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 = $wb", []> {
718 let DecoderMethod = "DecodeVLDST1Instruction";
722 def VLD1d8T : VLD1D3<{0,0,0,?}, "8">;
723 def VLD1d16T : VLD1D3<{0,1,0,?}, "16">;
724 def VLD1d32T : VLD1D3<{1,0,0,?}, "32">;
725 def VLD1d64T : VLD1D3<{1,1,0,?}, "64">;
727 defm VLD1d8Twb : VLD1D3WB<{0,0,0,?}, "8">;
728 defm VLD1d16Twb : VLD1D3WB<{0,1,0,?}, "16">;
729 defm VLD1d32Twb : VLD1D3WB<{1,0,0,?}, "32">;
730 defm VLD1d64Twb : VLD1D3WB<{1,1,0,?}, "64">;
732 def VLD1d64TPseudo : VLDQQPseudo<IIC_VLD1x3>;
733 def VLD1d64TPseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD1x3>;
734 def VLD1d64TPseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD1x3>;
736 // ...with 4 registers
737 class VLD1D4<bits<4> op7_4, string Dt>
738 : NLdSt<0, 0b10, 0b0010, op7_4, (outs VecListFourD:$Vd),
739 (ins addrmode6:$Rn), IIC_VLD1x4, "vld1", Dt,
740 "$Vd, $Rn", "", []> {
742 let Inst{5-4} = Rn{5-4};
743 let DecoderMethod = "DecodeVLDST1Instruction";
745 multiclass VLD1D4WB<bits<4> op7_4, string Dt> {
746 def _fixed : NLdSt<0,0b10,0b0010, op7_4, (outs VecListFourD:$Vd, GPR:$wb),
747 (ins addrmode6:$Rn), IIC_VLD1x2u,
748 "vld1", Dt, "$Vd, $Rn!",
749 "$Rn.addr = $wb", []> {
750 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
751 let Inst{5-4} = Rn{5-4};
752 let DecoderMethod = "DecodeVLDST1Instruction";
754 def _register : NLdSt<0,0b10,0b0010,op7_4, (outs VecListFourD:$Vd, GPR:$wb),
755 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1x2u,
756 "vld1", Dt, "$Vd, $Rn, $Rm",
757 "$Rn.addr = $wb", []> {
758 let Inst{5-4} = Rn{5-4};
759 let DecoderMethod = "DecodeVLDST1Instruction";
763 def VLD1d8Q : VLD1D4<{0,0,?,?}, "8">;
764 def VLD1d16Q : VLD1D4<{0,1,?,?}, "16">;
765 def VLD1d32Q : VLD1D4<{1,0,?,?}, "32">;
766 def VLD1d64Q : VLD1D4<{1,1,?,?}, "64">;
768 defm VLD1d8Qwb : VLD1D4WB<{0,0,?,?}, "8">;
769 defm VLD1d16Qwb : VLD1D4WB<{0,1,?,?}, "16">;
770 defm VLD1d32Qwb : VLD1D4WB<{1,0,?,?}, "32">;
771 defm VLD1d64Qwb : VLD1D4WB<{1,1,?,?}, "64">;
773 def VLD1d64QPseudo : VLDQQPseudo<IIC_VLD1x4>;
774 def VLD1d64QPseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD1x4>;
775 def VLD1d64QPseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD1x4>;
777 // VLD2 : Vector Load (multiple 2-element structures)
778 class VLD2<bits<4> op11_8, bits<4> op7_4, string Dt, RegisterOperand VdTy,
780 : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd),
781 (ins addrmode6:$Rn), itin,
782 "vld2", Dt, "$Vd, $Rn", "", []> {
784 let Inst{5-4} = Rn{5-4};
785 let DecoderMethod = "DecodeVLDST2Instruction";
788 def VLD2d8 : VLD2<0b1000, {0,0,?,?}, "8", VecListDPair, IIC_VLD2>;
789 def VLD2d16 : VLD2<0b1000, {0,1,?,?}, "16", VecListDPair, IIC_VLD2>;
790 def VLD2d32 : VLD2<0b1000, {1,0,?,?}, "32", VecListDPair, IIC_VLD2>;
792 def VLD2q8 : VLD2<0b0011, {0,0,?,?}, "8", VecListFourD, IIC_VLD2x2>;
793 def VLD2q16 : VLD2<0b0011, {0,1,?,?}, "16", VecListFourD, IIC_VLD2x2>;
794 def VLD2q32 : VLD2<0b0011, {1,0,?,?}, "32", VecListFourD, IIC_VLD2x2>;
796 def VLD2q8Pseudo : VLDQQPseudo<IIC_VLD2x2>;
797 def VLD2q16Pseudo : VLDQQPseudo<IIC_VLD2x2>;
798 def VLD2q32Pseudo : VLDQQPseudo<IIC_VLD2x2>;
800 // ...with address register writeback:
801 multiclass VLD2WB<bits<4> op11_8, bits<4> op7_4, string Dt,
802 RegisterOperand VdTy, InstrItinClass itin> {
803 def _fixed : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd, GPR:$wb),
804 (ins addrmode6:$Rn), itin,
805 "vld2", Dt, "$Vd, $Rn!",
806 "$Rn.addr = $wb", []> {
807 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
808 let Inst{5-4} = Rn{5-4};
809 let DecoderMethod = "DecodeVLDST2Instruction";
811 def _register : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd, GPR:$wb),
812 (ins addrmode6:$Rn, rGPR:$Rm), itin,
813 "vld2", Dt, "$Vd, $Rn, $Rm",
814 "$Rn.addr = $wb", []> {
815 let Inst{5-4} = Rn{5-4};
816 let DecoderMethod = "DecodeVLDST2Instruction";
820 defm VLD2d8wb : VLD2WB<0b1000, {0,0,?,?}, "8", VecListDPair, IIC_VLD2u>;
821 defm VLD2d16wb : VLD2WB<0b1000, {0,1,?,?}, "16", VecListDPair, IIC_VLD2u>;
822 defm VLD2d32wb : VLD2WB<0b1000, {1,0,?,?}, "32", VecListDPair, IIC_VLD2u>;
824 defm VLD2q8wb : VLD2WB<0b0011, {0,0,?,?}, "8", VecListFourD, IIC_VLD2x2u>;
825 defm VLD2q16wb : VLD2WB<0b0011, {0,1,?,?}, "16", VecListFourD, IIC_VLD2x2u>;
826 defm VLD2q32wb : VLD2WB<0b0011, {1,0,?,?}, "32", VecListFourD, IIC_VLD2x2u>;
828 def VLD2q8PseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD2x2u>;
829 def VLD2q16PseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD2x2u>;
830 def VLD2q32PseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD2x2u>;
831 def VLD2q8PseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD2x2u>;
832 def VLD2q16PseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD2x2u>;
833 def VLD2q32PseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD2x2u>;
835 // ...with double-spaced registers
836 def VLD2b8 : VLD2<0b1001, {0,0,?,?}, "8", VecListDPairSpaced, IIC_VLD2>;
837 def VLD2b16 : VLD2<0b1001, {0,1,?,?}, "16", VecListDPairSpaced, IIC_VLD2>;
838 def VLD2b32 : VLD2<0b1001, {1,0,?,?}, "32", VecListDPairSpaced, IIC_VLD2>;
839 defm VLD2b8wb : VLD2WB<0b1001, {0,0,?,?}, "8", VecListDPairSpaced, IIC_VLD2u>;
840 defm VLD2b16wb : VLD2WB<0b1001, {0,1,?,?}, "16", VecListDPairSpaced, IIC_VLD2u>;
841 defm VLD2b32wb : VLD2WB<0b1001, {1,0,?,?}, "32", VecListDPairSpaced, IIC_VLD2u>;
843 // VLD3 : Vector Load (multiple 3-element structures)
844 class VLD3D<bits<4> op11_8, bits<4> op7_4, string Dt>
845 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
846 (ins addrmode6:$Rn), IIC_VLD3,
847 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn", "", []> {
850 let DecoderMethod = "DecodeVLDST3Instruction";
853 def VLD3d8 : VLD3D<0b0100, {0,0,0,?}, "8">;
854 def VLD3d16 : VLD3D<0b0100, {0,1,0,?}, "16">;
855 def VLD3d32 : VLD3D<0b0100, {1,0,0,?}, "32">;
857 def VLD3d8Pseudo : VLDQQPseudo<IIC_VLD3>;
858 def VLD3d16Pseudo : VLDQQPseudo<IIC_VLD3>;
859 def VLD3d32Pseudo : VLDQQPseudo<IIC_VLD3>;
861 // ...with address register writeback:
862 class VLD3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
863 : NLdSt<0, 0b10, op11_8, op7_4,
864 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
865 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD3u,
866 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn$Rm",
867 "$Rn.addr = $wb", []> {
869 let DecoderMethod = "DecodeVLDST3Instruction";
872 def VLD3d8_UPD : VLD3DWB<0b0100, {0,0,0,?}, "8">;
873 def VLD3d16_UPD : VLD3DWB<0b0100, {0,1,0,?}, "16">;
874 def VLD3d32_UPD : VLD3DWB<0b0100, {1,0,0,?}, "32">;
876 def VLD3d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
877 def VLD3d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
878 def VLD3d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
880 // ...with double-spaced registers:
881 def VLD3q8 : VLD3D<0b0101, {0,0,0,?}, "8">;
882 def VLD3q16 : VLD3D<0b0101, {0,1,0,?}, "16">;
883 def VLD3q32 : VLD3D<0b0101, {1,0,0,?}, "32">;
884 def VLD3q8_UPD : VLD3DWB<0b0101, {0,0,0,?}, "8">;
885 def VLD3q16_UPD : VLD3DWB<0b0101, {0,1,0,?}, "16">;
886 def VLD3q32_UPD : VLD3DWB<0b0101, {1,0,0,?}, "32">;
888 def VLD3q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
889 def VLD3q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
890 def VLD3q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
892 // ...alternate versions to be allocated odd register numbers:
893 def VLD3q8oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
894 def VLD3q16oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
895 def VLD3q32oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
897 def VLD3q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
898 def VLD3q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
899 def VLD3q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
901 // VLD4 : Vector Load (multiple 4-element structures)
902 class VLD4D<bits<4> op11_8, bits<4> op7_4, string Dt>
903 : NLdSt<0, 0b10, op11_8, op7_4,
904 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
905 (ins addrmode6:$Rn), IIC_VLD4,
906 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> {
908 let Inst{5-4} = Rn{5-4};
909 let DecoderMethod = "DecodeVLDST4Instruction";
912 def VLD4d8 : VLD4D<0b0000, {0,0,?,?}, "8">;
913 def VLD4d16 : VLD4D<0b0000, {0,1,?,?}, "16">;
914 def VLD4d32 : VLD4D<0b0000, {1,0,?,?}, "32">;
916 def VLD4d8Pseudo : VLDQQPseudo<IIC_VLD4>;
917 def VLD4d16Pseudo : VLDQQPseudo<IIC_VLD4>;
918 def VLD4d32Pseudo : VLDQQPseudo<IIC_VLD4>;
920 // ...with address register writeback:
921 class VLD4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
922 : NLdSt<0, 0b10, op11_8, op7_4,
923 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
924 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD4u,
925 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm",
926 "$Rn.addr = $wb", []> {
927 let Inst{5-4} = Rn{5-4};
928 let DecoderMethod = "DecodeVLDST4Instruction";
931 def VLD4d8_UPD : VLD4DWB<0b0000, {0,0,?,?}, "8">;
932 def VLD4d16_UPD : VLD4DWB<0b0000, {0,1,?,?}, "16">;
933 def VLD4d32_UPD : VLD4DWB<0b0000, {1,0,?,?}, "32">;
935 def VLD4d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
936 def VLD4d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
937 def VLD4d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
939 // ...with double-spaced registers:
940 def VLD4q8 : VLD4D<0b0001, {0,0,?,?}, "8">;
941 def VLD4q16 : VLD4D<0b0001, {0,1,?,?}, "16">;
942 def VLD4q32 : VLD4D<0b0001, {1,0,?,?}, "32">;
943 def VLD4q8_UPD : VLD4DWB<0b0001, {0,0,?,?}, "8">;
944 def VLD4q16_UPD : VLD4DWB<0b0001, {0,1,?,?}, "16">;
945 def VLD4q32_UPD : VLD4DWB<0b0001, {1,0,?,?}, "32">;
947 def VLD4q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
948 def VLD4q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
949 def VLD4q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
951 // ...alternate versions to be allocated odd register numbers:
952 def VLD4q8oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
953 def VLD4q16oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
954 def VLD4q32oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
956 def VLD4q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
957 def VLD4q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
958 def VLD4q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
960 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
962 // Classes for VLD*LN pseudo-instructions with multi-register operands.
963 // These are expanded to real instructions after register allocation.
964 class VLDQLNPseudo<InstrItinClass itin>
965 : PseudoNLdSt<(outs QPR:$dst),
966 (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
967 itin, "$src = $dst">;
968 class VLDQLNWBPseudo<InstrItinClass itin>
969 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
970 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
971 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
972 class VLDQQLNPseudo<InstrItinClass itin>
973 : PseudoNLdSt<(outs QQPR:$dst),
974 (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
975 itin, "$src = $dst">;
976 class VLDQQLNWBPseudo<InstrItinClass itin>
977 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
978 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
979 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
980 class VLDQQQQLNPseudo<InstrItinClass itin>
981 : PseudoNLdSt<(outs QQQQPR:$dst),
982 (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
983 itin, "$src = $dst">;
984 class VLDQQQQLNWBPseudo<InstrItinClass itin>
985 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
986 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
987 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
989 // VLD1LN : Vector Load (single element to one lane)
990 class VLD1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
992 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
993 (ins addrmode6:$Rn, DPR:$src, nohash_imm:$lane),
994 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
996 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
997 (i32 (LoadOp addrmode6:$Rn)),
1000 let DecoderMethod = "DecodeVLD1LN";
1002 class VLD1LN32<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1004 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
1005 (ins addrmode6oneL32:$Rn, DPR:$src, nohash_imm:$lane),
1006 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
1008 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
1009 (i32 (LoadOp addrmode6oneL32:$Rn)),
1012 let DecoderMethod = "DecodeVLD1LN";
1014 class VLD1QLNPseudo<ValueType Ty, PatFrag LoadOp> : VLDQLNPseudo<IIC_VLD1ln> {
1015 let Pattern = [(set QPR:$dst, (vector_insert (Ty QPR:$src),
1016 (i32 (LoadOp addrmode6:$addr)),
1020 def VLD1LNd8 : VLD1LN<0b0000, {?,?,?,0}, "8", v8i8, extloadi8> {
1021 let Inst{7-5} = lane{2-0};
1023 def VLD1LNd16 : VLD1LN<0b0100, {?,?,0,?}, "16", v4i16, extloadi16> {
1024 let Inst{7-6} = lane{1-0};
1025 let Inst{5-4} = Rn{5-4};
1027 def VLD1LNd32 : VLD1LN32<0b1000, {?,0,?,?}, "32", v2i32, load> {
1028 let Inst{7} = lane{0};
1029 let Inst{5-4} = Rn{5-4};
1032 def VLD1LNq8Pseudo : VLD1QLNPseudo<v16i8, extloadi8>;
1033 def VLD1LNq16Pseudo : VLD1QLNPseudo<v8i16, extloadi16>;
1034 def VLD1LNq32Pseudo : VLD1QLNPseudo<v4i32, load>;
1036 def : Pat<(vector_insert (v2f32 DPR:$src),
1037 (f32 (load addrmode6:$addr)), imm:$lane),
1038 (VLD1LNd32 addrmode6:$addr, DPR:$src, imm:$lane)>;
1039 def : Pat<(vector_insert (v4f32 QPR:$src),
1040 (f32 (load addrmode6:$addr)), imm:$lane),
1041 (VLD1LNq32Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
1043 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
1045 // ...with address register writeback:
1046 class VLD1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1047 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, GPR:$wb),
1048 (ins addrmode6:$Rn, am6offset:$Rm,
1049 DPR:$src, nohash_imm:$lane), IIC_VLD1lnu, "vld1", Dt,
1050 "\\{$Vd[$lane]\\}, $Rn$Rm",
1051 "$src = $Vd, $Rn.addr = $wb", []> {
1052 let DecoderMethod = "DecodeVLD1LN";
1055 def VLD1LNd8_UPD : VLD1LNWB<0b0000, {?,?,?,0}, "8"> {
1056 let Inst{7-5} = lane{2-0};
1058 def VLD1LNd16_UPD : VLD1LNWB<0b0100, {?,?,0,?}, "16"> {
1059 let Inst{7-6} = lane{1-0};
1060 let Inst{4} = Rn{4};
1062 def VLD1LNd32_UPD : VLD1LNWB<0b1000, {?,0,?,?}, "32"> {
1063 let Inst{7} = lane{0};
1064 let Inst{5} = Rn{4};
1065 let Inst{4} = Rn{4};
1068 def VLD1LNq8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
1069 def VLD1LNq16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
1070 def VLD1LNq32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
1072 // VLD2LN : Vector Load (single 2-element structure to one lane)
1073 class VLD2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1074 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2),
1075 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, nohash_imm:$lane),
1076 IIC_VLD2ln, "vld2", Dt, "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn",
1077 "$src1 = $Vd, $src2 = $dst2", []> {
1079 let Inst{4} = Rn{4};
1080 let DecoderMethod = "DecodeVLD2LN";
1083 def VLD2LNd8 : VLD2LN<0b0001, {?,?,?,?}, "8"> {
1084 let Inst{7-5} = lane{2-0};
1086 def VLD2LNd16 : VLD2LN<0b0101, {?,?,0,?}, "16"> {
1087 let Inst{7-6} = lane{1-0};
1089 def VLD2LNd32 : VLD2LN<0b1001, {?,0,0,?}, "32"> {
1090 let Inst{7} = lane{0};
1093 def VLD2LNd8Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
1094 def VLD2LNd16Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
1095 def VLD2LNd32Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
1097 // ...with double-spaced registers:
1098 def VLD2LNq16 : VLD2LN<0b0101, {?,?,1,?}, "16"> {
1099 let Inst{7-6} = lane{1-0};
1101 def VLD2LNq32 : VLD2LN<0b1001, {?,1,0,?}, "32"> {
1102 let Inst{7} = lane{0};
1105 def VLD2LNq16Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
1106 def VLD2LNq32Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
1108 // ...with address register writeback:
1109 class VLD2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1110 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
1111 (ins addrmode6:$Rn, am6offset:$Rm,
1112 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VLD2lnu, "vld2", Dt,
1113 "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn$Rm",
1114 "$src1 = $Vd, $src2 = $dst2, $Rn.addr = $wb", []> {
1115 let Inst{4} = Rn{4};
1116 let DecoderMethod = "DecodeVLD2LN";
1119 def VLD2LNd8_UPD : VLD2LNWB<0b0001, {?,?,?,?}, "8"> {
1120 let Inst{7-5} = lane{2-0};
1122 def VLD2LNd16_UPD : VLD2LNWB<0b0101, {?,?,0,?}, "16"> {
1123 let Inst{7-6} = lane{1-0};
1125 def VLD2LNd32_UPD : VLD2LNWB<0b1001, {?,0,0,?}, "32"> {
1126 let Inst{7} = lane{0};
1129 def VLD2LNd8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
1130 def VLD2LNd16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
1131 def VLD2LNd32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
1133 def VLD2LNq16_UPD : VLD2LNWB<0b0101, {?,?,1,?}, "16"> {
1134 let Inst{7-6} = lane{1-0};
1136 def VLD2LNq32_UPD : VLD2LNWB<0b1001, {?,1,0,?}, "32"> {
1137 let Inst{7} = lane{0};
1140 def VLD2LNq16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
1141 def VLD2LNq32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
1143 // VLD3LN : Vector Load (single 3-element structure to one lane)
1144 class VLD3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1145 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
1146 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3,
1147 nohash_imm:$lane), IIC_VLD3ln, "vld3", Dt,
1148 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn",
1149 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3", []> {
1151 let DecoderMethod = "DecodeVLD3LN";
1154 def VLD3LNd8 : VLD3LN<0b0010, {?,?,?,0}, "8"> {
1155 let Inst{7-5} = lane{2-0};
1157 def VLD3LNd16 : VLD3LN<0b0110, {?,?,0,0}, "16"> {
1158 let Inst{7-6} = lane{1-0};
1160 def VLD3LNd32 : VLD3LN<0b1010, {?,0,0,0}, "32"> {
1161 let Inst{7} = lane{0};
1164 def VLD3LNd8Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
1165 def VLD3LNd16Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
1166 def VLD3LNd32Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
1168 // ...with double-spaced registers:
1169 def VLD3LNq16 : VLD3LN<0b0110, {?,?,1,0}, "16"> {
1170 let Inst{7-6} = lane{1-0};
1172 def VLD3LNq32 : VLD3LN<0b1010, {?,1,0,0}, "32"> {
1173 let Inst{7} = lane{0};
1176 def VLD3LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
1177 def VLD3LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
1179 // ...with address register writeback:
1180 class VLD3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1181 : NLdStLn<1, 0b10, op11_8, op7_4,
1182 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
1183 (ins addrmode6:$Rn, am6offset:$Rm,
1184 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
1185 IIC_VLD3lnu, "vld3", Dt,
1186 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn$Rm",
1187 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $Rn.addr = $wb",
1189 let DecoderMethod = "DecodeVLD3LN";
1192 def VLD3LNd8_UPD : VLD3LNWB<0b0010, {?,?,?,0}, "8"> {
1193 let Inst{7-5} = lane{2-0};
1195 def VLD3LNd16_UPD : VLD3LNWB<0b0110, {?,?,0,0}, "16"> {
1196 let Inst{7-6} = lane{1-0};
1198 def VLD3LNd32_UPD : VLD3LNWB<0b1010, {?,0,0,0}, "32"> {
1199 let Inst{7} = lane{0};
1202 def VLD3LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
1203 def VLD3LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
1204 def VLD3LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
1206 def VLD3LNq16_UPD : VLD3LNWB<0b0110, {?,?,1,0}, "16"> {
1207 let Inst{7-6} = lane{1-0};
1209 def VLD3LNq32_UPD : VLD3LNWB<0b1010, {?,1,0,0}, "32"> {
1210 let Inst{7} = lane{0};
1213 def VLD3LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
1214 def VLD3LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
1216 // VLD4LN : Vector Load (single 4-element structure to one lane)
1217 class VLD4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1218 : NLdStLn<1, 0b10, op11_8, op7_4,
1219 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
1220 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
1221 nohash_imm:$lane), IIC_VLD4ln, "vld4", Dt,
1222 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn",
1223 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4", []> {
1225 let Inst{4} = Rn{4};
1226 let DecoderMethod = "DecodeVLD4LN";
1229 def VLD4LNd8 : VLD4LN<0b0011, {?,?,?,?}, "8"> {
1230 let Inst{7-5} = lane{2-0};
1232 def VLD4LNd16 : VLD4LN<0b0111, {?,?,0,?}, "16"> {
1233 let Inst{7-6} = lane{1-0};
1235 def VLD4LNd32 : VLD4LN<0b1011, {?,0,?,?}, "32"> {
1236 let Inst{7} = lane{0};
1237 let Inst{5} = Rn{5};
1240 def VLD4LNd8Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
1241 def VLD4LNd16Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
1242 def VLD4LNd32Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
1244 // ...with double-spaced registers:
1245 def VLD4LNq16 : VLD4LN<0b0111, {?,?,1,?}, "16"> {
1246 let Inst{7-6} = lane{1-0};
1248 def VLD4LNq32 : VLD4LN<0b1011, {?,1,?,?}, "32"> {
1249 let Inst{7} = lane{0};
1250 let Inst{5} = Rn{5};
1253 def VLD4LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
1254 def VLD4LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
1256 // ...with address register writeback:
1257 class VLD4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1258 : NLdStLn<1, 0b10, op11_8, op7_4,
1259 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
1260 (ins addrmode6:$Rn, am6offset:$Rm,
1261 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
1262 IIC_VLD4lnu, "vld4", Dt,
1263 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn$Rm",
1264 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4, $Rn.addr = $wb",
1266 let Inst{4} = Rn{4};
1267 let DecoderMethod = "DecodeVLD4LN" ;
1270 def VLD4LNd8_UPD : VLD4LNWB<0b0011, {?,?,?,?}, "8"> {
1271 let Inst{7-5} = lane{2-0};
1273 def VLD4LNd16_UPD : VLD4LNWB<0b0111, {?,?,0,?}, "16"> {
1274 let Inst{7-6} = lane{1-0};
1276 def VLD4LNd32_UPD : VLD4LNWB<0b1011, {?,0,?,?}, "32"> {
1277 let Inst{7} = lane{0};
1278 let Inst{5} = Rn{5};
1281 def VLD4LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
1282 def VLD4LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
1283 def VLD4LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
1285 def VLD4LNq16_UPD : VLD4LNWB<0b0111, {?,?,1,?}, "16"> {
1286 let Inst{7-6} = lane{1-0};
1288 def VLD4LNq32_UPD : VLD4LNWB<0b1011, {?,1,?,?}, "32"> {
1289 let Inst{7} = lane{0};
1290 let Inst{5} = Rn{5};
1293 def VLD4LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
1294 def VLD4LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
1296 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
1298 // VLD1DUP : Vector Load (single element to all lanes)
1299 class VLD1DUP<bits<4> op7_4, string Dt, ValueType Ty, PatFrag LoadOp>
1300 : NLdSt<1, 0b10, 0b1100, op7_4, (outs VecListOneDAllLanes:$Vd),
1301 (ins addrmode6dup:$Rn),
1302 IIC_VLD1dup, "vld1", Dt, "$Vd, $Rn", "",
1303 [(set VecListOneDAllLanes:$Vd,
1304 (Ty (NEONvdup (i32 (LoadOp addrmode6dup:$Rn)))))]> {
1306 let Inst{4} = Rn{4};
1307 let DecoderMethod = "DecodeVLD1DupInstruction";
1309 def VLD1DUPd8 : VLD1DUP<{0,0,0,?}, "8", v8i8, extloadi8>;
1310 def VLD1DUPd16 : VLD1DUP<{0,1,0,?}, "16", v4i16, extloadi16>;
1311 def VLD1DUPd32 : VLD1DUP<{1,0,0,?}, "32", v2i32, load>;
1313 def : Pat<(v2f32 (NEONvdup (f32 (load addrmode6dup:$addr)))),
1314 (VLD1DUPd32 addrmode6:$addr)>;
1316 class VLD1QDUP<bits<4> op7_4, string Dt, ValueType Ty, PatFrag LoadOp>
1317 : NLdSt<1, 0b10, 0b1100, op7_4, (outs VecListDPairAllLanes:$Vd),
1318 (ins addrmode6dup:$Rn), IIC_VLD1dup,
1319 "vld1", Dt, "$Vd, $Rn", "",
1320 [(set VecListDPairAllLanes:$Vd,
1321 (Ty (NEONvdup (i32 (LoadOp addrmode6dup:$Rn)))))]> {
1323 let Inst{4} = Rn{4};
1324 let DecoderMethod = "DecodeVLD1DupInstruction";
1327 def VLD1DUPq8 : VLD1QDUP<{0,0,1,0}, "8", v16i8, extloadi8>;
1328 def VLD1DUPq16 : VLD1QDUP<{0,1,1,?}, "16", v8i16, extloadi16>;
1329 def VLD1DUPq32 : VLD1QDUP<{1,0,1,?}, "32", v4i32, load>;
1331 def : Pat<(v4f32 (NEONvdup (f32 (load addrmode6dup:$addr)))),
1332 (VLD1DUPq32 addrmode6:$addr)>;
1334 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
1335 // ...with address register writeback:
1336 multiclass VLD1DUPWB<bits<4> op7_4, string Dt> {
1337 def _fixed : NLdSt<1, 0b10, 0b1100, op7_4,
1338 (outs VecListOneDAllLanes:$Vd, GPR:$wb),
1339 (ins addrmode6dup:$Rn), IIC_VLD1dupu,
1340 "vld1", Dt, "$Vd, $Rn!",
1341 "$Rn.addr = $wb", []> {
1342 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1343 let Inst{4} = Rn{4};
1344 let DecoderMethod = "DecodeVLD1DupInstruction";
1346 def _register : NLdSt<1, 0b10, 0b1100, op7_4,
1347 (outs VecListOneDAllLanes:$Vd, GPR:$wb),
1348 (ins addrmode6dup:$Rn, rGPR:$Rm), IIC_VLD1dupu,
1349 "vld1", Dt, "$Vd, $Rn, $Rm",
1350 "$Rn.addr = $wb", []> {
1351 let Inst{4} = Rn{4};
1352 let DecoderMethod = "DecodeVLD1DupInstruction";
1355 multiclass VLD1QDUPWB<bits<4> op7_4, string Dt> {
1356 def _fixed : NLdSt<1, 0b10, 0b1100, op7_4,
1357 (outs VecListDPairAllLanes:$Vd, GPR:$wb),
1358 (ins addrmode6dup:$Rn), IIC_VLD1dupu,
1359 "vld1", Dt, "$Vd, $Rn!",
1360 "$Rn.addr = $wb", []> {
1361 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1362 let Inst{4} = Rn{4};
1363 let DecoderMethod = "DecodeVLD1DupInstruction";
1365 def _register : NLdSt<1, 0b10, 0b1100, op7_4,
1366 (outs VecListDPairAllLanes:$Vd, GPR:$wb),
1367 (ins addrmode6dup:$Rn, rGPR:$Rm), IIC_VLD1dupu,
1368 "vld1", Dt, "$Vd, $Rn, $Rm",
1369 "$Rn.addr = $wb", []> {
1370 let Inst{4} = Rn{4};
1371 let DecoderMethod = "DecodeVLD1DupInstruction";
1375 defm VLD1DUPd8wb : VLD1DUPWB<{0,0,0,0}, "8">;
1376 defm VLD1DUPd16wb : VLD1DUPWB<{0,1,0,?}, "16">;
1377 defm VLD1DUPd32wb : VLD1DUPWB<{1,0,0,?}, "32">;
1379 defm VLD1DUPq8wb : VLD1QDUPWB<{0,0,1,0}, "8">;
1380 defm VLD1DUPq16wb : VLD1QDUPWB<{0,1,1,?}, "16">;
1381 defm VLD1DUPq32wb : VLD1QDUPWB<{1,0,1,?}, "32">;
1383 // VLD2DUP : Vector Load (single 2-element structure to all lanes)
1384 class VLD2DUP<bits<4> op7_4, string Dt, RegisterOperand VdTy>
1385 : NLdSt<1, 0b10, 0b1101, op7_4, (outs VdTy:$Vd),
1386 (ins addrmode6dup:$Rn), IIC_VLD2dup,
1387 "vld2", Dt, "$Vd, $Rn", "", []> {
1389 let Inst{4} = Rn{4};
1390 let DecoderMethod = "DecodeVLD2DupInstruction";
1393 def VLD2DUPd8 : VLD2DUP<{0,0,0,?}, "8", VecListDPairAllLanes>;
1394 def VLD2DUPd16 : VLD2DUP<{0,1,0,?}, "16", VecListDPairAllLanes>;
1395 def VLD2DUPd32 : VLD2DUP<{1,0,0,?}, "32", VecListDPairAllLanes>;
1397 // ...with double-spaced registers
1398 def VLD2DUPd8x2 : VLD2DUP<{0,0,1,?}, "8", VecListDPairSpacedAllLanes>;
1399 def VLD2DUPd16x2 : VLD2DUP<{0,1,1,?}, "16", VecListDPairSpacedAllLanes>;
1400 def VLD2DUPd32x2 : VLD2DUP<{1,0,1,?}, "32", VecListDPairSpacedAllLanes>;
1402 // ...with address register writeback:
1403 multiclass VLD2DUPWB<bits<4> op7_4, string Dt, RegisterOperand VdTy> {
1404 def _fixed : NLdSt<1, 0b10, 0b1101, op7_4,
1405 (outs VdTy:$Vd, GPR:$wb),
1406 (ins addrmode6dup:$Rn), IIC_VLD2dupu,
1407 "vld2", Dt, "$Vd, $Rn!",
1408 "$Rn.addr = $wb", []> {
1409 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1410 let Inst{4} = Rn{4};
1411 let DecoderMethod = "DecodeVLD2DupInstruction";
1413 def _register : NLdSt<1, 0b10, 0b1101, op7_4,
1414 (outs VdTy:$Vd, GPR:$wb),
1415 (ins addrmode6dup:$Rn, rGPR:$Rm), IIC_VLD2dupu,
1416 "vld2", Dt, "$Vd, $Rn, $Rm",
1417 "$Rn.addr = $wb", []> {
1418 let Inst{4} = Rn{4};
1419 let DecoderMethod = "DecodeVLD2DupInstruction";
1423 defm VLD2DUPd8wb : VLD2DUPWB<{0,0,0,0}, "8", VecListDPairAllLanes>;
1424 defm VLD2DUPd16wb : VLD2DUPWB<{0,1,0,?}, "16", VecListDPairAllLanes>;
1425 defm VLD2DUPd32wb : VLD2DUPWB<{1,0,0,?}, "32", VecListDPairAllLanes>;
1427 defm VLD2DUPd8x2wb : VLD2DUPWB<{0,0,1,0}, "8", VecListDPairSpacedAllLanes>;
1428 defm VLD2DUPd16x2wb : VLD2DUPWB<{0,1,1,?}, "16", VecListDPairSpacedAllLanes>;
1429 defm VLD2DUPd32x2wb : VLD2DUPWB<{1,0,1,?}, "32", VecListDPairSpacedAllLanes>;
1431 // VLD3DUP : Vector Load (single 3-element structure to all lanes)
1432 class VLD3DUP<bits<4> op7_4, string Dt>
1433 : NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
1434 (ins addrmode6dup:$Rn), IIC_VLD3dup,
1435 "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn", "", []> {
1438 let DecoderMethod = "DecodeVLD3DupInstruction";
1441 def VLD3DUPd8 : VLD3DUP<{0,0,0,?}, "8">;
1442 def VLD3DUPd16 : VLD3DUP<{0,1,0,?}, "16">;
1443 def VLD3DUPd32 : VLD3DUP<{1,0,0,?}, "32">;
1445 def VLD3DUPd8Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1446 def VLD3DUPd16Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1447 def VLD3DUPd32Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1449 // ...with double-spaced registers (not used for codegen):
1450 def VLD3DUPq8 : VLD3DUP<{0,0,1,?}, "8">;
1451 def VLD3DUPq16 : VLD3DUP<{0,1,1,?}, "16">;
1452 def VLD3DUPq32 : VLD3DUP<{1,0,1,?}, "32">;
1454 // ...with address register writeback:
1455 class VLD3DUPWB<bits<4> op7_4, string Dt>
1456 : NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
1457 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD3dupu,
1458 "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn$Rm",
1459 "$Rn.addr = $wb", []> {
1461 let DecoderMethod = "DecodeVLD3DupInstruction";
1464 def VLD3DUPd8_UPD : VLD3DUPWB<{0,0,0,0}, "8">;
1465 def VLD3DUPd16_UPD : VLD3DUPWB<{0,1,0,?}, "16">;
1466 def VLD3DUPd32_UPD : VLD3DUPWB<{1,0,0,?}, "32">;
1468 def VLD3DUPq8_UPD : VLD3DUPWB<{0,0,1,0}, "8">;
1469 def VLD3DUPq16_UPD : VLD3DUPWB<{0,1,1,?}, "16">;
1470 def VLD3DUPq32_UPD : VLD3DUPWB<{1,0,1,?}, "32">;
1472 def VLD3DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1473 def VLD3DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1474 def VLD3DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1476 // VLD4DUP : Vector Load (single 4-element structure to all lanes)
1477 class VLD4DUP<bits<4> op7_4, string Dt>
1478 : NLdSt<1, 0b10, 0b1111, op7_4,
1479 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
1480 (ins addrmode6dup:$Rn), IIC_VLD4dup,
1481 "vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn", "", []> {
1483 let Inst{4} = Rn{4};
1484 let DecoderMethod = "DecodeVLD4DupInstruction";
1487 def VLD4DUPd8 : VLD4DUP<{0,0,0,?}, "8">;
1488 def VLD4DUPd16 : VLD4DUP<{0,1,0,?}, "16">;
1489 def VLD4DUPd32 : VLD4DUP<{1,?,0,?}, "32"> { let Inst{6} = Rn{5}; }
1491 def VLD4DUPd8Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1492 def VLD4DUPd16Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1493 def VLD4DUPd32Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1495 // ...with double-spaced registers (not used for codegen):
1496 def VLD4DUPq8 : VLD4DUP<{0,0,1,?}, "8">;
1497 def VLD4DUPq16 : VLD4DUP<{0,1,1,?}, "16">;
1498 def VLD4DUPq32 : VLD4DUP<{1,?,1,?}, "32"> { let Inst{6} = Rn{5}; }
1500 // ...with address register writeback:
1501 class VLD4DUPWB<bits<4> op7_4, string Dt>
1502 : NLdSt<1, 0b10, 0b1111, op7_4,
1503 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
1504 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD4dupu,
1505 "vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn$Rm",
1506 "$Rn.addr = $wb", []> {
1507 let Inst{4} = Rn{4};
1508 let DecoderMethod = "DecodeVLD4DupInstruction";
1511 def VLD4DUPd8_UPD : VLD4DUPWB<{0,0,0,0}, "8">;
1512 def VLD4DUPd16_UPD : VLD4DUPWB<{0,1,0,?}, "16">;
1513 def VLD4DUPd32_UPD : VLD4DUPWB<{1,?,0,?}, "32"> { let Inst{6} = Rn{5}; }
1515 def VLD4DUPq8_UPD : VLD4DUPWB<{0,0,1,0}, "8">;
1516 def VLD4DUPq16_UPD : VLD4DUPWB<{0,1,1,?}, "16">;
1517 def VLD4DUPq32_UPD : VLD4DUPWB<{1,?,1,?}, "32"> { let Inst{6} = Rn{5}; }
1519 def VLD4DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1520 def VLD4DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1521 def VLD4DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1523 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
1525 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
1527 // Classes for VST* pseudo-instructions with multi-register operands.
1528 // These are expanded to real instructions after register allocation.
1529 class VSTQPseudo<InstrItinClass itin>
1530 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src), itin, "">;
1531 class VSTQWBPseudo<InstrItinClass itin>
1532 : PseudoNLdSt<(outs GPR:$wb),
1533 (ins addrmode6:$addr, am6offset:$offset, QPR:$src), itin,
1534 "$addr.addr = $wb">;
1535 class VSTQWBfixedPseudo<InstrItinClass itin>
1536 : PseudoNLdSt<(outs GPR:$wb),
1537 (ins addrmode6:$addr, QPR:$src), itin,
1538 "$addr.addr = $wb">;
1539 class VSTQWBregisterPseudo<InstrItinClass itin>
1540 : PseudoNLdSt<(outs GPR:$wb),
1541 (ins addrmode6:$addr, rGPR:$offset, QPR:$src), itin,
1542 "$addr.addr = $wb">;
1543 class VSTQQPseudo<InstrItinClass itin>
1544 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src), itin, "">;
1545 class VSTQQWBPseudo<InstrItinClass itin>
1546 : PseudoNLdSt<(outs GPR:$wb),
1547 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src), itin,
1548 "$addr.addr = $wb">;
1549 class VSTQQWBfixedPseudo<InstrItinClass itin>
1550 : PseudoNLdSt<(outs GPR:$wb),
1551 (ins addrmode6:$addr, QQPR:$src), itin,
1552 "$addr.addr = $wb">;
1553 class VSTQQWBregisterPseudo<InstrItinClass itin>
1554 : PseudoNLdSt<(outs GPR:$wb),
1555 (ins addrmode6:$addr, rGPR:$offset, QQPR:$src), itin,
1556 "$addr.addr = $wb">;
1558 class VSTQQQQPseudo<InstrItinClass itin>
1559 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src), itin, "">;
1560 class VSTQQQQWBPseudo<InstrItinClass itin>
1561 : PseudoNLdSt<(outs GPR:$wb),
1562 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
1563 "$addr.addr = $wb">;
1565 // VST1 : Vector Store (multiple single elements)
1566 class VST1D<bits<4> op7_4, string Dt>
1567 : NLdSt<0,0b00,0b0111,op7_4, (outs), (ins addrmode6:$Rn, VecListOneD:$Vd),
1568 IIC_VST1, "vst1", Dt, "$Vd, $Rn", "", []> {
1570 let Inst{4} = Rn{4};
1571 let DecoderMethod = "DecodeVLDST1Instruction";
1573 class VST1Q<bits<4> op7_4, string Dt>
1574 : NLdSt<0,0b00,0b1010,op7_4, (outs), (ins addrmode6:$Rn, VecListDPair:$Vd),
1575 IIC_VST1x2, "vst1", Dt, "$Vd, $Rn", "", []> {
1577 let Inst{5-4} = Rn{5-4};
1578 let DecoderMethod = "DecodeVLDST1Instruction";
1581 def VST1d8 : VST1D<{0,0,0,?}, "8">;
1582 def VST1d16 : VST1D<{0,1,0,?}, "16">;
1583 def VST1d32 : VST1D<{1,0,0,?}, "32">;
1584 def VST1d64 : VST1D<{1,1,0,?}, "64">;
1586 def VST1q8 : VST1Q<{0,0,?,?}, "8">;
1587 def VST1q16 : VST1Q<{0,1,?,?}, "16">;
1588 def VST1q32 : VST1Q<{1,0,?,?}, "32">;
1589 def VST1q64 : VST1Q<{1,1,?,?}, "64">;
1591 // ...with address register writeback:
1592 multiclass VST1DWB<bits<4> op7_4, string Dt> {
1593 def _fixed : NLdSt<0,0b00, 0b0111,op7_4, (outs GPR:$wb),
1594 (ins addrmode6:$Rn, VecListOneD:$Vd), IIC_VLD1u,
1595 "vst1", Dt, "$Vd, $Rn!",
1596 "$Rn.addr = $wb", []> {
1597 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1598 let Inst{4} = Rn{4};
1599 let DecoderMethod = "DecodeVLDST1Instruction";
1601 def _register : NLdSt<0,0b00,0b0111,op7_4, (outs GPR:$wb),
1602 (ins addrmode6:$Rn, rGPR:$Rm, VecListOneD:$Vd),
1604 "vst1", Dt, "$Vd, $Rn, $Rm",
1605 "$Rn.addr = $wb", []> {
1606 let Inst{4} = Rn{4};
1607 let DecoderMethod = "DecodeVLDST1Instruction";
1610 multiclass VST1QWB<bits<4> op7_4, string Dt> {
1611 def _fixed : NLdSt<0,0b00,0b1010,op7_4, (outs GPR:$wb),
1612 (ins addrmode6:$Rn, VecListDPair:$Vd), IIC_VLD1x2u,
1613 "vst1", Dt, "$Vd, $Rn!",
1614 "$Rn.addr = $wb", []> {
1615 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1616 let Inst{5-4} = Rn{5-4};
1617 let DecoderMethod = "DecodeVLDST1Instruction";
1619 def _register : NLdSt<0,0b00,0b1010,op7_4, (outs GPR:$wb),
1620 (ins addrmode6:$Rn, rGPR:$Rm, VecListDPair:$Vd),
1622 "vst1", Dt, "$Vd, $Rn, $Rm",
1623 "$Rn.addr = $wb", []> {
1624 let Inst{5-4} = Rn{5-4};
1625 let DecoderMethod = "DecodeVLDST1Instruction";
1629 defm VST1d8wb : VST1DWB<{0,0,0,?}, "8">;
1630 defm VST1d16wb : VST1DWB<{0,1,0,?}, "16">;
1631 defm VST1d32wb : VST1DWB<{1,0,0,?}, "32">;
1632 defm VST1d64wb : VST1DWB<{1,1,0,?}, "64">;
1634 defm VST1q8wb : VST1QWB<{0,0,?,?}, "8">;
1635 defm VST1q16wb : VST1QWB<{0,1,?,?}, "16">;
1636 defm VST1q32wb : VST1QWB<{1,0,?,?}, "32">;
1637 defm VST1q64wb : VST1QWB<{1,1,?,?}, "64">;
1639 // ...with 3 registers
1640 class VST1D3<bits<4> op7_4, string Dt>
1641 : NLdSt<0, 0b00, 0b0110, op7_4, (outs),
1642 (ins addrmode6:$Rn, VecListThreeD:$Vd),
1643 IIC_VST1x3, "vst1", Dt, "$Vd, $Rn", "", []> {
1645 let Inst{4} = Rn{4};
1646 let DecoderMethod = "DecodeVLDST1Instruction";
1648 multiclass VST1D3WB<bits<4> op7_4, string Dt> {
1649 def _fixed : NLdSt<0,0b00,0b0110,op7_4, (outs GPR:$wb),
1650 (ins addrmode6:$Rn, VecListThreeD:$Vd), IIC_VLD1x3u,
1651 "vst1", Dt, "$Vd, $Rn!",
1652 "$Rn.addr = $wb", []> {
1653 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1654 let Inst{5-4} = Rn{5-4};
1655 let DecoderMethod = "DecodeVLDST1Instruction";
1657 def _register : NLdSt<0,0b00,0b0110,op7_4, (outs GPR:$wb),
1658 (ins addrmode6:$Rn, rGPR:$Rm, VecListThreeD:$Vd),
1660 "vst1", Dt, "$Vd, $Rn, $Rm",
1661 "$Rn.addr = $wb", []> {
1662 let Inst{5-4} = Rn{5-4};
1663 let DecoderMethod = "DecodeVLDST1Instruction";
1667 def VST1d8T : VST1D3<{0,0,0,?}, "8">;
1668 def VST1d16T : VST1D3<{0,1,0,?}, "16">;
1669 def VST1d32T : VST1D3<{1,0,0,?}, "32">;
1670 def VST1d64T : VST1D3<{1,1,0,?}, "64">;
1672 defm VST1d8Twb : VST1D3WB<{0,0,0,?}, "8">;
1673 defm VST1d16Twb : VST1D3WB<{0,1,0,?}, "16">;
1674 defm VST1d32Twb : VST1D3WB<{1,0,0,?}, "32">;
1675 defm VST1d64Twb : VST1D3WB<{1,1,0,?}, "64">;
1677 def VST1d64TPseudo : VSTQQPseudo<IIC_VST1x3>;
1678 def VST1d64TPseudoWB_fixed : VSTQQWBfixedPseudo<IIC_VST1x3u>;
1679 def VST1d64TPseudoWB_register : VSTQQWBPseudo<IIC_VST1x3u>;
1681 // ...with 4 registers
1682 class VST1D4<bits<4> op7_4, string Dt>
1683 : NLdSt<0, 0b00, 0b0010, op7_4, (outs),
1684 (ins addrmode6:$Rn, VecListFourD:$Vd),
1685 IIC_VST1x4, "vst1", Dt, "$Vd, $Rn", "",
1688 let Inst{5-4} = Rn{5-4};
1689 let DecoderMethod = "DecodeVLDST1Instruction";
1691 multiclass VST1D4WB<bits<4> op7_4, string Dt> {
1692 def _fixed : NLdSt<0,0b00,0b0010,op7_4, (outs GPR:$wb),
1693 (ins addrmode6:$Rn, VecListFourD:$Vd), IIC_VLD1x4u,
1694 "vst1", Dt, "$Vd, $Rn!",
1695 "$Rn.addr = $wb", []> {
1696 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1697 let Inst{5-4} = Rn{5-4};
1698 let DecoderMethod = "DecodeVLDST1Instruction";
1700 def _register : NLdSt<0,0b00,0b0010,op7_4, (outs GPR:$wb),
1701 (ins addrmode6:$Rn, rGPR:$Rm, VecListFourD:$Vd),
1703 "vst1", Dt, "$Vd, $Rn, $Rm",
1704 "$Rn.addr = $wb", []> {
1705 let Inst{5-4} = Rn{5-4};
1706 let DecoderMethod = "DecodeVLDST1Instruction";
1710 def VST1d8Q : VST1D4<{0,0,?,?}, "8">;
1711 def VST1d16Q : VST1D4<{0,1,?,?}, "16">;
1712 def VST1d32Q : VST1D4<{1,0,?,?}, "32">;
1713 def VST1d64Q : VST1D4<{1,1,?,?}, "64">;
1715 defm VST1d8Qwb : VST1D4WB<{0,0,?,?}, "8">;
1716 defm VST1d16Qwb : VST1D4WB<{0,1,?,?}, "16">;
1717 defm VST1d32Qwb : VST1D4WB<{1,0,?,?}, "32">;
1718 defm VST1d64Qwb : VST1D4WB<{1,1,?,?}, "64">;
1720 def VST1d64QPseudo : VSTQQPseudo<IIC_VST1x4>;
1721 def VST1d64QPseudoWB_fixed : VSTQQWBfixedPseudo<IIC_VST1x4u>;
1722 def VST1d64QPseudoWB_register : VSTQQWBPseudo<IIC_VST1x4u>;
1724 // VST2 : Vector Store (multiple 2-element structures)
1725 class VST2<bits<4> op11_8, bits<4> op7_4, string Dt, RegisterOperand VdTy,
1726 InstrItinClass itin>
1727 : NLdSt<0, 0b00, op11_8, op7_4, (outs), (ins addrmode6:$Rn, VdTy:$Vd),
1728 itin, "vst2", Dt, "$Vd, $Rn", "", []> {
1730 let Inst{5-4} = Rn{5-4};
1731 let DecoderMethod = "DecodeVLDST2Instruction";
1734 def VST2d8 : VST2<0b1000, {0,0,?,?}, "8", VecListDPair, IIC_VST2>;
1735 def VST2d16 : VST2<0b1000, {0,1,?,?}, "16", VecListDPair, IIC_VST2>;
1736 def VST2d32 : VST2<0b1000, {1,0,?,?}, "32", VecListDPair, IIC_VST2>;
1738 def VST2q8 : VST2<0b0011, {0,0,?,?}, "8", VecListFourD, IIC_VST2x2>;
1739 def VST2q16 : VST2<0b0011, {0,1,?,?}, "16", VecListFourD, IIC_VST2x2>;
1740 def VST2q32 : VST2<0b0011, {1,0,?,?}, "32", VecListFourD, IIC_VST2x2>;
1742 def VST2q8Pseudo : VSTQQPseudo<IIC_VST2x2>;
1743 def VST2q16Pseudo : VSTQQPseudo<IIC_VST2x2>;
1744 def VST2q32Pseudo : VSTQQPseudo<IIC_VST2x2>;
1746 // ...with address register writeback:
1747 multiclass VST2DWB<bits<4> op11_8, bits<4> op7_4, string Dt,
1748 RegisterOperand VdTy> {
1749 def _fixed : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1750 (ins addrmode6:$Rn, VdTy:$Vd), IIC_VLD1u,
1751 "vst2", Dt, "$Vd, $Rn!",
1752 "$Rn.addr = $wb", []> {
1753 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1754 let Inst{5-4} = Rn{5-4};
1755 let DecoderMethod = "DecodeVLDST2Instruction";
1757 def _register : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1758 (ins addrmode6:$Rn, rGPR:$Rm, VdTy:$Vd), IIC_VLD1u,
1759 "vst2", Dt, "$Vd, $Rn, $Rm",
1760 "$Rn.addr = $wb", []> {
1761 let Inst{5-4} = Rn{5-4};
1762 let DecoderMethod = "DecodeVLDST2Instruction";
1765 multiclass VST2QWB<bits<4> op7_4, string Dt> {
1766 def _fixed : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
1767 (ins addrmode6:$Rn, VecListFourD:$Vd), IIC_VLD1u,
1768 "vst2", Dt, "$Vd, $Rn!",
1769 "$Rn.addr = $wb", []> {
1770 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1771 let Inst{5-4} = Rn{5-4};
1772 let DecoderMethod = "DecodeVLDST2Instruction";
1774 def _register : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
1775 (ins addrmode6:$Rn, rGPR:$Rm, VecListFourD:$Vd),
1777 "vst2", Dt, "$Vd, $Rn, $Rm",
1778 "$Rn.addr = $wb", []> {
1779 let Inst{5-4} = Rn{5-4};
1780 let DecoderMethod = "DecodeVLDST2Instruction";
1784 defm VST2d8wb : VST2DWB<0b1000, {0,0,?,?}, "8", VecListDPair>;
1785 defm VST2d16wb : VST2DWB<0b1000, {0,1,?,?}, "16", VecListDPair>;
1786 defm VST2d32wb : VST2DWB<0b1000, {1,0,?,?}, "32", VecListDPair>;
1788 defm VST2q8wb : VST2QWB<{0,0,?,?}, "8">;
1789 defm VST2q16wb : VST2QWB<{0,1,?,?}, "16">;
1790 defm VST2q32wb : VST2QWB<{1,0,?,?}, "32">;
1792 def VST2q8PseudoWB_fixed : VSTQQWBfixedPseudo<IIC_VST2x2u>;
1793 def VST2q16PseudoWB_fixed : VSTQQWBfixedPseudo<IIC_VST2x2u>;
1794 def VST2q32PseudoWB_fixed : VSTQQWBfixedPseudo<IIC_VST2x2u>;
1795 def VST2q8PseudoWB_register : VSTQQWBregisterPseudo<IIC_VST2x2u>;
1796 def VST2q16PseudoWB_register : VSTQQWBregisterPseudo<IIC_VST2x2u>;
1797 def VST2q32PseudoWB_register : VSTQQWBregisterPseudo<IIC_VST2x2u>;
1799 // ...with double-spaced registers
1800 def VST2b8 : VST2<0b1001, {0,0,?,?}, "8", VecListDPairSpaced, IIC_VST2>;
1801 def VST2b16 : VST2<0b1001, {0,1,?,?}, "16", VecListDPairSpaced, IIC_VST2>;
1802 def VST2b32 : VST2<0b1001, {1,0,?,?}, "32", VecListDPairSpaced, IIC_VST2>;
1803 defm VST2b8wb : VST2DWB<0b1001, {0,0,?,?}, "8", VecListDPairSpaced>;
1804 defm VST2b16wb : VST2DWB<0b1001, {0,1,?,?}, "16", VecListDPairSpaced>;
1805 defm VST2b32wb : VST2DWB<0b1001, {1,0,?,?}, "32", VecListDPairSpaced>;
1807 // VST3 : Vector Store (multiple 3-element structures)
1808 class VST3D<bits<4> op11_8, bits<4> op7_4, string Dt>
1809 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1810 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3,
1811 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn", "", []> {
1813 let Inst{4} = Rn{4};
1814 let DecoderMethod = "DecodeVLDST3Instruction";
1817 def VST3d8 : VST3D<0b0100, {0,0,0,?}, "8">;
1818 def VST3d16 : VST3D<0b0100, {0,1,0,?}, "16">;
1819 def VST3d32 : VST3D<0b0100, {1,0,0,?}, "32">;
1821 def VST3d8Pseudo : VSTQQPseudo<IIC_VST3>;
1822 def VST3d16Pseudo : VSTQQPseudo<IIC_VST3>;
1823 def VST3d32Pseudo : VSTQQPseudo<IIC_VST3>;
1825 // ...with address register writeback:
1826 class VST3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1827 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1828 (ins addrmode6:$Rn, am6offset:$Rm,
1829 DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3u,
1830 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn$Rm",
1831 "$Rn.addr = $wb", []> {
1832 let Inst{4} = Rn{4};
1833 let DecoderMethod = "DecodeVLDST3Instruction";
1836 def VST3d8_UPD : VST3DWB<0b0100, {0,0,0,?}, "8">;
1837 def VST3d16_UPD : VST3DWB<0b0100, {0,1,0,?}, "16">;
1838 def VST3d32_UPD : VST3DWB<0b0100, {1,0,0,?}, "32">;
1840 def VST3d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1841 def VST3d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1842 def VST3d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1844 // ...with double-spaced registers:
1845 def VST3q8 : VST3D<0b0101, {0,0,0,?}, "8">;
1846 def VST3q16 : VST3D<0b0101, {0,1,0,?}, "16">;
1847 def VST3q32 : VST3D<0b0101, {1,0,0,?}, "32">;
1848 def VST3q8_UPD : VST3DWB<0b0101, {0,0,0,?}, "8">;
1849 def VST3q16_UPD : VST3DWB<0b0101, {0,1,0,?}, "16">;
1850 def VST3q32_UPD : VST3DWB<0b0101, {1,0,0,?}, "32">;
1852 def VST3q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1853 def VST3q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1854 def VST3q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1856 // ...alternate versions to be allocated odd register numbers:
1857 def VST3q8oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1858 def VST3q16oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1859 def VST3q32oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1861 def VST3q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1862 def VST3q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1863 def VST3q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1865 // VST4 : Vector Store (multiple 4-element structures)
1866 class VST4D<bits<4> op11_8, bits<4> op7_4, string Dt>
1867 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1868 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1869 IIC_VST4, "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn",
1872 let Inst{5-4} = Rn{5-4};
1873 let DecoderMethod = "DecodeVLDST4Instruction";
1876 def VST4d8 : VST4D<0b0000, {0,0,?,?}, "8">;
1877 def VST4d16 : VST4D<0b0000, {0,1,?,?}, "16">;
1878 def VST4d32 : VST4D<0b0000, {1,0,?,?}, "32">;
1880 def VST4d8Pseudo : VSTQQPseudo<IIC_VST4>;
1881 def VST4d16Pseudo : VSTQQPseudo<IIC_VST4>;
1882 def VST4d32Pseudo : VSTQQPseudo<IIC_VST4>;
1884 // ...with address register writeback:
1885 class VST4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1886 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1887 (ins addrmode6:$Rn, am6offset:$Rm,
1888 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST4u,
1889 "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1890 "$Rn.addr = $wb", []> {
1891 let Inst{5-4} = Rn{5-4};
1892 let DecoderMethod = "DecodeVLDST4Instruction";
1895 def VST4d8_UPD : VST4DWB<0b0000, {0,0,?,?}, "8">;
1896 def VST4d16_UPD : VST4DWB<0b0000, {0,1,?,?}, "16">;
1897 def VST4d32_UPD : VST4DWB<0b0000, {1,0,?,?}, "32">;
1899 def VST4d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1900 def VST4d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1901 def VST4d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1903 // ...with double-spaced registers:
1904 def VST4q8 : VST4D<0b0001, {0,0,?,?}, "8">;
1905 def VST4q16 : VST4D<0b0001, {0,1,?,?}, "16">;
1906 def VST4q32 : VST4D<0b0001, {1,0,?,?}, "32">;
1907 def VST4q8_UPD : VST4DWB<0b0001, {0,0,?,?}, "8">;
1908 def VST4q16_UPD : VST4DWB<0b0001, {0,1,?,?}, "16">;
1909 def VST4q32_UPD : VST4DWB<0b0001, {1,0,?,?}, "32">;
1911 def VST4q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1912 def VST4q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1913 def VST4q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1915 // ...alternate versions to be allocated odd register numbers:
1916 def VST4q8oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1917 def VST4q16oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1918 def VST4q32oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1920 def VST4q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1921 def VST4q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1922 def VST4q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1924 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
1926 // Classes for VST*LN pseudo-instructions with multi-register operands.
1927 // These are expanded to real instructions after register allocation.
1928 class VSTQLNPseudo<InstrItinClass itin>
1929 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
1931 class VSTQLNWBPseudo<InstrItinClass itin>
1932 : PseudoNLdSt<(outs GPR:$wb),
1933 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
1934 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1935 class VSTQQLNPseudo<InstrItinClass itin>
1936 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
1938 class VSTQQLNWBPseudo<InstrItinClass itin>
1939 : PseudoNLdSt<(outs GPR:$wb),
1940 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
1941 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1942 class VSTQQQQLNPseudo<InstrItinClass itin>
1943 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
1945 class VSTQQQQLNWBPseudo<InstrItinClass itin>
1946 : PseudoNLdSt<(outs GPR:$wb),
1947 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
1948 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1950 // VST1LN : Vector Store (single element from one lane)
1951 class VST1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1952 PatFrag StoreOp, SDNode ExtractOp, Operand AddrMode>
1953 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1954 (ins AddrMode:$Rn, DPR:$Vd, nohash_imm:$lane),
1955 IIC_VST1ln, "vst1", Dt, "\\{$Vd[$lane]\\}, $Rn", "",
1956 [(StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane), AddrMode:$Rn)]> {
1958 let DecoderMethod = "DecodeVST1LN";
1960 class VST1QLNPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
1961 : VSTQLNPseudo<IIC_VST1ln> {
1962 let Pattern = [(StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
1966 def VST1LNd8 : VST1LN<0b0000, {?,?,?,0}, "8", v8i8, truncstorei8,
1967 NEONvgetlaneu, addrmode6> {
1968 let Inst{7-5} = lane{2-0};
1970 def VST1LNd16 : VST1LN<0b0100, {?,?,0,?}, "16", v4i16, truncstorei16,
1971 NEONvgetlaneu, addrmode6> {
1972 let Inst{7-6} = lane{1-0};
1973 let Inst{4} = Rn{4};
1976 def VST1LNd32 : VST1LN<0b1000, {?,0,?,?}, "32", v2i32, store, extractelt,
1978 let Inst{7} = lane{0};
1979 let Inst{5-4} = Rn{5-4};
1982 def VST1LNq8Pseudo : VST1QLNPseudo<v16i8, truncstorei8, NEONvgetlaneu>;
1983 def VST1LNq16Pseudo : VST1QLNPseudo<v8i16, truncstorei16, NEONvgetlaneu>;
1984 def VST1LNq32Pseudo : VST1QLNPseudo<v4i32, store, extractelt>;
1986 def : Pat<(store (extractelt (v2f32 DPR:$src), imm:$lane), addrmode6:$addr),
1987 (VST1LNd32 addrmode6:$addr, DPR:$src, imm:$lane)>;
1988 def : Pat<(store (extractelt (v4f32 QPR:$src), imm:$lane), addrmode6:$addr),
1989 (VST1LNq32Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
1991 // ...with address register writeback:
1992 class VST1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1993 PatFrag StoreOp, SDNode ExtractOp, Operand AdrMode>
1994 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1995 (ins AdrMode:$Rn, am6offset:$Rm,
1996 DPR:$Vd, nohash_imm:$lane), IIC_VST1lnu, "vst1", Dt,
1997 "\\{$Vd[$lane]\\}, $Rn$Rm",
1999 [(set GPR:$wb, (StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane),
2000 AdrMode:$Rn, am6offset:$Rm))]> {
2001 let DecoderMethod = "DecodeVST1LN";
2003 class VST1QLNWBPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
2004 : VSTQLNWBPseudo<IIC_VST1lnu> {
2005 let Pattern = [(set GPR:$wb, (StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
2006 addrmode6:$addr, am6offset:$offset))];
2009 def VST1LNd8_UPD : VST1LNWB<0b0000, {?,?,?,0}, "8", v8i8, post_truncsti8,
2010 NEONvgetlaneu, addrmode6> {
2011 let Inst{7-5} = lane{2-0};
2013 def VST1LNd16_UPD : VST1LNWB<0b0100, {?,?,0,?}, "16", v4i16, post_truncsti16,
2014 NEONvgetlaneu, addrmode6> {
2015 let Inst{7-6} = lane{1-0};
2016 let Inst{4} = Rn{4};
2018 def VST1LNd32_UPD : VST1LNWB<0b1000, {?,0,?,?}, "32", v2i32, post_store,
2019 extractelt, addrmode6oneL32> {
2020 let Inst{7} = lane{0};
2021 let Inst{5-4} = Rn{5-4};
2024 def VST1LNq8Pseudo_UPD : VST1QLNWBPseudo<v16i8, post_truncsti8, NEONvgetlaneu>;
2025 def VST1LNq16Pseudo_UPD : VST1QLNWBPseudo<v8i16, post_truncsti16,NEONvgetlaneu>;
2026 def VST1LNq32Pseudo_UPD : VST1QLNWBPseudo<v4i32, post_store, extractelt>;
2028 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
2030 // VST2LN : Vector Store (single 2-element structure from one lane)
2031 class VST2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
2032 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
2033 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, nohash_imm:$lane),
2034 IIC_VST2ln, "vst2", Dt, "\\{$Vd[$lane], $src2[$lane]\\}, $Rn",
2037 let Inst{4} = Rn{4};
2038 let DecoderMethod = "DecodeVST2LN";
2041 def VST2LNd8 : VST2LN<0b0001, {?,?,?,?}, "8"> {
2042 let Inst{7-5} = lane{2-0};
2044 def VST2LNd16 : VST2LN<0b0101, {?,?,0,?}, "16"> {
2045 let Inst{7-6} = lane{1-0};
2047 def VST2LNd32 : VST2LN<0b1001, {?,0,0,?}, "32"> {
2048 let Inst{7} = lane{0};
2051 def VST2LNd8Pseudo : VSTQLNPseudo<IIC_VST2ln>;
2052 def VST2LNd16Pseudo : VSTQLNPseudo<IIC_VST2ln>;
2053 def VST2LNd32Pseudo : VSTQLNPseudo<IIC_VST2ln>;
2055 // ...with double-spaced registers:
2056 def VST2LNq16 : VST2LN<0b0101, {?,?,1,?}, "16"> {
2057 let Inst{7-6} = lane{1-0};
2058 let Inst{4} = Rn{4};
2060 def VST2LNq32 : VST2LN<0b1001, {?,1,0,?}, "32"> {
2061 let Inst{7} = lane{0};
2062 let Inst{4} = Rn{4};
2065 def VST2LNq16Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
2066 def VST2LNq32Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
2068 // ...with address register writeback:
2069 class VST2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
2070 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
2071 (ins addrmode6:$Rn, am6offset:$Rm,
2072 DPR:$Vd, DPR:$src2, nohash_imm:$lane), IIC_VST2lnu, "vst2", Dt,
2073 "\\{$Vd[$lane], $src2[$lane]\\}, $Rn$Rm",
2074 "$Rn.addr = $wb", []> {
2075 let Inst{4} = Rn{4};
2076 let DecoderMethod = "DecodeVST2LN";
2079 def VST2LNd8_UPD : VST2LNWB<0b0001, {?,?,?,?}, "8"> {
2080 let Inst{7-5} = lane{2-0};
2082 def VST2LNd16_UPD : VST2LNWB<0b0101, {?,?,0,?}, "16"> {
2083 let Inst{7-6} = lane{1-0};
2085 def VST2LNd32_UPD : VST2LNWB<0b1001, {?,0,0,?}, "32"> {
2086 let Inst{7} = lane{0};
2089 def VST2LNd8Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
2090 def VST2LNd16Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
2091 def VST2LNd32Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
2093 def VST2LNq16_UPD : VST2LNWB<0b0101, {?,?,1,?}, "16"> {
2094 let Inst{7-6} = lane{1-0};
2096 def VST2LNq32_UPD : VST2LNWB<0b1001, {?,1,0,?}, "32"> {
2097 let Inst{7} = lane{0};
2100 def VST2LNq16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
2101 def VST2LNq32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
2103 // VST3LN : Vector Store (single 3-element structure from one lane)
2104 class VST3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
2105 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
2106 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3,
2107 nohash_imm:$lane), IIC_VST3ln, "vst3", Dt,
2108 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn", "", []> {
2110 let DecoderMethod = "DecodeVST3LN";
2113 def VST3LNd8 : VST3LN<0b0010, {?,?,?,0}, "8"> {
2114 let Inst{7-5} = lane{2-0};
2116 def VST3LNd16 : VST3LN<0b0110, {?,?,0,0}, "16"> {
2117 let Inst{7-6} = lane{1-0};
2119 def VST3LNd32 : VST3LN<0b1010, {?,0,0,0}, "32"> {
2120 let Inst{7} = lane{0};
2123 def VST3LNd8Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
2124 def VST3LNd16Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
2125 def VST3LNd32Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
2127 // ...with double-spaced registers:
2128 def VST3LNq16 : VST3LN<0b0110, {?,?,1,0}, "16"> {
2129 let Inst{7-6} = lane{1-0};
2131 def VST3LNq32 : VST3LN<0b1010, {?,1,0,0}, "32"> {
2132 let Inst{7} = lane{0};
2135 def VST3LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
2136 def VST3LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
2138 // ...with address register writeback:
2139 class VST3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
2140 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
2141 (ins addrmode6:$Rn, am6offset:$Rm,
2142 DPR:$Vd, DPR:$src2, DPR:$src3, nohash_imm:$lane),
2143 IIC_VST3lnu, "vst3", Dt,
2144 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn$Rm",
2145 "$Rn.addr = $wb", []> {
2146 let DecoderMethod = "DecodeVST3LN";
2149 def VST3LNd8_UPD : VST3LNWB<0b0010, {?,?,?,0}, "8"> {
2150 let Inst{7-5} = lane{2-0};
2152 def VST3LNd16_UPD : VST3LNWB<0b0110, {?,?,0,0}, "16"> {
2153 let Inst{7-6} = lane{1-0};
2155 def VST3LNd32_UPD : VST3LNWB<0b1010, {?,0,0,0}, "32"> {
2156 let Inst{7} = lane{0};
2159 def VST3LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
2160 def VST3LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
2161 def VST3LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
2163 def VST3LNq16_UPD : VST3LNWB<0b0110, {?,?,1,0}, "16"> {
2164 let Inst{7-6} = lane{1-0};
2166 def VST3LNq32_UPD : VST3LNWB<0b1010, {?,1,0,0}, "32"> {
2167 let Inst{7} = lane{0};
2170 def VST3LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
2171 def VST3LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
2173 // VST4LN : Vector Store (single 4-element structure from one lane)
2174 class VST4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
2175 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
2176 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4,
2177 nohash_imm:$lane), IIC_VST4ln, "vst4", Dt,
2178 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn",
2181 let Inst{4} = Rn{4};
2182 let DecoderMethod = "DecodeVST4LN";
2185 def VST4LNd8 : VST4LN<0b0011, {?,?,?,?}, "8"> {
2186 let Inst{7-5} = lane{2-0};
2188 def VST4LNd16 : VST4LN<0b0111, {?,?,0,?}, "16"> {
2189 let Inst{7-6} = lane{1-0};
2191 def VST4LNd32 : VST4LN<0b1011, {?,0,?,?}, "32"> {
2192 let Inst{7} = lane{0};
2193 let Inst{5} = Rn{5};
2196 def VST4LNd8Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
2197 def VST4LNd16Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
2198 def VST4LNd32Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
2200 // ...with double-spaced registers:
2201 def VST4LNq16 : VST4LN<0b0111, {?,?,1,?}, "16"> {
2202 let Inst{7-6} = lane{1-0};
2204 def VST4LNq32 : VST4LN<0b1011, {?,1,?,?}, "32"> {
2205 let Inst{7} = lane{0};
2206 let Inst{5} = Rn{5};
2209 def VST4LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
2210 def VST4LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
2212 // ...with address register writeback:
2213 class VST4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
2214 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
2215 (ins addrmode6:$Rn, am6offset:$Rm,
2216 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
2217 IIC_VST4lnu, "vst4", Dt,
2218 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn$Rm",
2219 "$Rn.addr = $wb", []> {
2220 let Inst{4} = Rn{4};
2221 let DecoderMethod = "DecodeVST4LN";
2224 def VST4LNd8_UPD : VST4LNWB<0b0011, {?,?,?,?}, "8"> {
2225 let Inst{7-5} = lane{2-0};
2227 def VST4LNd16_UPD : VST4LNWB<0b0111, {?,?,0,?}, "16"> {
2228 let Inst{7-6} = lane{1-0};
2230 def VST4LNd32_UPD : VST4LNWB<0b1011, {?,0,?,?}, "32"> {
2231 let Inst{7} = lane{0};
2232 let Inst{5} = Rn{5};
2235 def VST4LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
2236 def VST4LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
2237 def VST4LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
2239 def VST4LNq16_UPD : VST4LNWB<0b0111, {?,?,1,?}, "16"> {
2240 let Inst{7-6} = lane{1-0};
2242 def VST4LNq32_UPD : VST4LNWB<0b1011, {?,1,?,?}, "32"> {
2243 let Inst{7} = lane{0};
2244 let Inst{5} = Rn{5};
2247 def VST4LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
2248 def VST4LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
2250 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
2252 // Use vld1/vst1 for unaligned f64 load / store
2253 def : Pat<(f64 (hword_alignedload addrmode6:$addr)),
2254 (VLD1d16 addrmode6:$addr)>, Requires<[IsLE]>;
2255 def : Pat<(hword_alignedstore (f64 DPR:$value), addrmode6:$addr),
2256 (VST1d16 addrmode6:$addr, DPR:$value)>, Requires<[IsLE]>;
2257 def : Pat<(f64 (byte_alignedload addrmode6:$addr)),
2258 (VLD1d8 addrmode6:$addr)>, Requires<[IsLE]>;
2259 def : Pat<(byte_alignedstore (f64 DPR:$value), addrmode6:$addr),
2260 (VST1d8 addrmode6:$addr, DPR:$value)>, Requires<[IsLE]>;
2261 def : Pat<(f64 (non_word_alignedload addrmode6:$addr)),
2262 (VLD1d64 addrmode6:$addr)>, Requires<[IsBE]>;
2263 def : Pat<(non_word_alignedstore (f64 DPR:$value), addrmode6:$addr),
2264 (VST1d64 addrmode6:$addr, DPR:$value)>, Requires<[IsBE]>;
2266 // Use vld1/vst1 for Q and QQ. Also use them for unaligned v2f64
2267 // load / store if it's legal.
2268 def : Pat<(v2f64 (dword_alignedload addrmode6:$addr)),
2269 (VLD1q64 addrmode6:$addr)>;
2270 def : Pat<(dword_alignedstore (v2f64 QPR:$value), addrmode6:$addr),
2271 (VST1q64 addrmode6:$addr, QPR:$value)>;
2272 def : Pat<(v2f64 (word_alignedload addrmode6:$addr)),
2273 (VLD1q32 addrmode6:$addr)>;
2274 def : Pat<(word_alignedstore (v2f64 QPR:$value), addrmode6:$addr),
2275 (VST1q32 addrmode6:$addr, QPR:$value)>;
2276 def : Pat<(v2f64 (hword_alignedload addrmode6:$addr)),
2277 (VLD1q16 addrmode6:$addr)>, Requires<[IsLE]>;
2278 def : Pat<(hword_alignedstore (v2f64 QPR:$value), addrmode6:$addr),
2279 (VST1q16 addrmode6:$addr, QPR:$value)>, Requires<[IsLE]>;
2280 def : Pat<(v2f64 (byte_alignedload addrmode6:$addr)),
2281 (VLD1q8 addrmode6:$addr)>, Requires<[IsLE]>;
2282 def : Pat<(byte_alignedstore (v2f64 QPR:$value), addrmode6:$addr),
2283 (VST1q8 addrmode6:$addr, QPR:$value)>, Requires<[IsLE]>;
2285 //===----------------------------------------------------------------------===//
2286 // NEON pattern fragments
2287 //===----------------------------------------------------------------------===//
2289 // Extract D sub-registers of Q registers.
2290 def DSubReg_i8_reg : SDNodeXForm<imm, [{
2291 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2292 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/8, MVT::i32);
2294 def DSubReg_i16_reg : SDNodeXForm<imm, [{
2295 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2296 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/4, MVT::i32);
2298 def DSubReg_i32_reg : SDNodeXForm<imm, [{
2299 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2300 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/2, MVT::i32);
2302 def DSubReg_f64_reg : SDNodeXForm<imm, [{
2303 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2304 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue(), MVT::i32);
2307 // Extract S sub-registers of Q/D registers.
2308 def SSubReg_f32_reg : SDNodeXForm<imm, [{
2309 assert(ARM::ssub_3 == ARM::ssub_0+3 && "Unexpected subreg numbering");
2310 return CurDAG->getTargetConstant(ARM::ssub_0 + N->getZExtValue(), MVT::i32);
2313 // Translate lane numbers from Q registers to D subregs.
2314 def SubReg_i8_lane : SDNodeXForm<imm, [{
2315 return CurDAG->getTargetConstant(N->getZExtValue() & 7, MVT::i32);
2317 def SubReg_i16_lane : SDNodeXForm<imm, [{
2318 return CurDAG->getTargetConstant(N->getZExtValue() & 3, MVT::i32);
2320 def SubReg_i32_lane : SDNodeXForm<imm, [{
2321 return CurDAG->getTargetConstant(N->getZExtValue() & 1, MVT::i32);
2324 //===----------------------------------------------------------------------===//
2325 // Instruction Classes
2326 //===----------------------------------------------------------------------===//
2328 // Basic 2-register operations: double- and quad-register.
2329 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2330 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
2331 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
2332 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2333 (ins DPR:$Vm), IIC_VUNAD, OpcodeStr, Dt,"$Vd, $Vm", "",
2334 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm))))]>;
2335 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2336 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
2337 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
2338 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2339 (ins QPR:$Vm), IIC_VUNAQ, OpcodeStr, Dt,"$Vd, $Vm", "",
2340 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vm))))]>;
2342 // Basic 2-register intrinsics, both double- and quad-register.
2343 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2344 bits<2> op17_16, bits<5> op11_7, bit op4,
2345 InstrItinClass itin, string OpcodeStr, string Dt,
2346 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2347 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2348 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2349 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
2350 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2351 bits<2> op17_16, bits<5> op11_7, bit op4,
2352 InstrItinClass itin, string OpcodeStr, string Dt,
2353 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2354 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2355 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2356 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
2358 // Same as above, but not predicated.
2359 class N2VDIntnp<bits<2> op17_16, bits<3> op10_8, bit op7,
2360 InstrItinClass itin, string OpcodeStr, string Dt,
2361 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2362 : N2Vnp<0b10, op17_16, op10_8, op7, 0, (outs DPR:$Vd), (ins DPR:$Vm),
2363 itin, OpcodeStr, Dt, ResTy, OpTy,
2364 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
2366 class N2VQIntnp<bits<2> op17_16, bits<3> op10_8, bit op7,
2367 InstrItinClass itin, string OpcodeStr, string Dt,
2368 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2369 : N2Vnp<0b10, op17_16, op10_8, op7, 1, (outs QPR:$Vd), (ins QPR:$Vm),
2370 itin, OpcodeStr, Dt, ResTy, OpTy,
2371 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
2373 // Similar to NV2VQIntnp with some more encoding bits exposed (crypto).
2374 class N2VQIntXnp<bits<2> op19_18, bits<2> op17_16, bits<3> op10_8, bit op6,
2375 bit op7, InstrItinClass itin, string OpcodeStr, string Dt,
2376 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2377 : N2Vnp<op19_18, op17_16, op10_8, op7, op6, (outs QPR:$Vd), (ins QPR:$Vm),
2378 itin, OpcodeStr, Dt, ResTy, OpTy,
2379 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
2381 // Same as N2VQIntXnp but with Vd as a src register.
2382 class N2VQIntX2np<bits<2> op19_18, bits<2> op17_16, bits<3> op10_8, bit op6,
2383 bit op7, InstrItinClass itin, string OpcodeStr, string Dt,
2384 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2385 : N2Vnp<op19_18, op17_16, op10_8, op7, op6,
2386 (outs QPR:$Vd), (ins QPR:$src, QPR:$Vm),
2387 itin, OpcodeStr, Dt, ResTy, OpTy,
2388 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$src), (OpTy QPR:$Vm))))]> {
2389 let Constraints = "$src = $Vd";
2392 // Narrow 2-register operations.
2393 class N2VN<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2394 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2395 InstrItinClass itin, string OpcodeStr, string Dt,
2396 ValueType TyD, ValueType TyQ, SDNode OpNode>
2397 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$Vd),
2398 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2399 [(set DPR:$Vd, (TyD (OpNode (TyQ QPR:$Vm))))]>;
2401 // Narrow 2-register intrinsics.
2402 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2403 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2404 InstrItinClass itin, string OpcodeStr, string Dt,
2405 ValueType TyD, ValueType TyQ, SDPatternOperator IntOp>
2406 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$Vd),
2407 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2408 [(set DPR:$Vd, (TyD (IntOp (TyQ QPR:$Vm))))]>;
2410 // Long 2-register operations (currently only used for VMOVL).
2411 class N2VL<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2412 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2413 InstrItinClass itin, string OpcodeStr, string Dt,
2414 ValueType TyQ, ValueType TyD, SDNode OpNode>
2415 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$Vd),
2416 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2417 [(set QPR:$Vd, (TyQ (OpNode (TyD DPR:$Vm))))]>;
2419 // Long 2-register intrinsics.
2420 class N2VLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2421 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2422 InstrItinClass itin, string OpcodeStr, string Dt,
2423 ValueType TyQ, ValueType TyD, SDPatternOperator IntOp>
2424 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$Vd),
2425 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2426 [(set QPR:$Vd, (TyQ (IntOp (TyD DPR:$Vm))))]>;
2428 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
2429 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
2430 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$Vd, DPR:$Vm),
2431 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
2432 OpcodeStr, Dt, "$Vd, $Vm",
2433 "$src1 = $Vd, $src2 = $Vm", []>;
2434 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
2435 InstrItinClass itin, string OpcodeStr, string Dt>
2436 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$Vd, QPR:$Vm),
2437 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$Vd, $Vm",
2438 "$src1 = $Vd, $src2 = $Vm", []>;
2440 // Basic 3-register operations: double- and quad-register.
2441 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2442 InstrItinClass itin, string OpcodeStr, string Dt,
2443 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2444 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2445 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2446 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2447 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
2448 // All of these have a two-operand InstAlias.
2449 let TwoOperandAliasConstraint = "$Vn = $Vd";
2450 let isCommutable = Commutable;
2452 // Same as N3VD but no data type.
2453 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2454 InstrItinClass itin, string OpcodeStr,
2455 ValueType ResTy, ValueType OpTy,
2456 SDNode OpNode, bit Commutable>
2457 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
2458 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2459 OpcodeStr, "$Vd, $Vn, $Vm", "",
2460 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>{
2461 // All of these have a two-operand InstAlias.
2462 let TwoOperandAliasConstraint = "$Vn = $Vd";
2463 let isCommutable = Commutable;
2466 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
2467 InstrItinClass itin, string OpcodeStr, string Dt,
2468 ValueType Ty, SDNode ShOp>
2469 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2470 (outs DPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2471 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2473 (Ty (ShOp (Ty DPR:$Vn),
2474 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),imm:$lane)))))]> {
2475 // All of these have a two-operand InstAlias.
2476 let TwoOperandAliasConstraint = "$Vn = $Vd";
2477 let isCommutable = 0;
2479 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
2480 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
2481 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2482 (outs DPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2483 NVMulSLFrm, IIC_VMULi16D, OpcodeStr, Dt,"$Vd, $Vn, $Vm$lane","",
2485 (Ty (ShOp (Ty DPR:$Vn),
2486 (Ty (NEONvduplane (Ty DPR_8:$Vm), imm:$lane)))))]> {
2487 // All of these have a two-operand InstAlias.
2488 let TwoOperandAliasConstraint = "$Vn = $Vd";
2489 let isCommutable = 0;
2492 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2493 InstrItinClass itin, string OpcodeStr, string Dt,
2494 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2495 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2496 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2497 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2498 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
2499 // All of these have a two-operand InstAlias.
2500 let TwoOperandAliasConstraint = "$Vn = $Vd";
2501 let isCommutable = Commutable;
2503 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2504 InstrItinClass itin, string OpcodeStr,
2505 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2506 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
2507 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2508 OpcodeStr, "$Vd, $Vn, $Vm", "",
2509 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>{
2510 // All of these have a two-operand InstAlias.
2511 let TwoOperandAliasConstraint = "$Vn = $Vd";
2512 let isCommutable = Commutable;
2514 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
2515 InstrItinClass itin, string OpcodeStr, string Dt,
2516 ValueType ResTy, ValueType OpTy, SDNode ShOp>
2517 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2518 (outs QPR:$Vd), (ins QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2519 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2520 [(set (ResTy QPR:$Vd),
2521 (ResTy (ShOp (ResTy QPR:$Vn),
2522 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2524 // All of these have a two-operand InstAlias.
2525 let TwoOperandAliasConstraint = "$Vn = $Vd";
2526 let isCommutable = 0;
2528 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
2529 ValueType ResTy, ValueType OpTy, SDNode ShOp>
2530 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2531 (outs QPR:$Vd), (ins QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2532 NVMulSLFrm, IIC_VMULi16Q, OpcodeStr, Dt,"$Vd, $Vn, $Vm$lane", "",
2533 [(set (ResTy QPR:$Vd),
2534 (ResTy (ShOp (ResTy QPR:$Vn),
2535 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2537 // All of these have a two-operand InstAlias.
2538 let TwoOperandAliasConstraint = "$Vn = $Vd";
2539 let isCommutable = 0;
2542 // Basic 3-register intrinsics, both double- and quad-register.
2543 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2544 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2545 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp, bit Commutable>
2546 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2547 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), f, itin,
2548 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2549 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
2550 // All of these have a two-operand InstAlias.
2551 let TwoOperandAliasConstraint = "$Vn = $Vd";
2552 let isCommutable = Commutable;
2555 class N3VDIntnp<bits<5> op27_23, bits<2> op21_20, bits<4> op11_8, bit op6,
2556 bit op4, Format f, InstrItinClass itin, string OpcodeStr,
2557 string Dt, ValueType ResTy, ValueType OpTy,
2558 SDPatternOperator IntOp, bit Commutable>
2559 : N3Vnp<op27_23, op21_20, op11_8, op6, op4,
2560 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin, OpcodeStr, Dt,
2561 ResTy, OpTy, IntOp, Commutable,
2562 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>;
2564 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2565 string OpcodeStr, string Dt, ValueType Ty, SDPatternOperator IntOp>
2566 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2567 (outs DPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2568 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2570 (Ty (IntOp (Ty DPR:$Vn),
2571 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),
2573 let isCommutable = 0;
2576 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2577 string OpcodeStr, string Dt, ValueType Ty, SDPatternOperator IntOp>
2578 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2579 (outs DPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2580 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2582 (Ty (IntOp (Ty DPR:$Vn),
2583 (Ty (NEONvduplane (Ty DPR_8:$Vm), imm:$lane)))))]> {
2584 let isCommutable = 0;
2586 class N3VDIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2587 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2588 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2589 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2590 (outs DPR:$Vd), (ins DPR:$Vm, DPR:$Vn), f, itin,
2591 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
2592 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (OpTy DPR:$Vn))))]> {
2593 let TwoOperandAliasConstraint = "$Vm = $Vd";
2594 let isCommutable = 0;
2597 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2598 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2599 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp, bit Commutable>
2600 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2601 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), f, itin,
2602 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2603 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
2604 // All of these have a two-operand InstAlias.
2605 let TwoOperandAliasConstraint = "$Vn = $Vd";
2606 let isCommutable = Commutable;
2609 class N3VQIntnp<bits<5> op27_23, bits<2> op21_20, bits<4> op11_8, bit op6,
2610 bit op4, Format f, InstrItinClass itin, string OpcodeStr,
2611 string Dt, ValueType ResTy, ValueType OpTy,
2612 SDPatternOperator IntOp, bit Commutable>
2613 : N3Vnp<op27_23, op21_20, op11_8, op6, op4,
2614 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), f, itin, OpcodeStr, Dt,
2615 ResTy, OpTy, IntOp, Commutable,
2616 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>;
2618 // Same as N3VQIntnp but with Vd as a src register.
2619 class N3VQInt3np<bits<5> op27_23, bits<2> op21_20, bits<4> op11_8, bit op6,
2620 bit op4, Format f, InstrItinClass itin, string OpcodeStr,
2621 string Dt, ValueType ResTy, ValueType OpTy,
2622 SDPatternOperator IntOp, bit Commutable>
2623 : N3Vnp<op27_23, op21_20, op11_8, op6, op4,
2624 (outs QPR:$Vd), (ins QPR:$src, QPR:$Vn, QPR:$Vm), f, itin, OpcodeStr,
2625 Dt, ResTy, OpTy, IntOp, Commutable,
2626 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$src), (OpTy QPR:$Vn),
2627 (OpTy QPR:$Vm))))]> {
2628 let Constraints = "$src = $Vd";
2631 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2632 string OpcodeStr, string Dt,
2633 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2634 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2635 (outs QPR:$Vd), (ins QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2636 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2637 [(set (ResTy QPR:$Vd),
2638 (ResTy (IntOp (ResTy QPR:$Vn),
2639 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2641 let isCommutable = 0;
2643 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2644 string OpcodeStr, string Dt,
2645 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2646 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2647 (outs QPR:$Vd), (ins QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2648 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2649 [(set (ResTy QPR:$Vd),
2650 (ResTy (IntOp (ResTy QPR:$Vn),
2651 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2653 let isCommutable = 0;
2655 class N3VQIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2656 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2657 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2658 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2659 (outs QPR:$Vd), (ins QPR:$Vm, QPR:$Vn), f, itin,
2660 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
2661 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (OpTy QPR:$Vn))))]> {
2662 let TwoOperandAliasConstraint = "$Vm = $Vd";
2663 let isCommutable = 0;
2666 // Multiply-Add/Sub operations: double- and quad-register.
2667 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2668 InstrItinClass itin, string OpcodeStr, string Dt,
2669 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator OpNode>
2670 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2671 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2672 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2673 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
2674 (Ty (MulOp DPR:$Vn, DPR:$Vm)))))]>;
2676 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2677 string OpcodeStr, string Dt,
2678 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator ShOp>
2679 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2681 (ins DPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2683 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2685 (Ty (ShOp (Ty DPR:$src1),
2687 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),
2689 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2690 string OpcodeStr, string Dt,
2691 ValueType Ty, SDNode MulOp, SDNode ShOp>
2692 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2694 (ins DPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2696 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2698 (Ty (ShOp (Ty DPR:$src1),
2700 (Ty (NEONvduplane (Ty DPR_8:$Vm),
2703 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2704 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
2705 SDPatternOperator MulOp, SDPatternOperator OpNode>
2706 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2707 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2708 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2709 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
2710 (Ty (MulOp QPR:$Vn, QPR:$Vm)))))]>;
2711 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2712 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2713 SDPatternOperator MulOp, SDPatternOperator ShOp>
2714 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2716 (ins QPR:$src1, QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2718 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2719 [(set (ResTy QPR:$Vd),
2720 (ResTy (ShOp (ResTy QPR:$src1),
2721 (ResTy (MulOp QPR:$Vn,
2722 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2724 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2725 string OpcodeStr, string Dt,
2726 ValueType ResTy, ValueType OpTy,
2727 SDNode MulOp, SDNode ShOp>
2728 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2730 (ins QPR:$src1, QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2732 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2733 [(set (ResTy QPR:$Vd),
2734 (ResTy (ShOp (ResTy QPR:$src1),
2735 (ResTy (MulOp QPR:$Vn,
2736 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2739 // Neon Intrinsic-Op instructions (VABA): double- and quad-register.
2740 class N3VDIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2741 InstrItinClass itin, string OpcodeStr, string Dt,
2742 ValueType Ty, SDPatternOperator IntOp, SDNode OpNode>
2743 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2744 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2745 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2746 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
2747 (Ty (IntOp (Ty DPR:$Vn), (Ty DPR:$Vm))))))]>;
2748 class N3VQIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2749 InstrItinClass itin, string OpcodeStr, string Dt,
2750 ValueType Ty, SDPatternOperator IntOp, SDNode OpNode>
2751 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2752 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2753 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2754 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
2755 (Ty (IntOp (Ty QPR:$Vn), (Ty QPR:$Vm))))))]>;
2757 // Neon 3-argument intrinsics, both double- and quad-register.
2758 // The destination register is also used as the first source operand register.
2759 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2760 InstrItinClass itin, string OpcodeStr, string Dt,
2761 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2762 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2763 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2764 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2765 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$src1),
2766 (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>;
2767 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2768 InstrItinClass itin, string OpcodeStr, string Dt,
2769 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2770 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2771 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2772 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2773 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$src1),
2774 (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>;
2776 // Long Multiply-Add/Sub operations.
2777 class N3VLMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2778 InstrItinClass itin, string OpcodeStr, string Dt,
2779 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2780 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2781 (outs QPR:$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:$src1),
2784 (TyQ (MulOp (TyD DPR:$Vn),
2785 (TyD DPR:$Vm)))))]>;
2786 class N3VLMulOpSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2787 InstrItinClass itin, string OpcodeStr, string Dt,
2788 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2789 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$Vd),
2790 (ins QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2792 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2794 (OpNode (TyQ QPR:$src1),
2795 (TyQ (MulOp (TyD DPR:$Vn),
2796 (TyD (NEONvduplane (TyD DPR_VFP2:$Vm),
2798 class N3VLMulOpSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2799 InstrItinClass itin, string OpcodeStr, string Dt,
2800 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2801 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$Vd),
2802 (ins QPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2804 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2806 (OpNode (TyQ QPR:$src1),
2807 (TyQ (MulOp (TyD DPR:$Vn),
2808 (TyD (NEONvduplane (TyD DPR_8:$Vm),
2811 // Long Intrinsic-Op vector operations with explicit extend (VABAL).
2812 class N3VLIntExtOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2813 InstrItinClass itin, string OpcodeStr, string Dt,
2814 ValueType TyQ, ValueType TyD, SDPatternOperator IntOp, SDNode ExtOp,
2816 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2817 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2818 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2819 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
2820 (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
2821 (TyD DPR:$Vm)))))))]>;
2823 // Neon Long 3-argument intrinsic. The destination register is
2824 // a quad-register and is also used as the first source operand register.
2825 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2826 InstrItinClass itin, string OpcodeStr, string Dt,
2827 ValueType TyQ, ValueType TyD, SDPatternOperator IntOp>
2828 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2829 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2830 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2832 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$Vn), (TyD DPR:$Vm))))]>;
2833 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2834 string OpcodeStr, string Dt,
2835 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2836 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2838 (ins QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2840 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2841 [(set (ResTy QPR:$Vd),
2842 (ResTy (IntOp (ResTy QPR:$src1),
2844 (OpTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2846 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2847 InstrItinClass itin, string OpcodeStr, string Dt,
2848 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2849 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2851 (ins QPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2853 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2854 [(set (ResTy QPR:$Vd),
2855 (ResTy (IntOp (ResTy QPR:$src1),
2857 (OpTy (NEONvduplane (OpTy DPR_8:$Vm),
2860 // Narrowing 3-register intrinsics.
2861 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2862 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
2863 SDPatternOperator IntOp, bit Commutable>
2864 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2865 (outs DPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINi4D,
2866 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2867 [(set DPR:$Vd, (TyD (IntOp (TyQ QPR:$Vn), (TyQ QPR:$Vm))))]> {
2868 let isCommutable = Commutable;
2871 // Long 3-register operations.
2872 class N3VL<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2873 InstrItinClass itin, string OpcodeStr, string Dt,
2874 ValueType TyQ, ValueType TyD, SDNode OpNode, bit Commutable>
2875 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2876 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2877 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2878 [(set QPR:$Vd, (TyQ (OpNode (TyD DPR:$Vn), (TyD DPR:$Vm))))]> {
2879 let isCommutable = Commutable;
2882 class N3VLSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2883 InstrItinClass itin, string OpcodeStr, string Dt,
2884 ValueType TyQ, ValueType TyD, SDNode OpNode>
2885 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2886 (outs QPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2887 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2889 (TyQ (OpNode (TyD DPR:$Vn),
2890 (TyD (NEONvduplane (TyD DPR_VFP2:$Vm),imm:$lane)))))]>;
2891 class N3VLSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2892 InstrItinClass itin, string OpcodeStr, string Dt,
2893 ValueType TyQ, ValueType TyD, SDNode OpNode>
2894 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2895 (outs QPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2896 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2898 (TyQ (OpNode (TyD DPR:$Vn),
2899 (TyD (NEONvduplane (TyD DPR_8:$Vm), imm:$lane)))))]>;
2901 // Long 3-register operations with explicitly extended operands.
2902 class N3VLExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2903 InstrItinClass itin, string OpcodeStr, string Dt,
2904 ValueType TyQ, ValueType TyD, SDNode OpNode, SDNode ExtOp,
2906 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2907 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2908 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2909 [(set QPR:$Vd, (OpNode (TyQ (ExtOp (TyD DPR:$Vn))),
2910 (TyQ (ExtOp (TyD DPR:$Vm)))))]> {
2911 let isCommutable = Commutable;
2914 // Long 3-register intrinsics with explicit extend (VABDL).
2915 class N3VLIntExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2916 InstrItinClass itin, string OpcodeStr, string Dt,
2917 ValueType TyQ, ValueType TyD, SDPatternOperator IntOp, SDNode ExtOp,
2919 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2920 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2921 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2922 [(set QPR:$Vd, (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
2923 (TyD DPR:$Vm))))))]> {
2924 let isCommutable = Commutable;
2927 // Long 3-register intrinsics.
2928 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2929 InstrItinClass itin, string OpcodeStr, string Dt,
2930 ValueType TyQ, ValueType TyD, SDPatternOperator IntOp, bit Commutable>
2931 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2932 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2933 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2934 [(set QPR:$Vd, (TyQ (IntOp (TyD DPR:$Vn), (TyD DPR:$Vm))))]> {
2935 let isCommutable = Commutable;
2938 // Same as above, but not predicated.
2939 class N3VLIntnp<bits<5> op27_23, bits<2> op21_20, bits<4> op11_8, bit op6,
2940 bit op4, InstrItinClass itin, string OpcodeStr,
2941 string Dt, ValueType ResTy, ValueType OpTy,
2942 SDPatternOperator IntOp, bit Commutable>
2943 : N3Vnp<op27_23, op21_20, op11_8, op6, op4,
2944 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin, OpcodeStr, Dt,
2945 ResTy, OpTy, IntOp, Commutable,
2946 [(set QPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>;
2948 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2949 string OpcodeStr, string Dt,
2950 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2951 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2952 (outs QPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2953 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2954 [(set (ResTy QPR:$Vd),
2955 (ResTy (IntOp (OpTy DPR:$Vn),
2956 (OpTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2958 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2959 InstrItinClass itin, string OpcodeStr, string Dt,
2960 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2961 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2962 (outs QPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2963 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2964 [(set (ResTy QPR:$Vd),
2965 (ResTy (IntOp (OpTy DPR:$Vn),
2966 (OpTy (NEONvduplane (OpTy DPR_8:$Vm),
2969 // Wide 3-register operations.
2970 class N3VW<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2971 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
2972 SDNode OpNode, SDNode ExtOp, bit Commutable>
2973 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2974 (outs QPR:$Vd), (ins QPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VSUBiD,
2975 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2976 [(set QPR:$Vd, (OpNode (TyQ QPR:$Vn),
2977 (TyQ (ExtOp (TyD DPR:$Vm)))))]> {
2978 // All of these have a two-operand InstAlias.
2979 let TwoOperandAliasConstraint = "$Vn = $Vd";
2980 let isCommutable = Commutable;
2983 // Pairwise long 2-register intrinsics, both double- and quad-register.
2984 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2985 bits<2> op17_16, bits<5> op11_7, bit op4,
2986 string OpcodeStr, string Dt,
2987 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2988 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2989 (ins DPR:$Vm), IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
2990 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
2991 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2992 bits<2> op17_16, bits<5> op11_7, bit op4,
2993 string OpcodeStr, string Dt,
2994 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2995 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2996 (ins QPR:$Vm), IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
2997 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
2999 // Pairwise long 2-register accumulate intrinsics,
3000 // both double- and quad-register.
3001 // The destination register is also used as the first source operand register.
3002 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
3003 bits<2> op17_16, bits<5> op11_7, bit op4,
3004 string OpcodeStr, string Dt,
3005 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
3006 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
3007 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vm), IIC_VPALiD,
3008 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
3009 [(set DPR:$Vd, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$Vm))))]>;
3010 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
3011 bits<2> op17_16, bits<5> op11_7, bit op4,
3012 string OpcodeStr, string Dt,
3013 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
3014 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
3015 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vm), IIC_VPALiQ,
3016 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
3017 [(set QPR:$Vd, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$Vm))))]>;
3019 // Shift by immediate,
3020 // both double- and quad-register.
3021 let TwoOperandAliasConstraint = "$Vm = $Vd" in {
3022 class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
3023 Format f, InstrItinClass itin, Operand ImmTy,
3024 string OpcodeStr, string Dt, ValueType Ty, SDNode OpNode>
3025 : N2VImm<op24, op23, op11_8, op7, 0, op4,
3026 (outs DPR:$Vd), (ins DPR:$Vm, ImmTy:$SIMM), f, itin,
3027 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
3028 [(set DPR:$Vd, (Ty (OpNode (Ty DPR:$Vm), (i32 imm:$SIMM))))]>;
3029 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
3030 Format f, InstrItinClass itin, Operand ImmTy,
3031 string OpcodeStr, string Dt, ValueType Ty, SDNode OpNode>
3032 : N2VImm<op24, op23, op11_8, op7, 1, op4,
3033 (outs QPR:$Vd), (ins QPR:$Vm, ImmTy:$SIMM), f, itin,
3034 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
3035 [(set QPR:$Vd, (Ty (OpNode (Ty QPR:$Vm), (i32 imm:$SIMM))))]>;
3038 // Long shift by immediate.
3039 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
3040 string OpcodeStr, string Dt,
3041 ValueType ResTy, ValueType OpTy, Operand ImmTy, SDNode OpNode>
3042 : N2VImm<op24, op23, op11_8, op7, op6, op4,
3043 (outs QPR:$Vd), (ins DPR:$Vm, ImmTy:$SIMM), N2RegVShLFrm,
3044 IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
3045 [(set QPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm),
3046 (i32 imm:$SIMM))))]>;
3048 // Narrow shift by immediate.
3049 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
3050 InstrItinClass itin, string OpcodeStr, string Dt,
3051 ValueType ResTy, ValueType OpTy, Operand ImmTy,
3052 SDPatternOperator OpNode>
3053 : N2VImm<op24, op23, op11_8, op7, op6, op4,
3054 (outs DPR:$Vd), (ins QPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, itin,
3055 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
3056 [(set DPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vm),
3057 (i32 ImmTy:$SIMM))))]>;
3059 // Shift right by immediate and accumulate,
3060 // both double- and quad-register.
3061 let TwoOperandAliasConstraint = "$Vm = $Vd" in {
3062 class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
3063 Operand ImmTy, string OpcodeStr, string Dt,
3064 ValueType Ty, SDNode ShOp>
3065 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
3066 (ins DPR:$src1, DPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, IIC_VPALiD,
3067 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
3068 [(set DPR:$Vd, (Ty (add DPR:$src1,
3069 (Ty (ShOp DPR:$Vm, (i32 imm:$SIMM))))))]>;
3070 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
3071 Operand ImmTy, string OpcodeStr, string Dt,
3072 ValueType Ty, SDNode ShOp>
3073 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
3074 (ins QPR:$src1, QPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, IIC_VPALiD,
3075 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
3076 [(set QPR:$Vd, (Ty (add QPR:$src1,
3077 (Ty (ShOp QPR:$Vm, (i32 imm:$SIMM))))))]>;
3080 // Shift by immediate and insert,
3081 // both double- and quad-register.
3082 let TwoOperandAliasConstraint = "$Vm = $Vd" in {
3083 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
3084 Operand ImmTy, Format f, string OpcodeStr, string Dt,
3085 ValueType Ty,SDNode ShOp>
3086 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
3087 (ins DPR:$src1, DPR:$Vm, ImmTy:$SIMM), f, IIC_VSHLiD,
3088 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
3089 [(set DPR:$Vd, (Ty (ShOp DPR:$src1, DPR:$Vm, (i32 imm:$SIMM))))]>;
3090 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
3091 Operand ImmTy, Format f, string OpcodeStr, string Dt,
3092 ValueType Ty,SDNode ShOp>
3093 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
3094 (ins QPR:$src1, QPR:$Vm, ImmTy:$SIMM), f, IIC_VSHLiQ,
3095 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
3096 [(set QPR:$Vd, (Ty (ShOp QPR:$src1, QPR:$Vm, (i32 imm:$SIMM))))]>;
3099 // Convert, with fractional bits immediate,
3100 // both double- and quad-register.
3101 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
3102 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
3103 SDPatternOperator IntOp>
3104 : N2VImm<op24, op23, op11_8, op7, 0, op4,
3105 (outs DPR:$Vd), (ins DPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
3106 IIC_VUNAD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
3107 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (i32 imm:$SIMM))))]>;
3108 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
3109 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
3110 SDPatternOperator IntOp>
3111 : N2VImm<op24, op23, op11_8, op7, 1, op4,
3112 (outs QPR:$Vd), (ins QPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
3113 IIC_VUNAQ, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
3114 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (i32 imm:$SIMM))))]>;
3116 //===----------------------------------------------------------------------===//
3118 //===----------------------------------------------------------------------===//
3120 // Abbreviations used in multiclass suffixes:
3121 // Q = quarter int (8 bit) elements
3122 // H = half int (16 bit) elements
3123 // S = single int (32 bit) elements
3124 // D = double int (64 bit) elements
3126 // Neon 2-register vector operations and intrinsics.
3128 // Neon 2-register comparisons.
3129 // source operand element sizes of 8, 16 and 32 bits:
3130 multiclass N2V_QHS_cmp<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3131 bits<5> op11_7, bit op4, string opc, string Dt,
3132 string asm, SDNode OpNode> {
3133 // 64-bit vector types.
3134 def v8i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 0, op4,
3135 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
3136 opc, !strconcat(Dt, "8"), asm, "",
3137 [(set DPR:$Vd, (v8i8 (OpNode (v8i8 DPR:$Vm))))]>;
3138 def v4i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
3139 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
3140 opc, !strconcat(Dt, "16"), asm, "",
3141 [(set DPR:$Vd, (v4i16 (OpNode (v4i16 DPR:$Vm))))]>;
3142 def v2i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
3143 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
3144 opc, !strconcat(Dt, "32"), asm, "",
3145 [(set DPR:$Vd, (v2i32 (OpNode (v2i32 DPR:$Vm))))]>;
3146 def v2f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
3147 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
3148 opc, "f32", asm, "",
3149 [(set DPR:$Vd, (v2i32 (OpNode (v2f32 DPR:$Vm))))]> {
3150 let Inst{10} = 1; // overwrite F = 1
3153 // 128-bit vector types.
3154 def v16i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 1, op4,
3155 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
3156 opc, !strconcat(Dt, "8"), asm, "",
3157 [(set QPR:$Vd, (v16i8 (OpNode (v16i8 QPR:$Vm))))]>;
3158 def v8i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
3159 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
3160 opc, !strconcat(Dt, "16"), asm, "",
3161 [(set QPR:$Vd, (v8i16 (OpNode (v8i16 QPR:$Vm))))]>;
3162 def v4i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
3163 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
3164 opc, !strconcat(Dt, "32"), asm, "",
3165 [(set QPR:$Vd, (v4i32 (OpNode (v4i32 QPR:$Vm))))]>;
3166 def v4f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
3167 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
3168 opc, "f32", asm, "",
3169 [(set QPR:$Vd, (v4i32 (OpNode (v4f32 QPR:$Vm))))]> {
3170 let Inst{10} = 1; // overwrite F = 1
3175 // Neon 2-register vector intrinsics,
3176 // element sizes of 8, 16 and 32 bits:
3177 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3178 bits<5> op11_7, bit op4,
3179 InstrItinClass itinD, InstrItinClass itinQ,
3180 string OpcodeStr, string Dt, SDPatternOperator IntOp> {
3181 // 64-bit vector types.
3182 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3183 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
3184 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3185 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
3186 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3187 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
3189 // 128-bit vector types.
3190 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3191 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
3192 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3193 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
3194 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3195 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
3199 // Neon Narrowing 2-register vector operations,
3200 // source operand element sizes of 16, 32 and 64 bits:
3201 multiclass N2VN_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3202 bits<5> op11_7, bit op6, bit op4,
3203 InstrItinClass itin, string OpcodeStr, string Dt,
3205 def v8i8 : N2VN<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
3206 itin, OpcodeStr, !strconcat(Dt, "16"),
3207 v8i8, v8i16, OpNode>;
3208 def v4i16 : N2VN<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
3209 itin, OpcodeStr, !strconcat(Dt, "32"),
3210 v4i16, v4i32, OpNode>;
3211 def v2i32 : N2VN<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
3212 itin, OpcodeStr, !strconcat(Dt, "64"),
3213 v2i32, v2i64, OpNode>;
3216 // Neon Narrowing 2-register vector intrinsics,
3217 // source operand element sizes of 16, 32 and 64 bits:
3218 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3219 bits<5> op11_7, bit op6, bit op4,
3220 InstrItinClass itin, string OpcodeStr, string Dt,
3221 SDPatternOperator IntOp> {
3222 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
3223 itin, OpcodeStr, !strconcat(Dt, "16"),
3224 v8i8, v8i16, IntOp>;
3225 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
3226 itin, OpcodeStr, !strconcat(Dt, "32"),
3227 v4i16, v4i32, IntOp>;
3228 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
3229 itin, OpcodeStr, !strconcat(Dt, "64"),
3230 v2i32, v2i64, IntOp>;
3234 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
3235 // source operand element sizes of 16, 32 and 64 bits:
3236 multiclass N2VL_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
3237 string OpcodeStr, string Dt, SDNode OpNode> {
3238 def v8i16 : N2VL<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
3239 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode>;
3240 def v4i32 : N2VL<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
3241 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
3242 def v2i64 : N2VL<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
3243 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
3247 // Neon 3-register vector operations.
3249 // First with only element sizes of 8, 16 and 32 bits:
3250 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3251 InstrItinClass itinD16, InstrItinClass itinD32,
3252 InstrItinClass itinQ16, InstrItinClass itinQ32,
3253 string OpcodeStr, string Dt,
3254 SDNode OpNode, bit Commutable = 0> {
3255 // 64-bit vector types.
3256 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
3257 OpcodeStr, !strconcat(Dt, "8"),
3258 v8i8, v8i8, OpNode, Commutable>;
3259 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
3260 OpcodeStr, !strconcat(Dt, "16"),
3261 v4i16, v4i16, OpNode, Commutable>;
3262 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
3263 OpcodeStr, !strconcat(Dt, "32"),
3264 v2i32, v2i32, OpNode, Commutable>;
3266 // 128-bit vector types.
3267 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
3268 OpcodeStr, !strconcat(Dt, "8"),
3269 v16i8, v16i8, OpNode, Commutable>;
3270 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
3271 OpcodeStr, !strconcat(Dt, "16"),
3272 v8i16, v8i16, OpNode, Commutable>;
3273 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
3274 OpcodeStr, !strconcat(Dt, "32"),
3275 v4i32, v4i32, OpNode, Commutable>;
3278 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, SDNode ShOp> {
3279 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, "i16", v4i16, ShOp>;
3280 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, "i32", v2i32, ShOp>;
3281 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, "i16", v8i16, v4i16, ShOp>;
3282 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, "i32",
3283 v4i32, v2i32, ShOp>;
3286 // ....then also with element size 64 bits:
3287 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3288 InstrItinClass itinD, InstrItinClass itinQ,
3289 string OpcodeStr, string Dt,
3290 SDNode OpNode, bit Commutable = 0>
3291 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
3292 OpcodeStr, Dt, OpNode, Commutable> {
3293 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
3294 OpcodeStr, !strconcat(Dt, "64"),
3295 v1i64, v1i64, OpNode, Commutable>;
3296 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
3297 OpcodeStr, !strconcat(Dt, "64"),
3298 v2i64, v2i64, OpNode, Commutable>;
3302 // Neon 3-register vector intrinsics.
3304 // First with only element sizes of 16 and 32 bits:
3305 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3306 InstrItinClass itinD16, InstrItinClass itinD32,
3307 InstrItinClass itinQ16, InstrItinClass itinQ32,
3308 string OpcodeStr, string Dt,
3309 SDPatternOperator IntOp, bit Commutable = 0> {
3310 // 64-bit vector types.
3311 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, f, itinD16,
3312 OpcodeStr, !strconcat(Dt, "16"),
3313 v4i16, v4i16, IntOp, Commutable>;
3314 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, f, itinD32,
3315 OpcodeStr, !strconcat(Dt, "32"),
3316 v2i32, v2i32, IntOp, Commutable>;
3318 // 128-bit vector types.
3319 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, f, itinQ16,
3320 OpcodeStr, !strconcat(Dt, "16"),
3321 v8i16, v8i16, IntOp, Commutable>;
3322 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, f, itinQ32,
3323 OpcodeStr, !strconcat(Dt, "32"),
3324 v4i32, v4i32, IntOp, Commutable>;
3326 multiclass N3VInt_HSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3327 InstrItinClass itinD16, InstrItinClass itinD32,
3328 InstrItinClass itinQ16, InstrItinClass itinQ32,
3329 string OpcodeStr, string Dt,
3330 SDPatternOperator IntOp> {
3331 // 64-bit vector types.
3332 def v4i16 : N3VDIntSh<op24, op23, 0b01, op11_8, op4, f, itinD16,
3333 OpcodeStr, !strconcat(Dt, "16"),
3334 v4i16, v4i16, IntOp>;
3335 def v2i32 : N3VDIntSh<op24, op23, 0b10, op11_8, op4, f, itinD32,
3336 OpcodeStr, !strconcat(Dt, "32"),
3337 v2i32, v2i32, IntOp>;
3339 // 128-bit vector types.
3340 def v8i16 : N3VQIntSh<op24, op23, 0b01, op11_8, op4, f, itinQ16,
3341 OpcodeStr, !strconcat(Dt, "16"),
3342 v8i16, v8i16, IntOp>;
3343 def v4i32 : N3VQIntSh<op24, op23, 0b10, op11_8, op4, f, itinQ32,
3344 OpcodeStr, !strconcat(Dt, "32"),
3345 v4i32, v4i32, IntOp>;
3348 multiclass N3VIntSL_HS<bits<4> op11_8,
3349 InstrItinClass itinD16, InstrItinClass itinD32,
3350 InstrItinClass itinQ16, InstrItinClass itinQ32,
3351 string OpcodeStr, string Dt, SDPatternOperator IntOp> {
3352 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
3353 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
3354 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
3355 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
3356 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
3357 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
3358 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
3359 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
3362 // ....then also with element size of 8 bits:
3363 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3364 InstrItinClass itinD16, InstrItinClass itinD32,
3365 InstrItinClass itinQ16, InstrItinClass itinQ32,
3366 string OpcodeStr, string Dt,
3367 SDPatternOperator IntOp, bit Commutable = 0>
3368 : N3VInt_HS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3369 OpcodeStr, Dt, IntOp, Commutable> {
3370 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, f, itinD16,
3371 OpcodeStr, !strconcat(Dt, "8"),
3372 v8i8, v8i8, IntOp, Commutable>;
3373 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, f, itinQ16,
3374 OpcodeStr, !strconcat(Dt, "8"),
3375 v16i8, v16i8, IntOp, Commutable>;
3377 multiclass N3VInt_QHSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3378 InstrItinClass itinD16, InstrItinClass itinD32,
3379 InstrItinClass itinQ16, InstrItinClass itinQ32,
3380 string OpcodeStr, string Dt,
3381 SDPatternOperator IntOp>
3382 : N3VInt_HSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3383 OpcodeStr, Dt, IntOp> {
3384 def v8i8 : N3VDIntSh<op24, op23, 0b00, op11_8, op4, f, itinD16,
3385 OpcodeStr, !strconcat(Dt, "8"),
3387 def v16i8 : N3VQIntSh<op24, op23, 0b00, op11_8, op4, f, itinQ16,
3388 OpcodeStr, !strconcat(Dt, "8"),
3389 v16i8, v16i8, IntOp>;
3393 // ....then also with element size of 64 bits:
3394 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3395 InstrItinClass itinD16, InstrItinClass itinD32,
3396 InstrItinClass itinQ16, InstrItinClass itinQ32,
3397 string OpcodeStr, string Dt,
3398 SDPatternOperator IntOp, bit Commutable = 0>
3399 : N3VInt_QHS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3400 OpcodeStr, Dt, IntOp, Commutable> {
3401 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, f, itinD32,
3402 OpcodeStr, !strconcat(Dt, "64"),
3403 v1i64, v1i64, IntOp, Commutable>;
3404 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, f, itinQ32,
3405 OpcodeStr, !strconcat(Dt, "64"),
3406 v2i64, v2i64, IntOp, Commutable>;
3408 multiclass N3VInt_QHSDSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3409 InstrItinClass itinD16, InstrItinClass itinD32,
3410 InstrItinClass itinQ16, InstrItinClass itinQ32,
3411 string OpcodeStr, string Dt,
3412 SDPatternOperator IntOp>
3413 : N3VInt_QHSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3414 OpcodeStr, Dt, IntOp> {
3415 def v1i64 : N3VDIntSh<op24, op23, 0b11, op11_8, op4, f, itinD32,
3416 OpcodeStr, !strconcat(Dt, "64"),
3417 v1i64, v1i64, IntOp>;
3418 def v2i64 : N3VQIntSh<op24, op23, 0b11, op11_8, op4, f, itinQ32,
3419 OpcodeStr, !strconcat(Dt, "64"),
3420 v2i64, v2i64, IntOp>;
3423 // Neon Narrowing 3-register vector intrinsics,
3424 // source operand element sizes of 16, 32 and 64 bits:
3425 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3426 string OpcodeStr, string Dt,
3427 SDPatternOperator IntOp, bit Commutable = 0> {
3428 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
3429 OpcodeStr, !strconcat(Dt, "16"),
3430 v8i8, v8i16, IntOp, Commutable>;
3431 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
3432 OpcodeStr, !strconcat(Dt, "32"),
3433 v4i16, v4i32, IntOp, Commutable>;
3434 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
3435 OpcodeStr, !strconcat(Dt, "64"),
3436 v2i32, v2i64, IntOp, Commutable>;
3440 // Neon Long 3-register vector operations.
3442 multiclass N3VL_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3443 InstrItinClass itin16, InstrItinClass itin32,
3444 string OpcodeStr, string Dt,
3445 SDNode OpNode, bit Commutable = 0> {
3446 def v8i16 : N3VL<op24, op23, 0b00, op11_8, op4, itin16,
3447 OpcodeStr, !strconcat(Dt, "8"),
3448 v8i16, v8i8, OpNode, Commutable>;
3449 def v4i32 : N3VL<op24, op23, 0b01, op11_8, op4, itin16,
3450 OpcodeStr, !strconcat(Dt, "16"),
3451 v4i32, v4i16, OpNode, Commutable>;
3452 def v2i64 : N3VL<op24, op23, 0b10, op11_8, op4, itin32,
3453 OpcodeStr, !strconcat(Dt, "32"),
3454 v2i64, v2i32, OpNode, Commutable>;
3457 multiclass N3VLSL_HS<bit op24, bits<4> op11_8,
3458 InstrItinClass itin, string OpcodeStr, string Dt,
3460 def v4i16 : N3VLSL16<op24, 0b01, op11_8, itin, OpcodeStr,
3461 !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
3462 def v2i32 : N3VLSL<op24, 0b10, op11_8, itin, OpcodeStr,
3463 !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
3466 multiclass N3VLExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3467 InstrItinClass itin16, InstrItinClass itin32,
3468 string OpcodeStr, string Dt,
3469 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
3470 def v8i16 : N3VLExt<op24, op23, 0b00, op11_8, op4, itin16,
3471 OpcodeStr, !strconcat(Dt, "8"),
3472 v8i16, v8i8, OpNode, ExtOp, Commutable>;
3473 def v4i32 : N3VLExt<op24, op23, 0b01, op11_8, op4, itin16,
3474 OpcodeStr, !strconcat(Dt, "16"),
3475 v4i32, v4i16, OpNode, ExtOp, Commutable>;
3476 def v2i64 : N3VLExt<op24, op23, 0b10, op11_8, op4, itin32,
3477 OpcodeStr, !strconcat(Dt, "32"),
3478 v2i64, v2i32, OpNode, ExtOp, Commutable>;
3481 // Neon Long 3-register vector intrinsics.
3483 // First with only element sizes of 16 and 32 bits:
3484 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
3485 InstrItinClass itin16, InstrItinClass itin32,
3486 string OpcodeStr, string Dt,
3487 SDPatternOperator IntOp, bit Commutable = 0> {
3488 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin16,
3489 OpcodeStr, !strconcat(Dt, "16"),
3490 v4i32, v4i16, IntOp, Commutable>;
3491 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin32,
3492 OpcodeStr, !strconcat(Dt, "32"),
3493 v2i64, v2i32, IntOp, Commutable>;
3496 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
3497 InstrItinClass itin, string OpcodeStr, string Dt,
3498 SDPatternOperator IntOp> {
3499 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
3500 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
3501 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
3502 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3505 // ....then also with element size of 8 bits:
3506 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3507 InstrItinClass itin16, InstrItinClass itin32,
3508 string OpcodeStr, string Dt,
3509 SDPatternOperator IntOp, bit Commutable = 0>
3510 : N3VLInt_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt,
3511 IntOp, Commutable> {
3512 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin16,
3513 OpcodeStr, !strconcat(Dt, "8"),
3514 v8i16, v8i8, IntOp, Commutable>;
3517 // ....with explicit extend (VABDL).
3518 multiclass N3VLIntExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3519 InstrItinClass itin, string OpcodeStr, string Dt,
3520 SDPatternOperator IntOp, SDNode ExtOp, bit Commutable = 0> {
3521 def v8i16 : N3VLIntExt<op24, op23, 0b00, op11_8, op4, itin,
3522 OpcodeStr, !strconcat(Dt, "8"),
3523 v8i16, v8i8, IntOp, ExtOp, Commutable>;
3524 def v4i32 : N3VLIntExt<op24, op23, 0b01, op11_8, op4, itin,
3525 OpcodeStr, !strconcat(Dt, "16"),
3526 v4i32, v4i16, IntOp, ExtOp, Commutable>;
3527 def v2i64 : N3VLIntExt<op24, op23, 0b10, op11_8, op4, itin,
3528 OpcodeStr, !strconcat(Dt, "32"),
3529 v2i64, v2i32, IntOp, ExtOp, Commutable>;
3533 // Neon Wide 3-register vector intrinsics,
3534 // source operand element sizes of 8, 16 and 32 bits:
3535 multiclass N3VW_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3536 string OpcodeStr, string Dt,
3537 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
3538 def v8i16 : N3VW<op24, op23, 0b00, op11_8, op4,
3539 OpcodeStr, !strconcat(Dt, "8"),
3540 v8i16, v8i8, OpNode, ExtOp, Commutable>;
3541 def v4i32 : N3VW<op24, op23, 0b01, op11_8, op4,
3542 OpcodeStr, !strconcat(Dt, "16"),
3543 v4i32, v4i16, OpNode, ExtOp, Commutable>;
3544 def v2i64 : N3VW<op24, op23, 0b10, op11_8, op4,
3545 OpcodeStr, !strconcat(Dt, "32"),
3546 v2i64, v2i32, OpNode, ExtOp, Commutable>;
3550 // Neon Multiply-Op vector operations,
3551 // element sizes of 8, 16 and 32 bits:
3552 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3553 InstrItinClass itinD16, InstrItinClass itinD32,
3554 InstrItinClass itinQ16, InstrItinClass itinQ32,
3555 string OpcodeStr, string Dt, SDNode OpNode> {
3556 // 64-bit vector types.
3557 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
3558 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
3559 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
3560 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
3561 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
3562 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
3564 // 128-bit vector types.
3565 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
3566 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
3567 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
3568 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
3569 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
3570 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
3573 multiclass N3VMulOpSL_HS<bits<4> op11_8,
3574 InstrItinClass itinD16, InstrItinClass itinD32,
3575 InstrItinClass itinQ16, InstrItinClass itinQ32,
3576 string OpcodeStr, string Dt, SDNode ShOp> {
3577 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
3578 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
3579 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
3580 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
3581 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
3582 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
3584 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
3585 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
3589 // Neon Intrinsic-Op vector operations,
3590 // element sizes of 8, 16 and 32 bits:
3591 multiclass N3VIntOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3592 InstrItinClass itinD, InstrItinClass itinQ,
3593 string OpcodeStr, string Dt, SDPatternOperator IntOp,
3595 // 64-bit vector types.
3596 def v8i8 : N3VDIntOp<op24, op23, 0b00, op11_8, op4, itinD,
3597 OpcodeStr, !strconcat(Dt, "8"), v8i8, IntOp, OpNode>;
3598 def v4i16 : N3VDIntOp<op24, op23, 0b01, op11_8, op4, itinD,
3599 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp, OpNode>;
3600 def v2i32 : N3VDIntOp<op24, op23, 0b10, op11_8, op4, itinD,
3601 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp, OpNode>;
3603 // 128-bit vector types.
3604 def v16i8 : N3VQIntOp<op24, op23, 0b00, op11_8, op4, itinQ,
3605 OpcodeStr, !strconcat(Dt, "8"), v16i8, IntOp, OpNode>;
3606 def v8i16 : N3VQIntOp<op24, op23, 0b01, op11_8, op4, itinQ,
3607 OpcodeStr, !strconcat(Dt, "16"), v8i16, IntOp, OpNode>;
3608 def v4i32 : N3VQIntOp<op24, op23, 0b10, op11_8, op4, itinQ,
3609 OpcodeStr, !strconcat(Dt, "32"), v4i32, IntOp, OpNode>;
3612 // Neon 3-argument intrinsics,
3613 // element sizes of 8, 16 and 32 bits:
3614 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3615 InstrItinClass itinD, InstrItinClass itinQ,
3616 string OpcodeStr, string Dt, SDPatternOperator IntOp> {
3617 // 64-bit vector types.
3618 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, itinD,
3619 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
3620 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, itinD,
3621 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
3622 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, itinD,
3623 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
3625 // 128-bit vector types.
3626 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, itinQ,
3627 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
3628 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, itinQ,
3629 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
3630 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, itinQ,
3631 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
3635 // Neon Long Multiply-Op vector operations,
3636 // element sizes of 8, 16 and 32 bits:
3637 multiclass N3VLMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3638 InstrItinClass itin16, InstrItinClass itin32,
3639 string OpcodeStr, string Dt, SDNode MulOp,
3641 def v8i16 : N3VLMulOp<op24, op23, 0b00, op11_8, op4, itin16, OpcodeStr,
3642 !strconcat(Dt, "8"), v8i16, v8i8, MulOp, OpNode>;
3643 def v4i32 : N3VLMulOp<op24, op23, 0b01, op11_8, op4, itin16, OpcodeStr,
3644 !strconcat(Dt, "16"), v4i32, v4i16, MulOp, OpNode>;
3645 def v2i64 : N3VLMulOp<op24, op23, 0b10, op11_8, op4, itin32, OpcodeStr,
3646 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
3649 multiclass N3VLMulOpSL_HS<bit op24, bits<4> op11_8, string OpcodeStr,
3650 string Dt, SDNode MulOp, SDNode OpNode> {
3651 def v4i16 : N3VLMulOpSL16<op24, 0b01, op11_8, IIC_VMACi16D, OpcodeStr,
3652 !strconcat(Dt,"16"), v4i32, v4i16, MulOp, OpNode>;
3653 def v2i32 : N3VLMulOpSL<op24, 0b10, op11_8, IIC_VMACi32D, OpcodeStr,
3654 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
3658 // Neon Long 3-argument intrinsics.
3660 // First with only element sizes of 16 and 32 bits:
3661 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
3662 InstrItinClass itin16, InstrItinClass itin32,
3663 string OpcodeStr, string Dt, SDPatternOperator IntOp> {
3664 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, itin16,
3665 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
3666 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, itin32,
3667 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3670 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
3671 string OpcodeStr, string Dt, SDPatternOperator IntOp> {
3672 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
3673 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
3674 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
3675 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3678 // ....then also with element size of 8 bits:
3679 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3680 InstrItinClass itin16, InstrItinClass itin32,
3681 string OpcodeStr, string Dt, SDPatternOperator IntOp>
3682 : N3VLInt3_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt, IntOp> {
3683 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, itin16,
3684 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
3687 // ....with explicit extend (VABAL).
3688 multiclass N3VLIntExtOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3689 InstrItinClass itin, string OpcodeStr, string Dt,
3690 SDPatternOperator IntOp, SDNode ExtOp, SDNode OpNode> {
3691 def v8i16 : N3VLIntExtOp<op24, op23, 0b00, op11_8, op4, itin,
3692 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8,
3693 IntOp, ExtOp, OpNode>;
3694 def v4i32 : N3VLIntExtOp<op24, op23, 0b01, op11_8, op4, itin,
3695 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16,
3696 IntOp, ExtOp, OpNode>;
3697 def v2i64 : N3VLIntExtOp<op24, op23, 0b10, op11_8, op4, itin,
3698 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32,
3699 IntOp, ExtOp, OpNode>;
3703 // Neon Pairwise long 2-register intrinsics,
3704 // element sizes of 8, 16 and 32 bits:
3705 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3706 bits<5> op11_7, bit op4,
3707 string OpcodeStr, string Dt, SDPatternOperator IntOp> {
3708 // 64-bit vector types.
3709 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3710 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
3711 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3712 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
3713 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3714 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
3716 // 128-bit vector types.
3717 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3718 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
3719 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3720 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
3721 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3722 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
3726 // Neon Pairwise long 2-register accumulate intrinsics,
3727 // element sizes of 8, 16 and 32 bits:
3728 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3729 bits<5> op11_7, bit op4,
3730 string OpcodeStr, string Dt, SDPatternOperator IntOp> {
3731 // 64-bit vector types.
3732 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3733 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
3734 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3735 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
3736 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3737 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
3739 // 128-bit vector types.
3740 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3741 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
3742 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3743 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
3744 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3745 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
3749 // Neon 2-register vector shift by immediate,
3750 // with f of either N2RegVShLFrm or N2RegVShRFrm
3751 // element sizes of 8, 16, 32 and 64 bits:
3752 multiclass N2VShL_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3753 InstrItinClass itin, string OpcodeStr, string Dt,
3755 // 64-bit vector types.
3756 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3757 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
3758 let Inst{21-19} = 0b001; // imm6 = 001xxx
3760 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3761 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
3762 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3764 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3765 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
3766 let Inst{21} = 0b1; // imm6 = 1xxxxx
3768 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, N2RegVShLFrm, itin, i32imm,
3769 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
3772 // 128-bit vector types.
3773 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3774 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
3775 let Inst{21-19} = 0b001; // imm6 = 001xxx
3777 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3778 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
3779 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3781 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3782 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
3783 let Inst{21} = 0b1; // imm6 = 1xxxxx
3785 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, N2RegVShLFrm, itin, i32imm,
3786 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
3789 multiclass N2VShR_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3790 InstrItinClass itin, string OpcodeStr, string Dt,
3791 string baseOpc, SDNode OpNode> {
3792 // 64-bit vector types.
3793 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm8,
3794 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
3795 let Inst{21-19} = 0b001; // imm6 = 001xxx
3797 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm16,
3798 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
3799 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3801 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm32,
3802 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
3803 let Inst{21} = 0b1; // imm6 = 1xxxxx
3805 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, N2RegVShRFrm, itin, shr_imm64,
3806 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
3809 // 128-bit vector types.
3810 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm8,
3811 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
3812 let Inst{21-19} = 0b001; // imm6 = 001xxx
3814 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm16,
3815 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
3816 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3818 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm32,
3819 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
3820 let Inst{21} = 0b1; // imm6 = 1xxxxx
3822 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, N2RegVShRFrm, itin, shr_imm64,
3823 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
3827 // Neon Shift-Accumulate vector operations,
3828 // element sizes of 8, 16, 32 and 64 bits:
3829 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3830 string OpcodeStr, string Dt, SDNode ShOp> {
3831 // 64-bit vector types.
3832 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm8,
3833 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
3834 let Inst{21-19} = 0b001; // imm6 = 001xxx
3836 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm16,
3837 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
3838 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3840 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm32,
3841 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
3842 let Inst{21} = 0b1; // imm6 = 1xxxxx
3844 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4, shr_imm64,
3845 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
3848 // 128-bit vector types.
3849 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm8,
3850 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
3851 let Inst{21-19} = 0b001; // imm6 = 001xxx
3853 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm16,
3854 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
3855 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3857 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm32,
3858 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
3859 let Inst{21} = 0b1; // imm6 = 1xxxxx
3861 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4, shr_imm64,
3862 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
3866 // Neon Shift-Insert vector operations,
3867 // with f of either N2RegVShLFrm or N2RegVShRFrm
3868 // element sizes of 8, 16, 32 and 64 bits:
3869 multiclass N2VShInsL_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3871 // 64-bit vector types.
3872 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3873 N2RegVShLFrm, OpcodeStr, "8", v8i8, NEONvsli> {
3874 let Inst{21-19} = 0b001; // imm6 = 001xxx
3876 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3877 N2RegVShLFrm, OpcodeStr, "16", v4i16, NEONvsli> {
3878 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3880 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3881 N2RegVShLFrm, OpcodeStr, "32", v2i32, NEONvsli> {
3882 let Inst{21} = 0b1; // imm6 = 1xxxxx
3884 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4, i32imm,
3885 N2RegVShLFrm, OpcodeStr, "64", v1i64, NEONvsli>;
3888 // 128-bit vector types.
3889 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3890 N2RegVShLFrm, OpcodeStr, "8", v16i8, NEONvsli> {
3891 let Inst{21-19} = 0b001; // imm6 = 001xxx
3893 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3894 N2RegVShLFrm, OpcodeStr, "16", v8i16, NEONvsli> {
3895 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3897 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3898 N2RegVShLFrm, OpcodeStr, "32", v4i32, NEONvsli> {
3899 let Inst{21} = 0b1; // imm6 = 1xxxxx
3901 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4, i32imm,
3902 N2RegVShLFrm, OpcodeStr, "64", v2i64, NEONvsli>;
3905 multiclass N2VShInsR_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3907 // 64-bit vector types.
3908 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm8,
3909 N2RegVShRFrm, OpcodeStr, "8", v8i8, NEONvsri> {
3910 let Inst{21-19} = 0b001; // imm6 = 001xxx
3912 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm16,
3913 N2RegVShRFrm, OpcodeStr, "16", v4i16, NEONvsri> {
3914 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3916 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm32,
3917 N2RegVShRFrm, OpcodeStr, "32", v2i32, NEONvsri> {
3918 let Inst{21} = 0b1; // imm6 = 1xxxxx
3920 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4, shr_imm64,
3921 N2RegVShRFrm, OpcodeStr, "64", v1i64, NEONvsri>;
3924 // 128-bit vector types.
3925 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm8,
3926 N2RegVShRFrm, OpcodeStr, "8", v16i8, NEONvsri> {
3927 let Inst{21-19} = 0b001; // imm6 = 001xxx
3929 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm16,
3930 N2RegVShRFrm, OpcodeStr, "16", v8i16, NEONvsri> {
3931 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3933 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm32,
3934 N2RegVShRFrm, OpcodeStr, "32", v4i32, NEONvsri> {
3935 let Inst{21} = 0b1; // imm6 = 1xxxxx
3937 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4, shr_imm64,
3938 N2RegVShRFrm, OpcodeStr, "64", v2i64, NEONvsri>;
3942 // Neon Shift Long operations,
3943 // element sizes of 8, 16, 32 bits:
3944 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
3945 bit op4, string OpcodeStr, string Dt, SDNode OpNode> {
3946 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3947 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, imm1_7, OpNode> {
3948 let Inst{21-19} = 0b001; // imm6 = 001xxx
3950 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3951 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, imm1_15, OpNode> {
3952 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3954 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3955 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, imm1_31, OpNode> {
3956 let Inst{21} = 0b1; // imm6 = 1xxxxx
3960 // Neon Shift Narrow operations,
3961 // element sizes of 16, 32, 64 bits:
3962 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
3963 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
3964 SDPatternOperator OpNode> {
3965 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3966 OpcodeStr, !strconcat(Dt, "16"),
3967 v8i8, v8i16, shr_imm8, OpNode> {
3968 let Inst{21-19} = 0b001; // imm6 = 001xxx
3970 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3971 OpcodeStr, !strconcat(Dt, "32"),
3972 v4i16, v4i32, shr_imm16, OpNode> {
3973 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3975 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3976 OpcodeStr, !strconcat(Dt, "64"),
3977 v2i32, v2i64, shr_imm32, OpNode> {
3978 let Inst{21} = 0b1; // imm6 = 1xxxxx
3982 //===----------------------------------------------------------------------===//
3983 // Instruction Definitions.
3984 //===----------------------------------------------------------------------===//
3986 // Vector Add Operations.
3988 // VADD : Vector Add (integer and floating-point)
3989 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
3991 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
3992 v2f32, v2f32, fadd, 1>;
3993 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
3994 v4f32, v4f32, fadd, 1>;
3995 // VADDL : Vector Add Long (Q = D + D)
3996 defm VADDLs : N3VLExt_QHS<0,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3997 "vaddl", "s", add, sext, 1>;
3998 defm VADDLu : N3VLExt_QHS<1,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3999 "vaddl", "u", add, zext, 1>;
4000 // VADDW : Vector Add Wide (Q = Q + D)
4001 defm VADDWs : N3VW_QHS<0,1,0b0001,0, "vaddw", "s", add, sext, 0>;
4002 defm VADDWu : N3VW_QHS<1,1,0b0001,0, "vaddw", "u", add, zext, 0>;
4003 // VHADD : Vector Halving Add
4004 defm VHADDs : N3VInt_QHS<0, 0, 0b0000, 0, N3RegFrm,
4005 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
4006 "vhadd", "s", int_arm_neon_vhadds, 1>;
4007 defm VHADDu : N3VInt_QHS<1, 0, 0b0000, 0, N3RegFrm,
4008 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
4009 "vhadd", "u", int_arm_neon_vhaddu, 1>;
4010 // VRHADD : Vector Rounding Halving Add
4011 defm VRHADDs : N3VInt_QHS<0, 0, 0b0001, 0, N3RegFrm,
4012 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
4013 "vrhadd", "s", int_arm_neon_vrhadds, 1>;
4014 defm VRHADDu : N3VInt_QHS<1, 0, 0b0001, 0, N3RegFrm,
4015 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
4016 "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
4017 // VQADD : Vector Saturating Add
4018 defm VQADDs : N3VInt_QHSD<0, 0, 0b0000, 1, N3RegFrm,
4019 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
4020 "vqadd", "s", int_arm_neon_vqadds, 1>;
4021 defm VQADDu : N3VInt_QHSD<1, 0, 0b0000, 1, N3RegFrm,
4022 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
4023 "vqadd", "u", int_arm_neon_vqaddu, 1>;
4024 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
4025 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i", null_frag, 1>;
4026 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
4027 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
4028 int_arm_neon_vraddhn, 1>;
4030 def : Pat<(v8i8 (trunc (NEONvshru (add (v8i16 QPR:$Vn), QPR:$Vm), 8))),
4031 (VADDHNv8i8 QPR:$Vn, QPR:$Vm)>;
4032 def : Pat<(v4i16 (trunc (NEONvshru (add (v4i32 QPR:$Vn), QPR:$Vm), 16))),
4033 (VADDHNv4i16 QPR:$Vn, QPR:$Vm)>;
4034 def : Pat<(v2i32 (trunc (NEONvshru (add (v2i64 QPR:$Vn), QPR:$Vm), 32))),
4035 (VADDHNv2i32 QPR:$Vn, QPR:$Vm)>;
4037 // Vector Multiply Operations.
4039 // VMUL : Vector Multiply (integer, polynomial and floating-point)
4040 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
4041 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
4042 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16D, "vmul",
4043 "p8", v8i8, v8i8, int_arm_neon_vmulp, 1>;
4044 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16Q, "vmul",
4045 "p8", v16i8, v16i8, int_arm_neon_vmulp, 1>;
4046 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VFMULD, "vmul", "f32",
4047 v2f32, v2f32, fmul, 1>;
4048 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VFMULQ, "vmul", "f32",
4049 v4f32, v4f32, fmul, 1>;
4050 defm VMULsl : N3VSL_HS<0b1000, "vmul", mul>;
4051 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
4052 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
4055 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
4056 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
4057 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
4058 (v4i16 (EXTRACT_SUBREG QPR:$src2,
4059 (DSubReg_i16_reg imm:$lane))),
4060 (SubReg_i16_lane imm:$lane)))>;
4061 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
4062 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
4063 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
4064 (v2i32 (EXTRACT_SUBREG QPR:$src2,
4065 (DSubReg_i32_reg imm:$lane))),
4066 (SubReg_i32_lane imm:$lane)))>;
4067 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
4068 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
4069 (v4f32 (VMULslfq (v4f32 QPR:$src1),
4070 (v2f32 (EXTRACT_SUBREG QPR:$src2,
4071 (DSubReg_i32_reg imm:$lane))),
4072 (SubReg_i32_lane imm:$lane)))>;
4075 def : Pat<(v2f32 (fmul DPR:$Rn, (NEONvdup (f32 SPR:$Rm)))),
4077 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$Rm, ssub_0),
4079 def : Pat<(v4f32 (fmul QPR:$Rn, (NEONvdup (f32 SPR:$Rm)))),
4081 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$Rm, ssub_0),
4085 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
4086 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, N3RegFrm, IIC_VMULi16D, IIC_VMULi32D,
4087 IIC_VMULi16Q, IIC_VMULi32Q,
4088 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
4089 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
4090 IIC_VMULi16Q, IIC_VMULi32Q,
4091 "vqdmulh", "s", int_arm_neon_vqdmulh>;
4092 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
4093 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
4095 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
4096 (v4i16 (EXTRACT_SUBREG QPR:$src2,
4097 (DSubReg_i16_reg imm:$lane))),
4098 (SubReg_i16_lane imm:$lane)))>;
4099 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
4100 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
4102 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
4103 (v2i32 (EXTRACT_SUBREG QPR:$src2,
4104 (DSubReg_i32_reg imm:$lane))),
4105 (SubReg_i32_lane imm:$lane)))>;
4107 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
4108 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, N3RegFrm,
4109 IIC_VMULi16D,IIC_VMULi32D,IIC_VMULi16Q,IIC_VMULi32Q,
4110 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
4111 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
4112 IIC_VMULi16Q, IIC_VMULi32Q,
4113 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
4114 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
4115 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
4117 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
4118 (v4i16 (EXTRACT_SUBREG QPR:$src2,
4119 (DSubReg_i16_reg imm:$lane))),
4120 (SubReg_i16_lane imm:$lane)))>;
4121 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
4122 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
4124 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
4125 (v2i32 (EXTRACT_SUBREG QPR:$src2,
4126 (DSubReg_i32_reg imm:$lane))),
4127 (SubReg_i32_lane imm:$lane)))>;
4129 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
4130 let PostEncoderMethod = "NEONThumb2DataIPostEncoder",
4131 DecoderNamespace = "NEONData" in {
4132 defm VMULLs : N3VL_QHS<0,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
4133 "vmull", "s", NEONvmulls, 1>;
4134 defm VMULLu : N3VL_QHS<1,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
4135 "vmull", "u", NEONvmullu, 1>;
4136 def VMULLp8 : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
4137 v8i16, v8i8, int_arm_neon_vmullp, 1>;
4138 def VMULLp64 : N3VLIntnp<0b00101, 0b10, 0b1110, 0, 0, NoItinerary,
4139 "vmull", "p64", v2i64, v1i64, int_arm_neon_vmullp, 1>,
4140 Requires<[HasV8, HasCrypto]>;
4142 defm VMULLsls : N3VLSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s", NEONvmulls>;
4143 defm VMULLslu : N3VLSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u", NEONvmullu>;
4145 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
4146 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, IIC_VMULi32D,
4147 "vqdmull", "s", int_arm_neon_vqdmull, 1>;
4148 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D,
4149 "vqdmull", "s", int_arm_neon_vqdmull>;
4151 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
4153 // VMLA : Vector Multiply Accumulate (integer and floating-point)
4154 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
4155 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
4156 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
4157 v2f32, fmul_su, fadd_mlx>,
4158 Requires<[HasNEON, UseFPVMLx, DontUseFusedMAC]>;
4159 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
4160 v4f32, fmul_su, fadd_mlx>,
4161 Requires<[HasNEON, UseFPVMLx, DontUseFusedMAC]>;
4162 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
4163 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
4164 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
4165 v2f32, fmul_su, fadd_mlx>,
4166 Requires<[HasNEON, UseFPVMLx]>;
4167 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
4168 v4f32, v2f32, fmul_su, fadd_mlx>,
4169 Requires<[HasNEON, UseFPVMLx]>;
4171 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
4172 (mul (v8i16 QPR:$src2),
4173 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
4174 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
4175 (v4i16 (EXTRACT_SUBREG QPR:$src3,
4176 (DSubReg_i16_reg imm:$lane))),
4177 (SubReg_i16_lane imm:$lane)))>;
4179 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
4180 (mul (v4i32 QPR:$src2),
4181 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
4182 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
4183 (v2i32 (EXTRACT_SUBREG QPR:$src3,
4184 (DSubReg_i32_reg imm:$lane))),
4185 (SubReg_i32_lane imm:$lane)))>;
4187 def : Pat<(v4f32 (fadd_mlx (v4f32 QPR:$src1),
4188 (fmul_su (v4f32 QPR:$src2),
4189 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
4190 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
4192 (v2f32 (EXTRACT_SUBREG QPR:$src3,
4193 (DSubReg_i32_reg imm:$lane))),
4194 (SubReg_i32_lane imm:$lane)))>,
4195 Requires<[HasNEON, UseFPVMLx]>;
4197 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
4198 defm VMLALs : N3VLMulOp_QHS<0,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
4199 "vmlal", "s", NEONvmulls, add>;
4200 defm VMLALu : N3VLMulOp_QHS<1,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
4201 "vmlal", "u", NEONvmullu, add>;
4203 defm VMLALsls : N3VLMulOpSL_HS<0, 0b0010, "vmlal", "s", NEONvmulls, add>;
4204 defm VMLALslu : N3VLMulOpSL_HS<1, 0b0010, "vmlal", "u", NEONvmullu, add>;
4206 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
4207 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
4208 "vqdmlal", "s", null_frag>;
4209 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", null_frag>;
4211 def : Pat<(v4i32 (int_arm_neon_vqadds (v4i32 QPR:$src1),
4212 (v4i32 (int_arm_neon_vqdmull (v4i16 DPR:$Vn),
4213 (v4i16 DPR:$Vm))))),
4214 (VQDMLALv4i32 QPR:$src1, DPR:$Vn, DPR:$Vm)>;
4215 def : Pat<(v2i64 (int_arm_neon_vqadds (v2i64 QPR:$src1),
4216 (v2i64 (int_arm_neon_vqdmull (v2i32 DPR:$Vn),
4217 (v2i32 DPR:$Vm))))),
4218 (VQDMLALv2i64 QPR:$src1, DPR:$Vn, DPR:$Vm)>;
4219 def : Pat<(v4i32 (int_arm_neon_vqadds (v4i32 QPR:$src1),
4220 (v4i32 (int_arm_neon_vqdmull (v4i16 DPR:$Vn),
4221 (v4i16 (NEONvduplane (v4i16 DPR_8:$Vm),
4223 (VQDMLALslv4i16 QPR:$src1, DPR:$Vn, DPR_8:$Vm, imm:$lane)>;
4224 def : Pat<(v2i64 (int_arm_neon_vqadds (v2i64 QPR:$src1),
4225 (v2i64 (int_arm_neon_vqdmull (v2i32 DPR:$Vn),
4226 (v2i32 (NEONvduplane (v2i32 DPR_VFP2:$Vm),
4228 (VQDMLALslv2i32 QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, imm:$lane)>;
4230 // VMLS : Vector Multiply Subtract (integer and floating-point)
4231 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
4232 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
4233 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
4234 v2f32, fmul_su, fsub_mlx>,
4235 Requires<[HasNEON, UseFPVMLx, DontUseFusedMAC]>;
4236 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
4237 v4f32, fmul_su, fsub_mlx>,
4238 Requires<[HasNEON, UseFPVMLx, DontUseFusedMAC]>;
4239 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
4240 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
4241 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
4242 v2f32, fmul_su, fsub_mlx>,
4243 Requires<[HasNEON, UseFPVMLx]>;
4244 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
4245 v4f32, v2f32, fmul_su, fsub_mlx>,
4246 Requires<[HasNEON, UseFPVMLx]>;
4248 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
4249 (mul (v8i16 QPR:$src2),
4250 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
4251 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
4252 (v4i16 (EXTRACT_SUBREG QPR:$src3,
4253 (DSubReg_i16_reg imm:$lane))),
4254 (SubReg_i16_lane imm:$lane)))>;
4256 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
4257 (mul (v4i32 QPR:$src2),
4258 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
4259 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
4260 (v2i32 (EXTRACT_SUBREG QPR:$src3,
4261 (DSubReg_i32_reg imm:$lane))),
4262 (SubReg_i32_lane imm:$lane)))>;
4264 def : Pat<(v4f32 (fsub_mlx (v4f32 QPR:$src1),
4265 (fmul_su (v4f32 QPR:$src2),
4266 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
4267 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
4268 (v2f32 (EXTRACT_SUBREG QPR:$src3,
4269 (DSubReg_i32_reg imm:$lane))),
4270 (SubReg_i32_lane imm:$lane)))>,
4271 Requires<[HasNEON, UseFPVMLx]>;
4273 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
4274 defm VMLSLs : N3VLMulOp_QHS<0,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
4275 "vmlsl", "s", NEONvmulls, sub>;
4276 defm VMLSLu : N3VLMulOp_QHS<1,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
4277 "vmlsl", "u", NEONvmullu, sub>;
4279 defm VMLSLsls : N3VLMulOpSL_HS<0, 0b0110, "vmlsl", "s", NEONvmulls, sub>;
4280 defm VMLSLslu : N3VLMulOpSL_HS<1, 0b0110, "vmlsl", "u", NEONvmullu, sub>;
4282 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
4283 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, IIC_VMACi16D, IIC_VMACi32D,
4284 "vqdmlsl", "s", null_frag>;
4285 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl", "s", null_frag>;
4287 def : Pat<(v4i32 (int_arm_neon_vqsubs (v4i32 QPR:$src1),
4288 (v4i32 (int_arm_neon_vqdmull (v4i16 DPR:$Vn),
4289 (v4i16 DPR:$Vm))))),
4290 (VQDMLSLv4i32 QPR:$src1, DPR:$Vn, DPR:$Vm)>;
4291 def : Pat<(v2i64 (int_arm_neon_vqsubs (v2i64 QPR:$src1),
4292 (v2i64 (int_arm_neon_vqdmull (v2i32 DPR:$Vn),
4293 (v2i32 DPR:$Vm))))),
4294 (VQDMLSLv2i64 QPR:$src1, DPR:$Vn, DPR:$Vm)>;
4295 def : Pat<(v4i32 (int_arm_neon_vqsubs (v4i32 QPR:$src1),
4296 (v4i32 (int_arm_neon_vqdmull (v4i16 DPR:$Vn),
4297 (v4i16 (NEONvduplane (v4i16 DPR_8:$Vm),
4299 (VQDMLSLslv4i16 QPR:$src1, DPR:$Vn, DPR_8:$Vm, imm:$lane)>;
4300 def : Pat<(v2i64 (int_arm_neon_vqsubs (v2i64 QPR:$src1),
4301 (v2i64 (int_arm_neon_vqdmull (v2i32 DPR:$Vn),
4302 (v2i32 (NEONvduplane (v2i32 DPR_VFP2:$Vm),
4304 (VQDMLSLslv2i32 QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, imm:$lane)>;
4306 // Fused Vector Multiply-Accumulate and Fused Multiply-Subtract Operations.
4307 def VFMAfd : N3VDMulOp<0, 0, 0b00, 0b1100, 1, IIC_VFMACD, "vfma", "f32",
4308 v2f32, fmul_su, fadd_mlx>,
4309 Requires<[HasNEON,HasVFP4,UseFusedMAC]>;
4311 def VFMAfq : N3VQMulOp<0, 0, 0b00, 0b1100, 1, IIC_VFMACQ, "vfma", "f32",
4312 v4f32, fmul_su, fadd_mlx>,
4313 Requires<[HasNEON,HasVFP4,UseFusedMAC]>;
4315 // Fused Vector Multiply Subtract (floating-point)
4316 def VFMSfd : N3VDMulOp<0, 0, 0b10, 0b1100, 1, IIC_VFMACD, "vfms", "f32",
4317 v2f32, fmul_su, fsub_mlx>,
4318 Requires<[HasNEON,HasVFP4,UseFusedMAC]>;
4319 def VFMSfq : N3VQMulOp<0, 0, 0b10, 0b1100, 1, IIC_VFMACQ, "vfms", "f32",
4320 v4f32, fmul_su, fsub_mlx>,
4321 Requires<[HasNEON,HasVFP4,UseFusedMAC]>;
4323 // Match @llvm.fma.* intrinsics
4324 def : Pat<(v2f32 (fma DPR:$Vn, DPR:$Vm, DPR:$src1)),
4325 (VFMAfd DPR:$src1, DPR:$Vn, DPR:$Vm)>,
4326 Requires<[HasVFP4]>;
4327 def : Pat<(v4f32 (fma QPR:$Vn, QPR:$Vm, QPR:$src1)),
4328 (VFMAfq QPR:$src1, QPR:$Vn, QPR:$Vm)>,
4329 Requires<[HasVFP4]>;
4330 def : Pat<(v2f32 (fma (fneg DPR:$Vn), DPR:$Vm, DPR:$src1)),
4331 (VFMSfd DPR:$src1, DPR:$Vn, DPR:$Vm)>,
4332 Requires<[HasVFP4]>;
4333 def : Pat<(v4f32 (fma (fneg QPR:$Vn), QPR:$Vm, QPR:$src1)),
4334 (VFMSfq QPR:$src1, QPR:$Vn, QPR:$Vm)>,
4335 Requires<[HasVFP4]>;
4337 // Vector Subtract Operations.
4339 // VSUB : Vector Subtract (integer and floating-point)
4340 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
4341 "vsub", "i", sub, 0>;
4342 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
4343 v2f32, v2f32, fsub, 0>;
4344 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
4345 v4f32, v4f32, fsub, 0>;
4346 // VSUBL : Vector Subtract Long (Q = D - D)
4347 defm VSUBLs : N3VLExt_QHS<0,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
4348 "vsubl", "s", sub, sext, 0>;
4349 defm VSUBLu : N3VLExt_QHS<1,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
4350 "vsubl", "u", sub, zext, 0>;
4351 // VSUBW : Vector Subtract Wide (Q = Q - D)
4352 defm VSUBWs : N3VW_QHS<0,1,0b0011,0, "vsubw", "s", sub, sext, 0>;
4353 defm VSUBWu : N3VW_QHS<1,1,0b0011,0, "vsubw", "u", sub, zext, 0>;
4354 // VHSUB : Vector Halving Subtract
4355 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, N3RegFrm,
4356 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4357 "vhsub", "s", int_arm_neon_vhsubs, 0>;
4358 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, N3RegFrm,
4359 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4360 "vhsub", "u", int_arm_neon_vhsubu, 0>;
4361 // VQSUB : Vector Saturing Subtract
4362 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, N3RegFrm,
4363 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4364 "vqsub", "s", int_arm_neon_vqsubs, 0>;
4365 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, N3RegFrm,
4366 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4367 "vqsub", "u", int_arm_neon_vqsubu, 0>;
4368 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
4369 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i", null_frag, 0>;
4370 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
4371 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
4372 int_arm_neon_vrsubhn, 0>;
4374 def : Pat<(v8i8 (trunc (NEONvshru (sub (v8i16 QPR:$Vn), QPR:$Vm), 8))),
4375 (VSUBHNv8i8 QPR:$Vn, QPR:$Vm)>;
4376 def : Pat<(v4i16 (trunc (NEONvshru (sub (v4i32 QPR:$Vn), QPR:$Vm), 16))),
4377 (VSUBHNv4i16 QPR:$Vn, QPR:$Vm)>;
4378 def : Pat<(v2i32 (trunc (NEONvshru (sub (v2i64 QPR:$Vn), QPR:$Vm), 32))),
4379 (VSUBHNv2i32 QPR:$Vn, QPR:$Vm)>;
4381 // Vector Comparisons.
4383 // VCEQ : Vector Compare Equal
4384 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
4385 IIC_VSUBi4Q, "vceq", "i", NEONvceq, 1>;
4386 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
4388 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
4391 let TwoOperandAliasConstraint = "$Vm = $Vd" in
4392 defm VCEQz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
4393 "$Vd, $Vm, #0", NEONvceqz>;
4395 // VCGE : Vector Compare Greater Than or Equal
4396 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
4397 IIC_VSUBi4Q, "vcge", "s", NEONvcge, 0>;
4398 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
4399 IIC_VSUBi4Q, "vcge", "u", NEONvcgeu, 0>;
4400 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32", v2i32, v2f32,
4402 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
4405 let TwoOperandAliasConstraint = "$Vm = $Vd" in {
4406 defm VCGEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
4407 "$Vd, $Vm, #0", NEONvcgez>;
4408 defm VCLEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
4409 "$Vd, $Vm, #0", NEONvclez>;
4412 // VCGT : Vector Compare Greater Than
4413 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
4414 IIC_VSUBi4Q, "vcgt", "s", NEONvcgt, 0>;
4415 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
4416 IIC_VSUBi4Q, "vcgt", "u", NEONvcgtu, 0>;
4417 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
4419 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
4422 let TwoOperandAliasConstraint = "$Vm = $Vd" in {
4423 defm VCGTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
4424 "$Vd, $Vm, #0", NEONvcgtz>;
4425 defm VCLTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
4426 "$Vd, $Vm, #0", NEONvcltz>;
4429 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
4430 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacge",
4431 "f32", v2i32, v2f32, int_arm_neon_vacge, 0>;
4432 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacge",
4433 "f32", v4i32, v4f32, int_arm_neon_vacge, 0>;
4434 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
4435 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacgt",
4436 "f32", v2i32, v2f32, int_arm_neon_vacgt, 0>;
4437 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacgt",
4438 "f32", v4i32, v4f32, int_arm_neon_vacgt, 0>;
4439 // VTST : Vector Test Bits
4440 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
4441 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
4443 def: NEONInstAlias<"vaclt${p}.f32 $Vd, $Vn, $Vm",
4444 (VACGTd DPR:$Vd, DPR:$Vm, DPR:$Vn, pred:$p)>;
4445 def: NEONInstAlias<"vaclt${p}.f32 $Vd, $Vn, $Vm",
4446 (VACGTq QPR:$Vd, QPR:$Vm, QPR:$Vn, pred:$p)>;
4447 def: NEONInstAlias<"vacle${p}.f32 $Vd, $Vn, $Vm",
4448 (VACGEd DPR:$Vd, DPR:$Vm, DPR:$Vn, pred:$p)>;
4449 def: NEONInstAlias<"vacle${p}.f32 $Vd, $Vn, $Vm",
4450 (VACGEq QPR:$Vd, QPR:$Vm, QPR:$Vn, pred:$p)>;
4452 def: NEONInstAlias<"vaclt${p}.f32 $Vd, $Vm",
4453 (VACGTd DPR:$Vd, DPR:$Vm, DPR:$Vd, pred:$p)>;
4454 def: NEONInstAlias<"vaclt${p}.f32 $Vd, $Vm",
4455 (VACGTq QPR:$Vd, QPR:$Vm, QPR:$Vd, pred:$p)>;
4456 def: NEONInstAlias<"vacle${p}.f32 $Vd, $Vm",
4457 (VACGEd DPR:$Vd, DPR:$Vm, DPR:$Vd, pred:$p)>;
4458 def: NEONInstAlias<"vacle${p}.f32 $Vd, $Vm",
4459 (VACGEq QPR:$Vd, QPR:$Vm, QPR:$Vd, pred:$p)>;
4461 // Vector Bitwise Operations.
4463 def vnotd : PatFrag<(ops node:$in),
4464 (xor node:$in, (bitconvert (v8i8 NEONimmAllOnesV)))>;
4465 def vnotq : PatFrag<(ops node:$in),
4466 (xor node:$in, (bitconvert (v16i8 NEONimmAllOnesV)))>;
4469 // VAND : Vector Bitwise AND
4470 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
4471 v2i32, v2i32, and, 1>;
4472 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
4473 v4i32, v4i32, and, 1>;
4475 // VEOR : Vector Bitwise Exclusive OR
4476 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
4477 v2i32, v2i32, xor, 1>;
4478 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
4479 v4i32, v4i32, xor, 1>;
4481 // VORR : Vector Bitwise OR
4482 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
4483 v2i32, v2i32, or, 1>;
4484 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
4485 v4i32, v4i32, or, 1>;
4487 def VORRiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 0, 1,
4488 (outs DPR:$Vd), (ins nImmSplatI16:$SIMM, DPR:$src),
4490 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
4492 (v4i16 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
4493 let Inst{9} = SIMM{9};
4496 def VORRiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 0, 1,
4497 (outs DPR:$Vd), (ins nImmSplatI32:$SIMM, DPR:$src),
4499 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
4501 (v2i32 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
4502 let Inst{10-9} = SIMM{10-9};
4505 def VORRiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 0, 1,
4506 (outs QPR:$Vd), (ins nImmSplatI16:$SIMM, QPR:$src),
4508 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
4510 (v8i16 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
4511 let Inst{9} = SIMM{9};
4514 def VORRiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 0, 1,
4515 (outs QPR:$Vd), (ins nImmSplatI32:$SIMM, QPR:$src),
4517 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
4519 (v4i32 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
4520 let Inst{10-9} = SIMM{10-9};
4524 // VBIC : Vector Bitwise Bit Clear (AND NOT)
4525 let TwoOperandAliasConstraint = "$Vn = $Vd" in {
4526 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
4527 (ins DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VBINiD,
4528 "vbic", "$Vd, $Vn, $Vm", "",
4529 [(set DPR:$Vd, (v2i32 (and DPR:$Vn,
4530 (vnotd DPR:$Vm))))]>;
4531 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
4532 (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINiQ,
4533 "vbic", "$Vd, $Vn, $Vm", "",
4534 [(set QPR:$Vd, (v4i32 (and QPR:$Vn,
4535 (vnotq QPR:$Vm))))]>;
4538 def VBICiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 1, 1,
4539 (outs DPR:$Vd), (ins nImmSplatI16:$SIMM, DPR:$src),
4541 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
4543 (v4i16 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
4544 let Inst{9} = SIMM{9};
4547 def VBICiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 1, 1,
4548 (outs DPR:$Vd), (ins nImmSplatI32:$SIMM, DPR:$src),
4550 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
4552 (v2i32 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
4553 let Inst{10-9} = SIMM{10-9};
4556 def VBICiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 1, 1,
4557 (outs QPR:$Vd), (ins nImmSplatI16:$SIMM, QPR:$src),
4559 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
4561 (v8i16 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
4562 let Inst{9} = SIMM{9};
4565 def VBICiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 1, 1,
4566 (outs QPR:$Vd), (ins nImmSplatI32:$SIMM, QPR:$src),
4568 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
4570 (v4i32 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
4571 let Inst{10-9} = SIMM{10-9};
4574 // VORN : Vector Bitwise OR NOT
4575 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$Vd),
4576 (ins DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VBINiD,
4577 "vorn", "$Vd, $Vn, $Vm", "",
4578 [(set DPR:$Vd, (v2i32 (or DPR:$Vn,
4579 (vnotd DPR:$Vm))))]>;
4580 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$Vd),
4581 (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINiQ,
4582 "vorn", "$Vd, $Vn, $Vm", "",
4583 [(set QPR:$Vd, (v4i32 (or QPR:$Vn,
4584 (vnotq QPR:$Vm))))]>;
4586 // VMVN : Vector Bitwise NOT (Immediate)
4588 let isReMaterializable = 1 in {
4590 def VMVNv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 1, 1, (outs DPR:$Vd),
4591 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4592 "vmvn", "i16", "$Vd, $SIMM", "",
4593 [(set DPR:$Vd, (v4i16 (NEONvmvnImm timm:$SIMM)))]> {
4594 let Inst{9} = SIMM{9};
4597 def VMVNv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 1, 1, (outs QPR:$Vd),
4598 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4599 "vmvn", "i16", "$Vd, $SIMM", "",
4600 [(set QPR:$Vd, (v8i16 (NEONvmvnImm timm:$SIMM)))]> {
4601 let Inst{9} = SIMM{9};
4604 def VMVNv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 1, 1, (outs DPR:$Vd),
4605 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4606 "vmvn", "i32", "$Vd, $SIMM", "",
4607 [(set DPR:$Vd, (v2i32 (NEONvmvnImm timm:$SIMM)))]> {
4608 let Inst{11-8} = SIMM{11-8};
4611 def VMVNv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 1, 1, (outs QPR:$Vd),
4612 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4613 "vmvn", "i32", "$Vd, $SIMM", "",
4614 [(set QPR:$Vd, (v4i32 (NEONvmvnImm timm:$SIMM)))]> {
4615 let Inst{11-8} = SIMM{11-8};
4619 // VMVN : Vector Bitwise NOT
4620 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
4621 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VSUBiD,
4622 "vmvn", "$Vd, $Vm", "",
4623 [(set DPR:$Vd, (v2i32 (vnotd DPR:$Vm)))]>;
4624 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
4625 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VSUBiD,
4626 "vmvn", "$Vd, $Vm", "",
4627 [(set QPR:$Vd, (v4i32 (vnotq QPR:$Vm)))]>;
4628 def : Pat<(v2i32 (vnotd DPR:$src)), (VMVNd DPR:$src)>;
4629 def : Pat<(v4i32 (vnotq QPR:$src)), (VMVNq QPR:$src)>;
4631 // VBSL : Vector Bitwise Select
4632 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
4633 (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4634 N3RegFrm, IIC_VCNTiD,
4635 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4637 (v2i32 (NEONvbsl DPR:$src1, DPR:$Vn, DPR:$Vm)))]>;
4638 def : Pat<(v8i8 (int_arm_neon_vbsl (v8i8 DPR:$src1),
4639 (v8i8 DPR:$Vn), (v8i8 DPR:$Vm))),
4640 (VBSLd DPR:$src1, DPR:$Vn, DPR:$Vm)>,
4641 Requires<[HasNEON]>;
4642 def : Pat<(v4i16 (int_arm_neon_vbsl (v4i16 DPR:$src1),
4643 (v4i16 DPR:$Vn), (v4i16 DPR:$Vm))),
4644 (VBSLd DPR:$src1, DPR:$Vn, DPR:$Vm)>,
4645 Requires<[HasNEON]>;
4646 def : Pat<(v2i32 (int_arm_neon_vbsl (v2i32 DPR:$src1),
4647 (v2i32 DPR:$Vn), (v2i32 DPR:$Vm))),
4648 (VBSLd DPR:$src1, DPR:$Vn, DPR:$Vm)>,
4649 Requires<[HasNEON]>;
4650 def : Pat<(v2f32 (int_arm_neon_vbsl (v2f32 DPR:$src1),
4651 (v2f32 DPR:$Vn), (v2f32 DPR:$Vm))),
4652 (VBSLd DPR:$src1, DPR:$Vn, DPR:$Vm)>,
4653 Requires<[HasNEON]>;
4654 def : Pat<(v1i64 (int_arm_neon_vbsl (v1i64 DPR:$src1),
4655 (v1i64 DPR:$Vn), (v1i64 DPR:$Vm))),
4656 (VBSLd DPR:$src1, DPR:$Vn, DPR:$Vm)>,
4657 Requires<[HasNEON]>;
4659 def : Pat<(v2i32 (or (and DPR:$Vn, DPR:$Vd),
4660 (and DPR:$Vm, (vnotd DPR:$Vd)))),
4661 (VBSLd DPR:$Vd, DPR:$Vn, DPR:$Vm)>,
4662 Requires<[HasNEON]>;
4664 def : Pat<(v1i64 (or (and DPR:$Vn, DPR:$Vd),
4665 (and DPR:$Vm, (vnotd DPR:$Vd)))),
4666 (VBSLd DPR:$Vd, DPR:$Vn, DPR:$Vm)>,
4667 Requires<[HasNEON]>;
4669 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
4670 (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4671 N3RegFrm, IIC_VCNTiQ,
4672 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4674 (v4i32 (NEONvbsl QPR:$src1, QPR:$Vn, QPR:$Vm)))]>;
4676 def : Pat<(v16i8 (int_arm_neon_vbsl (v16i8 QPR:$src1),
4677 (v16i8 QPR:$Vn), (v16i8 QPR:$Vm))),
4678 (VBSLq QPR:$src1, QPR:$Vn, QPR:$Vm)>,
4679 Requires<[HasNEON]>;
4680 def : Pat<(v8i16 (int_arm_neon_vbsl (v8i16 QPR:$src1),
4681 (v8i16 QPR:$Vn), (v8i16 QPR:$Vm))),
4682 (VBSLq QPR:$src1, QPR:$Vn, QPR:$Vm)>,
4683 Requires<[HasNEON]>;
4684 def : Pat<(v4i32 (int_arm_neon_vbsl (v4i32 QPR:$src1),
4685 (v4i32 QPR:$Vn), (v4i32 QPR:$Vm))),
4686 (VBSLq QPR:$src1, QPR:$Vn, QPR:$Vm)>,
4687 Requires<[HasNEON]>;
4688 def : Pat<(v4f32 (int_arm_neon_vbsl (v4f32 QPR:$src1),
4689 (v4f32 QPR:$Vn), (v4f32 QPR:$Vm))),
4690 (VBSLq QPR:$src1, QPR:$Vn, QPR:$Vm)>,
4691 Requires<[HasNEON]>;
4692 def : Pat<(v2i64 (int_arm_neon_vbsl (v2i64 QPR:$src1),
4693 (v2i64 QPR:$Vn), (v2i64 QPR:$Vm))),
4694 (VBSLq QPR:$src1, QPR:$Vn, QPR:$Vm)>,
4695 Requires<[HasNEON]>;
4697 def : Pat<(v4i32 (or (and QPR:$Vn, QPR:$Vd),
4698 (and QPR:$Vm, (vnotq QPR:$Vd)))),
4699 (VBSLq QPR:$Vd, QPR:$Vn, QPR:$Vm)>,
4700 Requires<[HasNEON]>;
4701 def : Pat<(v2i64 (or (and QPR:$Vn, QPR:$Vd),
4702 (and QPR:$Vm, (vnotq QPR:$Vd)))),
4703 (VBSLq QPR:$Vd, QPR:$Vn, QPR:$Vm)>,
4704 Requires<[HasNEON]>;
4706 // VBIF : Vector Bitwise Insert if False
4707 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
4708 // FIXME: This instruction's encoding MAY NOT BE correct.
4709 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
4710 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4711 N3RegFrm, IIC_VBINiD,
4712 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4714 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
4715 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4716 N3RegFrm, IIC_VBINiQ,
4717 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4720 // VBIT : Vector Bitwise Insert if True
4721 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
4722 // FIXME: This instruction's encoding MAY NOT BE correct.
4723 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
4724 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4725 N3RegFrm, IIC_VBINiD,
4726 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4728 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
4729 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4730 N3RegFrm, IIC_VBINiQ,
4731 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4734 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
4735 // for equivalent operations with different register constraints; it just
4738 // Vector Absolute Differences.
4740 // VABD : Vector Absolute Difference
4741 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, N3RegFrm,
4742 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4743 "vabd", "s", int_arm_neon_vabds, 1>;
4744 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, N3RegFrm,
4745 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4746 "vabd", "u", int_arm_neon_vabdu, 1>;
4747 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBIND,
4748 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 1>;
4749 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBINQ,
4750 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 1>;
4752 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
4753 defm VABDLs : N3VLIntExt_QHS<0,1,0b0111,0, IIC_VSUBi4Q,
4754 "vabdl", "s", int_arm_neon_vabds, zext, 1>;
4755 defm VABDLu : N3VLIntExt_QHS<1,1,0b0111,0, IIC_VSUBi4Q,
4756 "vabdl", "u", int_arm_neon_vabdu, zext, 1>;
4758 // VABA : Vector Absolute Difference and Accumulate
4759 defm VABAs : N3VIntOp_QHS<0,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
4760 "vaba", "s", int_arm_neon_vabds, add>;
4761 defm VABAu : N3VIntOp_QHS<1,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
4762 "vaba", "u", int_arm_neon_vabdu, add>;
4764 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
4765 defm VABALs : N3VLIntExtOp_QHS<0,1,0b0101,0, IIC_VABAD,
4766 "vabal", "s", int_arm_neon_vabds, zext, add>;
4767 defm VABALu : N3VLIntExtOp_QHS<1,1,0b0101,0, IIC_VABAD,
4768 "vabal", "u", int_arm_neon_vabdu, zext, add>;
4770 // Vector Maximum and Minimum.
4772 // VMAX : Vector Maximum
4773 defm VMAXs : N3VInt_QHS<0, 0, 0b0110, 0, N3RegFrm,
4774 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4775 "vmax", "s", int_arm_neon_vmaxs, 1>;
4776 defm VMAXu : N3VInt_QHS<1, 0, 0b0110, 0, N3RegFrm,
4777 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4778 "vmax", "u", int_arm_neon_vmaxu, 1>;
4779 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBIND,
4781 v2f32, v2f32, int_arm_neon_vmaxs, 1>;
4782 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBINQ,
4784 v4f32, v4f32, int_arm_neon_vmaxs, 1>;
4787 let PostEncoderMethod = "NEONThumb2V8PostEncoder", DecoderNamespace = "v8NEON" in {
4788 def VMAXNMND : N3VDIntnp<0b00110, 0b00, 0b1111, 0, 1,
4789 N3RegFrm, NoItinerary, "vmaxnm", "f32",
4790 v2f32, v2f32, int_arm_neon_vmaxnm, 1>,
4791 Requires<[HasV8, HasNEON]>;
4792 def VMAXNMNQ : N3VQIntnp<0b00110, 0b00, 0b1111, 1, 1,
4793 N3RegFrm, NoItinerary, "vmaxnm", "f32",
4794 v4f32, v4f32, int_arm_neon_vmaxnm, 1>,
4795 Requires<[HasV8, HasNEON]>;
4798 // VMIN : Vector Minimum
4799 defm VMINs : N3VInt_QHS<0, 0, 0b0110, 1, N3RegFrm,
4800 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4801 "vmin", "s", int_arm_neon_vmins, 1>;
4802 defm VMINu : N3VInt_QHS<1, 0, 0b0110, 1, N3RegFrm,
4803 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4804 "vmin", "u", int_arm_neon_vminu, 1>;
4805 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBIND,
4807 v2f32, v2f32, int_arm_neon_vmins, 1>;
4808 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBINQ,
4810 v4f32, v4f32, int_arm_neon_vmins, 1>;
4813 let PostEncoderMethod = "NEONThumb2V8PostEncoder", DecoderNamespace = "v8NEON" in {
4814 def VMINNMND : N3VDIntnp<0b00110, 0b10, 0b1111, 0, 1,
4815 N3RegFrm, NoItinerary, "vminnm", "f32",
4816 v2f32, v2f32, int_arm_neon_vminnm, 1>,
4817 Requires<[HasV8, HasNEON]>;
4818 def VMINNMNQ : N3VQIntnp<0b00110, 0b10, 0b1111, 1, 1,
4819 N3RegFrm, NoItinerary, "vminnm", "f32",
4820 v4f32, v4f32, int_arm_neon_vminnm, 1>,
4821 Requires<[HasV8, HasNEON]>;
4824 // Vector Pairwise Operations.
4826 // VPADD : Vector Pairwise Add
4827 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4829 v8i8, v8i8, int_arm_neon_vpadd, 0>;
4830 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4832 v4i16, v4i16, int_arm_neon_vpadd, 0>;
4833 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4835 v2i32, v2i32, int_arm_neon_vpadd, 0>;
4836 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, N3RegFrm,
4837 IIC_VPBIND, "vpadd", "f32",
4838 v2f32, v2f32, int_arm_neon_vpadd, 0>;
4840 // VPADDL : Vector Pairwise Add Long
4841 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
4842 int_arm_neon_vpaddls>;
4843 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
4844 int_arm_neon_vpaddlu>;
4846 // VPADAL : Vector Pairwise Add and Accumulate Long
4847 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
4848 int_arm_neon_vpadals>;
4849 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
4850 int_arm_neon_vpadalu>;
4852 // VPMAX : Vector Pairwise Maximum
4853 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4854 "s8", v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
4855 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4856 "s16", v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
4857 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4858 "s32", v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
4859 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4860 "u8", v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
4861 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4862 "u16", v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
4863 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4864 "u32", v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
4865 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmax",
4866 "f32", v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
4868 // VPMIN : Vector Pairwise Minimum
4869 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4870 "s8", v8i8, v8i8, int_arm_neon_vpmins, 0>;
4871 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4872 "s16", v4i16, v4i16, int_arm_neon_vpmins, 0>;
4873 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4874 "s32", v2i32, v2i32, int_arm_neon_vpmins, 0>;
4875 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4876 "u8", v8i8, v8i8, int_arm_neon_vpminu, 0>;
4877 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4878 "u16", v4i16, v4i16, int_arm_neon_vpminu, 0>;
4879 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4880 "u32", v2i32, v2i32, int_arm_neon_vpminu, 0>;
4881 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmin",
4882 "f32", v2f32, v2f32, int_arm_neon_vpmins, 0>;
4884 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
4886 // VRECPE : Vector Reciprocal Estimate
4887 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
4888 IIC_VUNAD, "vrecpe", "u32",
4889 v2i32, v2i32, int_arm_neon_vrecpe>;
4890 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
4891 IIC_VUNAQ, "vrecpe", "u32",
4892 v4i32, v4i32, int_arm_neon_vrecpe>;
4893 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
4894 IIC_VUNAD, "vrecpe", "f32",
4895 v2f32, v2f32, int_arm_neon_vrecpe>;
4896 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
4897 IIC_VUNAQ, "vrecpe", "f32",
4898 v4f32, v4f32, int_arm_neon_vrecpe>;
4900 // VRECPS : Vector Reciprocal Step
4901 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
4902 IIC_VRECSD, "vrecps", "f32",
4903 v2f32, v2f32, int_arm_neon_vrecps, 1>;
4904 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
4905 IIC_VRECSQ, "vrecps", "f32",
4906 v4f32, v4f32, int_arm_neon_vrecps, 1>;
4908 // VRSQRTE : Vector Reciprocal Square Root Estimate
4909 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
4910 IIC_VUNAD, "vrsqrte", "u32",
4911 v2i32, v2i32, int_arm_neon_vrsqrte>;
4912 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
4913 IIC_VUNAQ, "vrsqrte", "u32",
4914 v4i32, v4i32, int_arm_neon_vrsqrte>;
4915 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
4916 IIC_VUNAD, "vrsqrte", "f32",
4917 v2f32, v2f32, int_arm_neon_vrsqrte>;
4918 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
4919 IIC_VUNAQ, "vrsqrte", "f32",
4920 v4f32, v4f32, int_arm_neon_vrsqrte>;
4922 // VRSQRTS : Vector Reciprocal Square Root Step
4923 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
4924 IIC_VRECSD, "vrsqrts", "f32",
4925 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
4926 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
4927 IIC_VRECSQ, "vrsqrts", "f32",
4928 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
4932 // VSHL : Vector Shift
4933 defm VSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 0, N3RegVShFrm,
4934 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
4935 "vshl", "s", int_arm_neon_vshifts>;
4936 defm VSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 0, N3RegVShFrm,
4937 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
4938 "vshl", "u", int_arm_neon_vshiftu>;
4940 // VSHL : Vector Shift Left (Immediate)
4941 defm VSHLi : N2VShL_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl>;
4943 // VSHR : Vector Shift Right (Immediate)
4944 defm VSHRs : N2VShR_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s", "VSHRs",
4946 defm VSHRu : N2VShR_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u", "VSHRu",
4949 // VSHLL : Vector Shift Left Long
4950 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>;
4951 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>;
4953 // VSHLL : Vector Shift Left Long (with maximum shift count)
4954 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
4955 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
4956 ValueType OpTy, Operand ImmTy, SDNode OpNode>
4957 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
4958 ResTy, OpTy, ImmTy, OpNode> {
4959 let Inst{21-16} = op21_16;
4960 let DecoderMethod = "DecodeVSHLMaxInstruction";
4962 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
4963 v8i16, v8i8, imm8, NEONvshlli>;
4964 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
4965 v4i32, v4i16, imm16, NEONvshlli>;
4966 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
4967 v2i64, v2i32, imm32, NEONvshlli>;
4969 // VSHRN : Vector Shift Right and Narrow
4970 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
4971 PatFrag<(ops node:$Rn, node:$amt),
4972 (trunc (NEONvshrs node:$Rn, node:$amt))>>;
4974 def : Pat<(v8i8 (trunc (NEONvshru (v8i16 QPR:$Vn), shr_imm8:$amt))),
4975 (VSHRNv8i8 QPR:$Vn, shr_imm8:$amt)>;
4976 def : Pat<(v4i16 (trunc (NEONvshru (v4i32 QPR:$Vn), shr_imm16:$amt))),
4977 (VSHRNv4i16 QPR:$Vn, shr_imm16:$amt)>;
4978 def : Pat<(v2i32 (trunc (NEONvshru (v2i64 QPR:$Vn), shr_imm32:$amt))),
4979 (VSHRNv2i32 QPR:$Vn, shr_imm32:$amt)>;
4981 // VRSHL : Vector Rounding Shift
4982 defm VRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 0, N3RegVShFrm,
4983 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4984 "vrshl", "s", int_arm_neon_vrshifts>;
4985 defm VRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 0, N3RegVShFrm,
4986 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4987 "vrshl", "u", int_arm_neon_vrshiftu>;
4988 // VRSHR : Vector Rounding Shift Right
4989 defm VRSHRs : N2VShR_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s", "VRSHRs",
4991 defm VRSHRu : N2VShR_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u", "VRSHRu",
4994 // VRSHRN : Vector Rounding Shift Right and Narrow
4995 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
4998 // VQSHL : Vector Saturating Shift
4999 defm VQSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 1, N3RegVShFrm,
5000 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
5001 "vqshl", "s", int_arm_neon_vqshifts>;
5002 defm VQSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 1, N3RegVShFrm,
5003 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
5004 "vqshl", "u", int_arm_neon_vqshiftu>;
5005 // VQSHL : Vector Saturating Shift Left (Immediate)
5006 defm VQSHLsi : N2VShL_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s",NEONvqshls>;
5007 defm VQSHLui : N2VShL_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u",NEONvqshlu>;
5009 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
5010 defm VQSHLsu : N2VShL_QHSD<1,1,0b0110,1, IIC_VSHLi4D,"vqshlu","s",NEONvqshlsu>;
5012 // VQSHRN : Vector Saturating Shift Right and Narrow
5013 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
5015 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
5018 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
5019 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
5022 // VQRSHL : Vector Saturating Rounding Shift
5023 defm VQRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 1, N3RegVShFrm,
5024 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
5025 "vqrshl", "s", int_arm_neon_vqrshifts>;
5026 defm VQRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 1, N3RegVShFrm,
5027 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
5028 "vqrshl", "u", int_arm_neon_vqrshiftu>;
5030 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
5031 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
5033 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
5036 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
5037 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
5040 // VSRA : Vector Shift Right and Accumulate
5041 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
5042 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
5043 // VRSRA : Vector Rounding Shift Right and Accumulate
5044 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
5045 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
5047 // VSLI : Vector Shift Left and Insert
5048 defm VSLI : N2VShInsL_QHSD<1, 1, 0b0101, 1, "vsli">;
5050 // VSRI : Vector Shift Right and Insert
5051 defm VSRI : N2VShInsR_QHSD<1, 1, 0b0100, 1, "vsri">;
5053 // Vector Absolute and Saturating Absolute.
5055 // VABS : Vector Absolute Value
5056 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
5057 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
5059 def VABSfd : N2VD<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
5061 v2f32, v2f32, fabs>;
5062 def VABSfq : N2VQ<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
5064 v4f32, v4f32, fabs>;
5066 def : Pat<(xor (v2i32 (bitconvert (v8i8 (NEONvshrs DPR:$src, (i32 7))))),
5067 (v2i32 (bitconvert (v8i8 (add DPR:$src,
5068 (NEONvshrs DPR:$src, (i32 7))))))),
5069 (VABSv8i8 DPR:$src)>;
5070 def : Pat<(xor (v2i32 (bitconvert (v4i16 (NEONvshrs DPR:$src, (i32 15))))),
5071 (v2i32 (bitconvert (v4i16 (add DPR:$src,
5072 (NEONvshrs DPR:$src, (i32 15))))))),
5073 (VABSv4i16 DPR:$src)>;
5074 def : Pat<(xor (v2i32 (NEONvshrs DPR:$src, (i32 31))),
5075 (v2i32 (add DPR:$src, (NEONvshrs DPR:$src, (i32 31))))),
5076 (VABSv2i32 DPR:$src)>;
5077 def : Pat<(xor (v4i32 (bitconvert (v16i8 (NEONvshrs QPR:$src, (i32 7))))),
5078 (v4i32 (bitconvert (v16i8 (add QPR:$src,
5079 (NEONvshrs QPR:$src, (i32 7))))))),
5080 (VABSv16i8 QPR:$src)>;
5081 def : Pat<(xor (v4i32 (bitconvert (v8i16 (NEONvshrs QPR:$src, (i32 15))))),
5082 (v4i32 (bitconvert (v8i16 (add QPR:$src,
5083 (NEONvshrs QPR:$src, (i32 15))))))),
5084 (VABSv8i16 QPR:$src)>;
5085 def : Pat<(xor (v4i32 (NEONvshrs QPR:$src, (i32 31))),
5086 (v4i32 (add QPR:$src, (NEONvshrs QPR:$src, (i32 31))))),
5087 (VABSv4i32 QPR:$src)>;
5089 def : Pat<(v2f32 (int_arm_neon_vabs (v2f32 DPR:$src))), (VABSfd DPR:$src)>;
5090 def : Pat<(v4f32 (int_arm_neon_vabs (v4f32 QPR:$src))), (VABSfq QPR:$src)>;
5092 // VQABS : Vector Saturating Absolute Value
5093 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
5094 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
5095 int_arm_neon_vqabs>;
5099 def vnegd : PatFrag<(ops node:$in),
5100 (sub (bitconvert (v2i32 NEONimmAllZerosV)), node:$in)>;
5101 def vnegq : PatFrag<(ops node:$in),
5102 (sub (bitconvert (v4i32 NEONimmAllZerosV)), node:$in)>;
5104 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
5105 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$Vd), (ins DPR:$Vm),
5106 IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
5107 [(set DPR:$Vd, (Ty (vnegd DPR:$Vm)))]>;
5108 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
5109 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$Vd), (ins QPR:$Vm),
5110 IIC_VSHLiQ, OpcodeStr, Dt, "$Vd, $Vm", "",
5111 [(set QPR:$Vd, (Ty (vnegq QPR:$Vm)))]>;
5113 // VNEG : Vector Negate (integer)
5114 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
5115 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
5116 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
5117 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
5118 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
5119 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
5121 // VNEG : Vector Negate (floating-point)
5122 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
5123 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VUNAD,
5124 "vneg", "f32", "$Vd, $Vm", "",
5125 [(set DPR:$Vd, (v2f32 (fneg DPR:$Vm)))]>;
5126 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
5127 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VUNAQ,
5128 "vneg", "f32", "$Vd, $Vm", "",
5129 [(set QPR:$Vd, (v4f32 (fneg QPR:$Vm)))]>;
5131 def : Pat<(v8i8 (vnegd DPR:$src)), (VNEGs8d DPR:$src)>;
5132 def : Pat<(v4i16 (vnegd DPR:$src)), (VNEGs16d DPR:$src)>;
5133 def : Pat<(v2i32 (vnegd DPR:$src)), (VNEGs32d DPR:$src)>;
5134 def : Pat<(v16i8 (vnegq QPR:$src)), (VNEGs8q QPR:$src)>;
5135 def : Pat<(v8i16 (vnegq QPR:$src)), (VNEGs16q QPR:$src)>;
5136 def : Pat<(v4i32 (vnegq QPR:$src)), (VNEGs32q QPR:$src)>;
5138 // VQNEG : Vector Saturating Negate
5139 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
5140 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
5141 int_arm_neon_vqneg>;
5143 // Vector Bit Counting Operations.
5145 // VCLS : Vector Count Leading Sign Bits
5146 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
5147 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
5149 // VCLZ : Vector Count Leading Zeros
5150 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
5151 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
5153 // VCNT : Vector Count One Bits
5154 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
5155 IIC_VCNTiD, "vcnt", "8",
5157 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
5158 IIC_VCNTiQ, "vcnt", "8",
5159 v16i8, v16i8, ctpop>;
5162 def VSWPd : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 0, 0,
5163 (outs DPR:$Vd, DPR:$Vm), (ins DPR:$in1, DPR:$in2),
5164 NoItinerary, "vswp", "$Vd, $Vm", "$in1 = $Vd, $in2 = $Vm",
5166 def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0,
5167 (outs QPR:$Vd, QPR:$Vm), (ins QPR:$in1, QPR:$in2),
5168 NoItinerary, "vswp", "$Vd, $Vm", "$in1 = $Vd, $in2 = $Vm",
5171 // Vector Move Operations.
5173 // VMOV : Vector Move (Register)
5174 def : NEONInstAlias<"vmov${p} $Vd, $Vm",
5175 (VORRd DPR:$Vd, DPR:$Vm, DPR:$Vm, pred:$p)>;
5176 def : NEONInstAlias<"vmov${p} $Vd, $Vm",
5177 (VORRq QPR:$Vd, QPR:$Vm, QPR:$Vm, pred:$p)>;
5179 // VMOV : Vector Move (Immediate)
5181 let isReMaterializable = 1 in {
5182 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$Vd),
5183 (ins nImmSplatI8:$SIMM), IIC_VMOVImm,
5184 "vmov", "i8", "$Vd, $SIMM", "",
5185 [(set DPR:$Vd, (v8i8 (NEONvmovImm timm:$SIMM)))]>;
5186 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$Vd),
5187 (ins nImmSplatI8:$SIMM), IIC_VMOVImm,
5188 "vmov", "i8", "$Vd, $SIMM", "",
5189 [(set QPR:$Vd, (v16i8 (NEONvmovImm timm:$SIMM)))]>;
5191 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 0, 1, (outs DPR:$Vd),
5192 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
5193 "vmov", "i16", "$Vd, $SIMM", "",
5194 [(set DPR:$Vd, (v4i16 (NEONvmovImm timm:$SIMM)))]> {
5195 let Inst{9} = SIMM{9};
5198 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 0, 1, (outs QPR:$Vd),
5199 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
5200 "vmov", "i16", "$Vd, $SIMM", "",
5201 [(set QPR:$Vd, (v8i16 (NEONvmovImm timm:$SIMM)))]> {
5202 let Inst{9} = SIMM{9};
5205 def VMOVv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 0, 1, (outs DPR:$Vd),
5206 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
5207 "vmov", "i32", "$Vd, $SIMM", "",
5208 [(set DPR:$Vd, (v2i32 (NEONvmovImm timm:$SIMM)))]> {
5209 let Inst{11-8} = SIMM{11-8};
5212 def VMOVv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 0, 1, (outs QPR:$Vd),
5213 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
5214 "vmov", "i32", "$Vd, $SIMM", "",
5215 [(set QPR:$Vd, (v4i32 (NEONvmovImm timm:$SIMM)))]> {
5216 let Inst{11-8} = SIMM{11-8};
5219 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$Vd),
5220 (ins nImmSplatI64:$SIMM), IIC_VMOVImm,
5221 "vmov", "i64", "$Vd, $SIMM", "",
5222 [(set DPR:$Vd, (v1i64 (NEONvmovImm timm:$SIMM)))]>;
5223 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$Vd),
5224 (ins nImmSplatI64:$SIMM), IIC_VMOVImm,
5225 "vmov", "i64", "$Vd, $SIMM", "",
5226 [(set QPR:$Vd, (v2i64 (NEONvmovImm timm:$SIMM)))]>;
5228 def VMOVv2f32 : N1ModImm<1, 0b000, 0b1111, 0, 0, 0, 1, (outs DPR:$Vd),
5229 (ins nImmVMOVF32:$SIMM), IIC_VMOVImm,
5230 "vmov", "f32", "$Vd, $SIMM", "",
5231 [(set DPR:$Vd, (v2f32 (NEONvmovFPImm timm:$SIMM)))]>;
5232 def VMOVv4f32 : N1ModImm<1, 0b000, 0b1111, 0, 1, 0, 1, (outs QPR:$Vd),
5233 (ins nImmVMOVF32:$SIMM), IIC_VMOVImm,
5234 "vmov", "f32", "$Vd, $SIMM", "",
5235 [(set QPR:$Vd, (v4f32 (NEONvmovFPImm timm:$SIMM)))]>;
5236 } // isReMaterializable
5238 // VMOV : Vector Get Lane (move scalar to ARM core register)
5240 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
5241 (outs GPR:$R), (ins DPR:$V, VectorIndex8:$lane),
5242 IIC_VMOVSI, "vmov", "s8", "$R, $V$lane",
5243 [(set GPR:$R, (NEONvgetlanes (v8i8 DPR:$V),
5245 let Inst{21} = lane{2};
5246 let Inst{6-5} = lane{1-0};
5248 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
5249 (outs GPR:$R), (ins DPR:$V, VectorIndex16:$lane),
5250 IIC_VMOVSI, "vmov", "s16", "$R, $V$lane",
5251 [(set GPR:$R, (NEONvgetlanes (v4i16 DPR:$V),
5253 let Inst{21} = lane{1};
5254 let Inst{6} = lane{0};
5256 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
5257 (outs GPR:$R), (ins DPR:$V, VectorIndex8:$lane),
5258 IIC_VMOVSI, "vmov", "u8", "$R, $V$lane",
5259 [(set GPR:$R, (NEONvgetlaneu (v8i8 DPR:$V),
5261 let Inst{21} = lane{2};
5262 let Inst{6-5} = lane{1-0};
5264 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
5265 (outs GPR:$R), (ins DPR:$V, VectorIndex16:$lane),
5266 IIC_VMOVSI, "vmov", "u16", "$R, $V$lane",
5267 [(set GPR:$R, (NEONvgetlaneu (v4i16 DPR:$V),
5269 let Inst{21} = lane{1};
5270 let Inst{6} = lane{0};
5272 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
5273 (outs GPR:$R), (ins DPR:$V, VectorIndex32:$lane),
5274 IIC_VMOVSI, "vmov", "32", "$R, $V$lane",
5275 [(set GPR:$R, (extractelt (v2i32 DPR:$V),
5277 Requires<[HasNEON, HasFastVGETLNi32]> {
5278 let Inst{21} = lane{0};
5280 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
5281 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
5282 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
5283 (DSubReg_i8_reg imm:$lane))),
5284 (SubReg_i8_lane imm:$lane))>;
5285 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
5286 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
5287 (DSubReg_i16_reg imm:$lane))),
5288 (SubReg_i16_lane imm:$lane))>;
5289 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
5290 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
5291 (DSubReg_i8_reg imm:$lane))),
5292 (SubReg_i8_lane imm:$lane))>;
5293 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
5294 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
5295 (DSubReg_i16_reg imm:$lane))),
5296 (SubReg_i16_lane imm:$lane))>;
5297 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
5298 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
5299 (DSubReg_i32_reg imm:$lane))),
5300 (SubReg_i32_lane imm:$lane))>,
5301 Requires<[HasNEON, HasFastVGETLNi32]>;
5302 def : Pat<(extractelt (v2i32 DPR:$src), imm:$lane),
5304 (i32 (EXTRACT_SUBREG DPR:$src, (SSubReg_f32_reg imm:$lane))), GPR)>,
5305 Requires<[HasNEON, HasSlowVGETLNi32]>;
5306 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
5308 (i32 (EXTRACT_SUBREG QPR:$src, (SSubReg_f32_reg imm:$lane))), GPR)>,
5309 Requires<[HasNEON, HasSlowVGETLNi32]>;
5310 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
5311 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
5312 (SSubReg_f32_reg imm:$src2))>;
5313 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
5314 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
5315 (SSubReg_f32_reg imm:$src2))>;
5316 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
5317 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
5318 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
5319 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
5322 // VMOV : Vector Set Lane (move ARM core register to scalar)
5324 let Constraints = "$src1 = $V" in {
5325 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$V),
5326 (ins DPR:$src1, GPR:$R, VectorIndex8:$lane),
5327 IIC_VMOVISL, "vmov", "8", "$V$lane, $R",
5328 [(set DPR:$V, (vector_insert (v8i8 DPR:$src1),
5329 GPR:$R, imm:$lane))]> {
5330 let Inst{21} = lane{2};
5331 let Inst{6-5} = lane{1-0};
5333 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$V),
5334 (ins DPR:$src1, GPR:$R, VectorIndex16:$lane),
5335 IIC_VMOVISL, "vmov", "16", "$V$lane, $R",
5336 [(set DPR:$V, (vector_insert (v4i16 DPR:$src1),
5337 GPR:$R, imm:$lane))]> {
5338 let Inst{21} = lane{1};
5339 let Inst{6} = lane{0};
5341 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$V),
5342 (ins DPR:$src1, GPR:$R, VectorIndex32:$lane),
5343 IIC_VMOVISL, "vmov", "32", "$V$lane, $R",
5344 [(set DPR:$V, (insertelt (v2i32 DPR:$src1),
5345 GPR:$R, imm:$lane))]> {
5346 let Inst{21} = lane{0};
5349 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
5350 (v16i8 (INSERT_SUBREG QPR:$src1,
5351 (v8i8 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
5352 (DSubReg_i8_reg imm:$lane))),
5353 GPR:$src2, (SubReg_i8_lane imm:$lane))),
5354 (DSubReg_i8_reg imm:$lane)))>;
5355 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
5356 (v8i16 (INSERT_SUBREG QPR:$src1,
5357 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
5358 (DSubReg_i16_reg imm:$lane))),
5359 GPR:$src2, (SubReg_i16_lane imm:$lane))),
5360 (DSubReg_i16_reg imm:$lane)))>;
5361 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
5362 (v4i32 (INSERT_SUBREG QPR:$src1,
5363 (v2i32 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
5364 (DSubReg_i32_reg imm:$lane))),
5365 GPR:$src2, (SubReg_i32_lane imm:$lane))),
5366 (DSubReg_i32_reg imm:$lane)))>;
5368 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
5369 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
5370 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
5371 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
5372 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
5373 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
5375 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
5376 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
5377 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
5378 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
5380 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
5381 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
5382 def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
5383 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
5384 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
5385 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
5387 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
5388 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
5389 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
5390 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
5391 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
5392 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
5394 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
5395 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
5396 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
5398 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
5399 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
5400 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
5402 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
5403 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
5404 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
5407 // VDUP : Vector Duplicate (from ARM core register to all elements)
5409 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
5410 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$V), (ins GPR:$R),
5411 IIC_VMOVIS, "vdup", Dt, "$V, $R",
5412 [(set DPR:$V, (Ty (NEONvdup (i32 GPR:$R))))]>;
5413 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
5414 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$V), (ins GPR:$R),
5415 IIC_VMOVIS, "vdup", Dt, "$V, $R",
5416 [(set QPR:$V, (Ty (NEONvdup (i32 GPR:$R))))]>;
5418 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
5419 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
5420 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>,
5421 Requires<[HasNEON, HasFastVDUP32]>;
5422 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
5423 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
5424 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
5426 // NEONvdup patterns for uarchs with fast VDUP.32.
5427 def : Pat<(v2f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VDUP32d GPR:$R)>,
5428 Requires<[HasNEON,HasFastVDUP32]>;
5429 def : Pat<(v4f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VDUP32q GPR:$R)>;
5431 // NEONvdup patterns for uarchs with slow VDUP.32 - use VMOVDRR instead.
5432 def : Pat<(v2i32 (NEONvdup (i32 GPR:$R))), (VMOVDRR GPR:$R, GPR:$R)>,
5433 Requires<[HasNEON,HasSlowVDUP32]>;
5434 def : Pat<(v2f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VMOVDRR GPR:$R, GPR:$R)>,
5435 Requires<[HasNEON,HasSlowVDUP32]>;
5437 // VDUP : Vector Duplicate Lane (from scalar to all elements)
5439 class VDUPLND<bits<4> op19_16, string OpcodeStr, string Dt,
5440 ValueType Ty, Operand IdxTy>
5441 : NVDupLane<op19_16, 0, (outs DPR:$Vd), (ins DPR:$Vm, IdxTy:$lane),
5442 IIC_VMOVD, OpcodeStr, Dt, "$Vd, $Vm$lane",
5443 [(set DPR:$Vd, (Ty (NEONvduplane (Ty DPR:$Vm), imm:$lane)))]>;
5445 class VDUPLNQ<bits<4> op19_16, string OpcodeStr, string Dt,
5446 ValueType ResTy, ValueType OpTy, Operand IdxTy>
5447 : NVDupLane<op19_16, 1, (outs QPR:$Vd), (ins DPR:$Vm, IdxTy:$lane),
5448 IIC_VMOVQ, OpcodeStr, Dt, "$Vd, $Vm$lane",
5449 [(set QPR:$Vd, (ResTy (NEONvduplane (OpTy DPR:$Vm),
5450 VectorIndex32:$lane)))]>;
5452 // Inst{19-16} is partially specified depending on the element size.
5454 def VDUPLN8d : VDUPLND<{?,?,?,1}, "vdup", "8", v8i8, VectorIndex8> {
5456 let Inst{19-17} = lane{2-0};
5458 def VDUPLN16d : VDUPLND<{?,?,1,0}, "vdup", "16", v4i16, VectorIndex16> {
5460 let Inst{19-18} = lane{1-0};
5462 def VDUPLN32d : VDUPLND<{?,1,0,0}, "vdup", "32", v2i32, VectorIndex32> {
5464 let Inst{19} = lane{0};
5466 def VDUPLN8q : VDUPLNQ<{?,?,?,1}, "vdup", "8", v16i8, v8i8, VectorIndex8> {
5468 let Inst{19-17} = lane{2-0};
5470 def VDUPLN16q : VDUPLNQ<{?,?,1,0}, "vdup", "16", v8i16, v4i16, VectorIndex16> {
5472 let Inst{19-18} = lane{1-0};
5474 def VDUPLN32q : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4i32, v2i32, VectorIndex32> {
5476 let Inst{19} = lane{0};
5479 def : Pat<(v2f32 (NEONvduplane (v2f32 DPR:$Vm), imm:$lane)),
5480 (VDUPLN32d DPR:$Vm, imm:$lane)>;
5482 def : Pat<(v4f32 (NEONvduplane (v2f32 DPR:$Vm), imm:$lane)),
5483 (VDUPLN32q DPR:$Vm, imm:$lane)>;
5485 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
5486 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
5487 (DSubReg_i8_reg imm:$lane))),
5488 (SubReg_i8_lane imm:$lane)))>;
5489 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
5490 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
5491 (DSubReg_i16_reg imm:$lane))),
5492 (SubReg_i16_lane imm:$lane)))>;
5493 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
5494 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
5495 (DSubReg_i32_reg imm:$lane))),
5496 (SubReg_i32_lane imm:$lane)))>;
5497 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
5498 (v4f32 (VDUPLN32q (v2f32 (EXTRACT_SUBREG QPR:$src,
5499 (DSubReg_i32_reg imm:$lane))),
5500 (SubReg_i32_lane imm:$lane)))>;
5502 def : Pat<(v2f32 (NEONvdup (f32 SPR:$src))),
5503 (v2f32 (VDUPLN32d (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
5504 SPR:$src, ssub_0), (i32 0)))>;
5505 def : Pat<(v4f32 (NEONvdup (f32 SPR:$src))),
5506 (v4f32 (VDUPLN32q (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
5507 SPR:$src, ssub_0), (i32 0)))>;
5509 // VMOVN : Vector Narrowing Move
5510 defm VMOVN : N2VN_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVN,
5511 "vmovn", "i", trunc>;
5512 // VQMOVN : Vector Saturating Narrowing Move
5513 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
5514 "vqmovn", "s", int_arm_neon_vqmovns>;
5515 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
5516 "vqmovn", "u", int_arm_neon_vqmovnu>;
5517 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
5518 "vqmovun", "s", int_arm_neon_vqmovnsu>;
5519 // VMOVL : Vector Lengthening Move
5520 defm VMOVLs : N2VL_QHS<0b01,0b10100,0,1, "vmovl", "s", sext>;
5521 defm VMOVLu : N2VL_QHS<0b11,0b10100,0,1, "vmovl", "u", zext>;
5522 def : Pat<(v8i16 (anyext (v8i8 DPR:$Vm))), (VMOVLuv8i16 DPR:$Vm)>;
5523 def : Pat<(v4i32 (anyext (v4i16 DPR:$Vm))), (VMOVLuv4i32 DPR:$Vm)>;
5524 def : Pat<(v2i64 (anyext (v2i32 DPR:$Vm))), (VMOVLuv2i64 DPR:$Vm)>;
5526 // Vector Conversions.
5528 // VCVT : Vector Convert Between Floating-Point and Integers
5529 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
5530 v2i32, v2f32, fp_to_sint>;
5531 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
5532 v2i32, v2f32, fp_to_uint>;
5533 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
5534 v2f32, v2i32, sint_to_fp>;
5535 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
5536 v2f32, v2i32, uint_to_fp>;
5538 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
5539 v4i32, v4f32, fp_to_sint>;
5540 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
5541 v4i32, v4f32, fp_to_uint>;
5542 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
5543 v4f32, v4i32, sint_to_fp>;
5544 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
5545 v4f32, v4i32, uint_to_fp>;
5548 multiclass VCVT_FPI<string op, bits<3> op10_8, SDPatternOperator IntS,
5549 SDPatternOperator IntU> {
5550 let PostEncoderMethod = "NEONThumb2V8PostEncoder", DecoderNamespace = "v8NEON" in {
5551 def SD : N2VDIntnp<0b11, op10_8, 0, NoItinerary, !strconcat("vcvt", op),
5552 "s32.f32", v2i32, v2f32, IntS>, Requires<[HasV8, HasNEON]>;
5553 def SQ : N2VQIntnp<0b11, op10_8, 0, NoItinerary, !strconcat("vcvt", op),
5554 "s32.f32", v4i32, v4f32, IntS>, Requires<[HasV8, HasNEON]>;
5555 def UD : N2VDIntnp<0b11, op10_8, 1, NoItinerary, !strconcat("vcvt", op),
5556 "u32.f32", v2i32, v2f32, IntU>, Requires<[HasV8, HasNEON]>;
5557 def UQ : N2VQIntnp<0b11, op10_8, 1, NoItinerary, !strconcat("vcvt", op),
5558 "u32.f32", v4i32, v4f32, IntU>, Requires<[HasV8, HasNEON]>;
5562 defm VCVTAN : VCVT_FPI<"a", 0b000, int_arm_neon_vcvtas, int_arm_neon_vcvtau>;
5563 defm VCVTNN : VCVT_FPI<"n", 0b001, int_arm_neon_vcvtns, int_arm_neon_vcvtnu>;
5564 defm VCVTPN : VCVT_FPI<"p", 0b010, int_arm_neon_vcvtps, int_arm_neon_vcvtpu>;
5565 defm VCVTMN : VCVT_FPI<"m", 0b011, int_arm_neon_vcvtms, int_arm_neon_vcvtmu>;
5567 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
5568 let DecoderMethod = "DecodeVCVTD" in {
5569 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
5570 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
5571 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
5572 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
5573 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
5574 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
5575 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
5576 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
5579 let DecoderMethod = "DecodeVCVTQ" in {
5580 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
5581 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
5582 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
5583 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
5584 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
5585 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
5586 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
5587 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
5590 def : NEONInstAlias<"vcvt${p}.s32.f32 $Dd, $Dm, #0",
5591 (VCVTf2sd DPR:$Dd, DPR:$Dm, pred:$p)>;
5592 def : NEONInstAlias<"vcvt${p}.u32.f32 $Dd, $Dm, #0",
5593 (VCVTf2ud DPR:$Dd, DPR:$Dm, pred:$p)>;
5594 def : NEONInstAlias<"vcvt${p}.f32.s32 $Dd, $Dm, #0",
5595 (VCVTs2fd DPR:$Dd, DPR:$Dm, pred:$p)>;
5596 def : NEONInstAlias<"vcvt${p}.f32.u32 $Dd, $Dm, #0",
5597 (VCVTu2fd DPR:$Dd, DPR:$Dm, pred:$p)>;
5599 def : NEONInstAlias<"vcvt${p}.s32.f32 $Qd, $Qm, #0",
5600 (VCVTf2sq QPR:$Qd, QPR:$Qm, pred:$p)>;
5601 def : NEONInstAlias<"vcvt${p}.u32.f32 $Qd, $Qm, #0",
5602 (VCVTf2uq QPR:$Qd, QPR:$Qm, pred:$p)>;
5603 def : NEONInstAlias<"vcvt${p}.f32.s32 $Qd, $Qm, #0",
5604 (VCVTs2fq QPR:$Qd, QPR:$Qm, pred:$p)>;
5605 def : NEONInstAlias<"vcvt${p}.f32.u32 $Qd, $Qm, #0",
5606 (VCVTu2fq QPR:$Qd, QPR:$Qm, pred:$p)>;
5609 // VCVT : Vector Convert Between Half-Precision and Single-Precision.
5610 def VCVTf2h : N2VNInt<0b11, 0b11, 0b01, 0b10, 0b01100, 0, 0,
5611 IIC_VUNAQ, "vcvt", "f16.f32",
5612 v4i16, v4f32, int_arm_neon_vcvtfp2hf>,
5613 Requires<[HasNEON, HasFP16]>;
5614 def VCVTh2f : N2VLInt<0b11, 0b11, 0b01, 0b10, 0b01110, 0, 0,
5615 IIC_VUNAQ, "vcvt", "f32.f16",
5616 v4f32, v4i16, int_arm_neon_vcvthf2fp>,
5617 Requires<[HasNEON, HasFP16]>;
5621 // VREV64 : Vector Reverse elements within 64-bit doublewords
5623 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5624 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$Vd),
5625 (ins DPR:$Vm), IIC_VMOVD,
5626 OpcodeStr, Dt, "$Vd, $Vm", "",
5627 [(set DPR:$Vd, (Ty (NEONvrev64 (Ty DPR:$Vm))))]>;
5628 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5629 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$Vd),
5630 (ins QPR:$Vm), IIC_VMOVQ,
5631 OpcodeStr, Dt, "$Vd, $Vm", "",
5632 [(set QPR:$Vd, (Ty (NEONvrev64 (Ty QPR:$Vm))))]>;
5634 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
5635 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
5636 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
5637 def : Pat<(v2f32 (NEONvrev64 (v2f32 DPR:$Vm))), (VREV64d32 DPR:$Vm)>;
5639 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
5640 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
5641 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
5642 def : Pat<(v4f32 (NEONvrev64 (v4f32 QPR:$Vm))), (VREV64q32 QPR:$Vm)>;
5644 // VREV32 : Vector Reverse elements within 32-bit words
5646 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5647 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$Vd),
5648 (ins DPR:$Vm), IIC_VMOVD,
5649 OpcodeStr, Dt, "$Vd, $Vm", "",
5650 [(set DPR:$Vd, (Ty (NEONvrev32 (Ty DPR:$Vm))))]>;
5651 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5652 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$Vd),
5653 (ins QPR:$Vm), IIC_VMOVQ,
5654 OpcodeStr, Dt, "$Vd, $Vm", "",
5655 [(set QPR:$Vd, (Ty (NEONvrev32 (Ty QPR:$Vm))))]>;
5657 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
5658 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
5660 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
5661 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
5663 // VREV16 : Vector Reverse elements within 16-bit halfwords
5665 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5666 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$Vd),
5667 (ins DPR:$Vm), IIC_VMOVD,
5668 OpcodeStr, Dt, "$Vd, $Vm", "",
5669 [(set DPR:$Vd, (Ty (NEONvrev16 (Ty DPR:$Vm))))]>;
5670 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5671 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$Vd),
5672 (ins QPR:$Vm), IIC_VMOVQ,
5673 OpcodeStr, Dt, "$Vd, $Vm", "",
5674 [(set QPR:$Vd, (Ty (NEONvrev16 (Ty QPR:$Vm))))]>;
5676 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
5677 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
5679 // Other Vector Shuffles.
5681 // Aligned extractions: really just dropping registers
5683 class AlignedVEXTq<ValueType DestTy, ValueType SrcTy, SDNodeXForm LaneCVT>
5684 : Pat<(DestTy (vector_extract_subvec (SrcTy QPR:$src), (i32 imm:$start))),
5685 (EXTRACT_SUBREG (SrcTy QPR:$src), (LaneCVT imm:$start))>;
5687 def : AlignedVEXTq<v8i8, v16i8, DSubReg_i8_reg>;
5689 def : AlignedVEXTq<v4i16, v8i16, DSubReg_i16_reg>;
5691 def : AlignedVEXTq<v2i32, v4i32, DSubReg_i32_reg>;
5693 def : AlignedVEXTq<v1i64, v2i64, DSubReg_f64_reg>;
5695 def : AlignedVEXTq<v2f32, v4f32, DSubReg_i32_reg>;
5698 // VEXT : Vector Extract
5701 // All of these have a two-operand InstAlias.
5702 let TwoOperandAliasConstraint = "$Vn = $Vd" in {
5703 class VEXTd<string OpcodeStr, string Dt, ValueType Ty, Operand immTy>
5704 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$Vd),
5705 (ins DPR:$Vn, DPR:$Vm, immTy:$index), NVExtFrm,
5706 IIC_VEXTD, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
5707 [(set DPR:$Vd, (Ty (NEONvext (Ty DPR:$Vn),
5708 (Ty DPR:$Vm), imm:$index)))]> {
5711 let Inst{10-8} = index{2-0};
5714 class VEXTq<string OpcodeStr, string Dt, ValueType Ty, Operand immTy>
5715 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$Vd),
5716 (ins QPR:$Vn, QPR:$Vm, imm0_15:$index), NVExtFrm,
5717 IIC_VEXTQ, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
5718 [(set QPR:$Vd, (Ty (NEONvext (Ty QPR:$Vn),
5719 (Ty QPR:$Vm), imm:$index)))]> {
5721 let Inst{11-8} = index{3-0};
5725 def VEXTd8 : VEXTd<"vext", "8", v8i8, imm0_7> {
5726 let Inst{10-8} = index{2-0};
5728 def VEXTd16 : VEXTd<"vext", "16", v4i16, imm0_3> {
5729 let Inst{10-9} = index{1-0};
5732 def VEXTd32 : VEXTd<"vext", "32", v2i32, imm0_1> {
5733 let Inst{10} = index{0};
5734 let Inst{9-8} = 0b00;
5736 def : Pat<(v2f32 (NEONvext (v2f32 DPR:$Vn),
5739 (VEXTd32 DPR:$Vn, DPR:$Vm, imm:$index)>;
5741 def VEXTq8 : VEXTq<"vext", "8", v16i8, imm0_15> {
5742 let Inst{11-8} = index{3-0};
5744 def VEXTq16 : VEXTq<"vext", "16", v8i16, imm0_7> {
5745 let Inst{11-9} = index{2-0};
5748 def VEXTq32 : VEXTq<"vext", "32", v4i32, imm0_3> {
5749 let Inst{11-10} = index{1-0};
5750 let Inst{9-8} = 0b00;
5752 def VEXTq64 : VEXTq<"vext", "64", v2i64, imm0_1> {
5753 let Inst{11} = index{0};
5754 let Inst{10-8} = 0b000;
5756 def : Pat<(v4f32 (NEONvext (v4f32 QPR:$Vn),
5759 (VEXTq32 QPR:$Vn, QPR:$Vm, imm:$index)>;
5761 // VTRN : Vector Transpose
5763 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
5764 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
5765 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
5767 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
5768 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
5769 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
5771 // VUZP : Vector Unzip (Deinterleave)
5773 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
5774 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
5775 // vuzp.32 Dd, Dm is a pseudo-instruction expanded to vtrn.32 Dd, Dm.
5776 def : NEONInstAlias<"vuzp${p}.32 $Dd, $Dm",
5777 (VTRNd32 DPR:$Dd, DPR:$Dm, pred:$p)>;
5779 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
5780 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
5781 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
5783 // VZIP : Vector Zip (Interleave)
5785 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
5786 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
5787 // vzip.32 Dd, Dm is a pseudo-instruction expanded to vtrn.32 Dd, Dm.
5788 def : NEONInstAlias<"vzip${p}.32 $Dd, $Dm",
5789 (VTRNd32 DPR:$Dd, DPR:$Dm, pred:$p)>;
5791 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
5792 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
5793 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
5795 // Vector Table Lookup and Table Extension.
5797 // VTBL : Vector Table Lookup
5798 let DecoderMethod = "DecodeTBLInstruction" in {
5800 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$Vd),
5801 (ins VecListOneD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB1,
5802 "vtbl", "8", "$Vd, $Vn, $Vm", "",
5803 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbl1 VecListOneD:$Vn, DPR:$Vm)))]>;
5804 let hasExtraSrcRegAllocReq = 1 in {
5806 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$Vd),
5807 (ins VecListDPair:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB2,
5808 "vtbl", "8", "$Vd, $Vn, $Vm", "", []>;
5810 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$Vd),
5811 (ins VecListThreeD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB3,
5812 "vtbl", "8", "$Vd, $Vn, $Vm", "", []>;
5814 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$Vd),
5815 (ins VecListFourD:$Vn, DPR:$Vm),
5817 "vtbl", "8", "$Vd, $Vn, $Vm", "", []>;
5818 } // hasExtraSrcRegAllocReq = 1
5821 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB3, "", []>;
5823 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB4, "", []>;
5825 // VTBX : Vector Table Extension
5827 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$Vd),
5828 (ins DPR:$orig, VecListOneD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX1,
5829 "vtbx", "8", "$Vd, $Vn, $Vm", "$orig = $Vd",
5830 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbx1
5831 DPR:$orig, VecListOneD:$Vn, DPR:$Vm)))]>;
5832 let hasExtraSrcRegAllocReq = 1 in {
5834 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$Vd),
5835 (ins DPR:$orig, VecListDPair:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX2,
5836 "vtbx", "8", "$Vd, $Vn, $Vm", "$orig = $Vd", []>;
5838 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$Vd),
5839 (ins DPR:$orig, VecListThreeD:$Vn, DPR:$Vm),
5840 NVTBLFrm, IIC_VTBX3,
5841 "vtbx", "8", "$Vd, $Vn, $Vm",
5844 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$Vd),
5845 (ins DPR:$orig, VecListFourD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX4,
5846 "vtbx", "8", "$Vd, $Vn, $Vm",
5848 } // hasExtraSrcRegAllocReq = 1
5851 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
5852 IIC_VTBX3, "$orig = $dst", []>;
5854 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
5855 IIC_VTBX4, "$orig = $dst", []>;
5856 } // DecoderMethod = "DecodeTBLInstruction"
5858 // VRINT : Vector Rounding
5859 multiclass VRINT_FPI<string op, bits<3> op9_7, SDPatternOperator Int> {
5860 let PostEncoderMethod = "NEONThumb2V8PostEncoder", DecoderNamespace = "v8NEON" in {
5861 def D : N2VDIntnp<0b10, 0b100, 0, NoItinerary,
5862 !strconcat("vrint", op), "f32",
5863 v2f32, v2f32, Int>, Requires<[HasV8, HasNEON]> {
5864 let Inst{9-7} = op9_7;
5866 def Q : N2VQIntnp<0b10, 0b100, 0, NoItinerary,
5867 !strconcat("vrint", op), "f32",
5868 v4f32, v4f32, Int>, Requires<[HasV8, HasNEON]> {
5869 let Inst{9-7} = op9_7;
5873 def : NEONInstAlias<!strconcat("vrint", op, ".f32.f32\t$Dd, $Dm"),
5874 (!cast<Instruction>(NAME#"D") DPR:$Dd, DPR:$Dm)>;
5875 def : NEONInstAlias<!strconcat("vrint", op, ".f32.f32\t$Qd, $Qm"),
5876 (!cast<Instruction>(NAME#"Q") QPR:$Qd, QPR:$Qm)>;
5879 defm VRINTNN : VRINT_FPI<"n", 0b000, int_arm_neon_vrintn>;
5880 defm VRINTXN : VRINT_FPI<"x", 0b001, int_arm_neon_vrintx>;
5881 defm VRINTAN : VRINT_FPI<"a", 0b010, int_arm_neon_vrinta>;
5882 defm VRINTZN : VRINT_FPI<"z", 0b011, int_arm_neon_vrintz>;
5883 defm VRINTMN : VRINT_FPI<"m", 0b101, int_arm_neon_vrintm>;
5884 defm VRINTPN : VRINT_FPI<"p", 0b111, int_arm_neon_vrintp>;
5886 // Cryptography instructions
5887 let PostEncoderMethod = "NEONThumb2DataIPostEncoder",
5888 DecoderNamespace = "v8Crypto", hasSideEffects = 0 in {
5889 class AES<string op, bit op7, bit op6, SDPatternOperator Int>
5890 : N2VQIntXnp<0b00, 0b00, 0b011, op6, op7, NoItinerary,
5891 !strconcat("aes", op), "8", v16i8, v16i8, Int>,
5892 Requires<[HasV8, HasCrypto]>;
5893 class AES2Op<string op, bit op7, bit op6, SDPatternOperator Int>
5894 : N2VQIntX2np<0b00, 0b00, 0b011, op6, op7, NoItinerary,
5895 !strconcat("aes", op), "8", v16i8, v16i8, Int>,
5896 Requires<[HasV8, HasCrypto]>;
5897 class N2SHA<string op, bits<2> op17_16, bits<3> op10_8, bit op7, bit op6,
5898 SDPatternOperator Int>
5899 : N2VQIntXnp<0b10, op17_16, op10_8, op6, op7, NoItinerary,
5900 !strconcat("sha", op), "32", v4i32, v4i32, Int>,
5901 Requires<[HasV8, HasCrypto]>;
5902 class N2SHA2Op<string op, bits<2> op17_16, bits<3> op10_8, bit op7, bit op6,
5903 SDPatternOperator Int>
5904 : N2VQIntX2np<0b10, op17_16, op10_8, op6, op7, NoItinerary,
5905 !strconcat("sha", op), "32", v4i32, v4i32, Int>,
5906 Requires<[HasV8, HasCrypto]>;
5907 class N3SHA3Op<string op, bits<5> op27_23, bits<2> op21_20, SDPatternOperator Int>
5908 : N3VQInt3np<op27_23, op21_20, 0b1100, 1, 0, N3RegFrm, NoItinerary,
5909 !strconcat("sha", op), "32", v4i32, v4i32, Int, 0>,
5910 Requires<[HasV8, HasCrypto]>;
5913 def AESD : AES2Op<"d", 0, 1, int_arm_neon_aesd>;
5914 def AESE : AES2Op<"e", 0, 0, int_arm_neon_aese>;
5915 def AESIMC : AES<"imc", 1, 1, int_arm_neon_aesimc>;
5916 def AESMC : AES<"mc", 1, 0, int_arm_neon_aesmc>;
5918 def SHA1H : N2SHA<"1h", 0b01, 0b010, 1, 1, null_frag>;
5919 def SHA1SU1 : N2SHA2Op<"1su1", 0b10, 0b011, 1, 0, int_arm_neon_sha1su1>;
5920 def SHA256SU0 : N2SHA2Op<"256su0", 0b10, 0b011, 1, 1, int_arm_neon_sha256su0>;
5921 def SHA1C : N3SHA3Op<"1c", 0b00100, 0b00, null_frag>;
5922 def SHA1M : N3SHA3Op<"1m", 0b00100, 0b10, null_frag>;
5923 def SHA1P : N3SHA3Op<"1p", 0b00100, 0b01, null_frag>;
5924 def SHA1SU0 : N3SHA3Op<"1su0", 0b00100, 0b11, int_arm_neon_sha1su0>;
5925 def SHA256H : N3SHA3Op<"256h", 0b00110, 0b00, int_arm_neon_sha256h>;
5926 def SHA256H2 : N3SHA3Op<"256h2", 0b00110, 0b01, int_arm_neon_sha256h2>;
5927 def SHA256SU1 : N3SHA3Op<"256su1", 0b00110, 0b10, int_arm_neon_sha256su1>;
5929 def : Pat<(i32 (int_arm_neon_sha1h i32:$Rn)),
5930 (COPY_TO_REGCLASS (f32 (EXTRACT_SUBREG
5931 (SHA1H (SUBREG_TO_REG (i64 0),
5932 (f32 (COPY_TO_REGCLASS i32:$Rn, SPR)),
5936 def : Pat<(v4i32 (int_arm_neon_sha1c v4i32:$hash_abcd, i32:$hash_e, v4i32:$wk)),
5937 (SHA1C v4i32:$hash_abcd,
5938 (SUBREG_TO_REG (i64 0),
5939 (f32 (COPY_TO_REGCLASS i32:$hash_e, SPR)),
5943 def : Pat<(v4i32 (int_arm_neon_sha1m v4i32:$hash_abcd, i32:$hash_e, v4i32:$wk)),
5944 (SHA1M v4i32:$hash_abcd,
5945 (SUBREG_TO_REG (i64 0),
5946 (f32 (COPY_TO_REGCLASS i32:$hash_e, SPR)),
5950 def : Pat<(v4i32 (int_arm_neon_sha1p v4i32:$hash_abcd, i32:$hash_e, v4i32:$wk)),
5951 (SHA1P v4i32:$hash_abcd,
5952 (SUBREG_TO_REG (i64 0),
5953 (f32 (COPY_TO_REGCLASS i32:$hash_e, SPR)),
5957 //===----------------------------------------------------------------------===//
5958 // NEON instructions for single-precision FP math
5959 //===----------------------------------------------------------------------===//
5961 class N2VSPat<SDNode OpNode, NeonI Inst>
5962 : NEONFPPat<(f32 (OpNode SPR:$a)),
5964 (v2f32 (COPY_TO_REGCLASS (Inst
5966 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5967 SPR:$a, ssub_0)), DPR_VFP2)), ssub_0)>;
5969 class N3VSPat<SDNode OpNode, NeonI Inst>
5970 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
5972 (v2f32 (COPY_TO_REGCLASS (Inst
5974 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5977 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5978 SPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>;
5980 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
5981 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
5983 (v2f32 (COPY_TO_REGCLASS (Inst
5985 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5988 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5991 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5992 SPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>;
5994 def : N3VSPat<fadd, VADDfd>;
5995 def : N3VSPat<fsub, VSUBfd>;
5996 def : N3VSPat<fmul, VMULfd>;
5997 def : N3VSMulOpPat<fmul, fadd, VMLAfd>,
5998 Requires<[HasNEON, UseNEONForFP, UseFPVMLx, DontUseFusedMAC]>;
5999 def : N3VSMulOpPat<fmul, fsub, VMLSfd>,
6000 Requires<[HasNEON, UseNEONForFP, UseFPVMLx, DontUseFusedMAC]>;
6001 def : N3VSMulOpPat<fmul, fadd, VFMAfd>,
6002 Requires<[HasVFP4, UseNEONForFP, UseFusedMAC]>;
6003 def : N3VSMulOpPat<fmul, fsub, VFMSfd>,
6004 Requires<[HasVFP4, UseNEONForFP, UseFusedMAC]>;
6005 def : N2VSPat<fabs, VABSfd>;
6006 def : N2VSPat<fneg, VNEGfd>;
6007 def : N3VSPat<NEONfmax, VMAXfd>;
6008 def : N3VSPat<NEONfmin, VMINfd>;
6009 def : N2VSPat<arm_ftosi, VCVTf2sd>;
6010 def : N2VSPat<arm_ftoui, VCVTf2ud>;
6011 def : N2VSPat<arm_sitof, VCVTs2fd>;
6012 def : N2VSPat<arm_uitof, VCVTu2fd>;
6014 // Prefer VMOVDRR for i32 -> f32 bitcasts, it can write all DPR registers.
6015 def : Pat<(f32 (bitconvert GPR:$a)),
6016 (EXTRACT_SUBREG (VMOVDRR GPR:$a, GPR:$a), ssub_0)>,
6017 Requires<[HasNEON, DontUseVMOVSR]>;
6019 //===----------------------------------------------------------------------===//
6020 // Non-Instruction Patterns
6021 //===----------------------------------------------------------------------===//
6024 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
6025 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
6026 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
6027 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
6028 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
6029 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
6030 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
6031 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
6032 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
6033 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
6034 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
6035 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
6036 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
6037 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
6038 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
6039 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
6040 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
6041 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
6042 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
6043 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
6044 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
6045 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
6046 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
6047 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
6048 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
6049 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
6050 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
6051 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
6052 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
6053 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
6055 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
6056 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
6057 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
6058 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
6059 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
6060 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
6061 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
6062 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
6063 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
6064 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
6065 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
6066 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
6067 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
6068 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
6069 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
6070 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
6071 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
6072 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
6073 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
6074 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
6075 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
6076 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
6077 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
6078 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
6079 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
6080 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
6081 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
6082 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
6083 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
6084 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;
6086 // Fold extracting an element out of a v2i32 into a vfp register.
6087 def : Pat<(f32 (bitconvert (i32 (extractelt (v2i32 DPR:$src), imm:$lane)))),
6088 (f32 (EXTRACT_SUBREG DPR:$src, (SSubReg_f32_reg imm:$lane)))>;
6090 // Vector lengthening move with load, matching extending loads.
6092 // extload, zextload and sextload for a standard lengthening load. Example:
6093 // Lengthen_Single<"8", "i16", "8"> =
6094 // Pat<(v8i16 (extloadvi8 addrmode6:$addr))
6095 // (VMOVLuv8i16 (VLD1d8 addrmode6:$addr,
6096 // (f64 (IMPLICIT_DEF)), (i32 0)))>;
6097 multiclass Lengthen_Single<string DestLanes, string DestTy, string SrcTy> {
6098 let AddedComplexity = 10 in {
6099 def _Any : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
6100 (!cast<PatFrag>("extloadvi" # SrcTy) addrmode6:$addr)),
6101 (!cast<Instruction>("VMOVLuv" # DestLanes # DestTy)
6102 (!cast<Instruction>("VLD1d" # SrcTy) addrmode6:$addr))>;
6104 def _Z : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
6105 (!cast<PatFrag>("zextloadvi" # SrcTy) addrmode6:$addr)),
6106 (!cast<Instruction>("VMOVLuv" # DestLanes # DestTy)
6107 (!cast<Instruction>("VLD1d" # SrcTy) addrmode6:$addr))>;
6109 def _S : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
6110 (!cast<PatFrag>("sextloadvi" # SrcTy) addrmode6:$addr)),
6111 (!cast<Instruction>("VMOVLsv" # DestLanes # DestTy)
6112 (!cast<Instruction>("VLD1d" # SrcTy) addrmode6:$addr))>;
6116 // extload, zextload and sextload for a lengthening load which only uses
6117 // half the lanes available. Example:
6118 // Lengthen_HalfSingle<"4", "i16", "8", "i16", "i8"> =
6119 // Pat<(v4i16 (extloadvi8 addrmode6oneL32:$addr)),
6120 // (EXTRACT_SUBREG (VMOVLuv8i16 (VLD1LNd32 addrmode6oneL32:$addr,
6121 // (f64 (IMPLICIT_DEF)), (i32 0))),
6123 multiclass Lengthen_HalfSingle<string DestLanes, string DestTy, string SrcTy,
6124 string InsnLanes, string InsnTy> {
6125 def _Any : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
6126 (!cast<PatFrag>("extloadv" # SrcTy) addrmode6oneL32:$addr)),
6127 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # InsnLanes # InsnTy)
6128 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
6130 def _Z : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
6131 (!cast<PatFrag>("zextloadv" # SrcTy) addrmode6oneL32:$addr)),
6132 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # InsnLanes # InsnTy)
6133 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
6135 def _S : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
6136 (!cast<PatFrag>("sextloadv" # SrcTy) addrmode6oneL32:$addr)),
6137 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLsv" # InsnLanes # InsnTy)
6138 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
6142 // extload, zextload and sextload for a lengthening load followed by another
6143 // lengthening load, to quadruple the initial length.
6145 // Lengthen_Double<"4", "i32", "i8", "8", "i16", "4", "i32"> =
6146 // Pat<(v4i32 (extloadvi8 addrmode6oneL32:$addr))
6147 // (EXTRACT_SUBREG (VMOVLuv4i32
6148 // (EXTRACT_SUBREG (VMOVLuv8i16 (VLD1LNd32 addrmode6oneL32:$addr,
6149 // (f64 (IMPLICIT_DEF)),
6153 multiclass Lengthen_Double<string DestLanes, string DestTy, string SrcTy,
6154 string Insn1Lanes, string Insn1Ty, string Insn2Lanes,
6156 def _Any : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
6157 (!cast<PatFrag>("extloadv" # SrcTy) addrmode6oneL32:$addr)),
6158 (!cast<Instruction>("VMOVLuv" # Insn2Lanes # Insn2Ty)
6159 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn1Lanes # Insn1Ty)
6160 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
6162 def _Z : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
6163 (!cast<PatFrag>("zextloadv" # SrcTy) addrmode6oneL32:$addr)),
6164 (!cast<Instruction>("VMOVLuv" # Insn2Lanes # Insn2Ty)
6165 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn1Lanes # Insn1Ty)
6166 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
6168 def _S : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
6169 (!cast<PatFrag>("sextloadv" # SrcTy) addrmode6oneL32:$addr)),
6170 (!cast<Instruction>("VMOVLsv" # Insn2Lanes # Insn2Ty)
6171 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLsv" # Insn1Lanes # Insn1Ty)
6172 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
6176 // extload, zextload and sextload for a lengthening load followed by another
6177 // lengthening load, to quadruple the initial length, but which ends up only
6178 // requiring half the available lanes (a 64-bit outcome instead of a 128-bit).
6180 // Lengthen_HalfDouble<"2", "i32", "i8", "8", "i16", "4", "i32"> =
6181 // Pat<(v2i32 (extloadvi8 addrmode6:$addr))
6182 // (EXTRACT_SUBREG (VMOVLuv4i32
6183 // (EXTRACT_SUBREG (VMOVLuv8i16 (VLD1LNd16 addrmode6:$addr,
6184 // (f64 (IMPLICIT_DEF)), (i32 0))),
6187 multiclass Lengthen_HalfDouble<string DestLanes, string DestTy, string SrcTy,
6188 string Insn1Lanes, string Insn1Ty, string Insn2Lanes,
6190 def _Any : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
6191 (!cast<PatFrag>("extloadv" # SrcTy) addrmode6:$addr)),
6192 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn2Lanes # Insn2Ty)
6193 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn1Lanes # Insn1Ty)
6194 (VLD1LNd16 addrmode6:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
6197 def _Z : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
6198 (!cast<PatFrag>("zextloadv" # SrcTy) addrmode6:$addr)),
6199 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn2Lanes # Insn2Ty)
6200 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn1Lanes # Insn1Ty)
6201 (VLD1LNd16 addrmode6:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
6204 def _S : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
6205 (!cast<PatFrag>("sextloadv" # SrcTy) addrmode6:$addr)),
6206 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLsv" # Insn2Lanes # Insn2Ty)
6207 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLsv" # Insn1Lanes # Insn1Ty)
6208 (VLD1LNd16 addrmode6:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
6213 defm : Lengthen_Single<"8", "i16", "8">; // v8i8 -> v8i16
6214 defm : Lengthen_Single<"4", "i32", "16">; // v4i16 -> v4i32
6215 defm : Lengthen_Single<"2", "i64", "32">; // v2i32 -> v2i64
6217 defm : Lengthen_HalfSingle<"4", "i16", "i8", "8", "i16">; // v4i8 -> v4i16
6218 defm : Lengthen_HalfSingle<"2", "i32", "i16", "4", "i32">; // v2i16 -> v2i32
6220 // Double lengthening - v4i8 -> v4i16 -> v4i32
6221 defm : Lengthen_Double<"4", "i32", "i8", "8", "i16", "4", "i32">;
6222 // v2i8 -> v2i16 -> v2i32
6223 defm : Lengthen_HalfDouble<"2", "i32", "i8", "8", "i16", "4", "i32">;
6224 // v2i16 -> v2i32 -> v2i64
6225 defm : Lengthen_Double<"2", "i64", "i16", "4", "i32", "2", "i64">;
6227 // Triple lengthening - v2i8 -> v2i16 -> v2i32 -> v2i64
6228 def : Pat<(v2i64 (extloadvi8 addrmode6:$addr)),
6229 (VMOVLuv2i64 (EXTRACT_SUBREG (VMOVLuv4i32 (EXTRACT_SUBREG (VMOVLuv8i16
6230 (VLD1LNd16 addrmode6:$addr,
6231 (f64 (IMPLICIT_DEF)), (i32 0))), dsub_0)), dsub_0))>;
6232 def : Pat<(v2i64 (zextloadvi8 addrmode6:$addr)),
6233 (VMOVLuv2i64 (EXTRACT_SUBREG (VMOVLuv4i32 (EXTRACT_SUBREG (VMOVLuv8i16
6234 (VLD1LNd16 addrmode6:$addr,
6235 (f64 (IMPLICIT_DEF)), (i32 0))), dsub_0)), dsub_0))>;
6236 def : Pat<(v2i64 (sextloadvi8 addrmode6:$addr)),
6237 (VMOVLsv2i64 (EXTRACT_SUBREG (VMOVLsv4i32 (EXTRACT_SUBREG (VMOVLsv8i16
6238 (VLD1LNd16 addrmode6:$addr,
6239 (f64 (IMPLICIT_DEF)), (i32 0))), dsub_0)), dsub_0))>;
6241 //===----------------------------------------------------------------------===//
6242 // Assembler aliases
6245 def : VFP2InstAlias<"fmdhr${p} $Dd, $Rn",
6246 (VSETLNi32 DPR:$Dd, GPR:$Rn, 1, pred:$p)>;
6247 def : VFP2InstAlias<"fmdlr${p} $Dd, $Rn",
6248 (VSETLNi32 DPR:$Dd, GPR:$Rn, 0, pred:$p)>;
6250 // VAND/VBIC/VEOR/VORR accept but do not require a type suffix.
6251 defm : NEONDTAnyInstAlias<"vand${p}", "$Vd, $Vn, $Vm",
6252 (VANDd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
6253 defm : NEONDTAnyInstAlias<"vand${p}", "$Vd, $Vn, $Vm",
6254 (VANDq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
6255 defm : NEONDTAnyInstAlias<"vbic${p}", "$Vd, $Vn, $Vm",
6256 (VBICd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
6257 defm : NEONDTAnyInstAlias<"vbic${p}", "$Vd, $Vn, $Vm",
6258 (VBICq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
6259 defm : NEONDTAnyInstAlias<"veor${p}", "$Vd, $Vn, $Vm",
6260 (VEORd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
6261 defm : NEONDTAnyInstAlias<"veor${p}", "$Vd, $Vn, $Vm",
6262 (VEORq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
6263 defm : NEONDTAnyInstAlias<"vorr${p}", "$Vd, $Vn, $Vm",
6264 (VORRd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
6265 defm : NEONDTAnyInstAlias<"vorr${p}", "$Vd, $Vn, $Vm",
6266 (VORRq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
6267 // ... two-operand aliases
6268 defm : NEONDTAnyInstAlias<"vand${p}", "$Vdn, $Vm",
6269 (VANDd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
6270 defm : NEONDTAnyInstAlias<"vand${p}", "$Vdn, $Vm",
6271 (VANDq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
6272 defm : NEONDTAnyInstAlias<"veor${p}", "$Vdn, $Vm",
6273 (VEORd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
6274 defm : NEONDTAnyInstAlias<"veor${p}", "$Vdn, $Vm",
6275 (VEORq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
6276 defm : NEONDTAnyInstAlias<"vorr${p}", "$Vdn, $Vm",
6277 (VORRd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
6278 defm : NEONDTAnyInstAlias<"vorr${p}", "$Vdn, $Vm",
6279 (VORRq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
6281 // VLD1 single-lane pseudo-instructions. These need special handling for
6282 // the lane index that an InstAlias can't handle, so we use these instead.
6283 def VLD1LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vld1${p}", ".8", "$list, $addr",
6284 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6285 def VLD1LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vld1${p}", ".16", "$list, $addr",
6286 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6287 def VLD1LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vld1${p}", ".32", "$list, $addr",
6288 (ins VecListOneDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6290 def VLD1LNdWB_fixed_Asm_8 :
6291 NEONDataTypeAsmPseudoInst<"vld1${p}", ".8", "$list, $addr!",
6292 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6293 def VLD1LNdWB_fixed_Asm_16 :
6294 NEONDataTypeAsmPseudoInst<"vld1${p}", ".16", "$list, $addr!",
6295 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6296 def VLD1LNdWB_fixed_Asm_32 :
6297 NEONDataTypeAsmPseudoInst<"vld1${p}", ".32", "$list, $addr!",
6298 (ins VecListOneDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6299 def VLD1LNdWB_register_Asm_8 :
6300 NEONDataTypeAsmPseudoInst<"vld1${p}", ".8", "$list, $addr, $Rm",
6301 (ins VecListOneDByteIndexed:$list, addrmode6:$addr,
6302 rGPR:$Rm, pred:$p)>;
6303 def VLD1LNdWB_register_Asm_16 :
6304 NEONDataTypeAsmPseudoInst<"vld1${p}", ".16", "$list, $addr, $Rm",
6305 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr,
6306 rGPR:$Rm, pred:$p)>;
6307 def VLD1LNdWB_register_Asm_32 :
6308 NEONDataTypeAsmPseudoInst<"vld1${p}", ".32", "$list, $addr, $Rm",
6309 (ins VecListOneDWordIndexed:$list, addrmode6:$addr,
6310 rGPR:$Rm, pred:$p)>;
6313 // VST1 single-lane pseudo-instructions. These need special handling for
6314 // the lane index that an InstAlias can't handle, so we use these instead.
6315 def VST1LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vst1${p}", ".8", "$list, $addr",
6316 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6317 def VST1LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vst1${p}", ".16", "$list, $addr",
6318 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6319 def VST1LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vst1${p}", ".32", "$list, $addr",
6320 (ins VecListOneDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6322 def VST1LNdWB_fixed_Asm_8 :
6323 NEONDataTypeAsmPseudoInst<"vst1${p}", ".8", "$list, $addr!",
6324 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6325 def VST1LNdWB_fixed_Asm_16 :
6326 NEONDataTypeAsmPseudoInst<"vst1${p}", ".16", "$list, $addr!",
6327 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6328 def VST1LNdWB_fixed_Asm_32 :
6329 NEONDataTypeAsmPseudoInst<"vst1${p}", ".32", "$list, $addr!",
6330 (ins VecListOneDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6331 def VST1LNdWB_register_Asm_8 :
6332 NEONDataTypeAsmPseudoInst<"vst1${p}", ".8", "$list, $addr, $Rm",
6333 (ins VecListOneDByteIndexed:$list, addrmode6:$addr,
6334 rGPR:$Rm, pred:$p)>;
6335 def VST1LNdWB_register_Asm_16 :
6336 NEONDataTypeAsmPseudoInst<"vst1${p}", ".16", "$list, $addr, $Rm",
6337 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr,
6338 rGPR:$Rm, pred:$p)>;
6339 def VST1LNdWB_register_Asm_32 :
6340 NEONDataTypeAsmPseudoInst<"vst1${p}", ".32", "$list, $addr, $Rm",
6341 (ins VecListOneDWordIndexed:$list, addrmode6:$addr,
6342 rGPR:$Rm, pred:$p)>;
6344 // VLD2 single-lane pseudo-instructions. These need special handling for
6345 // the lane index that an InstAlias can't handle, so we use these instead.
6346 def VLD2LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vld2${p}", ".8", "$list, $addr",
6347 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6348 def VLD2LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr",
6349 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6350 def VLD2LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr",
6351 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6352 def VLD2LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr",
6353 (ins VecListTwoQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6354 def VLD2LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr",
6355 (ins VecListTwoQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6357 def VLD2LNdWB_fixed_Asm_8 :
6358 NEONDataTypeAsmPseudoInst<"vld2${p}", ".8", "$list, $addr!",
6359 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6360 def VLD2LNdWB_fixed_Asm_16 :
6361 NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr!",
6362 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6363 def VLD2LNdWB_fixed_Asm_32 :
6364 NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr!",
6365 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6366 def VLD2LNqWB_fixed_Asm_16 :
6367 NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr!",
6368 (ins VecListTwoQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6369 def VLD2LNqWB_fixed_Asm_32 :
6370 NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr!",
6371 (ins VecListTwoQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6372 def VLD2LNdWB_register_Asm_8 :
6373 NEONDataTypeAsmPseudoInst<"vld2${p}", ".8", "$list, $addr, $Rm",
6374 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr,
6375 rGPR:$Rm, pred:$p)>;
6376 def VLD2LNdWB_register_Asm_16 :
6377 NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr, $Rm",
6378 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr,
6379 rGPR:$Rm, pred:$p)>;
6380 def VLD2LNdWB_register_Asm_32 :
6381 NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr, $Rm",
6382 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr,
6383 rGPR:$Rm, pred:$p)>;
6384 def VLD2LNqWB_register_Asm_16 :
6385 NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr, $Rm",
6386 (ins VecListTwoQHWordIndexed:$list, addrmode6:$addr,
6387 rGPR:$Rm, pred:$p)>;
6388 def VLD2LNqWB_register_Asm_32 :
6389 NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr, $Rm",
6390 (ins VecListTwoQWordIndexed:$list, addrmode6:$addr,
6391 rGPR:$Rm, pred:$p)>;
6394 // VST2 single-lane pseudo-instructions. These need special handling for
6395 // the lane index that an InstAlias can't handle, so we use these instead.
6396 def VST2LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vst2${p}", ".8", "$list, $addr",
6397 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6398 def VST2LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vst2${p}", ".16", "$list, $addr",
6399 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6400 def VST2LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr",
6401 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6402 def VST2LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vst2${p}", ".16", "$list, $addr",
6403 (ins VecListTwoQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6404 def VST2LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr",
6405 (ins VecListTwoQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6407 def VST2LNdWB_fixed_Asm_8 :
6408 NEONDataTypeAsmPseudoInst<"vst2${p}", ".8", "$list, $addr!",
6409 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6410 def VST2LNdWB_fixed_Asm_16 :
6411 NEONDataTypeAsmPseudoInst<"vst2${p}", ".16", "$list, $addr!",
6412 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6413 def VST2LNdWB_fixed_Asm_32 :
6414 NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr!",
6415 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6416 def VST2LNqWB_fixed_Asm_16 :
6417 NEONDataTypeAsmPseudoInst<"vst2${p}", ".16", "$list, $addr!",
6418 (ins VecListTwoQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6419 def VST2LNqWB_fixed_Asm_32 :
6420 NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr!",
6421 (ins VecListTwoQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6422 def VST2LNdWB_register_Asm_8 :
6423 NEONDataTypeAsmPseudoInst<"vst2${p}", ".8", "$list, $addr, $Rm",
6424 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr,
6425 rGPR:$Rm, pred:$p)>;
6426 def VST2LNdWB_register_Asm_16 :
6427 NEONDataTypeAsmPseudoInst<"vst2${p}", ".16","$list, $addr, $Rm",
6428 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr,
6429 rGPR:$Rm, pred:$p)>;
6430 def VST2LNdWB_register_Asm_32 :
6431 NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr, $Rm",
6432 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr,
6433 rGPR:$Rm, pred:$p)>;
6434 def VST2LNqWB_register_Asm_16 :
6435 NEONDataTypeAsmPseudoInst<"vst2${p}", ".16","$list, $addr, $Rm",
6436 (ins VecListTwoQHWordIndexed:$list, addrmode6:$addr,
6437 rGPR:$Rm, pred:$p)>;
6438 def VST2LNqWB_register_Asm_32 :
6439 NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr, $Rm",
6440 (ins VecListTwoQWordIndexed:$list, addrmode6:$addr,
6441 rGPR:$Rm, pred:$p)>;
6443 // VLD3 all-lanes pseudo-instructions. These need special handling for
6444 // the lane index that an InstAlias can't handle, so we use these instead.
6445 def VLD3DUPdAsm_8 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr",
6446 (ins VecListThreeDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6447 def VLD3DUPdAsm_16: NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
6448 (ins VecListThreeDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6449 def VLD3DUPdAsm_32: NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
6450 (ins VecListThreeDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6451 def VLD3DUPqAsm_8 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr",
6452 (ins VecListThreeQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6453 def VLD3DUPqAsm_16: NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
6454 (ins VecListThreeQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6455 def VLD3DUPqAsm_32: NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
6456 (ins VecListThreeQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6458 def VLD3DUPdWB_fixed_Asm_8 :
6459 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr!",
6460 (ins VecListThreeDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6461 def VLD3DUPdWB_fixed_Asm_16 :
6462 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
6463 (ins VecListThreeDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6464 def VLD3DUPdWB_fixed_Asm_32 :
6465 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
6466 (ins VecListThreeDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6467 def VLD3DUPqWB_fixed_Asm_8 :
6468 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr!",
6469 (ins VecListThreeQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6470 def VLD3DUPqWB_fixed_Asm_16 :
6471 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
6472 (ins VecListThreeQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6473 def VLD3DUPqWB_fixed_Asm_32 :
6474 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
6475 (ins VecListThreeQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6476 def VLD3DUPdWB_register_Asm_8 :
6477 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr, $Rm",
6478 (ins VecListThreeDAllLanes:$list, addrmode6:$addr,
6479 rGPR:$Rm, pred:$p)>;
6480 def VLD3DUPdWB_register_Asm_16 :
6481 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
6482 (ins VecListThreeDAllLanes:$list, addrmode6:$addr,
6483 rGPR:$Rm, pred:$p)>;
6484 def VLD3DUPdWB_register_Asm_32 :
6485 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
6486 (ins VecListThreeDAllLanes:$list, addrmode6:$addr,
6487 rGPR:$Rm, pred:$p)>;
6488 def VLD3DUPqWB_register_Asm_8 :
6489 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr, $Rm",
6490 (ins VecListThreeQAllLanes:$list, addrmode6:$addr,
6491 rGPR:$Rm, pred:$p)>;
6492 def VLD3DUPqWB_register_Asm_16 :
6493 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
6494 (ins VecListThreeQAllLanes:$list, addrmode6:$addr,
6495 rGPR:$Rm, pred:$p)>;
6496 def VLD3DUPqWB_register_Asm_32 :
6497 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
6498 (ins VecListThreeQAllLanes:$list, addrmode6:$addr,
6499 rGPR:$Rm, pred:$p)>;
6502 // VLD3 single-lane pseudo-instructions. These need special handling for
6503 // the lane index that an InstAlias can't handle, so we use these instead.
6504 def VLD3LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr",
6505 (ins VecListThreeDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6506 def VLD3LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
6507 (ins VecListThreeDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6508 def VLD3LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
6509 (ins VecListThreeDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6510 def VLD3LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
6511 (ins VecListThreeQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6512 def VLD3LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
6513 (ins VecListThreeQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6515 def VLD3LNdWB_fixed_Asm_8 :
6516 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr!",
6517 (ins VecListThreeDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6518 def VLD3LNdWB_fixed_Asm_16 :
6519 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
6520 (ins VecListThreeDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6521 def VLD3LNdWB_fixed_Asm_32 :
6522 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
6523 (ins VecListThreeDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6524 def VLD3LNqWB_fixed_Asm_16 :
6525 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
6526 (ins VecListThreeQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6527 def VLD3LNqWB_fixed_Asm_32 :
6528 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
6529 (ins VecListThreeQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6530 def VLD3LNdWB_register_Asm_8 :
6531 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr, $Rm",
6532 (ins VecListThreeDByteIndexed:$list, addrmode6:$addr,
6533 rGPR:$Rm, pred:$p)>;
6534 def VLD3LNdWB_register_Asm_16 :
6535 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
6536 (ins VecListThreeDHWordIndexed:$list, addrmode6:$addr,
6537 rGPR:$Rm, pred:$p)>;
6538 def VLD3LNdWB_register_Asm_32 :
6539 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
6540 (ins VecListThreeDWordIndexed:$list, addrmode6:$addr,
6541 rGPR:$Rm, pred:$p)>;
6542 def VLD3LNqWB_register_Asm_16 :
6543 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
6544 (ins VecListThreeQHWordIndexed:$list, addrmode6:$addr,
6545 rGPR:$Rm, pred:$p)>;
6546 def VLD3LNqWB_register_Asm_32 :
6547 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
6548 (ins VecListThreeQWordIndexed:$list, addrmode6:$addr,
6549 rGPR:$Rm, pred:$p)>;
6551 // VLD3 multiple structure pseudo-instructions. These need special handling for
6552 // the vector operands that the normal instructions don't yet model.
6553 // FIXME: Remove these when the register classes and instructions are updated.
6554 def VLD3dAsm_8 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr",
6555 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6556 def VLD3dAsm_16 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
6557 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6558 def VLD3dAsm_32 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
6559 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6560 def VLD3qAsm_8 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr",
6561 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6562 def VLD3qAsm_16 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
6563 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6564 def VLD3qAsm_32 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
6565 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6567 def VLD3dWB_fixed_Asm_8 :
6568 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr!",
6569 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6570 def VLD3dWB_fixed_Asm_16 :
6571 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
6572 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6573 def VLD3dWB_fixed_Asm_32 :
6574 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
6575 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6576 def VLD3qWB_fixed_Asm_8 :
6577 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr!",
6578 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6579 def VLD3qWB_fixed_Asm_16 :
6580 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
6581 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6582 def VLD3qWB_fixed_Asm_32 :
6583 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
6584 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6585 def VLD3dWB_register_Asm_8 :
6586 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr, $Rm",
6587 (ins VecListThreeD:$list, addrmode6:$addr,
6588 rGPR:$Rm, pred:$p)>;
6589 def VLD3dWB_register_Asm_16 :
6590 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
6591 (ins VecListThreeD:$list, addrmode6:$addr,
6592 rGPR:$Rm, pred:$p)>;
6593 def VLD3dWB_register_Asm_32 :
6594 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
6595 (ins VecListThreeD:$list, addrmode6:$addr,
6596 rGPR:$Rm, pred:$p)>;
6597 def VLD3qWB_register_Asm_8 :
6598 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr, $Rm",
6599 (ins VecListThreeQ:$list, addrmode6:$addr,
6600 rGPR:$Rm, pred:$p)>;
6601 def VLD3qWB_register_Asm_16 :
6602 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
6603 (ins VecListThreeQ:$list, addrmode6:$addr,
6604 rGPR:$Rm, pred:$p)>;
6605 def VLD3qWB_register_Asm_32 :
6606 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
6607 (ins VecListThreeQ:$list, addrmode6:$addr,
6608 rGPR:$Rm, pred:$p)>;
6610 // VST3 single-lane pseudo-instructions. These need special handling for
6611 // the lane index that an InstAlias can't handle, so we use these instead.
6612 def VST3LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr",
6613 (ins VecListThreeDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6614 def VST3LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr",
6615 (ins VecListThreeDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6616 def VST3LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr",
6617 (ins VecListThreeDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6618 def VST3LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr",
6619 (ins VecListThreeQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6620 def VST3LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr",
6621 (ins VecListThreeQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6623 def VST3LNdWB_fixed_Asm_8 :
6624 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr!",
6625 (ins VecListThreeDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6626 def VST3LNdWB_fixed_Asm_16 :
6627 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr!",
6628 (ins VecListThreeDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6629 def VST3LNdWB_fixed_Asm_32 :
6630 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr!",
6631 (ins VecListThreeDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6632 def VST3LNqWB_fixed_Asm_16 :
6633 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr!",
6634 (ins VecListThreeQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6635 def VST3LNqWB_fixed_Asm_32 :
6636 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr!",
6637 (ins VecListThreeQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6638 def VST3LNdWB_register_Asm_8 :
6639 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr, $Rm",
6640 (ins VecListThreeDByteIndexed:$list, addrmode6:$addr,
6641 rGPR:$Rm, pred:$p)>;
6642 def VST3LNdWB_register_Asm_16 :
6643 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr, $Rm",
6644 (ins VecListThreeDHWordIndexed:$list, addrmode6:$addr,
6645 rGPR:$Rm, pred:$p)>;
6646 def VST3LNdWB_register_Asm_32 :
6647 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr, $Rm",
6648 (ins VecListThreeDWordIndexed:$list, addrmode6:$addr,
6649 rGPR:$Rm, pred:$p)>;
6650 def VST3LNqWB_register_Asm_16 :
6651 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr, $Rm",
6652 (ins VecListThreeQHWordIndexed:$list, addrmode6:$addr,
6653 rGPR:$Rm, pred:$p)>;
6654 def VST3LNqWB_register_Asm_32 :
6655 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr, $Rm",
6656 (ins VecListThreeQWordIndexed:$list, addrmode6:$addr,
6657 rGPR:$Rm, pred:$p)>;
6660 // VST3 multiple structure pseudo-instructions. These need special handling for
6661 // the vector operands that the normal instructions don't yet model.
6662 // FIXME: Remove these when the register classes and instructions are updated.
6663 def VST3dAsm_8 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr",
6664 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6665 def VST3dAsm_16 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr",
6666 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6667 def VST3dAsm_32 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr",
6668 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6669 def VST3qAsm_8 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr",
6670 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6671 def VST3qAsm_16 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr",
6672 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6673 def VST3qAsm_32 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr",
6674 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6676 def VST3dWB_fixed_Asm_8 :
6677 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr!",
6678 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6679 def VST3dWB_fixed_Asm_16 :
6680 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr!",
6681 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6682 def VST3dWB_fixed_Asm_32 :
6683 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr!",
6684 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6685 def VST3qWB_fixed_Asm_8 :
6686 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr!",
6687 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6688 def VST3qWB_fixed_Asm_16 :
6689 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr!",
6690 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6691 def VST3qWB_fixed_Asm_32 :
6692 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr!",
6693 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6694 def VST3dWB_register_Asm_8 :
6695 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr, $Rm",
6696 (ins VecListThreeD:$list, addrmode6:$addr,
6697 rGPR:$Rm, pred:$p)>;
6698 def VST3dWB_register_Asm_16 :
6699 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr, $Rm",
6700 (ins VecListThreeD:$list, addrmode6:$addr,
6701 rGPR:$Rm, pred:$p)>;
6702 def VST3dWB_register_Asm_32 :
6703 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr, $Rm",
6704 (ins VecListThreeD:$list, addrmode6:$addr,
6705 rGPR:$Rm, pred:$p)>;
6706 def VST3qWB_register_Asm_8 :
6707 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr, $Rm",
6708 (ins VecListThreeQ:$list, addrmode6:$addr,
6709 rGPR:$Rm, pred:$p)>;
6710 def VST3qWB_register_Asm_16 :
6711 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr, $Rm",
6712 (ins VecListThreeQ:$list, addrmode6:$addr,
6713 rGPR:$Rm, pred:$p)>;
6714 def VST3qWB_register_Asm_32 :
6715 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr, $Rm",
6716 (ins VecListThreeQ:$list, addrmode6:$addr,
6717 rGPR:$Rm, pred:$p)>;
6719 // VLD4 all-lanes pseudo-instructions. These need special handling for
6720 // the lane index that an InstAlias can't handle, so we use these instead.
6721 def VLD4DUPdAsm_8 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr",
6722 (ins VecListFourDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6723 def VLD4DUPdAsm_16: NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
6724 (ins VecListFourDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6725 def VLD4DUPdAsm_32: NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
6726 (ins VecListFourDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6727 def VLD4DUPqAsm_8 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr",
6728 (ins VecListFourQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6729 def VLD4DUPqAsm_16: NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
6730 (ins VecListFourQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6731 def VLD4DUPqAsm_32: NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
6732 (ins VecListFourQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6734 def VLD4DUPdWB_fixed_Asm_8 :
6735 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr!",
6736 (ins VecListFourDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6737 def VLD4DUPdWB_fixed_Asm_16 :
6738 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
6739 (ins VecListFourDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6740 def VLD4DUPdWB_fixed_Asm_32 :
6741 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
6742 (ins VecListFourDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6743 def VLD4DUPqWB_fixed_Asm_8 :
6744 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr!",
6745 (ins VecListFourQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6746 def VLD4DUPqWB_fixed_Asm_16 :
6747 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
6748 (ins VecListFourQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6749 def VLD4DUPqWB_fixed_Asm_32 :
6750 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
6751 (ins VecListFourQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6752 def VLD4DUPdWB_register_Asm_8 :
6753 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr, $Rm",
6754 (ins VecListFourDAllLanes:$list, addrmode6:$addr,
6755 rGPR:$Rm, pred:$p)>;
6756 def VLD4DUPdWB_register_Asm_16 :
6757 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
6758 (ins VecListFourDAllLanes:$list, addrmode6:$addr,
6759 rGPR:$Rm, pred:$p)>;
6760 def VLD4DUPdWB_register_Asm_32 :
6761 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
6762 (ins VecListFourDAllLanes:$list, addrmode6:$addr,
6763 rGPR:$Rm, pred:$p)>;
6764 def VLD4DUPqWB_register_Asm_8 :
6765 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr, $Rm",
6766 (ins VecListFourQAllLanes:$list, addrmode6:$addr,
6767 rGPR:$Rm, pred:$p)>;
6768 def VLD4DUPqWB_register_Asm_16 :
6769 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
6770 (ins VecListFourQAllLanes:$list, addrmode6:$addr,
6771 rGPR:$Rm, pred:$p)>;
6772 def VLD4DUPqWB_register_Asm_32 :
6773 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
6774 (ins VecListFourQAllLanes:$list, addrmode6:$addr,
6775 rGPR:$Rm, pred:$p)>;
6778 // VLD4 single-lane pseudo-instructions. These need special handling for
6779 // the lane index that an InstAlias can't handle, so we use these instead.
6780 def VLD4LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr",
6781 (ins VecListFourDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6782 def VLD4LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
6783 (ins VecListFourDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6784 def VLD4LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
6785 (ins VecListFourDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6786 def VLD4LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
6787 (ins VecListFourQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6788 def VLD4LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
6789 (ins VecListFourQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6791 def VLD4LNdWB_fixed_Asm_8 :
6792 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr!",
6793 (ins VecListFourDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6794 def VLD4LNdWB_fixed_Asm_16 :
6795 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
6796 (ins VecListFourDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6797 def VLD4LNdWB_fixed_Asm_32 :
6798 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
6799 (ins VecListFourDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6800 def VLD4LNqWB_fixed_Asm_16 :
6801 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
6802 (ins VecListFourQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6803 def VLD4LNqWB_fixed_Asm_32 :
6804 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
6805 (ins VecListFourQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6806 def VLD4LNdWB_register_Asm_8 :
6807 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr, $Rm",
6808 (ins VecListFourDByteIndexed:$list, addrmode6:$addr,
6809 rGPR:$Rm, pred:$p)>;
6810 def VLD4LNdWB_register_Asm_16 :
6811 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
6812 (ins VecListFourDHWordIndexed:$list, addrmode6:$addr,
6813 rGPR:$Rm, pred:$p)>;
6814 def VLD4LNdWB_register_Asm_32 :
6815 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
6816 (ins VecListFourDWordIndexed:$list, addrmode6:$addr,
6817 rGPR:$Rm, pred:$p)>;
6818 def VLD4LNqWB_register_Asm_16 :
6819 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
6820 (ins VecListFourQHWordIndexed:$list, addrmode6:$addr,
6821 rGPR:$Rm, pred:$p)>;
6822 def VLD4LNqWB_register_Asm_32 :
6823 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
6824 (ins VecListFourQWordIndexed:$list, addrmode6:$addr,
6825 rGPR:$Rm, pred:$p)>;
6829 // VLD4 multiple structure pseudo-instructions. These need special handling for
6830 // the vector operands that the normal instructions don't yet model.
6831 // FIXME: Remove these when the register classes and instructions are updated.
6832 def VLD4dAsm_8 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr",
6833 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6834 def VLD4dAsm_16 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
6835 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6836 def VLD4dAsm_32 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
6837 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6838 def VLD4qAsm_8 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr",
6839 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6840 def VLD4qAsm_16 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
6841 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6842 def VLD4qAsm_32 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
6843 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6845 def VLD4dWB_fixed_Asm_8 :
6846 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr!",
6847 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6848 def VLD4dWB_fixed_Asm_16 :
6849 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
6850 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6851 def VLD4dWB_fixed_Asm_32 :
6852 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
6853 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6854 def VLD4qWB_fixed_Asm_8 :
6855 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr!",
6856 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6857 def VLD4qWB_fixed_Asm_16 :
6858 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
6859 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6860 def VLD4qWB_fixed_Asm_32 :
6861 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
6862 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6863 def VLD4dWB_register_Asm_8 :
6864 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr, $Rm",
6865 (ins VecListFourD:$list, addrmode6:$addr,
6866 rGPR:$Rm, pred:$p)>;
6867 def VLD4dWB_register_Asm_16 :
6868 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
6869 (ins VecListFourD:$list, addrmode6:$addr,
6870 rGPR:$Rm, pred:$p)>;
6871 def VLD4dWB_register_Asm_32 :
6872 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
6873 (ins VecListFourD:$list, addrmode6:$addr,
6874 rGPR:$Rm, pred:$p)>;
6875 def VLD4qWB_register_Asm_8 :
6876 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr, $Rm",
6877 (ins VecListFourQ:$list, addrmode6:$addr,
6878 rGPR:$Rm, pred:$p)>;
6879 def VLD4qWB_register_Asm_16 :
6880 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
6881 (ins VecListFourQ:$list, addrmode6:$addr,
6882 rGPR:$Rm, pred:$p)>;
6883 def VLD4qWB_register_Asm_32 :
6884 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
6885 (ins VecListFourQ:$list, addrmode6:$addr,
6886 rGPR:$Rm, pred:$p)>;
6888 // VST4 single-lane pseudo-instructions. These need special handling for
6889 // the lane index that an InstAlias can't handle, so we use these instead.
6890 def VST4LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr",
6891 (ins VecListFourDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6892 def VST4LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr",
6893 (ins VecListFourDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6894 def VST4LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr",
6895 (ins VecListFourDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6896 def VST4LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr",
6897 (ins VecListFourQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6898 def VST4LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr",
6899 (ins VecListFourQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6901 def VST4LNdWB_fixed_Asm_8 :
6902 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr!",
6903 (ins VecListFourDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6904 def VST4LNdWB_fixed_Asm_16 :
6905 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr!",
6906 (ins VecListFourDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6907 def VST4LNdWB_fixed_Asm_32 :
6908 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr!",
6909 (ins VecListFourDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6910 def VST4LNqWB_fixed_Asm_16 :
6911 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr!",
6912 (ins VecListFourQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6913 def VST4LNqWB_fixed_Asm_32 :
6914 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr!",
6915 (ins VecListFourQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6916 def VST4LNdWB_register_Asm_8 :
6917 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr, $Rm",
6918 (ins VecListFourDByteIndexed:$list, addrmode6:$addr,
6919 rGPR:$Rm, pred:$p)>;
6920 def VST4LNdWB_register_Asm_16 :
6921 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr, $Rm",
6922 (ins VecListFourDHWordIndexed:$list, addrmode6:$addr,
6923 rGPR:$Rm, pred:$p)>;
6924 def VST4LNdWB_register_Asm_32 :
6925 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr, $Rm",
6926 (ins VecListFourDWordIndexed:$list, addrmode6:$addr,
6927 rGPR:$Rm, pred:$p)>;
6928 def VST4LNqWB_register_Asm_16 :
6929 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr, $Rm",
6930 (ins VecListFourQHWordIndexed:$list, addrmode6:$addr,
6931 rGPR:$Rm, pred:$p)>;
6932 def VST4LNqWB_register_Asm_32 :
6933 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr, $Rm",
6934 (ins VecListFourQWordIndexed:$list, addrmode6:$addr,
6935 rGPR:$Rm, pred:$p)>;
6938 // VST4 multiple structure pseudo-instructions. These need special handling for
6939 // the vector operands that the normal instructions don't yet model.
6940 // FIXME: Remove these when the register classes and instructions are updated.
6941 def VST4dAsm_8 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr",
6942 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6943 def VST4dAsm_16 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr",
6944 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6945 def VST4dAsm_32 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr",
6946 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6947 def VST4qAsm_8 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr",
6948 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6949 def VST4qAsm_16 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr",
6950 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6951 def VST4qAsm_32 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr",
6952 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6954 def VST4dWB_fixed_Asm_8 :
6955 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr!",
6956 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6957 def VST4dWB_fixed_Asm_16 :
6958 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr!",
6959 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6960 def VST4dWB_fixed_Asm_32 :
6961 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr!",
6962 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6963 def VST4qWB_fixed_Asm_8 :
6964 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr!",
6965 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6966 def VST4qWB_fixed_Asm_16 :
6967 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr!",
6968 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6969 def VST4qWB_fixed_Asm_32 :
6970 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr!",
6971 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6972 def VST4dWB_register_Asm_8 :
6973 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr, $Rm",
6974 (ins VecListFourD:$list, addrmode6:$addr,
6975 rGPR:$Rm, pred:$p)>;
6976 def VST4dWB_register_Asm_16 :
6977 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr, $Rm",
6978 (ins VecListFourD:$list, addrmode6:$addr,
6979 rGPR:$Rm, pred:$p)>;
6980 def VST4dWB_register_Asm_32 :
6981 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr, $Rm",
6982 (ins VecListFourD:$list, addrmode6:$addr,
6983 rGPR:$Rm, pred:$p)>;
6984 def VST4qWB_register_Asm_8 :
6985 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr, $Rm",
6986 (ins VecListFourQ:$list, addrmode6:$addr,
6987 rGPR:$Rm, pred:$p)>;
6988 def VST4qWB_register_Asm_16 :
6989 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr, $Rm",
6990 (ins VecListFourQ:$list, addrmode6:$addr,
6991 rGPR:$Rm, pred:$p)>;
6992 def VST4qWB_register_Asm_32 :
6993 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr, $Rm",
6994 (ins VecListFourQ:$list, addrmode6:$addr,
6995 rGPR:$Rm, pred:$p)>;
6997 // VMOV/VMVN takes an optional datatype suffix
6998 defm : NEONDTAnyInstAlias<"vmov${p}", "$Vd, $Vm",
6999 (VORRd DPR:$Vd, DPR:$Vm, DPR:$Vm, pred:$p)>;
7000 defm : NEONDTAnyInstAlias<"vmov${p}", "$Vd, $Vm",
7001 (VORRq QPR:$Vd, QPR:$Vm, QPR:$Vm, pred:$p)>;
7003 defm : NEONDTAnyInstAlias<"vmvn${p}", "$Vd, $Vm",
7004 (VMVNd DPR:$Vd, DPR:$Vm, pred:$p)>;
7005 defm : NEONDTAnyInstAlias<"vmvn${p}", "$Vd, $Vm",
7006 (VMVNq QPR:$Vd, QPR:$Vm, pred:$p)>;
7008 // VCLT (register) is an assembler alias for VCGT w/ the operands reversed.
7009 // D-register versions.
7010 def : NEONInstAlias<"vcle${p}.s8 $Dd, $Dn, $Dm",
7011 (VCGEsv8i8 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
7012 def : NEONInstAlias<"vcle${p}.s16 $Dd, $Dn, $Dm",
7013 (VCGEsv4i16 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
7014 def : NEONInstAlias<"vcle${p}.s32 $Dd, $Dn, $Dm",
7015 (VCGEsv2i32 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
7016 def : NEONInstAlias<"vcle${p}.u8 $Dd, $Dn, $Dm",
7017 (VCGEuv8i8 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
7018 def : NEONInstAlias<"vcle${p}.u16 $Dd, $Dn, $Dm",
7019 (VCGEuv4i16 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
7020 def : NEONInstAlias<"vcle${p}.u32 $Dd, $Dn, $Dm",
7021 (VCGEuv2i32 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
7022 def : NEONInstAlias<"vcle${p}.f32 $Dd, $Dn, $Dm",
7023 (VCGEfd DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
7024 // Q-register versions.
7025 def : NEONInstAlias<"vcle${p}.s8 $Qd, $Qn, $Qm",
7026 (VCGEsv16i8 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
7027 def : NEONInstAlias<"vcle${p}.s16 $Qd, $Qn, $Qm",
7028 (VCGEsv8i16 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
7029 def : NEONInstAlias<"vcle${p}.s32 $Qd, $Qn, $Qm",
7030 (VCGEsv4i32 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
7031 def : NEONInstAlias<"vcle${p}.u8 $Qd, $Qn, $Qm",
7032 (VCGEuv16i8 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
7033 def : NEONInstAlias<"vcle${p}.u16 $Qd, $Qn, $Qm",
7034 (VCGEuv8i16 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
7035 def : NEONInstAlias<"vcle${p}.u32 $Qd, $Qn, $Qm",
7036 (VCGEuv4i32 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
7037 def : NEONInstAlias<"vcle${p}.f32 $Qd, $Qn, $Qm",
7038 (VCGEfq QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
7040 // VCLT (register) is an assembler alias for VCGT w/ the operands reversed.
7041 // D-register versions.
7042 def : NEONInstAlias<"vclt${p}.s8 $Dd, $Dn, $Dm",
7043 (VCGTsv8i8 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
7044 def : NEONInstAlias<"vclt${p}.s16 $Dd, $Dn, $Dm",
7045 (VCGTsv4i16 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
7046 def : NEONInstAlias<"vclt${p}.s32 $Dd, $Dn, $Dm",
7047 (VCGTsv2i32 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
7048 def : NEONInstAlias<"vclt${p}.u8 $Dd, $Dn, $Dm",
7049 (VCGTuv8i8 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
7050 def : NEONInstAlias<"vclt${p}.u16 $Dd, $Dn, $Dm",
7051 (VCGTuv4i16 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
7052 def : NEONInstAlias<"vclt${p}.u32 $Dd, $Dn, $Dm",
7053 (VCGTuv2i32 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
7054 def : NEONInstAlias<"vclt${p}.f32 $Dd, $Dn, $Dm",
7055 (VCGTfd DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
7056 // Q-register versions.
7057 def : NEONInstAlias<"vclt${p}.s8 $Qd, $Qn, $Qm",
7058 (VCGTsv16i8 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
7059 def : NEONInstAlias<"vclt${p}.s16 $Qd, $Qn, $Qm",
7060 (VCGTsv8i16 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
7061 def : NEONInstAlias<"vclt${p}.s32 $Qd, $Qn, $Qm",
7062 (VCGTsv4i32 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
7063 def : NEONInstAlias<"vclt${p}.u8 $Qd, $Qn, $Qm",
7064 (VCGTuv16i8 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
7065 def : NEONInstAlias<"vclt${p}.u16 $Qd, $Qn, $Qm",
7066 (VCGTuv8i16 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
7067 def : NEONInstAlias<"vclt${p}.u32 $Qd, $Qn, $Qm",
7068 (VCGTuv4i32 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
7069 def : NEONInstAlias<"vclt${p}.f32 $Qd, $Qn, $Qm",
7070 (VCGTfq QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
7072 // VSWP allows, but does not require, a type suffix.
7073 defm : NEONDTAnyInstAlias<"vswp${p}", "$Vd, $Vm",
7074 (VSWPd DPR:$Vd, DPR:$Vm, pred:$p)>;
7075 defm : NEONDTAnyInstAlias<"vswp${p}", "$Vd, $Vm",
7076 (VSWPq QPR:$Vd, QPR:$Vm, pred:$p)>;
7078 // VBIF, VBIT, and VBSL allow, but do not require, a type suffix.
7079 defm : NEONDTAnyInstAlias<"vbif${p}", "$Vd, $Vn, $Vm",
7080 (VBIFd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
7081 defm : NEONDTAnyInstAlias<"vbit${p}", "$Vd, $Vn, $Vm",
7082 (VBITd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
7083 defm : NEONDTAnyInstAlias<"vbsl${p}", "$Vd, $Vn, $Vm",
7084 (VBSLd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
7085 defm : NEONDTAnyInstAlias<"vbif${p}", "$Vd, $Vn, $Vm",
7086 (VBIFq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
7087 defm : NEONDTAnyInstAlias<"vbit${p}", "$Vd, $Vn, $Vm",
7088 (VBITq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
7089 defm : NEONDTAnyInstAlias<"vbsl${p}", "$Vd, $Vn, $Vm",
7090 (VBSLq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
7092 // "vmov Rd, #-imm" can be handled via "vmvn".
7093 def : NEONInstAlias<"vmov${p}.i32 $Vd, $imm",
7094 (VMVNv2i32 DPR:$Vd, nImmVMOVI32Neg:$imm, pred:$p)>;
7095 def : NEONInstAlias<"vmov${p}.i32 $Vd, $imm",
7096 (VMVNv4i32 QPR:$Vd, nImmVMOVI32Neg:$imm, pred:$p)>;
7097 def : NEONInstAlias<"vmvn${p}.i32 $Vd, $imm",
7098 (VMOVv2i32 DPR:$Vd, nImmVMOVI32Neg:$imm, pred:$p)>;
7099 def : NEONInstAlias<"vmvn${p}.i32 $Vd, $imm",
7100 (VMOVv4i32 QPR:$Vd, nImmVMOVI32Neg:$imm, pred:$p)>;
7102 // 'gas' compatibility aliases for quad-word instructions. Strictly speaking,
7103 // these should restrict to just the Q register variants, but the register
7104 // classes are enough to match correctly regardless, so we keep it simple
7105 // and just use MnemonicAlias.
7106 def : NEONMnemonicAlias<"vbicq", "vbic">;
7107 def : NEONMnemonicAlias<"vandq", "vand">;
7108 def : NEONMnemonicAlias<"veorq", "veor">;
7109 def : NEONMnemonicAlias<"vorrq", "vorr">;
7111 def : NEONMnemonicAlias<"vmovq", "vmov">;
7112 def : NEONMnemonicAlias<"vmvnq", "vmvn">;
7113 // Explicit versions for floating point so that the FPImm variants get
7114 // handled early. The parser gets confused otherwise.
7115 def : NEONMnemonicAlias<"vmovq.f32", "vmov.f32">;
7116 def : NEONMnemonicAlias<"vmovq.f64", "vmov.f64">;
7118 def : NEONMnemonicAlias<"vaddq", "vadd">;
7119 def : NEONMnemonicAlias<"vsubq", "vsub">;
7121 def : NEONMnemonicAlias<"vminq", "vmin">;
7122 def : NEONMnemonicAlias<"vmaxq", "vmax">;
7124 def : NEONMnemonicAlias<"vmulq", "vmul">;
7126 def : NEONMnemonicAlias<"vabsq", "vabs">;
7128 def : NEONMnemonicAlias<"vshlq", "vshl">;
7129 def : NEONMnemonicAlias<"vshrq", "vshr">;
7131 def : NEONMnemonicAlias<"vcvtq", "vcvt">;
7133 def : NEONMnemonicAlias<"vcleq", "vcle">;
7134 def : NEONMnemonicAlias<"vceqq", "vceq">;
7136 def : NEONMnemonicAlias<"vzipq", "vzip">;
7137 def : NEONMnemonicAlias<"vswpq", "vswp">;
7139 def : NEONMnemonicAlias<"vrecpeq.f32", "vrecpe.f32">;
7140 def : NEONMnemonicAlias<"vrecpeq.u32", "vrecpe.u32">;
7143 // Alias for loading floating point immediates that aren't representable
7144 // using the vmov.f32 encoding but the bitpattern is representable using
7145 // the .i32 encoding.
7146 def : NEONInstAlias<"vmov${p}.f32 $Vd, $imm",
7147 (VMOVv4i32 QPR:$Vd, nImmVMOVI32:$imm, pred:$p)>;
7148 def : NEONInstAlias<"vmov${p}.f32 $Vd, $imm",
7149 (VMOVv2i32 DPR:$Vd, nImmVMOVI32:$imm, pred:$p)>;