1 //===-- ARMInstrNEON.td - NEON support for ARM -------------*- tablegen -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file describes the ARM NEON instruction set.
12 //===----------------------------------------------------------------------===//
15 //===----------------------------------------------------------------------===//
16 // NEON-specific Operands.
17 //===----------------------------------------------------------------------===//
18 def nModImm : Operand<i32> {
19 let PrintMethod = "printNEONModImmOperand";
22 def nImmSplatI8AsmOperand : AsmOperandClass { let Name = "NEONi8splat"; }
23 def nImmSplatI8 : Operand<i32> {
24 let PrintMethod = "printNEONModImmOperand";
25 let ParserMatchClass = nImmSplatI8AsmOperand;
27 def nImmSplatI16AsmOperand : AsmOperandClass { let Name = "NEONi16splat"; }
28 def nImmSplatI16 : Operand<i32> {
29 let PrintMethod = "printNEONModImmOperand";
30 let ParserMatchClass = nImmSplatI16AsmOperand;
32 def nImmSplatI32AsmOperand : AsmOperandClass { let Name = "NEONi32splat"; }
33 def nImmSplatI32 : Operand<i32> {
34 let PrintMethod = "printNEONModImmOperand";
35 let ParserMatchClass = nImmSplatI32AsmOperand;
37 def nImmVMOVI32AsmOperand : AsmOperandClass { let Name = "NEONi32vmov"; }
38 def nImmVMOVI32 : Operand<i32> {
39 let PrintMethod = "printNEONModImmOperand";
40 let ParserMatchClass = nImmVMOVI32AsmOperand;
42 def nImmVMOVI32NegAsmOperand : AsmOperandClass { let Name = "NEONi32vmovNeg"; }
43 def nImmVMOVI32Neg : Operand<i32> {
44 let PrintMethod = "printNEONModImmOperand";
45 let ParserMatchClass = nImmVMOVI32NegAsmOperand;
47 def nImmVMOVF32 : Operand<i32> {
48 let PrintMethod = "printFPImmOperand";
49 let ParserMatchClass = FPImmOperand;
51 def nImmSplatI64AsmOperand : AsmOperandClass { let Name = "NEONi64splat"; }
52 def nImmSplatI64 : Operand<i32> {
53 let PrintMethod = "printNEONModImmOperand";
54 let ParserMatchClass = nImmSplatI64AsmOperand;
57 def VectorIndex8Operand : AsmOperandClass { let Name = "VectorIndex8"; }
58 def VectorIndex16Operand : AsmOperandClass { let Name = "VectorIndex16"; }
59 def VectorIndex32Operand : AsmOperandClass { let Name = "VectorIndex32"; }
60 def VectorIndex8 : Operand<i32>, ImmLeaf<i32, [{
61 return ((uint64_t)Imm) < 8;
63 let ParserMatchClass = VectorIndex8Operand;
64 let PrintMethod = "printVectorIndex";
65 let MIOperandInfo = (ops i32imm);
67 def VectorIndex16 : Operand<i32>, ImmLeaf<i32, [{
68 return ((uint64_t)Imm) < 4;
70 let ParserMatchClass = VectorIndex16Operand;
71 let PrintMethod = "printVectorIndex";
72 let MIOperandInfo = (ops i32imm);
74 def VectorIndex32 : Operand<i32>, ImmLeaf<i32, [{
75 return ((uint64_t)Imm) < 2;
77 let ParserMatchClass = VectorIndex32Operand;
78 let PrintMethod = "printVectorIndex";
79 let MIOperandInfo = (ops i32imm);
82 // Register list of one D register.
83 def VecListOneDAsmOperand : AsmOperandClass {
84 let Name = "VecListOneD";
85 let ParserMethod = "parseVectorList";
86 let RenderMethod = "addVecListOperands";
88 def VecListOneD : RegisterOperand<DPR, "printVectorListOne"> {
89 let ParserMatchClass = VecListOneDAsmOperand;
91 // Register list of two sequential D registers.
92 def VecListDPairAsmOperand : AsmOperandClass {
93 let Name = "VecListDPair";
94 let ParserMethod = "parseVectorList";
95 let RenderMethod = "addVecListOperands";
97 def VecListDPair : RegisterOperand<DPair, "printVectorListTwo"> {
98 let ParserMatchClass = VecListDPairAsmOperand;
100 // Register list of three sequential D registers.
101 def VecListThreeDAsmOperand : AsmOperandClass {
102 let Name = "VecListThreeD";
103 let ParserMethod = "parseVectorList";
104 let RenderMethod = "addVecListOperands";
106 def VecListThreeD : RegisterOperand<DPR, "printVectorListThree"> {
107 let ParserMatchClass = VecListThreeDAsmOperand;
109 // Register list of four sequential D registers.
110 def VecListFourDAsmOperand : AsmOperandClass {
111 let Name = "VecListFourD";
112 let ParserMethod = "parseVectorList";
113 let RenderMethod = "addVecListOperands";
115 def VecListFourD : RegisterOperand<DPR, "printVectorListFour"> {
116 let ParserMatchClass = VecListFourDAsmOperand;
118 // Register list of two D registers spaced by 2 (two sequential Q registers).
119 def VecListDPairSpacedAsmOperand : AsmOperandClass {
120 let Name = "VecListDPairSpaced";
121 let ParserMethod = "parseVectorList";
122 let RenderMethod = "addVecListOperands";
124 def VecListDPairSpaced : RegisterOperand<DPair, "printVectorListTwoSpaced"> {
125 let ParserMatchClass = VecListDPairSpacedAsmOperand;
127 // Register list of three D registers spaced by 2 (three Q registers).
128 def VecListThreeQAsmOperand : AsmOperandClass {
129 let Name = "VecListThreeQ";
130 let ParserMethod = "parseVectorList";
131 let RenderMethod = "addVecListOperands";
133 def VecListThreeQ : RegisterOperand<DPR, "printVectorListThreeSpaced"> {
134 let ParserMatchClass = VecListThreeQAsmOperand;
136 // Register list of three D registers spaced by 2 (three Q registers).
137 def VecListFourQAsmOperand : AsmOperandClass {
138 let Name = "VecListFourQ";
139 let ParserMethod = "parseVectorList";
140 let RenderMethod = "addVecListOperands";
142 def VecListFourQ : RegisterOperand<DPR, "printVectorListFourSpaced"> {
143 let ParserMatchClass = VecListFourQAsmOperand;
146 // Register list of one D register, with "all lanes" subscripting.
147 def VecListOneDAllLanesAsmOperand : AsmOperandClass {
148 let Name = "VecListOneDAllLanes";
149 let ParserMethod = "parseVectorList";
150 let RenderMethod = "addVecListOperands";
152 def VecListOneDAllLanes : RegisterOperand<DPR, "printVectorListOneAllLanes"> {
153 let ParserMatchClass = VecListOneDAllLanesAsmOperand;
155 // Register list of two D registers, with "all lanes" subscripting.
156 def VecListDPairAllLanesAsmOperand : AsmOperandClass {
157 let Name = "VecListDPairAllLanes";
158 let ParserMethod = "parseVectorList";
159 let RenderMethod = "addVecListOperands";
161 def VecListDPairAllLanes : RegisterOperand<DPair,
162 "printVectorListTwoAllLanes"> {
163 let ParserMatchClass = VecListDPairAllLanesAsmOperand;
165 // Register list of two D registers spaced by 2 (two sequential Q registers).
166 def VecListDPairSpacedAllLanesAsmOperand : AsmOperandClass {
167 let Name = "VecListDPairSpacedAllLanes";
168 let ParserMethod = "parseVectorList";
169 let RenderMethod = "addVecListOperands";
171 def VecListDPairSpacedAllLanes : RegisterOperand<DPair,
172 "printVectorListTwoSpacedAllLanes"> {
173 let ParserMatchClass = VecListDPairSpacedAllLanesAsmOperand;
175 // Register list of three D registers, with "all lanes" subscripting.
176 def VecListThreeDAllLanesAsmOperand : AsmOperandClass {
177 let Name = "VecListThreeDAllLanes";
178 let ParserMethod = "parseVectorList";
179 let RenderMethod = "addVecListOperands";
181 def VecListThreeDAllLanes : RegisterOperand<DPR,
182 "printVectorListThreeAllLanes"> {
183 let ParserMatchClass = VecListThreeDAllLanesAsmOperand;
185 // Register list of three D registers spaced by 2 (three sequential Q regs).
186 def VecListThreeQAllLanesAsmOperand : AsmOperandClass {
187 let Name = "VecListThreeQAllLanes";
188 let ParserMethod = "parseVectorList";
189 let RenderMethod = "addVecListOperands";
191 def VecListThreeQAllLanes : RegisterOperand<DPR,
192 "printVectorListThreeSpacedAllLanes"> {
193 let ParserMatchClass = VecListThreeQAllLanesAsmOperand;
195 // Register list of four D registers, with "all lanes" subscripting.
196 def VecListFourDAllLanesAsmOperand : AsmOperandClass {
197 let Name = "VecListFourDAllLanes";
198 let ParserMethod = "parseVectorList";
199 let RenderMethod = "addVecListOperands";
201 def VecListFourDAllLanes : RegisterOperand<DPR, "printVectorListFourAllLanes"> {
202 let ParserMatchClass = VecListFourDAllLanesAsmOperand;
204 // Register list of four D registers spaced by 2 (four sequential Q regs).
205 def VecListFourQAllLanesAsmOperand : AsmOperandClass {
206 let Name = "VecListFourQAllLanes";
207 let ParserMethod = "parseVectorList";
208 let RenderMethod = "addVecListOperands";
210 def VecListFourQAllLanes : RegisterOperand<DPR,
211 "printVectorListFourSpacedAllLanes"> {
212 let ParserMatchClass = VecListFourQAllLanesAsmOperand;
216 // Register list of one D register, with byte lane subscripting.
217 def VecListOneDByteIndexAsmOperand : AsmOperandClass {
218 let Name = "VecListOneDByteIndexed";
219 let ParserMethod = "parseVectorList";
220 let RenderMethod = "addVecListIndexedOperands";
222 def VecListOneDByteIndexed : Operand<i32> {
223 let ParserMatchClass = VecListOneDByteIndexAsmOperand;
224 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
226 // ...with half-word lane subscripting.
227 def VecListOneDHWordIndexAsmOperand : AsmOperandClass {
228 let Name = "VecListOneDHWordIndexed";
229 let ParserMethod = "parseVectorList";
230 let RenderMethod = "addVecListIndexedOperands";
232 def VecListOneDHWordIndexed : Operand<i32> {
233 let ParserMatchClass = VecListOneDHWordIndexAsmOperand;
234 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
236 // ...with word lane subscripting.
237 def VecListOneDWordIndexAsmOperand : AsmOperandClass {
238 let Name = "VecListOneDWordIndexed";
239 let ParserMethod = "parseVectorList";
240 let RenderMethod = "addVecListIndexedOperands";
242 def VecListOneDWordIndexed : Operand<i32> {
243 let ParserMatchClass = VecListOneDWordIndexAsmOperand;
244 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
247 // Register list of two D registers with byte lane subscripting.
248 def VecListTwoDByteIndexAsmOperand : AsmOperandClass {
249 let Name = "VecListTwoDByteIndexed";
250 let ParserMethod = "parseVectorList";
251 let RenderMethod = "addVecListIndexedOperands";
253 def VecListTwoDByteIndexed : Operand<i32> {
254 let ParserMatchClass = VecListTwoDByteIndexAsmOperand;
255 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
257 // ...with half-word lane subscripting.
258 def VecListTwoDHWordIndexAsmOperand : AsmOperandClass {
259 let Name = "VecListTwoDHWordIndexed";
260 let ParserMethod = "parseVectorList";
261 let RenderMethod = "addVecListIndexedOperands";
263 def VecListTwoDHWordIndexed : Operand<i32> {
264 let ParserMatchClass = VecListTwoDHWordIndexAsmOperand;
265 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
267 // ...with word lane subscripting.
268 def VecListTwoDWordIndexAsmOperand : AsmOperandClass {
269 let Name = "VecListTwoDWordIndexed";
270 let ParserMethod = "parseVectorList";
271 let RenderMethod = "addVecListIndexedOperands";
273 def VecListTwoDWordIndexed : Operand<i32> {
274 let ParserMatchClass = VecListTwoDWordIndexAsmOperand;
275 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
277 // Register list of two Q registers with half-word lane subscripting.
278 def VecListTwoQHWordIndexAsmOperand : AsmOperandClass {
279 let Name = "VecListTwoQHWordIndexed";
280 let ParserMethod = "parseVectorList";
281 let RenderMethod = "addVecListIndexedOperands";
283 def VecListTwoQHWordIndexed : Operand<i32> {
284 let ParserMatchClass = VecListTwoQHWordIndexAsmOperand;
285 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
287 // ...with word lane subscripting.
288 def VecListTwoQWordIndexAsmOperand : AsmOperandClass {
289 let Name = "VecListTwoQWordIndexed";
290 let ParserMethod = "parseVectorList";
291 let RenderMethod = "addVecListIndexedOperands";
293 def VecListTwoQWordIndexed : Operand<i32> {
294 let ParserMatchClass = VecListTwoQWordIndexAsmOperand;
295 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
299 // Register list of three D registers with byte lane subscripting.
300 def VecListThreeDByteIndexAsmOperand : AsmOperandClass {
301 let Name = "VecListThreeDByteIndexed";
302 let ParserMethod = "parseVectorList";
303 let RenderMethod = "addVecListIndexedOperands";
305 def VecListThreeDByteIndexed : Operand<i32> {
306 let ParserMatchClass = VecListThreeDByteIndexAsmOperand;
307 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
309 // ...with half-word lane subscripting.
310 def VecListThreeDHWordIndexAsmOperand : AsmOperandClass {
311 let Name = "VecListThreeDHWordIndexed";
312 let ParserMethod = "parseVectorList";
313 let RenderMethod = "addVecListIndexedOperands";
315 def VecListThreeDHWordIndexed : Operand<i32> {
316 let ParserMatchClass = VecListThreeDHWordIndexAsmOperand;
317 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
319 // ...with word lane subscripting.
320 def VecListThreeDWordIndexAsmOperand : AsmOperandClass {
321 let Name = "VecListThreeDWordIndexed";
322 let ParserMethod = "parseVectorList";
323 let RenderMethod = "addVecListIndexedOperands";
325 def VecListThreeDWordIndexed : Operand<i32> {
326 let ParserMatchClass = VecListThreeDWordIndexAsmOperand;
327 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
329 // Register list of three Q registers with half-word lane subscripting.
330 def VecListThreeQHWordIndexAsmOperand : AsmOperandClass {
331 let Name = "VecListThreeQHWordIndexed";
332 let ParserMethod = "parseVectorList";
333 let RenderMethod = "addVecListIndexedOperands";
335 def VecListThreeQHWordIndexed : Operand<i32> {
336 let ParserMatchClass = VecListThreeQHWordIndexAsmOperand;
337 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
339 // ...with word lane subscripting.
340 def VecListThreeQWordIndexAsmOperand : AsmOperandClass {
341 let Name = "VecListThreeQWordIndexed";
342 let ParserMethod = "parseVectorList";
343 let RenderMethod = "addVecListIndexedOperands";
345 def VecListThreeQWordIndexed : Operand<i32> {
346 let ParserMatchClass = VecListThreeQWordIndexAsmOperand;
347 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
350 // Register list of four D registers with byte lane subscripting.
351 def VecListFourDByteIndexAsmOperand : AsmOperandClass {
352 let Name = "VecListFourDByteIndexed";
353 let ParserMethod = "parseVectorList";
354 let RenderMethod = "addVecListIndexedOperands";
356 def VecListFourDByteIndexed : Operand<i32> {
357 let ParserMatchClass = VecListFourDByteIndexAsmOperand;
358 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
360 // ...with half-word lane subscripting.
361 def VecListFourDHWordIndexAsmOperand : AsmOperandClass {
362 let Name = "VecListFourDHWordIndexed";
363 let ParserMethod = "parseVectorList";
364 let RenderMethod = "addVecListIndexedOperands";
366 def VecListFourDHWordIndexed : Operand<i32> {
367 let ParserMatchClass = VecListFourDHWordIndexAsmOperand;
368 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
370 // ...with word lane subscripting.
371 def VecListFourDWordIndexAsmOperand : AsmOperandClass {
372 let Name = "VecListFourDWordIndexed";
373 let ParserMethod = "parseVectorList";
374 let RenderMethod = "addVecListIndexedOperands";
376 def VecListFourDWordIndexed : Operand<i32> {
377 let ParserMatchClass = VecListFourDWordIndexAsmOperand;
378 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
380 // Register list of four Q registers with half-word lane subscripting.
381 def VecListFourQHWordIndexAsmOperand : AsmOperandClass {
382 let Name = "VecListFourQHWordIndexed";
383 let ParserMethod = "parseVectorList";
384 let RenderMethod = "addVecListIndexedOperands";
386 def VecListFourQHWordIndexed : Operand<i32> {
387 let ParserMatchClass = VecListFourQHWordIndexAsmOperand;
388 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
390 // ...with word lane subscripting.
391 def VecListFourQWordIndexAsmOperand : AsmOperandClass {
392 let Name = "VecListFourQWordIndexed";
393 let ParserMethod = "parseVectorList";
394 let RenderMethod = "addVecListIndexedOperands";
396 def VecListFourQWordIndexed : Operand<i32> {
397 let ParserMatchClass = VecListFourQWordIndexAsmOperand;
398 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
401 def dword_alignedload : PatFrag<(ops node:$ptr), (load node:$ptr), [{
402 return cast<LoadSDNode>(N)->getAlignment() >= 8;
404 def dword_alignedstore : PatFrag<(ops node:$val, node:$ptr),
405 (store node:$val, node:$ptr), [{
406 return cast<StoreSDNode>(N)->getAlignment() >= 8;
408 def word_alignedload : PatFrag<(ops node:$ptr), (load node:$ptr), [{
409 return cast<LoadSDNode>(N)->getAlignment() == 4;
411 def word_alignedstore : PatFrag<(ops node:$val, node:$ptr),
412 (store node:$val, node:$ptr), [{
413 return cast<StoreSDNode>(N)->getAlignment() == 4;
415 def hword_alignedload : PatFrag<(ops node:$ptr), (load node:$ptr), [{
416 return cast<LoadSDNode>(N)->getAlignment() == 2;
418 def hword_alignedstore : PatFrag<(ops node:$val, node:$ptr),
419 (store node:$val, node:$ptr), [{
420 return cast<StoreSDNode>(N)->getAlignment() == 2;
422 def byte_alignedload : PatFrag<(ops node:$ptr), (load node:$ptr), [{
423 return cast<LoadSDNode>(N)->getAlignment() == 1;
425 def byte_alignedstore : PatFrag<(ops node:$val, node:$ptr),
426 (store node:$val, node:$ptr), [{
427 return cast<StoreSDNode>(N)->getAlignment() == 1;
429 def non_word_alignedload : PatFrag<(ops node:$ptr), (load node:$ptr), [{
430 return cast<LoadSDNode>(N)->getAlignment() < 4;
432 def non_word_alignedstore : PatFrag<(ops node:$val, node:$ptr),
433 (store node:$val, node:$ptr), [{
434 return cast<StoreSDNode>(N)->getAlignment() < 4;
437 //===----------------------------------------------------------------------===//
438 // NEON-specific DAG Nodes.
439 //===----------------------------------------------------------------------===//
441 def SDTARMVCMP : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<1, 2>]>;
442 def SDTARMVCMPZ : SDTypeProfile<1, 1, []>;
444 def NEONvceq : SDNode<"ARMISD::VCEQ", SDTARMVCMP>;
445 def NEONvceqz : SDNode<"ARMISD::VCEQZ", SDTARMVCMPZ>;
446 def NEONvcge : SDNode<"ARMISD::VCGE", SDTARMVCMP>;
447 def NEONvcgez : SDNode<"ARMISD::VCGEZ", SDTARMVCMPZ>;
448 def NEONvclez : SDNode<"ARMISD::VCLEZ", SDTARMVCMPZ>;
449 def NEONvcgeu : SDNode<"ARMISD::VCGEU", SDTARMVCMP>;
450 def NEONvcgt : SDNode<"ARMISD::VCGT", SDTARMVCMP>;
451 def NEONvcgtz : SDNode<"ARMISD::VCGTZ", SDTARMVCMPZ>;
452 def NEONvcltz : SDNode<"ARMISD::VCLTZ", SDTARMVCMPZ>;
453 def NEONvcgtu : SDNode<"ARMISD::VCGTU", SDTARMVCMP>;
454 def NEONvtst : SDNode<"ARMISD::VTST", SDTARMVCMP>;
456 // Types for vector shift by immediates. The "SHX" version is for long and
457 // narrow operations where the source and destination vectors have different
458 // types. The "SHINS" version is for shift and insert operations.
459 def SDTARMVSH : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
461 def SDTARMVSHX : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
463 def SDTARMVSHINS : SDTypeProfile<1, 3, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
464 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
466 def NEONvshl : SDNode<"ARMISD::VSHL", SDTARMVSH>;
467 def NEONvshrs : SDNode<"ARMISD::VSHRs", SDTARMVSH>;
468 def NEONvshru : SDNode<"ARMISD::VSHRu", SDTARMVSH>;
469 def NEONvshlls : SDNode<"ARMISD::VSHLLs", SDTARMVSHX>;
470 def NEONvshllu : SDNode<"ARMISD::VSHLLu", SDTARMVSHX>;
471 def NEONvshlli : SDNode<"ARMISD::VSHLLi", SDTARMVSHX>;
472 def NEONvshrn : SDNode<"ARMISD::VSHRN", SDTARMVSHX>;
474 def NEONvrshrs : SDNode<"ARMISD::VRSHRs", SDTARMVSH>;
475 def NEONvrshru : SDNode<"ARMISD::VRSHRu", SDTARMVSH>;
476 def NEONvrshrn : SDNode<"ARMISD::VRSHRN", SDTARMVSHX>;
478 def NEONvqshls : SDNode<"ARMISD::VQSHLs", SDTARMVSH>;
479 def NEONvqshlu : SDNode<"ARMISD::VQSHLu", SDTARMVSH>;
480 def NEONvqshlsu : SDNode<"ARMISD::VQSHLsu", SDTARMVSH>;
481 def NEONvqshrns : SDNode<"ARMISD::VQSHRNs", SDTARMVSHX>;
482 def NEONvqshrnu : SDNode<"ARMISD::VQSHRNu", SDTARMVSHX>;
483 def NEONvqshrnsu : SDNode<"ARMISD::VQSHRNsu", SDTARMVSHX>;
485 def NEONvqrshrns : SDNode<"ARMISD::VQRSHRNs", SDTARMVSHX>;
486 def NEONvqrshrnu : SDNode<"ARMISD::VQRSHRNu", SDTARMVSHX>;
487 def NEONvqrshrnsu : SDNode<"ARMISD::VQRSHRNsu", SDTARMVSHX>;
489 def NEONvsli : SDNode<"ARMISD::VSLI", SDTARMVSHINS>;
490 def NEONvsri : SDNode<"ARMISD::VSRI", SDTARMVSHINS>;
492 def SDTARMVGETLN : SDTypeProfile<1, 2, [SDTCisVT<0, i32>, SDTCisInt<1>,
494 def NEONvgetlaneu : SDNode<"ARMISD::VGETLANEu", SDTARMVGETLN>;
495 def NEONvgetlanes : SDNode<"ARMISD::VGETLANEs", SDTARMVGETLN>;
497 def SDTARMVMOVIMM : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisVT<1, i32>]>;
498 def NEONvmovImm : SDNode<"ARMISD::VMOVIMM", SDTARMVMOVIMM>;
499 def NEONvmvnImm : SDNode<"ARMISD::VMVNIMM", SDTARMVMOVIMM>;
500 def NEONvmovFPImm : SDNode<"ARMISD::VMOVFPIMM", SDTARMVMOVIMM>;
502 def SDTARMVORRIMM : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
504 def NEONvorrImm : SDNode<"ARMISD::VORRIMM", SDTARMVORRIMM>;
505 def NEONvbicImm : SDNode<"ARMISD::VBICIMM", SDTARMVORRIMM>;
507 def NEONvbsl : SDNode<"ARMISD::VBSL",
508 SDTypeProfile<1, 3, [SDTCisVec<0>,
511 SDTCisSameAs<0, 3>]>>;
513 def NEONvdup : SDNode<"ARMISD::VDUP", SDTypeProfile<1, 1, [SDTCisVec<0>]>>;
515 // VDUPLANE can produce a quad-register result from a double-register source,
516 // so the result is not constrained to match the source.
517 def NEONvduplane : SDNode<"ARMISD::VDUPLANE",
518 SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisVec<1>,
521 def SDTARMVEXT : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
522 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
523 def NEONvext : SDNode<"ARMISD::VEXT", SDTARMVEXT>;
525 def SDTARMVSHUF : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0, 1>]>;
526 def NEONvrev64 : SDNode<"ARMISD::VREV64", SDTARMVSHUF>;
527 def NEONvrev32 : SDNode<"ARMISD::VREV32", SDTARMVSHUF>;
528 def NEONvrev16 : SDNode<"ARMISD::VREV16", SDTARMVSHUF>;
530 def SDTARMVSHUF2 : SDTypeProfile<2, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
532 SDTCisSameAs<0, 3>]>;
533 def NEONzip : SDNode<"ARMISD::VZIP", SDTARMVSHUF2>;
534 def NEONuzp : SDNode<"ARMISD::VUZP", SDTARMVSHUF2>;
535 def NEONtrn : SDNode<"ARMISD::VTRN", SDTARMVSHUF2>;
537 def SDTARMVMULL : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
538 SDTCisSameAs<1, 2>]>;
539 def NEONvmulls : SDNode<"ARMISD::VMULLs", SDTARMVMULL>;
540 def NEONvmullu : SDNode<"ARMISD::VMULLu", SDTARMVMULL>;
542 def SDTARMFMAX : SDTypeProfile<1, 2, [SDTCisVT<0, f32>, SDTCisSameAs<0, 1>,
543 SDTCisSameAs<0, 2>]>;
544 def NEONfmax : SDNode<"ARMISD::FMAX", SDTARMFMAX>;
545 def NEONfmin : SDNode<"ARMISD::FMIN", SDTARMFMAX>;
547 def NEONimmAllZerosV: PatLeaf<(NEONvmovImm (i32 timm)), [{
548 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
549 unsigned EltBits = 0;
550 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
551 return (EltBits == 32 && EltVal == 0);
554 def NEONimmAllOnesV: PatLeaf<(NEONvmovImm (i32 timm)), [{
555 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
556 unsigned EltBits = 0;
557 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
558 return (EltBits == 8 && EltVal == 0xff);
561 //===----------------------------------------------------------------------===//
562 // NEON load / store instructions
563 //===----------------------------------------------------------------------===//
565 // Use VLDM to load a Q register as a D register pair.
566 // This is a pseudo instruction that is expanded to VLDMD after reg alloc.
568 : PseudoVFPLdStM<(outs DPair:$dst), (ins GPR:$Rn),
570 [(set DPair:$dst, (v2f64 (load GPR:$Rn)))]>;
572 // Use VSTM to store a Q register as a D register pair.
573 // This is a pseudo instruction that is expanded to VSTMD after reg alloc.
575 : PseudoVFPLdStM<(outs), (ins DPair:$src, GPR:$Rn),
577 [(store (v2f64 DPair:$src), GPR:$Rn)]>;
579 // Classes for VLD* pseudo-instructions with multi-register operands.
580 // These are expanded to real instructions after register allocation.
581 class VLDQPseudo<InstrItinClass itin>
582 : PseudoNLdSt<(outs QPR:$dst), (ins addrmode6:$addr), itin, "">;
583 class VLDQWBPseudo<InstrItinClass itin>
584 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
585 (ins addrmode6:$addr, am6offset:$offset), itin,
587 class VLDQWBfixedPseudo<InstrItinClass itin>
588 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
589 (ins addrmode6:$addr), itin,
591 class VLDQWBregisterPseudo<InstrItinClass itin>
592 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
593 (ins addrmode6:$addr, rGPR:$offset), itin,
596 class VLDQQPseudo<InstrItinClass itin>
597 : PseudoNLdSt<(outs QQPR:$dst), (ins addrmode6:$addr), itin, "">;
598 class VLDQQWBPseudo<InstrItinClass itin>
599 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
600 (ins addrmode6:$addr, am6offset:$offset), itin,
602 class VLDQQWBfixedPseudo<InstrItinClass itin>
603 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
604 (ins addrmode6:$addr), itin,
606 class VLDQQWBregisterPseudo<InstrItinClass itin>
607 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
608 (ins addrmode6:$addr, rGPR:$offset), itin,
612 class VLDQQQQPseudo<InstrItinClass itin>
613 : PseudoNLdSt<(outs QQQQPR:$dst), (ins addrmode6:$addr, QQQQPR:$src),itin,
615 class VLDQQQQWBPseudo<InstrItinClass itin>
616 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
617 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
618 "$addr.addr = $wb, $src = $dst">;
620 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
622 // VLD1 : Vector Load (multiple single elements)
623 class VLD1D<bits<4> op7_4, string Dt>
624 : NLdSt<0,0b10,0b0111,op7_4, (outs VecListOneD:$Vd),
625 (ins addrmode6:$Rn), IIC_VLD1,
626 "vld1", Dt, "$Vd, $Rn", "", []> {
629 let DecoderMethod = "DecodeVLDST1Instruction";
631 class VLD1Q<bits<4> op7_4, string Dt>
632 : NLdSt<0,0b10,0b1010,op7_4, (outs VecListDPair:$Vd),
633 (ins addrmode6:$Rn), IIC_VLD1x2,
634 "vld1", Dt, "$Vd, $Rn", "", []> {
636 let Inst{5-4} = Rn{5-4};
637 let DecoderMethod = "DecodeVLDST1Instruction";
640 def VLD1d8 : VLD1D<{0,0,0,?}, "8">;
641 def VLD1d16 : VLD1D<{0,1,0,?}, "16">;
642 def VLD1d32 : VLD1D<{1,0,0,?}, "32">;
643 def VLD1d64 : VLD1D<{1,1,0,?}, "64">;
645 def VLD1q8 : VLD1Q<{0,0,?,?}, "8">;
646 def VLD1q16 : VLD1Q<{0,1,?,?}, "16">;
647 def VLD1q32 : VLD1Q<{1,0,?,?}, "32">;
648 def VLD1q64 : VLD1Q<{1,1,?,?}, "64">;
650 // ...with address register writeback:
651 multiclass VLD1DWB<bits<4> op7_4, string Dt> {
652 def _fixed : NLdSt<0,0b10, 0b0111,op7_4, (outs VecListOneD:$Vd, GPR:$wb),
653 (ins addrmode6:$Rn), IIC_VLD1u,
654 "vld1", Dt, "$Vd, $Rn!",
655 "$Rn.addr = $wb", []> {
656 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
658 let DecoderMethod = "DecodeVLDST1Instruction";
660 def _register : NLdSt<0,0b10,0b0111,op7_4, (outs VecListOneD:$Vd, GPR:$wb),
661 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1u,
662 "vld1", Dt, "$Vd, $Rn, $Rm",
663 "$Rn.addr = $wb", []> {
665 let DecoderMethod = "DecodeVLDST1Instruction";
668 multiclass VLD1QWB<bits<4> op7_4, string Dt> {
669 def _fixed : NLdSt<0,0b10,0b1010,op7_4, (outs VecListDPair:$Vd, GPR:$wb),
670 (ins addrmode6:$Rn), IIC_VLD1x2u,
671 "vld1", Dt, "$Vd, $Rn!",
672 "$Rn.addr = $wb", []> {
673 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
674 let Inst{5-4} = Rn{5-4};
675 let DecoderMethod = "DecodeVLDST1Instruction";
677 def _register : NLdSt<0,0b10,0b1010,op7_4, (outs VecListDPair:$Vd, GPR:$wb),
678 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1x2u,
679 "vld1", Dt, "$Vd, $Rn, $Rm",
680 "$Rn.addr = $wb", []> {
681 let Inst{5-4} = Rn{5-4};
682 let DecoderMethod = "DecodeVLDST1Instruction";
686 defm VLD1d8wb : VLD1DWB<{0,0,0,?}, "8">;
687 defm VLD1d16wb : VLD1DWB<{0,1,0,?}, "16">;
688 defm VLD1d32wb : VLD1DWB<{1,0,0,?}, "32">;
689 defm VLD1d64wb : VLD1DWB<{1,1,0,?}, "64">;
690 defm VLD1q8wb : VLD1QWB<{0,0,?,?}, "8">;
691 defm VLD1q16wb : VLD1QWB<{0,1,?,?}, "16">;
692 defm VLD1q32wb : VLD1QWB<{1,0,?,?}, "32">;
693 defm VLD1q64wb : VLD1QWB<{1,1,?,?}, "64">;
695 // ...with 3 registers
696 class VLD1D3<bits<4> op7_4, string Dt>
697 : NLdSt<0,0b10,0b0110,op7_4, (outs VecListThreeD:$Vd),
698 (ins addrmode6:$Rn), IIC_VLD1x3, "vld1", Dt,
699 "$Vd, $Rn", "", []> {
702 let DecoderMethod = "DecodeVLDST1Instruction";
704 multiclass VLD1D3WB<bits<4> op7_4, string Dt> {
705 def _fixed : NLdSt<0,0b10,0b0110, op7_4, (outs VecListThreeD:$Vd, GPR:$wb),
706 (ins addrmode6:$Rn), IIC_VLD1x2u,
707 "vld1", Dt, "$Vd, $Rn!",
708 "$Rn.addr = $wb", []> {
709 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
711 let DecoderMethod = "DecodeVLDST1Instruction";
713 def _register : NLdSt<0,0b10,0b0110,op7_4, (outs VecListThreeD:$Vd, GPR:$wb),
714 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1x2u,
715 "vld1", Dt, "$Vd, $Rn, $Rm",
716 "$Rn.addr = $wb", []> {
718 let DecoderMethod = "DecodeVLDST1Instruction";
722 def VLD1d8T : VLD1D3<{0,0,0,?}, "8">;
723 def VLD1d16T : VLD1D3<{0,1,0,?}, "16">;
724 def VLD1d32T : VLD1D3<{1,0,0,?}, "32">;
725 def VLD1d64T : VLD1D3<{1,1,0,?}, "64">;
727 defm VLD1d8Twb : VLD1D3WB<{0,0,0,?}, "8">;
728 defm VLD1d16Twb : VLD1D3WB<{0,1,0,?}, "16">;
729 defm VLD1d32Twb : VLD1D3WB<{1,0,0,?}, "32">;
730 defm VLD1d64Twb : VLD1D3WB<{1,1,0,?}, "64">;
732 def VLD1d64TPseudo : VLDQQPseudo<IIC_VLD1x3>;
734 // ...with 4 registers
735 class VLD1D4<bits<4> op7_4, string Dt>
736 : NLdSt<0, 0b10, 0b0010, op7_4, (outs VecListFourD:$Vd),
737 (ins addrmode6:$Rn), IIC_VLD1x4, "vld1", Dt,
738 "$Vd, $Rn", "", []> {
740 let Inst{5-4} = Rn{5-4};
741 let DecoderMethod = "DecodeVLDST1Instruction";
743 multiclass VLD1D4WB<bits<4> op7_4, string Dt> {
744 def _fixed : NLdSt<0,0b10,0b0010, op7_4, (outs VecListFourD:$Vd, GPR:$wb),
745 (ins addrmode6:$Rn), IIC_VLD1x2u,
746 "vld1", Dt, "$Vd, $Rn!",
747 "$Rn.addr = $wb", []> {
748 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
749 let Inst{5-4} = Rn{5-4};
750 let DecoderMethod = "DecodeVLDST1Instruction";
752 def _register : NLdSt<0,0b10,0b0010,op7_4, (outs VecListFourD:$Vd, GPR:$wb),
753 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1x2u,
754 "vld1", Dt, "$Vd, $Rn, $Rm",
755 "$Rn.addr = $wb", []> {
756 let Inst{5-4} = Rn{5-4};
757 let DecoderMethod = "DecodeVLDST1Instruction";
761 def VLD1d8Q : VLD1D4<{0,0,?,?}, "8">;
762 def VLD1d16Q : VLD1D4<{0,1,?,?}, "16">;
763 def VLD1d32Q : VLD1D4<{1,0,?,?}, "32">;
764 def VLD1d64Q : VLD1D4<{1,1,?,?}, "64">;
766 defm VLD1d8Qwb : VLD1D4WB<{0,0,?,?}, "8">;
767 defm VLD1d16Qwb : VLD1D4WB<{0,1,?,?}, "16">;
768 defm VLD1d32Qwb : VLD1D4WB<{1,0,?,?}, "32">;
769 defm VLD1d64Qwb : VLD1D4WB<{1,1,?,?}, "64">;
771 def VLD1d64QPseudo : VLDQQPseudo<IIC_VLD1x4>;
773 // VLD2 : Vector Load (multiple 2-element structures)
774 class VLD2<bits<4> op11_8, bits<4> op7_4, string Dt, RegisterOperand VdTy,
776 : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd),
777 (ins addrmode6:$Rn), itin,
778 "vld2", Dt, "$Vd, $Rn", "", []> {
780 let Inst{5-4} = Rn{5-4};
781 let DecoderMethod = "DecodeVLDST2Instruction";
784 def VLD2d8 : VLD2<0b1000, {0,0,?,?}, "8", VecListDPair, IIC_VLD2>;
785 def VLD2d16 : VLD2<0b1000, {0,1,?,?}, "16", VecListDPair, IIC_VLD2>;
786 def VLD2d32 : VLD2<0b1000, {1,0,?,?}, "32", VecListDPair, IIC_VLD2>;
788 def VLD2q8 : VLD2<0b0011, {0,0,?,?}, "8", VecListFourD, IIC_VLD2x2>;
789 def VLD2q16 : VLD2<0b0011, {0,1,?,?}, "16", VecListFourD, IIC_VLD2x2>;
790 def VLD2q32 : VLD2<0b0011, {1,0,?,?}, "32", VecListFourD, IIC_VLD2x2>;
792 def VLD2q8Pseudo : VLDQQPseudo<IIC_VLD2x2>;
793 def VLD2q16Pseudo : VLDQQPseudo<IIC_VLD2x2>;
794 def VLD2q32Pseudo : VLDQQPseudo<IIC_VLD2x2>;
796 // ...with address register writeback:
797 multiclass VLD2WB<bits<4> op11_8, bits<4> op7_4, string Dt,
798 RegisterOperand VdTy, InstrItinClass itin> {
799 def _fixed : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd, GPR:$wb),
800 (ins addrmode6:$Rn), itin,
801 "vld2", Dt, "$Vd, $Rn!",
802 "$Rn.addr = $wb", []> {
803 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
804 let Inst{5-4} = Rn{5-4};
805 let DecoderMethod = "DecodeVLDST2Instruction";
807 def _register : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd, GPR:$wb),
808 (ins addrmode6:$Rn, rGPR:$Rm), itin,
809 "vld2", Dt, "$Vd, $Rn, $Rm",
810 "$Rn.addr = $wb", []> {
811 let Inst{5-4} = Rn{5-4};
812 let DecoderMethod = "DecodeVLDST2Instruction";
816 defm VLD2d8wb : VLD2WB<0b1000, {0,0,?,?}, "8", VecListDPair, IIC_VLD2u>;
817 defm VLD2d16wb : VLD2WB<0b1000, {0,1,?,?}, "16", VecListDPair, IIC_VLD2u>;
818 defm VLD2d32wb : VLD2WB<0b1000, {1,0,?,?}, "32", VecListDPair, IIC_VLD2u>;
820 defm VLD2q8wb : VLD2WB<0b0011, {0,0,?,?}, "8", VecListFourD, IIC_VLD2x2u>;
821 defm VLD2q16wb : VLD2WB<0b0011, {0,1,?,?}, "16", VecListFourD, IIC_VLD2x2u>;
822 defm VLD2q32wb : VLD2WB<0b0011, {1,0,?,?}, "32", VecListFourD, IIC_VLD2x2u>;
824 def VLD2q8PseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD2x2u>;
825 def VLD2q16PseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD2x2u>;
826 def VLD2q32PseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD2x2u>;
827 def VLD2q8PseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD2x2u>;
828 def VLD2q16PseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD2x2u>;
829 def VLD2q32PseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD2x2u>;
831 // ...with double-spaced registers
832 def VLD2b8 : VLD2<0b1001, {0,0,?,?}, "8", VecListDPairSpaced, IIC_VLD2>;
833 def VLD2b16 : VLD2<0b1001, {0,1,?,?}, "16", VecListDPairSpaced, IIC_VLD2>;
834 def VLD2b32 : VLD2<0b1001, {1,0,?,?}, "32", VecListDPairSpaced, IIC_VLD2>;
835 defm VLD2b8wb : VLD2WB<0b1001, {0,0,?,?}, "8", VecListDPairSpaced, IIC_VLD2u>;
836 defm VLD2b16wb : VLD2WB<0b1001, {0,1,?,?}, "16", VecListDPairSpaced, IIC_VLD2u>;
837 defm VLD2b32wb : VLD2WB<0b1001, {1,0,?,?}, "32", VecListDPairSpaced, IIC_VLD2u>;
839 // VLD3 : Vector Load (multiple 3-element structures)
840 class VLD3D<bits<4> op11_8, bits<4> op7_4, string Dt>
841 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
842 (ins addrmode6:$Rn), IIC_VLD3,
843 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn", "", []> {
846 let DecoderMethod = "DecodeVLDST3Instruction";
849 def VLD3d8 : VLD3D<0b0100, {0,0,0,?}, "8">;
850 def VLD3d16 : VLD3D<0b0100, {0,1,0,?}, "16">;
851 def VLD3d32 : VLD3D<0b0100, {1,0,0,?}, "32">;
853 def VLD3d8Pseudo : VLDQQPseudo<IIC_VLD3>;
854 def VLD3d16Pseudo : VLDQQPseudo<IIC_VLD3>;
855 def VLD3d32Pseudo : VLDQQPseudo<IIC_VLD3>;
857 // ...with address register writeback:
858 class VLD3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
859 : NLdSt<0, 0b10, op11_8, op7_4,
860 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
861 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD3u,
862 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn$Rm",
863 "$Rn.addr = $wb", []> {
865 let DecoderMethod = "DecodeVLDST3Instruction";
868 def VLD3d8_UPD : VLD3DWB<0b0100, {0,0,0,?}, "8">;
869 def VLD3d16_UPD : VLD3DWB<0b0100, {0,1,0,?}, "16">;
870 def VLD3d32_UPD : VLD3DWB<0b0100, {1,0,0,?}, "32">;
872 def VLD3d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
873 def VLD3d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
874 def VLD3d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
876 // ...with double-spaced registers:
877 def VLD3q8 : VLD3D<0b0101, {0,0,0,?}, "8">;
878 def VLD3q16 : VLD3D<0b0101, {0,1,0,?}, "16">;
879 def VLD3q32 : VLD3D<0b0101, {1,0,0,?}, "32">;
880 def VLD3q8_UPD : VLD3DWB<0b0101, {0,0,0,?}, "8">;
881 def VLD3q16_UPD : VLD3DWB<0b0101, {0,1,0,?}, "16">;
882 def VLD3q32_UPD : VLD3DWB<0b0101, {1,0,0,?}, "32">;
884 def VLD3q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
885 def VLD3q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
886 def VLD3q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
888 // ...alternate versions to be allocated odd register numbers:
889 def VLD3q8oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
890 def VLD3q16oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
891 def VLD3q32oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
893 def VLD3q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
894 def VLD3q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
895 def VLD3q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
897 // VLD4 : Vector Load (multiple 4-element structures)
898 class VLD4D<bits<4> op11_8, bits<4> op7_4, string Dt>
899 : NLdSt<0, 0b10, op11_8, op7_4,
900 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
901 (ins addrmode6:$Rn), IIC_VLD4,
902 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> {
904 let Inst{5-4} = Rn{5-4};
905 let DecoderMethod = "DecodeVLDST4Instruction";
908 def VLD4d8 : VLD4D<0b0000, {0,0,?,?}, "8">;
909 def VLD4d16 : VLD4D<0b0000, {0,1,?,?}, "16">;
910 def VLD4d32 : VLD4D<0b0000, {1,0,?,?}, "32">;
912 def VLD4d8Pseudo : VLDQQPseudo<IIC_VLD4>;
913 def VLD4d16Pseudo : VLDQQPseudo<IIC_VLD4>;
914 def VLD4d32Pseudo : VLDQQPseudo<IIC_VLD4>;
916 // ...with address register writeback:
917 class VLD4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
918 : NLdSt<0, 0b10, op11_8, op7_4,
919 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
920 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD4u,
921 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm",
922 "$Rn.addr = $wb", []> {
923 let Inst{5-4} = Rn{5-4};
924 let DecoderMethod = "DecodeVLDST4Instruction";
927 def VLD4d8_UPD : VLD4DWB<0b0000, {0,0,?,?}, "8">;
928 def VLD4d16_UPD : VLD4DWB<0b0000, {0,1,?,?}, "16">;
929 def VLD4d32_UPD : VLD4DWB<0b0000, {1,0,?,?}, "32">;
931 def VLD4d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
932 def VLD4d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
933 def VLD4d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
935 // ...with double-spaced registers:
936 def VLD4q8 : VLD4D<0b0001, {0,0,?,?}, "8">;
937 def VLD4q16 : VLD4D<0b0001, {0,1,?,?}, "16">;
938 def VLD4q32 : VLD4D<0b0001, {1,0,?,?}, "32">;
939 def VLD4q8_UPD : VLD4DWB<0b0001, {0,0,?,?}, "8">;
940 def VLD4q16_UPD : VLD4DWB<0b0001, {0,1,?,?}, "16">;
941 def VLD4q32_UPD : VLD4DWB<0b0001, {1,0,?,?}, "32">;
943 def VLD4q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
944 def VLD4q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
945 def VLD4q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
947 // ...alternate versions to be allocated odd register numbers:
948 def VLD4q8oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
949 def VLD4q16oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
950 def VLD4q32oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
952 def VLD4q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
953 def VLD4q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
954 def VLD4q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
956 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
958 // Classes for VLD*LN pseudo-instructions with multi-register operands.
959 // These are expanded to real instructions after register allocation.
960 class VLDQLNPseudo<InstrItinClass itin>
961 : PseudoNLdSt<(outs QPR:$dst),
962 (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
963 itin, "$src = $dst">;
964 class VLDQLNWBPseudo<InstrItinClass itin>
965 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
966 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
967 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
968 class VLDQQLNPseudo<InstrItinClass itin>
969 : PseudoNLdSt<(outs QQPR:$dst),
970 (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
971 itin, "$src = $dst">;
972 class VLDQQLNWBPseudo<InstrItinClass itin>
973 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
974 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
975 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
976 class VLDQQQQLNPseudo<InstrItinClass itin>
977 : PseudoNLdSt<(outs QQQQPR:$dst),
978 (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
979 itin, "$src = $dst">;
980 class VLDQQQQLNWBPseudo<InstrItinClass itin>
981 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
982 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
983 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
985 // VLD1LN : Vector Load (single element to one lane)
986 class VLD1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
988 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
989 (ins addrmode6:$Rn, DPR:$src, nohash_imm:$lane),
990 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
992 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
993 (i32 (LoadOp addrmode6:$Rn)),
996 let DecoderMethod = "DecodeVLD1LN";
998 class VLD1LN32<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1000 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
1001 (ins addrmode6oneL32:$Rn, DPR:$src, nohash_imm:$lane),
1002 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
1004 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
1005 (i32 (LoadOp addrmode6oneL32:$Rn)),
1008 let DecoderMethod = "DecodeVLD1LN";
1010 class VLD1QLNPseudo<ValueType Ty, PatFrag LoadOp> : VLDQLNPseudo<IIC_VLD1ln> {
1011 let Pattern = [(set QPR:$dst, (vector_insert (Ty QPR:$src),
1012 (i32 (LoadOp addrmode6:$addr)),
1016 def VLD1LNd8 : VLD1LN<0b0000, {?,?,?,0}, "8", v8i8, extloadi8> {
1017 let Inst{7-5} = lane{2-0};
1019 def VLD1LNd16 : VLD1LN<0b0100, {?,?,0,?}, "16", v4i16, extloadi16> {
1020 let Inst{7-6} = lane{1-0};
1021 let Inst{5-4} = Rn{5-4};
1023 def VLD1LNd32 : VLD1LN32<0b1000, {?,0,?,?}, "32", v2i32, load> {
1024 let Inst{7} = lane{0};
1025 let Inst{5-4} = Rn{5-4};
1028 def VLD1LNq8Pseudo : VLD1QLNPseudo<v16i8, extloadi8>;
1029 def VLD1LNq16Pseudo : VLD1QLNPseudo<v8i16, extloadi16>;
1030 def VLD1LNq32Pseudo : VLD1QLNPseudo<v4i32, load>;
1032 def : Pat<(vector_insert (v2f32 DPR:$src),
1033 (f32 (load addrmode6:$addr)), imm:$lane),
1034 (VLD1LNd32 addrmode6:$addr, DPR:$src, imm:$lane)>;
1035 def : Pat<(vector_insert (v4f32 QPR:$src),
1036 (f32 (load addrmode6:$addr)), imm:$lane),
1037 (VLD1LNq32Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
1039 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
1041 // ...with address register writeback:
1042 class VLD1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1043 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, GPR:$wb),
1044 (ins addrmode6:$Rn, am6offset:$Rm,
1045 DPR:$src, nohash_imm:$lane), IIC_VLD1lnu, "vld1", Dt,
1046 "\\{$Vd[$lane]\\}, $Rn$Rm",
1047 "$src = $Vd, $Rn.addr = $wb", []> {
1048 let DecoderMethod = "DecodeVLD1LN";
1051 def VLD1LNd8_UPD : VLD1LNWB<0b0000, {?,?,?,0}, "8"> {
1052 let Inst{7-5} = lane{2-0};
1054 def VLD1LNd16_UPD : VLD1LNWB<0b0100, {?,?,0,?}, "16"> {
1055 let Inst{7-6} = lane{1-0};
1056 let Inst{4} = Rn{4};
1058 def VLD1LNd32_UPD : VLD1LNWB<0b1000, {?,0,?,?}, "32"> {
1059 let Inst{7} = lane{0};
1060 let Inst{5} = Rn{4};
1061 let Inst{4} = Rn{4};
1064 def VLD1LNq8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
1065 def VLD1LNq16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
1066 def VLD1LNq32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
1068 // VLD2LN : Vector Load (single 2-element structure to one lane)
1069 class VLD2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1070 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2),
1071 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, nohash_imm:$lane),
1072 IIC_VLD2ln, "vld2", Dt, "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn",
1073 "$src1 = $Vd, $src2 = $dst2", []> {
1075 let Inst{4} = Rn{4};
1076 let DecoderMethod = "DecodeVLD2LN";
1079 def VLD2LNd8 : VLD2LN<0b0001, {?,?,?,?}, "8"> {
1080 let Inst{7-5} = lane{2-0};
1082 def VLD2LNd16 : VLD2LN<0b0101, {?,?,0,?}, "16"> {
1083 let Inst{7-6} = lane{1-0};
1085 def VLD2LNd32 : VLD2LN<0b1001, {?,0,0,?}, "32"> {
1086 let Inst{7} = lane{0};
1089 def VLD2LNd8Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
1090 def VLD2LNd16Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
1091 def VLD2LNd32Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
1093 // ...with double-spaced registers:
1094 def VLD2LNq16 : VLD2LN<0b0101, {?,?,1,?}, "16"> {
1095 let Inst{7-6} = lane{1-0};
1097 def VLD2LNq32 : VLD2LN<0b1001, {?,1,0,?}, "32"> {
1098 let Inst{7} = lane{0};
1101 def VLD2LNq16Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
1102 def VLD2LNq32Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
1104 // ...with address register writeback:
1105 class VLD2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1106 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
1107 (ins addrmode6:$Rn, am6offset:$Rm,
1108 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VLD2lnu, "vld2", Dt,
1109 "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn$Rm",
1110 "$src1 = $Vd, $src2 = $dst2, $Rn.addr = $wb", []> {
1111 let Inst{4} = Rn{4};
1112 let DecoderMethod = "DecodeVLD2LN";
1115 def VLD2LNd8_UPD : VLD2LNWB<0b0001, {?,?,?,?}, "8"> {
1116 let Inst{7-5} = lane{2-0};
1118 def VLD2LNd16_UPD : VLD2LNWB<0b0101, {?,?,0,?}, "16"> {
1119 let Inst{7-6} = lane{1-0};
1121 def VLD2LNd32_UPD : VLD2LNWB<0b1001, {?,0,0,?}, "32"> {
1122 let Inst{7} = lane{0};
1125 def VLD2LNd8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
1126 def VLD2LNd16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
1127 def VLD2LNd32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
1129 def VLD2LNq16_UPD : VLD2LNWB<0b0101, {?,?,1,?}, "16"> {
1130 let Inst{7-6} = lane{1-0};
1132 def VLD2LNq32_UPD : VLD2LNWB<0b1001, {?,1,0,?}, "32"> {
1133 let Inst{7} = lane{0};
1136 def VLD2LNq16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
1137 def VLD2LNq32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
1139 // VLD3LN : Vector Load (single 3-element structure to one lane)
1140 class VLD3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1141 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
1142 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3,
1143 nohash_imm:$lane), IIC_VLD3ln, "vld3", Dt,
1144 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn",
1145 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3", []> {
1147 let DecoderMethod = "DecodeVLD3LN";
1150 def VLD3LNd8 : VLD3LN<0b0010, {?,?,?,0}, "8"> {
1151 let Inst{7-5} = lane{2-0};
1153 def VLD3LNd16 : VLD3LN<0b0110, {?,?,0,0}, "16"> {
1154 let Inst{7-6} = lane{1-0};
1156 def VLD3LNd32 : VLD3LN<0b1010, {?,0,0,0}, "32"> {
1157 let Inst{7} = lane{0};
1160 def VLD3LNd8Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
1161 def VLD3LNd16Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
1162 def VLD3LNd32Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
1164 // ...with double-spaced registers:
1165 def VLD3LNq16 : VLD3LN<0b0110, {?,?,1,0}, "16"> {
1166 let Inst{7-6} = lane{1-0};
1168 def VLD3LNq32 : VLD3LN<0b1010, {?,1,0,0}, "32"> {
1169 let Inst{7} = lane{0};
1172 def VLD3LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
1173 def VLD3LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
1175 // ...with address register writeback:
1176 class VLD3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1177 : NLdStLn<1, 0b10, op11_8, op7_4,
1178 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
1179 (ins addrmode6:$Rn, am6offset:$Rm,
1180 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
1181 IIC_VLD3lnu, "vld3", Dt,
1182 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn$Rm",
1183 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $Rn.addr = $wb",
1185 let DecoderMethod = "DecodeVLD3LN";
1188 def VLD3LNd8_UPD : VLD3LNWB<0b0010, {?,?,?,0}, "8"> {
1189 let Inst{7-5} = lane{2-0};
1191 def VLD3LNd16_UPD : VLD3LNWB<0b0110, {?,?,0,0}, "16"> {
1192 let Inst{7-6} = lane{1-0};
1194 def VLD3LNd32_UPD : VLD3LNWB<0b1010, {?,0,0,0}, "32"> {
1195 let Inst{7} = lane{0};
1198 def VLD3LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
1199 def VLD3LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
1200 def VLD3LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
1202 def VLD3LNq16_UPD : VLD3LNWB<0b0110, {?,?,1,0}, "16"> {
1203 let Inst{7-6} = lane{1-0};
1205 def VLD3LNq32_UPD : VLD3LNWB<0b1010, {?,1,0,0}, "32"> {
1206 let Inst{7} = lane{0};
1209 def VLD3LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
1210 def VLD3LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
1212 // VLD4LN : Vector Load (single 4-element structure to one lane)
1213 class VLD4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1214 : NLdStLn<1, 0b10, op11_8, op7_4,
1215 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
1216 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
1217 nohash_imm:$lane), IIC_VLD4ln, "vld4", Dt,
1218 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn",
1219 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4", []> {
1221 let Inst{4} = Rn{4};
1222 let DecoderMethod = "DecodeVLD4LN";
1225 def VLD4LNd8 : VLD4LN<0b0011, {?,?,?,?}, "8"> {
1226 let Inst{7-5} = lane{2-0};
1228 def VLD4LNd16 : VLD4LN<0b0111, {?,?,0,?}, "16"> {
1229 let Inst{7-6} = lane{1-0};
1231 def VLD4LNd32 : VLD4LN<0b1011, {?,0,?,?}, "32"> {
1232 let Inst{7} = lane{0};
1233 let Inst{5} = Rn{5};
1236 def VLD4LNd8Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
1237 def VLD4LNd16Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
1238 def VLD4LNd32Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
1240 // ...with double-spaced registers:
1241 def VLD4LNq16 : VLD4LN<0b0111, {?,?,1,?}, "16"> {
1242 let Inst{7-6} = lane{1-0};
1244 def VLD4LNq32 : VLD4LN<0b1011, {?,1,?,?}, "32"> {
1245 let Inst{7} = lane{0};
1246 let Inst{5} = Rn{5};
1249 def VLD4LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
1250 def VLD4LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
1252 // ...with address register writeback:
1253 class VLD4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1254 : NLdStLn<1, 0b10, op11_8, op7_4,
1255 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
1256 (ins addrmode6:$Rn, am6offset:$Rm,
1257 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
1258 IIC_VLD4lnu, "vld4", Dt,
1259 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn$Rm",
1260 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4, $Rn.addr = $wb",
1262 let Inst{4} = Rn{4};
1263 let DecoderMethod = "DecodeVLD4LN" ;
1266 def VLD4LNd8_UPD : VLD4LNWB<0b0011, {?,?,?,?}, "8"> {
1267 let Inst{7-5} = lane{2-0};
1269 def VLD4LNd16_UPD : VLD4LNWB<0b0111, {?,?,0,?}, "16"> {
1270 let Inst{7-6} = lane{1-0};
1272 def VLD4LNd32_UPD : VLD4LNWB<0b1011, {?,0,?,?}, "32"> {
1273 let Inst{7} = lane{0};
1274 let Inst{5} = Rn{5};
1277 def VLD4LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
1278 def VLD4LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
1279 def VLD4LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
1281 def VLD4LNq16_UPD : VLD4LNWB<0b0111, {?,?,1,?}, "16"> {
1282 let Inst{7-6} = lane{1-0};
1284 def VLD4LNq32_UPD : VLD4LNWB<0b1011, {?,1,?,?}, "32"> {
1285 let Inst{7} = lane{0};
1286 let Inst{5} = Rn{5};
1289 def VLD4LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
1290 def VLD4LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
1292 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
1294 // VLD1DUP : Vector Load (single element to all lanes)
1295 class VLD1DUP<bits<4> op7_4, string Dt, ValueType Ty, PatFrag LoadOp>
1296 : NLdSt<1, 0b10, 0b1100, op7_4, (outs VecListOneDAllLanes:$Vd),
1297 (ins addrmode6dup:$Rn),
1298 IIC_VLD1dup, "vld1", Dt, "$Vd, $Rn", "",
1299 [(set VecListOneDAllLanes:$Vd,
1300 (Ty (NEONvdup (i32 (LoadOp addrmode6dup:$Rn)))))]> {
1302 let Inst{4} = Rn{4};
1303 let DecoderMethod = "DecodeVLD1DupInstruction";
1305 def VLD1DUPd8 : VLD1DUP<{0,0,0,?}, "8", v8i8, extloadi8>;
1306 def VLD1DUPd16 : VLD1DUP<{0,1,0,?}, "16", v4i16, extloadi16>;
1307 def VLD1DUPd32 : VLD1DUP<{1,0,0,?}, "32", v2i32, load>;
1309 def : Pat<(v2f32 (NEONvdup (f32 (load addrmode6dup:$addr)))),
1310 (VLD1DUPd32 addrmode6:$addr)>;
1312 class VLD1QDUP<bits<4> op7_4, string Dt, ValueType Ty, PatFrag LoadOp>
1313 : NLdSt<1, 0b10, 0b1100, op7_4, (outs VecListDPairAllLanes:$Vd),
1314 (ins addrmode6dup:$Rn), IIC_VLD1dup,
1315 "vld1", Dt, "$Vd, $Rn", "",
1316 [(set VecListDPairAllLanes:$Vd,
1317 (Ty (NEONvdup (i32 (LoadOp addrmode6dup:$Rn)))))]> {
1319 let Inst{4} = Rn{4};
1320 let DecoderMethod = "DecodeVLD1DupInstruction";
1323 def VLD1DUPq8 : VLD1QDUP<{0,0,1,0}, "8", v16i8, extloadi8>;
1324 def VLD1DUPq16 : VLD1QDUP<{0,1,1,?}, "16", v8i16, extloadi16>;
1325 def VLD1DUPq32 : VLD1QDUP<{1,0,1,?}, "32", v4i32, load>;
1327 def : Pat<(v4f32 (NEONvdup (f32 (load addrmode6dup:$addr)))),
1328 (VLD1DUPq32 addrmode6:$addr)>;
1330 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
1331 // ...with address register writeback:
1332 multiclass VLD1DUPWB<bits<4> op7_4, string Dt> {
1333 def _fixed : NLdSt<1, 0b10, 0b1100, op7_4,
1334 (outs VecListOneDAllLanes:$Vd, GPR:$wb),
1335 (ins addrmode6dup:$Rn), IIC_VLD1dupu,
1336 "vld1", Dt, "$Vd, $Rn!",
1337 "$Rn.addr = $wb", []> {
1338 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1339 let Inst{4} = Rn{4};
1340 let DecoderMethod = "DecodeVLD1DupInstruction";
1342 def _register : NLdSt<1, 0b10, 0b1100, op7_4,
1343 (outs VecListOneDAllLanes:$Vd, GPR:$wb),
1344 (ins addrmode6dup:$Rn, rGPR:$Rm), IIC_VLD1dupu,
1345 "vld1", Dt, "$Vd, $Rn, $Rm",
1346 "$Rn.addr = $wb", []> {
1347 let Inst{4} = Rn{4};
1348 let DecoderMethod = "DecodeVLD1DupInstruction";
1351 multiclass VLD1QDUPWB<bits<4> op7_4, string Dt> {
1352 def _fixed : NLdSt<1, 0b10, 0b1100, op7_4,
1353 (outs VecListDPairAllLanes:$Vd, GPR:$wb),
1354 (ins addrmode6dup:$Rn), IIC_VLD1dupu,
1355 "vld1", Dt, "$Vd, $Rn!",
1356 "$Rn.addr = $wb", []> {
1357 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1358 let Inst{4} = Rn{4};
1359 let DecoderMethod = "DecodeVLD1DupInstruction";
1361 def _register : NLdSt<1, 0b10, 0b1100, op7_4,
1362 (outs VecListDPairAllLanes:$Vd, GPR:$wb),
1363 (ins addrmode6dup:$Rn, rGPR:$Rm), IIC_VLD1dupu,
1364 "vld1", Dt, "$Vd, $Rn, $Rm",
1365 "$Rn.addr = $wb", []> {
1366 let Inst{4} = Rn{4};
1367 let DecoderMethod = "DecodeVLD1DupInstruction";
1371 defm VLD1DUPd8wb : VLD1DUPWB<{0,0,0,0}, "8">;
1372 defm VLD1DUPd16wb : VLD1DUPWB<{0,1,0,?}, "16">;
1373 defm VLD1DUPd32wb : VLD1DUPWB<{1,0,0,?}, "32">;
1375 defm VLD1DUPq8wb : VLD1QDUPWB<{0,0,1,0}, "8">;
1376 defm VLD1DUPq16wb : VLD1QDUPWB<{0,1,1,?}, "16">;
1377 defm VLD1DUPq32wb : VLD1QDUPWB<{1,0,1,?}, "32">;
1379 // VLD2DUP : Vector Load (single 2-element structure to all lanes)
1380 class VLD2DUP<bits<4> op7_4, string Dt, RegisterOperand VdTy>
1381 : NLdSt<1, 0b10, 0b1101, op7_4, (outs VdTy:$Vd),
1382 (ins addrmode6dup:$Rn), IIC_VLD2dup,
1383 "vld2", Dt, "$Vd, $Rn", "", []> {
1385 let Inst{4} = Rn{4};
1386 let DecoderMethod = "DecodeVLD2DupInstruction";
1389 def VLD2DUPd8 : VLD2DUP<{0,0,0,?}, "8", VecListDPairAllLanes>;
1390 def VLD2DUPd16 : VLD2DUP<{0,1,0,?}, "16", VecListDPairAllLanes>;
1391 def VLD2DUPd32 : VLD2DUP<{1,0,0,?}, "32", VecListDPairAllLanes>;
1393 // ...with double-spaced registers
1394 def VLD2DUPd8x2 : VLD2DUP<{0,0,1,?}, "8", VecListDPairSpacedAllLanes>;
1395 def VLD2DUPd16x2 : VLD2DUP<{0,1,1,?}, "16", VecListDPairSpacedAllLanes>;
1396 def VLD2DUPd32x2 : VLD2DUP<{1,0,1,?}, "32", VecListDPairSpacedAllLanes>;
1398 // ...with address register writeback:
1399 multiclass VLD2DUPWB<bits<4> op7_4, string Dt, RegisterOperand VdTy> {
1400 def _fixed : NLdSt<1, 0b10, 0b1101, op7_4,
1401 (outs VdTy:$Vd, GPR:$wb),
1402 (ins addrmode6dup:$Rn), IIC_VLD2dupu,
1403 "vld2", Dt, "$Vd, $Rn!",
1404 "$Rn.addr = $wb", []> {
1405 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1406 let Inst{4} = Rn{4};
1407 let DecoderMethod = "DecodeVLD2DupInstruction";
1409 def _register : NLdSt<1, 0b10, 0b1101, op7_4,
1410 (outs VdTy:$Vd, GPR:$wb),
1411 (ins addrmode6dup:$Rn, rGPR:$Rm), IIC_VLD2dupu,
1412 "vld2", Dt, "$Vd, $Rn, $Rm",
1413 "$Rn.addr = $wb", []> {
1414 let Inst{4} = Rn{4};
1415 let DecoderMethod = "DecodeVLD2DupInstruction";
1419 defm VLD2DUPd8wb : VLD2DUPWB<{0,0,0,0}, "8", VecListDPairAllLanes>;
1420 defm VLD2DUPd16wb : VLD2DUPWB<{0,1,0,?}, "16", VecListDPairAllLanes>;
1421 defm VLD2DUPd32wb : VLD2DUPWB<{1,0,0,?}, "32", VecListDPairAllLanes>;
1423 defm VLD2DUPd8x2wb : VLD2DUPWB<{0,0,1,0}, "8", VecListDPairSpacedAllLanes>;
1424 defm VLD2DUPd16x2wb : VLD2DUPWB<{0,1,1,?}, "16", VecListDPairSpacedAllLanes>;
1425 defm VLD2DUPd32x2wb : VLD2DUPWB<{1,0,1,?}, "32", VecListDPairSpacedAllLanes>;
1427 // VLD3DUP : Vector Load (single 3-element structure to all lanes)
1428 class VLD3DUP<bits<4> op7_4, string Dt>
1429 : NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
1430 (ins addrmode6dup:$Rn), IIC_VLD3dup,
1431 "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn", "", []> {
1434 let DecoderMethod = "DecodeVLD3DupInstruction";
1437 def VLD3DUPd8 : VLD3DUP<{0,0,0,?}, "8">;
1438 def VLD3DUPd16 : VLD3DUP<{0,1,0,?}, "16">;
1439 def VLD3DUPd32 : VLD3DUP<{1,0,0,?}, "32">;
1441 def VLD3DUPd8Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1442 def VLD3DUPd16Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1443 def VLD3DUPd32Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1445 // ...with double-spaced registers (not used for codegen):
1446 def VLD3DUPq8 : VLD3DUP<{0,0,1,?}, "8">;
1447 def VLD3DUPq16 : VLD3DUP<{0,1,1,?}, "16">;
1448 def VLD3DUPq32 : VLD3DUP<{1,0,1,?}, "32">;
1450 // ...with address register writeback:
1451 class VLD3DUPWB<bits<4> op7_4, string Dt>
1452 : NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
1453 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD3dupu,
1454 "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn$Rm",
1455 "$Rn.addr = $wb", []> {
1457 let DecoderMethod = "DecodeVLD3DupInstruction";
1460 def VLD3DUPd8_UPD : VLD3DUPWB<{0,0,0,0}, "8">;
1461 def VLD3DUPd16_UPD : VLD3DUPWB<{0,1,0,?}, "16">;
1462 def VLD3DUPd32_UPD : VLD3DUPWB<{1,0,0,?}, "32">;
1464 def VLD3DUPq8_UPD : VLD3DUPWB<{0,0,1,0}, "8">;
1465 def VLD3DUPq16_UPD : VLD3DUPWB<{0,1,1,?}, "16">;
1466 def VLD3DUPq32_UPD : VLD3DUPWB<{1,0,1,?}, "32">;
1468 def VLD3DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1469 def VLD3DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1470 def VLD3DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1472 // VLD4DUP : Vector Load (single 4-element structure to all lanes)
1473 class VLD4DUP<bits<4> op7_4, string Dt>
1474 : NLdSt<1, 0b10, 0b1111, op7_4,
1475 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
1476 (ins addrmode6dup:$Rn), IIC_VLD4dup,
1477 "vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn", "", []> {
1479 let Inst{4} = Rn{4};
1480 let DecoderMethod = "DecodeVLD4DupInstruction";
1483 def VLD4DUPd8 : VLD4DUP<{0,0,0,?}, "8">;
1484 def VLD4DUPd16 : VLD4DUP<{0,1,0,?}, "16">;
1485 def VLD4DUPd32 : VLD4DUP<{1,?,0,?}, "32"> { let Inst{6} = Rn{5}; }
1487 def VLD4DUPd8Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1488 def VLD4DUPd16Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1489 def VLD4DUPd32Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1491 // ...with double-spaced registers (not used for codegen):
1492 def VLD4DUPq8 : VLD4DUP<{0,0,1,?}, "8">;
1493 def VLD4DUPq16 : VLD4DUP<{0,1,1,?}, "16">;
1494 def VLD4DUPq32 : VLD4DUP<{1,?,1,?}, "32"> { let Inst{6} = Rn{5}; }
1496 // ...with address register writeback:
1497 class VLD4DUPWB<bits<4> op7_4, string Dt>
1498 : NLdSt<1, 0b10, 0b1111, op7_4,
1499 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
1500 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD4dupu,
1501 "vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn$Rm",
1502 "$Rn.addr = $wb", []> {
1503 let Inst{4} = Rn{4};
1504 let DecoderMethod = "DecodeVLD4DupInstruction";
1507 def VLD4DUPd8_UPD : VLD4DUPWB<{0,0,0,0}, "8">;
1508 def VLD4DUPd16_UPD : VLD4DUPWB<{0,1,0,?}, "16">;
1509 def VLD4DUPd32_UPD : VLD4DUPWB<{1,?,0,?}, "32"> { let Inst{6} = Rn{5}; }
1511 def VLD4DUPq8_UPD : VLD4DUPWB<{0,0,1,0}, "8">;
1512 def VLD4DUPq16_UPD : VLD4DUPWB<{0,1,1,?}, "16">;
1513 def VLD4DUPq32_UPD : VLD4DUPWB<{1,?,1,?}, "32"> { let Inst{6} = Rn{5}; }
1515 def VLD4DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1516 def VLD4DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1517 def VLD4DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1519 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
1521 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
1523 // Classes for VST* pseudo-instructions with multi-register operands.
1524 // These are expanded to real instructions after register allocation.
1525 class VSTQPseudo<InstrItinClass itin>
1526 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src), itin, "">;
1527 class VSTQWBPseudo<InstrItinClass itin>
1528 : PseudoNLdSt<(outs GPR:$wb),
1529 (ins addrmode6:$addr, am6offset:$offset, QPR:$src), itin,
1530 "$addr.addr = $wb">;
1531 class VSTQWBfixedPseudo<InstrItinClass itin>
1532 : PseudoNLdSt<(outs GPR:$wb),
1533 (ins addrmode6:$addr, QPR:$src), itin,
1534 "$addr.addr = $wb">;
1535 class VSTQWBregisterPseudo<InstrItinClass itin>
1536 : PseudoNLdSt<(outs GPR:$wb),
1537 (ins addrmode6:$addr, rGPR:$offset, QPR:$src), itin,
1538 "$addr.addr = $wb">;
1539 class VSTQQPseudo<InstrItinClass itin>
1540 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src), itin, "">;
1541 class VSTQQWBPseudo<InstrItinClass itin>
1542 : PseudoNLdSt<(outs GPR:$wb),
1543 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src), itin,
1544 "$addr.addr = $wb">;
1545 class VSTQQWBfixedPseudo<InstrItinClass itin>
1546 : PseudoNLdSt<(outs GPR:$wb),
1547 (ins addrmode6:$addr, QQPR:$src), itin,
1548 "$addr.addr = $wb">;
1549 class VSTQQWBregisterPseudo<InstrItinClass itin>
1550 : PseudoNLdSt<(outs GPR:$wb),
1551 (ins addrmode6:$addr, rGPR:$offset, QQPR:$src), itin,
1552 "$addr.addr = $wb">;
1554 class VSTQQQQPseudo<InstrItinClass itin>
1555 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src), itin, "">;
1556 class VSTQQQQWBPseudo<InstrItinClass itin>
1557 : PseudoNLdSt<(outs GPR:$wb),
1558 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
1559 "$addr.addr = $wb">;
1561 // VST1 : Vector Store (multiple single elements)
1562 class VST1D<bits<4> op7_4, string Dt>
1563 : NLdSt<0,0b00,0b0111,op7_4, (outs), (ins addrmode6:$Rn, VecListOneD:$Vd),
1564 IIC_VST1, "vst1", Dt, "$Vd, $Rn", "", []> {
1566 let Inst{4} = Rn{4};
1567 let DecoderMethod = "DecodeVLDST1Instruction";
1569 class VST1Q<bits<4> op7_4, string Dt>
1570 : NLdSt<0,0b00,0b1010,op7_4, (outs), (ins addrmode6:$Rn, VecListDPair:$Vd),
1571 IIC_VST1x2, "vst1", Dt, "$Vd, $Rn", "", []> {
1573 let Inst{5-4} = Rn{5-4};
1574 let DecoderMethod = "DecodeVLDST1Instruction";
1577 def VST1d8 : VST1D<{0,0,0,?}, "8">;
1578 def VST1d16 : VST1D<{0,1,0,?}, "16">;
1579 def VST1d32 : VST1D<{1,0,0,?}, "32">;
1580 def VST1d64 : VST1D<{1,1,0,?}, "64">;
1582 def VST1q8 : VST1Q<{0,0,?,?}, "8">;
1583 def VST1q16 : VST1Q<{0,1,?,?}, "16">;
1584 def VST1q32 : VST1Q<{1,0,?,?}, "32">;
1585 def VST1q64 : VST1Q<{1,1,?,?}, "64">;
1587 // ...with address register writeback:
1588 multiclass VST1DWB<bits<4> op7_4, string Dt> {
1589 def _fixed : NLdSt<0,0b00, 0b0111,op7_4, (outs GPR:$wb),
1590 (ins addrmode6:$Rn, VecListOneD:$Vd), IIC_VLD1u,
1591 "vst1", Dt, "$Vd, $Rn!",
1592 "$Rn.addr = $wb", []> {
1593 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1594 let Inst{4} = Rn{4};
1595 let DecoderMethod = "DecodeVLDST1Instruction";
1597 def _register : NLdSt<0,0b00,0b0111,op7_4, (outs GPR:$wb),
1598 (ins addrmode6:$Rn, rGPR:$Rm, VecListOneD:$Vd),
1600 "vst1", Dt, "$Vd, $Rn, $Rm",
1601 "$Rn.addr = $wb", []> {
1602 let Inst{4} = Rn{4};
1603 let DecoderMethod = "DecodeVLDST1Instruction";
1606 multiclass VST1QWB<bits<4> op7_4, string Dt> {
1607 def _fixed : NLdSt<0,0b00,0b1010,op7_4, (outs GPR:$wb),
1608 (ins addrmode6:$Rn, VecListDPair:$Vd), IIC_VLD1x2u,
1609 "vst1", Dt, "$Vd, $Rn!",
1610 "$Rn.addr = $wb", []> {
1611 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1612 let Inst{5-4} = Rn{5-4};
1613 let DecoderMethod = "DecodeVLDST1Instruction";
1615 def _register : NLdSt<0,0b00,0b1010,op7_4, (outs GPR:$wb),
1616 (ins addrmode6:$Rn, rGPR:$Rm, VecListDPair:$Vd),
1618 "vst1", Dt, "$Vd, $Rn, $Rm",
1619 "$Rn.addr = $wb", []> {
1620 let Inst{5-4} = Rn{5-4};
1621 let DecoderMethod = "DecodeVLDST1Instruction";
1625 defm VST1d8wb : VST1DWB<{0,0,0,?}, "8">;
1626 defm VST1d16wb : VST1DWB<{0,1,0,?}, "16">;
1627 defm VST1d32wb : VST1DWB<{1,0,0,?}, "32">;
1628 defm VST1d64wb : VST1DWB<{1,1,0,?}, "64">;
1630 defm VST1q8wb : VST1QWB<{0,0,?,?}, "8">;
1631 defm VST1q16wb : VST1QWB<{0,1,?,?}, "16">;
1632 defm VST1q32wb : VST1QWB<{1,0,?,?}, "32">;
1633 defm VST1q64wb : VST1QWB<{1,1,?,?}, "64">;
1635 // ...with 3 registers
1636 class VST1D3<bits<4> op7_4, string Dt>
1637 : NLdSt<0, 0b00, 0b0110, op7_4, (outs),
1638 (ins addrmode6:$Rn, VecListThreeD:$Vd),
1639 IIC_VST1x3, "vst1", Dt, "$Vd, $Rn", "", []> {
1641 let Inst{4} = Rn{4};
1642 let DecoderMethod = "DecodeVLDST1Instruction";
1644 multiclass VST1D3WB<bits<4> op7_4, string Dt> {
1645 def _fixed : NLdSt<0,0b00,0b0110,op7_4, (outs GPR:$wb),
1646 (ins addrmode6:$Rn, VecListThreeD:$Vd), IIC_VLD1x3u,
1647 "vst1", Dt, "$Vd, $Rn!",
1648 "$Rn.addr = $wb", []> {
1649 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1650 let Inst{5-4} = Rn{5-4};
1651 let DecoderMethod = "DecodeVLDST1Instruction";
1653 def _register : NLdSt<0,0b00,0b0110,op7_4, (outs GPR:$wb),
1654 (ins addrmode6:$Rn, rGPR:$Rm, VecListThreeD:$Vd),
1656 "vst1", Dt, "$Vd, $Rn, $Rm",
1657 "$Rn.addr = $wb", []> {
1658 let Inst{5-4} = Rn{5-4};
1659 let DecoderMethod = "DecodeVLDST1Instruction";
1663 def VST1d8T : VST1D3<{0,0,0,?}, "8">;
1664 def VST1d16T : VST1D3<{0,1,0,?}, "16">;
1665 def VST1d32T : VST1D3<{1,0,0,?}, "32">;
1666 def VST1d64T : VST1D3<{1,1,0,?}, "64">;
1668 defm VST1d8Twb : VST1D3WB<{0,0,0,?}, "8">;
1669 defm VST1d16Twb : VST1D3WB<{0,1,0,?}, "16">;
1670 defm VST1d32Twb : VST1D3WB<{1,0,0,?}, "32">;
1671 defm VST1d64Twb : VST1D3WB<{1,1,0,?}, "64">;
1673 def VST1d64TPseudo : VSTQQPseudo<IIC_VST1x3>;
1674 def VST1d64TPseudoWB_fixed : VSTQQWBPseudo<IIC_VST1x3u>;
1675 def VST1d64TPseudoWB_register : VSTQQWBPseudo<IIC_VST1x3u>;
1677 // ...with 4 registers
1678 class VST1D4<bits<4> op7_4, string Dt>
1679 : NLdSt<0, 0b00, 0b0010, op7_4, (outs),
1680 (ins addrmode6:$Rn, VecListFourD:$Vd),
1681 IIC_VST1x4, "vst1", Dt, "$Vd, $Rn", "",
1684 let Inst{5-4} = Rn{5-4};
1685 let DecoderMethod = "DecodeVLDST1Instruction";
1687 multiclass VST1D4WB<bits<4> op7_4, string Dt> {
1688 def _fixed : NLdSt<0,0b00,0b0010,op7_4, (outs GPR:$wb),
1689 (ins addrmode6:$Rn, VecListFourD:$Vd), IIC_VLD1x4u,
1690 "vst1", Dt, "$Vd, $Rn!",
1691 "$Rn.addr = $wb", []> {
1692 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1693 let Inst{5-4} = Rn{5-4};
1694 let DecoderMethod = "DecodeVLDST1Instruction";
1696 def _register : NLdSt<0,0b00,0b0010,op7_4, (outs GPR:$wb),
1697 (ins addrmode6:$Rn, rGPR:$Rm, VecListFourD:$Vd),
1699 "vst1", Dt, "$Vd, $Rn, $Rm",
1700 "$Rn.addr = $wb", []> {
1701 let Inst{5-4} = Rn{5-4};
1702 let DecoderMethod = "DecodeVLDST1Instruction";
1706 def VST1d8Q : VST1D4<{0,0,?,?}, "8">;
1707 def VST1d16Q : VST1D4<{0,1,?,?}, "16">;
1708 def VST1d32Q : VST1D4<{1,0,?,?}, "32">;
1709 def VST1d64Q : VST1D4<{1,1,?,?}, "64">;
1711 defm VST1d8Qwb : VST1D4WB<{0,0,?,?}, "8">;
1712 defm VST1d16Qwb : VST1D4WB<{0,1,?,?}, "16">;
1713 defm VST1d32Qwb : VST1D4WB<{1,0,?,?}, "32">;
1714 defm VST1d64Qwb : VST1D4WB<{1,1,?,?}, "64">;
1716 def VST1d64QPseudo : VSTQQPseudo<IIC_VST1x4>;
1717 def VST1d64QPseudoWB_fixed : VSTQQWBPseudo<IIC_VST1x4u>;
1718 def VST1d64QPseudoWB_register : VSTQQWBPseudo<IIC_VST1x4u>;
1720 // VST2 : Vector Store (multiple 2-element structures)
1721 class VST2<bits<4> op11_8, bits<4> op7_4, string Dt, RegisterOperand VdTy,
1722 InstrItinClass itin>
1723 : NLdSt<0, 0b00, op11_8, op7_4, (outs), (ins addrmode6:$Rn, VdTy:$Vd),
1724 itin, "vst2", Dt, "$Vd, $Rn", "", []> {
1726 let Inst{5-4} = Rn{5-4};
1727 let DecoderMethod = "DecodeVLDST2Instruction";
1730 def VST2d8 : VST2<0b1000, {0,0,?,?}, "8", VecListDPair, IIC_VST2>;
1731 def VST2d16 : VST2<0b1000, {0,1,?,?}, "16", VecListDPair, IIC_VST2>;
1732 def VST2d32 : VST2<0b1000, {1,0,?,?}, "32", VecListDPair, IIC_VST2>;
1734 def VST2q8 : VST2<0b0011, {0,0,?,?}, "8", VecListFourD, IIC_VST2x2>;
1735 def VST2q16 : VST2<0b0011, {0,1,?,?}, "16", VecListFourD, IIC_VST2x2>;
1736 def VST2q32 : VST2<0b0011, {1,0,?,?}, "32", VecListFourD, IIC_VST2x2>;
1738 def VST2q8Pseudo : VSTQQPseudo<IIC_VST2x2>;
1739 def VST2q16Pseudo : VSTQQPseudo<IIC_VST2x2>;
1740 def VST2q32Pseudo : VSTQQPseudo<IIC_VST2x2>;
1742 // ...with address register writeback:
1743 multiclass VST2DWB<bits<4> op11_8, bits<4> op7_4, string Dt,
1744 RegisterOperand VdTy> {
1745 def _fixed : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1746 (ins addrmode6:$Rn, VdTy:$Vd), IIC_VLD1u,
1747 "vst2", Dt, "$Vd, $Rn!",
1748 "$Rn.addr = $wb", []> {
1749 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1750 let Inst{5-4} = Rn{5-4};
1751 let DecoderMethod = "DecodeVLDST2Instruction";
1753 def _register : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1754 (ins addrmode6:$Rn, rGPR:$Rm, VdTy:$Vd), IIC_VLD1u,
1755 "vst2", Dt, "$Vd, $Rn, $Rm",
1756 "$Rn.addr = $wb", []> {
1757 let Inst{5-4} = Rn{5-4};
1758 let DecoderMethod = "DecodeVLDST2Instruction";
1761 multiclass VST2QWB<bits<4> op7_4, string Dt> {
1762 def _fixed : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
1763 (ins addrmode6:$Rn, VecListFourD:$Vd), IIC_VLD1u,
1764 "vst2", Dt, "$Vd, $Rn!",
1765 "$Rn.addr = $wb", []> {
1766 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1767 let Inst{5-4} = Rn{5-4};
1768 let DecoderMethod = "DecodeVLDST2Instruction";
1770 def _register : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
1771 (ins addrmode6:$Rn, rGPR:$Rm, VecListFourD:$Vd),
1773 "vst2", Dt, "$Vd, $Rn, $Rm",
1774 "$Rn.addr = $wb", []> {
1775 let Inst{5-4} = Rn{5-4};
1776 let DecoderMethod = "DecodeVLDST2Instruction";
1780 defm VST2d8wb : VST2DWB<0b1000, {0,0,?,?}, "8", VecListDPair>;
1781 defm VST2d16wb : VST2DWB<0b1000, {0,1,?,?}, "16", VecListDPair>;
1782 defm VST2d32wb : VST2DWB<0b1000, {1,0,?,?}, "32", VecListDPair>;
1784 defm VST2q8wb : VST2QWB<{0,0,?,?}, "8">;
1785 defm VST2q16wb : VST2QWB<{0,1,?,?}, "16">;
1786 defm VST2q32wb : VST2QWB<{1,0,?,?}, "32">;
1788 def VST2q8PseudoWB_fixed : VSTQQWBfixedPseudo<IIC_VST2x2u>;
1789 def VST2q16PseudoWB_fixed : VSTQQWBfixedPseudo<IIC_VST2x2u>;
1790 def VST2q32PseudoWB_fixed : VSTQQWBfixedPseudo<IIC_VST2x2u>;
1791 def VST2q8PseudoWB_register : VSTQQWBregisterPseudo<IIC_VST2x2u>;
1792 def VST2q16PseudoWB_register : VSTQQWBregisterPseudo<IIC_VST2x2u>;
1793 def VST2q32PseudoWB_register : VSTQQWBregisterPseudo<IIC_VST2x2u>;
1795 // ...with double-spaced registers
1796 def VST2b8 : VST2<0b1001, {0,0,?,?}, "8", VecListDPairSpaced, IIC_VST2>;
1797 def VST2b16 : VST2<0b1001, {0,1,?,?}, "16", VecListDPairSpaced, IIC_VST2>;
1798 def VST2b32 : VST2<0b1001, {1,0,?,?}, "32", VecListDPairSpaced, IIC_VST2>;
1799 defm VST2b8wb : VST2DWB<0b1001, {0,0,?,?}, "8", VecListDPairSpaced>;
1800 defm VST2b16wb : VST2DWB<0b1001, {0,1,?,?}, "16", VecListDPairSpaced>;
1801 defm VST2b32wb : VST2DWB<0b1001, {1,0,?,?}, "32", VecListDPairSpaced>;
1803 // VST3 : Vector Store (multiple 3-element structures)
1804 class VST3D<bits<4> op11_8, bits<4> op7_4, string Dt>
1805 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1806 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3,
1807 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn", "", []> {
1809 let Inst{4} = Rn{4};
1810 let DecoderMethod = "DecodeVLDST3Instruction";
1813 def VST3d8 : VST3D<0b0100, {0,0,0,?}, "8">;
1814 def VST3d16 : VST3D<0b0100, {0,1,0,?}, "16">;
1815 def VST3d32 : VST3D<0b0100, {1,0,0,?}, "32">;
1817 def VST3d8Pseudo : VSTQQPseudo<IIC_VST3>;
1818 def VST3d16Pseudo : VSTQQPseudo<IIC_VST3>;
1819 def VST3d32Pseudo : VSTQQPseudo<IIC_VST3>;
1821 // ...with address register writeback:
1822 class VST3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1823 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1824 (ins addrmode6:$Rn, am6offset:$Rm,
1825 DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3u,
1826 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn$Rm",
1827 "$Rn.addr = $wb", []> {
1828 let Inst{4} = Rn{4};
1829 let DecoderMethod = "DecodeVLDST3Instruction";
1832 def VST3d8_UPD : VST3DWB<0b0100, {0,0,0,?}, "8">;
1833 def VST3d16_UPD : VST3DWB<0b0100, {0,1,0,?}, "16">;
1834 def VST3d32_UPD : VST3DWB<0b0100, {1,0,0,?}, "32">;
1836 def VST3d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1837 def VST3d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1838 def VST3d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1840 // ...with double-spaced registers:
1841 def VST3q8 : VST3D<0b0101, {0,0,0,?}, "8">;
1842 def VST3q16 : VST3D<0b0101, {0,1,0,?}, "16">;
1843 def VST3q32 : VST3D<0b0101, {1,0,0,?}, "32">;
1844 def VST3q8_UPD : VST3DWB<0b0101, {0,0,0,?}, "8">;
1845 def VST3q16_UPD : VST3DWB<0b0101, {0,1,0,?}, "16">;
1846 def VST3q32_UPD : VST3DWB<0b0101, {1,0,0,?}, "32">;
1848 def VST3q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1849 def VST3q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1850 def VST3q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1852 // ...alternate versions to be allocated odd register numbers:
1853 def VST3q8oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1854 def VST3q16oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1855 def VST3q32oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1857 def VST3q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1858 def VST3q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1859 def VST3q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1861 // VST4 : Vector Store (multiple 4-element structures)
1862 class VST4D<bits<4> op11_8, bits<4> op7_4, string Dt>
1863 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1864 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1865 IIC_VST4, "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn",
1868 let Inst{5-4} = Rn{5-4};
1869 let DecoderMethod = "DecodeVLDST4Instruction";
1872 def VST4d8 : VST4D<0b0000, {0,0,?,?}, "8">;
1873 def VST4d16 : VST4D<0b0000, {0,1,?,?}, "16">;
1874 def VST4d32 : VST4D<0b0000, {1,0,?,?}, "32">;
1876 def VST4d8Pseudo : VSTQQPseudo<IIC_VST4>;
1877 def VST4d16Pseudo : VSTQQPseudo<IIC_VST4>;
1878 def VST4d32Pseudo : VSTQQPseudo<IIC_VST4>;
1880 // ...with address register writeback:
1881 class VST4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1882 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1883 (ins addrmode6:$Rn, am6offset:$Rm,
1884 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST4u,
1885 "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1886 "$Rn.addr = $wb", []> {
1887 let Inst{5-4} = Rn{5-4};
1888 let DecoderMethod = "DecodeVLDST4Instruction";
1891 def VST4d8_UPD : VST4DWB<0b0000, {0,0,?,?}, "8">;
1892 def VST4d16_UPD : VST4DWB<0b0000, {0,1,?,?}, "16">;
1893 def VST4d32_UPD : VST4DWB<0b0000, {1,0,?,?}, "32">;
1895 def VST4d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1896 def VST4d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1897 def VST4d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1899 // ...with double-spaced registers:
1900 def VST4q8 : VST4D<0b0001, {0,0,?,?}, "8">;
1901 def VST4q16 : VST4D<0b0001, {0,1,?,?}, "16">;
1902 def VST4q32 : VST4D<0b0001, {1,0,?,?}, "32">;
1903 def VST4q8_UPD : VST4DWB<0b0001, {0,0,?,?}, "8">;
1904 def VST4q16_UPD : VST4DWB<0b0001, {0,1,?,?}, "16">;
1905 def VST4q32_UPD : VST4DWB<0b0001, {1,0,?,?}, "32">;
1907 def VST4q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1908 def VST4q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1909 def VST4q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1911 // ...alternate versions to be allocated odd register numbers:
1912 def VST4q8oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1913 def VST4q16oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1914 def VST4q32oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1916 def VST4q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1917 def VST4q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1918 def VST4q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1920 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
1922 // Classes for VST*LN pseudo-instructions with multi-register operands.
1923 // These are expanded to real instructions after register allocation.
1924 class VSTQLNPseudo<InstrItinClass itin>
1925 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
1927 class VSTQLNWBPseudo<InstrItinClass itin>
1928 : PseudoNLdSt<(outs GPR:$wb),
1929 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
1930 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1931 class VSTQQLNPseudo<InstrItinClass itin>
1932 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
1934 class VSTQQLNWBPseudo<InstrItinClass itin>
1935 : PseudoNLdSt<(outs GPR:$wb),
1936 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
1937 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1938 class VSTQQQQLNPseudo<InstrItinClass itin>
1939 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
1941 class VSTQQQQLNWBPseudo<InstrItinClass itin>
1942 : PseudoNLdSt<(outs GPR:$wb),
1943 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
1944 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1946 // VST1LN : Vector Store (single element from one lane)
1947 class VST1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1948 PatFrag StoreOp, SDNode ExtractOp, Operand AddrMode>
1949 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1950 (ins AddrMode:$Rn, DPR:$Vd, nohash_imm:$lane),
1951 IIC_VST1ln, "vst1", Dt, "\\{$Vd[$lane]\\}, $Rn", "",
1952 [(StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane), AddrMode:$Rn)]> {
1954 let DecoderMethod = "DecodeVST1LN";
1956 class VST1QLNPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
1957 : VSTQLNPseudo<IIC_VST1ln> {
1958 let Pattern = [(StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
1962 def VST1LNd8 : VST1LN<0b0000, {?,?,?,0}, "8", v8i8, truncstorei8,
1963 NEONvgetlaneu, addrmode6> {
1964 let Inst{7-5} = lane{2-0};
1966 def VST1LNd16 : VST1LN<0b0100, {?,?,0,?}, "16", v4i16, truncstorei16,
1967 NEONvgetlaneu, addrmode6> {
1968 let Inst{7-6} = lane{1-0};
1969 let Inst{4} = Rn{4};
1972 def VST1LNd32 : VST1LN<0b1000, {?,0,?,?}, "32", v2i32, store, extractelt,
1974 let Inst{7} = lane{0};
1975 let Inst{5-4} = Rn{5-4};
1978 def VST1LNq8Pseudo : VST1QLNPseudo<v16i8, truncstorei8, NEONvgetlaneu>;
1979 def VST1LNq16Pseudo : VST1QLNPseudo<v8i16, truncstorei16, NEONvgetlaneu>;
1980 def VST1LNq32Pseudo : VST1QLNPseudo<v4i32, store, extractelt>;
1982 def : Pat<(store (extractelt (v2f32 DPR:$src), imm:$lane), addrmode6:$addr),
1983 (VST1LNd32 addrmode6:$addr, DPR:$src, imm:$lane)>;
1984 def : Pat<(store (extractelt (v4f32 QPR:$src), imm:$lane), addrmode6:$addr),
1985 (VST1LNq32Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
1987 // ...with address register writeback:
1988 class VST1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1989 PatFrag StoreOp, SDNode ExtractOp, Operand AdrMode>
1990 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1991 (ins AdrMode:$Rn, am6offset:$Rm,
1992 DPR:$Vd, nohash_imm:$lane), IIC_VST1lnu, "vst1", Dt,
1993 "\\{$Vd[$lane]\\}, $Rn$Rm",
1995 [(set GPR:$wb, (StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane),
1996 AdrMode:$Rn, am6offset:$Rm))]> {
1997 let DecoderMethod = "DecodeVST1LN";
1999 class VST1QLNWBPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
2000 : VSTQLNWBPseudo<IIC_VST1lnu> {
2001 let Pattern = [(set GPR:$wb, (StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
2002 addrmode6:$addr, am6offset:$offset))];
2005 def VST1LNd8_UPD : VST1LNWB<0b0000, {?,?,?,0}, "8", v8i8, post_truncsti8,
2006 NEONvgetlaneu, addrmode6> {
2007 let Inst{7-5} = lane{2-0};
2009 def VST1LNd16_UPD : VST1LNWB<0b0100, {?,?,0,?}, "16", v4i16, post_truncsti16,
2010 NEONvgetlaneu, addrmode6> {
2011 let Inst{7-6} = lane{1-0};
2012 let Inst{4} = Rn{4};
2014 def VST1LNd32_UPD : VST1LNWB<0b1000, {?,0,?,?}, "32", v2i32, post_store,
2015 extractelt, addrmode6oneL32> {
2016 let Inst{7} = lane{0};
2017 let Inst{5-4} = Rn{5-4};
2020 def VST1LNq8Pseudo_UPD : VST1QLNWBPseudo<v16i8, post_truncsti8, NEONvgetlaneu>;
2021 def VST1LNq16Pseudo_UPD : VST1QLNWBPseudo<v8i16, post_truncsti16,NEONvgetlaneu>;
2022 def VST1LNq32Pseudo_UPD : VST1QLNWBPseudo<v4i32, post_store, extractelt>;
2024 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
2026 // VST2LN : Vector Store (single 2-element structure from one lane)
2027 class VST2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
2028 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
2029 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, nohash_imm:$lane),
2030 IIC_VST2ln, "vst2", Dt, "\\{$Vd[$lane], $src2[$lane]\\}, $Rn",
2033 let Inst{4} = Rn{4};
2034 let DecoderMethod = "DecodeVST2LN";
2037 def VST2LNd8 : VST2LN<0b0001, {?,?,?,?}, "8"> {
2038 let Inst{7-5} = lane{2-0};
2040 def VST2LNd16 : VST2LN<0b0101, {?,?,0,?}, "16"> {
2041 let Inst{7-6} = lane{1-0};
2043 def VST2LNd32 : VST2LN<0b1001, {?,0,0,?}, "32"> {
2044 let Inst{7} = lane{0};
2047 def VST2LNd8Pseudo : VSTQLNPseudo<IIC_VST2ln>;
2048 def VST2LNd16Pseudo : VSTQLNPseudo<IIC_VST2ln>;
2049 def VST2LNd32Pseudo : VSTQLNPseudo<IIC_VST2ln>;
2051 // ...with double-spaced registers:
2052 def VST2LNq16 : VST2LN<0b0101, {?,?,1,?}, "16"> {
2053 let Inst{7-6} = lane{1-0};
2054 let Inst{4} = Rn{4};
2056 def VST2LNq32 : VST2LN<0b1001, {?,1,0,?}, "32"> {
2057 let Inst{7} = lane{0};
2058 let Inst{4} = Rn{4};
2061 def VST2LNq16Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
2062 def VST2LNq32Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
2064 // ...with address register writeback:
2065 class VST2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
2066 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
2067 (ins addrmode6:$Rn, am6offset:$Rm,
2068 DPR:$Vd, DPR:$src2, nohash_imm:$lane), IIC_VST2lnu, "vst2", Dt,
2069 "\\{$Vd[$lane], $src2[$lane]\\}, $Rn$Rm",
2070 "$Rn.addr = $wb", []> {
2071 let Inst{4} = Rn{4};
2072 let DecoderMethod = "DecodeVST2LN";
2075 def VST2LNd8_UPD : VST2LNWB<0b0001, {?,?,?,?}, "8"> {
2076 let Inst{7-5} = lane{2-0};
2078 def VST2LNd16_UPD : VST2LNWB<0b0101, {?,?,0,?}, "16"> {
2079 let Inst{7-6} = lane{1-0};
2081 def VST2LNd32_UPD : VST2LNWB<0b1001, {?,0,0,?}, "32"> {
2082 let Inst{7} = lane{0};
2085 def VST2LNd8Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
2086 def VST2LNd16Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
2087 def VST2LNd32Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
2089 def VST2LNq16_UPD : VST2LNWB<0b0101, {?,?,1,?}, "16"> {
2090 let Inst{7-6} = lane{1-0};
2092 def VST2LNq32_UPD : VST2LNWB<0b1001, {?,1,0,?}, "32"> {
2093 let Inst{7} = lane{0};
2096 def VST2LNq16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
2097 def VST2LNq32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
2099 // VST3LN : Vector Store (single 3-element structure from one lane)
2100 class VST3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
2101 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
2102 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3,
2103 nohash_imm:$lane), IIC_VST3ln, "vst3", Dt,
2104 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn", "", []> {
2106 let DecoderMethod = "DecodeVST3LN";
2109 def VST3LNd8 : VST3LN<0b0010, {?,?,?,0}, "8"> {
2110 let Inst{7-5} = lane{2-0};
2112 def VST3LNd16 : VST3LN<0b0110, {?,?,0,0}, "16"> {
2113 let Inst{7-6} = lane{1-0};
2115 def VST3LNd32 : VST3LN<0b1010, {?,0,0,0}, "32"> {
2116 let Inst{7} = lane{0};
2119 def VST3LNd8Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
2120 def VST3LNd16Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
2121 def VST3LNd32Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
2123 // ...with double-spaced registers:
2124 def VST3LNq16 : VST3LN<0b0110, {?,?,1,0}, "16"> {
2125 let Inst{7-6} = lane{1-0};
2127 def VST3LNq32 : VST3LN<0b1010, {?,1,0,0}, "32"> {
2128 let Inst{7} = lane{0};
2131 def VST3LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
2132 def VST3LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
2134 // ...with address register writeback:
2135 class VST3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
2136 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
2137 (ins addrmode6:$Rn, am6offset:$Rm,
2138 DPR:$Vd, DPR:$src2, DPR:$src3, nohash_imm:$lane),
2139 IIC_VST3lnu, "vst3", Dt,
2140 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn$Rm",
2141 "$Rn.addr = $wb", []> {
2142 let DecoderMethod = "DecodeVST3LN";
2145 def VST3LNd8_UPD : VST3LNWB<0b0010, {?,?,?,0}, "8"> {
2146 let Inst{7-5} = lane{2-0};
2148 def VST3LNd16_UPD : VST3LNWB<0b0110, {?,?,0,0}, "16"> {
2149 let Inst{7-6} = lane{1-0};
2151 def VST3LNd32_UPD : VST3LNWB<0b1010, {?,0,0,0}, "32"> {
2152 let Inst{7} = lane{0};
2155 def VST3LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
2156 def VST3LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
2157 def VST3LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
2159 def VST3LNq16_UPD : VST3LNWB<0b0110, {?,?,1,0}, "16"> {
2160 let Inst{7-6} = lane{1-0};
2162 def VST3LNq32_UPD : VST3LNWB<0b1010, {?,1,0,0}, "32"> {
2163 let Inst{7} = lane{0};
2166 def VST3LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
2167 def VST3LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
2169 // VST4LN : Vector Store (single 4-element structure from one lane)
2170 class VST4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
2171 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
2172 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4,
2173 nohash_imm:$lane), IIC_VST4ln, "vst4", Dt,
2174 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn",
2177 let Inst{4} = Rn{4};
2178 let DecoderMethod = "DecodeVST4LN";
2181 def VST4LNd8 : VST4LN<0b0011, {?,?,?,?}, "8"> {
2182 let Inst{7-5} = lane{2-0};
2184 def VST4LNd16 : VST4LN<0b0111, {?,?,0,?}, "16"> {
2185 let Inst{7-6} = lane{1-0};
2187 def VST4LNd32 : VST4LN<0b1011, {?,0,?,?}, "32"> {
2188 let Inst{7} = lane{0};
2189 let Inst{5} = Rn{5};
2192 def VST4LNd8Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
2193 def VST4LNd16Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
2194 def VST4LNd32Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
2196 // ...with double-spaced registers:
2197 def VST4LNq16 : VST4LN<0b0111, {?,?,1,?}, "16"> {
2198 let Inst{7-6} = lane{1-0};
2200 def VST4LNq32 : VST4LN<0b1011, {?,1,?,?}, "32"> {
2201 let Inst{7} = lane{0};
2202 let Inst{5} = Rn{5};
2205 def VST4LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
2206 def VST4LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
2208 // ...with address register writeback:
2209 class VST4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
2210 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
2211 (ins addrmode6:$Rn, am6offset:$Rm,
2212 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
2213 IIC_VST4lnu, "vst4", Dt,
2214 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn$Rm",
2215 "$Rn.addr = $wb", []> {
2216 let Inst{4} = Rn{4};
2217 let DecoderMethod = "DecodeVST4LN";
2220 def VST4LNd8_UPD : VST4LNWB<0b0011, {?,?,?,?}, "8"> {
2221 let Inst{7-5} = lane{2-0};
2223 def VST4LNd16_UPD : VST4LNWB<0b0111, {?,?,0,?}, "16"> {
2224 let Inst{7-6} = lane{1-0};
2226 def VST4LNd32_UPD : VST4LNWB<0b1011, {?,0,?,?}, "32"> {
2227 let Inst{7} = lane{0};
2228 let Inst{5} = Rn{5};
2231 def VST4LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
2232 def VST4LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
2233 def VST4LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
2235 def VST4LNq16_UPD : VST4LNWB<0b0111, {?,?,1,?}, "16"> {
2236 let Inst{7-6} = lane{1-0};
2238 def VST4LNq32_UPD : VST4LNWB<0b1011, {?,1,?,?}, "32"> {
2239 let Inst{7} = lane{0};
2240 let Inst{5} = Rn{5};
2243 def VST4LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
2244 def VST4LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
2246 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
2248 // Use vld1/vst1 for unaligned f64 load / store
2249 def : Pat<(f64 (hword_alignedload addrmode6:$addr)),
2250 (VLD1d16 addrmode6:$addr)>, Requires<[IsLE]>;
2251 def : Pat<(hword_alignedstore (f64 DPR:$value), addrmode6:$addr),
2252 (VST1d16 addrmode6:$addr, DPR:$value)>, Requires<[IsLE]>;
2253 def : Pat<(f64 (byte_alignedload addrmode6:$addr)),
2254 (VLD1d8 addrmode6:$addr)>, Requires<[IsLE]>;
2255 def : Pat<(byte_alignedstore (f64 DPR:$value), addrmode6:$addr),
2256 (VST1d8 addrmode6:$addr, DPR:$value)>, Requires<[IsLE]>;
2257 def : Pat<(f64 (non_word_alignedload addrmode6:$addr)),
2258 (VLD1d64 addrmode6:$addr)>, Requires<[IsBE]>;
2259 def : Pat<(non_word_alignedstore (f64 DPR:$value), addrmode6:$addr),
2260 (VST1d64 addrmode6:$addr, DPR:$value)>, Requires<[IsBE]>;
2262 // Use vld1/vst1 for Q and QQ. Also use them for unaligned v2f64
2263 // load / store if it's legal.
2264 def : Pat<(v2f64 (dword_alignedload addrmode6:$addr)),
2265 (VLD1q64 addrmode6:$addr)>;
2266 def : Pat<(dword_alignedstore (v2f64 QPR:$value), addrmode6:$addr),
2267 (VST1q64 addrmode6:$addr, QPR:$value)>;
2268 def : Pat<(v2f64 (word_alignedload addrmode6:$addr)),
2269 (VLD1q32 addrmode6:$addr)>;
2270 def : Pat<(word_alignedstore (v2f64 QPR:$value), addrmode6:$addr),
2271 (VST1q32 addrmode6:$addr, QPR:$value)>;
2272 def : Pat<(v2f64 (hword_alignedload addrmode6:$addr)),
2273 (VLD1q16 addrmode6:$addr)>, Requires<[IsLE]>;
2274 def : Pat<(hword_alignedstore (v2f64 QPR:$value), addrmode6:$addr),
2275 (VST1q16 addrmode6:$addr, QPR:$value)>, Requires<[IsLE]>;
2276 def : Pat<(v2f64 (byte_alignedload addrmode6:$addr)),
2277 (VLD1q8 addrmode6:$addr)>, Requires<[IsLE]>;
2278 def : Pat<(byte_alignedstore (v2f64 QPR:$value), addrmode6:$addr),
2279 (VST1q8 addrmode6:$addr, QPR:$value)>, Requires<[IsLE]>;
2281 //===----------------------------------------------------------------------===//
2282 // NEON pattern fragments
2283 //===----------------------------------------------------------------------===//
2285 // Extract D sub-registers of Q registers.
2286 def DSubReg_i8_reg : SDNodeXForm<imm, [{
2287 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2288 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/8, MVT::i32);
2290 def DSubReg_i16_reg : SDNodeXForm<imm, [{
2291 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2292 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/4, MVT::i32);
2294 def DSubReg_i32_reg : SDNodeXForm<imm, [{
2295 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2296 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/2, MVT::i32);
2298 def DSubReg_f64_reg : SDNodeXForm<imm, [{
2299 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2300 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue(), MVT::i32);
2303 // Extract S sub-registers of Q/D registers.
2304 def SSubReg_f32_reg : SDNodeXForm<imm, [{
2305 assert(ARM::ssub_3 == ARM::ssub_0+3 && "Unexpected subreg numbering");
2306 return CurDAG->getTargetConstant(ARM::ssub_0 + N->getZExtValue(), MVT::i32);
2309 // Translate lane numbers from Q registers to D subregs.
2310 def SubReg_i8_lane : SDNodeXForm<imm, [{
2311 return CurDAG->getTargetConstant(N->getZExtValue() & 7, MVT::i32);
2313 def SubReg_i16_lane : SDNodeXForm<imm, [{
2314 return CurDAG->getTargetConstant(N->getZExtValue() & 3, MVT::i32);
2316 def SubReg_i32_lane : SDNodeXForm<imm, [{
2317 return CurDAG->getTargetConstant(N->getZExtValue() & 1, MVT::i32);
2320 //===----------------------------------------------------------------------===//
2321 // Instruction Classes
2322 //===----------------------------------------------------------------------===//
2324 // Basic 2-register operations: double- and quad-register.
2325 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2326 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
2327 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
2328 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2329 (ins DPR:$Vm), IIC_VUNAD, OpcodeStr, Dt,"$Vd, $Vm", "",
2330 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm))))]>;
2331 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2332 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
2333 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
2334 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2335 (ins QPR:$Vm), IIC_VUNAQ, OpcodeStr, Dt,"$Vd, $Vm", "",
2336 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vm))))]>;
2338 // Basic 2-register intrinsics, both double- and quad-register.
2339 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2340 bits<2> op17_16, bits<5> op11_7, bit op4,
2341 InstrItinClass itin, string OpcodeStr, string Dt,
2342 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2343 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2344 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2345 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
2346 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2347 bits<2> op17_16, bits<5> op11_7, bit op4,
2348 InstrItinClass itin, string OpcodeStr, string Dt,
2349 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2350 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2351 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2352 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
2354 // Same as above, but not predicated.
2355 class N2VDIntnp<bits<2> op17_16, bits<3> op10_8, bit op7,
2356 InstrItinClass itin, string OpcodeStr, string Dt,
2357 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2358 : N2Vnp<op17_16, op10_8, op7, 0, (outs DPR:$Vd), (ins DPR:$Vm),
2359 itin, OpcodeStr, Dt, ResTy, OpTy,
2360 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
2362 class N2VQIntnp<bits<2> op17_16, bits<3> op10_8, bit op7,
2363 InstrItinClass itin, string OpcodeStr, string Dt,
2364 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2365 : N2Vnp<op17_16, op10_8, op7, 1, (outs QPR:$Vd), (ins QPR:$Vm),
2366 itin, OpcodeStr, Dt, ResTy, OpTy,
2367 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
2369 // Narrow 2-register operations.
2370 class N2VN<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2371 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2372 InstrItinClass itin, string OpcodeStr, string Dt,
2373 ValueType TyD, ValueType TyQ, SDNode OpNode>
2374 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$Vd),
2375 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2376 [(set DPR:$Vd, (TyD (OpNode (TyQ QPR:$Vm))))]>;
2378 // Narrow 2-register intrinsics.
2379 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2380 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2381 InstrItinClass itin, string OpcodeStr, string Dt,
2382 ValueType TyD, ValueType TyQ, SDPatternOperator IntOp>
2383 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$Vd),
2384 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2385 [(set DPR:$Vd, (TyD (IntOp (TyQ QPR:$Vm))))]>;
2387 // Long 2-register operations (currently only used for VMOVL).
2388 class N2VL<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2389 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2390 InstrItinClass itin, string OpcodeStr, string Dt,
2391 ValueType TyQ, ValueType TyD, SDNode OpNode>
2392 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$Vd),
2393 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2394 [(set QPR:$Vd, (TyQ (OpNode (TyD DPR:$Vm))))]>;
2396 // Long 2-register intrinsics.
2397 class N2VLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2398 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2399 InstrItinClass itin, string OpcodeStr, string Dt,
2400 ValueType TyQ, ValueType TyD, SDPatternOperator IntOp>
2401 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$Vd),
2402 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2403 [(set QPR:$Vd, (TyQ (IntOp (TyD DPR:$Vm))))]>;
2405 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
2406 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
2407 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$Vd, DPR:$Vm),
2408 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
2409 OpcodeStr, Dt, "$Vd, $Vm",
2410 "$src1 = $Vd, $src2 = $Vm", []>;
2411 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
2412 InstrItinClass itin, string OpcodeStr, string Dt>
2413 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$Vd, QPR:$Vm),
2414 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$Vd, $Vm",
2415 "$src1 = $Vd, $src2 = $Vm", []>;
2417 // Basic 3-register operations: double- and quad-register.
2418 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2419 InstrItinClass itin, string OpcodeStr, string Dt,
2420 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2421 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2422 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2423 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2424 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
2425 // All of these have a two-operand InstAlias.
2426 let TwoOperandAliasConstraint = "$Vn = $Vd";
2427 let isCommutable = Commutable;
2429 // Same as N3VD but no data type.
2430 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2431 InstrItinClass itin, string OpcodeStr,
2432 ValueType ResTy, ValueType OpTy,
2433 SDNode OpNode, bit Commutable>
2434 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
2435 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2436 OpcodeStr, "$Vd, $Vn, $Vm", "",
2437 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>{
2438 // All of these have a two-operand InstAlias.
2439 let TwoOperandAliasConstraint = "$Vn = $Vd";
2440 let isCommutable = Commutable;
2443 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
2444 InstrItinClass itin, string OpcodeStr, string Dt,
2445 ValueType Ty, SDNode ShOp>
2446 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2447 (outs DPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2448 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2450 (Ty (ShOp (Ty DPR:$Vn),
2451 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),imm:$lane)))))]> {
2452 // All of these have a two-operand InstAlias.
2453 let TwoOperandAliasConstraint = "$Vn = $Vd";
2454 let isCommutable = 0;
2456 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
2457 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
2458 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2459 (outs DPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2460 NVMulSLFrm, IIC_VMULi16D, OpcodeStr, Dt,"$Vd, $Vn, $Vm$lane","",
2462 (Ty (ShOp (Ty DPR:$Vn),
2463 (Ty (NEONvduplane (Ty DPR_8:$Vm), imm:$lane)))))]> {
2464 // All of these have a two-operand InstAlias.
2465 let TwoOperandAliasConstraint = "$Vn = $Vd";
2466 let isCommutable = 0;
2469 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2470 InstrItinClass itin, string OpcodeStr, string Dt,
2471 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2472 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2473 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2474 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2475 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
2476 // All of these have a two-operand InstAlias.
2477 let TwoOperandAliasConstraint = "$Vn = $Vd";
2478 let isCommutable = Commutable;
2480 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2481 InstrItinClass itin, string OpcodeStr,
2482 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2483 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
2484 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2485 OpcodeStr, "$Vd, $Vn, $Vm", "",
2486 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>{
2487 // All of these have a two-operand InstAlias.
2488 let TwoOperandAliasConstraint = "$Vn = $Vd";
2489 let isCommutable = Commutable;
2491 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
2492 InstrItinClass itin, string OpcodeStr, string Dt,
2493 ValueType ResTy, ValueType OpTy, SDNode ShOp>
2494 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2495 (outs QPR:$Vd), (ins QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2496 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2497 [(set (ResTy QPR:$Vd),
2498 (ResTy (ShOp (ResTy QPR:$Vn),
2499 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2501 // All of these have a two-operand InstAlias.
2502 let TwoOperandAliasConstraint = "$Vn = $Vd";
2503 let isCommutable = 0;
2505 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
2506 ValueType ResTy, ValueType OpTy, SDNode ShOp>
2507 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2508 (outs QPR:$Vd), (ins QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2509 NVMulSLFrm, IIC_VMULi16Q, OpcodeStr, Dt,"$Vd, $Vn, $Vm$lane", "",
2510 [(set (ResTy QPR:$Vd),
2511 (ResTy (ShOp (ResTy QPR:$Vn),
2512 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2514 // All of these have a two-operand InstAlias.
2515 let TwoOperandAliasConstraint = "$Vn = $Vd";
2516 let isCommutable = 0;
2519 // Basic 3-register intrinsics, both double- and quad-register.
2520 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2521 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2522 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp, bit Commutable>
2523 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2524 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), f, itin,
2525 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2526 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
2527 // All of these have a two-operand InstAlias.
2528 let TwoOperandAliasConstraint = "$Vn = $Vd";
2529 let isCommutable = Commutable;
2532 class N3VDIntnp<bits<5> op27_23, bits<2> op21_20, bits<4> op11_8, bit op6,
2533 bit op4, Format f, InstrItinClass itin, string OpcodeStr,
2534 string Dt, ValueType ResTy, ValueType OpTy,
2535 SDPatternOperator IntOp, bit Commutable>
2536 : N3Vnp<op27_23, op21_20, op11_8, op6, op4,
2537 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), f, itin, OpcodeStr, Dt,
2538 ResTy, OpTy, IntOp, Commutable,
2539 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>;
2541 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2542 string OpcodeStr, string Dt, ValueType Ty, SDPatternOperator IntOp>
2543 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2544 (outs DPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2545 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2547 (Ty (IntOp (Ty DPR:$Vn),
2548 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),
2550 let isCommutable = 0;
2553 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2554 string OpcodeStr, string Dt, ValueType Ty, SDPatternOperator IntOp>
2555 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2556 (outs DPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2557 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2559 (Ty (IntOp (Ty DPR:$Vn),
2560 (Ty (NEONvduplane (Ty DPR_8:$Vm), imm:$lane)))))]> {
2561 let isCommutable = 0;
2563 class N3VDIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2564 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2565 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2566 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2567 (outs DPR:$Vd), (ins DPR:$Vm, DPR:$Vn), f, itin,
2568 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
2569 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (OpTy DPR:$Vn))))]> {
2570 let TwoOperandAliasConstraint = "$Vm = $Vd";
2571 let isCommutable = 0;
2574 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2575 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2576 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp, bit Commutable>
2577 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2578 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), f, itin,
2579 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2580 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
2581 // All of these have a two-operand InstAlias.
2582 let TwoOperandAliasConstraint = "$Vn = $Vd";
2583 let isCommutable = Commutable;
2586 class N3VQIntnp<bits<5> op27_23, bits<2> op21_20, bits<4> op11_8, bit op6,
2587 bit op4, Format f, InstrItinClass itin, string OpcodeStr,
2588 string Dt, ValueType ResTy, ValueType OpTy,
2589 SDPatternOperator IntOp, bit Commutable>
2590 : N3Vnp<op27_23, op21_20, op11_8, op6, op4,
2591 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), f, itin, OpcodeStr, Dt,
2592 ResTy, OpTy, IntOp, Commutable,
2593 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>;
2595 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2596 string OpcodeStr, string Dt,
2597 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2598 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2599 (outs QPR:$Vd), (ins QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2600 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2601 [(set (ResTy QPR:$Vd),
2602 (ResTy (IntOp (ResTy QPR:$Vn),
2603 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2605 let isCommutable = 0;
2607 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2608 string OpcodeStr, string Dt,
2609 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2610 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2611 (outs QPR:$Vd), (ins QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2612 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2613 [(set (ResTy QPR:$Vd),
2614 (ResTy (IntOp (ResTy QPR:$Vn),
2615 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2617 let isCommutable = 0;
2619 class N3VQIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2620 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2621 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2622 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2623 (outs QPR:$Vd), (ins QPR:$Vm, QPR:$Vn), f, itin,
2624 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
2625 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (OpTy QPR:$Vn))))]> {
2626 let TwoOperandAliasConstraint = "$Vm = $Vd";
2627 let isCommutable = 0;
2630 // Multiply-Add/Sub operations: double- and quad-register.
2631 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2632 InstrItinClass itin, string OpcodeStr, string Dt,
2633 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator OpNode>
2634 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2635 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2636 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2637 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
2638 (Ty (MulOp DPR:$Vn, DPR:$Vm)))))]>;
2640 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2641 string OpcodeStr, string Dt,
2642 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator ShOp>
2643 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2645 (ins DPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2647 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2649 (Ty (ShOp (Ty DPR:$src1),
2651 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),
2653 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2654 string OpcodeStr, string Dt,
2655 ValueType Ty, SDNode MulOp, SDNode ShOp>
2656 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2658 (ins DPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2660 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2662 (Ty (ShOp (Ty DPR:$src1),
2664 (Ty (NEONvduplane (Ty DPR_8:$Vm),
2667 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2668 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
2669 SDPatternOperator MulOp, SDPatternOperator OpNode>
2670 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2671 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2672 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2673 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
2674 (Ty (MulOp QPR:$Vn, QPR:$Vm)))))]>;
2675 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2676 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2677 SDPatternOperator MulOp, SDPatternOperator ShOp>
2678 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2680 (ins QPR:$src1, QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2682 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2683 [(set (ResTy QPR:$Vd),
2684 (ResTy (ShOp (ResTy QPR:$src1),
2685 (ResTy (MulOp QPR:$Vn,
2686 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2688 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2689 string OpcodeStr, string Dt,
2690 ValueType ResTy, ValueType OpTy,
2691 SDNode MulOp, SDNode ShOp>
2692 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2694 (ins QPR:$src1, QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2696 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2697 [(set (ResTy QPR:$Vd),
2698 (ResTy (ShOp (ResTy QPR:$src1),
2699 (ResTy (MulOp QPR:$Vn,
2700 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2703 // Neon Intrinsic-Op instructions (VABA): double- and quad-register.
2704 class N3VDIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2705 InstrItinClass itin, string OpcodeStr, string Dt,
2706 ValueType Ty, SDPatternOperator IntOp, SDNode OpNode>
2707 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2708 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2709 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2710 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
2711 (Ty (IntOp (Ty DPR:$Vn), (Ty DPR:$Vm))))))]>;
2712 class N3VQIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2713 InstrItinClass itin, string OpcodeStr, string Dt,
2714 ValueType Ty, SDPatternOperator IntOp, SDNode OpNode>
2715 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2716 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2717 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2718 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
2719 (Ty (IntOp (Ty QPR:$Vn), (Ty QPR:$Vm))))))]>;
2721 // Neon 3-argument intrinsics, both double- and quad-register.
2722 // The destination register is also used as the first source operand register.
2723 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2724 InstrItinClass itin, string OpcodeStr, string Dt,
2725 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2726 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2727 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2728 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2729 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$src1),
2730 (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>;
2731 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2732 InstrItinClass itin, string OpcodeStr, string Dt,
2733 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2734 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2735 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2736 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2737 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$src1),
2738 (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>;
2740 // Long Multiply-Add/Sub operations.
2741 class N3VLMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2742 InstrItinClass itin, string OpcodeStr, string Dt,
2743 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2744 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2745 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2746 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2747 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
2748 (TyQ (MulOp (TyD DPR:$Vn),
2749 (TyD DPR:$Vm)))))]>;
2750 class N3VLMulOpSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2751 InstrItinClass itin, string OpcodeStr, string Dt,
2752 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2753 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$Vd),
2754 (ins QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2756 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2758 (OpNode (TyQ QPR:$src1),
2759 (TyQ (MulOp (TyD DPR:$Vn),
2760 (TyD (NEONvduplane (TyD DPR_VFP2:$Vm),
2762 class N3VLMulOpSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2763 InstrItinClass itin, string OpcodeStr, string Dt,
2764 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2765 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$Vd),
2766 (ins QPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2768 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2770 (OpNode (TyQ QPR:$src1),
2771 (TyQ (MulOp (TyD DPR:$Vn),
2772 (TyD (NEONvduplane (TyD DPR_8:$Vm),
2775 // Long Intrinsic-Op vector operations with explicit extend (VABAL).
2776 class N3VLIntExtOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2777 InstrItinClass itin, string OpcodeStr, string Dt,
2778 ValueType TyQ, ValueType TyD, SDPatternOperator IntOp, SDNode ExtOp,
2780 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2781 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2782 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2783 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
2784 (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
2785 (TyD DPR:$Vm)))))))]>;
2787 // Neon Long 3-argument intrinsic. The destination register is
2788 // a quad-register and is also used as the first source operand register.
2789 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2790 InstrItinClass itin, string OpcodeStr, string Dt,
2791 ValueType TyQ, ValueType TyD, SDPatternOperator IntOp>
2792 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2793 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2794 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2796 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$Vn), (TyD DPR:$Vm))))]>;
2797 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2798 string OpcodeStr, string Dt,
2799 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2800 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2802 (ins QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2804 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2805 [(set (ResTy QPR:$Vd),
2806 (ResTy (IntOp (ResTy QPR:$src1),
2808 (OpTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2810 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2811 InstrItinClass itin, string OpcodeStr, string Dt,
2812 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2813 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2815 (ins QPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2817 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2818 [(set (ResTy QPR:$Vd),
2819 (ResTy (IntOp (ResTy QPR:$src1),
2821 (OpTy (NEONvduplane (OpTy DPR_8:$Vm),
2824 // Narrowing 3-register intrinsics.
2825 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2826 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
2827 SDPatternOperator IntOp, bit Commutable>
2828 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2829 (outs DPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINi4D,
2830 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2831 [(set DPR:$Vd, (TyD (IntOp (TyQ QPR:$Vn), (TyQ QPR:$Vm))))]> {
2832 let isCommutable = Commutable;
2835 // Long 3-register operations.
2836 class N3VL<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2837 InstrItinClass itin, string OpcodeStr, string Dt,
2838 ValueType TyQ, ValueType TyD, SDNode OpNode, bit Commutable>
2839 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2840 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2841 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2842 [(set QPR:$Vd, (TyQ (OpNode (TyD DPR:$Vn), (TyD DPR:$Vm))))]> {
2843 let isCommutable = Commutable;
2845 class N3VLSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2846 InstrItinClass itin, string OpcodeStr, string Dt,
2847 ValueType TyQ, ValueType TyD, SDNode OpNode>
2848 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2849 (outs QPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2850 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2852 (TyQ (OpNode (TyD DPR:$Vn),
2853 (TyD (NEONvduplane (TyD DPR_VFP2:$Vm),imm:$lane)))))]>;
2854 class N3VLSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2855 InstrItinClass itin, string OpcodeStr, string Dt,
2856 ValueType TyQ, ValueType TyD, SDNode OpNode>
2857 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2858 (outs QPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2859 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2861 (TyQ (OpNode (TyD DPR:$Vn),
2862 (TyD (NEONvduplane (TyD DPR_8:$Vm), imm:$lane)))))]>;
2864 // Long 3-register operations with explicitly extended operands.
2865 class N3VLExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2866 InstrItinClass itin, string OpcodeStr, string Dt,
2867 ValueType TyQ, ValueType TyD, SDNode OpNode, SDNode ExtOp,
2869 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2870 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2871 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2872 [(set QPR:$Vd, (OpNode (TyQ (ExtOp (TyD DPR:$Vn))),
2873 (TyQ (ExtOp (TyD DPR:$Vm)))))]> {
2874 let isCommutable = Commutable;
2877 // Long 3-register intrinsics with explicit extend (VABDL).
2878 class N3VLIntExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2879 InstrItinClass itin, string OpcodeStr, string Dt,
2880 ValueType TyQ, ValueType TyD, SDPatternOperator IntOp, SDNode ExtOp,
2882 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2883 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2884 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2885 [(set QPR:$Vd, (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
2886 (TyD DPR:$Vm))))))]> {
2887 let isCommutable = Commutable;
2890 // Long 3-register intrinsics.
2891 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2892 InstrItinClass itin, string OpcodeStr, string Dt,
2893 ValueType TyQ, ValueType TyD, SDPatternOperator IntOp, bit Commutable>
2894 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2895 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2896 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2897 [(set QPR:$Vd, (TyQ (IntOp (TyD DPR:$Vn), (TyD DPR:$Vm))))]> {
2898 let isCommutable = Commutable;
2900 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2901 string OpcodeStr, string Dt,
2902 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2903 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2904 (outs QPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2905 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2906 [(set (ResTy QPR:$Vd),
2907 (ResTy (IntOp (OpTy DPR:$Vn),
2908 (OpTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2910 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2911 InstrItinClass itin, string OpcodeStr, string Dt,
2912 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2913 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2914 (outs QPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2915 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2916 [(set (ResTy QPR:$Vd),
2917 (ResTy (IntOp (OpTy DPR:$Vn),
2918 (OpTy (NEONvduplane (OpTy DPR_8:$Vm),
2921 // Wide 3-register operations.
2922 class N3VW<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2923 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
2924 SDNode OpNode, SDNode ExtOp, bit Commutable>
2925 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2926 (outs QPR:$Vd), (ins QPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VSUBiD,
2927 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2928 [(set QPR:$Vd, (OpNode (TyQ QPR:$Vn),
2929 (TyQ (ExtOp (TyD DPR:$Vm)))))]> {
2930 // All of these have a two-operand InstAlias.
2931 let TwoOperandAliasConstraint = "$Vn = $Vd";
2932 let isCommutable = Commutable;
2935 // Pairwise long 2-register intrinsics, both double- and quad-register.
2936 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2937 bits<2> op17_16, bits<5> op11_7, bit op4,
2938 string OpcodeStr, string Dt,
2939 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2940 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2941 (ins DPR:$Vm), IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
2942 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
2943 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2944 bits<2> op17_16, bits<5> op11_7, bit op4,
2945 string OpcodeStr, string Dt,
2946 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2947 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2948 (ins QPR:$Vm), IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
2949 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
2951 // Pairwise long 2-register accumulate intrinsics,
2952 // both double- and quad-register.
2953 // The destination register is also used as the first source operand register.
2954 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2955 bits<2> op17_16, bits<5> op11_7, bit op4,
2956 string OpcodeStr, string Dt,
2957 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2958 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
2959 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vm), IIC_VPALiD,
2960 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2961 [(set DPR:$Vd, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$Vm))))]>;
2962 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2963 bits<2> op17_16, bits<5> op11_7, bit op4,
2964 string OpcodeStr, string Dt,
2965 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2966 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
2967 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vm), IIC_VPALiQ,
2968 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2969 [(set QPR:$Vd, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$Vm))))]>;
2971 // Shift by immediate,
2972 // both double- and quad-register.
2973 let TwoOperandAliasConstraint = "$Vm = $Vd" in {
2974 class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2975 Format f, InstrItinClass itin, Operand ImmTy,
2976 string OpcodeStr, string Dt, ValueType Ty, SDNode OpNode>
2977 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2978 (outs DPR:$Vd), (ins DPR:$Vm, ImmTy:$SIMM), f, itin,
2979 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2980 [(set DPR:$Vd, (Ty (OpNode (Ty DPR:$Vm), (i32 imm:$SIMM))))]>;
2981 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2982 Format f, InstrItinClass itin, Operand ImmTy,
2983 string OpcodeStr, string Dt, ValueType Ty, SDNode OpNode>
2984 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2985 (outs QPR:$Vd), (ins QPR:$Vm, ImmTy:$SIMM), f, itin,
2986 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2987 [(set QPR:$Vd, (Ty (OpNode (Ty QPR:$Vm), (i32 imm:$SIMM))))]>;
2990 // Long shift by immediate.
2991 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2992 string OpcodeStr, string Dt,
2993 ValueType ResTy, ValueType OpTy, Operand ImmTy, SDNode OpNode>
2994 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2995 (outs QPR:$Vd), (ins DPR:$Vm, ImmTy:$SIMM), N2RegVShLFrm,
2996 IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2997 [(set QPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm),
2998 (i32 imm:$SIMM))))]>;
3000 // Narrow shift by immediate.
3001 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
3002 InstrItinClass itin, string OpcodeStr, string Dt,
3003 ValueType ResTy, ValueType OpTy, Operand ImmTy, SDNode OpNode>
3004 : N2VImm<op24, op23, op11_8, op7, op6, op4,
3005 (outs DPR:$Vd), (ins QPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, itin,
3006 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
3007 [(set DPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vm),
3008 (i32 imm:$SIMM))))]>;
3010 // Shift right by immediate and accumulate,
3011 // both double- and quad-register.
3012 let TwoOperandAliasConstraint = "$Vm = $Vd" in {
3013 class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
3014 Operand ImmTy, string OpcodeStr, string Dt,
3015 ValueType Ty, SDNode ShOp>
3016 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
3017 (ins DPR:$src1, DPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, IIC_VPALiD,
3018 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
3019 [(set DPR:$Vd, (Ty (add DPR:$src1,
3020 (Ty (ShOp DPR:$Vm, (i32 imm:$SIMM))))))]>;
3021 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
3022 Operand ImmTy, string OpcodeStr, string Dt,
3023 ValueType Ty, SDNode ShOp>
3024 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
3025 (ins QPR:$src1, QPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, IIC_VPALiD,
3026 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
3027 [(set QPR:$Vd, (Ty (add QPR:$src1,
3028 (Ty (ShOp QPR:$Vm, (i32 imm:$SIMM))))))]>;
3031 // Shift by immediate and insert,
3032 // both double- and quad-register.
3033 let TwoOperandAliasConstraint = "$Vm = $Vd" in {
3034 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
3035 Operand ImmTy, Format f, string OpcodeStr, string Dt,
3036 ValueType Ty,SDNode ShOp>
3037 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
3038 (ins DPR:$src1, DPR:$Vm, ImmTy:$SIMM), f, IIC_VSHLiD,
3039 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
3040 [(set DPR:$Vd, (Ty (ShOp DPR:$src1, DPR:$Vm, (i32 imm:$SIMM))))]>;
3041 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
3042 Operand ImmTy, Format f, string OpcodeStr, string Dt,
3043 ValueType Ty,SDNode ShOp>
3044 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
3045 (ins QPR:$src1, QPR:$Vm, ImmTy:$SIMM), f, IIC_VSHLiQ,
3046 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
3047 [(set QPR:$Vd, (Ty (ShOp QPR:$src1, QPR:$Vm, (i32 imm:$SIMM))))]>;
3050 // Convert, with fractional bits immediate,
3051 // both double- and quad-register.
3052 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
3053 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
3054 SDPatternOperator IntOp>
3055 : N2VImm<op24, op23, op11_8, op7, 0, op4,
3056 (outs DPR:$Vd), (ins DPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
3057 IIC_VUNAD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
3058 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (i32 imm:$SIMM))))]>;
3059 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
3060 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
3061 SDPatternOperator IntOp>
3062 : N2VImm<op24, op23, op11_8, op7, 1, op4,
3063 (outs QPR:$Vd), (ins QPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
3064 IIC_VUNAQ, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
3065 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (i32 imm:$SIMM))))]>;
3067 //===----------------------------------------------------------------------===//
3069 //===----------------------------------------------------------------------===//
3071 // Abbreviations used in multiclass suffixes:
3072 // Q = quarter int (8 bit) elements
3073 // H = half int (16 bit) elements
3074 // S = single int (32 bit) elements
3075 // D = double int (64 bit) elements
3077 // Neon 2-register vector operations and intrinsics.
3079 // Neon 2-register comparisons.
3080 // source operand element sizes of 8, 16 and 32 bits:
3081 multiclass N2V_QHS_cmp<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3082 bits<5> op11_7, bit op4, string opc, string Dt,
3083 string asm, SDNode OpNode> {
3084 // 64-bit vector types.
3085 def v8i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 0, op4,
3086 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
3087 opc, !strconcat(Dt, "8"), asm, "",
3088 [(set DPR:$Vd, (v8i8 (OpNode (v8i8 DPR:$Vm))))]>;
3089 def v4i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
3090 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
3091 opc, !strconcat(Dt, "16"), asm, "",
3092 [(set DPR:$Vd, (v4i16 (OpNode (v4i16 DPR:$Vm))))]>;
3093 def v2i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
3094 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
3095 opc, !strconcat(Dt, "32"), asm, "",
3096 [(set DPR:$Vd, (v2i32 (OpNode (v2i32 DPR:$Vm))))]>;
3097 def v2f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
3098 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
3099 opc, "f32", asm, "",
3100 [(set DPR:$Vd, (v2i32 (OpNode (v2f32 DPR:$Vm))))]> {
3101 let Inst{10} = 1; // overwrite F = 1
3104 // 128-bit vector types.
3105 def v16i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 1, op4,
3106 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
3107 opc, !strconcat(Dt, "8"), asm, "",
3108 [(set QPR:$Vd, (v16i8 (OpNode (v16i8 QPR:$Vm))))]>;
3109 def v8i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
3110 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
3111 opc, !strconcat(Dt, "16"), asm, "",
3112 [(set QPR:$Vd, (v8i16 (OpNode (v8i16 QPR:$Vm))))]>;
3113 def v4i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
3114 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
3115 opc, !strconcat(Dt, "32"), asm, "",
3116 [(set QPR:$Vd, (v4i32 (OpNode (v4i32 QPR:$Vm))))]>;
3117 def v4f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
3118 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
3119 opc, "f32", asm, "",
3120 [(set QPR:$Vd, (v4i32 (OpNode (v4f32 QPR:$Vm))))]> {
3121 let Inst{10} = 1; // overwrite F = 1
3126 // Neon 2-register vector intrinsics,
3127 // element sizes of 8, 16 and 32 bits:
3128 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3129 bits<5> op11_7, bit op4,
3130 InstrItinClass itinD, InstrItinClass itinQ,
3131 string OpcodeStr, string Dt, SDPatternOperator IntOp> {
3132 // 64-bit vector types.
3133 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3134 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
3135 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3136 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
3137 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3138 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
3140 // 128-bit vector types.
3141 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3142 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
3143 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3144 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
3145 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3146 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
3150 // Neon Narrowing 2-register vector operations,
3151 // source operand element sizes of 16, 32 and 64 bits:
3152 multiclass N2VN_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3153 bits<5> op11_7, bit op6, bit op4,
3154 InstrItinClass itin, string OpcodeStr, string Dt,
3156 def v8i8 : N2VN<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
3157 itin, OpcodeStr, !strconcat(Dt, "16"),
3158 v8i8, v8i16, OpNode>;
3159 def v4i16 : N2VN<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
3160 itin, OpcodeStr, !strconcat(Dt, "32"),
3161 v4i16, v4i32, OpNode>;
3162 def v2i32 : N2VN<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
3163 itin, OpcodeStr, !strconcat(Dt, "64"),
3164 v2i32, v2i64, OpNode>;
3167 // Neon Narrowing 2-register vector intrinsics,
3168 // source operand element sizes of 16, 32 and 64 bits:
3169 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3170 bits<5> op11_7, bit op6, bit op4,
3171 InstrItinClass itin, string OpcodeStr, string Dt,
3172 SDPatternOperator IntOp> {
3173 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
3174 itin, OpcodeStr, !strconcat(Dt, "16"),
3175 v8i8, v8i16, IntOp>;
3176 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
3177 itin, OpcodeStr, !strconcat(Dt, "32"),
3178 v4i16, v4i32, IntOp>;
3179 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
3180 itin, OpcodeStr, !strconcat(Dt, "64"),
3181 v2i32, v2i64, IntOp>;
3185 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
3186 // source operand element sizes of 16, 32 and 64 bits:
3187 multiclass N2VL_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
3188 string OpcodeStr, string Dt, SDNode OpNode> {
3189 def v8i16 : N2VL<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
3190 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode>;
3191 def v4i32 : N2VL<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
3192 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
3193 def v2i64 : N2VL<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
3194 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
3198 // Neon 3-register vector operations.
3200 // First with only element sizes of 8, 16 and 32 bits:
3201 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3202 InstrItinClass itinD16, InstrItinClass itinD32,
3203 InstrItinClass itinQ16, InstrItinClass itinQ32,
3204 string OpcodeStr, string Dt,
3205 SDNode OpNode, bit Commutable = 0> {
3206 // 64-bit vector types.
3207 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
3208 OpcodeStr, !strconcat(Dt, "8"),
3209 v8i8, v8i8, OpNode, Commutable>;
3210 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
3211 OpcodeStr, !strconcat(Dt, "16"),
3212 v4i16, v4i16, OpNode, Commutable>;
3213 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
3214 OpcodeStr, !strconcat(Dt, "32"),
3215 v2i32, v2i32, OpNode, Commutable>;
3217 // 128-bit vector types.
3218 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
3219 OpcodeStr, !strconcat(Dt, "8"),
3220 v16i8, v16i8, OpNode, Commutable>;
3221 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
3222 OpcodeStr, !strconcat(Dt, "16"),
3223 v8i16, v8i16, OpNode, Commutable>;
3224 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
3225 OpcodeStr, !strconcat(Dt, "32"),
3226 v4i32, v4i32, OpNode, Commutable>;
3229 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, SDNode ShOp> {
3230 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, "i16", v4i16, ShOp>;
3231 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, "i32", v2i32, ShOp>;
3232 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, "i16", v8i16, v4i16, ShOp>;
3233 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, "i32",
3234 v4i32, v2i32, ShOp>;
3237 // ....then also with element size 64 bits:
3238 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3239 InstrItinClass itinD, InstrItinClass itinQ,
3240 string OpcodeStr, string Dt,
3241 SDNode OpNode, bit Commutable = 0>
3242 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
3243 OpcodeStr, Dt, OpNode, Commutable> {
3244 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
3245 OpcodeStr, !strconcat(Dt, "64"),
3246 v1i64, v1i64, OpNode, Commutable>;
3247 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
3248 OpcodeStr, !strconcat(Dt, "64"),
3249 v2i64, v2i64, OpNode, Commutable>;
3253 // Neon 3-register vector intrinsics.
3255 // First with only element sizes of 16 and 32 bits:
3256 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3257 InstrItinClass itinD16, InstrItinClass itinD32,
3258 InstrItinClass itinQ16, InstrItinClass itinQ32,
3259 string OpcodeStr, string Dt,
3260 SDPatternOperator IntOp, bit Commutable = 0> {
3261 // 64-bit vector types.
3262 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, f, itinD16,
3263 OpcodeStr, !strconcat(Dt, "16"),
3264 v4i16, v4i16, IntOp, Commutable>;
3265 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, f, itinD32,
3266 OpcodeStr, !strconcat(Dt, "32"),
3267 v2i32, v2i32, IntOp, Commutable>;
3269 // 128-bit vector types.
3270 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, f, itinQ16,
3271 OpcodeStr, !strconcat(Dt, "16"),
3272 v8i16, v8i16, IntOp, Commutable>;
3273 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, f, itinQ32,
3274 OpcodeStr, !strconcat(Dt, "32"),
3275 v4i32, v4i32, IntOp, Commutable>;
3277 multiclass N3VInt_HSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3278 InstrItinClass itinD16, InstrItinClass itinD32,
3279 InstrItinClass itinQ16, InstrItinClass itinQ32,
3280 string OpcodeStr, string Dt,
3281 SDPatternOperator IntOp> {
3282 // 64-bit vector types.
3283 def v4i16 : N3VDIntSh<op24, op23, 0b01, op11_8, op4, f, itinD16,
3284 OpcodeStr, !strconcat(Dt, "16"),
3285 v4i16, v4i16, IntOp>;
3286 def v2i32 : N3VDIntSh<op24, op23, 0b10, op11_8, op4, f, itinD32,
3287 OpcodeStr, !strconcat(Dt, "32"),
3288 v2i32, v2i32, IntOp>;
3290 // 128-bit vector types.
3291 def v8i16 : N3VQIntSh<op24, op23, 0b01, op11_8, op4, f, itinQ16,
3292 OpcodeStr, !strconcat(Dt, "16"),
3293 v8i16, v8i16, IntOp>;
3294 def v4i32 : N3VQIntSh<op24, op23, 0b10, op11_8, op4, f, itinQ32,
3295 OpcodeStr, !strconcat(Dt, "32"),
3296 v4i32, v4i32, IntOp>;
3299 multiclass N3VIntSL_HS<bits<4> op11_8,
3300 InstrItinClass itinD16, InstrItinClass itinD32,
3301 InstrItinClass itinQ16, InstrItinClass itinQ32,
3302 string OpcodeStr, string Dt, SDPatternOperator IntOp> {
3303 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
3304 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
3305 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
3306 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
3307 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
3308 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
3309 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
3310 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
3313 // ....then also with element size of 8 bits:
3314 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3315 InstrItinClass itinD16, InstrItinClass itinD32,
3316 InstrItinClass itinQ16, InstrItinClass itinQ32,
3317 string OpcodeStr, string Dt,
3318 SDPatternOperator IntOp, bit Commutable = 0>
3319 : N3VInt_HS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3320 OpcodeStr, Dt, IntOp, Commutable> {
3321 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, f, itinD16,
3322 OpcodeStr, !strconcat(Dt, "8"),
3323 v8i8, v8i8, IntOp, Commutable>;
3324 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, f, itinQ16,
3325 OpcodeStr, !strconcat(Dt, "8"),
3326 v16i8, v16i8, IntOp, Commutable>;
3328 multiclass N3VInt_QHSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3329 InstrItinClass itinD16, InstrItinClass itinD32,
3330 InstrItinClass itinQ16, InstrItinClass itinQ32,
3331 string OpcodeStr, string Dt,
3332 SDPatternOperator IntOp>
3333 : N3VInt_HSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3334 OpcodeStr, Dt, IntOp> {
3335 def v8i8 : N3VDIntSh<op24, op23, 0b00, op11_8, op4, f, itinD16,
3336 OpcodeStr, !strconcat(Dt, "8"),
3338 def v16i8 : N3VQIntSh<op24, op23, 0b00, op11_8, op4, f, itinQ16,
3339 OpcodeStr, !strconcat(Dt, "8"),
3340 v16i8, v16i8, IntOp>;
3344 // ....then also with element size of 64 bits:
3345 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3346 InstrItinClass itinD16, InstrItinClass itinD32,
3347 InstrItinClass itinQ16, InstrItinClass itinQ32,
3348 string OpcodeStr, string Dt,
3349 SDPatternOperator IntOp, bit Commutable = 0>
3350 : N3VInt_QHS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3351 OpcodeStr, Dt, IntOp, Commutable> {
3352 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, f, itinD32,
3353 OpcodeStr, !strconcat(Dt, "64"),
3354 v1i64, v1i64, IntOp, Commutable>;
3355 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, f, itinQ32,
3356 OpcodeStr, !strconcat(Dt, "64"),
3357 v2i64, v2i64, IntOp, Commutable>;
3359 multiclass N3VInt_QHSDSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3360 InstrItinClass itinD16, InstrItinClass itinD32,
3361 InstrItinClass itinQ16, InstrItinClass itinQ32,
3362 string OpcodeStr, string Dt,
3363 SDPatternOperator IntOp>
3364 : N3VInt_QHSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3365 OpcodeStr, Dt, IntOp> {
3366 def v1i64 : N3VDIntSh<op24, op23, 0b11, op11_8, op4, f, itinD32,
3367 OpcodeStr, !strconcat(Dt, "64"),
3368 v1i64, v1i64, IntOp>;
3369 def v2i64 : N3VQIntSh<op24, op23, 0b11, op11_8, op4, f, itinQ32,
3370 OpcodeStr, !strconcat(Dt, "64"),
3371 v2i64, v2i64, IntOp>;
3374 // Neon Narrowing 3-register vector intrinsics,
3375 // source operand element sizes of 16, 32 and 64 bits:
3376 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3377 string OpcodeStr, string Dt,
3378 SDPatternOperator IntOp, bit Commutable = 0> {
3379 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
3380 OpcodeStr, !strconcat(Dt, "16"),
3381 v8i8, v8i16, IntOp, Commutable>;
3382 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
3383 OpcodeStr, !strconcat(Dt, "32"),
3384 v4i16, v4i32, IntOp, Commutable>;
3385 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
3386 OpcodeStr, !strconcat(Dt, "64"),
3387 v2i32, v2i64, IntOp, Commutable>;
3391 // Neon Long 3-register vector operations.
3393 multiclass N3VL_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3394 InstrItinClass itin16, InstrItinClass itin32,
3395 string OpcodeStr, string Dt,
3396 SDNode OpNode, bit Commutable = 0> {
3397 def v8i16 : N3VL<op24, op23, 0b00, op11_8, op4, itin16,
3398 OpcodeStr, !strconcat(Dt, "8"),
3399 v8i16, v8i8, OpNode, Commutable>;
3400 def v4i32 : N3VL<op24, op23, 0b01, op11_8, op4, itin16,
3401 OpcodeStr, !strconcat(Dt, "16"),
3402 v4i32, v4i16, OpNode, Commutable>;
3403 def v2i64 : N3VL<op24, op23, 0b10, op11_8, op4, itin32,
3404 OpcodeStr, !strconcat(Dt, "32"),
3405 v2i64, v2i32, OpNode, Commutable>;
3408 multiclass N3VLSL_HS<bit op24, bits<4> op11_8,
3409 InstrItinClass itin, string OpcodeStr, string Dt,
3411 def v4i16 : N3VLSL16<op24, 0b01, op11_8, itin, OpcodeStr,
3412 !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
3413 def v2i32 : N3VLSL<op24, 0b10, op11_8, itin, OpcodeStr,
3414 !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
3417 multiclass N3VLExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3418 InstrItinClass itin16, InstrItinClass itin32,
3419 string OpcodeStr, string Dt,
3420 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
3421 def v8i16 : N3VLExt<op24, op23, 0b00, op11_8, op4, itin16,
3422 OpcodeStr, !strconcat(Dt, "8"),
3423 v8i16, v8i8, OpNode, ExtOp, Commutable>;
3424 def v4i32 : N3VLExt<op24, op23, 0b01, op11_8, op4, itin16,
3425 OpcodeStr, !strconcat(Dt, "16"),
3426 v4i32, v4i16, OpNode, ExtOp, Commutable>;
3427 def v2i64 : N3VLExt<op24, op23, 0b10, op11_8, op4, itin32,
3428 OpcodeStr, !strconcat(Dt, "32"),
3429 v2i64, v2i32, OpNode, ExtOp, Commutable>;
3432 // Neon Long 3-register vector intrinsics.
3434 // First with only element sizes of 16 and 32 bits:
3435 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
3436 InstrItinClass itin16, InstrItinClass itin32,
3437 string OpcodeStr, string Dt,
3438 SDPatternOperator IntOp, bit Commutable = 0> {
3439 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin16,
3440 OpcodeStr, !strconcat(Dt, "16"),
3441 v4i32, v4i16, IntOp, Commutable>;
3442 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin32,
3443 OpcodeStr, !strconcat(Dt, "32"),
3444 v2i64, v2i32, IntOp, Commutable>;
3447 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
3448 InstrItinClass itin, string OpcodeStr, string Dt,
3449 SDPatternOperator IntOp> {
3450 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
3451 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
3452 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
3453 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3456 // ....then also with element size of 8 bits:
3457 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3458 InstrItinClass itin16, InstrItinClass itin32,
3459 string OpcodeStr, string Dt,
3460 SDPatternOperator IntOp, bit Commutable = 0>
3461 : N3VLInt_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt,
3462 IntOp, Commutable> {
3463 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin16,
3464 OpcodeStr, !strconcat(Dt, "8"),
3465 v8i16, v8i8, IntOp, Commutable>;
3468 // ....with explicit extend (VABDL).
3469 multiclass N3VLIntExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3470 InstrItinClass itin, string OpcodeStr, string Dt,
3471 SDPatternOperator IntOp, SDNode ExtOp, bit Commutable = 0> {
3472 def v8i16 : N3VLIntExt<op24, op23, 0b00, op11_8, op4, itin,
3473 OpcodeStr, !strconcat(Dt, "8"),
3474 v8i16, v8i8, IntOp, ExtOp, Commutable>;
3475 def v4i32 : N3VLIntExt<op24, op23, 0b01, op11_8, op4, itin,
3476 OpcodeStr, !strconcat(Dt, "16"),
3477 v4i32, v4i16, IntOp, ExtOp, Commutable>;
3478 def v2i64 : N3VLIntExt<op24, op23, 0b10, op11_8, op4, itin,
3479 OpcodeStr, !strconcat(Dt, "32"),
3480 v2i64, v2i32, IntOp, ExtOp, Commutable>;
3484 // Neon Wide 3-register vector intrinsics,
3485 // source operand element sizes of 8, 16 and 32 bits:
3486 multiclass N3VW_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3487 string OpcodeStr, string Dt,
3488 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
3489 def v8i16 : N3VW<op24, op23, 0b00, op11_8, op4,
3490 OpcodeStr, !strconcat(Dt, "8"),
3491 v8i16, v8i8, OpNode, ExtOp, Commutable>;
3492 def v4i32 : N3VW<op24, op23, 0b01, op11_8, op4,
3493 OpcodeStr, !strconcat(Dt, "16"),
3494 v4i32, v4i16, OpNode, ExtOp, Commutable>;
3495 def v2i64 : N3VW<op24, op23, 0b10, op11_8, op4,
3496 OpcodeStr, !strconcat(Dt, "32"),
3497 v2i64, v2i32, OpNode, ExtOp, Commutable>;
3501 // Neon Multiply-Op vector operations,
3502 // element sizes of 8, 16 and 32 bits:
3503 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3504 InstrItinClass itinD16, InstrItinClass itinD32,
3505 InstrItinClass itinQ16, InstrItinClass itinQ32,
3506 string OpcodeStr, string Dt, SDNode OpNode> {
3507 // 64-bit vector types.
3508 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
3509 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
3510 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
3511 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
3512 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
3513 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
3515 // 128-bit vector types.
3516 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
3517 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
3518 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
3519 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
3520 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
3521 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
3524 multiclass N3VMulOpSL_HS<bits<4> op11_8,
3525 InstrItinClass itinD16, InstrItinClass itinD32,
3526 InstrItinClass itinQ16, InstrItinClass itinQ32,
3527 string OpcodeStr, string Dt, SDNode ShOp> {
3528 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
3529 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
3530 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
3531 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
3532 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
3533 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
3535 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
3536 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
3540 // Neon Intrinsic-Op vector operations,
3541 // element sizes of 8, 16 and 32 bits:
3542 multiclass N3VIntOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3543 InstrItinClass itinD, InstrItinClass itinQ,
3544 string OpcodeStr, string Dt, SDPatternOperator IntOp,
3546 // 64-bit vector types.
3547 def v8i8 : N3VDIntOp<op24, op23, 0b00, op11_8, op4, itinD,
3548 OpcodeStr, !strconcat(Dt, "8"), v8i8, IntOp, OpNode>;
3549 def v4i16 : N3VDIntOp<op24, op23, 0b01, op11_8, op4, itinD,
3550 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp, OpNode>;
3551 def v2i32 : N3VDIntOp<op24, op23, 0b10, op11_8, op4, itinD,
3552 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp, OpNode>;
3554 // 128-bit vector types.
3555 def v16i8 : N3VQIntOp<op24, op23, 0b00, op11_8, op4, itinQ,
3556 OpcodeStr, !strconcat(Dt, "8"), v16i8, IntOp, OpNode>;
3557 def v8i16 : N3VQIntOp<op24, op23, 0b01, op11_8, op4, itinQ,
3558 OpcodeStr, !strconcat(Dt, "16"), v8i16, IntOp, OpNode>;
3559 def v4i32 : N3VQIntOp<op24, op23, 0b10, op11_8, op4, itinQ,
3560 OpcodeStr, !strconcat(Dt, "32"), v4i32, IntOp, OpNode>;
3563 // Neon 3-argument intrinsics,
3564 // element sizes of 8, 16 and 32 bits:
3565 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3566 InstrItinClass itinD, InstrItinClass itinQ,
3567 string OpcodeStr, string Dt, SDPatternOperator IntOp> {
3568 // 64-bit vector types.
3569 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, itinD,
3570 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
3571 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, itinD,
3572 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
3573 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, itinD,
3574 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
3576 // 128-bit vector types.
3577 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, itinQ,
3578 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
3579 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, itinQ,
3580 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
3581 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, itinQ,
3582 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
3586 // Neon Long Multiply-Op vector operations,
3587 // element sizes of 8, 16 and 32 bits:
3588 multiclass N3VLMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3589 InstrItinClass itin16, InstrItinClass itin32,
3590 string OpcodeStr, string Dt, SDNode MulOp,
3592 def v8i16 : N3VLMulOp<op24, op23, 0b00, op11_8, op4, itin16, OpcodeStr,
3593 !strconcat(Dt, "8"), v8i16, v8i8, MulOp, OpNode>;
3594 def v4i32 : N3VLMulOp<op24, op23, 0b01, op11_8, op4, itin16, OpcodeStr,
3595 !strconcat(Dt, "16"), v4i32, v4i16, MulOp, OpNode>;
3596 def v2i64 : N3VLMulOp<op24, op23, 0b10, op11_8, op4, itin32, OpcodeStr,
3597 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
3600 multiclass N3VLMulOpSL_HS<bit op24, bits<4> op11_8, string OpcodeStr,
3601 string Dt, SDNode MulOp, SDNode OpNode> {
3602 def v4i16 : N3VLMulOpSL16<op24, 0b01, op11_8, IIC_VMACi16D, OpcodeStr,
3603 !strconcat(Dt,"16"), v4i32, v4i16, MulOp, OpNode>;
3604 def v2i32 : N3VLMulOpSL<op24, 0b10, op11_8, IIC_VMACi32D, OpcodeStr,
3605 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
3609 // Neon Long 3-argument intrinsics.
3611 // First with only element sizes of 16 and 32 bits:
3612 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
3613 InstrItinClass itin16, InstrItinClass itin32,
3614 string OpcodeStr, string Dt, SDPatternOperator IntOp> {
3615 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, itin16,
3616 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
3617 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, itin32,
3618 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3621 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
3622 string OpcodeStr, string Dt, SDPatternOperator IntOp> {
3623 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
3624 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
3625 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
3626 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3629 // ....then also with element size of 8 bits:
3630 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3631 InstrItinClass itin16, InstrItinClass itin32,
3632 string OpcodeStr, string Dt, SDPatternOperator IntOp>
3633 : N3VLInt3_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt, IntOp> {
3634 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, itin16,
3635 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
3638 // ....with explicit extend (VABAL).
3639 multiclass N3VLIntExtOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3640 InstrItinClass itin, string OpcodeStr, string Dt,
3641 SDPatternOperator IntOp, SDNode ExtOp, SDNode OpNode> {
3642 def v8i16 : N3VLIntExtOp<op24, op23, 0b00, op11_8, op4, itin,
3643 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8,
3644 IntOp, ExtOp, OpNode>;
3645 def v4i32 : N3VLIntExtOp<op24, op23, 0b01, op11_8, op4, itin,
3646 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16,
3647 IntOp, ExtOp, OpNode>;
3648 def v2i64 : N3VLIntExtOp<op24, op23, 0b10, op11_8, op4, itin,
3649 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32,
3650 IntOp, ExtOp, OpNode>;
3654 // Neon Pairwise long 2-register intrinsics,
3655 // element sizes of 8, 16 and 32 bits:
3656 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3657 bits<5> op11_7, bit op4,
3658 string OpcodeStr, string Dt, SDPatternOperator IntOp> {
3659 // 64-bit vector types.
3660 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3661 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
3662 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3663 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
3664 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3665 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
3667 // 128-bit vector types.
3668 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3669 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
3670 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3671 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
3672 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3673 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
3677 // Neon Pairwise long 2-register accumulate intrinsics,
3678 // element sizes of 8, 16 and 32 bits:
3679 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3680 bits<5> op11_7, bit op4,
3681 string OpcodeStr, string Dt, SDPatternOperator IntOp> {
3682 // 64-bit vector types.
3683 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3684 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
3685 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3686 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
3687 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3688 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
3690 // 128-bit vector types.
3691 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3692 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
3693 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3694 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
3695 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3696 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
3700 // Neon 2-register vector shift by immediate,
3701 // with f of either N2RegVShLFrm or N2RegVShRFrm
3702 // element sizes of 8, 16, 32 and 64 bits:
3703 multiclass N2VShL_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3704 InstrItinClass itin, string OpcodeStr, string Dt,
3706 // 64-bit vector types.
3707 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3708 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
3709 let Inst{21-19} = 0b001; // imm6 = 001xxx
3711 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3712 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
3713 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3715 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3716 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
3717 let Inst{21} = 0b1; // imm6 = 1xxxxx
3719 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, N2RegVShLFrm, itin, i32imm,
3720 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
3723 // 128-bit vector types.
3724 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3725 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
3726 let Inst{21-19} = 0b001; // imm6 = 001xxx
3728 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3729 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
3730 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3732 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3733 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
3734 let Inst{21} = 0b1; // imm6 = 1xxxxx
3736 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, N2RegVShLFrm, itin, i32imm,
3737 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
3740 multiclass N2VShR_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3741 InstrItinClass itin, string OpcodeStr, string Dt,
3742 string baseOpc, SDNode OpNode> {
3743 // 64-bit vector types.
3744 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm8,
3745 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
3746 let Inst{21-19} = 0b001; // imm6 = 001xxx
3748 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm16,
3749 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
3750 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3752 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm32,
3753 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
3754 let Inst{21} = 0b1; // imm6 = 1xxxxx
3756 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, N2RegVShRFrm, itin, shr_imm64,
3757 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
3760 // 128-bit vector types.
3761 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm8,
3762 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
3763 let Inst{21-19} = 0b001; // imm6 = 001xxx
3765 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm16,
3766 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
3767 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3769 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm32,
3770 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
3771 let Inst{21} = 0b1; // imm6 = 1xxxxx
3773 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, N2RegVShRFrm, itin, shr_imm64,
3774 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
3778 // Neon Shift-Accumulate vector operations,
3779 // element sizes of 8, 16, 32 and 64 bits:
3780 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3781 string OpcodeStr, string Dt, SDNode ShOp> {
3782 // 64-bit vector types.
3783 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm8,
3784 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
3785 let Inst{21-19} = 0b001; // imm6 = 001xxx
3787 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm16,
3788 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
3789 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3791 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm32,
3792 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
3793 let Inst{21} = 0b1; // imm6 = 1xxxxx
3795 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4, shr_imm64,
3796 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
3799 // 128-bit vector types.
3800 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm8,
3801 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
3802 let Inst{21-19} = 0b001; // imm6 = 001xxx
3804 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm16,
3805 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
3806 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3808 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm32,
3809 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
3810 let Inst{21} = 0b1; // imm6 = 1xxxxx
3812 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4, shr_imm64,
3813 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
3817 // Neon Shift-Insert vector operations,
3818 // with f of either N2RegVShLFrm or N2RegVShRFrm
3819 // element sizes of 8, 16, 32 and 64 bits:
3820 multiclass N2VShInsL_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3822 // 64-bit vector types.
3823 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3824 N2RegVShLFrm, OpcodeStr, "8", v8i8, NEONvsli> {
3825 let Inst{21-19} = 0b001; // imm6 = 001xxx
3827 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3828 N2RegVShLFrm, OpcodeStr, "16", v4i16, NEONvsli> {
3829 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3831 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3832 N2RegVShLFrm, OpcodeStr, "32", v2i32, NEONvsli> {
3833 let Inst{21} = 0b1; // imm6 = 1xxxxx
3835 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4, i32imm,
3836 N2RegVShLFrm, OpcodeStr, "64", v1i64, NEONvsli>;
3839 // 128-bit vector types.
3840 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3841 N2RegVShLFrm, OpcodeStr, "8", v16i8, NEONvsli> {
3842 let Inst{21-19} = 0b001; // imm6 = 001xxx
3844 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3845 N2RegVShLFrm, OpcodeStr, "16", v8i16, NEONvsli> {
3846 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3848 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3849 N2RegVShLFrm, OpcodeStr, "32", v4i32, NEONvsli> {
3850 let Inst{21} = 0b1; // imm6 = 1xxxxx
3852 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4, i32imm,
3853 N2RegVShLFrm, OpcodeStr, "64", v2i64, NEONvsli>;
3856 multiclass N2VShInsR_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3858 // 64-bit vector types.
3859 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm8,
3860 N2RegVShRFrm, OpcodeStr, "8", v8i8, NEONvsri> {
3861 let Inst{21-19} = 0b001; // imm6 = 001xxx
3863 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm16,
3864 N2RegVShRFrm, OpcodeStr, "16", v4i16, NEONvsri> {
3865 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3867 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm32,
3868 N2RegVShRFrm, OpcodeStr, "32", v2i32, NEONvsri> {
3869 let Inst{21} = 0b1; // imm6 = 1xxxxx
3871 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4, shr_imm64,
3872 N2RegVShRFrm, OpcodeStr, "64", v1i64, NEONvsri>;
3875 // 128-bit vector types.
3876 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm8,
3877 N2RegVShRFrm, OpcodeStr, "8", v16i8, NEONvsri> {
3878 let Inst{21-19} = 0b001; // imm6 = 001xxx
3880 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm16,
3881 N2RegVShRFrm, OpcodeStr, "16", v8i16, NEONvsri> {
3882 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3884 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm32,
3885 N2RegVShRFrm, OpcodeStr, "32", v4i32, NEONvsri> {
3886 let Inst{21} = 0b1; // imm6 = 1xxxxx
3888 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4, shr_imm64,
3889 N2RegVShRFrm, OpcodeStr, "64", v2i64, NEONvsri>;
3893 // Neon Shift Long operations,
3894 // element sizes of 8, 16, 32 bits:
3895 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
3896 bit op4, string OpcodeStr, string Dt, SDNode OpNode> {
3897 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3898 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, imm1_7, OpNode> {
3899 let Inst{21-19} = 0b001; // imm6 = 001xxx
3901 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3902 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, imm1_15, OpNode> {
3903 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3905 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3906 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, imm1_31, OpNode> {
3907 let Inst{21} = 0b1; // imm6 = 1xxxxx
3911 // Neon Shift Narrow operations,
3912 // element sizes of 16, 32, 64 bits:
3913 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
3914 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
3916 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3917 OpcodeStr, !strconcat(Dt, "16"),
3918 v8i8, v8i16, shr_imm8, OpNode> {
3919 let Inst{21-19} = 0b001; // imm6 = 001xxx
3921 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3922 OpcodeStr, !strconcat(Dt, "32"),
3923 v4i16, v4i32, shr_imm16, OpNode> {
3924 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3926 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3927 OpcodeStr, !strconcat(Dt, "64"),
3928 v2i32, v2i64, shr_imm32, OpNode> {
3929 let Inst{21} = 0b1; // imm6 = 1xxxxx
3933 //===----------------------------------------------------------------------===//
3934 // Instruction Definitions.
3935 //===----------------------------------------------------------------------===//
3937 // Vector Add Operations.
3939 // VADD : Vector Add (integer and floating-point)
3940 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
3942 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
3943 v2f32, v2f32, fadd, 1>;
3944 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
3945 v4f32, v4f32, fadd, 1>;
3946 // VADDL : Vector Add Long (Q = D + D)
3947 defm VADDLs : N3VLExt_QHS<0,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3948 "vaddl", "s", add, sext, 1>;
3949 defm VADDLu : N3VLExt_QHS<1,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3950 "vaddl", "u", add, zext, 1>;
3951 // VADDW : Vector Add Wide (Q = Q + D)
3952 defm VADDWs : N3VW_QHS<0,1,0b0001,0, "vaddw", "s", add, sext, 0>;
3953 defm VADDWu : N3VW_QHS<1,1,0b0001,0, "vaddw", "u", add, zext, 0>;
3954 // VHADD : Vector Halving Add
3955 defm VHADDs : N3VInt_QHS<0, 0, 0b0000, 0, N3RegFrm,
3956 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3957 "vhadd", "s", int_arm_neon_vhadds, 1>;
3958 defm VHADDu : N3VInt_QHS<1, 0, 0b0000, 0, N3RegFrm,
3959 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3960 "vhadd", "u", int_arm_neon_vhaddu, 1>;
3961 // VRHADD : Vector Rounding Halving Add
3962 defm VRHADDs : N3VInt_QHS<0, 0, 0b0001, 0, N3RegFrm,
3963 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3964 "vrhadd", "s", int_arm_neon_vrhadds, 1>;
3965 defm VRHADDu : N3VInt_QHS<1, 0, 0b0001, 0, N3RegFrm,
3966 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3967 "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
3968 // VQADD : Vector Saturating Add
3969 defm VQADDs : N3VInt_QHSD<0, 0, 0b0000, 1, N3RegFrm,
3970 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3971 "vqadd", "s", int_arm_neon_vqadds, 1>;
3972 defm VQADDu : N3VInt_QHSD<1, 0, 0b0000, 1, N3RegFrm,
3973 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3974 "vqadd", "u", int_arm_neon_vqaddu, 1>;
3975 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
3976 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i", null_frag, 1>;
3977 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
3978 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
3979 int_arm_neon_vraddhn, 1>;
3981 def : Pat<(v8i8 (trunc (NEONvshru (add (v8i16 QPR:$Vn), QPR:$Vm), 8))),
3982 (VADDHNv8i8 QPR:$Vn, QPR:$Vm)>;
3983 def : Pat<(v4i16 (trunc (NEONvshru (add (v4i32 QPR:$Vn), QPR:$Vm), 16))),
3984 (VADDHNv4i16 QPR:$Vn, QPR:$Vm)>;
3985 def : Pat<(v2i32 (trunc (NEONvshru (add (v2i64 QPR:$Vn), QPR:$Vm), 32))),
3986 (VADDHNv2i32 QPR:$Vn, QPR:$Vm)>;
3988 // Vector Multiply Operations.
3990 // VMUL : Vector Multiply (integer, polynomial and floating-point)
3991 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
3992 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
3993 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16D, "vmul",
3994 "p8", v8i8, v8i8, int_arm_neon_vmulp, 1>;
3995 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16Q, "vmul",
3996 "p8", v16i8, v16i8, int_arm_neon_vmulp, 1>;
3997 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VFMULD, "vmul", "f32",
3998 v2f32, v2f32, fmul, 1>;
3999 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VFMULQ, "vmul", "f32",
4000 v4f32, v4f32, fmul, 1>;
4001 defm VMULsl : N3VSL_HS<0b1000, "vmul", mul>;
4002 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
4003 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
4006 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
4007 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
4008 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
4009 (v4i16 (EXTRACT_SUBREG QPR:$src2,
4010 (DSubReg_i16_reg imm:$lane))),
4011 (SubReg_i16_lane imm:$lane)))>;
4012 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
4013 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
4014 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
4015 (v2i32 (EXTRACT_SUBREG QPR:$src2,
4016 (DSubReg_i32_reg imm:$lane))),
4017 (SubReg_i32_lane imm:$lane)))>;
4018 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
4019 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
4020 (v4f32 (VMULslfq (v4f32 QPR:$src1),
4021 (v2f32 (EXTRACT_SUBREG QPR:$src2,
4022 (DSubReg_i32_reg imm:$lane))),
4023 (SubReg_i32_lane imm:$lane)))>;
4025 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
4026 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, N3RegFrm, IIC_VMULi16D, IIC_VMULi32D,
4027 IIC_VMULi16Q, IIC_VMULi32Q,
4028 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
4029 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
4030 IIC_VMULi16Q, IIC_VMULi32Q,
4031 "vqdmulh", "s", int_arm_neon_vqdmulh>;
4032 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
4033 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
4035 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
4036 (v4i16 (EXTRACT_SUBREG QPR:$src2,
4037 (DSubReg_i16_reg imm:$lane))),
4038 (SubReg_i16_lane imm:$lane)))>;
4039 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
4040 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
4042 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
4043 (v2i32 (EXTRACT_SUBREG QPR:$src2,
4044 (DSubReg_i32_reg imm:$lane))),
4045 (SubReg_i32_lane imm:$lane)))>;
4047 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
4048 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, N3RegFrm,
4049 IIC_VMULi16D,IIC_VMULi32D,IIC_VMULi16Q,IIC_VMULi32Q,
4050 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
4051 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
4052 IIC_VMULi16Q, IIC_VMULi32Q,
4053 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
4054 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
4055 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
4057 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
4058 (v4i16 (EXTRACT_SUBREG QPR:$src2,
4059 (DSubReg_i16_reg imm:$lane))),
4060 (SubReg_i16_lane imm:$lane)))>;
4061 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
4062 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
4064 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
4065 (v2i32 (EXTRACT_SUBREG QPR:$src2,
4066 (DSubReg_i32_reg imm:$lane))),
4067 (SubReg_i32_lane imm:$lane)))>;
4069 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
4070 defm VMULLs : N3VL_QHS<0,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
4071 "vmull", "s", NEONvmulls, 1>;
4072 defm VMULLu : N3VL_QHS<1,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
4073 "vmull", "u", NEONvmullu, 1>;
4074 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
4075 v8i16, v8i8, int_arm_neon_vmullp, 1>;
4076 defm VMULLsls : N3VLSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s", NEONvmulls>;
4077 defm VMULLslu : N3VLSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u", NEONvmullu>;
4079 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
4080 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, IIC_VMULi32D,
4081 "vqdmull", "s", int_arm_neon_vqdmull, 1>;
4082 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D,
4083 "vqdmull", "s", int_arm_neon_vqdmull>;
4085 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
4087 // VMLA : Vector Multiply Accumulate (integer and floating-point)
4088 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
4089 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
4090 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
4091 v2f32, fmul_su, fadd_mlx>,
4092 Requires<[HasNEON, UseFPVMLx, DontUseFusedMAC]>;
4093 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
4094 v4f32, fmul_su, fadd_mlx>,
4095 Requires<[HasNEON, UseFPVMLx, DontUseFusedMAC]>;
4096 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
4097 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
4098 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
4099 v2f32, fmul_su, fadd_mlx>,
4100 Requires<[HasNEON, UseFPVMLx]>;
4101 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
4102 v4f32, v2f32, fmul_su, fadd_mlx>,
4103 Requires<[HasNEON, UseFPVMLx]>;
4105 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
4106 (mul (v8i16 QPR:$src2),
4107 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
4108 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
4109 (v4i16 (EXTRACT_SUBREG QPR:$src3,
4110 (DSubReg_i16_reg imm:$lane))),
4111 (SubReg_i16_lane imm:$lane)))>;
4113 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
4114 (mul (v4i32 QPR:$src2),
4115 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
4116 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
4117 (v2i32 (EXTRACT_SUBREG QPR:$src3,
4118 (DSubReg_i32_reg imm:$lane))),
4119 (SubReg_i32_lane imm:$lane)))>;
4121 def : Pat<(v4f32 (fadd_mlx (v4f32 QPR:$src1),
4122 (fmul_su (v4f32 QPR:$src2),
4123 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
4124 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
4126 (v2f32 (EXTRACT_SUBREG QPR:$src3,
4127 (DSubReg_i32_reg imm:$lane))),
4128 (SubReg_i32_lane imm:$lane)))>,
4129 Requires<[HasNEON, UseFPVMLx]>;
4131 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
4132 defm VMLALs : N3VLMulOp_QHS<0,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
4133 "vmlal", "s", NEONvmulls, add>;
4134 defm VMLALu : N3VLMulOp_QHS<1,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
4135 "vmlal", "u", NEONvmullu, add>;
4137 defm VMLALsls : N3VLMulOpSL_HS<0, 0b0010, "vmlal", "s", NEONvmulls, add>;
4138 defm VMLALslu : N3VLMulOpSL_HS<1, 0b0010, "vmlal", "u", NEONvmullu, add>;
4140 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
4141 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
4142 "vqdmlal", "s", null_frag>;
4143 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", null_frag>;
4145 def : Pat<(v4i32 (int_arm_neon_vqadds (v4i32 QPR:$src1),
4146 (v4i32 (int_arm_neon_vqdmull (v4i16 DPR:$Vn),
4147 (v4i16 DPR:$Vm))))),
4148 (VQDMLALv4i32 QPR:$src1, DPR:$Vn, DPR:$Vm)>;
4149 def : Pat<(v2i64 (int_arm_neon_vqadds (v2i64 QPR:$src1),
4150 (v2i64 (int_arm_neon_vqdmull (v2i32 DPR:$Vn),
4151 (v2i32 DPR:$Vm))))),
4152 (VQDMLALv2i64 QPR:$src1, DPR:$Vn, DPR:$Vm)>;
4153 def : Pat<(v4i32 (int_arm_neon_vqadds (v4i32 QPR:$src1),
4154 (v4i32 (int_arm_neon_vqdmull (v4i16 DPR:$Vn),
4155 (v4i16 (NEONvduplane (v4i16 DPR_8:$Vm),
4157 (VQDMLALslv4i16 QPR:$src1, DPR:$Vn, DPR_8:$Vm, imm:$lane)>;
4158 def : Pat<(v2i64 (int_arm_neon_vqadds (v2i64 QPR:$src1),
4159 (v2i64 (int_arm_neon_vqdmull (v2i32 DPR:$Vn),
4160 (v2i32 (NEONvduplane (v2i32 DPR_VFP2:$Vm),
4162 (VQDMLALslv2i32 QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, imm:$lane)>;
4164 // VMLS : Vector Multiply Subtract (integer and floating-point)
4165 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
4166 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
4167 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
4168 v2f32, fmul_su, fsub_mlx>,
4169 Requires<[HasNEON, UseFPVMLx, DontUseFusedMAC]>;
4170 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
4171 v4f32, fmul_su, fsub_mlx>,
4172 Requires<[HasNEON, UseFPVMLx, DontUseFusedMAC]>;
4173 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
4174 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
4175 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
4176 v2f32, fmul_su, fsub_mlx>,
4177 Requires<[HasNEON, UseFPVMLx]>;
4178 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
4179 v4f32, v2f32, fmul_su, fsub_mlx>,
4180 Requires<[HasNEON, UseFPVMLx]>;
4182 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
4183 (mul (v8i16 QPR:$src2),
4184 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
4185 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
4186 (v4i16 (EXTRACT_SUBREG QPR:$src3,
4187 (DSubReg_i16_reg imm:$lane))),
4188 (SubReg_i16_lane imm:$lane)))>;
4190 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
4191 (mul (v4i32 QPR:$src2),
4192 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
4193 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
4194 (v2i32 (EXTRACT_SUBREG QPR:$src3,
4195 (DSubReg_i32_reg imm:$lane))),
4196 (SubReg_i32_lane imm:$lane)))>;
4198 def : Pat<(v4f32 (fsub_mlx (v4f32 QPR:$src1),
4199 (fmul_su (v4f32 QPR:$src2),
4200 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
4201 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
4202 (v2f32 (EXTRACT_SUBREG QPR:$src3,
4203 (DSubReg_i32_reg imm:$lane))),
4204 (SubReg_i32_lane imm:$lane)))>,
4205 Requires<[HasNEON, UseFPVMLx]>;
4207 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
4208 defm VMLSLs : N3VLMulOp_QHS<0,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
4209 "vmlsl", "s", NEONvmulls, sub>;
4210 defm VMLSLu : N3VLMulOp_QHS<1,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
4211 "vmlsl", "u", NEONvmullu, sub>;
4213 defm VMLSLsls : N3VLMulOpSL_HS<0, 0b0110, "vmlsl", "s", NEONvmulls, sub>;
4214 defm VMLSLslu : N3VLMulOpSL_HS<1, 0b0110, "vmlsl", "u", NEONvmullu, sub>;
4216 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
4217 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, IIC_VMACi16D, IIC_VMACi32D,
4218 "vqdmlsl", "s", null_frag>;
4219 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl", "s", null_frag>;
4221 def : Pat<(v4i32 (int_arm_neon_vqsubs (v4i32 QPR:$src1),
4222 (v4i32 (int_arm_neon_vqdmull (v4i16 DPR:$Vn),
4223 (v4i16 DPR:$Vm))))),
4224 (VQDMLSLv4i32 QPR:$src1, DPR:$Vn, DPR:$Vm)>;
4225 def : Pat<(v2i64 (int_arm_neon_vqsubs (v2i64 QPR:$src1),
4226 (v2i64 (int_arm_neon_vqdmull (v2i32 DPR:$Vn),
4227 (v2i32 DPR:$Vm))))),
4228 (VQDMLSLv2i64 QPR:$src1, DPR:$Vn, DPR:$Vm)>;
4229 def : Pat<(v4i32 (int_arm_neon_vqsubs (v4i32 QPR:$src1),
4230 (v4i32 (int_arm_neon_vqdmull (v4i16 DPR:$Vn),
4231 (v4i16 (NEONvduplane (v4i16 DPR_8:$Vm),
4233 (VQDMLSLslv4i16 QPR:$src1, DPR:$Vn, DPR_8:$Vm, imm:$lane)>;
4234 def : Pat<(v2i64 (int_arm_neon_vqsubs (v2i64 QPR:$src1),
4235 (v2i64 (int_arm_neon_vqdmull (v2i32 DPR:$Vn),
4236 (v2i32 (NEONvduplane (v2i32 DPR_VFP2:$Vm),
4238 (VQDMLSLslv2i32 QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, imm:$lane)>;
4240 // Fused Vector Multiply-Accumulate and Fused Multiply-Subtract Operations.
4241 def VFMAfd : N3VDMulOp<0, 0, 0b00, 0b1100, 1, IIC_VFMACD, "vfma", "f32",
4242 v2f32, fmul_su, fadd_mlx>,
4243 Requires<[HasVFP4,UseFusedMAC]>;
4245 def VFMAfq : N3VQMulOp<0, 0, 0b00, 0b1100, 1, IIC_VFMACQ, "vfma", "f32",
4246 v4f32, fmul_su, fadd_mlx>,
4247 Requires<[HasVFP4,UseFusedMAC]>;
4249 // Fused Vector Multiply Subtract (floating-point)
4250 def VFMSfd : N3VDMulOp<0, 0, 0b10, 0b1100, 1, IIC_VFMACD, "vfms", "f32",
4251 v2f32, fmul_su, fsub_mlx>,
4252 Requires<[HasVFP4,UseFusedMAC]>;
4253 def VFMSfq : N3VQMulOp<0, 0, 0b10, 0b1100, 1, IIC_VFMACQ, "vfms", "f32",
4254 v4f32, fmul_su, fsub_mlx>,
4255 Requires<[HasVFP4,UseFusedMAC]>;
4257 // Match @llvm.fma.* intrinsics
4258 def : Pat<(v2f32 (fma DPR:$Vn, DPR:$Vm, DPR:$src1)),
4259 (VFMAfd DPR:$src1, DPR:$Vn, DPR:$Vm)>,
4260 Requires<[HasVFP4]>;
4261 def : Pat<(v4f32 (fma QPR:$Vn, QPR:$Vm, QPR:$src1)),
4262 (VFMAfq QPR:$src1, QPR:$Vn, QPR:$Vm)>,
4263 Requires<[HasVFP4]>;
4264 def : Pat<(v2f32 (fma (fneg DPR:$Vn), DPR:$Vm, DPR:$src1)),
4265 (VFMSfd DPR:$src1, DPR:$Vn, DPR:$Vm)>,
4266 Requires<[HasVFP4]>;
4267 def : Pat<(v4f32 (fma (fneg QPR:$Vn), QPR:$Vm, QPR:$src1)),
4268 (VFMSfq QPR:$src1, QPR:$Vn, QPR:$Vm)>,
4269 Requires<[HasVFP4]>;
4271 // Vector Subtract Operations.
4273 // VSUB : Vector Subtract (integer and floating-point)
4274 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
4275 "vsub", "i", sub, 0>;
4276 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
4277 v2f32, v2f32, fsub, 0>;
4278 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
4279 v4f32, v4f32, fsub, 0>;
4280 // VSUBL : Vector Subtract Long (Q = D - D)
4281 defm VSUBLs : N3VLExt_QHS<0,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
4282 "vsubl", "s", sub, sext, 0>;
4283 defm VSUBLu : N3VLExt_QHS<1,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
4284 "vsubl", "u", sub, zext, 0>;
4285 // VSUBW : Vector Subtract Wide (Q = Q - D)
4286 defm VSUBWs : N3VW_QHS<0,1,0b0011,0, "vsubw", "s", sub, sext, 0>;
4287 defm VSUBWu : N3VW_QHS<1,1,0b0011,0, "vsubw", "u", sub, zext, 0>;
4288 // VHSUB : Vector Halving Subtract
4289 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, N3RegFrm,
4290 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4291 "vhsub", "s", int_arm_neon_vhsubs, 0>;
4292 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, N3RegFrm,
4293 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4294 "vhsub", "u", int_arm_neon_vhsubu, 0>;
4295 // VQSUB : Vector Saturing Subtract
4296 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, N3RegFrm,
4297 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4298 "vqsub", "s", int_arm_neon_vqsubs, 0>;
4299 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, N3RegFrm,
4300 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4301 "vqsub", "u", int_arm_neon_vqsubu, 0>;
4302 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
4303 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i", null_frag, 0>;
4304 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
4305 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
4306 int_arm_neon_vrsubhn, 0>;
4308 def : Pat<(v8i8 (trunc (NEONvshru (sub (v8i16 QPR:$Vn), QPR:$Vm), 8))),
4309 (VSUBHNv8i8 QPR:$Vn, QPR:$Vm)>;
4310 def : Pat<(v4i16 (trunc (NEONvshru (sub (v4i32 QPR:$Vn), QPR:$Vm), 16))),
4311 (VSUBHNv4i16 QPR:$Vn, QPR:$Vm)>;
4312 def : Pat<(v2i32 (trunc (NEONvshru (sub (v2i64 QPR:$Vn), QPR:$Vm), 32))),
4313 (VSUBHNv2i32 QPR:$Vn, QPR:$Vm)>;
4315 // Vector Comparisons.
4317 // VCEQ : Vector Compare Equal
4318 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
4319 IIC_VSUBi4Q, "vceq", "i", NEONvceq, 1>;
4320 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
4322 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
4325 let TwoOperandAliasConstraint = "$Vm = $Vd" in
4326 defm VCEQz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
4327 "$Vd, $Vm, #0", NEONvceqz>;
4329 // VCGE : Vector Compare Greater Than or Equal
4330 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
4331 IIC_VSUBi4Q, "vcge", "s", NEONvcge, 0>;
4332 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
4333 IIC_VSUBi4Q, "vcge", "u", NEONvcgeu, 0>;
4334 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32", v2i32, v2f32,
4336 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
4339 let TwoOperandAliasConstraint = "$Vm = $Vd" in {
4340 defm VCGEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
4341 "$Vd, $Vm, #0", NEONvcgez>;
4342 defm VCLEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
4343 "$Vd, $Vm, #0", NEONvclez>;
4346 // VCGT : Vector Compare Greater Than
4347 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
4348 IIC_VSUBi4Q, "vcgt", "s", NEONvcgt, 0>;
4349 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
4350 IIC_VSUBi4Q, "vcgt", "u", NEONvcgtu, 0>;
4351 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
4353 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
4356 let TwoOperandAliasConstraint = "$Vm = $Vd" in {
4357 defm VCGTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
4358 "$Vd, $Vm, #0", NEONvcgtz>;
4359 defm VCLTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
4360 "$Vd, $Vm, #0", NEONvcltz>;
4363 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
4364 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacge",
4365 "f32", v2i32, v2f32, int_arm_neon_vacged, 0>;
4366 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacge",
4367 "f32", v4i32, v4f32, int_arm_neon_vacgeq, 0>;
4368 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
4369 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacgt",
4370 "f32", v2i32, v2f32, int_arm_neon_vacgtd, 0>;
4371 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacgt",
4372 "f32", v4i32, v4f32, int_arm_neon_vacgtq, 0>;
4373 // VTST : Vector Test Bits
4374 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
4375 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
4377 def: NEONInstAlias<"vaclt${p}.f32 $Vd, $Vn, $Vm",
4378 (VACGTd DPR:$Vd, DPR:$Vm, DPR:$Vn, pred:$p)>;
4379 def: NEONInstAlias<"vaclt${p}.f32 $Vd, $Vn, $Vm",
4380 (VACGTq QPR:$Vd, QPR:$Vm, QPR:$Vn, pred:$p)>;
4381 def: NEONInstAlias<"vacle${p}.f32 $Vd, $Vn, $Vm",
4382 (VACGEd DPR:$Vd, DPR:$Vm, DPR:$Vn, pred:$p)>;
4383 def: NEONInstAlias<"vacle${p}.f32 $Vd, $Vn, $Vm",
4384 (VACGEq QPR:$Vd, QPR:$Vm, QPR:$Vn, pred:$p)>;
4386 def: NEONInstAlias<"vaclt${p}.f32 $Vd, $Vm",
4387 (VACGTd DPR:$Vd, DPR:$Vm, DPR:$Vd, pred:$p)>;
4388 def: NEONInstAlias<"vaclt${p}.f32 $Vd, $Vm",
4389 (VACGTq QPR:$Vd, QPR:$Vm, QPR:$Vd, pred:$p)>;
4390 def: NEONInstAlias<"vacle${p}.f32 $Vd, $Vm",
4391 (VACGEd DPR:$Vd, DPR:$Vm, DPR:$Vd, pred:$p)>;
4392 def: NEONInstAlias<"vacle${p}.f32 $Vd, $Vm",
4393 (VACGEq QPR:$Vd, QPR:$Vm, QPR:$Vd, pred:$p)>;
4395 // Vector Bitwise Operations.
4397 def vnotd : PatFrag<(ops node:$in),
4398 (xor node:$in, (bitconvert (v8i8 NEONimmAllOnesV)))>;
4399 def vnotq : PatFrag<(ops node:$in),
4400 (xor node:$in, (bitconvert (v16i8 NEONimmAllOnesV)))>;
4403 // VAND : Vector Bitwise AND
4404 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
4405 v2i32, v2i32, and, 1>;
4406 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
4407 v4i32, v4i32, and, 1>;
4409 // VEOR : Vector Bitwise Exclusive OR
4410 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
4411 v2i32, v2i32, xor, 1>;
4412 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
4413 v4i32, v4i32, xor, 1>;
4415 // VORR : Vector Bitwise OR
4416 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
4417 v2i32, v2i32, or, 1>;
4418 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
4419 v4i32, v4i32, or, 1>;
4421 def VORRiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 0, 1,
4422 (outs DPR:$Vd), (ins nImmSplatI16:$SIMM, DPR:$src),
4424 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
4426 (v4i16 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
4427 let Inst{9} = SIMM{9};
4430 def VORRiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 0, 1,
4431 (outs DPR:$Vd), (ins nImmSplatI32:$SIMM, DPR:$src),
4433 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
4435 (v2i32 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
4436 let Inst{10-9} = SIMM{10-9};
4439 def VORRiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 0, 1,
4440 (outs QPR:$Vd), (ins nImmSplatI16:$SIMM, QPR:$src),
4442 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
4444 (v8i16 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
4445 let Inst{9} = SIMM{9};
4448 def VORRiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 0, 1,
4449 (outs QPR:$Vd), (ins nImmSplatI32:$SIMM, QPR:$src),
4451 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
4453 (v4i32 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
4454 let Inst{10-9} = SIMM{10-9};
4458 // VBIC : Vector Bitwise Bit Clear (AND NOT)
4459 let TwoOperandAliasConstraint = "$Vn = $Vd" in {
4460 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
4461 (ins DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VBINiD,
4462 "vbic", "$Vd, $Vn, $Vm", "",
4463 [(set DPR:$Vd, (v2i32 (and DPR:$Vn,
4464 (vnotd DPR:$Vm))))]>;
4465 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
4466 (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINiQ,
4467 "vbic", "$Vd, $Vn, $Vm", "",
4468 [(set QPR:$Vd, (v4i32 (and QPR:$Vn,
4469 (vnotq QPR:$Vm))))]>;
4472 def VBICiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 1, 1,
4473 (outs DPR:$Vd), (ins nImmSplatI16:$SIMM, DPR:$src),
4475 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
4477 (v4i16 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
4478 let Inst{9} = SIMM{9};
4481 def VBICiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 1, 1,
4482 (outs DPR:$Vd), (ins nImmSplatI32:$SIMM, DPR:$src),
4484 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
4486 (v2i32 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
4487 let Inst{10-9} = SIMM{10-9};
4490 def VBICiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 1, 1,
4491 (outs QPR:$Vd), (ins nImmSplatI16:$SIMM, QPR:$src),
4493 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
4495 (v8i16 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
4496 let Inst{9} = SIMM{9};
4499 def VBICiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 1, 1,
4500 (outs QPR:$Vd), (ins nImmSplatI32:$SIMM, QPR:$src),
4502 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
4504 (v4i32 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
4505 let Inst{10-9} = SIMM{10-9};
4508 // VORN : Vector Bitwise OR NOT
4509 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$Vd),
4510 (ins DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VBINiD,
4511 "vorn", "$Vd, $Vn, $Vm", "",
4512 [(set DPR:$Vd, (v2i32 (or DPR:$Vn,
4513 (vnotd DPR:$Vm))))]>;
4514 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$Vd),
4515 (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINiQ,
4516 "vorn", "$Vd, $Vn, $Vm", "",
4517 [(set QPR:$Vd, (v4i32 (or QPR:$Vn,
4518 (vnotq QPR:$Vm))))]>;
4520 // VMVN : Vector Bitwise NOT (Immediate)
4522 let isReMaterializable = 1 in {
4524 def VMVNv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 1, 1, (outs DPR:$Vd),
4525 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4526 "vmvn", "i16", "$Vd, $SIMM", "",
4527 [(set DPR:$Vd, (v4i16 (NEONvmvnImm timm:$SIMM)))]> {
4528 let Inst{9} = SIMM{9};
4531 def VMVNv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 1, 1, (outs QPR:$Vd),
4532 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4533 "vmvn", "i16", "$Vd, $SIMM", "",
4534 [(set QPR:$Vd, (v8i16 (NEONvmvnImm timm:$SIMM)))]> {
4535 let Inst{9} = SIMM{9};
4538 def VMVNv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 1, 1, (outs DPR:$Vd),
4539 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4540 "vmvn", "i32", "$Vd, $SIMM", "",
4541 [(set DPR:$Vd, (v2i32 (NEONvmvnImm timm:$SIMM)))]> {
4542 let Inst{11-8} = SIMM{11-8};
4545 def VMVNv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 1, 1, (outs QPR:$Vd),
4546 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4547 "vmvn", "i32", "$Vd, $SIMM", "",
4548 [(set QPR:$Vd, (v4i32 (NEONvmvnImm timm:$SIMM)))]> {
4549 let Inst{11-8} = SIMM{11-8};
4553 // VMVN : Vector Bitwise NOT
4554 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
4555 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VSUBiD,
4556 "vmvn", "$Vd, $Vm", "",
4557 [(set DPR:$Vd, (v2i32 (vnotd DPR:$Vm)))]>;
4558 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
4559 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VSUBiD,
4560 "vmvn", "$Vd, $Vm", "",
4561 [(set QPR:$Vd, (v4i32 (vnotq QPR:$Vm)))]>;
4562 def : Pat<(v2i32 (vnotd DPR:$src)), (VMVNd DPR:$src)>;
4563 def : Pat<(v4i32 (vnotq QPR:$src)), (VMVNq QPR:$src)>;
4565 // VBSL : Vector Bitwise Select
4566 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
4567 (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4568 N3RegFrm, IIC_VCNTiD,
4569 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4571 (v2i32 (NEONvbsl DPR:$src1, DPR:$Vn, DPR:$Vm)))]>;
4572 def : Pat<(v8i8 (int_arm_neon_vbsl (v8i8 DPR:$src1),
4573 (v8i8 DPR:$Vn), (v8i8 DPR:$Vm))),
4574 (VBSLd DPR:$src1, DPR:$Vn, DPR:$Vm)>,
4575 Requires<[HasNEON]>;
4576 def : Pat<(v4i16 (int_arm_neon_vbsl (v4i16 DPR:$src1),
4577 (v4i16 DPR:$Vn), (v4i16 DPR:$Vm))),
4578 (VBSLd DPR:$src1, DPR:$Vn, DPR:$Vm)>,
4579 Requires<[HasNEON]>;
4580 def : Pat<(v2i32 (int_arm_neon_vbsl (v2i32 DPR:$src1),
4581 (v2i32 DPR:$Vn), (v2i32 DPR:$Vm))),
4582 (VBSLd DPR:$src1, DPR:$Vn, DPR:$Vm)>,
4583 Requires<[HasNEON]>;
4584 def : Pat<(v2f32 (int_arm_neon_vbsl (v2f32 DPR:$src1),
4585 (v2f32 DPR:$Vn), (v2f32 DPR:$Vm))),
4586 (VBSLd DPR:$src1, DPR:$Vn, DPR:$Vm)>,
4587 Requires<[HasNEON]>;
4588 def : Pat<(v1i64 (int_arm_neon_vbsl (v1i64 DPR:$src1),
4589 (v1i64 DPR:$Vn), (v1i64 DPR:$Vm))),
4590 (VBSLd DPR:$src1, DPR:$Vn, DPR:$Vm)>,
4591 Requires<[HasNEON]>;
4593 def : Pat<(v2i32 (or (and DPR:$Vn, DPR:$Vd),
4594 (and DPR:$Vm, (vnotd DPR:$Vd)))),
4595 (VBSLd DPR:$Vd, DPR:$Vn, DPR:$Vm)>,
4596 Requires<[HasNEON]>;
4598 def : Pat<(v1i64 (or (and DPR:$Vn, DPR:$Vd),
4599 (and DPR:$Vm, (vnotd DPR:$Vd)))),
4600 (VBSLd DPR:$Vd, DPR:$Vn, DPR:$Vm)>,
4601 Requires<[HasNEON]>;
4603 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
4604 (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4605 N3RegFrm, IIC_VCNTiQ,
4606 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4608 (v4i32 (NEONvbsl QPR:$src1, QPR:$Vn, QPR:$Vm)))]>;
4610 def : Pat<(v16i8 (int_arm_neon_vbsl (v16i8 QPR:$src1),
4611 (v16i8 QPR:$Vn), (v16i8 QPR:$Vm))),
4612 (VBSLq QPR:$src1, QPR:$Vn, QPR:$Vm)>,
4613 Requires<[HasNEON]>;
4614 def : Pat<(v8i16 (int_arm_neon_vbsl (v8i16 QPR:$src1),
4615 (v8i16 QPR:$Vn), (v8i16 QPR:$Vm))),
4616 (VBSLq QPR:$src1, QPR:$Vn, QPR:$Vm)>,
4617 Requires<[HasNEON]>;
4618 def : Pat<(v4i32 (int_arm_neon_vbsl (v4i32 QPR:$src1),
4619 (v4i32 QPR:$Vn), (v4i32 QPR:$Vm))),
4620 (VBSLq QPR:$src1, QPR:$Vn, QPR:$Vm)>,
4621 Requires<[HasNEON]>;
4622 def : Pat<(v4f32 (int_arm_neon_vbsl (v4f32 QPR:$src1),
4623 (v4f32 QPR:$Vn), (v4f32 QPR:$Vm))),
4624 (VBSLq QPR:$src1, QPR:$Vn, QPR:$Vm)>,
4625 Requires<[HasNEON]>;
4626 def : Pat<(v2i64 (int_arm_neon_vbsl (v2i64 QPR:$src1),
4627 (v2i64 QPR:$Vn), (v2i64 QPR:$Vm))),
4628 (VBSLq QPR:$src1, QPR:$Vn, QPR:$Vm)>,
4629 Requires<[HasNEON]>;
4631 def : Pat<(v4i32 (or (and QPR:$Vn, QPR:$Vd),
4632 (and QPR:$Vm, (vnotq QPR:$Vd)))),
4633 (VBSLq QPR:$Vd, QPR:$Vn, QPR:$Vm)>,
4634 Requires<[HasNEON]>;
4635 def : Pat<(v2i64 (or (and QPR:$Vn, QPR:$Vd),
4636 (and QPR:$Vm, (vnotq QPR:$Vd)))),
4637 (VBSLq QPR:$Vd, QPR:$Vn, QPR:$Vm)>,
4638 Requires<[HasNEON]>;
4640 // VBIF : Vector Bitwise Insert if False
4641 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
4642 // FIXME: This instruction's encoding MAY NOT BE correct.
4643 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
4644 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4645 N3RegFrm, IIC_VBINiD,
4646 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4648 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
4649 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4650 N3RegFrm, IIC_VBINiQ,
4651 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4654 // VBIT : Vector Bitwise Insert if True
4655 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
4656 // FIXME: This instruction's encoding MAY NOT BE correct.
4657 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
4658 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4659 N3RegFrm, IIC_VBINiD,
4660 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4662 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
4663 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4664 N3RegFrm, IIC_VBINiQ,
4665 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4668 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
4669 // for equivalent operations with different register constraints; it just
4672 // Vector Absolute Differences.
4674 // VABD : Vector Absolute Difference
4675 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, N3RegFrm,
4676 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4677 "vabd", "s", int_arm_neon_vabds, 1>;
4678 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, N3RegFrm,
4679 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4680 "vabd", "u", int_arm_neon_vabdu, 1>;
4681 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBIND,
4682 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 1>;
4683 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBINQ,
4684 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 1>;
4686 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
4687 defm VABDLs : N3VLIntExt_QHS<0,1,0b0111,0, IIC_VSUBi4Q,
4688 "vabdl", "s", int_arm_neon_vabds, zext, 1>;
4689 defm VABDLu : N3VLIntExt_QHS<1,1,0b0111,0, IIC_VSUBi4Q,
4690 "vabdl", "u", int_arm_neon_vabdu, zext, 1>;
4692 // VABA : Vector Absolute Difference and Accumulate
4693 defm VABAs : N3VIntOp_QHS<0,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
4694 "vaba", "s", int_arm_neon_vabds, add>;
4695 defm VABAu : N3VIntOp_QHS<1,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
4696 "vaba", "u", int_arm_neon_vabdu, add>;
4698 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
4699 defm VABALs : N3VLIntExtOp_QHS<0,1,0b0101,0, IIC_VABAD,
4700 "vabal", "s", int_arm_neon_vabds, zext, add>;
4701 defm VABALu : N3VLIntExtOp_QHS<1,1,0b0101,0, IIC_VABAD,
4702 "vabal", "u", int_arm_neon_vabdu, zext, add>;
4704 // Vector Maximum and Minimum.
4706 // VMAX : Vector Maximum
4707 defm VMAXs : N3VInt_QHS<0, 0, 0b0110, 0, N3RegFrm,
4708 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4709 "vmax", "s", int_arm_neon_vmaxs, 1>;
4710 defm VMAXu : N3VInt_QHS<1, 0, 0b0110, 0, N3RegFrm,
4711 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4712 "vmax", "u", int_arm_neon_vmaxu, 1>;
4713 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBIND,
4715 v2f32, v2f32, int_arm_neon_vmaxs, 1>;
4716 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBINQ,
4718 v4f32, v4f32, int_arm_neon_vmaxs, 1>;
4721 let PostEncoderMethod = "NEONThumb2V8PostEncoder", DecoderNamespace = "v8NEON" in {
4722 def VMAXNMND : N3VDIntnp<0b00110, 0b00, 0b1111, 0, 1,
4723 N3RegFrm, NoItinerary, "vmaxnm", "f32",
4724 v2f32, v2f32, int_arm_neon_vmaxnm, 1>,
4725 Requires<[HasV8, HasNEON]>;
4726 def VMAXNMNQ : N3VQIntnp<0b00110, 0b00, 0b1111, 1, 1,
4727 N3RegFrm, NoItinerary, "vmaxnm", "f32",
4728 v4f32, v4f32, int_arm_neon_vmaxnm, 1>,
4729 Requires<[HasV8, HasNEON]>;
4732 // VMIN : Vector Minimum
4733 defm VMINs : N3VInt_QHS<0, 0, 0b0110, 1, N3RegFrm,
4734 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4735 "vmin", "s", int_arm_neon_vmins, 1>;
4736 defm VMINu : N3VInt_QHS<1, 0, 0b0110, 1, N3RegFrm,
4737 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4738 "vmin", "u", int_arm_neon_vminu, 1>;
4739 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBIND,
4741 v2f32, v2f32, int_arm_neon_vmins, 1>;
4742 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBINQ,
4744 v4f32, v4f32, int_arm_neon_vmins, 1>;
4747 let PostEncoderMethod = "NEONThumb2V8PostEncoder", DecoderNamespace = "v8NEON" in {
4748 def VMINNMND : N3VDIntnp<0b00110, 0b10, 0b1111, 0, 1,
4749 N3RegFrm, NoItinerary, "vminnm", "f32",
4750 v2f32, v2f32, int_arm_neon_vminnm, 1>,
4751 Requires<[HasV8, HasNEON]>;
4752 def VMINNMNQ : N3VQIntnp<0b00110, 0b10, 0b1111, 1, 1,
4753 N3RegFrm, NoItinerary, "vminnm", "f32",
4754 v4f32, v4f32, int_arm_neon_vminnm, 1>,
4755 Requires<[HasV8, HasNEON]>;
4758 // Vector Pairwise Operations.
4760 // VPADD : Vector Pairwise Add
4761 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4763 v8i8, v8i8, int_arm_neon_vpadd, 0>;
4764 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4766 v4i16, v4i16, int_arm_neon_vpadd, 0>;
4767 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4769 v2i32, v2i32, int_arm_neon_vpadd, 0>;
4770 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, N3RegFrm,
4771 IIC_VPBIND, "vpadd", "f32",
4772 v2f32, v2f32, int_arm_neon_vpadd, 0>;
4774 // VPADDL : Vector Pairwise Add Long
4775 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
4776 int_arm_neon_vpaddls>;
4777 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
4778 int_arm_neon_vpaddlu>;
4780 // VPADAL : Vector Pairwise Add and Accumulate Long
4781 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
4782 int_arm_neon_vpadals>;
4783 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
4784 int_arm_neon_vpadalu>;
4786 // VPMAX : Vector Pairwise Maximum
4787 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4788 "s8", v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
4789 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4790 "s16", v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
4791 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4792 "s32", v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
4793 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4794 "u8", v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
4795 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4796 "u16", v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
4797 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4798 "u32", v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
4799 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmax",
4800 "f32", v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
4802 // VPMIN : Vector Pairwise Minimum
4803 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4804 "s8", v8i8, v8i8, int_arm_neon_vpmins, 0>;
4805 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4806 "s16", v4i16, v4i16, int_arm_neon_vpmins, 0>;
4807 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4808 "s32", v2i32, v2i32, int_arm_neon_vpmins, 0>;
4809 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4810 "u8", v8i8, v8i8, int_arm_neon_vpminu, 0>;
4811 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4812 "u16", v4i16, v4i16, int_arm_neon_vpminu, 0>;
4813 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4814 "u32", v2i32, v2i32, int_arm_neon_vpminu, 0>;
4815 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmin",
4816 "f32", v2f32, v2f32, int_arm_neon_vpmins, 0>;
4818 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
4820 // VRECPE : Vector Reciprocal Estimate
4821 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
4822 IIC_VUNAD, "vrecpe", "u32",
4823 v2i32, v2i32, int_arm_neon_vrecpe>;
4824 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
4825 IIC_VUNAQ, "vrecpe", "u32",
4826 v4i32, v4i32, int_arm_neon_vrecpe>;
4827 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
4828 IIC_VUNAD, "vrecpe", "f32",
4829 v2f32, v2f32, int_arm_neon_vrecpe>;
4830 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
4831 IIC_VUNAQ, "vrecpe", "f32",
4832 v4f32, v4f32, int_arm_neon_vrecpe>;
4834 // VRECPS : Vector Reciprocal Step
4835 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
4836 IIC_VRECSD, "vrecps", "f32",
4837 v2f32, v2f32, int_arm_neon_vrecps, 1>;
4838 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
4839 IIC_VRECSQ, "vrecps", "f32",
4840 v4f32, v4f32, int_arm_neon_vrecps, 1>;
4842 // VRSQRTE : Vector Reciprocal Square Root Estimate
4843 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
4844 IIC_VUNAD, "vrsqrte", "u32",
4845 v2i32, v2i32, int_arm_neon_vrsqrte>;
4846 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
4847 IIC_VUNAQ, "vrsqrte", "u32",
4848 v4i32, v4i32, int_arm_neon_vrsqrte>;
4849 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
4850 IIC_VUNAD, "vrsqrte", "f32",
4851 v2f32, v2f32, int_arm_neon_vrsqrte>;
4852 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
4853 IIC_VUNAQ, "vrsqrte", "f32",
4854 v4f32, v4f32, int_arm_neon_vrsqrte>;
4856 // VRSQRTS : Vector Reciprocal Square Root Step
4857 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
4858 IIC_VRECSD, "vrsqrts", "f32",
4859 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
4860 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
4861 IIC_VRECSQ, "vrsqrts", "f32",
4862 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
4866 // VSHL : Vector Shift
4867 defm VSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 0, N3RegVShFrm,
4868 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
4869 "vshl", "s", int_arm_neon_vshifts>;
4870 defm VSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 0, N3RegVShFrm,
4871 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
4872 "vshl", "u", int_arm_neon_vshiftu>;
4874 // VSHL : Vector Shift Left (Immediate)
4875 defm VSHLi : N2VShL_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl>;
4877 // VSHR : Vector Shift Right (Immediate)
4878 defm VSHRs : N2VShR_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s", "VSHRs",
4880 defm VSHRu : N2VShR_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u", "VSHRu",
4883 // VSHLL : Vector Shift Left Long
4884 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>;
4885 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>;
4887 // VSHLL : Vector Shift Left Long (with maximum shift count)
4888 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
4889 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
4890 ValueType OpTy, Operand ImmTy, SDNode OpNode>
4891 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
4892 ResTy, OpTy, ImmTy, OpNode> {
4893 let Inst{21-16} = op21_16;
4894 let DecoderMethod = "DecodeVSHLMaxInstruction";
4896 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
4897 v8i16, v8i8, imm8, NEONvshlli>;
4898 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
4899 v4i32, v4i16, imm16, NEONvshlli>;
4900 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
4901 v2i64, v2i32, imm32, NEONvshlli>;
4903 // VSHRN : Vector Shift Right and Narrow
4904 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
4907 // VRSHL : Vector Rounding Shift
4908 defm VRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 0, N3RegVShFrm,
4909 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4910 "vrshl", "s", int_arm_neon_vrshifts>;
4911 defm VRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 0, N3RegVShFrm,
4912 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4913 "vrshl", "u", int_arm_neon_vrshiftu>;
4914 // VRSHR : Vector Rounding Shift Right
4915 defm VRSHRs : N2VShR_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s", "VRSHRs",
4917 defm VRSHRu : N2VShR_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u", "VRSHRu",
4920 // VRSHRN : Vector Rounding Shift Right and Narrow
4921 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
4924 // VQSHL : Vector Saturating Shift
4925 defm VQSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 1, N3RegVShFrm,
4926 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4927 "vqshl", "s", int_arm_neon_vqshifts>;
4928 defm VQSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 1, N3RegVShFrm,
4929 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4930 "vqshl", "u", int_arm_neon_vqshiftu>;
4931 // VQSHL : Vector Saturating Shift Left (Immediate)
4932 defm VQSHLsi : N2VShL_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s",NEONvqshls>;
4933 defm VQSHLui : N2VShL_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u",NEONvqshlu>;
4935 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
4936 defm VQSHLsu : N2VShL_QHSD<1,1,0b0110,1, IIC_VSHLi4D,"vqshlu","s",NEONvqshlsu>;
4938 // VQSHRN : Vector Saturating Shift Right and Narrow
4939 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
4941 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
4944 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
4945 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
4948 // VQRSHL : Vector Saturating Rounding Shift
4949 defm VQRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 1, N3RegVShFrm,
4950 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4951 "vqrshl", "s", int_arm_neon_vqrshifts>;
4952 defm VQRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 1, N3RegVShFrm,
4953 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4954 "vqrshl", "u", int_arm_neon_vqrshiftu>;
4956 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
4957 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
4959 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
4962 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
4963 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
4966 // VSRA : Vector Shift Right and Accumulate
4967 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
4968 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
4969 // VRSRA : Vector Rounding Shift Right and Accumulate
4970 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
4971 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
4973 // VSLI : Vector Shift Left and Insert
4974 defm VSLI : N2VShInsL_QHSD<1, 1, 0b0101, 1, "vsli">;
4976 // VSRI : Vector Shift Right and Insert
4977 defm VSRI : N2VShInsR_QHSD<1, 1, 0b0100, 1, "vsri">;
4979 // Vector Absolute and Saturating Absolute.
4981 // VABS : Vector Absolute Value
4982 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
4983 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
4985 def VABSfd : N2VD<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
4987 v2f32, v2f32, fabs>;
4988 def VABSfq : N2VQ<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
4990 v4f32, v4f32, fabs>;
4992 def : Pat<(xor (v2i32 (bitconvert (v8i8 (NEONvshrs DPR:$src, (i32 7))))),
4993 (v2i32 (bitconvert (v8i8 (add DPR:$src,
4994 (NEONvshrs DPR:$src, (i32 7))))))),
4995 (VABSv8i8 DPR:$src)>;
4996 def : Pat<(xor (v2i32 (bitconvert (v4i16 (NEONvshrs DPR:$src, (i32 15))))),
4997 (v2i32 (bitconvert (v4i16 (add DPR:$src,
4998 (NEONvshrs DPR:$src, (i32 15))))))),
4999 (VABSv4i16 DPR:$src)>;
5000 def : Pat<(xor (v2i32 (NEONvshrs DPR:$src, (i32 31))),
5001 (v2i32 (add DPR:$src, (NEONvshrs DPR:$src, (i32 31))))),
5002 (VABSv2i32 DPR:$src)>;
5003 def : Pat<(xor (v4i32 (bitconvert (v16i8 (NEONvshrs QPR:$src, (i32 7))))),
5004 (v4i32 (bitconvert (v16i8 (add QPR:$src,
5005 (NEONvshrs QPR:$src, (i32 7))))))),
5006 (VABSv16i8 QPR:$src)>;
5007 def : Pat<(xor (v4i32 (bitconvert (v8i16 (NEONvshrs QPR:$src, (i32 15))))),
5008 (v4i32 (bitconvert (v8i16 (add QPR:$src,
5009 (NEONvshrs QPR:$src, (i32 15))))))),
5010 (VABSv8i16 QPR:$src)>;
5011 def : Pat<(xor (v4i32 (NEONvshrs QPR:$src, (i32 31))),
5012 (v4i32 (add QPR:$src, (NEONvshrs QPR:$src, (i32 31))))),
5013 (VABSv4i32 QPR:$src)>;
5015 def : Pat<(v2f32 (int_arm_neon_vabs (v2f32 DPR:$src))), (VABSfd DPR:$src)>;
5016 def : Pat<(v4f32 (int_arm_neon_vabs (v4f32 QPR:$src))), (VABSfq QPR:$src)>;
5018 // VQABS : Vector Saturating Absolute Value
5019 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
5020 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
5021 int_arm_neon_vqabs>;
5025 def vnegd : PatFrag<(ops node:$in),
5026 (sub (bitconvert (v2i32 NEONimmAllZerosV)), node:$in)>;
5027 def vnegq : PatFrag<(ops node:$in),
5028 (sub (bitconvert (v4i32 NEONimmAllZerosV)), node:$in)>;
5030 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
5031 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$Vd), (ins DPR:$Vm),
5032 IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
5033 [(set DPR:$Vd, (Ty (vnegd DPR:$Vm)))]>;
5034 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
5035 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$Vd), (ins QPR:$Vm),
5036 IIC_VSHLiQ, OpcodeStr, Dt, "$Vd, $Vm", "",
5037 [(set QPR:$Vd, (Ty (vnegq QPR:$Vm)))]>;
5039 // VNEG : Vector Negate (integer)
5040 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
5041 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
5042 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
5043 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
5044 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
5045 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
5047 // VNEG : Vector Negate (floating-point)
5048 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
5049 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VUNAD,
5050 "vneg", "f32", "$Vd, $Vm", "",
5051 [(set DPR:$Vd, (v2f32 (fneg DPR:$Vm)))]>;
5052 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
5053 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VUNAQ,
5054 "vneg", "f32", "$Vd, $Vm", "",
5055 [(set QPR:$Vd, (v4f32 (fneg QPR:$Vm)))]>;
5057 def : Pat<(v8i8 (vnegd DPR:$src)), (VNEGs8d DPR:$src)>;
5058 def : Pat<(v4i16 (vnegd DPR:$src)), (VNEGs16d DPR:$src)>;
5059 def : Pat<(v2i32 (vnegd DPR:$src)), (VNEGs32d DPR:$src)>;
5060 def : Pat<(v16i8 (vnegq QPR:$src)), (VNEGs8q QPR:$src)>;
5061 def : Pat<(v8i16 (vnegq QPR:$src)), (VNEGs16q QPR:$src)>;
5062 def : Pat<(v4i32 (vnegq QPR:$src)), (VNEGs32q QPR:$src)>;
5064 // VQNEG : Vector Saturating Negate
5065 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
5066 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
5067 int_arm_neon_vqneg>;
5069 // Vector Bit Counting Operations.
5071 // VCLS : Vector Count Leading Sign Bits
5072 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
5073 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
5075 // VCLZ : Vector Count Leading Zeros
5076 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
5077 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
5079 // VCNT : Vector Count One Bits
5080 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
5081 IIC_VCNTiD, "vcnt", "8",
5083 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
5084 IIC_VCNTiQ, "vcnt", "8",
5085 v16i8, v16i8, ctpop>;
5088 def VSWPd : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 0, 0,
5089 (outs DPR:$Vd, DPR:$Vm), (ins DPR:$in1, DPR:$in2),
5090 NoItinerary, "vswp", "$Vd, $Vm", "$in1 = $Vd, $in2 = $Vm",
5092 def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0,
5093 (outs QPR:$Vd, QPR:$Vm), (ins QPR:$in1, QPR:$in2),
5094 NoItinerary, "vswp", "$Vd, $Vm", "$in1 = $Vd, $in2 = $Vm",
5097 // Vector Move Operations.
5099 // VMOV : Vector Move (Register)
5100 def : InstAlias<"vmov${p} $Vd, $Vm",
5101 (VORRd DPR:$Vd, DPR:$Vm, DPR:$Vm, pred:$p)>;
5102 def : InstAlias<"vmov${p} $Vd, $Vm",
5103 (VORRq QPR:$Vd, QPR:$Vm, QPR:$Vm, pred:$p)>;
5105 // VMOV : Vector Move (Immediate)
5107 let isReMaterializable = 1 in {
5108 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$Vd),
5109 (ins nImmSplatI8:$SIMM), IIC_VMOVImm,
5110 "vmov", "i8", "$Vd, $SIMM", "",
5111 [(set DPR:$Vd, (v8i8 (NEONvmovImm timm:$SIMM)))]>;
5112 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$Vd),
5113 (ins nImmSplatI8:$SIMM), IIC_VMOVImm,
5114 "vmov", "i8", "$Vd, $SIMM", "",
5115 [(set QPR:$Vd, (v16i8 (NEONvmovImm timm:$SIMM)))]>;
5117 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 0, 1, (outs DPR:$Vd),
5118 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
5119 "vmov", "i16", "$Vd, $SIMM", "",
5120 [(set DPR:$Vd, (v4i16 (NEONvmovImm timm:$SIMM)))]> {
5121 let Inst{9} = SIMM{9};
5124 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 0, 1, (outs QPR:$Vd),
5125 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
5126 "vmov", "i16", "$Vd, $SIMM", "",
5127 [(set QPR:$Vd, (v8i16 (NEONvmovImm timm:$SIMM)))]> {
5128 let Inst{9} = SIMM{9};
5131 def VMOVv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 0, 1, (outs DPR:$Vd),
5132 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
5133 "vmov", "i32", "$Vd, $SIMM", "",
5134 [(set DPR:$Vd, (v2i32 (NEONvmovImm timm:$SIMM)))]> {
5135 let Inst{11-8} = SIMM{11-8};
5138 def VMOVv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 0, 1, (outs QPR:$Vd),
5139 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
5140 "vmov", "i32", "$Vd, $SIMM", "",
5141 [(set QPR:$Vd, (v4i32 (NEONvmovImm timm:$SIMM)))]> {
5142 let Inst{11-8} = SIMM{11-8};
5145 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$Vd),
5146 (ins nImmSplatI64:$SIMM), IIC_VMOVImm,
5147 "vmov", "i64", "$Vd, $SIMM", "",
5148 [(set DPR:$Vd, (v1i64 (NEONvmovImm timm:$SIMM)))]>;
5149 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$Vd),
5150 (ins nImmSplatI64:$SIMM), IIC_VMOVImm,
5151 "vmov", "i64", "$Vd, $SIMM", "",
5152 [(set QPR:$Vd, (v2i64 (NEONvmovImm timm:$SIMM)))]>;
5154 def VMOVv2f32 : N1ModImm<1, 0b000, 0b1111, 0, 0, 0, 1, (outs DPR:$Vd),
5155 (ins nImmVMOVF32:$SIMM), IIC_VMOVImm,
5156 "vmov", "f32", "$Vd, $SIMM", "",
5157 [(set DPR:$Vd, (v2f32 (NEONvmovFPImm timm:$SIMM)))]>;
5158 def VMOVv4f32 : N1ModImm<1, 0b000, 0b1111, 0, 1, 0, 1, (outs QPR:$Vd),
5159 (ins nImmVMOVF32:$SIMM), IIC_VMOVImm,
5160 "vmov", "f32", "$Vd, $SIMM", "",
5161 [(set QPR:$Vd, (v4f32 (NEONvmovFPImm timm:$SIMM)))]>;
5162 } // isReMaterializable
5164 // VMOV : Vector Get Lane (move scalar to ARM core register)
5166 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
5167 (outs GPR:$R), (ins DPR:$V, VectorIndex8:$lane),
5168 IIC_VMOVSI, "vmov", "s8", "$R, $V$lane",
5169 [(set GPR:$R, (NEONvgetlanes (v8i8 DPR:$V),
5171 let Inst{21} = lane{2};
5172 let Inst{6-5} = lane{1-0};
5174 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
5175 (outs GPR:$R), (ins DPR:$V, VectorIndex16:$lane),
5176 IIC_VMOVSI, "vmov", "s16", "$R, $V$lane",
5177 [(set GPR:$R, (NEONvgetlanes (v4i16 DPR:$V),
5179 let Inst{21} = lane{1};
5180 let Inst{6} = lane{0};
5182 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
5183 (outs GPR:$R), (ins DPR:$V, VectorIndex8:$lane),
5184 IIC_VMOVSI, "vmov", "u8", "$R, $V$lane",
5185 [(set GPR:$R, (NEONvgetlaneu (v8i8 DPR:$V),
5187 let Inst{21} = lane{2};
5188 let Inst{6-5} = lane{1-0};
5190 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
5191 (outs GPR:$R), (ins DPR:$V, VectorIndex16:$lane),
5192 IIC_VMOVSI, "vmov", "u16", "$R, $V$lane",
5193 [(set GPR:$R, (NEONvgetlaneu (v4i16 DPR:$V),
5195 let Inst{21} = lane{1};
5196 let Inst{6} = lane{0};
5198 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
5199 (outs GPR:$R), (ins DPR:$V, VectorIndex32:$lane),
5200 IIC_VMOVSI, "vmov", "32", "$R, $V$lane",
5201 [(set GPR:$R, (extractelt (v2i32 DPR:$V),
5203 Requires<[HasNEON, HasFastVGETLNi32]> {
5204 let Inst{21} = lane{0};
5206 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
5207 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
5208 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
5209 (DSubReg_i8_reg imm:$lane))),
5210 (SubReg_i8_lane imm:$lane))>;
5211 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
5212 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
5213 (DSubReg_i16_reg imm:$lane))),
5214 (SubReg_i16_lane imm:$lane))>;
5215 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
5216 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
5217 (DSubReg_i8_reg imm:$lane))),
5218 (SubReg_i8_lane imm:$lane))>;
5219 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
5220 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
5221 (DSubReg_i16_reg imm:$lane))),
5222 (SubReg_i16_lane imm:$lane))>;
5223 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
5224 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
5225 (DSubReg_i32_reg imm:$lane))),
5226 (SubReg_i32_lane imm:$lane))>,
5227 Requires<[HasNEON, HasFastVGETLNi32]>;
5228 def : Pat<(extractelt (v2i32 DPR:$src), imm:$lane),
5230 (i32 (EXTRACT_SUBREG DPR:$src, (SSubReg_f32_reg imm:$lane))), GPR)>,
5231 Requires<[HasNEON, HasSlowVGETLNi32]>;
5232 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
5234 (i32 (EXTRACT_SUBREG QPR:$src, (SSubReg_f32_reg imm:$lane))), GPR)>,
5235 Requires<[HasNEON, HasSlowVGETLNi32]>;
5236 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
5237 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
5238 (SSubReg_f32_reg imm:$src2))>;
5239 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
5240 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
5241 (SSubReg_f32_reg imm:$src2))>;
5242 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
5243 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
5244 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
5245 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
5248 // VMOV : Vector Set Lane (move ARM core register to scalar)
5250 let Constraints = "$src1 = $V" in {
5251 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$V),
5252 (ins DPR:$src1, GPR:$R, VectorIndex8:$lane),
5253 IIC_VMOVISL, "vmov", "8", "$V$lane, $R",
5254 [(set DPR:$V, (vector_insert (v8i8 DPR:$src1),
5255 GPR:$R, imm:$lane))]> {
5256 let Inst{21} = lane{2};
5257 let Inst{6-5} = lane{1-0};
5259 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$V),
5260 (ins DPR:$src1, GPR:$R, VectorIndex16:$lane),
5261 IIC_VMOVISL, "vmov", "16", "$V$lane, $R",
5262 [(set DPR:$V, (vector_insert (v4i16 DPR:$src1),
5263 GPR:$R, imm:$lane))]> {
5264 let Inst{21} = lane{1};
5265 let Inst{6} = lane{0};
5267 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$V),
5268 (ins DPR:$src1, GPR:$R, VectorIndex32:$lane),
5269 IIC_VMOVISL, "vmov", "32", "$V$lane, $R",
5270 [(set DPR:$V, (insertelt (v2i32 DPR:$src1),
5271 GPR:$R, imm:$lane))]> {
5272 let Inst{21} = lane{0};
5275 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
5276 (v16i8 (INSERT_SUBREG QPR:$src1,
5277 (v8i8 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
5278 (DSubReg_i8_reg imm:$lane))),
5279 GPR:$src2, (SubReg_i8_lane imm:$lane))),
5280 (DSubReg_i8_reg imm:$lane)))>;
5281 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
5282 (v8i16 (INSERT_SUBREG QPR:$src1,
5283 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
5284 (DSubReg_i16_reg imm:$lane))),
5285 GPR:$src2, (SubReg_i16_lane imm:$lane))),
5286 (DSubReg_i16_reg imm:$lane)))>;
5287 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
5288 (v4i32 (INSERT_SUBREG QPR:$src1,
5289 (v2i32 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
5290 (DSubReg_i32_reg imm:$lane))),
5291 GPR:$src2, (SubReg_i32_lane imm:$lane))),
5292 (DSubReg_i32_reg imm:$lane)))>;
5294 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
5295 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
5296 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
5297 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
5298 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
5299 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
5301 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
5302 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
5303 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
5304 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
5306 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
5307 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
5308 def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
5309 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
5310 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
5311 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
5313 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
5314 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
5315 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
5316 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
5317 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
5318 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
5320 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
5321 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
5322 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
5324 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
5325 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
5326 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
5328 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
5329 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
5330 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
5333 // VDUP : Vector Duplicate (from ARM core register to all elements)
5335 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
5336 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$V), (ins GPR:$R),
5337 IIC_VMOVIS, "vdup", Dt, "$V, $R",
5338 [(set DPR:$V, (Ty (NEONvdup (i32 GPR:$R))))]>;
5339 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
5340 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$V), (ins GPR:$R),
5341 IIC_VMOVIS, "vdup", Dt, "$V, $R",
5342 [(set QPR:$V, (Ty (NEONvdup (i32 GPR:$R))))]>;
5344 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
5345 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
5346 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>,
5347 Requires<[HasNEON, HasFastVDUP32]>;
5348 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
5349 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
5350 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
5352 // NEONvdup patterns for uarchs with fast VDUP.32.
5353 def : Pat<(v2f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VDUP32d GPR:$R)>,
5354 Requires<[HasNEON,HasFastVDUP32]>;
5355 def : Pat<(v4f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VDUP32q GPR:$R)>;
5357 // NEONvdup patterns for uarchs with slow VDUP.32 - use VMOVDRR instead.
5358 def : Pat<(v2i32 (NEONvdup (i32 GPR:$R))), (VMOVDRR GPR:$R, GPR:$R)>,
5359 Requires<[HasNEON,HasSlowVDUP32]>;
5360 def : Pat<(v2f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VMOVDRR GPR:$R, GPR:$R)>,
5361 Requires<[HasNEON,HasSlowVDUP32]>;
5363 // VDUP : Vector Duplicate Lane (from scalar to all elements)
5365 class VDUPLND<bits<4> op19_16, string OpcodeStr, string Dt,
5366 ValueType Ty, Operand IdxTy>
5367 : NVDupLane<op19_16, 0, (outs DPR:$Vd), (ins DPR:$Vm, IdxTy:$lane),
5368 IIC_VMOVD, OpcodeStr, Dt, "$Vd, $Vm$lane",
5369 [(set DPR:$Vd, (Ty (NEONvduplane (Ty DPR:$Vm), imm:$lane)))]>;
5371 class VDUPLNQ<bits<4> op19_16, string OpcodeStr, string Dt,
5372 ValueType ResTy, ValueType OpTy, Operand IdxTy>
5373 : NVDupLane<op19_16, 1, (outs QPR:$Vd), (ins DPR:$Vm, IdxTy:$lane),
5374 IIC_VMOVQ, OpcodeStr, Dt, "$Vd, $Vm$lane",
5375 [(set QPR:$Vd, (ResTy (NEONvduplane (OpTy DPR:$Vm),
5376 VectorIndex32:$lane)))]>;
5378 // Inst{19-16} is partially specified depending on the element size.
5380 def VDUPLN8d : VDUPLND<{?,?,?,1}, "vdup", "8", v8i8, VectorIndex8> {
5382 let Inst{19-17} = lane{2-0};
5384 def VDUPLN16d : VDUPLND<{?,?,1,0}, "vdup", "16", v4i16, VectorIndex16> {
5386 let Inst{19-18} = lane{1-0};
5388 def VDUPLN32d : VDUPLND<{?,1,0,0}, "vdup", "32", v2i32, VectorIndex32> {
5390 let Inst{19} = lane{0};
5392 def VDUPLN8q : VDUPLNQ<{?,?,?,1}, "vdup", "8", v16i8, v8i8, VectorIndex8> {
5394 let Inst{19-17} = lane{2-0};
5396 def VDUPLN16q : VDUPLNQ<{?,?,1,0}, "vdup", "16", v8i16, v4i16, VectorIndex16> {
5398 let Inst{19-18} = lane{1-0};
5400 def VDUPLN32q : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4i32, v2i32, VectorIndex32> {
5402 let Inst{19} = lane{0};
5405 def : Pat<(v2f32 (NEONvduplane (v2f32 DPR:$Vm), imm:$lane)),
5406 (VDUPLN32d DPR:$Vm, imm:$lane)>;
5408 def : Pat<(v4f32 (NEONvduplane (v2f32 DPR:$Vm), imm:$lane)),
5409 (VDUPLN32q DPR:$Vm, imm:$lane)>;
5411 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
5412 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
5413 (DSubReg_i8_reg imm:$lane))),
5414 (SubReg_i8_lane imm:$lane)))>;
5415 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
5416 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
5417 (DSubReg_i16_reg imm:$lane))),
5418 (SubReg_i16_lane imm:$lane)))>;
5419 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
5420 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
5421 (DSubReg_i32_reg imm:$lane))),
5422 (SubReg_i32_lane imm:$lane)))>;
5423 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
5424 (v4f32 (VDUPLN32q (v2f32 (EXTRACT_SUBREG QPR:$src,
5425 (DSubReg_i32_reg imm:$lane))),
5426 (SubReg_i32_lane imm:$lane)))>;
5428 def VDUPfdf : PseudoNeonI<(outs DPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
5429 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
5430 def VDUPfqf : PseudoNeonI<(outs QPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
5431 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
5433 // VMOVN : Vector Narrowing Move
5434 defm VMOVN : N2VN_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVN,
5435 "vmovn", "i", trunc>;
5436 // VQMOVN : Vector Saturating Narrowing Move
5437 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
5438 "vqmovn", "s", int_arm_neon_vqmovns>;
5439 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
5440 "vqmovn", "u", int_arm_neon_vqmovnu>;
5441 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
5442 "vqmovun", "s", int_arm_neon_vqmovnsu>;
5443 // VMOVL : Vector Lengthening Move
5444 defm VMOVLs : N2VL_QHS<0b01,0b10100,0,1, "vmovl", "s", sext>;
5445 defm VMOVLu : N2VL_QHS<0b11,0b10100,0,1, "vmovl", "u", zext>;
5446 def : Pat<(v8i16 (anyext (v8i8 DPR:$Vm))), (VMOVLuv8i16 DPR:$Vm)>;
5447 def : Pat<(v4i32 (anyext (v4i16 DPR:$Vm))), (VMOVLuv4i32 DPR:$Vm)>;
5448 def : Pat<(v2i64 (anyext (v2i32 DPR:$Vm))), (VMOVLuv2i64 DPR:$Vm)>;
5450 // Vector Conversions.
5452 // VCVT : Vector Convert Between Floating-Point and Integers
5453 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
5454 v2i32, v2f32, fp_to_sint>;
5455 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
5456 v2i32, v2f32, fp_to_uint>;
5457 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
5458 v2f32, v2i32, sint_to_fp>;
5459 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
5460 v2f32, v2i32, uint_to_fp>;
5462 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
5463 v4i32, v4f32, fp_to_sint>;
5464 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
5465 v4i32, v4f32, fp_to_uint>;
5466 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
5467 v4f32, v4i32, sint_to_fp>;
5468 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
5469 v4f32, v4i32, uint_to_fp>;
5472 multiclass VCVT_FPI<string op, bits<3> op10_8, SDPatternOperator IntS,
5473 SDPatternOperator IntU> {
5474 let PostEncoderMethod = "NEONThumb2V8PostEncoder", DecoderNamespace = "v8NEON" in {
5475 def SD : N2VDIntnp<0b11, op10_8, 0, NoItinerary, !strconcat("vcvt", op),
5476 "s32.f32", v2i32, v2f32, IntS>, Requires<[HasV8, HasNEON]>;
5477 def SQ : N2VQIntnp<0b11, op10_8, 0, NoItinerary, !strconcat("vcvt", op),
5478 "s32.f32", v4i32, v4f32, IntS>, Requires<[HasV8, HasNEON]>;
5479 def UD : N2VDIntnp<0b11, op10_8, 1, NoItinerary, !strconcat("vcvt", op),
5480 "u32.f32", v2i32, v2f32, IntU>, Requires<[HasV8, HasNEON]>;
5481 def UQ : N2VQIntnp<0b11, op10_8, 1, NoItinerary, !strconcat("vcvt", op),
5482 "u32.f32", v4i32, v4f32, IntU>, Requires<[HasV8, HasNEON]>;
5486 defm VCVTAN : VCVT_FPI<"a", 0b000, int_arm_neon_vcvtas, int_arm_neon_vcvtau>;
5487 defm VCVTNN : VCVT_FPI<"n", 0b001, int_arm_neon_vcvtns, int_arm_neon_vcvtnu>;
5488 defm VCVTPN : VCVT_FPI<"p", 0b010, int_arm_neon_vcvtps, int_arm_neon_vcvtpu>;
5489 defm VCVTMN : VCVT_FPI<"m", 0b011, int_arm_neon_vcvtms, int_arm_neon_vcvtmu>;
5491 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
5492 let DecoderMethod = "DecodeVCVTD" in {
5493 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
5494 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
5495 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
5496 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
5497 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
5498 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
5499 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
5500 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
5503 let DecoderMethod = "DecodeVCVTQ" in {
5504 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
5505 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
5506 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
5507 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
5508 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
5509 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
5510 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
5511 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
5514 def : NEONInstAlias<"vcvt${p}.s32.f32 $Dd, $Dm, #0",
5515 (VCVTf2sd DPR:$Dd, DPR:$Dm, pred:$p)>;
5516 def : NEONInstAlias<"vcvt${p}.u32.f32 $Dd, $Dm, #0",
5517 (VCVTf2ud DPR:$Dd, DPR:$Dm, pred:$p)>;
5518 def : NEONInstAlias<"vcvt${p}.f32.s32 $Dd, $Dm, #0",
5519 (VCVTs2fd DPR:$Dd, DPR:$Dm, pred:$p)>;
5520 def : NEONInstAlias<"vcvt${p}.f32.u32 $Dd, $Dm, #0",
5521 (VCVTu2fd DPR:$Dd, DPR:$Dm, pred:$p)>;
5523 def : NEONInstAlias<"vcvt${p}.s32.f32 $Qd, $Qm, #0",
5524 (VCVTf2sq QPR:$Qd, QPR:$Qm, pred:$p)>;
5525 def : NEONInstAlias<"vcvt${p}.u32.f32 $Qd, $Qm, #0",
5526 (VCVTf2uq QPR:$Qd, QPR:$Qm, pred:$p)>;
5527 def : NEONInstAlias<"vcvt${p}.f32.s32 $Qd, $Qm, #0",
5528 (VCVTs2fq QPR:$Qd, QPR:$Qm, pred:$p)>;
5529 def : NEONInstAlias<"vcvt${p}.f32.u32 $Qd, $Qm, #0",
5530 (VCVTu2fq QPR:$Qd, QPR:$Qm, pred:$p)>;
5533 // VCVT : Vector Convert Between Half-Precision and Single-Precision.
5534 def VCVTf2h : N2VNInt<0b11, 0b11, 0b01, 0b10, 0b01100, 0, 0,
5535 IIC_VUNAQ, "vcvt", "f16.f32",
5536 v4i16, v4f32, int_arm_neon_vcvtfp2hf>,
5537 Requires<[HasNEON, HasFP16]>;
5538 def VCVTh2f : N2VLInt<0b11, 0b11, 0b01, 0b10, 0b01110, 0, 0,
5539 IIC_VUNAQ, "vcvt", "f32.f16",
5540 v4f32, v4i16, int_arm_neon_vcvthf2fp>,
5541 Requires<[HasNEON, HasFP16]>;
5545 // VREV64 : Vector Reverse elements within 64-bit doublewords
5547 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5548 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$Vd),
5549 (ins DPR:$Vm), IIC_VMOVD,
5550 OpcodeStr, Dt, "$Vd, $Vm", "",
5551 [(set DPR:$Vd, (Ty (NEONvrev64 (Ty DPR:$Vm))))]>;
5552 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5553 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$Vd),
5554 (ins QPR:$Vm), IIC_VMOVQ,
5555 OpcodeStr, Dt, "$Vd, $Vm", "",
5556 [(set QPR:$Vd, (Ty (NEONvrev64 (Ty QPR:$Vm))))]>;
5558 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
5559 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
5560 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
5561 def : Pat<(v2f32 (NEONvrev64 (v2f32 DPR:$Vm))), (VREV64d32 DPR:$Vm)>;
5563 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
5564 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
5565 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
5566 def : Pat<(v4f32 (NEONvrev64 (v4f32 QPR:$Vm))), (VREV64q32 QPR:$Vm)>;
5568 // VREV32 : Vector Reverse elements within 32-bit words
5570 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5571 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$Vd),
5572 (ins DPR:$Vm), IIC_VMOVD,
5573 OpcodeStr, Dt, "$Vd, $Vm", "",
5574 [(set DPR:$Vd, (Ty (NEONvrev32 (Ty DPR:$Vm))))]>;
5575 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5576 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$Vd),
5577 (ins QPR:$Vm), IIC_VMOVQ,
5578 OpcodeStr, Dt, "$Vd, $Vm", "",
5579 [(set QPR:$Vd, (Ty (NEONvrev32 (Ty QPR:$Vm))))]>;
5581 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
5582 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
5584 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
5585 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
5587 // VREV16 : Vector Reverse elements within 16-bit halfwords
5589 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5590 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$Vd),
5591 (ins DPR:$Vm), IIC_VMOVD,
5592 OpcodeStr, Dt, "$Vd, $Vm", "",
5593 [(set DPR:$Vd, (Ty (NEONvrev16 (Ty DPR:$Vm))))]>;
5594 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5595 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$Vd),
5596 (ins QPR:$Vm), IIC_VMOVQ,
5597 OpcodeStr, Dt, "$Vd, $Vm", "",
5598 [(set QPR:$Vd, (Ty (NEONvrev16 (Ty QPR:$Vm))))]>;
5600 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
5601 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
5603 // Other Vector Shuffles.
5605 // Aligned extractions: really just dropping registers
5607 class AlignedVEXTq<ValueType DestTy, ValueType SrcTy, SDNodeXForm LaneCVT>
5608 : Pat<(DestTy (vector_extract_subvec (SrcTy QPR:$src), (i32 imm:$start))),
5609 (EXTRACT_SUBREG (SrcTy QPR:$src), (LaneCVT imm:$start))>;
5611 def : AlignedVEXTq<v8i8, v16i8, DSubReg_i8_reg>;
5613 def : AlignedVEXTq<v4i16, v8i16, DSubReg_i16_reg>;
5615 def : AlignedVEXTq<v2i32, v4i32, DSubReg_i32_reg>;
5617 def : AlignedVEXTq<v1i64, v2i64, DSubReg_f64_reg>;
5619 def : AlignedVEXTq<v2f32, v4f32, DSubReg_i32_reg>;
5622 // VEXT : Vector Extract
5625 // All of these have a two-operand InstAlias.
5626 let TwoOperandAliasConstraint = "$Vn = $Vd" in {
5627 class VEXTd<string OpcodeStr, string Dt, ValueType Ty, Operand immTy>
5628 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$Vd),
5629 (ins DPR:$Vn, DPR:$Vm, immTy:$index), NVExtFrm,
5630 IIC_VEXTD, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
5631 [(set DPR:$Vd, (Ty (NEONvext (Ty DPR:$Vn),
5632 (Ty DPR:$Vm), imm:$index)))]> {
5635 let Inst{10-8} = index{2-0};
5638 class VEXTq<string OpcodeStr, string Dt, ValueType Ty, Operand immTy>
5639 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$Vd),
5640 (ins QPR:$Vn, QPR:$Vm, imm0_15:$index), NVExtFrm,
5641 IIC_VEXTQ, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
5642 [(set QPR:$Vd, (Ty (NEONvext (Ty QPR:$Vn),
5643 (Ty QPR:$Vm), imm:$index)))]> {
5645 let Inst{11-8} = index{3-0};
5649 def VEXTd8 : VEXTd<"vext", "8", v8i8, imm0_7> {
5650 let Inst{10-8} = index{2-0};
5652 def VEXTd16 : VEXTd<"vext", "16", v4i16, imm0_3> {
5653 let Inst{10-9} = index{1-0};
5656 def VEXTd32 : VEXTd<"vext", "32", v2i32, imm0_1> {
5657 let Inst{10} = index{0};
5658 let Inst{9-8} = 0b00;
5660 def : Pat<(v2f32 (NEONvext (v2f32 DPR:$Vn),
5663 (VEXTd32 DPR:$Vn, DPR:$Vm, imm:$index)>;
5665 def VEXTq8 : VEXTq<"vext", "8", v16i8, imm0_15> {
5666 let Inst{11-8} = index{3-0};
5668 def VEXTq16 : VEXTq<"vext", "16", v8i16, imm0_7> {
5669 let Inst{11-9} = index{2-0};
5672 def VEXTq32 : VEXTq<"vext", "32", v4i32, imm0_3> {
5673 let Inst{11-10} = index{1-0};
5674 let Inst{9-8} = 0b00;
5676 def VEXTq64 : VEXTq<"vext", "64", v2i64, imm0_1> {
5677 let Inst{11} = index{0};
5678 let Inst{10-8} = 0b000;
5680 def : Pat<(v4f32 (NEONvext (v4f32 QPR:$Vn),
5683 (VEXTq32 QPR:$Vn, QPR:$Vm, imm:$index)>;
5685 // VTRN : Vector Transpose
5687 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
5688 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
5689 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
5691 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
5692 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
5693 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
5695 // VUZP : Vector Unzip (Deinterleave)
5697 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
5698 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
5699 // vuzp.32 Dd, Dm is a pseudo-instruction expanded to vtrn.32 Dd, Dm.
5700 def : NEONInstAlias<"vuzp${p}.32 $Dd, $Dm",
5701 (VTRNd32 DPR:$Dd, DPR:$Dm, pred:$p)>;
5703 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
5704 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
5705 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
5707 // VZIP : Vector Zip (Interleave)
5709 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
5710 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
5711 // vzip.32 Dd, Dm is a pseudo-instruction expanded to vtrn.32 Dd, Dm.
5712 def : NEONInstAlias<"vzip${p}.32 $Dd, $Dm",
5713 (VTRNd32 DPR:$Dd, DPR:$Dm, pred:$p)>;
5715 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
5716 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
5717 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
5719 // Vector Table Lookup and Table Extension.
5721 // VTBL : Vector Table Lookup
5722 let DecoderMethod = "DecodeTBLInstruction" in {
5724 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$Vd),
5725 (ins VecListOneD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB1,
5726 "vtbl", "8", "$Vd, $Vn, $Vm", "",
5727 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbl1 VecListOneD:$Vn, DPR:$Vm)))]>;
5728 let hasExtraSrcRegAllocReq = 1 in {
5730 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$Vd),
5731 (ins VecListDPair:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB2,
5732 "vtbl", "8", "$Vd, $Vn, $Vm", "", []>;
5734 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$Vd),
5735 (ins VecListThreeD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB3,
5736 "vtbl", "8", "$Vd, $Vn, $Vm", "", []>;
5738 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$Vd),
5739 (ins VecListFourD:$Vn, DPR:$Vm),
5741 "vtbl", "8", "$Vd, $Vn, $Vm", "", []>;
5742 } // hasExtraSrcRegAllocReq = 1
5745 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB3, "", []>;
5747 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB4, "", []>;
5749 // VTBX : Vector Table Extension
5751 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$Vd),
5752 (ins DPR:$orig, VecListOneD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX1,
5753 "vtbx", "8", "$Vd, $Vn, $Vm", "$orig = $Vd",
5754 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbx1
5755 DPR:$orig, VecListOneD:$Vn, DPR:$Vm)))]>;
5756 let hasExtraSrcRegAllocReq = 1 in {
5758 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$Vd),
5759 (ins DPR:$orig, VecListDPair:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX2,
5760 "vtbx", "8", "$Vd, $Vn, $Vm", "$orig = $Vd", []>;
5762 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$Vd),
5763 (ins DPR:$orig, VecListThreeD:$Vn, DPR:$Vm),
5764 NVTBLFrm, IIC_VTBX3,
5765 "vtbx", "8", "$Vd, $Vn, $Vm",
5768 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$Vd),
5769 (ins DPR:$orig, VecListFourD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX4,
5770 "vtbx", "8", "$Vd, $Vn, $Vm",
5772 } // hasExtraSrcRegAllocReq = 1
5775 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
5776 IIC_VTBX3, "$orig = $dst", []>;
5778 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
5779 IIC_VTBX4, "$orig = $dst", []>;
5780 } // DecoderMethod = "DecodeTBLInstruction"
5782 // VRINT : Vector Rounding
5783 multiclass VRINT_FPI<string op, bits<3> op9_7, SDPatternOperator Int> {
5784 let PostEncoderMethod = "NEONThumb2V8PostEncoder", DecoderNamespace = "v8NEON" in {
5785 def D : N2VDIntnp<0b10, 0b100, 0, NoItinerary,
5786 !strconcat("vrint", op), "f32",
5787 v2f32, v2f32, Int>, Requires<[HasV8, HasNEON]> {
5788 let Inst{9-7} = op9_7;
5790 def Q : N2VQIntnp<0b10, 0b100, 0, NoItinerary,
5791 !strconcat("vrint", op), "f32",
5792 v4f32, v4f32, Int>, Requires<[HasV8, HasNEON]> {
5793 let Inst{9-7} = op9_7;
5797 def : NEONInstAlias<!strconcat("vrint", op, ".f32.f32\t$Dd, $Dm"),
5798 (!cast<Instruction>(NAME#"D") DPR:$Dd, DPR:$Dm)>;
5799 def : NEONInstAlias<!strconcat("vrint", op, ".f32.f32\t$Qd, $Qm"),
5800 (!cast<Instruction>(NAME#"Q") QPR:$Qd, QPR:$Qm)>;
5803 defm VRINTNN : VRINT_FPI<"n", 0b000, int_arm_neon_vrintn>;
5804 defm VRINTXN : VRINT_FPI<"x", 0b001, int_arm_neon_vrintx>;
5805 defm VRINTAN : VRINT_FPI<"a", 0b010, int_arm_neon_vrinta>;
5806 defm VRINTZN : VRINT_FPI<"z", 0b011, int_arm_neon_vrintz>;
5807 defm VRINTMN : VRINT_FPI<"m", 0b101, int_arm_neon_vrintm>;
5808 defm VRINTPN : VRINT_FPI<"p", 0b111, int_arm_neon_vrintp>;
5810 //===----------------------------------------------------------------------===//
5811 // NEON instructions for single-precision FP math
5812 //===----------------------------------------------------------------------===//
5814 class N2VSPat<SDNode OpNode, NeonI Inst>
5815 : NEONFPPat<(f32 (OpNode SPR:$a)),
5817 (v2f32 (COPY_TO_REGCLASS (Inst
5819 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5820 SPR:$a, ssub_0)), DPR_VFP2)), ssub_0)>;
5822 class N3VSPat<SDNode OpNode, NeonI Inst>
5823 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
5825 (v2f32 (COPY_TO_REGCLASS (Inst
5827 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5830 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5831 SPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>;
5833 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
5834 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
5836 (v2f32 (COPY_TO_REGCLASS (Inst
5838 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5841 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5844 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5845 SPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>;
5847 def : N3VSPat<fadd, VADDfd>;
5848 def : N3VSPat<fsub, VSUBfd>;
5849 def : N3VSPat<fmul, VMULfd>;
5850 def : N3VSMulOpPat<fmul, fadd, VMLAfd>,
5851 Requires<[HasNEON, UseNEONForFP, UseFPVMLx, DontUseFusedMAC]>;
5852 def : N3VSMulOpPat<fmul, fsub, VMLSfd>,
5853 Requires<[HasNEON, UseNEONForFP, UseFPVMLx, DontUseFusedMAC]>;
5854 def : N3VSMulOpPat<fmul, fadd, VFMAfd>,
5855 Requires<[HasVFP4, UseNEONForFP, UseFusedMAC]>;
5856 def : N3VSMulOpPat<fmul, fsub, VFMSfd>,
5857 Requires<[HasVFP4, UseNEONForFP, UseFusedMAC]>;
5858 def : N2VSPat<fabs, VABSfd>;
5859 def : N2VSPat<fneg, VNEGfd>;
5860 def : N3VSPat<NEONfmax, VMAXfd>;
5861 def : N3VSPat<NEONfmin, VMINfd>;
5862 def : N2VSPat<arm_ftosi, VCVTf2sd>;
5863 def : N2VSPat<arm_ftoui, VCVTf2ud>;
5864 def : N2VSPat<arm_sitof, VCVTs2fd>;
5865 def : N2VSPat<arm_uitof, VCVTu2fd>;
5867 // Prefer VMOVDRR for i32 -> f32 bitcasts, it can write all DPR registers.
5868 def : Pat<(f32 (bitconvert GPR:$a)),
5869 (EXTRACT_SUBREG (VMOVDRR GPR:$a, GPR:$a), ssub_0)>,
5870 Requires<[HasNEON, DontUseVMOVSR]>;
5872 //===----------------------------------------------------------------------===//
5873 // Non-Instruction Patterns
5874 //===----------------------------------------------------------------------===//
5877 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
5878 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
5879 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
5880 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
5881 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
5882 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
5883 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
5884 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
5885 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
5886 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
5887 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
5888 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
5889 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
5890 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
5891 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
5892 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
5893 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
5894 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
5895 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
5896 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
5897 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
5898 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
5899 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
5900 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
5901 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
5902 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
5903 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
5904 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
5905 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
5906 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
5908 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
5909 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
5910 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
5911 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
5912 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
5913 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
5914 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
5915 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
5916 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
5917 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
5918 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
5919 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
5920 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
5921 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
5922 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
5923 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
5924 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
5925 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
5926 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
5927 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
5928 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
5929 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
5930 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
5931 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
5932 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
5933 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
5934 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
5935 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
5936 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
5937 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;
5939 // Fold extracting an element out of a v2i32 into a vfp register.
5940 def : Pat<(f32 (bitconvert (i32 (extractelt (v2i32 DPR:$src), imm:$lane)))),
5941 (f32 (EXTRACT_SUBREG DPR:$src, (SSubReg_f32_reg imm:$lane)))>;
5943 // Vector lengthening move with load, matching extending loads.
5945 // extload, zextload and sextload for a standard lengthening load. Example:
5946 // Lengthen_Single<"8", "i16", "8"> =
5947 // Pat<(v8i16 (extloadvi8 addrmode6:$addr))
5948 // (VMOVLuv8i16 (VLD1d8 addrmode6:$addr,
5949 // (f64 (IMPLICIT_DEF)), (i32 0)))>;
5950 multiclass Lengthen_Single<string DestLanes, string DestTy, string SrcTy> {
5951 let AddedComplexity = 10 in {
5952 def _Any : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5953 (!cast<PatFrag>("extloadvi" # SrcTy) addrmode6:$addr)),
5954 (!cast<Instruction>("VMOVLuv" # DestLanes # DestTy)
5955 (!cast<Instruction>("VLD1d" # SrcTy) addrmode6:$addr))>;
5957 def _Z : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5958 (!cast<PatFrag>("zextloadvi" # SrcTy) addrmode6:$addr)),
5959 (!cast<Instruction>("VMOVLuv" # DestLanes # DestTy)
5960 (!cast<Instruction>("VLD1d" # SrcTy) addrmode6:$addr))>;
5962 def _S : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5963 (!cast<PatFrag>("sextloadvi" # SrcTy) addrmode6:$addr)),
5964 (!cast<Instruction>("VMOVLsv" # DestLanes # DestTy)
5965 (!cast<Instruction>("VLD1d" # SrcTy) addrmode6:$addr))>;
5969 // extload, zextload and sextload for a lengthening load which only uses
5970 // half the lanes available. Example:
5971 // Lengthen_HalfSingle<"4", "i16", "8", "i16", "i8"> =
5972 // Pat<(v4i16 (extloadvi8 addrmode6oneL32:$addr)),
5973 // (EXTRACT_SUBREG (VMOVLuv8i16 (VLD1LNd32 addrmode6oneL32:$addr,
5974 // (f64 (IMPLICIT_DEF)), (i32 0))),
5976 multiclass Lengthen_HalfSingle<string DestLanes, string DestTy, string SrcTy,
5977 string InsnLanes, string InsnTy> {
5978 def _Any : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5979 (!cast<PatFrag>("extloadv" # SrcTy) addrmode6oneL32:$addr)),
5980 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # InsnLanes # InsnTy)
5981 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
5983 def _Z : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5984 (!cast<PatFrag>("zextloadv" # SrcTy) addrmode6oneL32:$addr)),
5985 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # InsnLanes # InsnTy)
5986 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
5988 def _S : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5989 (!cast<PatFrag>("sextloadv" # SrcTy) addrmode6oneL32:$addr)),
5990 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLsv" # InsnLanes # InsnTy)
5991 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
5995 // extload, zextload and sextload for a lengthening load followed by another
5996 // lengthening load, to quadruple the initial length.
5998 // Lengthen_Double<"4", "i32", "i8", "8", "i16", "4", "i32"> =
5999 // Pat<(v4i32 (extloadvi8 addrmode6oneL32:$addr))
6000 // (EXTRACT_SUBREG (VMOVLuv4i32
6001 // (EXTRACT_SUBREG (VMOVLuv8i16 (VLD1LNd32 addrmode6oneL32:$addr,
6002 // (f64 (IMPLICIT_DEF)),
6006 multiclass Lengthen_Double<string DestLanes, string DestTy, string SrcTy,
6007 string Insn1Lanes, string Insn1Ty, string Insn2Lanes,
6009 def _Any : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
6010 (!cast<PatFrag>("extloadv" # SrcTy) addrmode6oneL32:$addr)),
6011 (!cast<Instruction>("VMOVLuv" # Insn2Lanes # Insn2Ty)
6012 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn1Lanes # Insn1Ty)
6013 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
6015 def _Z : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
6016 (!cast<PatFrag>("zextloadv" # SrcTy) addrmode6oneL32:$addr)),
6017 (!cast<Instruction>("VMOVLuv" # Insn2Lanes # Insn2Ty)
6018 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn1Lanes # Insn1Ty)
6019 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
6021 def _S : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
6022 (!cast<PatFrag>("sextloadv" # SrcTy) addrmode6oneL32:$addr)),
6023 (!cast<Instruction>("VMOVLsv" # Insn2Lanes # Insn2Ty)
6024 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLsv" # Insn1Lanes # Insn1Ty)
6025 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
6029 // extload, zextload and sextload for a lengthening load followed by another
6030 // lengthening load, to quadruple the initial length, but which ends up only
6031 // requiring half the available lanes (a 64-bit outcome instead of a 128-bit).
6033 // Lengthen_HalfDouble<"2", "i32", "i8", "8", "i16", "4", "i32"> =
6034 // Pat<(v2i32 (extloadvi8 addrmode6:$addr))
6035 // (EXTRACT_SUBREG (VMOVLuv4i32
6036 // (EXTRACT_SUBREG (VMOVLuv8i16 (VLD1LNd16 addrmode6:$addr,
6037 // (f64 (IMPLICIT_DEF)), (i32 0))),
6040 multiclass Lengthen_HalfDouble<string DestLanes, string DestTy, string SrcTy,
6041 string Insn1Lanes, string Insn1Ty, string Insn2Lanes,
6043 def _Any : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
6044 (!cast<PatFrag>("extloadv" # SrcTy) addrmode6:$addr)),
6045 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn2Lanes # Insn2Ty)
6046 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn1Lanes # Insn1Ty)
6047 (VLD1LNd16 addrmode6:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
6050 def _Z : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
6051 (!cast<PatFrag>("zextloadv" # SrcTy) addrmode6:$addr)),
6052 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn2Lanes # Insn2Ty)
6053 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn1Lanes # Insn1Ty)
6054 (VLD1LNd16 addrmode6:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
6057 def _S : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
6058 (!cast<PatFrag>("sextloadv" # SrcTy) addrmode6:$addr)),
6059 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLsv" # Insn2Lanes # Insn2Ty)
6060 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLsv" # Insn1Lanes # Insn1Ty)
6061 (VLD1LNd16 addrmode6:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
6066 defm : Lengthen_Single<"8", "i16", "8">; // v8i8 -> v8i16
6067 defm : Lengthen_Single<"4", "i32", "16">; // v4i16 -> v4i32
6068 defm : Lengthen_Single<"2", "i64", "32">; // v2i32 -> v2i64
6070 defm : Lengthen_HalfSingle<"4", "i16", "i8", "8", "i16">; // v4i8 -> v4i16
6071 defm : Lengthen_HalfSingle<"2", "i32", "i16", "4", "i32">; // v2i16 -> v2i32
6073 // Double lengthening - v4i8 -> v4i16 -> v4i32
6074 defm : Lengthen_Double<"4", "i32", "i8", "8", "i16", "4", "i32">;
6075 // v2i8 -> v2i16 -> v2i32
6076 defm : Lengthen_HalfDouble<"2", "i32", "i8", "8", "i16", "4", "i32">;
6077 // v2i16 -> v2i32 -> v2i64
6078 defm : Lengthen_Double<"2", "i64", "i16", "4", "i32", "2", "i64">;
6080 // Triple lengthening - v2i8 -> v2i16 -> v2i32 -> v2i64
6081 def : Pat<(v2i64 (extloadvi8 addrmode6:$addr)),
6082 (VMOVLuv2i64 (EXTRACT_SUBREG (VMOVLuv4i32 (EXTRACT_SUBREG (VMOVLuv8i16
6083 (VLD1LNd16 addrmode6:$addr,
6084 (f64 (IMPLICIT_DEF)), (i32 0))), dsub_0)), dsub_0))>;
6085 def : Pat<(v2i64 (zextloadvi8 addrmode6:$addr)),
6086 (VMOVLuv2i64 (EXTRACT_SUBREG (VMOVLuv4i32 (EXTRACT_SUBREG (VMOVLuv8i16
6087 (VLD1LNd16 addrmode6:$addr,
6088 (f64 (IMPLICIT_DEF)), (i32 0))), dsub_0)), dsub_0))>;
6089 def : Pat<(v2i64 (sextloadvi8 addrmode6:$addr)),
6090 (VMOVLsv2i64 (EXTRACT_SUBREG (VMOVLsv4i32 (EXTRACT_SUBREG (VMOVLsv8i16
6091 (VLD1LNd16 addrmode6:$addr,
6092 (f64 (IMPLICIT_DEF)), (i32 0))), dsub_0)), dsub_0))>;
6094 //===----------------------------------------------------------------------===//
6095 // Assembler aliases
6098 def : VFP2InstAlias<"fmdhr${p} $Dd, $Rn",
6099 (VSETLNi32 DPR:$Dd, GPR:$Rn, 1, pred:$p)>;
6100 def : VFP2InstAlias<"fmdlr${p} $Dd, $Rn",
6101 (VSETLNi32 DPR:$Dd, GPR:$Rn, 0, pred:$p)>;
6103 // VAND/VBIC/VEOR/VORR accept but do not require a type suffix.
6104 defm : NEONDTAnyInstAlias<"vand${p}", "$Vd, $Vn, $Vm",
6105 (VANDd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
6106 defm : NEONDTAnyInstAlias<"vand${p}", "$Vd, $Vn, $Vm",
6107 (VANDq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
6108 defm : NEONDTAnyInstAlias<"vbic${p}", "$Vd, $Vn, $Vm",
6109 (VBICd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
6110 defm : NEONDTAnyInstAlias<"vbic${p}", "$Vd, $Vn, $Vm",
6111 (VBICq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
6112 defm : NEONDTAnyInstAlias<"veor${p}", "$Vd, $Vn, $Vm",
6113 (VEORd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
6114 defm : NEONDTAnyInstAlias<"veor${p}", "$Vd, $Vn, $Vm",
6115 (VEORq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
6116 defm : NEONDTAnyInstAlias<"vorr${p}", "$Vd, $Vn, $Vm",
6117 (VORRd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
6118 defm : NEONDTAnyInstAlias<"vorr${p}", "$Vd, $Vn, $Vm",
6119 (VORRq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
6120 // ... two-operand aliases
6121 defm : NEONDTAnyInstAlias<"vand${p}", "$Vdn, $Vm",
6122 (VANDd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
6123 defm : NEONDTAnyInstAlias<"vand${p}", "$Vdn, $Vm",
6124 (VANDq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
6125 defm : NEONDTAnyInstAlias<"veor${p}", "$Vdn, $Vm",
6126 (VEORd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
6127 defm : NEONDTAnyInstAlias<"veor${p}", "$Vdn, $Vm",
6128 (VEORq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
6129 defm : NEONDTAnyInstAlias<"vorr${p}", "$Vdn, $Vm",
6130 (VORRd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
6131 defm : NEONDTAnyInstAlias<"vorr${p}", "$Vdn, $Vm",
6132 (VORRq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
6134 // VLD1 single-lane pseudo-instructions. These need special handling for
6135 // the lane index that an InstAlias can't handle, so we use these instead.
6136 def VLD1LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vld1${p}", ".8", "$list, $addr",
6137 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6138 def VLD1LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vld1${p}", ".16", "$list, $addr",
6139 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6140 def VLD1LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vld1${p}", ".32", "$list, $addr",
6141 (ins VecListOneDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6143 def VLD1LNdWB_fixed_Asm_8 :
6144 NEONDataTypeAsmPseudoInst<"vld1${p}", ".8", "$list, $addr!",
6145 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6146 def VLD1LNdWB_fixed_Asm_16 :
6147 NEONDataTypeAsmPseudoInst<"vld1${p}", ".16", "$list, $addr!",
6148 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6149 def VLD1LNdWB_fixed_Asm_32 :
6150 NEONDataTypeAsmPseudoInst<"vld1${p}", ".32", "$list, $addr!",
6151 (ins VecListOneDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6152 def VLD1LNdWB_register_Asm_8 :
6153 NEONDataTypeAsmPseudoInst<"vld1${p}", ".8", "$list, $addr, $Rm",
6154 (ins VecListOneDByteIndexed:$list, addrmode6:$addr,
6155 rGPR:$Rm, pred:$p)>;
6156 def VLD1LNdWB_register_Asm_16 :
6157 NEONDataTypeAsmPseudoInst<"vld1${p}", ".16", "$list, $addr, $Rm",
6158 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr,
6159 rGPR:$Rm, pred:$p)>;
6160 def VLD1LNdWB_register_Asm_32 :
6161 NEONDataTypeAsmPseudoInst<"vld1${p}", ".32", "$list, $addr, $Rm",
6162 (ins VecListOneDWordIndexed:$list, addrmode6:$addr,
6163 rGPR:$Rm, pred:$p)>;
6166 // VST1 single-lane pseudo-instructions. These need special handling for
6167 // the lane index that an InstAlias can't handle, so we use these instead.
6168 def VST1LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vst1${p}", ".8", "$list, $addr",
6169 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6170 def VST1LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vst1${p}", ".16", "$list, $addr",
6171 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6172 def VST1LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vst1${p}", ".32", "$list, $addr",
6173 (ins VecListOneDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6175 def VST1LNdWB_fixed_Asm_8 :
6176 NEONDataTypeAsmPseudoInst<"vst1${p}", ".8", "$list, $addr!",
6177 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6178 def VST1LNdWB_fixed_Asm_16 :
6179 NEONDataTypeAsmPseudoInst<"vst1${p}", ".16", "$list, $addr!",
6180 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6181 def VST1LNdWB_fixed_Asm_32 :
6182 NEONDataTypeAsmPseudoInst<"vst1${p}", ".32", "$list, $addr!",
6183 (ins VecListOneDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6184 def VST1LNdWB_register_Asm_8 :
6185 NEONDataTypeAsmPseudoInst<"vst1${p}", ".8", "$list, $addr, $Rm",
6186 (ins VecListOneDByteIndexed:$list, addrmode6:$addr,
6187 rGPR:$Rm, pred:$p)>;
6188 def VST1LNdWB_register_Asm_16 :
6189 NEONDataTypeAsmPseudoInst<"vst1${p}", ".16", "$list, $addr, $Rm",
6190 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr,
6191 rGPR:$Rm, pred:$p)>;
6192 def VST1LNdWB_register_Asm_32 :
6193 NEONDataTypeAsmPseudoInst<"vst1${p}", ".32", "$list, $addr, $Rm",
6194 (ins VecListOneDWordIndexed:$list, addrmode6:$addr,
6195 rGPR:$Rm, pred:$p)>;
6197 // VLD2 single-lane pseudo-instructions. These need special handling for
6198 // the lane index that an InstAlias can't handle, so we use these instead.
6199 def VLD2LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vld2${p}", ".8", "$list, $addr",
6200 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6201 def VLD2LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr",
6202 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6203 def VLD2LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr",
6204 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6205 def VLD2LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr",
6206 (ins VecListTwoQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6207 def VLD2LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr",
6208 (ins VecListTwoQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6210 def VLD2LNdWB_fixed_Asm_8 :
6211 NEONDataTypeAsmPseudoInst<"vld2${p}", ".8", "$list, $addr!",
6212 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6213 def VLD2LNdWB_fixed_Asm_16 :
6214 NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr!",
6215 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6216 def VLD2LNdWB_fixed_Asm_32 :
6217 NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr!",
6218 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6219 def VLD2LNqWB_fixed_Asm_16 :
6220 NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr!",
6221 (ins VecListTwoQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6222 def VLD2LNqWB_fixed_Asm_32 :
6223 NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr!",
6224 (ins VecListTwoQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6225 def VLD2LNdWB_register_Asm_8 :
6226 NEONDataTypeAsmPseudoInst<"vld2${p}", ".8", "$list, $addr, $Rm",
6227 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr,
6228 rGPR:$Rm, pred:$p)>;
6229 def VLD2LNdWB_register_Asm_16 :
6230 NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr, $Rm",
6231 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr,
6232 rGPR:$Rm, pred:$p)>;
6233 def VLD2LNdWB_register_Asm_32 :
6234 NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr, $Rm",
6235 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr,
6236 rGPR:$Rm, pred:$p)>;
6237 def VLD2LNqWB_register_Asm_16 :
6238 NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr, $Rm",
6239 (ins VecListTwoQHWordIndexed:$list, addrmode6:$addr,
6240 rGPR:$Rm, pred:$p)>;
6241 def VLD2LNqWB_register_Asm_32 :
6242 NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr, $Rm",
6243 (ins VecListTwoQWordIndexed:$list, addrmode6:$addr,
6244 rGPR:$Rm, pred:$p)>;
6247 // VST2 single-lane pseudo-instructions. These need special handling for
6248 // the lane index that an InstAlias can't handle, so we use these instead.
6249 def VST2LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vst2${p}", ".8", "$list, $addr",
6250 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6251 def VST2LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vst2${p}", ".16", "$list, $addr",
6252 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6253 def VST2LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr",
6254 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6255 def VST2LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vst2${p}", ".16", "$list, $addr",
6256 (ins VecListTwoQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6257 def VST2LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr",
6258 (ins VecListTwoQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6260 def VST2LNdWB_fixed_Asm_8 :
6261 NEONDataTypeAsmPseudoInst<"vst2${p}", ".8", "$list, $addr!",
6262 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6263 def VST2LNdWB_fixed_Asm_16 :
6264 NEONDataTypeAsmPseudoInst<"vst2${p}", ".16", "$list, $addr!",
6265 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6266 def VST2LNdWB_fixed_Asm_32 :
6267 NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr!",
6268 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6269 def VST2LNqWB_fixed_Asm_16 :
6270 NEONDataTypeAsmPseudoInst<"vst2${p}", ".16", "$list, $addr!",
6271 (ins VecListTwoQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6272 def VST2LNqWB_fixed_Asm_32 :
6273 NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr!",
6274 (ins VecListTwoQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6275 def VST2LNdWB_register_Asm_8 :
6276 NEONDataTypeAsmPseudoInst<"vst2${p}", ".8", "$list, $addr, $Rm",
6277 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr,
6278 rGPR:$Rm, pred:$p)>;
6279 def VST2LNdWB_register_Asm_16 :
6280 NEONDataTypeAsmPseudoInst<"vst2${p}", ".16","$list, $addr, $Rm",
6281 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr,
6282 rGPR:$Rm, pred:$p)>;
6283 def VST2LNdWB_register_Asm_32 :
6284 NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr, $Rm",
6285 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr,
6286 rGPR:$Rm, pred:$p)>;
6287 def VST2LNqWB_register_Asm_16 :
6288 NEONDataTypeAsmPseudoInst<"vst2${p}", ".16","$list, $addr, $Rm",
6289 (ins VecListTwoQHWordIndexed:$list, addrmode6:$addr,
6290 rGPR:$Rm, pred:$p)>;
6291 def VST2LNqWB_register_Asm_32 :
6292 NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr, $Rm",
6293 (ins VecListTwoQWordIndexed:$list, addrmode6:$addr,
6294 rGPR:$Rm, pred:$p)>;
6296 // VLD3 all-lanes pseudo-instructions. These need special handling for
6297 // the lane index that an InstAlias can't handle, so we use these instead.
6298 def VLD3DUPdAsm_8 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr",
6299 (ins VecListThreeDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6300 def VLD3DUPdAsm_16: NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
6301 (ins VecListThreeDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6302 def VLD3DUPdAsm_32: NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
6303 (ins VecListThreeDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6304 def VLD3DUPqAsm_8 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr",
6305 (ins VecListThreeQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6306 def VLD3DUPqAsm_16: NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
6307 (ins VecListThreeQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6308 def VLD3DUPqAsm_32: NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
6309 (ins VecListThreeQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6311 def VLD3DUPdWB_fixed_Asm_8 :
6312 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr!",
6313 (ins VecListThreeDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6314 def VLD3DUPdWB_fixed_Asm_16 :
6315 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
6316 (ins VecListThreeDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6317 def VLD3DUPdWB_fixed_Asm_32 :
6318 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
6319 (ins VecListThreeDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6320 def VLD3DUPqWB_fixed_Asm_8 :
6321 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr!",
6322 (ins VecListThreeQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6323 def VLD3DUPqWB_fixed_Asm_16 :
6324 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
6325 (ins VecListThreeQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6326 def VLD3DUPqWB_fixed_Asm_32 :
6327 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
6328 (ins VecListThreeQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6329 def VLD3DUPdWB_register_Asm_8 :
6330 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr, $Rm",
6331 (ins VecListThreeDAllLanes:$list, addrmode6:$addr,
6332 rGPR:$Rm, pred:$p)>;
6333 def VLD3DUPdWB_register_Asm_16 :
6334 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
6335 (ins VecListThreeDAllLanes:$list, addrmode6:$addr,
6336 rGPR:$Rm, pred:$p)>;
6337 def VLD3DUPdWB_register_Asm_32 :
6338 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
6339 (ins VecListThreeDAllLanes:$list, addrmode6:$addr,
6340 rGPR:$Rm, pred:$p)>;
6341 def VLD3DUPqWB_register_Asm_8 :
6342 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr, $Rm",
6343 (ins VecListThreeQAllLanes:$list, addrmode6:$addr,
6344 rGPR:$Rm, pred:$p)>;
6345 def VLD3DUPqWB_register_Asm_16 :
6346 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
6347 (ins VecListThreeQAllLanes:$list, addrmode6:$addr,
6348 rGPR:$Rm, pred:$p)>;
6349 def VLD3DUPqWB_register_Asm_32 :
6350 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
6351 (ins VecListThreeQAllLanes:$list, addrmode6:$addr,
6352 rGPR:$Rm, pred:$p)>;
6355 // VLD3 single-lane pseudo-instructions. These need special handling for
6356 // the lane index that an InstAlias can't handle, so we use these instead.
6357 def VLD3LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr",
6358 (ins VecListThreeDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6359 def VLD3LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
6360 (ins VecListThreeDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6361 def VLD3LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
6362 (ins VecListThreeDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6363 def VLD3LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
6364 (ins VecListThreeQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6365 def VLD3LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
6366 (ins VecListThreeQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6368 def VLD3LNdWB_fixed_Asm_8 :
6369 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr!",
6370 (ins VecListThreeDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6371 def VLD3LNdWB_fixed_Asm_16 :
6372 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
6373 (ins VecListThreeDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6374 def VLD3LNdWB_fixed_Asm_32 :
6375 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
6376 (ins VecListThreeDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6377 def VLD3LNqWB_fixed_Asm_16 :
6378 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
6379 (ins VecListThreeQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6380 def VLD3LNqWB_fixed_Asm_32 :
6381 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
6382 (ins VecListThreeQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6383 def VLD3LNdWB_register_Asm_8 :
6384 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr, $Rm",
6385 (ins VecListThreeDByteIndexed:$list, addrmode6:$addr,
6386 rGPR:$Rm, pred:$p)>;
6387 def VLD3LNdWB_register_Asm_16 :
6388 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
6389 (ins VecListThreeDHWordIndexed:$list, addrmode6:$addr,
6390 rGPR:$Rm, pred:$p)>;
6391 def VLD3LNdWB_register_Asm_32 :
6392 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
6393 (ins VecListThreeDWordIndexed:$list, addrmode6:$addr,
6394 rGPR:$Rm, pred:$p)>;
6395 def VLD3LNqWB_register_Asm_16 :
6396 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
6397 (ins VecListThreeQHWordIndexed:$list, addrmode6:$addr,
6398 rGPR:$Rm, pred:$p)>;
6399 def VLD3LNqWB_register_Asm_32 :
6400 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
6401 (ins VecListThreeQWordIndexed:$list, addrmode6:$addr,
6402 rGPR:$Rm, pred:$p)>;
6404 // VLD3 multiple structure pseudo-instructions. These need special handling for
6405 // the vector operands that the normal instructions don't yet model.
6406 // FIXME: Remove these when the register classes and instructions are updated.
6407 def VLD3dAsm_8 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr",
6408 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6409 def VLD3dAsm_16 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
6410 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6411 def VLD3dAsm_32 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
6412 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6413 def VLD3qAsm_8 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr",
6414 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6415 def VLD3qAsm_16 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
6416 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6417 def VLD3qAsm_32 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
6418 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6420 def VLD3dWB_fixed_Asm_8 :
6421 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr!",
6422 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6423 def VLD3dWB_fixed_Asm_16 :
6424 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
6425 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6426 def VLD3dWB_fixed_Asm_32 :
6427 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
6428 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6429 def VLD3qWB_fixed_Asm_8 :
6430 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr!",
6431 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6432 def VLD3qWB_fixed_Asm_16 :
6433 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
6434 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6435 def VLD3qWB_fixed_Asm_32 :
6436 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
6437 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6438 def VLD3dWB_register_Asm_8 :
6439 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr, $Rm",
6440 (ins VecListThreeD:$list, addrmode6:$addr,
6441 rGPR:$Rm, pred:$p)>;
6442 def VLD3dWB_register_Asm_16 :
6443 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
6444 (ins VecListThreeD:$list, addrmode6:$addr,
6445 rGPR:$Rm, pred:$p)>;
6446 def VLD3dWB_register_Asm_32 :
6447 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
6448 (ins VecListThreeD:$list, addrmode6:$addr,
6449 rGPR:$Rm, pred:$p)>;
6450 def VLD3qWB_register_Asm_8 :
6451 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr, $Rm",
6452 (ins VecListThreeQ:$list, addrmode6:$addr,
6453 rGPR:$Rm, pred:$p)>;
6454 def VLD3qWB_register_Asm_16 :
6455 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
6456 (ins VecListThreeQ:$list, addrmode6:$addr,
6457 rGPR:$Rm, pred:$p)>;
6458 def VLD3qWB_register_Asm_32 :
6459 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
6460 (ins VecListThreeQ:$list, addrmode6:$addr,
6461 rGPR:$Rm, pred:$p)>;
6463 // VST3 single-lane pseudo-instructions. These need special handling for
6464 // the lane index that an InstAlias can't handle, so we use these instead.
6465 def VST3LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr",
6466 (ins VecListThreeDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6467 def VST3LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr",
6468 (ins VecListThreeDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6469 def VST3LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr",
6470 (ins VecListThreeDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6471 def VST3LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr",
6472 (ins VecListThreeQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6473 def VST3LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr",
6474 (ins VecListThreeQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6476 def VST3LNdWB_fixed_Asm_8 :
6477 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr!",
6478 (ins VecListThreeDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6479 def VST3LNdWB_fixed_Asm_16 :
6480 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr!",
6481 (ins VecListThreeDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6482 def VST3LNdWB_fixed_Asm_32 :
6483 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr!",
6484 (ins VecListThreeDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6485 def VST3LNqWB_fixed_Asm_16 :
6486 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr!",
6487 (ins VecListThreeQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6488 def VST3LNqWB_fixed_Asm_32 :
6489 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr!",
6490 (ins VecListThreeQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6491 def VST3LNdWB_register_Asm_8 :
6492 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr, $Rm",
6493 (ins VecListThreeDByteIndexed:$list, addrmode6:$addr,
6494 rGPR:$Rm, pred:$p)>;
6495 def VST3LNdWB_register_Asm_16 :
6496 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr, $Rm",
6497 (ins VecListThreeDHWordIndexed:$list, addrmode6:$addr,
6498 rGPR:$Rm, pred:$p)>;
6499 def VST3LNdWB_register_Asm_32 :
6500 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr, $Rm",
6501 (ins VecListThreeDWordIndexed:$list, addrmode6:$addr,
6502 rGPR:$Rm, pred:$p)>;
6503 def VST3LNqWB_register_Asm_16 :
6504 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr, $Rm",
6505 (ins VecListThreeQHWordIndexed:$list, addrmode6:$addr,
6506 rGPR:$Rm, pred:$p)>;
6507 def VST3LNqWB_register_Asm_32 :
6508 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr, $Rm",
6509 (ins VecListThreeQWordIndexed:$list, addrmode6:$addr,
6510 rGPR:$Rm, pred:$p)>;
6513 // VST3 multiple structure pseudo-instructions. These need special handling for
6514 // the vector operands that the normal instructions don't yet model.
6515 // FIXME: Remove these when the register classes and instructions are updated.
6516 def VST3dAsm_8 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr",
6517 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6518 def VST3dAsm_16 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr",
6519 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6520 def VST3dAsm_32 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr",
6521 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6522 def VST3qAsm_8 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr",
6523 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6524 def VST3qAsm_16 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr",
6525 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6526 def VST3qAsm_32 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr",
6527 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6529 def VST3dWB_fixed_Asm_8 :
6530 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr!",
6531 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6532 def VST3dWB_fixed_Asm_16 :
6533 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr!",
6534 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6535 def VST3dWB_fixed_Asm_32 :
6536 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr!",
6537 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6538 def VST3qWB_fixed_Asm_8 :
6539 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr!",
6540 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6541 def VST3qWB_fixed_Asm_16 :
6542 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr!",
6543 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6544 def VST3qWB_fixed_Asm_32 :
6545 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr!",
6546 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6547 def VST3dWB_register_Asm_8 :
6548 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr, $Rm",
6549 (ins VecListThreeD:$list, addrmode6:$addr,
6550 rGPR:$Rm, pred:$p)>;
6551 def VST3dWB_register_Asm_16 :
6552 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr, $Rm",
6553 (ins VecListThreeD:$list, addrmode6:$addr,
6554 rGPR:$Rm, pred:$p)>;
6555 def VST3dWB_register_Asm_32 :
6556 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr, $Rm",
6557 (ins VecListThreeD:$list, addrmode6:$addr,
6558 rGPR:$Rm, pred:$p)>;
6559 def VST3qWB_register_Asm_8 :
6560 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr, $Rm",
6561 (ins VecListThreeQ:$list, addrmode6:$addr,
6562 rGPR:$Rm, pred:$p)>;
6563 def VST3qWB_register_Asm_16 :
6564 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr, $Rm",
6565 (ins VecListThreeQ:$list, addrmode6:$addr,
6566 rGPR:$Rm, pred:$p)>;
6567 def VST3qWB_register_Asm_32 :
6568 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr, $Rm",
6569 (ins VecListThreeQ:$list, addrmode6:$addr,
6570 rGPR:$Rm, pred:$p)>;
6572 // VLD4 all-lanes pseudo-instructions. These need special handling for
6573 // the lane index that an InstAlias can't handle, so we use these instead.
6574 def VLD4DUPdAsm_8 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr",
6575 (ins VecListFourDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6576 def VLD4DUPdAsm_16: NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
6577 (ins VecListFourDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6578 def VLD4DUPdAsm_32: NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
6579 (ins VecListFourDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6580 def VLD4DUPqAsm_8 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr",
6581 (ins VecListFourQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6582 def VLD4DUPqAsm_16: NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
6583 (ins VecListFourQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6584 def VLD4DUPqAsm_32: NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
6585 (ins VecListFourQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6587 def VLD4DUPdWB_fixed_Asm_8 :
6588 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr!",
6589 (ins VecListFourDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6590 def VLD4DUPdWB_fixed_Asm_16 :
6591 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
6592 (ins VecListFourDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6593 def VLD4DUPdWB_fixed_Asm_32 :
6594 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
6595 (ins VecListFourDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6596 def VLD4DUPqWB_fixed_Asm_8 :
6597 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr!",
6598 (ins VecListFourQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6599 def VLD4DUPqWB_fixed_Asm_16 :
6600 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
6601 (ins VecListFourQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6602 def VLD4DUPqWB_fixed_Asm_32 :
6603 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
6604 (ins VecListFourQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6605 def VLD4DUPdWB_register_Asm_8 :
6606 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr, $Rm",
6607 (ins VecListFourDAllLanes:$list, addrmode6:$addr,
6608 rGPR:$Rm, pred:$p)>;
6609 def VLD4DUPdWB_register_Asm_16 :
6610 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
6611 (ins VecListFourDAllLanes:$list, addrmode6:$addr,
6612 rGPR:$Rm, pred:$p)>;
6613 def VLD4DUPdWB_register_Asm_32 :
6614 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
6615 (ins VecListFourDAllLanes:$list, addrmode6:$addr,
6616 rGPR:$Rm, pred:$p)>;
6617 def VLD4DUPqWB_register_Asm_8 :
6618 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr, $Rm",
6619 (ins VecListFourQAllLanes:$list, addrmode6:$addr,
6620 rGPR:$Rm, pred:$p)>;
6621 def VLD4DUPqWB_register_Asm_16 :
6622 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
6623 (ins VecListFourQAllLanes:$list, addrmode6:$addr,
6624 rGPR:$Rm, pred:$p)>;
6625 def VLD4DUPqWB_register_Asm_32 :
6626 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
6627 (ins VecListFourQAllLanes:$list, addrmode6:$addr,
6628 rGPR:$Rm, pred:$p)>;
6631 // VLD4 single-lane pseudo-instructions. These need special handling for
6632 // the lane index that an InstAlias can't handle, so we use these instead.
6633 def VLD4LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr",
6634 (ins VecListFourDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6635 def VLD4LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
6636 (ins VecListFourDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6637 def VLD4LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
6638 (ins VecListFourDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6639 def VLD4LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
6640 (ins VecListFourQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6641 def VLD4LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
6642 (ins VecListFourQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6644 def VLD4LNdWB_fixed_Asm_8 :
6645 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr!",
6646 (ins VecListFourDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6647 def VLD4LNdWB_fixed_Asm_16 :
6648 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
6649 (ins VecListFourDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6650 def VLD4LNdWB_fixed_Asm_32 :
6651 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
6652 (ins VecListFourDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6653 def VLD4LNqWB_fixed_Asm_16 :
6654 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
6655 (ins VecListFourQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6656 def VLD4LNqWB_fixed_Asm_32 :
6657 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
6658 (ins VecListFourQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6659 def VLD4LNdWB_register_Asm_8 :
6660 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr, $Rm",
6661 (ins VecListFourDByteIndexed:$list, addrmode6:$addr,
6662 rGPR:$Rm, pred:$p)>;
6663 def VLD4LNdWB_register_Asm_16 :
6664 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
6665 (ins VecListFourDHWordIndexed:$list, addrmode6:$addr,
6666 rGPR:$Rm, pred:$p)>;
6667 def VLD4LNdWB_register_Asm_32 :
6668 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
6669 (ins VecListFourDWordIndexed:$list, addrmode6:$addr,
6670 rGPR:$Rm, pred:$p)>;
6671 def VLD4LNqWB_register_Asm_16 :
6672 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
6673 (ins VecListFourQHWordIndexed:$list, addrmode6:$addr,
6674 rGPR:$Rm, pred:$p)>;
6675 def VLD4LNqWB_register_Asm_32 :
6676 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
6677 (ins VecListFourQWordIndexed:$list, addrmode6:$addr,
6678 rGPR:$Rm, pred:$p)>;
6682 // VLD4 multiple structure pseudo-instructions. These need special handling for
6683 // the vector operands that the normal instructions don't yet model.
6684 // FIXME: Remove these when the register classes and instructions are updated.
6685 def VLD4dAsm_8 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr",
6686 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6687 def VLD4dAsm_16 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
6688 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6689 def VLD4dAsm_32 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
6690 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6691 def VLD4qAsm_8 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr",
6692 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6693 def VLD4qAsm_16 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
6694 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6695 def VLD4qAsm_32 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
6696 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6698 def VLD4dWB_fixed_Asm_8 :
6699 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr!",
6700 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6701 def VLD4dWB_fixed_Asm_16 :
6702 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
6703 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6704 def VLD4dWB_fixed_Asm_32 :
6705 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
6706 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6707 def VLD4qWB_fixed_Asm_8 :
6708 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr!",
6709 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6710 def VLD4qWB_fixed_Asm_16 :
6711 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
6712 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6713 def VLD4qWB_fixed_Asm_32 :
6714 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
6715 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6716 def VLD4dWB_register_Asm_8 :
6717 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr, $Rm",
6718 (ins VecListFourD:$list, addrmode6:$addr,
6719 rGPR:$Rm, pred:$p)>;
6720 def VLD4dWB_register_Asm_16 :
6721 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
6722 (ins VecListFourD:$list, addrmode6:$addr,
6723 rGPR:$Rm, pred:$p)>;
6724 def VLD4dWB_register_Asm_32 :
6725 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
6726 (ins VecListFourD:$list, addrmode6:$addr,
6727 rGPR:$Rm, pred:$p)>;
6728 def VLD4qWB_register_Asm_8 :
6729 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr, $Rm",
6730 (ins VecListFourQ:$list, addrmode6:$addr,
6731 rGPR:$Rm, pred:$p)>;
6732 def VLD4qWB_register_Asm_16 :
6733 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
6734 (ins VecListFourQ:$list, addrmode6:$addr,
6735 rGPR:$Rm, pred:$p)>;
6736 def VLD4qWB_register_Asm_32 :
6737 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
6738 (ins VecListFourQ:$list, addrmode6:$addr,
6739 rGPR:$Rm, pred:$p)>;
6741 // VST4 single-lane pseudo-instructions. These need special handling for
6742 // the lane index that an InstAlias can't handle, so we use these instead.
6743 def VST4LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr",
6744 (ins VecListFourDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6745 def VST4LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr",
6746 (ins VecListFourDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6747 def VST4LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr",
6748 (ins VecListFourDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6749 def VST4LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr",
6750 (ins VecListFourQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6751 def VST4LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr",
6752 (ins VecListFourQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6754 def VST4LNdWB_fixed_Asm_8 :
6755 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr!",
6756 (ins VecListFourDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6757 def VST4LNdWB_fixed_Asm_16 :
6758 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr!",
6759 (ins VecListFourDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6760 def VST4LNdWB_fixed_Asm_32 :
6761 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr!",
6762 (ins VecListFourDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6763 def VST4LNqWB_fixed_Asm_16 :
6764 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr!",
6765 (ins VecListFourQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6766 def VST4LNqWB_fixed_Asm_32 :
6767 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr!",
6768 (ins VecListFourQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6769 def VST4LNdWB_register_Asm_8 :
6770 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr, $Rm",
6771 (ins VecListFourDByteIndexed:$list, addrmode6:$addr,
6772 rGPR:$Rm, pred:$p)>;
6773 def VST4LNdWB_register_Asm_16 :
6774 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr, $Rm",
6775 (ins VecListFourDHWordIndexed:$list, addrmode6:$addr,
6776 rGPR:$Rm, pred:$p)>;
6777 def VST4LNdWB_register_Asm_32 :
6778 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr, $Rm",
6779 (ins VecListFourDWordIndexed:$list, addrmode6:$addr,
6780 rGPR:$Rm, pred:$p)>;
6781 def VST4LNqWB_register_Asm_16 :
6782 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr, $Rm",
6783 (ins VecListFourQHWordIndexed:$list, addrmode6:$addr,
6784 rGPR:$Rm, pred:$p)>;
6785 def VST4LNqWB_register_Asm_32 :
6786 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr, $Rm",
6787 (ins VecListFourQWordIndexed:$list, addrmode6:$addr,
6788 rGPR:$Rm, pred:$p)>;
6791 // VST4 multiple structure pseudo-instructions. These need special handling for
6792 // the vector operands that the normal instructions don't yet model.
6793 // FIXME: Remove these when the register classes and instructions are updated.
6794 def VST4dAsm_8 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr",
6795 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6796 def VST4dAsm_16 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr",
6797 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6798 def VST4dAsm_32 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr",
6799 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6800 def VST4qAsm_8 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr",
6801 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6802 def VST4qAsm_16 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr",
6803 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6804 def VST4qAsm_32 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr",
6805 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6807 def VST4dWB_fixed_Asm_8 :
6808 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr!",
6809 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6810 def VST4dWB_fixed_Asm_16 :
6811 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr!",
6812 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6813 def VST4dWB_fixed_Asm_32 :
6814 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr!",
6815 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6816 def VST4qWB_fixed_Asm_8 :
6817 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr!",
6818 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6819 def VST4qWB_fixed_Asm_16 :
6820 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr!",
6821 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6822 def VST4qWB_fixed_Asm_32 :
6823 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr!",
6824 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6825 def VST4dWB_register_Asm_8 :
6826 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr, $Rm",
6827 (ins VecListFourD:$list, addrmode6:$addr,
6828 rGPR:$Rm, pred:$p)>;
6829 def VST4dWB_register_Asm_16 :
6830 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr, $Rm",
6831 (ins VecListFourD:$list, addrmode6:$addr,
6832 rGPR:$Rm, pred:$p)>;
6833 def VST4dWB_register_Asm_32 :
6834 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr, $Rm",
6835 (ins VecListFourD:$list, addrmode6:$addr,
6836 rGPR:$Rm, pred:$p)>;
6837 def VST4qWB_register_Asm_8 :
6838 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr, $Rm",
6839 (ins VecListFourQ:$list, addrmode6:$addr,
6840 rGPR:$Rm, pred:$p)>;
6841 def VST4qWB_register_Asm_16 :
6842 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr, $Rm",
6843 (ins VecListFourQ:$list, addrmode6:$addr,
6844 rGPR:$Rm, pred:$p)>;
6845 def VST4qWB_register_Asm_32 :
6846 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr, $Rm",
6847 (ins VecListFourQ:$list, addrmode6:$addr,
6848 rGPR:$Rm, pred:$p)>;
6850 // VMOV/VMVN takes an optional datatype suffix
6851 defm : NEONDTAnyInstAlias<"vmov${p}", "$Vd, $Vm",
6852 (VORRd DPR:$Vd, DPR:$Vm, DPR:$Vm, pred:$p)>;
6853 defm : NEONDTAnyInstAlias<"vmov${p}", "$Vd, $Vm",
6854 (VORRq QPR:$Vd, QPR:$Vm, QPR:$Vm, pred:$p)>;
6856 defm : NEONDTAnyInstAlias<"vmvn${p}", "$Vd, $Vm",
6857 (VMVNd DPR:$Vd, DPR:$Vm, pred:$p)>;
6858 defm : NEONDTAnyInstAlias<"vmvn${p}", "$Vd, $Vm",
6859 (VMVNq QPR:$Vd, QPR:$Vm, pred:$p)>;
6861 // VCLT (register) is an assembler alias for VCGT w/ the operands reversed.
6862 // D-register versions.
6863 def : NEONInstAlias<"vcle${p}.s8 $Dd, $Dn, $Dm",
6864 (VCGEsv8i8 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6865 def : NEONInstAlias<"vcle${p}.s16 $Dd, $Dn, $Dm",
6866 (VCGEsv4i16 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6867 def : NEONInstAlias<"vcle${p}.s32 $Dd, $Dn, $Dm",
6868 (VCGEsv2i32 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6869 def : NEONInstAlias<"vcle${p}.u8 $Dd, $Dn, $Dm",
6870 (VCGEuv8i8 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6871 def : NEONInstAlias<"vcle${p}.u16 $Dd, $Dn, $Dm",
6872 (VCGEuv4i16 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6873 def : NEONInstAlias<"vcle${p}.u32 $Dd, $Dn, $Dm",
6874 (VCGEuv2i32 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6875 def : NEONInstAlias<"vcle${p}.f32 $Dd, $Dn, $Dm",
6876 (VCGEfd DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6877 // Q-register versions.
6878 def : NEONInstAlias<"vcle${p}.s8 $Qd, $Qn, $Qm",
6879 (VCGEsv16i8 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6880 def : NEONInstAlias<"vcle${p}.s16 $Qd, $Qn, $Qm",
6881 (VCGEsv8i16 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6882 def : NEONInstAlias<"vcle${p}.s32 $Qd, $Qn, $Qm",
6883 (VCGEsv4i32 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6884 def : NEONInstAlias<"vcle${p}.u8 $Qd, $Qn, $Qm",
6885 (VCGEuv16i8 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6886 def : NEONInstAlias<"vcle${p}.u16 $Qd, $Qn, $Qm",
6887 (VCGEuv8i16 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6888 def : NEONInstAlias<"vcle${p}.u32 $Qd, $Qn, $Qm",
6889 (VCGEuv4i32 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6890 def : NEONInstAlias<"vcle${p}.f32 $Qd, $Qn, $Qm",
6891 (VCGEfq QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6893 // VCLT (register) is an assembler alias for VCGT w/ the operands reversed.
6894 // D-register versions.
6895 def : NEONInstAlias<"vclt${p}.s8 $Dd, $Dn, $Dm",
6896 (VCGTsv8i8 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6897 def : NEONInstAlias<"vclt${p}.s16 $Dd, $Dn, $Dm",
6898 (VCGTsv4i16 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6899 def : NEONInstAlias<"vclt${p}.s32 $Dd, $Dn, $Dm",
6900 (VCGTsv2i32 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6901 def : NEONInstAlias<"vclt${p}.u8 $Dd, $Dn, $Dm",
6902 (VCGTuv8i8 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6903 def : NEONInstAlias<"vclt${p}.u16 $Dd, $Dn, $Dm",
6904 (VCGTuv4i16 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6905 def : NEONInstAlias<"vclt${p}.u32 $Dd, $Dn, $Dm",
6906 (VCGTuv2i32 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6907 def : NEONInstAlias<"vclt${p}.f32 $Dd, $Dn, $Dm",
6908 (VCGTfd DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6909 // Q-register versions.
6910 def : NEONInstAlias<"vclt${p}.s8 $Qd, $Qn, $Qm",
6911 (VCGTsv16i8 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6912 def : NEONInstAlias<"vclt${p}.s16 $Qd, $Qn, $Qm",
6913 (VCGTsv8i16 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6914 def : NEONInstAlias<"vclt${p}.s32 $Qd, $Qn, $Qm",
6915 (VCGTsv4i32 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6916 def : NEONInstAlias<"vclt${p}.u8 $Qd, $Qn, $Qm",
6917 (VCGTuv16i8 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6918 def : NEONInstAlias<"vclt${p}.u16 $Qd, $Qn, $Qm",
6919 (VCGTuv8i16 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6920 def : NEONInstAlias<"vclt${p}.u32 $Qd, $Qn, $Qm",
6921 (VCGTuv4i32 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6922 def : NEONInstAlias<"vclt${p}.f32 $Qd, $Qn, $Qm",
6923 (VCGTfq QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6925 // VSWP allows, but does not require, a type suffix.
6926 defm : NEONDTAnyInstAlias<"vswp${p}", "$Vd, $Vm",
6927 (VSWPd DPR:$Vd, DPR:$Vm, pred:$p)>;
6928 defm : NEONDTAnyInstAlias<"vswp${p}", "$Vd, $Vm",
6929 (VSWPq QPR:$Vd, QPR:$Vm, pred:$p)>;
6931 // VBIF, VBIT, and VBSL allow, but do not require, a type suffix.
6932 defm : NEONDTAnyInstAlias<"vbif${p}", "$Vd, $Vn, $Vm",
6933 (VBIFd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
6934 defm : NEONDTAnyInstAlias<"vbit${p}", "$Vd, $Vn, $Vm",
6935 (VBITd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
6936 defm : NEONDTAnyInstAlias<"vbsl${p}", "$Vd, $Vn, $Vm",
6937 (VBSLd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
6938 defm : NEONDTAnyInstAlias<"vbif${p}", "$Vd, $Vn, $Vm",
6939 (VBIFq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
6940 defm : NEONDTAnyInstAlias<"vbit${p}", "$Vd, $Vn, $Vm",
6941 (VBITq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
6942 defm : NEONDTAnyInstAlias<"vbsl${p}", "$Vd, $Vn, $Vm",
6943 (VBSLq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
6945 // "vmov Rd, #-imm" can be handled via "vmvn".
6946 def : NEONInstAlias<"vmov${p}.i32 $Vd, $imm",
6947 (VMVNv2i32 DPR:$Vd, nImmVMOVI32Neg:$imm, pred:$p)>;
6948 def : NEONInstAlias<"vmov${p}.i32 $Vd, $imm",
6949 (VMVNv4i32 QPR:$Vd, nImmVMOVI32Neg:$imm, pred:$p)>;
6950 def : NEONInstAlias<"vmvn${p}.i32 $Vd, $imm",
6951 (VMOVv2i32 DPR:$Vd, nImmVMOVI32Neg:$imm, pred:$p)>;
6952 def : NEONInstAlias<"vmvn${p}.i32 $Vd, $imm",
6953 (VMOVv4i32 QPR:$Vd, nImmVMOVI32Neg:$imm, pred:$p)>;
6955 // 'gas' compatibility aliases for quad-word instructions. Strictly speaking,
6956 // these should restrict to just the Q register variants, but the register
6957 // classes are enough to match correctly regardless, so we keep it simple
6958 // and just use MnemonicAlias.
6959 def : NEONMnemonicAlias<"vbicq", "vbic">;
6960 def : NEONMnemonicAlias<"vandq", "vand">;
6961 def : NEONMnemonicAlias<"veorq", "veor">;
6962 def : NEONMnemonicAlias<"vorrq", "vorr">;
6964 def : NEONMnemonicAlias<"vmovq", "vmov">;
6965 def : NEONMnemonicAlias<"vmvnq", "vmvn">;
6966 // Explicit versions for floating point so that the FPImm variants get
6967 // handled early. The parser gets confused otherwise.
6968 def : NEONMnemonicAlias<"vmovq.f32", "vmov.f32">;
6969 def : NEONMnemonicAlias<"vmovq.f64", "vmov.f64">;
6971 def : NEONMnemonicAlias<"vaddq", "vadd">;
6972 def : NEONMnemonicAlias<"vsubq", "vsub">;
6974 def : NEONMnemonicAlias<"vminq", "vmin">;
6975 def : NEONMnemonicAlias<"vmaxq", "vmax">;
6977 def : NEONMnemonicAlias<"vmulq", "vmul">;
6979 def : NEONMnemonicAlias<"vabsq", "vabs">;
6981 def : NEONMnemonicAlias<"vshlq", "vshl">;
6982 def : NEONMnemonicAlias<"vshrq", "vshr">;
6984 def : NEONMnemonicAlias<"vcvtq", "vcvt">;
6986 def : NEONMnemonicAlias<"vcleq", "vcle">;
6987 def : NEONMnemonicAlias<"vceqq", "vceq">;
6989 def : NEONMnemonicAlias<"vzipq", "vzip">;
6990 def : NEONMnemonicAlias<"vswpq", "vswp">;
6992 def : NEONMnemonicAlias<"vrecpeq.f32", "vrecpe.f32">;
6993 def : NEONMnemonicAlias<"vrecpeq.u32", "vrecpe.u32">;
6996 // Alias for loading floating point immediates that aren't representable
6997 // using the vmov.f32 encoding but the bitpattern is representable using
6998 // the .i32 encoding.
6999 def : NEONInstAlias<"vmov${p}.f32 $Vd, $imm",
7000 (VMOVv4i32 QPR:$Vd, nImmVMOVI32:$imm, pred:$p)>;
7001 def : NEONInstAlias<"vmov${p}.f32 $Vd, $imm",
7002 (VMOVv2i32 DPR:$Vd, nImmVMOVI32:$imm, pred:$p)>;