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 nImmVMOVI32AsmOperand : AsmOperandClass { let Name = "NEONi32vmov"; }
38 def nImmVMOVI32 : Operand<i32> {
39 let PrintMethod = "printNEONModImmOperand";
40 let ParserMatchClass = nImmVMOVI32AsmOperand;
42 def nImmVMOVI32NegAsmOperand : AsmOperandClass { let Name = "NEONi32vmovNeg"; }
43 def nImmVMOVI32Neg : Operand<i32> {
44 let PrintMethod = "printNEONModImmOperand";
45 let ParserMatchClass = nImmVMOVI32NegAsmOperand;
47 def nImmVMOVF32 : Operand<i32> {
48 let PrintMethod = "printFPImmOperand";
49 let ParserMatchClass = FPImmOperand;
51 def nImmSplatI64AsmOperand : AsmOperandClass { let Name = "NEONi64splat"; }
52 def nImmSplatI64 : Operand<i32> {
53 let PrintMethod = "printNEONModImmOperand";
54 let ParserMatchClass = nImmSplatI64AsmOperand;
57 def VectorIndex8Operand : AsmOperandClass { let Name = "VectorIndex8"; }
58 def VectorIndex16Operand : AsmOperandClass { let Name = "VectorIndex16"; }
59 def VectorIndex32Operand : AsmOperandClass { let Name = "VectorIndex32"; }
60 def VectorIndex8 : Operand<i32>, ImmLeaf<i32, [{
61 return ((uint64_t)Imm) < 8;
63 let ParserMatchClass = VectorIndex8Operand;
64 let PrintMethod = "printVectorIndex";
65 let MIOperandInfo = (ops i32imm);
67 def VectorIndex16 : Operand<i32>, ImmLeaf<i32, [{
68 return ((uint64_t)Imm) < 4;
70 let ParserMatchClass = VectorIndex16Operand;
71 let PrintMethod = "printVectorIndex";
72 let MIOperandInfo = (ops i32imm);
74 def VectorIndex32 : Operand<i32>, ImmLeaf<i32, [{
75 return ((uint64_t)Imm) < 2;
77 let ParserMatchClass = VectorIndex32Operand;
78 let PrintMethod = "printVectorIndex";
79 let MIOperandInfo = (ops i32imm);
82 // Register list of one D register.
83 def VecListOneDAsmOperand : AsmOperandClass {
84 let Name = "VecListOneD";
85 let ParserMethod = "parseVectorList";
86 let RenderMethod = "addVecListOperands";
88 def VecListOneD : RegisterOperand<DPR, "printVectorListOne"> {
89 let ParserMatchClass = VecListOneDAsmOperand;
91 // Register list of two sequential D registers.
92 def VecListDPairAsmOperand : AsmOperandClass {
93 let Name = "VecListDPair";
94 let ParserMethod = "parseVectorList";
95 let RenderMethod = "addVecListOperands";
97 def VecListDPair : RegisterOperand<DPair, "printVectorListTwo"> {
98 let ParserMatchClass = VecListDPairAsmOperand;
100 // Register list of three sequential D registers.
101 def VecListThreeDAsmOperand : AsmOperandClass {
102 let Name = "VecListThreeD";
103 let ParserMethod = "parseVectorList";
104 let RenderMethod = "addVecListOperands";
106 def VecListThreeD : RegisterOperand<DPR, "printVectorListThree"> {
107 let ParserMatchClass = VecListThreeDAsmOperand;
109 // Register list of four sequential D registers.
110 def VecListFourDAsmOperand : AsmOperandClass {
111 let Name = "VecListFourD";
112 let ParserMethod = "parseVectorList";
113 let RenderMethod = "addVecListOperands";
115 def VecListFourD : RegisterOperand<DPR, "printVectorListFour"> {
116 let ParserMatchClass = VecListFourDAsmOperand;
118 // Register list of two D registers spaced by 2 (two sequential Q registers).
119 def VecListDPairSpacedAsmOperand : AsmOperandClass {
120 let Name = "VecListDPairSpaced";
121 let ParserMethod = "parseVectorList";
122 let RenderMethod = "addVecListOperands";
124 def VecListDPairSpaced : RegisterOperand<DPair, "printVectorListTwoSpaced"> {
125 let ParserMatchClass = VecListDPairSpacedAsmOperand;
127 // Register list of three D registers spaced by 2 (three Q registers).
128 def VecListThreeQAsmOperand : AsmOperandClass {
129 let Name = "VecListThreeQ";
130 let ParserMethod = "parseVectorList";
131 let RenderMethod = "addVecListOperands";
133 def VecListThreeQ : RegisterOperand<DPR, "printVectorListThreeSpaced"> {
134 let ParserMatchClass = VecListThreeQAsmOperand;
136 // Register list of three D registers spaced by 2 (three Q registers).
137 def VecListFourQAsmOperand : AsmOperandClass {
138 let Name = "VecListFourQ";
139 let ParserMethod = "parseVectorList";
140 let RenderMethod = "addVecListOperands";
142 def VecListFourQ : RegisterOperand<DPR, "printVectorListFourSpaced"> {
143 let ParserMatchClass = VecListFourQAsmOperand;
146 // Register list of one D register, with "all lanes" subscripting.
147 def VecListOneDAllLanesAsmOperand : AsmOperandClass {
148 let Name = "VecListOneDAllLanes";
149 let ParserMethod = "parseVectorList";
150 let RenderMethod = "addVecListOperands";
152 def VecListOneDAllLanes : RegisterOperand<DPR, "printVectorListOneAllLanes"> {
153 let ParserMatchClass = VecListOneDAllLanesAsmOperand;
155 // Register list of two D registers, with "all lanes" subscripting.
156 def VecListDPairAllLanesAsmOperand : AsmOperandClass {
157 let Name = "VecListDPairAllLanes";
158 let ParserMethod = "parseVectorList";
159 let RenderMethod = "addVecListOperands";
161 def VecListDPairAllLanes : RegisterOperand<DPair,
162 "printVectorListTwoAllLanes"> {
163 let ParserMatchClass = VecListDPairAllLanesAsmOperand;
165 // Register list of two D registers spaced by 2 (two sequential Q registers).
166 def VecListDPairSpacedAllLanesAsmOperand : AsmOperandClass {
167 let Name = "VecListDPairSpacedAllLanes";
168 let ParserMethod = "parseVectorList";
169 let RenderMethod = "addVecListOperands";
171 def VecListDPairSpacedAllLanes : RegisterOperand<DPair,
172 "printVectorListTwoSpacedAllLanes"> {
173 let ParserMatchClass = VecListDPairSpacedAllLanesAsmOperand;
175 // Register list of three D registers, with "all lanes" subscripting.
176 def VecListThreeDAllLanesAsmOperand : AsmOperandClass {
177 let Name = "VecListThreeDAllLanes";
178 let ParserMethod = "parseVectorList";
179 let RenderMethod = "addVecListOperands";
181 def VecListThreeDAllLanes : RegisterOperand<DPR,
182 "printVectorListThreeAllLanes"> {
183 let ParserMatchClass = VecListThreeDAllLanesAsmOperand;
185 // Register list of three D registers spaced by 2 (three sequential Q regs).
186 def VecListThreeQAllLanesAsmOperand : AsmOperandClass {
187 let Name = "VecListThreeQAllLanes";
188 let ParserMethod = "parseVectorList";
189 let RenderMethod = "addVecListOperands";
191 def VecListThreeQAllLanes : RegisterOperand<DPR,
192 "printVectorListThreeSpacedAllLanes"> {
193 let ParserMatchClass = VecListThreeQAllLanesAsmOperand;
195 // Register list of four D registers, with "all lanes" subscripting.
196 def VecListFourDAllLanesAsmOperand : AsmOperandClass {
197 let Name = "VecListFourDAllLanes";
198 let ParserMethod = "parseVectorList";
199 let RenderMethod = "addVecListOperands";
201 def VecListFourDAllLanes : RegisterOperand<DPR, "printVectorListFourAllLanes"> {
202 let ParserMatchClass = VecListFourDAllLanesAsmOperand;
204 // Register list of four D registers spaced by 2 (four sequential Q regs).
205 def VecListFourQAllLanesAsmOperand : AsmOperandClass {
206 let Name = "VecListFourQAllLanes";
207 let ParserMethod = "parseVectorList";
208 let RenderMethod = "addVecListOperands";
210 def VecListFourQAllLanes : RegisterOperand<DPR,
211 "printVectorListFourSpacedAllLanes"> {
212 let ParserMatchClass = VecListFourQAllLanesAsmOperand;
216 // Register list of one D register, with byte lane subscripting.
217 def VecListOneDByteIndexAsmOperand : AsmOperandClass {
218 let Name = "VecListOneDByteIndexed";
219 let ParserMethod = "parseVectorList";
220 let RenderMethod = "addVecListIndexedOperands";
222 def VecListOneDByteIndexed : Operand<i32> {
223 let ParserMatchClass = VecListOneDByteIndexAsmOperand;
224 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
226 // ...with half-word lane subscripting.
227 def VecListOneDHWordIndexAsmOperand : AsmOperandClass {
228 let Name = "VecListOneDHWordIndexed";
229 let ParserMethod = "parseVectorList";
230 let RenderMethod = "addVecListIndexedOperands";
232 def VecListOneDHWordIndexed : Operand<i32> {
233 let ParserMatchClass = VecListOneDHWordIndexAsmOperand;
234 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
236 // ...with word lane subscripting.
237 def VecListOneDWordIndexAsmOperand : AsmOperandClass {
238 let Name = "VecListOneDWordIndexed";
239 let ParserMethod = "parseVectorList";
240 let RenderMethod = "addVecListIndexedOperands";
242 def VecListOneDWordIndexed : Operand<i32> {
243 let ParserMatchClass = VecListOneDWordIndexAsmOperand;
244 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
247 // Register list of two D registers with byte lane subscripting.
248 def VecListTwoDByteIndexAsmOperand : AsmOperandClass {
249 let Name = "VecListTwoDByteIndexed";
250 let ParserMethod = "parseVectorList";
251 let RenderMethod = "addVecListIndexedOperands";
253 def VecListTwoDByteIndexed : Operand<i32> {
254 let ParserMatchClass = VecListTwoDByteIndexAsmOperand;
255 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
257 // ...with half-word lane subscripting.
258 def VecListTwoDHWordIndexAsmOperand : AsmOperandClass {
259 let Name = "VecListTwoDHWordIndexed";
260 let ParserMethod = "parseVectorList";
261 let RenderMethod = "addVecListIndexedOperands";
263 def VecListTwoDHWordIndexed : Operand<i32> {
264 let ParserMatchClass = VecListTwoDHWordIndexAsmOperand;
265 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
267 // ...with word lane subscripting.
268 def VecListTwoDWordIndexAsmOperand : AsmOperandClass {
269 let Name = "VecListTwoDWordIndexed";
270 let ParserMethod = "parseVectorList";
271 let RenderMethod = "addVecListIndexedOperands";
273 def VecListTwoDWordIndexed : Operand<i32> {
274 let ParserMatchClass = VecListTwoDWordIndexAsmOperand;
275 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
277 // Register list of two Q registers with half-word lane subscripting.
278 def VecListTwoQHWordIndexAsmOperand : AsmOperandClass {
279 let Name = "VecListTwoQHWordIndexed";
280 let ParserMethod = "parseVectorList";
281 let RenderMethod = "addVecListIndexedOperands";
283 def VecListTwoQHWordIndexed : Operand<i32> {
284 let ParserMatchClass = VecListTwoQHWordIndexAsmOperand;
285 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
287 // ...with word lane subscripting.
288 def VecListTwoQWordIndexAsmOperand : AsmOperandClass {
289 let Name = "VecListTwoQWordIndexed";
290 let ParserMethod = "parseVectorList";
291 let RenderMethod = "addVecListIndexedOperands";
293 def VecListTwoQWordIndexed : Operand<i32> {
294 let ParserMatchClass = VecListTwoQWordIndexAsmOperand;
295 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
299 // Register list of three D registers with byte lane subscripting.
300 def VecListThreeDByteIndexAsmOperand : AsmOperandClass {
301 let Name = "VecListThreeDByteIndexed";
302 let ParserMethod = "parseVectorList";
303 let RenderMethod = "addVecListIndexedOperands";
305 def VecListThreeDByteIndexed : Operand<i32> {
306 let ParserMatchClass = VecListThreeDByteIndexAsmOperand;
307 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
309 // ...with half-word lane subscripting.
310 def VecListThreeDHWordIndexAsmOperand : AsmOperandClass {
311 let Name = "VecListThreeDHWordIndexed";
312 let ParserMethod = "parseVectorList";
313 let RenderMethod = "addVecListIndexedOperands";
315 def VecListThreeDHWordIndexed : Operand<i32> {
316 let ParserMatchClass = VecListThreeDHWordIndexAsmOperand;
317 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
319 // ...with word lane subscripting.
320 def VecListThreeDWordIndexAsmOperand : AsmOperandClass {
321 let Name = "VecListThreeDWordIndexed";
322 let ParserMethod = "parseVectorList";
323 let RenderMethod = "addVecListIndexedOperands";
325 def VecListThreeDWordIndexed : Operand<i32> {
326 let ParserMatchClass = VecListThreeDWordIndexAsmOperand;
327 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
329 // Register list of three Q registers with half-word lane subscripting.
330 def VecListThreeQHWordIndexAsmOperand : AsmOperandClass {
331 let Name = "VecListThreeQHWordIndexed";
332 let ParserMethod = "parseVectorList";
333 let RenderMethod = "addVecListIndexedOperands";
335 def VecListThreeQHWordIndexed : Operand<i32> {
336 let ParserMatchClass = VecListThreeQHWordIndexAsmOperand;
337 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
339 // ...with word lane subscripting.
340 def VecListThreeQWordIndexAsmOperand : AsmOperandClass {
341 let Name = "VecListThreeQWordIndexed";
342 let ParserMethod = "parseVectorList";
343 let RenderMethod = "addVecListIndexedOperands";
345 def VecListThreeQWordIndexed : Operand<i32> {
346 let ParserMatchClass = VecListThreeQWordIndexAsmOperand;
347 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
350 // Register list of four D registers with byte lane subscripting.
351 def VecListFourDByteIndexAsmOperand : AsmOperandClass {
352 let Name = "VecListFourDByteIndexed";
353 let ParserMethod = "parseVectorList";
354 let RenderMethod = "addVecListIndexedOperands";
356 def VecListFourDByteIndexed : Operand<i32> {
357 let ParserMatchClass = VecListFourDByteIndexAsmOperand;
358 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
360 // ...with half-word lane subscripting.
361 def VecListFourDHWordIndexAsmOperand : AsmOperandClass {
362 let Name = "VecListFourDHWordIndexed";
363 let ParserMethod = "parseVectorList";
364 let RenderMethod = "addVecListIndexedOperands";
366 def VecListFourDHWordIndexed : Operand<i32> {
367 let ParserMatchClass = VecListFourDHWordIndexAsmOperand;
368 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
370 // ...with word lane subscripting.
371 def VecListFourDWordIndexAsmOperand : AsmOperandClass {
372 let Name = "VecListFourDWordIndexed";
373 let ParserMethod = "parseVectorList";
374 let RenderMethod = "addVecListIndexedOperands";
376 def VecListFourDWordIndexed : Operand<i32> {
377 let ParserMatchClass = VecListFourDWordIndexAsmOperand;
378 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
380 // Register list of four Q registers with half-word lane subscripting.
381 def VecListFourQHWordIndexAsmOperand : AsmOperandClass {
382 let Name = "VecListFourQHWordIndexed";
383 let ParserMethod = "parseVectorList";
384 let RenderMethod = "addVecListIndexedOperands";
386 def VecListFourQHWordIndexed : Operand<i32> {
387 let ParserMatchClass = VecListFourQHWordIndexAsmOperand;
388 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
390 // ...with word lane subscripting.
391 def VecListFourQWordIndexAsmOperand : AsmOperandClass {
392 let Name = "VecListFourQWordIndexed";
393 let ParserMethod = "parseVectorList";
394 let RenderMethod = "addVecListIndexedOperands";
396 def VecListFourQWordIndexed : Operand<i32> {
397 let ParserMatchClass = VecListFourQWordIndexAsmOperand;
398 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
401 def dword_alignedload : PatFrag<(ops node:$ptr), (load node:$ptr), [{
402 return cast<LoadSDNode>(N)->getAlignment() >= 8;
404 def dword_alignedstore : PatFrag<(ops node:$val, node:$ptr),
405 (store node:$val, node:$ptr), [{
406 return cast<StoreSDNode>(N)->getAlignment() >= 8;
408 def word_alignedload : PatFrag<(ops node:$ptr), (load node:$ptr), [{
409 return cast<LoadSDNode>(N)->getAlignment() == 4;
411 def word_alignedstore : PatFrag<(ops node:$val, node:$ptr),
412 (store node:$val, node:$ptr), [{
413 return cast<StoreSDNode>(N)->getAlignment() == 4;
415 def hword_alignedload : PatFrag<(ops node:$ptr), (load node:$ptr), [{
416 return cast<LoadSDNode>(N)->getAlignment() == 2;
418 def hword_alignedstore : PatFrag<(ops node:$val, node:$ptr),
419 (store node:$val, node:$ptr), [{
420 return cast<StoreSDNode>(N)->getAlignment() == 2;
422 def byte_alignedload : PatFrag<(ops node:$ptr), (load node:$ptr), [{
423 return cast<LoadSDNode>(N)->getAlignment() == 1;
425 def byte_alignedstore : PatFrag<(ops node:$val, node:$ptr),
426 (store node:$val, node:$ptr), [{
427 return cast<StoreSDNode>(N)->getAlignment() == 1;
429 def non_word_alignedload : PatFrag<(ops node:$ptr), (load node:$ptr), [{
430 return cast<LoadSDNode>(N)->getAlignment() < 4;
432 def non_word_alignedstore : PatFrag<(ops node:$val, node:$ptr),
433 (store node:$val, node:$ptr), [{
434 return cast<StoreSDNode>(N)->getAlignment() < 4;
437 //===----------------------------------------------------------------------===//
438 // NEON-specific DAG Nodes.
439 //===----------------------------------------------------------------------===//
441 def SDTARMVCMP : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<1, 2>]>;
442 def SDTARMVCMPZ : SDTypeProfile<1, 1, []>;
444 def NEONvceq : SDNode<"ARMISD::VCEQ", SDTARMVCMP>;
445 def NEONvceqz : SDNode<"ARMISD::VCEQZ", SDTARMVCMPZ>;
446 def NEONvcge : SDNode<"ARMISD::VCGE", SDTARMVCMP>;
447 def NEONvcgez : SDNode<"ARMISD::VCGEZ", SDTARMVCMPZ>;
448 def NEONvclez : SDNode<"ARMISD::VCLEZ", SDTARMVCMPZ>;
449 def NEONvcgeu : SDNode<"ARMISD::VCGEU", SDTARMVCMP>;
450 def NEONvcgt : SDNode<"ARMISD::VCGT", SDTARMVCMP>;
451 def NEONvcgtz : SDNode<"ARMISD::VCGTZ", SDTARMVCMPZ>;
452 def NEONvcltz : SDNode<"ARMISD::VCLTZ", SDTARMVCMPZ>;
453 def NEONvcgtu : SDNode<"ARMISD::VCGTU", SDTARMVCMP>;
454 def NEONvtst : SDNode<"ARMISD::VTST", SDTARMVCMP>;
456 // Types for vector shift by immediates. The "SHX" version is for long and
457 // narrow operations where the source and destination vectors have different
458 // types. The "SHINS" version is for shift and insert operations.
459 def SDTARMVSH : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
461 def SDTARMVSHX : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
463 def SDTARMVSHINS : SDTypeProfile<1, 3, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
464 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
466 def NEONvshl : SDNode<"ARMISD::VSHL", SDTARMVSH>;
467 def NEONvshrs : SDNode<"ARMISD::VSHRs", SDTARMVSH>;
468 def NEONvshru : SDNode<"ARMISD::VSHRu", SDTARMVSH>;
469 def NEONvshlls : SDNode<"ARMISD::VSHLLs", SDTARMVSHX>;
470 def NEONvshllu : SDNode<"ARMISD::VSHLLu", SDTARMVSHX>;
471 def NEONvshlli : SDNode<"ARMISD::VSHLLi", SDTARMVSHX>;
472 def NEONvshrn : SDNode<"ARMISD::VSHRN", SDTARMVSHX>;
474 def NEONvrshrs : SDNode<"ARMISD::VRSHRs", SDTARMVSH>;
475 def NEONvrshru : SDNode<"ARMISD::VRSHRu", SDTARMVSH>;
476 def NEONvrshrn : SDNode<"ARMISD::VRSHRN", SDTARMVSHX>;
478 def NEONvqshls : SDNode<"ARMISD::VQSHLs", SDTARMVSH>;
479 def NEONvqshlu : SDNode<"ARMISD::VQSHLu", SDTARMVSH>;
480 def NEONvqshlsu : SDNode<"ARMISD::VQSHLsu", SDTARMVSH>;
481 def NEONvqshrns : SDNode<"ARMISD::VQSHRNs", SDTARMVSHX>;
482 def NEONvqshrnu : SDNode<"ARMISD::VQSHRNu", SDTARMVSHX>;
483 def NEONvqshrnsu : SDNode<"ARMISD::VQSHRNsu", SDTARMVSHX>;
485 def NEONvqrshrns : SDNode<"ARMISD::VQRSHRNs", SDTARMVSHX>;
486 def NEONvqrshrnu : SDNode<"ARMISD::VQRSHRNu", SDTARMVSHX>;
487 def NEONvqrshrnsu : SDNode<"ARMISD::VQRSHRNsu", SDTARMVSHX>;
489 def NEONvsli : SDNode<"ARMISD::VSLI", SDTARMVSHINS>;
490 def NEONvsri : SDNode<"ARMISD::VSRI", SDTARMVSHINS>;
492 def SDTARMVGETLN : SDTypeProfile<1, 2, [SDTCisVT<0, i32>, SDTCisInt<1>,
494 def NEONvgetlaneu : SDNode<"ARMISD::VGETLANEu", SDTARMVGETLN>;
495 def NEONvgetlanes : SDNode<"ARMISD::VGETLANEs", SDTARMVGETLN>;
497 def SDTARMVMOVIMM : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisVT<1, i32>]>;
498 def NEONvmovImm : SDNode<"ARMISD::VMOVIMM", SDTARMVMOVIMM>;
499 def NEONvmvnImm : SDNode<"ARMISD::VMVNIMM", SDTARMVMOVIMM>;
500 def NEONvmovFPImm : SDNode<"ARMISD::VMOVFPIMM", SDTARMVMOVIMM>;
502 def SDTARMVORRIMM : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
504 def NEONvorrImm : SDNode<"ARMISD::VORRIMM", SDTARMVORRIMM>;
505 def NEONvbicImm : SDNode<"ARMISD::VBICIMM", SDTARMVORRIMM>;
507 def NEONvbsl : SDNode<"ARMISD::VBSL",
508 SDTypeProfile<1, 3, [SDTCisVec<0>,
511 SDTCisSameAs<0, 3>]>>;
513 def NEONvdup : SDNode<"ARMISD::VDUP", SDTypeProfile<1, 1, [SDTCisVec<0>]>>;
515 // VDUPLANE can produce a quad-register result from a double-register source,
516 // so the result is not constrained to match the source.
517 def NEONvduplane : SDNode<"ARMISD::VDUPLANE",
518 SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisVec<1>,
521 def SDTARMVEXT : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
522 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
523 def NEONvext : SDNode<"ARMISD::VEXT", SDTARMVEXT>;
525 def SDTARMVSHUF : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0, 1>]>;
526 def NEONvrev64 : SDNode<"ARMISD::VREV64", SDTARMVSHUF>;
527 def NEONvrev32 : SDNode<"ARMISD::VREV32", SDTARMVSHUF>;
528 def NEONvrev16 : SDNode<"ARMISD::VREV16", SDTARMVSHUF>;
530 def SDTARMVSHUF2 : SDTypeProfile<2, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
532 SDTCisSameAs<0, 3>]>;
533 def NEONzip : SDNode<"ARMISD::VZIP", SDTARMVSHUF2>;
534 def NEONuzp : SDNode<"ARMISD::VUZP", SDTARMVSHUF2>;
535 def NEONtrn : SDNode<"ARMISD::VTRN", SDTARMVSHUF2>;
537 def SDTARMVMULL : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
538 SDTCisSameAs<1, 2>]>;
539 def NEONvmulls : SDNode<"ARMISD::VMULLs", SDTARMVMULL>;
540 def NEONvmullu : SDNode<"ARMISD::VMULLu", SDTARMVMULL>;
542 def SDTARMFMAX : SDTypeProfile<1, 2, [SDTCisVT<0, f32>, SDTCisSameAs<0, 1>,
543 SDTCisSameAs<0, 2>]>;
544 def NEONfmax : SDNode<"ARMISD::FMAX", SDTARMFMAX>;
545 def NEONfmin : SDNode<"ARMISD::FMIN", SDTARMFMAX>;
547 def NEONimmAllZerosV: PatLeaf<(NEONvmovImm (i32 timm)), [{
548 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
549 unsigned EltBits = 0;
550 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
551 return (EltBits == 32 && EltVal == 0);
554 def NEONimmAllOnesV: PatLeaf<(NEONvmovImm (i32 timm)), [{
555 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
556 unsigned EltBits = 0;
557 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
558 return (EltBits == 8 && EltVal == 0xff);
561 //===----------------------------------------------------------------------===//
562 // NEON load / store instructions
563 //===----------------------------------------------------------------------===//
565 // Use VLDM to load a Q register as a D register pair.
566 // This is a pseudo instruction that is expanded to VLDMD after reg alloc.
568 : PseudoVFPLdStM<(outs DPair:$dst), (ins GPR:$Rn),
570 [(set DPair:$dst, (v2f64 (load GPR:$Rn)))]>;
572 // Use VSTM to store a Q register as a D register pair.
573 // This is a pseudo instruction that is expanded to VSTMD after reg alloc.
575 : PseudoVFPLdStM<(outs), (ins DPair:$src, GPR:$Rn),
577 [(store (v2f64 DPair:$src), GPR:$Rn)]>;
579 // Classes for VLD* pseudo-instructions with multi-register operands.
580 // These are expanded to real instructions after register allocation.
581 class VLDQPseudo<InstrItinClass itin>
582 : PseudoNLdSt<(outs QPR:$dst), (ins addrmode6:$addr), itin, "">;
583 class VLDQWBPseudo<InstrItinClass itin>
584 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
585 (ins addrmode6:$addr, am6offset:$offset), itin,
587 class VLDQWBfixedPseudo<InstrItinClass itin>
588 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
589 (ins addrmode6:$addr), itin,
591 class VLDQWBregisterPseudo<InstrItinClass itin>
592 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
593 (ins addrmode6:$addr, rGPR:$offset), itin,
596 class VLDQQPseudo<InstrItinClass itin>
597 : PseudoNLdSt<(outs QQPR:$dst), (ins addrmode6:$addr), itin, "">;
598 class VLDQQWBPseudo<InstrItinClass itin>
599 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
600 (ins addrmode6:$addr, am6offset:$offset), itin,
602 class VLDQQWBfixedPseudo<InstrItinClass itin>
603 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
604 (ins addrmode6:$addr), itin,
606 class VLDQQWBregisterPseudo<InstrItinClass itin>
607 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
608 (ins addrmode6:$addr, rGPR:$offset), itin,
612 class VLDQQQQPseudo<InstrItinClass itin>
613 : PseudoNLdSt<(outs QQQQPR:$dst), (ins addrmode6:$addr, QQQQPR:$src),itin,
615 class VLDQQQQWBPseudo<InstrItinClass itin>
616 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
617 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
618 "$addr.addr = $wb, $src = $dst">;
620 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
622 // VLD1 : Vector Load (multiple single elements)
623 class VLD1D<bits<4> op7_4, string Dt>
624 : NLdSt<0,0b10,0b0111,op7_4, (outs VecListOneD:$Vd),
625 (ins addrmode6:$Rn), IIC_VLD1,
626 "vld1", Dt, "$Vd, $Rn", "", []> {
629 let DecoderMethod = "DecodeVLDST1Instruction";
631 class VLD1Q<bits<4> op7_4, string Dt>
632 : NLdSt<0,0b10,0b1010,op7_4, (outs VecListDPair:$Vd),
633 (ins addrmode6:$Rn), IIC_VLD1x2,
634 "vld1", Dt, "$Vd, $Rn", "", []> {
636 let Inst{5-4} = Rn{5-4};
637 let DecoderMethod = "DecodeVLDST1Instruction";
640 def VLD1d8 : VLD1D<{0,0,0,?}, "8">;
641 def VLD1d16 : VLD1D<{0,1,0,?}, "16">;
642 def VLD1d32 : VLD1D<{1,0,0,?}, "32">;
643 def VLD1d64 : VLD1D<{1,1,0,?}, "64">;
645 def VLD1q8 : VLD1Q<{0,0,?,?}, "8">;
646 def VLD1q16 : VLD1Q<{0,1,?,?}, "16">;
647 def VLD1q32 : VLD1Q<{1,0,?,?}, "32">;
648 def VLD1q64 : VLD1Q<{1,1,?,?}, "64">;
650 // ...with address register writeback:
651 multiclass VLD1DWB<bits<4> op7_4, string Dt> {
652 def _fixed : NLdSt<0,0b10, 0b0111,op7_4, (outs VecListOneD:$Vd, GPR:$wb),
653 (ins addrmode6:$Rn), IIC_VLD1u,
654 "vld1", Dt, "$Vd, $Rn!",
655 "$Rn.addr = $wb", []> {
656 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
658 let DecoderMethod = "DecodeVLDST1Instruction";
659 let AsmMatchConverter = "cvtVLDwbFixed";
661 def _register : NLdSt<0,0b10,0b0111,op7_4, (outs VecListOneD:$Vd, GPR:$wb),
662 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1u,
663 "vld1", Dt, "$Vd, $Rn, $Rm",
664 "$Rn.addr = $wb", []> {
666 let DecoderMethod = "DecodeVLDST1Instruction";
667 let AsmMatchConverter = "cvtVLDwbRegister";
670 multiclass VLD1QWB<bits<4> op7_4, string Dt> {
671 def _fixed : NLdSt<0,0b10,0b1010,op7_4, (outs VecListDPair:$Vd, GPR:$wb),
672 (ins addrmode6:$Rn), IIC_VLD1x2u,
673 "vld1", Dt, "$Vd, $Rn!",
674 "$Rn.addr = $wb", []> {
675 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
676 let Inst{5-4} = Rn{5-4};
677 let DecoderMethod = "DecodeVLDST1Instruction";
678 let AsmMatchConverter = "cvtVLDwbFixed";
680 def _register : NLdSt<0,0b10,0b1010,op7_4, (outs VecListDPair:$Vd, GPR:$wb),
681 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1x2u,
682 "vld1", Dt, "$Vd, $Rn, $Rm",
683 "$Rn.addr = $wb", []> {
684 let Inst{5-4} = Rn{5-4};
685 let DecoderMethod = "DecodeVLDST1Instruction";
686 let AsmMatchConverter = "cvtVLDwbRegister";
690 defm VLD1d8wb : VLD1DWB<{0,0,0,?}, "8">;
691 defm VLD1d16wb : VLD1DWB<{0,1,0,?}, "16">;
692 defm VLD1d32wb : VLD1DWB<{1,0,0,?}, "32">;
693 defm VLD1d64wb : VLD1DWB<{1,1,0,?}, "64">;
694 defm VLD1q8wb : VLD1QWB<{0,0,?,?}, "8">;
695 defm VLD1q16wb : VLD1QWB<{0,1,?,?}, "16">;
696 defm VLD1q32wb : VLD1QWB<{1,0,?,?}, "32">;
697 defm VLD1q64wb : VLD1QWB<{1,1,?,?}, "64">;
699 // ...with 3 registers
700 class VLD1D3<bits<4> op7_4, string Dt>
701 : NLdSt<0,0b10,0b0110,op7_4, (outs VecListThreeD:$Vd),
702 (ins addrmode6:$Rn), IIC_VLD1x3, "vld1", Dt,
703 "$Vd, $Rn", "", []> {
706 let DecoderMethod = "DecodeVLDST1Instruction";
708 multiclass VLD1D3WB<bits<4> op7_4, string Dt> {
709 def _fixed : NLdSt<0,0b10,0b0110, op7_4, (outs VecListThreeD:$Vd, GPR:$wb),
710 (ins addrmode6:$Rn), IIC_VLD1x2u,
711 "vld1", Dt, "$Vd, $Rn!",
712 "$Rn.addr = $wb", []> {
713 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
715 let DecoderMethod = "DecodeVLDST1Instruction";
716 let AsmMatchConverter = "cvtVLDwbFixed";
718 def _register : NLdSt<0,0b10,0b0110,op7_4, (outs VecListThreeD:$Vd, GPR:$wb),
719 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1x2u,
720 "vld1", Dt, "$Vd, $Rn, $Rm",
721 "$Rn.addr = $wb", []> {
723 let DecoderMethod = "DecodeVLDST1Instruction";
724 let AsmMatchConverter = "cvtVLDwbRegister";
728 def VLD1d8T : VLD1D3<{0,0,0,?}, "8">;
729 def VLD1d16T : VLD1D3<{0,1,0,?}, "16">;
730 def VLD1d32T : VLD1D3<{1,0,0,?}, "32">;
731 def VLD1d64T : VLD1D3<{1,1,0,?}, "64">;
733 defm VLD1d8Twb : VLD1D3WB<{0,0,0,?}, "8">;
734 defm VLD1d16Twb : VLD1D3WB<{0,1,0,?}, "16">;
735 defm VLD1d32Twb : VLD1D3WB<{1,0,0,?}, "32">;
736 defm VLD1d64Twb : VLD1D3WB<{1,1,0,?}, "64">;
738 def VLD1d64TPseudo : VLDQQPseudo<IIC_VLD1x3>;
740 // ...with 4 registers
741 class VLD1D4<bits<4> op7_4, string Dt>
742 : NLdSt<0, 0b10, 0b0010, op7_4, (outs VecListFourD:$Vd),
743 (ins addrmode6:$Rn), IIC_VLD1x4, "vld1", Dt,
744 "$Vd, $Rn", "", []> {
746 let Inst{5-4} = Rn{5-4};
747 let DecoderMethod = "DecodeVLDST1Instruction";
749 multiclass VLD1D4WB<bits<4> op7_4, string Dt> {
750 def _fixed : NLdSt<0,0b10,0b0010, op7_4, (outs VecListFourD:$Vd, GPR:$wb),
751 (ins addrmode6:$Rn), IIC_VLD1x2u,
752 "vld1", Dt, "$Vd, $Rn!",
753 "$Rn.addr = $wb", []> {
754 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
755 let Inst{5-4} = Rn{5-4};
756 let DecoderMethod = "DecodeVLDST1Instruction";
757 let AsmMatchConverter = "cvtVLDwbFixed";
759 def _register : NLdSt<0,0b10,0b0010,op7_4, (outs VecListFourD:$Vd, GPR:$wb),
760 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1x2u,
761 "vld1", Dt, "$Vd, $Rn, $Rm",
762 "$Rn.addr = $wb", []> {
763 let Inst{5-4} = Rn{5-4};
764 let DecoderMethod = "DecodeVLDST1Instruction";
765 let AsmMatchConverter = "cvtVLDwbRegister";
769 def VLD1d8Q : VLD1D4<{0,0,?,?}, "8">;
770 def VLD1d16Q : VLD1D4<{0,1,?,?}, "16">;
771 def VLD1d32Q : VLD1D4<{1,0,?,?}, "32">;
772 def VLD1d64Q : VLD1D4<{1,1,?,?}, "64">;
774 defm VLD1d8Qwb : VLD1D4WB<{0,0,?,?}, "8">;
775 defm VLD1d16Qwb : VLD1D4WB<{0,1,?,?}, "16">;
776 defm VLD1d32Qwb : VLD1D4WB<{1,0,?,?}, "32">;
777 defm VLD1d64Qwb : VLD1D4WB<{1,1,?,?}, "64">;
779 def VLD1d64QPseudo : VLDQQPseudo<IIC_VLD1x4>;
781 // VLD2 : Vector Load (multiple 2-element structures)
782 class VLD2<bits<4> op11_8, bits<4> op7_4, string Dt, RegisterOperand VdTy,
784 : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd),
785 (ins addrmode6:$Rn), itin,
786 "vld2", Dt, "$Vd, $Rn", "", []> {
788 let Inst{5-4} = Rn{5-4};
789 let DecoderMethod = "DecodeVLDST2Instruction";
792 def VLD2d8 : VLD2<0b1000, {0,0,?,?}, "8", VecListDPair, IIC_VLD2>;
793 def VLD2d16 : VLD2<0b1000, {0,1,?,?}, "16", VecListDPair, IIC_VLD2>;
794 def VLD2d32 : VLD2<0b1000, {1,0,?,?}, "32", VecListDPair, IIC_VLD2>;
796 def VLD2q8 : VLD2<0b0011, {0,0,?,?}, "8", VecListFourD, IIC_VLD2x2>;
797 def VLD2q16 : VLD2<0b0011, {0,1,?,?}, "16", VecListFourD, IIC_VLD2x2>;
798 def VLD2q32 : VLD2<0b0011, {1,0,?,?}, "32", VecListFourD, IIC_VLD2x2>;
800 def VLD2q8Pseudo : VLDQQPseudo<IIC_VLD2x2>;
801 def VLD2q16Pseudo : VLDQQPseudo<IIC_VLD2x2>;
802 def VLD2q32Pseudo : VLDQQPseudo<IIC_VLD2x2>;
804 // ...with address register writeback:
805 multiclass VLD2WB<bits<4> op11_8, bits<4> op7_4, string Dt,
806 RegisterOperand VdTy, InstrItinClass itin> {
807 def _fixed : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd, GPR:$wb),
808 (ins addrmode6:$Rn), itin,
809 "vld2", Dt, "$Vd, $Rn!",
810 "$Rn.addr = $wb", []> {
811 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
812 let Inst{5-4} = Rn{5-4};
813 let DecoderMethod = "DecodeVLDST2Instruction";
814 let AsmMatchConverter = "cvtVLDwbFixed";
816 def _register : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd, GPR:$wb),
817 (ins addrmode6:$Rn, rGPR:$Rm), itin,
818 "vld2", Dt, "$Vd, $Rn, $Rm",
819 "$Rn.addr = $wb", []> {
820 let Inst{5-4} = Rn{5-4};
821 let DecoderMethod = "DecodeVLDST2Instruction";
822 let AsmMatchConverter = "cvtVLDwbRegister";
826 defm VLD2d8wb : VLD2WB<0b1000, {0,0,?,?}, "8", VecListDPair, IIC_VLD2u>;
827 defm VLD2d16wb : VLD2WB<0b1000, {0,1,?,?}, "16", VecListDPair, IIC_VLD2u>;
828 defm VLD2d32wb : VLD2WB<0b1000, {1,0,?,?}, "32", VecListDPair, IIC_VLD2u>;
830 defm VLD2q8wb : VLD2WB<0b0011, {0,0,?,?}, "8", VecListFourD, IIC_VLD2x2u>;
831 defm VLD2q16wb : VLD2WB<0b0011, {0,1,?,?}, "16", VecListFourD, IIC_VLD2x2u>;
832 defm VLD2q32wb : VLD2WB<0b0011, {1,0,?,?}, "32", VecListFourD, IIC_VLD2x2u>;
834 def VLD2q8PseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD2x2u>;
835 def VLD2q16PseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD2x2u>;
836 def VLD2q32PseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD2x2u>;
837 def VLD2q8PseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD2x2u>;
838 def VLD2q16PseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD2x2u>;
839 def VLD2q32PseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD2x2u>;
841 // ...with double-spaced registers
842 def VLD2b8 : VLD2<0b1001, {0,0,?,?}, "8", VecListDPairSpaced, IIC_VLD2>;
843 def VLD2b16 : VLD2<0b1001, {0,1,?,?}, "16", VecListDPairSpaced, IIC_VLD2>;
844 def VLD2b32 : VLD2<0b1001, {1,0,?,?}, "32", VecListDPairSpaced, IIC_VLD2>;
845 defm VLD2b8wb : VLD2WB<0b1001, {0,0,?,?}, "8", VecListDPairSpaced, IIC_VLD2u>;
846 defm VLD2b16wb : VLD2WB<0b1001, {0,1,?,?}, "16", VecListDPairSpaced, IIC_VLD2u>;
847 defm VLD2b32wb : VLD2WB<0b1001, {1,0,?,?}, "32", VecListDPairSpaced, IIC_VLD2u>;
849 // VLD3 : Vector Load (multiple 3-element structures)
850 class VLD3D<bits<4> op11_8, bits<4> op7_4, string Dt>
851 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
852 (ins addrmode6:$Rn), IIC_VLD3,
853 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn", "", []> {
856 let DecoderMethod = "DecodeVLDST3Instruction";
859 def VLD3d8 : VLD3D<0b0100, {0,0,0,?}, "8">;
860 def VLD3d16 : VLD3D<0b0100, {0,1,0,?}, "16">;
861 def VLD3d32 : VLD3D<0b0100, {1,0,0,?}, "32">;
863 def VLD3d8Pseudo : VLDQQPseudo<IIC_VLD3>;
864 def VLD3d16Pseudo : VLDQQPseudo<IIC_VLD3>;
865 def VLD3d32Pseudo : VLDQQPseudo<IIC_VLD3>;
867 // ...with address register writeback:
868 class VLD3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
869 : NLdSt<0, 0b10, op11_8, op7_4,
870 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
871 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD3u,
872 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn$Rm",
873 "$Rn.addr = $wb", []> {
875 let DecoderMethod = "DecodeVLDST3Instruction";
878 def VLD3d8_UPD : VLD3DWB<0b0100, {0,0,0,?}, "8">;
879 def VLD3d16_UPD : VLD3DWB<0b0100, {0,1,0,?}, "16">;
880 def VLD3d32_UPD : VLD3DWB<0b0100, {1,0,0,?}, "32">;
882 def VLD3d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
883 def VLD3d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
884 def VLD3d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
886 // ...with double-spaced registers:
887 def VLD3q8 : VLD3D<0b0101, {0,0,0,?}, "8">;
888 def VLD3q16 : VLD3D<0b0101, {0,1,0,?}, "16">;
889 def VLD3q32 : VLD3D<0b0101, {1,0,0,?}, "32">;
890 def VLD3q8_UPD : VLD3DWB<0b0101, {0,0,0,?}, "8">;
891 def VLD3q16_UPD : VLD3DWB<0b0101, {0,1,0,?}, "16">;
892 def VLD3q32_UPD : VLD3DWB<0b0101, {1,0,0,?}, "32">;
894 def VLD3q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
895 def VLD3q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
896 def VLD3q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
898 // ...alternate versions to be allocated odd register numbers:
899 def VLD3q8oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
900 def VLD3q16oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
901 def VLD3q32oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
903 def VLD3q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
904 def VLD3q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
905 def VLD3q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
907 // VLD4 : Vector Load (multiple 4-element structures)
908 class VLD4D<bits<4> op11_8, bits<4> op7_4, string Dt>
909 : NLdSt<0, 0b10, op11_8, op7_4,
910 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
911 (ins addrmode6:$Rn), IIC_VLD4,
912 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> {
914 let Inst{5-4} = Rn{5-4};
915 let DecoderMethod = "DecodeVLDST4Instruction";
918 def VLD4d8 : VLD4D<0b0000, {0,0,?,?}, "8">;
919 def VLD4d16 : VLD4D<0b0000, {0,1,?,?}, "16">;
920 def VLD4d32 : VLD4D<0b0000, {1,0,?,?}, "32">;
922 def VLD4d8Pseudo : VLDQQPseudo<IIC_VLD4>;
923 def VLD4d16Pseudo : VLDQQPseudo<IIC_VLD4>;
924 def VLD4d32Pseudo : VLDQQPseudo<IIC_VLD4>;
926 // ...with address register writeback:
927 class VLD4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
928 : NLdSt<0, 0b10, op11_8, op7_4,
929 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
930 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD4u,
931 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm",
932 "$Rn.addr = $wb", []> {
933 let Inst{5-4} = Rn{5-4};
934 let DecoderMethod = "DecodeVLDST4Instruction";
937 def VLD4d8_UPD : VLD4DWB<0b0000, {0,0,?,?}, "8">;
938 def VLD4d16_UPD : VLD4DWB<0b0000, {0,1,?,?}, "16">;
939 def VLD4d32_UPD : VLD4DWB<0b0000, {1,0,?,?}, "32">;
941 def VLD4d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
942 def VLD4d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
943 def VLD4d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
945 // ...with double-spaced registers:
946 def VLD4q8 : VLD4D<0b0001, {0,0,?,?}, "8">;
947 def VLD4q16 : VLD4D<0b0001, {0,1,?,?}, "16">;
948 def VLD4q32 : VLD4D<0b0001, {1,0,?,?}, "32">;
949 def VLD4q8_UPD : VLD4DWB<0b0001, {0,0,?,?}, "8">;
950 def VLD4q16_UPD : VLD4DWB<0b0001, {0,1,?,?}, "16">;
951 def VLD4q32_UPD : VLD4DWB<0b0001, {1,0,?,?}, "32">;
953 def VLD4q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
954 def VLD4q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
955 def VLD4q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
957 // ...alternate versions to be allocated odd register numbers:
958 def VLD4q8oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
959 def VLD4q16oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
960 def VLD4q32oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
962 def VLD4q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
963 def VLD4q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
964 def VLD4q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
966 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
968 // Classes for VLD*LN pseudo-instructions with multi-register operands.
969 // These are expanded to real instructions after register allocation.
970 class VLDQLNPseudo<InstrItinClass itin>
971 : PseudoNLdSt<(outs QPR:$dst),
972 (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
973 itin, "$src = $dst">;
974 class VLDQLNWBPseudo<InstrItinClass itin>
975 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
976 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
977 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
978 class VLDQQLNPseudo<InstrItinClass itin>
979 : PseudoNLdSt<(outs QQPR:$dst),
980 (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
981 itin, "$src = $dst">;
982 class VLDQQLNWBPseudo<InstrItinClass itin>
983 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
984 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
985 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
986 class VLDQQQQLNPseudo<InstrItinClass itin>
987 : PseudoNLdSt<(outs QQQQPR:$dst),
988 (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
989 itin, "$src = $dst">;
990 class VLDQQQQLNWBPseudo<InstrItinClass itin>
991 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
992 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
993 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
995 // VLD1LN : Vector Load (single element to one lane)
996 class VLD1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
998 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
999 (ins addrmode6:$Rn, DPR:$src, nohash_imm:$lane),
1000 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
1002 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
1003 (i32 (LoadOp addrmode6:$Rn)),
1006 let DecoderMethod = "DecodeVLD1LN";
1008 class VLD1LN32<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1010 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
1011 (ins addrmode6oneL32:$Rn, DPR:$src, nohash_imm:$lane),
1012 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
1014 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
1015 (i32 (LoadOp addrmode6oneL32:$Rn)),
1018 let DecoderMethod = "DecodeVLD1LN";
1020 class VLD1QLNPseudo<ValueType Ty, PatFrag LoadOp> : VLDQLNPseudo<IIC_VLD1ln> {
1021 let Pattern = [(set QPR:$dst, (vector_insert (Ty QPR:$src),
1022 (i32 (LoadOp addrmode6:$addr)),
1026 def VLD1LNd8 : VLD1LN<0b0000, {?,?,?,0}, "8", v8i8, extloadi8> {
1027 let Inst{7-5} = lane{2-0};
1029 def VLD1LNd16 : VLD1LN<0b0100, {?,?,0,?}, "16", v4i16, extloadi16> {
1030 let Inst{7-6} = lane{1-0};
1031 let Inst{5-4} = Rn{5-4};
1033 def VLD1LNd32 : VLD1LN32<0b1000, {?,0,?,?}, "32", v2i32, load> {
1034 let Inst{7} = lane{0};
1035 let Inst{5-4} = Rn{5-4};
1038 def VLD1LNq8Pseudo : VLD1QLNPseudo<v16i8, extloadi8>;
1039 def VLD1LNq16Pseudo : VLD1QLNPseudo<v8i16, extloadi16>;
1040 def VLD1LNq32Pseudo : VLD1QLNPseudo<v4i32, load>;
1042 def : Pat<(vector_insert (v2f32 DPR:$src),
1043 (f32 (load addrmode6:$addr)), imm:$lane),
1044 (VLD1LNd32 addrmode6:$addr, DPR:$src, imm:$lane)>;
1045 def : Pat<(vector_insert (v4f32 QPR:$src),
1046 (f32 (load addrmode6:$addr)), imm:$lane),
1047 (VLD1LNq32Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
1049 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
1051 // ...with address register writeback:
1052 class VLD1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1053 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, GPR:$wb),
1054 (ins addrmode6:$Rn, am6offset:$Rm,
1055 DPR:$src, nohash_imm:$lane), IIC_VLD1lnu, "vld1", Dt,
1056 "\\{$Vd[$lane]\\}, $Rn$Rm",
1057 "$src = $Vd, $Rn.addr = $wb", []> {
1058 let DecoderMethod = "DecodeVLD1LN";
1061 def VLD1LNd8_UPD : VLD1LNWB<0b0000, {?,?,?,0}, "8"> {
1062 let Inst{7-5} = lane{2-0};
1064 def VLD1LNd16_UPD : VLD1LNWB<0b0100, {?,?,0,?}, "16"> {
1065 let Inst{7-6} = lane{1-0};
1066 let Inst{4} = Rn{4};
1068 def VLD1LNd32_UPD : VLD1LNWB<0b1000, {?,0,?,?}, "32"> {
1069 let Inst{7} = lane{0};
1070 let Inst{5} = Rn{4};
1071 let Inst{4} = Rn{4};
1074 def VLD1LNq8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
1075 def VLD1LNq16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
1076 def VLD1LNq32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
1078 // VLD2LN : Vector Load (single 2-element structure to one lane)
1079 class VLD2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1080 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2),
1081 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, nohash_imm:$lane),
1082 IIC_VLD2ln, "vld2", Dt, "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn",
1083 "$src1 = $Vd, $src2 = $dst2", []> {
1085 let Inst{4} = Rn{4};
1086 let DecoderMethod = "DecodeVLD2LN";
1089 def VLD2LNd8 : VLD2LN<0b0001, {?,?,?,?}, "8"> {
1090 let Inst{7-5} = lane{2-0};
1092 def VLD2LNd16 : VLD2LN<0b0101, {?,?,0,?}, "16"> {
1093 let Inst{7-6} = lane{1-0};
1095 def VLD2LNd32 : VLD2LN<0b1001, {?,0,0,?}, "32"> {
1096 let Inst{7} = lane{0};
1099 def VLD2LNd8Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
1100 def VLD2LNd16Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
1101 def VLD2LNd32Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
1103 // ...with double-spaced registers:
1104 def VLD2LNq16 : VLD2LN<0b0101, {?,?,1,?}, "16"> {
1105 let Inst{7-6} = lane{1-0};
1107 def VLD2LNq32 : VLD2LN<0b1001, {?,1,0,?}, "32"> {
1108 let Inst{7} = lane{0};
1111 def VLD2LNq16Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
1112 def VLD2LNq32Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
1114 // ...with address register writeback:
1115 class VLD2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1116 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
1117 (ins addrmode6:$Rn, am6offset:$Rm,
1118 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VLD2lnu, "vld2", Dt,
1119 "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn$Rm",
1120 "$src1 = $Vd, $src2 = $dst2, $Rn.addr = $wb", []> {
1121 let Inst{4} = Rn{4};
1122 let DecoderMethod = "DecodeVLD2LN";
1125 def VLD2LNd8_UPD : VLD2LNWB<0b0001, {?,?,?,?}, "8"> {
1126 let Inst{7-5} = lane{2-0};
1128 def VLD2LNd16_UPD : VLD2LNWB<0b0101, {?,?,0,?}, "16"> {
1129 let Inst{7-6} = lane{1-0};
1131 def VLD2LNd32_UPD : VLD2LNWB<0b1001, {?,0,0,?}, "32"> {
1132 let Inst{7} = lane{0};
1135 def VLD2LNd8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
1136 def VLD2LNd16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
1137 def VLD2LNd32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
1139 def VLD2LNq16_UPD : VLD2LNWB<0b0101, {?,?,1,?}, "16"> {
1140 let Inst{7-6} = lane{1-0};
1142 def VLD2LNq32_UPD : VLD2LNWB<0b1001, {?,1,0,?}, "32"> {
1143 let Inst{7} = lane{0};
1146 def VLD2LNq16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
1147 def VLD2LNq32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
1149 // VLD3LN : Vector Load (single 3-element structure to one lane)
1150 class VLD3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1151 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
1152 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3,
1153 nohash_imm:$lane), IIC_VLD3ln, "vld3", Dt,
1154 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn",
1155 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3", []> {
1157 let DecoderMethod = "DecodeVLD3LN";
1160 def VLD3LNd8 : VLD3LN<0b0010, {?,?,?,0}, "8"> {
1161 let Inst{7-5} = lane{2-0};
1163 def VLD3LNd16 : VLD3LN<0b0110, {?,?,0,0}, "16"> {
1164 let Inst{7-6} = lane{1-0};
1166 def VLD3LNd32 : VLD3LN<0b1010, {?,0,0,0}, "32"> {
1167 let Inst{7} = lane{0};
1170 def VLD3LNd8Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
1171 def VLD3LNd16Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
1172 def VLD3LNd32Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
1174 // ...with double-spaced registers:
1175 def VLD3LNq16 : VLD3LN<0b0110, {?,?,1,0}, "16"> {
1176 let Inst{7-6} = lane{1-0};
1178 def VLD3LNq32 : VLD3LN<0b1010, {?,1,0,0}, "32"> {
1179 let Inst{7} = lane{0};
1182 def VLD3LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
1183 def VLD3LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
1185 // ...with address register writeback:
1186 class VLD3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1187 : NLdStLn<1, 0b10, op11_8, op7_4,
1188 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
1189 (ins addrmode6:$Rn, am6offset:$Rm,
1190 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
1191 IIC_VLD3lnu, "vld3", Dt,
1192 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn$Rm",
1193 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $Rn.addr = $wb",
1195 let DecoderMethod = "DecodeVLD3LN";
1198 def VLD3LNd8_UPD : VLD3LNWB<0b0010, {?,?,?,0}, "8"> {
1199 let Inst{7-5} = lane{2-0};
1201 def VLD3LNd16_UPD : VLD3LNWB<0b0110, {?,?,0,0}, "16"> {
1202 let Inst{7-6} = lane{1-0};
1204 def VLD3LNd32_UPD : VLD3LNWB<0b1010, {?,0,0,0}, "32"> {
1205 let Inst{7} = lane{0};
1208 def VLD3LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
1209 def VLD3LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
1210 def VLD3LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
1212 def VLD3LNq16_UPD : VLD3LNWB<0b0110, {?,?,1,0}, "16"> {
1213 let Inst{7-6} = lane{1-0};
1215 def VLD3LNq32_UPD : VLD3LNWB<0b1010, {?,1,0,0}, "32"> {
1216 let Inst{7} = lane{0};
1219 def VLD3LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
1220 def VLD3LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
1222 // VLD4LN : Vector Load (single 4-element structure to one lane)
1223 class VLD4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1224 : NLdStLn<1, 0b10, op11_8, op7_4,
1225 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
1226 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
1227 nohash_imm:$lane), IIC_VLD4ln, "vld4", Dt,
1228 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn",
1229 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4", []> {
1231 let Inst{4} = Rn{4};
1232 let DecoderMethod = "DecodeVLD4LN";
1235 def VLD4LNd8 : VLD4LN<0b0011, {?,?,?,?}, "8"> {
1236 let Inst{7-5} = lane{2-0};
1238 def VLD4LNd16 : VLD4LN<0b0111, {?,?,0,?}, "16"> {
1239 let Inst{7-6} = lane{1-0};
1241 def VLD4LNd32 : VLD4LN<0b1011, {?,0,?,?}, "32"> {
1242 let Inst{7} = lane{0};
1243 let Inst{5} = Rn{5};
1246 def VLD4LNd8Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
1247 def VLD4LNd16Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
1248 def VLD4LNd32Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
1250 // ...with double-spaced registers:
1251 def VLD4LNq16 : VLD4LN<0b0111, {?,?,1,?}, "16"> {
1252 let Inst{7-6} = lane{1-0};
1254 def VLD4LNq32 : VLD4LN<0b1011, {?,1,?,?}, "32"> {
1255 let Inst{7} = lane{0};
1256 let Inst{5} = Rn{5};
1259 def VLD4LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
1260 def VLD4LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
1262 // ...with address register writeback:
1263 class VLD4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1264 : NLdStLn<1, 0b10, op11_8, op7_4,
1265 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
1266 (ins addrmode6:$Rn, am6offset:$Rm,
1267 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
1268 IIC_VLD4lnu, "vld4", Dt,
1269 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn$Rm",
1270 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4, $Rn.addr = $wb",
1272 let Inst{4} = Rn{4};
1273 let DecoderMethod = "DecodeVLD4LN" ;
1276 def VLD4LNd8_UPD : VLD4LNWB<0b0011, {?,?,?,?}, "8"> {
1277 let Inst{7-5} = lane{2-0};
1279 def VLD4LNd16_UPD : VLD4LNWB<0b0111, {?,?,0,?}, "16"> {
1280 let Inst{7-6} = lane{1-0};
1282 def VLD4LNd32_UPD : VLD4LNWB<0b1011, {?,0,?,?}, "32"> {
1283 let Inst{7} = lane{0};
1284 let Inst{5} = Rn{5};
1287 def VLD4LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
1288 def VLD4LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
1289 def VLD4LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
1291 def VLD4LNq16_UPD : VLD4LNWB<0b0111, {?,?,1,?}, "16"> {
1292 let Inst{7-6} = lane{1-0};
1294 def VLD4LNq32_UPD : VLD4LNWB<0b1011, {?,1,?,?}, "32"> {
1295 let Inst{7} = lane{0};
1296 let Inst{5} = Rn{5};
1299 def VLD4LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
1300 def VLD4LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
1302 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
1304 // VLD1DUP : Vector Load (single element to all lanes)
1305 class VLD1DUP<bits<4> op7_4, string Dt, ValueType Ty, PatFrag LoadOp>
1306 : NLdSt<1, 0b10, 0b1100, op7_4, (outs VecListOneDAllLanes:$Vd),
1307 (ins addrmode6dup:$Rn),
1308 IIC_VLD1dup, "vld1", Dt, "$Vd, $Rn", "",
1309 [(set VecListOneDAllLanes:$Vd,
1310 (Ty (NEONvdup (i32 (LoadOp addrmode6dup:$Rn)))))]> {
1312 let Inst{4} = Rn{4};
1313 let DecoderMethod = "DecodeVLD1DupInstruction";
1315 def VLD1DUPd8 : VLD1DUP<{0,0,0,?}, "8", v8i8, extloadi8>;
1316 def VLD1DUPd16 : VLD1DUP<{0,1,0,?}, "16", v4i16, extloadi16>;
1317 def VLD1DUPd32 : VLD1DUP<{1,0,0,?}, "32", v2i32, load>;
1319 def : Pat<(v2f32 (NEONvdup (f32 (load addrmode6dup:$addr)))),
1320 (VLD1DUPd32 addrmode6:$addr)>;
1322 class VLD1QDUP<bits<4> op7_4, string Dt, ValueType Ty, PatFrag LoadOp>
1323 : NLdSt<1, 0b10, 0b1100, op7_4, (outs VecListDPairAllLanes:$Vd),
1324 (ins addrmode6dup:$Rn), IIC_VLD1dup,
1325 "vld1", Dt, "$Vd, $Rn", "",
1326 [(set VecListDPairAllLanes:$Vd,
1327 (Ty (NEONvdup (i32 (LoadOp addrmode6dup:$Rn)))))]> {
1329 let Inst{4} = Rn{4};
1330 let DecoderMethod = "DecodeVLD1DupInstruction";
1333 def VLD1DUPq8 : VLD1QDUP<{0,0,1,0}, "8", v16i8, extloadi8>;
1334 def VLD1DUPq16 : VLD1QDUP<{0,1,1,?}, "16", v8i16, extloadi16>;
1335 def VLD1DUPq32 : VLD1QDUP<{1,0,1,?}, "32", v4i32, load>;
1337 def : Pat<(v4f32 (NEONvdup (f32 (load addrmode6dup:$addr)))),
1338 (VLD1DUPq32 addrmode6:$addr)>;
1340 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
1341 // ...with address register writeback:
1342 multiclass VLD1DUPWB<bits<4> op7_4, string Dt> {
1343 def _fixed : NLdSt<1, 0b10, 0b1100, op7_4,
1344 (outs VecListOneDAllLanes:$Vd, GPR:$wb),
1345 (ins addrmode6dup:$Rn), IIC_VLD1dupu,
1346 "vld1", Dt, "$Vd, $Rn!",
1347 "$Rn.addr = $wb", []> {
1348 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1349 let Inst{4} = Rn{4};
1350 let DecoderMethod = "DecodeVLD1DupInstruction";
1351 let AsmMatchConverter = "cvtVLDwbFixed";
1353 def _register : NLdSt<1, 0b10, 0b1100, op7_4,
1354 (outs VecListOneDAllLanes:$Vd, GPR:$wb),
1355 (ins addrmode6dup:$Rn, rGPR:$Rm), IIC_VLD1dupu,
1356 "vld1", Dt, "$Vd, $Rn, $Rm",
1357 "$Rn.addr = $wb", []> {
1358 let Inst{4} = Rn{4};
1359 let DecoderMethod = "DecodeVLD1DupInstruction";
1360 let AsmMatchConverter = "cvtVLDwbRegister";
1363 multiclass VLD1QDUPWB<bits<4> op7_4, string Dt> {
1364 def _fixed : NLdSt<1, 0b10, 0b1100, op7_4,
1365 (outs VecListDPairAllLanes:$Vd, GPR:$wb),
1366 (ins addrmode6dup:$Rn), IIC_VLD1dupu,
1367 "vld1", Dt, "$Vd, $Rn!",
1368 "$Rn.addr = $wb", []> {
1369 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1370 let Inst{4} = Rn{4};
1371 let DecoderMethod = "DecodeVLD1DupInstruction";
1372 let AsmMatchConverter = "cvtVLDwbFixed";
1374 def _register : NLdSt<1, 0b10, 0b1100, op7_4,
1375 (outs VecListDPairAllLanes:$Vd, GPR:$wb),
1376 (ins addrmode6dup:$Rn, rGPR:$Rm), IIC_VLD1dupu,
1377 "vld1", Dt, "$Vd, $Rn, $Rm",
1378 "$Rn.addr = $wb", []> {
1379 let Inst{4} = Rn{4};
1380 let DecoderMethod = "DecodeVLD1DupInstruction";
1381 let AsmMatchConverter = "cvtVLDwbRegister";
1385 defm VLD1DUPd8wb : VLD1DUPWB<{0,0,0,0}, "8">;
1386 defm VLD1DUPd16wb : VLD1DUPWB<{0,1,0,?}, "16">;
1387 defm VLD1DUPd32wb : VLD1DUPWB<{1,0,0,?}, "32">;
1389 defm VLD1DUPq8wb : VLD1QDUPWB<{0,0,1,0}, "8">;
1390 defm VLD1DUPq16wb : VLD1QDUPWB<{0,1,1,?}, "16">;
1391 defm VLD1DUPq32wb : VLD1QDUPWB<{1,0,1,?}, "32">;
1393 // VLD2DUP : Vector Load (single 2-element structure to all lanes)
1394 class VLD2DUP<bits<4> op7_4, string Dt, RegisterOperand VdTy>
1395 : NLdSt<1, 0b10, 0b1101, op7_4, (outs VdTy:$Vd),
1396 (ins addrmode6dup:$Rn), IIC_VLD2dup,
1397 "vld2", Dt, "$Vd, $Rn", "", []> {
1399 let Inst{4} = Rn{4};
1400 let DecoderMethod = "DecodeVLD2DupInstruction";
1403 def VLD2DUPd8 : VLD2DUP<{0,0,0,?}, "8", VecListDPairAllLanes>;
1404 def VLD2DUPd16 : VLD2DUP<{0,1,0,?}, "16", VecListDPairAllLanes>;
1405 def VLD2DUPd32 : VLD2DUP<{1,0,0,?}, "32", VecListDPairAllLanes>;
1407 // ...with double-spaced registers
1408 def VLD2DUPd8x2 : VLD2DUP<{0,0,1,?}, "8", VecListDPairSpacedAllLanes>;
1409 def VLD2DUPd16x2 : VLD2DUP<{0,1,1,?}, "16", VecListDPairSpacedAllLanes>;
1410 def VLD2DUPd32x2 : VLD2DUP<{1,0,1,?}, "32", VecListDPairSpacedAllLanes>;
1412 // ...with address register writeback:
1413 multiclass VLD2DUPWB<bits<4> op7_4, string Dt, RegisterOperand VdTy> {
1414 def _fixed : NLdSt<1, 0b10, 0b1101, op7_4,
1415 (outs VdTy:$Vd, GPR:$wb),
1416 (ins addrmode6dup:$Rn), IIC_VLD2dupu,
1417 "vld2", Dt, "$Vd, $Rn!",
1418 "$Rn.addr = $wb", []> {
1419 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1420 let Inst{4} = Rn{4};
1421 let DecoderMethod = "DecodeVLD2DupInstruction";
1422 let AsmMatchConverter = "cvtVLDwbFixed";
1424 def _register : NLdSt<1, 0b10, 0b1101, op7_4,
1425 (outs VdTy:$Vd, GPR:$wb),
1426 (ins addrmode6dup:$Rn, rGPR:$Rm), IIC_VLD2dupu,
1427 "vld2", Dt, "$Vd, $Rn, $Rm",
1428 "$Rn.addr = $wb", []> {
1429 let Inst{4} = Rn{4};
1430 let DecoderMethod = "DecodeVLD2DupInstruction";
1431 let AsmMatchConverter = "cvtVLDwbRegister";
1435 defm VLD2DUPd8wb : VLD2DUPWB<{0,0,0,0}, "8", VecListDPairAllLanes>;
1436 defm VLD2DUPd16wb : VLD2DUPWB<{0,1,0,?}, "16", VecListDPairAllLanes>;
1437 defm VLD2DUPd32wb : VLD2DUPWB<{1,0,0,?}, "32", VecListDPairAllLanes>;
1439 defm VLD2DUPd8x2wb : VLD2DUPWB<{0,0,1,0}, "8", VecListDPairSpacedAllLanes>;
1440 defm VLD2DUPd16x2wb : VLD2DUPWB<{0,1,1,?}, "16", VecListDPairSpacedAllLanes>;
1441 defm VLD2DUPd32x2wb : VLD2DUPWB<{1,0,1,?}, "32", VecListDPairSpacedAllLanes>;
1443 // VLD3DUP : Vector Load (single 3-element structure to all lanes)
1444 class VLD3DUP<bits<4> op7_4, string Dt>
1445 : NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
1446 (ins addrmode6dup:$Rn), IIC_VLD3dup,
1447 "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn", "", []> {
1450 let DecoderMethod = "DecodeVLD3DupInstruction";
1453 def VLD3DUPd8 : VLD3DUP<{0,0,0,?}, "8">;
1454 def VLD3DUPd16 : VLD3DUP<{0,1,0,?}, "16">;
1455 def VLD3DUPd32 : VLD3DUP<{1,0,0,?}, "32">;
1457 def VLD3DUPd8Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1458 def VLD3DUPd16Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1459 def VLD3DUPd32Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1461 // ...with double-spaced registers (not used for codegen):
1462 def VLD3DUPq8 : VLD3DUP<{0,0,1,?}, "8">;
1463 def VLD3DUPq16 : VLD3DUP<{0,1,1,?}, "16">;
1464 def VLD3DUPq32 : VLD3DUP<{1,0,1,?}, "32">;
1466 // ...with address register writeback:
1467 class VLD3DUPWB<bits<4> op7_4, string Dt>
1468 : NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
1469 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD3dupu,
1470 "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn$Rm",
1471 "$Rn.addr = $wb", []> {
1473 let DecoderMethod = "DecodeVLD3DupInstruction";
1476 def VLD3DUPd8_UPD : VLD3DUPWB<{0,0,0,0}, "8">;
1477 def VLD3DUPd16_UPD : VLD3DUPWB<{0,1,0,?}, "16">;
1478 def VLD3DUPd32_UPD : VLD3DUPWB<{1,0,0,?}, "32">;
1480 def VLD3DUPq8_UPD : VLD3DUPWB<{0,0,1,0}, "8">;
1481 def VLD3DUPq16_UPD : VLD3DUPWB<{0,1,1,?}, "16">;
1482 def VLD3DUPq32_UPD : VLD3DUPWB<{1,0,1,?}, "32">;
1484 def VLD3DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1485 def VLD3DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1486 def VLD3DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1488 // VLD4DUP : Vector Load (single 4-element structure to all lanes)
1489 class VLD4DUP<bits<4> op7_4, string Dt>
1490 : NLdSt<1, 0b10, 0b1111, op7_4,
1491 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
1492 (ins addrmode6dup:$Rn), IIC_VLD4dup,
1493 "vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn", "", []> {
1495 let Inst{4} = Rn{4};
1496 let DecoderMethod = "DecodeVLD4DupInstruction";
1499 def VLD4DUPd8 : VLD4DUP<{0,0,0,?}, "8">;
1500 def VLD4DUPd16 : VLD4DUP<{0,1,0,?}, "16">;
1501 def VLD4DUPd32 : VLD4DUP<{1,?,0,?}, "32"> { let Inst{6} = Rn{5}; }
1503 def VLD4DUPd8Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1504 def VLD4DUPd16Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1505 def VLD4DUPd32Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1507 // ...with double-spaced registers (not used for codegen):
1508 def VLD4DUPq8 : VLD4DUP<{0,0,1,?}, "8">;
1509 def VLD4DUPq16 : VLD4DUP<{0,1,1,?}, "16">;
1510 def VLD4DUPq32 : VLD4DUP<{1,?,1,?}, "32"> { let Inst{6} = Rn{5}; }
1512 // ...with address register writeback:
1513 class VLD4DUPWB<bits<4> op7_4, string Dt>
1514 : NLdSt<1, 0b10, 0b1111, op7_4,
1515 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
1516 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD4dupu,
1517 "vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn$Rm",
1518 "$Rn.addr = $wb", []> {
1519 let Inst{4} = Rn{4};
1520 let DecoderMethod = "DecodeVLD4DupInstruction";
1523 def VLD4DUPd8_UPD : VLD4DUPWB<{0,0,0,0}, "8">;
1524 def VLD4DUPd16_UPD : VLD4DUPWB<{0,1,0,?}, "16">;
1525 def VLD4DUPd32_UPD : VLD4DUPWB<{1,?,0,?}, "32"> { let Inst{6} = Rn{5}; }
1527 def VLD4DUPq8_UPD : VLD4DUPWB<{0,0,1,0}, "8">;
1528 def VLD4DUPq16_UPD : VLD4DUPWB<{0,1,1,?}, "16">;
1529 def VLD4DUPq32_UPD : VLD4DUPWB<{1,?,1,?}, "32"> { let Inst{6} = Rn{5}; }
1531 def VLD4DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1532 def VLD4DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1533 def VLD4DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1535 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
1537 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
1539 // Classes for VST* pseudo-instructions with multi-register operands.
1540 // These are expanded to real instructions after register allocation.
1541 class VSTQPseudo<InstrItinClass itin>
1542 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src), itin, "">;
1543 class VSTQWBPseudo<InstrItinClass itin>
1544 : PseudoNLdSt<(outs GPR:$wb),
1545 (ins addrmode6:$addr, am6offset:$offset, QPR:$src), itin,
1546 "$addr.addr = $wb">;
1547 class VSTQWBfixedPseudo<InstrItinClass itin>
1548 : PseudoNLdSt<(outs GPR:$wb),
1549 (ins addrmode6:$addr, QPR:$src), itin,
1550 "$addr.addr = $wb">;
1551 class VSTQWBregisterPseudo<InstrItinClass itin>
1552 : PseudoNLdSt<(outs GPR:$wb),
1553 (ins addrmode6:$addr, rGPR:$offset, QPR:$src), itin,
1554 "$addr.addr = $wb">;
1555 class VSTQQPseudo<InstrItinClass itin>
1556 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src), itin, "">;
1557 class VSTQQWBPseudo<InstrItinClass itin>
1558 : PseudoNLdSt<(outs GPR:$wb),
1559 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src), itin,
1560 "$addr.addr = $wb">;
1561 class VSTQQWBfixedPseudo<InstrItinClass itin>
1562 : PseudoNLdSt<(outs GPR:$wb),
1563 (ins addrmode6:$addr, QQPR:$src), itin,
1564 "$addr.addr = $wb">;
1565 class VSTQQWBregisterPseudo<InstrItinClass itin>
1566 : PseudoNLdSt<(outs GPR:$wb),
1567 (ins addrmode6:$addr, rGPR:$offset, QQPR:$src), itin,
1568 "$addr.addr = $wb">;
1570 class VSTQQQQPseudo<InstrItinClass itin>
1571 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src), itin, "">;
1572 class VSTQQQQWBPseudo<InstrItinClass itin>
1573 : PseudoNLdSt<(outs GPR:$wb),
1574 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
1575 "$addr.addr = $wb">;
1577 // VST1 : Vector Store (multiple single elements)
1578 class VST1D<bits<4> op7_4, string Dt>
1579 : NLdSt<0,0b00,0b0111,op7_4, (outs), (ins addrmode6:$Rn, VecListOneD:$Vd),
1580 IIC_VST1, "vst1", Dt, "$Vd, $Rn", "", []> {
1582 let Inst{4} = Rn{4};
1583 let DecoderMethod = "DecodeVLDST1Instruction";
1585 class VST1Q<bits<4> op7_4, string Dt>
1586 : NLdSt<0,0b00,0b1010,op7_4, (outs), (ins addrmode6:$Rn, VecListDPair:$Vd),
1587 IIC_VST1x2, "vst1", Dt, "$Vd, $Rn", "", []> {
1589 let Inst{5-4} = Rn{5-4};
1590 let DecoderMethod = "DecodeVLDST1Instruction";
1593 def VST1d8 : VST1D<{0,0,0,?}, "8">;
1594 def VST1d16 : VST1D<{0,1,0,?}, "16">;
1595 def VST1d32 : VST1D<{1,0,0,?}, "32">;
1596 def VST1d64 : VST1D<{1,1,0,?}, "64">;
1598 def VST1q8 : VST1Q<{0,0,?,?}, "8">;
1599 def VST1q16 : VST1Q<{0,1,?,?}, "16">;
1600 def VST1q32 : VST1Q<{1,0,?,?}, "32">;
1601 def VST1q64 : VST1Q<{1,1,?,?}, "64">;
1603 // ...with address register writeback:
1604 multiclass VST1DWB<bits<4> op7_4, string Dt> {
1605 def _fixed : NLdSt<0,0b00, 0b0111,op7_4, (outs GPR:$wb),
1606 (ins addrmode6:$Rn, VecListOneD:$Vd), IIC_VLD1u,
1607 "vst1", Dt, "$Vd, $Rn!",
1608 "$Rn.addr = $wb", []> {
1609 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1610 let Inst{4} = Rn{4};
1611 let DecoderMethod = "DecodeVLDST1Instruction";
1612 let AsmMatchConverter = "cvtVSTwbFixed";
1614 def _register : NLdSt<0,0b00,0b0111,op7_4, (outs GPR:$wb),
1615 (ins addrmode6:$Rn, rGPR:$Rm, VecListOneD:$Vd),
1617 "vst1", Dt, "$Vd, $Rn, $Rm",
1618 "$Rn.addr = $wb", []> {
1619 let Inst{4} = Rn{4};
1620 let DecoderMethod = "DecodeVLDST1Instruction";
1621 let AsmMatchConverter = "cvtVSTwbRegister";
1624 multiclass VST1QWB<bits<4> op7_4, string Dt> {
1625 def _fixed : NLdSt<0,0b00,0b1010,op7_4, (outs GPR:$wb),
1626 (ins addrmode6:$Rn, VecListDPair:$Vd), IIC_VLD1x2u,
1627 "vst1", Dt, "$Vd, $Rn!",
1628 "$Rn.addr = $wb", []> {
1629 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1630 let Inst{5-4} = Rn{5-4};
1631 let DecoderMethod = "DecodeVLDST1Instruction";
1632 let AsmMatchConverter = "cvtVSTwbFixed";
1634 def _register : NLdSt<0,0b00,0b1010,op7_4, (outs GPR:$wb),
1635 (ins addrmode6:$Rn, rGPR:$Rm, VecListDPair:$Vd),
1637 "vst1", Dt, "$Vd, $Rn, $Rm",
1638 "$Rn.addr = $wb", []> {
1639 let Inst{5-4} = Rn{5-4};
1640 let DecoderMethod = "DecodeVLDST1Instruction";
1641 let AsmMatchConverter = "cvtVSTwbRegister";
1645 defm VST1d8wb : VST1DWB<{0,0,0,?}, "8">;
1646 defm VST1d16wb : VST1DWB<{0,1,0,?}, "16">;
1647 defm VST1d32wb : VST1DWB<{1,0,0,?}, "32">;
1648 defm VST1d64wb : VST1DWB<{1,1,0,?}, "64">;
1650 defm VST1q8wb : VST1QWB<{0,0,?,?}, "8">;
1651 defm VST1q16wb : VST1QWB<{0,1,?,?}, "16">;
1652 defm VST1q32wb : VST1QWB<{1,0,?,?}, "32">;
1653 defm VST1q64wb : VST1QWB<{1,1,?,?}, "64">;
1655 // ...with 3 registers
1656 class VST1D3<bits<4> op7_4, string Dt>
1657 : NLdSt<0, 0b00, 0b0110, op7_4, (outs),
1658 (ins addrmode6:$Rn, VecListThreeD:$Vd),
1659 IIC_VST1x3, "vst1", Dt, "$Vd, $Rn", "", []> {
1661 let Inst{4} = Rn{4};
1662 let DecoderMethod = "DecodeVLDST1Instruction";
1664 multiclass VST1D3WB<bits<4> op7_4, string Dt> {
1665 def _fixed : NLdSt<0,0b00,0b0110,op7_4, (outs GPR:$wb),
1666 (ins addrmode6:$Rn, VecListThreeD:$Vd), IIC_VLD1x3u,
1667 "vst1", Dt, "$Vd, $Rn!",
1668 "$Rn.addr = $wb", []> {
1669 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1670 let Inst{5-4} = Rn{5-4};
1671 let DecoderMethod = "DecodeVLDST1Instruction";
1672 let AsmMatchConverter = "cvtVSTwbFixed";
1674 def _register : NLdSt<0,0b00,0b0110,op7_4, (outs GPR:$wb),
1675 (ins addrmode6:$Rn, rGPR:$Rm, VecListThreeD:$Vd),
1677 "vst1", Dt, "$Vd, $Rn, $Rm",
1678 "$Rn.addr = $wb", []> {
1679 let Inst{5-4} = Rn{5-4};
1680 let DecoderMethod = "DecodeVLDST1Instruction";
1681 let AsmMatchConverter = "cvtVSTwbRegister";
1685 def VST1d8T : VST1D3<{0,0,0,?}, "8">;
1686 def VST1d16T : VST1D3<{0,1,0,?}, "16">;
1687 def VST1d32T : VST1D3<{1,0,0,?}, "32">;
1688 def VST1d64T : VST1D3<{1,1,0,?}, "64">;
1690 defm VST1d8Twb : VST1D3WB<{0,0,0,?}, "8">;
1691 defm VST1d16Twb : VST1D3WB<{0,1,0,?}, "16">;
1692 defm VST1d32Twb : VST1D3WB<{1,0,0,?}, "32">;
1693 defm VST1d64Twb : VST1D3WB<{1,1,0,?}, "64">;
1695 def VST1d64TPseudo : VSTQQPseudo<IIC_VST1x3>;
1696 def VST1d64TPseudoWB_fixed : VSTQQWBPseudo<IIC_VST1x3u>;
1697 def VST1d64TPseudoWB_register : VSTQQWBPseudo<IIC_VST1x3u>;
1699 // ...with 4 registers
1700 class VST1D4<bits<4> op7_4, string Dt>
1701 : NLdSt<0, 0b00, 0b0010, op7_4, (outs),
1702 (ins addrmode6:$Rn, VecListFourD:$Vd),
1703 IIC_VST1x4, "vst1", Dt, "$Vd, $Rn", "",
1706 let Inst{5-4} = Rn{5-4};
1707 let DecoderMethod = "DecodeVLDST1Instruction";
1709 multiclass VST1D4WB<bits<4> op7_4, string Dt> {
1710 def _fixed : NLdSt<0,0b00,0b0010,op7_4, (outs GPR:$wb),
1711 (ins addrmode6:$Rn, VecListFourD:$Vd), IIC_VLD1x4u,
1712 "vst1", Dt, "$Vd, $Rn!",
1713 "$Rn.addr = $wb", []> {
1714 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1715 let Inst{5-4} = Rn{5-4};
1716 let DecoderMethod = "DecodeVLDST1Instruction";
1717 let AsmMatchConverter = "cvtVSTwbFixed";
1719 def _register : NLdSt<0,0b00,0b0010,op7_4, (outs GPR:$wb),
1720 (ins addrmode6:$Rn, rGPR:$Rm, VecListFourD:$Vd),
1722 "vst1", Dt, "$Vd, $Rn, $Rm",
1723 "$Rn.addr = $wb", []> {
1724 let Inst{5-4} = Rn{5-4};
1725 let DecoderMethod = "DecodeVLDST1Instruction";
1726 let AsmMatchConverter = "cvtVSTwbRegister";
1730 def VST1d8Q : VST1D4<{0,0,?,?}, "8">;
1731 def VST1d16Q : VST1D4<{0,1,?,?}, "16">;
1732 def VST1d32Q : VST1D4<{1,0,?,?}, "32">;
1733 def VST1d64Q : VST1D4<{1,1,?,?}, "64">;
1735 defm VST1d8Qwb : VST1D4WB<{0,0,?,?}, "8">;
1736 defm VST1d16Qwb : VST1D4WB<{0,1,?,?}, "16">;
1737 defm VST1d32Qwb : VST1D4WB<{1,0,?,?}, "32">;
1738 defm VST1d64Qwb : VST1D4WB<{1,1,?,?}, "64">;
1740 def VST1d64QPseudo : VSTQQPseudo<IIC_VST1x4>;
1741 def VST1d64QPseudoWB_fixed : VSTQQWBPseudo<IIC_VST1x4u>;
1742 def VST1d64QPseudoWB_register : VSTQQWBPseudo<IIC_VST1x4u>;
1744 // VST2 : Vector Store (multiple 2-element structures)
1745 class VST2<bits<4> op11_8, bits<4> op7_4, string Dt, RegisterOperand VdTy,
1746 InstrItinClass itin>
1747 : NLdSt<0, 0b00, op11_8, op7_4, (outs), (ins addrmode6:$Rn, VdTy:$Vd),
1748 itin, "vst2", Dt, "$Vd, $Rn", "", []> {
1750 let Inst{5-4} = Rn{5-4};
1751 let DecoderMethod = "DecodeVLDST2Instruction";
1754 def VST2d8 : VST2<0b1000, {0,0,?,?}, "8", VecListDPair, IIC_VST2>;
1755 def VST2d16 : VST2<0b1000, {0,1,?,?}, "16", VecListDPair, IIC_VST2>;
1756 def VST2d32 : VST2<0b1000, {1,0,?,?}, "32", VecListDPair, IIC_VST2>;
1758 def VST2q8 : VST2<0b0011, {0,0,?,?}, "8", VecListFourD, IIC_VST2x2>;
1759 def VST2q16 : VST2<0b0011, {0,1,?,?}, "16", VecListFourD, IIC_VST2x2>;
1760 def VST2q32 : VST2<0b0011, {1,0,?,?}, "32", VecListFourD, IIC_VST2x2>;
1762 def VST2q8Pseudo : VSTQQPseudo<IIC_VST2x2>;
1763 def VST2q16Pseudo : VSTQQPseudo<IIC_VST2x2>;
1764 def VST2q32Pseudo : VSTQQPseudo<IIC_VST2x2>;
1766 // ...with address register writeback:
1767 multiclass VST2DWB<bits<4> op11_8, bits<4> op7_4, string Dt,
1768 RegisterOperand VdTy> {
1769 def _fixed : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1770 (ins addrmode6:$Rn, VdTy:$Vd), IIC_VLD1u,
1771 "vst2", Dt, "$Vd, $Rn!",
1772 "$Rn.addr = $wb", []> {
1773 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1774 let Inst{5-4} = Rn{5-4};
1775 let DecoderMethod = "DecodeVLDST2Instruction";
1776 let AsmMatchConverter = "cvtVSTwbFixed";
1778 def _register : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1779 (ins addrmode6:$Rn, rGPR:$Rm, VdTy:$Vd), IIC_VLD1u,
1780 "vst2", Dt, "$Vd, $Rn, $Rm",
1781 "$Rn.addr = $wb", []> {
1782 let Inst{5-4} = Rn{5-4};
1783 let DecoderMethod = "DecodeVLDST2Instruction";
1784 let AsmMatchConverter = "cvtVSTwbRegister";
1787 multiclass VST2QWB<bits<4> op7_4, string Dt> {
1788 def _fixed : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
1789 (ins addrmode6:$Rn, VecListFourD:$Vd), IIC_VLD1u,
1790 "vst2", Dt, "$Vd, $Rn!",
1791 "$Rn.addr = $wb", []> {
1792 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1793 let Inst{5-4} = Rn{5-4};
1794 let DecoderMethod = "DecodeVLDST2Instruction";
1795 let AsmMatchConverter = "cvtVSTwbFixed";
1797 def _register : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
1798 (ins addrmode6:$Rn, rGPR:$Rm, VecListFourD:$Vd),
1800 "vst2", Dt, "$Vd, $Rn, $Rm",
1801 "$Rn.addr = $wb", []> {
1802 let Inst{5-4} = Rn{5-4};
1803 let DecoderMethod = "DecodeVLDST2Instruction";
1804 let AsmMatchConverter = "cvtVSTwbRegister";
1808 defm VST2d8wb : VST2DWB<0b1000, {0,0,?,?}, "8", VecListDPair>;
1809 defm VST2d16wb : VST2DWB<0b1000, {0,1,?,?}, "16", VecListDPair>;
1810 defm VST2d32wb : VST2DWB<0b1000, {1,0,?,?}, "32", VecListDPair>;
1812 defm VST2q8wb : VST2QWB<{0,0,?,?}, "8">;
1813 defm VST2q16wb : VST2QWB<{0,1,?,?}, "16">;
1814 defm VST2q32wb : VST2QWB<{1,0,?,?}, "32">;
1816 def VST2q8PseudoWB_fixed : VSTQQWBfixedPseudo<IIC_VST2x2u>;
1817 def VST2q16PseudoWB_fixed : VSTQQWBfixedPseudo<IIC_VST2x2u>;
1818 def VST2q32PseudoWB_fixed : VSTQQWBfixedPseudo<IIC_VST2x2u>;
1819 def VST2q8PseudoWB_register : VSTQQWBregisterPseudo<IIC_VST2x2u>;
1820 def VST2q16PseudoWB_register : VSTQQWBregisterPseudo<IIC_VST2x2u>;
1821 def VST2q32PseudoWB_register : VSTQQWBregisterPseudo<IIC_VST2x2u>;
1823 // ...with double-spaced registers
1824 def VST2b8 : VST2<0b1001, {0,0,?,?}, "8", VecListDPairSpaced, IIC_VST2>;
1825 def VST2b16 : VST2<0b1001, {0,1,?,?}, "16", VecListDPairSpaced, IIC_VST2>;
1826 def VST2b32 : VST2<0b1001, {1,0,?,?}, "32", VecListDPairSpaced, IIC_VST2>;
1827 defm VST2b8wb : VST2DWB<0b1001, {0,0,?,?}, "8", VecListDPairSpaced>;
1828 defm VST2b16wb : VST2DWB<0b1001, {0,1,?,?}, "16", VecListDPairSpaced>;
1829 defm VST2b32wb : VST2DWB<0b1001, {1,0,?,?}, "32", VecListDPairSpaced>;
1831 // VST3 : Vector Store (multiple 3-element structures)
1832 class VST3D<bits<4> op11_8, bits<4> op7_4, string Dt>
1833 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1834 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3,
1835 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn", "", []> {
1837 let Inst{4} = Rn{4};
1838 let DecoderMethod = "DecodeVLDST3Instruction";
1841 def VST3d8 : VST3D<0b0100, {0,0,0,?}, "8">;
1842 def VST3d16 : VST3D<0b0100, {0,1,0,?}, "16">;
1843 def VST3d32 : VST3D<0b0100, {1,0,0,?}, "32">;
1845 def VST3d8Pseudo : VSTQQPseudo<IIC_VST3>;
1846 def VST3d16Pseudo : VSTQQPseudo<IIC_VST3>;
1847 def VST3d32Pseudo : VSTQQPseudo<IIC_VST3>;
1849 // ...with address register writeback:
1850 class VST3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1851 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1852 (ins addrmode6:$Rn, am6offset:$Rm,
1853 DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3u,
1854 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn$Rm",
1855 "$Rn.addr = $wb", []> {
1856 let Inst{4} = Rn{4};
1857 let DecoderMethod = "DecodeVLDST3Instruction";
1860 def VST3d8_UPD : VST3DWB<0b0100, {0,0,0,?}, "8">;
1861 def VST3d16_UPD : VST3DWB<0b0100, {0,1,0,?}, "16">;
1862 def VST3d32_UPD : VST3DWB<0b0100, {1,0,0,?}, "32">;
1864 def VST3d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1865 def VST3d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1866 def VST3d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1868 // ...with double-spaced registers:
1869 def VST3q8 : VST3D<0b0101, {0,0,0,?}, "8">;
1870 def VST3q16 : VST3D<0b0101, {0,1,0,?}, "16">;
1871 def VST3q32 : VST3D<0b0101, {1,0,0,?}, "32">;
1872 def VST3q8_UPD : VST3DWB<0b0101, {0,0,0,?}, "8">;
1873 def VST3q16_UPD : VST3DWB<0b0101, {0,1,0,?}, "16">;
1874 def VST3q32_UPD : VST3DWB<0b0101, {1,0,0,?}, "32">;
1876 def VST3q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1877 def VST3q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1878 def VST3q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1880 // ...alternate versions to be allocated odd register numbers:
1881 def VST3q8oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1882 def VST3q16oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1883 def VST3q32oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1885 def VST3q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1886 def VST3q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1887 def VST3q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1889 // VST4 : Vector Store (multiple 4-element structures)
1890 class VST4D<bits<4> op11_8, bits<4> op7_4, string Dt>
1891 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1892 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1893 IIC_VST4, "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn",
1896 let Inst{5-4} = Rn{5-4};
1897 let DecoderMethod = "DecodeVLDST4Instruction";
1900 def VST4d8 : VST4D<0b0000, {0,0,?,?}, "8">;
1901 def VST4d16 : VST4D<0b0000, {0,1,?,?}, "16">;
1902 def VST4d32 : VST4D<0b0000, {1,0,?,?}, "32">;
1904 def VST4d8Pseudo : VSTQQPseudo<IIC_VST4>;
1905 def VST4d16Pseudo : VSTQQPseudo<IIC_VST4>;
1906 def VST4d32Pseudo : VSTQQPseudo<IIC_VST4>;
1908 // ...with address register writeback:
1909 class VST4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1910 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1911 (ins addrmode6:$Rn, am6offset:$Rm,
1912 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST4u,
1913 "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1914 "$Rn.addr = $wb", []> {
1915 let Inst{5-4} = Rn{5-4};
1916 let DecoderMethod = "DecodeVLDST4Instruction";
1919 def VST4d8_UPD : VST4DWB<0b0000, {0,0,?,?}, "8">;
1920 def VST4d16_UPD : VST4DWB<0b0000, {0,1,?,?}, "16">;
1921 def VST4d32_UPD : VST4DWB<0b0000, {1,0,?,?}, "32">;
1923 def VST4d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1924 def VST4d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1925 def VST4d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1927 // ...with double-spaced registers:
1928 def VST4q8 : VST4D<0b0001, {0,0,?,?}, "8">;
1929 def VST4q16 : VST4D<0b0001, {0,1,?,?}, "16">;
1930 def VST4q32 : VST4D<0b0001, {1,0,?,?}, "32">;
1931 def VST4q8_UPD : VST4DWB<0b0001, {0,0,?,?}, "8">;
1932 def VST4q16_UPD : VST4DWB<0b0001, {0,1,?,?}, "16">;
1933 def VST4q32_UPD : VST4DWB<0b0001, {1,0,?,?}, "32">;
1935 def VST4q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1936 def VST4q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1937 def VST4q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1939 // ...alternate versions to be allocated odd register numbers:
1940 def VST4q8oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1941 def VST4q16oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1942 def VST4q32oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1944 def VST4q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1945 def VST4q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1946 def VST4q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1948 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
1950 // Classes for VST*LN pseudo-instructions with multi-register operands.
1951 // These are expanded to real instructions after register allocation.
1952 class VSTQLNPseudo<InstrItinClass itin>
1953 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
1955 class VSTQLNWBPseudo<InstrItinClass itin>
1956 : PseudoNLdSt<(outs GPR:$wb),
1957 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
1958 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1959 class VSTQQLNPseudo<InstrItinClass itin>
1960 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
1962 class VSTQQLNWBPseudo<InstrItinClass itin>
1963 : PseudoNLdSt<(outs GPR:$wb),
1964 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
1965 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1966 class VSTQQQQLNPseudo<InstrItinClass itin>
1967 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
1969 class VSTQQQQLNWBPseudo<InstrItinClass itin>
1970 : PseudoNLdSt<(outs GPR:$wb),
1971 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
1972 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1974 // VST1LN : Vector Store (single element from one lane)
1975 class VST1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1976 PatFrag StoreOp, SDNode ExtractOp, Operand AddrMode>
1977 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1978 (ins AddrMode:$Rn, DPR:$Vd, nohash_imm:$lane),
1979 IIC_VST1ln, "vst1", Dt, "\\{$Vd[$lane]\\}, $Rn", "",
1980 [(StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane), AddrMode:$Rn)]> {
1982 let DecoderMethod = "DecodeVST1LN";
1984 class VST1QLNPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
1985 : VSTQLNPseudo<IIC_VST1ln> {
1986 let Pattern = [(StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
1990 def VST1LNd8 : VST1LN<0b0000, {?,?,?,0}, "8", v8i8, truncstorei8,
1991 NEONvgetlaneu, addrmode6> {
1992 let Inst{7-5} = lane{2-0};
1994 def VST1LNd16 : VST1LN<0b0100, {?,?,0,?}, "16", v4i16, truncstorei16,
1995 NEONvgetlaneu, addrmode6> {
1996 let Inst{7-6} = lane{1-0};
1997 let Inst{4} = Rn{4};
2000 def VST1LNd32 : VST1LN<0b1000, {?,0,?,?}, "32", v2i32, store, extractelt,
2002 let Inst{7} = lane{0};
2003 let Inst{5-4} = Rn{5-4};
2006 def VST1LNq8Pseudo : VST1QLNPseudo<v16i8, truncstorei8, NEONvgetlaneu>;
2007 def VST1LNq16Pseudo : VST1QLNPseudo<v8i16, truncstorei16, NEONvgetlaneu>;
2008 def VST1LNq32Pseudo : VST1QLNPseudo<v4i32, store, extractelt>;
2010 def : Pat<(store (extractelt (v2f32 DPR:$src), imm:$lane), addrmode6:$addr),
2011 (VST1LNd32 addrmode6:$addr, DPR:$src, imm:$lane)>;
2012 def : Pat<(store (extractelt (v4f32 QPR:$src), imm:$lane), addrmode6:$addr),
2013 (VST1LNq32Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
2015 // ...with address register writeback:
2016 class VST1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
2017 PatFrag StoreOp, SDNode ExtractOp, Operand AdrMode>
2018 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
2019 (ins AdrMode:$Rn, am6offset:$Rm,
2020 DPR:$Vd, nohash_imm:$lane), IIC_VST1lnu, "vst1", Dt,
2021 "\\{$Vd[$lane]\\}, $Rn$Rm",
2023 [(set GPR:$wb, (StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane),
2024 AdrMode:$Rn, am6offset:$Rm))]> {
2025 let DecoderMethod = "DecodeVST1LN";
2027 class VST1QLNWBPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
2028 : VSTQLNWBPseudo<IIC_VST1lnu> {
2029 let Pattern = [(set GPR:$wb, (StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
2030 addrmode6:$addr, am6offset:$offset))];
2033 def VST1LNd8_UPD : VST1LNWB<0b0000, {?,?,?,0}, "8", v8i8, post_truncsti8,
2034 NEONvgetlaneu, addrmode6> {
2035 let Inst{7-5} = lane{2-0};
2037 def VST1LNd16_UPD : VST1LNWB<0b0100, {?,?,0,?}, "16", v4i16, post_truncsti16,
2038 NEONvgetlaneu, addrmode6> {
2039 let Inst{7-6} = lane{1-0};
2040 let Inst{4} = Rn{4};
2042 def VST1LNd32_UPD : VST1LNWB<0b1000, {?,0,?,?}, "32", v2i32, post_store,
2043 extractelt, addrmode6oneL32> {
2044 let Inst{7} = lane{0};
2045 let Inst{5-4} = Rn{5-4};
2048 def VST1LNq8Pseudo_UPD : VST1QLNWBPseudo<v16i8, post_truncsti8, NEONvgetlaneu>;
2049 def VST1LNq16Pseudo_UPD : VST1QLNWBPseudo<v8i16, post_truncsti16,NEONvgetlaneu>;
2050 def VST1LNq32Pseudo_UPD : VST1QLNWBPseudo<v4i32, post_store, extractelt>;
2052 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
2054 // VST2LN : Vector Store (single 2-element structure from one lane)
2055 class VST2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
2056 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
2057 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, nohash_imm:$lane),
2058 IIC_VST2ln, "vst2", Dt, "\\{$Vd[$lane], $src2[$lane]\\}, $Rn",
2061 let Inst{4} = Rn{4};
2062 let DecoderMethod = "DecodeVST2LN";
2065 def VST2LNd8 : VST2LN<0b0001, {?,?,?,?}, "8"> {
2066 let Inst{7-5} = lane{2-0};
2068 def VST2LNd16 : VST2LN<0b0101, {?,?,0,?}, "16"> {
2069 let Inst{7-6} = lane{1-0};
2071 def VST2LNd32 : VST2LN<0b1001, {?,0,0,?}, "32"> {
2072 let Inst{7} = lane{0};
2075 def VST2LNd8Pseudo : VSTQLNPseudo<IIC_VST2ln>;
2076 def VST2LNd16Pseudo : VSTQLNPseudo<IIC_VST2ln>;
2077 def VST2LNd32Pseudo : VSTQLNPseudo<IIC_VST2ln>;
2079 // ...with double-spaced registers:
2080 def VST2LNq16 : VST2LN<0b0101, {?,?,1,?}, "16"> {
2081 let Inst{7-6} = lane{1-0};
2082 let Inst{4} = Rn{4};
2084 def VST2LNq32 : VST2LN<0b1001, {?,1,0,?}, "32"> {
2085 let Inst{7} = lane{0};
2086 let Inst{4} = Rn{4};
2089 def VST2LNq16Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
2090 def VST2LNq32Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
2092 // ...with address register writeback:
2093 class VST2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
2094 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
2095 (ins addrmode6:$Rn, am6offset:$Rm,
2096 DPR:$Vd, DPR:$src2, nohash_imm:$lane), IIC_VST2lnu, "vst2", Dt,
2097 "\\{$Vd[$lane], $src2[$lane]\\}, $Rn$Rm",
2098 "$Rn.addr = $wb", []> {
2099 let Inst{4} = Rn{4};
2100 let DecoderMethod = "DecodeVST2LN";
2103 def VST2LNd8_UPD : VST2LNWB<0b0001, {?,?,?,?}, "8"> {
2104 let Inst{7-5} = lane{2-0};
2106 def VST2LNd16_UPD : VST2LNWB<0b0101, {?,?,0,?}, "16"> {
2107 let Inst{7-6} = lane{1-0};
2109 def VST2LNd32_UPD : VST2LNWB<0b1001, {?,0,0,?}, "32"> {
2110 let Inst{7} = lane{0};
2113 def VST2LNd8Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
2114 def VST2LNd16Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
2115 def VST2LNd32Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
2117 def VST2LNq16_UPD : VST2LNWB<0b0101, {?,?,1,?}, "16"> {
2118 let Inst{7-6} = lane{1-0};
2120 def VST2LNq32_UPD : VST2LNWB<0b1001, {?,1,0,?}, "32"> {
2121 let Inst{7} = lane{0};
2124 def VST2LNq16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
2125 def VST2LNq32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
2127 // VST3LN : Vector Store (single 3-element structure from one lane)
2128 class VST3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
2129 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
2130 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3,
2131 nohash_imm:$lane), IIC_VST3ln, "vst3", Dt,
2132 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn", "", []> {
2134 let DecoderMethod = "DecodeVST3LN";
2137 def VST3LNd8 : VST3LN<0b0010, {?,?,?,0}, "8"> {
2138 let Inst{7-5} = lane{2-0};
2140 def VST3LNd16 : VST3LN<0b0110, {?,?,0,0}, "16"> {
2141 let Inst{7-6} = lane{1-0};
2143 def VST3LNd32 : VST3LN<0b1010, {?,0,0,0}, "32"> {
2144 let Inst{7} = lane{0};
2147 def VST3LNd8Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
2148 def VST3LNd16Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
2149 def VST3LNd32Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
2151 // ...with double-spaced registers:
2152 def VST3LNq16 : VST3LN<0b0110, {?,?,1,0}, "16"> {
2153 let Inst{7-6} = lane{1-0};
2155 def VST3LNq32 : VST3LN<0b1010, {?,1,0,0}, "32"> {
2156 let Inst{7} = lane{0};
2159 def VST3LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
2160 def VST3LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
2162 // ...with address register writeback:
2163 class VST3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
2164 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
2165 (ins addrmode6:$Rn, am6offset:$Rm,
2166 DPR:$Vd, DPR:$src2, DPR:$src3, nohash_imm:$lane),
2167 IIC_VST3lnu, "vst3", Dt,
2168 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn$Rm",
2169 "$Rn.addr = $wb", []> {
2170 let DecoderMethod = "DecodeVST3LN";
2173 def VST3LNd8_UPD : VST3LNWB<0b0010, {?,?,?,0}, "8"> {
2174 let Inst{7-5} = lane{2-0};
2176 def VST3LNd16_UPD : VST3LNWB<0b0110, {?,?,0,0}, "16"> {
2177 let Inst{7-6} = lane{1-0};
2179 def VST3LNd32_UPD : VST3LNWB<0b1010, {?,0,0,0}, "32"> {
2180 let Inst{7} = lane{0};
2183 def VST3LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
2184 def VST3LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
2185 def VST3LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
2187 def VST3LNq16_UPD : VST3LNWB<0b0110, {?,?,1,0}, "16"> {
2188 let Inst{7-6} = lane{1-0};
2190 def VST3LNq32_UPD : VST3LNWB<0b1010, {?,1,0,0}, "32"> {
2191 let Inst{7} = lane{0};
2194 def VST3LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
2195 def VST3LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
2197 // VST4LN : Vector Store (single 4-element structure from one lane)
2198 class VST4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
2199 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
2200 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4,
2201 nohash_imm:$lane), IIC_VST4ln, "vst4", Dt,
2202 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn",
2205 let Inst{4} = Rn{4};
2206 let DecoderMethod = "DecodeVST4LN";
2209 def VST4LNd8 : VST4LN<0b0011, {?,?,?,?}, "8"> {
2210 let Inst{7-5} = lane{2-0};
2212 def VST4LNd16 : VST4LN<0b0111, {?,?,0,?}, "16"> {
2213 let Inst{7-6} = lane{1-0};
2215 def VST4LNd32 : VST4LN<0b1011, {?,0,?,?}, "32"> {
2216 let Inst{7} = lane{0};
2217 let Inst{5} = Rn{5};
2220 def VST4LNd8Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
2221 def VST4LNd16Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
2222 def VST4LNd32Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
2224 // ...with double-spaced registers:
2225 def VST4LNq16 : VST4LN<0b0111, {?,?,1,?}, "16"> {
2226 let Inst{7-6} = lane{1-0};
2228 def VST4LNq32 : VST4LN<0b1011, {?,1,?,?}, "32"> {
2229 let Inst{7} = lane{0};
2230 let Inst{5} = Rn{5};
2233 def VST4LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
2234 def VST4LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
2236 // ...with address register writeback:
2237 class VST4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
2238 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
2239 (ins addrmode6:$Rn, am6offset:$Rm,
2240 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
2241 IIC_VST4lnu, "vst4", Dt,
2242 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn$Rm",
2243 "$Rn.addr = $wb", []> {
2244 let Inst{4} = Rn{4};
2245 let DecoderMethod = "DecodeVST4LN";
2248 def VST4LNd8_UPD : VST4LNWB<0b0011, {?,?,?,?}, "8"> {
2249 let Inst{7-5} = lane{2-0};
2251 def VST4LNd16_UPD : VST4LNWB<0b0111, {?,?,0,?}, "16"> {
2252 let Inst{7-6} = lane{1-0};
2254 def VST4LNd32_UPD : VST4LNWB<0b1011, {?,0,?,?}, "32"> {
2255 let Inst{7} = lane{0};
2256 let Inst{5} = Rn{5};
2259 def VST4LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
2260 def VST4LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
2261 def VST4LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
2263 def VST4LNq16_UPD : VST4LNWB<0b0111, {?,?,1,?}, "16"> {
2264 let Inst{7-6} = lane{1-0};
2266 def VST4LNq32_UPD : VST4LNWB<0b1011, {?,1,?,?}, "32"> {
2267 let Inst{7} = lane{0};
2268 let Inst{5} = Rn{5};
2271 def VST4LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
2272 def VST4LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
2274 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
2276 // Use vld1/vst1 for unaligned f64 load / store
2277 def : Pat<(f64 (hword_alignedload addrmode6:$addr)),
2278 (VLD1d16 addrmode6:$addr)>, Requires<[IsLE]>;
2279 def : Pat<(hword_alignedstore (f64 DPR:$value), addrmode6:$addr),
2280 (VST1d16 addrmode6:$addr, DPR:$value)>, Requires<[IsLE]>;
2281 def : Pat<(f64 (byte_alignedload addrmode6:$addr)),
2282 (VLD1d8 addrmode6:$addr)>, Requires<[IsLE]>;
2283 def : Pat<(byte_alignedstore (f64 DPR:$value), addrmode6:$addr),
2284 (VST1d8 addrmode6:$addr, DPR:$value)>, Requires<[IsLE]>;
2285 def : Pat<(f64 (non_word_alignedload addrmode6:$addr)),
2286 (VLD1d64 addrmode6:$addr)>, Requires<[IsBE]>;
2287 def : Pat<(non_word_alignedstore (f64 DPR:$value), addrmode6:$addr),
2288 (VST1d64 addrmode6:$addr, DPR:$value)>, Requires<[IsBE]>;
2290 // Use vld1/vst1 for Q and QQ. Also use them for unaligned v2f64
2291 // load / store if it's legal.
2292 def : Pat<(v2f64 (dword_alignedload addrmode6:$addr)),
2293 (VLD1q64 addrmode6:$addr)>;
2294 def : Pat<(dword_alignedstore (v2f64 QPR:$value), addrmode6:$addr),
2295 (VST1q64 addrmode6:$addr, QPR:$value)>;
2296 def : Pat<(v2f64 (word_alignedload addrmode6:$addr)),
2297 (VLD1q32 addrmode6:$addr)>;
2298 def : Pat<(word_alignedstore (v2f64 QPR:$value), addrmode6:$addr),
2299 (VST1q32 addrmode6:$addr, QPR:$value)>;
2300 def : Pat<(v2f64 (hword_alignedload addrmode6:$addr)),
2301 (VLD1q16 addrmode6:$addr)>, Requires<[IsLE]>;
2302 def : Pat<(hword_alignedstore (v2f64 QPR:$value), addrmode6:$addr),
2303 (VST1q16 addrmode6:$addr, QPR:$value)>, Requires<[IsLE]>;
2304 def : Pat<(v2f64 (byte_alignedload addrmode6:$addr)),
2305 (VLD1q8 addrmode6:$addr)>, Requires<[IsLE]>;
2306 def : Pat<(byte_alignedstore (v2f64 QPR:$value), addrmode6:$addr),
2307 (VST1q8 addrmode6:$addr, QPR:$value)>, Requires<[IsLE]>;
2309 //===----------------------------------------------------------------------===//
2310 // NEON pattern fragments
2311 //===----------------------------------------------------------------------===//
2313 // Extract D sub-registers of Q registers.
2314 def DSubReg_i8_reg : SDNodeXForm<imm, [{
2315 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2316 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/8, MVT::i32);
2318 def DSubReg_i16_reg : SDNodeXForm<imm, [{
2319 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2320 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/4, MVT::i32);
2322 def DSubReg_i32_reg : SDNodeXForm<imm, [{
2323 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2324 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/2, MVT::i32);
2326 def DSubReg_f64_reg : SDNodeXForm<imm, [{
2327 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2328 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue(), MVT::i32);
2331 // Extract S sub-registers of Q/D registers.
2332 def SSubReg_f32_reg : SDNodeXForm<imm, [{
2333 assert(ARM::ssub_3 == ARM::ssub_0+3 && "Unexpected subreg numbering");
2334 return CurDAG->getTargetConstant(ARM::ssub_0 + N->getZExtValue(), MVT::i32);
2337 // Translate lane numbers from Q registers to D subregs.
2338 def SubReg_i8_lane : SDNodeXForm<imm, [{
2339 return CurDAG->getTargetConstant(N->getZExtValue() & 7, MVT::i32);
2341 def SubReg_i16_lane : SDNodeXForm<imm, [{
2342 return CurDAG->getTargetConstant(N->getZExtValue() & 3, MVT::i32);
2344 def SubReg_i32_lane : SDNodeXForm<imm, [{
2345 return CurDAG->getTargetConstant(N->getZExtValue() & 1, MVT::i32);
2348 //===----------------------------------------------------------------------===//
2349 // Instruction Classes
2350 //===----------------------------------------------------------------------===//
2352 // Basic 2-register operations: double- and quad-register.
2353 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2354 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
2355 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
2356 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2357 (ins DPR:$Vm), IIC_VUNAD, OpcodeStr, Dt,"$Vd, $Vm", "",
2358 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm))))]>;
2359 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2360 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
2361 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
2362 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2363 (ins QPR:$Vm), IIC_VUNAQ, OpcodeStr, Dt,"$Vd, $Vm", "",
2364 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vm))))]>;
2366 // Basic 2-register intrinsics, both double- and quad-register.
2367 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2368 bits<2> op17_16, bits<5> op11_7, bit op4,
2369 InstrItinClass itin, string OpcodeStr, string Dt,
2370 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2371 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2372 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2373 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
2374 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2375 bits<2> op17_16, bits<5> op11_7, bit op4,
2376 InstrItinClass itin, string OpcodeStr, string Dt,
2377 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2378 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2379 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2380 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
2382 // Narrow 2-register operations.
2383 class N2VN<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2384 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2385 InstrItinClass itin, string OpcodeStr, string Dt,
2386 ValueType TyD, ValueType TyQ, SDNode OpNode>
2387 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$Vd),
2388 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2389 [(set DPR:$Vd, (TyD (OpNode (TyQ QPR:$Vm))))]>;
2391 // Narrow 2-register intrinsics.
2392 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2393 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2394 InstrItinClass itin, string OpcodeStr, string Dt,
2395 ValueType TyD, ValueType TyQ, SDPatternOperator IntOp>
2396 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$Vd),
2397 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2398 [(set DPR:$Vd, (TyD (IntOp (TyQ QPR:$Vm))))]>;
2400 // Long 2-register operations (currently only used for VMOVL).
2401 class N2VL<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2402 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2403 InstrItinClass itin, string OpcodeStr, string Dt,
2404 ValueType TyQ, ValueType TyD, SDNode OpNode>
2405 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$Vd),
2406 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2407 [(set QPR:$Vd, (TyQ (OpNode (TyD DPR:$Vm))))]>;
2409 // Long 2-register intrinsics.
2410 class N2VLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2411 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2412 InstrItinClass itin, string OpcodeStr, string Dt,
2413 ValueType TyQ, ValueType TyD, SDPatternOperator IntOp>
2414 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$Vd),
2415 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2416 [(set QPR:$Vd, (TyQ (IntOp (TyD DPR:$Vm))))]>;
2418 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
2419 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
2420 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$Vd, DPR:$Vm),
2421 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
2422 OpcodeStr, Dt, "$Vd, $Vm",
2423 "$src1 = $Vd, $src2 = $Vm", []>;
2424 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
2425 InstrItinClass itin, string OpcodeStr, string Dt>
2426 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$Vd, QPR:$Vm),
2427 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$Vd, $Vm",
2428 "$src1 = $Vd, $src2 = $Vm", []>;
2430 // Basic 3-register operations: double- and quad-register.
2431 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2432 InstrItinClass itin, string OpcodeStr, string Dt,
2433 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2434 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2435 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2436 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2437 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
2438 // All of these have a two-operand InstAlias.
2439 let TwoOperandAliasConstraint = "$Vn = $Vd";
2440 let isCommutable = Commutable;
2442 // Same as N3VD but no data type.
2443 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2444 InstrItinClass itin, string OpcodeStr,
2445 ValueType ResTy, ValueType OpTy,
2446 SDNode OpNode, bit Commutable>
2447 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
2448 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2449 OpcodeStr, "$Vd, $Vn, $Vm", "",
2450 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>{
2451 // All of these have a two-operand InstAlias.
2452 let TwoOperandAliasConstraint = "$Vn = $Vd";
2453 let isCommutable = Commutable;
2456 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
2457 InstrItinClass itin, string OpcodeStr, string Dt,
2458 ValueType Ty, SDNode ShOp>
2459 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2460 (outs DPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2461 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2463 (Ty (ShOp (Ty DPR:$Vn),
2464 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),imm:$lane)))))]> {
2465 // All of these have a two-operand InstAlias.
2466 let TwoOperandAliasConstraint = "$Vn = $Vd";
2467 let isCommutable = 0;
2469 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
2470 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
2471 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2472 (outs DPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2473 NVMulSLFrm, IIC_VMULi16D, OpcodeStr, Dt,"$Vd, $Vn, $Vm$lane","",
2475 (Ty (ShOp (Ty DPR:$Vn),
2476 (Ty (NEONvduplane (Ty DPR_8:$Vm), imm:$lane)))))]> {
2477 // All of these have a two-operand InstAlias.
2478 let TwoOperandAliasConstraint = "$Vn = $Vd";
2479 let isCommutable = 0;
2482 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2483 InstrItinClass itin, string OpcodeStr, string Dt,
2484 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2485 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2486 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2487 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2488 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
2489 // All of these have a two-operand InstAlias.
2490 let TwoOperandAliasConstraint = "$Vn = $Vd";
2491 let isCommutable = Commutable;
2493 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2494 InstrItinClass itin, string OpcodeStr,
2495 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2496 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
2497 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2498 OpcodeStr, "$Vd, $Vn, $Vm", "",
2499 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>{
2500 // All of these have a two-operand InstAlias.
2501 let TwoOperandAliasConstraint = "$Vn = $Vd";
2502 let isCommutable = Commutable;
2504 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
2505 InstrItinClass itin, string OpcodeStr, string Dt,
2506 ValueType ResTy, ValueType OpTy, SDNode ShOp>
2507 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2508 (outs QPR:$Vd), (ins QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2509 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2510 [(set (ResTy QPR:$Vd),
2511 (ResTy (ShOp (ResTy QPR:$Vn),
2512 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2514 // All of these have a two-operand InstAlias.
2515 let TwoOperandAliasConstraint = "$Vn = $Vd";
2516 let isCommutable = 0;
2518 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
2519 ValueType ResTy, ValueType OpTy, SDNode ShOp>
2520 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2521 (outs QPR:$Vd), (ins QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2522 NVMulSLFrm, IIC_VMULi16Q, OpcodeStr, Dt,"$Vd, $Vn, $Vm$lane", "",
2523 [(set (ResTy QPR:$Vd),
2524 (ResTy (ShOp (ResTy QPR:$Vn),
2525 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2527 // All of these have a two-operand InstAlias.
2528 let TwoOperandAliasConstraint = "$Vn = $Vd";
2529 let isCommutable = 0;
2532 // Basic 3-register intrinsics, both double- and quad-register.
2533 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2534 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2535 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp, bit Commutable>
2536 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2537 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), f, itin,
2538 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2539 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
2540 // All of these have a two-operand InstAlias.
2541 let TwoOperandAliasConstraint = "$Vn = $Vd";
2542 let isCommutable = Commutable;
2545 class N3VDIntnp<bits<5> op27_23, bits<2> op21_20, bits<4> op11_8, bit op6,
2546 bit op4, Format f, InstrItinClass itin, string OpcodeStr,
2547 string Dt, ValueType ResTy, ValueType OpTy,
2548 SDPatternOperator IntOp, bit Commutable>
2549 : N3Vnp<op27_23, op21_20, op11_8, op6, op4,
2550 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), f, itin, OpcodeStr, Dt,
2551 ResTy, OpTy, IntOp, Commutable,
2552 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>;
2554 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2555 string OpcodeStr, string Dt, ValueType Ty, SDPatternOperator IntOp>
2556 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2557 (outs DPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2558 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2560 (Ty (IntOp (Ty DPR:$Vn),
2561 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),
2563 let isCommutable = 0;
2566 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2567 string OpcodeStr, string Dt, ValueType Ty, SDPatternOperator IntOp>
2568 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2569 (outs DPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2570 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2572 (Ty (IntOp (Ty DPR:$Vn),
2573 (Ty (NEONvduplane (Ty DPR_8:$Vm), imm:$lane)))))]> {
2574 let isCommutable = 0;
2576 class N3VDIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2577 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2578 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2579 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2580 (outs DPR:$Vd), (ins DPR:$Vm, DPR:$Vn), f, itin,
2581 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
2582 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (OpTy DPR:$Vn))))]> {
2583 let TwoOperandAliasConstraint = "$Vm = $Vd";
2584 let isCommutable = 0;
2587 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2588 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2589 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp, bit Commutable>
2590 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2591 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), f, itin,
2592 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2593 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
2594 // All of these have a two-operand InstAlias.
2595 let TwoOperandAliasConstraint = "$Vn = $Vd";
2596 let isCommutable = Commutable;
2599 class N3VQIntnp<bits<5> op27_23, bits<2> op21_20, bits<4> op11_8, bit op6,
2600 bit op4, Format f, InstrItinClass itin, string OpcodeStr,
2601 string Dt, ValueType ResTy, ValueType OpTy,
2602 SDPatternOperator IntOp, bit Commutable>
2603 : N3Vnp<op27_23, op21_20, op11_8, op6, op4,
2604 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), f, itin, OpcodeStr, Dt,
2605 ResTy, OpTy, IntOp, Commutable,
2606 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>;
2608 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2609 string OpcodeStr, string Dt,
2610 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2611 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2612 (outs QPR:$Vd), (ins QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2613 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2614 [(set (ResTy QPR:$Vd),
2615 (ResTy (IntOp (ResTy QPR:$Vn),
2616 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2618 let isCommutable = 0;
2620 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2621 string OpcodeStr, string Dt,
2622 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2623 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2624 (outs QPR:$Vd), (ins QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2625 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2626 [(set (ResTy QPR:$Vd),
2627 (ResTy (IntOp (ResTy QPR:$Vn),
2628 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2630 let isCommutable = 0;
2632 class N3VQIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2633 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2634 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2635 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2636 (outs QPR:$Vd), (ins QPR:$Vm, QPR:$Vn), f, itin,
2637 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
2638 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (OpTy QPR:$Vn))))]> {
2639 let TwoOperandAliasConstraint = "$Vm = $Vd";
2640 let isCommutable = 0;
2643 // Multiply-Add/Sub operations: double- and quad-register.
2644 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2645 InstrItinClass itin, string OpcodeStr, string Dt,
2646 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator OpNode>
2647 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2648 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2649 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2650 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
2651 (Ty (MulOp DPR:$Vn, DPR:$Vm)))))]>;
2653 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2654 string OpcodeStr, string Dt,
2655 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator ShOp>
2656 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2658 (ins DPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2660 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2662 (Ty (ShOp (Ty DPR:$src1),
2664 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),
2666 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2667 string OpcodeStr, string Dt,
2668 ValueType Ty, SDNode MulOp, SDNode ShOp>
2669 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2671 (ins DPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2673 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2675 (Ty (ShOp (Ty DPR:$src1),
2677 (Ty (NEONvduplane (Ty DPR_8:$Vm),
2680 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2681 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
2682 SDPatternOperator MulOp, SDPatternOperator OpNode>
2683 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2684 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2685 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2686 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
2687 (Ty (MulOp QPR:$Vn, QPR:$Vm)))))]>;
2688 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2689 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2690 SDPatternOperator MulOp, SDPatternOperator ShOp>
2691 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2693 (ins QPR:$src1, QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2695 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2696 [(set (ResTy QPR:$Vd),
2697 (ResTy (ShOp (ResTy QPR:$src1),
2698 (ResTy (MulOp QPR:$Vn,
2699 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2701 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2702 string OpcodeStr, string Dt,
2703 ValueType ResTy, ValueType OpTy,
2704 SDNode MulOp, SDNode ShOp>
2705 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2707 (ins QPR:$src1, QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2709 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2710 [(set (ResTy QPR:$Vd),
2711 (ResTy (ShOp (ResTy QPR:$src1),
2712 (ResTy (MulOp QPR:$Vn,
2713 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2716 // Neon Intrinsic-Op instructions (VABA): double- and quad-register.
2717 class N3VDIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2718 InstrItinClass itin, string OpcodeStr, string Dt,
2719 ValueType Ty, SDPatternOperator IntOp, SDNode OpNode>
2720 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2721 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2722 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2723 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
2724 (Ty (IntOp (Ty DPR:$Vn), (Ty DPR:$Vm))))))]>;
2725 class N3VQIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2726 InstrItinClass itin, string OpcodeStr, string Dt,
2727 ValueType Ty, SDPatternOperator IntOp, SDNode OpNode>
2728 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2729 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2730 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2731 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
2732 (Ty (IntOp (Ty QPR:$Vn), (Ty QPR:$Vm))))))]>;
2734 // Neon 3-argument intrinsics, both double- and quad-register.
2735 // The destination register is also used as the first source operand register.
2736 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2737 InstrItinClass itin, string OpcodeStr, string Dt,
2738 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2739 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2740 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2741 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2742 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$src1),
2743 (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>;
2744 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2745 InstrItinClass itin, string OpcodeStr, string Dt,
2746 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2747 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2748 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2749 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2750 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$src1),
2751 (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>;
2753 // Long Multiply-Add/Sub operations.
2754 class N3VLMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2755 InstrItinClass itin, string OpcodeStr, string Dt,
2756 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2757 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2758 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2759 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2760 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
2761 (TyQ (MulOp (TyD DPR:$Vn),
2762 (TyD DPR:$Vm)))))]>;
2763 class N3VLMulOpSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2764 InstrItinClass itin, string OpcodeStr, string Dt,
2765 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2766 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$Vd),
2767 (ins QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2769 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2771 (OpNode (TyQ QPR:$src1),
2772 (TyQ (MulOp (TyD DPR:$Vn),
2773 (TyD (NEONvduplane (TyD DPR_VFP2:$Vm),
2775 class N3VLMulOpSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2776 InstrItinClass itin, string OpcodeStr, string Dt,
2777 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2778 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$Vd),
2779 (ins QPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2781 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2783 (OpNode (TyQ QPR:$src1),
2784 (TyQ (MulOp (TyD DPR:$Vn),
2785 (TyD (NEONvduplane (TyD DPR_8:$Vm),
2788 // Long Intrinsic-Op vector operations with explicit extend (VABAL).
2789 class N3VLIntExtOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2790 InstrItinClass itin, string OpcodeStr, string Dt,
2791 ValueType TyQ, ValueType TyD, SDPatternOperator IntOp, SDNode ExtOp,
2793 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2794 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2795 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2796 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
2797 (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
2798 (TyD DPR:$Vm)))))))]>;
2800 // Neon Long 3-argument intrinsic. The destination register is
2801 // a quad-register and is also used as the first source operand register.
2802 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2803 InstrItinClass itin, string OpcodeStr, string Dt,
2804 ValueType TyQ, ValueType TyD, SDPatternOperator IntOp>
2805 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2806 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2807 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2809 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$Vn), (TyD DPR:$Vm))))]>;
2810 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2811 string OpcodeStr, string Dt,
2812 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2813 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2815 (ins QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2817 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2818 [(set (ResTy QPR:$Vd),
2819 (ResTy (IntOp (ResTy QPR:$src1),
2821 (OpTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2823 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2824 InstrItinClass itin, string OpcodeStr, string Dt,
2825 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2826 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2828 (ins QPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2830 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2831 [(set (ResTy QPR:$Vd),
2832 (ResTy (IntOp (ResTy QPR:$src1),
2834 (OpTy (NEONvduplane (OpTy DPR_8:$Vm),
2837 // Narrowing 3-register intrinsics.
2838 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2839 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
2840 SDPatternOperator IntOp, bit Commutable>
2841 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2842 (outs DPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINi4D,
2843 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2844 [(set DPR:$Vd, (TyD (IntOp (TyQ QPR:$Vn), (TyQ QPR:$Vm))))]> {
2845 let isCommutable = Commutable;
2848 // Long 3-register operations.
2849 class N3VL<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2850 InstrItinClass itin, string OpcodeStr, string Dt,
2851 ValueType TyQ, ValueType TyD, SDNode OpNode, bit Commutable>
2852 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2853 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2854 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2855 [(set QPR:$Vd, (TyQ (OpNode (TyD DPR:$Vn), (TyD DPR:$Vm))))]> {
2856 let isCommutable = Commutable;
2858 class N3VLSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2859 InstrItinClass itin, string OpcodeStr, string Dt,
2860 ValueType TyQ, ValueType TyD, SDNode OpNode>
2861 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2862 (outs QPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2863 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2865 (TyQ (OpNode (TyD DPR:$Vn),
2866 (TyD (NEONvduplane (TyD DPR_VFP2:$Vm),imm:$lane)))))]>;
2867 class N3VLSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2868 InstrItinClass itin, string OpcodeStr, string Dt,
2869 ValueType TyQ, ValueType TyD, SDNode OpNode>
2870 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2871 (outs QPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2872 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2874 (TyQ (OpNode (TyD DPR:$Vn),
2875 (TyD (NEONvduplane (TyD DPR_8:$Vm), imm:$lane)))))]>;
2877 // Long 3-register operations with explicitly extended operands.
2878 class N3VLExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2879 InstrItinClass itin, string OpcodeStr, string Dt,
2880 ValueType TyQ, ValueType TyD, SDNode OpNode, SDNode ExtOp,
2882 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2883 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2884 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2885 [(set QPR:$Vd, (OpNode (TyQ (ExtOp (TyD DPR:$Vn))),
2886 (TyQ (ExtOp (TyD DPR:$Vm)))))]> {
2887 let isCommutable = Commutable;
2890 // Long 3-register intrinsics with explicit extend (VABDL).
2891 class N3VLIntExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2892 InstrItinClass itin, string OpcodeStr, string Dt,
2893 ValueType TyQ, ValueType TyD, SDPatternOperator IntOp, SDNode ExtOp,
2895 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2896 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2897 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2898 [(set QPR:$Vd, (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
2899 (TyD DPR:$Vm))))))]> {
2900 let isCommutable = Commutable;
2903 // Long 3-register intrinsics.
2904 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2905 InstrItinClass itin, string OpcodeStr, string Dt,
2906 ValueType TyQ, ValueType TyD, SDPatternOperator IntOp, bit Commutable>
2907 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2908 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2909 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2910 [(set QPR:$Vd, (TyQ (IntOp (TyD DPR:$Vn), (TyD DPR:$Vm))))]> {
2911 let isCommutable = Commutable;
2913 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2914 string OpcodeStr, string Dt,
2915 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2916 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2917 (outs QPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2918 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2919 [(set (ResTy QPR:$Vd),
2920 (ResTy (IntOp (OpTy DPR:$Vn),
2921 (OpTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2923 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2924 InstrItinClass itin, string OpcodeStr, string Dt,
2925 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2926 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2927 (outs QPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2928 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2929 [(set (ResTy QPR:$Vd),
2930 (ResTy (IntOp (OpTy DPR:$Vn),
2931 (OpTy (NEONvduplane (OpTy DPR_8:$Vm),
2934 // Wide 3-register operations.
2935 class N3VW<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2936 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
2937 SDNode OpNode, SDNode ExtOp, bit Commutable>
2938 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2939 (outs QPR:$Vd), (ins QPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VSUBiD,
2940 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2941 [(set QPR:$Vd, (OpNode (TyQ QPR:$Vn),
2942 (TyQ (ExtOp (TyD DPR:$Vm)))))]> {
2943 // All of these have a two-operand InstAlias.
2944 let TwoOperandAliasConstraint = "$Vn = $Vd";
2945 let isCommutable = Commutable;
2948 // Pairwise long 2-register intrinsics, both double- and quad-register.
2949 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2950 bits<2> op17_16, bits<5> op11_7, bit op4,
2951 string OpcodeStr, string Dt,
2952 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2953 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2954 (ins DPR:$Vm), IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
2955 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
2956 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2957 bits<2> op17_16, bits<5> op11_7, bit op4,
2958 string OpcodeStr, string Dt,
2959 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2960 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2961 (ins QPR:$Vm), IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
2962 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
2964 // Pairwise long 2-register accumulate intrinsics,
2965 // both double- and quad-register.
2966 // The destination register is also used as the first source operand register.
2967 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2968 bits<2> op17_16, bits<5> op11_7, bit op4,
2969 string OpcodeStr, string Dt,
2970 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2971 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
2972 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vm), IIC_VPALiD,
2973 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2974 [(set DPR:$Vd, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$Vm))))]>;
2975 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2976 bits<2> op17_16, bits<5> op11_7, bit op4,
2977 string OpcodeStr, string Dt,
2978 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2979 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
2980 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vm), IIC_VPALiQ,
2981 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2982 [(set QPR:$Vd, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$Vm))))]>;
2984 // Shift by immediate,
2985 // both double- and quad-register.
2986 let TwoOperandAliasConstraint = "$Vm = $Vd" in {
2987 class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2988 Format f, InstrItinClass itin, Operand ImmTy,
2989 string OpcodeStr, string Dt, ValueType Ty, SDNode OpNode>
2990 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2991 (outs DPR:$Vd), (ins DPR:$Vm, ImmTy:$SIMM), f, itin,
2992 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2993 [(set DPR:$Vd, (Ty (OpNode (Ty DPR:$Vm), (i32 imm:$SIMM))))]>;
2994 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2995 Format f, InstrItinClass itin, Operand ImmTy,
2996 string OpcodeStr, string Dt, ValueType Ty, SDNode OpNode>
2997 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2998 (outs QPR:$Vd), (ins QPR:$Vm, ImmTy:$SIMM), f, itin,
2999 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
3000 [(set QPR:$Vd, (Ty (OpNode (Ty QPR:$Vm), (i32 imm:$SIMM))))]>;
3003 // Long shift by immediate.
3004 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
3005 string OpcodeStr, string Dt,
3006 ValueType ResTy, ValueType OpTy, Operand ImmTy, SDNode OpNode>
3007 : N2VImm<op24, op23, op11_8, op7, op6, op4,
3008 (outs QPR:$Vd), (ins DPR:$Vm, ImmTy:$SIMM), N2RegVShLFrm,
3009 IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
3010 [(set QPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm),
3011 (i32 imm:$SIMM))))]>;
3013 // Narrow shift by immediate.
3014 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
3015 InstrItinClass itin, string OpcodeStr, string Dt,
3016 ValueType ResTy, ValueType OpTy, Operand ImmTy, SDNode OpNode>
3017 : N2VImm<op24, op23, op11_8, op7, op6, op4,
3018 (outs DPR:$Vd), (ins QPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, itin,
3019 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
3020 [(set DPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vm),
3021 (i32 imm:$SIMM))))]>;
3023 // Shift right by immediate and accumulate,
3024 // both double- and quad-register.
3025 let TwoOperandAliasConstraint = "$Vm = $Vd" in {
3026 class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
3027 Operand ImmTy, string OpcodeStr, string Dt,
3028 ValueType Ty, SDNode ShOp>
3029 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
3030 (ins DPR:$src1, DPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, IIC_VPALiD,
3031 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
3032 [(set DPR:$Vd, (Ty (add DPR:$src1,
3033 (Ty (ShOp DPR:$Vm, (i32 imm:$SIMM))))))]>;
3034 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
3035 Operand ImmTy, string OpcodeStr, string Dt,
3036 ValueType Ty, SDNode ShOp>
3037 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
3038 (ins QPR:$src1, QPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, IIC_VPALiD,
3039 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
3040 [(set QPR:$Vd, (Ty (add QPR:$src1,
3041 (Ty (ShOp QPR:$Vm, (i32 imm:$SIMM))))))]>;
3044 // Shift by immediate and insert,
3045 // both double- and quad-register.
3046 let TwoOperandAliasConstraint = "$Vm = $Vd" in {
3047 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
3048 Operand ImmTy, Format f, string OpcodeStr, string Dt,
3049 ValueType Ty,SDNode ShOp>
3050 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
3051 (ins DPR:$src1, DPR:$Vm, ImmTy:$SIMM), f, IIC_VSHLiD,
3052 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
3053 [(set DPR:$Vd, (Ty (ShOp DPR:$src1, DPR:$Vm, (i32 imm:$SIMM))))]>;
3054 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
3055 Operand ImmTy, Format f, string OpcodeStr, string Dt,
3056 ValueType Ty,SDNode ShOp>
3057 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
3058 (ins QPR:$src1, QPR:$Vm, ImmTy:$SIMM), f, IIC_VSHLiQ,
3059 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
3060 [(set QPR:$Vd, (Ty (ShOp QPR:$src1, QPR:$Vm, (i32 imm:$SIMM))))]>;
3063 // Convert, with fractional bits immediate,
3064 // both double- and quad-register.
3065 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
3066 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
3067 SDPatternOperator IntOp>
3068 : N2VImm<op24, op23, op11_8, op7, 0, op4,
3069 (outs DPR:$Vd), (ins DPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
3070 IIC_VUNAD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
3071 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (i32 imm:$SIMM))))]>;
3072 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
3073 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
3074 SDPatternOperator IntOp>
3075 : N2VImm<op24, op23, op11_8, op7, 1, op4,
3076 (outs QPR:$Vd), (ins QPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
3077 IIC_VUNAQ, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
3078 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (i32 imm:$SIMM))))]>;
3080 //===----------------------------------------------------------------------===//
3082 //===----------------------------------------------------------------------===//
3084 // Abbreviations used in multiclass suffixes:
3085 // Q = quarter int (8 bit) elements
3086 // H = half int (16 bit) elements
3087 // S = single int (32 bit) elements
3088 // D = double int (64 bit) elements
3090 // Neon 2-register vector operations and intrinsics.
3092 // Neon 2-register comparisons.
3093 // source operand element sizes of 8, 16 and 32 bits:
3094 multiclass N2V_QHS_cmp<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3095 bits<5> op11_7, bit op4, string opc, string Dt,
3096 string asm, SDNode OpNode> {
3097 // 64-bit vector types.
3098 def v8i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 0, op4,
3099 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
3100 opc, !strconcat(Dt, "8"), asm, "",
3101 [(set DPR:$Vd, (v8i8 (OpNode (v8i8 DPR:$Vm))))]>;
3102 def v4i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
3103 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
3104 opc, !strconcat(Dt, "16"), asm, "",
3105 [(set DPR:$Vd, (v4i16 (OpNode (v4i16 DPR:$Vm))))]>;
3106 def v2i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
3107 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
3108 opc, !strconcat(Dt, "32"), asm, "",
3109 [(set DPR:$Vd, (v2i32 (OpNode (v2i32 DPR:$Vm))))]>;
3110 def v2f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
3111 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
3112 opc, "f32", asm, "",
3113 [(set DPR:$Vd, (v2i32 (OpNode (v2f32 DPR:$Vm))))]> {
3114 let Inst{10} = 1; // overwrite F = 1
3117 // 128-bit vector types.
3118 def v16i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 1, op4,
3119 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
3120 opc, !strconcat(Dt, "8"), asm, "",
3121 [(set QPR:$Vd, (v16i8 (OpNode (v16i8 QPR:$Vm))))]>;
3122 def v8i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
3123 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
3124 opc, !strconcat(Dt, "16"), asm, "",
3125 [(set QPR:$Vd, (v8i16 (OpNode (v8i16 QPR:$Vm))))]>;
3126 def v4i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
3127 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
3128 opc, !strconcat(Dt, "32"), asm, "",
3129 [(set QPR:$Vd, (v4i32 (OpNode (v4i32 QPR:$Vm))))]>;
3130 def v4f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
3131 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
3132 opc, "f32", asm, "",
3133 [(set QPR:$Vd, (v4i32 (OpNode (v4f32 QPR:$Vm))))]> {
3134 let Inst{10} = 1; // overwrite F = 1
3139 // Neon 2-register vector intrinsics,
3140 // element sizes of 8, 16 and 32 bits:
3141 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3142 bits<5> op11_7, bit op4,
3143 InstrItinClass itinD, InstrItinClass itinQ,
3144 string OpcodeStr, string Dt, SDPatternOperator IntOp> {
3145 // 64-bit vector types.
3146 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3147 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
3148 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3149 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
3150 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3151 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
3153 // 128-bit vector types.
3154 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3155 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
3156 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3157 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
3158 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3159 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
3163 // Neon Narrowing 2-register vector operations,
3164 // source operand element sizes of 16, 32 and 64 bits:
3165 multiclass N2VN_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3166 bits<5> op11_7, bit op6, bit op4,
3167 InstrItinClass itin, string OpcodeStr, string Dt,
3169 def v8i8 : N2VN<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
3170 itin, OpcodeStr, !strconcat(Dt, "16"),
3171 v8i8, v8i16, OpNode>;
3172 def v4i16 : N2VN<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
3173 itin, OpcodeStr, !strconcat(Dt, "32"),
3174 v4i16, v4i32, OpNode>;
3175 def v2i32 : N2VN<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
3176 itin, OpcodeStr, !strconcat(Dt, "64"),
3177 v2i32, v2i64, OpNode>;
3180 // Neon Narrowing 2-register vector intrinsics,
3181 // source operand element sizes of 16, 32 and 64 bits:
3182 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3183 bits<5> op11_7, bit op6, bit op4,
3184 InstrItinClass itin, string OpcodeStr, string Dt,
3185 SDPatternOperator IntOp> {
3186 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
3187 itin, OpcodeStr, !strconcat(Dt, "16"),
3188 v8i8, v8i16, IntOp>;
3189 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
3190 itin, OpcodeStr, !strconcat(Dt, "32"),
3191 v4i16, v4i32, IntOp>;
3192 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
3193 itin, OpcodeStr, !strconcat(Dt, "64"),
3194 v2i32, v2i64, IntOp>;
3198 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
3199 // source operand element sizes of 16, 32 and 64 bits:
3200 multiclass N2VL_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
3201 string OpcodeStr, string Dt, SDNode OpNode> {
3202 def v8i16 : N2VL<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
3203 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode>;
3204 def v4i32 : N2VL<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
3205 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
3206 def v2i64 : N2VL<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
3207 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
3211 // Neon 3-register vector operations.
3213 // First with only element sizes of 8, 16 and 32 bits:
3214 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3215 InstrItinClass itinD16, InstrItinClass itinD32,
3216 InstrItinClass itinQ16, InstrItinClass itinQ32,
3217 string OpcodeStr, string Dt,
3218 SDNode OpNode, bit Commutable = 0> {
3219 // 64-bit vector types.
3220 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
3221 OpcodeStr, !strconcat(Dt, "8"),
3222 v8i8, v8i8, OpNode, Commutable>;
3223 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
3224 OpcodeStr, !strconcat(Dt, "16"),
3225 v4i16, v4i16, OpNode, Commutable>;
3226 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
3227 OpcodeStr, !strconcat(Dt, "32"),
3228 v2i32, v2i32, OpNode, Commutable>;
3230 // 128-bit vector types.
3231 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
3232 OpcodeStr, !strconcat(Dt, "8"),
3233 v16i8, v16i8, OpNode, Commutable>;
3234 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
3235 OpcodeStr, !strconcat(Dt, "16"),
3236 v8i16, v8i16, OpNode, Commutable>;
3237 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
3238 OpcodeStr, !strconcat(Dt, "32"),
3239 v4i32, v4i32, OpNode, Commutable>;
3242 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, SDNode ShOp> {
3243 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, "i16", v4i16, ShOp>;
3244 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, "i32", v2i32, ShOp>;
3245 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, "i16", v8i16, v4i16, ShOp>;
3246 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, "i32",
3247 v4i32, v2i32, ShOp>;
3250 // ....then also with element size 64 bits:
3251 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3252 InstrItinClass itinD, InstrItinClass itinQ,
3253 string OpcodeStr, string Dt,
3254 SDNode OpNode, bit Commutable = 0>
3255 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
3256 OpcodeStr, Dt, OpNode, Commutable> {
3257 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
3258 OpcodeStr, !strconcat(Dt, "64"),
3259 v1i64, v1i64, OpNode, Commutable>;
3260 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
3261 OpcodeStr, !strconcat(Dt, "64"),
3262 v2i64, v2i64, OpNode, Commutable>;
3266 // Neon 3-register vector intrinsics.
3268 // First with only element sizes of 16 and 32 bits:
3269 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3270 InstrItinClass itinD16, InstrItinClass itinD32,
3271 InstrItinClass itinQ16, InstrItinClass itinQ32,
3272 string OpcodeStr, string Dt,
3273 SDPatternOperator IntOp, bit Commutable = 0> {
3274 // 64-bit vector types.
3275 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, f, itinD16,
3276 OpcodeStr, !strconcat(Dt, "16"),
3277 v4i16, v4i16, IntOp, Commutable>;
3278 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, f, itinD32,
3279 OpcodeStr, !strconcat(Dt, "32"),
3280 v2i32, v2i32, IntOp, Commutable>;
3282 // 128-bit vector types.
3283 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, f, itinQ16,
3284 OpcodeStr, !strconcat(Dt, "16"),
3285 v8i16, v8i16, IntOp, Commutable>;
3286 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, f, itinQ32,
3287 OpcodeStr, !strconcat(Dt, "32"),
3288 v4i32, v4i32, IntOp, Commutable>;
3290 multiclass N3VInt_HSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3291 InstrItinClass itinD16, InstrItinClass itinD32,
3292 InstrItinClass itinQ16, InstrItinClass itinQ32,
3293 string OpcodeStr, string Dt,
3294 SDPatternOperator IntOp> {
3295 // 64-bit vector types.
3296 def v4i16 : N3VDIntSh<op24, op23, 0b01, op11_8, op4, f, itinD16,
3297 OpcodeStr, !strconcat(Dt, "16"),
3298 v4i16, v4i16, IntOp>;
3299 def v2i32 : N3VDIntSh<op24, op23, 0b10, op11_8, op4, f, itinD32,
3300 OpcodeStr, !strconcat(Dt, "32"),
3301 v2i32, v2i32, IntOp>;
3303 // 128-bit vector types.
3304 def v8i16 : N3VQIntSh<op24, op23, 0b01, op11_8, op4, f, itinQ16,
3305 OpcodeStr, !strconcat(Dt, "16"),
3306 v8i16, v8i16, IntOp>;
3307 def v4i32 : N3VQIntSh<op24, op23, 0b10, op11_8, op4, f, itinQ32,
3308 OpcodeStr, !strconcat(Dt, "32"),
3309 v4i32, v4i32, IntOp>;
3312 multiclass N3VIntSL_HS<bits<4> op11_8,
3313 InstrItinClass itinD16, InstrItinClass itinD32,
3314 InstrItinClass itinQ16, InstrItinClass itinQ32,
3315 string OpcodeStr, string Dt, SDPatternOperator IntOp> {
3316 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
3317 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
3318 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
3319 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
3320 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
3321 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
3322 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
3323 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
3326 // ....then also with element size of 8 bits:
3327 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3328 InstrItinClass itinD16, InstrItinClass itinD32,
3329 InstrItinClass itinQ16, InstrItinClass itinQ32,
3330 string OpcodeStr, string Dt,
3331 SDPatternOperator IntOp, bit Commutable = 0>
3332 : N3VInt_HS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3333 OpcodeStr, Dt, IntOp, Commutable> {
3334 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, f, itinD16,
3335 OpcodeStr, !strconcat(Dt, "8"),
3336 v8i8, v8i8, IntOp, Commutable>;
3337 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, f, itinQ16,
3338 OpcodeStr, !strconcat(Dt, "8"),
3339 v16i8, v16i8, IntOp, Commutable>;
3341 multiclass N3VInt_QHSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3342 InstrItinClass itinD16, InstrItinClass itinD32,
3343 InstrItinClass itinQ16, InstrItinClass itinQ32,
3344 string OpcodeStr, string Dt,
3345 SDPatternOperator IntOp>
3346 : N3VInt_HSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3347 OpcodeStr, Dt, IntOp> {
3348 def v8i8 : N3VDIntSh<op24, op23, 0b00, op11_8, op4, f, itinD16,
3349 OpcodeStr, !strconcat(Dt, "8"),
3351 def v16i8 : N3VQIntSh<op24, op23, 0b00, op11_8, op4, f, itinQ16,
3352 OpcodeStr, !strconcat(Dt, "8"),
3353 v16i8, v16i8, IntOp>;
3357 // ....then also with element size of 64 bits:
3358 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3359 InstrItinClass itinD16, InstrItinClass itinD32,
3360 InstrItinClass itinQ16, InstrItinClass itinQ32,
3361 string OpcodeStr, string Dt,
3362 SDPatternOperator IntOp, bit Commutable = 0>
3363 : N3VInt_QHS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3364 OpcodeStr, Dt, IntOp, Commutable> {
3365 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, f, itinD32,
3366 OpcodeStr, !strconcat(Dt, "64"),
3367 v1i64, v1i64, IntOp, Commutable>;
3368 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, f, itinQ32,
3369 OpcodeStr, !strconcat(Dt, "64"),
3370 v2i64, v2i64, IntOp, Commutable>;
3372 multiclass N3VInt_QHSDSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3373 InstrItinClass itinD16, InstrItinClass itinD32,
3374 InstrItinClass itinQ16, InstrItinClass itinQ32,
3375 string OpcodeStr, string Dt,
3376 SDPatternOperator IntOp>
3377 : N3VInt_QHSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3378 OpcodeStr, Dt, IntOp> {
3379 def v1i64 : N3VDIntSh<op24, op23, 0b11, op11_8, op4, f, itinD32,
3380 OpcodeStr, !strconcat(Dt, "64"),
3381 v1i64, v1i64, IntOp>;
3382 def v2i64 : N3VQIntSh<op24, op23, 0b11, op11_8, op4, f, itinQ32,
3383 OpcodeStr, !strconcat(Dt, "64"),
3384 v2i64, v2i64, IntOp>;
3387 // Neon Narrowing 3-register vector intrinsics,
3388 // source operand element sizes of 16, 32 and 64 bits:
3389 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3390 string OpcodeStr, string Dt,
3391 SDPatternOperator IntOp, bit Commutable = 0> {
3392 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
3393 OpcodeStr, !strconcat(Dt, "16"),
3394 v8i8, v8i16, IntOp, Commutable>;
3395 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
3396 OpcodeStr, !strconcat(Dt, "32"),
3397 v4i16, v4i32, IntOp, Commutable>;
3398 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
3399 OpcodeStr, !strconcat(Dt, "64"),
3400 v2i32, v2i64, IntOp, Commutable>;
3404 // Neon Long 3-register vector operations.
3406 multiclass N3VL_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3407 InstrItinClass itin16, InstrItinClass itin32,
3408 string OpcodeStr, string Dt,
3409 SDNode OpNode, bit Commutable = 0> {
3410 def v8i16 : N3VL<op24, op23, 0b00, op11_8, op4, itin16,
3411 OpcodeStr, !strconcat(Dt, "8"),
3412 v8i16, v8i8, OpNode, Commutable>;
3413 def v4i32 : N3VL<op24, op23, 0b01, op11_8, op4, itin16,
3414 OpcodeStr, !strconcat(Dt, "16"),
3415 v4i32, v4i16, OpNode, Commutable>;
3416 def v2i64 : N3VL<op24, op23, 0b10, op11_8, op4, itin32,
3417 OpcodeStr, !strconcat(Dt, "32"),
3418 v2i64, v2i32, OpNode, Commutable>;
3421 multiclass N3VLSL_HS<bit op24, bits<4> op11_8,
3422 InstrItinClass itin, string OpcodeStr, string Dt,
3424 def v4i16 : N3VLSL16<op24, 0b01, op11_8, itin, OpcodeStr,
3425 !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
3426 def v2i32 : N3VLSL<op24, 0b10, op11_8, itin, OpcodeStr,
3427 !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
3430 multiclass N3VLExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3431 InstrItinClass itin16, InstrItinClass itin32,
3432 string OpcodeStr, string Dt,
3433 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
3434 def v8i16 : N3VLExt<op24, op23, 0b00, op11_8, op4, itin16,
3435 OpcodeStr, !strconcat(Dt, "8"),
3436 v8i16, v8i8, OpNode, ExtOp, Commutable>;
3437 def v4i32 : N3VLExt<op24, op23, 0b01, op11_8, op4, itin16,
3438 OpcodeStr, !strconcat(Dt, "16"),
3439 v4i32, v4i16, OpNode, ExtOp, Commutable>;
3440 def v2i64 : N3VLExt<op24, op23, 0b10, op11_8, op4, itin32,
3441 OpcodeStr, !strconcat(Dt, "32"),
3442 v2i64, v2i32, OpNode, ExtOp, Commutable>;
3445 // Neon Long 3-register vector intrinsics.
3447 // First with only element sizes of 16 and 32 bits:
3448 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
3449 InstrItinClass itin16, InstrItinClass itin32,
3450 string OpcodeStr, string Dt,
3451 SDPatternOperator IntOp, bit Commutable = 0> {
3452 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin16,
3453 OpcodeStr, !strconcat(Dt, "16"),
3454 v4i32, v4i16, IntOp, Commutable>;
3455 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin32,
3456 OpcodeStr, !strconcat(Dt, "32"),
3457 v2i64, v2i32, IntOp, Commutable>;
3460 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
3461 InstrItinClass itin, string OpcodeStr, string Dt,
3462 SDPatternOperator IntOp> {
3463 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
3464 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
3465 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
3466 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3469 // ....then also with element size of 8 bits:
3470 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3471 InstrItinClass itin16, InstrItinClass itin32,
3472 string OpcodeStr, string Dt,
3473 SDPatternOperator IntOp, bit Commutable = 0>
3474 : N3VLInt_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt,
3475 IntOp, Commutable> {
3476 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin16,
3477 OpcodeStr, !strconcat(Dt, "8"),
3478 v8i16, v8i8, IntOp, Commutable>;
3481 // ....with explicit extend (VABDL).
3482 multiclass N3VLIntExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3483 InstrItinClass itin, string OpcodeStr, string Dt,
3484 SDPatternOperator IntOp, SDNode ExtOp, bit Commutable = 0> {
3485 def v8i16 : N3VLIntExt<op24, op23, 0b00, op11_8, op4, itin,
3486 OpcodeStr, !strconcat(Dt, "8"),
3487 v8i16, v8i8, IntOp, ExtOp, Commutable>;
3488 def v4i32 : N3VLIntExt<op24, op23, 0b01, op11_8, op4, itin,
3489 OpcodeStr, !strconcat(Dt, "16"),
3490 v4i32, v4i16, IntOp, ExtOp, Commutable>;
3491 def v2i64 : N3VLIntExt<op24, op23, 0b10, op11_8, op4, itin,
3492 OpcodeStr, !strconcat(Dt, "32"),
3493 v2i64, v2i32, IntOp, ExtOp, Commutable>;
3497 // Neon Wide 3-register vector intrinsics,
3498 // source operand element sizes of 8, 16 and 32 bits:
3499 multiclass N3VW_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3500 string OpcodeStr, string Dt,
3501 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
3502 def v8i16 : N3VW<op24, op23, 0b00, op11_8, op4,
3503 OpcodeStr, !strconcat(Dt, "8"),
3504 v8i16, v8i8, OpNode, ExtOp, Commutable>;
3505 def v4i32 : N3VW<op24, op23, 0b01, op11_8, op4,
3506 OpcodeStr, !strconcat(Dt, "16"),
3507 v4i32, v4i16, OpNode, ExtOp, Commutable>;
3508 def v2i64 : N3VW<op24, op23, 0b10, op11_8, op4,
3509 OpcodeStr, !strconcat(Dt, "32"),
3510 v2i64, v2i32, OpNode, ExtOp, Commutable>;
3514 // Neon Multiply-Op vector operations,
3515 // element sizes of 8, 16 and 32 bits:
3516 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3517 InstrItinClass itinD16, InstrItinClass itinD32,
3518 InstrItinClass itinQ16, InstrItinClass itinQ32,
3519 string OpcodeStr, string Dt, SDNode OpNode> {
3520 // 64-bit vector types.
3521 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
3522 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
3523 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
3524 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
3525 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
3526 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
3528 // 128-bit vector types.
3529 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
3530 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
3531 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
3532 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
3533 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
3534 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
3537 multiclass N3VMulOpSL_HS<bits<4> op11_8,
3538 InstrItinClass itinD16, InstrItinClass itinD32,
3539 InstrItinClass itinQ16, InstrItinClass itinQ32,
3540 string OpcodeStr, string Dt, SDNode ShOp> {
3541 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
3542 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
3543 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
3544 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
3545 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
3546 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
3548 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
3549 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
3553 // Neon Intrinsic-Op vector operations,
3554 // element sizes of 8, 16 and 32 bits:
3555 multiclass N3VIntOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3556 InstrItinClass itinD, InstrItinClass itinQ,
3557 string OpcodeStr, string Dt, SDPatternOperator IntOp,
3559 // 64-bit vector types.
3560 def v8i8 : N3VDIntOp<op24, op23, 0b00, op11_8, op4, itinD,
3561 OpcodeStr, !strconcat(Dt, "8"), v8i8, IntOp, OpNode>;
3562 def v4i16 : N3VDIntOp<op24, op23, 0b01, op11_8, op4, itinD,
3563 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp, OpNode>;
3564 def v2i32 : N3VDIntOp<op24, op23, 0b10, op11_8, op4, itinD,
3565 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp, OpNode>;
3567 // 128-bit vector types.
3568 def v16i8 : N3VQIntOp<op24, op23, 0b00, op11_8, op4, itinQ,
3569 OpcodeStr, !strconcat(Dt, "8"), v16i8, IntOp, OpNode>;
3570 def v8i16 : N3VQIntOp<op24, op23, 0b01, op11_8, op4, itinQ,
3571 OpcodeStr, !strconcat(Dt, "16"), v8i16, IntOp, OpNode>;
3572 def v4i32 : N3VQIntOp<op24, op23, 0b10, op11_8, op4, itinQ,
3573 OpcodeStr, !strconcat(Dt, "32"), v4i32, IntOp, OpNode>;
3576 // Neon 3-argument intrinsics,
3577 // element sizes of 8, 16 and 32 bits:
3578 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3579 InstrItinClass itinD, InstrItinClass itinQ,
3580 string OpcodeStr, string Dt, SDPatternOperator IntOp> {
3581 // 64-bit vector types.
3582 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, itinD,
3583 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
3584 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, itinD,
3585 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
3586 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, itinD,
3587 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
3589 // 128-bit vector types.
3590 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, itinQ,
3591 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
3592 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, itinQ,
3593 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
3594 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, itinQ,
3595 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
3599 // Neon Long Multiply-Op vector operations,
3600 // element sizes of 8, 16 and 32 bits:
3601 multiclass N3VLMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3602 InstrItinClass itin16, InstrItinClass itin32,
3603 string OpcodeStr, string Dt, SDNode MulOp,
3605 def v8i16 : N3VLMulOp<op24, op23, 0b00, op11_8, op4, itin16, OpcodeStr,
3606 !strconcat(Dt, "8"), v8i16, v8i8, MulOp, OpNode>;
3607 def v4i32 : N3VLMulOp<op24, op23, 0b01, op11_8, op4, itin16, OpcodeStr,
3608 !strconcat(Dt, "16"), v4i32, v4i16, MulOp, OpNode>;
3609 def v2i64 : N3VLMulOp<op24, op23, 0b10, op11_8, op4, itin32, OpcodeStr,
3610 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
3613 multiclass N3VLMulOpSL_HS<bit op24, bits<4> op11_8, string OpcodeStr,
3614 string Dt, SDNode MulOp, SDNode OpNode> {
3615 def v4i16 : N3VLMulOpSL16<op24, 0b01, op11_8, IIC_VMACi16D, OpcodeStr,
3616 !strconcat(Dt,"16"), v4i32, v4i16, MulOp, OpNode>;
3617 def v2i32 : N3VLMulOpSL<op24, 0b10, op11_8, IIC_VMACi32D, OpcodeStr,
3618 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
3622 // Neon Long 3-argument intrinsics.
3624 // First with only element sizes of 16 and 32 bits:
3625 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
3626 InstrItinClass itin16, InstrItinClass itin32,
3627 string OpcodeStr, string Dt, SDPatternOperator IntOp> {
3628 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, itin16,
3629 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
3630 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, itin32,
3631 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3634 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
3635 string OpcodeStr, string Dt, SDPatternOperator IntOp> {
3636 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
3637 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
3638 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
3639 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3642 // ....then also with element size of 8 bits:
3643 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3644 InstrItinClass itin16, InstrItinClass itin32,
3645 string OpcodeStr, string Dt, SDPatternOperator IntOp>
3646 : N3VLInt3_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt, IntOp> {
3647 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, itin16,
3648 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
3651 // ....with explicit extend (VABAL).
3652 multiclass N3VLIntExtOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3653 InstrItinClass itin, string OpcodeStr, string Dt,
3654 SDPatternOperator IntOp, SDNode ExtOp, SDNode OpNode> {
3655 def v8i16 : N3VLIntExtOp<op24, op23, 0b00, op11_8, op4, itin,
3656 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8,
3657 IntOp, ExtOp, OpNode>;
3658 def v4i32 : N3VLIntExtOp<op24, op23, 0b01, op11_8, op4, itin,
3659 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16,
3660 IntOp, ExtOp, OpNode>;
3661 def v2i64 : N3VLIntExtOp<op24, op23, 0b10, op11_8, op4, itin,
3662 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32,
3663 IntOp, ExtOp, OpNode>;
3667 // Neon Pairwise long 2-register intrinsics,
3668 // element sizes of 8, 16 and 32 bits:
3669 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3670 bits<5> op11_7, bit op4,
3671 string OpcodeStr, string Dt, SDPatternOperator IntOp> {
3672 // 64-bit vector types.
3673 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3674 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
3675 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3676 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
3677 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3678 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
3680 // 128-bit vector types.
3681 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3682 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
3683 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3684 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
3685 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3686 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
3690 // Neon Pairwise long 2-register accumulate intrinsics,
3691 // element sizes of 8, 16 and 32 bits:
3692 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3693 bits<5> op11_7, bit op4,
3694 string OpcodeStr, string Dt, SDPatternOperator IntOp> {
3695 // 64-bit vector types.
3696 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3697 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
3698 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3699 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
3700 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3701 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
3703 // 128-bit vector types.
3704 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3705 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
3706 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3707 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
3708 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3709 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
3713 // Neon 2-register vector shift by immediate,
3714 // with f of either N2RegVShLFrm or N2RegVShRFrm
3715 // element sizes of 8, 16, 32 and 64 bits:
3716 multiclass N2VShL_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3717 InstrItinClass itin, string OpcodeStr, string Dt,
3719 // 64-bit vector types.
3720 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3721 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
3722 let Inst{21-19} = 0b001; // imm6 = 001xxx
3724 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3725 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
3726 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3728 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3729 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
3730 let Inst{21} = 0b1; // imm6 = 1xxxxx
3732 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, N2RegVShLFrm, itin, i32imm,
3733 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
3736 // 128-bit vector types.
3737 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3738 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
3739 let Inst{21-19} = 0b001; // imm6 = 001xxx
3741 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3742 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
3743 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3745 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3746 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
3747 let Inst{21} = 0b1; // imm6 = 1xxxxx
3749 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, N2RegVShLFrm, itin, i32imm,
3750 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
3753 multiclass N2VShR_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3754 InstrItinClass itin, string OpcodeStr, string Dt,
3755 string baseOpc, SDNode OpNode> {
3756 // 64-bit vector types.
3757 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm8,
3758 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
3759 let Inst{21-19} = 0b001; // imm6 = 001xxx
3761 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm16,
3762 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
3763 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3765 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm32,
3766 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
3767 let Inst{21} = 0b1; // imm6 = 1xxxxx
3769 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, N2RegVShRFrm, itin, shr_imm64,
3770 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
3773 // 128-bit vector types.
3774 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm8,
3775 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
3776 let Inst{21-19} = 0b001; // imm6 = 001xxx
3778 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm16,
3779 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
3780 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3782 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm32,
3783 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
3784 let Inst{21} = 0b1; // imm6 = 1xxxxx
3786 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, N2RegVShRFrm, itin, shr_imm64,
3787 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
3791 // Neon Shift-Accumulate vector operations,
3792 // element sizes of 8, 16, 32 and 64 bits:
3793 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3794 string OpcodeStr, string Dt, SDNode ShOp> {
3795 // 64-bit vector types.
3796 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm8,
3797 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
3798 let Inst{21-19} = 0b001; // imm6 = 001xxx
3800 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm16,
3801 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
3802 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3804 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm32,
3805 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
3806 let Inst{21} = 0b1; // imm6 = 1xxxxx
3808 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4, shr_imm64,
3809 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
3812 // 128-bit vector types.
3813 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm8,
3814 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
3815 let Inst{21-19} = 0b001; // imm6 = 001xxx
3817 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm16,
3818 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
3819 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3821 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm32,
3822 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
3823 let Inst{21} = 0b1; // imm6 = 1xxxxx
3825 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4, shr_imm64,
3826 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
3830 // Neon Shift-Insert vector operations,
3831 // with f of either N2RegVShLFrm or N2RegVShRFrm
3832 // element sizes of 8, 16, 32 and 64 bits:
3833 multiclass N2VShInsL_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3835 // 64-bit vector types.
3836 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3837 N2RegVShLFrm, OpcodeStr, "8", v8i8, NEONvsli> {
3838 let Inst{21-19} = 0b001; // imm6 = 001xxx
3840 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3841 N2RegVShLFrm, OpcodeStr, "16", v4i16, NEONvsli> {
3842 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3844 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3845 N2RegVShLFrm, OpcodeStr, "32", v2i32, NEONvsli> {
3846 let Inst{21} = 0b1; // imm6 = 1xxxxx
3848 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4, i32imm,
3849 N2RegVShLFrm, OpcodeStr, "64", v1i64, NEONvsli>;
3852 // 128-bit vector types.
3853 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3854 N2RegVShLFrm, OpcodeStr, "8", v16i8, NEONvsli> {
3855 let Inst{21-19} = 0b001; // imm6 = 001xxx
3857 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3858 N2RegVShLFrm, OpcodeStr, "16", v8i16, NEONvsli> {
3859 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3861 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3862 N2RegVShLFrm, OpcodeStr, "32", v4i32, NEONvsli> {
3863 let Inst{21} = 0b1; // imm6 = 1xxxxx
3865 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4, i32imm,
3866 N2RegVShLFrm, OpcodeStr, "64", v2i64, NEONvsli>;
3869 multiclass N2VShInsR_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3871 // 64-bit vector types.
3872 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm8,
3873 N2RegVShRFrm, OpcodeStr, "8", v8i8, NEONvsri> {
3874 let Inst{21-19} = 0b001; // imm6 = 001xxx
3876 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm16,
3877 N2RegVShRFrm, OpcodeStr, "16", v4i16, NEONvsri> {
3878 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3880 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm32,
3881 N2RegVShRFrm, OpcodeStr, "32", v2i32, NEONvsri> {
3882 let Inst{21} = 0b1; // imm6 = 1xxxxx
3884 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4, shr_imm64,
3885 N2RegVShRFrm, OpcodeStr, "64", v1i64, NEONvsri>;
3888 // 128-bit vector types.
3889 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm8,
3890 N2RegVShRFrm, OpcodeStr, "8", v16i8, NEONvsri> {
3891 let Inst{21-19} = 0b001; // imm6 = 001xxx
3893 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm16,
3894 N2RegVShRFrm, OpcodeStr, "16", v8i16, NEONvsri> {
3895 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3897 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm32,
3898 N2RegVShRFrm, OpcodeStr, "32", v4i32, NEONvsri> {
3899 let Inst{21} = 0b1; // imm6 = 1xxxxx
3901 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4, shr_imm64,
3902 N2RegVShRFrm, OpcodeStr, "64", v2i64, NEONvsri>;
3906 // Neon Shift Long operations,
3907 // element sizes of 8, 16, 32 bits:
3908 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
3909 bit op4, string OpcodeStr, string Dt, SDNode OpNode> {
3910 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3911 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, imm1_7, OpNode> {
3912 let Inst{21-19} = 0b001; // imm6 = 001xxx
3914 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3915 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, imm1_15, OpNode> {
3916 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3918 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3919 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, imm1_31, OpNode> {
3920 let Inst{21} = 0b1; // imm6 = 1xxxxx
3924 // Neon Shift Narrow operations,
3925 // element sizes of 16, 32, 64 bits:
3926 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
3927 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
3929 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3930 OpcodeStr, !strconcat(Dt, "16"),
3931 v8i8, v8i16, shr_imm8, OpNode> {
3932 let Inst{21-19} = 0b001; // imm6 = 001xxx
3934 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3935 OpcodeStr, !strconcat(Dt, "32"),
3936 v4i16, v4i32, shr_imm16, OpNode> {
3937 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3939 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3940 OpcodeStr, !strconcat(Dt, "64"),
3941 v2i32, v2i64, shr_imm32, OpNode> {
3942 let Inst{21} = 0b1; // imm6 = 1xxxxx
3946 //===----------------------------------------------------------------------===//
3947 // Instruction Definitions.
3948 //===----------------------------------------------------------------------===//
3950 // Vector Add Operations.
3952 // VADD : Vector Add (integer and floating-point)
3953 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
3955 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
3956 v2f32, v2f32, fadd, 1>;
3957 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
3958 v4f32, v4f32, fadd, 1>;
3959 // VADDL : Vector Add Long (Q = D + D)
3960 defm VADDLs : N3VLExt_QHS<0,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3961 "vaddl", "s", add, sext, 1>;
3962 defm VADDLu : N3VLExt_QHS<1,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3963 "vaddl", "u", add, zext, 1>;
3964 // VADDW : Vector Add Wide (Q = Q + D)
3965 defm VADDWs : N3VW_QHS<0,1,0b0001,0, "vaddw", "s", add, sext, 0>;
3966 defm VADDWu : N3VW_QHS<1,1,0b0001,0, "vaddw", "u", add, zext, 0>;
3967 // VHADD : Vector Halving Add
3968 defm VHADDs : N3VInt_QHS<0, 0, 0b0000, 0, N3RegFrm,
3969 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3970 "vhadd", "s", int_arm_neon_vhadds, 1>;
3971 defm VHADDu : N3VInt_QHS<1, 0, 0b0000, 0, N3RegFrm,
3972 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3973 "vhadd", "u", int_arm_neon_vhaddu, 1>;
3974 // VRHADD : Vector Rounding Halving Add
3975 defm VRHADDs : N3VInt_QHS<0, 0, 0b0001, 0, N3RegFrm,
3976 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3977 "vrhadd", "s", int_arm_neon_vrhadds, 1>;
3978 defm VRHADDu : N3VInt_QHS<1, 0, 0b0001, 0, N3RegFrm,
3979 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3980 "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
3981 // VQADD : Vector Saturating Add
3982 defm VQADDs : N3VInt_QHSD<0, 0, 0b0000, 1, N3RegFrm,
3983 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3984 "vqadd", "s", int_arm_neon_vqadds, 1>;
3985 defm VQADDu : N3VInt_QHSD<1, 0, 0b0000, 1, N3RegFrm,
3986 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3987 "vqadd", "u", int_arm_neon_vqaddu, 1>;
3988 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
3989 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i",
3990 int_arm_neon_vaddhn, 1>;
3991 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
3992 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
3993 int_arm_neon_vraddhn, 1>;
3995 // Vector Multiply Operations.
3997 // VMUL : Vector Multiply (integer, polynomial and floating-point)
3998 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
3999 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
4000 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16D, "vmul",
4001 "p8", v8i8, v8i8, int_arm_neon_vmulp, 1>;
4002 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16Q, "vmul",
4003 "p8", v16i8, v16i8, int_arm_neon_vmulp, 1>;
4004 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VFMULD, "vmul", "f32",
4005 v2f32, v2f32, fmul, 1>;
4006 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VFMULQ, "vmul", "f32",
4007 v4f32, v4f32, fmul, 1>;
4008 defm VMULsl : N3VSL_HS<0b1000, "vmul", mul>;
4009 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
4010 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
4013 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
4014 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
4015 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
4016 (v4i16 (EXTRACT_SUBREG QPR:$src2,
4017 (DSubReg_i16_reg imm:$lane))),
4018 (SubReg_i16_lane imm:$lane)))>;
4019 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
4020 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
4021 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
4022 (v2i32 (EXTRACT_SUBREG QPR:$src2,
4023 (DSubReg_i32_reg imm:$lane))),
4024 (SubReg_i32_lane imm:$lane)))>;
4025 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
4026 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
4027 (v4f32 (VMULslfq (v4f32 QPR:$src1),
4028 (v2f32 (EXTRACT_SUBREG QPR:$src2,
4029 (DSubReg_i32_reg imm:$lane))),
4030 (SubReg_i32_lane imm:$lane)))>;
4032 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
4033 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, N3RegFrm, IIC_VMULi16D, IIC_VMULi32D,
4034 IIC_VMULi16Q, IIC_VMULi32Q,
4035 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
4036 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
4037 IIC_VMULi16Q, IIC_VMULi32Q,
4038 "vqdmulh", "s", int_arm_neon_vqdmulh>;
4039 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
4040 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
4042 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
4043 (v4i16 (EXTRACT_SUBREG QPR:$src2,
4044 (DSubReg_i16_reg imm:$lane))),
4045 (SubReg_i16_lane imm:$lane)))>;
4046 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
4047 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
4049 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
4050 (v2i32 (EXTRACT_SUBREG QPR:$src2,
4051 (DSubReg_i32_reg imm:$lane))),
4052 (SubReg_i32_lane imm:$lane)))>;
4054 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
4055 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, N3RegFrm,
4056 IIC_VMULi16D,IIC_VMULi32D,IIC_VMULi16Q,IIC_VMULi32Q,
4057 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
4058 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
4059 IIC_VMULi16Q, IIC_VMULi32Q,
4060 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
4061 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
4062 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
4064 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
4065 (v4i16 (EXTRACT_SUBREG QPR:$src2,
4066 (DSubReg_i16_reg imm:$lane))),
4067 (SubReg_i16_lane imm:$lane)))>;
4068 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
4069 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
4071 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
4072 (v2i32 (EXTRACT_SUBREG QPR:$src2,
4073 (DSubReg_i32_reg imm:$lane))),
4074 (SubReg_i32_lane imm:$lane)))>;
4076 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
4077 defm VMULLs : N3VL_QHS<0,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
4078 "vmull", "s", NEONvmulls, 1>;
4079 defm VMULLu : N3VL_QHS<1,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
4080 "vmull", "u", NEONvmullu, 1>;
4081 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
4082 v8i16, v8i8, int_arm_neon_vmullp, 1>;
4083 defm VMULLsls : N3VLSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s", NEONvmulls>;
4084 defm VMULLslu : N3VLSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u", NEONvmullu>;
4086 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
4087 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, IIC_VMULi32D,
4088 "vqdmull", "s", int_arm_neon_vqdmull, 1>;
4089 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D,
4090 "vqdmull", "s", int_arm_neon_vqdmull>;
4092 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
4094 // VMLA : Vector Multiply Accumulate (integer and floating-point)
4095 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
4096 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
4097 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
4098 v2f32, fmul_su, fadd_mlx>,
4099 Requires<[HasNEON, UseFPVMLx, DontUseFusedMAC]>;
4100 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
4101 v4f32, fmul_su, fadd_mlx>,
4102 Requires<[HasNEON, UseFPVMLx, DontUseFusedMAC]>;
4103 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
4104 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
4105 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
4106 v2f32, fmul_su, fadd_mlx>,
4107 Requires<[HasNEON, UseFPVMLx]>;
4108 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
4109 v4f32, v2f32, fmul_su, fadd_mlx>,
4110 Requires<[HasNEON, UseFPVMLx]>;
4112 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
4113 (mul (v8i16 QPR:$src2),
4114 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
4115 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
4116 (v4i16 (EXTRACT_SUBREG QPR:$src3,
4117 (DSubReg_i16_reg imm:$lane))),
4118 (SubReg_i16_lane imm:$lane)))>;
4120 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
4121 (mul (v4i32 QPR:$src2),
4122 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
4123 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
4124 (v2i32 (EXTRACT_SUBREG QPR:$src3,
4125 (DSubReg_i32_reg imm:$lane))),
4126 (SubReg_i32_lane imm:$lane)))>;
4128 def : Pat<(v4f32 (fadd_mlx (v4f32 QPR:$src1),
4129 (fmul_su (v4f32 QPR:$src2),
4130 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
4131 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
4133 (v2f32 (EXTRACT_SUBREG QPR:$src3,
4134 (DSubReg_i32_reg imm:$lane))),
4135 (SubReg_i32_lane imm:$lane)))>,
4136 Requires<[HasNEON, UseFPVMLx]>;
4138 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
4139 defm VMLALs : N3VLMulOp_QHS<0,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
4140 "vmlal", "s", NEONvmulls, add>;
4141 defm VMLALu : N3VLMulOp_QHS<1,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
4142 "vmlal", "u", NEONvmullu, add>;
4144 defm VMLALsls : N3VLMulOpSL_HS<0, 0b0010, "vmlal", "s", NEONvmulls, add>;
4145 defm VMLALslu : N3VLMulOpSL_HS<1, 0b0010, "vmlal", "u", NEONvmullu, add>;
4147 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
4148 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
4149 "vqdmlal", "s", int_arm_neon_vqdmlal>;
4150 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", int_arm_neon_vqdmlal>;
4152 // VMLS : Vector Multiply Subtract (integer and floating-point)
4153 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
4154 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
4155 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
4156 v2f32, fmul_su, fsub_mlx>,
4157 Requires<[HasNEON, UseFPVMLx, DontUseFusedMAC]>;
4158 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
4159 v4f32, fmul_su, fsub_mlx>,
4160 Requires<[HasNEON, UseFPVMLx, DontUseFusedMAC]>;
4161 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
4162 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
4163 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
4164 v2f32, fmul_su, fsub_mlx>,
4165 Requires<[HasNEON, UseFPVMLx]>;
4166 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
4167 v4f32, v2f32, fmul_su, fsub_mlx>,
4168 Requires<[HasNEON, UseFPVMLx]>;
4170 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
4171 (mul (v8i16 QPR:$src2),
4172 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
4173 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
4174 (v4i16 (EXTRACT_SUBREG QPR:$src3,
4175 (DSubReg_i16_reg imm:$lane))),
4176 (SubReg_i16_lane imm:$lane)))>;
4178 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
4179 (mul (v4i32 QPR:$src2),
4180 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
4181 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
4182 (v2i32 (EXTRACT_SUBREG QPR:$src3,
4183 (DSubReg_i32_reg imm:$lane))),
4184 (SubReg_i32_lane imm:$lane)))>;
4186 def : Pat<(v4f32 (fsub_mlx (v4f32 QPR:$src1),
4187 (fmul_su (v4f32 QPR:$src2),
4188 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
4189 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
4190 (v2f32 (EXTRACT_SUBREG QPR:$src3,
4191 (DSubReg_i32_reg imm:$lane))),
4192 (SubReg_i32_lane imm:$lane)))>,
4193 Requires<[HasNEON, UseFPVMLx]>;
4195 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
4196 defm VMLSLs : N3VLMulOp_QHS<0,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
4197 "vmlsl", "s", NEONvmulls, sub>;
4198 defm VMLSLu : N3VLMulOp_QHS<1,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
4199 "vmlsl", "u", NEONvmullu, sub>;
4201 defm VMLSLsls : N3VLMulOpSL_HS<0, 0b0110, "vmlsl", "s", NEONvmulls, sub>;
4202 defm VMLSLslu : N3VLMulOpSL_HS<1, 0b0110, "vmlsl", "u", NEONvmullu, sub>;
4204 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
4205 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, IIC_VMACi16D, IIC_VMACi32D,
4206 "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
4207 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
4209 // Fused Vector Multiply-Accumulate and Fused Multiply-Subtract Operations.
4210 def VFMAfd : N3VDMulOp<0, 0, 0b00, 0b1100, 1, IIC_VFMACD, "vfma", "f32",
4211 v2f32, fmul_su, fadd_mlx>,
4212 Requires<[HasVFP4,UseFusedMAC]>;
4214 def VFMAfq : N3VQMulOp<0, 0, 0b00, 0b1100, 1, IIC_VFMACQ, "vfma", "f32",
4215 v4f32, fmul_su, fadd_mlx>,
4216 Requires<[HasVFP4,UseFusedMAC]>;
4218 // Fused Vector Multiply Subtract (floating-point)
4219 def VFMSfd : N3VDMulOp<0, 0, 0b10, 0b1100, 1, IIC_VFMACD, "vfms", "f32",
4220 v2f32, fmul_su, fsub_mlx>,
4221 Requires<[HasVFP4,UseFusedMAC]>;
4222 def VFMSfq : N3VQMulOp<0, 0, 0b10, 0b1100, 1, IIC_VFMACQ, "vfms", "f32",
4223 v4f32, fmul_su, fsub_mlx>,
4224 Requires<[HasVFP4,UseFusedMAC]>;
4226 // Match @llvm.fma.* intrinsics
4227 def : Pat<(v2f32 (fma DPR:$Vn, DPR:$Vm, DPR:$src1)),
4228 (VFMAfd DPR:$src1, DPR:$Vn, DPR:$Vm)>,
4229 Requires<[HasVFP4]>;
4230 def : Pat<(v4f32 (fma QPR:$Vn, QPR:$Vm, QPR:$src1)),
4231 (VFMAfq QPR:$src1, QPR:$Vn, QPR:$Vm)>,
4232 Requires<[HasVFP4]>;
4233 def : Pat<(v2f32 (fma (fneg DPR:$Vn), DPR:$Vm, DPR:$src1)),
4234 (VFMSfd DPR:$src1, DPR:$Vn, DPR:$Vm)>,
4235 Requires<[HasVFP4]>;
4236 def : Pat<(v4f32 (fma (fneg QPR:$Vn), QPR:$Vm, QPR:$src1)),
4237 (VFMSfq QPR:$src1, QPR:$Vn, QPR:$Vm)>,
4238 Requires<[HasVFP4]>;
4240 // Vector Subtract Operations.
4242 // VSUB : Vector Subtract (integer and floating-point)
4243 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
4244 "vsub", "i", sub, 0>;
4245 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
4246 v2f32, v2f32, fsub, 0>;
4247 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
4248 v4f32, v4f32, fsub, 0>;
4249 // VSUBL : Vector Subtract Long (Q = D - D)
4250 defm VSUBLs : N3VLExt_QHS<0,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
4251 "vsubl", "s", sub, sext, 0>;
4252 defm VSUBLu : N3VLExt_QHS<1,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
4253 "vsubl", "u", sub, zext, 0>;
4254 // VSUBW : Vector Subtract Wide (Q = Q - D)
4255 defm VSUBWs : N3VW_QHS<0,1,0b0011,0, "vsubw", "s", sub, sext, 0>;
4256 defm VSUBWu : N3VW_QHS<1,1,0b0011,0, "vsubw", "u", sub, zext, 0>;
4257 // VHSUB : Vector Halving Subtract
4258 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, N3RegFrm,
4259 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4260 "vhsub", "s", int_arm_neon_vhsubs, 0>;
4261 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, N3RegFrm,
4262 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4263 "vhsub", "u", int_arm_neon_vhsubu, 0>;
4264 // VQSUB : Vector Saturing Subtract
4265 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, N3RegFrm,
4266 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4267 "vqsub", "s", int_arm_neon_vqsubs, 0>;
4268 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, N3RegFrm,
4269 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4270 "vqsub", "u", int_arm_neon_vqsubu, 0>;
4271 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
4272 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i",
4273 int_arm_neon_vsubhn, 0>;
4274 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
4275 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
4276 int_arm_neon_vrsubhn, 0>;
4278 // Vector Comparisons.
4280 // VCEQ : Vector Compare Equal
4281 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
4282 IIC_VSUBi4Q, "vceq", "i", NEONvceq, 1>;
4283 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
4285 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
4288 let TwoOperandAliasConstraint = "$Vm = $Vd" in
4289 defm VCEQz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
4290 "$Vd, $Vm, #0", NEONvceqz>;
4292 // VCGE : Vector Compare Greater Than or Equal
4293 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
4294 IIC_VSUBi4Q, "vcge", "s", NEONvcge, 0>;
4295 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
4296 IIC_VSUBi4Q, "vcge", "u", NEONvcgeu, 0>;
4297 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32", v2i32, v2f32,
4299 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
4302 let TwoOperandAliasConstraint = "$Vm = $Vd" in {
4303 defm VCGEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
4304 "$Vd, $Vm, #0", NEONvcgez>;
4305 defm VCLEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
4306 "$Vd, $Vm, #0", NEONvclez>;
4309 // VCGT : Vector Compare Greater Than
4310 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
4311 IIC_VSUBi4Q, "vcgt", "s", NEONvcgt, 0>;
4312 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
4313 IIC_VSUBi4Q, "vcgt", "u", NEONvcgtu, 0>;
4314 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
4316 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
4319 let TwoOperandAliasConstraint = "$Vm = $Vd" in {
4320 defm VCGTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
4321 "$Vd, $Vm, #0", NEONvcgtz>;
4322 defm VCLTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
4323 "$Vd, $Vm, #0", NEONvcltz>;
4326 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
4327 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacge",
4328 "f32", v2i32, v2f32, int_arm_neon_vacged, 0>;
4329 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacge",
4330 "f32", v4i32, v4f32, int_arm_neon_vacgeq, 0>;
4331 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
4332 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacgt",
4333 "f32", v2i32, v2f32, int_arm_neon_vacgtd, 0>;
4334 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacgt",
4335 "f32", v4i32, v4f32, int_arm_neon_vacgtq, 0>;
4336 // VTST : Vector Test Bits
4337 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
4338 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
4340 def: NEONInstAlias<"vaclt${p}.f32 $Vd, $Vn, $Vm",
4341 (VACGTd DPR:$Vd, DPR:$Vm, DPR:$Vn, pred:$p)>;
4342 def: NEONInstAlias<"vaclt${p}.f32 $Vd, $Vn, $Vm",
4343 (VACGTq QPR:$Vd, QPR:$Vm, QPR:$Vn, pred:$p)>;
4344 def: NEONInstAlias<"vacle${p}.f32 $Vd, $Vn, $Vm",
4345 (VACGEd DPR:$Vd, DPR:$Vm, DPR:$Vn, pred:$p)>;
4346 def: NEONInstAlias<"vacle${p}.f32 $Vd, $Vn, $Vm",
4347 (VACGEq QPR:$Vd, QPR:$Vm, QPR:$Vn, pred:$p)>;
4349 def: NEONInstAlias<"vaclt${p}.f32 $Vd, $Vm",
4350 (VACGTd DPR:$Vd, DPR:$Vm, DPR:$Vd, pred:$p)>;
4351 def: NEONInstAlias<"vaclt${p}.f32 $Vd, $Vm",
4352 (VACGTq QPR:$Vd, QPR:$Vm, QPR:$Vd, pred:$p)>;
4353 def: NEONInstAlias<"vacle${p}.f32 $Vd, $Vm",
4354 (VACGEd DPR:$Vd, DPR:$Vm, DPR:$Vd, pred:$p)>;
4355 def: NEONInstAlias<"vacle${p}.f32 $Vd, $Vm",
4356 (VACGEq QPR:$Vd, QPR:$Vm, QPR:$Vd, pred:$p)>;
4358 // Vector Bitwise Operations.
4360 def vnotd : PatFrag<(ops node:$in),
4361 (xor node:$in, (bitconvert (v8i8 NEONimmAllOnesV)))>;
4362 def vnotq : PatFrag<(ops node:$in),
4363 (xor node:$in, (bitconvert (v16i8 NEONimmAllOnesV)))>;
4366 // VAND : Vector Bitwise AND
4367 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
4368 v2i32, v2i32, and, 1>;
4369 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
4370 v4i32, v4i32, and, 1>;
4372 // VEOR : Vector Bitwise Exclusive OR
4373 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
4374 v2i32, v2i32, xor, 1>;
4375 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
4376 v4i32, v4i32, xor, 1>;
4378 // VORR : Vector Bitwise OR
4379 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
4380 v2i32, v2i32, or, 1>;
4381 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
4382 v4i32, v4i32, or, 1>;
4384 def VORRiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 0, 1,
4385 (outs DPR:$Vd), (ins nImmSplatI16:$SIMM, DPR:$src),
4387 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
4389 (v4i16 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
4390 let Inst{9} = SIMM{9};
4393 def VORRiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 0, 1,
4394 (outs DPR:$Vd), (ins nImmSplatI32:$SIMM, DPR:$src),
4396 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
4398 (v2i32 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
4399 let Inst{10-9} = SIMM{10-9};
4402 def VORRiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 0, 1,
4403 (outs QPR:$Vd), (ins nImmSplatI16:$SIMM, QPR:$src),
4405 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
4407 (v8i16 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
4408 let Inst{9} = SIMM{9};
4411 def VORRiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 0, 1,
4412 (outs QPR:$Vd), (ins nImmSplatI32:$SIMM, QPR:$src),
4414 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
4416 (v4i32 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
4417 let Inst{10-9} = SIMM{10-9};
4421 // VBIC : Vector Bitwise Bit Clear (AND NOT)
4422 let TwoOperandAliasConstraint = "$Vn = $Vd" in {
4423 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
4424 (ins DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VBINiD,
4425 "vbic", "$Vd, $Vn, $Vm", "",
4426 [(set DPR:$Vd, (v2i32 (and DPR:$Vn,
4427 (vnotd DPR:$Vm))))]>;
4428 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
4429 (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINiQ,
4430 "vbic", "$Vd, $Vn, $Vm", "",
4431 [(set QPR:$Vd, (v4i32 (and QPR:$Vn,
4432 (vnotq QPR:$Vm))))]>;
4435 def VBICiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 1, 1,
4436 (outs DPR:$Vd), (ins nImmSplatI16:$SIMM, DPR:$src),
4438 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
4440 (v4i16 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
4441 let Inst{9} = SIMM{9};
4444 def VBICiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 1, 1,
4445 (outs DPR:$Vd), (ins nImmSplatI32:$SIMM, DPR:$src),
4447 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
4449 (v2i32 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
4450 let Inst{10-9} = SIMM{10-9};
4453 def VBICiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 1, 1,
4454 (outs QPR:$Vd), (ins nImmSplatI16:$SIMM, QPR:$src),
4456 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
4458 (v8i16 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
4459 let Inst{9} = SIMM{9};
4462 def VBICiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 1, 1,
4463 (outs QPR:$Vd), (ins nImmSplatI32:$SIMM, QPR:$src),
4465 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
4467 (v4i32 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
4468 let Inst{10-9} = SIMM{10-9};
4471 // VORN : Vector Bitwise OR NOT
4472 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$Vd),
4473 (ins DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VBINiD,
4474 "vorn", "$Vd, $Vn, $Vm", "",
4475 [(set DPR:$Vd, (v2i32 (or DPR:$Vn,
4476 (vnotd DPR:$Vm))))]>;
4477 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$Vd),
4478 (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINiQ,
4479 "vorn", "$Vd, $Vn, $Vm", "",
4480 [(set QPR:$Vd, (v4i32 (or QPR:$Vn,
4481 (vnotq QPR:$Vm))))]>;
4483 // VMVN : Vector Bitwise NOT (Immediate)
4485 let isReMaterializable = 1 in {
4487 def VMVNv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 1, 1, (outs DPR:$Vd),
4488 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4489 "vmvn", "i16", "$Vd, $SIMM", "",
4490 [(set DPR:$Vd, (v4i16 (NEONvmvnImm timm:$SIMM)))]> {
4491 let Inst{9} = SIMM{9};
4494 def VMVNv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 1, 1, (outs QPR:$Vd),
4495 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4496 "vmvn", "i16", "$Vd, $SIMM", "",
4497 [(set QPR:$Vd, (v8i16 (NEONvmvnImm timm:$SIMM)))]> {
4498 let Inst{9} = SIMM{9};
4501 def VMVNv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 1, 1, (outs DPR:$Vd),
4502 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4503 "vmvn", "i32", "$Vd, $SIMM", "",
4504 [(set DPR:$Vd, (v2i32 (NEONvmvnImm timm:$SIMM)))]> {
4505 let Inst{11-8} = SIMM{11-8};
4508 def VMVNv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 1, 1, (outs QPR:$Vd),
4509 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4510 "vmvn", "i32", "$Vd, $SIMM", "",
4511 [(set QPR:$Vd, (v4i32 (NEONvmvnImm timm:$SIMM)))]> {
4512 let Inst{11-8} = SIMM{11-8};
4516 // VMVN : Vector Bitwise NOT
4517 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
4518 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VSUBiD,
4519 "vmvn", "$Vd, $Vm", "",
4520 [(set DPR:$Vd, (v2i32 (vnotd DPR:$Vm)))]>;
4521 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
4522 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VSUBiD,
4523 "vmvn", "$Vd, $Vm", "",
4524 [(set QPR:$Vd, (v4i32 (vnotq QPR:$Vm)))]>;
4525 def : Pat<(v2i32 (vnotd DPR:$src)), (VMVNd DPR:$src)>;
4526 def : Pat<(v4i32 (vnotq QPR:$src)), (VMVNq QPR:$src)>;
4528 // VBSL : Vector Bitwise Select
4529 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
4530 (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4531 N3RegFrm, IIC_VCNTiD,
4532 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4534 (v2i32 (NEONvbsl DPR:$src1, DPR:$Vn, DPR:$Vm)))]>;
4535 def : Pat<(v8i8 (int_arm_neon_vbsl (v8i8 DPR:$src1),
4536 (v8i8 DPR:$Vn), (v8i8 DPR:$Vm))),
4537 (VBSLd DPR:$src1, DPR:$Vn, DPR:$Vm)>,
4538 Requires<[HasNEON]>;
4539 def : Pat<(v4i16 (int_arm_neon_vbsl (v4i16 DPR:$src1),
4540 (v4i16 DPR:$Vn), (v4i16 DPR:$Vm))),
4541 (VBSLd DPR:$src1, DPR:$Vn, DPR:$Vm)>,
4542 Requires<[HasNEON]>;
4543 def : Pat<(v2i32 (int_arm_neon_vbsl (v2i32 DPR:$src1),
4544 (v2i32 DPR:$Vn), (v2i32 DPR:$Vm))),
4545 (VBSLd DPR:$src1, DPR:$Vn, DPR:$Vm)>,
4546 Requires<[HasNEON]>;
4547 def : Pat<(v2f32 (int_arm_neon_vbsl (v2f32 DPR:$src1),
4548 (v2f32 DPR:$Vn), (v2f32 DPR:$Vm))),
4549 (VBSLd DPR:$src1, DPR:$Vn, DPR:$Vm)>,
4550 Requires<[HasNEON]>;
4551 def : Pat<(v1i64 (int_arm_neon_vbsl (v1i64 DPR:$src1),
4552 (v1i64 DPR:$Vn), (v1i64 DPR:$Vm))),
4553 (VBSLd DPR:$src1, DPR:$Vn, DPR:$Vm)>,
4554 Requires<[HasNEON]>;
4556 def : Pat<(v2i32 (or (and DPR:$Vn, DPR:$Vd),
4557 (and DPR:$Vm, (vnotd DPR:$Vd)))),
4558 (VBSLd DPR:$Vd, DPR:$Vn, DPR:$Vm)>,
4559 Requires<[HasNEON]>;
4561 def : Pat<(v1i64 (or (and DPR:$Vn, DPR:$Vd),
4562 (and DPR:$Vm, (vnotd DPR:$Vd)))),
4563 (VBSLd DPR:$Vd, DPR:$Vn, DPR:$Vm)>,
4564 Requires<[HasNEON]>;
4566 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
4567 (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4568 N3RegFrm, IIC_VCNTiQ,
4569 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4571 (v4i32 (NEONvbsl QPR:$src1, QPR:$Vn, QPR:$Vm)))]>;
4573 def : Pat<(v16i8 (int_arm_neon_vbsl (v16i8 QPR:$src1),
4574 (v16i8 QPR:$Vn), (v16i8 QPR:$Vm))),
4575 (VBSLq QPR:$src1, QPR:$Vn, QPR:$Vm)>,
4576 Requires<[HasNEON]>;
4577 def : Pat<(v8i16 (int_arm_neon_vbsl (v8i16 QPR:$src1),
4578 (v8i16 QPR:$Vn), (v8i16 QPR:$Vm))),
4579 (VBSLq QPR:$src1, QPR:$Vn, QPR:$Vm)>,
4580 Requires<[HasNEON]>;
4581 def : Pat<(v4i32 (int_arm_neon_vbsl (v4i32 QPR:$src1),
4582 (v4i32 QPR:$Vn), (v4i32 QPR:$Vm))),
4583 (VBSLq QPR:$src1, QPR:$Vn, QPR:$Vm)>,
4584 Requires<[HasNEON]>;
4585 def : Pat<(v4f32 (int_arm_neon_vbsl (v4f32 QPR:$src1),
4586 (v4f32 QPR:$Vn), (v4f32 QPR:$Vm))),
4587 (VBSLq QPR:$src1, QPR:$Vn, QPR:$Vm)>,
4588 Requires<[HasNEON]>;
4589 def : Pat<(v2i64 (int_arm_neon_vbsl (v2i64 QPR:$src1),
4590 (v2i64 QPR:$Vn), (v2i64 QPR:$Vm))),
4591 (VBSLq QPR:$src1, QPR:$Vn, QPR:$Vm)>,
4592 Requires<[HasNEON]>;
4594 def : Pat<(v4i32 (or (and QPR:$Vn, QPR:$Vd),
4595 (and QPR:$Vm, (vnotq QPR:$Vd)))),
4596 (VBSLq QPR:$Vd, QPR:$Vn, QPR:$Vm)>,
4597 Requires<[HasNEON]>;
4598 def : Pat<(v2i64 (or (and QPR:$Vn, QPR:$Vd),
4599 (and QPR:$Vm, (vnotq QPR:$Vd)))),
4600 (VBSLq QPR:$Vd, QPR:$Vn, QPR:$Vm)>,
4601 Requires<[HasNEON]>;
4603 // VBIF : Vector Bitwise Insert if False
4604 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
4605 // FIXME: This instruction's encoding MAY NOT BE correct.
4606 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
4607 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4608 N3RegFrm, IIC_VBINiD,
4609 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4611 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
4612 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4613 N3RegFrm, IIC_VBINiQ,
4614 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4617 // VBIT : Vector Bitwise Insert if True
4618 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
4619 // FIXME: This instruction's encoding MAY NOT BE correct.
4620 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
4621 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4622 N3RegFrm, IIC_VBINiD,
4623 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4625 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
4626 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4627 N3RegFrm, IIC_VBINiQ,
4628 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4631 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
4632 // for equivalent operations with different register constraints; it just
4635 // Vector Absolute Differences.
4637 // VABD : Vector Absolute Difference
4638 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, N3RegFrm,
4639 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4640 "vabd", "s", int_arm_neon_vabds, 1>;
4641 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, N3RegFrm,
4642 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4643 "vabd", "u", int_arm_neon_vabdu, 1>;
4644 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBIND,
4645 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 1>;
4646 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBINQ,
4647 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 1>;
4649 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
4650 defm VABDLs : N3VLIntExt_QHS<0,1,0b0111,0, IIC_VSUBi4Q,
4651 "vabdl", "s", int_arm_neon_vabds, zext, 1>;
4652 defm VABDLu : N3VLIntExt_QHS<1,1,0b0111,0, IIC_VSUBi4Q,
4653 "vabdl", "u", int_arm_neon_vabdu, zext, 1>;
4655 // VABA : Vector Absolute Difference and Accumulate
4656 defm VABAs : N3VIntOp_QHS<0,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
4657 "vaba", "s", int_arm_neon_vabds, add>;
4658 defm VABAu : N3VIntOp_QHS<1,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
4659 "vaba", "u", int_arm_neon_vabdu, add>;
4661 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
4662 defm VABALs : N3VLIntExtOp_QHS<0,1,0b0101,0, IIC_VABAD,
4663 "vabal", "s", int_arm_neon_vabds, zext, add>;
4664 defm VABALu : N3VLIntExtOp_QHS<1,1,0b0101,0, IIC_VABAD,
4665 "vabal", "u", int_arm_neon_vabdu, zext, add>;
4667 // Vector Maximum and Minimum.
4669 // VMAX : Vector Maximum
4670 defm VMAXs : N3VInt_QHS<0, 0, 0b0110, 0, N3RegFrm,
4671 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4672 "vmax", "s", int_arm_neon_vmaxs, 1>;
4673 defm VMAXu : N3VInt_QHS<1, 0, 0b0110, 0, N3RegFrm,
4674 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4675 "vmax", "u", int_arm_neon_vmaxu, 1>;
4676 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBIND,
4678 v2f32, v2f32, int_arm_neon_vmaxs, 1>;
4679 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBINQ,
4681 v4f32, v4f32, int_arm_neon_vmaxs, 1>;
4684 let PostEncoderMethod = "NEONThumb2V8PostEncoder", DecoderNamespace = "v8NEON" in {
4685 def VMAXNMND : N3VDIntnp<0b000110, 0b00, 0b1111, 0, 1,
4686 N3RegFrm, NoItinerary, "vmaxnm", "f32",
4687 v2f32, v2f32, int_arm_neon_vmaxnm, 1>,
4688 Requires<[HasV8, HasNEON]>;
4689 def VMAXNMNQ : N3VQIntnp<0b00110, 0b00, 0b1111, 1, 1,
4690 N3RegFrm, NoItinerary, "vmaxnm", "f32",
4691 v4f32, v4f32, int_arm_neon_vmaxnm, 1>,
4692 Requires<[HasV8, HasNEON]>;
4695 // VMIN : Vector Minimum
4696 defm VMINs : N3VInt_QHS<0, 0, 0b0110, 1, N3RegFrm,
4697 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4698 "vmin", "s", int_arm_neon_vmins, 1>;
4699 defm VMINu : N3VInt_QHS<1, 0, 0b0110, 1, N3RegFrm,
4700 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4701 "vmin", "u", int_arm_neon_vminu, 1>;
4702 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBIND,
4704 v2f32, v2f32, int_arm_neon_vmins, 1>;
4705 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBINQ,
4707 v4f32, v4f32, int_arm_neon_vmins, 1>;
4710 let PostEncoderMethod = "NEONThumb2V8PostEncoder", DecoderNamespace = "v8NEON" in {
4711 def VMINNMND : N3VDIntnp<0b00110, 0b10, 0b1111, 0, 1,
4712 N3RegFrm, NoItinerary, "vminnm", "f32",
4713 v2f32, v2f32, int_arm_neon_vminnm, 1>,
4714 Requires<[HasV8, HasNEON]>;
4715 def VMINNMNQ : N3VQIntnp<0b00110, 0b10, 0b1111, 1, 1,
4716 N3RegFrm, NoItinerary, "vminnm", "f32",
4717 v4f32, v4f32, int_arm_neon_vminnm, 1>,
4718 Requires<[HasV8, HasNEON]>;
4721 // Vector Pairwise Operations.
4723 // VPADD : Vector Pairwise Add
4724 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4726 v8i8, v8i8, int_arm_neon_vpadd, 0>;
4727 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4729 v4i16, v4i16, int_arm_neon_vpadd, 0>;
4730 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4732 v2i32, v2i32, int_arm_neon_vpadd, 0>;
4733 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, N3RegFrm,
4734 IIC_VPBIND, "vpadd", "f32",
4735 v2f32, v2f32, int_arm_neon_vpadd, 0>;
4737 // VPADDL : Vector Pairwise Add Long
4738 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
4739 int_arm_neon_vpaddls>;
4740 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
4741 int_arm_neon_vpaddlu>;
4743 // VPADAL : Vector Pairwise Add and Accumulate Long
4744 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
4745 int_arm_neon_vpadals>;
4746 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
4747 int_arm_neon_vpadalu>;
4749 // VPMAX : Vector Pairwise Maximum
4750 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4751 "s8", v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
4752 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4753 "s16", v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
4754 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4755 "s32", v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
4756 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4757 "u8", v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
4758 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4759 "u16", v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
4760 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4761 "u32", v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
4762 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmax",
4763 "f32", v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
4765 // VPMIN : Vector Pairwise Minimum
4766 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4767 "s8", v8i8, v8i8, int_arm_neon_vpmins, 0>;
4768 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4769 "s16", v4i16, v4i16, int_arm_neon_vpmins, 0>;
4770 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4771 "s32", v2i32, v2i32, int_arm_neon_vpmins, 0>;
4772 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4773 "u8", v8i8, v8i8, int_arm_neon_vpminu, 0>;
4774 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4775 "u16", v4i16, v4i16, int_arm_neon_vpminu, 0>;
4776 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4777 "u32", v2i32, v2i32, int_arm_neon_vpminu, 0>;
4778 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmin",
4779 "f32", v2f32, v2f32, int_arm_neon_vpmins, 0>;
4781 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
4783 // VRECPE : Vector Reciprocal Estimate
4784 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
4785 IIC_VUNAD, "vrecpe", "u32",
4786 v2i32, v2i32, int_arm_neon_vrecpe>;
4787 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
4788 IIC_VUNAQ, "vrecpe", "u32",
4789 v4i32, v4i32, int_arm_neon_vrecpe>;
4790 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
4791 IIC_VUNAD, "vrecpe", "f32",
4792 v2f32, v2f32, int_arm_neon_vrecpe>;
4793 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
4794 IIC_VUNAQ, "vrecpe", "f32",
4795 v4f32, v4f32, int_arm_neon_vrecpe>;
4797 // VRECPS : Vector Reciprocal Step
4798 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
4799 IIC_VRECSD, "vrecps", "f32",
4800 v2f32, v2f32, int_arm_neon_vrecps, 1>;
4801 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
4802 IIC_VRECSQ, "vrecps", "f32",
4803 v4f32, v4f32, int_arm_neon_vrecps, 1>;
4805 // VRSQRTE : Vector Reciprocal Square Root Estimate
4806 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
4807 IIC_VUNAD, "vrsqrte", "u32",
4808 v2i32, v2i32, int_arm_neon_vrsqrte>;
4809 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
4810 IIC_VUNAQ, "vrsqrte", "u32",
4811 v4i32, v4i32, int_arm_neon_vrsqrte>;
4812 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
4813 IIC_VUNAD, "vrsqrte", "f32",
4814 v2f32, v2f32, int_arm_neon_vrsqrte>;
4815 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
4816 IIC_VUNAQ, "vrsqrte", "f32",
4817 v4f32, v4f32, int_arm_neon_vrsqrte>;
4819 // VRSQRTS : Vector Reciprocal Square Root Step
4820 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
4821 IIC_VRECSD, "vrsqrts", "f32",
4822 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
4823 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
4824 IIC_VRECSQ, "vrsqrts", "f32",
4825 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
4829 // VSHL : Vector Shift
4830 defm VSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 0, N3RegVShFrm,
4831 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
4832 "vshl", "s", int_arm_neon_vshifts>;
4833 defm VSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 0, N3RegVShFrm,
4834 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
4835 "vshl", "u", int_arm_neon_vshiftu>;
4837 // VSHL : Vector Shift Left (Immediate)
4838 defm VSHLi : N2VShL_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl>;
4840 // VSHR : Vector Shift Right (Immediate)
4841 defm VSHRs : N2VShR_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s", "VSHRs",
4843 defm VSHRu : N2VShR_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u", "VSHRu",
4846 // VSHLL : Vector Shift Left Long
4847 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>;
4848 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>;
4850 // VSHLL : Vector Shift Left Long (with maximum shift count)
4851 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
4852 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
4853 ValueType OpTy, Operand ImmTy, SDNode OpNode>
4854 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
4855 ResTy, OpTy, ImmTy, OpNode> {
4856 let Inst{21-16} = op21_16;
4857 let DecoderMethod = "DecodeVSHLMaxInstruction";
4859 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
4860 v8i16, v8i8, imm8, NEONvshlli>;
4861 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
4862 v4i32, v4i16, imm16, NEONvshlli>;
4863 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
4864 v2i64, v2i32, imm32, NEONvshlli>;
4866 // VSHRN : Vector Shift Right and Narrow
4867 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
4870 // VRSHL : Vector Rounding Shift
4871 defm VRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 0, N3RegVShFrm,
4872 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4873 "vrshl", "s", int_arm_neon_vrshifts>;
4874 defm VRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 0, N3RegVShFrm,
4875 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4876 "vrshl", "u", int_arm_neon_vrshiftu>;
4877 // VRSHR : Vector Rounding Shift Right
4878 defm VRSHRs : N2VShR_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s", "VRSHRs",
4880 defm VRSHRu : N2VShR_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u", "VRSHRu",
4883 // VRSHRN : Vector Rounding Shift Right and Narrow
4884 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
4887 // VQSHL : Vector Saturating Shift
4888 defm VQSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 1, N3RegVShFrm,
4889 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4890 "vqshl", "s", int_arm_neon_vqshifts>;
4891 defm VQSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 1, N3RegVShFrm,
4892 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4893 "vqshl", "u", int_arm_neon_vqshiftu>;
4894 // VQSHL : Vector Saturating Shift Left (Immediate)
4895 defm VQSHLsi : N2VShL_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s",NEONvqshls>;
4896 defm VQSHLui : N2VShL_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u",NEONvqshlu>;
4898 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
4899 defm VQSHLsu : N2VShL_QHSD<1,1,0b0110,1, IIC_VSHLi4D,"vqshlu","s",NEONvqshlsu>;
4901 // VQSHRN : Vector Saturating Shift Right and Narrow
4902 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
4904 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
4907 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
4908 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
4911 // VQRSHL : Vector Saturating Rounding Shift
4912 defm VQRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 1, N3RegVShFrm,
4913 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4914 "vqrshl", "s", int_arm_neon_vqrshifts>;
4915 defm VQRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 1, N3RegVShFrm,
4916 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4917 "vqrshl", "u", int_arm_neon_vqrshiftu>;
4919 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
4920 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
4922 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
4925 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
4926 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
4929 // VSRA : Vector Shift Right and Accumulate
4930 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
4931 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
4932 // VRSRA : Vector Rounding Shift Right and Accumulate
4933 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
4934 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
4936 // VSLI : Vector Shift Left and Insert
4937 defm VSLI : N2VShInsL_QHSD<1, 1, 0b0101, 1, "vsli">;
4939 // VSRI : Vector Shift Right and Insert
4940 defm VSRI : N2VShInsR_QHSD<1, 1, 0b0100, 1, "vsri">;
4942 // Vector Absolute and Saturating Absolute.
4944 // VABS : Vector Absolute Value
4945 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
4946 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
4948 def VABSfd : N2VD<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
4950 v2f32, v2f32, fabs>;
4951 def VABSfq : N2VQ<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
4953 v4f32, v4f32, fabs>;
4955 def : Pat<(xor (v2i32 (bitconvert (v8i8 (NEONvshrs DPR:$src, (i32 7))))),
4956 (v2i32 (bitconvert (v8i8 (add DPR:$src,
4957 (NEONvshrs DPR:$src, (i32 7))))))),
4958 (VABSv8i8 DPR:$src)>;
4959 def : Pat<(xor (v2i32 (bitconvert (v4i16 (NEONvshrs DPR:$src, (i32 15))))),
4960 (v2i32 (bitconvert (v4i16 (add DPR:$src,
4961 (NEONvshrs DPR:$src, (i32 15))))))),
4962 (VABSv4i16 DPR:$src)>;
4963 def : Pat<(xor (v2i32 (NEONvshrs DPR:$src, (i32 31))),
4964 (v2i32 (add DPR:$src, (NEONvshrs DPR:$src, (i32 31))))),
4965 (VABSv2i32 DPR:$src)>;
4966 def : Pat<(xor (v4i32 (bitconvert (v16i8 (NEONvshrs QPR:$src, (i32 7))))),
4967 (v4i32 (bitconvert (v16i8 (add QPR:$src,
4968 (NEONvshrs QPR:$src, (i32 7))))))),
4969 (VABSv16i8 QPR:$src)>;
4970 def : Pat<(xor (v4i32 (bitconvert (v8i16 (NEONvshrs QPR:$src, (i32 15))))),
4971 (v4i32 (bitconvert (v8i16 (add QPR:$src,
4972 (NEONvshrs QPR:$src, (i32 15))))))),
4973 (VABSv8i16 QPR:$src)>;
4974 def : Pat<(xor (v4i32 (NEONvshrs QPR:$src, (i32 31))),
4975 (v4i32 (add QPR:$src, (NEONvshrs QPR:$src, (i32 31))))),
4976 (VABSv4i32 QPR:$src)>;
4978 def : Pat<(v2f32 (int_arm_neon_vabs (v2f32 DPR:$src))), (VABSfd DPR:$src)>;
4979 def : Pat<(v4f32 (int_arm_neon_vabs (v4f32 QPR:$src))), (VABSfq QPR:$src)>;
4981 // VQABS : Vector Saturating Absolute Value
4982 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
4983 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
4984 int_arm_neon_vqabs>;
4988 def vnegd : PatFrag<(ops node:$in),
4989 (sub (bitconvert (v2i32 NEONimmAllZerosV)), node:$in)>;
4990 def vnegq : PatFrag<(ops node:$in),
4991 (sub (bitconvert (v4i32 NEONimmAllZerosV)), node:$in)>;
4993 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
4994 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$Vd), (ins DPR:$Vm),
4995 IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
4996 [(set DPR:$Vd, (Ty (vnegd DPR:$Vm)))]>;
4997 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
4998 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$Vd), (ins QPR:$Vm),
4999 IIC_VSHLiQ, OpcodeStr, Dt, "$Vd, $Vm", "",
5000 [(set QPR:$Vd, (Ty (vnegq QPR:$Vm)))]>;
5002 // VNEG : Vector Negate (integer)
5003 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
5004 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
5005 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
5006 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
5007 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
5008 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
5010 // VNEG : Vector Negate (floating-point)
5011 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
5012 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VUNAD,
5013 "vneg", "f32", "$Vd, $Vm", "",
5014 [(set DPR:$Vd, (v2f32 (fneg DPR:$Vm)))]>;
5015 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
5016 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VUNAQ,
5017 "vneg", "f32", "$Vd, $Vm", "",
5018 [(set QPR:$Vd, (v4f32 (fneg QPR:$Vm)))]>;
5020 def : Pat<(v8i8 (vnegd DPR:$src)), (VNEGs8d DPR:$src)>;
5021 def : Pat<(v4i16 (vnegd DPR:$src)), (VNEGs16d DPR:$src)>;
5022 def : Pat<(v2i32 (vnegd DPR:$src)), (VNEGs32d DPR:$src)>;
5023 def : Pat<(v16i8 (vnegq QPR:$src)), (VNEGs8q QPR:$src)>;
5024 def : Pat<(v8i16 (vnegq QPR:$src)), (VNEGs16q QPR:$src)>;
5025 def : Pat<(v4i32 (vnegq QPR:$src)), (VNEGs32q QPR:$src)>;
5027 // VQNEG : Vector Saturating Negate
5028 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
5029 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
5030 int_arm_neon_vqneg>;
5032 // Vector Bit Counting Operations.
5034 // VCLS : Vector Count Leading Sign Bits
5035 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
5036 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
5038 // VCLZ : Vector Count Leading Zeros
5039 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
5040 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
5042 // VCNT : Vector Count One Bits
5043 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
5044 IIC_VCNTiD, "vcnt", "8",
5046 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
5047 IIC_VCNTiQ, "vcnt", "8",
5048 v16i8, v16i8, ctpop>;
5051 def VSWPd : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 0, 0,
5052 (outs DPR:$Vd, DPR:$Vm), (ins DPR:$in1, DPR:$in2),
5053 NoItinerary, "vswp", "$Vd, $Vm", "$in1 = $Vd, $in2 = $Vm",
5055 def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0,
5056 (outs QPR:$Vd, QPR:$Vm), (ins QPR:$in1, QPR:$in2),
5057 NoItinerary, "vswp", "$Vd, $Vm", "$in1 = $Vd, $in2 = $Vm",
5060 // Vector Move Operations.
5062 // VMOV : Vector Move (Register)
5063 def : InstAlias<"vmov${p} $Vd, $Vm",
5064 (VORRd DPR:$Vd, DPR:$Vm, DPR:$Vm, pred:$p)>;
5065 def : InstAlias<"vmov${p} $Vd, $Vm",
5066 (VORRq QPR:$Vd, QPR:$Vm, QPR:$Vm, pred:$p)>;
5068 // VMOV : Vector Move (Immediate)
5070 let isReMaterializable = 1 in {
5071 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$Vd),
5072 (ins nImmSplatI8:$SIMM), IIC_VMOVImm,
5073 "vmov", "i8", "$Vd, $SIMM", "",
5074 [(set DPR:$Vd, (v8i8 (NEONvmovImm timm:$SIMM)))]>;
5075 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$Vd),
5076 (ins nImmSplatI8:$SIMM), IIC_VMOVImm,
5077 "vmov", "i8", "$Vd, $SIMM", "",
5078 [(set QPR:$Vd, (v16i8 (NEONvmovImm timm:$SIMM)))]>;
5080 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 0, 1, (outs DPR:$Vd),
5081 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
5082 "vmov", "i16", "$Vd, $SIMM", "",
5083 [(set DPR:$Vd, (v4i16 (NEONvmovImm timm:$SIMM)))]> {
5084 let Inst{9} = SIMM{9};
5087 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 0, 1, (outs QPR:$Vd),
5088 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
5089 "vmov", "i16", "$Vd, $SIMM", "",
5090 [(set QPR:$Vd, (v8i16 (NEONvmovImm timm:$SIMM)))]> {
5091 let Inst{9} = SIMM{9};
5094 def VMOVv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 0, 1, (outs DPR:$Vd),
5095 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
5096 "vmov", "i32", "$Vd, $SIMM", "",
5097 [(set DPR:$Vd, (v2i32 (NEONvmovImm timm:$SIMM)))]> {
5098 let Inst{11-8} = SIMM{11-8};
5101 def VMOVv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 0, 1, (outs QPR:$Vd),
5102 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
5103 "vmov", "i32", "$Vd, $SIMM", "",
5104 [(set QPR:$Vd, (v4i32 (NEONvmovImm timm:$SIMM)))]> {
5105 let Inst{11-8} = SIMM{11-8};
5108 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$Vd),
5109 (ins nImmSplatI64:$SIMM), IIC_VMOVImm,
5110 "vmov", "i64", "$Vd, $SIMM", "",
5111 [(set DPR:$Vd, (v1i64 (NEONvmovImm timm:$SIMM)))]>;
5112 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$Vd),
5113 (ins nImmSplatI64:$SIMM), IIC_VMOVImm,
5114 "vmov", "i64", "$Vd, $SIMM", "",
5115 [(set QPR:$Vd, (v2i64 (NEONvmovImm timm:$SIMM)))]>;
5117 def VMOVv2f32 : N1ModImm<1, 0b000, 0b1111, 0, 0, 0, 1, (outs DPR:$Vd),
5118 (ins nImmVMOVF32:$SIMM), IIC_VMOVImm,
5119 "vmov", "f32", "$Vd, $SIMM", "",
5120 [(set DPR:$Vd, (v2f32 (NEONvmovFPImm timm:$SIMM)))]>;
5121 def VMOVv4f32 : N1ModImm<1, 0b000, 0b1111, 0, 1, 0, 1, (outs QPR:$Vd),
5122 (ins nImmVMOVF32:$SIMM), IIC_VMOVImm,
5123 "vmov", "f32", "$Vd, $SIMM", "",
5124 [(set QPR:$Vd, (v4f32 (NEONvmovFPImm timm:$SIMM)))]>;
5125 } // isReMaterializable
5127 // VMOV : Vector Get Lane (move scalar to ARM core register)
5129 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
5130 (outs GPR:$R), (ins DPR:$V, VectorIndex8:$lane),
5131 IIC_VMOVSI, "vmov", "s8", "$R, $V$lane",
5132 [(set GPR:$R, (NEONvgetlanes (v8i8 DPR:$V),
5134 let Inst{21} = lane{2};
5135 let Inst{6-5} = lane{1-0};
5137 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
5138 (outs GPR:$R), (ins DPR:$V, VectorIndex16:$lane),
5139 IIC_VMOVSI, "vmov", "s16", "$R, $V$lane",
5140 [(set GPR:$R, (NEONvgetlanes (v4i16 DPR:$V),
5142 let Inst{21} = lane{1};
5143 let Inst{6} = lane{0};
5145 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
5146 (outs GPR:$R), (ins DPR:$V, VectorIndex8:$lane),
5147 IIC_VMOVSI, "vmov", "u8", "$R, $V$lane",
5148 [(set GPR:$R, (NEONvgetlaneu (v8i8 DPR:$V),
5150 let Inst{21} = lane{2};
5151 let Inst{6-5} = lane{1-0};
5153 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
5154 (outs GPR:$R), (ins DPR:$V, VectorIndex16:$lane),
5155 IIC_VMOVSI, "vmov", "u16", "$R, $V$lane",
5156 [(set GPR:$R, (NEONvgetlaneu (v4i16 DPR:$V),
5158 let Inst{21} = lane{1};
5159 let Inst{6} = lane{0};
5161 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
5162 (outs GPR:$R), (ins DPR:$V, VectorIndex32:$lane),
5163 IIC_VMOVSI, "vmov", "32", "$R, $V$lane",
5164 [(set GPR:$R, (extractelt (v2i32 DPR:$V),
5166 Requires<[HasNEON, HasFastVGETLNi32]> {
5167 let Inst{21} = lane{0};
5169 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
5170 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
5171 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
5172 (DSubReg_i8_reg imm:$lane))),
5173 (SubReg_i8_lane imm:$lane))>;
5174 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
5175 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
5176 (DSubReg_i16_reg imm:$lane))),
5177 (SubReg_i16_lane imm:$lane))>;
5178 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
5179 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
5180 (DSubReg_i8_reg imm:$lane))),
5181 (SubReg_i8_lane imm:$lane))>;
5182 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
5183 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
5184 (DSubReg_i16_reg imm:$lane))),
5185 (SubReg_i16_lane imm:$lane))>;
5186 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
5187 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
5188 (DSubReg_i32_reg imm:$lane))),
5189 (SubReg_i32_lane imm:$lane))>,
5190 Requires<[HasNEON, HasFastVGETLNi32]>;
5191 def : Pat<(extractelt (v2i32 DPR:$src), imm:$lane),
5193 (i32 (EXTRACT_SUBREG DPR:$src, (SSubReg_f32_reg imm:$lane))), GPR)>,
5194 Requires<[HasNEON, HasSlowVGETLNi32]>;
5195 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
5197 (i32 (EXTRACT_SUBREG QPR:$src, (SSubReg_f32_reg imm:$lane))), GPR)>,
5198 Requires<[HasNEON, HasSlowVGETLNi32]>;
5199 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
5200 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
5201 (SSubReg_f32_reg imm:$src2))>;
5202 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
5203 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
5204 (SSubReg_f32_reg imm:$src2))>;
5205 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
5206 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
5207 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
5208 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
5211 // VMOV : Vector Set Lane (move ARM core register to scalar)
5213 let Constraints = "$src1 = $V" in {
5214 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$V),
5215 (ins DPR:$src1, GPR:$R, VectorIndex8:$lane),
5216 IIC_VMOVISL, "vmov", "8", "$V$lane, $R",
5217 [(set DPR:$V, (vector_insert (v8i8 DPR:$src1),
5218 GPR:$R, imm:$lane))]> {
5219 let Inst{21} = lane{2};
5220 let Inst{6-5} = lane{1-0};
5222 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$V),
5223 (ins DPR:$src1, GPR:$R, VectorIndex16:$lane),
5224 IIC_VMOVISL, "vmov", "16", "$V$lane, $R",
5225 [(set DPR:$V, (vector_insert (v4i16 DPR:$src1),
5226 GPR:$R, imm:$lane))]> {
5227 let Inst{21} = lane{1};
5228 let Inst{6} = lane{0};
5230 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$V),
5231 (ins DPR:$src1, GPR:$R, VectorIndex32:$lane),
5232 IIC_VMOVISL, "vmov", "32", "$V$lane, $R",
5233 [(set DPR:$V, (insertelt (v2i32 DPR:$src1),
5234 GPR:$R, imm:$lane))]> {
5235 let Inst{21} = lane{0};
5238 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
5239 (v16i8 (INSERT_SUBREG QPR:$src1,
5240 (v8i8 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
5241 (DSubReg_i8_reg imm:$lane))),
5242 GPR:$src2, (SubReg_i8_lane imm:$lane))),
5243 (DSubReg_i8_reg imm:$lane)))>;
5244 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
5245 (v8i16 (INSERT_SUBREG QPR:$src1,
5246 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
5247 (DSubReg_i16_reg imm:$lane))),
5248 GPR:$src2, (SubReg_i16_lane imm:$lane))),
5249 (DSubReg_i16_reg imm:$lane)))>;
5250 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
5251 (v4i32 (INSERT_SUBREG QPR:$src1,
5252 (v2i32 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
5253 (DSubReg_i32_reg imm:$lane))),
5254 GPR:$src2, (SubReg_i32_lane imm:$lane))),
5255 (DSubReg_i32_reg imm:$lane)))>;
5257 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
5258 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
5259 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
5260 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
5261 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
5262 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
5264 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
5265 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
5266 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
5267 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
5269 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
5270 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
5271 def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
5272 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
5273 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
5274 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
5276 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
5277 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
5278 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
5279 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
5280 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
5281 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
5283 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
5284 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
5285 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
5287 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
5288 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
5289 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
5291 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
5292 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
5293 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
5296 // VDUP : Vector Duplicate (from ARM core register to all elements)
5298 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
5299 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$V), (ins GPR:$R),
5300 IIC_VMOVIS, "vdup", Dt, "$V, $R",
5301 [(set DPR:$V, (Ty (NEONvdup (i32 GPR:$R))))]>;
5302 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
5303 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$V), (ins GPR:$R),
5304 IIC_VMOVIS, "vdup", Dt, "$V, $R",
5305 [(set QPR:$V, (Ty (NEONvdup (i32 GPR:$R))))]>;
5307 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
5308 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
5309 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>,
5310 Requires<[HasNEON, HasFastVDUP32]>;
5311 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
5312 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
5313 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
5315 // NEONvdup patterns for uarchs with fast VDUP.32.
5316 def : Pat<(v2f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VDUP32d GPR:$R)>,
5317 Requires<[HasNEON,HasFastVDUP32]>;
5318 def : Pat<(v4f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VDUP32q GPR:$R)>;
5320 // NEONvdup patterns for uarchs with slow VDUP.32 - use VMOVDRR instead.
5321 def : Pat<(v2i32 (NEONvdup (i32 GPR:$R))), (VMOVDRR GPR:$R, GPR:$R)>,
5322 Requires<[HasNEON,HasSlowVDUP32]>;
5323 def : Pat<(v2f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VMOVDRR GPR:$R, GPR:$R)>,
5324 Requires<[HasNEON,HasSlowVDUP32]>;
5326 // VDUP : Vector Duplicate Lane (from scalar to all elements)
5328 class VDUPLND<bits<4> op19_16, string OpcodeStr, string Dt,
5329 ValueType Ty, Operand IdxTy>
5330 : NVDupLane<op19_16, 0, (outs DPR:$Vd), (ins DPR:$Vm, IdxTy:$lane),
5331 IIC_VMOVD, OpcodeStr, Dt, "$Vd, $Vm$lane",
5332 [(set DPR:$Vd, (Ty (NEONvduplane (Ty DPR:$Vm), imm:$lane)))]>;
5334 class VDUPLNQ<bits<4> op19_16, string OpcodeStr, string Dt,
5335 ValueType ResTy, ValueType OpTy, Operand IdxTy>
5336 : NVDupLane<op19_16, 1, (outs QPR:$Vd), (ins DPR:$Vm, IdxTy:$lane),
5337 IIC_VMOVQ, OpcodeStr, Dt, "$Vd, $Vm$lane",
5338 [(set QPR:$Vd, (ResTy (NEONvduplane (OpTy DPR:$Vm),
5339 VectorIndex32:$lane)))]>;
5341 // Inst{19-16} is partially specified depending on the element size.
5343 def VDUPLN8d : VDUPLND<{?,?,?,1}, "vdup", "8", v8i8, VectorIndex8> {
5345 let Inst{19-17} = lane{2-0};
5347 def VDUPLN16d : VDUPLND<{?,?,1,0}, "vdup", "16", v4i16, VectorIndex16> {
5349 let Inst{19-18} = lane{1-0};
5351 def VDUPLN32d : VDUPLND<{?,1,0,0}, "vdup", "32", v2i32, VectorIndex32> {
5353 let Inst{19} = lane{0};
5355 def VDUPLN8q : VDUPLNQ<{?,?,?,1}, "vdup", "8", v16i8, v8i8, VectorIndex8> {
5357 let Inst{19-17} = lane{2-0};
5359 def VDUPLN16q : VDUPLNQ<{?,?,1,0}, "vdup", "16", v8i16, v4i16, VectorIndex16> {
5361 let Inst{19-18} = lane{1-0};
5363 def VDUPLN32q : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4i32, v2i32, VectorIndex32> {
5365 let Inst{19} = lane{0};
5368 def : Pat<(v2f32 (NEONvduplane (v2f32 DPR:$Vm), imm:$lane)),
5369 (VDUPLN32d DPR:$Vm, imm:$lane)>;
5371 def : Pat<(v4f32 (NEONvduplane (v2f32 DPR:$Vm), imm:$lane)),
5372 (VDUPLN32q DPR:$Vm, imm:$lane)>;
5374 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
5375 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
5376 (DSubReg_i8_reg imm:$lane))),
5377 (SubReg_i8_lane imm:$lane)))>;
5378 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
5379 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
5380 (DSubReg_i16_reg imm:$lane))),
5381 (SubReg_i16_lane imm:$lane)))>;
5382 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
5383 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
5384 (DSubReg_i32_reg imm:$lane))),
5385 (SubReg_i32_lane imm:$lane)))>;
5386 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
5387 (v4f32 (VDUPLN32q (v2f32 (EXTRACT_SUBREG QPR:$src,
5388 (DSubReg_i32_reg imm:$lane))),
5389 (SubReg_i32_lane imm:$lane)))>;
5391 def VDUPfdf : PseudoNeonI<(outs DPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
5392 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
5393 def VDUPfqf : PseudoNeonI<(outs QPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
5394 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
5396 // VMOVN : Vector Narrowing Move
5397 defm VMOVN : N2VN_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVN,
5398 "vmovn", "i", trunc>;
5399 // VQMOVN : Vector Saturating Narrowing Move
5400 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
5401 "vqmovn", "s", int_arm_neon_vqmovns>;
5402 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
5403 "vqmovn", "u", int_arm_neon_vqmovnu>;
5404 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
5405 "vqmovun", "s", int_arm_neon_vqmovnsu>;
5406 // VMOVL : Vector Lengthening Move
5407 defm VMOVLs : N2VL_QHS<0b01,0b10100,0,1, "vmovl", "s", sext>;
5408 defm VMOVLu : N2VL_QHS<0b11,0b10100,0,1, "vmovl", "u", zext>;
5409 def : Pat<(v8i16 (anyext (v8i8 DPR:$Vm))), (VMOVLuv8i16 DPR:$Vm)>;
5410 def : Pat<(v4i32 (anyext (v4i16 DPR:$Vm))), (VMOVLuv4i32 DPR:$Vm)>;
5411 def : Pat<(v2i64 (anyext (v2i32 DPR:$Vm))), (VMOVLuv2i64 DPR:$Vm)>;
5413 // Vector Conversions.
5415 // VCVT : Vector Convert Between Floating-Point and Integers
5416 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
5417 v2i32, v2f32, fp_to_sint>;
5418 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
5419 v2i32, v2f32, fp_to_uint>;
5420 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
5421 v2f32, v2i32, sint_to_fp>;
5422 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
5423 v2f32, v2i32, uint_to_fp>;
5425 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
5426 v4i32, v4f32, fp_to_sint>;
5427 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
5428 v4i32, v4f32, fp_to_uint>;
5429 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
5430 v4f32, v4i32, sint_to_fp>;
5431 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
5432 v4f32, v4i32, uint_to_fp>;
5434 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
5435 let DecoderMethod = "DecodeVCVTD" in {
5436 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
5437 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
5438 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
5439 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
5440 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
5441 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
5442 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
5443 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
5446 let DecoderMethod = "DecodeVCVTQ" in {
5447 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
5448 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
5449 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
5450 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
5451 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
5452 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
5453 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
5454 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
5457 // VCVT : Vector Convert Between Half-Precision and Single-Precision.
5458 def VCVTf2h : N2VNInt<0b11, 0b11, 0b01, 0b10, 0b01100, 0, 0,
5459 IIC_VUNAQ, "vcvt", "f16.f32",
5460 v4i16, v4f32, int_arm_neon_vcvtfp2hf>,
5461 Requires<[HasNEON, HasFP16]>;
5462 def VCVTh2f : N2VLInt<0b11, 0b11, 0b01, 0b10, 0b01110, 0, 0,
5463 IIC_VUNAQ, "vcvt", "f32.f16",
5464 v4f32, v4i16, int_arm_neon_vcvthf2fp>,
5465 Requires<[HasNEON, HasFP16]>;
5469 // VREV64 : Vector Reverse elements within 64-bit doublewords
5471 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5472 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$Vd),
5473 (ins DPR:$Vm), IIC_VMOVD,
5474 OpcodeStr, Dt, "$Vd, $Vm", "",
5475 [(set DPR:$Vd, (Ty (NEONvrev64 (Ty DPR:$Vm))))]>;
5476 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5477 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$Vd),
5478 (ins QPR:$Vm), IIC_VMOVQ,
5479 OpcodeStr, Dt, "$Vd, $Vm", "",
5480 [(set QPR:$Vd, (Ty (NEONvrev64 (Ty QPR:$Vm))))]>;
5482 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
5483 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
5484 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
5485 def : Pat<(v2f32 (NEONvrev64 (v2f32 DPR:$Vm))), (VREV64d32 DPR:$Vm)>;
5487 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
5488 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
5489 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
5490 def : Pat<(v4f32 (NEONvrev64 (v4f32 QPR:$Vm))), (VREV64q32 QPR:$Vm)>;
5492 // VREV32 : Vector Reverse elements within 32-bit words
5494 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5495 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$Vd),
5496 (ins DPR:$Vm), IIC_VMOVD,
5497 OpcodeStr, Dt, "$Vd, $Vm", "",
5498 [(set DPR:$Vd, (Ty (NEONvrev32 (Ty DPR:$Vm))))]>;
5499 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5500 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$Vd),
5501 (ins QPR:$Vm), IIC_VMOVQ,
5502 OpcodeStr, Dt, "$Vd, $Vm", "",
5503 [(set QPR:$Vd, (Ty (NEONvrev32 (Ty QPR:$Vm))))]>;
5505 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
5506 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
5508 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
5509 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
5511 // VREV16 : Vector Reverse elements within 16-bit halfwords
5513 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5514 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$Vd),
5515 (ins DPR:$Vm), IIC_VMOVD,
5516 OpcodeStr, Dt, "$Vd, $Vm", "",
5517 [(set DPR:$Vd, (Ty (NEONvrev16 (Ty DPR:$Vm))))]>;
5518 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5519 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$Vd),
5520 (ins QPR:$Vm), IIC_VMOVQ,
5521 OpcodeStr, Dt, "$Vd, $Vm", "",
5522 [(set QPR:$Vd, (Ty (NEONvrev16 (Ty QPR:$Vm))))]>;
5524 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
5525 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
5527 // Other Vector Shuffles.
5529 // Aligned extractions: really just dropping registers
5531 class AlignedVEXTq<ValueType DestTy, ValueType SrcTy, SDNodeXForm LaneCVT>
5532 : Pat<(DestTy (vector_extract_subvec (SrcTy QPR:$src), (i32 imm:$start))),
5533 (EXTRACT_SUBREG (SrcTy QPR:$src), (LaneCVT imm:$start))>;
5535 def : AlignedVEXTq<v8i8, v16i8, DSubReg_i8_reg>;
5537 def : AlignedVEXTq<v4i16, v8i16, DSubReg_i16_reg>;
5539 def : AlignedVEXTq<v2i32, v4i32, DSubReg_i32_reg>;
5541 def : AlignedVEXTq<v1i64, v2i64, DSubReg_f64_reg>;
5543 def : AlignedVEXTq<v2f32, v4f32, DSubReg_i32_reg>;
5546 // VEXT : Vector Extract
5549 // All of these have a two-operand InstAlias.
5550 let TwoOperandAliasConstraint = "$Vn = $Vd" in {
5551 class VEXTd<string OpcodeStr, string Dt, ValueType Ty, Operand immTy>
5552 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$Vd),
5553 (ins DPR:$Vn, DPR:$Vm, immTy:$index), NVExtFrm,
5554 IIC_VEXTD, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
5555 [(set DPR:$Vd, (Ty (NEONvext (Ty DPR:$Vn),
5556 (Ty DPR:$Vm), imm:$index)))]> {
5559 let Inst{10-8} = index{2-0};
5562 class VEXTq<string OpcodeStr, string Dt, ValueType Ty, Operand immTy>
5563 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$Vd),
5564 (ins QPR:$Vn, QPR:$Vm, imm0_15:$index), NVExtFrm,
5565 IIC_VEXTQ, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
5566 [(set QPR:$Vd, (Ty (NEONvext (Ty QPR:$Vn),
5567 (Ty QPR:$Vm), imm:$index)))]> {
5569 let Inst{11-8} = index{3-0};
5573 def VEXTd8 : VEXTd<"vext", "8", v8i8, imm0_7> {
5574 let Inst{10-8} = index{2-0};
5576 def VEXTd16 : VEXTd<"vext", "16", v4i16, imm0_3> {
5577 let Inst{10-9} = index{1-0};
5580 def VEXTd32 : VEXTd<"vext", "32", v2i32, imm0_1> {
5581 let Inst{10} = index{0};
5582 let Inst{9-8} = 0b00;
5584 def : Pat<(v2f32 (NEONvext (v2f32 DPR:$Vn),
5587 (VEXTd32 DPR:$Vn, DPR:$Vm, imm:$index)>;
5589 def VEXTq8 : VEXTq<"vext", "8", v16i8, imm0_15> {
5590 let Inst{11-8} = index{3-0};
5592 def VEXTq16 : VEXTq<"vext", "16", v8i16, imm0_7> {
5593 let Inst{11-9} = index{2-0};
5596 def VEXTq32 : VEXTq<"vext", "32", v4i32, imm0_3> {
5597 let Inst{11-10} = index{1-0};
5598 let Inst{9-8} = 0b00;
5600 def VEXTq64 : VEXTq<"vext", "64", v2i64, imm0_1> {
5601 let Inst{11} = index{0};
5602 let Inst{10-8} = 0b000;
5604 def : Pat<(v4f32 (NEONvext (v4f32 QPR:$Vn),
5607 (VEXTq32 QPR:$Vn, QPR:$Vm, imm:$index)>;
5609 // VTRN : Vector Transpose
5611 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
5612 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
5613 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
5615 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
5616 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
5617 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
5619 // VUZP : Vector Unzip (Deinterleave)
5621 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
5622 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
5623 // vuzp.32 Dd, Dm is a pseudo-instruction expanded to vtrn.32 Dd, Dm.
5624 def : NEONInstAlias<"vuzp${p}.32 $Dd, $Dm",
5625 (VTRNd32 DPR:$Dd, DPR:$Dm, pred:$p)>;
5627 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
5628 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
5629 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
5631 // VZIP : Vector Zip (Interleave)
5633 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
5634 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
5635 // vzip.32 Dd, Dm is a pseudo-instruction expanded to vtrn.32 Dd, Dm.
5636 def : NEONInstAlias<"vzip${p}.32 $Dd, $Dm",
5637 (VTRNd32 DPR:$Dd, DPR:$Dm, pred:$p)>;
5639 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
5640 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
5641 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
5643 // Vector Table Lookup and Table Extension.
5645 // VTBL : Vector Table Lookup
5646 let DecoderMethod = "DecodeTBLInstruction" in {
5648 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$Vd),
5649 (ins VecListOneD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB1,
5650 "vtbl", "8", "$Vd, $Vn, $Vm", "",
5651 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbl1 VecListOneD:$Vn, DPR:$Vm)))]>;
5652 let hasExtraSrcRegAllocReq = 1 in {
5654 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$Vd),
5655 (ins VecListDPair:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB2,
5656 "vtbl", "8", "$Vd, $Vn, $Vm", "", []>;
5658 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$Vd),
5659 (ins VecListThreeD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB3,
5660 "vtbl", "8", "$Vd, $Vn, $Vm", "", []>;
5662 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$Vd),
5663 (ins VecListFourD:$Vn, DPR:$Vm),
5665 "vtbl", "8", "$Vd, $Vn, $Vm", "", []>;
5666 } // hasExtraSrcRegAllocReq = 1
5669 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB3, "", []>;
5671 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB4, "", []>;
5673 // VTBX : Vector Table Extension
5675 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$Vd),
5676 (ins DPR:$orig, VecListOneD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX1,
5677 "vtbx", "8", "$Vd, $Vn, $Vm", "$orig = $Vd",
5678 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbx1
5679 DPR:$orig, VecListOneD:$Vn, DPR:$Vm)))]>;
5680 let hasExtraSrcRegAllocReq = 1 in {
5682 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$Vd),
5683 (ins DPR:$orig, VecListDPair:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX2,
5684 "vtbx", "8", "$Vd, $Vn, $Vm", "$orig = $Vd", []>;
5686 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$Vd),
5687 (ins DPR:$orig, VecListThreeD:$Vn, DPR:$Vm),
5688 NVTBLFrm, IIC_VTBX3,
5689 "vtbx", "8", "$Vd, $Vn, $Vm",
5692 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$Vd),
5693 (ins DPR:$orig, VecListFourD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX4,
5694 "vtbx", "8", "$Vd, $Vn, $Vm",
5696 } // hasExtraSrcRegAllocReq = 1
5699 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
5700 IIC_VTBX3, "$orig = $dst", []>;
5702 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
5703 IIC_VTBX4, "$orig = $dst", []>;
5704 } // DecoderMethod = "DecodeTBLInstruction"
5706 //===----------------------------------------------------------------------===//
5707 // NEON instructions for single-precision FP math
5708 //===----------------------------------------------------------------------===//
5710 class N2VSPat<SDNode OpNode, NeonI Inst>
5711 : NEONFPPat<(f32 (OpNode SPR:$a)),
5713 (v2f32 (COPY_TO_REGCLASS (Inst
5715 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5716 SPR:$a, ssub_0)), DPR_VFP2)), ssub_0)>;
5718 class N3VSPat<SDNode OpNode, NeonI Inst>
5719 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
5721 (v2f32 (COPY_TO_REGCLASS (Inst
5723 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5726 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5727 SPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>;
5729 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
5730 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
5732 (v2f32 (COPY_TO_REGCLASS (Inst
5734 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5737 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5740 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5741 SPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>;
5743 def : N3VSPat<fadd, VADDfd>;
5744 def : N3VSPat<fsub, VSUBfd>;
5745 def : N3VSPat<fmul, VMULfd>;
5746 def : N3VSMulOpPat<fmul, fadd, VMLAfd>,
5747 Requires<[HasNEON, UseNEONForFP, UseFPVMLx, DontUseFusedMAC]>;
5748 def : N3VSMulOpPat<fmul, fsub, VMLSfd>,
5749 Requires<[HasNEON, UseNEONForFP, UseFPVMLx, DontUseFusedMAC]>;
5750 def : N3VSMulOpPat<fmul, fadd, VFMAfd>,
5751 Requires<[HasVFP4, UseNEONForFP, UseFusedMAC]>;
5752 def : N3VSMulOpPat<fmul, fsub, VFMSfd>,
5753 Requires<[HasVFP4, UseNEONForFP, UseFusedMAC]>;
5754 def : N2VSPat<fabs, VABSfd>;
5755 def : N2VSPat<fneg, VNEGfd>;
5756 def : N3VSPat<NEONfmax, VMAXfd>;
5757 def : N3VSPat<NEONfmin, VMINfd>;
5758 def : N2VSPat<arm_ftosi, VCVTf2sd>;
5759 def : N2VSPat<arm_ftoui, VCVTf2ud>;
5760 def : N2VSPat<arm_sitof, VCVTs2fd>;
5761 def : N2VSPat<arm_uitof, VCVTu2fd>;
5763 // Prefer VMOVDRR for i32 -> f32 bitcasts, it can write all DPR registers.
5764 def : Pat<(f32 (bitconvert GPR:$a)),
5765 (EXTRACT_SUBREG (VMOVDRR GPR:$a, GPR:$a), ssub_0)>,
5766 Requires<[HasNEON, DontUseVMOVSR]>;
5768 //===----------------------------------------------------------------------===//
5769 // Non-Instruction Patterns
5770 //===----------------------------------------------------------------------===//
5773 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
5774 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
5775 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
5776 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
5777 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
5778 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
5779 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
5780 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
5781 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
5782 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
5783 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
5784 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
5785 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
5786 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
5787 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
5788 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
5789 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
5790 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
5791 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
5792 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
5793 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
5794 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
5795 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
5796 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
5797 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
5798 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
5799 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
5800 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
5801 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
5802 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
5804 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
5805 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
5806 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
5807 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
5808 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
5809 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
5810 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
5811 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
5812 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
5813 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
5814 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
5815 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
5816 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
5817 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
5818 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
5819 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
5820 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
5821 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
5822 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
5823 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
5824 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
5825 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
5826 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
5827 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
5828 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
5829 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
5830 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
5831 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
5832 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
5833 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;
5835 // Fold extracting an element out of a v2i32 into a vfp register.
5836 def : Pat<(f32 (bitconvert (i32 (extractelt (v2i32 DPR:$src), imm:$lane)))),
5837 (f32 (EXTRACT_SUBREG DPR:$src, (SSubReg_f32_reg imm:$lane)))>;
5839 // Vector lengthening move with load, matching extending loads.
5841 // extload, zextload and sextload for a standard lengthening load. Example:
5842 // Lengthen_Single<"8", "i16", "8"> =
5843 // Pat<(v8i16 (extloadvi8 addrmode6:$addr))
5844 // (VMOVLuv8i16 (VLD1d8 addrmode6:$addr,
5845 // (f64 (IMPLICIT_DEF)), (i32 0)))>;
5846 multiclass Lengthen_Single<string DestLanes, string DestTy, string SrcTy> {
5847 let AddedComplexity = 10 in {
5848 def _Any : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5849 (!cast<PatFrag>("extloadvi" # SrcTy) addrmode6:$addr)),
5850 (!cast<Instruction>("VMOVLuv" # DestLanes # DestTy)
5851 (!cast<Instruction>("VLD1d" # SrcTy) addrmode6:$addr))>;
5853 def _Z : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5854 (!cast<PatFrag>("zextloadvi" # SrcTy) addrmode6:$addr)),
5855 (!cast<Instruction>("VMOVLuv" # DestLanes # DestTy)
5856 (!cast<Instruction>("VLD1d" # SrcTy) addrmode6:$addr))>;
5858 def _S : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5859 (!cast<PatFrag>("sextloadvi" # SrcTy) addrmode6:$addr)),
5860 (!cast<Instruction>("VMOVLsv" # DestLanes # DestTy)
5861 (!cast<Instruction>("VLD1d" # SrcTy) addrmode6:$addr))>;
5865 // extload, zextload and sextload for a lengthening load which only uses
5866 // half the lanes available. Example:
5867 // Lengthen_HalfSingle<"4", "i16", "8", "i16", "i8"> =
5868 // Pat<(v4i16 (extloadvi8 addrmode6oneL32:$addr)),
5869 // (EXTRACT_SUBREG (VMOVLuv8i16 (VLD1LNd32 addrmode6oneL32:$addr,
5870 // (f64 (IMPLICIT_DEF)), (i32 0))),
5872 multiclass Lengthen_HalfSingle<string DestLanes, string DestTy, string SrcTy,
5873 string InsnLanes, string InsnTy> {
5874 def _Any : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5875 (!cast<PatFrag>("extloadv" # SrcTy) addrmode6oneL32:$addr)),
5876 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # InsnLanes # InsnTy)
5877 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
5879 def _Z : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5880 (!cast<PatFrag>("zextloadv" # SrcTy) addrmode6oneL32:$addr)),
5881 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # InsnLanes # InsnTy)
5882 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
5884 def _S : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5885 (!cast<PatFrag>("sextloadv" # SrcTy) addrmode6oneL32:$addr)),
5886 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLsv" # InsnLanes # InsnTy)
5887 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
5891 // extload, zextload and sextload for a lengthening load followed by another
5892 // lengthening load, to quadruple the initial length.
5894 // Lengthen_Double<"4", "i32", "i8", "8", "i16", "4", "i32"> =
5895 // Pat<(v4i32 (extloadvi8 addrmode6oneL32:$addr))
5896 // (EXTRACT_SUBREG (VMOVLuv4i32
5897 // (EXTRACT_SUBREG (VMOVLuv8i16 (VLD1LNd32 addrmode6oneL32:$addr,
5898 // (f64 (IMPLICIT_DEF)),
5902 multiclass Lengthen_Double<string DestLanes, string DestTy, string SrcTy,
5903 string Insn1Lanes, string Insn1Ty, string Insn2Lanes,
5905 def _Any : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5906 (!cast<PatFrag>("extloadv" # SrcTy) addrmode6oneL32:$addr)),
5907 (!cast<Instruction>("VMOVLuv" # Insn2Lanes # Insn2Ty)
5908 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn1Lanes # Insn1Ty)
5909 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
5911 def _Z : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5912 (!cast<PatFrag>("zextloadv" # SrcTy) addrmode6oneL32:$addr)),
5913 (!cast<Instruction>("VMOVLuv" # Insn2Lanes # Insn2Ty)
5914 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn1Lanes # Insn1Ty)
5915 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
5917 def _S : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5918 (!cast<PatFrag>("sextloadv" # SrcTy) addrmode6oneL32:$addr)),
5919 (!cast<Instruction>("VMOVLsv" # Insn2Lanes # Insn2Ty)
5920 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLsv" # Insn1Lanes # Insn1Ty)
5921 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
5925 // extload, zextload and sextload for a lengthening load followed by another
5926 // lengthening load, to quadruple the initial length, but which ends up only
5927 // requiring half the available lanes (a 64-bit outcome instead of a 128-bit).
5929 // Lengthen_HalfDouble<"2", "i32", "i8", "8", "i16", "4", "i32"> =
5930 // Pat<(v2i32 (extloadvi8 addrmode6:$addr))
5931 // (EXTRACT_SUBREG (VMOVLuv4i32
5932 // (EXTRACT_SUBREG (VMOVLuv8i16 (VLD1LNd16 addrmode6:$addr,
5933 // (f64 (IMPLICIT_DEF)), (i32 0))),
5936 multiclass Lengthen_HalfDouble<string DestLanes, string DestTy, string SrcTy,
5937 string Insn1Lanes, string Insn1Ty, string Insn2Lanes,
5939 def _Any : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5940 (!cast<PatFrag>("extloadv" # SrcTy) addrmode6:$addr)),
5941 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn2Lanes # Insn2Ty)
5942 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn1Lanes # Insn1Ty)
5943 (VLD1LNd16 addrmode6:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
5946 def _Z : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5947 (!cast<PatFrag>("zextloadv" # SrcTy) addrmode6:$addr)),
5948 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn2Lanes # Insn2Ty)
5949 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn1Lanes # Insn1Ty)
5950 (VLD1LNd16 addrmode6:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
5953 def _S : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5954 (!cast<PatFrag>("sextloadv" # SrcTy) addrmode6:$addr)),
5955 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLsv" # Insn2Lanes # Insn2Ty)
5956 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLsv" # Insn1Lanes # Insn1Ty)
5957 (VLD1LNd16 addrmode6:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
5962 defm : Lengthen_Single<"8", "i16", "8">; // v8i8 -> v8i16
5963 defm : Lengthen_Single<"4", "i32", "16">; // v4i16 -> v4i32
5964 defm : Lengthen_Single<"2", "i64", "32">; // v2i32 -> v2i64
5966 defm : Lengthen_HalfSingle<"4", "i16", "i8", "8", "i16">; // v4i8 -> v4i16
5967 defm : Lengthen_HalfSingle<"2", "i32", "i16", "4", "i32">; // v2i16 -> v2i32
5969 // Double lengthening - v4i8 -> v4i16 -> v4i32
5970 defm : Lengthen_Double<"4", "i32", "i8", "8", "i16", "4", "i32">;
5971 // v2i8 -> v2i16 -> v2i32
5972 defm : Lengthen_HalfDouble<"2", "i32", "i8", "8", "i16", "4", "i32">;
5973 // v2i16 -> v2i32 -> v2i64
5974 defm : Lengthen_Double<"2", "i64", "i16", "4", "i32", "2", "i64">;
5976 // Triple lengthening - v2i8 -> v2i16 -> v2i32 -> v2i64
5977 def : Pat<(v2i64 (extloadvi8 addrmode6:$addr)),
5978 (VMOVLuv2i64 (EXTRACT_SUBREG (VMOVLuv4i32 (EXTRACT_SUBREG (VMOVLuv8i16
5979 (VLD1LNd16 addrmode6:$addr,
5980 (f64 (IMPLICIT_DEF)), (i32 0))), dsub_0)), dsub_0))>;
5981 def : Pat<(v2i64 (zextloadvi8 addrmode6:$addr)),
5982 (VMOVLuv2i64 (EXTRACT_SUBREG (VMOVLuv4i32 (EXTRACT_SUBREG (VMOVLuv8i16
5983 (VLD1LNd16 addrmode6:$addr,
5984 (f64 (IMPLICIT_DEF)), (i32 0))), dsub_0)), dsub_0))>;
5985 def : Pat<(v2i64 (sextloadvi8 addrmode6:$addr)),
5986 (VMOVLsv2i64 (EXTRACT_SUBREG (VMOVLsv4i32 (EXTRACT_SUBREG (VMOVLsv8i16
5987 (VLD1LNd16 addrmode6:$addr,
5988 (f64 (IMPLICIT_DEF)), (i32 0))), dsub_0)), dsub_0))>;
5990 //===----------------------------------------------------------------------===//
5991 // Assembler aliases
5994 def : VFP2InstAlias<"fmdhr${p} $Dd, $Rn",
5995 (VSETLNi32 DPR:$Dd, GPR:$Rn, 1, pred:$p)>;
5996 def : VFP2InstAlias<"fmdlr${p} $Dd, $Rn",
5997 (VSETLNi32 DPR:$Dd, GPR:$Rn, 0, pred:$p)>;
5999 // VAND/VBIC/VEOR/VORR accept but do not require a type suffix.
6000 defm : NEONDTAnyInstAlias<"vand${p}", "$Vd, $Vn, $Vm",
6001 (VANDd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
6002 defm : NEONDTAnyInstAlias<"vand${p}", "$Vd, $Vn, $Vm",
6003 (VANDq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
6004 defm : NEONDTAnyInstAlias<"vbic${p}", "$Vd, $Vn, $Vm",
6005 (VBICd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
6006 defm : NEONDTAnyInstAlias<"vbic${p}", "$Vd, $Vn, $Vm",
6007 (VBICq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
6008 defm : NEONDTAnyInstAlias<"veor${p}", "$Vd, $Vn, $Vm",
6009 (VEORd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
6010 defm : NEONDTAnyInstAlias<"veor${p}", "$Vd, $Vn, $Vm",
6011 (VEORq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
6012 defm : NEONDTAnyInstAlias<"vorr${p}", "$Vd, $Vn, $Vm",
6013 (VORRd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
6014 defm : NEONDTAnyInstAlias<"vorr${p}", "$Vd, $Vn, $Vm",
6015 (VORRq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
6016 // ... two-operand aliases
6017 defm : NEONDTAnyInstAlias<"vand${p}", "$Vdn, $Vm",
6018 (VANDd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
6019 defm : NEONDTAnyInstAlias<"vand${p}", "$Vdn, $Vm",
6020 (VANDq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
6021 defm : NEONDTAnyInstAlias<"veor${p}", "$Vdn, $Vm",
6022 (VEORd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
6023 defm : NEONDTAnyInstAlias<"veor${p}", "$Vdn, $Vm",
6024 (VEORq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
6025 defm : NEONDTAnyInstAlias<"vorr${p}", "$Vdn, $Vm",
6026 (VORRd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
6027 defm : NEONDTAnyInstAlias<"vorr${p}", "$Vdn, $Vm",
6028 (VORRq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
6030 // VLD1 single-lane pseudo-instructions. These need special handling for
6031 // the lane index that an InstAlias can't handle, so we use these instead.
6032 def VLD1LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vld1${p}", ".8", "$list, $addr",
6033 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6034 def VLD1LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vld1${p}", ".16", "$list, $addr",
6035 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6036 def VLD1LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vld1${p}", ".32", "$list, $addr",
6037 (ins VecListOneDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6039 def VLD1LNdWB_fixed_Asm_8 :
6040 NEONDataTypeAsmPseudoInst<"vld1${p}", ".8", "$list, $addr!",
6041 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6042 def VLD1LNdWB_fixed_Asm_16 :
6043 NEONDataTypeAsmPseudoInst<"vld1${p}", ".16", "$list, $addr!",
6044 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6045 def VLD1LNdWB_fixed_Asm_32 :
6046 NEONDataTypeAsmPseudoInst<"vld1${p}", ".32", "$list, $addr!",
6047 (ins VecListOneDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6048 def VLD1LNdWB_register_Asm_8 :
6049 NEONDataTypeAsmPseudoInst<"vld1${p}", ".8", "$list, $addr, $Rm",
6050 (ins VecListOneDByteIndexed:$list, addrmode6:$addr,
6051 rGPR:$Rm, pred:$p)>;
6052 def VLD1LNdWB_register_Asm_16 :
6053 NEONDataTypeAsmPseudoInst<"vld1${p}", ".16", "$list, $addr, $Rm",
6054 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr,
6055 rGPR:$Rm, pred:$p)>;
6056 def VLD1LNdWB_register_Asm_32 :
6057 NEONDataTypeAsmPseudoInst<"vld1${p}", ".32", "$list, $addr, $Rm",
6058 (ins VecListOneDWordIndexed:$list, addrmode6:$addr,
6059 rGPR:$Rm, pred:$p)>;
6062 // VST1 single-lane pseudo-instructions. These need special handling for
6063 // the lane index that an InstAlias can't handle, so we use these instead.
6064 def VST1LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vst1${p}", ".8", "$list, $addr",
6065 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6066 def VST1LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vst1${p}", ".16", "$list, $addr",
6067 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6068 def VST1LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vst1${p}", ".32", "$list, $addr",
6069 (ins VecListOneDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6071 def VST1LNdWB_fixed_Asm_8 :
6072 NEONDataTypeAsmPseudoInst<"vst1${p}", ".8", "$list, $addr!",
6073 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6074 def VST1LNdWB_fixed_Asm_16 :
6075 NEONDataTypeAsmPseudoInst<"vst1${p}", ".16", "$list, $addr!",
6076 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6077 def VST1LNdWB_fixed_Asm_32 :
6078 NEONDataTypeAsmPseudoInst<"vst1${p}", ".32", "$list, $addr!",
6079 (ins VecListOneDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6080 def VST1LNdWB_register_Asm_8 :
6081 NEONDataTypeAsmPseudoInst<"vst1${p}", ".8", "$list, $addr, $Rm",
6082 (ins VecListOneDByteIndexed:$list, addrmode6:$addr,
6083 rGPR:$Rm, pred:$p)>;
6084 def VST1LNdWB_register_Asm_16 :
6085 NEONDataTypeAsmPseudoInst<"vst1${p}", ".16", "$list, $addr, $Rm",
6086 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr,
6087 rGPR:$Rm, pred:$p)>;
6088 def VST1LNdWB_register_Asm_32 :
6089 NEONDataTypeAsmPseudoInst<"vst1${p}", ".32", "$list, $addr, $Rm",
6090 (ins VecListOneDWordIndexed:$list, addrmode6:$addr,
6091 rGPR:$Rm, pred:$p)>;
6093 // VLD2 single-lane pseudo-instructions. These need special handling for
6094 // the lane index that an InstAlias can't handle, so we use these instead.
6095 def VLD2LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vld2${p}", ".8", "$list, $addr",
6096 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6097 def VLD2LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr",
6098 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6099 def VLD2LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr",
6100 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6101 def VLD2LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr",
6102 (ins VecListTwoQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6103 def VLD2LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr",
6104 (ins VecListTwoQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6106 def VLD2LNdWB_fixed_Asm_8 :
6107 NEONDataTypeAsmPseudoInst<"vld2${p}", ".8", "$list, $addr!",
6108 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6109 def VLD2LNdWB_fixed_Asm_16 :
6110 NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr!",
6111 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6112 def VLD2LNdWB_fixed_Asm_32 :
6113 NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr!",
6114 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6115 def VLD2LNqWB_fixed_Asm_16 :
6116 NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr!",
6117 (ins VecListTwoQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6118 def VLD2LNqWB_fixed_Asm_32 :
6119 NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr!",
6120 (ins VecListTwoQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6121 def VLD2LNdWB_register_Asm_8 :
6122 NEONDataTypeAsmPseudoInst<"vld2${p}", ".8", "$list, $addr, $Rm",
6123 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr,
6124 rGPR:$Rm, pred:$p)>;
6125 def VLD2LNdWB_register_Asm_16 :
6126 NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr, $Rm",
6127 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr,
6128 rGPR:$Rm, pred:$p)>;
6129 def VLD2LNdWB_register_Asm_32 :
6130 NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr, $Rm",
6131 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr,
6132 rGPR:$Rm, pred:$p)>;
6133 def VLD2LNqWB_register_Asm_16 :
6134 NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr, $Rm",
6135 (ins VecListTwoQHWordIndexed:$list, addrmode6:$addr,
6136 rGPR:$Rm, pred:$p)>;
6137 def VLD2LNqWB_register_Asm_32 :
6138 NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr, $Rm",
6139 (ins VecListTwoQWordIndexed:$list, addrmode6:$addr,
6140 rGPR:$Rm, pred:$p)>;
6143 // VST2 single-lane pseudo-instructions. These need special handling for
6144 // the lane index that an InstAlias can't handle, so we use these instead.
6145 def VST2LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vst2${p}", ".8", "$list, $addr",
6146 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6147 def VST2LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vst2${p}", ".16", "$list, $addr",
6148 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6149 def VST2LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr",
6150 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6151 def VST2LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vst2${p}", ".16", "$list, $addr",
6152 (ins VecListTwoQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6153 def VST2LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr",
6154 (ins VecListTwoQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6156 def VST2LNdWB_fixed_Asm_8 :
6157 NEONDataTypeAsmPseudoInst<"vst2${p}", ".8", "$list, $addr!",
6158 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6159 def VST2LNdWB_fixed_Asm_16 :
6160 NEONDataTypeAsmPseudoInst<"vst2${p}", ".16", "$list, $addr!",
6161 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6162 def VST2LNdWB_fixed_Asm_32 :
6163 NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr!",
6164 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6165 def VST2LNqWB_fixed_Asm_16 :
6166 NEONDataTypeAsmPseudoInst<"vst2${p}", ".16", "$list, $addr!",
6167 (ins VecListTwoQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6168 def VST2LNqWB_fixed_Asm_32 :
6169 NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr!",
6170 (ins VecListTwoQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6171 def VST2LNdWB_register_Asm_8 :
6172 NEONDataTypeAsmPseudoInst<"vst2${p}", ".8", "$list, $addr, $Rm",
6173 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr,
6174 rGPR:$Rm, pred:$p)>;
6175 def VST2LNdWB_register_Asm_16 :
6176 NEONDataTypeAsmPseudoInst<"vst2${p}", ".16","$list, $addr, $Rm",
6177 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr,
6178 rGPR:$Rm, pred:$p)>;
6179 def VST2LNdWB_register_Asm_32 :
6180 NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr, $Rm",
6181 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr,
6182 rGPR:$Rm, pred:$p)>;
6183 def VST2LNqWB_register_Asm_16 :
6184 NEONDataTypeAsmPseudoInst<"vst2${p}", ".16","$list, $addr, $Rm",
6185 (ins VecListTwoQHWordIndexed:$list, addrmode6:$addr,
6186 rGPR:$Rm, pred:$p)>;
6187 def VST2LNqWB_register_Asm_32 :
6188 NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr, $Rm",
6189 (ins VecListTwoQWordIndexed:$list, addrmode6:$addr,
6190 rGPR:$Rm, pred:$p)>;
6192 // VLD3 all-lanes pseudo-instructions. These need special handling for
6193 // the lane index that an InstAlias can't handle, so we use these instead.
6194 def VLD3DUPdAsm_8 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr",
6195 (ins VecListThreeDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6196 def VLD3DUPdAsm_16: NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
6197 (ins VecListThreeDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6198 def VLD3DUPdAsm_32: NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
6199 (ins VecListThreeDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6200 def VLD3DUPqAsm_8 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr",
6201 (ins VecListThreeQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6202 def VLD3DUPqAsm_16: NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
6203 (ins VecListThreeQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6204 def VLD3DUPqAsm_32: NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
6205 (ins VecListThreeQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6207 def VLD3DUPdWB_fixed_Asm_8 :
6208 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr!",
6209 (ins VecListThreeDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6210 def VLD3DUPdWB_fixed_Asm_16 :
6211 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
6212 (ins VecListThreeDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6213 def VLD3DUPdWB_fixed_Asm_32 :
6214 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
6215 (ins VecListThreeDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6216 def VLD3DUPqWB_fixed_Asm_8 :
6217 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr!",
6218 (ins VecListThreeQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6219 def VLD3DUPqWB_fixed_Asm_16 :
6220 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
6221 (ins VecListThreeQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6222 def VLD3DUPqWB_fixed_Asm_32 :
6223 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
6224 (ins VecListThreeQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6225 def VLD3DUPdWB_register_Asm_8 :
6226 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr, $Rm",
6227 (ins VecListThreeDAllLanes:$list, addrmode6:$addr,
6228 rGPR:$Rm, pred:$p)>;
6229 def VLD3DUPdWB_register_Asm_16 :
6230 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
6231 (ins VecListThreeDAllLanes:$list, addrmode6:$addr,
6232 rGPR:$Rm, pred:$p)>;
6233 def VLD3DUPdWB_register_Asm_32 :
6234 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
6235 (ins VecListThreeDAllLanes:$list, addrmode6:$addr,
6236 rGPR:$Rm, pred:$p)>;
6237 def VLD3DUPqWB_register_Asm_8 :
6238 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr, $Rm",
6239 (ins VecListThreeQAllLanes:$list, addrmode6:$addr,
6240 rGPR:$Rm, pred:$p)>;
6241 def VLD3DUPqWB_register_Asm_16 :
6242 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
6243 (ins VecListThreeQAllLanes:$list, addrmode6:$addr,
6244 rGPR:$Rm, pred:$p)>;
6245 def VLD3DUPqWB_register_Asm_32 :
6246 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
6247 (ins VecListThreeQAllLanes:$list, addrmode6:$addr,
6248 rGPR:$Rm, pred:$p)>;
6251 // VLD3 single-lane pseudo-instructions. These need special handling for
6252 // the lane index that an InstAlias can't handle, so we use these instead.
6253 def VLD3LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr",
6254 (ins VecListThreeDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6255 def VLD3LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
6256 (ins VecListThreeDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6257 def VLD3LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
6258 (ins VecListThreeDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6259 def VLD3LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
6260 (ins VecListThreeQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6261 def VLD3LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
6262 (ins VecListThreeQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6264 def VLD3LNdWB_fixed_Asm_8 :
6265 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr!",
6266 (ins VecListThreeDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6267 def VLD3LNdWB_fixed_Asm_16 :
6268 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
6269 (ins VecListThreeDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6270 def VLD3LNdWB_fixed_Asm_32 :
6271 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
6272 (ins VecListThreeDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6273 def VLD3LNqWB_fixed_Asm_16 :
6274 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
6275 (ins VecListThreeQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6276 def VLD3LNqWB_fixed_Asm_32 :
6277 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
6278 (ins VecListThreeQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6279 def VLD3LNdWB_register_Asm_8 :
6280 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr, $Rm",
6281 (ins VecListThreeDByteIndexed:$list, addrmode6:$addr,
6282 rGPR:$Rm, pred:$p)>;
6283 def VLD3LNdWB_register_Asm_16 :
6284 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
6285 (ins VecListThreeDHWordIndexed:$list, addrmode6:$addr,
6286 rGPR:$Rm, pred:$p)>;
6287 def VLD3LNdWB_register_Asm_32 :
6288 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
6289 (ins VecListThreeDWordIndexed:$list, addrmode6:$addr,
6290 rGPR:$Rm, pred:$p)>;
6291 def VLD3LNqWB_register_Asm_16 :
6292 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
6293 (ins VecListThreeQHWordIndexed:$list, addrmode6:$addr,
6294 rGPR:$Rm, pred:$p)>;
6295 def VLD3LNqWB_register_Asm_32 :
6296 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
6297 (ins VecListThreeQWordIndexed:$list, addrmode6:$addr,
6298 rGPR:$Rm, pred:$p)>;
6300 // VLD3 multiple structure pseudo-instructions. These need special handling for
6301 // the vector operands that the normal instructions don't yet model.
6302 // FIXME: Remove these when the register classes and instructions are updated.
6303 def VLD3dAsm_8 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr",
6304 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6305 def VLD3dAsm_16 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
6306 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6307 def VLD3dAsm_32 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
6308 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6309 def VLD3qAsm_8 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr",
6310 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6311 def VLD3qAsm_16 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
6312 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6313 def VLD3qAsm_32 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
6314 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6316 def VLD3dWB_fixed_Asm_8 :
6317 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr!",
6318 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6319 def VLD3dWB_fixed_Asm_16 :
6320 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
6321 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6322 def VLD3dWB_fixed_Asm_32 :
6323 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
6324 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6325 def VLD3qWB_fixed_Asm_8 :
6326 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr!",
6327 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6328 def VLD3qWB_fixed_Asm_16 :
6329 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
6330 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6331 def VLD3qWB_fixed_Asm_32 :
6332 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
6333 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6334 def VLD3dWB_register_Asm_8 :
6335 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr, $Rm",
6336 (ins VecListThreeD:$list, addrmode6:$addr,
6337 rGPR:$Rm, pred:$p)>;
6338 def VLD3dWB_register_Asm_16 :
6339 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
6340 (ins VecListThreeD:$list, addrmode6:$addr,
6341 rGPR:$Rm, pred:$p)>;
6342 def VLD3dWB_register_Asm_32 :
6343 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
6344 (ins VecListThreeD:$list, addrmode6:$addr,
6345 rGPR:$Rm, pred:$p)>;
6346 def VLD3qWB_register_Asm_8 :
6347 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr, $Rm",
6348 (ins VecListThreeQ:$list, addrmode6:$addr,
6349 rGPR:$Rm, pred:$p)>;
6350 def VLD3qWB_register_Asm_16 :
6351 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
6352 (ins VecListThreeQ:$list, addrmode6:$addr,
6353 rGPR:$Rm, pred:$p)>;
6354 def VLD3qWB_register_Asm_32 :
6355 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
6356 (ins VecListThreeQ:$list, addrmode6:$addr,
6357 rGPR:$Rm, pred:$p)>;
6359 // VST3 single-lane pseudo-instructions. These need special handling for
6360 // the lane index that an InstAlias can't handle, so we use these instead.
6361 def VST3LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr",
6362 (ins VecListThreeDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6363 def VST3LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr",
6364 (ins VecListThreeDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6365 def VST3LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr",
6366 (ins VecListThreeDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6367 def VST3LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr",
6368 (ins VecListThreeQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6369 def VST3LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr",
6370 (ins VecListThreeQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6372 def VST3LNdWB_fixed_Asm_8 :
6373 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr!",
6374 (ins VecListThreeDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6375 def VST3LNdWB_fixed_Asm_16 :
6376 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr!",
6377 (ins VecListThreeDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6378 def VST3LNdWB_fixed_Asm_32 :
6379 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr!",
6380 (ins VecListThreeDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6381 def VST3LNqWB_fixed_Asm_16 :
6382 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr!",
6383 (ins VecListThreeQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6384 def VST3LNqWB_fixed_Asm_32 :
6385 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr!",
6386 (ins VecListThreeQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6387 def VST3LNdWB_register_Asm_8 :
6388 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr, $Rm",
6389 (ins VecListThreeDByteIndexed:$list, addrmode6:$addr,
6390 rGPR:$Rm, pred:$p)>;
6391 def VST3LNdWB_register_Asm_16 :
6392 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr, $Rm",
6393 (ins VecListThreeDHWordIndexed:$list, addrmode6:$addr,
6394 rGPR:$Rm, pred:$p)>;
6395 def VST3LNdWB_register_Asm_32 :
6396 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr, $Rm",
6397 (ins VecListThreeDWordIndexed:$list, addrmode6:$addr,
6398 rGPR:$Rm, pred:$p)>;
6399 def VST3LNqWB_register_Asm_16 :
6400 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr, $Rm",
6401 (ins VecListThreeQHWordIndexed:$list, addrmode6:$addr,
6402 rGPR:$Rm, pred:$p)>;
6403 def VST3LNqWB_register_Asm_32 :
6404 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr, $Rm",
6405 (ins VecListThreeQWordIndexed:$list, addrmode6:$addr,
6406 rGPR:$Rm, pred:$p)>;
6409 // VST3 multiple structure pseudo-instructions. These need special handling for
6410 // the vector operands that the normal instructions don't yet model.
6411 // FIXME: Remove these when the register classes and instructions are updated.
6412 def VST3dAsm_8 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr",
6413 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6414 def VST3dAsm_16 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr",
6415 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6416 def VST3dAsm_32 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr",
6417 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6418 def VST3qAsm_8 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr",
6419 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6420 def VST3qAsm_16 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr",
6421 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6422 def VST3qAsm_32 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr",
6423 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6425 def VST3dWB_fixed_Asm_8 :
6426 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr!",
6427 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6428 def VST3dWB_fixed_Asm_16 :
6429 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr!",
6430 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6431 def VST3dWB_fixed_Asm_32 :
6432 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr!",
6433 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6434 def VST3qWB_fixed_Asm_8 :
6435 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr!",
6436 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6437 def VST3qWB_fixed_Asm_16 :
6438 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr!",
6439 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6440 def VST3qWB_fixed_Asm_32 :
6441 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr!",
6442 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6443 def VST3dWB_register_Asm_8 :
6444 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr, $Rm",
6445 (ins VecListThreeD:$list, addrmode6:$addr,
6446 rGPR:$Rm, pred:$p)>;
6447 def VST3dWB_register_Asm_16 :
6448 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr, $Rm",
6449 (ins VecListThreeD:$list, addrmode6:$addr,
6450 rGPR:$Rm, pred:$p)>;
6451 def VST3dWB_register_Asm_32 :
6452 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr, $Rm",
6453 (ins VecListThreeD:$list, addrmode6:$addr,
6454 rGPR:$Rm, pred:$p)>;
6455 def VST3qWB_register_Asm_8 :
6456 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr, $Rm",
6457 (ins VecListThreeQ:$list, addrmode6:$addr,
6458 rGPR:$Rm, pred:$p)>;
6459 def VST3qWB_register_Asm_16 :
6460 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr, $Rm",
6461 (ins VecListThreeQ:$list, addrmode6:$addr,
6462 rGPR:$Rm, pred:$p)>;
6463 def VST3qWB_register_Asm_32 :
6464 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr, $Rm",
6465 (ins VecListThreeQ:$list, addrmode6:$addr,
6466 rGPR:$Rm, pred:$p)>;
6468 // VLD4 all-lanes pseudo-instructions. These need special handling for
6469 // the lane index that an InstAlias can't handle, so we use these instead.
6470 def VLD4DUPdAsm_8 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr",
6471 (ins VecListFourDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6472 def VLD4DUPdAsm_16: NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
6473 (ins VecListFourDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6474 def VLD4DUPdAsm_32: NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
6475 (ins VecListFourDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6476 def VLD4DUPqAsm_8 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr",
6477 (ins VecListFourQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6478 def VLD4DUPqAsm_16: NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
6479 (ins VecListFourQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6480 def VLD4DUPqAsm_32: NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
6481 (ins VecListFourQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6483 def VLD4DUPdWB_fixed_Asm_8 :
6484 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr!",
6485 (ins VecListFourDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6486 def VLD4DUPdWB_fixed_Asm_16 :
6487 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
6488 (ins VecListFourDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6489 def VLD4DUPdWB_fixed_Asm_32 :
6490 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
6491 (ins VecListFourDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6492 def VLD4DUPqWB_fixed_Asm_8 :
6493 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr!",
6494 (ins VecListFourQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6495 def VLD4DUPqWB_fixed_Asm_16 :
6496 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
6497 (ins VecListFourQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6498 def VLD4DUPqWB_fixed_Asm_32 :
6499 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
6500 (ins VecListFourQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6501 def VLD4DUPdWB_register_Asm_8 :
6502 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr, $Rm",
6503 (ins VecListFourDAllLanes:$list, addrmode6:$addr,
6504 rGPR:$Rm, pred:$p)>;
6505 def VLD4DUPdWB_register_Asm_16 :
6506 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
6507 (ins VecListFourDAllLanes:$list, addrmode6:$addr,
6508 rGPR:$Rm, pred:$p)>;
6509 def VLD4DUPdWB_register_Asm_32 :
6510 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
6511 (ins VecListFourDAllLanes:$list, addrmode6:$addr,
6512 rGPR:$Rm, pred:$p)>;
6513 def VLD4DUPqWB_register_Asm_8 :
6514 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr, $Rm",
6515 (ins VecListFourQAllLanes:$list, addrmode6:$addr,
6516 rGPR:$Rm, pred:$p)>;
6517 def VLD4DUPqWB_register_Asm_16 :
6518 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
6519 (ins VecListFourQAllLanes:$list, addrmode6:$addr,
6520 rGPR:$Rm, pred:$p)>;
6521 def VLD4DUPqWB_register_Asm_32 :
6522 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
6523 (ins VecListFourQAllLanes:$list, addrmode6:$addr,
6524 rGPR:$Rm, pred:$p)>;
6527 // VLD4 single-lane pseudo-instructions. These need special handling for
6528 // the lane index that an InstAlias can't handle, so we use these instead.
6529 def VLD4LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr",
6530 (ins VecListFourDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6531 def VLD4LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
6532 (ins VecListFourDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6533 def VLD4LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
6534 (ins VecListFourDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6535 def VLD4LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
6536 (ins VecListFourQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6537 def VLD4LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
6538 (ins VecListFourQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6540 def VLD4LNdWB_fixed_Asm_8 :
6541 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr!",
6542 (ins VecListFourDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6543 def VLD4LNdWB_fixed_Asm_16 :
6544 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
6545 (ins VecListFourDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6546 def VLD4LNdWB_fixed_Asm_32 :
6547 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
6548 (ins VecListFourDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6549 def VLD4LNqWB_fixed_Asm_16 :
6550 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
6551 (ins VecListFourQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6552 def VLD4LNqWB_fixed_Asm_32 :
6553 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
6554 (ins VecListFourQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6555 def VLD4LNdWB_register_Asm_8 :
6556 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr, $Rm",
6557 (ins VecListFourDByteIndexed:$list, addrmode6:$addr,
6558 rGPR:$Rm, pred:$p)>;
6559 def VLD4LNdWB_register_Asm_16 :
6560 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
6561 (ins VecListFourDHWordIndexed:$list, addrmode6:$addr,
6562 rGPR:$Rm, pred:$p)>;
6563 def VLD4LNdWB_register_Asm_32 :
6564 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
6565 (ins VecListFourDWordIndexed:$list, addrmode6:$addr,
6566 rGPR:$Rm, pred:$p)>;
6567 def VLD4LNqWB_register_Asm_16 :
6568 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
6569 (ins VecListFourQHWordIndexed:$list, addrmode6:$addr,
6570 rGPR:$Rm, pred:$p)>;
6571 def VLD4LNqWB_register_Asm_32 :
6572 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
6573 (ins VecListFourQWordIndexed:$list, addrmode6:$addr,
6574 rGPR:$Rm, pred:$p)>;
6578 // VLD4 multiple structure pseudo-instructions. These need special handling for
6579 // the vector operands that the normal instructions don't yet model.
6580 // FIXME: Remove these when the register classes and instructions are updated.
6581 def VLD4dAsm_8 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr",
6582 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6583 def VLD4dAsm_16 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
6584 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6585 def VLD4dAsm_32 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
6586 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6587 def VLD4qAsm_8 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr",
6588 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6589 def VLD4qAsm_16 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
6590 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6591 def VLD4qAsm_32 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
6592 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6594 def VLD4dWB_fixed_Asm_8 :
6595 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr!",
6596 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6597 def VLD4dWB_fixed_Asm_16 :
6598 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
6599 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6600 def VLD4dWB_fixed_Asm_32 :
6601 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
6602 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6603 def VLD4qWB_fixed_Asm_8 :
6604 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr!",
6605 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6606 def VLD4qWB_fixed_Asm_16 :
6607 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
6608 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6609 def VLD4qWB_fixed_Asm_32 :
6610 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
6611 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6612 def VLD4dWB_register_Asm_8 :
6613 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr, $Rm",
6614 (ins VecListFourD:$list, addrmode6:$addr,
6615 rGPR:$Rm, pred:$p)>;
6616 def VLD4dWB_register_Asm_16 :
6617 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
6618 (ins VecListFourD:$list, addrmode6:$addr,
6619 rGPR:$Rm, pred:$p)>;
6620 def VLD4dWB_register_Asm_32 :
6621 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
6622 (ins VecListFourD:$list, addrmode6:$addr,
6623 rGPR:$Rm, pred:$p)>;
6624 def VLD4qWB_register_Asm_8 :
6625 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr, $Rm",
6626 (ins VecListFourQ:$list, addrmode6:$addr,
6627 rGPR:$Rm, pred:$p)>;
6628 def VLD4qWB_register_Asm_16 :
6629 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
6630 (ins VecListFourQ:$list, addrmode6:$addr,
6631 rGPR:$Rm, pred:$p)>;
6632 def VLD4qWB_register_Asm_32 :
6633 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
6634 (ins VecListFourQ:$list, addrmode6:$addr,
6635 rGPR:$Rm, pred:$p)>;
6637 // VST4 single-lane pseudo-instructions. These need special handling for
6638 // the lane index that an InstAlias can't handle, so we use these instead.
6639 def VST4LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr",
6640 (ins VecListFourDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6641 def VST4LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr",
6642 (ins VecListFourDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6643 def VST4LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr",
6644 (ins VecListFourDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6645 def VST4LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr",
6646 (ins VecListFourQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6647 def VST4LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr",
6648 (ins VecListFourQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6650 def VST4LNdWB_fixed_Asm_8 :
6651 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr!",
6652 (ins VecListFourDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6653 def VST4LNdWB_fixed_Asm_16 :
6654 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr!",
6655 (ins VecListFourDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6656 def VST4LNdWB_fixed_Asm_32 :
6657 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr!",
6658 (ins VecListFourDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6659 def VST4LNqWB_fixed_Asm_16 :
6660 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr!",
6661 (ins VecListFourQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6662 def VST4LNqWB_fixed_Asm_32 :
6663 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr!",
6664 (ins VecListFourQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6665 def VST4LNdWB_register_Asm_8 :
6666 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr, $Rm",
6667 (ins VecListFourDByteIndexed:$list, addrmode6:$addr,
6668 rGPR:$Rm, pred:$p)>;
6669 def VST4LNdWB_register_Asm_16 :
6670 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr, $Rm",
6671 (ins VecListFourDHWordIndexed:$list, addrmode6:$addr,
6672 rGPR:$Rm, pred:$p)>;
6673 def VST4LNdWB_register_Asm_32 :
6674 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr, $Rm",
6675 (ins VecListFourDWordIndexed:$list, addrmode6:$addr,
6676 rGPR:$Rm, pred:$p)>;
6677 def VST4LNqWB_register_Asm_16 :
6678 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr, $Rm",
6679 (ins VecListFourQHWordIndexed:$list, addrmode6:$addr,
6680 rGPR:$Rm, pred:$p)>;
6681 def VST4LNqWB_register_Asm_32 :
6682 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr, $Rm",
6683 (ins VecListFourQWordIndexed:$list, addrmode6:$addr,
6684 rGPR:$Rm, pred:$p)>;
6687 // VST4 multiple structure pseudo-instructions. These need special handling for
6688 // the vector operands that the normal instructions don't yet model.
6689 // FIXME: Remove these when the register classes and instructions are updated.
6690 def VST4dAsm_8 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr",
6691 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6692 def VST4dAsm_16 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr",
6693 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6694 def VST4dAsm_32 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr",
6695 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6696 def VST4qAsm_8 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr",
6697 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6698 def VST4qAsm_16 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr",
6699 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6700 def VST4qAsm_32 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr",
6701 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6703 def VST4dWB_fixed_Asm_8 :
6704 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr!",
6705 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6706 def VST4dWB_fixed_Asm_16 :
6707 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr!",
6708 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6709 def VST4dWB_fixed_Asm_32 :
6710 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr!",
6711 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6712 def VST4qWB_fixed_Asm_8 :
6713 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr!",
6714 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6715 def VST4qWB_fixed_Asm_16 :
6716 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr!",
6717 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6718 def VST4qWB_fixed_Asm_32 :
6719 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr!",
6720 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6721 def VST4dWB_register_Asm_8 :
6722 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr, $Rm",
6723 (ins VecListFourD:$list, addrmode6:$addr,
6724 rGPR:$Rm, pred:$p)>;
6725 def VST4dWB_register_Asm_16 :
6726 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr, $Rm",
6727 (ins VecListFourD:$list, addrmode6:$addr,
6728 rGPR:$Rm, pred:$p)>;
6729 def VST4dWB_register_Asm_32 :
6730 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr, $Rm",
6731 (ins VecListFourD:$list, addrmode6:$addr,
6732 rGPR:$Rm, pred:$p)>;
6733 def VST4qWB_register_Asm_8 :
6734 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr, $Rm",
6735 (ins VecListFourQ:$list, addrmode6:$addr,
6736 rGPR:$Rm, pred:$p)>;
6737 def VST4qWB_register_Asm_16 :
6738 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr, $Rm",
6739 (ins VecListFourQ:$list, addrmode6:$addr,
6740 rGPR:$Rm, pred:$p)>;
6741 def VST4qWB_register_Asm_32 :
6742 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr, $Rm",
6743 (ins VecListFourQ:$list, addrmode6:$addr,
6744 rGPR:$Rm, pred:$p)>;
6746 // VMOV/VMVN takes an optional datatype suffix
6747 defm : NEONDTAnyInstAlias<"vmov${p}", "$Vd, $Vm",
6748 (VORRd DPR:$Vd, DPR:$Vm, DPR:$Vm, pred:$p)>;
6749 defm : NEONDTAnyInstAlias<"vmov${p}", "$Vd, $Vm",
6750 (VORRq QPR:$Vd, QPR:$Vm, QPR:$Vm, pred:$p)>;
6752 defm : NEONDTAnyInstAlias<"vmvn${p}", "$Vd, $Vm",
6753 (VMVNd DPR:$Vd, DPR:$Vm, pred:$p)>;
6754 defm : NEONDTAnyInstAlias<"vmvn${p}", "$Vd, $Vm",
6755 (VMVNq QPR:$Vd, QPR:$Vm, pred:$p)>;
6757 // VCLT (register) is an assembler alias for VCGT w/ the operands reversed.
6758 // D-register versions.
6759 def : NEONInstAlias<"vcle${p}.s8 $Dd, $Dn, $Dm",
6760 (VCGEsv8i8 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6761 def : NEONInstAlias<"vcle${p}.s16 $Dd, $Dn, $Dm",
6762 (VCGEsv4i16 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6763 def : NEONInstAlias<"vcle${p}.s32 $Dd, $Dn, $Dm",
6764 (VCGEsv2i32 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6765 def : NEONInstAlias<"vcle${p}.u8 $Dd, $Dn, $Dm",
6766 (VCGEuv8i8 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6767 def : NEONInstAlias<"vcle${p}.u16 $Dd, $Dn, $Dm",
6768 (VCGEuv4i16 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6769 def : NEONInstAlias<"vcle${p}.u32 $Dd, $Dn, $Dm",
6770 (VCGEuv2i32 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6771 def : NEONInstAlias<"vcle${p}.f32 $Dd, $Dn, $Dm",
6772 (VCGEfd DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6773 // Q-register versions.
6774 def : NEONInstAlias<"vcle${p}.s8 $Qd, $Qn, $Qm",
6775 (VCGEsv16i8 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6776 def : NEONInstAlias<"vcle${p}.s16 $Qd, $Qn, $Qm",
6777 (VCGEsv8i16 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6778 def : NEONInstAlias<"vcle${p}.s32 $Qd, $Qn, $Qm",
6779 (VCGEsv4i32 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6780 def : NEONInstAlias<"vcle${p}.u8 $Qd, $Qn, $Qm",
6781 (VCGEuv16i8 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6782 def : NEONInstAlias<"vcle${p}.u16 $Qd, $Qn, $Qm",
6783 (VCGEuv8i16 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6784 def : NEONInstAlias<"vcle${p}.u32 $Qd, $Qn, $Qm",
6785 (VCGEuv4i32 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6786 def : NEONInstAlias<"vcle${p}.f32 $Qd, $Qn, $Qm",
6787 (VCGEfq QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6789 // VCLT (register) is an assembler alias for VCGT w/ the operands reversed.
6790 // D-register versions.
6791 def : NEONInstAlias<"vclt${p}.s8 $Dd, $Dn, $Dm",
6792 (VCGTsv8i8 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6793 def : NEONInstAlias<"vclt${p}.s16 $Dd, $Dn, $Dm",
6794 (VCGTsv4i16 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6795 def : NEONInstAlias<"vclt${p}.s32 $Dd, $Dn, $Dm",
6796 (VCGTsv2i32 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6797 def : NEONInstAlias<"vclt${p}.u8 $Dd, $Dn, $Dm",
6798 (VCGTuv8i8 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6799 def : NEONInstAlias<"vclt${p}.u16 $Dd, $Dn, $Dm",
6800 (VCGTuv4i16 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6801 def : NEONInstAlias<"vclt${p}.u32 $Dd, $Dn, $Dm",
6802 (VCGTuv2i32 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6803 def : NEONInstAlias<"vclt${p}.f32 $Dd, $Dn, $Dm",
6804 (VCGTfd DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6805 // Q-register versions.
6806 def : NEONInstAlias<"vclt${p}.s8 $Qd, $Qn, $Qm",
6807 (VCGTsv16i8 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6808 def : NEONInstAlias<"vclt${p}.s16 $Qd, $Qn, $Qm",
6809 (VCGTsv8i16 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6810 def : NEONInstAlias<"vclt${p}.s32 $Qd, $Qn, $Qm",
6811 (VCGTsv4i32 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6812 def : NEONInstAlias<"vclt${p}.u8 $Qd, $Qn, $Qm",
6813 (VCGTuv16i8 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6814 def : NEONInstAlias<"vclt${p}.u16 $Qd, $Qn, $Qm",
6815 (VCGTuv8i16 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6816 def : NEONInstAlias<"vclt${p}.u32 $Qd, $Qn, $Qm",
6817 (VCGTuv4i32 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6818 def : NEONInstAlias<"vclt${p}.f32 $Qd, $Qn, $Qm",
6819 (VCGTfq QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6821 // VSWP allows, but does not require, a type suffix.
6822 defm : NEONDTAnyInstAlias<"vswp${p}", "$Vd, $Vm",
6823 (VSWPd DPR:$Vd, DPR:$Vm, pred:$p)>;
6824 defm : NEONDTAnyInstAlias<"vswp${p}", "$Vd, $Vm",
6825 (VSWPq QPR:$Vd, QPR:$Vm, pred:$p)>;
6827 // VBIF, VBIT, and VBSL allow, but do not require, a type suffix.
6828 defm : NEONDTAnyInstAlias<"vbif${p}", "$Vd, $Vn, $Vm",
6829 (VBIFd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
6830 defm : NEONDTAnyInstAlias<"vbit${p}", "$Vd, $Vn, $Vm",
6831 (VBITd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
6832 defm : NEONDTAnyInstAlias<"vbsl${p}", "$Vd, $Vn, $Vm",
6833 (VBSLd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
6834 defm : NEONDTAnyInstAlias<"vbif${p}", "$Vd, $Vn, $Vm",
6835 (VBIFq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
6836 defm : NEONDTAnyInstAlias<"vbit${p}", "$Vd, $Vn, $Vm",
6837 (VBITq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
6838 defm : NEONDTAnyInstAlias<"vbsl${p}", "$Vd, $Vn, $Vm",
6839 (VBSLq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
6841 // "vmov Rd, #-imm" can be handled via "vmvn".
6842 def : NEONInstAlias<"vmov${p}.i32 $Vd, $imm",
6843 (VMVNv2i32 DPR:$Vd, nImmVMOVI32Neg:$imm, pred:$p)>;
6844 def : NEONInstAlias<"vmov${p}.i32 $Vd, $imm",
6845 (VMVNv4i32 QPR:$Vd, nImmVMOVI32Neg:$imm, pred:$p)>;
6846 def : NEONInstAlias<"vmvn${p}.i32 $Vd, $imm",
6847 (VMOVv2i32 DPR:$Vd, nImmVMOVI32Neg:$imm, pred:$p)>;
6848 def : NEONInstAlias<"vmvn${p}.i32 $Vd, $imm",
6849 (VMOVv4i32 QPR:$Vd, nImmVMOVI32Neg:$imm, pred:$p)>;
6851 // 'gas' compatibility aliases for quad-word instructions. Strictly speaking,
6852 // these should restrict to just the Q register variants, but the register
6853 // classes are enough to match correctly regardless, so we keep it simple
6854 // and just use MnemonicAlias.
6855 def : NEONMnemonicAlias<"vbicq", "vbic">;
6856 def : NEONMnemonicAlias<"vandq", "vand">;
6857 def : NEONMnemonicAlias<"veorq", "veor">;
6858 def : NEONMnemonicAlias<"vorrq", "vorr">;
6860 def : NEONMnemonicAlias<"vmovq", "vmov">;
6861 def : NEONMnemonicAlias<"vmvnq", "vmvn">;
6862 // Explicit versions for floating point so that the FPImm variants get
6863 // handled early. The parser gets confused otherwise.
6864 def : NEONMnemonicAlias<"vmovq.f32", "vmov.f32">;
6865 def : NEONMnemonicAlias<"vmovq.f64", "vmov.f64">;
6867 def : NEONMnemonicAlias<"vaddq", "vadd">;
6868 def : NEONMnemonicAlias<"vsubq", "vsub">;
6870 def : NEONMnemonicAlias<"vminq", "vmin">;
6871 def : NEONMnemonicAlias<"vmaxq", "vmax">;
6873 def : NEONMnemonicAlias<"vmulq", "vmul">;
6875 def : NEONMnemonicAlias<"vabsq", "vabs">;
6877 def : NEONMnemonicAlias<"vshlq", "vshl">;
6878 def : NEONMnemonicAlias<"vshrq", "vshr">;
6880 def : NEONMnemonicAlias<"vcvtq", "vcvt">;
6882 def : NEONMnemonicAlias<"vcleq", "vcle">;
6883 def : NEONMnemonicAlias<"vceqq", "vceq">;
6885 def : NEONMnemonicAlias<"vzipq", "vzip">;
6886 def : NEONMnemonicAlias<"vswpq", "vswp">;
6888 def : NEONMnemonicAlias<"vrecpeq.f32", "vrecpe.f32">;
6889 def : NEONMnemonicAlias<"vrecpeq.u32", "vrecpe.u32">;
6892 // Alias for loading floating point immediates that aren't representable
6893 // using the vmov.f32 encoding but the bitpattern is representable using
6894 // the .i32 encoding.
6895 def : NEONInstAlias<"vmov${p}.f32 $Vd, $imm",
6896 (VMOVv4i32 QPR:$Vd, nImmVMOVI32:$imm, pred:$p)>;
6897 def : NEONInstAlias<"vmov${p}.f32 $Vd, $imm",
6898 (VMOVv2i32 DPR:$Vd, nImmVMOVI32:$imm, pred:$p)>;