1 //===- ARM64InstrInfo.td - Describe the ARM64 Instructions -*- tablegen -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // ARM64 Instruction definitions.
12 //===----------------------------------------------------------------------===//
14 //===----------------------------------------------------------------------===//
15 // ARM64-specific DAG Nodes.
18 // SDTBinaryArithWithFlagsOut - RES1, FLAGS = op LHS, RHS
19 def SDTBinaryArithWithFlagsOut : SDTypeProfile<2, 2,
22 SDTCisInt<0>, SDTCisVT<1, i32>]>;
24 // SDTBinaryArithWithFlagsIn - RES1, FLAGS = op LHS, RHS, FLAGS
25 def SDTBinaryArithWithFlagsIn : SDTypeProfile<1, 3,
31 // SDTBinaryArithWithFlagsInOut - RES1, FLAGS = op LHS, RHS, FLAGS
32 def SDTBinaryArithWithFlagsInOut : SDTypeProfile<2, 3,
39 def SDT_ARM64Brcond : SDTypeProfile<0, 3,
40 [SDTCisVT<0, OtherVT>, SDTCisVT<1, i32>,
42 def SDT_ARM64cbz : SDTypeProfile<0, 2, [SDTCisInt<0>, SDTCisVT<1, OtherVT>]>;
43 def SDT_ARM64tbz : SDTypeProfile<0, 3, [SDTCisVT<0, i64>, SDTCisVT<1, i64>,
44 SDTCisVT<2, OtherVT>]>;
47 def SDT_ARM64CSel : SDTypeProfile<1, 4,
52 def SDT_ARM64FCmp : SDTypeProfile<0, 2,
55 def SDT_ARM64Dup : SDTypeProfile<1, 1, [SDTCisVec<0>]>;
56 def SDT_ARM64DupLane : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisInt<2>]>;
57 def SDT_ARM64Zip : SDTypeProfile<1, 2, [SDTCisVec<0>,
60 def SDT_ARM64MOVIedit : SDTypeProfile<1, 1, [SDTCisInt<1>]>;
61 def SDT_ARM64MOVIshift : SDTypeProfile<1, 2, [SDTCisInt<1>, SDTCisInt<2>]>;
62 def SDT_ARM64vecimm : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0,1>,
63 SDTCisInt<2>, SDTCisInt<3>]>;
64 def SDT_ARM64UnaryVec: SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0,1>]>;
65 def SDT_ARM64ExtVec: SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0,1>,
66 SDTCisSameAs<0,2>, SDTCisInt<3>]>;
67 def SDT_ARM64vshift : SDTypeProfile<1, 2, [SDTCisSameAs<0,1>, SDTCisInt<2>]>;
69 def SDT_ARM64unvec : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0,1>]>;
70 def SDT_ARM64fcmpz : SDTypeProfile<1, 1, []>;
71 def SDT_ARM64fcmp : SDTypeProfile<1, 2, [SDTCisSameAs<1,2>]>;
72 def SDT_ARM64binvec : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisSameAs<0,1>,
74 def SDT_ARM64trivec : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0,1>,
77 def SDT_ARM64TCRET : SDTypeProfile<0, 1, [SDTCisPtrTy<0>]>;
78 def SDT_ARM64PREFETCH : SDTypeProfile<0, 2, [SDTCisVT<0, i32>, SDTCisPtrTy<1>]>;
80 def SDT_ARM64ITOF : SDTypeProfile<1, 1, [SDTCisFP<0>, SDTCisSameAs<0,1>]>;
82 def SDT_ARM64TLSDescCall : SDTypeProfile<0, -2, [SDTCisPtrTy<0>,
84 def SDT_ARM64WrapperLarge : SDTypeProfile<1, 4,
85 [SDTCisVT<0, i64>, SDTCisVT<1, i32>,
86 SDTCisSameAs<1, 2>, SDTCisSameAs<1, 3>,
91 def ARM64adrp : SDNode<"ARM64ISD::ADRP", SDTIntUnaryOp, []>;
92 def ARM64addlow : SDNode<"ARM64ISD::ADDlow", SDTIntBinOp, []>;
93 def ARM64LOADgot : SDNode<"ARM64ISD::LOADgot", SDTIntUnaryOp>;
94 def ARM64callseq_start : SDNode<"ISD::CALLSEQ_START",
95 SDCallSeqStart<[ SDTCisVT<0, i32> ]>,
96 [SDNPHasChain, SDNPOutGlue]>;
97 def ARM64callseq_end : SDNode<"ISD::CALLSEQ_END",
98 SDCallSeqEnd<[ SDTCisVT<0, i32>,
100 [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue]>;
101 def ARM64call : SDNode<"ARM64ISD::CALL",
102 SDTypeProfile<0, -1, [SDTCisPtrTy<0>]>,
103 [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue,
105 def ARM64brcond : SDNode<"ARM64ISD::BRCOND", SDT_ARM64Brcond,
107 def ARM64cbz : SDNode<"ARM64ISD::CBZ", SDT_ARM64cbz,
109 def ARM64cbnz : SDNode<"ARM64ISD::CBNZ", SDT_ARM64cbz,
111 def ARM64tbz : SDNode<"ARM64ISD::TBZ", SDT_ARM64tbz,
113 def ARM64tbnz : SDNode<"ARM64ISD::TBNZ", SDT_ARM64tbz,
117 def ARM64csel : SDNode<"ARM64ISD::CSEL", SDT_ARM64CSel>;
118 def ARM64csinv : SDNode<"ARM64ISD::CSINV", SDT_ARM64CSel>;
119 def ARM64csneg : SDNode<"ARM64ISD::CSNEG", SDT_ARM64CSel>;
120 def ARM64csinc : SDNode<"ARM64ISD::CSINC", SDT_ARM64CSel>;
121 def ARM64retflag : SDNode<"ARM64ISD::RET_FLAG", SDTNone,
122 [SDNPHasChain, SDNPOptInGlue, SDNPVariadic]>;
123 def ARM64adc : SDNode<"ARM64ISD::ADC", SDTBinaryArithWithFlagsIn >;
124 def ARM64sbc : SDNode<"ARM64ISD::SBC", SDTBinaryArithWithFlagsIn>;
125 def ARM64add_flag : SDNode<"ARM64ISD::ADDS", SDTBinaryArithWithFlagsOut,
127 def ARM64sub_flag : SDNode<"ARM64ISD::SUBS", SDTBinaryArithWithFlagsOut>;
128 def ARM64and_flag : SDNode<"ARM64ISD::ANDS", SDTBinaryArithWithFlagsOut>;
129 def ARM64adc_flag : SDNode<"ARM64ISD::ADCS", SDTBinaryArithWithFlagsInOut>;
130 def ARM64sbc_flag : SDNode<"ARM64ISD::SBCS", SDTBinaryArithWithFlagsInOut>;
132 def ARM64threadpointer : SDNode<"ARM64ISD::THREAD_POINTER", SDTPtrLeaf>;
134 def ARM64fcmp : SDNode<"ARM64ISD::FCMP", SDT_ARM64FCmp>;
136 def ARM64fmax : SDNode<"ARM64ISD::FMAX", SDTFPBinOp>;
137 def ARM64fmin : SDNode<"ARM64ISD::FMIN", SDTFPBinOp>;
139 def ARM64dup : SDNode<"ARM64ISD::DUP", SDT_ARM64Dup>;
140 def ARM64duplane8 : SDNode<"ARM64ISD::DUPLANE8", SDT_ARM64DupLane>;
141 def ARM64duplane16 : SDNode<"ARM64ISD::DUPLANE16", SDT_ARM64DupLane>;
142 def ARM64duplane32 : SDNode<"ARM64ISD::DUPLANE32", SDT_ARM64DupLane>;
143 def ARM64duplane64 : SDNode<"ARM64ISD::DUPLANE64", SDT_ARM64DupLane>;
145 def ARM64zip1 : SDNode<"ARM64ISD::ZIP1", SDT_ARM64Zip>;
146 def ARM64zip2 : SDNode<"ARM64ISD::ZIP2", SDT_ARM64Zip>;
147 def ARM64uzp1 : SDNode<"ARM64ISD::UZP1", SDT_ARM64Zip>;
148 def ARM64uzp2 : SDNode<"ARM64ISD::UZP2", SDT_ARM64Zip>;
149 def ARM64trn1 : SDNode<"ARM64ISD::TRN1", SDT_ARM64Zip>;
150 def ARM64trn2 : SDNode<"ARM64ISD::TRN2", SDT_ARM64Zip>;
152 def ARM64movi_edit : SDNode<"ARM64ISD::MOVIedit", SDT_ARM64MOVIedit>;
153 def ARM64movi_shift : SDNode<"ARM64ISD::MOVIshift", SDT_ARM64MOVIshift>;
154 def ARM64movi_msl : SDNode<"ARM64ISD::MOVImsl", SDT_ARM64MOVIshift>;
155 def ARM64mvni_shift : SDNode<"ARM64ISD::MVNIshift", SDT_ARM64MOVIshift>;
156 def ARM64mvni_msl : SDNode<"ARM64ISD::MVNImsl", SDT_ARM64MOVIshift>;
157 def ARM64movi : SDNode<"ARM64ISD::MOVI", SDT_ARM64MOVIedit>;
158 def ARM64fmov : SDNode<"ARM64ISD::FMOV", SDT_ARM64MOVIedit>;
160 def ARM64rev16 : SDNode<"ARM64ISD::REV16", SDT_ARM64UnaryVec>;
161 def ARM64rev32 : SDNode<"ARM64ISD::REV32", SDT_ARM64UnaryVec>;
162 def ARM64rev64 : SDNode<"ARM64ISD::REV64", SDT_ARM64UnaryVec>;
163 def ARM64ext : SDNode<"ARM64ISD::EXT", SDT_ARM64ExtVec>;
165 def ARM64vashr : SDNode<"ARM64ISD::VASHR", SDT_ARM64vshift>;
166 def ARM64vlshr : SDNode<"ARM64ISD::VLSHR", SDT_ARM64vshift>;
167 def ARM64vshl : SDNode<"ARM64ISD::VSHL", SDT_ARM64vshift>;
168 def ARM64sqshli : SDNode<"ARM64ISD::SQSHL_I", SDT_ARM64vshift>;
169 def ARM64uqshli : SDNode<"ARM64ISD::UQSHL_I", SDT_ARM64vshift>;
170 def ARM64sqshlui : SDNode<"ARM64ISD::SQSHLU_I", SDT_ARM64vshift>;
171 def ARM64srshri : SDNode<"ARM64ISD::SRSHR_I", SDT_ARM64vshift>;
172 def ARM64urshri : SDNode<"ARM64ISD::URSHR_I", SDT_ARM64vshift>;
174 def ARM64not: SDNode<"ARM64ISD::NOT", SDT_ARM64unvec>;
175 def ARM64bit: SDNode<"ARM64ISD::BIT", SDT_ARM64trivec>;
177 def ARM64cmeq: SDNode<"ARM64ISD::CMEQ", SDT_ARM64binvec>;
178 def ARM64cmge: SDNode<"ARM64ISD::CMGE", SDT_ARM64binvec>;
179 def ARM64cmgt: SDNode<"ARM64ISD::CMGT", SDT_ARM64binvec>;
180 def ARM64cmhi: SDNode<"ARM64ISD::CMHI", SDT_ARM64binvec>;
181 def ARM64cmhs: SDNode<"ARM64ISD::CMHS", SDT_ARM64binvec>;
183 def ARM64fcmeq: SDNode<"ARM64ISD::FCMEQ", SDT_ARM64fcmp>;
184 def ARM64fcmge: SDNode<"ARM64ISD::FCMGE", SDT_ARM64fcmp>;
185 def ARM64fcmgt: SDNode<"ARM64ISD::FCMGT", SDT_ARM64fcmp>;
187 def ARM64cmeqz: SDNode<"ARM64ISD::CMEQz", SDT_ARM64unvec>;
188 def ARM64cmgez: SDNode<"ARM64ISD::CMGEz", SDT_ARM64unvec>;
189 def ARM64cmgtz: SDNode<"ARM64ISD::CMGTz", SDT_ARM64unvec>;
190 def ARM64cmlez: SDNode<"ARM64ISD::CMLEz", SDT_ARM64unvec>;
191 def ARM64cmltz: SDNode<"ARM64ISD::CMLTz", SDT_ARM64unvec>;
192 def ARM64cmtst : PatFrag<(ops node:$LHS, node:$RHS),
193 (ARM64not (ARM64cmeqz (and node:$LHS, node:$RHS)))>;
195 def ARM64fcmeqz: SDNode<"ARM64ISD::FCMEQz", SDT_ARM64fcmpz>;
196 def ARM64fcmgez: SDNode<"ARM64ISD::FCMGEz", SDT_ARM64fcmpz>;
197 def ARM64fcmgtz: SDNode<"ARM64ISD::FCMGTz", SDT_ARM64fcmpz>;
198 def ARM64fcmlez: SDNode<"ARM64ISD::FCMLEz", SDT_ARM64fcmpz>;
199 def ARM64fcmltz: SDNode<"ARM64ISD::FCMLTz", SDT_ARM64fcmpz>;
201 def ARM64bici: SDNode<"ARM64ISD::BICi", SDT_ARM64vecimm>;
202 def ARM64orri: SDNode<"ARM64ISD::ORRi", SDT_ARM64vecimm>;
204 def ARM64neg : SDNode<"ARM64ISD::NEG", SDT_ARM64unvec>;
206 def ARM64tcret: SDNode<"ARM64ISD::TC_RETURN", SDT_ARM64TCRET,
207 [SDNPHasChain, SDNPOptInGlue, SDNPVariadic]>;
209 def ARM64Prefetch : SDNode<"ARM64ISD::PREFETCH", SDT_ARM64PREFETCH,
210 [SDNPHasChain, SDNPSideEffect]>;
212 def ARM64sitof: SDNode<"ARM64ISD::SITOF", SDT_ARM64ITOF>;
213 def ARM64uitof: SDNode<"ARM64ISD::UITOF", SDT_ARM64ITOF>;
215 def ARM64tlsdesc_call : SDNode<"ARM64ISD::TLSDESC_CALL", SDT_ARM64TLSDescCall,
216 [SDNPInGlue, SDNPOutGlue, SDNPHasChain,
219 def ARM64WrapperLarge : SDNode<"ARM64ISD::WrapperLarge", SDT_ARM64WrapperLarge>;
222 //===----------------------------------------------------------------------===//
224 //===----------------------------------------------------------------------===//
226 // ARM64 Instruction Predicate Definitions.
228 def HasZCZ : Predicate<"Subtarget->hasZeroCycleZeroing()">;
229 def NoZCZ : Predicate<"!Subtarget->hasZeroCycleZeroing()">;
230 def IsDarwin : Predicate<"Subtarget->isTargetDarwin()">;
231 def IsNotDarwin: Predicate<"!Subtarget->isTargetDarwin()">;
232 def ForCodeSize : Predicate<"ForCodeSize">;
233 def NotForCodeSize : Predicate<"!ForCodeSize">;
235 include "ARM64InstrFormats.td"
237 //===----------------------------------------------------------------------===//
239 //===----------------------------------------------------------------------===//
240 // Miscellaneous instructions.
241 //===----------------------------------------------------------------------===//
243 let Defs = [SP], Uses = [SP], hasSideEffects = 1, isCodeGenOnly = 1 in {
244 def ADJCALLSTACKDOWN : Pseudo<(outs), (ins i32imm:$amt),
245 [(ARM64callseq_start timm:$amt)]>;
246 def ADJCALLSTACKUP : Pseudo<(outs), (ins i32imm:$amt1, i32imm:$amt2),
247 [(ARM64callseq_end timm:$amt1, timm:$amt2)]>;
248 } // Defs = [SP], Uses = [SP], hasSideEffects = 1, isCodeGenOnly = 1
250 let isReMaterializable = 1, isCodeGenOnly = 1 in {
251 // FIXME: The following pseudo instructions are only needed because remat
252 // cannot handle multiple instructions. When that changes, they can be
253 // removed, along with the ARM64Wrapper node.
255 let AddedComplexity = 10 in
256 def LOADgot : Pseudo<(outs GPR64:$dst), (ins i64imm:$addr),
257 [(set GPR64:$dst, (ARM64LOADgot tglobaladdr:$addr))]>,
260 // The MOVaddr instruction should match only when the add is not folded
261 // into a load or store address.
263 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
264 [(set GPR64:$dst, (ARM64addlow (ARM64adrp tglobaladdr:$hi),
265 tglobaladdr:$low))]>,
266 Sched<[WriteAdrAdr]>;
268 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
269 [(set GPR64:$dst, (ARM64addlow (ARM64adrp tjumptable:$hi),
271 Sched<[WriteAdrAdr]>;
273 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
274 [(set GPR64:$dst, (ARM64addlow (ARM64adrp tconstpool:$hi),
276 Sched<[WriteAdrAdr]>;
278 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
279 [(set GPR64:$dst, (ARM64addlow (ARM64adrp tblockaddress:$hi),
280 tblockaddress:$low))]>,
281 Sched<[WriteAdrAdr]>;
283 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
284 [(set GPR64:$dst, (ARM64addlow (ARM64adrp tglobaltlsaddr:$hi),
285 tglobaltlsaddr:$low))]>,
286 Sched<[WriteAdrAdr]>;
288 : Pseudo<(outs GPR64:$dst), (ins i64imm:$hi, i64imm:$low),
289 [(set GPR64:$dst, (ARM64addlow (ARM64adrp texternalsym:$hi),
290 texternalsym:$low))]>,
291 Sched<[WriteAdrAdr]>;
293 } // isReMaterializable, isCodeGenOnly
295 def : Pat<(ARM64LOADgot tglobaltlsaddr:$addr),
296 (LOADgot tglobaltlsaddr:$addr)>;
298 def : Pat<(ARM64LOADgot texternalsym:$addr),
299 (LOADgot texternalsym:$addr)>;
301 def : Pat<(ARM64LOADgot tconstpool:$addr),
302 (LOADgot tconstpool:$addr)>;
304 //===----------------------------------------------------------------------===//
305 // System instructions.
306 //===----------------------------------------------------------------------===//
308 def HINT : HintI<"hint">;
309 def : InstAlias<"nop", (HINT 0b000)>;
310 def : InstAlias<"yield",(HINT 0b001)>;
311 def : InstAlias<"wfe", (HINT 0b010)>;
312 def : InstAlias<"wfi", (HINT 0b011)>;
313 def : InstAlias<"sev", (HINT 0b100)>;
314 def : InstAlias<"sevl", (HINT 0b101)>;
316 // As far as LLVM is concerned this writes to the system's exclusive monitors.
317 let mayLoad = 1, mayStore = 1 in
318 def CLREX : CRmSystemI<imm0_15, 0b010, "clrex">;
320 def DMB : CRmSystemI<barrier_op, 0b101, "dmb">;
321 def DSB : CRmSystemI<barrier_op, 0b100, "dsb">;
322 def ISB : CRmSystemI<barrier_op, 0b110, "isb">;
323 def : InstAlias<"clrex", (CLREX 0xf)>;
324 def : InstAlias<"isb", (ISB 0xf)>;
328 def MSRcpsr: MSRcpsrI;
330 // The thread pointer (on Linux, at least, where this has been implemented) is
332 def : Pat<(ARM64threadpointer), (MRS 0xde82)>;
334 // Generic system instructions
335 def SYSxt : SystemXtI<0, "sys">;
336 def SYSLxt : SystemLXtI<1, "sysl">;
338 def : InstAlias<"sys $op1, $Cn, $Cm, $op2",
339 (SYSxt imm0_7:$op1, sys_cr_op:$Cn,
340 sys_cr_op:$Cm, imm0_7:$op2, XZR)>;
342 //===----------------------------------------------------------------------===//
343 // Move immediate instructions.
344 //===----------------------------------------------------------------------===//
346 defm MOVK : InsertImmediate<0b11, "movk">;
347 defm MOVN : MoveImmediate<0b00, "movn">;
349 let PostEncoderMethod = "fixMOVZ" in
350 defm MOVZ : MoveImmediate<0b10, "movz">;
352 def : InstAlias<"movk $dst, $imm", (MOVKWi GPR32:$dst, imm0_65535:$imm, 0)>;
353 def : InstAlias<"movk $dst, $imm", (MOVKXi GPR64:$dst, imm0_65535:$imm, 0)>;
354 def : InstAlias<"movn $dst, $imm", (MOVNWi GPR32:$dst, imm0_65535:$imm, 0)>;
355 def : InstAlias<"movn $dst, $imm", (MOVNXi GPR64:$dst, imm0_65535:$imm, 0)>;
356 def : InstAlias<"movz $dst, $imm", (MOVZWi GPR32:$dst, imm0_65535:$imm, 0)>;
357 def : InstAlias<"movz $dst, $imm", (MOVZXi GPR64:$dst, imm0_65535:$imm, 0)>;
359 def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movz_symbol_g3:$sym, 48)>;
360 def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movz_symbol_g2:$sym, 32)>;
361 def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movz_symbol_g1:$sym, 16)>;
362 def : InstAlias<"movz $Rd, $sym", (MOVZXi GPR64:$Rd, movz_symbol_g0:$sym, 0)>;
364 def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movz_symbol_g3:$sym, 48)>;
365 def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movz_symbol_g2:$sym, 32)>;
366 def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movz_symbol_g1:$sym, 16)>;
367 def : InstAlias<"movn $Rd, $sym", (MOVNXi GPR64:$Rd, movz_symbol_g0:$sym, 0)>;
369 def : InstAlias<"movz $Rd, $sym", (MOVZWi GPR32:$Rd, movz_symbol_g3:$sym, 48)>;
370 def : InstAlias<"movz $Rd, $sym", (MOVZWi GPR32:$Rd, movz_symbol_g2:$sym, 32)>;
371 def : InstAlias<"movz $Rd, $sym", (MOVZWi GPR32:$Rd, movz_symbol_g1:$sym, 16)>;
372 def : InstAlias<"movz $Rd, $sym", (MOVZWi GPR32:$Rd, movz_symbol_g0:$sym, 0)>;
374 def : InstAlias<"movk $Rd, $sym", (MOVKXi GPR64:$Rd, movk_symbol_g2:$sym, 32)>;
375 def : InstAlias<"movk $Rd, $sym", (MOVKXi GPR64:$Rd, movk_symbol_g1:$sym, 16)>;
376 def : InstAlias<"movk $Rd, $sym", (MOVKXi GPR64:$Rd, movk_symbol_g0:$sym, 0)>;
378 def : InstAlias<"movk $Rd, $sym", (MOVKWi GPR32:$Rd, movk_symbol_g2:$sym, 32)>;
379 def : InstAlias<"movk $Rd, $sym", (MOVKWi GPR32:$Rd, movk_symbol_g1:$sym, 16)>;
380 def : InstAlias<"movk $Rd, $sym", (MOVKWi GPR32:$Rd, movk_symbol_g0:$sym, 0)>;
382 let isReMaterializable = 1, isCodeGenOnly = 1, isMoveImm = 1,
383 isAsCheapAsAMove = 1 in {
384 // FIXME: The following pseudo instructions are only needed because remat
385 // cannot handle multiple instructions. When that changes, we can select
386 // directly to the real instructions and get rid of these pseudos.
389 : Pseudo<(outs GPR32:$dst), (ins i32imm:$src),
390 [(set GPR32:$dst, imm:$src)]>,
393 : Pseudo<(outs GPR64:$dst), (ins i64imm:$src),
394 [(set GPR64:$dst, imm:$src)]>,
396 } // isReMaterializable, isCodeGenOnly
398 def : Pat<(ARM64WrapperLarge tglobaladdr:$g3, tglobaladdr:$g2,
399 tglobaladdr:$g1, tglobaladdr:$g0),
400 (MOVKXi (MOVKXi (MOVKXi (MOVZXi tglobaladdr:$g3, 48),
401 tglobaladdr:$g2, 32),
402 tglobaladdr:$g1, 16),
403 tglobaladdr:$g0, 0)>;
405 def : Pat<(ARM64WrapperLarge tblockaddress:$g3, tblockaddress:$g2,
406 tblockaddress:$g1, tblockaddress:$g0),
407 (MOVKXi (MOVKXi (MOVKXi (MOVZXi tblockaddress:$g3, 48),
408 tblockaddress:$g2, 32),
409 tblockaddress:$g1, 16),
410 tblockaddress:$g0, 0)>;
412 def : Pat<(ARM64WrapperLarge tconstpool:$g3, tconstpool:$g2,
413 tconstpool:$g1, tconstpool:$g0),
414 (MOVKXi (MOVKXi (MOVKXi (MOVZXi tconstpool:$g3, 48),
420 //===----------------------------------------------------------------------===//
421 // Arithmetic instructions.
422 //===----------------------------------------------------------------------===//
424 // Add/subtract with carry.
425 defm ADC : AddSubCarry<0, "adc", "adcs", ARM64adc, ARM64adc_flag>;
426 defm SBC : AddSubCarry<1, "sbc", "sbcs", ARM64sbc, ARM64sbc_flag>;
428 def : InstAlias<"ngc $dst, $src", (SBCWr GPR32:$dst, WZR, GPR32:$src)>;
429 def : InstAlias<"ngc $dst, $src", (SBCXr GPR64:$dst, XZR, GPR64:$src)>;
430 def : InstAlias<"ngcs $dst, $src", (SBCSWr GPR32:$dst, WZR, GPR32:$src)>;
431 def : InstAlias<"ngcs $dst, $src", (SBCSXr GPR64:$dst, XZR, GPR64:$src)>;
434 defm ADD : AddSub<0, "add", add>;
435 defm SUB : AddSub<1, "sub">;
437 defm ADDS : AddSubS<0, "adds", ARM64add_flag>;
438 defm SUBS : AddSubS<1, "subs", ARM64sub_flag>;
440 // Use SUBS instead of SUB to enable CSE between SUBS and SUB.
441 def : Pat<(sub GPR32sp:$Rn, addsub_shifted_imm32:$imm),
442 (SUBSWri GPR32sp:$Rn, addsub_shifted_imm32:$imm)>;
443 def : Pat<(sub GPR64sp:$Rn, addsub_shifted_imm64:$imm),
444 (SUBSXri GPR64sp:$Rn, addsub_shifted_imm64:$imm)>;
445 def : Pat<(sub GPR32:$Rn, GPR32:$Rm),
446 (SUBSWrr GPR32:$Rn, GPR32:$Rm)>;
447 def : Pat<(sub GPR64:$Rn, GPR64:$Rm),
448 (SUBSXrr GPR64:$Rn, GPR64:$Rm)>;
449 def : Pat<(sub GPR32:$Rn, arith_shifted_reg32:$Rm),
450 (SUBSWrs GPR32:$Rn, arith_shifted_reg32:$Rm)>;
451 def : Pat<(sub GPR64:$Rn, arith_shifted_reg64:$Rm),
452 (SUBSXrs GPR64:$Rn, arith_shifted_reg64:$Rm)>;
453 def : Pat<(sub GPR32sp:$R2, arith_extended_reg32<i32>:$R3),
454 (SUBSWrx GPR32sp:$R2, arith_extended_reg32<i32>:$R3)>;
455 def : Pat<(sub GPR64sp:$R2, arith_extended_reg32to64<i64>:$R3),
456 (SUBSXrx GPR64sp:$R2, arith_extended_reg32to64<i64>:$R3)>;
458 // Because of the immediate format for add/sub-imm instructions, the
459 // expression (add x, -1) must be transformed to (SUB{W,X}ri x, 1).
460 // These patterns capture that transformation.
461 let AddedComplexity = 1 in {
462 def : Pat<(add GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
463 (SUBSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
464 def : Pat<(add GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
465 (SUBSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
466 def : Pat<(sub GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
467 (ADDWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
468 def : Pat<(sub GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
469 (ADDXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
472 def : InstAlias<"neg $dst, $src", (SUBWrs GPR32:$dst, WZR, GPR32:$src, 0)>;
473 def : InstAlias<"neg $dst, $src", (SUBXrs GPR64:$dst, XZR, GPR64:$src, 0)>;
474 def : InstAlias<"neg $dst, $src, $shift",
475 (SUBWrs GPR32:$dst, WZR, GPR32:$src, arith_shift:$shift)>;
476 def : InstAlias<"neg $dst, $src, $shift",
477 (SUBXrs GPR64:$dst, XZR, GPR64:$src, arith_shift:$shift)>;
479 // Because of the immediate format for add/sub-imm instructions, the
480 // expression (add x, -1) must be transformed to (SUB{W,X}ri x, 1).
481 // These patterns capture that transformation.
482 let AddedComplexity = 1 in {
483 def : Pat<(ARM64add_flag GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
484 (SUBSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
485 def : Pat<(ARM64add_flag GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
486 (SUBSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
487 def : Pat<(ARM64sub_flag GPR32:$Rn, neg_addsub_shifted_imm32:$imm),
488 (ADDSWri GPR32:$Rn, neg_addsub_shifted_imm32:$imm)>;
489 def : Pat<(ARM64sub_flag GPR64:$Rn, neg_addsub_shifted_imm64:$imm),
490 (ADDSXri GPR64:$Rn, neg_addsub_shifted_imm64:$imm)>;
493 def : InstAlias<"negs $dst, $src", (SUBSWrs GPR32:$dst, WZR, GPR32:$src, 0)>;
494 def : InstAlias<"negs $dst, $src", (SUBSXrs GPR64:$dst, XZR, GPR64:$src, 0)>;
495 def : InstAlias<"negs $dst, $src, $shift",
496 (SUBSWrs GPR32:$dst, WZR, GPR32:$src, arith_shift:$shift)>;
497 def : InstAlias<"negs $dst, $src, $shift",
498 (SUBSXrs GPR64:$dst, XZR, GPR64:$src, arith_shift:$shift)>;
500 // Unsigned/Signed divide
501 defm UDIV : Div<0, "udiv", udiv>;
502 defm SDIV : Div<1, "sdiv", sdiv>;
503 let isCodeGenOnly = 1 in {
504 defm UDIV_Int : Div<0, "udiv", int_arm64_udiv>;
505 defm SDIV_Int : Div<1, "sdiv", int_arm64_sdiv>;
509 defm ASRV : Shift<0b10, "asrv", sra>;
510 defm LSLV : Shift<0b00, "lslv", shl>;
511 defm LSRV : Shift<0b01, "lsrv", srl>;
512 defm RORV : Shift<0b11, "rorv", rotr>;
514 def : ShiftAlias<"asr", ASRVWr, GPR32>;
515 def : ShiftAlias<"asr", ASRVXr, GPR64>;
516 def : ShiftAlias<"lsl", LSLVWr, GPR32>;
517 def : ShiftAlias<"lsl", LSLVXr, GPR64>;
518 def : ShiftAlias<"lsr", LSRVWr, GPR32>;
519 def : ShiftAlias<"lsr", LSRVXr, GPR64>;
520 def : ShiftAlias<"ror", RORVWr, GPR32>;
521 def : ShiftAlias<"ror", RORVXr, GPR64>;
524 let AddedComplexity = 7 in {
525 defm MADD : MulAccum<0, "madd", add>;
526 defm MSUB : MulAccum<1, "msub", sub>;
528 def : Pat<(i32 (mul GPR32:$Rn, GPR32:$Rm)),
529 (MADDWrrr GPR32:$Rn, GPR32:$Rm, WZR)>;
530 def : Pat<(i64 (mul GPR64:$Rn, GPR64:$Rm)),
531 (MADDXrrr GPR64:$Rn, GPR64:$Rm, XZR)>;
533 def : Pat<(i32 (ineg (mul GPR32:$Rn, GPR32:$Rm))),
534 (MSUBWrrr GPR32:$Rn, GPR32:$Rm, WZR)>;
535 def : Pat<(i64 (ineg (mul GPR64:$Rn, GPR64:$Rm))),
536 (MSUBXrrr GPR64:$Rn, GPR64:$Rm, XZR)>;
537 } // AddedComplexity = 7
539 let AddedComplexity = 5 in {
540 def SMADDLrrr : WideMulAccum<0, 0b001, "smaddl", add, sext>;
541 def SMSUBLrrr : WideMulAccum<1, 0b001, "smsubl", sub, sext>;
542 def UMADDLrrr : WideMulAccum<0, 0b101, "umaddl", add, zext>;
543 def UMSUBLrrr : WideMulAccum<1, 0b101, "umsubl", sub, zext>;
545 def : Pat<(i64 (mul (sext GPR32:$Rn), (sext GPR32:$Rm))),
546 (SMADDLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
547 def : Pat<(i64 (mul (zext GPR32:$Rn), (zext GPR32:$Rm))),
548 (UMADDLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
550 def : Pat<(i64 (ineg (mul (sext GPR32:$Rn), (sext GPR32:$Rm)))),
551 (SMSUBLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
552 def : Pat<(i64 (ineg (mul (zext GPR32:$Rn), (zext GPR32:$Rm)))),
553 (UMSUBLrrr GPR32:$Rn, GPR32:$Rm, XZR)>;
554 } // AddedComplexity = 5
556 def : MulAccumWAlias<"mul", MADDWrrr>;
557 def : MulAccumXAlias<"mul", MADDXrrr>;
558 def : MulAccumWAlias<"mneg", MSUBWrrr>;
559 def : MulAccumXAlias<"mneg", MSUBXrrr>;
560 def : WideMulAccumAlias<"smull", SMADDLrrr>;
561 def : WideMulAccumAlias<"smnegl", SMSUBLrrr>;
562 def : WideMulAccumAlias<"umull", UMADDLrrr>;
563 def : WideMulAccumAlias<"umnegl", UMSUBLrrr>;
566 def SMULHrr : MulHi<0b010, "smulh", mulhs>;
567 def UMULHrr : MulHi<0b110, "umulh", mulhu>;
570 def CRC32Brr : BaseCRC32<0, 0b00, 0, GPR32, int_arm64_crc32b, "crc32b">;
571 def CRC32Hrr : BaseCRC32<0, 0b01, 0, GPR32, int_arm64_crc32h, "crc32h">;
572 def CRC32Wrr : BaseCRC32<0, 0b10, 0, GPR32, int_arm64_crc32w, "crc32w">;
573 def CRC32Xrr : BaseCRC32<1, 0b11, 0, GPR64, int_arm64_crc32x, "crc32x">;
575 def CRC32CBrr : BaseCRC32<0, 0b00, 1, GPR32, int_arm64_crc32cb, "crc32cb">;
576 def CRC32CHrr : BaseCRC32<0, 0b01, 1, GPR32, int_arm64_crc32ch, "crc32ch">;
577 def CRC32CWrr : BaseCRC32<0, 0b10, 1, GPR32, int_arm64_crc32cw, "crc32cw">;
578 def CRC32CXrr : BaseCRC32<1, 0b11, 1, GPR64, int_arm64_crc32cx, "crc32cx">;
581 //===----------------------------------------------------------------------===//
582 // Logical instructions.
583 //===----------------------------------------------------------------------===//
586 defm ANDS : LogicalImmS<0b11, "ands", ARM64and_flag>;
587 defm AND : LogicalImm<0b00, "and", and>;
588 defm EOR : LogicalImm<0b10, "eor", xor>;
589 defm ORR : LogicalImm<0b01, "orr", or>;
591 def : InstAlias<"mov $dst, $imm", (ORRWri GPR32sp:$dst, WZR,
592 logical_imm32:$imm)>;
593 def : InstAlias<"mov $dst, $imm", (ORRXri GPR64sp:$dst, XZR,
594 logical_imm64:$imm)>;
598 defm ANDS : LogicalRegS<0b11, 0, "ands">;
599 defm BICS : LogicalRegS<0b11, 1, "bics">;
600 defm AND : LogicalReg<0b00, 0, "and", and>;
601 defm BIC : LogicalReg<0b00, 1, "bic",
602 BinOpFrag<(and node:$LHS, (not node:$RHS))>>;
603 defm EON : LogicalReg<0b10, 1, "eon",
604 BinOpFrag<(xor node:$LHS, (not node:$RHS))>>;
605 defm EOR : LogicalReg<0b10, 0, "eor", xor>;
606 defm ORN : LogicalReg<0b01, 1, "orn",
607 BinOpFrag<(or node:$LHS, (not node:$RHS))>>;
608 defm ORR : LogicalReg<0b01, 0, "orr", or>;
610 def : InstAlias<"tst $src1, $src2",
611 (ANDSWri WZR, GPR32:$src1, logical_imm32:$src2)>;
612 def : InstAlias<"tst $src1, $src2",
613 (ANDSXri XZR, GPR64:$src1, logical_imm64:$src2)>;
615 def : InstAlias<"tst $src1, $src2",
616 (ANDSWrs WZR, GPR32:$src1, GPR32:$src2, 0)>;
617 def : InstAlias<"tst $src1, $src2",
618 (ANDSXrs XZR, GPR64:$src1, GPR64:$src2, 0)>;
620 def : InstAlias<"tst $src1, $src2, $sh",
621 (ANDSWrs WZR, GPR32:$src1, GPR32:$src2, logical_shift:$sh)>;
622 def : InstAlias<"tst $src1, $src2, $sh",
623 (ANDSXrs XZR, GPR64:$src1, GPR64:$src2, logical_shift:$sh)>;
625 def : InstAlias<"mvn $Wd, $Wm",
626 (ORNWrs GPR32:$Wd, WZR, GPR32:$Wm, 0)>;
627 def : InstAlias<"mvn $Xd, $Xm",
628 (ORNXrs GPR64:$Xd, XZR, GPR64:$Xm, 0)>;
630 def : InstAlias<"mvn $Wd, $Wm, $sh",
631 (ORNWrs GPR32:$Wd, WZR, GPR32:$Wm, logical_shift:$sh)>;
632 def : InstAlias<"mvn $Xd, $Xm, $sh",
633 (ORNXrs GPR64:$Xd, XZR, GPR64:$Xm, logical_shift:$sh)>;
635 def : Pat<(not GPR32:$Wm), (ORNWrr WZR, GPR32:$Wm)>;
636 def : Pat<(not GPR64:$Xm), (ORNXrr XZR, GPR64:$Xm)>;
639 //===----------------------------------------------------------------------===//
640 // One operand data processing instructions.
641 //===----------------------------------------------------------------------===//
643 defm CLS : OneOperandData<0b101, "cls">;
644 defm CLZ : OneOperandData<0b100, "clz", ctlz>;
645 defm RBIT : OneOperandData<0b000, "rbit">;
646 def REV16Wr : OneWRegData<0b001, "rev16",
647 UnOpFrag<(rotr (bswap node:$LHS), (i64 16))>>;
648 def REV16Xr : OneXRegData<0b001, "rev16", null_frag>;
650 def : Pat<(cttz GPR32:$Rn),
651 (CLZWr (RBITWr GPR32:$Rn))>;
652 def : Pat<(cttz GPR64:$Rn),
653 (CLZXr (RBITXr GPR64:$Rn))>;
654 def : Pat<(ctlz (or (shl (xor (sra GPR32:$Rn, (i64 31)), GPR32:$Rn), (i64 1)),
657 def : Pat<(ctlz (or (shl (xor (sra GPR64:$Rn, (i64 63)), GPR64:$Rn), (i64 1)),
661 // Unlike the other one operand instructions, the instructions with the "rev"
662 // mnemonic do *not* just different in the size bit, but actually use different
663 // opcode bits for the different sizes.
664 def REVWr : OneWRegData<0b010, "rev", bswap>;
665 def REVXr : OneXRegData<0b011, "rev", bswap>;
666 def REV32Xr : OneXRegData<0b010, "rev32",
667 UnOpFrag<(rotr (bswap node:$LHS), (i64 32))>>;
669 // The bswap commutes with the rotr so we want a pattern for both possible
671 def : Pat<(bswap (rotr GPR32:$Rn, (i64 16))), (REV16Wr GPR32:$Rn)>;
672 def : Pat<(bswap (rotr GPR64:$Rn, (i64 32))), (REV32Xr GPR64:$Rn)>;
674 //===----------------------------------------------------------------------===//
675 // Bitfield immediate extraction instruction.
676 //===----------------------------------------------------------------------===//
677 let neverHasSideEffects = 1 in
678 defm EXTR : ExtractImm<"extr">;
679 def : InstAlias<"ror $dst, $src, $shift",
680 (EXTRWrri GPR32:$dst, GPR32:$src, GPR32:$src, imm0_31:$shift)>;
681 def : InstAlias<"ror $dst, $src, $shift",
682 (EXTRXrri GPR64:$dst, GPR64:$src, GPR64:$src, imm0_63:$shift)>;
684 def : Pat<(rotr GPR32:$Rn, (i64 imm0_31:$imm)),
685 (EXTRWrri GPR32:$Rn, GPR32:$Rn, imm0_31:$imm)>;
686 def : Pat<(rotr GPR64:$Rn, (i64 imm0_63:$imm)),
687 (EXTRXrri GPR64:$Rn, GPR64:$Rn, imm0_63:$imm)>;
689 //===----------------------------------------------------------------------===//
690 // Other bitfield immediate instructions.
691 //===----------------------------------------------------------------------===//
692 let neverHasSideEffects = 1 in {
693 defm BFM : BitfieldImmWith2RegArgs<0b01, "bfm">;
694 defm SBFM : BitfieldImm<0b00, "sbfm">;
695 defm UBFM : BitfieldImm<0b10, "ubfm">;
698 def i32shift_a : Operand<i64>, SDNodeXForm<imm, [{
699 uint64_t enc = (32 - N->getZExtValue()) & 0x1f;
700 return CurDAG->getTargetConstant(enc, MVT::i64);
703 def i32shift_b : Operand<i64>, SDNodeXForm<imm, [{
704 uint64_t enc = 31 - N->getZExtValue();
705 return CurDAG->getTargetConstant(enc, MVT::i64);
708 // min(7, 31 - shift_amt)
709 def i32shift_sext_i8 : Operand<i64>, SDNodeXForm<imm, [{
710 uint64_t enc = 31 - N->getZExtValue();
711 enc = enc > 7 ? 7 : enc;
712 return CurDAG->getTargetConstant(enc, MVT::i64);
715 // min(15, 31 - shift_amt)
716 def i32shift_sext_i16 : Operand<i64>, SDNodeXForm<imm, [{
717 uint64_t enc = 31 - N->getZExtValue();
718 enc = enc > 15 ? 15 : enc;
719 return CurDAG->getTargetConstant(enc, MVT::i64);
722 def i64shift_a : Operand<i64>, SDNodeXForm<imm, [{
723 uint64_t enc = (64 - N->getZExtValue()) & 0x3f;
724 return CurDAG->getTargetConstant(enc, MVT::i64);
727 def i64shift_b : Operand<i64>, SDNodeXForm<imm, [{
728 uint64_t enc = 63 - N->getZExtValue();
729 return CurDAG->getTargetConstant(enc, MVT::i64);
732 // min(7, 63 - shift_amt)
733 def i64shift_sext_i8 : Operand<i64>, SDNodeXForm<imm, [{
734 uint64_t enc = 63 - N->getZExtValue();
735 enc = enc > 7 ? 7 : enc;
736 return CurDAG->getTargetConstant(enc, MVT::i64);
739 // min(15, 63 - shift_amt)
740 def i64shift_sext_i16 : Operand<i64>, SDNodeXForm<imm, [{
741 uint64_t enc = 63 - N->getZExtValue();
742 enc = enc > 15 ? 15 : enc;
743 return CurDAG->getTargetConstant(enc, MVT::i64);
746 // min(31, 63 - shift_amt)
747 def i64shift_sext_i32 : Operand<i64>, SDNodeXForm<imm, [{
748 uint64_t enc = 63 - N->getZExtValue();
749 enc = enc > 31 ? 31 : enc;
750 return CurDAG->getTargetConstant(enc, MVT::i64);
753 def : Pat<(shl GPR32:$Rn, (i64 imm0_31:$imm)),
754 (UBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
755 (i64 (i32shift_b imm0_31:$imm)))>;
756 def : Pat<(shl GPR64:$Rn, (i64 imm0_63:$imm)),
757 (UBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
758 (i64 (i64shift_b imm0_63:$imm)))>;
760 let AddedComplexity = 10 in {
761 def : Pat<(sra GPR32:$Rn, (i64 imm0_31:$imm)),
762 (SBFMWri GPR32:$Rn, imm0_31:$imm, 31)>;
763 def : Pat<(sra GPR64:$Rn, (i64 imm0_63:$imm)),
764 (SBFMXri GPR64:$Rn, imm0_63:$imm, 63)>;
767 def : InstAlias<"asr $dst, $src, $shift",
768 (SBFMWri GPR32:$dst, GPR32:$src, imm0_31:$shift, 31)>;
769 def : InstAlias<"asr $dst, $src, $shift",
770 (SBFMXri GPR64:$dst, GPR64:$src, imm0_63:$shift, 63)>;
771 def : InstAlias<"sxtb $dst, $src", (SBFMWri GPR32:$dst, GPR32:$src, 0, 7)>;
772 def : InstAlias<"sxtb $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 7)>;
773 def : InstAlias<"sxth $dst, $src", (SBFMWri GPR32:$dst, GPR32:$src, 0, 15)>;
774 def : InstAlias<"sxth $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 15)>;
775 def : InstAlias<"sxtw $dst, $src", (SBFMXri GPR64:$dst, GPR64:$src, 0, 31)>;
777 def : Pat<(srl GPR32:$Rn, (i64 imm0_31:$imm)),
778 (UBFMWri GPR32:$Rn, imm0_31:$imm, 31)>;
779 def : Pat<(srl GPR64:$Rn, (i64 imm0_63:$imm)),
780 (UBFMXri GPR64:$Rn, imm0_63:$imm, 63)>;
782 def : InstAlias<"lsr $dst, $src, $shift",
783 (UBFMWri GPR32:$dst, GPR32:$src, imm0_31:$shift, 31)>;
784 def : InstAlias<"lsr $dst, $src, $shift",
785 (UBFMXri GPR64:$dst, GPR64:$src, imm0_63:$shift, 63)>;
786 def : InstAlias<"uxtb $dst, $src", (UBFMWri GPR32:$dst, GPR32:$src, 0, 7)>;
787 def : InstAlias<"uxtb $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 7)>;
788 def : InstAlias<"uxth $dst, $src", (UBFMWri GPR32:$dst, GPR32:$src, 0, 15)>;
789 def : InstAlias<"uxth $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 15)>;
790 def : InstAlias<"uxtw $dst, $src", (UBFMXri GPR64:$dst, GPR64:$src, 0, 31)>;
792 //===----------------------------------------------------------------------===//
793 // Conditionally set flags instructions.
794 //===----------------------------------------------------------------------===//
795 defm CCMN : CondSetFlagsImm<0, "ccmn">;
796 defm CCMP : CondSetFlagsImm<1, "ccmp">;
798 defm CCMN : CondSetFlagsReg<0, "ccmn">;
799 defm CCMP : CondSetFlagsReg<1, "ccmp">;
801 //===----------------------------------------------------------------------===//
802 // Conditional select instructions.
803 //===----------------------------------------------------------------------===//
804 defm CSEL : CondSelect<0, 0b00, "csel">;
806 def inc : PatFrag<(ops node:$in), (add node:$in, 1)>;
807 defm CSINC : CondSelectOp<0, 0b01, "csinc", inc>;
808 defm CSINV : CondSelectOp<1, 0b00, "csinv", not>;
809 defm CSNEG : CondSelectOp<1, 0b01, "csneg", ineg>;
811 def : Pat<(ARM64csinv GPR32:$tval, GPR32:$fval, (i32 imm:$cc), CPSR),
812 (CSINVWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
813 def : Pat<(ARM64csinv GPR64:$tval, GPR64:$fval, (i32 imm:$cc), CPSR),
814 (CSINVXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
815 def : Pat<(ARM64csneg GPR32:$tval, GPR32:$fval, (i32 imm:$cc), CPSR),
816 (CSNEGWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
817 def : Pat<(ARM64csneg GPR64:$tval, GPR64:$fval, (i32 imm:$cc), CPSR),
818 (CSNEGXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
819 def : Pat<(ARM64csinc GPR32:$tval, GPR32:$fval, (i32 imm:$cc), CPSR),
820 (CSINCWr GPR32:$tval, GPR32:$fval, (i32 imm:$cc))>;
821 def : Pat<(ARM64csinc GPR64:$tval, GPR64:$fval, (i32 imm:$cc), CPSR),
822 (CSINCXr GPR64:$tval, GPR64:$fval, (i32 imm:$cc))>;
824 def : Pat<(ARM64csel (i32 0), (i32 1), (i32 imm:$cc), CPSR),
825 (CSINCWr WZR, WZR, (i32 imm:$cc))>;
826 def : Pat<(ARM64csel (i64 0), (i64 1), (i32 imm:$cc), CPSR),
827 (CSINCXr XZR, XZR, (i32 imm:$cc))>;
828 def : Pat<(ARM64csel (i32 0), (i32 -1), (i32 imm:$cc), CPSR),
829 (CSINVWr WZR, WZR, (i32 imm:$cc))>;
830 def : Pat<(ARM64csel (i64 0), (i64 -1), (i32 imm:$cc), CPSR),
831 (CSINVXr XZR, XZR, (i32 imm:$cc))>;
833 // The inverse of the condition code from the alias instruction is what is used
834 // in the aliased instruction. The parser all ready inverts the condition code
835 // for these aliases.
836 // FIXME: Is this the correct way to handle these aliases?
837 def : InstAlias<"cset $dst, $cc", (CSINCWr GPR32:$dst, WZR, WZR, ccode:$cc)>;
838 def : InstAlias<"cset $dst, $cc", (CSINCXr GPR64:$dst, XZR, XZR, ccode:$cc)>;
840 def : InstAlias<"csetm $dst, $cc", (CSINVWr GPR32:$dst, WZR, WZR, ccode:$cc)>;
841 def : InstAlias<"csetm $dst, $cc", (CSINVXr GPR64:$dst, XZR, XZR, ccode:$cc)>;
843 def : InstAlias<"cinc $dst, $src, $cc",
844 (CSINCWr GPR32:$dst, GPR32:$src, GPR32:$src, ccode:$cc)>;
845 def : InstAlias<"cinc $dst, $src, $cc",
846 (CSINCXr GPR64:$dst, GPR64:$src, GPR64:$src, ccode:$cc)>;
848 def : InstAlias<"cinv $dst, $src, $cc",
849 (CSINVWr GPR32:$dst, GPR32:$src, GPR32:$src, ccode:$cc)>;
850 def : InstAlias<"cinv $dst, $src, $cc",
851 (CSINVXr GPR64:$dst, GPR64:$src, GPR64:$src, ccode:$cc)>;
853 def : InstAlias<"cneg $dst, $src, $cc",
854 (CSNEGWr GPR32:$dst, GPR32:$src, GPR32:$src, ccode:$cc)>;
855 def : InstAlias<"cneg $dst, $src, $cc",
856 (CSNEGXr GPR64:$dst, GPR64:$src, GPR64:$src, ccode:$cc)>;
858 //===----------------------------------------------------------------------===//
859 // PC-relative instructions.
860 //===----------------------------------------------------------------------===//
861 let isReMaterializable = 1 in {
862 let neverHasSideEffects = 1, mayStore = 0, mayLoad = 0 in {
863 def ADR : ADRI<0, "adr", adrlabel, []>;
864 } // neverHasSideEffects = 1
866 def ADRP : ADRI<1, "adrp", adrplabel,
867 [(set GPR64:$Xd, (ARM64adrp tglobaladdr:$label))]>;
868 } // isReMaterializable = 1
870 // page address of a constant pool entry, block address
871 def : Pat<(ARM64adrp tconstpool:$cp), (ADRP tconstpool:$cp)>;
872 def : Pat<(ARM64adrp tblockaddress:$cp), (ADRP tblockaddress:$cp)>;
874 //===----------------------------------------------------------------------===//
875 // Unconditional branch (register) instructions.
876 //===----------------------------------------------------------------------===//
878 let isReturn = 1, isTerminator = 1, isBarrier = 1 in {
879 def RET : BranchReg<0b0010, "ret", []>;
880 def DRPS : SpecialReturn<0b0101, "drps">;
881 def ERET : SpecialReturn<0b0100, "eret">;
882 } // isReturn = 1, isTerminator = 1, isBarrier = 1
884 // Default to the LR register.
885 def : InstAlias<"ret", (RET LR)>;
887 let isCall = 1, Defs = [LR], Uses = [SP] in {
888 def BLR : BranchReg<0b0001, "blr", [(ARM64call GPR64:$Rn)]>;
891 let isBranch = 1, isTerminator = 1, isBarrier = 1, isIndirectBranch = 1 in {
892 def BR : BranchReg<0b0000, "br", [(brind GPR64:$Rn)]>;
893 } // isBranch, isTerminator, isBarrier, isIndirectBranch
895 // Create a separate pseudo-instruction for codegen to use so that we don't
896 // flag lr as used in every function. It'll be restored before the RET by the
897 // epilogue if it's legitimately used.
898 def RET_ReallyLR : Pseudo<(outs), (ins), [(ARM64retflag)]> {
899 let isTerminator = 1;
904 // This is a directive-like pseudo-instruction. The purpose is to insert an
905 // R_AARCH64_TLSDESC_CALL relocation at the offset of the following instruction
906 // (which in the usual case is a BLR).
907 let hasSideEffects = 1 in
908 def TLSDESCCALL : Pseudo<(outs), (ins i64imm:$sym), []> {
909 let AsmString = ".tlsdesccall $sym";
912 // Pseudo-instruction representing a BLR with attached TLSDESC relocation. It
913 // gets expanded to two MCInsts during lowering.
914 let isCall = 1, Defs = [LR] in
916 : Pseudo<(outs), (ins GPR64:$dest, i64imm:$sym),
917 [(ARM64tlsdesc_call GPR64:$dest, tglobaltlsaddr:$sym)]>;
919 def : Pat<(ARM64tlsdesc_call GPR64:$dest, texternalsym:$sym),
920 (TLSDESC_BLR GPR64:$dest, texternalsym:$sym)>;
921 //===----------------------------------------------------------------------===//
922 // Conditional branch (immediate) instruction.
923 //===----------------------------------------------------------------------===//
924 def Bcc : BranchCond;
926 //===----------------------------------------------------------------------===//
927 // Compare-and-branch instructions.
928 //===----------------------------------------------------------------------===//
929 defm CBZ : CmpBranch<0, "cbz", ARM64cbz>;
930 defm CBNZ : CmpBranch<1, "cbnz", ARM64cbnz>;
932 //===----------------------------------------------------------------------===//
933 // Test-bit-and-branch instructions.
934 //===----------------------------------------------------------------------===//
935 def TBZ : TestBranch<0, "tbz", ARM64tbz>;
936 def TBNZ : TestBranch<1, "tbnz", ARM64tbnz>;
938 //===----------------------------------------------------------------------===//
939 // Unconditional branch (immediate) instructions.
940 //===----------------------------------------------------------------------===//
941 let isBranch = 1, isTerminator = 1, isBarrier = 1 in {
942 def B : BranchImm<0, "b", [(br bb:$addr)]>;
943 } // isBranch, isTerminator, isBarrier
945 let isCall = 1, Defs = [LR], Uses = [SP] in {
946 def BL : CallImm<1, "bl", [(ARM64call tglobaladdr:$addr)]>;
948 def : Pat<(ARM64call texternalsym:$func), (BL texternalsym:$func)>;
950 //===----------------------------------------------------------------------===//
951 // Exception generation instructions.
952 //===----------------------------------------------------------------------===//
953 def BRK : ExceptionGeneration<0b001, 0b00, "brk">;
954 def DCPS1 : ExceptionGeneration<0b101, 0b01, "dcps1">;
955 def DCPS2 : ExceptionGeneration<0b101, 0b10, "dcps2">;
956 def DCPS3 : ExceptionGeneration<0b101, 0b11, "dcps3">;
957 def HLT : ExceptionGeneration<0b010, 0b00, "hlt">;
958 def HVC : ExceptionGeneration<0b000, 0b10, "hvc">;
959 def SMC : ExceptionGeneration<0b000, 0b11, "smc">;
960 def SVC : ExceptionGeneration<0b000, 0b01, "svc">;
962 // DCPSn defaults to an immediate operand of zero if unspecified.
963 def : InstAlias<"dcps1", (DCPS1 0)>;
964 def : InstAlias<"dcps2", (DCPS2 0)>;
965 def : InstAlias<"dcps3", (DCPS3 0)>;
967 //===----------------------------------------------------------------------===//
968 // Load instructions.
969 //===----------------------------------------------------------------------===//
971 // Pair (indexed, offset)
972 def LDPWi : LoadPairOffset<0b00, 0, GPR32, am_indexed32simm7, "ldp">;
973 def LDPXi : LoadPairOffset<0b10, 0, GPR64, am_indexed64simm7, "ldp">;
974 def LDPSi : LoadPairOffset<0b00, 1, FPR32, am_indexed32simm7, "ldp">;
975 def LDPDi : LoadPairOffset<0b01, 1, FPR64, am_indexed64simm7, "ldp">;
976 def LDPQi : LoadPairOffset<0b10, 1, FPR128, am_indexed128simm7, "ldp">;
978 def LDPSWi : LoadPairOffset<0b01, 0, GPR64, am_indexed32simm7, "ldpsw">;
980 // Pair (pre-indexed)
981 def LDPWpre : LoadPairPreIdx<0b00, 0, GPR32, am_indexed32simm7_wb, "ldp">;
982 def LDPXpre : LoadPairPreIdx<0b10, 0, GPR64, am_indexed64simm7_wb, "ldp">;
983 def LDPSpre : LoadPairPreIdx<0b00, 1, FPR32, am_indexed32simm7_wb, "ldp">;
984 def LDPDpre : LoadPairPreIdx<0b01, 1, FPR64, am_indexed64simm7_wb, "ldp">;
985 def LDPQpre : LoadPairPreIdx<0b10, 1, FPR128, am_indexed128simm7_wb, "ldp">;
987 def LDPSWpre : LoadPairPreIdx<0b01, 0, GPR64, am_indexed32simm7_wb, "ldpsw">;
989 // Pair (post-indexed)
990 def LDPWpost : LoadPairPostIdx<0b00, 0, GPR32, simm7s4, "ldp">;
991 def LDPXpost : LoadPairPostIdx<0b10, 0, GPR64, simm7s8, "ldp">;
992 def LDPSpost : LoadPairPostIdx<0b00, 1, FPR32, simm7s4, "ldp">;
993 def LDPDpost : LoadPairPostIdx<0b01, 1, FPR64, simm7s8, "ldp">;
994 def LDPQpost : LoadPairPostIdx<0b10, 1, FPR128, simm7s16, "ldp">;
996 def LDPSWpost : LoadPairPostIdx<0b01, 0, GPR64, simm7s4, "ldpsw">;
999 // Pair (no allocate)
1000 def LDNPWi : LoadPairNoAlloc<0b00, 0, GPR32, am_indexed32simm7, "ldnp">;
1001 def LDNPXi : LoadPairNoAlloc<0b10, 0, GPR64, am_indexed64simm7, "ldnp">;
1002 def LDNPSi : LoadPairNoAlloc<0b00, 1, FPR32, am_indexed32simm7, "ldnp">;
1003 def LDNPDi : LoadPairNoAlloc<0b01, 1, FPR64, am_indexed64simm7, "ldnp">;
1004 def LDNPQi : LoadPairNoAlloc<0b10, 1, FPR128, am_indexed128simm7, "ldnp">;
1007 // (register offset)
1010 let AddedComplexity = 10 in {
1012 def LDRBBro : Load8RO<0b00, 0, 0b01, GPR32, "ldrb",
1013 [(set GPR32:$Rt, (zextloadi8 ro_indexed8:$addr))]>;
1014 def LDRHHro : Load16RO<0b01, 0, 0b01, GPR32, "ldrh",
1015 [(set GPR32:$Rt, (zextloadi16 ro_indexed16:$addr))]>;
1016 def LDRWro : Load32RO<0b10, 0, 0b01, GPR32, "ldr",
1017 [(set GPR32:$Rt, (load ro_indexed32:$addr))]>;
1018 def LDRXro : Load64RO<0b11, 0, 0b01, GPR64, "ldr",
1019 [(set GPR64:$Rt, (load ro_indexed64:$addr))]>;
1022 def LDRBro : Load8RO<0b00, 1, 0b01, FPR8, "ldr",
1023 [(set FPR8:$Rt, (load ro_indexed8:$addr))]>;
1024 def LDRHro : Load16RO<0b01, 1, 0b01, FPR16, "ldr",
1025 [(set (f16 FPR16:$Rt), (load ro_indexed16:$addr))]>;
1026 def LDRSro : Load32RO<0b10, 1, 0b01, FPR32, "ldr",
1027 [(set (f32 FPR32:$Rt), (load ro_indexed32:$addr))]>;
1028 def LDRDro : Load64RO<0b11, 1, 0b01, FPR64, "ldr",
1029 [(set (f64 FPR64:$Rt), (load ro_indexed64:$addr))]>;
1030 def LDRQro : Load128RO<0b00, 1, 0b11, FPR128, "ldr", []> {
1034 // For regular load, we do not have any alignment requirement.
1035 // Thus, it is safe to directly map the vector loads with interesting
1036 // addressing modes.
1037 // FIXME: We could do the same for bitconvert to floating point vectors.
1038 def : Pat <(v8i8 (scalar_to_vector (i32 (extloadi8 ro_indexed8:$addr)))),
1039 (INSERT_SUBREG (v8i8 (IMPLICIT_DEF)),
1040 (LDRBro ro_indexed8:$addr), bsub)>;
1041 def : Pat <(v16i8 (scalar_to_vector (i32 (extloadi8 ro_indexed8:$addr)))),
1042 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
1043 (LDRBro ro_indexed8:$addr), bsub)>;
1044 def : Pat <(v4i16 (scalar_to_vector (i32 (extloadi16 ro_indexed16:$addr)))),
1045 (INSERT_SUBREG (v4i16 (IMPLICIT_DEF)),
1046 (LDRHro ro_indexed16:$addr), hsub)>;
1047 def : Pat <(v8i16 (scalar_to_vector (i32 (extloadi16 ro_indexed16:$addr)))),
1048 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
1049 (LDRHro ro_indexed16:$addr), hsub)>;
1050 def : Pat <(v2i32 (scalar_to_vector (i32 (load ro_indexed32:$addr)))),
1051 (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)),
1052 (LDRSro ro_indexed32:$addr), ssub)>;
1053 def : Pat <(v4i32 (scalar_to_vector (i32 (load ro_indexed32:$addr)))),
1054 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
1055 (LDRSro ro_indexed32:$addr), ssub)>;
1056 def : Pat <(v1i64 (scalar_to_vector (i64 (load ro_indexed64:$addr)))),
1057 (LDRDro ro_indexed64:$addr)>;
1058 def : Pat <(v2i64 (scalar_to_vector (i64 (load ro_indexed64:$addr)))),
1059 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
1060 (LDRDro ro_indexed64:$addr), dsub)>;
1062 // Match all load 64 bits width whose type is compatible with FPR64
1063 def : Pat<(v2f32 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1064 def : Pat<(v1f64 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1065 def : Pat<(v8i8 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1066 def : Pat<(v4i16 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1067 def : Pat<(v2i32 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1068 def : Pat<(v1i64 (load ro_indexed64:$addr)), (LDRDro ro_indexed64:$addr)>;
1070 // Match all load 128 bits width whose type is compatible with FPR128
1071 def : Pat<(v4f32 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1072 def : Pat<(v2f64 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1073 def : Pat<(v16i8 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1074 def : Pat<(v8i16 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1075 def : Pat<(v4i32 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1076 def : Pat<(v2i64 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1077 def : Pat<(f128 (load ro_indexed128:$addr)), (LDRQro ro_indexed128:$addr)>;
1079 // Load sign-extended half-word
1080 def LDRSHWro : Load16RO<0b01, 0, 0b11, GPR32, "ldrsh",
1081 [(set GPR32:$Rt, (sextloadi16 ro_indexed16:$addr))]>;
1082 def LDRSHXro : Load16RO<0b01, 0, 0b10, GPR64, "ldrsh",
1083 [(set GPR64:$Rt, (sextloadi16 ro_indexed16:$addr))]>;
1085 // Load sign-extended byte
1086 def LDRSBWro : Load8RO<0b00, 0, 0b11, GPR32, "ldrsb",
1087 [(set GPR32:$Rt, (sextloadi8 ro_indexed8:$addr))]>;
1088 def LDRSBXro : Load8RO<0b00, 0, 0b10, GPR64, "ldrsb",
1089 [(set GPR64:$Rt, (sextloadi8 ro_indexed8:$addr))]>;
1091 // Load sign-extended word
1092 def LDRSWro : Load32RO<0b10, 0, 0b10, GPR64, "ldrsw",
1093 [(set GPR64:$Rt, (sextloadi32 ro_indexed32:$addr))]>;
1096 def PRFMro : PrefetchRO<0b11, 0, 0b10, "prfm",
1097 [(ARM64Prefetch imm:$Rt, ro_indexed64:$addr)]>;
1100 def : Pat<(i64 (zextloadi8 ro_indexed8:$addr)),
1101 (SUBREG_TO_REG (i64 0), (LDRBBro ro_indexed8:$addr), sub_32)>;
1102 def : Pat<(i64 (zextloadi16 ro_indexed16:$addr)),
1103 (SUBREG_TO_REG (i64 0), (LDRHHro ro_indexed16:$addr), sub_32)>;
1105 // zextloadi1 -> zextloadi8
1106 def : Pat<(i32 (zextloadi1 ro_indexed8:$addr)), (LDRBBro ro_indexed8:$addr)>;
1107 def : Pat<(i64 (zextloadi1 ro_indexed8:$addr)),
1108 (SUBREG_TO_REG (i64 0), (LDRBBro ro_indexed8:$addr), sub_32)>;
1110 // extload -> zextload
1111 def : Pat<(i32 (extloadi16 ro_indexed16:$addr)), (LDRHHro ro_indexed16:$addr)>;
1112 def : Pat<(i32 (extloadi8 ro_indexed8:$addr)), (LDRBBro ro_indexed8:$addr)>;
1113 def : Pat<(i32 (extloadi1 ro_indexed8:$addr)), (LDRBBro ro_indexed8:$addr)>;
1114 def : Pat<(i64 (extloadi32 ro_indexed32:$addr)),
1115 (SUBREG_TO_REG (i64 0), (LDRWro ro_indexed32:$addr), sub_32)>;
1116 def : Pat<(i64 (extloadi16 ro_indexed16:$addr)),
1117 (SUBREG_TO_REG (i64 0), (LDRHHro ro_indexed16:$addr), sub_32)>;
1118 def : Pat<(i64 (extloadi8 ro_indexed8:$addr)),
1119 (SUBREG_TO_REG (i64 0), (LDRBBro ro_indexed8:$addr), sub_32)>;
1120 def : Pat<(i64 (extloadi1 ro_indexed8:$addr)),
1121 (SUBREG_TO_REG (i64 0), (LDRBBro ro_indexed8:$addr), sub_32)>;
1123 } // AddedComplexity = 10
1126 // (unsigned immediate)
1128 def LDRXui : LoadUI<0b11, 0, 0b01, GPR64, am_indexed64, "ldr",
1129 [(set GPR64:$Rt, (load am_indexed64:$addr))]>;
1130 def LDRWui : LoadUI<0b10, 0, 0b01, GPR32, am_indexed32, "ldr",
1131 [(set GPR32:$Rt, (load am_indexed32:$addr))]>;
1132 def LDRBui : LoadUI<0b00, 1, 0b01, FPR8, am_indexed8, "ldr",
1133 [(set FPR8:$Rt, (load am_indexed8:$addr))]>;
1134 def LDRHui : LoadUI<0b01, 1, 0b01, FPR16, am_indexed16, "ldr",
1135 [(set (f16 FPR16:$Rt), (load am_indexed16:$addr))]>;
1136 def LDRSui : LoadUI<0b10, 1, 0b01, FPR32, am_indexed32, "ldr",
1137 [(set (f32 FPR32:$Rt), (load am_indexed32:$addr))]>;
1138 def LDRDui : LoadUI<0b11, 1, 0b01, FPR64, am_indexed64, "ldr",
1139 [(set (f64 FPR64:$Rt), (load am_indexed64:$addr))]>;
1140 def LDRQui : LoadUI<0b00, 1, 0b11, FPR128, am_indexed128, "ldr",
1141 [(set (f128 FPR128:$Rt), (load am_indexed128:$addr))]>;
1143 // For regular load, we do not have any alignment requirement.
1144 // Thus, it is safe to directly map the vector loads with interesting
1145 // addressing modes.
1146 // FIXME: We could do the same for bitconvert to floating point vectors.
1147 def : Pat <(v8i8 (scalar_to_vector (i32 (extloadi8 am_indexed8:$addr)))),
1148 (INSERT_SUBREG (v8i8 (IMPLICIT_DEF)),
1149 (LDRBui am_indexed8:$addr), bsub)>;
1150 def : Pat <(v16i8 (scalar_to_vector (i32 (extloadi8 am_indexed8:$addr)))),
1151 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
1152 (LDRBui am_indexed8:$addr), bsub)>;
1153 def : Pat <(v4i16 (scalar_to_vector (i32 (extloadi16 am_indexed16:$addr)))),
1154 (INSERT_SUBREG (v4i16 (IMPLICIT_DEF)),
1155 (LDRHui am_indexed16:$addr), hsub)>;
1156 def : Pat <(v8i16 (scalar_to_vector (i32 (extloadi16 am_indexed16:$addr)))),
1157 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
1158 (LDRHui am_indexed16:$addr), hsub)>;
1159 def : Pat <(v2i32 (scalar_to_vector (i32 (load am_indexed32:$addr)))),
1160 (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)),
1161 (LDRSui am_indexed32:$addr), ssub)>;
1162 def : Pat <(v4i32 (scalar_to_vector (i32 (load am_indexed32:$addr)))),
1163 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
1164 (LDRSui am_indexed32:$addr), ssub)>;
1165 def : Pat <(v1i64 (scalar_to_vector (i64 (load am_indexed64:$addr)))),
1166 (LDRDui am_indexed64:$addr)>;
1167 def : Pat <(v2i64 (scalar_to_vector (i64 (load am_indexed64:$addr)))),
1168 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
1169 (LDRDui am_indexed64:$addr), dsub)>;
1171 // Match all load 64 bits width whose type is compatible with FPR64
1172 def : Pat<(v2f32 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1173 def : Pat<(v1f64 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1174 def : Pat<(v8i8 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1175 def : Pat<(v4i16 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1176 def : Pat<(v2i32 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1177 def : Pat<(v1i64 (load am_indexed64:$addr)), (LDRDui am_indexed64:$addr)>;
1179 // Match all load 128 bits width whose type is compatible with FPR128
1180 def : Pat<(v4f32 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1181 def : Pat<(v2f64 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1182 def : Pat<(v16i8 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1183 def : Pat<(v8i16 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1184 def : Pat<(v4i32 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1185 def : Pat<(v2i64 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1186 def : Pat<(f128 (load am_indexed128:$addr)), (LDRQui am_indexed128:$addr)>;
1188 def LDRHHui : LoadUI<0b01, 0, 0b01, GPR32, am_indexed16, "ldrh",
1189 [(set GPR32:$Rt, (zextloadi16 am_indexed16:$addr))]>;
1190 def LDRBBui : LoadUI<0b00, 0, 0b01, GPR32, am_indexed8, "ldrb",
1191 [(set GPR32:$Rt, (zextloadi8 am_indexed8:$addr))]>;
1193 def : Pat<(i64 (zextloadi8 am_indexed8:$addr)),
1194 (SUBREG_TO_REG (i64 0), (LDRBBui am_indexed8:$addr), sub_32)>;
1195 def : Pat<(i64 (zextloadi16 am_indexed16:$addr)),
1196 (SUBREG_TO_REG (i64 0), (LDRHHui am_indexed16:$addr), sub_32)>;
1198 // zextloadi1 -> zextloadi8
1199 def : Pat<(i32 (zextloadi1 am_indexed8:$addr)), (LDRBBui am_indexed8:$addr)>;
1200 def : Pat<(i64 (zextloadi1 am_indexed8:$addr)),
1201 (SUBREG_TO_REG (i64 0), (LDRBBui am_indexed8:$addr), sub_32)>;
1203 // extload -> zextload
1204 def : Pat<(i32 (extloadi16 am_indexed16:$addr)), (LDRHHui am_indexed16:$addr)>;
1205 def : Pat<(i32 (extloadi8 am_indexed8:$addr)), (LDRBBui am_indexed8:$addr)>;
1206 def : Pat<(i32 (extloadi1 am_indexed8:$addr)), (LDRBBui am_indexed8:$addr)>;
1207 def : Pat<(i64 (extloadi32 am_indexed32:$addr)),
1208 (SUBREG_TO_REG (i64 0), (LDRWui am_indexed32:$addr), sub_32)>;
1209 def : Pat<(i64 (extloadi16 am_indexed16:$addr)),
1210 (SUBREG_TO_REG (i64 0), (LDRHHui am_indexed16:$addr), sub_32)>;
1211 def : Pat<(i64 (extloadi8 am_indexed8:$addr)),
1212 (SUBREG_TO_REG (i64 0), (LDRBBui am_indexed8:$addr), sub_32)>;
1213 def : Pat<(i64 (extloadi1 am_indexed8:$addr)),
1214 (SUBREG_TO_REG (i64 0), (LDRBBui am_indexed8:$addr), sub_32)>;
1216 // load sign-extended half-word
1217 def LDRSHWui : LoadUI<0b01, 0, 0b11, GPR32, am_indexed16, "ldrsh",
1218 [(set GPR32:$Rt, (sextloadi16 am_indexed16:$addr))]>;
1219 def LDRSHXui : LoadUI<0b01, 0, 0b10, GPR64, am_indexed16, "ldrsh",
1220 [(set GPR64:$Rt, (sextloadi16 am_indexed16:$addr))]>;
1222 // load sign-extended byte
1223 def LDRSBWui : LoadUI<0b00, 0, 0b11, GPR32, am_indexed8, "ldrsb",
1224 [(set GPR32:$Rt, (sextloadi8 am_indexed8:$addr))]>;
1225 def LDRSBXui : LoadUI<0b00, 0, 0b10, GPR64, am_indexed8, "ldrsb",
1226 [(set GPR64:$Rt, (sextloadi8 am_indexed8:$addr))]>;
1228 // load sign-extended word
1229 def LDRSWui : LoadUI<0b10, 0, 0b10, GPR64, am_indexed32, "ldrsw",
1230 [(set GPR64:$Rt, (sextloadi32 am_indexed32:$addr))]>;
1232 // load zero-extended word
1233 def : Pat<(i64 (zextloadi32 am_indexed32:$addr)),
1234 (SUBREG_TO_REG (i64 0), (LDRWui am_indexed32:$addr), sub_32)>;
1237 def PRFMui : PrefetchUI<0b11, 0, 0b10, "prfm",
1238 [(ARM64Prefetch imm:$Rt, am_indexed64:$addr)]>;
1242 def LDRWl : LoadLiteral<0b00, 0, GPR32, "ldr">;
1243 def LDRXl : LoadLiteral<0b01, 0, GPR64, "ldr">;
1244 def LDRSl : LoadLiteral<0b00, 1, FPR32, "ldr">;
1245 def LDRDl : LoadLiteral<0b01, 1, FPR64, "ldr">;
1246 def LDRQl : LoadLiteral<0b10, 1, FPR128, "ldr">;
1248 // load sign-extended word
1249 def LDRSWl : LoadLiteral<0b10, 0, GPR64, "ldrsw">;
1252 def PRFMl : PrefetchLiteral<0b11, 0, "prfm", []>;
1253 // [(ARM64Prefetch imm:$Rt, tglobaladdr:$label)]>;
1256 // (unscaled immediate)
1257 def LDURXi : LoadUnscaled<0b11, 0, 0b01, GPR64, am_unscaled64, "ldur",
1258 [(set GPR64:$Rt, (load am_unscaled64:$addr))]>;
1259 def LDURWi : LoadUnscaled<0b10, 0, 0b01, GPR32, am_unscaled32, "ldur",
1260 [(set GPR32:$Rt, (load am_unscaled32:$addr))]>;
1261 def LDURBi : LoadUnscaled<0b00, 1, 0b01, FPR8, am_unscaled8, "ldur",
1262 [(set FPR8:$Rt, (load am_unscaled8:$addr))]>;
1263 def LDURHi : LoadUnscaled<0b01, 1, 0b01, FPR16, am_unscaled16, "ldur",
1264 [(set (f16 FPR16:$Rt), (load am_unscaled16:$addr))]>;
1265 def LDURSi : LoadUnscaled<0b10, 1, 0b01, FPR32, am_unscaled32, "ldur",
1266 [(set (f32 FPR32:$Rt), (load am_unscaled32:$addr))]>;
1267 def LDURDi : LoadUnscaled<0b11, 1, 0b01, FPR64, am_unscaled64, "ldur",
1268 [(set (f64 FPR64:$Rt), (load am_unscaled64:$addr))]>;
1269 def LDURQi : LoadUnscaled<0b00, 1, 0b11, FPR128, am_unscaled128, "ldur",
1270 [(set (v2f64 FPR128:$Rt), (load am_unscaled128:$addr))]>;
1273 : LoadUnscaled<0b01, 0, 0b01, GPR32, am_unscaled16, "ldurh",
1274 [(set GPR32:$Rt, (zextloadi16 am_unscaled16:$addr))]>;
1276 : LoadUnscaled<0b00, 0, 0b01, GPR32, am_unscaled8, "ldurb",
1277 [(set GPR32:$Rt, (zextloadi8 am_unscaled8:$addr))]>;
1279 // Match all load 64 bits width whose type is compatible with FPR64
1280 def : Pat<(v2f32 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1281 def : Pat<(v1f64 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1282 def : Pat<(v8i8 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1283 def : Pat<(v4i16 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1284 def : Pat<(v2i32 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1285 def : Pat<(v1i64 (load am_unscaled64:$addr)), (LDURDi am_unscaled64:$addr)>;
1287 // Match all load 128 bits width whose type is compatible with FPR128
1288 def : Pat<(v4f32 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1289 def : Pat<(v2f64 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1290 def : Pat<(v16i8 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1291 def : Pat<(v8i16 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1292 def : Pat<(v4i32 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1293 def : Pat<(v2i64 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1294 def : Pat<(f128 (load am_unscaled128:$addr)), (LDURQi am_unscaled128:$addr)>;
1297 def : Pat<(i32 (extloadi16 am_unscaled16:$addr)), (LDURHHi am_unscaled16:$addr)>;
1298 def : Pat<(i32 (extloadi8 am_unscaled8:$addr)), (LDURBBi am_unscaled8:$addr)>;
1299 def : Pat<(i32 (extloadi1 am_unscaled8:$addr)), (LDURBBi am_unscaled8:$addr)>;
1300 def : Pat<(i64 (extloadi32 am_unscaled32:$addr)),
1301 (SUBREG_TO_REG (i64 0), (LDURWi am_unscaled32:$addr), sub_32)>;
1302 def : Pat<(i64 (extloadi16 am_unscaled16:$addr)),
1303 (SUBREG_TO_REG (i64 0), (LDURHHi am_unscaled16:$addr), sub_32)>;
1304 def : Pat<(i64 (extloadi8 am_unscaled8:$addr)),
1305 (SUBREG_TO_REG (i64 0), (LDURBBi am_unscaled8:$addr), sub_32)>;
1306 def : Pat<(i64 (extloadi1 am_unscaled8:$addr)),
1307 (SUBREG_TO_REG (i64 0), (LDURBBi am_unscaled8:$addr), sub_32)>;
1309 def : Pat<(i32 (zextloadi16 am_unscaled16:$addr)),
1310 (LDURHHi am_unscaled16:$addr)>;
1311 def : Pat<(i32 (zextloadi8 am_unscaled8:$addr)),
1312 (LDURBBi am_unscaled8:$addr)>;
1313 def : Pat<(i32 (zextloadi1 am_unscaled8:$addr)),
1314 (LDURBBi am_unscaled8:$addr)>;
1315 def : Pat<(i64 (zextloadi32 am_unscaled32:$addr)),
1316 (SUBREG_TO_REG (i64 0), (LDURWi am_unscaled32:$addr), sub_32)>;
1317 def : Pat<(i64 (zextloadi16 am_unscaled16:$addr)),
1318 (SUBREG_TO_REG (i64 0), (LDURHHi am_unscaled16:$addr), sub_32)>;
1319 def : Pat<(i64 (zextloadi8 am_unscaled8:$addr)),
1320 (SUBREG_TO_REG (i64 0), (LDURBBi am_unscaled8:$addr), sub_32)>;
1321 def : Pat<(i64 (zextloadi1 am_unscaled8:$addr)),
1322 (SUBREG_TO_REG (i64 0), (LDURBBi am_unscaled8:$addr), sub_32)>;
1326 // LDR mnemonics fall back to LDUR for negative or unaligned offsets.
1328 // Define new assembler match classes as we want to only match these when
1329 // the don't otherwise match the scaled addressing mode for LDR/STR. Don't
1330 // associate a DiagnosticType either, as we want the diagnostic for the
1331 // canonical form (the scaled operand) to take precedence.
1332 def MemoryUnscaledFB8Operand : AsmOperandClass {
1333 let Name = "MemoryUnscaledFB8";
1334 let RenderMethod = "addMemoryUnscaledOperands";
1336 def MemoryUnscaledFB16Operand : AsmOperandClass {
1337 let Name = "MemoryUnscaledFB16";
1338 let RenderMethod = "addMemoryUnscaledOperands";
1340 def MemoryUnscaledFB32Operand : AsmOperandClass {
1341 let Name = "MemoryUnscaledFB32";
1342 let RenderMethod = "addMemoryUnscaledOperands";
1344 def MemoryUnscaledFB64Operand : AsmOperandClass {
1345 let Name = "MemoryUnscaledFB64";
1346 let RenderMethod = "addMemoryUnscaledOperands";
1348 def MemoryUnscaledFB128Operand : AsmOperandClass {
1349 let Name = "MemoryUnscaledFB128";
1350 let RenderMethod = "addMemoryUnscaledOperands";
1352 def am_unscaled_fb8 : Operand<i64> {
1353 let ParserMatchClass = MemoryUnscaledFB8Operand;
1354 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
1356 def am_unscaled_fb16 : Operand<i64> {
1357 let ParserMatchClass = MemoryUnscaledFB16Operand;
1358 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
1360 def am_unscaled_fb32 : Operand<i64> {
1361 let ParserMatchClass = MemoryUnscaledFB32Operand;
1362 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
1364 def am_unscaled_fb64 : Operand<i64> {
1365 let ParserMatchClass = MemoryUnscaledFB64Operand;
1366 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
1368 def am_unscaled_fb128 : Operand<i64> {
1369 let ParserMatchClass = MemoryUnscaledFB128Operand;
1370 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
1372 def : InstAlias<"ldr $Rt, $addr", (LDURXi GPR64:$Rt, am_unscaled_fb64:$addr)>;
1373 def : InstAlias<"ldr $Rt, $addr", (LDURWi GPR32:$Rt, am_unscaled_fb32:$addr)>;
1374 def : InstAlias<"ldr $Rt, $addr", (LDURBi FPR8:$Rt, am_unscaled_fb8:$addr)>;
1375 def : InstAlias<"ldr $Rt, $addr", (LDURHi FPR16:$Rt, am_unscaled_fb16:$addr)>;
1376 def : InstAlias<"ldr $Rt, $addr", (LDURSi FPR32:$Rt, am_unscaled_fb32:$addr)>;
1377 def : InstAlias<"ldr $Rt, $addr", (LDURDi FPR64:$Rt, am_unscaled_fb64:$addr)>;
1378 def : InstAlias<"ldr $Rt, $addr", (LDURQi FPR128:$Rt, am_unscaled_fb128:$addr)>;
1381 def : Pat<(i64 (zextloadi8 am_unscaled8:$addr)),
1382 (SUBREG_TO_REG (i64 0), (LDURBBi am_unscaled8:$addr), sub_32)>;
1383 def : Pat<(i64 (zextloadi16 am_unscaled16:$addr)),
1384 (SUBREG_TO_REG (i64 0), (LDURHHi am_unscaled16:$addr), sub_32)>;
1386 // load sign-extended half-word
1388 : LoadUnscaled<0b01, 0, 0b11, GPR32, am_unscaled16, "ldursh",
1389 [(set GPR32:$Rt, (sextloadi16 am_unscaled16:$addr))]>;
1391 : LoadUnscaled<0b01, 0, 0b10, GPR64, am_unscaled16, "ldursh",
1392 [(set GPR64:$Rt, (sextloadi16 am_unscaled16:$addr))]>;
1394 // load sign-extended byte
1396 : LoadUnscaled<0b00, 0, 0b11, GPR32, am_unscaled8, "ldursb",
1397 [(set GPR32:$Rt, (sextloadi8 am_unscaled8:$addr))]>;
1399 : LoadUnscaled<0b00, 0, 0b10, GPR64, am_unscaled8, "ldursb",
1400 [(set GPR64:$Rt, (sextloadi8 am_unscaled8:$addr))]>;
1402 // load sign-extended word
1404 : LoadUnscaled<0b10, 0, 0b10, GPR64, am_unscaled32, "ldursw",
1405 [(set GPR64:$Rt, (sextloadi32 am_unscaled32:$addr))]>;
1407 // zero and sign extending aliases from generic LDR* mnemonics to LDUR*.
1408 def : InstAlias<"ldrb $Rt, $addr", (LDURBBi GPR32:$Rt, am_unscaled_fb8:$addr)>;
1409 def : InstAlias<"ldrh $Rt, $addr", (LDURHHi GPR32:$Rt, am_unscaled_fb16:$addr)>;
1410 def : InstAlias<"ldrsb $Rt, $addr", (LDURSBWi GPR32:$Rt, am_unscaled_fb8:$addr)>;
1411 def : InstAlias<"ldrsb $Rt, $addr", (LDURSBXi GPR64:$Rt, am_unscaled_fb8:$addr)>;
1412 def : InstAlias<"ldrsh $Rt, $addr", (LDURSHWi GPR32:$Rt, am_unscaled_fb16:$addr)>;
1413 def : InstAlias<"ldrsh $Rt, $addr", (LDURSHXi GPR64:$Rt, am_unscaled_fb16:$addr)>;
1414 def : InstAlias<"ldrsw $Rt, $addr", (LDURSWi GPR64:$Rt, am_unscaled_fb32:$addr)>;
1417 def PRFUMi : PrefetchUnscaled<0b11, 0, 0b10, "prfum",
1418 [(ARM64Prefetch imm:$Rt, am_unscaled64:$addr)]>;
1421 // (unscaled immediate, unprivileged)
1422 def LDTRXi : LoadUnprivileged<0b11, 0, 0b01, GPR64, "ldtr">;
1423 def LDTRWi : LoadUnprivileged<0b10, 0, 0b01, GPR32, "ldtr">;
1425 def LDTRHi : LoadUnprivileged<0b01, 0, 0b01, GPR32, "ldtrh">;
1426 def LDTRBi : LoadUnprivileged<0b00, 0, 0b01, GPR32, "ldtrb">;
1428 // load sign-extended half-word
1429 def LDTRSHWi : LoadUnprivileged<0b01, 0, 0b11, GPR32, "ldtrsh">;
1430 def LDTRSHXi : LoadUnprivileged<0b01, 0, 0b10, GPR64, "ldtrsh">;
1432 // load sign-extended byte
1433 def LDTRSBWi : LoadUnprivileged<0b00, 0, 0b11, GPR32, "ldtrsb">;
1434 def LDTRSBXi : LoadUnprivileged<0b00, 0, 0b10, GPR64, "ldtrsb">;
1436 // load sign-extended word
1437 def LDTRSWi : LoadUnprivileged<0b10, 0, 0b10, GPR64, "ldtrsw">;
1440 // (immediate pre-indexed)
1441 def LDRWpre : LoadPreIdx<0b10, 0, 0b01, GPR32, "ldr">;
1442 def LDRXpre : LoadPreIdx<0b11, 0, 0b01, GPR64, "ldr">;
1443 def LDRBpre : LoadPreIdx<0b00, 1, 0b01, FPR8, "ldr">;
1444 def LDRHpre : LoadPreIdx<0b01, 1, 0b01, FPR16, "ldr">;
1445 def LDRSpre : LoadPreIdx<0b10, 1, 0b01, FPR32, "ldr">;
1446 def LDRDpre : LoadPreIdx<0b11, 1, 0b01, FPR64, "ldr">;
1447 def LDRQpre : LoadPreIdx<0b00, 1, 0b11, FPR128, "ldr">;
1449 // load sign-extended half-word
1450 def LDRSHWpre : LoadPreIdx<0b01, 0, 0b11, GPR32, "ldrsh">;
1451 def LDRSHXpre : LoadPreIdx<0b01, 0, 0b10, GPR64, "ldrsh">;
1453 // load sign-extended byte
1454 def LDRSBWpre : LoadPreIdx<0b00, 0, 0b11, GPR32, "ldrsb">;
1455 def LDRSBXpre : LoadPreIdx<0b00, 0, 0b10, GPR64, "ldrsb">;
1457 // load zero-extended byte
1458 def LDRBBpre : LoadPreIdx<0b00, 0, 0b01, GPR32, "ldrb">;
1459 def LDRHHpre : LoadPreIdx<0b01, 0, 0b01, GPR32, "ldrh">;
1461 // load sign-extended word
1462 def LDRSWpre : LoadPreIdx<0b10, 0, 0b10, GPR64, "ldrsw">;
1464 // ISel pseudos and patterns. See expanded comment on LoadPreIdxPseudo.
1465 def LDRDpre_isel : LoadPreIdxPseudo<FPR64>;
1466 def LDRSpre_isel : LoadPreIdxPseudo<FPR32>;
1467 def LDRXpre_isel : LoadPreIdxPseudo<GPR64>;
1468 def LDRWpre_isel : LoadPreIdxPseudo<GPR32>;
1469 def LDRHHpre_isel : LoadPreIdxPseudo<GPR32>;
1470 def LDRBBpre_isel : LoadPreIdxPseudo<GPR32>;
1472 def LDRSWpre_isel : LoadPreIdxPseudo<GPR64>;
1473 def LDRSHWpre_isel : LoadPreIdxPseudo<GPR32>;
1474 def LDRSHXpre_isel : LoadPreIdxPseudo<GPR64>;
1475 def LDRSBWpre_isel : LoadPreIdxPseudo<GPR32>;
1476 def LDRSBXpre_isel : LoadPreIdxPseudo<GPR64>;
1479 // (immediate post-indexed)
1480 def LDRWpost : LoadPostIdx<0b10, 0, 0b01, GPR32, "ldr">;
1481 def LDRXpost : LoadPostIdx<0b11, 0, 0b01, GPR64, "ldr">;
1482 def LDRBpost : LoadPostIdx<0b00, 1, 0b01, FPR8, "ldr">;
1483 def LDRHpost : LoadPostIdx<0b01, 1, 0b01, FPR16, "ldr">;
1484 def LDRSpost : LoadPostIdx<0b10, 1, 0b01, FPR32, "ldr">;
1485 def LDRDpost : LoadPostIdx<0b11, 1, 0b01, FPR64, "ldr">;
1486 def LDRQpost : LoadPostIdx<0b00, 1, 0b11, FPR128, "ldr">;
1488 // load sign-extended half-word
1489 def LDRSHWpost : LoadPostIdx<0b01, 0, 0b11, GPR32, "ldrsh">;
1490 def LDRSHXpost : LoadPostIdx<0b01, 0, 0b10, GPR64, "ldrsh">;
1492 // load sign-extended byte
1493 def LDRSBWpost : LoadPostIdx<0b00, 0, 0b11, GPR32, "ldrsb">;
1494 def LDRSBXpost : LoadPostIdx<0b00, 0, 0b10, GPR64, "ldrsb">;
1496 // load zero-extended byte
1497 def LDRBBpost : LoadPostIdx<0b00, 0, 0b01, GPR32, "ldrb">;
1498 def LDRHHpost : LoadPostIdx<0b01, 0, 0b01, GPR32, "ldrh">;
1500 // load sign-extended word
1501 def LDRSWpost : LoadPostIdx<0b10, 0, 0b10, GPR64, "ldrsw">;
1503 // ISel pseudos and patterns. See expanded comment on LoadPostIdxPseudo.
1504 def LDRDpost_isel : LoadPostIdxPseudo<FPR64>;
1505 def LDRSpost_isel : LoadPostIdxPseudo<FPR32>;
1506 def LDRXpost_isel : LoadPostIdxPseudo<GPR64>;
1507 def LDRWpost_isel : LoadPostIdxPseudo<GPR32>;
1508 def LDRHHpost_isel : LoadPostIdxPseudo<GPR32>;
1509 def LDRBBpost_isel : LoadPostIdxPseudo<GPR32>;
1511 def LDRSWpost_isel : LoadPostIdxPseudo<GPR64>;
1512 def LDRSHWpost_isel : LoadPostIdxPseudo<GPR32>;
1513 def LDRSHXpost_isel : LoadPostIdxPseudo<GPR64>;
1514 def LDRSBWpost_isel : LoadPostIdxPseudo<GPR32>;
1515 def LDRSBXpost_isel : LoadPostIdxPseudo<GPR64>;
1517 //===----------------------------------------------------------------------===//
1518 // Store instructions.
1519 //===----------------------------------------------------------------------===//
1521 // Pair (indexed, offset)
1522 // FIXME: Use dedicated range-checked addressing mode operand here.
1523 def STPWi : StorePairOffset<0b00, 0, GPR32, am_indexed32simm7, "stp">;
1524 def STPXi : StorePairOffset<0b10, 0, GPR64, am_indexed64simm7, "stp">;
1525 def STPSi : StorePairOffset<0b00, 1, FPR32, am_indexed32simm7, "stp">;
1526 def STPDi : StorePairOffset<0b01, 1, FPR64, am_indexed64simm7, "stp">;
1527 def STPQi : StorePairOffset<0b10, 1, FPR128, am_indexed128simm7, "stp">;
1529 // Pair (pre-indexed)
1530 def STPWpre : StorePairPreIdx<0b00, 0, GPR32, am_indexed32simm7_wb, "stp">;
1531 def STPXpre : StorePairPreIdx<0b10, 0, GPR64, am_indexed64simm7_wb, "stp">;
1532 def STPSpre : StorePairPreIdx<0b00, 1, FPR32, am_indexed32simm7_wb, "stp">;
1533 def STPDpre : StorePairPreIdx<0b01, 1, FPR64, am_indexed64simm7_wb, "stp">;
1534 def STPQpre : StorePairPreIdx<0b10, 1, FPR128, am_indexed128simm7_wb, "stp">;
1536 // Pair (pre-indexed)
1537 def STPWpost : StorePairPostIdx<0b00, 0, GPR32, simm7s4, "stp">;
1538 def STPXpost : StorePairPostIdx<0b10, 0, GPR64, simm7s8, "stp">;
1539 def STPSpost : StorePairPostIdx<0b00, 1, FPR32, simm7s4, "stp">;
1540 def STPDpost : StorePairPostIdx<0b01, 1, FPR64, simm7s8, "stp">;
1541 def STPQpost : StorePairPostIdx<0b10, 1, FPR128, simm7s16, "stp">;
1543 // Pair (no allocate)
1544 def STNPWi : StorePairNoAlloc<0b00, 0, GPR32, am_indexed32simm7, "stnp">;
1545 def STNPXi : StorePairNoAlloc<0b10, 0, GPR64, am_indexed64simm7, "stnp">;
1546 def STNPSi : StorePairNoAlloc<0b00, 1, FPR32, am_indexed32simm7, "stnp">;
1547 def STNPDi : StorePairNoAlloc<0b01, 1, FPR64, am_indexed64simm7, "stnp">;
1548 def STNPQi : StorePairNoAlloc<0b10, 1, FPR128, am_indexed128simm7, "stnp">;
1551 // (Register offset)
1553 let AddedComplexity = 10 in {
1556 def STRHHro : Store16RO<0b01, 0, 0b00, GPR32, "strh",
1557 [(truncstorei16 GPR32:$Rt, ro_indexed16:$addr)]>;
1558 def STRBBro : Store8RO<0b00, 0, 0b00, GPR32, "strb",
1559 [(truncstorei8 GPR32:$Rt, ro_indexed8:$addr)]>;
1560 def STRWro : Store32RO<0b10, 0, 0b00, GPR32, "str",
1561 [(store GPR32:$Rt, ro_indexed32:$addr)]>;
1562 def STRXro : Store64RO<0b11, 0, 0b00, GPR64, "str",
1563 [(store GPR64:$Rt, ro_indexed64:$addr)]>;
1566 def : Pat<(truncstorei8 GPR64:$Rt, ro_indexed8:$addr),
1567 (STRBBro (EXTRACT_SUBREG GPR64:$Rt, sub_32), ro_indexed8:$addr)>;
1568 def : Pat<(truncstorei16 GPR64:$Rt, ro_indexed16:$addr),
1569 (STRHHro (EXTRACT_SUBREG GPR64:$Rt, sub_32), ro_indexed16:$addr)>;
1570 def : Pat<(truncstorei32 GPR64:$Rt, ro_indexed32:$addr),
1571 (STRWro (EXTRACT_SUBREG GPR64:$Rt, sub_32), ro_indexed32:$addr)>;
1575 def STRBro : Store8RO<0b00, 1, 0b00, FPR8, "str",
1576 [(store FPR8:$Rt, ro_indexed8:$addr)]>;
1577 def STRHro : Store16RO<0b01, 1, 0b00, FPR16, "str",
1578 [(store (f16 FPR16:$Rt), ro_indexed16:$addr)]>;
1579 def STRSro : Store32RO<0b10, 1, 0b00, FPR32, "str",
1580 [(store (f32 FPR32:$Rt), ro_indexed32:$addr)]>;
1581 def STRDro : Store64RO<0b11, 1, 0b00, FPR64, "str",
1582 [(store (f64 FPR64:$Rt), ro_indexed64:$addr)]>;
1583 def STRQro : Store128RO<0b00, 1, 0b10, FPR128, "str", []> {
1587 // Match all store 64 bits width whose type is compatible with FPR64
1588 def : Pat<(store (v2f32 FPR64:$Rn), ro_indexed64:$addr),
1589 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1590 def : Pat<(store (v1f64 FPR64:$Rn), ro_indexed64:$addr),
1591 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1592 def : Pat<(store (v8i8 FPR64:$Rn), ro_indexed64:$addr),
1593 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1594 def : Pat<(store (v4i16 FPR64:$Rn), ro_indexed64:$addr),
1595 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1596 def : Pat<(store (v2i32 FPR64:$Rn), ro_indexed64:$addr),
1597 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1598 def : Pat<(store (v1i64 FPR64:$Rn), ro_indexed64:$addr),
1599 (STRDro FPR64:$Rn, ro_indexed64:$addr)>;
1601 // Match all store 128 bits width whose type is compatible with FPR128
1602 def : Pat<(store (v4f32 FPR128:$Rn), ro_indexed128:$addr),
1603 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1604 def : Pat<(store (v2f64 FPR128:$Rn), ro_indexed128:$addr),
1605 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1606 def : Pat<(store (v16i8 FPR128:$Rn), ro_indexed128:$addr),
1607 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1608 def : Pat<(store (v8i16 FPR128:$Rn), ro_indexed128:$addr),
1609 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1610 def : Pat<(store (v4i32 FPR128:$Rn), ro_indexed128:$addr),
1611 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1612 def : Pat<(store (v2i64 FPR128:$Rn), ro_indexed128:$addr),
1613 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1614 def : Pat<(store (f128 FPR128:$Rn), ro_indexed128:$addr),
1615 (STRQro FPR128:$Rn, ro_indexed128:$addr)>;
1618 // (unsigned immediate)
1619 def STRXui : StoreUI<0b11, 0, 0b00, GPR64, am_indexed64, "str",
1620 [(store GPR64:$Rt, am_indexed64:$addr)]>;
1621 def STRWui : StoreUI<0b10, 0, 0b00, GPR32, am_indexed32, "str",
1622 [(store GPR32:$Rt, am_indexed32:$addr)]>;
1623 def STRBui : StoreUI<0b00, 1, 0b00, FPR8, am_indexed8, "str",
1624 [(store FPR8:$Rt, am_indexed8:$addr)]>;
1625 def STRHui : StoreUI<0b01, 1, 0b00, FPR16, am_indexed16, "str",
1626 [(store (f16 FPR16:$Rt), am_indexed16:$addr)]>;
1627 def STRSui : StoreUI<0b10, 1, 0b00, FPR32, am_indexed32, "str",
1628 [(store (f32 FPR32:$Rt), am_indexed32:$addr)]>;
1629 def STRDui : StoreUI<0b11, 1, 0b00, FPR64, am_indexed64, "str",
1630 [(store (f64 FPR64:$Rt), am_indexed64:$addr)]>;
1631 def STRQui : StoreUI<0b00, 1, 0b10, FPR128, am_indexed128, "str", []> {
1635 // Match all store 64 bits width whose type is compatible with FPR64
1636 def : Pat<(store (v2f32 FPR64:$Rn), am_indexed64:$addr),
1637 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1638 def : Pat<(store (v1f64 FPR64:$Rn), am_indexed64:$addr),
1639 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1640 def : Pat<(store (v8i8 FPR64:$Rn), am_indexed64:$addr),
1641 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1642 def : Pat<(store (v4i16 FPR64:$Rn), am_indexed64:$addr),
1643 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1644 def : Pat<(store (v2i32 FPR64:$Rn), am_indexed64:$addr),
1645 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1646 def : Pat<(store (v1i64 FPR64:$Rn), am_indexed64:$addr),
1647 (STRDui FPR64:$Rn, am_indexed64:$addr)>;
1649 // Match all store 128 bits width whose type is compatible with FPR128
1650 def : Pat<(store (v4f32 FPR128:$Rn), am_indexed128:$addr),
1651 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1652 def : Pat<(store (v2f64 FPR128:$Rn), am_indexed128:$addr),
1653 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1654 def : Pat<(store (v16i8 FPR128:$Rn), am_indexed128:$addr),
1655 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1656 def : Pat<(store (v8i16 FPR128:$Rn), am_indexed128:$addr),
1657 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1658 def : Pat<(store (v4i32 FPR128:$Rn), am_indexed128:$addr),
1659 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1660 def : Pat<(store (v2i64 FPR128:$Rn), am_indexed128:$addr),
1661 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1662 def : Pat<(store (f128 FPR128:$Rn), am_indexed128:$addr),
1663 (STRQui FPR128:$Rn, am_indexed128:$addr)>;
1665 def STRHHui : StoreUI<0b01, 0, 0b00, GPR32, am_indexed16, "strh",
1666 [(truncstorei16 GPR32:$Rt, am_indexed16:$addr)]>;
1667 def STRBBui : StoreUI<0b00, 0, 0b00, GPR32, am_indexed8, "strb",
1668 [(truncstorei8 GPR32:$Rt, am_indexed8:$addr)]>;
1671 def : Pat<(truncstorei32 GPR64:$Rt, am_indexed32:$addr),
1672 (STRWui (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_indexed32:$addr)>;
1673 def : Pat<(truncstorei16 GPR64:$Rt, am_indexed16:$addr),
1674 (STRHHui (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_indexed16:$addr)>;
1675 def : Pat<(truncstorei8 GPR64:$Rt, am_indexed8:$addr),
1676 (STRBBui (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_indexed8:$addr)>;
1678 } // AddedComplexity = 10
1681 // (unscaled immediate)
1682 def STURXi : StoreUnscaled<0b11, 0, 0b00, GPR64, am_unscaled64, "stur",
1683 [(store GPR64:$Rt, am_unscaled64:$addr)]>;
1684 def STURWi : StoreUnscaled<0b10, 0, 0b00, GPR32, am_unscaled32, "stur",
1685 [(store GPR32:$Rt, am_unscaled32:$addr)]>;
1686 def STURBi : StoreUnscaled<0b00, 1, 0b00, FPR8, am_unscaled8, "stur",
1687 [(store FPR8:$Rt, am_unscaled8:$addr)]>;
1688 def STURHi : StoreUnscaled<0b01, 1, 0b00, FPR16, am_unscaled16, "stur",
1689 [(store (f16 FPR16:$Rt), am_unscaled16:$addr)]>;
1690 def STURSi : StoreUnscaled<0b10, 1, 0b00, FPR32, am_unscaled32, "stur",
1691 [(store (f32 FPR32:$Rt), am_unscaled32:$addr)]>;
1692 def STURDi : StoreUnscaled<0b11, 1, 0b00, FPR64, am_unscaled64, "stur",
1693 [(store (f64 FPR64:$Rt), am_unscaled64:$addr)]>;
1694 def STURQi : StoreUnscaled<0b00, 1, 0b10, FPR128, am_unscaled128, "stur",
1695 [(store (v2f64 FPR128:$Rt), am_unscaled128:$addr)]>;
1696 def STURHHi : StoreUnscaled<0b01, 0, 0b00, GPR32, am_unscaled16, "sturh",
1697 [(truncstorei16 GPR32:$Rt, am_unscaled16:$addr)]>;
1698 def STURBBi : StoreUnscaled<0b00, 0, 0b00, GPR32, am_unscaled8, "sturb",
1699 [(truncstorei8 GPR32:$Rt, am_unscaled8:$addr)]>;
1701 // Match all store 64 bits width whose type is compatible with FPR64
1702 def : Pat<(store (v2f32 FPR64:$Rn), am_unscaled64:$addr),
1703 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1704 def : Pat<(store (v1f64 FPR64:$Rn), am_unscaled64:$addr),
1705 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1706 def : Pat<(store (v8i8 FPR64:$Rn), am_unscaled64:$addr),
1707 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1708 def : Pat<(store (v4i16 FPR64:$Rn), am_unscaled64:$addr),
1709 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1710 def : Pat<(store (v2i32 FPR64:$Rn), am_unscaled64:$addr),
1711 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1712 def : Pat<(store (v1i64 FPR64:$Rn), am_unscaled64:$addr),
1713 (STURDi FPR64:$Rn, am_unscaled64:$addr)>;
1715 // Match all store 128 bits width whose type is compatible with FPR128
1716 def : Pat<(store (v4f32 FPR128:$Rn), am_unscaled128:$addr),
1717 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1718 def : Pat<(store (v2f64 FPR128:$Rn), am_unscaled128:$addr),
1719 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1720 def : Pat<(store (v16i8 FPR128:$Rn), am_unscaled128:$addr),
1721 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1722 def : Pat<(store (v8i16 FPR128:$Rn), am_unscaled128:$addr),
1723 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1724 def : Pat<(store (v4i32 FPR128:$Rn), am_unscaled128:$addr),
1725 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1726 def : Pat<(store (v2i64 FPR128:$Rn), am_unscaled128:$addr),
1727 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1728 def : Pat<(store (f128 FPR128:$Rn), am_unscaled128:$addr),
1729 (STURQi FPR128:$Rn, am_unscaled128:$addr)>;
1731 // unscaled i64 truncating stores
1732 def : Pat<(truncstorei32 GPR64:$Rt, am_unscaled32:$addr),
1733 (STURWi (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_unscaled32:$addr)>;
1734 def : Pat<(truncstorei16 GPR64:$Rt, am_unscaled16:$addr),
1735 (STURHHi (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_unscaled16:$addr)>;
1736 def : Pat<(truncstorei8 GPR64:$Rt, am_unscaled8:$addr),
1737 (STURBBi (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_unscaled8:$addr)>;
1740 // STR mnemonics fall back to STUR for negative or unaligned offsets.
1741 def : InstAlias<"str $Rt, $addr", (STURXi GPR64:$Rt, am_unscaled_fb64:$addr)>;
1742 def : InstAlias<"str $Rt, $addr", (STURWi GPR32:$Rt, am_unscaled_fb32:$addr)>;
1743 def : InstAlias<"str $Rt, $addr", (STURBi FPR8:$Rt, am_unscaled_fb8:$addr)>;
1744 def : InstAlias<"str $Rt, $addr", (STURHi FPR16:$Rt, am_unscaled_fb16:$addr)>;
1745 def : InstAlias<"str $Rt, $addr", (STURSi FPR32:$Rt, am_unscaled_fb32:$addr)>;
1746 def : InstAlias<"str $Rt, $addr", (STURDi FPR64:$Rt, am_unscaled_fb64:$addr)>;
1747 def : InstAlias<"str $Rt, $addr", (STURQi FPR128:$Rt, am_unscaled_fb128:$addr)>;
1749 def : InstAlias<"strb $Rt, $addr", (STURBBi GPR32:$Rt, am_unscaled_fb8:$addr)>;
1750 def : InstAlias<"strh $Rt, $addr", (STURHHi GPR32:$Rt, am_unscaled_fb16:$addr)>;
1753 // (unscaled immediate, unprivileged)
1754 def STTRWi : StoreUnprivileged<0b10, 0, 0b00, GPR32, "sttr">;
1755 def STTRXi : StoreUnprivileged<0b11, 0, 0b00, GPR64, "sttr">;
1757 def STTRHi : StoreUnprivileged<0b01, 0, 0b00, GPR32, "sttrh">;
1758 def STTRBi : StoreUnprivileged<0b00, 0, 0b00, GPR32, "sttrb">;
1761 // (immediate pre-indexed)
1762 def STRWpre : StorePreIdx<0b10, 0, 0b00, GPR32, "str">;
1763 def STRXpre : StorePreIdx<0b11, 0, 0b00, GPR64, "str">;
1764 def STRBpre : StorePreIdx<0b00, 1, 0b00, FPR8, "str">;
1765 def STRHpre : StorePreIdx<0b01, 1, 0b00, FPR16, "str">;
1766 def STRSpre : StorePreIdx<0b10, 1, 0b00, FPR32, "str">;
1767 def STRDpre : StorePreIdx<0b11, 1, 0b00, FPR64, "str">;
1768 def STRQpre : StorePreIdx<0b00, 1, 0b10, FPR128, "str">;
1770 def STRBBpre : StorePreIdx<0b00, 0, 0b00, GPR32, "strb">;
1771 def STRHHpre : StorePreIdx<0b01, 0, 0b00, GPR32, "strh">;
1773 // ISel pseudos and patterns. See expanded comment on StorePreIdxPseudo.
1774 defm STRDpre : StorePreIdxPseudo<FPR64, f64, pre_store>;
1775 defm STRSpre : StorePreIdxPseudo<FPR32, f32, pre_store>;
1776 defm STRXpre : StorePreIdxPseudo<GPR64, i64, pre_store>;
1777 defm STRWpre : StorePreIdxPseudo<GPR32, i32, pre_store>;
1778 defm STRHHpre : StorePreIdxPseudo<GPR32, i32, pre_truncsti16>;
1779 defm STRBBpre : StorePreIdxPseudo<GPR32, i32, pre_truncsti8>;
1781 def : Pat<(pre_truncsti32 GPR64:$Rt, am_noindex:$addr, simm9:$off),
1782 (STRWpre_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
1784 def : Pat<(pre_truncsti16 GPR64:$Rt, am_noindex:$addr, simm9:$off),
1785 (STRHHpre_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
1787 def : Pat<(pre_truncsti8 GPR64:$Rt, am_noindex:$addr, simm9:$off),
1788 (STRBBpre_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
1792 // (immediate post-indexed)
1793 def STRWpost : StorePostIdx<0b10, 0, 0b00, GPR32, "str">;
1794 def STRXpost : StorePostIdx<0b11, 0, 0b00, GPR64, "str">;
1795 def STRBpost : StorePostIdx<0b00, 1, 0b00, FPR8, "str">;
1796 def STRHpost : StorePostIdx<0b01, 1, 0b00, FPR16, "str">;
1797 def STRSpost : StorePostIdx<0b10, 1, 0b00, FPR32, "str">;
1798 def STRDpost : StorePostIdx<0b11, 1, 0b00, FPR64, "str">;
1799 def STRQpost : StorePostIdx<0b00, 1, 0b10, FPR128, "str">;
1801 def STRBBpost : StorePostIdx<0b00, 0, 0b00, GPR32, "strb">;
1802 def STRHHpost : StorePostIdx<0b01, 0, 0b00, GPR32, "strh">;
1804 // ISel pseudos and patterns. See expanded comment on StorePostIdxPseudo.
1805 defm STRDpost : StorePostIdxPseudo<FPR64, f64, post_store, STRDpost>;
1806 defm STRSpost : StorePostIdxPseudo<FPR32, f32, post_store, STRSpost>;
1807 defm STRXpost : StorePostIdxPseudo<GPR64, i64, post_store, STRXpost>;
1808 defm STRWpost : StorePostIdxPseudo<GPR32, i32, post_store, STRWpost>;
1809 defm STRHHpost : StorePostIdxPseudo<GPR32, i32, post_truncsti16, STRHHpost>;
1810 defm STRBBpost : StorePostIdxPseudo<GPR32, i32, post_truncsti8, STRBBpost>;
1812 def : Pat<(post_truncsti32 GPR64:$Rt, am_noindex:$addr, simm9:$off),
1813 (STRWpost_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
1815 def : Pat<(post_truncsti16 GPR64:$Rt, am_noindex:$addr, simm9:$off),
1816 (STRHHpost_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
1818 def : Pat<(post_truncsti8 GPR64:$Rt, am_noindex:$addr, simm9:$off),
1819 (STRBBpost_isel (EXTRACT_SUBREG GPR64:$Rt, sub_32), am_noindex:$addr,
1823 //===----------------------------------------------------------------------===//
1824 // Load/store exclusive instructions.
1825 //===----------------------------------------------------------------------===//
1827 def LDARW : LoadAcquire <0b10, 1, 1, 0, 1, GPR32, "ldar">;
1828 def LDARX : LoadAcquire <0b11, 1, 1, 0, 1, GPR64, "ldar">;
1829 def LDARB : LoadAcquire <0b00, 1, 1, 0, 1, GPR32, "ldarb">;
1830 def LDARH : LoadAcquire <0b01, 1, 1, 0, 1, GPR32, "ldarh">;
1832 def LDAXRW : LoadExclusive <0b10, 0, 1, 0, 1, GPR32, "ldaxr">;
1833 def LDAXRX : LoadExclusive <0b11, 0, 1, 0, 1, GPR64, "ldaxr">;
1834 def LDAXRB : LoadExclusive <0b00, 0, 1, 0, 1, GPR32, "ldaxrb">;
1835 def LDAXRH : LoadExclusive <0b01, 0, 1, 0, 1, GPR32, "ldaxrh">;
1837 def LDXRW : LoadExclusive <0b10, 0, 1, 0, 0, GPR32, "ldxr">;
1838 def LDXRX : LoadExclusive <0b11, 0, 1, 0, 0, GPR64, "ldxr">;
1839 def LDXRB : LoadExclusive <0b00, 0, 1, 0, 0, GPR32, "ldxrb">;
1840 def LDXRH : LoadExclusive <0b01, 0, 1, 0, 0, GPR32, "ldxrh">;
1842 def STLRW : StoreRelease <0b10, 1, 0, 0, 1, GPR32, "stlr">;
1843 def STLRX : StoreRelease <0b11, 1, 0, 0, 1, GPR64, "stlr">;
1844 def STLRB : StoreRelease <0b00, 1, 0, 0, 1, GPR32, "stlrb">;
1845 def STLRH : StoreRelease <0b01, 1, 0, 0, 1, GPR32, "stlrh">;
1847 def STLXRW : StoreExclusive<0b10, 0, 0, 0, 1, GPR32, "stlxr">;
1848 def STLXRX : StoreExclusive<0b11, 0, 0, 0, 1, GPR64, "stlxr">;
1849 def STLXRB : StoreExclusive<0b00, 0, 0, 0, 1, GPR32, "stlxrb">;
1850 def STLXRH : StoreExclusive<0b01, 0, 0, 0, 1, GPR32, "stlxrh">;
1852 def STXRW : StoreExclusive<0b10, 0, 0, 0, 0, GPR32, "stxr">;
1853 def STXRX : StoreExclusive<0b11, 0, 0, 0, 0, GPR64, "stxr">;
1854 def STXRB : StoreExclusive<0b00, 0, 0, 0, 0, GPR32, "stxrb">;
1855 def STXRH : StoreExclusive<0b01, 0, 0, 0, 0, GPR32, "stxrh">;
1857 def LDAXPW : LoadExclusivePair<0b10, 0, 1, 1, 1, GPR32, "ldaxp">;
1858 def LDAXPX : LoadExclusivePair<0b11, 0, 1, 1, 1, GPR64, "ldaxp">;
1860 def LDXPW : LoadExclusivePair<0b10, 0, 1, 1, 0, GPR32, "ldxp">;
1861 def LDXPX : LoadExclusivePair<0b11, 0, 1, 1, 0, GPR64, "ldxp">;
1863 def STLXPW : StoreExclusivePair<0b10, 0, 0, 1, 1, GPR32, "stlxp">;
1864 def STLXPX : StoreExclusivePair<0b11, 0, 0, 1, 1, GPR64, "stlxp">;
1866 def STXPW : StoreExclusivePair<0b10, 0, 0, 1, 0, GPR32, "stxp">;
1867 def STXPX : StoreExclusivePair<0b11, 0, 0, 1, 0, GPR64, "stxp">;
1869 //===----------------------------------------------------------------------===//
1870 // Scaled floating point to integer conversion instructions.
1871 //===----------------------------------------------------------------------===//
1873 defm FCVTAS : FPToIntegerUnscaled<0b00, 0b100, "fcvtas", int_arm64_neon_fcvtas>;
1874 defm FCVTAU : FPToIntegerUnscaled<0b00, 0b101, "fcvtau", int_arm64_neon_fcvtau>;
1875 defm FCVTMS : FPToIntegerUnscaled<0b10, 0b000, "fcvtms", int_arm64_neon_fcvtms>;
1876 defm FCVTMU : FPToIntegerUnscaled<0b10, 0b001, "fcvtmu", int_arm64_neon_fcvtmu>;
1877 defm FCVTNS : FPToIntegerUnscaled<0b00, 0b000, "fcvtns", int_arm64_neon_fcvtns>;
1878 defm FCVTNU : FPToIntegerUnscaled<0b00, 0b001, "fcvtnu", int_arm64_neon_fcvtnu>;
1879 defm FCVTPS : FPToIntegerUnscaled<0b01, 0b000, "fcvtps", int_arm64_neon_fcvtps>;
1880 defm FCVTPU : FPToIntegerUnscaled<0b01, 0b001, "fcvtpu", int_arm64_neon_fcvtpu>;
1881 defm FCVTZS : FPToIntegerUnscaled<0b11, 0b000, "fcvtzs", fp_to_sint>;
1882 defm FCVTZU : FPToIntegerUnscaled<0b11, 0b001, "fcvtzu", fp_to_uint>;
1883 defm FCVTZS : FPToIntegerScaled<0b11, 0b000, "fcvtzs", fp_to_sint>;
1884 defm FCVTZU : FPToIntegerScaled<0b11, 0b001, "fcvtzu", fp_to_uint>;
1885 let isCodeGenOnly = 1 in {
1886 defm FCVTZS_Int : FPToIntegerUnscaled<0b11, 0b000, "fcvtzs", int_arm64_neon_fcvtzs>;
1887 defm FCVTZU_Int : FPToIntegerUnscaled<0b11, 0b001, "fcvtzu", int_arm64_neon_fcvtzu>;
1888 defm FCVTZS_Int : FPToIntegerScaled<0b11, 0b000, "fcvtzs", int_arm64_neon_fcvtzs>;
1889 defm FCVTZU_Int : FPToIntegerScaled<0b11, 0b001, "fcvtzu", int_arm64_neon_fcvtzu>;
1892 //===----------------------------------------------------------------------===//
1893 // Scaled integer to floating point conversion instructions.
1894 //===----------------------------------------------------------------------===//
1896 defm SCVTF : IntegerToFP<0, "scvtf", sint_to_fp>;
1897 defm UCVTF : IntegerToFP<1, "ucvtf", uint_to_fp>;
1899 //===----------------------------------------------------------------------===//
1900 // Unscaled integer to floating point conversion instruction.
1901 //===----------------------------------------------------------------------===//
1903 defm FMOV : UnscaledConversion<"fmov">;
1905 def : Pat<(f32 (fpimm0)), (FMOVWSr WZR)>, Requires<[NoZCZ]>;
1906 def : Pat<(f64 (fpimm0)), (FMOVXDr XZR)>, Requires<[NoZCZ]>;
1908 def : Pat<(v8i8 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1909 def : Pat<(v4i16 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1910 def : Pat<(v2i32 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1911 def : Pat<(v1i64 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1912 def : Pat<(v2f32 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1913 def : Pat<(v1f64 (bitconvert GPR64:$Xn)), (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1914 def : Pat<(v1i64 (scalar_to_vector GPR64:$Xn)),
1915 (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1916 def : Pat<(v1f64 (scalar_to_vector GPR64:$Xn)),
1917 (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1918 def : Pat<(v1f64 (scalar_to_vector (f64 FPR64:$Xn))), (v1f64 FPR64:$Xn)>;
1920 def : Pat<(i64 (bitconvert (v8i8 V64:$Vn))),
1921 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
1922 def : Pat<(i64 (bitconvert (v4i16 V64:$Vn))),
1923 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
1924 def : Pat<(i64 (bitconvert (v2i32 V64:$Vn))),
1925 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
1926 def : Pat<(i64 (bitconvert (v1i64 V64:$Vn))),
1927 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
1928 def : Pat<(i64 (bitconvert (v2f32 V64:$Vn))),
1929 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
1930 def : Pat<(i64 (bitconvert (v1f64 V64:$Vn))),
1931 (COPY_TO_REGCLASS V64:$Vn, GPR64)>;
1933 def : Pat<(f32 (bitconvert (i32 GPR32:$Xn))),
1934 (COPY_TO_REGCLASS GPR32:$Xn, FPR32)>;
1935 def : Pat<(i32 (bitconvert (f32 FPR32:$Xn))),
1936 (COPY_TO_REGCLASS FPR32:$Xn, GPR32)>;
1937 def : Pat<(f64 (bitconvert (i64 GPR64:$Xn))),
1938 (COPY_TO_REGCLASS GPR64:$Xn, FPR64)>;
1939 def : Pat<(i64 (bitconvert (f64 FPR64:$Xn))),
1940 (COPY_TO_REGCLASS FPR64:$Xn, GPR64)>;
1942 //===----------------------------------------------------------------------===//
1943 // Floating point conversion instruction.
1944 //===----------------------------------------------------------------------===//
1946 defm FCVT : FPConversion<"fcvt">;
1948 def : Pat<(f32_to_f16 FPR32:$Rn),
1949 (i32 (COPY_TO_REGCLASS
1950 (f32 (SUBREG_TO_REG (i32 0), (FCVTHSr FPR32:$Rn), hsub)),
1953 def FCVTSHpseudo : Pseudo<(outs FPR32:$Rd), (ins FPR32:$Rn),
1954 [(set (f32 FPR32:$Rd), (f16_to_f32 i32:$Rn))]>;
1956 //===----------------------------------------------------------------------===//
1957 // Floating point single operand instructions.
1958 //===----------------------------------------------------------------------===//
1960 defm FABS : SingleOperandFPData<0b0001, "fabs", fabs>;
1961 defm FMOV : SingleOperandFPData<0b0000, "fmov">;
1962 defm FNEG : SingleOperandFPData<0b0010, "fneg", fneg>;
1963 defm FRINTA : SingleOperandFPData<0b1100, "frinta", frnd>;
1964 defm FRINTI : SingleOperandFPData<0b1111, "frinti", fnearbyint>;
1965 defm FRINTM : SingleOperandFPData<0b1010, "frintm", ffloor>;
1966 defm FRINTN : SingleOperandFPData<0b1000, "frintn", int_arm64_neon_frintn>;
1967 defm FRINTP : SingleOperandFPData<0b1001, "frintp", fceil>;
1969 def : Pat<(v1f64 (int_arm64_neon_frintn (v1f64 FPR64:$Rn))),
1970 (FRINTNDr FPR64:$Rn)>;
1972 // FRINTX is inserted to set the flags as required by FENV_ACCESS ON behavior
1973 // in the C spec. Setting hasSideEffects ensures it is not DCE'd.
1974 // <rdar://problem/13715968>
1975 // TODO: We should really model the FPSR flags correctly. This is really ugly.
1976 let hasSideEffects = 1 in {
1977 defm FRINTX : SingleOperandFPData<0b1110, "frintx", frint>;
1980 defm FRINTZ : SingleOperandFPData<0b1011, "frintz", ftrunc>;
1982 let SchedRW = [WriteFDiv] in {
1983 defm FSQRT : SingleOperandFPData<0b0011, "fsqrt", fsqrt>;
1986 //===----------------------------------------------------------------------===//
1987 // Floating point two operand instructions.
1988 //===----------------------------------------------------------------------===//
1990 defm FADD : TwoOperandFPData<0b0010, "fadd", fadd>;
1991 let SchedRW = [WriteFDiv] in {
1992 defm FDIV : TwoOperandFPData<0b0001, "fdiv", fdiv>;
1994 defm FMAXNM : TwoOperandFPData<0b0110, "fmaxnm", int_arm64_neon_fmaxnm>;
1995 defm FMAX : TwoOperandFPData<0b0100, "fmax", ARM64fmax>;
1996 defm FMINNM : TwoOperandFPData<0b0111, "fminnm", int_arm64_neon_fminnm>;
1997 defm FMIN : TwoOperandFPData<0b0101, "fmin", ARM64fmin>;
1998 let SchedRW = [WriteFMul] in {
1999 defm FMUL : TwoOperandFPData<0b0000, "fmul", fmul>;
2000 defm FNMUL : TwoOperandFPDataNeg<0b1000, "fnmul", fmul>;
2002 defm FSUB : TwoOperandFPData<0b0011, "fsub", fsub>;
2004 def : Pat<(v1f64 (ARM64fmax (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2005 (FMAXDrr FPR64:$Rn, FPR64:$Rm)>;
2006 def : Pat<(v1f64 (ARM64fmin (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2007 (FMINDrr FPR64:$Rn, FPR64:$Rm)>;
2008 def : Pat<(v1f64 (int_arm64_neon_fmaxnm (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2009 (FMAXNMDrr FPR64:$Rn, FPR64:$Rm)>;
2010 def : Pat<(v1f64 (int_arm64_neon_fminnm (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2011 (FMINNMDrr FPR64:$Rn, FPR64:$Rm)>;
2013 //===----------------------------------------------------------------------===//
2014 // Floating point three operand instructions.
2015 //===----------------------------------------------------------------------===//
2017 defm FMADD : ThreeOperandFPData<0, 0, "fmadd", fma>;
2018 defm FMSUB : ThreeOperandFPData<0, 1, "fmsub",
2019 TriOpFrag<(fma node:$LHS, (fneg node:$MHS), node:$RHS)> >;
2020 defm FNMADD : ThreeOperandFPData<1, 0, "fnmadd",
2021 TriOpFrag<(fneg (fma node:$LHS, node:$MHS, node:$RHS))> >;
2022 defm FNMSUB : ThreeOperandFPData<1, 1, "fnmsub",
2023 TriOpFrag<(fma node:$LHS, node:$MHS, (fneg node:$RHS))> >;
2025 // The following def pats catch the case where the LHS of an FMA is negated.
2026 // The TriOpFrag above catches the case where the middle operand is negated.
2028 // N.b. FMSUB etc have the accumulator at the *end* of (outs), unlike
2029 // the NEON variant.
2030 def : Pat<(f32 (fma (fneg FPR32:$Rn), FPR32:$Rm, FPR32:$Ra)),
2031 (FMSUBSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
2033 def : Pat<(f64 (fma (fneg FPR64:$Rn), FPR64:$Rm, FPR64:$Ra)),
2034 (FMSUBDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
2036 // We handled -(a + b*c) for FNMADD above, now it's time for "(-a) + (-b)*c" and
2038 def : Pat<(f32 (fma (fneg FPR32:$Rn), FPR32:$Rm, (fneg FPR32:$Ra))),
2039 (FNMADDSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
2041 def : Pat<(f64 (fma (fneg FPR64:$Rn), FPR64:$Rm, (fneg FPR64:$Ra))),
2042 (FNMADDDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
2044 def : Pat<(f32 (fma FPR32:$Rn, (fneg FPR32:$Rm), (fneg FPR32:$Ra))),
2045 (FNMADDSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
2047 def : Pat<(f64 (fma FPR64:$Rn, (fneg FPR64:$Rm), (fneg FPR64:$Ra))),
2048 (FNMADDDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
2050 //===----------------------------------------------------------------------===//
2051 // Floating point comparison instructions.
2052 //===----------------------------------------------------------------------===//
2054 defm FCMPE : FPComparison<1, "fcmpe">;
2055 defm FCMP : FPComparison<0, "fcmp", ARM64fcmp>;
2057 //===----------------------------------------------------------------------===//
2058 // Floating point conditional comparison instructions.
2059 //===----------------------------------------------------------------------===//
2061 defm FCCMPE : FPCondComparison<1, "fccmpe">;
2062 defm FCCMP : FPCondComparison<0, "fccmp">;
2064 //===----------------------------------------------------------------------===//
2065 // Floating point conditional select instruction.
2066 //===----------------------------------------------------------------------===//
2068 defm FCSEL : FPCondSelect<"fcsel">;
2070 // CSEL instructions providing f128 types need to be handled by a
2071 // pseudo-instruction since the eventual code will need to introduce basic
2072 // blocks and control flow.
2073 def F128CSEL : Pseudo<(outs FPR128:$Rd),
2074 (ins FPR128:$Rn, FPR128:$Rm, ccode:$cond),
2075 [(set (f128 FPR128:$Rd),
2076 (ARM64csel FPR128:$Rn, FPR128:$Rm,
2077 (i32 imm:$cond), CPSR))]> {
2079 let usesCustomInserter = 1;
2083 //===----------------------------------------------------------------------===//
2084 // Floating point immediate move.
2085 //===----------------------------------------------------------------------===//
2087 let isReMaterializable = 1 in {
2088 defm FMOV : FPMoveImmediate<"fmov">;
2091 //===----------------------------------------------------------------------===//
2092 // Advanced SIMD two vector instructions.
2093 //===----------------------------------------------------------------------===//
2095 defm ABS : SIMDTwoVectorBHSD<0, 0b01011, "abs", int_arm64_neon_abs>;
2096 defm CLS : SIMDTwoVectorBHS<0, 0b00100, "cls", int_arm64_neon_cls>;
2097 defm CLZ : SIMDTwoVectorBHS<1, 0b00100, "clz", ctlz>;
2098 defm CMEQ : SIMDCmpTwoVector<0, 0b01001, "cmeq", ARM64cmeqz>;
2099 defm CMGE : SIMDCmpTwoVector<1, 0b01000, "cmge", ARM64cmgez>;
2100 defm CMGT : SIMDCmpTwoVector<0, 0b01000, "cmgt", ARM64cmgtz>;
2101 defm CMLE : SIMDCmpTwoVector<1, 0b01001, "cmle", ARM64cmlez>;
2102 defm CMLT : SIMDCmpTwoVector<0, 0b01010, "cmlt", ARM64cmltz>;
2103 defm CNT : SIMDTwoVectorB<0, 0b00, 0b00101, "cnt", ctpop>;
2104 defm FABS : SIMDTwoVectorFP<0, 1, 0b01111, "fabs", fabs>;
2106 defm FCMEQ : SIMDFPCmpTwoVector<0, 1, 0b01101, "fcmeq", ARM64fcmeqz>;
2107 defm FCMGE : SIMDFPCmpTwoVector<1, 1, 0b01100, "fcmge", ARM64fcmgez>;
2108 defm FCMGT : SIMDFPCmpTwoVector<0, 1, 0b01100, "fcmgt", ARM64fcmgtz>;
2109 defm FCMLE : SIMDFPCmpTwoVector<1, 1, 0b01101, "fcmle", ARM64fcmlez>;
2110 defm FCMLT : SIMDFPCmpTwoVector<0, 1, 0b01110, "fcmlt", ARM64fcmltz>;
2111 defm FCVTAS : SIMDTwoVectorFPToInt<0,0,0b11100, "fcvtas",int_arm64_neon_fcvtas>;
2112 defm FCVTAU : SIMDTwoVectorFPToInt<1,0,0b11100, "fcvtau",int_arm64_neon_fcvtau>;
2113 defm FCVTL : SIMDFPWidenTwoVector<0, 0, 0b10111, "fcvtl">;
2114 def : Pat<(v4f32 (int_arm64_neon_vcvthf2fp (v4i16 V64:$Rn))),
2115 (FCVTLv4i16 V64:$Rn)>;
2116 def : Pat<(v4f32 (int_arm64_neon_vcvthf2fp (extract_subvector (v8i16 V128:$Rn),
2118 (FCVTLv8i16 V128:$Rn)>;
2119 def : Pat<(v2f64 (fextend (v2f32 V64:$Rn))), (FCVTLv2i32 V64:$Rn)>;
2120 def : Pat<(v2f64 (fextend (v2f32 (extract_subvector (v4f32 V128:$Rn),
2122 (FCVTLv4i32 V128:$Rn)>;
2124 defm FCVTMS : SIMDTwoVectorFPToInt<0,0,0b11011, "fcvtms",int_arm64_neon_fcvtms>;
2125 defm FCVTMU : SIMDTwoVectorFPToInt<1,0,0b11011, "fcvtmu",int_arm64_neon_fcvtmu>;
2126 defm FCVTNS : SIMDTwoVectorFPToInt<0,0,0b11010, "fcvtns",int_arm64_neon_fcvtns>;
2127 defm FCVTNU : SIMDTwoVectorFPToInt<1,0,0b11010, "fcvtnu",int_arm64_neon_fcvtnu>;
2128 defm FCVTN : SIMDFPNarrowTwoVector<0, 0, 0b10110, "fcvtn">;
2129 def : Pat<(v4i16 (int_arm64_neon_vcvtfp2hf (v4f32 V128:$Rn))),
2130 (FCVTNv4i16 V128:$Rn)>;
2131 def : Pat<(concat_vectors V64:$Rd,
2132 (v4i16 (int_arm64_neon_vcvtfp2hf (v4f32 V128:$Rn)))),
2133 (FCVTNv8i16 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
2134 def : Pat<(v2f32 (fround (v2f64 V128:$Rn))), (FCVTNv2i32 V128:$Rn)>;
2135 def : Pat<(concat_vectors V64:$Rd, (v2f32 (fround (v2f64 V128:$Rn)))),
2136 (FCVTNv4i32 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
2137 defm FCVTPS : SIMDTwoVectorFPToInt<0,1,0b11010, "fcvtps",int_arm64_neon_fcvtps>;
2138 defm FCVTPU : SIMDTwoVectorFPToInt<1,1,0b11010, "fcvtpu",int_arm64_neon_fcvtpu>;
2139 defm FCVTXN : SIMDFPInexactCvtTwoVector<1, 0, 0b10110, "fcvtxn",
2140 int_arm64_neon_fcvtxn>;
2141 defm FCVTZS : SIMDTwoVectorFPToInt<0, 1, 0b11011, "fcvtzs", fp_to_sint>;
2142 defm FCVTZU : SIMDTwoVectorFPToInt<1, 1, 0b11011, "fcvtzu", fp_to_uint>;
2143 let isCodeGenOnly = 1 in {
2144 defm FCVTZS_Int : SIMDTwoVectorFPToInt<0, 1, 0b11011, "fcvtzs",
2145 int_arm64_neon_fcvtzs>;
2146 defm FCVTZU_Int : SIMDTwoVectorFPToInt<1, 1, 0b11011, "fcvtzu",
2147 int_arm64_neon_fcvtzu>;
2149 defm FNEG : SIMDTwoVectorFP<1, 1, 0b01111, "fneg", fneg>;
2150 defm FRECPE : SIMDTwoVectorFP<0, 1, 0b11101, "frecpe", int_arm64_neon_frecpe>;
2151 defm FRINTA : SIMDTwoVectorFP<1, 0, 0b11000, "frinta", frnd>;
2152 defm FRINTI : SIMDTwoVectorFP<1, 1, 0b11001, "frinti", fnearbyint>;
2153 defm FRINTM : SIMDTwoVectorFP<0, 0, 0b11001, "frintm", ffloor>;
2154 defm FRINTN : SIMDTwoVectorFP<0, 0, 0b11000, "frintn", int_arm64_neon_frintn>;
2155 defm FRINTP : SIMDTwoVectorFP<0, 1, 0b11000, "frintp", fceil>;
2156 defm FRINTX : SIMDTwoVectorFP<1, 0, 0b11001, "frintx", frint>;
2157 defm FRINTZ : SIMDTwoVectorFP<0, 1, 0b11001, "frintz", ftrunc>;
2158 defm FRSQRTE: SIMDTwoVectorFP<1, 1, 0b11101, "frsqrte", int_arm64_neon_frsqrte>;
2159 defm FSQRT : SIMDTwoVectorFP<1, 1, 0b11111, "fsqrt", fsqrt>;
2160 defm NEG : SIMDTwoVectorBHSD<1, 0b01011, "neg",
2161 UnOpFrag<(sub immAllZerosV, node:$LHS)> >;
2162 defm NOT : SIMDTwoVectorB<1, 0b00, 0b00101, "not", vnot>;
2163 // Aliases for MVN -> NOT.
2164 def : InstAlias<"mvn.8b $Vd, $Vn", (NOTv8i8 V64:$Vd, V64:$Vn)>;
2165 def : InstAlias<"mvn.16b $Vd, $Vn", (NOTv16i8 V128:$Vd, V128:$Vn)>;
2166 def : InstAlias<"mvn $Vd.8b, $Vn.8b", (NOTv8i8 V64:$Vd, V64:$Vn)>;
2167 def : InstAlias<"mvn $Vd.16b, $Vn.16b", (NOTv16i8 V128:$Vd, V128:$Vn)>;
2169 def : Pat<(ARM64neg (v8i8 V64:$Rn)), (NEGv8i8 V64:$Rn)>;
2170 def : Pat<(ARM64neg (v16i8 V128:$Rn)), (NEGv16i8 V128:$Rn)>;
2171 def : Pat<(ARM64neg (v4i16 V64:$Rn)), (NEGv4i16 V64:$Rn)>;
2172 def : Pat<(ARM64neg (v8i16 V128:$Rn)), (NEGv8i16 V128:$Rn)>;
2173 def : Pat<(ARM64neg (v2i32 V64:$Rn)), (NEGv2i32 V64:$Rn)>;
2174 def : Pat<(ARM64neg (v4i32 V128:$Rn)), (NEGv4i32 V128:$Rn)>;
2175 def : Pat<(ARM64neg (v2i64 V128:$Rn)), (NEGv2i64 V128:$Rn)>;
2177 def : Pat<(ARM64not (v8i8 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2178 def : Pat<(ARM64not (v16i8 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2179 def : Pat<(ARM64not (v4i16 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2180 def : Pat<(ARM64not (v8i16 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2181 def : Pat<(ARM64not (v2i32 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2182 def : Pat<(ARM64not (v1i64 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2183 def : Pat<(ARM64not (v4i32 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2184 def : Pat<(ARM64not (v2i64 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2186 def : Pat<(vnot (v4i16 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2187 def : Pat<(vnot (v8i16 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2188 def : Pat<(vnot (v2i32 V64:$Rn)), (NOTv8i8 V64:$Rn)>;
2189 def : Pat<(vnot (v4i32 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2190 def : Pat<(vnot (v2i64 V128:$Rn)), (NOTv16i8 V128:$Rn)>;
2192 defm RBIT : SIMDTwoVectorB<1, 0b01, 0b00101, "rbit", int_arm64_neon_rbit>;
2193 defm REV16 : SIMDTwoVectorB<0, 0b00, 0b00001, "rev16", ARM64rev16>;
2194 defm REV32 : SIMDTwoVectorBH<1, 0b00000, "rev32", ARM64rev32>;
2195 defm REV64 : SIMDTwoVectorBHS<0, 0b00000, "rev64", ARM64rev64>;
2196 defm SADALP : SIMDLongTwoVectorTied<0, 0b00110, "sadalp",
2197 BinOpFrag<(add node:$LHS, (int_arm64_neon_saddlp node:$RHS))> >;
2198 defm SADDLP : SIMDLongTwoVector<0, 0b00010, "saddlp", int_arm64_neon_saddlp>;
2199 defm SCVTF : SIMDTwoVectorIntToFP<0, 0, 0b11101, "scvtf", sint_to_fp>;
2200 defm SHLL : SIMDVectorLShiftLongBySizeBHS;
2201 defm SQABS : SIMDTwoVectorBHSD<0, 0b00111, "sqabs", int_arm64_neon_sqabs>;
2202 defm SQNEG : SIMDTwoVectorBHSD<1, 0b00111, "sqneg", int_arm64_neon_sqneg>;
2203 defm SQXTN : SIMDMixedTwoVector<0, 0b10100, "sqxtn", int_arm64_neon_sqxtn>;
2204 defm SQXTUN : SIMDMixedTwoVector<1, 0b10010, "sqxtun", int_arm64_neon_sqxtun>;
2205 defm SUQADD : SIMDTwoVectorBHSDTied<0, 0b00011, "suqadd",int_arm64_neon_suqadd>;
2206 defm UADALP : SIMDLongTwoVectorTied<1, 0b00110, "uadalp",
2207 BinOpFrag<(add node:$LHS, (int_arm64_neon_uaddlp node:$RHS))> >;
2208 defm UADDLP : SIMDLongTwoVector<1, 0b00010, "uaddlp",
2209 int_arm64_neon_uaddlp>;
2210 defm UCVTF : SIMDTwoVectorIntToFP<1, 0, 0b11101, "ucvtf", uint_to_fp>;
2211 defm UQXTN : SIMDMixedTwoVector<1, 0b10100, "uqxtn", int_arm64_neon_uqxtn>;
2212 defm URECPE : SIMDTwoVectorS<0, 1, 0b11100, "urecpe", int_arm64_neon_urecpe>;
2213 defm URSQRTE: SIMDTwoVectorS<1, 1, 0b11100, "ursqrte", int_arm64_neon_ursqrte>;
2214 defm USQADD : SIMDTwoVectorBHSDTied<1, 0b00011, "usqadd",int_arm64_neon_usqadd>;
2215 defm XTN : SIMDMixedTwoVector<0, 0b10010, "xtn", trunc>;
2217 def : Pat<(v2f32 (ARM64rev64 V64:$Rn)), (REV64v2i32 V64:$Rn)>;
2218 def : Pat<(v4f32 (ARM64rev64 V128:$Rn)), (REV64v4i32 V128:$Rn)>;
2220 // Patterns for vector long shift (by element width). These need to match all
2221 // three of zext, sext and anyext so it's easier to pull the patterns out of the
2223 multiclass SIMDVectorLShiftLongBySizeBHSPats<SDPatternOperator ext> {
2224 def : Pat<(ARM64vshl (v8i16 (ext (v8i8 V64:$Rn))), (i32 8)),
2225 (SHLLv8i8 V64:$Rn)>;
2226 def : Pat<(ARM64vshl (v8i16 (ext (extract_high_v16i8 V128:$Rn))), (i32 8)),
2227 (SHLLv16i8 V128:$Rn)>;
2228 def : Pat<(ARM64vshl (v4i32 (ext (v4i16 V64:$Rn))), (i32 16)),
2229 (SHLLv4i16 V64:$Rn)>;
2230 def : Pat<(ARM64vshl (v4i32 (ext (extract_high_v8i16 V128:$Rn))), (i32 16)),
2231 (SHLLv8i16 V128:$Rn)>;
2232 def : Pat<(ARM64vshl (v2i64 (ext (v2i32 V64:$Rn))), (i32 32)),
2233 (SHLLv2i32 V64:$Rn)>;
2234 def : Pat<(ARM64vshl (v2i64 (ext (extract_high_v4i32 V128:$Rn))), (i32 32)),
2235 (SHLLv4i32 V128:$Rn)>;
2238 defm : SIMDVectorLShiftLongBySizeBHSPats<anyext>;
2239 defm : SIMDVectorLShiftLongBySizeBHSPats<zext>;
2240 defm : SIMDVectorLShiftLongBySizeBHSPats<sext>;
2242 //===----------------------------------------------------------------------===//
2243 // Advanced SIMD three vector instructions.
2244 //===----------------------------------------------------------------------===//
2246 defm ADD : SIMDThreeSameVector<0, 0b10000, "add", add>;
2247 defm ADDP : SIMDThreeSameVector<0, 0b10111, "addp", int_arm64_neon_addp>;
2248 defm CMEQ : SIMDThreeSameVector<1, 0b10001, "cmeq", ARM64cmeq>;
2249 defm CMGE : SIMDThreeSameVector<0, 0b00111, "cmge", ARM64cmge>;
2250 defm CMGT : SIMDThreeSameVector<0, 0b00110, "cmgt", ARM64cmgt>;
2251 defm CMHI : SIMDThreeSameVector<1, 0b00110, "cmhi", ARM64cmhi>;
2252 defm CMHS : SIMDThreeSameVector<1, 0b00111, "cmhs", ARM64cmhs>;
2253 defm CMTST : SIMDThreeSameVector<0, 0b10001, "cmtst", ARM64cmtst>;
2254 defm FABD : SIMDThreeSameVectorFP<1,1,0b11010,"fabd", int_arm64_neon_fabd>;
2255 defm FACGE : SIMDThreeSameVectorFPCmp<1,0,0b11101,"facge",int_arm64_neon_facge>;
2256 defm FACGT : SIMDThreeSameVectorFPCmp<1,1,0b11101,"facgt",int_arm64_neon_facgt>;
2257 defm FADDP : SIMDThreeSameVectorFP<1,0,0b11010,"faddp",int_arm64_neon_addp>;
2258 defm FADD : SIMDThreeSameVectorFP<0,0,0b11010,"fadd", fadd>;
2259 defm FCMEQ : SIMDThreeSameVectorFPCmp<0, 0, 0b11100, "fcmeq", ARM64fcmeq>;
2260 defm FCMGE : SIMDThreeSameVectorFPCmp<1, 0, 0b11100, "fcmge", ARM64fcmge>;
2261 defm FCMGT : SIMDThreeSameVectorFPCmp<1, 1, 0b11100, "fcmgt", ARM64fcmgt>;
2262 defm FDIV : SIMDThreeSameVectorFP<1,0,0b11111,"fdiv", fdiv>;
2263 defm FMAXNMP : SIMDThreeSameVectorFP<1,0,0b11000,"fmaxnmp", int_arm64_neon_fmaxnmp>;
2264 defm FMAXNM : SIMDThreeSameVectorFP<0,0,0b11000,"fmaxnm", int_arm64_neon_fmaxnm>;
2265 defm FMAXP : SIMDThreeSameVectorFP<1,0,0b11110,"fmaxp", int_arm64_neon_fmaxp>;
2266 defm FMAX : SIMDThreeSameVectorFP<0,0,0b11110,"fmax", ARM64fmax>;
2267 defm FMINNMP : SIMDThreeSameVectorFP<1,1,0b11000,"fminnmp", int_arm64_neon_fminnmp>;
2268 defm FMINNM : SIMDThreeSameVectorFP<0,1,0b11000,"fminnm", int_arm64_neon_fminnm>;
2269 defm FMINP : SIMDThreeSameVectorFP<1,1,0b11110,"fminp", int_arm64_neon_fminp>;
2270 defm FMIN : SIMDThreeSameVectorFP<0,1,0b11110,"fmin", ARM64fmin>;
2272 // NOTE: The operands of the PatFrag are reordered on FMLA/FMLS because the
2273 // instruction expects the addend first, while the fma intrinsic puts it last.
2274 defm FMLA : SIMDThreeSameVectorFPTied<0, 0, 0b11001, "fmla",
2275 TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)> >;
2276 defm FMLS : SIMDThreeSameVectorFPTied<0, 1, 0b11001, "fmls",
2277 TriOpFrag<(fma node:$MHS, (fneg node:$RHS), node:$LHS)> >;
2279 // The following def pats catch the case where the LHS of an FMA is negated.
2280 // The TriOpFrag above catches the case where the middle operand is negated.
2281 def : Pat<(v2f32 (fma (fneg V64:$Rn), V64:$Rm, V64:$Rd)),
2282 (FMLSv2f32 V64:$Rd, V64:$Rn, V64:$Rm)>;
2284 def : Pat<(v4f32 (fma (fneg V128:$Rn), V128:$Rm, V128:$Rd)),
2285 (FMLSv4f32 V128:$Rd, V128:$Rn, V128:$Rm)>;
2287 def : Pat<(v2f64 (fma (fneg V128:$Rn), V128:$Rm, V128:$Rd)),
2288 (FMLSv2f64 V128:$Rd, V128:$Rn, V128:$Rm)>;
2290 defm FMULX : SIMDThreeSameVectorFP<0,0,0b11011,"fmulx", int_arm64_neon_fmulx>;
2291 defm FMUL : SIMDThreeSameVectorFP<1,0,0b11011,"fmul", fmul>;
2292 defm FRECPS : SIMDThreeSameVectorFP<0,0,0b11111,"frecps", int_arm64_neon_frecps>;
2293 defm FRSQRTS : SIMDThreeSameVectorFP<0,1,0b11111,"frsqrts", int_arm64_neon_frsqrts>;
2294 defm FSUB : SIMDThreeSameVectorFP<0,1,0b11010,"fsub", fsub>;
2295 defm MLA : SIMDThreeSameVectorBHSTied<0, 0b10010, "mla",
2296 TriOpFrag<(add node:$LHS, (mul node:$MHS, node:$RHS))> >;
2297 defm MLS : SIMDThreeSameVectorBHSTied<1, 0b10010, "mls",
2298 TriOpFrag<(sub node:$LHS, (mul node:$MHS, node:$RHS))> >;
2299 defm MUL : SIMDThreeSameVectorBHS<0, 0b10011, "mul", mul>;
2300 defm PMUL : SIMDThreeSameVectorB<1, 0b10011, "pmul", int_arm64_neon_pmul>;
2301 defm SABA : SIMDThreeSameVectorBHSTied<0, 0b01111, "saba",
2302 TriOpFrag<(add node:$LHS, (int_arm64_neon_sabd node:$MHS, node:$RHS))> >;
2303 defm SABD : SIMDThreeSameVectorBHS<0,0b01110,"sabd", int_arm64_neon_sabd>;
2304 defm SHADD : SIMDThreeSameVectorBHS<0,0b00000,"shadd", int_arm64_neon_shadd>;
2305 defm SHSUB : SIMDThreeSameVectorBHS<0,0b00100,"shsub", int_arm64_neon_shsub>;
2306 defm SMAXP : SIMDThreeSameVectorBHS<0,0b10100,"smaxp", int_arm64_neon_smaxp>;
2307 defm SMAX : SIMDThreeSameVectorBHS<0,0b01100,"smax", int_arm64_neon_smax>;
2308 defm SMINP : SIMDThreeSameVectorBHS<0,0b10101,"sminp", int_arm64_neon_sminp>;
2309 defm SMIN : SIMDThreeSameVectorBHS<0,0b01101,"smin", int_arm64_neon_smin>;
2310 defm SQADD : SIMDThreeSameVector<0,0b00001,"sqadd", int_arm64_neon_sqadd>;
2311 defm SQDMULH : SIMDThreeSameVectorHS<0,0b10110,"sqdmulh",int_arm64_neon_sqdmulh>;
2312 defm SQRDMULH : SIMDThreeSameVectorHS<1,0b10110,"sqrdmulh",int_arm64_neon_sqrdmulh>;
2313 defm SQRSHL : SIMDThreeSameVector<0,0b01011,"sqrshl", int_arm64_neon_sqrshl>;
2314 defm SQSHL : SIMDThreeSameVector<0,0b01001,"sqshl", int_arm64_neon_sqshl>;
2315 defm SQSUB : SIMDThreeSameVector<0,0b00101,"sqsub", int_arm64_neon_sqsub>;
2316 defm SRHADD : SIMDThreeSameVectorBHS<0,0b00010,"srhadd",int_arm64_neon_srhadd>;
2317 defm SRSHL : SIMDThreeSameVector<0,0b01010,"srshl", int_arm64_neon_srshl>;
2318 defm SSHL : SIMDThreeSameVector<0,0b01000,"sshl", int_arm64_neon_sshl>;
2319 defm SUB : SIMDThreeSameVector<1,0b10000,"sub", sub>;
2320 defm UABA : SIMDThreeSameVectorBHSTied<1, 0b01111, "uaba",
2321 TriOpFrag<(add node:$LHS, (int_arm64_neon_uabd node:$MHS, node:$RHS))> >;
2322 defm UABD : SIMDThreeSameVectorBHS<1,0b01110,"uabd", int_arm64_neon_uabd>;
2323 defm UHADD : SIMDThreeSameVectorBHS<1,0b00000,"uhadd", int_arm64_neon_uhadd>;
2324 defm UHSUB : SIMDThreeSameVectorBHS<1,0b00100,"uhsub", int_arm64_neon_uhsub>;
2325 defm UMAXP : SIMDThreeSameVectorBHS<1,0b10100,"umaxp", int_arm64_neon_umaxp>;
2326 defm UMAX : SIMDThreeSameVectorBHS<1,0b01100,"umax", int_arm64_neon_umax>;
2327 defm UMINP : SIMDThreeSameVectorBHS<1,0b10101,"uminp", int_arm64_neon_uminp>;
2328 defm UMIN : SIMDThreeSameVectorBHS<1,0b01101,"umin", int_arm64_neon_umin>;
2329 defm UQADD : SIMDThreeSameVector<1,0b00001,"uqadd", int_arm64_neon_uqadd>;
2330 defm UQRSHL : SIMDThreeSameVector<1,0b01011,"uqrshl", int_arm64_neon_uqrshl>;
2331 defm UQSHL : SIMDThreeSameVector<1,0b01001,"uqshl", int_arm64_neon_uqshl>;
2332 defm UQSUB : SIMDThreeSameVector<1,0b00101,"uqsub", int_arm64_neon_uqsub>;
2333 defm URHADD : SIMDThreeSameVectorBHS<1,0b00010,"urhadd", int_arm64_neon_urhadd>;
2334 defm URSHL : SIMDThreeSameVector<1,0b01010,"urshl", int_arm64_neon_urshl>;
2335 defm USHL : SIMDThreeSameVector<1,0b01000,"ushl", int_arm64_neon_ushl>;
2337 defm AND : SIMDLogicalThreeVector<0, 0b00, "and", and>;
2338 defm BIC : SIMDLogicalThreeVector<0, 0b01, "bic",
2339 BinOpFrag<(and node:$LHS, (vnot node:$RHS))> >;
2340 defm BIF : SIMDLogicalThreeVector<1, 0b11, "bif">;
2341 defm BIT : SIMDLogicalThreeVectorTied<1, 0b10, "bit", ARM64bit>;
2342 defm BSL : SIMDLogicalThreeVectorTied<1, 0b01, "bsl",
2343 TriOpFrag<(or (and node:$LHS, node:$MHS), (and (vnot node:$LHS), node:$RHS))>>;
2344 defm EOR : SIMDLogicalThreeVector<1, 0b00, "eor", xor>;
2345 defm ORN : SIMDLogicalThreeVector<0, 0b11, "orn",
2346 BinOpFrag<(or node:$LHS, (vnot node:$RHS))> >;
2347 defm ORR : SIMDLogicalThreeVector<0, 0b10, "orr", or>;
2349 // FIXME: the .16b and .8b variantes should be emitted by the
2350 // AsmWriter. TableGen's AsmWriter-generator doesn't deal with variant syntaxes
2351 // in aliases yet though.
2352 def : InstAlias<"mov{\t$dst.16b, $src.16b|.16b\t$dst, $src}",
2353 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2354 def : InstAlias<"{mov\t$dst.8h, $src.8h|mov.8h\t$dst, $src}",
2355 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2356 def : InstAlias<"{mov\t$dst.4s, $src.4s|mov.4s\t$dst, $src}",
2357 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2358 def : InstAlias<"{mov\t$dst.2d, $src.2d|mov.2d\t$dst, $src}",
2359 (ORRv16i8 V128:$dst, V128:$src, V128:$src), 0>;
2361 def : InstAlias<"{mov\t$dst.8b, $src.8b|mov.8b\t$dst, $src}",
2362 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2363 def : InstAlias<"{mov\t$dst.4h, $src.4h|mov.4h\t$dst, $src}",
2364 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2365 def : InstAlias<"{mov\t$dst.2s, $src.2s|mov.2s\t$dst, $src}",
2366 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2367 def : InstAlias<"{mov\t$dst.1d, $src.1d|mov.1d\t$dst, $src}",
2368 (ORRv8i8 V64:$dst, V64:$src, V64:$src), 0>;
2370 def : InstAlias<"{cmls\t$dst.8b, $src1.8b, $src2.8b" #
2371 "|cmls.8b\t$dst, $src1, $src2}",
2372 (CMHSv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2373 def : InstAlias<"{cmls\t$dst.16b, $src1.16b, $src2.16b" #
2374 "|cmls.16b\t$dst, $src1, $src2}",
2375 (CMHSv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2376 def : InstAlias<"{cmls\t$dst.4h, $src1.4h, $src2.4h" #
2377 "|cmls.4h\t$dst, $src1, $src2}",
2378 (CMHSv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2379 def : InstAlias<"{cmls\t$dst.8h, $src1.8h, $src2.8h" #
2380 "|cmls.8h\t$dst, $src1, $src2}",
2381 (CMHSv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2382 def : InstAlias<"{cmls\t$dst.2s, $src1.2s, $src2.2s" #
2383 "|cmls.2s\t$dst, $src1, $src2}",
2384 (CMHSv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2385 def : InstAlias<"{cmls\t$dst.4s, $src1.4s, $src2.4s" #
2386 "|cmls.4s\t$dst, $src1, $src2}",
2387 (CMHSv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2388 def : InstAlias<"{cmls\t$dst.2d, $src1.2d, $src2.2d" #
2389 "|cmls.2d\t$dst, $src1, $src2}",
2390 (CMHSv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2392 def : InstAlias<"{cmlo\t$dst.8b, $src1.8b, $src2.8b" #
2393 "|cmlo.8b\t$dst, $src1, $src2}",
2394 (CMHIv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2395 def : InstAlias<"{cmlo\t$dst.16b, $src1.16b, $src2.16b" #
2396 "|cmlo.16b\t$dst, $src1, $src2}",
2397 (CMHIv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2398 def : InstAlias<"{cmlo\t$dst.4h, $src1.4h, $src2.4h" #
2399 "|cmlo.4h\t$dst, $src1, $src2}",
2400 (CMHIv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2401 def : InstAlias<"{cmlo\t$dst.8h, $src1.8h, $src2.8h" #
2402 "|cmlo.8h\t$dst, $src1, $src2}",
2403 (CMHIv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2404 def : InstAlias<"{cmlo\t$dst.2s, $src1.2s, $src2.2s" #
2405 "|cmlo.2s\t$dst, $src1, $src2}",
2406 (CMHIv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2407 def : InstAlias<"{cmlo\t$dst.4s, $src1.4s, $src2.4s" #
2408 "|cmlo.4s\t$dst, $src1, $src2}",
2409 (CMHIv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2410 def : InstAlias<"{cmlo\t$dst.2d, $src1.2d, $src2.2d" #
2411 "|cmlo.2d\t$dst, $src1, $src2}",
2412 (CMHIv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2414 def : InstAlias<"{cmle\t$dst.8b, $src1.8b, $src2.8b" #
2415 "|cmle.8b\t$dst, $src1, $src2}",
2416 (CMGEv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2417 def : InstAlias<"{cmle\t$dst.16b, $src1.16b, $src2.16b" #
2418 "|cmle.16b\t$dst, $src1, $src2}",
2419 (CMGEv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2420 def : InstAlias<"{cmle\t$dst.4h, $src1.4h, $src2.4h" #
2421 "|cmle.4h\t$dst, $src1, $src2}",
2422 (CMGEv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2423 def : InstAlias<"{cmle\t$dst.8h, $src1.8h, $src2.8h" #
2424 "|cmle.8h\t$dst, $src1, $src2}",
2425 (CMGEv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2426 def : InstAlias<"{cmle\t$dst.2s, $src1.2s, $src2.2s" #
2427 "|cmle.2s\t$dst, $src1, $src2}",
2428 (CMGEv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2429 def : InstAlias<"{cmle\t$dst.4s, $src1.4s, $src2.4s" #
2430 "|cmle.4s\t$dst, $src1, $src2}",
2431 (CMGEv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2432 def : InstAlias<"{cmle\t$dst.2d, $src1.2d, $src2.2d" #
2433 "|cmle.2d\t$dst, $src1, $src2}",
2434 (CMGEv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2436 def : InstAlias<"{cmlt\t$dst.8b, $src1.8b, $src2.8b" #
2437 "|cmlt.8b\t$dst, $src1, $src2}",
2438 (CMGTv8i8 V64:$dst, V64:$src2, V64:$src1), 0>;
2439 def : InstAlias<"{cmlt\t$dst.16b, $src1.16b, $src2.16b" #
2440 "|cmlt.16b\t$dst, $src1, $src2}",
2441 (CMGTv16i8 V128:$dst, V128:$src2, V128:$src1), 0>;
2442 def : InstAlias<"{cmlt\t$dst.4h, $src1.4h, $src2.4h" #
2443 "|cmlt.4h\t$dst, $src1, $src2}",
2444 (CMGTv4i16 V64:$dst, V64:$src2, V64:$src1), 0>;
2445 def : InstAlias<"{cmlt\t$dst.8h, $src1.8h, $src2.8h" #
2446 "|cmlt.8h\t$dst, $src1, $src2}",
2447 (CMGTv8i16 V128:$dst, V128:$src2, V128:$src1), 0>;
2448 def : InstAlias<"{cmlt\t$dst.2s, $src1.2s, $src2.2s" #
2449 "|cmlt.2s\t$dst, $src1, $src2}",
2450 (CMGTv2i32 V64:$dst, V64:$src2, V64:$src1), 0>;
2451 def : InstAlias<"{cmlt\t$dst.4s, $src1.4s, $src2.4s" #
2452 "|cmlt.4s\t$dst, $src1, $src2}",
2453 (CMGTv4i32 V128:$dst, V128:$src2, V128:$src1), 0>;
2454 def : InstAlias<"{cmlt\t$dst.2d, $src1.2d, $src2.2d" #
2455 "|cmlt.2d\t$dst, $src1, $src2}",
2456 (CMGTv2i64 V128:$dst, V128:$src2, V128:$src1), 0>;
2458 def : InstAlias<"{fcmle\t$dst.2s, $src1.2s, $src2.2s" #
2459 "|fcmle.2s\t$dst, $src1, $src2}",
2460 (FCMGEv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
2461 def : InstAlias<"{fcmle\t$dst.4s, $src1.4s, $src2.4s" #
2462 "|fcmle.4s\t$dst, $src1, $src2}",
2463 (FCMGEv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
2464 def : InstAlias<"{fcmle\t$dst.2d, $src1.2d, $src2.2d" #
2465 "|fcmle.2d\t$dst, $src1, $src2}",
2466 (FCMGEv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
2468 def : InstAlias<"{fcmlt\t$dst.2s, $src1.2s, $src2.2s" #
2469 "|fcmlt.2s\t$dst, $src1, $src2}",
2470 (FCMGTv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
2471 def : InstAlias<"{fcmlt\t$dst.4s, $src1.4s, $src2.4s" #
2472 "|fcmlt.4s\t$dst, $src1, $src2}",
2473 (FCMGTv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
2474 def : InstAlias<"{fcmlt\t$dst.2d, $src1.2d, $src2.2d" #
2475 "|fcmlt.2d\t$dst, $src1, $src2}",
2476 (FCMGTv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
2478 def : InstAlias<"{facle\t$dst.2s, $src1.2s, $src2.2s" #
2479 "|facle.2s\t$dst, $src1, $src2}",
2480 (FACGEv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
2481 def : InstAlias<"{facle\t$dst.4s, $src1.4s, $src2.4s" #
2482 "|facle.4s\t$dst, $src1, $src2}",
2483 (FACGEv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
2484 def : InstAlias<"{facle\t$dst.2d, $src1.2d, $src2.2d" #
2485 "|facle.2d\t$dst, $src1, $src2}",
2486 (FACGEv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
2488 def : InstAlias<"{faclt\t$dst.2s, $src1.2s, $src2.2s" #
2489 "|faclt.2s\t$dst, $src1, $src2}",
2490 (FACGTv2f32 V64:$dst, V64:$src2, V64:$src1), 0>;
2491 def : InstAlias<"{faclt\t$dst.4s, $src1.4s, $src2.4s" #
2492 "|faclt.4s\t$dst, $src1, $src2}",
2493 (FACGTv4f32 V128:$dst, V128:$src2, V128:$src1), 0>;
2494 def : InstAlias<"{faclt\t$dst.2d, $src1.2d, $src2.2d" #
2495 "|faclt.2d\t$dst, $src1, $src2}",
2496 (FACGTv2f64 V128:$dst, V128:$src2, V128:$src1), 0>;
2498 //===----------------------------------------------------------------------===//
2499 // Advanced SIMD three scalar instructions.
2500 //===----------------------------------------------------------------------===//
2502 defm ADD : SIMDThreeScalarD<0, 0b10000, "add", add>;
2503 defm CMEQ : SIMDThreeScalarD<1, 0b10001, "cmeq", ARM64cmeq>;
2504 defm CMGE : SIMDThreeScalarD<0, 0b00111, "cmge", ARM64cmge>;
2505 defm CMGT : SIMDThreeScalarD<0, 0b00110, "cmgt", ARM64cmgt>;
2506 defm CMHI : SIMDThreeScalarD<1, 0b00110, "cmhi", ARM64cmhi>;
2507 defm CMHS : SIMDThreeScalarD<1, 0b00111, "cmhs", ARM64cmhs>;
2508 defm CMTST : SIMDThreeScalarD<0, 0b10001, "cmtst", ARM64cmtst>;
2509 defm FABD : SIMDThreeScalarSD<1, 1, 0b11010, "fabd", int_arm64_sisd_fabd>;
2510 def : Pat<(v1f64 (int_arm64_neon_fabd (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
2511 (FABD64 FPR64:$Rn, FPR64:$Rm)>;
2512 defm FACGE : SIMDThreeScalarFPCmp<1, 0, 0b11101, "facge",
2513 int_arm64_neon_facge>;
2514 defm FACGT : SIMDThreeScalarFPCmp<1, 1, 0b11101, "facgt",
2515 int_arm64_neon_facgt>;
2516 defm FCMEQ : SIMDThreeScalarFPCmp<0, 0, 0b11100, "fcmeq", ARM64fcmeq>;
2517 defm FCMGE : SIMDThreeScalarFPCmp<1, 0, 0b11100, "fcmge", ARM64fcmge>;
2518 defm FCMGT : SIMDThreeScalarFPCmp<1, 1, 0b11100, "fcmgt", ARM64fcmgt>;
2519 defm FMULX : SIMDThreeScalarSD<0, 0, 0b11011, "fmulx", int_arm64_neon_fmulx>;
2520 defm FRECPS : SIMDThreeScalarSD<0, 0, 0b11111, "frecps", int_arm64_neon_frecps>;
2521 defm FRSQRTS : SIMDThreeScalarSD<0, 1, 0b11111, "frsqrts", int_arm64_neon_frsqrts>;
2522 defm SQADD : SIMDThreeScalarBHSD<0, 0b00001, "sqadd", int_arm64_neon_sqadd>;
2523 defm SQDMULH : SIMDThreeScalarHS< 0, 0b10110, "sqdmulh", int_arm64_neon_sqdmulh>;
2524 defm SQRDMULH : SIMDThreeScalarHS< 1, 0b10110, "sqrdmulh", int_arm64_neon_sqrdmulh>;
2525 defm SQRSHL : SIMDThreeScalarBHSD<0, 0b01011, "sqrshl",int_arm64_neon_sqrshl>;
2526 defm SQSHL : SIMDThreeScalarBHSD<0, 0b01001, "sqshl", int_arm64_neon_sqshl>;
2527 defm SQSUB : SIMDThreeScalarBHSD<0, 0b00101, "sqsub", int_arm64_neon_sqsub>;
2528 defm SRSHL : SIMDThreeScalarD< 0, 0b01010, "srshl", int_arm64_neon_srshl>;
2529 defm SSHL : SIMDThreeScalarD< 0, 0b01000, "sshl", int_arm64_neon_sshl>;
2530 defm SUB : SIMDThreeScalarD< 1, 0b10000, "sub", sub>;
2531 defm UQADD : SIMDThreeScalarBHSD<1, 0b00001, "uqadd", int_arm64_neon_uqadd>;
2532 defm UQRSHL : SIMDThreeScalarBHSD<1, 0b01011, "uqrshl",int_arm64_neon_uqrshl>;
2533 defm UQSHL : SIMDThreeScalarBHSD<1, 0b01001, "uqshl", int_arm64_neon_uqshl>;
2534 defm UQSUB : SIMDThreeScalarBHSD<1, 0b00101, "uqsub", int_arm64_neon_uqsub>;
2535 defm URSHL : SIMDThreeScalarD< 1, 0b01010, "urshl", int_arm64_neon_urshl>;
2536 defm USHL : SIMDThreeScalarD< 1, 0b01000, "ushl", int_arm64_neon_ushl>;
2538 def : InstAlias<"cmls $dst, $src1, $src2",
2539 (CMHSv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2540 def : InstAlias<"cmle $dst, $src1, $src2",
2541 (CMGEv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2542 def : InstAlias<"cmlo $dst, $src1, $src2",
2543 (CMHIv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2544 def : InstAlias<"cmlt $dst, $src1, $src2",
2545 (CMGTv1i64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2546 def : InstAlias<"fcmle $dst, $src1, $src2",
2547 (FCMGE32 FPR32:$dst, FPR32:$src2, FPR32:$src1)>;
2548 def : InstAlias<"fcmle $dst, $src1, $src2",
2549 (FCMGE64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2550 def : InstAlias<"fcmlt $dst, $src1, $src2",
2551 (FCMGT32 FPR32:$dst, FPR32:$src2, FPR32:$src1)>;
2552 def : InstAlias<"fcmlt $dst, $src1, $src2",
2553 (FCMGT64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2554 def : InstAlias<"facle $dst, $src1, $src2",
2555 (FACGE32 FPR32:$dst, FPR32:$src2, FPR32:$src1)>;
2556 def : InstAlias<"facle $dst, $src1, $src2",
2557 (FACGE64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2558 def : InstAlias<"faclt $dst, $src1, $src2",
2559 (FACGT32 FPR32:$dst, FPR32:$src2, FPR32:$src1)>;
2560 def : InstAlias<"faclt $dst, $src1, $src2",
2561 (FACGT64 FPR64:$dst, FPR64:$src2, FPR64:$src1)>;
2563 //===----------------------------------------------------------------------===//
2564 // Advanced SIMD three scalar instructions (mixed operands).
2565 //===----------------------------------------------------------------------===//
2566 defm SQDMULL : SIMDThreeScalarMixedHS<0, 0b11010, "sqdmull",
2567 int_arm64_neon_sqdmulls_scalar>;
2568 defm SQDMLAL : SIMDThreeScalarMixedTiedHS<0, 0b10010, "sqdmlal">;
2569 defm SQDMLSL : SIMDThreeScalarMixedTiedHS<0, 0b10110, "sqdmlsl">;
2571 def : Pat<(i64 (int_arm64_neon_sqadd (i64 FPR64:$Rd),
2572 (i64 (int_arm64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
2573 (i32 FPR32:$Rm))))),
2574 (SQDMLALi32 FPR64:$Rd, FPR32:$Rn, FPR32:$Rm)>;
2575 def : Pat<(i64 (int_arm64_neon_sqsub (i64 FPR64:$Rd),
2576 (i64 (int_arm64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
2577 (i32 FPR32:$Rm))))),
2578 (SQDMLSLi32 FPR64:$Rd, FPR32:$Rn, FPR32:$Rm)>;
2580 //===----------------------------------------------------------------------===//
2581 // Advanced SIMD two scalar instructions.
2582 //===----------------------------------------------------------------------===//
2584 defm ABS : SIMDTwoScalarD< 0, 0b01011, "abs", int_arm64_neon_abs>;
2585 defm CMEQ : SIMDCmpTwoScalarD< 0, 0b01001, "cmeq", ARM64cmeqz>;
2586 defm CMGE : SIMDCmpTwoScalarD< 1, 0b01000, "cmge", ARM64cmgez>;
2587 defm CMGT : SIMDCmpTwoScalarD< 0, 0b01000, "cmgt", ARM64cmgtz>;
2588 defm CMLE : SIMDCmpTwoScalarD< 1, 0b01001, "cmle", ARM64cmlez>;
2589 defm CMLT : SIMDCmpTwoScalarD< 0, 0b01010, "cmlt", ARM64cmltz>;
2590 defm FCMEQ : SIMDCmpTwoScalarSD<0, 1, 0b01101, "fcmeq", ARM64fcmeqz>;
2591 defm FCMGE : SIMDCmpTwoScalarSD<1, 1, 0b01100, "fcmge", ARM64fcmgez>;
2592 defm FCMGT : SIMDCmpTwoScalarSD<0, 1, 0b01100, "fcmgt", ARM64fcmgtz>;
2593 defm FCMLE : SIMDCmpTwoScalarSD<1, 1, 0b01101, "fcmle", ARM64fcmlez>;
2594 defm FCMLT : SIMDCmpTwoScalarSD<0, 1, 0b01110, "fcmlt", ARM64fcmltz>;
2595 defm FCVTAS : SIMDTwoScalarSD< 0, 0, 0b11100, "fcvtas">;
2596 defm FCVTAU : SIMDTwoScalarSD< 1, 0, 0b11100, "fcvtau">;
2597 defm FCVTMS : SIMDTwoScalarSD< 0, 0, 0b11011, "fcvtms">;
2598 defm FCVTMU : SIMDTwoScalarSD< 1, 0, 0b11011, "fcvtmu">;
2599 defm FCVTNS : SIMDTwoScalarSD< 0, 0, 0b11010, "fcvtns">;
2600 defm FCVTNU : SIMDTwoScalarSD< 1, 0, 0b11010, "fcvtnu">;
2601 defm FCVTPS : SIMDTwoScalarSD< 0, 1, 0b11010, "fcvtps">;
2602 defm FCVTPU : SIMDTwoScalarSD< 1, 1, 0b11010, "fcvtpu">;
2603 def FCVTXNv1i64 : SIMDInexactCvtTwoScalar<0b10110, "fcvtxn">;
2604 defm FCVTZS : SIMDTwoScalarSD< 0, 1, 0b11011, "fcvtzs">;
2605 defm FCVTZU : SIMDTwoScalarSD< 1, 1, 0b11011, "fcvtzu">;
2606 defm FRECPE : SIMDTwoScalarSD< 0, 1, 0b11101, "frecpe">;
2607 defm FRECPX : SIMDTwoScalarSD< 0, 1, 0b11111, "frecpx">;
2608 defm FRSQRTE : SIMDTwoScalarSD< 1, 1, 0b11101, "frsqrte">;
2609 defm NEG : SIMDTwoScalarD< 1, 0b01011, "neg",
2610 UnOpFrag<(sub immAllZerosV, node:$LHS)> >;
2611 defm SCVTF : SIMDTwoScalarCVTSD< 0, 0, 0b11101, "scvtf", ARM64sitof>;
2612 defm SQABS : SIMDTwoScalarBHSD< 0, 0b00111, "sqabs", int_arm64_neon_sqabs>;
2613 defm SQNEG : SIMDTwoScalarBHSD< 1, 0b00111, "sqneg", int_arm64_neon_sqneg>;
2614 defm SQXTN : SIMDTwoScalarMixedBHS< 0, 0b10100, "sqxtn", int_arm64_neon_scalar_sqxtn>;
2615 defm SQXTUN : SIMDTwoScalarMixedBHS< 1, 0b10010, "sqxtun", int_arm64_neon_scalar_sqxtun>;
2616 defm SUQADD : SIMDTwoScalarBHSDTied< 0, 0b00011, "suqadd",
2617 int_arm64_neon_suqadd>;
2618 defm UCVTF : SIMDTwoScalarCVTSD< 1, 0, 0b11101, "ucvtf", ARM64uitof>;
2619 defm UQXTN : SIMDTwoScalarMixedBHS<1, 0b10100, "uqxtn", int_arm64_neon_scalar_uqxtn>;
2620 defm USQADD : SIMDTwoScalarBHSDTied< 1, 0b00011, "usqadd",
2621 int_arm64_neon_usqadd>;
2623 def : Pat<(v1i64 (int_arm64_neon_fcvtas (v1f64 FPR64:$Rn))),
2624 (FCVTASv1i64 FPR64:$Rn)>;
2625 def : Pat<(v1i64 (int_arm64_neon_fcvtau (v1f64 FPR64:$Rn))),
2626 (FCVTAUv1i64 FPR64:$Rn)>;
2627 def : Pat<(v1i64 (int_arm64_neon_fcvtms (v1f64 FPR64:$Rn))),
2628 (FCVTMSv1i64 FPR64:$Rn)>;
2629 def : Pat<(v1i64 (int_arm64_neon_fcvtmu (v1f64 FPR64:$Rn))),
2630 (FCVTMUv1i64 FPR64:$Rn)>;
2631 def : Pat<(v1i64 (int_arm64_neon_fcvtns (v1f64 FPR64:$Rn))),
2632 (FCVTNSv1i64 FPR64:$Rn)>;
2633 def : Pat<(v1i64 (int_arm64_neon_fcvtnu (v1f64 FPR64:$Rn))),
2634 (FCVTNUv1i64 FPR64:$Rn)>;
2635 def : Pat<(v1i64 (int_arm64_neon_fcvtps (v1f64 FPR64:$Rn))),
2636 (FCVTPSv1i64 FPR64:$Rn)>;
2637 def : Pat<(v1i64 (int_arm64_neon_fcvtpu (v1f64 FPR64:$Rn))),
2638 (FCVTPUv1i64 FPR64:$Rn)>;
2640 def : Pat<(f32 (int_arm64_neon_frecpe (f32 FPR32:$Rn))),
2641 (FRECPEv1i32 FPR32:$Rn)>;
2642 def : Pat<(f64 (int_arm64_neon_frecpe (f64 FPR64:$Rn))),
2643 (FRECPEv1i64 FPR64:$Rn)>;
2644 def : Pat<(v1f64 (int_arm64_neon_frecpe (v1f64 FPR64:$Rn))),
2645 (FRECPEv1i64 FPR64:$Rn)>;
2647 def : Pat<(f32 (int_arm64_neon_frecpx (f32 FPR32:$Rn))),
2648 (FRECPXv1i32 FPR32:$Rn)>;
2649 def : Pat<(f64 (int_arm64_neon_frecpx (f64 FPR64:$Rn))),
2650 (FRECPXv1i64 FPR64:$Rn)>;
2652 def : Pat<(f32 (int_arm64_neon_frsqrte (f32 FPR32:$Rn))),
2653 (FRSQRTEv1i32 FPR32:$Rn)>;
2654 def : Pat<(f64 (int_arm64_neon_frsqrte (f64 FPR64:$Rn))),
2655 (FRSQRTEv1i64 FPR64:$Rn)>;
2656 def : Pat<(v1f64 (int_arm64_neon_frsqrte (v1f64 FPR64:$Rn))),
2657 (FRSQRTEv1i64 FPR64:$Rn)>;
2659 // If an integer is about to be converted to a floating point value,
2660 // just load it on the floating point unit.
2661 // Here are the patterns for 8 and 16-bits to float.
2663 def : Pat <(f32 (uint_to_fp (i32 (zextloadi8 ro_indexed8:$addr)))),
2664 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2665 (LDRBro ro_indexed8:$addr), bsub))>;
2666 def : Pat <(f32 (uint_to_fp (i32 (zextloadi8 am_indexed8:$addr)))),
2667 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2668 (LDRBui am_indexed8:$addr), bsub))>;
2669 def : Pat <(f32 (uint_to_fp (i32 (zextloadi8 am_unscaled8:$addr)))),
2670 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2671 (LDURBi am_unscaled8:$addr), bsub))>;
2672 // 16-bits -> float.
2673 def : Pat <(f32 (uint_to_fp (i32 (zextloadi16 ro_indexed16:$addr)))),
2674 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2675 (LDRHro ro_indexed16:$addr), hsub))>;
2676 def : Pat <(f32 (uint_to_fp (i32 (zextloadi16 am_indexed16:$addr)))),
2677 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2678 (LDRHui am_indexed16:$addr), hsub))>;
2679 def : Pat <(f32 (uint_to_fp (i32 (zextloadi16 am_unscaled16:$addr)))),
2680 (UCVTFv1i32 (INSERT_SUBREG (f32 (IMPLICIT_DEF)),
2681 (LDURHi am_unscaled16:$addr), hsub))>;
2682 // 32-bits are handled in target specific dag combine:
2683 // performIntToFpCombine.
2684 // 64-bits integer to 32-bits floating point, not possible with
2685 // UCVTF on floating point registers (both source and destination
2686 // must have the same size).
2688 // Here are the patterns for 8, 16, 32, and 64-bits to double.
2689 // 8-bits -> double.
2690 def : Pat <(f64 (uint_to_fp (i32 (zextloadi8 ro_indexed8:$addr)))),
2691 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2692 (LDRBro ro_indexed8:$addr), bsub))>;
2693 def : Pat <(f64 (uint_to_fp (i32 (zextloadi8 am_indexed8:$addr)))),
2694 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2695 (LDRBui am_indexed8:$addr), bsub))>;
2696 def : Pat <(f64 (uint_to_fp (i32 (zextloadi8 am_unscaled8:$addr)))),
2697 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2698 (LDURBi am_unscaled8:$addr), bsub))>;
2699 // 16-bits -> double.
2700 def : Pat <(f64 (uint_to_fp (i32 (zextloadi16 ro_indexed16:$addr)))),
2701 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2702 (LDRHro ro_indexed16:$addr), hsub))>;
2703 def : Pat <(f64 (uint_to_fp (i32 (zextloadi16 am_indexed16:$addr)))),
2704 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2705 (LDRHui am_indexed16:$addr), hsub))>;
2706 def : Pat <(f64 (uint_to_fp (i32 (zextloadi16 am_unscaled16:$addr)))),
2707 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2708 (LDURHi am_unscaled16:$addr), hsub))>;
2709 // 32-bits -> double.
2710 def : Pat <(f64 (uint_to_fp (i32 (load ro_indexed32:$addr)))),
2711 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2712 (LDRSro ro_indexed32:$addr), ssub))>;
2713 def : Pat <(f64 (uint_to_fp (i32 (load am_indexed32:$addr)))),
2714 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2715 (LDRSui am_indexed32:$addr), ssub))>;
2716 def : Pat <(f64 (uint_to_fp (i32 (load am_unscaled32:$addr)))),
2717 (UCVTFv1i64 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
2718 (LDURSi am_unscaled32:$addr), ssub))>;
2719 // 64-bits -> double are handled in target specific dag combine:
2720 // performIntToFpCombine.
2722 //===----------------------------------------------------------------------===//
2723 // Advanced SIMD three different-sized vector instructions.
2724 //===----------------------------------------------------------------------===//
2726 defm ADDHN : SIMDNarrowThreeVectorBHS<0,0b0100,"addhn", int_arm64_neon_addhn>;
2727 defm SUBHN : SIMDNarrowThreeVectorBHS<0,0b0110,"subhn", int_arm64_neon_subhn>;
2728 defm RADDHN : SIMDNarrowThreeVectorBHS<1,0b0100,"raddhn",int_arm64_neon_raddhn>;
2729 defm RSUBHN : SIMDNarrowThreeVectorBHS<1,0b0110,"rsubhn",int_arm64_neon_rsubhn>;
2730 defm PMULL : SIMDDifferentThreeVectorBD<0,0b1110,"pmull",int_arm64_neon_pmull>;
2731 defm SABAL : SIMDLongThreeVectorTiedBHSabal<0,0b0101,"sabal",
2732 int_arm64_neon_sabd>;
2733 defm SABDL : SIMDLongThreeVectorBHSabdl<0, 0b0111, "sabdl",
2734 int_arm64_neon_sabd>;
2735 defm SADDL : SIMDLongThreeVectorBHS< 0, 0b0000, "saddl",
2736 BinOpFrag<(add (sext node:$LHS), (sext node:$RHS))>>;
2737 defm SADDW : SIMDWideThreeVectorBHS< 0, 0b0001, "saddw",
2738 BinOpFrag<(add node:$LHS, (sext node:$RHS))>>;
2739 defm SMLAL : SIMDLongThreeVectorTiedBHS<0, 0b1000, "smlal",
2740 TriOpFrag<(add node:$LHS, (int_arm64_neon_smull node:$MHS, node:$RHS))>>;
2741 defm SMLSL : SIMDLongThreeVectorTiedBHS<0, 0b1010, "smlsl",
2742 TriOpFrag<(sub node:$LHS, (int_arm64_neon_smull node:$MHS, node:$RHS))>>;
2743 defm SMULL : SIMDLongThreeVectorBHS<0, 0b1100, "smull", int_arm64_neon_smull>;
2744 defm SQDMLAL : SIMDLongThreeVectorSQDMLXTiedHS<0, 0b1001, "sqdmlal",
2745 int_arm64_neon_sqadd>;
2746 defm SQDMLSL : SIMDLongThreeVectorSQDMLXTiedHS<0, 0b1011, "sqdmlsl",
2747 int_arm64_neon_sqsub>;
2748 defm SQDMULL : SIMDLongThreeVectorHS<0, 0b1101, "sqdmull",
2749 int_arm64_neon_sqdmull>;
2750 defm SSUBL : SIMDLongThreeVectorBHS<0, 0b0010, "ssubl",
2751 BinOpFrag<(sub (sext node:$LHS), (sext node:$RHS))>>;
2752 defm SSUBW : SIMDWideThreeVectorBHS<0, 0b0011, "ssubw",
2753 BinOpFrag<(sub node:$LHS, (sext node:$RHS))>>;
2754 defm UABAL : SIMDLongThreeVectorTiedBHSabal<1, 0b0101, "uabal",
2755 int_arm64_neon_uabd>;
2756 defm UABDL : SIMDLongThreeVectorBHSabdl<1, 0b0111, "uabdl",
2757 int_arm64_neon_uabd>;
2758 defm UADDL : SIMDLongThreeVectorBHS<1, 0b0000, "uaddl",
2759 BinOpFrag<(add (zext node:$LHS), (zext node:$RHS))>>;
2760 defm UADDW : SIMDWideThreeVectorBHS<1, 0b0001, "uaddw",
2761 BinOpFrag<(add node:$LHS, (zext node:$RHS))>>;
2762 defm UMLAL : SIMDLongThreeVectorTiedBHS<1, 0b1000, "umlal",
2763 TriOpFrag<(add node:$LHS, (int_arm64_neon_umull node:$MHS, node:$RHS))>>;
2764 defm UMLSL : SIMDLongThreeVectorTiedBHS<1, 0b1010, "umlsl",
2765 TriOpFrag<(sub node:$LHS, (int_arm64_neon_umull node:$MHS, node:$RHS))>>;
2766 defm UMULL : SIMDLongThreeVectorBHS<1, 0b1100, "umull", int_arm64_neon_umull>;
2767 defm USUBL : SIMDLongThreeVectorBHS<1, 0b0010, "usubl",
2768 BinOpFrag<(sub (zext node:$LHS), (zext node:$RHS))>>;
2769 defm USUBW : SIMDWideThreeVectorBHS< 1, 0b0011, "usubw",
2770 BinOpFrag<(sub node:$LHS, (zext node:$RHS))>>;
2772 // Patterns for 64-bit pmull
2773 def : Pat<(int_arm64_neon_pmull64 V64:$Rn, V64:$Rm),
2774 (PMULLv1i64 V64:$Rn, V64:$Rm)>;
2775 def : Pat<(int_arm64_neon_pmull64 (vector_extract (v2i64 V128:$Rn), (i64 1)),
2776 (vector_extract (v2i64 V128:$Rm), (i64 1))),
2777 (PMULLv2i64 V128:$Rn, V128:$Rm)>;
2779 // CodeGen patterns for addhn and subhn instructions, which can actually be
2780 // written in LLVM IR without too much difficulty.
2783 def : Pat<(v8i8 (trunc (v8i16 (ARM64vlshr (add V128:$Rn, V128:$Rm), (i32 8))))),
2784 (ADDHNv8i16_v8i8 V128:$Rn, V128:$Rm)>;
2785 def : Pat<(v4i16 (trunc (v4i32 (ARM64vlshr (add V128:$Rn, V128:$Rm),
2787 (ADDHNv4i32_v4i16 V128:$Rn, V128:$Rm)>;
2788 def : Pat<(v2i32 (trunc (v2i64 (ARM64vlshr (add V128:$Rn, V128:$Rm),
2790 (ADDHNv2i64_v2i32 V128:$Rn, V128:$Rm)>;
2791 def : Pat<(concat_vectors (v8i8 V64:$Rd),
2792 (trunc (v8i16 (ARM64vlshr (add V128:$Rn, V128:$Rm),
2794 (ADDHNv8i16_v16i8 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
2795 V128:$Rn, V128:$Rm)>;
2796 def : Pat<(concat_vectors (v4i16 V64:$Rd),
2797 (trunc (v4i32 (ARM64vlshr (add V128:$Rn, V128:$Rm),
2799 (ADDHNv4i32_v8i16 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
2800 V128:$Rn, V128:$Rm)>;
2801 def : Pat<(concat_vectors (v2i32 V64:$Rd),
2802 (trunc (v2i64 (ARM64vlshr (add V128:$Rn, V128:$Rm),
2804 (ADDHNv2i64_v4i32 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
2805 V128:$Rn, V128:$Rm)>;
2808 def : Pat<(v8i8 (trunc (v8i16 (ARM64vlshr (sub V128:$Rn, V128:$Rm), (i32 8))))),
2809 (SUBHNv8i16_v8i8 V128:$Rn, V128:$Rm)>;
2810 def : Pat<(v4i16 (trunc (v4i32 (ARM64vlshr (sub V128:$Rn, V128:$Rm),
2812 (SUBHNv4i32_v4i16 V128:$Rn, V128:$Rm)>;
2813 def : Pat<(v2i32 (trunc (v2i64 (ARM64vlshr (sub V128:$Rn, V128:$Rm),
2815 (SUBHNv2i64_v2i32 V128:$Rn, V128:$Rm)>;
2816 def : Pat<(concat_vectors (v8i8 V64:$Rd),
2817 (trunc (v8i16 (ARM64vlshr (sub V128:$Rn, V128:$Rm),
2819 (SUBHNv8i16_v16i8 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
2820 V128:$Rn, V128:$Rm)>;
2821 def : Pat<(concat_vectors (v4i16 V64:$Rd),
2822 (trunc (v4i32 (ARM64vlshr (sub V128:$Rn, V128:$Rm),
2824 (SUBHNv4i32_v8i16 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
2825 V128:$Rn, V128:$Rm)>;
2826 def : Pat<(concat_vectors (v2i32 V64:$Rd),
2827 (trunc (v2i64 (ARM64vlshr (sub V128:$Rn, V128:$Rm),
2829 (SUBHNv2i64_v4i32 (SUBREG_TO_REG (i32 0), V64:$Rd, dsub),
2830 V128:$Rn, V128:$Rm)>;
2832 //----------------------------------------------------------------------------
2833 // AdvSIMD bitwise extract from vector instruction.
2834 //----------------------------------------------------------------------------
2836 defm EXT : SIMDBitwiseExtract<"ext">;
2838 def : Pat<(v4i16 (ARM64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
2839 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
2840 def : Pat<(v8i16 (ARM64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
2841 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
2842 def : Pat<(v2i32 (ARM64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
2843 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
2844 def : Pat<(v2f32 (ARM64ext V64:$Rn, V64:$Rm, (i32 imm:$imm))),
2845 (EXTv8i8 V64:$Rn, V64:$Rm, imm:$imm)>;
2846 def : Pat<(v4i32 (ARM64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
2847 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
2848 def : Pat<(v4f32 (ARM64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
2849 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
2850 def : Pat<(v2i64 (ARM64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
2851 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
2852 def : Pat<(v2f64 (ARM64ext V128:$Rn, V128:$Rm, (i32 imm:$imm))),
2853 (EXTv16i8 V128:$Rn, V128:$Rm, imm:$imm)>;
2855 // We use EXT to handle extract_subvector to copy the upper 64-bits of a
2857 def : Pat<(v8i8 (extract_subvector V128:$Rn, (i64 8))),
2858 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
2859 def : Pat<(v4i16 (extract_subvector V128:$Rn, (i64 4))),
2860 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
2861 def : Pat<(v2i32 (extract_subvector V128:$Rn, (i64 2))),
2862 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
2863 def : Pat<(v1i64 (extract_subvector V128:$Rn, (i64 1))),
2864 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
2865 def : Pat<(v2f32 (extract_subvector V128:$Rn, (i64 2))),
2866 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
2867 def : Pat<(v1f64 (extract_subvector V128:$Rn, (i64 1))),
2868 (EXTRACT_SUBREG (EXTv16i8 V128:$Rn, V128:$Rn, 8), dsub)>;
2871 //----------------------------------------------------------------------------
2872 // AdvSIMD zip vector
2873 //----------------------------------------------------------------------------
2875 defm TRN1 : SIMDZipVector<0b010, "trn1", ARM64trn1>;
2876 defm TRN2 : SIMDZipVector<0b110, "trn2", ARM64trn2>;
2877 defm UZP1 : SIMDZipVector<0b001, "uzp1", ARM64uzp1>;
2878 defm UZP2 : SIMDZipVector<0b101, "uzp2", ARM64uzp2>;
2879 defm ZIP1 : SIMDZipVector<0b011, "zip1", ARM64zip1>;
2880 defm ZIP2 : SIMDZipVector<0b111, "zip2", ARM64zip2>;
2882 //----------------------------------------------------------------------------
2883 // AdvSIMD TBL/TBX instructions
2884 //----------------------------------------------------------------------------
2886 defm TBL : SIMDTableLookup< 0, "tbl">;
2887 defm TBX : SIMDTableLookupTied<1, "tbx">;
2889 def : Pat<(v8i8 (int_arm64_neon_tbl1 (v16i8 VecListOne128:$Rn), (v8i8 V64:$Ri))),
2890 (TBLv8i8One VecListOne128:$Rn, V64:$Ri)>;
2891 def : Pat<(v16i8 (int_arm64_neon_tbl1 (v16i8 V128:$Ri), (v16i8 V128:$Rn))),
2892 (TBLv16i8One V128:$Ri, V128:$Rn)>;
2894 def : Pat<(v8i8 (int_arm64_neon_tbx1 (v8i8 V64:$Rd),
2895 (v16i8 VecListOne128:$Rn), (v8i8 V64:$Ri))),
2896 (TBXv8i8One V64:$Rd, VecListOne128:$Rn, V64:$Ri)>;
2897 def : Pat<(v16i8 (int_arm64_neon_tbx1 (v16i8 V128:$Rd),
2898 (v16i8 V128:$Ri), (v16i8 V128:$Rn))),
2899 (TBXv16i8One V128:$Rd, V128:$Ri, V128:$Rn)>;
2902 //----------------------------------------------------------------------------
2903 // AdvSIMD scalar CPY instruction
2904 //----------------------------------------------------------------------------
2906 defm CPY : SIMDScalarCPY<"cpy">;
2908 //----------------------------------------------------------------------------
2909 // AdvSIMD scalar pairwise instructions
2910 //----------------------------------------------------------------------------
2912 defm ADDP : SIMDPairwiseScalarD<0, 0b11011, "addp">;
2913 defm FADDP : SIMDPairwiseScalarSD<1, 0, 0b01101, "faddp">;
2914 defm FMAXNMP : SIMDPairwiseScalarSD<1, 0, 0b01100, "fmaxnmp">;
2915 defm FMAXP : SIMDPairwiseScalarSD<1, 0, 0b01111, "fmaxp">;
2916 defm FMINNMP : SIMDPairwiseScalarSD<1, 1, 0b01100, "fminnmp">;
2917 defm FMINP : SIMDPairwiseScalarSD<1, 1, 0b01111, "fminp">;
2918 def : Pat<(i64 (int_arm64_neon_saddv (v2i64 V128:$Rn))),
2919 (ADDPv2i64p V128:$Rn)>;
2920 def : Pat<(i64 (int_arm64_neon_uaddv (v2i64 V128:$Rn))),
2921 (ADDPv2i64p V128:$Rn)>;
2922 def : Pat<(f32 (int_arm64_neon_faddv (v2f32 V64:$Rn))),
2923 (FADDPv2i32p V64:$Rn)>;
2924 def : Pat<(f32 (int_arm64_neon_faddv (v4f32 V128:$Rn))),
2925 (FADDPv2i32p (EXTRACT_SUBREG (FADDPv4f32 V128:$Rn, V128:$Rn), dsub))>;
2926 def : Pat<(f64 (int_arm64_neon_faddv (v2f64 V128:$Rn))),
2927 (FADDPv2i64p V128:$Rn)>;
2928 def : Pat<(f32 (int_arm64_neon_fmaxnmv (v2f32 V64:$Rn))),
2929 (FMAXNMPv2i32p V64:$Rn)>;
2930 def : Pat<(f64 (int_arm64_neon_fmaxnmv (v2f64 V128:$Rn))),
2931 (FMAXNMPv2i64p V128:$Rn)>;
2932 def : Pat<(f32 (int_arm64_neon_fmaxv (v2f32 V64:$Rn))),
2933 (FMAXPv2i32p V64:$Rn)>;
2934 def : Pat<(f64 (int_arm64_neon_fmaxv (v2f64 V128:$Rn))),
2935 (FMAXPv2i64p V128:$Rn)>;
2936 def : Pat<(f32 (int_arm64_neon_fminnmv (v2f32 V64:$Rn))),
2937 (FMINNMPv2i32p V64:$Rn)>;
2938 def : Pat<(f64 (int_arm64_neon_fminnmv (v2f64 V128:$Rn))),
2939 (FMINNMPv2i64p V128:$Rn)>;
2940 def : Pat<(f32 (int_arm64_neon_fminv (v2f32 V64:$Rn))),
2941 (FMINPv2i32p V64:$Rn)>;
2942 def : Pat<(f64 (int_arm64_neon_fminv (v2f64 V128:$Rn))),
2943 (FMINPv2i64p V128:$Rn)>;
2945 //----------------------------------------------------------------------------
2946 // AdvSIMD INS/DUP instructions
2947 //----------------------------------------------------------------------------
2949 def DUPv8i8gpr : SIMDDupFromMain<0, 0b00001, ".8b", v8i8, V64, GPR32>;
2950 def DUPv16i8gpr : SIMDDupFromMain<1, 0b00001, ".16b", v16i8, V128, GPR32>;
2951 def DUPv4i16gpr : SIMDDupFromMain<0, 0b00010, ".4h", v4i16, V64, GPR32>;
2952 def DUPv8i16gpr : SIMDDupFromMain<1, 0b00010, ".8h", v8i16, V128, GPR32>;
2953 def DUPv2i32gpr : SIMDDupFromMain<0, 0b00100, ".2s", v2i32, V64, GPR32>;
2954 def DUPv4i32gpr : SIMDDupFromMain<1, 0b00100, ".4s", v4i32, V128, GPR32>;
2955 def DUPv2i64gpr : SIMDDupFromMain<1, 0b01000, ".2d", v2i64, V128, GPR64>;
2957 def DUPv2i64lane : SIMDDup64FromElement;
2958 def DUPv2i32lane : SIMDDup32FromElement<0, ".2s", v2i32, V64>;
2959 def DUPv4i32lane : SIMDDup32FromElement<1, ".4s", v4i32, V128>;
2960 def DUPv4i16lane : SIMDDup16FromElement<0, ".4h", v4i16, V64>;
2961 def DUPv8i16lane : SIMDDup16FromElement<1, ".8h", v8i16, V128>;
2962 def DUPv8i8lane : SIMDDup8FromElement <0, ".8b", v8i8, V64>;
2963 def DUPv16i8lane : SIMDDup8FromElement <1, ".16b", v16i8, V128>;
2965 def : Pat<(v2f32 (ARM64dup (f32 FPR32:$Rn))),
2966 (v2f32 (DUPv2i32lane
2967 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rn, ssub),
2969 def : Pat<(v4f32 (ARM64dup (f32 FPR32:$Rn))),
2970 (v4f32 (DUPv4i32lane
2971 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rn, ssub),
2973 def : Pat<(v2f64 (ARM64dup (f64 FPR64:$Rn))),
2974 (v2f64 (DUPv2i64lane
2975 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$Rn, dsub),
2978 def : Pat<(v2f32 (ARM64duplane32 (v4f32 V128:$Rn), VectorIndexS:$imm)),
2979 (DUPv2i32lane V128:$Rn, VectorIndexS:$imm)>;
2980 def : Pat<(v4f32 (ARM64duplane32 (v4f32 V128:$Rn), VectorIndexS:$imm)),
2981 (DUPv4i32lane V128:$Rn, VectorIndexS:$imm)>;
2982 def : Pat<(v2f64 (ARM64duplane64 (v2f64 V128:$Rn), VectorIndexD:$imm)),
2983 (DUPv2i64lane V128:$Rn, VectorIndexD:$imm)>;
2988 def : Pat<(sext_inreg (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx), i8),
2989 (i32 (SMOVvi8to32 V128:$Rn, VectorIndexB:$idx))>;
2990 def : Pat<(sext_inreg (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx), i8),
2991 (i64 (SMOVvi8to64 V128:$Rn, VectorIndexB:$idx))>;
2992 def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
2993 (i32 (SMOVvi16to32 V128:$Rn, VectorIndexH:$idx))>;
2994 def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
2995 (i64 (SMOVvi16to64 V128:$Rn, VectorIndexH:$idx))>;
2996 def : Pat<(sext_inreg (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),i16),
2997 (i32 (SMOVvi16to32 V128:$Rn, VectorIndexH:$idx))>;
2998 def : Pat<(sext (i32 (vector_extract (v4i32 V128:$Rn), VectorIndexS:$idx))),
2999 (i64 (SMOVvi32to64 V128:$Rn, VectorIndexS:$idx))>;
3001 // Extracting i8 or i16 elements will have the zero-extend transformed to
3002 // an 'and' mask by type legalization since neither i8 nor i16 are legal types
3003 // for ARM64. Match these patterns here since UMOV already zeroes out the high
3004 // bits of the destination register.
3005 def : Pat<(and (vector_extract (v16i8 V128:$Rn), VectorIndexB:$idx),
3007 (i32 (UMOVvi8 V128:$Rn, VectorIndexB:$idx))>;
3008 def : Pat<(and (vector_extract (v8i16 V128:$Rn), VectorIndexH:$idx),
3010 (i32 (UMOVvi16 V128:$Rn, VectorIndexH:$idx))>;
3014 def : Pat<(v16i8 (scalar_to_vector GPR32:$Rn)),
3015 (SUBREG_TO_REG (i32 0),
3016 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3017 def : Pat<(v8i8 (scalar_to_vector GPR32:$Rn)),
3018 (SUBREG_TO_REG (i32 0),
3019 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3021 def : Pat<(v8i16 (scalar_to_vector GPR32:$Rn)),
3022 (SUBREG_TO_REG (i32 0),
3023 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3024 def : Pat<(v4i16 (scalar_to_vector GPR32:$Rn)),
3025 (SUBREG_TO_REG (i32 0),
3026 (f32 (COPY_TO_REGCLASS GPR32:$Rn, FPR32)), ssub)>;
3028 def : Pat<(v2i32 (scalar_to_vector (i32 FPR32:$Rn))),
3029 (v2i32 (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)),
3030 (i32 FPR32:$Rn), ssub))>;
3031 def : Pat<(v4i32 (scalar_to_vector (i32 FPR32:$Rn))),
3032 (v4i32 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
3033 (i32 FPR32:$Rn), ssub))>;
3034 def : Pat<(v2i64 (scalar_to_vector (i64 FPR64:$Rn))),
3035 (v2i64 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
3036 (i64 FPR64:$Rn), dsub))>;
3038 def : Pat<(v4f32 (scalar_to_vector (f32 FPR32:$Rn))),
3039 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rn, ssub)>;
3040 def : Pat<(v2f32 (scalar_to_vector (f32 FPR32:$Rn))),
3041 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), FPR32:$Rn, ssub)>;
3042 def : Pat<(v2f64 (scalar_to_vector (f64 FPR64:$Rn))),
3043 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$Rn, dsub)>;
3045 def : Pat<(v2f32 (vector_insert (v2f32 V64:$Rn),
3046 (f32 FPR32:$Rm), (i64 VectorIndexS:$imm))),
3049 (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), V64:$Rn, dsub)),
3051 (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rm, ssub)),
3054 def : Pat<(v4f32 (vector_insert (v4f32 V128:$Rn),
3055 (f32 FPR32:$Rm), (i64 VectorIndexS:$imm))),
3057 V128:$Rn, VectorIndexS:$imm,
3058 (v4f32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR32:$Rm, ssub)),
3060 def : Pat<(v2f64 (vector_insert (v2f64 V128:$Rn),
3061 (f64 FPR64:$Rm), (i64 VectorIndexD:$imm))),
3063 V128:$Rn, VectorIndexD:$imm,
3064 (v2f64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$Rm, dsub)),
3067 // Copy an element at a constant index in one vector into a constant indexed
3068 // element of another.
3069 // FIXME refactor to a shared class/dev parameterized on vector type, vector
3070 // index type and INS extension
3071 def : Pat<(v16i8 (int_arm64_neon_vcopy_lane
3072 (v16i8 V128:$Vd), VectorIndexB:$idx, (v16i8 V128:$Vs),
3073 VectorIndexB:$idx2)),
3075 V128:$Vd, VectorIndexB:$idx, V128:$Vs, VectorIndexB:$idx2)
3077 def : Pat<(v8i16 (int_arm64_neon_vcopy_lane
3078 (v8i16 V128:$Vd), VectorIndexH:$idx, (v8i16 V128:$Vs),
3079 VectorIndexH:$idx2)),
3081 V128:$Vd, VectorIndexH:$idx, V128:$Vs, VectorIndexH:$idx2)
3083 def : Pat<(v4i32 (int_arm64_neon_vcopy_lane
3084 (v4i32 V128:$Vd), VectorIndexS:$idx, (v4i32 V128:$Vs),
3085 VectorIndexS:$idx2)),
3087 V128:$Vd, VectorIndexS:$idx, V128:$Vs, VectorIndexS:$idx2)
3089 def : Pat<(v2i64 (int_arm64_neon_vcopy_lane
3090 (v2i64 V128:$Vd), VectorIndexD:$idx, (v2i64 V128:$Vs),
3091 VectorIndexD:$idx2)),
3093 V128:$Vd, VectorIndexD:$idx, V128:$Vs, VectorIndexD:$idx2)
3096 // Floating point vector extractions are codegen'd as either a sequence of
3097 // subregister extractions, possibly fed by an INS if the lane number is
3098 // anything other than zero.
3099 def : Pat<(vector_extract (v2f64 V128:$Rn), 0),
3100 (f64 (EXTRACT_SUBREG V128:$Rn, dsub))>;
3101 def : Pat<(vector_extract (v4f32 V128:$Rn), 0),
3102 (f32 (EXTRACT_SUBREG V128:$Rn, ssub))>;
3103 def : Pat<(vector_extract (v2f64 V128:$Rn), VectorIndexD:$idx),
3104 (f64 (EXTRACT_SUBREG
3105 (INSvi64lane (v2f64 (IMPLICIT_DEF)), 0,
3106 V128:$Rn, VectorIndexD:$idx),
3108 def : Pat<(vector_extract (v4f32 V128:$Rn), VectorIndexS:$idx),
3109 (f32 (EXTRACT_SUBREG
3110 (INSvi32lane (v4f32 (IMPLICIT_DEF)), 0,
3111 V128:$Rn, VectorIndexS:$idx),
3114 // All concat_vectors operations are canonicalised to act on i64 vectors for
3115 // ARM64. In the general case we need an instruction, which had just as well be
3117 class ConcatPat<ValueType DstTy, ValueType SrcTy>
3118 : Pat<(DstTy (concat_vectors (SrcTy V64:$Rd), V64:$Rn)),
3119 (INSvi64lane (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), 1,
3120 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rn, dsub), 0)>;
3122 def : ConcatPat<v2i64, v1i64>;
3123 def : ConcatPat<v2f64, v1f64>;
3124 def : ConcatPat<v4i32, v2i32>;
3125 def : ConcatPat<v4f32, v2f32>;
3126 def : ConcatPat<v8i16, v4i16>;
3127 def : ConcatPat<v16i8, v8i8>;
3129 // If the high lanes are undef, though, we can just ignore them:
3130 class ConcatUndefPat<ValueType DstTy, ValueType SrcTy>
3131 : Pat<(DstTy (concat_vectors (SrcTy V64:$Rn), undef)),
3132 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rn, dsub)>;
3134 def : ConcatUndefPat<v2i64, v1i64>;
3135 def : ConcatUndefPat<v2f64, v1f64>;
3136 def : ConcatUndefPat<v4i32, v2i32>;
3137 def : ConcatUndefPat<v4f32, v2f32>;
3138 def : ConcatUndefPat<v8i16, v4i16>;
3139 def : ConcatUndefPat<v16i8, v8i8>;
3141 //----------------------------------------------------------------------------
3142 // AdvSIMD across lanes instructions
3143 //----------------------------------------------------------------------------
3145 defm ADDV : SIMDAcrossLanesBHS<0, 0b11011, "addv">;
3146 defm SMAXV : SIMDAcrossLanesBHS<0, 0b01010, "smaxv">;
3147 defm SMINV : SIMDAcrossLanesBHS<0, 0b11010, "sminv">;
3148 defm UMAXV : SIMDAcrossLanesBHS<1, 0b01010, "umaxv">;
3149 defm UMINV : SIMDAcrossLanesBHS<1, 0b11010, "uminv">;
3150 defm SADDLV : SIMDAcrossLanesHSD<0, 0b00011, "saddlv">;
3151 defm UADDLV : SIMDAcrossLanesHSD<1, 0b00011, "uaddlv">;
3152 defm FMAXNMV : SIMDAcrossLanesS<0b01100, 0, "fmaxnmv", int_arm64_neon_fmaxnmv>;
3153 defm FMAXV : SIMDAcrossLanesS<0b01111, 0, "fmaxv", int_arm64_neon_fmaxv>;
3154 defm FMINNMV : SIMDAcrossLanesS<0b01100, 1, "fminnmv", int_arm64_neon_fminnmv>;
3155 defm FMINV : SIMDAcrossLanesS<0b01111, 1, "fminv", int_arm64_neon_fminv>;
3157 multiclass SIMDAcrossLanesSignedIntrinsic<string baseOpc, Intrinsic intOp> {
3158 // If there is a sign extension after this intrinsic, consume it as smov already
3160 def : Pat<(i32 (sext_inreg (i32 (intOp (v8i8 V64:$Rn))), i8)),
3162 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3163 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3165 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
3167 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3168 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3170 // If there is a sign extension after this intrinsic, consume it as smov already
3172 def : Pat<(i32 (sext_inreg (i32 (intOp (v16i8 V128:$Rn))), i8)),
3174 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3175 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3177 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
3179 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3180 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3182 // If there is a sign extension after this intrinsic, consume it as smov already
3184 def : Pat<(i32 (sext_inreg (i32 (intOp (v4i16 V64:$Rn))), i16)),
3186 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3187 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3189 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
3191 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3192 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3194 // If there is a sign extension after this intrinsic, consume it as smov already
3196 def : Pat<(i32 (sext_inreg (i32 (intOp (v8i16 V128:$Rn))), i16)),
3198 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3199 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3201 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
3203 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3204 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3207 def : Pat<(i32 (intOp (v4i32 V128:$Rn))),
3208 (i32 (EXTRACT_SUBREG
3209 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3210 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), ssub),
3214 multiclass SIMDAcrossLanesUnsignedIntrinsic<string baseOpc, Intrinsic intOp> {
3215 // If there is a masking operation keeping only what has been actually
3216 // generated, consume it.
3217 def : Pat<(i32 (and (i32 (intOp (v8i8 V64:$Rn))), maski8_or_more)),
3218 (i32 (EXTRACT_SUBREG
3219 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3220 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3222 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
3223 (i32 (EXTRACT_SUBREG
3224 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3225 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), bsub),
3227 // If there is a masking operation keeping only what has been actually
3228 // generated, consume it.
3229 def : Pat<(i32 (and (i32 (intOp (v16i8 V128:$Rn))), maski8_or_more)),
3230 (i32 (EXTRACT_SUBREG
3231 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3232 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3234 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
3235 (i32 (EXTRACT_SUBREG
3236 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3237 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), bsub),
3240 // If there is a masking operation keeping only what has been actually
3241 // generated, consume it.
3242 def : Pat<(i32 (and (i32 (intOp (v4i16 V64:$Rn))), maski16_or_more)),
3243 (i32 (EXTRACT_SUBREG
3244 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3245 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3247 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
3248 (i32 (EXTRACT_SUBREG
3249 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3250 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), hsub),
3252 // If there is a masking operation keeping only what has been actually
3253 // generated, consume it.
3254 def : Pat<(i32 (and (i32 (intOp (v8i16 V128:$Rn))), maski16_or_more)),
3255 (i32 (EXTRACT_SUBREG
3256 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3257 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3259 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
3260 (i32 (EXTRACT_SUBREG
3261 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3262 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), hsub),
3265 def : Pat<(i32 (intOp (v4i32 V128:$Rn))),
3266 (i32 (EXTRACT_SUBREG
3267 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3268 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), ssub),
3273 multiclass SIMDAcrossLanesSignedLongIntrinsic<string baseOpc, Intrinsic intOp> {
3274 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
3276 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3277 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), hsub),
3279 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
3281 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3282 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), hsub),
3285 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
3286 (i32 (EXTRACT_SUBREG
3287 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3288 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), ssub),
3290 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
3291 (i32 (EXTRACT_SUBREG
3292 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3293 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), ssub),
3296 def : Pat<(i64 (intOp (v4i32 V128:$Rn))),
3297 (i64 (EXTRACT_SUBREG
3298 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3299 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), dsub),
3303 multiclass SIMDAcrossLanesUnsignedLongIntrinsic<string baseOpc,
3305 def : Pat<(i32 (intOp (v8i8 V64:$Rn))),
3306 (i32 (EXTRACT_SUBREG
3307 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3308 (!cast<Instruction>(!strconcat(baseOpc, "v8i8v")) V64:$Rn), hsub),
3310 def : Pat<(i32 (intOp (v16i8 V128:$Rn))),
3311 (i32 (EXTRACT_SUBREG
3312 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3313 (!cast<Instruction>(!strconcat(baseOpc, "v16i8v")) V128:$Rn), hsub),
3316 def : Pat<(i32 (intOp (v4i16 V64:$Rn))),
3317 (i32 (EXTRACT_SUBREG
3318 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3319 (!cast<Instruction>(!strconcat(baseOpc, "v4i16v")) V64:$Rn), ssub),
3321 def : Pat<(i32 (intOp (v8i16 V128:$Rn))),
3322 (i32 (EXTRACT_SUBREG
3323 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3324 (!cast<Instruction>(!strconcat(baseOpc, "v8i16v")) V128:$Rn), ssub),
3327 def : Pat<(i64 (intOp (v4i32 V128:$Rn))),
3328 (i64 (EXTRACT_SUBREG
3329 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3330 (!cast<Instruction>(!strconcat(baseOpc, "v4i32v")) V128:$Rn), dsub),
3334 defm : SIMDAcrossLanesSignedIntrinsic<"ADDV", int_arm64_neon_saddv>;
3335 // vaddv_[su]32 is special; -> ADDP Vd.2S,Vn.2S,Vm.2S; return Vd.s[0];Vn==Vm
3336 def : Pat<(i32 (int_arm64_neon_saddv (v2i32 V64:$Rn))),
3337 (EXTRACT_SUBREG (ADDPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3339 defm : SIMDAcrossLanesUnsignedIntrinsic<"ADDV", int_arm64_neon_uaddv>;
3340 // vaddv_[su]32 is special; -> ADDP Vd.2S,Vn.2S,Vm.2S; return Vd.s[0];Vn==Vm
3341 def : Pat<(i32 (int_arm64_neon_uaddv (v2i32 V64:$Rn))),
3342 (EXTRACT_SUBREG (ADDPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3344 defm : SIMDAcrossLanesSignedIntrinsic<"SMAXV", int_arm64_neon_smaxv>;
3345 def : Pat<(i32 (int_arm64_neon_smaxv (v2i32 V64:$Rn))),
3346 (EXTRACT_SUBREG (SMAXPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3348 defm : SIMDAcrossLanesSignedIntrinsic<"SMINV", int_arm64_neon_sminv>;
3349 def : Pat<(i32 (int_arm64_neon_sminv (v2i32 V64:$Rn))),
3350 (EXTRACT_SUBREG (SMINPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3352 defm : SIMDAcrossLanesUnsignedIntrinsic<"UMAXV", int_arm64_neon_umaxv>;
3353 def : Pat<(i32 (int_arm64_neon_umaxv (v2i32 V64:$Rn))),
3354 (EXTRACT_SUBREG (UMAXPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3356 defm : SIMDAcrossLanesUnsignedIntrinsic<"UMINV", int_arm64_neon_uminv>;
3357 def : Pat<(i32 (int_arm64_neon_uminv (v2i32 V64:$Rn))),
3358 (EXTRACT_SUBREG (UMINPv2i32 V64:$Rn, V64:$Rn), ssub)>;
3360 defm : SIMDAcrossLanesSignedLongIntrinsic<"SADDLV", int_arm64_neon_saddlv>;
3361 defm : SIMDAcrossLanesUnsignedLongIntrinsic<"UADDLV", int_arm64_neon_uaddlv>;
3363 // The vaddlv_s32 intrinsic gets mapped to SADDLP.
3364 def : Pat<(i64 (int_arm64_neon_saddlv (v2i32 V64:$Rn))),
3365 (i64 (EXTRACT_SUBREG
3366 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3367 (SADDLPv2i32_v1i64 V64:$Rn), dsub),
3369 // The vaddlv_u32 intrinsic gets mapped to UADDLP.
3370 def : Pat<(i64 (int_arm64_neon_uaddlv (v2i32 V64:$Rn))),
3371 (i64 (EXTRACT_SUBREG
3372 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3373 (UADDLPv2i32_v1i64 V64:$Rn), dsub),
3376 //------------------------------------------------------------------------------
3377 // AdvSIMD modified immediate instructions
3378 //------------------------------------------------------------------------------
3381 defm BIC : SIMDModifiedImmVectorShiftTied<1, 0b11, 0b01, "bic", ARM64bici>;
3383 defm ORR : SIMDModifiedImmVectorShiftTied<0, 0b11, 0b01, "orr", ARM64orri>;
3387 def FMOVv2f64_ns : SIMDModifiedImmVectorNoShift<1, 1, 0b1111, V128, fpimm8,
3389 [(set (v2f64 V128:$Rd), (ARM64fmov imm0_255:$imm8))]>;
3390 def FMOVv2f32_ns : SIMDModifiedImmVectorNoShift<0, 0, 0b1111, V64, fpimm8,
3392 [(set (v2f32 V64:$Rd), (ARM64fmov imm0_255:$imm8))]>;
3393 def FMOVv4f32_ns : SIMDModifiedImmVectorNoShift<1, 0, 0b1111, V128, fpimm8,
3395 [(set (v4f32 V128:$Rd), (ARM64fmov imm0_255:$imm8))]>;
3399 // EDIT byte mask: scalar
3400 let isReMaterializable = 1, isAsCheapAsAMove = 1 in
3401 def MOVID : SIMDModifiedImmScalarNoShift<0, 1, 0b1110, "movi",
3402 [(set FPR64:$Rd, simdimmtype10:$imm8)]>;
3403 // The movi_edit node has the immediate value already encoded, so we use
3404 // a plain imm0_255 here.
3405 def : Pat<(f64 (ARM64movi_edit imm0_255:$shift)),
3406 (MOVID imm0_255:$shift)>;
3408 def : Pat<(v1i64 immAllZerosV), (MOVID (i32 0))>;
3409 def : Pat<(v2i32 immAllZerosV), (MOVID (i32 0))>;
3410 def : Pat<(v4i16 immAllZerosV), (MOVID (i32 0))>;
3411 def : Pat<(v8i8 immAllZerosV), (MOVID (i32 0))>;
3413 def : Pat<(v1i64 immAllOnesV), (MOVID (i32 255))>;
3414 def : Pat<(v2i32 immAllOnesV), (MOVID (i32 255))>;
3415 def : Pat<(v4i16 immAllOnesV), (MOVID (i32 255))>;
3416 def : Pat<(v8i8 immAllOnesV), (MOVID (i32 255))>;
3418 // EDIT byte mask: 2d
3420 // The movi_edit node has the immediate value already encoded, so we use
3421 // a plain imm0_255 in the pattern
3422 let isReMaterializable = 1, isAsCheapAsAMove = 1 in
3423 def MOVIv2d_ns : SIMDModifiedImmVectorNoShift<1, 1, 0b1110, V128,
3426 [(set (v2i64 V128:$Rd), (ARM64movi_edit imm0_255:$imm8))]>;
3429 // Use movi.2d to materialize 0.0 if the HW does zero-cycle zeroing.
3430 // Complexity is added to break a tie with a plain MOVI.
3431 let AddedComplexity = 1 in {
3432 def : Pat<(f32 fpimm0),
3433 (f32 (EXTRACT_SUBREG (v2i64 (MOVIv2d_ns (i32 0))), ssub))>,
3435 def : Pat<(f64 fpimm0),
3436 (f64 (EXTRACT_SUBREG (v2i64 (MOVIv2d_ns (i32 0))), dsub))>,
3440 def : Pat<(v2i64 immAllZerosV), (MOVIv2d_ns (i32 0))>;
3441 def : Pat<(v4i32 immAllZerosV), (MOVIv2d_ns (i32 0))>;
3442 def : Pat<(v8i16 immAllZerosV), (MOVIv2d_ns (i32 0))>;
3443 def : Pat<(v16i8 immAllZerosV), (MOVIv2d_ns (i32 0))>;
3445 def : Pat<(v2i64 immAllOnesV), (MOVIv2d_ns (i32 255))>;
3446 def : Pat<(v4i32 immAllOnesV), (MOVIv2d_ns (i32 255))>;
3447 def : Pat<(v8i16 immAllOnesV), (MOVIv2d_ns (i32 255))>;
3448 def : Pat<(v16i8 immAllOnesV), (MOVIv2d_ns (i32 255))>;
3450 // EDIT per word & halfword: 2s, 4h, 4s, & 8h
3451 defm MOVI : SIMDModifiedImmVectorShift<0, 0b10, 0b00, "movi">;
3452 def : Pat<(v2i32 (ARM64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
3453 (MOVIv2i32 imm0_255:$imm8, imm:$shift)>;
3454 def : Pat<(v4i32 (ARM64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
3455 (MOVIv4i32 imm0_255:$imm8, imm:$shift)>;
3456 def : Pat<(v4i16 (ARM64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
3457 (MOVIv4i16 imm0_255:$imm8, imm:$shift)>;
3458 def : Pat<(v8i16 (ARM64movi_shift imm0_255:$imm8, (i32 imm:$shift))),
3459 (MOVIv8i16 imm0_255:$imm8, imm:$shift)>;
3461 // EDIT per word: 2s & 4s with MSL shifter
3462 def MOVIv2s_msl : SIMDModifiedImmMoveMSL<0, 0, {1,1,0,?}, V64, "movi", ".2s",
3463 [(set (v2i32 V64:$Rd),
3464 (ARM64movi_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
3465 def MOVIv4s_msl : SIMDModifiedImmMoveMSL<1, 0, {1,1,0,?}, V128, "movi", ".4s",
3466 [(set (v4i32 V128:$Rd),
3467 (ARM64movi_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
3469 // Per byte: 8b & 16b
3470 def MOVIv8b_ns : SIMDModifiedImmVectorNoShift<0, 0, 0b1110, V64, imm0_255,
3472 [(set (v8i8 V64:$Rd), (ARM64movi imm0_255:$imm8))]>;
3473 def MOVIv16b_ns : SIMDModifiedImmVectorNoShift<1, 0, 0b1110, V128, imm0_255,
3475 [(set (v16i8 V128:$Rd), (ARM64movi imm0_255:$imm8))]>;
3479 // EDIT per word & halfword: 2s, 4h, 4s, & 8h
3480 defm MVNI : SIMDModifiedImmVectorShift<1, 0b10, 0b00, "mvni">;
3481 def : Pat<(v2i32 (ARM64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
3482 (MVNIv2i32 imm0_255:$imm8, imm:$shift)>;
3483 def : Pat<(v4i32 (ARM64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
3484 (MVNIv4i32 imm0_255:$imm8, imm:$shift)>;
3485 def : Pat<(v4i16 (ARM64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
3486 (MVNIv4i16 imm0_255:$imm8, imm:$shift)>;
3487 def : Pat<(v8i16 (ARM64mvni_shift imm0_255:$imm8, (i32 imm:$shift))),
3488 (MVNIv8i16 imm0_255:$imm8, imm:$shift)>;
3490 // EDIT per word: 2s & 4s with MSL shifter
3491 def MVNIv2s_msl : SIMDModifiedImmMoveMSL<0, 1, {1,1,0,?}, V64, "mvni", ".2s",
3492 [(set (v2i32 V64:$Rd),
3493 (ARM64mvni_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
3494 def MVNIv4s_msl : SIMDModifiedImmMoveMSL<1, 1, {1,1,0,?}, V128, "mvni", ".4s",
3495 [(set (v4i32 V128:$Rd),
3496 (ARM64mvni_msl imm0_255:$imm8, (i32 imm:$shift)))]>;
3498 //----------------------------------------------------------------------------
3499 // AdvSIMD indexed element
3500 //----------------------------------------------------------------------------
3502 let neverHasSideEffects = 1 in {
3503 defm FMLA : SIMDFPIndexedSDTied<0, 0b0001, "fmla">;
3504 defm FMLS : SIMDFPIndexedSDTied<0, 0b0101, "fmls">;
3507 // NOTE: Operands are reordered in the FMLA/FMLS PatFrags because the
3508 // instruction expects the addend first, while the intrinsic expects it last.
3510 // On the other hand, there are quite a few valid combinatorial options due to
3511 // the commutativity of multiplication and the fact that (-x) * y = x * (-y).
3512 defm : SIMDFPIndexedSDTiedPatterns<"FMLA",
3513 TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)>>;
3514 defm : SIMDFPIndexedSDTiedPatterns<"FMLA",
3515 TriOpFrag<(fma node:$MHS, node:$RHS, node:$LHS)>>;
3517 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
3518 TriOpFrag<(fma node:$MHS, (fneg node:$RHS), node:$LHS)> >;
3519 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
3520 TriOpFrag<(fma node:$RHS, (fneg node:$MHS), node:$LHS)> >;
3521 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
3522 TriOpFrag<(fma (fneg node:$RHS), node:$MHS, node:$LHS)> >;
3523 defm : SIMDFPIndexedSDTiedPatterns<"FMLS",
3524 TriOpFrag<(fma (fneg node:$MHS), node:$RHS, node:$LHS)> >;
3526 multiclass FMLSIndexedAfterNegPatterns<SDPatternOperator OpNode> {
3527 // 3 variants for the .2s version: DUPLANE from 128-bit, DUPLANE from 64-bit
3529 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
3530 (ARM64duplane32 (v4f32 (fneg V128:$Rm)),
3531 VectorIndexS:$idx))),
3532 (FMLSv2i32_indexed V64:$Rd, V64:$Rn, V128:$Rm, VectorIndexS:$idx)>;
3533 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
3534 (v2f32 (ARM64duplane32
3535 (v4f32 (insert_subvector undef,
3536 (v2f32 (fneg V64:$Rm)),
3538 VectorIndexS:$idx)))),
3539 (FMLSv2i32_indexed V64:$Rd, V64:$Rn,
3540 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub),
3541 VectorIndexS:$idx)>;
3542 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
3543 (ARM64dup (f32 (fneg FPR32Op:$Rm))))),
3544 (FMLSv2i32_indexed V64:$Rd, V64:$Rn,
3545 (SUBREG_TO_REG (i32 0), FPR32Op:$Rm, ssub), (i64 0))>;
3547 // 3 variants for the .4s version: DUPLANE from 128-bit, DUPLANE from 64-bit
3549 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
3550 (ARM64duplane32 (v4f32 (fneg V128:$Rm)),
3551 VectorIndexS:$idx))),
3552 (FMLSv4i32_indexed V128:$Rd, V128:$Rn, V128:$Rm,
3553 VectorIndexS:$idx)>;
3554 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
3555 (v4f32 (ARM64duplane32
3556 (v4f32 (insert_subvector undef,
3557 (v2f32 (fneg V64:$Rm)),
3559 VectorIndexS:$idx)))),
3560 (FMLSv4i32_indexed V128:$Rd, V128:$Rn,
3561 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub),
3562 VectorIndexS:$idx)>;
3563 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
3564 (ARM64dup (f32 (fneg FPR32Op:$Rm))))),
3565 (FMLSv4i32_indexed V128:$Rd, V128:$Rn,
3566 (SUBREG_TO_REG (i32 0), FPR32Op:$Rm, ssub), (i64 0))>;
3568 // 2 variants for the .2d version: DUPLANE from 128-bit, and DUP scalar
3569 // (DUPLANE from 64-bit would be trivial).
3570 def : Pat<(v2f64 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn),
3571 (ARM64duplane64 (v2f64 (fneg V128:$Rm)),
3572 VectorIndexD:$idx))),
3574 V128:$Rd, V128:$Rn, V128:$Rm, VectorIndexS:$idx)>;
3575 def : Pat<(v2f64 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn),
3576 (ARM64dup (f64 (fneg FPR64Op:$Rm))))),
3577 (FMLSv2i64_indexed V128:$Rd, V128:$Rn,
3578 (SUBREG_TO_REG (i32 0), FPR64Op:$Rm, dsub), (i64 0))>;
3580 // 2 variants for 32-bit scalar version: extract from .2s or from .4s
3581 def : Pat<(f32 (OpNode (f32 FPR32:$Rd), (f32 FPR32:$Rn),
3582 (vector_extract (v4f32 (fneg V128:$Rm)),
3583 VectorIndexS:$idx))),
3584 (FMLSv1i32_indexed FPR32:$Rd, FPR32:$Rn,
3585 V128:$Rm, VectorIndexS:$idx)>;
3586 def : Pat<(f32 (OpNode (f32 FPR32:$Rd), (f32 FPR32:$Rn),
3587 (vector_extract (v2f32 (fneg V64:$Rm)),
3588 VectorIndexS:$idx))),
3589 (FMLSv1i32_indexed FPR32:$Rd, FPR32:$Rn,
3590 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub), VectorIndexS:$idx)>;
3592 // 1 variant for 64-bit scalar version: extract from .1d or from .2d
3593 def : Pat<(f64 (OpNode (f64 FPR64:$Rd), (f64 FPR64:$Rn),
3594 (vector_extract (v2f64 (fneg V128:$Rm)),
3595 VectorIndexS:$idx))),
3596 (FMLSv1i64_indexed FPR64:$Rd, FPR64:$Rn,
3597 V128:$Rm, VectorIndexS:$idx)>;
3600 defm : FMLSIndexedAfterNegPatterns<
3601 TriOpFrag<(fma node:$RHS, node:$MHS, node:$LHS)> >;
3602 defm : FMLSIndexedAfterNegPatterns<
3603 TriOpFrag<(fma node:$MHS, node:$RHS, node:$LHS)> >;
3605 defm FMULX : SIMDFPIndexedSD<1, 0b1001, "fmulx", int_arm64_neon_fmulx>;
3606 defm FMUL : SIMDFPIndexedSD<0, 0b1001, "fmul", fmul>;
3608 def : Pat<(v2f32 (fmul V64:$Rn, (ARM64dup (f32 FPR32:$Rm)))),
3609 (FMULv2i32_indexed V64:$Rn,
3610 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rm, ssub),
3612 def : Pat<(v4f32 (fmul V128:$Rn, (ARM64dup (f32 FPR32:$Rm)))),
3613 (FMULv4i32_indexed V128:$Rn,
3614 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR32:$Rm, ssub),
3616 def : Pat<(v2f64 (fmul V128:$Rn, (ARM64dup (f64 FPR64:$Rm)))),
3617 (FMULv2i64_indexed V128:$Rn,
3618 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$Rm, dsub),
3621 defm SQDMULH : SIMDIndexedHS<0, 0b1100, "sqdmulh", int_arm64_neon_sqdmulh>;
3622 defm SQRDMULH : SIMDIndexedHS<0, 0b1101, "sqrdmulh", int_arm64_neon_sqrdmulh>;
3623 defm MLA : SIMDVectorIndexedHSTied<1, 0b0000, "mla",
3624 TriOpFrag<(add node:$LHS, (mul node:$MHS, node:$RHS))>>;
3625 defm MLS : SIMDVectorIndexedHSTied<1, 0b0100, "mls",
3626 TriOpFrag<(sub node:$LHS, (mul node:$MHS, node:$RHS))>>;
3627 defm MUL : SIMDVectorIndexedHS<0, 0b1000, "mul", mul>;
3628 defm SMLAL : SIMDVectorIndexedLongSDTied<0, 0b0010, "smlal",
3629 TriOpFrag<(add node:$LHS, (int_arm64_neon_smull node:$MHS, node:$RHS))>>;
3630 defm SMLSL : SIMDVectorIndexedLongSDTied<0, 0b0110, "smlsl",
3631 TriOpFrag<(sub node:$LHS, (int_arm64_neon_smull node:$MHS, node:$RHS))>>;
3632 defm SMULL : SIMDVectorIndexedLongSD<0, 0b1010, "smull",
3633 int_arm64_neon_smull>;
3634 defm SQDMLAL : SIMDIndexedLongSQDMLXSDTied<0, 0b0011, "sqdmlal",
3635 int_arm64_neon_sqadd>;
3636 defm SQDMLSL : SIMDIndexedLongSQDMLXSDTied<0, 0b0111, "sqdmlsl",
3637 int_arm64_neon_sqsub>;
3638 defm SQDMULL : SIMDIndexedLongSD<0, 0b1011, "sqdmull", int_arm64_neon_sqdmull>;
3639 defm UMLAL : SIMDVectorIndexedLongSDTied<1, 0b0010, "umlal",
3640 TriOpFrag<(add node:$LHS, (int_arm64_neon_umull node:$MHS, node:$RHS))>>;
3641 defm UMLSL : SIMDVectorIndexedLongSDTied<1, 0b0110, "umlsl",
3642 TriOpFrag<(sub node:$LHS, (int_arm64_neon_umull node:$MHS, node:$RHS))>>;
3643 defm UMULL : SIMDVectorIndexedLongSD<1, 0b1010, "umull",
3644 int_arm64_neon_umull>;
3646 // A scalar sqdmull with the second operand being a vector lane can be
3647 // handled directly with the indexed instruction encoding.
3648 def : Pat<(int_arm64_neon_sqdmulls_scalar (i32 FPR32:$Rn),
3649 (vector_extract (v4i32 V128:$Vm),
3650 VectorIndexS:$idx)),
3651 (SQDMULLv1i64_indexed FPR32:$Rn, V128:$Vm, VectorIndexS:$idx)>;
3653 //----------------------------------------------------------------------------
3654 // AdvSIMD scalar shift instructions
3655 //----------------------------------------------------------------------------
3656 defm FCVTZS : SIMDScalarRShiftSD<0, 0b11111, "fcvtzs">;
3657 defm FCVTZU : SIMDScalarRShiftSD<1, 0b11111, "fcvtzu">;
3658 defm SCVTF : SIMDScalarRShiftSD<0, 0b11100, "scvtf">;
3659 defm UCVTF : SIMDScalarRShiftSD<1, 0b11100, "ucvtf">;
3660 // Codegen patterns for the above. We don't put these directly on the
3661 // instructions because TableGen's type inference can't handle the truth.
3662 // Having the same base pattern for fp <--> int totally freaks it out.
3663 def : Pat<(int_arm64_neon_vcvtfp2fxs FPR32:$Rn, vecshiftR32:$imm),
3664 (FCVTZSs FPR32:$Rn, vecshiftR32:$imm)>;
3665 def : Pat<(int_arm64_neon_vcvtfp2fxu FPR32:$Rn, vecshiftR32:$imm),
3666 (FCVTZUs FPR32:$Rn, vecshiftR32:$imm)>;
3667 def : Pat<(i64 (int_arm64_neon_vcvtfp2fxs (f64 FPR64:$Rn), vecshiftR64:$imm)),
3668 (FCVTZSd FPR64:$Rn, vecshiftR64:$imm)>;
3669 def : Pat<(i64 (int_arm64_neon_vcvtfp2fxu (f64 FPR64:$Rn), vecshiftR64:$imm)),
3670 (FCVTZUd FPR64:$Rn, vecshiftR64:$imm)>;
3671 def : Pat<(v1i64 (int_arm64_neon_vcvtfp2fxs (v1f64 FPR64:$Rn),
3673 (FCVTZSd FPR64:$Rn, vecshiftR64:$imm)>;
3674 def : Pat<(v1i64 (int_arm64_neon_vcvtfp2fxu (v1f64 FPR64:$Rn),
3676 (FCVTZUd FPR64:$Rn, vecshiftR64:$imm)>;
3677 def : Pat<(int_arm64_neon_vcvtfxs2fp FPR32:$Rn, vecshiftR32:$imm),
3678 (SCVTFs FPR32:$Rn, vecshiftR32:$imm)>;
3679 def : Pat<(int_arm64_neon_vcvtfxu2fp FPR32:$Rn, vecshiftR32:$imm),
3680 (UCVTFs FPR32:$Rn, vecshiftR32:$imm)>;
3681 def : Pat<(f64 (int_arm64_neon_vcvtfxs2fp (i64 FPR64:$Rn), vecshiftR64:$imm)),
3682 (SCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
3683 def : Pat<(f64 (int_arm64_neon_vcvtfxu2fp (i64 FPR64:$Rn), vecshiftR64:$imm)),
3684 (UCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
3685 def : Pat<(v1f64 (int_arm64_neon_vcvtfxs2fp (v1i64 FPR64:$Rn),
3687 (SCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
3688 def : Pat<(v1f64 (int_arm64_neon_vcvtfxu2fp (v1i64 FPR64:$Rn),
3690 (UCVTFd FPR64:$Rn, vecshiftR64:$imm)>;
3692 defm SHL : SIMDScalarLShiftD< 0, 0b01010, "shl", ARM64vshl>;
3693 defm SLI : SIMDScalarLShiftDTied<1, 0b01010, "sli">;
3694 defm SQRSHRN : SIMDScalarRShiftBHS< 0, 0b10011, "sqrshrn",
3695 int_arm64_neon_sqrshrn>;
3696 defm SQRSHRUN : SIMDScalarRShiftBHS< 1, 0b10001, "sqrshrun",
3697 int_arm64_neon_sqrshrun>;
3698 defm SQSHLU : SIMDScalarLShiftBHSD<1, 0b01100, "sqshlu", ARM64sqshlui>;
3699 defm SQSHL : SIMDScalarLShiftBHSD<0, 0b01110, "sqshl", ARM64sqshli>;
3700 defm SQSHRN : SIMDScalarRShiftBHS< 0, 0b10010, "sqshrn",
3701 int_arm64_neon_sqshrn>;
3702 defm SQSHRUN : SIMDScalarRShiftBHS< 1, 0b10000, "sqshrun",
3703 int_arm64_neon_sqshrun>;
3704 defm SRI : SIMDScalarRShiftDTied< 1, 0b01000, "sri">;
3705 defm SRSHR : SIMDScalarRShiftD< 0, 0b00100, "srshr", ARM64srshri>;
3706 defm SRSRA : SIMDScalarRShiftDTied< 0, 0b00110, "srsra",
3707 TriOpFrag<(add node:$LHS,
3708 (ARM64srshri node:$MHS, node:$RHS))>>;
3709 defm SSHR : SIMDScalarRShiftD< 0, 0b00000, "sshr", ARM64vashr>;
3710 defm SSRA : SIMDScalarRShiftDTied< 0, 0b00010, "ssra",
3711 TriOpFrag<(add node:$LHS,
3712 (ARM64vashr node:$MHS, node:$RHS))>>;
3713 defm UQRSHRN : SIMDScalarRShiftBHS< 1, 0b10011, "uqrshrn",
3714 int_arm64_neon_uqrshrn>;
3715 defm UQSHL : SIMDScalarLShiftBHSD<1, 0b01110, "uqshl", ARM64uqshli>;
3716 defm UQSHRN : SIMDScalarRShiftBHS< 1, 0b10010, "uqshrn",
3717 int_arm64_neon_uqshrn>;
3718 defm URSHR : SIMDScalarRShiftD< 1, 0b00100, "urshr", ARM64urshri>;
3719 defm URSRA : SIMDScalarRShiftDTied< 1, 0b00110, "ursra",
3720 TriOpFrag<(add node:$LHS,
3721 (ARM64urshri node:$MHS, node:$RHS))>>;
3722 defm USHR : SIMDScalarRShiftD< 1, 0b00000, "ushr", ARM64vlshr>;
3723 defm USRA : SIMDScalarRShiftDTied< 1, 0b00010, "usra",
3724 TriOpFrag<(add node:$LHS,
3725 (ARM64vlshr node:$MHS, node:$RHS))>>;
3727 //----------------------------------------------------------------------------
3728 // AdvSIMD vector shift instructions
3729 //----------------------------------------------------------------------------
3730 defm FCVTZS:SIMDVectorRShiftSD<0, 0b11111, "fcvtzs", int_arm64_neon_vcvtfp2fxs>;
3731 defm FCVTZU:SIMDVectorRShiftSD<1, 0b11111, "fcvtzu", int_arm64_neon_vcvtfp2fxu>;
3732 defm SCVTF: SIMDVectorRShiftSDToFP<0, 0b11100, "scvtf",
3733 int_arm64_neon_vcvtfxs2fp>;
3734 defm RSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10001, "rshrn",
3735 int_arm64_neon_rshrn>;
3736 defm SHL : SIMDVectorLShiftBHSD<0, 0b01010, "shl", ARM64vshl>;
3737 defm SHRN : SIMDVectorRShiftNarrowBHS<0, 0b10000, "shrn",
3738 BinOpFrag<(trunc (ARM64vashr node:$LHS, node:$RHS))>>;
3739 defm SLI : SIMDVectorLShiftBHSDTied<1, 0b01010, "sli", int_arm64_neon_vsli>;
3740 def : Pat<(v1i64 (int_arm64_neon_vsli (v1i64 FPR64:$Rd), (v1i64 FPR64:$Rn),
3741 (i32 vecshiftL64:$imm))),
3742 (SLId FPR64:$Rd, FPR64:$Rn, vecshiftL64:$imm)>;
3743 defm SQRSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10011, "sqrshrn",
3744 int_arm64_neon_sqrshrn>;
3745 defm SQRSHRUN: SIMDVectorRShiftNarrowBHS<1, 0b10001, "sqrshrun",
3746 int_arm64_neon_sqrshrun>;
3747 defm SQSHLU : SIMDVectorLShiftBHSD<1, 0b01100, "sqshlu", ARM64sqshlui>;
3748 defm SQSHL : SIMDVectorLShiftBHSD<0, 0b01110, "sqshl", ARM64sqshli>;
3749 defm SQSHRN : SIMDVectorRShiftNarrowBHS<0, 0b10010, "sqshrn",
3750 int_arm64_neon_sqshrn>;
3751 defm SQSHRUN : SIMDVectorRShiftNarrowBHS<1, 0b10000, "sqshrun",
3752 int_arm64_neon_sqshrun>;
3753 defm SRI : SIMDVectorRShiftBHSDTied<1, 0b01000, "sri", int_arm64_neon_vsri>;
3754 def : Pat<(v1i64 (int_arm64_neon_vsri (v1i64 FPR64:$Rd), (v1i64 FPR64:$Rn),
3755 (i32 vecshiftR64:$imm))),
3756 (SRId FPR64:$Rd, FPR64:$Rn, vecshiftR64:$imm)>;
3757 defm SRSHR : SIMDVectorRShiftBHSD<0, 0b00100, "srshr", ARM64srshri>;
3758 defm SRSRA : SIMDVectorRShiftBHSDTied<0, 0b00110, "srsra",
3759 TriOpFrag<(add node:$LHS,
3760 (ARM64srshri node:$MHS, node:$RHS))> >;
3761 defm SSHLL : SIMDVectorLShiftLongBHSD<0, 0b10100, "sshll",
3762 BinOpFrag<(ARM64vshl (sext node:$LHS), node:$RHS)>>;
3764 defm SSHR : SIMDVectorRShiftBHSD<0, 0b00000, "sshr", ARM64vashr>;
3765 defm SSRA : SIMDVectorRShiftBHSDTied<0, 0b00010, "ssra",
3766 TriOpFrag<(add node:$LHS, (ARM64vashr node:$MHS, node:$RHS))>>;
3767 defm UCVTF : SIMDVectorRShiftSDToFP<1, 0b11100, "ucvtf",
3768 int_arm64_neon_vcvtfxu2fp>;
3769 defm UQRSHRN : SIMDVectorRShiftNarrowBHS<1, 0b10011, "uqrshrn",
3770 int_arm64_neon_uqrshrn>;
3771 defm UQSHL : SIMDVectorLShiftBHSD<1, 0b01110, "uqshl", ARM64uqshli>;
3772 defm UQSHRN : SIMDVectorRShiftNarrowBHS<1, 0b10010, "uqshrn",
3773 int_arm64_neon_uqshrn>;
3774 defm URSHR : SIMDVectorRShiftBHSD<1, 0b00100, "urshr", ARM64urshri>;
3775 defm URSRA : SIMDVectorRShiftBHSDTied<1, 0b00110, "ursra",
3776 TriOpFrag<(add node:$LHS,
3777 (ARM64urshri node:$MHS, node:$RHS))> >;
3778 defm USHLL : SIMDVectorLShiftLongBHSD<1, 0b10100, "ushll",
3779 BinOpFrag<(ARM64vshl (zext node:$LHS), node:$RHS)>>;
3780 defm USHR : SIMDVectorRShiftBHSD<1, 0b00000, "ushr", ARM64vlshr>;
3781 defm USRA : SIMDVectorRShiftBHSDTied<1, 0b00010, "usra",
3782 TriOpFrag<(add node:$LHS, (ARM64vlshr node:$MHS, node:$RHS))> >;
3784 // SHRN patterns for when a logical right shift was used instead of arithmetic
3785 // (the immediate guarantees no sign bits actually end up in the result so it
3787 def : Pat<(v8i8 (trunc (ARM64vlshr (v8i16 V128:$Rn), vecshiftR16Narrow:$imm))),
3788 (SHRNv8i8_shift V128:$Rn, vecshiftR16Narrow:$imm)>;
3789 def : Pat<(v4i16 (trunc (ARM64vlshr (v4i32 V128:$Rn), vecshiftR32Narrow:$imm))),
3790 (SHRNv4i16_shift V128:$Rn, vecshiftR32Narrow:$imm)>;
3791 def : Pat<(v2i32 (trunc (ARM64vlshr (v2i64 V128:$Rn), vecshiftR64Narrow:$imm))),
3792 (SHRNv2i32_shift V128:$Rn, vecshiftR64Narrow:$imm)>;
3794 def : Pat<(v16i8 (concat_vectors (v8i8 V64:$Rd),
3795 (trunc (ARM64vlshr (v8i16 V128:$Rn),
3796 vecshiftR16Narrow:$imm)))),
3797 (SHRNv16i8_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
3798 V128:$Rn, vecshiftR16Narrow:$imm)>;
3799 def : Pat<(v8i16 (concat_vectors (v4i16 V64:$Rd),
3800 (trunc (ARM64vlshr (v4i32 V128:$Rn),
3801 vecshiftR32Narrow:$imm)))),
3802 (SHRNv8i16_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
3803 V128:$Rn, vecshiftR32Narrow:$imm)>;
3804 def : Pat<(v4i32 (concat_vectors (v2i32 V64:$Rd),
3805 (trunc (ARM64vlshr (v2i64 V128:$Rn),
3806 vecshiftR64Narrow:$imm)))),
3807 (SHRNv4i32_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
3808 V128:$Rn, vecshiftR32Narrow:$imm)>;
3810 // Vector sign and zero extensions are implemented with SSHLL and USSHLL.
3811 // Anyexts are implemented as zexts.
3812 def : Pat<(v8i16 (sext (v8i8 V64:$Rn))), (SSHLLv8i8_shift V64:$Rn, (i32 0))>;
3813 def : Pat<(v8i16 (zext (v8i8 V64:$Rn))), (USHLLv8i8_shift V64:$Rn, (i32 0))>;
3814 def : Pat<(v8i16 (anyext (v8i8 V64:$Rn))), (USHLLv8i8_shift V64:$Rn, (i32 0))>;
3815 def : Pat<(v4i32 (sext (v4i16 V64:$Rn))), (SSHLLv4i16_shift V64:$Rn, (i32 0))>;
3816 def : Pat<(v4i32 (zext (v4i16 V64:$Rn))), (USHLLv4i16_shift V64:$Rn, (i32 0))>;
3817 def : Pat<(v4i32 (anyext (v4i16 V64:$Rn))), (USHLLv4i16_shift V64:$Rn, (i32 0))>;
3818 def : Pat<(v2i64 (sext (v2i32 V64:$Rn))), (SSHLLv2i32_shift V64:$Rn, (i32 0))>;
3819 def : Pat<(v2i64 (zext (v2i32 V64:$Rn))), (USHLLv2i32_shift V64:$Rn, (i32 0))>;
3820 def : Pat<(v2i64 (anyext (v2i32 V64:$Rn))), (USHLLv2i32_shift V64:$Rn, (i32 0))>;
3821 // Also match an extend from the upper half of a 128 bit source register.
3822 def : Pat<(v8i16 (anyext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
3823 (USHLLv16i8_shift V128:$Rn, (i32 0))>;
3824 def : Pat<(v8i16 (zext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
3825 (USHLLv16i8_shift V128:$Rn, (i32 0))>;
3826 def : Pat<(v8i16 (sext (v8i8 (extract_subvector V128:$Rn, (i64 8)) ))),
3827 (SSHLLv16i8_shift V128:$Rn, (i32 0))>;
3828 def : Pat<(v4i32 (anyext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
3829 (USHLLv8i16_shift V128:$Rn, (i32 0))>;
3830 def : Pat<(v4i32 (zext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
3831 (USHLLv8i16_shift V128:$Rn, (i32 0))>;
3832 def : Pat<(v4i32 (sext (v4i16 (extract_subvector V128:$Rn, (i64 4)) ))),
3833 (SSHLLv8i16_shift V128:$Rn, (i32 0))>;
3834 def : Pat<(v2i64 (anyext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
3835 (USHLLv4i32_shift V128:$Rn, (i32 0))>;
3836 def : Pat<(v2i64 (zext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
3837 (USHLLv4i32_shift V128:$Rn, (i32 0))>;
3838 def : Pat<(v2i64 (sext (v2i32 (extract_subvector V128:$Rn, (i64 2)) ))),
3839 (SSHLLv4i32_shift V128:$Rn, (i32 0))>;
3841 // Vector shift sxtl aliases
3842 def : InstAlias<"sxtl.8h $dst, $src1",
3843 (SSHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
3844 def : InstAlias<"sxtl $dst.8h, $src1.8b",
3845 (SSHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
3846 def : InstAlias<"sxtl.4s $dst, $src1",
3847 (SSHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
3848 def : InstAlias<"sxtl $dst.4s, $src1.4h",
3849 (SSHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
3850 def : InstAlias<"sxtl.2d $dst, $src1",
3851 (SSHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
3852 def : InstAlias<"sxtl $dst.2d, $src1.2s",
3853 (SSHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
3855 // Vector shift sxtl2 aliases
3856 def : InstAlias<"sxtl2.8h $dst, $src1",
3857 (SSHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
3858 def : InstAlias<"sxtl2 $dst.8h, $src1.16b",
3859 (SSHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
3860 def : InstAlias<"sxtl2.4s $dst, $src1",
3861 (SSHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
3862 def : InstAlias<"sxtl2 $dst.4s, $src1.8h",
3863 (SSHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
3864 def : InstAlias<"sxtl2.2d $dst, $src1",
3865 (SSHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
3866 def : InstAlias<"sxtl2 $dst.2d, $src1.4s",
3867 (SSHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
3869 // Vector shift uxtl aliases
3870 def : InstAlias<"uxtl.8h $dst, $src1",
3871 (USHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
3872 def : InstAlias<"uxtl $dst.8h, $src1.8b",
3873 (USHLLv8i8_shift V128:$dst, V64:$src1, 0)>;
3874 def : InstAlias<"uxtl.4s $dst, $src1",
3875 (USHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
3876 def : InstAlias<"uxtl $dst.4s, $src1.4h",
3877 (USHLLv4i16_shift V128:$dst, V64:$src1, 0)>;
3878 def : InstAlias<"uxtl.2d $dst, $src1",
3879 (USHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
3880 def : InstAlias<"uxtl $dst.2d, $src1.2s",
3881 (USHLLv2i32_shift V128:$dst, V64:$src1, 0)>;
3883 // Vector shift uxtl2 aliases
3884 def : InstAlias<"uxtl2.8h $dst, $src1",
3885 (USHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
3886 def : InstAlias<"uxtl2 $dst.8h, $src1.16b",
3887 (USHLLv16i8_shift V128:$dst, V128:$src1, 0)>;
3888 def : InstAlias<"uxtl2.4s $dst, $src1",
3889 (USHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
3890 def : InstAlias<"uxtl2 $dst.4s, $src1.8h",
3891 (USHLLv8i16_shift V128:$dst, V128:$src1, 0)>;
3892 def : InstAlias<"uxtl2.2d $dst, $src1",
3893 (USHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
3894 def : InstAlias<"uxtl2 $dst.2d, $src1.4s",
3895 (USHLLv4i32_shift V128:$dst, V128:$src1, 0)>;
3897 // If an integer is about to be converted to a floating point value,
3898 // just load it on the floating point unit.
3899 // These patterns are more complex because floating point loads do not
3900 // support sign extension.
3901 // The sign extension has to be explicitly added and is only supported for
3902 // one step: byte-to-half, half-to-word, word-to-doubleword.
3903 // SCVTF GPR -> FPR is 9 cycles.
3904 // SCVTF FPR -> FPR is 4 cyclces.
3905 // (sign extension with lengthen) SXTL FPR -> FPR is 2 cycles.
3906 // Therefore, we can do 2 sign extensions and one SCVTF FPR -> FPR
3907 // and still being faster.
3908 // However, this is not good for code size.
3909 // 8-bits -> float. 2 sizes step-up.
3910 def : Pat <(f32 (sint_to_fp (i32 (sextloadi8 ro_indexed8:$addr)))),
3911 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
3916 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3917 (LDRBro ro_indexed8:$addr),
3922 ssub)))>, Requires<[NotForCodeSize]>;
3923 def : Pat <(f32 (sint_to_fp (i32 (sextloadi8 am_indexed8:$addr)))),
3924 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
3929 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3930 (LDRBui am_indexed8:$addr),
3935 ssub)))>, Requires<[NotForCodeSize]>;
3936 def : Pat <(f32 (sint_to_fp (i32 (sextloadi8 am_unscaled8:$addr)))),
3937 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
3942 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3943 (LDURBi am_unscaled8:$addr),
3948 ssub)))>, Requires<[NotForCodeSize]>;
3949 // 16-bits -> float. 1 size step-up.
3950 def : Pat <(f32 (sint_to_fp (i32 (sextloadi16 ro_indexed16:$addr)))),
3951 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
3953 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3954 (LDRHro ro_indexed16:$addr),
3957 ssub)))>, Requires<[NotForCodeSize]>;
3958 def : Pat <(f32 (sint_to_fp (i32 (sextloadi16 am_indexed16:$addr)))),
3959 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
3961 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3962 (LDRHui am_indexed16:$addr),
3965 ssub)))>, Requires<[NotForCodeSize]>;
3966 def : Pat <(f32 (sint_to_fp (i32 (sextloadi16 am_unscaled16:$addr)))),
3967 (SCVTFv1i32 (f32 (EXTRACT_SUBREG
3969 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3970 (LDURHi am_unscaled16:$addr),
3973 ssub)))>, Requires<[NotForCodeSize]>;
3974 // 32-bits to 32-bits are handled in target specific dag combine:
3975 // performIntToFpCombine.
3976 // 64-bits integer to 32-bits floating point, not possible with
3977 // SCVTF on floating point registers (both source and destination
3978 // must have the same size).
3980 // Here are the patterns for 8, 16, 32, and 64-bits to double.
3981 // 8-bits -> double. 3 size step-up: give up.
3982 // 16-bits -> double. 2 size step.
3983 def : Pat <(f64 (sint_to_fp (i32 (sextloadi16 ro_indexed16:$addr)))),
3984 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
3989 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
3990 (LDRHro ro_indexed16:$addr),
3995 dsub)))>, Requires<[NotForCodeSize]>;
3996 def : Pat <(f64 (sint_to_fp (i32 (sextloadi16 am_indexed16:$addr)))),
3997 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4002 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4003 (LDRHui am_indexed16:$addr),
4008 dsub)))>, Requires<[NotForCodeSize]>;
4009 def : Pat <(f64 (sint_to_fp (i32 (sextloadi16 am_unscaled16:$addr)))),
4010 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4015 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4016 (LDURHi am_unscaled16:$addr),
4021 dsub)))>, Requires<[NotForCodeSize]>;
4022 // 32-bits -> double. 1 size step-up.
4023 def : Pat <(f64 (sint_to_fp (i32 (load ro_indexed32:$addr)))),
4024 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4026 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4027 (LDRSro ro_indexed32:$addr),
4030 dsub)))>, Requires<[NotForCodeSize]>;
4031 def : Pat <(f64 (sint_to_fp (i32 (load am_indexed32:$addr)))),
4032 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4034 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4035 (LDRSui am_indexed32:$addr),
4038 dsub)))>, Requires<[NotForCodeSize]>;
4039 def : Pat <(f64 (sint_to_fp (i32 (load am_unscaled32:$addr)))),
4040 (SCVTFv1i64 (f64 (EXTRACT_SUBREG
4042 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
4043 (LDURSi am_unscaled32:$addr),
4046 dsub)))>, Requires<[NotForCodeSize]>;
4047 // 64-bits -> double are handled in target specific dag combine:
4048 // performIntToFpCombine.
4051 //----------------------------------------------------------------------------
4052 // AdvSIMD Load-Store Structure
4053 //----------------------------------------------------------------------------
4054 defm LD1 : SIMDLd1Multiple<"ld1">;
4055 defm LD2 : SIMDLd2Multiple<"ld2">;
4056 defm LD3 : SIMDLd3Multiple<"ld3">;
4057 defm LD4 : SIMDLd4Multiple<"ld4">;
4059 defm ST1 : SIMDSt1Multiple<"st1">;
4060 defm ST2 : SIMDSt2Multiple<"st2">;
4061 defm ST3 : SIMDSt3Multiple<"st3">;
4062 defm ST4 : SIMDSt4Multiple<"st4">;
4064 class Ld1Pat<ValueType ty, Instruction INST>
4065 : Pat<(ty (load am_simdnoindex:$vaddr)), (INST am_simdnoindex:$vaddr)>;
4067 def : Ld1Pat<v16i8, LD1Onev16b>;
4068 def : Ld1Pat<v8i16, LD1Onev8h>;
4069 def : Ld1Pat<v4i32, LD1Onev4s>;
4070 def : Ld1Pat<v2i64, LD1Onev2d>;
4071 def : Ld1Pat<v8i8, LD1Onev8b>;
4072 def : Ld1Pat<v4i16, LD1Onev4h>;
4073 def : Ld1Pat<v2i32, LD1Onev2s>;
4074 def : Ld1Pat<v1i64, LD1Onev1d>;
4076 class St1Pat<ValueType ty, Instruction INST>
4077 : Pat<(store ty:$Vt, am_simdnoindex:$vaddr),
4078 (INST ty:$Vt, am_simdnoindex:$vaddr)>;
4080 def : St1Pat<v16i8, ST1Onev16b>;
4081 def : St1Pat<v8i16, ST1Onev8h>;
4082 def : St1Pat<v4i32, ST1Onev4s>;
4083 def : St1Pat<v2i64, ST1Onev2d>;
4084 def : St1Pat<v8i8, ST1Onev8b>;
4085 def : St1Pat<v4i16, ST1Onev4h>;
4086 def : St1Pat<v2i32, ST1Onev2s>;
4087 def : St1Pat<v1i64, ST1Onev1d>;
4093 defm LD1R : SIMDLdR<0, 0b110, 0, "ld1r", "One", 1, 2, 4, 8>;
4094 defm LD2R : SIMDLdR<1, 0b110, 0, "ld2r", "Two", 2, 4, 8, 16>;
4095 defm LD3R : SIMDLdR<0, 0b111, 0, "ld3r", "Three", 3, 6, 12, 24>;
4096 defm LD4R : SIMDLdR<1, 0b111, 0, "ld4r", "Four", 4, 8, 16, 32>;
4097 let mayLoad = 1, neverHasSideEffects = 1 in {
4098 defm LD1 : SIMDLdSingleBTied<0, 0b000, "ld1", VecListOneb, GPR64pi1>;
4099 defm LD1 : SIMDLdSingleHTied<0, 0b010, 0, "ld1", VecListOneh, GPR64pi2>;
4100 defm LD1 : SIMDLdSingleSTied<0, 0b100, 0b00, "ld1", VecListOnes, GPR64pi4>;
4101 defm LD1 : SIMDLdSingleDTied<0, 0b100, 0b01, "ld1", VecListOned, GPR64pi8>;
4102 defm LD2 : SIMDLdSingleBTied<1, 0b000, "ld2", VecListTwob, GPR64pi2>;
4103 defm LD2 : SIMDLdSingleHTied<1, 0b010, 0, "ld2", VecListTwoh, GPR64pi4>;
4104 defm LD2 : SIMDLdSingleSTied<1, 0b100, 0b00, "ld2", VecListTwos, GPR64pi8>;
4105 defm LD2 : SIMDLdSingleDTied<1, 0b100, 0b01, "ld2", VecListTwod, GPR64pi16>;
4106 defm LD3 : SIMDLdSingleBTied<0, 0b001, "ld3", VecListThreeb, GPR64pi3>;
4107 defm LD3 : SIMDLdSingleHTied<0, 0b011, 0, "ld3", VecListThreeh, GPR64pi6>;
4108 defm LD3 : SIMDLdSingleSTied<0, 0b101, 0b00, "ld3", VecListThrees, GPR64pi12>;
4109 defm LD3 : SIMDLdSingleDTied<0, 0b101, 0b01, "ld3", VecListThreed, GPR64pi24>;
4110 defm LD4 : SIMDLdSingleBTied<1, 0b001, "ld4", VecListFourb, GPR64pi4>;
4111 defm LD4 : SIMDLdSingleHTied<1, 0b011, 0, "ld4", VecListFourh, GPR64pi8>;
4112 defm LD4 : SIMDLdSingleSTied<1, 0b101, 0b00, "ld4", VecListFours, GPR64pi16>;
4113 defm LD4 : SIMDLdSingleDTied<1, 0b101, 0b01, "ld4", VecListFourd, GPR64pi32>;
4116 def : Pat<(v8i8 (ARM64dup (i32 (extloadi8 am_simdnoindex:$vaddr)))),
4117 (LD1Rv8b am_simdnoindex:$vaddr)>;
4118 def : Pat<(v16i8 (ARM64dup (i32 (extloadi8 am_simdnoindex:$vaddr)))),
4119 (LD1Rv16b am_simdnoindex:$vaddr)>;
4120 def : Pat<(v4i16 (ARM64dup (i32 (extloadi16 am_simdnoindex:$vaddr)))),
4121 (LD1Rv4h am_simdnoindex:$vaddr)>;
4122 def : Pat<(v8i16 (ARM64dup (i32 (extloadi16 am_simdnoindex:$vaddr)))),
4123 (LD1Rv8h am_simdnoindex:$vaddr)>;
4124 def : Pat<(v2i32 (ARM64dup (i32 (load am_simdnoindex:$vaddr)))),
4125 (LD1Rv2s am_simdnoindex:$vaddr)>;
4126 def : Pat<(v4i32 (ARM64dup (i32 (load am_simdnoindex:$vaddr)))),
4127 (LD1Rv4s am_simdnoindex:$vaddr)>;
4128 def : Pat<(v2i64 (ARM64dup (i64 (load am_simdnoindex:$vaddr)))),
4129 (LD1Rv2d am_simdnoindex:$vaddr)>;
4130 def : Pat<(v1i64 (ARM64dup (i64 (load am_simdnoindex:$vaddr)))),
4131 (LD1Rv1d am_simdnoindex:$vaddr)>;
4132 // Grab the floating point version too
4133 def : Pat<(v2f32 (ARM64dup (f32 (load am_simdnoindex:$vaddr)))),
4134 (LD1Rv2s am_simdnoindex:$vaddr)>;
4135 def : Pat<(v4f32 (ARM64dup (f32 (load am_simdnoindex:$vaddr)))),
4136 (LD1Rv4s am_simdnoindex:$vaddr)>;
4137 def : Pat<(v2f64 (ARM64dup (f64 (load am_simdnoindex:$vaddr)))),
4138 (LD1Rv2d am_simdnoindex:$vaddr)>;
4139 def : Pat<(v1f64 (ARM64dup (f64 (load am_simdnoindex:$vaddr)))),
4140 (LD1Rv1d am_simdnoindex:$vaddr)>;
4142 class Ld1Lane128Pat<SDPatternOperator scalar_load, Operand VecIndex,
4143 ValueType VTy, ValueType STy, Instruction LD1>
4144 : Pat<(vector_insert (VTy VecListOne128:$Rd),
4145 (STy (scalar_load am_simdnoindex:$vaddr)), VecIndex:$idx),
4146 (LD1 VecListOne128:$Rd, VecIndex:$idx, am_simdnoindex:$vaddr)>;
4148 def : Ld1Lane128Pat<extloadi8, VectorIndexB, v16i8, i32, LD1i8>;
4149 def : Ld1Lane128Pat<extloadi16, VectorIndexH, v8i16, i32, LD1i16>;
4150 def : Ld1Lane128Pat<load, VectorIndexS, v4i32, i32, LD1i32>;
4151 def : Ld1Lane128Pat<load, VectorIndexS, v4f32, f32, LD1i32>;
4152 def : Ld1Lane128Pat<load, VectorIndexD, v2i64, i64, LD1i64>;
4153 def : Ld1Lane128Pat<load, VectorIndexD, v2f64, f64, LD1i64>;
4155 class Ld1Lane64Pat<SDPatternOperator scalar_load, Operand VecIndex,
4156 ValueType VTy, ValueType STy, Instruction LD1>
4157 : Pat<(vector_insert (VTy VecListOne64:$Rd),
4158 (STy (scalar_load am_simdnoindex:$vaddr)), VecIndex:$idx),
4160 (LD1 (SUBREG_TO_REG (i32 0), VecListOne64:$Rd, dsub),
4161 VecIndex:$idx, am_simdnoindex:$vaddr),
4164 def : Ld1Lane64Pat<extloadi8, VectorIndexB, v8i8, i32, LD1i8>;
4165 def : Ld1Lane64Pat<extloadi16, VectorIndexH, v4i16, i32, LD1i16>;
4166 def : Ld1Lane64Pat<load, VectorIndexS, v2i32, i32, LD1i32>;
4167 def : Ld1Lane64Pat<load, VectorIndexS, v2f32, f32, LD1i32>;
4170 defm LD1 : SIMDLdSt1SingleAliases<"ld1">;
4171 defm LD2 : SIMDLdSt2SingleAliases<"ld2">;
4172 defm LD3 : SIMDLdSt3SingleAliases<"ld3">;
4173 defm LD4 : SIMDLdSt4SingleAliases<"ld4">;
4176 defm ST1 : SIMDStSingleB<0, 0b000, "st1", VecListOneb, GPR64pi1>;
4177 defm ST1 : SIMDStSingleH<0, 0b010, 0, "st1", VecListOneh, GPR64pi2>;
4178 defm ST1 : SIMDStSingleS<0, 0b100, 0b00, "st1", VecListOnes, GPR64pi4>;
4179 defm ST1 : SIMDStSingleD<0, 0b100, 0b01, "st1", VecListOned, GPR64pi8>;
4181 let AddedComplexity = 8 in
4182 class St1Lane128Pat<SDPatternOperator scalar_store, Operand VecIndex,
4183 ValueType VTy, ValueType STy, Instruction ST1>
4185 (STy (vector_extract (VTy VecListOne128:$Vt), VecIndex:$idx)),
4186 am_simdnoindex:$vaddr),
4187 (ST1 VecListOne128:$Vt, VecIndex:$idx, am_simdnoindex:$vaddr)>;
4189 def : St1Lane128Pat<truncstorei8, VectorIndexB, v16i8, i32, ST1i8>;
4190 def : St1Lane128Pat<truncstorei16, VectorIndexH, v8i16, i32, ST1i16>;
4191 def : St1Lane128Pat<store, VectorIndexS, v4i32, i32, ST1i32>;
4192 def : St1Lane128Pat<store, VectorIndexS, v4f32, f32, ST1i32>;
4193 def : St1Lane128Pat<store, VectorIndexD, v2i64, i64, ST1i64>;
4194 def : St1Lane128Pat<store, VectorIndexD, v2f64, f64, ST1i64>;
4196 let AddedComplexity = 8 in
4197 class St1Lane64Pat<SDPatternOperator scalar_store, Operand VecIndex,
4198 ValueType VTy, ValueType STy, Instruction ST1>
4200 (STy (vector_extract (VTy VecListOne64:$Vt), VecIndex:$idx)),
4201 am_simdnoindex:$vaddr),
4202 (ST1 (SUBREG_TO_REG (i32 0), VecListOne64:$Vt, dsub),
4203 VecIndex:$idx, am_simdnoindex:$vaddr)>;
4205 def : St1Lane64Pat<truncstorei8, VectorIndexB, v8i8, i32, ST1i8>;
4206 def : St1Lane64Pat<truncstorei16, VectorIndexH, v4i16, i32, ST1i16>;
4207 def : St1Lane64Pat<store, VectorIndexS, v2i32, i32, ST1i32>;
4208 def : St1Lane64Pat<store, VectorIndexS, v2f32, f32, ST1i32>;
4210 let mayStore = 1, neverHasSideEffects = 1 in {
4211 defm ST2 : SIMDStSingleB<1, 0b000, "st2", VecListTwob, GPR64pi2>;
4212 defm ST2 : SIMDStSingleH<1, 0b010, 0, "st2", VecListTwoh, GPR64pi4>;
4213 defm ST2 : SIMDStSingleS<1, 0b100, 0b00, "st2", VecListTwos, GPR64pi8>;
4214 defm ST2 : SIMDStSingleD<1, 0b100, 0b01, "st2", VecListTwod, GPR64pi16>;
4215 defm ST3 : SIMDStSingleB<0, 0b001, "st3", VecListThreeb, GPR64pi3>;
4216 defm ST3 : SIMDStSingleH<0, 0b011, 0, "st3", VecListThreeh, GPR64pi6>;
4217 defm ST3 : SIMDStSingleS<0, 0b101, 0b00, "st3", VecListThrees, GPR64pi12>;
4218 defm ST3 : SIMDStSingleD<0, 0b101, 0b01, "st3", VecListThreed, GPR64pi24>;
4219 defm ST4 : SIMDStSingleB<1, 0b001, "st4", VecListFourb, GPR64pi4>;
4220 defm ST4 : SIMDStSingleH<1, 0b011, 0, "st4", VecListFourh, GPR64pi8>;
4221 defm ST4 : SIMDStSingleS<1, 0b101, 0b00, "st4", VecListFours, GPR64pi16>;
4222 defm ST4 : SIMDStSingleD<1, 0b101, 0b01, "st4", VecListFourd, GPR64pi32>;
4225 defm ST1 : SIMDLdSt1SingleAliases<"st1">;
4226 defm ST2 : SIMDLdSt2SingleAliases<"st2">;
4227 defm ST3 : SIMDLdSt3SingleAliases<"st3">;
4228 defm ST4 : SIMDLdSt4SingleAliases<"st4">;
4230 //----------------------------------------------------------------------------
4231 // Crypto extensions
4232 //----------------------------------------------------------------------------
4234 def AESErr : AESTiedInst<0b0100, "aese", int_arm64_crypto_aese>;
4235 def AESDrr : AESTiedInst<0b0101, "aesd", int_arm64_crypto_aesd>;
4236 def AESMCrr : AESInst< 0b0110, "aesmc", int_arm64_crypto_aesmc>;
4237 def AESIMCrr : AESInst< 0b0111, "aesimc", int_arm64_crypto_aesimc>;
4239 def SHA1Crrr : SHATiedInstQSV<0b000, "sha1c", int_arm64_crypto_sha1c>;
4240 def SHA1Prrr : SHATiedInstQSV<0b001, "sha1p", int_arm64_crypto_sha1p>;
4241 def SHA1Mrrr : SHATiedInstQSV<0b010, "sha1m", int_arm64_crypto_sha1m>;
4242 def SHA1SU0rrr : SHATiedInstVVV<0b011, "sha1su0", int_arm64_crypto_sha1su0>;
4243 def SHA256Hrrr : SHATiedInstQQV<0b100, "sha256h", int_arm64_crypto_sha256h>;
4244 def SHA256H2rrr : SHATiedInstQQV<0b101, "sha256h2",int_arm64_crypto_sha256h2>;
4245 def SHA256SU1rrr :SHATiedInstVVV<0b110, "sha256su1",int_arm64_crypto_sha256su1>;
4247 def SHA1Hrr : SHAInstSS< 0b0000, "sha1h", int_arm64_crypto_sha1h>;
4248 def SHA1SU1rr : SHATiedInstVV<0b0001, "sha1su1", int_arm64_crypto_sha1su1>;
4249 def SHA256SU0rr : SHATiedInstVV<0b0010, "sha256su0",int_arm64_crypto_sha256su0>;
4251 //----------------------------------------------------------------------------
4253 //----------------------------------------------------------------------------
4254 // FIXME: Like for X86, these should go in their own separate .td file.
4256 // Any instruction that defines a 32-bit result leaves the high half of the
4257 // register. Truncate can be lowered to EXTRACT_SUBREG. CopyFromReg may
4258 // be copying from a truncate. But any other 32-bit operation will zero-extend
4260 // FIXME: X86 also checks for CMOV here. Do we need something similar?
4261 def def32 : PatLeaf<(i32 GPR32:$src), [{
4262 return N->getOpcode() != ISD::TRUNCATE &&
4263 N->getOpcode() != TargetOpcode::EXTRACT_SUBREG &&
4264 N->getOpcode() != ISD::CopyFromReg;
4267 // In the case of a 32-bit def that is known to implicitly zero-extend,
4268 // we can use a SUBREG_TO_REG.
4269 def : Pat<(i64 (zext def32:$src)), (SUBREG_TO_REG (i64 0), GPR32:$src, sub_32)>;
4271 // For an anyext, we don't care what the high bits are, so we can perform an
4272 // INSERT_SUBREF into an IMPLICIT_DEF.
4273 def : Pat<(i64 (anyext GPR32:$src)),
4274 (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32)>;
4276 // When we need to explicitly zero-extend, we use an unsigned bitfield move
4277 // instruction (UBFM) on the enclosing super-reg.
4278 def : Pat<(i64 (zext GPR32:$src)),
4279 (UBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32), 0, 31)>;
4281 // To sign extend, we use a signed bitfield move instruction (SBFM) on the
4282 // containing super-reg.
4283 def : Pat<(i64 (sext GPR32:$src)),
4284 (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32), 0, 31)>;
4285 def : Pat<(i64 (sext_inreg GPR64:$src, i32)), (SBFMXri GPR64:$src, 0, 31)>;
4286 def : Pat<(i64 (sext_inreg GPR64:$src, i16)), (SBFMXri GPR64:$src, 0, 15)>;
4287 def : Pat<(i64 (sext_inreg GPR64:$src, i8)), (SBFMXri GPR64:$src, 0, 7)>;
4288 def : Pat<(i64 (sext_inreg GPR64:$src, i1)), (SBFMXri GPR64:$src, 0, 0)>;
4289 def : Pat<(i32 (sext_inreg GPR32:$src, i16)), (SBFMWri GPR32:$src, 0, 15)>;
4290 def : Pat<(i32 (sext_inreg GPR32:$src, i8)), (SBFMWri GPR32:$src, 0, 7)>;
4291 def : Pat<(i32 (sext_inreg GPR32:$src, i1)), (SBFMWri GPR32:$src, 0, 0)>;
4293 def : Pat<(shl (sext_inreg GPR32:$Rn, i8), (i64 imm0_31:$imm)),
4294 (SBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
4295 (i64 (i32shift_sext_i8 imm0_31:$imm)))>;
4296 def : Pat<(shl (sext_inreg GPR64:$Rn, i8), (i64 imm0_63:$imm)),
4297 (SBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
4298 (i64 (i64shift_sext_i8 imm0_63:$imm)))>;
4300 def : Pat<(shl (sext_inreg GPR32:$Rn, i16), (i64 imm0_31:$imm)),
4301 (SBFMWri GPR32:$Rn, (i64 (i32shift_a imm0_31:$imm)),
4302 (i64 (i32shift_sext_i16 imm0_31:$imm)))>;
4303 def : Pat<(shl (sext_inreg GPR64:$Rn, i16), (i64 imm0_63:$imm)),
4304 (SBFMXri GPR64:$Rn, (i64 (i64shift_a imm0_63:$imm)),
4305 (i64 (i64shift_sext_i16 imm0_63:$imm)))>;
4307 def : Pat<(shl (i64 (sext GPR32:$Rn)), (i64 imm0_63:$imm)),
4308 (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$Rn, sub_32),
4309 (i64 (i64shift_a imm0_63:$imm)),
4310 (i64 (i64shift_sext_i32 imm0_63:$imm)))>;
4312 // sra patterns have an AddedComplexity of 10, so make sure we have a higher
4313 // AddedComplexity for the following patterns since we want to match sext + sra
4314 // patterns before we attempt to match a single sra node.
4315 let AddedComplexity = 20 in {
4316 // We support all sext + sra combinations which preserve at least one bit of the
4317 // original value which is to be sign extended. E.g. we support shifts up to
4319 def : Pat<(sra (sext_inreg GPR32:$Rn, i8), (i64 imm0_7:$imm)),
4320 (SBFMWri GPR32:$Rn, (i64 imm0_7:$imm), 7)>;
4321 def : Pat<(sra (sext_inreg GPR64:$Rn, i8), (i64 imm0_7:$imm)),
4322 (SBFMXri GPR64:$Rn, (i64 imm0_7:$imm), 7)>;
4324 def : Pat<(sra (sext_inreg GPR32:$Rn, i16), (i64 imm0_15:$imm)),
4325 (SBFMWri GPR32:$Rn, (i64 imm0_15:$imm), 15)>;
4326 def : Pat<(sra (sext_inreg GPR64:$Rn, i16), (i64 imm0_15:$imm)),
4327 (SBFMXri GPR64:$Rn, (i64 imm0_15:$imm), 15)>;
4329 def : Pat<(sra (i64 (sext GPR32:$Rn)), (i64 imm0_31:$imm)),
4330 (SBFMXri (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$Rn, sub_32),
4331 (i64 imm0_31:$imm), 31)>;
4332 } // AddedComplexity = 20
4334 // To truncate, we can simply extract from a subregister.
4335 def : Pat<(i32 (trunc GPR64sp:$src)),
4336 (i32 (EXTRACT_SUBREG GPR64sp:$src, sub_32))>;
4338 // __builtin_trap() uses the BRK instruction on ARM64.
4339 def : Pat<(trap), (BRK 1)>;
4341 // Conversions within AdvSIMD types in the same register size are free.
4343 def : Pat<(v1i64 (bitconvert (v2i32 FPR64:$src))), (v1i64 FPR64:$src)>;
4344 def : Pat<(v1i64 (bitconvert (v4i16 FPR64:$src))), (v1i64 FPR64:$src)>;
4345 def : Pat<(v1i64 (bitconvert (v8i8 FPR64:$src))), (v1i64 FPR64:$src)>;
4346 def : Pat<(v1i64 (bitconvert (f64 FPR64:$src))), (v1i64 FPR64:$src)>;
4347 def : Pat<(v1i64 (bitconvert (v2f32 FPR64:$src))), (v1i64 FPR64:$src)>;
4348 def : Pat<(v1i64 (bitconvert (v1f64 FPR64:$src))), (v1i64 FPR64:$src)>;
4350 def : Pat<(v2i32 (bitconvert (v1i64 FPR64:$src))), (v2i32 FPR64:$src)>;
4351 def : Pat<(v2i32 (bitconvert (v4i16 FPR64:$src))), (v2i32 FPR64:$src)>;
4352 def : Pat<(v2i32 (bitconvert (v8i8 FPR64:$src))), (v2i32 FPR64:$src)>;
4353 def : Pat<(v2i32 (bitconvert (f64 FPR64:$src))), (v2i32 FPR64:$src)>;
4354 def : Pat<(v2i32 (bitconvert (v2f32 FPR64:$src))), (v2i32 FPR64:$src)>;
4355 def : Pat<(v2i32 (bitconvert (v1f64 FPR64:$src))), (v2i32 FPR64:$src)>;
4357 def : Pat<(v4i16 (bitconvert (v1i64 FPR64:$src))), (v4i16 FPR64:$src)>;
4358 def : Pat<(v4i16 (bitconvert (v2i32 FPR64:$src))), (v4i16 FPR64:$src)>;
4359 def : Pat<(v4i16 (bitconvert (v8i8 FPR64:$src))), (v4i16 FPR64:$src)>;
4360 def : Pat<(v4i16 (bitconvert (f64 FPR64:$src))), (v4i16 FPR64:$src)>;
4361 def : Pat<(v4i16 (bitconvert (v2f32 FPR64:$src))), (v4i16 FPR64:$src)>;
4362 def : Pat<(v4i16 (bitconvert (v1f64 FPR64:$src))), (v4i16 FPR64:$src)>;
4364 def : Pat<(v8i8 (bitconvert (v1i64 FPR64:$src))), (v8i8 FPR64:$src)>;
4365 def : Pat<(v8i8 (bitconvert (v2i32 FPR64:$src))), (v8i8 FPR64:$src)>;
4366 def : Pat<(v8i8 (bitconvert (v4i16 FPR64:$src))), (v8i8 FPR64:$src)>;
4367 def : Pat<(v8i8 (bitconvert (f64 FPR64:$src))), (v8i8 FPR64:$src)>;
4368 def : Pat<(v8i8 (bitconvert (v2f32 FPR64:$src))), (v8i8 FPR64:$src)>;
4369 def : Pat<(v8i8 (bitconvert (v1f64 FPR64:$src))), (v8i8 FPR64:$src)>;
4371 def : Pat<(f64 (bitconvert (v1i64 FPR64:$src))), (f64 FPR64:$src)>;
4372 def : Pat<(f64 (bitconvert (v2i32 FPR64:$src))), (f64 FPR64:$src)>;
4373 def : Pat<(f64 (bitconvert (v4i16 FPR64:$src))), (f64 FPR64:$src)>;
4374 def : Pat<(f64 (bitconvert (v8i8 FPR64:$src))), (f64 FPR64:$src)>;
4375 def : Pat<(f64 (bitconvert (v2f32 FPR64:$src))), (f64 FPR64:$src)>;
4376 def : Pat<(f64 (bitconvert (v1f64 FPR64:$src))), (f64 FPR64:$src)>;
4378 def : Pat<(v1f64 (bitconvert (v1i64 FPR64:$src))), (v1f64 FPR64:$src)>;
4379 def : Pat<(v1f64 (bitconvert (v2i32 FPR64:$src))), (v1f64 FPR64:$src)>;
4380 def : Pat<(v1f64 (bitconvert (v4i16 FPR64:$src))), (v1f64 FPR64:$src)>;
4381 def : Pat<(v1f64 (bitconvert (v8i8 FPR64:$src))), (v1f64 FPR64:$src)>;
4382 def : Pat<(v1f64 (bitconvert (f64 FPR64:$src))), (v1f64 FPR64:$src)>;
4383 def : Pat<(v1f64 (bitconvert (v2f32 FPR64:$src))), (v1f64 FPR64:$src)>;
4385 def : Pat<(v2f32 (bitconvert (f64 FPR64:$src))), (v2f32 FPR64:$src)>;
4386 def : Pat<(v2f32 (bitconvert (v1i64 FPR64:$src))), (v2f32 FPR64:$src)>;
4387 def : Pat<(v2f32 (bitconvert (v2i32 FPR64:$src))), (v2f32 FPR64:$src)>;
4388 def : Pat<(v2f32 (bitconvert (v4i16 FPR64:$src))), (v2f32 FPR64:$src)>;
4389 def : Pat<(v2f32 (bitconvert (v8i8 FPR64:$src))), (v2f32 FPR64:$src)>;
4390 def : Pat<(v2f32 (bitconvert (v1f64 FPR64:$src))), (v2f32 FPR64:$src)>;
4393 def : Pat<(f128 (bitconvert (v2i64 FPR128:$src))), (f128 FPR128:$src)>;
4394 def : Pat<(f128 (bitconvert (v4i32 FPR128:$src))), (f128 FPR128:$src)>;
4395 def : Pat<(f128 (bitconvert (v8i16 FPR128:$src))), (f128 FPR128:$src)>;
4396 def : Pat<(f128 (bitconvert (v2f64 FPR128:$src))), (f128 FPR128:$src)>;
4397 def : Pat<(f128 (bitconvert (v4f32 FPR128:$src))), (f128 FPR128:$src)>;
4399 def : Pat<(v2f64 (bitconvert (f128 FPR128:$src))), (v2f64 FPR128:$src)>;
4400 def : Pat<(v2f64 (bitconvert (v4i32 FPR128:$src))), (v2f64 FPR128:$src)>;
4401 def : Pat<(v2f64 (bitconvert (v8i16 FPR128:$src))), (v2f64 FPR128:$src)>;
4402 def : Pat<(v2f64 (bitconvert (v16i8 FPR128:$src))), (v2f64 FPR128:$src)>;
4403 def : Pat<(v2f64 (bitconvert (v2i64 FPR128:$src))), (v2f64 FPR128:$src)>;
4404 def : Pat<(v2f64 (bitconvert (v4f32 FPR128:$src))), (v2f64 FPR128:$src)>;
4406 def : Pat<(v4f32 (bitconvert (f128 FPR128:$src))), (v4f32 FPR128:$src)>;
4407 def : Pat<(v4f32 (bitconvert (v4i32 FPR128:$src))), (v4f32 FPR128:$src)>;
4408 def : Pat<(v4f32 (bitconvert (v8i16 FPR128:$src))), (v4f32 FPR128:$src)>;
4409 def : Pat<(v4f32 (bitconvert (v16i8 FPR128:$src))), (v4f32 FPR128:$src)>;
4410 def : Pat<(v4f32 (bitconvert (v2i64 FPR128:$src))), (v4f32 FPR128:$src)>;
4411 def : Pat<(v4f32 (bitconvert (v2f64 FPR128:$src))), (v4f32 FPR128:$src)>;
4413 def : Pat<(v2i64 (bitconvert (f128 FPR128:$src))), (v2i64 FPR128:$src)>;
4414 def : Pat<(v2i64 (bitconvert (v4i32 FPR128:$src))), (v2i64 FPR128:$src)>;
4415 def : Pat<(v2i64 (bitconvert (v8i16 FPR128:$src))), (v2i64 FPR128:$src)>;
4416 def : Pat<(v2i64 (bitconvert (v16i8 FPR128:$src))), (v2i64 FPR128:$src)>;
4417 def : Pat<(v2i64 (bitconvert (v2f64 FPR128:$src))), (v2i64 FPR128:$src)>;
4418 def : Pat<(v2i64 (bitconvert (v4f32 FPR128:$src))), (v2i64 FPR128:$src)>;
4420 def : Pat<(v4i32 (bitconvert (f128 FPR128:$src))), (v4i32 FPR128:$src)>;
4421 def : Pat<(v4i32 (bitconvert (v2i64 FPR128:$src))), (v4i32 FPR128:$src)>;
4422 def : Pat<(v4i32 (bitconvert (v8i16 FPR128:$src))), (v4i32 FPR128:$src)>;
4423 def : Pat<(v4i32 (bitconvert (v16i8 FPR128:$src))), (v4i32 FPR128:$src)>;
4424 def : Pat<(v4i32 (bitconvert (v2f64 FPR128:$src))), (v4i32 FPR128:$src)>;
4425 def : Pat<(v4i32 (bitconvert (v4f32 FPR128:$src))), (v4i32 FPR128:$src)>;
4427 def : Pat<(v8i16 (bitconvert (f128 FPR128:$src))), (v8i16 FPR128:$src)>;
4428 def : Pat<(v8i16 (bitconvert (v2i64 FPR128:$src))), (v8i16 FPR128:$src)>;
4429 def : Pat<(v8i16 (bitconvert (v4i32 FPR128:$src))), (v8i16 FPR128:$src)>;
4430 def : Pat<(v8i16 (bitconvert (v16i8 FPR128:$src))), (v8i16 FPR128:$src)>;
4431 def : Pat<(v8i16 (bitconvert (v2f64 FPR128:$src))), (v8i16 FPR128:$src)>;
4432 def : Pat<(v8i16 (bitconvert (v4f32 FPR128:$src))), (v8i16 FPR128:$src)>;
4434 def : Pat<(v16i8 (bitconvert (f128 FPR128:$src))), (v16i8 FPR128:$src)>;
4435 def : Pat<(v16i8 (bitconvert (v2i64 FPR128:$src))), (v16i8 FPR128:$src)>;
4436 def : Pat<(v16i8 (bitconvert (v4i32 FPR128:$src))), (v16i8 FPR128:$src)>;
4437 def : Pat<(v16i8 (bitconvert (v8i16 FPR128:$src))), (v16i8 FPR128:$src)>;
4438 def : Pat<(v16i8 (bitconvert (v2f64 FPR128:$src))), (v16i8 FPR128:$src)>;
4439 def : Pat<(v16i8 (bitconvert (v4f32 FPR128:$src))), (v16i8 FPR128:$src)>;
4441 def : Pat<(v8i8 (extract_subvector (v16i8 FPR128:$Rn), (i64 1))),
4442 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
4443 def : Pat<(v4i16 (extract_subvector (v8i16 FPR128:$Rn), (i64 1))),
4444 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
4445 def : Pat<(v2i32 (extract_subvector (v4i32 FPR128:$Rn), (i64 1))),
4446 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
4447 def : Pat<(v1i64 (extract_subvector (v2i64 FPR128:$Rn), (i64 1))),
4448 (EXTRACT_SUBREG (DUPv2i64lane FPR128:$Rn, 1), dsub)>;
4450 // A 64-bit subvector insert to the first 128-bit vector position
4451 // is a subregister copy that needs no instruction.
4452 def : Pat<(insert_subvector undef, (v1i64 FPR64:$src), (i32 0)),
4453 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
4454 def : Pat<(insert_subvector undef, (v1f64 FPR64:$src), (i32 0)),
4455 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
4456 def : Pat<(insert_subvector undef, (v2i32 FPR64:$src), (i32 0)),
4457 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
4458 def : Pat<(insert_subvector undef, (v2f32 FPR64:$src), (i32 0)),
4459 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
4460 def : Pat<(insert_subvector undef, (v4i16 FPR64:$src), (i32 0)),
4461 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
4462 def : Pat<(insert_subvector undef, (v8i8 FPR64:$src), (i32 0)),
4463 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)), FPR64:$src, dsub)>;
4465 // Use pair-wise add instructions when summing up the lanes for v2f64, v2i64
4467 def : Pat<(i64 (add (vector_extract (v2i64 FPR128:$Rn), (i64 0)),
4468 (vector_extract (v2i64 FPR128:$Rn), (i64 1)))),
4469 (i64 (ADDPv2i64p (v2i64 FPR128:$Rn)))>;
4470 def : Pat<(f64 (fadd (vector_extract (v2f64 FPR128:$Rn), (i64 0)),
4471 (vector_extract (v2f64 FPR128:$Rn), (i64 1)))),
4472 (f64 (FADDPv2i64p (v2f64 FPR128:$Rn)))>;
4473 // vector_extract on 64-bit vectors gets promoted to a 128 bit vector,
4474 // so we match on v4f32 here, not v2f32. This will also catch adding
4475 // the low two lanes of a true v4f32 vector.
4476 def : Pat<(fadd (vector_extract (v4f32 FPR128:$Rn), (i64 0)),
4477 (vector_extract (v4f32 FPR128:$Rn), (i64 1))),
4478 (f32 (FADDPv2i32p (EXTRACT_SUBREG FPR128:$Rn, dsub)))>;
4480 // Scalar 64-bit shifts in FPR64 registers.
4481 def : Pat<(i64 (int_arm64_neon_sshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
4482 (SSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
4483 def : Pat<(i64 (int_arm64_neon_ushl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
4484 (USHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
4485 def : Pat<(i64 (int_arm64_neon_srshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
4486 (SRSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
4487 def : Pat<(i64 (int_arm64_neon_urshl (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
4488 (URSHLv1i64 FPR64:$Rn, FPR64:$Rm)>;
4490 // Tail call return handling. These are all compiler pseudo-instructions,
4491 // so no encoding information or anything like that.
4492 let isCall = 1, isTerminator = 1, isReturn = 1, isBarrier = 1, Uses = [SP] in {
4493 def TCRETURNdi : Pseudo<(outs), (ins i64imm:$dst), []>;
4494 def TCRETURNri : Pseudo<(outs), (ins tcGPR64:$dst), []>;
4497 def : Pat<(ARM64tcret tcGPR64:$dst), (TCRETURNri tcGPR64:$dst)>;
4498 def : Pat<(ARM64tcret (i64 tglobaladdr:$dst)), (TCRETURNdi texternalsym:$dst)>;
4499 def : Pat<(ARM64tcret (i64 texternalsym:$dst)), (TCRETURNdi texternalsym:$dst)>;
4501 include "ARM64InstrAtomics.td"