1 //===-- ARMInstrNEON.td - NEON support for ARM -------------*- tablegen -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file describes the ARM NEON instruction set.
12 //===----------------------------------------------------------------------===//
15 //===----------------------------------------------------------------------===//
16 // NEON-specific Operands.
17 //===----------------------------------------------------------------------===//
18 def nModImm : Operand<i32> {
19 let PrintMethod = "printNEONModImmOperand";
22 def nImmSplatI8AsmOperand : AsmOperandClass { let Name = "NEONi8splat"; }
23 def nImmSplatI8 : Operand<i32> {
24 let PrintMethod = "printNEONModImmOperand";
25 let ParserMatchClass = nImmSplatI8AsmOperand;
27 def nImmSplatI16AsmOperand : AsmOperandClass { let Name = "NEONi16splat"; }
28 def nImmSplatI16 : Operand<i32> {
29 let PrintMethod = "printNEONModImmOperand";
30 let ParserMatchClass = nImmSplatI16AsmOperand;
32 def nImmSplatI32AsmOperand : AsmOperandClass { let Name = "NEONi32splat"; }
33 def nImmSplatI32 : Operand<i32> {
34 let PrintMethod = "printNEONModImmOperand";
35 let ParserMatchClass = nImmSplatI32AsmOperand;
37 def nImmSplatNotI16AsmOperand : AsmOperandClass { let Name = "NEONi16splatNot"; }
38 def nImmSplatNotI16 : Operand<i32> {
39 let ParserMatchClass = nImmSplatNotI16AsmOperand;
41 def nImmSplatNotI32AsmOperand : AsmOperandClass { let Name = "NEONi32splatNot"; }
42 def nImmSplatNotI32 : Operand<i32> {
43 let ParserMatchClass = nImmSplatNotI32AsmOperand;
45 def nImmVMOVI32AsmOperand : AsmOperandClass { let Name = "NEONi32vmov"; }
46 def nImmVMOVI32 : Operand<i32> {
47 let PrintMethod = "printNEONModImmOperand";
48 let ParserMatchClass = nImmVMOVI32AsmOperand;
51 def nImmVMOVI16AsmOperandByteReplicate :
53 let Name = "NEONi16vmovByteReplicate";
54 let PredicateMethod = "isNEONi16ByteReplicate";
55 let RenderMethod = "addNEONvmovByteReplicateOperands";
57 def nImmVMOVI32AsmOperandByteReplicate :
59 let Name = "NEONi32vmovByteReplicate";
60 let PredicateMethod = "isNEONi32ByteReplicate";
61 let RenderMethod = "addNEONvmovByteReplicateOperands";
63 def nImmVMVNI16AsmOperandByteReplicate :
65 let Name = "NEONi16invByteReplicate";
66 let PredicateMethod = "isNEONi16ByteReplicate";
67 let RenderMethod = "addNEONinvByteReplicateOperands";
69 def nImmVMVNI32AsmOperandByteReplicate :
71 let Name = "NEONi32invByteReplicate";
72 let PredicateMethod = "isNEONi32ByteReplicate";
73 let RenderMethod = "addNEONinvByteReplicateOperands";
76 def nImmVMOVI16ByteReplicate : Operand<i32> {
77 let PrintMethod = "printNEONModImmOperand";
78 let ParserMatchClass = nImmVMOVI16AsmOperandByteReplicate;
80 def nImmVMOVI32ByteReplicate : Operand<i32> {
81 let PrintMethod = "printNEONModImmOperand";
82 let ParserMatchClass = nImmVMOVI32AsmOperandByteReplicate;
84 def nImmVMVNI16ByteReplicate : Operand<i32> {
85 let PrintMethod = "printNEONModImmOperand";
86 let ParserMatchClass = nImmVMVNI16AsmOperandByteReplicate;
88 def nImmVMVNI32ByteReplicate : Operand<i32> {
89 let PrintMethod = "printNEONModImmOperand";
90 let ParserMatchClass = nImmVMVNI32AsmOperandByteReplicate;
93 def nImmVMOVI32NegAsmOperand : AsmOperandClass { let Name = "NEONi32vmovNeg"; }
94 def nImmVMOVI32Neg : Operand<i32> {
95 let PrintMethod = "printNEONModImmOperand";
96 let ParserMatchClass = nImmVMOVI32NegAsmOperand;
98 def nImmVMOVF32 : Operand<i32> {
99 let PrintMethod = "printFPImmOperand";
100 let ParserMatchClass = FPImmOperand;
102 def nImmSplatI64AsmOperand : AsmOperandClass { let Name = "NEONi64splat"; }
103 def nImmSplatI64 : Operand<i32> {
104 let PrintMethod = "printNEONModImmOperand";
105 let ParserMatchClass = nImmSplatI64AsmOperand;
108 def VectorIndex8Operand : AsmOperandClass { let Name = "VectorIndex8"; }
109 def VectorIndex16Operand : AsmOperandClass { let Name = "VectorIndex16"; }
110 def VectorIndex32Operand : AsmOperandClass { let Name = "VectorIndex32"; }
111 def VectorIndex64Operand : AsmOperandClass { let Name = "VectorIndex64"; }
112 def VectorIndex8 : Operand<i32>, ImmLeaf<i32, [{
113 return ((uint64_t)Imm) < 8;
115 let ParserMatchClass = VectorIndex8Operand;
116 let PrintMethod = "printVectorIndex";
117 let MIOperandInfo = (ops i32imm);
119 def VectorIndex16 : Operand<i32>, ImmLeaf<i32, [{
120 return ((uint64_t)Imm) < 4;
122 let ParserMatchClass = VectorIndex16Operand;
123 let PrintMethod = "printVectorIndex";
124 let MIOperandInfo = (ops i32imm);
126 def VectorIndex32 : Operand<i32>, ImmLeaf<i32, [{
127 return ((uint64_t)Imm) < 2;
129 let ParserMatchClass = VectorIndex32Operand;
130 let PrintMethod = "printVectorIndex";
131 let MIOperandInfo = (ops i32imm);
133 def VectorIndex64 : Operand<i32>, ImmLeaf<i32, [{
134 return ((uint64_t)Imm) < 1;
136 let ParserMatchClass = VectorIndex64Operand;
137 let PrintMethod = "printVectorIndex";
138 let MIOperandInfo = (ops i32imm);
141 // Register list of one D register.
142 def VecListOneDAsmOperand : AsmOperandClass {
143 let Name = "VecListOneD";
144 let ParserMethod = "parseVectorList";
145 let RenderMethod = "addVecListOperands";
147 def VecListOneD : RegisterOperand<DPR, "printVectorListOne"> {
148 let ParserMatchClass = VecListOneDAsmOperand;
150 // Register list of two sequential D registers.
151 def VecListDPairAsmOperand : AsmOperandClass {
152 let Name = "VecListDPair";
153 let ParserMethod = "parseVectorList";
154 let RenderMethod = "addVecListOperands";
156 def VecListDPair : RegisterOperand<DPair, "printVectorListTwo"> {
157 let ParserMatchClass = VecListDPairAsmOperand;
159 // Register list of three sequential D registers.
160 def VecListThreeDAsmOperand : AsmOperandClass {
161 let Name = "VecListThreeD";
162 let ParserMethod = "parseVectorList";
163 let RenderMethod = "addVecListOperands";
165 def VecListThreeD : RegisterOperand<DPR, "printVectorListThree"> {
166 let ParserMatchClass = VecListThreeDAsmOperand;
168 // Register list of four sequential D registers.
169 def VecListFourDAsmOperand : AsmOperandClass {
170 let Name = "VecListFourD";
171 let ParserMethod = "parseVectorList";
172 let RenderMethod = "addVecListOperands";
174 def VecListFourD : RegisterOperand<DPR, "printVectorListFour"> {
175 let ParserMatchClass = VecListFourDAsmOperand;
177 // Register list of two D registers spaced by 2 (two sequential Q registers).
178 def VecListDPairSpacedAsmOperand : AsmOperandClass {
179 let Name = "VecListDPairSpaced";
180 let ParserMethod = "parseVectorList";
181 let RenderMethod = "addVecListOperands";
183 def VecListDPairSpaced : RegisterOperand<DPair, "printVectorListTwoSpaced"> {
184 let ParserMatchClass = VecListDPairSpacedAsmOperand;
186 // Register list of three D registers spaced by 2 (three Q registers).
187 def VecListThreeQAsmOperand : AsmOperandClass {
188 let Name = "VecListThreeQ";
189 let ParserMethod = "parseVectorList";
190 let RenderMethod = "addVecListOperands";
192 def VecListThreeQ : RegisterOperand<DPR, "printVectorListThreeSpaced"> {
193 let ParserMatchClass = VecListThreeQAsmOperand;
195 // Register list of three D registers spaced by 2 (three Q registers).
196 def VecListFourQAsmOperand : AsmOperandClass {
197 let Name = "VecListFourQ";
198 let ParserMethod = "parseVectorList";
199 let RenderMethod = "addVecListOperands";
201 def VecListFourQ : RegisterOperand<DPR, "printVectorListFourSpaced"> {
202 let ParserMatchClass = VecListFourQAsmOperand;
205 // Register list of one D register, with "all lanes" subscripting.
206 def VecListOneDAllLanesAsmOperand : AsmOperandClass {
207 let Name = "VecListOneDAllLanes";
208 let ParserMethod = "parseVectorList";
209 let RenderMethod = "addVecListOperands";
211 def VecListOneDAllLanes : RegisterOperand<DPR, "printVectorListOneAllLanes"> {
212 let ParserMatchClass = VecListOneDAllLanesAsmOperand;
214 // Register list of two D registers, with "all lanes" subscripting.
215 def VecListDPairAllLanesAsmOperand : AsmOperandClass {
216 let Name = "VecListDPairAllLanes";
217 let ParserMethod = "parseVectorList";
218 let RenderMethod = "addVecListOperands";
220 def VecListDPairAllLanes : RegisterOperand<DPair,
221 "printVectorListTwoAllLanes"> {
222 let ParserMatchClass = VecListDPairAllLanesAsmOperand;
224 // Register list of two D registers spaced by 2 (two sequential Q registers).
225 def VecListDPairSpacedAllLanesAsmOperand : AsmOperandClass {
226 let Name = "VecListDPairSpacedAllLanes";
227 let ParserMethod = "parseVectorList";
228 let RenderMethod = "addVecListOperands";
230 def VecListDPairSpacedAllLanes : RegisterOperand<DPair,
231 "printVectorListTwoSpacedAllLanes"> {
232 let ParserMatchClass = VecListDPairSpacedAllLanesAsmOperand;
234 // Register list of three D registers, with "all lanes" subscripting.
235 def VecListThreeDAllLanesAsmOperand : AsmOperandClass {
236 let Name = "VecListThreeDAllLanes";
237 let ParserMethod = "parseVectorList";
238 let RenderMethod = "addVecListOperands";
240 def VecListThreeDAllLanes : RegisterOperand<DPR,
241 "printVectorListThreeAllLanes"> {
242 let ParserMatchClass = VecListThreeDAllLanesAsmOperand;
244 // Register list of three D registers spaced by 2 (three sequential Q regs).
245 def VecListThreeQAllLanesAsmOperand : AsmOperandClass {
246 let Name = "VecListThreeQAllLanes";
247 let ParserMethod = "parseVectorList";
248 let RenderMethod = "addVecListOperands";
250 def VecListThreeQAllLanes : RegisterOperand<DPR,
251 "printVectorListThreeSpacedAllLanes"> {
252 let ParserMatchClass = VecListThreeQAllLanesAsmOperand;
254 // Register list of four D registers, with "all lanes" subscripting.
255 def VecListFourDAllLanesAsmOperand : AsmOperandClass {
256 let Name = "VecListFourDAllLanes";
257 let ParserMethod = "parseVectorList";
258 let RenderMethod = "addVecListOperands";
260 def VecListFourDAllLanes : RegisterOperand<DPR, "printVectorListFourAllLanes"> {
261 let ParserMatchClass = VecListFourDAllLanesAsmOperand;
263 // Register list of four D registers spaced by 2 (four sequential Q regs).
264 def VecListFourQAllLanesAsmOperand : AsmOperandClass {
265 let Name = "VecListFourQAllLanes";
266 let ParserMethod = "parseVectorList";
267 let RenderMethod = "addVecListOperands";
269 def VecListFourQAllLanes : RegisterOperand<DPR,
270 "printVectorListFourSpacedAllLanes"> {
271 let ParserMatchClass = VecListFourQAllLanesAsmOperand;
275 // Register list of one D register, with byte lane subscripting.
276 def VecListOneDByteIndexAsmOperand : AsmOperandClass {
277 let Name = "VecListOneDByteIndexed";
278 let ParserMethod = "parseVectorList";
279 let RenderMethod = "addVecListIndexedOperands";
281 def VecListOneDByteIndexed : Operand<i32> {
282 let ParserMatchClass = VecListOneDByteIndexAsmOperand;
283 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
285 // ...with half-word lane subscripting.
286 def VecListOneDHWordIndexAsmOperand : AsmOperandClass {
287 let Name = "VecListOneDHWordIndexed";
288 let ParserMethod = "parseVectorList";
289 let RenderMethod = "addVecListIndexedOperands";
291 def VecListOneDHWordIndexed : Operand<i32> {
292 let ParserMatchClass = VecListOneDHWordIndexAsmOperand;
293 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
295 // ...with word lane subscripting.
296 def VecListOneDWordIndexAsmOperand : AsmOperandClass {
297 let Name = "VecListOneDWordIndexed";
298 let ParserMethod = "parseVectorList";
299 let RenderMethod = "addVecListIndexedOperands";
301 def VecListOneDWordIndexed : Operand<i32> {
302 let ParserMatchClass = VecListOneDWordIndexAsmOperand;
303 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
306 // Register list of two D registers with byte lane subscripting.
307 def VecListTwoDByteIndexAsmOperand : AsmOperandClass {
308 let Name = "VecListTwoDByteIndexed";
309 let ParserMethod = "parseVectorList";
310 let RenderMethod = "addVecListIndexedOperands";
312 def VecListTwoDByteIndexed : Operand<i32> {
313 let ParserMatchClass = VecListTwoDByteIndexAsmOperand;
314 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
316 // ...with half-word lane subscripting.
317 def VecListTwoDHWordIndexAsmOperand : AsmOperandClass {
318 let Name = "VecListTwoDHWordIndexed";
319 let ParserMethod = "parseVectorList";
320 let RenderMethod = "addVecListIndexedOperands";
322 def VecListTwoDHWordIndexed : Operand<i32> {
323 let ParserMatchClass = VecListTwoDHWordIndexAsmOperand;
324 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
326 // ...with word lane subscripting.
327 def VecListTwoDWordIndexAsmOperand : AsmOperandClass {
328 let Name = "VecListTwoDWordIndexed";
329 let ParserMethod = "parseVectorList";
330 let RenderMethod = "addVecListIndexedOperands";
332 def VecListTwoDWordIndexed : Operand<i32> {
333 let ParserMatchClass = VecListTwoDWordIndexAsmOperand;
334 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
336 // Register list of two Q registers with half-word lane subscripting.
337 def VecListTwoQHWordIndexAsmOperand : AsmOperandClass {
338 let Name = "VecListTwoQHWordIndexed";
339 let ParserMethod = "parseVectorList";
340 let RenderMethod = "addVecListIndexedOperands";
342 def VecListTwoQHWordIndexed : Operand<i32> {
343 let ParserMatchClass = VecListTwoQHWordIndexAsmOperand;
344 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
346 // ...with word lane subscripting.
347 def VecListTwoQWordIndexAsmOperand : AsmOperandClass {
348 let Name = "VecListTwoQWordIndexed";
349 let ParserMethod = "parseVectorList";
350 let RenderMethod = "addVecListIndexedOperands";
352 def VecListTwoQWordIndexed : Operand<i32> {
353 let ParserMatchClass = VecListTwoQWordIndexAsmOperand;
354 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
358 // Register list of three D registers with byte lane subscripting.
359 def VecListThreeDByteIndexAsmOperand : AsmOperandClass {
360 let Name = "VecListThreeDByteIndexed";
361 let ParserMethod = "parseVectorList";
362 let RenderMethod = "addVecListIndexedOperands";
364 def VecListThreeDByteIndexed : Operand<i32> {
365 let ParserMatchClass = VecListThreeDByteIndexAsmOperand;
366 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
368 // ...with half-word lane subscripting.
369 def VecListThreeDHWordIndexAsmOperand : AsmOperandClass {
370 let Name = "VecListThreeDHWordIndexed";
371 let ParserMethod = "parseVectorList";
372 let RenderMethod = "addVecListIndexedOperands";
374 def VecListThreeDHWordIndexed : Operand<i32> {
375 let ParserMatchClass = VecListThreeDHWordIndexAsmOperand;
376 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
378 // ...with word lane subscripting.
379 def VecListThreeDWordIndexAsmOperand : AsmOperandClass {
380 let Name = "VecListThreeDWordIndexed";
381 let ParserMethod = "parseVectorList";
382 let RenderMethod = "addVecListIndexedOperands";
384 def VecListThreeDWordIndexed : Operand<i32> {
385 let ParserMatchClass = VecListThreeDWordIndexAsmOperand;
386 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
388 // Register list of three Q registers with half-word lane subscripting.
389 def VecListThreeQHWordIndexAsmOperand : AsmOperandClass {
390 let Name = "VecListThreeQHWordIndexed";
391 let ParserMethod = "parseVectorList";
392 let RenderMethod = "addVecListIndexedOperands";
394 def VecListThreeQHWordIndexed : Operand<i32> {
395 let ParserMatchClass = VecListThreeQHWordIndexAsmOperand;
396 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
398 // ...with word lane subscripting.
399 def VecListThreeQWordIndexAsmOperand : AsmOperandClass {
400 let Name = "VecListThreeQWordIndexed";
401 let ParserMethod = "parseVectorList";
402 let RenderMethod = "addVecListIndexedOperands";
404 def VecListThreeQWordIndexed : Operand<i32> {
405 let ParserMatchClass = VecListThreeQWordIndexAsmOperand;
406 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
409 // Register list of four D registers with byte lane subscripting.
410 def VecListFourDByteIndexAsmOperand : AsmOperandClass {
411 let Name = "VecListFourDByteIndexed";
412 let ParserMethod = "parseVectorList";
413 let RenderMethod = "addVecListIndexedOperands";
415 def VecListFourDByteIndexed : Operand<i32> {
416 let ParserMatchClass = VecListFourDByteIndexAsmOperand;
417 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
419 // ...with half-word lane subscripting.
420 def VecListFourDHWordIndexAsmOperand : AsmOperandClass {
421 let Name = "VecListFourDHWordIndexed";
422 let ParserMethod = "parseVectorList";
423 let RenderMethod = "addVecListIndexedOperands";
425 def VecListFourDHWordIndexed : Operand<i32> {
426 let ParserMatchClass = VecListFourDHWordIndexAsmOperand;
427 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
429 // ...with word lane subscripting.
430 def VecListFourDWordIndexAsmOperand : AsmOperandClass {
431 let Name = "VecListFourDWordIndexed";
432 let ParserMethod = "parseVectorList";
433 let RenderMethod = "addVecListIndexedOperands";
435 def VecListFourDWordIndexed : Operand<i32> {
436 let ParserMatchClass = VecListFourDWordIndexAsmOperand;
437 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
439 // Register list of four Q registers with half-word lane subscripting.
440 def VecListFourQHWordIndexAsmOperand : AsmOperandClass {
441 let Name = "VecListFourQHWordIndexed";
442 let ParserMethod = "parseVectorList";
443 let RenderMethod = "addVecListIndexedOperands";
445 def VecListFourQHWordIndexed : Operand<i32> {
446 let ParserMatchClass = VecListFourQHWordIndexAsmOperand;
447 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
449 // ...with word lane subscripting.
450 def VecListFourQWordIndexAsmOperand : AsmOperandClass {
451 let Name = "VecListFourQWordIndexed";
452 let ParserMethod = "parseVectorList";
453 let RenderMethod = "addVecListIndexedOperands";
455 def VecListFourQWordIndexed : Operand<i32> {
456 let ParserMatchClass = VecListFourQWordIndexAsmOperand;
457 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
460 def dword_alignedload : PatFrag<(ops node:$ptr), (load node:$ptr), [{
461 return cast<LoadSDNode>(N)->getAlignment() >= 8;
463 def dword_alignedstore : PatFrag<(ops node:$val, node:$ptr),
464 (store node:$val, node:$ptr), [{
465 return cast<StoreSDNode>(N)->getAlignment() >= 8;
467 def word_alignedload : PatFrag<(ops node:$ptr), (load node:$ptr), [{
468 return cast<LoadSDNode>(N)->getAlignment() == 4;
470 def word_alignedstore : PatFrag<(ops node:$val, node:$ptr),
471 (store node:$val, node:$ptr), [{
472 return cast<StoreSDNode>(N)->getAlignment() == 4;
474 def hword_alignedload : PatFrag<(ops node:$ptr), (load node:$ptr), [{
475 return cast<LoadSDNode>(N)->getAlignment() == 2;
477 def hword_alignedstore : PatFrag<(ops node:$val, node:$ptr),
478 (store node:$val, node:$ptr), [{
479 return cast<StoreSDNode>(N)->getAlignment() == 2;
481 def byte_alignedload : PatFrag<(ops node:$ptr), (load node:$ptr), [{
482 return cast<LoadSDNode>(N)->getAlignment() == 1;
484 def byte_alignedstore : PatFrag<(ops node:$val, node:$ptr),
485 (store node:$val, node:$ptr), [{
486 return cast<StoreSDNode>(N)->getAlignment() == 1;
488 def non_word_alignedload : PatFrag<(ops node:$ptr), (load node:$ptr), [{
489 return cast<LoadSDNode>(N)->getAlignment() < 4;
491 def non_word_alignedstore : PatFrag<(ops node:$val, node:$ptr),
492 (store node:$val, node:$ptr), [{
493 return cast<StoreSDNode>(N)->getAlignment() < 4;
496 //===----------------------------------------------------------------------===//
497 // NEON-specific DAG Nodes.
498 //===----------------------------------------------------------------------===//
500 def SDTARMVCMP : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<1, 2>]>;
501 def SDTARMVCMPZ : SDTypeProfile<1, 1, []>;
503 def NEONvceq : SDNode<"ARMISD::VCEQ", SDTARMVCMP>;
504 def NEONvceqz : SDNode<"ARMISD::VCEQZ", SDTARMVCMPZ>;
505 def NEONvcge : SDNode<"ARMISD::VCGE", SDTARMVCMP>;
506 def NEONvcgez : SDNode<"ARMISD::VCGEZ", SDTARMVCMPZ>;
507 def NEONvclez : SDNode<"ARMISD::VCLEZ", SDTARMVCMPZ>;
508 def NEONvcgeu : SDNode<"ARMISD::VCGEU", SDTARMVCMP>;
509 def NEONvcgt : SDNode<"ARMISD::VCGT", SDTARMVCMP>;
510 def NEONvcgtz : SDNode<"ARMISD::VCGTZ", SDTARMVCMPZ>;
511 def NEONvcltz : SDNode<"ARMISD::VCLTZ", SDTARMVCMPZ>;
512 def NEONvcgtu : SDNode<"ARMISD::VCGTU", SDTARMVCMP>;
513 def NEONvtst : SDNode<"ARMISD::VTST", SDTARMVCMP>;
515 // Types for vector shift by immediates. The "SHX" version is for long and
516 // narrow operations where the source and destination vectors have different
517 // types. The "SHINS" version is for shift and insert operations.
518 def SDTARMVSH : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
520 def SDTARMVSHX : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
522 def SDTARMVSHINS : SDTypeProfile<1, 3, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
523 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
525 def NEONvshl : SDNode<"ARMISD::VSHL", SDTARMVSH>;
526 def NEONvshrs : SDNode<"ARMISD::VSHRs", SDTARMVSH>;
527 def NEONvshru : SDNode<"ARMISD::VSHRu", SDTARMVSH>;
528 def NEONvshrn : SDNode<"ARMISD::VSHRN", SDTARMVSHX>;
530 def NEONvrshrs : SDNode<"ARMISD::VRSHRs", SDTARMVSH>;
531 def NEONvrshru : SDNode<"ARMISD::VRSHRu", SDTARMVSH>;
532 def NEONvrshrn : SDNode<"ARMISD::VRSHRN", SDTARMVSHX>;
534 def NEONvqshls : SDNode<"ARMISD::VQSHLs", SDTARMVSH>;
535 def NEONvqshlu : SDNode<"ARMISD::VQSHLu", SDTARMVSH>;
536 def NEONvqshlsu : SDNode<"ARMISD::VQSHLsu", SDTARMVSH>;
537 def NEONvqshrns : SDNode<"ARMISD::VQSHRNs", SDTARMVSHX>;
538 def NEONvqshrnu : SDNode<"ARMISD::VQSHRNu", SDTARMVSHX>;
539 def NEONvqshrnsu : SDNode<"ARMISD::VQSHRNsu", SDTARMVSHX>;
541 def NEONvqrshrns : SDNode<"ARMISD::VQRSHRNs", SDTARMVSHX>;
542 def NEONvqrshrnu : SDNode<"ARMISD::VQRSHRNu", SDTARMVSHX>;
543 def NEONvqrshrnsu : SDNode<"ARMISD::VQRSHRNsu", SDTARMVSHX>;
545 def NEONvsli : SDNode<"ARMISD::VSLI", SDTARMVSHINS>;
546 def NEONvsri : SDNode<"ARMISD::VSRI", SDTARMVSHINS>;
548 def SDTARMVGETLN : SDTypeProfile<1, 2, [SDTCisVT<0, i32>, SDTCisInt<1>,
550 def NEONvgetlaneu : SDNode<"ARMISD::VGETLANEu", SDTARMVGETLN>;
551 def NEONvgetlanes : SDNode<"ARMISD::VGETLANEs", SDTARMVGETLN>;
553 def SDTARMVMOVIMM : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisVT<1, i32>]>;
554 def NEONvmovImm : SDNode<"ARMISD::VMOVIMM", SDTARMVMOVIMM>;
555 def NEONvmvnImm : SDNode<"ARMISD::VMVNIMM", SDTARMVMOVIMM>;
556 def NEONvmovFPImm : SDNode<"ARMISD::VMOVFPIMM", SDTARMVMOVIMM>;
558 def SDTARMVORRIMM : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
560 def NEONvorrImm : SDNode<"ARMISD::VORRIMM", SDTARMVORRIMM>;
561 def NEONvbicImm : SDNode<"ARMISD::VBICIMM", SDTARMVORRIMM>;
563 def NEONvbsl : SDNode<"ARMISD::VBSL",
564 SDTypeProfile<1, 3, [SDTCisVec<0>,
567 SDTCisSameAs<0, 3>]>>;
569 def NEONvdup : SDNode<"ARMISD::VDUP", SDTypeProfile<1, 1, [SDTCisVec<0>]>>;
571 // VDUPLANE can produce a quad-register result from a double-register source,
572 // so the result is not constrained to match the source.
573 def NEONvduplane : SDNode<"ARMISD::VDUPLANE",
574 SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisVec<1>,
577 def SDTARMVEXT : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
578 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
579 def NEONvext : SDNode<"ARMISD::VEXT", SDTARMVEXT>;
581 def SDTARMVSHUF : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0, 1>]>;
582 def NEONvrev64 : SDNode<"ARMISD::VREV64", SDTARMVSHUF>;
583 def NEONvrev32 : SDNode<"ARMISD::VREV32", SDTARMVSHUF>;
584 def NEONvrev16 : SDNode<"ARMISD::VREV16", SDTARMVSHUF>;
586 def SDTARMVSHUF2 : SDTypeProfile<2, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
588 SDTCisSameAs<0, 3>]>;
589 def NEONzip : SDNode<"ARMISD::VZIP", SDTARMVSHUF2>;
590 def NEONuzp : SDNode<"ARMISD::VUZP", SDTARMVSHUF2>;
591 def NEONtrn : SDNode<"ARMISD::VTRN", SDTARMVSHUF2>;
593 def SDTARMVMULL : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
594 SDTCisSameAs<1, 2>]>;
595 def NEONvmulls : SDNode<"ARMISD::VMULLs", SDTARMVMULL>;
596 def NEONvmullu : SDNode<"ARMISD::VMULLu", SDTARMVMULL>;
598 def SDTARMVTBL1 : SDTypeProfile<1, 2, [SDTCisVT<0, v8i8>, SDTCisVT<1, v8i8>,
600 def SDTARMVTBL2 : SDTypeProfile<1, 3, [SDTCisVT<0, v8i8>, SDTCisVT<1, v8i8>,
601 SDTCisVT<2, v8i8>, SDTCisVT<3, v8i8>]>;
602 def NEONvtbl1 : SDNode<"ARMISD::VTBL1", SDTARMVTBL1>;
603 def NEONvtbl2 : SDNode<"ARMISD::VTBL2", SDTARMVTBL2>;
606 def NEONimmAllZerosV: PatLeaf<(NEONvmovImm (i32 timm)), [{
607 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
608 unsigned EltBits = 0;
609 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
610 return (EltBits == 32 && EltVal == 0);
613 def NEONimmAllOnesV: PatLeaf<(NEONvmovImm (i32 timm)), [{
614 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
615 unsigned EltBits = 0;
616 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
617 return (EltBits == 8 && EltVal == 0xff);
620 //===----------------------------------------------------------------------===//
621 // NEON load / store instructions
622 //===----------------------------------------------------------------------===//
624 // Use VLDM to load a Q register as a D register pair.
625 // This is a pseudo instruction that is expanded to VLDMD after reg alloc.
627 : PseudoVFPLdStM<(outs DPair:$dst), (ins GPR:$Rn),
629 [(set DPair:$dst, (v2f64 (word_alignedload GPR:$Rn)))]>;
631 // Use VSTM to store a Q register as a D register pair.
632 // This is a pseudo instruction that is expanded to VSTMD after reg alloc.
634 : PseudoVFPLdStM<(outs), (ins DPair:$src, GPR:$Rn),
636 [(word_alignedstore (v2f64 DPair:$src), GPR:$Rn)]>;
638 // Classes for VLD* pseudo-instructions with multi-register operands.
639 // These are expanded to real instructions after register allocation.
640 class VLDQPseudo<InstrItinClass itin>
641 : PseudoNLdSt<(outs QPR:$dst), (ins addrmode6:$addr), itin, "">;
642 class VLDQWBPseudo<InstrItinClass itin>
643 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
644 (ins addrmode6:$addr, am6offset:$offset), itin,
646 class VLDQWBfixedPseudo<InstrItinClass itin>
647 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
648 (ins addrmode6:$addr), itin,
650 class VLDQWBregisterPseudo<InstrItinClass itin>
651 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
652 (ins addrmode6:$addr, rGPR:$offset), itin,
655 class VLDQQPseudo<InstrItinClass itin>
656 : PseudoNLdSt<(outs QQPR:$dst), (ins addrmode6:$addr), itin, "">;
657 class VLDQQWBPseudo<InstrItinClass itin>
658 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
659 (ins addrmode6:$addr, am6offset:$offset), itin,
661 class VLDQQWBfixedPseudo<InstrItinClass itin>
662 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
663 (ins addrmode6:$addr), itin,
665 class VLDQQWBregisterPseudo<InstrItinClass itin>
666 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
667 (ins addrmode6:$addr, rGPR:$offset), itin,
671 class VLDQQQQPseudo<InstrItinClass itin>
672 : PseudoNLdSt<(outs QQQQPR:$dst), (ins addrmode6:$addr, QQQQPR:$src),itin,
674 class VLDQQQQWBPseudo<InstrItinClass itin>
675 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
676 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
677 "$addr.addr = $wb, $src = $dst">;
679 let mayLoad = 1, hasSideEffects = 0, hasExtraDefRegAllocReq = 1 in {
681 // VLD1 : Vector Load (multiple single elements)
682 class VLD1D<bits<4> op7_4, string Dt, Operand AddrMode>
683 : NLdSt<0,0b10,0b0111,op7_4, (outs VecListOneD:$Vd),
684 (ins AddrMode:$Rn), IIC_VLD1,
685 "vld1", Dt, "$Vd, $Rn", "", []>, Sched<[WriteVLD1]> {
688 let DecoderMethod = "DecodeVLDST1Instruction";
690 class VLD1Q<bits<4> op7_4, string Dt, Operand AddrMode>
691 : NLdSt<0,0b10,0b1010,op7_4, (outs VecListDPair:$Vd),
692 (ins AddrMode:$Rn), IIC_VLD1x2,
693 "vld1", Dt, "$Vd, $Rn", "", []>, Sched<[WriteVLD2]> {
695 let Inst{5-4} = Rn{5-4};
696 let DecoderMethod = "DecodeVLDST1Instruction";
699 def VLD1d8 : VLD1D<{0,0,0,?}, "8", addrmode6align64>;
700 def VLD1d16 : VLD1D<{0,1,0,?}, "16", addrmode6align64>;
701 def VLD1d32 : VLD1D<{1,0,0,?}, "32", addrmode6align64>;
702 def VLD1d64 : VLD1D<{1,1,0,?}, "64", addrmode6align64>;
704 def VLD1q8 : VLD1Q<{0,0,?,?}, "8", addrmode6align64or128>;
705 def VLD1q16 : VLD1Q<{0,1,?,?}, "16", addrmode6align64or128>;
706 def VLD1q32 : VLD1Q<{1,0,?,?}, "32", addrmode6align64or128>;
707 def VLD1q64 : VLD1Q<{1,1,?,?}, "64", addrmode6align64or128>;
709 // ...with address register writeback:
710 multiclass VLD1DWB<bits<4> op7_4, string Dt, Operand AddrMode> {
711 def _fixed : NLdSt<0,0b10, 0b0111,op7_4, (outs VecListOneD:$Vd, GPR:$wb),
712 (ins AddrMode:$Rn), IIC_VLD1u,
713 "vld1", Dt, "$Vd, $Rn!",
714 "$Rn.addr = $wb", []>, Sched<[WriteVLD1]> {
715 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
717 let DecoderMethod = "DecodeVLDST1Instruction";
719 def _register : NLdSt<0,0b10,0b0111,op7_4, (outs VecListOneD:$Vd, GPR:$wb),
720 (ins AddrMode:$Rn, rGPR:$Rm), IIC_VLD1u,
721 "vld1", Dt, "$Vd, $Rn, $Rm",
722 "$Rn.addr = $wb", []>, Sched<[WriteVLD1]> {
724 let DecoderMethod = "DecodeVLDST1Instruction";
727 multiclass VLD1QWB<bits<4> op7_4, string Dt, Operand AddrMode> {
728 def _fixed : NLdSt<0,0b10,0b1010,op7_4, (outs VecListDPair:$Vd, GPR:$wb),
729 (ins AddrMode:$Rn), IIC_VLD1x2u,
730 "vld1", Dt, "$Vd, $Rn!",
731 "$Rn.addr = $wb", []>, Sched<[WriteVLD2]> {
732 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
733 let Inst{5-4} = Rn{5-4};
734 let DecoderMethod = "DecodeVLDST1Instruction";
736 def _register : NLdSt<0,0b10,0b1010,op7_4, (outs VecListDPair:$Vd, GPR:$wb),
737 (ins AddrMode:$Rn, rGPR:$Rm), IIC_VLD1x2u,
738 "vld1", Dt, "$Vd, $Rn, $Rm",
739 "$Rn.addr = $wb", []>, Sched<[WriteVLD2]> {
740 let Inst{5-4} = Rn{5-4};
741 let DecoderMethod = "DecodeVLDST1Instruction";
745 defm VLD1d8wb : VLD1DWB<{0,0,0,?}, "8", addrmode6align64>;
746 defm VLD1d16wb : VLD1DWB<{0,1,0,?}, "16", addrmode6align64>;
747 defm VLD1d32wb : VLD1DWB<{1,0,0,?}, "32", addrmode6align64>;
748 defm VLD1d64wb : VLD1DWB<{1,1,0,?}, "64", addrmode6align64>;
749 defm VLD1q8wb : VLD1QWB<{0,0,?,?}, "8", addrmode6align64or128>;
750 defm VLD1q16wb : VLD1QWB<{0,1,?,?}, "16", addrmode6align64or128>;
751 defm VLD1q32wb : VLD1QWB<{1,0,?,?}, "32", addrmode6align64or128>;
752 defm VLD1q64wb : VLD1QWB<{1,1,?,?}, "64", addrmode6align64or128>;
754 // ...with 3 registers
755 class VLD1D3<bits<4> op7_4, string Dt, Operand AddrMode>
756 : NLdSt<0,0b10,0b0110,op7_4, (outs VecListThreeD:$Vd),
757 (ins AddrMode:$Rn), IIC_VLD1x3, "vld1", Dt,
758 "$Vd, $Rn", "", []>, Sched<[WriteVLD3]> {
761 let DecoderMethod = "DecodeVLDST1Instruction";
763 multiclass VLD1D3WB<bits<4> op7_4, string Dt, Operand AddrMode> {
764 def _fixed : NLdSt<0,0b10,0b0110, op7_4, (outs VecListThreeD:$Vd, GPR:$wb),
765 (ins AddrMode:$Rn), IIC_VLD1x2u,
766 "vld1", Dt, "$Vd, $Rn!",
767 "$Rn.addr = $wb", []>, Sched<[WriteVLD3]> {
768 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
770 let DecoderMethod = "DecodeVLDST1Instruction";
772 def _register : NLdSt<0,0b10,0b0110,op7_4, (outs VecListThreeD:$Vd, GPR:$wb),
773 (ins AddrMode:$Rn, rGPR:$Rm), IIC_VLD1x2u,
774 "vld1", Dt, "$Vd, $Rn, $Rm",
775 "$Rn.addr = $wb", []>, Sched<[WriteVLD3]> {
777 let DecoderMethod = "DecodeVLDST1Instruction";
781 def VLD1d8T : VLD1D3<{0,0,0,?}, "8", addrmode6align64>;
782 def VLD1d16T : VLD1D3<{0,1,0,?}, "16", addrmode6align64>;
783 def VLD1d32T : VLD1D3<{1,0,0,?}, "32", addrmode6align64>;
784 def VLD1d64T : VLD1D3<{1,1,0,?}, "64", addrmode6align64>;
786 defm VLD1d8Twb : VLD1D3WB<{0,0,0,?}, "8", addrmode6align64>;
787 defm VLD1d16Twb : VLD1D3WB<{0,1,0,?}, "16", addrmode6align64>;
788 defm VLD1d32Twb : VLD1D3WB<{1,0,0,?}, "32", addrmode6align64>;
789 defm VLD1d64Twb : VLD1D3WB<{1,1,0,?}, "64", addrmode6align64>;
791 def VLD1d64TPseudo : VLDQQPseudo<IIC_VLD1x3>, Sched<[WriteVLD3]>;
792 def VLD1d64TPseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD1x3>, Sched<[WriteVLD3]>;
793 def VLD1d64TPseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD1x3>, Sched<[WriteVLD3]>;
795 // ...with 4 registers
796 class VLD1D4<bits<4> op7_4, string Dt, Operand AddrMode>
797 : NLdSt<0, 0b10, 0b0010, op7_4, (outs VecListFourD:$Vd),
798 (ins AddrMode:$Rn), IIC_VLD1x4, "vld1", Dt,
799 "$Vd, $Rn", "", []>, Sched<[WriteVLD4]> {
801 let Inst{5-4} = Rn{5-4};
802 let DecoderMethod = "DecodeVLDST1Instruction";
804 multiclass VLD1D4WB<bits<4> op7_4, string Dt, Operand AddrMode> {
805 def _fixed : NLdSt<0,0b10,0b0010, op7_4, (outs VecListFourD:$Vd, GPR:$wb),
806 (ins AddrMode:$Rn), IIC_VLD1x2u,
807 "vld1", Dt, "$Vd, $Rn!",
808 "$Rn.addr = $wb", []>, Sched<[WriteVLD4]> {
809 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
810 let Inst{5-4} = Rn{5-4};
811 let DecoderMethod = "DecodeVLDST1Instruction";
813 def _register : NLdSt<0,0b10,0b0010,op7_4, (outs VecListFourD:$Vd, GPR:$wb),
814 (ins AddrMode:$Rn, rGPR:$Rm), IIC_VLD1x2u,
815 "vld1", Dt, "$Vd, $Rn, $Rm",
816 "$Rn.addr = $wb", []>, Sched<[WriteVLD4]> {
817 let Inst{5-4} = Rn{5-4};
818 let DecoderMethod = "DecodeVLDST1Instruction";
822 def VLD1d8Q : VLD1D4<{0,0,?,?}, "8", addrmode6align64or128or256>;
823 def VLD1d16Q : VLD1D4<{0,1,?,?}, "16", addrmode6align64or128or256>;
824 def VLD1d32Q : VLD1D4<{1,0,?,?}, "32", addrmode6align64or128or256>;
825 def VLD1d64Q : VLD1D4<{1,1,?,?}, "64", addrmode6align64or128or256>;
827 defm VLD1d8Qwb : VLD1D4WB<{0,0,?,?}, "8", addrmode6align64or128or256>;
828 defm VLD1d16Qwb : VLD1D4WB<{0,1,?,?}, "16", addrmode6align64or128or256>;
829 defm VLD1d32Qwb : VLD1D4WB<{1,0,?,?}, "32", addrmode6align64or128or256>;
830 defm VLD1d64Qwb : VLD1D4WB<{1,1,?,?}, "64", addrmode6align64or128or256>;
832 def VLD1d64QPseudo : VLDQQPseudo<IIC_VLD1x4>, Sched<[WriteVLD4]>;
833 def VLD1d64QPseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD1x4>, Sched<[WriteVLD4]>;
834 def VLD1d64QPseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD1x4>, Sched<[WriteVLD4]>;
836 // VLD2 : Vector Load (multiple 2-element structures)
837 class VLD2<bits<4> op11_8, bits<4> op7_4, string Dt, RegisterOperand VdTy,
838 InstrItinClass itin, Operand AddrMode>
839 : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd),
840 (ins AddrMode:$Rn), itin,
841 "vld2", Dt, "$Vd, $Rn", "", []> {
843 let Inst{5-4} = Rn{5-4};
844 let DecoderMethod = "DecodeVLDST2Instruction";
847 def VLD2d8 : VLD2<0b1000, {0,0,?,?}, "8", VecListDPair, IIC_VLD2,
848 addrmode6align64or128>, Sched<[WriteVLD2]>;
849 def VLD2d16 : VLD2<0b1000, {0,1,?,?}, "16", VecListDPair, IIC_VLD2,
850 addrmode6align64or128>, Sched<[WriteVLD2]>;
851 def VLD2d32 : VLD2<0b1000, {1,0,?,?}, "32", VecListDPair, IIC_VLD2,
852 addrmode6align64or128>, Sched<[WriteVLD2]>;
854 def VLD2q8 : VLD2<0b0011, {0,0,?,?}, "8", VecListFourD, IIC_VLD2x2,
855 addrmode6align64or128or256>, Sched<[WriteVLD4]>;
856 def VLD2q16 : VLD2<0b0011, {0,1,?,?}, "16", VecListFourD, IIC_VLD2x2,
857 addrmode6align64or128or256>, Sched<[WriteVLD4]>;
858 def VLD2q32 : VLD2<0b0011, {1,0,?,?}, "32", VecListFourD, IIC_VLD2x2,
859 addrmode6align64or128or256>, Sched<[WriteVLD4]>;
861 def VLD2q8Pseudo : VLDQQPseudo<IIC_VLD2x2>, Sched<[WriteVLD4]>;
862 def VLD2q16Pseudo : VLDQQPseudo<IIC_VLD2x2>, Sched<[WriteVLD4]>;
863 def VLD2q32Pseudo : VLDQQPseudo<IIC_VLD2x2>, Sched<[WriteVLD4]>;
865 // ...with address register writeback:
866 multiclass VLD2WB<bits<4> op11_8, bits<4> op7_4, string Dt,
867 RegisterOperand VdTy, InstrItinClass itin, Operand AddrMode> {
868 def _fixed : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd, GPR:$wb),
869 (ins AddrMode:$Rn), itin,
870 "vld2", Dt, "$Vd, $Rn!",
871 "$Rn.addr = $wb", []> {
872 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
873 let Inst{5-4} = Rn{5-4};
874 let DecoderMethod = "DecodeVLDST2Instruction";
876 def _register : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd, GPR:$wb),
877 (ins AddrMode:$Rn, rGPR:$Rm), itin,
878 "vld2", Dt, "$Vd, $Rn, $Rm",
879 "$Rn.addr = $wb", []> {
880 let Inst{5-4} = Rn{5-4};
881 let DecoderMethod = "DecodeVLDST2Instruction";
885 defm VLD2d8wb : VLD2WB<0b1000, {0,0,?,?}, "8", VecListDPair, IIC_VLD2u,
886 addrmode6align64or128>, Sched<[WriteVLD2]>;
887 defm VLD2d16wb : VLD2WB<0b1000, {0,1,?,?}, "16", VecListDPair, IIC_VLD2u,
888 addrmode6align64or128>, Sched<[WriteVLD2]>;
889 defm VLD2d32wb : VLD2WB<0b1000, {1,0,?,?}, "32", VecListDPair, IIC_VLD2u,
890 addrmode6align64or128>, Sched<[WriteVLD2]>;
892 defm VLD2q8wb : VLD2WB<0b0011, {0,0,?,?}, "8", VecListFourD, IIC_VLD2x2u,
893 addrmode6align64or128or256>, Sched<[WriteVLD4]>;
894 defm VLD2q16wb : VLD2WB<0b0011, {0,1,?,?}, "16", VecListFourD, IIC_VLD2x2u,
895 addrmode6align64or128or256>, Sched<[WriteVLD4]>;
896 defm VLD2q32wb : VLD2WB<0b0011, {1,0,?,?}, "32", VecListFourD, IIC_VLD2x2u,
897 addrmode6align64or128or256>, Sched<[WriteVLD4]>;
899 def VLD2q8PseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD2x2u>, Sched<[WriteVLD4]>;
900 def VLD2q16PseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD2x2u>, Sched<[WriteVLD4]>;
901 def VLD2q32PseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD2x2u>, Sched<[WriteVLD4]>;
902 def VLD2q8PseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD2x2u>, Sched<[WriteVLD4]>;
903 def VLD2q16PseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD2x2u>, Sched<[WriteVLD4]>;
904 def VLD2q32PseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD2x2u>, Sched<[WriteVLD4]>;
906 // ...with double-spaced registers
907 def VLD2b8 : VLD2<0b1001, {0,0,?,?}, "8", VecListDPairSpaced, IIC_VLD2,
908 addrmode6align64or128>, Sched<[WriteVLD2]>;
909 def VLD2b16 : VLD2<0b1001, {0,1,?,?}, "16", VecListDPairSpaced, IIC_VLD2,
910 addrmode6align64or128>, Sched<[WriteVLD2]>;
911 def VLD2b32 : VLD2<0b1001, {1,0,?,?}, "32", VecListDPairSpaced, IIC_VLD2,
912 addrmode6align64or128>, Sched<[WriteVLD2]>;
913 defm VLD2b8wb : VLD2WB<0b1001, {0,0,?,?}, "8", VecListDPairSpaced, IIC_VLD2u,
914 addrmode6align64or128>, Sched<[WriteVLD2]>;
915 defm VLD2b16wb : VLD2WB<0b1001, {0,1,?,?}, "16", VecListDPairSpaced, IIC_VLD2u,
916 addrmode6align64or128>, Sched<[WriteVLD2]>;
917 defm VLD2b32wb : VLD2WB<0b1001, {1,0,?,?}, "32", VecListDPairSpaced, IIC_VLD2u,
918 addrmode6align64or128>, Sched<[WriteVLD2]>;
920 // VLD3 : Vector Load (multiple 3-element structures)
921 class VLD3D<bits<4> op11_8, bits<4> op7_4, string Dt>
922 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
923 (ins addrmode6:$Rn), IIC_VLD3,
924 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn", "", []>, Sched<[WriteVLD3]> {
927 let DecoderMethod = "DecodeVLDST3Instruction";
930 def VLD3d8 : VLD3D<0b0100, {0,0,0,?}, "8">;
931 def VLD3d16 : VLD3D<0b0100, {0,1,0,?}, "16">;
932 def VLD3d32 : VLD3D<0b0100, {1,0,0,?}, "32">;
934 def VLD3d8Pseudo : VLDQQPseudo<IIC_VLD3>, Sched<[WriteVLD3]>;
935 def VLD3d16Pseudo : VLDQQPseudo<IIC_VLD3>, Sched<[WriteVLD3]>;
936 def VLD3d32Pseudo : VLDQQPseudo<IIC_VLD3>, Sched<[WriteVLD3]>;
938 // ...with address register writeback:
939 class VLD3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
940 : NLdSt<0, 0b10, op11_8, op7_4,
941 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
942 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD3u,
943 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn$Rm",
944 "$Rn.addr = $wb", []>, Sched<[WriteVLD3]> {
946 let DecoderMethod = "DecodeVLDST3Instruction";
949 def VLD3d8_UPD : VLD3DWB<0b0100, {0,0,0,?}, "8">;
950 def VLD3d16_UPD : VLD3DWB<0b0100, {0,1,0,?}, "16">;
951 def VLD3d32_UPD : VLD3DWB<0b0100, {1,0,0,?}, "32">;
953 def VLD3d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>, Sched<[WriteVLD3]>;
954 def VLD3d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>, Sched<[WriteVLD3]>;
955 def VLD3d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>, Sched<[WriteVLD3]>;
957 // ...with double-spaced registers:
958 def VLD3q8 : VLD3D<0b0101, {0,0,0,?}, "8">;
959 def VLD3q16 : VLD3D<0b0101, {0,1,0,?}, "16">;
960 def VLD3q32 : VLD3D<0b0101, {1,0,0,?}, "32">;
961 def VLD3q8_UPD : VLD3DWB<0b0101, {0,0,0,?}, "8">;
962 def VLD3q16_UPD : VLD3DWB<0b0101, {0,1,0,?}, "16">;
963 def VLD3q32_UPD : VLD3DWB<0b0101, {1,0,0,?}, "32">;
965 def VLD3q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>, Sched<[WriteVLD3]>;
966 def VLD3q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>, Sched<[WriteVLD3]>;
967 def VLD3q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>, Sched<[WriteVLD3]>;
969 // ...alternate versions to be allocated odd register numbers:
970 def VLD3q8oddPseudo : VLDQQQQPseudo<IIC_VLD3>, Sched<[WriteVLD3]>;
971 def VLD3q16oddPseudo : VLDQQQQPseudo<IIC_VLD3>, Sched<[WriteVLD3]>;
972 def VLD3q32oddPseudo : VLDQQQQPseudo<IIC_VLD3>, Sched<[WriteVLD3]>;
974 def VLD3q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>, Sched<[WriteVLD3]>;
975 def VLD3q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>, Sched<[WriteVLD3]>;
976 def VLD3q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>, Sched<[WriteVLD3]>;
978 // VLD4 : Vector Load (multiple 4-element structures)
979 class VLD4D<bits<4> op11_8, bits<4> op7_4, string Dt>
980 : NLdSt<0, 0b10, op11_8, op7_4,
981 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
982 (ins addrmode6:$Rn), IIC_VLD4,
983 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []>,
986 let Inst{5-4} = Rn{5-4};
987 let DecoderMethod = "DecodeVLDST4Instruction";
990 def VLD4d8 : VLD4D<0b0000, {0,0,?,?}, "8">;
991 def VLD4d16 : VLD4D<0b0000, {0,1,?,?}, "16">;
992 def VLD4d32 : VLD4D<0b0000, {1,0,?,?}, "32">;
994 def VLD4d8Pseudo : VLDQQPseudo<IIC_VLD4>, Sched<[WriteVLD4]>;
995 def VLD4d16Pseudo : VLDQQPseudo<IIC_VLD4>, Sched<[WriteVLD4]>;
996 def VLD4d32Pseudo : VLDQQPseudo<IIC_VLD4>, Sched<[WriteVLD4]>;
998 // ...with address register writeback:
999 class VLD4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1000 : NLdSt<0, 0b10, op11_8, op7_4,
1001 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
1002 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD4u,
1003 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm",
1004 "$Rn.addr = $wb", []>, Sched<[WriteVLD4]> {
1005 let Inst{5-4} = Rn{5-4};
1006 let DecoderMethod = "DecodeVLDST4Instruction";
1009 def VLD4d8_UPD : VLD4DWB<0b0000, {0,0,?,?}, "8">;
1010 def VLD4d16_UPD : VLD4DWB<0b0000, {0,1,?,?}, "16">;
1011 def VLD4d32_UPD : VLD4DWB<0b0000, {1,0,?,?}, "32">;
1013 def VLD4d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>, Sched<[WriteVLD4]>;
1014 def VLD4d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>, Sched<[WriteVLD4]>;
1015 def VLD4d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>, Sched<[WriteVLD4]>;
1017 // ...with double-spaced registers:
1018 def VLD4q8 : VLD4D<0b0001, {0,0,?,?}, "8">;
1019 def VLD4q16 : VLD4D<0b0001, {0,1,?,?}, "16">;
1020 def VLD4q32 : VLD4D<0b0001, {1,0,?,?}, "32">;
1021 def VLD4q8_UPD : VLD4DWB<0b0001, {0,0,?,?}, "8">;
1022 def VLD4q16_UPD : VLD4DWB<0b0001, {0,1,?,?}, "16">;
1023 def VLD4q32_UPD : VLD4DWB<0b0001, {1,0,?,?}, "32">;
1025 def VLD4q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>, Sched<[WriteVLD4]>;
1026 def VLD4q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>, Sched<[WriteVLD4]>;
1027 def VLD4q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>, Sched<[WriteVLD4]>;
1029 // ...alternate versions to be allocated odd register numbers:
1030 def VLD4q8oddPseudo : VLDQQQQPseudo<IIC_VLD4>, Sched<[WriteVLD4]>;
1031 def VLD4q16oddPseudo : VLDQQQQPseudo<IIC_VLD4>, Sched<[WriteVLD4]>;
1032 def VLD4q32oddPseudo : VLDQQQQPseudo<IIC_VLD4>, Sched<[WriteVLD4]>;
1034 def VLD4q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>, Sched<[WriteVLD4]>;
1035 def VLD4q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>, Sched<[WriteVLD4]>;
1036 def VLD4q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>, Sched<[WriteVLD4]>;
1038 } // mayLoad = 1, hasSideEffects = 0, hasExtraDefRegAllocReq = 1
1040 // Classes for VLD*LN pseudo-instructions with multi-register operands.
1041 // These are expanded to real instructions after register allocation.
1042 class VLDQLNPseudo<InstrItinClass itin>
1043 : PseudoNLdSt<(outs QPR:$dst),
1044 (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
1045 itin, "$src = $dst">;
1046 class VLDQLNWBPseudo<InstrItinClass itin>
1047 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
1048 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
1049 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
1050 class VLDQQLNPseudo<InstrItinClass itin>
1051 : PseudoNLdSt<(outs QQPR:$dst),
1052 (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
1053 itin, "$src = $dst">;
1054 class VLDQQLNWBPseudo<InstrItinClass itin>
1055 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
1056 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
1057 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
1058 class VLDQQQQLNPseudo<InstrItinClass itin>
1059 : PseudoNLdSt<(outs QQQQPR:$dst),
1060 (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
1061 itin, "$src = $dst">;
1062 class VLDQQQQLNWBPseudo<InstrItinClass itin>
1063 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
1064 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
1065 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
1067 // VLD1LN : Vector Load (single element to one lane)
1068 class VLD1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1070 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
1071 (ins addrmode6:$Rn, DPR:$src, nohash_imm:$lane),
1072 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
1074 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
1075 (i32 (LoadOp addrmode6:$Rn)),
1078 let DecoderMethod = "DecodeVLD1LN";
1080 class VLD1LN32<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1082 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
1083 (ins addrmode6oneL32:$Rn, DPR:$src, nohash_imm:$lane),
1084 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
1086 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
1087 (i32 (LoadOp addrmode6oneL32:$Rn)),
1088 imm:$lane))]>, Sched<[WriteVLD1]> {
1090 let DecoderMethod = "DecodeVLD1LN";
1092 class VLD1QLNPseudo<ValueType Ty, PatFrag LoadOp> : VLDQLNPseudo<IIC_VLD1ln>,
1093 Sched<[WriteVLD1]> {
1094 let Pattern = [(set QPR:$dst, (vector_insert (Ty QPR:$src),
1095 (i32 (LoadOp addrmode6:$addr)),
1099 def VLD1LNd8 : VLD1LN<0b0000, {?,?,?,0}, "8", v8i8, extloadi8> {
1100 let Inst{7-5} = lane{2-0};
1102 def VLD1LNd16 : VLD1LN<0b0100, {?,?,0,?}, "16", v4i16, extloadi16> {
1103 let Inst{7-6} = lane{1-0};
1104 let Inst{5-4} = Rn{5-4};
1106 def VLD1LNd32 : VLD1LN32<0b1000, {?,0,?,?}, "32", v2i32, load> {
1107 let Inst{7} = lane{0};
1108 let Inst{5-4} = Rn{5-4};
1111 def VLD1LNq8Pseudo : VLD1QLNPseudo<v16i8, extloadi8>;
1112 def VLD1LNq16Pseudo : VLD1QLNPseudo<v8i16, extloadi16>;
1113 def VLD1LNq32Pseudo : VLD1QLNPseudo<v4i32, load>;
1115 def : Pat<(vector_insert (v2f32 DPR:$src),
1116 (f32 (load addrmode6:$addr)), imm:$lane),
1117 (VLD1LNd32 addrmode6:$addr, DPR:$src, imm:$lane)>;
1118 def : Pat<(vector_insert (v4f32 QPR:$src),
1119 (f32 (load addrmode6:$addr)), imm:$lane),
1120 (VLD1LNq32Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
1122 // A 64-bit subvector insert to the first 128-bit vector position
1123 // is a subregister copy that needs no instruction.
1124 def : Pat<(insert_subvector undef, (v1i64 DPR:$src), (i32 0)),
1125 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
1126 def : Pat<(insert_subvector undef, (v2i32 DPR:$src), (i32 0)),
1127 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
1128 def : Pat<(insert_subvector undef, (v2f32 DPR:$src), (i32 0)),
1129 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
1130 def : Pat<(insert_subvector undef, (v4i16 DPR:$src), (i32 0)),
1131 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
1132 def : Pat<(insert_subvector undef, (v4f16 DPR:$src), (i32 0)),
1133 (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
1134 def : Pat<(insert_subvector (v16i8 undef), (v8i8 DPR:$src), (i32 0)),
1135 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
1138 let mayLoad = 1, hasSideEffects = 0, hasExtraDefRegAllocReq = 1 in {
1140 // ...with address register writeback:
1141 class VLD1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1142 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, GPR:$wb),
1143 (ins addrmode6:$Rn, am6offset:$Rm,
1144 DPR:$src, nohash_imm:$lane), IIC_VLD1lnu, "vld1", Dt,
1145 "\\{$Vd[$lane]\\}, $Rn$Rm",
1146 "$src = $Vd, $Rn.addr = $wb", []>, Sched<[WriteVLD1]> {
1147 let DecoderMethod = "DecodeVLD1LN";
1150 def VLD1LNd8_UPD : VLD1LNWB<0b0000, {?,?,?,0}, "8"> {
1151 let Inst{7-5} = lane{2-0};
1153 def VLD1LNd16_UPD : VLD1LNWB<0b0100, {?,?,0,?}, "16"> {
1154 let Inst{7-6} = lane{1-0};
1155 let Inst{4} = Rn{4};
1157 def VLD1LNd32_UPD : VLD1LNWB<0b1000, {?,0,?,?}, "32"> {
1158 let Inst{7} = lane{0};
1159 let Inst{5} = Rn{4};
1160 let Inst{4} = Rn{4};
1163 def VLD1LNq8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>, Sched<[WriteVLD1]>;
1164 def VLD1LNq16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>, Sched<[WriteVLD1]>;
1165 def VLD1LNq32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>, Sched<[WriteVLD1]>;
1167 // VLD2LN : Vector Load (single 2-element structure to one lane)
1168 class VLD2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1169 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2),
1170 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, nohash_imm:$lane),
1171 IIC_VLD2ln, "vld2", Dt, "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn",
1172 "$src1 = $Vd, $src2 = $dst2", []>, Sched<[WriteVLD1]> {
1174 let Inst{4} = Rn{4};
1175 let DecoderMethod = "DecodeVLD2LN";
1178 def VLD2LNd8 : VLD2LN<0b0001, {?,?,?,?}, "8"> {
1179 let Inst{7-5} = lane{2-0};
1181 def VLD2LNd16 : VLD2LN<0b0101, {?,?,0,?}, "16"> {
1182 let Inst{7-6} = lane{1-0};
1184 def VLD2LNd32 : VLD2LN<0b1001, {?,0,0,?}, "32"> {
1185 let Inst{7} = lane{0};
1188 def VLD2LNd8Pseudo : VLDQLNPseudo<IIC_VLD2ln>, Sched<[WriteVLD1]>;
1189 def VLD2LNd16Pseudo : VLDQLNPseudo<IIC_VLD2ln>, Sched<[WriteVLD1]>;
1190 def VLD2LNd32Pseudo : VLDQLNPseudo<IIC_VLD2ln>, Sched<[WriteVLD1]>;
1192 // ...with double-spaced registers:
1193 def VLD2LNq16 : VLD2LN<0b0101, {?,?,1,?}, "16"> {
1194 let Inst{7-6} = lane{1-0};
1196 def VLD2LNq32 : VLD2LN<0b1001, {?,1,0,?}, "32"> {
1197 let Inst{7} = lane{0};
1200 def VLD2LNq16Pseudo : VLDQQLNPseudo<IIC_VLD2ln>, Sched<[WriteVLD1]>;
1201 def VLD2LNq32Pseudo : VLDQQLNPseudo<IIC_VLD2ln>, Sched<[WriteVLD1]>;
1203 // ...with address register writeback:
1204 class VLD2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1205 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
1206 (ins addrmode6:$Rn, am6offset:$Rm,
1207 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VLD2lnu, "vld2", Dt,
1208 "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn$Rm",
1209 "$src1 = $Vd, $src2 = $dst2, $Rn.addr = $wb", []> {
1210 let Inst{4} = Rn{4};
1211 let DecoderMethod = "DecodeVLD2LN";
1214 def VLD2LNd8_UPD : VLD2LNWB<0b0001, {?,?,?,?}, "8"> {
1215 let Inst{7-5} = lane{2-0};
1217 def VLD2LNd16_UPD : VLD2LNWB<0b0101, {?,?,0,?}, "16"> {
1218 let Inst{7-6} = lane{1-0};
1220 def VLD2LNd32_UPD : VLD2LNWB<0b1001, {?,0,0,?}, "32"> {
1221 let Inst{7} = lane{0};
1224 def VLD2LNd8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>, Sched<[WriteVLD1]>;
1225 def VLD2LNd16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>, Sched<[WriteVLD1]>;
1226 def VLD2LNd32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>, Sched<[WriteVLD1]>;
1228 def VLD2LNq16_UPD : VLD2LNWB<0b0101, {?,?,1,?}, "16"> {
1229 let Inst{7-6} = lane{1-0};
1231 def VLD2LNq32_UPD : VLD2LNWB<0b1001, {?,1,0,?}, "32"> {
1232 let Inst{7} = lane{0};
1235 def VLD2LNq16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>, Sched<[WriteVLD1]>;
1236 def VLD2LNq32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>, Sched<[WriteVLD1]>;
1238 // VLD3LN : Vector Load (single 3-element structure to one lane)
1239 class VLD3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1240 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
1241 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3,
1242 nohash_imm:$lane), IIC_VLD3ln, "vld3", Dt,
1243 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn",
1244 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3", []>, Sched<[WriteVLD2]> {
1246 let DecoderMethod = "DecodeVLD3LN";
1249 def VLD3LNd8 : VLD3LN<0b0010, {?,?,?,0}, "8"> {
1250 let Inst{7-5} = lane{2-0};
1252 def VLD3LNd16 : VLD3LN<0b0110, {?,?,0,0}, "16"> {
1253 let Inst{7-6} = lane{1-0};
1255 def VLD3LNd32 : VLD3LN<0b1010, {?,0,0,0}, "32"> {
1256 let Inst{7} = lane{0};
1259 def VLD3LNd8Pseudo : VLDQQLNPseudo<IIC_VLD3ln>, Sched<[WriteVLD2]>;
1260 def VLD3LNd16Pseudo : VLDQQLNPseudo<IIC_VLD3ln>, Sched<[WriteVLD2]>;
1261 def VLD3LNd32Pseudo : VLDQQLNPseudo<IIC_VLD3ln>, Sched<[WriteVLD2]>;
1263 // ...with double-spaced registers:
1264 def VLD3LNq16 : VLD3LN<0b0110, {?,?,1,0}, "16"> {
1265 let Inst{7-6} = lane{1-0};
1267 def VLD3LNq32 : VLD3LN<0b1010, {?,1,0,0}, "32"> {
1268 let Inst{7} = lane{0};
1271 def VLD3LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>, Sched<[WriteVLD2]>;
1272 def VLD3LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>, Sched<[WriteVLD2]>;
1274 // ...with address register writeback:
1275 class VLD3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1276 : NLdStLn<1, 0b10, op11_8, op7_4,
1277 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
1278 (ins addrmode6:$Rn, am6offset:$Rm,
1279 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
1280 IIC_VLD3lnu, "vld3", Dt,
1281 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn$Rm",
1282 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $Rn.addr = $wb",
1283 []>, Sched<[WriteVLD2]> {
1284 let DecoderMethod = "DecodeVLD3LN";
1287 def VLD3LNd8_UPD : VLD3LNWB<0b0010, {?,?,?,0}, "8"> {
1288 let Inst{7-5} = lane{2-0};
1290 def VLD3LNd16_UPD : VLD3LNWB<0b0110, {?,?,0,0}, "16"> {
1291 let Inst{7-6} = lane{1-0};
1293 def VLD3LNd32_UPD : VLD3LNWB<0b1010, {?,0,0,0}, "32"> {
1294 let Inst{7} = lane{0};
1297 def VLD3LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>, Sched<[WriteVLD2]>;
1298 def VLD3LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>, Sched<[WriteVLD2]>;
1299 def VLD3LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>, Sched<[WriteVLD2]>;
1301 def VLD3LNq16_UPD : VLD3LNWB<0b0110, {?,?,1,0}, "16"> {
1302 let Inst{7-6} = lane{1-0};
1304 def VLD3LNq32_UPD : VLD3LNWB<0b1010, {?,1,0,0}, "32"> {
1305 let Inst{7} = lane{0};
1308 def VLD3LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>, Sched<[WriteVLD2]>;
1309 def VLD3LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>, Sched<[WriteVLD2]>;
1311 // VLD4LN : Vector Load (single 4-element structure to one lane)
1312 class VLD4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1313 : NLdStLn<1, 0b10, op11_8, op7_4,
1314 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
1315 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
1316 nohash_imm:$lane), IIC_VLD4ln, "vld4", Dt,
1317 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn",
1318 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4", []>,
1319 Sched<[WriteVLD2]> {
1321 let Inst{4} = Rn{4};
1322 let DecoderMethod = "DecodeVLD4LN";
1325 def VLD4LNd8 : VLD4LN<0b0011, {?,?,?,?}, "8"> {
1326 let Inst{7-5} = lane{2-0};
1328 def VLD4LNd16 : VLD4LN<0b0111, {?,?,0,?}, "16"> {
1329 let Inst{7-6} = lane{1-0};
1331 def VLD4LNd32 : VLD4LN<0b1011, {?,0,?,?}, "32"> {
1332 let Inst{7} = lane{0};
1333 let Inst{5} = Rn{5};
1336 def VLD4LNd8Pseudo : VLDQQLNPseudo<IIC_VLD4ln>, Sched<[WriteVLD2]>;
1337 def VLD4LNd16Pseudo : VLDQQLNPseudo<IIC_VLD4ln>, Sched<[WriteVLD2]>;
1338 def VLD4LNd32Pseudo : VLDQQLNPseudo<IIC_VLD4ln>, Sched<[WriteVLD2]>;
1340 // ...with double-spaced registers:
1341 def VLD4LNq16 : VLD4LN<0b0111, {?,?,1,?}, "16"> {
1342 let Inst{7-6} = lane{1-0};
1344 def VLD4LNq32 : VLD4LN<0b1011, {?,1,?,?}, "32"> {
1345 let Inst{7} = lane{0};
1346 let Inst{5} = Rn{5};
1349 def VLD4LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>, Sched<[WriteVLD2]>;
1350 def VLD4LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>, Sched<[WriteVLD2]>;
1352 // ...with address register writeback:
1353 class VLD4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1354 : NLdStLn<1, 0b10, op11_8, op7_4,
1355 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
1356 (ins addrmode6:$Rn, am6offset:$Rm,
1357 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
1358 IIC_VLD4lnu, "vld4", Dt,
1359 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn$Rm",
1360 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4, $Rn.addr = $wb",
1362 let Inst{4} = Rn{4};
1363 let DecoderMethod = "DecodeVLD4LN" ;
1366 def VLD4LNd8_UPD : VLD4LNWB<0b0011, {?,?,?,?}, "8"> {
1367 let Inst{7-5} = lane{2-0};
1369 def VLD4LNd16_UPD : VLD4LNWB<0b0111, {?,?,0,?}, "16"> {
1370 let Inst{7-6} = lane{1-0};
1372 def VLD4LNd32_UPD : VLD4LNWB<0b1011, {?,0,?,?}, "32"> {
1373 let Inst{7} = lane{0};
1374 let Inst{5} = Rn{5};
1377 def VLD4LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>, Sched<[WriteVLD2]>;
1378 def VLD4LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>, Sched<[WriteVLD2]>;
1379 def VLD4LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>, Sched<[WriteVLD2]>;
1381 def VLD4LNq16_UPD : VLD4LNWB<0b0111, {?,?,1,?}, "16"> {
1382 let Inst{7-6} = lane{1-0};
1384 def VLD4LNq32_UPD : VLD4LNWB<0b1011, {?,1,?,?}, "32"> {
1385 let Inst{7} = lane{0};
1386 let Inst{5} = Rn{5};
1389 def VLD4LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>, Sched<[WriteVLD2]>;
1390 def VLD4LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>, Sched<[WriteVLD2]>;
1392 } // mayLoad = 1, hasSideEffects = 0, hasExtraDefRegAllocReq = 1
1394 // VLD1DUP : Vector Load (single element to all lanes)
1395 class VLD1DUP<bits<4> op7_4, string Dt, ValueType Ty, PatFrag LoadOp,
1397 : NLdSt<1, 0b10, 0b1100, op7_4, (outs VecListOneDAllLanes:$Vd),
1399 IIC_VLD1dup, "vld1", Dt, "$Vd, $Rn", "",
1400 [(set VecListOneDAllLanes:$Vd,
1401 (Ty (NEONvdup (i32 (LoadOp AddrMode:$Rn)))))]>,
1402 Sched<[WriteVLD2]> {
1404 let Inst{4} = Rn{4};
1405 let DecoderMethod = "DecodeVLD1DupInstruction";
1407 def VLD1DUPd8 : VLD1DUP<{0,0,0,?}, "8", v8i8, extloadi8,
1408 addrmode6dupalignNone>;
1409 def VLD1DUPd16 : VLD1DUP<{0,1,0,?}, "16", v4i16, extloadi16,
1410 addrmode6dupalign16>;
1411 def VLD1DUPd32 : VLD1DUP<{1,0,0,?}, "32", v2i32, load,
1412 addrmode6dupalign32>;
1414 def : Pat<(v2f32 (NEONvdup (f32 (load addrmode6dup:$addr)))),
1415 (VLD1DUPd32 addrmode6:$addr)>;
1417 class VLD1QDUP<bits<4> op7_4, string Dt, ValueType Ty, PatFrag LoadOp,
1419 : NLdSt<1, 0b10, 0b1100, op7_4, (outs VecListDPairAllLanes:$Vd),
1420 (ins AddrMode:$Rn), IIC_VLD1dup,
1421 "vld1", Dt, "$Vd, $Rn", "",
1422 [(set VecListDPairAllLanes:$Vd,
1423 (Ty (NEONvdup (i32 (LoadOp AddrMode:$Rn)))))]> {
1425 let Inst{4} = Rn{4};
1426 let DecoderMethod = "DecodeVLD1DupInstruction";
1429 def VLD1DUPq8 : VLD1QDUP<{0,0,1,0}, "8", v16i8, extloadi8,
1430 addrmode6dupalignNone>;
1431 def VLD1DUPq16 : VLD1QDUP<{0,1,1,?}, "16", v8i16, extloadi16,
1432 addrmode6dupalign16>;
1433 def VLD1DUPq32 : VLD1QDUP<{1,0,1,?}, "32", v4i32, load,
1434 addrmode6dupalign32>;
1436 def : Pat<(v4f32 (NEONvdup (f32 (load addrmode6dup:$addr)))),
1437 (VLD1DUPq32 addrmode6:$addr)>;
1439 let mayLoad = 1, hasSideEffects = 0, hasExtraDefRegAllocReq = 1 in {
1440 // ...with address register writeback:
1441 multiclass VLD1DUPWB<bits<4> op7_4, string Dt, Operand AddrMode> {
1442 def _fixed : NLdSt<1, 0b10, 0b1100, op7_4,
1443 (outs VecListOneDAllLanes:$Vd, GPR:$wb),
1444 (ins AddrMode:$Rn), IIC_VLD1dupu,
1445 "vld1", Dt, "$Vd, $Rn!",
1446 "$Rn.addr = $wb", []> {
1447 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1448 let Inst{4} = Rn{4};
1449 let DecoderMethod = "DecodeVLD1DupInstruction";
1451 def _register : NLdSt<1, 0b10, 0b1100, op7_4,
1452 (outs VecListOneDAllLanes:$Vd, GPR:$wb),
1453 (ins AddrMode:$Rn, rGPR:$Rm), IIC_VLD1dupu,
1454 "vld1", Dt, "$Vd, $Rn, $Rm",
1455 "$Rn.addr = $wb", []> {
1456 let Inst{4} = Rn{4};
1457 let DecoderMethod = "DecodeVLD1DupInstruction";
1460 multiclass VLD1QDUPWB<bits<4> op7_4, string Dt, Operand AddrMode> {
1461 def _fixed : NLdSt<1, 0b10, 0b1100, op7_4,
1462 (outs VecListDPairAllLanes:$Vd, GPR:$wb),
1463 (ins AddrMode:$Rn), IIC_VLD1dupu,
1464 "vld1", Dt, "$Vd, $Rn!",
1465 "$Rn.addr = $wb", []>, Sched<[WriteVLD1]> {
1466 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1467 let Inst{4} = Rn{4};
1468 let DecoderMethod = "DecodeVLD1DupInstruction";
1470 def _register : NLdSt<1, 0b10, 0b1100, op7_4,
1471 (outs VecListDPairAllLanes:$Vd, GPR:$wb),
1472 (ins AddrMode:$Rn, rGPR:$Rm), IIC_VLD1dupu,
1473 "vld1", Dt, "$Vd, $Rn, $Rm",
1474 "$Rn.addr = $wb", []> {
1475 let Inst{4} = Rn{4};
1476 let DecoderMethod = "DecodeVLD1DupInstruction";
1480 defm VLD1DUPd8wb : VLD1DUPWB<{0,0,0,0}, "8", addrmode6dupalignNone>;
1481 defm VLD1DUPd16wb : VLD1DUPWB<{0,1,0,?}, "16", addrmode6dupalign16>;
1482 defm VLD1DUPd32wb : VLD1DUPWB<{1,0,0,?}, "32", addrmode6dupalign32>;
1484 defm VLD1DUPq8wb : VLD1QDUPWB<{0,0,1,0}, "8", addrmode6dupalignNone>;
1485 defm VLD1DUPq16wb : VLD1QDUPWB<{0,1,1,?}, "16", addrmode6dupalign16>;
1486 defm VLD1DUPq32wb : VLD1QDUPWB<{1,0,1,?}, "32", addrmode6dupalign32>;
1488 // VLD2DUP : Vector Load (single 2-element structure to all lanes)
1489 class VLD2DUP<bits<4> op7_4, string Dt, RegisterOperand VdTy, Operand AddrMode>
1490 : NLdSt<1, 0b10, 0b1101, op7_4, (outs VdTy:$Vd),
1491 (ins AddrMode:$Rn), IIC_VLD2dup,
1492 "vld2", Dt, "$Vd, $Rn", "", []> {
1494 let Inst{4} = Rn{4};
1495 let DecoderMethod = "DecodeVLD2DupInstruction";
1498 def VLD2DUPd8 : VLD2DUP<{0,0,0,?}, "8", VecListDPairAllLanes,
1499 addrmode6dupalign16>;
1500 def VLD2DUPd16 : VLD2DUP<{0,1,0,?}, "16", VecListDPairAllLanes,
1501 addrmode6dupalign32>;
1502 def VLD2DUPd32 : VLD2DUP<{1,0,0,?}, "32", VecListDPairAllLanes,
1503 addrmode6dupalign64>;
1505 // HACK this one, VLD2DUPd8x2 must be changed at the same time with VLD2b8 or
1506 // "vld2.8 {d0[], d2[]}, [r4:32]" will become "vld2.8 {d0, d2}, [r4:32]".
1507 // ...with double-spaced registers
1508 def VLD2DUPd8x2 : VLD2DUP<{0,0,1,?}, "8", VecListDPairSpacedAllLanes,
1509 addrmode6dupalign16>;
1510 def VLD2DUPd16x2 : VLD2DUP<{0,1,1,?}, "16", VecListDPairSpacedAllLanes,
1511 addrmode6dupalign32>;
1512 def VLD2DUPd32x2 : VLD2DUP<{1,0,1,?}, "32", VecListDPairSpacedAllLanes,
1513 addrmode6dupalign64>;
1515 // ...with address register writeback:
1516 multiclass VLD2DUPWB<bits<4> op7_4, string Dt, RegisterOperand VdTy,
1518 def _fixed : NLdSt<1, 0b10, 0b1101, op7_4,
1519 (outs VdTy:$Vd, GPR:$wb),
1520 (ins AddrMode:$Rn), IIC_VLD2dupu,
1521 "vld2", Dt, "$Vd, $Rn!",
1522 "$Rn.addr = $wb", []>, Sched<[WriteVLD1]> {
1523 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1524 let Inst{4} = Rn{4};
1525 let DecoderMethod = "DecodeVLD2DupInstruction";
1527 def _register : NLdSt<1, 0b10, 0b1101, op7_4,
1528 (outs VdTy:$Vd, GPR:$wb),
1529 (ins AddrMode:$Rn, rGPR:$Rm), IIC_VLD2dupu,
1530 "vld2", Dt, "$Vd, $Rn, $Rm",
1531 "$Rn.addr = $wb", []>, Sched<[WriteVLD1]> {
1532 let Inst{4} = Rn{4};
1533 let DecoderMethod = "DecodeVLD2DupInstruction";
1537 defm VLD2DUPd8wb : VLD2DUPWB<{0,0,0,0}, "8", VecListDPairAllLanes,
1538 addrmode6dupalign16>;
1539 defm VLD2DUPd16wb : VLD2DUPWB<{0,1,0,?}, "16", VecListDPairAllLanes,
1540 addrmode6dupalign32>;
1541 defm VLD2DUPd32wb : VLD2DUPWB<{1,0,0,?}, "32", VecListDPairAllLanes,
1542 addrmode6dupalign64>;
1544 defm VLD2DUPd8x2wb : VLD2DUPWB<{0,0,1,0}, "8", VecListDPairSpacedAllLanes,
1545 addrmode6dupalign16>;
1546 defm VLD2DUPd16x2wb : VLD2DUPWB<{0,1,1,?}, "16", VecListDPairSpacedAllLanes,
1547 addrmode6dupalign32>;
1548 defm VLD2DUPd32x2wb : VLD2DUPWB<{1,0,1,?}, "32", VecListDPairSpacedAllLanes,
1549 addrmode6dupalign64>;
1551 // VLD3DUP : Vector Load (single 3-element structure to all lanes)
1552 class VLD3DUP<bits<4> op7_4, string Dt>
1553 : NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
1554 (ins addrmode6dup:$Rn), IIC_VLD3dup,
1555 "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn", "", []>,
1556 Sched<[WriteVLD2]> {
1559 let DecoderMethod = "DecodeVLD3DupInstruction";
1562 def VLD3DUPd8 : VLD3DUP<{0,0,0,?}, "8">;
1563 def VLD3DUPd16 : VLD3DUP<{0,1,0,?}, "16">;
1564 def VLD3DUPd32 : VLD3DUP<{1,0,0,?}, "32">;
1566 def VLD3DUPd8Pseudo : VLDQQPseudo<IIC_VLD3dup>, Sched<[WriteVLD2]>;
1567 def VLD3DUPd16Pseudo : VLDQQPseudo<IIC_VLD3dup>, Sched<[WriteVLD2]>;
1568 def VLD3DUPd32Pseudo : VLDQQPseudo<IIC_VLD3dup>, Sched<[WriteVLD2]>;
1570 // ...with double-spaced registers (not used for codegen):
1571 def VLD3DUPq8 : VLD3DUP<{0,0,1,?}, "8">;
1572 def VLD3DUPq16 : VLD3DUP<{0,1,1,?}, "16">;
1573 def VLD3DUPq32 : VLD3DUP<{1,0,1,?}, "32">;
1575 // ...with address register writeback:
1576 class VLD3DUPWB<bits<4> op7_4, string Dt, Operand AddrMode>
1577 : NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
1578 (ins AddrMode:$Rn, am6offset:$Rm), IIC_VLD3dupu,
1579 "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn$Rm",
1580 "$Rn.addr = $wb", []>, Sched<[WriteVLD2]> {
1582 let DecoderMethod = "DecodeVLD3DupInstruction";
1585 def VLD3DUPd8_UPD : VLD3DUPWB<{0,0,0,0}, "8", addrmode6dupalign64>;
1586 def VLD3DUPd16_UPD : VLD3DUPWB<{0,1,0,?}, "16", addrmode6dupalign64>;
1587 def VLD3DUPd32_UPD : VLD3DUPWB<{1,0,0,?}, "32", addrmode6dupalign64>;
1589 def VLD3DUPq8_UPD : VLD3DUPWB<{0,0,1,0}, "8", addrmode6dupalign64>;
1590 def VLD3DUPq16_UPD : VLD3DUPWB<{0,1,1,?}, "16", addrmode6dupalign64>;
1591 def VLD3DUPq32_UPD : VLD3DUPWB<{1,0,1,?}, "32", addrmode6dupalign64>;
1593 def VLD3DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>, Sched<[WriteVLD2]>;
1594 def VLD3DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>, Sched<[WriteVLD2]>;
1595 def VLD3DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>, Sched<[WriteVLD2]>;
1597 // VLD4DUP : Vector Load (single 4-element structure to all lanes)
1598 class VLD4DUP<bits<4> op7_4, string Dt>
1599 : NLdSt<1, 0b10, 0b1111, op7_4,
1600 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
1601 (ins addrmode6dup:$Rn), IIC_VLD4dup,
1602 "vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn", "", []> {
1604 let Inst{4} = Rn{4};
1605 let DecoderMethod = "DecodeVLD4DupInstruction";
1608 def VLD4DUPd8 : VLD4DUP<{0,0,0,?}, "8">;
1609 def VLD4DUPd16 : VLD4DUP<{0,1,0,?}, "16">;
1610 def VLD4DUPd32 : VLD4DUP<{1,?,0,?}, "32"> { let Inst{6} = Rn{5}; }
1612 def VLD4DUPd8Pseudo : VLDQQPseudo<IIC_VLD4dup>, Sched<[WriteVLD2]>;
1613 def VLD4DUPd16Pseudo : VLDQQPseudo<IIC_VLD4dup>, Sched<[WriteVLD2]>;
1614 def VLD4DUPd32Pseudo : VLDQQPseudo<IIC_VLD4dup>, Sched<[WriteVLD2]>;
1616 // ...with double-spaced registers (not used for codegen):
1617 def VLD4DUPq8 : VLD4DUP<{0,0,1,?}, "8">;
1618 def VLD4DUPq16 : VLD4DUP<{0,1,1,?}, "16">;
1619 def VLD4DUPq32 : VLD4DUP<{1,?,1,?}, "32"> { let Inst{6} = Rn{5}; }
1621 // ...with address register writeback:
1622 class VLD4DUPWB<bits<4> op7_4, string Dt>
1623 : NLdSt<1, 0b10, 0b1111, op7_4,
1624 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
1625 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD4dupu,
1626 "vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn$Rm",
1627 "$Rn.addr = $wb", []>, Sched<[WriteVLD2]> {
1628 let Inst{4} = Rn{4};
1629 let DecoderMethod = "DecodeVLD4DupInstruction";
1632 def VLD4DUPd8_UPD : VLD4DUPWB<{0,0,0,0}, "8">;
1633 def VLD4DUPd16_UPD : VLD4DUPWB<{0,1,0,?}, "16">;
1634 def VLD4DUPd32_UPD : VLD4DUPWB<{1,?,0,?}, "32"> { let Inst{6} = Rn{5}; }
1636 def VLD4DUPq8_UPD : VLD4DUPWB<{0,0,1,0}, "8">;
1637 def VLD4DUPq16_UPD : VLD4DUPWB<{0,1,1,?}, "16">;
1638 def VLD4DUPq32_UPD : VLD4DUPWB<{1,?,1,?}, "32"> { let Inst{6} = Rn{5}; }
1640 def VLD4DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>, Sched<[WriteVLD2]>;
1641 def VLD4DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>, Sched<[WriteVLD2]>;
1642 def VLD4DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>, Sched<[WriteVLD2]>;
1644 } // mayLoad = 1, hasSideEffects = 0, hasExtraDefRegAllocReq = 1
1646 let mayStore = 1, hasSideEffects = 0, hasExtraSrcRegAllocReq = 1 in {
1648 // Classes for VST* pseudo-instructions with multi-register operands.
1649 // These are expanded to real instructions after register allocation.
1650 class VSTQPseudo<InstrItinClass itin>
1651 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src), itin, "">;
1652 class VSTQWBPseudo<InstrItinClass itin>
1653 : PseudoNLdSt<(outs GPR:$wb),
1654 (ins addrmode6:$addr, am6offset:$offset, QPR:$src), itin,
1655 "$addr.addr = $wb">;
1656 class VSTQWBfixedPseudo<InstrItinClass itin>
1657 : PseudoNLdSt<(outs GPR:$wb),
1658 (ins addrmode6:$addr, QPR:$src), itin,
1659 "$addr.addr = $wb">;
1660 class VSTQWBregisterPseudo<InstrItinClass itin>
1661 : PseudoNLdSt<(outs GPR:$wb),
1662 (ins addrmode6:$addr, rGPR:$offset, QPR:$src), itin,
1663 "$addr.addr = $wb">;
1664 class VSTQQPseudo<InstrItinClass itin>
1665 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src), itin, "">;
1666 class VSTQQWBPseudo<InstrItinClass itin>
1667 : PseudoNLdSt<(outs GPR:$wb),
1668 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src), itin,
1669 "$addr.addr = $wb">;
1670 class VSTQQWBfixedPseudo<InstrItinClass itin>
1671 : PseudoNLdSt<(outs GPR:$wb),
1672 (ins addrmode6:$addr, QQPR:$src), itin,
1673 "$addr.addr = $wb">;
1674 class VSTQQWBregisterPseudo<InstrItinClass itin>
1675 : PseudoNLdSt<(outs GPR:$wb),
1676 (ins addrmode6:$addr, rGPR:$offset, QQPR:$src), itin,
1677 "$addr.addr = $wb">;
1679 class VSTQQQQPseudo<InstrItinClass itin>
1680 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src), itin, "">;
1681 class VSTQQQQWBPseudo<InstrItinClass itin>
1682 : PseudoNLdSt<(outs GPR:$wb),
1683 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
1684 "$addr.addr = $wb">;
1686 // VST1 : Vector Store (multiple single elements)
1687 class VST1D<bits<4> op7_4, string Dt, Operand AddrMode>
1688 : NLdSt<0,0b00,0b0111,op7_4, (outs), (ins AddrMode:$Rn, VecListOneD:$Vd),
1689 IIC_VST1, "vst1", Dt, "$Vd, $Rn", "", []>, Sched<[WriteVST1]> {
1691 let Inst{4} = Rn{4};
1692 let DecoderMethod = "DecodeVLDST1Instruction";
1694 class VST1Q<bits<4> op7_4, string Dt, Operand AddrMode>
1695 : NLdSt<0,0b00,0b1010,op7_4, (outs), (ins AddrMode:$Rn, VecListDPair:$Vd),
1696 IIC_VST1x2, "vst1", Dt, "$Vd, $Rn", "", []>, Sched<[WriteVST2]> {
1698 let Inst{5-4} = Rn{5-4};
1699 let DecoderMethod = "DecodeVLDST1Instruction";
1702 def VST1d8 : VST1D<{0,0,0,?}, "8", addrmode6align64>;
1703 def VST1d16 : VST1D<{0,1,0,?}, "16", addrmode6align64>;
1704 def VST1d32 : VST1D<{1,0,0,?}, "32", addrmode6align64>;
1705 def VST1d64 : VST1D<{1,1,0,?}, "64", addrmode6align64>;
1707 def VST1q8 : VST1Q<{0,0,?,?}, "8", addrmode6align64or128>;
1708 def VST1q16 : VST1Q<{0,1,?,?}, "16", addrmode6align64or128>;
1709 def VST1q32 : VST1Q<{1,0,?,?}, "32", addrmode6align64or128>;
1710 def VST1q64 : VST1Q<{1,1,?,?}, "64", addrmode6align64or128>;
1712 // ...with address register writeback:
1713 multiclass VST1DWB<bits<4> op7_4, string Dt, Operand AddrMode> {
1714 def _fixed : NLdSt<0,0b00, 0b0111,op7_4, (outs GPR:$wb),
1715 (ins AddrMode:$Rn, VecListOneD:$Vd), IIC_VLD1u,
1716 "vst1", Dt, "$Vd, $Rn!",
1717 "$Rn.addr = $wb", []>, Sched<[WriteVST1]> {
1718 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1719 let Inst{4} = Rn{4};
1720 let DecoderMethod = "DecodeVLDST1Instruction";
1722 def _register : NLdSt<0,0b00,0b0111,op7_4, (outs GPR:$wb),
1723 (ins AddrMode:$Rn, rGPR:$Rm, VecListOneD:$Vd),
1725 "vst1", Dt, "$Vd, $Rn, $Rm",
1726 "$Rn.addr = $wb", []>, Sched<[WriteVST1]> {
1727 let Inst{4} = Rn{4};
1728 let DecoderMethod = "DecodeVLDST1Instruction";
1731 multiclass VST1QWB<bits<4> op7_4, string Dt, Operand AddrMode> {
1732 def _fixed : NLdSt<0,0b00,0b1010,op7_4, (outs GPR:$wb),
1733 (ins AddrMode:$Rn, VecListDPair:$Vd), IIC_VLD1x2u,
1734 "vst1", Dt, "$Vd, $Rn!",
1735 "$Rn.addr = $wb", []>, Sched<[WriteVST2]> {
1736 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1737 let Inst{5-4} = Rn{5-4};
1738 let DecoderMethod = "DecodeVLDST1Instruction";
1740 def _register : NLdSt<0,0b00,0b1010,op7_4, (outs GPR:$wb),
1741 (ins AddrMode:$Rn, rGPR:$Rm, VecListDPair:$Vd),
1743 "vst1", Dt, "$Vd, $Rn, $Rm",
1744 "$Rn.addr = $wb", []>, Sched<[WriteVST2]> {
1745 let Inst{5-4} = Rn{5-4};
1746 let DecoderMethod = "DecodeVLDST1Instruction";
1750 defm VST1d8wb : VST1DWB<{0,0,0,?}, "8", addrmode6align64>;
1751 defm VST1d16wb : VST1DWB<{0,1,0,?}, "16", addrmode6align64>;
1752 defm VST1d32wb : VST1DWB<{1,0,0,?}, "32", addrmode6align64>;
1753 defm VST1d64wb : VST1DWB<{1,1,0,?}, "64", addrmode6align64>;
1755 defm VST1q8wb : VST1QWB<{0,0,?,?}, "8", addrmode6align64or128>;
1756 defm VST1q16wb : VST1QWB<{0,1,?,?}, "16", addrmode6align64or128>;
1757 defm VST1q32wb : VST1QWB<{1,0,?,?}, "32", addrmode6align64or128>;
1758 defm VST1q64wb : VST1QWB<{1,1,?,?}, "64", addrmode6align64or128>;
1760 // ...with 3 registers
1761 class VST1D3<bits<4> op7_4, string Dt, Operand AddrMode>
1762 : NLdSt<0, 0b00, 0b0110, op7_4, (outs),
1763 (ins AddrMode:$Rn, VecListThreeD:$Vd),
1764 IIC_VST1x3, "vst1", Dt, "$Vd, $Rn", "", []>, Sched<[WriteVST3]> {
1766 let Inst{4} = Rn{4};
1767 let DecoderMethod = "DecodeVLDST1Instruction";
1769 multiclass VST1D3WB<bits<4> op7_4, string Dt, Operand AddrMode> {
1770 def _fixed : NLdSt<0,0b00,0b0110,op7_4, (outs GPR:$wb),
1771 (ins AddrMode:$Rn, VecListThreeD:$Vd), IIC_VLD1x3u,
1772 "vst1", Dt, "$Vd, $Rn!",
1773 "$Rn.addr = $wb", []>, Sched<[WriteVST3]> {
1774 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1775 let Inst{5-4} = Rn{5-4};
1776 let DecoderMethod = "DecodeVLDST1Instruction";
1778 def _register : NLdSt<0,0b00,0b0110,op7_4, (outs GPR:$wb),
1779 (ins AddrMode:$Rn, rGPR:$Rm, VecListThreeD:$Vd),
1781 "vst1", Dt, "$Vd, $Rn, $Rm",
1782 "$Rn.addr = $wb", []>, Sched<[WriteVST3]> {
1783 let Inst{5-4} = Rn{5-4};
1784 let DecoderMethod = "DecodeVLDST1Instruction";
1788 def VST1d8T : VST1D3<{0,0,0,?}, "8", addrmode6align64>;
1789 def VST1d16T : VST1D3<{0,1,0,?}, "16", addrmode6align64>;
1790 def VST1d32T : VST1D3<{1,0,0,?}, "32", addrmode6align64>;
1791 def VST1d64T : VST1D3<{1,1,0,?}, "64", addrmode6align64>;
1793 defm VST1d8Twb : VST1D3WB<{0,0,0,?}, "8", addrmode6align64>;
1794 defm VST1d16Twb : VST1D3WB<{0,1,0,?}, "16", addrmode6align64>;
1795 defm VST1d32Twb : VST1D3WB<{1,0,0,?}, "32", addrmode6align64>;
1796 defm VST1d64Twb : VST1D3WB<{1,1,0,?}, "64", addrmode6align64>;
1798 def VST1d64TPseudo : VSTQQPseudo<IIC_VST1x3>, Sched<[WriteVST3]>;
1799 def VST1d64TPseudoWB_fixed : VSTQQWBfixedPseudo<IIC_VST1x3u>, Sched<[WriteVST3]>;
1800 def VST1d64TPseudoWB_register : VSTQQWBPseudo<IIC_VST1x3u>, Sched<[WriteVST3]>;
1802 // ...with 4 registers
1803 class VST1D4<bits<4> op7_4, string Dt, Operand AddrMode>
1804 : NLdSt<0, 0b00, 0b0010, op7_4, (outs),
1805 (ins AddrMode:$Rn, VecListFourD:$Vd),
1806 IIC_VST1x4, "vst1", Dt, "$Vd, $Rn", "",
1807 []>, Sched<[WriteVST4]> {
1809 let Inst{5-4} = Rn{5-4};
1810 let DecoderMethod = "DecodeVLDST1Instruction";
1812 multiclass VST1D4WB<bits<4> op7_4, string Dt, Operand AddrMode> {
1813 def _fixed : NLdSt<0,0b00,0b0010,op7_4, (outs GPR:$wb),
1814 (ins AddrMode:$Rn, VecListFourD:$Vd), IIC_VLD1x4u,
1815 "vst1", Dt, "$Vd, $Rn!",
1816 "$Rn.addr = $wb", []>, Sched<[WriteVST4]> {
1817 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1818 let Inst{5-4} = Rn{5-4};
1819 let DecoderMethod = "DecodeVLDST1Instruction";
1821 def _register : NLdSt<0,0b00,0b0010,op7_4, (outs GPR:$wb),
1822 (ins AddrMode:$Rn, rGPR:$Rm, VecListFourD:$Vd),
1824 "vst1", Dt, "$Vd, $Rn, $Rm",
1825 "$Rn.addr = $wb", []>, Sched<[WriteVST4]> {
1826 let Inst{5-4} = Rn{5-4};
1827 let DecoderMethod = "DecodeVLDST1Instruction";
1831 def VST1d8Q : VST1D4<{0,0,?,?}, "8", addrmode6align64or128or256>;
1832 def VST1d16Q : VST1D4<{0,1,?,?}, "16", addrmode6align64or128or256>;
1833 def VST1d32Q : VST1D4<{1,0,?,?}, "32", addrmode6align64or128or256>;
1834 def VST1d64Q : VST1D4<{1,1,?,?}, "64", addrmode6align64or128or256>;
1836 defm VST1d8Qwb : VST1D4WB<{0,0,?,?}, "8", addrmode6align64or128or256>;
1837 defm VST1d16Qwb : VST1D4WB<{0,1,?,?}, "16", addrmode6align64or128or256>;
1838 defm VST1d32Qwb : VST1D4WB<{1,0,?,?}, "32", addrmode6align64or128or256>;
1839 defm VST1d64Qwb : VST1D4WB<{1,1,?,?}, "64", addrmode6align64or128or256>;
1841 def VST1d64QPseudo : VSTQQPseudo<IIC_VST1x4>, Sched<[WriteVST4]>;
1842 def VST1d64QPseudoWB_fixed : VSTQQWBfixedPseudo<IIC_VST1x4u>, Sched<[WriteVST4]>;
1843 def VST1d64QPseudoWB_register : VSTQQWBPseudo<IIC_VST1x4u>, Sched<[WriteVST4]>;
1845 // VST2 : Vector Store (multiple 2-element structures)
1846 class VST2<bits<4> op11_8, bits<4> op7_4, string Dt, RegisterOperand VdTy,
1847 InstrItinClass itin, Operand AddrMode>
1848 : NLdSt<0, 0b00, op11_8, op7_4, (outs), (ins AddrMode:$Rn, VdTy:$Vd),
1849 itin, "vst2", Dt, "$Vd, $Rn", "", []> {
1851 let Inst{5-4} = Rn{5-4};
1852 let DecoderMethod = "DecodeVLDST2Instruction";
1855 def VST2d8 : VST2<0b1000, {0,0,?,?}, "8", VecListDPair, IIC_VST2,
1856 addrmode6align64or128>, Sched<[WriteVST2]>;
1857 def VST2d16 : VST2<0b1000, {0,1,?,?}, "16", VecListDPair, IIC_VST2,
1858 addrmode6align64or128>, Sched<[WriteVST2]>;
1859 def VST2d32 : VST2<0b1000, {1,0,?,?}, "32", VecListDPair, IIC_VST2,
1860 addrmode6align64or128>, Sched<[WriteVST2]>;
1862 def VST2q8 : VST2<0b0011, {0,0,?,?}, "8", VecListFourD, IIC_VST2x2,
1863 addrmode6align64or128or256>, Sched<[WriteVST4]>;
1864 def VST2q16 : VST2<0b0011, {0,1,?,?}, "16", VecListFourD, IIC_VST2x2,
1865 addrmode6align64or128or256>, Sched<[WriteVST4]>;
1866 def VST2q32 : VST2<0b0011, {1,0,?,?}, "32", VecListFourD, IIC_VST2x2,
1867 addrmode6align64or128or256>, Sched<[WriteVST4]>;
1869 def VST2q8Pseudo : VSTQQPseudo<IIC_VST2x2>, Sched<[WriteVST4]>;
1870 def VST2q16Pseudo : VSTQQPseudo<IIC_VST2x2>, Sched<[WriteVST4]>;
1871 def VST2q32Pseudo : VSTQQPseudo<IIC_VST2x2>, Sched<[WriteVST4]>;
1873 // ...with address register writeback:
1874 multiclass VST2DWB<bits<4> op11_8, bits<4> op7_4, string Dt,
1875 RegisterOperand VdTy, Operand AddrMode> {
1876 def _fixed : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1877 (ins AddrMode:$Rn, VdTy:$Vd), IIC_VLD1u,
1878 "vst2", Dt, "$Vd, $Rn!",
1879 "$Rn.addr = $wb", []>, Sched<[WriteVST2]> {
1880 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1881 let Inst{5-4} = Rn{5-4};
1882 let DecoderMethod = "DecodeVLDST2Instruction";
1884 def _register : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1885 (ins AddrMode:$Rn, rGPR:$Rm, VdTy:$Vd), IIC_VLD1u,
1886 "vst2", Dt, "$Vd, $Rn, $Rm",
1887 "$Rn.addr = $wb", []>, Sched<[WriteVST2]> {
1888 let Inst{5-4} = Rn{5-4};
1889 let DecoderMethod = "DecodeVLDST2Instruction";
1892 multiclass VST2QWB<bits<4> op7_4, string Dt, Operand AddrMode> {
1893 def _fixed : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
1894 (ins AddrMode:$Rn, VecListFourD:$Vd), IIC_VLD1u,
1895 "vst2", Dt, "$Vd, $Rn!",
1896 "$Rn.addr = $wb", []>, Sched<[WriteVST4]> {
1897 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1898 let Inst{5-4} = Rn{5-4};
1899 let DecoderMethod = "DecodeVLDST2Instruction";
1901 def _register : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
1902 (ins AddrMode:$Rn, rGPR:$Rm, VecListFourD:$Vd),
1904 "vst2", Dt, "$Vd, $Rn, $Rm",
1905 "$Rn.addr = $wb", []>, Sched<[WriteVST4]> {
1906 let Inst{5-4} = Rn{5-4};
1907 let DecoderMethod = "DecodeVLDST2Instruction";
1911 defm VST2d8wb : VST2DWB<0b1000, {0,0,?,?}, "8", VecListDPair,
1912 addrmode6align64or128>;
1913 defm VST2d16wb : VST2DWB<0b1000, {0,1,?,?}, "16", VecListDPair,
1914 addrmode6align64or128>;
1915 defm VST2d32wb : VST2DWB<0b1000, {1,0,?,?}, "32", VecListDPair,
1916 addrmode6align64or128>;
1918 defm VST2q8wb : VST2QWB<{0,0,?,?}, "8", addrmode6align64or128or256>;
1919 defm VST2q16wb : VST2QWB<{0,1,?,?}, "16", addrmode6align64or128or256>;
1920 defm VST2q32wb : VST2QWB<{1,0,?,?}, "32", addrmode6align64or128or256>;
1922 def VST2q8PseudoWB_fixed : VSTQQWBfixedPseudo<IIC_VST2x2u>, Sched<[WriteVST4]>;
1923 def VST2q16PseudoWB_fixed : VSTQQWBfixedPseudo<IIC_VST2x2u>, Sched<[WriteVST4]>;
1924 def VST2q32PseudoWB_fixed : VSTQQWBfixedPseudo<IIC_VST2x2u>, Sched<[WriteVST4]>;
1925 def VST2q8PseudoWB_register : VSTQQWBregisterPseudo<IIC_VST2x2u>, Sched<[WriteVST4]>;
1926 def VST2q16PseudoWB_register : VSTQQWBregisterPseudo<IIC_VST2x2u>, Sched<[WriteVST4]>;
1927 def VST2q32PseudoWB_register : VSTQQWBregisterPseudo<IIC_VST2x2u>, Sched<[WriteVST4]>;
1929 // ...with double-spaced registers
1930 def VST2b8 : VST2<0b1001, {0,0,?,?}, "8", VecListDPairSpaced, IIC_VST2,
1931 addrmode6align64or128>;
1932 def VST2b16 : VST2<0b1001, {0,1,?,?}, "16", VecListDPairSpaced, IIC_VST2,
1933 addrmode6align64or128>;
1934 def VST2b32 : VST2<0b1001, {1,0,?,?}, "32", VecListDPairSpaced, IIC_VST2,
1935 addrmode6align64or128>;
1936 defm VST2b8wb : VST2DWB<0b1001, {0,0,?,?}, "8", VecListDPairSpaced,
1937 addrmode6align64or128>;
1938 defm VST2b16wb : VST2DWB<0b1001, {0,1,?,?}, "16", VecListDPairSpaced,
1939 addrmode6align64or128>;
1940 defm VST2b32wb : VST2DWB<0b1001, {1,0,?,?}, "32", VecListDPairSpaced,
1941 addrmode6align64or128>;
1943 // VST3 : Vector Store (multiple 3-element structures)
1944 class VST3D<bits<4> op11_8, bits<4> op7_4, string Dt>
1945 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1946 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3,
1947 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn", "", []>, Sched<[WriteVST3]> {
1949 let Inst{4} = Rn{4};
1950 let DecoderMethod = "DecodeVLDST3Instruction";
1953 def VST3d8 : VST3D<0b0100, {0,0,0,?}, "8">;
1954 def VST3d16 : VST3D<0b0100, {0,1,0,?}, "16">;
1955 def VST3d32 : VST3D<0b0100, {1,0,0,?}, "32">;
1957 def VST3d8Pseudo : VSTQQPseudo<IIC_VST3>, Sched<[WriteVST3]>;
1958 def VST3d16Pseudo : VSTQQPseudo<IIC_VST3>, Sched<[WriteVST3]>;
1959 def VST3d32Pseudo : VSTQQPseudo<IIC_VST3>, Sched<[WriteVST3]>;
1961 // ...with address register writeback:
1962 class VST3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1963 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1964 (ins addrmode6:$Rn, am6offset:$Rm,
1965 DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3u,
1966 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn$Rm",
1967 "$Rn.addr = $wb", []>, Sched<[WriteVST3]> {
1968 let Inst{4} = Rn{4};
1969 let DecoderMethod = "DecodeVLDST3Instruction";
1972 def VST3d8_UPD : VST3DWB<0b0100, {0,0,0,?}, "8">;
1973 def VST3d16_UPD : VST3DWB<0b0100, {0,1,0,?}, "16">;
1974 def VST3d32_UPD : VST3DWB<0b0100, {1,0,0,?}, "32">;
1976 def VST3d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>, Sched<[WriteVST3]>;
1977 def VST3d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>, Sched<[WriteVST3]>;
1978 def VST3d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>, Sched<[WriteVST3]>;
1980 // ...with double-spaced registers:
1981 def VST3q8 : VST3D<0b0101, {0,0,0,?}, "8">;
1982 def VST3q16 : VST3D<0b0101, {0,1,0,?}, "16">;
1983 def VST3q32 : VST3D<0b0101, {1,0,0,?}, "32">;
1984 def VST3q8_UPD : VST3DWB<0b0101, {0,0,0,?}, "8">;
1985 def VST3q16_UPD : VST3DWB<0b0101, {0,1,0,?}, "16">;
1986 def VST3q32_UPD : VST3DWB<0b0101, {1,0,0,?}, "32">;
1988 def VST3q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>, Sched<[WriteVST3]>;
1989 def VST3q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>, Sched<[WriteVST3]>;
1990 def VST3q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>, Sched<[WriteVST3]>;
1992 // ...alternate versions to be allocated odd register numbers:
1993 def VST3q8oddPseudo : VSTQQQQPseudo<IIC_VST3>, Sched<[WriteVST3]>;
1994 def VST3q16oddPseudo : VSTQQQQPseudo<IIC_VST3>, Sched<[WriteVST3]>;
1995 def VST3q32oddPseudo : VSTQQQQPseudo<IIC_VST3>, Sched<[WriteVST3]>;
1997 def VST3q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>, Sched<[WriteVST3]>;
1998 def VST3q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>, Sched<[WriteVST3]>;
1999 def VST3q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>, Sched<[WriteVST3]>;
2001 // VST4 : Vector Store (multiple 4-element structures)
2002 class VST4D<bits<4> op11_8, bits<4> op7_4, string Dt>
2003 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
2004 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
2005 IIC_VST4, "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn",
2006 "", []>, Sched<[WriteVST4]> {
2008 let Inst{5-4} = Rn{5-4};
2009 let DecoderMethod = "DecodeVLDST4Instruction";
2012 def VST4d8 : VST4D<0b0000, {0,0,?,?}, "8">;
2013 def VST4d16 : VST4D<0b0000, {0,1,?,?}, "16">;
2014 def VST4d32 : VST4D<0b0000, {1,0,?,?}, "32">;
2016 def VST4d8Pseudo : VSTQQPseudo<IIC_VST4>, Sched<[WriteVST4]>;
2017 def VST4d16Pseudo : VSTQQPseudo<IIC_VST4>, Sched<[WriteVST4]>;
2018 def VST4d32Pseudo : VSTQQPseudo<IIC_VST4>, Sched<[WriteVST4]>;
2020 // ...with address register writeback:
2021 class VST4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
2022 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
2023 (ins addrmode6:$Rn, am6offset:$Rm,
2024 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST4u,
2025 "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
2026 "$Rn.addr = $wb", []>, Sched<[WriteVST4]> {
2027 let Inst{5-4} = Rn{5-4};
2028 let DecoderMethod = "DecodeVLDST4Instruction";
2031 def VST4d8_UPD : VST4DWB<0b0000, {0,0,?,?}, "8">;
2032 def VST4d16_UPD : VST4DWB<0b0000, {0,1,?,?}, "16">;
2033 def VST4d32_UPD : VST4DWB<0b0000, {1,0,?,?}, "32">;
2035 def VST4d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>, Sched<[WriteVST4]>;
2036 def VST4d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>, Sched<[WriteVST4]>;
2037 def VST4d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>, Sched<[WriteVST4]>;
2039 // ...with double-spaced registers:
2040 def VST4q8 : VST4D<0b0001, {0,0,?,?}, "8">;
2041 def VST4q16 : VST4D<0b0001, {0,1,?,?}, "16">;
2042 def VST4q32 : VST4D<0b0001, {1,0,?,?}, "32">;
2043 def VST4q8_UPD : VST4DWB<0b0001, {0,0,?,?}, "8">;
2044 def VST4q16_UPD : VST4DWB<0b0001, {0,1,?,?}, "16">;
2045 def VST4q32_UPD : VST4DWB<0b0001, {1,0,?,?}, "32">;
2047 def VST4q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>, Sched<[WriteVST4]>;
2048 def VST4q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>, Sched<[WriteVST4]>;
2049 def VST4q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>, Sched<[WriteVST4]>;
2051 // ...alternate versions to be allocated odd register numbers:
2052 def VST4q8oddPseudo : VSTQQQQPseudo<IIC_VST4>, Sched<[WriteVST4]>;
2053 def VST4q16oddPseudo : VSTQQQQPseudo<IIC_VST4>, Sched<[WriteVST4]>;
2054 def VST4q32oddPseudo : VSTQQQQPseudo<IIC_VST4>, Sched<[WriteVST4]>;
2056 def VST4q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>, Sched<[WriteVST4]>;
2057 def VST4q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>, Sched<[WriteVST4]>;
2058 def VST4q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>, Sched<[WriteVST4]>;
2060 } // mayStore = 1, hasSideEffects = 0, hasExtraSrcRegAllocReq = 1
2062 // Classes for VST*LN pseudo-instructions with multi-register operands.
2063 // These are expanded to real instructions after register allocation.
2064 class VSTQLNPseudo<InstrItinClass itin>
2065 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
2067 class VSTQLNWBPseudo<InstrItinClass itin>
2068 : PseudoNLdSt<(outs GPR:$wb),
2069 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
2070 nohash_imm:$lane), itin, "$addr.addr = $wb">;
2071 class VSTQQLNPseudo<InstrItinClass itin>
2072 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
2074 class VSTQQLNWBPseudo<InstrItinClass itin>
2075 : PseudoNLdSt<(outs GPR:$wb),
2076 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
2077 nohash_imm:$lane), itin, "$addr.addr = $wb">;
2078 class VSTQQQQLNPseudo<InstrItinClass itin>
2079 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
2081 class VSTQQQQLNWBPseudo<InstrItinClass itin>
2082 : PseudoNLdSt<(outs GPR:$wb),
2083 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
2084 nohash_imm:$lane), itin, "$addr.addr = $wb">;
2086 // VST1LN : Vector Store (single element from one lane)
2087 class VST1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
2088 PatFrag StoreOp, SDNode ExtractOp, Operand AddrMode>
2089 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
2090 (ins AddrMode:$Rn, DPR:$Vd, nohash_imm:$lane),
2091 IIC_VST1ln, "vst1", Dt, "\\{$Vd[$lane]\\}, $Rn", "",
2092 [(StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane), AddrMode:$Rn)]>,
2093 Sched<[WriteVST1]> {
2095 let DecoderMethod = "DecodeVST1LN";
2097 class VST1QLNPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
2098 : VSTQLNPseudo<IIC_VST1ln>, Sched<[WriteVST1]> {
2099 let Pattern = [(StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
2103 def VST1LNd8 : VST1LN<0b0000, {?,?,?,0}, "8", v8i8, truncstorei8,
2104 NEONvgetlaneu, addrmode6> {
2105 let Inst{7-5} = lane{2-0};
2107 def VST1LNd16 : VST1LN<0b0100, {?,?,0,?}, "16", v4i16, truncstorei16,
2108 NEONvgetlaneu, addrmode6> {
2109 let Inst{7-6} = lane{1-0};
2110 let Inst{4} = Rn{4};
2113 def VST1LNd32 : VST1LN<0b1000, {?,0,?,?}, "32", v2i32, store, extractelt,
2115 let Inst{7} = lane{0};
2116 let Inst{5-4} = Rn{5-4};
2119 def VST1LNq8Pseudo : VST1QLNPseudo<v16i8, truncstorei8, NEONvgetlaneu>;
2120 def VST1LNq16Pseudo : VST1QLNPseudo<v8i16, truncstorei16, NEONvgetlaneu>;
2121 def VST1LNq32Pseudo : VST1QLNPseudo<v4i32, store, extractelt>;
2123 def : Pat<(store (extractelt (v2f32 DPR:$src), imm:$lane), addrmode6:$addr),
2124 (VST1LNd32 addrmode6:$addr, DPR:$src, imm:$lane)>;
2125 def : Pat<(store (extractelt (v4f32 QPR:$src), imm:$lane), addrmode6:$addr),
2126 (VST1LNq32Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
2128 // ...with address register writeback:
2129 class VST1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
2130 PatFrag StoreOp, SDNode ExtractOp, Operand AdrMode>
2131 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
2132 (ins AdrMode:$Rn, am6offset:$Rm,
2133 DPR:$Vd, nohash_imm:$lane), IIC_VST1lnu, "vst1", Dt,
2134 "\\{$Vd[$lane]\\}, $Rn$Rm",
2136 [(set GPR:$wb, (StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane),
2137 AdrMode:$Rn, am6offset:$Rm))]>,
2138 Sched<[WriteVST1]> {
2139 let DecoderMethod = "DecodeVST1LN";
2141 class VST1QLNWBPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
2142 : VSTQLNWBPseudo<IIC_VST1lnu>, Sched<[WriteVST1]> {
2143 let Pattern = [(set GPR:$wb, (StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
2144 addrmode6:$addr, am6offset:$offset))];
2147 def VST1LNd8_UPD : VST1LNWB<0b0000, {?,?,?,0}, "8", v8i8, post_truncsti8,
2148 NEONvgetlaneu, addrmode6> {
2149 let Inst{7-5} = lane{2-0};
2151 def VST1LNd16_UPD : VST1LNWB<0b0100, {?,?,0,?}, "16", v4i16, post_truncsti16,
2152 NEONvgetlaneu, addrmode6> {
2153 let Inst{7-6} = lane{1-0};
2154 let Inst{4} = Rn{4};
2156 def VST1LNd32_UPD : VST1LNWB<0b1000, {?,0,?,?}, "32", v2i32, post_store,
2157 extractelt, addrmode6oneL32> {
2158 let Inst{7} = lane{0};
2159 let Inst{5-4} = Rn{5-4};
2162 def VST1LNq8Pseudo_UPD : VST1QLNWBPseudo<v16i8, post_truncsti8, NEONvgetlaneu>;
2163 def VST1LNq16Pseudo_UPD : VST1QLNWBPseudo<v8i16, post_truncsti16,NEONvgetlaneu>;
2164 def VST1LNq32Pseudo_UPD : VST1QLNWBPseudo<v4i32, post_store, extractelt>;
2166 let mayStore = 1, hasSideEffects = 0, hasExtraSrcRegAllocReq = 1 in {
2168 // VST2LN : Vector Store (single 2-element structure from one lane)
2169 class VST2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
2170 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
2171 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, nohash_imm:$lane),
2172 IIC_VST2ln, "vst2", Dt, "\\{$Vd[$lane], $src2[$lane]\\}, $Rn",
2173 "", []>, Sched<[WriteVST1]> {
2175 let Inst{4} = Rn{4};
2176 let DecoderMethod = "DecodeVST2LN";
2179 def VST2LNd8 : VST2LN<0b0001, {?,?,?,?}, "8"> {
2180 let Inst{7-5} = lane{2-0};
2182 def VST2LNd16 : VST2LN<0b0101, {?,?,0,?}, "16"> {
2183 let Inst{7-6} = lane{1-0};
2185 def VST2LNd32 : VST2LN<0b1001, {?,0,0,?}, "32"> {
2186 let Inst{7} = lane{0};
2189 def VST2LNd8Pseudo : VSTQLNPseudo<IIC_VST2ln>, Sched<[WriteVST1]>;
2190 def VST2LNd16Pseudo : VSTQLNPseudo<IIC_VST2ln>, Sched<[WriteVST1]>;
2191 def VST2LNd32Pseudo : VSTQLNPseudo<IIC_VST2ln>, Sched<[WriteVST1]>;
2193 // ...with double-spaced registers:
2194 def VST2LNq16 : VST2LN<0b0101, {?,?,1,?}, "16"> {
2195 let Inst{7-6} = lane{1-0};
2196 let Inst{4} = Rn{4};
2198 def VST2LNq32 : VST2LN<0b1001, {?,1,0,?}, "32"> {
2199 let Inst{7} = lane{0};
2200 let Inst{4} = Rn{4};
2203 def VST2LNq16Pseudo : VSTQQLNPseudo<IIC_VST2ln>, Sched<[WriteVST1]>;
2204 def VST2LNq32Pseudo : VSTQQLNPseudo<IIC_VST2ln>, Sched<[WriteVST1]>;
2206 // ...with address register writeback:
2207 class VST2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
2208 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
2209 (ins addrmode6:$Rn, am6offset:$Rm,
2210 DPR:$Vd, DPR:$src2, nohash_imm:$lane), IIC_VST2lnu, "vst2", Dt,
2211 "\\{$Vd[$lane], $src2[$lane]\\}, $Rn$Rm",
2212 "$Rn.addr = $wb", []> {
2213 let Inst{4} = Rn{4};
2214 let DecoderMethod = "DecodeVST2LN";
2217 def VST2LNd8_UPD : VST2LNWB<0b0001, {?,?,?,?}, "8"> {
2218 let Inst{7-5} = lane{2-0};
2220 def VST2LNd16_UPD : VST2LNWB<0b0101, {?,?,0,?}, "16"> {
2221 let Inst{7-6} = lane{1-0};
2223 def VST2LNd32_UPD : VST2LNWB<0b1001, {?,0,0,?}, "32"> {
2224 let Inst{7} = lane{0};
2227 def VST2LNd8Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>, Sched<[WriteVST1]>;
2228 def VST2LNd16Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>, Sched<[WriteVST1]>;
2229 def VST2LNd32Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>, Sched<[WriteVST1]>;
2231 def VST2LNq16_UPD : VST2LNWB<0b0101, {?,?,1,?}, "16"> {
2232 let Inst{7-6} = lane{1-0};
2234 def VST2LNq32_UPD : VST2LNWB<0b1001, {?,1,0,?}, "32"> {
2235 let Inst{7} = lane{0};
2238 def VST2LNq16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>, Sched<[WriteVST1]>;
2239 def VST2LNq32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>, Sched<[WriteVST1]>;
2241 // VST3LN : Vector Store (single 3-element structure from one lane)
2242 class VST3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
2243 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
2244 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3,
2245 nohash_imm:$lane), IIC_VST3ln, "vst3", Dt,
2246 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn", "", []>,
2247 Sched<[WriteVST2]> {
2249 let DecoderMethod = "DecodeVST3LN";
2252 def VST3LNd8 : VST3LN<0b0010, {?,?,?,0}, "8"> {
2253 let Inst{7-5} = lane{2-0};
2255 def VST3LNd16 : VST3LN<0b0110, {?,?,0,0}, "16"> {
2256 let Inst{7-6} = lane{1-0};
2258 def VST3LNd32 : VST3LN<0b1010, {?,0,0,0}, "32"> {
2259 let Inst{7} = lane{0};
2262 def VST3LNd8Pseudo : VSTQQLNPseudo<IIC_VST3ln>, Sched<[WriteVST2]>;
2263 def VST3LNd16Pseudo : VSTQQLNPseudo<IIC_VST3ln>, Sched<[WriteVST2]>;
2264 def VST3LNd32Pseudo : VSTQQLNPseudo<IIC_VST3ln>, Sched<[WriteVST2]>;
2266 // ...with double-spaced registers:
2267 def VST3LNq16 : VST3LN<0b0110, {?,?,1,0}, "16"> {
2268 let Inst{7-6} = lane{1-0};
2270 def VST3LNq32 : VST3LN<0b1010, {?,1,0,0}, "32"> {
2271 let Inst{7} = lane{0};
2274 def VST3LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
2275 def VST3LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
2277 // ...with address register writeback:
2278 class VST3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
2279 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
2280 (ins addrmode6:$Rn, am6offset:$Rm,
2281 DPR:$Vd, DPR:$src2, DPR:$src3, nohash_imm:$lane),
2282 IIC_VST3lnu, "vst3", Dt,
2283 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn$Rm",
2284 "$Rn.addr = $wb", []> {
2285 let DecoderMethod = "DecodeVST3LN";
2288 def VST3LNd8_UPD : VST3LNWB<0b0010, {?,?,?,0}, "8"> {
2289 let Inst{7-5} = lane{2-0};
2291 def VST3LNd16_UPD : VST3LNWB<0b0110, {?,?,0,0}, "16"> {
2292 let Inst{7-6} = lane{1-0};
2294 def VST3LNd32_UPD : VST3LNWB<0b1010, {?,0,0,0}, "32"> {
2295 let Inst{7} = lane{0};
2298 def VST3LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>, Sched<[WriteVST2]>;
2299 def VST3LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>, Sched<[WriteVST2]>;
2300 def VST3LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>, Sched<[WriteVST2]>;
2302 def VST3LNq16_UPD : VST3LNWB<0b0110, {?,?,1,0}, "16"> {
2303 let Inst{7-6} = lane{1-0};
2305 def VST3LNq32_UPD : VST3LNWB<0b1010, {?,1,0,0}, "32"> {
2306 let Inst{7} = lane{0};
2309 def VST3LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>, Sched<[WriteVST2]>;
2310 def VST3LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>, Sched<[WriteVST2]>;
2312 // VST4LN : Vector Store (single 4-element structure from one lane)
2313 class VST4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
2314 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
2315 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4,
2316 nohash_imm:$lane), IIC_VST4ln, "vst4", Dt,
2317 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn",
2318 "", []>, Sched<[WriteVST2]> {
2320 let Inst{4} = Rn{4};
2321 let DecoderMethod = "DecodeVST4LN";
2324 def VST4LNd8 : VST4LN<0b0011, {?,?,?,?}, "8"> {
2325 let Inst{7-5} = lane{2-0};
2327 def VST4LNd16 : VST4LN<0b0111, {?,?,0,?}, "16"> {
2328 let Inst{7-6} = lane{1-0};
2330 def VST4LNd32 : VST4LN<0b1011, {?,0,?,?}, "32"> {
2331 let Inst{7} = lane{0};
2332 let Inst{5} = Rn{5};
2335 def VST4LNd8Pseudo : VSTQQLNPseudo<IIC_VST4ln>, Sched<[WriteVST2]>;
2336 def VST4LNd16Pseudo : VSTQQLNPseudo<IIC_VST4ln>, Sched<[WriteVST2]>;
2337 def VST4LNd32Pseudo : VSTQQLNPseudo<IIC_VST4ln>, Sched<[WriteVST2]>;
2339 // ...with double-spaced registers:
2340 def VST4LNq16 : VST4LN<0b0111, {?,?,1,?}, "16"> {
2341 let Inst{7-6} = lane{1-0};
2343 def VST4LNq32 : VST4LN<0b1011, {?,1,?,?}, "32"> {
2344 let Inst{7} = lane{0};
2345 let Inst{5} = Rn{5};
2348 def VST4LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>, Sched<[WriteVST2]>;
2349 def VST4LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>, Sched<[WriteVST2]>;
2351 // ...with address register writeback:
2352 class VST4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
2353 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
2354 (ins addrmode6:$Rn, am6offset:$Rm,
2355 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
2356 IIC_VST4lnu, "vst4", Dt,
2357 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn$Rm",
2358 "$Rn.addr = $wb", []> {
2359 let Inst{4} = Rn{4};
2360 let DecoderMethod = "DecodeVST4LN";
2363 def VST4LNd8_UPD : VST4LNWB<0b0011, {?,?,?,?}, "8"> {
2364 let Inst{7-5} = lane{2-0};
2366 def VST4LNd16_UPD : VST4LNWB<0b0111, {?,?,0,?}, "16"> {
2367 let Inst{7-6} = lane{1-0};
2369 def VST4LNd32_UPD : VST4LNWB<0b1011, {?,0,?,?}, "32"> {
2370 let Inst{7} = lane{0};
2371 let Inst{5} = Rn{5};
2374 def VST4LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>, Sched<[WriteVST2]>;
2375 def VST4LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>, Sched<[WriteVST2]>;
2376 def VST4LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>, Sched<[WriteVST2]>;
2378 def VST4LNq16_UPD : VST4LNWB<0b0111, {?,?,1,?}, "16"> {
2379 let Inst{7-6} = lane{1-0};
2381 def VST4LNq32_UPD : VST4LNWB<0b1011, {?,1,?,?}, "32"> {
2382 let Inst{7} = lane{0};
2383 let Inst{5} = Rn{5};
2386 def VST4LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>, Sched<[WriteVST2]>;
2387 def VST4LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>, Sched<[WriteVST2]>;
2389 } // mayStore = 1, hasSideEffects = 0, hasExtraSrcRegAllocReq = 1
2391 // Use vld1/vst1 for unaligned f64 load / store
2392 def : Pat<(f64 (hword_alignedload addrmode6:$addr)),
2393 (VLD1d16 addrmode6:$addr)>, Requires<[IsLE]>;
2394 def : Pat<(hword_alignedstore (f64 DPR:$value), addrmode6:$addr),
2395 (VST1d16 addrmode6:$addr, DPR:$value)>, Requires<[IsLE]>;
2396 def : Pat<(f64 (byte_alignedload addrmode6:$addr)),
2397 (VLD1d8 addrmode6:$addr)>, Requires<[IsLE]>;
2398 def : Pat<(byte_alignedstore (f64 DPR:$value), addrmode6:$addr),
2399 (VST1d8 addrmode6:$addr, DPR:$value)>, Requires<[IsLE]>;
2400 def : Pat<(f64 (non_word_alignedload addrmode6:$addr)),
2401 (VLD1d64 addrmode6:$addr)>, Requires<[IsBE]>;
2402 def : Pat<(non_word_alignedstore (f64 DPR:$value), addrmode6:$addr),
2403 (VST1d64 addrmode6:$addr, DPR:$value)>, Requires<[IsBE]>;
2405 // Use vld1/vst1 for Q and QQ. Also use them for unaligned v2f64
2406 // load / store if it's legal.
2407 def : Pat<(v2f64 (dword_alignedload addrmode6:$addr)),
2408 (VLD1q64 addrmode6:$addr)>;
2409 def : Pat<(dword_alignedstore (v2f64 QPR:$value), addrmode6:$addr),
2410 (VST1q64 addrmode6:$addr, QPR:$value)>;
2411 def : Pat<(v2f64 (word_alignedload addrmode6:$addr)),
2412 (VLD1q32 addrmode6:$addr)>, Requires<[IsLE]>;
2413 def : Pat<(word_alignedstore (v2f64 QPR:$value), addrmode6:$addr),
2414 (VST1q32 addrmode6:$addr, QPR:$value)>, Requires<[IsLE]>;
2415 def : Pat<(v2f64 (hword_alignedload addrmode6:$addr)),
2416 (VLD1q16 addrmode6:$addr)>, Requires<[IsLE]>;
2417 def : Pat<(hword_alignedstore (v2f64 QPR:$value), addrmode6:$addr),
2418 (VST1q16 addrmode6:$addr, QPR:$value)>, Requires<[IsLE]>;
2419 def : Pat<(v2f64 (byte_alignedload addrmode6:$addr)),
2420 (VLD1q8 addrmode6:$addr)>, Requires<[IsLE]>;
2421 def : Pat<(byte_alignedstore (v2f64 QPR:$value), addrmode6:$addr),
2422 (VST1q8 addrmode6:$addr, QPR:$value)>, Requires<[IsLE]>;
2424 //===----------------------------------------------------------------------===//
2425 // NEON pattern fragments
2426 //===----------------------------------------------------------------------===//
2428 // Extract D sub-registers of Q registers.
2429 def DSubReg_i8_reg : SDNodeXForm<imm, [{
2430 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2431 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/8, SDLoc(N),
2434 def DSubReg_i16_reg : SDNodeXForm<imm, [{
2435 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2436 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/4, SDLoc(N),
2439 def DSubReg_i32_reg : SDNodeXForm<imm, [{
2440 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2441 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/2, SDLoc(N),
2444 def DSubReg_f64_reg : SDNodeXForm<imm, [{
2445 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2446 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue(), SDLoc(N),
2450 // Extract S sub-registers of Q/D registers.
2451 def SSubReg_f32_reg : SDNodeXForm<imm, [{
2452 assert(ARM::ssub_3 == ARM::ssub_0+3 && "Unexpected subreg numbering");
2453 return CurDAG->getTargetConstant(ARM::ssub_0 + N->getZExtValue(), SDLoc(N),
2457 // Translate lane numbers from Q registers to D subregs.
2458 def SubReg_i8_lane : SDNodeXForm<imm, [{
2459 return CurDAG->getTargetConstant(N->getZExtValue() & 7, SDLoc(N), MVT::i32);
2461 def SubReg_i16_lane : SDNodeXForm<imm, [{
2462 return CurDAG->getTargetConstant(N->getZExtValue() & 3, SDLoc(N), MVT::i32);
2464 def SubReg_i32_lane : SDNodeXForm<imm, [{
2465 return CurDAG->getTargetConstant(N->getZExtValue() & 1, SDLoc(N), MVT::i32);
2468 //===----------------------------------------------------------------------===//
2469 // Instruction Classes
2470 //===----------------------------------------------------------------------===//
2472 // Basic 2-register operations: double- and quad-register.
2473 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2474 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
2475 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
2476 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2477 (ins DPR:$Vm), IIC_VUNAD, OpcodeStr, Dt,"$Vd, $Vm", "",
2478 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm))))]>;
2479 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2480 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
2481 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
2482 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2483 (ins QPR:$Vm), IIC_VUNAQ, OpcodeStr, Dt,"$Vd, $Vm", "",
2484 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vm))))]>;
2486 // Basic 2-register intrinsics, both double- and quad-register.
2487 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2488 bits<2> op17_16, bits<5> op11_7, bit op4,
2489 InstrItinClass itin, string OpcodeStr, string Dt,
2490 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2491 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2492 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2493 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
2494 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2495 bits<2> op17_16, bits<5> op11_7, bit op4,
2496 InstrItinClass itin, string OpcodeStr, string Dt,
2497 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2498 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2499 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2500 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
2502 // Same as above, but not predicated.
2503 class N2VDIntnp<bits<2> op19_18, bits<2> op17_16, bits<3> op10_8, bit op7,
2504 InstrItinClass itin, string OpcodeStr, string Dt,
2505 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2506 : N2Vnp<op19_18, op17_16, op10_8, op7, 0, (outs DPR:$Vd), (ins DPR:$Vm),
2507 itin, OpcodeStr, Dt,
2508 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
2510 class N2VQIntnp<bits<2> op19_18, bits<2> op17_16, bits<3> op10_8, bit op7,
2511 InstrItinClass itin, string OpcodeStr, string Dt,
2512 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2513 : N2Vnp<op19_18, op17_16, op10_8, op7, 1, (outs QPR:$Vd), (ins QPR:$Vm),
2514 itin, OpcodeStr, Dt,
2515 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
2517 // Similar to NV2VQIntnp with some more encoding bits exposed (crypto).
2518 class N2VQIntXnp<bits<2> op19_18, bits<2> op17_16, bits<3> op10_8, bit op6,
2519 bit op7, InstrItinClass itin, string OpcodeStr, string Dt,
2520 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2521 : N2Vnp<op19_18, op17_16, op10_8, op7, op6, (outs QPR:$Vd), (ins QPR:$Vm),
2522 itin, OpcodeStr, Dt,
2523 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
2525 // Same as N2VQIntXnp but with Vd as a src register.
2526 class N2VQIntX2np<bits<2> op19_18, bits<2> op17_16, bits<3> op10_8, bit op6,
2527 bit op7, InstrItinClass itin, string OpcodeStr, string Dt,
2528 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2529 : N2Vnp<op19_18, op17_16, op10_8, op7, op6,
2530 (outs QPR:$Vd), (ins QPR:$src, QPR:$Vm),
2531 itin, OpcodeStr, Dt,
2532 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$src), (OpTy QPR:$Vm))))]> {
2533 let Constraints = "$src = $Vd";
2536 // Narrow 2-register operations.
2537 class N2VN<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2538 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2539 InstrItinClass itin, string OpcodeStr, string Dt,
2540 ValueType TyD, ValueType TyQ, SDNode OpNode>
2541 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$Vd),
2542 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2543 [(set DPR:$Vd, (TyD (OpNode (TyQ QPR:$Vm))))]>;
2545 // Narrow 2-register intrinsics.
2546 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2547 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2548 InstrItinClass itin, string OpcodeStr, string Dt,
2549 ValueType TyD, ValueType TyQ, SDPatternOperator IntOp>
2550 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$Vd),
2551 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2552 [(set DPR:$Vd, (TyD (IntOp (TyQ QPR:$Vm))))]>;
2554 // Long 2-register operations (currently only used for VMOVL).
2555 class N2VL<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2556 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2557 InstrItinClass itin, string OpcodeStr, string Dt,
2558 ValueType TyQ, ValueType TyD, SDNode OpNode>
2559 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$Vd),
2560 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2561 [(set QPR:$Vd, (TyQ (OpNode (TyD DPR:$Vm))))]>;
2563 // Long 2-register intrinsics.
2564 class N2VLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2565 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2566 InstrItinClass itin, string OpcodeStr, string Dt,
2567 ValueType TyQ, ValueType TyD, SDPatternOperator IntOp>
2568 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$Vd),
2569 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2570 [(set QPR:$Vd, (TyQ (IntOp (TyD DPR:$Vm))))]>;
2572 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
2573 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
2574 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$Vd, DPR:$Vm),
2575 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
2576 OpcodeStr, Dt, "$Vd, $Vm",
2577 "$src1 = $Vd, $src2 = $Vm", []>;
2578 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
2579 InstrItinClass itin, string OpcodeStr, string Dt>
2580 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$Vd, QPR:$Vm),
2581 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$Vd, $Vm",
2582 "$src1 = $Vd, $src2 = $Vm", []>;
2584 // Basic 3-register operations: double- and quad-register.
2585 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2586 InstrItinClass itin, string OpcodeStr, string Dt,
2587 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2588 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2589 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2590 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2591 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
2592 // All of these have a two-operand InstAlias.
2593 let TwoOperandAliasConstraint = "$Vn = $Vd";
2594 let isCommutable = Commutable;
2596 // Same as N3VD but no data type.
2597 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2598 InstrItinClass itin, string OpcodeStr,
2599 ValueType ResTy, ValueType OpTy,
2600 SDNode OpNode, bit Commutable>
2601 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
2602 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2603 OpcodeStr, "$Vd, $Vn, $Vm", "",
2604 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>{
2605 // All of these have a two-operand InstAlias.
2606 let TwoOperandAliasConstraint = "$Vn = $Vd";
2607 let isCommutable = Commutable;
2610 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
2611 InstrItinClass itin, string OpcodeStr, string Dt,
2612 ValueType Ty, SDNode ShOp>
2613 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2614 (outs DPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2615 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2617 (Ty (ShOp (Ty DPR:$Vn),
2618 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),imm:$lane)))))]> {
2619 // All of these have a two-operand InstAlias.
2620 let TwoOperandAliasConstraint = "$Vn = $Vd";
2621 let isCommutable = 0;
2623 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
2624 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
2625 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2626 (outs DPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2627 NVMulSLFrm, IIC_VMULi16D, OpcodeStr, Dt,"$Vd, $Vn, $Vm$lane","",
2629 (Ty (ShOp (Ty DPR:$Vn),
2630 (Ty (NEONvduplane (Ty DPR_8:$Vm), imm:$lane)))))]> {
2631 // All of these have a two-operand InstAlias.
2632 let TwoOperandAliasConstraint = "$Vn = $Vd";
2633 let isCommutable = 0;
2636 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2637 InstrItinClass itin, string OpcodeStr, string Dt,
2638 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2639 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2640 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2641 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2642 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
2643 // All of these have a two-operand InstAlias.
2644 let TwoOperandAliasConstraint = "$Vn = $Vd";
2645 let isCommutable = Commutable;
2647 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2648 InstrItinClass itin, string OpcodeStr,
2649 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2650 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
2651 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2652 OpcodeStr, "$Vd, $Vn, $Vm", "",
2653 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>{
2654 // All of these have a two-operand InstAlias.
2655 let TwoOperandAliasConstraint = "$Vn = $Vd";
2656 let isCommutable = Commutable;
2658 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
2659 InstrItinClass itin, string OpcodeStr, string Dt,
2660 ValueType ResTy, ValueType OpTy, SDNode ShOp>
2661 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2662 (outs QPR:$Vd), (ins QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2663 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2664 [(set (ResTy QPR:$Vd),
2665 (ResTy (ShOp (ResTy QPR:$Vn),
2666 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2668 // All of these have a two-operand InstAlias.
2669 let TwoOperandAliasConstraint = "$Vn = $Vd";
2670 let isCommutable = 0;
2672 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
2673 ValueType ResTy, ValueType OpTy, SDNode ShOp>
2674 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2675 (outs QPR:$Vd), (ins QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2676 NVMulSLFrm, IIC_VMULi16Q, OpcodeStr, Dt,"$Vd, $Vn, $Vm$lane", "",
2677 [(set (ResTy QPR:$Vd),
2678 (ResTy (ShOp (ResTy QPR:$Vn),
2679 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2681 // All of these have a two-operand InstAlias.
2682 let TwoOperandAliasConstraint = "$Vn = $Vd";
2683 let isCommutable = 0;
2686 // Basic 3-register intrinsics, both double- and quad-register.
2687 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2688 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2689 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp, bit Commutable>
2690 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2691 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), f, itin,
2692 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2693 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
2694 // All of these have a two-operand InstAlias.
2695 let TwoOperandAliasConstraint = "$Vn = $Vd";
2696 let isCommutable = Commutable;
2699 class N3VDIntnp<bits<5> op27_23, bits<2> op21_20, bits<4> op11_8, bit op6,
2700 bit op4, Format f, InstrItinClass itin, string OpcodeStr,
2701 string Dt, ValueType ResTy, ValueType OpTy,
2702 SDPatternOperator IntOp, bit Commutable>
2703 : N3Vnp<op27_23, op21_20, op11_8, op6, op4,
2704 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin, OpcodeStr, Dt,
2705 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>;
2707 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2708 string OpcodeStr, string Dt, ValueType Ty, SDPatternOperator IntOp>
2709 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2710 (outs DPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2711 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2713 (Ty (IntOp (Ty DPR:$Vn),
2714 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),
2716 let isCommutable = 0;
2719 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2720 string OpcodeStr, string Dt, ValueType Ty, SDPatternOperator IntOp>
2721 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2722 (outs DPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2723 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2725 (Ty (IntOp (Ty DPR:$Vn),
2726 (Ty (NEONvduplane (Ty DPR_8:$Vm), imm:$lane)))))]> {
2727 let isCommutable = 0;
2729 class N3VDIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2730 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2731 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2732 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2733 (outs DPR:$Vd), (ins DPR:$Vm, DPR:$Vn), f, itin,
2734 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
2735 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (OpTy DPR:$Vn))))]> {
2736 let TwoOperandAliasConstraint = "$Vm = $Vd";
2737 let isCommutable = 0;
2740 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2741 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2742 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp, bit Commutable>
2743 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2744 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), f, itin,
2745 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2746 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
2747 // All of these have a two-operand InstAlias.
2748 let TwoOperandAliasConstraint = "$Vn = $Vd";
2749 let isCommutable = Commutable;
2752 class N3VQIntnp<bits<5> op27_23, bits<2> op21_20, bits<4> op11_8, bit op6,
2753 bit op4, Format f, InstrItinClass itin, string OpcodeStr,
2754 string Dt, ValueType ResTy, ValueType OpTy,
2755 SDPatternOperator IntOp, bit Commutable>
2756 : N3Vnp<op27_23, op21_20, op11_8, op6, op4,
2757 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), f, itin, OpcodeStr, Dt,
2758 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>;
2760 // Same as N3VQIntnp but with Vd as a src register.
2761 class N3VQInt3np<bits<5> op27_23, bits<2> op21_20, bits<4> op11_8, bit op6,
2762 bit op4, Format f, InstrItinClass itin, string OpcodeStr,
2763 string Dt, ValueType ResTy, ValueType OpTy,
2764 SDPatternOperator IntOp, bit Commutable>
2765 : N3Vnp<op27_23, op21_20, op11_8, op6, op4,
2766 (outs QPR:$Vd), (ins QPR:$src, QPR:$Vn, QPR:$Vm),
2767 f, itin, OpcodeStr, Dt,
2768 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$src), (OpTy QPR:$Vn),
2769 (OpTy QPR:$Vm))))]> {
2770 let Constraints = "$src = $Vd";
2773 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2774 string OpcodeStr, string Dt,
2775 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2776 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2777 (outs QPR:$Vd), (ins QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2778 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2779 [(set (ResTy QPR:$Vd),
2780 (ResTy (IntOp (ResTy QPR:$Vn),
2781 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2783 let isCommutable = 0;
2785 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2786 string OpcodeStr, string Dt,
2787 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2788 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2789 (outs QPR:$Vd), (ins QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2790 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2791 [(set (ResTy QPR:$Vd),
2792 (ResTy (IntOp (ResTy QPR:$Vn),
2793 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2795 let isCommutable = 0;
2797 class N3VQIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2798 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2799 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2800 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2801 (outs QPR:$Vd), (ins QPR:$Vm, QPR:$Vn), f, itin,
2802 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
2803 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (OpTy QPR:$Vn))))]> {
2804 let TwoOperandAliasConstraint = "$Vm = $Vd";
2805 let isCommutable = 0;
2808 // Multiply-Add/Sub operations: double- and quad-register.
2809 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2810 InstrItinClass itin, string OpcodeStr, string Dt,
2811 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator OpNode>
2812 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2813 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2814 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2815 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
2816 (Ty (MulOp DPR:$Vn, DPR:$Vm)))))]>;
2818 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2819 string OpcodeStr, string Dt,
2820 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator ShOp>
2821 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2823 (ins DPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2825 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2827 (Ty (ShOp (Ty DPR:$src1),
2829 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),
2831 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2832 string OpcodeStr, string Dt,
2833 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator ShOp>
2834 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2836 (ins DPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2838 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2840 (Ty (ShOp (Ty DPR:$src1),
2842 (Ty (NEONvduplane (Ty DPR_8:$Vm),
2845 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2846 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
2847 SDPatternOperator MulOp, SDPatternOperator OpNode>
2848 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2849 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2850 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2851 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
2852 (Ty (MulOp QPR:$Vn, QPR:$Vm)))))]>;
2853 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2854 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2855 SDPatternOperator MulOp, SDPatternOperator ShOp>
2856 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2858 (ins QPR:$src1, QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2860 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2861 [(set (ResTy QPR:$Vd),
2862 (ResTy (ShOp (ResTy QPR:$src1),
2863 (ResTy (MulOp QPR:$Vn,
2864 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2866 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2867 string OpcodeStr, string Dt,
2868 ValueType ResTy, ValueType OpTy,
2869 SDPatternOperator MulOp, SDPatternOperator ShOp>
2870 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2872 (ins QPR:$src1, QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2874 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2875 [(set (ResTy QPR:$Vd),
2876 (ResTy (ShOp (ResTy QPR:$src1),
2877 (ResTy (MulOp QPR:$Vn,
2878 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2881 // Neon Intrinsic-Op instructions (VABA): double- and quad-register.
2882 class N3VDIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2883 InstrItinClass itin, string OpcodeStr, string Dt,
2884 ValueType Ty, SDPatternOperator IntOp, SDNode OpNode>
2885 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2886 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2887 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2888 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
2889 (Ty (IntOp (Ty DPR:$Vn), (Ty DPR:$Vm))))))]>;
2890 class N3VQIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2891 InstrItinClass itin, string OpcodeStr, string Dt,
2892 ValueType Ty, SDPatternOperator IntOp, SDNode OpNode>
2893 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2894 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2895 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2896 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
2897 (Ty (IntOp (Ty QPR:$Vn), (Ty QPR:$Vm))))))]>;
2899 // Neon 3-argument intrinsics, both double- and quad-register.
2900 // The destination register is also used as the first source operand register.
2901 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2902 InstrItinClass itin, string OpcodeStr, string Dt,
2903 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2904 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2905 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2906 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2907 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$src1),
2908 (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>;
2909 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2910 InstrItinClass itin, string OpcodeStr, string Dt,
2911 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2912 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2913 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2914 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2915 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$src1),
2916 (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>;
2918 // Long Multiply-Add/Sub operations.
2919 class N3VLMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2920 InstrItinClass itin, string OpcodeStr, string Dt,
2921 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2922 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2923 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2924 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2925 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
2926 (TyQ (MulOp (TyD DPR:$Vn),
2927 (TyD DPR:$Vm)))))]>;
2928 class N3VLMulOpSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2929 InstrItinClass itin, string OpcodeStr, string Dt,
2930 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2931 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$Vd),
2932 (ins QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2934 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2936 (OpNode (TyQ QPR:$src1),
2937 (TyQ (MulOp (TyD DPR:$Vn),
2938 (TyD (NEONvduplane (TyD DPR_VFP2:$Vm),
2940 class N3VLMulOpSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2941 InstrItinClass itin, string OpcodeStr, string Dt,
2942 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2943 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$Vd),
2944 (ins QPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2946 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2948 (OpNode (TyQ QPR:$src1),
2949 (TyQ (MulOp (TyD DPR:$Vn),
2950 (TyD (NEONvduplane (TyD DPR_8:$Vm),
2953 // Long Intrinsic-Op vector operations with explicit extend (VABAL).
2954 class N3VLIntExtOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2955 InstrItinClass itin, string OpcodeStr, string Dt,
2956 ValueType TyQ, ValueType TyD, SDPatternOperator IntOp, SDNode ExtOp,
2958 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2959 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2960 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2961 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
2962 (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
2963 (TyD DPR:$Vm)))))))]>;
2965 // Neon Long 3-argument intrinsic. The destination register is
2966 // a quad-register and is also used as the first source operand register.
2967 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2968 InstrItinClass itin, string OpcodeStr, string Dt,
2969 ValueType TyQ, ValueType TyD, SDPatternOperator IntOp>
2970 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2971 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2972 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2974 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$Vn), (TyD DPR:$Vm))))]>;
2975 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2976 string OpcodeStr, string Dt,
2977 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2978 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2980 (ins QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2982 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2983 [(set (ResTy QPR:$Vd),
2984 (ResTy (IntOp (ResTy QPR:$src1),
2986 (OpTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2988 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2989 InstrItinClass itin, string OpcodeStr, string Dt,
2990 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2991 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2993 (ins QPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2995 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2996 [(set (ResTy QPR:$Vd),
2997 (ResTy (IntOp (ResTy QPR:$src1),
2999 (OpTy (NEONvduplane (OpTy DPR_8:$Vm),
3002 // Narrowing 3-register intrinsics.
3003 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
3004 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
3005 SDPatternOperator IntOp, bit Commutable>
3006 : N3V<op24, op23, op21_20, op11_8, 0, op4,
3007 (outs DPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINi4D,
3008 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
3009 [(set DPR:$Vd, (TyD (IntOp (TyQ QPR:$Vn), (TyQ QPR:$Vm))))]> {
3010 let isCommutable = Commutable;
3013 // Long 3-register operations.
3014 class N3VL<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
3015 InstrItinClass itin, string OpcodeStr, string Dt,
3016 ValueType TyQ, ValueType TyD, SDNode OpNode, bit Commutable>
3017 : N3V<op24, op23, op21_20, op11_8, 0, op4,
3018 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
3019 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
3020 [(set QPR:$Vd, (TyQ (OpNode (TyD DPR:$Vn), (TyD DPR:$Vm))))]> {
3021 let isCommutable = Commutable;
3024 class N3VLSL<bit op24, bits<2> op21_20, bits<4> op11_8,
3025 InstrItinClass itin, string OpcodeStr, string Dt,
3026 ValueType TyQ, ValueType TyD, SDNode OpNode>
3027 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
3028 (outs QPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
3029 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
3031 (TyQ (OpNode (TyD DPR:$Vn),
3032 (TyD (NEONvduplane (TyD DPR_VFP2:$Vm),imm:$lane)))))]>;
3033 class N3VLSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
3034 InstrItinClass itin, string OpcodeStr, string Dt,
3035 ValueType TyQ, ValueType TyD, SDNode OpNode>
3036 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
3037 (outs QPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
3038 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
3040 (TyQ (OpNode (TyD DPR:$Vn),
3041 (TyD (NEONvduplane (TyD DPR_8:$Vm), imm:$lane)))))]>;
3043 // Long 3-register operations with explicitly extended operands.
3044 class N3VLExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
3045 InstrItinClass itin, string OpcodeStr, string Dt,
3046 ValueType TyQ, ValueType TyD, SDNode OpNode, SDNode ExtOp,
3048 : N3V<op24, op23, op21_20, op11_8, 0, op4,
3049 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
3050 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
3051 [(set QPR:$Vd, (OpNode (TyQ (ExtOp (TyD DPR:$Vn))),
3052 (TyQ (ExtOp (TyD DPR:$Vm)))))]> {
3053 let isCommutable = Commutable;
3056 // Long 3-register intrinsics with explicit extend (VABDL).
3057 class N3VLIntExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
3058 InstrItinClass itin, string OpcodeStr, string Dt,
3059 ValueType TyQ, ValueType TyD, SDPatternOperator IntOp, SDNode ExtOp,
3061 : N3V<op24, op23, op21_20, op11_8, 0, op4,
3062 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
3063 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
3064 [(set QPR:$Vd, (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
3065 (TyD DPR:$Vm))))))]> {
3066 let isCommutable = Commutable;
3069 // Long 3-register intrinsics.
3070 class N3VLInt<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, SDPatternOperator IntOp, 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 (IntOp (TyD DPR:$Vn), (TyD DPR:$Vm))))]> {
3077 let isCommutable = Commutable;
3080 // Same as above, but not predicated.
3081 class N3VLIntnp<bits<5> op27_23, bits<2> op21_20, bits<4> op11_8, bit op6,
3082 bit op4, InstrItinClass itin, string OpcodeStr,
3083 string Dt, ValueType ResTy, ValueType OpTy,
3084 SDPatternOperator IntOp, bit Commutable>
3085 : N3Vnp<op27_23, op21_20, op11_8, op6, op4,
3086 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin, OpcodeStr, Dt,
3087 [(set QPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>;
3089 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
3090 string OpcodeStr, string Dt,
3091 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
3092 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
3093 (outs QPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
3094 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
3095 [(set (ResTy QPR:$Vd),
3096 (ResTy (IntOp (OpTy DPR:$Vn),
3097 (OpTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
3099 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
3100 InstrItinClass itin, string OpcodeStr, string Dt,
3101 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
3102 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
3103 (outs QPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
3104 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
3105 [(set (ResTy QPR:$Vd),
3106 (ResTy (IntOp (OpTy DPR:$Vn),
3107 (OpTy (NEONvduplane (OpTy DPR_8:$Vm),
3110 // Wide 3-register operations.
3111 class N3VW<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
3112 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
3113 SDNode OpNode, SDNode ExtOp, bit Commutable>
3114 : N3V<op24, op23, op21_20, op11_8, 0, op4,
3115 (outs QPR:$Vd), (ins QPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VSUBiD,
3116 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
3117 [(set QPR:$Vd, (OpNode (TyQ QPR:$Vn),
3118 (TyQ (ExtOp (TyD DPR:$Vm)))))]> {
3119 // All of these have a two-operand InstAlias.
3120 let TwoOperandAliasConstraint = "$Vn = $Vd";
3121 let isCommutable = Commutable;
3124 // Pairwise long 2-register intrinsics, both double- and quad-register.
3125 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
3126 bits<2> op17_16, bits<5> op11_7, bit op4,
3127 string OpcodeStr, string Dt,
3128 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
3129 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
3130 (ins DPR:$Vm), IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
3131 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
3132 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
3133 bits<2> op17_16, bits<5> op11_7, bit op4,
3134 string OpcodeStr, string Dt,
3135 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
3136 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
3137 (ins QPR:$Vm), IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
3138 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
3140 // Pairwise long 2-register accumulate intrinsics,
3141 // both double- and quad-register.
3142 // The destination register is also used as the first source operand register.
3143 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
3144 bits<2> op17_16, bits<5> op11_7, bit op4,
3145 string OpcodeStr, string Dt,
3146 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
3147 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
3148 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vm), IIC_VPALiD,
3149 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
3150 [(set DPR:$Vd, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$Vm))))]>;
3151 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
3152 bits<2> op17_16, bits<5> op11_7, bit op4,
3153 string OpcodeStr, string Dt,
3154 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
3155 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
3156 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vm), IIC_VPALiQ,
3157 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
3158 [(set QPR:$Vd, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$Vm))))]>;
3160 // Shift by immediate,
3161 // both double- and quad-register.
3162 let TwoOperandAliasConstraint = "$Vm = $Vd" in {
3163 class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
3164 Format f, InstrItinClass itin, Operand ImmTy,
3165 string OpcodeStr, string Dt, ValueType Ty, SDNode OpNode>
3166 : N2VImm<op24, op23, op11_8, op7, 0, op4,
3167 (outs DPR:$Vd), (ins DPR:$Vm, ImmTy:$SIMM), f, itin,
3168 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
3169 [(set DPR:$Vd, (Ty (OpNode (Ty DPR:$Vm), (i32 imm:$SIMM))))]>;
3170 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
3171 Format f, InstrItinClass itin, Operand ImmTy,
3172 string OpcodeStr, string Dt, ValueType Ty, SDNode OpNode>
3173 : N2VImm<op24, op23, op11_8, op7, 1, op4,
3174 (outs QPR:$Vd), (ins QPR:$Vm, ImmTy:$SIMM), f, itin,
3175 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
3176 [(set QPR:$Vd, (Ty (OpNode (Ty QPR:$Vm), (i32 imm:$SIMM))))]>;
3179 // Long shift by immediate.
3180 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
3181 string OpcodeStr, string Dt,
3182 ValueType ResTy, ValueType OpTy, Operand ImmTy,
3183 SDPatternOperator OpNode>
3184 : N2VImm<op24, op23, op11_8, op7, op6, op4,
3185 (outs QPR:$Vd), (ins DPR:$Vm, ImmTy:$SIMM), N2RegVShLFrm,
3186 IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
3187 [(set QPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm), ImmTy:$SIMM)))]>;
3189 // Narrow shift by immediate.
3190 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
3191 InstrItinClass itin, string OpcodeStr, string Dt,
3192 ValueType ResTy, ValueType OpTy, Operand ImmTy,
3193 SDPatternOperator OpNode>
3194 : N2VImm<op24, op23, op11_8, op7, op6, op4,
3195 (outs DPR:$Vd), (ins QPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, itin,
3196 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
3197 [(set DPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vm),
3198 (i32 ImmTy:$SIMM))))]>;
3200 // Shift right by immediate and accumulate,
3201 // both double- and quad-register.
3202 let TwoOperandAliasConstraint = "$Vm = $Vd" in {
3203 class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
3204 Operand ImmTy, string OpcodeStr, string Dt,
3205 ValueType Ty, SDNode ShOp>
3206 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
3207 (ins DPR:$src1, DPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, IIC_VPALiD,
3208 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
3209 [(set DPR:$Vd, (Ty (add DPR:$src1,
3210 (Ty (ShOp DPR:$Vm, (i32 imm:$SIMM))))))]>;
3211 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
3212 Operand ImmTy, string OpcodeStr, string Dt,
3213 ValueType Ty, SDNode ShOp>
3214 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
3215 (ins QPR:$src1, QPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, IIC_VPALiD,
3216 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
3217 [(set QPR:$Vd, (Ty (add QPR:$src1,
3218 (Ty (ShOp QPR:$Vm, (i32 imm:$SIMM))))))]>;
3221 // Shift by immediate and insert,
3222 // both double- and quad-register.
3223 let TwoOperandAliasConstraint = "$Vm = $Vd" in {
3224 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
3225 Operand ImmTy, Format f, string OpcodeStr, string Dt,
3226 ValueType Ty,SDNode ShOp>
3227 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
3228 (ins DPR:$src1, DPR:$Vm, ImmTy:$SIMM), f, IIC_VSHLiD,
3229 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
3230 [(set DPR:$Vd, (Ty (ShOp DPR:$src1, DPR:$Vm, (i32 imm:$SIMM))))]>;
3231 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
3232 Operand ImmTy, Format f, string OpcodeStr, string Dt,
3233 ValueType Ty,SDNode ShOp>
3234 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
3235 (ins QPR:$src1, QPR:$Vm, ImmTy:$SIMM), f, IIC_VSHLiQ,
3236 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
3237 [(set QPR:$Vd, (Ty (ShOp QPR:$src1, QPR:$Vm, (i32 imm:$SIMM))))]>;
3240 // Convert, with fractional bits immediate,
3241 // both double- and quad-register.
3242 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
3243 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
3244 SDPatternOperator IntOp>
3245 : N2VImm<op24, op23, op11_8, op7, 0, op4,
3246 (outs DPR:$Vd), (ins DPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
3247 IIC_VUNAD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
3248 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (i32 imm:$SIMM))))]>;
3249 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
3250 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
3251 SDPatternOperator IntOp>
3252 : N2VImm<op24, op23, op11_8, op7, 1, op4,
3253 (outs QPR:$Vd), (ins QPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
3254 IIC_VUNAQ, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
3255 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (i32 imm:$SIMM))))]>;
3257 //===----------------------------------------------------------------------===//
3259 //===----------------------------------------------------------------------===//
3261 // Abbreviations used in multiclass suffixes:
3262 // Q = quarter int (8 bit) elements
3263 // H = half int (16 bit) elements
3264 // S = single int (32 bit) elements
3265 // D = double int (64 bit) elements
3267 // Neon 2-register vector operations and intrinsics.
3269 // Neon 2-register comparisons.
3270 // source operand element sizes of 8, 16 and 32 bits:
3271 multiclass N2V_QHS_cmp<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3272 bits<5> op11_7, bit op4, string opc, string Dt,
3273 string asm, SDNode OpNode> {
3274 // 64-bit vector types.
3275 def v8i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 0, op4,
3276 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
3277 opc, !strconcat(Dt, "8"), asm, "",
3278 [(set DPR:$Vd, (v8i8 (OpNode (v8i8 DPR:$Vm))))]>;
3279 def v4i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
3280 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
3281 opc, !strconcat(Dt, "16"), asm, "",
3282 [(set DPR:$Vd, (v4i16 (OpNode (v4i16 DPR:$Vm))))]>;
3283 def v2i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
3284 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
3285 opc, !strconcat(Dt, "32"), asm, "",
3286 [(set DPR:$Vd, (v2i32 (OpNode (v2i32 DPR:$Vm))))]>;
3287 def v2f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
3288 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
3289 opc, "f32", asm, "",
3290 [(set DPR:$Vd, (v2i32 (OpNode (v2f32 DPR:$Vm))))]> {
3291 let Inst{10} = 1; // overwrite F = 1
3293 def v4f16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
3294 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
3295 opc, "f16", asm, "",
3296 [(set DPR:$Vd, (v4i16 (OpNode (v4f16 DPR:$Vm))))]>,
3297 Requires<[HasNEON,HasFullFP16]> {
3298 let Inst{10} = 1; // overwrite F = 1
3301 // 128-bit vector types.
3302 def v16i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 1, op4,
3303 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
3304 opc, !strconcat(Dt, "8"), asm, "",
3305 [(set QPR:$Vd, (v16i8 (OpNode (v16i8 QPR:$Vm))))]>;
3306 def v8i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
3307 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
3308 opc, !strconcat(Dt, "16"), asm, "",
3309 [(set QPR:$Vd, (v8i16 (OpNode (v8i16 QPR:$Vm))))]>;
3310 def v4i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
3311 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
3312 opc, !strconcat(Dt, "32"), asm, "",
3313 [(set QPR:$Vd, (v4i32 (OpNode (v4i32 QPR:$Vm))))]>;
3314 def v4f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
3315 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
3316 opc, "f32", asm, "",
3317 [(set QPR:$Vd, (v4i32 (OpNode (v4f32 QPR:$Vm))))]> {
3318 let Inst{10} = 1; // overwrite F = 1
3320 def v8f16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
3321 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
3322 opc, "f16", asm, "",
3323 [(set QPR:$Vd, (v8i16 (OpNode (v8f16 QPR:$Vm))))]>,
3324 Requires<[HasNEON,HasFullFP16]> {
3325 let Inst{10} = 1; // overwrite F = 1
3330 // Neon 2-register vector intrinsics,
3331 // element sizes of 8, 16 and 32 bits:
3332 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3333 bits<5> op11_7, bit op4,
3334 InstrItinClass itinD, InstrItinClass itinQ,
3335 string OpcodeStr, string Dt, SDPatternOperator IntOp> {
3336 // 64-bit vector types.
3337 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3338 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
3339 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3340 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
3341 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3342 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
3344 // 128-bit vector types.
3345 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3346 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
3347 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3348 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
3349 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3350 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
3354 // Neon Narrowing 2-register vector operations,
3355 // source operand element sizes of 16, 32 and 64 bits:
3356 multiclass N2VN_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3357 bits<5> op11_7, bit op6, bit op4,
3358 InstrItinClass itin, string OpcodeStr, string Dt,
3360 def v8i8 : N2VN<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
3361 itin, OpcodeStr, !strconcat(Dt, "16"),
3362 v8i8, v8i16, OpNode>;
3363 def v4i16 : N2VN<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
3364 itin, OpcodeStr, !strconcat(Dt, "32"),
3365 v4i16, v4i32, OpNode>;
3366 def v2i32 : N2VN<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
3367 itin, OpcodeStr, !strconcat(Dt, "64"),
3368 v2i32, v2i64, OpNode>;
3371 // Neon Narrowing 2-register vector intrinsics,
3372 // source operand element sizes of 16, 32 and 64 bits:
3373 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3374 bits<5> op11_7, bit op6, bit op4,
3375 InstrItinClass itin, string OpcodeStr, string Dt,
3376 SDPatternOperator IntOp> {
3377 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
3378 itin, OpcodeStr, !strconcat(Dt, "16"),
3379 v8i8, v8i16, IntOp>;
3380 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
3381 itin, OpcodeStr, !strconcat(Dt, "32"),
3382 v4i16, v4i32, IntOp>;
3383 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
3384 itin, OpcodeStr, !strconcat(Dt, "64"),
3385 v2i32, v2i64, IntOp>;
3389 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
3390 // source operand element sizes of 16, 32 and 64 bits:
3391 multiclass N2VL_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
3392 string OpcodeStr, string Dt, SDNode OpNode> {
3393 def v8i16 : N2VL<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
3394 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode>;
3395 def v4i32 : N2VL<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
3396 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
3397 def v2i64 : N2VL<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
3398 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
3402 // Neon 3-register vector operations.
3404 // First with only element sizes of 8, 16 and 32 bits:
3405 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3406 InstrItinClass itinD16, InstrItinClass itinD32,
3407 InstrItinClass itinQ16, InstrItinClass itinQ32,
3408 string OpcodeStr, string Dt,
3409 SDNode OpNode, bit Commutable = 0> {
3410 // 64-bit vector types.
3411 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
3412 OpcodeStr, !strconcat(Dt, "8"),
3413 v8i8, v8i8, OpNode, Commutable>;
3414 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
3415 OpcodeStr, !strconcat(Dt, "16"),
3416 v4i16, v4i16, OpNode, Commutable>;
3417 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
3418 OpcodeStr, !strconcat(Dt, "32"),
3419 v2i32, v2i32, OpNode, Commutable>;
3421 // 128-bit vector types.
3422 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
3423 OpcodeStr, !strconcat(Dt, "8"),
3424 v16i8, v16i8, OpNode, Commutable>;
3425 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
3426 OpcodeStr, !strconcat(Dt, "16"),
3427 v8i16, v8i16, OpNode, Commutable>;
3428 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
3429 OpcodeStr, !strconcat(Dt, "32"),
3430 v4i32, v4i32, OpNode, Commutable>;
3433 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, SDNode ShOp> {
3434 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, "i16", v4i16, ShOp>;
3435 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, "i32", v2i32, ShOp>;
3436 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, "i16", v8i16, v4i16, ShOp>;
3437 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, "i32",
3438 v4i32, v2i32, ShOp>;
3441 // ....then also with element size 64 bits:
3442 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3443 InstrItinClass itinD, InstrItinClass itinQ,
3444 string OpcodeStr, string Dt,
3445 SDNode OpNode, bit Commutable = 0>
3446 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
3447 OpcodeStr, Dt, OpNode, Commutable> {
3448 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
3449 OpcodeStr, !strconcat(Dt, "64"),
3450 v1i64, v1i64, OpNode, Commutable>;
3451 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
3452 OpcodeStr, !strconcat(Dt, "64"),
3453 v2i64, v2i64, OpNode, Commutable>;
3457 // Neon 3-register vector intrinsics.
3459 // First with only element sizes of 16 and 32 bits:
3460 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3461 InstrItinClass itinD16, InstrItinClass itinD32,
3462 InstrItinClass itinQ16, InstrItinClass itinQ32,
3463 string OpcodeStr, string Dt,
3464 SDPatternOperator IntOp, bit Commutable = 0> {
3465 // 64-bit vector types.
3466 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, f, itinD16,
3467 OpcodeStr, !strconcat(Dt, "16"),
3468 v4i16, v4i16, IntOp, Commutable>;
3469 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, f, itinD32,
3470 OpcodeStr, !strconcat(Dt, "32"),
3471 v2i32, v2i32, IntOp, Commutable>;
3473 // 128-bit vector types.
3474 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, f, itinQ16,
3475 OpcodeStr, !strconcat(Dt, "16"),
3476 v8i16, v8i16, IntOp, Commutable>;
3477 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, f, itinQ32,
3478 OpcodeStr, !strconcat(Dt, "32"),
3479 v4i32, v4i32, IntOp, Commutable>;
3481 multiclass N3VInt_HSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3482 InstrItinClass itinD16, InstrItinClass itinD32,
3483 InstrItinClass itinQ16, InstrItinClass itinQ32,
3484 string OpcodeStr, string Dt,
3485 SDPatternOperator IntOp> {
3486 // 64-bit vector types.
3487 def v4i16 : N3VDIntSh<op24, op23, 0b01, op11_8, op4, f, itinD16,
3488 OpcodeStr, !strconcat(Dt, "16"),
3489 v4i16, v4i16, IntOp>;
3490 def v2i32 : N3VDIntSh<op24, op23, 0b10, op11_8, op4, f, itinD32,
3491 OpcodeStr, !strconcat(Dt, "32"),
3492 v2i32, v2i32, IntOp>;
3494 // 128-bit vector types.
3495 def v8i16 : N3VQIntSh<op24, op23, 0b01, op11_8, op4, f, itinQ16,
3496 OpcodeStr, !strconcat(Dt, "16"),
3497 v8i16, v8i16, IntOp>;
3498 def v4i32 : N3VQIntSh<op24, op23, 0b10, op11_8, op4, f, itinQ32,
3499 OpcodeStr, !strconcat(Dt, "32"),
3500 v4i32, v4i32, IntOp>;
3503 multiclass N3VIntSL_HS<bits<4> op11_8,
3504 InstrItinClass itinD16, InstrItinClass itinD32,
3505 InstrItinClass itinQ16, InstrItinClass itinQ32,
3506 string OpcodeStr, string Dt, SDPatternOperator IntOp> {
3507 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
3508 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
3509 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
3510 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
3511 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
3512 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
3513 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
3514 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
3517 // ....then also with element size of 8 bits:
3518 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3519 InstrItinClass itinD16, InstrItinClass itinD32,
3520 InstrItinClass itinQ16, InstrItinClass itinQ32,
3521 string OpcodeStr, string Dt,
3522 SDPatternOperator IntOp, bit Commutable = 0>
3523 : N3VInt_HS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3524 OpcodeStr, Dt, IntOp, Commutable> {
3525 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, f, itinD16,
3526 OpcodeStr, !strconcat(Dt, "8"),
3527 v8i8, v8i8, IntOp, Commutable>;
3528 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, f, itinQ16,
3529 OpcodeStr, !strconcat(Dt, "8"),
3530 v16i8, v16i8, IntOp, Commutable>;
3532 multiclass N3VInt_QHSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3533 InstrItinClass itinD16, InstrItinClass itinD32,
3534 InstrItinClass itinQ16, InstrItinClass itinQ32,
3535 string OpcodeStr, string Dt,
3536 SDPatternOperator IntOp>
3537 : N3VInt_HSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3538 OpcodeStr, Dt, IntOp> {
3539 def v8i8 : N3VDIntSh<op24, op23, 0b00, op11_8, op4, f, itinD16,
3540 OpcodeStr, !strconcat(Dt, "8"),
3542 def v16i8 : N3VQIntSh<op24, op23, 0b00, op11_8, op4, f, itinQ16,
3543 OpcodeStr, !strconcat(Dt, "8"),
3544 v16i8, v16i8, IntOp>;
3548 // ....then also with element size of 64 bits:
3549 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3550 InstrItinClass itinD16, InstrItinClass itinD32,
3551 InstrItinClass itinQ16, InstrItinClass itinQ32,
3552 string OpcodeStr, string Dt,
3553 SDPatternOperator IntOp, bit Commutable = 0>
3554 : N3VInt_QHS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3555 OpcodeStr, Dt, IntOp, Commutable> {
3556 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, f, itinD32,
3557 OpcodeStr, !strconcat(Dt, "64"),
3558 v1i64, v1i64, IntOp, Commutable>;
3559 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, f, itinQ32,
3560 OpcodeStr, !strconcat(Dt, "64"),
3561 v2i64, v2i64, IntOp, Commutable>;
3563 multiclass N3VInt_QHSDSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3564 InstrItinClass itinD16, InstrItinClass itinD32,
3565 InstrItinClass itinQ16, InstrItinClass itinQ32,
3566 string OpcodeStr, string Dt,
3567 SDPatternOperator IntOp>
3568 : N3VInt_QHSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3569 OpcodeStr, Dt, IntOp> {
3570 def v1i64 : N3VDIntSh<op24, op23, 0b11, op11_8, op4, f, itinD32,
3571 OpcodeStr, !strconcat(Dt, "64"),
3572 v1i64, v1i64, IntOp>;
3573 def v2i64 : N3VQIntSh<op24, op23, 0b11, op11_8, op4, f, itinQ32,
3574 OpcodeStr, !strconcat(Dt, "64"),
3575 v2i64, v2i64, IntOp>;
3578 // Neon Narrowing 3-register vector intrinsics,
3579 // source operand element sizes of 16, 32 and 64 bits:
3580 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3581 string OpcodeStr, string Dt,
3582 SDPatternOperator IntOp, bit Commutable = 0> {
3583 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
3584 OpcodeStr, !strconcat(Dt, "16"),
3585 v8i8, v8i16, IntOp, Commutable>;
3586 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
3587 OpcodeStr, !strconcat(Dt, "32"),
3588 v4i16, v4i32, IntOp, Commutable>;
3589 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
3590 OpcodeStr, !strconcat(Dt, "64"),
3591 v2i32, v2i64, IntOp, Commutable>;
3595 // Neon Long 3-register vector operations.
3597 multiclass N3VL_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3598 InstrItinClass itin16, InstrItinClass itin32,
3599 string OpcodeStr, string Dt,
3600 SDNode OpNode, bit Commutable = 0> {
3601 def v8i16 : N3VL<op24, op23, 0b00, op11_8, op4, itin16,
3602 OpcodeStr, !strconcat(Dt, "8"),
3603 v8i16, v8i8, OpNode, Commutable>;
3604 def v4i32 : N3VL<op24, op23, 0b01, op11_8, op4, itin16,
3605 OpcodeStr, !strconcat(Dt, "16"),
3606 v4i32, v4i16, OpNode, Commutable>;
3607 def v2i64 : N3VL<op24, op23, 0b10, op11_8, op4, itin32,
3608 OpcodeStr, !strconcat(Dt, "32"),
3609 v2i64, v2i32, OpNode, Commutable>;
3612 multiclass N3VLSL_HS<bit op24, bits<4> op11_8,
3613 InstrItinClass itin, string OpcodeStr, string Dt,
3615 def v4i16 : N3VLSL16<op24, 0b01, op11_8, itin, OpcodeStr,
3616 !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
3617 def v2i32 : N3VLSL<op24, 0b10, op11_8, itin, OpcodeStr,
3618 !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
3621 multiclass N3VLExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3622 InstrItinClass itin16, InstrItinClass itin32,
3623 string OpcodeStr, string Dt,
3624 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
3625 def v8i16 : N3VLExt<op24, op23, 0b00, op11_8, op4, itin16,
3626 OpcodeStr, !strconcat(Dt, "8"),
3627 v8i16, v8i8, OpNode, ExtOp, Commutable>;
3628 def v4i32 : N3VLExt<op24, op23, 0b01, op11_8, op4, itin16,
3629 OpcodeStr, !strconcat(Dt, "16"),
3630 v4i32, v4i16, OpNode, ExtOp, Commutable>;
3631 def v2i64 : N3VLExt<op24, op23, 0b10, op11_8, op4, itin32,
3632 OpcodeStr, !strconcat(Dt, "32"),
3633 v2i64, v2i32, OpNode, ExtOp, Commutable>;
3636 // Neon Long 3-register vector intrinsics.
3638 // First with only element sizes of 16 and 32 bits:
3639 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
3640 InstrItinClass itin16, InstrItinClass itin32,
3641 string OpcodeStr, string Dt,
3642 SDPatternOperator IntOp, bit Commutable = 0> {
3643 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin16,
3644 OpcodeStr, !strconcat(Dt, "16"),
3645 v4i32, v4i16, IntOp, Commutable>;
3646 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin32,
3647 OpcodeStr, !strconcat(Dt, "32"),
3648 v2i64, v2i32, IntOp, Commutable>;
3651 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
3652 InstrItinClass itin, string OpcodeStr, string Dt,
3653 SDPatternOperator IntOp> {
3654 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
3655 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
3656 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
3657 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3660 // ....then also with element size of 8 bits:
3661 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3662 InstrItinClass itin16, InstrItinClass itin32,
3663 string OpcodeStr, string Dt,
3664 SDPatternOperator IntOp, bit Commutable = 0>
3665 : N3VLInt_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt,
3666 IntOp, Commutable> {
3667 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin16,
3668 OpcodeStr, !strconcat(Dt, "8"),
3669 v8i16, v8i8, IntOp, Commutable>;
3672 // ....with explicit extend (VABDL).
3673 multiclass N3VLIntExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3674 InstrItinClass itin, string OpcodeStr, string Dt,
3675 SDPatternOperator IntOp, SDNode ExtOp, bit Commutable = 0> {
3676 def v8i16 : N3VLIntExt<op24, op23, 0b00, op11_8, op4, itin,
3677 OpcodeStr, !strconcat(Dt, "8"),
3678 v8i16, v8i8, IntOp, ExtOp, Commutable>;
3679 def v4i32 : N3VLIntExt<op24, op23, 0b01, op11_8, op4, itin,
3680 OpcodeStr, !strconcat(Dt, "16"),
3681 v4i32, v4i16, IntOp, ExtOp, Commutable>;
3682 def v2i64 : N3VLIntExt<op24, op23, 0b10, op11_8, op4, itin,
3683 OpcodeStr, !strconcat(Dt, "32"),
3684 v2i64, v2i32, IntOp, ExtOp, Commutable>;
3688 // Neon Wide 3-register vector intrinsics,
3689 // source operand element sizes of 8, 16 and 32 bits:
3690 multiclass N3VW_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3691 string OpcodeStr, string Dt,
3692 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
3693 def v8i16 : N3VW<op24, op23, 0b00, op11_8, op4,
3694 OpcodeStr, !strconcat(Dt, "8"),
3695 v8i16, v8i8, OpNode, ExtOp, Commutable>;
3696 def v4i32 : N3VW<op24, op23, 0b01, op11_8, op4,
3697 OpcodeStr, !strconcat(Dt, "16"),
3698 v4i32, v4i16, OpNode, ExtOp, Commutable>;
3699 def v2i64 : N3VW<op24, op23, 0b10, op11_8, op4,
3700 OpcodeStr, !strconcat(Dt, "32"),
3701 v2i64, v2i32, OpNode, ExtOp, Commutable>;
3705 // Neon Multiply-Op vector operations,
3706 // element sizes of 8, 16 and 32 bits:
3707 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3708 InstrItinClass itinD16, InstrItinClass itinD32,
3709 InstrItinClass itinQ16, InstrItinClass itinQ32,
3710 string OpcodeStr, string Dt, SDNode OpNode> {
3711 // 64-bit vector types.
3712 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
3713 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
3714 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
3715 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
3716 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
3717 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
3719 // 128-bit vector types.
3720 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
3721 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
3722 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
3723 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
3724 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
3725 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
3728 multiclass N3VMulOpSL_HS<bits<4> op11_8,
3729 InstrItinClass itinD16, InstrItinClass itinD32,
3730 InstrItinClass itinQ16, InstrItinClass itinQ32,
3731 string OpcodeStr, string Dt, SDPatternOperator ShOp> {
3732 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
3733 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
3734 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
3735 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
3736 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
3737 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
3739 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
3740 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
3744 // Neon Intrinsic-Op vector operations,
3745 // element sizes of 8, 16 and 32 bits:
3746 multiclass N3VIntOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3747 InstrItinClass itinD, InstrItinClass itinQ,
3748 string OpcodeStr, string Dt, SDPatternOperator IntOp,
3750 // 64-bit vector types.
3751 def v8i8 : N3VDIntOp<op24, op23, 0b00, op11_8, op4, itinD,
3752 OpcodeStr, !strconcat(Dt, "8"), v8i8, IntOp, OpNode>;
3753 def v4i16 : N3VDIntOp<op24, op23, 0b01, op11_8, op4, itinD,
3754 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp, OpNode>;
3755 def v2i32 : N3VDIntOp<op24, op23, 0b10, op11_8, op4, itinD,
3756 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp, OpNode>;
3758 // 128-bit vector types.
3759 def v16i8 : N3VQIntOp<op24, op23, 0b00, op11_8, op4, itinQ,
3760 OpcodeStr, !strconcat(Dt, "8"), v16i8, IntOp, OpNode>;
3761 def v8i16 : N3VQIntOp<op24, op23, 0b01, op11_8, op4, itinQ,
3762 OpcodeStr, !strconcat(Dt, "16"), v8i16, IntOp, OpNode>;
3763 def v4i32 : N3VQIntOp<op24, op23, 0b10, op11_8, op4, itinQ,
3764 OpcodeStr, !strconcat(Dt, "32"), v4i32, IntOp, OpNode>;
3767 // Neon 3-argument intrinsics,
3768 // element sizes of 16 and 32 bits:
3769 multiclass N3VInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
3770 InstrItinClass itinD16, InstrItinClass itinD32,
3771 InstrItinClass itinQ16, InstrItinClass itinQ32,
3772 string OpcodeStr, string Dt, SDPatternOperator IntOp> {
3773 // 64-bit vector types.
3774 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, itinD16,
3775 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
3776 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, itinD32,
3777 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
3779 // 128-bit vector types.
3780 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, itinQ16,
3781 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
3782 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, itinQ32,
3783 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
3786 // element sizes of 8, 16 and 32 bits:
3787 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3788 InstrItinClass itinD16, InstrItinClass itinD32,
3789 InstrItinClass itinQ16, InstrItinClass itinQ32,
3790 string OpcodeStr, string Dt, SDPatternOperator IntOp>
3791 :N3VInt3_HS <op24, op23, op11_8, op4, itinD16, itinD32,
3792 itinQ16, itinQ32, OpcodeStr, Dt, IntOp>{
3793 // 64-bit vector types.
3794 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, itinD16,
3795 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
3796 // 128-bit vector types.
3797 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, itinQ16,
3798 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
3801 // Neon Long Multiply-Op vector operations,
3802 // element sizes of 8, 16 and 32 bits:
3803 multiclass N3VLMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3804 InstrItinClass itin16, InstrItinClass itin32,
3805 string OpcodeStr, string Dt, SDNode MulOp,
3807 def v8i16 : N3VLMulOp<op24, op23, 0b00, op11_8, op4, itin16, OpcodeStr,
3808 !strconcat(Dt, "8"), v8i16, v8i8, MulOp, OpNode>;
3809 def v4i32 : N3VLMulOp<op24, op23, 0b01, op11_8, op4, itin16, OpcodeStr,
3810 !strconcat(Dt, "16"), v4i32, v4i16, MulOp, OpNode>;
3811 def v2i64 : N3VLMulOp<op24, op23, 0b10, op11_8, op4, itin32, OpcodeStr,
3812 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
3815 multiclass N3VLMulOpSL_HS<bit op24, bits<4> op11_8, string OpcodeStr,
3816 string Dt, SDNode MulOp, SDNode OpNode> {
3817 def v4i16 : N3VLMulOpSL16<op24, 0b01, op11_8, IIC_VMACi16D, OpcodeStr,
3818 !strconcat(Dt,"16"), v4i32, v4i16, MulOp, OpNode>;
3819 def v2i32 : N3VLMulOpSL<op24, 0b10, op11_8, IIC_VMACi32D, OpcodeStr,
3820 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
3824 // Neon Long 3-argument intrinsics.
3826 // First with only element sizes of 16 and 32 bits:
3827 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
3828 InstrItinClass itin16, InstrItinClass itin32,
3829 string OpcodeStr, string Dt, SDPatternOperator IntOp> {
3830 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, itin16,
3831 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
3832 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, itin32,
3833 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3836 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
3837 string OpcodeStr, string Dt, SDPatternOperator IntOp> {
3838 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
3839 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
3840 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
3841 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3844 // ....then also with element size of 8 bits:
3845 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3846 InstrItinClass itin16, InstrItinClass itin32,
3847 string OpcodeStr, string Dt, SDPatternOperator IntOp>
3848 : N3VLInt3_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt, IntOp> {
3849 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, itin16,
3850 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
3853 // ....with explicit extend (VABAL).
3854 multiclass N3VLIntExtOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3855 InstrItinClass itin, string OpcodeStr, string Dt,
3856 SDPatternOperator IntOp, SDNode ExtOp, SDNode OpNode> {
3857 def v8i16 : N3VLIntExtOp<op24, op23, 0b00, op11_8, op4, itin,
3858 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8,
3859 IntOp, ExtOp, OpNode>;
3860 def v4i32 : N3VLIntExtOp<op24, op23, 0b01, op11_8, op4, itin,
3861 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16,
3862 IntOp, ExtOp, OpNode>;
3863 def v2i64 : N3VLIntExtOp<op24, op23, 0b10, op11_8, op4, itin,
3864 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32,
3865 IntOp, ExtOp, OpNode>;
3869 // Neon Pairwise long 2-register intrinsics,
3870 // element sizes of 8, 16 and 32 bits:
3871 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3872 bits<5> op11_7, bit op4,
3873 string OpcodeStr, string Dt, SDPatternOperator IntOp> {
3874 // 64-bit vector types.
3875 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3876 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
3877 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3878 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
3879 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3880 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
3882 // 128-bit vector types.
3883 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3884 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
3885 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3886 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
3887 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3888 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
3892 // Neon Pairwise long 2-register accumulate intrinsics,
3893 // element sizes of 8, 16 and 32 bits:
3894 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3895 bits<5> op11_7, bit op4,
3896 string OpcodeStr, string Dt, SDPatternOperator IntOp> {
3897 // 64-bit vector types.
3898 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3899 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
3900 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3901 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
3902 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3903 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
3905 // 128-bit vector types.
3906 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3907 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
3908 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3909 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
3910 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3911 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
3915 // Neon 2-register vector shift by immediate,
3916 // with f of either N2RegVShLFrm or N2RegVShRFrm
3917 // element sizes of 8, 16, 32 and 64 bits:
3918 multiclass N2VShL_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3919 InstrItinClass itin, string OpcodeStr, string Dt,
3921 // 64-bit vector types.
3922 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3923 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
3924 let Inst{21-19} = 0b001; // imm6 = 001xxx
3926 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3927 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
3928 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3930 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3931 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
3932 let Inst{21} = 0b1; // imm6 = 1xxxxx
3934 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, N2RegVShLFrm, itin, i32imm,
3935 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
3938 // 128-bit vector types.
3939 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3940 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
3941 let Inst{21-19} = 0b001; // imm6 = 001xxx
3943 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3944 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
3945 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3947 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3948 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
3949 let Inst{21} = 0b1; // imm6 = 1xxxxx
3951 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, N2RegVShLFrm, itin, i32imm,
3952 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
3955 multiclass N2VShR_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3956 InstrItinClass itin, string OpcodeStr, string Dt,
3957 string baseOpc, SDNode OpNode> {
3958 // 64-bit vector types.
3959 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm8,
3960 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
3961 let Inst{21-19} = 0b001; // imm6 = 001xxx
3963 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm16,
3964 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
3965 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3967 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm32,
3968 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
3969 let Inst{21} = 0b1; // imm6 = 1xxxxx
3971 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, N2RegVShRFrm, itin, shr_imm64,
3972 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
3975 // 128-bit vector types.
3976 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm8,
3977 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
3978 let Inst{21-19} = 0b001; // imm6 = 001xxx
3980 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm16,
3981 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
3982 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3984 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm32,
3985 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
3986 let Inst{21} = 0b1; // imm6 = 1xxxxx
3988 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, N2RegVShRFrm, itin, shr_imm64,
3989 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
3993 // Neon Shift-Accumulate vector operations,
3994 // element sizes of 8, 16, 32 and 64 bits:
3995 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3996 string OpcodeStr, string Dt, SDNode ShOp> {
3997 // 64-bit vector types.
3998 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm8,
3999 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
4000 let Inst{21-19} = 0b001; // imm6 = 001xxx
4002 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm16,
4003 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
4004 let Inst{21-20} = 0b01; // imm6 = 01xxxx
4006 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm32,
4007 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
4008 let Inst{21} = 0b1; // imm6 = 1xxxxx
4010 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4, shr_imm64,
4011 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
4014 // 128-bit vector types.
4015 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm8,
4016 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
4017 let Inst{21-19} = 0b001; // imm6 = 001xxx
4019 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm16,
4020 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
4021 let Inst{21-20} = 0b01; // imm6 = 01xxxx
4023 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm32,
4024 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
4025 let Inst{21} = 0b1; // imm6 = 1xxxxx
4027 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4, shr_imm64,
4028 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
4032 // Neon Shift-Insert vector operations,
4033 // with f of either N2RegVShLFrm or N2RegVShRFrm
4034 // element sizes of 8, 16, 32 and 64 bits:
4035 multiclass N2VShInsL_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
4037 // 64-bit vector types.
4038 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
4039 N2RegVShLFrm, OpcodeStr, "8", v8i8, NEONvsli> {
4040 let Inst{21-19} = 0b001; // imm6 = 001xxx
4042 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
4043 N2RegVShLFrm, OpcodeStr, "16", v4i16, NEONvsli> {
4044 let Inst{21-20} = 0b01; // imm6 = 01xxxx
4046 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
4047 N2RegVShLFrm, OpcodeStr, "32", v2i32, NEONvsli> {
4048 let Inst{21} = 0b1; // imm6 = 1xxxxx
4050 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4, i32imm,
4051 N2RegVShLFrm, OpcodeStr, "64", v1i64, NEONvsli>;
4054 // 128-bit vector types.
4055 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
4056 N2RegVShLFrm, OpcodeStr, "8", v16i8, NEONvsli> {
4057 let Inst{21-19} = 0b001; // imm6 = 001xxx
4059 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
4060 N2RegVShLFrm, OpcodeStr, "16", v8i16, NEONvsli> {
4061 let Inst{21-20} = 0b01; // imm6 = 01xxxx
4063 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
4064 N2RegVShLFrm, OpcodeStr, "32", v4i32, NEONvsli> {
4065 let Inst{21} = 0b1; // imm6 = 1xxxxx
4067 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4, i32imm,
4068 N2RegVShLFrm, OpcodeStr, "64", v2i64, NEONvsli>;
4071 multiclass N2VShInsR_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
4073 // 64-bit vector types.
4074 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm8,
4075 N2RegVShRFrm, OpcodeStr, "8", v8i8, NEONvsri> {
4076 let Inst{21-19} = 0b001; // imm6 = 001xxx
4078 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm16,
4079 N2RegVShRFrm, OpcodeStr, "16", v4i16, NEONvsri> {
4080 let Inst{21-20} = 0b01; // imm6 = 01xxxx
4082 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm32,
4083 N2RegVShRFrm, OpcodeStr, "32", v2i32, NEONvsri> {
4084 let Inst{21} = 0b1; // imm6 = 1xxxxx
4086 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4, shr_imm64,
4087 N2RegVShRFrm, OpcodeStr, "64", v1i64, NEONvsri>;
4090 // 128-bit vector types.
4091 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm8,
4092 N2RegVShRFrm, OpcodeStr, "8", v16i8, NEONvsri> {
4093 let Inst{21-19} = 0b001; // imm6 = 001xxx
4095 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm16,
4096 N2RegVShRFrm, OpcodeStr, "16", v8i16, NEONvsri> {
4097 let Inst{21-20} = 0b01; // imm6 = 01xxxx
4099 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm32,
4100 N2RegVShRFrm, OpcodeStr, "32", v4i32, NEONvsri> {
4101 let Inst{21} = 0b1; // imm6 = 1xxxxx
4103 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4, shr_imm64,
4104 N2RegVShRFrm, OpcodeStr, "64", v2i64, NEONvsri>;
4108 // Neon Shift Long operations,
4109 // element sizes of 8, 16, 32 bits:
4110 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
4111 bit op4, string OpcodeStr, string Dt,
4112 SDPatternOperator OpNode> {
4113 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
4114 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, imm1_7, OpNode> {
4115 let Inst{21-19} = 0b001; // imm6 = 001xxx
4117 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
4118 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, imm1_15, OpNode> {
4119 let Inst{21-20} = 0b01; // imm6 = 01xxxx
4121 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
4122 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, imm1_31, OpNode> {
4123 let Inst{21} = 0b1; // imm6 = 1xxxxx
4127 // Neon Shift Narrow operations,
4128 // element sizes of 16, 32, 64 bits:
4129 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
4130 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
4131 SDPatternOperator OpNode> {
4132 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
4133 OpcodeStr, !strconcat(Dt, "16"),
4134 v8i8, v8i16, shr_imm8, OpNode> {
4135 let Inst{21-19} = 0b001; // imm6 = 001xxx
4137 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
4138 OpcodeStr, !strconcat(Dt, "32"),
4139 v4i16, v4i32, shr_imm16, OpNode> {
4140 let Inst{21-20} = 0b01; // imm6 = 01xxxx
4142 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
4143 OpcodeStr, !strconcat(Dt, "64"),
4144 v2i32, v2i64, shr_imm32, OpNode> {
4145 let Inst{21} = 0b1; // imm6 = 1xxxxx
4149 //===----------------------------------------------------------------------===//
4150 // Instruction Definitions.
4151 //===----------------------------------------------------------------------===//
4153 // Vector Add Operations.
4155 // VADD : Vector Add (integer and floating-point)
4156 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
4158 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
4159 v2f32, v2f32, fadd, 1>;
4160 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
4161 v4f32, v4f32, fadd, 1>;
4162 def VADDhd : N3VD<0, 0, 0b01, 0b1101, 0, IIC_VBIND, "vadd", "f16",
4163 v4f16, v4f16, fadd, 1>,
4164 Requires<[HasNEON,HasFullFP16]>;
4165 def VADDhq : N3VQ<0, 0, 0b01, 0b1101, 0, IIC_VBINQ, "vadd", "f16",
4166 v8f16, v8f16, fadd, 1>,
4167 Requires<[HasNEON,HasFullFP16]>;
4168 // VADDL : Vector Add Long (Q = D + D)
4169 defm VADDLs : N3VLExt_QHS<0,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
4170 "vaddl", "s", add, sext, 1>;
4171 defm VADDLu : N3VLExt_QHS<1,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
4172 "vaddl", "u", add, zext, 1>;
4173 // VADDW : Vector Add Wide (Q = Q + D)
4174 defm VADDWs : N3VW_QHS<0,1,0b0001,0, "vaddw", "s", add, sext, 0>;
4175 defm VADDWu : N3VW_QHS<1,1,0b0001,0, "vaddw", "u", add, zext, 0>;
4176 // VHADD : Vector Halving Add
4177 defm VHADDs : N3VInt_QHS<0, 0, 0b0000, 0, N3RegFrm,
4178 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
4179 "vhadd", "s", int_arm_neon_vhadds, 1>;
4180 defm VHADDu : N3VInt_QHS<1, 0, 0b0000, 0, N3RegFrm,
4181 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
4182 "vhadd", "u", int_arm_neon_vhaddu, 1>;
4183 // VRHADD : Vector Rounding Halving Add
4184 defm VRHADDs : N3VInt_QHS<0, 0, 0b0001, 0, N3RegFrm,
4185 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
4186 "vrhadd", "s", int_arm_neon_vrhadds, 1>;
4187 defm VRHADDu : N3VInt_QHS<1, 0, 0b0001, 0, N3RegFrm,
4188 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
4189 "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
4190 // VQADD : Vector Saturating Add
4191 defm VQADDs : N3VInt_QHSD<0, 0, 0b0000, 1, N3RegFrm,
4192 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
4193 "vqadd", "s", int_arm_neon_vqadds, 1>;
4194 defm VQADDu : N3VInt_QHSD<1, 0, 0b0000, 1, N3RegFrm,
4195 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
4196 "vqadd", "u", int_arm_neon_vqaddu, 1>;
4197 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
4198 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i", null_frag, 1>;
4199 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
4200 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
4201 int_arm_neon_vraddhn, 1>;
4203 def : Pat<(v8i8 (trunc (NEONvshru (add (v8i16 QPR:$Vn), QPR:$Vm), 8))),
4204 (VADDHNv8i8 QPR:$Vn, QPR:$Vm)>;
4205 def : Pat<(v4i16 (trunc (NEONvshru (add (v4i32 QPR:$Vn), QPR:$Vm), 16))),
4206 (VADDHNv4i16 QPR:$Vn, QPR:$Vm)>;
4207 def : Pat<(v2i32 (trunc (NEONvshru (add (v2i64 QPR:$Vn), QPR:$Vm), 32))),
4208 (VADDHNv2i32 QPR:$Vn, QPR:$Vm)>;
4210 // Vector Multiply Operations.
4212 // VMUL : Vector Multiply (integer, polynomial and floating-point)
4213 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
4214 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
4215 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16D, "vmul",
4216 "p8", v8i8, v8i8, int_arm_neon_vmulp, 1>;
4217 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16Q, "vmul",
4218 "p8", v16i8, v16i8, int_arm_neon_vmulp, 1>;
4219 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VFMULD, "vmul", "f32",
4220 v2f32, v2f32, fmul, 1>;
4221 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VFMULQ, "vmul", "f32",
4222 v4f32, v4f32, fmul, 1>;
4223 def VMULhd : N3VD<1, 0, 0b01, 0b1101, 1, IIC_VFMULD, "vmul", "f16",
4224 v4f16, v4f16, fmul, 1>,
4225 Requires<[HasNEON,HasFullFP16]>;
4226 def VMULhq : N3VQ<1, 0, 0b01, 0b1101, 1, IIC_VFMULQ, "vmul", "f16",
4227 v8f16, v8f16, fmul, 1>,
4228 Requires<[HasNEON,HasFullFP16]>;
4229 defm VMULsl : N3VSL_HS<0b1000, "vmul", mul>;
4230 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
4231 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
4233 def VMULslhd : N3VDSL16<0b01, 0b1001, "vmul", "f16", v4f16, fmul>,
4234 Requires<[HasNEON,HasFullFP16]>;
4235 def VMULslhq : N3VQSL16<0b01, 0b1001, "vmul", "f16", v8f16,
4237 Requires<[HasNEON,HasFullFP16]>;
4239 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
4240 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
4241 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
4242 (v4i16 (EXTRACT_SUBREG QPR:$src2,
4243 (DSubReg_i16_reg imm:$lane))),
4244 (SubReg_i16_lane imm:$lane)))>;
4245 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
4246 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
4247 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
4248 (v2i32 (EXTRACT_SUBREG QPR:$src2,
4249 (DSubReg_i32_reg imm:$lane))),
4250 (SubReg_i32_lane imm:$lane)))>;
4251 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
4252 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
4253 (v4f32 (VMULslfq (v4f32 QPR:$src1),
4254 (v2f32 (EXTRACT_SUBREG QPR:$src2,
4255 (DSubReg_i32_reg imm:$lane))),
4256 (SubReg_i32_lane imm:$lane)))>;
4259 def : Pat<(v2f32 (fmul DPR:$Rn, (NEONvdup (f32 SPR:$Rm)))),
4261 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$Rm, ssub_0),
4263 def : Pat<(v4f32 (fmul QPR:$Rn, (NEONvdup (f32 SPR:$Rm)))),
4265 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$Rm, ssub_0),
4269 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
4270 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, N3RegFrm, IIC_VMULi16D, IIC_VMULi32D,
4271 IIC_VMULi16Q, IIC_VMULi32Q,
4272 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
4273 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
4274 IIC_VMULi16Q, IIC_VMULi32Q,
4275 "vqdmulh", "s", int_arm_neon_vqdmulh>;
4276 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
4277 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
4279 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
4280 (v4i16 (EXTRACT_SUBREG QPR:$src2,
4281 (DSubReg_i16_reg imm:$lane))),
4282 (SubReg_i16_lane imm:$lane)))>;
4283 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
4284 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
4286 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
4287 (v2i32 (EXTRACT_SUBREG QPR:$src2,
4288 (DSubReg_i32_reg imm:$lane))),
4289 (SubReg_i32_lane imm:$lane)))>;
4291 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
4292 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, N3RegFrm,
4293 IIC_VMULi16D,IIC_VMULi32D,IIC_VMULi16Q,IIC_VMULi32Q,
4294 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
4295 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
4296 IIC_VMULi16Q, IIC_VMULi32Q,
4297 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
4298 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
4299 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
4301 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
4302 (v4i16 (EXTRACT_SUBREG QPR:$src2,
4303 (DSubReg_i16_reg imm:$lane))),
4304 (SubReg_i16_lane imm:$lane)))>;
4305 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
4306 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
4308 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
4309 (v2i32 (EXTRACT_SUBREG QPR:$src2,
4310 (DSubReg_i32_reg imm:$lane))),
4311 (SubReg_i32_lane imm:$lane)))>;
4313 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
4314 let PostEncoderMethod = "NEONThumb2DataIPostEncoder",
4315 DecoderNamespace = "NEONData" in {
4316 defm VMULLs : N3VL_QHS<0,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
4317 "vmull", "s", NEONvmulls, 1>;
4318 defm VMULLu : N3VL_QHS<1,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
4319 "vmull", "u", NEONvmullu, 1>;
4320 def VMULLp8 : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
4321 v8i16, v8i8, int_arm_neon_vmullp, 1>;
4322 def VMULLp64 : N3VLIntnp<0b00101, 0b10, 0b1110, 0, 0, NoItinerary,
4323 "vmull", "p64", v2i64, v1i64, int_arm_neon_vmullp, 1>,
4324 Requires<[HasV8, HasCrypto]>;
4326 defm VMULLsls : N3VLSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s", NEONvmulls>;
4327 defm VMULLslu : N3VLSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u", NEONvmullu>;
4329 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
4330 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, IIC_VMULi32D,
4331 "vqdmull", "s", int_arm_neon_vqdmull, 1>;
4332 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D,
4333 "vqdmull", "s", int_arm_neon_vqdmull>;
4335 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
4337 // VMLA : Vector Multiply Accumulate (integer and floating-point)
4338 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
4339 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
4340 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
4341 v2f32, fmul_su, fadd_mlx>,
4342 Requires<[HasNEON, UseFPVMLx, DontUseFusedMAC]>;
4343 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
4344 v4f32, fmul_su, fadd_mlx>,
4345 Requires<[HasNEON, UseFPVMLx, DontUseFusedMAC]>;
4346 def VMLAhd : N3VDMulOp<0, 0, 0b01, 0b1101, 1, IIC_VMACD, "vmla", "f16",
4347 v4f16, fmul_su, fadd_mlx>,
4348 Requires<[HasNEON, HasFullFP16, UseFPVMLx, DontUseFusedMAC]>;
4349 def VMLAhq : N3VQMulOp<0, 0, 0b01, 0b1101, 1, IIC_VMACQ, "vmla", "f16",
4350 v8f16, fmul_su, fadd_mlx>,
4351 Requires<[HasNEON, HasFullFP16, UseFPVMLx, DontUseFusedMAC]>;
4352 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
4353 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
4354 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
4355 v2f32, fmul_su, fadd_mlx>,
4356 Requires<[HasNEON, UseFPVMLx]>;
4357 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
4358 v4f32, v2f32, fmul_su, fadd_mlx>,
4359 Requires<[HasNEON, UseFPVMLx]>;
4360 def VMLAslhd : N3VDMulOpSL16<0b01, 0b0001, IIC_VMACD, "vmla", "f16",
4362 Requires<[HasNEON, HasFullFP16, UseFPVMLx]>;
4363 def VMLAslhq : N3VQMulOpSL16<0b01, 0b0001, IIC_VMACQ, "vmla", "f16",
4364 v8f16, v4f16, fmul, fadd>,
4365 Requires<[HasNEON, HasFullFP16, UseFPVMLx]>;
4367 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
4368 (mul (v8i16 QPR:$src2),
4369 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
4370 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
4371 (v4i16 (EXTRACT_SUBREG QPR:$src3,
4372 (DSubReg_i16_reg imm:$lane))),
4373 (SubReg_i16_lane imm:$lane)))>;
4375 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
4376 (mul (v4i32 QPR:$src2),
4377 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
4378 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
4379 (v2i32 (EXTRACT_SUBREG QPR:$src3,
4380 (DSubReg_i32_reg imm:$lane))),
4381 (SubReg_i32_lane imm:$lane)))>;
4383 def : Pat<(v4f32 (fadd_mlx (v4f32 QPR:$src1),
4384 (fmul_su (v4f32 QPR:$src2),
4385 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
4386 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
4388 (v2f32 (EXTRACT_SUBREG QPR:$src3,
4389 (DSubReg_i32_reg imm:$lane))),
4390 (SubReg_i32_lane imm:$lane)))>,
4391 Requires<[HasNEON, UseFPVMLx]>;
4393 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
4394 defm VMLALs : N3VLMulOp_QHS<0,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
4395 "vmlal", "s", NEONvmulls, add>;
4396 defm VMLALu : N3VLMulOp_QHS<1,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
4397 "vmlal", "u", NEONvmullu, add>;
4399 defm VMLALsls : N3VLMulOpSL_HS<0, 0b0010, "vmlal", "s", NEONvmulls, add>;
4400 defm VMLALslu : N3VLMulOpSL_HS<1, 0b0010, "vmlal", "u", NEONvmullu, add>;
4402 let Predicates = [HasNEON, HasV8_1a] in {
4403 // v8.1a Neon Rounding Double Multiply-Op vector operations,
4404 // VQRDMLAH : Vector Saturating Rounding Doubling Multiply Accumulate Long
4406 defm VQRDMLAH : N3VInt3_HS<1, 0, 0b1011, 1, IIC_VMACi16D, IIC_VMACi32D,
4407 IIC_VMACi16Q, IIC_VMACi32Q, "vqrdmlah", "s",
4409 def : Pat<(v4i16 (int_arm_neon_vqadds
4411 (v4i16 (int_arm_neon_vqrdmulh (v4i16 DPR:$Vn),
4412 (v4i16 DPR:$Vm))))),
4413 (v4i16 (VQRDMLAHv4i16 DPR:$src1, DPR:$Vn, DPR:$Vm))>;
4414 def : Pat<(v2i32 (int_arm_neon_vqadds
4416 (v2i32 (int_arm_neon_vqrdmulh (v2i32 DPR:$Vn),
4417 (v2i32 DPR:$Vm))))),
4418 (v2i32 (VQRDMLAHv2i32 DPR:$src1, DPR:$Vn, DPR:$Vm))>;
4419 def : Pat<(v8i16 (int_arm_neon_vqadds
4421 (v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$Vn),
4422 (v8i16 QPR:$Vm))))),
4423 (v8i16 (VQRDMLAHv8i16 QPR:$src1, QPR:$Vn, QPR:$Vm))>;
4424 def : Pat<(v4i32 (int_arm_neon_vqadds
4426 (v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$Vn),
4427 (v4i32 QPR:$Vm))))),
4428 (v4i32 (VQRDMLAHv4i32 QPR:$src1, QPR:$Vn, QPR:$Vm))>;
4430 defm VQRDMLAHsl : N3VMulOpSL_HS<0b1110, IIC_VMACi16D, IIC_VMACi32D,
4431 IIC_VMACi16Q, IIC_VMACi32Q, "vqrdmlah", "s",
4433 def : Pat<(v4i16 (int_arm_neon_vqadds
4435 (v4i16 (int_arm_neon_vqrdmulh
4437 (v4i16 (NEONvduplane (v4i16 DPR_8:$Vm),
4439 (v4i16 (VQRDMLAHslv4i16 DPR:$src1, DPR:$Vn, DPR_8:$Vm,
4441 def : Pat<(v2i32 (int_arm_neon_vqadds
4443 (v2i32 (int_arm_neon_vqrdmulh
4445 (v2i32 (NEONvduplane (v2i32 DPR_VFP2:$Vm),
4447 (v2i32 (VQRDMLAHslv2i32 DPR:$src1, DPR:$Vn, DPR_VFP2:$Vm,
4449 def : Pat<(v8i16 (int_arm_neon_vqadds
4451 (v8i16 (int_arm_neon_vqrdmulh
4453 (v8i16 (NEONvduplane (v8i16 QPR:$src3),
4455 (v8i16 (VQRDMLAHslv8i16 (v8i16 QPR:$src1),
4457 (v4i16 (EXTRACT_SUBREG
4459 (DSubReg_i16_reg imm:$lane))),
4460 (SubReg_i16_lane imm:$lane)))>;
4461 def : Pat<(v4i32 (int_arm_neon_vqadds
4463 (v4i32 (int_arm_neon_vqrdmulh
4465 (v4i32 (NEONvduplane (v4i32 QPR:$src3),
4467 (v4i32 (VQRDMLAHslv4i32 (v4i32 QPR:$src1),
4469 (v2i32 (EXTRACT_SUBREG
4471 (DSubReg_i32_reg imm:$lane))),
4472 (SubReg_i32_lane imm:$lane)))>;
4474 // VQRDMLSH : Vector Saturating Rounding Doubling Multiply Subtract Long
4476 defm VQRDMLSH : N3VInt3_HS<1, 0, 0b1100, 1, IIC_VMACi16D, IIC_VMACi32D,
4477 IIC_VMACi16Q, IIC_VMACi32Q, "vqrdmlsh", "s",
4479 def : Pat<(v4i16 (int_arm_neon_vqsubs
4481 (v4i16 (int_arm_neon_vqrdmulh (v4i16 DPR:$Vn),
4482 (v4i16 DPR:$Vm))))),
4483 (v4i16 (VQRDMLSHv4i16 DPR:$src1, DPR:$Vn, DPR:$Vm))>;
4484 def : Pat<(v2i32 (int_arm_neon_vqsubs
4486 (v2i32 (int_arm_neon_vqrdmulh (v2i32 DPR:$Vn),
4487 (v2i32 DPR:$Vm))))),
4488 (v2i32 (VQRDMLSHv2i32 DPR:$src1, DPR:$Vn, DPR:$Vm))>;
4489 def : Pat<(v8i16 (int_arm_neon_vqsubs
4491 (v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$Vn),
4492 (v8i16 QPR:$Vm))))),
4493 (v8i16 (VQRDMLSHv8i16 QPR:$src1, QPR:$Vn, QPR:$Vm))>;
4494 def : Pat<(v4i32 (int_arm_neon_vqsubs
4496 (v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$Vn),
4497 (v4i32 QPR:$Vm))))),
4498 (v4i32 (VQRDMLSHv4i32 QPR:$src1, QPR:$Vn, QPR:$Vm))>;
4500 defm VQRDMLSHsl : N3VMulOpSL_HS<0b1111, IIC_VMACi16D, IIC_VMACi32D,
4501 IIC_VMACi16Q, IIC_VMACi32Q, "vqrdmlsh", "s",
4503 def : Pat<(v4i16 (int_arm_neon_vqsubs
4505 (v4i16 (int_arm_neon_vqrdmulh
4507 (v4i16 (NEONvduplane (v4i16 DPR_8:$Vm),
4509 (v4i16 (VQRDMLSHslv4i16 DPR:$src1, DPR:$Vn, DPR_8:$Vm, imm:$lane))>;
4510 def : Pat<(v2i32 (int_arm_neon_vqsubs
4512 (v2i32 (int_arm_neon_vqrdmulh
4514 (v2i32 (NEONvduplane (v2i32 DPR_VFP2:$Vm),
4516 (v2i32 (VQRDMLSHslv2i32 DPR:$src1, DPR:$Vn, DPR_VFP2:$Vm,
4518 def : Pat<(v8i16 (int_arm_neon_vqsubs
4520 (v8i16 (int_arm_neon_vqrdmulh
4522 (v8i16 (NEONvduplane (v8i16 QPR:$src3),
4524 (v8i16 (VQRDMLSHslv8i16 (v8i16 QPR:$src1),
4526 (v4i16 (EXTRACT_SUBREG
4528 (DSubReg_i16_reg imm:$lane))),
4529 (SubReg_i16_lane imm:$lane)))>;
4530 def : Pat<(v4i32 (int_arm_neon_vqsubs
4532 (v4i32 (int_arm_neon_vqrdmulh
4534 (v4i32 (NEONvduplane (v4i32 QPR:$src3),
4536 (v4i32 (VQRDMLSHslv4i32 (v4i32 QPR:$src1),
4538 (v2i32 (EXTRACT_SUBREG
4540 (DSubReg_i32_reg imm:$lane))),
4541 (SubReg_i32_lane imm:$lane)))>;
4543 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
4544 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
4545 "vqdmlal", "s", null_frag>;
4546 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", null_frag>;
4548 def : Pat<(v4i32 (int_arm_neon_vqadds (v4i32 QPR:$src1),
4549 (v4i32 (int_arm_neon_vqdmull (v4i16 DPR:$Vn),
4550 (v4i16 DPR:$Vm))))),
4551 (VQDMLALv4i32 QPR:$src1, DPR:$Vn, DPR:$Vm)>;
4552 def : Pat<(v2i64 (int_arm_neon_vqadds (v2i64 QPR:$src1),
4553 (v2i64 (int_arm_neon_vqdmull (v2i32 DPR:$Vn),
4554 (v2i32 DPR:$Vm))))),
4555 (VQDMLALv2i64 QPR:$src1, DPR:$Vn, DPR:$Vm)>;
4556 def : Pat<(v4i32 (int_arm_neon_vqadds (v4i32 QPR:$src1),
4557 (v4i32 (int_arm_neon_vqdmull (v4i16 DPR:$Vn),
4558 (v4i16 (NEONvduplane (v4i16 DPR_8:$Vm),
4560 (VQDMLALslv4i16 QPR:$src1, DPR:$Vn, DPR_8:$Vm, imm:$lane)>;
4561 def : Pat<(v2i64 (int_arm_neon_vqadds (v2i64 QPR:$src1),
4562 (v2i64 (int_arm_neon_vqdmull (v2i32 DPR:$Vn),
4563 (v2i32 (NEONvduplane (v2i32 DPR_VFP2:$Vm),
4565 (VQDMLALslv2i32 QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, imm:$lane)>;
4567 // VMLS : Vector Multiply Subtract (integer and floating-point)
4568 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
4569 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
4570 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
4571 v2f32, fmul_su, fsub_mlx>,
4572 Requires<[HasNEON, UseFPVMLx, DontUseFusedMAC]>;
4573 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
4574 v4f32, fmul_su, fsub_mlx>,
4575 Requires<[HasNEON, UseFPVMLx, DontUseFusedMAC]>;
4576 def VMLShd : N3VDMulOp<0, 0, 0b11, 0b1101, 1, IIC_VMACD, "vmls", "f16",
4578 Requires<[HasNEON, HasFullFP16, UseFPVMLx, DontUseFusedMAC]>;
4579 def VMLShq : N3VQMulOp<0, 0, 0b11, 0b1101, 1, IIC_VMACQ, "vmls", "f16",
4581 Requires<[HasNEON, HasFullFP16, UseFPVMLx, DontUseFusedMAC]>;
4582 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
4583 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
4584 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
4585 v2f32, fmul_su, fsub_mlx>,
4586 Requires<[HasNEON, UseFPVMLx]>;
4587 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
4588 v4f32, v2f32, fmul_su, fsub_mlx>,
4589 Requires<[HasNEON, UseFPVMLx]>;
4590 def VMLSslhd : N3VDMulOpSL16<0b01, 0b0101, IIC_VMACD, "vmls", "f16",
4592 Requires<[HasNEON, HasFullFP16, UseFPVMLx]>;
4593 def VMLSslhq : N3VQMulOpSL16<0b01, 0b0101, IIC_VMACQ, "vmls", "f16",
4594 v8f16, v4f16, fmul, fsub>,
4595 Requires<[HasNEON, HasFullFP16, UseFPVMLx]>;
4597 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
4598 (mul (v8i16 QPR:$src2),
4599 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
4600 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
4601 (v4i16 (EXTRACT_SUBREG QPR:$src3,
4602 (DSubReg_i16_reg imm:$lane))),
4603 (SubReg_i16_lane imm:$lane)))>;
4605 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
4606 (mul (v4i32 QPR:$src2),
4607 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
4608 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
4609 (v2i32 (EXTRACT_SUBREG QPR:$src3,
4610 (DSubReg_i32_reg imm:$lane))),
4611 (SubReg_i32_lane imm:$lane)))>;
4613 def : Pat<(v4f32 (fsub_mlx (v4f32 QPR:$src1),
4614 (fmul_su (v4f32 QPR:$src2),
4615 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
4616 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
4617 (v2f32 (EXTRACT_SUBREG QPR:$src3,
4618 (DSubReg_i32_reg imm:$lane))),
4619 (SubReg_i32_lane imm:$lane)))>,
4620 Requires<[HasNEON, UseFPVMLx]>;
4622 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
4623 defm VMLSLs : N3VLMulOp_QHS<0,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
4624 "vmlsl", "s", NEONvmulls, sub>;
4625 defm VMLSLu : N3VLMulOp_QHS<1,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
4626 "vmlsl", "u", NEONvmullu, sub>;
4628 defm VMLSLsls : N3VLMulOpSL_HS<0, 0b0110, "vmlsl", "s", NEONvmulls, sub>;
4629 defm VMLSLslu : N3VLMulOpSL_HS<1, 0b0110, "vmlsl", "u", NEONvmullu, sub>;
4631 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
4632 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, IIC_VMACi16D, IIC_VMACi32D,
4633 "vqdmlsl", "s", null_frag>;
4634 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b0111, "vqdmlsl", "s", null_frag>;
4636 def : Pat<(v4i32 (int_arm_neon_vqsubs (v4i32 QPR:$src1),
4637 (v4i32 (int_arm_neon_vqdmull (v4i16 DPR:$Vn),
4638 (v4i16 DPR:$Vm))))),
4639 (VQDMLSLv4i32 QPR:$src1, DPR:$Vn, DPR:$Vm)>;
4640 def : Pat<(v2i64 (int_arm_neon_vqsubs (v2i64 QPR:$src1),
4641 (v2i64 (int_arm_neon_vqdmull (v2i32 DPR:$Vn),
4642 (v2i32 DPR:$Vm))))),
4643 (VQDMLSLv2i64 QPR:$src1, DPR:$Vn, DPR:$Vm)>;
4644 def : Pat<(v4i32 (int_arm_neon_vqsubs (v4i32 QPR:$src1),
4645 (v4i32 (int_arm_neon_vqdmull (v4i16 DPR:$Vn),
4646 (v4i16 (NEONvduplane (v4i16 DPR_8:$Vm),
4648 (VQDMLSLslv4i16 QPR:$src1, DPR:$Vn, DPR_8:$Vm, imm:$lane)>;
4649 def : Pat<(v2i64 (int_arm_neon_vqsubs (v2i64 QPR:$src1),
4650 (v2i64 (int_arm_neon_vqdmull (v2i32 DPR:$Vn),
4651 (v2i32 (NEONvduplane (v2i32 DPR_VFP2:$Vm),
4653 (VQDMLSLslv2i32 QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, imm:$lane)>;
4655 // Fused Vector Multiply-Accumulate and Fused Multiply-Subtract Operations.
4656 def VFMAfd : N3VDMulOp<0, 0, 0b00, 0b1100, 1, IIC_VFMACD, "vfma", "f32",
4657 v2f32, fmul_su, fadd_mlx>,
4658 Requires<[HasNEON,HasVFP4,UseFusedMAC]>;
4660 def VFMAfq : N3VQMulOp<0, 0, 0b00, 0b1100, 1, IIC_VFMACQ, "vfma", "f32",
4661 v4f32, fmul_su, fadd_mlx>,
4662 Requires<[HasNEON,HasVFP4,UseFusedMAC]>;
4663 def VFMAhd : N3VDMulOp<0, 0, 0b01, 0b1100, 1, IIC_VFMACD, "vfma", "f16",
4665 Requires<[HasNEON,HasFullFP16,UseFusedMAC]>;
4667 def VFMAhq : N3VQMulOp<0, 0, 0b01, 0b1100, 1, IIC_VFMACQ, "vfma", "f16",
4669 Requires<[HasNEON,HasFullFP16,UseFusedMAC]>;
4671 // Fused Vector Multiply Subtract (floating-point)
4672 def VFMSfd : N3VDMulOp<0, 0, 0b10, 0b1100, 1, IIC_VFMACD, "vfms", "f32",
4673 v2f32, fmul_su, fsub_mlx>,
4674 Requires<[HasNEON,HasVFP4,UseFusedMAC]>;
4675 def VFMSfq : N3VQMulOp<0, 0, 0b10, 0b1100, 1, IIC_VFMACQ, "vfms", "f32",
4676 v4f32, fmul_su, fsub_mlx>,
4677 Requires<[HasNEON,HasVFP4,UseFusedMAC]>;
4678 def VFMShd : N3VDMulOp<0, 0, 0b11, 0b1100, 1, IIC_VFMACD, "vfms", "f16",
4680 Requires<[HasNEON,HasFullFP16,UseFusedMAC]>;
4681 def VFMShq : N3VQMulOp<0, 0, 0b11, 0b1100, 1, IIC_VFMACQ, "vfms", "f16",
4683 Requires<[HasNEON,HasFullFP16,UseFusedMAC]>;
4685 // Match @llvm.fma.* intrinsics
4686 def : Pat<(v2f32 (fma DPR:$Vn, DPR:$Vm, DPR:$src1)),
4687 (VFMAfd DPR:$src1, DPR:$Vn, DPR:$Vm)>,
4688 Requires<[HasVFP4]>;
4689 def : Pat<(v4f32 (fma QPR:$Vn, QPR:$Vm, QPR:$src1)),
4690 (VFMAfq QPR:$src1, QPR:$Vn, QPR:$Vm)>,
4691 Requires<[HasVFP4]>;
4692 def : Pat<(v2f32 (fma (fneg DPR:$Vn), DPR:$Vm, DPR:$src1)),
4693 (VFMSfd DPR:$src1, DPR:$Vn, DPR:$Vm)>,
4694 Requires<[HasVFP4]>;
4695 def : Pat<(v4f32 (fma (fneg QPR:$Vn), QPR:$Vm, QPR:$src1)),
4696 (VFMSfq QPR:$src1, QPR:$Vn, QPR:$Vm)>,
4697 Requires<[HasVFP4]>;
4699 // ARMv8.2a dot product instructions.
4700 // We put them in the VFPV8 decoder namespace because the ARM and Thumb
4701 // encodings are the same and thus no further bit twiddling is necessary
4702 // in the disassembler.
4703 let Predicates = [HasDotProd], DecoderNamespace = "VFPV8" in {
4705 def VUDOTD : N3Vnp<0b11000, 0b10, 0b1101, 0b0, 0b1,
4706 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm),
4707 N3RegFrm, IIC_VDOTPROD, "vudot", "u8", []>;
4708 def VSDOTD : N3Vnp<0b11000, 0b10, 0b1101, 0b0, 0b0,
4709 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm),
4710 N3RegFrm, IIC_VDOTPROD, "vsdot", "s8", []>;
4711 def VUDOTQ : N3Vnp<0b11000, 0b10, 0b1101, 0b1, 0b1,
4712 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm),
4713 N3RegFrm, IIC_VDOTPROD, "vudot", "u8", []>;
4714 def VSDOTQ : N3Vnp<0b11000, 0b10, 0b1101, 0b1, 0b0,
4715 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm),
4716 N3RegFrm, IIC_VDOTPROD, "vsdot", "s8", []>;
4718 // Indexed dot product instructions:
4719 class DOTI<string opc, string dt, bit Q, bit U, RegisterClass Ty> :
4720 N3Vnp<0b11100, 0b10, 0b1101, Q, U,
4721 (outs Ty:$Vd), (ins Ty:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
4722 N3RegFrm, IIC_VDOTPROD, opc, dt, []> {
4725 let AsmString = !strconcat(opc, ".", dt, "\t$Vd, $Vn, $Vm$lane");
4728 def VUDOTDI : DOTI<"vudot", "u8", 0b0, 0b1, DPR>;
4729 def VSDOTDI : DOTI<"vsdot", "s8", 0b0, 0b0, DPR>;
4730 def VUDOTQI : DOTI<"vudot", "u8", 0b1, 0b1, QPR>;
4731 def VSDOTQI : DOTI<"vsdot", "s8", 0b1, 0b0, QPR>;
4735 // ARMv8.3 complex operations
4736 class BaseN3VCP8ComplexTied<bit op21, bit op4, bit s, bit q,
4737 InstrItinClass itin, dag oops, dag iops,
4738 string opc, string dt, list<dag> pattern>
4739 : N3VCP8<{?,?}, {op21,s}, q, op4, oops,
4740 iops, itin, opc, dt, "$Vd, $Vn, $Vm, $rot", "$src1 = $Vd", pattern>{
4742 let Inst{24-23} = rot;
4745 class BaseN3VCP8ComplexOdd<bit op23, bit op21, bit op4, bit s, bit q,
4746 InstrItinClass itin, dag oops, dag iops, string opc,
4747 string dt, list<dag> pattern>
4748 : N3VCP8<{?,op23}, {op21,s}, q, op4, oops,
4749 iops, itin, opc, dt, "$Vd, $Vn, $Vm, $rot", "", pattern> {
4754 class BaseN3VCP8ComplexTiedLane32<bit op4, bit s, bit q, InstrItinClass itin,
4755 dag oops, dag iops, string opc, string dt,
4757 : N3VLaneCP8<s, {?,?}, q, op4, oops, iops, itin, opc, dt,
4758 "$Vd, $Vn, $Vm$lane, $rot", "$src1 = $Vd", pattern> {
4762 let Inst{21-20} = rot;
4766 class BaseN3VCP8ComplexTiedLane64<bit op4, bit s, bit q, InstrItinClass itin,
4767 dag oops, dag iops, string opc, string dt,
4769 : N3VLaneCP8<s, {?,?}, q, op4, oops, iops, itin, opc, dt,
4770 "$Vd, $Vn, $Vm$lane, $rot", "$src1 = $Vd", pattern> {
4774 let Inst{21-20} = rot;
4775 let Inst{5} = Vm{4};
4776 // This is needed because the lane operand does not have any bits in the
4777 // encoding (it only has one possible value), so we need to manually set it
4778 // to it's default value.
4779 let DecoderMethod = "DecodeNEONComplexLane64Instruction";
4782 multiclass N3VCP8ComplexTied<bit op21, bit op4,
4783 string OpcodeStr, SDPatternOperator Op> {
4784 let Predicates = [HasNEON,HasV8_3a,HasFullFP16] in {
4785 def v4f16 : BaseN3VCP8ComplexTied<op21, op4, 0, 0, IIC_VMACD, (outs DPR:$Vd),
4786 (ins DPR:$src1, DPR:$Vn, DPR:$Vm, complexrotateop:$rot),
4787 OpcodeStr, "f16", []>;
4788 def v8f16 : BaseN3VCP8ComplexTied<op21, op4, 0, 1, IIC_VMACQ, (outs QPR:$Vd),
4789 (ins QPR:$src1, QPR:$Vn, QPR:$Vm, complexrotateop:$rot),
4790 OpcodeStr, "f16", []>;
4792 let Predicates = [HasNEON,HasV8_3a] in {
4793 def v2f32 : BaseN3VCP8ComplexTied<op21, op4, 1, 0, IIC_VMACD, (outs DPR:$Vd),
4794 (ins DPR:$src1, DPR:$Vn, DPR:$Vm, complexrotateop:$rot),
4795 OpcodeStr, "f32", []>;
4796 def v4f32 : BaseN3VCP8ComplexTied<op21, op4, 1, 1, IIC_VMACQ, (outs QPR:$Vd),
4797 (ins QPR:$src1, QPR:$Vn, QPR:$Vm, complexrotateop:$rot),
4798 OpcodeStr, "f32", []>;
4802 multiclass N3VCP8ComplexOdd<bit op23, bit op21, bit op4,
4803 string OpcodeStr, SDPatternOperator Op> {
4804 let Predicates = [HasNEON,HasV8_3a,HasFullFP16] in {
4805 def v4f16 : BaseN3VCP8ComplexOdd<op23, op21, op4, 0, 0, IIC_VMACD,
4807 (ins DPR:$Vn, DPR:$Vm, complexrotateopodd:$rot),
4808 OpcodeStr, "f16", []>;
4809 def v8f16 : BaseN3VCP8ComplexOdd<op23, op21, op4, 0, 1, IIC_VMACQ,
4811 (ins QPR:$Vn, QPR:$Vm, complexrotateopodd:$rot),
4812 OpcodeStr, "f16", []>;
4814 let Predicates = [HasNEON,HasV8_3a] in {
4815 def v2f32 : BaseN3VCP8ComplexOdd<op23, op21, op4, 1, 0, IIC_VMACD,
4817 (ins DPR:$Vn, DPR:$Vm, complexrotateopodd:$rot),
4818 OpcodeStr, "f32", []>;
4819 def v4f32 : BaseN3VCP8ComplexOdd<op23, op21, op4, 1, 1, IIC_VMACQ,
4821 (ins QPR:$Vn, QPR:$Vm, complexrotateopodd:$rot),
4822 OpcodeStr, "f32", []>;
4826 // These instructions index by pairs of lanes, so the VectorIndexes are twice
4827 // as wide as the data types.
4828 multiclass N3VCP8ComplexTiedLane<bit op4, string OpcodeStr,
4829 SDPatternOperator Op> {
4830 let Predicates = [HasNEON,HasV8_3a,HasFullFP16] in {
4831 def v4f16_indexed : BaseN3VCP8ComplexTiedLane32<op4, 0, 0, IIC_VMACD,
4833 (ins DPR:$src1, DPR:$Vn, DPR_VFP2:$Vm,
4834 VectorIndex32:$lane, complexrotateop:$rot),
4835 OpcodeStr, "f16", []>;
4836 def v8f16_indexed : BaseN3VCP8ComplexTiedLane32<op4, 0, 1, IIC_VMACQ,
4838 (ins QPR:$src1, QPR:$Vn, DPR_VFP2:$Vm,
4839 VectorIndex32:$lane, complexrotateop:$rot),
4840 OpcodeStr, "f16", []>;
4842 let Predicates = [HasNEON,HasV8_3a] in {
4843 def v2f32_indexed : BaseN3VCP8ComplexTiedLane64<op4, 1, 0, IIC_VMACD,
4845 (ins DPR:$src1, DPR:$Vn, DPR:$Vm, VectorIndex64:$lane,
4846 complexrotateop:$rot),
4847 OpcodeStr, "f32", []>;
4848 def v4f32_indexed : BaseN3VCP8ComplexTiedLane64<op4, 1, 1, IIC_VMACQ,
4850 (ins QPR:$src1, QPR:$Vn, DPR:$Vm, VectorIndex64:$lane,
4851 complexrotateop:$rot),
4852 OpcodeStr, "f32", []>;
4856 defm VCMLA : N3VCP8ComplexTied<1, 0, "vcmla", null_frag>;
4857 defm VCADD : N3VCP8ComplexOdd<1, 0, 0, "vcadd", null_frag>;
4858 defm VCMLA : N3VCP8ComplexTiedLane<0, "vcmla", null_frag>;
4860 // Vector Subtract Operations.
4862 // VSUB : Vector Subtract (integer and floating-point)
4863 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
4864 "vsub", "i", sub, 0>;
4865 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
4866 v2f32, v2f32, fsub, 0>;
4867 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
4868 v4f32, v4f32, fsub, 0>;
4869 def VSUBhd : N3VD<0, 0, 0b11, 0b1101, 0, IIC_VBIND, "vsub", "f16",
4870 v4f16, v4f16, fsub, 0>,
4871 Requires<[HasNEON,HasFullFP16]>;
4872 def VSUBhq : N3VQ<0, 0, 0b11, 0b1101, 0, IIC_VBINQ, "vsub", "f16",
4873 v8f16, v8f16, fsub, 0>,
4874 Requires<[HasNEON,HasFullFP16]>;
4875 // VSUBL : Vector Subtract Long (Q = D - D)
4876 defm VSUBLs : N3VLExt_QHS<0,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
4877 "vsubl", "s", sub, sext, 0>;
4878 defm VSUBLu : N3VLExt_QHS<1,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
4879 "vsubl", "u", sub, zext, 0>;
4880 // VSUBW : Vector Subtract Wide (Q = Q - D)
4881 defm VSUBWs : N3VW_QHS<0,1,0b0011,0, "vsubw", "s", sub, sext, 0>;
4882 defm VSUBWu : N3VW_QHS<1,1,0b0011,0, "vsubw", "u", sub, zext, 0>;
4883 // VHSUB : Vector Halving Subtract
4884 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, N3RegFrm,
4885 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4886 "vhsub", "s", int_arm_neon_vhsubs, 0>;
4887 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, N3RegFrm,
4888 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4889 "vhsub", "u", int_arm_neon_vhsubu, 0>;
4890 // VQSUB : Vector Saturing Subtract
4891 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, N3RegFrm,
4892 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4893 "vqsub", "s", int_arm_neon_vqsubs, 0>;
4894 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, N3RegFrm,
4895 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4896 "vqsub", "u", int_arm_neon_vqsubu, 0>;
4897 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
4898 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i", null_frag, 0>;
4899 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
4900 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
4901 int_arm_neon_vrsubhn, 0>;
4903 def : Pat<(v8i8 (trunc (NEONvshru (sub (v8i16 QPR:$Vn), QPR:$Vm), 8))),
4904 (VSUBHNv8i8 QPR:$Vn, QPR:$Vm)>;
4905 def : Pat<(v4i16 (trunc (NEONvshru (sub (v4i32 QPR:$Vn), QPR:$Vm), 16))),
4906 (VSUBHNv4i16 QPR:$Vn, QPR:$Vm)>;
4907 def : Pat<(v2i32 (trunc (NEONvshru (sub (v2i64 QPR:$Vn), QPR:$Vm), 32))),
4908 (VSUBHNv2i32 QPR:$Vn, QPR:$Vm)>;
4910 // Vector Comparisons.
4912 // VCEQ : Vector Compare Equal
4913 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
4914 IIC_VSUBi4Q, "vceq", "i", NEONvceq, 1>;
4915 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
4917 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
4919 def VCEQhd : N3VD<0,0,0b01,0b1110,0, IIC_VBIND, "vceq", "f16", v4i16, v4f16,
4921 Requires<[HasNEON, HasFullFP16]>;
4922 def VCEQhq : N3VQ<0,0,0b01,0b1110,0, IIC_VBINQ, "vceq", "f16", v8i16, v8f16,
4924 Requires<[HasNEON, HasFullFP16]>;
4926 let TwoOperandAliasConstraint = "$Vm = $Vd" in
4927 defm VCEQz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
4928 "$Vd, $Vm, #0", NEONvceqz>;
4930 // VCGE : Vector Compare Greater Than or Equal
4931 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
4932 IIC_VSUBi4Q, "vcge", "s", NEONvcge, 0>;
4933 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
4934 IIC_VSUBi4Q, "vcge", "u", NEONvcgeu, 0>;
4935 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32", v2i32, v2f32,
4937 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
4939 def VCGEhd : N3VD<1,0,0b01,0b1110,0, IIC_VBIND, "vcge", "f16", v4i16, v4f16,
4941 Requires<[HasNEON, HasFullFP16]>;
4942 def VCGEhq : N3VQ<1,0,0b01,0b1110,0, IIC_VBINQ, "vcge", "f16", v8i16, v8f16,
4944 Requires<[HasNEON, HasFullFP16]>;
4946 let TwoOperandAliasConstraint = "$Vm = $Vd" in {
4947 defm VCGEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
4948 "$Vd, $Vm, #0", NEONvcgez>;
4949 defm VCLEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
4950 "$Vd, $Vm, #0", NEONvclez>;
4953 // VCGT : Vector Compare Greater Than
4954 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
4955 IIC_VSUBi4Q, "vcgt", "s", NEONvcgt, 0>;
4956 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
4957 IIC_VSUBi4Q, "vcgt", "u", NEONvcgtu, 0>;
4958 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
4960 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
4962 def VCGThd : N3VD<1,0,0b11,0b1110,0, IIC_VBIND, "vcgt", "f16", v4i16, v4f16,
4964 Requires<[HasNEON, HasFullFP16]>;
4965 def VCGThq : N3VQ<1,0,0b11,0b1110,0, IIC_VBINQ, "vcgt", "f16", v8i16, v8f16,
4967 Requires<[HasNEON, HasFullFP16]>;
4969 let TwoOperandAliasConstraint = "$Vm = $Vd" in {
4970 defm VCGTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
4971 "$Vd, $Vm, #0", NEONvcgtz>;
4972 defm VCLTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
4973 "$Vd, $Vm, #0", NEONvcltz>;
4976 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
4977 def VACGEfd : N3VDInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacge",
4978 "f32", v2i32, v2f32, int_arm_neon_vacge, 0>;
4979 def VACGEfq : N3VQInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacge",
4980 "f32", v4i32, v4f32, int_arm_neon_vacge, 0>;
4981 def VACGEhd : N3VDInt<1, 0, 0b01, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacge",
4982 "f16", v4i16, v4f16, int_arm_neon_vacge, 0>,
4983 Requires<[HasNEON, HasFullFP16]>;
4984 def VACGEhq : N3VQInt<1, 0, 0b01, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacge",
4985 "f16", v8i16, v8f16, int_arm_neon_vacge, 0>,
4986 Requires<[HasNEON, HasFullFP16]>;
4987 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
4988 def VACGTfd : N3VDInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacgt",
4989 "f32", v2i32, v2f32, int_arm_neon_vacgt, 0>;
4990 def VACGTfq : N3VQInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacgt",
4991 "f32", v4i32, v4f32, int_arm_neon_vacgt, 0>;
4992 def VACGThd : N3VDInt<1, 0, 0b11, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacgt",
4993 "f16", v4i16, v4f16, int_arm_neon_vacgt, 0>,
4994 Requires<[HasNEON, HasFullFP16]>;
4995 def VACGThq : N3VQInt<1, 0, 0b11, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacgt",
4996 "f16", v8f16, v8f16, int_arm_neon_vacgt, 0>,
4997 Requires<[HasNEON, HasFullFP16]>;
4998 // VTST : Vector Test Bits
4999 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
5000 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
5002 def: NEONInstAlias<"vaclt${p}.f32 $Vd, $Vn, $Vm",
5003 (VACGTfd DPR:$Vd, DPR:$Vm, DPR:$Vn, pred:$p)>;
5004 def: NEONInstAlias<"vaclt${p}.f32 $Vd, $Vn, $Vm",
5005 (VACGTfq QPR:$Vd, QPR:$Vm, QPR:$Vn, pred:$p)>;
5006 def: NEONInstAlias<"vacle${p}.f32 $Vd, $Vn, $Vm",
5007 (VACGEfd DPR:$Vd, DPR:$Vm, DPR:$Vn, pred:$p)>;
5008 def: NEONInstAlias<"vacle${p}.f32 $Vd, $Vn, $Vm",
5009 (VACGEfq QPR:$Vd, QPR:$Vm, QPR:$Vn, pred:$p)>;
5010 let Predicates = [HasNEON, HasFullFP16] in {
5011 def: NEONInstAlias<"vaclt${p}.f16 $Vd, $Vn, $Vm",
5012 (VACGThd DPR:$Vd, DPR:$Vm, DPR:$Vn, pred:$p)>;
5013 def: NEONInstAlias<"vaclt${p}.f16 $Vd, $Vn, $Vm",
5014 (VACGThq QPR:$Vd, QPR:$Vm, QPR:$Vn, pred:$p)>;
5015 def: NEONInstAlias<"vacle${p}.f16 $Vd, $Vn, $Vm",
5016 (VACGEhd DPR:$Vd, DPR:$Vm, DPR:$Vn, pred:$p)>;
5017 def: NEONInstAlias<"vacle${p}.f16 $Vd, $Vn, $Vm",
5018 (VACGEhq QPR:$Vd, QPR:$Vm, QPR:$Vn, pred:$p)>;
5021 def: NEONInstAlias<"vaclt${p}.f32 $Vd, $Vm",
5022 (VACGTfd DPR:$Vd, DPR:$Vm, DPR:$Vd, pred:$p)>;
5023 def: NEONInstAlias<"vaclt${p}.f32 $Vd, $Vm",
5024 (VACGTfq QPR:$Vd, QPR:$Vm, QPR:$Vd, pred:$p)>;
5025 def: NEONInstAlias<"vacle${p}.f32 $Vd, $Vm",
5026 (VACGEfd DPR:$Vd, DPR:$Vm, DPR:$Vd, pred:$p)>;
5027 def: NEONInstAlias<"vacle${p}.f32 $Vd, $Vm",
5028 (VACGEfq QPR:$Vd, QPR:$Vm, QPR:$Vd, pred:$p)>;
5029 let Predicates = [HasNEON, HasFullFP16] in {
5030 def: NEONInstAlias<"vaclt${p}.f16 $Vd, $Vm",
5031 (VACGThd DPR:$Vd, DPR:$Vm, DPR:$Vd, pred:$p)>;
5032 def: NEONInstAlias<"vaclt${p}.f16 $Vd, $Vm",
5033 (VACGThq QPR:$Vd, QPR:$Vm, QPR:$Vd, pred:$p)>;
5034 def: NEONInstAlias<"vacle${p}.f16 $Vd, $Vm",
5035 (VACGEhd DPR:$Vd, DPR:$Vm, DPR:$Vd, pred:$p)>;
5036 def: NEONInstAlias<"vacle${p}.f16 $Vd, $Vm",
5037 (VACGEhq QPR:$Vd, QPR:$Vm, QPR:$Vd, pred:$p)>;
5040 // Vector Bitwise Operations.
5042 def vnotd : PatFrag<(ops node:$in),
5043 (xor node:$in, (bitconvert (v8i8 NEONimmAllOnesV)))>;
5044 def vnotq : PatFrag<(ops node:$in),
5045 (xor node:$in, (bitconvert (v16i8 NEONimmAllOnesV)))>;
5048 // VAND : Vector Bitwise AND
5049 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
5050 v2i32, v2i32, and, 1>;
5051 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
5052 v4i32, v4i32, and, 1>;
5054 // VEOR : Vector Bitwise Exclusive OR
5055 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
5056 v2i32, v2i32, xor, 1>;
5057 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
5058 v4i32, v4i32, xor, 1>;
5060 // VORR : Vector Bitwise OR
5061 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
5062 v2i32, v2i32, or, 1>;
5063 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
5064 v4i32, v4i32, or, 1>;
5066 def VORRiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 0, 1,
5067 (outs DPR:$Vd), (ins nImmSplatI16:$SIMM, DPR:$src),
5069 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
5071 (v4i16 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
5072 let Inst{9} = SIMM{9};
5075 def VORRiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 0, 1,
5076 (outs DPR:$Vd), (ins nImmSplatI32:$SIMM, DPR:$src),
5078 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
5080 (v2i32 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
5081 let Inst{10-9} = SIMM{10-9};
5084 def VORRiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 0, 1,
5085 (outs QPR:$Vd), (ins nImmSplatI16:$SIMM, QPR:$src),
5087 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
5089 (v8i16 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
5090 let Inst{9} = SIMM{9};
5093 def VORRiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 0, 1,
5094 (outs QPR:$Vd), (ins nImmSplatI32:$SIMM, QPR:$src),
5096 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
5098 (v4i32 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
5099 let Inst{10-9} = SIMM{10-9};
5103 // VBIC : Vector Bitwise Bit Clear (AND NOT)
5104 let TwoOperandAliasConstraint = "$Vn = $Vd" in {
5105 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
5106 (ins DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VBINiD,
5107 "vbic", "$Vd, $Vn, $Vm", "",
5108 [(set DPR:$Vd, (v2i32 (and DPR:$Vn,
5109 (vnotd DPR:$Vm))))]>;
5110 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
5111 (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINiQ,
5112 "vbic", "$Vd, $Vn, $Vm", "",
5113 [(set QPR:$Vd, (v4i32 (and QPR:$Vn,
5114 (vnotq QPR:$Vm))))]>;
5117 def VBICiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 1, 1,
5118 (outs DPR:$Vd), (ins nImmSplatI16:$SIMM, DPR:$src),
5120 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
5122 (v4i16 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
5123 let Inst{9} = SIMM{9};
5126 def VBICiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 1, 1,
5127 (outs DPR:$Vd), (ins nImmSplatI32:$SIMM, DPR:$src),
5129 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
5131 (v2i32 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
5132 let Inst{10-9} = SIMM{10-9};
5135 def VBICiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 1, 1,
5136 (outs QPR:$Vd), (ins nImmSplatI16:$SIMM, QPR:$src),
5138 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
5140 (v8i16 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
5141 let Inst{9} = SIMM{9};
5144 def VBICiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 1, 1,
5145 (outs QPR:$Vd), (ins nImmSplatI32:$SIMM, QPR:$src),
5147 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
5149 (v4i32 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
5150 let Inst{10-9} = SIMM{10-9};
5153 // VORN : Vector Bitwise OR NOT
5154 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$Vd),
5155 (ins DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VBINiD,
5156 "vorn", "$Vd, $Vn, $Vm", "",
5157 [(set DPR:$Vd, (v2i32 (or DPR:$Vn,
5158 (vnotd DPR:$Vm))))]>;
5159 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$Vd),
5160 (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINiQ,
5161 "vorn", "$Vd, $Vn, $Vm", "",
5162 [(set QPR:$Vd, (v4i32 (or QPR:$Vn,
5163 (vnotq QPR:$Vm))))]>;
5165 // VMVN : Vector Bitwise NOT (Immediate)
5167 let isReMaterializable = 1 in {
5169 def VMVNv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 1, 1, (outs DPR:$Vd),
5170 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
5171 "vmvn", "i16", "$Vd, $SIMM", "",
5172 [(set DPR:$Vd, (v4i16 (NEONvmvnImm timm:$SIMM)))]> {
5173 let Inst{9} = SIMM{9};
5176 def VMVNv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 1, 1, (outs QPR:$Vd),
5177 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
5178 "vmvn", "i16", "$Vd, $SIMM", "",
5179 [(set QPR:$Vd, (v8i16 (NEONvmvnImm timm:$SIMM)))]> {
5180 let Inst{9} = SIMM{9};
5183 def VMVNv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 1, 1, (outs DPR:$Vd),
5184 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
5185 "vmvn", "i32", "$Vd, $SIMM", "",
5186 [(set DPR:$Vd, (v2i32 (NEONvmvnImm timm:$SIMM)))]> {
5187 let Inst{11-8} = SIMM{11-8};
5190 def VMVNv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 1, 1, (outs QPR:$Vd),
5191 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
5192 "vmvn", "i32", "$Vd, $SIMM", "",
5193 [(set QPR:$Vd, (v4i32 (NEONvmvnImm timm:$SIMM)))]> {
5194 let Inst{11-8} = SIMM{11-8};
5198 // VMVN : Vector Bitwise NOT
5199 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
5200 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VSUBiD,
5201 "vmvn", "$Vd, $Vm", "",
5202 [(set DPR:$Vd, (v2i32 (vnotd DPR:$Vm)))]>;
5203 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
5204 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VSUBiD,
5205 "vmvn", "$Vd, $Vm", "",
5206 [(set QPR:$Vd, (v4i32 (vnotq QPR:$Vm)))]>;
5207 def : Pat<(v2i32 (vnotd DPR:$src)), (VMVNd DPR:$src)>;
5208 def : Pat<(v4i32 (vnotq QPR:$src)), (VMVNq QPR:$src)>;
5210 // VBSL : Vector Bitwise Select
5211 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
5212 (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
5213 N3RegFrm, IIC_VCNTiD,
5214 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
5216 (v2i32 (NEONvbsl DPR:$src1, DPR:$Vn, DPR:$Vm)))]>;
5217 def : Pat<(v8i8 (int_arm_neon_vbsl (v8i8 DPR:$src1),
5218 (v8i8 DPR:$Vn), (v8i8 DPR:$Vm))),
5219 (VBSLd DPR:$src1, DPR:$Vn, DPR:$Vm)>,
5220 Requires<[HasNEON]>;
5221 def : Pat<(v4i16 (int_arm_neon_vbsl (v4i16 DPR:$src1),
5222 (v4i16 DPR:$Vn), (v4i16 DPR:$Vm))),
5223 (VBSLd DPR:$src1, DPR:$Vn, DPR:$Vm)>,
5224 Requires<[HasNEON]>;
5225 def : Pat<(v2i32 (int_arm_neon_vbsl (v2i32 DPR:$src1),
5226 (v2i32 DPR:$Vn), (v2i32 DPR:$Vm))),
5227 (VBSLd DPR:$src1, DPR:$Vn, DPR:$Vm)>,
5228 Requires<[HasNEON]>;
5229 def : Pat<(v2f32 (int_arm_neon_vbsl (v2f32 DPR:$src1),
5230 (v2f32 DPR:$Vn), (v2f32 DPR:$Vm))),
5231 (VBSLd DPR:$src1, DPR:$Vn, DPR:$Vm)>,
5232 Requires<[HasNEON]>;
5233 def : Pat<(v1i64 (int_arm_neon_vbsl (v1i64 DPR:$src1),
5234 (v1i64 DPR:$Vn), (v1i64 DPR:$Vm))),
5235 (VBSLd DPR:$src1, DPR:$Vn, DPR:$Vm)>,
5236 Requires<[HasNEON]>;
5238 def : Pat<(v2i32 (or (and DPR:$Vn, DPR:$Vd),
5239 (and DPR:$Vm, (vnotd DPR:$Vd)))),
5240 (VBSLd DPR:$Vd, DPR:$Vn, DPR:$Vm)>,
5241 Requires<[HasNEON]>;
5243 def : Pat<(v1i64 (or (and DPR:$Vn, DPR:$Vd),
5244 (and DPR:$Vm, (vnotd DPR:$Vd)))),
5245 (VBSLd DPR:$Vd, DPR:$Vn, DPR:$Vm)>,
5246 Requires<[HasNEON]>;
5248 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
5249 (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
5250 N3RegFrm, IIC_VCNTiQ,
5251 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
5253 (v4i32 (NEONvbsl QPR:$src1, QPR:$Vn, QPR:$Vm)))]>;
5255 def : Pat<(v16i8 (int_arm_neon_vbsl (v16i8 QPR:$src1),
5256 (v16i8 QPR:$Vn), (v16i8 QPR:$Vm))),
5257 (VBSLq QPR:$src1, QPR:$Vn, QPR:$Vm)>,
5258 Requires<[HasNEON]>;
5259 def : Pat<(v8i16 (int_arm_neon_vbsl (v8i16 QPR:$src1),
5260 (v8i16 QPR:$Vn), (v8i16 QPR:$Vm))),
5261 (VBSLq QPR:$src1, QPR:$Vn, QPR:$Vm)>,
5262 Requires<[HasNEON]>;
5263 def : Pat<(v4i32 (int_arm_neon_vbsl (v4i32 QPR:$src1),
5264 (v4i32 QPR:$Vn), (v4i32 QPR:$Vm))),
5265 (VBSLq QPR:$src1, QPR:$Vn, QPR:$Vm)>,
5266 Requires<[HasNEON]>;
5267 def : Pat<(v4f32 (int_arm_neon_vbsl (v4f32 QPR:$src1),
5268 (v4f32 QPR:$Vn), (v4f32 QPR:$Vm))),
5269 (VBSLq QPR:$src1, QPR:$Vn, QPR:$Vm)>,
5270 Requires<[HasNEON]>;
5271 def : Pat<(v2i64 (int_arm_neon_vbsl (v2i64 QPR:$src1),
5272 (v2i64 QPR:$Vn), (v2i64 QPR:$Vm))),
5273 (VBSLq QPR:$src1, QPR:$Vn, QPR:$Vm)>,
5274 Requires<[HasNEON]>;
5276 def : Pat<(v4i32 (or (and QPR:$Vn, QPR:$Vd),
5277 (and QPR:$Vm, (vnotq QPR:$Vd)))),
5278 (VBSLq QPR:$Vd, QPR:$Vn, QPR:$Vm)>,
5279 Requires<[HasNEON]>;
5280 def : Pat<(v2i64 (or (and QPR:$Vn, QPR:$Vd),
5281 (and QPR:$Vm, (vnotq QPR:$Vd)))),
5282 (VBSLq QPR:$Vd, QPR:$Vn, QPR:$Vm)>,
5283 Requires<[HasNEON]>;
5285 // VBIF : Vector Bitwise Insert if False
5286 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
5287 // FIXME: This instruction's encoding MAY NOT BE correct.
5288 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
5289 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
5290 N3RegFrm, IIC_VBINiD,
5291 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
5293 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
5294 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
5295 N3RegFrm, IIC_VBINiQ,
5296 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
5299 // VBIT : Vector Bitwise Insert if True
5300 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
5301 // FIXME: This instruction's encoding MAY NOT BE correct.
5302 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
5303 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
5304 N3RegFrm, IIC_VBINiD,
5305 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
5307 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
5308 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
5309 N3RegFrm, IIC_VBINiQ,
5310 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
5313 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
5314 // for equivalent operations with different register constraints; it just
5317 // Vector Absolute Differences.
5319 // VABD : Vector Absolute Difference
5320 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, N3RegFrm,
5321 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
5322 "vabd", "s", int_arm_neon_vabds, 1>;
5323 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, N3RegFrm,
5324 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
5325 "vabd", "u", int_arm_neon_vabdu, 1>;
5326 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBIND,
5327 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 1>;
5328 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBINQ,
5329 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 1>;
5330 def VABDhd : N3VDInt<1, 0, 0b11, 0b1101, 0, N3RegFrm, IIC_VBIND,
5331 "vabd", "f16", v4f16, v4f16, int_arm_neon_vabds, 1>,
5332 Requires<[HasNEON, HasFullFP16]>;
5333 def VABDhq : N3VQInt<1, 0, 0b11, 0b1101, 0, N3RegFrm, IIC_VBINQ,
5334 "vabd", "f16", v8f16, v8f16, int_arm_neon_vabds, 1>,
5335 Requires<[HasNEON, HasFullFP16]>;
5337 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
5338 defm VABDLs : N3VLIntExt_QHS<0,1,0b0111,0, IIC_VSUBi4Q,
5339 "vabdl", "s", int_arm_neon_vabds, zext, 1>;
5340 defm VABDLu : N3VLIntExt_QHS<1,1,0b0111,0, IIC_VSUBi4Q,
5341 "vabdl", "u", int_arm_neon_vabdu, zext, 1>;
5344 PatFrag<(ops node:$in1, node:$in2, node:$shift),
5345 (NEONvshrs (sub (zext node:$in1),
5346 (zext node:$in2)), (i32 $shift))>;
5348 def : Pat<(xor (v4i32 (bitconvert (v8i16 (abd_shr (v8i8 DPR:$opA), (v8i8 DPR:$opB), 15)))),
5349 (v4i32 (bitconvert (v8i16 (add (sub (zext (v8i8 DPR:$opA)),
5350 (zext (v8i8 DPR:$opB))),
5351 (v8i16 (abd_shr (v8i8 DPR:$opA), (v8i8 DPR:$opB), 15))))))),
5352 (VABDLuv8i16 DPR:$opA, DPR:$opB)>;
5354 def : Pat<(xor (v4i32 (abd_shr (v4i16 DPR:$opA), (v4i16 DPR:$opB), 31)),
5355 (v4i32 (add (sub (zext (v4i16 DPR:$opA)),
5356 (zext (v4i16 DPR:$opB))),
5357 (abd_shr (v4i16 DPR:$opA), (v4i16 DPR:$opB), 31)))),
5358 (VABDLuv4i32 DPR:$opA, DPR:$opB)>;
5360 def : Pat<(xor (v4i32 (bitconvert (v2i64 (abd_shr (v2i32 DPR:$opA), (v2i32 DPR:$opB), 63)))),
5361 (v4i32 (bitconvert (v2i64 (add (sub (zext (v2i32 DPR:$opA)),
5362 (zext (v2i32 DPR:$opB))),
5363 (abd_shr (v2i32 DPR:$opA), (v2i32 DPR:$opB), 63)))))),
5364 (VABDLuv2i64 DPR:$opA, DPR:$opB)>;
5366 // VABA : Vector Absolute Difference and Accumulate
5367 defm VABAs : N3VIntOp_QHS<0,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
5368 "vaba", "s", int_arm_neon_vabds, add>;
5369 defm VABAu : N3VIntOp_QHS<1,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
5370 "vaba", "u", int_arm_neon_vabdu, add>;
5372 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
5373 defm VABALs : N3VLIntExtOp_QHS<0,1,0b0101,0, IIC_VABAD,
5374 "vabal", "s", int_arm_neon_vabds, zext, add>;
5375 defm VABALu : N3VLIntExtOp_QHS<1,1,0b0101,0, IIC_VABAD,
5376 "vabal", "u", int_arm_neon_vabdu, zext, add>;
5378 // Vector Maximum and Minimum.
5380 // VMAX : Vector Maximum
5381 defm VMAXs : N3VInt_QHS<0, 0, 0b0110, 0, N3RegFrm,
5382 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
5383 "vmax", "s", smax, 1>;
5384 defm VMAXu : N3VInt_QHS<1, 0, 0b0110, 0, N3RegFrm,
5385 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
5386 "vmax", "u", umax, 1>;
5387 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBIND,
5389 v2f32, v2f32, fmaxnan, 1>;
5390 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBINQ,
5392 v4f32, v4f32, fmaxnan, 1>;
5393 def VMAXhd : N3VDInt<0, 0, 0b01, 0b1111, 0, N3RegFrm, IIC_VBIND,
5395 v4f16, v4f16, fmaxnan, 1>,
5396 Requires<[HasNEON, HasFullFP16]>;
5397 def VMAXhq : N3VQInt<0, 0, 0b01, 0b1111, 0, N3RegFrm, IIC_VBINQ,
5399 v8f16, v8f16, fmaxnan, 1>,
5400 Requires<[HasNEON, HasFullFP16]>;
5403 let PostEncoderMethod = "NEONThumb2V8PostEncoder", DecoderNamespace = "v8NEON" in {
5404 def VMAXNMNDf : N3VDIntnp<0b00110, 0b00, 0b1111, 0, 1,
5405 N3RegFrm, NoItinerary, "vmaxnm", "f32",
5406 v2f32, v2f32, fmaxnum, 1>,
5407 Requires<[HasV8, HasNEON]>;
5408 def VMAXNMNQf : N3VQIntnp<0b00110, 0b00, 0b1111, 1, 1,
5409 N3RegFrm, NoItinerary, "vmaxnm", "f32",
5410 v4f32, v4f32, fmaxnum, 1>,
5411 Requires<[HasV8, HasNEON]>;
5412 def VMAXNMNDh : N3VDIntnp<0b00110, 0b01, 0b1111, 0, 1,
5413 N3RegFrm, NoItinerary, "vmaxnm", "f16",
5414 v4f16, v4f16, fmaxnum, 1>,
5415 Requires<[HasV8, HasNEON, HasFullFP16]>;
5416 def VMAXNMNQh : N3VQIntnp<0b00110, 0b01, 0b1111, 1, 1,
5417 N3RegFrm, NoItinerary, "vmaxnm", "f16",
5418 v8f16, v8f16, fmaxnum, 1>,
5419 Requires<[HasV8, HasNEON, HasFullFP16]>;
5422 // VMIN : Vector Minimum
5423 defm VMINs : N3VInt_QHS<0, 0, 0b0110, 1, N3RegFrm,
5424 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
5425 "vmin", "s", smin, 1>;
5426 defm VMINu : N3VInt_QHS<1, 0, 0b0110, 1, N3RegFrm,
5427 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
5428 "vmin", "u", umin, 1>;
5429 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBIND,
5431 v2f32, v2f32, fminnan, 1>;
5432 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBINQ,
5434 v4f32, v4f32, fminnan, 1>;
5435 def VMINhd : N3VDInt<0, 0, 0b11, 0b1111, 0, N3RegFrm, IIC_VBIND,
5437 v4f16, v4f16, fminnan, 1>,
5438 Requires<[HasNEON, HasFullFP16]>;
5439 def VMINhq : N3VQInt<0, 0, 0b11, 0b1111, 0, N3RegFrm, IIC_VBINQ,
5441 v8f16, v8f16, fminnan, 1>,
5442 Requires<[HasNEON, HasFullFP16]>;
5445 let PostEncoderMethod = "NEONThumb2V8PostEncoder", DecoderNamespace = "v8NEON" in {
5446 def VMINNMNDf : N3VDIntnp<0b00110, 0b10, 0b1111, 0, 1,
5447 N3RegFrm, NoItinerary, "vminnm", "f32",
5448 v2f32, v2f32, fminnum, 1>,
5449 Requires<[HasV8, HasNEON]>;
5450 def VMINNMNQf : N3VQIntnp<0b00110, 0b10, 0b1111, 1, 1,
5451 N3RegFrm, NoItinerary, "vminnm", "f32",
5452 v4f32, v4f32, fminnum, 1>,
5453 Requires<[HasV8, HasNEON]>;
5454 def VMINNMNDh : N3VDIntnp<0b00110, 0b11, 0b1111, 0, 1,
5455 N3RegFrm, NoItinerary, "vminnm", "f16",
5456 v4f16, v4f16, fminnum, 1>,
5457 Requires<[HasV8, HasNEON, HasFullFP16]>;
5458 def VMINNMNQh : N3VQIntnp<0b00110, 0b11, 0b1111, 1, 1,
5459 N3RegFrm, NoItinerary, "vminnm", "f16",
5460 v8f16, v8f16, fminnum, 1>,
5461 Requires<[HasV8, HasNEON, HasFullFP16]>;
5464 // Vector Pairwise Operations.
5466 // VPADD : Vector Pairwise Add
5467 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
5469 v8i8, v8i8, int_arm_neon_vpadd, 0>;
5470 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
5472 v4i16, v4i16, int_arm_neon_vpadd, 0>;
5473 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
5475 v2i32, v2i32, int_arm_neon_vpadd, 0>;
5476 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, N3RegFrm,
5477 IIC_VPBIND, "vpadd", "f32",
5478 v2f32, v2f32, int_arm_neon_vpadd, 0>;
5479 def VPADDh : N3VDInt<1, 0, 0b01, 0b1101, 0, N3RegFrm,
5480 IIC_VPBIND, "vpadd", "f16",
5481 v4f16, v4f16, int_arm_neon_vpadd, 0>,
5482 Requires<[HasNEON, HasFullFP16]>;
5484 // VPADDL : Vector Pairwise Add Long
5485 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
5486 int_arm_neon_vpaddls>;
5487 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
5488 int_arm_neon_vpaddlu>;
5490 // VPADAL : Vector Pairwise Add and Accumulate Long
5491 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
5492 int_arm_neon_vpadals>;
5493 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
5494 int_arm_neon_vpadalu>;
5496 // VPMAX : Vector Pairwise Maximum
5497 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
5498 "s8", v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
5499 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
5500 "s16", v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
5501 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
5502 "s32", v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
5503 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
5504 "u8", v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
5505 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
5506 "u16", v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
5507 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
5508 "u32", v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
5509 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmax",
5510 "f32", v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
5511 def VPMAXh : N3VDInt<1, 0, 0b01, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmax",
5512 "f16", v4f16, v4f16, int_arm_neon_vpmaxs, 0>,
5513 Requires<[HasNEON, HasFullFP16]>;
5515 // VPMIN : Vector Pairwise Minimum
5516 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
5517 "s8", v8i8, v8i8, int_arm_neon_vpmins, 0>;
5518 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
5519 "s16", v4i16, v4i16, int_arm_neon_vpmins, 0>;
5520 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
5521 "s32", v2i32, v2i32, int_arm_neon_vpmins, 0>;
5522 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
5523 "u8", v8i8, v8i8, int_arm_neon_vpminu, 0>;
5524 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
5525 "u16", v4i16, v4i16, int_arm_neon_vpminu, 0>;
5526 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
5527 "u32", v2i32, v2i32, int_arm_neon_vpminu, 0>;
5528 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmin",
5529 "f32", v2f32, v2f32, int_arm_neon_vpmins, 0>;
5530 def VPMINh : N3VDInt<1, 0, 0b11, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmin",
5531 "f16", v4f16, v4f16, int_arm_neon_vpmins, 0>,
5532 Requires<[HasNEON, HasFullFP16]>;
5534 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
5536 // VRECPE : Vector Reciprocal Estimate
5537 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
5538 IIC_VUNAD, "vrecpe", "u32",
5539 v2i32, v2i32, int_arm_neon_vrecpe>;
5540 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
5541 IIC_VUNAQ, "vrecpe", "u32",
5542 v4i32, v4i32, int_arm_neon_vrecpe>;
5543 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
5544 IIC_VUNAD, "vrecpe", "f32",
5545 v2f32, v2f32, int_arm_neon_vrecpe>;
5546 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
5547 IIC_VUNAQ, "vrecpe", "f32",
5548 v4f32, v4f32, int_arm_neon_vrecpe>;
5549 def VRECPEhd : N2VDInt<0b11, 0b11, 0b01, 0b11, 0b01010, 0,
5550 IIC_VUNAD, "vrecpe", "f16",
5551 v4f16, v4f16, int_arm_neon_vrecpe>,
5552 Requires<[HasNEON, HasFullFP16]>;
5553 def VRECPEhq : N2VQInt<0b11, 0b11, 0b01, 0b11, 0b01010, 0,
5554 IIC_VUNAQ, "vrecpe", "f16",
5555 v8f16, v8f16, int_arm_neon_vrecpe>,
5556 Requires<[HasNEON, HasFullFP16]>;
5558 // VRECPS : Vector Reciprocal Step
5559 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
5560 IIC_VRECSD, "vrecps", "f32",
5561 v2f32, v2f32, int_arm_neon_vrecps, 1>;
5562 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
5563 IIC_VRECSQ, "vrecps", "f32",
5564 v4f32, v4f32, int_arm_neon_vrecps, 1>;
5565 def VRECPShd : N3VDInt<0, 0, 0b01, 0b1111, 1, N3RegFrm,
5566 IIC_VRECSD, "vrecps", "f16",
5567 v4f16, v4f16, int_arm_neon_vrecps, 1>,
5568 Requires<[HasNEON, HasFullFP16]>;
5569 def VRECPShq : N3VQInt<0, 0, 0b01, 0b1111, 1, N3RegFrm,
5570 IIC_VRECSQ, "vrecps", "f16",
5571 v8f16, v8f16, int_arm_neon_vrecps, 1>,
5572 Requires<[HasNEON, HasFullFP16]>;
5574 // VRSQRTE : Vector Reciprocal Square Root Estimate
5575 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
5576 IIC_VUNAD, "vrsqrte", "u32",
5577 v2i32, v2i32, int_arm_neon_vrsqrte>;
5578 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
5579 IIC_VUNAQ, "vrsqrte", "u32",
5580 v4i32, v4i32, int_arm_neon_vrsqrte>;
5581 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
5582 IIC_VUNAD, "vrsqrte", "f32",
5583 v2f32, v2f32, int_arm_neon_vrsqrte>;
5584 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
5585 IIC_VUNAQ, "vrsqrte", "f32",
5586 v4f32, v4f32, int_arm_neon_vrsqrte>;
5587 def VRSQRTEhd : N2VDInt<0b11, 0b11, 0b01, 0b11, 0b01011, 0,
5588 IIC_VUNAD, "vrsqrte", "f16",
5589 v4f16, v4f16, int_arm_neon_vrsqrte>,
5590 Requires<[HasNEON, HasFullFP16]>;
5591 def VRSQRTEhq : N2VQInt<0b11, 0b11, 0b01, 0b11, 0b01011, 0,
5592 IIC_VUNAQ, "vrsqrte", "f16",
5593 v8f16, v8f16, int_arm_neon_vrsqrte>,
5594 Requires<[HasNEON, HasFullFP16]>;
5596 // VRSQRTS : Vector Reciprocal Square Root Step
5597 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
5598 IIC_VRECSD, "vrsqrts", "f32",
5599 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
5600 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
5601 IIC_VRECSQ, "vrsqrts", "f32",
5602 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
5603 def VRSQRTShd : N3VDInt<0, 0, 0b11, 0b1111, 1, N3RegFrm,
5604 IIC_VRECSD, "vrsqrts", "f16",
5605 v4f16, v4f16, int_arm_neon_vrsqrts, 1>,
5606 Requires<[HasNEON, HasFullFP16]>;
5607 def VRSQRTShq : N3VQInt<0, 0, 0b11, 0b1111, 1, N3RegFrm,
5608 IIC_VRECSQ, "vrsqrts", "f16",
5609 v8f16, v8f16, int_arm_neon_vrsqrts, 1>,
5610 Requires<[HasNEON, HasFullFP16]>;
5614 // VSHL : Vector Shift
5615 defm VSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 0, N3RegVShFrm,
5616 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
5617 "vshl", "s", int_arm_neon_vshifts>;
5618 defm VSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 0, N3RegVShFrm,
5619 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
5620 "vshl", "u", int_arm_neon_vshiftu>;
5622 // VSHL : Vector Shift Left (Immediate)
5623 defm VSHLi : N2VShL_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl>;
5625 // VSHR : Vector Shift Right (Immediate)
5626 defm VSHRs : N2VShR_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s", "VSHRs",
5628 defm VSHRu : N2VShR_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u", "VSHRu",
5631 // VSHLL : Vector Shift Left Long
5632 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s",
5633 PatFrag<(ops node:$LHS, node:$RHS), (NEONvshl (sext node:$LHS), node:$RHS)>>;
5634 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u",
5635 PatFrag<(ops node:$LHS, node:$RHS), (NEONvshl (zext node:$LHS), node:$RHS)>>;
5637 // VSHLL : Vector Shift Left Long (with maximum shift count)
5638 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
5639 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
5640 ValueType OpTy, Operand ImmTy>
5641 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
5642 ResTy, OpTy, ImmTy, null_frag> {
5643 let Inst{21-16} = op21_16;
5644 let DecoderMethod = "DecodeVSHLMaxInstruction";
5646 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
5648 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
5649 v4i32, v4i16, imm16>;
5650 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
5651 v2i64, v2i32, imm32>;
5653 def : Pat<(v8i16 (NEONvshl (zext (v8i8 DPR:$Rn)), (i32 8))),
5654 (VSHLLi8 DPR:$Rn, 8)>;
5655 def : Pat<(v4i32 (NEONvshl (zext (v4i16 DPR:$Rn)), (i32 16))),
5656 (VSHLLi16 DPR:$Rn, 16)>;
5657 def : Pat<(v2i64 (NEONvshl (zext (v2i32 DPR:$Rn)), (i32 32))),
5658 (VSHLLi32 DPR:$Rn, 32)>;
5659 def : Pat<(v8i16 (NEONvshl (sext (v8i8 DPR:$Rn)), (i32 8))),
5660 (VSHLLi8 DPR:$Rn, 8)>;
5661 def : Pat<(v4i32 (NEONvshl (sext (v4i16 DPR:$Rn)), (i32 16))),
5662 (VSHLLi16 DPR:$Rn, 16)>;
5663 def : Pat<(v2i64 (NEONvshl (sext (v2i32 DPR:$Rn)), (i32 32))),
5664 (VSHLLi32 DPR:$Rn, 32)>;
5665 def : Pat<(v8i16 (NEONvshl (anyext (v8i8 DPR:$Rn)), (i32 8))),
5666 (VSHLLi8 DPR:$Rn, 8)>;
5667 def : Pat<(v4i32 (NEONvshl (anyext (v4i16 DPR:$Rn)), (i32 16))),
5668 (VSHLLi16 DPR:$Rn, 16)>;
5669 def : Pat<(v2i64 (NEONvshl (anyext (v2i32 DPR:$Rn)), (i32 32))),
5670 (VSHLLi32 DPR:$Rn, 32)>;
5672 // VSHRN : Vector Shift Right and Narrow
5673 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
5674 PatFrag<(ops node:$Rn, node:$amt),
5675 (trunc (NEONvshrs node:$Rn, node:$amt))>>;
5677 def : Pat<(v8i8 (trunc (NEONvshru (v8i16 QPR:$Vn), shr_imm8:$amt))),
5678 (VSHRNv8i8 QPR:$Vn, shr_imm8:$amt)>;
5679 def : Pat<(v4i16 (trunc (NEONvshru (v4i32 QPR:$Vn), shr_imm16:$amt))),
5680 (VSHRNv4i16 QPR:$Vn, shr_imm16:$amt)>;
5681 def : Pat<(v2i32 (trunc (NEONvshru (v2i64 QPR:$Vn), shr_imm32:$amt))),
5682 (VSHRNv2i32 QPR:$Vn, shr_imm32:$amt)>;
5684 // VRSHL : Vector Rounding Shift
5685 defm VRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 0, N3RegVShFrm,
5686 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
5687 "vrshl", "s", int_arm_neon_vrshifts>;
5688 defm VRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 0, N3RegVShFrm,
5689 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
5690 "vrshl", "u", int_arm_neon_vrshiftu>;
5691 // VRSHR : Vector Rounding Shift Right
5692 defm VRSHRs : N2VShR_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s", "VRSHRs",
5694 defm VRSHRu : N2VShR_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u", "VRSHRu",
5697 // VRSHRN : Vector Rounding Shift Right and Narrow
5698 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
5701 // VQSHL : Vector Saturating Shift
5702 defm VQSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 1, N3RegVShFrm,
5703 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
5704 "vqshl", "s", int_arm_neon_vqshifts>;
5705 defm VQSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 1, N3RegVShFrm,
5706 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
5707 "vqshl", "u", int_arm_neon_vqshiftu>;
5708 // VQSHL : Vector Saturating Shift Left (Immediate)
5709 defm VQSHLsi : N2VShL_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s",NEONvqshls>;
5710 defm VQSHLui : N2VShL_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u",NEONvqshlu>;
5712 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
5713 defm VQSHLsu : N2VShL_QHSD<1,1,0b0110,1, IIC_VSHLi4D,"vqshlu","s",NEONvqshlsu>;
5715 // VQSHRN : Vector Saturating Shift Right and Narrow
5716 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
5718 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
5721 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
5722 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
5725 // VQRSHL : Vector Saturating Rounding Shift
5726 defm VQRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 1, N3RegVShFrm,
5727 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
5728 "vqrshl", "s", int_arm_neon_vqrshifts>;
5729 defm VQRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 1, N3RegVShFrm,
5730 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
5731 "vqrshl", "u", int_arm_neon_vqrshiftu>;
5733 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
5734 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
5736 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
5739 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
5740 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
5743 // VSRA : Vector Shift Right and Accumulate
5744 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
5745 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
5746 // VRSRA : Vector Rounding Shift Right and Accumulate
5747 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
5748 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
5750 // VSLI : Vector Shift Left and Insert
5751 defm VSLI : N2VShInsL_QHSD<1, 1, 0b0101, 1, "vsli">;
5753 // VSRI : Vector Shift Right and Insert
5754 defm VSRI : N2VShInsR_QHSD<1, 1, 0b0100, 1, "vsri">;
5756 // Vector Absolute and Saturating Absolute.
5758 // VABS : Vector Absolute Value
5759 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
5760 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s", abs>;
5761 def VABSfd : N2VD<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
5763 v2f32, v2f32, fabs>;
5764 def VABSfq : N2VQ<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
5766 v4f32, v4f32, fabs>;
5767 def VABShd : N2VD<0b11, 0b11, 0b01, 0b01, 0b01110, 0,
5769 v4f16, v4f16, fabs>,
5770 Requires<[HasNEON, HasFullFP16]>;
5771 def VABShq : N2VQ<0b11, 0b11, 0b01, 0b01, 0b01110, 0,
5773 v8f16, v8f16, fabs>,
5774 Requires<[HasNEON, HasFullFP16]>;
5776 // VQABS : Vector Saturating Absolute Value
5777 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
5778 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
5779 int_arm_neon_vqabs>;
5783 def vnegd : PatFrag<(ops node:$in),
5784 (sub (bitconvert (v2i32 NEONimmAllZerosV)), node:$in)>;
5785 def vnegq : PatFrag<(ops node:$in),
5786 (sub (bitconvert (v4i32 NEONimmAllZerosV)), node:$in)>;
5788 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
5789 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$Vd), (ins DPR:$Vm),
5790 IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
5791 [(set DPR:$Vd, (Ty (vnegd DPR:$Vm)))]>;
5792 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
5793 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$Vd), (ins QPR:$Vm),
5794 IIC_VSHLiQ, OpcodeStr, Dt, "$Vd, $Vm", "",
5795 [(set QPR:$Vd, (Ty (vnegq QPR:$Vm)))]>;
5797 // VNEG : Vector Negate (integer)
5798 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
5799 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
5800 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
5801 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
5802 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
5803 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
5805 // VNEG : Vector Negate (floating-point)
5806 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
5807 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VUNAD,
5808 "vneg", "f32", "$Vd, $Vm", "",
5809 [(set DPR:$Vd, (v2f32 (fneg DPR:$Vm)))]>;
5810 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
5811 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VUNAQ,
5812 "vneg", "f32", "$Vd, $Vm", "",
5813 [(set QPR:$Vd, (v4f32 (fneg QPR:$Vm)))]>;
5814 def VNEGhd : N2V<0b11, 0b11, 0b01, 0b01, 0b01111, 0, 0,
5815 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VUNAD,
5816 "vneg", "f16", "$Vd, $Vm", "",
5817 [(set DPR:$Vd, (v4f16 (fneg DPR:$Vm)))]>,
5818 Requires<[HasNEON, HasFullFP16]>;
5819 def VNEGhq : N2V<0b11, 0b11, 0b01, 0b01, 0b01111, 1, 0,
5820 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VUNAQ,
5821 "vneg", "f16", "$Vd, $Vm", "",
5822 [(set QPR:$Vd, (v8f16 (fneg QPR:$Vm)))]>,
5823 Requires<[HasNEON, HasFullFP16]>;
5825 def : Pat<(v8i8 (vnegd DPR:$src)), (VNEGs8d DPR:$src)>;
5826 def : Pat<(v4i16 (vnegd DPR:$src)), (VNEGs16d DPR:$src)>;
5827 def : Pat<(v2i32 (vnegd DPR:$src)), (VNEGs32d DPR:$src)>;
5828 def : Pat<(v16i8 (vnegq QPR:$src)), (VNEGs8q QPR:$src)>;
5829 def : Pat<(v8i16 (vnegq QPR:$src)), (VNEGs16q QPR:$src)>;
5830 def : Pat<(v4i32 (vnegq QPR:$src)), (VNEGs32q QPR:$src)>;
5832 // VQNEG : Vector Saturating Negate
5833 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
5834 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
5835 int_arm_neon_vqneg>;
5837 // Vector Bit Counting Operations.
5839 // VCLS : Vector Count Leading Sign Bits
5840 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
5841 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
5843 // VCLZ : Vector Count Leading Zeros
5844 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
5845 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
5847 // VCNT : Vector Count One Bits
5848 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
5849 IIC_VCNTiD, "vcnt", "8",
5851 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
5852 IIC_VCNTiQ, "vcnt", "8",
5853 v16i8, v16i8, ctpop>;
5856 def VSWPd : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 0, 0,
5857 (outs DPR:$Vd, DPR:$Vm), (ins DPR:$in1, DPR:$in2),
5858 NoItinerary, "vswp", "$Vd, $Vm", "$in1 = $Vd, $in2 = $Vm",
5860 def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0,
5861 (outs QPR:$Vd, QPR:$Vm), (ins QPR:$in1, QPR:$in2),
5862 NoItinerary, "vswp", "$Vd, $Vm", "$in1 = $Vd, $in2 = $Vm",
5865 // Vector Move Operations.
5867 // VMOV : Vector Move (Register)
5868 def : NEONInstAlias<"vmov${p} $Vd, $Vm",
5869 (VORRd DPR:$Vd, DPR:$Vm, DPR:$Vm, pred:$p)>;
5870 def : NEONInstAlias<"vmov${p} $Vd, $Vm",
5871 (VORRq QPR:$Vd, QPR:$Vm, QPR:$Vm, pred:$p)>;
5873 // VMOV : Vector Move (Immediate)
5875 // Although VMOVs are not strictly speaking cheap, they are as expensive
5876 // as their copies counterpart (VORR), so we should prefer rematerialization
5877 // over splitting when it applies.
5878 let isReMaterializable = 1, isAsCheapAsAMove=1 in {
5879 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$Vd),
5880 (ins nImmSplatI8:$SIMM), IIC_VMOVImm,
5881 "vmov", "i8", "$Vd, $SIMM", "",
5882 [(set DPR:$Vd, (v8i8 (NEONvmovImm timm:$SIMM)))]>;
5883 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$Vd),
5884 (ins nImmSplatI8:$SIMM), IIC_VMOVImm,
5885 "vmov", "i8", "$Vd, $SIMM", "",
5886 [(set QPR:$Vd, (v16i8 (NEONvmovImm timm:$SIMM)))]>;
5888 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 0, 1, (outs DPR:$Vd),
5889 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
5890 "vmov", "i16", "$Vd, $SIMM", "",
5891 [(set DPR:$Vd, (v4i16 (NEONvmovImm timm:$SIMM)))]> {
5892 let Inst{9} = SIMM{9};
5895 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 0, 1, (outs QPR:$Vd),
5896 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
5897 "vmov", "i16", "$Vd, $SIMM", "",
5898 [(set QPR:$Vd, (v8i16 (NEONvmovImm timm:$SIMM)))]> {
5899 let Inst{9} = SIMM{9};
5902 def VMOVv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 0, 1, (outs DPR:$Vd),
5903 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
5904 "vmov", "i32", "$Vd, $SIMM", "",
5905 [(set DPR:$Vd, (v2i32 (NEONvmovImm timm:$SIMM)))]> {
5906 let Inst{11-8} = SIMM{11-8};
5909 def VMOVv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 0, 1, (outs QPR:$Vd),
5910 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
5911 "vmov", "i32", "$Vd, $SIMM", "",
5912 [(set QPR:$Vd, (v4i32 (NEONvmovImm timm:$SIMM)))]> {
5913 let Inst{11-8} = SIMM{11-8};
5916 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$Vd),
5917 (ins nImmSplatI64:$SIMM), IIC_VMOVImm,
5918 "vmov", "i64", "$Vd, $SIMM", "",
5919 [(set DPR:$Vd, (v1i64 (NEONvmovImm timm:$SIMM)))]>;
5920 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$Vd),
5921 (ins nImmSplatI64:$SIMM), IIC_VMOVImm,
5922 "vmov", "i64", "$Vd, $SIMM", "",
5923 [(set QPR:$Vd, (v2i64 (NEONvmovImm timm:$SIMM)))]>;
5925 def VMOVv2f32 : N1ModImm<1, 0b000, 0b1111, 0, 0, 0, 1, (outs DPR:$Vd),
5926 (ins nImmVMOVF32:$SIMM), IIC_VMOVImm,
5927 "vmov", "f32", "$Vd, $SIMM", "",
5928 [(set DPR:$Vd, (v2f32 (NEONvmovFPImm timm:$SIMM)))]>;
5929 def VMOVv4f32 : N1ModImm<1, 0b000, 0b1111, 0, 1, 0, 1, (outs QPR:$Vd),
5930 (ins nImmVMOVF32:$SIMM), IIC_VMOVImm,
5931 "vmov", "f32", "$Vd, $SIMM", "",
5932 [(set QPR:$Vd, (v4f32 (NEONvmovFPImm timm:$SIMM)))]>;
5933 } // isReMaterializable, isAsCheapAsAMove
5935 // Add support for bytes replication feature, so it could be GAS compatible.
5936 // E.g. instructions below:
5937 // "vmov.i32 d0, 0xffffffff"
5938 // "vmov.i32 d0, 0xabababab"
5939 // "vmov.i16 d0, 0xabab"
5940 // are incorrect, but we could deal with such cases.
5941 // For last two instructions, for example, it should emit:
5942 // "vmov.i8 d0, 0xab"
5943 def : NEONInstAlias<"vmov${p}.i16 $Vd, $Vm",
5944 (VMOVv8i8 DPR:$Vd, nImmVMOVI16ByteReplicate:$Vm, pred:$p)>;
5945 def : NEONInstAlias<"vmov${p}.i32 $Vd, $Vm",
5946 (VMOVv8i8 DPR:$Vd, nImmVMOVI32ByteReplicate:$Vm, pred:$p)>;
5947 def : NEONInstAlias<"vmov${p}.i16 $Vd, $Vm",
5948 (VMOVv16i8 QPR:$Vd, nImmVMOVI16ByteReplicate:$Vm, pred:$p)>;
5949 def : NEONInstAlias<"vmov${p}.i32 $Vd, $Vm",
5950 (VMOVv16i8 QPR:$Vd, nImmVMOVI32ByteReplicate:$Vm, pred:$p)>;
5952 // Also add same support for VMVN instructions. So instruction:
5953 // "vmvn.i32 d0, 0xabababab"
5955 // "vmov.i8 d0, 0x54"
5956 def : NEONInstAlias<"vmvn${p}.i16 $Vd, $Vm",
5957 (VMOVv8i8 DPR:$Vd, nImmVMVNI16ByteReplicate:$Vm, pred:$p)>;
5958 def : NEONInstAlias<"vmvn${p}.i32 $Vd, $Vm",
5959 (VMOVv8i8 DPR:$Vd, nImmVMVNI32ByteReplicate:$Vm, pred:$p)>;
5960 def : NEONInstAlias<"vmvn${p}.i16 $Vd, $Vm",
5961 (VMOVv16i8 QPR:$Vd, nImmVMVNI16ByteReplicate:$Vm, pred:$p)>;
5962 def : NEONInstAlias<"vmvn${p}.i32 $Vd, $Vm",
5963 (VMOVv16i8 QPR:$Vd, nImmVMVNI32ByteReplicate:$Vm, pred:$p)>;
5965 // On some CPUs the two instructions "vmov.i32 dD, #0" and "vmov.i32 qD, #0"
5966 // require zero cycles to execute so they should be used wherever possible for
5967 // setting a register to zero.
5969 // Even without these pseudo-insts we would probably end up with the correct
5970 // instruction, but we could not mark the general ones with "isAsCheapAsAMove"
5971 // since they are sometimes rather expensive (in general).
5973 let AddedComplexity = 50, isAsCheapAsAMove = 1, isReMaterializable = 1 in {
5974 def VMOVD0 : ARMPseudoExpand<(outs DPR:$Vd), (ins), 4, IIC_VMOVImm,
5975 [(set DPR:$Vd, (v2i32 NEONimmAllZerosV))],
5976 (VMOVv2i32 DPR:$Vd, 0, (ops 14, zero_reg))>,
5978 def VMOVQ0 : ARMPseudoExpand<(outs QPR:$Vd), (ins), 4, IIC_VMOVImm,
5979 [(set QPR:$Vd, (v4i32 NEONimmAllZerosV))],
5980 (VMOVv4i32 QPR:$Vd, 0, (ops 14, zero_reg))>,
5984 // VMOV : Vector Get Lane (move scalar to ARM core register)
5986 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
5987 (outs GPR:$R), (ins DPR:$V, VectorIndex8:$lane),
5988 IIC_VMOVSI, "vmov", "s8", "$R, $V$lane",
5989 [(set GPR:$R, (NEONvgetlanes (v8i8 DPR:$V),
5991 let Inst{21} = lane{2};
5992 let Inst{6-5} = lane{1-0};
5994 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
5995 (outs GPR:$R), (ins DPR:$V, VectorIndex16:$lane),
5996 IIC_VMOVSI, "vmov", "s16", "$R, $V$lane",
5997 [(set GPR:$R, (NEONvgetlanes (v4i16 DPR:$V),
5999 let Inst{21} = lane{1};
6000 let Inst{6} = lane{0};
6002 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
6003 (outs GPR:$R), (ins DPR:$V, VectorIndex8:$lane),
6004 IIC_VMOVSI, "vmov", "u8", "$R, $V$lane",
6005 [(set GPR:$R, (NEONvgetlaneu (v8i8 DPR:$V),
6007 let Inst{21} = lane{2};
6008 let Inst{6-5} = lane{1-0};
6010 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
6011 (outs GPR:$R), (ins DPR:$V, VectorIndex16:$lane),
6012 IIC_VMOVSI, "vmov", "u16", "$R, $V$lane",
6013 [(set GPR:$R, (NEONvgetlaneu (v4i16 DPR:$V),
6015 let Inst{21} = lane{1};
6016 let Inst{6} = lane{0};
6018 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
6019 (outs GPR:$R), (ins DPR:$V, VectorIndex32:$lane),
6020 IIC_VMOVSI, "vmov", "32", "$R, $V$lane",
6021 [(set GPR:$R, (extractelt (v2i32 DPR:$V),
6023 Requires<[HasVFP2, HasFastVGETLNi32]> {
6024 let Inst{21} = lane{0};
6026 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
6027 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
6028 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
6029 (DSubReg_i8_reg imm:$lane))),
6030 (SubReg_i8_lane imm:$lane))>;
6031 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
6032 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
6033 (DSubReg_i16_reg imm:$lane))),
6034 (SubReg_i16_lane imm:$lane))>;
6035 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
6036 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
6037 (DSubReg_i8_reg imm:$lane))),
6038 (SubReg_i8_lane imm:$lane))>;
6039 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
6040 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
6041 (DSubReg_i16_reg imm:$lane))),
6042 (SubReg_i16_lane imm:$lane))>;
6043 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
6044 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
6045 (DSubReg_i32_reg imm:$lane))),
6046 (SubReg_i32_lane imm:$lane))>,
6047 Requires<[HasNEON, HasFastVGETLNi32]>;
6048 def : Pat<(extractelt (v2i32 DPR:$src), imm:$lane),
6050 (i32 (EXTRACT_SUBREG DPR:$src, (SSubReg_f32_reg imm:$lane))), GPR)>,
6051 Requires<[HasNEON, HasSlowVGETLNi32]>;
6052 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
6054 (i32 (EXTRACT_SUBREG QPR:$src, (SSubReg_f32_reg imm:$lane))), GPR)>,
6055 Requires<[HasNEON, HasSlowVGETLNi32]>;
6056 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
6057 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
6058 (SSubReg_f32_reg imm:$src2))>;
6059 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
6060 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
6061 (SSubReg_f32_reg imm:$src2))>;
6062 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
6063 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
6064 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
6065 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
6068 // VMOV : Vector Set Lane (move ARM core register to scalar)
6070 let Constraints = "$src1 = $V" in {
6071 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$V),
6072 (ins DPR:$src1, GPR:$R, VectorIndex8:$lane),
6073 IIC_VMOVISL, "vmov", "8", "$V$lane, $R",
6074 [(set DPR:$V, (vector_insert (v8i8 DPR:$src1),
6075 GPR:$R, imm:$lane))]> {
6076 let Inst{21} = lane{2};
6077 let Inst{6-5} = lane{1-0};
6079 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$V),
6080 (ins DPR:$src1, GPR:$R, VectorIndex16:$lane),
6081 IIC_VMOVISL, "vmov", "16", "$V$lane, $R",
6082 [(set DPR:$V, (vector_insert (v4i16 DPR:$src1),
6083 GPR:$R, imm:$lane))]> {
6084 let Inst{21} = lane{1};
6085 let Inst{6} = lane{0};
6087 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$V),
6088 (ins DPR:$src1, GPR:$R, VectorIndex32:$lane),
6089 IIC_VMOVISL, "vmov", "32", "$V$lane, $R",
6090 [(set DPR:$V, (insertelt (v2i32 DPR:$src1),
6091 GPR:$R, imm:$lane))]>,
6092 Requires<[HasVFP2]> {
6093 let Inst{21} = lane{0};
6094 // This instruction is equivalent as
6095 // $V = INSERT_SUBREG $src1, $R, translateImmToSubIdx($imm)
6096 let isInsertSubreg = 1;
6099 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
6100 (v16i8 (INSERT_SUBREG QPR:$src1,
6101 (v8i8 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
6102 (DSubReg_i8_reg imm:$lane))),
6103 GPR:$src2, (SubReg_i8_lane imm:$lane))),
6104 (DSubReg_i8_reg imm:$lane)))>;
6105 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
6106 (v8i16 (INSERT_SUBREG QPR:$src1,
6107 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
6108 (DSubReg_i16_reg imm:$lane))),
6109 GPR:$src2, (SubReg_i16_lane imm:$lane))),
6110 (DSubReg_i16_reg imm:$lane)))>;
6111 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
6112 (v4i32 (INSERT_SUBREG QPR:$src1,
6113 (v2i32 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
6114 (DSubReg_i32_reg imm:$lane))),
6115 GPR:$src2, (SubReg_i32_lane imm:$lane))),
6116 (DSubReg_i32_reg imm:$lane)))>;
6118 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
6119 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
6120 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
6121 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
6122 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
6123 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
6125 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
6126 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
6127 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
6128 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
6130 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
6131 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
6132 def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
6133 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
6134 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
6135 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
6137 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
6138 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
6139 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
6140 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
6141 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
6142 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
6144 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
6145 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
6146 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
6148 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
6149 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
6150 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
6152 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
6153 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
6154 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
6157 // VDUP : Vector Duplicate (from ARM core register to all elements)
6159 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
6160 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$V), (ins GPR:$R),
6161 IIC_VMOVIS, "vdup", Dt, "$V, $R",
6162 [(set DPR:$V, (Ty (NEONvdup (i32 GPR:$R))))]>;
6163 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
6164 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$V), (ins GPR:$R),
6165 IIC_VMOVIS, "vdup", Dt, "$V, $R",
6166 [(set QPR:$V, (Ty (NEONvdup (i32 GPR:$R))))]>;
6168 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
6169 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
6170 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>,
6171 Requires<[HasNEON, HasFastVDUP32]>;
6172 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
6173 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
6174 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
6176 // NEONvdup patterns for uarchs with fast VDUP.32.
6177 def : Pat<(v2f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VDUP32d GPR:$R)>,
6178 Requires<[HasNEON,HasFastVDUP32]>;
6179 def : Pat<(v4f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VDUP32q GPR:$R)>;
6181 // NEONvdup patterns for uarchs with slow VDUP.32 - use VMOVDRR instead.
6182 def : Pat<(v2i32 (NEONvdup (i32 GPR:$R))), (VMOVDRR GPR:$R, GPR:$R)>,
6183 Requires<[HasNEON,HasSlowVDUP32]>;
6184 def : Pat<(v2f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VMOVDRR GPR:$R, GPR:$R)>,
6185 Requires<[HasNEON,HasSlowVDUP32]>;
6187 // VDUP : Vector Duplicate Lane (from scalar to all elements)
6189 class VDUPLND<bits<4> op19_16, string OpcodeStr, string Dt,
6190 ValueType Ty, Operand IdxTy>
6191 : NVDupLane<op19_16, 0, (outs DPR:$Vd), (ins DPR:$Vm, IdxTy:$lane),
6192 IIC_VMOVD, OpcodeStr, Dt, "$Vd, $Vm$lane",
6193 [(set DPR:$Vd, (Ty (NEONvduplane (Ty DPR:$Vm), imm:$lane)))]>;
6195 class VDUPLNQ<bits<4> op19_16, string OpcodeStr, string Dt,
6196 ValueType ResTy, ValueType OpTy, Operand IdxTy>
6197 : NVDupLane<op19_16, 1, (outs QPR:$Vd), (ins DPR:$Vm, IdxTy:$lane),
6198 IIC_VMOVQ, OpcodeStr, Dt, "$Vd, $Vm$lane",
6199 [(set QPR:$Vd, (ResTy (NEONvduplane (OpTy DPR:$Vm),
6200 VectorIndex32:$lane)))]>;
6202 // Inst{19-16} is partially specified depending on the element size.
6204 def VDUPLN8d : VDUPLND<{?,?,?,1}, "vdup", "8", v8i8, VectorIndex8> {
6206 let Inst{19-17} = lane{2-0};
6208 def VDUPLN16d : VDUPLND<{?,?,1,0}, "vdup", "16", v4i16, VectorIndex16> {
6210 let Inst{19-18} = lane{1-0};
6212 def VDUPLN32d : VDUPLND<{?,1,0,0}, "vdup", "32", v2i32, VectorIndex32> {
6214 let Inst{19} = lane{0};
6216 def VDUPLN8q : VDUPLNQ<{?,?,?,1}, "vdup", "8", v16i8, v8i8, VectorIndex8> {
6218 let Inst{19-17} = lane{2-0};
6220 def VDUPLN16q : VDUPLNQ<{?,?,1,0}, "vdup", "16", v8i16, v4i16, VectorIndex16> {
6222 let Inst{19-18} = lane{1-0};
6224 def VDUPLN32q : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4i32, v2i32, VectorIndex32> {
6226 let Inst{19} = lane{0};
6229 def : Pat<(v2f32 (NEONvduplane (v2f32 DPR:$Vm), imm:$lane)),
6230 (VDUPLN32d DPR:$Vm, imm:$lane)>;
6232 def : Pat<(v4f32 (NEONvduplane (v2f32 DPR:$Vm), imm:$lane)),
6233 (VDUPLN32q DPR:$Vm, imm:$lane)>;
6235 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
6236 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
6237 (DSubReg_i8_reg imm:$lane))),
6238 (SubReg_i8_lane imm:$lane)))>;
6239 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
6240 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
6241 (DSubReg_i16_reg imm:$lane))),
6242 (SubReg_i16_lane imm:$lane)))>;
6243 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
6244 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
6245 (DSubReg_i32_reg imm:$lane))),
6246 (SubReg_i32_lane imm:$lane)))>;
6247 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
6248 (v4f32 (VDUPLN32q (v2f32 (EXTRACT_SUBREG QPR:$src,
6249 (DSubReg_i32_reg imm:$lane))),
6250 (SubReg_i32_lane imm:$lane)))>;
6252 def : Pat<(v2f32 (NEONvdup (f32 SPR:$src))),
6253 (v2f32 (VDUPLN32d (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
6254 SPR:$src, ssub_0), (i32 0)))>;
6255 def : Pat<(v4f32 (NEONvdup (f32 SPR:$src))),
6256 (v4f32 (VDUPLN32q (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
6257 SPR:$src, ssub_0), (i32 0)))>;
6259 // VMOVN : Vector Narrowing Move
6260 defm VMOVN : N2VN_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVN,
6261 "vmovn", "i", trunc>;
6262 // VQMOVN : Vector Saturating Narrowing Move
6263 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
6264 "vqmovn", "s", int_arm_neon_vqmovns>;
6265 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
6266 "vqmovn", "u", int_arm_neon_vqmovnu>;
6267 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
6268 "vqmovun", "s", int_arm_neon_vqmovnsu>;
6269 // VMOVL : Vector Lengthening Move
6270 defm VMOVLs : N2VL_QHS<0b01,0b10100,0,1, "vmovl", "s", sext>;
6271 defm VMOVLu : N2VL_QHS<0b11,0b10100,0,1, "vmovl", "u", zext>;
6272 def : Pat<(v8i16 (anyext (v8i8 DPR:$Vm))), (VMOVLuv8i16 DPR:$Vm)>;
6273 def : Pat<(v4i32 (anyext (v4i16 DPR:$Vm))), (VMOVLuv4i32 DPR:$Vm)>;
6274 def : Pat<(v2i64 (anyext (v2i32 DPR:$Vm))), (VMOVLuv2i64 DPR:$Vm)>;
6276 // Vector Conversions.
6278 // VCVT : Vector Convert Between Floating-Point and Integers
6279 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
6280 v2i32, v2f32, fp_to_sint>;
6281 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
6282 v2i32, v2f32, fp_to_uint>;
6283 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
6284 v2f32, v2i32, sint_to_fp>;
6285 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
6286 v2f32, v2i32, uint_to_fp>;
6288 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
6289 v4i32, v4f32, fp_to_sint>;
6290 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
6291 v4i32, v4f32, fp_to_uint>;
6292 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
6293 v4f32, v4i32, sint_to_fp>;
6294 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
6295 v4f32, v4i32, uint_to_fp>;
6297 def VCVTh2sd : N2VD<0b11, 0b11, 0b01, 0b11, 0b01110, 0, "vcvt", "s16.f16",
6298 v4i16, v4f16, fp_to_sint>,
6299 Requires<[HasNEON, HasFullFP16]>;
6300 def VCVTh2ud : N2VD<0b11, 0b11, 0b01, 0b11, 0b01111, 0, "vcvt", "u16.f16",
6301 v4i16, v4f16, fp_to_uint>,
6302 Requires<[HasNEON, HasFullFP16]>;
6303 def VCVTs2hd : N2VD<0b11, 0b11, 0b01, 0b11, 0b01100, 0, "vcvt", "f16.s16",
6304 v4f16, v4i16, sint_to_fp>,
6305 Requires<[HasNEON, HasFullFP16]>;
6306 def VCVTu2hd : N2VD<0b11, 0b11, 0b01, 0b11, 0b01101, 0, "vcvt", "f16.u16",
6307 v4f16, v4i16, uint_to_fp>,
6308 Requires<[HasNEON, HasFullFP16]>;
6310 def VCVTh2sq : N2VQ<0b11, 0b11, 0b01, 0b11, 0b01110, 0, "vcvt", "s16.f16",
6311 v8i16, v8f16, fp_to_sint>,
6312 Requires<[HasNEON, HasFullFP16]>;
6313 def VCVTh2uq : N2VQ<0b11, 0b11, 0b01, 0b11, 0b01111, 0, "vcvt", "u16.f16",
6314 v8i16, v8f16, fp_to_uint>,
6315 Requires<[HasNEON, HasFullFP16]>;
6316 def VCVTs2hq : N2VQ<0b11, 0b11, 0b01, 0b11, 0b01100, 0, "vcvt", "f16.s16",
6317 v8f16, v8i16, sint_to_fp>,
6318 Requires<[HasNEON, HasFullFP16]>;
6319 def VCVTu2hq : N2VQ<0b11, 0b11, 0b01, 0b11, 0b01101, 0, "vcvt", "f16.u16",
6320 v8f16, v8i16, uint_to_fp>,
6321 Requires<[HasNEON, HasFullFP16]>;
6324 multiclass VCVT_FPI<string op, bits<3> op10_8, SDPatternOperator IntS,
6325 SDPatternOperator IntU> {
6326 let PostEncoderMethod = "NEONThumb2V8PostEncoder", DecoderNamespace = "v8NEON" in {
6327 def SDf : N2VDIntnp<0b10, 0b11, op10_8, 0, NoItinerary, !strconcat("vcvt", op),
6328 "s32.f32", v2i32, v2f32, IntS>, Requires<[HasV8, HasNEON]>;
6329 def SQf : N2VQIntnp<0b10, 0b11, op10_8, 0, NoItinerary, !strconcat("vcvt", op),
6330 "s32.f32", v4i32, v4f32, IntS>, Requires<[HasV8, HasNEON]>;
6331 def UDf : N2VDIntnp<0b10, 0b11, op10_8, 1, NoItinerary, !strconcat("vcvt", op),
6332 "u32.f32", v2i32, v2f32, IntU>, Requires<[HasV8, HasNEON]>;
6333 def UQf : N2VQIntnp<0b10, 0b11, op10_8, 1, NoItinerary, !strconcat("vcvt", op),
6334 "u32.f32", v4i32, v4f32, IntU>, Requires<[HasV8, HasNEON]>;
6335 def SDh : N2VDIntnp<0b01, 0b11, op10_8, 0, NoItinerary, !strconcat("vcvt", op),
6336 "s16.f16", v4i16, v4f16, IntS>,
6337 Requires<[HasV8, HasNEON, HasFullFP16]>;
6338 def SQh : N2VQIntnp<0b01, 0b11, op10_8, 0, NoItinerary, !strconcat("vcvt", op),
6339 "s16.f16", v8i16, v8f16, IntS>,
6340 Requires<[HasV8, HasNEON, HasFullFP16]>;
6341 def UDh : N2VDIntnp<0b01, 0b11, op10_8, 1, NoItinerary, !strconcat("vcvt", op),
6342 "u16.f16", v4i16, v4f16, IntU>,
6343 Requires<[HasV8, HasNEON, HasFullFP16]>;
6344 def UQh : N2VQIntnp<0b01, 0b11, op10_8, 1, NoItinerary, !strconcat("vcvt", op),
6345 "u16.f16", v8i16, v8f16, IntU>,
6346 Requires<[HasV8, HasNEON, HasFullFP16]>;
6350 defm VCVTAN : VCVT_FPI<"a", 0b000, int_arm_neon_vcvtas, int_arm_neon_vcvtau>;
6351 defm VCVTNN : VCVT_FPI<"n", 0b001, int_arm_neon_vcvtns, int_arm_neon_vcvtnu>;
6352 defm VCVTPN : VCVT_FPI<"p", 0b010, int_arm_neon_vcvtps, int_arm_neon_vcvtpu>;
6353 defm VCVTMN : VCVT_FPI<"m", 0b011, int_arm_neon_vcvtms, int_arm_neon_vcvtmu>;
6355 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
6356 let DecoderMethod = "DecodeVCVTD" in {
6357 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
6358 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
6359 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
6360 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
6361 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
6362 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
6363 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
6364 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
6365 let Predicates = [HasNEON, HasFullFP16] in {
6366 def VCVTh2xsd : N2VCvtD<0, 1, 0b1101, 0, 1, "vcvt", "s16.f16",
6367 v4i16, v4f16, int_arm_neon_vcvtfp2fxs>;
6368 def VCVTh2xud : N2VCvtD<1, 1, 0b1101, 0, 1, "vcvt", "u16.f16",
6369 v4i16, v4f16, int_arm_neon_vcvtfp2fxu>;
6370 def VCVTxs2hd : N2VCvtD<0, 1, 0b1100, 0, 1, "vcvt", "f16.s16",
6371 v4f16, v4i16, int_arm_neon_vcvtfxs2fp>;
6372 def VCVTxu2hd : N2VCvtD<1, 1, 0b1100, 0, 1, "vcvt", "f16.u16",
6373 v4f16, v4i16, int_arm_neon_vcvtfxu2fp>;
6374 } // Predicates = [HasNEON, HasFullFP16]
6377 let DecoderMethod = "DecodeVCVTQ" in {
6378 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
6379 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
6380 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
6381 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
6382 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
6383 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
6384 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
6385 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
6386 let Predicates = [HasNEON, HasFullFP16] in {
6387 def VCVTh2xsq : N2VCvtQ<0, 1, 0b1101, 0, 1, "vcvt", "s16.f16",
6388 v8i16, v8f16, int_arm_neon_vcvtfp2fxs>;
6389 def VCVTh2xuq : N2VCvtQ<1, 1, 0b1101, 0, 1, "vcvt", "u16.f16",
6390 v8i16, v8f16, int_arm_neon_vcvtfp2fxu>;
6391 def VCVTxs2hq : N2VCvtQ<0, 1, 0b1100, 0, 1, "vcvt", "f16.s16",
6392 v8f16, v8i16, int_arm_neon_vcvtfxs2fp>;
6393 def VCVTxu2hq : N2VCvtQ<1, 1, 0b1100, 0, 1, "vcvt", "f16.u16",
6394 v8f16, v8i16, int_arm_neon_vcvtfxu2fp>;
6395 } // Predicates = [HasNEON, HasFullFP16]
6398 def : NEONInstAlias<"vcvt${p}.s32.f32 $Dd, $Dm, #0",
6399 (VCVTf2sd DPR:$Dd, DPR:$Dm, pred:$p)>;
6400 def : NEONInstAlias<"vcvt${p}.u32.f32 $Dd, $Dm, #0",
6401 (VCVTf2ud DPR:$Dd, DPR:$Dm, pred:$p)>;
6402 def : NEONInstAlias<"vcvt${p}.f32.s32 $Dd, $Dm, #0",
6403 (VCVTs2fd DPR:$Dd, DPR:$Dm, pred:$p)>;
6404 def : NEONInstAlias<"vcvt${p}.f32.u32 $Dd, $Dm, #0",
6405 (VCVTu2fd DPR:$Dd, DPR:$Dm, pred:$p)>;
6407 def : NEONInstAlias<"vcvt${p}.s32.f32 $Qd, $Qm, #0",
6408 (VCVTf2sq QPR:$Qd, QPR:$Qm, pred:$p)>;
6409 def : NEONInstAlias<"vcvt${p}.u32.f32 $Qd, $Qm, #0",
6410 (VCVTf2uq QPR:$Qd, QPR:$Qm, pred:$p)>;
6411 def : NEONInstAlias<"vcvt${p}.f32.s32 $Qd, $Qm, #0",
6412 (VCVTs2fq QPR:$Qd, QPR:$Qm, pred:$p)>;
6413 def : NEONInstAlias<"vcvt${p}.f32.u32 $Qd, $Qm, #0",
6414 (VCVTu2fq QPR:$Qd, QPR:$Qm, pred:$p)>;
6416 def : NEONInstAlias<"vcvt${p}.s16.f16 $Dd, $Dm, #0",
6417 (VCVTh2sd DPR:$Dd, DPR:$Dm, pred:$p)>;
6418 def : NEONInstAlias<"vcvt${p}.u16.f16 $Dd, $Dm, #0",
6419 (VCVTh2ud DPR:$Dd, DPR:$Dm, pred:$p)>;
6420 def : NEONInstAlias<"vcvt${p}.f16.s16 $Dd, $Dm, #0",
6421 (VCVTs2hd DPR:$Dd, DPR:$Dm, pred:$p)>;
6422 def : NEONInstAlias<"vcvt${p}.f16.u16 $Dd, $Dm, #0",
6423 (VCVTu2hd DPR:$Dd, DPR:$Dm, pred:$p)>;
6425 def : NEONInstAlias<"vcvt${p}.s16.f16 $Qd, $Qm, #0",
6426 (VCVTh2sq QPR:$Qd, QPR:$Qm, pred:$p)>;
6427 def : NEONInstAlias<"vcvt${p}.u16.f16 $Qd, $Qm, #0",
6428 (VCVTh2uq QPR:$Qd, QPR:$Qm, pred:$p)>;
6429 def : NEONInstAlias<"vcvt${p}.f16.s16 $Qd, $Qm, #0",
6430 (VCVTs2hq QPR:$Qd, QPR:$Qm, pred:$p)>;
6431 def : NEONInstAlias<"vcvt${p}.f16.u16 $Qd, $Qm, #0",
6432 (VCVTu2hq QPR:$Qd, QPR:$Qm, pred:$p)>;
6435 // VCVT : Vector Convert Between Half-Precision and Single-Precision.
6436 def VCVTf2h : N2VNInt<0b11, 0b11, 0b01, 0b10, 0b01100, 0, 0,
6437 IIC_VUNAQ, "vcvt", "f16.f32",
6438 v4i16, v4f32, int_arm_neon_vcvtfp2hf>,
6439 Requires<[HasNEON, HasFP16]>;
6440 def VCVTh2f : N2VLInt<0b11, 0b11, 0b01, 0b10, 0b01110, 0, 0,
6441 IIC_VUNAQ, "vcvt", "f32.f16",
6442 v4f32, v4i16, int_arm_neon_vcvthf2fp>,
6443 Requires<[HasNEON, HasFP16]>;
6447 // VREV64 : Vector Reverse elements within 64-bit doublewords
6449 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
6450 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$Vd),
6451 (ins DPR:$Vm), IIC_VMOVD,
6452 OpcodeStr, Dt, "$Vd, $Vm", "",
6453 [(set DPR:$Vd, (Ty (NEONvrev64 (Ty DPR:$Vm))))]>;
6454 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
6455 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$Vd),
6456 (ins QPR:$Vm), IIC_VMOVQ,
6457 OpcodeStr, Dt, "$Vd, $Vm", "",
6458 [(set QPR:$Vd, (Ty (NEONvrev64 (Ty QPR:$Vm))))]>;
6460 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
6461 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
6462 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
6463 def : Pat<(v2f32 (NEONvrev64 (v2f32 DPR:$Vm))), (VREV64d32 DPR:$Vm)>;
6465 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
6466 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
6467 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
6468 def : Pat<(v4f32 (NEONvrev64 (v4f32 QPR:$Vm))), (VREV64q32 QPR:$Vm)>;
6470 // VREV32 : Vector Reverse elements within 32-bit words
6472 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
6473 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$Vd),
6474 (ins DPR:$Vm), IIC_VMOVD,
6475 OpcodeStr, Dt, "$Vd, $Vm", "",
6476 [(set DPR:$Vd, (Ty (NEONvrev32 (Ty DPR:$Vm))))]>;
6477 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
6478 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$Vd),
6479 (ins QPR:$Vm), IIC_VMOVQ,
6480 OpcodeStr, Dt, "$Vd, $Vm", "",
6481 [(set QPR:$Vd, (Ty (NEONvrev32 (Ty QPR:$Vm))))]>;
6483 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
6484 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
6486 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
6487 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
6489 // VREV16 : Vector Reverse elements within 16-bit halfwords
6491 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
6492 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$Vd),
6493 (ins DPR:$Vm), IIC_VMOVD,
6494 OpcodeStr, Dt, "$Vd, $Vm", "",
6495 [(set DPR:$Vd, (Ty (NEONvrev16 (Ty DPR:$Vm))))]>;
6496 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
6497 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$Vd),
6498 (ins QPR:$Vm), IIC_VMOVQ,
6499 OpcodeStr, Dt, "$Vd, $Vm", "",
6500 [(set QPR:$Vd, (Ty (NEONvrev16 (Ty QPR:$Vm))))]>;
6502 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
6503 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
6505 // Other Vector Shuffles.
6507 // Aligned extractions: really just dropping registers
6509 class AlignedVEXTq<ValueType DestTy, ValueType SrcTy, SDNodeXForm LaneCVT>
6510 : Pat<(DestTy (vector_extract_subvec (SrcTy QPR:$src), (i32 imm:$start))),
6511 (EXTRACT_SUBREG (SrcTy QPR:$src), (LaneCVT imm:$start))>;
6513 def : AlignedVEXTq<v8i8, v16i8, DSubReg_i8_reg>;
6515 def : AlignedVEXTq<v4i16, v8i16, DSubReg_i16_reg>;
6517 def : AlignedVEXTq<v2i32, v4i32, DSubReg_i32_reg>;
6519 def : AlignedVEXTq<v1i64, v2i64, DSubReg_f64_reg>;
6521 def : AlignedVEXTq<v2f32, v4f32, DSubReg_i32_reg>;
6524 // VEXT : Vector Extract
6527 // All of these have a two-operand InstAlias.
6528 let TwoOperandAliasConstraint = "$Vn = $Vd" in {
6529 class VEXTd<string OpcodeStr, string Dt, ValueType Ty, Operand immTy>
6530 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$Vd),
6531 (ins DPR:$Vn, DPR:$Vm, immTy:$index), NVExtFrm,
6532 IIC_VEXTD, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
6533 [(set DPR:$Vd, (Ty (NEONvext (Ty DPR:$Vn),
6534 (Ty DPR:$Vm), imm:$index)))]> {
6537 let Inst{10-8} = index{2-0};
6540 class VEXTq<string OpcodeStr, string Dt, ValueType Ty, Operand immTy>
6541 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$Vd),
6542 (ins QPR:$Vn, QPR:$Vm, imm0_15:$index), NVExtFrm,
6543 IIC_VEXTQ, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
6544 [(set QPR:$Vd, (Ty (NEONvext (Ty QPR:$Vn),
6545 (Ty QPR:$Vm), imm:$index)))]> {
6547 let Inst{11-8} = index{3-0};
6551 def VEXTd8 : VEXTd<"vext", "8", v8i8, imm0_7> {
6552 let Inst{10-8} = index{2-0};
6554 def VEXTd16 : VEXTd<"vext", "16", v4i16, imm0_3> {
6555 let Inst{10-9} = index{1-0};
6558 def VEXTd32 : VEXTd<"vext", "32", v2i32, imm0_1> {
6559 let Inst{10} = index{0};
6560 let Inst{9-8} = 0b00;
6562 def : Pat<(v2f32 (NEONvext (v2f32 DPR:$Vn),
6565 (VEXTd32 DPR:$Vn, DPR:$Vm, imm:$index)>;
6567 def VEXTq8 : VEXTq<"vext", "8", v16i8, imm0_15> {
6568 let Inst{11-8} = index{3-0};
6570 def VEXTq16 : VEXTq<"vext", "16", v8i16, imm0_7> {
6571 let Inst{11-9} = index{2-0};
6574 def VEXTq32 : VEXTq<"vext", "32", v4i32, imm0_3> {
6575 let Inst{11-10} = index{1-0};
6576 let Inst{9-8} = 0b00;
6578 def VEXTq64 : VEXTq<"vext", "64", v2i64, imm0_1> {
6579 let Inst{11} = index{0};
6580 let Inst{10-8} = 0b000;
6582 def : Pat<(v4f32 (NEONvext (v4f32 QPR:$Vn),
6585 (VEXTq32 QPR:$Vn, QPR:$Vm, imm:$index)>;
6587 // VTRN : Vector Transpose
6589 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
6590 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
6591 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
6593 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
6594 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
6595 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
6597 // VUZP : Vector Unzip (Deinterleave)
6599 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
6600 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
6601 // vuzp.32 Dd, Dm is a pseudo-instruction expanded to vtrn.32 Dd, Dm.
6602 def : NEONInstAlias<"vuzp${p}.32 $Dd, $Dm",
6603 (VTRNd32 DPR:$Dd, DPR:$Dm, pred:$p)>;
6605 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
6606 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
6607 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
6609 // VZIP : Vector Zip (Interleave)
6611 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
6612 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
6613 // vzip.32 Dd, Dm is a pseudo-instruction expanded to vtrn.32 Dd, Dm.
6614 def : NEONInstAlias<"vzip${p}.32 $Dd, $Dm",
6615 (VTRNd32 DPR:$Dd, DPR:$Dm, pred:$p)>;
6617 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
6618 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
6619 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
6621 // Vector Table Lookup and Table Extension.
6623 // VTBL : Vector Table Lookup
6624 let DecoderMethod = "DecodeTBLInstruction" in {
6626 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$Vd),
6627 (ins VecListOneD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB1,
6628 "vtbl", "8", "$Vd, $Vn, $Vm", "",
6629 [(set DPR:$Vd, (v8i8 (NEONvtbl1 VecListOneD:$Vn, DPR:$Vm)))]>;
6631 let hasExtraSrcRegAllocReq = 1 in {
6633 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$Vd),
6634 (ins VecListDPair:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB2,
6635 "vtbl", "8", "$Vd, $Vn, $Vm", "", []>;
6637 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$Vd),
6638 (ins VecListThreeD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB3,
6639 "vtbl", "8", "$Vd, $Vn, $Vm", "", []>;
6641 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$Vd),
6642 (ins VecListFourD:$Vn, DPR:$Vm),
6644 "vtbl", "8", "$Vd, $Vn, $Vm", "", []>;
6645 } // hasExtraSrcRegAllocReq = 1
6648 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB3, "", []>;
6650 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB4, "", []>;
6652 // VTBX : Vector Table Extension
6654 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$Vd),
6655 (ins DPR:$orig, VecListOneD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX1,
6656 "vtbx", "8", "$Vd, $Vn, $Vm", "$orig = $Vd",
6657 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbx1
6658 DPR:$orig, VecListOneD:$Vn, DPR:$Vm)))]>;
6659 let hasExtraSrcRegAllocReq = 1 in {
6661 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$Vd),
6662 (ins DPR:$orig, VecListDPair:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX2,
6663 "vtbx", "8", "$Vd, $Vn, $Vm", "$orig = $Vd", []>;
6665 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$Vd),
6666 (ins DPR:$orig, VecListThreeD:$Vn, DPR:$Vm),
6667 NVTBLFrm, IIC_VTBX3,
6668 "vtbx", "8", "$Vd, $Vn, $Vm",
6671 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$Vd),
6672 (ins DPR:$orig, VecListFourD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX4,
6673 "vtbx", "8", "$Vd, $Vn, $Vm",
6675 } // hasExtraSrcRegAllocReq = 1
6678 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
6679 IIC_VTBX3, "$orig = $dst", []>;
6681 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
6682 IIC_VTBX4, "$orig = $dst", []>;
6683 } // DecoderMethod = "DecodeTBLInstruction"
6685 def : Pat<(v8i8 (NEONvtbl2 v8i8:$Vn0, v8i8:$Vn1, v8i8:$Vm)),
6686 (v8i8 (VTBL2 (REG_SEQUENCE DPair, v8i8:$Vn0, dsub_0,
6689 def : Pat<(v8i8 (int_arm_neon_vtbx2 v8i8:$orig, v8i8:$Vn0, v8i8:$Vn1,
6691 (v8i8 (VTBX2 v8i8:$orig,
6692 (REG_SEQUENCE DPair, v8i8:$Vn0, dsub_0,
6696 def : Pat<(v8i8 (int_arm_neon_vtbl3 v8i8:$Vn0, v8i8:$Vn1,
6697 v8i8:$Vn2, v8i8:$Vm)),
6698 (v8i8 (VTBL3Pseudo (REG_SEQUENCE QQPR, v8i8:$Vn0, dsub_0,
6701 (v8i8 (IMPLICIT_DEF)), dsub_3),
6703 def : Pat<(v8i8 (int_arm_neon_vtbx3 v8i8:$orig, v8i8:$Vn0, v8i8:$Vn1,
6704 v8i8:$Vn2, v8i8:$Vm)),
6705 (v8i8 (VTBX3Pseudo v8i8:$orig,
6706 (REG_SEQUENCE QQPR, v8i8:$Vn0, dsub_0,
6709 (v8i8 (IMPLICIT_DEF)), dsub_3),
6712 def : Pat<(v8i8 (int_arm_neon_vtbl4 v8i8:$Vn0, v8i8:$Vn1,
6713 v8i8:$Vn2, v8i8:$Vn3, v8i8:$Vm)),
6714 (v8i8 (VTBL4Pseudo (REG_SEQUENCE QQPR, v8i8:$Vn0, dsub_0,
6719 def : Pat<(v8i8 (int_arm_neon_vtbx4 v8i8:$orig, v8i8:$Vn0, v8i8:$Vn1,
6720 v8i8:$Vn2, v8i8:$Vn3, v8i8:$Vm)),
6721 (v8i8 (VTBX4Pseudo v8i8:$orig,
6722 (REG_SEQUENCE QQPR, v8i8:$Vn0, dsub_0,
6728 // VRINT : Vector Rounding
6729 multiclass VRINT_FPI<string op, bits<3> op9_7, SDPatternOperator Int> {
6730 let PostEncoderMethod = "NEONThumb2V8PostEncoder", DecoderNamespace = "v8NEON" in {
6731 def Df : N2VDIntnp<0b10, 0b10, 0b100, 0, NoItinerary,
6732 !strconcat("vrint", op), "f32",
6733 v2f32, v2f32, Int>, Requires<[HasV8, HasNEON]> {
6734 let Inst{9-7} = op9_7;
6736 def Qf : N2VQIntnp<0b10, 0b10, 0b100, 0, NoItinerary,
6737 !strconcat("vrint", op), "f32",
6738 v4f32, v4f32, Int>, Requires<[HasV8, HasNEON]> {
6739 let Inst{9-7} = op9_7;
6741 def Dh : N2VDIntnp<0b01, 0b10, 0b100, 0, NoItinerary,
6742 !strconcat("vrint", op), "f16",
6744 Requires<[HasV8, HasNEON, HasFullFP16]> {
6745 let Inst{9-7} = op9_7;
6747 def Qh : N2VQIntnp<0b01, 0b10, 0b100, 0, NoItinerary,
6748 !strconcat("vrint", op), "f16",
6750 Requires<[HasV8, HasNEON, HasFullFP16]> {
6751 let Inst{9-7} = op9_7;
6755 def : NEONInstAlias<!strconcat("vrint", op, ".f32.f32\t$Dd, $Dm"),
6756 (!cast<Instruction>(NAME#"Df") DPR:$Dd, DPR:$Dm)>;
6757 def : NEONInstAlias<!strconcat("vrint", op, ".f32.f32\t$Qd, $Qm"),
6758 (!cast<Instruction>(NAME#"Qf") QPR:$Qd, QPR:$Qm)>;
6759 let Predicates = [HasNEON, HasFullFP16] in {
6760 def : NEONInstAlias<!strconcat("vrint", op, ".f16.f16\t$Dd, $Dm"),
6761 (!cast<Instruction>(NAME#"Dh") DPR:$Dd, DPR:$Dm)>;
6762 def : NEONInstAlias<!strconcat("vrint", op, ".f16.f16\t$Qd, $Qm"),
6763 (!cast<Instruction>(NAME#"Qh") QPR:$Qd, QPR:$Qm)>;
6767 defm VRINTNN : VRINT_FPI<"n", 0b000, int_arm_neon_vrintn>;
6768 defm VRINTXN : VRINT_FPI<"x", 0b001, int_arm_neon_vrintx>;
6769 defm VRINTAN : VRINT_FPI<"a", 0b010, int_arm_neon_vrinta>;
6770 defm VRINTZN : VRINT_FPI<"z", 0b011, int_arm_neon_vrintz>;
6771 defm VRINTMN : VRINT_FPI<"m", 0b101, int_arm_neon_vrintm>;
6772 defm VRINTPN : VRINT_FPI<"p", 0b111, int_arm_neon_vrintp>;
6774 // Cryptography instructions
6775 let PostEncoderMethod = "NEONThumb2DataIPostEncoder",
6776 DecoderNamespace = "v8Crypto", hasSideEffects = 0 in {
6777 class AES<string op, bit op7, bit op6, SDPatternOperator Int>
6778 : N2VQIntXnp<0b00, 0b00, 0b011, op6, op7, NoItinerary,
6779 !strconcat("aes", op), "8", v16i8, v16i8, Int>,
6780 Requires<[HasV8, HasCrypto]>;
6781 class AES2Op<string op, bit op7, bit op6, SDPatternOperator Int>
6782 : N2VQIntX2np<0b00, 0b00, 0b011, op6, op7, NoItinerary,
6783 !strconcat("aes", op), "8", v16i8, v16i8, Int>,
6784 Requires<[HasV8, HasCrypto]>;
6785 class N2SHA<string op, bits<2> op17_16, bits<3> op10_8, bit op7, bit op6,
6786 SDPatternOperator Int>
6787 : N2VQIntXnp<0b10, op17_16, op10_8, op6, op7, NoItinerary,
6788 !strconcat("sha", op), "32", v4i32, v4i32, Int>,
6789 Requires<[HasV8, HasCrypto]>;
6790 class N2SHA2Op<string op, bits<2> op17_16, bits<3> op10_8, bit op7, bit op6,
6791 SDPatternOperator Int>
6792 : N2VQIntX2np<0b10, op17_16, op10_8, op6, op7, NoItinerary,
6793 !strconcat("sha", op), "32", v4i32, v4i32, Int>,
6794 Requires<[HasV8, HasCrypto]>;
6795 class N3SHA3Op<string op, bits<5> op27_23, bits<2> op21_20, SDPatternOperator Int>
6796 : N3VQInt3np<op27_23, op21_20, 0b1100, 1, 0, N3RegFrm, NoItinerary,
6797 !strconcat("sha", op), "32", v4i32, v4i32, Int, 0>,
6798 Requires<[HasV8, HasCrypto]>;
6801 def AESD : AES2Op<"d", 0, 1, int_arm_neon_aesd>;
6802 def AESE : AES2Op<"e", 0, 0, int_arm_neon_aese>;
6803 def AESIMC : AES<"imc", 1, 1, int_arm_neon_aesimc>;
6804 def AESMC : AES<"mc", 1, 0, int_arm_neon_aesmc>;
6806 def SHA1H : N2SHA<"1h", 0b01, 0b010, 1, 1, null_frag>;
6807 def SHA1SU1 : N2SHA2Op<"1su1", 0b10, 0b011, 1, 0, int_arm_neon_sha1su1>;
6808 def SHA256SU0 : N2SHA2Op<"256su0", 0b10, 0b011, 1, 1, int_arm_neon_sha256su0>;
6809 def SHA1C : N3SHA3Op<"1c", 0b00100, 0b00, null_frag>;
6810 def SHA1M : N3SHA3Op<"1m", 0b00100, 0b10, null_frag>;
6811 def SHA1P : N3SHA3Op<"1p", 0b00100, 0b01, null_frag>;
6812 def SHA1SU0 : N3SHA3Op<"1su0", 0b00100, 0b11, int_arm_neon_sha1su0>;
6813 def SHA256H : N3SHA3Op<"256h", 0b00110, 0b00, int_arm_neon_sha256h>;
6814 def SHA256H2 : N3SHA3Op<"256h2", 0b00110, 0b01, int_arm_neon_sha256h2>;
6815 def SHA256SU1 : N3SHA3Op<"256su1", 0b00110, 0b10, int_arm_neon_sha256su1>;
6817 def : Pat<(i32 (int_arm_neon_sha1h i32:$Rn)),
6818 (COPY_TO_REGCLASS (f32 (EXTRACT_SUBREG
6819 (SHA1H (SUBREG_TO_REG (i64 0),
6820 (f32 (COPY_TO_REGCLASS i32:$Rn, SPR)),
6824 def : Pat<(v4i32 (int_arm_neon_sha1c v4i32:$hash_abcd, i32:$hash_e, v4i32:$wk)),
6825 (SHA1C v4i32:$hash_abcd,
6826 (SUBREG_TO_REG (i64 0),
6827 (f32 (COPY_TO_REGCLASS i32:$hash_e, SPR)),
6831 def : Pat<(v4i32 (int_arm_neon_sha1m v4i32:$hash_abcd, i32:$hash_e, v4i32:$wk)),
6832 (SHA1M v4i32:$hash_abcd,
6833 (SUBREG_TO_REG (i64 0),
6834 (f32 (COPY_TO_REGCLASS i32:$hash_e, SPR)),
6838 def : Pat<(v4i32 (int_arm_neon_sha1p v4i32:$hash_abcd, i32:$hash_e, v4i32:$wk)),
6839 (SHA1P v4i32:$hash_abcd,
6840 (SUBREG_TO_REG (i64 0),
6841 (f32 (COPY_TO_REGCLASS i32:$hash_e, SPR)),
6845 //===----------------------------------------------------------------------===//
6846 // NEON instructions for single-precision FP math
6847 //===----------------------------------------------------------------------===//
6849 class N2VSPat<SDNode OpNode, NeonI Inst>
6850 : NEONFPPat<(f32 (OpNode SPR:$a)),
6852 (v2f32 (COPY_TO_REGCLASS (Inst
6854 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
6855 SPR:$a, ssub_0)), DPR_VFP2)), ssub_0)>;
6857 class N3VSPat<SDNode OpNode, NeonI Inst>
6858 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
6860 (v2f32 (COPY_TO_REGCLASS (Inst
6862 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
6865 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
6866 SPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>;
6868 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
6869 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
6871 (v2f32 (COPY_TO_REGCLASS (Inst
6873 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
6876 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
6879 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
6880 SPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>;
6882 class NVCVTIFPat<SDNode OpNode, NeonI Inst>
6883 : NEONFPPat<(f32 (OpNode GPR:$a)),
6884 (f32 (EXTRACT_SUBREG
6887 (v2f32 (IMPLICIT_DEF)),
6888 (i32 (COPY_TO_REGCLASS GPR:$a, SPR)), ssub_0))),
6890 class NVCVTFIPat<SDNode OpNode, NeonI Inst>
6891 : NEONFPPat<(i32 (OpNode SPR:$a)),
6892 (i32 (EXTRACT_SUBREG
6893 (v2f32 (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
6897 def : N3VSPat<fadd, VADDfd>;
6898 def : N3VSPat<fsub, VSUBfd>;
6899 def : N3VSPat<fmul, VMULfd>;
6900 def : N3VSMulOpPat<fmul, fadd, VMLAfd>,
6901 Requires<[HasNEON, UseNEONForFP, UseFPVMLx, DontUseFusedMAC]>;
6902 def : N3VSMulOpPat<fmul, fsub, VMLSfd>,
6903 Requires<[HasNEON, UseNEONForFP, UseFPVMLx, DontUseFusedMAC]>;
6904 def : N3VSMulOpPat<fmul, fadd, VFMAfd>,
6905 Requires<[HasVFP4, UseNEONForFP, UseFusedMAC]>;
6906 def : N3VSMulOpPat<fmul, fsub, VFMSfd>,
6907 Requires<[HasVFP4, UseNEONForFP, UseFusedMAC]>;
6908 def : N2VSPat<fabs, VABSfd>;
6909 def : N2VSPat<fneg, VNEGfd>;
6910 def : N3VSPat<fmaxnan, VMAXfd>, Requires<[HasNEON]>;
6911 def : N3VSPat<fminnan, VMINfd>, Requires<[HasNEON]>;
6912 def : NVCVTFIPat<fp_to_sint, VCVTf2sd>;
6913 def : NVCVTFIPat<fp_to_uint, VCVTf2ud>;
6914 def : NVCVTIFPat<sint_to_fp, VCVTs2fd>;
6915 def : NVCVTIFPat<uint_to_fp, VCVTu2fd>;
6917 // NEON doesn't have any f64 conversions, so provide patterns to make
6918 // sure the VFP conversions match when extracting from a vector.
6919 def : VFPPat<(f64 (sint_to_fp (extractelt (v2i32 DPR:$src), imm:$lane))),
6920 (VSITOD (EXTRACT_SUBREG DPR:$src, (SSubReg_f32_reg imm:$lane)))>;
6921 def : VFPPat<(f64 (sint_to_fp (extractelt (v4i32 QPR:$src), imm:$lane))),
6922 (VSITOD (EXTRACT_SUBREG QPR:$src, (SSubReg_f32_reg imm:$lane)))>;
6923 def : VFPPat<(f64 (uint_to_fp (extractelt (v2i32 DPR:$src), imm:$lane))),
6924 (VUITOD (EXTRACT_SUBREG DPR:$src, (SSubReg_f32_reg imm:$lane)))>;
6925 def : VFPPat<(f64 (uint_to_fp (extractelt (v4i32 QPR:$src), imm:$lane))),
6926 (VUITOD (EXTRACT_SUBREG QPR:$src, (SSubReg_f32_reg imm:$lane)))>;
6929 // Prefer VMOVDRR for i32 -> f32 bitcasts, it can write all DPR registers.
6930 def : Pat<(f32 (bitconvert GPR:$a)),
6931 (EXTRACT_SUBREG (VMOVDRR GPR:$a, GPR:$a), ssub_0)>,
6932 Requires<[HasNEON, DontUseVMOVSR]>;
6934 //===----------------------------------------------------------------------===//
6935 // Non-Instruction Patterns
6936 //===----------------------------------------------------------------------===//
6939 let Predicates = [IsLE] in {
6940 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
6941 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
6942 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
6944 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
6945 let Predicates = [IsLE] in {
6946 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
6947 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
6948 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
6949 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
6950 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
6952 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
6953 let Predicates = [IsLE] in {
6954 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
6955 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
6956 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
6957 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
6958 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
6959 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
6960 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
6961 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
6962 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
6963 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
6965 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
6966 let Predicates = [IsLE] in {
6967 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
6968 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
6969 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
6970 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
6971 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
6972 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
6974 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
6975 let Predicates = [IsLE] in {
6976 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
6977 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
6980 let Predicates = [IsLE] in {
6981 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
6982 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
6983 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
6985 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
6986 let Predicates = [IsLE] in {
6987 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
6988 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
6989 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
6990 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
6991 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
6993 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
6994 let Predicates = [IsLE] in {
6995 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
6996 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
6997 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
6998 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
6999 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
7000 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
7001 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
7002 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
7003 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
7004 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
7005 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
7007 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
7008 let Predicates = [IsLE] in {
7009 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
7010 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
7011 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
7013 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
7014 let Predicates = [IsLE] in {
7015 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
7016 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
7017 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
7018 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;
7021 let Predicates = [IsBE] in {
7022 // 64 bit conversions
7023 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (VREV64d32 DPR:$src)>;
7024 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (VREV64d16 DPR:$src)>;
7025 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (VREV64d8 DPR:$src)>;
7026 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (VREV64d32 DPR:$src)>;
7027 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (VREV64d32 DPR:$src)>;
7028 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (VREV32d16 DPR:$src)>;
7029 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (VREV32d8 DPR:$src)>;
7030 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (VREV64d32 DPR:$src)>;
7031 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (VREV64d16 DPR:$src)>;
7032 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (VREV32d16 DPR:$src)>;
7033 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (VREV16d8 DPR:$src)>;
7034 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (VREV64d16 DPR:$src)>;
7035 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (VREV32d16 DPR:$src)>;
7036 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (VREV64d8 DPR:$src)>;
7037 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (VREV32d8 DPR:$src)>;
7038 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (VREV16d8 DPR:$src)>;
7039 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (VREV64d8 DPR:$src)>;
7040 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (VREV32d8 DPR:$src)>;
7041 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (VREV64d32 DPR:$src)>;
7042 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (VREV64d16 DPR:$src)>;
7043 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (VREV64d8 DPR:$src)>;
7044 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (VREV64d32 DPR:$src)>;
7045 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (VREV64d32 DPR:$src)>;
7046 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (VREV64d32 DPR:$src)>;
7047 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (VREV32d16 DPR:$src)>;
7048 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (VREV32d8 DPR:$src)>;
7050 // 128 bit conversions
7051 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (VREV64q32 QPR:$src)>;
7052 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (VREV64q16 QPR:$src)>;
7053 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (VREV64q8 QPR:$src)>;
7054 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (VREV64q32 QPR:$src)>;
7055 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (VREV64q32 QPR:$src)>;
7056 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (VREV32q16 QPR:$src)>;
7057 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (VREV32q8 QPR:$src)>;
7058 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (VREV64q32 QPR:$src)>;
7059 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (VREV64q16 QPR:$src)>;
7060 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (VREV32q16 QPR:$src)>;
7061 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (VREV16q8 QPR:$src)>;
7062 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (VREV64q16 QPR:$src)>;
7063 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (VREV32q16 QPR:$src)>;
7064 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (VREV64q8 QPR:$src)>;
7065 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (VREV32q8 QPR:$src)>;
7066 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (VREV16q8 QPR:$src)>;
7067 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (VREV64q8 QPR:$src)>;
7068 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (VREV32q8 QPR:$src)>;
7069 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (VREV64q32 QPR:$src)>;
7070 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (VREV32q16 QPR:$src)>;
7071 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (VREV32q8 QPR:$src)>;
7072 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (VREV64q32 QPR:$src)>;
7073 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (VREV64q32 QPR:$src)>;
7074 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (VREV64q16 QPR:$src)>;
7075 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (VREV64q8 QPR:$src)>;
7076 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (VREV64q32 QPR:$src)>;
7079 // Use VLD1/VST1 + VREV for non-word-aligned v2f64 load/store on Big Endian
7080 def : Pat<(v2f64 (byte_alignedload addrmode6:$addr)),
7081 (VREV64q8 (VLD1q8 addrmode6:$addr))>, Requires<[IsBE]>;
7082 def : Pat<(byte_alignedstore (v2f64 QPR:$value), addrmode6:$addr),
7083 (VST1q8 addrmode6:$addr, (VREV64q8 QPR:$value))>, Requires<[IsBE]>;
7084 def : Pat<(v2f64 (hword_alignedload addrmode6:$addr)),
7085 (VREV64q16 (VLD1q16 addrmode6:$addr))>, Requires<[IsBE]>;
7086 def : Pat<(hword_alignedstore (v2f64 QPR:$value), addrmode6:$addr),
7087 (VST1q16 addrmode6:$addr, (VREV64q16 QPR:$value))>, Requires<[IsBE]>;
7089 // Fold extracting an element out of a v2i32 into a vfp register.
7090 def : Pat<(f32 (bitconvert (i32 (extractelt (v2i32 DPR:$src), imm:$lane)))),
7091 (f32 (EXTRACT_SUBREG DPR:$src, (SSubReg_f32_reg imm:$lane)))>;
7093 // Vector lengthening move with load, matching extending loads.
7095 // extload, zextload and sextload for a standard lengthening load. Example:
7096 // Lengthen_Single<"8", "i16", "8"> =
7097 // Pat<(v8i16 (extloadvi8 addrmode6:$addr))
7098 // (VMOVLuv8i16 (VLD1d8 addrmode6:$addr,
7099 // (f64 (IMPLICIT_DEF)), (i32 0)))>;
7100 multiclass Lengthen_Single<string DestLanes, string DestTy, string SrcTy> {
7101 let AddedComplexity = 10 in {
7102 def _Any : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
7103 (!cast<PatFrag>("extloadvi" # SrcTy) addrmode6:$addr)),
7104 (!cast<Instruction>("VMOVLuv" # DestLanes # DestTy)
7105 (!cast<Instruction>("VLD1d" # SrcTy) addrmode6:$addr))>;
7107 def _Z : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
7108 (!cast<PatFrag>("zextloadvi" # SrcTy) addrmode6:$addr)),
7109 (!cast<Instruction>("VMOVLuv" # DestLanes # DestTy)
7110 (!cast<Instruction>("VLD1d" # SrcTy) addrmode6:$addr))>;
7112 def _S : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
7113 (!cast<PatFrag>("sextloadvi" # SrcTy) addrmode6:$addr)),
7114 (!cast<Instruction>("VMOVLsv" # DestLanes # DestTy)
7115 (!cast<Instruction>("VLD1d" # SrcTy) addrmode6:$addr))>;
7119 // extload, zextload and sextload for a lengthening load which only uses
7120 // half the lanes available. Example:
7121 // Lengthen_HalfSingle<"4", "i16", "8", "i16", "i8"> =
7122 // Pat<(v4i16 (extloadvi8 addrmode6oneL32:$addr)),
7123 // (EXTRACT_SUBREG (VMOVLuv8i16 (VLD1LNd32 addrmode6oneL32:$addr,
7124 // (f64 (IMPLICIT_DEF)), (i32 0))),
7126 multiclass Lengthen_HalfSingle<string DestLanes, string DestTy, string SrcTy,
7127 string InsnLanes, string InsnTy> {
7128 def _Any : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
7129 (!cast<PatFrag>("extloadv" # SrcTy) addrmode6oneL32:$addr)),
7130 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # InsnLanes # InsnTy)
7131 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
7133 def _Z : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
7134 (!cast<PatFrag>("zextloadv" # SrcTy) addrmode6oneL32:$addr)),
7135 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # InsnLanes # InsnTy)
7136 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
7138 def _S : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
7139 (!cast<PatFrag>("sextloadv" # SrcTy) addrmode6oneL32:$addr)),
7140 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLsv" # InsnLanes # InsnTy)
7141 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
7145 // The following class definition is basically a copy of the
7146 // Lengthen_HalfSingle definition above, however with an additional parameter
7147 // "RevLanes" to select the correct VREV32dXX instruction. This is to convert
7148 // data loaded by VLD1LN into proper vector format in big endian mode.
7149 multiclass Lengthen_HalfSingle_Big_Endian<string DestLanes, string DestTy, string SrcTy,
7150 string InsnLanes, string InsnTy, string RevLanes> {
7151 def _Any : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
7152 (!cast<PatFrag>("extloadv" # SrcTy) addrmode6oneL32:$addr)),
7153 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # InsnLanes # InsnTy)
7154 (!cast<Instruction>("VREV32d" # RevLanes)
7155 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0)))),
7157 def _Z : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
7158 (!cast<PatFrag>("zextloadv" # SrcTy) addrmode6oneL32:$addr)),
7159 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # InsnLanes # InsnTy)
7160 (!cast<Instruction>("VREV32d" # RevLanes)
7161 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0)))),
7163 def _S : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
7164 (!cast<PatFrag>("sextloadv" # SrcTy) addrmode6oneL32:$addr)),
7165 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLsv" # InsnLanes # InsnTy)
7166 (!cast<Instruction>("VREV32d" # RevLanes)
7167 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0)))),
7171 // extload, zextload and sextload for a lengthening load followed by another
7172 // lengthening load, to quadruple the initial length.
7174 // Lengthen_Double<"4", "i32", "i8", "8", "i16", "4", "i32"> =
7175 // Pat<(v4i32 (extloadvi8 addrmode6oneL32:$addr))
7176 // (EXTRACT_SUBREG (VMOVLuv4i32
7177 // (EXTRACT_SUBREG (VMOVLuv8i16 (VLD1LNd32 addrmode6oneL32:$addr,
7178 // (f64 (IMPLICIT_DEF)),
7182 multiclass Lengthen_Double<string DestLanes, string DestTy, string SrcTy,
7183 string Insn1Lanes, string Insn1Ty, string Insn2Lanes,
7185 def _Any : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
7186 (!cast<PatFrag>("extloadv" # SrcTy) addrmode6oneL32:$addr)),
7187 (!cast<Instruction>("VMOVLuv" # Insn2Lanes # Insn2Ty)
7188 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn1Lanes # Insn1Ty)
7189 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
7191 def _Z : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
7192 (!cast<PatFrag>("zextloadv" # SrcTy) addrmode6oneL32:$addr)),
7193 (!cast<Instruction>("VMOVLuv" # Insn2Lanes # Insn2Ty)
7194 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn1Lanes # Insn1Ty)
7195 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
7197 def _S : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
7198 (!cast<PatFrag>("sextloadv" # SrcTy) addrmode6oneL32:$addr)),
7199 (!cast<Instruction>("VMOVLsv" # Insn2Lanes # Insn2Ty)
7200 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLsv" # Insn1Lanes # Insn1Ty)
7201 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
7205 // The following class definition is basically a copy of the
7206 // Lengthen_Double definition above, however with an additional parameter
7207 // "RevLanes" to select the correct VREV32dXX instruction. This is to convert
7208 // data loaded by VLD1LN into proper vector format in big endian mode.
7209 multiclass Lengthen_Double_Big_Endian<string DestLanes, string DestTy, string SrcTy,
7210 string Insn1Lanes, string Insn1Ty, string Insn2Lanes,
7211 string Insn2Ty, string RevLanes> {
7212 def _Any : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
7213 (!cast<PatFrag>("extloadv" # SrcTy) addrmode6oneL32:$addr)),
7214 (!cast<Instruction>("VMOVLuv" # Insn2Lanes # Insn2Ty)
7215 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn1Lanes # Insn1Ty)
7216 (!cast<Instruction>("VREV32d" # RevLanes)
7217 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0)))),
7219 def _Z : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
7220 (!cast<PatFrag>("zextloadv" # SrcTy) addrmode6oneL32:$addr)),
7221 (!cast<Instruction>("VMOVLuv" # Insn2Lanes # Insn2Ty)
7222 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn1Lanes # Insn1Ty)
7223 (!cast<Instruction>("VREV32d" # RevLanes)
7224 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0)))),
7226 def _S : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
7227 (!cast<PatFrag>("sextloadv" # SrcTy) addrmode6oneL32:$addr)),
7228 (!cast<Instruction>("VMOVLsv" # Insn2Lanes # Insn2Ty)
7229 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLsv" # Insn1Lanes # Insn1Ty)
7230 (!cast<Instruction>("VREV32d" # RevLanes)
7231 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0)))),
7235 // extload, zextload and sextload for a lengthening load followed by another
7236 // lengthening load, to quadruple the initial length, but which ends up only
7237 // requiring half the available lanes (a 64-bit outcome instead of a 128-bit).
7239 // Lengthen_HalfDouble<"2", "i32", "i8", "8", "i16", "4", "i32"> =
7240 // Pat<(v2i32 (extloadvi8 addrmode6:$addr))
7241 // (EXTRACT_SUBREG (VMOVLuv4i32
7242 // (EXTRACT_SUBREG (VMOVLuv8i16 (VLD1LNd16 addrmode6:$addr,
7243 // (f64 (IMPLICIT_DEF)), (i32 0))),
7246 multiclass Lengthen_HalfDouble<string DestLanes, string DestTy, string SrcTy,
7247 string Insn1Lanes, string Insn1Ty, string Insn2Lanes,
7249 def _Any : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
7250 (!cast<PatFrag>("extloadv" # SrcTy) addrmode6:$addr)),
7251 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn2Lanes # Insn2Ty)
7252 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn1Lanes # Insn1Ty)
7253 (VLD1LNd16 addrmode6:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
7256 def _Z : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
7257 (!cast<PatFrag>("zextloadv" # SrcTy) addrmode6:$addr)),
7258 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn2Lanes # Insn2Ty)
7259 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn1Lanes # Insn1Ty)
7260 (VLD1LNd16 addrmode6:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
7263 def _S : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
7264 (!cast<PatFrag>("sextloadv" # SrcTy) addrmode6:$addr)),
7265 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLsv" # Insn2Lanes # Insn2Ty)
7266 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLsv" # Insn1Lanes # Insn1Ty)
7267 (VLD1LNd16 addrmode6:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
7272 // The following class definition is basically a copy of the
7273 // Lengthen_HalfDouble definition above, however with an additional VREV16d8
7274 // instruction to convert data loaded by VLD1LN into proper vector format
7275 // in big endian mode.
7276 multiclass Lengthen_HalfDouble_Big_Endian<string DestLanes, string DestTy, string SrcTy,
7277 string Insn1Lanes, string Insn1Ty, string Insn2Lanes,
7279 def _Any : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
7280 (!cast<PatFrag>("extloadv" # SrcTy) addrmode6:$addr)),
7281 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn2Lanes # Insn2Ty)
7282 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn1Lanes # Insn1Ty)
7283 (!cast<Instruction>("VREV16d8")
7284 (VLD1LNd16 addrmode6:$addr, (f64 (IMPLICIT_DEF)), (i32 0)))),
7287 def _Z : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
7288 (!cast<PatFrag>("zextloadv" # SrcTy) addrmode6:$addr)),
7289 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn2Lanes # Insn2Ty)
7290 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn1Lanes # Insn1Ty)
7291 (!cast<Instruction>("VREV16d8")
7292 (VLD1LNd16 addrmode6:$addr, (f64 (IMPLICIT_DEF)), (i32 0)))),
7295 def _S : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
7296 (!cast<PatFrag>("sextloadv" # SrcTy) addrmode6:$addr)),
7297 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLsv" # Insn2Lanes # Insn2Ty)
7298 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLsv" # Insn1Lanes # Insn1Ty)
7299 (!cast<Instruction>("VREV16d8")
7300 (VLD1LNd16 addrmode6:$addr, (f64 (IMPLICIT_DEF)), (i32 0)))),
7305 defm : Lengthen_Single<"8", "i16", "8">; // v8i8 -> v8i16
7306 defm : Lengthen_Single<"4", "i32", "16">; // v4i16 -> v4i32
7307 defm : Lengthen_Single<"2", "i64", "32">; // v2i32 -> v2i64
7309 let Predicates = [IsLE] in {
7310 defm : Lengthen_HalfSingle<"4", "i16", "i8", "8", "i16">; // v4i8 -> v4i16
7311 defm : Lengthen_HalfSingle<"2", "i32", "i16", "4", "i32">; // v2i16 -> v2i32
7313 // Double lengthening - v4i8 -> v4i16 -> v4i32
7314 defm : Lengthen_Double<"4", "i32", "i8", "8", "i16", "4", "i32">;
7315 // v2i8 -> v2i16 -> v2i32
7316 defm : Lengthen_HalfDouble<"2", "i32", "i8", "8", "i16", "4", "i32">;
7317 // v2i16 -> v2i32 -> v2i64
7318 defm : Lengthen_Double<"2", "i64", "i16", "4", "i32", "2", "i64">;
7321 let Predicates = [IsBE] in {
7322 defm : Lengthen_HalfSingle_Big_Endian<"4", "i16", "i8", "8", "i16", "8">; // v4i8 -> v4i16
7323 defm : Lengthen_HalfSingle_Big_Endian<"2", "i32", "i16", "4", "i32", "16">; // v2i16 -> v2i32
7325 // Double lengthening - v4i8 -> v4i16 -> v4i32
7326 defm : Lengthen_Double_Big_Endian<"4", "i32", "i8", "8", "i16", "4", "i32", "8">;
7327 // v2i8 -> v2i16 -> v2i32
7328 defm : Lengthen_HalfDouble_Big_Endian<"2", "i32", "i8", "8", "i16", "4", "i32">;
7329 // v2i16 -> v2i32 -> v2i64
7330 defm : Lengthen_Double_Big_Endian<"2", "i64", "i16", "4", "i32", "2", "i64", "16">;
7333 // Triple lengthening - v2i8 -> v2i16 -> v2i32 -> v2i64
7334 let Predicates = [IsLE] in {
7335 def : Pat<(v2i64 (extloadvi8 addrmode6:$addr)),
7336 (VMOVLuv2i64 (EXTRACT_SUBREG (VMOVLuv4i32 (EXTRACT_SUBREG (VMOVLuv8i16
7337 (VLD1LNd16 addrmode6:$addr,
7338 (f64 (IMPLICIT_DEF)), (i32 0))), dsub_0)), dsub_0))>;
7339 def : Pat<(v2i64 (zextloadvi8 addrmode6:$addr)),
7340 (VMOVLuv2i64 (EXTRACT_SUBREG (VMOVLuv4i32 (EXTRACT_SUBREG (VMOVLuv8i16
7341 (VLD1LNd16 addrmode6:$addr,
7342 (f64 (IMPLICIT_DEF)), (i32 0))), dsub_0)), dsub_0))>;
7343 def : Pat<(v2i64 (sextloadvi8 addrmode6:$addr)),
7344 (VMOVLsv2i64 (EXTRACT_SUBREG (VMOVLsv4i32 (EXTRACT_SUBREG (VMOVLsv8i16
7345 (VLD1LNd16 addrmode6:$addr,
7346 (f64 (IMPLICIT_DEF)), (i32 0))), dsub_0)), dsub_0))>;
7348 // The following patterns are basically a copy of the patterns above,
7349 // however with an additional VREV16d instruction to convert data
7350 // loaded by VLD1LN into proper vector format in big endian mode.
7351 let Predicates = [IsBE] in {
7352 def : Pat<(v2i64 (extloadvi8 addrmode6:$addr)),
7353 (VMOVLuv2i64 (EXTRACT_SUBREG (VMOVLuv4i32 (EXTRACT_SUBREG (VMOVLuv8i16
7354 (!cast<Instruction>("VREV16d8")
7355 (VLD1LNd16 addrmode6:$addr,
7356 (f64 (IMPLICIT_DEF)), (i32 0)))), dsub_0)), dsub_0))>;
7357 def : Pat<(v2i64 (zextloadvi8 addrmode6:$addr)),
7358 (VMOVLuv2i64 (EXTRACT_SUBREG (VMOVLuv4i32 (EXTRACT_SUBREG (VMOVLuv8i16
7359 (!cast<Instruction>("VREV16d8")
7360 (VLD1LNd16 addrmode6:$addr,
7361 (f64 (IMPLICIT_DEF)), (i32 0)))), dsub_0)), dsub_0))>;
7362 def : Pat<(v2i64 (sextloadvi8 addrmode6:$addr)),
7363 (VMOVLsv2i64 (EXTRACT_SUBREG (VMOVLsv4i32 (EXTRACT_SUBREG (VMOVLsv8i16
7364 (!cast<Instruction>("VREV16d8")
7365 (VLD1LNd16 addrmode6:$addr,
7366 (f64 (IMPLICIT_DEF)), (i32 0)))), dsub_0)), dsub_0))>;
7369 def : Pat<(v2i64 (concat_vectors DPR:$Dn, DPR:$Dm)),
7370 (REG_SEQUENCE QPR, DPR:$Dn, dsub_0, DPR:$Dm, dsub_1)>;
7371 def : Pat<(v4i32 (concat_vectors DPR:$Dn, DPR:$Dm)),
7372 (REG_SEQUENCE QPR, DPR:$Dn, dsub_0, DPR:$Dm, dsub_1)>;
7373 def : Pat<(v8i16 (concat_vectors DPR:$Dn, DPR:$Dm)),
7374 (REG_SEQUENCE QPR, DPR:$Dn, dsub_0, DPR:$Dm, dsub_1)>;
7375 def : Pat<(v16i8 (concat_vectors DPR:$Dn, DPR:$Dm)),
7376 (REG_SEQUENCE QPR, DPR:$Dn, dsub_0, DPR:$Dm, dsub_1)>;
7377 def : Pat<(v4f32 (concat_vectors DPR:$Dn, DPR:$Dm)),
7378 (REG_SEQUENCE QPR, DPR:$Dn, dsub_0, DPR:$Dm, dsub_1)>;
7380 //===----------------------------------------------------------------------===//
7381 // Assembler aliases
7384 def : VFP2InstAlias<"fmdhr${p} $Dd, $Rn",
7385 (VSETLNi32 DPR:$Dd, GPR:$Rn, 1, pred:$p)>;
7386 def : VFP2InstAlias<"fmdlr${p} $Dd, $Rn",
7387 (VSETLNi32 DPR:$Dd, GPR:$Rn, 0, pred:$p)>;
7389 // VAND/VBIC/VEOR/VORR accept but do not require a type suffix.
7390 defm : NEONDTAnyInstAlias<"vand${p}", "$Vd, $Vn, $Vm",
7391 (VANDd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
7392 defm : NEONDTAnyInstAlias<"vand${p}", "$Vd, $Vn, $Vm",
7393 (VANDq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
7394 defm : NEONDTAnyInstAlias<"vbic${p}", "$Vd, $Vn, $Vm",
7395 (VBICd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
7396 defm : NEONDTAnyInstAlias<"vbic${p}", "$Vd, $Vn, $Vm",
7397 (VBICq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
7398 defm : NEONDTAnyInstAlias<"veor${p}", "$Vd, $Vn, $Vm",
7399 (VEORd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
7400 defm : NEONDTAnyInstAlias<"veor${p}", "$Vd, $Vn, $Vm",
7401 (VEORq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
7402 defm : NEONDTAnyInstAlias<"vorr${p}", "$Vd, $Vn, $Vm",
7403 (VORRd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
7404 defm : NEONDTAnyInstAlias<"vorr${p}", "$Vd, $Vn, $Vm",
7405 (VORRq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
7406 // ... two-operand aliases
7407 defm : NEONDTAnyInstAlias<"vand${p}", "$Vdn, $Vm",
7408 (VANDd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
7409 defm : NEONDTAnyInstAlias<"vand${p}", "$Vdn, $Vm",
7410 (VANDq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
7411 defm : NEONDTAnyInstAlias<"veor${p}", "$Vdn, $Vm",
7412 (VEORd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
7413 defm : NEONDTAnyInstAlias<"veor${p}", "$Vdn, $Vm",
7414 (VEORq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
7415 defm : NEONDTAnyInstAlias<"vorr${p}", "$Vdn, $Vm",
7416 (VORRd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
7417 defm : NEONDTAnyInstAlias<"vorr${p}", "$Vdn, $Vm",
7418 (VORRq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
7420 def : NEONInstAlias<"vand${p}.i16 $Vd, $imm",
7421 (VBICiv4i16 DPR:$Vd, nImmSplatNotI16:$imm, pred:$p)>;
7422 def : NEONInstAlias<"vand${p}.i32 $Vd, $imm",
7423 (VBICiv2i32 DPR:$Vd, nImmSplatNotI32:$imm, pred:$p)>;
7424 def : NEONInstAlias<"vand${p}.i16 $Vd, $imm",
7425 (VBICiv8i16 QPR:$Vd, nImmSplatNotI16:$imm, pred:$p)>;
7426 def : NEONInstAlias<"vand${p}.i32 $Vd, $imm",
7427 (VBICiv4i32 QPR:$Vd, nImmSplatNotI32:$imm, pred:$p)>;
7430 // VLD1 single-lane pseudo-instructions. These need special handling for
7431 // the lane index that an InstAlias can't handle, so we use these instead.
7432 def VLD1LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vld1${p}", ".8", "$list, $addr",
7433 (ins VecListOneDByteIndexed:$list, addrmode6alignNone:$addr,
7435 def VLD1LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vld1${p}", ".16", "$list, $addr",
7436 (ins VecListOneDHWordIndexed:$list, addrmode6align16:$addr,
7438 def VLD1LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vld1${p}", ".32", "$list, $addr",
7439 (ins VecListOneDWordIndexed:$list, addrmode6align32:$addr,
7442 def VLD1LNdWB_fixed_Asm_8 :
7443 NEONDataTypeAsmPseudoInst<"vld1${p}", ".8", "$list, $addr!",
7444 (ins VecListOneDByteIndexed:$list, addrmode6alignNone:$addr,
7446 def VLD1LNdWB_fixed_Asm_16 :
7447 NEONDataTypeAsmPseudoInst<"vld1${p}", ".16", "$list, $addr!",
7448 (ins VecListOneDHWordIndexed:$list, addrmode6align16:$addr,
7450 def VLD1LNdWB_fixed_Asm_32 :
7451 NEONDataTypeAsmPseudoInst<"vld1${p}", ".32", "$list, $addr!",
7452 (ins VecListOneDWordIndexed:$list, addrmode6align32:$addr,
7454 def VLD1LNdWB_register_Asm_8 :
7455 NEONDataTypeAsmPseudoInst<"vld1${p}", ".8", "$list, $addr, $Rm",
7456 (ins VecListOneDByteIndexed:$list, addrmode6alignNone:$addr,
7457 rGPR:$Rm, pred:$p)>;
7458 def VLD1LNdWB_register_Asm_16 :
7459 NEONDataTypeAsmPseudoInst<"vld1${p}", ".16", "$list, $addr, $Rm",
7460 (ins VecListOneDHWordIndexed:$list, addrmode6align16:$addr,
7461 rGPR:$Rm, pred:$p)>;
7462 def VLD1LNdWB_register_Asm_32 :
7463 NEONDataTypeAsmPseudoInst<"vld1${p}", ".32", "$list, $addr, $Rm",
7464 (ins VecListOneDWordIndexed:$list, addrmode6align32:$addr,
7465 rGPR:$Rm, pred:$p)>;
7468 // VST1 single-lane pseudo-instructions. These need special handling for
7469 // the lane index that an InstAlias can't handle, so we use these instead.
7470 def VST1LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vst1${p}", ".8", "$list, $addr",
7471 (ins VecListOneDByteIndexed:$list, addrmode6alignNone:$addr,
7473 def VST1LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vst1${p}", ".16", "$list, $addr",
7474 (ins VecListOneDHWordIndexed:$list, addrmode6align16:$addr,
7476 def VST1LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vst1${p}", ".32", "$list, $addr",
7477 (ins VecListOneDWordIndexed:$list, addrmode6align32:$addr,
7480 def VST1LNdWB_fixed_Asm_8 :
7481 NEONDataTypeAsmPseudoInst<"vst1${p}", ".8", "$list, $addr!",
7482 (ins VecListOneDByteIndexed:$list, addrmode6alignNone:$addr,
7484 def VST1LNdWB_fixed_Asm_16 :
7485 NEONDataTypeAsmPseudoInst<"vst1${p}", ".16", "$list, $addr!",
7486 (ins VecListOneDHWordIndexed:$list, addrmode6align16:$addr,
7488 def VST1LNdWB_fixed_Asm_32 :
7489 NEONDataTypeAsmPseudoInst<"vst1${p}", ".32", "$list, $addr!",
7490 (ins VecListOneDWordIndexed:$list, addrmode6align32:$addr,
7492 def VST1LNdWB_register_Asm_8 :
7493 NEONDataTypeAsmPseudoInst<"vst1${p}", ".8", "$list, $addr, $Rm",
7494 (ins VecListOneDByteIndexed:$list, addrmode6alignNone:$addr,
7495 rGPR:$Rm, pred:$p)>;
7496 def VST1LNdWB_register_Asm_16 :
7497 NEONDataTypeAsmPseudoInst<"vst1${p}", ".16", "$list, $addr, $Rm",
7498 (ins VecListOneDHWordIndexed:$list, addrmode6align16:$addr,
7499 rGPR:$Rm, pred:$p)>;
7500 def VST1LNdWB_register_Asm_32 :
7501 NEONDataTypeAsmPseudoInst<"vst1${p}", ".32", "$list, $addr, $Rm",
7502 (ins VecListOneDWordIndexed:$list, addrmode6align32:$addr,
7503 rGPR:$Rm, pred:$p)>;
7505 // VLD2 single-lane pseudo-instructions. These need special handling for
7506 // the lane index that an InstAlias can't handle, so we use these instead.
7507 def VLD2LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vld2${p}", ".8", "$list, $addr",
7508 (ins VecListTwoDByteIndexed:$list, addrmode6align16:$addr,
7510 def VLD2LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr",
7511 (ins VecListTwoDHWordIndexed:$list, addrmode6align32:$addr,
7513 def VLD2LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr",
7514 (ins VecListTwoDWordIndexed:$list, addrmode6align64:$addr, pred:$p)>;
7515 def VLD2LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr",
7516 (ins VecListTwoQHWordIndexed:$list, addrmode6align32:$addr,
7518 def VLD2LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr",
7519 (ins VecListTwoQWordIndexed:$list, addrmode6align64:$addr,
7522 def VLD2LNdWB_fixed_Asm_8 :
7523 NEONDataTypeAsmPseudoInst<"vld2${p}", ".8", "$list, $addr!",
7524 (ins VecListTwoDByteIndexed:$list, addrmode6align16:$addr,
7526 def VLD2LNdWB_fixed_Asm_16 :
7527 NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr!",
7528 (ins VecListTwoDHWordIndexed:$list, addrmode6align32:$addr,
7530 def VLD2LNdWB_fixed_Asm_32 :
7531 NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr!",
7532 (ins VecListTwoDWordIndexed:$list, addrmode6align64:$addr,
7534 def VLD2LNqWB_fixed_Asm_16 :
7535 NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr!",
7536 (ins VecListTwoQHWordIndexed:$list, addrmode6align32:$addr,
7538 def VLD2LNqWB_fixed_Asm_32 :
7539 NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr!",
7540 (ins VecListTwoQWordIndexed:$list, addrmode6align64:$addr,
7542 def VLD2LNdWB_register_Asm_8 :
7543 NEONDataTypeAsmPseudoInst<"vld2${p}", ".8", "$list, $addr, $Rm",
7544 (ins VecListTwoDByteIndexed:$list, addrmode6align16:$addr,
7545 rGPR:$Rm, pred:$p)>;
7546 def VLD2LNdWB_register_Asm_16 :
7547 NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr, $Rm",
7548 (ins VecListTwoDHWordIndexed:$list, addrmode6align32:$addr,
7549 rGPR:$Rm, pred:$p)>;
7550 def VLD2LNdWB_register_Asm_32 :
7551 NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr, $Rm",
7552 (ins VecListTwoDWordIndexed:$list, addrmode6align64:$addr,
7553 rGPR:$Rm, pred:$p)>;
7554 def VLD2LNqWB_register_Asm_16 :
7555 NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr, $Rm",
7556 (ins VecListTwoQHWordIndexed:$list, addrmode6align32:$addr,
7557 rGPR:$Rm, pred:$p)>;
7558 def VLD2LNqWB_register_Asm_32 :
7559 NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr, $Rm",
7560 (ins VecListTwoQWordIndexed:$list, addrmode6align64:$addr,
7561 rGPR:$Rm, pred:$p)>;
7564 // VST2 single-lane pseudo-instructions. These need special handling for
7565 // the lane index that an InstAlias can't handle, so we use these instead.
7566 def VST2LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vst2${p}", ".8", "$list, $addr",
7567 (ins VecListTwoDByteIndexed:$list, addrmode6align16:$addr,
7569 def VST2LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vst2${p}", ".16", "$list, $addr",
7570 (ins VecListTwoDHWordIndexed:$list, addrmode6align32:$addr,
7572 def VST2LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr",
7573 (ins VecListTwoDWordIndexed:$list, addrmode6align64:$addr,
7575 def VST2LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vst2${p}", ".16", "$list, $addr",
7576 (ins VecListTwoQHWordIndexed:$list, addrmode6align32:$addr,
7578 def VST2LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr",
7579 (ins VecListTwoQWordIndexed:$list, addrmode6align64:$addr,
7582 def VST2LNdWB_fixed_Asm_8 :
7583 NEONDataTypeAsmPseudoInst<"vst2${p}", ".8", "$list, $addr!",
7584 (ins VecListTwoDByteIndexed:$list, addrmode6align16:$addr,
7586 def VST2LNdWB_fixed_Asm_16 :
7587 NEONDataTypeAsmPseudoInst<"vst2${p}", ".16", "$list, $addr!",
7588 (ins VecListTwoDHWordIndexed:$list, addrmode6align32:$addr,
7590 def VST2LNdWB_fixed_Asm_32 :
7591 NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr!",
7592 (ins VecListTwoDWordIndexed:$list, addrmode6align64:$addr,
7594 def VST2LNqWB_fixed_Asm_16 :
7595 NEONDataTypeAsmPseudoInst<"vst2${p}", ".16", "$list, $addr!",
7596 (ins VecListTwoQHWordIndexed:$list, addrmode6align32:$addr,
7598 def VST2LNqWB_fixed_Asm_32 :
7599 NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr!",
7600 (ins VecListTwoQWordIndexed:$list, addrmode6align64:$addr,
7602 def VST2LNdWB_register_Asm_8 :
7603 NEONDataTypeAsmPseudoInst<"vst2${p}", ".8", "$list, $addr, $Rm",
7604 (ins VecListTwoDByteIndexed:$list, addrmode6align16:$addr,
7605 rGPR:$Rm, pred:$p)>;
7606 def VST2LNdWB_register_Asm_16 :
7607 NEONDataTypeAsmPseudoInst<"vst2${p}", ".16","$list, $addr, $Rm",
7608 (ins VecListTwoDHWordIndexed:$list, addrmode6align32:$addr,
7609 rGPR:$Rm, pred:$p)>;
7610 def VST2LNdWB_register_Asm_32 :
7611 NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr, $Rm",
7612 (ins VecListTwoDWordIndexed:$list, addrmode6align64:$addr,
7613 rGPR:$Rm, pred:$p)>;
7614 def VST2LNqWB_register_Asm_16 :
7615 NEONDataTypeAsmPseudoInst<"vst2${p}", ".16","$list, $addr, $Rm",
7616 (ins VecListTwoQHWordIndexed:$list, addrmode6align32:$addr,
7617 rGPR:$Rm, pred:$p)>;
7618 def VST2LNqWB_register_Asm_32 :
7619 NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr, $Rm",
7620 (ins VecListTwoQWordIndexed:$list, addrmode6align64:$addr,
7621 rGPR:$Rm, pred:$p)>;
7623 // VLD3 all-lanes pseudo-instructions. These need special handling for
7624 // the lane index that an InstAlias can't handle, so we use these instead.
7625 def VLD3DUPdAsm_8 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr",
7626 (ins VecListThreeDAllLanes:$list, addrmode6dupalignNone:$addr,
7628 def VLD3DUPdAsm_16: NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
7629 (ins VecListThreeDAllLanes:$list, addrmode6dupalignNone:$addr,
7631 def VLD3DUPdAsm_32: NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
7632 (ins VecListThreeDAllLanes:$list, addrmode6dupalignNone:$addr,
7634 def VLD3DUPqAsm_8 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr",
7635 (ins VecListThreeQAllLanes:$list, addrmode6dupalignNone:$addr,
7637 def VLD3DUPqAsm_16: NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
7638 (ins VecListThreeQAllLanes:$list, addrmode6dupalignNone:$addr,
7640 def VLD3DUPqAsm_32: NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
7641 (ins VecListThreeQAllLanes:$list, addrmode6dupalignNone:$addr,
7644 def VLD3DUPdWB_fixed_Asm_8 :
7645 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr!",
7646 (ins VecListThreeDAllLanes:$list, addrmode6dupalignNone:$addr,
7648 def VLD3DUPdWB_fixed_Asm_16 :
7649 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
7650 (ins VecListThreeDAllLanes:$list, addrmode6dupalignNone:$addr,
7652 def VLD3DUPdWB_fixed_Asm_32 :
7653 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
7654 (ins VecListThreeDAllLanes:$list, addrmode6dupalignNone:$addr,
7656 def VLD3DUPqWB_fixed_Asm_8 :
7657 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr!",
7658 (ins VecListThreeQAllLanes:$list, addrmode6dupalignNone:$addr,
7660 def VLD3DUPqWB_fixed_Asm_16 :
7661 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
7662 (ins VecListThreeQAllLanes:$list, addrmode6dupalignNone:$addr,
7664 def VLD3DUPqWB_fixed_Asm_32 :
7665 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
7666 (ins VecListThreeQAllLanes:$list, addrmode6dupalignNone:$addr,
7668 def VLD3DUPdWB_register_Asm_8 :
7669 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr, $Rm",
7670 (ins VecListThreeDAllLanes:$list, addrmode6dupalignNone:$addr,
7671 rGPR:$Rm, pred:$p)>;
7672 def VLD3DUPdWB_register_Asm_16 :
7673 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
7674 (ins VecListThreeDAllLanes:$list, addrmode6dupalignNone:$addr,
7675 rGPR:$Rm, pred:$p)>;
7676 def VLD3DUPdWB_register_Asm_32 :
7677 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
7678 (ins VecListThreeDAllLanes:$list, addrmode6dupalignNone:$addr,
7679 rGPR:$Rm, pred:$p)>;
7680 def VLD3DUPqWB_register_Asm_8 :
7681 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr, $Rm",
7682 (ins VecListThreeQAllLanes:$list, addrmode6dupalignNone:$addr,
7683 rGPR:$Rm, pred:$p)>;
7684 def VLD3DUPqWB_register_Asm_16 :
7685 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
7686 (ins VecListThreeQAllLanes:$list, addrmode6dupalignNone:$addr,
7687 rGPR:$Rm, pred:$p)>;
7688 def VLD3DUPqWB_register_Asm_32 :
7689 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
7690 (ins VecListThreeQAllLanes:$list, addrmode6dupalignNone:$addr,
7691 rGPR:$Rm, pred:$p)>;
7694 // VLD3 single-lane pseudo-instructions. These need special handling for
7695 // the lane index that an InstAlias can't handle, so we use these instead.
7696 def VLD3LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr",
7697 (ins VecListThreeDByteIndexed:$list, addrmode6alignNone:$addr,
7699 def VLD3LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
7700 (ins VecListThreeDHWordIndexed:$list, addrmode6alignNone:$addr,
7702 def VLD3LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
7703 (ins VecListThreeDWordIndexed:$list, addrmode6alignNone:$addr,
7705 def VLD3LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
7706 (ins VecListThreeQHWordIndexed:$list, addrmode6alignNone:$addr,
7708 def VLD3LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
7709 (ins VecListThreeQWordIndexed:$list, addrmode6alignNone:$addr,
7712 def VLD3LNdWB_fixed_Asm_8 :
7713 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr!",
7714 (ins VecListThreeDByteIndexed:$list, addrmode6alignNone:$addr,
7716 def VLD3LNdWB_fixed_Asm_16 :
7717 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
7718 (ins VecListThreeDHWordIndexed:$list, addrmode6alignNone:$addr,
7720 def VLD3LNdWB_fixed_Asm_32 :
7721 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
7722 (ins VecListThreeDWordIndexed:$list, addrmode6alignNone:$addr,
7724 def VLD3LNqWB_fixed_Asm_16 :
7725 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
7726 (ins VecListThreeQHWordIndexed:$list, addrmode6alignNone:$addr,
7728 def VLD3LNqWB_fixed_Asm_32 :
7729 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
7730 (ins VecListThreeQWordIndexed:$list, addrmode6alignNone:$addr,
7732 def VLD3LNdWB_register_Asm_8 :
7733 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr, $Rm",
7734 (ins VecListThreeDByteIndexed:$list, addrmode6alignNone:$addr,
7735 rGPR:$Rm, pred:$p)>;
7736 def VLD3LNdWB_register_Asm_16 :
7737 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
7738 (ins VecListThreeDHWordIndexed:$list,
7739 addrmode6alignNone:$addr, rGPR:$Rm, pred:$p)>;
7740 def VLD3LNdWB_register_Asm_32 :
7741 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
7742 (ins VecListThreeDWordIndexed:$list, addrmode6alignNone:$addr,
7743 rGPR:$Rm, pred:$p)>;
7744 def VLD3LNqWB_register_Asm_16 :
7745 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
7746 (ins VecListThreeQHWordIndexed:$list,
7747 addrmode6alignNone:$addr, rGPR:$Rm, pred:$p)>;
7748 def VLD3LNqWB_register_Asm_32 :
7749 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
7750 (ins VecListThreeQWordIndexed:$list, addrmode6alignNone:$addr,
7751 rGPR:$Rm, pred:$p)>;
7753 // VLD3 multiple structure pseudo-instructions. These need special handling for
7754 // the vector operands that the normal instructions don't yet model.
7755 // FIXME: Remove these when the register classes and instructions are updated.
7756 def VLD3dAsm_8 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr",
7757 (ins VecListThreeD:$list, addrmode6align64:$addr, pred:$p)>;
7758 def VLD3dAsm_16 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
7759 (ins VecListThreeD:$list, addrmode6align64:$addr, pred:$p)>;
7760 def VLD3dAsm_32 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
7761 (ins VecListThreeD:$list, addrmode6align64:$addr, pred:$p)>;
7762 def VLD3qAsm_8 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr",
7763 (ins VecListThreeQ:$list, addrmode6align64:$addr, pred:$p)>;
7764 def VLD3qAsm_16 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
7765 (ins VecListThreeQ:$list, addrmode6align64:$addr, pred:$p)>;
7766 def VLD3qAsm_32 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
7767 (ins VecListThreeQ:$list, addrmode6align64:$addr, pred:$p)>;
7769 def VLD3dWB_fixed_Asm_8 :
7770 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr!",
7771 (ins VecListThreeD:$list, addrmode6align64:$addr, pred:$p)>;
7772 def VLD3dWB_fixed_Asm_16 :
7773 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
7774 (ins VecListThreeD:$list, addrmode6align64:$addr, pred:$p)>;
7775 def VLD3dWB_fixed_Asm_32 :
7776 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
7777 (ins VecListThreeD:$list, addrmode6align64:$addr, pred:$p)>;
7778 def VLD3qWB_fixed_Asm_8 :
7779 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr!",
7780 (ins VecListThreeQ:$list, addrmode6align64:$addr, pred:$p)>;
7781 def VLD3qWB_fixed_Asm_16 :
7782 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
7783 (ins VecListThreeQ:$list, addrmode6align64:$addr, pred:$p)>;
7784 def VLD3qWB_fixed_Asm_32 :
7785 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
7786 (ins VecListThreeQ:$list, addrmode6align64:$addr, pred:$p)>;
7787 def VLD3dWB_register_Asm_8 :
7788 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr, $Rm",
7789 (ins VecListThreeD:$list, addrmode6align64:$addr,
7790 rGPR:$Rm, pred:$p)>;
7791 def VLD3dWB_register_Asm_16 :
7792 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
7793 (ins VecListThreeD:$list, addrmode6align64:$addr,
7794 rGPR:$Rm, pred:$p)>;
7795 def VLD3dWB_register_Asm_32 :
7796 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
7797 (ins VecListThreeD:$list, addrmode6align64:$addr,
7798 rGPR:$Rm, pred:$p)>;
7799 def VLD3qWB_register_Asm_8 :
7800 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr, $Rm",
7801 (ins VecListThreeQ:$list, addrmode6align64:$addr,
7802 rGPR:$Rm, pred:$p)>;
7803 def VLD3qWB_register_Asm_16 :
7804 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
7805 (ins VecListThreeQ:$list, addrmode6align64:$addr,
7806 rGPR:$Rm, pred:$p)>;
7807 def VLD3qWB_register_Asm_32 :
7808 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
7809 (ins VecListThreeQ:$list, addrmode6align64:$addr,
7810 rGPR:$Rm, pred:$p)>;
7812 // VST3 single-lane pseudo-instructions. These need special handling for
7813 // the lane index that an InstAlias can't handle, so we use these instead.
7814 def VST3LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr",
7815 (ins VecListThreeDByteIndexed:$list, addrmode6alignNone:$addr,
7817 def VST3LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr",
7818 (ins VecListThreeDHWordIndexed:$list, addrmode6alignNone:$addr,
7820 def VST3LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr",
7821 (ins VecListThreeDWordIndexed:$list, addrmode6alignNone:$addr,
7823 def VST3LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr",
7824 (ins VecListThreeQHWordIndexed:$list, addrmode6alignNone:$addr,
7826 def VST3LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr",
7827 (ins VecListThreeQWordIndexed:$list, addrmode6alignNone:$addr,
7830 def VST3LNdWB_fixed_Asm_8 :
7831 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr!",
7832 (ins VecListThreeDByteIndexed:$list, addrmode6alignNone:$addr,
7834 def VST3LNdWB_fixed_Asm_16 :
7835 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr!",
7836 (ins VecListThreeDHWordIndexed:$list, addrmode6alignNone:$addr,
7838 def VST3LNdWB_fixed_Asm_32 :
7839 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr!",
7840 (ins VecListThreeDWordIndexed:$list, addrmode6alignNone:$addr,
7842 def VST3LNqWB_fixed_Asm_16 :
7843 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr!",
7844 (ins VecListThreeQHWordIndexed:$list, addrmode6alignNone:$addr,
7846 def VST3LNqWB_fixed_Asm_32 :
7847 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr!",
7848 (ins VecListThreeQWordIndexed:$list, addrmode6alignNone:$addr,
7850 def VST3LNdWB_register_Asm_8 :
7851 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr, $Rm",
7852 (ins VecListThreeDByteIndexed:$list, addrmode6alignNone:$addr,
7853 rGPR:$Rm, pred:$p)>;
7854 def VST3LNdWB_register_Asm_16 :
7855 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr, $Rm",
7856 (ins VecListThreeDHWordIndexed:$list,
7857 addrmode6alignNone:$addr, rGPR:$Rm, pred:$p)>;
7858 def VST3LNdWB_register_Asm_32 :
7859 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr, $Rm",
7860 (ins VecListThreeDWordIndexed:$list, addrmode6alignNone:$addr,
7861 rGPR:$Rm, pred:$p)>;
7862 def VST3LNqWB_register_Asm_16 :
7863 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr, $Rm",
7864 (ins VecListThreeQHWordIndexed:$list,
7865 addrmode6alignNone:$addr, rGPR:$Rm, pred:$p)>;
7866 def VST3LNqWB_register_Asm_32 :
7867 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr, $Rm",
7868 (ins VecListThreeQWordIndexed:$list, addrmode6alignNone:$addr,
7869 rGPR:$Rm, pred:$p)>;
7872 // VST3 multiple structure pseudo-instructions. These need special handling for
7873 // the vector operands that the normal instructions don't yet model.
7874 // FIXME: Remove these when the register classes and instructions are updated.
7875 def VST3dAsm_8 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr",
7876 (ins VecListThreeD:$list, addrmode6align64:$addr, pred:$p)>;
7877 def VST3dAsm_16 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr",
7878 (ins VecListThreeD:$list, addrmode6align64:$addr, pred:$p)>;
7879 def VST3dAsm_32 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr",
7880 (ins VecListThreeD:$list, addrmode6align64:$addr, pred:$p)>;
7881 def VST3qAsm_8 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr",
7882 (ins VecListThreeQ:$list, addrmode6align64:$addr, pred:$p)>;
7883 def VST3qAsm_16 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr",
7884 (ins VecListThreeQ:$list, addrmode6align64:$addr, pred:$p)>;
7885 def VST3qAsm_32 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr",
7886 (ins VecListThreeQ:$list, addrmode6align64:$addr, pred:$p)>;
7888 def VST3dWB_fixed_Asm_8 :
7889 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr!",
7890 (ins VecListThreeD:$list, addrmode6align64:$addr, pred:$p)>;
7891 def VST3dWB_fixed_Asm_16 :
7892 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr!",
7893 (ins VecListThreeD:$list, addrmode6align64:$addr, pred:$p)>;
7894 def VST3dWB_fixed_Asm_32 :
7895 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr!",
7896 (ins VecListThreeD:$list, addrmode6align64:$addr, pred:$p)>;
7897 def VST3qWB_fixed_Asm_8 :
7898 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr!",
7899 (ins VecListThreeQ:$list, addrmode6align64:$addr, pred:$p)>;
7900 def VST3qWB_fixed_Asm_16 :
7901 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr!",
7902 (ins VecListThreeQ:$list, addrmode6align64:$addr, pred:$p)>;
7903 def VST3qWB_fixed_Asm_32 :
7904 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr!",
7905 (ins VecListThreeQ:$list, addrmode6align64:$addr, pred:$p)>;
7906 def VST3dWB_register_Asm_8 :
7907 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr, $Rm",
7908 (ins VecListThreeD:$list, addrmode6align64:$addr,
7909 rGPR:$Rm, pred:$p)>;
7910 def VST3dWB_register_Asm_16 :
7911 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr, $Rm",
7912 (ins VecListThreeD:$list, addrmode6align64:$addr,
7913 rGPR:$Rm, pred:$p)>;
7914 def VST3dWB_register_Asm_32 :
7915 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr, $Rm",
7916 (ins VecListThreeD:$list, addrmode6align64:$addr,
7917 rGPR:$Rm, pred:$p)>;
7918 def VST3qWB_register_Asm_8 :
7919 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr, $Rm",
7920 (ins VecListThreeQ:$list, addrmode6align64:$addr,
7921 rGPR:$Rm, pred:$p)>;
7922 def VST3qWB_register_Asm_16 :
7923 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr, $Rm",
7924 (ins VecListThreeQ:$list, addrmode6align64:$addr,
7925 rGPR:$Rm, pred:$p)>;
7926 def VST3qWB_register_Asm_32 :
7927 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr, $Rm",
7928 (ins VecListThreeQ:$list, addrmode6align64:$addr,
7929 rGPR:$Rm, pred:$p)>;
7931 // VLD4 all-lanes pseudo-instructions. These need special handling for
7932 // the lane index that an InstAlias can't handle, so we use these instead.
7933 def VLD4DUPdAsm_8 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr",
7934 (ins VecListFourDAllLanes:$list, addrmode6dupalign32:$addr,
7936 def VLD4DUPdAsm_16: NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
7937 (ins VecListFourDAllLanes:$list, addrmode6dupalign64:$addr,
7939 def VLD4DUPdAsm_32: NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
7940 (ins VecListFourDAllLanes:$list, addrmode6dupalign64or128:$addr,
7942 def VLD4DUPqAsm_8 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr",
7943 (ins VecListFourQAllLanes:$list, addrmode6dupalign32:$addr,
7945 def VLD4DUPqAsm_16: NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
7946 (ins VecListFourQAllLanes:$list, addrmode6dupalign64:$addr,
7948 def VLD4DUPqAsm_32: NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
7949 (ins VecListFourQAllLanes:$list, addrmode6dupalign64or128:$addr,
7952 def VLD4DUPdWB_fixed_Asm_8 :
7953 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr!",
7954 (ins VecListFourDAllLanes:$list, addrmode6dupalign32:$addr,
7956 def VLD4DUPdWB_fixed_Asm_16 :
7957 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
7958 (ins VecListFourDAllLanes:$list, addrmode6dupalign64:$addr,
7960 def VLD4DUPdWB_fixed_Asm_32 :
7961 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
7962 (ins VecListFourDAllLanes:$list, addrmode6dupalign64or128:$addr,
7964 def VLD4DUPqWB_fixed_Asm_8 :
7965 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr!",
7966 (ins VecListFourQAllLanes:$list, addrmode6dupalign32:$addr,
7968 def VLD4DUPqWB_fixed_Asm_16 :
7969 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
7970 (ins VecListFourQAllLanes:$list, addrmode6dupalign64:$addr,
7972 def VLD4DUPqWB_fixed_Asm_32 :
7973 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
7974 (ins VecListFourQAllLanes:$list, addrmode6dupalign64or128:$addr,
7976 def VLD4DUPdWB_register_Asm_8 :
7977 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr, $Rm",
7978 (ins VecListFourDAllLanes:$list, addrmode6dupalign32:$addr,
7979 rGPR:$Rm, pred:$p)>;
7980 def VLD4DUPdWB_register_Asm_16 :
7981 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
7982 (ins VecListFourDAllLanes:$list, addrmode6dupalign64:$addr,
7983 rGPR:$Rm, pred:$p)>;
7984 def VLD4DUPdWB_register_Asm_32 :
7985 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
7986 (ins VecListFourDAllLanes:$list,
7987 addrmode6dupalign64or128:$addr, rGPR:$Rm, pred:$p)>;
7988 def VLD4DUPqWB_register_Asm_8 :
7989 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr, $Rm",
7990 (ins VecListFourQAllLanes:$list, addrmode6dupalign32:$addr,
7991 rGPR:$Rm, pred:$p)>;
7992 def VLD4DUPqWB_register_Asm_16 :
7993 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
7994 (ins VecListFourQAllLanes:$list, addrmode6dupalign64:$addr,
7995 rGPR:$Rm, pred:$p)>;
7996 def VLD4DUPqWB_register_Asm_32 :
7997 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
7998 (ins VecListFourQAllLanes:$list,
7999 addrmode6dupalign64or128:$addr, rGPR:$Rm, pred:$p)>;
8002 // VLD4 single-lane pseudo-instructions. These need special handling for
8003 // the lane index that an InstAlias can't handle, so we use these instead.
8004 def VLD4LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr",
8005 (ins VecListFourDByteIndexed:$list, addrmode6align32:$addr,
8007 def VLD4LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
8008 (ins VecListFourDHWordIndexed:$list, addrmode6align64:$addr,
8010 def VLD4LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
8011 (ins VecListFourDWordIndexed:$list, addrmode6align64or128:$addr,
8013 def VLD4LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
8014 (ins VecListFourQHWordIndexed:$list, addrmode6align64:$addr,
8016 def VLD4LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
8017 (ins VecListFourQWordIndexed:$list, addrmode6align64or128:$addr,
8020 def VLD4LNdWB_fixed_Asm_8 :
8021 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr!",
8022 (ins VecListFourDByteIndexed:$list, addrmode6align32:$addr,
8024 def VLD4LNdWB_fixed_Asm_16 :
8025 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
8026 (ins VecListFourDHWordIndexed:$list, addrmode6align64:$addr,
8028 def VLD4LNdWB_fixed_Asm_32 :
8029 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
8030 (ins VecListFourDWordIndexed:$list, addrmode6align64or128:$addr,
8032 def VLD4LNqWB_fixed_Asm_16 :
8033 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
8034 (ins VecListFourQHWordIndexed:$list, addrmode6align64:$addr,
8036 def VLD4LNqWB_fixed_Asm_32 :
8037 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
8038 (ins VecListFourQWordIndexed:$list, addrmode6align64or128:$addr,
8040 def VLD4LNdWB_register_Asm_8 :
8041 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr, $Rm",
8042 (ins VecListFourDByteIndexed:$list, addrmode6align32:$addr,
8043 rGPR:$Rm, pred:$p)>;
8044 def VLD4LNdWB_register_Asm_16 :
8045 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
8046 (ins VecListFourDHWordIndexed:$list, addrmode6align64:$addr,
8047 rGPR:$Rm, pred:$p)>;
8048 def VLD4LNdWB_register_Asm_32 :
8049 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
8050 (ins VecListFourDWordIndexed:$list,
8051 addrmode6align64or128:$addr, rGPR:$Rm, pred:$p)>;
8052 def VLD4LNqWB_register_Asm_16 :
8053 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
8054 (ins VecListFourQHWordIndexed:$list, addrmode6align64:$addr,
8055 rGPR:$Rm, pred:$p)>;
8056 def VLD4LNqWB_register_Asm_32 :
8057 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
8058 (ins VecListFourQWordIndexed:$list,
8059 addrmode6align64or128:$addr, rGPR:$Rm, pred:$p)>;
8063 // VLD4 multiple structure pseudo-instructions. These need special handling for
8064 // the vector operands that the normal instructions don't yet model.
8065 // FIXME: Remove these when the register classes and instructions are updated.
8066 def VLD4dAsm_8 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr",
8067 (ins VecListFourD:$list, addrmode6align64or128or256:$addr,
8069 def VLD4dAsm_16 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
8070 (ins VecListFourD:$list, addrmode6align64or128or256:$addr,
8072 def VLD4dAsm_32 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
8073 (ins VecListFourD:$list, addrmode6align64or128or256:$addr,
8075 def VLD4qAsm_8 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr",
8076 (ins VecListFourQ:$list, addrmode6align64or128or256:$addr,
8078 def VLD4qAsm_16 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
8079 (ins VecListFourQ:$list, addrmode6align64or128or256:$addr,
8081 def VLD4qAsm_32 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
8082 (ins VecListFourQ:$list, addrmode6align64or128or256:$addr,
8085 def VLD4dWB_fixed_Asm_8 :
8086 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr!",
8087 (ins VecListFourD:$list, addrmode6align64or128or256:$addr,
8089 def VLD4dWB_fixed_Asm_16 :
8090 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
8091 (ins VecListFourD:$list, addrmode6align64or128or256:$addr,
8093 def VLD4dWB_fixed_Asm_32 :
8094 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
8095 (ins VecListFourD:$list, addrmode6align64or128or256:$addr,
8097 def VLD4qWB_fixed_Asm_8 :
8098 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr!",
8099 (ins VecListFourQ:$list, addrmode6align64or128or256:$addr,
8101 def VLD4qWB_fixed_Asm_16 :
8102 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
8103 (ins VecListFourQ:$list, addrmode6align64or128or256:$addr,
8105 def VLD4qWB_fixed_Asm_32 :
8106 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
8107 (ins VecListFourQ:$list, addrmode6align64or128or256:$addr,
8109 def VLD4dWB_register_Asm_8 :
8110 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr, $Rm",
8111 (ins VecListFourD:$list, addrmode6align64or128or256:$addr,
8112 rGPR:$Rm, pred:$p)>;
8113 def VLD4dWB_register_Asm_16 :
8114 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
8115 (ins VecListFourD:$list, addrmode6align64or128or256:$addr,
8116 rGPR:$Rm, pred:$p)>;
8117 def VLD4dWB_register_Asm_32 :
8118 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
8119 (ins VecListFourD:$list, addrmode6align64or128or256:$addr,
8120 rGPR:$Rm, pred:$p)>;
8121 def VLD4qWB_register_Asm_8 :
8122 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr, $Rm",
8123 (ins VecListFourQ:$list, addrmode6align64or128or256:$addr,
8124 rGPR:$Rm, pred:$p)>;
8125 def VLD4qWB_register_Asm_16 :
8126 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
8127 (ins VecListFourQ:$list, addrmode6align64or128or256:$addr,
8128 rGPR:$Rm, pred:$p)>;
8129 def VLD4qWB_register_Asm_32 :
8130 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
8131 (ins VecListFourQ:$list, addrmode6align64or128or256:$addr,
8132 rGPR:$Rm, pred:$p)>;
8134 // VST4 single-lane pseudo-instructions. These need special handling for
8135 // the lane index that an InstAlias can't handle, so we use these instead.
8136 def VST4LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr",
8137 (ins VecListFourDByteIndexed:$list, addrmode6align32:$addr,
8139 def VST4LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr",
8140 (ins VecListFourDHWordIndexed:$list, addrmode6align64:$addr,
8142 def VST4LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr",
8143 (ins VecListFourDWordIndexed:$list, addrmode6align64or128:$addr,
8145 def VST4LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr",
8146 (ins VecListFourQHWordIndexed:$list, addrmode6align64:$addr,
8148 def VST4LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr",
8149 (ins VecListFourQWordIndexed:$list, addrmode6align64or128:$addr,
8152 def VST4LNdWB_fixed_Asm_8 :
8153 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr!",
8154 (ins VecListFourDByteIndexed:$list, addrmode6align32:$addr,
8156 def VST4LNdWB_fixed_Asm_16 :
8157 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr!",
8158 (ins VecListFourDHWordIndexed:$list, addrmode6align64:$addr,
8160 def VST4LNdWB_fixed_Asm_32 :
8161 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr!",
8162 (ins VecListFourDWordIndexed:$list, addrmode6align64or128:$addr,
8164 def VST4LNqWB_fixed_Asm_16 :
8165 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr!",
8166 (ins VecListFourQHWordIndexed:$list, addrmode6align64:$addr,
8168 def VST4LNqWB_fixed_Asm_32 :
8169 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr!",
8170 (ins VecListFourQWordIndexed:$list, addrmode6align64or128:$addr,
8172 def VST4LNdWB_register_Asm_8 :
8173 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr, $Rm",
8174 (ins VecListFourDByteIndexed:$list, addrmode6align32:$addr,
8175 rGPR:$Rm, pred:$p)>;
8176 def VST4LNdWB_register_Asm_16 :
8177 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr, $Rm",
8178 (ins VecListFourDHWordIndexed:$list, addrmode6align64:$addr,
8179 rGPR:$Rm, pred:$p)>;
8180 def VST4LNdWB_register_Asm_32 :
8181 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr, $Rm",
8182 (ins VecListFourDWordIndexed:$list,
8183 addrmode6align64or128:$addr, rGPR:$Rm, pred:$p)>;
8184 def VST4LNqWB_register_Asm_16 :
8185 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr, $Rm",
8186 (ins VecListFourQHWordIndexed:$list, addrmode6align64:$addr,
8187 rGPR:$Rm, pred:$p)>;
8188 def VST4LNqWB_register_Asm_32 :
8189 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr, $Rm",
8190 (ins VecListFourQWordIndexed:$list,
8191 addrmode6align64or128:$addr, rGPR:$Rm, pred:$p)>;
8194 // VST4 multiple structure pseudo-instructions. These need special handling for
8195 // the vector operands that the normal instructions don't yet model.
8196 // FIXME: Remove these when the register classes and instructions are updated.
8197 def VST4dAsm_8 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr",
8198 (ins VecListFourD:$list, addrmode6align64or128or256:$addr,
8200 def VST4dAsm_16 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr",
8201 (ins VecListFourD:$list, addrmode6align64or128or256:$addr,
8203 def VST4dAsm_32 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr",
8204 (ins VecListFourD:$list, addrmode6align64or128or256:$addr,
8206 def VST4qAsm_8 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr",
8207 (ins VecListFourQ:$list, addrmode6align64or128or256:$addr,
8209 def VST4qAsm_16 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr",
8210 (ins VecListFourQ:$list, addrmode6align64or128or256:$addr,
8212 def VST4qAsm_32 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr",
8213 (ins VecListFourQ:$list, addrmode6align64or128or256:$addr,
8216 def VST4dWB_fixed_Asm_8 :
8217 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr!",
8218 (ins VecListFourD:$list, addrmode6align64or128or256:$addr,
8220 def VST4dWB_fixed_Asm_16 :
8221 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr!",
8222 (ins VecListFourD:$list, addrmode6align64or128or256:$addr,
8224 def VST4dWB_fixed_Asm_32 :
8225 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr!",
8226 (ins VecListFourD:$list, addrmode6align64or128or256:$addr,
8228 def VST4qWB_fixed_Asm_8 :
8229 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr!",
8230 (ins VecListFourQ:$list, addrmode6align64or128or256:$addr,
8232 def VST4qWB_fixed_Asm_16 :
8233 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr!",
8234 (ins VecListFourQ:$list, addrmode6align64or128or256:$addr,
8236 def VST4qWB_fixed_Asm_32 :
8237 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr!",
8238 (ins VecListFourQ:$list, addrmode6align64or128or256:$addr,
8240 def VST4dWB_register_Asm_8 :
8241 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr, $Rm",
8242 (ins VecListFourD:$list, addrmode6align64or128or256:$addr,
8243 rGPR:$Rm, pred:$p)>;
8244 def VST4dWB_register_Asm_16 :
8245 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr, $Rm",
8246 (ins VecListFourD:$list, addrmode6align64or128or256:$addr,
8247 rGPR:$Rm, pred:$p)>;
8248 def VST4dWB_register_Asm_32 :
8249 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr, $Rm",
8250 (ins VecListFourD:$list, addrmode6align64or128or256:$addr,
8251 rGPR:$Rm, pred:$p)>;
8252 def VST4qWB_register_Asm_8 :
8253 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr, $Rm",
8254 (ins VecListFourQ:$list, addrmode6align64or128or256:$addr,
8255 rGPR:$Rm, pred:$p)>;
8256 def VST4qWB_register_Asm_16 :
8257 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr, $Rm",
8258 (ins VecListFourQ:$list, addrmode6align64or128or256:$addr,
8259 rGPR:$Rm, pred:$p)>;
8260 def VST4qWB_register_Asm_32 :
8261 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr, $Rm",
8262 (ins VecListFourQ:$list, addrmode6align64or128or256:$addr,
8263 rGPR:$Rm, pred:$p)>;
8265 // VMOV/VMVN takes an optional datatype suffix
8266 defm : NEONDTAnyInstAlias<"vmov${p}", "$Vd, $Vm",
8267 (VORRd DPR:$Vd, DPR:$Vm, DPR:$Vm, pred:$p)>;
8268 defm : NEONDTAnyInstAlias<"vmov${p}", "$Vd, $Vm",
8269 (VORRq QPR:$Vd, QPR:$Vm, QPR:$Vm, pred:$p)>;
8271 defm : NEONDTAnyInstAlias<"vmvn${p}", "$Vd, $Vm",
8272 (VMVNd DPR:$Vd, DPR:$Vm, pred:$p)>;
8273 defm : NEONDTAnyInstAlias<"vmvn${p}", "$Vd, $Vm",
8274 (VMVNq QPR:$Vd, QPR:$Vm, pred:$p)>;
8276 // VCLT (register) is an assembler alias for VCGT w/ the operands reversed.
8277 // D-register versions.
8278 def : NEONInstAlias<"vcle${p}.s8 $Dd, $Dn, $Dm",
8279 (VCGEsv8i8 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
8280 def : NEONInstAlias<"vcle${p}.s16 $Dd, $Dn, $Dm",
8281 (VCGEsv4i16 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
8282 def : NEONInstAlias<"vcle${p}.s32 $Dd, $Dn, $Dm",
8283 (VCGEsv2i32 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
8284 def : NEONInstAlias<"vcle${p}.u8 $Dd, $Dn, $Dm",
8285 (VCGEuv8i8 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
8286 def : NEONInstAlias<"vcle${p}.u16 $Dd, $Dn, $Dm",
8287 (VCGEuv4i16 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
8288 def : NEONInstAlias<"vcle${p}.u32 $Dd, $Dn, $Dm",
8289 (VCGEuv2i32 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
8290 def : NEONInstAlias<"vcle${p}.f32 $Dd, $Dn, $Dm",
8291 (VCGEfd DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
8292 let Predicates = [HasNEON, HasFullFP16] in
8293 def : NEONInstAlias<"vcle${p}.f16 $Dd, $Dn, $Dm",
8294 (VCGEhd DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
8295 // Q-register versions.
8296 def : NEONInstAlias<"vcle${p}.s8 $Qd, $Qn, $Qm",
8297 (VCGEsv16i8 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
8298 def : NEONInstAlias<"vcle${p}.s16 $Qd, $Qn, $Qm",
8299 (VCGEsv8i16 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
8300 def : NEONInstAlias<"vcle${p}.s32 $Qd, $Qn, $Qm",
8301 (VCGEsv4i32 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
8302 def : NEONInstAlias<"vcle${p}.u8 $Qd, $Qn, $Qm",
8303 (VCGEuv16i8 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
8304 def : NEONInstAlias<"vcle${p}.u16 $Qd, $Qn, $Qm",
8305 (VCGEuv8i16 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
8306 def : NEONInstAlias<"vcle${p}.u32 $Qd, $Qn, $Qm",
8307 (VCGEuv4i32 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
8308 def : NEONInstAlias<"vcle${p}.f32 $Qd, $Qn, $Qm",
8309 (VCGEfq QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
8310 let Predicates = [HasNEON, HasFullFP16] in
8311 def : NEONInstAlias<"vcle${p}.f16 $Qd, $Qn, $Qm",
8312 (VCGEhq QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
8314 // VCLT (register) is an assembler alias for VCGT w/ the operands reversed.
8315 // D-register versions.
8316 def : NEONInstAlias<"vclt${p}.s8 $Dd, $Dn, $Dm",
8317 (VCGTsv8i8 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
8318 def : NEONInstAlias<"vclt${p}.s16 $Dd, $Dn, $Dm",
8319 (VCGTsv4i16 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
8320 def : NEONInstAlias<"vclt${p}.s32 $Dd, $Dn, $Dm",
8321 (VCGTsv2i32 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
8322 def : NEONInstAlias<"vclt${p}.u8 $Dd, $Dn, $Dm",
8323 (VCGTuv8i8 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
8324 def : NEONInstAlias<"vclt${p}.u16 $Dd, $Dn, $Dm",
8325 (VCGTuv4i16 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
8326 def : NEONInstAlias<"vclt${p}.u32 $Dd, $Dn, $Dm",
8327 (VCGTuv2i32 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
8328 def : NEONInstAlias<"vclt${p}.f32 $Dd, $Dn, $Dm",
8329 (VCGTfd DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
8330 let Predicates = [HasNEON, HasFullFP16] in
8331 def : NEONInstAlias<"vclt${p}.f16 $Dd, $Dn, $Dm",
8332 (VCGThd DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
8333 // Q-register versions.
8334 def : NEONInstAlias<"vclt${p}.s8 $Qd, $Qn, $Qm",
8335 (VCGTsv16i8 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
8336 def : NEONInstAlias<"vclt${p}.s16 $Qd, $Qn, $Qm",
8337 (VCGTsv8i16 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
8338 def : NEONInstAlias<"vclt${p}.s32 $Qd, $Qn, $Qm",
8339 (VCGTsv4i32 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
8340 def : NEONInstAlias<"vclt${p}.u8 $Qd, $Qn, $Qm",
8341 (VCGTuv16i8 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
8342 def : NEONInstAlias<"vclt${p}.u16 $Qd, $Qn, $Qm",
8343 (VCGTuv8i16 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
8344 def : NEONInstAlias<"vclt${p}.u32 $Qd, $Qn, $Qm",
8345 (VCGTuv4i32 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
8346 def : NEONInstAlias<"vclt${p}.f32 $Qd, $Qn, $Qm",
8347 (VCGTfq QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
8348 let Predicates = [HasNEON, HasFullFP16] in
8349 def : NEONInstAlias<"vclt${p}.f16 $Qd, $Qn, $Qm",
8350 (VCGThq QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
8352 // VSWP allows, but does not require, a type suffix.
8353 defm : NEONDTAnyInstAlias<"vswp${p}", "$Vd, $Vm",
8354 (VSWPd DPR:$Vd, DPR:$Vm, pred:$p)>;
8355 defm : NEONDTAnyInstAlias<"vswp${p}", "$Vd, $Vm",
8356 (VSWPq QPR:$Vd, QPR:$Vm, pred:$p)>;
8358 // VBIF, VBIT, and VBSL allow, but do not require, a type suffix.
8359 defm : NEONDTAnyInstAlias<"vbif${p}", "$Vd, $Vn, $Vm",
8360 (VBIFd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
8361 defm : NEONDTAnyInstAlias<"vbit${p}", "$Vd, $Vn, $Vm",
8362 (VBITd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
8363 defm : NEONDTAnyInstAlias<"vbsl${p}", "$Vd, $Vn, $Vm",
8364 (VBSLd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
8365 defm : NEONDTAnyInstAlias<"vbif${p}", "$Vd, $Vn, $Vm",
8366 (VBIFq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
8367 defm : NEONDTAnyInstAlias<"vbit${p}", "$Vd, $Vn, $Vm",
8368 (VBITq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
8369 defm : NEONDTAnyInstAlias<"vbsl${p}", "$Vd, $Vn, $Vm",
8370 (VBSLq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
8372 // "vmov Rd, #-imm" can be handled via "vmvn".
8373 def : NEONInstAlias<"vmov${p}.i32 $Vd, $imm",
8374 (VMVNv2i32 DPR:$Vd, nImmVMOVI32Neg:$imm, pred:$p)>;
8375 def : NEONInstAlias<"vmov${p}.i32 $Vd, $imm",
8376 (VMVNv4i32 QPR:$Vd, nImmVMOVI32Neg:$imm, pred:$p)>;
8377 def : NEONInstAlias<"vmvn${p}.i32 $Vd, $imm",
8378 (VMOVv2i32 DPR:$Vd, nImmVMOVI32Neg:$imm, pred:$p)>;
8379 def : NEONInstAlias<"vmvn${p}.i32 $Vd, $imm",
8380 (VMOVv4i32 QPR:$Vd, nImmVMOVI32Neg:$imm, pred:$p)>;
8382 // 'gas' compatibility aliases for quad-word instructions. Strictly speaking,
8383 // these should restrict to just the Q register variants, but the register
8384 // classes are enough to match correctly regardless, so we keep it simple
8385 // and just use MnemonicAlias.
8386 def : NEONMnemonicAlias<"vbicq", "vbic">;
8387 def : NEONMnemonicAlias<"vandq", "vand">;
8388 def : NEONMnemonicAlias<"veorq", "veor">;
8389 def : NEONMnemonicAlias<"vorrq", "vorr">;
8391 def : NEONMnemonicAlias<"vmovq", "vmov">;
8392 def : NEONMnemonicAlias<"vmvnq", "vmvn">;
8393 // Explicit versions for floating point so that the FPImm variants get
8394 // handled early. The parser gets confused otherwise.
8395 def : NEONMnemonicAlias<"vmovq.f32", "vmov.f32">;
8396 def : NEONMnemonicAlias<"vmovq.f64", "vmov.f64">;
8398 def : NEONMnemonicAlias<"vaddq", "vadd">;
8399 def : NEONMnemonicAlias<"vsubq", "vsub">;
8401 def : NEONMnemonicAlias<"vminq", "vmin">;
8402 def : NEONMnemonicAlias<"vmaxq", "vmax">;
8404 def : NEONMnemonicAlias<"vmulq", "vmul">;
8406 def : NEONMnemonicAlias<"vabsq", "vabs">;
8408 def : NEONMnemonicAlias<"vshlq", "vshl">;
8409 def : NEONMnemonicAlias<"vshrq", "vshr">;
8411 def : NEONMnemonicAlias<"vcvtq", "vcvt">;
8413 def : NEONMnemonicAlias<"vcleq", "vcle">;
8414 def : NEONMnemonicAlias<"vceqq", "vceq">;
8416 def : NEONMnemonicAlias<"vzipq", "vzip">;
8417 def : NEONMnemonicAlias<"vswpq", "vswp">;
8419 def : NEONMnemonicAlias<"vrecpeq.f32", "vrecpe.f32">;
8420 def : NEONMnemonicAlias<"vrecpeq.u32", "vrecpe.u32">;
8423 // Alias for loading floating point immediates that aren't representable
8424 // using the vmov.f32 encoding but the bitpattern is representable using
8425 // the .i32 encoding.
8426 def : NEONInstAlias<"vmov${p}.f32 $Vd, $imm",
8427 (VMOVv4i32 QPR:$Vd, nImmVMOVI32:$imm, pred:$p)>;
8428 def : NEONInstAlias<"vmov${p}.f32 $Vd, $imm",
8429 (VMOVv2i32 DPR:$Vd, nImmVMOVI32:$imm, pred:$p)>;