dst.2d, $src1.2s",
3857 (USHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
3859 // Vector shift uxtl2 aliases
3860 def : InstAlias<"uxtl2.8h $dst, $src1",
3861 (USHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
3862 def : InstAlias<"uxtl2 $dst.8h, $src1.16b",
3863 (USHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
3864 def : InstAlias<"uxtl2.4s $dst, $src1",
3865 (USHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
3866 def : InstAlias<"uxtl2 $dst.4s, $src1.8h",
3867 (USHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
3868 def : InstAlias<"uxtl2.2d $dst, $src1",
3869 (USHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
3870 def : InstAlias<"uxtl2 $dst.2d, $src1.4s",
3871 (USHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
3873 // If an integer is about to be converted to a floating point value,
3874 // just load it on the floating point unit.
3875 // These patterns are more complex because floating point loads do not
3876 // support sign extension.
3877 // The sign extension has to be explicitly added and is only supported for
3878 // one step: byte-to-half, half-to-word, word-to-doubleword.
3879 // SCVTF GPR -> FPR is 9 cycles.
3880 // SCVTF FPR -> FPR is 4 cyclces.
3881 // (sign extension with lengthen) SXTL FPR -> FPR is 2 cycles.
3882 // Therefore, we can do 2 sign extensions and one SCVTF FPR -> FPR
3883 // and still being faster.
3884 // However, this is not good for code size.
3885 // 8-bits -> float. 2 sizes step-up.
3886 def : Pat <(f32 (sint_to_fp (i32 (sextloadi8 ro_indexed8:$addr)))),
3887 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
3892 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3893 (LDRBro ro_indexed8:$addr),
3898 ssub)))>, Requires<[NotForCodeSize]>;
3899 def : Pat <(f32 (sint_to_fp (i32 (sextloadi8 am_indexed8:$addr)))),
3900 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
3905 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3906 (LDRBui am_indexed8:$addr),
3911 ssub)))>, Requires<[NotForCodeSize]>;
3912 def : Pat <(f32 (sint_to_fp (i32 (sextloadi8 am_unscaled8:$addr)))),
3913 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
3918 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3919 (LDURBi am_unscaled8:$addr),
3924 ssub)))>, Requires<[NotForCodeSize]>;
3925 // 16-bits -> float. 1 size step-up.
3926 def : Pat <(f32 (sint_to_fp (i32 (sextloadi16 ro_indexed16:$addr)))),
3927 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
3929 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3930 (LDRHro ro_indexed16:$addr),
3933 ssub)))>, Requires<[NotForCodeSize]>;
3934 def : Pat <(f32 (sint_to_fp (i32 (sextloadi16 am_indexed16:$addr)))),
3935 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
3937 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3938 (LDRHui am_indexed16:$addr),
3941 ssub)))>, Requires<[NotForCodeSize]>;
3942 def : Pat <(f32 (sint_to_fp (i32 (sextloadi16 am_unscaled16:$addr)))),
3943 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
3945 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3946 (LDURHi am_unscaled16:$addr),
3949 ssub)))>, Requires<[NotForCodeSize]>;
3950 // 32-bits to 32-bits are handled in target specific dag combine:
3951 // performIntToFpCombine.
3952 // 64-bits integer to 32-bits floating point, not possible with
3953 // SCVTF on floating point registers (both source and destination
3954 // must have the same size).
3956 // Here are the patterns for 8, 16, 32, and 64-bits to double.
3957 // 8-bits -> double. 3 size step-up: give up.
3958 // 16-bits -> double. 2 size step.
3959 def : Pat <(f64 (sint_to_fp (i32 (sextloadi16 ro_indexed16:$addr)))),
3960 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
3965 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3966 (LDRHro ro_indexed16:$addr),
3971 dsub)))>, Requires<[NotForCodeSize]>;
3972 def : Pat <(f64 (sint_to_fp (i32 (sextloadi16 am_indexed16:$addr)))),
3973 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
3978 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3979 (LDRHui am_indexed16:$addr),
3984 dsub)))>, Requires<[NotForCodeSize]>;
3985 def : Pat <(f64 (sint_to_fp (i32 (sextloadi16 am_unscaled16:$addr)))),
3986 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
3991 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3992 (LDURHi am_unscaled16:$addr),
3997 dsub)))>, Requires<[NotForCodeSize]>;
3998 // 32-bits -> double. 1 size step-up.
3999 def : Pat <(f64 (sint_to_fp (i32 (load ro_indexed32:$addr)))),
4000 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4002 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4003 (LDRSro ro_indexed32:$addr),
4006 dsub)))>, Requires<[NotForCodeSize]>;
4007 def : Pat <(f64 (sint_to_fp (i32 (load am_indexed32:$addr)))),
4008 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4010 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4011 (LDRSui am_indexed32:$addr),
4014 dsub)))>, Requires<[NotForCodeSize]>;
4015 def : Pat <(f64 (sint_to_fp (i32 (load am_unscaled32:$addr)))),
4016 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4018 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4019 (LDURSi am_unscaled32:$addr),
4022 dsub)))>, Requires<[NotForCodeSize]>;
4023 // 64-bits -> double are handled in target specific dag combine:
4024 // performIntToFpCombine.
4027 //----------------------------------------------------------------------------
4028 // AdvSIMD Load-Store Structure
4029 //----------------------------------------------------------------------------
4030 defm LD1 : SIMDLd1Multiple<"ld1">;
4031 defm LD2 : SIMDLd2Multiple<"ld2">;
4032 defm LD3 : SIMDLd3Multiple<"ld3">;
4033 defm LD4 : SIMDLd4Multiple<"ld4">;
4035 defm ST1 : SIMDSt1Multiple<"st1">;
4036 defm ST2 : SIMDSt2Multiple<"st2">;
4037 defm ST3 : SIMDSt3Multiple<"st3">;
4038 defm ST4 : SIMDSt4Multiple<"st4">;
4040 class Ld1Pat<ValueType ty, Instruction INST>
4041 : Pat<(ty (load am_simdnoindex:$vaddr)), (INST am_simdnoindex:$vaddr)>;
4043 def : Ld1Pat<v16i8, LD1Onev16b>;
4044 def : Ld1Pat<v8i16, LD1Onev8h>;
4045 def : Ld1Pat<v4i32, LD1Onev4s>;
4046 def : Ld1Pat<v2i64, LD1Onev2d>;
4047 def : Ld1Pat<v8i8, LD1Onev8b>;
4048 def : Ld1Pat<v4i16, LD1Onev4h>;
4049 def : Ld1Pat<v2i32, LD1Onev2s>;
4050 def : Ld1Pat<v1i64, LD1Onev1d>;
4052 class St1Pat<ValueType ty, Instruction INST>
4053 : Pat<(store ty:$Vt, am_simdnoindex:$vaddr),
4054 (INST ty:$Vt, am_simdnoindex:$vaddr)>;
4056 def : St1Pat<v16i8, ST1Onev16b>;
4057 def : St1Pat<v8i16, ST1Onev8h>;
4058 def : St1Pat<v4i32, ST1Onev4s>;
4059 def : St1Pat<v2i64, ST1Onev2d>;
4060 def : St1Pat<v8i8, ST1Onev8b>;
4061 def : St1Pat<v4i16, ST1Onev4h>;
4062 def : St1Pat<v2i32, ST1Onev2s>;
4063 def : St1Pat<v1i64, ST1Onev1d>;
4069 defm LD1R : SIMDLdR<0, 0b110, 0, "ld1r", "One", 1, 2, 4, 8>;
4070 defm LD2R : SIMDLdR<1, 0b110, 0, "ld2r", "Two", 2, 4, 8, 16>;
4071 defm LD3R : SIMDLdR<0, 0b111, 0, "ld3r", "Three", 3, 6, 12, 24>;
4072 defm LD4R : SIMDLdR<1, 0b111, 0, "ld4r", "Four", 4, 8, 16, 32>;
4073 let mayLoad = 1, neverHasSideEffects = 1 in {
4074 defm LD1 : SIMDLdSingleBTied<0, 0b000, "ld1", VecListOneb, GPR64pi1>;
4075 defm LD1 : SIMDLdSingleHTied<0, 0b010, 0, "ld1", VecListOneh, GPR64pi2>;
4076 defm LD1 : SIMDLdSingleSTied<0, 0b100, 0b00, "ld1", VecListOnes, GPR64pi4>;
4077 defm LD1 : SIMDLdSingleDTied<0, 0b100, 0b01, "ld1", VecListOned, GPR64pi8>;
4078 defm LD2 : SIMDLdSingleBTied<1, 0b000, "ld2", VecListTwob, GPR64pi2>;
4079 defm LD2 : SIMDLdSingleHTied<1, 0b010, 0, "ld2", VecListTwoh, GPR64pi4>;
4080 defm LD2 : SIMDLdSingleSTied<1, 0b100, 0b00, "ld2", VecListTwos, GPR64pi8>;
4081 defm LD2 : SIMDLdSingleDTied<1, 0b100, 0b01, "ld2", VecListTwod, GPR64pi16>;
4082 defm LD3 : SIMDLdSingleBTied<0, 0b001, "ld3", VecListThreeb, GPR64pi3>;
4083 defm LD3 : SIMDLdSingleHTied<0, 0b011, 0, "ld3", VecListThreeh, GPR64pi6>;
4084 defm LD3 : SIMDLdSingleSTied<0, 0b101, 0b00, "ld3", VecListThrees, GPR64pi12>;
4085 defm LD3 : SIMDLdSingleDTied<0, 0b101, 0b01, "ld3", VecListThreed, GPR64pi24>;
4086 defm LD4 : SIMDLdSingleBTied<1, 0b001, "ld4", VecListFourb, GPR64pi4>;
4087 defm LD4 : SIMDLdSingleHTied<1, 0b011, 0, "ld4", VecListFourh, GPR64pi8>;
4088 defm LD4 : SIMDLdSingleSTied<1, 0b101, 0b00, "ld4", VecListFours, GPR64pi16>;
4089 defm LD4 : SIMDLdSingleDTied<1, 0b101, 0b01, "ld4", VecListFourd, GPR64pi32>;
4092 def : Pat<(v8i8 (ARM64dup (i32 (extloadi8 am_simdnoindex:$vaddr)))),
4093 (LD1Rv8b am_simdnoindex:$vaddr)>;
4094 def : Pat<(v16i8 (ARM64dup (i32 (extloadi8 am_simdnoindex:$vaddr)))),
4095 (LD1Rv16b am_simdnoindex:$vaddr)>;
4096 def : Pat<(v4i16 (ARM64dup (i32 (extloadi16 am_simdnoindex:$vaddr)))),
4097 (LD1Rv4h am_simdnoindex:$vaddr)>;
4098 def : Pat<(v8i16 (ARM64dup (i32 (extloadi16 am_simdnoindex:$vaddr)))),
4099 (LD1Rv8h am_simdnoindex:$vaddr)>;
4100 def : Pat<(v2i32 (ARM64dup (i32 (load am_simdnoindex:$vaddr)))),
4101 (LD1Rv2s am_simdnoindex:$vaddr)>;
4102 def : Pat<(v4i32 (ARM64dup (i32 (load am_simdnoindex:$vaddr)))),
4103 (LD1Rv4s am_simdnoindex:$vaddr)>;
4104 def : Pat<(v2i64 (ARM64dup (i64 (load am_simdnoindex:$vaddr)))),
4105 (LD1Rv2d am_simdnoindex:$vaddr)>;
4106 def : Pat<(v1i64 (ARM64dup (i64 (load am_simdnoindex:$vaddr)))),
4107 (LD1Rv1d am_simdnoindex:$vaddr)>;
4108 // Grab the floating point version too
4109 def : Pat<(v2f32 (ARM64dup (f32 (load am_simdnoindex:$vaddr)))),
4110 (LD1Rv2s am_simdnoindex:$vaddr)>;
4111 def : Pat<(v4f32 (ARM64dup (f32 (load am_simdnoindex:$vaddr)))),
4112 (LD1Rv4s am_simdnoindex:$vaddr)>;
4113 def : Pat<(v2f64 (ARM64dup (f64 (load am_simdnoindex:$vaddr)))),
4114 (LD1Rv2d am_simdnoindex:$vaddr)>;
4115 def : Pat<(v1f64 (ARM64dup (f64 (load am_simdnoindex:$vaddr)))),
4116 (LD1Rv1d am_simdnoindex:$vaddr)>;
4118 class Ld1Lane128Pat<SDPatternOperator scalar_load, Operand VecIndex,
4119 ValueType VTy, ValueType STy, Instruction LD1>
4120 : Pat<(vector_insert (VTy VecListOne128:$Rd),
4121 (STy (scalar_load am_simdnoindex:$vaddr)), VecIndex:$idx),
4122 (LD1 VecListOne128:$Rd, VecIndex:$idx, am_simdnoindex:$vaddr)>;
4124 def : Ld1Lane128Pat<extloadi8, VectorIndexB, v16i8, i32, LD1i8>;
4125 def : Ld1Lane128Pat<extloadi16, VectorIndexH, v8i16, i32, LD1i16>;
4126 def : Ld1Lane128Pat<load, VectorIndexS, v4i32, i32, LD1i32>;
4127 def : Ld1Lane128Pat<load, VectorIndexS, v4f32, f32, LD1i32>;
4128 def : Ld1Lane128Pat<load, VectorIndexD, v2i64, i64, LD1i64>;
4129 def : Ld1Lane128Pat<load, VectorIndexD, v2f64, f64, LD1i64>;
4131 class Ld1Lane64Pat<SDPatternOperator scalar_load, Operand VecIndex,
4132 ValueType VTy, ValueType STy, Instruction LD1>
4133 : Pat<(vector_insert (VTy VecListOne64:$Rd),
4134 (STy (scalar_load am_simdnoindex:$vaddr)), VecIndex:$idx),
4136 (LD1 (SUBREG_TO_REG (i32 0), VecListOne64:$Rd, dsub),
4137 VecIndex:$idx, am_simdnoindex:$vaddr),
4140 def : Ld1Lane64Pat<extloadi8, VectorIndexB, v8i8, i32, LD1i8>;
4141 def : Ld1Lane64Pat<extloadi16, VectorIndexH, v4i16, i32, LD1i16>;
4142 def : Ld1Lane64Pat<load, VectorIndexS, v2i32, i32, LD1i32>;
4143 def : Ld1Lane64Pat<load, VectorIndexS, v2f32, f32, LD1i32>;
4146 defm LD1 : SIMDLdSt1SingleAliases<"ld1">;
4147 defm LD2 : SIMDLdSt2SingleAliases<"ld2">;
4148 defm LD3 : SIMDLdSt3SingleAliases<"ld3">;
4149 defm LD4 : SIMDLdSt4SingleAliases<"ld4">;
4152 defm ST1 : SIMDStSingleB<0, 0b000, "st1", VecListOneb, GPR64pi1>;
4153 defm ST1 : SIMDStSingleH<0, 0b010, 0, "st1", VecListOneh, GPR64pi2>;
4154 defm ST1 : SIMDStSingleS<0, 0b100, 0b00, "st1", VecListOnes, GPR64pi4>;
4155 defm ST1 : SIMDStSingleD<0, 0b100, 0b01, "st1", VecListOned, GPR64pi8>;
4157 let AddedComplexity = 8 in
4158 class St1Lane128Pat<SDPatternOperator scalar_store, Operand VecIndex,
4159 ValueType VTy, ValueType STy, Instruction ST1>
4161 (STy (vector_extract (VTy VecListOne128:$Vt), VecIndex:$idx)),
4162 am_simdnoindex:$vaddr),
4163 (ST1 VecListOne128:$Vt, VecIndex:$idx, am_simdnoindex:$vaddr)>;
4165 def : St1Lane128Pat<truncstorei8, VectorIndexB, v16i8, i32, ST1i8>;
4166 def : St1Lane128Pat<truncstorei16, VectorIndexH, v8i16, i32, ST1i16>;
4167 def : St1Lane128Pat<store, VectorIndexS, v4i32, i32, ST1i32>;
4168 def : St1Lane128Pat<store, VectorIndexS, v4f32, f32, ST1i32>;
4169 def : St1Lane128Pat<store, VectorIndexD, v2i64, i64, ST1i64>;
4170 def : St1Lane128Pat<store, VectorIndexD, v2f64, f64, ST1i64>;
4172 let AddedComplexity = 8 in
4173 class St1Lane64Pat<SDPatternOperator scalar_store, Operand VecIndex,
4174 ValueType VTy, ValueType STy, Instruction ST1>
4176 (STy (vector_extract (VTy VecListOne64:$Vt), VecIndex:$idx)),
4177 am_simdnoindex:$vaddr),
4178 (ST1 (SUBREG_TO_REG (i32 0), VecListOne64:$Vt, dsub),
4179 VecIndex:$idx, am_simdnoindex:$vaddr)>;
4181 def : St1Lane64Pat<truncstorei8, VectorIndexB, v8i8, i32, ST1i8>;
4182 def : St1Lane64Pat<truncstorei16, VectorIndexH, v4i16, i32, ST1i16>;
4183 def : St1Lane64Pat<store, VectorIndexS, v2i32, i32, ST1i32>;
4184 def : St1Lane64Pat<store, VectorIndexS, v2f32, f32, ST1i32>;
4186 let mayStore = 1, neverHasSideEffects = 1 in {
4187 defm ST2 : SIMDStSingleB<1, 0b000, "st2", VecListTwob, GPR64pi2>;
4188 defm ST2 : SIMDStSingleH<1, 0b010, 0, "st2", VecListTwoh, GPR64pi4>;
4189 defm ST2 : SIMDStSingleS<1, 0b100, 0b00, "st2", VecListTwos, GPR64pi8>;
4190 defm ST2 : SIMDStSingleD<1, 0b100, 0b01, "st2", VecListTwod, GPR64pi16>;
4191 defm ST3 : SIMDStSingleB<0, 0b001, "st3", VecListThreeb, GPR64pi3>;
4192 defm ST3 : SIMDStSingleH<0, 0b011, 0, "st3", VecListThreeh, GPR64pi6>;
4193 defm ST3 : SIMDStSingleS<0, 0b101, 0b00, "st3", VecListThrees, GPR64pi12>;
4194 defm ST3 : SIMDStSingleD<0, 0b101, 0b01, "st3", VecListThreed, GPR64pi24>;
4195 defm ST4 : SIMDStSingleB<1, 0b001, "st4", VecListFourb, GPR64pi4>;
4196 defm ST4 : SIMDStSingleH<1, 0b011, 0, "st4", VecListFourh, GPR64pi8>;
4197 defm ST4 : SIMDStSingleS<1, 0b101, 0b00, "st4", VecListFours, GPR64pi16>;
4198 defm ST4 : SIMDStSingleD<1, 0b101, 0b01, "st4", VecListFourd, GPR64pi32>;
4201 defm ST1 : SIMDLdSt1SingleAliases<"st1">;
4202 defm ST2 : SIMDLdSt2SingleAliases<"st2">;
4203 defm ST3 : SIMDLdSt3SingleAliases<"st3">;
4204 defm ST4 : SIMDLdSt4SingleAliases<"st4">;
4206 //----------------------------------------------------------------------------
4207 // Crypto extensions
4208 //----------------------------------------------------------------------------
4210 def AESErr : AESTiedInst<0b0100, "aese", int_arm64_crypto_aese>;
4211 def AESDrr : AESTiedInst<0b0101, "aesd", int_arm64_crypto_aesd>;
4212 def AESMCrr : AESInst< 0b0110, "aesmc", int_arm64_crypto_aesmc>;
4213 def AESIMCrr : AESInst< 0b0111, "aesimc", int_arm64_crypto_aesimc>;
4215 def SHA1Crrr : SHATiedInstQSV<0b000, "sha1c", int_arm64_crypto_sha1c>;
4216 def SHA1Prrr : SHATiedInstQSV<0b001, "sha1p", int_arm64_crypto_sha1p>;
4217 def SHA1Mrrr : SHATiedInstQSV<0b010, "sha1m", int_arm64_crypto_sha1m>;
4218 def SHA1SU0rrr : SHATiedInstVVV<0b011, "sha1su0", int_arm64_crypto_sha1su0>;
4219 def SHA256Hrrr : SHATiedInstQQV<0b100, "sha256h", int_arm64_crypto_sha256h>;
4220 def SHA256H2rrr : SHATiedInstQQV<0b101, "sha256h2",int_arm64_crypto_sha256h2>;
4221 def SHA256SU1rrr :SHATiedInstVVV<0b110, "sha256su1",int_arm64_crypto_sha256su1>;
4223 def SHA1Hrr : SHAInstSS< 0b0000, "sha1h", int_arm64_crypto_sha1h>;
4224 def SHA1SU1rr : SHATiedInstVV<0b0001, "sha1su1", int_arm64_crypto_sha1su1>;
4225 def SHA256SU0rr : SHATiedInstVV<0b0010, "sha256su0",int_arm64_crypto_sha256su0>;
4227 //----------------------------------------------------------------------------
4229 //----------------------------------------------------------------------------
4230 // FIXME: Like for X86, these should go in their own separate .td file.
4232 // Any instruction that defines a 32-bit result leaves the high half of the
4233 // register. Truncate can be lowered to EXTRACT_SUBREG. CopyFromReg may
4234 // be copying from a truncate. But any other 32-bit operation will zero-extend
4236 // FIXME: X86 also checks for CMOV here. Do we need something similar?
4237 def def32 : PatLeaf<(i32 GPR32:$src), [{
4238 return N->getOpcode() != ISD::TRUNCATE &&
4239 N->getOpcode() != TargetOpcode::EXTRACT_SUBREG &&
4240 N->getOpcode() != ISD::CopyFromReg;
4243 // In the case of a 32-bit def that is known to implicitly zero-extend,
4244 // we can use a SUBREG_TO_REG.
4245 def : Pat<(i64 (zext def32:$src)), (SUBREG_TO_REG (i64 0), GPR32:$src, sub_32)>;
4247 // For an anyext, we don't care what the high bits are, so we can perform an
4248 // INSERT_SUBREF into an IMPLICIT_DEF.
4249 def : Pat<(i64 (anyext GPR32:$src)),
4250 (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32)>;
4252 // When we need to explicitly zero-extend, we use an unsigned bitfield move
4253 // instruction (UBFM) on the enclosing super-reg.
4254 def : Pat<(i64 (zext GPR32:$src)),
4255 (UBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32), 0, 31)>;
4257 // To sign extend, we use a signed bitfield move instruction (SBFM) on the
4258 // containing super-reg.
4259 def : Pat<(i64 (sext GPR32:$src)),
4260 (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32), 0, 31)>;
4261 def : Pat<(i64 (sext_inreg GPR64:$src, i32)), (SBFMXri GPR64:$src, 0, 31)>;
4262 def : Pat<(i64 (sext_inreg GPR64:$src, i16)), (SBFMXri GPR64:$src, 0, 15)>;
4263 def : Pat<(i64 (sext_inreg GPR64:$src, i8)), (SBFMXri GPR64:$src, 0, 7)>;
4264 def : Pat<(i64 (sext_inreg GPR64:$src, i1)), (SBFMXri GPR64:$src, 0, 0)>;
4265 def : Pat<(i32 (sext_inreg GPR32:$src, i16)), (SBFMWri GPR32:$src, 0, 15)>;
4266 def : Pat<(i32 (sext_inreg GPR32:$src, i8)), (SBFMWri GPR32:$src, 0, 7)>;
4267 def : Pat<(i32 (sext_inreg GPR32:$src, i1)), (SBFMWri GPR32:$src, 0, 0)>;
4269 def : Pat<(shl (sext_inreg GPR32:$Rn, i8), (i64 imm0_31:$imm)),
4270 (SBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
4271 (i64 (i32shift_sext_i8 imm0_31:$imm)))>;
4272 def : Pat<(shl (sext_inreg GPR64:$Rn, i8), (i64 imm0_63:$imm)),
4273 (SBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
4274 (i64 (i64shift_sext_i8 imm0_63:$imm)))>;
4276 def : Pat<(shl (sext_inreg GPR32:$Rn, i16), (i64 imm0_31:$imm)),
4277 (SBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
4278 (i64 (i32shift_sext_i16 imm0_31:$imm)))>;
4279 def : Pat<(shl (sext_inreg GPR64:$Rn, i16), (i64 imm0_63:$imm)),
4280 (SBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
4281 (i64 (i64shift_sext_i16 imm0_63:$imm)))>;
4283 def : Pat<(shl (i64 (sext GPR32:$Rn)), (i64 imm0_63:$imm)),
4284 (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$Rn, sub_32),
4285 (i64 (i64shift_a imm0_63:$imm)),
4286 (i64 (i64shift_sext_i32 imm0_63:$imm)))>;
4288 // sra patterns have an AddedComplexity of 10, so make sure we have a higher
4289 // AddedComplexity for the following patterns since we want to match sext + sra
4290 // patterns before we attempt to match a single sra node.
4291 let AddedComplexity = 20 in {
4292 // We support all sext + sra combinations which preserve at least one bit of the
4293 // original value which is to be sign extended. E.g. we support shifts up to
4295 def : Pat<(sra (sext_inreg GPR32:$Rn, i8), (i64 imm0_7:$imm)),
4296 (SBFMWri GPR32:$Rn, (i64 imm0_7:$imm), 7)>;
4297 def : Pat<(sra (sext_inreg GPR64:$Rn, i8), (i64 imm0_7:$imm)),
4298 (SBFMXri GPR64:$Rn, (i64 imm0_7:$imm), 7)>;
4300 def : Pat<(sra (sext_inreg GPR32:$Rn, i16), (i64 imm0_15:$imm)),
4301 (SBFMWri GPR32:$Rn, (i64 imm0_15:$imm), 15)>;
4302 def : Pat<(sra (sext_inreg GPR64:$Rn, i16), (i64 imm0_15:$imm)),
4303 (SBFMXri GPR64:$Rn, (i64 imm0_15:$imm), 15)>;
4305 def : Pat<(sra (i64 (sext GPR32:$Rn)), (i64 imm0_31:$imm)),
4306 (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$Rn, sub_32),
4307 (i64 imm0_31:$imm), 31)>;
4308 } // AddedComplexity = 20
4310 // To truncate, we can simply extract from a subregister.
4311 def : Pat<(i32 (trunc GPR64sp:$src)),
4312 (i32 (EXTRACT_SUBREG GPR64sp:$src, sub_32))>;
4314 // __builtin_trap() uses the BRK instruction on ARM64.
4315 def : Pat<(trap), (BRK 1)>;
4317 // Conversions within AdvSIMD types in the same register size are free.
4319 def : Pat<(v1i64 (bitconvert (v2i32 FPR64:$src))), (v1i64 FPR64:$src)>;
4320 def : Pat<(v1i64 (bitconvert (v4i16 FPR64:$src))), (v1i64 FPR64:$src)>;
4321 def : Pat<(v1i64 (bitconvert (v8i8 FPR64:$src))), (v1i64 FPR64:$src)>;
4322 def : Pat<(v1i64 (bitconvert (f64 FPR64:$src))), (v1i64 FPR64:$src)>;
4323 def : Pat<(v1i64 (bitconvert (v2f32 FPR64:$src))), (v1i64 FPR64:$src)>;
4324 def : Pat<(v1i64 (bitconvert (v1f64 FPR64:$src))), (v1i64 FPR64:$src)>;
4326 def : Pat<(v2i32 (bitconvert (v1i64 FPR64:$src))), (v2i32 FPR64:$src)>;
4327 def : Pat<(v2i32 (bitconvert (v4i16 FPR64:$src))), (v2i32 FPR64:$src)>;
4328 def : Pat<(v2i32 (bitconvert (v8i8 FPR64:$src))), (v2i32 FPR64:$src)>;
4329 def : Pat<(v2i32 (bitconvert (f64 FPR64:$src))), (v2i32 FPR64:$src)>;
4330 def : Pat<(v2i32 (bitconvert (v2f32 FPR64:$src))), (v2i32 FPR64:$src)>;
4331 def : Pat<(v2i32 (bitconvert (v1f64 FPR64:$src))), (v2i32 FPR64:$src)>;
4333 def : Pat<(v4i16 (bitconvert (v1i64 FPR64:$src))), (v4i16 FPR64:$src)>;
4334 def : Pat<(v4i16 (bitconvert (v2i32 FPR64:$src))), (v4i16 FPR64:$src)>;
4335 def : Pat<(v4i16 (bitconvert (v8i8 FPR64:$src))), (v4i16 FPR64:$src)>;
4336 def : Pat<(v4i16 (bitconvert (f64 FPR64:$src))), (v4i16 FPR64:$src)>;
4337 def : Pat<(v4i16 (bitconvert (v2f32 FPR64:$src))), (v4i16 FPR64:$src)>;
4338 def : Pat<(v4i16 (bitconvert (v1f64 FPR64:$src))), (v4i16 FPR64:$src)>;
4340 def : Pat<(v8i8 (bitconvert (v1i64 FPR64:$src))), (v8i8 FPR64:$src)>;
4341 def : Pat<(v8i8 (bitconvert (v2i32 FPR64:$src))), (v8i8 FPR64:$src)>;
4342 def : Pat<(v8i8 (bitconvert (v4i16 FPR64:$src))), (v8i8 FPR64:$src)>;
4343 def : Pat<(v8i8 (bitconvert (f64 FPR64:$src))), (v8i8 FPR64:$src)>;
4344 def : Pat<(v8i8 (bitconvert (v2f32 FPR64:$src))), (v8i8 FPR64:$src)>;
4345 def : Pat<(v8i8 (bitconvert (v1f64 FPR64:$src))), (v8i8 FPR64:$src)>;
4347 def : Pat<(f64 (bitconvert (v1i64 FPR64:$src))), (f64 FPR64:$src)>;
4348 def : Pat<(f64 (bitconvert (v2i32 FPR64:$src))), (f64 FPR64:$src)>;
4349 def : Pat<(f64 (bitconvert (v4i16 FPR64:$src))), (f64 FPR64:$src)>;
4350 def : Pat<(f64 (bitconvert (v8i8 FPR64:$src))), (f64 FPR64:$src)>;
4351 def : Pat<(f64 (bitconvert (v2f32 FPR64:$src))), (f64 FPR64:$src)>;
4352 def : Pat<(f64 (bitconvert (v1f64 FPR64:$src))), (f64 FPR64:$src)>;
4354 def : Pat<(v1f64 (bitconvert (v1i64 FPR64:$src))), (v1f64 FPR64:$src)>;
4355 def : Pat<(v1f64 (bitconvert (v2i32 FPR64:$src))), (v1f64 FPR64:$src)>;
4356 def : Pat<(v1f64 (bitconvert (v4i16 FPR64:$src))), (v1f64 FPR64:$src)>;
4357 def : Pat<(v1f64 (bitconvert (v8i8 FPR64:$src))), (v1f64 FPR64:$src)>;
4358 def : Pat<(v1f64 (bitconvert (f64 FPR64:$src))), (v1f64 FPR64:$src)>;
4359 def : Pat<(v1f64 (bitconvert (v2f32 FPR64:$src))), (v1f64 FPR64:$src)>;
4361 def : Pat<(v2f32 (bitconvert (f64 FPR64:$src))), (v2f32 FPR64:$src)>;
4362 def : Pat<(v2f32 (bitconvert (v1i64 FPR64:$src))), (v2f32 FPR64:$src)>;
4363 def : Pat<(v2f32 (bitconvert (v2i32 FPR64:$src))), (v2f32 FPR64:$src)>;
4364 def : Pat<(v2f32 (bitconvert (v4i16 FPR64:$src))), (v2f32 FPR64:$src)>;
4365 def : Pat<(v2f32 (bitconvert (v8i8 FPR64:$src))), (v2f32 FPR64:$src)>;
4366 def : Pat<(v2f32 (bitconvert (v1f64 FPR64:$src))), (v2f32 FPR64:$src)>;
4369 def : Pat<(f128 (bitconvert (v2i64 FPR128:$src))), (f128 FPR128:$src)>;
4370 def : Pat<(f128 (bitconvert (v4i32 FPR128:$src))), (f128 FPR128:$src)>;
4371 def : Pat<(f128 (bitconvert (v8i16 FPR128:$src))), (f128 FPR128:$src)>;
4372 def : Pat<(f128 (bitconvert (v2f64 FPR128:$src))), (f128 FPR128:$src)>;
4373 def : Pat<(f128 (bitconvert (v4f32 FPR128:$src))), (f128 FPR128:$src)>;
4375 def : Pat<(v2f64 (bitconvert (f128 FPR128:$src))), (v2f64 FPR128:$src)>;
4376 def : Pat<(v2f64 (bitconvert (v4i32 FPR128:$src))), (v2f64 FPR128:$src)>;
4377 def : Pat<(v2f64 (bitconvert (v8i16 FPR128:$src))), (v2f64 FPR128:$src)>;
4378 def : Pat<(v2f64 (bitconvert (v16i8 FPR128:$src))), (v2f64 FPR128:$src)>;
4379 def : Pat<(v2f64 (bitconvert (v2i64 FPR128:$src))), (v2f64 FPR128:$src)>;
4380 def : Pat<(v2f64 (bitconvert (v4f32 FPR128:$src))), (v2f64 FPR128:$src)>;
4382 def : Pat<(v4f32 (bitconvert (f128 FPR128:$src))), (v4f32 FPR128:$src)>;
4383 def : Pat<(v4f32 (bitconvert (v4i32 FPR128:$src))), (v4f32 FPR128:$src)>;
4384 def : Pat<(v4f32 (bitconvert (v8i16 FPR128:$src))), (v4f32 FPR128:$src)>;
4385 def : Pat<(v4f32 (bitconvert (v16i8 FPR128:$src))), (v4f32 FPR128:$src)>;
4386 def : Pat<(v4f32 (bitconvert (v2i64 FPR128:$src))), (v4f32 FPR128:$src)>;
4387 def : Pat<(v4f32 (bitconvert (v2f64 FPR128:$src))), (v4f32 FPR128:$src)>;
4389 def : Pat<(v2i64 (bitconvert (f128 FPR128:$src))), (v2i64 FPR128:$src)>;
4390 def : Pat<(v2i64 (bitconvert (v4i32 FPR128:$src))), (v2i64 FPR128:$src)>;
4391 def : Pat<(v2i64 (bitconvert (v8i16 FPR128:$src))), (v2i64 FPR128:$src)>;
4392 def : Pat<(v2i64 (bitconvert (v16i8 FPR128:$src))), (v2i64 FPR128:$src)>;
4393 def : Pat<(v2i64 (bitconvert (v2f64 FPR128:$src))), (v2i64 FPR128:$src)>;
4394 def : Pat<(v2i64 (bitconvert (v4f32 FPR128:$src))), (v2i64 FPR128:$src)>;
4396 def : Pat<(v4i32 (bitconvert (f128 FPR128:$src))), (v4i32 FPR128:$src)>;
4397 def : Pat<(v4i32 (bitconvert (v2i64 FPR128:$src))), (v4i32 FPR128:$src)>;
4398 def : Pat<(v4i32 (bitconvert (v8i16 FPR128:$src))), (v4i32 FPR128:$src)>;
4399 def : Pat<(v4i32 (bitconvert (v16i8 FPR128:$src))), (v4i32 FPR128:$src)>;
4400 def : Pat<(v4i32 (bitconvert (v2f64 FPR128:$src))), (v4i32 FPR128:$src)>;
4401 def : Pat<(v4i32 (bitconvert (v4f32 FPR128:$src))), (v4i32 FPR128:$src)>;
4403 def : Pat<(v8i16 (bitconvert (f128 FPR128:$src))), (v8i16 FPR128:$src)>;
4404 def : Pat<(v8i16 (bitconvert (v2i64 FPR128:$src))), (v8i16 FPR128:$src)>;
4405 def : Pat<(v8i16 (bitconvert (v4i32 FPR128:$src))), (v8i16 FPR128:$src)>;
4406 def : Pat<(v8i16 (bitconvert (v16i8 FPR128:$src))), (v8i16 FPR128:$src)>;
4407 def : Pat<(v8i16 (bitconvert (v2f64 FPR128:$src))), (v8i16 FPR128:$src)>;
4408 def : Pat<(v8i16 (bitconvert (v4f32 FPR128:$src))), (v8i16 FPR128:$src)>;
4410 def : Pat<(v16i8 (bitconvert (f128 FPR128:$src))), (v16i8 FPR128:$src)>;
4411 def : Pat<(v16i8 (bitconvert (v2i64 FPR128:$src))), (v16i8 FPR128:$src)>;
4412 def : Pat<(v16i8 (bitconvert (v4i32 FPR128:$src))), (v16i8 FPR128:$src)>;
4413 def : Pat<(v16i8 (bitconvert (v8i16 FPR128:$src))), (v16i8 FPR128:$src)>;
4414 def : Pat<(v16i8 (bitconvert (v2f64 FPR128:$src))), (v16i8 FPR128:$src)>;
4415 def : Pat<(v16i8 (bitconvert (v4f32 FPR128:$src))), (v16i8 FPR128:$src)>;
4417 def : Pat<(v8i8 (extract_subvector (v16i8 FPR128:$Rn), (i64 1))),
4418 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
4419 def : Pat<(v4i16 (extract_subvector (v8i16 FPR128:$Rn), (i64 1))),
4420 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
4421 def : Pat<(v2i32 (extract_subvector (v4i32 FPR128:$Rn), (i64 1))),
4422 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
4423 def : Pat<(v1i64 (extract_subvector (v2i64 FPR128:$Rn), (i64 1))),
4424 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
4426 // A 64-bit subvector insert to the first 128-bit vector position
4427 // is a subregister copy that needs no instruction.
4428 def : Pat<(insert_subvector undef, (v1i64 FPR64:$src), (i32 0)),
4429 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
4430 def : Pat<(insert_subvector undef, (v1f64 FPR64:$src), (i32 0)),
4431 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
4432 def : Pat<(insert_subvector undef, (v2i32 FPR64:$src), (i32 0)),
4433 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
4434 def : Pat<(insert_subvector undef, (v2f32 FPR64:$src), (i32 0)),
4435 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
4436 def : Pat<(insert_subvector undef, (v4i16 FPR64:$src), (i32 0)),
4437 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
4438 def : Pat<(insert_subvector undef, (v8i8 FPR64:$src), (i32 0)),
4439 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
4441 // Use pair-wise add instructions when summing up the lanes for v2f64, v2i64
4443 def : Pat<(i64 (add (vector_extract (v2i64 FPR128:$Rn), (i64 0)),
4444 (vector_extract (v2i64 FPR128:$Rn), (i64 1)))),
4445 (i64 (ADDPv2i64p (v2i64 FPR128:$Rn)))>;
4446 def : Pat<(f64 (fadd (vector_extract (v2f64 FPR128:$Rn), (i64 0)),
4447 (vector_extract (v2f64 FPR128:$Rn), (i64 1)))),
4448 (f64 (FADDPv2i64p (v2f64 FPR128:$Rn)))>;
4449 // vector_extract on 64-bit vectors gets promoted to a 128 bit vector,
4450 // so we match on v4f32 here, not v2f32. This will also catch adding
4451 // the low two lanes of a true v4f32 vector.
4452 def : Pat<(fadd (vector_extract (v4f32 FPR128:$Rn), (i64 0)),
4453 (vector_extract (v4f32 FPR128:$Rn), (i64 1))),
4454 (f32 (FADDPv2i32p (EXTRACT_SUBREG FPR128:$Rn, dsub)))>;
4456 // Scalar 64-bit shifts in FPR64 registers.
4457 def : Pat<(i64 (int_arm64_neon_sshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
4458 (SSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
4459 def : Pat<(i64 (int_arm64_neon_ushl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
4460 (USHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
4461 def : Pat<(i64 (int_arm64_neon_srshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
4462 (SRSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
4463 def : Pat<(i64 (int_arm64_neon_urshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
4464 (URSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
4466 // Tail call return handling. These are all compiler pseudo-instructions,
4467 // so no encoding information or anything like that.
4468 let isCall = 1, isTerminator = 1, isReturn = 1, isBarrier = 1, Uses = [SP] in {
4469 def TCRETURNdi : Pseudo<(outs), (ins i64imm:$dst), []>;
4470 def TCRETURNri : Pseudo<(outs), (ins tcGPR64:$dst), []>;
4473 def : Pat<(ARM64tcret tcGPR64:$dst), (TCRETURNri tcGPR64:$dst)>;
4474 def : Pat<(ARM64tcret (i64 tglobaladdr:$dst)), (TCRETURNdi texternalsym:$dst)>;
4475 def : Pat<(ARM64tcret (i64 texternalsym:$dst)), (TCRETURNdi texternalsym:$dst)>;
4477 include "ARM64InstrAtomics.td"