1 //===-- ARMInstrNEON.td - NEON support for ARM -------------*- tablegen -*-===//
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
7 //===----------------------------------------------------------------------===//
9 // This file describes the ARM NEON instruction set.
11 //===----------------------------------------------------------------------===//
14 //===----------------------------------------------------------------------===//
15 // NEON-specific Operands.
16 //===----------------------------------------------------------------------===//
17 def nModImm : Operand<i32> {
18 let PrintMethod = "printNEONModImmOperand";
21 def nImmSplatI8AsmOperand : AsmOperandClass { let Name = "NEONi8splat"; }
22 def nImmSplatI8 : Operand<i32> {
23 let PrintMethod = "printNEONModImmOperand";
24 let ParserMatchClass = nImmSplatI8AsmOperand;
26 def nImmSplatI16AsmOperand : AsmOperandClass { let Name = "NEONi16splat"; }
27 def nImmSplatI16 : Operand<i32> {
28 let PrintMethod = "printNEONModImmOperand";
29 let ParserMatchClass = nImmSplatI16AsmOperand;
31 def nImmSplatI32AsmOperand : AsmOperandClass { let Name = "NEONi32splat"; }
32 def nImmSplatI32 : Operand<i32> {
33 let PrintMethod = "printNEONModImmOperand";
34 let ParserMatchClass = nImmSplatI32AsmOperand;
36 def nImmSplatNotI16AsmOperand : AsmOperandClass { let Name = "NEONi16splatNot"; }
37 def nImmSplatNotI16 : Operand<i32> {
38 let ParserMatchClass = nImmSplatNotI16AsmOperand;
40 def nImmSplatNotI32AsmOperand : AsmOperandClass { let Name = "NEONi32splatNot"; }
41 def nImmSplatNotI32 : Operand<i32> {
42 let ParserMatchClass = nImmSplatNotI32AsmOperand;
44 def nImmVMOVI32AsmOperand : AsmOperandClass { let Name = "NEONi32vmov"; }
45 def nImmVMOVI32 : Operand<i32> {
46 let PrintMethod = "printNEONModImmOperand";
47 let ParserMatchClass = nImmVMOVI32AsmOperand;
50 class nImmVMOVIAsmOperandReplicate<ValueType From, ValueType To>
52 let Name = "NEONi" # To.Size # "vmovi" # From.Size # "Replicate";
53 let PredicateMethod = "isNEONmovReplicate<" # From.Size # ", " # To.Size # ">";
54 let RenderMethod = "addNEONvmovi" # From.Size # "ReplicateOperands";
57 class nImmVINVIAsmOperandReplicate<ValueType From, ValueType To>
59 let Name = "NEONi" # To.Size # "invi" # From.Size # "Replicate";
60 let PredicateMethod = "isNEONinvReplicate<" # From.Size # ", " # To.Size # ">";
61 let RenderMethod = "addNEONinvi" # From.Size # "ReplicateOperands";
64 class nImmVMOVIReplicate<ValueType From, ValueType To> : Operand<i32> {
65 let PrintMethod = "printNEONModImmOperand";
66 let ParserMatchClass = nImmVMOVIAsmOperandReplicate<From, To>;
69 class nImmVINVIReplicate<ValueType From, ValueType To> : Operand<i32> {
70 let PrintMethod = "printNEONModImmOperand";
71 let ParserMatchClass = nImmVINVIAsmOperandReplicate<From, To>;
74 def nImmVMOVI32NegAsmOperand : AsmOperandClass { let Name = "NEONi32vmovNeg"; }
75 def nImmVMOVI32Neg : Operand<i32> {
76 let PrintMethod = "printNEONModImmOperand";
77 let ParserMatchClass = nImmVMOVI32NegAsmOperand;
79 def nImmVMOVF32 : Operand<i32> {
80 let PrintMethod = "printFPImmOperand";
81 let ParserMatchClass = FPImmOperand;
83 def nImmSplatI64AsmOperand : AsmOperandClass { let Name = "NEONi64splat"; }
84 def nImmSplatI64 : Operand<i32> {
85 let PrintMethod = "printNEONModImmOperand";
86 let ParserMatchClass = nImmSplatI64AsmOperand;
89 def VectorIndex8Operand : AsmOperandClass { let Name = "VectorIndex8"; }
90 def VectorIndex16Operand : AsmOperandClass { let Name = "VectorIndex16"; }
91 def VectorIndex32Operand : AsmOperandClass { let Name = "VectorIndex32"; }
92 def VectorIndex64Operand : AsmOperandClass { let Name = "VectorIndex64"; }
93 def VectorIndex8 : Operand<i32>, ImmLeaf<i32, [{
94 return ((uint64_t)Imm) < 8;
96 let ParserMatchClass = VectorIndex8Operand;
97 let PrintMethod = "printVectorIndex";
98 let MIOperandInfo = (ops i32imm);
100 def VectorIndex16 : Operand<i32>, ImmLeaf<i32, [{
101 return ((uint64_t)Imm) < 4;
103 let ParserMatchClass = VectorIndex16Operand;
104 let PrintMethod = "printVectorIndex";
105 let MIOperandInfo = (ops i32imm);
107 def VectorIndex32 : Operand<i32>, ImmLeaf<i32, [{
108 return ((uint64_t)Imm) < 2;
110 let ParserMatchClass = VectorIndex32Operand;
111 let PrintMethod = "printVectorIndex";
112 let MIOperandInfo = (ops i32imm);
114 def VectorIndex64 : Operand<i32>, ImmLeaf<i32, [{
115 return ((uint64_t)Imm) < 1;
117 let ParserMatchClass = VectorIndex64Operand;
118 let PrintMethod = "printVectorIndex";
119 let MIOperandInfo = (ops i32imm);
122 // Register list of one D register.
123 def VecListOneDAsmOperand : AsmOperandClass {
124 let Name = "VecListOneD";
125 let ParserMethod = "parseVectorList";
126 let RenderMethod = "addVecListOperands";
128 def VecListOneD : RegisterOperand<DPR, "printVectorListOne"> {
129 let ParserMatchClass = VecListOneDAsmOperand;
131 // Register list of two sequential D registers.
132 def VecListDPairAsmOperand : AsmOperandClass {
133 let Name = "VecListDPair";
134 let ParserMethod = "parseVectorList";
135 let RenderMethod = "addVecListOperands";
137 def VecListDPair : RegisterOperand<DPair, "printVectorListTwo"> {
138 let ParserMatchClass = VecListDPairAsmOperand;
140 // Register list of three sequential D registers.
141 def VecListThreeDAsmOperand : AsmOperandClass {
142 let Name = "VecListThreeD";
143 let ParserMethod = "parseVectorList";
144 let RenderMethod = "addVecListOperands";
146 def VecListThreeD : RegisterOperand<DPR, "printVectorListThree"> {
147 let ParserMatchClass = VecListThreeDAsmOperand;
149 // Register list of four sequential D registers.
150 def VecListFourDAsmOperand : AsmOperandClass {
151 let Name = "VecListFourD";
152 let ParserMethod = "parseVectorList";
153 let RenderMethod = "addVecListOperands";
155 def VecListFourD : RegisterOperand<DPR, "printVectorListFour"> {
156 let ParserMatchClass = VecListFourDAsmOperand;
158 // Register list of two D registers spaced by 2 (two sequential Q registers).
159 def VecListDPairSpacedAsmOperand : AsmOperandClass {
160 let Name = "VecListDPairSpaced";
161 let ParserMethod = "parseVectorList";
162 let RenderMethod = "addVecListOperands";
164 def VecListDPairSpaced : RegisterOperand<DPair, "printVectorListTwoSpaced"> {
165 let ParserMatchClass = VecListDPairSpacedAsmOperand;
167 // Register list of three D registers spaced by 2 (three Q registers).
168 def VecListThreeQAsmOperand : AsmOperandClass {
169 let Name = "VecListThreeQ";
170 let ParserMethod = "parseVectorList";
171 let RenderMethod = "addVecListOperands";
173 def VecListThreeQ : RegisterOperand<DPR, "printVectorListThreeSpaced"> {
174 let ParserMatchClass = VecListThreeQAsmOperand;
176 // Register list of three D registers spaced by 2 (three Q registers).
177 def VecListFourQAsmOperand : AsmOperandClass {
178 let Name = "VecListFourQ";
179 let ParserMethod = "parseVectorList";
180 let RenderMethod = "addVecListOperands";
182 def VecListFourQ : RegisterOperand<DPR, "printVectorListFourSpaced"> {
183 let ParserMatchClass = VecListFourQAsmOperand;
186 // Register list of one D register, with "all lanes" subscripting.
187 def VecListOneDAllLanesAsmOperand : AsmOperandClass {
188 let Name = "VecListOneDAllLanes";
189 let ParserMethod = "parseVectorList";
190 let RenderMethod = "addVecListOperands";
192 def VecListOneDAllLanes : RegisterOperand<DPR, "printVectorListOneAllLanes"> {
193 let ParserMatchClass = VecListOneDAllLanesAsmOperand;
195 // Register list of two D registers, with "all lanes" subscripting.
196 def VecListDPairAllLanesAsmOperand : AsmOperandClass {
197 let Name = "VecListDPairAllLanes";
198 let ParserMethod = "parseVectorList";
199 let RenderMethod = "addVecListOperands";
201 def VecListDPairAllLanes : RegisterOperand<DPair,
202 "printVectorListTwoAllLanes"> {
203 let ParserMatchClass = VecListDPairAllLanesAsmOperand;
205 // Register list of two D registers spaced by 2 (two sequential Q registers).
206 def VecListDPairSpacedAllLanesAsmOperand : AsmOperandClass {
207 let Name = "VecListDPairSpacedAllLanes";
208 let ParserMethod = "parseVectorList";
209 let RenderMethod = "addVecListOperands";
211 def VecListDPairSpacedAllLanes : RegisterOperand<DPairSpc,
212 "printVectorListTwoSpacedAllLanes"> {
213 let ParserMatchClass = VecListDPairSpacedAllLanesAsmOperand;
215 // Register list of three D registers, with "all lanes" subscripting.
216 def VecListThreeDAllLanesAsmOperand : AsmOperandClass {
217 let Name = "VecListThreeDAllLanes";
218 let ParserMethod = "parseVectorList";
219 let RenderMethod = "addVecListOperands";
221 def VecListThreeDAllLanes : RegisterOperand<DPR,
222 "printVectorListThreeAllLanes"> {
223 let ParserMatchClass = VecListThreeDAllLanesAsmOperand;
225 // Register list of three D registers spaced by 2 (three sequential Q regs).
226 def VecListThreeQAllLanesAsmOperand : AsmOperandClass {
227 let Name = "VecListThreeQAllLanes";
228 let ParserMethod = "parseVectorList";
229 let RenderMethod = "addVecListOperands";
231 def VecListThreeQAllLanes : RegisterOperand<DPR,
232 "printVectorListThreeSpacedAllLanes"> {
233 let ParserMatchClass = VecListThreeQAllLanesAsmOperand;
235 // Register list of four D registers, with "all lanes" subscripting.
236 def VecListFourDAllLanesAsmOperand : AsmOperandClass {
237 let Name = "VecListFourDAllLanes";
238 let ParserMethod = "parseVectorList";
239 let RenderMethod = "addVecListOperands";
241 def VecListFourDAllLanes : RegisterOperand<DPR, "printVectorListFourAllLanes"> {
242 let ParserMatchClass = VecListFourDAllLanesAsmOperand;
244 // Register list of four D registers spaced by 2 (four sequential Q regs).
245 def VecListFourQAllLanesAsmOperand : AsmOperandClass {
246 let Name = "VecListFourQAllLanes";
247 let ParserMethod = "parseVectorList";
248 let RenderMethod = "addVecListOperands";
250 def VecListFourQAllLanes : RegisterOperand<DPR,
251 "printVectorListFourSpacedAllLanes"> {
252 let ParserMatchClass = VecListFourQAllLanesAsmOperand;
256 // Register list of one D register, with byte lane subscripting.
257 def VecListOneDByteIndexAsmOperand : AsmOperandClass {
258 let Name = "VecListOneDByteIndexed";
259 let ParserMethod = "parseVectorList";
260 let RenderMethod = "addVecListIndexedOperands";
262 def VecListOneDByteIndexed : Operand<i32> {
263 let ParserMatchClass = VecListOneDByteIndexAsmOperand;
264 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
266 // ...with half-word lane subscripting.
267 def VecListOneDHWordIndexAsmOperand : AsmOperandClass {
268 let Name = "VecListOneDHWordIndexed";
269 let ParserMethod = "parseVectorList";
270 let RenderMethod = "addVecListIndexedOperands";
272 def VecListOneDHWordIndexed : Operand<i32> {
273 let ParserMatchClass = VecListOneDHWordIndexAsmOperand;
274 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
276 // ...with word lane subscripting.
277 def VecListOneDWordIndexAsmOperand : AsmOperandClass {
278 let Name = "VecListOneDWordIndexed";
279 let ParserMethod = "parseVectorList";
280 let RenderMethod = "addVecListIndexedOperands";
282 def VecListOneDWordIndexed : Operand<i32> {
283 let ParserMatchClass = VecListOneDWordIndexAsmOperand;
284 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
287 // Register list of two D registers with byte lane subscripting.
288 def VecListTwoDByteIndexAsmOperand : AsmOperandClass {
289 let Name = "VecListTwoDByteIndexed";
290 let ParserMethod = "parseVectorList";
291 let RenderMethod = "addVecListIndexedOperands";
293 def VecListTwoDByteIndexed : Operand<i32> {
294 let ParserMatchClass = VecListTwoDByteIndexAsmOperand;
295 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
297 // ...with half-word lane subscripting.
298 def VecListTwoDHWordIndexAsmOperand : AsmOperandClass {
299 let Name = "VecListTwoDHWordIndexed";
300 let ParserMethod = "parseVectorList";
301 let RenderMethod = "addVecListIndexedOperands";
303 def VecListTwoDHWordIndexed : Operand<i32> {
304 let ParserMatchClass = VecListTwoDHWordIndexAsmOperand;
305 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
307 // ...with word lane subscripting.
308 def VecListTwoDWordIndexAsmOperand : AsmOperandClass {
309 let Name = "VecListTwoDWordIndexed";
310 let ParserMethod = "parseVectorList";
311 let RenderMethod = "addVecListIndexedOperands";
313 def VecListTwoDWordIndexed : Operand<i32> {
314 let ParserMatchClass = VecListTwoDWordIndexAsmOperand;
315 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
317 // Register list of two Q registers with half-word lane subscripting.
318 def VecListTwoQHWordIndexAsmOperand : AsmOperandClass {
319 let Name = "VecListTwoQHWordIndexed";
320 let ParserMethod = "parseVectorList";
321 let RenderMethod = "addVecListIndexedOperands";
323 def VecListTwoQHWordIndexed : Operand<i32> {
324 let ParserMatchClass = VecListTwoQHWordIndexAsmOperand;
325 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
327 // ...with word lane subscripting.
328 def VecListTwoQWordIndexAsmOperand : AsmOperandClass {
329 let Name = "VecListTwoQWordIndexed";
330 let ParserMethod = "parseVectorList";
331 let RenderMethod = "addVecListIndexedOperands";
333 def VecListTwoQWordIndexed : Operand<i32> {
334 let ParserMatchClass = VecListTwoQWordIndexAsmOperand;
335 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
339 // Register list of three D registers with byte lane subscripting.
340 def VecListThreeDByteIndexAsmOperand : AsmOperandClass {
341 let Name = "VecListThreeDByteIndexed";
342 let ParserMethod = "parseVectorList";
343 let RenderMethod = "addVecListIndexedOperands";
345 def VecListThreeDByteIndexed : Operand<i32> {
346 let ParserMatchClass = VecListThreeDByteIndexAsmOperand;
347 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
349 // ...with half-word lane subscripting.
350 def VecListThreeDHWordIndexAsmOperand : AsmOperandClass {
351 let Name = "VecListThreeDHWordIndexed";
352 let ParserMethod = "parseVectorList";
353 let RenderMethod = "addVecListIndexedOperands";
355 def VecListThreeDHWordIndexed : Operand<i32> {
356 let ParserMatchClass = VecListThreeDHWordIndexAsmOperand;
357 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
359 // ...with word lane subscripting.
360 def VecListThreeDWordIndexAsmOperand : AsmOperandClass {
361 let Name = "VecListThreeDWordIndexed";
362 let ParserMethod = "parseVectorList";
363 let RenderMethod = "addVecListIndexedOperands";
365 def VecListThreeDWordIndexed : Operand<i32> {
366 let ParserMatchClass = VecListThreeDWordIndexAsmOperand;
367 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
369 // Register list of three Q registers with half-word lane subscripting.
370 def VecListThreeQHWordIndexAsmOperand : AsmOperandClass {
371 let Name = "VecListThreeQHWordIndexed";
372 let ParserMethod = "parseVectorList";
373 let RenderMethod = "addVecListIndexedOperands";
375 def VecListThreeQHWordIndexed : Operand<i32> {
376 let ParserMatchClass = VecListThreeQHWordIndexAsmOperand;
377 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
379 // ...with word lane subscripting.
380 def VecListThreeQWordIndexAsmOperand : AsmOperandClass {
381 let Name = "VecListThreeQWordIndexed";
382 let ParserMethod = "parseVectorList";
383 let RenderMethod = "addVecListIndexedOperands";
385 def VecListThreeQWordIndexed : Operand<i32> {
386 let ParserMatchClass = VecListThreeQWordIndexAsmOperand;
387 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
390 // Register list of four D registers with byte lane subscripting.
391 def VecListFourDByteIndexAsmOperand : AsmOperandClass {
392 let Name = "VecListFourDByteIndexed";
393 let ParserMethod = "parseVectorList";
394 let RenderMethod = "addVecListIndexedOperands";
396 def VecListFourDByteIndexed : Operand<i32> {
397 let ParserMatchClass = VecListFourDByteIndexAsmOperand;
398 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
400 // ...with half-word lane subscripting.
401 def VecListFourDHWordIndexAsmOperand : AsmOperandClass {
402 let Name = "VecListFourDHWordIndexed";
403 let ParserMethod = "parseVectorList";
404 let RenderMethod = "addVecListIndexedOperands";
406 def VecListFourDHWordIndexed : Operand<i32> {
407 let ParserMatchClass = VecListFourDHWordIndexAsmOperand;
408 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
410 // ...with word lane subscripting.
411 def VecListFourDWordIndexAsmOperand : AsmOperandClass {
412 let Name = "VecListFourDWordIndexed";
413 let ParserMethod = "parseVectorList";
414 let RenderMethod = "addVecListIndexedOperands";
416 def VecListFourDWordIndexed : Operand<i32> {
417 let ParserMatchClass = VecListFourDWordIndexAsmOperand;
418 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
420 // Register list of four Q registers with half-word lane subscripting.
421 def VecListFourQHWordIndexAsmOperand : AsmOperandClass {
422 let Name = "VecListFourQHWordIndexed";
423 let ParserMethod = "parseVectorList";
424 let RenderMethod = "addVecListIndexedOperands";
426 def VecListFourQHWordIndexed : Operand<i32> {
427 let ParserMatchClass = VecListFourQHWordIndexAsmOperand;
428 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
430 // ...with word lane subscripting.
431 def VecListFourQWordIndexAsmOperand : AsmOperandClass {
432 let Name = "VecListFourQWordIndexed";
433 let ParserMethod = "parseVectorList";
434 let RenderMethod = "addVecListIndexedOperands";
436 def VecListFourQWordIndexed : Operand<i32> {
437 let ParserMatchClass = VecListFourQWordIndexAsmOperand;
438 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
441 def dword_alignedload : PatFrag<(ops node:$ptr), (load node:$ptr), [{
442 return cast<LoadSDNode>(N)->getAlignment() >= 8;
444 def dword_alignedstore : PatFrag<(ops node:$val, node:$ptr),
445 (store node:$val, node:$ptr), [{
446 return cast<StoreSDNode>(N)->getAlignment() >= 8;
448 def word_alignedload : PatFrag<(ops node:$ptr), (load node:$ptr), [{
449 return cast<LoadSDNode>(N)->getAlignment() == 4;
451 def word_alignedstore : PatFrag<(ops node:$val, node:$ptr),
452 (store node:$val, node:$ptr), [{
453 return cast<StoreSDNode>(N)->getAlignment() == 4;
455 def hword_alignedload : PatFrag<(ops node:$ptr), (load node:$ptr), [{
456 return cast<LoadSDNode>(N)->getAlignment() == 2;
458 def hword_alignedstore : PatFrag<(ops node:$val, node:$ptr),
459 (store node:$val, node:$ptr), [{
460 return cast<StoreSDNode>(N)->getAlignment() == 2;
462 def byte_alignedload : PatFrag<(ops node:$ptr), (load node:$ptr), [{
463 return cast<LoadSDNode>(N)->getAlignment() == 1;
465 def byte_alignedstore : PatFrag<(ops node:$val, node:$ptr),
466 (store node:$val, node:$ptr), [{
467 return cast<StoreSDNode>(N)->getAlignment() == 1;
469 def non_word_alignedload : PatFrag<(ops node:$ptr), (load node:$ptr), [{
470 return cast<LoadSDNode>(N)->getAlignment() < 4;
472 def non_word_alignedstore : PatFrag<(ops node:$val, node:$ptr),
473 (store node:$val, node:$ptr), [{
474 return cast<StoreSDNode>(N)->getAlignment() < 4;
477 //===----------------------------------------------------------------------===//
478 // NEON-specific DAG Nodes.
479 //===----------------------------------------------------------------------===//
481 def SDTARMVCMP : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<1, 2>]>;
482 def SDTARMVCMPZ : SDTypeProfile<1, 1, []>;
484 def NEONvceq : SDNode<"ARMISD::VCEQ", SDTARMVCMP>;
485 def NEONvceqz : SDNode<"ARMISD::VCEQZ", SDTARMVCMPZ>;
486 def NEONvcge : SDNode<"ARMISD::VCGE", SDTARMVCMP>;
487 def NEONvcgez : SDNode<"ARMISD::VCGEZ", SDTARMVCMPZ>;
488 def NEONvclez : SDNode<"ARMISD::VCLEZ", SDTARMVCMPZ>;
489 def NEONvcgeu : SDNode<"ARMISD::VCGEU", SDTARMVCMP>;
490 def NEONvcgt : SDNode<"ARMISD::VCGT", SDTARMVCMP>;
491 def NEONvcgtz : SDNode<"ARMISD::VCGTZ", SDTARMVCMPZ>;
492 def NEONvcltz : SDNode<"ARMISD::VCLTZ", SDTARMVCMPZ>;
493 def NEONvcgtu : SDNode<"ARMISD::VCGTU", SDTARMVCMP>;
494 def NEONvtst : SDNode<"ARMISD::VTST", SDTARMVCMP>;
496 // Types for vector shift by immediates. The "SHX" version is for long and
497 // narrow operations where the source and destination vectors have different
498 // types. The "SHINS" version is for shift and insert operations.
499 def SDTARMVSHXIMM : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
501 def SDTARMVSHINSIMM : SDTypeProfile<1, 3, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
502 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
504 def NEONvshrnImm : SDNode<"ARMISD::VSHRNIMM", SDTARMVSHXIMM>;
506 def NEONvrshrsImm : SDNode<"ARMISD::VRSHRsIMM", SDTARMVSHIMM>;
507 def NEONvrshruImm : SDNode<"ARMISD::VRSHRuIMM", SDTARMVSHIMM>;
508 def NEONvrshrnImm : SDNode<"ARMISD::VRSHRNIMM", SDTARMVSHXIMM>;
510 def NEONvqshlsImm : SDNode<"ARMISD::VQSHLsIMM", SDTARMVSHIMM>;
511 def NEONvqshluImm : SDNode<"ARMISD::VQSHLuIMM", SDTARMVSHIMM>;
512 def NEONvqshlsuImm : SDNode<"ARMISD::VQSHLsuIMM", SDTARMVSHIMM>;
513 def NEONvqshrnsImm : SDNode<"ARMISD::VQSHRNsIMM", SDTARMVSHXIMM>;
514 def NEONvqshrnuImm : SDNode<"ARMISD::VQSHRNuIMM", SDTARMVSHXIMM>;
515 def NEONvqshrnsuImm : SDNode<"ARMISD::VQSHRNsuIMM", SDTARMVSHXIMM>;
517 def NEONvqrshrnsImm : SDNode<"ARMISD::VQRSHRNsIMM", SDTARMVSHXIMM>;
518 def NEONvqrshrnuImm : SDNode<"ARMISD::VQRSHRNuIMM", SDTARMVSHXIMM>;
519 def NEONvqrshrnsuImm : SDNode<"ARMISD::VQRSHRNsuIMM", SDTARMVSHXIMM>;
521 def NEONvsliImm : SDNode<"ARMISD::VSLIIMM", SDTARMVSHINSIMM>;
522 def NEONvsriImm : SDNode<"ARMISD::VSRIIMM", SDTARMVSHINSIMM>;
524 def SDTARMVORRIMM : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
526 def NEONvorrImm : SDNode<"ARMISD::VORRIMM", SDTARMVORRIMM>;
527 def NEONvbicImm : SDNode<"ARMISD::VBICIMM", SDTARMVORRIMM>;
529 def NEONvbsl : SDNode<"ARMISD::VBSL",
530 SDTypeProfile<1, 3, [SDTCisVec<0>,
533 SDTCisSameAs<0, 3>]>>;
535 def SDTARMVEXT : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
536 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
537 def NEONvext : SDNode<"ARMISD::VEXT", SDTARMVEXT>;
539 def SDTARMVSHUF2 : SDTypeProfile<2, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
541 SDTCisSameAs<0, 3>]>;
542 def NEONzip : SDNode<"ARMISD::VZIP", SDTARMVSHUF2>;
543 def NEONuzp : SDNode<"ARMISD::VUZP", SDTARMVSHUF2>;
544 def NEONtrn : SDNode<"ARMISD::VTRN", SDTARMVSHUF2>;
546 def SDTARMVMULL : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
547 SDTCisSameAs<1, 2>]>;
548 def NEONvmulls : SDNode<"ARMISD::VMULLs", SDTARMVMULL>;
549 def NEONvmullu : SDNode<"ARMISD::VMULLu", SDTARMVMULL>;
551 def SDTARMVTBL1 : SDTypeProfile<1, 2, [SDTCisVT<0, v8i8>, SDTCisVT<1, v8i8>,
553 def SDTARMVTBL2 : SDTypeProfile<1, 3, [SDTCisVT<0, v8i8>, SDTCisVT<1, v8i8>,
554 SDTCisVT<2, v8i8>, SDTCisVT<3, v8i8>]>;
555 def NEONvtbl1 : SDNode<"ARMISD::VTBL1", SDTARMVTBL1>;
556 def NEONvtbl2 : SDNode<"ARMISD::VTBL2", SDTARMVTBL2>;
559 def NEONimmAllZerosV: PatLeaf<(ARMvmovImm (i32 timm)), [{
560 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
561 unsigned EltBits = 0;
562 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
563 return (EltBits == 32 && EltVal == 0);
566 def NEONimmAllOnesV: PatLeaf<(ARMvmovImm (i32 timm)), [{
567 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
568 unsigned EltBits = 0;
569 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
570 return (EltBits == 8 && EltVal == 0xff);
573 //===----------------------------------------------------------------------===//
574 // NEON load / store instructions
575 //===----------------------------------------------------------------------===//
577 // Use VLDM to load a Q register as a D register pair.
578 // This is a pseudo instruction that is expanded to VLDMD after reg alloc.
580 : PseudoVFPLdStM<(outs DPair:$dst), (ins GPR:$Rn),
582 [(set DPair:$dst, (v2f64 (word_alignedload GPR:$Rn)))]>;
584 // Use VSTM to store a Q register as a D register pair.
585 // This is a pseudo instruction that is expanded to VSTMD after reg alloc.
587 : PseudoVFPLdStM<(outs), (ins DPair:$src, GPR:$Rn),
589 [(word_alignedstore (v2f64 DPair:$src), GPR:$Rn)]>;
591 // Classes for VLD* pseudo-instructions with multi-register operands.
592 // These are expanded to real instructions after register allocation.
593 class VLDQPseudo<InstrItinClass itin>
594 : PseudoNLdSt<(outs QPR:$dst), (ins addrmode6:$addr), itin, "">;
595 class VLDQWBPseudo<InstrItinClass itin>
596 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
597 (ins addrmode6:$addr, am6offset:$offset), itin,
599 class VLDQWBfixedPseudo<InstrItinClass itin>
600 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
601 (ins addrmode6:$addr), itin,
603 class VLDQWBregisterPseudo<InstrItinClass itin>
604 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
605 (ins addrmode6:$addr, rGPR:$offset), itin,
608 class VLDQQPseudo<InstrItinClass itin>
609 : PseudoNLdSt<(outs QQPR:$dst), (ins addrmode6:$addr), itin, "">;
610 class VLDQQWBPseudo<InstrItinClass itin>
611 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
612 (ins addrmode6:$addr, am6offset:$offset), itin,
614 class VLDQQWBfixedPseudo<InstrItinClass itin>
615 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
616 (ins addrmode6:$addr), itin,
618 class VLDQQWBregisterPseudo<InstrItinClass itin>
619 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
620 (ins addrmode6:$addr, rGPR:$offset), itin,
624 class VLDQQQQPseudo<InstrItinClass itin>
625 : PseudoNLdSt<(outs QQQQPR:$dst), (ins addrmode6:$addr, QQQQPR:$src),itin,
627 class VLDQQQQWBPseudo<InstrItinClass itin>
628 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
629 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
630 "$addr.addr = $wb, $src = $dst">;
632 let mayLoad = 1, hasSideEffects = 0, hasExtraDefRegAllocReq = 1 in {
634 // VLD1 : Vector Load (multiple single elements)
635 class VLD1D<bits<4> op7_4, string Dt, Operand AddrMode>
636 : NLdSt<0,0b10,0b0111,op7_4, (outs VecListOneD:$Vd),
637 (ins AddrMode:$Rn), IIC_VLD1,
638 "vld1", Dt, "$Vd, $Rn", "", []>, Sched<[WriteVLD1]> {
641 let DecoderMethod = "DecodeVLDST1Instruction";
643 class VLD1Q<bits<4> op7_4, string Dt, Operand AddrMode>
644 : NLdSt<0,0b10,0b1010,op7_4, (outs VecListDPair:$Vd),
645 (ins AddrMode:$Rn), IIC_VLD1x2,
646 "vld1", Dt, "$Vd, $Rn", "", []>, Sched<[WriteVLD2]> {
648 let Inst{5-4} = Rn{5-4};
649 let DecoderMethod = "DecodeVLDST1Instruction";
652 def VLD1d8 : VLD1D<{0,0,0,?}, "8", addrmode6align64>;
653 def VLD1d16 : VLD1D<{0,1,0,?}, "16", addrmode6align64>;
654 def VLD1d32 : VLD1D<{1,0,0,?}, "32", addrmode6align64>;
655 def VLD1d64 : VLD1D<{1,1,0,?}, "64", addrmode6align64>;
657 def VLD1q8 : VLD1Q<{0,0,?,?}, "8", addrmode6align64or128>;
658 def VLD1q16 : VLD1Q<{0,1,?,?}, "16", addrmode6align64or128>;
659 def VLD1q32 : VLD1Q<{1,0,?,?}, "32", addrmode6align64or128>;
660 def VLD1q64 : VLD1Q<{1,1,?,?}, "64", addrmode6align64or128>;
662 // ...with address register writeback:
663 multiclass VLD1DWB<bits<4> op7_4, string Dt, Operand AddrMode> {
664 def _fixed : NLdSt<0,0b10, 0b0111,op7_4, (outs VecListOneD:$Vd, GPR:$wb),
665 (ins AddrMode:$Rn), IIC_VLD1u,
666 "vld1", Dt, "$Vd, $Rn!",
667 "$Rn.addr = $wb", []>, Sched<[WriteVLD1]> {
668 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
670 let DecoderMethod = "DecodeVLDST1Instruction";
672 def _register : NLdSt<0,0b10,0b0111,op7_4, (outs VecListOneD:$Vd, GPR:$wb),
673 (ins AddrMode:$Rn, rGPR:$Rm), IIC_VLD1u,
674 "vld1", Dt, "$Vd, $Rn, $Rm",
675 "$Rn.addr = $wb", []>, Sched<[WriteVLD1]> {
677 let DecoderMethod = "DecodeVLDST1Instruction";
680 multiclass VLD1QWB<bits<4> op7_4, string Dt, Operand AddrMode> {
681 def _fixed : NLdSt<0,0b10,0b1010,op7_4, (outs VecListDPair:$Vd, GPR:$wb),
682 (ins AddrMode:$Rn), IIC_VLD1x2u,
683 "vld1", Dt, "$Vd, $Rn!",
684 "$Rn.addr = $wb", []>, Sched<[WriteVLD2]> {
685 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
686 let Inst{5-4} = Rn{5-4};
687 let DecoderMethod = "DecodeVLDST1Instruction";
689 def _register : NLdSt<0,0b10,0b1010,op7_4, (outs VecListDPair:$Vd, GPR:$wb),
690 (ins AddrMode:$Rn, rGPR:$Rm), IIC_VLD1x2u,
691 "vld1", Dt, "$Vd, $Rn, $Rm",
692 "$Rn.addr = $wb", []>, Sched<[WriteVLD2]> {
693 let Inst{5-4} = Rn{5-4};
694 let DecoderMethod = "DecodeVLDST1Instruction";
698 defm VLD1d8wb : VLD1DWB<{0,0,0,?}, "8", addrmode6align64>;
699 defm VLD1d16wb : VLD1DWB<{0,1,0,?}, "16", addrmode6align64>;
700 defm VLD1d32wb : VLD1DWB<{1,0,0,?}, "32", addrmode6align64>;
701 defm VLD1d64wb : VLD1DWB<{1,1,0,?}, "64", addrmode6align64>;
702 defm VLD1q8wb : VLD1QWB<{0,0,?,?}, "8", addrmode6align64or128>;
703 defm VLD1q16wb : VLD1QWB<{0,1,?,?}, "16", addrmode6align64or128>;
704 defm VLD1q32wb : VLD1QWB<{1,0,?,?}, "32", addrmode6align64or128>;
705 defm VLD1q64wb : VLD1QWB<{1,1,?,?}, "64", addrmode6align64or128>;
707 // ...with 3 registers
708 class VLD1D3<bits<4> op7_4, string Dt, Operand AddrMode>
709 : NLdSt<0,0b10,0b0110,op7_4, (outs VecListThreeD:$Vd),
710 (ins AddrMode:$Rn), IIC_VLD1x3, "vld1", Dt,
711 "$Vd, $Rn", "", []>, Sched<[WriteVLD3]> {
714 let DecoderMethod = "DecodeVLDST1Instruction";
716 multiclass VLD1D3WB<bits<4> op7_4, string Dt, Operand AddrMode> {
717 def _fixed : NLdSt<0,0b10,0b0110, op7_4, (outs VecListThreeD:$Vd, GPR:$wb),
718 (ins AddrMode:$Rn), IIC_VLD1x2u,
719 "vld1", Dt, "$Vd, $Rn!",
720 "$Rn.addr = $wb", []>, Sched<[WriteVLD3]> {
721 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
723 let DecoderMethod = "DecodeVLDST1Instruction";
725 def _register : NLdSt<0,0b10,0b0110,op7_4, (outs VecListThreeD:$Vd, GPR:$wb),
726 (ins AddrMode:$Rn, rGPR:$Rm), IIC_VLD1x2u,
727 "vld1", Dt, "$Vd, $Rn, $Rm",
728 "$Rn.addr = $wb", []>, Sched<[WriteVLD3]> {
730 let DecoderMethod = "DecodeVLDST1Instruction";
734 def VLD1d8T : VLD1D3<{0,0,0,?}, "8", addrmode6align64>;
735 def VLD1d16T : VLD1D3<{0,1,0,?}, "16", addrmode6align64>;
736 def VLD1d32T : VLD1D3<{1,0,0,?}, "32", addrmode6align64>;
737 def VLD1d64T : VLD1D3<{1,1,0,?}, "64", addrmode6align64>;
739 defm VLD1d8Twb : VLD1D3WB<{0,0,0,?}, "8", addrmode6align64>;
740 defm VLD1d16Twb : VLD1D3WB<{0,1,0,?}, "16", addrmode6align64>;
741 defm VLD1d32Twb : VLD1D3WB<{1,0,0,?}, "32", addrmode6align64>;
742 defm VLD1d64Twb : VLD1D3WB<{1,1,0,?}, "64", addrmode6align64>;
744 def VLD1d8TPseudo : VLDQQPseudo<IIC_VLD1x3>, Sched<[WriteVLD3]>;
745 def VLD1d16TPseudo : VLDQQPseudo<IIC_VLD1x3>, Sched<[WriteVLD3]>;
746 def VLD1d32TPseudo : VLDQQPseudo<IIC_VLD1x3>, Sched<[WriteVLD3]>;
747 def VLD1d64TPseudo : VLDQQPseudo<IIC_VLD1x3>, Sched<[WriteVLD3]>;
748 def VLD1d64TPseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD1x3>, Sched<[WriteVLD3]>;
749 def VLD1d64TPseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD1x3>, Sched<[WriteVLD3]>;
751 def VLD1q8HighTPseudo : VLDQQQQPseudo<IIC_VLD1x3>, Sched<[WriteVLD3]>;
752 def VLD1q8LowTPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD1x3>, Sched<[WriteVLD3]>;
753 def VLD1q16HighTPseudo : VLDQQQQPseudo<IIC_VLD1x3>, Sched<[WriteVLD3]>;
754 def VLD1q16LowTPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD1x3>, Sched<[WriteVLD3]>;
755 def VLD1q32HighTPseudo : VLDQQQQPseudo<IIC_VLD1x3>, Sched<[WriteVLD3]>;
756 def VLD1q32LowTPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD1x3>, Sched<[WriteVLD3]>;
757 def VLD1q64HighTPseudo : VLDQQQQPseudo<IIC_VLD1x3>, Sched<[WriteVLD3]>;
758 def VLD1q64LowTPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD1x3>, Sched<[WriteVLD3]>;
760 // ...with 4 registers
761 class VLD1D4<bits<4> op7_4, string Dt, Operand AddrMode>
762 : NLdSt<0, 0b10, 0b0010, op7_4, (outs VecListFourD:$Vd),
763 (ins AddrMode:$Rn), IIC_VLD1x4, "vld1", Dt,
764 "$Vd, $Rn", "", []>, Sched<[WriteVLD4]> {
766 let Inst{5-4} = Rn{5-4};
767 let DecoderMethod = "DecodeVLDST1Instruction";
769 multiclass VLD1D4WB<bits<4> op7_4, string Dt, Operand AddrMode> {
770 def _fixed : NLdSt<0,0b10,0b0010, op7_4, (outs VecListFourD:$Vd, GPR:$wb),
771 (ins AddrMode:$Rn), IIC_VLD1x2u,
772 "vld1", Dt, "$Vd, $Rn!",
773 "$Rn.addr = $wb", []>, Sched<[WriteVLD4]> {
774 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
775 let Inst{5-4} = Rn{5-4};
776 let DecoderMethod = "DecodeVLDST1Instruction";
778 def _register : NLdSt<0,0b10,0b0010,op7_4, (outs VecListFourD:$Vd, GPR:$wb),
779 (ins AddrMode:$Rn, rGPR:$Rm), IIC_VLD1x2u,
780 "vld1", Dt, "$Vd, $Rn, $Rm",
781 "$Rn.addr = $wb", []>, Sched<[WriteVLD4]> {
782 let Inst{5-4} = Rn{5-4};
783 let DecoderMethod = "DecodeVLDST1Instruction";
787 def VLD1d8Q : VLD1D4<{0,0,?,?}, "8", addrmode6align64or128or256>;
788 def VLD1d16Q : VLD1D4<{0,1,?,?}, "16", addrmode6align64or128or256>;
789 def VLD1d32Q : VLD1D4<{1,0,?,?}, "32", addrmode6align64or128or256>;
790 def VLD1d64Q : VLD1D4<{1,1,?,?}, "64", addrmode6align64or128or256>;
792 defm VLD1d8Qwb : VLD1D4WB<{0,0,?,?}, "8", addrmode6align64or128or256>;
793 defm VLD1d16Qwb : VLD1D4WB<{0,1,?,?}, "16", addrmode6align64or128or256>;
794 defm VLD1d32Qwb : VLD1D4WB<{1,0,?,?}, "32", addrmode6align64or128or256>;
795 defm VLD1d64Qwb : VLD1D4WB<{1,1,?,?}, "64", addrmode6align64or128or256>;
797 def VLD1d8QPseudo : VLDQQPseudo<IIC_VLD1x4>, Sched<[WriteVLD4]>;
798 def VLD1d16QPseudo : VLDQQPseudo<IIC_VLD1x4>, Sched<[WriteVLD4]>;
799 def VLD1d32QPseudo : VLDQQPseudo<IIC_VLD1x4>, Sched<[WriteVLD4]>;
800 def VLD1d64QPseudo : VLDQQPseudo<IIC_VLD1x4>, Sched<[WriteVLD4]>;
801 def VLD1d64QPseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD1x4>, Sched<[WriteVLD4]>;
802 def VLD1d64QPseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD1x4>, Sched<[WriteVLD4]>;
804 def VLD1q8LowQPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD1x4>, Sched<[WriteVLD4]>;
805 def VLD1q8HighQPseudo : VLDQQQQPseudo<IIC_VLD1x4>, Sched<[WriteVLD4]>;
806 def VLD1q16LowQPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD1x4>, Sched<[WriteVLD4]>;
807 def VLD1q16HighQPseudo : VLDQQQQPseudo<IIC_VLD1x4>, Sched<[WriteVLD4]>;
808 def VLD1q32LowQPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD1x4>, Sched<[WriteVLD4]>;
809 def VLD1q32HighQPseudo : VLDQQQQPseudo<IIC_VLD1x4>, Sched<[WriteVLD4]>;
810 def VLD1q64LowQPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD1x4>, Sched<[WriteVLD4]>;
811 def VLD1q64HighQPseudo : VLDQQQQPseudo<IIC_VLD1x4>, Sched<[WriteVLD4]>;
813 // VLD2 : Vector Load (multiple 2-element structures)
814 class VLD2<bits<4> op11_8, bits<4> op7_4, string Dt, RegisterOperand VdTy,
815 InstrItinClass itin, Operand AddrMode>
816 : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd),
817 (ins AddrMode:$Rn), itin,
818 "vld2", Dt, "$Vd, $Rn", "", []> {
820 let Inst{5-4} = Rn{5-4};
821 let DecoderMethod = "DecodeVLDST2Instruction";
824 def VLD2d8 : VLD2<0b1000, {0,0,?,?}, "8", VecListDPair, IIC_VLD2,
825 addrmode6align64or128>, Sched<[WriteVLD2]>;
826 def VLD2d16 : VLD2<0b1000, {0,1,?,?}, "16", VecListDPair, IIC_VLD2,
827 addrmode6align64or128>, Sched<[WriteVLD2]>;
828 def VLD2d32 : VLD2<0b1000, {1,0,?,?}, "32", VecListDPair, IIC_VLD2,
829 addrmode6align64or128>, Sched<[WriteVLD2]>;
831 def VLD2q8 : VLD2<0b0011, {0,0,?,?}, "8", VecListFourD, IIC_VLD2x2,
832 addrmode6align64or128or256>, Sched<[WriteVLD4]>;
833 def VLD2q16 : VLD2<0b0011, {0,1,?,?}, "16", VecListFourD, IIC_VLD2x2,
834 addrmode6align64or128or256>, Sched<[WriteVLD4]>;
835 def VLD2q32 : VLD2<0b0011, {1,0,?,?}, "32", VecListFourD, IIC_VLD2x2,
836 addrmode6align64or128or256>, Sched<[WriteVLD4]>;
838 def VLD2q8Pseudo : VLDQQPseudo<IIC_VLD2x2>, Sched<[WriteVLD4]>;
839 def VLD2q16Pseudo : VLDQQPseudo<IIC_VLD2x2>, Sched<[WriteVLD4]>;
840 def VLD2q32Pseudo : VLDQQPseudo<IIC_VLD2x2>, Sched<[WriteVLD4]>;
842 // ...with address register writeback:
843 multiclass VLD2WB<bits<4> op11_8, bits<4> op7_4, string Dt,
844 RegisterOperand VdTy, InstrItinClass itin, Operand AddrMode> {
845 def _fixed : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd, GPR:$wb),
846 (ins AddrMode:$Rn), itin,
847 "vld2", Dt, "$Vd, $Rn!",
848 "$Rn.addr = $wb", []> {
849 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
850 let Inst{5-4} = Rn{5-4};
851 let DecoderMethod = "DecodeVLDST2Instruction";
853 def _register : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd, GPR:$wb),
854 (ins AddrMode:$Rn, rGPR:$Rm), itin,
855 "vld2", Dt, "$Vd, $Rn, $Rm",
856 "$Rn.addr = $wb", []> {
857 let Inst{5-4} = Rn{5-4};
858 let DecoderMethod = "DecodeVLDST2Instruction";
862 defm VLD2d8wb : VLD2WB<0b1000, {0,0,?,?}, "8", VecListDPair, IIC_VLD2u,
863 addrmode6align64or128>, Sched<[WriteVLD2]>;
864 defm VLD2d16wb : VLD2WB<0b1000, {0,1,?,?}, "16", VecListDPair, IIC_VLD2u,
865 addrmode6align64or128>, Sched<[WriteVLD2]>;
866 defm VLD2d32wb : VLD2WB<0b1000, {1,0,?,?}, "32", VecListDPair, IIC_VLD2u,
867 addrmode6align64or128>, Sched<[WriteVLD2]>;
869 defm VLD2q8wb : VLD2WB<0b0011, {0,0,?,?}, "8", VecListFourD, IIC_VLD2x2u,
870 addrmode6align64or128or256>, Sched<[WriteVLD4]>;
871 defm VLD2q16wb : VLD2WB<0b0011, {0,1,?,?}, "16", VecListFourD, IIC_VLD2x2u,
872 addrmode6align64or128or256>, Sched<[WriteVLD4]>;
873 defm VLD2q32wb : VLD2WB<0b0011, {1,0,?,?}, "32", VecListFourD, IIC_VLD2x2u,
874 addrmode6align64or128or256>, Sched<[WriteVLD4]>;
876 def VLD2q8PseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD2x2u>, Sched<[WriteVLD4]>;
877 def VLD2q16PseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD2x2u>, Sched<[WriteVLD4]>;
878 def VLD2q32PseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD2x2u>, Sched<[WriteVLD4]>;
879 def VLD2q8PseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD2x2u>, Sched<[WriteVLD4]>;
880 def VLD2q16PseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD2x2u>, Sched<[WriteVLD4]>;
881 def VLD2q32PseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD2x2u>, Sched<[WriteVLD4]>;
883 // ...with double-spaced registers
884 def VLD2b8 : VLD2<0b1001, {0,0,?,?}, "8", VecListDPairSpaced, IIC_VLD2,
885 addrmode6align64or128>, Sched<[WriteVLD2]>;
886 def VLD2b16 : VLD2<0b1001, {0,1,?,?}, "16", VecListDPairSpaced, IIC_VLD2,
887 addrmode6align64or128>, Sched<[WriteVLD2]>;
888 def VLD2b32 : VLD2<0b1001, {1,0,?,?}, "32", VecListDPairSpaced, IIC_VLD2,
889 addrmode6align64or128>, Sched<[WriteVLD2]>;
890 defm VLD2b8wb : VLD2WB<0b1001, {0,0,?,?}, "8", VecListDPairSpaced, IIC_VLD2u,
891 addrmode6align64or128>, Sched<[WriteVLD2]>;
892 defm VLD2b16wb : VLD2WB<0b1001, {0,1,?,?}, "16", VecListDPairSpaced, IIC_VLD2u,
893 addrmode6align64or128>, Sched<[WriteVLD2]>;
894 defm VLD2b32wb : VLD2WB<0b1001, {1,0,?,?}, "32", VecListDPairSpaced, IIC_VLD2u,
895 addrmode6align64or128>, Sched<[WriteVLD2]>;
897 // VLD3 : Vector Load (multiple 3-element structures)
898 class VLD3D<bits<4> op11_8, bits<4> op7_4, string Dt>
899 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
900 (ins addrmode6:$Rn), IIC_VLD3,
901 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn", "", []>, Sched<[WriteVLD3]> {
904 let DecoderMethod = "DecodeVLDST3Instruction";
907 def VLD3d8 : VLD3D<0b0100, {0,0,0,?}, "8">;
908 def VLD3d16 : VLD3D<0b0100, {0,1,0,?}, "16">;
909 def VLD3d32 : VLD3D<0b0100, {1,0,0,?}, "32">;
911 def VLD3d8Pseudo : VLDQQPseudo<IIC_VLD3>, Sched<[WriteVLD3]>;
912 def VLD3d16Pseudo : VLDQQPseudo<IIC_VLD3>, Sched<[WriteVLD3]>;
913 def VLD3d32Pseudo : VLDQQPseudo<IIC_VLD3>, Sched<[WriteVLD3]>;
915 // ...with address register writeback:
916 class VLD3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
917 : NLdSt<0, 0b10, op11_8, op7_4,
918 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
919 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD3u,
920 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn$Rm",
921 "$Rn.addr = $wb", []>, Sched<[WriteVLD3]> {
923 let DecoderMethod = "DecodeVLDST3Instruction";
926 def VLD3d8_UPD : VLD3DWB<0b0100, {0,0,0,?}, "8">;
927 def VLD3d16_UPD : VLD3DWB<0b0100, {0,1,0,?}, "16">;
928 def VLD3d32_UPD : VLD3DWB<0b0100, {1,0,0,?}, "32">;
930 def VLD3d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>, Sched<[WriteVLD3]>;
931 def VLD3d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>, Sched<[WriteVLD3]>;
932 def VLD3d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>, Sched<[WriteVLD3]>;
934 // ...with double-spaced registers:
935 def VLD3q8 : VLD3D<0b0101, {0,0,0,?}, "8">;
936 def VLD3q16 : VLD3D<0b0101, {0,1,0,?}, "16">;
937 def VLD3q32 : VLD3D<0b0101, {1,0,0,?}, "32">;
938 def VLD3q8_UPD : VLD3DWB<0b0101, {0,0,0,?}, "8">;
939 def VLD3q16_UPD : VLD3DWB<0b0101, {0,1,0,?}, "16">;
940 def VLD3q32_UPD : VLD3DWB<0b0101, {1,0,0,?}, "32">;
942 def VLD3q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>, Sched<[WriteVLD3]>;
943 def VLD3q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>, Sched<[WriteVLD3]>;
944 def VLD3q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>, Sched<[WriteVLD3]>;
946 // ...alternate versions to be allocated odd register numbers:
947 def VLD3q8oddPseudo : VLDQQQQPseudo<IIC_VLD3>, Sched<[WriteVLD3]>;
948 def VLD3q16oddPseudo : VLDQQQQPseudo<IIC_VLD3>, Sched<[WriteVLD3]>;
949 def VLD3q32oddPseudo : VLDQQQQPseudo<IIC_VLD3>, Sched<[WriteVLD3]>;
951 def VLD3q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>, Sched<[WriteVLD3]>;
952 def VLD3q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>, Sched<[WriteVLD3]>;
953 def VLD3q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>, Sched<[WriteVLD3]>;
955 // VLD4 : Vector Load (multiple 4-element structures)
956 class VLD4D<bits<4> op11_8, bits<4> op7_4, string Dt>
957 : NLdSt<0, 0b10, op11_8, op7_4,
958 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
959 (ins addrmode6:$Rn), IIC_VLD4,
960 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []>,
963 let Inst{5-4} = Rn{5-4};
964 let DecoderMethod = "DecodeVLDST4Instruction";
967 def VLD4d8 : VLD4D<0b0000, {0,0,?,?}, "8">;
968 def VLD4d16 : VLD4D<0b0000, {0,1,?,?}, "16">;
969 def VLD4d32 : VLD4D<0b0000, {1,0,?,?}, "32">;
971 def VLD4d8Pseudo : VLDQQPseudo<IIC_VLD4>, Sched<[WriteVLD4]>;
972 def VLD4d16Pseudo : VLDQQPseudo<IIC_VLD4>, Sched<[WriteVLD4]>;
973 def VLD4d32Pseudo : VLDQQPseudo<IIC_VLD4>, Sched<[WriteVLD4]>;
975 // ...with address register writeback:
976 class VLD4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
977 : NLdSt<0, 0b10, op11_8, op7_4,
978 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
979 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD4u,
980 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm",
981 "$Rn.addr = $wb", []>, Sched<[WriteVLD4]> {
982 let Inst{5-4} = Rn{5-4};
983 let DecoderMethod = "DecodeVLDST4Instruction";
986 def VLD4d8_UPD : VLD4DWB<0b0000, {0,0,?,?}, "8">;
987 def VLD4d16_UPD : VLD4DWB<0b0000, {0,1,?,?}, "16">;
988 def VLD4d32_UPD : VLD4DWB<0b0000, {1,0,?,?}, "32">;
990 def VLD4d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>, Sched<[WriteVLD4]>;
991 def VLD4d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>, Sched<[WriteVLD4]>;
992 def VLD4d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>, Sched<[WriteVLD4]>;
994 // ...with double-spaced registers:
995 def VLD4q8 : VLD4D<0b0001, {0,0,?,?}, "8">;
996 def VLD4q16 : VLD4D<0b0001, {0,1,?,?}, "16">;
997 def VLD4q32 : VLD4D<0b0001, {1,0,?,?}, "32">;
998 def VLD4q8_UPD : VLD4DWB<0b0001, {0,0,?,?}, "8">;
999 def VLD4q16_UPD : VLD4DWB<0b0001, {0,1,?,?}, "16">;
1000 def VLD4q32_UPD : VLD4DWB<0b0001, {1,0,?,?}, "32">;
1002 def VLD4q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>, Sched<[WriteVLD4]>;
1003 def VLD4q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>, Sched<[WriteVLD4]>;
1004 def VLD4q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>, Sched<[WriteVLD4]>;
1006 // ...alternate versions to be allocated odd register numbers:
1007 def VLD4q8oddPseudo : VLDQQQQPseudo<IIC_VLD4>, Sched<[WriteVLD4]>;
1008 def VLD4q16oddPseudo : VLDQQQQPseudo<IIC_VLD4>, Sched<[WriteVLD4]>;
1009 def VLD4q32oddPseudo : VLDQQQQPseudo<IIC_VLD4>, Sched<[WriteVLD4]>;
1011 def VLD4q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>, Sched<[WriteVLD4]>;
1012 def VLD4q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>, Sched<[WriteVLD4]>;
1013 def VLD4q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>, Sched<[WriteVLD4]>;
1015 } // mayLoad = 1, hasSideEffects = 0, hasExtraDefRegAllocReq = 1
1017 // Classes for VLD*LN pseudo-instructions with multi-register operands.
1018 // These are expanded to real instructions after register allocation.
1019 class VLDQLNPseudo<InstrItinClass itin>
1020 : PseudoNLdSt<(outs QPR:$dst),
1021 (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
1022 itin, "$src = $dst">;
1023 class VLDQLNWBPseudo<InstrItinClass itin>
1024 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
1025 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
1026 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
1027 class VLDQQLNPseudo<InstrItinClass itin>
1028 : PseudoNLdSt<(outs QQPR:$dst),
1029 (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
1030 itin, "$src = $dst">;
1031 class VLDQQLNWBPseudo<InstrItinClass itin>
1032 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
1033 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
1034 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
1035 class VLDQQQQLNPseudo<InstrItinClass itin>
1036 : PseudoNLdSt<(outs QQQQPR:$dst),
1037 (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
1038 itin, "$src = $dst">;
1039 class VLDQQQQLNWBPseudo<InstrItinClass itin>
1040 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
1041 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
1042 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
1044 // VLD1LN : Vector Load (single element to one lane)
1045 class VLD1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1047 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
1048 (ins addrmode6:$Rn, DPR:$src, nohash_imm:$lane),
1049 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
1051 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
1052 (i32 (LoadOp addrmode6:$Rn)),
1055 let DecoderMethod = "DecodeVLD1LN";
1057 class VLD1LN32<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1059 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
1060 (ins addrmode6oneL32:$Rn, DPR:$src, nohash_imm:$lane),
1061 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
1063 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
1064 (i32 (LoadOp addrmode6oneL32:$Rn)),
1065 imm:$lane))]>, Sched<[WriteVLD1]> {
1067 let DecoderMethod = "DecodeVLD1LN";
1069 class VLD1QLNPseudo<ValueType Ty, PatFrag LoadOp> : VLDQLNPseudo<IIC_VLD1ln>,
1070 Sched<[WriteVLD1]> {
1071 let Pattern = [(set QPR:$dst, (vector_insert (Ty QPR:$src),
1072 (i32 (LoadOp addrmode6:$addr)),
1076 def VLD1LNd8 : VLD1LN<0b0000, {?,?,?,0}, "8", v8i8, extloadi8> {
1077 let Inst{7-5} = lane{2-0};
1079 def VLD1LNd16 : VLD1LN<0b0100, {?,?,0,?}, "16", v4i16, extloadi16> {
1080 let Inst{7-6} = lane{1-0};
1081 let Inst{5-4} = Rn{5-4};
1083 def VLD1LNd32 : VLD1LN32<0b1000, {?,0,?,?}, "32", v2i32, load> {
1084 let Inst{7} = lane{0};
1085 let Inst{5-4} = Rn{5-4};
1088 def VLD1LNq8Pseudo : VLD1QLNPseudo<v16i8, extloadi8>;
1089 def VLD1LNq16Pseudo : VLD1QLNPseudo<v8i16, extloadi16>;
1090 def VLD1LNq32Pseudo : VLD1QLNPseudo<v4i32, load>;
1092 let Predicates = [HasNEON] in {
1093 def : Pat<(vector_insert (v4f16 DPR:$src),
1094 (f16 (load addrmode6:$addr)), imm:$lane),
1095 (VLD1LNd16 addrmode6:$addr, DPR:$src, imm:$lane)>;
1096 def : Pat<(vector_insert (v8f16 QPR:$src),
1097 (f16 (load addrmode6:$addr)), imm:$lane),
1098 (VLD1LNq16Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
1099 def : Pat<(vector_insert (v2f32 DPR:$src),
1100 (f32 (load addrmode6:$addr)), imm:$lane),
1101 (VLD1LNd32 addrmode6:$addr, DPR:$src, imm:$lane)>;
1102 def : Pat<(vector_insert (v4f32 QPR:$src),
1103 (f32 (load addrmode6:$addr)), imm:$lane),
1104 (VLD1LNq32Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
1106 // A 64-bit subvector insert to the first 128-bit vector position
1107 // is a subregister copy that needs no instruction.
1108 def : Pat<(insert_subvector undef, (v1i64 DPR:$src), (i32 0)),
1109 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
1110 def : Pat<(insert_subvector undef, (v2i32 DPR:$src), (i32 0)),
1111 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
1112 def : Pat<(insert_subvector undef, (v2f32 DPR:$src), (i32 0)),
1113 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
1114 def : Pat<(insert_subvector undef, (v4i16 DPR:$src), (i32 0)),
1115 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
1116 def : Pat<(insert_subvector undef, (v4f16 DPR:$src), (i32 0)),
1117 (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
1118 def : Pat<(insert_subvector (v16i8 undef), (v8i8 DPR:$src), (i32 0)),
1119 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
1123 let mayLoad = 1, hasSideEffects = 0, hasExtraDefRegAllocReq = 1 in {
1125 // ...with address register writeback:
1126 class VLD1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1127 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, GPR:$wb),
1128 (ins addrmode6:$Rn, am6offset:$Rm,
1129 DPR:$src, nohash_imm:$lane), IIC_VLD1lnu, "vld1", Dt,
1130 "\\{$Vd[$lane]\\}, $Rn$Rm",
1131 "$src = $Vd, $Rn.addr = $wb", []>, Sched<[WriteVLD1]> {
1132 let DecoderMethod = "DecodeVLD1LN";
1135 def VLD1LNd8_UPD : VLD1LNWB<0b0000, {?,?,?,0}, "8"> {
1136 let Inst{7-5} = lane{2-0};
1138 def VLD1LNd16_UPD : VLD1LNWB<0b0100, {?,?,0,?}, "16"> {
1139 let Inst{7-6} = lane{1-0};
1140 let Inst{4} = Rn{4};
1142 def VLD1LNd32_UPD : VLD1LNWB<0b1000, {?,0,?,?}, "32"> {
1143 let Inst{7} = lane{0};
1144 let Inst{5} = Rn{4};
1145 let Inst{4} = Rn{4};
1148 def VLD1LNq8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>, Sched<[WriteVLD1]>;
1149 def VLD1LNq16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>, Sched<[WriteVLD1]>;
1150 def VLD1LNq32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>, Sched<[WriteVLD1]>;
1152 // VLD2LN : Vector Load (single 2-element structure to one lane)
1153 class VLD2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1154 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2),
1155 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, nohash_imm:$lane),
1156 IIC_VLD2ln, "vld2", Dt, "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn",
1157 "$src1 = $Vd, $src2 = $dst2", []>, Sched<[WriteVLD1]> {
1159 let Inst{4} = Rn{4};
1160 let DecoderMethod = "DecodeVLD2LN";
1163 def VLD2LNd8 : VLD2LN<0b0001, {?,?,?,?}, "8"> {
1164 let Inst{7-5} = lane{2-0};
1166 def VLD2LNd16 : VLD2LN<0b0101, {?,?,0,?}, "16"> {
1167 let Inst{7-6} = lane{1-0};
1169 def VLD2LNd32 : VLD2LN<0b1001, {?,0,0,?}, "32"> {
1170 let Inst{7} = lane{0};
1173 def VLD2LNd8Pseudo : VLDQLNPseudo<IIC_VLD2ln>, Sched<[WriteVLD1]>;
1174 def VLD2LNd16Pseudo : VLDQLNPseudo<IIC_VLD2ln>, Sched<[WriteVLD1]>;
1175 def VLD2LNd32Pseudo : VLDQLNPseudo<IIC_VLD2ln>, Sched<[WriteVLD1]>;
1177 // ...with double-spaced registers:
1178 def VLD2LNq16 : VLD2LN<0b0101, {?,?,1,?}, "16"> {
1179 let Inst{7-6} = lane{1-0};
1181 def VLD2LNq32 : VLD2LN<0b1001, {?,1,0,?}, "32"> {
1182 let Inst{7} = lane{0};
1185 def VLD2LNq16Pseudo : VLDQQLNPseudo<IIC_VLD2ln>, Sched<[WriteVLD1]>;
1186 def VLD2LNq32Pseudo : VLDQQLNPseudo<IIC_VLD2ln>, Sched<[WriteVLD1]>;
1188 // ...with address register writeback:
1189 class VLD2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1190 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
1191 (ins addrmode6:$Rn, am6offset:$Rm,
1192 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VLD2lnu, "vld2", Dt,
1193 "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn$Rm",
1194 "$src1 = $Vd, $src2 = $dst2, $Rn.addr = $wb", []> {
1195 let Inst{4} = Rn{4};
1196 let DecoderMethod = "DecodeVLD2LN";
1199 def VLD2LNd8_UPD : VLD2LNWB<0b0001, {?,?,?,?}, "8"> {
1200 let Inst{7-5} = lane{2-0};
1202 def VLD2LNd16_UPD : VLD2LNWB<0b0101, {?,?,0,?}, "16"> {
1203 let Inst{7-6} = lane{1-0};
1205 def VLD2LNd32_UPD : VLD2LNWB<0b1001, {?,0,0,?}, "32"> {
1206 let Inst{7} = lane{0};
1209 def VLD2LNd8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>, Sched<[WriteVLD1]>;
1210 def VLD2LNd16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>, Sched<[WriteVLD1]>;
1211 def VLD2LNd32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>, Sched<[WriteVLD1]>;
1213 def VLD2LNq16_UPD : VLD2LNWB<0b0101, {?,?,1,?}, "16"> {
1214 let Inst{7-6} = lane{1-0};
1216 def VLD2LNq32_UPD : VLD2LNWB<0b1001, {?,1,0,?}, "32"> {
1217 let Inst{7} = lane{0};
1220 def VLD2LNq16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>, Sched<[WriteVLD1]>;
1221 def VLD2LNq32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>, Sched<[WriteVLD1]>;
1223 // VLD3LN : Vector Load (single 3-element structure to one lane)
1224 class VLD3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1225 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
1226 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3,
1227 nohash_imm:$lane), IIC_VLD3ln, "vld3", Dt,
1228 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn",
1229 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3", []>, Sched<[WriteVLD2]> {
1231 let DecoderMethod = "DecodeVLD3LN";
1234 def VLD3LNd8 : VLD3LN<0b0010, {?,?,?,0}, "8"> {
1235 let Inst{7-5} = lane{2-0};
1237 def VLD3LNd16 : VLD3LN<0b0110, {?,?,0,0}, "16"> {
1238 let Inst{7-6} = lane{1-0};
1240 def VLD3LNd32 : VLD3LN<0b1010, {?,0,0,0}, "32"> {
1241 let Inst{7} = lane{0};
1244 def VLD3LNd8Pseudo : VLDQQLNPseudo<IIC_VLD3ln>, Sched<[WriteVLD2]>;
1245 def VLD3LNd16Pseudo : VLDQQLNPseudo<IIC_VLD3ln>, Sched<[WriteVLD2]>;
1246 def VLD3LNd32Pseudo : VLDQQLNPseudo<IIC_VLD3ln>, Sched<[WriteVLD2]>;
1248 // ...with double-spaced registers:
1249 def VLD3LNq16 : VLD3LN<0b0110, {?,?,1,0}, "16"> {
1250 let Inst{7-6} = lane{1-0};
1252 def VLD3LNq32 : VLD3LN<0b1010, {?,1,0,0}, "32"> {
1253 let Inst{7} = lane{0};
1256 def VLD3LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>, Sched<[WriteVLD2]>;
1257 def VLD3LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>, Sched<[WriteVLD2]>;
1259 // ...with address register writeback:
1260 class VLD3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1261 : NLdStLn<1, 0b10, op11_8, op7_4,
1262 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
1263 (ins addrmode6:$Rn, am6offset:$Rm,
1264 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
1265 IIC_VLD3lnu, "vld3", Dt,
1266 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn$Rm",
1267 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $Rn.addr = $wb",
1268 []>, Sched<[WriteVLD2]> {
1269 let DecoderMethod = "DecodeVLD3LN";
1272 def VLD3LNd8_UPD : VLD3LNWB<0b0010, {?,?,?,0}, "8"> {
1273 let Inst{7-5} = lane{2-0};
1275 def VLD3LNd16_UPD : VLD3LNWB<0b0110, {?,?,0,0}, "16"> {
1276 let Inst{7-6} = lane{1-0};
1278 def VLD3LNd32_UPD : VLD3LNWB<0b1010, {?,0,0,0}, "32"> {
1279 let Inst{7} = lane{0};
1282 def VLD3LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>, Sched<[WriteVLD2]>;
1283 def VLD3LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>, Sched<[WriteVLD2]>;
1284 def VLD3LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>, Sched<[WriteVLD2]>;
1286 def VLD3LNq16_UPD : VLD3LNWB<0b0110, {?,?,1,0}, "16"> {
1287 let Inst{7-6} = lane{1-0};
1289 def VLD3LNq32_UPD : VLD3LNWB<0b1010, {?,1,0,0}, "32"> {
1290 let Inst{7} = lane{0};
1293 def VLD3LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>, Sched<[WriteVLD2]>;
1294 def VLD3LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>, Sched<[WriteVLD2]>;
1296 // VLD4LN : Vector Load (single 4-element structure to one lane)
1297 class VLD4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1298 : NLdStLn<1, 0b10, op11_8, op7_4,
1299 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
1300 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
1301 nohash_imm:$lane), IIC_VLD4ln, "vld4", Dt,
1302 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn",
1303 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4", []>,
1304 Sched<[WriteVLD2]> {
1306 let Inst{4} = Rn{4};
1307 let DecoderMethod = "DecodeVLD4LN";
1310 def VLD4LNd8 : VLD4LN<0b0011, {?,?,?,?}, "8"> {
1311 let Inst{7-5} = lane{2-0};
1313 def VLD4LNd16 : VLD4LN<0b0111, {?,?,0,?}, "16"> {
1314 let Inst{7-6} = lane{1-0};
1316 def VLD4LNd32 : VLD4LN<0b1011, {?,0,?,?}, "32"> {
1317 let Inst{7} = lane{0};
1318 let Inst{5} = Rn{5};
1321 def VLD4LNd8Pseudo : VLDQQLNPseudo<IIC_VLD4ln>, Sched<[WriteVLD2]>;
1322 def VLD4LNd16Pseudo : VLDQQLNPseudo<IIC_VLD4ln>, Sched<[WriteVLD2]>;
1323 def VLD4LNd32Pseudo : VLDQQLNPseudo<IIC_VLD4ln>, Sched<[WriteVLD2]>;
1325 // ...with double-spaced registers:
1326 def VLD4LNq16 : VLD4LN<0b0111, {?,?,1,?}, "16"> {
1327 let Inst{7-6} = lane{1-0};
1329 def VLD4LNq32 : VLD4LN<0b1011, {?,1,?,?}, "32"> {
1330 let Inst{7} = lane{0};
1331 let Inst{5} = Rn{5};
1334 def VLD4LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>, Sched<[WriteVLD2]>;
1335 def VLD4LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>, Sched<[WriteVLD2]>;
1337 // ...with address register writeback:
1338 class VLD4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1339 : NLdStLn<1, 0b10, op11_8, op7_4,
1340 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
1341 (ins addrmode6:$Rn, am6offset:$Rm,
1342 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
1343 IIC_VLD4lnu, "vld4", Dt,
1344 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn$Rm",
1345 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4, $Rn.addr = $wb",
1347 let Inst{4} = Rn{4};
1348 let DecoderMethod = "DecodeVLD4LN" ;
1351 def VLD4LNd8_UPD : VLD4LNWB<0b0011, {?,?,?,?}, "8"> {
1352 let Inst{7-5} = lane{2-0};
1354 def VLD4LNd16_UPD : VLD4LNWB<0b0111, {?,?,0,?}, "16"> {
1355 let Inst{7-6} = lane{1-0};
1357 def VLD4LNd32_UPD : VLD4LNWB<0b1011, {?,0,?,?}, "32"> {
1358 let Inst{7} = lane{0};
1359 let Inst{5} = Rn{5};
1362 def VLD4LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>, Sched<[WriteVLD2]>;
1363 def VLD4LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>, Sched<[WriteVLD2]>;
1364 def VLD4LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>, Sched<[WriteVLD2]>;
1366 def VLD4LNq16_UPD : VLD4LNWB<0b0111, {?,?,1,?}, "16"> {
1367 let Inst{7-6} = lane{1-0};
1369 def VLD4LNq32_UPD : VLD4LNWB<0b1011, {?,1,?,?}, "32"> {
1370 let Inst{7} = lane{0};
1371 let Inst{5} = Rn{5};
1374 def VLD4LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>, Sched<[WriteVLD2]>;
1375 def VLD4LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>, Sched<[WriteVLD2]>;
1377 } // mayLoad = 1, hasSideEffects = 0, hasExtraDefRegAllocReq = 1
1379 // VLD1DUP : Vector Load (single element to all lanes)
1380 class VLD1DUP<bits<4> op7_4, string Dt, ValueType Ty, PatFrag LoadOp,
1382 : NLdSt<1, 0b10, 0b1100, op7_4, (outs VecListOneDAllLanes:$Vd),
1384 IIC_VLD1dup, "vld1", Dt, "$Vd, $Rn", "",
1385 [(set VecListOneDAllLanes:$Vd,
1386 (Ty (ARMvdup (i32 (LoadOp AddrMode:$Rn)))))]>,
1387 Sched<[WriteVLD2]> {
1389 let Inst{4} = Rn{4};
1390 let DecoderMethod = "DecodeVLD1DupInstruction";
1392 def VLD1DUPd8 : VLD1DUP<{0,0,0,?}, "8", v8i8, extloadi8,
1393 addrmode6dupalignNone>;
1394 def VLD1DUPd16 : VLD1DUP<{0,1,0,?}, "16", v4i16, extloadi16,
1395 addrmode6dupalign16>;
1396 def VLD1DUPd32 : VLD1DUP<{1,0,0,?}, "32", v2i32, load,
1397 addrmode6dupalign32>;
1399 let Predicates = [HasNEON] in {
1400 def : Pat<(v2f32 (ARMvdup (f32 (load addrmode6dup:$addr)))),
1401 (VLD1DUPd32 addrmode6:$addr)>;
1404 class VLD1QDUP<bits<4> op7_4, string Dt, ValueType Ty, PatFrag LoadOp,
1406 : NLdSt<1, 0b10, 0b1100, op7_4, (outs VecListDPairAllLanes:$Vd),
1407 (ins AddrMode:$Rn), IIC_VLD1dup,
1408 "vld1", Dt, "$Vd, $Rn", "",
1409 [(set VecListDPairAllLanes:$Vd,
1410 (Ty (ARMvdup (i32 (LoadOp AddrMode:$Rn)))))]> {
1412 let Inst{4} = Rn{4};
1413 let DecoderMethod = "DecodeVLD1DupInstruction";
1416 def VLD1DUPq8 : VLD1QDUP<{0,0,1,0}, "8", v16i8, extloadi8,
1417 addrmode6dupalignNone>;
1418 def VLD1DUPq16 : VLD1QDUP<{0,1,1,?}, "16", v8i16, extloadi16,
1419 addrmode6dupalign16>;
1420 def VLD1DUPq32 : VLD1QDUP<{1,0,1,?}, "32", v4i32, load,
1421 addrmode6dupalign32>;
1423 let Predicates = [HasNEON] in {
1424 def : Pat<(v4f32 (ARMvdup (f32 (load addrmode6dup:$addr)))),
1425 (VLD1DUPq32 addrmode6:$addr)>;
1428 let mayLoad = 1, hasSideEffects = 0, hasExtraDefRegAllocReq = 1 in {
1429 // ...with address register writeback:
1430 multiclass VLD1DUPWB<bits<4> op7_4, string Dt, Operand AddrMode> {
1431 def _fixed : NLdSt<1, 0b10, 0b1100, op7_4,
1432 (outs VecListOneDAllLanes:$Vd, GPR:$wb),
1433 (ins AddrMode:$Rn), IIC_VLD1dupu,
1434 "vld1", Dt, "$Vd, $Rn!",
1435 "$Rn.addr = $wb", []> {
1436 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1437 let Inst{4} = Rn{4};
1438 let DecoderMethod = "DecodeVLD1DupInstruction";
1440 def _register : NLdSt<1, 0b10, 0b1100, op7_4,
1441 (outs VecListOneDAllLanes:$Vd, GPR:$wb),
1442 (ins AddrMode:$Rn, rGPR:$Rm), IIC_VLD1dupu,
1443 "vld1", Dt, "$Vd, $Rn, $Rm",
1444 "$Rn.addr = $wb", []> {
1445 let Inst{4} = Rn{4};
1446 let DecoderMethod = "DecodeVLD1DupInstruction";
1449 multiclass VLD1QDUPWB<bits<4> op7_4, string Dt, Operand AddrMode> {
1450 def _fixed : NLdSt<1, 0b10, 0b1100, op7_4,
1451 (outs VecListDPairAllLanes:$Vd, GPR:$wb),
1452 (ins AddrMode:$Rn), IIC_VLD1dupu,
1453 "vld1", Dt, "$Vd, $Rn!",
1454 "$Rn.addr = $wb", []>, Sched<[WriteVLD1]> {
1455 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1456 let Inst{4} = Rn{4};
1457 let DecoderMethod = "DecodeVLD1DupInstruction";
1459 def _register : NLdSt<1, 0b10, 0b1100, op7_4,
1460 (outs VecListDPairAllLanes:$Vd, GPR:$wb),
1461 (ins AddrMode:$Rn, rGPR:$Rm), IIC_VLD1dupu,
1462 "vld1", Dt, "$Vd, $Rn, $Rm",
1463 "$Rn.addr = $wb", []> {
1464 let Inst{4} = Rn{4};
1465 let DecoderMethod = "DecodeVLD1DupInstruction";
1469 defm VLD1DUPd8wb : VLD1DUPWB<{0,0,0,0}, "8", addrmode6dupalignNone>;
1470 defm VLD1DUPd16wb : VLD1DUPWB<{0,1,0,?}, "16", addrmode6dupalign16>;
1471 defm VLD1DUPd32wb : VLD1DUPWB<{1,0,0,?}, "32", addrmode6dupalign32>;
1473 defm VLD1DUPq8wb : VLD1QDUPWB<{0,0,1,0}, "8", addrmode6dupalignNone>;
1474 defm VLD1DUPq16wb : VLD1QDUPWB<{0,1,1,?}, "16", addrmode6dupalign16>;
1475 defm VLD1DUPq32wb : VLD1QDUPWB<{1,0,1,?}, "32", addrmode6dupalign32>;
1477 // VLD2DUP : Vector Load (single 2-element structure to all lanes)
1478 class VLD2DUP<bits<4> op7_4, string Dt, RegisterOperand VdTy, Operand AddrMode>
1479 : NLdSt<1, 0b10, 0b1101, op7_4, (outs VdTy:$Vd),
1480 (ins AddrMode:$Rn), IIC_VLD2dup,
1481 "vld2", Dt, "$Vd, $Rn", "", []> {
1483 let Inst{4} = Rn{4};
1484 let DecoderMethod = "DecodeVLD2DupInstruction";
1487 def VLD2DUPd8 : VLD2DUP<{0,0,0,?}, "8", VecListDPairAllLanes,
1488 addrmode6dupalign16>;
1489 def VLD2DUPd16 : VLD2DUP<{0,1,0,?}, "16", VecListDPairAllLanes,
1490 addrmode6dupalign32>;
1491 def VLD2DUPd32 : VLD2DUP<{1,0,0,?}, "32", VecListDPairAllLanes,
1492 addrmode6dupalign64>;
1494 // HACK this one, VLD2DUPd8x2 must be changed at the same time with VLD2b8 or
1495 // "vld2.8 {d0[], d2[]}, [r4:32]" will become "vld2.8 {d0, d2}, [r4:32]".
1496 // ...with double-spaced registers
1497 def VLD2DUPd8x2 : VLD2DUP<{0,0,1,?}, "8", VecListDPairSpacedAllLanes,
1498 addrmode6dupalign16>;
1499 def VLD2DUPd16x2 : VLD2DUP<{0,1,1,?}, "16", VecListDPairSpacedAllLanes,
1500 addrmode6dupalign32>;
1501 def VLD2DUPd32x2 : VLD2DUP<{1,0,1,?}, "32", VecListDPairSpacedAllLanes,
1502 addrmode6dupalign64>;
1504 def VLD2DUPq8EvenPseudo : VLDQQPseudo<IIC_VLD2dup>, Sched<[WriteVLD2]>;
1505 def VLD2DUPq8OddPseudo : VLDQQPseudo<IIC_VLD2dup>, Sched<[WriteVLD2]>;
1506 def VLD2DUPq16EvenPseudo : VLDQQPseudo<IIC_VLD2dup>, Sched<[WriteVLD2]>;
1507 def VLD2DUPq16OddPseudo : VLDQQPseudo<IIC_VLD2dup>, Sched<[WriteVLD2]>;
1508 def VLD2DUPq32EvenPseudo : VLDQQPseudo<IIC_VLD2dup>, Sched<[WriteVLD2]>;
1509 def VLD2DUPq32OddPseudo : VLDQQPseudo<IIC_VLD2dup>, Sched<[WriteVLD2]>;
1511 // ...with address register writeback:
1512 multiclass VLD2DUPWB<bits<4> op7_4, string Dt, RegisterOperand VdTy,
1514 def _fixed : NLdSt<1, 0b10, 0b1101, op7_4,
1515 (outs VdTy:$Vd, GPR:$wb),
1516 (ins AddrMode:$Rn), IIC_VLD2dupu,
1517 "vld2", Dt, "$Vd, $Rn!",
1518 "$Rn.addr = $wb", []>, Sched<[WriteVLD1]> {
1519 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1520 let Inst{4} = Rn{4};
1521 let DecoderMethod = "DecodeVLD2DupInstruction";
1523 def _register : NLdSt<1, 0b10, 0b1101, op7_4,
1524 (outs VdTy:$Vd, GPR:$wb),
1525 (ins AddrMode:$Rn, rGPR:$Rm), IIC_VLD2dupu,
1526 "vld2", Dt, "$Vd, $Rn, $Rm",
1527 "$Rn.addr = $wb", []>, Sched<[WriteVLD1]> {
1528 let Inst{4} = Rn{4};
1529 let DecoderMethod = "DecodeVLD2DupInstruction";
1533 defm VLD2DUPd8wb : VLD2DUPWB<{0,0,0,0}, "8", VecListDPairAllLanes,
1534 addrmode6dupalign16>;
1535 defm VLD2DUPd16wb : VLD2DUPWB<{0,1,0,?}, "16", VecListDPairAllLanes,
1536 addrmode6dupalign32>;
1537 defm VLD2DUPd32wb : VLD2DUPWB<{1,0,0,?}, "32", VecListDPairAllLanes,
1538 addrmode6dupalign64>;
1540 defm VLD2DUPd8x2wb : VLD2DUPWB<{0,0,1,0}, "8", VecListDPairSpacedAllLanes,
1541 addrmode6dupalign16>;
1542 defm VLD2DUPd16x2wb : VLD2DUPWB<{0,1,1,?}, "16", VecListDPairSpacedAllLanes,
1543 addrmode6dupalign32>;
1544 defm VLD2DUPd32x2wb : VLD2DUPWB<{1,0,1,?}, "32", VecListDPairSpacedAllLanes,
1545 addrmode6dupalign64>;
1547 // VLD3DUP : Vector Load (single 3-element structure to all lanes)
1548 class VLD3DUP<bits<4> op7_4, string Dt>
1549 : NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
1550 (ins addrmode6dup:$Rn), IIC_VLD3dup,
1551 "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn", "", []>,
1552 Sched<[WriteVLD2]> {
1555 let DecoderMethod = "DecodeVLD3DupInstruction";
1558 def VLD3DUPd8 : VLD3DUP<{0,0,0,?}, "8">;
1559 def VLD3DUPd16 : VLD3DUP<{0,1,0,?}, "16">;
1560 def VLD3DUPd32 : VLD3DUP<{1,0,0,?}, "32">;
1562 def VLD3DUPd8Pseudo : VLDQQPseudo<IIC_VLD3dup>, Sched<[WriteVLD2]>;
1563 def VLD3DUPd16Pseudo : VLDQQPseudo<IIC_VLD3dup>, Sched<[WriteVLD2]>;
1564 def VLD3DUPd32Pseudo : VLDQQPseudo<IIC_VLD3dup>, Sched<[WriteVLD2]>;
1566 // ...with double-spaced registers (not used for codegen):
1567 def VLD3DUPq8 : VLD3DUP<{0,0,1,?}, "8">;
1568 def VLD3DUPq16 : VLD3DUP<{0,1,1,?}, "16">;
1569 def VLD3DUPq32 : VLD3DUP<{1,0,1,?}, "32">;
1571 def VLD3DUPq8EvenPseudo : VLDQQQQPseudo<IIC_VLD3dup>, Sched<[WriteVLD2]>;
1572 def VLD3DUPq8OddPseudo : VLDQQQQPseudo<IIC_VLD3dup>, Sched<[WriteVLD2]>;
1573 def VLD3DUPq16EvenPseudo : VLDQQQQPseudo<IIC_VLD3dup>, Sched<[WriteVLD2]>;
1574 def VLD3DUPq16OddPseudo : VLDQQQQPseudo<IIC_VLD3dup>, Sched<[WriteVLD2]>;
1575 def VLD3DUPq32EvenPseudo : VLDQQQQPseudo<IIC_VLD3dup>, Sched<[WriteVLD2]>;
1576 def VLD3DUPq32OddPseudo : VLDQQQQPseudo<IIC_VLD3dup>, Sched<[WriteVLD2]>;
1578 // ...with address register writeback:
1579 class VLD3DUPWB<bits<4> op7_4, string Dt, Operand AddrMode>
1580 : NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
1581 (ins AddrMode:$Rn, am6offset:$Rm), IIC_VLD3dupu,
1582 "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn$Rm",
1583 "$Rn.addr = $wb", []>, Sched<[WriteVLD2]> {
1585 let DecoderMethod = "DecodeVLD3DupInstruction";
1588 def VLD3DUPd8_UPD : VLD3DUPWB<{0,0,0,0}, "8", addrmode6dupalign64>;
1589 def VLD3DUPd16_UPD : VLD3DUPWB<{0,1,0,?}, "16", addrmode6dupalign64>;
1590 def VLD3DUPd32_UPD : VLD3DUPWB<{1,0,0,?}, "32", addrmode6dupalign64>;
1592 def VLD3DUPq8_UPD : VLD3DUPWB<{0,0,1,0}, "8", addrmode6dupalign64>;
1593 def VLD3DUPq16_UPD : VLD3DUPWB<{0,1,1,?}, "16", addrmode6dupalign64>;
1594 def VLD3DUPq32_UPD : VLD3DUPWB<{1,0,1,?}, "32", addrmode6dupalign64>;
1596 def VLD3DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>, Sched<[WriteVLD2]>;
1597 def VLD3DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>, Sched<[WriteVLD2]>;
1598 def VLD3DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>, Sched<[WriteVLD2]>;
1600 // VLD4DUP : Vector Load (single 4-element structure to all lanes)
1601 class VLD4DUP<bits<4> op7_4, string Dt>
1602 : NLdSt<1, 0b10, 0b1111, op7_4,
1603 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
1604 (ins addrmode6dup:$Rn), IIC_VLD4dup,
1605 "vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn", "", []> {
1607 let Inst{4} = Rn{4};
1608 let DecoderMethod = "DecodeVLD4DupInstruction";
1611 def VLD4DUPd8 : VLD4DUP<{0,0,0,?}, "8">;
1612 def VLD4DUPd16 : VLD4DUP<{0,1,0,?}, "16">;
1613 def VLD4DUPd32 : VLD4DUP<{1,?,0,?}, "32"> { let Inst{6} = Rn{5}; }
1615 def VLD4DUPd8Pseudo : VLDQQPseudo<IIC_VLD4dup>, Sched<[WriteVLD2]>;
1616 def VLD4DUPd16Pseudo : VLDQQPseudo<IIC_VLD4dup>, Sched<[WriteVLD2]>;
1617 def VLD4DUPd32Pseudo : VLDQQPseudo<IIC_VLD4dup>, Sched<[WriteVLD2]>;
1619 // ...with double-spaced registers (not used for codegen):
1620 def VLD4DUPq8 : VLD4DUP<{0,0,1,?}, "8">;
1621 def VLD4DUPq16 : VLD4DUP<{0,1,1,?}, "16">;
1622 def VLD4DUPq32 : VLD4DUP<{1,?,1,?}, "32"> { let Inst{6} = Rn{5}; }
1624 def VLD4DUPq8EvenPseudo : VLDQQQQPseudo<IIC_VLD4dup>, Sched<[WriteVLD2]>;
1625 def VLD4DUPq8OddPseudo : VLDQQQQPseudo<IIC_VLD4dup>, Sched<[WriteVLD2]>;
1626 def VLD4DUPq16EvenPseudo : VLDQQQQPseudo<IIC_VLD4dup>, Sched<[WriteVLD2]>;
1627 def VLD4DUPq16OddPseudo : VLDQQQQPseudo<IIC_VLD4dup>, Sched<[WriteVLD2]>;
1628 def VLD4DUPq32EvenPseudo : VLDQQQQPseudo<IIC_VLD4dup>, Sched<[WriteVLD2]>;
1629 def VLD4DUPq32OddPseudo : VLDQQQQPseudo<IIC_VLD4dup>, Sched<[WriteVLD2]>;
1631 // ...with address register writeback:
1632 class VLD4DUPWB<bits<4> op7_4, string Dt>
1633 : NLdSt<1, 0b10, 0b1111, op7_4,
1634 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
1635 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD4dupu,
1636 "vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn$Rm",
1637 "$Rn.addr = $wb", []>, Sched<[WriteVLD2]> {
1638 let Inst{4} = Rn{4};
1639 let DecoderMethod = "DecodeVLD4DupInstruction";
1642 def VLD4DUPd8_UPD : VLD4DUPWB<{0,0,0,0}, "8">;
1643 def VLD4DUPd16_UPD : VLD4DUPWB<{0,1,0,?}, "16">;
1644 def VLD4DUPd32_UPD : VLD4DUPWB<{1,?,0,?}, "32"> { let Inst{6} = Rn{5}; }
1646 def VLD4DUPq8_UPD : VLD4DUPWB<{0,0,1,0}, "8">;
1647 def VLD4DUPq16_UPD : VLD4DUPWB<{0,1,1,?}, "16">;
1648 def VLD4DUPq32_UPD : VLD4DUPWB<{1,?,1,?}, "32"> { let Inst{6} = Rn{5}; }
1650 def VLD4DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>, Sched<[WriteVLD2]>;
1651 def VLD4DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>, Sched<[WriteVLD2]>;
1652 def VLD4DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>, Sched<[WriteVLD2]>;
1654 } // mayLoad = 1, hasSideEffects = 0, hasExtraDefRegAllocReq = 1
1656 let mayStore = 1, hasSideEffects = 0, hasExtraSrcRegAllocReq = 1 in {
1658 // Classes for VST* pseudo-instructions with multi-register operands.
1659 // These are expanded to real instructions after register allocation.
1660 class VSTQPseudo<InstrItinClass itin>
1661 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src), itin, "">;
1662 class VSTQWBPseudo<InstrItinClass itin>
1663 : PseudoNLdSt<(outs GPR:$wb),
1664 (ins addrmode6:$addr, am6offset:$offset, QPR:$src), itin,
1665 "$addr.addr = $wb">;
1666 class VSTQWBfixedPseudo<InstrItinClass itin>
1667 : PseudoNLdSt<(outs GPR:$wb),
1668 (ins addrmode6:$addr, QPR:$src), itin,
1669 "$addr.addr = $wb">;
1670 class VSTQWBregisterPseudo<InstrItinClass itin>
1671 : PseudoNLdSt<(outs GPR:$wb),
1672 (ins addrmode6:$addr, rGPR:$offset, QPR:$src), itin,
1673 "$addr.addr = $wb">;
1674 class VSTQQPseudo<InstrItinClass itin>
1675 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src), itin, "">;
1676 class VSTQQWBPseudo<InstrItinClass itin>
1677 : PseudoNLdSt<(outs GPR:$wb),
1678 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src), itin,
1679 "$addr.addr = $wb">;
1680 class VSTQQWBfixedPseudo<InstrItinClass itin>
1681 : PseudoNLdSt<(outs GPR:$wb),
1682 (ins addrmode6:$addr, QQPR:$src), itin,
1683 "$addr.addr = $wb">;
1684 class VSTQQWBregisterPseudo<InstrItinClass itin>
1685 : PseudoNLdSt<(outs GPR:$wb),
1686 (ins addrmode6:$addr, rGPR:$offset, QQPR:$src), itin,
1687 "$addr.addr = $wb">;
1689 class VSTQQQQPseudo<InstrItinClass itin>
1690 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src), itin, "">;
1691 class VSTQQQQWBPseudo<InstrItinClass itin>
1692 : PseudoNLdSt<(outs GPR:$wb),
1693 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
1694 "$addr.addr = $wb">;
1696 // VST1 : Vector Store (multiple single elements)
1697 class VST1D<bits<4> op7_4, string Dt, Operand AddrMode>
1698 : NLdSt<0,0b00,0b0111,op7_4, (outs), (ins AddrMode:$Rn, VecListOneD:$Vd),
1699 IIC_VST1, "vst1", Dt, "$Vd, $Rn", "", []>, Sched<[WriteVST1]> {
1701 let Inst{4} = Rn{4};
1702 let DecoderMethod = "DecodeVLDST1Instruction";
1704 class VST1Q<bits<4> op7_4, string Dt, Operand AddrMode>
1705 : NLdSt<0,0b00,0b1010,op7_4, (outs), (ins AddrMode:$Rn, VecListDPair:$Vd),
1706 IIC_VST1x2, "vst1", Dt, "$Vd, $Rn", "", []>, Sched<[WriteVST2]> {
1708 let Inst{5-4} = Rn{5-4};
1709 let DecoderMethod = "DecodeVLDST1Instruction";
1712 def VST1d8 : VST1D<{0,0,0,?}, "8", addrmode6align64>;
1713 def VST1d16 : VST1D<{0,1,0,?}, "16", addrmode6align64>;
1714 def VST1d32 : VST1D<{1,0,0,?}, "32", addrmode6align64>;
1715 def VST1d64 : VST1D<{1,1,0,?}, "64", addrmode6align64>;
1717 def VST1q8 : VST1Q<{0,0,?,?}, "8", addrmode6align64or128>;
1718 def VST1q16 : VST1Q<{0,1,?,?}, "16", addrmode6align64or128>;
1719 def VST1q32 : VST1Q<{1,0,?,?}, "32", addrmode6align64or128>;
1720 def VST1q64 : VST1Q<{1,1,?,?}, "64", addrmode6align64or128>;
1722 // ...with address register writeback:
1723 multiclass VST1DWB<bits<4> op7_4, string Dt, Operand AddrMode> {
1724 def _fixed : NLdSt<0,0b00, 0b0111,op7_4, (outs GPR:$wb),
1725 (ins AddrMode:$Rn, VecListOneD:$Vd), IIC_VLD1u,
1726 "vst1", Dt, "$Vd, $Rn!",
1727 "$Rn.addr = $wb", []>, Sched<[WriteVST1]> {
1728 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1729 let Inst{4} = Rn{4};
1730 let DecoderMethod = "DecodeVLDST1Instruction";
1732 def _register : NLdSt<0,0b00,0b0111,op7_4, (outs GPR:$wb),
1733 (ins AddrMode:$Rn, rGPR:$Rm, VecListOneD:$Vd),
1735 "vst1", Dt, "$Vd, $Rn, $Rm",
1736 "$Rn.addr = $wb", []>, Sched<[WriteVST1]> {
1737 let Inst{4} = Rn{4};
1738 let DecoderMethod = "DecodeVLDST1Instruction";
1741 multiclass VST1QWB<bits<4> op7_4, string Dt, Operand AddrMode> {
1742 def _fixed : NLdSt<0,0b00,0b1010,op7_4, (outs GPR:$wb),
1743 (ins AddrMode:$Rn, VecListDPair:$Vd), IIC_VLD1x2u,
1744 "vst1", Dt, "$Vd, $Rn!",
1745 "$Rn.addr = $wb", []>, Sched<[WriteVST2]> {
1746 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1747 let Inst{5-4} = Rn{5-4};
1748 let DecoderMethod = "DecodeVLDST1Instruction";
1750 def _register : NLdSt<0,0b00,0b1010,op7_4, (outs GPR:$wb),
1751 (ins AddrMode:$Rn, rGPR:$Rm, VecListDPair:$Vd),
1753 "vst1", Dt, "$Vd, $Rn, $Rm",
1754 "$Rn.addr = $wb", []>, Sched<[WriteVST2]> {
1755 let Inst{5-4} = Rn{5-4};
1756 let DecoderMethod = "DecodeVLDST1Instruction";
1760 defm VST1d8wb : VST1DWB<{0,0,0,?}, "8", addrmode6align64>;
1761 defm VST1d16wb : VST1DWB<{0,1,0,?}, "16", addrmode6align64>;
1762 defm VST1d32wb : VST1DWB<{1,0,0,?}, "32", addrmode6align64>;
1763 defm VST1d64wb : VST1DWB<{1,1,0,?}, "64", addrmode6align64>;
1765 defm VST1q8wb : VST1QWB<{0,0,?,?}, "8", addrmode6align64or128>;
1766 defm VST1q16wb : VST1QWB<{0,1,?,?}, "16", addrmode6align64or128>;
1767 defm VST1q32wb : VST1QWB<{1,0,?,?}, "32", addrmode6align64or128>;
1768 defm VST1q64wb : VST1QWB<{1,1,?,?}, "64", addrmode6align64or128>;
1770 // ...with 3 registers
1771 class VST1D3<bits<4> op7_4, string Dt, Operand AddrMode>
1772 : NLdSt<0, 0b00, 0b0110, op7_4, (outs),
1773 (ins AddrMode:$Rn, VecListThreeD:$Vd),
1774 IIC_VST1x3, "vst1", Dt, "$Vd, $Rn", "", []>, Sched<[WriteVST3]> {
1776 let Inst{4} = Rn{4};
1777 let DecoderMethod = "DecodeVLDST1Instruction";
1779 multiclass VST1D3WB<bits<4> op7_4, string Dt, Operand AddrMode> {
1780 def _fixed : NLdSt<0,0b00,0b0110,op7_4, (outs GPR:$wb),
1781 (ins AddrMode:$Rn, VecListThreeD:$Vd), IIC_VLD1x3u,
1782 "vst1", Dt, "$Vd, $Rn!",
1783 "$Rn.addr = $wb", []>, Sched<[WriteVST3]> {
1784 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1785 let Inst{5-4} = Rn{5-4};
1786 let DecoderMethod = "DecodeVLDST1Instruction";
1788 def _register : NLdSt<0,0b00,0b0110,op7_4, (outs GPR:$wb),
1789 (ins AddrMode:$Rn, rGPR:$Rm, VecListThreeD:$Vd),
1791 "vst1", Dt, "$Vd, $Rn, $Rm",
1792 "$Rn.addr = $wb", []>, Sched<[WriteVST3]> {
1793 let Inst{5-4} = Rn{5-4};
1794 let DecoderMethod = "DecodeVLDST1Instruction";
1798 def VST1d8T : VST1D3<{0,0,0,?}, "8", addrmode6align64>;
1799 def VST1d16T : VST1D3<{0,1,0,?}, "16", addrmode6align64>;
1800 def VST1d32T : VST1D3<{1,0,0,?}, "32", addrmode6align64>;
1801 def VST1d64T : VST1D3<{1,1,0,?}, "64", addrmode6align64>;
1803 defm VST1d8Twb : VST1D3WB<{0,0,0,?}, "8", addrmode6align64>;
1804 defm VST1d16Twb : VST1D3WB<{0,1,0,?}, "16", addrmode6align64>;
1805 defm VST1d32Twb : VST1D3WB<{1,0,0,?}, "32", addrmode6align64>;
1806 defm VST1d64Twb : VST1D3WB<{1,1,0,?}, "64", addrmode6align64>;
1808 def VST1d8TPseudo : VSTQQPseudo<IIC_VST1x3>, Sched<[WriteVST3]>;
1809 def VST1d16TPseudo : VSTQQPseudo<IIC_VST1x3>, Sched<[WriteVST3]>;
1810 def VST1d32TPseudo : VSTQQPseudo<IIC_VST1x3>, Sched<[WriteVST3]>;
1811 def VST1d64TPseudo : VSTQQPseudo<IIC_VST1x3>, Sched<[WriteVST3]>;
1812 def VST1d64TPseudoWB_fixed : VSTQQWBfixedPseudo<IIC_VST1x3u>, Sched<[WriteVST3]>;
1813 def VST1d64TPseudoWB_register : VSTQQWBPseudo<IIC_VST1x3u>, Sched<[WriteVST3]>;
1815 def VST1q8HighTPseudo : VSTQQQQPseudo<IIC_VST1x3>, Sched<[WriteVST3]>;
1816 def VST1q8LowTPseudo_UPD : VSTQQQQWBPseudo<IIC_VST1x3>, Sched<[WriteVST3]>;
1817 def VST1q16HighTPseudo : VSTQQQQPseudo<IIC_VST1x3>, Sched<[WriteVST3]>;
1818 def VST1q16LowTPseudo_UPD : VSTQQQQWBPseudo<IIC_VST1x3>, Sched<[WriteVST3]>;
1819 def VST1q32HighTPseudo : VSTQQQQPseudo<IIC_VST1x3>, Sched<[WriteVST3]>;
1820 def VST1q32LowTPseudo_UPD : VSTQQQQWBPseudo<IIC_VST1x3>, Sched<[WriteVST3]>;
1821 def VST1q64HighTPseudo : VSTQQQQPseudo<IIC_VST1x3>, Sched<[WriteVST3]>;
1822 def VST1q64LowTPseudo_UPD : VSTQQQQWBPseudo<IIC_VST1x3>, Sched<[WriteVST3]>;
1824 // ...with 4 registers
1825 class VST1D4<bits<4> op7_4, string Dt, Operand AddrMode>
1826 : NLdSt<0, 0b00, 0b0010, op7_4, (outs),
1827 (ins AddrMode:$Rn, VecListFourD:$Vd),
1828 IIC_VST1x4, "vst1", Dt, "$Vd, $Rn", "",
1829 []>, Sched<[WriteVST4]> {
1831 let Inst{5-4} = Rn{5-4};
1832 let DecoderMethod = "DecodeVLDST1Instruction";
1834 multiclass VST1D4WB<bits<4> op7_4, string Dt, Operand AddrMode> {
1835 def _fixed : NLdSt<0,0b00,0b0010,op7_4, (outs GPR:$wb),
1836 (ins AddrMode:$Rn, VecListFourD:$Vd), IIC_VLD1x4u,
1837 "vst1", Dt, "$Vd, $Rn!",
1838 "$Rn.addr = $wb", []>, Sched<[WriteVST4]> {
1839 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1840 let Inst{5-4} = Rn{5-4};
1841 let DecoderMethod = "DecodeVLDST1Instruction";
1843 def _register : NLdSt<0,0b00,0b0010,op7_4, (outs GPR:$wb),
1844 (ins AddrMode:$Rn, rGPR:$Rm, VecListFourD:$Vd),
1846 "vst1", Dt, "$Vd, $Rn, $Rm",
1847 "$Rn.addr = $wb", []>, Sched<[WriteVST4]> {
1848 let Inst{5-4} = Rn{5-4};
1849 let DecoderMethod = "DecodeVLDST1Instruction";
1853 def VST1d8Q : VST1D4<{0,0,?,?}, "8", addrmode6align64or128or256>;
1854 def VST1d16Q : VST1D4<{0,1,?,?}, "16", addrmode6align64or128or256>;
1855 def VST1d32Q : VST1D4<{1,0,?,?}, "32", addrmode6align64or128or256>;
1856 def VST1d64Q : VST1D4<{1,1,?,?}, "64", addrmode6align64or128or256>;
1858 defm VST1d8Qwb : VST1D4WB<{0,0,?,?}, "8", addrmode6align64or128or256>;
1859 defm VST1d16Qwb : VST1D4WB<{0,1,?,?}, "16", addrmode6align64or128or256>;
1860 defm VST1d32Qwb : VST1D4WB<{1,0,?,?}, "32", addrmode6align64or128or256>;
1861 defm VST1d64Qwb : VST1D4WB<{1,1,?,?}, "64", addrmode6align64or128or256>;
1863 def VST1d8QPseudo : VSTQQPseudo<IIC_VST1x4>, Sched<[WriteVST4]>;
1864 def VST1d16QPseudo : VSTQQPseudo<IIC_VST1x4>, Sched<[WriteVST4]>;
1865 def VST1d32QPseudo : VSTQQPseudo<IIC_VST1x4>, Sched<[WriteVST4]>;
1866 def VST1d64QPseudo : VSTQQPseudo<IIC_VST1x4>, Sched<[WriteVST4]>;
1867 def VST1d64QPseudoWB_fixed : VSTQQWBfixedPseudo<IIC_VST1x4u>, Sched<[WriteVST4]>;
1868 def VST1d64QPseudoWB_register : VSTQQWBPseudo<IIC_VST1x4u>, Sched<[WriteVST4]>;
1870 def VST1q8HighQPseudo : VSTQQQQPseudo<IIC_VST1x4>, Sched<[WriteVST4]>;
1871 def VST1q8LowQPseudo_UPD : VSTQQQQWBPseudo<IIC_VST1x4>, Sched<[WriteVST4]>;
1872 def VST1q16HighQPseudo : VSTQQQQPseudo<IIC_VST1x4>, Sched<[WriteVST4]>;
1873 def VST1q16LowQPseudo_UPD : VSTQQQQWBPseudo<IIC_VST1x4>, Sched<[WriteVST4]>;
1874 def VST1q32HighQPseudo : VSTQQQQPseudo<IIC_VST1x4>, Sched<[WriteVST4]>;
1875 def VST1q32LowQPseudo_UPD : VSTQQQQWBPseudo<IIC_VST1x4>, Sched<[WriteVST4]>;
1876 def VST1q64HighQPseudo : VSTQQQQPseudo<IIC_VST1x4>, Sched<[WriteVST4]>;
1877 def VST1q64LowQPseudo_UPD : VSTQQQQWBPseudo<IIC_VST1x4>, Sched<[WriteVST4]>;
1879 // VST2 : Vector Store (multiple 2-element structures)
1880 class VST2<bits<4> op11_8, bits<4> op7_4, string Dt, RegisterOperand VdTy,
1881 InstrItinClass itin, Operand AddrMode>
1882 : NLdSt<0, 0b00, op11_8, op7_4, (outs), (ins AddrMode:$Rn, VdTy:$Vd),
1883 itin, "vst2", Dt, "$Vd, $Rn", "", []> {
1885 let Inst{5-4} = Rn{5-4};
1886 let DecoderMethod = "DecodeVLDST2Instruction";
1889 def VST2d8 : VST2<0b1000, {0,0,?,?}, "8", VecListDPair, IIC_VST2,
1890 addrmode6align64or128>, Sched<[WriteVST2]>;
1891 def VST2d16 : VST2<0b1000, {0,1,?,?}, "16", VecListDPair, IIC_VST2,
1892 addrmode6align64or128>, Sched<[WriteVST2]>;
1893 def VST2d32 : VST2<0b1000, {1,0,?,?}, "32", VecListDPair, IIC_VST2,
1894 addrmode6align64or128>, Sched<[WriteVST2]>;
1896 def VST2q8 : VST2<0b0011, {0,0,?,?}, "8", VecListFourD, IIC_VST2x2,
1897 addrmode6align64or128or256>, Sched<[WriteVST4]>;
1898 def VST2q16 : VST2<0b0011, {0,1,?,?}, "16", VecListFourD, IIC_VST2x2,
1899 addrmode6align64or128or256>, Sched<[WriteVST4]>;
1900 def VST2q32 : VST2<0b0011, {1,0,?,?}, "32", VecListFourD, IIC_VST2x2,
1901 addrmode6align64or128or256>, Sched<[WriteVST4]>;
1903 def VST2q8Pseudo : VSTQQPseudo<IIC_VST2x2>, Sched<[WriteVST4]>;
1904 def VST2q16Pseudo : VSTQQPseudo<IIC_VST2x2>, Sched<[WriteVST4]>;
1905 def VST2q32Pseudo : VSTQQPseudo<IIC_VST2x2>, Sched<[WriteVST4]>;
1907 // ...with address register writeback:
1908 multiclass VST2DWB<bits<4> op11_8, bits<4> op7_4, string Dt,
1909 RegisterOperand VdTy, Operand AddrMode> {
1910 def _fixed : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1911 (ins AddrMode:$Rn, VdTy:$Vd), IIC_VLD1u,
1912 "vst2", Dt, "$Vd, $Rn!",
1913 "$Rn.addr = $wb", []>, Sched<[WriteVST2]> {
1914 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1915 let Inst{5-4} = Rn{5-4};
1916 let DecoderMethod = "DecodeVLDST2Instruction";
1918 def _register : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1919 (ins AddrMode:$Rn, rGPR:$Rm, VdTy:$Vd), IIC_VLD1u,
1920 "vst2", Dt, "$Vd, $Rn, $Rm",
1921 "$Rn.addr = $wb", []>, Sched<[WriteVST2]> {
1922 let Inst{5-4} = Rn{5-4};
1923 let DecoderMethod = "DecodeVLDST2Instruction";
1926 multiclass VST2QWB<bits<4> op7_4, string Dt, Operand AddrMode> {
1927 def _fixed : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
1928 (ins AddrMode:$Rn, VecListFourD:$Vd), IIC_VLD1u,
1929 "vst2", Dt, "$Vd, $Rn!",
1930 "$Rn.addr = $wb", []>, Sched<[WriteVST4]> {
1931 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1932 let Inst{5-4} = Rn{5-4};
1933 let DecoderMethod = "DecodeVLDST2Instruction";
1935 def _register : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
1936 (ins AddrMode:$Rn, rGPR:$Rm, VecListFourD:$Vd),
1938 "vst2", Dt, "$Vd, $Rn, $Rm",
1939 "$Rn.addr = $wb", []>, Sched<[WriteVST4]> {
1940 let Inst{5-4} = Rn{5-4};
1941 let DecoderMethod = "DecodeVLDST2Instruction";
1945 defm VST2d8wb : VST2DWB<0b1000, {0,0,?,?}, "8", VecListDPair,
1946 addrmode6align64or128>;
1947 defm VST2d16wb : VST2DWB<0b1000, {0,1,?,?}, "16", VecListDPair,
1948 addrmode6align64or128>;
1949 defm VST2d32wb : VST2DWB<0b1000, {1,0,?,?}, "32", VecListDPair,
1950 addrmode6align64or128>;
1952 defm VST2q8wb : VST2QWB<{0,0,?,?}, "8", addrmode6align64or128or256>;
1953 defm VST2q16wb : VST2QWB<{0,1,?,?}, "16", addrmode6align64or128or256>;
1954 defm VST2q32wb : VST2QWB<{1,0,?,?}, "32", addrmode6align64or128or256>;
1956 def VST2q8PseudoWB_fixed : VSTQQWBfixedPseudo<IIC_VST2x2u>, Sched<[WriteVST4]>;
1957 def VST2q16PseudoWB_fixed : VSTQQWBfixedPseudo<IIC_VST2x2u>, Sched<[WriteVST4]>;
1958 def VST2q32PseudoWB_fixed : VSTQQWBfixedPseudo<IIC_VST2x2u>, Sched<[WriteVST4]>;
1959 def VST2q8PseudoWB_register : VSTQQWBregisterPseudo<IIC_VST2x2u>, Sched<[WriteVST4]>;
1960 def VST2q16PseudoWB_register : VSTQQWBregisterPseudo<IIC_VST2x2u>, Sched<[WriteVST4]>;
1961 def VST2q32PseudoWB_register : VSTQQWBregisterPseudo<IIC_VST2x2u>, Sched<[WriteVST4]>;
1963 // ...with double-spaced registers
1964 def VST2b8 : VST2<0b1001, {0,0,?,?}, "8", VecListDPairSpaced, IIC_VST2,
1965 addrmode6align64or128>;
1966 def VST2b16 : VST2<0b1001, {0,1,?,?}, "16", VecListDPairSpaced, IIC_VST2,
1967 addrmode6align64or128>;
1968 def VST2b32 : VST2<0b1001, {1,0,?,?}, "32", VecListDPairSpaced, IIC_VST2,
1969 addrmode6align64or128>;
1970 defm VST2b8wb : VST2DWB<0b1001, {0,0,?,?}, "8", VecListDPairSpaced,
1971 addrmode6align64or128>;
1972 defm VST2b16wb : VST2DWB<0b1001, {0,1,?,?}, "16", VecListDPairSpaced,
1973 addrmode6align64or128>;
1974 defm VST2b32wb : VST2DWB<0b1001, {1,0,?,?}, "32", VecListDPairSpaced,
1975 addrmode6align64or128>;
1977 // VST3 : Vector Store (multiple 3-element structures)
1978 class VST3D<bits<4> op11_8, bits<4> op7_4, string Dt>
1979 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1980 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3,
1981 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn", "", []>, Sched<[WriteVST3]> {
1983 let Inst{4} = Rn{4};
1984 let DecoderMethod = "DecodeVLDST3Instruction";
1987 def VST3d8 : VST3D<0b0100, {0,0,0,?}, "8">;
1988 def VST3d16 : VST3D<0b0100, {0,1,0,?}, "16">;
1989 def VST3d32 : VST3D<0b0100, {1,0,0,?}, "32">;
1991 def VST3d8Pseudo : VSTQQPseudo<IIC_VST3>, Sched<[WriteVST3]>;
1992 def VST3d16Pseudo : VSTQQPseudo<IIC_VST3>, Sched<[WriteVST3]>;
1993 def VST3d32Pseudo : VSTQQPseudo<IIC_VST3>, Sched<[WriteVST3]>;
1995 // ...with address register writeback:
1996 class VST3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1997 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1998 (ins addrmode6:$Rn, am6offset:$Rm,
1999 DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3u,
2000 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn$Rm",
2001 "$Rn.addr = $wb", []>, Sched<[WriteVST3]> {
2002 let Inst{4} = Rn{4};
2003 let DecoderMethod = "DecodeVLDST3Instruction";
2006 def VST3d8_UPD : VST3DWB<0b0100, {0,0,0,?}, "8">;
2007 def VST3d16_UPD : VST3DWB<0b0100, {0,1,0,?}, "16">;
2008 def VST3d32_UPD : VST3DWB<0b0100, {1,0,0,?}, "32">;
2010 def VST3d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>, Sched<[WriteVST3]>;
2011 def VST3d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>, Sched<[WriteVST3]>;
2012 def VST3d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>, Sched<[WriteVST3]>;
2014 // ...with double-spaced registers:
2015 def VST3q8 : VST3D<0b0101, {0,0,0,?}, "8">;
2016 def VST3q16 : VST3D<0b0101, {0,1,0,?}, "16">;
2017 def VST3q32 : VST3D<0b0101, {1,0,0,?}, "32">;
2018 def VST3q8_UPD : VST3DWB<0b0101, {0,0,0,?}, "8">;
2019 def VST3q16_UPD : VST3DWB<0b0101, {0,1,0,?}, "16">;
2020 def VST3q32_UPD : VST3DWB<0b0101, {1,0,0,?}, "32">;
2022 def VST3q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>, Sched<[WriteVST3]>;
2023 def VST3q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>, Sched<[WriteVST3]>;
2024 def VST3q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>, Sched<[WriteVST3]>;
2026 // ...alternate versions to be allocated odd register numbers:
2027 def VST3q8oddPseudo : VSTQQQQPseudo<IIC_VST3>, Sched<[WriteVST3]>;
2028 def VST3q16oddPseudo : VSTQQQQPseudo<IIC_VST3>, Sched<[WriteVST3]>;
2029 def VST3q32oddPseudo : VSTQQQQPseudo<IIC_VST3>, Sched<[WriteVST3]>;
2031 def VST3q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>, Sched<[WriteVST3]>;
2032 def VST3q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>, Sched<[WriteVST3]>;
2033 def VST3q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>, Sched<[WriteVST3]>;
2035 // VST4 : Vector Store (multiple 4-element structures)
2036 class VST4D<bits<4> op11_8, bits<4> op7_4, string Dt>
2037 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
2038 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
2039 IIC_VST4, "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn",
2040 "", []>, Sched<[WriteVST4]> {
2042 let Inst{5-4} = Rn{5-4};
2043 let DecoderMethod = "DecodeVLDST4Instruction";
2046 def VST4d8 : VST4D<0b0000, {0,0,?,?}, "8">;
2047 def VST4d16 : VST4D<0b0000, {0,1,?,?}, "16">;
2048 def VST4d32 : VST4D<0b0000, {1,0,?,?}, "32">;
2050 def VST4d8Pseudo : VSTQQPseudo<IIC_VST4>, Sched<[WriteVST4]>;
2051 def VST4d16Pseudo : VSTQQPseudo<IIC_VST4>, Sched<[WriteVST4]>;
2052 def VST4d32Pseudo : VSTQQPseudo<IIC_VST4>, Sched<[WriteVST4]>;
2054 // ...with address register writeback:
2055 class VST4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
2056 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
2057 (ins addrmode6:$Rn, am6offset:$Rm,
2058 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST4u,
2059 "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
2060 "$Rn.addr = $wb", []>, Sched<[WriteVST4]> {
2061 let Inst{5-4} = Rn{5-4};
2062 let DecoderMethod = "DecodeVLDST4Instruction";
2065 def VST4d8_UPD : VST4DWB<0b0000, {0,0,?,?}, "8">;
2066 def VST4d16_UPD : VST4DWB<0b0000, {0,1,?,?}, "16">;
2067 def VST4d32_UPD : VST4DWB<0b0000, {1,0,?,?}, "32">;
2069 def VST4d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>, Sched<[WriteVST4]>;
2070 def VST4d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>, Sched<[WriteVST4]>;
2071 def VST4d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>, Sched<[WriteVST4]>;
2073 // ...with double-spaced registers:
2074 def VST4q8 : VST4D<0b0001, {0,0,?,?}, "8">;
2075 def VST4q16 : VST4D<0b0001, {0,1,?,?}, "16">;
2076 def VST4q32 : VST4D<0b0001, {1,0,?,?}, "32">;
2077 def VST4q8_UPD : VST4DWB<0b0001, {0,0,?,?}, "8">;
2078 def VST4q16_UPD : VST4DWB<0b0001, {0,1,?,?}, "16">;
2079 def VST4q32_UPD : VST4DWB<0b0001, {1,0,?,?}, "32">;
2081 def VST4q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>, Sched<[WriteVST4]>;
2082 def VST4q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>, Sched<[WriteVST4]>;
2083 def VST4q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>, Sched<[WriteVST4]>;
2085 // ...alternate versions to be allocated odd register numbers:
2086 def VST4q8oddPseudo : VSTQQQQPseudo<IIC_VST4>, Sched<[WriteVST4]>;
2087 def VST4q16oddPseudo : VSTQQQQPseudo<IIC_VST4>, Sched<[WriteVST4]>;
2088 def VST4q32oddPseudo : VSTQQQQPseudo<IIC_VST4>, Sched<[WriteVST4]>;
2090 def VST4q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>, Sched<[WriteVST4]>;
2091 def VST4q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>, Sched<[WriteVST4]>;
2092 def VST4q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>, Sched<[WriteVST4]>;
2094 } // mayStore = 1, hasSideEffects = 0, hasExtraSrcRegAllocReq = 1
2096 // Classes for VST*LN pseudo-instructions with multi-register operands.
2097 // These are expanded to real instructions after register allocation.
2098 class VSTQLNPseudo<InstrItinClass itin>
2099 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
2101 class VSTQLNWBPseudo<InstrItinClass itin>
2102 : PseudoNLdSt<(outs GPR:$wb),
2103 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
2104 nohash_imm:$lane), itin, "$addr.addr = $wb">;
2105 class VSTQQLNPseudo<InstrItinClass itin>
2106 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
2108 class VSTQQLNWBPseudo<InstrItinClass itin>
2109 : PseudoNLdSt<(outs GPR:$wb),
2110 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
2111 nohash_imm:$lane), itin, "$addr.addr = $wb">;
2112 class VSTQQQQLNPseudo<InstrItinClass itin>
2113 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
2115 class VSTQQQQLNWBPseudo<InstrItinClass itin>
2116 : PseudoNLdSt<(outs GPR:$wb),
2117 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
2118 nohash_imm:$lane), itin, "$addr.addr = $wb">;
2120 // VST1LN : Vector Store (single element from one lane)
2121 class VST1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
2122 PatFrag StoreOp, SDNode ExtractOp, Operand AddrMode>
2123 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
2124 (ins AddrMode:$Rn, DPR:$Vd, nohash_imm:$lane),
2125 IIC_VST1ln, "vst1", Dt, "\\{$Vd[$lane]\\}, $Rn", "",
2126 [(StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane), AddrMode:$Rn)]>,
2127 Sched<[WriteVST1]> {
2129 let DecoderMethod = "DecodeVST1LN";
2131 class VST1QLNPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
2132 : VSTQLNPseudo<IIC_VST1ln>, Sched<[WriteVST1]> {
2133 let Pattern = [(StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
2137 def VST1LNd8 : VST1LN<0b0000, {?,?,?,0}, "8", v8i8, truncstorei8,
2138 ARMvgetlaneu, addrmode6> {
2139 let Inst{7-5} = lane{2-0};
2141 def VST1LNd16 : VST1LN<0b0100, {?,?,0,?}, "16", v4i16, truncstorei16,
2142 ARMvgetlaneu, addrmode6> {
2143 let Inst{7-6} = lane{1-0};
2144 let Inst{4} = Rn{4};
2147 def VST1LNd32 : VST1LN<0b1000, {?,0,?,?}, "32", v2i32, store, extractelt,
2149 let Inst{7} = lane{0};
2150 let Inst{5-4} = Rn{5-4};
2153 def VST1LNq8Pseudo : VST1QLNPseudo<v16i8, truncstorei8, ARMvgetlaneu>;
2154 def VST1LNq16Pseudo : VST1QLNPseudo<v8i16, truncstorei16, ARMvgetlaneu>;
2155 def VST1LNq32Pseudo : VST1QLNPseudo<v4i32, store, extractelt>;
2157 let Predicates = [HasNEON] in {
2158 def : Pat<(store (extractelt (v2f32 DPR:$src), imm:$lane), addrmode6:$addr),
2159 (VST1LNd32 addrmode6:$addr, DPR:$src, imm:$lane)>;
2160 def : Pat<(store (extractelt (v4f32 QPR:$src), imm:$lane), addrmode6:$addr),
2161 (VST1LNq32Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
2163 def : Pat<(store (extractelt (v4f16 DPR:$src), imm:$lane), addrmode6:$addr),
2164 (VST1LNd16 addrmode6:$addr, DPR:$src, imm:$lane)>;
2165 def : Pat<(store (extractelt (v8f16 QPR:$src), imm:$lane), addrmode6:$addr),
2166 (VST1LNq16Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
2169 // ...with address register writeback:
2170 class VST1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
2171 PatFrag StoreOp, SDNode ExtractOp, Operand AdrMode>
2172 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
2173 (ins AdrMode:$Rn, am6offset:$Rm,
2174 DPR:$Vd, nohash_imm:$lane), IIC_VST1lnu, "vst1", Dt,
2175 "\\{$Vd[$lane]\\}, $Rn$Rm",
2177 [(set GPR:$wb, (StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane),
2178 AdrMode:$Rn, am6offset:$Rm))]>,
2179 Sched<[WriteVST1]> {
2180 let DecoderMethod = "DecodeVST1LN";
2182 class VST1QLNWBPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
2183 : VSTQLNWBPseudo<IIC_VST1lnu>, Sched<[WriteVST1]> {
2184 let Pattern = [(set GPR:$wb, (StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
2185 addrmode6:$addr, am6offset:$offset))];
2188 def VST1LNd8_UPD : VST1LNWB<0b0000, {?,?,?,0}, "8", v8i8, post_truncsti8,
2189 ARMvgetlaneu, addrmode6> {
2190 let Inst{7-5} = lane{2-0};
2192 def VST1LNd16_UPD : VST1LNWB<0b0100, {?,?,0,?}, "16", v4i16, post_truncsti16,
2193 ARMvgetlaneu, addrmode6> {
2194 let Inst{7-6} = lane{1-0};
2195 let Inst{4} = Rn{4};
2197 def VST1LNd32_UPD : VST1LNWB<0b1000, {?,0,?,?}, "32", v2i32, post_store,
2198 extractelt, addrmode6oneL32> {
2199 let Inst{7} = lane{0};
2200 let Inst{5-4} = Rn{5-4};
2203 def VST1LNq8Pseudo_UPD : VST1QLNWBPseudo<v16i8, post_truncsti8, ARMvgetlaneu>;
2204 def VST1LNq16Pseudo_UPD : VST1QLNWBPseudo<v8i16, post_truncsti16,ARMvgetlaneu>;
2205 def VST1LNq32Pseudo_UPD : VST1QLNWBPseudo<v4i32, post_store, extractelt>;
2207 let mayStore = 1, hasSideEffects = 0, hasExtraSrcRegAllocReq = 1 in {
2209 // VST2LN : Vector Store (single 2-element structure from one lane)
2210 class VST2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
2211 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
2212 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, nohash_imm:$lane),
2213 IIC_VST2ln, "vst2", Dt, "\\{$Vd[$lane], $src2[$lane]\\}, $Rn",
2214 "", []>, Sched<[WriteVST1]> {
2216 let Inst{4} = Rn{4};
2217 let DecoderMethod = "DecodeVST2LN";
2220 def VST2LNd8 : VST2LN<0b0001, {?,?,?,?}, "8"> {
2221 let Inst{7-5} = lane{2-0};
2223 def VST2LNd16 : VST2LN<0b0101, {?,?,0,?}, "16"> {
2224 let Inst{7-6} = lane{1-0};
2226 def VST2LNd32 : VST2LN<0b1001, {?,0,0,?}, "32"> {
2227 let Inst{7} = lane{0};
2230 def VST2LNd8Pseudo : VSTQLNPseudo<IIC_VST2ln>, Sched<[WriteVST1]>;
2231 def VST2LNd16Pseudo : VSTQLNPseudo<IIC_VST2ln>, Sched<[WriteVST1]>;
2232 def VST2LNd32Pseudo : VSTQLNPseudo<IIC_VST2ln>, Sched<[WriteVST1]>;
2234 // ...with double-spaced registers:
2235 def VST2LNq16 : VST2LN<0b0101, {?,?,1,?}, "16"> {
2236 let Inst{7-6} = lane{1-0};
2237 let Inst{4} = Rn{4};
2239 def VST2LNq32 : VST2LN<0b1001, {?,1,0,?}, "32"> {
2240 let Inst{7} = lane{0};
2241 let Inst{4} = Rn{4};
2244 def VST2LNq16Pseudo : VSTQQLNPseudo<IIC_VST2ln>, Sched<[WriteVST1]>;
2245 def VST2LNq32Pseudo : VSTQQLNPseudo<IIC_VST2ln>, Sched<[WriteVST1]>;
2247 // ...with address register writeback:
2248 class VST2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
2249 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
2250 (ins addrmode6:$Rn, am6offset:$Rm,
2251 DPR:$Vd, DPR:$src2, nohash_imm:$lane), IIC_VST2lnu, "vst2", Dt,
2252 "\\{$Vd[$lane], $src2[$lane]\\}, $Rn$Rm",
2253 "$Rn.addr = $wb", []> {
2254 let Inst{4} = Rn{4};
2255 let DecoderMethod = "DecodeVST2LN";
2258 def VST2LNd8_UPD : VST2LNWB<0b0001, {?,?,?,?}, "8"> {
2259 let Inst{7-5} = lane{2-0};
2261 def VST2LNd16_UPD : VST2LNWB<0b0101, {?,?,0,?}, "16"> {
2262 let Inst{7-6} = lane{1-0};
2264 def VST2LNd32_UPD : VST2LNWB<0b1001, {?,0,0,?}, "32"> {
2265 let Inst{7} = lane{0};
2268 def VST2LNd8Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>, Sched<[WriteVST1]>;
2269 def VST2LNd16Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>, Sched<[WriteVST1]>;
2270 def VST2LNd32Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>, Sched<[WriteVST1]>;
2272 def VST2LNq16_UPD : VST2LNWB<0b0101, {?,?,1,?}, "16"> {
2273 let Inst{7-6} = lane{1-0};
2275 def VST2LNq32_UPD : VST2LNWB<0b1001, {?,1,0,?}, "32"> {
2276 let Inst{7} = lane{0};
2279 def VST2LNq16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>, Sched<[WriteVST1]>;
2280 def VST2LNq32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>, Sched<[WriteVST1]>;
2282 // VST3LN : Vector Store (single 3-element structure from one lane)
2283 class VST3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
2284 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
2285 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3,
2286 nohash_imm:$lane), IIC_VST3ln, "vst3", Dt,
2287 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn", "", []>,
2288 Sched<[WriteVST2]> {
2290 let DecoderMethod = "DecodeVST3LN";
2293 def VST3LNd8 : VST3LN<0b0010, {?,?,?,0}, "8"> {
2294 let Inst{7-5} = lane{2-0};
2296 def VST3LNd16 : VST3LN<0b0110, {?,?,0,0}, "16"> {
2297 let Inst{7-6} = lane{1-0};
2299 def VST3LNd32 : VST3LN<0b1010, {?,0,0,0}, "32"> {
2300 let Inst{7} = lane{0};
2303 def VST3LNd8Pseudo : VSTQQLNPseudo<IIC_VST3ln>, Sched<[WriteVST2]>;
2304 def VST3LNd16Pseudo : VSTQQLNPseudo<IIC_VST3ln>, Sched<[WriteVST2]>;
2305 def VST3LNd32Pseudo : VSTQQLNPseudo<IIC_VST3ln>, Sched<[WriteVST2]>;
2307 // ...with double-spaced registers:
2308 def VST3LNq16 : VST3LN<0b0110, {?,?,1,0}, "16"> {
2309 let Inst{7-6} = lane{1-0};
2311 def VST3LNq32 : VST3LN<0b1010, {?,1,0,0}, "32"> {
2312 let Inst{7} = lane{0};
2315 def VST3LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
2316 def VST3LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
2318 // ...with address register writeback:
2319 class VST3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
2320 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
2321 (ins addrmode6:$Rn, am6offset:$Rm,
2322 DPR:$Vd, DPR:$src2, DPR:$src3, nohash_imm:$lane),
2323 IIC_VST3lnu, "vst3", Dt,
2324 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn$Rm",
2325 "$Rn.addr = $wb", []> {
2326 let DecoderMethod = "DecodeVST3LN";
2329 def VST3LNd8_UPD : VST3LNWB<0b0010, {?,?,?,0}, "8"> {
2330 let Inst{7-5} = lane{2-0};
2332 def VST3LNd16_UPD : VST3LNWB<0b0110, {?,?,0,0}, "16"> {
2333 let Inst{7-6} = lane{1-0};
2335 def VST3LNd32_UPD : VST3LNWB<0b1010, {?,0,0,0}, "32"> {
2336 let Inst{7} = lane{0};
2339 def VST3LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>, Sched<[WriteVST2]>;
2340 def VST3LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>, Sched<[WriteVST2]>;
2341 def VST3LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>, Sched<[WriteVST2]>;
2343 def VST3LNq16_UPD : VST3LNWB<0b0110, {?,?,1,0}, "16"> {
2344 let Inst{7-6} = lane{1-0};
2346 def VST3LNq32_UPD : VST3LNWB<0b1010, {?,1,0,0}, "32"> {
2347 let Inst{7} = lane{0};
2350 def VST3LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>, Sched<[WriteVST2]>;
2351 def VST3LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>, Sched<[WriteVST2]>;
2353 // VST4LN : Vector Store (single 4-element structure from one lane)
2354 class VST4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
2355 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
2356 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4,
2357 nohash_imm:$lane), IIC_VST4ln, "vst4", Dt,
2358 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn",
2359 "", []>, Sched<[WriteVST2]> {
2361 let Inst{4} = Rn{4};
2362 let DecoderMethod = "DecodeVST4LN";
2365 def VST4LNd8 : VST4LN<0b0011, {?,?,?,?}, "8"> {
2366 let Inst{7-5} = lane{2-0};
2368 def VST4LNd16 : VST4LN<0b0111, {?,?,0,?}, "16"> {
2369 let Inst{7-6} = lane{1-0};
2371 def VST4LNd32 : VST4LN<0b1011, {?,0,?,?}, "32"> {
2372 let Inst{7} = lane{0};
2373 let Inst{5} = Rn{5};
2376 def VST4LNd8Pseudo : VSTQQLNPseudo<IIC_VST4ln>, Sched<[WriteVST2]>;
2377 def VST4LNd16Pseudo : VSTQQLNPseudo<IIC_VST4ln>, Sched<[WriteVST2]>;
2378 def VST4LNd32Pseudo : VSTQQLNPseudo<IIC_VST4ln>, Sched<[WriteVST2]>;
2380 // ...with double-spaced registers:
2381 def VST4LNq16 : VST4LN<0b0111, {?,?,1,?}, "16"> {
2382 let Inst{7-6} = lane{1-0};
2384 def VST4LNq32 : VST4LN<0b1011, {?,1,?,?}, "32"> {
2385 let Inst{7} = lane{0};
2386 let Inst{5} = Rn{5};
2389 def VST4LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>, Sched<[WriteVST2]>;
2390 def VST4LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>, Sched<[WriteVST2]>;
2392 // ...with address register writeback:
2393 class VST4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
2394 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
2395 (ins addrmode6:$Rn, am6offset:$Rm,
2396 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
2397 IIC_VST4lnu, "vst4", Dt,
2398 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn$Rm",
2399 "$Rn.addr = $wb", []> {
2400 let Inst{4} = Rn{4};
2401 let DecoderMethod = "DecodeVST4LN";
2404 def VST4LNd8_UPD : VST4LNWB<0b0011, {?,?,?,?}, "8"> {
2405 let Inst{7-5} = lane{2-0};
2407 def VST4LNd16_UPD : VST4LNWB<0b0111, {?,?,0,?}, "16"> {
2408 let Inst{7-6} = lane{1-0};
2410 def VST4LNd32_UPD : VST4LNWB<0b1011, {?,0,?,?}, "32"> {
2411 let Inst{7} = lane{0};
2412 let Inst{5} = Rn{5};
2415 def VST4LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>, Sched<[WriteVST2]>;
2416 def VST4LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>, Sched<[WriteVST2]>;
2417 def VST4LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>, Sched<[WriteVST2]>;
2419 def VST4LNq16_UPD : VST4LNWB<0b0111, {?,?,1,?}, "16"> {
2420 let Inst{7-6} = lane{1-0};
2422 def VST4LNq32_UPD : VST4LNWB<0b1011, {?,1,?,?}, "32"> {
2423 let Inst{7} = lane{0};
2424 let Inst{5} = Rn{5};
2427 def VST4LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>, Sched<[WriteVST2]>;
2428 def VST4LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>, Sched<[WriteVST2]>;
2430 } // mayStore = 1, hasSideEffects = 0, hasExtraSrcRegAllocReq = 1
2432 // Use vld1/vst1 for unaligned f64 load / store
2433 let Predicates = [IsLE,HasNEON] in {
2434 def : Pat<(f64 (hword_alignedload addrmode6:$addr)),
2435 (VLD1d16 addrmode6:$addr)>;
2436 def : Pat<(hword_alignedstore (f64 DPR:$value), addrmode6:$addr),
2437 (VST1d16 addrmode6:$addr, DPR:$value)>;
2438 def : Pat<(f64 (byte_alignedload addrmode6:$addr)),
2439 (VLD1d8 addrmode6:$addr)>;
2440 def : Pat<(byte_alignedstore (f64 DPR:$value), addrmode6:$addr),
2441 (VST1d8 addrmode6:$addr, DPR:$value)>;
2443 let Predicates = [IsBE,HasNEON] in {
2444 def : Pat<(f64 (non_word_alignedload addrmode6:$addr)),
2445 (VLD1d64 addrmode6:$addr)>;
2446 def : Pat<(non_word_alignedstore (f64 DPR:$value), addrmode6:$addr),
2447 (VST1d64 addrmode6:$addr, DPR:$value)>;
2450 // Use vld1/vst1 for Q and QQ. Also use them for unaligned v2f64
2451 // load / store if it's legal.
2452 let Predicates = [HasNEON] in {
2453 def : Pat<(v2f64 (dword_alignedload addrmode6:$addr)),
2454 (VLD1q64 addrmode6:$addr)>;
2455 def : Pat<(dword_alignedstore (v2f64 QPR:$value), addrmode6:$addr),
2456 (VST1q64 addrmode6:$addr, QPR:$value)>;
2458 let Predicates = [IsLE,HasNEON] in {
2459 def : Pat<(v2f64 (word_alignedload addrmode6:$addr)),
2460 (VLD1q32 addrmode6:$addr)>;
2461 def : Pat<(word_alignedstore (v2f64 QPR:$value), addrmode6:$addr),
2462 (VST1q32 addrmode6:$addr, QPR:$value)>;
2463 def : Pat<(v2f64 (hword_alignedload addrmode6:$addr)),
2464 (VLD1q16 addrmode6:$addr)>;
2465 def : Pat<(hword_alignedstore (v2f64 QPR:$value), addrmode6:$addr),
2466 (VST1q16 addrmode6:$addr, QPR:$value)>;
2467 def : Pat<(v2f64 (byte_alignedload addrmode6:$addr)),
2468 (VLD1q8 addrmode6:$addr)>;
2469 def : Pat<(byte_alignedstore (v2f64 QPR:$value), addrmode6:$addr),
2470 (VST1q8 addrmode6:$addr, QPR:$value)>;
2473 //===----------------------------------------------------------------------===//
2474 // NEON pattern fragments
2475 //===----------------------------------------------------------------------===//
2477 // Extract D sub-registers of Q registers.
2478 def DSubReg_i8_reg : SDNodeXForm<imm, [{
2479 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2480 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/8, SDLoc(N),
2483 def DSubReg_i16_reg : SDNodeXForm<imm, [{
2484 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2485 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/4, SDLoc(N),
2488 def DSubReg_i32_reg : SDNodeXForm<imm, [{
2489 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2490 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/2, SDLoc(N),
2493 def DSubReg_f64_reg : SDNodeXForm<imm, [{
2494 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2495 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue(), SDLoc(N),
2499 // Extract S sub-registers of Q/D registers.
2500 def SSubReg_f32_reg : SDNodeXForm<imm, [{
2501 assert(ARM::ssub_3 == ARM::ssub_0+3 && "Unexpected subreg numbering");
2502 return CurDAG->getTargetConstant(ARM::ssub_0 + N->getZExtValue(), SDLoc(N),
2506 // Extract S sub-registers of Q/D registers containing a given f16 lane.
2507 def SSubReg_f16_reg : SDNodeXForm<imm, [{
2508 assert(ARM::ssub_3 == ARM::ssub_0+3 && "Unexpected subreg numbering");
2509 return CurDAG->getTargetConstant(ARM::ssub_0 + N->getZExtValue()/2, SDLoc(N),
2513 // Translate lane numbers from Q registers to D subregs.
2514 def SubReg_i8_lane : SDNodeXForm<imm, [{
2515 return CurDAG->getTargetConstant(N->getZExtValue() & 7, SDLoc(N), MVT::i32);
2517 def SubReg_i16_lane : SDNodeXForm<imm, [{
2518 return CurDAG->getTargetConstant(N->getZExtValue() & 3, SDLoc(N), MVT::i32);
2520 def SubReg_i32_lane : SDNodeXForm<imm, [{
2521 return CurDAG->getTargetConstant(N->getZExtValue() & 1, SDLoc(N), MVT::i32);
2524 //===----------------------------------------------------------------------===//
2525 // Instruction Classes
2526 //===----------------------------------------------------------------------===//
2528 // Basic 2-register operations: double- and quad-register.
2529 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2530 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
2531 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
2532 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2533 (ins DPR:$Vm), IIC_VUNAD, OpcodeStr, Dt,"$Vd, $Vm", "",
2534 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm))))]>;
2535 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2536 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
2537 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
2538 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2539 (ins QPR:$Vm), IIC_VUNAQ, OpcodeStr, Dt,"$Vd, $Vm", "",
2540 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vm))))]>;
2542 // Basic 2-register intrinsics, both double- and quad-register.
2543 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2544 bits<2> op17_16, bits<5> op11_7, bit op4,
2545 InstrItinClass itin, string OpcodeStr, string Dt,
2546 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2547 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2548 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2549 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
2550 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2551 bits<2> op17_16, bits<5> op11_7, bit op4,
2552 InstrItinClass itin, string OpcodeStr, string Dt,
2553 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2554 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2555 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2556 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
2558 // Same as above, but not predicated.
2559 class N2VDIntnp<bits<2> op19_18, bits<2> op17_16, bits<3> op10_8, bit op7,
2560 InstrItinClass itin, string OpcodeStr, string Dt,
2561 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2562 : N2Vnp<op19_18, op17_16, op10_8, op7, 0, (outs DPR:$Vd), (ins DPR:$Vm),
2563 itin, OpcodeStr, Dt,
2564 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
2566 class N2VQIntnp<bits<2> op19_18, bits<2> op17_16, bits<3> op10_8, bit op7,
2567 InstrItinClass itin, string OpcodeStr, string Dt,
2568 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2569 : N2Vnp<op19_18, op17_16, op10_8, op7, 1, (outs QPR:$Vd), (ins QPR:$Vm),
2570 itin, OpcodeStr, Dt,
2571 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
2573 // Similar to NV2VQIntnp with some more encoding bits exposed (crypto).
2574 class N2VQIntXnp<bits<2> op19_18, bits<2> op17_16, bits<3> op10_8, bit op6,
2575 bit op7, InstrItinClass itin, string OpcodeStr, string Dt,
2576 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2577 : N2Vnp<op19_18, op17_16, op10_8, op7, op6, (outs QPR:$Vd), (ins QPR:$Vm),
2578 itin, OpcodeStr, Dt,
2579 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
2581 // Same as N2VQIntXnp but with Vd as a src register.
2582 class N2VQIntX2np<bits<2> op19_18, bits<2> op17_16, bits<3> op10_8, bit op6,
2583 bit op7, InstrItinClass itin, string OpcodeStr, string Dt,
2584 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2585 : N2Vnp<op19_18, op17_16, op10_8, op7, op6,
2586 (outs QPR:$Vd), (ins QPR:$src, QPR:$Vm),
2587 itin, OpcodeStr, Dt,
2588 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$src), (OpTy QPR:$Vm))))]> {
2589 let Constraints = "$src = $Vd";
2592 // Narrow 2-register operations.
2593 class N2VN<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2594 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2595 InstrItinClass itin, string OpcodeStr, string Dt,
2596 ValueType TyD, ValueType TyQ, SDNode OpNode>
2597 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$Vd),
2598 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2599 [(set DPR:$Vd, (TyD (OpNode (TyQ QPR:$Vm))))]>;
2601 // Narrow 2-register intrinsics.
2602 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2603 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2604 InstrItinClass itin, string OpcodeStr, string Dt,
2605 ValueType TyD, ValueType TyQ, SDPatternOperator IntOp>
2606 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$Vd),
2607 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2608 [(set DPR:$Vd, (TyD (IntOp (TyQ QPR:$Vm))))]>;
2610 // Long 2-register operations (currently only used for VMOVL).
2611 class N2VL<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2612 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2613 InstrItinClass itin, string OpcodeStr, string Dt,
2614 ValueType TyQ, ValueType TyD, SDNode OpNode>
2615 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$Vd),
2616 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2617 [(set QPR:$Vd, (TyQ (OpNode (TyD DPR:$Vm))))]>;
2619 // Long 2-register intrinsics.
2620 class N2VLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2621 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2622 InstrItinClass itin, string OpcodeStr, string Dt,
2623 ValueType TyQ, ValueType TyD, SDPatternOperator IntOp>
2624 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$Vd),
2625 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2626 [(set QPR:$Vd, (TyQ (IntOp (TyD DPR:$Vm))))]>;
2628 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
2629 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
2630 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$Vd, DPR:$Vm),
2631 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
2632 OpcodeStr, Dt, "$Vd, $Vm",
2633 "$src1 = $Vd, $src2 = $Vm", []>;
2634 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
2635 InstrItinClass itin, string OpcodeStr, string Dt>
2636 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$Vd, QPR:$Vm),
2637 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$Vd, $Vm",
2638 "$src1 = $Vd, $src2 = $Vm", []>;
2640 // Basic 3-register operations: double- and quad-register.
2641 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2642 InstrItinClass itin, string OpcodeStr, string Dt,
2643 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2644 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2645 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2646 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2647 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
2648 // All of these have a two-operand InstAlias.
2649 let TwoOperandAliasConstraint = "$Vn = $Vd";
2650 let isCommutable = Commutable;
2652 // Same as N3VD but no data type.
2653 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2654 InstrItinClass itin, string OpcodeStr,
2655 ValueType ResTy, ValueType OpTy,
2656 SDNode OpNode, bit Commutable>
2657 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
2658 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2659 OpcodeStr, "$Vd, $Vn, $Vm", "",
2660 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>{
2661 // All of these have a two-operand InstAlias.
2662 let TwoOperandAliasConstraint = "$Vn = $Vd";
2663 let isCommutable = Commutable;
2666 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
2667 InstrItinClass itin, string OpcodeStr, string Dt,
2668 ValueType Ty, SDNode ShOp>
2669 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2670 (outs DPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2671 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2673 (Ty (ShOp (Ty DPR:$Vn),
2674 (Ty (ARMvduplane (Ty DPR_VFP2:$Vm),imm:$lane)))))]> {
2675 // All of these have a two-operand InstAlias.
2676 let TwoOperandAliasConstraint = "$Vn = $Vd";
2677 let isCommutable = 0;
2679 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
2680 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
2681 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2682 (outs DPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2683 NVMulSLFrm, IIC_VMULi16D, OpcodeStr, Dt,"$Vd, $Vn, $Vm$lane","",
2685 (Ty (ShOp (Ty DPR:$Vn),
2686 (Ty (ARMvduplane (Ty DPR_8:$Vm), imm:$lane)))))]> {
2687 // All of these have a two-operand InstAlias.
2688 let TwoOperandAliasConstraint = "$Vn = $Vd";
2689 let isCommutable = 0;
2692 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2693 InstrItinClass itin, string OpcodeStr, string Dt,
2694 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2695 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2696 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2697 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2698 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
2699 // All of these have a two-operand InstAlias.
2700 let TwoOperandAliasConstraint = "$Vn = $Vd";
2701 let isCommutable = Commutable;
2703 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2704 InstrItinClass itin, string OpcodeStr,
2705 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2706 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
2707 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2708 OpcodeStr, "$Vd, $Vn, $Vm", "",
2709 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>{
2710 // All of these have a two-operand InstAlias.
2711 let TwoOperandAliasConstraint = "$Vn = $Vd";
2712 let isCommutable = Commutable;
2714 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
2715 InstrItinClass itin, string OpcodeStr, string Dt,
2716 ValueType ResTy, ValueType OpTy, SDNode ShOp>
2717 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2718 (outs QPR:$Vd), (ins QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2719 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2720 [(set (ResTy QPR:$Vd),
2721 (ResTy (ShOp (ResTy QPR:$Vn),
2722 (ResTy (ARMvduplane (OpTy DPR_VFP2:$Vm),
2724 // All of these have a two-operand InstAlias.
2725 let TwoOperandAliasConstraint = "$Vn = $Vd";
2726 let isCommutable = 0;
2728 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
2729 ValueType ResTy, ValueType OpTy, SDNode ShOp>
2730 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2731 (outs QPR:$Vd), (ins QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2732 NVMulSLFrm, IIC_VMULi16Q, OpcodeStr, Dt,"$Vd, $Vn, $Vm$lane", "",
2733 [(set (ResTy QPR:$Vd),
2734 (ResTy (ShOp (ResTy QPR:$Vn),
2735 (ResTy (ARMvduplane (OpTy DPR_8:$Vm),
2737 // All of these have a two-operand InstAlias.
2738 let TwoOperandAliasConstraint = "$Vn = $Vd";
2739 let isCommutable = 0;
2742 // Basic 3-register intrinsics, both double- and quad-register.
2743 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2744 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2745 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp, bit Commutable>
2746 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2747 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), f, itin,
2748 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2749 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
2750 // All of these have a two-operand InstAlias.
2751 let TwoOperandAliasConstraint = "$Vn = $Vd";
2752 let isCommutable = Commutable;
2755 class N3VDIntnp<bits<5> op27_23, bits<2> op21_20, bits<4> op11_8, bit op6,
2756 bit op4, Format f, InstrItinClass itin, string OpcodeStr,
2757 string Dt, ValueType ResTy, ValueType OpTy,
2758 SDPatternOperator IntOp, bit Commutable>
2759 : N3Vnp<op27_23, op21_20, op11_8, op6, op4,
2760 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin, OpcodeStr, Dt,
2761 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>;
2763 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2764 string OpcodeStr, string Dt, ValueType Ty, SDPatternOperator IntOp>
2765 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2766 (outs DPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2767 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2769 (Ty (IntOp (Ty DPR:$Vn),
2770 (Ty (ARMvduplane (Ty DPR_VFP2:$Vm),
2772 let isCommutable = 0;
2775 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2776 string OpcodeStr, string Dt, ValueType Ty, SDPatternOperator IntOp>
2777 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2778 (outs DPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2779 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2781 (Ty (IntOp (Ty DPR:$Vn),
2782 (Ty (ARMvduplane (Ty DPR_8:$Vm), imm:$lane)))))]> {
2783 let isCommutable = 0;
2785 class N3VDIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2786 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2787 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2788 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2789 (outs DPR:$Vd), (ins DPR:$Vm, DPR:$Vn), f, itin,
2790 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
2791 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (OpTy DPR:$Vn))))]> {
2792 let TwoOperandAliasConstraint = "$Vm = $Vd";
2793 let isCommutable = 0;
2796 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2797 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2798 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp, bit Commutable>
2799 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2800 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), f, itin,
2801 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2802 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
2803 // All of these have a two-operand InstAlias.
2804 let TwoOperandAliasConstraint = "$Vn = $Vd";
2805 let isCommutable = Commutable;
2808 class N3VQIntnp<bits<5> op27_23, bits<2> op21_20, bits<4> op11_8, bit op6,
2809 bit op4, Format f, InstrItinClass itin, string OpcodeStr,
2810 string Dt, ValueType ResTy, ValueType OpTy,
2811 SDPatternOperator IntOp, bit Commutable>
2812 : N3Vnp<op27_23, op21_20, op11_8, op6, op4,
2813 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), f, itin, OpcodeStr, Dt,
2814 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>;
2816 // Same as N3VQIntnp but with Vd as a src register.
2817 class N3VQInt3np<bits<5> op27_23, bits<2> op21_20, bits<4> op11_8, bit op6,
2818 bit op4, Format f, InstrItinClass itin, string OpcodeStr,
2819 string Dt, ValueType ResTy, ValueType OpTy,
2820 SDPatternOperator IntOp, bit Commutable>
2821 : N3Vnp<op27_23, op21_20, op11_8, op6, op4,
2822 (outs QPR:$Vd), (ins QPR:$src, QPR:$Vn, QPR:$Vm),
2823 f, itin, OpcodeStr, Dt,
2824 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$src), (OpTy QPR:$Vn),
2825 (OpTy QPR:$Vm))))]> {
2826 let Constraints = "$src = $Vd";
2829 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2830 string OpcodeStr, string Dt,
2831 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2832 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2833 (outs QPR:$Vd), (ins QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2834 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2835 [(set (ResTy QPR:$Vd),
2836 (ResTy (IntOp (ResTy QPR:$Vn),
2837 (ResTy (ARMvduplane (OpTy DPR_VFP2:$Vm),
2839 let isCommutable = 0;
2841 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2842 string OpcodeStr, string Dt,
2843 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2844 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2845 (outs QPR:$Vd), (ins QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2846 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2847 [(set (ResTy QPR:$Vd),
2848 (ResTy (IntOp (ResTy QPR:$Vn),
2849 (ResTy (ARMvduplane (OpTy DPR_8:$Vm),
2851 let isCommutable = 0;
2853 class N3VQIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2854 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2855 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2856 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2857 (outs QPR:$Vd), (ins QPR:$Vm, QPR:$Vn), f, itin,
2858 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
2859 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (OpTy QPR:$Vn))))]> {
2860 let TwoOperandAliasConstraint = "$Vm = $Vd";
2861 let isCommutable = 0;
2864 // Multiply-Add/Sub operations: double- and quad-register.
2865 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2866 InstrItinClass itin, string OpcodeStr, string Dt,
2867 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator OpNode>
2868 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2869 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2870 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2871 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
2872 (Ty (MulOp DPR:$Vn, DPR:$Vm)))))]>;
2874 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2875 string OpcodeStr, string Dt,
2876 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator ShOp>
2877 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2879 (ins DPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2881 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2883 (Ty (ShOp (Ty DPR:$src1),
2885 (Ty (ARMvduplane (Ty DPR_VFP2:$Vm),
2887 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2888 string OpcodeStr, string Dt,
2889 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator ShOp>
2890 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2892 (ins DPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2894 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2896 (Ty (ShOp (Ty DPR:$src1),
2898 (Ty (ARMvduplane (Ty DPR_8:$Vm),
2901 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2902 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
2903 SDPatternOperator MulOp, SDPatternOperator OpNode>
2904 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2905 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2906 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2907 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
2908 (Ty (MulOp QPR:$Vn, QPR:$Vm)))))]>;
2909 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2910 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2911 SDPatternOperator MulOp, SDPatternOperator ShOp>
2912 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2914 (ins QPR:$src1, QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2916 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2917 [(set (ResTy QPR:$Vd),
2918 (ResTy (ShOp (ResTy QPR:$src1),
2919 (ResTy (MulOp QPR:$Vn,
2920 (ResTy (ARMvduplane (OpTy DPR_VFP2:$Vm),
2922 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2923 string OpcodeStr, string Dt,
2924 ValueType ResTy, ValueType OpTy,
2925 SDPatternOperator MulOp, SDPatternOperator ShOp>
2926 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2928 (ins QPR:$src1, QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2930 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2931 [(set (ResTy QPR:$Vd),
2932 (ResTy (ShOp (ResTy QPR:$src1),
2933 (ResTy (MulOp QPR:$Vn,
2934 (ResTy (ARMvduplane (OpTy DPR_8:$Vm),
2937 // Neon Intrinsic-Op instructions (VABA): double- and quad-register.
2938 class N3VDIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2939 InstrItinClass itin, string OpcodeStr, string Dt,
2940 ValueType Ty, SDPatternOperator IntOp, SDNode OpNode>
2941 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2942 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2943 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2944 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
2945 (Ty (IntOp (Ty DPR:$Vn), (Ty DPR:$Vm))))))]>;
2946 class N3VQIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2947 InstrItinClass itin, string OpcodeStr, string Dt,
2948 ValueType Ty, SDPatternOperator IntOp, SDNode OpNode>
2949 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2950 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2951 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2952 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
2953 (Ty (IntOp (Ty QPR:$Vn), (Ty QPR:$Vm))))))]>;
2955 // Neon 3-argument intrinsics, both double- and quad-register.
2956 // The destination register is also used as the first source operand register.
2957 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2958 InstrItinClass itin, string OpcodeStr, string Dt,
2959 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2960 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2961 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2962 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2963 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$src1),
2964 (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>;
2965 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2966 InstrItinClass itin, string OpcodeStr, string Dt,
2967 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2968 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2969 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2970 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2971 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$src1),
2972 (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>;
2974 // Long Multiply-Add/Sub operations.
2975 class N3VLMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2976 InstrItinClass itin, string OpcodeStr, string Dt,
2977 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2978 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2979 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2980 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2981 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
2982 (TyQ (MulOp (TyD DPR:$Vn),
2983 (TyD DPR:$Vm)))))]>;
2984 class N3VLMulOpSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2985 InstrItinClass itin, string OpcodeStr, string Dt,
2986 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2987 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$Vd),
2988 (ins QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2990 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2992 (OpNode (TyQ QPR:$src1),
2993 (TyQ (MulOp (TyD DPR:$Vn),
2994 (TyD (ARMvduplane (TyD DPR_VFP2:$Vm),
2996 class N3VLMulOpSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2997 InstrItinClass itin, string OpcodeStr, string Dt,
2998 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2999 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$Vd),
3000 (ins QPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
3002 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
3004 (OpNode (TyQ QPR:$src1),
3005 (TyQ (MulOp (TyD DPR:$Vn),
3006 (TyD (ARMvduplane (TyD DPR_8:$Vm),
3009 // Long Intrinsic-Op vector operations with explicit extend (VABAL).
3010 class N3VLIntExtOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
3011 InstrItinClass itin, string OpcodeStr, string Dt,
3012 ValueType TyQ, ValueType TyD, SDPatternOperator IntOp, SDNode ExtOp,
3014 : N3V<op24, op23, op21_20, op11_8, 0, op4,
3015 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
3016 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
3017 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
3018 (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
3019 (TyD DPR:$Vm)))))))]>;
3021 // Neon Long 3-argument intrinsic. The destination register is
3022 // a quad-register and is also used as the first source operand register.
3023 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
3024 InstrItinClass itin, string OpcodeStr, string Dt,
3025 ValueType TyQ, ValueType TyD, SDPatternOperator IntOp>
3026 : N3V<op24, op23, op21_20, op11_8, 0, op4,
3027 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
3028 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
3030 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$Vn), (TyD DPR:$Vm))))]>;
3031 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
3032 string OpcodeStr, string Dt,
3033 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
3034 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
3036 (ins QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
3038 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
3039 [(set (ResTy QPR:$Vd),
3040 (ResTy (IntOp (ResTy QPR:$src1),
3042 (OpTy (ARMvduplane (OpTy DPR_VFP2:$Vm),
3044 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
3045 InstrItinClass itin, string OpcodeStr, string Dt,
3046 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
3047 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
3049 (ins QPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
3051 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
3052 [(set (ResTy QPR:$Vd),
3053 (ResTy (IntOp (ResTy QPR:$src1),
3055 (OpTy (ARMvduplane (OpTy DPR_8:$Vm),
3058 // Narrowing 3-register intrinsics.
3059 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
3060 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
3061 SDPatternOperator IntOp, bit Commutable>
3062 : N3V<op24, op23, op21_20, op11_8, 0, op4,
3063 (outs DPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINi4D,
3064 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
3065 [(set DPR:$Vd, (TyD (IntOp (TyQ QPR:$Vn), (TyQ QPR:$Vm))))]> {
3066 let isCommutable = Commutable;
3069 // Long 3-register operations.
3070 class N3VL<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
3071 InstrItinClass itin, string OpcodeStr, string Dt,
3072 ValueType TyQ, ValueType TyD, SDNode OpNode, bit Commutable>
3073 : N3V<op24, op23, op21_20, op11_8, 0, op4,
3074 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
3075 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
3076 [(set QPR:$Vd, (TyQ (OpNode (TyD DPR:$Vn), (TyD DPR:$Vm))))]> {
3077 let isCommutable = Commutable;
3080 class N3VLSL<bit op24, bits<2> op21_20, bits<4> op11_8,
3081 InstrItinClass itin, string OpcodeStr, string Dt,
3082 ValueType TyQ, ValueType TyD, SDNode OpNode>
3083 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
3084 (outs QPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
3085 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
3087 (TyQ (OpNode (TyD DPR:$Vn),
3088 (TyD (ARMvduplane (TyD DPR_VFP2:$Vm),imm:$lane)))))]>;
3089 class N3VLSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
3090 InstrItinClass itin, string OpcodeStr, string Dt,
3091 ValueType TyQ, ValueType TyD, SDNode OpNode>
3092 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
3093 (outs QPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
3094 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
3096 (TyQ (OpNode (TyD DPR:$Vn),
3097 (TyD (ARMvduplane (TyD DPR_8:$Vm), imm:$lane)))))]>;
3099 // Long 3-register operations with explicitly extended operands.
3100 class N3VLExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
3101 InstrItinClass itin, string OpcodeStr, string Dt,
3102 ValueType TyQ, ValueType TyD, SDNode OpNode, SDNode ExtOp,
3104 : N3V<op24, op23, op21_20, op11_8, 0, op4,
3105 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
3106 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
3107 [(set QPR:$Vd, (OpNode (TyQ (ExtOp (TyD DPR:$Vn))),
3108 (TyQ (ExtOp (TyD DPR:$Vm)))))]> {
3109 let isCommutable = Commutable;
3112 // Long 3-register intrinsics with explicit extend (VABDL).
3113 class N3VLIntExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
3114 InstrItinClass itin, string OpcodeStr, string Dt,
3115 ValueType TyQ, ValueType TyD, SDPatternOperator IntOp, SDNode ExtOp,
3117 : N3V<op24, op23, op21_20, op11_8, 0, op4,
3118 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
3119 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
3120 [(set QPR:$Vd, (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
3121 (TyD DPR:$Vm))))))]> {
3122 let isCommutable = Commutable;
3125 // Long 3-register intrinsics.
3126 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
3127 InstrItinClass itin, string OpcodeStr, string Dt,
3128 ValueType TyQ, ValueType TyD, SDPatternOperator IntOp, bit Commutable>
3129 : N3V<op24, op23, op21_20, op11_8, 0, op4,
3130 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
3131 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
3132 [(set QPR:$Vd, (TyQ (IntOp (TyD DPR:$Vn), (TyD DPR:$Vm))))]> {
3133 let isCommutable = Commutable;
3136 // Same as above, but not predicated.
3137 class N3VLIntnp<bits<5> op27_23, bits<2> op21_20, bits<4> op11_8, bit op6,
3138 bit op4, InstrItinClass itin, string OpcodeStr,
3139 string Dt, ValueType ResTy, ValueType OpTy,
3140 SDPatternOperator IntOp, bit Commutable>
3141 : N3Vnp<op27_23, op21_20, op11_8, op6, op4,
3142 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin, OpcodeStr, Dt,
3143 [(set QPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>;
3145 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
3146 string OpcodeStr, string Dt,
3147 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
3148 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
3149 (outs QPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
3150 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
3151 [(set (ResTy QPR:$Vd),
3152 (ResTy (IntOp (OpTy DPR:$Vn),
3153 (OpTy (ARMvduplane (OpTy DPR_VFP2:$Vm),
3155 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
3156 InstrItinClass itin, string OpcodeStr, string Dt,
3157 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
3158 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
3159 (outs QPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
3160 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
3161 [(set (ResTy QPR:$Vd),
3162 (ResTy (IntOp (OpTy DPR:$Vn),
3163 (OpTy (ARMvduplane (OpTy DPR_8:$Vm),
3166 // Wide 3-register operations.
3167 class N3VW<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
3168 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
3169 SDNode OpNode, SDNode ExtOp, bit Commutable>
3170 : N3V<op24, op23, op21_20, op11_8, 0, op4,
3171 (outs QPR:$Vd), (ins QPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VSUBiD,
3172 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
3173 [(set QPR:$Vd, (OpNode (TyQ QPR:$Vn),
3174 (TyQ (ExtOp (TyD DPR:$Vm)))))]> {
3175 // All of these have a two-operand InstAlias.
3176 let TwoOperandAliasConstraint = "$Vn = $Vd";
3177 let isCommutable = Commutable;
3180 // Pairwise long 2-register intrinsics, both double- and quad-register.
3181 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
3182 bits<2> op17_16, bits<5> op11_7, bit op4,
3183 string OpcodeStr, string Dt,
3184 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
3185 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
3186 (ins DPR:$Vm), IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
3187 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
3188 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
3189 bits<2> op17_16, bits<5> op11_7, bit op4,
3190 string OpcodeStr, string Dt,
3191 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
3192 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
3193 (ins QPR:$Vm), IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
3194 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
3196 // Pairwise long 2-register accumulate intrinsics,
3197 // both double- and quad-register.
3198 // The destination register is also used as the first source operand register.
3199 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
3200 bits<2> op17_16, bits<5> op11_7, bit op4,
3201 string OpcodeStr, string Dt,
3202 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
3203 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
3204 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vm), IIC_VPALiD,
3205 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
3206 [(set DPR:$Vd, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$Vm))))]>;
3207 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
3208 bits<2> op17_16, bits<5> op11_7, bit op4,
3209 string OpcodeStr, string Dt,
3210 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
3211 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
3212 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vm), IIC_VPALiQ,
3213 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
3214 [(set QPR:$Vd, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$Vm))))]>;
3216 // Shift by immediate,
3217 // both double- and quad-register.
3218 let TwoOperandAliasConstraint = "$Vm = $Vd" in {
3219 class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
3220 Format f, InstrItinClass itin, Operand ImmTy,
3221 string OpcodeStr, string Dt, ValueType Ty, SDNode OpNode>
3222 : N2VImm<op24, op23, op11_8, op7, 0, op4,
3223 (outs DPR:$Vd), (ins DPR:$Vm, ImmTy:$SIMM), f, itin,
3224 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
3225 [(set DPR:$Vd, (Ty (OpNode (Ty DPR:$Vm), (i32 imm:$SIMM))))]>;
3226 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
3227 Format f, InstrItinClass itin, Operand ImmTy,
3228 string OpcodeStr, string Dt, ValueType Ty, SDNode OpNode>
3229 : N2VImm<op24, op23, op11_8, op7, 1, op4,
3230 (outs QPR:$Vd), (ins QPR:$Vm, ImmTy:$SIMM), f, itin,
3231 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
3232 [(set QPR:$Vd, (Ty (OpNode (Ty QPR:$Vm), (i32 imm:$SIMM))))]>;
3235 // Long shift by immediate.
3236 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
3237 string OpcodeStr, string Dt,
3238 ValueType ResTy, ValueType OpTy, Operand ImmTy,
3239 SDPatternOperator OpNode>
3240 : N2VImm<op24, op23, op11_8, op7, op6, op4,
3241 (outs QPR:$Vd), (ins DPR:$Vm, ImmTy:$SIMM), N2RegVShLFrm,
3242 IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
3243 [(set QPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm), ImmTy:$SIMM)))]>;
3245 // Narrow shift by immediate.
3246 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
3247 InstrItinClass itin, string OpcodeStr, string Dt,
3248 ValueType ResTy, ValueType OpTy, Operand ImmTy,
3249 SDPatternOperator OpNode>
3250 : N2VImm<op24, op23, op11_8, op7, op6, op4,
3251 (outs DPR:$Vd), (ins QPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, itin,
3252 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
3253 [(set DPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vm),
3254 (i32 ImmTy:$SIMM))))]>;
3256 // Shift right by immediate and accumulate,
3257 // both double- and quad-register.
3258 let TwoOperandAliasConstraint = "$Vm = $Vd" in {
3259 class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
3260 Operand ImmTy, string OpcodeStr, string Dt,
3261 ValueType Ty, SDNode ShOp>
3262 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
3263 (ins DPR:$src1, DPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, IIC_VPALiD,
3264 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
3265 [(set DPR:$Vd, (Ty (add DPR:$src1,
3266 (Ty (ShOp DPR:$Vm, (i32 imm:$SIMM))))))]>;
3267 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
3268 Operand ImmTy, string OpcodeStr, string Dt,
3269 ValueType Ty, SDNode ShOp>
3270 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
3271 (ins QPR:$src1, QPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, IIC_VPALiD,
3272 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
3273 [(set QPR:$Vd, (Ty (add QPR:$src1,
3274 (Ty (ShOp QPR:$Vm, (i32 imm:$SIMM))))))]>;
3277 // Shift by immediate and insert,
3278 // both double- and quad-register.
3279 let TwoOperandAliasConstraint = "$Vm = $Vd" in {
3280 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
3281 Operand ImmTy, Format f, string OpcodeStr, string Dt,
3282 ValueType Ty,SDNode ShOp>
3283 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
3284 (ins DPR:$src1, DPR:$Vm, ImmTy:$SIMM), f, IIC_VSHLiD,
3285 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
3286 [(set DPR:$Vd, (Ty (ShOp DPR:$src1, DPR:$Vm, (i32 imm:$SIMM))))]>;
3287 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
3288 Operand ImmTy, Format f, string OpcodeStr, string Dt,
3289 ValueType Ty,SDNode ShOp>
3290 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
3291 (ins QPR:$src1, QPR:$Vm, ImmTy:$SIMM), f, IIC_VSHLiQ,
3292 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
3293 [(set QPR:$Vd, (Ty (ShOp QPR:$src1, QPR:$Vm, (i32 imm:$SIMM))))]>;
3296 // Convert, with fractional bits immediate,
3297 // both double- and quad-register.
3298 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
3299 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
3300 SDPatternOperator IntOp>
3301 : N2VImm<op24, op23, op11_8, op7, 0, op4,
3302 (outs DPR:$Vd), (ins DPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
3303 IIC_VUNAD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
3304 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (i32 imm:$SIMM))))]>;
3305 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
3306 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
3307 SDPatternOperator IntOp>
3308 : N2VImm<op24, op23, op11_8, op7, 1, op4,
3309 (outs QPR:$Vd), (ins QPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
3310 IIC_VUNAQ, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
3311 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (i32 imm:$SIMM))))]>;
3313 //===----------------------------------------------------------------------===//
3315 //===----------------------------------------------------------------------===//
3317 // Abbreviations used in multiclass suffixes:
3318 // Q = quarter int (8 bit) elements
3319 // H = half int (16 bit) elements
3320 // S = single int (32 bit) elements
3321 // D = double int (64 bit) elements
3323 // Neon 2-register vector operations and intrinsics.
3325 // Neon 2-register comparisons.
3326 // source operand element sizes of 8, 16 and 32 bits:
3327 multiclass N2V_QHS_cmp<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3328 bits<5> op11_7, bit op4, string opc, string Dt,
3329 string asm, SDNode OpNode> {
3330 // 64-bit vector types.
3331 def v8i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 0, op4,
3332 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
3333 opc, !strconcat(Dt, "8"), asm, "",
3334 [(set DPR:$Vd, (v8i8 (OpNode (v8i8 DPR:$Vm))))]>;
3335 def v4i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
3336 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
3337 opc, !strconcat(Dt, "16"), asm, "",
3338 [(set DPR:$Vd, (v4i16 (OpNode (v4i16 DPR:$Vm))))]>;
3339 def v2i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
3340 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
3341 opc, !strconcat(Dt, "32"), asm, "",
3342 [(set DPR:$Vd, (v2i32 (OpNode (v2i32 DPR:$Vm))))]>;
3343 def v2f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
3344 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
3345 opc, "f32", asm, "",
3346 [(set DPR:$Vd, (v2i32 (OpNode (v2f32 DPR:$Vm))))]> {
3347 let Inst{10} = 1; // overwrite F = 1
3349 def v4f16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
3350 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
3351 opc, "f16", asm, "",
3352 [(set DPR:$Vd, (v4i16 (OpNode (v4f16 DPR:$Vm))))]>,
3353 Requires<[HasNEON,HasFullFP16]> {
3354 let Inst{10} = 1; // overwrite F = 1
3357 // 128-bit vector types.
3358 def v16i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 1, op4,
3359 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
3360 opc, !strconcat(Dt, "8"), asm, "",
3361 [(set QPR:$Vd, (v16i8 (OpNode (v16i8 QPR:$Vm))))]>;
3362 def v8i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
3363 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
3364 opc, !strconcat(Dt, "16"), asm, "",
3365 [(set QPR:$Vd, (v8i16 (OpNode (v8i16 QPR:$Vm))))]>;
3366 def v4i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
3367 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
3368 opc, !strconcat(Dt, "32"), asm, "",
3369 [(set QPR:$Vd, (v4i32 (OpNode (v4i32 QPR:$Vm))))]>;
3370 def v4f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
3371 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
3372 opc, "f32", asm, "",
3373 [(set QPR:$Vd, (v4i32 (OpNode (v4f32 QPR:$Vm))))]> {
3374 let Inst{10} = 1; // overwrite F = 1
3376 def v8f16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
3377 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
3378 opc, "f16", asm, "",
3379 [(set QPR:$Vd, (v8i16 (OpNode (v8f16 QPR:$Vm))))]>,
3380 Requires<[HasNEON,HasFullFP16]> {
3381 let Inst{10} = 1; // overwrite F = 1
3386 // Neon 2-register vector intrinsics,
3387 // element sizes of 8, 16 and 32 bits:
3388 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3389 bits<5> op11_7, bit op4,
3390 InstrItinClass itinD, InstrItinClass itinQ,
3391 string OpcodeStr, string Dt, SDPatternOperator IntOp> {
3392 // 64-bit vector types.
3393 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3394 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
3395 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3396 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
3397 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3398 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
3400 // 128-bit vector types.
3401 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3402 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
3403 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3404 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
3405 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3406 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
3410 // Neon Narrowing 2-register vector operations,
3411 // source operand element sizes of 16, 32 and 64 bits:
3412 multiclass N2VN_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3413 bits<5> op11_7, bit op6, bit op4,
3414 InstrItinClass itin, string OpcodeStr, string Dt,
3416 def v8i8 : N2VN<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
3417 itin, OpcodeStr, !strconcat(Dt, "16"),
3418 v8i8, v8i16, OpNode>;
3419 def v4i16 : N2VN<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
3420 itin, OpcodeStr, !strconcat(Dt, "32"),
3421 v4i16, v4i32, OpNode>;
3422 def v2i32 : N2VN<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
3423 itin, OpcodeStr, !strconcat(Dt, "64"),
3424 v2i32, v2i64, OpNode>;
3427 // Neon Narrowing 2-register vector intrinsics,
3428 // source operand element sizes of 16, 32 and 64 bits:
3429 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3430 bits<5> op11_7, bit op6, bit op4,
3431 InstrItinClass itin, string OpcodeStr, string Dt,
3432 SDPatternOperator IntOp> {
3433 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
3434 itin, OpcodeStr, !strconcat(Dt, "16"),
3435 v8i8, v8i16, IntOp>;
3436 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
3437 itin, OpcodeStr, !strconcat(Dt, "32"),
3438 v4i16, v4i32, IntOp>;
3439 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
3440 itin, OpcodeStr, !strconcat(Dt, "64"),
3441 v2i32, v2i64, IntOp>;
3445 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
3446 // source operand element sizes of 16, 32 and 64 bits:
3447 multiclass N2VL_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
3448 string OpcodeStr, string Dt, SDNode OpNode> {
3449 def v8i16 : N2VL<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
3450 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode>;
3451 def v4i32 : N2VL<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
3452 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
3453 def v2i64 : N2VL<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
3454 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
3458 // Neon 3-register vector operations.
3460 // First with only element sizes of 8, 16 and 32 bits:
3461 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3462 InstrItinClass itinD16, InstrItinClass itinD32,
3463 InstrItinClass itinQ16, InstrItinClass itinQ32,
3464 string OpcodeStr, string Dt,
3465 SDNode OpNode, bit Commutable = 0> {
3466 // 64-bit vector types.
3467 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
3468 OpcodeStr, !strconcat(Dt, "8"),
3469 v8i8, v8i8, OpNode, Commutable>;
3470 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
3471 OpcodeStr, !strconcat(Dt, "16"),
3472 v4i16, v4i16, OpNode, Commutable>;
3473 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
3474 OpcodeStr, !strconcat(Dt, "32"),
3475 v2i32, v2i32, OpNode, Commutable>;
3477 // 128-bit vector types.
3478 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
3479 OpcodeStr, !strconcat(Dt, "8"),
3480 v16i8, v16i8, OpNode, Commutable>;
3481 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
3482 OpcodeStr, !strconcat(Dt, "16"),
3483 v8i16, v8i16, OpNode, Commutable>;
3484 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
3485 OpcodeStr, !strconcat(Dt, "32"),
3486 v4i32, v4i32, OpNode, Commutable>;
3489 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, SDNode ShOp> {
3490 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, "i16", v4i16, ShOp>;
3491 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, "i32", v2i32, ShOp>;
3492 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, "i16", v8i16, v4i16, ShOp>;
3493 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, "i32",
3494 v4i32, v2i32, ShOp>;
3497 // ....then also with element size 64 bits:
3498 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3499 InstrItinClass itinD, InstrItinClass itinQ,
3500 string OpcodeStr, string Dt,
3501 SDNode OpNode, bit Commutable = 0>
3502 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
3503 OpcodeStr, Dt, OpNode, Commutable> {
3504 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
3505 OpcodeStr, !strconcat(Dt, "64"),
3506 v1i64, v1i64, OpNode, Commutable>;
3507 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
3508 OpcodeStr, !strconcat(Dt, "64"),
3509 v2i64, v2i64, OpNode, Commutable>;
3513 // Neon 3-register vector intrinsics.
3515 // First with only element sizes of 16 and 32 bits:
3516 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3517 InstrItinClass itinD16, InstrItinClass itinD32,
3518 InstrItinClass itinQ16, InstrItinClass itinQ32,
3519 string OpcodeStr, string Dt,
3520 SDPatternOperator IntOp, bit Commutable = 0> {
3521 // 64-bit vector types.
3522 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, f, itinD16,
3523 OpcodeStr, !strconcat(Dt, "16"),
3524 v4i16, v4i16, IntOp, Commutable>;
3525 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, f, itinD32,
3526 OpcodeStr, !strconcat(Dt, "32"),
3527 v2i32, v2i32, IntOp, Commutable>;
3529 // 128-bit vector types.
3530 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, f, itinQ16,
3531 OpcodeStr, !strconcat(Dt, "16"),
3532 v8i16, v8i16, IntOp, Commutable>;
3533 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, f, itinQ32,
3534 OpcodeStr, !strconcat(Dt, "32"),
3535 v4i32, v4i32, IntOp, Commutable>;
3537 multiclass N3VInt_HSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3538 InstrItinClass itinD16, InstrItinClass itinD32,
3539 InstrItinClass itinQ16, InstrItinClass itinQ32,
3540 string OpcodeStr, string Dt,
3541 SDPatternOperator IntOp> {
3542 // 64-bit vector types.
3543 def v4i16 : N3VDIntSh<op24, op23, 0b01, op11_8, op4, f, itinD16,
3544 OpcodeStr, !strconcat(Dt, "16"),
3545 v4i16, v4i16, IntOp>;
3546 def v2i32 : N3VDIntSh<op24, op23, 0b10, op11_8, op4, f, itinD32,
3547 OpcodeStr, !strconcat(Dt, "32"),
3548 v2i32, v2i32, IntOp>;
3550 // 128-bit vector types.
3551 def v8i16 : N3VQIntSh<op24, op23, 0b01, op11_8, op4, f, itinQ16,
3552 OpcodeStr, !strconcat(Dt, "16"),
3553 v8i16, v8i16, IntOp>;
3554 def v4i32 : N3VQIntSh<op24, op23, 0b10, op11_8, op4, f, itinQ32,
3555 OpcodeStr, !strconcat(Dt, "32"),
3556 v4i32, v4i32, IntOp>;
3559 multiclass N3VIntSL_HS<bits<4> op11_8,
3560 InstrItinClass itinD16, InstrItinClass itinD32,
3561 InstrItinClass itinQ16, InstrItinClass itinQ32,
3562 string OpcodeStr, string Dt, SDPatternOperator IntOp> {
3563 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
3564 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
3565 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
3566 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
3567 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
3568 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
3569 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
3570 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
3573 // ....then also with element size of 8 bits:
3574 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3575 InstrItinClass itinD16, InstrItinClass itinD32,
3576 InstrItinClass itinQ16, InstrItinClass itinQ32,
3577 string OpcodeStr, string Dt,
3578 SDPatternOperator IntOp, bit Commutable = 0>
3579 : N3VInt_HS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3580 OpcodeStr, Dt, IntOp, Commutable> {
3581 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, f, itinD16,
3582 OpcodeStr, !strconcat(Dt, "8"),
3583 v8i8, v8i8, IntOp, Commutable>;
3584 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, f, itinQ16,
3585 OpcodeStr, !strconcat(Dt, "8"),
3586 v16i8, v16i8, IntOp, Commutable>;
3588 multiclass N3VInt_QHSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3589 InstrItinClass itinD16, InstrItinClass itinD32,
3590 InstrItinClass itinQ16, InstrItinClass itinQ32,
3591 string OpcodeStr, string Dt,
3592 SDPatternOperator IntOp>
3593 : N3VInt_HSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3594 OpcodeStr, Dt, IntOp> {
3595 def v8i8 : N3VDIntSh<op24, op23, 0b00, op11_8, op4, f, itinD16,
3596 OpcodeStr, !strconcat(Dt, "8"),
3598 def v16i8 : N3VQIntSh<op24, op23, 0b00, op11_8, op4, f, itinQ16,
3599 OpcodeStr, !strconcat(Dt, "8"),
3600 v16i8, v16i8, IntOp>;
3604 // ....then also with element size of 64 bits:
3605 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3606 InstrItinClass itinD16, InstrItinClass itinD32,
3607 InstrItinClass itinQ16, InstrItinClass itinQ32,
3608 string OpcodeStr, string Dt,
3609 SDPatternOperator IntOp, bit Commutable = 0>
3610 : N3VInt_QHS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3611 OpcodeStr, Dt, IntOp, Commutable> {
3612 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, f, itinD32,
3613 OpcodeStr, !strconcat(Dt, "64"),
3614 v1i64, v1i64, IntOp, Commutable>;
3615 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, f, itinQ32,
3616 OpcodeStr, !strconcat(Dt, "64"),
3617 v2i64, v2i64, IntOp, Commutable>;
3619 multiclass N3VInt_QHSDSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3620 InstrItinClass itinD16, InstrItinClass itinD32,
3621 InstrItinClass itinQ16, InstrItinClass itinQ32,
3622 string OpcodeStr, string Dt,
3623 SDPatternOperator IntOp>
3624 : N3VInt_QHSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3625 OpcodeStr, Dt, IntOp> {
3626 def v1i64 : N3VDIntSh<op24, op23, 0b11, op11_8, op4, f, itinD32,
3627 OpcodeStr, !strconcat(Dt, "64"),
3628 v1i64, v1i64, IntOp>;
3629 def v2i64 : N3VQIntSh<op24, op23, 0b11, op11_8, op4, f, itinQ32,
3630 OpcodeStr, !strconcat(Dt, "64"),
3631 v2i64, v2i64, IntOp>;
3634 // Neon Narrowing 3-register vector intrinsics,
3635 // source operand element sizes of 16, 32 and 64 bits:
3636 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3637 string OpcodeStr, string Dt,
3638 SDPatternOperator IntOp, bit Commutable = 0> {
3639 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
3640 OpcodeStr, !strconcat(Dt, "16"),
3641 v8i8, v8i16, IntOp, Commutable>;
3642 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
3643 OpcodeStr, !strconcat(Dt, "32"),
3644 v4i16, v4i32, IntOp, Commutable>;
3645 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
3646 OpcodeStr, !strconcat(Dt, "64"),
3647 v2i32, v2i64, IntOp, Commutable>;
3651 // Neon Long 3-register vector operations.
3653 multiclass N3VL_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3654 InstrItinClass itin16, InstrItinClass itin32,
3655 string OpcodeStr, string Dt,
3656 SDNode OpNode, bit Commutable = 0> {
3657 def v8i16 : N3VL<op24, op23, 0b00, op11_8, op4, itin16,
3658 OpcodeStr, !strconcat(Dt, "8"),
3659 v8i16, v8i8, OpNode, Commutable>;
3660 def v4i32 : N3VL<op24, op23, 0b01, op11_8, op4, itin16,
3661 OpcodeStr, !strconcat(Dt, "16"),
3662 v4i32, v4i16, OpNode, Commutable>;
3663 def v2i64 : N3VL<op24, op23, 0b10, op11_8, op4, itin32,
3664 OpcodeStr, !strconcat(Dt, "32"),
3665 v2i64, v2i32, OpNode, Commutable>;
3668 multiclass N3VLSL_HS<bit op24, bits<4> op11_8,
3669 InstrItinClass itin, string OpcodeStr, string Dt,
3671 def v4i16 : N3VLSL16<op24, 0b01, op11_8, itin, OpcodeStr,
3672 !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
3673 def v2i32 : N3VLSL<op24, 0b10, op11_8, itin, OpcodeStr,
3674 !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
3677 multiclass N3VLExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3678 InstrItinClass itin16, InstrItinClass itin32,
3679 string OpcodeStr, string Dt,
3680 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
3681 def v8i16 : N3VLExt<op24, op23, 0b00, op11_8, op4, itin16,
3682 OpcodeStr, !strconcat(Dt, "8"),
3683 v8i16, v8i8, OpNode, ExtOp, Commutable>;
3684 def v4i32 : N3VLExt<op24, op23, 0b01, op11_8, op4, itin16,
3685 OpcodeStr, !strconcat(Dt, "16"),
3686 v4i32, v4i16, OpNode, ExtOp, Commutable>;
3687 def v2i64 : N3VLExt<op24, op23, 0b10, op11_8, op4, itin32,
3688 OpcodeStr, !strconcat(Dt, "32"),
3689 v2i64, v2i32, OpNode, ExtOp, Commutable>;
3692 // Neon Long 3-register vector intrinsics.
3694 // First with only element sizes of 16 and 32 bits:
3695 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
3696 InstrItinClass itin16, InstrItinClass itin32,
3697 string OpcodeStr, string Dt,
3698 SDPatternOperator IntOp, bit Commutable = 0> {
3699 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin16,
3700 OpcodeStr, !strconcat(Dt, "16"),
3701 v4i32, v4i16, IntOp, Commutable>;
3702 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin32,
3703 OpcodeStr, !strconcat(Dt, "32"),
3704 v2i64, v2i32, IntOp, Commutable>;
3707 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
3708 InstrItinClass itin, string OpcodeStr, string Dt,
3709 SDPatternOperator IntOp> {
3710 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
3711 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
3712 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
3713 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3716 // ....then also with element size of 8 bits:
3717 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3718 InstrItinClass itin16, InstrItinClass itin32,
3719 string OpcodeStr, string Dt,
3720 SDPatternOperator IntOp, bit Commutable = 0>
3721 : N3VLInt_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt,
3722 IntOp, Commutable> {
3723 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin16,
3724 OpcodeStr, !strconcat(Dt, "8"),
3725 v8i16, v8i8, IntOp, Commutable>;
3728 // ....with explicit extend (VABDL).
3729 multiclass N3VLIntExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3730 InstrItinClass itin, string OpcodeStr, string Dt,
3731 SDPatternOperator IntOp, SDNode ExtOp, bit Commutable = 0> {
3732 def v8i16 : N3VLIntExt<op24, op23, 0b00, op11_8, op4, itin,
3733 OpcodeStr, !strconcat(Dt, "8"),
3734 v8i16, v8i8, IntOp, ExtOp, Commutable>;
3735 def v4i32 : N3VLIntExt<op24, op23, 0b01, op11_8, op4, itin,
3736 OpcodeStr, !strconcat(Dt, "16"),
3737 v4i32, v4i16, IntOp, ExtOp, Commutable>;
3738 def v2i64 : N3VLIntExt<op24, op23, 0b10, op11_8, op4, itin,
3739 OpcodeStr, !strconcat(Dt, "32"),
3740 v2i64, v2i32, IntOp, ExtOp, Commutable>;
3744 // Neon Wide 3-register vector intrinsics,
3745 // source operand element sizes of 8, 16 and 32 bits:
3746 multiclass N3VW_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3747 string OpcodeStr, string Dt,
3748 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
3749 def v8i16 : N3VW<op24, op23, 0b00, op11_8, op4,
3750 OpcodeStr, !strconcat(Dt, "8"),
3751 v8i16, v8i8, OpNode, ExtOp, Commutable>;
3752 def v4i32 : N3VW<op24, op23, 0b01, op11_8, op4,
3753 OpcodeStr, !strconcat(Dt, "16"),
3754 v4i32, v4i16, OpNode, ExtOp, Commutable>;
3755 def v2i64 : N3VW<op24, op23, 0b10, op11_8, op4,
3756 OpcodeStr, !strconcat(Dt, "32"),
3757 v2i64, v2i32, OpNode, ExtOp, Commutable>;
3761 // Neon Multiply-Op vector operations,
3762 // element sizes of 8, 16 and 32 bits:
3763 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3764 InstrItinClass itinD16, InstrItinClass itinD32,
3765 InstrItinClass itinQ16, InstrItinClass itinQ32,
3766 string OpcodeStr, string Dt, SDNode OpNode> {
3767 // 64-bit vector types.
3768 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
3769 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
3770 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
3771 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
3772 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
3773 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
3775 // 128-bit vector types.
3776 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
3777 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
3778 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
3779 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
3780 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
3781 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
3784 multiclass N3VMulOpSL_HS<bits<4> op11_8,
3785 InstrItinClass itinD16, InstrItinClass itinD32,
3786 InstrItinClass itinQ16, InstrItinClass itinQ32,
3787 string OpcodeStr, string Dt, SDPatternOperator ShOp> {
3788 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
3789 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
3790 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
3791 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
3792 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
3793 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
3795 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
3796 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
3800 // Neon Intrinsic-Op vector operations,
3801 // element sizes of 8, 16 and 32 bits:
3802 multiclass N3VIntOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3803 InstrItinClass itinD, InstrItinClass itinQ,
3804 string OpcodeStr, string Dt, SDPatternOperator IntOp,
3806 // 64-bit vector types.
3807 def v8i8 : N3VDIntOp<op24, op23, 0b00, op11_8, op4, itinD,
3808 OpcodeStr, !strconcat(Dt, "8"), v8i8, IntOp, OpNode>;
3809 def v4i16 : N3VDIntOp<op24, op23, 0b01, op11_8, op4, itinD,
3810 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp, OpNode>;
3811 def v2i32 : N3VDIntOp<op24, op23, 0b10, op11_8, op4, itinD,
3812 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp, OpNode>;
3814 // 128-bit vector types.
3815 def v16i8 : N3VQIntOp<op24, op23, 0b00, op11_8, op4, itinQ,
3816 OpcodeStr, !strconcat(Dt, "8"), v16i8, IntOp, OpNode>;
3817 def v8i16 : N3VQIntOp<op24, op23, 0b01, op11_8, op4, itinQ,
3818 OpcodeStr, !strconcat(Dt, "16"), v8i16, IntOp, OpNode>;
3819 def v4i32 : N3VQIntOp<op24, op23, 0b10, op11_8, op4, itinQ,
3820 OpcodeStr, !strconcat(Dt, "32"), v4i32, IntOp, OpNode>;
3823 // Neon 3-argument intrinsics,
3824 // element sizes of 16 and 32 bits:
3825 multiclass N3VInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
3826 InstrItinClass itinD16, InstrItinClass itinD32,
3827 InstrItinClass itinQ16, InstrItinClass itinQ32,
3828 string OpcodeStr, string Dt, SDPatternOperator IntOp> {
3829 // 64-bit vector types.
3830 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, itinD16,
3831 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
3832 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, itinD32,
3833 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
3835 // 128-bit vector types.
3836 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, itinQ16,
3837 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
3838 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, itinQ32,
3839 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
3842 // element sizes of 8, 16 and 32 bits:
3843 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3844 InstrItinClass itinD16, InstrItinClass itinD32,
3845 InstrItinClass itinQ16, InstrItinClass itinQ32,
3846 string OpcodeStr, string Dt, SDPatternOperator IntOp>
3847 :N3VInt3_HS <op24, op23, op11_8, op4, itinD16, itinD32,
3848 itinQ16, itinQ32, OpcodeStr, Dt, IntOp>{
3849 // 64-bit vector types.
3850 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, itinD16,
3851 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
3852 // 128-bit vector types.
3853 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, itinQ16,
3854 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
3857 // Neon Long Multiply-Op vector operations,
3858 // element sizes of 8, 16 and 32 bits:
3859 multiclass N3VLMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3860 InstrItinClass itin16, InstrItinClass itin32,
3861 string OpcodeStr, string Dt, SDNode MulOp,
3863 def v8i16 : N3VLMulOp<op24, op23, 0b00, op11_8, op4, itin16, OpcodeStr,
3864 !strconcat(Dt, "8"), v8i16, v8i8, MulOp, OpNode>;
3865 def v4i32 : N3VLMulOp<op24, op23, 0b01, op11_8, op4, itin16, OpcodeStr,
3866 !strconcat(Dt, "16"), v4i32, v4i16, MulOp, OpNode>;
3867 def v2i64 : N3VLMulOp<op24, op23, 0b10, op11_8, op4, itin32, OpcodeStr,
3868 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
3871 multiclass N3VLMulOpSL_HS<bit op24, bits<4> op11_8, string OpcodeStr,
3872 string Dt, SDNode MulOp, SDNode OpNode> {
3873 def v4i16 : N3VLMulOpSL16<op24, 0b01, op11_8, IIC_VMACi16D, OpcodeStr,
3874 !strconcat(Dt,"16"), v4i32, v4i16, MulOp, OpNode>;
3875 def v2i32 : N3VLMulOpSL<op24, 0b10, op11_8, IIC_VMACi32D, OpcodeStr,
3876 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
3880 // Neon Long 3-argument intrinsics.
3882 // First with only element sizes of 16 and 32 bits:
3883 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
3884 InstrItinClass itin16, InstrItinClass itin32,
3885 string OpcodeStr, string Dt, SDPatternOperator IntOp> {
3886 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, itin16,
3887 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
3888 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, itin32,
3889 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3892 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
3893 string OpcodeStr, string Dt, SDPatternOperator IntOp> {
3894 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
3895 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
3896 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
3897 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3900 // ....then also with element size of 8 bits:
3901 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3902 InstrItinClass itin16, InstrItinClass itin32,
3903 string OpcodeStr, string Dt, SDPatternOperator IntOp>
3904 : N3VLInt3_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt, IntOp> {
3905 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, itin16,
3906 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
3909 // ....with explicit extend (VABAL).
3910 multiclass N3VLIntExtOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3911 InstrItinClass itin, string OpcodeStr, string Dt,
3912 SDPatternOperator IntOp, SDNode ExtOp, SDNode OpNode> {
3913 def v8i16 : N3VLIntExtOp<op24, op23, 0b00, op11_8, op4, itin,
3914 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8,
3915 IntOp, ExtOp, OpNode>;
3916 def v4i32 : N3VLIntExtOp<op24, op23, 0b01, op11_8, op4, itin,
3917 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16,
3918 IntOp, ExtOp, OpNode>;
3919 def v2i64 : N3VLIntExtOp<op24, op23, 0b10, op11_8, op4, itin,
3920 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32,
3921 IntOp, ExtOp, OpNode>;
3925 // Neon Pairwise long 2-register intrinsics,
3926 // element sizes of 8, 16 and 32 bits:
3927 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3928 bits<5> op11_7, bit op4,
3929 string OpcodeStr, string Dt, SDPatternOperator IntOp> {
3930 // 64-bit vector types.
3931 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3932 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
3933 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3934 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
3935 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3936 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
3938 // 128-bit vector types.
3939 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3940 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
3941 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3942 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
3943 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3944 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
3948 // Neon Pairwise long 2-register accumulate intrinsics,
3949 // element sizes of 8, 16 and 32 bits:
3950 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3951 bits<5> op11_7, bit op4,
3952 string OpcodeStr, string Dt, SDPatternOperator IntOp> {
3953 // 64-bit vector types.
3954 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3955 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
3956 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3957 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
3958 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3959 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
3961 // 128-bit vector types.
3962 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3963 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
3964 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3965 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
3966 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3967 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
3971 // Neon 2-register vector shift by immediate,
3972 // with f of either N2RegVShLFrm or N2RegVShRFrm
3973 // element sizes of 8, 16, 32 and 64 bits:
3974 multiclass N2VShL_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3975 InstrItinClass itin, string OpcodeStr, string Dt,
3977 // 64-bit vector types.
3978 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3979 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
3980 let Inst{21-19} = 0b001; // imm6 = 001xxx
3982 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3983 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
3984 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3986 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3987 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
3988 let Inst{21} = 0b1; // imm6 = 1xxxxx
3990 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, N2RegVShLFrm, itin, i32imm,
3991 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
3994 // 128-bit vector types.
3995 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3996 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
3997 let Inst{21-19} = 0b001; // imm6 = 001xxx
3999 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
4000 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
4001 let Inst{21-20} = 0b01; // imm6 = 01xxxx
4003 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
4004 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
4005 let Inst{21} = 0b1; // imm6 = 1xxxxx
4007 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, N2RegVShLFrm, itin, i32imm,
4008 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
4011 multiclass N2VShR_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
4012 InstrItinClass itin, string OpcodeStr, string Dt,
4013 string baseOpc, SDNode OpNode> {
4014 // 64-bit vector types.
4015 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm8,
4016 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
4017 let Inst{21-19} = 0b001; // imm6 = 001xxx
4019 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm16,
4020 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
4021 let Inst{21-20} = 0b01; // imm6 = 01xxxx
4023 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm32,
4024 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
4025 let Inst{21} = 0b1; // imm6 = 1xxxxx
4027 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, N2RegVShRFrm, itin, shr_imm64,
4028 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
4031 // 128-bit vector types.
4032 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm8,
4033 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
4034 let Inst{21-19} = 0b001; // imm6 = 001xxx
4036 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm16,
4037 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
4038 let Inst{21-20} = 0b01; // imm6 = 01xxxx
4040 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm32,
4041 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
4042 let Inst{21} = 0b1; // imm6 = 1xxxxx
4044 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, N2RegVShRFrm, itin, shr_imm64,
4045 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
4049 // Neon Shift-Accumulate vector operations,
4050 // element sizes of 8, 16, 32 and 64 bits:
4051 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
4052 string OpcodeStr, string Dt, SDNode ShOp> {
4053 // 64-bit vector types.
4054 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm8,
4055 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
4056 let Inst{21-19} = 0b001; // imm6 = 001xxx
4058 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm16,
4059 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
4060 let Inst{21-20} = 0b01; // imm6 = 01xxxx
4062 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm32,
4063 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
4064 let Inst{21} = 0b1; // imm6 = 1xxxxx
4066 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4, shr_imm64,
4067 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
4070 // 128-bit vector types.
4071 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm8,
4072 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
4073 let Inst{21-19} = 0b001; // imm6 = 001xxx
4075 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm16,
4076 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
4077 let Inst{21-20} = 0b01; // imm6 = 01xxxx
4079 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm32,
4080 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
4081 let Inst{21} = 0b1; // imm6 = 1xxxxx
4083 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4, shr_imm64,
4084 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
4088 // Neon Shift-Insert vector operations,
4089 // with f of either N2RegVShLFrm or N2RegVShRFrm
4090 // element sizes of 8, 16, 32 and 64 bits:
4091 multiclass N2VShInsL_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
4093 // 64-bit vector types.
4094 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
4095 N2RegVShLFrm, OpcodeStr, "8", v8i8, NEONvsliImm> {
4096 let Inst{21-19} = 0b001; // imm6 = 001xxx
4098 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
4099 N2RegVShLFrm, OpcodeStr, "16", v4i16, NEONvsliImm> {
4100 let Inst{21-20} = 0b01; // imm6 = 01xxxx
4102 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
4103 N2RegVShLFrm, OpcodeStr, "32", v2i32, NEONvsliImm> {
4104 let Inst{21} = 0b1; // imm6 = 1xxxxx
4106 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4, i32imm,
4107 N2RegVShLFrm, OpcodeStr, "64", v1i64, NEONvsliImm>;
4110 // 128-bit vector types.
4111 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
4112 N2RegVShLFrm, OpcodeStr, "8", v16i8, NEONvsliImm> {
4113 let Inst{21-19} = 0b001; // imm6 = 001xxx
4115 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
4116 N2RegVShLFrm, OpcodeStr, "16", v8i16, NEONvsliImm> {
4117 let Inst{21-20} = 0b01; // imm6 = 01xxxx
4119 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
4120 N2RegVShLFrm, OpcodeStr, "32", v4i32, NEONvsliImm> {
4121 let Inst{21} = 0b1; // imm6 = 1xxxxx
4123 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4, i32imm,
4124 N2RegVShLFrm, OpcodeStr, "64", v2i64, NEONvsliImm>;
4127 multiclass N2VShInsR_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
4129 // 64-bit vector types.
4130 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm8,
4131 N2RegVShRFrm, OpcodeStr, "8", v8i8, NEONvsriImm> {
4132 let Inst{21-19} = 0b001; // imm6 = 001xxx
4134 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm16,
4135 N2RegVShRFrm, OpcodeStr, "16", v4i16, NEONvsriImm> {
4136 let Inst{21-20} = 0b01; // imm6 = 01xxxx
4138 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm32,
4139 N2RegVShRFrm, OpcodeStr, "32", v2i32, NEONvsriImm> {
4140 let Inst{21} = 0b1; // imm6 = 1xxxxx
4142 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4, shr_imm64,
4143 N2RegVShRFrm, OpcodeStr, "64", v1i64, NEONvsriImm>;
4146 // 128-bit vector types.
4147 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm8,
4148 N2RegVShRFrm, OpcodeStr, "8", v16i8, NEONvsriImm> {
4149 let Inst{21-19} = 0b001; // imm6 = 001xxx
4151 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm16,
4152 N2RegVShRFrm, OpcodeStr, "16", v8i16, NEONvsriImm> {
4153 let Inst{21-20} = 0b01; // imm6 = 01xxxx
4155 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm32,
4156 N2RegVShRFrm, OpcodeStr, "32", v4i32, NEONvsriImm> {
4157 let Inst{21} = 0b1; // imm6 = 1xxxxx
4159 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4, shr_imm64,
4160 N2RegVShRFrm, OpcodeStr, "64", v2i64, NEONvsriImm>;
4164 // Neon Shift Long operations,
4165 // element sizes of 8, 16, 32 bits:
4166 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
4167 bit op4, string OpcodeStr, string Dt,
4168 SDPatternOperator OpNode> {
4169 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
4170 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, imm1_7, OpNode> {
4171 let Inst{21-19} = 0b001; // imm6 = 001xxx
4173 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
4174 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, imm1_15, OpNode> {
4175 let Inst{21-20} = 0b01; // imm6 = 01xxxx
4177 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
4178 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, imm1_31, OpNode> {
4179 let Inst{21} = 0b1; // imm6 = 1xxxxx
4183 // Neon Shift Narrow operations,
4184 // element sizes of 16, 32, 64 bits:
4185 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
4186 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
4187 SDPatternOperator OpNode> {
4188 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
4189 OpcodeStr, !strconcat(Dt, "16"),
4190 v8i8, v8i16, shr_imm8, OpNode> {
4191 let Inst{21-19} = 0b001; // imm6 = 001xxx
4193 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
4194 OpcodeStr, !strconcat(Dt, "32"),
4195 v4i16, v4i32, shr_imm16, OpNode> {
4196 let Inst{21-20} = 0b01; // imm6 = 01xxxx
4198 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
4199 OpcodeStr, !strconcat(Dt, "64"),
4200 v2i32, v2i64, shr_imm32, OpNode> {
4201 let Inst{21} = 0b1; // imm6 = 1xxxxx
4205 //===----------------------------------------------------------------------===//
4206 // Instruction Definitions.
4207 //===----------------------------------------------------------------------===//
4209 // Vector Add Operations.
4211 // VADD : Vector Add (integer and floating-point)
4212 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
4214 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
4215 v2f32, v2f32, fadd, 1>;
4216 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
4217 v4f32, v4f32, fadd, 1>;
4218 def VADDhd : N3VD<0, 0, 0b01, 0b1101, 0, IIC_VBIND, "vadd", "f16",
4219 v4f16, v4f16, fadd, 1>,
4220 Requires<[HasNEON,HasFullFP16]>;
4221 def VADDhq : N3VQ<0, 0, 0b01, 0b1101, 0, IIC_VBINQ, "vadd", "f16",
4222 v8f16, v8f16, fadd, 1>,
4223 Requires<[HasNEON,HasFullFP16]>;
4224 // VADDL : Vector Add Long (Q = D + D)
4225 defm VADDLs : N3VLExt_QHS<0,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
4226 "vaddl", "s", add, sext, 1>;
4227 defm VADDLu : N3VLExt_QHS<1,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
4228 "vaddl", "u", add, zext, 1>;
4229 // VADDW : Vector Add Wide (Q = Q + D)
4230 defm VADDWs : N3VW_QHS<0,1,0b0001,0, "vaddw", "s", add, sext, 0>;
4231 defm VADDWu : N3VW_QHS<1,1,0b0001,0, "vaddw", "u", add, zext, 0>;
4232 // VHADD : Vector Halving Add
4233 defm VHADDs : N3VInt_QHS<0, 0, 0b0000, 0, N3RegFrm,
4234 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
4235 "vhadd", "s", int_arm_neon_vhadds, 1>;
4236 defm VHADDu : N3VInt_QHS<1, 0, 0b0000, 0, N3RegFrm,
4237 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
4238 "vhadd", "u", int_arm_neon_vhaddu, 1>;
4239 // VRHADD : Vector Rounding Halving Add
4240 defm VRHADDs : N3VInt_QHS<0, 0, 0b0001, 0, N3RegFrm,
4241 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
4242 "vrhadd", "s", int_arm_neon_vrhadds, 1>;
4243 defm VRHADDu : N3VInt_QHS<1, 0, 0b0001, 0, N3RegFrm,
4244 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
4245 "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
4246 // VQADD : Vector Saturating Add
4247 defm VQADDs : N3VInt_QHSD<0, 0, 0b0000, 1, N3RegFrm,
4248 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
4249 "vqadd", "s", int_arm_neon_vqadds, 1>;
4250 defm VQADDu : N3VInt_QHSD<1, 0, 0b0000, 1, N3RegFrm,
4251 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
4252 "vqadd", "u", int_arm_neon_vqaddu, 1>;
4253 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
4254 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i", null_frag, 1>;
4255 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
4256 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
4257 int_arm_neon_vraddhn, 1>;
4259 let Predicates = [HasNEON] in {
4260 def : Pat<(v8i8 (trunc (ARMvshruImm (add (v8i16 QPR:$Vn), QPR:$Vm), 8))),
4261 (VADDHNv8i8 QPR:$Vn, QPR:$Vm)>;
4262 def : Pat<(v4i16 (trunc (ARMvshruImm (add (v4i32 QPR:$Vn), QPR:$Vm), 16))),
4263 (VADDHNv4i16 QPR:$Vn, QPR:$Vm)>;
4264 def : Pat<(v2i32 (trunc (ARMvshruImm (add (v2i64 QPR:$Vn), QPR:$Vm), 32))),
4265 (VADDHNv2i32 QPR:$Vn, QPR:$Vm)>;
4268 // Vector Multiply Operations.
4270 // VMUL : Vector Multiply (integer, polynomial and floating-point)
4271 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
4272 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
4273 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16D, "vmul",
4274 "p8", v8i8, v8i8, int_arm_neon_vmulp, 1>;
4275 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16Q, "vmul",
4276 "p8", v16i8, v16i8, int_arm_neon_vmulp, 1>;
4277 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VFMULD, "vmul", "f32",
4278 v2f32, v2f32, fmul, 1>;
4279 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VFMULQ, "vmul", "f32",
4280 v4f32, v4f32, fmul, 1>;
4281 def VMULhd : N3VD<1, 0, 0b01, 0b1101, 1, IIC_VFMULD, "vmul", "f16",
4282 v4f16, v4f16, fmul, 1>,
4283 Requires<[HasNEON,HasFullFP16]>;
4284 def VMULhq : N3VQ<1, 0, 0b01, 0b1101, 1, IIC_VFMULQ, "vmul", "f16",
4285 v8f16, v8f16, fmul, 1>,
4286 Requires<[HasNEON,HasFullFP16]>;
4287 defm VMULsl : N3VSL_HS<0b1000, "vmul", mul>;
4288 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
4289 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
4291 def VMULslhd : N3VDSL16<0b01, 0b1001, "vmul", "f16", v4f16, fmul>,
4292 Requires<[HasNEON,HasFullFP16]>;
4293 def VMULslhq : N3VQSL16<0b01, 0b1001, "vmul", "f16", v8f16,
4295 Requires<[HasNEON,HasFullFP16]>;
4297 let Predicates = [HasNEON] in {
4298 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
4299 (v8i16 (ARMvduplane (v8i16 QPR:$src2), imm:$lane)))),
4300 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
4301 (v4i16 (EXTRACT_SUBREG QPR:$src2,
4302 (DSubReg_i16_reg imm:$lane))),
4303 (SubReg_i16_lane imm:$lane)))>;
4304 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
4305 (v4i32 (ARMvduplane (v4i32 QPR:$src2), imm:$lane)))),
4306 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
4307 (v2i32 (EXTRACT_SUBREG QPR:$src2,
4308 (DSubReg_i32_reg imm:$lane))),
4309 (SubReg_i32_lane imm:$lane)))>;
4310 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
4311 (v4f32 (ARMvduplane (v4f32 QPR:$src2), imm:$lane)))),
4312 (v4f32 (VMULslfq (v4f32 QPR:$src1),
4313 (v2f32 (EXTRACT_SUBREG QPR:$src2,
4314 (DSubReg_i32_reg imm:$lane))),
4315 (SubReg_i32_lane imm:$lane)))>;
4316 def : Pat<(v8f16 (fmul (v8f16 QPR:$src1),
4317 (v8f16 (ARMvduplane (v8f16 QPR:$src2), imm:$lane)))),
4318 (v8f16 (VMULslhq(v8f16 QPR:$src1),
4319 (v4f16 (EXTRACT_SUBREG QPR:$src2,
4320 (DSubReg_i16_reg imm:$lane))),
4321 (SubReg_i16_lane imm:$lane)))>;
4323 def : Pat<(v2f32 (fmul DPR:$Rn, (ARMvdup (f32 SPR:$Rm)))),
4325 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$Rm, ssub_0),
4327 def : Pat<(v4f16 (fmul DPR:$Rn, (ARMvdup (f16 HPR:$Rm)))),
4329 (INSERT_SUBREG (v4f16 (IMPLICIT_DEF)), HPR:$Rm, ssub_0),
4331 def : Pat<(v4f32 (fmul QPR:$Rn, (ARMvdup (f32 SPR:$Rm)))),
4333 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$Rm, ssub_0),
4335 def : Pat<(v8f16 (fmul QPR:$Rn, (ARMvdup (f16 HPR:$Rm)))),
4337 (INSERT_SUBREG (v4f16 (IMPLICIT_DEF)), HPR:$Rm, ssub_0),
4341 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
4342 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, N3RegFrm, IIC_VMULi16D, IIC_VMULi32D,
4343 IIC_VMULi16Q, IIC_VMULi32Q,
4344 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
4345 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
4346 IIC_VMULi16Q, IIC_VMULi32Q,
4347 "vqdmulh", "s", int_arm_neon_vqdmulh>;
4349 let Predicates = [HasNEON] in {
4350 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
4351 (v8i16 (ARMvduplane (v8i16 QPR:$src2),
4353 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
4354 (v4i16 (EXTRACT_SUBREG QPR:$src2,
4355 (DSubReg_i16_reg imm:$lane))),
4356 (SubReg_i16_lane imm:$lane)))>;
4357 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
4358 (v4i32 (ARMvduplane (v4i32 QPR:$src2),
4360 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
4361 (v2i32 (EXTRACT_SUBREG QPR:$src2,
4362 (DSubReg_i32_reg imm:$lane))),
4363 (SubReg_i32_lane imm:$lane)))>;
4366 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
4367 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, N3RegFrm,
4368 IIC_VMULi16D,IIC_VMULi32D,IIC_VMULi16Q,IIC_VMULi32Q,
4369 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
4370 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
4371 IIC_VMULi16Q, IIC_VMULi32Q,
4372 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
4374 let Predicates = [HasNEON] in {
4375 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
4376 (v8i16 (ARMvduplane (v8i16 QPR:$src2),
4378 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
4379 (v4i16 (EXTRACT_SUBREG QPR:$src2,
4380 (DSubReg_i16_reg imm:$lane))),
4381 (SubReg_i16_lane imm:$lane)))>;
4382 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
4383 (v4i32 (ARMvduplane (v4i32 QPR:$src2),
4385 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
4386 (v2i32 (EXTRACT_SUBREG QPR:$src2,
4387 (DSubReg_i32_reg imm:$lane))),
4388 (SubReg_i32_lane imm:$lane)))>;
4391 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
4392 let PostEncoderMethod = "NEONThumb2DataIPostEncoder",
4393 DecoderNamespace = "NEONData" in {
4394 defm VMULLs : N3VL_QHS<0,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
4395 "vmull", "s", NEONvmulls, 1>;
4396 defm VMULLu : N3VL_QHS<1,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
4397 "vmull", "u", NEONvmullu, 1>;
4398 def VMULLp8 : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
4399 v8i16, v8i8, int_arm_neon_vmullp, 1>;
4400 def VMULLp64 : N3VLIntnp<0b00101, 0b10, 0b1110, 0, 0, NoItinerary,
4401 "vmull", "p64", v2i64, v1i64, int_arm_neon_vmullp, 1>,
4402 Requires<[HasV8, HasCrypto]>;
4404 defm VMULLsls : N3VLSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s", NEONvmulls>;
4405 defm VMULLslu : N3VLSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u", NEONvmullu>;
4407 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
4408 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, IIC_VMULi32D,
4409 "vqdmull", "s", int_arm_neon_vqdmull, 1>;
4410 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D,
4411 "vqdmull", "s", int_arm_neon_vqdmull>;
4413 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
4415 // VMLA : Vector Multiply Accumulate (integer and floating-point)
4416 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
4417 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
4418 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
4419 v2f32, fmul_su, fadd_mlx>,
4420 Requires<[HasNEON, UseFPVMLx]>;
4421 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
4422 v4f32, fmul_su, fadd_mlx>,
4423 Requires<[HasNEON, UseFPVMLx]>;
4424 def VMLAhd : N3VDMulOp<0, 0, 0b01, 0b1101, 1, IIC_VMACD, "vmla", "f16",
4425 v4f16, fmul_su, fadd_mlx>,
4426 Requires<[HasNEON, HasFullFP16, UseFPVMLx]>;
4427 def VMLAhq : N3VQMulOp<0, 0, 0b01, 0b1101, 1, IIC_VMACQ, "vmla", "f16",
4428 v8f16, fmul_su, fadd_mlx>,
4429 Requires<[HasNEON, HasFullFP16, UseFPVMLx]>;
4430 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
4431 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
4432 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
4433 v2f32, fmul_su, fadd_mlx>,
4434 Requires<[HasNEON, UseFPVMLx]>;
4435 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
4436 v4f32, v2f32, fmul_su, fadd_mlx>,
4437 Requires<[HasNEON, UseFPVMLx]>;
4438 def VMLAslhd : N3VDMulOpSL16<0b01, 0b0001, IIC_VMACD, "vmla", "f16",
4440 Requires<[HasNEON, HasFullFP16, UseFPVMLx]>;
4441 def VMLAslhq : N3VQMulOpSL16<0b01, 0b0001, IIC_VMACQ, "vmla", "f16",
4442 v8f16, v4f16, fmul, fadd>,
4443 Requires<[HasNEON, HasFullFP16, UseFPVMLx]>;
4445 let Predicates = [HasNEON] in {
4446 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
4447 (mul (v8i16 QPR:$src2),
4448 (v8i16 (ARMvduplane (v8i16 QPR:$src3), imm:$lane))))),
4449 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
4450 (v4i16 (EXTRACT_SUBREG QPR:$src3,
4451 (DSubReg_i16_reg imm:$lane))),
4452 (SubReg_i16_lane imm:$lane)))>;
4454 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
4455 (mul (v4i32 QPR:$src2),
4456 (v4i32 (ARMvduplane (v4i32 QPR:$src3), imm:$lane))))),
4457 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
4458 (v2i32 (EXTRACT_SUBREG QPR:$src3,
4459 (DSubReg_i32_reg imm:$lane))),
4460 (SubReg_i32_lane imm:$lane)))>;
4463 def : Pat<(v4f32 (fadd_mlx (v4f32 QPR:$src1),
4464 (fmul_su (v4f32 QPR:$src2),
4465 (v4f32 (ARMvduplane (v4f32 QPR:$src3), imm:$lane))))),
4466 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
4468 (v2f32 (EXTRACT_SUBREG QPR:$src3,
4469 (DSubReg_i32_reg imm:$lane))),
4470 (SubReg_i32_lane imm:$lane)))>,
4471 Requires<[HasNEON, UseFPVMLx]>;
4473 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
4474 defm VMLALs : N3VLMulOp_QHS<0,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
4475 "vmlal", "s", NEONvmulls, add>;
4476 defm VMLALu : N3VLMulOp_QHS<1,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
4477 "vmlal", "u", NEONvmullu, add>;
4479 defm VMLALsls : N3VLMulOpSL_HS<0, 0b0010, "vmlal", "s", NEONvmulls, add>;
4480 defm VMLALslu : N3VLMulOpSL_HS<1, 0b0010, "vmlal", "u", NEONvmullu, add>;
4482 let Predicates = [HasNEON, HasV8_1a] in {
4483 // v8.1a Neon Rounding Double Multiply-Op vector operations,
4484 // VQRDMLAH : Vector Saturating Rounding Doubling Multiply Accumulate Long
4486 defm VQRDMLAH : N3VInt3_HS<1, 0, 0b1011, 1, IIC_VMACi16D, IIC_VMACi32D,
4487 IIC_VMACi16Q, IIC_VMACi32Q, "vqrdmlah", "s",
4489 def : Pat<(v4i16 (int_arm_neon_vqadds
4491 (v4i16 (int_arm_neon_vqrdmulh (v4i16 DPR:$Vn),
4492 (v4i16 DPR:$Vm))))),
4493 (v4i16 (VQRDMLAHv4i16 DPR:$src1, DPR:$Vn, DPR:$Vm))>;
4494 def : Pat<(v2i32 (int_arm_neon_vqadds
4496 (v2i32 (int_arm_neon_vqrdmulh (v2i32 DPR:$Vn),
4497 (v2i32 DPR:$Vm))))),
4498 (v2i32 (VQRDMLAHv2i32 DPR:$src1, DPR:$Vn, DPR:$Vm))>;
4499 def : Pat<(v8i16 (int_arm_neon_vqadds
4501 (v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$Vn),
4502 (v8i16 QPR:$Vm))))),
4503 (v8i16 (VQRDMLAHv8i16 QPR:$src1, QPR:$Vn, QPR:$Vm))>;
4504 def : Pat<(v4i32 (int_arm_neon_vqadds
4506 (v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$Vn),
4507 (v4i32 QPR:$Vm))))),
4508 (v4i32 (VQRDMLAHv4i32 QPR:$src1, QPR:$Vn, QPR:$Vm))>;
4510 defm VQRDMLAHsl : N3VMulOpSL_HS<0b1110, IIC_VMACi16D, IIC_VMACi32D,
4511 IIC_VMACi16Q, IIC_VMACi32Q, "vqrdmlah", "s",
4513 def : Pat<(v4i16 (int_arm_neon_vqadds
4515 (v4i16 (int_arm_neon_vqrdmulh
4517 (v4i16 (ARMvduplane (v4i16 DPR_8:$Vm),
4519 (v4i16 (VQRDMLAHslv4i16 DPR:$src1, DPR:$Vn, DPR_8:$Vm,
4521 def : Pat<(v2i32 (int_arm_neon_vqadds
4523 (v2i32 (int_arm_neon_vqrdmulh
4525 (v2i32 (ARMvduplane (v2i32 DPR_VFP2:$Vm),
4527 (v2i32 (VQRDMLAHslv2i32 DPR:$src1, DPR:$Vn, DPR_VFP2:$Vm,
4529 def : Pat<(v8i16 (int_arm_neon_vqadds
4531 (v8i16 (int_arm_neon_vqrdmulh
4533 (v8i16 (ARMvduplane (v8i16 QPR:$src3),
4535 (v8i16 (VQRDMLAHslv8i16 (v8i16 QPR:$src1),
4537 (v4i16 (EXTRACT_SUBREG
4539 (DSubReg_i16_reg imm:$lane))),
4540 (SubReg_i16_lane imm:$lane)))>;
4541 def : Pat<(v4i32 (int_arm_neon_vqadds
4543 (v4i32 (int_arm_neon_vqrdmulh
4545 (v4i32 (ARMvduplane (v4i32 QPR:$src3),
4547 (v4i32 (VQRDMLAHslv4i32 (v4i32 QPR:$src1),
4549 (v2i32 (EXTRACT_SUBREG
4551 (DSubReg_i32_reg imm:$lane))),
4552 (SubReg_i32_lane imm:$lane)))>;
4554 // VQRDMLSH : Vector Saturating Rounding Doubling Multiply Subtract Long
4556 defm VQRDMLSH : N3VInt3_HS<1, 0, 0b1100, 1, IIC_VMACi16D, IIC_VMACi32D,
4557 IIC_VMACi16Q, IIC_VMACi32Q, "vqrdmlsh", "s",
4559 def : Pat<(v4i16 (int_arm_neon_vqsubs
4561 (v4i16 (int_arm_neon_vqrdmulh (v4i16 DPR:$Vn),
4562 (v4i16 DPR:$Vm))))),
4563 (v4i16 (VQRDMLSHv4i16 DPR:$src1, DPR:$Vn, DPR:$Vm))>;
4564 def : Pat<(v2i32 (int_arm_neon_vqsubs
4566 (v2i32 (int_arm_neon_vqrdmulh (v2i32 DPR:$Vn),
4567 (v2i32 DPR:$Vm))))),
4568 (v2i32 (VQRDMLSHv2i32 DPR:$src1, DPR:$Vn, DPR:$Vm))>;
4569 def : Pat<(v8i16 (int_arm_neon_vqsubs
4571 (v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$Vn),
4572 (v8i16 QPR:$Vm))))),
4573 (v8i16 (VQRDMLSHv8i16 QPR:$src1, QPR:$Vn, QPR:$Vm))>;
4574 def : Pat<(v4i32 (int_arm_neon_vqsubs
4576 (v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$Vn),
4577 (v4i32 QPR:$Vm))))),
4578 (v4i32 (VQRDMLSHv4i32 QPR:$src1, QPR:$Vn, QPR:$Vm))>;
4580 defm VQRDMLSHsl : N3VMulOpSL_HS<0b1111, IIC_VMACi16D, IIC_VMACi32D,
4581 IIC_VMACi16Q, IIC_VMACi32Q, "vqrdmlsh", "s",
4583 def : Pat<(v4i16 (int_arm_neon_vqsubs
4585 (v4i16 (int_arm_neon_vqrdmulh
4587 (v4i16 (ARMvduplane (v4i16 DPR_8:$Vm),
4589 (v4i16 (VQRDMLSHslv4i16 DPR:$src1, DPR:$Vn, DPR_8:$Vm, imm:$lane))>;
4590 def : Pat<(v2i32 (int_arm_neon_vqsubs
4592 (v2i32 (int_arm_neon_vqrdmulh
4594 (v2i32 (ARMvduplane (v2i32 DPR_VFP2:$Vm),
4596 (v2i32 (VQRDMLSHslv2i32 DPR:$src1, DPR:$Vn, DPR_VFP2:$Vm,
4598 def : Pat<(v8i16 (int_arm_neon_vqsubs
4600 (v8i16 (int_arm_neon_vqrdmulh
4602 (v8i16 (ARMvduplane (v8i16 QPR:$src3),
4604 (v8i16 (VQRDMLSHslv8i16 (v8i16 QPR:$src1),
4606 (v4i16 (EXTRACT_SUBREG
4608 (DSubReg_i16_reg imm:$lane))),
4609 (SubReg_i16_lane imm:$lane)))>;
4610 def : Pat<(v4i32 (int_arm_neon_vqsubs
4612 (v4i32 (int_arm_neon_vqrdmulh
4614 (v4i32 (ARMvduplane (v4i32 QPR:$src3),
4616 (v4i32 (VQRDMLSHslv4i32 (v4i32 QPR:$src1),
4618 (v2i32 (EXTRACT_SUBREG
4620 (DSubReg_i32_reg imm:$lane))),
4621 (SubReg_i32_lane imm:$lane)))>;
4623 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
4624 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
4625 "vqdmlal", "s", null_frag>;
4626 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", null_frag>;
4628 let Predicates = [HasNEON] in {
4629 def : Pat<(v4i32 (int_arm_neon_vqadds (v4i32 QPR:$src1),
4630 (v4i32 (int_arm_neon_vqdmull (v4i16 DPR:$Vn),
4631 (v4i16 DPR:$Vm))))),
4632 (VQDMLALv4i32 QPR:$src1, DPR:$Vn, DPR:$Vm)>;
4633 def : Pat<(v2i64 (int_arm_neon_vqadds (v2i64 QPR:$src1),
4634 (v2i64 (int_arm_neon_vqdmull (v2i32 DPR:$Vn),
4635 (v2i32 DPR:$Vm))))),
4636 (VQDMLALv2i64 QPR:$src1, DPR:$Vn, DPR:$Vm)>;
4637 def : Pat<(v4i32 (int_arm_neon_vqadds (v4i32 QPR:$src1),
4638 (v4i32 (int_arm_neon_vqdmull (v4i16 DPR:$Vn),
4639 (v4i16 (ARMvduplane (v4i16 DPR_8:$Vm),
4641 (VQDMLALslv4i16 QPR:$src1, DPR:$Vn, DPR_8:$Vm, imm:$lane)>;
4642 def : Pat<(v2i64 (int_arm_neon_vqadds (v2i64 QPR:$src1),
4643 (v2i64 (int_arm_neon_vqdmull (v2i32 DPR:$Vn),
4644 (v2i32 (ARMvduplane (v2i32 DPR_VFP2:$Vm),
4646 (VQDMLALslv2i32 QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, imm:$lane)>;
4649 // VMLS : Vector Multiply Subtract (integer and floating-point)
4650 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
4651 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
4652 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
4653 v2f32, fmul_su, fsub_mlx>,
4654 Requires<[HasNEON, UseFPVMLx]>;
4655 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
4656 v4f32, fmul_su, fsub_mlx>,
4657 Requires<[HasNEON, UseFPVMLx]>;
4658 def VMLShd : N3VDMulOp<0, 0, 0b11, 0b1101, 1, IIC_VMACD, "vmls", "f16",
4660 Requires<[HasNEON, HasFullFP16, UseFPVMLx]>;
4661 def VMLShq : N3VQMulOp<0, 0, 0b11, 0b1101, 1, IIC_VMACQ, "vmls", "f16",
4663 Requires<[HasNEON, HasFullFP16, UseFPVMLx]>;
4664 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
4665 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
4666 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
4667 v2f32, fmul_su, fsub_mlx>,
4668 Requires<[HasNEON, UseFPVMLx]>;
4669 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
4670 v4f32, v2f32, fmul_su, fsub_mlx>,
4671 Requires<[HasNEON, UseFPVMLx]>;
4672 def VMLSslhd : N3VDMulOpSL16<0b01, 0b0101, IIC_VMACD, "vmls", "f16",
4674 Requires<[HasNEON, HasFullFP16, UseFPVMLx]>;
4675 def VMLSslhq : N3VQMulOpSL16<0b01, 0b0101, IIC_VMACQ, "vmls", "f16",
4676 v8f16, v4f16, fmul, fsub>,
4677 Requires<[HasNEON, HasFullFP16, UseFPVMLx]>;
4679 let Predicates = [HasNEON] in {
4680 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
4681 (mul (v8i16 QPR:$src2),
4682 (v8i16 (ARMvduplane (v8i16 QPR:$src3), imm:$lane))))),
4683 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
4684 (v4i16 (EXTRACT_SUBREG QPR:$src3,
4685 (DSubReg_i16_reg imm:$lane))),
4686 (SubReg_i16_lane imm:$lane)))>;
4688 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
4689 (mul (v4i32 QPR:$src2),
4690 (v4i32 (ARMvduplane (v4i32 QPR:$src3), imm:$lane))))),
4691 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
4692 (v2i32 (EXTRACT_SUBREG QPR:$src3,
4693 (DSubReg_i32_reg imm:$lane))),
4694 (SubReg_i32_lane imm:$lane)))>;
4697 def : Pat<(v4f32 (fsub_mlx (v4f32 QPR:$src1),
4698 (fmul_su (v4f32 QPR:$src2),
4699 (v4f32 (ARMvduplane (v4f32 QPR:$src3), imm:$lane))))),
4700 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
4701 (v2f32 (EXTRACT_SUBREG QPR:$src3,
4702 (DSubReg_i32_reg imm:$lane))),
4703 (SubReg_i32_lane imm:$lane)))>,
4704 Requires<[HasNEON, UseFPVMLx]>;
4706 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
4707 defm VMLSLs : N3VLMulOp_QHS<0,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
4708 "vmlsl", "s", NEONvmulls, sub>;
4709 defm VMLSLu : N3VLMulOp_QHS<1,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
4710 "vmlsl", "u", NEONvmullu, sub>;
4712 defm VMLSLsls : N3VLMulOpSL_HS<0, 0b0110, "vmlsl", "s", NEONvmulls, sub>;
4713 defm VMLSLslu : N3VLMulOpSL_HS<1, 0b0110, "vmlsl", "u", NEONvmullu, sub>;
4715 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
4716 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, IIC_VMACi16D, IIC_VMACi32D,
4717 "vqdmlsl", "s", null_frag>;
4718 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b0111, "vqdmlsl", "s", null_frag>;
4720 let Predicates = [HasNEON] in {
4721 def : Pat<(v4i32 (int_arm_neon_vqsubs (v4i32 QPR:$src1),
4722 (v4i32 (int_arm_neon_vqdmull (v4i16 DPR:$Vn),
4723 (v4i16 DPR:$Vm))))),
4724 (VQDMLSLv4i32 QPR:$src1, DPR:$Vn, DPR:$Vm)>;
4725 def : Pat<(v2i64 (int_arm_neon_vqsubs (v2i64 QPR:$src1),
4726 (v2i64 (int_arm_neon_vqdmull (v2i32 DPR:$Vn),
4727 (v2i32 DPR:$Vm))))),
4728 (VQDMLSLv2i64 QPR:$src1, DPR:$Vn, DPR:$Vm)>;
4729 def : Pat<(v4i32 (int_arm_neon_vqsubs (v4i32 QPR:$src1),
4730 (v4i32 (int_arm_neon_vqdmull (v4i16 DPR:$Vn),
4731 (v4i16 (ARMvduplane (v4i16 DPR_8:$Vm),
4733 (VQDMLSLslv4i16 QPR:$src1, DPR:$Vn, DPR_8:$Vm, imm:$lane)>;
4734 def : Pat<(v2i64 (int_arm_neon_vqsubs (v2i64 QPR:$src1),
4735 (v2i64 (int_arm_neon_vqdmull (v2i32 DPR:$Vn),
4736 (v2i32 (ARMvduplane (v2i32 DPR_VFP2:$Vm),
4738 (VQDMLSLslv2i32 QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, imm:$lane)>;
4741 // Fused Vector Multiply-Accumulate and Fused Multiply-Subtract Operations.
4742 def VFMAfd : N3VDMulOp<0, 0, 0b00, 0b1100, 1, IIC_VFMACD, "vfma", "f32",
4743 v2f32, fmul_su, fadd_mlx>,
4744 Requires<[HasNEON,HasVFP4,UseFusedMAC]>;
4746 def VFMAfq : N3VQMulOp<0, 0, 0b00, 0b1100, 1, IIC_VFMACQ, "vfma", "f32",
4747 v4f32, fmul_su, fadd_mlx>,
4748 Requires<[HasNEON,HasVFP4,UseFusedMAC]>;
4749 def VFMAhd : N3VDMulOp<0, 0, 0b01, 0b1100, 1, IIC_VFMACD, "vfma", "f16",
4751 Requires<[HasNEON,HasFullFP16,UseFusedMAC]>;
4753 def VFMAhq : N3VQMulOp<0, 0, 0b01, 0b1100, 1, IIC_VFMACQ, "vfma", "f16",
4755 Requires<[HasNEON,HasFullFP16,UseFusedMAC]>;
4757 // Fused Vector Multiply Subtract (floating-point)
4758 def VFMSfd : N3VDMulOp<0, 0, 0b10, 0b1100, 1, IIC_VFMACD, "vfms", "f32",
4759 v2f32, fmul_su, fsub_mlx>,
4760 Requires<[HasNEON,HasVFP4,UseFusedMAC]>;
4761 def VFMSfq : N3VQMulOp<0, 0, 0b10, 0b1100, 1, IIC_VFMACQ, "vfms", "f32",
4762 v4f32, fmul_su, fsub_mlx>,
4763 Requires<[HasNEON,HasVFP4,UseFusedMAC]>;
4764 def VFMShd : N3VDMulOp<0, 0, 0b11, 0b1100, 1, IIC_VFMACD, "vfms", "f16",
4766 Requires<[HasNEON,HasFullFP16,UseFusedMAC]>;
4767 def VFMShq : N3VQMulOp<0, 0, 0b11, 0b1100, 1, IIC_VFMACQ, "vfms", "f16",
4769 Requires<[HasNEON,HasFullFP16,UseFusedMAC]>;
4771 // Match @llvm.fma.* intrinsics
4772 def : Pat<(v4f16 (fma DPR:$Vn, DPR:$Vm, DPR:$src1)),
4773 (VFMAhd DPR:$src1, DPR:$Vn, DPR:$Vm)>,
4774 Requires<[HasNEON,HasFullFP16]>;
4775 def : Pat<(v8f16 (fma QPR:$Vn, QPR:$Vm, QPR:$src1)),
4776 (VFMAhq QPR:$src1, QPR:$Vn, QPR:$Vm)>,
4777 Requires<[HasNEON,HasFullFP16]>;
4778 def : Pat<(v2f32 (fma DPR:$Vn, DPR:$Vm, DPR:$src1)),
4779 (VFMAfd DPR:$src1, DPR:$Vn, DPR:$Vm)>,
4780 Requires<[HasNEON,HasVFP4]>;
4781 def : Pat<(v4f32 (fma QPR:$Vn, QPR:$Vm, QPR:$src1)),
4782 (VFMAfq QPR:$src1, QPR:$Vn, QPR:$Vm)>,
4783 Requires<[HasNEON,HasVFP4]>;
4784 def : Pat<(v2f32 (fma (fneg DPR:$Vn), DPR:$Vm, DPR:$src1)),
4785 (VFMSfd DPR:$src1, DPR:$Vn, DPR:$Vm)>,
4786 Requires<[HasNEON,HasVFP4]>;
4787 def : Pat<(v4f32 (fma (fneg QPR:$Vn), QPR:$Vm, QPR:$src1)),
4788 (VFMSfq QPR:$src1, QPR:$Vn, QPR:$Vm)>,
4789 Requires<[HasNEON,HasVFP4]>;
4791 // ARMv8.2a dot product instructions.
4792 // We put them in the VFPV8 decoder namespace because the ARM and Thumb
4793 // encodings are the same and thus no further bit twiddling is necessary
4794 // in the disassembler.
4795 class VDOT<bit op6, bit op4, RegisterClass RegTy, string Asm, string AsmTy,
4796 ValueType AccumTy, ValueType InputTy,
4797 SDPatternOperator OpNode> :
4798 N3Vnp<0b11000, 0b10, 0b1101, op6, op4, (outs RegTy:$dst),
4799 (ins RegTy:$Vd, RegTy:$Vn, RegTy:$Vm), N3RegFrm, IIC_VDOTPROD,
4801 [(set (AccumTy RegTy:$dst),
4802 (OpNode (AccumTy RegTy:$Vd),
4803 (InputTy RegTy:$Vn),
4804 (InputTy RegTy:$Vm)))]> {
4805 let Predicates = [HasDotProd];
4806 let DecoderNamespace = "VFPV8";
4807 let Constraints = "$dst = $Vd";
4810 def VUDOTD : VDOT<0, 1, DPR, "vudot", "u8", v2i32, v8i8, int_arm_neon_udot>;
4811 def VSDOTD : VDOT<0, 0, DPR, "vsdot", "s8", v2i32, v8i8, int_arm_neon_sdot>;
4812 def VUDOTQ : VDOT<1, 1, QPR, "vudot", "u8", v4i32, v16i8, int_arm_neon_udot>;
4813 def VSDOTQ : VDOT<1, 0, QPR, "vsdot", "s8", v4i32, v16i8, int_arm_neon_sdot>;
4815 // Indexed dot product instructions:
4816 multiclass DOTI<string opc, string dt, bit Q, bit U, RegisterClass Ty,
4817 ValueType AccumType, ValueType InputType, SDPatternOperator OpNode,
4819 def "" : N3Vnp<0b11100, 0b10, 0b1101, Q, U, (outs Ty:$dst),
4820 (ins Ty:$Vd, Ty:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
4821 N3RegFrm, IIC_VDOTPROD, opc, dt, []> {
4824 let AsmString = !strconcat(opc, ".", dt, "\t$Vd, $Vn, $Vm$lane");
4825 let Constraints = "$dst = $Vd";
4826 let Predicates = [HasDotProd];
4827 let DecoderNamespace = "VFPV8";
4831 (AccumType (OpNode (AccumType Ty:$Vd),
4833 (InputType (bitconvert (AccumType
4834 (ARMvduplane (AccumType Ty:$Vm),
4835 VectorIndex32:$lane)))))),
4836 (!cast<Instruction>(NAME) Ty:$Vd, Ty:$Vn, RHS, VectorIndex32:$lane)>;
4839 defm VUDOTDI : DOTI<"vudot", "u8", 0b0, 0b1, DPR, v2i32, v8i8,
4840 int_arm_neon_udot, (v2i32 DPR_VFP2:$Vm)>;
4841 defm VSDOTDI : DOTI<"vsdot", "s8", 0b0, 0b0, DPR, v2i32, v8i8,
4842 int_arm_neon_sdot, (v2i32 DPR_VFP2:$Vm)>;
4843 defm VUDOTQI : DOTI<"vudot", "u8", 0b1, 0b1, QPR, v4i32, v16i8,
4844 int_arm_neon_udot, (EXTRACT_SUBREG QPR:$Vm, dsub_0)>;
4845 defm VSDOTQI : DOTI<"vsdot", "s8", 0b1, 0b0, QPR, v4i32, v16i8,
4846 int_arm_neon_sdot, (EXTRACT_SUBREG QPR:$Vm, dsub_0)>;
4849 // ARMv8.3 complex operations
4850 class BaseN3VCP8ComplexTied<bit op21, bit op4, bit s, bit q,
4851 InstrItinClass itin, dag oops, dag iops,
4852 string opc, string dt, list<dag> pattern>
4853 : N3VCP8<{?,?}, {op21,s}, q, op4, oops,
4854 iops, itin, opc, dt, "$Vd, $Vn, $Vm, $rot", "$src1 = $Vd", pattern>{
4856 let Inst{24-23} = rot;
4859 class BaseN3VCP8ComplexOdd<bit op23, bit op21, bit op4, bit s, bit q,
4860 InstrItinClass itin, dag oops, dag iops, string opc,
4861 string dt, list<dag> pattern>
4862 : N3VCP8<{?,op23}, {op21,s}, q, op4, oops,
4863 iops, itin, opc, dt, "$Vd, $Vn, $Vm, $rot", "", pattern> {
4868 class BaseN3VCP8ComplexTiedLane32<bit op4, bit s, bit q, InstrItinClass itin,
4869 dag oops, dag iops, string opc, string dt,
4871 : N3VLaneCP8<s, {?,?}, q, op4, oops, iops, itin, opc, dt,
4872 "$Vd, $Vn, $Vm$lane, $rot", "$src1 = $Vd", pattern> {
4876 let Inst{21-20} = rot;
4880 class BaseN3VCP8ComplexTiedLane64<bit op4, bit s, bit q, InstrItinClass itin,
4881 dag oops, dag iops, string opc, string dt,
4883 : N3VLaneCP8<s, {?,?}, q, op4, oops, iops, itin, opc, dt,
4884 "$Vd, $Vn, $Vm$lane, $rot", "$src1 = $Vd", pattern> {
4888 let Inst{21-20} = rot;
4889 let Inst{5} = Vm{4};
4890 // This is needed because the lane operand does not have any bits in the
4891 // encoding (it only has one possible value), so we need to manually set it
4892 // to it's default value.
4893 let DecoderMethod = "DecodeNEONComplexLane64Instruction";
4896 multiclass N3VCP8ComplexTied<bit op21, bit op4,
4897 string OpcodeStr, SDPatternOperator Op> {
4898 let Predicates = [HasNEON,HasV8_3a,HasFullFP16] in {
4899 def v4f16 : BaseN3VCP8ComplexTied<op21, op4, 0, 0, IIC_VMACD, (outs DPR:$Vd),
4900 (ins DPR:$src1, DPR:$Vn, DPR:$Vm, complexrotateop:$rot),
4901 OpcodeStr, "f16", []>;
4902 def v8f16 : BaseN3VCP8ComplexTied<op21, op4, 0, 1, IIC_VMACQ, (outs QPR:$Vd),
4903 (ins QPR:$src1, QPR:$Vn, QPR:$Vm, complexrotateop:$rot),
4904 OpcodeStr, "f16", []>;
4906 let Predicates = [HasNEON,HasV8_3a] in {
4907 def v2f32 : BaseN3VCP8ComplexTied<op21, op4, 1, 0, IIC_VMACD, (outs DPR:$Vd),
4908 (ins DPR:$src1, DPR:$Vn, DPR:$Vm, complexrotateop:$rot),
4909 OpcodeStr, "f32", []>;
4910 def v4f32 : BaseN3VCP8ComplexTied<op21, op4, 1, 1, IIC_VMACQ, (outs QPR:$Vd),
4911 (ins QPR:$src1, QPR:$Vn, QPR:$Vm, complexrotateop:$rot),
4912 OpcodeStr, "f32", []>;
4916 multiclass N3VCP8ComplexOdd<bit op23, bit op21, bit op4,
4917 string OpcodeStr, SDPatternOperator Op> {
4918 let Predicates = [HasNEON,HasV8_3a,HasFullFP16] in {
4919 def v4f16 : BaseN3VCP8ComplexOdd<op23, op21, op4, 0, 0, IIC_VMACD,
4921 (ins DPR:$Vn, DPR:$Vm, complexrotateopodd:$rot),
4922 OpcodeStr, "f16", []>;
4923 def v8f16 : BaseN3VCP8ComplexOdd<op23, op21, op4, 0, 1, IIC_VMACQ,
4925 (ins QPR:$Vn, QPR:$Vm, complexrotateopodd:$rot),
4926 OpcodeStr, "f16", []>;
4928 let Predicates = [HasNEON,HasV8_3a] in {
4929 def v2f32 : BaseN3VCP8ComplexOdd<op23, op21, op4, 1, 0, IIC_VMACD,
4931 (ins DPR:$Vn, DPR:$Vm, complexrotateopodd:$rot),
4932 OpcodeStr, "f32", []>;
4933 def v4f32 : BaseN3VCP8ComplexOdd<op23, op21, op4, 1, 1, IIC_VMACQ,
4935 (ins QPR:$Vn, QPR:$Vm, complexrotateopodd:$rot),
4936 OpcodeStr, "f32", []>;
4940 // These instructions index by pairs of lanes, so the VectorIndexes are twice
4941 // as wide as the data types.
4942 multiclass N3VCP8ComplexTiedLane<bit op4, string OpcodeStr,
4943 SDPatternOperator Op> {
4944 let Predicates = [HasNEON,HasV8_3a,HasFullFP16] in {
4945 def v4f16_indexed : BaseN3VCP8ComplexTiedLane32<op4, 0, 0, IIC_VMACD,
4947 (ins DPR:$src1, DPR:$Vn, DPR_VFP2:$Vm,
4948 VectorIndex32:$lane, complexrotateop:$rot),
4949 OpcodeStr, "f16", []>;
4950 def v8f16_indexed : BaseN3VCP8ComplexTiedLane32<op4, 0, 1, IIC_VMACQ,
4952 (ins QPR:$src1, QPR:$Vn, DPR_VFP2:$Vm,
4953 VectorIndex32:$lane, complexrotateop:$rot),
4954 OpcodeStr, "f16", []>;
4956 let Predicates = [HasNEON,HasV8_3a] in {
4957 def v2f32_indexed : BaseN3VCP8ComplexTiedLane64<op4, 1, 0, IIC_VMACD,
4959 (ins DPR:$src1, DPR:$Vn, DPR:$Vm, VectorIndex64:$lane,
4960 complexrotateop:$rot),
4961 OpcodeStr, "f32", []>;
4962 def v4f32_indexed : BaseN3VCP8ComplexTiedLane64<op4, 1, 1, IIC_VMACQ,
4964 (ins QPR:$src1, QPR:$Vn, DPR:$Vm, VectorIndex64:$lane,
4965 complexrotateop:$rot),
4966 OpcodeStr, "f32", []>;
4970 defm VCMLA : N3VCP8ComplexTied<1, 0, "vcmla", null_frag>;
4971 defm VCADD : N3VCP8ComplexOdd<1, 0, 0, "vcadd", null_frag>;
4972 defm VCMLA : N3VCP8ComplexTiedLane<0, "vcmla", null_frag>;
4974 // Vector Subtract Operations.
4976 // VSUB : Vector Subtract (integer and floating-point)
4977 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
4978 "vsub", "i", sub, 0>;
4979 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
4980 v2f32, v2f32, fsub, 0>;
4981 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
4982 v4f32, v4f32, fsub, 0>;
4983 def VSUBhd : N3VD<0, 0, 0b11, 0b1101, 0, IIC_VBIND, "vsub", "f16",
4984 v4f16, v4f16, fsub, 0>,
4985 Requires<[HasNEON,HasFullFP16]>;
4986 def VSUBhq : N3VQ<0, 0, 0b11, 0b1101, 0, IIC_VBINQ, "vsub", "f16",
4987 v8f16, v8f16, fsub, 0>,
4988 Requires<[HasNEON,HasFullFP16]>;
4989 // VSUBL : Vector Subtract Long (Q = D - D)
4990 defm VSUBLs : N3VLExt_QHS<0,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
4991 "vsubl", "s", sub, sext, 0>;
4992 defm VSUBLu : N3VLExt_QHS<1,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
4993 "vsubl", "u", sub, zext, 0>;
4994 // VSUBW : Vector Subtract Wide (Q = Q - D)
4995 defm VSUBWs : N3VW_QHS<0,1,0b0011,0, "vsubw", "s", sub, sext, 0>;
4996 defm VSUBWu : N3VW_QHS<1,1,0b0011,0, "vsubw", "u", sub, zext, 0>;
4997 // VHSUB : Vector Halving Subtract
4998 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, N3RegFrm,
4999 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
5000 "vhsub", "s", int_arm_neon_vhsubs, 0>;
5001 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, N3RegFrm,
5002 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
5003 "vhsub", "u", int_arm_neon_vhsubu, 0>;
5004 // VQSUB : Vector Saturing Subtract
5005 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, N3RegFrm,
5006 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
5007 "vqsub", "s", int_arm_neon_vqsubs, 0>;
5008 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, N3RegFrm,
5009 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
5010 "vqsub", "u", int_arm_neon_vqsubu, 0>;
5011 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
5012 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i", null_frag, 0>;
5013 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
5014 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
5015 int_arm_neon_vrsubhn, 0>;
5017 let Predicates = [HasNEON] in {
5018 def : Pat<(v8i8 (trunc (ARMvshruImm (sub (v8i16 QPR:$Vn), QPR:$Vm), 8))),
5019 (VSUBHNv8i8 QPR:$Vn, QPR:$Vm)>;
5020 def : Pat<(v4i16 (trunc (ARMvshruImm (sub (v4i32 QPR:$Vn), QPR:$Vm), 16))),
5021 (VSUBHNv4i16 QPR:$Vn, QPR:$Vm)>;
5022 def : Pat<(v2i32 (trunc (ARMvshruImm (sub (v2i64 QPR:$Vn), QPR:$Vm), 32))),
5023 (VSUBHNv2i32 QPR:$Vn, QPR:$Vm)>;
5026 // Vector Comparisons.
5028 // VCEQ : Vector Compare Equal
5029 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
5030 IIC_VSUBi4Q, "vceq", "i", NEONvceq, 1>;
5031 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
5033 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
5035 def VCEQhd : N3VD<0,0,0b01,0b1110,0, IIC_VBIND, "vceq", "f16", v4i16, v4f16,
5037 Requires<[HasNEON, HasFullFP16]>;
5038 def VCEQhq : N3VQ<0,0,0b01,0b1110,0, IIC_VBINQ, "vceq", "f16", v8i16, v8f16,
5040 Requires<[HasNEON, HasFullFP16]>;
5042 let TwoOperandAliasConstraint = "$Vm = $Vd" in
5043 defm VCEQz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
5044 "$Vd, $Vm, #0", NEONvceqz>;
5046 // VCGE : Vector Compare Greater Than or Equal
5047 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
5048 IIC_VSUBi4Q, "vcge", "s", NEONvcge, 0>;
5049 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
5050 IIC_VSUBi4Q, "vcge", "u", NEONvcgeu, 0>;
5051 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32", v2i32, v2f32,
5053 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
5055 def VCGEhd : N3VD<1,0,0b01,0b1110,0, IIC_VBIND, "vcge", "f16", v4i16, v4f16,
5057 Requires<[HasNEON, HasFullFP16]>;
5058 def VCGEhq : N3VQ<1,0,0b01,0b1110,0, IIC_VBINQ, "vcge", "f16", v8i16, v8f16,
5060 Requires<[HasNEON, HasFullFP16]>;
5062 let TwoOperandAliasConstraint = "$Vm = $Vd" in {
5063 defm VCGEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
5064 "$Vd, $Vm, #0", NEONvcgez>;
5065 defm VCLEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
5066 "$Vd, $Vm, #0", NEONvclez>;
5069 // VCGT : Vector Compare Greater Than
5070 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
5071 IIC_VSUBi4Q, "vcgt", "s", NEONvcgt, 0>;
5072 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
5073 IIC_VSUBi4Q, "vcgt", "u", NEONvcgtu, 0>;
5074 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
5076 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
5078 def VCGThd : N3VD<1,0,0b11,0b1110,0, IIC_VBIND, "vcgt", "f16", v4i16, v4f16,
5080 Requires<[HasNEON, HasFullFP16]>;
5081 def VCGThq : N3VQ<1,0,0b11,0b1110,0, IIC_VBINQ, "vcgt", "f16", v8i16, v8f16,
5083 Requires<[HasNEON, HasFullFP16]>;
5085 let TwoOperandAliasConstraint = "$Vm = $Vd" in {
5086 defm VCGTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
5087 "$Vd, $Vm, #0", NEONvcgtz>;
5088 defm VCLTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
5089 "$Vd, $Vm, #0", NEONvcltz>;
5092 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
5093 def VACGEfd : N3VDInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacge",
5094 "f32", v2i32, v2f32, int_arm_neon_vacge, 0>;
5095 def VACGEfq : N3VQInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacge",
5096 "f32", v4i32, v4f32, int_arm_neon_vacge, 0>;
5097 def VACGEhd : N3VDInt<1, 0, 0b01, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacge",
5098 "f16", v4i16, v4f16, int_arm_neon_vacge, 0>,
5099 Requires<[HasNEON, HasFullFP16]>;
5100 def VACGEhq : N3VQInt<1, 0, 0b01, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacge",
5101 "f16", v8i16, v8f16, int_arm_neon_vacge, 0>,
5102 Requires<[HasNEON, HasFullFP16]>;
5103 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
5104 def VACGTfd : N3VDInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacgt",
5105 "f32", v2i32, v2f32, int_arm_neon_vacgt, 0>;
5106 def VACGTfq : N3VQInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacgt",
5107 "f32", v4i32, v4f32, int_arm_neon_vacgt, 0>;
5108 def VACGThd : N3VDInt<1, 0, 0b11, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacgt",
5109 "f16", v4i16, v4f16, int_arm_neon_vacgt, 0>,
5110 Requires<[HasNEON, HasFullFP16]>;
5111 def VACGThq : N3VQInt<1, 0, 0b11, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacgt",
5112 "f16", v8i16, v8f16, int_arm_neon_vacgt, 0>,
5113 Requires<[HasNEON, HasFullFP16]>;
5114 // VTST : Vector Test Bits
5115 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
5116 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
5118 def: NEONInstAlias<"vaclt${p}.f32 $Vd, $Vn, $Vm",
5119 (VACGTfd DPR:$Vd, DPR:$Vm, DPR:$Vn, pred:$p)>;
5120 def: NEONInstAlias<"vaclt${p}.f32 $Vd, $Vn, $Vm",
5121 (VACGTfq QPR:$Vd, QPR:$Vm, QPR:$Vn, pred:$p)>;
5122 def: NEONInstAlias<"vacle${p}.f32 $Vd, $Vn, $Vm",
5123 (VACGEfd DPR:$Vd, DPR:$Vm, DPR:$Vn, pred:$p)>;
5124 def: NEONInstAlias<"vacle${p}.f32 $Vd, $Vn, $Vm",
5125 (VACGEfq QPR:$Vd, QPR:$Vm, QPR:$Vn, pred:$p)>;
5126 let Predicates = [HasNEON, HasFullFP16] in {
5127 def: NEONInstAlias<"vaclt${p}.f16 $Vd, $Vn, $Vm",
5128 (VACGThd DPR:$Vd, DPR:$Vm, DPR:$Vn, pred:$p)>;
5129 def: NEONInstAlias<"vaclt${p}.f16 $Vd, $Vn, $Vm",
5130 (VACGThq QPR:$Vd, QPR:$Vm, QPR:$Vn, pred:$p)>;
5131 def: NEONInstAlias<"vacle${p}.f16 $Vd, $Vn, $Vm",
5132 (VACGEhd DPR:$Vd, DPR:$Vm, DPR:$Vn, pred:$p)>;
5133 def: NEONInstAlias<"vacle${p}.f16 $Vd, $Vn, $Vm",
5134 (VACGEhq QPR:$Vd, QPR:$Vm, QPR:$Vn, pred:$p)>;
5137 // +fp16fml Floating Point Multiplication Variants
5138 let Predicates = [HasNEON, HasFP16FML], DecoderNamespace= "VFPV8" in {
5140 class N3VCP8F16Q1<string asm, RegisterClass Td, RegisterClass Tn,
5141 RegisterClass Tm, bits<2> op1, bits<2> op2, bit op3>
5142 : N3VCP8<op1, op2, 1, op3, (outs Td:$Vd), (ins Tn:$Vn, Tm:$Vm), NoItinerary,
5143 asm, "f16", "$Vd, $Vn, $Vm", "", []>;
5145 class N3VCP8F16Q0<string asm, RegisterClass Td, RegisterClass Tn,
5146 RegisterClass Tm, bits<2> op1, bits<2> op2, bit op3>
5147 : N3VCP8Q0<op1, op2, 0, op3, (outs Td:$Vd), (ins Tn:$Vn, Tm:$Vm), NoItinerary,
5148 asm, "f16", "$Vd, $Vn, $Vm", "", []>;
5150 // Vd, Vs, Vs[0-15], Idx[0-1]
5151 class VFMD<string opc, string type, bits<2> S>
5152 : N3VLaneCP8<0, S, 0, 1, (outs DPR:$Vd),
5153 (ins SPR:$Vn, SPR_8:$Vm, VectorIndex32:$idx),
5154 IIC_VMACD, opc, type, "$Vd, $Vn, $Vm$idx", "", []> {
5157 let Inst{19-16} = Vn{4-1};
5158 let Inst{7} = Vn{0};
5159 let Inst{5} = Vm{0};
5160 let Inst{2-0} = Vm{3-1};
5163 // Vq, Vd, Vd[0-7], Idx[0-3]
5164 class VFMQ<string opc, string type, bits<2> S>
5165 : N3VLaneCP8<0, S, 1, 1, (outs QPR:$Vd),
5166 (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$idx),
5167 IIC_VMACD, opc, type, "$Vd, $Vn, $Vm$idx", "", []> {
5169 let Inst{5} = idx{1};
5170 let Inst{3} = idx{0};
5173 let hasNoSchedulingInfo = 1 in {
5175 def VFMALD : N3VCP8F16Q0<"vfmal", DPR, SPR, SPR, 0b00, 0b10, 1>;
5176 def VFMSLD : N3VCP8F16Q0<"vfmsl", DPR, SPR, SPR, 0b01, 0b10, 1>;
5177 def VFMALQ : N3VCP8F16Q1<"vfmal", QPR, DPR, DPR, 0b00, 0b10, 1>;
5178 def VFMSLQ : N3VCP8F16Q1<"vfmsl", QPR, DPR, DPR, 0b01, 0b10, 1>;
5179 def VFMALDI : VFMD<"vfmal", "f16", 0b00>;
5180 def VFMSLDI : VFMD<"vfmsl", "f16", 0b01>;
5181 def VFMALQI : VFMQ<"vfmal", "f16", 0b00>;
5182 def VFMSLQI : VFMQ<"vfmsl", "f16", 0b01>;
5184 } // HasNEON, HasFP16FML
5187 def: NEONInstAlias<"vaclt${p}.f32 $Vd, $Vm",
5188 (VACGTfd DPR:$Vd, DPR:$Vm, DPR:$Vd, pred:$p)>;
5189 def: NEONInstAlias<"vaclt${p}.f32 $Vd, $Vm",
5190 (VACGTfq QPR:$Vd, QPR:$Vm, QPR:$Vd, pred:$p)>;
5191 def: NEONInstAlias<"vacle${p}.f32 $Vd, $Vm",
5192 (VACGEfd DPR:$Vd, DPR:$Vm, DPR:$Vd, pred:$p)>;
5193 def: NEONInstAlias<"vacle${p}.f32 $Vd, $Vm",
5194 (VACGEfq QPR:$Vd, QPR:$Vm, QPR:$Vd, pred:$p)>;
5195 let Predicates = [HasNEON, HasFullFP16] in {
5196 def: NEONInstAlias<"vaclt${p}.f16 $Vd, $Vm",
5197 (VACGThd DPR:$Vd, DPR:$Vm, DPR:$Vd, pred:$p)>;
5198 def: NEONInstAlias<"vaclt${p}.f16 $Vd, $Vm",
5199 (VACGThq QPR:$Vd, QPR:$Vm, QPR:$Vd, pred:$p)>;
5200 def: NEONInstAlias<"vacle${p}.f16 $Vd, $Vm",
5201 (VACGEhd DPR:$Vd, DPR:$Vm, DPR:$Vd, pred:$p)>;
5202 def: NEONInstAlias<"vacle${p}.f16 $Vd, $Vm",
5203 (VACGEhq QPR:$Vd, QPR:$Vm, QPR:$Vd, pred:$p)>;
5206 // Vector Bitwise Operations.
5208 def vnotd : PatFrag<(ops node:$in),
5209 (xor node:$in, (bitconvert (v8i8 NEONimmAllOnesV)))>;
5210 def vnotq : PatFrag<(ops node:$in),
5211 (xor node:$in, (bitconvert (v16i8 NEONimmAllOnesV)))>;
5214 // VAND : Vector Bitwise AND
5215 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
5216 v2i32, v2i32, and, 1>;
5217 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
5218 v4i32, v4i32, and, 1>;
5220 // VEOR : Vector Bitwise Exclusive OR
5221 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
5222 v2i32, v2i32, xor, 1>;
5223 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
5224 v4i32, v4i32, xor, 1>;
5226 // VORR : Vector Bitwise OR
5227 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
5228 v2i32, v2i32, or, 1>;
5229 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
5230 v4i32, v4i32, or, 1>;
5232 def VORRiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 0, 1,
5233 (outs DPR:$Vd), (ins nImmSplatI16:$SIMM, DPR:$src),
5235 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
5237 (v4i16 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
5238 let Inst{9} = SIMM{9};
5241 def VORRiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 0, 1,
5242 (outs DPR:$Vd), (ins nImmSplatI32:$SIMM, DPR:$src),
5244 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
5246 (v2i32 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
5247 let Inst{10-9} = SIMM{10-9};
5250 def VORRiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 0, 1,
5251 (outs QPR:$Vd), (ins nImmSplatI16:$SIMM, QPR:$src),
5253 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
5255 (v8i16 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
5256 let Inst{9} = SIMM{9};
5259 def VORRiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 0, 1,
5260 (outs QPR:$Vd), (ins nImmSplatI32:$SIMM, QPR:$src),
5262 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
5264 (v4i32 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
5265 let Inst{10-9} = SIMM{10-9};
5269 // VBIC : Vector Bitwise Bit Clear (AND NOT)
5270 let TwoOperandAliasConstraint = "$Vn = $Vd" in {
5271 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
5272 (ins DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VBINiD,
5273 "vbic", "$Vd, $Vn, $Vm", "",
5274 [(set DPR:$Vd, (v2i32 (and DPR:$Vn,
5275 (vnotd DPR:$Vm))))]>;
5276 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
5277 (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINiQ,
5278 "vbic", "$Vd, $Vn, $Vm", "",
5279 [(set QPR:$Vd, (v4i32 (and QPR:$Vn,
5280 (vnotq QPR:$Vm))))]>;
5283 def VBICiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 1, 1,
5284 (outs DPR:$Vd), (ins nImmSplatI16:$SIMM, DPR:$src),
5286 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
5288 (v4i16 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
5289 let Inst{9} = SIMM{9};
5292 def VBICiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 1, 1,
5293 (outs DPR:$Vd), (ins nImmSplatI32:$SIMM, DPR:$src),
5295 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
5297 (v2i32 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
5298 let Inst{10-9} = SIMM{10-9};
5301 def VBICiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 1, 1,
5302 (outs QPR:$Vd), (ins nImmSplatI16:$SIMM, QPR:$src),
5304 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
5306 (v8i16 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
5307 let Inst{9} = SIMM{9};
5310 def VBICiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 1, 1,
5311 (outs QPR:$Vd), (ins nImmSplatI32:$SIMM, QPR:$src),
5313 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
5315 (v4i32 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
5316 let Inst{10-9} = SIMM{10-9};
5319 // VORN : Vector Bitwise OR NOT
5320 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$Vd),
5321 (ins DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VBINiD,
5322 "vorn", "$Vd, $Vn, $Vm", "",
5323 [(set DPR:$Vd, (v2i32 (or DPR:$Vn,
5324 (vnotd DPR:$Vm))))]>;
5325 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$Vd),
5326 (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINiQ,
5327 "vorn", "$Vd, $Vn, $Vm", "",
5328 [(set QPR:$Vd, (v4i32 (or QPR:$Vn,
5329 (vnotq QPR:$Vm))))]>;
5331 // VMVN : Vector Bitwise NOT (Immediate)
5333 let isReMaterializable = 1 in {
5335 def VMVNv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 1, 1, (outs DPR:$Vd),
5336 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
5337 "vmvn", "i16", "$Vd, $SIMM", "",
5338 [(set DPR:$Vd, (v4i16 (ARMvmvnImm timm:$SIMM)))]> {
5339 let Inst{9} = SIMM{9};
5342 def VMVNv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 1, 1, (outs QPR:$Vd),
5343 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
5344 "vmvn", "i16", "$Vd, $SIMM", "",
5345 [(set QPR:$Vd, (v8i16 (ARMvmvnImm timm:$SIMM)))]> {
5346 let Inst{9} = SIMM{9};
5349 def VMVNv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 1, 1, (outs DPR:$Vd),
5350 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
5351 "vmvn", "i32", "$Vd, $SIMM", "",
5352 [(set DPR:$Vd, (v2i32 (ARMvmvnImm timm:$SIMM)))]> {
5353 let Inst{11-8} = SIMM{11-8};
5356 def VMVNv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 1, 1, (outs QPR:$Vd),
5357 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
5358 "vmvn", "i32", "$Vd, $SIMM", "",
5359 [(set QPR:$Vd, (v4i32 (ARMvmvnImm timm:$SIMM)))]> {
5360 let Inst{11-8} = SIMM{11-8};
5364 // VMVN : Vector Bitwise NOT
5365 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
5366 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VSUBiD,
5367 "vmvn", "$Vd, $Vm", "",
5368 [(set DPR:$Vd, (v2i32 (vnotd DPR:$Vm)))]>;
5369 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
5370 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VSUBiD,
5371 "vmvn", "$Vd, $Vm", "",
5372 [(set QPR:$Vd, (v4i32 (vnotq QPR:$Vm)))]>;
5373 let Predicates = [HasNEON] in {
5374 def : Pat<(v2i32 (vnotd DPR:$src)), (VMVNd DPR:$src)>;
5375 def : Pat<(v4i32 (vnotq QPR:$src)), (VMVNq QPR:$src)>;
5378 // VBSL : Vector Bitwise Select
5379 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
5380 (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
5381 N3RegFrm, IIC_VCNTiD,
5382 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
5384 (v2i32 (NEONvbsl DPR:$src1, DPR:$Vn, DPR:$Vm)))]>;
5385 let Predicates = [HasNEON] in {
5386 def : Pat<(v8i8 (int_arm_neon_vbsl (v8i8 DPR:$src1),
5387 (v8i8 DPR:$Vn), (v8i8 DPR:$Vm))),
5388 (VBSLd DPR:$src1, DPR:$Vn, DPR:$Vm)>;
5389 def : Pat<(v4i16 (int_arm_neon_vbsl (v4i16 DPR:$src1),
5390 (v4i16 DPR:$Vn), (v4i16 DPR:$Vm))),
5391 (VBSLd DPR:$src1, DPR:$Vn, DPR:$Vm)>;
5392 def : Pat<(v2i32 (int_arm_neon_vbsl (v2i32 DPR:$src1),
5393 (v2i32 DPR:$Vn), (v2i32 DPR:$Vm))),
5394 (VBSLd DPR:$src1, DPR:$Vn, DPR:$Vm)>;
5395 def : Pat<(v2f32 (int_arm_neon_vbsl (v2f32 DPR:$src1),
5396 (v2f32 DPR:$Vn), (v2f32 DPR:$Vm))),
5397 (VBSLd DPR:$src1, DPR:$Vn, DPR:$Vm)>;
5398 def : Pat<(v1i64 (int_arm_neon_vbsl (v1i64 DPR:$src1),
5399 (v1i64 DPR:$Vn), (v1i64 DPR:$Vm))),
5400 (VBSLd DPR:$src1, DPR:$Vn, DPR:$Vm)>;
5402 def : Pat<(v2i32 (or (and DPR:$Vn, DPR:$Vd),
5403 (and DPR:$Vm, (vnotd DPR:$Vd)))),
5404 (VBSLd DPR:$Vd, DPR:$Vn, DPR:$Vm)>;
5406 def : Pat<(v1i64 (or (and DPR:$Vn, DPR:$Vd),
5407 (and DPR:$Vm, (vnotd DPR:$Vd)))),
5408 (VBSLd DPR:$Vd, DPR:$Vn, DPR:$Vm)>;
5411 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
5412 (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
5413 N3RegFrm, IIC_VCNTiQ,
5414 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
5416 (v4i32 (NEONvbsl QPR:$src1, QPR:$Vn, QPR:$Vm)))]>;
5418 let Predicates = [HasNEON] in {
5419 def : Pat<(v16i8 (int_arm_neon_vbsl (v16i8 QPR:$src1),
5420 (v16i8 QPR:$Vn), (v16i8 QPR:$Vm))),
5421 (VBSLq QPR:$src1, QPR:$Vn, QPR:$Vm)>;
5422 def : Pat<(v8i16 (int_arm_neon_vbsl (v8i16 QPR:$src1),
5423 (v8i16 QPR:$Vn), (v8i16 QPR:$Vm))),
5424 (VBSLq QPR:$src1, QPR:$Vn, QPR:$Vm)>;
5425 def : Pat<(v4i32 (int_arm_neon_vbsl (v4i32 QPR:$src1),
5426 (v4i32 QPR:$Vn), (v4i32 QPR:$Vm))),
5427 (VBSLq QPR:$src1, QPR:$Vn, QPR:$Vm)>;
5428 def : Pat<(v4f32 (int_arm_neon_vbsl (v4f32 QPR:$src1),
5429 (v4f32 QPR:$Vn), (v4f32 QPR:$Vm))),
5430 (VBSLq QPR:$src1, QPR:$Vn, QPR:$Vm)>;
5431 def : Pat<(v2i64 (int_arm_neon_vbsl (v2i64 QPR:$src1),
5432 (v2i64 QPR:$Vn), (v2i64 QPR:$Vm))),
5433 (VBSLq QPR:$src1, QPR:$Vn, QPR:$Vm)>;
5435 def : Pat<(v4i32 (or (and QPR:$Vn, QPR:$Vd),
5436 (and QPR:$Vm, (vnotq QPR:$Vd)))),
5437 (VBSLq QPR:$Vd, QPR:$Vn, QPR:$Vm)>;
5438 def : Pat<(v2i64 (or (and QPR:$Vn, QPR:$Vd),
5439 (and QPR:$Vm, (vnotq QPR:$Vd)))),
5440 (VBSLq QPR:$Vd, QPR:$Vn, QPR:$Vm)>;
5443 // VBIF : Vector Bitwise Insert if False
5444 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
5445 // FIXME: This instruction's encoding MAY NOT BE correct.
5446 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
5447 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
5448 N3RegFrm, IIC_VBINiD,
5449 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
5451 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
5452 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
5453 N3RegFrm, IIC_VBINiQ,
5454 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
5457 // VBIT : Vector Bitwise Insert if True
5458 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
5459 // FIXME: This instruction's encoding MAY NOT BE correct.
5460 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
5461 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
5462 N3RegFrm, IIC_VBINiD,
5463 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
5465 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
5466 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
5467 N3RegFrm, IIC_VBINiQ,
5468 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
5471 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
5472 // for equivalent operations with different register constraints; it just
5475 // Vector Absolute Differences.
5477 // VABD : Vector Absolute Difference
5478 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, N3RegFrm,
5479 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
5480 "vabd", "s", int_arm_neon_vabds, 1>;
5481 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, N3RegFrm,
5482 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
5483 "vabd", "u", int_arm_neon_vabdu, 1>;
5484 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBIND,
5485 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 1>;
5486 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBINQ,
5487 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 1>;
5488 def VABDhd : N3VDInt<1, 0, 0b11, 0b1101, 0, N3RegFrm, IIC_VBIND,
5489 "vabd", "f16", v4f16, v4f16, int_arm_neon_vabds, 1>,
5490 Requires<[HasNEON, HasFullFP16]>;
5491 def VABDhq : N3VQInt<1, 0, 0b11, 0b1101, 0, N3RegFrm, IIC_VBINQ,
5492 "vabd", "f16", v8f16, v8f16, int_arm_neon_vabds, 1>,
5493 Requires<[HasNEON, HasFullFP16]>;
5495 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
5496 defm VABDLs : N3VLIntExt_QHS<0,1,0b0111,0, IIC_VSUBi4Q,
5497 "vabdl", "s", int_arm_neon_vabds, zext, 1>;
5498 defm VABDLu : N3VLIntExt_QHS<1,1,0b0111,0, IIC_VSUBi4Q,
5499 "vabdl", "u", int_arm_neon_vabdu, zext, 1>;
5501 let Predicates = [HasNEON] in {
5502 def : Pat<(v8i16 (abs (sub (zext (v8i8 DPR:$opA)), (zext (v8i8 DPR:$opB))))),
5503 (VABDLuv8i16 DPR:$opA, DPR:$opB)>;
5504 def : Pat<(v4i32 (abs (sub (zext (v4i16 DPR:$opA)), (zext (v4i16 DPR:$opB))))),
5505 (VABDLuv4i32 DPR:$opA, DPR:$opB)>;
5508 // ISD::ABS is not legal for v2i64, so VABDL needs to be matched from the
5509 // shift/xor pattern for ABS.
5512 PatFrag<(ops node:$in1, node:$in2, node:$shift),
5513 (ARMvshrsImm (sub (zext node:$in1),
5514 (zext node:$in2)), (i32 $shift))>;
5516 let Predicates = [HasNEON] in {
5517 def : Pat<(xor (v4i32 (bitconvert (v2i64 (abd_shr (v2i32 DPR:$opA), (v2i32 DPR:$opB), 63)))),
5518 (v4i32 (bitconvert (v2i64 (add (sub (zext (v2i32 DPR:$opA)),
5519 (zext (v2i32 DPR:$opB))),
5520 (abd_shr (v2i32 DPR:$opA), (v2i32 DPR:$opB), 63)))))),
5521 (VABDLuv2i64 DPR:$opA, DPR:$opB)>;
5524 // VABA : Vector Absolute Difference and Accumulate
5525 defm VABAs : N3VIntOp_QHS<0,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
5526 "vaba", "s", int_arm_neon_vabds, add>;
5527 defm VABAu : N3VIntOp_QHS<1,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
5528 "vaba", "u", int_arm_neon_vabdu, add>;
5530 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
5531 defm VABALs : N3VLIntExtOp_QHS<0,1,0b0101,0, IIC_VABAD,
5532 "vabal", "s", int_arm_neon_vabds, zext, add>;
5533 defm VABALu : N3VLIntExtOp_QHS<1,1,0b0101,0, IIC_VABAD,
5534 "vabal", "u", int_arm_neon_vabdu, zext, add>;
5536 // Vector Maximum and Minimum.
5538 // VMAX : Vector Maximum
5539 defm VMAXs : N3VInt_QHS<0, 0, 0b0110, 0, N3RegFrm,
5540 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
5541 "vmax", "s", smax, 1>;
5542 defm VMAXu : N3VInt_QHS<1, 0, 0b0110, 0, N3RegFrm,
5543 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
5544 "vmax", "u", umax, 1>;
5545 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBIND,
5547 v2f32, v2f32, fmaximum, 1>;
5548 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBINQ,
5550 v4f32, v4f32, fmaximum, 1>;
5551 def VMAXhd : N3VDInt<0, 0, 0b01, 0b1111, 0, N3RegFrm, IIC_VBIND,
5553 v4f16, v4f16, fmaximum, 1>,
5554 Requires<[HasNEON, HasFullFP16]>;
5555 def VMAXhq : N3VQInt<0, 0, 0b01, 0b1111, 0, N3RegFrm, IIC_VBINQ,
5557 v8f16, v8f16, fmaximum, 1>,
5558 Requires<[HasNEON, HasFullFP16]>;
5561 let PostEncoderMethod = "NEONThumb2V8PostEncoder", DecoderNamespace = "v8NEON" in {
5562 def NEON_VMAXNMNDf : N3VDIntnp<0b00110, 0b00, 0b1111, 0, 1,
5563 N3RegFrm, NoItinerary, "vmaxnm", "f32",
5564 v2f32, v2f32, fmaxnum, 1>,
5565 Requires<[HasV8, HasNEON]>;
5566 def NEON_VMAXNMNQf : N3VQIntnp<0b00110, 0b00, 0b1111, 1, 1,
5567 N3RegFrm, NoItinerary, "vmaxnm", "f32",
5568 v4f32, v4f32, fmaxnum, 1>,
5569 Requires<[HasV8, HasNEON]>;
5570 def NEON_VMAXNMNDh : N3VDIntnp<0b00110, 0b01, 0b1111, 0, 1,
5571 N3RegFrm, NoItinerary, "vmaxnm", "f16",
5572 v4f16, v4f16, fmaxnum, 1>,
5573 Requires<[HasV8, HasNEON, HasFullFP16]>;
5574 def NEON_VMAXNMNQh : N3VQIntnp<0b00110, 0b01, 0b1111, 1, 1,
5575 N3RegFrm, NoItinerary, "vmaxnm", "f16",
5576 v8f16, v8f16, fmaxnum, 1>,
5577 Requires<[HasV8, HasNEON, HasFullFP16]>;
5580 // VMIN : Vector Minimum
5581 defm VMINs : N3VInt_QHS<0, 0, 0b0110, 1, N3RegFrm,
5582 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
5583 "vmin", "s", smin, 1>;
5584 defm VMINu : N3VInt_QHS<1, 0, 0b0110, 1, N3RegFrm,
5585 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
5586 "vmin", "u", umin, 1>;
5587 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBIND,
5589 v2f32, v2f32, fminimum, 1>;
5590 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBINQ,
5592 v4f32, v4f32, fminimum, 1>;
5593 def VMINhd : N3VDInt<0, 0, 0b11, 0b1111, 0, N3RegFrm, IIC_VBIND,
5595 v4f16, v4f16, fminimum, 1>,
5596 Requires<[HasNEON, HasFullFP16]>;
5597 def VMINhq : N3VQInt<0, 0, 0b11, 0b1111, 0, N3RegFrm, IIC_VBINQ,
5599 v8f16, v8f16, fminimum, 1>,
5600 Requires<[HasNEON, HasFullFP16]>;
5603 let PostEncoderMethod = "NEONThumb2V8PostEncoder", DecoderNamespace = "v8NEON" in {
5604 def NEON_VMINNMNDf : N3VDIntnp<0b00110, 0b10, 0b1111, 0, 1,
5605 N3RegFrm, NoItinerary, "vminnm", "f32",
5606 v2f32, v2f32, fminnum, 1>,
5607 Requires<[HasV8, HasNEON]>;
5608 def NEON_VMINNMNQf : N3VQIntnp<0b00110, 0b10, 0b1111, 1, 1,
5609 N3RegFrm, NoItinerary, "vminnm", "f32",
5610 v4f32, v4f32, fminnum, 1>,
5611 Requires<[HasV8, HasNEON]>;
5612 def NEON_VMINNMNDh : N3VDIntnp<0b00110, 0b11, 0b1111, 0, 1,
5613 N3RegFrm, NoItinerary, "vminnm", "f16",
5614 v4f16, v4f16, fminnum, 1>,
5615 Requires<[HasV8, HasNEON, HasFullFP16]>;
5616 def NEON_VMINNMNQh : N3VQIntnp<0b00110, 0b11, 0b1111, 1, 1,
5617 N3RegFrm, NoItinerary, "vminnm", "f16",
5618 v8f16, v8f16, fminnum, 1>,
5619 Requires<[HasV8, HasNEON, HasFullFP16]>;
5622 // Vector Pairwise Operations.
5624 // VPADD : Vector Pairwise Add
5625 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
5627 v8i8, v8i8, int_arm_neon_vpadd, 0>;
5628 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
5630 v4i16, v4i16, int_arm_neon_vpadd, 0>;
5631 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
5633 v2i32, v2i32, int_arm_neon_vpadd, 0>;
5634 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, N3RegFrm,
5635 IIC_VPBIND, "vpadd", "f32",
5636 v2f32, v2f32, int_arm_neon_vpadd, 0>;
5637 def VPADDh : N3VDInt<1, 0, 0b01, 0b1101, 0, N3RegFrm,
5638 IIC_VPBIND, "vpadd", "f16",
5639 v4f16, v4f16, int_arm_neon_vpadd, 0>,
5640 Requires<[HasNEON, HasFullFP16]>;
5642 // VPADDL : Vector Pairwise Add Long
5643 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
5644 int_arm_neon_vpaddls>;
5645 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
5646 int_arm_neon_vpaddlu>;
5648 // VPADAL : Vector Pairwise Add and Accumulate Long
5649 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
5650 int_arm_neon_vpadals>;
5651 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
5652 int_arm_neon_vpadalu>;
5654 // VPMAX : Vector Pairwise Maximum
5655 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
5656 "s8", v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
5657 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
5658 "s16", v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
5659 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
5660 "s32", v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
5661 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
5662 "u8", v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
5663 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
5664 "u16", v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
5665 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
5666 "u32", v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
5667 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmax",
5668 "f32", v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
5669 def VPMAXh : N3VDInt<1, 0, 0b01, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmax",
5670 "f16", v4f16, v4f16, int_arm_neon_vpmaxs, 0>,
5671 Requires<[HasNEON, HasFullFP16]>;
5673 // VPMIN : Vector Pairwise Minimum
5674 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
5675 "s8", v8i8, v8i8, int_arm_neon_vpmins, 0>;
5676 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
5677 "s16", v4i16, v4i16, int_arm_neon_vpmins, 0>;
5678 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
5679 "s32", v2i32, v2i32, int_arm_neon_vpmins, 0>;
5680 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
5681 "u8", v8i8, v8i8, int_arm_neon_vpminu, 0>;
5682 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
5683 "u16", v4i16, v4i16, int_arm_neon_vpminu, 0>;
5684 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
5685 "u32", v2i32, v2i32, int_arm_neon_vpminu, 0>;
5686 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmin",
5687 "f32", v2f32, v2f32, int_arm_neon_vpmins, 0>;
5688 def VPMINh : N3VDInt<1, 0, 0b11, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmin",
5689 "f16", v4f16, v4f16, int_arm_neon_vpmins, 0>,
5690 Requires<[HasNEON, HasFullFP16]>;
5692 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
5694 // VRECPE : Vector Reciprocal Estimate
5695 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
5696 IIC_VUNAD, "vrecpe", "u32",
5697 v2i32, v2i32, int_arm_neon_vrecpe>;
5698 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
5699 IIC_VUNAQ, "vrecpe", "u32",
5700 v4i32, v4i32, int_arm_neon_vrecpe>;
5701 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
5702 IIC_VUNAD, "vrecpe", "f32",
5703 v2f32, v2f32, int_arm_neon_vrecpe>;
5704 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
5705 IIC_VUNAQ, "vrecpe", "f32",
5706 v4f32, v4f32, int_arm_neon_vrecpe>;
5707 def VRECPEhd : N2VDInt<0b11, 0b11, 0b01, 0b11, 0b01010, 0,
5708 IIC_VUNAD, "vrecpe", "f16",
5709 v4f16, v4f16, int_arm_neon_vrecpe>,
5710 Requires<[HasNEON, HasFullFP16]>;
5711 def VRECPEhq : N2VQInt<0b11, 0b11, 0b01, 0b11, 0b01010, 0,
5712 IIC_VUNAQ, "vrecpe", "f16",
5713 v8f16, v8f16, int_arm_neon_vrecpe>,
5714 Requires<[HasNEON, HasFullFP16]>;
5716 // VRECPS : Vector Reciprocal Step
5717 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
5718 IIC_VRECSD, "vrecps", "f32",
5719 v2f32, v2f32, int_arm_neon_vrecps, 1>;
5720 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
5721 IIC_VRECSQ, "vrecps", "f32",
5722 v4f32, v4f32, int_arm_neon_vrecps, 1>;
5723 def VRECPShd : N3VDInt<0, 0, 0b01, 0b1111, 1, N3RegFrm,
5724 IIC_VRECSD, "vrecps", "f16",
5725 v4f16, v4f16, int_arm_neon_vrecps, 1>,
5726 Requires<[HasNEON, HasFullFP16]>;
5727 def VRECPShq : N3VQInt<0, 0, 0b01, 0b1111, 1, N3RegFrm,
5728 IIC_VRECSQ, "vrecps", "f16",
5729 v8f16, v8f16, int_arm_neon_vrecps, 1>,
5730 Requires<[HasNEON, HasFullFP16]>;
5732 // VRSQRTE : Vector Reciprocal Square Root Estimate
5733 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
5734 IIC_VUNAD, "vrsqrte", "u32",
5735 v2i32, v2i32, int_arm_neon_vrsqrte>;
5736 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
5737 IIC_VUNAQ, "vrsqrte", "u32",
5738 v4i32, v4i32, int_arm_neon_vrsqrte>;
5739 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
5740 IIC_VUNAD, "vrsqrte", "f32",
5741 v2f32, v2f32, int_arm_neon_vrsqrte>;
5742 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
5743 IIC_VUNAQ, "vrsqrte", "f32",
5744 v4f32, v4f32, int_arm_neon_vrsqrte>;
5745 def VRSQRTEhd : N2VDInt<0b11, 0b11, 0b01, 0b11, 0b01011, 0,
5746 IIC_VUNAD, "vrsqrte", "f16",
5747 v4f16, v4f16, int_arm_neon_vrsqrte>,
5748 Requires<[HasNEON, HasFullFP16]>;
5749 def VRSQRTEhq : N2VQInt<0b11, 0b11, 0b01, 0b11, 0b01011, 0,
5750 IIC_VUNAQ, "vrsqrte", "f16",
5751 v8f16, v8f16, int_arm_neon_vrsqrte>,
5752 Requires<[HasNEON, HasFullFP16]>;
5754 // VRSQRTS : Vector Reciprocal Square Root Step
5755 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
5756 IIC_VRECSD, "vrsqrts", "f32",
5757 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
5758 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
5759 IIC_VRECSQ, "vrsqrts", "f32",
5760 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
5761 def VRSQRTShd : N3VDInt<0, 0, 0b11, 0b1111, 1, N3RegFrm,
5762 IIC_VRECSD, "vrsqrts", "f16",
5763 v4f16, v4f16, int_arm_neon_vrsqrts, 1>,
5764 Requires<[HasNEON, HasFullFP16]>;
5765 def VRSQRTShq : N3VQInt<0, 0, 0b11, 0b1111, 1, N3RegFrm,
5766 IIC_VRECSQ, "vrsqrts", "f16",
5767 v8f16, v8f16, int_arm_neon_vrsqrts, 1>,
5768 Requires<[HasNEON, HasFullFP16]>;
5772 // VSHL : Vector Shift
5773 defm VSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 0, N3RegVShFrm,
5774 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
5775 "vshl", "s", int_arm_neon_vshifts>;
5776 defm VSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 0, N3RegVShFrm,
5777 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
5778 "vshl", "u", int_arm_neon_vshiftu>;
5780 let Predicates = [HasNEON] in {
5781 def : Pat<(v8i8 (ARMvshls (v8i8 DPR:$Dn), (v8i8 DPR:$Dm))),
5782 (VSHLsv8i8 DPR:$Dn, DPR:$Dm)>;
5783 def : Pat<(v4i16 (ARMvshls (v4i16 DPR:$Dn), (v4i16 DPR:$Dm))),
5784 (VSHLsv4i16 DPR:$Dn, DPR:$Dm)>;
5785 def : Pat<(v2i32 (ARMvshls (v2i32 DPR:$Dn), (v2i32 DPR:$Dm))),
5786 (VSHLsv2i32 DPR:$Dn, DPR:$Dm)>;
5787 def : Pat<(v1i64 (ARMvshls (v1i64 DPR:$Dn), (v1i64 DPR:$Dm))),
5788 (VSHLsv1i64 DPR:$Dn, DPR:$Dm)>;
5789 def : Pat<(v16i8 (ARMvshls (v16i8 QPR:$Dn), (v16i8 QPR:$Dm))),
5790 (VSHLsv16i8 QPR:$Dn, QPR:$Dm)>;
5791 def : Pat<(v8i16 (ARMvshls (v8i16 QPR:$Dn), (v8i16 QPR:$Dm))),
5792 (VSHLsv8i16 QPR:$Dn, QPR:$Dm)>;
5793 def : Pat<(v4i32 (ARMvshls (v4i32 QPR:$Dn), (v4i32 QPR:$Dm))),
5794 (VSHLsv4i32 QPR:$Dn, QPR:$Dm)>;
5795 def : Pat<(v2i64 (ARMvshls (v2i64 QPR:$Dn), (v2i64 QPR:$Dm))),
5796 (VSHLsv2i64 QPR:$Dn, QPR:$Dm)>;
5798 def : Pat<(v8i8 (ARMvshlu (v8i8 DPR:$Dn), (v8i8 DPR:$Dm))),
5799 (VSHLuv8i8 DPR:$Dn, DPR:$Dm)>;
5800 def : Pat<(v4i16 (ARMvshlu (v4i16 DPR:$Dn), (v4i16 DPR:$Dm))),
5801 (VSHLuv4i16 DPR:$Dn, DPR:$Dm)>;
5802 def : Pat<(v2i32 (ARMvshlu (v2i32 DPR:$Dn), (v2i32 DPR:$Dm))),
5803 (VSHLuv2i32 DPR:$Dn, DPR:$Dm)>;
5804 def : Pat<(v1i64 (ARMvshlu (v1i64 DPR:$Dn), (v1i64 DPR:$Dm))),
5805 (VSHLuv1i64 DPR:$Dn, DPR:$Dm)>;
5806 def : Pat<(v16i8 (ARMvshlu (v16i8 QPR:$Dn), (v16i8 QPR:$Dm))),
5807 (VSHLuv16i8 QPR:$Dn, QPR:$Dm)>;
5808 def : Pat<(v8i16 (ARMvshlu (v8i16 QPR:$Dn), (v8i16 QPR:$Dm))),
5809 (VSHLuv8i16 QPR:$Dn, QPR:$Dm)>;
5810 def : Pat<(v4i32 (ARMvshlu (v4i32 QPR:$Dn), (v4i32 QPR:$Dm))),
5811 (VSHLuv4i32 QPR:$Dn, QPR:$Dm)>;
5812 def : Pat<(v2i64 (ARMvshlu (v2i64 QPR:$Dn), (v2i64 QPR:$Dm))),
5813 (VSHLuv2i64 QPR:$Dn, QPR:$Dm)>;
5817 // VSHL : Vector Shift Left (Immediate)
5818 defm VSHLi : N2VShL_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", ARMvshlImm>;
5820 // VSHR : Vector Shift Right (Immediate)
5821 defm VSHRs : N2VShR_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s", "VSHRs",
5823 defm VSHRu : N2VShR_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u", "VSHRu",
5826 // VSHLL : Vector Shift Left Long
5827 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s",
5828 PatFrag<(ops node:$LHS, node:$RHS), (ARMvshlImm (sext node:$LHS), node:$RHS)>>;
5829 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u",
5830 PatFrag<(ops node:$LHS, node:$RHS), (ARMvshlImm (zext node:$LHS), node:$RHS)>>;
5832 // VSHLL : Vector Shift Left Long (with maximum shift count)
5833 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
5834 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
5835 ValueType OpTy, Operand ImmTy>
5836 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
5837 ResTy, OpTy, ImmTy, null_frag> {
5838 let Inst{21-16} = op21_16;
5839 let DecoderMethod = "DecodeVSHLMaxInstruction";
5841 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
5843 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
5844 v4i32, v4i16, imm16>;
5845 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
5846 v2i64, v2i32, imm32>;
5848 let Predicates = [HasNEON] in {
5849 def : Pat<(v8i16 (ARMvshlImm (zext (v8i8 DPR:$Rn)), (i32 8))),
5850 (VSHLLi8 DPR:$Rn, 8)>;
5851 def : Pat<(v4i32 (ARMvshlImm (zext (v4i16 DPR:$Rn)), (i32 16))),
5852 (VSHLLi16 DPR:$Rn, 16)>;
5853 def : Pat<(v2i64 (ARMvshlImm (zext (v2i32 DPR:$Rn)), (i32 32))),
5854 (VSHLLi32 DPR:$Rn, 32)>;
5855 def : Pat<(v8i16 (ARMvshlImm (sext (v8i8 DPR:$Rn)), (i32 8))),
5856 (VSHLLi8 DPR:$Rn, 8)>;
5857 def : Pat<(v4i32 (ARMvshlImm (sext (v4i16 DPR:$Rn)), (i32 16))),
5858 (VSHLLi16 DPR:$Rn, 16)>;
5859 def : Pat<(v2i64 (ARMvshlImm (sext (v2i32 DPR:$Rn)), (i32 32))),
5860 (VSHLLi32 DPR:$Rn, 32)>;
5861 def : Pat<(v8i16 (ARMvshlImm (anyext (v8i8 DPR:$Rn)), (i32 8))),
5862 (VSHLLi8 DPR:$Rn, 8)>;
5863 def : Pat<(v4i32 (ARMvshlImm (anyext (v4i16 DPR:$Rn)), (i32 16))),
5864 (VSHLLi16 DPR:$Rn, 16)>;
5865 def : Pat<(v2i64 (ARMvshlImm (anyext (v2i32 DPR:$Rn)), (i32 32))),
5866 (VSHLLi32 DPR:$Rn, 32)>;
5869 // VSHRN : Vector Shift Right and Narrow
5870 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
5871 PatFrag<(ops node:$Rn, node:$amt),
5872 (trunc (ARMvshrsImm node:$Rn, node:$amt))>>;
5874 let Predicates = [HasNEON] in {
5875 def : Pat<(v8i8 (trunc (ARMvshruImm (v8i16 QPR:$Vn), shr_imm8:$amt))),
5876 (VSHRNv8i8 QPR:$Vn, shr_imm8:$amt)>;
5877 def : Pat<(v4i16 (trunc (ARMvshruImm (v4i32 QPR:$Vn), shr_imm16:$amt))),
5878 (VSHRNv4i16 QPR:$Vn, shr_imm16:$amt)>;
5879 def : Pat<(v2i32 (trunc (ARMvshruImm (v2i64 QPR:$Vn), shr_imm32:$amt))),
5880 (VSHRNv2i32 QPR:$Vn, shr_imm32:$amt)>;
5883 // VRSHL : Vector Rounding Shift
5884 defm VRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 0, N3RegVShFrm,
5885 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
5886 "vrshl", "s", int_arm_neon_vrshifts>;
5887 defm VRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 0, N3RegVShFrm,
5888 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
5889 "vrshl", "u", int_arm_neon_vrshiftu>;
5890 // VRSHR : Vector Rounding Shift Right
5891 defm VRSHRs : N2VShR_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s", "VRSHRs",
5893 defm VRSHRu : N2VShR_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u", "VRSHRu",
5896 // VRSHRN : Vector Rounding Shift Right and Narrow
5897 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
5900 // VQSHL : Vector Saturating Shift
5901 defm VQSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 1, N3RegVShFrm,
5902 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
5903 "vqshl", "s", int_arm_neon_vqshifts>;
5904 defm VQSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 1, N3RegVShFrm,
5905 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
5906 "vqshl", "u", int_arm_neon_vqshiftu>;
5907 // VQSHL : Vector Saturating Shift Left (Immediate)
5908 defm VQSHLsi : N2VShL_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s",NEONvqshlsImm>;
5909 defm VQSHLui : N2VShL_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u",NEONvqshluImm>;
5911 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
5912 defm VQSHLsu : N2VShL_QHSD<1,1,0b0110,1, IIC_VSHLi4D,"vqshlu","s",NEONvqshlsuImm>;
5914 // VQSHRN : Vector Saturating Shift Right and Narrow
5915 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
5917 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
5920 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
5921 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
5924 // VQRSHL : Vector Saturating Rounding Shift
5925 defm VQRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 1, N3RegVShFrm,
5926 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
5927 "vqrshl", "s", int_arm_neon_vqrshifts>;
5928 defm VQRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 1, N3RegVShFrm,
5929 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
5930 "vqrshl", "u", int_arm_neon_vqrshiftu>;
5932 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
5933 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
5935 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
5938 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
5939 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
5942 // VSRA : Vector Shift Right and Accumulate
5943 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", ARMvshrsImm>;
5944 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", ARMvshruImm>;
5945 // VRSRA : Vector Rounding Shift Right and Accumulate
5946 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrsImm>;
5947 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshruImm>;
5949 // VSLI : Vector Shift Left and Insert
5950 defm VSLI : N2VShInsL_QHSD<1, 1, 0b0101, 1, "vsli">;
5952 // VSRI : Vector Shift Right and Insert
5953 defm VSRI : N2VShInsR_QHSD<1, 1, 0b0100, 1, "vsri">;
5955 // Vector Absolute and Saturating Absolute.
5957 // VABS : Vector Absolute Value
5958 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
5959 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s", abs>;
5960 def VABSfd : N2VD<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
5962 v2f32, v2f32, fabs>;
5963 def VABSfq : N2VQ<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
5965 v4f32, v4f32, fabs>;
5966 def VABShd : N2VD<0b11, 0b11, 0b01, 0b01, 0b01110, 0,
5968 v4f16, v4f16, fabs>,
5969 Requires<[HasNEON, HasFullFP16]>;
5970 def VABShq : N2VQ<0b11, 0b11, 0b01, 0b01, 0b01110, 0,
5972 v8f16, v8f16, fabs>,
5973 Requires<[HasNEON, HasFullFP16]>;
5975 // VQABS : Vector Saturating Absolute Value
5976 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
5977 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
5978 int_arm_neon_vqabs>;
5982 def vnegd : PatFrag<(ops node:$in),
5983 (sub (bitconvert (v2i32 NEONimmAllZerosV)), node:$in)>;
5984 def vnegq : PatFrag<(ops node:$in),
5985 (sub (bitconvert (v4i32 NEONimmAllZerosV)), node:$in)>;
5987 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
5988 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$Vd), (ins DPR:$Vm),
5989 IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
5990 [(set DPR:$Vd, (Ty (vnegd DPR:$Vm)))]>;
5991 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
5992 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$Vd), (ins QPR:$Vm),
5993 IIC_VSHLiQ, OpcodeStr, Dt, "$Vd, $Vm", "",
5994 [(set QPR:$Vd, (Ty (vnegq QPR:$Vm)))]>;
5996 // VNEG : Vector Negate (integer)
5997 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
5998 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
5999 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
6000 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
6001 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
6002 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
6004 // VNEG : Vector Negate (floating-point)
6005 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
6006 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VUNAD,
6007 "vneg", "f32", "$Vd, $Vm", "",
6008 [(set DPR:$Vd, (v2f32 (fneg DPR:$Vm)))]>;
6009 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
6010 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VUNAQ,
6011 "vneg", "f32", "$Vd, $Vm", "",
6012 [(set QPR:$Vd, (v4f32 (fneg QPR:$Vm)))]>;
6013 def VNEGhd : N2V<0b11, 0b11, 0b01, 0b01, 0b01111, 0, 0,
6014 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VUNAD,
6015 "vneg", "f16", "$Vd, $Vm", "",
6016 [(set DPR:$Vd, (v4f16 (fneg DPR:$Vm)))]>,
6017 Requires<[HasNEON, HasFullFP16]>;
6018 def VNEGhq : N2V<0b11, 0b11, 0b01, 0b01, 0b01111, 1, 0,
6019 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VUNAQ,
6020 "vneg", "f16", "$Vd, $Vm", "",
6021 [(set QPR:$Vd, (v8f16 (fneg QPR:$Vm)))]>,
6022 Requires<[HasNEON, HasFullFP16]>;
6024 let Predicates = [HasNEON] in {
6025 def : Pat<(v8i8 (vnegd DPR:$src)), (VNEGs8d DPR:$src)>;
6026 def : Pat<(v4i16 (vnegd DPR:$src)), (VNEGs16d DPR:$src)>;
6027 def : Pat<(v2i32 (vnegd DPR:$src)), (VNEGs32d DPR:$src)>;
6028 def : Pat<(v16i8 (vnegq QPR:$src)), (VNEGs8q QPR:$src)>;
6029 def : Pat<(v8i16 (vnegq QPR:$src)), (VNEGs16q QPR:$src)>;
6030 def : Pat<(v4i32 (vnegq QPR:$src)), (VNEGs32q QPR:$src)>;
6033 // VQNEG : Vector Saturating Negate
6034 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
6035 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
6036 int_arm_neon_vqneg>;
6038 // Vector Bit Counting Operations.
6040 // VCLS : Vector Count Leading Sign Bits
6041 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
6042 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
6044 // VCLZ : Vector Count Leading Zeros
6045 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
6046 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
6048 // VCNT : Vector Count One Bits
6049 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
6050 IIC_VCNTiD, "vcnt", "8",
6052 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
6053 IIC_VCNTiQ, "vcnt", "8",
6054 v16i8, v16i8, ctpop>;
6057 def VSWPd : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 0, 0,
6058 (outs DPR:$Vd, DPR:$Vm), (ins DPR:$in1, DPR:$in2),
6059 NoItinerary, "vswp", "$Vd, $Vm", "$in1 = $Vd, $in2 = $Vm",
6061 def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0,
6062 (outs QPR:$Vd, QPR:$Vm), (ins QPR:$in1, QPR:$in2),
6063 NoItinerary, "vswp", "$Vd, $Vm", "$in1 = $Vd, $in2 = $Vm",
6066 // Vector Move Operations.
6068 // VMOV : Vector Move (Register)
6069 def : NEONInstAlias<"vmov${p} $Vd, $Vm",
6070 (VORRd DPR:$Vd, DPR:$Vm, DPR:$Vm, pred:$p)>;
6071 def : NEONInstAlias<"vmov${p} $Vd, $Vm",
6072 (VORRq QPR:$Vd, QPR:$Vm, QPR:$Vm, pred:$p)>;
6074 // VMOV : Vector Move (Immediate)
6076 // Although VMOVs are not strictly speaking cheap, they are as expensive
6077 // as their copies counterpart (VORR), so we should prefer rematerialization
6078 // over splitting when it applies.
6079 let isReMaterializable = 1, isAsCheapAsAMove=1 in {
6080 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$Vd),
6081 (ins nImmSplatI8:$SIMM), IIC_VMOVImm,
6082 "vmov", "i8", "$Vd, $SIMM", "",
6083 [(set DPR:$Vd, (v8i8 (ARMvmovImm timm:$SIMM)))]>;
6084 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$Vd),
6085 (ins nImmSplatI8:$SIMM), IIC_VMOVImm,
6086 "vmov", "i8", "$Vd, $SIMM", "",
6087 [(set QPR:$Vd, (v16i8 (ARMvmovImm timm:$SIMM)))]>;
6089 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 0, 1, (outs DPR:$Vd),
6090 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
6091 "vmov", "i16", "$Vd, $SIMM", "",
6092 [(set DPR:$Vd, (v4i16 (ARMvmovImm timm:$SIMM)))]> {
6093 let Inst{9} = SIMM{9};
6096 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 0, 1, (outs QPR:$Vd),
6097 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
6098 "vmov", "i16", "$Vd, $SIMM", "",
6099 [(set QPR:$Vd, (v8i16 (ARMvmovImm timm:$SIMM)))]> {
6100 let Inst{9} = SIMM{9};
6103 def VMOVv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 0, 1, (outs DPR:$Vd),
6104 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
6105 "vmov", "i32", "$Vd, $SIMM", "",
6106 [(set DPR:$Vd, (v2i32 (ARMvmovImm timm:$SIMM)))]> {
6107 let Inst{11-8} = SIMM{11-8};
6110 def VMOVv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 0, 1, (outs QPR:$Vd),
6111 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
6112 "vmov", "i32", "$Vd, $SIMM", "",
6113 [(set QPR:$Vd, (v4i32 (ARMvmovImm timm:$SIMM)))]> {
6114 let Inst{11-8} = SIMM{11-8};
6117 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$Vd),
6118 (ins nImmSplatI64:$SIMM), IIC_VMOVImm,
6119 "vmov", "i64", "$Vd, $SIMM", "",
6120 [(set DPR:$Vd, (v1i64 (ARMvmovImm timm:$SIMM)))]>;
6121 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$Vd),
6122 (ins nImmSplatI64:$SIMM), IIC_VMOVImm,
6123 "vmov", "i64", "$Vd, $SIMM", "",
6124 [(set QPR:$Vd, (v2i64 (ARMvmovImm timm:$SIMM)))]>;
6126 def VMOVv2f32 : N1ModImm<1, 0b000, 0b1111, 0, 0, 0, 1, (outs DPR:$Vd),
6127 (ins nImmVMOVF32:$SIMM), IIC_VMOVImm,
6128 "vmov", "f32", "$Vd, $SIMM", "",
6129 [(set DPR:$Vd, (v2f32 (ARMvmovFPImm timm:$SIMM)))]>;
6130 def VMOVv4f32 : N1ModImm<1, 0b000, 0b1111, 0, 1, 0, 1, (outs QPR:$Vd),
6131 (ins nImmVMOVF32:$SIMM), IIC_VMOVImm,
6132 "vmov", "f32", "$Vd, $SIMM", "",
6133 [(set QPR:$Vd, (v4f32 (ARMvmovFPImm timm:$SIMM)))]>;
6134 } // isReMaterializable, isAsCheapAsAMove
6136 // Add support for bytes replication feature, so it could be GAS compatible.
6137 multiclass NEONImmReplicateI8InstAlias<ValueType To> {
6138 // E.g. instructions below:
6139 // "vmov.i32 d0, #0xffffffff"
6140 // "vmov.i32 d0, #0xabababab"
6141 // "vmov.i16 d0, #0xabab"
6142 // are incorrect, but we could deal with such cases.
6143 // For last two instructions, for example, it should emit:
6144 // "vmov.i8 d0, #0xab"
6145 def : NEONInstAlias<"vmov${p}.i" # To.Size # " $Vd, $Vm",
6146 (VMOVv8i8 DPR:$Vd, nImmVMOVIReplicate<i8, To>:$Vm, pred:$p)>;
6147 def : NEONInstAlias<"vmov${p}.i" # To.Size # " $Vd, $Vm",
6148 (VMOVv16i8 QPR:$Vd, nImmVMOVIReplicate<i8, To>:$Vm, pred:$p)>;
6149 // Also add same support for VMVN instructions. So instruction:
6150 // "vmvn.i32 d0, #0xabababab"
6152 // "vmov.i8 d0, #0x54"
6153 def : NEONInstAlias<"vmvn${p}.i" # To.Size # " $Vd, $Vm",
6154 (VMOVv8i8 DPR:$Vd, nImmVINVIReplicate<i8, To>:$Vm, pred:$p)>;
6155 def : NEONInstAlias<"vmvn${p}.i" # To.Size # " $Vd, $Vm",
6156 (VMOVv16i8 QPR:$Vd, nImmVINVIReplicate<i8, To>:$Vm, pred:$p)>;
6159 defm : NEONImmReplicateI8InstAlias<i16>;
6160 defm : NEONImmReplicateI8InstAlias<i32>;
6161 defm : NEONImmReplicateI8InstAlias<i64>;
6163 // Similar to above for types other than i8, e.g.:
6164 // "vmov.i32 d0, #0xab00ab00" -> "vmov.i16 d0, #0xab00"
6165 // "vmvn.i64 q0, #0xab000000ab000000" -> "vmvn.i32 q0, #0xab000000"
6166 // In this case we do not canonicalize VMVN to VMOV
6167 multiclass NEONImmReplicateInstAlias<ValueType From, NeonI V8, NeonI V16,
6168 NeonI NV8, NeonI NV16, ValueType To> {
6169 def : NEONInstAlias<"vmov${p}.i" # To.Size # " $Vd, $Vm",
6170 (V8 DPR:$Vd, nImmVMOVIReplicate<From, To>:$Vm, pred:$p)>;
6171 def : NEONInstAlias<"vmov${p}.i" # To.Size # " $Vd, $Vm",
6172 (V16 QPR:$Vd, nImmVMOVIReplicate<From, To>:$Vm, pred:$p)>;
6173 def : NEONInstAlias<"vmvn${p}.i" # To.Size # " $Vd, $Vm",
6174 (NV8 DPR:$Vd, nImmVMOVIReplicate<From, To>:$Vm, pred:$p)>;
6175 def : NEONInstAlias<"vmvn${p}.i" # To.Size # " $Vd, $Vm",
6176 (NV16 QPR:$Vd, nImmVMOVIReplicate<From, To>:$Vm, pred:$p)>;
6179 defm : NEONImmReplicateInstAlias<i16, VMOVv4i16, VMOVv8i16,
6180 VMVNv4i16, VMVNv8i16, i32>;
6181 defm : NEONImmReplicateInstAlias<i16, VMOVv4i16, VMOVv8i16,
6182 VMVNv4i16, VMVNv8i16, i64>;
6183 defm : NEONImmReplicateInstAlias<i32, VMOVv2i32, VMOVv4i32,
6184 VMVNv2i32, VMVNv4i32, i64>;
6185 // TODO: add "VMOV <-> VMVN" conversion for cases like
6186 // "vmov.i32 d0, #0xffaaffaa" -> "vmvn.i16 d0, #0x55"
6187 // "vmvn.i32 d0, #0xaaffaaff" -> "vmov.i16 d0, #0xff00"
6189 // On some CPUs the two instructions "vmov.i32 dD, #0" and "vmov.i32 qD, #0"
6190 // require zero cycles to execute so they should be used wherever possible for
6191 // setting a register to zero.
6193 // Even without these pseudo-insts we would probably end up with the correct
6194 // instruction, but we could not mark the general ones with "isAsCheapAsAMove"
6195 // since they are sometimes rather expensive (in general).
6197 let AddedComplexity = 50, isAsCheapAsAMove = 1, isReMaterializable = 1 in {
6198 def VMOVD0 : ARMPseudoExpand<(outs DPR:$Vd), (ins), 4, IIC_VMOVImm,
6199 [(set DPR:$Vd, (v2i32 NEONimmAllZerosV))],
6200 (VMOVv2i32 DPR:$Vd, 0, (ops 14, zero_reg))>,
6202 def VMOVQ0 : ARMPseudoExpand<(outs QPR:$Vd), (ins), 4, IIC_VMOVImm,
6203 [(set QPR:$Vd, (v4i32 NEONimmAllZerosV))],
6204 (VMOVv4i32 QPR:$Vd, 0, (ops 14, zero_reg))>,
6208 // VMOV : Vector Get Lane (move scalar to ARM core register)
6210 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
6211 (outs GPR:$R), (ins DPR:$V, VectorIndex8:$lane),
6212 IIC_VMOVSI, "vmov", "s8", "$R, $V$lane",
6213 [(set GPR:$R, (ARMvgetlanes (v8i8 DPR:$V),
6215 let Inst{21} = lane{2};
6216 let Inst{6-5} = lane{1-0};
6218 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
6219 (outs GPR:$R), (ins DPR:$V, VectorIndex16:$lane),
6220 IIC_VMOVSI, "vmov", "s16", "$R, $V$lane",
6221 [(set GPR:$R, (ARMvgetlanes (v4i16 DPR:$V),
6223 let Inst{21} = lane{1};
6224 let Inst{6} = lane{0};
6226 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
6227 (outs GPR:$R), (ins DPR:$V, VectorIndex8:$lane),
6228 IIC_VMOVSI, "vmov", "u8", "$R, $V$lane",
6229 [(set GPR:$R, (ARMvgetlaneu (v8i8 DPR:$V),
6231 let Inst{21} = lane{2};
6232 let Inst{6-5} = lane{1-0};
6234 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
6235 (outs GPR:$R), (ins DPR:$V, VectorIndex16:$lane),
6236 IIC_VMOVSI, "vmov", "u16", "$R, $V$lane",
6237 [(set GPR:$R, (ARMvgetlaneu (v4i16 DPR:$V),
6239 let Inst{21} = lane{1};
6240 let Inst{6} = lane{0};
6242 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
6243 (outs GPR:$R), (ins DPR:$V, VectorIndex32:$lane),
6244 IIC_VMOVSI, "vmov", "32", "$R, $V$lane",
6245 [(set GPR:$R, (extractelt (v2i32 DPR:$V),
6247 Requires<[HasFPRegs, HasFastVGETLNi32]> {
6248 let Inst{21} = lane{0};
6250 let Predicates = [HasNEON] in {
6251 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
6252 def : Pat<(ARMvgetlanes (v16i8 QPR:$src), imm:$lane),
6253 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
6254 (DSubReg_i8_reg imm:$lane))),
6255 (SubReg_i8_lane imm:$lane))>;
6256 def : Pat<(ARMvgetlanes (v8i16 QPR:$src), imm:$lane),
6257 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
6258 (DSubReg_i16_reg imm:$lane))),
6259 (SubReg_i16_lane imm:$lane))>;
6260 def : Pat<(ARMvgetlaneu (v16i8 QPR:$src), imm:$lane),
6261 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
6262 (DSubReg_i8_reg imm:$lane))),
6263 (SubReg_i8_lane imm:$lane))>;
6264 def : Pat<(ARMvgetlaneu (v8i16 QPR:$src), imm:$lane),
6265 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
6266 (DSubReg_i16_reg imm:$lane))),
6267 (SubReg_i16_lane imm:$lane))>;
6269 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
6270 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
6271 (DSubReg_i32_reg imm:$lane))),
6272 (SubReg_i32_lane imm:$lane))>,
6273 Requires<[HasNEON, HasFastVGETLNi32]>;
6274 def : Pat<(extractelt (v2i32 DPR:$src), imm:$lane),
6276 (i32 (EXTRACT_SUBREG DPR:$src, (SSubReg_f32_reg imm:$lane))), GPR)>,
6277 Requires<[HasNEON, HasSlowVGETLNi32]>;
6278 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
6280 (i32 (EXTRACT_SUBREG QPR:$src, (SSubReg_f32_reg imm:$lane))), GPR)>,
6281 Requires<[HasNEON, HasSlowVGETLNi32]>;
6282 let Predicates = [HasNEON] in {
6283 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
6284 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
6285 (SSubReg_f32_reg imm:$src2))>;
6286 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
6287 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
6288 (SSubReg_f32_reg imm:$src2))>;
6289 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
6290 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
6291 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
6292 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
6295 def imm_even : ImmLeaf<i32, [{ return (Imm & 1) == 0; }]>;
6296 def imm_odd : ImmLeaf<i32, [{ return (Imm & 1) == 1; }]>;
6298 let Predicates = [HasNEON] in {
6299 def : Pat<(extractelt (v4f16 DPR:$src), imm_even:$lane),
6301 (v2f32 (COPY_TO_REGCLASS (v4f16 DPR:$src), DPR_VFP2)),
6302 (SSubReg_f16_reg imm_even:$lane))>;
6304 def : Pat<(extractelt (v4f16 DPR:$src), imm_odd:$lane),
6306 (VMOVH (EXTRACT_SUBREG
6307 (v2f32 (COPY_TO_REGCLASS (v4f16 DPR:$src), DPR_VFP2)),
6308 (SSubReg_f16_reg imm_odd:$lane))),
6311 def : Pat<(extractelt (v8f16 QPR:$src), imm_even:$lane),
6313 (v4f32 (COPY_TO_REGCLASS (v8f16 QPR:$src), QPR_VFP2)),
6314 (SSubReg_f16_reg imm_even:$lane))>;
6316 def : Pat<(extractelt (v8f16 QPR:$src), imm_odd:$lane),
6318 (VMOVH (EXTRACT_SUBREG
6319 (v4f32 (COPY_TO_REGCLASS (v8f16 QPR:$src), QPR_VFP2)),
6320 (SSubReg_f16_reg imm_odd:$lane))),
6324 // VMOV : Vector Set Lane (move ARM core register to scalar)
6326 let Constraints = "$src1 = $V" in {
6327 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$V),
6328 (ins DPR:$src1, GPR:$R, VectorIndex8:$lane),
6329 IIC_VMOVISL, "vmov", "8", "$V$lane, $R",
6330 [(set DPR:$V, (vector_insert (v8i8 DPR:$src1),
6331 GPR:$R, imm:$lane))]> {
6332 let Inst{21} = lane{2};
6333 let Inst{6-5} = lane{1-0};
6335 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$V),
6336 (ins DPR:$src1, GPR:$R, VectorIndex16:$lane),
6337 IIC_VMOVISL, "vmov", "16", "$V$lane, $R",
6338 [(set DPR:$V, (vector_insert (v4i16 DPR:$src1),
6339 GPR:$R, imm:$lane))]> {
6340 let Inst{21} = lane{1};
6341 let Inst{6} = lane{0};
6343 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$V),
6344 (ins DPR:$src1, GPR:$R, VectorIndex32:$lane),
6345 IIC_VMOVISL, "vmov", "32", "$V$lane, $R",
6346 [(set DPR:$V, (insertelt (v2i32 DPR:$src1),
6347 GPR:$R, imm:$lane))]>,
6348 Requires<[HasVFP2]> {
6349 let Inst{21} = lane{0};
6350 // This instruction is equivalent as
6351 // $V = INSERT_SUBREG $src1, $R, translateImmToSubIdx($imm)
6352 let isInsertSubreg = 1;
6356 let Predicates = [HasNEON] in {
6357 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
6358 (v16i8 (INSERT_SUBREG QPR:$src1,
6359 (v8i8 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
6360 (DSubReg_i8_reg imm:$lane))),
6361 GPR:$src2, (SubReg_i8_lane imm:$lane))),
6362 (DSubReg_i8_reg imm:$lane)))>;
6363 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
6364 (v8i16 (INSERT_SUBREG QPR:$src1,
6365 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
6366 (DSubReg_i16_reg imm:$lane))),
6367 GPR:$src2, (SubReg_i16_lane imm:$lane))),
6368 (DSubReg_i16_reg imm:$lane)))>;
6369 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
6370 (v4i32 (INSERT_SUBREG QPR:$src1,
6371 (v2i32 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
6372 (DSubReg_i32_reg imm:$lane))),
6373 GPR:$src2, (SubReg_i32_lane imm:$lane))),
6374 (DSubReg_i32_reg imm:$lane)))>;
6376 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
6377 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
6378 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
6379 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
6380 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
6381 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
6383 def : Pat<(insertelt (v4f16 DPR:$src1), HPR:$src2, imm:$lane),
6384 (v4f16 (VSETLNi16 DPR:$src1, (VMOVRH $src2), imm:$lane))>;
6385 def : Pat<(insertelt (v8f16 QPR:$src1), HPR:$src2, imm:$lane),
6386 (v8f16 (INSERT_SUBREG QPR:$src1,
6387 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
6388 (DSubReg_i16_reg imm:$lane))),
6389 (VMOVRH $src2), (SubReg_i16_lane imm:$lane))),
6390 (DSubReg_i16_reg imm:$lane)))>;
6392 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
6393 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
6394 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
6395 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
6397 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
6398 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
6399 def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
6400 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
6401 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
6402 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
6404 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
6405 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
6406 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
6407 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
6408 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
6409 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
6411 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
6412 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
6413 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
6415 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
6416 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
6417 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
6419 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
6420 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
6421 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
6425 // VDUP : Vector Duplicate (from ARM core register to all elements)
6427 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
6428 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$V), (ins GPR:$R),
6429 IIC_VMOVIS, "vdup", Dt, "$V, $R",
6430 [(set DPR:$V, (Ty (ARMvdup (i32 GPR:$R))))]>;
6431 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
6432 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$V), (ins GPR:$R),
6433 IIC_VMOVIS, "vdup", Dt, "$V, $R",
6434 [(set QPR:$V, (Ty (ARMvdup (i32 GPR:$R))))]>;
6436 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
6437 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
6438 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>,
6439 Requires<[HasNEON, HasFastVDUP32]>;
6440 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
6441 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
6442 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
6444 // ARMvdup patterns for uarchs with fast VDUP.32.
6445 def : Pat<(v2f32 (ARMvdup (f32 (bitconvert GPR:$R)))), (VDUP32d GPR:$R)>,
6446 Requires<[HasNEON,HasFastVDUP32]>;
6447 def : Pat<(v4f32 (ARMvdup (f32 (bitconvert GPR:$R)))), (VDUP32q GPR:$R)>,
6448 Requires<[HasNEON]>;
6450 // ARMvdup patterns for uarchs with slow VDUP.32 - use VMOVDRR instead.
6451 def : Pat<(v2i32 (ARMvdup (i32 GPR:$R))), (VMOVDRR GPR:$R, GPR:$R)>,
6452 Requires<[HasNEON,HasSlowVDUP32]>;
6453 def : Pat<(v2f32 (ARMvdup (f32 (bitconvert GPR:$R)))), (VMOVDRR GPR:$R, GPR:$R)>,
6454 Requires<[HasNEON,HasSlowVDUP32]>;
6456 // VDUP : Vector Duplicate Lane (from scalar to all elements)
6458 class VDUPLND<bits<4> op19_16, string OpcodeStr, string Dt,
6459 ValueType Ty, Operand IdxTy>
6460 : NVDupLane<op19_16, 0, (outs DPR:$Vd), (ins DPR:$Vm, IdxTy:$lane),
6461 IIC_VMOVD, OpcodeStr, Dt, "$Vd, $Vm$lane",
6462 [(set DPR:$Vd, (Ty (ARMvduplane (Ty DPR:$Vm), imm:$lane)))]>;
6464 class VDUPLNQ<bits<4> op19_16, string OpcodeStr, string Dt,
6465 ValueType ResTy, ValueType OpTy, Operand IdxTy>
6466 : NVDupLane<op19_16, 1, (outs QPR:$Vd), (ins DPR:$Vm, IdxTy:$lane),
6467 IIC_VMOVQ, OpcodeStr, Dt, "$Vd, $Vm$lane",
6468 [(set QPR:$Vd, (ResTy (ARMvduplane (OpTy DPR:$Vm),
6469 VectorIndex32:$lane)))]>;
6471 // Inst{19-16} is partially specified depending on the element size.
6473 def VDUPLN8d : VDUPLND<{?,?,?,1}, "vdup", "8", v8i8, VectorIndex8> {
6475 let Inst{19-17} = lane{2-0};
6477 def VDUPLN16d : VDUPLND<{?,?,1,0}, "vdup", "16", v4i16, VectorIndex16> {
6479 let Inst{19-18} = lane{1-0};
6481 def VDUPLN32d : VDUPLND<{?,1,0,0}, "vdup", "32", v2i32, VectorIndex32> {
6483 let Inst{19} = lane{0};
6485 def VDUPLN8q : VDUPLNQ<{?,?,?,1}, "vdup", "8", v16i8, v8i8, VectorIndex8> {
6487 let Inst{19-17} = lane{2-0};
6489 def VDUPLN16q : VDUPLNQ<{?,?,1,0}, "vdup", "16", v8i16, v4i16, VectorIndex16> {
6491 let Inst{19-18} = lane{1-0};
6493 def VDUPLN32q : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4i32, v2i32, VectorIndex32> {
6495 let Inst{19} = lane{0};
6498 let Predicates = [HasNEON] in {
6499 def : Pat<(v4f16 (ARMvduplane (v4f16 DPR:$Vm), imm:$lane)),
6500 (VDUPLN32d DPR:$Vm, imm:$lane)>;
6502 def : Pat<(v2f32 (ARMvduplane (v2f32 DPR:$Vm), imm:$lane)),
6503 (VDUPLN32d DPR:$Vm, imm:$lane)>;
6505 def : Pat<(v4f32 (ARMvduplane (v2f32 DPR:$Vm), imm:$lane)),
6506 (VDUPLN32q DPR:$Vm, imm:$lane)>;
6508 def : Pat<(v16i8 (ARMvduplane (v16i8 QPR:$src), imm:$lane)),
6509 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
6510 (DSubReg_i8_reg imm:$lane))),
6511 (SubReg_i8_lane imm:$lane)))>;
6512 def : Pat<(v8i16 (ARMvduplane (v8i16 QPR:$src), imm:$lane)),
6513 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
6514 (DSubReg_i16_reg imm:$lane))),
6515 (SubReg_i16_lane imm:$lane)))>;
6516 def : Pat<(v8f16 (ARMvduplane (v8f16 QPR:$src), imm:$lane)),
6517 (v8f16 (VDUPLN16q (v4f16 (EXTRACT_SUBREG QPR:$src,
6518 (DSubReg_i16_reg imm:$lane))),
6519 (SubReg_i16_lane imm:$lane)))>;
6520 def : Pat<(v4i32 (ARMvduplane (v4i32 QPR:$src), imm:$lane)),
6521 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
6522 (DSubReg_i32_reg imm:$lane))),
6523 (SubReg_i32_lane imm:$lane)))>;
6524 def : Pat<(v4f32 (ARMvduplane (v4f32 QPR:$src), imm:$lane)),
6525 (v4f32 (VDUPLN32q (v2f32 (EXTRACT_SUBREG QPR:$src,
6526 (DSubReg_i32_reg imm:$lane))),
6527 (SubReg_i32_lane imm:$lane)))>;
6529 def : Pat<(v4f16 (ARMvdup HPR:$src)),
6530 (v4f16 (VDUPLN16d (INSERT_SUBREG (v4f16 (IMPLICIT_DEF)),
6531 HPR:$src, ssub_0), (i32 0)))>;
6532 def : Pat<(v2f32 (ARMvdup (f32 SPR:$src))),
6533 (v2f32 (VDUPLN32d (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
6534 SPR:$src, ssub_0), (i32 0)))>;
6535 def : Pat<(v4f32 (ARMvdup (f32 SPR:$src))),
6536 (v4f32 (VDUPLN32q (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
6537 SPR:$src, ssub_0), (i32 0)))>;
6538 def : Pat<(v8f16 (ARMvdup HPR:$src)),
6539 (v8f16 (VDUPLN16q (INSERT_SUBREG (v4f16 (IMPLICIT_DEF)),
6540 HPR:$src, ssub_0), (i32 0)))>;
6543 // VMOVN : Vector Narrowing Move
6544 defm VMOVN : N2VN_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVN,
6545 "vmovn", "i", trunc>;
6546 // VQMOVN : Vector Saturating Narrowing Move
6547 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
6548 "vqmovn", "s", int_arm_neon_vqmovns>;
6549 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
6550 "vqmovn", "u", int_arm_neon_vqmovnu>;
6551 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
6552 "vqmovun", "s", int_arm_neon_vqmovnsu>;
6553 // VMOVL : Vector Lengthening Move
6554 defm VMOVLs : N2VL_QHS<0b01,0b10100,0,1, "vmovl", "s", sext>;
6555 defm VMOVLu : N2VL_QHS<0b11,0b10100,0,1, "vmovl", "u", zext>;
6557 let Predicates = [HasNEON] in {
6558 def : Pat<(v8i16 (anyext (v8i8 DPR:$Vm))), (VMOVLuv8i16 DPR:$Vm)>;
6559 def : Pat<(v4i32 (anyext (v4i16 DPR:$Vm))), (VMOVLuv4i32 DPR:$Vm)>;
6560 def : Pat<(v2i64 (anyext (v2i32 DPR:$Vm))), (VMOVLuv2i64 DPR:$Vm)>;
6563 // Vector Conversions.
6565 // VCVT : Vector Convert Between Floating-Point and Integers
6566 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
6567 v2i32, v2f32, fp_to_sint>;
6568 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
6569 v2i32, v2f32, fp_to_uint>;
6570 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
6571 v2f32, v2i32, sint_to_fp>;
6572 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
6573 v2f32, v2i32, uint_to_fp>;
6575 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
6576 v4i32, v4f32, fp_to_sint>;
6577 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
6578 v4i32, v4f32, fp_to_uint>;
6579 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
6580 v4f32, v4i32, sint_to_fp>;
6581 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
6582 v4f32, v4i32, uint_to_fp>;
6584 def VCVTh2sd : N2VD<0b11, 0b11, 0b01, 0b11, 0b01110, 0, "vcvt", "s16.f16",
6585 v4i16, v4f16, fp_to_sint>,
6586 Requires<[HasNEON, HasFullFP16]>;
6587 def VCVTh2ud : N2VD<0b11, 0b11, 0b01, 0b11, 0b01111, 0, "vcvt", "u16.f16",
6588 v4i16, v4f16, fp_to_uint>,
6589 Requires<[HasNEON, HasFullFP16]>;
6590 def VCVTs2hd : N2VD<0b11, 0b11, 0b01, 0b11, 0b01100, 0, "vcvt", "f16.s16",
6591 v4f16, v4i16, sint_to_fp>,
6592 Requires<[HasNEON, HasFullFP16]>;
6593 def VCVTu2hd : N2VD<0b11, 0b11, 0b01, 0b11, 0b01101, 0, "vcvt", "f16.u16",
6594 v4f16, v4i16, uint_to_fp>,
6595 Requires<[HasNEON, HasFullFP16]>;
6597 def VCVTh2sq : N2VQ<0b11, 0b11, 0b01, 0b11, 0b01110, 0, "vcvt", "s16.f16",
6598 v8i16, v8f16, fp_to_sint>,
6599 Requires<[HasNEON, HasFullFP16]>;
6600 def VCVTh2uq : N2VQ<0b11, 0b11, 0b01, 0b11, 0b01111, 0, "vcvt", "u16.f16",
6601 v8i16, v8f16, fp_to_uint>,
6602 Requires<[HasNEON, HasFullFP16]>;
6603 def VCVTs2hq : N2VQ<0b11, 0b11, 0b01, 0b11, 0b01100, 0, "vcvt", "f16.s16",
6604 v8f16, v8i16, sint_to_fp>,
6605 Requires<[HasNEON, HasFullFP16]>;
6606 def VCVTu2hq : N2VQ<0b11, 0b11, 0b01, 0b11, 0b01101, 0, "vcvt", "f16.u16",
6607 v8f16, v8i16, uint_to_fp>,
6608 Requires<[HasNEON, HasFullFP16]>;
6611 multiclass VCVT_FPI<string op, bits<3> op10_8, SDPatternOperator IntS,
6612 SDPatternOperator IntU> {
6613 let PostEncoderMethod = "NEONThumb2V8PostEncoder", DecoderNamespace = "v8NEON" in {
6614 def SDf : N2VDIntnp<0b10, 0b11, op10_8, 0, NoItinerary, !strconcat("vcvt", op),
6615 "s32.f32", v2i32, v2f32, IntS>, Requires<[HasV8, HasNEON]>;
6616 def SQf : N2VQIntnp<0b10, 0b11, op10_8, 0, NoItinerary, !strconcat("vcvt", op),
6617 "s32.f32", v4i32, v4f32, IntS>, Requires<[HasV8, HasNEON]>;
6618 def UDf : N2VDIntnp<0b10, 0b11, op10_8, 1, NoItinerary, !strconcat("vcvt", op),
6619 "u32.f32", v2i32, v2f32, IntU>, Requires<[HasV8, HasNEON]>;
6620 def UQf : N2VQIntnp<0b10, 0b11, op10_8, 1, NoItinerary, !strconcat("vcvt", op),
6621 "u32.f32", v4i32, v4f32, IntU>, Requires<[HasV8, HasNEON]>;
6622 def SDh : N2VDIntnp<0b01, 0b11, op10_8, 0, NoItinerary, !strconcat("vcvt", op),
6623 "s16.f16", v4i16, v4f16, IntS>,
6624 Requires<[HasV8, HasNEON, HasFullFP16]>;
6625 def SQh : N2VQIntnp<0b01, 0b11, op10_8, 0, NoItinerary, !strconcat("vcvt", op),
6626 "s16.f16", v8i16, v8f16, IntS>,
6627 Requires<[HasV8, HasNEON, HasFullFP16]>;
6628 def UDh : N2VDIntnp<0b01, 0b11, op10_8, 1, NoItinerary, !strconcat("vcvt", op),
6629 "u16.f16", v4i16, v4f16, IntU>,
6630 Requires<[HasV8, HasNEON, HasFullFP16]>;
6631 def UQh : N2VQIntnp<0b01, 0b11, op10_8, 1, NoItinerary, !strconcat("vcvt", op),
6632 "u16.f16", v8i16, v8f16, IntU>,
6633 Requires<[HasV8, HasNEON, HasFullFP16]>;
6637 defm VCVTAN : VCVT_FPI<"a", 0b000, int_arm_neon_vcvtas, int_arm_neon_vcvtau>;
6638 defm VCVTNN : VCVT_FPI<"n", 0b001, int_arm_neon_vcvtns, int_arm_neon_vcvtnu>;
6639 defm VCVTPN : VCVT_FPI<"p", 0b010, int_arm_neon_vcvtps, int_arm_neon_vcvtpu>;
6640 defm VCVTMN : VCVT_FPI<"m", 0b011, int_arm_neon_vcvtms, int_arm_neon_vcvtmu>;
6642 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
6643 let DecoderMethod = "DecodeVCVTD" in {
6644 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
6645 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
6646 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
6647 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
6648 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
6649 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
6650 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
6651 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
6652 let Predicates = [HasNEON, HasFullFP16] in {
6653 def VCVTh2xsd : N2VCvtD<0, 1, 0b1101, 0, 1, "vcvt", "s16.f16",
6654 v4i16, v4f16, int_arm_neon_vcvtfp2fxs>;
6655 def VCVTh2xud : N2VCvtD<1, 1, 0b1101, 0, 1, "vcvt", "u16.f16",
6656 v4i16, v4f16, int_arm_neon_vcvtfp2fxu>;
6657 def VCVTxs2hd : N2VCvtD<0, 1, 0b1100, 0, 1, "vcvt", "f16.s16",
6658 v4f16, v4i16, int_arm_neon_vcvtfxs2fp>;
6659 def VCVTxu2hd : N2VCvtD<1, 1, 0b1100, 0, 1, "vcvt", "f16.u16",
6660 v4f16, v4i16, int_arm_neon_vcvtfxu2fp>;
6661 } // Predicates = [HasNEON, HasFullFP16]
6664 let DecoderMethod = "DecodeVCVTQ" in {
6665 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
6666 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
6667 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
6668 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
6669 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
6670 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
6671 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
6672 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
6673 let Predicates = [HasNEON, HasFullFP16] in {
6674 def VCVTh2xsq : N2VCvtQ<0, 1, 0b1101, 0, 1, "vcvt", "s16.f16",
6675 v8i16, v8f16, int_arm_neon_vcvtfp2fxs>;
6676 def VCVTh2xuq : N2VCvtQ<1, 1, 0b1101, 0, 1, "vcvt", "u16.f16",
6677 v8i16, v8f16, int_arm_neon_vcvtfp2fxu>;
6678 def VCVTxs2hq : N2VCvtQ<0, 1, 0b1100, 0, 1, "vcvt", "f16.s16",
6679 v8f16, v8i16, int_arm_neon_vcvtfxs2fp>;
6680 def VCVTxu2hq : N2VCvtQ<1, 1, 0b1100, 0, 1, "vcvt", "f16.u16",
6681 v8f16, v8i16, int_arm_neon_vcvtfxu2fp>;
6682 } // Predicates = [HasNEON, HasFullFP16]
6685 def : NEONInstAlias<"vcvt${p}.s32.f32 $Dd, $Dm, #0",
6686 (VCVTf2sd DPR:$Dd, DPR:$Dm, pred:$p)>;
6687 def : NEONInstAlias<"vcvt${p}.u32.f32 $Dd, $Dm, #0",
6688 (VCVTf2ud DPR:$Dd, DPR:$Dm, pred:$p)>;
6689 def : NEONInstAlias<"vcvt${p}.f32.s32 $Dd, $Dm, #0",
6690 (VCVTs2fd DPR:$Dd, DPR:$Dm, pred:$p)>;
6691 def : NEONInstAlias<"vcvt${p}.f32.u32 $Dd, $Dm, #0",
6692 (VCVTu2fd DPR:$Dd, DPR:$Dm, pred:$p)>;
6694 def : NEONInstAlias<"vcvt${p}.s32.f32 $Qd, $Qm, #0",
6695 (VCVTf2sq QPR:$Qd, QPR:$Qm, pred:$p)>;
6696 def : NEONInstAlias<"vcvt${p}.u32.f32 $Qd, $Qm, #0",
6697 (VCVTf2uq QPR:$Qd, QPR:$Qm, pred:$p)>;
6698 def : NEONInstAlias<"vcvt${p}.f32.s32 $Qd, $Qm, #0",
6699 (VCVTs2fq QPR:$Qd, QPR:$Qm, pred:$p)>;
6700 def : NEONInstAlias<"vcvt${p}.f32.u32 $Qd, $Qm, #0",
6701 (VCVTu2fq QPR:$Qd, QPR:$Qm, pred:$p)>;
6703 def : NEONInstAlias<"vcvt${p}.s16.f16 $Dd, $Dm, #0",
6704 (VCVTh2sd DPR:$Dd, DPR:$Dm, pred:$p)>;
6705 def : NEONInstAlias<"vcvt${p}.u16.f16 $Dd, $Dm, #0",
6706 (VCVTh2ud DPR:$Dd, DPR:$Dm, pred:$p)>;
6707 def : NEONInstAlias<"vcvt${p}.f16.s16 $Dd, $Dm, #0",
6708 (VCVTs2hd DPR:$Dd, DPR:$Dm, pred:$p)>;
6709 def : NEONInstAlias<"vcvt${p}.f16.u16 $Dd, $Dm, #0",
6710 (VCVTu2hd DPR:$Dd, DPR:$Dm, pred:$p)>;
6712 def : NEONInstAlias<"vcvt${p}.s16.f16 $Qd, $Qm, #0",
6713 (VCVTh2sq QPR:$Qd, QPR:$Qm, pred:$p)>;
6714 def : NEONInstAlias<"vcvt${p}.u16.f16 $Qd, $Qm, #0",
6715 (VCVTh2uq QPR:$Qd, QPR:$Qm, pred:$p)>;
6716 def : NEONInstAlias<"vcvt${p}.f16.s16 $Qd, $Qm, #0",
6717 (VCVTs2hq QPR:$Qd, QPR:$Qm, pred:$p)>;
6718 def : NEONInstAlias<"vcvt${p}.f16.u16 $Qd, $Qm, #0",
6719 (VCVTu2hq QPR:$Qd, QPR:$Qm, pred:$p)>;
6722 // VCVT : Vector Convert Between Half-Precision and Single-Precision.
6723 def VCVTf2h : N2VNInt<0b11, 0b11, 0b01, 0b10, 0b01100, 0, 0,
6724 IIC_VUNAQ, "vcvt", "f16.f32",
6725 v4i16, v4f32, int_arm_neon_vcvtfp2hf>,
6726 Requires<[HasNEON, HasFP16]>;
6727 def VCVTh2f : N2VLInt<0b11, 0b11, 0b01, 0b10, 0b01110, 0, 0,
6728 IIC_VUNAQ, "vcvt", "f32.f16",
6729 v4f32, v4i16, int_arm_neon_vcvthf2fp>,
6730 Requires<[HasNEON, HasFP16]>;
6734 // VREV64 : Vector Reverse elements within 64-bit doublewords
6736 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
6737 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$Vd),
6738 (ins DPR:$Vm), IIC_VMOVD,
6739 OpcodeStr, Dt, "$Vd, $Vm", "",
6740 [(set DPR:$Vd, (Ty (ARMvrev64 (Ty DPR:$Vm))))]>;
6741 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
6742 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$Vd),
6743 (ins QPR:$Vm), IIC_VMOVQ,
6744 OpcodeStr, Dt, "$Vd, $Vm", "",
6745 [(set QPR:$Vd, (Ty (ARMvrev64 (Ty QPR:$Vm))))]>;
6747 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
6748 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
6749 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
6750 let Predicates = [HasNEON] in {
6751 def : Pat<(v2f32 (ARMvrev64 (v2f32 DPR:$Vm))), (VREV64d32 DPR:$Vm)>;
6754 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
6755 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
6756 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
6758 let Predicates = [HasNEON] in {
6759 def : Pat<(v4f32 (ARMvrev64 (v4f32 QPR:$Vm))), (VREV64q32 QPR:$Vm)>;
6760 def : Pat<(v8f16 (ARMvrev64 (v8f16 QPR:$Vm))), (VREV64q16 QPR:$Vm)>;
6761 def : Pat<(v4f16 (ARMvrev64 (v4f16 DPR:$Vm))), (VREV64d16 DPR:$Vm)>;
6764 // VREV32 : Vector Reverse elements within 32-bit words
6766 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
6767 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$Vd),
6768 (ins DPR:$Vm), IIC_VMOVD,
6769 OpcodeStr, Dt, "$Vd, $Vm", "",
6770 [(set DPR:$Vd, (Ty (ARMvrev32 (Ty DPR:$Vm))))]>;
6771 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
6772 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$Vd),
6773 (ins QPR:$Vm), IIC_VMOVQ,
6774 OpcodeStr, Dt, "$Vd, $Vm", "",
6775 [(set QPR:$Vd, (Ty (ARMvrev32 (Ty QPR:$Vm))))]>;
6777 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
6778 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
6780 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
6781 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
6783 // VREV16 : Vector Reverse elements within 16-bit halfwords
6785 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
6786 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$Vd),
6787 (ins DPR:$Vm), IIC_VMOVD,
6788 OpcodeStr, Dt, "$Vd, $Vm", "",
6789 [(set DPR:$Vd, (Ty (ARMvrev16 (Ty DPR:$Vm))))]>;
6790 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
6791 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$Vd),
6792 (ins QPR:$Vm), IIC_VMOVQ,
6793 OpcodeStr, Dt, "$Vd, $Vm", "",
6794 [(set QPR:$Vd, (Ty (ARMvrev16 (Ty QPR:$Vm))))]>;
6796 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
6797 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
6799 // Other Vector Shuffles.
6801 // Aligned extractions: really just dropping registers
6803 class AlignedVEXTq<ValueType DestTy, ValueType SrcTy, SDNodeXForm LaneCVT>
6804 : Pat<(DestTy (vector_extract_subvec (SrcTy QPR:$src), (i32 imm:$start))),
6805 (EXTRACT_SUBREG (SrcTy QPR:$src), (LaneCVT imm:$start))>,
6806 Requires<[HasNEON]>;
6808 def : AlignedVEXTq<v8i8, v16i8, DSubReg_i8_reg>;
6810 def : AlignedVEXTq<v4i16, v8i16, DSubReg_i16_reg>;
6812 def : AlignedVEXTq<v2i32, v4i32, DSubReg_i32_reg>;
6814 def : AlignedVEXTq<v1i64, v2i64, DSubReg_f64_reg>;
6816 def : AlignedVEXTq<v2f32, v4f32, DSubReg_i32_reg>;
6818 def : AlignedVEXTq<v4f16, v8f16, DSubReg_i16_reg>; // v8f16 -> v4f16
6820 // VEXT : Vector Extract
6823 // All of these have a two-operand InstAlias.
6824 let TwoOperandAliasConstraint = "$Vn = $Vd" in {
6825 class VEXTd<string OpcodeStr, string Dt, ValueType Ty, Operand immTy>
6826 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$Vd),
6827 (ins DPR:$Vn, DPR:$Vm, immTy:$index), NVExtFrm,
6828 IIC_VEXTD, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
6829 [(set DPR:$Vd, (Ty (NEONvext (Ty DPR:$Vn),
6830 (Ty DPR:$Vm), imm:$index)))]> {
6833 let Inst{10-8} = index{2-0};
6836 class VEXTq<string OpcodeStr, string Dt, ValueType Ty, Operand immTy>
6837 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$Vd),
6838 (ins QPR:$Vn, QPR:$Vm, imm0_15:$index), NVExtFrm,
6839 IIC_VEXTQ, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
6840 [(set QPR:$Vd, (Ty (NEONvext (Ty QPR:$Vn),
6841 (Ty QPR:$Vm), imm:$index)))]> {
6843 let Inst{11-8} = index{3-0};
6847 def VEXTd8 : VEXTd<"vext", "8", v8i8, imm0_7> {
6848 let Inst{10-8} = index{2-0};
6850 def VEXTd16 : VEXTd<"vext", "16", v4i16, imm0_3> {
6851 let Inst{10-9} = index{1-0};
6854 let Predicates = [HasNEON] in {
6855 def : Pat<(v4f16 (NEONvext (v4f16 DPR:$Vn), (v4f16 DPR:$Vm), (i32 imm:$index))),
6856 (VEXTd16 DPR:$Vn, DPR:$Vm, imm:$index)>;
6859 def VEXTd32 : VEXTd<"vext", "32", v2i32, imm0_1> {
6860 let Inst{10} = index{0};
6861 let Inst{9-8} = 0b00;
6863 let Predicates = [HasNEON] in {
6864 def : Pat<(v2f32 (NEONvext (v2f32 DPR:$Vn), (v2f32 DPR:$Vm), (i32 imm:$index))),
6865 (VEXTd32 DPR:$Vn, DPR:$Vm, imm:$index)>;
6868 def VEXTq8 : VEXTq<"vext", "8", v16i8, imm0_15> {
6869 let Inst{11-8} = index{3-0};
6871 def VEXTq16 : VEXTq<"vext", "16", v8i16, imm0_7> {
6872 let Inst{11-9} = index{2-0};
6875 let Predicates = [HasNEON] in {
6876 def : Pat<(v8f16 (NEONvext (v8f16 QPR:$Vn), (v8f16 QPR:$Vm), (i32 imm:$index))),
6877 (VEXTq16 QPR:$Vn, QPR:$Vm, imm:$index)>;
6880 def VEXTq32 : VEXTq<"vext", "32", v4i32, imm0_3> {
6881 let Inst{11-10} = index{1-0};
6882 let Inst{9-8} = 0b00;
6884 def VEXTq64 : VEXTq<"vext", "64", v2i64, imm0_1> {
6885 let Inst{11} = index{0};
6886 let Inst{10-8} = 0b000;
6888 let Predicates = [HasNEON] in {
6889 def : Pat<(v4f32 (NEONvext (v4f32 QPR:$Vn), (v4f32 QPR:$Vm), (i32 imm:$index))),
6890 (VEXTq32 QPR:$Vn, QPR:$Vm, imm:$index)>;
6893 // VTRN : Vector Transpose
6895 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
6896 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
6897 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
6899 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
6900 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
6901 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
6903 // VUZP : Vector Unzip (Deinterleave)
6905 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
6906 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
6907 // vuzp.32 Dd, Dm is a pseudo-instruction expanded to vtrn.32 Dd, Dm.
6908 def : NEONInstAlias<"vuzp${p}.32 $Dd, $Dm",
6909 (VTRNd32 DPR:$Dd, DPR:$Dm, pred:$p)>;
6911 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
6912 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
6913 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
6915 // VZIP : Vector Zip (Interleave)
6917 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
6918 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
6919 // vzip.32 Dd, Dm is a pseudo-instruction expanded to vtrn.32 Dd, Dm.
6920 def : NEONInstAlias<"vzip${p}.32 $Dd, $Dm",
6921 (VTRNd32 DPR:$Dd, DPR:$Dm, pred:$p)>;
6923 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
6924 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
6925 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
6927 // Vector Table Lookup and Table Extension.
6929 // VTBL : Vector Table Lookup
6930 let DecoderMethod = "DecodeTBLInstruction" in {
6932 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$Vd),
6933 (ins VecListOneD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB1,
6934 "vtbl", "8", "$Vd, $Vn, $Vm", "",
6935 [(set DPR:$Vd, (v8i8 (NEONvtbl1 VecListOneD:$Vn, DPR:$Vm)))]>;
6937 let hasExtraSrcRegAllocReq = 1 in {
6939 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$Vd),
6940 (ins VecListDPair:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB2,
6941 "vtbl", "8", "$Vd, $Vn, $Vm", "", []>;
6943 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$Vd),
6944 (ins VecListThreeD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB3,
6945 "vtbl", "8", "$Vd, $Vn, $Vm", "", []>;
6947 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$Vd),
6948 (ins VecListFourD:$Vn, DPR:$Vm),
6950 "vtbl", "8", "$Vd, $Vn, $Vm", "", []>;
6951 } // hasExtraSrcRegAllocReq = 1
6954 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB3, "", []>;
6956 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB4, "", []>;
6958 // VTBX : Vector Table Extension
6960 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$Vd),
6961 (ins DPR:$orig, VecListOneD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX1,
6962 "vtbx", "8", "$Vd, $Vn, $Vm", "$orig = $Vd",
6963 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbx1
6964 DPR:$orig, VecListOneD:$Vn, DPR:$Vm)))]>;
6965 let hasExtraSrcRegAllocReq = 1 in {
6967 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$Vd),
6968 (ins DPR:$orig, VecListDPair:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX2,
6969 "vtbx", "8", "$Vd, $Vn, $Vm", "$orig = $Vd", []>;
6971 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$Vd),
6972 (ins DPR:$orig, VecListThreeD:$Vn, DPR:$Vm),
6973 NVTBLFrm, IIC_VTBX3,
6974 "vtbx", "8", "$Vd, $Vn, $Vm",
6977 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$Vd),
6978 (ins DPR:$orig, VecListFourD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX4,
6979 "vtbx", "8", "$Vd, $Vn, $Vm",
6981 } // hasExtraSrcRegAllocReq = 1
6984 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
6985 IIC_VTBX3, "$orig = $dst", []>;
6987 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
6988 IIC_VTBX4, "$orig = $dst", []>;
6989 } // DecoderMethod = "DecodeTBLInstruction"
6991 let Predicates = [HasNEON] in {
6992 def : Pat<(v8i8 (NEONvtbl2 v8i8:$Vn0, v8i8:$Vn1, v8i8:$Vm)),
6993 (v8i8 (VTBL2 (REG_SEQUENCE DPair, v8i8:$Vn0, dsub_0,
6996 def : Pat<(v8i8 (int_arm_neon_vtbx2 v8i8:$orig, v8i8:$Vn0, v8i8:$Vn1,
6998 (v8i8 (VTBX2 v8i8:$orig,
6999 (REG_SEQUENCE DPair, v8i8:$Vn0, dsub_0,
7003 def : Pat<(v8i8 (int_arm_neon_vtbl3 v8i8:$Vn0, v8i8:$Vn1,
7004 v8i8:$Vn2, v8i8:$Vm)),
7005 (v8i8 (VTBL3Pseudo (REG_SEQUENCE QQPR, v8i8:$Vn0, dsub_0,
7008 (v8i8 (IMPLICIT_DEF)), dsub_3),
7010 def : Pat<(v8i8 (int_arm_neon_vtbx3 v8i8:$orig, v8i8:$Vn0, v8i8:$Vn1,
7011 v8i8:$Vn2, v8i8:$Vm)),
7012 (v8i8 (VTBX3Pseudo v8i8:$orig,
7013 (REG_SEQUENCE QQPR, v8i8:$Vn0, dsub_0,
7016 (v8i8 (IMPLICIT_DEF)), dsub_3),
7019 def : Pat<(v8i8 (int_arm_neon_vtbl4 v8i8:$Vn0, v8i8:$Vn1,
7020 v8i8:$Vn2, v8i8:$Vn3, v8i8:$Vm)),
7021 (v8i8 (VTBL4Pseudo (REG_SEQUENCE QQPR, v8i8:$Vn0, dsub_0,
7026 def : Pat<(v8i8 (int_arm_neon_vtbx4 v8i8:$orig, v8i8:$Vn0, v8i8:$Vn1,
7027 v8i8:$Vn2, v8i8:$Vn3, v8i8:$Vm)),
7028 (v8i8 (VTBX4Pseudo v8i8:$orig,
7029 (REG_SEQUENCE QQPR, v8i8:$Vn0, dsub_0,
7036 // VRINT : Vector Rounding
7037 multiclass VRINT_FPI<string op, bits<3> op9_7, SDPatternOperator Int> {
7038 let PostEncoderMethod = "NEONThumb2V8PostEncoder", DecoderNamespace = "v8NEON" in {
7039 def Df : N2VDIntnp<0b10, 0b10, 0b100, 0, NoItinerary,
7040 !strconcat("vrint", op), "f32",
7041 v2f32, v2f32, Int>, Requires<[HasV8, HasNEON]> {
7042 let Inst{9-7} = op9_7;
7044 def Qf : N2VQIntnp<0b10, 0b10, 0b100, 0, NoItinerary,
7045 !strconcat("vrint", op), "f32",
7046 v4f32, v4f32, Int>, Requires<[HasV8, HasNEON]> {
7047 let Inst{9-7} = op9_7;
7049 def Dh : N2VDIntnp<0b01, 0b10, 0b100, 0, NoItinerary,
7050 !strconcat("vrint", op), "f16",
7052 Requires<[HasV8, HasNEON, HasFullFP16]> {
7053 let Inst{9-7} = op9_7;
7055 def Qh : N2VQIntnp<0b01, 0b10, 0b100, 0, NoItinerary,
7056 !strconcat("vrint", op), "f16",
7058 Requires<[HasV8, HasNEON, HasFullFP16]> {
7059 let Inst{9-7} = op9_7;
7063 def : NEONInstAlias<!strconcat("vrint", op, ".f32.f32\t$Dd, $Dm"),
7064 (!cast<Instruction>(NAME#"Df") DPR:$Dd, DPR:$Dm)>;
7065 def : NEONInstAlias<!strconcat("vrint", op, ".f32.f32\t$Qd, $Qm"),
7066 (!cast<Instruction>(NAME#"Qf") QPR:$Qd, QPR:$Qm)>;
7067 let Predicates = [HasNEON, HasFullFP16] in {
7068 def : NEONInstAlias<!strconcat("vrint", op, ".f16.f16\t$Dd, $Dm"),
7069 (!cast<Instruction>(NAME#"Dh") DPR:$Dd, DPR:$Dm)>;
7070 def : NEONInstAlias<!strconcat("vrint", op, ".f16.f16\t$Qd, $Qm"),
7071 (!cast<Instruction>(NAME#"Qh") QPR:$Qd, QPR:$Qm)>;
7075 defm VRINTNN : VRINT_FPI<"n", 0b000, int_arm_neon_vrintn>;
7076 defm VRINTXN : VRINT_FPI<"x", 0b001, int_arm_neon_vrintx>;
7077 defm VRINTAN : VRINT_FPI<"a", 0b010, int_arm_neon_vrinta>;
7078 defm VRINTZN : VRINT_FPI<"z", 0b011, int_arm_neon_vrintz>;
7079 defm VRINTMN : VRINT_FPI<"m", 0b101, int_arm_neon_vrintm>;
7080 defm VRINTPN : VRINT_FPI<"p", 0b111, int_arm_neon_vrintp>;
7082 // Cryptography instructions
7083 let PostEncoderMethod = "NEONThumb2DataIPostEncoder",
7084 DecoderNamespace = "v8Crypto", hasSideEffects = 0 in {
7085 class AES<string op, bit op7, bit op6, SDPatternOperator Int>
7086 : N2VQIntXnp<0b00, 0b00, 0b011, op6, op7, NoItinerary,
7087 !strconcat("aes", op), "8", v16i8, v16i8, Int>,
7088 Requires<[HasV8, HasCrypto]>;
7089 class AES2Op<string op, bit op7, bit op6, SDPatternOperator Int>
7090 : N2VQIntX2np<0b00, 0b00, 0b011, op6, op7, NoItinerary,
7091 !strconcat("aes", op), "8", v16i8, v16i8, Int>,
7092 Requires<[HasV8, HasCrypto]>;
7093 class N2SHA<string op, bits<2> op17_16, bits<3> op10_8, bit op7, bit op6,
7094 SDPatternOperator Int>
7095 : N2VQIntXnp<0b10, op17_16, op10_8, op6, op7, NoItinerary,
7096 !strconcat("sha", op), "32", v4i32, v4i32, Int>,
7097 Requires<[HasV8, HasCrypto]>;
7098 class N2SHA2Op<string op, bits<2> op17_16, bits<3> op10_8, bit op7, bit op6,
7099 SDPatternOperator Int>
7100 : N2VQIntX2np<0b10, op17_16, op10_8, op6, op7, NoItinerary,
7101 !strconcat("sha", op), "32", v4i32, v4i32, Int>,
7102 Requires<[HasV8, HasCrypto]>;
7103 class N3SHA3Op<string op, bits<5> op27_23, bits<2> op21_20, SDPatternOperator Int>
7104 : N3VQInt3np<op27_23, op21_20, 0b1100, 1, 0, N3RegFrm, NoItinerary,
7105 !strconcat("sha", op), "32", v4i32, v4i32, Int, 0>,
7106 Requires<[HasV8, HasCrypto]>;
7109 def AESD : AES2Op<"d", 0, 1, int_arm_neon_aesd>;
7110 def AESE : AES2Op<"e", 0, 0, int_arm_neon_aese>;
7111 def AESIMC : AES<"imc", 1, 1, int_arm_neon_aesimc>;
7112 def AESMC : AES<"mc", 1, 0, int_arm_neon_aesmc>;
7114 def SHA1H : N2SHA<"1h", 0b01, 0b010, 1, 1, null_frag>;
7115 def SHA1SU1 : N2SHA2Op<"1su1", 0b10, 0b011, 1, 0, int_arm_neon_sha1su1>;
7116 def SHA256SU0 : N2SHA2Op<"256su0", 0b10, 0b011, 1, 1, int_arm_neon_sha256su0>;
7117 def SHA1C : N3SHA3Op<"1c", 0b00100, 0b00, null_frag>;
7118 def SHA1M : N3SHA3Op<"1m", 0b00100, 0b10, null_frag>;
7119 def SHA1P : N3SHA3Op<"1p", 0b00100, 0b01, null_frag>;
7120 def SHA1SU0 : N3SHA3Op<"1su0", 0b00100, 0b11, int_arm_neon_sha1su0>;
7121 def SHA256H : N3SHA3Op<"256h", 0b00110, 0b00, int_arm_neon_sha256h>;
7122 def SHA256H2 : N3SHA3Op<"256h2", 0b00110, 0b01, int_arm_neon_sha256h2>;
7123 def SHA256SU1 : N3SHA3Op<"256su1", 0b00110, 0b10, int_arm_neon_sha256su1>;
7125 let Predicates = [HasNEON] in {
7126 def : Pat<(i32 (int_arm_neon_sha1h i32:$Rn)),
7127 (COPY_TO_REGCLASS (f32 (EXTRACT_SUBREG
7128 (SHA1H (SUBREG_TO_REG (i64 0),
7129 (f32 (COPY_TO_REGCLASS i32:$Rn, SPR)),
7133 def : Pat<(v4i32 (int_arm_neon_sha1c v4i32:$hash_abcd, i32:$hash_e, v4i32:$wk)),
7134 (SHA1C v4i32:$hash_abcd,
7135 (SUBREG_TO_REG (i64 0),
7136 (f32 (COPY_TO_REGCLASS i32:$hash_e, SPR)),
7140 def : Pat<(v4i32 (int_arm_neon_sha1m v4i32:$hash_abcd, i32:$hash_e, v4i32:$wk)),
7141 (SHA1M v4i32:$hash_abcd,
7142 (SUBREG_TO_REG (i64 0),
7143 (f32 (COPY_TO_REGCLASS i32:$hash_e, SPR)),
7147 def : Pat<(v4i32 (int_arm_neon_sha1p v4i32:$hash_abcd, i32:$hash_e, v4i32:$wk)),
7148 (SHA1P v4i32:$hash_abcd,
7149 (SUBREG_TO_REG (i64 0),
7150 (f32 (COPY_TO_REGCLASS i32:$hash_e, SPR)),
7155 //===----------------------------------------------------------------------===//
7156 // NEON instructions for single-precision FP math
7157 //===----------------------------------------------------------------------===//
7159 class N2VSPat<SDNode OpNode, NeonI Inst>
7160 : NEONFPPat<(f32 (OpNode SPR:$a)),
7162 (v2f32 (COPY_TO_REGCLASS (Inst
7164 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
7165 SPR:$a, ssub_0)), DPR_VFP2)), ssub_0)>;
7167 class N3VSPat<SDNode OpNode, NeonI Inst>
7168 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
7170 (v2f32 (COPY_TO_REGCLASS (Inst
7172 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
7175 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
7176 SPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>;
7178 class N3VSPatFP16<SDNode OpNode, NeonI Inst>
7179 : NEONFPPat<(f16 (OpNode HPR:$a, HPR:$b)),
7181 (v4f16 (COPY_TO_REGCLASS (Inst
7183 (v4f16 (COPY_TO_REGCLASS (v4f16 (IMPLICIT_DEF)), DPR_VFP2)),
7186 (v4f16 (COPY_TO_REGCLASS (v4f16 (IMPLICIT_DEF)), DPR_VFP2)),
7187 HPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>;
7189 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
7190 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
7192 (v2f32 (COPY_TO_REGCLASS (Inst
7194 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
7197 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
7200 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
7201 SPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>;
7203 class NVCVTIFPat<SDNode OpNode, NeonI Inst>
7204 : NEONFPPat<(f32 (OpNode GPR:$a)),
7205 (f32 (EXTRACT_SUBREG
7208 (v2f32 (IMPLICIT_DEF)),
7209 (i32 (COPY_TO_REGCLASS GPR:$a, SPR)), ssub_0))),
7211 class NVCVTFIPat<SDNode OpNode, NeonI Inst>
7212 : NEONFPPat<(i32 (OpNode SPR:$a)),
7213 (i32 (EXTRACT_SUBREG
7214 (v2f32 (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
7218 def : N3VSPat<fadd, VADDfd>;
7219 def : N3VSPat<fsub, VSUBfd>;
7220 def : N3VSPat<fmul, VMULfd>;
7221 def : N3VSMulOpPat<fmul, fadd, VMLAfd>,
7222 Requires<[HasNEON, UseNEONForFP, UseFPVMLx]>;
7223 def : N3VSMulOpPat<fmul, fsub, VMLSfd>,
7224 Requires<[HasNEON, UseNEONForFP, UseFPVMLx]>;
7225 def : N3VSMulOpPat<fmul, fadd, VFMAfd>,
7226 Requires<[HasVFP4, UseNEONForFP, UseFusedMAC]>;
7227 def : N3VSMulOpPat<fmul, fsub, VFMSfd>,
7228 Requires<[HasVFP4, UseNEONForFP, UseFusedMAC]>;
7229 def : N2VSPat<fabs, VABSfd>;
7230 def : N2VSPat<fneg, VNEGfd>;
7231 def : N3VSPatFP16<fmaximum, VMAXhd>, Requires<[HasFullFP16]>;
7232 def : N3VSPatFP16<fminimum, VMINhd>, Requires<[HasFullFP16]>;
7233 def : N3VSPat<fmaximum, VMAXfd>, Requires<[HasNEON]>;
7234 def : N3VSPat<fminimum, VMINfd>, Requires<[HasNEON]>;
7235 def : NVCVTFIPat<fp_to_sint, VCVTf2sd>;
7236 def : NVCVTFIPat<fp_to_uint, VCVTf2ud>;
7237 def : NVCVTIFPat<sint_to_fp, VCVTs2fd>;
7238 def : NVCVTIFPat<uint_to_fp, VCVTu2fd>;
7240 // NEON doesn't have any f64 conversions, so provide patterns to make
7241 // sure the VFP conversions match when extracting from a vector.
7242 def : VFPPat<(f64 (sint_to_fp (extractelt (v2i32 DPR:$src), imm:$lane))),
7243 (VSITOD (EXTRACT_SUBREG DPR:$src, (SSubReg_f32_reg imm:$lane)))>;
7244 def : VFPPat<(f64 (sint_to_fp (extractelt (v4i32 QPR:$src), imm:$lane))),
7245 (VSITOD (EXTRACT_SUBREG QPR:$src, (SSubReg_f32_reg imm:$lane)))>;
7246 def : VFPPat<(f64 (uint_to_fp (extractelt (v2i32 DPR:$src), imm:$lane))),
7247 (VUITOD (EXTRACT_SUBREG DPR:$src, (SSubReg_f32_reg imm:$lane)))>;
7248 def : VFPPat<(f64 (uint_to_fp (extractelt (v4i32 QPR:$src), imm:$lane))),
7249 (VUITOD (EXTRACT_SUBREG QPR:$src, (SSubReg_f32_reg imm:$lane)))>;
7252 // Prefer VMOVDRR for i32 -> f32 bitcasts, it can write all DPR registers.
7253 def : Pat<(f32 (bitconvert GPR:$a)),
7254 (EXTRACT_SUBREG (VMOVDRR GPR:$a, GPR:$a), ssub_0)>,
7255 Requires<[HasNEON, DontUseVMOVSR]>;
7256 def : Pat<(arm_vmovsr GPR:$a),
7257 (EXTRACT_SUBREG (VMOVDRR GPR:$a, GPR:$a), ssub_0)>,
7258 Requires<[HasNEON, DontUseVMOVSR]>;
7260 //===----------------------------------------------------------------------===//
7261 // Non-Instruction Patterns or Endiness - Revert Patterns
7262 //===----------------------------------------------------------------------===//
7265 // 64 bit conversions
7266 let Predicates = [HasNEON] in {
7267 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
7268 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
7270 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
7271 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
7273 def : Pat<(v4i16 (bitconvert (v4f16 DPR:$src))), (v4i16 DPR:$src)>;
7274 def : Pat<(v4f16 (bitconvert (v4i16 DPR:$src))), (v4f16 DPR:$src)>;
7276 // 128 bit conversions
7277 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
7278 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
7280 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
7281 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
7283 def : Pat<(v8i16 (bitconvert (v8f16 QPR:$src))), (v8i16 QPR:$src)>;
7284 def : Pat<(v8f16 (bitconvert (v8i16 QPR:$src))), (v8f16 QPR:$src)>;
7287 let Predicates = [IsLE,HasNEON] in {
7288 // 64 bit conversions
7289 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
7290 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
7291 def : Pat<(f64 (bitconvert (v4f16 DPR:$src))), (f64 DPR:$src)>;
7292 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
7293 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
7295 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
7296 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
7297 def : Pat<(v1i64 (bitconvert (v4f16 DPR:$src))), (v1i64 DPR:$src)>;
7298 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
7299 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
7301 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
7302 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
7303 def : Pat<(v2f32 (bitconvert (v4f16 DPR:$src))), (v2f32 DPR:$src)>;
7304 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
7305 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
7307 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
7308 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
7309 def : Pat<(v2i32 (bitconvert (v4f16 DPR:$src))), (v2i32 DPR:$src)>;
7310 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
7311 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
7313 def : Pat<(v4f16 (bitconvert (f64 DPR:$src))), (v4f16 DPR:$src)>;
7314 def : Pat<(v4f16 (bitconvert (v1i64 DPR:$src))), (v4f16 DPR:$src)>;
7315 def : Pat<(v4f16 (bitconvert (v2f32 DPR:$src))), (v4f16 DPR:$src)>;
7316 def : Pat<(v4f16 (bitconvert (v2i32 DPR:$src))), (v4f16 DPR:$src)>;
7317 def : Pat<(v4f16 (bitconvert (v8i8 DPR:$src))), (v4f16 DPR:$src)>;
7319 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
7320 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
7321 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
7322 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
7323 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
7325 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
7326 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
7327 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
7328 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
7329 def : Pat<(v8i8 (bitconvert (v4f16 DPR:$src))), (v8i8 DPR:$src)>;
7330 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
7332 // 128 bit conversions
7333 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;
7334 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
7335 def : Pat<(v2f64 (bitconvert (v8f16 QPR:$src))), (v2f64 QPR:$src)>;
7336 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
7337 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
7339 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
7340 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
7341 def : Pat<(v2i64 (bitconvert (v8f16 QPR:$src))), (v2i64 QPR:$src)>;
7342 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
7343 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
7345 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
7346 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
7347 def : Pat<(v4f32 (bitconvert (v8f16 QPR:$src))), (v4f32 QPR:$src)>;
7348 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
7349 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
7351 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
7352 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
7353 def : Pat<(v4i32 (bitconvert (v8f16 QPR:$src))), (v4i32 QPR:$src)>;
7354 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
7355 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
7357 def : Pat<(v8f16 (bitconvert (v2f64 QPR:$src))), (v8f16 QPR:$src)>;
7358 def : Pat<(v8f16 (bitconvert (v2i64 QPR:$src))), (v8f16 QPR:$src)>;
7359 def : Pat<(v8f16 (bitconvert (v4f32 QPR:$src))), (v8f16 QPR:$src)>;
7360 def : Pat<(v8f16 (bitconvert (v4i32 QPR:$src))), (v8f16 QPR:$src)>;
7361 def : Pat<(v8f16 (bitconvert (v16i8 QPR:$src))), (v8f16 QPR:$src)>;
7363 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
7364 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
7365 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
7366 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
7367 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
7369 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
7370 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
7371 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
7372 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
7373 def : Pat<(v16i8 (bitconvert (v8f16 QPR:$src))), (v16i8 QPR:$src)>;
7374 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
7377 let Predicates = [IsBE,HasNEON] in {
7378 // 64 bit conversions
7379 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (VREV64d32 DPR:$src)>;
7380 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (VREV64d32 DPR:$src)>;
7381 def : Pat<(f64 (bitconvert (v4f16 DPR:$src))), (VREV64d16 DPR:$src)>;
7382 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (VREV64d16 DPR:$src)>;
7383 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (VREV64d8 DPR:$src)>;
7385 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (VREV64d32 DPR:$src)>;
7386 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (VREV64d32 DPR:$src)>;
7387 def : Pat<(v1i64 (bitconvert (v4f16 DPR:$src))), (VREV64d16 DPR:$src)>;
7388 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (VREV64d16 DPR:$src)>;
7389 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (VREV64d8 DPR:$src)>;
7391 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (VREV64d32 DPR:$src)>;
7392 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (VREV64d32 DPR:$src)>;
7393 def : Pat<(v2f32 (bitconvert (v4f16 DPR:$src))), (VREV32d16 DPR:$src)>;
7394 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (VREV32d16 DPR:$src)>;
7395 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (VREV32d8 DPR:$src)>;
7397 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (VREV64d32 DPR:$src)>;
7398 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (VREV64d32 DPR:$src)>;
7399 def : Pat<(v2i32 (bitconvert (v4f16 DPR:$src))), (VREV32d16 DPR:$src)>;
7400 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (VREV32d16 DPR:$src)>;
7401 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (VREV32d8 DPR:$src)>;
7403 def : Pat<(v4f16 (bitconvert (f64 DPR:$src))), (VREV64d16 DPR:$src)>;
7404 def : Pat<(v4f16 (bitconvert (v1i64 DPR:$src))), (VREV64d16 DPR:$src)>;
7405 def : Pat<(v4f16 (bitconvert (v2f32 DPR:$src))), (VREV32d16 DPR:$src)>;
7406 def : Pat<(v4f16 (bitconvert (v2i32 DPR:$src))), (VREV32d16 DPR:$src)>;
7407 def : Pat<(v4f16 (bitconvert (v8i8 DPR:$src))), (VREV16d8 DPR:$src)>;
7409 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (VREV64d16 DPR:$src)>;
7410 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (VREV64d16 DPR:$src)>;
7411 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (VREV32d16 DPR:$src)>;
7412 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (VREV32d16 DPR:$src)>;
7413 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (VREV16d8 DPR:$src)>;
7415 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (VREV64d8 DPR:$src)>;
7416 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (VREV64d8 DPR:$src)>;
7417 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (VREV32d8 DPR:$src)>;
7418 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (VREV32d8 DPR:$src)>;
7419 def : Pat<(v8i8 (bitconvert (v4f16 DPR:$src))), (VREV16d8 DPR:$src)>;
7420 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (VREV16d8 DPR:$src)>;
7422 // 128 bit conversions
7423 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (VREV64q32 QPR:$src)>;
7424 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (VREV64q32 QPR:$src)>;
7425 def : Pat<(v2f64 (bitconvert (v8f16 QPR:$src))), (VREV64q16 QPR:$src)>;
7426 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (VREV64q16 QPR:$src)>;
7427 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (VREV64q8 QPR:$src)>;
7429 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (VREV64q32 QPR:$src)>;
7430 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (VREV64q32 QPR:$src)>;
7431 def : Pat<(v2i64 (bitconvert (v8f16 QPR:$src))), (VREV64q16 QPR:$src)>;
7432 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (VREV64q16 QPR:$src)>;
7433 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (VREV64q8 QPR:$src)>;
7435 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (VREV64q32 QPR:$src)>;
7436 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (VREV64q32 QPR:$src)>;
7437 def : Pat<(v4f32 (bitconvert (v8f16 QPR:$src))), (VREV32q16 QPR:$src)>;
7438 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (VREV32q16 QPR:$src)>;
7439 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (VREV32q8 QPR:$src)>;
7441 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (VREV64q32 QPR:$src)>;
7442 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (VREV64q32 QPR:$src)>;
7443 def : Pat<(v4i32 (bitconvert (v8f16 QPR:$src))), (VREV32q16 QPR:$src)>;
7444 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (VREV32q16 QPR:$src)>;
7445 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (VREV32q8 QPR:$src)>;
7447 def : Pat<(v8f16 (bitconvert (v2f64 QPR:$src))), (VREV64q16 QPR:$src)>;
7448 def : Pat<(v8f16 (bitconvert (v2i64 QPR:$src))), (VREV64q16 QPR:$src)>;
7449 def : Pat<(v8f16 (bitconvert (v4f32 QPR:$src))), (VREV32q16 QPR:$src)>;
7450 def : Pat<(v8f16 (bitconvert (v4i32 QPR:$src))), (VREV32q16 QPR:$src)>;
7451 def : Pat<(v8f16 (bitconvert (v16i8 QPR:$src))), (VREV16q8 QPR:$src)>;
7453 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (VREV64q16 QPR:$src)>;
7454 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (VREV64q16 QPR:$src)>;
7455 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (VREV32q16 QPR:$src)>;
7456 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (VREV32q16 QPR:$src)>;
7457 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (VREV16q8 QPR:$src)>;
7459 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (VREV64q8 QPR:$src)>;
7460 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (VREV64q8 QPR:$src)>;
7461 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (VREV32q8 QPR:$src)>;
7462 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (VREV32q8 QPR:$src)>;
7463 def : Pat<(v16i8 (bitconvert (v8f16 QPR:$src))), (VREV16q8 QPR:$src)>;
7464 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (VREV16q8 QPR:$src)>;
7467 // Use VLD1/VST1 + VREV for non-word-aligned v2f64 load/store on Big Endian
7468 let Predicates = [IsBE,HasNEON] in {
7469 def : Pat<(v2f64 (byte_alignedload addrmode6:$addr)),
7470 (VREV64q8 (VLD1q8 addrmode6:$addr))>;
7471 def : Pat<(byte_alignedstore (v2f64 QPR:$value), addrmode6:$addr),
7472 (VST1q8 addrmode6:$addr, (VREV64q8 QPR:$value))>;
7473 def : Pat<(v2f64 (hword_alignedload addrmode6:$addr)),
7474 (VREV64q16 (VLD1q16 addrmode6:$addr))>;
7475 def : Pat<(hword_alignedstore (v2f64 QPR:$value), addrmode6:$addr),
7476 (VST1q16 addrmode6:$addr, (VREV64q16 QPR:$value))>;
7479 // Fold extracting an element out of a v2i32 into a vfp register.
7480 def : Pat<(f32 (bitconvert (i32 (extractelt (v2i32 DPR:$src), imm:$lane)))),
7481 (f32 (EXTRACT_SUBREG DPR:$src, (SSubReg_f32_reg imm:$lane)))>,
7482 Requires<[HasNEON]>;
7484 // Vector lengthening move with load, matching extending loads.
7486 // extload, zextload and sextload for a standard lengthening load. Example:
7487 // Lengthen_Single<"8", "i16", "8"> =
7488 // Pat<(v8i16 (extloadvi8 addrmode6:$addr))
7489 // (VMOVLuv8i16 (VLD1d8 addrmode6:$addr,
7490 // (f64 (IMPLICIT_DEF)), (i32 0)))>;
7491 multiclass Lengthen_Single<string DestLanes, string DestTy, string SrcTy> {
7492 let AddedComplexity = 10 in {
7493 def _Any : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
7494 (!cast<PatFrag>("extloadvi" # SrcTy) addrmode6:$addr)),
7495 (!cast<Instruction>("VMOVLuv" # DestLanes # DestTy)
7496 (!cast<Instruction>("VLD1d" # SrcTy) addrmode6:$addr))>,
7497 Requires<[HasNEON]>;
7499 def _Z : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
7500 (!cast<PatFrag>("zextloadvi" # SrcTy) addrmode6:$addr)),
7501 (!cast<Instruction>("VMOVLuv" # DestLanes # DestTy)
7502 (!cast<Instruction>("VLD1d" # SrcTy) addrmode6:$addr))>,
7503 Requires<[HasNEON]>;
7505 def _S : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
7506 (!cast<PatFrag>("sextloadvi" # SrcTy) addrmode6:$addr)),
7507 (!cast<Instruction>("VMOVLsv" # DestLanes # DestTy)
7508 (!cast<Instruction>("VLD1d" # SrcTy) addrmode6:$addr))>,
7509 Requires<[HasNEON]>;
7513 // extload, zextload and sextload for a lengthening load which only uses
7514 // half the lanes available. Example:
7515 // Lengthen_HalfSingle<"4", "i16", "8", "i16", "i8"> =
7516 // Pat<(v4i16 (extloadvi8 addrmode6oneL32:$addr)),
7517 // (EXTRACT_SUBREG (VMOVLuv8i16 (VLD1LNd32 addrmode6oneL32:$addr,
7518 // (f64 (IMPLICIT_DEF)), (i32 0))),
7520 multiclass Lengthen_HalfSingle<string DestLanes, string DestTy, string SrcTy,
7521 string InsnLanes, string InsnTy> {
7522 def _Any : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
7523 (!cast<PatFrag>("extloadv" # SrcTy) addrmode6oneL32:$addr)),
7524 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # InsnLanes # InsnTy)
7525 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
7527 Requires<[HasNEON]>;
7528 def _Z : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
7529 (!cast<PatFrag>("zextloadv" # SrcTy) addrmode6oneL32:$addr)),
7530 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # InsnLanes # InsnTy)
7531 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
7533 Requires<[HasNEON]>;
7534 def _S : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
7535 (!cast<PatFrag>("sextloadv" # SrcTy) addrmode6oneL32:$addr)),
7536 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLsv" # InsnLanes # InsnTy)
7537 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
7539 Requires<[HasNEON]>;
7542 // The following class definition is basically a copy of the
7543 // Lengthen_HalfSingle definition above, however with an additional parameter
7544 // "RevLanes" to select the correct VREV32dXX instruction. This is to convert
7545 // data loaded by VLD1LN into proper vector format in big endian mode.
7546 multiclass Lengthen_HalfSingle_Big_Endian<string DestLanes, string DestTy, string SrcTy,
7547 string InsnLanes, string InsnTy, string RevLanes> {
7548 def _Any : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
7549 (!cast<PatFrag>("extloadv" # SrcTy) addrmode6oneL32:$addr)),
7550 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # InsnLanes # InsnTy)
7551 (!cast<Instruction>("VREV32d" # RevLanes)
7552 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0)))),
7554 Requires<[HasNEON]>;
7555 def _Z : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
7556 (!cast<PatFrag>("zextloadv" # SrcTy) addrmode6oneL32:$addr)),
7557 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # InsnLanes # InsnTy)
7558 (!cast<Instruction>("VREV32d" # RevLanes)
7559 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0)))),
7561 Requires<[HasNEON]>;
7562 def _S : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
7563 (!cast<PatFrag>("sextloadv" # SrcTy) addrmode6oneL32:$addr)),
7564 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLsv" # InsnLanes # InsnTy)
7565 (!cast<Instruction>("VREV32d" # RevLanes)
7566 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0)))),
7568 Requires<[HasNEON]>;
7571 // extload, zextload and sextload for a lengthening load followed by another
7572 // lengthening load, to quadruple the initial length.
7574 // Lengthen_Double<"4", "i32", "i8", "8", "i16", "4", "i32"> =
7575 // Pat<(v4i32 (extloadvi8 addrmode6oneL32:$addr))
7576 // (EXTRACT_SUBREG (VMOVLuv4i32
7577 // (EXTRACT_SUBREG (VMOVLuv8i16 (VLD1LNd32 addrmode6oneL32:$addr,
7578 // (f64 (IMPLICIT_DEF)),
7582 multiclass Lengthen_Double<string DestLanes, string DestTy, string SrcTy,
7583 string Insn1Lanes, string Insn1Ty, string Insn2Lanes,
7585 def _Any : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
7586 (!cast<PatFrag>("extloadv" # SrcTy) addrmode6oneL32:$addr)),
7587 (!cast<Instruction>("VMOVLuv" # Insn2Lanes # Insn2Ty)
7588 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn1Lanes # Insn1Ty)
7589 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
7591 Requires<[HasNEON]>;
7592 def _Z : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
7593 (!cast<PatFrag>("zextloadv" # SrcTy) addrmode6oneL32:$addr)),
7594 (!cast<Instruction>("VMOVLuv" # Insn2Lanes # Insn2Ty)
7595 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn1Lanes # Insn1Ty)
7596 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
7598 Requires<[HasNEON]>;
7599 def _S : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
7600 (!cast<PatFrag>("sextloadv" # SrcTy) addrmode6oneL32:$addr)),
7601 (!cast<Instruction>("VMOVLsv" # Insn2Lanes # Insn2Ty)
7602 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLsv" # Insn1Lanes # Insn1Ty)
7603 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
7605 Requires<[HasNEON]>;
7608 // The following class definition is basically a copy of the
7609 // Lengthen_Double definition above, however with an additional parameter
7610 // "RevLanes" to select the correct VREV32dXX instruction. This is to convert
7611 // data loaded by VLD1LN into proper vector format in big endian mode.
7612 multiclass Lengthen_Double_Big_Endian<string DestLanes, string DestTy, string SrcTy,
7613 string Insn1Lanes, string Insn1Ty, string Insn2Lanes,
7614 string Insn2Ty, string RevLanes> {
7615 def _Any : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
7616 (!cast<PatFrag>("extloadv" # SrcTy) addrmode6oneL32:$addr)),
7617 (!cast<Instruction>("VMOVLuv" # Insn2Lanes # Insn2Ty)
7618 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn1Lanes # Insn1Ty)
7619 (!cast<Instruction>("VREV32d" # RevLanes)
7620 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0)))),
7622 Requires<[HasNEON]>;
7623 def _Z : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
7624 (!cast<PatFrag>("zextloadv" # SrcTy) addrmode6oneL32:$addr)),
7625 (!cast<Instruction>("VMOVLuv" # Insn2Lanes # Insn2Ty)
7626 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn1Lanes # Insn1Ty)
7627 (!cast<Instruction>("VREV32d" # RevLanes)
7628 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0)))),
7630 Requires<[HasNEON]>;
7631 def _S : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
7632 (!cast<PatFrag>("sextloadv" # SrcTy) addrmode6oneL32:$addr)),
7633 (!cast<Instruction>("VMOVLsv" # Insn2Lanes # Insn2Ty)
7634 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLsv" # Insn1Lanes # Insn1Ty)
7635 (!cast<Instruction>("VREV32d" # RevLanes)
7636 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0)))),
7638 Requires<[HasNEON]>;
7641 // extload, zextload and sextload for a lengthening load followed by another
7642 // lengthening load, to quadruple the initial length, but which ends up only
7643 // requiring half the available lanes (a 64-bit outcome instead of a 128-bit).
7645 // Lengthen_HalfDouble<"2", "i32", "i8", "8", "i16", "4", "i32"> =
7646 // Pat<(v2i32 (extloadvi8 addrmode6:$addr))
7647 // (EXTRACT_SUBREG (VMOVLuv4i32
7648 // (EXTRACT_SUBREG (VMOVLuv8i16 (VLD1LNd16 addrmode6:$addr,
7649 // (f64 (IMPLICIT_DEF)), (i32 0))),
7652 multiclass Lengthen_HalfDouble<string DestLanes, string DestTy, string SrcTy,
7653 string Insn1Lanes, string Insn1Ty, string Insn2Lanes,
7655 def _Any : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
7656 (!cast<PatFrag>("extloadv" # SrcTy) addrmode6:$addr)),
7657 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn2Lanes # Insn2Ty)
7658 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn1Lanes # Insn1Ty)
7659 (VLD1LNd16 addrmode6:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
7662 Requires<[HasNEON]>;
7663 def _Z : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
7664 (!cast<PatFrag>("zextloadv" # SrcTy) addrmode6:$addr)),
7665 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn2Lanes # Insn2Ty)
7666 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn1Lanes # Insn1Ty)
7667 (VLD1LNd16 addrmode6:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
7670 Requires<[HasNEON]>;
7671 def _S : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
7672 (!cast<PatFrag>("sextloadv" # SrcTy) addrmode6:$addr)),
7673 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLsv" # Insn2Lanes # Insn2Ty)
7674 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLsv" # Insn1Lanes # Insn1Ty)
7675 (VLD1LNd16 addrmode6:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
7678 Requires<[HasNEON]>;
7681 // The following class definition is basically a copy of the
7682 // Lengthen_HalfDouble definition above, however with an additional VREV16d8
7683 // instruction to convert data loaded by VLD1LN into proper vector format
7684 // in big endian mode.
7685 multiclass Lengthen_HalfDouble_Big_Endian<string DestLanes, string DestTy, string SrcTy,
7686 string Insn1Lanes, string Insn1Ty, string Insn2Lanes,
7688 def _Any : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
7689 (!cast<PatFrag>("extloadv" # SrcTy) addrmode6:$addr)),
7690 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn2Lanes # Insn2Ty)
7691 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn1Lanes # Insn1Ty)
7692 (!cast<Instruction>("VREV16d8")
7693 (VLD1LNd16 addrmode6:$addr, (f64 (IMPLICIT_DEF)), (i32 0)))),
7696 Requires<[HasNEON]>;
7697 def _Z : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
7698 (!cast<PatFrag>("zextloadv" # SrcTy) addrmode6:$addr)),
7699 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn2Lanes # Insn2Ty)
7700 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn1Lanes # Insn1Ty)
7701 (!cast<Instruction>("VREV16d8")
7702 (VLD1LNd16 addrmode6:$addr, (f64 (IMPLICIT_DEF)), (i32 0)))),
7705 Requires<[HasNEON]>;
7706 def _S : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
7707 (!cast<PatFrag>("sextloadv" # SrcTy) addrmode6:$addr)),
7708 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLsv" # Insn2Lanes # Insn2Ty)
7709 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLsv" # Insn1Lanes # Insn1Ty)
7710 (!cast<Instruction>("VREV16d8")
7711 (VLD1LNd16 addrmode6:$addr, (f64 (IMPLICIT_DEF)), (i32 0)))),
7714 Requires<[HasNEON]>;
7717 defm : Lengthen_Single<"8", "i16", "8">; // v8i8 -> v8i16
7718 defm : Lengthen_Single<"4", "i32", "16">; // v4i16 -> v4i32
7719 defm : Lengthen_Single<"2", "i64", "32">; // v2i32 -> v2i64
7721 let Predicates = [HasNEON,IsLE] in {
7722 defm : Lengthen_HalfSingle<"4", "i16", "i8", "8", "i16">; // v4i8 -> v4i16
7723 defm : Lengthen_HalfSingle<"2", "i32", "i16", "4", "i32">; // v2i16 -> v2i32
7725 // Double lengthening - v4i8 -> v4i16 -> v4i32
7726 defm : Lengthen_Double<"4", "i32", "i8", "8", "i16", "4", "i32">;
7727 // v2i8 -> v2i16 -> v2i32
7728 defm : Lengthen_HalfDouble<"2", "i32", "i8", "8", "i16", "4", "i32">;
7729 // v2i16 -> v2i32 -> v2i64
7730 defm : Lengthen_Double<"2", "i64", "i16", "4", "i32", "2", "i64">;
7733 let Predicates = [HasNEON,IsBE] in {
7734 defm : Lengthen_HalfSingle_Big_Endian<"4", "i16", "i8", "8", "i16", "8">; // v4i8 -> v4i16
7735 defm : Lengthen_HalfSingle_Big_Endian<"2", "i32", "i16", "4", "i32", "16">; // v2i16 -> v2i32
7737 // Double lengthening - v4i8 -> v4i16 -> v4i32
7738 defm : Lengthen_Double_Big_Endian<"4", "i32", "i8", "8", "i16", "4", "i32", "8">;
7739 // v2i8 -> v2i16 -> v2i32
7740 defm : Lengthen_HalfDouble_Big_Endian<"2", "i32", "i8", "8", "i16", "4", "i32">;
7741 // v2i16 -> v2i32 -> v2i64
7742 defm : Lengthen_Double_Big_Endian<"2", "i64", "i16", "4", "i32", "2", "i64", "16">;
7745 // Triple lengthening - v2i8 -> v2i16 -> v2i32 -> v2i64
7746 let Predicates = [HasNEON,IsLE] in {
7747 def : Pat<(v2i64 (extloadvi8 addrmode6:$addr)),
7748 (VMOVLuv2i64 (EXTRACT_SUBREG (VMOVLuv4i32 (EXTRACT_SUBREG (VMOVLuv8i16
7749 (VLD1LNd16 addrmode6:$addr,
7750 (f64 (IMPLICIT_DEF)), (i32 0))), dsub_0)), dsub_0))>;
7751 def : Pat<(v2i64 (zextloadvi8 addrmode6:$addr)),
7752 (VMOVLuv2i64 (EXTRACT_SUBREG (VMOVLuv4i32 (EXTRACT_SUBREG (VMOVLuv8i16
7753 (VLD1LNd16 addrmode6:$addr,
7754 (f64 (IMPLICIT_DEF)), (i32 0))), dsub_0)), dsub_0))>;
7755 def : Pat<(v2i64 (sextloadvi8 addrmode6:$addr)),
7756 (VMOVLsv2i64 (EXTRACT_SUBREG (VMOVLsv4i32 (EXTRACT_SUBREG (VMOVLsv8i16
7757 (VLD1LNd16 addrmode6:$addr,
7758 (f64 (IMPLICIT_DEF)), (i32 0))), dsub_0)), dsub_0))>;
7760 // The following patterns are basically a copy of the patterns above,
7761 // however with an additional VREV16d instruction to convert data
7762 // loaded by VLD1LN into proper vector format in big endian mode.
7763 let Predicates = [HasNEON,IsBE] in {
7764 def : Pat<(v2i64 (extloadvi8 addrmode6:$addr)),
7765 (VMOVLuv2i64 (EXTRACT_SUBREG (VMOVLuv4i32 (EXTRACT_SUBREG (VMOVLuv8i16
7766 (!cast<Instruction>("VREV16d8")
7767 (VLD1LNd16 addrmode6:$addr,
7768 (f64 (IMPLICIT_DEF)), (i32 0)))), dsub_0)), dsub_0))>;
7769 def : Pat<(v2i64 (zextloadvi8 addrmode6:$addr)),
7770 (VMOVLuv2i64 (EXTRACT_SUBREG (VMOVLuv4i32 (EXTRACT_SUBREG (VMOVLuv8i16
7771 (!cast<Instruction>("VREV16d8")
7772 (VLD1LNd16 addrmode6:$addr,
7773 (f64 (IMPLICIT_DEF)), (i32 0)))), dsub_0)), dsub_0))>;
7774 def : Pat<(v2i64 (sextloadvi8 addrmode6:$addr)),
7775 (VMOVLsv2i64 (EXTRACT_SUBREG (VMOVLsv4i32 (EXTRACT_SUBREG (VMOVLsv8i16
7776 (!cast<Instruction>("VREV16d8")
7777 (VLD1LNd16 addrmode6:$addr,
7778 (f64 (IMPLICIT_DEF)), (i32 0)))), dsub_0)), dsub_0))>;
7781 let Predicates = [HasNEON] in {
7782 def : Pat<(v2i64 (concat_vectors DPR:$Dn, DPR:$Dm)),
7783 (REG_SEQUENCE QPR, DPR:$Dn, dsub_0, DPR:$Dm, dsub_1)>;
7784 def : Pat<(v4i32 (concat_vectors DPR:$Dn, DPR:$Dm)),
7785 (REG_SEQUENCE QPR, DPR:$Dn, dsub_0, DPR:$Dm, dsub_1)>;
7786 def : Pat<(v8i16 (concat_vectors DPR:$Dn, DPR:$Dm)),
7787 (REG_SEQUENCE QPR, DPR:$Dn, dsub_0, DPR:$Dm, dsub_1)>;
7788 def : Pat<(v16i8 (concat_vectors DPR:$Dn, DPR:$Dm)),
7789 (REG_SEQUENCE QPR, DPR:$Dn, dsub_0, DPR:$Dm, dsub_1)>;
7790 def : Pat<(v4f32 (concat_vectors DPR:$Dn, DPR:$Dm)),
7791 (REG_SEQUENCE QPR, DPR:$Dn, dsub_0, DPR:$Dm, dsub_1)>;
7792 def : Pat<(v8f16 (concat_vectors DPR:$Dn, DPR:$Dm)),
7793 (REG_SEQUENCE QPR, DPR:$Dn, dsub_0, DPR:$Dm, dsub_1)>;
7796 //===----------------------------------------------------------------------===//
7797 // Assembler aliases
7800 def : VFP2InstAlias<"fmdhr${p} $Dd, $Rn",
7801 (VSETLNi32 DPR:$Dd, GPR:$Rn, 1, pred:$p)>;
7802 def : VFP2InstAlias<"fmdlr${p} $Dd, $Rn",
7803 (VSETLNi32 DPR:$Dd, GPR:$Rn, 0, pred:$p)>;
7805 // VAND/VBIC/VEOR/VORR accept but do not require a type suffix.
7806 defm : NEONDTAnyInstAlias<"vand${p}", "$Vd, $Vn, $Vm",
7807 (VANDd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
7808 defm : NEONDTAnyInstAlias<"vand${p}", "$Vd, $Vn, $Vm",
7809 (VANDq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
7810 defm : NEONDTAnyInstAlias<"vbic${p}", "$Vd, $Vn, $Vm",
7811 (VBICd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
7812 defm : NEONDTAnyInstAlias<"vbic${p}", "$Vd, $Vn, $Vm",
7813 (VBICq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
7814 defm : NEONDTAnyInstAlias<"veor${p}", "$Vd, $Vn, $Vm",
7815 (VEORd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
7816 defm : NEONDTAnyInstAlias<"veor${p}", "$Vd, $Vn, $Vm",
7817 (VEORq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
7818 defm : NEONDTAnyInstAlias<"vorr${p}", "$Vd, $Vn, $Vm",
7819 (VORRd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
7820 defm : NEONDTAnyInstAlias<"vorr${p}", "$Vd, $Vn, $Vm",
7821 (VORRq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
7822 // ... two-operand aliases
7823 defm : NEONDTAnyInstAlias<"vand${p}", "$Vdn, $Vm",
7824 (VANDd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
7825 defm : NEONDTAnyInstAlias<"vand${p}", "$Vdn, $Vm",
7826 (VANDq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
7827 defm : NEONDTAnyInstAlias<"veor${p}", "$Vdn, $Vm",
7828 (VEORd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
7829 defm : NEONDTAnyInstAlias<"veor${p}", "$Vdn, $Vm",
7830 (VEORq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
7831 defm : NEONDTAnyInstAlias<"vorr${p}", "$Vdn, $Vm",
7832 (VORRd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
7833 defm : NEONDTAnyInstAlias<"vorr${p}", "$Vdn, $Vm",
7834 (VORRq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
7836 def : NEONInstAlias<"vand${p}.i16 $Vd, $imm",
7837 (VBICiv4i16 DPR:$Vd, nImmSplatNotI16:$imm, pred:$p)>;
7838 def : NEONInstAlias<"vand${p}.i32 $Vd, $imm",
7839 (VBICiv2i32 DPR:$Vd, nImmSplatNotI32:$imm, pred:$p)>;
7840 def : NEONInstAlias<"vand${p}.i16 $Vd, $imm",
7841 (VBICiv8i16 QPR:$Vd, nImmSplatNotI16:$imm, pred:$p)>;
7842 def : NEONInstAlias<"vand${p}.i32 $Vd, $imm",
7843 (VBICiv4i32 QPR:$Vd, nImmSplatNotI32:$imm, pred:$p)>;
7846 // VLD1 single-lane pseudo-instructions. These need special handling for
7847 // the lane index that an InstAlias can't handle, so we use these instead.
7848 def VLD1LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vld1${p}", ".8", "$list, $addr",
7849 (ins VecListOneDByteIndexed:$list, addrmode6alignNone:$addr,
7851 def VLD1LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vld1${p}", ".16", "$list, $addr",
7852 (ins VecListOneDHWordIndexed:$list, addrmode6align16:$addr,
7854 def VLD1LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vld1${p}", ".32", "$list, $addr",
7855 (ins VecListOneDWordIndexed:$list, addrmode6align32:$addr,
7858 def VLD1LNdWB_fixed_Asm_8 :
7859 NEONDataTypeAsmPseudoInst<"vld1${p}", ".8", "$list, $addr!",
7860 (ins VecListOneDByteIndexed:$list, addrmode6alignNone:$addr,
7862 def VLD1LNdWB_fixed_Asm_16 :
7863 NEONDataTypeAsmPseudoInst<"vld1${p}", ".16", "$list, $addr!",
7864 (ins VecListOneDHWordIndexed:$list, addrmode6align16:$addr,
7866 def VLD1LNdWB_fixed_Asm_32 :
7867 NEONDataTypeAsmPseudoInst<"vld1${p}", ".32", "$list, $addr!",
7868 (ins VecListOneDWordIndexed:$list, addrmode6align32:$addr,
7870 def VLD1LNdWB_register_Asm_8 :
7871 NEONDataTypeAsmPseudoInst<"vld1${p}", ".8", "$list, $addr, $Rm",
7872 (ins VecListOneDByteIndexed:$list, addrmode6alignNone:$addr,
7873 rGPR:$Rm, pred:$p)>;
7874 def VLD1LNdWB_register_Asm_16 :
7875 NEONDataTypeAsmPseudoInst<"vld1${p}", ".16", "$list, $addr, $Rm",
7876 (ins VecListOneDHWordIndexed:$list, addrmode6align16:$addr,
7877 rGPR:$Rm, pred:$p)>;
7878 def VLD1LNdWB_register_Asm_32 :
7879 NEONDataTypeAsmPseudoInst<"vld1${p}", ".32", "$list, $addr, $Rm",
7880 (ins VecListOneDWordIndexed:$list, addrmode6align32:$addr,
7881 rGPR:$Rm, pred:$p)>;
7884 // VST1 single-lane pseudo-instructions. These need special handling for
7885 // the lane index that an InstAlias can't handle, so we use these instead.
7886 def VST1LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vst1${p}", ".8", "$list, $addr",
7887 (ins VecListOneDByteIndexed:$list, addrmode6alignNone:$addr,
7889 def VST1LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vst1${p}", ".16", "$list, $addr",
7890 (ins VecListOneDHWordIndexed:$list, addrmode6align16:$addr,
7892 def VST1LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vst1${p}", ".32", "$list, $addr",
7893 (ins VecListOneDWordIndexed:$list, addrmode6align32:$addr,
7896 def VST1LNdWB_fixed_Asm_8 :
7897 NEONDataTypeAsmPseudoInst<"vst1${p}", ".8", "$list, $addr!",
7898 (ins VecListOneDByteIndexed:$list, addrmode6alignNone:$addr,
7900 def VST1LNdWB_fixed_Asm_16 :
7901 NEONDataTypeAsmPseudoInst<"vst1${p}", ".16", "$list, $addr!",
7902 (ins VecListOneDHWordIndexed:$list, addrmode6align16:$addr,
7904 def VST1LNdWB_fixed_Asm_32 :
7905 NEONDataTypeAsmPseudoInst<"vst1${p}", ".32", "$list, $addr!",
7906 (ins VecListOneDWordIndexed:$list, addrmode6align32:$addr,
7908 def VST1LNdWB_register_Asm_8 :
7909 NEONDataTypeAsmPseudoInst<"vst1${p}", ".8", "$list, $addr, $Rm",
7910 (ins VecListOneDByteIndexed:$list, addrmode6alignNone:$addr,
7911 rGPR:$Rm, pred:$p)>;
7912 def VST1LNdWB_register_Asm_16 :
7913 NEONDataTypeAsmPseudoInst<"vst1${p}", ".16", "$list, $addr, $Rm",
7914 (ins VecListOneDHWordIndexed:$list, addrmode6align16:$addr,
7915 rGPR:$Rm, pred:$p)>;
7916 def VST1LNdWB_register_Asm_32 :
7917 NEONDataTypeAsmPseudoInst<"vst1${p}", ".32", "$list, $addr, $Rm",
7918 (ins VecListOneDWordIndexed:$list, addrmode6align32:$addr,
7919 rGPR:$Rm, pred:$p)>;
7921 // VLD2 single-lane pseudo-instructions. These need special handling for
7922 // the lane index that an InstAlias can't handle, so we use these instead.
7923 def VLD2LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vld2${p}", ".8", "$list, $addr",
7924 (ins VecListTwoDByteIndexed:$list, addrmode6align16:$addr,
7926 def VLD2LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr",
7927 (ins VecListTwoDHWordIndexed:$list, addrmode6align32:$addr,
7929 def VLD2LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr",
7930 (ins VecListTwoDWordIndexed:$list, addrmode6align64:$addr, pred:$p)>;
7931 def VLD2LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr",
7932 (ins VecListTwoQHWordIndexed:$list, addrmode6align32:$addr,
7934 def VLD2LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr",
7935 (ins VecListTwoQWordIndexed:$list, addrmode6align64:$addr,
7938 def VLD2LNdWB_fixed_Asm_8 :
7939 NEONDataTypeAsmPseudoInst<"vld2${p}", ".8", "$list, $addr!",
7940 (ins VecListTwoDByteIndexed:$list, addrmode6align16:$addr,
7942 def VLD2LNdWB_fixed_Asm_16 :
7943 NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr!",
7944 (ins VecListTwoDHWordIndexed:$list, addrmode6align32:$addr,
7946 def VLD2LNdWB_fixed_Asm_32 :
7947 NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr!",
7948 (ins VecListTwoDWordIndexed:$list, addrmode6align64:$addr,
7950 def VLD2LNqWB_fixed_Asm_16 :
7951 NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr!",
7952 (ins VecListTwoQHWordIndexed:$list, addrmode6align32:$addr,
7954 def VLD2LNqWB_fixed_Asm_32 :
7955 NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr!",
7956 (ins VecListTwoQWordIndexed:$list, addrmode6align64:$addr,
7958 def VLD2LNdWB_register_Asm_8 :
7959 NEONDataTypeAsmPseudoInst<"vld2${p}", ".8", "$list, $addr, $Rm",
7960 (ins VecListTwoDByteIndexed:$list, addrmode6align16:$addr,
7961 rGPR:$Rm, pred:$p)>;
7962 def VLD2LNdWB_register_Asm_16 :
7963 NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr, $Rm",
7964 (ins VecListTwoDHWordIndexed:$list, addrmode6align32:$addr,
7965 rGPR:$Rm, pred:$p)>;
7966 def VLD2LNdWB_register_Asm_32 :
7967 NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr, $Rm",
7968 (ins VecListTwoDWordIndexed:$list, addrmode6align64:$addr,
7969 rGPR:$Rm, pred:$p)>;
7970 def VLD2LNqWB_register_Asm_16 :
7971 NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr, $Rm",
7972 (ins VecListTwoQHWordIndexed:$list, addrmode6align32:$addr,
7973 rGPR:$Rm, pred:$p)>;
7974 def VLD2LNqWB_register_Asm_32 :
7975 NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr, $Rm",
7976 (ins VecListTwoQWordIndexed:$list, addrmode6align64:$addr,
7977 rGPR:$Rm, pred:$p)>;
7980 // VST2 single-lane pseudo-instructions. These need special handling for
7981 // the lane index that an InstAlias can't handle, so we use these instead.
7982 def VST2LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vst2${p}", ".8", "$list, $addr",
7983 (ins VecListTwoDByteIndexed:$list, addrmode6align16:$addr,
7985 def VST2LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vst2${p}", ".16", "$list, $addr",
7986 (ins VecListTwoDHWordIndexed:$list, addrmode6align32:$addr,
7988 def VST2LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr",
7989 (ins VecListTwoDWordIndexed:$list, addrmode6align64:$addr,
7991 def VST2LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vst2${p}", ".16", "$list, $addr",
7992 (ins VecListTwoQHWordIndexed:$list, addrmode6align32:$addr,
7994 def VST2LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr",
7995 (ins VecListTwoQWordIndexed:$list, addrmode6align64:$addr,
7998 def VST2LNdWB_fixed_Asm_8 :
7999 NEONDataTypeAsmPseudoInst<"vst2${p}", ".8", "$list, $addr!",
8000 (ins VecListTwoDByteIndexed:$list, addrmode6align16:$addr,
8002 def VST2LNdWB_fixed_Asm_16 :
8003 NEONDataTypeAsmPseudoInst<"vst2${p}", ".16", "$list, $addr!",
8004 (ins VecListTwoDHWordIndexed:$list, addrmode6align32:$addr,
8006 def VST2LNdWB_fixed_Asm_32 :
8007 NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr!",
8008 (ins VecListTwoDWordIndexed:$list, addrmode6align64:$addr,
8010 def VST2LNqWB_fixed_Asm_16 :
8011 NEONDataTypeAsmPseudoInst<"vst2${p}", ".16", "$list, $addr!",
8012 (ins VecListTwoQHWordIndexed:$list, addrmode6align32:$addr,
8014 def VST2LNqWB_fixed_Asm_32 :
8015 NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr!",
8016 (ins VecListTwoQWordIndexed:$list, addrmode6align64:$addr,
8018 def VST2LNdWB_register_Asm_8 :
8019 NEONDataTypeAsmPseudoInst<"vst2${p}", ".8", "$list, $addr, $Rm",
8020 (ins VecListTwoDByteIndexed:$list, addrmode6align16:$addr,
8021 rGPR:$Rm, pred:$p)>;
8022 def VST2LNdWB_register_Asm_16 :
8023 NEONDataTypeAsmPseudoInst<"vst2${p}", ".16","$list, $addr, $Rm",
8024 (ins VecListTwoDHWordIndexed:$list, addrmode6align32:$addr,
8025 rGPR:$Rm, pred:$p)>;
8026 def VST2LNdWB_register_Asm_32 :
8027 NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr, $Rm",
8028 (ins VecListTwoDWordIndexed:$list, addrmode6align64:$addr,
8029 rGPR:$Rm, pred:$p)>;
8030 def VST2LNqWB_register_Asm_16 :
8031 NEONDataTypeAsmPseudoInst<"vst2${p}", ".16","$list, $addr, $Rm",
8032 (ins VecListTwoQHWordIndexed:$list, addrmode6align32:$addr,
8033 rGPR:$Rm, pred:$p)>;
8034 def VST2LNqWB_register_Asm_32 :
8035 NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr, $Rm",
8036 (ins VecListTwoQWordIndexed:$list, addrmode6align64:$addr,
8037 rGPR:$Rm, pred:$p)>;
8039 // VLD3 all-lanes pseudo-instructions. These need special handling for
8040 // the lane index that an InstAlias can't handle, so we use these instead.
8041 def VLD3DUPdAsm_8 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr",
8042 (ins VecListThreeDAllLanes:$list, addrmode6dupalignNone:$addr,
8044 def VLD3DUPdAsm_16: NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
8045 (ins VecListThreeDAllLanes:$list, addrmode6dupalignNone:$addr,
8047 def VLD3DUPdAsm_32: NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
8048 (ins VecListThreeDAllLanes:$list, addrmode6dupalignNone:$addr,
8050 def VLD3DUPqAsm_8 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr",
8051 (ins VecListThreeQAllLanes:$list, addrmode6dupalignNone:$addr,
8053 def VLD3DUPqAsm_16: NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
8054 (ins VecListThreeQAllLanes:$list, addrmode6dupalignNone:$addr,
8056 def VLD3DUPqAsm_32: NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
8057 (ins VecListThreeQAllLanes:$list, addrmode6dupalignNone:$addr,
8060 def VLD3DUPdWB_fixed_Asm_8 :
8061 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr!",
8062 (ins VecListThreeDAllLanes:$list, addrmode6dupalignNone:$addr,
8064 def VLD3DUPdWB_fixed_Asm_16 :
8065 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
8066 (ins VecListThreeDAllLanes:$list, addrmode6dupalignNone:$addr,
8068 def VLD3DUPdWB_fixed_Asm_32 :
8069 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
8070 (ins VecListThreeDAllLanes:$list, addrmode6dupalignNone:$addr,
8072 def VLD3DUPqWB_fixed_Asm_8 :
8073 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr!",
8074 (ins VecListThreeQAllLanes:$list, addrmode6dupalignNone:$addr,
8076 def VLD3DUPqWB_fixed_Asm_16 :
8077 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
8078 (ins VecListThreeQAllLanes:$list, addrmode6dupalignNone:$addr,
8080 def VLD3DUPqWB_fixed_Asm_32 :
8081 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
8082 (ins VecListThreeQAllLanes:$list, addrmode6dupalignNone:$addr,
8084 def VLD3DUPdWB_register_Asm_8 :
8085 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr, $Rm",
8086 (ins VecListThreeDAllLanes:$list, addrmode6dupalignNone:$addr,
8087 rGPR:$Rm, pred:$p)>;
8088 def VLD3DUPdWB_register_Asm_16 :
8089 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
8090 (ins VecListThreeDAllLanes:$list, addrmode6dupalignNone:$addr,
8091 rGPR:$Rm, pred:$p)>;
8092 def VLD3DUPdWB_register_Asm_32 :
8093 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
8094 (ins VecListThreeDAllLanes:$list, addrmode6dupalignNone:$addr,
8095 rGPR:$Rm, pred:$p)>;
8096 def VLD3DUPqWB_register_Asm_8 :
8097 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr, $Rm",
8098 (ins VecListThreeQAllLanes:$list, addrmode6dupalignNone:$addr,
8099 rGPR:$Rm, pred:$p)>;
8100 def VLD3DUPqWB_register_Asm_16 :
8101 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
8102 (ins VecListThreeQAllLanes:$list, addrmode6dupalignNone:$addr,
8103 rGPR:$Rm, pred:$p)>;
8104 def VLD3DUPqWB_register_Asm_32 :
8105 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
8106 (ins VecListThreeQAllLanes:$list, addrmode6dupalignNone:$addr,
8107 rGPR:$Rm, pred:$p)>;
8110 // VLD3 single-lane pseudo-instructions. These need special handling for
8111 // the lane index that an InstAlias can't handle, so we use these instead.
8112 def VLD3LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr",
8113 (ins VecListThreeDByteIndexed:$list, addrmode6alignNone:$addr,
8115 def VLD3LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
8116 (ins VecListThreeDHWordIndexed:$list, addrmode6alignNone:$addr,
8118 def VLD3LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
8119 (ins VecListThreeDWordIndexed:$list, addrmode6alignNone:$addr,
8121 def VLD3LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
8122 (ins VecListThreeQHWordIndexed:$list, addrmode6alignNone:$addr,
8124 def VLD3LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
8125 (ins VecListThreeQWordIndexed:$list, addrmode6alignNone:$addr,
8128 def VLD3LNdWB_fixed_Asm_8 :
8129 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr!",
8130 (ins VecListThreeDByteIndexed:$list, addrmode6alignNone:$addr,
8132 def VLD3LNdWB_fixed_Asm_16 :
8133 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
8134 (ins VecListThreeDHWordIndexed:$list, addrmode6alignNone:$addr,
8136 def VLD3LNdWB_fixed_Asm_32 :
8137 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
8138 (ins VecListThreeDWordIndexed:$list, addrmode6alignNone:$addr,
8140 def VLD3LNqWB_fixed_Asm_16 :
8141 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
8142 (ins VecListThreeQHWordIndexed:$list, addrmode6alignNone:$addr,
8144 def VLD3LNqWB_fixed_Asm_32 :
8145 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
8146 (ins VecListThreeQWordIndexed:$list, addrmode6alignNone:$addr,
8148 def VLD3LNdWB_register_Asm_8 :
8149 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr, $Rm",
8150 (ins VecListThreeDByteIndexed:$list, addrmode6alignNone:$addr,
8151 rGPR:$Rm, pred:$p)>;
8152 def VLD3LNdWB_register_Asm_16 :
8153 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
8154 (ins VecListThreeDHWordIndexed:$list,
8155 addrmode6alignNone:$addr, rGPR:$Rm, pred:$p)>;
8156 def VLD3LNdWB_register_Asm_32 :
8157 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
8158 (ins VecListThreeDWordIndexed:$list, addrmode6alignNone:$addr,
8159 rGPR:$Rm, pred:$p)>;
8160 def VLD3LNqWB_register_Asm_16 :
8161 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
8162 (ins VecListThreeQHWordIndexed:$list,
8163 addrmode6alignNone:$addr, rGPR:$Rm, pred:$p)>;
8164 def VLD3LNqWB_register_Asm_32 :
8165 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
8166 (ins VecListThreeQWordIndexed:$list, addrmode6alignNone:$addr,
8167 rGPR:$Rm, pred:$p)>;
8169 // VLD3 multiple structure pseudo-instructions. These need special handling for
8170 // the vector operands that the normal instructions don't yet model.
8171 // FIXME: Remove these when the register classes and instructions are updated.
8172 def VLD3dAsm_8 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr",
8173 (ins VecListThreeD:$list, addrmode6align64:$addr, pred:$p)>;
8174 def VLD3dAsm_16 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
8175 (ins VecListThreeD:$list, addrmode6align64:$addr, pred:$p)>;
8176 def VLD3dAsm_32 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
8177 (ins VecListThreeD:$list, addrmode6align64:$addr, pred:$p)>;
8178 def VLD3qAsm_8 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr",
8179 (ins VecListThreeQ:$list, addrmode6align64:$addr, pred:$p)>;
8180 def VLD3qAsm_16 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
8181 (ins VecListThreeQ:$list, addrmode6align64:$addr, pred:$p)>;
8182 def VLD3qAsm_32 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
8183 (ins VecListThreeQ:$list, addrmode6align64:$addr, pred:$p)>;
8185 def VLD3dWB_fixed_Asm_8 :
8186 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr!",
8187 (ins VecListThreeD:$list, addrmode6align64:$addr, pred:$p)>;
8188 def VLD3dWB_fixed_Asm_16 :
8189 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
8190 (ins VecListThreeD:$list, addrmode6align64:$addr, pred:$p)>;
8191 def VLD3dWB_fixed_Asm_32 :
8192 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
8193 (ins VecListThreeD:$list, addrmode6align64:$addr, pred:$p)>;
8194 def VLD3qWB_fixed_Asm_8 :
8195 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr!",
8196 (ins VecListThreeQ:$list, addrmode6align64:$addr, pred:$p)>;
8197 def VLD3qWB_fixed_Asm_16 :
8198 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
8199 (ins VecListThreeQ:$list, addrmode6align64:$addr, pred:$p)>;
8200 def VLD3qWB_fixed_Asm_32 :
8201 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
8202 (ins VecListThreeQ:$list, addrmode6align64:$addr, pred:$p)>;
8203 def VLD3dWB_register_Asm_8 :
8204 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr, $Rm",
8205 (ins VecListThreeD:$list, addrmode6align64:$addr,
8206 rGPR:$Rm, pred:$p)>;
8207 def VLD3dWB_register_Asm_16 :
8208 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
8209 (ins VecListThreeD:$list, addrmode6align64:$addr,
8210 rGPR:$Rm, pred:$p)>;
8211 def VLD3dWB_register_Asm_32 :
8212 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
8213 (ins VecListThreeD:$list, addrmode6align64:$addr,
8214 rGPR:$Rm, pred:$p)>;
8215 def VLD3qWB_register_Asm_8 :
8216 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr, $Rm",
8217 (ins VecListThreeQ:$list, addrmode6align64:$addr,
8218 rGPR:$Rm, pred:$p)>;
8219 def VLD3qWB_register_Asm_16 :
8220 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
8221 (ins VecListThreeQ:$list, addrmode6align64:$addr,
8222 rGPR:$Rm, pred:$p)>;
8223 def VLD3qWB_register_Asm_32 :
8224 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
8225 (ins VecListThreeQ:$list, addrmode6align64:$addr,
8226 rGPR:$Rm, pred:$p)>;
8228 // VST3 single-lane pseudo-instructions. These need special handling for
8229 // the lane index that an InstAlias can't handle, so we use these instead.
8230 def VST3LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr",
8231 (ins VecListThreeDByteIndexed:$list, addrmode6alignNone:$addr,
8233 def VST3LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr",
8234 (ins VecListThreeDHWordIndexed:$list, addrmode6alignNone:$addr,
8236 def VST3LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr",
8237 (ins VecListThreeDWordIndexed:$list, addrmode6alignNone:$addr,
8239 def VST3LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr",
8240 (ins VecListThreeQHWordIndexed:$list, addrmode6alignNone:$addr,
8242 def VST3LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr",
8243 (ins VecListThreeQWordIndexed:$list, addrmode6alignNone:$addr,
8246 def VST3LNdWB_fixed_Asm_8 :
8247 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr!",
8248 (ins VecListThreeDByteIndexed:$list, addrmode6alignNone:$addr,
8250 def VST3LNdWB_fixed_Asm_16 :
8251 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr!",
8252 (ins VecListThreeDHWordIndexed:$list, addrmode6alignNone:$addr,
8254 def VST3LNdWB_fixed_Asm_32 :
8255 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr!",
8256 (ins VecListThreeDWordIndexed:$list, addrmode6alignNone:$addr,
8258 def VST3LNqWB_fixed_Asm_16 :
8259 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr!",
8260 (ins VecListThreeQHWordIndexed:$list, addrmode6alignNone:$addr,
8262 def VST3LNqWB_fixed_Asm_32 :
8263 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr!",
8264 (ins VecListThreeQWordIndexed:$list, addrmode6alignNone:$addr,
8266 def VST3LNdWB_register_Asm_8 :
8267 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr, $Rm",
8268 (ins VecListThreeDByteIndexed:$list, addrmode6alignNone:$addr,
8269 rGPR:$Rm, pred:$p)>;
8270 def VST3LNdWB_register_Asm_16 :
8271 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr, $Rm",
8272 (ins VecListThreeDHWordIndexed:$list,
8273 addrmode6alignNone:$addr, rGPR:$Rm, pred:$p)>;
8274 def VST3LNdWB_register_Asm_32 :
8275 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr, $Rm",
8276 (ins VecListThreeDWordIndexed:$list, addrmode6alignNone:$addr,
8277 rGPR:$Rm, pred:$p)>;
8278 def VST3LNqWB_register_Asm_16 :
8279 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr, $Rm",
8280 (ins VecListThreeQHWordIndexed:$list,
8281 addrmode6alignNone:$addr, rGPR:$Rm, pred:$p)>;
8282 def VST3LNqWB_register_Asm_32 :
8283 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr, $Rm",
8284 (ins VecListThreeQWordIndexed:$list, addrmode6alignNone:$addr,
8285 rGPR:$Rm, pred:$p)>;
8288 // VST3 multiple structure pseudo-instructions. These need special handling for
8289 // the vector operands that the normal instructions don't yet model.
8290 // FIXME: Remove these when the register classes and instructions are updated.
8291 def VST3dAsm_8 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr",
8292 (ins VecListThreeD:$list, addrmode6align64:$addr, pred:$p)>;
8293 def VST3dAsm_16 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr",
8294 (ins VecListThreeD:$list, addrmode6align64:$addr, pred:$p)>;
8295 def VST3dAsm_32 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr",
8296 (ins VecListThreeD:$list, addrmode6align64:$addr, pred:$p)>;
8297 def VST3qAsm_8 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr",
8298 (ins VecListThreeQ:$list, addrmode6align64:$addr, pred:$p)>;
8299 def VST3qAsm_16 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr",
8300 (ins VecListThreeQ:$list, addrmode6align64:$addr, pred:$p)>;
8301 def VST3qAsm_32 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr",
8302 (ins VecListThreeQ:$list, addrmode6align64:$addr, pred:$p)>;
8304 def VST3dWB_fixed_Asm_8 :
8305 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr!",
8306 (ins VecListThreeD:$list, addrmode6align64:$addr, pred:$p)>;
8307 def VST3dWB_fixed_Asm_16 :
8308 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr!",
8309 (ins VecListThreeD:$list, addrmode6align64:$addr, pred:$p)>;
8310 def VST3dWB_fixed_Asm_32 :
8311 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr!",
8312 (ins VecListThreeD:$list, addrmode6align64:$addr, pred:$p)>;
8313 def VST3qWB_fixed_Asm_8 :
8314 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr!",
8315 (ins VecListThreeQ:$list, addrmode6align64:$addr, pred:$p)>;
8316 def VST3qWB_fixed_Asm_16 :
8317 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr!",
8318 (ins VecListThreeQ:$list, addrmode6align64:$addr, pred:$p)>;
8319 def VST3qWB_fixed_Asm_32 :
8320 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr!",
8321 (ins VecListThreeQ:$list, addrmode6align64:$addr, pred:$p)>;
8322 def VST3dWB_register_Asm_8 :
8323 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr, $Rm",
8324 (ins VecListThreeD:$list, addrmode6align64:$addr,
8325 rGPR:$Rm, pred:$p)>;
8326 def VST3dWB_register_Asm_16 :
8327 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr, $Rm",
8328 (ins VecListThreeD:$list, addrmode6align64:$addr,
8329 rGPR:$Rm, pred:$p)>;
8330 def VST3dWB_register_Asm_32 :
8331 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr, $Rm",
8332 (ins VecListThreeD:$list, addrmode6align64:$addr,
8333 rGPR:$Rm, pred:$p)>;
8334 def VST3qWB_register_Asm_8 :
8335 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr, $Rm",
8336 (ins VecListThreeQ:$list, addrmode6align64:$addr,
8337 rGPR:$Rm, pred:$p)>;
8338 def VST3qWB_register_Asm_16 :
8339 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr, $Rm",
8340 (ins VecListThreeQ:$list, addrmode6align64:$addr,
8341 rGPR:$Rm, pred:$p)>;
8342 def VST3qWB_register_Asm_32 :
8343 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr, $Rm",
8344 (ins VecListThreeQ:$list, addrmode6align64:$addr,
8345 rGPR:$Rm, pred:$p)>;
8347 // VLD4 all-lanes pseudo-instructions. These need special handling for
8348 // the lane index that an InstAlias can't handle, so we use these instead.
8349 def VLD4DUPdAsm_8 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr",
8350 (ins VecListFourDAllLanes:$list, addrmode6dupalign32:$addr,
8352 def VLD4DUPdAsm_16: NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
8353 (ins VecListFourDAllLanes:$list, addrmode6dupalign64:$addr,
8355 def VLD4DUPdAsm_32: NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
8356 (ins VecListFourDAllLanes:$list, addrmode6dupalign64or128:$addr,
8358 def VLD4DUPqAsm_8 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr",
8359 (ins VecListFourQAllLanes:$list, addrmode6dupalign32:$addr,
8361 def VLD4DUPqAsm_16: NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
8362 (ins VecListFourQAllLanes:$list, addrmode6dupalign64:$addr,
8364 def VLD4DUPqAsm_32: NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
8365 (ins VecListFourQAllLanes:$list, addrmode6dupalign64or128:$addr,
8368 def VLD4DUPdWB_fixed_Asm_8 :
8369 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr!",
8370 (ins VecListFourDAllLanes:$list, addrmode6dupalign32:$addr,
8372 def VLD4DUPdWB_fixed_Asm_16 :
8373 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
8374 (ins VecListFourDAllLanes:$list, addrmode6dupalign64:$addr,
8376 def VLD4DUPdWB_fixed_Asm_32 :
8377 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
8378 (ins VecListFourDAllLanes:$list, addrmode6dupalign64or128:$addr,
8380 def VLD4DUPqWB_fixed_Asm_8 :
8381 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr!",
8382 (ins VecListFourQAllLanes:$list, addrmode6dupalign32:$addr,
8384 def VLD4DUPqWB_fixed_Asm_16 :
8385 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
8386 (ins VecListFourQAllLanes:$list, addrmode6dupalign64:$addr,
8388 def VLD4DUPqWB_fixed_Asm_32 :
8389 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
8390 (ins VecListFourQAllLanes:$list, addrmode6dupalign64or128:$addr,
8392 def VLD4DUPdWB_register_Asm_8 :
8393 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr, $Rm",
8394 (ins VecListFourDAllLanes:$list, addrmode6dupalign32:$addr,
8395 rGPR:$Rm, pred:$p)>;
8396 def VLD4DUPdWB_register_Asm_16 :
8397 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
8398 (ins VecListFourDAllLanes:$list, addrmode6dupalign64:$addr,
8399 rGPR:$Rm, pred:$p)>;
8400 def VLD4DUPdWB_register_Asm_32 :
8401 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
8402 (ins VecListFourDAllLanes:$list,
8403 addrmode6dupalign64or128:$addr, rGPR:$Rm, pred:$p)>;
8404 def VLD4DUPqWB_register_Asm_8 :
8405 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr, $Rm",
8406 (ins VecListFourQAllLanes:$list, addrmode6dupalign32:$addr,
8407 rGPR:$Rm, pred:$p)>;
8408 def VLD4DUPqWB_register_Asm_16 :
8409 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
8410 (ins VecListFourQAllLanes:$list, addrmode6dupalign64:$addr,
8411 rGPR:$Rm, pred:$p)>;
8412 def VLD4DUPqWB_register_Asm_32 :
8413 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
8414 (ins VecListFourQAllLanes:$list,
8415 addrmode6dupalign64or128:$addr, rGPR:$Rm, pred:$p)>;
8418 // VLD4 single-lane pseudo-instructions. These need special handling for
8419 // the lane index that an InstAlias can't handle, so we use these instead.
8420 def VLD4LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr",
8421 (ins VecListFourDByteIndexed:$list, addrmode6align32:$addr,
8423 def VLD4LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
8424 (ins VecListFourDHWordIndexed:$list, addrmode6align64:$addr,
8426 def VLD4LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
8427 (ins VecListFourDWordIndexed:$list, addrmode6align64or128:$addr,
8429 def VLD4LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
8430 (ins VecListFourQHWordIndexed:$list, addrmode6align64:$addr,
8432 def VLD4LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
8433 (ins VecListFourQWordIndexed:$list, addrmode6align64or128:$addr,
8436 def VLD4LNdWB_fixed_Asm_8 :
8437 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr!",
8438 (ins VecListFourDByteIndexed:$list, addrmode6align32:$addr,
8440 def VLD4LNdWB_fixed_Asm_16 :
8441 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
8442 (ins VecListFourDHWordIndexed:$list, addrmode6align64:$addr,
8444 def VLD4LNdWB_fixed_Asm_32 :
8445 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
8446 (ins VecListFourDWordIndexed:$list, addrmode6align64or128:$addr,
8448 def VLD4LNqWB_fixed_Asm_16 :
8449 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
8450 (ins VecListFourQHWordIndexed:$list, addrmode6align64:$addr,
8452 def VLD4LNqWB_fixed_Asm_32 :
8453 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
8454 (ins VecListFourQWordIndexed:$list, addrmode6align64or128:$addr,
8456 def VLD4LNdWB_register_Asm_8 :
8457 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr, $Rm",
8458 (ins VecListFourDByteIndexed:$list, addrmode6align32:$addr,
8459 rGPR:$Rm, pred:$p)>;
8460 def VLD4LNdWB_register_Asm_16 :
8461 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
8462 (ins VecListFourDHWordIndexed:$list, addrmode6align64:$addr,
8463 rGPR:$Rm, pred:$p)>;
8464 def VLD4LNdWB_register_Asm_32 :
8465 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
8466 (ins VecListFourDWordIndexed:$list,
8467 addrmode6align64or128:$addr, rGPR:$Rm, pred:$p)>;
8468 def VLD4LNqWB_register_Asm_16 :
8469 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
8470 (ins VecListFourQHWordIndexed:$list, addrmode6align64:$addr,
8471 rGPR:$Rm, pred:$p)>;
8472 def VLD4LNqWB_register_Asm_32 :
8473 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
8474 (ins VecListFourQWordIndexed:$list,
8475 addrmode6align64or128:$addr, rGPR:$Rm, pred:$p)>;
8479 // VLD4 multiple structure pseudo-instructions. These need special handling for
8480 // the vector operands that the normal instructions don't yet model.
8481 // FIXME: Remove these when the register classes and instructions are updated.
8482 def VLD4dAsm_8 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr",
8483 (ins VecListFourD:$list, addrmode6align64or128or256:$addr,
8485 def VLD4dAsm_16 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
8486 (ins VecListFourD:$list, addrmode6align64or128or256:$addr,
8488 def VLD4dAsm_32 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
8489 (ins VecListFourD:$list, addrmode6align64or128or256:$addr,
8491 def VLD4qAsm_8 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr",
8492 (ins VecListFourQ:$list, addrmode6align64or128or256:$addr,
8494 def VLD4qAsm_16 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
8495 (ins VecListFourQ:$list, addrmode6align64or128or256:$addr,
8497 def VLD4qAsm_32 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
8498 (ins VecListFourQ:$list, addrmode6align64or128or256:$addr,
8501 def VLD4dWB_fixed_Asm_8 :
8502 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr!",
8503 (ins VecListFourD:$list, addrmode6align64or128or256:$addr,
8505 def VLD4dWB_fixed_Asm_16 :
8506 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
8507 (ins VecListFourD:$list, addrmode6align64or128or256:$addr,
8509 def VLD4dWB_fixed_Asm_32 :
8510 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
8511 (ins VecListFourD:$list, addrmode6align64or128or256:$addr,
8513 def VLD4qWB_fixed_Asm_8 :
8514 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr!",
8515 (ins VecListFourQ:$list, addrmode6align64or128or256:$addr,
8517 def VLD4qWB_fixed_Asm_16 :
8518 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
8519 (ins VecListFourQ:$list, addrmode6align64or128or256:$addr,
8521 def VLD4qWB_fixed_Asm_32 :
8522 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
8523 (ins VecListFourQ:$list, addrmode6align64or128or256:$addr,
8525 def VLD4dWB_register_Asm_8 :
8526 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr, $Rm",
8527 (ins VecListFourD:$list, addrmode6align64or128or256:$addr,
8528 rGPR:$Rm, pred:$p)>;
8529 def VLD4dWB_register_Asm_16 :
8530 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
8531 (ins VecListFourD:$list, addrmode6align64or128or256:$addr,
8532 rGPR:$Rm, pred:$p)>;
8533 def VLD4dWB_register_Asm_32 :
8534 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
8535 (ins VecListFourD:$list, addrmode6align64or128or256:$addr,
8536 rGPR:$Rm, pred:$p)>;
8537 def VLD4qWB_register_Asm_8 :
8538 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr, $Rm",
8539 (ins VecListFourQ:$list, addrmode6align64or128or256:$addr,
8540 rGPR:$Rm, pred:$p)>;
8541 def VLD4qWB_register_Asm_16 :
8542 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
8543 (ins VecListFourQ:$list, addrmode6align64or128or256:$addr,
8544 rGPR:$Rm, pred:$p)>;
8545 def VLD4qWB_register_Asm_32 :
8546 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
8547 (ins VecListFourQ:$list, addrmode6align64or128or256:$addr,
8548 rGPR:$Rm, pred:$p)>;
8550 // VST4 single-lane pseudo-instructions. These need special handling for
8551 // the lane index that an InstAlias can't handle, so we use these instead.
8552 def VST4LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr",
8553 (ins VecListFourDByteIndexed:$list, addrmode6align32:$addr,
8555 def VST4LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr",
8556 (ins VecListFourDHWordIndexed:$list, addrmode6align64:$addr,
8558 def VST4LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr",
8559 (ins VecListFourDWordIndexed:$list, addrmode6align64or128:$addr,
8561 def VST4LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr",
8562 (ins VecListFourQHWordIndexed:$list, addrmode6align64:$addr,
8564 def VST4LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr",
8565 (ins VecListFourQWordIndexed:$list, addrmode6align64or128:$addr,
8568 def VST4LNdWB_fixed_Asm_8 :
8569 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr!",
8570 (ins VecListFourDByteIndexed:$list, addrmode6align32:$addr,
8572 def VST4LNdWB_fixed_Asm_16 :
8573 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr!",
8574 (ins VecListFourDHWordIndexed:$list, addrmode6align64:$addr,
8576 def VST4LNdWB_fixed_Asm_32 :
8577 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr!",
8578 (ins VecListFourDWordIndexed:$list, addrmode6align64or128:$addr,
8580 def VST4LNqWB_fixed_Asm_16 :
8581 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr!",
8582 (ins VecListFourQHWordIndexed:$list, addrmode6align64:$addr,
8584 def VST4LNqWB_fixed_Asm_32 :
8585 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr!",
8586 (ins VecListFourQWordIndexed:$list, addrmode6align64or128:$addr,
8588 def VST4LNdWB_register_Asm_8 :
8589 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr, $Rm",
8590 (ins VecListFourDByteIndexed:$list, addrmode6align32:$addr,
8591 rGPR:$Rm, pred:$p)>;
8592 def VST4LNdWB_register_Asm_16 :
8593 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr, $Rm",
8594 (ins VecListFourDHWordIndexed:$list, addrmode6align64:$addr,
8595 rGPR:$Rm, pred:$p)>;
8596 def VST4LNdWB_register_Asm_32 :
8597 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr, $Rm",
8598 (ins VecListFourDWordIndexed:$list,
8599 addrmode6align64or128:$addr, rGPR:$Rm, pred:$p)>;
8600 def VST4LNqWB_register_Asm_16 :
8601 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr, $Rm",
8602 (ins VecListFourQHWordIndexed:$list, addrmode6align64:$addr,
8603 rGPR:$Rm, pred:$p)>;
8604 def VST4LNqWB_register_Asm_32 :
8605 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr, $Rm",
8606 (ins VecListFourQWordIndexed:$list,
8607 addrmode6align64or128:$addr, rGPR:$Rm, pred:$p)>;
8610 // VST4 multiple structure pseudo-instructions. These need special handling for
8611 // the vector operands that the normal instructions don't yet model.
8612 // FIXME: Remove these when the register classes and instructions are updated.
8613 def VST4dAsm_8 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr",
8614 (ins VecListFourD:$list, addrmode6align64or128or256:$addr,
8616 def VST4dAsm_16 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr",
8617 (ins VecListFourD:$list, addrmode6align64or128or256:$addr,
8619 def VST4dAsm_32 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr",
8620 (ins VecListFourD:$list, addrmode6align64or128or256:$addr,
8622 def VST4qAsm_8 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr",
8623 (ins VecListFourQ:$list, addrmode6align64or128or256:$addr,
8625 def VST4qAsm_16 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr",
8626 (ins VecListFourQ:$list, addrmode6align64or128or256:$addr,
8628 def VST4qAsm_32 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr",
8629 (ins VecListFourQ:$list, addrmode6align64or128or256:$addr,
8632 def VST4dWB_fixed_Asm_8 :
8633 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr!",
8634 (ins VecListFourD:$list, addrmode6align64or128or256:$addr,
8636 def VST4dWB_fixed_Asm_16 :
8637 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr!",
8638 (ins VecListFourD:$list, addrmode6align64or128or256:$addr,
8640 def VST4dWB_fixed_Asm_32 :
8641 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr!",
8642 (ins VecListFourD:$list, addrmode6align64or128or256:$addr,
8644 def VST4qWB_fixed_Asm_8 :
8645 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr!",
8646 (ins VecListFourQ:$list, addrmode6align64or128or256:$addr,
8648 def VST4qWB_fixed_Asm_16 :
8649 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr!",
8650 (ins VecListFourQ:$list, addrmode6align64or128or256:$addr,
8652 def VST4qWB_fixed_Asm_32 :
8653 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr!",
8654 (ins VecListFourQ:$list, addrmode6align64or128or256:$addr,
8656 def VST4dWB_register_Asm_8 :
8657 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr, $Rm",
8658 (ins VecListFourD:$list, addrmode6align64or128or256:$addr,
8659 rGPR:$Rm, pred:$p)>;
8660 def VST4dWB_register_Asm_16 :
8661 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr, $Rm",
8662 (ins VecListFourD:$list, addrmode6align64or128or256:$addr,
8663 rGPR:$Rm, pred:$p)>;
8664 def VST4dWB_register_Asm_32 :
8665 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr, $Rm",
8666 (ins VecListFourD:$list, addrmode6align64or128or256:$addr,
8667 rGPR:$Rm, pred:$p)>;
8668 def VST4qWB_register_Asm_8 :
8669 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr, $Rm",
8670 (ins VecListFourQ:$list, addrmode6align64or128or256:$addr,
8671 rGPR:$Rm, pred:$p)>;
8672 def VST4qWB_register_Asm_16 :
8673 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr, $Rm",
8674 (ins VecListFourQ:$list, addrmode6align64or128or256:$addr,
8675 rGPR:$Rm, pred:$p)>;
8676 def VST4qWB_register_Asm_32 :
8677 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr, $Rm",
8678 (ins VecListFourQ:$list, addrmode6align64or128or256:$addr,
8679 rGPR:$Rm, pred:$p)>;
8681 // VMOV/VMVN takes an optional datatype suffix
8682 defm : NEONDTAnyInstAlias<"vmov${p}", "$Vd, $Vm",
8683 (VORRd DPR:$Vd, DPR:$Vm, DPR:$Vm, pred:$p)>;
8684 defm : NEONDTAnyInstAlias<"vmov${p}", "$Vd, $Vm",
8685 (VORRq QPR:$Vd, QPR:$Vm, QPR:$Vm, pred:$p)>;
8687 defm : NEONDTAnyInstAlias<"vmvn${p}", "$Vd, $Vm",
8688 (VMVNd DPR:$Vd, DPR:$Vm, pred:$p)>;
8689 defm : NEONDTAnyInstAlias<"vmvn${p}", "$Vd, $Vm",
8690 (VMVNq QPR:$Vd, QPR:$Vm, pred:$p)>;
8692 // VCLT (register) is an assembler alias for VCGT w/ the operands reversed.
8693 // D-register versions.
8694 def : NEONInstAlias<"vcle${p}.s8 $Dd, $Dn, $Dm",
8695 (VCGEsv8i8 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
8696 def : NEONInstAlias<"vcle${p}.s16 $Dd, $Dn, $Dm",
8697 (VCGEsv4i16 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
8698 def : NEONInstAlias<"vcle${p}.s32 $Dd, $Dn, $Dm",
8699 (VCGEsv2i32 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
8700 def : NEONInstAlias<"vcle${p}.u8 $Dd, $Dn, $Dm",
8701 (VCGEuv8i8 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
8702 def : NEONInstAlias<"vcle${p}.u16 $Dd, $Dn, $Dm",
8703 (VCGEuv4i16 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
8704 def : NEONInstAlias<"vcle${p}.u32 $Dd, $Dn, $Dm",
8705 (VCGEuv2i32 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
8706 def : NEONInstAlias<"vcle${p}.f32 $Dd, $Dn, $Dm",
8707 (VCGEfd DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
8708 let Predicates = [HasNEON, HasFullFP16] in
8709 def : NEONInstAlias<"vcle${p}.f16 $Dd, $Dn, $Dm",
8710 (VCGEhd DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
8711 // Q-register versions.
8712 def : NEONInstAlias<"vcle${p}.s8 $Qd, $Qn, $Qm",
8713 (VCGEsv16i8 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
8714 def : NEONInstAlias<"vcle${p}.s16 $Qd, $Qn, $Qm",
8715 (VCGEsv8i16 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
8716 def : NEONInstAlias<"vcle${p}.s32 $Qd, $Qn, $Qm",
8717 (VCGEsv4i32 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
8718 def : NEONInstAlias<"vcle${p}.u8 $Qd, $Qn, $Qm",
8719 (VCGEuv16i8 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
8720 def : NEONInstAlias<"vcle${p}.u16 $Qd, $Qn, $Qm",
8721 (VCGEuv8i16 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
8722 def : NEONInstAlias<"vcle${p}.u32 $Qd, $Qn, $Qm",
8723 (VCGEuv4i32 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
8724 def : NEONInstAlias<"vcle${p}.f32 $Qd, $Qn, $Qm",
8725 (VCGEfq QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
8726 let Predicates = [HasNEON, HasFullFP16] in
8727 def : NEONInstAlias<"vcle${p}.f16 $Qd, $Qn, $Qm",
8728 (VCGEhq QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
8730 // VCLT (register) is an assembler alias for VCGT w/ the operands reversed.
8731 // D-register versions.
8732 def : NEONInstAlias<"vclt${p}.s8 $Dd, $Dn, $Dm",
8733 (VCGTsv8i8 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
8734 def : NEONInstAlias<"vclt${p}.s16 $Dd, $Dn, $Dm",
8735 (VCGTsv4i16 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
8736 def : NEONInstAlias<"vclt${p}.s32 $Dd, $Dn, $Dm",
8737 (VCGTsv2i32 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
8738 def : NEONInstAlias<"vclt${p}.u8 $Dd, $Dn, $Dm",
8739 (VCGTuv8i8 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
8740 def : NEONInstAlias<"vclt${p}.u16 $Dd, $Dn, $Dm",
8741 (VCGTuv4i16 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
8742 def : NEONInstAlias<"vclt${p}.u32 $Dd, $Dn, $Dm",
8743 (VCGTuv2i32 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
8744 def : NEONInstAlias<"vclt${p}.f32 $Dd, $Dn, $Dm",
8745 (VCGTfd DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
8746 let Predicates = [HasNEON, HasFullFP16] in
8747 def : NEONInstAlias<"vclt${p}.f16 $Dd, $Dn, $Dm",
8748 (VCGThd DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
8749 // Q-register versions.
8750 def : NEONInstAlias<"vclt${p}.s8 $Qd, $Qn, $Qm",
8751 (VCGTsv16i8 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
8752 def : NEONInstAlias<"vclt${p}.s16 $Qd, $Qn, $Qm",
8753 (VCGTsv8i16 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
8754 def : NEONInstAlias<"vclt${p}.s32 $Qd, $Qn, $Qm",
8755 (VCGTsv4i32 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
8756 def : NEONInstAlias<"vclt${p}.u8 $Qd, $Qn, $Qm",
8757 (VCGTuv16i8 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
8758 def : NEONInstAlias<"vclt${p}.u16 $Qd, $Qn, $Qm",
8759 (VCGTuv8i16 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
8760 def : NEONInstAlias<"vclt${p}.u32 $Qd, $Qn, $Qm",
8761 (VCGTuv4i32 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
8762 def : NEONInstAlias<"vclt${p}.f32 $Qd, $Qn, $Qm",
8763 (VCGTfq QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
8764 let Predicates = [HasNEON, HasFullFP16] in
8765 def : NEONInstAlias<"vclt${p}.f16 $Qd, $Qn, $Qm",
8766 (VCGThq QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
8768 // VSWP allows, but does not require, a type suffix.
8769 defm : NEONDTAnyInstAlias<"vswp${p}", "$Vd, $Vm",
8770 (VSWPd DPR:$Vd, DPR:$Vm, pred:$p)>;
8771 defm : NEONDTAnyInstAlias<"vswp${p}", "$Vd, $Vm",
8772 (VSWPq QPR:$Vd, QPR:$Vm, pred:$p)>;
8774 // VBIF, VBIT, and VBSL allow, but do not require, a type suffix.
8775 defm : NEONDTAnyInstAlias<"vbif${p}", "$Vd, $Vn, $Vm",
8776 (VBIFd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
8777 defm : NEONDTAnyInstAlias<"vbit${p}", "$Vd, $Vn, $Vm",
8778 (VBITd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
8779 defm : NEONDTAnyInstAlias<"vbsl${p}", "$Vd, $Vn, $Vm",
8780 (VBSLd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
8781 defm : NEONDTAnyInstAlias<"vbif${p}", "$Vd, $Vn, $Vm",
8782 (VBIFq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
8783 defm : NEONDTAnyInstAlias<"vbit${p}", "$Vd, $Vn, $Vm",
8784 (VBITq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
8785 defm : NEONDTAnyInstAlias<"vbsl${p}", "$Vd, $Vn, $Vm",
8786 (VBSLq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
8788 // "vmov Rd, #-imm" can be handled via "vmvn".
8789 def : NEONInstAlias<"vmov${p}.i32 $Vd, $imm",
8790 (VMVNv2i32 DPR:$Vd, nImmVMOVI32Neg:$imm, pred:$p)>;
8791 def : NEONInstAlias<"vmov${p}.i32 $Vd, $imm",
8792 (VMVNv4i32 QPR:$Vd, nImmVMOVI32Neg:$imm, pred:$p)>;
8793 def : NEONInstAlias<"vmvn${p}.i32 $Vd, $imm",
8794 (VMOVv2i32 DPR:$Vd, nImmVMOVI32Neg:$imm, pred:$p)>;
8795 def : NEONInstAlias<"vmvn${p}.i32 $Vd, $imm",
8796 (VMOVv4i32 QPR:$Vd, nImmVMOVI32Neg:$imm, pred:$p)>;
8798 // 'gas' compatibility aliases for quad-word instructions. Strictly speaking,
8799 // these should restrict to just the Q register variants, but the register
8800 // classes are enough to match correctly regardless, so we keep it simple
8801 // and just use MnemonicAlias.
8802 def : NEONMnemonicAlias<"vbicq", "vbic">;
8803 def : NEONMnemonicAlias<"vandq", "vand">;
8804 def : NEONMnemonicAlias<"veorq", "veor">;
8805 def : NEONMnemonicAlias<"vorrq", "vorr">;
8807 def : NEONMnemonicAlias<"vmovq", "vmov">;
8808 def : NEONMnemonicAlias<"vmvnq", "vmvn">;
8809 // Explicit versions for floating point so that the FPImm variants get
8810 // handled early. The parser gets confused otherwise.
8811 def : NEONMnemonicAlias<"vmovq.f32", "vmov.f32">;
8812 def : NEONMnemonicAlias<"vmovq.f64", "vmov.f64">;
8814 def : NEONMnemonicAlias<"vaddq", "vadd">;
8815 def : NEONMnemonicAlias<"vsubq", "vsub">;
8817 def : NEONMnemonicAlias<"vminq", "vmin">;
8818 def : NEONMnemonicAlias<"vmaxq", "vmax">;
8820 def : NEONMnemonicAlias<"vmulq", "vmul">;
8822 def : NEONMnemonicAlias<"vabsq", "vabs">;
8824 def : NEONMnemonicAlias<"vshlq", "vshl">;
8825 def : NEONMnemonicAlias<"vshrq", "vshr">;
8827 def : NEONMnemonicAlias<"vcvtq", "vcvt">;
8829 def : NEONMnemonicAlias<"vcleq", "vcle">;
8830 def : NEONMnemonicAlias<"vceqq", "vceq">;
8832 def : NEONMnemonicAlias<"vzipq", "vzip">;
8833 def : NEONMnemonicAlias<"vswpq", "vswp">;
8835 def : NEONMnemonicAlias<"vrecpeq.f32", "vrecpe.f32">;
8836 def : NEONMnemonicAlias<"vrecpeq.u32", "vrecpe.u32">;
8839 // Alias for loading floating point immediates that aren't representable
8840 // using the vmov.f32 encoding but the bitpattern is representable using
8841 // the .i32 encoding.
8842 def : NEONInstAlias<"vmov${p}.f32 $Vd, $imm",
8843 (VMOVv4i32 QPR:$Vd, nImmVMOVI32:$imm, pred:$p)>;
8844 def : NEONInstAlias<"vmov${p}.f32 $Vd, $imm",
8845 (VMOVv2i32 DPR:$Vd, nImmVMOVI32:$imm, pred:$p)>;