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 SDTARMVLONG2: SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
538 SDTCisSameAs<1, 2>]>;
539 def NEONvmulls : SDNode<"ARMISD::VMULLs", SDTARMVLONG2>;
540 def NEONvmullu : SDNode<"ARMISD::VMULLu", SDTARMVLONG2>;
542 def NEONvaddls : SDNode<"ARMISD::VADDLs", SDTARMVLONG2>;
543 def NEONvaddlu : SDNode<"ARMISD::VADDLu", SDTARMVLONG2>;
545 def NEONvsubls : SDNode<"ARMISD::VSUBLs", SDTARMVLONG2>;
546 def NEONvsublu : SDNode<"ARMISD::VSUBLu", SDTARMVLONG2>;
548 def SDTARMFMAX : SDTypeProfile<1, 2, [SDTCisVT<0, f32>, SDTCisSameAs<0, 1>,
549 SDTCisSameAs<0, 2>]>;
550 def NEONfmax : SDNode<"ARMISD::FMAX", SDTARMFMAX>;
551 def NEONfmin : SDNode<"ARMISD::FMIN", SDTARMFMAX>;
553 def NEONimmAllZerosV: PatLeaf<(NEONvmovImm (i32 timm)), [{
554 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
555 unsigned EltBits = 0;
556 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
557 return (EltBits == 32 && EltVal == 0);
560 def NEONimmAllOnesV: PatLeaf<(NEONvmovImm (i32 timm)), [{
561 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
562 unsigned EltBits = 0;
563 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
564 return (EltBits == 8 && EltVal == 0xff);
567 //===----------------------------------------------------------------------===//
568 // NEON load / store instructions
569 //===----------------------------------------------------------------------===//
571 // Use VLDM to load a Q register as a D register pair.
572 // This is a pseudo instruction that is expanded to VLDMD after reg alloc.
574 : PseudoVFPLdStM<(outs DPair:$dst), (ins GPR:$Rn),
576 [(set DPair:$dst, (v2f64 (load GPR:$Rn)))]>;
578 // Use VSTM to store a Q register as a D register pair.
579 // This is a pseudo instruction that is expanded to VSTMD after reg alloc.
581 : PseudoVFPLdStM<(outs), (ins DPair:$src, GPR:$Rn),
583 [(store (v2f64 DPair:$src), GPR:$Rn)]>;
585 // Classes for VLD* pseudo-instructions with multi-register operands.
586 // These are expanded to real instructions after register allocation.
587 class VLDQPseudo<InstrItinClass itin>
588 : PseudoNLdSt<(outs QPR:$dst), (ins addrmode6:$addr), itin, "">;
589 class VLDQWBPseudo<InstrItinClass itin>
590 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
591 (ins addrmode6:$addr, am6offset:$offset), itin,
593 class VLDQWBfixedPseudo<InstrItinClass itin>
594 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
595 (ins addrmode6:$addr), itin,
597 class VLDQWBregisterPseudo<InstrItinClass itin>
598 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
599 (ins addrmode6:$addr, rGPR:$offset), itin,
602 class VLDQQPseudo<InstrItinClass itin>
603 : PseudoNLdSt<(outs QQPR:$dst), (ins addrmode6:$addr), itin, "">;
604 class VLDQQWBPseudo<InstrItinClass itin>
605 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
606 (ins addrmode6:$addr, am6offset:$offset), itin,
608 class VLDQQWBfixedPseudo<InstrItinClass itin>
609 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
610 (ins addrmode6:$addr), itin,
612 class VLDQQWBregisterPseudo<InstrItinClass itin>
613 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
614 (ins addrmode6:$addr, rGPR:$offset), itin,
618 class VLDQQQQPseudo<InstrItinClass itin>
619 : PseudoNLdSt<(outs QQQQPR:$dst), (ins addrmode6:$addr, QQQQPR:$src),itin,
621 class VLDQQQQWBPseudo<InstrItinClass itin>
622 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
623 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
624 "$addr.addr = $wb, $src = $dst">;
626 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
628 // VLD1 : Vector Load (multiple single elements)
629 class VLD1D<bits<4> op7_4, string Dt>
630 : NLdSt<0,0b10,0b0111,op7_4, (outs VecListOneD:$Vd),
631 (ins addrmode6:$Rn), IIC_VLD1,
632 "vld1", Dt, "$Vd, $Rn", "", []> {
635 let DecoderMethod = "DecodeVLDInstruction";
637 class VLD1Q<bits<4> op7_4, string Dt>
638 : NLdSt<0,0b10,0b1010,op7_4, (outs VecListDPair:$Vd),
639 (ins addrmode6:$Rn), IIC_VLD1x2,
640 "vld1", Dt, "$Vd, $Rn", "", []> {
642 let Inst{5-4} = Rn{5-4};
643 let DecoderMethod = "DecodeVLDInstruction";
646 def VLD1d8 : VLD1D<{0,0,0,?}, "8">;
647 def VLD1d16 : VLD1D<{0,1,0,?}, "16">;
648 def VLD1d32 : VLD1D<{1,0,0,?}, "32">;
649 def VLD1d64 : VLD1D<{1,1,0,?}, "64">;
651 def VLD1q8 : VLD1Q<{0,0,?,?}, "8">;
652 def VLD1q16 : VLD1Q<{0,1,?,?}, "16">;
653 def VLD1q32 : VLD1Q<{1,0,?,?}, "32">;
654 def VLD1q64 : VLD1Q<{1,1,?,?}, "64">;
656 // ...with address register writeback:
657 multiclass VLD1DWB<bits<4> op7_4, string Dt> {
658 def _fixed : NLdSt<0,0b10, 0b0111,op7_4, (outs VecListOneD:$Vd, GPR:$wb),
659 (ins addrmode6:$Rn), IIC_VLD1u,
660 "vld1", Dt, "$Vd, $Rn!",
661 "$Rn.addr = $wb", []> {
662 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
664 let DecoderMethod = "DecodeVLDInstruction";
665 let AsmMatchConverter = "cvtVLDwbFixed";
667 def _register : NLdSt<0,0b10,0b0111,op7_4, (outs VecListOneD:$Vd, GPR:$wb),
668 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1u,
669 "vld1", Dt, "$Vd, $Rn, $Rm",
670 "$Rn.addr = $wb", []> {
672 let DecoderMethod = "DecodeVLDInstruction";
673 let AsmMatchConverter = "cvtVLDwbRegister";
676 multiclass VLD1QWB<bits<4> op7_4, string Dt> {
677 def _fixed : NLdSt<0,0b10,0b1010,op7_4, (outs VecListDPair:$Vd, GPR:$wb),
678 (ins addrmode6:$Rn), IIC_VLD1x2u,
679 "vld1", Dt, "$Vd, $Rn!",
680 "$Rn.addr = $wb", []> {
681 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
682 let Inst{5-4} = Rn{5-4};
683 let DecoderMethod = "DecodeVLDInstruction";
684 let AsmMatchConverter = "cvtVLDwbFixed";
686 def _register : NLdSt<0,0b10,0b1010,op7_4, (outs VecListDPair:$Vd, GPR:$wb),
687 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1x2u,
688 "vld1", Dt, "$Vd, $Rn, $Rm",
689 "$Rn.addr = $wb", []> {
690 let Inst{5-4} = Rn{5-4};
691 let DecoderMethod = "DecodeVLDInstruction";
692 let AsmMatchConverter = "cvtVLDwbRegister";
696 defm VLD1d8wb : VLD1DWB<{0,0,0,?}, "8">;
697 defm VLD1d16wb : VLD1DWB<{0,1,0,?}, "16">;
698 defm VLD1d32wb : VLD1DWB<{1,0,0,?}, "32">;
699 defm VLD1d64wb : VLD1DWB<{1,1,0,?}, "64">;
700 defm VLD1q8wb : VLD1QWB<{0,0,?,?}, "8">;
701 defm VLD1q16wb : VLD1QWB<{0,1,?,?}, "16">;
702 defm VLD1q32wb : VLD1QWB<{1,0,?,?}, "32">;
703 defm VLD1q64wb : VLD1QWB<{1,1,?,?}, "64">;
705 // ...with 3 registers
706 class VLD1D3<bits<4> op7_4, string Dt>
707 : NLdSt<0,0b10,0b0110,op7_4, (outs VecListThreeD:$Vd),
708 (ins addrmode6:$Rn), IIC_VLD1x3, "vld1", Dt,
709 "$Vd, $Rn", "", []> {
712 let DecoderMethod = "DecodeVLDInstruction";
714 multiclass VLD1D3WB<bits<4> op7_4, string Dt> {
715 def _fixed : NLdSt<0,0b10,0b0110, op7_4, (outs VecListThreeD:$Vd, GPR:$wb),
716 (ins addrmode6:$Rn), IIC_VLD1x2u,
717 "vld1", Dt, "$Vd, $Rn!",
718 "$Rn.addr = $wb", []> {
719 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
721 let DecoderMethod = "DecodeVLDInstruction";
722 let AsmMatchConverter = "cvtVLDwbFixed";
724 def _register : NLdSt<0,0b10,0b0110,op7_4, (outs VecListThreeD:$Vd, GPR:$wb),
725 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1x2u,
726 "vld1", Dt, "$Vd, $Rn, $Rm",
727 "$Rn.addr = $wb", []> {
729 let DecoderMethod = "DecodeVLDInstruction";
730 let AsmMatchConverter = "cvtVLDwbRegister";
734 def VLD1d8T : VLD1D3<{0,0,0,?}, "8">;
735 def VLD1d16T : VLD1D3<{0,1,0,?}, "16">;
736 def VLD1d32T : VLD1D3<{1,0,0,?}, "32">;
737 def VLD1d64T : VLD1D3<{1,1,0,?}, "64">;
739 defm VLD1d8Twb : VLD1D3WB<{0,0,0,?}, "8">;
740 defm VLD1d16Twb : VLD1D3WB<{0,1,0,?}, "16">;
741 defm VLD1d32Twb : VLD1D3WB<{1,0,0,?}, "32">;
742 defm VLD1d64Twb : VLD1D3WB<{1,1,0,?}, "64">;
744 def VLD1d64TPseudo : VLDQQPseudo<IIC_VLD1x3>;
746 // ...with 4 registers
747 class VLD1D4<bits<4> op7_4, string Dt>
748 : NLdSt<0, 0b10, 0b0010, op7_4, (outs VecListFourD:$Vd),
749 (ins addrmode6:$Rn), IIC_VLD1x4, "vld1", Dt,
750 "$Vd, $Rn", "", []> {
752 let Inst{5-4} = Rn{5-4};
753 let DecoderMethod = "DecodeVLDInstruction";
755 multiclass VLD1D4WB<bits<4> op7_4, string Dt> {
756 def _fixed : NLdSt<0,0b10,0b0010, op7_4, (outs VecListFourD:$Vd, GPR:$wb),
757 (ins addrmode6:$Rn), IIC_VLD1x2u,
758 "vld1", Dt, "$Vd, $Rn!",
759 "$Rn.addr = $wb", []> {
760 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
761 let Inst{5-4} = Rn{5-4};
762 let DecoderMethod = "DecodeVLDInstruction";
763 let AsmMatchConverter = "cvtVLDwbFixed";
765 def _register : NLdSt<0,0b10,0b0010,op7_4, (outs VecListFourD:$Vd, GPR:$wb),
766 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1x2u,
767 "vld1", Dt, "$Vd, $Rn, $Rm",
768 "$Rn.addr = $wb", []> {
769 let Inst{5-4} = Rn{5-4};
770 let DecoderMethod = "DecodeVLDInstruction";
771 let AsmMatchConverter = "cvtVLDwbRegister";
775 def VLD1d8Q : VLD1D4<{0,0,?,?}, "8">;
776 def VLD1d16Q : VLD1D4<{0,1,?,?}, "16">;
777 def VLD1d32Q : VLD1D4<{1,0,?,?}, "32">;
778 def VLD1d64Q : VLD1D4<{1,1,?,?}, "64">;
780 defm VLD1d8Qwb : VLD1D4WB<{0,0,?,?}, "8">;
781 defm VLD1d16Qwb : VLD1D4WB<{0,1,?,?}, "16">;
782 defm VLD1d32Qwb : VLD1D4WB<{1,0,?,?}, "32">;
783 defm VLD1d64Qwb : VLD1D4WB<{1,1,?,?}, "64">;
785 def VLD1d64QPseudo : VLDQQPseudo<IIC_VLD1x4>;
787 // VLD2 : Vector Load (multiple 2-element structures)
788 class VLD2<bits<4> op11_8, bits<4> op7_4, string Dt, RegisterOperand VdTy,
790 : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd),
791 (ins addrmode6:$Rn), itin,
792 "vld2", Dt, "$Vd, $Rn", "", []> {
794 let Inst{5-4} = Rn{5-4};
795 let DecoderMethod = "DecodeVLDInstruction";
798 def VLD2d8 : VLD2<0b1000, {0,0,?,?}, "8", VecListDPair, IIC_VLD2>;
799 def VLD2d16 : VLD2<0b1000, {0,1,?,?}, "16", VecListDPair, IIC_VLD2>;
800 def VLD2d32 : VLD2<0b1000, {1,0,?,?}, "32", VecListDPair, IIC_VLD2>;
802 def VLD2q8 : VLD2<0b0011, {0,0,?,?}, "8", VecListFourD, IIC_VLD2x2>;
803 def VLD2q16 : VLD2<0b0011, {0,1,?,?}, "16", VecListFourD, IIC_VLD2x2>;
804 def VLD2q32 : VLD2<0b0011, {1,0,?,?}, "32", VecListFourD, IIC_VLD2x2>;
806 def VLD2q8Pseudo : VLDQQPseudo<IIC_VLD2x2>;
807 def VLD2q16Pseudo : VLDQQPseudo<IIC_VLD2x2>;
808 def VLD2q32Pseudo : VLDQQPseudo<IIC_VLD2x2>;
810 // ...with address register writeback:
811 multiclass VLD2WB<bits<4> op11_8, bits<4> op7_4, string Dt,
812 RegisterOperand VdTy, InstrItinClass itin> {
813 def _fixed : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd, GPR:$wb),
814 (ins addrmode6:$Rn), itin,
815 "vld2", Dt, "$Vd, $Rn!",
816 "$Rn.addr = $wb", []> {
817 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
818 let Inst{5-4} = Rn{5-4};
819 let DecoderMethod = "DecodeVLDInstruction";
820 let AsmMatchConverter = "cvtVLDwbFixed";
822 def _register : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd, GPR:$wb),
823 (ins addrmode6:$Rn, rGPR:$Rm), itin,
824 "vld2", Dt, "$Vd, $Rn, $Rm",
825 "$Rn.addr = $wb", []> {
826 let Inst{5-4} = Rn{5-4};
827 let DecoderMethod = "DecodeVLDInstruction";
828 let AsmMatchConverter = "cvtVLDwbRegister";
832 defm VLD2d8wb : VLD2WB<0b1000, {0,0,?,?}, "8", VecListDPair, IIC_VLD2u>;
833 defm VLD2d16wb : VLD2WB<0b1000, {0,1,?,?}, "16", VecListDPair, IIC_VLD2u>;
834 defm VLD2d32wb : VLD2WB<0b1000, {1,0,?,?}, "32", VecListDPair, IIC_VLD2u>;
836 defm VLD2q8wb : VLD2WB<0b0011, {0,0,?,?}, "8", VecListFourD, IIC_VLD2x2u>;
837 defm VLD2q16wb : VLD2WB<0b0011, {0,1,?,?}, "16", VecListFourD, IIC_VLD2x2u>;
838 defm VLD2q32wb : VLD2WB<0b0011, {1,0,?,?}, "32", VecListFourD, IIC_VLD2x2u>;
840 def VLD2q8PseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD2x2u>;
841 def VLD2q16PseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD2x2u>;
842 def VLD2q32PseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD2x2u>;
843 def VLD2q8PseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD2x2u>;
844 def VLD2q16PseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD2x2u>;
845 def VLD2q32PseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD2x2u>;
847 // ...with double-spaced registers
848 def VLD2b8 : VLD2<0b1001, {0,0,?,?}, "8", VecListDPairSpaced, IIC_VLD2>;
849 def VLD2b16 : VLD2<0b1001, {0,1,?,?}, "16", VecListDPairSpaced, IIC_VLD2>;
850 def VLD2b32 : VLD2<0b1001, {1,0,?,?}, "32", VecListDPairSpaced, IIC_VLD2>;
851 defm VLD2b8wb : VLD2WB<0b1001, {0,0,?,?}, "8", VecListDPairSpaced, IIC_VLD2u>;
852 defm VLD2b16wb : VLD2WB<0b1001, {0,1,?,?}, "16", VecListDPairSpaced, IIC_VLD2u>;
853 defm VLD2b32wb : VLD2WB<0b1001, {1,0,?,?}, "32", VecListDPairSpaced, IIC_VLD2u>;
855 // VLD3 : Vector Load (multiple 3-element structures)
856 class VLD3D<bits<4> op11_8, bits<4> op7_4, string Dt>
857 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
858 (ins addrmode6:$Rn), IIC_VLD3,
859 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn", "", []> {
862 let DecoderMethod = "DecodeVLDInstruction";
865 def VLD3d8 : VLD3D<0b0100, {0,0,0,?}, "8">;
866 def VLD3d16 : VLD3D<0b0100, {0,1,0,?}, "16">;
867 def VLD3d32 : VLD3D<0b0100, {1,0,0,?}, "32">;
869 def VLD3d8Pseudo : VLDQQPseudo<IIC_VLD3>;
870 def VLD3d16Pseudo : VLDQQPseudo<IIC_VLD3>;
871 def VLD3d32Pseudo : VLDQQPseudo<IIC_VLD3>;
873 // ...with address register writeback:
874 class VLD3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
875 : NLdSt<0, 0b10, op11_8, op7_4,
876 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
877 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD3u,
878 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn$Rm",
879 "$Rn.addr = $wb", []> {
881 let DecoderMethod = "DecodeVLDInstruction";
884 def VLD3d8_UPD : VLD3DWB<0b0100, {0,0,0,?}, "8">;
885 def VLD3d16_UPD : VLD3DWB<0b0100, {0,1,0,?}, "16">;
886 def VLD3d32_UPD : VLD3DWB<0b0100, {1,0,0,?}, "32">;
888 def VLD3d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
889 def VLD3d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
890 def VLD3d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
892 // ...with double-spaced registers:
893 def VLD3q8 : VLD3D<0b0101, {0,0,0,?}, "8">;
894 def VLD3q16 : VLD3D<0b0101, {0,1,0,?}, "16">;
895 def VLD3q32 : VLD3D<0b0101, {1,0,0,?}, "32">;
896 def VLD3q8_UPD : VLD3DWB<0b0101, {0,0,0,?}, "8">;
897 def VLD3q16_UPD : VLD3DWB<0b0101, {0,1,0,?}, "16">;
898 def VLD3q32_UPD : VLD3DWB<0b0101, {1,0,0,?}, "32">;
900 def VLD3q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
901 def VLD3q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
902 def VLD3q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
904 // ...alternate versions to be allocated odd register numbers:
905 def VLD3q8oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
906 def VLD3q16oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
907 def VLD3q32oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
909 def VLD3q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
910 def VLD3q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
911 def VLD3q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
913 // VLD4 : Vector Load (multiple 4-element structures)
914 class VLD4D<bits<4> op11_8, bits<4> op7_4, string Dt>
915 : NLdSt<0, 0b10, op11_8, op7_4,
916 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
917 (ins addrmode6:$Rn), IIC_VLD4,
918 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> {
920 let Inst{5-4} = Rn{5-4};
921 let DecoderMethod = "DecodeVLDInstruction";
924 def VLD4d8 : VLD4D<0b0000, {0,0,?,?}, "8">;
925 def VLD4d16 : VLD4D<0b0000, {0,1,?,?}, "16">;
926 def VLD4d32 : VLD4D<0b0000, {1,0,?,?}, "32">;
928 def VLD4d8Pseudo : VLDQQPseudo<IIC_VLD4>;
929 def VLD4d16Pseudo : VLDQQPseudo<IIC_VLD4>;
930 def VLD4d32Pseudo : VLDQQPseudo<IIC_VLD4>;
932 // ...with address register writeback:
933 class VLD4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
934 : NLdSt<0, 0b10, op11_8, op7_4,
935 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
936 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD4u,
937 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm",
938 "$Rn.addr = $wb", []> {
939 let Inst{5-4} = Rn{5-4};
940 let DecoderMethod = "DecodeVLDInstruction";
943 def VLD4d8_UPD : VLD4DWB<0b0000, {0,0,?,?}, "8">;
944 def VLD4d16_UPD : VLD4DWB<0b0000, {0,1,?,?}, "16">;
945 def VLD4d32_UPD : VLD4DWB<0b0000, {1,0,?,?}, "32">;
947 def VLD4d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
948 def VLD4d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
949 def VLD4d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
951 // ...with double-spaced registers:
952 def VLD4q8 : VLD4D<0b0001, {0,0,?,?}, "8">;
953 def VLD4q16 : VLD4D<0b0001, {0,1,?,?}, "16">;
954 def VLD4q32 : VLD4D<0b0001, {1,0,?,?}, "32">;
955 def VLD4q8_UPD : VLD4DWB<0b0001, {0,0,?,?}, "8">;
956 def VLD4q16_UPD : VLD4DWB<0b0001, {0,1,?,?}, "16">;
957 def VLD4q32_UPD : VLD4DWB<0b0001, {1,0,?,?}, "32">;
959 def VLD4q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
960 def VLD4q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
961 def VLD4q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
963 // ...alternate versions to be allocated odd register numbers:
964 def VLD4q8oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
965 def VLD4q16oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
966 def VLD4q32oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
968 def VLD4q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
969 def VLD4q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
970 def VLD4q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
972 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
974 // Classes for VLD*LN pseudo-instructions with multi-register operands.
975 // These are expanded to real instructions after register allocation.
976 class VLDQLNPseudo<InstrItinClass itin>
977 : PseudoNLdSt<(outs QPR:$dst),
978 (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
979 itin, "$src = $dst">;
980 class VLDQLNWBPseudo<InstrItinClass itin>
981 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
982 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
983 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
984 class VLDQQLNPseudo<InstrItinClass itin>
985 : PseudoNLdSt<(outs QQPR:$dst),
986 (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
987 itin, "$src = $dst">;
988 class VLDQQLNWBPseudo<InstrItinClass itin>
989 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
990 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
991 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
992 class VLDQQQQLNPseudo<InstrItinClass itin>
993 : PseudoNLdSt<(outs QQQQPR:$dst),
994 (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
995 itin, "$src = $dst">;
996 class VLDQQQQLNWBPseudo<InstrItinClass itin>
997 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
998 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
999 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
1001 // VLD1LN : Vector Load (single element to one lane)
1002 class VLD1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1004 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
1005 (ins addrmode6:$Rn, DPR:$src, nohash_imm:$lane),
1006 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
1008 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
1009 (i32 (LoadOp addrmode6:$Rn)),
1012 let DecoderMethod = "DecodeVLD1LN";
1014 class VLD1LN32<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1016 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
1017 (ins addrmode6oneL32:$Rn, DPR:$src, nohash_imm:$lane),
1018 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
1020 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
1021 (i32 (LoadOp addrmode6oneL32:$Rn)),
1024 let DecoderMethod = "DecodeVLD1LN";
1026 class VLD1QLNPseudo<ValueType Ty, PatFrag LoadOp> : VLDQLNPseudo<IIC_VLD1ln> {
1027 let Pattern = [(set QPR:$dst, (vector_insert (Ty QPR:$src),
1028 (i32 (LoadOp addrmode6:$addr)),
1032 def VLD1LNd8 : VLD1LN<0b0000, {?,?,?,0}, "8", v8i8, extloadi8> {
1033 let Inst{7-5} = lane{2-0};
1035 def VLD1LNd16 : VLD1LN<0b0100, {?,?,0,?}, "16", v4i16, extloadi16> {
1036 let Inst{7-6} = lane{1-0};
1037 let Inst{5-4} = Rn{5-4};
1039 def VLD1LNd32 : VLD1LN32<0b1000, {?,0,?,?}, "32", v2i32, load> {
1040 let Inst{7} = lane{0};
1041 let Inst{5-4} = Rn{5-4};
1044 def VLD1LNq8Pseudo : VLD1QLNPseudo<v16i8, extloadi8>;
1045 def VLD1LNq16Pseudo : VLD1QLNPseudo<v8i16, extloadi16>;
1046 def VLD1LNq32Pseudo : VLD1QLNPseudo<v4i32, load>;
1048 def : Pat<(vector_insert (v2f32 DPR:$src),
1049 (f32 (load addrmode6:$addr)), imm:$lane),
1050 (VLD1LNd32 addrmode6:$addr, DPR:$src, imm:$lane)>;
1051 def : Pat<(vector_insert (v4f32 QPR:$src),
1052 (f32 (load addrmode6:$addr)), imm:$lane),
1053 (VLD1LNq32Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
1055 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
1057 // ...with address register writeback:
1058 class VLD1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1059 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, GPR:$wb),
1060 (ins addrmode6:$Rn, am6offset:$Rm,
1061 DPR:$src, nohash_imm:$lane), IIC_VLD1lnu, "vld1", Dt,
1062 "\\{$Vd[$lane]\\}, $Rn$Rm",
1063 "$src = $Vd, $Rn.addr = $wb", []> {
1064 let DecoderMethod = "DecodeVLD1LN";
1067 def VLD1LNd8_UPD : VLD1LNWB<0b0000, {?,?,?,0}, "8"> {
1068 let Inst{7-5} = lane{2-0};
1070 def VLD1LNd16_UPD : VLD1LNWB<0b0100, {?,?,0,?}, "16"> {
1071 let Inst{7-6} = lane{1-0};
1072 let Inst{4} = Rn{4};
1074 def VLD1LNd32_UPD : VLD1LNWB<0b1000, {?,0,?,?}, "32"> {
1075 let Inst{7} = lane{0};
1076 let Inst{5} = Rn{4};
1077 let Inst{4} = Rn{4};
1080 def VLD1LNq8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
1081 def VLD1LNq16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
1082 def VLD1LNq32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
1084 // VLD2LN : Vector Load (single 2-element structure to one lane)
1085 class VLD2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1086 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2),
1087 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, nohash_imm:$lane),
1088 IIC_VLD2ln, "vld2", Dt, "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn",
1089 "$src1 = $Vd, $src2 = $dst2", []> {
1091 let Inst{4} = Rn{4};
1092 let DecoderMethod = "DecodeVLD2LN";
1095 def VLD2LNd8 : VLD2LN<0b0001, {?,?,?,?}, "8"> {
1096 let Inst{7-5} = lane{2-0};
1098 def VLD2LNd16 : VLD2LN<0b0101, {?,?,0,?}, "16"> {
1099 let Inst{7-6} = lane{1-0};
1101 def VLD2LNd32 : VLD2LN<0b1001, {?,0,0,?}, "32"> {
1102 let Inst{7} = lane{0};
1105 def VLD2LNd8Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
1106 def VLD2LNd16Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
1107 def VLD2LNd32Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
1109 // ...with double-spaced registers:
1110 def VLD2LNq16 : VLD2LN<0b0101, {?,?,1,?}, "16"> {
1111 let Inst{7-6} = lane{1-0};
1113 def VLD2LNq32 : VLD2LN<0b1001, {?,1,0,?}, "32"> {
1114 let Inst{7} = lane{0};
1117 def VLD2LNq16Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
1118 def VLD2LNq32Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
1120 // ...with address register writeback:
1121 class VLD2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1122 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
1123 (ins addrmode6:$Rn, am6offset:$Rm,
1124 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VLD2lnu, "vld2", Dt,
1125 "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn$Rm",
1126 "$src1 = $Vd, $src2 = $dst2, $Rn.addr = $wb", []> {
1127 let Inst{4} = Rn{4};
1128 let DecoderMethod = "DecodeVLD2LN";
1131 def VLD2LNd8_UPD : VLD2LNWB<0b0001, {?,?,?,?}, "8"> {
1132 let Inst{7-5} = lane{2-0};
1134 def VLD2LNd16_UPD : VLD2LNWB<0b0101, {?,?,0,?}, "16"> {
1135 let Inst{7-6} = lane{1-0};
1137 def VLD2LNd32_UPD : VLD2LNWB<0b1001, {?,0,0,?}, "32"> {
1138 let Inst{7} = lane{0};
1141 def VLD2LNd8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
1142 def VLD2LNd16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
1143 def VLD2LNd32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
1145 def VLD2LNq16_UPD : VLD2LNWB<0b0101, {?,?,1,?}, "16"> {
1146 let Inst{7-6} = lane{1-0};
1148 def VLD2LNq32_UPD : VLD2LNWB<0b1001, {?,1,0,?}, "32"> {
1149 let Inst{7} = lane{0};
1152 def VLD2LNq16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
1153 def VLD2LNq32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
1155 // VLD3LN : Vector Load (single 3-element structure to one lane)
1156 class VLD3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1157 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
1158 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3,
1159 nohash_imm:$lane), IIC_VLD3ln, "vld3", Dt,
1160 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn",
1161 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3", []> {
1163 let DecoderMethod = "DecodeVLD3LN";
1166 def VLD3LNd8 : VLD3LN<0b0010, {?,?,?,0}, "8"> {
1167 let Inst{7-5} = lane{2-0};
1169 def VLD3LNd16 : VLD3LN<0b0110, {?,?,0,0}, "16"> {
1170 let Inst{7-6} = lane{1-0};
1172 def VLD3LNd32 : VLD3LN<0b1010, {?,0,0,0}, "32"> {
1173 let Inst{7} = lane{0};
1176 def VLD3LNd8Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
1177 def VLD3LNd16Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
1178 def VLD3LNd32Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
1180 // ...with double-spaced registers:
1181 def VLD3LNq16 : VLD3LN<0b0110, {?,?,1,0}, "16"> {
1182 let Inst{7-6} = lane{1-0};
1184 def VLD3LNq32 : VLD3LN<0b1010, {?,1,0,0}, "32"> {
1185 let Inst{7} = lane{0};
1188 def VLD3LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
1189 def VLD3LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
1191 // ...with address register writeback:
1192 class VLD3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1193 : NLdStLn<1, 0b10, op11_8, op7_4,
1194 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
1195 (ins addrmode6:$Rn, am6offset:$Rm,
1196 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
1197 IIC_VLD3lnu, "vld3", Dt,
1198 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn$Rm",
1199 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $Rn.addr = $wb",
1201 let DecoderMethod = "DecodeVLD3LN";
1204 def VLD3LNd8_UPD : VLD3LNWB<0b0010, {?,?,?,0}, "8"> {
1205 let Inst{7-5} = lane{2-0};
1207 def VLD3LNd16_UPD : VLD3LNWB<0b0110, {?,?,0,0}, "16"> {
1208 let Inst{7-6} = lane{1-0};
1210 def VLD3LNd32_UPD : VLD3LNWB<0b1010, {?,0,0,0}, "32"> {
1211 let Inst{7} = lane{0};
1214 def VLD3LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
1215 def VLD3LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
1216 def VLD3LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
1218 def VLD3LNq16_UPD : VLD3LNWB<0b0110, {?,?,1,0}, "16"> {
1219 let Inst{7-6} = lane{1-0};
1221 def VLD3LNq32_UPD : VLD3LNWB<0b1010, {?,1,0,0}, "32"> {
1222 let Inst{7} = lane{0};
1225 def VLD3LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
1226 def VLD3LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
1228 // VLD4LN : Vector Load (single 4-element structure to one lane)
1229 class VLD4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1230 : NLdStLn<1, 0b10, op11_8, op7_4,
1231 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
1232 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
1233 nohash_imm:$lane), IIC_VLD4ln, "vld4", Dt,
1234 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn",
1235 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4", []> {
1237 let Inst{4} = Rn{4};
1238 let DecoderMethod = "DecodeVLD4LN";
1241 def VLD4LNd8 : VLD4LN<0b0011, {?,?,?,?}, "8"> {
1242 let Inst{7-5} = lane{2-0};
1244 def VLD4LNd16 : VLD4LN<0b0111, {?,?,0,?}, "16"> {
1245 let Inst{7-6} = lane{1-0};
1247 def VLD4LNd32 : VLD4LN<0b1011, {?,0,?,?}, "32"> {
1248 let Inst{7} = lane{0};
1249 let Inst{5} = Rn{5};
1252 def VLD4LNd8Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
1253 def VLD4LNd16Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
1254 def VLD4LNd32Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
1256 // ...with double-spaced registers:
1257 def VLD4LNq16 : VLD4LN<0b0111, {?,?,1,?}, "16"> {
1258 let Inst{7-6} = lane{1-0};
1260 def VLD4LNq32 : VLD4LN<0b1011, {?,1,?,?}, "32"> {
1261 let Inst{7} = lane{0};
1262 let Inst{5} = Rn{5};
1265 def VLD4LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
1266 def VLD4LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
1268 // ...with address register writeback:
1269 class VLD4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1270 : NLdStLn<1, 0b10, op11_8, op7_4,
1271 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
1272 (ins addrmode6:$Rn, am6offset:$Rm,
1273 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
1274 IIC_VLD4lnu, "vld4", Dt,
1275 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn$Rm",
1276 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4, $Rn.addr = $wb",
1278 let Inst{4} = Rn{4};
1279 let DecoderMethod = "DecodeVLD4LN" ;
1282 def VLD4LNd8_UPD : VLD4LNWB<0b0011, {?,?,?,?}, "8"> {
1283 let Inst{7-5} = lane{2-0};
1285 def VLD4LNd16_UPD : VLD4LNWB<0b0111, {?,?,0,?}, "16"> {
1286 let Inst{7-6} = lane{1-0};
1288 def VLD4LNd32_UPD : VLD4LNWB<0b1011, {?,0,?,?}, "32"> {
1289 let Inst{7} = lane{0};
1290 let Inst{5} = Rn{5};
1293 def VLD4LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
1294 def VLD4LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
1295 def VLD4LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
1297 def VLD4LNq16_UPD : VLD4LNWB<0b0111, {?,?,1,?}, "16"> {
1298 let Inst{7-6} = lane{1-0};
1300 def VLD4LNq32_UPD : VLD4LNWB<0b1011, {?,1,?,?}, "32"> {
1301 let Inst{7} = lane{0};
1302 let Inst{5} = Rn{5};
1305 def VLD4LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
1306 def VLD4LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
1308 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
1310 // VLD1DUP : Vector Load (single element to all lanes)
1311 class VLD1DUP<bits<4> op7_4, string Dt, ValueType Ty, PatFrag LoadOp>
1312 : NLdSt<1, 0b10, 0b1100, op7_4, (outs VecListOneDAllLanes:$Vd),
1313 (ins addrmode6dup:$Rn),
1314 IIC_VLD1dup, "vld1", Dt, "$Vd, $Rn", "",
1315 [(set VecListOneDAllLanes:$Vd,
1316 (Ty (NEONvdup (i32 (LoadOp addrmode6dup:$Rn)))))]> {
1318 let Inst{4} = Rn{4};
1319 let DecoderMethod = "DecodeVLD1DupInstruction";
1321 def VLD1DUPd8 : VLD1DUP<{0,0,0,?}, "8", v8i8, extloadi8>;
1322 def VLD1DUPd16 : VLD1DUP<{0,1,0,?}, "16", v4i16, extloadi16>;
1323 def VLD1DUPd32 : VLD1DUP<{1,0,0,?}, "32", v2i32, load>;
1325 def : Pat<(v2f32 (NEONvdup (f32 (load addrmode6dup:$addr)))),
1326 (VLD1DUPd32 addrmode6:$addr)>;
1328 class VLD1QDUP<bits<4> op7_4, string Dt, ValueType Ty, PatFrag LoadOp>
1329 : NLdSt<1, 0b10, 0b1100, op7_4, (outs VecListDPairAllLanes:$Vd),
1330 (ins addrmode6dup:$Rn), IIC_VLD1dup,
1331 "vld1", Dt, "$Vd, $Rn", "",
1332 [(set VecListDPairAllLanes:$Vd,
1333 (Ty (NEONvdup (i32 (LoadOp addrmode6dup:$Rn)))))]> {
1335 let Inst{4} = Rn{4};
1336 let DecoderMethod = "DecodeVLD1DupInstruction";
1339 def VLD1DUPq8 : VLD1QDUP<{0,0,1,0}, "8", v16i8, extloadi8>;
1340 def VLD1DUPq16 : VLD1QDUP<{0,1,1,?}, "16", v8i16, extloadi16>;
1341 def VLD1DUPq32 : VLD1QDUP<{1,0,1,?}, "32", v4i32, load>;
1343 def : Pat<(v4f32 (NEONvdup (f32 (load addrmode6dup:$addr)))),
1344 (VLD1DUPq32 addrmode6:$addr)>;
1346 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
1347 // ...with address register writeback:
1348 multiclass VLD1DUPWB<bits<4> op7_4, string Dt> {
1349 def _fixed : NLdSt<1, 0b10, 0b1100, op7_4,
1350 (outs VecListOneDAllLanes:$Vd, GPR:$wb),
1351 (ins addrmode6dup:$Rn), IIC_VLD1dupu,
1352 "vld1", Dt, "$Vd, $Rn!",
1353 "$Rn.addr = $wb", []> {
1354 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1355 let Inst{4} = Rn{4};
1356 let DecoderMethod = "DecodeVLD1DupInstruction";
1357 let AsmMatchConverter = "cvtVLDwbFixed";
1359 def _register : NLdSt<1, 0b10, 0b1100, op7_4,
1360 (outs VecListOneDAllLanes:$Vd, GPR:$wb),
1361 (ins addrmode6dup:$Rn, rGPR:$Rm), IIC_VLD1dupu,
1362 "vld1", Dt, "$Vd, $Rn, $Rm",
1363 "$Rn.addr = $wb", []> {
1364 let Inst{4} = Rn{4};
1365 let DecoderMethod = "DecodeVLD1DupInstruction";
1366 let AsmMatchConverter = "cvtVLDwbRegister";
1369 multiclass VLD1QDUPWB<bits<4> op7_4, string Dt> {
1370 def _fixed : NLdSt<1, 0b10, 0b1100, op7_4,
1371 (outs VecListDPairAllLanes:$Vd, GPR:$wb),
1372 (ins addrmode6dup:$Rn), IIC_VLD1dupu,
1373 "vld1", Dt, "$Vd, $Rn!",
1374 "$Rn.addr = $wb", []> {
1375 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1376 let Inst{4} = Rn{4};
1377 let DecoderMethod = "DecodeVLD1DupInstruction";
1378 let AsmMatchConverter = "cvtVLDwbFixed";
1380 def _register : NLdSt<1, 0b10, 0b1100, op7_4,
1381 (outs VecListDPairAllLanes:$Vd, GPR:$wb),
1382 (ins addrmode6dup:$Rn, rGPR:$Rm), IIC_VLD1dupu,
1383 "vld1", Dt, "$Vd, $Rn, $Rm",
1384 "$Rn.addr = $wb", []> {
1385 let Inst{4} = Rn{4};
1386 let DecoderMethod = "DecodeVLD1DupInstruction";
1387 let AsmMatchConverter = "cvtVLDwbRegister";
1391 defm VLD1DUPd8wb : VLD1DUPWB<{0,0,0,0}, "8">;
1392 defm VLD1DUPd16wb : VLD1DUPWB<{0,1,0,?}, "16">;
1393 defm VLD1DUPd32wb : VLD1DUPWB<{1,0,0,?}, "32">;
1395 defm VLD1DUPq8wb : VLD1QDUPWB<{0,0,1,0}, "8">;
1396 defm VLD1DUPq16wb : VLD1QDUPWB<{0,1,1,?}, "16">;
1397 defm VLD1DUPq32wb : VLD1QDUPWB<{1,0,1,?}, "32">;
1399 // VLD2DUP : Vector Load (single 2-element structure to all lanes)
1400 class VLD2DUP<bits<4> op7_4, string Dt, RegisterOperand VdTy>
1401 : NLdSt<1, 0b10, 0b1101, op7_4, (outs VdTy:$Vd),
1402 (ins addrmode6dup:$Rn), IIC_VLD2dup,
1403 "vld2", Dt, "$Vd, $Rn", "", []> {
1405 let Inst{4} = Rn{4};
1406 let DecoderMethod = "DecodeVLD2DupInstruction";
1409 def VLD2DUPd8 : VLD2DUP<{0,0,0,?}, "8", VecListDPairAllLanes>;
1410 def VLD2DUPd16 : VLD2DUP<{0,1,0,?}, "16", VecListDPairAllLanes>;
1411 def VLD2DUPd32 : VLD2DUP<{1,0,0,?}, "32", VecListDPairAllLanes>;
1413 // ...with double-spaced registers
1414 def VLD2DUPd8x2 : VLD2DUP<{0,0,1,?}, "8", VecListDPairSpacedAllLanes>;
1415 def VLD2DUPd16x2 : VLD2DUP<{0,1,1,?}, "16", VecListDPairSpacedAllLanes>;
1416 def VLD2DUPd32x2 : VLD2DUP<{1,0,1,?}, "32", VecListDPairSpacedAllLanes>;
1418 // ...with address register writeback:
1419 multiclass VLD2DUPWB<bits<4> op7_4, string Dt, RegisterOperand VdTy> {
1420 def _fixed : NLdSt<1, 0b10, 0b1101, op7_4,
1421 (outs VdTy:$Vd, GPR:$wb),
1422 (ins addrmode6dup:$Rn), IIC_VLD2dupu,
1423 "vld2", Dt, "$Vd, $Rn!",
1424 "$Rn.addr = $wb", []> {
1425 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1426 let Inst{4} = Rn{4};
1427 let DecoderMethod = "DecodeVLD2DupInstruction";
1428 let AsmMatchConverter = "cvtVLDwbFixed";
1430 def _register : NLdSt<1, 0b10, 0b1101, op7_4,
1431 (outs VdTy:$Vd, GPR:$wb),
1432 (ins addrmode6dup:$Rn, rGPR:$Rm), IIC_VLD2dupu,
1433 "vld2", Dt, "$Vd, $Rn, $Rm",
1434 "$Rn.addr = $wb", []> {
1435 let Inst{4} = Rn{4};
1436 let DecoderMethod = "DecodeVLD2DupInstruction";
1437 let AsmMatchConverter = "cvtVLDwbRegister";
1441 defm VLD2DUPd8wb : VLD2DUPWB<{0,0,0,0}, "8", VecListDPairAllLanes>;
1442 defm VLD2DUPd16wb : VLD2DUPWB<{0,1,0,?}, "16", VecListDPairAllLanes>;
1443 defm VLD2DUPd32wb : VLD2DUPWB<{1,0,0,?}, "32", VecListDPairAllLanes>;
1445 defm VLD2DUPd8x2wb : VLD2DUPWB<{0,0,1,0}, "8", VecListDPairSpacedAllLanes>;
1446 defm VLD2DUPd16x2wb : VLD2DUPWB<{0,1,1,?}, "16", VecListDPairSpacedAllLanes>;
1447 defm VLD2DUPd32x2wb : VLD2DUPWB<{1,0,1,?}, "32", VecListDPairSpacedAllLanes>;
1449 // VLD3DUP : Vector Load (single 3-element structure to all lanes)
1450 class VLD3DUP<bits<4> op7_4, string Dt>
1451 : NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
1452 (ins addrmode6dup:$Rn), IIC_VLD3dup,
1453 "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn", "", []> {
1456 let DecoderMethod = "DecodeVLD3DupInstruction";
1459 def VLD3DUPd8 : VLD3DUP<{0,0,0,?}, "8">;
1460 def VLD3DUPd16 : VLD3DUP<{0,1,0,?}, "16">;
1461 def VLD3DUPd32 : VLD3DUP<{1,0,0,?}, "32">;
1463 def VLD3DUPd8Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1464 def VLD3DUPd16Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1465 def VLD3DUPd32Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1467 // ...with double-spaced registers (not used for codegen):
1468 def VLD3DUPq8 : VLD3DUP<{0,0,1,?}, "8">;
1469 def VLD3DUPq16 : VLD3DUP<{0,1,1,?}, "16">;
1470 def VLD3DUPq32 : VLD3DUP<{1,0,1,?}, "32">;
1472 // ...with address register writeback:
1473 class VLD3DUPWB<bits<4> op7_4, string Dt>
1474 : NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
1475 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD3dupu,
1476 "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn$Rm",
1477 "$Rn.addr = $wb", []> {
1479 let DecoderMethod = "DecodeVLD3DupInstruction";
1482 def VLD3DUPd8_UPD : VLD3DUPWB<{0,0,0,0}, "8">;
1483 def VLD3DUPd16_UPD : VLD3DUPWB<{0,1,0,?}, "16">;
1484 def VLD3DUPd32_UPD : VLD3DUPWB<{1,0,0,?}, "32">;
1486 def VLD3DUPq8_UPD : VLD3DUPWB<{0,0,1,0}, "8">;
1487 def VLD3DUPq16_UPD : VLD3DUPWB<{0,1,1,?}, "16">;
1488 def VLD3DUPq32_UPD : VLD3DUPWB<{1,0,1,?}, "32">;
1490 def VLD3DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1491 def VLD3DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1492 def VLD3DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1494 // VLD4DUP : Vector Load (single 4-element structure to all lanes)
1495 class VLD4DUP<bits<4> op7_4, string Dt>
1496 : NLdSt<1, 0b10, 0b1111, op7_4,
1497 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
1498 (ins addrmode6dup:$Rn), IIC_VLD4dup,
1499 "vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn", "", []> {
1501 let Inst{4} = Rn{4};
1502 let DecoderMethod = "DecodeVLD4DupInstruction";
1505 def VLD4DUPd8 : VLD4DUP<{0,0,0,?}, "8">;
1506 def VLD4DUPd16 : VLD4DUP<{0,1,0,?}, "16">;
1507 def VLD4DUPd32 : VLD4DUP<{1,?,0,?}, "32"> { let Inst{6} = Rn{5}; }
1509 def VLD4DUPd8Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1510 def VLD4DUPd16Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1511 def VLD4DUPd32Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1513 // ...with double-spaced registers (not used for codegen):
1514 def VLD4DUPq8 : VLD4DUP<{0,0,1,?}, "8">;
1515 def VLD4DUPq16 : VLD4DUP<{0,1,1,?}, "16">;
1516 def VLD4DUPq32 : VLD4DUP<{1,?,1,?}, "32"> { let Inst{6} = Rn{5}; }
1518 // ...with address register writeback:
1519 class VLD4DUPWB<bits<4> op7_4, string Dt>
1520 : NLdSt<1, 0b10, 0b1111, op7_4,
1521 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
1522 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD4dupu,
1523 "vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn$Rm",
1524 "$Rn.addr = $wb", []> {
1525 let Inst{4} = Rn{4};
1526 let DecoderMethod = "DecodeVLD4DupInstruction";
1529 def VLD4DUPd8_UPD : VLD4DUPWB<{0,0,0,0}, "8">;
1530 def VLD4DUPd16_UPD : VLD4DUPWB<{0,1,0,?}, "16">;
1531 def VLD4DUPd32_UPD : VLD4DUPWB<{1,?,0,?}, "32"> { let Inst{6} = Rn{5}; }
1533 def VLD4DUPq8_UPD : VLD4DUPWB<{0,0,1,0}, "8">;
1534 def VLD4DUPq16_UPD : VLD4DUPWB<{0,1,1,?}, "16">;
1535 def VLD4DUPq32_UPD : VLD4DUPWB<{1,?,1,?}, "32"> { let Inst{6} = Rn{5}; }
1537 def VLD4DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1538 def VLD4DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1539 def VLD4DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1541 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
1543 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
1545 // Classes for VST* pseudo-instructions with multi-register operands.
1546 // These are expanded to real instructions after register allocation.
1547 class VSTQPseudo<InstrItinClass itin>
1548 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src), itin, "">;
1549 class VSTQWBPseudo<InstrItinClass itin>
1550 : PseudoNLdSt<(outs GPR:$wb),
1551 (ins addrmode6:$addr, am6offset:$offset, QPR:$src), itin,
1552 "$addr.addr = $wb">;
1553 class VSTQWBfixedPseudo<InstrItinClass itin>
1554 : PseudoNLdSt<(outs GPR:$wb),
1555 (ins addrmode6:$addr, QPR:$src), itin,
1556 "$addr.addr = $wb">;
1557 class VSTQWBregisterPseudo<InstrItinClass itin>
1558 : PseudoNLdSt<(outs GPR:$wb),
1559 (ins addrmode6:$addr, rGPR:$offset, QPR:$src), itin,
1560 "$addr.addr = $wb">;
1561 class VSTQQPseudo<InstrItinClass itin>
1562 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src), itin, "">;
1563 class VSTQQWBPseudo<InstrItinClass itin>
1564 : PseudoNLdSt<(outs GPR:$wb),
1565 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src), itin,
1566 "$addr.addr = $wb">;
1567 class VSTQQWBfixedPseudo<InstrItinClass itin>
1568 : PseudoNLdSt<(outs GPR:$wb),
1569 (ins addrmode6:$addr, QQPR:$src), itin,
1570 "$addr.addr = $wb">;
1571 class VSTQQWBregisterPseudo<InstrItinClass itin>
1572 : PseudoNLdSt<(outs GPR:$wb),
1573 (ins addrmode6:$addr, rGPR:$offset, QQPR:$src), itin,
1574 "$addr.addr = $wb">;
1576 class VSTQQQQPseudo<InstrItinClass itin>
1577 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src), itin, "">;
1578 class VSTQQQQWBPseudo<InstrItinClass itin>
1579 : PseudoNLdSt<(outs GPR:$wb),
1580 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
1581 "$addr.addr = $wb">;
1583 // VST1 : Vector Store (multiple single elements)
1584 class VST1D<bits<4> op7_4, string Dt>
1585 : NLdSt<0,0b00,0b0111,op7_4, (outs), (ins addrmode6:$Rn, VecListOneD:$Vd),
1586 IIC_VST1, "vst1", Dt, "$Vd, $Rn", "", []> {
1588 let Inst{4} = Rn{4};
1589 let DecoderMethod = "DecodeVSTInstruction";
1591 class VST1Q<bits<4> op7_4, string Dt>
1592 : NLdSt<0,0b00,0b1010,op7_4, (outs), (ins addrmode6:$Rn, VecListDPair:$Vd),
1593 IIC_VST1x2, "vst1", Dt, "$Vd, $Rn", "", []> {
1595 let Inst{5-4} = Rn{5-4};
1596 let DecoderMethod = "DecodeVSTInstruction";
1599 def VST1d8 : VST1D<{0,0,0,?}, "8">;
1600 def VST1d16 : VST1D<{0,1,0,?}, "16">;
1601 def VST1d32 : VST1D<{1,0,0,?}, "32">;
1602 def VST1d64 : VST1D<{1,1,0,?}, "64">;
1604 def VST1q8 : VST1Q<{0,0,?,?}, "8">;
1605 def VST1q16 : VST1Q<{0,1,?,?}, "16">;
1606 def VST1q32 : VST1Q<{1,0,?,?}, "32">;
1607 def VST1q64 : VST1Q<{1,1,?,?}, "64">;
1609 // ...with address register writeback:
1610 multiclass VST1DWB<bits<4> op7_4, string Dt> {
1611 def _fixed : NLdSt<0,0b00, 0b0111,op7_4, (outs GPR:$wb),
1612 (ins addrmode6:$Rn, VecListOneD:$Vd), IIC_VLD1u,
1613 "vst1", Dt, "$Vd, $Rn!",
1614 "$Rn.addr = $wb", []> {
1615 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1616 let Inst{4} = Rn{4};
1617 let DecoderMethod = "DecodeVSTInstruction";
1618 let AsmMatchConverter = "cvtVSTwbFixed";
1620 def _register : NLdSt<0,0b00,0b0111,op7_4, (outs GPR:$wb),
1621 (ins addrmode6:$Rn, rGPR:$Rm, VecListOneD:$Vd),
1623 "vst1", Dt, "$Vd, $Rn, $Rm",
1624 "$Rn.addr = $wb", []> {
1625 let Inst{4} = Rn{4};
1626 let DecoderMethod = "DecodeVSTInstruction";
1627 let AsmMatchConverter = "cvtVSTwbRegister";
1630 multiclass VST1QWB<bits<4> op7_4, string Dt> {
1631 def _fixed : NLdSt<0,0b00,0b1010,op7_4, (outs GPR:$wb),
1632 (ins addrmode6:$Rn, VecListDPair:$Vd), IIC_VLD1x2u,
1633 "vst1", Dt, "$Vd, $Rn!",
1634 "$Rn.addr = $wb", []> {
1635 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1636 let Inst{5-4} = Rn{5-4};
1637 let DecoderMethod = "DecodeVSTInstruction";
1638 let AsmMatchConverter = "cvtVSTwbFixed";
1640 def _register : NLdSt<0,0b00,0b1010,op7_4, (outs GPR:$wb),
1641 (ins addrmode6:$Rn, rGPR:$Rm, VecListDPair:$Vd),
1643 "vst1", Dt, "$Vd, $Rn, $Rm",
1644 "$Rn.addr = $wb", []> {
1645 let Inst{5-4} = Rn{5-4};
1646 let DecoderMethod = "DecodeVSTInstruction";
1647 let AsmMatchConverter = "cvtVSTwbRegister";
1651 defm VST1d8wb : VST1DWB<{0,0,0,?}, "8">;
1652 defm VST1d16wb : VST1DWB<{0,1,0,?}, "16">;
1653 defm VST1d32wb : VST1DWB<{1,0,0,?}, "32">;
1654 defm VST1d64wb : VST1DWB<{1,1,0,?}, "64">;
1656 defm VST1q8wb : VST1QWB<{0,0,?,?}, "8">;
1657 defm VST1q16wb : VST1QWB<{0,1,?,?}, "16">;
1658 defm VST1q32wb : VST1QWB<{1,0,?,?}, "32">;
1659 defm VST1q64wb : VST1QWB<{1,1,?,?}, "64">;
1661 // ...with 3 registers
1662 class VST1D3<bits<4> op7_4, string Dt>
1663 : NLdSt<0, 0b00, 0b0110, op7_4, (outs),
1664 (ins addrmode6:$Rn, VecListThreeD:$Vd),
1665 IIC_VST1x3, "vst1", Dt, "$Vd, $Rn", "", []> {
1667 let Inst{4} = Rn{4};
1668 let DecoderMethod = "DecodeVSTInstruction";
1670 multiclass VST1D3WB<bits<4> op7_4, string Dt> {
1671 def _fixed : NLdSt<0,0b00,0b0110,op7_4, (outs GPR:$wb),
1672 (ins addrmode6:$Rn, VecListThreeD:$Vd), IIC_VLD1x3u,
1673 "vst1", Dt, "$Vd, $Rn!",
1674 "$Rn.addr = $wb", []> {
1675 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1676 let Inst{5-4} = Rn{5-4};
1677 let DecoderMethod = "DecodeVSTInstruction";
1678 let AsmMatchConverter = "cvtVSTwbFixed";
1680 def _register : NLdSt<0,0b00,0b0110,op7_4, (outs GPR:$wb),
1681 (ins addrmode6:$Rn, rGPR:$Rm, VecListThreeD:$Vd),
1683 "vst1", Dt, "$Vd, $Rn, $Rm",
1684 "$Rn.addr = $wb", []> {
1685 let Inst{5-4} = Rn{5-4};
1686 let DecoderMethod = "DecodeVSTInstruction";
1687 let AsmMatchConverter = "cvtVSTwbRegister";
1691 def VST1d8T : VST1D3<{0,0,0,?}, "8">;
1692 def VST1d16T : VST1D3<{0,1,0,?}, "16">;
1693 def VST1d32T : VST1D3<{1,0,0,?}, "32">;
1694 def VST1d64T : VST1D3<{1,1,0,?}, "64">;
1696 defm VST1d8Twb : VST1D3WB<{0,0,0,?}, "8">;
1697 defm VST1d16Twb : VST1D3WB<{0,1,0,?}, "16">;
1698 defm VST1d32Twb : VST1D3WB<{1,0,0,?}, "32">;
1699 defm VST1d64Twb : VST1D3WB<{1,1,0,?}, "64">;
1701 def VST1d64TPseudo : VSTQQPseudo<IIC_VST1x3>;
1702 def VST1d64TPseudoWB_fixed : VSTQQWBPseudo<IIC_VST1x3u>;
1703 def VST1d64TPseudoWB_register : VSTQQWBPseudo<IIC_VST1x3u>;
1705 // ...with 4 registers
1706 class VST1D4<bits<4> op7_4, string Dt>
1707 : NLdSt<0, 0b00, 0b0010, op7_4, (outs),
1708 (ins addrmode6:$Rn, VecListFourD:$Vd),
1709 IIC_VST1x4, "vst1", Dt, "$Vd, $Rn", "",
1712 let Inst{5-4} = Rn{5-4};
1713 let DecoderMethod = "DecodeVSTInstruction";
1715 multiclass VST1D4WB<bits<4> op7_4, string Dt> {
1716 def _fixed : NLdSt<0,0b00,0b0010,op7_4, (outs GPR:$wb),
1717 (ins addrmode6:$Rn, VecListFourD:$Vd), IIC_VLD1x4u,
1718 "vst1", Dt, "$Vd, $Rn!",
1719 "$Rn.addr = $wb", []> {
1720 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1721 let Inst{5-4} = Rn{5-4};
1722 let DecoderMethod = "DecodeVSTInstruction";
1723 let AsmMatchConverter = "cvtVSTwbFixed";
1725 def _register : NLdSt<0,0b00,0b0010,op7_4, (outs GPR:$wb),
1726 (ins addrmode6:$Rn, rGPR:$Rm, VecListFourD:$Vd),
1728 "vst1", Dt, "$Vd, $Rn, $Rm",
1729 "$Rn.addr = $wb", []> {
1730 let Inst{5-4} = Rn{5-4};
1731 let DecoderMethod = "DecodeVSTInstruction";
1732 let AsmMatchConverter = "cvtVSTwbRegister";
1736 def VST1d8Q : VST1D4<{0,0,?,?}, "8">;
1737 def VST1d16Q : VST1D4<{0,1,?,?}, "16">;
1738 def VST1d32Q : VST1D4<{1,0,?,?}, "32">;
1739 def VST1d64Q : VST1D4<{1,1,?,?}, "64">;
1741 defm VST1d8Qwb : VST1D4WB<{0,0,?,?}, "8">;
1742 defm VST1d16Qwb : VST1D4WB<{0,1,?,?}, "16">;
1743 defm VST1d32Qwb : VST1D4WB<{1,0,?,?}, "32">;
1744 defm VST1d64Qwb : VST1D4WB<{1,1,?,?}, "64">;
1746 def VST1d64QPseudo : VSTQQPseudo<IIC_VST1x4>;
1747 def VST1d64QPseudoWB_fixed : VSTQQWBPseudo<IIC_VST1x4u>;
1748 def VST1d64QPseudoWB_register : VSTQQWBPseudo<IIC_VST1x4u>;
1750 // VST2 : Vector Store (multiple 2-element structures)
1751 class VST2<bits<4> op11_8, bits<4> op7_4, string Dt, RegisterOperand VdTy,
1752 InstrItinClass itin>
1753 : NLdSt<0, 0b00, op11_8, op7_4, (outs), (ins addrmode6:$Rn, VdTy:$Vd),
1754 itin, "vst2", Dt, "$Vd, $Rn", "", []> {
1756 let Inst{5-4} = Rn{5-4};
1757 let DecoderMethod = "DecodeVSTInstruction";
1760 def VST2d8 : VST2<0b1000, {0,0,?,?}, "8", VecListDPair, IIC_VST2>;
1761 def VST2d16 : VST2<0b1000, {0,1,?,?}, "16", VecListDPair, IIC_VST2>;
1762 def VST2d32 : VST2<0b1000, {1,0,?,?}, "32", VecListDPair, IIC_VST2>;
1764 def VST2q8 : VST2<0b0011, {0,0,?,?}, "8", VecListFourD, IIC_VST2x2>;
1765 def VST2q16 : VST2<0b0011, {0,1,?,?}, "16", VecListFourD, IIC_VST2x2>;
1766 def VST2q32 : VST2<0b0011, {1,0,?,?}, "32", VecListFourD, IIC_VST2x2>;
1768 def VST2q8Pseudo : VSTQQPseudo<IIC_VST2x2>;
1769 def VST2q16Pseudo : VSTQQPseudo<IIC_VST2x2>;
1770 def VST2q32Pseudo : VSTQQPseudo<IIC_VST2x2>;
1772 // ...with address register writeback:
1773 multiclass VST2DWB<bits<4> op11_8, bits<4> op7_4, string Dt,
1774 RegisterOperand VdTy> {
1775 def _fixed : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1776 (ins addrmode6:$Rn, VdTy:$Vd), IIC_VLD1u,
1777 "vst2", Dt, "$Vd, $Rn!",
1778 "$Rn.addr = $wb", []> {
1779 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1780 let Inst{5-4} = Rn{5-4};
1781 let DecoderMethod = "DecodeVSTInstruction";
1782 let AsmMatchConverter = "cvtVSTwbFixed";
1784 def _register : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1785 (ins addrmode6:$Rn, rGPR:$Rm, VdTy:$Vd), IIC_VLD1u,
1786 "vst2", Dt, "$Vd, $Rn, $Rm",
1787 "$Rn.addr = $wb", []> {
1788 let Inst{5-4} = Rn{5-4};
1789 let DecoderMethod = "DecodeVSTInstruction";
1790 let AsmMatchConverter = "cvtVSTwbRegister";
1793 multiclass VST2QWB<bits<4> op7_4, string Dt> {
1794 def _fixed : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
1795 (ins addrmode6:$Rn, VecListFourD:$Vd), IIC_VLD1u,
1796 "vst2", Dt, "$Vd, $Rn!",
1797 "$Rn.addr = $wb", []> {
1798 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1799 let Inst{5-4} = Rn{5-4};
1800 let DecoderMethod = "DecodeVSTInstruction";
1801 let AsmMatchConverter = "cvtVSTwbFixed";
1803 def _register : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
1804 (ins addrmode6:$Rn, rGPR:$Rm, VecListFourD:$Vd),
1806 "vst2", Dt, "$Vd, $Rn, $Rm",
1807 "$Rn.addr = $wb", []> {
1808 let Inst{5-4} = Rn{5-4};
1809 let DecoderMethod = "DecodeVSTInstruction";
1810 let AsmMatchConverter = "cvtVSTwbRegister";
1814 defm VST2d8wb : VST2DWB<0b1000, {0,0,?,?}, "8", VecListDPair>;
1815 defm VST2d16wb : VST2DWB<0b1000, {0,1,?,?}, "16", VecListDPair>;
1816 defm VST2d32wb : VST2DWB<0b1000, {1,0,?,?}, "32", VecListDPair>;
1818 defm VST2q8wb : VST2QWB<{0,0,?,?}, "8">;
1819 defm VST2q16wb : VST2QWB<{0,1,?,?}, "16">;
1820 defm VST2q32wb : VST2QWB<{1,0,?,?}, "32">;
1822 def VST2q8PseudoWB_fixed : VSTQQWBfixedPseudo<IIC_VST2x2u>;
1823 def VST2q16PseudoWB_fixed : VSTQQWBfixedPseudo<IIC_VST2x2u>;
1824 def VST2q32PseudoWB_fixed : VSTQQWBfixedPseudo<IIC_VST2x2u>;
1825 def VST2q8PseudoWB_register : VSTQQWBregisterPseudo<IIC_VST2x2u>;
1826 def VST2q16PseudoWB_register : VSTQQWBregisterPseudo<IIC_VST2x2u>;
1827 def VST2q32PseudoWB_register : VSTQQWBregisterPseudo<IIC_VST2x2u>;
1829 // ...with double-spaced registers
1830 def VST2b8 : VST2<0b1001, {0,0,?,?}, "8", VecListDPairSpaced, IIC_VST2>;
1831 def VST2b16 : VST2<0b1001, {0,1,?,?}, "16", VecListDPairSpaced, IIC_VST2>;
1832 def VST2b32 : VST2<0b1001, {1,0,?,?}, "32", VecListDPairSpaced, IIC_VST2>;
1833 defm VST2b8wb : VST2DWB<0b1001, {0,0,?,?}, "8", VecListDPairSpaced>;
1834 defm VST2b16wb : VST2DWB<0b1001, {0,1,?,?}, "16", VecListDPairSpaced>;
1835 defm VST2b32wb : VST2DWB<0b1001, {1,0,?,?}, "32", VecListDPairSpaced>;
1837 // VST3 : Vector Store (multiple 3-element structures)
1838 class VST3D<bits<4> op11_8, bits<4> op7_4, string Dt>
1839 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1840 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3,
1841 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn", "", []> {
1843 let Inst{4} = Rn{4};
1844 let DecoderMethod = "DecodeVSTInstruction";
1847 def VST3d8 : VST3D<0b0100, {0,0,0,?}, "8">;
1848 def VST3d16 : VST3D<0b0100, {0,1,0,?}, "16">;
1849 def VST3d32 : VST3D<0b0100, {1,0,0,?}, "32">;
1851 def VST3d8Pseudo : VSTQQPseudo<IIC_VST3>;
1852 def VST3d16Pseudo : VSTQQPseudo<IIC_VST3>;
1853 def VST3d32Pseudo : VSTQQPseudo<IIC_VST3>;
1855 // ...with address register writeback:
1856 class VST3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1857 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1858 (ins addrmode6:$Rn, am6offset:$Rm,
1859 DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3u,
1860 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn$Rm",
1861 "$Rn.addr = $wb", []> {
1862 let Inst{4} = Rn{4};
1863 let DecoderMethod = "DecodeVSTInstruction";
1866 def VST3d8_UPD : VST3DWB<0b0100, {0,0,0,?}, "8">;
1867 def VST3d16_UPD : VST3DWB<0b0100, {0,1,0,?}, "16">;
1868 def VST3d32_UPD : VST3DWB<0b0100, {1,0,0,?}, "32">;
1870 def VST3d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1871 def VST3d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1872 def VST3d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1874 // ...with double-spaced registers:
1875 def VST3q8 : VST3D<0b0101, {0,0,0,?}, "8">;
1876 def VST3q16 : VST3D<0b0101, {0,1,0,?}, "16">;
1877 def VST3q32 : VST3D<0b0101, {1,0,0,?}, "32">;
1878 def VST3q8_UPD : VST3DWB<0b0101, {0,0,0,?}, "8">;
1879 def VST3q16_UPD : VST3DWB<0b0101, {0,1,0,?}, "16">;
1880 def VST3q32_UPD : VST3DWB<0b0101, {1,0,0,?}, "32">;
1882 def VST3q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1883 def VST3q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1884 def VST3q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1886 // ...alternate versions to be allocated odd register numbers:
1887 def VST3q8oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1888 def VST3q16oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1889 def VST3q32oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1891 def VST3q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1892 def VST3q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1893 def VST3q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1895 // VST4 : Vector Store (multiple 4-element structures)
1896 class VST4D<bits<4> op11_8, bits<4> op7_4, string Dt>
1897 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1898 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1899 IIC_VST4, "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn",
1902 let Inst{5-4} = Rn{5-4};
1903 let DecoderMethod = "DecodeVSTInstruction";
1906 def VST4d8 : VST4D<0b0000, {0,0,?,?}, "8">;
1907 def VST4d16 : VST4D<0b0000, {0,1,?,?}, "16">;
1908 def VST4d32 : VST4D<0b0000, {1,0,?,?}, "32">;
1910 def VST4d8Pseudo : VSTQQPseudo<IIC_VST4>;
1911 def VST4d16Pseudo : VSTQQPseudo<IIC_VST4>;
1912 def VST4d32Pseudo : VSTQQPseudo<IIC_VST4>;
1914 // ...with address register writeback:
1915 class VST4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1916 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1917 (ins addrmode6:$Rn, am6offset:$Rm,
1918 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST4u,
1919 "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1920 "$Rn.addr = $wb", []> {
1921 let Inst{5-4} = Rn{5-4};
1922 let DecoderMethod = "DecodeVSTInstruction";
1925 def VST4d8_UPD : VST4DWB<0b0000, {0,0,?,?}, "8">;
1926 def VST4d16_UPD : VST4DWB<0b0000, {0,1,?,?}, "16">;
1927 def VST4d32_UPD : VST4DWB<0b0000, {1,0,?,?}, "32">;
1929 def VST4d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1930 def VST4d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1931 def VST4d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1933 // ...with double-spaced registers:
1934 def VST4q8 : VST4D<0b0001, {0,0,?,?}, "8">;
1935 def VST4q16 : VST4D<0b0001, {0,1,?,?}, "16">;
1936 def VST4q32 : VST4D<0b0001, {1,0,?,?}, "32">;
1937 def VST4q8_UPD : VST4DWB<0b0001, {0,0,?,?}, "8">;
1938 def VST4q16_UPD : VST4DWB<0b0001, {0,1,?,?}, "16">;
1939 def VST4q32_UPD : VST4DWB<0b0001, {1,0,?,?}, "32">;
1941 def VST4q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1942 def VST4q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1943 def VST4q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1945 // ...alternate versions to be allocated odd register numbers:
1946 def VST4q8oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1947 def VST4q16oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1948 def VST4q32oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1950 def VST4q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1951 def VST4q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1952 def VST4q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1954 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
1956 // Classes for VST*LN pseudo-instructions with multi-register operands.
1957 // These are expanded to real instructions after register allocation.
1958 class VSTQLNPseudo<InstrItinClass itin>
1959 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
1961 class VSTQLNWBPseudo<InstrItinClass itin>
1962 : PseudoNLdSt<(outs GPR:$wb),
1963 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
1964 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1965 class VSTQQLNPseudo<InstrItinClass itin>
1966 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
1968 class VSTQQLNWBPseudo<InstrItinClass itin>
1969 : PseudoNLdSt<(outs GPR:$wb),
1970 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
1971 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1972 class VSTQQQQLNPseudo<InstrItinClass itin>
1973 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
1975 class VSTQQQQLNWBPseudo<InstrItinClass itin>
1976 : PseudoNLdSt<(outs GPR:$wb),
1977 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
1978 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1980 // VST1LN : Vector Store (single element from one lane)
1981 class VST1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1982 PatFrag StoreOp, SDNode ExtractOp, Operand AddrMode>
1983 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1984 (ins AddrMode:$Rn, DPR:$Vd, nohash_imm:$lane),
1985 IIC_VST1ln, "vst1", Dt, "\\{$Vd[$lane]\\}, $Rn", "",
1986 [(StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane), AddrMode:$Rn)]> {
1988 let DecoderMethod = "DecodeVST1LN";
1990 class VST1QLNPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
1991 : VSTQLNPseudo<IIC_VST1ln> {
1992 let Pattern = [(StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
1996 def VST1LNd8 : VST1LN<0b0000, {?,?,?,0}, "8", v8i8, truncstorei8,
1997 NEONvgetlaneu, addrmode6> {
1998 let Inst{7-5} = lane{2-0};
2000 def VST1LNd16 : VST1LN<0b0100, {?,?,0,?}, "16", v4i16, truncstorei16,
2001 NEONvgetlaneu, addrmode6> {
2002 let Inst{7-6} = lane{1-0};
2003 let Inst{4} = Rn{4};
2006 def VST1LNd32 : VST1LN<0b1000, {?,0,?,?}, "32", v2i32, store, extractelt,
2008 let Inst{7} = lane{0};
2009 let Inst{5-4} = Rn{5-4};
2012 def VST1LNq8Pseudo : VST1QLNPseudo<v16i8, truncstorei8, NEONvgetlaneu>;
2013 def VST1LNq16Pseudo : VST1QLNPseudo<v8i16, truncstorei16, NEONvgetlaneu>;
2014 def VST1LNq32Pseudo : VST1QLNPseudo<v4i32, store, extractelt>;
2016 def : Pat<(store (extractelt (v2f32 DPR:$src), imm:$lane), addrmode6:$addr),
2017 (VST1LNd32 addrmode6:$addr, DPR:$src, imm:$lane)>;
2018 def : Pat<(store (extractelt (v4f32 QPR:$src), imm:$lane), addrmode6:$addr),
2019 (VST1LNq32Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
2021 // ...with address register writeback:
2022 class VST1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
2023 PatFrag StoreOp, SDNode ExtractOp, Operand AdrMode>
2024 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
2025 (ins AdrMode:$Rn, am6offset:$Rm,
2026 DPR:$Vd, nohash_imm:$lane), IIC_VST1lnu, "vst1", Dt,
2027 "\\{$Vd[$lane]\\}, $Rn$Rm",
2029 [(set GPR:$wb, (StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane),
2030 AdrMode:$Rn, am6offset:$Rm))]> {
2031 let DecoderMethod = "DecodeVST1LN";
2033 class VST1QLNWBPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
2034 : VSTQLNWBPseudo<IIC_VST1lnu> {
2035 let Pattern = [(set GPR:$wb, (StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
2036 addrmode6:$addr, am6offset:$offset))];
2039 def VST1LNd8_UPD : VST1LNWB<0b0000, {?,?,?,0}, "8", v8i8, post_truncsti8,
2040 NEONvgetlaneu, addrmode6> {
2041 let Inst{7-5} = lane{2-0};
2043 def VST1LNd16_UPD : VST1LNWB<0b0100, {?,?,0,?}, "16", v4i16, post_truncsti16,
2044 NEONvgetlaneu, addrmode6> {
2045 let Inst{7-6} = lane{1-0};
2046 let Inst{4} = Rn{4};
2048 def VST1LNd32_UPD : VST1LNWB<0b1000, {?,0,?,?}, "32", v2i32, post_store,
2049 extractelt, addrmode6oneL32> {
2050 let Inst{7} = lane{0};
2051 let Inst{5-4} = Rn{5-4};
2054 def VST1LNq8Pseudo_UPD : VST1QLNWBPseudo<v16i8, post_truncsti8, NEONvgetlaneu>;
2055 def VST1LNq16Pseudo_UPD : VST1QLNWBPseudo<v8i16, post_truncsti16,NEONvgetlaneu>;
2056 def VST1LNq32Pseudo_UPD : VST1QLNWBPseudo<v4i32, post_store, extractelt>;
2058 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
2060 // VST2LN : Vector Store (single 2-element structure from one lane)
2061 class VST2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
2062 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
2063 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, nohash_imm:$lane),
2064 IIC_VST2ln, "vst2", Dt, "\\{$Vd[$lane], $src2[$lane]\\}, $Rn",
2067 let Inst{4} = Rn{4};
2068 let DecoderMethod = "DecodeVST2LN";
2071 def VST2LNd8 : VST2LN<0b0001, {?,?,?,?}, "8"> {
2072 let Inst{7-5} = lane{2-0};
2074 def VST2LNd16 : VST2LN<0b0101, {?,?,0,?}, "16"> {
2075 let Inst{7-6} = lane{1-0};
2077 def VST2LNd32 : VST2LN<0b1001, {?,0,0,?}, "32"> {
2078 let Inst{7} = lane{0};
2081 def VST2LNd8Pseudo : VSTQLNPseudo<IIC_VST2ln>;
2082 def VST2LNd16Pseudo : VSTQLNPseudo<IIC_VST2ln>;
2083 def VST2LNd32Pseudo : VSTQLNPseudo<IIC_VST2ln>;
2085 // ...with double-spaced registers:
2086 def VST2LNq16 : VST2LN<0b0101, {?,?,1,?}, "16"> {
2087 let Inst{7-6} = lane{1-0};
2088 let Inst{4} = Rn{4};
2090 def VST2LNq32 : VST2LN<0b1001, {?,1,0,?}, "32"> {
2091 let Inst{7} = lane{0};
2092 let Inst{4} = Rn{4};
2095 def VST2LNq16Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
2096 def VST2LNq32Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
2098 // ...with address register writeback:
2099 class VST2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
2100 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
2101 (ins addrmode6:$Rn, am6offset:$Rm,
2102 DPR:$Vd, DPR:$src2, nohash_imm:$lane), IIC_VST2lnu, "vst2", Dt,
2103 "\\{$Vd[$lane], $src2[$lane]\\}, $Rn$Rm",
2104 "$Rn.addr = $wb", []> {
2105 let Inst{4} = Rn{4};
2106 let DecoderMethod = "DecodeVST2LN";
2109 def VST2LNd8_UPD : VST2LNWB<0b0001, {?,?,?,?}, "8"> {
2110 let Inst{7-5} = lane{2-0};
2112 def VST2LNd16_UPD : VST2LNWB<0b0101, {?,?,0,?}, "16"> {
2113 let Inst{7-6} = lane{1-0};
2115 def VST2LNd32_UPD : VST2LNWB<0b1001, {?,0,0,?}, "32"> {
2116 let Inst{7} = lane{0};
2119 def VST2LNd8Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
2120 def VST2LNd16Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
2121 def VST2LNd32Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
2123 def VST2LNq16_UPD : VST2LNWB<0b0101, {?,?,1,?}, "16"> {
2124 let Inst{7-6} = lane{1-0};
2126 def VST2LNq32_UPD : VST2LNWB<0b1001, {?,1,0,?}, "32"> {
2127 let Inst{7} = lane{0};
2130 def VST2LNq16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
2131 def VST2LNq32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
2133 // VST3LN : Vector Store (single 3-element structure from one lane)
2134 class VST3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
2135 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
2136 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3,
2137 nohash_imm:$lane), IIC_VST3ln, "vst3", Dt,
2138 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn", "", []> {
2140 let DecoderMethod = "DecodeVST3LN";
2143 def VST3LNd8 : VST3LN<0b0010, {?,?,?,0}, "8"> {
2144 let Inst{7-5} = lane{2-0};
2146 def VST3LNd16 : VST3LN<0b0110, {?,?,0,0}, "16"> {
2147 let Inst{7-6} = lane{1-0};
2149 def VST3LNd32 : VST3LN<0b1010, {?,0,0,0}, "32"> {
2150 let Inst{7} = lane{0};
2153 def VST3LNd8Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
2154 def VST3LNd16Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
2155 def VST3LNd32Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
2157 // ...with double-spaced registers:
2158 def VST3LNq16 : VST3LN<0b0110, {?,?,1,0}, "16"> {
2159 let Inst{7-6} = lane{1-0};
2161 def VST3LNq32 : VST3LN<0b1010, {?,1,0,0}, "32"> {
2162 let Inst{7} = lane{0};
2165 def VST3LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
2166 def VST3LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
2168 // ...with address register writeback:
2169 class VST3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
2170 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
2171 (ins addrmode6:$Rn, am6offset:$Rm,
2172 DPR:$Vd, DPR:$src2, DPR:$src3, nohash_imm:$lane),
2173 IIC_VST3lnu, "vst3", Dt,
2174 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn$Rm",
2175 "$Rn.addr = $wb", []> {
2176 let DecoderMethod = "DecodeVST3LN";
2179 def VST3LNd8_UPD : VST3LNWB<0b0010, {?,?,?,0}, "8"> {
2180 let Inst{7-5} = lane{2-0};
2182 def VST3LNd16_UPD : VST3LNWB<0b0110, {?,?,0,0}, "16"> {
2183 let Inst{7-6} = lane{1-0};
2185 def VST3LNd32_UPD : VST3LNWB<0b1010, {?,0,0,0}, "32"> {
2186 let Inst{7} = lane{0};
2189 def VST3LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
2190 def VST3LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
2191 def VST3LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
2193 def VST3LNq16_UPD : VST3LNWB<0b0110, {?,?,1,0}, "16"> {
2194 let Inst{7-6} = lane{1-0};
2196 def VST3LNq32_UPD : VST3LNWB<0b1010, {?,1,0,0}, "32"> {
2197 let Inst{7} = lane{0};
2200 def VST3LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
2201 def VST3LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
2203 // VST4LN : Vector Store (single 4-element structure from one lane)
2204 class VST4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
2205 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
2206 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4,
2207 nohash_imm:$lane), IIC_VST4ln, "vst4", Dt,
2208 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn",
2211 let Inst{4} = Rn{4};
2212 let DecoderMethod = "DecodeVST4LN";
2215 def VST4LNd8 : VST4LN<0b0011, {?,?,?,?}, "8"> {
2216 let Inst{7-5} = lane{2-0};
2218 def VST4LNd16 : VST4LN<0b0111, {?,?,0,?}, "16"> {
2219 let Inst{7-6} = lane{1-0};
2221 def VST4LNd32 : VST4LN<0b1011, {?,0,?,?}, "32"> {
2222 let Inst{7} = lane{0};
2223 let Inst{5} = Rn{5};
2226 def VST4LNd8Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
2227 def VST4LNd16Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
2228 def VST4LNd32Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
2230 // ...with double-spaced registers:
2231 def VST4LNq16 : VST4LN<0b0111, {?,?,1,?}, "16"> {
2232 let Inst{7-6} = lane{1-0};
2234 def VST4LNq32 : VST4LN<0b1011, {?,1,?,?}, "32"> {
2235 let Inst{7} = lane{0};
2236 let Inst{5} = Rn{5};
2239 def VST4LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
2240 def VST4LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
2242 // ...with address register writeback:
2243 class VST4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
2244 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
2245 (ins addrmode6:$Rn, am6offset:$Rm,
2246 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
2247 IIC_VST4lnu, "vst4", Dt,
2248 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn$Rm",
2249 "$Rn.addr = $wb", []> {
2250 let Inst{4} = Rn{4};
2251 let DecoderMethod = "DecodeVST4LN";
2254 def VST4LNd8_UPD : VST4LNWB<0b0011, {?,?,?,?}, "8"> {
2255 let Inst{7-5} = lane{2-0};
2257 def VST4LNd16_UPD : VST4LNWB<0b0111, {?,?,0,?}, "16"> {
2258 let Inst{7-6} = lane{1-0};
2260 def VST4LNd32_UPD : VST4LNWB<0b1011, {?,0,?,?}, "32"> {
2261 let Inst{7} = lane{0};
2262 let Inst{5} = Rn{5};
2265 def VST4LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
2266 def VST4LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
2267 def VST4LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
2269 def VST4LNq16_UPD : VST4LNWB<0b0111, {?,?,1,?}, "16"> {
2270 let Inst{7-6} = lane{1-0};
2272 def VST4LNq32_UPD : VST4LNWB<0b1011, {?,1,?,?}, "32"> {
2273 let Inst{7} = lane{0};
2274 let Inst{5} = Rn{5};
2277 def VST4LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
2278 def VST4LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
2280 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
2282 // Use vld1/vst1 for unaligned f64 load / store
2283 def : Pat<(f64 (hword_alignedload addrmode6:$addr)),
2284 (VLD1d16 addrmode6:$addr)>, Requires<[IsLE]>;
2285 def : Pat<(hword_alignedstore (f64 DPR:$value), addrmode6:$addr),
2286 (VST1d16 addrmode6:$addr, DPR:$value)>, Requires<[IsLE]>;
2287 def : Pat<(f64 (byte_alignedload addrmode6:$addr)),
2288 (VLD1d8 addrmode6:$addr)>, Requires<[IsLE]>;
2289 def : Pat<(byte_alignedstore (f64 DPR:$value), addrmode6:$addr),
2290 (VST1d8 addrmode6:$addr, DPR:$value)>, Requires<[IsLE]>;
2291 def : Pat<(f64 (non_word_alignedload addrmode6:$addr)),
2292 (VLD1d64 addrmode6:$addr)>, Requires<[IsBE]>;
2293 def : Pat<(non_word_alignedstore (f64 DPR:$value), addrmode6:$addr),
2294 (VST1d64 addrmode6:$addr, DPR:$value)>, Requires<[IsBE]>;
2296 // Use vld1/vst1 for Q and QQ. Also use them for unaligned v2f64
2297 // load / store if it's legal.
2298 def : Pat<(v2f64 (dword_alignedload addrmode6:$addr)),
2299 (VLD1q64 addrmode6:$addr)>;
2300 def : Pat<(dword_alignedstore (v2f64 QPR:$value), addrmode6:$addr),
2301 (VST1q64 addrmode6:$addr, QPR:$value)>;
2302 def : Pat<(v2f64 (word_alignedload addrmode6:$addr)),
2303 (VLD1q32 addrmode6:$addr)>;
2304 def : Pat<(word_alignedstore (v2f64 QPR:$value), addrmode6:$addr),
2305 (VST1q32 addrmode6:$addr, QPR:$value)>;
2306 def : Pat<(v2f64 (hword_alignedload addrmode6:$addr)),
2307 (VLD1q16 addrmode6:$addr)>, Requires<[IsLE]>;
2308 def : Pat<(hword_alignedstore (v2f64 QPR:$value), addrmode6:$addr),
2309 (VST1q16 addrmode6:$addr, QPR:$value)>, Requires<[IsLE]>;
2310 def : Pat<(v2f64 (byte_alignedload addrmode6:$addr)),
2311 (VLD1q8 addrmode6:$addr)>, Requires<[IsLE]>;
2312 def : Pat<(byte_alignedstore (v2f64 QPR:$value), addrmode6:$addr),
2313 (VST1q8 addrmode6:$addr, QPR:$value)>, Requires<[IsLE]>;
2315 //===----------------------------------------------------------------------===//
2316 // NEON pattern fragments
2317 //===----------------------------------------------------------------------===//
2319 // Extract D sub-registers of Q registers.
2320 def DSubReg_i8_reg : SDNodeXForm<imm, [{
2321 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2322 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/8, MVT::i32);
2324 def DSubReg_i16_reg : SDNodeXForm<imm, [{
2325 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2326 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/4, MVT::i32);
2328 def DSubReg_i32_reg : SDNodeXForm<imm, [{
2329 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2330 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/2, MVT::i32);
2332 def DSubReg_f64_reg : SDNodeXForm<imm, [{
2333 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2334 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue(), MVT::i32);
2337 // Extract S sub-registers of Q/D registers.
2338 def SSubReg_f32_reg : SDNodeXForm<imm, [{
2339 assert(ARM::ssub_3 == ARM::ssub_0+3 && "Unexpected subreg numbering");
2340 return CurDAG->getTargetConstant(ARM::ssub_0 + N->getZExtValue(), MVT::i32);
2343 // Translate lane numbers from Q registers to D subregs.
2344 def SubReg_i8_lane : SDNodeXForm<imm, [{
2345 return CurDAG->getTargetConstant(N->getZExtValue() & 7, MVT::i32);
2347 def SubReg_i16_lane : SDNodeXForm<imm, [{
2348 return CurDAG->getTargetConstant(N->getZExtValue() & 3, MVT::i32);
2350 def SubReg_i32_lane : SDNodeXForm<imm, [{
2351 return CurDAG->getTargetConstant(N->getZExtValue() & 1, MVT::i32);
2354 //===----------------------------------------------------------------------===//
2355 // Instruction Classes
2356 //===----------------------------------------------------------------------===//
2358 // Basic 2-register operations: double- and quad-register.
2359 class N2VD<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, 0, op4, (outs DPR:$Vd),
2363 (ins DPR:$Vm), IIC_VUNAD, OpcodeStr, Dt,"$Vd, $Vm", "",
2364 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm))))]>;
2365 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2366 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
2367 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
2368 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2369 (ins QPR:$Vm), IIC_VUNAQ, OpcodeStr, Dt,"$Vd, $Vm", "",
2370 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vm))))]>;
2372 // Basic 2-register intrinsics, both double- and quad-register.
2373 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2374 bits<2> op17_16, bits<5> op11_7, bit op4,
2375 InstrItinClass itin, string OpcodeStr, string Dt,
2376 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2377 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2378 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2379 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
2380 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2381 bits<2> op17_16, bits<5> op11_7, bit op4,
2382 InstrItinClass itin, string OpcodeStr, string Dt,
2383 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2384 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2385 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2386 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
2388 // Narrow 2-register operations.
2389 class N2VN<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2390 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2391 InstrItinClass itin, string OpcodeStr, string Dt,
2392 ValueType TyD, ValueType TyQ, SDNode OpNode>
2393 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$Vd),
2394 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2395 [(set DPR:$Vd, (TyD (OpNode (TyQ QPR:$Vm))))]>;
2397 // Narrow 2-register intrinsics.
2398 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2399 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2400 InstrItinClass itin, string OpcodeStr, string Dt,
2401 ValueType TyD, ValueType TyQ, SDPatternOperator IntOp>
2402 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$Vd),
2403 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2404 [(set DPR:$Vd, (TyD (IntOp (TyQ QPR:$Vm))))]>;
2406 // Long 2-register operations (currently only used for VMOVL).
2407 class N2VL<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2408 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2409 InstrItinClass itin, string OpcodeStr, string Dt,
2410 ValueType TyQ, ValueType TyD, SDNode OpNode>
2411 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$Vd),
2412 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2413 [(set QPR:$Vd, (TyQ (OpNode (TyD DPR:$Vm))))]>;
2415 // Long 2-register intrinsics.
2416 class N2VLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2417 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2418 InstrItinClass itin, string OpcodeStr, string Dt,
2419 ValueType TyQ, ValueType TyD, SDPatternOperator IntOp>
2420 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$Vd),
2421 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2422 [(set QPR:$Vd, (TyQ (IntOp (TyD DPR:$Vm))))]>;
2424 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
2425 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
2426 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$Vd, DPR:$Vm),
2427 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
2428 OpcodeStr, Dt, "$Vd, $Vm",
2429 "$src1 = $Vd, $src2 = $Vm", []>;
2430 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
2431 InstrItinClass itin, string OpcodeStr, string Dt>
2432 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$Vd, QPR:$Vm),
2433 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$Vd, $Vm",
2434 "$src1 = $Vd, $src2 = $Vm", []>;
2436 // Basic 3-register operations: double- and quad-register.
2437 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2438 InstrItinClass itin, string OpcodeStr, string Dt,
2439 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2440 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2441 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2442 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2443 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
2444 // All of these have a two-operand InstAlias.
2445 let TwoOperandAliasConstraint = "$Vn = $Vd";
2446 let isCommutable = Commutable;
2448 // Same as N3VD but no data type.
2449 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2450 InstrItinClass itin, string OpcodeStr,
2451 ValueType ResTy, ValueType OpTy,
2452 SDNode OpNode, bit Commutable>
2453 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
2454 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2455 OpcodeStr, "$Vd, $Vn, $Vm", "",
2456 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>{
2457 // All of these have a two-operand InstAlias.
2458 let TwoOperandAliasConstraint = "$Vn = $Vd";
2459 let isCommutable = Commutable;
2462 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
2463 InstrItinClass itin, string OpcodeStr, string Dt,
2464 ValueType Ty, SDNode ShOp>
2465 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2466 (outs DPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2467 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2469 (Ty (ShOp (Ty DPR:$Vn),
2470 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),imm:$lane)))))]> {
2471 // All of these have a two-operand InstAlias.
2472 let TwoOperandAliasConstraint = "$Vn = $Vd";
2473 let isCommutable = 0;
2475 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
2476 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
2477 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2478 (outs DPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2479 NVMulSLFrm, IIC_VMULi16D, OpcodeStr, Dt,"$Vd, $Vn, $Vm$lane","",
2481 (Ty (ShOp (Ty DPR:$Vn),
2482 (Ty (NEONvduplane (Ty DPR_8:$Vm), imm:$lane)))))]> {
2483 // All of these have a two-operand InstAlias.
2484 let TwoOperandAliasConstraint = "$Vn = $Vd";
2485 let isCommutable = 0;
2488 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2489 InstrItinClass itin, string OpcodeStr, string Dt,
2490 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2491 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2492 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2493 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2494 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
2495 // All of these have a two-operand InstAlias.
2496 let TwoOperandAliasConstraint = "$Vn = $Vd";
2497 let isCommutable = Commutable;
2499 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2500 InstrItinClass itin, string OpcodeStr,
2501 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2502 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
2503 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2504 OpcodeStr, "$Vd, $Vn, $Vm", "",
2505 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>{
2506 // All of these have a two-operand InstAlias.
2507 let TwoOperandAliasConstraint = "$Vn = $Vd";
2508 let isCommutable = Commutable;
2510 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
2511 InstrItinClass itin, string OpcodeStr, string Dt,
2512 ValueType ResTy, ValueType OpTy, SDNode ShOp>
2513 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2514 (outs QPR:$Vd), (ins QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2515 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2516 [(set (ResTy QPR:$Vd),
2517 (ResTy (ShOp (ResTy QPR:$Vn),
2518 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2520 // All of these have a two-operand InstAlias.
2521 let TwoOperandAliasConstraint = "$Vn = $Vd";
2522 let isCommutable = 0;
2524 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
2525 ValueType ResTy, ValueType OpTy, SDNode ShOp>
2526 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2527 (outs QPR:$Vd), (ins QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2528 NVMulSLFrm, IIC_VMULi16Q, OpcodeStr, Dt,"$Vd, $Vn, $Vm$lane", "",
2529 [(set (ResTy QPR:$Vd),
2530 (ResTy (ShOp (ResTy QPR:$Vn),
2531 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2533 // All of these have a two-operand InstAlias.
2534 let TwoOperandAliasConstraint = "$Vn = $Vd";
2535 let isCommutable = 0;
2538 // Basic 3-register intrinsics, both double- and quad-register.
2539 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2540 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2541 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp, bit Commutable>
2542 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2543 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), f, itin,
2544 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2545 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
2546 // All of these have a two-operand InstAlias.
2547 let TwoOperandAliasConstraint = "$Vn = $Vd";
2548 let isCommutable = Commutable;
2550 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2551 string OpcodeStr, string Dt, ValueType Ty, SDPatternOperator IntOp>
2552 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2553 (outs DPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2554 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2556 (Ty (IntOp (Ty DPR:$Vn),
2557 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),
2559 let isCommutable = 0;
2561 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2562 string OpcodeStr, string Dt, ValueType Ty, SDPatternOperator IntOp>
2563 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2564 (outs DPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2565 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2567 (Ty (IntOp (Ty DPR:$Vn),
2568 (Ty (NEONvduplane (Ty DPR_8:$Vm), imm:$lane)))))]> {
2569 let isCommutable = 0;
2571 class N3VDIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2572 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2573 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2574 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2575 (outs DPR:$Vd), (ins DPR:$Vm, DPR:$Vn), f, itin,
2576 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
2577 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (OpTy DPR:$Vn))))]> {
2578 let TwoOperandAliasConstraint = "$Vm = $Vd";
2579 let isCommutable = 0;
2582 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2583 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2584 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp, bit Commutable>
2585 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2586 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), f, itin,
2587 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2588 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
2589 // All of these have a two-operand InstAlias.
2590 let TwoOperandAliasConstraint = "$Vn = $Vd";
2591 let isCommutable = Commutable;
2593 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2594 string OpcodeStr, string Dt,
2595 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2596 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2597 (outs QPR:$Vd), (ins QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2598 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2599 [(set (ResTy QPR:$Vd),
2600 (ResTy (IntOp (ResTy QPR:$Vn),
2601 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2603 let isCommutable = 0;
2605 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2606 string OpcodeStr, string Dt,
2607 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2608 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2609 (outs QPR:$Vd), (ins QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2610 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2611 [(set (ResTy QPR:$Vd),
2612 (ResTy (IntOp (ResTy QPR:$Vn),
2613 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2615 let isCommutable = 0;
2617 class N3VQIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2618 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2619 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2620 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2621 (outs QPR:$Vd), (ins QPR:$Vm, QPR:$Vn), f, itin,
2622 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
2623 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (OpTy QPR:$Vn))))]> {
2624 let TwoOperandAliasConstraint = "$Vm = $Vd";
2625 let isCommutable = 0;
2628 // Multiply-Add/Sub operations: double- and quad-register.
2629 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2630 InstrItinClass itin, string OpcodeStr, string Dt,
2631 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator OpNode>
2632 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2633 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2634 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2635 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
2636 (Ty (MulOp DPR:$Vn, DPR:$Vm)))))]>;
2638 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2639 string OpcodeStr, string Dt,
2640 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator ShOp>
2641 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2643 (ins DPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2645 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2647 (Ty (ShOp (Ty DPR:$src1),
2649 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),
2651 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2652 string OpcodeStr, string Dt,
2653 ValueType Ty, SDNode MulOp, SDNode ShOp>
2654 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2656 (ins DPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2658 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2660 (Ty (ShOp (Ty DPR:$src1),
2662 (Ty (NEONvduplane (Ty DPR_8:$Vm),
2665 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2666 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
2667 SDPatternOperator MulOp, SDPatternOperator OpNode>
2668 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2669 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2670 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2671 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
2672 (Ty (MulOp QPR:$Vn, QPR:$Vm)))))]>;
2673 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2674 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2675 SDPatternOperator MulOp, SDPatternOperator ShOp>
2676 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2678 (ins QPR:$src1, QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2680 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2681 [(set (ResTy QPR:$Vd),
2682 (ResTy (ShOp (ResTy QPR:$src1),
2683 (ResTy (MulOp QPR:$Vn,
2684 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2686 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2687 string OpcodeStr, string Dt,
2688 ValueType ResTy, ValueType OpTy,
2689 SDNode MulOp, SDNode ShOp>
2690 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2692 (ins QPR:$src1, QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2694 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2695 [(set (ResTy QPR:$Vd),
2696 (ResTy (ShOp (ResTy QPR:$src1),
2697 (ResTy (MulOp QPR:$Vn,
2698 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2701 // Neon Intrinsic-Op instructions (VABA): double- and quad-register.
2702 class N3VDIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2703 InstrItinClass itin, string OpcodeStr, string Dt,
2704 ValueType Ty, SDPatternOperator IntOp, SDNode OpNode>
2705 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2706 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2707 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2708 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
2709 (Ty (IntOp (Ty DPR:$Vn), (Ty DPR:$Vm))))))]>;
2710 class N3VQIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2711 InstrItinClass itin, string OpcodeStr, string Dt,
2712 ValueType Ty, SDPatternOperator IntOp, SDNode OpNode>
2713 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2714 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2715 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2716 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
2717 (Ty (IntOp (Ty QPR:$Vn), (Ty QPR:$Vm))))))]>;
2719 // Neon 3-argument intrinsics, both double- and quad-register.
2720 // The destination register is also used as the first source operand register.
2721 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2722 InstrItinClass itin, string OpcodeStr, string Dt,
2723 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2724 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2725 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2726 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2727 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$src1),
2728 (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>;
2729 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2730 InstrItinClass itin, string OpcodeStr, string Dt,
2731 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2732 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2733 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2734 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2735 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$src1),
2736 (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>;
2738 // Long Multiply-Add/Sub operations.
2739 class N3VLMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2740 InstrItinClass itin, string OpcodeStr, string Dt,
2741 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2742 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2743 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2744 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2745 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
2746 (TyQ (MulOp (TyD DPR:$Vn),
2747 (TyD DPR:$Vm)))))]>;
2748 class N3VLMulOpSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2749 InstrItinClass itin, string OpcodeStr, string Dt,
2750 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2751 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$Vd),
2752 (ins QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2754 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2756 (OpNode (TyQ QPR:$src1),
2757 (TyQ (MulOp (TyD DPR:$Vn),
2758 (TyD (NEONvduplane (TyD DPR_VFP2:$Vm),
2760 class N3VLMulOpSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2761 InstrItinClass itin, string OpcodeStr, string Dt,
2762 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2763 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$Vd),
2764 (ins QPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2766 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2768 (OpNode (TyQ QPR:$src1),
2769 (TyQ (MulOp (TyD DPR:$Vn),
2770 (TyD (NEONvduplane (TyD DPR_8:$Vm),
2773 // Long Intrinsic-Op vector operations with explicit extend (VABAL).
2774 class N3VLIntExtOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2775 InstrItinClass itin, string OpcodeStr, string Dt,
2776 ValueType TyQ, ValueType TyD, SDPatternOperator IntOp, SDNode ExtOp,
2778 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2779 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2780 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2781 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
2782 (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
2783 (TyD DPR:$Vm)))))))]>;
2785 // Neon Long 3-argument intrinsic. The destination register is
2786 // a quad-register and is also used as the first source operand register.
2787 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2788 InstrItinClass itin, string OpcodeStr, string Dt,
2789 ValueType TyQ, ValueType TyD, SDPatternOperator IntOp>
2790 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2791 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2792 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2794 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$Vn), (TyD DPR:$Vm))))]>;
2795 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2796 string OpcodeStr, string Dt,
2797 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2798 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2800 (ins QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2802 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2803 [(set (ResTy QPR:$Vd),
2804 (ResTy (IntOp (ResTy QPR:$src1),
2806 (OpTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2808 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2809 InstrItinClass itin, string OpcodeStr, string Dt,
2810 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2811 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2813 (ins QPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2815 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2816 [(set (ResTy QPR:$Vd),
2817 (ResTy (IntOp (ResTy QPR:$src1),
2819 (OpTy (NEONvduplane (OpTy DPR_8:$Vm),
2822 // Narrowing 3-register intrinsics.
2823 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2824 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
2825 SDPatternOperator IntOp, bit Commutable>
2826 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2827 (outs DPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINi4D,
2828 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2829 [(set DPR:$Vd, (TyD (IntOp (TyQ QPR:$Vn), (TyQ QPR:$Vm))))]> {
2830 let isCommutable = Commutable;
2833 // Long 3-register operations.
2834 class N3VL<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2835 InstrItinClass itin, string OpcodeStr, string Dt,
2836 ValueType TyQ, ValueType TyD, SDNode OpNode, bit Commutable>
2837 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2838 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2839 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2840 [(set QPR:$Vd, (TyQ (OpNode (TyD DPR:$Vn), (TyD DPR:$Vm))))]> {
2841 let isCommutable = Commutable;
2843 class N3VLSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2844 InstrItinClass itin, string OpcodeStr, string Dt,
2845 ValueType TyQ, ValueType TyD, SDNode OpNode>
2846 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2847 (outs QPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2848 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2850 (TyQ (OpNode (TyD DPR:$Vn),
2851 (TyD (NEONvduplane (TyD DPR_VFP2:$Vm),imm:$lane)))))]>;
2852 class N3VLSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2853 InstrItinClass itin, string OpcodeStr, string Dt,
2854 ValueType TyQ, ValueType TyD, SDNode OpNode>
2855 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2856 (outs QPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2857 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2859 (TyQ (OpNode (TyD DPR:$Vn),
2860 (TyD (NEONvduplane (TyD DPR_8:$Vm), imm:$lane)))))]>;
2862 // Long 3-register operations with explicitly extended operands.
2863 class N3VLExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2864 InstrItinClass itin, string OpcodeStr, string Dt,
2865 ValueType TyQ, ValueType TyD, SDNode OpNode, SDNode ExtOp,
2867 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2868 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2869 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2870 [(set QPR:$Vd, (OpNode (TyQ (ExtOp (TyD DPR:$Vn))),
2871 (TyQ (ExtOp (TyD DPR:$Vm)))))]> {
2872 let isCommutable = Commutable;
2875 // Long 3-register intrinsics with explicit extend (VABDL).
2876 class N3VLIntExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2877 InstrItinClass itin, string OpcodeStr, string Dt,
2878 ValueType TyQ, ValueType TyD, SDPatternOperator IntOp, SDNode ExtOp,
2880 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2881 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2882 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2883 [(set QPR:$Vd, (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
2884 (TyD DPR:$Vm))))))]> {
2885 let isCommutable = Commutable;
2888 // Long 3-register intrinsics.
2889 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2890 InstrItinClass itin, string OpcodeStr, string Dt,
2891 ValueType TyQ, ValueType TyD, SDPatternOperator IntOp, bit Commutable>
2892 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2893 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2894 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2895 [(set QPR:$Vd, (TyQ (IntOp (TyD DPR:$Vn), (TyD DPR:$Vm))))]> {
2896 let isCommutable = Commutable;
2898 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2899 string OpcodeStr, string Dt,
2900 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2901 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2902 (outs QPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2903 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2904 [(set (ResTy QPR:$Vd),
2905 (ResTy (IntOp (OpTy DPR:$Vn),
2906 (OpTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2908 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2909 InstrItinClass itin, string OpcodeStr, string Dt,
2910 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2911 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2912 (outs QPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2913 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2914 [(set (ResTy QPR:$Vd),
2915 (ResTy (IntOp (OpTy DPR:$Vn),
2916 (OpTy (NEONvduplane (OpTy DPR_8:$Vm),
2919 // Wide 3-register operations.
2920 class N3VW<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2921 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
2922 SDNode OpNode, SDNode ExtOp, bit Commutable>
2923 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2924 (outs QPR:$Vd), (ins QPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VSUBiD,
2925 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2926 [(set QPR:$Vd, (OpNode (TyQ QPR:$Vn),
2927 (TyQ (ExtOp (TyD DPR:$Vm)))))]> {
2928 // All of these have a two-operand InstAlias.
2929 let TwoOperandAliasConstraint = "$Vn = $Vd";
2930 let isCommutable = Commutable;
2933 // Pairwise long 2-register intrinsics, both double- and quad-register.
2934 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2935 bits<2> op17_16, bits<5> op11_7, bit op4,
2936 string OpcodeStr, string Dt,
2937 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2938 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2939 (ins DPR:$Vm), IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
2940 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
2941 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2942 bits<2> op17_16, bits<5> op11_7, bit op4,
2943 string OpcodeStr, string Dt,
2944 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2945 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2946 (ins QPR:$Vm), IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
2947 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
2949 // Pairwise long 2-register accumulate intrinsics,
2950 // both double- and quad-register.
2951 // The destination register is also used as the first source operand register.
2952 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2953 bits<2> op17_16, bits<5> op11_7, bit op4,
2954 string OpcodeStr, string Dt,
2955 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2956 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
2957 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vm), IIC_VPALiD,
2958 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2959 [(set DPR:$Vd, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$Vm))))]>;
2960 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2961 bits<2> op17_16, bits<5> op11_7, bit op4,
2962 string OpcodeStr, string Dt,
2963 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2964 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
2965 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vm), IIC_VPALiQ,
2966 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2967 [(set QPR:$Vd, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$Vm))))]>;
2969 // Shift by immediate,
2970 // both double- and quad-register.
2971 let TwoOperandAliasConstraint = "$Vm = $Vd" in {
2972 class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2973 Format f, InstrItinClass itin, Operand ImmTy,
2974 string OpcodeStr, string Dt, ValueType Ty, SDNode OpNode>
2975 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2976 (outs DPR:$Vd), (ins DPR:$Vm, ImmTy:$SIMM), f, itin,
2977 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2978 [(set DPR:$Vd, (Ty (OpNode (Ty DPR:$Vm), (i32 imm:$SIMM))))]>;
2979 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2980 Format f, InstrItinClass itin, Operand ImmTy,
2981 string OpcodeStr, string Dt, ValueType Ty, SDNode OpNode>
2982 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2983 (outs QPR:$Vd), (ins QPR:$Vm, ImmTy:$SIMM), f, itin,
2984 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2985 [(set QPR:$Vd, (Ty (OpNode (Ty QPR:$Vm), (i32 imm:$SIMM))))]>;
2988 // Long shift by immediate.
2989 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2990 string OpcodeStr, string Dt,
2991 ValueType ResTy, ValueType OpTy, Operand ImmTy, SDNode OpNode>
2992 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2993 (outs QPR:$Vd), (ins DPR:$Vm, ImmTy:$SIMM), N2RegVShLFrm,
2994 IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2995 [(set QPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm),
2996 (i32 imm:$SIMM))))]>;
2998 // Narrow shift by immediate.
2999 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
3000 InstrItinClass itin, string OpcodeStr, string Dt,
3001 ValueType ResTy, ValueType OpTy, Operand ImmTy, SDNode OpNode>
3002 : N2VImm<op24, op23, op11_8, op7, op6, op4,
3003 (outs DPR:$Vd), (ins QPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, itin,
3004 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
3005 [(set DPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vm),
3006 (i32 imm:$SIMM))))]>;
3008 // Shift right by immediate and accumulate,
3009 // both double- and quad-register.
3010 let TwoOperandAliasConstraint = "$Vm = $Vd" in {
3011 class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
3012 Operand ImmTy, string OpcodeStr, string Dt,
3013 ValueType Ty, SDNode ShOp>
3014 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
3015 (ins DPR:$src1, DPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, IIC_VPALiD,
3016 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
3017 [(set DPR:$Vd, (Ty (add DPR:$src1,
3018 (Ty (ShOp DPR:$Vm, (i32 imm:$SIMM))))))]>;
3019 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
3020 Operand ImmTy, string OpcodeStr, string Dt,
3021 ValueType Ty, SDNode ShOp>
3022 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
3023 (ins QPR:$src1, QPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, IIC_VPALiD,
3024 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
3025 [(set QPR:$Vd, (Ty (add QPR:$src1,
3026 (Ty (ShOp QPR:$Vm, (i32 imm:$SIMM))))))]>;
3029 // Shift by immediate and insert,
3030 // both double- and quad-register.
3031 let TwoOperandAliasConstraint = "$Vm = $Vd" in {
3032 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
3033 Operand ImmTy, Format f, string OpcodeStr, string Dt,
3034 ValueType Ty,SDNode ShOp>
3035 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
3036 (ins DPR:$src1, DPR:$Vm, ImmTy:$SIMM), f, IIC_VSHLiD,
3037 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
3038 [(set DPR:$Vd, (Ty (ShOp DPR:$src1, DPR:$Vm, (i32 imm:$SIMM))))]>;
3039 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
3040 Operand ImmTy, Format f, string OpcodeStr, string Dt,
3041 ValueType Ty,SDNode ShOp>
3042 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
3043 (ins QPR:$src1, QPR:$Vm, ImmTy:$SIMM), f, IIC_VSHLiQ,
3044 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
3045 [(set QPR:$Vd, (Ty (ShOp QPR:$src1, QPR:$Vm, (i32 imm:$SIMM))))]>;
3048 // Convert, with fractional bits immediate,
3049 // both double- and quad-register.
3050 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
3051 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
3052 SDPatternOperator IntOp>
3053 : N2VImm<op24, op23, op11_8, op7, 0, op4,
3054 (outs DPR:$Vd), (ins DPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
3055 IIC_VUNAD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
3056 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (i32 imm:$SIMM))))]>;
3057 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
3058 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
3059 SDPatternOperator IntOp>
3060 : N2VImm<op24, op23, op11_8, op7, 1, op4,
3061 (outs QPR:$Vd), (ins QPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
3062 IIC_VUNAQ, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
3063 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (i32 imm:$SIMM))))]>;
3065 //===----------------------------------------------------------------------===//
3067 //===----------------------------------------------------------------------===//
3069 // Abbreviations used in multiclass suffixes:
3070 // Q = quarter int (8 bit) elements
3071 // H = half int (16 bit) elements
3072 // S = single int (32 bit) elements
3073 // D = double int (64 bit) elements
3075 // Neon 2-register vector operations and intrinsics.
3077 // Neon 2-register comparisons.
3078 // source operand element sizes of 8, 16 and 32 bits:
3079 multiclass N2V_QHS_cmp<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3080 bits<5> op11_7, bit op4, string opc, string Dt,
3081 string asm, SDNode OpNode> {
3082 // 64-bit vector types.
3083 def v8i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 0, op4,
3084 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
3085 opc, !strconcat(Dt, "8"), asm, "",
3086 [(set DPR:$Vd, (v8i8 (OpNode (v8i8 DPR:$Vm))))]>;
3087 def v4i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
3088 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
3089 opc, !strconcat(Dt, "16"), asm, "",
3090 [(set DPR:$Vd, (v4i16 (OpNode (v4i16 DPR:$Vm))))]>;
3091 def v2i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
3092 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
3093 opc, !strconcat(Dt, "32"), asm, "",
3094 [(set DPR:$Vd, (v2i32 (OpNode (v2i32 DPR:$Vm))))]>;
3095 def v2f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
3096 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
3097 opc, "f32", asm, "",
3098 [(set DPR:$Vd, (v2i32 (OpNode (v2f32 DPR:$Vm))))]> {
3099 let Inst{10} = 1; // overwrite F = 1
3102 // 128-bit vector types.
3103 def v16i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 1, op4,
3104 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
3105 opc, !strconcat(Dt, "8"), asm, "",
3106 [(set QPR:$Vd, (v16i8 (OpNode (v16i8 QPR:$Vm))))]>;
3107 def v8i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
3108 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
3109 opc, !strconcat(Dt, "16"), asm, "",
3110 [(set QPR:$Vd, (v8i16 (OpNode (v8i16 QPR:$Vm))))]>;
3111 def v4i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
3112 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
3113 opc, !strconcat(Dt, "32"), asm, "",
3114 [(set QPR:$Vd, (v4i32 (OpNode (v4i32 QPR:$Vm))))]>;
3115 def v4f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
3116 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
3117 opc, "f32", asm, "",
3118 [(set QPR:$Vd, (v4i32 (OpNode (v4f32 QPR:$Vm))))]> {
3119 let Inst{10} = 1; // overwrite F = 1
3124 // Neon 2-register vector intrinsics,
3125 // element sizes of 8, 16 and 32 bits:
3126 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3127 bits<5> op11_7, bit op4,
3128 InstrItinClass itinD, InstrItinClass itinQ,
3129 string OpcodeStr, string Dt, SDPatternOperator IntOp> {
3130 // 64-bit vector types.
3131 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3132 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
3133 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3134 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
3135 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3136 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
3138 // 128-bit vector types.
3139 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3140 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
3141 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3142 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
3143 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3144 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
3148 // Neon Narrowing 2-register vector operations,
3149 // source operand element sizes of 16, 32 and 64 bits:
3150 multiclass N2VN_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3151 bits<5> op11_7, bit op6, bit op4,
3152 InstrItinClass itin, string OpcodeStr, string Dt,
3154 def v8i8 : N2VN<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
3155 itin, OpcodeStr, !strconcat(Dt, "16"),
3156 v8i8, v8i16, OpNode>;
3157 def v4i16 : N2VN<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
3158 itin, OpcodeStr, !strconcat(Dt, "32"),
3159 v4i16, v4i32, OpNode>;
3160 def v2i32 : N2VN<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
3161 itin, OpcodeStr, !strconcat(Dt, "64"),
3162 v2i32, v2i64, OpNode>;
3165 // Neon Narrowing 2-register vector intrinsics,
3166 // source operand element sizes of 16, 32 and 64 bits:
3167 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3168 bits<5> op11_7, bit op6, bit op4,
3169 InstrItinClass itin, string OpcodeStr, string Dt,
3170 SDPatternOperator IntOp> {
3171 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
3172 itin, OpcodeStr, !strconcat(Dt, "16"),
3173 v8i8, v8i16, IntOp>;
3174 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
3175 itin, OpcodeStr, !strconcat(Dt, "32"),
3176 v4i16, v4i32, IntOp>;
3177 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
3178 itin, OpcodeStr, !strconcat(Dt, "64"),
3179 v2i32, v2i64, IntOp>;
3183 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
3184 // source operand element sizes of 16, 32 and 64 bits:
3185 multiclass N2VL_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
3186 string OpcodeStr, string Dt, SDNode OpNode> {
3187 def v8i16 : N2VL<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
3188 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode>;
3189 def v4i32 : N2VL<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
3190 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
3191 def v2i64 : N2VL<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
3192 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
3196 // Neon 3-register vector operations.
3198 // First with only element sizes of 8, 16 and 32 bits:
3199 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3200 InstrItinClass itinD16, InstrItinClass itinD32,
3201 InstrItinClass itinQ16, InstrItinClass itinQ32,
3202 string OpcodeStr, string Dt,
3203 SDNode OpNode, bit Commutable = 0> {
3204 // 64-bit vector types.
3205 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
3206 OpcodeStr, !strconcat(Dt, "8"),
3207 v8i8, v8i8, OpNode, Commutable>;
3208 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
3209 OpcodeStr, !strconcat(Dt, "16"),
3210 v4i16, v4i16, OpNode, Commutable>;
3211 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
3212 OpcodeStr, !strconcat(Dt, "32"),
3213 v2i32, v2i32, OpNode, Commutable>;
3215 // 128-bit vector types.
3216 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
3217 OpcodeStr, !strconcat(Dt, "8"),
3218 v16i8, v16i8, OpNode, Commutable>;
3219 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
3220 OpcodeStr, !strconcat(Dt, "16"),
3221 v8i16, v8i16, OpNode, Commutable>;
3222 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
3223 OpcodeStr, !strconcat(Dt, "32"),
3224 v4i32, v4i32, OpNode, Commutable>;
3227 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, SDNode ShOp> {
3228 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, "i16", v4i16, ShOp>;
3229 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, "i32", v2i32, ShOp>;
3230 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, "i16", v8i16, v4i16, ShOp>;
3231 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, "i32",
3232 v4i32, v2i32, ShOp>;
3235 // ....then also with element size 64 bits:
3236 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3237 InstrItinClass itinD, InstrItinClass itinQ,
3238 string OpcodeStr, string Dt,
3239 SDNode OpNode, bit Commutable = 0>
3240 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
3241 OpcodeStr, Dt, OpNode, Commutable> {
3242 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
3243 OpcodeStr, !strconcat(Dt, "64"),
3244 v1i64, v1i64, OpNode, Commutable>;
3245 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
3246 OpcodeStr, !strconcat(Dt, "64"),
3247 v2i64, v2i64, OpNode, Commutable>;
3251 // Neon 3-register vector intrinsics.
3253 // First with only element sizes of 16 and 32 bits:
3254 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3255 InstrItinClass itinD16, InstrItinClass itinD32,
3256 InstrItinClass itinQ16, InstrItinClass itinQ32,
3257 string OpcodeStr, string Dt,
3258 SDPatternOperator IntOp, bit Commutable = 0> {
3259 // 64-bit vector types.
3260 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, f, itinD16,
3261 OpcodeStr, !strconcat(Dt, "16"),
3262 v4i16, v4i16, IntOp, Commutable>;
3263 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, f, itinD32,
3264 OpcodeStr, !strconcat(Dt, "32"),
3265 v2i32, v2i32, IntOp, Commutable>;
3267 // 128-bit vector types.
3268 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, f, itinQ16,
3269 OpcodeStr, !strconcat(Dt, "16"),
3270 v8i16, v8i16, IntOp, Commutable>;
3271 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, f, itinQ32,
3272 OpcodeStr, !strconcat(Dt, "32"),
3273 v4i32, v4i32, IntOp, Commutable>;
3275 multiclass N3VInt_HSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3276 InstrItinClass itinD16, InstrItinClass itinD32,
3277 InstrItinClass itinQ16, InstrItinClass itinQ32,
3278 string OpcodeStr, string Dt,
3279 SDPatternOperator IntOp> {
3280 // 64-bit vector types.
3281 def v4i16 : N3VDIntSh<op24, op23, 0b01, op11_8, op4, f, itinD16,
3282 OpcodeStr, !strconcat(Dt, "16"),
3283 v4i16, v4i16, IntOp>;
3284 def v2i32 : N3VDIntSh<op24, op23, 0b10, op11_8, op4, f, itinD32,
3285 OpcodeStr, !strconcat(Dt, "32"),
3286 v2i32, v2i32, IntOp>;
3288 // 128-bit vector types.
3289 def v8i16 : N3VQIntSh<op24, op23, 0b01, op11_8, op4, f, itinQ16,
3290 OpcodeStr, !strconcat(Dt, "16"),
3291 v8i16, v8i16, IntOp>;
3292 def v4i32 : N3VQIntSh<op24, op23, 0b10, op11_8, op4, f, itinQ32,
3293 OpcodeStr, !strconcat(Dt, "32"),
3294 v4i32, v4i32, IntOp>;
3297 multiclass N3VIntSL_HS<bits<4> op11_8,
3298 InstrItinClass itinD16, InstrItinClass itinD32,
3299 InstrItinClass itinQ16, InstrItinClass itinQ32,
3300 string OpcodeStr, string Dt, SDPatternOperator IntOp> {
3301 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
3302 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
3303 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
3304 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
3305 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
3306 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
3307 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
3308 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
3311 // ....then also with element size of 8 bits:
3312 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3313 InstrItinClass itinD16, InstrItinClass itinD32,
3314 InstrItinClass itinQ16, InstrItinClass itinQ32,
3315 string OpcodeStr, string Dt,
3316 SDPatternOperator IntOp, bit Commutable = 0>
3317 : N3VInt_HS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3318 OpcodeStr, Dt, IntOp, Commutable> {
3319 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, f, itinD16,
3320 OpcodeStr, !strconcat(Dt, "8"),
3321 v8i8, v8i8, IntOp, Commutable>;
3322 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, f, itinQ16,
3323 OpcodeStr, !strconcat(Dt, "8"),
3324 v16i8, v16i8, IntOp, Commutable>;
3326 multiclass N3VInt_QHSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3327 InstrItinClass itinD16, InstrItinClass itinD32,
3328 InstrItinClass itinQ16, InstrItinClass itinQ32,
3329 string OpcodeStr, string Dt,
3330 SDPatternOperator IntOp>
3331 : N3VInt_HSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3332 OpcodeStr, Dt, IntOp> {
3333 def v8i8 : N3VDIntSh<op24, op23, 0b00, op11_8, op4, f, itinD16,
3334 OpcodeStr, !strconcat(Dt, "8"),
3336 def v16i8 : N3VQIntSh<op24, op23, 0b00, op11_8, op4, f, itinQ16,
3337 OpcodeStr, !strconcat(Dt, "8"),
3338 v16i8, v16i8, IntOp>;
3342 // ....then also with element size of 64 bits:
3343 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3344 InstrItinClass itinD16, InstrItinClass itinD32,
3345 InstrItinClass itinQ16, InstrItinClass itinQ32,
3346 string OpcodeStr, string Dt,
3347 SDPatternOperator IntOp, bit Commutable = 0>
3348 : N3VInt_QHS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3349 OpcodeStr, Dt, IntOp, Commutable> {
3350 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, f, itinD32,
3351 OpcodeStr, !strconcat(Dt, "64"),
3352 v1i64, v1i64, IntOp, Commutable>;
3353 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, f, itinQ32,
3354 OpcodeStr, !strconcat(Dt, "64"),
3355 v2i64, v2i64, IntOp, Commutable>;
3357 multiclass N3VInt_QHSDSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3358 InstrItinClass itinD16, InstrItinClass itinD32,
3359 InstrItinClass itinQ16, InstrItinClass itinQ32,
3360 string OpcodeStr, string Dt,
3361 SDPatternOperator IntOp>
3362 : N3VInt_QHSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3363 OpcodeStr, Dt, IntOp> {
3364 def v1i64 : N3VDIntSh<op24, op23, 0b11, op11_8, op4, f, itinD32,
3365 OpcodeStr, !strconcat(Dt, "64"),
3366 v1i64, v1i64, IntOp>;
3367 def v2i64 : N3VQIntSh<op24, op23, 0b11, op11_8, op4, f, itinQ32,
3368 OpcodeStr, !strconcat(Dt, "64"),
3369 v2i64, v2i64, IntOp>;
3372 // Neon Narrowing 3-register vector intrinsics,
3373 // source operand element sizes of 16, 32 and 64 bits:
3374 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3375 string OpcodeStr, string Dt,
3376 SDPatternOperator IntOp, bit Commutable = 0> {
3377 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
3378 OpcodeStr, !strconcat(Dt, "16"),
3379 v8i8, v8i16, IntOp, Commutable>;
3380 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
3381 OpcodeStr, !strconcat(Dt, "32"),
3382 v4i16, v4i32, IntOp, Commutable>;
3383 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
3384 OpcodeStr, !strconcat(Dt, "64"),
3385 v2i32, v2i64, IntOp, Commutable>;
3389 // Neon Long 3-register vector operations.
3391 multiclass N3VL_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3392 InstrItinClass itin16, InstrItinClass itin32,
3393 string OpcodeStr, string Dt,
3394 SDNode OpNode, bit Commutable = 0> {
3395 def v8i16 : N3VL<op24, op23, 0b00, op11_8, op4, itin16,
3396 OpcodeStr, !strconcat(Dt, "8"),
3397 v8i16, v8i8, OpNode, Commutable>;
3398 def v4i32 : N3VL<op24, op23, 0b01, op11_8, op4, itin16,
3399 OpcodeStr, !strconcat(Dt, "16"),
3400 v4i32, v4i16, OpNode, Commutable>;
3401 def v2i64 : N3VL<op24, op23, 0b10, op11_8, op4, itin32,
3402 OpcodeStr, !strconcat(Dt, "32"),
3403 v2i64, v2i32, OpNode, Commutable>;
3406 multiclass N3VLSL_HS<bit op24, bits<4> op11_8,
3407 InstrItinClass itin, string OpcodeStr, string Dt,
3409 def v4i16 : N3VLSL16<op24, 0b01, op11_8, itin, OpcodeStr,
3410 !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
3411 def v2i32 : N3VLSL<op24, 0b10, op11_8, itin, OpcodeStr,
3412 !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
3415 multiclass N3VLExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3416 InstrItinClass itin16, InstrItinClass itin32,
3417 string OpcodeStr, string Dt,
3418 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
3419 def v8i16 : N3VLExt<op24, op23, 0b00, op11_8, op4, itin16,
3420 OpcodeStr, !strconcat(Dt, "8"),
3421 v8i16, v8i8, OpNode, ExtOp, Commutable>;
3422 def v4i32 : N3VLExt<op24, op23, 0b01, op11_8, op4, itin16,
3423 OpcodeStr, !strconcat(Dt, "16"),
3424 v4i32, v4i16, OpNode, ExtOp, Commutable>;
3425 def v2i64 : N3VLExt<op24, op23, 0b10, op11_8, op4, itin32,
3426 OpcodeStr, !strconcat(Dt, "32"),
3427 v2i64, v2i32, OpNode, ExtOp, Commutable>;
3430 // Neon Long 3-register vector intrinsics.
3432 // First with only element sizes of 16 and 32 bits:
3433 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
3434 InstrItinClass itin16, InstrItinClass itin32,
3435 string OpcodeStr, string Dt,
3436 SDPatternOperator IntOp, bit Commutable = 0> {
3437 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin16,
3438 OpcodeStr, !strconcat(Dt, "16"),
3439 v4i32, v4i16, IntOp, Commutable>;
3440 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin32,
3441 OpcodeStr, !strconcat(Dt, "32"),
3442 v2i64, v2i32, IntOp, Commutable>;
3445 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
3446 InstrItinClass itin, string OpcodeStr, string Dt,
3447 SDPatternOperator IntOp> {
3448 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
3449 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
3450 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
3451 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3454 // ....then also with element size of 8 bits:
3455 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3456 InstrItinClass itin16, InstrItinClass itin32,
3457 string OpcodeStr, string Dt,
3458 SDPatternOperator IntOp, bit Commutable = 0>
3459 : N3VLInt_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt,
3460 IntOp, Commutable> {
3461 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin16,
3462 OpcodeStr, !strconcat(Dt, "8"),
3463 v8i16, v8i8, IntOp, Commutable>;
3466 // ....with explicit extend (VABDL).
3467 multiclass N3VLIntExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3468 InstrItinClass itin, string OpcodeStr, string Dt,
3469 SDPatternOperator IntOp, SDNode ExtOp, bit Commutable = 0> {
3470 def v8i16 : N3VLIntExt<op24, op23, 0b00, op11_8, op4, itin,
3471 OpcodeStr, !strconcat(Dt, "8"),
3472 v8i16, v8i8, IntOp, ExtOp, Commutable>;
3473 def v4i32 : N3VLIntExt<op24, op23, 0b01, op11_8, op4, itin,
3474 OpcodeStr, !strconcat(Dt, "16"),
3475 v4i32, v4i16, IntOp, ExtOp, Commutable>;
3476 def v2i64 : N3VLIntExt<op24, op23, 0b10, op11_8, op4, itin,
3477 OpcodeStr, !strconcat(Dt, "32"),
3478 v2i64, v2i32, IntOp, ExtOp, Commutable>;
3482 // Neon Wide 3-register vector intrinsics,
3483 // source operand element sizes of 8, 16 and 32 bits:
3484 multiclass N3VW_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3485 string OpcodeStr, string Dt,
3486 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
3487 def v8i16 : N3VW<op24, op23, 0b00, op11_8, op4,
3488 OpcodeStr, !strconcat(Dt, "8"),
3489 v8i16, v8i8, OpNode, ExtOp, Commutable>;
3490 def v4i32 : N3VW<op24, op23, 0b01, op11_8, op4,
3491 OpcodeStr, !strconcat(Dt, "16"),
3492 v4i32, v4i16, OpNode, ExtOp, Commutable>;
3493 def v2i64 : N3VW<op24, op23, 0b10, op11_8, op4,
3494 OpcodeStr, !strconcat(Dt, "32"),
3495 v2i64, v2i32, OpNode, ExtOp, Commutable>;
3499 // Neon Multiply-Op vector operations,
3500 // element sizes of 8, 16 and 32 bits:
3501 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3502 InstrItinClass itinD16, InstrItinClass itinD32,
3503 InstrItinClass itinQ16, InstrItinClass itinQ32,
3504 string OpcodeStr, string Dt, SDNode OpNode> {
3505 // 64-bit vector types.
3506 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
3507 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
3508 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
3509 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
3510 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
3511 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
3513 // 128-bit vector types.
3514 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
3515 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
3516 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
3517 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
3518 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
3519 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
3522 multiclass N3VMulOpSL_HS<bits<4> op11_8,
3523 InstrItinClass itinD16, InstrItinClass itinD32,
3524 InstrItinClass itinQ16, InstrItinClass itinQ32,
3525 string OpcodeStr, string Dt, SDNode ShOp> {
3526 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
3527 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
3528 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
3529 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
3530 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
3531 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
3533 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
3534 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
3538 // Neon Intrinsic-Op vector operations,
3539 // element sizes of 8, 16 and 32 bits:
3540 multiclass N3VIntOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3541 InstrItinClass itinD, InstrItinClass itinQ,
3542 string OpcodeStr, string Dt, SDPatternOperator IntOp,
3544 // 64-bit vector types.
3545 def v8i8 : N3VDIntOp<op24, op23, 0b00, op11_8, op4, itinD,
3546 OpcodeStr, !strconcat(Dt, "8"), v8i8, IntOp, OpNode>;
3547 def v4i16 : N3VDIntOp<op24, op23, 0b01, op11_8, op4, itinD,
3548 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp, OpNode>;
3549 def v2i32 : N3VDIntOp<op24, op23, 0b10, op11_8, op4, itinD,
3550 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp, OpNode>;
3552 // 128-bit vector types.
3553 def v16i8 : N3VQIntOp<op24, op23, 0b00, op11_8, op4, itinQ,
3554 OpcodeStr, !strconcat(Dt, "8"), v16i8, IntOp, OpNode>;
3555 def v8i16 : N3VQIntOp<op24, op23, 0b01, op11_8, op4, itinQ,
3556 OpcodeStr, !strconcat(Dt, "16"), v8i16, IntOp, OpNode>;
3557 def v4i32 : N3VQIntOp<op24, op23, 0b10, op11_8, op4, itinQ,
3558 OpcodeStr, !strconcat(Dt, "32"), v4i32, IntOp, OpNode>;
3561 // Neon 3-argument intrinsics,
3562 // element sizes of 8, 16 and 32 bits:
3563 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3564 InstrItinClass itinD, InstrItinClass itinQ,
3565 string OpcodeStr, string Dt, SDPatternOperator IntOp> {
3566 // 64-bit vector types.
3567 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, itinD,
3568 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
3569 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, itinD,
3570 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
3571 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, itinD,
3572 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
3574 // 128-bit vector types.
3575 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, itinQ,
3576 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
3577 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, itinQ,
3578 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
3579 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, itinQ,
3580 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
3584 // Neon Long Multiply-Op vector operations,
3585 // element sizes of 8, 16 and 32 bits:
3586 multiclass N3VLMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3587 InstrItinClass itin16, InstrItinClass itin32,
3588 string OpcodeStr, string Dt, SDNode MulOp,
3590 def v8i16 : N3VLMulOp<op24, op23, 0b00, op11_8, op4, itin16, OpcodeStr,
3591 !strconcat(Dt, "8"), v8i16, v8i8, MulOp, OpNode>;
3592 def v4i32 : N3VLMulOp<op24, op23, 0b01, op11_8, op4, itin16, OpcodeStr,
3593 !strconcat(Dt, "16"), v4i32, v4i16, MulOp, OpNode>;
3594 def v2i64 : N3VLMulOp<op24, op23, 0b10, op11_8, op4, itin32, OpcodeStr,
3595 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
3598 multiclass N3VLMulOpSL_HS<bit op24, bits<4> op11_8, string OpcodeStr,
3599 string Dt, SDNode MulOp, SDNode OpNode> {
3600 def v4i16 : N3VLMulOpSL16<op24, 0b01, op11_8, IIC_VMACi16D, OpcodeStr,
3601 !strconcat(Dt,"16"), v4i32, v4i16, MulOp, OpNode>;
3602 def v2i32 : N3VLMulOpSL<op24, 0b10, op11_8, IIC_VMACi32D, OpcodeStr,
3603 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
3607 // Neon Long 3-argument intrinsics.
3609 // First with only element sizes of 16 and 32 bits:
3610 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
3611 InstrItinClass itin16, InstrItinClass itin32,
3612 string OpcodeStr, string Dt, SDPatternOperator IntOp> {
3613 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, itin16,
3614 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
3615 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, itin32,
3616 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3619 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
3620 string OpcodeStr, string Dt, SDPatternOperator IntOp> {
3621 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
3622 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
3623 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
3624 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3627 // ....then also with element size of 8 bits:
3628 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3629 InstrItinClass itin16, InstrItinClass itin32,
3630 string OpcodeStr, string Dt, SDPatternOperator IntOp>
3631 : N3VLInt3_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt, IntOp> {
3632 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, itin16,
3633 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
3636 // ....with explicit extend (VABAL).
3637 multiclass N3VLIntExtOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3638 InstrItinClass itin, string OpcodeStr, string Dt,
3639 SDPatternOperator IntOp, SDNode ExtOp, SDNode OpNode> {
3640 def v8i16 : N3VLIntExtOp<op24, op23, 0b00, op11_8, op4, itin,
3641 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8,
3642 IntOp, ExtOp, OpNode>;
3643 def v4i32 : N3VLIntExtOp<op24, op23, 0b01, op11_8, op4, itin,
3644 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16,
3645 IntOp, ExtOp, OpNode>;
3646 def v2i64 : N3VLIntExtOp<op24, op23, 0b10, op11_8, op4, itin,
3647 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32,
3648 IntOp, ExtOp, OpNode>;
3652 // Neon Pairwise long 2-register intrinsics,
3653 // element sizes of 8, 16 and 32 bits:
3654 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3655 bits<5> op11_7, bit op4,
3656 string OpcodeStr, string Dt, SDPatternOperator IntOp> {
3657 // 64-bit vector types.
3658 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3659 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
3660 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3661 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
3662 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3663 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
3665 // 128-bit vector types.
3666 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3667 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
3668 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3669 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
3670 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3671 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
3675 // Neon Pairwise long 2-register accumulate intrinsics,
3676 // element sizes of 8, 16 and 32 bits:
3677 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3678 bits<5> op11_7, bit op4,
3679 string OpcodeStr, string Dt, SDPatternOperator IntOp> {
3680 // 64-bit vector types.
3681 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3682 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
3683 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3684 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
3685 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3686 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
3688 // 128-bit vector types.
3689 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3690 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
3691 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3692 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
3693 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3694 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
3698 // Neon 2-register vector shift by immediate,
3699 // with f of either N2RegVShLFrm or N2RegVShRFrm
3700 // element sizes of 8, 16, 32 and 64 bits:
3701 multiclass N2VShL_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3702 InstrItinClass itin, string OpcodeStr, string Dt,
3704 // 64-bit vector types.
3705 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3706 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
3707 let Inst{21-19} = 0b001; // imm6 = 001xxx
3709 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3710 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
3711 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3713 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3714 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
3715 let Inst{21} = 0b1; // imm6 = 1xxxxx
3717 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, N2RegVShLFrm, itin, i32imm,
3718 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
3721 // 128-bit vector types.
3722 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3723 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
3724 let Inst{21-19} = 0b001; // imm6 = 001xxx
3726 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3727 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
3728 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3730 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3731 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
3732 let Inst{21} = 0b1; // imm6 = 1xxxxx
3734 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, N2RegVShLFrm, itin, i32imm,
3735 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
3738 multiclass N2VShR_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3739 InstrItinClass itin, string OpcodeStr, string Dt,
3740 string baseOpc, SDNode OpNode> {
3741 // 64-bit vector types.
3742 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm8,
3743 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
3744 let Inst{21-19} = 0b001; // imm6 = 001xxx
3746 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm16,
3747 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
3748 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3750 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm32,
3751 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
3752 let Inst{21} = 0b1; // imm6 = 1xxxxx
3754 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, N2RegVShRFrm, itin, shr_imm64,
3755 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
3758 // 128-bit vector types.
3759 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm8,
3760 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
3761 let Inst{21-19} = 0b001; // imm6 = 001xxx
3763 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm16,
3764 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
3765 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3767 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm32,
3768 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
3769 let Inst{21} = 0b1; // imm6 = 1xxxxx
3771 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, N2RegVShRFrm, itin, shr_imm64,
3772 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
3776 // Neon Shift-Accumulate vector operations,
3777 // element sizes of 8, 16, 32 and 64 bits:
3778 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3779 string OpcodeStr, string Dt, SDNode ShOp> {
3780 // 64-bit vector types.
3781 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm8,
3782 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
3783 let Inst{21-19} = 0b001; // imm6 = 001xxx
3785 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm16,
3786 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
3787 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3789 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm32,
3790 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
3791 let Inst{21} = 0b1; // imm6 = 1xxxxx
3793 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4, shr_imm64,
3794 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
3797 // 128-bit vector types.
3798 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm8,
3799 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
3800 let Inst{21-19} = 0b001; // imm6 = 001xxx
3802 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm16,
3803 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
3804 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3806 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm32,
3807 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
3808 let Inst{21} = 0b1; // imm6 = 1xxxxx
3810 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4, shr_imm64,
3811 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
3815 // Neon Shift-Insert vector operations,
3816 // with f of either N2RegVShLFrm or N2RegVShRFrm
3817 // element sizes of 8, 16, 32 and 64 bits:
3818 multiclass N2VShInsL_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3820 // 64-bit vector types.
3821 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3822 N2RegVShLFrm, OpcodeStr, "8", v8i8, NEONvsli> {
3823 let Inst{21-19} = 0b001; // imm6 = 001xxx
3825 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3826 N2RegVShLFrm, OpcodeStr, "16", v4i16, NEONvsli> {
3827 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3829 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3830 N2RegVShLFrm, OpcodeStr, "32", v2i32, NEONvsli> {
3831 let Inst{21} = 0b1; // imm6 = 1xxxxx
3833 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4, i32imm,
3834 N2RegVShLFrm, OpcodeStr, "64", v1i64, NEONvsli>;
3837 // 128-bit vector types.
3838 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3839 N2RegVShLFrm, OpcodeStr, "8", v16i8, NEONvsli> {
3840 let Inst{21-19} = 0b001; // imm6 = 001xxx
3842 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3843 N2RegVShLFrm, OpcodeStr, "16", v8i16, NEONvsli> {
3844 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3846 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3847 N2RegVShLFrm, OpcodeStr, "32", v4i32, NEONvsli> {
3848 let Inst{21} = 0b1; // imm6 = 1xxxxx
3850 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4, i32imm,
3851 N2RegVShLFrm, OpcodeStr, "64", v2i64, NEONvsli>;
3854 multiclass N2VShInsR_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3856 // 64-bit vector types.
3857 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm8,
3858 N2RegVShRFrm, OpcodeStr, "8", v8i8, NEONvsri> {
3859 let Inst{21-19} = 0b001; // imm6 = 001xxx
3861 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm16,
3862 N2RegVShRFrm, OpcodeStr, "16", v4i16, NEONvsri> {
3863 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3865 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm32,
3866 N2RegVShRFrm, OpcodeStr, "32", v2i32, NEONvsri> {
3867 let Inst{21} = 0b1; // imm6 = 1xxxxx
3869 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4, shr_imm64,
3870 N2RegVShRFrm, OpcodeStr, "64", v1i64, NEONvsri>;
3873 // 128-bit vector types.
3874 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm8,
3875 N2RegVShRFrm, OpcodeStr, "8", v16i8, NEONvsri> {
3876 let Inst{21-19} = 0b001; // imm6 = 001xxx
3878 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm16,
3879 N2RegVShRFrm, OpcodeStr, "16", v8i16, NEONvsri> {
3880 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3882 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm32,
3883 N2RegVShRFrm, OpcodeStr, "32", v4i32, NEONvsri> {
3884 let Inst{21} = 0b1; // imm6 = 1xxxxx
3886 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4, shr_imm64,
3887 N2RegVShRFrm, OpcodeStr, "64", v2i64, NEONvsri>;
3891 // Neon Shift Long operations,
3892 // element sizes of 8, 16, 32 bits:
3893 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
3894 bit op4, string OpcodeStr, string Dt, SDNode OpNode> {
3895 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3896 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, imm1_7, OpNode> {
3897 let Inst{21-19} = 0b001; // imm6 = 001xxx
3899 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3900 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, imm1_15, OpNode> {
3901 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3903 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3904 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, imm1_31, OpNode> {
3905 let Inst{21} = 0b1; // imm6 = 1xxxxx
3909 // Neon Shift Narrow operations,
3910 // element sizes of 16, 32, 64 bits:
3911 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
3912 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
3914 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3915 OpcodeStr, !strconcat(Dt, "16"),
3916 v8i8, v8i16, shr_imm8, OpNode> {
3917 let Inst{21-19} = 0b001; // imm6 = 001xxx
3919 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3920 OpcodeStr, !strconcat(Dt, "32"),
3921 v4i16, v4i32, shr_imm16, OpNode> {
3922 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3924 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3925 OpcodeStr, !strconcat(Dt, "64"),
3926 v2i32, v2i64, shr_imm32, OpNode> {
3927 let Inst{21} = 0b1; // imm6 = 1xxxxx
3931 //===----------------------------------------------------------------------===//
3932 // Instruction Definitions.
3933 //===----------------------------------------------------------------------===//
3935 // Vector Add Operations.
3937 // VADD : Vector Add (integer and floating-point)
3938 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
3940 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
3941 v2f32, v2f32, fadd, 1>;
3942 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
3943 v4f32, v4f32, fadd, 1>;
3944 // VADDL : Vector Add Long (Q = D + D)
3945 defm VADDLs : N3VLExt_QHS<0,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3946 "vaddl", "s", add, sext, 1>;
3947 defm VADDLu : N3VLExt_QHS<1,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3948 "vaddl", "u", add, zext, 1>;
3950 def : Pat<(v4i32 (NEONvaddlu (v4i16 DPR:$src1), (v4i16 DPR:$src2))),
3951 (v4i32 (VADDLuv4i32 DPR:$src1, DPR:$src2))>;
3952 def : Pat<(v8i16 (NEONvaddlu (v8i8 DPR:$src1), (v8i8 DPR:$src2))),
3953 (v8i16 (VADDLuv8i16 DPR:$src1, DPR:$src2))>;
3954 def : Pat<(v2i64 (NEONvaddlu (v2i32 DPR:$src1), (v2i32 DPR:$src2))),
3955 (v2i64 (VADDLuv2i64 DPR:$src1, DPR:$src2))>;
3957 def : Pat<(v4i32 (NEONvaddls (v4i16 DPR:$src1), (v4i16 DPR:$src2))),
3958 (v4i32 (VADDLsv4i32 DPR:$src1, DPR:$src2))>;
3959 def : Pat<(v8i16 (NEONvaddls (v8i8 DPR:$src1), (v8i8 DPR:$src2))),
3960 (v8i16 (VADDLsv8i16 DPR:$src1, DPR:$src2))>;
3961 def : Pat<(v2i64 (NEONvaddls (v2i32 DPR:$src1), (v2i32 DPR:$src2))),
3962 (v2i64 (VADDLsv2i64 DPR:$src1, DPR:$src2))>;
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>;
4255 def : Pat<(v4i32 (NEONvsublu (v4i16 DPR:$src1), (v4i16 DPR:$src2))),
4256 (v4i32 (VSUBLuv4i32 DPR:$src1, DPR:$src2))>;
4257 def : Pat<(v8i16 (NEONvsublu (v8i8 DPR:$src1), (v8i8 DPR:$src2))),
4258 (v8i16 (VSUBLuv8i16 DPR:$src1, DPR:$src2))>;
4259 def : Pat<(v2i64 (NEONvsublu (v2i32 DPR:$src1), (v2i32 DPR:$src2))),
4260 (v2i64 (VSUBLuv2i64 DPR:$src1, DPR:$src2))>;
4262 def : Pat<(v4i32 (NEONvsubls (v4i16 DPR:$src1), (v4i16 DPR:$src2))),
4263 (v4i32 (VSUBLsv4i32 DPR:$src1, DPR:$src2))>;
4264 def : Pat<(v8i16 (NEONvsubls (v8i8 DPR:$src1), (v8i8 DPR:$src2))),
4265 (v8i16 (VSUBLsv8i16 DPR:$src1, DPR:$src2))>;
4266 def : Pat<(v2i64 (NEONvsubls (v2i32 DPR:$src1), (v2i32 DPR:$src2))),
4267 (v2i64 (VSUBLsv2i64 DPR:$src1, DPR:$src2))>;
4269 // VSUBW : Vector Subtract Wide (Q = Q - D)
4270 defm VSUBWs : N3VW_QHS<0,1,0b0011,0, "vsubw", "s", sub, sext, 0>;
4271 defm VSUBWu : N3VW_QHS<1,1,0b0011,0, "vsubw", "u", sub, zext, 0>;
4272 // VHSUB : Vector Halving Subtract
4273 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, N3RegFrm,
4274 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4275 "vhsub", "s", int_arm_neon_vhsubs, 0>;
4276 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, N3RegFrm,
4277 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4278 "vhsub", "u", int_arm_neon_vhsubu, 0>;
4279 // VQSUB : Vector Saturing Subtract
4280 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, N3RegFrm,
4281 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4282 "vqsub", "s", int_arm_neon_vqsubs, 0>;
4283 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, N3RegFrm,
4284 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4285 "vqsub", "u", int_arm_neon_vqsubu, 0>;
4286 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
4287 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i",
4288 int_arm_neon_vsubhn, 0>;
4289 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
4290 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
4291 int_arm_neon_vrsubhn, 0>;
4293 // Vector Comparisons.
4295 // VCEQ : Vector Compare Equal
4296 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
4297 IIC_VSUBi4Q, "vceq", "i", NEONvceq, 1>;
4298 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
4300 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
4303 defm VCEQz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
4304 "$Vd, $Vm, #0", NEONvceqz>;
4306 // VCGE : Vector Compare Greater Than or Equal
4307 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
4308 IIC_VSUBi4Q, "vcge", "s", NEONvcge, 0>;
4309 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
4310 IIC_VSUBi4Q, "vcge", "u", NEONvcgeu, 0>;
4311 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32", v2i32, v2f32,
4313 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
4316 defm VCGEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
4317 "$Vd, $Vm, #0", NEONvcgez>;
4318 defm VCLEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
4319 "$Vd, $Vm, #0", NEONvclez>;
4321 // VCGT : Vector Compare Greater Than
4322 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
4323 IIC_VSUBi4Q, "vcgt", "s", NEONvcgt, 0>;
4324 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
4325 IIC_VSUBi4Q, "vcgt", "u", NEONvcgtu, 0>;
4326 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
4328 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
4331 defm VCGTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
4332 "$Vd, $Vm, #0", NEONvcgtz>;
4333 defm VCLTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
4334 "$Vd, $Vm, #0", NEONvcltz>;
4336 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
4337 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacge",
4338 "f32", v2i32, v2f32, int_arm_neon_vacged, 0>;
4339 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacge",
4340 "f32", v4i32, v4f32, int_arm_neon_vacgeq, 0>;
4341 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
4342 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacgt",
4343 "f32", v2i32, v2f32, int_arm_neon_vacgtd, 0>;
4344 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacgt",
4345 "f32", v4i32, v4f32, int_arm_neon_vacgtq, 0>;
4346 // VTST : Vector Test Bits
4347 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
4348 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
4350 // Vector Bitwise Operations.
4352 def vnotd : PatFrag<(ops node:$in),
4353 (xor node:$in, (bitconvert (v8i8 NEONimmAllOnesV)))>;
4354 def vnotq : PatFrag<(ops node:$in),
4355 (xor node:$in, (bitconvert (v16i8 NEONimmAllOnesV)))>;
4358 // VAND : Vector Bitwise AND
4359 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
4360 v2i32, v2i32, and, 1>;
4361 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
4362 v4i32, v4i32, and, 1>;
4364 // VEOR : Vector Bitwise Exclusive OR
4365 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
4366 v2i32, v2i32, xor, 1>;
4367 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
4368 v4i32, v4i32, xor, 1>;
4370 // VORR : Vector Bitwise OR
4371 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
4372 v2i32, v2i32, or, 1>;
4373 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
4374 v4i32, v4i32, or, 1>;
4376 def VORRiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 0, 1,
4377 (outs DPR:$Vd), (ins nImmSplatI16:$SIMM, DPR:$src),
4379 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
4381 (v4i16 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
4382 let Inst{9} = SIMM{9};
4385 def VORRiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 0, 1,
4386 (outs DPR:$Vd), (ins nImmSplatI32:$SIMM, DPR:$src),
4388 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
4390 (v2i32 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
4391 let Inst{10-9} = SIMM{10-9};
4394 def VORRiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 0, 1,
4395 (outs QPR:$Vd), (ins nImmSplatI16:$SIMM, QPR:$src),
4397 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
4399 (v8i16 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
4400 let Inst{9} = SIMM{9};
4403 def VORRiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 0, 1,
4404 (outs QPR:$Vd), (ins nImmSplatI32:$SIMM, QPR:$src),
4406 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
4408 (v4i32 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
4409 let Inst{10-9} = SIMM{10-9};
4413 // VBIC : Vector Bitwise Bit Clear (AND NOT)
4414 let TwoOperandAliasConstraint = "$Vn = $Vd" in {
4415 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
4416 (ins DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VBINiD,
4417 "vbic", "$Vd, $Vn, $Vm", "",
4418 [(set DPR:$Vd, (v2i32 (and DPR:$Vn,
4419 (vnotd DPR:$Vm))))]>;
4420 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
4421 (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINiQ,
4422 "vbic", "$Vd, $Vn, $Vm", "",
4423 [(set QPR:$Vd, (v4i32 (and QPR:$Vn,
4424 (vnotq QPR:$Vm))))]>;
4427 def VBICiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 1, 1,
4428 (outs DPR:$Vd), (ins nImmSplatI16:$SIMM, DPR:$src),
4430 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
4432 (v4i16 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
4433 let Inst{9} = SIMM{9};
4436 def VBICiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 1, 1,
4437 (outs DPR:$Vd), (ins nImmSplatI32:$SIMM, DPR:$src),
4439 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
4441 (v2i32 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
4442 let Inst{10-9} = SIMM{10-9};
4445 def VBICiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 1, 1,
4446 (outs QPR:$Vd), (ins nImmSplatI16:$SIMM, QPR:$src),
4448 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
4450 (v8i16 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
4451 let Inst{9} = SIMM{9};
4454 def VBICiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 1, 1,
4455 (outs QPR:$Vd), (ins nImmSplatI32:$SIMM, QPR:$src),
4457 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
4459 (v4i32 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
4460 let Inst{10-9} = SIMM{10-9};
4463 // VORN : Vector Bitwise OR NOT
4464 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$Vd),
4465 (ins DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VBINiD,
4466 "vorn", "$Vd, $Vn, $Vm", "",
4467 [(set DPR:$Vd, (v2i32 (or DPR:$Vn,
4468 (vnotd DPR:$Vm))))]>;
4469 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$Vd),
4470 (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINiQ,
4471 "vorn", "$Vd, $Vn, $Vm", "",
4472 [(set QPR:$Vd, (v4i32 (or QPR:$Vn,
4473 (vnotq QPR:$Vm))))]>;
4475 // VMVN : Vector Bitwise NOT (Immediate)
4477 let isReMaterializable = 1 in {
4479 def VMVNv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 1, 1, (outs DPR:$Vd),
4480 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4481 "vmvn", "i16", "$Vd, $SIMM", "",
4482 [(set DPR:$Vd, (v4i16 (NEONvmvnImm timm:$SIMM)))]> {
4483 let Inst{9} = SIMM{9};
4486 def VMVNv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 1, 1, (outs QPR:$Vd),
4487 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4488 "vmvn", "i16", "$Vd, $SIMM", "",
4489 [(set QPR:$Vd, (v8i16 (NEONvmvnImm timm:$SIMM)))]> {
4490 let Inst{9} = SIMM{9};
4493 def VMVNv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 1, 1, (outs DPR:$Vd),
4494 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4495 "vmvn", "i32", "$Vd, $SIMM", "",
4496 [(set DPR:$Vd, (v2i32 (NEONvmvnImm timm:$SIMM)))]> {
4497 let Inst{11-8} = SIMM{11-8};
4500 def VMVNv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 1, 1, (outs QPR:$Vd),
4501 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4502 "vmvn", "i32", "$Vd, $SIMM", "",
4503 [(set QPR:$Vd, (v4i32 (NEONvmvnImm timm:$SIMM)))]> {
4504 let Inst{11-8} = SIMM{11-8};
4508 // VMVN : Vector Bitwise NOT
4509 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
4510 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VSUBiD,
4511 "vmvn", "$Vd, $Vm", "",
4512 [(set DPR:$Vd, (v2i32 (vnotd DPR:$Vm)))]>;
4513 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
4514 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VSUBiD,
4515 "vmvn", "$Vd, $Vm", "",
4516 [(set QPR:$Vd, (v4i32 (vnotq QPR:$Vm)))]>;
4517 def : Pat<(v2i32 (vnotd DPR:$src)), (VMVNd DPR:$src)>;
4518 def : Pat<(v4i32 (vnotq QPR:$src)), (VMVNq QPR:$src)>;
4520 // VBSL : Vector Bitwise Select
4521 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
4522 (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4523 N3RegFrm, IIC_VCNTiD,
4524 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4526 (v2i32 (NEONvbsl DPR:$src1, DPR:$Vn, DPR:$Vm)))]>;
4527 def : Pat<(v8i8 (int_arm_neon_vbsl (v8i8 DPR:$src1),
4528 (v8i8 DPR:$Vn), (v8i8 DPR:$Vm))),
4529 (VBSLd DPR:$src1, DPR:$Vn, DPR:$Vm)>,
4530 Requires<[HasNEON]>;
4531 def : Pat<(v4i16 (int_arm_neon_vbsl (v4i16 DPR:$src1),
4532 (v4i16 DPR:$Vn), (v4i16 DPR:$Vm))),
4533 (VBSLd DPR:$src1, DPR:$Vn, DPR:$Vm)>,
4534 Requires<[HasNEON]>;
4535 def : Pat<(v2i32 (int_arm_neon_vbsl (v2i32 DPR:$src1),
4536 (v2i32 DPR:$Vn), (v2i32 DPR:$Vm))),
4537 (VBSLd DPR:$src1, DPR:$Vn, DPR:$Vm)>,
4538 Requires<[HasNEON]>;
4539 def : Pat<(v2f32 (int_arm_neon_vbsl (v2f32 DPR:$src1),
4540 (v2f32 DPR:$Vn), (v2f32 DPR:$Vm))),
4541 (VBSLd DPR:$src1, DPR:$Vn, DPR:$Vm)>,
4542 Requires<[HasNEON]>;
4543 def : Pat<(v1i64 (int_arm_neon_vbsl (v1i64 DPR:$src1),
4544 (v1i64 DPR:$Vn), (v1i64 DPR:$Vm))),
4545 (VBSLd DPR:$src1, DPR:$Vn, DPR:$Vm)>,
4546 Requires<[HasNEON]>;
4548 def : Pat<(v2i32 (or (and DPR:$Vn, DPR:$Vd),
4549 (and DPR:$Vm, (vnotd DPR:$Vd)))),
4550 (VBSLd DPR:$Vd, DPR:$Vn, DPR:$Vm)>,
4551 Requires<[HasNEON]>;
4553 def : Pat<(v1i64 (or (and DPR:$Vn, DPR:$Vd),
4554 (and DPR:$Vm, (vnotd DPR:$Vd)))),
4555 (VBSLd DPR:$Vd, DPR:$Vn, DPR:$Vm)>,
4556 Requires<[HasNEON]>;
4558 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
4559 (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4560 N3RegFrm, IIC_VCNTiQ,
4561 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4563 (v4i32 (NEONvbsl QPR:$src1, QPR:$Vn, QPR:$Vm)))]>;
4565 def : Pat<(v16i8 (int_arm_neon_vbsl (v16i8 QPR:$src1),
4566 (v16i8 QPR:$Vn), (v16i8 QPR:$Vm))),
4567 (VBSLq QPR:$src1, QPR:$Vn, QPR:$Vm)>,
4568 Requires<[HasNEON]>;
4569 def : Pat<(v8i16 (int_arm_neon_vbsl (v8i16 QPR:$src1),
4570 (v8i16 QPR:$Vn), (v8i16 QPR:$Vm))),
4571 (VBSLq QPR:$src1, QPR:$Vn, QPR:$Vm)>,
4572 Requires<[HasNEON]>;
4573 def : Pat<(v4i32 (int_arm_neon_vbsl (v4i32 QPR:$src1),
4574 (v4i32 QPR:$Vn), (v4i32 QPR:$Vm))),
4575 (VBSLq QPR:$src1, QPR:$Vn, QPR:$Vm)>,
4576 Requires<[HasNEON]>;
4577 def : Pat<(v4f32 (int_arm_neon_vbsl (v4f32 QPR:$src1),
4578 (v4f32 QPR:$Vn), (v4f32 QPR:$Vm))),
4579 (VBSLq QPR:$src1, QPR:$Vn, QPR:$Vm)>,
4580 Requires<[HasNEON]>;
4581 def : Pat<(v2i64 (int_arm_neon_vbsl (v2i64 QPR:$src1),
4582 (v2i64 QPR:$Vn), (v2i64 QPR:$Vm))),
4583 (VBSLq QPR:$src1, QPR:$Vn, QPR:$Vm)>,
4584 Requires<[HasNEON]>;
4586 def : Pat<(v4i32 (or (and QPR:$Vn, QPR:$Vd),
4587 (and QPR:$Vm, (vnotq QPR:$Vd)))),
4588 (VBSLq QPR:$Vd, QPR:$Vn, QPR:$Vm)>,
4589 Requires<[HasNEON]>;
4590 def : Pat<(v2i64 (or (and QPR:$Vn, QPR:$Vd),
4591 (and QPR:$Vm, (vnotq QPR:$Vd)))),
4592 (VBSLq QPR:$Vd, QPR:$Vn, QPR:$Vm)>,
4593 Requires<[HasNEON]>;
4595 // VBIF : Vector Bitwise Insert if False
4596 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
4597 // FIXME: This instruction's encoding MAY NOT BE correct.
4598 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
4599 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4600 N3RegFrm, IIC_VBINiD,
4601 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4603 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
4604 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4605 N3RegFrm, IIC_VBINiQ,
4606 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4609 // VBIT : Vector Bitwise Insert if True
4610 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
4611 // FIXME: This instruction's encoding MAY NOT BE correct.
4612 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
4613 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4614 N3RegFrm, IIC_VBINiD,
4615 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4617 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
4618 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4619 N3RegFrm, IIC_VBINiQ,
4620 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4623 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
4624 // for equivalent operations with different register constraints; it just
4627 // Vector Absolute Differences.
4629 // VABD : Vector Absolute Difference
4630 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, N3RegFrm,
4631 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4632 "vabd", "s", int_arm_neon_vabds, 1>;
4633 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, N3RegFrm,
4634 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4635 "vabd", "u", int_arm_neon_vabdu, 1>;
4636 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBIND,
4637 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 1>;
4638 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBINQ,
4639 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 1>;
4641 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
4642 defm VABDLs : N3VLIntExt_QHS<0,1,0b0111,0, IIC_VSUBi4Q,
4643 "vabdl", "s", int_arm_neon_vabds, zext, 1>;
4644 defm VABDLu : N3VLIntExt_QHS<1,1,0b0111,0, IIC_VSUBi4Q,
4645 "vabdl", "u", int_arm_neon_vabdu, zext, 1>;
4647 // VABA : Vector Absolute Difference and Accumulate
4648 defm VABAs : N3VIntOp_QHS<0,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
4649 "vaba", "s", int_arm_neon_vabds, add>;
4650 defm VABAu : N3VIntOp_QHS<1,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
4651 "vaba", "u", int_arm_neon_vabdu, add>;
4653 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
4654 defm VABALs : N3VLIntExtOp_QHS<0,1,0b0101,0, IIC_VABAD,
4655 "vabal", "s", int_arm_neon_vabds, zext, add>;
4656 defm VABALu : N3VLIntExtOp_QHS<1,1,0b0101,0, IIC_VABAD,
4657 "vabal", "u", int_arm_neon_vabdu, zext, add>;
4659 // Vector Maximum and Minimum.
4661 // VMAX : Vector Maximum
4662 defm VMAXs : N3VInt_QHS<0, 0, 0b0110, 0, N3RegFrm,
4663 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4664 "vmax", "s", int_arm_neon_vmaxs, 1>;
4665 defm VMAXu : N3VInt_QHS<1, 0, 0b0110, 0, N3RegFrm,
4666 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4667 "vmax", "u", int_arm_neon_vmaxu, 1>;
4668 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBIND,
4670 v2f32, v2f32, int_arm_neon_vmaxs, 1>;
4671 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBINQ,
4673 v4f32, v4f32, int_arm_neon_vmaxs, 1>;
4675 // VMIN : Vector Minimum
4676 defm VMINs : N3VInt_QHS<0, 0, 0b0110, 1, N3RegFrm,
4677 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4678 "vmin", "s", int_arm_neon_vmins, 1>;
4679 defm VMINu : N3VInt_QHS<1, 0, 0b0110, 1, N3RegFrm,
4680 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4681 "vmin", "u", int_arm_neon_vminu, 1>;
4682 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBIND,
4684 v2f32, v2f32, int_arm_neon_vmins, 1>;
4685 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBINQ,
4687 v4f32, v4f32, int_arm_neon_vmins, 1>;
4689 // Vector Pairwise Operations.
4691 // VPADD : Vector Pairwise Add
4692 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4694 v8i8, v8i8, int_arm_neon_vpadd, 0>;
4695 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4697 v4i16, v4i16, int_arm_neon_vpadd, 0>;
4698 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4700 v2i32, v2i32, int_arm_neon_vpadd, 0>;
4701 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, N3RegFrm,
4702 IIC_VPBIND, "vpadd", "f32",
4703 v2f32, v2f32, int_arm_neon_vpadd, 0>;
4705 // VPADDL : Vector Pairwise Add Long
4706 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
4707 int_arm_neon_vpaddls>;
4708 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
4709 int_arm_neon_vpaddlu>;
4711 // VPADAL : Vector Pairwise Add and Accumulate Long
4712 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
4713 int_arm_neon_vpadals>;
4714 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
4715 int_arm_neon_vpadalu>;
4717 // VPMAX : Vector Pairwise Maximum
4718 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4719 "s8", v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
4720 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4721 "s16", v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
4722 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4723 "s32", v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
4724 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4725 "u8", v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
4726 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4727 "u16", v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
4728 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4729 "u32", v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
4730 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmax",
4731 "f32", v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
4733 // VPMIN : Vector Pairwise Minimum
4734 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4735 "s8", v8i8, v8i8, int_arm_neon_vpmins, 0>;
4736 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4737 "s16", v4i16, v4i16, int_arm_neon_vpmins, 0>;
4738 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4739 "s32", v2i32, v2i32, int_arm_neon_vpmins, 0>;
4740 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4741 "u8", v8i8, v8i8, int_arm_neon_vpminu, 0>;
4742 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4743 "u16", v4i16, v4i16, int_arm_neon_vpminu, 0>;
4744 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4745 "u32", v2i32, v2i32, int_arm_neon_vpminu, 0>;
4746 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmin",
4747 "f32", v2f32, v2f32, int_arm_neon_vpmins, 0>;
4749 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
4751 // VRECPE : Vector Reciprocal Estimate
4752 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
4753 IIC_VUNAD, "vrecpe", "u32",
4754 v2i32, v2i32, int_arm_neon_vrecpe>;
4755 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
4756 IIC_VUNAQ, "vrecpe", "u32",
4757 v4i32, v4i32, int_arm_neon_vrecpe>;
4758 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
4759 IIC_VUNAD, "vrecpe", "f32",
4760 v2f32, v2f32, int_arm_neon_vrecpe>;
4761 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
4762 IIC_VUNAQ, "vrecpe", "f32",
4763 v4f32, v4f32, int_arm_neon_vrecpe>;
4765 // VRECPS : Vector Reciprocal Step
4766 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
4767 IIC_VRECSD, "vrecps", "f32",
4768 v2f32, v2f32, int_arm_neon_vrecps, 1>;
4769 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
4770 IIC_VRECSQ, "vrecps", "f32",
4771 v4f32, v4f32, int_arm_neon_vrecps, 1>;
4773 // VRSQRTE : Vector Reciprocal Square Root Estimate
4774 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
4775 IIC_VUNAD, "vrsqrte", "u32",
4776 v2i32, v2i32, int_arm_neon_vrsqrte>;
4777 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
4778 IIC_VUNAQ, "vrsqrte", "u32",
4779 v4i32, v4i32, int_arm_neon_vrsqrte>;
4780 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
4781 IIC_VUNAD, "vrsqrte", "f32",
4782 v2f32, v2f32, int_arm_neon_vrsqrte>;
4783 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
4784 IIC_VUNAQ, "vrsqrte", "f32",
4785 v4f32, v4f32, int_arm_neon_vrsqrte>;
4787 // VRSQRTS : Vector Reciprocal Square Root Step
4788 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
4789 IIC_VRECSD, "vrsqrts", "f32",
4790 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
4791 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
4792 IIC_VRECSQ, "vrsqrts", "f32",
4793 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
4797 // VSHL : Vector Shift
4798 defm VSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 0, N3RegVShFrm,
4799 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
4800 "vshl", "s", int_arm_neon_vshifts>;
4801 defm VSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 0, N3RegVShFrm,
4802 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
4803 "vshl", "u", int_arm_neon_vshiftu>;
4805 // VSHL : Vector Shift Left (Immediate)
4806 defm VSHLi : N2VShL_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl>;
4808 // VSHR : Vector Shift Right (Immediate)
4809 defm VSHRs : N2VShR_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s", "VSHRs",
4811 defm VSHRu : N2VShR_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u", "VSHRu",
4814 // VSHLL : Vector Shift Left Long
4815 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>;
4816 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>;
4818 // VSHLL : Vector Shift Left Long (with maximum shift count)
4819 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
4820 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
4821 ValueType OpTy, Operand ImmTy, SDNode OpNode>
4822 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
4823 ResTy, OpTy, ImmTy, OpNode> {
4824 let Inst{21-16} = op21_16;
4825 let DecoderMethod = "DecodeVSHLMaxInstruction";
4827 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
4828 v8i16, v8i8, imm8, NEONvshlli>;
4829 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
4830 v4i32, v4i16, imm16, NEONvshlli>;
4831 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
4832 v2i64, v2i32, imm32, NEONvshlli>;
4834 // VSHRN : Vector Shift Right and Narrow
4835 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
4838 // VRSHL : Vector Rounding Shift
4839 defm VRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 0, N3RegVShFrm,
4840 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4841 "vrshl", "s", int_arm_neon_vrshifts>;
4842 defm VRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 0, N3RegVShFrm,
4843 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4844 "vrshl", "u", int_arm_neon_vrshiftu>;
4845 // VRSHR : Vector Rounding Shift Right
4846 defm VRSHRs : N2VShR_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s", "VRSHRs",
4848 defm VRSHRu : N2VShR_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u", "VRSHRu",
4851 // VRSHRN : Vector Rounding Shift Right and Narrow
4852 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
4855 // VQSHL : Vector Saturating Shift
4856 defm VQSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 1, N3RegVShFrm,
4857 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4858 "vqshl", "s", int_arm_neon_vqshifts>;
4859 defm VQSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 1, N3RegVShFrm,
4860 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4861 "vqshl", "u", int_arm_neon_vqshiftu>;
4862 // VQSHL : Vector Saturating Shift Left (Immediate)
4863 defm VQSHLsi : N2VShL_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s",NEONvqshls>;
4864 defm VQSHLui : N2VShL_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u",NEONvqshlu>;
4866 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
4867 defm VQSHLsu : N2VShL_QHSD<1,1,0b0110,1, IIC_VSHLi4D,"vqshlu","s",NEONvqshlsu>;
4869 // VQSHRN : Vector Saturating Shift Right and Narrow
4870 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
4872 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
4875 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
4876 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
4879 // VQRSHL : Vector Saturating Rounding Shift
4880 defm VQRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 1, N3RegVShFrm,
4881 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4882 "vqrshl", "s", int_arm_neon_vqrshifts>;
4883 defm VQRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 1, N3RegVShFrm,
4884 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4885 "vqrshl", "u", int_arm_neon_vqrshiftu>;
4887 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
4888 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
4890 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
4893 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
4894 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
4897 // VSRA : Vector Shift Right and Accumulate
4898 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
4899 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
4900 // VRSRA : Vector Rounding Shift Right and Accumulate
4901 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
4902 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
4904 // VSLI : Vector Shift Left and Insert
4905 defm VSLI : N2VShInsL_QHSD<1, 1, 0b0101, 1, "vsli">;
4907 // VSRI : Vector Shift Right and Insert
4908 defm VSRI : N2VShInsR_QHSD<1, 1, 0b0100, 1, "vsri">;
4910 // Vector Absolute and Saturating Absolute.
4912 // VABS : Vector Absolute Value
4913 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
4914 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
4916 def VABSfd : N2VD<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
4918 v2f32, v2f32, fabs>;
4919 def VABSfq : N2VQ<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
4921 v4f32, v4f32, fabs>;
4923 def : Pat<(v2f32 (int_arm_neon_vabs (v2f32 DPR:$src))), (VABSfd DPR:$src)>;
4924 def : Pat<(v4f32 (int_arm_neon_vabs (v4f32 QPR:$src))), (VABSfq QPR:$src)>;
4926 // VQABS : Vector Saturating Absolute Value
4927 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
4928 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
4929 int_arm_neon_vqabs>;
4933 def vnegd : PatFrag<(ops node:$in),
4934 (sub (bitconvert (v2i32 NEONimmAllZerosV)), node:$in)>;
4935 def vnegq : PatFrag<(ops node:$in),
4936 (sub (bitconvert (v4i32 NEONimmAllZerosV)), node:$in)>;
4938 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
4939 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$Vd), (ins DPR:$Vm),
4940 IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
4941 [(set DPR:$Vd, (Ty (vnegd DPR:$Vm)))]>;
4942 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
4943 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$Vd), (ins QPR:$Vm),
4944 IIC_VSHLiQ, OpcodeStr, Dt, "$Vd, $Vm", "",
4945 [(set QPR:$Vd, (Ty (vnegq QPR:$Vm)))]>;
4947 // VNEG : Vector Negate (integer)
4948 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
4949 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
4950 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
4951 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
4952 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
4953 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
4955 // VNEG : Vector Negate (floating-point)
4956 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
4957 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VUNAD,
4958 "vneg", "f32", "$Vd, $Vm", "",
4959 [(set DPR:$Vd, (v2f32 (fneg DPR:$Vm)))]>;
4960 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
4961 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VUNAQ,
4962 "vneg", "f32", "$Vd, $Vm", "",
4963 [(set QPR:$Vd, (v4f32 (fneg QPR:$Vm)))]>;
4965 def : Pat<(v8i8 (vnegd DPR:$src)), (VNEGs8d DPR:$src)>;
4966 def : Pat<(v4i16 (vnegd DPR:$src)), (VNEGs16d DPR:$src)>;
4967 def : Pat<(v2i32 (vnegd DPR:$src)), (VNEGs32d DPR:$src)>;
4968 def : Pat<(v16i8 (vnegq QPR:$src)), (VNEGs8q QPR:$src)>;
4969 def : Pat<(v8i16 (vnegq QPR:$src)), (VNEGs16q QPR:$src)>;
4970 def : Pat<(v4i32 (vnegq QPR:$src)), (VNEGs32q QPR:$src)>;
4972 // VQNEG : Vector Saturating Negate
4973 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
4974 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
4975 int_arm_neon_vqneg>;
4977 // Vector Bit Counting Operations.
4979 // VCLS : Vector Count Leading Sign Bits
4980 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
4981 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
4983 // VCLZ : Vector Count Leading Zeros
4984 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
4985 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
4987 // VCNT : Vector Count One Bits
4988 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
4989 IIC_VCNTiD, "vcnt", "8",
4991 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
4992 IIC_VCNTiQ, "vcnt", "8",
4993 v16i8, v16i8, ctpop>;
4996 def VSWPd : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 0, 0,
4997 (outs DPR:$Vd, DPR:$Vm), (ins DPR:$in1, DPR:$in2),
4998 NoItinerary, "vswp", "$Vd, $Vm", "$in1 = $Vd, $in2 = $Vm",
5000 def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0,
5001 (outs QPR:$Vd, QPR:$Vm), (ins QPR:$in1, QPR:$in2),
5002 NoItinerary, "vswp", "$Vd, $Vm", "$in1 = $Vd, $in2 = $Vm",
5005 // Vector Move Operations.
5007 // VMOV : Vector Move (Register)
5008 def : InstAlias<"vmov${p} $Vd, $Vm",
5009 (VORRd DPR:$Vd, DPR:$Vm, DPR:$Vm, pred:$p)>;
5010 def : InstAlias<"vmov${p} $Vd, $Vm",
5011 (VORRq QPR:$Vd, QPR:$Vm, QPR:$Vm, pred:$p)>;
5013 // VMOV : Vector Move (Immediate)
5015 let isReMaterializable = 1 in {
5016 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$Vd),
5017 (ins nImmSplatI8:$SIMM), IIC_VMOVImm,
5018 "vmov", "i8", "$Vd, $SIMM", "",
5019 [(set DPR:$Vd, (v8i8 (NEONvmovImm timm:$SIMM)))]>;
5020 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$Vd),
5021 (ins nImmSplatI8:$SIMM), IIC_VMOVImm,
5022 "vmov", "i8", "$Vd, $SIMM", "",
5023 [(set QPR:$Vd, (v16i8 (NEONvmovImm timm:$SIMM)))]>;
5025 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 0, 1, (outs DPR:$Vd),
5026 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
5027 "vmov", "i16", "$Vd, $SIMM", "",
5028 [(set DPR:$Vd, (v4i16 (NEONvmovImm timm:$SIMM)))]> {
5029 let Inst{9} = SIMM{9};
5032 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 0, 1, (outs QPR:$Vd),
5033 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
5034 "vmov", "i16", "$Vd, $SIMM", "",
5035 [(set QPR:$Vd, (v8i16 (NEONvmovImm timm:$SIMM)))]> {
5036 let Inst{9} = SIMM{9};
5039 def VMOVv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 0, 1, (outs DPR:$Vd),
5040 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
5041 "vmov", "i32", "$Vd, $SIMM", "",
5042 [(set DPR:$Vd, (v2i32 (NEONvmovImm timm:$SIMM)))]> {
5043 let Inst{11-8} = SIMM{11-8};
5046 def VMOVv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 0, 1, (outs QPR:$Vd),
5047 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
5048 "vmov", "i32", "$Vd, $SIMM", "",
5049 [(set QPR:$Vd, (v4i32 (NEONvmovImm timm:$SIMM)))]> {
5050 let Inst{11-8} = SIMM{11-8};
5053 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$Vd),
5054 (ins nImmSplatI64:$SIMM), IIC_VMOVImm,
5055 "vmov", "i64", "$Vd, $SIMM", "",
5056 [(set DPR:$Vd, (v1i64 (NEONvmovImm timm:$SIMM)))]>;
5057 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$Vd),
5058 (ins nImmSplatI64:$SIMM), IIC_VMOVImm,
5059 "vmov", "i64", "$Vd, $SIMM", "",
5060 [(set QPR:$Vd, (v2i64 (NEONvmovImm timm:$SIMM)))]>;
5062 def VMOVv2f32 : N1ModImm<1, 0b000, 0b1111, 0, 0, 0, 1, (outs DPR:$Vd),
5063 (ins nImmVMOVF32:$SIMM), IIC_VMOVImm,
5064 "vmov", "f32", "$Vd, $SIMM", "",
5065 [(set DPR:$Vd, (v2f32 (NEONvmovFPImm timm:$SIMM)))]>;
5066 def VMOVv4f32 : N1ModImm<1, 0b000, 0b1111, 0, 1, 0, 1, (outs QPR:$Vd),
5067 (ins nImmVMOVF32:$SIMM), IIC_VMOVImm,
5068 "vmov", "f32", "$Vd, $SIMM", "",
5069 [(set QPR:$Vd, (v4f32 (NEONvmovFPImm timm:$SIMM)))]>;
5070 } // isReMaterializable
5072 // VMOV : Vector Get Lane (move scalar to ARM core register)
5074 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
5075 (outs GPR:$R), (ins DPR:$V, VectorIndex8:$lane),
5076 IIC_VMOVSI, "vmov", "s8", "$R, $V$lane",
5077 [(set GPR:$R, (NEONvgetlanes (v8i8 DPR:$V),
5079 let Inst{21} = lane{2};
5080 let Inst{6-5} = lane{1-0};
5082 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
5083 (outs GPR:$R), (ins DPR:$V, VectorIndex16:$lane),
5084 IIC_VMOVSI, "vmov", "s16", "$R, $V$lane",
5085 [(set GPR:$R, (NEONvgetlanes (v4i16 DPR:$V),
5087 let Inst{21} = lane{1};
5088 let Inst{6} = lane{0};
5090 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
5091 (outs GPR:$R), (ins DPR:$V, VectorIndex8:$lane),
5092 IIC_VMOVSI, "vmov", "u8", "$R, $V$lane",
5093 [(set GPR:$R, (NEONvgetlaneu (v8i8 DPR:$V),
5095 let Inst{21} = lane{2};
5096 let Inst{6-5} = lane{1-0};
5098 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
5099 (outs GPR:$R), (ins DPR:$V, VectorIndex16:$lane),
5100 IIC_VMOVSI, "vmov", "u16", "$R, $V$lane",
5101 [(set GPR:$R, (NEONvgetlaneu (v4i16 DPR:$V),
5103 let Inst{21} = lane{1};
5104 let Inst{6} = lane{0};
5106 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
5107 (outs GPR:$R), (ins DPR:$V, VectorIndex32:$lane),
5108 IIC_VMOVSI, "vmov", "32", "$R, $V$lane",
5109 [(set GPR:$R, (extractelt (v2i32 DPR:$V),
5111 Requires<[HasNEON, HasFastVGETLNi32]> {
5112 let Inst{21} = lane{0};
5114 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
5115 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
5116 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
5117 (DSubReg_i8_reg imm:$lane))),
5118 (SubReg_i8_lane imm:$lane))>;
5119 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
5120 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
5121 (DSubReg_i16_reg imm:$lane))),
5122 (SubReg_i16_lane imm:$lane))>;
5123 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
5124 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
5125 (DSubReg_i8_reg imm:$lane))),
5126 (SubReg_i8_lane imm:$lane))>;
5127 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
5128 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
5129 (DSubReg_i16_reg imm:$lane))),
5130 (SubReg_i16_lane imm:$lane))>;
5131 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
5132 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
5133 (DSubReg_i32_reg imm:$lane))),
5134 (SubReg_i32_lane imm:$lane))>,
5135 Requires<[HasNEON, HasFastVGETLNi32]>;
5136 def : Pat<(extractelt (v2i32 DPR:$src), imm:$lane),
5138 (i32 (EXTRACT_SUBREG DPR:$src, (SSubReg_f32_reg imm:$lane))), GPR)>,
5139 Requires<[HasNEON, HasSlowVGETLNi32]>;
5140 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
5142 (i32 (EXTRACT_SUBREG QPR:$src, (SSubReg_f32_reg imm:$lane))), GPR)>,
5143 Requires<[HasNEON, HasSlowVGETLNi32]>;
5144 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
5145 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
5146 (SSubReg_f32_reg imm:$src2))>;
5147 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
5148 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
5149 (SSubReg_f32_reg imm:$src2))>;
5150 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
5151 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
5152 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
5153 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
5156 // VMOV : Vector Set Lane (move ARM core register to scalar)
5158 let Constraints = "$src1 = $V" in {
5159 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$V),
5160 (ins DPR:$src1, GPR:$R, VectorIndex8:$lane),
5161 IIC_VMOVISL, "vmov", "8", "$V$lane, $R",
5162 [(set DPR:$V, (vector_insert (v8i8 DPR:$src1),
5163 GPR:$R, imm:$lane))]> {
5164 let Inst{21} = lane{2};
5165 let Inst{6-5} = lane{1-0};
5167 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$V),
5168 (ins DPR:$src1, GPR:$R, VectorIndex16:$lane),
5169 IIC_VMOVISL, "vmov", "16", "$V$lane, $R",
5170 [(set DPR:$V, (vector_insert (v4i16 DPR:$src1),
5171 GPR:$R, imm:$lane))]> {
5172 let Inst{21} = lane{1};
5173 let Inst{6} = lane{0};
5175 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$V),
5176 (ins DPR:$src1, GPR:$R, VectorIndex32:$lane),
5177 IIC_VMOVISL, "vmov", "32", "$V$lane, $R",
5178 [(set DPR:$V, (insertelt (v2i32 DPR:$src1),
5179 GPR:$R, imm:$lane))]> {
5180 let Inst{21} = lane{0};
5183 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
5184 (v16i8 (INSERT_SUBREG QPR:$src1,
5185 (v8i8 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
5186 (DSubReg_i8_reg imm:$lane))),
5187 GPR:$src2, (SubReg_i8_lane imm:$lane))),
5188 (DSubReg_i8_reg imm:$lane)))>;
5189 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
5190 (v8i16 (INSERT_SUBREG QPR:$src1,
5191 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
5192 (DSubReg_i16_reg imm:$lane))),
5193 GPR:$src2, (SubReg_i16_lane imm:$lane))),
5194 (DSubReg_i16_reg imm:$lane)))>;
5195 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
5196 (v4i32 (INSERT_SUBREG QPR:$src1,
5197 (v2i32 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
5198 (DSubReg_i32_reg imm:$lane))),
5199 GPR:$src2, (SubReg_i32_lane imm:$lane))),
5200 (DSubReg_i32_reg imm:$lane)))>;
5202 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
5203 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
5204 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
5205 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
5206 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
5207 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
5209 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
5210 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
5211 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
5212 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
5214 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
5215 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
5216 def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
5217 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
5218 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
5219 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
5221 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
5222 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
5223 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
5224 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
5225 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
5226 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
5228 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
5229 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
5230 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
5232 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
5233 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
5234 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
5236 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
5237 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
5238 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
5241 // VDUP : Vector Duplicate (from ARM core register to all elements)
5243 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
5244 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$V), (ins GPR:$R),
5245 IIC_VMOVIS, "vdup", Dt, "$V, $R",
5246 [(set DPR:$V, (Ty (NEONvdup (i32 GPR:$R))))]>;
5247 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
5248 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$V), (ins GPR:$R),
5249 IIC_VMOVIS, "vdup", Dt, "$V, $R",
5250 [(set QPR:$V, (Ty (NEONvdup (i32 GPR:$R))))]>;
5252 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
5253 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
5254 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>,
5255 Requires<[HasNEON, HasFastVDUP32]>;
5256 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
5257 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
5258 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
5260 // NEONvdup patterns for uarchs with fast VDUP.32.
5261 def : Pat<(v2f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VDUP32d GPR:$R)>,
5262 Requires<[HasNEON,HasFastVDUP32]>;
5263 def : Pat<(v4f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VDUP32q GPR:$R)>;
5265 // NEONvdup patterns for uarchs with slow VDUP.32 - use VMOVDRR instead.
5266 def : Pat<(v2i32 (NEONvdup (i32 GPR:$R))), (VMOVDRR GPR:$R, GPR:$R)>,
5267 Requires<[HasNEON,HasSlowVDUP32]>;
5268 def : Pat<(v2f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VMOVDRR GPR:$R, GPR:$R)>,
5269 Requires<[HasNEON,HasSlowVDUP32]>;
5271 // VDUP : Vector Duplicate Lane (from scalar to all elements)
5273 class VDUPLND<bits<4> op19_16, string OpcodeStr, string Dt,
5274 ValueType Ty, Operand IdxTy>
5275 : NVDupLane<op19_16, 0, (outs DPR:$Vd), (ins DPR:$Vm, IdxTy:$lane),
5276 IIC_VMOVD, OpcodeStr, Dt, "$Vd, $Vm$lane",
5277 [(set DPR:$Vd, (Ty (NEONvduplane (Ty DPR:$Vm), imm:$lane)))]>;
5279 class VDUPLNQ<bits<4> op19_16, string OpcodeStr, string Dt,
5280 ValueType ResTy, ValueType OpTy, Operand IdxTy>
5281 : NVDupLane<op19_16, 1, (outs QPR:$Vd), (ins DPR:$Vm, IdxTy:$lane),
5282 IIC_VMOVQ, OpcodeStr, Dt, "$Vd, $Vm$lane",
5283 [(set QPR:$Vd, (ResTy (NEONvduplane (OpTy DPR:$Vm),
5284 VectorIndex32:$lane)))]>;
5286 // Inst{19-16} is partially specified depending on the element size.
5288 def VDUPLN8d : VDUPLND<{?,?,?,1}, "vdup", "8", v8i8, VectorIndex8> {
5290 let Inst{19-17} = lane{2-0};
5292 def VDUPLN16d : VDUPLND<{?,?,1,0}, "vdup", "16", v4i16, VectorIndex16> {
5294 let Inst{19-18} = lane{1-0};
5296 def VDUPLN32d : VDUPLND<{?,1,0,0}, "vdup", "32", v2i32, VectorIndex32> {
5298 let Inst{19} = lane{0};
5300 def VDUPLN8q : VDUPLNQ<{?,?,?,1}, "vdup", "8", v16i8, v8i8, VectorIndex8> {
5302 let Inst{19-17} = lane{2-0};
5304 def VDUPLN16q : VDUPLNQ<{?,?,1,0}, "vdup", "16", v8i16, v4i16, VectorIndex16> {
5306 let Inst{19-18} = lane{1-0};
5308 def VDUPLN32q : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4i32, v2i32, VectorIndex32> {
5310 let Inst{19} = lane{0};
5313 def : Pat<(v2f32 (NEONvduplane (v2f32 DPR:$Vm), imm:$lane)),
5314 (VDUPLN32d DPR:$Vm, imm:$lane)>;
5316 def : Pat<(v4f32 (NEONvduplane (v2f32 DPR:$Vm), imm:$lane)),
5317 (VDUPLN32q DPR:$Vm, imm:$lane)>;
5319 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
5320 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
5321 (DSubReg_i8_reg imm:$lane))),
5322 (SubReg_i8_lane imm:$lane)))>;
5323 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
5324 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
5325 (DSubReg_i16_reg imm:$lane))),
5326 (SubReg_i16_lane imm:$lane)))>;
5327 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
5328 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
5329 (DSubReg_i32_reg imm:$lane))),
5330 (SubReg_i32_lane imm:$lane)))>;
5331 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
5332 (v4f32 (VDUPLN32q (v2f32 (EXTRACT_SUBREG QPR:$src,
5333 (DSubReg_i32_reg imm:$lane))),
5334 (SubReg_i32_lane imm:$lane)))>;
5336 def VDUPfdf : PseudoNeonI<(outs DPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
5337 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
5338 def VDUPfqf : PseudoNeonI<(outs QPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
5339 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
5341 // VMOVN : Vector Narrowing Move
5342 defm VMOVN : N2VN_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVN,
5343 "vmovn", "i", trunc>;
5344 // VQMOVN : Vector Saturating Narrowing Move
5345 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
5346 "vqmovn", "s", int_arm_neon_vqmovns>;
5347 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
5348 "vqmovn", "u", int_arm_neon_vqmovnu>;
5349 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
5350 "vqmovun", "s", int_arm_neon_vqmovnsu>;
5351 // VMOVL : Vector Lengthening Move
5352 defm VMOVLs : N2VL_QHS<0b01,0b10100,0,1, "vmovl", "s", sext>;
5353 defm VMOVLu : N2VL_QHS<0b11,0b10100,0,1, "vmovl", "u", zext>;
5354 def : Pat<(v8i16 (anyext (v8i8 DPR:$Vm))), (VMOVLuv8i16 DPR:$Vm)>;
5355 def : Pat<(v4i32 (anyext (v4i16 DPR:$Vm))), (VMOVLuv4i32 DPR:$Vm)>;
5356 def : Pat<(v2i64 (anyext (v2i32 DPR:$Vm))), (VMOVLuv2i64 DPR:$Vm)>;
5358 // Vector Conversions.
5360 // VCVT : Vector Convert Between Floating-Point and Integers
5361 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
5362 v2i32, v2f32, fp_to_sint>;
5363 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
5364 v2i32, v2f32, fp_to_uint>;
5365 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
5366 v2f32, v2i32, sint_to_fp>;
5367 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
5368 v2f32, v2i32, uint_to_fp>;
5370 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
5371 v4i32, v4f32, fp_to_sint>;
5372 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
5373 v4i32, v4f32, fp_to_uint>;
5374 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
5375 v4f32, v4i32, sint_to_fp>;
5376 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
5377 v4f32, v4i32, uint_to_fp>;
5379 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
5380 let DecoderMethod = "DecodeVCVTD" in {
5381 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
5382 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
5383 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
5384 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
5385 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
5386 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
5387 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
5388 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
5391 let DecoderMethod = "DecodeVCVTQ" in {
5392 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
5393 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
5394 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
5395 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
5396 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
5397 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
5398 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
5399 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
5402 // VCVT : Vector Convert Between Half-Precision and Single-Precision.
5403 def VCVTf2h : N2VNInt<0b11, 0b11, 0b01, 0b10, 0b01100, 0, 0,
5404 IIC_VUNAQ, "vcvt", "f16.f32",
5405 v4i16, v4f32, int_arm_neon_vcvtfp2hf>,
5406 Requires<[HasNEON, HasFP16]>;
5407 def VCVTh2f : N2VLInt<0b11, 0b11, 0b01, 0b10, 0b01110, 0, 0,
5408 IIC_VUNAQ, "vcvt", "f32.f16",
5409 v4f32, v4i16, int_arm_neon_vcvthf2fp>,
5410 Requires<[HasNEON, HasFP16]>;
5414 // VREV64 : Vector Reverse elements within 64-bit doublewords
5416 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5417 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$Vd),
5418 (ins DPR:$Vm), IIC_VMOVD,
5419 OpcodeStr, Dt, "$Vd, $Vm", "",
5420 [(set DPR:$Vd, (Ty (NEONvrev64 (Ty DPR:$Vm))))]>;
5421 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5422 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$Vd),
5423 (ins QPR:$Vm), IIC_VMOVQ,
5424 OpcodeStr, Dt, "$Vd, $Vm", "",
5425 [(set QPR:$Vd, (Ty (NEONvrev64 (Ty QPR:$Vm))))]>;
5427 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
5428 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
5429 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
5430 def : Pat<(v2f32 (NEONvrev64 (v2f32 DPR:$Vm))), (VREV64d32 DPR:$Vm)>;
5432 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
5433 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
5434 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
5435 def : Pat<(v4f32 (NEONvrev64 (v4f32 QPR:$Vm))), (VREV64q32 QPR:$Vm)>;
5437 // VREV32 : Vector Reverse elements within 32-bit words
5439 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5440 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$Vd),
5441 (ins DPR:$Vm), IIC_VMOVD,
5442 OpcodeStr, Dt, "$Vd, $Vm", "",
5443 [(set DPR:$Vd, (Ty (NEONvrev32 (Ty DPR:$Vm))))]>;
5444 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5445 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$Vd),
5446 (ins QPR:$Vm), IIC_VMOVQ,
5447 OpcodeStr, Dt, "$Vd, $Vm", "",
5448 [(set QPR:$Vd, (Ty (NEONvrev32 (Ty QPR:$Vm))))]>;
5450 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
5451 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
5453 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
5454 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
5456 // VREV16 : Vector Reverse elements within 16-bit halfwords
5458 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5459 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$Vd),
5460 (ins DPR:$Vm), IIC_VMOVD,
5461 OpcodeStr, Dt, "$Vd, $Vm", "",
5462 [(set DPR:$Vd, (Ty (NEONvrev16 (Ty DPR:$Vm))))]>;
5463 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5464 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$Vd),
5465 (ins QPR:$Vm), IIC_VMOVQ,
5466 OpcodeStr, Dt, "$Vd, $Vm", "",
5467 [(set QPR:$Vd, (Ty (NEONvrev16 (Ty QPR:$Vm))))]>;
5469 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
5470 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
5472 // Other Vector Shuffles.
5474 // Aligned extractions: really just dropping registers
5476 class AlignedVEXTq<ValueType DestTy, ValueType SrcTy, SDNodeXForm LaneCVT>
5477 : Pat<(DestTy (vector_extract_subvec (SrcTy QPR:$src), (i32 imm:$start))),
5478 (EXTRACT_SUBREG (SrcTy QPR:$src), (LaneCVT imm:$start))>;
5480 def : AlignedVEXTq<v8i8, v16i8, DSubReg_i8_reg>;
5482 def : AlignedVEXTq<v4i16, v8i16, DSubReg_i16_reg>;
5484 def : AlignedVEXTq<v2i32, v4i32, DSubReg_i32_reg>;
5486 def : AlignedVEXTq<v1i64, v2i64, DSubReg_f64_reg>;
5488 def : AlignedVEXTq<v2f32, v4f32, DSubReg_i32_reg>;
5491 // VEXT : Vector Extract
5494 // All of these have a two-operand InstAlias.
5495 let TwoOperandAliasConstraint = "$Vn = $Vd" in {
5496 class VEXTd<string OpcodeStr, string Dt, ValueType Ty, Operand immTy>
5497 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$Vd),
5498 (ins DPR:$Vn, DPR:$Vm, immTy:$index), NVExtFrm,
5499 IIC_VEXTD, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
5500 [(set DPR:$Vd, (Ty (NEONvext (Ty DPR:$Vn),
5501 (Ty DPR:$Vm), imm:$index)))]> {
5503 let Inst{11-8} = index{3-0};
5506 class VEXTq<string OpcodeStr, string Dt, ValueType Ty, Operand immTy>
5507 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$Vd),
5508 (ins QPR:$Vn, QPR:$Vm, imm0_15:$index), NVExtFrm,
5509 IIC_VEXTQ, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
5510 [(set QPR:$Vd, (Ty (NEONvext (Ty QPR:$Vn),
5511 (Ty QPR:$Vm), imm:$index)))]> {
5513 let Inst{11-8} = index{3-0};
5517 def VEXTd8 : VEXTd<"vext", "8", v8i8, imm0_7> {
5518 let Inst{11-8} = index{3-0};
5520 def VEXTd16 : VEXTd<"vext", "16", v4i16, imm0_3> {
5521 let Inst{11-9} = index{2-0};
5524 def VEXTd32 : VEXTd<"vext", "32", v2i32, imm0_1> {
5525 let Inst{11-10} = index{1-0};
5526 let Inst{9-8} = 0b00;
5528 def : Pat<(v2f32 (NEONvext (v2f32 DPR:$Vn),
5531 (VEXTd32 DPR:$Vn, DPR:$Vm, imm:$index)>;
5533 def VEXTq8 : VEXTq<"vext", "8", v16i8, imm0_15> {
5534 let Inst{11-8} = index{3-0};
5536 def VEXTq16 : VEXTq<"vext", "16", v8i16, imm0_7> {
5537 let Inst{11-9} = index{2-0};
5540 def VEXTq32 : VEXTq<"vext", "32", v4i32, imm0_3> {
5541 let Inst{11-10} = index{1-0};
5542 let Inst{9-8} = 0b00;
5544 def VEXTq64 : VEXTq<"vext", "64", v2i64, imm0_1> {
5545 let Inst{11} = index{0};
5546 let Inst{10-8} = 0b000;
5548 def : Pat<(v4f32 (NEONvext (v4f32 QPR:$Vn),
5551 (VEXTq32 QPR:$Vn, QPR:$Vm, imm:$index)>;
5553 // VTRN : Vector Transpose
5555 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
5556 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
5557 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
5559 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
5560 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
5561 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
5563 // VUZP : Vector Unzip (Deinterleave)
5565 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
5566 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
5567 // vuzp.32 Dd, Dm is a pseudo-instruction expanded to vtrn.32 Dd, Dm.
5568 def : NEONInstAlias<"vuzp${p}.32 $Dd, $Dm",
5569 (VTRNd32 DPR:$Dd, DPR:$Dm, pred:$p)>;
5571 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
5572 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
5573 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
5575 // VZIP : Vector Zip (Interleave)
5577 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
5578 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
5579 // vzip.32 Dd, Dm is a pseudo-instruction expanded to vtrn.32 Dd, Dm.
5580 def : NEONInstAlias<"vzip${p}.32 $Dd, $Dm",
5581 (VTRNd32 DPR:$Dd, DPR:$Dm, pred:$p)>;
5583 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
5584 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
5585 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
5587 // Vector Table Lookup and Table Extension.
5589 // VTBL : Vector Table Lookup
5590 let DecoderMethod = "DecodeTBLInstruction" in {
5592 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$Vd),
5593 (ins VecListOneD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB1,
5594 "vtbl", "8", "$Vd, $Vn, $Vm", "",
5595 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbl1 VecListOneD:$Vn, DPR:$Vm)))]>;
5596 let hasExtraSrcRegAllocReq = 1 in {
5598 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$Vd),
5599 (ins VecListDPair:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB2,
5600 "vtbl", "8", "$Vd, $Vn, $Vm", "", []>;
5602 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$Vd),
5603 (ins VecListThreeD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB3,
5604 "vtbl", "8", "$Vd, $Vn, $Vm", "", []>;
5606 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$Vd),
5607 (ins VecListFourD:$Vn, DPR:$Vm),
5609 "vtbl", "8", "$Vd, $Vn, $Vm", "", []>;
5610 } // hasExtraSrcRegAllocReq = 1
5613 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB3, "", []>;
5615 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB4, "", []>;
5617 // VTBX : Vector Table Extension
5619 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$Vd),
5620 (ins DPR:$orig, VecListOneD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX1,
5621 "vtbx", "8", "$Vd, $Vn, $Vm", "$orig = $Vd",
5622 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbx1
5623 DPR:$orig, VecListOneD:$Vn, DPR:$Vm)))]>;
5624 let hasExtraSrcRegAllocReq = 1 in {
5626 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$Vd),
5627 (ins DPR:$orig, VecListDPair:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX2,
5628 "vtbx", "8", "$Vd, $Vn, $Vm", "$orig = $Vd", []>;
5630 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$Vd),
5631 (ins DPR:$orig, VecListThreeD:$Vn, DPR:$Vm),
5632 NVTBLFrm, IIC_VTBX3,
5633 "vtbx", "8", "$Vd, $Vn, $Vm",
5636 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$Vd),
5637 (ins DPR:$orig, VecListFourD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX4,
5638 "vtbx", "8", "$Vd, $Vn, $Vm",
5640 } // hasExtraSrcRegAllocReq = 1
5643 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
5644 IIC_VTBX3, "$orig = $dst", []>;
5646 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
5647 IIC_VTBX4, "$orig = $dst", []>;
5648 } // DecoderMethod = "DecodeTBLInstruction"
5650 //===----------------------------------------------------------------------===//
5651 // NEON instructions for single-precision FP math
5652 //===----------------------------------------------------------------------===//
5654 class N2VSPat<SDNode OpNode, NeonI Inst>
5655 : NEONFPPat<(f32 (OpNode SPR:$a)),
5657 (v2f32 (COPY_TO_REGCLASS (Inst
5659 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5660 SPR:$a, ssub_0)), DPR_VFP2)), ssub_0)>;
5662 class N3VSPat<SDNode OpNode, NeonI Inst>
5663 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
5665 (v2f32 (COPY_TO_REGCLASS (Inst
5667 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5670 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5671 SPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>;
5673 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
5674 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
5676 (v2f32 (COPY_TO_REGCLASS (Inst
5678 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5681 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5684 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5685 SPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>;
5687 def : N3VSPat<fadd, VADDfd>;
5688 def : N3VSPat<fsub, VSUBfd>;
5689 def : N3VSPat<fmul, VMULfd>;
5690 def : N3VSMulOpPat<fmul, fadd, VMLAfd>,
5691 Requires<[HasNEON, UseNEONForFP, UseFPVMLx, DontUseFusedMAC]>;
5692 def : N3VSMulOpPat<fmul, fsub, VMLSfd>,
5693 Requires<[HasNEON, UseNEONForFP, UseFPVMLx, DontUseFusedMAC]>;
5694 def : N3VSMulOpPat<fmul, fadd, VFMAfd>,
5695 Requires<[HasVFP4, UseNEONForFP, UseFusedMAC]>;
5696 def : N3VSMulOpPat<fmul, fsub, VFMSfd>,
5697 Requires<[HasVFP4, UseNEONForFP, UseFusedMAC]>;
5698 def : N2VSPat<fabs, VABSfd>;
5699 def : N2VSPat<fneg, VNEGfd>;
5700 def : N3VSPat<NEONfmax, VMAXfd>;
5701 def : N3VSPat<NEONfmin, VMINfd>;
5702 def : N2VSPat<arm_ftosi, VCVTf2sd>;
5703 def : N2VSPat<arm_ftoui, VCVTf2ud>;
5704 def : N2VSPat<arm_sitof, VCVTs2fd>;
5705 def : N2VSPat<arm_uitof, VCVTu2fd>;
5707 // Prefer VMOVDRR for i32 -> f32 bitcasts, it can write all DPR registers.
5708 def : Pat<(f32 (bitconvert GPR:$a)),
5709 (EXTRACT_SUBREG (VMOVDRR GPR:$a, GPR:$a), ssub_0)>,
5710 Requires<[HasNEON, DontUseVMOVSR]>;
5712 //===----------------------------------------------------------------------===//
5713 // Non-Instruction Patterns
5714 //===----------------------------------------------------------------------===//
5717 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
5718 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
5719 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
5720 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
5721 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
5722 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
5723 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
5724 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
5725 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
5726 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
5727 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
5728 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
5729 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
5730 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
5731 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
5732 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
5733 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
5734 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
5735 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
5736 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
5737 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
5738 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
5739 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
5740 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
5741 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
5742 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
5743 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
5744 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
5745 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
5746 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
5748 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
5749 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
5750 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
5751 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
5752 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
5753 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
5754 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
5755 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
5756 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
5757 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
5758 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
5759 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
5760 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
5761 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
5762 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
5763 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
5764 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
5765 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
5766 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
5767 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
5768 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
5769 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
5770 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
5771 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
5772 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
5773 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
5774 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
5775 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
5776 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
5777 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;
5779 // Vector lengthening move with load, matching extending loads.
5781 // extload, zextload and sextload for a standard lengthening load. Example:
5782 // Lengthen_Single<"8", "i16", "8"> =
5783 // Pat<(v8i16 (extloadvi8 addrmode6:$addr))
5784 // (VMOVLuv8i16 (VLD1d8 addrmode6:$addr,
5785 // (f64 (IMPLICIT_DEF)), (i32 0)))>;
5786 multiclass Lengthen_Single<string DestLanes, string DestTy, string SrcTy> {
5787 let AddedComplexity = 10 in {
5788 def _Any : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5789 (!cast<PatFrag>("extloadvi" # SrcTy) addrmode6:$addr)),
5790 (!cast<Instruction>("VMOVLuv" # DestLanes # DestTy)
5791 (!cast<Instruction>("VLD1d" # SrcTy) addrmode6:$addr))>;
5793 def _Z : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5794 (!cast<PatFrag>("zextloadvi" # SrcTy) addrmode6:$addr)),
5795 (!cast<Instruction>("VMOVLuv" # DestLanes # DestTy)
5796 (!cast<Instruction>("VLD1d" # SrcTy) addrmode6:$addr))>;
5798 def _S : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5799 (!cast<PatFrag>("sextloadvi" # SrcTy) addrmode6:$addr)),
5800 (!cast<Instruction>("VMOVLsv" # DestLanes # DestTy)
5801 (!cast<Instruction>("VLD1d" # SrcTy) addrmode6:$addr))>;
5805 // extload, zextload and sextload for a lengthening load which only uses
5806 // half the lanes available. Example:
5807 // Lengthen_HalfSingle<"4", "i16", "8", "i16", "i8"> =
5808 // Pat<(v4i16 (extloadvi8 addrmode6oneL32:$addr)),
5809 // (EXTRACT_SUBREG (VMOVLuv8i16 (VLD1LNd32 addrmode6oneL32:$addr,
5810 // (f64 (IMPLICIT_DEF)), (i32 0))),
5812 multiclass Lengthen_HalfSingle<string DestLanes, string DestTy, string SrcTy,
5813 string InsnLanes, string InsnTy> {
5814 def _Any : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5815 (!cast<PatFrag>("extloadv" # SrcTy) addrmode6oneL32:$addr)),
5816 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # InsnLanes # InsnTy)
5817 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
5819 def _Z : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5820 (!cast<PatFrag>("zextloadv" # SrcTy) addrmode6oneL32:$addr)),
5821 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # InsnLanes # InsnTy)
5822 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
5824 def _S : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5825 (!cast<PatFrag>("sextloadv" # SrcTy) addrmode6oneL32:$addr)),
5826 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLsv" # InsnLanes # InsnTy)
5827 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
5831 // extload, zextload and sextload for a lengthening load followed by another
5832 // lengthening load, to quadruple the initial length.
5834 // Lengthen_Double<"4", "i32", "i8", "8", "i16", "4", "i32"> =
5835 // Pat<(v4i32 (extloadvi8 addrmode6oneL32:$addr))
5836 // (EXTRACT_SUBREG (VMOVLuv4i32
5837 // (EXTRACT_SUBREG (VMOVLuv8i16 (VLD1LNd32 addrmode6oneL32:$addr,
5838 // (f64 (IMPLICIT_DEF)),
5842 multiclass Lengthen_Double<string DestLanes, string DestTy, string SrcTy,
5843 string Insn1Lanes, string Insn1Ty, string Insn2Lanes,
5845 def _Any : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5846 (!cast<PatFrag>("extloadv" # SrcTy) addrmode6oneL32:$addr)),
5847 (!cast<Instruction>("VMOVLuv" # Insn2Lanes # Insn2Ty)
5848 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn1Lanes # Insn1Ty)
5849 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
5851 def _Z : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5852 (!cast<PatFrag>("zextloadv" # SrcTy) addrmode6oneL32:$addr)),
5853 (!cast<Instruction>("VMOVLuv" # Insn2Lanes # Insn2Ty)
5854 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn1Lanes # Insn1Ty)
5855 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
5857 def _S : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5858 (!cast<PatFrag>("sextloadv" # SrcTy) addrmode6oneL32:$addr)),
5859 (!cast<Instruction>("VMOVLsv" # Insn2Lanes # Insn2Ty)
5860 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLsv" # Insn1Lanes # Insn1Ty)
5861 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
5865 // extload, zextload and sextload for a lengthening load followed by another
5866 // lengthening load, to quadruple the initial length, but which ends up only
5867 // requiring half the available lanes (a 64-bit outcome instead of a 128-bit).
5869 // Lengthen_HalfDouble<"2", "i32", "i8", "8", "i16", "4", "i32"> =
5870 // Pat<(v2i32 (extloadvi8 addrmode6:$addr))
5871 // (EXTRACT_SUBREG (VMOVLuv4i32
5872 // (EXTRACT_SUBREG (VMOVLuv8i16 (VLD1LNd16 addrmode6:$addr,
5873 // (f64 (IMPLICIT_DEF)), (i32 0))),
5876 multiclass Lengthen_HalfDouble<string DestLanes, string DestTy, string SrcTy,
5877 string Insn1Lanes, string Insn1Ty, string Insn2Lanes,
5879 def _Any : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5880 (!cast<PatFrag>("extloadv" # SrcTy) addrmode6:$addr)),
5881 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn2Lanes # Insn2Ty)
5882 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn1Lanes # Insn1Ty)
5883 (VLD1LNd16 addrmode6:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
5886 def _Z : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5887 (!cast<PatFrag>("zextloadv" # SrcTy) addrmode6:$addr)),
5888 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn2Lanes # Insn2Ty)
5889 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn1Lanes # Insn1Ty)
5890 (VLD1LNd16 addrmode6:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
5893 def _S : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5894 (!cast<PatFrag>("sextloadv" # SrcTy) addrmode6:$addr)),
5895 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLsv" # Insn2Lanes # Insn2Ty)
5896 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLsv" # Insn1Lanes # Insn1Ty)
5897 (VLD1LNd16 addrmode6:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
5902 defm : Lengthen_Single<"8", "i16", "8">; // v8i8 -> v8i16
5903 defm : Lengthen_Single<"4", "i32", "16">; // v4i16 -> v4i32
5904 defm : Lengthen_Single<"2", "i64", "32">; // v2i32 -> v2i64
5906 defm : Lengthen_HalfSingle<"4", "i16", "i8", "8", "i16">; // v4i8 -> v4i16
5907 defm : Lengthen_HalfSingle<"2", "i32", "i16", "4", "i32">; // v2i16 -> v2i32
5909 // Double lengthening - v4i8 -> v4i16 -> v4i32
5910 defm : Lengthen_Double<"4", "i32", "i8", "8", "i16", "4", "i32">;
5911 // v2i8 -> v2i16 -> v2i32
5912 defm : Lengthen_HalfDouble<"2", "i32", "i8", "8", "i16", "4", "i32">;
5913 // v2i16 -> v2i32 -> v2i64
5914 defm : Lengthen_Double<"2", "i64", "i16", "4", "i32", "2", "i64">;
5916 // Triple lengthening - v2i8 -> v2i16 -> v2i32 -> v2i64
5917 def : Pat<(v2i64 (extloadvi8 addrmode6:$addr)),
5918 (VMOVLuv2i64 (EXTRACT_SUBREG (VMOVLuv4i32 (EXTRACT_SUBREG (VMOVLuv8i16
5919 (VLD1LNd16 addrmode6:$addr,
5920 (f64 (IMPLICIT_DEF)), (i32 0))), dsub_0)), dsub_0))>;
5921 def : Pat<(v2i64 (zextloadvi8 addrmode6:$addr)),
5922 (VMOVLuv2i64 (EXTRACT_SUBREG (VMOVLuv4i32 (EXTRACT_SUBREG (VMOVLuv8i16
5923 (VLD1LNd16 addrmode6:$addr,
5924 (f64 (IMPLICIT_DEF)), (i32 0))), dsub_0)), dsub_0))>;
5925 def : Pat<(v2i64 (sextloadvi8 addrmode6:$addr)),
5926 (VMOVLsv2i64 (EXTRACT_SUBREG (VMOVLsv4i32 (EXTRACT_SUBREG (VMOVLsv8i16
5927 (VLD1LNd16 addrmode6:$addr,
5928 (f64 (IMPLICIT_DEF)), (i32 0))), dsub_0)), dsub_0))>;
5930 //===----------------------------------------------------------------------===//
5931 // Assembler aliases
5934 def : VFP2InstAlias<"fmdhr${p} $Dd, $Rn",
5935 (VSETLNi32 DPR:$Dd, GPR:$Rn, 1, pred:$p)>;
5936 def : VFP2InstAlias<"fmdlr${p} $Dd, $Rn",
5937 (VSETLNi32 DPR:$Dd, GPR:$Rn, 0, pred:$p)>;
5939 // VAND/VBIC/VEOR/VORR accept but do not require a type suffix.
5940 defm : NEONDTAnyInstAlias<"vand${p}", "$Vd, $Vn, $Vm",
5941 (VANDd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
5942 defm : NEONDTAnyInstAlias<"vand${p}", "$Vd, $Vn, $Vm",
5943 (VANDq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
5944 defm : NEONDTAnyInstAlias<"vbic${p}", "$Vd, $Vn, $Vm",
5945 (VBICd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
5946 defm : NEONDTAnyInstAlias<"vbic${p}", "$Vd, $Vn, $Vm",
5947 (VBICq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
5948 defm : NEONDTAnyInstAlias<"veor${p}", "$Vd, $Vn, $Vm",
5949 (VEORd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
5950 defm : NEONDTAnyInstAlias<"veor${p}", "$Vd, $Vn, $Vm",
5951 (VEORq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
5952 defm : NEONDTAnyInstAlias<"vorr${p}", "$Vd, $Vn, $Vm",
5953 (VORRd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
5954 defm : NEONDTAnyInstAlias<"vorr${p}", "$Vd, $Vn, $Vm",
5955 (VORRq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
5956 // ... two-operand aliases
5957 defm : NEONDTAnyInstAlias<"vand${p}", "$Vdn, $Vm",
5958 (VANDd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5959 defm : NEONDTAnyInstAlias<"vand${p}", "$Vdn, $Vm",
5960 (VANDq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5961 defm : NEONDTAnyInstAlias<"veor${p}", "$Vdn, $Vm",
5962 (VEORd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5963 defm : NEONDTAnyInstAlias<"veor${p}", "$Vdn, $Vm",
5964 (VEORq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5965 defm : NEONDTAnyInstAlias<"vorr${p}", "$Vdn, $Vm",
5966 (VORRd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5967 defm : NEONDTAnyInstAlias<"vorr${p}", "$Vdn, $Vm",
5968 (VORRq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5970 // VLD1 single-lane pseudo-instructions. These need special handling for
5971 // the lane index that an InstAlias can't handle, so we use these instead.
5972 def VLD1LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vld1${p}", ".8", "$list, $addr",
5973 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5974 def VLD1LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vld1${p}", ".16", "$list, $addr",
5975 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5976 def VLD1LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vld1${p}", ".32", "$list, $addr",
5977 (ins VecListOneDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5979 def VLD1LNdWB_fixed_Asm_8 :
5980 NEONDataTypeAsmPseudoInst<"vld1${p}", ".8", "$list, $addr!",
5981 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5982 def VLD1LNdWB_fixed_Asm_16 :
5983 NEONDataTypeAsmPseudoInst<"vld1${p}", ".16", "$list, $addr!",
5984 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5985 def VLD1LNdWB_fixed_Asm_32 :
5986 NEONDataTypeAsmPseudoInst<"vld1${p}", ".32", "$list, $addr!",
5987 (ins VecListOneDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5988 def VLD1LNdWB_register_Asm_8 :
5989 NEONDataTypeAsmPseudoInst<"vld1${p}", ".8", "$list, $addr, $Rm",
5990 (ins VecListOneDByteIndexed:$list, addrmode6:$addr,
5991 rGPR:$Rm, pred:$p)>;
5992 def VLD1LNdWB_register_Asm_16 :
5993 NEONDataTypeAsmPseudoInst<"vld1${p}", ".16", "$list, $addr, $Rm",
5994 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr,
5995 rGPR:$Rm, pred:$p)>;
5996 def VLD1LNdWB_register_Asm_32 :
5997 NEONDataTypeAsmPseudoInst<"vld1${p}", ".32", "$list, $addr, $Rm",
5998 (ins VecListOneDWordIndexed:$list, addrmode6:$addr,
5999 rGPR:$Rm, pred:$p)>;
6002 // VST1 single-lane pseudo-instructions. These need special handling for
6003 // the lane index that an InstAlias can't handle, so we use these instead.
6004 def VST1LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vst1${p}", ".8", "$list, $addr",
6005 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6006 def VST1LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vst1${p}", ".16", "$list, $addr",
6007 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6008 def VST1LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vst1${p}", ".32", "$list, $addr",
6009 (ins VecListOneDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6011 def VST1LNdWB_fixed_Asm_8 :
6012 NEONDataTypeAsmPseudoInst<"vst1${p}", ".8", "$list, $addr!",
6013 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6014 def VST1LNdWB_fixed_Asm_16 :
6015 NEONDataTypeAsmPseudoInst<"vst1${p}", ".16", "$list, $addr!",
6016 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6017 def VST1LNdWB_fixed_Asm_32 :
6018 NEONDataTypeAsmPseudoInst<"vst1${p}", ".32", "$list, $addr!",
6019 (ins VecListOneDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6020 def VST1LNdWB_register_Asm_8 :
6021 NEONDataTypeAsmPseudoInst<"vst1${p}", ".8", "$list, $addr, $Rm",
6022 (ins VecListOneDByteIndexed:$list, addrmode6:$addr,
6023 rGPR:$Rm, pred:$p)>;
6024 def VST1LNdWB_register_Asm_16 :
6025 NEONDataTypeAsmPseudoInst<"vst1${p}", ".16", "$list, $addr, $Rm",
6026 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr,
6027 rGPR:$Rm, pred:$p)>;
6028 def VST1LNdWB_register_Asm_32 :
6029 NEONDataTypeAsmPseudoInst<"vst1${p}", ".32", "$list, $addr, $Rm",
6030 (ins VecListOneDWordIndexed:$list, addrmode6:$addr,
6031 rGPR:$Rm, pred:$p)>;
6033 // VLD2 single-lane pseudo-instructions. These need special handling for
6034 // the lane index that an InstAlias can't handle, so we use these instead.
6035 def VLD2LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vld2${p}", ".8", "$list, $addr",
6036 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6037 def VLD2LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr",
6038 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6039 def VLD2LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr",
6040 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6041 def VLD2LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr",
6042 (ins VecListTwoQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6043 def VLD2LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr",
6044 (ins VecListTwoQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6046 def VLD2LNdWB_fixed_Asm_8 :
6047 NEONDataTypeAsmPseudoInst<"vld2${p}", ".8", "$list, $addr!",
6048 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6049 def VLD2LNdWB_fixed_Asm_16 :
6050 NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr!",
6051 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6052 def VLD2LNdWB_fixed_Asm_32 :
6053 NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr!",
6054 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6055 def VLD2LNqWB_fixed_Asm_16 :
6056 NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr!",
6057 (ins VecListTwoQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6058 def VLD2LNqWB_fixed_Asm_32 :
6059 NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr!",
6060 (ins VecListTwoQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6061 def VLD2LNdWB_register_Asm_8 :
6062 NEONDataTypeAsmPseudoInst<"vld2${p}", ".8", "$list, $addr, $Rm",
6063 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr,
6064 rGPR:$Rm, pred:$p)>;
6065 def VLD2LNdWB_register_Asm_16 :
6066 NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr, $Rm",
6067 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr,
6068 rGPR:$Rm, pred:$p)>;
6069 def VLD2LNdWB_register_Asm_32 :
6070 NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr, $Rm",
6071 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr,
6072 rGPR:$Rm, pred:$p)>;
6073 def VLD2LNqWB_register_Asm_16 :
6074 NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr, $Rm",
6075 (ins VecListTwoQHWordIndexed:$list, addrmode6:$addr,
6076 rGPR:$Rm, pred:$p)>;
6077 def VLD2LNqWB_register_Asm_32 :
6078 NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr, $Rm",
6079 (ins VecListTwoQWordIndexed:$list, addrmode6:$addr,
6080 rGPR:$Rm, pred:$p)>;
6083 // VST2 single-lane pseudo-instructions. These need special handling for
6084 // the lane index that an InstAlias can't handle, so we use these instead.
6085 def VST2LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vst2${p}", ".8", "$list, $addr",
6086 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6087 def VST2LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vst2${p}", ".16", "$list, $addr",
6088 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6089 def VST2LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr",
6090 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6091 def VST2LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vst2${p}", ".16", "$list, $addr",
6092 (ins VecListTwoQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6093 def VST2LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr",
6094 (ins VecListTwoQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6096 def VST2LNdWB_fixed_Asm_8 :
6097 NEONDataTypeAsmPseudoInst<"vst2${p}", ".8", "$list, $addr!",
6098 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6099 def VST2LNdWB_fixed_Asm_16 :
6100 NEONDataTypeAsmPseudoInst<"vst2${p}", ".16", "$list, $addr!",
6101 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6102 def VST2LNdWB_fixed_Asm_32 :
6103 NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr!",
6104 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6105 def VST2LNqWB_fixed_Asm_16 :
6106 NEONDataTypeAsmPseudoInst<"vst2${p}", ".16", "$list, $addr!",
6107 (ins VecListTwoQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6108 def VST2LNqWB_fixed_Asm_32 :
6109 NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr!",
6110 (ins VecListTwoQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6111 def VST2LNdWB_register_Asm_8 :
6112 NEONDataTypeAsmPseudoInst<"vst2${p}", ".8", "$list, $addr, $Rm",
6113 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr,
6114 rGPR:$Rm, pred:$p)>;
6115 def VST2LNdWB_register_Asm_16 :
6116 NEONDataTypeAsmPseudoInst<"vst2${p}", ".16","$list, $addr, $Rm",
6117 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr,
6118 rGPR:$Rm, pred:$p)>;
6119 def VST2LNdWB_register_Asm_32 :
6120 NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr, $Rm",
6121 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr,
6122 rGPR:$Rm, pred:$p)>;
6123 def VST2LNqWB_register_Asm_16 :
6124 NEONDataTypeAsmPseudoInst<"vst2${p}", ".16","$list, $addr, $Rm",
6125 (ins VecListTwoQHWordIndexed:$list, addrmode6:$addr,
6126 rGPR:$Rm, pred:$p)>;
6127 def VST2LNqWB_register_Asm_32 :
6128 NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr, $Rm",
6129 (ins VecListTwoQWordIndexed:$list, addrmode6:$addr,
6130 rGPR:$Rm, pred:$p)>;
6132 // VLD3 all-lanes pseudo-instructions. These need special handling for
6133 // the lane index that an InstAlias can't handle, so we use these instead.
6134 def VLD3DUPdAsm_8 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr",
6135 (ins VecListThreeDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6136 def VLD3DUPdAsm_16: NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
6137 (ins VecListThreeDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6138 def VLD3DUPdAsm_32: NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
6139 (ins VecListThreeDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6140 def VLD3DUPqAsm_8 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr",
6141 (ins VecListThreeQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6142 def VLD3DUPqAsm_16: NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
6143 (ins VecListThreeQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6144 def VLD3DUPqAsm_32: NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
6145 (ins VecListThreeQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6147 def VLD3DUPdWB_fixed_Asm_8 :
6148 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr!",
6149 (ins VecListThreeDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6150 def VLD3DUPdWB_fixed_Asm_16 :
6151 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
6152 (ins VecListThreeDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6153 def VLD3DUPdWB_fixed_Asm_32 :
6154 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
6155 (ins VecListThreeDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6156 def VLD3DUPqWB_fixed_Asm_8 :
6157 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr!",
6158 (ins VecListThreeQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6159 def VLD3DUPqWB_fixed_Asm_16 :
6160 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
6161 (ins VecListThreeQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6162 def VLD3DUPqWB_fixed_Asm_32 :
6163 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
6164 (ins VecListThreeQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6165 def VLD3DUPdWB_register_Asm_8 :
6166 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr, $Rm",
6167 (ins VecListThreeDAllLanes:$list, addrmode6:$addr,
6168 rGPR:$Rm, pred:$p)>;
6169 def VLD3DUPdWB_register_Asm_16 :
6170 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
6171 (ins VecListThreeDAllLanes:$list, addrmode6:$addr,
6172 rGPR:$Rm, pred:$p)>;
6173 def VLD3DUPdWB_register_Asm_32 :
6174 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
6175 (ins VecListThreeDAllLanes:$list, addrmode6:$addr,
6176 rGPR:$Rm, pred:$p)>;
6177 def VLD3DUPqWB_register_Asm_8 :
6178 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr, $Rm",
6179 (ins VecListThreeQAllLanes:$list, addrmode6:$addr,
6180 rGPR:$Rm, pred:$p)>;
6181 def VLD3DUPqWB_register_Asm_16 :
6182 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
6183 (ins VecListThreeQAllLanes:$list, addrmode6:$addr,
6184 rGPR:$Rm, pred:$p)>;
6185 def VLD3DUPqWB_register_Asm_32 :
6186 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
6187 (ins VecListThreeQAllLanes:$list, addrmode6:$addr,
6188 rGPR:$Rm, pred:$p)>;
6191 // VLD3 single-lane pseudo-instructions. These need special handling for
6192 // the lane index that an InstAlias can't handle, so we use these instead.
6193 def VLD3LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr",
6194 (ins VecListThreeDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6195 def VLD3LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
6196 (ins VecListThreeDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6197 def VLD3LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
6198 (ins VecListThreeDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6199 def VLD3LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
6200 (ins VecListThreeQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6201 def VLD3LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
6202 (ins VecListThreeQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6204 def VLD3LNdWB_fixed_Asm_8 :
6205 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr!",
6206 (ins VecListThreeDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6207 def VLD3LNdWB_fixed_Asm_16 :
6208 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
6209 (ins VecListThreeDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6210 def VLD3LNdWB_fixed_Asm_32 :
6211 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
6212 (ins VecListThreeDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6213 def VLD3LNqWB_fixed_Asm_16 :
6214 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
6215 (ins VecListThreeQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6216 def VLD3LNqWB_fixed_Asm_32 :
6217 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
6218 (ins VecListThreeQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6219 def VLD3LNdWB_register_Asm_8 :
6220 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr, $Rm",
6221 (ins VecListThreeDByteIndexed:$list, addrmode6:$addr,
6222 rGPR:$Rm, pred:$p)>;
6223 def VLD3LNdWB_register_Asm_16 :
6224 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
6225 (ins VecListThreeDHWordIndexed:$list, addrmode6:$addr,
6226 rGPR:$Rm, pred:$p)>;
6227 def VLD3LNdWB_register_Asm_32 :
6228 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
6229 (ins VecListThreeDWordIndexed:$list, addrmode6:$addr,
6230 rGPR:$Rm, pred:$p)>;
6231 def VLD3LNqWB_register_Asm_16 :
6232 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
6233 (ins VecListThreeQHWordIndexed:$list, addrmode6:$addr,
6234 rGPR:$Rm, pred:$p)>;
6235 def VLD3LNqWB_register_Asm_32 :
6236 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
6237 (ins VecListThreeQWordIndexed:$list, addrmode6:$addr,
6238 rGPR:$Rm, pred:$p)>;
6240 // VLD3 multiple structure pseudo-instructions. These need special handling for
6241 // the vector operands that the normal instructions don't yet model.
6242 // FIXME: Remove these when the register classes and instructions are updated.
6243 def VLD3dAsm_8 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr",
6244 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6245 def VLD3dAsm_16 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
6246 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6247 def VLD3dAsm_32 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
6248 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6249 def VLD3qAsm_8 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr",
6250 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6251 def VLD3qAsm_16 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
6252 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6253 def VLD3qAsm_32 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
6254 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6256 def VLD3dWB_fixed_Asm_8 :
6257 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr!",
6258 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6259 def VLD3dWB_fixed_Asm_16 :
6260 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
6261 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6262 def VLD3dWB_fixed_Asm_32 :
6263 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
6264 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6265 def VLD3qWB_fixed_Asm_8 :
6266 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr!",
6267 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6268 def VLD3qWB_fixed_Asm_16 :
6269 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
6270 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6271 def VLD3qWB_fixed_Asm_32 :
6272 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
6273 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6274 def VLD3dWB_register_Asm_8 :
6275 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr, $Rm",
6276 (ins VecListThreeD:$list, addrmode6:$addr,
6277 rGPR:$Rm, pred:$p)>;
6278 def VLD3dWB_register_Asm_16 :
6279 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
6280 (ins VecListThreeD:$list, addrmode6:$addr,
6281 rGPR:$Rm, pred:$p)>;
6282 def VLD3dWB_register_Asm_32 :
6283 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
6284 (ins VecListThreeD:$list, addrmode6:$addr,
6285 rGPR:$Rm, pred:$p)>;
6286 def VLD3qWB_register_Asm_8 :
6287 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr, $Rm",
6288 (ins VecListThreeQ:$list, addrmode6:$addr,
6289 rGPR:$Rm, pred:$p)>;
6290 def VLD3qWB_register_Asm_16 :
6291 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
6292 (ins VecListThreeQ:$list, addrmode6:$addr,
6293 rGPR:$Rm, pred:$p)>;
6294 def VLD3qWB_register_Asm_32 :
6295 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
6296 (ins VecListThreeQ:$list, addrmode6:$addr,
6297 rGPR:$Rm, pred:$p)>;
6299 // VST3 single-lane pseudo-instructions. These need special handling for
6300 // the lane index that an InstAlias can't handle, so we use these instead.
6301 def VST3LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr",
6302 (ins VecListThreeDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6303 def VST3LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr",
6304 (ins VecListThreeDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6305 def VST3LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr",
6306 (ins VecListThreeDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6307 def VST3LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr",
6308 (ins VecListThreeQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6309 def VST3LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr",
6310 (ins VecListThreeQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6312 def VST3LNdWB_fixed_Asm_8 :
6313 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr!",
6314 (ins VecListThreeDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6315 def VST3LNdWB_fixed_Asm_16 :
6316 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr!",
6317 (ins VecListThreeDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6318 def VST3LNdWB_fixed_Asm_32 :
6319 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr!",
6320 (ins VecListThreeDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6321 def VST3LNqWB_fixed_Asm_16 :
6322 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr!",
6323 (ins VecListThreeQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6324 def VST3LNqWB_fixed_Asm_32 :
6325 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr!",
6326 (ins VecListThreeQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6327 def VST3LNdWB_register_Asm_8 :
6328 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr, $Rm",
6329 (ins VecListThreeDByteIndexed:$list, addrmode6:$addr,
6330 rGPR:$Rm, pred:$p)>;
6331 def VST3LNdWB_register_Asm_16 :
6332 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr, $Rm",
6333 (ins VecListThreeDHWordIndexed:$list, addrmode6:$addr,
6334 rGPR:$Rm, pred:$p)>;
6335 def VST3LNdWB_register_Asm_32 :
6336 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr, $Rm",
6337 (ins VecListThreeDWordIndexed:$list, addrmode6:$addr,
6338 rGPR:$Rm, pred:$p)>;
6339 def VST3LNqWB_register_Asm_16 :
6340 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr, $Rm",
6341 (ins VecListThreeQHWordIndexed:$list, addrmode6:$addr,
6342 rGPR:$Rm, pred:$p)>;
6343 def VST3LNqWB_register_Asm_32 :
6344 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr, $Rm",
6345 (ins VecListThreeQWordIndexed:$list, addrmode6:$addr,
6346 rGPR:$Rm, pred:$p)>;
6349 // VST3 multiple structure pseudo-instructions. These need special handling for
6350 // the vector operands that the normal instructions don't yet model.
6351 // FIXME: Remove these when the register classes and instructions are updated.
6352 def VST3dAsm_8 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr",
6353 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6354 def VST3dAsm_16 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr",
6355 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6356 def VST3dAsm_32 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr",
6357 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6358 def VST3qAsm_8 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr",
6359 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6360 def VST3qAsm_16 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr",
6361 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6362 def VST3qAsm_32 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr",
6363 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6365 def VST3dWB_fixed_Asm_8 :
6366 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr!",
6367 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6368 def VST3dWB_fixed_Asm_16 :
6369 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr!",
6370 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6371 def VST3dWB_fixed_Asm_32 :
6372 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr!",
6373 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6374 def VST3qWB_fixed_Asm_8 :
6375 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr!",
6376 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6377 def VST3qWB_fixed_Asm_16 :
6378 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr!",
6379 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6380 def VST3qWB_fixed_Asm_32 :
6381 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr!",
6382 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6383 def VST3dWB_register_Asm_8 :
6384 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr, $Rm",
6385 (ins VecListThreeD:$list, addrmode6:$addr,
6386 rGPR:$Rm, pred:$p)>;
6387 def VST3dWB_register_Asm_16 :
6388 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr, $Rm",
6389 (ins VecListThreeD:$list, addrmode6:$addr,
6390 rGPR:$Rm, pred:$p)>;
6391 def VST3dWB_register_Asm_32 :
6392 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr, $Rm",
6393 (ins VecListThreeD:$list, addrmode6:$addr,
6394 rGPR:$Rm, pred:$p)>;
6395 def VST3qWB_register_Asm_8 :
6396 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr, $Rm",
6397 (ins VecListThreeQ:$list, addrmode6:$addr,
6398 rGPR:$Rm, pred:$p)>;
6399 def VST3qWB_register_Asm_16 :
6400 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr, $Rm",
6401 (ins VecListThreeQ:$list, addrmode6:$addr,
6402 rGPR:$Rm, pred:$p)>;
6403 def VST3qWB_register_Asm_32 :
6404 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr, $Rm",
6405 (ins VecListThreeQ:$list, addrmode6:$addr,
6406 rGPR:$Rm, pred:$p)>;
6408 // VLD4 all-lanes pseudo-instructions. These need special handling for
6409 // the lane index that an InstAlias can't handle, so we use these instead.
6410 def VLD4DUPdAsm_8 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr",
6411 (ins VecListFourDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6412 def VLD4DUPdAsm_16: NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
6413 (ins VecListFourDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6414 def VLD4DUPdAsm_32: NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
6415 (ins VecListFourDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6416 def VLD4DUPqAsm_8 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr",
6417 (ins VecListFourQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6418 def VLD4DUPqAsm_16: NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
6419 (ins VecListFourQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6420 def VLD4DUPqAsm_32: NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
6421 (ins VecListFourQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6423 def VLD4DUPdWB_fixed_Asm_8 :
6424 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr!",
6425 (ins VecListFourDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6426 def VLD4DUPdWB_fixed_Asm_16 :
6427 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
6428 (ins VecListFourDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6429 def VLD4DUPdWB_fixed_Asm_32 :
6430 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
6431 (ins VecListFourDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6432 def VLD4DUPqWB_fixed_Asm_8 :
6433 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr!",
6434 (ins VecListFourQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6435 def VLD4DUPqWB_fixed_Asm_16 :
6436 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
6437 (ins VecListFourQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6438 def VLD4DUPqWB_fixed_Asm_32 :
6439 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
6440 (ins VecListFourQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6441 def VLD4DUPdWB_register_Asm_8 :
6442 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr, $Rm",
6443 (ins VecListFourDAllLanes:$list, addrmode6:$addr,
6444 rGPR:$Rm, pred:$p)>;
6445 def VLD4DUPdWB_register_Asm_16 :
6446 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
6447 (ins VecListFourDAllLanes:$list, addrmode6:$addr,
6448 rGPR:$Rm, pred:$p)>;
6449 def VLD4DUPdWB_register_Asm_32 :
6450 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
6451 (ins VecListFourDAllLanes:$list, addrmode6:$addr,
6452 rGPR:$Rm, pred:$p)>;
6453 def VLD4DUPqWB_register_Asm_8 :
6454 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr, $Rm",
6455 (ins VecListFourQAllLanes:$list, addrmode6:$addr,
6456 rGPR:$Rm, pred:$p)>;
6457 def VLD4DUPqWB_register_Asm_16 :
6458 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
6459 (ins VecListFourQAllLanes:$list, addrmode6:$addr,
6460 rGPR:$Rm, pred:$p)>;
6461 def VLD4DUPqWB_register_Asm_32 :
6462 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
6463 (ins VecListFourQAllLanes:$list, addrmode6:$addr,
6464 rGPR:$Rm, pred:$p)>;
6467 // VLD4 single-lane pseudo-instructions. These need special handling for
6468 // the lane index that an InstAlias can't handle, so we use these instead.
6469 def VLD4LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr",
6470 (ins VecListFourDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6471 def VLD4LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
6472 (ins VecListFourDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6473 def VLD4LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
6474 (ins VecListFourDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6475 def VLD4LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
6476 (ins VecListFourQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6477 def VLD4LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
6478 (ins VecListFourQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6480 def VLD4LNdWB_fixed_Asm_8 :
6481 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr!",
6482 (ins VecListFourDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6483 def VLD4LNdWB_fixed_Asm_16 :
6484 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
6485 (ins VecListFourDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6486 def VLD4LNdWB_fixed_Asm_32 :
6487 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
6488 (ins VecListFourDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6489 def VLD4LNqWB_fixed_Asm_16 :
6490 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
6491 (ins VecListFourQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6492 def VLD4LNqWB_fixed_Asm_32 :
6493 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
6494 (ins VecListFourQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6495 def VLD4LNdWB_register_Asm_8 :
6496 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr, $Rm",
6497 (ins VecListFourDByteIndexed:$list, addrmode6:$addr,
6498 rGPR:$Rm, pred:$p)>;
6499 def VLD4LNdWB_register_Asm_16 :
6500 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
6501 (ins VecListFourDHWordIndexed:$list, addrmode6:$addr,
6502 rGPR:$Rm, pred:$p)>;
6503 def VLD4LNdWB_register_Asm_32 :
6504 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
6505 (ins VecListFourDWordIndexed:$list, addrmode6:$addr,
6506 rGPR:$Rm, pred:$p)>;
6507 def VLD4LNqWB_register_Asm_16 :
6508 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
6509 (ins VecListFourQHWordIndexed:$list, addrmode6:$addr,
6510 rGPR:$Rm, pred:$p)>;
6511 def VLD4LNqWB_register_Asm_32 :
6512 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
6513 (ins VecListFourQWordIndexed:$list, addrmode6:$addr,
6514 rGPR:$Rm, pred:$p)>;
6518 // VLD4 multiple structure pseudo-instructions. These need special handling for
6519 // the vector operands that the normal instructions don't yet model.
6520 // FIXME: Remove these when the register classes and instructions are updated.
6521 def VLD4dAsm_8 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr",
6522 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6523 def VLD4dAsm_16 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
6524 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6525 def VLD4dAsm_32 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
6526 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6527 def VLD4qAsm_8 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr",
6528 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6529 def VLD4qAsm_16 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
6530 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6531 def VLD4qAsm_32 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
6532 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6534 def VLD4dWB_fixed_Asm_8 :
6535 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr!",
6536 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6537 def VLD4dWB_fixed_Asm_16 :
6538 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
6539 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6540 def VLD4dWB_fixed_Asm_32 :
6541 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
6542 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6543 def VLD4qWB_fixed_Asm_8 :
6544 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr!",
6545 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6546 def VLD4qWB_fixed_Asm_16 :
6547 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
6548 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6549 def VLD4qWB_fixed_Asm_32 :
6550 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
6551 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6552 def VLD4dWB_register_Asm_8 :
6553 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr, $Rm",
6554 (ins VecListFourD:$list, addrmode6:$addr,
6555 rGPR:$Rm, pred:$p)>;
6556 def VLD4dWB_register_Asm_16 :
6557 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
6558 (ins VecListFourD:$list, addrmode6:$addr,
6559 rGPR:$Rm, pred:$p)>;
6560 def VLD4dWB_register_Asm_32 :
6561 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
6562 (ins VecListFourD:$list, addrmode6:$addr,
6563 rGPR:$Rm, pred:$p)>;
6564 def VLD4qWB_register_Asm_8 :
6565 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr, $Rm",
6566 (ins VecListFourQ:$list, addrmode6:$addr,
6567 rGPR:$Rm, pred:$p)>;
6568 def VLD4qWB_register_Asm_16 :
6569 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
6570 (ins VecListFourQ:$list, addrmode6:$addr,
6571 rGPR:$Rm, pred:$p)>;
6572 def VLD4qWB_register_Asm_32 :
6573 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
6574 (ins VecListFourQ:$list, addrmode6:$addr,
6575 rGPR:$Rm, pred:$p)>;
6577 // VST4 single-lane pseudo-instructions. These need special handling for
6578 // the lane index that an InstAlias can't handle, so we use these instead.
6579 def VST4LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr",
6580 (ins VecListFourDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6581 def VST4LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr",
6582 (ins VecListFourDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6583 def VST4LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr",
6584 (ins VecListFourDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6585 def VST4LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr",
6586 (ins VecListFourQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6587 def VST4LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr",
6588 (ins VecListFourQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6590 def VST4LNdWB_fixed_Asm_8 :
6591 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr!",
6592 (ins VecListFourDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6593 def VST4LNdWB_fixed_Asm_16 :
6594 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr!",
6595 (ins VecListFourDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6596 def VST4LNdWB_fixed_Asm_32 :
6597 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr!",
6598 (ins VecListFourDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6599 def VST4LNqWB_fixed_Asm_16 :
6600 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr!",
6601 (ins VecListFourQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6602 def VST4LNqWB_fixed_Asm_32 :
6603 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr!",
6604 (ins VecListFourQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6605 def VST4LNdWB_register_Asm_8 :
6606 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr, $Rm",
6607 (ins VecListFourDByteIndexed:$list, addrmode6:$addr,
6608 rGPR:$Rm, pred:$p)>;
6609 def VST4LNdWB_register_Asm_16 :
6610 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr, $Rm",
6611 (ins VecListFourDHWordIndexed:$list, addrmode6:$addr,
6612 rGPR:$Rm, pred:$p)>;
6613 def VST4LNdWB_register_Asm_32 :
6614 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr, $Rm",
6615 (ins VecListFourDWordIndexed:$list, addrmode6:$addr,
6616 rGPR:$Rm, pred:$p)>;
6617 def VST4LNqWB_register_Asm_16 :
6618 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr, $Rm",
6619 (ins VecListFourQHWordIndexed:$list, addrmode6:$addr,
6620 rGPR:$Rm, pred:$p)>;
6621 def VST4LNqWB_register_Asm_32 :
6622 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr, $Rm",
6623 (ins VecListFourQWordIndexed:$list, addrmode6:$addr,
6624 rGPR:$Rm, pred:$p)>;
6627 // VST4 multiple structure pseudo-instructions. These need special handling for
6628 // the vector operands that the normal instructions don't yet model.
6629 // FIXME: Remove these when the register classes and instructions are updated.
6630 def VST4dAsm_8 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr",
6631 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6632 def VST4dAsm_16 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr",
6633 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6634 def VST4dAsm_32 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr",
6635 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6636 def VST4qAsm_8 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr",
6637 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6638 def VST4qAsm_16 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr",
6639 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6640 def VST4qAsm_32 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr",
6641 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6643 def VST4dWB_fixed_Asm_8 :
6644 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr!",
6645 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6646 def VST4dWB_fixed_Asm_16 :
6647 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr!",
6648 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6649 def VST4dWB_fixed_Asm_32 :
6650 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr!",
6651 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6652 def VST4qWB_fixed_Asm_8 :
6653 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr!",
6654 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6655 def VST4qWB_fixed_Asm_16 :
6656 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr!",
6657 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6658 def VST4qWB_fixed_Asm_32 :
6659 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr!",
6660 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6661 def VST4dWB_register_Asm_8 :
6662 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr, $Rm",
6663 (ins VecListFourD:$list, addrmode6:$addr,
6664 rGPR:$Rm, pred:$p)>;
6665 def VST4dWB_register_Asm_16 :
6666 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr, $Rm",
6667 (ins VecListFourD:$list, addrmode6:$addr,
6668 rGPR:$Rm, pred:$p)>;
6669 def VST4dWB_register_Asm_32 :
6670 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr, $Rm",
6671 (ins VecListFourD:$list, addrmode6:$addr,
6672 rGPR:$Rm, pred:$p)>;
6673 def VST4qWB_register_Asm_8 :
6674 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr, $Rm",
6675 (ins VecListFourQ:$list, addrmode6:$addr,
6676 rGPR:$Rm, pred:$p)>;
6677 def VST4qWB_register_Asm_16 :
6678 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr, $Rm",
6679 (ins VecListFourQ:$list, addrmode6:$addr,
6680 rGPR:$Rm, pred:$p)>;
6681 def VST4qWB_register_Asm_32 :
6682 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr, $Rm",
6683 (ins VecListFourQ:$list, addrmode6:$addr,
6684 rGPR:$Rm, pred:$p)>;
6686 // VMOV takes an optional datatype suffix
6687 defm : NEONDTAnyInstAlias<"vmov${p}", "$Vd, $Vm",
6688 (VORRd DPR:$Vd, DPR:$Vm, DPR:$Vm, pred:$p)>;
6689 defm : NEONDTAnyInstAlias<"vmov${p}", "$Vd, $Vm",
6690 (VORRq QPR:$Vd, QPR:$Vm, QPR:$Vm, pred:$p)>;
6692 // VCLT (register) is an assembler alias for VCGT w/ the operands reversed.
6693 // D-register versions.
6694 def : NEONInstAlias<"vcle${p}.s8 $Dd, $Dn, $Dm",
6695 (VCGEsv8i8 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6696 def : NEONInstAlias<"vcle${p}.s16 $Dd, $Dn, $Dm",
6697 (VCGEsv4i16 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6698 def : NEONInstAlias<"vcle${p}.s32 $Dd, $Dn, $Dm",
6699 (VCGEsv2i32 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6700 def : NEONInstAlias<"vcle${p}.u8 $Dd, $Dn, $Dm",
6701 (VCGEuv8i8 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6702 def : NEONInstAlias<"vcle${p}.u16 $Dd, $Dn, $Dm",
6703 (VCGEuv4i16 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6704 def : NEONInstAlias<"vcle${p}.u32 $Dd, $Dn, $Dm",
6705 (VCGEuv2i32 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6706 def : NEONInstAlias<"vcle${p}.f32 $Dd, $Dn, $Dm",
6707 (VCGEfd DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6708 // Q-register versions.
6709 def : NEONInstAlias<"vcle${p}.s8 $Qd, $Qn, $Qm",
6710 (VCGEsv16i8 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6711 def : NEONInstAlias<"vcle${p}.s16 $Qd, $Qn, $Qm",
6712 (VCGEsv8i16 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6713 def : NEONInstAlias<"vcle${p}.s32 $Qd, $Qn, $Qm",
6714 (VCGEsv4i32 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6715 def : NEONInstAlias<"vcle${p}.u8 $Qd, $Qn, $Qm",
6716 (VCGEuv16i8 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6717 def : NEONInstAlias<"vcle${p}.u16 $Qd, $Qn, $Qm",
6718 (VCGEuv8i16 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6719 def : NEONInstAlias<"vcle${p}.u32 $Qd, $Qn, $Qm",
6720 (VCGEuv4i32 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6721 def : NEONInstAlias<"vcle${p}.f32 $Qd, $Qn, $Qm",
6722 (VCGEfq QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6724 // VCLT (register) is an assembler alias for VCGT w/ the operands reversed.
6725 // D-register versions.
6726 def : NEONInstAlias<"vclt${p}.s8 $Dd, $Dn, $Dm",
6727 (VCGTsv8i8 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6728 def : NEONInstAlias<"vclt${p}.s16 $Dd, $Dn, $Dm",
6729 (VCGTsv4i16 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6730 def : NEONInstAlias<"vclt${p}.s32 $Dd, $Dn, $Dm",
6731 (VCGTsv2i32 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6732 def : NEONInstAlias<"vclt${p}.u8 $Dd, $Dn, $Dm",
6733 (VCGTuv8i8 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6734 def : NEONInstAlias<"vclt${p}.u16 $Dd, $Dn, $Dm",
6735 (VCGTuv4i16 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6736 def : NEONInstAlias<"vclt${p}.u32 $Dd, $Dn, $Dm",
6737 (VCGTuv2i32 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6738 def : NEONInstAlias<"vclt${p}.f32 $Dd, $Dn, $Dm",
6739 (VCGTfd DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6740 // Q-register versions.
6741 def : NEONInstAlias<"vclt${p}.s8 $Qd, $Qn, $Qm",
6742 (VCGTsv16i8 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6743 def : NEONInstAlias<"vclt${p}.s16 $Qd, $Qn, $Qm",
6744 (VCGTsv8i16 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6745 def : NEONInstAlias<"vclt${p}.s32 $Qd, $Qn, $Qm",
6746 (VCGTsv4i32 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6747 def : NEONInstAlias<"vclt${p}.u8 $Qd, $Qn, $Qm",
6748 (VCGTuv16i8 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6749 def : NEONInstAlias<"vclt${p}.u16 $Qd, $Qn, $Qm",
6750 (VCGTuv8i16 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6751 def : NEONInstAlias<"vclt${p}.u32 $Qd, $Qn, $Qm",
6752 (VCGTuv4i32 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6753 def : NEONInstAlias<"vclt${p}.f32 $Qd, $Qn, $Qm",
6754 (VCGTfq QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6756 // VSWP allows, but does not require, a type suffix.
6757 defm : NEONDTAnyInstAlias<"vswp${p}", "$Vd, $Vm",
6758 (VSWPd DPR:$Vd, DPR:$Vm, pred:$p)>;
6759 defm : NEONDTAnyInstAlias<"vswp${p}", "$Vd, $Vm",
6760 (VSWPq QPR:$Vd, QPR:$Vm, pred:$p)>;
6762 // VBIF, VBIT, and VBSL allow, but do not require, a type suffix.
6763 defm : NEONDTAnyInstAlias<"vbif${p}", "$Vd, $Vn, $Vm",
6764 (VBIFd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
6765 defm : NEONDTAnyInstAlias<"vbit${p}", "$Vd, $Vn, $Vm",
6766 (VBITd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
6767 defm : NEONDTAnyInstAlias<"vbsl${p}", "$Vd, $Vn, $Vm",
6768 (VBSLd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
6769 defm : NEONDTAnyInstAlias<"vbif${p}", "$Vd, $Vn, $Vm",
6770 (VBIFq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
6771 defm : NEONDTAnyInstAlias<"vbit${p}", "$Vd, $Vn, $Vm",
6772 (VBITq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
6773 defm : NEONDTAnyInstAlias<"vbsl${p}", "$Vd, $Vn, $Vm",
6774 (VBSLq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
6776 // "vmov Rd, #-imm" can be handled via "vmvn".
6777 def : NEONInstAlias<"vmov${p}.i32 $Vd, $imm",
6778 (VMVNv2i32 DPR:$Vd, nImmVMOVI32Neg:$imm, pred:$p)>;
6779 def : NEONInstAlias<"vmov${p}.i32 $Vd, $imm",
6780 (VMVNv4i32 QPR:$Vd, nImmVMOVI32Neg:$imm, pred:$p)>;
6781 def : NEONInstAlias<"vmvn${p}.i32 $Vd, $imm",
6782 (VMOVv2i32 DPR:$Vd, nImmVMOVI32Neg:$imm, pred:$p)>;
6783 def : NEONInstAlias<"vmvn${p}.i32 $Vd, $imm",
6784 (VMOVv4i32 QPR:$Vd, nImmVMOVI32Neg:$imm, pred:$p)>;
6786 // 'gas' compatibility aliases for quad-word instructions. Strictly speaking,
6787 // these should restrict to just the Q register variants, but the register
6788 // classes are enough to match correctly regardless, so we keep it simple
6789 // and just use MnemonicAlias.
6790 def : NEONMnemonicAlias<"vbicq", "vbic">;
6791 def : NEONMnemonicAlias<"vandq", "vand">;
6792 def : NEONMnemonicAlias<"veorq", "veor">;
6793 def : NEONMnemonicAlias<"vorrq", "vorr">;
6795 def : NEONMnemonicAlias<"vmovq", "vmov">;
6796 def : NEONMnemonicAlias<"vmvnq", "vmvn">;
6797 // Explicit versions for floating point so that the FPImm variants get
6798 // handled early. The parser gets confused otherwise.
6799 def : NEONMnemonicAlias<"vmovq.f32", "vmov.f32">;
6800 def : NEONMnemonicAlias<"vmovq.f64", "vmov.f64">;
6802 def : NEONMnemonicAlias<"vaddq", "vadd">;
6803 def : NEONMnemonicAlias<"vsubq", "vsub">;
6805 def : NEONMnemonicAlias<"vminq", "vmin">;
6806 def : NEONMnemonicAlias<"vmaxq", "vmax">;
6808 def : NEONMnemonicAlias<"vmulq", "vmul">;
6810 def : NEONMnemonicAlias<"vabsq", "vabs">;
6812 def : NEONMnemonicAlias<"vshlq", "vshl">;
6813 def : NEONMnemonicAlias<"vshrq", "vshr">;
6815 def : NEONMnemonicAlias<"vcvtq", "vcvt">;
6817 def : NEONMnemonicAlias<"vcleq", "vcle">;
6818 def : NEONMnemonicAlias<"vceqq", "vceq">;
6820 def : NEONMnemonicAlias<"vzipq", "vzip">;
6821 def : NEONMnemonicAlias<"vswpq", "vswp">;
6823 def : NEONMnemonicAlias<"vrecpeq.f32", "vrecpe.f32">;
6824 def : NEONMnemonicAlias<"vrecpeq.u32", "vrecpe.u32">;
6827 // Alias for loading floating point immediates that aren't representable
6828 // using the vmov.f32 encoding but the bitpattern is representable using
6829 // the .i32 encoding.
6830 def : NEONInstAlias<"vmov${p}.f32 $Vd, $imm",
6831 (VMOVv4i32 QPR:$Vd, nImmVMOVI32:$imm, pred:$p)>;
6832 def : NEONInstAlias<"vmov${p}.f32 $Vd, $imm",
6833 (VMOVv2i32 DPR:$Vd, nImmVMOVI32:$imm, pred:$p)>;